Polarization in Spectral Lines
Uploaded by
alejmar.perPolarization in Spectral Lines
Uploaded by
alejmar.perPOLARIZATION IN SPECTRAL LINES
ASTROPHYSICS AND
SPACE SCIENCE LIBRARY
VOLUME 307
EDITORIAL BOARD
Chairman
W.B. BURTON, National Radio Astronomy Observatory, Charlottesville, Virginia, U.S.A.
([email protected]); University of Leiden, The Netherlands ([email protected])
Executive Committee
J. M. E. KUIJPERS, Faculty of Science, Nijmegen, The Netherlands
E. P. J. VAN DEN HEUVEL, Astronomical Institute, University of Amsterdam,
The Netherlands
H. VAN DER LAAN, Astronomical Institute, University of Utrecht,
The Netherlands
MEMBERS
I. APPENZELLER, Landessternwarte Heidelberg-Königstuhl, Germany
J. N. BAHCALL, The Institute for Advanced Study, Princeton, U.S.A.
F. BERTOLA, Universitá di Padova, Italy
J. P. CASSINELLI, University of Wisconsin, Madison, U.S.A.
C. J. CESARSKY, Centre d'Etudes de Saclay, Gif-sur-Yvette Cedex, France
O. ENGVOLD, Institute of Theoretical Astrophysics, University of Oslo, Norway
R. McCRAY, University of Colorado, JILA, Boulder, U.S.A.
P. G. MURDIN, Institute of Astronomy, Cambridge, U.K.
F. PACINI, Istituto Astronomia Arcetri, Firenze, Italy
V. RADHAKRISHNAN, Raman Research Institute, Bangalore, India
K. SATO, School of Science, The University of Tokyo, Japan
F. H. SHU, University of California, Berkeley, U.S.A.
B. V. SOMOV, Astronomical Institute, Moscow State University, Russia
R. A. SUNYAEV, Space Research Institute, Moscow, Russia
Y. TANAKA, Institute of Space & Astronautical Science, Kanagawa, Japan
S. TREMAINE, CITA, Princeton University, U.S.A.
N. O. WEISS, University of Cambridge, U.K.
POLARIZATION
IN SPECTRAL LINES
by
EGIDIO LANDI DEGL’INNOCENTI
University of Firenze,
Firenze, Italy
and
MARCO LANDOLFI
Arcetri Observatory,
Firenze, Italy
KLUWER ACADEMIC PUBLISHERS
NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW
eBook ISBN: 1-4020-2415-0
Print ISBN: 1-4020-2414-2
©2005 Springer Science + Business Media, Inc.
Print ©2004 Kluwer Academic Publishers
Dordrecht
All rights reserved
No part of this eBook may be reproduced or transmitted in any form or by any means, electronic,
mechanical, recording, or otherwise, without written consent from the Publisher
Created in the United States of America
Visit Springer's eBookstore at: http://ebooks.springerlink.com
and the Springer Global Website Online at: http://www.springeronline.com
To Nadine and Vanessa,
and to Gianna
PREFACE
Quod si tam celebris est apud omnes gloria Adamantis, atque varia ista
opum gaudia, gemmae unionesque, ad ostentationem tantum placent, ut
digitis colloque circumferantur; non minori afficiendos speraverim gaudio
eos, quibus curiositatis conscientia quam deliciarum est potior, novitate
corporis alicujus, instar crystalli translucidi, quod ex Islandia nuper ad
nos perlatum est; cujus tam mira est constitutio, ut haud sciam, num
alias magis naturae apparuerit gratia.
Erasmus Bartholinus, Experimenta crystalli islandici disdiaclastici
Apart from a few objects of our immediate neighborhood (the solar system), all
the information on the physical phenomena taking place in the Universe comes
from the radiation that the astronomical objects send into space and that is finally
collected on earth by telescopes or other instruments. Among the different kinds
of radiation, electromagnetic waves have by far played the most important role in
the history of Astronomy – probably, it is not unrealistic to say that more than
99% of our present knowledge of the Universe derives from the analysis of the
electromagnetic radiation.
Such radiation contains three different kinds of information, encoded into as
many physical characteristics typical of any oscillatory propagation phenomenon:
the propagation direction, the frequency and amplitude of the oscillation, and the
oscillation direction – or polarization.
The first one is the most direct and the easiest to measure: the human eye is
itself a suitable instrument, though of limited accuracy. As time passed, positional
astronomy became more and more accurate thanks to the invention of the opti-
cal telescope, the introduction of photographic and digital techniques and, more
recently, the development of technologies for producing images of a given region
at different wavelengths via radio, infrared, X-ray and γ-ray telescopes, often op-
erating on board of spacecrafts. Our present knowledge of the morphology and
dynamics of the Universe, and of the different objects of which it is composed
(from planets to stars, from nebulae to globular clusters, from galaxies to Active
Galactic Nuclei and to clusters of galaxies) is based on the huge number of such
observations that have been accumulating during several centuries.
However, even if very accurate, the measurement of the propagation direction
of the electromagnetic radiation is inadequate to study other fundamental aspects
of the physical Universe such as the composition, structure, and evolution of the
different objects. To this aim, a detailed analysis of the frequency (or wavelength)
distribution of the energy carried by the electromagnetic radiation is required,
which was made possible by the invention of the spectrograph. Only through the
systematic use of spectroscopic methods it has been possible to obtain a direct
comprehension of the physical mechanisms which govern the equilibrium of stars,
their birth, evolution and death, and the complicated processes taking place in
the interstellar medium and in the nuclei of galaxies. Spectroscopy, which is also
viii PREFACE
at the basis of the idea of an expanding Universe, has played such a key role in
the comprehension of the physical Universe that a new name, astrophysics, was
introduced in the scientific lexicon to denote the astronomical research based on
this technique.
The third, and often neglected, characteristic of the electromagnetic radiation
is polarization. The earliest studies on polarization were performed around 1670
by the Danish scientist Erasmus Bartholinus, who was strongly impressed by the
properties of a newly discovered crystal, the Iceland spar (‘. . . whose behavior is so
surprising that, as far as I know, never the grace of nature appeared more clearly’)
and who immediately realized that those properties could prove useful to improve
human knowledge.
Since polarization is mainly related to the geometrical aspects of the emission
process (rather than to its energetics), and since polarization measurements are
often difficult to perform because of the intrinsic weakness of the signals, the study
of polarization found its place in the astronomical research with some difficulty.
But eventually the prediction of Erasmus Bartholinus was fully confirmed: some of
the most important astronomical discoveries of the 20th century were made thanks
to polarimetry – or, more properly, spectropolarimetry. Suffice it to quote the
discovery of magnetic fields in the sun, stars, and the interstellar medium.
The first application of spectropolarimetry to the astronomical research dates
back to 1908 when, using a Nicol prism as a polarizer and a Fresnel rhomb as a
quarter-wave plate, George Ellery Hale succeeded in taking two spectra of the same
area of a sunspot in opposite directions of circular polarization. The comparison
of the spectra showed the presence of the typical signature induced by a strong
magnetic field, the Zeeman effect.
Since 1908, things have considerably evolved from the technological point of view.
Spectropolarimetric observations of the solar spectrum have now attained a sensi-
tivity level which goes beyond the most optimistic expectations of only two or three
decades ago. The first dedicated instrument for the measurement of Stokes parame-
ters profiles in Fraunhofer lines, the ‘mythic’ Stokes-I polarimeter, developed in the
mid 1970s at the High Altitude Observatory, hardly attained a sensitivity of 1%.
Nowadays, sensitivity in solar spectropolarimetry has reached the level of 10−5 for
spatially unresolved observations, and approximately one order of magnitude less
for observations at high spatial and temporal resolution. It has to be expected that
these limits will be rapidly overcome by the next generation polarimeters and that
the same technologies will be adapted to galactic and extragalactic observations.
The dramatic increase of polarimetric sensitivity in solar observations has raised
a serious challenge to the theoretical interpretation. Polarization in spectral lines
is indeed a subtle phenomenon since, in astrophysical plasmas, there are several
physical mechanisms that can generate polarization signatures in line profiles and
many others that can modify them during the propagation. Some of these mech-
anisms have been known for a long time from laboratory atomic physics. They
are – just to mention the most remarkable – the Zeeman effect, resonance po-
larization, and the Hanle effect. Other mechanisms are characteristic of optically
thick plasmas, and are related to the propagation of radiation in anisotropic media.
PREFACE ix
They are known under the general names of dichroism and anomalous dispersion,
though in special cases different denominations are often used (inverse Zeeman
effect, magneto-optical effects, Faraday rotation, Faraday pulsation, etc.).
These processes have mostly been studied in specific physical contexts, for differ-
ent purposes and at different levels of sophistication, and the scientific literature
on the subject is scattered across books and journals, spanning almost a century
of active research. For this reason we felt that a book capable of describing, in
a unique and self-consistent formalism, all the known physical phenomena that
may affect the polarization signatures of spectral lines, might be welcome to the
scientific community. The diagnostic content of spectropolarimetry is high, but
the correct interpretation of the observations requires a full understanding of the
physics underlying the generation and transfer of polarized radiation.
The redaction of this book required several years. We might try to say, like Huy-
gens in the preface to his Treatise on light, ‘The reason is that I wrote it rather
carelessly in the Language in which it appears, with the intention of translating it
into Latin, so doing in order to obtain greater attention to the thing’,1 but we feel
it would be hardly believed. The true reason is that the theory of spectropolarime-
try is complicated, because it implies some knowledge of several subjects: atomic
physics, quantum mechanics (with special emphasis on the theory of angular mo-
mentum and of the density matrix), quantum electrodynamics, radiative transfer
(both under LTE and Non-LTE conditions).
Moreover, spectropolarimetry is full of traps: among all the disciplines in as-
trophysics, there can hardly be found one where more attention has to be payed
to each single definition, each transformation, each physical application. Sign er-
rors are especially insidious, as most remarkably shown by the classical example of
circular polarization in a given wing of a spectral line formed in a magnetic atmo-
sphere. There are four operations which produce a sign change in such polarization,
and obviously, there are as many possibilities to make a sign error. To have a sign
switch, one can: a) invert the direction of the magnetic field; b) interchange the red
with the blue wing; c) use the opposite definition of positive and negative circular
polarization; d) consider an emission line instead of an absorption line. This is just
an example, but it shows very well the subtleties of the subject. We tried to make
the exposition as clear as possible by using everywhere the same definitions and
conventions, and by carefully describing all the mathematical developments.
We hope that this book may be useful to the next generations of scholars that
will like to enter the field of spectropolarimetry, solar and non-solar. And we hope
that it may contribute to make this research field more accessible and less hermetic,
thus attracting more and more scientists to the fascinanting world of the Stokes
parameters profiles and of their interpretation.
Arcetri, March 2004 Egidio Landi Degl’Innocenti
Marco Landolfi
1
Christiaan Huygens, Treatise on light (1690), translated by S.P. Thompson, Dover
Publications, New York, 1962.
CONTENTS
CHAPTER 1. DESCRIPTION OF POLARIZED RADIATION 1
1.1. The Polarization Ellipse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Special Cases of the Polarization Ellipse . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3. Polarization Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4. Quasi-monochromatic Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5. Polarizers and Retarders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6. Stokes Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7. Measurements of the Stokes Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.8. Stokes Parameters and Polarization Tensor. . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.9. Properties of the Stokes Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.10. Photons and Stokes Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
CHAPTER 2. ANGULAR MOMENTUM AND RACAH ALGEBRA 29
2.1. Eigenvalues and Eigenvectors of Angular Momentum . . . . . . . . . . . . . . . . 29
2.2. Coupling of Two Angular Momenta: Vector-Coupling Coefficients
and 3-j Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3. Coupling of Three Angular Momenta: Racah Coefficients
and 6-j Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4. Coupling of Four Angular Momenta: 9-j Symbols . . . . . . . . . . . . . . . . . . . 45
2.5. Rotations and Euler Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.6. Rotation Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7. Irreducible Spherical Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.8. The Wigner-Eckart Theorem and its Consequences . . . . . . . . . . . . . . . . . . 65
2.9. Properties of Reduced Matrix Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
CHAPTER 3. ATOMIC SPECTROSCOPY 73
3.1. Zeeman Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2. Classical Theory of the Zeeman Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.3. Classification of Zeeman Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.4. The Paschen-Back Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.5. Magnetic Field and Hyperfine Structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.6. Atomic Level Polarization and Density Matrix . . . . . . . . . . . . . . . . . . . . . . 115
3.7. Multipole Moments of the Density Matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . 122
xii CONTENTS
CHAPTER 4. QUANTIZATION OF THE ELECTROMAGNETIC FIELD
(NON-RELATIVISTIC THEORY) 131
4.1. Quantization of the Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2. The Electromagnetic Field as a Superposition of Plane Waves . . . . . . . 134
4.3. Quantization of the Electromagnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.4. The Stokes Parameters in the Formalism of Second Quantization . . . . 141
4.5. The Density Operator of the Radiation Field. . . . . . . . . . . . . . . . . . . . . . . . 144
CHAPTER 5. INTERACTION OF MATERIAL SYSTEMS WITH POLARIZED
RADIATION (THE CLASSICAL APPROACH) 147
5.1. Propagation of Electromagnetic Waves in Anisotropic Media . . . . . . . . 147
5.2. Transfer Equations for Polarized Radiation. . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.3. Application to Magnetic Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.4. The Voigt Function and the Associated Dispersion Profile . . . . . . . . . . . 162
5.5. Symmetry Properties of the Transfer Equations
for Polarized Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.6. Geometrical Interpretation of the Transfer Equations
for Polarized Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.7. Resonance Scattering and the Hanle Effect . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.8. The Scattering Phase Matrix in a Particular Case . . . . . . . . . . . . . . . . . . . 185
5.9. Some Illustrations of the Hanle Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.10. The Scattering Phase Matrix Expressed in Terms
of Rotation Matrices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.11. Spherical Tensors for Polarimetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
5.12. Further Properties of the Scattering Phase Matrix . . . . . . . . . . . . . . . . . . 212
5.13. Understanding Scattering Experiments through Oscillator Models . . . 215
5.14. The Role of Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
5.15. Some Properties of the Collisional Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.16. Classification of the Physical Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
CHAPTER 6. INTERACTION OF MATERIAL SYSTEMS WITH POLARIZED
RADIATION (THE QUANTUM APPROACH) 237
6.1. Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.2. The Interaction Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
6.3. The Dipole Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.4. Approximate Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6.5. Evolution Equations for the Atomic System . . . . . . . . . . . . . . . . . . . . . . . . . 252
6.6. Evolution Equations for the Radiation Field . . . . . . . . . . . . . . . . . . . . . . . . 265
6.7. Evolution Equations for the Stokes Parameters . . . . . . . . . . . . . . . . . . . . . . 270
CONTENTS xiii
6.8. Magnetic Dipole Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
CHAPTER 7. STATISTICAL EQUILIBRIUM EQUATIONS AND RADIATIVE
TRANSFER COEFFICIENTS FOR ATOMIC SYSTEMS 277
7.1. The Multi-Level Atom in the Standard Representation . . . . . . . . . . . . . . 278
7.2. The Multi-Level Atom in the Spherical Statistical Tensor
Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
7.3. Conjugation Properties of the Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
7.4. The No-Coherence Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
7.5. The Multi-Term Atom in the Energy-Eigenvector Representation . . . . 295
7.6. The Multi-Term Atom in the Spherical Statistical Tensor
Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
7.7. Conjugation Properties of the Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
7.8. The Multi-Level Atom as a Special Case of the Multi-Term Atom . . . 312
7.9. The Multi-Level Atom with Hyperfine Structure . . . . . . . . . . . . . . . . . . . . 315
7.10. The Principle of Spectroscopic Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
7.11. Selection Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
7.12. Changing the Reference System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
7.13. Collisional Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
CHAPTER 8. RADIATIVE TRANSFER FOR POLARIZED RADIATION 349
8.1. Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
8.2. Formal Solution of the Radiative Transfer Equations:
the Evolution Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
8.3. Analytical Expressions for the Evolution Operator . . . . . . . . . . . . . . . . . . 353
8.4. Solution of the Radiative Transfer Equations by Diagonalization. . . . . 358
8.5. Evolution Operator for Purely Dichroic or Purely Dispersive Media. . 362
8.6. Evolution Operator and Mueller Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
8.7. Perturbative Solution of the Radiative Transfer Equations. . . . . . . . . . . 366
8.8. An Alternative Form of the Radiative Transfer Equations . . . . . . . . . . . 368
8.9. Solution of the Alternative Form of the Radiative Transfer
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
CHAPTER 9. LINE FORMATION IN A MAGNETIC FIELD 375
9.1. Transfer Equation for Polarized Radiation in a Magnetized Stellar
Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
9.2. Comparison with the Classical Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
xiv CONTENTS
9.3. Collisional and Doppler Broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
9.4. Different Forms of the Transfer Equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
9.5. Generalities and Symmetry Properties of the Transfer Equation . . . . . 389
9.6. The Weak Field Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
9.7. Formal Solution through the Evolution Operator . . . . . . . . . . . . . . . . . . . . 409
9.8. The Unno-Rachkovsky Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
9.9. More General Analytical Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
9.10. Solutions for Special Magnetic Field Orientations . . . . . . . . . . . . . . . . . . . 422
9.11. The Stepanov Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
9.12. The Intense Field Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
9.13. The Seares Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
9.14. The Magnetic Field with Variable Azimuth . . . . . . . . . . . . . . . . . . . . . . . . . 435
9.15. Numerical Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
9.16. Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
9.17. Contribution Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
9.18. Blends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
9.19. The Magnetic Intensification Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
9.20. Net Linear Polarization in Spectral Lines: the Differential
Saturation Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
9.21. Net Circular Polarization in Spectral Lines . . . . . . . . . . . . . . . . . . . . . . . . . . 473
9.22. The Importance of Magneto-Optical Effects . . . . . . . . . . . . . . . . . . . . . . . . . 488
9.23. Transfer Equation for Fine-Structured and Hyperfine-Structured
Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
9.24. Line Formation in Stochastic Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
9.25. Isotropic, Microturbulent Magnetic Field. . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
CHAPTER 10. NON-EQUILIBRIUM ATOMIC PHYSICS 509
10.1. The Two-Level Atom: Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
10.2. The Two-Level Atom: Resonance Polarization . . . . . . . . . . . . . . . . . . . . . . 513
10.3. The Two-Level Atom: the Hanle Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
10.4. The Two-Level Atom: Spectral Details of the Hanle Effect . . . . . . . . . . 524
10.5. The Two-Level Atom: Resonance Polarization for Strong
Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528
10.6. The Two-Level Atom: the Role of Collisions . . . . . . . . . . . . . . . . . . . . . . . . 532
10.7. The Two-Level Atom: The Role of Lower-Level Polarization. . . . . . . . . 535
10.8. The Two-Level Atom: the Hanle Effect with Lower-Level
Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
10.9. The Two-Level Atom: the Role of Stimulated Emission . . . . . . . . . . . . . 551
10.10. Three-Level Atoms: the Stepwise Excitation . . . . . . . . . . . . . . . . . . . . . . . . 557
10.11. Three-Level Atoms: the Raman Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
10.12. Three-Level Atoms: an Example Leading to Population Inversion . . . 563
10.13. The Weak Anisotropy Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
CONTENTS xv
10.14. The Two-Level Atom in the Weak Anisotropy Approximation . . . . . . . 570
10.15. The Two-Term Atom: Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 580
10.16. The Two-Term Atom: Resonance Polarization . . . . . . . . . . . . . . . . . . . . . . 584
10.17. The Two-Term Atom: Spectral Details of Resonance Polarization . . . 589
10.18. The Two-Term Atom: the Hanle Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
10.19. The Two-Term Atom: the Franken Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 605
10.20. The Two-Term Atom: the Alignment-to-Orientation Conversion
Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
10.21. The Two-Term Atom: the Role of Lower-Term Polarization . . . . . . . . . 609
10.22. The Two-Level Atom with Hyperfine Structure . . . . . . . . . . . . . . . . . . . . . 616
CHAPTER 11. ASTROPHYSICAL APPLICATIONS:
SOLAR MAGNETOMETRY 625
11.1. The Longitudinal Magnetograph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 626
11.2. The Vector Magnetograph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
11.3. The Unno-fit Technique. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634
11.4. The Bisector and the Center-of-Gravity Techniques . . . . . . . . . . . . . . . . . 639
11.5. Unresolved Fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 644
11.6. Other Inversion Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654
11.7. Disambiguation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 661
CHAPTER 12. ASTROPHYSICAL APPLICATIONS:
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 663
12.1. The Milne-Eddington Model Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
12.2. The Grey Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 668
12.3. Outer Atmospheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673
12.4. Symmetry-Breaking Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 677
CHAPTER 13. ASTROPHYSICAL APPLICATIONS:
THE OUTER LAYERS OF STELLAR ATMOSPHERES 689
13.1. The Flat-Spectrum Approximation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
13.2. Velocity/Density-Matrix Correlations and the Approximation of
Complete Redistribution on Velocities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
13.3. Resonance Polarization and the Hanle Effect in the Absence of
Velocity/Density-Matrix Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 698
13.4. Diagnostics of Magnetic Fields in Solar Prominences . . . . . . . . . . . . . . . . 710
13.5. Diagnostics of Magnetic Fields from Coronal Forbidden Lines . . . . . . . 715
xvi CONTENTS
13.6. Resonance Polarization in the Presence of Velocity/Density-Matrix
Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 720
13.7. The Hanle Effect in the Presence of Velocity/Density-Matrix
Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 730
CHAPTER 14. ASTROPHYSICAL APPLICATIONS:
STELLAR ATMOSPHERES 737
14.1. The Non-LTE Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 737
14.2. The Two-Level Atom: Non-LTE Theory for Weak Magnetic Fields
(Hanle Effect Regime) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 742
14.3. The Two-Level Atom: Non-LTE Theory for Strong Magnetic Fields 760
14.4. The Non-LTE Regime of Order 1.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767
14.5. The Non-LTE Problem for More Complicated Atomic Models . . . . . . . 769
14.6. Applications to Realistic Atmospheres: Polarization in the
Continuum Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774
14.7. Applications to Realistic Atmospheres: Approximate Results on the
Polarization of the Fraunhofer Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783
14.8. Alternative Methods for the Solution of the Non-LTE Problem . . . . . . 787
APPENDIX
A1. A Fortran Code for Computing 3-j, 6-j, and 9-j Symbols. . . . . . . . . . . . 791
A2. Sample Evaluation of a Quantity Involving the Contraction
of 3-j Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794
A3. Momentum and Angular Momentum of the Electromagnetic Field . . . 796
A4. Multipole Components of Collisional Rates . . . . . . . . . . . . . . . . . . . . . . . . . . 799
A5. Explicit Expression for the Exponential of the Propagation Matrix . . 803
A6. Diagonalization of the Propagation Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 809
A7. Formulae for the Calculation of the Evolution Operator . . . . . . . . . . . . . 812
A8. The Feautrier Method: Numerical Details . . . . . . . . . . . . . . . . . . . . . . . . . . . 813
A9. The Diagonal Element Lambda-Operator (DELO) Method:
Numerical Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 818
A10. Equivalent Width in the Presence of Depth-Dependent Line Shifts. . . 820
A11. Net Circular Polarization in Blends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 822
A12. Evolution Operator in Stochastic Media. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824
A13. Properties of the Generalized Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 829
A14. Properties of the Symbol [WKK Q (β L Sβu Lu ; B)]fs . . . . . . . . . . . . . . . . . 835
A15. A Property of the Hopf Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 838
A16. A Numerical Algorithm for the Solution of the Hopf Equation . . . . . . . 840
A17. Symmetry Properties of the Comoving-Frame Radiation Field
Tensor for a Cylindrically Symmetrical Atmosphere . . . . . . . . . . . . . . . . . 841
CONTENTS xvii
A18. Redistribution Matrix for a Maxwellian Distribution of Velocities. . . . 843
A19. Properties of the Kernel KQQ
K
(RB ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844
A20. The Multipole Coupling Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 845
A21. The Calculation of a Double
√ Integral. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 850
A22. The Generalization of the - Law. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853
A23. The Generalized Multipole Coupling Coefficients . . . . . . . . . . . . . . . . . . . . 859
A24. Reduced Matrix Elements for Photoionization Cross Sections . . . . . . . . 863
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 867
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 869
AUTHOR INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
SUBJECT INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 885
CHAPTER 1
DESCRIPTION OF POLARIZED RADIATION
According to Maxwell’s equations, the electric and magnetic fields associated with
a radiation beam propagating in vacuum are perpendicular to each other and to
the direction of propagation. The phenomena of polarization are connected with
the possibility that is left to the electric field vector (or, alternatively, to the asso-
ciated magnetic field vector) of spanning the plane perpendicular to the direction
of propagation.
The polarization properties of a radiation beam can be described in several dif-
ferent ways, each of them having its proper advantages and disadvantages. In the
following we will go through some of these descriptions showing how they are in-
terrelated and, most of all, how they can be translated into operational definitions
having a direct physical meaning in terms of measured quantities.
1.1. The Polarization Ellipse
We start from the classical (non-quantum) description of the electromagnetic field
and we refer to the idealized case of a pure monochromatic, plane wave propagating
in vacuum along the positive z-axis of a right-handed coordinate system. A full
description of the wave can be given either specifying its electric field vector or,
alternatively, its magnetic field vector, as the two quantities are related by the
equation
r, t) = n × E(
B( r, t) , (1.1)
where n is the unit vector in the direction of propagation, and where E and B are
both measured in c.g.s. units (the Gauss-Hertz system, that will be used – with
some few exceptions – throughout the whole book). As far as the cross product
is concerned, we will follow the almost universally accepted convention that is
referred to as the right-hand (or screwdriver ) rule, and that is illustrated in Fig. 1.1.
However, we remind the reader that the sign in Eq. (1.1) is related to the sign
conventions which have been historically adopted to define the positive directions
for E and B. These are, in turn, connected with the historical convention of
adopting the positive sign for the proton charge and, in magnetostatic experiments,
the positive sign for the idealized monopole which is found at the North-end of a
dipole oriented in the earth magnetic field (Maxwell, 1873). Note that, according
to this definition, the earth presently shows a South magnetic pole close to the
North geographic pole.
To define the polarization properties of an electromagnetic wave we choose its
representation in terms of electric field vibrations. Again, there are historical rea-
2 CHAPTER 1
Fig.1.1. Mutual relationship among instantaneous electric field vector, instantaneous magnetic
field vector and direction of propagation of an ordinary electromagnetic wave. The cross product
is defined according to the right-hand convention.
sons behind this choice; the description in terms of electric vibrations is nowadays
preferred because polarization measurements are mostly performed, at least in the
optical spectrum, by means of materials whose interaction with the electromagnetic
radiation is dominated by the electric vector. However, a description in terms of
magnetic vibrations would be as appropriate as the former.
In the right-handed coordinate system introduced above, with its x-axis pointing
in an arbitrary direction in the plane perpendicular to the direction of propagation,
the electromagnetic wave is described by the following expressions
Ex (r, t) = E1 cos(kz − ωt + φ1 )
Ey (r, t) = E2 cos(kz − ωt + φ2 ) , (1.2)
where E1 , E2 , φ1 , and φ2 are four positive constants specifying the amplitudes and
phases of the electric oscillations, and where k and ω have the usual meaning of
wavenumber and angular frequency.
In a given plane that is held fixed in space, for instance in the plane z = 0, the
electric field vector oscillates according to the equations
Ex (t) = E1 cos(ωt − φ1 )
Ey (t) = E2 cos(ωt − φ2 ) . (1.3)
The tip of the electric field vector rotates in the x-y plane describing an ellipse that
is called the polarization ellipse and whose characteristic parameters can be found
with the following algebraic manipulations.
Let us consider the couple of axes (x y ) that are obtained by rotating the old
couple (xy) through an angle α measured positively from the x-axis to the y-axis
(see Fig. 1.2). In this new system, with the position
C = cos α , S = sin α
we have
Ex (t) = (E1 C cos φ1 + E2 S cos φ2 ) cos ωt
+ (E1 C sin φ1 + E2 S sin φ2 ) sin ωt
Ey (t) = (−E1 S cos φ1 + E2 C cos φ2 ) cos ωt
+ (−E1 S sin φ1 + E2 C sin φ2 ) sin ωt . (1.4)
DESCRIPTION OF POLARIZED RADIATION 3
Fig.1.2. Polarization ellipse.
We now choose the angle α so as to make the x and y axes coincide with the
proper axes of the polarization ellipse, the long axis being directed along x . If a
and b are the semi-major and semi-minor axes, respectively, we have
Ex (t) = a cos(ωt − φ0 )
Ey (t) = b sin(ωt − φ0 ) , (1.5)
where the phase φ0 must be determined in such a way that a > 0 and a ≥ |b|.
Comparing Eqs. (1.4) and (1.5) we obtain
a cos φ0 = E1 C cos φ1 + E2 S cos φ2
a sin φ0 = E1 C sin φ1 + E2 S sin φ2
b cos φ0 = −E1 S sin φ1 + E2 C sin φ2
b sin φ0 = E1 S cos φ1 − E2 C cos φ2 .
From these equations we easily get
a2 + b2 = E12 + E22
ab = −E1 E2 sin(φ1 − φ2 )
a2 − b2 = (E12 − E22 ) cos 2α + 2E1 E2 cos(φ1 − φ2 ) sin 2α
(E12 − E22 ) sin 2α = 2E1 E2 cos(φ1 − φ2 ) cos 2α . (1.6)
Equations (1.6) show that the geometrical properties of the polarization ellipse
depend on four bilinear combinations of the electric field components. Introducing
the notations
PI = E12 + E22
PQ = E12 − E22
PU = 2E1 E2 cos(φ1 − φ2 )
PV = 2E1 E2 sin(φ1 − φ2 ) , (1.7)
4 CHAPTER 1
the angle α can be found from the equation
PQ sin 2α = PU cos 2α
added to the further constraint
a2 − b2 = PQ cos 2α + PU sin 2α ≥ 0 .
If we choose (0, π) as the interval of definition for α, we obtain, for PQ = 0
1 PU
α = arctan + α0 (1.8a)
2 PQ
where ⎧
⎪
⎨0 if PQ > 0 and PU ≥ 0
α0 = π if PQ > 0 and PU < 0 (1.8b)
⎪
⎩ π/2 if P < 0 ,
Q
and for PQ = 0
1
4π if PU > 0
α= 3
(1.8c)
4π if PU < 0 .
Obviously, in the case PQ = PU = 0 the angle α is left undefined.
From Eqs. (1.6) we can also find the values of the semiaxes of the polarization
ellipse. Since
(a + b)2 = PI − PV ,
the values of a and b can be found as the solutions of the second degree equation
1
x2 − PI − PV x − PV = 0 ,
2
which gives
1
a= PI − PV + PI + PV
2
1
b= PI − PV − PI + PV . (1.9)
2
Note that the four quantities defined in Eqs. (1.7) are not independent, being
related by the expression
PI2 = PQ2 + PU2 + PV2 . (1.10)
DESCRIPTION OF POLARIZED RADIATION 5
1.2. Special Cases of the Polarization Ellipse
For some special values of the parameters E1 , E2 , and (φ1 − φ2 ), the polarization
ellipse degenerates either into a segment or into a circle. The first case takes place
when PV = 0, which means that either E1 or E2 vanish, or, alternatively, that
φ1 = φ2 or φ1 = φ2 ± π. From Eqs. (1.9) we get
a = PI , b=0,
while the angle α can still be deduced from Eqs. (1.8). In this case the monochro-
matic wave is said to be linearly polarized, and its electric field oscillates in the
constant plane containing the direction of propagation and the direction, in the
normal plane, characterized by the angle α.
The other case corresponds to PQ = PU = 0, which means that E1 = E2 and that
(φ1 − φ2 ) = ± π/2; in other words the electric oscillations have the same amplitude
along the x and y axes and are in phase quadrature. For the fourth parameter we
then obtain
PV = ± PI ,
and correspondingly
a= PI /2 , b = ∓ PI /2 .
In both cases the polarization ellipse degenerates into a circle. When PV = +PI ,
the tip of the electric field vector rotates clockwise for an observer facing the ra-
diation source, as apparent from Eqs. (1.5) being a > 0, b < 0. Conversely, when
PV = −PI the rotation is counterclockwise as seen by the same observer. In this
book we will adopt the convention of referring to the first case as positive (or right-
handed ) circular polarization, and to the second case as negative (or left-handed )
circular polarization. In the first case a snapshot of the electromagnetic wave shows
that the end point of the electric field vector draws a helix that fits the thread of
a usual, right-handed screw; in the second case we have a left-handed screw.
Our convention, which is summarized in Fig. 1.3, agrees with those proposed in
the classical textbooks on polarized light by Shurcliff (1962) and by Clarke and
Grainger (1971). The same convention is also used, although with some few excep-
tions, by optical astronomers working in the field of polarimetry. Many radioas-
tronomers, on the other hand, use the opposite convention, so that the situation
in this field is still rather confusing (see for instance Clarke, 1974).
The sign of circular polarization is also connected with the helicity (or spin) of
photons. This connection is discussed in some detail in the following of this book
(see Sect. 4.4 and App. 3).
1.3. Polarization Tensor
In Sect. 1.1 we have described the electric vibration of a plane monochromatic
wave by means of the real quantities E1 , E2 , φ1 , φ2 . Alternatively, the electric
6 CHAPTER 1
positive, negative,
right-handed left-handed
Fig.1.3. Conventions for defining the sign and handedness of circular polarization. The tip of the
electric field vector rotates, as a function of time, in a fixed plane perpendicular to the direction
of propagation; the observer, receiving the radiation into his/her eye, sees a clockwise (a) or
counterclockwise (b) rotation. The corresponding snapshots are shown in the lower half of the
figure.
vibration can be described using complex numbers, with the usual convention that
the physical quantity is represented by the real part of the complex number. The
first of Eqs. (1.2) can then be rewritten in either of the two equivalent forms (i is
the imaginary unit)
i(kz−ωt)
Ex (r, t) = Re E1 e (1.11)
where E1 = E1 eiφ1 , or
i(−kz+ωt)
Ex (r, t) = Re E1 e (1.12)
where E1 = E1 e−iφ1 is the complex conjugate of E1 . Here and in the following we
will use the further convention of describing oscillating quantities with a temporal
exponential of the form e−iωt (rather than eiωt ), so that the expression for Ex (r, t)
in terms of complex notations is that of Eq. (1.11). With the same convention, we
have for the electric oscillation along the y-axis
i(kz−ωt)
Ey (r, t) = Re E2 e (1.13)
with E2 = E2 eiφ2 .
Using the complex quantities E1 and E2 it is possible to introduce a 2×2 Hermitian
matrix J, called the polarization tensor , defined by
Jij = Ei∗ Ej (i, j = 1, 2) ,
with
∗
Jij = Jji . (1.14)
The four parameters describing the geometrical properties of the polarization ellipse
(Eqs. (1.7)) are related to the components of the polarization tensor by the simple
DESCRIPTION OF POLARIZED RADIATION 7
expressions
PI = E1∗ E1 + E2∗ E2
PQ = E1∗ E1 − E2∗ E2
PU = E1∗ E2 + E2∗ E1
PV = i (E1∗ E2 − E2∗ E1 ) . (1.15)
Note that a minus sign would appear in the expression for PV if Eq. (1.12) were
used.
After inversion of Eqs. (1.15) the polarization tensor can be written as
1 PI + PQ PU − iPV
J= ,
2 PU + iPV PI − PQ
which can be cast into a more synthetic form. Defining the formal vector
⎛ ⎞ ⎛ ⎞
P0 PI
⎜ P1 ⎟ ⎜ PQ ⎟
⎝ ⎠=⎝ ⎠ ,
P2 PU
P3 PV
and introducing the 2 × 2 identity matrix σ0 and the Pauli spin matrices σ1 , σ2 ,
and σ3 , we can write1
1
3
J= P σ , (1.16)
2 i=0 i i
where
1 0 1 0 0 1 0 −i
σ0 = , σ1 = , σ2 = , σ3 = (1.17)
0 1 0 −1 1 0 i 0
with
σj σk = δjk σ0 + i jkl σl (j, k, l = 1, 2, 3) . (1.18)
l
In the last equation δjk is the Kronecker symbol – equal to 1 for j = k and to 0
for j = k – and jkl is the antisymmetric (or Levi-Civita) tensor – equal to +1 if
(j, k, l) is an even permutation of (1, 2, 3), to −1 if (j, k, l) is an odd permutation
of (1, 2, 3), and to 0 if two (or three) indices are equal.
Multiplying Eq. (1.16) by σj and using the relation
Tr (σj σi ) = 2 δji (i, j = 0, . . . , 3) , (1.19)
1
The representation that we have chosen here for the Pauli spin matrices differs from the
standard one, initially introduced by Pauli (1927) and then followed in classical textbooks on
Quantum Mechanics (Schiff, 1949; Messiah, 1961; etc.). The matrices introduced by Pauli (σx ,
σy , σz ) are connected with the ones defined here by the relations: σx = σ2 , σy = σ3 , σz = σ1 .
8 CHAPTER 1
we can rewrite Eqs. (1.15) in the form
Pj = Tr (σj J ) . (1.20)
Finally, Eq. (1.10) can be expressed in the compact form
det J = 0 . (1.21)
1.4. Quasi-monochromatic Wave
The pure monochromatic wave considered in Sect. 1.1 is nothing but a mathemati-
cal abstraction; in the physical world one always deals with radiation beams having
a non-zero angular spread and a finite frequency bandwidth. Such beams can be
described using wave packets of the form
r, t) = Re
E( k ) e i(k ·r−ω t) n(k ) d3k
E( ,
k ) is the complex electric field amplitude of the single wave propagating
where E(
with wavenumber k and angular frequency ω = c |k |, with
k ) · k = 0 ,
E(
and where n(k ) is the number density of waves in the three-dimensional wavenum-
ber space.
We now suppose the wave packet to be confined into a small range of wavenum-
bers ∆k centered around k. This means that our packet has an angular spread
in the solid angle ∆Ω ≈ |∆k|/|k| and a finite frequency bandwidth ∆ω ≈ c |∆k|.
Writing
k = k + δk , ω = ω + δω
we obtain
r, t) = Re e i(k·r−ωt) k + δk ) e i(δk·r−δω t) n(δk ) d3 (δk ) .
E( E( (1.22)
Let us consider a surface element Σ perpendicular to the k direction, and a co-
ordinate system (xyz) as shown in Fig. 1.4. At any point P of the surface Σ the
components of the electric field vector are given by
−iωt
Ex (P, t) = Re E1 (P, t) e
−iωt
Ey (P, t) = Re E2 (P, t) e , (1.23)
DESCRIPTION OF POLARIZED RADIATION 9
Σ ∆Ω
Fig.1.4. Coordinate system for defining the various components of the electric vibration of a
quasi-monochromatic wave. The vector k is directed along the z-axis.
where E1 (P, t) and E2 (P, t) are the x and y components, respectively, of the vector
resulting from the evaluation of the integral in Eq. (1.22). The third component
Ez (P, t) is much smaller than the x and y components (a typical ratio Ez /Ex being
of order ∆Ω) and will be neglected in the following.
If the linear dimensions L of the surface Σ are such that L|∆k| 1, the quan-
tities E1 (P, t) and E2 (P, t) are weakly varying functions of the point P. Moreover,
they are slowly varying functions of time, having temporal variations on typical
time-scales much larger than the period of the wave. As a result, the tip of the
electric field vector at the point P describes a polarization ellipse whose character-
istic parameters vary slowly with time; these parameters will also show a smooth
variation from point to point on the surface Σ.
Obviously, it is impossible to define an instantaneous polarization ellipse in this
situation. It is however possible to define appropriate average quantities by gener-
alizing Eqs. (1.15) in the form
PI = E1∗ (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
PQ = E1∗ (P, t) E1 (P, t) − E2∗ (P, t) E2 (P, t)
PU = E1∗ (P, t) E2 (P, t) + E2∗ (P, t) E1 (P, t)
PV = i E1∗ (P, t) E2 (P, t) − E2∗ (P, t) E1 (P, t) , (1.24)
where the symbol · · · means an average over a time interval much longer than
the wave period and an average over the surface Σ.
Similarly, one can generalize the definition of the polarization tensor by writing
Jij = Ei∗ (P, t) Ej (P, t) , (1.25)
so that Eqs. (1.16) and (1.20) are still valid.
The important difference from the case of the monochromatic wave lies in the
fact that Eqs. (1.10) and (1.21) no longer are valid. To prove this statement, let
us consider the complex quantity A defined by
A = E2 (P, t) E1∗ (P, t) E1 (P, t) − E1 (P, t) E1∗ (P, t) E2 (P, t) .
10 CHAPTER 1
For the average value of its square modulus we have, with easy transformations
|A|2 = E1∗ (P, t) E1 (P, t) E1∗ (P, t) E1 (P, t) E2∗ (P, t) E2 (P, t)
− E1∗ (P, t) E2 (P, t) E2∗ (P, t) E1 (P, t) .
From this equation, excluding the trivial case E1 (P, t) ≡ 0, we conclude that, being
∗ 2
E1 (P, t) E1 (P, t) > 0 and |A| ≥ 0 ,
we must have
∗
E1 (P, t) E1 (P, t) E2∗ (P, t) E2 (P, t) − E1∗ (P, t) E2 (P, t) E2∗ (P, t) E1 (P, t) ≥ 0 ,
that is
det J = PI2 − PQ2 − PU2 − PV2 ≥ 0 . (1.26)
The equal sign holds only in the case A ≡ 0, which implies
∗
E2 (P, t) E1 (P, t) E2 (P, t)
= ∗ ;
E1 (P, t) E1 (P, t) E1 (P, t)
this means either that one of the two components is identically zero, or that the
two components have, at any point P and any time t, the same amplitude ratio and
phase difference. In both cases the quasi-monochromatic wave is said to be totally
polarized. The opposite situation occurs when the two components E1 (P, t) and
E2 (P, t) have the same average amplitude and random phase difference, so that
E1∗ (P, t) E1 (P, t) = E2∗ (P, t) E2 (P, t)
∗
E1 (P, t) E2 (P, t) = E2∗ (P, t) E1 (P, t) = 0 .
We have in this case
PQ = PU = PV = 0 ,
and the quasi-monochromatic wave is said to be totally unpolarized (if the frequency
of the radiation beam is in the visible band of the spectrum, the beam is said to be
composed of natural light ). In intermediate cases the quasi-monochromatic wave
is said to be partially polarized with a polarization degree given by
p= PQ2 + PU2 + PV2 PI ,
with 0 ≤ p ≤ 1.
DESCRIPTION OF POLARIZED RADIATION 11
Fig.1.5. For a partially polarized quasi-monochromatic wave, the tip of the electric field vector
at a given point P describes, as a function of time, the trajectory shown in this graph. Note that
the characteristic shape of the polarization ellipse has a slow variation with time. The electric
field is given in arbitrary units.
In Figs. 1.5 and 1.6 the trajectory of the tip of the electric field vector is plotted
for two schematic situations corresponding to a partially polarized and a totally
polarized quasi-monochromatic wave, respectively.
1.5. Polarizers and Retarders
In the previous sections we have described the polarization properties of a radiation
beam using appropriate averages of the electric vibrations. Now we must give some
operational definitions capable of relating these properties to actual measurements
to be performed on the beam. For this purpose it is necessary to introduce the
concept of ideal polarizing filters, a concept that will be used in the following to
give an operational definition of the Stokes parameters.
We define an ideal linear polarizer (sometimes called analyzer ) as a device that
is totally transparent to the electric vibration along a given axis (called the trans-
mission – or acceptance – axis of the polarizer) and totally opaque to the electric
vibration along the axis perpendicular to the former. When such a device is in-
troduced in the optical path of a radiation beam, the components of the electric
vibration are modified inside the polarizer according to the equation
Ea iψ 1 0 Ea iψ Ea
=e =e ,
Eb 0 0 Eb 0
12 CHAPTER 1
Fig.1.6. Same as Fig.1.5 for a totally polarized quasi-monochromatic wave. Although the size of
the ellipse changes with time, its characteristic shape (i.e. the α angle defined in Fig.1.2 and the
ratio between the axes) remains fixed.
where Ea and Eb are the components, at the entrance of the polarizer, along the
transmission axis and the orthogonal axis, respectively; Ea and Eb are the same
components at the exit of the polarizer, and ψ is a phase difference that is totally
inessential in determining the polarization properties of the beam.
The simplest device whose performances approach in practice those of an ideal
polarizer (in the optical range of the electromagnetic spectrum) is the well-known
Polaroid which has been developed commercially in various forms by the Polaroid
Corporation, Cambridge, Massachusetts, U.S.A. This device is based on the prop-
erty shown by some organic, needle-shaped microcrystals of absorbing light prefer-
entially along the longest axis of the crystal. Roughly speaking, a Polaroid consists
of a large number of such microcrystals oriented through a procedure of unidirec-
tional stretching. Since the basic physical phenomenon involved in the operation
of a Polaroid is dichroism (the property of absorbing light to different extents de-
pending on the polarization of the incident beam), the Polaroid is often referred to
as a dichroic polarizer .
Linear polarizers of a different kind are the so-called birefringence polarizers
which consist of one or more birefringent crystals (generally calcite) suitably pre-
pared and arranged into a compact system. These polarizers operate according to
the following scheme: the incident beam is divided in two orthogonally polarized
components that propagate along different directions in the birefringent crystal;
one of the two components is physically removed by absorption or lateral deflec-
tion, so that a single, linearly polarized component is transmitted through the
whole system. Examples of birefringence polarizers are those bearing the names of
DESCRIPTION OF POLARIZED RADIATION 13
Ahrens, Wollaston, Rochon, Glan-Foucault, and Nicol. More details on birefrin-
gent polarizers, as well as on dichroic polarizers, can be found in Shurcliff (1962).
In the same book different kinds of linear polarizers, like reflection polarizers and
scattering polarizers, are also described.
Turning now our attention to retarders, we define an ideal retarder as a device
which divides the incident beam in two orthogonally polarized components, de-
phases one of them relative to the other without altering their amplitudes, and,
finally, reunites the two components giving rise to a beam whose polarization char-
acteristics are, in general, completely different from those of the incident beam.
An ideal retarder (sometimes called compensator ) is characterized by the fact that
the phase velocity of an electromagnetic wave propagating along the optical axis
depends on the direction of the electric vibration. For an electric vibration parallel
to the fast axis the retarder is characterized by the index of refraction nf , while
for an electric vibration parallel to the slow axis the index of refraction is ns , with
n s > nf .
Let us consider a monochromatic wave that propagates along the optical axis of
a retarder of given thickness l. The electric field carried by the wave is described
in complex notations by the equations
i(nf kz−ωt)
Ef (z, t) = Re Ef e
(0 ≤ z ≤ l)
i(ns kz−ωt)
Es (z, t) = Re Es e ,
where z is the coordinate along the optical axis of the retarder with the origin
(z = 0) at the entrance, Ef and Es are the complex amplitudes of the electric
vibration along the fast and slow axes, respectively, and k is the wavenumber in
vacuum.
Comparing the electric vibrations at the exit of the retarder with those at the
entrance, we have with obvious notations
Ef iψ 1 0 Ef iψ Ef
= e = e iδ , (1.27)
Es
iδ
0 e Es e Es
where ψ is a phase that does not affect the polarization properties of the wave,
while δ, the so-called retardance (or retardation), is given by
δ = (ns − nf ) kl = 2π (ns − nf ) l λ , (1.28)
λ being the wavelength in vacuum. When the thickness of the retarder is such that
δ = π/2, the retarder is called a quarter-wave plate, and, similarly, when δ = π the
retarder is called a half-wave plate.
It is interesting to note that an ideal quarter-wave plate followed by an ideal linear
polarizer whose transmission axis is directed at 45◦ from the fast axis in the coun-
terclockwise direction (as seen by an observer facing the radiation source) behaves
14 CHAPTER 1
fast axis
direction of
propagation
slow
axis
transmission
axis
quarter-wave linear
plate polarizer
Fig.1.7. Practical realization of a filter transparent to positive circular polarization and opaque
to negative circular polarization.
like an ideal filter that is totally transparent to positive (or right-handed) circular
polarization and totally opaque to negative (or left-handed) circular polarization.1
To prove this statement, let us consider a circularly polarized monochromatic
wave propagating along the direction shown in Fig. 1.7. If the wave is totally
circularly polarized in the positive direction, the components of the electric field
vibration along the unit vectors ex and ey before entering the retarder are (in
complex notations)
−iωt
Ex = Re A e
−iωt
Ey = Re −iA e , (1.29)
where A is a constant. At the exit of the retarder, we have from Eq. (1.27) (apart
from an inessential phase factor)
−iωt
Ex = Ey = Re A e , (1.30)
thus the component of the electric vibration along the transmission axis of the
polarizer is given by
1 √ −iωt
Ea = √ Ex + Ey = 2 Re A e .
2
Since this component is simply transmitted through the linear polarizer, the elec-
tric vibration at the exit is still described by Eq. (1.30). Therefore, although the
1
Obviously, when the transmission axis is inclined at 45 ◦ in the opposite direction, the optical
combination behaves in the opposite way with respect to circular polarization.
DESCRIPTION OF POLARIZED RADIATION 15
polarization has been transformed from circular to linear, the energy carried by
the wave has not been changed by the system (the overbar means a time average)
Ex2 + Ey2 = Ex2 + Ey2 = |A|2 .
It follows that the device schematized in Fig. 1.7 is totally transparent to positive
circular polarization.
Applying the same argument to a monochromatic wave of negative circular po-
larization (Ey in Eq. (1.29) must be changed in sign) we obtain that the wave is
totally absorbed inside the polarizer.
Most of the retarders that are used in practice are of the birefringent type; they
often consist of a crystal, like quartz or calcite, cut parallel to the optical axis. More
recently piezo-optical birefringence modulators have also been used. In these devices
the birefringence is induced by compressing or distending a material (otherwise
isotropic and uniform) along a given direction. According to Kemp (1969) a very
modest force, that can even be attained by simple finger pressure, is enough to
produce a quarter-wave plate by stressing a block of glass having the size of a
match-box.
Other devices that have been often used to produce retarders with variable or
modulated retardance are the so-called Pockels cells. Here the birefringence is
induced by application of high electric fields (of the order of several kV/cm) in ma-
terials such as ammonium dihydrogen phosphate (ADP) or potassium dihydrogen
phosphate (KDP).
1.6. Stokes Parameters
A simple operational definition of the Stokes parameters can be given in terms of a
set of ideal filters and an ideal detector capable of measuring, in absolute units, the
electromagnetic energy falling on its acceptance area. For this definition – or, more
precisely, for the definition of the two parameters connected with linear polariza-
tion – it is also necessary to fix a particular direction in the plane perpendicular to
the direction of propagation. Such direction will be referred to in the following as
the reference direction or the reference axis.
Given a quasi-monochromatic radiation beam having a small aperture dΩ, and
frequency contained in the interval (ν, ν + dν), we define the first Stokes parameter,
I, as the energy measured by our ideal detector per unit time and per unit cross-
sectional area. In formulae, if dW0 is the energy measured by the detector in the
time interval dt over the surface dS oriented perpendicularly to the beam, we have
dW0 = I dS dt dΩ dν .
This is just the usual definition of the radiation specific intensity, so that the first
Stokes parameter is simply called the intensity of the beam.
Let’s now insert in the beam path an ideal linear polarizer with its transmission
axis directed in succession at 0◦ , 45◦ , 90◦ , and 135◦ from the reference axis, all the
16 CHAPTER 1
angles being reckoned counterclockwise for an observer looking at the beam from
the detector. If dW1 , dW2 , dW3 , and dW4 are the corresponding energies measured
by the detector in the time interval dt over the surface dS, the second and third
Stokes parameters, Q and U , are defined by
dW1 − dW3 = Q dS dt dΩ dν
dW2 − dW4 = U dS dt dΩ dν .
Finally, for the definition of the fourth parameter we introduce two ideal filters,
the first being totally opaque to negative (left-handed) circular polarization and
totally transparent to positive (right-handed) circular polarization, and the second
behaving in the opposite way (such filters can be realized as shown in the former
section). If dW5 and dW6 are the energies measured by the detector (in the time
interval dt over the surface dS) with the first and second filters – respectively –
interposed, the definition of the fourth parameter, V , is the following
dW5 − dW6 = V dS dt dΩ dν .
The symbols used in this book to represent the Stokes parameters (I, Q, U , V )
were introduced for the first time by Walker (1954). In his original note, Stokes
(1852) used the symbols (A, B, C, D), while Jones (1941) and Perrin (1942) used the
notation (I, M, C, S). Walker’s notation seems nowadays to be preferred especially
in the astrophysical literature.
The operational definitions given above are not universally adopted. In partic-
ular, the opposite sign is used in the definition of V by those authors who name
positive circular polarization the one that we have defined as negative and vice
versa (see the discussion at the end of Sect. 1.2). Less frequently a sign inversion is
found in the definition of U . Figure 1.8 summarizes the conventions that are used
in this book.
Now that the Stokes parameters have been defined, it is necessary to establish
their connection with the description, given in the previous sections, of the polar-
ization properties of an electromagnetic wave in terms of electric fields. Since the
energy flux carried by an electromagnetic wave is given by its Poynting vector, the
energy ∆W measured in the time interval ∆t by an ideal detector having cross-
sectional area ∆S is connected with the electric field components by the relation
c 2
∆W = Ex + Ey2 ∆S ∆t ,
4π
where Ex and Ey are the components of the electric field along two axes perpen-
dicular to each other and to the direction of propagation, and where the brackets
mean an average over the time interval ∆t and over the surface ∆S of the detector.
Using Eqs. (1.3), (1.11), and (1.13), we can also write
c 2 c ∗
∆W = E1 + E22 ∆S ∆t = E1 E1 + E2∗ E2 ∆S ∆t . (1.31)
8π 8π
DESCRIPTION OF POLARIZED RADIATION 17
reference
direction
Fig.1.8. Schematic representation of the definition of the Stokes parameters. The observer is
supposed to face the radiation source.
Referring to the general case of a quasi-monochromatic wave, we introduce a couple
of orthogonal unit vectors ea and eb , with ea specifying the reference direction and
eb oriented in such a way that ea , eb , and the direction of propagation form a right-
handed coordinate system.1 If E1 (P, t) and E2 (P, t) are the complex amplitudes
of the electric field along the ea and eb directions, the energy measured by the
detector in the various configurations specified above is given by (see Eqs. (1.23)):
a) without any filter
∗ c
∆W0 = E1 (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t) ∆S ∆t ;
8π
b) with a linear polarizer interposed, having its transmission axis at 0◦ , 45◦ , 90◦ ,
and 135◦ , respectively
c
∆W1 = E1∗ (P, t) E1 (P, t) ∆S ∆t
8π
1 ∗
∆W2 = E1 (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
2
c
+ E1∗ (P, t) E2 (P, t) + E2∗ (P, t) E1 (P, t) ∆S ∆t
8π
c
∆W3 = E2∗ (P, t) E2 (P, t) ∆S ∆t
8π
1
In the following, ea will be referred to as the reference direction unit vector, while eb will
be referred to as the associated unit vector. Note that the definition of the Stokes parameters
remains the same if we replace ea by −ea and eb by −eb .
18 CHAPTER 1
1 ∗
∆W4 = E1 (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
2
c
− E1∗ (P, t) E2 (P, t) − E2∗ (P, t) E1 (P, t) ∆S ∆t ;
8π
c) with a filter for circular polarization interposed, opaque to negative polarization
(and transparent to positive), and vice versa
1 ∗
∆W5 = E1 (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
2
c
+ i E1∗ (P, t) E2 (P, t) − i E2∗ (P, t) E1 (P, t) ∆S ∆t
8π
1 ∗
∆W6 = E1 (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
2
c
− i E1∗ (P, t) E2 (P, t) + i E2∗ (P, t) E1 (P, t) ∆S ∆t .
8π
From these expressions, taking the limit for infinitesimal values of ∆S and ∆t, and
bearing in mind Eqs. (1.24) and the operational definitions of the Stokes parame-
ters, we obtain
I = kPI = k E1∗ (P, t) E1 (P, t) + E2∗ (P, t) E2 (P, t)
Q = kPQ = k E1∗ (P, t) E1 (P, t) − E2∗ (P, t) E2 (P, t)
U = kPU = k E1∗ (P, t) E2 (P, t) + E2∗ (P, t) E1 (P, t)
V = kPV = k i E1∗ (P, t) E2 (P, t) − E2∗ (P, t) E1 (P, t) , (1.32)
where k is a dimensional positive constant whose actual value is important only
when absolute measurements are to be performed.
It should be kept in mind that Eqs. (1.32) are valid only when the component E1
of the electric vibration refers to the reference axis defining the Stokes parameters,
and when the conventions implicit in Eqs. (1.11) and (1.13) are used. Note also
that Eqs. (1.32) give a deeper physical meaning to the quantities PI , PQ , PU , PV
introduced in former sections (see Eqs. (1.7), (1.15), and (1.24)).
1.7. Measurements of the Stokes Parameters
The Stokes parameters can be measured by several different techniques, and the
various instruments that have been devised for this purpose cannot be classified in
a unique scheme. Following Hauge (1976) it is however possible to describe a kind
of prototype instrument that summarizes the essential characteristics of most of
DESCRIPTION OF POLARIZED RADIATION 19
reference
retarder polarizer
direction
α detector
β
transmission
axis
fast axis
Fig.1.9. Schematic representation of a prototype instrument aimed at the measurement of the
Stokes parameters.
the devices that are used in practice, especially for the measurement of the Stokes
parameters in the visible or near-infrared range of the electromagnetic spectrum.
Referring to Fig. 1.9, if E1 and E2 are (in complex notations) the electric vibrations
along the unit vectors ea and eb (where ea is the reference direction), we have, after
the beam has crossed the retarder (inessential phase factors are omitted)
Ef = E1 cos α + E2 sin α
iδ
Es = e (−E1 sin α + E2 cos α) ,
where Ef and Es are the components of the electric vibrations along the fast and
slow axes, respectively, and δ is the retardance. The only component transmitted
by the polarizer is the one parallel to the transmission axis. This component is
given by
Ep = cos(β − α) Ef + sin(β − α) Es ,
so that for the signal D of the detector we obtain after some algebra
1
D(α, β, δ) = k Ep∗ Ep = I + (Q cos 2α + U sin 2α) cos 2(β − α)
2
− (Q sin 2α − U cos 2α) sin 2(β − α) cos δ
+ V sin 2(β − α) sin δ . (1.33)
A first method for measuring the Stokes parameters is to set the various devices
at particular angles. If we are concerned with linear polarization only, the retarder
can be removed (δ = 0◦ ), or, alternatively, its fast axis can be set parallel to the
transmission axis of the polarizer (α = β). In both cases the signal D of the
detector takes the simpler form
1
D (β) = I + Q cos 2β + U sin 2β ,
2
20 CHAPTER 1
and the Stokes parameters I, Q, and U can be found by the combination of four
measurements
I = D (0◦ ) + D (90◦ )
Q = D (0◦ ) − D (90◦ )
U = D (45◦ ) − D (135◦ ) ,
or, alternatively, with only three measurements
2 ◦
I= D (0 ) + D (60◦ ) + D (120◦ )
3
2
Q= 2D (0◦ ) − D (60◦ ) − D (120◦ )
3
2
U = √ D (60◦ ) − D (120◦ ) .
3
For the measurement of V the retarder is essential. Setting a quarter-wave plate
(δ = 90◦ ) at α = 0◦ we get from Eq. (1.33)
1
D (β) = I + Q cos 2β + V sin 2β ,
2
and the Stokes parameters I, Q, and V are found by
I = D (0◦ ) + D (90◦ )
Q = D (0◦ ) − D (90◦ )
V = D (45◦ ) − D (135◦ ) .
This technique, however, has the disadvantage of implying a rotation of the
exit polarizer, which is often a dangerous procedure especially if the polarization
analysis is to be followed by a spectral analysis (a grating spectrometer, for instance,
has a response that is rather sensitive to the polarization of the incident radiation).
To avoid these problems one can employ, for the measurement of linear polarization,
a half-wave plate (δ = 180◦ ) and keep the polarizer fixed at β = 0◦ . The signal is
now
1
D (α) = I + Q cos 4α + U sin 4α ,
2
and the Stokes parameters I, Q, and U can be obtained by the relations
I = D (0◦ ) + D (45◦ )
Q = D (0◦ ) − D (45◦ )
U = D (22.5◦ ) − D (67.5◦ ) .
DESCRIPTION OF POLARIZED RADIATION 21
For circular polarization, keeping the polarizer fixed at β = 0◦ , one gets (again
using a quarter-wave plate)
1
D (α) = I + Q cos2 2α + U sin 2α cos 2α − V sin 2α ,
2
and the Stokes parameters I, Q, U , and V can be found by the combination of five
measurements
I = D (45◦ ) + D (135◦ )
Q = 2 D (0◦ ) − D (45◦ ) − D (135◦ )
U = 2 D (22.5◦ ) − D (67.5◦ )
V = D (135◦ ) − D (45◦ ) .
A different method for measuring the Stokes parameters is based on rotating
devices. Using a quarter-wave plate rotating with angular frequency ω and setting
β = 0◦ , we obtain for the detector signal D(t) the expression
1 Q Q U
D(t) = I+ + cos 4ωt + sin 4ωt − V sin 2ωt ,
2 2 2 2
and the various Stokes parameters can be measured (for a stationary source) by
extracting the Fourier components of the signal D(t).
A third method is to use a compensator with modulated retardance (like a Pockels
cell) while keeping the optical components at fixed orientations. To maximize
the modulation effect it is convenient to set (β − α) = ±45◦ , and two sets of
measurements are necessary to get the four Stokes parameters. For instance, setting
α = 0◦ and β = 45◦ we have
1
D(t) = I + U cos δ(t) + V sin δ(t) ,
2
whereas setting α = −45◦ , β = 0◦ we get
1
D(t) = I + Q cos δ(t) + V sin δ(t) ,
2
and in both cases three of the four Stokes parameters can be obtained by an analysis
of the modulated signal.
We want to remark that the measurement of Stokes parameters is an art in itself
that can be pushed to a very high degree of sophistication. The few words that
have been spent here on this subject are not meant to be complete, but just to give
the reader an intuitive grasp on how this kind of measurements are performed in
practice.
22 CHAPTER 1
1.8. Stokes Parameters and Polarization Tensor
In Sect. 1.6 we have found the relation between the Stokes parameters and the
components of the electric vibration along two orthogonal unit vectors, ea and
eb , with ea the reference direction unit vector and eb the associated unit vector.
Equations (1.32) can be rewritten in terms of the polarization tensor defined in
Sects. 1.3 and 1.4. From Eq. (1.25) we have
I = k J11 + J22
Q = k J11 − J22
U = k J12 + J21
V = k i J12 − J21 . (1.34)
Introducing the so-called Stokes vector Si (i = 0, . . . , 3) via the equation
⎛ ⎞ ⎛ ⎞
S0 I
⎜ S1 ⎟ ⎜ Q ⎟
⎝ ⎠=⎝ ⎠
S2 U
S3 V
we have, in strict analogy with Eqs. (1.16) and (1.20)
1
3
J= S σ
2k j=0 j j
Sj = k Tr σj J (j = 0, . . . , 3) , (1.35)
where σj are the matrices defined in Eqs. (1.17).
The preceding relations between the Stokes vector and the polarization tensor
involve the dimensional quantity k introduced in Eqs. (1.32). This is a consequence
of the fact that the Stokes parameters have the dimensions of an intensity, while
the polarization tensor has the dimensions of a squared electric field. To simplify
the theoretical treatment of polarized radiation it is often convenient to leave out
the constant k, by defining a ‘new’ polarization tensor I having the dimensions of
intensity
I = kJ , (1.36)
where J is the tensor introduced in Sects. 1.3-1.4 and k is the quantity appearing
in Eqs. (1.32). In the following we will use the term ‘polarization tensor’, without
distinction, for both J and I. Obviously, the tensor I obeys the relations (cf.
Eqs. (1.14) and (1.26))
∗
Iij = Iji , (1.37)
and
det I ≥ 0 . (1.38)
DESCRIPTION OF POLARIZED RADIATION 23
In terms of I, Eqs. (1.34) and (1.35) read, respectively
I = I11 + I22
Q = I11 − I22
U = I12 + I21
V = i I12 − I21 (1.39)
and
1
3
I= S σ
2 j=0 j j
Sj = Tr σj I (j = 0, . . . , 3) . (1.40)
If the unit vectors chosen to define the polarization tensor differ from (ea , eb ), the
relations between the Stokes parameters and the components Iij are not as simple
as in Eqs. (1.39). Let us consider, in particular, the (complex) unit vectors e+ and
e− defined by
iφ
e+ = cos θ ea + sin θ e eb
iφ
e− = − sin θ ea + cos θ e eb , (1.41)
with
e+∗ · e+ = e−∗ · e− = 1
e+∗ · e− = e−∗ · e+ = 0 .
Denoting by E+ and E− the controvariant components of the electric vibration
along the unit vectors e+ and e− , defined in such a way that E = E1ea + E2eb =
E+e+ + E−e− , we have
E1 = cos θ E+ − sin θ E−
iφ
E2 = e sin θ E+ + cos θ E− ,
and simple relations can be established between the ‘old’ polarization tensor Iij =
k Ei∗ Ej (i, j = 1, 2) and the ‘new’ polarization tensor Iαβ = k Eα∗ Eβ (α, β =
+, −)
I11 = cos2 θ I++ + sin2 θ I−− − sin θ cos θ I+− + I−+
I22 = sin2 θ I++ + cos2 θ I−− + sin θ cos θ I+− + I−+
∗ iφ
I12 = I21 = e sin θ cos θ I++ − I−− + cos2 θ I+− − sin2 θ I−+ ,
24 CHAPTER 1
and conversely
−iφ iφ
I++ = cos2 θ I11 + sin2 θ I22 + sin θ cos θ e I12 + e I21
−iφ iφ
I−− = sin2 θ I11 + cos2 θ I22 − sin θ cos θ e I12 + e I21
∗
−iφ iφ
I+− = I−+ = − sin θ cos θ I11 − I22 + cos2 θ e I12 − sin2 θ e I21 .
Using these transformations we can express the Stokes parameters in the form
I = I++ + I−−
Q = cos 2θ I++ − I−− − sin 2θ I+− + I−+
U = sin 2θ cos φ I++ − I−− + cos 2θ cos φ I+− + I−+
+ i sin φ I+− − I−+
V = − sin 2θ sin φ I++ − I−− − cos 2θ sin φ I+− + I−+
+ i cos φ I+− − I−+ , (1.42)
with the inverse relations
I++ + I−− = I
I++ − I−− = cos 2θ Q + sin 2θ cos φ U − sin 2θ sin φ V
I+− + I−+ = − sin 2θ Q + cos 2θ cos φ U − cos 2θ sin φ V
i I+− − I−+ = sin φ U + cos φ V . (1.43)
These formulae show that it is possible to find a particular couple of complex unit
vectors of the form (1.41) such that the polarization tensor is diagonal. In fact, it
can be easily shown that for φ and θ implicitly defined by
tan φ = −V /U , tan 2θ = V 2 + U2 Q, (1.44)
the components of the polarization tensor satisfy the relations
I++ + I−− = I
I++ − I−− = ± Q2 + U 2 + V 2
I+− = I−+ = 0 ,
where the sign ambiguity in the right-hand side of the middle equation is connected
with the determination chosen for the φ and θ angles in Eqs. (1.44). On this basis
of complex unit vectors the electric vibration is decomposed in two independent
orthogonal components.
DESCRIPTION OF POLARIZED RADIATION 25
old reference
direction
new reference
direction
Fig.1.10. The rotation of the reference direction is specified by the angle α. The radiation beam
is propagating from behind the page to the reader’s eye.
1.9. Properties of the Stokes Parameters
Since the Stokes parameters are defined with respect to a given reference direction,
we must establish how they are changed when a different reference direction is
chosen. Referring to Fig. 1.10, and denoting by (I, Q, U, V ) and (I , Q , U , V ) the
Stokes parameters relative to the old and new reference direction, respectively, the
transformation law can be established with the help of the equations derived in the
previous section. Substituting α for θ and 0 for φ into Eqs. (1.41), and interpreting
the unit vectors e+ and e− as ea and eb , respectively, we obtain from Eqs. (1.43)
and (1.39)
I = I
Q = cos 2α Q + sin 2α U
U = − sin 2α Q + cos 2α U
V =V . (1.45)
These transformations show that I and V are invariant under rotation of the refer-
ence direction, while the linear polarization parameters undergo a rotation through
an angle 2αin the Q-U plane, which leaves unchanged the total linear polariza-
tion PL = Q2 + U 2 . It is also apparent that the reference direction has a 180◦
ambiguity, an obvious consequence of the definition of the Stokes parameters.
Introducing the position angle α0 , defined by
Q = PL cos 2α0
U = PL sin 2α0 , (1.46)
the former equations for Q and U become
Q = PL cos 2(α0 − α)
U = PL sin 2(α0 − α) .
26 CHAPTER 1
The two quantities PL and α0 are often used to characterize the linear polariza-
tion in place of Q and U . In astronomical observations the reference direction is
generally taken along the meridian through the observed object, and the position
angle is measured counterclockwise from the same meridian. In solar observations
the reference direction is sometimes chosen to represent a physical direction, like,
for instance, the radial direction from the center of the sun to the observed point,
or the parallel to the solar limb.
Another fundamental property of the Stokes parameters is their additivity for
incoherent beams. This property can be proved quite easily from Eqs. (1.32).
Indeed, if the electric vibrations of two beams are described (in complex notations)
by Ei(1) (P, t) and Ei(2) (P, t), respectively, and if the two beams are incoherent, so
that
(1)
Ei (P, t)∗ Ej(2) (P, t) = Ei(2) (P, t)∗ Ej(1) (P, t) = 0 (i, j = 1, 2) ,
we obtain for the Stokes parameters of the composite beam
Sk = Sk(1) + Sk(2) (k = 0, . . . , 3) ,
where Sk(1) and Sk(2) are the Stokes parameters of the separate beams. This property,
which can obviously be generalized to an arbitrary number of incoherent beams,
will be referred to in the following as the addition theorem.
As the Stokes parameters of a radiation beam satisfy the relation (see Eqs. (1.38)
and (1.40))
I 2 ≥ Q2 + U 2 + V 2 , (1.47)
we can write, using the addition theorem
⎛ ⎞ ⎛ ⎞ ⎛ 2 ⎞
I I − Q2 + U 2 + V 2 Q + U2 + V 2
⎜Q⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 0 ⎟ ⎜ Q ⎟
⎜ ⎟=⎜ ⎟+⎜ ⎟ .
⎝U ⎠ ⎝ 0 ⎠ ⎝ U ⎠
V 0 V
This means that any radiation beam can be considered as the incoherent super-
position of an unpolarized beam and a totally polarized beam. The latter is char-
acterized by a well-defined polarization ellipse, whose elements can be found from
Eqs. (1.8) and (1.9). The former, on the contrary, is composed of natural radiation.
Another interesting feature of the Stokes parameters stems from the possibility
of establishing a mapping between the Stokes vectors and the points of a three-
dimensional space that will be referred to as the Poincaré space. To the Stokes vec-
tor S = (I, Q, U, V )† we associate the point P having coordinates (Q/I, U/I, V /I),
as shown in Fig. 1.11. When the point P lies on the surface of the sphere of unit
radius (the Poincaré sphere), the corresponding Stokes vector represents a totally
polarized radiation beam, while the center of the sphere represents a beam of natu-
ral radiation. This mapping, proposed by Poincaré (1892) in a classical monograph,
is particularly suitable to visualize the effect produced on a light beam by a given
DESCRIPTION OF POLARIZED RADIATION 27
Fig.1.11. Mapping of a Stokes vector (I, Q, U, V ) into the representative point P inside the
Poincaré sphere of unit radius.
device (like a polarizer or a retarder) as a stepwise movement of the representa-
tive point on the sphere. In Chap. 5 we will see how it is possible to visualize
the transfer of polarized radiation as a continuous movement of the representative
point inside the Poincaré sphere.
1.10. Photons and Stokes Parameters
As we have seen in the former sections, the operational definition of the Stokes
parameters of a radiation beam involves at least four independent measurements
performed by interposing polarizers or retarders in the beam path. On the other
hand, according to the principles of Quantum Mechanics, a photon which interacts
with a polarizer is either absorbed or transmitted; in the latter case its state is in
general modified, the photon polarization being now parallel to the transmission
axis of the polarizer. All information about the original polarization state of the
photon is in any case lost, so that it is impossible to perform the four independent
measurements needed to determine its Stokes parameters. We conclude that to
speak about the Stokes parameters of a single photon has no physical meaning.
Stokes parameters can only be measured for a statistical ensemble of photons.
The connection between the measured Stokes parameters and the photon wave-
function can be easily established. Referring to Fig. 1.7, let us suppose that the
i-th photon has a wavefunction of the form
(i) (i)
|ψ (i) = c1 |ψ1 + c2 |ψ2 ,
where |ψ1 (|ψ2 ) is the wavefunction in the polarization space of a photon which
propagates along the z-axis and is transmitted through a polarizer having its trans-
mission axis in the ex (ey ) direction. Applying to the photon beam the same argu-
ments developed for wave packets in Sect. 1.6, we can express the Stokes parameters
28 CHAPTER 1
as
I = k c∗1 c1 + c∗2 c2
Q = k c∗1 c1 − c∗2 c2
U = k c∗1 c2 + c∗2 c1
V = k i c∗1 c2 − c∗2 c1 ,
where k is a dimensional constant and where the statistical averages c∗k cj are
defined by
1 (i) ∗ (i)
N
c∗k cj = c cj (k, j = 1, 2) ,
N i=1 k
N being the number of photons falling on the measuring device in a given time
interval ∆t.
(i)
Only when all the photons are in the same state, so that the amplitudes cj are
independent of i, it is possible to determine their wavefunction from the measured
Stokes parameters.1 This case, characterized by the equality
I 2 = Q2 + U 2 + V 2 ,
is the analogue of the macroscopic case of the monochromatic wave. On the con-
trary, when the states of the single photons are different from each other, we have
the more complicated situation described by Eq. (1.47), that has its macroscopic
analogue in the quasi-monochromatic wave. From the quantum-mechanical point
of view, the photons are said to be in a pure state in the former case and in a
mixture of states in the latter.
1
Apart from an inessential phase factor. According to the principles of Quantum Mechanics,
absolute phases of wavefunctions are not observable quantities.
CHAPTER 2
ANGULAR MOMENTUM AND RACAH ALGEBRA
This chapter, and the two following, are devoted to establish the quantum-mechan-
ical bases that will be needed to approach the theoretical problems involved in the
generation and transfer of polarized radiation. In particular, we will recall in the
present chapter some important concepts about angular momentum and Racah
algebra and will introduce a set of notations and definitions that will be useful in
the following.
These arguments have been treated in various classical textbooks or monographs
(see e.g. Edmonds, 1957; Messiah, 1961; Brink and Satchler, 1968; Varshalovich
et al., 1988). The reason why they are repeated here is to spare the reader the
painful job of translating notations and conventions from one book to the other.
Obviously, the reader who is already familiar with the theory of angular momentum
can simply skip this chapter.
2.1. Eigenvalues and Eigenvectors of Angular Momentum
Angular momentum is defined in Quantum Mechanics as a Hermitian vector op-
whose components along the axes (xyz) of a right-handed coordinate
erator, J,
system satisfy the following commutation rules1
[Jx , Jy ] = iJz
[Jy , Jz ] = iJx
[Jz , Jx ] = iJy . (2.1)
From these equations it can be easily shown that the square of the angular mo-
mentum,
J 2 = Jx2 + Jy2 + Jz2 ,
commutes with each of the three components
[J 2 , Jx ] = [J 2 , Jy ] = [J 2 , Jz ] = 0 .
Taking J 2 and one of the three components – say Jz – as a maximum set of com-
muting operators, we can look for the common eigenvectors of these two operators
that we denote by the symbol |λm
J 2 |λm = λ |λm , Jz |λm = m |λm .
1
The operator J considered here is the angular momentum expressed in units of the reduced
Planck constant h̄ = h/(2π).
30 CHAPTER 2
The Hermitian character of the operators Jx , Jy , and Jz , together with the com-
mutation rules (2.1), are sufficient to deduce the eigenvalue spectrum or, in other
words, the allowed values for λ and m.
Let us introduce the operators J+ and J− , usually called the shift operators,
defined by
J± = Jx ± iJy .
Directly from the definition it is possible to prove the following relations
(J± )† = J∓ (2.2a)
[J 2 , J± ] = 0 (2.2b)
[Jz , J± ] = ±J± (2.2c)
[J+ , J− ] = 2Jz (2.2d)
J+ J− = J 2 − Jz2 + Jz (2.2e)
J− J+ = J 2 − Jz2 − Jz . (2.2f)
Using these relations it can be proved that the vectors J± |λm are still eigenvectors
of J 2 and Jz , corresponding to the eigenvalues λ and (m ± 1), respectively. From
Eqs. (2.2b) and (2.2c) we have
J 2 [J± |λm ] = J± J 2 |λm = λ [J± |λm ]
Jz [J± |λm ] = [J± Jz ± J± ] |λm = (m ± 1) [J± |λm ] . (2.3)
On the other hand, if we consider the norms of the vectors J+ |λm and J− |λm
(that cannot be negative numbers), and take into account Eqs. (2.2a), (2.2e),
and (2.2f), we have
†
λm|J+ J+ |λm = λm|J− J+ |λm = (λ − m2 − m) λm|λm ≥ 0
†
λm|J− J− |λm = λm|J+ J− |λm = (λ − m2 + m) λm|λm ≥ 0 , (2.4)
so that
λ − m2 − m ≥ 0 , λ − m2 + m ≥ 0 .
Since λ and m are real numbers (being eigenvalues of Hermitian operators), these
two inequalities give
λ≥0, |m| ≤ λ . (2.5)
Therefore, once we fix a particular eigenvalue λ for J 2 , the eigenvalues of Jz must
be confined within a limited interval.
Starting now from a given eigenvector |λm and applying repeatedly the shift
operator J+ (or J− , respectively), we obtain two distinct chains of eigenvectors
corresponding to eigenvalues of Jz increasing (or decreasing) by unity as shown
in Eq. (2.3). These two chains have to stop at a certain point so that the second
ANGULAR MOMENTUM AND RACAH ALGEBRA 31
inequality in Eq. (2.5) can be satisfied. Thus there must be a maximum eigenvalue
for Jz – say j – such that
J+ |λj = 0 .
Equations (2.4) give for this eigenvector
λ − j2 − j = 0 , (2.6)
so that
λ = j(j + 1) .
Similarly, there must be a minimum eigenvalue for Jz – say j – such that
J− |λj = 0 .
From Eqs. (2.4) we obtain for this eigenvector
λ − j 2 + j = 0 ,
and comparing this equation with Eq. (2.6) we get the following relation between
j and j
j(j + 1) = j (j − 1) .
This is a second degree equation whose solutions are
j = −j
j = j + 1 .
The second solution must be rejected as it leads to j > j. We are then left with the
first solution which implies that the minimum eigenvalue of Jz is just the opposite
of the maximum eigenvalue.
On the other hand, we can obtain the eigenvector |λ −j from the eigenvector
|λ j by applying to the latter a finite number of times the operator J− ; this implies
that the number [j − (−j)] = 2j is an integer, so that the only allowed values for j
are integers or half-integers; moreover, the number of eigenvectors associated with
a given eigenvalue of J 2 is (2j + 1).
From now on, the eigenvectors of J 2 and Jz will be denoted by the symbol |jm ;
they are supposed to be normalized to unity and to satisfy the eigenvalue equations
J 2 |jm = j(j + 1) |jm
Jz |jm = m |jm (m = −j, −j + 1, . . . , j) . (2.7)
Note also that the eigenvectors |jm obey the orthogonality relation
jm|j m = δjj δmm , (2.8)
since J 2 and Jz are Hermitian operators.
32 CHAPTER 2
The phase relations among the various eigenvectors characterized by the same
j-value and different m-values must be established using suitable conventions. From
the properties of the shift operators we have
J+ |jm = α |j m+1 ,
where
|α|2 = j(j + 1) − m(m + 1) .
The phase of the α constant is conventionally set equal to zero, so that
J+ |jm = j(j + 1) − m(m + 1) |j m+1
= (j + m + 1)(j − m) |j m+1 , (2.9)
and, similarly
J− |jm = j(j + 1) − m(m − 1) |j m−1
= (j − m + 1)(j + m) |j m−1 . (2.10)
With these phase convention, it is easily found from Eq. (2.10) that any eigenvector
|jm is related to the ‘parent’ eigenvector |jj by the equation
(j + m)! j−m
|jm = J− |jj . (2.11)
(2j)! (j − m)!
2.2. Coupling of Two Angular Momenta: Vector-Coupling
Coefficients and 3-j Symbols
A fundamental problem encountered in Quantum Mechanics is the addition of the
angular momenta relative to two independent particles, or to two different degrees
of freedom (like orbital motion and spin) of the same particle. If J1 and J2 are the
angular momenta of the two separate systems, with
[J1i , J2j ] = 0 (i, j = x, y, z) ,
the angular momentum eigenvectors (normalized to unity) of the total system can
be written as
|j1 j2 m1 m2 ,
each eigenvector being a dyadic product of the form |j1 m1 |j2 m2 that satisfies
the eigenvalue equations
J12 |j1 j2 m1 m2 = j1 (j1 + 1) |j1 j2 m1 m2
J1z |j1 j2 m1 m2 = m1 |j1 j2 m1 m1
J22 |j1 j2 m1 m2 = j2 (j2 + 1) |j1 j2 m1 m2
J2z |j1 j2 m1 m2 = m2 |j1 j2 m1 m2 .
ANGULAR MOMENTUM AND RACAH ALGEBRA 33
On the other hand, the total system can also be described, from the point of
view of its angular momentum properties, in terms of a different set of operators,
namely J12 , J22 , J 2 , Jz , where
J = J1 + J2 .
It is easily seen that these four operators commute among each other; as a conse-
quence, we can introduce a new set of normalized eigenvectors of the form |j1 j2 JM
such that
J12 |j1 j2 JM = j1 (j1 + 1) |j1 j2 JM
J22 |j1 j2 JM = j2 (j2 + 1) |j1 j2 JM
J 2 |j1 j2 JM = J(J + 1) |j1 j2 JM
Jz |j1 j2 JM = M |j1 j2 JM .
Since the two different sets of eigenvectors span the same Hilbert space, they must
be connected by a unitary similarity transformation of the form
|j1 j2 JM = |j1 j2 m1 m2 j1 j2 m1 m2 |j1 j2 JM
m1 m2
|j1 j2 m1 m2 = |j1 j2 JM j1 j2 JM |j1 j2 m1 m2 . (2.12)
JM
The coefficients appearing in these expressions are called vector-coupling coeffi-
cients, or Wigner coefficients, or Clebsh-Gordan coefficients. In shorthand nota-
tions they are often denoted by the symbol j1 j2 m1 m2 |JM , as the repetition of
the arguments j1 and j2 in the ket is unnecessary.
To deduce the expression for these coefficients we must first of all establish suit-
able phase conventions. The phase relations among the eigenvectors of the form
|j1 j2 m1 m2 are the same as those of the former section, so that we can write (see
Eq. (2.11))
(j1 + m1 )! (j2 + m2 )! j −m j2 −m2
|j1 j2 m1 m2 = J 1 1 J2− |j1 j2 j1 j2 ,
(2j1 )! (2j2 )! (j1 − m1 )! (j2 − m2 )! 1−
and the same convention of the former section applies also to the phase relations
among the eigenvectors of the form |JM corresponding to a fixed J-value. The
remaining phase conventions will be established in the following.
Let us consider the eigenvector of the form |j1 j2 m1 m2 corresponding to the
maximum m-values (m1 = j1 , m2 = j2 ). We have
Jz |j1 j2 j1 j2 = [J1z + J2z ] |j1 j2 j1 j2 = (j1 + j2 )|j1 j2 j1 j2
J 2 |j1 j2 j1 j2 = [J12 + J22 + 2J1z J2z + J1+ J2− + J1− J2+ ] |j1 j2 j1 j2
= [j1 (j1 + 1) + j2 (j2 + 1) + 2j1 j2 ] |j1 j2 j1 j2
= (j1 + j2 )(j1 + j2 + 1) |j1 j2 j1 j2 .
34 CHAPTER 2
These equations show that the eigenvector |j1 j2 j1 j2 is also an eigenvector of the
form |JM , with J = M = (j1 + j2 ). By convention we take the same phase for
the two eigenvectors, so that
|j1 +j2 j1 +j2 = |j1 j2 j1 j2 .
The above value of J is the maximum of the possible J-values, as it can be argued
from the fact that the maximum eigenvalue of Jz is (j1 + j2 ). The eigenvectors
with the same J-value and lower M -values can be found by repeated application
of the operator J− , so that (see Eq. (2.11))
(j1 + j2 + M )!
|j1 +j2 M = [J + J2− ]j1 +j2 −M |j1 j2 j1 j2 .
(2j1 + 2j2 )! (j1 + j2 − M )! 1−
The eigenvectors of the form |JM with J < (j1 + j2 ) can be found using a set
of operators, On , that will be called the supershift operators, defined by
n
n−r n
On = (−1)n−r J1+
r
J2+ n
J1− J2− .
r=0
Let us consider the application of the operator On to the vector |j1 j2 j1 j2 . Using
Eqs. (2.9) and (2.10) we have, after some algebra
(n! )2 (2j1 )! (2j2 )!
On |j1 j2 j1 j2 =
(2j1 − n)! (2j2 − n)!
n
(2j1 − n + r)! (2j2 − r)!
× (−1)n−r |j1 j2 j1 −n+r j2 −r . (2.13)
r=0
r! (n − r)!
Applying to this vector the operator Jz we have
Jz On |j1 j2 j1 j2 = (j1 + j2 − n) On |j1 j2 j1 j2 , (2.14)
and applying the operator J 2 , after some heavy algebra that is left to the reader
as an exercise, we have
J 2 On |j1 j2 j1 j2 = (j1 + j2 − n)(j1 + j2 − n + 1) On |j1 j2 j1 j2 . (2.15)
Equations (2.14) and (2.15) show that the vector On |j1 j2 j1 j2 is, apart from a
normalization factor and a phase factor, a vector of the form |JJ with J =
(j1 + j2 − n). Note that n cannot be larger than the smaller of the values 2j1 and
2j2 (otherwise On |j1 j2 j1 j2 = 0), so that the allowed values for J must satisfy the
triangular condition
|j1 − j2 | ≤ J ≤ j1 + j2 .
ANGULAR MOMENTUM AND RACAH ALGEBRA 35
The norm Nn of the vector On |j1 j2 j1 j2 can be easily obtained from Eq. (2.13)
(n! )4 (2j1 )! (2j2 )! (2j1 − n + r)! (2j2 − r)!
n
Nn = .
[(2j1 − n)! (2j2 − n)! ]2 r=0 r! (n − r)!
The summation above can be evaluated using some properties of the binomial
coefficients (see App. 1 of Edmonds, 1957, for a formal proof)
n
(2j1 − n + r)! (2j2 − r)! (2j1 − n)! (2j2 − n)! (2j1 + 2j2 − n + 1)!
= ,
r=0
r! (n − r)! n! (2j1 + 2j2 − 2n + 1)!
and hence
(n! )3 (2j1 )! (2j2 )! (2j1 + 2j2 − n + 1)!
Nn = . (2.16)
(2j1 − n)! (2j2 − n)! (2j1 + 2j2 − 2n + 1)!
From Eqs. (2.13) and (2.16) we can express any vector |JJ (normalized to unity)
as a function of the vectors of the form |j1 j2 m1 m2 . Apart from a phase factor we
have
(j1 + j2 − J)! (2J + 1)!
|JJ =
(j1 − j2 + J)! (j2 − j1 + J)! (j1 + j2 + J + 1)!
j1 +j2 −J
(j1 − j2 + J + r)! (2j2 − r)!
× (−1)j1 +j2 −J−r
r=0
r! (j1 + j2 − J − r)!
× |j1 j2 J −j2 +r j2 −r . (2.17)
The phase convention that will be used in this book is that of assuming Eq. (2.17)
(with no phase factor in front) valid for any value of J. This convention agrees with
those given by Racah (1942), Edmonds (1957), Messiah (1961), Brink and Satch-
ler (1968), Varshalovich et al. (1988), and, together with the other conventions
established previously, leads to vector-coupling coefficients that are all real .
Now that the phase conventions have been fully established, we can turn to the
evaluation of the vector-coupling coefficients. From Eq. (2.11) we have
(J + M )!
|JM = J J−M |JJ . (2.18)
(2J)! (J − M )! −
Writing
J−M
(J − M )!
J−M
J− = [J1− + J2− ]J−M = J k J J−M−k
k! (J − M − k)! 1− 2−
k=0
36 CHAPTER 2
and introducing Eq. (2.17) into (2.18) we obtain
(j1 + j2 − J)! (J + M )! (J − M )! (2J + 1)
|JM =
(j1 − j2 + J)! (j2 − j1 + J)! (j1 + j2 + J + 1)!
−J
j1 +j
J−M 2
(j1 − j2 + J + r)! (2j2 − r)!
× (−1)j1 +j2 −J−r
r=0
k! (J − M − k)! r! (j1 + j2 − J − r)!
k=0
(j1 + j2 − J + k − r)! (J − M − k + r)!
×
(j1 − j2 + J − k + r)! (2j2 − J + M + k − r)!
× |j1 j2 J −j2 −k+r j2 −J +M +k−r .
From this expression the vector-coupling coefficients are easily calculated. Taking
the scalar product with j1 j2 m1 m2 |, all the terms in the right-hand side are zero
except those satisfying the conditions
J − j2 − k + r = m1
j2 − J + M + k − r = m2 .
Adding these two equations we find
M = m1 + m2 ,
and we can eliminate the sum over k by substituting
k = J − j2 − m1 + r = J − j2 − M + m2 + r .
Thus we obtain
j1 j2 m1 m2 |JM = δm +m2 ,M
1
(2J + 1) (j1 + j2 − J)! (J + M )! (J − M )! (j1 − m1 )! (j2 − m2 )!
×
(j1 − j2 + J)! (j2 − j1 + J)! (j1 + j2 + J + 1)! (j1 + m1 )! (j2 + m2 )!
j1 +j2 −J
(j1 − j2 + J + r)! (2j2 − r)!
× (−1)j1 +j2 −J−r .
r=0
(J − j2 − m1 + r)! (j2 − m2 − r)! (j1 + j2 − J − r)! r!
This expression can be transformed into a more symmetrical one (Racah, 1942)
j1 j2 m1 m2 |JM = δm +m2 ,M
1
(2J + 1)(j1 + j2 − J)! (j1 − j2 + J)! (−j1 + j2 + J)!
×
(j1 + j2 + J + 1)!
× (j1 + m1 )! (j1 − m1 )! (j2 + m2 )! (j2 − m2 )! (J + M )! (J − M )!
× (−1)t t! (j1 + j2 − J − t)! (j1 − m1 − t)! (j2 + m2 − t)!
t
−1
× (J − j2 + m1 + t)! (J − j1 − m2 + t)! , (2.19)
ANGULAR MOMENTUM AND RACAH ALGEBRA 37
Fig.2.1. The addition of two vectors having fixed projections on a given axis leads, in classical
physics, to a continuous distribution for the modulus of the resultant vector.
where the sum over t runs over all the values leading to non-negative factorials.
The vector-coupling coefficients have a very simple physical meaning. Given two
independent systems of angular momenta j1 and j2 , suppose that the z-axis pro-
jections of these momenta have been measured and found equal to m1 and m2 ,
respectively. The square modulus | j1 j2 m1 m2 |JM |2 represents the probability
of finding the values J and M , respectively, when measuring the angular momen-
tum of the total system and its z-projection. The Kronecker symbol in Eq. (2.19)
implies that the measured M -value must be equal to (m1 + m2 ), which is the same
result found for classical (non quantum-mechanical) angular momenta. In the clas-
sical case, however, the total angular momentum J has a continuous distribution of
values that depend on the relative orientation of the vectors j1 and j2 (see Fig. 2.1).
The vector-coupling coefficients can be regarded as the quantum-mechanical ana-
logue of this distribution; but the distribution is now discrete, which reflects the
quantum nature of the angular momenta.
The vector-coupling coefficients satisfy a large number of properties. Multiplying
Eqs. (2.12) by j1 j2 J M | and j1 j2 m1 m2 |, respectively, taking Eq. (2.8) into
account, and recalling that the vector-coupling coefficients are real, one gets the
orthogonality relations
j1 j2 m1 m2 |JM j1 j2 m1 m2 |J M = δJJ δMM (2.20a)
m1 m2
j1 j2 m1 m2 |JM j1 j2 m1 m2 |JM = δm m1 δm m2 . (2.20b)
1 2
JM
† †
The identity j1 j2 m1 m2 |J+ |JM = j1 j2 m1 m2 |J1− + J2− |JM leads to the re-
38 CHAPTER 2
cursion relation
J(J + 1) − M (M + 1) j1 j2 m1 m2 |J M +1 =
= j1 (j1 + 1) − m1 (m1 − 1) j1 j2 m1 −1 m2 |JM
+ j2 (j2 + 1) − m2 (m2 − 1) j1 j2 m1 m2 −1|JM . (2.21)
The other properties of the vector-coupling coefficients are more easily expressed
in terms of the 3-j symbols, which are defined by
√ j1 j2 J
j1 j2 m1 m2 |J −M = (−1)j1 −j2 −M 2J + 1 . (2.22)
m1 m2 M
Nowadays the 3-j coefficients are employed more often than the vector-coupling
coefficients. We just summarize below a set of relations that will be needed later.
Formal proofs can be easily obtained with the help of Eqs. (2.19), (2.20), and (2.22).
a) Orthogonality relations
a b c a b c
(2c + 1) = δcc δγγ (2.23a)
α β γ α β γ
αβ
a b c a b c
(2c + 1) = δαα δββ . (2.23b)
α β γ α β γ
cγ
b) The 3-j symbols are invariant under cyclic permutations of their columns and
are multiplied by (−1)a+b+c under non-cyclic ones
a b c b c a c b a
= = (−1)a+b+c , etc. (2.24)
α β γ β γ α γ β α
c) They are multiplied by (−1)a+b+c under sign inversion of the second row
a b c a b c
= (−1)a+b+c . (2.25)
α β γ −α −β −γ
d) By assigning particular values to some of the arguments, compact analytical
expressions can be easily obtained; we have for instance
a b 0 1
= (−1)a−α δab δα,−β √ (2.26a)
α β 0 2a + 1
a a + 12 1 a+α+1
2 = (−1)1−a+α (2.26b)
α −α − 12 12 (2a + 2)(2a + 1)
a a 1 (a − α)(a + α + 1)
= (−1)a−α (2.26c)
α −α − 1 1 2a(a + 1)(2a + 1)
ANGULAR MOMENTUM AND RACAH ALGEBRA 39
a a 1 α
= (−1)a−α (2.26d)
α −α 0 a(a + 1)(2a + 1)
a a+1 1 (a + α + 1)(a + α + 2)
= (−1)a−α (2.26e)
α −α − 1 1 (2a + 1)(2a + 2)(2a + 3)
a a+1 1 (a − α + 1)(a + α + 1)
= (−1)a−α−1 . (2.26f)
α −α 0 (a + 1)(2a + 1)(2a + 3)
Analytical expressions for the 3-j coefficients with one of the three elements of the
first row set at 3/2 or 2 can be found in Brink and Satchler (1968); analogous
formulae with one j-value equal to 5/2 are given by Saito and Morita (1955), and
with one j-value equal to 3 by Falkoff et al. (1952). These analytical expressions,
together with the various numerical tables of 3-j symbols (or vector-coupling coef-
ficients) prepared by several authors (Alder, 1952; Simon, 1954; Simon et al., 1954;
Rotenberg et al., 1959; Varshalovich et al., 1988) have nowadays lost some of their
interest due to the fact that, with modern computers, the direct computation of
3-j symbols via Eqs. (2.19) and (2.22) can be easily performed. A sample Fortran
code is given in App. 1.
2.3. Coupling of Three Angular Momenta: Racah Coefficients
and 6-j Symbols
In the previous section we have shown how it is possible to couple two angular
momenta J1 and J2 to get the resultant vector J. When three angular momenta
are present, say J1 , J2 , and J3 , one can follow the same line of reasoning, by first
adding two angular momenta and then adding the third one to the sum of the first
two to obtain, as a final result, the total angular momentum J.
In this procedure we are however faced with some ambiguity. In fact, we could
start from the couple (J1 , J2 ) to obtain J12 = J1 + J2 , and then add J12 to J3 to
obtain J. Alternatively, we could start from the couple (J , J ) to obtain J =
2 3 23
J2 + J3 , and then add J23 to J1 to obtain again the resultant J.
Finally, we could
introduce the intermediate vector J13 = J1 + J3 and then add J13 to J2 .
These different coupling schemes are related to the fact that, given three com-
muting angular momentum operators J1 , J2 , and J3 , three different sets of six
commuting operators can be considered, namely
I) J12 , J22 , J32 , J12
2
, J 2 , Jz
II) J12 , J22 , J32 , J23
2
, J 2 , Jz
III) J12 , J22 , J32 , J13
2
, J 2 , Jz ,
besides the standard set
IV) J12 , J22 , J32 , J1z , J2z , J3z .
40 CHAPTER 2
Expressing the eigenvectors of sets I, II, and III in terms of the eigenvectors of
set IV, it is possible to find the transformations connecting the different bases. The
coefficients entering these transformations are called the recoupling coefficients and
can be derived as follows.
Denoting the eigenvectors of set I by the symbol
|(j1 j2 )j12 , j3 , JM , (2.27)
we apply twice the first of Eqs. (2.12) to obtain
|(j1 j2 )j12 , j3 , JM = |j12 j3 m12 m3 j12 j3 m12 m3 |j12 j3 JM
m12 m3
= |j1 j2 j3 m1 m2 m3
m1 m2 m3 m12
× j1 j2 m1 m2 |j1 j2 j12 m12 j12 j3 m12 m3 |j12 j3 JM , (2.28)
where we have used the symbol
|j1 j2 j3 m1 m2 m3 = |j1 m1 |j2 m2 |j3 m3
to represent the eigenvectors of set IV.
Similarly, for a given eigenvector of set II of the form |j1 , (j2 j3 )j23 , J M we
obtain
|j1 , (j2 j3 )j23 , J M =
= |j1 j2 j3 m1 m2 m3
m1 m2 m3 m23
× j2 j3 m2 m3 |j2 j3 j23 m23 j1 j23 m1 m23 |j1 j23 J M . (2.29)
If we now take the scalar product of the two vectors in Eqs. (2.28) and (2.29) we
obtain, using the shorthand notation of the vector-coupling coefficients
(j1 j2 )j12 , j3 , JM |j1 , (j2 j3 )j23 , J M =
= j1 j2 m1 m2 |j12 m12 j12 j3 m12 m3 |JM
m1 m2 m3 m12 m23
× j2 j3 m2 m3 |j23 m23 j1 j23 m1 m23 |J M δJJ δMM . (2.30)
The scalar product in the left-hand side is independent of the M -value, as it can
be easily proved by evaluation of the matrix element
(j1 j2 )j12 , j3 , JM |J− J+ |j1 , (j2 j3 )j23 , JM
ANGULAR MOMENTUM AND RACAH ALGEBRA 41
Fig.2.2. When adding in classical physics three angular momenta j1 , j2 , j3 to give the resultant
J,
once the sum of two of them (j1 and j2 ) is specified, the modulus of (j2 +j3 ) shows a continuous
distribution.
via Eqs. (2.9), (2.2a), and (2.2f). On the other hand, owing to the properties of the
vector-coupling coefficients, the summation in the right-hand side can be restricted
to m1 , m2 , and m3 , with the condition (m1 + m2 + m3 ) = M .
The scalar product now derived is called the recoupling coefficient and has a
simple physical meaning. Given three vectors j1 , j2 , and j3 , that combine to give
the resultant J, the square modulus of the recoupling coefficient is the probability
that, if by measuring (j1 +j2 )2 we found the value j12 (j12 +1), then, when measuring
(j2 +j3 )2 , we find the value j23 (j23 +1). As illustrated in Fig. 2.2, the modulus of the
vector j23 is undetermined in classical physics too; it has a continuous distribution
of probabilities which originates from purely geometrical reasons. The quantum
nature of the angular momenta adds to this ‘geometrical indetermination’ to give
a discrete distribution for the probabilities.
Dropping the inessential parameter M , the recoupling coefficients can be used to
express the transformations between different bases in the form
|j1 , (j2 j3 )j23 , JM = |(j1 j2 )j12 , j3 , JM
j12
× (j1 j2 )j12 , j3 , J|j1 , (j2 j3 )j23 , J . (2.31)
In place of the recoupling coefficients it is customary to introduce different symbols
like the Racah coefficients (or Racah functions) defined by
W (j1 j2 Jj3 ; j12 j23 ) =
1
= (j1 j2 )j12 , j3 , J|j1 , (j2 j3 )j23 , J , (2.32)
(2j12 + 1)(2j23 + 1)
42 CHAPTER 2
or the more symmetrical 6-j symbols
!
j1 j2 j12
= (−1)j1 +j2 +j3 +J W (j1 j2 Jj3 ; j12 j23 ) . (2.33)
j3 J j23
Starting from Eq. (2.30), and using Eqs. (2.22), (2.32), and (2.33), it is possible
to express the 6-j symbol as a sum of products of four 3-j symbols
!
a b c
δee δ = (−1)b+c+d+β+γ+δ (2e + 1)
d e f
αβγδφ
c d e d b f b c a f a e
× . (2.34)
γ −δ δ −β φ β −γ α φ α
It should be remarked that, owing to the symmetry properties (2.24) and (2.25),
the right-hand side can be written in several equivalent ways, obtained by column
permutations or by sign change of the second row of each 3-j symbol. Since for
each 3-j symbol there are six permutations of the columns and two determinations
for the sign of the second row, 124 different realizations exist for the right-hand side
of Eq. (2.34), apart from the ordering of the 3-j coefficients. All these realizations
have however the same ‘topological invariant’ which can be visualized by graphical
methods. The theory of graphical methods for angular momentum problems can
be found in Edmonds (1957) or in Brink and Satchler (1968).
The 6-j symbol has several important properties. First of all, it is zero unless four
triangular conditions are satisfied. These conditions can be illustrated as follows
! ! ! !
◦ ◦ ◦ ◦ ◦ ◦
, , , .
◦ ◦ ◦ ◦ ◦ ◦
Another remarkable property is the invariance of the 6-j symbol both under inter-
change of any two columns and under interchange of the upper and lower arguments
in any two columns.
An analytical expression for the 6-j symbol can be obtained from Eq. (2.30) by
substituting the series expansion (2.19) of the vector-coupling coefficients. After a
long calculation, Racah (1942) gives the following expression
!
a b c
= ∆(abc) ∆(aef ) ∆(dbf ) ∆(dec) (−1)z (z + 1)!
d e f
z
× (z − a − b − c)! (z − a − e − f )! (z − d − b − f )! (z − d − e − c)!
−1
×(a + b + d + e − z)! (b + c + e + f − z)! (a + c + d + f − z)! , (2.35)
where
(a + b − c)! (a − b + c)! (−a + b + c)!
∆(abc) =
(a + b + c + 1)!
ANGULAR MOMENTUM AND RACAH ALGEBRA 43
and where the sum is extended to all the values of z leading to non-negative facto-
rials.
By specifying the value of one of the arguments, simple analytical formulae can
be obtained, e.g.:
!
a b 0 1
= δab δed (−1)a+e+f (2.36a)
d e f (2a + 1)(2d + 1)
!
a a+ 1 1 1 (a + b + c + 2)(a + b − c + 1)
1
2 2 = (−1)a+b+c+1 (2.36b)
b+ 2 b c 2 (a + 1)(2a + 1)(b + 1)(2b + 1)
!
a a+ 1 1 1 (a − b + c + 1)(c − a + b)
2 2 = (−1)a+b+c (2.36c)
b− 1
2 b c 2 (a + 1)(2a + 1)b(2b + 1)
!
a a 1 1 a(a + 1) + b(b + 1) − c(c + 1)
= (−1)a+b+c+1 (2.36d)
b b c 2 a(a + 1)(2a + 1)b(b + 1)(2b + 1)
!
a a+1 1
= (−1)a+b+c
b+1 b c
1 (a + b + c + 3)(a + b + c + 2)(a + b − c + 2)(a + b − c + 1)
× (2.36e)
2 (a + 1)(2a + 1)(2a + 3)(b + 1)(2b + 1)(2b + 3)
!
a a+1 1
= (−1)a+b+c+1
b b c
1 (a + b + c + 2)(a − b + c + 1)(a + b − c + 1)(c − a + b)
× (2.36f)
2 (a + 1)(2a + 1)(2a + 3)b(b + 1)(2b + 1)
!
a a+1 1
= (−1)a+b+c
b−1 b c
1 (c − a + b)(c − a + b − 1)(a − b + c + 2)(a − b + c + 1)
× (2.36g)
2 (a + 1)(2a + 1)(2a + 3)(2b − 1)b(2b + 1)
!
a a 2
= (−1)a+b+c
b b c
3 s(s + 1) − 43 a(a + 1)b(b + 1)
× (2.36h)
2 (2a − 1)a(a + 1)(2a + 1)(2a + 3)(2b − 1)b(b + 1)(2b + 1)(2b + 3)
where
s = c(c + 1) − a(a + 1) − b(b + 1) .
44 CHAPTER 2
Further analytical formulae have been given by Biedenharn et al. (1952), while
numerical tables have been prepared by Biedenharn (1952), Sharp et al. (1954),
Simon et al. (1954), Rotenberg et al. (1959), and Varshalovich et al. (1988). As
noted in the previous section, these tabulations have nowadays lost much of their
interest since modern computers provide the possibility of computing 6-j symbols
with simple numerical codes like that in App. 1.
Other important properties of the 6-j symbols, that can be proved either from
the analytical expression (2.35) or directly from their definition, are given below.
Sum rules:
!
a b c
(−1)2c (2c + 1) =1 (2.37)
a b f
c
!
a b c
(−1)a+b+c (2c + 1) = δf 0 (2a + 1)(2b + 1) (2.38)
b a f
c
! !
a b c a b c
(2c + 1)(2f + 1) = δf g (2.39)
d e f d e g
c
! ! !
a b c a b c a e f
(−1)f +g+c (2c + 1) = (2.40)
d e f e d g b d g
c
! ! !
a+b+c+d+e+f +g+h+i+j a b c a b c g h c
(−1) (2c + 1) =
d e f g h i e d j
c
! !
f i j f i j
= (2.41)
g d b h e a
Contraction of 3-j symbols:
c d e d b f b c a
(−1)b+c+d+β+γ+δ =
γ −δ δ −β φ β −γ α
βγδ
!
f a e f a e
= (2.42)
φ α c d b
!
a b e c a f b d f
(−1)a+b+c+d−e+f −α−δ (2f + 1) =
d c f γ α φ β δ −φ
f
a b e d c e
= . (2.43)
α β − δ γ
ANGULAR MOMENTUM AND RACAH ALGEBRA 45
2.4. Coupling of Four Angular Momenta: 9-j Symbols
When four angular momenta are present, say J1 , J2 , J3 , and J4 , different coupling
schemes can be followed, as in the case of three angular momenta discussed in the
previous section. For instance, one can start from the couples (J1 , J2 ) and (J3 , J4 )
to obtain the intermediate vectors J12 = J1 + J2 and J34 = J3 + J4 , and then add
J12 to J34 to obtain the resultant J. Alternatively, one can start from the couples
(J1 , J3 ) and (J2 , J4 ) to get the intermediate vectors J13 = J1 + J3 and J24 = J2 + J4 ,
and then add J13 to J24 to obtain again the resultant J. A third possibility would
be to introduce the intermediate vectors J14 = J1 + J4 and J23 = J2 + J3 .
These different coupling schemes are related to the fact that, given four indepen-
dent angular momenta, three different sets of eight commuting operators can be
constructed, namely
I) J12 , J22 , J32 , J42 , J12
2 2
, J34 , J 2 , Jz
II) J12 , J22 , J32 , J42 , J13
2 2
, J24 , J 2 , Jz
III) J12 , J22 , J32 , J42 , J14
2 2
, J23 , J 2 , Jz ,
besides the standard set
IV) J12 , J22 , J32 , J42 , J1z , J2z , J3z , J4z .
Denoting the eigenvectors of set I by the symbol
|(j1 j2 )j12 , (j3 j4 )j34 , JM
and those of set II by the symbol
|(j1 j3 )j13 , (j2 j4 )j24 , JM ,
the two bases are connected by the transformation
|(j1 j2 )j12 , (j3 j4 )j34 , JM = |(j1 j3 )j13 , (j2 j4 )j24 , JM
j13 j24
× (j1 j3 )j13 , (j2 j4 )j24 , JM |(j1 j2 )j12 , (j3 j4 )j34 , JM .
The recoupling coefficients defined in this equation are independent of M (see the
comments about Eq. (2.30)). In their place different symbols are usually intro-
duced, like the 9-j symbols defined by
(j1 j3 )j13 , (j2 j4 )j24 , J|(j1 j2 )j12 , (j3 j4 )j34 , J =
⎧ ⎫
⎨ j1 j3 j13 ⎬
= (2j12 + 1)(2j34 + 1)(2j13 + 1)(2j24 + 1) j j4 j24 . (2.44)
⎩ 2 ⎭
j12 j34 J
46 CHAPTER 2
The 9-j symbol can be easily expressed in terms of 3-j coefficients by means of a
procedure similar to that of the previous section for the 6-j coefficients (Eqs. (2.27)-
(2.30)). The final result is the compact formula
⎧ ⎫
⎨a b c⎬ a b c d e f g h i
d e f = (2a + 1)
⎩ ⎭ α β γ δ φ χ η ξ
g h i
a d g b e h c f i
× ,
α δ χ β η γ φ ξ
where the sum runs over all the Greek symbols except α, namely over β, γ, δ, ,
φ, χ, η, ξ.
The 9-j symbol can be also expressed in terms of 6-j symbols. We start from
the vector |(ad)g, (be)h, i and neglect the fact that the vector h is obtained by
combining b with e. From Eq. (2.31) we have
|(ad)g, (be)h, i = a, (dh)k, i|(ad)g, h, i |a, (dh)k, i . (2.45)
k
Recalling that h is the resultant of b and e, we recouple the three angular momenta
d, b, e to obtain
|a, (dh)k, i = b, (de)f, k|d, (be)h, k |a, (bf )k, i . (2.46)
f
Finally we recouple the three vectors a, b, f to get
|a, (bf )k, i = (ab)c, f, i|a, (bf )k, i |(ab)c, f, i . (2.47)
c
Substituting Eq. (2.47) into Eq. (2.46) and then into Eq. (2.45), and recalling that
the vector f is the combination of d and e, we obtain
|(ad)g, (be)h, i = a, (dh)k, i|(ad)g, h, i b, (de)f, k|d, (be)h, k
kf c
× (ab)c, f, i|a, (bf )k, i |(ab)c, (de)f, i .
The three recoupling coefficients can be expressed in terms of 6-j symbols. For this
purpose it is necessary to recall the definitions (2.32)-(2.33) and to observe that,
owing to the properties of the vector-coupling coefficients, the order of the various
angular momenta can be changed according to the following rules
|a, (bc)d, e = (−1)d−b−c |a, (cb)d, e
|a, (bc)d, e = (−1)e−a−d |(bc)d, a, e .
ANGULAR MOMENTUM AND RACAH ALGEBRA 47
Using these properties and the definition of the 9-j symbol given in Eq. (2.44), we
obtain the remarkable expression
⎧ ⎫
⎨a b c⎬ ! ! !
2k a i k b f k a i k
d e f = (−1) (2k +1) . (2.48)
⎩ ⎭ h d g d h e f b c
g h i k
The 9-j symbol satisfies several important properties. First of all, the symbol is
zero unless the six triangular conditions illustrated below are satisfied
⎧ ⎫ ⎧ ⎫
◦ ◦ ◦⎪
⎨◦
⎪ ◦ ◦⎪
⎬ ⎪
⎨ ⎬
◦ ◦ ◦ , ◦ ◦ ◦ .
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
◦ ◦ ◦ ◦ ◦ ◦
Moreover, it is invariant under reflection about either diagonal, and is multiplied
by (−1)t , t being the sum of its nine arguments, under interchange of two adjacent
rows or columns.
If one of the arguments is zero, the 9-j symbol reduces to a 6-j symbol times a
constant. This can be proved directly from Eq. (2.48), which gives
⎧ ⎫
⎨a b c⎬ !
1 a b c
d e f = δcf δgh (−1)b+c+d+g
. (2.49)
⎩ ⎭ (2c + 1)(2g + 1) e d g
g h 0
For the numerical evaluation of 9-j symbols Eq. (2.48) can be conveniently used,
and numerical tables are also available (Smith and Stevenson, 1957; Smith, 1958;
Varshalovich et al., 1988). However, as already mentioned about 3-j and 6-j sym-
bols, simple computer codes like that in App. 1 are nowadays generally preferred.
Other important properties of the 9-j symbols are given below.
Orthogonality:
⎧ ⎫⎧ ⎫
⎨a b c ⎬⎨ a b c⎬
(2c + 1)(2f + 1)(2g + 1)(2h + 1) d e f d e f = δgg δhh (2.50)
⎩ ⎭⎩ ⎭
cf g h i g h i
Sum rule:
⎧ ⎫⎧ ⎫ ⎧ ⎫
⎨a b c ⎬⎨ a d g ⎬ ⎨a b c⎬
(−1)2b+h+k−f (2g + 1)(2h + 1) d e f e b h = e d f (2.51)
⎩ ⎭⎩ ⎭ ⎩ ⎭
gh g h i j k i j k i
48 CHAPTER 2
Contraction of 3-j symbols:
⎧ ⎫
⎨a b c ⎬
a b c
d e f =
α β γ ⎩ ⎭
g h i
b e h c f i a d g
d e f
g h i
= (2.52)
β η γ φ ξ α δ χ δ φ χ η ξ
ηφξδχ
⎧ ⎫
⎨a b c ⎬
a b c b e h
(2b + 1) d e f =
α β γ β η ⎩ ⎭
b g h i
c f i a d g d e f g h i
= (2.53)
γ φ ξ α δ χ δ φ χ η ξ
φξδχ
⎧ ⎫
⎨a b c⎬
a b c b e h c f i
(2b + 1)(2c + 1) d e f =
α β γ β η γ φ ξ ⎩ ⎭
bc g h i
a d g d e f g h i
= (2.54)
α δ χ δ φ χ η ξ
Contraction of 6-j and 9-j symbols:
⎧ ⎫
⎨a b c ⎬ ! ! !
a b c a i k b f k
(2c + 1) d e f = (−1)2k (2.55)
⎩ ⎭ f i k h d g d h e
c g h i
(−1)s+c+d+f −b−g−h−l (2s + 1)(2t + 1)
st
⎧ ⎫
⎨a b c ⎬ ! ! !
c f i d e f b e h
× d t s =
⎩ ⎭ j k s j s t j l t
g l k
⎧ ⎫
⎨a b c ⎬ !
g h i
= d e f . (2.56)
⎩ ⎭ j k l
g h i
Finally, we want to remark that the methods illustrated so far for the coupling
of 2, 3, and 4 angular momentum operators can be directly generalized to any
number of operators. The addition of n angular momenta will involve recoupling
ANGULAR MOMENTUM AND RACAH ALGEBRA 49
coefficients depending on 3(n− 1) parameters. For n = 5 two different types of 12-j
symbols have been introduced (Jahn and Hope, 1954; Ord-Smith, 1954; Elliott and
Flowers, 1955). Fortunately we will not need such symbols in this book.
2.5. Rotations and Euler Angles
It is well-known that in Quantum Mechanics the state of a physical system is
described by a vector in the Hilbert space, while the dynamical variables of the
system are described by linear Hermitian operators. According to the postulates
of Quantum Mechanics, the result of any given measurement accomplished on the
system is expressed in terms of probabilities, which are calculated by taking the
square modulus of scalar products between appropriate vectors of the Hilbert space.
Similarly, the mean value of any given observable (its expectation value) is given by
the diagonal matrix element ψ|O|ψ , where |ψ is the state vector of the system
and O is the Hermitian operator associated with the measured observable.
When a rotation is performed in the ordinary three-dimensional space, two differ-
ent points of view can be followed: either a passive point of view which consists in
rotating the observer’s coordinate system leaving the physical system unchanged,
or an active point of view which consists in rotating the physical system leaving
the coordinate system unchanged.
Adopting the first point of view, the state vector |ψ of the physical system
remains unchanged, while the Hermitian operator O corresponding to a classical
observable changes into a different operator, O , connected with O by the same
transformation that holds for the corresponding classical observable (Correspon-
dence Principle). This can be performed by introducing a unitary similarity trans-
formation on the operators,
O = D(R) OD† (R) , (2.57)
where D(R) is an appropriate linear operator such that D† (R) = D−1 (R) and
D(R−1 ) = D−1 (R), where R−1 is the inverse rotation of R.
On the contrary, adopting the active point of view, the operators remain un-
changed, while the state vector |ψ changes into the new state vector |ψ given
by
|ψ = D(R)|ψ .
Obviously these transformations satisfy the property
ψ |O |ψ = ψ|O|ψ ,
which means that if the same rotation is performed both on the coordinate system
and on the physical system, the expectation value of any observable remains the
same.
Note that the formalism here introduced is consistent with the intuitive fact
that an arbitrary rotation R performed on the coordinate system is equivalent to
50 CHAPTER 2
positive
rotation
δ α
Fig.2.3. To find the variation δv of the components of the vector v under a positive rotation of
the coordinate system through an angle φ about the u-axis, we observe that this variation is the
same as that obtained by rotating the vector v through an angle −φ about the same axis. The
figure shows that for an infinitesimal rotation of the arbitrary vector w
through an infinitesimal
angle α we have δw = α u × w
, so that δv = − φ u × v.
the inverse rotation R−1 performed on the physical system. This can be formally
proved by considering the quantity
ψ|D(R) OD† (R)|ψ .
On one hand, this expression can be regarded as
ψ| D(R) OD† (R) |ψ = ψ|O |ψ ,
which represents the expectation value of the operator O after a rotation of the
coordinate system (passive point of view). On the other hand, the same expression
can be considered as
ψ|D(R) O D† (R)|ψ = ψ|D† (R−1 ) O D(R−1 )|ψ ,
which represents the expectation value of the same operator after the inverse rota-
tion of the physical system (active point of view).
The expression for the operator D(R) can be easily found in the following way.
We consider a physical system referred to a given coordinate system, and fix our
attention on a particular observable like the vector v (representing for instance the
position or the velocity or the angular momentum of a particle, etc.). We now
perform a positive rotation of the coordinate system through an infinitesimal angle
φ about the unit vector u, where by positive rotation we mean, here and in the
following, the one that makes a right-handed screw advance along the direction of
u. The vector v will be changed by the quantity (see Fig. 2.3)
δv = − φ u × v . (2.58)
ANGULAR MOMENTUM AND RACAH ALGEBRA 51
According to Eq. (2.57) the operator Du (φ) corresponding to this infinitesimal
rotation must therefore satisfy the relation
v = v + δv = Du (φ) v Du† (φ) , (2.59)
with δv given by Eq. (2.58).
Let’s now recall that the commutation relations between any dynamical variable
of vectorial character and the total angular momentum J are of the form1
[Jk , vl ] = i klm vm (k, l, m = x, y, z) . (2.60)
m
From Eqs. (2.59) and (2.60) it can be easily proved that
Du (φ) = 1 − i φJu , (2.61)
where Ju = J ·u is the projection of the total angular momentum on the u-direction
(note that Du† (φ) = Du−1 (φ) = Du (−φ)).
From Eq. (2.61) we can find the expression for the operator corresponding to the
rotation through a finite angle Ω. To this aim, we divide the angle Ω into a large
number n of equal parts, and perform the finite rotation as the succession of n
infinitesimal rotations of amplitude Ω/n. We have
Ω
n
n
n! (−iΩJu )r
Du (Ω) = lim 1 − i Ju = lim
n→∞ n n→∞
r=0
r! (n − r)! nr
∞
(−iΩJu )r n!
= lim ,
r! n→∞ (n − r)! nr
r=0
and since the limit gives 1, we obtain
∞
(−iΩJu )r −iΩJu
Du (Ω) = =e , (2.62)
r=0
r!
where, by definition, the exponential of an operator has the usual meaning given
by its Taylor expansion.
Having established the expression for the operator D corresponding to a finite
rotation about an arbitrary axis, we can easily find its expression for the most
1
For a system consisting of one particle, if r , p, l = h̄1 r × p, s, and
J = l + s are the
position, momentum, orbital angular momentum, spin, and total angular momentum operators,
respectively, Eq. (2.60), with v denoting any of these vectors as well as any linear combination of
them, can be easily deduced from the fundamental commutation rules
%
[ri , rj ] = [pi , pj ] = 0 , [ri , pj ] = ih̄ δij , [si , sj ] = i k
ijk sk .
The generalization to the case of many-particle systems is straightforward.
52 CHAPTER 2
α γ
β α
γ
Fig.2.4. Definition of the Euler angles.
general rotation of the reference system. But we must first introduce the Euler
angles, which are defined as follows.
Referring to Fig. 2.4, we first perform a rotation through an angle α (0 ≤ α < 2π)
about the z-axis of the original reference system. This rotation brings the y-axis
into a new axis, called the line of nodes, or w-axis. Next we perform a rotation
through an angle β (0 ≤ β < π) about the line of nodes. This rotation brings
the z-axis of the original system into a new axis, called the figure-axis, or z -axis.
Finally we perform a rotation through an angle γ (0 ≤ γ < 2π) about the figure-axis
to get the new system (x y z ).
For the general rotation specified by the three Euler angles (α, β, γ) we have
D(α, β, γ) = Dz (γ) Dw (β) Dz (α) , (2.63)
an expression which has the disadvantage that the rotations β and γ are relative to
axes that have been carried over by previous rotations. To overcome this drawback
we observe that1
Dw (β) = Dz (α) Dy (β) Dz (−α) (2.64)
1
Equation (2.64) is self-evident from geometrical considerations, since
Dw (β) Dz (α) = Dz (α) Dy (β) ;
however, it can also be proved analytically. From Eq. (2.62) we have
Dw (β) = e−iβJw ,
ANGULAR MOMENTUM AND RACAH ALGEBRA 53
and that
Dz (γ) = Dw (β) Dz (γ) Dw (−β) . (2.65)
Substituting Eqs. (2.65) and (2.64) into Eq. (2.63), and noting that two rotations
about the same axis commute, we obtain
−iαJz −iβJy −iγJz
D(α, β, γ) = Dz (α) Dy (β) Dz (γ) = e e e . (2.66)
This expression shows that the rotation described by Eq. (2.63) can also be achieved
by performing three successive rotations about the axes of the old reference system
in the following order: first a rotation through an angle γ about the z-axis, then
a rotation through an angle β about the y-axis, and finally a rotation through an
angle α about the z-axis. The former realization of the rotation, involving the
line of nodes and the figure-axis, is easier to visualize; the latter is more useful in
practice, since it involves rotation operators depending on the projections of the
angular momentum along the axes of a unique coordinate system.
2.6. Rotation Matrices
The matrix elements of the rotation operator D(α, β, γ) between eigenstates of the
total angular momentum are called rotation matrices. These quantities, which are
explicitly defined by the relation1
DMN
J
(R) = JM |D(R)|JN (2.67)
with the symbol R representing the triad (α, β, γ), are of fundamental importance
in angular momentum theory. Their physical meaning descends directly from that
of the rotation operator D(R) defined in the previous section. Given a physical
system of angular momentum J, the square modulus
|DMN
J
(R)|2
represents the probability of finding the value N for the z -axis projection of the
momentum after its z-axis projection has been measured and found equal to M ,
the system (x y z ) being obtained from (xyz) through the rotation R. Note that
where Jw is the operator obtained from Jy by a rotation through an angle α about the z-axis,
Jw = Dz (α) Jy Dz−1 (α) = Dz (α) Jy Dz (−α) .
Taking this expression into account we obtain, by a power series expansion of the exponential
−iβJy
Dw (β) = Dz (α) e Dz (−α) = Dz (α) Dy (β) Dz (−α) .
The same argument can be applied to deduce Eq. (2.65).
1
Note that the matrix elements JM |D(R)|J M , with J = J, identically vanish. This is
a consequence of the fact that the operator D(R) commutes with J 2 .
54 CHAPTER 2
if |JM is the eigenvector corresponding to the projection M of the angular mo-
mentum along the z-axis of the old reference system, the vector D(R)|JM is
the eigenvector corresponding to the same projection along the axis z of the new
reference system.
To find an explicit expression for the rotation matrices, we first define the reduced
rotation matrices dJMN . From Eq. (2.66) we have
−iαJz −iβJy −iγJz −i(αM+γN ) J
DMN
J
(αβγ) = JM |e e e |JN =e dMN (β) , (2.68)
where
−iβJy
dJMN (β) = JM |e |JN .
The reduced rotation matrices can be easily calculated in the case J = 1/2. In-
troducing the spinors | + and | − to represent the eigenstates |JM having
J = 1/2 and M equal to +1/2 and −1/2, respectively, we have from Eqs. (2.9)
and (2.10), taking into account that Jy = −i (J+ − J− )/2
i i
Jy | + = |− , Jy | − = − |+ ,
2 2
so that, with easy transformations
∞
−iβJy (−iβ)n n β β
e |+ = Jy | + = cos |+ + sin |− ,
n=0
n! 2 2
and similarly
−iβJy β β
e | − − sin | + .
|− = cos
2 2
Thus the reduced rotation matrices for J = 1/2 are the following
1 1 β
d 21 1 (β) = d−
2
1
−1
(β) = cos
2 2 2 2 2
1 1 β
1 1 (β) = −d 1
2 2
d− −1
(β) = sin .
2 2 2 2 2
Turning to the general case, let us consider an angular momentum eigenstate of
the form |JJ . This eigenstate can be expressed in terms of spinors by regarding
the angular momentum J as the result of the addition of 2J momenta J1 , J2 , . . . , J2J
each equal to 1/2. In this representation we have, with obvious notations
|JJ = | + 1| + 2 ···|+ 2J ,
where | + i is the normalized eigenvector of the operator Jiz corresponding to the
eigenvalue +1/2.
In the same representation, the eigenvector |JM can be obtained by repeated
application of the operator
J− = J1− + J2− + · · · + J2J− .
ANGULAR MOMENTUM AND RACAH ALGEBRA 55
Starting from the eigenvector |JJ we have from Eq. (2.11)
1 1
|J J −1 = √ J− |JJ = √ J1− + J2− + · · · + J2J− | + 1 | + 2 · · · | + 2J
2J 2J
1 &
= √ | − 1 | + 2 · · · | + 2J + | + 1 | − 2 · · · | + 2J + · · ·
2J
'
+ | + 1 | + 2 · · · | − 2J ,
and expressing in succession the eigenvectors |J J − 2 , |J J − 3 , . . . we finally
obtain
(J + M )! (J − M )! & '
|JM = | − | − ···|− | + | + ···|+ ,
(2J)! p
( )* +( )* +
(J−M)terms (J+M)terms
with the sum extended to all the permutations having (J − M ) spinors of the form
| − and (J + M ) spinors of the form | + .
Thus we can write
−iβJy (J + M )! (J − M )! (J + N )! (J − N )!
dMN (β) = JM |e
J
|JN =
(2J)!
& '
−iβ(J1y +J2y +···+J2Jy )
× − | − |··· − | + | + |··· + | e
p
( )* + ( )* +
(J−M)terms (J+M)terms
& '
× | − | − ···| − | + | + ···| + .
q
( )* +( )* +
(J−N )terms (J+N )terms
Let us consider a single term in the sum over q; the operator e−iβJy , acting on it,
will produce, apart from the ordering of the single vectors | + and | − , a state
vector of the form
J−N J+N
β β β β
cos | − − sin | + cos | + + sin | − =
2 2 2 2
J+N
J−N J−N −t+J+N −r t+r
J −N J +N β β
= (−1)t cos sin
t=0 r=0
t r 2 2
× | − | − ···|− | + | + ···|+ .
( )* +( )* +
(J−N −t+r)terms (J+N −r+t)terms
Whatever the ordering of the various indices, this state vector gives a non-zero
scalar product with one and only one of the terms in the sum over p, provided that
J −M =J −N −t+r
J +M =J +N −r+t.
56 CHAPTER 2
These equalities are both satisfied if r = N − M + t, and since the sum over q gives
(2J)! [(J + N )! (J − N )! ]−1 terms of the same kind
dJMN (β) = (J + M )! (J − M )! (J + N )! (J − N )!
, -2J+M−N −2t , -2t−M+N
cos β2 sin β2
× (−1)t , (2.69)
t
(J + M − t)! (J − N − t)! t! (t + N − M )!
where the sum runs over all the t values which lead to non-negative factorials.
The reduced rotation matrices have several important properties that can be
easily deduced either from their definition or from their analytical expression.
Reality:
dJMN (β)∗ = dJMN (β)
Closure:
dJMN (β1 ) dJN P (β2 ) = dJMP (β1 + β2 )
N
Symmetry:
dJMN (β) = (−1)M−N dJ−M−N (β)
= (−1)M−N dJN M (β)
= (−1)M−N dJMN (−β)
= (−1)J+M dJM−N (π − β)
= (−1)J+N dJM−N (π + β) . (2.70)
Explicit analytical formulae for J = 1 and 2 are given in Table 2.1. Additional
formulae for J = 3 to 6 can be found in Buckmaster (1964, 1966). An interesting
connection between reduced rotation matrices and Jacobi polynomials can be found
in Edmonds (1957).
Going back to the rotation matrices defined in Eq. (2.67), it is possible to establish
for them a number of important properties. First of all, as the rotation operator
D(R) is a unitary operator (D† (R) = D−1 (R)), we have
DMN
J
(αβγ)∗ = DN
J
M (−γ −β −α) , (2.71)
where (−γ −β −α) are the Euler angles characterizing the inverse rotation of (αβγ).
From the unitary character of the operator D(R) one gets the orthogonality
relations
DPJ N (R)∗ DPJ M (R) = δMN
P
∗
DMP
J
(R) DN
J
P (R) = δMN . (2.72)
P
ANGULAR MOMENTUM AND RACAH ALGEBRA 57
TABLE 2.1
Algebraic formulae for d1M N (β) (top) and d2M N (β) (bottom). C and S stand for cos β and sin β,
respectively.
N = −1 N =0 N = +1
M = −1 1
2
(1 + C) 1
√ S 1
2
(1 − C)
2
M =0 − √1 S C 1
√ S
2 2
M = +1 1
2
(1 − C) − √1 S 1
2
(1 + C)
2
N = −2 N = −1 N =0 N = +1 N = +2
3
M = −2 1
4
(1 + C)2 1
2
S(1 + C) 8
S2 1
2
S(1 − C) 1
4
(1 − C)2
3
M = −1 − 12 S(1 + C) (C − 12 )(C + 1) 2
SC (C + 12 )(1 − C) 1
2
S(1 − C)
3 3 3 3
M =0 8
S2 − 2
SC 1
2
(3C 2 − 1) 2
SC 8
S2
3
M = +1 − 12 S(1 − C) (C + 12 )(1 − C) − 2
SC (C − 12 )(C + 1) 1
2
S(1 + C)
3
M = +2 1
4
(1 − C)2 − 12 S(1 − C) 8
S2 − 12 S(1 + C) 1
4
(1 + C)2
From Eq. (2.68), using the symmetry properties (2.70), it can be easily shown
that
DMN
J
(αβγ)∗ = (−1)M−N D−M−N
J
(αβγ) . (2.73)
If R is a rotation composed of two consecutive rotations R1 and R2 , the operator
D(R) can be written as the product D(R1 ) D(R2 ).1 For the rotation matrices we
thus have, directly from their definition
DMN
J
(R) = JM |D(R1 )D(R2 )|JN = DMP
J
(R1 ) DPJ N (R2 ) , (2.74)
P
1
This statement is not trivial. Indeed the operator D(R) should be written
D(R) = D (R2 ) D(R1 ) ,
where
−iβ2 Jy
D (R2 ) = e−iα2 Jz e e−iγ2 Jz ,
with Jy , Jz the components of J on the axes y , z obtained from y, z through the rotation R1 .
Since
Jy = D(R1 ) Jy D −1 (R1 ) , Jz = D(R1 ) Jz D −1 (R1 ) ,
we have, by series expansion of the exponentials
−iβ2 Jy
D (R2 ) = D(R1 ) e−iα2 Jz D −1 (R1 ) D(R1 ) e D −1 (R1 ) D(R1 ) e−iγ2 Jz D −1 (R1 ) ,
so that
D(R) = D(R1 ) D(R2 ) .
58 CHAPTER 2
the so-called closure property of the rotation matrices.
Another important property follows from the addition rules of two angular mo-
menta. Let us consider the product
DMN
J
(αβγ) DM
J
N (αβγ) ;
according to the definition of rotation matrices, this product can be written in the
form
−iαJz −iβJy −iγJz −iαJz −iβJy −iγJz
JM |e e e |JN J M |e e e |J N .
If we now consider the angular momentum operators appearing in the two matrix
elements as referring to two different systems, we can write
−iαJz −iβJy −iγJz
DMN
J
(αβγ) DM
J
N (αβγ) = JJ M M |e e e |JJ N N ,
where the J-operator is the resultant of the angular momenta of the two systems.
On the other hand, bearing in mind the coupling rules of two angular momenta
and the definition of 3-j symbols (Eqs. (2.12) and (2.22)), one obtains with easy
transformations1
DMN
J
(αβγ) DM
J
N (αβγ) =
J J K J J K ∗
= (2K + 1) DQQ
K
(αβγ) , (2.75)
M M Q N N Q
K
a formula that shows how the product of two rotation matrices of rank J and J
can be reduced to the linear combination of rotation matrices having rank K such
that |J − J | ≤ K ≤ (J + J ).
From Eq. (2.75), using the orthogonality relation (2.23a) of the 3-j symbols, the
following expressions are also obtained
∗
DQQ
K
(αβγ) =
J J K J J K
= (2K + 1) DMN
J
(αβγ) DM
J
N (αβγ) ;
M M Q N N Q
MM N N
J J J J J
DMN
J
(αβγ) DM N (αβγ) DM N (αβγ) =
M M M
MM M
J J J
= .
N N N
1
Note that a formal summation over Q and Q might be added in the right-hand side of
Eq. (2.75).
ANGULAR MOMENTUM AND RACAH ALGEBRA 59
Another important property of the rotation matrices is the so-called Weyl theo-
rem
2π 2π π
J ∗
dα dγ dβ sin β DMN
J
(αβγ) DM N (αβγ) =
0 0 0
2
8π
= δ δ δ , (2.76)
2J + 1 JJ MM N N
which is indeed valid only when J and J are both integers or both half-integers.
To prove the theorem, we note that Eqs. (2.73) and (2.75) give
∗ M −N
DMN
J
(αβγ) DM
J
N (αβγ) = (−1)
J J K J J K ∗
× (2K + 1) DQQ
K
(αβγ) , (2.77)
M −M Q N −N Q
K
where, if we confine ourselves to one of the two cases specified above, the sum over
K runs only over integral values. The integral in Eq. (2.76) is therefore reduced to
the sum of various integrals of the form
2π 2π π
∗
dα dγ dβ sin β DQQ
K
(αβγ) ,
0 0 0
with K, Q, and Q integers. The integrals over α and γ can be easily performed
(see Eq. (2.68)) and give the value 4π 2 δQ0 δQ 0 . On the other hand, Eq. (2.69)
yields
, -2K−2t , -2t
cos β2 sin β2
dK
00 (β) = (K! )
2
(−1)t ,
t
[(K − t)! t! ]2
and since
π/2
a! b!
(sin x)2a+1 (cos x)2b+1 dx = ,
2 (a + b + 1)!
0
we have
π
2 K!
dK
00 (β) sin β dβ = (−1)t = 2 δK0 .
K+1 t (K − t)! t!
0
Thus the integral in Eq. (2.76) reduces to
2 M −N J J 0 J J 0
8π (−1)
M −M 0 N −N 0
60 CHAPTER 2
or, using Eq. (2.26a), to
8π 2
δ δ δ ,
2J + 1 JJ MM N N
which proves the Weyl theorem.
This theorem can be generalized to the case where one of the angular momenta
J and J appearing in the rotation matrices of Eq. (2.76) is a half-integer while
the other is an integer. In this case, however, the integrals over α and γ must be
extended to the interval (0, 4π) instead of (0, 2π), and the result is zero. Therefore
the general form of the Weyl theorem, valid for any value of J and J , is the
following
4π 4π π
J ∗ 32π 2
dα dγ dβ sin β DMN
J
(αβγ) DM N (αβγ) = δ δ δ .
2J + 1 JJ MM N N
0 0 0
From the Weyl theorem in the form of Eq. (2.76) it is possible to find a highly
symmetrical expression for the integral of the product of three rotation matrices
2π 2π π
J J2 J3
dα dγ dβ sin β DM1 N1 (αβγ) DM2 N2 (αβγ) DM3 N3 (αβγ) =
1
0 0 0
2 J1 J2 J3 J1 J2 J3
= 8π .
M1 M2 M3 N1 N2 N3
It should be remarked, however, that this expression is valid only when the three
J’s are integers or when two of them are half-integers and the third is an integer.
In different cases the expression in the right-hand side is meaningless.
2.7. Irreducible Spherical Tensors
In the ordinary three-dimensional space referred to a right-handed coordinate sys-
tem, a Cartesian tensor of rank k is defined as a set of 3k quantities (called the
components of the tensor) which, under rotation of the coordinate system, change
according to the linear relationship
T (i1 , i2 , . . . , ik ) = a(i1 , j1 ) a(i2 , j2 ) . . . a(ik , jk ) T (j1 , j2 , . . . , jk )
j1 j2 ...jk
(i1 , i2 , . . . ik , j1 , j2 , . . . , jk = 1, 2, 3) ,
where T (j1 , j2 , . . . , jk ) and T (i1 , i2 , . . . , ik ) are the tensor components in the old
and new reference system, respectively, and where the coefficients a(il , jl ) are the
direction cosines of the new axes relative to the old axes.
ANGULAR MOMENTUM AND RACAH ALGEBRA 61
satisfy a transforma-
For instance, the Cartesian components of a usual vector A
tion of the form
Ai = a(i, j) Aj ,
j
and B
thus a vector is a Cartesian tensor of rank 1. If A are two arbitrary vectors,
defining
Cij = Ai Bj
we have
Cij = a(i, k) a(j, l) Ckl ,
kl
which shows that the dyadic product of two vectors forms a Cartesian tensor of
rank 2.
An irreducible spherical tensor of rank k is defined as a set of (2k + 1) quanti-
ties (called the components of the spherical tensor) which, under rotation of the
coordinate system, change according to the linear relation
T kq = Tpk Dpq
k
(R) (q, p = −k, −k + 1, . . . , k) , (2.78)
p
where T kq and Tpk are the tensor components in the new and old reference system,
respectively, and where R is the rotation that brings the old system into the new
one.
Contrary to the case of Cartesian tensors, the dyadic product of two irreducible
spherical tensors no longer is an irreducible spherical tensor. In fact, if Rqk and Sqk
are two irreducible tensors, the (2k + 1)(2k + 1) components of the form Rqk Sqk
change, under the rotation R, into
(Rqk Sqk ) = (Rpk Spk ) Dpq
k
(R) Dpk q (R) .
pp
On the other hand, using Eq. (2.75) we can write
k k K k k K
(Rqk Sqk ) = (Rpk Spk ) (2K + 1) DQ
K ∗
P (R) .
p p Q q q P
pp K
Multiplying both sides by
k k K
(−1)k−k +Q ,
q q −Q
summing over q and q , and defining
k−k +Q k k K
TQK = (−1) (Rqk Sqk ) , (2.79)
q q −Q
62 CHAPTER 2
we obtain, with the help of Eq. (2.23a)
k k K
T K = (Rpk Spk ) (−1)k−k +Q DQ
K
−Q (R)
∗
Q p p Q
pp
∗
= K
T−Q (−1)
Q+Q
DQ
K
−Q (R) .
Q
Recalling the conjugation property of the rotation matrices (Eq. (2.73)) and chang-
ing the dummy index Q in −P , we obtain for TQK the typical transformation law
for spherical tensors,
T K
Q = TPK DPKQ (R) .
P
Equation (2.79) can thus be regarded as a kind of recipe to obtain a spherical
tensor of rank K from the dyadic product of two spherical tensors of rank k and
k . It should be noted, however, that definition (2.79) is by no means unique, as
the multiplication of the right-hand side by a factor of the form f (K), with f an
arbitrary function, does not affect the transformation law of the tensor TQK under
rotations.
The concept of spherical tensor can be easily extended to quantum-mechanical
operators. A spherical tensor operator of rank k is defined as a set of (2k + 1)
operators (called the components of the tensor) which, under rotation of the coor-
dinate system, change according to the linear relation (2.78). Since, on the other
hand, quantum-mechanical operators are transformed according to Eq. (2.57), the
components of a spherical tensor operator must satisfy the relation
D(R) Tqk D† (R) = Tpk Dpq
k
(R) . (2.80)
p
From this equation it is possible to deduce the commutation rules of a spherical
tensor operator with angular momentum. Let us consider an infinitesimal rotation
of the coordinate system through an angle φ about the unit vector u. We have (see
Eq. (2.61))
D(R) = 1 − iφJu
D† (R) = 1 + iφJu
Dpq
k
(R) = kp|1 − iφJu |kq = δpq − iφ kp|Ju |kq .
Substitution into Eq. (2.80) yields
[Ju , Tqk ] = Tpk kp|Ju |kq ,
p
and equating Ju with Jx , Jy , and Jz one gets (see Eqs. (2.7), (2.9), and (2.10))
[Jz , Tqk ] = q Tqk
[J± , Tqk ] = (k ± q + 1)(k ∓ q) Tq±1
k
. (2.81)
ANGULAR MOMENTUM AND RACAH ALGEBRA 63
Conversely,these commutation rules could be taken as basic definitions of a spher-
ical tensor operator. It can indeed be proved that from these equations one easily
gets back transformation (2.78).
From Eqs. (2.81) it is possible to find the spherical components, A1q , of a vector
Equation (2.60) gives
operator A.
[Jz , Az ] = 0 ,
hence we can write
A10 = Az .
Setting k = 1 and q = 0 in the second of Eqs. (2.81) we further obtain
√ 1
[J± , A10 ] = 2 A±1 ,
so that, summarizing
1
A1−1 = √ (Ax − iAy )
2
1
A0 = Az
1
A11 = − √ (Ax + iAy ) . (2.82)
2
Given two tensor operators of rank 1, it is possible to construct three tensor
operators of rank 0, 1, and 2, respectively, by considering the linear combinations
(2.79). For the 0-rank tensor one gets, using Eq. (2.26a)
1 1
T00 = √ (−1)1−q Rq1 S−q
1
= √ (R11 S−1
1
− R01 S01 + R−1
1
S11 )
3 q 3
1
= −√ R ·S , (2.83)
3
√
which shows that, apart from a numerical factor (−1/ 3), the 0-rank tensor is
simply the scalar product of the two vectors.
For the tensor of rank 1, a direct application of Eq. (2.79), with the 3-j symbols
computed from Eqs. (2.26c,d), gives
1
1
T−1 = √ (R01 S−11
− R−11
S01 )
6
1
T01 = √ (R11 S−1
1
− R−1
1
S11 )
6
1
T11 = √ (R11 S01 − R01 S11 ) ,
6
64 CHAPTER 2
or, in terms of Cartesian components (see Eqs. (2.82) and their inverse)
i
Tx = √ (Ry Sz − Rz Sy )
6
i
Ty = √ (Rz Sx − Rx Sz )
6
i
Tz = √ (Rx Sy − Ry Sx ) .
6
Therefore the spherical components of the tensor of rank 1 determine a vector T
which, apart from a numerical factor, is simply the cross product of the vectors R
and S,
i
T = √ R ×S .
6
Finally, the components of the tensor of rank 2 are given by
1 1 1
2
T−2 = √ R−1 S−1
5
1
2
T−1 = √ (R01 S−1
1 1
+ R−1 S01 )
10
1
T02 = √ (R−11
S11 + 2R01 S01 + R11 S−1
1
)
30
1
T12 = √ (R01 S11 + R11 S01 )
10
1 1 1
T2 = √ R1 S1 .
2
(2.84)
5
It should be remarked that the adjoint (or Hermitian conjugate) of a spherical
tensor operator Tqk no longer is a spherical tensor operator. This can be easily
proved by taking the Hermitian conjugates of Eqs. (2.81), recalling that
[A, B]† = −[A† , B † ] , (J± )† = J∓ ;
we have
[Jz , (Tqk )† ] = −q (Tqk )†
[J± , (Tqk )† ] = − (k ∓ q + 1)(k ± q) (Tq∓1
k
)† .
However, the operator
k †
Oqk = (−1)r−q (T−q ) ,
where r is zero for k integer and 1/2 for k half-integer, is indeed a spherical tensor
operator. In fact
k † k †
[Jz , Oqk ] = −(−1)r−q [Jz , T−q ] = (−1)r−q q (T−q ) = q Oqk
k †
[J± , Oqk ] = −(−1)r−q [J∓ , T−q ] = (−1)r−q∓1 (k ± q + 1)(k ∓ q) (T−q∓1
k
)†
= (k ± q + 1)(k ∓ q) Oq±1k
.
ANGULAR MOMENTUM AND RACAH ALGEBRA 65
The spherical tensor operator Oqk can be considered as the Hermitian conjugate of
the tensor Tqk ; employing for this tensor the notation T † we have
(T † )kq = (−1)r−q (T−q
k †
) . (2.85)
A Hermitian spherical tensor operator is defined as an operator such that
(T † )kq = Tqk ;
as a consequence, the components of such operators satisfy the relation
k †
Tqk = (−1)r−q (T−q ) . (2.86)
form
It is seen at once that the spherical components A1q of a Hermitian vector A
a Hermitian spherical tensor operator. From Eqs. (2.82) we have
1
(A1−1 )† = √ (Ax + iAy ) = −A11
2
(A10 )† = Az = A10
1
(A11 )† = − √ (Ax − iAy ) = −A1−1 ,
2
so that Eq. (2.86) is satisfied.
2.8. The Wigner-Eckart Theorem and its Consequences
The Wigner-Eckart theorem has a fundamental importance in angular momentum
theory, as it allows the matrix elements of spherical tensor operators to be reduced
to simple mathematical expressions.
For the evaluation of the matrix element
αJM |Tqk |α J M , (2.87)
where Tqk is the q-component of a spherical tensor operator of rank k, and where
α and α are supplementary quantum numbers (relative to Hermitian operators
commuting with J 2 and Jz ), we first consider the (2k + 1)(2J + 1) vectors
Tqk |α J M ,
or, more precisely, their linear combinations
Tqk |α J M J kM q|J M , (2.88)
Mq
66 CHAPTER 2
where J is a given angular momentum eigenvalue having the same character (in-
teger or half-integer) of (J + k), and limited between |J − k| and (J + k). We
will now show that the (2J + 1) vectors obtained by varying M in the linear
combinations (2.88) are eigenvectors of J 2 and Jz corresponding to the eigenvalues
J (J + 1) and M , respectively.
Denoting by |v(J , M ) the linear combination (2.88), we have
Jz |v(J , M ) = Jz Tqk |α J M J kM q|J M ,
Mq
and using the commutation rules of spherical tensors (Eqs. (2.81)) we obtain
Jz |v(J , M ) = (M + q) Tqk |α J M J kM q|J M
Mq
= M |v(J , M ) .
(2.89)
Similarly, one gets from Eqs. (2.81) and (2.9)
J+ |v(J , M ) =
= J (J + 1) − M (M + 1) Tqk |α J M +1 J kM q|J M
Mq
+ k(k + 1) − q(q + 1) Tq+1
k
|α J M J kM q|J M ,
Mq
and renaming the summation indices
J+ |v(J , M ) =
= Tqk |α J M J (J + 1) − M (M − 1) J k M −1 q|J M
Mq
+ k(k + 1) − q(q − 1) J kM q−1|J M ,
which can be rewritten, with the help of the recursion relation (2.21), in the form
J+ |v(J , M ) = J (J + 1) − M (M + 1) |v(J , M +1) . (2.90)
In a similar way one can prove that
J− |v(J , M ) = J (J + 1) − M (M − 1) |v(J , M −1) , (2.91)
so that, from Eqs. (2.89), (2.90), and (2.91)
1
J 2 |v(J , M ) = Jz2 + (J+ J− + J− J+ ) |v(J , M )
2
= J (J + 1) |v(J , M ) . (2.92)
ANGULAR MOMENTUM AND RACAH ALGEBRA 67
Equations (2.89) and (2.92) show that the (2J + 1) vectors of the form
|v(J , M ) are angular momentum eigenvectors (in general not normalized) cor-
responding to the eigenvalues J and M .
Recalling the definition of |v(J , M ) we have, after inversion of Eq. (2.88) via
the orthogonality relations of the vector-coupling coefficients (Eq. (2.20b))
Tqk |α J M = J kM q|J M |v(J , M ) ,
J
and hence the scalar products (2.87) can be written in the form
αJM |Tqk |α J M = J kM q|J M αJM |v(J , M ) . (2.93)
J
On the other hand, the scalar product αJM |v(J , M ) is zero unless J = J
and M = M . Moreover, it is independent of M (see the comments after Eq. (2.30))
so that we can write
αJM |v(J , M ) = δJJ δMM (−1)2k αJ T k α J , (2.94)
where the quantity αJ T k α J is the so-called reduced matrix element 1 of the
operator Tqk . Substitution of Eq. (2.94) into Eq. (2.93) leads to the final expression
of the Wigner-Eckart theorem
αJM |Tqk |α J M = (−1)2k αJ T k α J J kM q|JM , (2.95)
or, in terms of 3-j symbols
αJM |Tqk |α J M =
√ J J k
= (−1)J +k+M
2J + 1 αJ T k α J . (2.96)
−M M q
The reduced matrix element can be expressed in terms of ordinary matrix ele-
ments by inversion of Eq. (2.96). From the orthogonality relation of the vector-
coupling coefficients (Eq. (2.20a)) we have
αJ T k α J = (−1)2k J kM q|JM αJM |Tqk |α J M .
Mq
A different way for calculating the reduced matrix element is to substitute some
special values of M and M into Eq. (2.95). For the reduced matrix element of the
angular momentum operator J we have for instance (denoting J 1 by J )
αJM |J01 |α J M = M δαα δJJ δMM = αJ J α J J 1M 0|JM ,
1
The definition given here for the reduced matrix
√ element agrees with that given by Brink
and Satchler (1968), while it differs by a factor 2J + 1 from those given by Racah (1942) and
Edmonds (1957).
68 CHAPTER 2
and being
M
J1M 0|JM =
J(J + 1)
one obtains
αJ J α J = δαα δJJ J(J + 1) . (2.97)
An important property of the reduced matrix element can be easily deduced from
the orthogonality relation of the vector-coupling coefficients
| αJM |Tqk |α J M |2 = | αJ T k α J |2 .
Mq
The most striking consequence of the Wigner-Eckart theorem is the direct ap-
pearance of selection rules for the matrix elements of the components of spherical
tensor operators. Indeed, because of the presence of the 3-j symbol in Eq. (2.96),
the matrix element αJM |Tqk |α J M is identically zero unless
|J − k| ≤ J ≤ J + k , M = M − q .
In particular, we obtain from Eq. (2.26a) that the matrix elements of a zero-rank
spherical tensor operator are diagonal with respect to J and M , and are indepen-
dent of M .
2.9. Properties of Reduced Matrix Elements
From the Wigner-Eckart theorem an important relation can be deduced for the
reduced matrix elements of a Hermitian spherical tensor operator Hqk . Taking the
complex conjugate of Eq. (2.95) we have
∗ ∗
αJM |Hqk |α J M = (−1)2k αJ H k α J J kM q|JM . (2.98)
On the other hand, being Hqk Hermitian, from Eq. (2.86) we deduce
∗
αJM |Hqk |α J M = α J M |(Hqk )† |αJM
= (−1)r+q α J M |H−q
k
|αJM .
Applying again the Wigner-Eckart theorem to the last matrix element we get
∗
αJM |Hqk |α J M = (−1)r+q+2k α J H k αJ JkM −q|J M . (2.99)
From the symmetry properties of the vector-coupling coefficients (Eqs. (2.22),
(2.24), (2.25)) we have
.
J−J −M+M 2J + 1
JkM −q|J M = (−1) J kM q|JM , (2.100)
2J + 1
ANGULAR MOMENTUM AND RACAH ALGEBRA 69
and, finally, from Eqs. (2.98), (2.99), and (2.100) we get
√ ∗ √
2J + 1 αJ H k α J = (−1)J−J +r
2J + 1 α J H k αJ , (2.101)
which is the relation, valid for Hermitian spherical tensor operators only, between
reduced matrix elements of the form αJ H k α J and α J H k αJ .
In Sect. 2.7 we have shown how it is possible to construct an irreducible spherical
tensor TQK from the product of two irreducible tensors Rqk and Sqk . We want here
to show how the reduced matrix elements of these tensors are related among each
other. The Wigner-Eckart theorem, applied to the tensor TQK , gives
√ J J K
αJM |TQK |α J M = (−1)J +K+M
2J + 1 αJ T K α J .
−M M Q
From Eq. (2.79), introducing intermediate states between the tensors Rqk and Sqk
and applying again the Wigner-Eckart theorem, one obtains
√
+M +2k−K+Q
(−1)J 2J + 1
qq α J M
J J k J J k k k K
×
−M M q −M M q q q −Q
× αJ Rk α J α J S k α J =
J J K
= αJ T K α J .
−M M Q
Multiplication of both sides by
J J K
−M M Q
followed by summation over M and M gives, using Eqs. (2.23a) and (2.34)1
√ !
J J K
αJ T K α J =(−1)J+J +K
2J + 1
k k J
α J
× αJ Rk α J α J S k α J , (2.102)
which shows how the reduced matrix element of a ‘composite’ spherical tensor is
related to the reduced matrix elements of the constituent tensors.
1
The deduction of Eq. (2.102) is not trivial, as it involves the contraction of four 3-j symbols
which must be suitably manipulated to obtain the formula (Eq. (2.34)) leading to the final result.
In App. 2 we give, for the unexperienced reader, an example of how the various manipulations
can be performed.
70 CHAPTER 2
A direct application of this formula to the 0-rank tensor T 0 obtained from the
scalar product of an arbitrary vector v with the angular momentum J gives (see
Eq. (2.83))
1
αJ T 0 α J = − √ αJ v · J α J
3
!
√ J J 0
= (−1)J+J 2J + 1 αJ v α J α J J α J ,
1 1 J
α J
which, using Eqs. (2.36a) and (2.97), can be written in the form
αJ v · J α J = δJJ J(J + 1) αJ v α J .
From this equation a remarkable relation, often referred to as the projection theo-
rem, can be easily proved by a double application of the Wigner-Eckart theorem,
[J(J + 1)] αJM |v |α JM = αJM | (v · J ) J |α JM . (2.103)
We have indeed for the spherical components of the operator J
αJM | (v · J ) Jq1 |α JM =
= αJM |v · J |α J M α J M | Jq1 |α JM
α J M
= αJ v · J α J J0M 0|JM α J J α J J1M q|JM
M
= J(J + 1) αJ v α J J1M q|JM
= J(J + 1) αJM | vq1 |α JM , (2.104)
which proves Eq. (2.103).
The projection theorem has a very simple physical meaning that is depicted in
Fig. 2.5. According to the vectorial model, the vector v rotates rapidly about J,
so that its only effective component, veff , points in the J -direction. On the other
hand, veff can be written in the form
(v · J ) J
veff = ,
J2
thus the projection theorem can be considered as the direct generalization to Quan-
tum Mechanics of this simple geometrical expression.
The projection theorem can be expressed in the equivalent form (see Eq. (2.104))
αJM |v |α JM = gαα J (v ) α JM |J |α JM ,
ANGULAR MOMENTUM AND RACAH ALGEBRA 71
Fig.2.5. The vector v is rapidly spinning about
J, so that its only effective component, veff , is in
the
J -direction.
where
αJM | v · J |α JM
αJ v α J
gαα J (v ) = = (2.105)
J(J + 1) J(J + 1)
is the generalization of the Landé factor that will be encountered in the next chap-
ter.
A final application of the Wigner-Eckart theorem concerns the matrix elements of
a spherical tensor operator acting on the dynamical variables of a single component
of a composite system.
Let us consider a system composed of two parts of angular momenta J1 and J2 ,
respectively, having total angular momentum J = J1 + J2 . If Tqk (1) is a spher-
ical tensor operator (the argument (1) means that it is acting only on the first
component of the system), we have for its matrix elements
αj1 j2 JM | Tqk (1) |α j1 j2 J M =
= (−1)2k J kM q|JM αj1 j2 J T k (1) α j1 j2 J . (2.106)
On the other hand, if we change the coupling scheme using the vector-coupling
coefficients, we can write
αj1 j2 JM | Tqk (1) |α j1 j2 J M =
= j1 j2 m1 m2 |JM j1 j2 m1 m2 |J M
m1 m2 m1 m2
× αj1 j2 m1 m2 | Tqk (1) |α j1 j2 m1 m2 .
Since Tqk (1) acts only on the first component of the system, applying again the
Wigner-Eckart theorem we obtain
αj1 j2 m1 m2 | Tqk (1) |α j1 j2 m1 m2 =
= (−1)2k j1 k m1 q|j1 m1 αj1 T k (1) α j1 δj j δm m2 . (2.107)
2 2 2
72 CHAPTER 2
Comparison of Eqs. (2.106) and (2.107) gives the following relation between reduced
matrix elements
J kM q|JM αj1 j2 J T k (1) α j1 j2 J =
= j1 j2 m1 m2 |JM j1 j2 m1 m2 |J M j1 k m1 q|j1 m1
m1 m2 m1
× αj1 T k (1) α j1 δj j .
2 2
This relation can be expressed in a more compact form by a calculation similar to
that in App. 2. This calculation, which is left as an exercise to the reader, leads to
the following expression
αj1 j2 J T k (1) α j1 j2 J =
!
j1 j1 k
= (−1)j1 +j2 +J +k
(2J + 1)(2j1 + 1)
J J j2
× αj1 T k (1) α j1 δj j . (2.108)
2 2
Similarly, if Tqk (2) is a spherical tensor operator acting only on the second part of
the composite system, we obtain
αj1 j2 J T k (2) α j1 j2 J =
!
j1 +j2 +J+k j2 j2 k
= (−1) (2J + 1)(2j2 + 1)
J J j1
× αj2 T k (2) α j2 δj j . (2.109)
1 1
k k
More generally, if Rq11 (1) and Sq22 (2) are two spherical tensor operators acting on
the first and second part of the composite system, respectively, and if TQK is the
spherical tensor operator constructed according to Eq. (2.79), the reduced matrix
elements are connected by the following relation
αj1 j2 J T K α j1 j2 J =
⎧ ⎫
⎨J J K⎬
= (2j1 + 1)(2j2 + 1)(2J + 1) j1 j1 k1
⎩ ⎭
j2 j2 k2
× αj1 Rk1 (1) α j1 α j2 S k2 (2) α j2 .
α
The proof of this relation, that follows from Eqs. (2.34) and (2.48), is left as an
exercise to the reader.
CHAPTER 3
ATOMIC SPECTROSCOPY
Polarization can be introduced in spectral lines by several different mechanisms,
that are connected either with the presence of an external field (like for instance
a magnetic or an electric field) or with the existence of some kind of anisotropy in
the excitation of the atomic system (optical pumping, impact polarization, etc.).
While the detailed theory describing the interplay of the various mechanisms
will be tackled in the following chapters, it is first necessary to state some specific
spectroscopic notations, and to give a number of basic results concerning the inter-
action of an atomic system with an external field, in order to establish an adequate
physical background. In particular, we will review in this chapter the theory of
the Zeeman and Paschen-Back effects, including the case of hyperfine structure,
and we will give a description of atomic polarization in terms of the density (or
statistical) operator and of statistical tensors.
3.1. Zeeman Effect
According to the theory of atomic structure (see for instance Condon and Shortley,
1935; Condon and Odabaşi, 1980), the modifications produced in atomic spectra by
an external, uniform magnetic field can be described by adding to the unperturbed
Hamiltonian H0 of the atomic system an additional term HB , called the magnetic
Hamiltonian, given by
2
e0 h + e0 (B
)·B × r )2 ,
HB = (L + 2S (3.1)
4πmc 8mc2
where e0 is the absolute value of the electron charge, m is the electron mass, h and
c have their usual meaning of Planck constant and speed of light, L and S
are the
(dimensionless) total orbital angular momentum and total spin of the electronic
is the magnetic field vector, and r is the position operator defined by
cloud, B
r = ri ,
i
where ri is the position of the i-th optical electron relative to the nucleus. For one-
electron atoms, Eq. (3.1) can be derived by taking the lowest-order, non relativistic
limit of Dirac’s equation describing the motion of a particle in an electromagnetic
field (see for instance Schiff, 1949).
The second term in Eq. (3.1) is the so-called diamagnetic term. Its importance
is very limited in practice, at least for the magnetic fields that are usually found
74 CHAPTER 3
in laboratory or astrophysical plasmas. Indeed, comparing the two terms in the
right-hand side of Eq. (3.1) and taking, as an order of magnitude, the radius of the
first Bohr’s orbit for the expectation value of r, we obtain for their ratio
second term 1 h3
≈ B = 1.06 × 10−10 B , (3.2)
first term 32π 3 m2 c e30
where B is expressed in G. If we exclude the extremely large fields that are found in
magnetic white dwarfs (see e.g. Angel et al., 1981, or Schmidt, 1987, for a review
on the subject) or are supposed to exist in neutron stars, the diamagnetic term can
be safely neglected in Eq. (3.1), so that we can write
e0 h )·B
= µ (L
+ 2S
)·B
,
HB = (L + 2S 0 (3.3)
4πmc
where µ0 is the so-called Bohr magneton (µ0 = 9.27 × 10−21 erg G−1 ).
If the magnetic field is so weak as to keep the magnetic energy much smaller than
the energy intervals relative to the unperturbed Hamiltonian H0 , the effect of HB
can be computed by perturbation theory (see Messiah, 1961, for an introduction to
the argument). According to this theory, the shifts of an N -fold degenerate energy
level of H0 are obtained by diagonalization of the matrix
(α) (α)
(HB )ij = ui |HB |uj (i, j = 1, . . . , N ) ,
(α)
where |uk (k = 1, . . . , N ) are the degenerate eigenvectors of H0 corresponding
to the eigenvalue Eα .
In our case, since the Hamiltonian H0 is invariant under rotations, the total
angular momentum J and its projection M along an arbitrary z-axis are good
quantum numbers, so that the eigenvectors can be written in the form |αJM
(where α is a collection of inner quantum numbers of H0 ), with
H0 |αJM = EαJ |αJM
J |αJM = J(J + 1) |αJM
2
Jz |αJM = M |αJM . (3.4)
When we add to H0 the magnetic Hamiltonian HB , the corrections to the degen-
erate energy EαJ are found by diagonalization of the matrix
αJM |HB |αJM = µ0 αJM | (L
+ 2S |αJM .
)·B
The matrix elements can be evaluated for any direction of the magnetic field in
the reference system (xyz) chosen to describe the atomic system. Writing B in the
form B = B b, where b is a unit vector in the magnetic field direction, and writing
(L+2 ) in the form (J + S
S ), where J is the total angular momentum of the atomic
system, we have
αJM |HB |αJM = µ0 B αJM | (J + S
) · b |αJM .
ATOMIC SPECTROSCOPY 75
Introducing the spherical components of the three vectors, the scalar product can
be written in the form (see Eq. (2.83))
) · b =
(J + S (−1)q (Jq + Sq ) b−q ,
q
so that, applying the Wigner-Eckart theorem in the form of Eq. (2.96), and using
Eq. (2.97), one obtains
√ J 1 J
αJM |HB |αJM = µ0 B (−1) J+M+q+1
2J + 1
M q −M
q
, -
× αJ b .
J(J + 1) + αJ S (3.5)
−q
Owing to the selection rules of the 3-j symbol, these matrix elements are zero
unless (M − M ) = 0, or ±1. We have therefore a tridiagonal matrix that might
in principle be diagonalized by standard methods.
The diagonalization can however be avoided by aligning the z-axis of the reference
system with the direction of the magnetic field. In this case, in fact, the only non-
zero component of the unit vector b is b0 = 1, and we easily obtain, substituting
the value of the 3-j symbol given by Eq. (2.26d)
αJM |HB |αJM = µ0 B g M δMM , (3.6)
where g, the so-called Landé factor , can be written in the form1
αJ S αJ
g =1+ (J = 0). (3.7)
J(J + 1)
Equation (3.6) shows that the magnetic Hamiltonian is now represented by a di-
agonal matrix. This means that, to the first order of perturbation theory, the
eigenstates of the total Hamiltonian (H0 + HB ) are of the form |αJM , while the
eigenvalues are given by
EαJ + µ0 gBM (M = −J, −J + 1, . . . , J).
Thus the Hamiltonian HB removes the degeneracy, and any level characterized by
the quantum number J is split into (2J + 1) equally spaced sublevels, the splitting
being proportional to the Landé factor g and to the magnetic field.
We want to remark, however, that the states |αJM are eigenvectors of the total
Hamiltonian only when the z-axis of the reference system points in the magnetic
1
As apparent from the 3-j symbol in Eq. (3.5), the matrix elements of HB identically vanish
if J = 0. This simply means that the magnetic field does not affect, to the first order, the energy
of the atomic levels characterized by J = 0. Consequently, one can formally put g = 0 for such
levels.
76 CHAPTER 3
field direction. This means that a state having a definite value for the projection
of the total angular momentum along the direction of the magnetic field has a
well-defined energy value. When the z-axis of the reference system is not in the
direction of the magnetic field, a state of the form |αJM is an eigenvector of H0 ,
but not of (H0 + HB ).
The general expression for the Landé factor, given in Eq. (3.7), involves the
evaluation of the reduced matrix element of S. This can be easily done when
the levels are described by the Russel-Saunders (or L-S ) coupling scheme (see
e.g. Condon and Shortley, 1935). In this scheme the quantum numbers L and
S corresponding to the total orbital angular momentum and to the total spin,
respectively, are good quantum numbers, so that an eigenvector of the Hamiltonian
H0 can be written in the form
|βLSJM ,
where β is a collection of quantum numbers representing the electronic configura-
tion. The reduced matrix element is then
βLSJ ,
βLSJ S
and it can be evaluated in the easiest way with the help of the projection theorem.
From Eq. (2.105) we have
· J |βLSJM
βLSJM |S
βLSJ =
βLSJ S ,
J(J + 1)
· J in the form
and writing S
· J = 1 J 2 + S 2 − L2
S
2
we obtain for the Landé factor
1 J(J + 1) + S(S + 1) − L(L + 1)
gLS = 1 + . (3.8)
2 J(J + 1)
This formula can be rewritten in the form
gLS = 1 + γ(J, S, L) ,
where we have used the compact notation
A(A + 1) + B(B + 1) − C(C + 1)
γ(A, B, C) = .
2A(A + 1)
In the j-j coupling scheme (see e.g. Condon and Shortley, 1935) a simple expres-
sion for the Landé factor can be obtained for levels originating from two optical
ATOMIC SPECTROSCOPY 77
electrons of orbital angular momentum l1 and l2 , respectively. In this case the
reduced matrix element to be evaluated is the following
β (l s )j , (l s )j , J
β (l1 s1 )j1 , (l2 s2 )j2 , J S ,
1 1 1 2 2 2
where s1 = s2 = 1/2. Writing S in terms of the spins of the single electrons
= s + s ) and applying Eqs. (2.108), (2.109), and (2.36d), one obtains after
(S 1 2
some calculations
gjj = 1 + γ(J, j1 , j2 ) γ(j1 , 1/2, l1) + γ(J, j2 , j1 ) γ(j2 , 1/2, l2) .
Another case where it is possible to find an analytical expression for the Landé
factor is the case of the so-called J1 -l coupling. In this coupling scheme a ‘parent’
level of orbital angular momentum L1 and spin S1 couples its total angular mo-
mentum J1 with the orbital angular momentum l of a further electron, to give an
angular momentum K which in its turn couples with the electron spin to give the
total angular momentum J. Thus the reduced matrix element to be evaluated is
the following
β (L1 S1 )J1 , l K, s, J S β (L S )J , l K, s, J ,
1 1 1
where s = 1/2. After some tedious algebra, that is left as an exercise to the reader,
one gets the formula
gJ l = 1 + γ(J, 1/2, K) + γ(J, K, 1/2) γ(K, J1, l) γ(J1 , S1 , L1 ) .
1
In more complicated coupling schemes, or in intermediate coupling, no simple
analytical expression can be found for the Landé factor unless the eigenfunction
of the level is known in full detail. If we know, for instance, how to express the
eigenvector |αJM of a given level in terms of the L-S coupling eigenvectors (see
Condon and Shortley, 1935, for further details),
|αJM = c(β, L, S) |βLSJM ,
βLS
it can be shown by some Racah algebra that
gic = |c(β, L, S)|2 gLS ,
βLS
where gLS are the Landé factors computed from Eq. (3.8). This formula is however
of little use, because the coefficients are seldom known with a sufficient degree of
accuracy.
It should be remarked, however, that in many cases the Landé factors are known
from experimental work carried out in spectroscopy laboratories. A systematic
(but incomplete and non-updated) list of experimental g-values can be found in
78 CHAPTER 3
Moore (1949, 1952, 1958). For the FeI and FeII spectra, further g-values can be
found in Reader and Sugar (1975) and in Corliss and Sugar (1982).
For an atom in L-S coupling, the Landé factor is equal to 1 for a level having
S = 0, while it is equal to 2 for a level having L = 0. When both L and S are
non-zero, g is generally bounded between 1 and 2, although many exceptions are
found to this simple rule, especially for low J-values. In rare cases, the combination
of the quantum numbers may give a negative or vanishing value for g. L-S levels
having g = 0 are 4 D1/2 , 5 F1 , 6 G3/2 , 7 H2 , 8 I5/2 , etc., while L-S levels having g < 0
are 6 F1/2 , 7 G1 , 8 G1/2 , 8 H3/2 , etc. In other cases, large values of g can result; for
instance, L-S levels having g ≥ 2.5 are 4 P1/2 (g = 2.667), 5 P1 (g = 2.5), 6 D1/2
(g = 3.333), 7 D1 (g = 3), 8 D3/2 (g = 2.8), 8 F1/2 (g = 4), etc.
Let’s now consider the Zeeman pattern to be expected in the transition between
two atomic levels, both split by the presence of a magnetic field. If J and J are the
angular momentum quantum numbers of the lower and upper level, respectively,
and if g and g are the corresponding Landé factors, the spectral line originating
from the transition between the two levels splits, owing to the magnetic field, into
a collection of components whose frequencies are given by
µ0 B
JJ
νMM = ν0 + (g M − gM ) , (3.9)
h
where ν0 is the frequency of the unperturbed line and where M and M are the
magnetic quantum numbers of the lower and upper sublevels, respectively. Formula
(3.9) is often written in the form
= ν0 + νL (g M − gM ) ,
JJ
νMM
where
µ0 B e B
= 0
νL = (3.10)
h 4πmc
is the so-called Larmor frequency; numerically we have
νL = 1.3996 × 106 B , (3.11)
with B expressed in G and νL in s−1 .
The wavelengths of the transitions can be easily evaluated in the visible and
infrared, since in these cases the Larmor frequency is much smaller than ν0 (for
instance, for a line at 1.2 µm and for B = 104 G, the ratio νL /ν0 is of order 6×10−5),
so that we can write
MM = λ0 − ∆λB (g M − gM ) ,
λJJ (3.12)
where λ0 = c/ν0 is the wavelength of the unperturbed line, and where
νL λ2 e B
∆λB = λ20 = 0 0 2 . (3.13)
c 4πmc
ATOMIC SPECTROSCOPY 79
Numerically we have
∆λB = 4.6686 × 10−10 λ20 B , (3.14)
where ∆λB is in mÅ, λ0 in Å, and B in G.
It is important to note that, of the various Zeeman components (whose wave-
lengths are given by Eq. (3.12)) arising in a given transition, only a limited number
is actually observed: this is due to selection rules that depend on the interaction
between atoms and radiation field. The general properties of Zeeman patterns
(namely the number of observed lines, their wavelength position, their relative
strength and polarization features) depend on the kind of interaction (electric-
dipole, magnetic-dipole, electric-quadrupole, etc.). In the rest of this section we
summarize the main properties of the Zeeman patterns resulting from the electric-
dipole interaction, which is the simplest one and the most important in practice.
The reader is referred to Sect. 9.1 for a formal derivation of what is anticipated
here.
The selection rule for electric-dipole transitions is
∆M = M − M = 0, ±1 , (3.15)
from which three distinct groups of Zeeman components arise.
The components having ∆M = −1, that will be called in the following the σr
components, are generally displaced to longer wavelengths (or towards the red side
of the spectrum) from the unperturbed line. In emission, they give rise in general
to elliptically polarized radiation, which degenerates into circularly polarized ra-
diation when observed along the direction of the magnetic field, and into linearly
polarized radiation when observed in the plane perpendicular to the magnetic field.
In particular, referring to Fig. 3.1, any σr component produces in emission: right-
handed (or positive) circular polarization along the positive z-axis, left-handed (or
negative) circular polarization along the negative z-axis, and linear polarization,
perpendicular to the z-axis, along any direction in the x-y plane.1
The components having ∆M = +1 (σb components) are generally displaced to
shorter wavelengths (or towards the blue side of the spectrum) from the unper-
turbed line. Their behavior is similar to that of the σr components except for the
handedness of circular polarization. Referring again to Fig. 3.1, any σb component
produces in emission: left-handed (or negative) circular polarization along the pos-
itive z-axis, and right-handed (or positive) circular polarization along the negative
z-axis. In the x-y plane the σb components have the same properties as the σr
components.
Finally, the wavelengths of the components having ∆M = 0 (the so-called π
components) fall in between those of the σr and σb components. In emission, the
1
In absorption, the situation is more complicated because the polarization of the radiation
absorbed by the σr components depends also on the polarization of the incident radiation. For
the special case of an unpolarized incident beam, the absorbed radiation has the same polariza-
tion characteristics as the radiation that would be emitted in the direction of the beam. As a
consequence, the opposite polarization will be present in the beam after absorption.
80 CHAPTER 3
σ
σ
σ π σ
σ
π
σ
σ
σ
Fig.3.1. The polarization properties of the radiation emitted by different Zeeman components
depend on the type of component and on the angle between the observing direction and the
magnetic field vector. For the radiation emitted along the x, y, and z-axes, the tip of the wave
electric field vector draws, in a fixed point, the figures here shown.
π components give rise to linear polarization whose direction is always parallel to
the magnetic field, as represented schematically in Fig. 3.1.
In particular, for an observer located in the x-y plane, all the components pro-
duce linear polarization that is parallel (for the π components) or perpendicular
(for the σ components) to the direction of the magnetic field. This explains the
denominations π and σ, which follow from the German words parallel and senkrecht
(perpendicular).
As far as the number of Zeeman components is concerned, the simplest case (often
referred to as normal Zeeman effect ) occurs when the angular momentum of any
of the two levels involved in the transition is zero (J = 0 or J = 0), or when both
levels have the same Landé factor (g = g ).1 In both cases only three components
are left: one σr component at the wavelength (λ0 + g∆λB ), one σb component at
the wavelength (λ0 − g∆λB ), and one π component at the wavelength λ0 , where
the g-factor is either the Landé factor of the level having J = 0, or the Landé factor
common to the two levels.2
1
The two cases will be referred to in this book as normal Zeeman triplet and anomalous
Zeeman triplet, respectively. Other authors reserve both names for the case (J = 0 or J = 0),
and use normal or anomalous according as the Landé factor is unity or not.
2
The normal Zeeman effect (or, more properly, the normal Zeeman triplet case) can also
be treated by a classical (non quantum-mechanical) approach, that will be presented in the next
section.
ATOMIC SPECTROSCOPY 81
TABLE 3.1
Analytical expressions for the strengths of Zeeman components. J and M are the quantum
numbers of the lower sublevels, J and M the quantum numbers of the upper sublevels.
J = J + 1 J = J J = J − 1
3(J −M )(J +M +1) 3(J −M )(J −M −1)
M = M + 1
3(J +M +1)(J +M +2)
σb 2(J +1)(2J +1)(2J +3) 2J (J +1)(2J +1) 2J (2J −1)(2J +1)
3(J −M +1)(J +M +1) 3M 2 3(J −M )(J +M )
π M = M (J +1)(2J +1)(2J +3) J (J +1)(2J +1) J (2J −1)(2J +1)
3(J −M +1)(J −M +2) 3(J +M )(J −M +1) 3(J +M )(J +M −1)
σr M = M − 1 2(J +1)(2J +1)(2J +3) 2J (J +1)(2J +1) 2J (2J −1)(2J +1)
In the other cases (referred to as anomalous Zeeman effect ), more complicated
patterns occur. It will be proved later (see Sect. 9.1) that the relative strengths of
the various components are given by the expression
2
J J 1
SqJJ (M, M ) = 3 (q = −1, 0, +1) , (3.16)
−M M −q
where the index q = −(M − M ) = −∆M is equal to −1, 0, +1 for the σb , π, and
σr components, respectively. It can be easily seen that the relative strengths, as
defined by Eq. (3.16), are normalized to unity; from Eq. (2.23a) we have in fact
J J 1
2
SqJJ (M, M ) = 3 =1 (q = −1, 0, +1) . (3.17)
−M M −q
MM MM
Using the analytical expressions of the 3-j symbols (Eqs. (2.26)), the strengths of
the various components can be written as algebraic functions of M and M . The
results are summarized in Table 3.1.
From Eq. (3.16) one can prove some symmetry properties for the strengths. Equa-
tions (2.24) and (2.25) give
SqJJ (M, M ) = S−q
JJ
(−M, −M ) (3.18)
SqJJ (M, M ) = SqJ J (−M , −M ) . (3.19)
On the other hand, defining the wavelength displacement from line center via
Eq. (3.12),
MM = −∆λB (g M − gM ) ,
∆λJJ (3.20)
we also have
MM = −∆λ−M−M
∆λJJ JJ
(3.21)
∆λJJ J J
MM = ∆λ−M −M . (3.22)
Equations (3.18) and (3.21) show that for any σr (σb ) component there is a σb
(σr ) component of the same strength symmetrical about line center, and for any
82 CHAPTER 3
π component (except the one corresponding to M = M = 0, if it exists) there is
another π component of the same strength symmetrical about line center. From
Eqs. (3.19) and (3.22) we further deduce that the inversion of the upper and lower
level in a given transition yields the same combination of strengths and splittings
of the various components, or, in other words, the same Zeeman pattern; the
component connecting, in the original pattern, the lower sublevel (J, M ) with the
upper sublevel (J , M ) corresponds in the ‘inverted’ pattern to the component
connecting the lower sublevel (J , −M ) with the upper sublevel (J, −M ).
Some representative Zeeman patterns are shown in Fig. 3.2. An extensive numeri-
cal table of Zeeman patterns for a large number of electric-dipole atomic transitions
has been given by Beckers (1969b).1
3.2. Classical Theory of the Zeeman Effect
According to the classical theory of the electron, as developed by Lorentz in the
early years of this century, we schematize the emitting atom as a negative electric
charge (-e0 ) oscillating, at the frequency ν0 , around a point P under the action of
the motion of the
an elastic, restoring force. In the presence of a magnetic field B,
electron is described by the equation
d2 x e dx − γ dx ,
= −4π 2 ν02 x − 0 ×B (3.23)
dt2 mc dt dt
where γ is a constant which accounts for the damping of the electron due to its
energy loss by radiation.
The actual value of this constant, that is introduced here in a purely phenomeno-
logical way, can be deduced by equating the work per unit time done on the electron
by the friction force
2
dL dx
= mγ
dt dt
with the energy loss by radiation per unit time
2
2e2 d2 x
W = 30 .
3c dt2
By averaging these quantities over an oscillation period, and disregarding the slight
difference – due to the magnetic field – between the actual oscillation frequency
and ν0 , we have
8π 2 e20 2 8π πe20 1
γ= ν = . (3.24)
3 mc3 0 3 mc λ20
1
In Beckers’ table the Landé factors of the levels involved in the transitions are calculated
according to the L-S coupling scheme (Eq. (3.8)).
ATOMIC SPECTROSCOPY 83
Fig.3.2. Characteristic Zeeman patterns for different transitions. The Landé factors are computed
according to the L-S coupling scheme. Following the usual convention, the π components are
drawn upward and the σ components downward. The σb components in panels (c) and (f) are
dashed for clarity; the σr and σb components at line center in panel (c) are drawn somewhat
apart but they actually coincide. From left to right, top to bottom, we have patterns of Type 0
(a), Type II (b), Type III (c,d), and Type I (e,f) (see Sect. 3.3 for the definition of Type).
84 CHAPTER 3
To find the solution to Eq. (3.23) we introduce a real unit vector u0 parallel to
B and two real unit vectors ur and us in the plane perpendicular to B, in such a
way that (ur , us , u0 ) form, in this order, a right-handed coordinate system. From
ur and us we next define two complex unit vectors u±1 by the equation
1
u±1 = √ (∓ ur + i us ) . (3.25)
2
Expanding x on the basis uα (α = 0, ±1),
x = xα uα ,
α
and noting that
uα × u0 = −i α u
α , (3.26)
we obtain for xα the following decoupled equations
d2 xα dx dx
= −4π 2 ν02 xα + 4πi α νL α − γ α (α = 0, ±1) , (3.27)
dt2 dt dt
where νL is the Larmor frequency defined in Eq. (3.10). If we now look for a
solution of the form
−2πi να t
xα = Aα e ,
and observe that the quantities νL and γ are generally much smaller than the
frequency ν0 ,1 we obtain
γ
−2πi (ν0 −ανL ) t −2t
xα = Aα e e , (3.28)
where the constants Aα are to be determined from the initial conditions and depend
on the excitation mechanism of the classical dipole.
To determine the radiation emitted by our classical model atom we have just to
recall some important results from classical electrodynamics. It is well-known (see
e.g. Jackson, 1962) that an oscillating, monochromatic dipole p (t) produces in the
radiation zone an electromagnetic wave whose frequency is the same as that of the
dipole, and whose electric field is described by the equation
ikr , - 2 ikr
t) = k 2 e
E (r, Ω, × p (t) × Ω
Ω = k e p⊥ (t) , (3.29)
r r
where k is the wavenumber, r is the distance from the dipole, Ω is a unit vector in
p−Ω
the direction of propagation, and p⊥ = (Ω
·
p ) is the component of the dipole
1
For instance, for a spectral line at 5000 Å and a magnetic field of 103 G, we have (see
Eqs. (3.11) and (3.24))
νL 1.4 × 109 s−1 , γ 8.9 × 107 s−1 , ν0 6.0 × 1014 s−1 .
ATOMIC SPECTROSCOPY 85
Fig.3.3. Geometry for calculating the polarization properties of the radiation emitted by a classical
The unit vector e is the reference direction for defining the Stokes
dipole in a magnetic field B. 1
parameters.
in the plane perpendicular to Ω. From this equation it is possible to determine
the polarization of the electromagnetic wave emitted by a classical dipole in any
direction.
To define the Stokes parameters (see Sect. 1.6) we choose, in the plane perpen-
the reference direction unit vector e and the associated unit vector
dicular to Ω, 1
e2 , oriented as shown in Fig. 3.3. Since the oscillating dipole is p = −e0 x, the
components of the vector p⊥ along the directions e1 and e2 can be easily obtained
−2πi (ν0 −ανL ) t
γ
−2t
p⊥i (t) = pi (t) = −e0 Cαi Aα e e (i = 1, 2) , (3.30)
α
where the direction cosines Cαi are defined by1
Cαi = uα · ei ∗ . (3.31)
We must consider, however, that the dipole oscillation is not monochromatic. By
Fourier series expansion we get
∞
−2πiνt
pi (t) = −e0 Cαi Aα Fα (ν) e dν , (3.32)
α −∞
1
Indeed, in this formula, the direction cosines could be just defined as Cαi = uα · ei , since
the unit vectors e1 and e2 are real. However, Eq. (3.30) and the following ones up to Eq. (3.37)
are valid also when the unit vectors e1 , e2 are of the more general form of Eqs. (1.41).
86 CHAPTER 3
where
∞ γ
−2πi (ν0 −ανL −ν) t −2t i 1
Fα (ν) = e e dt = − (3.33)
2π (ν0 − ανL − ν) − iΓ
0
with Γ = γ/4π. Thus the component of the electric field along the unit vector ei
can be written in the form (as k = 2πν/c)
2 ∞
t) = − 4π e0
Ei (r, Ω, Cαi Aα Fα (ν) ν 2 e
i(kr−2πνt)
dν ,
rc2 α −∞
and hence
16π 4 e20 ∗
Ei∗ Ej = Cαi Cβj A∗α Aβ
r 2 c4
αβ
∞ ∞
i(k −k)r −2πi(ν −ν)t
× dν dν Fα (ν)∗ Fβ (ν ) ν 2 ν 2 e e .
−∞ −∞
To obtain the polarization tensor we must take the average of this quantity over a
time interval τ sufficiently long as specified in Sect. 1.4. This leads to
τ /2
−2πi(ν −ν)t 1 −2πi(ν −ν)t 1
e = lim e dt = δ(ν − ν ) ,
τ τ →∞ τ
−τ /2
with δ(x) the Dirac delta-function, so that (see Eq. (1.25))1
∞
16π 4 e20 ∗ A∗α Aβ
Jij (r) = 2 4 Cαi Cβj ν 4 Fα (ν)∗ Fβ (ν) dν . (3.34)
r c τ
αβ −∞
The quantities Aα depend on the initial conditions for the motion of the oscillator.
For random initial conditions (as one would have, for instance, if the oscillator
were collisionally excited by a population of perturbing particles with isotropically
distributed velocities), these quantities are uncorrelated with each other, so that,
performing a statistical average over the initial conditions, we can replace A∗α Aβ
by |A|2 δαβ . On the other hand, the quantity |A|2 is related to the average energy
Ē contained in each of the three degrees of freedom of the oscillator. In fact, since
1
Note that this formula contains the normalization time τ which is not related to any physical
quantity. We will see shortly (Eq. (3.35)) how this quantity disappears from the final results.
ATOMIC SPECTROSCOPY 87
for a classical oscillator the average kinetic energy is equal to the average potential
energy, writing Aα = |A| eiφ we have from Eq. (3.28)
2
Ē = 4π 2 ν02 m Re(xα ) = 4π 2 ν02 m |A|2 cos2 [2π(ν0 − ανL )t − φ] e−γt ,
and being the time interval τ much larger than the decay time γ −1 of the oscillator,
we can write
τ
1 1 −γt 2π 2 ν02 m
Ē = 4π 2 ν02 m |A|2 lim e dt = |A|2 . (3.35)
2 τ τ →∞ γτ
0
Assum-
Let us now introduce the emission coefficient in tensorial form, εij (ν, Ω).
ing N oscillators per unit volume, we have (see the analogous Eq. (1.31))
∞
dν dt dΩ = c
εij (ν, Ω) J (r) N dt r2 dΩ , (3.36)
8π ij
−∞
and using Eqs. (3.34) and (3.35) we get
2
= πe0 ν 2 γ N Ē
εij (ν, Ω) ∗
Cαi Cαj |Fα (ν)|2 ,
3 0
mc α
which via Eq. (3.33) can be written in the form
= πe20 ν2 ∗ 1 Γ
εij (ν, Ω) N 02 Ē Cαi Cαj . (3.37)
mc c α
π (ν0 − ανL − ν) + Γ
2 2
From Eqs. (1.39), defining the profiles φb , φp , φr (normalized to 1 in frequency) as
1 Γ
φb =
π (ν0 + νL − ν)2 + Γ 2
1 Γ
φp =
π (ν0 − ν)2 + Γ 2
1 Γ
φr = , (3.38)
π (ν0 − νL − ν)2 + Γ 2
and expressing the direction cosines for the geometry specified in Fig. 3.3,
1
C±1 1 = √ (∓ cos θ cos φ + i sin θ sin φ)
2
C0 1 = − sin θ
1
C±1 2 = √ (± sin φ + i cos φ)
2
C0 2 = 0 , (3.39)
88 CHAPTER 3
we finally obtain the emission coefficient in the four Stokes parameters
= ε (ν, Ω)
εI (ν, Ω) = k 1 φ sin2 θ + φb + φr (1 + cos2 θ)
+ ε (ν, Ω)
11 22
2 p 2
= ε (ν, Ω)
− ε (ν, Ω)
=k 1 φ + φ
εQ (ν, Ω) 11 22 φ − b r
sin2 θ
2 p 2
= ε (ν, Ω)
εU (ν, Ω) + ε (ν, Ω)
=0
12 21
, -
= i ε (ν, Ω)
εV (ν, Ω) = k 1 φ − φ cos θ ,
− ε (ν, Ω) (3.40)
12 21
2 r b
where
πe20 2ν 2
k= N 20 Ē .
mc c
These formulae show that our collection of randomly excited classical model
atoms emits along B (θ = 0) a radiation that is right-handed (or positively) circu-
larly polarized around ν = (ν0 − νL ) (in the ‘red’ component σr ) and left-handed
(or negatively) circularly polarized around ν = (ν0 + νL ) (in the ‘blue’ component
σb ). Obviously, the handedness of circular polarization changes when the direc-
tion is reversed (θ = π). For θ = π/2, on the contrary, the emitted radiation is
linearly polarized. The direction of linear polarization is parallel to the magnetic
field in the π component and perpendicular to the magnetic field in the σb and σr
components.
To sum up, we see that the classical theory provides a satisfactory explanation
for the normal Zeeman effect with respect both to the wavelength separation of the
Zeeman components and to their polarization characteristics. In the classical theory
the frequency separation of the components is found to be νL , which corresponds
to a Landé factor g = 1. This is quite obvious since g-values other than 1 are
produced by the spin (see Eq. (3.7)), that cannot be accounted for by a classical
theory.
We also want to remark that the results obtained for the polarization of the
various components depend on the negative sign that we have assumed for the
electric charge (see Eq. (3.23)). Assuming a positive charge for the electron, νL has
to be replaced by -νL in Eqs. (3.27) and following, and a further sign change must be
performed in Eq. (3.30). This last change is irrelevant, as the Stokes parameters
are quadratic functions of the dipole components, so that the net result is the
exchange of φr and φb in Eqs. (3.40), which leads to the opposite sign for εV (ν, Ω).
In other words, a sign inversion in the electron charge produces a sign inversion of
circular polarization. This also means that by measuring the handedness of circular
polarization in the various Zeeman components it is possible to determine the sign
of the electron charge.
This was just the procedure followed by Lorentz and Zeeman to give, for the
first time, the correct negative sign for the electron charge. It is curious to note,
however, that at first they deduced the wrong sign: hence the witty remark by Segrè
(1976) that ‘when signs are involved, even two Dutch physicists as scrupulous as
Lorentz and Zeeman may make errors’.
ATOMIC SPECTROSCOPY 89
3.3. Classification of Zeeman Patterns
The first parameter that can be used to characterize Zeeman patterns is the so-
called effective Landé factor ḡ,1 which represents the wavelength shift from line
center of the ‘center of gravity’ of the σr components in units of ∆λB (sometimes
also called Lorentz units). For an electric-dipole transition connecting a given level,
having angular momentum J1 and Landé factor g1 , with another level of angular
momentum J2 and Landé factor g2 , the value of ḡ can be easily derived from
Eqs. (3.12) and (3.16)
/ J1 J2 0
JJ λM M − λ0
1 2 1 2
ḡ = S1 (M1 , M2 )
∆λB
M1 M2
2
J2 J1 1
= 3 (g1 M1 − g2 M2 ) . (3.41)
−M2 M1 −1
M1 M2
This expression is valid both when the index 1 refers to the lower level and the
index 2 to the upper level, and in the opposite case. This is a consequence of the
symmetry properties discussed at the end of Sect. 3.1, and reflects the fact that
Zeeman patterns depend only on the quantum numbers of the levels involved in
the transition, and not on which of them is the lower or the upper.
Obviously, because of the symmetry properties just mentioned (Eqs. (3.18)-
(3.22)), the center of gravity of the σb components is -ḡ, while the center of gravity
of the π components is 0.
Equation (3.41) can be transformed into a simpler expression by means of some
Racah algebra. First we note that, with the help of Eq. (2.26d), we can write for
M1 and M2
J1 −M1 J1 J1 1
M1 = J1 (J1 + 1)(2J1 + 1) (−1) (3.42a)
M1 −M1 0
J2 J2 1
M2 = J2 (J2 + 1)(2J2 + 1) (−1)J2 −M2 . (3.42b)
M2 −M2 0
Next we observe that the term originating from M1 in Eq. (3.41) can be written,
using Eq. (2.42), in the form
J J1 1
2
J1 J1 1
2 (−1)J1 −M1 =
−M2 M1 −1 M1 −M1 0
M1 M2
J1 J1 1 J2 J1 1
= (−1)J1 −M1
M1 −M1 0 −M2 M1 −1
M1 M1 M2
J2 J1 1
× =
−M2 M1 −1
1
The quantity ḡ is often denoted by the symbol z in the literature concerning magnetic stars.
90 CHAPTER 3
!
1 1 1 1 1 1
= (−1)1+J1 +J2 . (3.43)
−1 1 0 J1 J1 J2
Performing the same transformation on the other term of Eq. (3.41), and taking
into account the analytical expressions of the 3-j and 6-j symbols (Eqs. (2.26d)
and (2.36d)), we obtain
1 1
ḡ = (g + g2 ) + (g1 − g2 ) d , (3.44)
2 1 4
where
d = J1 (J1 + 1) − J2 (J2 + 1) .
Note that this formula, originally due to Shenstone and Blair (1929), is invariant
under interchange of the indices 1 and 2. In most cases, Eq. (3.44) is employed
together with the Landé factors obtained from the L-S coupling scheme (Eq. (3.8)).
The resulting ḡ-value, that will be denoted in the following by ḡLS , can be easily
calculated for any given electric-dipole transition (see e.g. Beckers, 1969b). A more
appropriate value for ḡ can be obtained by substituting in Eq. (3.44) the values g1
and g2 deduced by spectroscopic measurements. In some cases, and especially for
lines belonging to complex spectra and involving levels of high excitation potential,
the L-S coupling scheme fails, so that the value ḡLS may be very different from ḡ.
Comparisons between ḡ and ḡLS values for several lines of the FeI and FeII spectra
have been published by Landi Degl’Innocenti (1982a) and by Solanki and Stenflo
(1985).
Typical ḡ values range from 0.5 to 2.5, but, for some transitions, ḡ can attain
null or even negative values, while for other transitions it can be larger than 2.5. A
list of transitions characterized by such ‘anomalous’ values of the effective Landé
factor is presented in Table 3.2. The list includes all the transitions between L-S
levels having S ≤ 7/2 and L ≤ 5, and subjected to the following limitations
∆J = 0, ±1 , ∆S = 0, ±1 , ∆L = 0, ±1, ±2 .
Lines having either exceptionally large values or null values for their effective Landé
factor are of great importance for the study of solar (and stellar) magnetism. Obvi-
ously, lines with large ḡ values produce strong polarization signals, and are partic-
ularly useful for the detection and measurement of magnetic fields. On the other
hand, lines with ḡ = 0 are especially useful for velocity measurements and for
calibration. A reduced list of lines with large ḡ-values has been given by Harvey
(1973), while Sistla and Harvey (1970) have presented a list of ḡ = 0 lines. A table
of close line pairs (less than 100 Å apart), with one line having a large ḡ value and
the other a small (or null) ḡ value, has been presented by Robinson (1980).
In some studies of stellar magnetism, it can be useful to compute an average
value for the effective Landé factor over the lines contained in a given statistical
sample or in an assigned spectral interval. For instance, considering the sample
of approximately 400 unblended lines of the FeI solar spectrum given by Stenflo
and Lindegren (1977), if the effective Landé factor of each line is weighted by the
ATOMIC SPECTROSCOPY 91
TABLE 3.2
Transitions having ‘anomalous’ values for the effective Landé factor ḡLS . The transitions listed
are those obeying the restrictions specified in the text.
Transition ḡLS Transition ḡLS Transition ḡLS
ḡLS < 0
4P
1 − 4F 3 −0.167 8D
3 − 6G 5 −0.6 8F
1 − 6G 3 −1.
2 2 2 2 2 2
5P
1 − 3 F2 −0.25 5F
1 − 7 G1 −0.25 8F
1 − 8H 3 −1.5
2 2
4D
1 − 6F
1 −0.333 6F
1 − 6F
1 −0.667 8F
3 − 6H
5 −1.
2 2 2 2 2 2
5D
0 − 7G
1 −0.5 6F
1 − 8G
1 −1. 8F
3 − 8H
5 −0.3
2 2 2 2
5D
1 − 5G
2 −0.25 6F
1 − 8H
3 −0.333 8F
5 − 6H
7 −0.286
2 2 2 2
6D
1 − 4F 3 −0.333 6F
3 − 6H 5 −0.3 6G
3 − 8H 3 −0.2
2 2 2 2 2 2
6D
1 − 6G 3 −0.833 6F
5 − 4H 7 −0.143 7G
1 − 7 G1 −0.5
2 2 2 2
6D
3 − 4G
5 −0.4 7F
0 − 7G
1 −0.5 8G
1 − 8G
1 −1.333
2 2 2 2
7D
1 − 5G
2 −1. 7F
1 − 5G
2 −0.25 8G
1 − 8H
3 −0.167
2 2
1 − G2 −0.25 − 7 H2 −0.75 − 6H 5 −0.2
7D 7 7F 8G
1 3
2 2
7D
2 − 5 G3 −0.167 7F
2 − 5 H3 −0.5 8H
3 − 8H 3 −0.4
2 2
ḡLS = 0
3P
0 − 5 F1 7D
1 − 5 F2 7F
2 − 7 H3
6P
3 − 4F 5 8D
5 − 6G 7 7F
3 − 5 H4
2 2 2 2
4D
1 − 4D 1 5F
1 − 5 F1 8F
3 − 6G 5
2 2 2 2
4D
1 − 6G
3
5F
1 − 7F
0
6G
3 − 6G 3
2 2 2 2
5D
0 − 5 F1 5F
1 − 7 H2 7H
2 − 7 H2
5D
2 − 3 G3 5F
2 − 5 H3
ḡLS ≥ 3
4P
1 − 6D 1 3. 6D
1 − 8F 1 3.667 8D
5 − 6G 3 3.6
2 2 2 2 2 2
6P
3 − 4D
1 3. 7D
1 − 7D
1 3. 8D
7 − 6G
5 3.
2 2 2 2
6P
3 − 6F
1 3.167 7D
1 − 7F
0 3. 7F
3 − 7H
2 3.
2 2
6P
5 − 4F
3 3. 7D
2 − 5F
1 3. 7F
4 − 5H
3 3.
2 2
7P
2 − 5 F1 3.5 7D
2 − G1
7 3.25 8F
1 − 8F 1 4.
2 2
8P
5 − 6F
3 3.2 7D
3 − 5G
2 3.167 8F
5 − 6G
3 3.
2 2 2 2
5D
0 − 7D
1 3. 8D
3 − 6F
1 3.667 8F
5 − 8H
3 3.3
2 2 2 2
6D
1 − 6D
1 3.333 8D
3 − 8G
1 3.833 8F
7 − 6H
5 3.286
2 2 2 2 2 2
92 CHAPTER 3
central depression observed for the same line, one finds the value ḡLS = 1.22
(Landi Degl’Innocenti, 1985a).1
Besides the effective Landé factor, higher order moments can be introduced to
characterize Zeeman patterns. For each type of component we define the barycen-
(n)
tric n-th order moment vq to be
JJ J J2 n
vq(n) (J1 , J2 ) = Sq 1 2 (M1 , M2 ) ∆λM1 M − q ḡ ∆λB , (3.45)
1 2
M1 M2
J J J J
where ∆λM1 M 2
and Sq 1 2 (M1 , M2 ) are defined in Eqs. (3.20) and (3.16), and where
1 2
q = −1, 0, +1 for the σb , π, and σr components, respectively.
Using the symmetry properties derived in Sect. 3.1 (Eqs. (3.18)-(3.22)), and
recalling that ḡ is invariant under interchange of J1 and J2 , one easily gets
vq(n) (J2 , J1 ) = vq(n) (J1 , J2 ) (3.46)
(n)
v−q (J1 , J2 ) = (−1)n vq(n) (J1 , J2 ) . (3.47)
Equation (3.46) shows that the various moments are not altered by the interchange
of the lower and the upper level – an obvious consequence of the same property
holding for the Zeeman patterns. Equation (3.47) shows that the odd-order mo-
ments of the π components identically vanish, while, for the σ components, the
even-order moments of the σb and σr components are the same, and the odd-order
moments have opposite sign. Obviously, all the even-order moments are positive
quantities.
We will now show that the barycentric moments defined in Eq. (3.45) can be
reduced to a simple form which contains 3-j and 6-j symbols only. Substitution of
Eqs. (3.20) and (3.44) into Eq. (3.45) leads after some algebra to the expression
vq(n) (J1 , J2 ) = ∆λnB (g1 − g2 )n Xq(n) (J1 , J2 ) , (3.48)
with
d+2
n
J J
Xq(n) (J1 , J2 ) = M1 − q Sq 1 2 (M1 , M2 )
4
M1 M2
n n−i
n d+2
= (−1)n−i q Zq(i) (J1 , J2 ) , (3.49)
i=0
i 4
(i)
where Zq (J1 , J2 ) is the i-th moment of M1 weighted by the strength of the q-
transition originating from M1 ,
J J
Zq(i) (J1 , J2 ) = M1i Sq 1 2 (M1 , M2 ) . (3.50)
M1 M2
1
Owing both to the L-S assumption and to the procedure itself by which it is obtained, this
value has only an approximate and statistical meaning.
ATOMIC SPECTROSCOPY 93
(i)
The quantities Zq (J1 , J2 ) can be evaluated by some Racah algebra. For this
purpose we note that the α-component of an angular momentum a can be written
in the form (see Eqs. (3.42))
a−α
a a 1
α = (−1) a(a + 1)(2a + 1) . (3.51)
α −α 0
Squaring this equation we get
a a 1 a a 1
α2 = a(a + 1)(2a + 1) ,
α −α 0 α −α 0
or, with the help of Eq. (2.43)
α2 = (−1)a+α a(a + 1)(2a + 1)
!
1 1 f 1 1 f a a f
× (2f + 1) , (3.52)
a a a 0 0 0 α −α 0
f
where the sum over f runs over the values 0 and 2 only (for f = 1 the first 3-j
symbol in the right-hand side is zero).
Repeated application of the same procedure gives the following remarkable for-
mula
n/2
αn = (−1)(2n−3)a+α a(a + 1)(2a + 1)
× (2f1 + 1)(2f2 + 1) · · · (2fn−1 + 1)
f1 f2 ···fn−1
! ! !
1 1 f1 1 f1 f2 1 fn−2 fn−1
× ···
a a a a a a a a a
1 1 f1 1 f1 f2 1 fn−2 fn−1
× ···
0 0 0 0 0 0 0 0 0
a a fn−1
× (n = 2, 3, . . .) .
α −α 0
Owing to the property (2.24) of the 3-j symbols, the indices f1 , f2 ,. . . are alternately
even and odd numbers. Moreover, they are chained in such a way that the allowed
values for fi+1 are (fi ± 1), starting from f1 = 0 and f1 = 2. For example, if n = 4
the chains of allowed values for f1 , f2 , and f3 are: (0,1,0), (0,1,2), (2,1,0), (2,1,2),
(2,3,2), (2, 3, 4).
We now use the above expansion to express the quantity M1i in Eq. (3.50), and
substitute the explicit expression for the strengths given by Eq. (3.16). Performing
94 CHAPTER 3
the sum over M1 and M2 by the same method that leads to Eq. (3.43), we obtain,
for i ≥ 2
Zq(i) (J1 , J2 ) =
i/2
= (−1)(2i−3)J1 −J2 +q 3 J1 (J1 + 1)(2J1 + 1)
× (2f1 + 1)(2f2 + 1) · · · (2fi−1 + 1)
f1 f2 ···fi−1
! ! !
1 f1 1 1 f1 f2 1 fi−2 fi−1
× ···
J1 J1 J1 J1 J1 J1 J1 J1 J1
1 1 f1 1 f1 f2 1 fi−2 fi−1
× ···
0 0 0 0 0 0 0 0 0
!
1 1 fi−1 1 1 fi−1
× ,
−q q 0 J1 J1 J2
while for i = 0 and i = 1 we have
Zq(0) (J1 , J2 ) = 1
Zq(1) (J1 , J2 ) = (−1)J1 +J2 +q 3 J1 (J1 + 1)(2J1 + 1)
!
1 1 1 1 1 1
× .
−q q 0 J1 J1 J2
These formulae, together with Eqs. (3.49) and (3.48), give the required expres-
sions for the barycentric moments. Using Eqs. (2.26) and (2.36) one can indeed
obtain simple analytical formulae for the low-order moments. The results, that
are contained in Table 3.3, show the characteristic symmetrical behavior of the
(n)
quantities Xq (J1 , J2 ) under interchange of J1 with J2 and of q with −q.
Besides the barycentric moments of the components, it is also convenient to
introduce the moments tout court. These moments are defined by
J J J J2 n
wq(n) (J1 , J2 ) = Sq 1 2 (M1 , M2 ) ∆λM1 M , (3.53)
1 2
M1 M2
and it can be easily seen that they satisfy the same symmetry properties found for
the barycentric moments (Eqs. (3.46) and (3.47)).
The relations between the two types of moments are the following
n
n n−i
vq(n) (J1 , J2 ) = (−1) n−i
wq(i) (J1 , J2 ) q ḡ ∆λB
i=0
i
n
n n−i
wq(n) (J1 , J2 ) = vq(i) (J1 , J2 ) q ḡ ∆λB .
i=0
i
ATOMIC SPECTROSCOPY 95
TABLE 3.3
Analytical expressions for the low-order barycentric moments of Zeeman components. (∗) The
expressions have been simplified taking into account that, for electric dipole transitions, the
quantity [d(2s − d2 )] is identically zero.
(0) (0)
Zq =1 Xq =1
(1) (1)
Z0 =0 X0 =0
(1) (1)
Z±1 = ± 1
4
(d + 2) X±1 = 0
(2) (2)
Z0 = 1
10
(−d2 + 3s − 2) X0 = 1
10
(−d2 + 3s − 2)
(2) (2)
Z±1 = 1
20
(d2 + 5d + 2s + 2) X±1 = 1
80
(−d2 + 8s − 12)
(3) (3)
Z0 =0 X0 =0
(3) (3)
Z±1 = ± 1
80
(3d3 + 6d2 + 12s + 8d − 8) X±1 = ± 1
160
d(4 − d2 )
(4) (4)
Z0 = 1
140
(−3d4 + 7d2 + 15s2 − 30s + 20) X0 = 1
140
(−3d4 + 7d2 + 15s2 − 30s + 20)
(4) (4)
Z±1 = 1
280
(3d4 + 21d3 + 14d2 + 6s2 + 16s X±1 = 1
8960
(−9d4 + 56d2 + 192s2 − 832s
−14d − 20) +816)
(∗)
s = [J1 (J1 + 1) + J2 (J2 + 1)] , d = [J1 (J1 + 1) − J2 (J2 + 1)]
(n)
After some algebraic manipulations, wq can be expressed in the form
wq(n) (J1 , J2 ) = ∆λnB G(n)
q (J1 , J2 ) , (3.54)
where
n
n
G(n)
q (J1 , J2 ) = (g1 − g2 )i (g2 q)n−i Zq(i) (J1 , J2 ) . (3.55)
i=0
i
(n)
Analytical formulae for the quantities Gq (J1 , J2 ) up to n = 4 are contained in
Table 3.4.
(n) (n)
An alternative derivation of the quantities vq and wq in terms of Bernoulli
polynomials has been given by Mathys and Stenflo (1987a,b), who employ the
(n) (q)
notations µq and n! (−∆λB )n Cn , respectively. In their second paper (1987b),
(n) (q)
these authors give also extensive numerical tables for µq and Cn ; these last
quantities are however evaluated, for any given transition, by assuming for the
Landé factors the values deduced from the L-S coupling scheme.
Zeeman patterns can be classified into four different types according to the value
of their barycentric moments, with particular emphasis on the third-order mo-
ment v1(3) .
First of all, we will call Type 0 the normal and anomalous Zeeman triplets (see
(n)
Sect. 3.1): obviously, these patterns are characterized by vq = 0 for n ≥ 1. A
complete list of transitions giving rise to Type 0 patterns has been published by
Mathys and Stenflo (1987b).
96 CHAPTER 3
TABLE 3.4
(∗)
Analytical expressions for the low-order moments of Zeeman components
(0)
Gq =1
(1)
G0 =0
(1)
G±1 = ± ( 12 gs + 14 dgd ) = ± ḡ
(2)
G0 = 1
10
(3s − d2 − 2)gd2
(2)
G±1 = 1
20
(d2 + 2s − 3)gd2 + 5dgs gd + 5gs2
(3)
G0 =0
(3)
G±1 = ± 1
80
(3d2 − 7)dgd3 + 6(d2 + 2s − 3)gd2 gs + 15dgd gs2 + 10gs3
(4)
G0 = 1
140
(−3d4 + 7d2 + 15s2 − 30s + 20)gd4
(4)
G±1 = 1
560
(6d4 − 14d2 + 12s2 − 52s + 51)gd4 + 14(3d2 − 7)dgd3 gs + 42(d2 + 2s − 3)gd2 gs2
+70dgd gs3 + 35gs4
(∗)
s = [J1 (J1 + 1) + J2 (J2 + 1)] , d = [J1 (J1 + 1) − J2 (J2 + 1)] , gs = (g1 + g2 ) , gd = (g1 − g2 )
For anomalous Zeeman patterns, we will call, following Back and Landé (1925),
Type I those having v1(3) > 0, Type II those having v1(3) < 0, and Type III those
with v1(3) = 0. As it can be argued from the definition of the barycentric moments
(Eq. (3.45)), patterns of Type I have their strongest σr component toward the blue
side of the spectrum, while patterns of Type II have their strongest σr component
toward the red side of the spectrum. Finally, Type III patterns have σ components
symmetrical about their center of gravity. Some examples of Zeeman patterns of
different types are shown in Fig. 3.2.
From the analytical expressions given in Table 3.3 and from Eq. (3.48) we have
1
v1(3) = ∆λ3B (g1 − g2 )3 d (4 − d2 ) .
160
This formula shows that Type III patterns can only originate when d = 0, or, in
other words, when J1 = J2 . By contrast, Type I and Type II patterns correspond
to transitions having ∆J = ±1. In particular, Type I patterns arise when the
Landé factor of the level having the smaller J is larger than the Landé factor of the
level having the larger J, while the opposite situation leads to Type II patterns.
It should be remarked that the classification of Zeeman patterns, although based
on the σ components, characterizes at the same time the behavior of the π com-
ponents. Indeed, with increasing distance from line center the strengths of the π
components decrease (quadratically) for Type I and Type II patterns, while they
increase (also quadratically) for Type III patterns. This is apparent from Table 3.1
and Eq. (3.20).
ATOMIC SPECTROSCOPY 97
Fig.3.4. Definition of the minimal equivalent Zeeman pattern.
It should also be mentioned that a useful concept for certain applications is that
of minimal equivalent Zeeman pattern. Given an arbitrary Zeeman pattern, we
define its minimal equivalent as the pattern having only two π components and
two σ components (both for the blue and the red wing) with such splittings and
strengths that the barycentric moments up to n = 3 of the original pattern are
reproduced. The minimal equivalent Zeeman pattern is sketched in Fig. 3.4. It can
be shown that the strengths Si and the splittings zi (expressed in units of ∆λB ,
and satisfying the condition z2 > z1 ) are given by
1
z0 = |gd | X0(2) S0 =
2
1 1 z
z1 = ḡ − |g | X1(2) 4 + z2 − z S1 = 1+ √
2 d 2 4 + z2
1 1 z
z2 = ḡ + |gd | X1(2) 4 + z2 + z S2 = 1− √ ,
2 2 4 + z2
where
gd X1(3)
|gd | X (2) 3/2
z= .
1
Note that the value of z remains undefined when X1(2) = 0; this is a degenerate
case corresponding either to a Zeeman triplet (Type 0 pattern) or to the transition
1/2 → 1/2. In both cases the two σ components merge into a single one with
z1 = z2 = ḡ, (S1 + S2 ) = 1.
3.4. The Paschen-Back Effect
As already stated in Sect. 3.1, when the magnetic field is so strong as to produce
on a J-level a splitting comparable with the energy separation between different
98 CHAPTER 3
J-levels, the perturbation theory presented for the Zeeman effect cannot be applied
any longer. In this regime, that is called the Paschen-Back effect regime, the energy
levels have to be found by diagonalization of the total Hamiltonian H given by
H = H0 + HB ,
where H0 and HB are the unperturbed and magnetic Hamiltonians, respectively,
with HB given by Eq. (3.3).
The diagonalization of the Hamiltonian H can be performed on the basis of the
eigenvectors |αJM of H0 (see Eqs. (3.4)). This implies the evaluation of matrix
elements of the form
αJM |HB |α J M = µ0 αJM |(L
+ 2S |α J M .
)·B
A calculation similar to that of Sect. 3.1 leads to the expression
J +M+q+1
√ J J 1
αJM |HB |α J M = µ0 B (−1) 2J + 1
−M M q
q
× J(J + 1) δαα δJJ + αJ α J
S b−q ,
and, in particular, when the quantization axis for J is in the magnetic field direction
αJM |HB |α J M =
= µ0 B M δαα δJJ
√ J J 1
+ (−1)J +M+1
2J + 1 α J
αJ S δMM . (3.56)
−M M 0
Equation (3.56) shows that the magnetic Hamiltonian HB connects only eigenstates
having the same value of the magnetic quantum number M : in other words, it is
block-diagonal, each block being characterized by an assigned M -value.
The general problem of finding the eigenvalues and the eigenvectors of the matrix
H can be solved only by numerical methods. There are, however, some special cases
where the analytical calculations can be pushed somewhat further. If we suppose,
for instance, that the atomic system is exactly described by the L-S coupling
scheme, the eigenstates of H0 are of the form |βLSJM (where β summarizes the
electronic configuration quantum numbers), so that Eq. (3.56) becomes
βLSJM |HB |β L S J M =
= µ0 B M δββ δLL δSS δJJ
J +M+1
√ J J 1
+ (−1) 2J + 1 β L S J
βLSJ S δMM ,
−M M 0
ATOMIC SPECTROSCOPY 99
and using Eqs. (2.109) and (2.97)
βLSJM |HB |β L S J M = µ0 B δββ δLL δSS δMM
× M δJJ + (−1)J+J +L+S+M (2J + 1)(2J + 1)S(S + 1)(2S + 1)
!
J J 1 J J 1
× . (3.57)
−M M 0 S S L
From this expression we obtain the important result that in the L-S coupling
scheme the magnetic Hamiltonian is block-diagonal, each block being now charac-
terized by the magnetic quantum number M and by the quantum numbers (β, L, S)
specifying the term. In other words, the magnetic field produces a J-mixing of the
various levels belonging to a particular term, so that the eigenvectors of the total
Hamiltonian will be of the form
j
|βLSjM = CJ (βLS, M ) |βLSJM , (3.58)
J
where the index j labels the different states of the N -fold degenerate subspace
corresponding to assigned values of (β, L, S, M ); N is given by
N = 1 + L + S − max(|M |, |L − S|) .
Of course, in contrast with the Zeeman effect regime, J is not a good quantum
number.
We now recall that in the L-S coupling scheme the energies EβLS (J) of the
different J-levels originating from a given term (β, L, S) are determined by the
spin-orbit Hamiltonian Hso (included in H0 ), which in most cases can be written
in the form (see Condon and Shortley, 1935)
·S
Hso = ζ(βLS) L , (3.59)
·S
where ζ is a constant having the dimensions of energy. Writing L in the form
= 1 J 2 − L2 − S 2 ,
·S
L
2
we have1
EβLS (J) = βLSJM |Hso |βLSJM
1
= ζ(βLS) J(J + 1) − L(L + 1) − S(S + 1) . (3.60)
2
1
Note in passing that the energy difference between two adjacent J-levels is given by
EβLS (J) − EβLS (J − 1) = ζ(βLS) J ,
which is the well-known Landé interval rule.
100 CHAPTER 3
Thus the whole problem of finding the energy levels of an atom described by the
L-S coupling scheme, embedded in a magnetic field, reduces to the diagonalization
of a set of matrices of the form
βLSJM |Hso + HB |βLSJ M .
From Eq. (3.57), using the analytical expressions of the 3-j and 6-j symbols
(Eqs. (2.26) and (2.36)), we have that the only non-zero matrix elements are those
of the form
βLSJM |Hso + HB |βLSJM = EβLS (J) + µ0 B gLS (J) M (3.61a)
βLS J −1 M |Hso + HB |βLSJM = βLSJM |Hso + HB |βLS J −1 M
µ0 B (J +S +L+1)(J −S +L)(J +S −L)(−J +S +L+1)(J 2 −M 2 )
=− , (3.61b)
2J (2J +1)(2J −1)
where gLS (J) is the Landé factor defined in Eq. (3.8).
Diagonalization of the matrix (3.61) (which is seen to be tridiagonal, real, and
symmetric) gives the energy eigenvalues λj (βLS, M ) and the corresponding eigen-
vectors expressed in terms of the coefficients CJj (βLS, M ) of Eq. (3.58). Since the
matrix is real and symmetric, the coefficients CJj can be chosen to be real. More-
over, from general theorems concerning the diagonalization of matrices one can
prove the relations
CJj (βLS, M ) CJj (βLS, M ) = δjj (3.62a)
J
CJj (βLS, M ) CJj (βLS, M ) = δJJ (3.62b)
j
CJj (βLS, M ) CJj (βLS, M )λj (βLS, M ) =
j
= βLSJM |Hso + HB |βLSJ M . (3.62c)
The diagonalization of matrix (3.61) can be performed analytically for doublet
terms only (S = 1/2). The energy eigenvalues are found to be
1 L+1
L ζ + µ0 B
λ1 = M
2 L + 12
for M = ±(L + 1/2), and
2
1 1 1
λ1,2 = − ζ + µ0 B M ± ζ2 L+ + 2ζµ0 BM + (µ0 B)2
4 2 2
ATOMIC SPECTROSCOPY 101
Fig.3.5. The energy levels of the term 2 P plotted as functions of the magnetic field strength. The
energy E (vertical scale) and the magnetic energy µ0 B (horizontal scale) are both normalized to
the fine-structure energy ζ (supposed positive). For an inverted multiplet (ζ < 0) the graph is
simply reversed about the E = 0 axis.
for M = ±(L + 1/2). In all the other cases, the diagonalization can be performed
only by numerical methods.
Figures 3.5 to 3.9 illustrate the behavior of the energy eigenvalues of the most
common L-S terms as a function of the magnetic field strength, parameterized
through the quantity γ defined by
µ0 B
γ= . (3.63)
ζ
For γ 1 the eigenvalues spread out linearly from their degenerate, zero-magnetic
field value (Zeeman effect regime). For intermediate values of γ, the eigenvalues
start crossing each other and the linearity with the magnetic field is, in general, lost
(incomplete Paschen-Back effect regime); finally, for large values of γ (γ 1), the
eigenvalues behave again linearly with γ, as shown in Fig. 3.10 (complete Paschen-
Back effect regime).
In this last regime, the spin-orbit interaction can be considered as a perturbation
in comparison with the magnetic interaction. On the other hand, the magnetic
Hamiltonian is diagonal on the basis |βLSMLMS , and the eigenvalues are given
by
βLSML MS |HB |βLSML MS = µ0 B (ML + 2MS ) ,
which explains their linear behavior with B for γ 1. As the magnetic field
increases from γ 1 to γ 1, the energy eigenvectors gradually evolve from
102 CHAPTER 3
Fig.3.6. Same as Fig.3.5 for the term 2 D.
Fig.3.7. Same as Fig.3.5 for the term 2 F.
ATOMIC SPECTROSCOPY 103
Fig.3.8. Same as Fig.3.5 for the term 3 P.
Fig.3.9. Same as Fig.3.5 for the term 3 D.
104 CHAPTER 3
Fig.3.10. For large values of the magnetic field, the energy eigenvalues spread out linearly. This
figure is just an enlargement of Fig.3.5 and is intended to show the behavior of the eigenvalues
for γ 1.
the form |βLSJM to the form |βLSML MS . In other words, the magnetic field
induces a gradual basis-transformation on the energy eigenstates.
As apparent from Figs. 3.5-3.9, the incomplete Paschen-Back effect regime is
characterized by several ‘crossings’ of the energy eigenvalues, each corresponding
to a well-defined value of the γ parameter. It will be shown in Sect. 10.18 that
such level-crossings induce important phenomena in resonance scattering. Table 3.5
gives, for the most common L-S terms, the values of γ relative to each level-crossing,
and the corresponding M -values of the levels involved. To illustrate how this table
should be used, consider the 3p 2 P term of NaI: from the tables of Moore (1949)
we get a value for the constant ζ equal to 11.4642 cm−1 , whence we deduce that
level-crossing takes place for B = 1.64 × 105 G and for B = 2.46 × 105 G.
In some cases, and particularly for light atoms such as He, the spin-orbit inter-
action cannot be described by the simple expression given in Eq. (3.59). In these
cases the energy levels in the presence of a magnetic field can still be obtained by
diagonalizing the matrix given in Eqs. (3.61), but the actual values EβLS (J) are
now to be found in tables of atomic energy levels (see e.g. Moore, 1949, 1952, 1958)
or in the specialized literature.
Let’s now consider how the splitting of the energy levels due to the magnetic
field affects, in the general case of the Paschen-Back regime, the spectral lines
originating in the transition between two given terms. The main properties of the
Paschen-Back patterns resulting from the electric-dipole interaction are described
ATOMIC SPECTROSCOPY 105
TABLE 3.5
Level-crossings for the most common L-S terms. The magnetic field is expressed through the
parameter γ defined in the text. The M -values refer to a normal multiplet (ζ > 0). For inverted
multiplets they must be changed in sign.
Term γ M M Term γ M M
2P 0.667 − 32 1
2
3P 0.45 −2 1
1. − 32 − 12 0.616 −2 0
2D 0.6 − 52 3
2
0.707 −1 1
0.75 − 52 1
2
0.859 −1 0
0.866 − 32 3
2
1.185 −1 0
1. − 52 −2 1
1.217 −2 0
1.257 − 32 1
2
1.5 −2 −1
1.5 − 52 −2 3
2.210 0 1
1.591 − 12 3
2
3D 0.476 −3 2
3. − 32 − 12 0.576 −3 1
2F 0.571 − 72 5
2
0.611 −1 2
0.667 − 72 3
2
0.724 −2 1
0.707 − 52 5
2
0.730 −3 0
0.8 − 72 1
2
0.786 −2 1
0.862 − 52 3
2
0.856 −1 2
0.928 − 32 5
2
0.930 −2 0
1. − 72 − 12 1. −3 −1
1.106 − 52 1
2
1.100 −2 0
1.225 − 32 3
2
1.157 −3 1
1.333 − 72 − 32 1.253 −1 1
1.356 − 12 5
2
1.287 −2 −1
1.562 − 52 − 12 1.370 −3 0
1.840 − 32 1
2
1.463 0 2
2. − 72 −2 5
1.667 −3 −1
2.174 − 12 3
2
1.760 −1 1
2.562 1
2
5
2
1.797 −2 1
3. − 52 − 32 1.813 −2 −1
6. − 32 − 12 2. −3 −2
2.5 −1 0
2.600 −2 0
4.773 0 1
106 CHAPTER 3
below.
First of all, since the projection M of the total angular momentum on the mag-
netic field direction is a good quantum number, the individual components can be
characterized by the value ∆M = (M − M ), where M refers to the lower level and
M to the upper one. Owing to the selection rule (3.15) we thus obtain that – in
strict analogy with the Zeeman effect – the components can be divided into three
distinct groups: σr (∆M = −1), π (∆M = 0), and σb (∆M = +1).
As far as the strengths of the components are concerned, it can be shown that
they are proportional to the quantities
M
SqjM,j = | βLSjM |rq |β L S j M |2 (q = −1, 0, +1) ,
where the unprimed quantum numbers refer to the lower level, the primed ones to
the upper level, and where rq are the spherical components of the position operator
of the optical electrons.
Expanding the eigenvectors as in Eq. (3.58), and taking into account that the
coefficients CJj are real, one obtains
M
SqjM,j = CJj (βLS, M ) CJj (βLS, M ) CJj (β L S , M ) CJj (β L S , M )
JJ J J
∗
× βLSJM |rq |β L S J M βLSJ M |rq |β L S J M .
Evaluating the matrix elements through the Wigner-Eckart theorem (Eq. (2.96)),
and noting that rq is an operator which acts only on the orbital part of the eigen-
vector, one gets from Eq. (2.108)
SqjM,j M =
= CJj (βLS, M ) CJj (βLS, M ) CJj (β L S, M ) CJj (β L S, M )
JJ J J
× (2L + 1) (2J + 1)(2J + 1)(2J + 1)(2J + 1)
! !
L L 1 L L 1
×
J J S J J S
J J 1 J J 1
× | βL r β L |2 δSS ,
−M M −q −M M −q
a formula that contains all the relevant selection rules, namely
∆L = 0, ±1 L = 0 L = 0
∆S = 0
∆M = 0, ±1 .
ATOMIC SPECTROSCOPY 107
Summation over all the possible transitions gives (see Eq. (3.62b))
!2
jM,j M L L 1
Sq = (2L + 1)(2J + 1)(2J + 1)
J J S
jj MM JJ MM
2
J J 1
× | βL r β L |2 ,
−M M −q
which can be reduced, using Eqs. (2.23a) and (2.39), to
1
M
SqjM,j = (2L + 1)(2S + 1) | βL r β L |2 .
3
jj MM
This property makes it possible to define the normalized strengths in the form
SqjM,j M =
3
= C j (βLS, M ) CJj (βLS, M ) CJj (β L S, M ) CJj (β L S, M )
2S + 1 J
JJ J J
× (2J + 1)(2J + 1)(2J + 1)(2J + 1)
! !
L L 1 L L 1
×
J J S J J S
J J 1 J J 1
× , (3.64)
−M M −q −M M −q
with
M
SqjM,j =1 (q = −1, 0, +1) .
jj MM
The strengths as defined by Eq. (3.64) have some important properties:
a) In the absence of magnetic fields, J is a good quantum number, and the eigenvec-
tors |βLSjM converge towards the states |βLSJM which are degenerate with
respect to M . For the coefficients CJj we have
CJj (βLS, M ) = δjJ ,
so that
!2 2
M 3 L L 1 J J 1
SqJM,J = (2J + 1)(2J + 1) ,
2S + 1 J J S −M M −q
and summing over the degeneracy parameters M and M we obtain the usual
expression for the relative strengths of a fine-structure multiplet
!2
J,J JM,J M (2J + 1)(2J + 1) L L 1
S = Sq = . (3.65)
2S + 1 J J S
MM
108 CHAPTER 3
b) The centers of gravity in frequency of the π, σr , and σb components (q = 0, +1,
and −1, respectively) are linear functions of the magnetic field
∆νq = −q νL , (3.66)
where νL is the Larmor frequency and ∆νq is defined by
M jj
∆νq = SqjM,j ∆νMM , (3.67)
jj MM
jj
with ∆νMM the frequency shift of the component corresponding to the transition
between the lower state |βLSjM and the upper state |β L Sj M from the
unperturbed frequency of the transition between the two terms (βLS) and (β L S).
To prove Eq. (3.66) we substitute into Eq. (3.67) the strengths of Eq. (3.64) and
the frequency shifts
jj λj (β L S, M ) − λj (βLS, M )
∆νMM = , (3.68)
h
where λj , λj are the eigenvalues of the matrix (3.61) for the lower and the upper
term, respectively. We then perform the sums over j and j using Eqs. (3.62b) and
(3.62c), to obtain
1 3
∆νq = (2J + 1)(2J + 1)(2J + 1)(2J + 1)
h 2S + 1
MM JJ J J
! !
L L 1 L L 1
×
J J S J J S
J J 1 J J 1
×
−M M −q −M M −q
× δJJ β L SJ M |Hso + HB |β L SJ M
− δJ J βLSJM |Hso + HB |βLSJ M . (3.69)
It is easy to prove that the spin-orbit Hamiltonian gives zero contribution. Indeed,
noting that this Hamiltonian is diagonal with respect to the quantum numbers J
and M , and that its matrix elements are independent of M , and performing the
sums over M and M via Eq. (2.23a), we have for this contribution
!2
1 (2J + 1)(2J + 1) L L 1
∆νq so =
h (2S + 1) J J S
JJ
× β L SJ M |Hso |β L SJ M − βLSJM |Hso |βLSJM .
ATOMIC SPECTROSCOPY 109
Summation over J for the first term, and over J for the second term gives, with
the use of Eq. (2.39)
1 2J + 1
∆νq so = β L SJ M |Hso |β L SJ M
h (2L + 1)(2S + 1)
J
2J + 1
− βLSJM |Hso |βLSJM .
(2L + 1)(2S + 1)
J
The first sum is simply the average energy of the fine-structure levels of the upper
term, while the second sum is the same quantity for the lower term. It can be
easily proved that both these quantities are identically zero when the spin-orbit
Hamiltonian is described by Eq. (3.59).1
The only non-vanishing contribution to Eq. (3.69) is due to the magnetic field. To
evaluate this contribution, we substitute the matrix elements given by Eq. (3.57)
and perform the sums over M and M by means of Eq. (2.42). After some tedious
Racah-algebra calculations which involve the use of Eqs. (2.38), (2.48), (2.55), and
(3.51), we obtain that only the first term in the square bracket of the matrix
element (3.57) gives a non-zero contribution, and we get the expression
!2
3 νL L L 1 1 1 1
∆νq = (−1)J+J +q (2J + 1)(2J + 1)
2S + 1 J J S q −q 0
JJ
!
1 1 1
× J (J + 1)(2J + 1)
J J J
!
1 1 1
+ J(J + 1)(2J + 1) .
J J J
Evaluating the 6-j symbols in the square bracket by Eq. (2.36d) and the 3-j symbol
by Eq. (2.26d), and performing the sums over J and J via Eq. (2.39), we finally
obtain Eq. (3.66).
The theorem just proved shows an interesting property of the Paschen-Back effect
which can be summarized by the following statement: The frequency shifts of the
centers of gravity of the σr , σb , and π components in the Paschen-Back effect regime
are those typical of the normal Zeeman effect between two spinless levels. In other
words, given a fine-structured lower term (βLS) and a fine-structured upper term
1
To prove this statement, one has just to recall the sum rules for integers
%n 1
%n 1
%n 1
k=1
k= 2
n (n + 1), k=1
k2 = 6
n (n + 1)(2n + 1), k=1
k3 = 4
n2 (n + 1)2 .
If, on the other hand, the spin-orbit Hamiltonian cannot be written in the form (3.59), the quantity
(∆νq )so may indeed differ from zero. In such cases, it represents a constant shift (independent
of q) that affects the unperturbed frequency of the transition. In other words, Eq. (3.66) still
holds provided Eq. (3.68) is replaced by
jj 1
∆νM M
= λ (β L S, M ) − λj (βLS, M ) − (∆νq )so .
h j
110 CHAPTER 3
(β L S), the centers of gravity of the σr , σb , and π components have the same
frequencies as the individual components that would originate from the transition
between a spinless lower level (βL) and a spinless upper level (β L ), both split
according to the normal Zeeman effect. This result is what should be expected
from the principle of spectroscopic stability,1 as it can be inferred from considering
the limit of negligible fine structure.
3.5. Magnetic Field and Hyperfine Structure
It is well-known that hyperfine structure results from the interaction of the elec-
tronic cloud with the atomic nucleus. The energy involved in this interaction is
approximately one thousand times smaller than the energy relative to the spin-
orbit interaction, so that the hyperfine-structure components of a spectral line
have typical wavelength separations of a few mÅ.
In most cases (and particularly for lines originating from astrophysical plasmas),
these components remain unresolved, because their separation is much smaller than
the line width due to different broadening mechanisms. For this reason the hyper-
fine structure of spectral lines is often neglected, especially in usual astrophysical
applications.
However, when polarization phenomena are involved, the situation is substan-
tially different, and hyperfine structure generally plays an important role (see
Sect. 9.23). We describe here the effect of a magnetic field on a hyperfine-structured
line, in order to state some basic properties of the resulting pattern.
We consider a particular isotope having nuclear spin I, and introduce the rep-
resentation |αJIF f to describe its energy eigenvectors. In this notation, apart
from the symbols already used, F is the total angular momentum (electronic plus
nuclear: F = J + I ) quantum number, while f represents its projection on the
z-axis of the coordinate system (f = −F, −F + 1, . . . , F ).
The hyperfine-structure interaction energy can be expressed as an infinite series
of electric and magnetic multipoles (see for instance Kopfermann, 1958). In most
cases, a very good approximation to the observed energy intervals is obtained by
retaining only the first two terms in the multipole series; these are the magnetic-
dipole and electric-quadrupole terms, which are given by
A(α, J, I)
αJIF f |Hhfs
(1)
|αJIF f = δF F δf f K
2
αJIF f |Hhfs
(2)
|αJIF f = δF F δf f B(α, J, I)
4
× K(K + 1) − J(J + 1)I(I + 1) , (3.70)
3
where A(α, J, I) and B(α, J, I) are the magnetic-dipole and electric-quadrupole hy-
perfine structure constants relative to a given atomic level of a given isotope, while
1
A formulation of this principle is given in Sect. 7.10.
ATOMIC SPECTROSCOPY 111
K is defined by
K = F (F + 1) − J(J + 1) − I(I + 1) . (3.71)
An extensive table of experimental A and B values can be found in Brix and
Kopfermann (1952). From such values, and from Eqs. (3.70), the energy of the
various hyperfine-structure sublevels can be easily computed. The sublevels are
characterized by the total angular momentum F and are degenerate with respect
to f .
If a magnetic field is present, so weak as to produce a splitting much smaller
than the energy differences between the various hyperfine-structure sublevels, its
effect can be deduced from perturbation theory, by diagonalizing the magnetic
Hamiltonian HB of Eq. (3.3) on the degenerate basis |αJIF f .1 Aligning the
z-axis of the reference system with the direction of the magnetic field, the matrix
elements to be evaluated are of the form
αJIF f |µ0 B(Lz + 2Sz )|αJIF f .
Let us evaluate the more general matrix element (that will be needed later)
between two states having different F -values. Writing (Lz + 2Sz ) in the form
(Jz + Sz ) and bearing in mind the definition of the Landé factor (Eq. (3.7)), we
obtain from Eqs. (2.96) (Wigner-Eckart theorem) and (2.108)
αJIF f |µ0 B(Jz + Sz )|αJIF f =
= µ0 B gαJ δf f (−1)J+I−f J(J + 1)(2J + 1)(2F + 1)(2F + 1)
!
F F 1 F F 1
× . (3.72)
J J I −f f 0
In particular, for F = F we have
αJIF f |µ0 B(Jz + Sz )|αJIF f = µ0 B gαJ ghfs (F ) f δf f ,
where the hyperfine-structure Landé factor, ghfs (F ), is given by
1 F (F + 1) + J(J + 1) − I(I + 1)
ghfs (F ) = (F = 0) . (3.73)
2 F (F + 1)
Thus for weak magnetic fields any hyperfine level of total angular momentum F
splits up into (2F + 1) magnetic sublevels, each characterized by its own f -value.
The splitting is proportional to the magnetic field and to the product of the usual
Landé factor gαJ times a Landé factor ghfs (F ) which depends on the particular
hyperfine level. This is the Zeeman effect for hyperfine structure.
1
The Hamiltonian describing the direct interaction between the magnetic field and the nuclear
spin can be safely neglected, since the corresponding energy is about 10 3 times smaller than the
energy involved in the interaction between the magnetic field and the electronic cloud.
112 CHAPTER 3
If the magnetic field is so strong that the Zeeman splitting is of the same order as
the energy difference between different hyperfine levels, perturbation theory cannot
be applied any longer, and one has to go back to the simultaneous diagonalization
of the hyperfine-structure and magnetic Hamiltonians, whose matrix elements are
given in Eqs. (3.70) and (3.72). The situation here is quite similar to that of the
Paschen-Back effect (Sect. 3.4): since the matrix elements are diagonal with respect
to f , the overall Hamiltonian can be factored into [2(I + J) + 1] submatrices, each
characterized by a particular f -value.1 Thus the eigenvalues and eigenvectors are
of the form
λi (αJI, f )
|αJIif = CFi (αJI, f ) |αJIF f (3.74)
F
(where the CFi coefficients can be chosen real) and can be found, in general,
by numerical diagonalization of the total Hamiltonian. Figure 3.11 illustrates
the behavior of the energy eigenvalues for the hyperfine-structure components of
the level 3 2 P3/2 of NaI as a function of the magnetic field (the values of the
hyperfine-structure constants are from Figger and Walther, 1974). As in the case
of the Paschen-Back effect, there are several level-crossings between the hyperfine-
structure magnetic sublevels. Such level-crossings induce important phenomena in
resonance scattering (see Sect. 10.22).
The strengths and the splittings of the various hyperfine-structure components
can be computed in strict analogy with the case of the Paschen-Back effect. For
the normalized strengths of the hyperfine components of the transition connecting
a lower level (αJ) with an upper level (α J ) one obtains
3
Sqif,i f = CFi (αJI, f ) CFi (αJI, f ) CFi (α J I, f ) CFi (α J I, f )
2I + 1
F F F F
× (2F + 1)(2F + 1)(2F + 1)(2F + 1)
! !
J J 1 J J 1
×
F F I F F I
F F 1 F F 1
× , (3.75)
−f f −q −f f −q
with
Sqif,i f = 1 (q = −1, 0, +1) .
ii f f
The relevant selection rules are
∆J = 0, ±1 J = 0 J = 0
∆f = 0, ±1 .
1
It is assumed here that the magnetic splitting is much smaller than the energy differences
among the various J-levels. The opposite case will not be considered in this book.
ATOMIC SPECTROSCOPY 113
Fig.3.11. Energies of the hyperfine sublevels of the level 3 2 P3/2 of NaI as functions of the magnetic
field strength.
Similarly to the Paschen-Back effect, the strengths defined in Eq. (3.75) have
some important properties, which can be proved by observing that the coefficients
CFi (αJI, f ) satisfy the following relations, similar to Eqs. (3.62)
CFi (αJI, f ) CFi (αJI, f ) = δii (3.76a)
F
CFi (αJI, f ) CFi (αJI, f ) = δF F (3.76b)
i
CFi (αJI, f ) CFi (αJI, f ) λi (αJI, f ) =
i
= αJIF f |Hhfs
(1) (2)
+ Hhfs + HB |αJIF f . (3.76c)
The properties of the strengths are the following:
a) In the absence of magnetic fields, F is a good quantum number, and the eigen-
vectors |αJIif converge towards the states |αJIF f that are degenerate with
respect to f ; the coefficients CFi are given by
CFi (αJI, f ) = δiF ,
114 CHAPTER 3
so that we obtain
!2 2
f 3 J J 1 F F 1
SqF f,F = (2F + 1)(2F + 1) ,
2I + 1 F F I −f f −q
and summing over the degeneracy parameters f and f one gets the usual expression
for the relative strengths of hyperfine-structure multiplets
!2
f (2F + 1)(2F + 1) J J 1
S F,F = SqF f,F = .
(2I + 1) F F I
ff
b) The centers of gravity in frequency of the π, σr , and σb components (q = 0, +1,
and −1, respectively) are linear functions of the magnetic field
∆νq = Sqif,i f ∆νfiif = −q ḡ νL , (3.77)
ii f f
where νL is the Larmor frequency, ∆νfiif is the frequency shift of the component
corresponding to the transition between the lower state |αJIif and the upper
state |α J Ii f from the unperturbed frequency of the transition between the
two levels (αJI) and (α J I),
λi (α J I, f ) − λi (αJI, f )
∆νfiif = , (3.78)
h
and where ḡ is given in terms of the Landé factors of the lower and upper level
1 1
ḡ = (g + gα J ) + (gαJ − gα J ) [J(J + 1) − J (J + 1)] .
2 αJ 4
Comparison with Eq. (3.44) shows that the factor ḡ is just the effective Landé
factor that one would obtain for the transition between the levels (αJ), (α J ) if
hyperfine structure were not present.
The proof of Eq. (3.77) is rather tedious and quite similar to the proof of the
analogous equation for the Paschen-Back effect. For this reason it will not be
given here and is left as an exercise to the reader.1 Equation (3.77) is particularly
important, and it can be expressed by the following statement: The frequency shifts
of the centers of gravity of the σr , σb , and π components of a hyperfine-structured
line are the same as those resulting from the same line without hyperfine structure.
This result is what should be expected from the principle of spectroscopic stability.
1
To prove that the hyperfine-structure Hamiltonian (Eqs. (3.70)) gives no contribution to
the quantity ∆νq , one needs the sum rules
%n %n
k=1
k4 = 1
30
n(n + 1)(2n + 1)(3n2 + 3n − 1), k=1
k5 = 1
12
n2 (n + 1)2 (2n2 + 2n − 1).
To evaluate the contribution of the magnetic Hamiltonian (Eq. (3.72)), the sum rule (2.41) must
be used (instead of (2.48), employed for the Paschen-Back calculation).
ATOMIC SPECTROSCOPY 115
Fig.3.12. Logarithm of the strengths of π (upper) and σr (lower) components against wavelength
for the line EuII λ 4205. The value of the magnetic field is 6 kG. Only the components having
strength larger than 10−8 are shown. The number of components of each type is approximately
400.
The numerical calculation of the strengths and splittings of the various hyperfine-
structure components of a spectral line requires the diagonalization of the total
Hamiltonian for the lower and for the upper level followed by the evaluation of
expressions (3.75). If the element has more than one isotope, the calculations must
be repeated for each of them and the splittings must be adjusted to allow for isotope
shift.
Landi Degl’Innocenti (1978a) has presented a computer program to perform such
calculations. The program has then been applied to the line EuII λ 4205 using the
data for the hyperfine-structure constants given by Krebs and Winckler (1960).
The result is shown in Fig. 3.12 for a magnetic field of 6 kG. The number of
components is indeed striking, with an intricate pattern recalling a ‘wild forest’.
The implications of such a pattern on the determination of the europium abundance
in magnetic stars have been discussed by Landi Degl’Innocenti (1975).
3.6. Atomic Level Polarization and Density Matrix
When an atomic system is excited through a physical process which, for any reason,
is not spatially isotropic, the various magnetic sublevels of the system, degenerate
or quasi-degenerate with respect to energy, are, in general, not evenly populated;
moreover, they are characterized by definite phase relations, and the atomic system
is said to be polarized. A typical example occurs when the atomic system is excited
by a unidirectional (or polarized) radiation beam, or by collisions with a collimated
beam of fast particles.
116 CHAPTER 3
In such situations, the description of the atomic excitation in terms of the usual
concept of populations of the atomic levels (or sublevels) is no longer adequate,
and one must resort to the more sophisticated concept of atomic density operator .
The density operator can be defined for any assigned quantum system. If pα
(pα ≥ 0) is the probability of finding the system in the pure state |ψ (α) (normalized
to 1), the density operator ρ is defined by
ρ= pα |ψ (α) ψ (α) | . (3.79)
α
Introducing a complete and orthogonal basis of unit vectors {|uk } for the Hilbert
space spanned by the vectors |ψ (α) , one has
um | ρ |un = ρmn = pα um |ψ (α) ψ (α) |un ,
α
where ρmn are the matrix elements of the density operator, also called density-
matrix elements.
Some important properties of the density operator follow directly from its defi-
nition:
a) The density operator is Hermitian,
ρ = ρ† ,
which leads to the remarkable symmetry property of the density matrix
ρmn = ρ∗nm . (3.80)
b) The density operator is positive definite. In fact, for an arbitrary state vector
|φ we have
φ| ρ |φ = pα | ψ (α) |φ |2 ≥ 0 ;
α
in particular, for |φ = |un
ρnn ≥ 0 , (3.81)
and hence the diagonal matrix elements are real and non-negative.
c) The trace of the density matrix is independent of the basis chosen to define its
matrix elements. In fact, since a complete, orthonormal basis satisfies the relation
|un un | = 1 , (3.82)
n
we obtain from Eq. (3.79)
Tr (ρ) = ρnn = pα un |ψ (α) ψ (α) |un
n n α
= pα ψ (α) | |un un | |ψ (α) = pα . (3.83)
α n α
ATOMIC SPECTROSCOPY 117
In particular, if the probabilities pα are normalized to unity one gets
Tr (ρ) = 1 . (3.84)
d) The density-matrix elements satisfy the following Schwarz inequality
|ρnm |2 ≤ ρnn ρmm . (3.85)
To prove this relation, let us consider the state vector
|φ = ρnm |un − ρnn |um ;
for the diagonal matrix element φ| ρ |φ one gets
, -
φ| ρ |φ = ρnn ρnn ρmm − |ρnm |2 ,
which proves Eq. (3.85) since φ| ρ |φ ≥ 0 and ρnn ≥ 0.
e) If the quantum system is in a pure state rather than a statistical mixture of
states, or, in other words, if the sum in Eq. (3.79) is restricted to only one state
|ψ (0) , the above inequality reduces to an equality
|ρnm |2 = | un |ψ (0) |2 | um |ψ (0) |2 = ρnn ρmm .
f) The expectation value of any dynamical variable is equal to the trace of the
product of the density operator times the operator associated to the variable
A = pα ψ (α) |A|ψ (α) = Tr (ρA) . (3.86)
α
Indeed, using Eq. (3.82) one obtains
Tr (ρA) = un |ρA|un
n
= un | pα |ψ (α) ψ (α) |A|un
n α
= pα ψ (α) |A|un un |ψ (α)
n α
= pα ψ (α) |A|ψ (α) = A .
α
g) If a system consists of two subsystems a and b, the trace of the density operator
of the system over the states of one subsystem is the density operator of the other
one. Indeed, let {| an } and {| bm } be two complete bases for the Hilbert spaces
118 CHAPTER 3
of subsystems a and b , respectively, and ρ the density operator of the compound
system. If O(a) is an operator acting on subsystem a only, we have
(a)
O = Tr ρ O(a) = an | bm | ρ O(a) | bm | an
nm
& '
= an | bm | ρ | bm O(a) | an
n m
= an | Tr(b) (ρ) O(a) | an .
n
On the other hand, if ρ(a) is the density operator of subsystem a, we also have
(a)
O = Tr(a) ρ(a) O(a) = an | ρ(a) O(a) | an .
n
Comparison of these two relations shows that
Tr(b) (ρ) = ρ(a) . (3.87)
h) The time evolution of the density operator in the Schrödinger representation is
described by the equation
d 2π
ρ= [H, ρ] , (3.88)
dt ih
where the symbol [A, B] denotes the commutator of the two operators A and B,
and where H is the Hamiltonian of the system.
This equation follows directly from the Schrödinger equation applied to the state
vectors |ψ (α) entering the definition of ρ. In fact, we have from Eq. (3.79)
d d (α) d
ρ= pα |ψ ψ (α) | + |ψ (α) ψ (α) |
dt α
dt dt
2π 2π (α)
= pα H|ψ (α) ψ (α) | − pα |ψ ψ (α) |H ,
α
ih α
ih
which proves Eq. (3.88).
As far as the density-matrix elements are concerned, their time evolution depends
on the particular basis {|un }. If we choose the basis of the energy eigenvectors,
or, in other words, if the vectors |un are such that
H|un = En |un ,
we have from Eq. (3.88)
d d 2π
ρ = u | ρ |un = um | [H, ρ] |un = −2πi νmn ρmn ,
dt mn dt m ih
ATOMIC SPECTROSCOPY 119
where the Bohr frequency νmn is defined by
Em − En
νmn = .
h
On the other hand, for a different basis one gets
d 2πi , -
ρmn = − Hmp ρpn − Hpn ρmp ,
dt h p
where
Hij = ui |H|uj .
For an atomic system, the most natural basis for defining the matrix elements
of the density operator is the basis of the eigenvectors of the angular momentum,
although in some cases this is not the basis of the energy eigenvectors (see for
instance the discussion of the Paschen-Back effect in Sect. 3.4). On this basis, the
elements of the density matrix are
ρ (αJM, α J M ) = αJM | ρ |α J M .
In this representation – usually referred to as the standard representation of the
atomic density operator – the diagonal terms, ρ (αJM, αJM ), represent the pop-
ulations of the various magnetic sublevels, with the overall population nJ of an
assigned J-level given by
nJ = ρ (αJM, αJM ) . (3.89)
M
Conversely, the off-diagonal terms ρ (αJM, α J M ) represent the so-called coher-
ences or phase-interferences between different magnetic sublevels. Their physical
meaning is illustrated, in a qualitative way, by the following example.
Suppose that a simple atomic system, namely a two-level atom with a lower level
having J = 0 and an upper level having J = 1, is excited by a stationary radiation
field tuned at the frequency of the atomic transition. If we choose a quantization
direction for the atomic system, and take a reference system (xyz) with the z-axis
in the quantization direction, the eigenvectors |αJM are of the form |1 − 1 ,
|1 0 , |1 1 for the upper level, and |0 0 for the lower level.
It will be proved later (see Sect. 7.1) that if the atom absorbs a photon propagat-
ing along any direction in the x-y plane and linearly polarized in the z-direction,
it is excited from the level |0 0 to the level |1 0 . Similarly, if the atom absorbs a
circularly polarized photon propagating along the z-direction, it is excited to the
level |1 1 , or |1 −1 , according to the handedness of circular polarization. In all
these cases, the diagonal matrix elements 1M | ρ |1M for the upper level attain
a definite value, but the coherences remain zero.
120 CHAPTER 3
For any other incidence direction, or any other polarization of the absorbed
photon, the atom is excited to an upper level which is a linear combination of the
form
c−1 |1 −1 + c0 |1 0 + c1 |1 1 = cM |1M ,
M
where the three coefficients cM depend on the direction and polarization of the
absorbed photon. In this case the coherences are non-zero and take the simple
form
1M | ρ |1M = cM c∗M .
The physical meaning of coherences is just contained in this simple example. We
will see in Sect. 5.13 how a qualitative analogy can be drawn between atomic
coherences and interferences in classical oscillators.
Although coherences can be defined for any pair of magnetic sublevels, no matter
how large is the energy difference between them, the most important are those
between degenerate or quasi-degenerate sublevels. For such coherences one can
introduce more compact symbols to simplify the notations; for instance, for the
magnetic sublevels pertaining to an assigned term α one can write
αJM | ρ |αJ M = ρα (JM, J M ) ,
and if the term is described by the L-S coupling scheme
βLSJM | ρ |βLSJ M = ρβLS (JM, J M ) . (3.90)
For the coherences between magnetic sublevels of the same J-level the following
notation can be used
αJM | ρ |αJM = ραJ (M, M ) , (3.91a)
and for the diagonal elements1
αJM | ρ |αJM = ραJ (M, M ) = ραJ (M ) . (3.91b)
The atomic density matrix can be defined in the presence of hyperfine structure as
well. With obvious notations, for the matrix elements between different hyperfine-
structure magnetic sublevels one can write in general
αJIF f | ρ |α J IF f = ρ (αJIF f, α J IF f ) ,
and for the coherences between hyperfine-structure sublevels belonging to the same
J-level
αJIF f | ρ |αJIF f = ραJI (F f, F f ) . (3.92)
1
The more compact notation ραJ (M ) can be conveniently used in the physical situations
where all the off-diagonal elements vanish.
ATOMIC SPECTROSCOPY 121
Important relations can be established between the different quantities here in-
troduced, like for instance between those defined in Eqs. (3.91) and those defined
in Eq. (3.92). We can consider an atom with hyperfine structure as consisting of
two separate parts, characterized by the eigenstates |αJM and |IMI (where MI
is the projection of the nuclear spin along the z-axis) and by the density operators
ρ(J) and ρ(I) , respectively. Since the quantum number I of the atomic system is
fixed, we have from Eq. (3.87)
IMI | ρ |IMI = ρ(J) ,
MI
where ρ is the density operator of the whole system. Taking the matrix element of
these operators between the eigenstates |αJM and |αJM we get
αJM | ρ(J) |αJM = αJIM MI | ρ |αJIM MI ,
MI
which relates the density-matrix elements in Eqs. (3.91) to those in Eq. (3.92). In
the following we will use this kind of relations by dropping the apex (J), and we
will simply write
ραJ (M, M ) = αJIM MI | ρ |αJIM MI .
MI
Changing the coupling scheme via Eq. (2.12) we have
ραJ (M, M ) = JIM MI |JIF f JIM MI |JIF f
F F f f MI
× αJIF f | ρ |αJIF f ,
or, in terms of 3-j symbols
ραJ (M, M ) = (−1)f −f (2F + 1)(2F + 1)
F F f f MI
J I F J I F
×
M MI −f M MI −f
× ραJI (F f, F f ) . (3.93)
A similar relation can be established between the density-matrix elements relative
to different J-levels of a term (Eq. (3.90)) and the density-matrix elements that
one can define for the same term by neglecting its fine structure. Setting first
βLML | ρ |βLML = ρβL (ML , ML ) ,
122 CHAPTER 3
where ML is the projection of the orbital angular momentum on the z-axis, and
then introducing the inner structure due to the spin, one obtains
ρβL (ML , ML ) = (−1)M−M (2J + 1)(2J + 1)
JJ MM MS
L S J L S J
×
ML MS −M ML MS −M
× ρβLS (JM, J M ) . (3.94)
We want to remark that the density-matrix elements on the basis of the eigen-
vectors of the angular momentum depend on the reference system chosen to define
such eigenvectors. If R is the rotation that brings a reference system (the ‘old’
one) into another reference system (the ‘new’ one), the eigenvectors of the angular
momentum in the new system are connected with those in the old system by the
relation
|JM new = D(R) |JM old ,
so that the transformation law for the density-matrix elements is the following
ρ (αJM, α J M ) =
new
= new αJM | ρ |α J M new = old αJM | D† ρD |α J M old
∗ J
= DN
J
M (R) DN M (R) ρ (αJN, α J N ) . (3.95)
old
NN
In particular, for the matrix elements diagonal with respect to α and J
ραJ (M, M ) = DN
J ∗ J
M (R) DN M (R) ραJ (N, N ) . (3.96)
new old
NN
3.7. Multipole Moments of the Density Matrix
As shown by Eq. (3.95), the transformation law for the density-matrix elements
on the basis of the eigenvectors of the angular momentum involves the product of
two rotation matrices. We can however construct – similarly to Sect. 2.7 – linear
combinations of these matrix elements whose transformation law involves just one
rotation matrix. By so doing we obtain the irreducible spherical components of the
density matrix, which are often referred to as the multipole moments of the density
matrix or the spherical statistical tensors.
ATOMIC SPECTROSCOPY 123
Defining the multipole moments of the density matrix by the expression1
√
ρKQ (αJ, α J ) = (−1)J−M 2K + 1
MM
J J K
× ρ (αJM, α J M ) , (3.97)
M −M −Q
it is easy to prove that, under the rotation R of the reference system, the multipole
moments change according to the law
∗
ρK
Q (αJ, α J ) = ρK
Q (αJ, α J ) DQK
Q (R) . (3.98)
new old
Q
The proof of Eq. (3.98) will not be given here, being quite similar to an √ analogous
proof given in Sect. 2.7. We just want to remark that the factor 2K + 1 in
Eq. (3.97) is inessential: in fact, if this factor is replaced by an arbitrary function
of K, Eq. (3.98) is still satisfied. The definition given in Eq. (3.97) is the same as
Omont’s (1977), but different authors give different definitions for the multipole
moments, and no definition has a special advantage.
Comparison of Eqs. (3.98) and (2.78) shows that the transformation law for the
multipole moments involves the complex conjugate of the rotation matrix, whereas
the transformation law for irreducible spherical tensors involves the rotation matrix
itself. Thus the multipole moments – as defined in Eq. (3.97) – are not, strictly
speaking, irreducible spherical tensors. A different definition could be given by
substituting ρ (α J M , αJM ) for ρ (αJM, α J M ) in Eq. (3.97), and the multipole
moments so defined would indeed satisfy the transformation law for irreducible
spherical tensors. However, this alternative definition has no advantages over ours.
Equation (3.97) can be easily inverted using the orthogonality relations of the
3-j symbols. From Eq. (2.23b) one obtains
√
ρ (αJM, α J M ) = (−1)J−M 2K + 1
KQ
J J K
× ρK
Q (αJ, α J ) . (3.99)
M −M −Q
The conjugation property of the multipole moments follows from Eq. (3.97) and
from the Hermitian character of the density operator
∗ −Q
ρK
Q (αJ, α J ) = (−1)
J−J
ρK
−Q (α J , αJ) . (3.100)
Similarly to Sect. 3.6, one can introduce shorthand notations for the multipole
moments when dealing with restricted subspaces. For instance, for the multipole
moments relative to the magnetic sublevels of a term one can write
ρK αK
Q (αJ, αJ ) = ρQ (J, J ) ,
1
Note that for any atom (or ion) the quantum numbers J and J are both integers or both
half-integers, thus the rank K of the multipole moments is always an integer.
124 CHAPTER 3
and for the magnetic sublevels of a given J-level
ρK K
Q (αJ, αJ) = ρQ (αJ)
√ J J K
= (−1)J−M 2K + 1 ραJ (M, M ) , (3.101)
M −M −Q
MM
with the conjugation property
∗
ρK Q K
Q (αJ) = (−1) ρ−Q (αJ) . (3.102)
The analytical expressions for the most common multipole moments, as functions
of the ordinary density-matrix elements, are given in Tables 3.6 and 3.7.
The multipole moments can be also defined for a J-level having hyperfine struc-
ture. With self-evident notations, we have
√
ρQ (F, F )
αJI K
= (−1)F −f 2K + 1
ff
F F K
× ραJI (F f, F f ) . (3.103)
f −f −Q
Relations similar to those in Eqs. (3.93) and (3.94) can be easily established for
the multipole moments. From Eq. (3.93), after some Racah algebra involving the
contraction of 3-j symbols (Eq. (2.42)), one obtains
ρK
Q (αJ) = (−1)J+I+F +K
(2F + 1)(2F + 1)
FF
!
F F K
× ρQ (F, F )
αJI K
. (3.104)
J J I
In a similar way, from Eq. (3.94) the following relation can be proved
ρK
Q (βL) = (−1)L+S+J +K
(2J + 1)(2J + 1)
JJ
!
J J K
× ρQ (J, J )
βLS K
, (3.105)
L L S
where ρK
Q (βL), the multipole moment for a spinless L-term, is defined by
√ L L K
ρK
Q (βL) = (−1)L−ML 2K + 1 ρβL (ML , ML ) .
ML −ML −Q
ML ML
An important property of the multipole moments follows from the evaluation
of the expectation value of an arbitrary spherical tensor operator TQK . Denoting
ATOMIC SPECTROSCOPY 125
TABLE 3.6
Analytical expressions for the multipole moments ρK Q (αJ) as functions of the matrix elements
ραJ (M, M ). The index (αJ) is suppressed for conciseness. The multipole moments for negative
values of Q can be obtained through the relation ρK Q K ∗
−Q (αJ) = (−1) ρQ (αJ) .
J =0 ρ00 = ρ(0, 0)
J= 1
2
ρ00 = √1 ρ( 12 , 12 ) + ρ(− 12 , − 12 )
2
ρ10 = √1 ρ( 12 , 12 ) − ρ(− 12 , − 12 )
2
ρ11 = −ρ( 12 , − 21 )
J =1 ρ00 = √1 ρ(1, 1) + ρ(0, 0) + ρ(−1, −1)
3
ρ10 = √1 ρ(1, 1) − ρ(−1, −1)
2
ρ11 = − √1 ρ(1, 0) + ρ(0, −1)
2
ρ20 = √1 ρ(1, 1) − 2 ρ(0, 0) + ρ(−1, −1)
6
ρ21 = − √1 ρ(1, 0) − ρ(0, −1)
2
ρ22 = ρ(1, −1)
J= 3
2
ρ00 = 1
2
ρ( 32 , 32 ) + ρ( 12 , 12 ) + ρ(− 12 , − 12 ) + ρ(− 32 , − 32 )
ρ10 = 1
√ 3 ρ( 32 , 32 ) + ρ( 12 , 12 ) − ρ(− 12 , − 12 ) − 3 ρ(− 32 , − 32 )
2 5
√ 3 1 √
ρ11 = − √1 3 ρ( 2 , 2 ) + 2 ρ( 12 , − 12 ) + 3 ρ(− 12 , − 32 )
103
ρ20 = 1
2
ρ( 2 , 32 ) − ρ( 12 , 12 ) − ρ(− 12 , − 12 ) + ρ(− 32 , − 32 )
ρ21 = − √1 ρ( 32 , 12 ) − ρ(− 12 , − 32 )
2
ρ22 = √1 ρ( 32 , − 12 ) + ρ( 12 , − 32 )
2
ρ30 = 1
√ ρ( 32 , 32 ) − 3 ρ( 12 , 12 ) + 3 ρ(− 12 , − 12 ) − ρ(− 32 , − 32 )
2 5
3 1 √ 1 1
ρ31 = − √1 ρ( 2 , 2 ) − 3 ρ( 2 , − 2 ) + ρ(− 12 , − 32 )
5
ρ32 = √1 ρ( 32 , − 12 ) − ρ( 12 , − 32 )
2
ρ33 = −ρ( 32 , − 23 )
J =2 ρ00 = √1 ρ(2, 2) + ρ(1, 1) + ρ(0, 0) + ρ(−1, −1) + ρ(−2, −2)
5
ρ10 = √1 2 ρ(2, 2) + ρ(1, 1) − ρ(−1, −1) − 2 ρ(−2, −2)
√
10
√ √ √
ρ11 = − √1 2 ρ(2, 1) + 3 ρ(1, 0) + 3 ρ(0, −1) + 2 ρ(−1, −2)
10
ρ20 = √1 2 ρ(2, 2) − ρ(1, 1) − 2 ρ(0, 0) − ρ(−1, −1) + 2 ρ(−2, −2)
14
√ √
ρ21 = − √1 6 ρ(2, 1) + ρ(1, 0) − ρ(0, −1) − 6 ρ(−1, −2)
√
14
√ √
ρ22 = √1 2 ρ(2, 0) + 3 ρ(1, −1) + 2 ρ(0, −2)
7
126 CHAPTER 3
TABLE 3.6
(continued)
J =2 ρ30 = √1 ρ(2, 2) − 2 ρ(1, 1) + 2 ρ(−1, −1) − ρ(−2, −2)
√
10
√ √ √
ρ31 = − √1 3 ρ(2, 1) − 2 ρ(1, 0) − 2 ρ(0, −1) + 3 ρ(−1, −2)
10
ρ32 = √1 ρ(2, 0) − ρ(0, −2)
2
ρ33 = − √1 ρ(2, −1) + ρ(1, −2)
2
ρ40 = √1 ρ(2, 2) − 4 ρ(1, 1) + 6 ρ(0, 0) − 4 ρ(−1, −1) + ρ(−2, −2)
70
√ √
ρ1 = − √
4 1
ρ(2, 1) − 6 ρ(1, 0) + 6 ρ(0, −1) − ρ(−1, −2)
14
√ √ √
ρ42 = √1 3 ρ(2, 0) − 2 2 ρ(1, −1) + 3 ρ(0, −2)
14
ρ43 = − √1 ρ(2, −1) − ρ(1, −2)
2
ρ44 = ρ(2, −2)
by TQK its expectation value, and referring, for simplicity, to an atom devoid of
hyperfine structure, we have
K , -
TQ = Tr ρ TQK = α J M | ρ TQK |α J M .
α J M
Introducing between ρ and TQK the completeness relation
1= |αJM αJM | ,
αJM
and applying the Wigner-Eckart theorem (Eq. (2.96)) and Eq. (3.99), one obtains
.
K 2J + 1
TQ = αJ T K α J ρK ∗
Q (αJ, α J ) , (3.106)
2K + 1
αα JJ
which shows that the expectation value of a spherical tensor operator is strictly
connected with the multipole moments of the same rank.
Let’s now consider the physical meaning of the multipole moments. For the
0-rank moment, we have from Eq. (3.101)
1
ρ00 (αJ) = √ ρ (M, M ) , (3.107)
2J + 1 M αJ
so that the overall population nαJ of the (αJ)-level is given by
√
nαJ = 2J + 1 ρ00 (αJ) . (3.108)
ATOMIC SPECTROSCOPY 127
TABLE 3.7
Analytical expressions for the multipole moments αρK
Q (J, J ) as functions of the matrix elements
ρα (JM, J M ). The index α is suppressed for conciseness. The multipole moments αρK
Q (J , J)
α K
can be obtained through the relation ρQ (J , J) = (−1) J −J +Q α K ∗
ρ−Q (J, J ) .
J =0 ρ11 (0, 1) = ρ(0 0, 1 −1)
J = 1 ρ10 (0, 1) = −ρ(0 0, 1 0)
ρ1−1 (0, 1) = ρ(0 0, 1 1)
√
J= 1
2
ρ11 ( 12 , 32 ) = 1
2
ρ( 12 1 3
,
2 2
− 12 ) + 3 ρ( 12 − 12 , 3
2
− 32 )
J = 3
2
ρ10 ( 12 , 32 ) = − √1 ρ( 12 12 , 32 12 ) + ρ( 12 − 12 , 32 − 12 )
√ 1 1 3 3
2
ρ1−1 ( 12 , 32 ) = 1
2
3 ρ( 2 2 , 2 2 ) + ρ( 12 − 12 , 32 12 )
ρ22 ( 12 , 32 ) = −ρ( 12 21 , 32 − 32 )
√ 1 1 3 1
ρ21 ( 12 , 32 ) = 12 3 ρ( 2 2 , 2 − 2 ) − ρ( 12 − 12 , 3
2
− 32 )
ρ20 ( 12 , 32 ) = − √1 ρ( 12 1 3 1
,
2 2 2
) − ρ( 12 − 12 , 32 − 12 )
2
√
ρ2−1 ( 12 , 32 ) = 1
2
ρ( 12 1 3
,
2 2
3
2
) − 3 ρ( 12 − 12 , 32 12 )
ρ2−2 ( 12 , 32 ) = ρ( 12 − 12 , 3 3
2 2
)
√ √
J =1 ρ11 (1, 2) = √1 ρ(1 1, 2 0) + 3 ρ(1 0, 2 −1) + 6 ρ(1 −1, 2 −2)
10
√ √
J = 2 ρ10 (1, 2) = − √1 3 ρ(1 1, 2 1) + 2 ρ(1 0, 2 0) + 3 ρ(1 −1, 2 −1)
√
10
√
ρ1−1 (1, 2) = √1 6 ρ(1 1, 2 2) + 3 ρ(1 0, 2 1) + ρ(1 −1, 2 0)
10
√
ρ22 (1, 2) = − √1 ρ(1 1, 2 −1) + 2 ρ(1 0, 2 −2)
√
3
√
ρ21 (1, 2) = √1 3 ρ(1 1, 2 0) + ρ(1 0, 2 −1) − 2 ρ(1 −1, 2 −2)
6
ρ20 (1, 2) = − √1 ρ(1 1, 2 1) − ρ(1 −1, 2 −1)
√
2
√
ρ2−1 (1, 2) = √1 2 ρ(1 1, 2 2) − ρ(1 0, 2 1) − 3 ρ(1 −1, 2 0)
6
√
ρ2−2 (1, 2) = √1 2 ρ(1 0, 2 2) + ρ(1 −1, 2 1)
3
ρ33 (1, 2) = ρ(1 1, 2 −2)
√
ρ32 (1, 2) = − √1 2 ρ(1 1, 2 −1) − ρ(1 0, 2 −2)
√
3
√
ρ31 (1, 2) = √1 6 ρ(1 1, 2 0) − 2 2 ρ(1 0, 2 −1) + ρ(1 −1, 2 −2)
15
√
ρ30 (1, 2) = − √1 ρ(1 1, 2 1) − 3 ρ(1 0, 2 0) + ρ(1 −1, 2 −1)
5
√ √
ρ3−1 (1, 2) = √1 ρ(1 1, 2 2) − 2 2 ρ(1 0, 2 1) + 6 ρ(1 −1, 2 0)
15
√
ρ3−2 (1, 2) = √1 ρ(1 0, 2 2) − 2 ρ(1 −1, 2 1)
3
ρ3−3 (1, 2) = ρ(1 −1, 2 2)
128 CHAPTER 3
To give a physical interpretation to the other multipole moments, let us consider
K
the spherical tensor operators JQ that can be constructed with the rectangular
components Jx , Jy , and Jz of the angular momentum according to Eqs. (2.82) and
(2.84). Assuming for simplicity the density-matrix elements to be non-zero for just
one (αJ)-level, from Eqs. (3.106) and (2.97) we obtain
.
1 J(J + 1)(2J + 1) 1
J0 = Jz = ρ0 (αJ)
3
.
1 1 1 2 J(J + 1)(2J + 1) 1
J1 = − √ (Jx + i Jy ) = ρ1 (αJ)∗ ,
2 3
which give after inversion1
√ 1
ρ10 (αJ) = 3 Jz
J(J + 1)(2J + 1)
.
3 1
Re ρ1 (αJ) = −
1
J
2 J(J + 1)(2J + 1) x
.
3 1
Im ρ11 (αJ) = J . (3.109)
2 J(J + 1)(2J + 1) y
These formulae show that the multipole moments of rank 1 are connected with
the average value of the angular momentum components on the axes x, y, and z.
When such values
are non-zero, there is a preferred direction in space, identified
by the vector J , along which the atom may be thought to be oriented. For this
reason the multipole moments of rank 1 are called the orientation components of
the density matrix.
For the tensors of rank 2 one obtains, again from Eq. (3.106)2
√
ρ20 (αJ) = 5 f (J) 2Jz2 − Jx2 − Jy2 (3.110a)
.
15
Re ρ21 (αJ) = − f (J) Jz Jx + Jx Jz (3.110b)
2
.
2 15
Im ρ1 (αJ) = f (J) Jy Jz + Jz Jy (3.110c)
2
.
15
Re ρ22 (αJ) = f (J) Jx2 − Jy2 (3.110d)
2
.
15
Im ρ2 (αJ) = −
2
f (J) Jx Jy + Jy Jx (3.110e)
2
1
Note that, for J = 0, the quantities ρ1Q are identically zero as it follows directly from
Eq. (3.101). Equations (3.109) are valid for J = 0 only. For the same reason Eqs. (3.110) are
valid for J ≥ 1 only.
2
Note that the operators (Ji Jk + Jk Ji ), (i, k = x, y, z) are Hermitian, so that their expectation
values are real.
ATOMIC SPECTROSCOPY 129
Fig.3.13. To get an intuitive grasp of the meaning of the orientation and alignment components of
the density matrix, we can regard the atomic angular momentum as having a uniform probability
of being found on the surface of the figures drawn above. For the sphere in panel (a), the atom
is neither oriented nor aligned; for the ellipsoid in panel (b) the atom is aligned but not oriented;
finally, for the ellipsoid in panel (c) the atom is both oriented and aligned. To have an example of
an atom oriented but not aligned we can go back to panel (a) and regard the angular momentum
as having more probability of being found on the upper hemisphere than on the lower one.
with
1
f (J) = ,
J(J + 1)(2J − 1)(2J + 1)(2J + 3)
the reduced matrix element αJ J 2 αJ having been evaluated via Eqs. (2.102)
and (2.36h).
The formulae now deduced show that the multipole moments of rank 2 are con-
nected with the average value of bilinear combinations of the angular momentum
components Jx , Jy , and Jz . These multipole moments are called the alignment
components of the density matrix. The word ‘alignment’, as opposed to ‘orienta-
tion’, is used in this context to mean that the atom may be regarded as aligned
with a particular straight line in space, irrespective, however, of which direction is
chosen on the line. In Fig. 3.13 we give an intuitive illustration of the meaning of
the multipole moments of rank 1 and 2.
Turning to multipole moments of higher rank, it is possible to show that those
of rank n are connected with the average values of polynomials of degree n in
the angular momentum components Jx , Jy , and Jz . No particular name has been
assigned to these moments.
130 CHAPTER 3
Another general property of multipole moments, which follows directly from the
definition (3.97), should be remarked: a moment of the form ρK Q (Q = 0) is a
linear combination of coherences between states characterized by projections of the
angular momentum differing by Q; a moment of the form ρK 0 is a linear combination
of populations.
Finally, some inequalities can be established for the multipole moments as a
consequence of Eq. (3.85). Taking the square modulus of Eq. (3.101) and summing
over K and Q we obtain
|ρK 2
Q (αJ)| = |ραJ (M, M )|2 ;
KQ MM
on the other hand, from Eqs. (3.107) and (3.85) we have
2 1
ρ00 (αJ) = ραJ (M, M ) ραJ (M , M )
2J + 1
MM
1
≥ |ραJ (M, M )|2 ,
2J + 1
MM
and from these two relations we get
2
|ρK
Q (αJ)| ≤ 2J ρ0 (αJ)
2 0
.
K=0 Q
Similarly, for the multipole moments connecting two levels (αJ), (α J ) one obtains,
starting from Eq. (3.97)
2
|ρK
Q (αJ, α J )| ≤ (2J + 1)(2J + 1) ρ00 (αJ) ρ00 (α J ) .
KQ
Another important inequality can be established for the quantities ρK
0 (αJ). From
Eq. (3.101) we have
√ J J K
ρK
0 (αJ) = 2K + 1 (−1)J−M
ραJ (M, M ) .
M −M 0
M
On the other hand, all the quantities ραJ (M, M ) are non-negative (see Eq. (3.81));
since their sum equals [2J + 1]1/2 ρ00 (αJ), we can write
3 3
3 J J K 33 0
|ρK
0 (αJ)| ≤ (2K + 1)(2J + 1) 33 ρ (αJ) ,
M̃ −M̃ 0 3 0
where M̃ is the M -value giving the largest absolute value for the 3-j symbol.
CHAPTER 4
QUANTIZATION OF THE ELECTROMAGNETIC FIELD
(NON-RELATIVISTIC THEORY)
Many of the physical phenomena involved in the generation and the transfer of po-
larized radiation in spectral lines can indeed be described by means of the classical
theories of the electron and of the radiation field. In Chap. 5 we will apply these
theories to obtain a simple derivation of the radiative transfer equations for polar-
ized radiation and of the law of resonance scattering in the presence of a magnetic
field.
However, only the most simplified cases are accurately described by the classical
theory. As far as the atomic system is concerned, the classical theory of the elec-
tron gives an appropriate representation only of a two-level atom with a normal
Zeeman triplet and an unpolarized ground level, so that its applicability is very lim-
ited. Similarly, the classical description of the radiation field prevents the correct
treatment of important phenomena such as, for instance, stimulated emission.
In the main body of this book the interaction between atomic systems and the
electromagnetic field will be described through a full quantum-mechanical formal-
ism: this will allow us to handle even the most complicated physical situations
without being obliged to introduce additional terms into the relevant equations by
using phenomenological or heuristic arguments. To this aim, we need however the
formalism of second quantization for the electromagnetic field, a formalism that we
are going to develop in this chapter.
We also want to remark that the formalism of second quantization for the radia-
tion field is a classical subsection of quantum field theory and has been fully treated
in excellent monographs (see for instance Dirac, 1935; Heitler, 1954; Akhiezer and
Berestetskii, 1965). The reason why these concepts are again developed in this
book is to establish on a firm basis a set of notations that will be employed in
the following. On the other hand, in view of the physical applications that will be
presented later, the formalism of second quantization needs not to be developed
in its general, relativistically invariant form. We will thus restrict attention to
the simpler non-relativistic formulation, while we refer the reader to the textbooks
quoted above for an introduction to invariant formalisms.
4.1. Quantization of the Harmonic Oscillator
The quantization of the harmonic oscillator is at the basis of the formalism of
second quantization, because, as it will be shown in the following, the Hamiltonian
of the electromagnetic field reduces to the sum of an infinite number of independent
Hamiltonians each formally equal to that of the harmonic oscillator.
132 CHAPTER 4
The classical Hamiltonian of a one-dimensional harmonic oscillator, expressed as
a function of the canonical variables q and p, is given by
p2 mω 2 2
H(q, p) = + q , (4.1)
2m 2
where m and ω are the mass and the angular frequency of the oscillating particle,
respectively. The Hamilton equations give for the particle motion
dq ∂H(q, p) p
= =
dt ∂p m
dp ∂H(q, p)
=− = −mω 2 q.
dt ∂q
To solve these coupled equations we introduce, following Dirac (1935), two linear
combinations of the variables q and p of the form
a = A (p − i mωq) , a∗ = A (p + i mωq) , (4.2)
where A is a real constant which, for the moment, does not need to be specified.
Taking the time derivative of these quantities we get the equations
da da∗
= −i ωa , = i ωa∗ ,
dt dt
which have the obvious solution
−iωt iωt
a = a0 e , a∗ = a∗0 e , (4.3)
with a0 to be determined from the initial conditions. Inversion of Eqs. (4.2) yields
the solution to the classical problem
i 1
q= (a − a∗ ) , p= (a + a∗ ) ,
2mωA 2A
with a and a∗ given by Eqs. (4.3).
The corresponding quantum-mechanical problem can be treated along the lines
of the Correspondence Principle. The Hamiltonian is still given by Eq. (4.1), but
q and p are now Hermitian operators obeying the commutation rule
[q, p] = ih̄ .
To solve the quantum-mechanical problem we introduce the operator a defined as
in Eqs. (4.2) and its Hermitian conjugate a† defined as a∗ in Eqs. (4.2). Expressing
the Hamiltonian in terms of a and a† we obtain
1
H= (a a† + a† a) ,
4mA2
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 133
and evaluating the commutator of a and a† we get
[a, a† ] = 2A2 mh̄ω .
We now choose the constant A in such a way as to make this commutator equal to
unity; expressing a a† in terms of a† a we get
1
H = h̄ω a† a +
2
with
[a, a† ] = 1 . (4.4)
To find the eigenvalues and the eigenvectors of H, first of all we observe that if
|v is an eigenvector of H corresponding to the eigenvalue v,
H|v = v |v ,
we have
† H 1 v 1
v|a a|v = v| − |v = − v|v .
h̄ω 2 h̄ω 2
Since the left-hand side of this equality is the square modulus of the vector a |v ,
it must be non-negative, so that we obtain
1
v≥ h̄ω , (4.5)
2
which means that all the eigenvalues are larger than h̄ω/2.
Next we observe that, owing to the commutation relations
[H, a] = −h̄ωa , [H, a† ] = h̄ωa† , (4.6)
we have
Ha |v = (aH − h̄ω a)|v = (v − h̄ω) a |v ,
which means that the vector a |v is an eigenvector of the Hamiltonian H corre-
sponding to the eigenvalue (v − h̄ω).
Similarly, it can be easily shown that the vector a2 |v is an eigenvector of H
corresponding to the eigenvalue (v − 2h̄ω), and, in general, that the vector an |v is
an eigenvector corresponding to the eigenvalue (v−nh̄ω). The chain of eigenvectors
a |v , a2 |v , . . . , an |v , . . ., corresponding to smaller and smaller eigenvalues, must
however be limited, because otherwise property (4.5) would be violated. This
means that it must exist an eigenvector, which we denote by the symbol |0 , such
that
a |0 = 0 . (4.7)
For this eigenvector we have
† 1 1
H |0 = h̄ω a a + |0 = h̄ω |0 .
2 2
134 CHAPTER 4
Conversely, from the eigenvector |0 one can construct different eigenvectors
corresponding to larger and larger eigenvalues. In fact, from Eq. (4.6) we obtain
3
Ha† |0 = a† H + h̄ω a† |0 = h̄ω a† |0 ,
2
and, in general
1
Ha†n |0 = n+ h̄ω a†n |0 .
2
The eigenvector a†n |0 , however, needs to be normalized. Assuming the eigenvector
|0 to be normalized,
0|0 = 1 ,
it is easy to prove via Eqs. (4.4) and (4.7) that
0|an a†n |0 = n! .
The normalized eigenvectors of the Hamiltonian H, which we denote by the symbol
|n , are therefore defined by
1
|n = √ a†n |0 ,
n!
and are such that
1
H |n = n + h̄ω |n
2
√
a |n = n |n − 1
√
a† |n = n + 1 |n + 1 . (4.8)
To conclude, the energy spectrum of the Hamiltonian of the harmonic oscillator
consists of a series of equispaced energy levels of the form (n+ 1/2) h̄ω. The integer
n can be regarded as the number of a sort of particles, or discrete quanta, of energy
h̄ω that can be created or destroyed and whose presence contributes to the total
energy of the harmonic oscillator. Formulae (4.8) show that the operator a, when
acting on an eigenvector, has the effect of decreasing the number of particles, while
the operator a† has the opposite effect of increasing it. For this reason the operators
a† and a are called the creation and the destruction (or annihilation) operators,
respectively.
4.2. The Electromagnetic Field as a Superposition of Plane Waves
Consider the electromagnetic field enclosed in a cubic cavity having its sides equal
to L. In the absence of charges and currents, one can choose a special case of the
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 135
Lorentz gauge where the scalar potential φ(r, t) is identically zero, so that the field
r, t) which obeys the wave equation
is described only by the vector potential A(
r, t) − 1 ∂2
∇2 A( A(r, t) = 0 . (4.9)
c2 ∂t2
The Lorentz gauge is now expressed by1
r, t) = 0 .
div A( (4.10)
r, t) by the relations
The electric and magnetic fields are connected with A(
r, t) = − 1 ∂ A(
E( r, t)
c ∂t
r, t) = curl A(
B( r, t) . (4.11)
At any given time t the vector potential, as well as any other physical function of
r, can be expanded in Fourier series. If we require that the vector potential obey
the so-called periodicity conditions
y, z, t) = A(x
A(x, + m L, y + m L, z + m L, t)
x y z
with mx , my , mz arbitrary integers, we obtain
r , t) =
A( (t) e ik·r ,
C (4.12)
k
k
where the summation is extended to all the values of k satisfying the periodicity
conditions
k = n 2π , n 2π , n 2π , (4.13)
x
L y L z L
with nx , ny , nz arbitrary integers. Because of the real character of the function
r, t), the complex vector C
A( (t) satisfies the conjugation property
k
(t)∗ = C
C (t) .
k −
k
We now require that the vector potential A obey the wave equation (4.9). This
means
(t)
d2 C k
= −ωk2 C (t) , (4.14)
dt2 k
where
ωk = c |k| . (4.15)
1
Note that Eq. (4.10) is not relativistically invariant.
136 CHAPTER 4
Equation (4.14) can be easily solved to give
(−) e−iωk t + C
(t) = C
C (+) e iωk t ,
k
k
k
(−) and C
where C (+) are constants. Substituting into Eq. (4.12) we obtain
k k
(−) e i(k·r−ωk t) +
r, t) =
A( C (+) e i(k·r+ωk t)
C (4.16)
k k
k
k
with
(−)∗ = C
C (+) .
k −k
The vector potential in Eq. (4.16) is decomposed into progressive and regressive
waves. Taking into account that ωk = ω−k , we write the exponential of the regres-
sive wave term in the form
i( r+ωk t)
k· −i(− r −ω−k t)
k·
e =e
and then we change the summation index of this term from k to -k to obtain
r, t) =
A( (−) e i(k·r−ωk t) + C
C (−)∗ e−i(k·r−ωk t) .
k
k
k
The supplementary condition (4.10) implies, for any k,
k · C
(−) = 0 ,
k
which is a transversality condition holding for any Fourier component of the vector
potential. This relation can be satisfied in the following way. For any wavevector
k we define two complex unit vectors e (λ = 1, 2) both perpendicular to k and
kλ
perpendicular to each other,
∗
ekλ · k = 0 , ekλ · ekλ = δλλ . (4.17)
Next we write
(−) e−iωk t =
C ckλ (t) ekλ ,
k
λ=1,2
where ckλ (t) are oscillatory functions satisfying the differential equation
d
c (t) = −i ωk ckλ (t) , (4.18)
dt kλ
and we finally obtain
i
k·
r −i
k·
r
r, t) =
A( ckλ (t) ekλ e + ckλ (t)∗ ekλ
∗
e , (4.19)
kλ
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 137
from which the electric and magnetic fields can be obtained via Eqs. (4.11)
r, t) = i i ∗ −ik·
k·
r r
E( ωk ckλ (t) ekλ e − ckλ (t)∗ ekλ e
c
kλ
k × c (t) e e ik·r − c (t)∗ e ∗ e−ik·r .
r, t) = i
B( (4.20)
kλ kλ kλ
kλ
kλ
The summations are extended to all the wavenumbers satisfying Eq. (4.13) and,
for each wavenumber, to the two possible polarization states. Each set of values
(k, λ) defines a mode of the radiation field inside the cavity. The number of modes
for which the modulus of k is contained within k and k + dk and its direction is
contained in the solid angle dΩ is given by
3
L 1
2
N (k) dk dΩ = 2 k dk dΩ = V k 2 dk dΩ , (4.21)
2π 4π 3
where V = L3 is the cavity volume.
4.3. Quantization of the Electromagnetic Field
We can now introduce the formalism of second quantization by interpreting the
coefficients ckλ (t) and ckλ (t)∗ in Eqs. (4.19) and (4.20) as two operators ckλ and
c† acting on a suitable Hilbert space. By so doing, the vector potential and the
kλ
electric and magnetic fields become quantum operators acting on the same Hilbert
space.
The Hamiltonian of the electromagnetic field can be found with the help of the
Correspondence Principle, starting from the classical expression
1
H= (E 2 + B 2 ) dV (4.22)
8π V
which gives the energy of the field contained in the volume V. Substituting for E
and B their expressions given in Eqs. (4.20) (with the coefficients c (t) and c (t)∗
kλ kλ
replaced by the operators ckλ and c† , respectively) the integral in Eq. (4.22)
kλ
reduces to the sum of integrals of the form
i(
k+k )·
r
e dV ,
V
which can be easily evaluated to give, owing to the periodicity conditions (4.13),
V δk,−k .
138 CHAPTER 4
We thus obtain
V ωk2
H=− ckλ c−kλ 2
ekλ · e−kλ − k × ekλ · k × e−kλ
8π c
kλλ
ωk2 ∗
+ c†kλ c†−kλ 2
ekλ · e−∗kλ − k × ekλ
∗
· k × e−∗kλ
c
ωk2
− ckλ c†kλ 2
∗
ekλ · ekλ
∗
+ k × ekλ · k × ekλ
c
4
ωk2 ∗
− c†kλ ckλ e · e ∗
+ k × ekλ · k × ekλ . (4.23)
c2 kλ kλ
Using Eqs. (4.15) and (4.17) we have that the first two terms in the right-hand side
identically vanish, while the last two terms give1
V 2
H= ωk c c† + c† c . (4.24)
4πc2 kλ kλ kλ kλ
kλ
Finally, performing the substitutions
. .
1 ωk V 1 ωk V †
akλ = c , a†kλ = c , (4.25)
c 2πh̄ kλ c 2πh̄ kλ
we can express the Hamiltonian of the electromagnetic field as the sum of an infinite
number of Hamiltonians (one for each mode) formally identical to the Hamiltonian
of the harmonic oscillator,
1
H= h̄ω a a† + a†kλ akλ .
2 k kλ kλ
kλ
The commutation properties of the operators akλ and a† can be derived from the
kλ
Correspondence Principle. Since these operators are, apart from a factor, the quan-
tum equivalent of the coefficients ckλ (t) which obey the differential equation (4.18),
we require that the same differential equation hold for the expectation value of the
operator akλ on an arbitrary state vector |φ of the Hilbert space,
d
φ|akλ |φ = −i ωk φ|akλ |φ .
dt
1
The cross products in Eq. (4.23) can be easily evaluated using the relations
% %
a × b =
jk ijk
aj bk ,
k ijk lmk
= δil δjm − δim δjl ,
i
where the antisymmetrical tensor ijk is defined on p.7.
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 139
Since the state vector |φ evolves according to the Schrödinger equation
d
ih̄ |φ = H |φ ,
dt
we obtain
φ| [akλ , H] |φ = h̄ωk φ|akλ |φ ,
or 1
ω φ| [a , a a† + a† a ] |φ = ωk φ|akλ |φ .
2 k kλ k λ k λ k λ k λ
k λ
Being |φ an arbitrary state vector, it can be proved that this equation is satisfied
if
[akλ , ak λ ] = [a , a† ] = 0 (4.26)
kλ k λ
k or λ = λ ), and if
when the operators refer to different modes (k =
[akλ , a†kλ ] = 1 . (4.27)
The Hamiltonian H can then be expressed in its final form
1
H= h̄ωk a† a + . (4.28)
kλ kλ 2
kλ
Equation (4.28), together with the commutation rules (4.26) and (4.27), proves
that the Hamiltonian of the electromagnetic field is the sum of an infinite number
of independent Hamiltonians, each formally identical to the Hamiltonian of the
harmonic oscillator.
If we now arrange the different modes of the electromagnetic field in some definite
order, and recall the results obtained in Sect. 4.1 for the harmonic oscillator, we
have that the eigenvectors of the Hamiltonian H can be described by a state vector
of the form
1
|n1 , n2 , . . , ni , . . = (a†1 )n1 (a†2 )n2 · · (a†i )ni · · |φ0 , (4.29)
n1 ! n2 ! · · ni ! · ·
where (n1 , n2 , . . . , ni , . . .) are the occupation numbers and
|φ0 = |0, 0, . . . , 0, . . .
is the so-called vacuum state.
The corresponding energy eigenvalues are given by
!
1
H |n1 , n2 , . . . , ni , . . . = h̄ωj nj + |n1 , n2 , . . . , ni , . . . .
j
2
140 CHAPTER 4
Note that the presence of the term 1/2 in this expression leads to a non-zero (and
indeed infinite) energy eigenvalue even when all the occupation numbers are zero.
This represents a zero-point energy that can be eliminated from the theory by
simply shifting the scale on which energy is measured. By so doing we can write
& '
H |n1 , n2 , . . . , ni , . . . = h̄ωj nj |n1 , n2 , . . . , ni , . . . .
j
The physical interpretation of the formalism introduced above is the following.
The occupation number ni is the number of photons in the i-th mode, thus the state
vector in Eq. (4.29) represents the physical state where n1 photons are present in
mode 1, n2 photons in mode 2, and so on. Since the occupation number ni relative
to the i-th mode can take all the values 0, 1, 2, . . ., and since, moreover, this number
is independent of the occupation numbers relative to all the other modes, it follows
that photons obey the Bose-Einstein statistics. This allows a direct deduction of
Planck’s law. In thermodynamic equilibrium at temperature T , the probability Pn
(normalized to unity) of having n photons in a given mode of angular frequency ω
is given by
−n h̄ω
kB T
e
Pn = ∞ ,
% −m h̄ω
kB T
e
m=0
where kB is the Boltzmann constant, so that the mean number of photons in the
mode is ∞
1
n̄ = n Pn = h̄ω .
n=0 e kB T
−1
Recalling Eq. (4.21), one obtains that the energy per unit volume with angular
frequency between ω and ω + dω is given by
5
dk 1 h̄ω 3 1
u(ω) = h̄ω n̄ N (k) dΩ = 2 3 h̄ω ,
dω V π c
e kB T
−1
and that the energy density with frequency between ν and ν + dν is
dω hν 3 1
u(ν) = u(ω) = 8π 3 hν
,
dν c
e kB T
−1
which is the classical Planck’s law .
The action of the operators ai and a†i on a given state vector can be deduced
from Eqs. (4.8)
√
ai |n1 , n2 , . . . , ni , . . . = ni |n1 , n2 , . . . , ni − 1, . . .
a†i |n1 , n2 , . . . , ni , . . . = ni + 1 |n1 , n2 , . . . , ni + 1, . . .
a†i ai |n1 , n2 , . . . , ni , . . . = ni |n1 , n2 , . . . , ni , . . . .
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 141
These equations fully justify the denomination of destruction and creation oper-
ators for ai and a†i , respectively. The last equation shows that a†i ai is a positive
definite Hermitian operator whose eigenvalues are the occupation numbers ni . For
this reason it is named the occupation number operator relative to the i-th mode.
4.4. The Stokes Parameters in the Formalism of Second Quantization
The formalism introduced in the previous sections can be slightly modified to make
it more suitable for the physical applications that will be presented in Chap. 6.
Instead of labelling the modes with the wavenumber k we characterize them by the
frequency ν and the direction Ω. Thus we replace the operators a and a† by
kλ kλ
λ) and a† (ν, Ω,
the operators a(ν, Ω, λ), respectively. Moreover, instead of the unit
which still depend on the direction
vectors ekλ we will use the unit vectors eλ (Ω)
but not any longer on the radiation frequency. With these notations, and taking
Ω
into account Eqs. (4.25) and (4.15), we can write for the operator A ˆ (r ) defined in
its classical form in Eq. (4.19)1
.
ˆ
(r ) =
h &
e ik·r + a† (ν, Ω,
λ) e (Ω) λ) e (Ω)
'
∗ e−ik·r , (4.30)
A c a(ν, Ω, λ λ
2πνV
ν Ωλ
where
k = 2πν Ω
c
·Ω
e (Ω) =0, ∗= δ
· e (Ω)
eλ (Ω) (λ, λ = 1, 2) , (4.31)
λ λ λλ
and
λ), a(ν , Ω
a(ν, Ω, , λ ) = a† (ν, Ω,
λ), a† (ν , Ω
, λ ) = 0
λ), a† (ν , Ω
a(ν, Ω, , λ ) = δ δ δ . (4.32)
νν ΩΩ λλ
Similarly, for the electric and magnetic field operators we have
.
ˆ 2πνh &
E (r ) = i e ik·r
λ) e (Ω)
a(ν, Ω,
V λ
'
ν Ωλ
− a† (ν, Ω, ∗ e−ik·r
λ) e (Ω)
λ
.
ˆ (r ) = i
2πνh & λ) Ω
e ik·r
× e (Ω)
B a(ν, Ω,
V λ
'
ν Ωλ
− a† (ν, Ω,
λ) Ω ∗ e−ik·r ,
× e (Ω) (4.33)
λ
1
To avoid any possible confusion, we use in this section the symbol ˆ to distinguish the
operators from the corresponding classical quantities.
142 CHAPTER 4
and for the Hamiltonian (with the vacuum energy removed)
Ĥ = hν a† (ν, Ω,
λ) a(ν, Ω,
λ) . (4.34)
ν Ωλ
From these expressions it is possible to derive, according to the Correspondence
Principle, the quantum-mechanical operators corresponding to a number of classical
dynamical variables. In App. 3 we present such derivations for the momentum and
for the angular momentum of the electromagnetic field.
Let us now consider the classical definition of the intensity of a radiation beam,
= dE
I(ν, Ω) ,
dS dt dν dΩ
where dE is the energy crossing the surface dS (perpendicular to the beam) in the
time interval dt, with frequency between ν and ν + dν and direction contained in
the solid angle dΩ around the direction Ω. The corresponding quantum operator
can be derived by the following arguments. The number of modes having frequency
contained in the interval (ν, ν + dν) and direction in the solid angle dΩ is given,
for each polarization state, by (see Eq. (4.21))
ν2
N (ν) dν dΩ = V dν dΩ . (4.35)
c3
λ) is the number of photons belonging to a particular mode, the number
If n(ν, Ω,
of such photons crossing the surface dS in the time interval dt is
λ) c dt dS
n(ν, Ω, ,
V
and since each photon carries the energy hν one obtains
= hν 3 λ) .
I(ν, Ω) n(ν, Ω, (4.36)
c2
λ=1,2
This expression immediately suggests the form of the corresponding quantum-
mechanical operator
3
ˆ Ω)
I(ν, = hν a† (ν, Ω,
λ) a(ν, Ω,
λ) ,
c2
λ=1,2
and the generalization to the polarization tensor defined in Eq. (1.36) is quite
obvious
3
= hν a† (ν, Ω,
Iˆαβ (ν, Ω) α) a(ν, Ω,
β) . (4.37)
c2
The relations connecting the Stokes parameters with the components of the po-
larization tensor depend on the reference direction chosen to define the Stokes
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 143
chosen to expand
parameters (see Sect. 1.6) and on the specific unit vectors eλ (Ω)
ˆ
the vector potential A (r ) (see Eq. (4.30)). Obviously, both choices are completely
arbitrary, and the relation between the Stokes parameters and the polarization
tensor is in general rather involved (see Eqs. (1.42)). It is however convenient, for
in the following way
future applications, to define the unit vectors eλ (Ω)
= √1 − e (Ω)
e+1 (Ω) + i e (Ω)
a b
2
= √1 e (Ω)
e−1 (Ω) + i e (Ω)
, (4.38)
a b
2
and e (Ω)
where ea (Ω) are the reference direction unit vector and the associated
b
unit vector, respectively (see Sect. 1.6). It can be easily shown that these vectors
satisfy Eqs. (4.17), namely
·Ω
e±1 (Ω) =0
∗ = e (Ω)
· e (Ω)
e+1 (Ω) ∗=1
· e (Ω)
+1 −1 −1
∗ = e (Ω)
· e (Ω)
e+1 (Ω) ∗=0.
· e (Ω)
−1 −1 +1
Moreover, they satisfy the relation
∗ = − e (Ω)
e±1 (Ω) .
∓1
– a choice that will appear clearer
With this choice for the unit vectors eλ (Ω)
in the following, see Sect. 5.10 – the relation between the Stokes parameters and
the polarization tensor can be derived from Eqs. (1.42) substituting θ = 5π/4,
φ = −π/2 (which makes Eqs. (1.41) coincident with Eqs. (4.38)). We have
ˆ Ω)
I(ν, = Iˆ (ν, Ω)
+ Iˆ (ν, Ω)
++ −−
= −Iˆ (ν, Ω)
Q̂ (ν, Ω) − Iˆ (ν, Ω)
+− −+
= −i Iˆ (ν, Ω)
Û (ν, Ω) + i Iˆ (ν, Ω)
+− −+
= Iˆ (ν, Ω)
V̂ (ν, Ω) − Iˆ (ν, Ω)
, (4.39)
++ −−
where we have shortened the notations writing Iˆ++ instead of Iˆ+1 +1 , and so on.
Finally, substituting Eq. (4.37) we obtain
3
ˆ Ω)
I(ν, = hν a† (ν, Ω, +1) + a† (ν, Ω,
+1) a(ν, Ω, −1) a(ν, Ω,
−1)
c 2
3
= − hν a† (ν, Ω,
Q̂ (ν, Ω) −1) + a† (ν, Ω,
+1) a(ν, Ω, −1) a(ν, Ω,
+1)
c 2
3
= −i hν a† (ν, Ω,
Û (ν, Ω) −1) − a† (ν, Ω,
+1) a(ν, Ω, −1) a(ν, Ω,
+1)
c 2
3
= hν a† (ν, Ω,
V̂ (ν, Ω) +1) − a† (ν, Ω,
+1) a(ν, Ω, −1) a(ν, Ω,
−1) ,
c 2
144 CHAPTER 4
which give the Stokes parameters of the radiation field in terms of creation and de-
and the angular
struction operators. The connection between the operator V̂ (ν, Ω)
momentum operator of the electromagnetic field is discussed in App. 3.
4.5. The Density Operator of the Radiation Field
The density operator ρ(R) of the radiation field can be introduced using the same
general formalism presented at the beginning of Sect. 3.6. The definition of ρ(R) is
indeed the same as that of Eq. (3.79),
(α) (α)
ρ(R) = pα |ψR ψR | ,
α
(α)
where |ψR is now a state vector of the Hilbert space considered in Sect. 4.3.
A complete basis of this Hilbert space is that of the energy eigenvectors given in
Eq. (4.29), 6 7
|n1 , n2 , . . . , ni , . . . ,
so that the matrix elements of the operator ρ(R) have the form
ρ(R)
(n ,n ,...,ni ,...) (n1 ,n2 ,...,ni ,...) = n1 , n2 , . . . , ni , . . . | ρ(R) |n1 , n2 , . . . , ni , . . . .
1 2
The expectation value of any dynamical variable can be expressed in terms of
the density operator ρ(R) . For instance, the expectation value of the electric field
in a given point is expressed by
ˆ
r ) = Tr E
E( (r ) ρ(R) ,
and analogous formulae can be written for the other physical variables associated
with the electromagnetic radiation field.
In some special cases, the density operator of the radiation field can be written
as the direct product of an infinite number of operators each spanning one mode of
1 In these
the radiation field characterized by the frequency ν and the direction Ω.
cases we have
) ⊗ ρ(R) (ν , Ω
ρ(R) = ρ(R) (ν1 , Ω ) · · · ⊗ ρ(R) (ν , Ω
) ··· .
1 2 2 i i
The expectation values of the polarization tensor components relative to the mode
) are then given by
(ν0 , Ω 0
Iαβ (ν0 , Ω ) ρ(R) = Tr Iˆ (ν , Ω
) = Tr Iˆ (ν , Ω ) ρ(R) (ν , Ω
) ,
0 αβ 0 0 αβ 0 0 0 0
1
The concept of mode is used here in a broader sense than in previous sections. Modes
are defined here irrespective of the photon polarization, so that, more precisely, they are double-
modes.
QUANTIZATION OF THE ELECTROMAGNETIC FIELD 145
or, with shorthand notations,
hν 3
Iαβ = Tr Iˆαβ ρ(R)
0 = 2 Tr a†α aβ ρ(R)
0 ,
c
where ρ(R)
0 is the ‘reduced’ density operator acting on the mode (ν0 , Ω0 ) and where
a†α and aβ are the operators a† (ν0 , Ω
, α) and a(ν , Ω
0 0
, β), respectively.
0
It should be remarked that the density operator described above is an operator
acting on the Hilbert space of the occupation numbers. Sometimes, a different
density operator is introduced, acting on the wavefunction space of the photons
(Fano, 1949; Blum, 1981). This latter operator will not be used in this book.
CHAPTER 5
INTERACTION OF MATERIAL SYSTEMS WITH POLARIZED
RADIATION (THE CLASSICAL APPROACH)
In this chapter we will discuss the physical mechanisms involved in the generation
and transfer of polarized radiation in spectral lines within the framework of classical
physics. In particular, the classical theory of the electron (already introduced in
Sect. 3.2) will be used to describe the atomic system interacting with polarized
radiation, and the electromagnetic radiation field will be described in terms of
classical electrodynamics.
The results that will be derived are of limited validity as, in most cases, the
correct results can only be obtained by a quantum-mechanical treatment. The
formalism of classical physics is, however, simpler and more transparent than the
quantum-mechanical one. Thus it is suited to give the reader a plain introduction
and an intuitive approach to the physical processes that will be treated in full
generality in the following chapters. Moreover, some of the results obtained from
classical physics coincide with special cases of the corresponding quantum results.
And it is indeed illuminating and encouraging to find, as a limiting case of an in-
volved quantum-mechanical calculation, the same result deduced from the simpler,
classical formalism.
This chapter is also intended to show the strict connection between some clas-
sical and quantum concepts. This subject – generally overlooked in textbooks on
Quantum Mechanics – can be conveniently illustrated by the physical processes
described here, such as resonance scattering and the Hanle effect. Although the
connection can be established only in a qualitative way, it is nevertheless important
for the correct understanding of polarization phenomena.
5.1. Propagation of Electromagnetic Waves in Anisotropic Media
The propagation properties of electromagnetic waves in a material medium can be
derived, in classical physics, from the Maxwell equations. In the following we will
restrict attention to a homogeneous medium having magnetic permeability µ = 1,
so that it is not necessary to distinguish between the two vectors B (magnetic
induction) and H (magnetic field).
In such a medium, with no free charges, the Maxwell equations take the form (in
the Gauss-Hertz system of units)
=0
divD (5.1a)
=0
divB (5.1b)
148 CHAPTER 5
= 1 ∂D
curl B (5.1c)
c ∂t
1 ∂B
=−
curl E , (5.1d)
c ∂t
is connected
where c is the velocity of light and where the electric displacement D
with E by the equation
=E
D + 4π P ,
P being the electric polarization. From Eqs. (5.1c) and (5.1d) it easily follows that
1 ∂2D
− grad div E
∇2 E = . (5.2)
c ∂t2
2
is related to E
For an isotropic medium, D by the simple equation
= E
D ,
where is the dielectric constant of the medium. Substitution into Eqs. (5.2) and
(5.1a) leads to the wave equation for the electric field
∂2E
=
∇2 E
c2 ∂t2
and to the transversality condition
=0.
divE
A particular solution to these equations is the well-known plane wave of frequency ν
propagating along the direction Ω
ν
e 2πi c (n r · Ω−ct) ,
=E
E (5.3)
0
√ is a constant such
where n = is the index of refraction of the medium and E0
that
·Ω
E =0.
0
Note that Eq. (5.3) gives for the spatial evolution of the electric field components
∂Ej ν
= 2πi n Ej ,
∂s c
along any direction perpendicular to Ω
where Ej is the component of E and s = r · Ω
is the coordinate measured along the ray path.
MATTER-RADIATION INTERACTION (CLASSICAL) 149
For an anisotropic medium, like for instance a birefringent crystal, or a vapor
permeated by a magnetic and/or electric field, the relation between D and E
becomes of tensorial form,
Di = ij Ej ,
j
where Di (i = 1, 2, 3) and Ej (j = 1, 2, 3) are the Cartesian components of the
vectors D and E, respectively, and where is the dielectric tensor.
ij
From general theorems on matrices it follows that the dielectric tensor can always
be diagonalized through a suitable similarity transformation. This means that any
medium is characterized by a triplet of axes uα (α = 1, 2, 3), called the principal
dielectric axes, such that when the vectors D and E are expanded on this basis,
=
D Dα uα , =
E Eα uα , (5.4)
α α
one simply has
Dα = α E α . (5.5)
The quantities α are called the principal dielectric constants. In the special case
where the medium is a crystal, the dielectric tensor is real and symmetric (Born
and Wolf, 1964). In this case the principal dielectric constants are real, and the
three unit vectors uα are also real and form an orthogonal triplet.
By contrast, these properties break down for absorbing anisotropic media. In this
case the principal dielectric constants are generally complex and the same holds for
the unit vectors uα , which no longer are orthogonal. It follows that if we expand
any vector v on the basis uα ,
v = vα uα , (5.6)
α
the components vα are given by
vα = uα · v , (5.7)
where the unit vectors uα (α = 1, 2, 3) are such that
uα · uβ = δαβ .
These vectors can be explicitly written in the form1
1 1 1
u1 = u × u3 , u2 = u × u1 , u3 = u × u2 , (5.8)
g 2 g 3 g 1
1
For a dielectric crystal we simply have uα = uα . In the case of a vapor permeated by a
magnetic field, it will be shown in Sect. 5.3 that uα =
uα∗ .
150 CHAPTER 5
where g = u1 · (u2 × u3 ).
The general solution to Eqs. (5.2) and (5.1a) in an anisotropic absorbing medium
is quite complicated, except for some particular directions of propagation and po-
larization. For instance, if we consider the wave
ν (n r · Ω−ct)
= E u e 2πi c
E α
α α
with
n2α = α (5.9)
and
=0,
uα · Ω
we easily see that Eqs. (5.2) and (5.1a) are satisfied. Obviously this is a very special
case because the wave is polarized along one of the principal dielectric axes and it
propagates along a direction perpendicular to the same axis.
Instead of looking for a general solution to the problem of wave propagation
within the medium,1 we will confine ourselves to establishing an evolution equation
for the components of the electric field. Our derivation will be further restricted
to media having principal dielectric constants very close to unity. In other words,
defining
α = 1 + ξα (α = 1, 2, 3) , (5.10)
we suppose that |ξα | 1.
Let us now consider a stationary plane wave of frequency ν propagating along
The vectors E
an arbitrary direction Ω. and D associated with the wave depend
) and on time t, thus Eq. (5.2)
only on s (the spatial coordinate measured along Ω
can be written in the form
∂2E 2 2 2
− ∂ (E
Ω = − 4π ν D
· Ω) . (5.11)
∂s2 ∂s2 c2
gives
Scalar multiplication by Ω
·Ω
D =0, (5.12)
so that the transversality condition (5.1a) is satisfied. These equations show that
is perpendicular to Ω
D while E has, in general, a non-zero component along the
direction of propagation.
Let’s now introduce two mutually orthogonal unit vectors ej perpendicular to
the direction of propagation,
ej∗ · ek = δjk , =0
ej · Ω (j, k = 1, 2) . (5.13)
1
This is a classical problem in crystal optics and leads to the so-called Fresnel equation of
wave normals (see Born and Wolf, 1964).
MATTER-RADIATION INTERACTION (CLASSICAL) 151
These vectors can be, for instance, the reference direction unit vector ea and the
associated unit vector eb defined in Sect. 1.6 or, more generally, the complex unit
vectors e+ and e− defined in Eqs. (1.41). Defining also
e3 = e3∗ = Ω
, (5.14)
and D
we can expand the vectors E on the basis (e , e , e ),
1 2 3
E = Ei ei , =
D Di ei ,
i i
where
Ei = ei∗ · E
, Di = ei∗ · D
.
The relation between the components Ei of the electric field in the reference system
(e1 , e2 , e3 ) and the components Eα along the triplet (u1 , u2 , u3 ) is the following
Ei = (ei∗ · uα ) Eα , Eα = (uα · ei ) Ei . (5.15)
α i
Analogous equations obviously hold for the components of the electric displace-
ment.
From the transversality condition (5.12) one obtains, using Eqs. (5.4), (5.5),
(5.15), (5.14), and (5.10)
(e3∗ · uα ) (uα · ej ) (1 + ξα ) Ej = 0 . (5.16)
j α
On the other hand, if we expand any vector ej on the basis uα via Eqs. (5.6) and
(5.7),
ej = (uα · ej ) uα ,
α
we obtain from Eqs. (5.13) and (5.14)
(ei∗ · uα ) (uα · ej ) = δij , (5.17)
α
therefore Eq. (5.16) can be written in the form
1+ (e3∗ · uα ) (uα · e3 ) ξα E3 =
α
=− (e3∗ · uα ) (uα · e1 ) ξα E1 − (e3∗ · uα ) (uα · e2 ) ξα E2 .
α α
This formula shows that the ratio of the longitudinal component E3 of the electric
field to the transverse components is of order ξα . Since we have supposed |ξα | 1,
we have
| E 3 | | Ei | (i = 1, 2) .
152 CHAPTER 5
Consider now Eq. (5.11). Performing the scalar product by ej∗ (j = 1, 2) and
using Eqs. (5.4), (5.5), (5.10), and (5.15), we get1
∂ 2 Ej 4π 2 ν 2
2
= − 2 Dj
∂s c
4π 2 ν 2 ∗
3
=− 2 (ej · uα ) (uα · ek ) (1 + ξα ) Ek (j = 1, 2) .
c α
k=1
The contribution from k = 3 has the form
4π 2 ν 2 ∗
− (ej · uα ) (uα · e3 ) ξα E3 ,
c2 α
and since E3 is itself of order ξα , this contribution can be neglected according to
the assumption | ξα | 1. Up to first-order terms in ξα we thus have
∂ 2 Ej 4π 2 ν 2 ∗
2
= − (ej · uα ) (uα · ek ) (1 + ξα ) Ek .
∂s2 c2 α
k=1
To the same order of approximation it can be easily shown, with the help of
Eq. (5.17), that this equation is equivalent to the following
∂Ej ν ∗
2 , 1 -
= ± 2πi (ej · uα ) (uα · ek ) 1 + ξα Ek . (5.18)
∂s c α
2
k=1
The ± sign appearing in this equation is a consequence of the existence of progres-
sive and regressive waves. If we choose to describe the temporal oscillation of the
wave through an exponential of the form e−2πiνt (consistently with the convention
of Sect. 1.3), we must take the positive sign. Since from Eqs. (5.9) and (5.10) we
also have
1
1 + ξα = nα , (5.19)
2
Eq. (5.18) can be finally written in the form
ν ∗
2
∂Ej
= 2πi (ej · uα ) (uα · ek ) nα Ek (j = 1, 2) . (5.20)
∂s c α k=1
This equation expresses the variation along the propagation direction of the trans-
verse components of the electric field. Its validity is limited to media whose prin-
cipal dielectric constants differ slightly from unity and it cannot be applied, in
general, to crystals. The same equation has been used (although without physical
explanation) by Jefferies et al. (1989) in their derivation of the transfer equations
1
The scalar product by e3∗ leads to the trivial identity 0 = 0.
MATTER-RADIATION INTERACTION (CLASSICAL) 153
for polarized radiation from classical physics. A simpler, heuristic derivation, that
is however basically inconsistent, has been given by Landi Degl’Innocenti (1992).
5.2. Transfer Equations for Polarized Radiation
Equation (5.20) describes the spatial evolution, along the ray path, of the transverse
components of the electric field associated with a stationary plane wave of frequency
ν. This equation can be rewritten in the form
∂
Ej (s, t) = − Gjk Ek (s, t) (j, k = 1, 2) , (5.21)
∂s
k
where
ν
Gjk = −2πi (uα · ej∗ ) (uα · ek ) nα . (5.22)
c α
The tensor G, which will be referred to in the following as the propagation tensor
of the electric field, depends both on the physical nature of the medium, specified
by the (complex) refractive indices nα , and on the geometry of the propagation
specified by the two scalar products. It is worth noticing that in an isotropic
medium, where nα = n, we get from Eq. (5.17)
ν
Gjk = −2πi n δjk . (5.23)
c
From Eq. (5.21) the transfer equation for the polarization tensor can be eas-
ily derived. Recalling the definition of polarization tensor given in Chap. 1 (see
Eq. (1.25) and its generalization in Sect. 1.8) we have
d Jjk (s + ∆s) − Jjk (s)
Jjk = lim
ds ∆s→0
∗ ∆s
Ej (s + ∆s, t) Ek (s + ∆s, t) − Ej∗ (s, t) Ek (s, t)
= lim
∆s→0 ∆s
8 ∗ 9 8 9
∂Ej (s, t) ∗ ∂Ek (s, t)
= Ek (s, t) + Ej (s, t) ,
∂s ∂s
and since the propagation tensor can be extracted from the statistical average
implied by the symbol · · ·, we obtain
d 2
Jjk = − G∗jl Jlk + Gkl Jjl (j, k = 1, 2) . (5.24)
ds
l=1
To derive the transfer equations for the Stokes parameters we need the relation
between these quantities and the components of the polarization tensor. We recall
154 CHAPTER 5
that this relation depends on the choice of the unit vectors e1 and e2 defined in
Eq. (5.13). The easiest choice is
e1 = ea , e2 = eb ,
where ea is the reference direction unit vector and eb the associated unit vector
(see Sect. 1.6). With this definition the relation between the Stokes parameters
and the components of the polarization tensor is given by Eqs. (1.34), which can
be inverted to give
1 1
J11 = I +Q J12 = U − iV
2k 2k
1 1
J22 = I −Q J21 = U + iV .
2k 2k
Substitution into Eqs. (5.24) leads to the following propagation equation for the
Stokes parameters
⎛ ⎞ ⎛ ⎞⎛ ⎞
I ηI ηQ ηU ηV I
d ⎜Q⎟ ⎜ ηQ ηI ρV −ρU ⎟ ⎜ Q ⎟
⎝ ⎠ = −⎝ ⎠⎝ ⎠ , (5.25)
ds U ηU −ρV ηI ρQ U
V ηV ρU −ρQ ηI V
where the seven independent quantities appearing in the propagation matrix are
given by
ηI = Re (G11 + G22 )
ηQ = Re (G11 − G22 ) ρQ = −Im (G11 − G22 )
ηU = Re (G12 + G21 ) ρU = −Im (G12 + G21 )
ηV = Im (G12 − G21 ) ρV = Re (G12 − G21 ) . (5.26)
Equations (5.25) and (5.26) are at the basis of radiative transfer for polarized
radiation and deserve some discussion. First of all it should be emphasized the
remarkable symmetry property of the 4 × 4 propagation matrix. This matrix is
constructed with only seven independent quantities and can be decomposed into a
diagonal matrix, proportional to ηI , and two off-diagonal matrices, one symmetric
and the other antisymmetric about the main diagonal.
Denoting the propagation matrix by K, we have
⎛ ⎞
ηI 0 0 0
⎜ 0 ηI 0 0 ⎟
K =⎝ ⎠+
0 0 ηI 0
0 0 0 ηI
⎛ ⎞ ⎛ ⎞
0 ηQ ηU ηV 0 0 0 0
⎜η 0 0 0 ⎟ ⎜0 0 ρV −ρU ⎟
+⎝ Q ⎠+⎝ ⎠ . (5.27)
ηU 0 0 0 0 −ρV 0 ρQ
ηV 0 0 0 0 ρU −ρQ 0
MATTER-RADIATION INTERACTION (CLASSICAL) 155
The first matrix is responsible for the absorption of the energy of the electromag-
netic wave irrespective of its polarization state. This matrix would produce by
itself an exponential decrease of the whole Stokes vector, so that the quantity ηI
can be regarded as the generalization to the polarized case of the absorption coef-
ficient encountered in the standard theory of radiative transfer. In this respect it
is worth noticing that for an isotropic medium we simply have from Eq. (5.23)
ν
ηI = 4π k,
c
k being the imaginary part of the refractive index which is connected with the
standard absorption coefficient. This first matrix will be referred to in the following
as the absorption matrix .
As apparent from Eq. (5.23), both the second and the third matrix are zero for
an isotropic medium. For reasons that will be clarified in the next section, these
matrices will be referred to in the following as dichroism matrix and dispersion
matrix , respectively.
The expression that we have obtained for the propagation matrix K, and in par-
ticular its symmetry properties, descend directly from the physical approximations
adopted. First of all we have assumed the principal dielectric constants α – which
describe the dielectric properties of the medium – to be independent of the ampli-
tude of the propagating electric field. This means that the polarization P induced
in the medium is linearly related to the electric field itself. The present theory is
therefore inadequate to treat the physical phenomena connected with non-linear
optics. Moreover, we have assumed the principal dielectric constants of the medium
to differ slightly from unity, so that the applicability of Eqs. (5.25)-(5.26) is limited
to such media. Finally, we have supposed the electromagnetic wave to propagate
in a homogeneous medium, whose properties do not depend on the spatial coordi-
nate r. It can however be shown that for non-homogeneous media the equations
now derived can still be applied provided that
3 3
3 3 3 3
3 grad n 3 3 n 2π 3 ,
α 3 α λ3
a condition which is generally well-satisfied for optical radiation propagating in
astrophysical or laboratory plasmas.
5.3. Application to Magnetic Lines
We now apply the results of the previous section to the case of an atomic vapor in
the presence of a magnetic field. We describe the atom by the classical theory of
the electron, going back to the oscillator model already used in Sect. 3.2. In the
presence of a constant magnetic field B and of an oscillating electric field of the
form
e−2πiνt ,
E
156 CHAPTER 5
the motion of the electron is described by the equation
d2 x e dx − γ dx − e0 E
e−2πiνt ,
= −4π 2 ν02 x − 0 ×B (5.28)
dt2 mc dt dt m
where ν0 is the frequency of the oscillator and γ is the damping constant.
To find the dipole induced by the oscillating electric field associated with the
incident electromagnetic wave we disregard the transient solution depending on
the initial conditions and we look for a solution of the form
−2πiνt
x = a e .
Substitution into Eq. (5.28) gives
e0
4π 2 (ν02 − ν 2 ) − 2πi νγ a − 8π 2 i ννL a × u0 = − E,
m
where νL is the Larmor frequency defined in Eq. (3.10) and u0 is a real unit vector
directed along the magnetic field.
Similarly to Sect. 3.2, we introduce the real orthogonal unit vectors ur , us (such
that (ur , us , u0 ) is a right-handed coordinate system), and their linear combinations
u±1 defined in Eqs. (3.25). Expanding the vectors a and E on the basis u (α =
α
0, ±1),
a = aα uα , E = Eα uα ,
α α
one easily obtains, with the help of Eq. (3.26)
aα = χα Eα , (5.29)
where χα is the electric susceptibility
e0 1
χα = − (α = 0, ±1) (5.30)
4π 2 m (ν02 − ν 2 ) − 2iνΓ − 2αννL
with Γ = γ/4π.
If there are N oscillators per unit volume, the electric polarization is
P = −N e0 a
and the electric displacement is
=E
D + 4π P = E
− 4πN e a .
0
Therefore the principal dielectric constants α defined in Eq. (5.5) are given by
N e20 1
α = 1 + (α = 0, ±1) .
πm (ν0 − ν ) − 2αννL − 2iνΓ
2 2
MATTER-RADIATION INTERACTION (CLASSICAL) 157
Since both νL and Γ are usually much smaller than ν0 (see Sect. 3.2), the principal
dielectric constants are sharp functions of the frequency ν peaked at ν0 , so that we
can replace (ν02 − ν 2 ) by 2ν (ν0 − ν).
If we now assume – consistently with the treatment of Sect. 5.2 – the principal
dielectric constants to be close to unity, and separate the refractive indices nα in
their real and imaginary parts,
nα = (1 + δα ) + i kα (5.31)
with δα and kα real, we obtain from Eqs. (5.9) and (5.19)
N e20 Γ
kα =
4πmν (ν0 − ανL − ν)2 + Γ 2
N e20 ν0 − ανL − ν
δα = . (5.32)
4πmν (ν0 − ανL − ν)2 + Γ 2
It should be emphasized that the basis uα (α = 0, ±1) introduced above is such
that the unit vectors uα appearing in the expression of the propagation tensor G
(Eq. (5.22)) are given by
uα = uα∗ ,
as it can be proved directly from Eqs. (5.8). As a consequence, the propagation
tensor reduces to
ν ∗
Gjk = −2πi Cαj Cαk nα , (5.33)
c α
where Cαj are the direction cosines defined in Eq. (3.31), and the completeness
relation (5.17) becomes
∗
Cαi Cαj = δij . (5.34)
α
Equations (5.33), (5.34), and (5.31) allow the coefficients of the propagation matrix
K given in Eq. (5.26) to be written in the form1
ν
ηI = 2π kα |Cα1 |2 + |Cα2 |2
c α
ν
ηQ = 2π kα |Cα1 |2 − |Cα2 |2
c α
ν
∗
ηU = 2π kα 2 Re Cα1 Cα2
c α
ν
∗
ηV = 2π kα 2 Im Cα1 Cα2
c α
1
Note that these expressions hold provided the unit vectors ei appearing in the direction cosines
Cαi are the reference direction unit vector ea and the associated unit vector eb (see Sect. 5.2).
158 CHAPTER 5
θ
Ω
Fig.5.1. The direction of the magnetic field B is specified by the polar angle θ and the azimuth
the unit vector
angle χ, reckoned from the reference direction ea . In the plane perpendicular to B,
and Ω;
ur lies in the plane containing B
us is defined accordingly.
ν
ρQ = 2π δα |Cα1 |2 − |Cα2 |2
c α
ν ∗
ρU = 2π δα 2 Re Cα1 Cα2
c α
ν ∗
ρV = 2π δα 2 Im Cα1 Cα2 . (5.35)
c α
Thus in the case considered in this section (propagation of electromagnetic waves
through an atomic vapor permeated by a magnetic field), the quantities ηI , ηQ ,
ηU , ηV depend only on the imaginary parts kα of the refractive indices, while ρQ ,
ρU , ρV depend only on the real parts δα . More precisely, ηQ , ηU , ηV depend upon
differences between the imaginary parts, as apparent from the fact that these quan-
tities vanish when the three kα are equal (isotropic case). Therefore the phenomena
described by the second matrix in the expansion (5.27) consist in a differential ab-
sorption of the different polarization states: hence the denomination of ‘dichroism
matrix’ introduced in Sect. 5.2. By contrast, the third matrix in Eq. (5.27) is
connected with differences between the real parts of the refractive indices, thus
it describes phenomena associated with the dephasing of the different polariza-
tion states in the propagation through the medium (hence the denomination of
‘dispersion matrix’).
The direction cosines Cαi can be explicitly evaluated once the geometry of the
magnetic field is specified. From Fig. 5.1 we easily obtain
ur = cos θ cos χ ea + cos θ sin χ eb − sin θ Ω
us = − sin χ ea + cos χ eb
u = sin θ cos χ e + sin θ sin χ e + cos θ Ω ,
0 a b
so that
1
u±1 = √ (∓ cos θ cos χ − i sin χ) ea + (∓ cos θ sin χ + i cos χ) eb ± sin θ Ω
.
2
MATTER-RADIATION INTERACTION (CLASSICAL) 159
TABLE 5.1
Expressions for the bilinear combinations of the quantities Cαi
α = −1 α=0 α = +1
|Cα1 |2 + |Cα2 |2 1
2
(1 + cos2 θ) sin2 θ 1
2
(1 + cos2 θ)
|Cα1 |2 − |Cα2 |2 − 12 sin2 θ cos 2χ sin2 θ cos 2χ − 12 sin2 θ cos 2χ
∗
2 Re Cα1 Cα2 − 12 sin2 θ sin 2χ sin2 θ sin 2χ − 12 sin2 θ sin 2χ
∗
2 Im Cα1 Cα2 − cos θ 0 cos θ
The Cαi combinations appearing in Eqs. (5.35) are summarized in Table 5.1. Using
Eqs. (5.32) one finally obtains the following expressions for the elements of the
propagation matrix
πe20 1 φb + φr
ηI = N φp sin2 θ + (1 + cos2 θ)
mc 2 2
πe20 1 φ + φr
ηQ = N φp − b sin2 θ cos 2χ
mc 2 2
πe20 1 φ + φr
ηU = N φp − b sin2 θ sin 2χ
mc 2 2
πe20 1
ηV = N φr − φb cos θ
mc 2
πe20 1 ψb + ψr
ρQ = N ψp − sin2 θ cos 2χ
mc 2 2
πe20 1 ψ + ψr
ρU = N ψp − b sin2 θ sin 2χ
mc 2 2
πe20 1
ρV = N ψr − ψb cos θ , (5.36)
mc 2
where
1 Γ 1 ν0 + νL − ν
φb = ψb =
π (ν0 + νL − ν)2 + Γ 2 π (ν0 + νL − ν)2 + Γ 2
1 Γ 1 ν0 − ν
φp = ψp =
π (ν0 − ν)2 + Γ 2 π (ν0 − ν)2 + Γ 2
1 Γ 1 ν0 − νL − ν
φr = ψr = . (5.37)
π (ν0 − νL − ν)2 + Γ 2 π (ν0 − νL − ν)2 + Γ 2
The functions φb , φp , φr – where the subscripts stand for ‘blue’, ‘parallel’, and ‘red’,
respectively – have already been encountered in Sect. 3.2 (see Eqs. (3.38)). They
are Lorentzian profiles centered at the frequencies (ν0 + νL ), ν0 , and (ν0 − νL ),
160 CHAPTER 5
Fig.5.2. Absorption profiles φ and dispersion profiles ψ in units of Γ −1 , for νL /Γ = 5.
respectively, and normalized to unity in frequency,
∞
φb,p,r dν = 1 .
−∞
The functions ψb , ψp , ψr are the corresponding dispersion profiles. An illustrative
example of such profiles is shown in Fig. 5.2. These profiles can be easily generalized
to include the effect of microscopic velocity fields that may be present in the medium
(see Sect. 5.4).
An important extension of the theory presented in this section concerns the
inclusion of the emission term into the propagation equation. This term has already
been derived in Sect. 3.2, although in a slightly different geometry (see Fig. 3.3).
For the geometrical configuration of Fig. 5.1 the expressions (3.39) for the direction
cosines must be replaced by corresponding expressions which lead to the bilinear
combinations of Table 5.1. The emission coefficient becomes accordingly
1 φ + φr
εI = k φp sin2 θ + b (1 + cos2 θ)
2 2
MATTER-RADIATION INTERACTION (CLASSICAL) 161
1 φ + φr
εQ = k φ − b sin2 θ cos 2χ
2 p 2
1 φ + φr
εU = k φ − b sin2 θ sin 2χ
2 p 2
1
εV = k φ − φb cos θ , (5.38)
2 r
where
πe20 2ν 2
k= N 20 Ē ,
mc c
Ē being the mean energy contained into each degree of freedom of the classical
oscillator.
Adding to Eq. (5.25) the contribution due to emission, we obtain the radiative
transfer equations for polarized radiation in the form
⎛ ⎞ ⎛ ⎞⎛ ⎞
I ηI ηQ ηU ηV I −S
d ⎜Q⎟ ⎜η ηI ρV −ρU ⎟ ⎜ Q ⎟
⎝ ⎠ = −⎝ Q ⎠⎝ ⎠ , (5.39)
ds U ηU −ρV ηI ρQ U
V ηV ρU −ρQ ηI V
where the scalar source function S is given by
2ν02
S= Ē . (5.40)
c2
In particular, under the hypothesis of Local Thermodynamic Equilibrium (LTE)
the mean energy Ē is given by kB T , where kB is the Boltzmann constant and T
the absolute temperature, so that
2ν02
SLTE = k T ,
c2 B
which is the classical expression for the Planck function BP in the limit h → 0
(approximation of Rayleigh and Jeans).
Obviously, we cannot expect to find by a classical theory the exact expression
for the Planck function, which is intimately connected with quantum concepts.
However, using a semiclassical approach, we can go back to the quantization of the
harmonic oscillator (see Sect. 4.1) and write
%
∞
−nx
ne
1
Ē = hν0 n=0
%
∞ = hν0
−nx ex − 1
e
n=0
where x = hν0 /kB T , and substituting this result into Eq. (5.40) we obtain
2hν03 1
SLTE = hν0
= BP (ν0 ) ,
c2
e kB T
−1
162 CHAPTER 5
an expression that could also be obtained by invoking the Kirchhoff law for the
emission of a source in LTE.
Equation (5.39), with the matrix coefficients given by Eqs. (5.36) and with
the Planck function in the place of the source function, was first obtained by
Unno (1956) by a heuristic approach which neglected anomalous dispersion effects
(ρQ = ρU = ρV = 0). These effects were subsequently introduced by Rachkovsky
(1962a) through classical considerations, similar to those presented in this section,
based on the refractive index. Another derivation has been recently presented by
Jefferies et al. (1989). In the meantime a quantum derivation of the same equation
was given by Landi Degl’Innocenti and Landi Degl’Innocenti (1972), who pointed
out some minor errors or ambiguities (mainly connected with sign conventions) ap-
pearing both in the original papers and in subsequent papers based on those (like
for instance in Beckers, 1969a). The reader is referred to Rees (1987) for a clear
presentation of the possible errors that may arise in the derivation of Eq. (5.39).
5.4. The Voigt Function and the Associated Dispersion Profile
In the previous section we have derived the absorption profiles φb , φp , φr and the
corresponding dispersion profiles ψb , ψp , ψr for a collection of atoms supposed at
rest in the laboratory frame (the frame where the Stokes parameters are defined).
These profiles are of the form p (ν0 − ν), where ν0 is the frequency of the classical
oscillator.
Actually, however, the atoms will always be in motion. In order to realize how
the profiles are affected by these motions, let us consider a radiation beam of
frequency ν propagating along a given direction, which we identify with the line of
sight. If w is the velocity component of the atom along the line of sight, the atomic
frequency ν0 is shifted, according to the classical formula of the Doppler effect, to
the new value , w-
ν0 = ν0 1 − ,
c
where we have assumed w c (non-relativistic approximation) and where we have
adopted the sign astrophysical convention according to which w > 0 when the atom
is receding from the observer.
Therefore, for a collection of atoms having a normalized distribution of velocity
components f (w), the profile p (ν0 − ν) must be replaced by the expression
∞
w
p ν0 − ν0 − ν f (w) dw . (5.41)
c
−∞
In many cases of astrophysical interest the velocity w can be decomposed in two
parts: the bulk (or macroscopic) velocity wA of the ambient medium, plus a random
velocity due to thermal or microturbulent motions, usually distributed according
MATTER-RADIATION INTERACTION (CLASSICAL) 163
to the Maxwellian law and characterized by the velocity wT . We thus assume for
f (w) the following expression
w−w 2
1 − A
f (w) = √ e wT
. (5.42)
π wT
Substituting Eq. (5.42) into Eq. (5.41), and introducing the reduced variables
wT Γ
∆νD = ν0 , a=
c ∆νD
νL wA ν w ν0 − ν
vB = , vA = = 0 A , v= (5.43)
∆νD wT c ∆νD ∆νD
one obtains with easy transformations
1
φα = √ H(v − vA + α vB , a)
π ∆νD
1
ψα = √ L(v − vA + α vB , a) , (5.44)
π ∆νD
where α = −1, 0, +1 for the ‘red’, ‘parallel’, and ‘blue’ component, respectively,
and where we have defined the functions
∞
a −y 2 1
H(v, a) = e dy
π (v − y)2 + a2
−∞
∞
1 −y 2 v−y
L(v, a) = e dy . (5.45)
π (v − y)2 + a2
−∞
In the above expressions ∆νD represents the Doppler width in frequency units,
and it is used to normalize all the other quantities: v is the so-called reduced
frequency, vB the normalized Zeeman splitting, vA the normalized shift due to the
bulk motion, and a the damping constant . Figure 5.3 shows the two sets of profiles
φ and ψ.
The functions defined in Eqs. (5.45) are the so-called Voigt function H(v, a) and
the associated dispersion profile L(v, a). In previous works on this subject the
function L(v, a) has been usually written in the form
L(v, a) = 2F (v, a) ,
and the name of Faraday-Voigt function has sometimes been employed for F (v, a).
It should be remarked that the quantities defined in Eqs. (5.43) can be expressed
in terms of wavelength (instead of frequency) displacements and broadenings. In-
troducing the Doppler width in wavelength units,
λ20 w
∆λD = ∆νD = λ0 T , (5.46)
c c
164 CHAPTER 5
−1
Fig.5.3. Absorption profiles φ and dispersion profiles ψ in units of ∆νD , corresponding to vB = 3,
vA = 0, and a = 0.05. Note that the magnetic field is 12 times stronger than in Fig.5.2.
Eqs. (5.43) can be written in the form
λ − λ0 λ20 Γ λ0 wA
v= , a= , vA =
∆λD c ∆λD c ∆λD
λ20 νL λ20 e0 B ∆λB
vB = = = , (5.47)
c ∆λD 4πmc2 ∆λD ∆λD
where ∆λB is defined in Eq. (3.13). With these positions Eqs. (5.44) are still valid.1
The thermal velocity wT in the above equations can be related to the kinetic
temperature T and to the microturbulent velocity ξ by the expression
2kB T
wT = + ξ2 , (5.48)
µM
1
Note, however, that the profiles φα defined in Eqs. (5.44) are normalized to unity in frequency.
The corresponding profiles, normalized to unity in wavelength, are obtained by substituting ∆λD
for ∆νD in the same equations.
MATTER-RADIATION INTERACTION (CLASSICAL) 165
where µ is the atomic weight of the atom which originates the spectral line and M
is the mass of unit atomic weight. Therefore, the normalized Zeeman splitting vB
can be expressed in the form
λ0 e0 B 1
vB =
4πmc 2kB T
+ ξ2
µM
or, numerically
1
vB = 1.400 × 10−7 λ0 B , (5.49)
1.663 × 10−2 T
µ + ξ2
where λ0 is in Å, B in G, T in K and ξ in km s−1 .
In the following of this section we discuss the main mathematical properties of
the functions H(v, a) and L(v, a) defined in Eqs. (5.45). These properties are easily
obtained by considering, rather than the separate functions, their complex linear
combination
W (v, a) = H(v, a) + i L(v, a) . (5.50)
Introducing the complex variable z defined as
z = v + ia , (5.51)
we obtain with easy transformations
∞ −y 2
i e
W (z) = dy , (5.52)
π z−y
−∞
where, being a > 0, we have Im z > 0. The function W (z) is an analytical function
of the complex variable z that can be related to the complementary error function
(see Abramowitz and Stegun, 1965).
By the substitution (z − y) = u we first transform the integral in Eq. (5.52) into
the following
−u2 +2uz
i −z2 e
W (z) = e du ,
π L u
where L is a straight line parallel to the real axis in the half plane Im u > 0 (see
Fig. 5.4).
Next we observe that
−u2 +2uz
d e −u2 +2uz z2 −(u−z)2 √ z2
du = 2 e du = 2 e e du = 2 π e ,
dz L u L L
whence we obtain by integration
−u2 +2uz z
e √ t2
du = C + 2 π e dt , (5.53)
L u
0
166 CHAPTER 5
Fig.5.4. Integration paths L and L in the plane of the complex variable u.
where C is a constant that can be determined by taking the limit z → 0 of Eq. (5.53)
−u2 +2uz −u2
e e
C = lim du = du .
z→0 L u L u
The last integral can be evaluated by observing that the integration variable u has
a pole at the origin. Deforming the integration path L into the path L shown in
Fig. 5.4, the contributions arising from the integration along the real axis cancel
out, while the contribution from the semicircle around the origin simply gives −iπ.
To conclude, we have obtained that the function W (z) can be written in the form
z
−z 2 2i t2 −z 2
W (z) = e 1+ √ e dt = e erfc (−iz) , (5.54)
π
0
where erfc (x) is the complementary error function defined as in Abramowitz and
Stegun (1965).
From Eqs. (5.52) and (5.54) some important properties of the functions H(v, a)
and L(v, a) can be easily derived:
i) Limiting case a = 0.
For a = 0, being z = v, Eq. (5.54) gives
v
−v 2 2i −v2 t2
W (z) = e +√ e e dt ,
π
0
thus from Eq. (5.50) one gets
−v 2 2
H(v, 0) = e , L(v, 0) = √ D(v) , (5.55)
π
MATTER-RADIATION INTERACTION (CLASSICAL) 167
where D(v) is the so-called Dawson integral
v
−v 2 t2
D(v) = e e dt . (5.56)
0
ii) Limiting case v = 0.
Being now z = ia we get
ia 2
a2 2i t
W (z) = e 1+ √ e dt ,
π
0
so that
a
a2 2 −t2 a2
H(0, a) = e 1− √ e dt = e erfc (a)
π
0
L(0, a) = 0 .
iii) Asymptotic expansion for large values of the argument: (v 2 + a2 ) 1.
In Eq. (5.52) one can write
−1 ∞
1 1 y yk
= 1− = ,
z−y z z z k+1
k=0
therefore
∞ ∞ ∞
i y k −y2 i Γ(n + 1/2)
W (z) = e dy = ,
π z k+1 π n=0 z 2n+1
k=0 −∞
where Γ(x) is the Euler Gamma function. By taking the leading terms of the
expansion in inverse powers of z one easily obtains the asymptotic behavior
1 a 1 3v 2 − a2
H(v, a) ∼ √ 2 2
1+ + ···
π a +v 2 (a2 + v 2 )2
1 v 1 v 2 − 3a2
L(v, a) ∼ √ 1+ + ··· . (5.57)
π a2 + v 2 2 (a2 + v 2 )2
iv) Derivatives.
From Eq. (5.54) we have
dW (z) 2i
= −2z W (z) + √ ,
dz π
168 CHAPTER 5
from which we deduce
∂ ∂ 2i
(H + i L) = −i (H + i L) = −2 (v + i a) (H + i L) + √ ,
∂v ∂a π
whence
∂H(v, a) ∂L(v, a)
= = 2 −vH(v, a) + aL(v, a)
∂v ∂a
∂H(v, a) ∂L(v, a) 1
=− = 2 − √ + aH(v, a) + vL(v, a) , (5.58)
∂a ∂v π
a set of formulae which allows to express the derivatives of H(v, a) and L(v, a) in
terms of the functions themselves.
v) Power series expansion.
For a 1, a useful series expansion can be easily derived from Eqs. (5.58). We
give here the expansion up to the fourth order in powers of a
−v 2 2a , - , - −v2
H(v, a) = e + √ 2vD(v) − 1 + a2 1 − 2v 2 e
π
4a 3
a4 , - −v2
+ √ v 2 − 1 + v(3 − 2v 2 )D(v) + 3 − 12v 2 + 4v 4 e + ···
3 π 6
2 −v 2 2 2 −v 2
L(v, a) = √ D(v) − 2av e + √ a2 v + (1 − 2v 2 )D(v) − a3 v(3 − 2v 2 ) e
π π 3
1
+ √ a4 v(5 − 2v 2 ) + (3 − 12v 2 + 4v 4 )D(v) + · · ·
3 π
vi) Symmetry properties.
From Eqs. (5.45) the following symmetry properties are easily derived
H(−v, a) = H(v, a) , L(−v, a) = −L(v, a) , (5.59)
which means that H is symmetrical in v, while L is antisymmetrical.
vii) Integral properties.
From Eq. (5.52), integrating in the complex plane z along the real axis R, and
using the residue theorem, we obtain
∞ ∞
i −y 2 1 −y 2 √
W (z) dz = dy e dz = e dy = π,
R π R z−y
−∞ −∞
whence
∞ ∞
√
H(v, a) dv = π , L(v, a) dv = 0 . (5.60)
−∞ −∞
MATTER-RADIATION INTERACTION (CLASSICAL) 169
This formula proves that the profiles φα in Eqs. (5.44) are normalized to unity in
frequency.
viii) Convolutions.
Consider the integral
∞
W (v − v0 , a) W (v − v0 , a ) dv .
−∞
Using Eqs. (5.51)-(5.52) and inverting the order of the integrals one gets
∞
W (v − v0 , a) W (v − v0 , a ) dv =
−∞
∞ ∞ ∞
1 −y 2 −y 2 1
=− 2 dy e dy e dv .
π (v − v0 − y + i a) (v − v0 − y + i a )
−∞ −∞ −∞
The last integral can be performed via the residue theorem. Since both poles lie in
the half plane Im v < 0, one obtains
∞
W (v − v0 , a) W (v − v0 , a ) dv = 0 . (5.61)
−∞
Similarly,
∞
W (v − v0 , a) W (v − v0 , a )∗ dv =
−∞
∞ ∞ ∞
1 −y 2 −y 2 1
= 2 dy e dy e dv ,
π (v − v0 − y + i a) (v − v0 − y − i a )
−∞ −∞ −∞
and using again the residue theorem we obtain
∞
W (v − v0 , a) W (v − v0 , a )∗ dv =
−∞
∞ ∞
2i −y 2 −y 2 1
= dy e dy e .
π v0 − v0 + y − y + i (a + a)
−∞ −∞
170 CHAPTER 5
The double integral can be evaluated by the substitution
1 1
x = √ y − y , x = √ y + y , (5.62)
2 2
which leads to
∞
W (v − v0 , a) W (v − v0 , a )∗ dv =
−∞
∞ ∞ −x2
2i −x 2 e
= e dx
√ dx .
π v0 − v0 − 2 x + i (a + a)
−∞ −∞
Performing the integral in dx and recalling Eq. (5.52), one finally obtains
∞
√ v0 − v0 a + a
W (v − v0 , a) W (v − v0 ∗
, a ) dv = 2π W √ , √ . (5.63)
2 2
−∞
Substitution of Eq. (5.50) into Eqs. (5.61) and (5.63) yields
∞ .
π v − v a + a
H(v − v0 , a) H(v − v0 , a ) dv = H 0√ 0 , √
2 2 2
−∞
∞ .
π v − v a + a
L(v − v0 , a) L(v − v0 , a ) dv = H 0√ 0 , √
2 2 2
−∞
∞ .
π v0 − v0 a + a
L(v − v0 , a) H(v − v0
, a ) dv = L √ , √ . (5.64)
2 2 2
−∞
These formulae can be easily generalized to more complicated convolutions. Con-
sidering the integrals
∞
v − v0 v − v0
W ,a W , a dv
b b
−∞
and
∞ ∗
v − v0 v − v0
W ,a W , a dv ,
b b
−∞
MATTER-RADIATION INTERACTION (CLASSICAL) 171
we obtain with similar manipulations1
∞
v − v0 v − v0
H ,a H , a dv =
b b
−∞
√
b b π v − v0 a b + ab
= √ H √0 ,√
b2 + b 2 b2 + b 2 b2 + b 2
∞
v − v0 v − v0
L ,a L , a dv =
b b
−∞
√
b b π v − v0 a b + ab
= √ H √0 ,√
b2 + b 2 b2 + b 2 b2 + b 2
∞
v − v0 v − v0
L ,a H , a dv =
b b
−∞
√
b b π v − v0 a b + ab
= √ L √0 ,√ . (5.65)
b2 + b 2 b2 + b 2 b2 + b 2
ix) Maxima and minima.
The function H(v, a), considered as a function of v, has a single maximum at v = 0.
The maxima and minima of the function L(v, a) with respect to v can be found by
solving the implicit equation (see Eqs. (5.58))
1
a H(v0 , a) + v0 L(v0 , a) = √ .
π
Using numerical methods, the value of v0 is found to be 0.9241 for a = 0, and to
increase almost linearly to 1.5372 for a = 1. For larger values of a, v0 converges
asymptotically towards a as it can be deduced from Eq. (5.57). Obviously, there is
a corresponding minimum at v = −v0 .
x) Numerical methods.
Several algorithms have been proposed for the numerical computation of the func-
tions H(v, a) and L(v, a). The most extensively used for astrophysical applications
are those by Reichel (1968), Hui et al. (1978), and Humlicek (1982). The values
obtained using these algorithms can be checked against the tabulated values of the
complex function W (z) given by Faddeeva and Terent’ev (1961).
1
Transformation (5.62) should be replaced by the following
by − b y b y + by
x= √ , x = √ .
2
b + b 2 b2 + b 2
172 CHAPTER 5
5.5. Symmetry Properties of the Transfer Equations
for Polarized Radiation
In Sect. 5.2 we have derived, in the framework of classical electrodynamics, a
compact expression for the propagation matrix K which describes the transfer of
polarized radiation, and we have seen that this matrix satisfies the remarkable
symmetry properties summarized in Eq. (5.27). It will be shown in Sect. 6.7 that
the quantum-mechanical treatment leads to the same symmetry properties, so that
it seems natural to ascribe them to general physical principles. Following Landi
Degl’Innocenti and Landi Degl’Innocenti (1981) we present here a series of argu-
ments leading to an alternative proof of Eq. (5.27).
Consider a radiation beam of frequency ν, characterized by the Stokes parameters
I, Q, U , and V , propagating through an arbitrary medium within the solid angle
By analogy with the usual transfer equation for unpolarized
dΩ in the direction Ω.
radiation we can write for the Stokes parameters a transfer equation of the form
d 3
S =− Kij Sj + εi (i = 0, . . . , 3) , (5.66)
ds i j=0
where S is the Stokes vector defined by
S † ≡ (S0 , S1 , S2 , S3 ) = (I, Q, U, V ) ,
K is a 4 × 4 matrix which is composed in principle of 16 independent quantities,
and ε is the emission vector
ε† ≡ (ε0 , ε1 , ε2 , ε3 ) = (εI , εQ , εU , εV ) .
We assume here that both K and ε do not depend on the Stokes vector S. This
may indeed appear as a rather restrictive assumption, because there are many
physical situations, mainly of astrophysical interest, where the properties of the
medium at any given point depend in fact upon the local radiation field. It should
be remarked, however, that the angular spread of the radiation beam that we are
considering can always be made sufficiently small for the properties of the medium
to be not affected by the beam itself. Keeping this argument in mind, we can
derive some important properties of K and ε by varying arbitrarily – in a sort of
Gedankenexperiment – the value of the Stokes vector S appearing in Eq. (5.66).
If we start by considering the limiting case of very intense radiation beams, the
contribution of the emission vector in Eq. (5.66) can be simply neglected and the
transfer equations reduce to
d 3
Si = − Kij Sj (i = 0, . . . , 3) , (5.67)
ds j=0
MATTER-RADIATION INTERACTION (CLASSICAL) 173
from which we deduce
1 d 2
I − Q2 − U 2 − V 2 =
2 ds
= −K00 I 2 + K11 Q2 + K22 U 2 + K33 V 2
+ (K10 − K01 )IQ + (K20 − K02 )IU + (K30 − K03 )IV
+ (K12 + K21 )QU + (K13 + K31 )QV + (K23 + K32 )U V . (5.68)
As proved in Chap. 1 (see Eq. (1.47)), the quantity (I 2 − Q2 − U 2 − V 2 ) is always
positive, except for the case of a totally polarized radiation beam when it reduces
to zero. Therefore, if we consider a totally polarized radiation beam,
S0† = (I0 , Q0 , U0 , V0 ) with I02 = Q20 + U02 + V02 ,
the right-hand side of Eq. (5.68) must be ≥ 0,
(K11 − K00 )Q20 + (K22 − K00 )U02 + (K33 − K00 )V02
+(K10 − K01 )I0 Q0 + (K20 − K02 )I0 U0 + (K30 − K03 )I0 V0
+(K12 + K21 )Q0 U0 + (K13 + K31 )Q0 V0 + (K23 + K32 )U0 V0 ≥ 0 . (5.69)
At this point we make a distinction between depolarizing and non-depolarizing
media according to whether the greater-than or the equal sign holds in Eq. (5.69).
In a non-depolarizing medium, by definition, a beam which is totally polarized
remains so in the process of propagation (provided that no other terms are added
to the right-hand side of Eq. (5.67)), while in a depolarizing medium this feature
is lost and the propagation shows a typical irreversible character.
For non-depolarizing media – the only case that will be considered here – equality
(5.69) must hold for whatever realization of the Stokes vector S0 , or, in other words,
for any values assigned to Q0 , U0 , V0 satisfying the condition I02 = (Q20 + U02 + V02 ).
Therefore all the brackets in the left-hand side of Eq. (5.69) must vanish, which
implies
K00 = K11 = K22 = K33
K01 = K10 , K02 = K20 , K03 = K30
K12 = −K21 , K13 = −K31 , K23 = −K32 .
These are just the symmetry relations contained in Eq. (5.27). In the light of the
arguments presented above we see that these relations are a direct consequence
of: a) the definition of the Stokes parameters (which implies the inequality I 2 ≥
Q2 + U 2 + V 2 ); b) the linear character of Eq. (5.67) (which is equivalent to the
assumption K independent of S); c) the reversible character of the interaction of
radiation with the ambient medium (which implies the equal sign in Eq. (5.69)).
174 CHAPTER 5
When conditions b) and c) are fulfilled the propagation matrix K can be written
in the form ⎛ ⎞
ηI ηQ ηU ηV
⎜η ηI ρV −ρU ⎟
K=⎝ Q ⎠ .
ηU −ρV ηI ρQ
ηV ρU −ρQ ηI
We will now prove that one, or even two, of the seven independent coefficients
appearing in the expression of K can be set equal to zero by an appropriate choice
of the reference direction which defines the Stokes parameters. Recalling the results
derived in Sect. 1.9 (see Eq. (1.45) and Fig. 1.10), the Stokes vector S relative to
the ‘new’ reference direction characterized by the angle α can be written in the
form
S = R(α) S ,
where ⎛ ⎞
1 0 0 0
⎜ 0 cos 2α sin 2α 0⎟
R(α) = ⎝ ⎠ .
0 − sin 2α cos 2α 0
0 0 0 1
Thus in the new reference frame the expression for the propagation matrix is
K = R(α) K R−1 (α) ,
whence we deduce after some easy algebra
ηI = ηI , ηV = ηV , ρV = ρV
ηQ = cos 2α ηQ + sin 2α ηU , ρQ = cos 2α ρQ + sin 2α ρU
ηU = − sin 2α ηQ + cos 2α ηU , ρU = − sin 2α ρQ + cos 2α ρU . (5.70)
These formulae show that while the quantities ηI , ηV , and ρV are invariant with
respect to the choice of the reference direction, ηQ and ηU change into each other
according to a rotation through an angle 2α in the ηQ -ηU plane, and the same
happens for ρQ and ρU . It follows that the two quantities ηL and ρL defined by
ηL2 = ηQ
2 2
+ ηU , ρ2L = ρ2Q + ρ2U
are invariant. Moreover, it is possible to choose a particular reference direction,
specified by the angle α1 , such that ηU is zero, and another reference direction,
specified by the angle α2 , such that ρU is zero. The angles α1 and α2 may be
different in principle, but in most cases of interest they are found to be equal.
When this happens we will call preferred reference direction the one for which ηU
and ρU are both zero.
Obviously, the expressions derived in Sect. 5.2 for the elements of the propagation
matrix K are consistent with the transformation law given in Eqs. (5.70). This
can be proved directly from the general equations (5.26) by taking into account
MATTER-RADIATION INTERACTION (CLASSICAL) 175
the transformation properties of the unit vectors ea , eb under rotation of the ref-
erence direction. The proof is straightforward and will not be given here. In the
special case of the propagation of radiation in a magnetized medium (Sect. 5.3) one
can easily see from Eqs. (5.36) that the preferred reference direction lies (indepen-
dently of frequency) in the plane containing the magnetic field and the direction
of propagation.1
The above remarks show that when a preferred reference direction can be defined
the propagation matrix K can be cast into the simpler form
⎛ ⎞
ηI ηQ 0 ηV
⎜ ηQ ηI ρV 0 ⎟
K=⎝ ⎠ .
0 −ρV ηI ρQ
ηV 0 −ρQ ηI
Going back to Eq. (5.66), we can also deduce some interesting properties of the
emission vector ε by considering the limiting case of very weak intensity of the
radiation beam. Under this limit the transfer equations reduce to
dSi
= εi (i = 0, . . . , 3) ,
ds
and by integration over the interval ∆s one gets
Si (s + ∆s) = Si (s) + εi ∆s .
Therefore, if we suppose Si (s) = 0 we obtain that the following inequality must be
satisfied
εI ≥ ε2Q + ε2U + ε2V . (5.71)
In the case of the propagation of radiation through a magnetized medium this
relation can be proved directly from Eqs. (5.38).
It should be noticed that the corresponding inequality for the components of the
dichroism matrix, namely
ηI ≥ 2 + η2 + η2 ,
ηQ U V (5.72)
does not hold in general. However, under Local Thermodynamic Equilibrium or,
more generally, when one can write
⎛ ⎞ ⎛ ⎞
εI ηI
⎜ εQ ⎟ ⎜η ⎟
⎝ ⎠=S⎝ Q⎠ , (5.73)
εU ηU
εV ηV
inequality (5.72) follows directly from (5.71).
1
Indeed, there is another preferred reference direction at 90 ◦ from the former.
176 CHAPTER 5
Another property of the emission vector concerns its transformation under a
rotation of the reference direction. It is easy to prove that the rotation leading to
Eqs. (5.70) induces the following transformation on the components of the emission
vector
εI = εI
εQ = cos 2α εQ + sin 2α εU
εU = − sin 2α εQ + cos 2α εU
εV = εV .
These formulae show that the quantity εL defined by
ε2L = ε2Q + ε2U
is invariant, and that it is possible to choose a particular reference direction, charac-
terized by the angle α3 , such that εU is zero. When a preferred reference direction
can be defined (ηU = ρU = 0) it often happens that εU is zero as well.
5.6. Geometrical Interpretation of the Transfer Equations
for Polarized Radiation
In Sect. 1.9 we have introduced the concept of Poincaré sphere as a mapping be-
tween Stokes vectors and points within a sphere of unit radius (see Fig. 1.11).
Thus it is quite natural to give a geometrical interpretation to the radiative trans-
fer equations for polarized radiation in terms of the motion of the representative
point within the Poincaré sphere. This is schematically illustrated in Fig. 5.5.
In a right-handed orthogonal system (xyz) we consider the formal vectors p , η ,
ρ , defined by
Q U V
p=
, ,
I I I
η = ηQ , ηU , ηV
ρ = ρQ , ρU , ρV
ε , εQ εU εV -
= = , , ,
I I I I
and the scalar quantity
ε
I = I .
I
With these positions the transfer equations for polarized radiation can be trans-
formed into the following
dI
= −(ηI +
η · p − I ) I (5.74a)
ds
d
p
= − η · p ) p + ρ
η + ( ×p + − I p . (5.74b)
ds
MATTER-RADIATION INTERACTION (CLASSICAL) 177
on the
surface
Fig.5.5. The transfer of polarized radiation in an arbitrary medium can be described in terms of
the motion of the representative point inside the Poincaré sphere. The point cannot pierce the
surface of the sphere but may describe a curve that is tangent to it, like in the point P. The
starting point P0 represents the radiation entering the medium.
The scalar equation describes the transfer of the beam intensity and it can be
considered as the equation which controls the scaling factor connecting the Stokes
parameters with their fractional value given by the vector p . It is interesting to
note that this equation differs from the standard equation for unpolarized radiation
only for the presence of the term ( η · p ) which adds to the (generalized) absorption
coefficient ηI . Note also that the dispersion term, ρ , does not appear directly
in this equation, nevertheless it affects the intensity via the term (η · p ) (since p
depends on ρ ).
The vector equation describes the motion of the representative point inside the
Poincaré sphere. The first term in the right-hand side tends to align the fractional
polarization p with the vector − η . If we consider for instance the simple case
where η does not depend on s, and if we take p = 0 (unpolarized radiation) as the
boundary condition at s = s0 , when the ray has traveled a distance ∆s from s0 we
obtain from this term
p (∆s) = − η ∆s ,
and, as ∆s grows, the representative point might pierce the Poincaré sphere. This
is however prevented by the second term, which is opposed to the first one and
whose relative importance grows as the point approaches the surface of the sphere.
Clearly this term describes a phenomenon of saturation for the fractional linear
polarization.
In this respect it is also interesting to consider the effect of the term (η · p )
in the scalar equation for the intensity. In the simple situation discussed above
(
η = const., p = 0 at s = s0 ) this term brings a negative contribution to the
absorption coefficient, since p is directed along −η . Therefore, as the radiation
beam propagates through the medium, the polarization grows and the medium
becomes more and more transparent. We are facing again a saturation phenomenon
which is strictly connected with the former.
178 CHAPTER 5
ρ
ρ
Fig.5.6. Geometrical interpretation of Faraday rotation (a) and Faraday pulsation (b). In case
(a) only the component ρV is non-zero and we have a transformation from Q to U , and vice versa.
In case (b) the component ρU induces a transformation from Q to V , and vice versa.
As an illustrative example, let us consider a low-temperature medium (i.e. a
medium with negligible emission) which is opaque to left-handed circular polariza-
tion, and suppose that an unpolarized radiation beam is entering the medium at
s = s0 . During the propagation, the left-handed circular polarization will decrease
exponentially, so that right-handed circular polarization only will remain in the
beam and the medium will be totally transparent.
Let’s now consider the third term in the right-hand side of Eq. (5.74b). This term
causes a precession of the vector p about the vector ρ ; in this motion the absolute
value of the fractional polarization is unchanged, while the type of polarization
changes. In particular, the z-component of the vector ρ , ρV , induces a rotation of
the representative point about the z-axis, which implies a rotation of the direction
of maximum linear polarization. The x and y components (ρQ and ρU ) give rise to
a rotation about an axis belonging to the x-y plane, which implies a transformation
from linear to circular polarization and vice versa. These two phenomena, which
are illustrated in Fig. 5.6, will be referred to in the following as Faraday rotation
and Faraday pulsation, respectively, although these names should more properly be
employed when dealing with the propagation of polarized radiation in a magnetized
medium far from resonances.
An important property of the term ( ρ× p ) is that it can be efficient in removing
the alignment between the vectors p and − η that tends to arise in the schematic
situation discussed previously. Thus the presence of dispersion reduces the impor-
tance of the saturation mechanisms that we have just pointed out. The only case
where the effects of dispersion can be neglected is when the three formal vectors η ,
ρ , and point in the same direction. In this case, if we add the further condition
that the radiation beam is unpolarized when entering the medium, the vector p
will also be aligned with the same direction, and the term ( ρ × p ) vanishes.
MATTER-RADIATION INTERACTION (CLASSICAL) 179
Comments similar to those on the terms − η and (η · p )
p can be made for the
last two terms in Eq. (5.74b), and −I p . The fourth term tends to align the
polarization p with , but when p starts growing the other term comes in and
prevents, through a saturation mechanism, the possibility of piercing the surface
of the Poincaré sphere.
Finally, let us consider the special case described by Eq. (5.73) (which includes
the LTE case). Defining the reduced scalar source function s̃ as
S
s̃ = ,
I
we have from Eq. (5.74b)
d
p
= −(1 − s̃) η + (
η · p − s̃ ηI ) p + ρ
× p ,
ds
an equation containing the driving term, the saturation term, and the dispersion
term.
5.7. Resonance Scattering and the Hanle Effect
We have seen in Sects. 5.1-5.2 how it is possible to deduce the transfer equations
for polarized radiation using the concepts of classical electrodynamics. The general
equations (5.25)-(5.26) have then been applied to the case of an atomic vapor
embedded in a magnetic field. By schematizing the atomic system as a negative
charge oscillating under the action of an elastic, restoring force, we have derived
Eqs. (5.36) which give the basic quantities describing the transfer of polarized
radiation in a magnetic field.
In this section we will use the same atomic model to deduce the laws of resonance
scattering and of the Hanle effect. Starting from Eq. (3.34), we can write for the
frequency-integrated radiation emitted per unit time in the solid angle dΩ by an
atomic oscillator embedded in a magnetic field
c
=
dI˜ij (Ω) J (r) r2 dΩ
8π ij
∞
2π 3 e20 ∗
A∗α Aβ
= dΩ Cαi Cβj ν 4 Fα (ν)∗ Fβ (ν) dν , (5.75)
c3 τ
αβ −∞
where the symbols have the same meaning as in Sect. 3.2.
To describe a scattering experiment we will now assume the initial values of
the amplitudes, Aα , to be proportional to the corresponding components Eα of
(see
the resonant electric field impinging on the oscillator from the direction Ω
Fig. 5.7). Note that the only difference between the present case and the case
considered in Sect. 3.2 lies in the assumption on the amplitudes Aα . To obtain
180 CHAPTER 5
atomic magnetic field
oscillator
incident scattered
radiation radiation
Ω Ω
Ω Ω
to the direction Ω
Fig.5.7. A radiation beam is scattered from the direction Ω in the presence of
a magnetic field.
the emission in the Zeeman effect we assumed the amplitudes to be completely
uncorrelated,
A∗α Aβ = |A|2 δαβ ;
by contrast, we now assume
A∗α Aβ = K Eα∗ Eβ dΩ , (5.76)
where dΩ is an infinitesimal solid angle around the direction Ω . Note that the
frequency dependence of the components Eα in the above equation is not specified:
they are supposed to be constant over a spectral interval centered at the reso-
nance frequency ν0 and much larger than the Larmor frequency νL (flat spectrum
approximation).
If we now introduce two mutually orthogonal unit vectors perpendicular to the
,
direction Ω
ei∗ · ej = δij , ei · Ω
= 0 (i, j = 1, 2) ,
we can write
Eγ = ∗
Cγi Ei , (5.77)
i
where the direction cosines Cγi are given by
Cγi = uγ · ei∗ .
Substituting Eq. (5.77) into Eq. (5.76), and averaging over all the possible realiza-
tions of the incident electric field, we have
K ∗
A∗α Aβ = C C I (Ω ) dΩ , (5.78)
k ij αi βj ij
where k is the constant appearing in Eqs. (1.34) and Iij (Ω ) is the polarization
tensor of the incident radiation.
MATTER-RADIATION INTERACTION (CLASSICAL) 181
The dimensional constant K can be determined by considering the special case
of thermodynamic equilibrium. In this case the incident radiation (isotropic and
unpolarized) at the resonant frequency ν0 is characterized by the polarization tensor
ν02 kB T
Iij = δij ,
c2
where we have used the classical expression for the Planck function. Substitution
into Eq. (5.78) gives
5
K ν02 kB T
A∗α Aβ =
Cαi ∗
Cβi dΩ
k c2 i
5
K ν02 kB T
= · u )(Ω
δαβ − (Ω · u ∗ ) dΩ
α β
k c2
8π K ν02 kB T
= δαβ . (5.79)
3 k c2
On the other hand, the amplitudes of the atomic oscillations in thermodynamic
equilibrium are given by (see Eq. (3.35) for a similar derivation)
γτ kB T
A∗α Aβ = δ . (5.80)
2π 2 ν02 m αβ
Comparison of expressions (5.79) and (5.80) leads to
K 3 γτ c2
= . (5.81)
k 16π 3 ν04 m
Using Eqs. (5.78) and (5.81), Eq. (5.75) becomes
dI˜ij (Ω)
=
∞
3 πe20 ∗
∗ dΩ
= γ Cαi Cβj Cαk Cβl Ikl (Ω ) dΩ Fα (ν)∗ Fβ (ν) dν , (5.82)
2 mc 4π
αβ kl −∞
where we have extracted the factor ν04 from the integral since the Fourier transforms
Fα (ν) are substantially non-zero only for ν ν0 (see Eq. (3.33)).
The frequency integral can be easily evaluated with the help of the residue the-
orem
∞
1 1
Fα (ν)∗ Fβ (ν) dν = , (5.83)
γ 1 + i (α − β)H
−∞
where
2πνL e B
H= = 0 . (5.84)
γ 2mcγ
182 CHAPTER 5
The quantity H is proportional to the magnetic field intensity; numerically we have
H = 0.879 B/γ , (5.85)
with B in G and γ in 107 s−1 .
Substitution of Eq. (5.83) into Eq. (5.82) leads to
dI˜ij (Ω)
= σ Ω
Tij,kl (Ω, ; B ) dΩ dΩ ,
) I (Ω (5.86)
cl kl
4π
kl
where
πe20
σcl =
mc
is the classical (frequency-integrated) cross section of the atomic oscillator, and T
– the so-called scattering phase matrix for the polarization tensor – is defined by
Ω
Tij,kl (Ω, )= 3
; B ∗
Cαi
Cβj Cαk ∗
Cβl
1
. (5.87)
2 1 + i (α − β)H
αβ
This matrix depends on the geometry of the scattering event and on the magnitude
and direction of the magnetic field vector; moreover, it depends on the choice of
the unit vectors (e1 , e2 ) and (e1 , e2 ). Note that expression (5.87) is valid both
for real unit vectors and, more generally, for complex unit vectors of the form of
Eqs. (1.41).
The following symmetry property can be proved directly from Eq. (5.87)
Ω
Tij,kl (Ω, ; B
)∗ = T
ji,lk (Ω, Ω ; B ) . (5.88)
If the direction of the magnetic field vector is reversed, one gets a sign inversion of
the term in H in Eq. (5.87) (see Eqs. (3.23), (3.27), and (3.33)). This leads to the
further property
Ω
Tij,kl (Ω, ; −B
)=T
lk,ji (Ω , Ω; B ) . (5.89)
In the case of zero magnetic field the expression for the matrix T reduces to a much
simpler form. Summation over the indices α and β yields
; 0) ≡ T 3
Ω
Tij,kl (Ω, ij,kl (Ω, Ω ) = D D∗ ,
2 ik jl
where the direction cosines Dik are given by
Dik = ei · ek∗ . (5.90)
Equation (5.86) can be transformed into an equivalent equation relating the
Stokes parameters of the incident and scattered radiation. As noticed several
times, the relation between Stokes parameters and polarization tensor depends
on the choice of the unit vectors ei and ei . The most natural choice is to take,
MATTER-RADIATION INTERACTION (CLASSICAL) 183
both for the incident and for the scattered radiation, the reference direction unit
vector and the associated unit vector defined in Sect. 1.6. In other words, we take
e1 = ea (Ω) e1 = ea (Ω
)
e = e (Ω)
2 b
e = e (Ω
2 b
) .
In this case the required relation is given by Eqs. (1.40), which imply that the
scattered in the solid angle dΩ by
frequency-integrated Stokes parameters dS̃i (Ω)
an atomic oscillator are given by
=
dS̃i (Ω) (σi )nm dI˜mn (Ω)
,
nm
with σi the Pauli spin matrices defined in Eq. (1.17). Using again Eqs. (1.40), we
obtain
=σ
dS̃i (Ω) Ω
Rij (Ω, ; B ) dΩ dΩ
) S (Ω (i, j = 0, . . . , 3) , (5.91)
cl j
j
4π
where the scattering matrix R for the Stokes parameters, sometimes called the
phase matrix , is given by
Ω
Rij (Ω, ; B
)=
1 Ω ; B)
= (σi )lk (σj )mn Tkl,mn (Ω,
2
klmn
3
∗ ∗ 1
= (σi )lk (σj )mn Cαk Cβl Cαm Cβn . (5.92)
4 1 + i (α − β)H
klmn αβ
In the case of zero magnetic field the matrix R reduces to the so-called Rayleigh
phase matrix
Ω
Rij (Ω, ; 0) ≡ R (Ω, ) = 3
Ω ∗
(σi )lk (σj )mn Dkm Dln ,
ij
4
klmn
where the direction cosines Dij are defined in Eq. (5.90).
Since the Stokes parameters are real, the scattering matrix must be real as well,
Ω
Rij (Ω, ; B
)∗ = R (Ω, ; B
Ω ); (5.93)
ij
this can be checked directly from Eq. (5.92) recalling that
(σi )∗nm = (σi )mn .
Moreover, from Eq. (5.89) one gets
Ω
Rij (Ω, ; −B
) = R (Ω , Ω;
B). (5.94)
ji
184 CHAPTER 5
scattering atom
incident
radiation
magnetic field
α
scattered radiation
Fig.5.8. The Hanle effect is illustrated by this simple scattering experiment. For zero magnetic
field the polarization of the scattered beam is perpendicular to the plane of scattering; when
a magnetic field is present, the polarization is reduced and its direction is rotated through an
angle α.
In the particular case of zero magnetic field this equation becomes
Ω
Rij (Ω, ) = R (Ω , Ω)
, (5.95)
ji
which is known as the Helmoltz principle of reciprocity for a scattering event.
Thus Eq. (5.94) represents the generalization of this principle to the case where a
magnetic field is present.
The equations that we have derived in this section describe the scattering of
polarized radiation by an atomic system, schematized as a harmonic oscillator, in
the presence of a magnetic field. In the limiting case of zero magnetic field the
results here obtained reduce to the classical results of Rayleigh for the scattering
of polarized radiation.1
In the presence of a magnetic field the laws of scattering are deeply modified and a
new phenomenon – known as Hanle effect – appears. Discovered in the laboratory
in the early years of this century (Hanle, 1924), this effect usually produces a
depolarization of the scattered radiation and a rotation of its plane of polarization
(see Fig. 5.8). However, this holds only for particular geometries of the scattering
event – probably the most used in laboratory experiments and the most important
for astrophysical applications. For different geometries the effect of the magnetic
field can be very different, and in fact the polarization of the scattered radiation
can even be increased relative to the non-magnetic situation. Thus, in general
terms, one can only state that the Hanle effect consists in a modification of the
scattered radiation due to the presence of a magnetic field . These points will be
further clarified in Sect. 5.9.
1
It should be remarked that Rayleigh’s law of scattering can be applied to a broader class
of phenomena and not only to the scattering by a harmonic oscillator – the simple case on which
we have based our deduction. Thomson scattering by free electrons, or scattering by a dielectric
sphere having a radius much smaller than the wavelength of the radiation, are just two examples
where Rayleigh’s law can still be applied in full generality. The only difference with the case
considered here is in the cross section σcl appearing in Eqs. (5.86) and (5.91).
MATTER-RADIATION INTERACTION (CLASSICAL) 185
θ
θ Ω
Ω θ
γ
γ
χ
χ
χ
Fig.5.9. The most general geometry for the scattering event. The unit vector 1, lying in the plane
is the reference direction unit vector e (Ω),
perpendicular to Ω, while 2 is the associated unit
a
The unit vectors 3 and 4 are e (Ω
vector eb (Ω). ) and e (Ω
), respectively.
a b
Equation (5.92) describes the Hanle effect from the classical point of view; we
will see in Chap. 10 how this equation is generalized in the quantum theory.
5.8. The Scattering Phase Matrix in a Particular Case
In the previous section we have obtained a general expression for the scattering
phase matrix that can be applied to any geometrical configuration of the scattering
event and of the magnetic field and to any choice of the reference directions for the
Stokes parameters. For practical applications it is however useful to have explicit
analytical expressions for the scattering phase matrix in particular geometrical
situations.
The most general geometry is illustrated in Fig. 5.9, where the three directions Ω,
, and B
Ω are specified by their polar angles and where the reference directions are
defined by the angles γ and γ . In this geometry the scattering phase matrix will
depend on 8 parameters, namely θ, θ , θB , (χ − χB ), (χ − χB ), γ, γ , and H, and
the calculation of one element Rij will require the sum of 144 (= 32 × 24 ) terms.
It is not worth here to develop in detail these calculations; the interested reader is
referred to Landi Degl’Innocenti and Landi Degl’Innocenti (1988), where an ana-
lytical expression for the phase matrix is derived under the assumption γ = γ = 0
(which means that the positive Q-directions for the incident and scattered beams
lie in the meridian planes of Fig. 5.9). Being deduced from the quantum theory,
186 CHAPTER 5
Ω
θ θ
Ω
Fig.5.10. A particular geometry for the calculation of the scattering phase matrix. The polar
and Ω
angles of the directions Ω are reckoned from the magnetic field direction. The meaning of
the unit vectors 1, 2, 3, 4 is the same as in Fig.5.9.
the expressions presented in the paper just quoted are in fact more general than
those obtainable from Eq. (5.92); however, they reduce to the classical results with
easy transformations.
Some general properties of the scattering phase matrix relative to the geometrical
configuration of Fig. 5.9 will be given in Sect. 5.12. Here we want to derive the
explicit expression of this matrix for the much simpler geometry illustrated in
Fig. 5.10.
Now the magnetic field vector is parallel to the z-axis and the reference direction
for the Stokes parameters lies in the meridian plane containing the magnetic field
and the propagation direction. The various unit vectors are given by
= e = cos θ cos χ ı + cos θ sin χ j − sin θ k
ea (Ω) 1
= e = − sin χ ı + cos χ j
eb (Ω) 2
ea (Ω
) = e = cos θ cos χ ı + cos θ sin χ j − sin θ k
1
eb (Ω
) = e = − sin χ ı + cos χ j
2
u0 = k
1
u±1 = √ ∓ ı + i j ,
2
and the direction cosines can be easily evaluated
MATTER-RADIATION INTERACTION (CLASSICAL) 187
1 −iχ i −iχ
C1 1 = − √ cos θ e C1 2 = √ e
2 2
C0 1 = − sin θ C0 2 = 0
1 iχ i iχ
C−1 1 = √ cos θ e C−1 2 = √ e ,
2 2
with analogous expressions for Cαi .
From these expressions the elements of the scattering phase matrix for the polar-
ization tensor, given by Eq. (5.87), can be evaluated via some heavy algebra that
is left to the reader as an exercise. The results are the following
3 3 1
T11,11 = 1 − µ2 − µ2 + µ2 µ2 + 2µµ 1 − µ2 1 − µ2 C1 + µ2 µ2 C2
2 2 2
3 1
T11,12 = T11,21 = −µ 1 − µ2 1 − µ2 S1 − µ2 µ S2
2 2
3 1
T12,11 = T21,11 = µ 1 − µ2 1 − µ2 S1 + µµ2 S2
2 2
3 1 2 1 2
T11,22 = µ − µ C2
2 2 2
3 1 2 1 2
T22,11 = µ − µ C2
2 2 2
3 1 1
T12,12 = T21,21 = µµ + 1 − µ2 1 − µ2 C1 + µµ C2
2 2 2
3 1 1
T12,21 = T21,12 = − µµ + µµ C2
2 2 2
3 1
T12,22 = T21,22 = − µ S2
2 2
3 1
T22,21 = T22,12 = µ S2
2 2
3 1 1
T22,22 = + C , (5.96)
2 2 2 2
where
µ = cos θ µ = cos θ
C1 = cos α1 cos α1 + χ − χ S1 = cos α1 sin α1 + χ − χ
C2 = cos α2 cos α2 + 2 (χ − χ) S2 = cos α2 sin α2 + 2 (χ − χ) (5.97)
and where the angles α1 and α2 , implicitly defined by
tan α1 = H , tan α2 = 2H , (5.98)
188 CHAPTER 5
are particularly useful to express the quantities (1 ± iH)−1 and (1 ± 2iH)−1 in the
form
∓ iα1
(1 ± iH)−1 = cos α1 e = cos2 α1 ∓ i sin α1 cos α1
∓ iα2
(1 ± 2iH)−1 = cos α2 e = cos2 α2 ∓ i sin α2 cos α2 .
Note that the intensity of the magnetic field is contained only in the quantities C1 ,
C2 , S1 , S2 . In particular, for zero magnetic field one gets
C1 = cos(χ − χ) S1 = sin(χ − χ)
C2 = cos 2(χ − χ) S2 = sin 2(χ − χ) ,
and for strong magnetic field (H 1)
C1 = S1 = C2 = S2 = 0 ,
so that the expressions of the matrix elements are considerably simplified.
Finally, substituting Eqs. (5.96) into Eq. (5.92) and using the explicit expressions
of the Pauli spin matrices (Eqs. (1.17)), one obtains for the scattering phase matrix
of the Stokes parameters the following expressions
Ω
R00 (Ω, ) = 3 3 − µ2 − µ2 + 3µ2 µ2 + 3 µµ 1 − µ2 1 − µ2 C
; B
1
8 2
3
+ 1 − µ2 1 − µ2 C2
8
Ω
R01 (Ω, ) = 3 1 − 3µ2 1 − µ2 + 3 µµ 1 − µ2 1 − µ2 C
; B
1
8 2
3
− 1 − µ2 1 + µ2 C2
8
Ω
R02 (Ω, ) = − 3 µ 1 − µ2 1 − µ2 S + 3 1 − µ2 µ S
; B
1 2
2 4
R (Ω, Ω ; B
) =0
03
Ω
R10 (Ω, ) = 3 1 − µ2 1 − 3µ2 + 3 µµ 1 − µ2 1 − µ2 C
; B
1
8 2
3
− 1 + µ2 1 − µ2 C2
8
Ω
R11 (Ω, ) = 9 1 − µ2 1 − µ2 + 3 µµ 1 − µ2 1 − µ2 C
; B
1
8 2
3
+ 1 + µ2 1 + µ2 C2
8
Ω
R12 (Ω, ) = − 3 µ 1 − µ2 1 − µ2 S − 3 1 + µ2 µ S
; B
1 2
2 4
R (Ω, Ω ; B
) =0
13
MATTER-RADIATION INTERACTION (CLASSICAL) 189
Ω
R20 (Ω, ) = 3 µ 1 − µ2 1 − µ2 S − 3 µ 1 − µ2 S
; B
1 2
2 4
Ω
R21 (Ω, ) = 3 µ 1 − µ2 1 − µ2 S + 3 µ 1 + µ2 S
; B
1 2
2 4
Ω
R22 (Ω, ) = 3 1 − µ2 1 − µ2 C + 3 µµ C
; B
1 2
2 2
R (Ω, Ω ; B
) =0
23
Ω
R30 (Ω, ; B
) =0
Ω
R31 (Ω, ; B
) =0
Ω
R32 (Ω, ; B
) =0
Ω ; B
)= 3 3
R33 (Ω, µµ + 1 − µ2 1 − µ2 C1 . (5.99)
2 2
These formulae show that the V Stokes parameter (S3 ) is totally decoupled from
the other Stokes parameters in a scattering event.
In the limiting case of strong magnetic field one gets
⎛8 1 2 2 ⎞
3 + 3 (1 − 3µ )(1 − 3µ ) (1 − 3µ )(1 − µ ) 0
2 2
0
⎜ (1 − µ2 )(1 − 3µ2 ) 3(1 − µ2 )(1 − µ2 ) 0 0 ⎟
Ω
R(Ω, → ∞) = 3 ⎜
; B ⎜
⎟
⎟
8⎝ 0 0 0 0 ⎠
0 0 0 4µµ
which shows that the U Stokes parameter of the scattered radiation is zero irre-
spective of the direction and polarization of the incoming beam. If, in addition, the
incoming beam is unpolarized, the fractional polarization of the scattered radiation
is found to be
Q (1 − µ2 )(1 − 3µ2 )
= ,
I 3 − µ2 − µ2 + 3µ2 µ2
thus it is positive, negative, or zero according as (µ2 − 1/3) is negative, positive,
or zero. Defining the Van Vleck angle θV according to the formula
√
cos θV = 1/ 3 , (5.100)
which implies
θV = 54◦ .74 , (5.101)
one obtains
Q )
> 0 (linear polarization parallel to B for θV < θ < π − θV
I
Q ) for 0 < θ < θ
< 0 (linear polarization perpendicular to B V
I
or π − θV < θ < π .
190 CHAPTER 5
Ω
Ω
scattered
reference
beam unpolarized
direction
incident beam
Fig.5.11. Geometry for a 90◦ scattering. The magnetic field lies in the plane perpendicular to
the incident beam and makes an angle β with the direction of the scattered beam. The reference
direction for the Stokes parameters of the scattered radiation is perpendicular to the scattering
plane.
In other words, for an unpolarized incoming beam and in the limit of large magnetic
field (H 1), the linear polarization of the scattered beam is either parallel
or perpendicular to the magnetic field according to the direction of the incident
radiation.
Another consequence that can be easily deduced from Eqs. (5.99) concerns the
particular case of an incident beam parallel (or antiparallel) to the magnetic field
vector. Setting µ = ± 1 in Eqs. (5.99) we see that the matrix elements of the
form Ri0 (i = 0, . . . , 3) are independent of the magnetic field. Therefore, if the
incident beam is unpolarized the scattered radiation is completely insensitive to
the magnetic field.
5.9. Some Illustrations of the Hanle Effect
To understand in further detail the role of the magnetic field in resonance scattering,
it is convenient to apply the results obtained in the previous section to some specific
geometrical configurations of the scattering process.
Consider first the configuration of Fig. 5.11, where we want to investigate the
polarization of the radiation scattered at 90◦ from an unpolarized incident beam.
The scattering event takes place in the presence of a magnetic field which lies in
the plane perpendicular to the incident beam.
Since the incident beam is unpolarized, the only elements of the scattering phase
matrix that must be computed are those of the form Ri0 (i = 0, . . . , 3). Taking as
reference direction the perpendicular to the scattering plane (see Fig. 5.11), these
matrix elements can be deduced from Eqs. (5.99) by the substitutions
π
µ = cos β , µ = 0 , χ − χ = .
2
This leads to the expressions
3
R00 = 3 − cos2 β − sin2 β cos2 α2
8
MATTER-RADIATION INTERACTION (CLASSICAL) 191
Fig.5.12. Polarization diagram (or Hanle diagram) relative to the scattering process illustrated in
Fig.5.11. Full lines correspond to β = const., while broken lines correspond to constant magnetic
field strength, parameterized through the quantity α2 (see Eqs.(5.84) and (5.98)).
3 2
R10 = sin β + (1 + cos2 β) cos2 α2
8
3
R20 = cos β sin α2 cos α2
4
R30 =0,
so that the fractional linear polarization of the scattered beam is given by
Q sin2 β + (1 + cos2 β) cos2 α2
=
I 3 − cos2 β − sin2 β cos2 α2
U 2 cos β sin α2 cos α2
= . (5.102)
I 3 − cos2 β − sin2 β cos2 α2
The results now deduced can be conveniently plotted in the polarization diagram
(or Hanle diagram) shown in Fig. 5.12. The solid lines are isoazimuth curves
(β = const.), while the broken lines are isostrength curves (α2 = const.). For
any assigned couple (β, α2 ) the diagram gives the polarization Q/I, U/I of the
scattered beam. Note that the values of β can be confined in the range [0, π] since
Eqs. (5.102) are invariant under the substitution β → −β.
There are several characteristics of Eqs. (5.102) that it is worthwhile to discuss
in some detail. First of all, in the case of zero magnetic field (α2 = 0) one gets
Q U
=1, =0, (5.103)
I I
192 CHAPTER 5
thus the scattered beam is totally polarized perpendicularly to the scattering plane.
On the other hand, for very strong magnetic field (α2 = π/2) one obtains
Q sin2 β U
= , =0,
I 3 − cos2 β I
so that the scattered beam is again polarized perpendicularly to the scattering
plane (consistently with our former discussion on the Van Vleck angle); however,
the fractional linear polarization varies now between 0 and 1/3 according to the
value of β.
For intermediate values of the magnetic field, the U Stokes parameter is in general
non-zero, which means that the direction of linear polarization is rotated from the
zero-field direction. The maximum rotation for a given magnetic field strength
occurs for β = 0 or π. In these cases we have
Q U
= cos2 α2 , = ± sin α2 cos α2 ,
I I
where the plus sign for U/I corresponds to β = 0 and the minus sign to β = π. In
both cases the total linear polarization is given by
2 2
Q U 1
pL = + = cos α2 = √ , (5.104)
I I 1 + 4H 2
while the position angle α0 (see Eqs. (1.46)) is
1 1
α0 = ± α2 = ± arctan 2H (5.105)
2 2
with the plus sign for β = 0 and the minus sign for β = π. This formula shows
that a magnetic field with the same direction as the scattered radiation produces
a counterclockwise rotation (for an observer looking at the scattering point) of the
plane of linear polarization; conversely, the rotation is clockwise if the magnetic
field is in the opposite direction.
Equations (5.104) and (5.105) are often used to discuss the role of the Hanle
effect in scattering polarization (see e.g. Mitchell and Zemansky, 1934). They are
however of limited use because they refer to an extremely particular case.
In the scattering configuration just considered the magnetic field produces, be-
sides a rotation of the polarization direction, a decrease of the polarization degree
(depolarization); but this is not always the case. A counter example is provided,
for instance, by forward scattering.
The relevant geometry is now illustrated in Fig. 5.13, where we consider the effect
of a magnetic field on the forward scattering of an unpolarized radiation beam. The
polarization of the scattered beam can again be deduced from Eqs. (5.99) by the
substitutions
µ = µ = cos β , χ = χ .
MATTER-RADIATION INTERACTION (CLASSICAL) 193
reference
direction
unpolarized
incident beam β scattered beam
Ω Ω
Fig.5.13. Forward scattering. The reference direction for the Stokes parameters lies in the plane
containing the magnetic field and the direction Ω.
We obtain
1 2 3 3
R00 = 1 + 3 cos2 β − 1 + sin2 β cos2 β cos2 α1 + sin4 β cos2 α2
8 2 8
3
R10 = sin2 β 1 − 3 cos2 β + 4 cos2 β cos2 α1 − (1 + cos2 β) cos2 α2
8
3
R20 = sin2 β cos β 2 sin α1 cos α1 − sin α2 cos α2
4
R30 = 0 . (5.106)
For zero magnetic field (α1 = α2 = 0) one gets
Q U
= =0,
I I
so that the scattered beam is unpolarized as the incident beam. For strong magnetic
field (α1 = α2 = π/2) one obtains
Q (1 − cos2 β)(1 − 3 cos2 β) U
= , =0
I 3 − 2 cos2 β + 3 cos4 β I
or, in other words, a linear polarization which is parallel or perpendicular – accord-
ing to the sign of the quantity (1 − 3 cos2 β) – to the plane containing the magnetic
field and the propagation direction. For intermediate values of the magnetic field
the U Stokes parameter is non-zero, which means a rotation of the plane of linear
polarization. The rotation angle is a complicated function of β and H, as apparent
from Eqs. (5.106).
This example shows that in some scattering configurations the magnetic field is
able to produce a definite amount of linear polarization in a radiation beam which
would be unpolarized if the magnetic field were not present.
194 CHAPTER 5
5.10. The Scattering Phase Matrix Expressed in Terms
of Rotation Matrices
The equations that we have derived in the preceding sections involve the evaluation
of several direction cosines relating the unit vectors uα , relative to the magnetic
needed for the definition of the Stokes parameters.
field, to the unit vectors ei (Ω)
These direction cosines and, in particular, their bilinear combinations can be conve-
niently expressed in terms of ordinary rotation matrices. This leads, in most cases,
to much more compact and handy expressions for the relevant physical quantities,
like the scattering phase matrix derived in Sect. 5.7.
The introduction of rotation matrices requires, however, some remarks about
the spherical components of ordinary vectors, an argument that we have already
treated in Sect. 2.7 and that we are going to develop in further detail here.
Given an arbitrary vector v , its spherical components vq are defined by (see
Eqs. (2.82))
1
v−1 = √ (vx − i vy )
2
v0 = vz
1
v1 = − √ (vx + i vy ) , (5.107)
2
where vx , vy , vz are the Cartesian components in a right-handed coordinate system.
Since in the following we will often deal with vectors of the form
v = a + i b (5.108)
with a and b real, we want to establish some properties of the spherical components
of such vectors.
Taking the complex conjugate of Eqs. (5.107), with v given by Eq. (5.108), we
have
1
(v−1 )∗ = √ (ax + by ) − i (bx − ay )
2
(v0 )∗ = az − i bz
1
(v1 )∗ = − √ (ax − by ) − i (bx + ay ) . (5.109)
2
On the other hand, taking the complex conjugate of Eq. (5.108),
v ∗ = a − i b ,
we have for the spherical components of the vector v ∗
1
(v ∗ )−1 = √ (ax − by ) − i (bx + ay )
2
(v ∗ )0 = az − i bz
1
(v ∗ )1 = − √ (ax + by ) − i (bx − ay ) . (5.110)
2
MATTER-RADIATION INTERACTION (CLASSICAL) 195
Comparison of Eqs. (5.110) and (5.109) shows that
(v ∗ )q = (−1)q (v−q )∗ , (5.111)
a formula that could also be deduced by application of Eq. (2.85) to ordinary vectors
(tensors of rank 1). Thus the complex conjugates of the spherical components of
the vector v do not form an irreducible spherical tensor; nevertheless they are
connected with the spherical components of the vector v ∗ by the simple relations
(5.111).
From now on we will use for simplicity the symbol vq∗ (without parentheses) to
denote the complex conjugate of the spherical component of the vector v ,
vq∗ ≡ (vq )∗ .
Consider now the direction cosines Cαi defined in the former sections. Since the
scalar product of two vectors v and w
can be written in the form (see Eqs. (2.83))
v · w
= (−1)q vq w−q , (5.112)
q
we have
Cαi = uα · ei∗ = (−1)q (uα )q (e∗i )−q = (uα )q (ei )∗q , (5.113)
q q
and for a bilinear product of two direction cosines
∗
Cαi Cβj = (uα )q (uβ )∗q (ei )∗q (ej )q .
This expression suggests the introduction of a tensor – the dyadic product of two
irreducible spherical tensors of rank 1 – which plays a very important role in the
description of polarization phenomena. Indeed, as we will see in the following, this
tensor – and the analogous ones that will be derived from it – appears as a natural
ingredient of almost any mathematical expression relevant to polarized radiation.
For this reason it deserves a careful definition.
In the right-handed reference system (xyz) of Fig. 5.14 we consider a particular
direction characterized by the real unit vector c. In the plane perpendicular to c
we introduce two real unit vectors, a and b, such that a, b, and c form, in this
order, a right-handed coordinate system. The orientation of (a, b, c ) relative to
(xyz) is specified by the angles θ, χ, and γ defined in Fig. 5.14, with
0≤θ≤π, 0 ≤ χ < 2π , 0 ≤ γ < 2π .
From the unit vectors a, b, c we then define the unit vectors c−1 , c0 , c+1 via the
equations
1
c−1 = √ (a + i b )
2
c0 = c
1
c+1 = √ (−a + i b ) , (5.114)
2
196 CHAPTER 5
plane
perpendicular
to
θ
γ
Fig.5.14. Geometry for the definition of the tensor Eqq (α, β,c ).
and, finally, we define the reducible spherical tensor Eqq relative to the triplet
(a, b, c ) as1
Eqq (α, β, c ) = (cα )q (cβ )∗q (α, β = ±1; q, q = 0, ±1) . (5.115)
The explicit expression for Eqq (α, β, c ) can be found by the following argument.
In the reference system (a, b, c ) we have, from Eqs. (5.107) applied to the unit
vectors defined in Eqs. (5.114)
(cα )q = δαq , (cβ )q = δβq .
Since (cα )q is an irreducible spherical tensor of rank 1, its expression in the reference
system (xyz) of Fig. 5.14 can be obtained by application of the transformation law
of irreducible spherical tensors under rotations. From Eq. (2.78) we get
(cα )q = δαp Dpq
1
(R) = Dαq1
(R) (5.116)
p
and, similarly
(cβ )∗q = Dβq
1 ∗
(R) ,
1
The quantity Eqq is neither an irreducible spherical tensor nor the dyadic product of
two irreducible spherical tensors. The denomination ‘reducible spherical tensor’ is due to the fact
that it can be reduced to the linear combination of irreducible spherical tensors (see Eq. (5.125)).
Note also that Eqq depends not only on the direction of the unit vector c (specified by the angles
θ and χ), but also on the direction (specified by the angle γ) of the unit vector a in the plane
perpendicular to c. To shorten notations, we keep only c as an explicit argument of the tensor.
MATTER-RADIATION INTERACTION (CLASSICAL) 197
where R ≡ (−γ, −θ, −χ) is the rotation bringing the system (a, b, c ) into the system
(xyz). Thus the expression for Eqq (α, β, c ) in the reference system (xyz) is
∗
Eqq (α, β, c ) = Dαq
1
(R) Dβq
1
(R) . (5.117)
Let’s now go back to the scattering phase matrix T appearing in Eqs. (5.86) and
(5.87), and consider the case where the polarization unit vectors (e1 , e2 ) relative to
the scattered radiation and the analogous vectors (e1 , e2 ) relative to the incident
radiation are defined by the expressions (note the similarity with Eqs. (5.114))
1
e1 = √ + i e (Ω)
ea (Ω) b
≡ e (Ω)
−1
2
1
e2 = √ + i e (Ω)
−ea (Ω) b
≡ e (Ω)
+1
2
1
e1 = √ ea (Ω
) + i e (Ω
b
) ≡ e (Ω
−1
)
2
1
e2 = √ −ea (Ω
) + i e (Ω
b
) ≡ e (Ω
+1
) , (5.118)
2
e (Ω
e (Ω),
where ea (Ω), ), e (Ω
) are the reference direction unit vectors and the
b a b
associated unit vectors defined as in Fig. 5.9.
With this choice Eq. (5.86) can be rewritten as
= σ
dI˜µν (Ω) Ω
Tµν,ρσ (Ω, ; B ) dΩ dΩ
) I (Ω
cl ρσ
ρσ
4π
where µ, ν, ρ, σ = ±1, with the elements of the matrix T given by (see Eq. (5.113))
Ω
Tµν,ρσ (Ω, ; B
)=
3
= (uα )∗q (eµ (Ω))
∗ ∗ ∗
q (uβ )q (eν (Ω))q (uα )q (eρ (Ω ))q (uβ )q (eσ (Ω ))q
2
αβ qq q q
1
× (α, β = 0, ±1) . (5.119)
1 + i (α − β)H
This formula can be evaluated in any reference system. If we choose the reference
system (ur , us , u0 ) used in Sects. 3.2 and 5.3 we simply have
(uα )q = δαq , (uβ )q = δβq . (5.120)
Thus Eq. (5.119) can be written, with the help of Eqs. (5.115) and (5.117), in the
form
Eqq (µ, ν, Ω) )
E (σ, ρ, Ω
Ω
Tµν,ρσ (Ω, )= 3
; B q q
2 1 + i (q − q )H
qq
3 Dµq
1
(R) Dνq
1 ∗ ∗
(R) Dρq (R ) Dσq (R )
1 1
= , (5.121)
2 1 + i (q − q )H
qq
198 CHAPTER 5
where R is the rotation bringing the system (ea (Ω), e (Ω),
Ω)
into the system
b
(ur , us , u0 ) and, similarly, R is the rotation bringing the system (ea (Ω
), e (Ω
b
), Ω
)
into the system (ur , us , u0 ).
In the general geometrical configuration of Fig. 5.9, R and R can be expressed
in the most convenient way as the result of two consecutive rotations,
R ≡ −γ, −θ, −χ × χB , θB , α
R ≡ −γ , −θ , −χ × χB , θB , α , (5.122)
where α is an arbitrary angle whose actual value is unimportant for the evalua-
tion of the quantity Tµν,ρσ and that can be set equal to zero. In the geometrical
configuration of Fig. 5.10, R and R are simply given by
R ≡ (0, −θ, −χ) , R ≡ (0, −θ , −χ ) . (5.123)
As shown in Sect. 2.7, the dyadic product of two irreducible spherical tensors of
rank k and k can be used to construct an irreducible spherical tensor of rank K.
Therefore, recalling Eq. (5.111), one can obtain an irreducible spherical tensor by
taking appropriate linear combinations of the quantities Eqq (α, β, c ) defined in
Eq. (5.115). According to Eq. (2.79) such combinations are given by1
1 1 K
EQ
K
(α, β, c ) = (−1)1+q 3(2K + 1) Eqq (α, β, c ) , (5.124)
q −q −Q
qq
with the inverse transformation
.
2K + 1 1 1 K
Eqq (α, β, c ) = (−1) 1+q
EQ
K
(α, β, c ) . (5.125)
3 q −q −Q
KQ
In terms of rotation matrices, one gets from Eq. (5.117)
1 1 K ∗
EQ
K
(α, β, c ) = (−1)1+q
3(2K + 1) Dαq
1
(R) Dβq
1
(R) ,
q −q −Q
qq
which, using Eqs. (2.77), (2.23a), (2.73) and the fact that α and β can only take
the values ±1, can be written in the form
1 1 K
EQ
K
(α, β, c ) = 3(2K + 1) DQ
K
Q (R) (α, β = ±1) . (5.126)
α −β −Q
1
The linear combinations of Eq. (2.79) have been multiplied by the factor f (K) = − 3(2K+1) .
This slightly simplifies some expressions that will be deduced in the following.
MATTER-RADIATION INTERACTION (CLASSICAL) 199
Substitution of Eq. (5.125) into Eq. (5.121) leads to the expression of the scat-
tering phase matrix T in terms of irreducible spherical tensors
Ω
Tµν,ρσ (Ω, ; B
)=
1 )
E K (σ, ρ, Ω 1
= (−1)Q EQ
K
(µ, ν, Ω) −Q
2 1 + i QH
KQ
3 1 1 K 1 1 K
= (−1)Q (2K + 1)
2 µ −ν −Q σ −ρ −Q
KQ
1
× DQ
K
Q (R) DQ −Q (R )
K
, (5.127)
1 + i QH
where R and R are the rotations given in Eqs. (5.122) or (5.123).
To deduce the scattering phase matrix R for the Stokes parameters we need the
relation between the Stokes parameters and the polarization tensor corresponding
to the basis of unit vectors (5.118). This relation can be obtained from Eqs. (1.42)
by the substitution θ = 5π/4, φ = −π/2. We have
I = I++ + I−− Q = − I+− + I−+
V = I++ − I−− U = − i I+− − I−+ ,
where we have shortened the notations writing I++ instead of I+1+1 and so on.
The inverse transformations are
1 1
I++ = I +V I+− = −Q + i U
2 2
1 1
I−− = I −V I−+ = −Q − i U .
2 2
If we take an appropriate representation for the Pauli spin matrices (which differs
from that of Eqs. (1.17)) we can write these transformations in a more compact
form. Defining
1 0 0 −1 0 −i −1 0
σ̂0 = , σ̂1 = , σ̂2 = , σ̂3 = (5.128)
0 1 −1 0 i 0 0 1
with
σ̂i σ̂j = δij σ̂0 + i ijk σ̂k (i, j, k = 1, 2, 3)
k
and
Tr (σ̂k σ̂l ) = 2 δkl (k, l = 0, 1, 2, 3) ,
200 CHAPTER 5
and labelling the rows and columns of the matrices in the order (−, +) (so that,for
instance, (σ̂i )+− = (σ̂i )21 ), we have1
1
3
Iαβ = S (σ̂ ) (α, β = ±1)
2 j=0 j j αβ
Sj = (σ̂j )αβ Iβα (j = 0, 1, 2, 3) . (5.129)
αβ
Since these transformations are identical to (1.40) provided the matrices σi are
exchanged with σ̂i , the scattering phase matrix for the Stokes parameters can be
easily obtained along the same lines of Sect. 5.7. The result is
Ω
Rij (Ω, )= 1
; B (σ̂ ) (σ̂ ) T Ω
(Ω, ; B
)
2 µνρσ=±1 i νµ j ρσ µν,ρσ
or, using Eq. (5.121)
Tqq (i, Ω) )
T (j, Ω
Ω
Rij (Ω, ; B
)=3 q q
, (5.130)
1 + i (q − q )H
qq
as
where we have defined the tensor Tqq (i, Ω)
1
=
Tqq (i, Ω) .
(σ̂ ) E (β, α, Ω) (5.131)
2 i αβ qq
αβ=±1
From the tensor Tqq it is possible to construct – similarly to what done for
Eqq – an irreducible spherical tensor, T Q
K
, that will play an important role in the
continuation of this book
1 1 K
TQK =
(i, Ω) (−1)1+q 3(2K + 1)
Tqq (i, Ω)
q −q −Q
qq
1 1 K
= (−1)1+q 3(2K + 1)
q −q −Q
qq αβ=±1
1
× (σ̂ ) E (β, α, Ω)
2 i αβ qq
1
= .
(σ̂ ) E K (β, α, Ω) (5.132)
2 i αβ Q
αβ=±1
1
Equations (5.129) have the same form as Eqs. (1.40). Note, however, that they involve the
polarization tensor defined on the basis of the unit vectors (5.118), while the polarization tensor
in Eqs. (1.40) is defined on the basis of the reference direction unit vector and the associated unit
vector.
MATTER-RADIATION INTERACTION (CLASSICAL) 201
Inversion of the first line of Eq. (5.132) gives
.
2K + 1 1 1 K
=
Tqq (i, Ω) (−1) 1+q
TQ
K ,
(i, Ω)
3 q −q −Q
KQ
which can be substituted into Eq. (5.130) to obtain the expression of the phase
matrix R in terms of the irreducible spherical tensor T Q
K
1
Ω
Rij (Ω, ; B
)= (−1)Q T Q
K
(i, Ω) )
T K (j, Ω . (5.133)
−Q
1 + i QH
KQ
Using Eq. (5.126) the tensor T Q K
can be explicitly written in terms of rotation
matrices
TQ =
P (i) DP Q (R) ,
K
(i, Ω) tK K
(5.134)
P
where
1
1 1 K
tK
P (i) = (σ̂i )αβ 3(2K + 1) . (5.135)
2 β −α −P
αβ=±1
From this equation, using some properties of the rotation matrices (Eqs. (2.71) and
∗
(2.73)) and noticing that tK K
P (i) = t−P (i), Eq. (5.133) can be rewritten in the form
1
Ω
Rij (Ω, ; B
)= tK K ∗
DPKQ (R) DQP
K −1
P (i) tP (j) (R ) , (5.136)
1 + i QH
KP P Q
which is our final equation expressing the phase matrix R in terms of rotation
matrices. In this equation we can isolate the contributions of the various K-values
(sometimes referred to as the multipole components of the phase matrix)
Ω
Rij (Ω, ; B
)= (K)
Rij (Ω, Ω ; B ) (5.137)
K
with
(K)
1
Rij (Ω, Ω ; B ) = (−1)Q T Q
K
(i, Ω) )
T K (j, Ω
−Q
1 + i QH
Q
1
∗ −1
= tK K
P (i) tP (j) DPKQ (R) DQP
K
(R ) . (5.138)
1 + i QH
PP Q
The tensors introduced in this section satisfy several properties, which are col-
lected in the following section together with a number of tables that are particularly
useful for applications. With the help of such tables we can easily get back expres-
sions (5.99) which give the scattering phase matrix in the reference system with
202 CHAPTER 5
the z-axis parallel to the magnetic field (see Fig. 5.10). In terms of multipole
components we get
⎛ ⎞
1 0 0 0
Ω
R(0) (Ω, )=⎜
; B ⎝
0 0 0 0⎟
⎠ ,
0 0 0 0
0 0 0 0
⎛ ⎞
0 0 0 0
0 0 0 0⎟
Ω
R(1) (Ω, ) = 3 µµ + 1 − µ2 1 − µ2 C ⎜
; B
1 ⎝0 0 0 0⎠ ,
2
0 0 0 1
⎛ ⎞
(1 − 3µ2 )(1 − 3µ2 ) (1 − 3µ2 )(3 − 3µ2 ) 0 0
⎜ (3 − 3µ2 )(1 − 3µ2 ) (3 − 3µ2 )(3 − 3µ2 ) 0 0⎟
Ω
R(2) (Ω, )= 1⎜
; B ⎜
⎟
⎟
8⎝ 0 0 0 0⎠
0 0 0 0
⎛ ⎞
µµ C1 µµ C1 −µ S1 0
⎜
0⎟
3 ⎜ µµ C µµ C1 −µ S1 ⎟
+ 1 − µ2 1 − µ2 ⎜ 1 ⎟
2 ⎝ µ S1 µ S1 C1 0⎠
0 0 0 0
⎛ 2 2 2 ⎞
(1 − µ )(1 − µ )C2 −(1 − µ )(1 + µ )C2 2(1 − µ )µ S2
2 2
0
⎜
3 ⎜−(1 + µ2 )(1 − µ2 )C2 (1 + µ2 )(1 + µ2 )C2 −2(1 + µ2 )µ S2 0⎟⎟
+ ⎜ ⎟ , (5.139)
8 ⎝ −2µ(1 − µ2 )S2 2µ(1 + µ2 )S2 4µµ C2 0⎠
0 0 0 0
where the notations are the same as in Eqs. (5.97). Adding the contributions of
the three matrices R(0) , R(1) , and R(2) , Eqs. (5.99) are easily obtained.
5.11. Spherical Tensors for Polarimetry
The various tensors defined in the former section are particularly suitable for the
description of polarization phenomena in spectral lines. As they will be heavily
used in the following of this book, their properties deserve a thorough discussion.
At the same time, we will take the opportunity of introducing some new tensors
that will also be useful later.
Starting with the reducible tensors, we have defined
= e (Ω)
Eqq (α, β, Ω) ∗
eβ (Ω) (q, q = 0, ±1; α, β = ±1) (5.140)
α q q
with
= √1 ∓ e (Ω)
e±1 (Ω) + i e (Ω)
, (5.141)
a b
2
MATTER-RADIATION INTERACTION (CLASSICAL) 203
and we have obtained, in terms of rotation matrices (see Eq. (5.117))
= D1 (R) D1 (R)∗ ,
Eqq (α, β, Ω) (5.142)
αq βq
where
R ≡ (−γ, −θ, −χ)
is the rotation bringing the coordinate system (ea (Ω), e (Ω),
Ω)
into the system
b
e (Ω),
(xyz) (refer to Fig. 5.14 with (a, b, c ) replaced by (ea (Ω), Ω)).
b
In terms of reduced rotation matrices (see Eq. (2.68)) one gets, with the help of
Eqs. (2.70)
= e i [(q−q )χ + (α−β)γ] d1 (θ) d1 (θ) .
E (α, β, Ω)
qq qα q β
From Eq. (5.142), using Eqs. (2.72), one obtains
= δ − D1 (R) D1 (R)∗ ,
Eqq (α, α, Ω) qq 0q 0q
α=±1
and by application of the Weyl theorem (Eq. (2.76))
5 dΩ
=2 δ .
Eqq (α, α, Ω) (5.143)
α=±1
4π 3 qq
Again from Eqs. (2.72) one gets the relation
=δ ,
Eqq (α, β, Ω) (5.144)
αβ
q
and directly from the definition (5.140)
∗ = E (β, α, Ω)
Eqq (α, β, Ω) . (5.145)
q q
as a function of χ, θ, and γ is
The explicit expression for the tensor Eqq (α, β, Ω)
given in Table 5.2.
defined by (see
The other reducible tensor that we have introduced is Tqq (i, Ω),
Eq. (5.131))
1 (q, q = 0, ±1)
=
Tqq (i, Ω)
(σ̂ ) E (β, α, Ω) (5.146)
2 i αβ qq (i = 0, . . . , 3) .
αβ=±1
Taking the complex conjugate, we have from Eq. (5.145)
1
∗=
Tqq (i, Ω) (σ̂ )∗ E (α, β, Ω)
,
2 i αβ q q
αβ=±1
204 CHAPTER 5
TABLE 5.2
Explicit expression for the tensor Eqq (α, β, Ω).
The remaining components can be obtained from
∗.
= E (β, α, Ω)
the relation Eq q (α, β, Ω) qq
E−1 −1 (−, −, Ω)
= 1
4
(1 + cos θ)2
E0 0 (−, −, Ω)
= 1
2
sin2 θ
E1 1 (−, −, Ω)
= 1
4
(1 − cos θ)2
E−1 0 (−, −, Ω)
=− 1
√ sin θ (1 + cos θ) e−iχ
2 2
E−1 1 (−, −, Ω)
= 1
4
sin2 θ e−2iχ
E0 1 (−, −, Ω)
=− 1
√ sin θ (1 − cos θ) e−iχ
2 2
E−1 −1 (−, +, Ω)
= 1
4
e−2iγ sin2 θ
E0 0 (−, +, Ω)
= − 12 e−2iγ sin2 θ
E1 1 (−, +, Ω)
= 1
4
e−2iγ sin2 θ
E−1 0 (−, +, Ω)
= 1
√ e−2iγ sin θ (1 + cos θ) e−iχ
2 2
E−1 1 (−, +, Ω)
= 1
4
e−2iγ (1 + cos θ)2 e−2iχ
E0 1 (−, +, Ω)
=− 1
√ e−2iγ sin θ (1 + cos θ) e−iχ
2 2
E−1 −1 (+, −, Ω)
= 1
4
e2iγ sin2 θ
E0 0 (+, −, Ω)
= − 12 e2iγ sin2 θ
E1 1 (+, −, Ω)
= 1
4
e2iγ sin2 θ
E−1 0 (+, −, Ω)
=− 1
√ e2iγ sin θ (1 − cos θ) e−iχ
2 2
E−1 1 (+, −, Ω)
= 1
4
e2iγ (1 − cos θ)2 e−2iχ
E0 1 (+, −, Ω)
= 1
√ e2iγ sin θ (1 − cos θ) e−iχ
2 2
E−1 −1 (+, +, Ω)
= 1
4
(1 − cos θ)2
E0 0 (+, +, Ω)
= 1
2
sin2 θ
E1 1 (+, +, Ω)
= 1
4
(1 + cos θ)2
E−1 0 (+, +, Ω)
= 1
√ sin θ (1 − cos θ) e−iχ
2 2
E−1 1 (+, +, Ω)
= 1
4
sin2 θ e−2iχ
E0 1 (+, +, Ω)
= 1
√ sin θ (1 + cos θ) e−iχ
2 2
MATTER-RADIATION INTERACTION (CLASSICAL) 205
and since the matrices σ̂i obey the conjugation property
(σ̂i )∗αβ = (σ̂i )βα , (5.147)
we get
∗ = T (i, Ω)
Tqq (i, Ω) . (5.148)
q q
Another property can be obtained from Eq. (5.144) using the trace properties of
the matrices σ̂i
1 1
=
Tqq (i, Ω) (σ̂i )αα = Tr (σ̂i ) = δi0 . (5.149)
q α=±1
2 2
The explicit expression for the tensor Tqq (i, Ω) is given in Table 5.3.
Using the tensors Eqq (α, β, Ω) and Tqq (i, Ω) we can construct a further reducible
which describes the polarization properties of the radiation field
tensor, Iqq (ν, Ω),
propagating along the direction Ω. If I (ν, Ω) is the polarization tensor (defined on
αβ
is the Stokes vector,
the basis of the unit vectors e±1 (Ω) of Eq. (5.141)) and Si (ν, Ω)
both relative to the frequency ν and the direction Ω, we define the tensor I (ν, Ω)
qq
according to either of the equivalent relations (see Eqs. (5.129) and (5.146))
=
Iqq (ν, Ω) I (ν, Ω)
Eqq (α, β, Ω)
βα
αβ=±1
3
= S (ν, Ω)
Tqq (i, Ω) . (5.150)
i
i=0
This tensor satisfies the conjugation property (easily deduced from Eq. (5.148))
∗ = I (ν, Ω)
Iqq (ν, Ω) , (5.151)
q q
and the trace property (consequence of Eq. (5.149))
= S (ν, Ω)
Iqq (ν, Ω) = I(ν, Ω)
. (5.152)
0
q
is given in Table 5.4. It is important
The explicit expression for the tensor Iqq (ν, Ω)
to remark that the linear polarization Stokes parameters, Q and U , enter the
expressions of Table 5.4 only via the linear combinations
Q̃ = cos 2γ Q − sin 2γ U , Ũ = cos 2γ U + sin 2γ Q .
These combinations are independent of the angle γ which defines the reference
direction for the Stokes parameters Q and U .
206 CHAPTER 5
TABLE 5.3
Explicit expression for the tensor Tqq (i, Ω).
The remaining components can be obtained from
∗.
= T (i, Ω)
the relation Tq q (i, Ω) qq
T−1 −1 (0, Ω)
= 1
4
(1 + cos2 θ)
T0 0 (0, Ω)
= 1
2
sin2 θ
T1 1 (0, Ω)
= 1
4
(1 + cos2 θ)
T−1 0 (0, Ω)
=− 1
√ sin θ cos θ e−iχ
2 2
T−1 1 (0, Ω)
= 1
4
sin2 θ e−2iχ
T0 1 (0, Ω)
= 1
√ sin θ cos θ e−iχ
2 2
T−1 −1 (1, Ω)
= − 1 cos 2γ sin2 θ
4
T0 0 (1, Ω)
= 1
2
cos 2γ sin2 θ
T1 1 (1, Ω)
= − 14 cos 2γ sin2 θ
T−1 0 (1, Ω)
=− 1
√ (cos 2γ cos θ − i sin 2γ) sin θ e−iχ
2 2
T−1 1 (1, Ω)
= − 4 cos 2γ (1 + cos2 θ) − 2i sin 2γ cos θ e−2iχ
1
T0 1 (1, Ω)
= 1
√ (cos 2γ cos θ − i sin 2γ) sin θ e−iχ
2 2
T−1 −1 (2, Ω)
= 1
4
sin 2γ sin2 θ
T0 0 (2, Ω)
= − 12 sin 2γ sin2 θ
T1 1 (2, Ω)
= 1
4
sin 2γ sin2 θ
T−1 0 (2, Ω)
= 1
√ (sin 2γ cos θ + i cos 2γ) sin θ e−iχ
2 2
T−1 1 (2, Ω)
= 1
4
sin 2γ (1 + cos2 θ) + 2i cos 2γ cos θ e−2iχ
T0 1 (2, Ω)
=− 1
√ (sin 2γ cos θ + i cos 2γ) sin θ e−iχ
2 2
T−1 −1 (3, Ω)
= − 1 cos θ
2
T0 0 (3, Ω)
=0
T1 1 (3, Ω)
= 1
2
cos θ
T−1 0 (3, Ω)
= 1
√ sin θ e−iχ
2 2
T−1 1 (3, Ω)
=0
T0 1 (3, Ω)
= 1
√ sin θ e−iχ
2 2
MATTER-RADIATION INTERACTION (CLASSICAL) 207
TABLE 5.4
Explicit expression for the tensor Iqq (ν, Ω).
The remaining components can be obtained from
∗.
= I (ν, Ω)
the relation Iq q (ν, Ω) qq
I−1 −1 (ν, Ω)
= 1
4
− sin2 θ Q̃(ν, Ω)
(1 + cos2 θ) I(ν, Ω) − 2 cos θ V (ν, Ω)
I0 0 (ν, Ω)
= 1
2
+ Q̃(ν, Ω)
sin2 θ I(ν, Ω)
I1 1 (ν, Ω)
= 1
4
− sin2 θ Q̃(ν, Ω)
(1 + cos2 θ) I(ν, Ω) + 2 cos θ V (ν, Ω)
I−1 0 (ν, Ω)
= 1
√ sin θ − cos θ I(ν, Ω)
− cos θ Q̃(ν, Ω)
+ i Ũ (ν, Ω) e−iχ
+ V (ν, Ω)
2 2
I−1 1 (ν, Ω)
= 1 − (1 + cos2 θ) Q̃(ν, Ω)
sin2 θ I(ν, Ω) e−2iχ
+ 2i cos θ Ũ(ν, Ω)
4
I0 1 (ν, Ω)
= 1
√ sin θ cos θ I(ν, Ω) − i Ũ (ν, Ω)
+ cos θ Q̃(ν, Ω) e−iχ
+ V (ν, Ω)
2 2
where Q̃(ν, Ω) − sin 2γ U (ν, Ω)
= cos 2γ Q(ν, Ω)
= cos 2γ U (ν, Ω)
Ũ(ν, Ω) + sin 2γ Q(ν, Ω)
one can define the tensor J (ν) by averaging
Finally, from the tensor Iqq (ν, Ω) qq
over the whole solid angle
5 5
dΩ
3
dΩ = S (ν, Ω)
.
Jqq (ν) = I (ν, Ω) T (i, Ω) (5.153)
4π qq 4π i=0 qq i
This tensor depends on the polarized radiation field propagating in all directions,
and will be called the (reducible) radiation field tensor . Its trace is given by (see
Eq. (5.152)) 5
dΩ
Jqq (ν) = = J(ν) ,
I(ν, Ω)
q
4π
which is the usual definition of the mean intensity of the radiation field over the
solid angle. From Eq. (5.151) it follows the conjugation property
Jqq (ν)∗ = Jq q (ν) . (5.154)
All the tensors defined so far are reducible tensors. Under rotation of the coor-
dinate system the new components, Tqq , are obtained from the old ones, Tpp , by
the transformation (see the derivation of Eq. (5.117))
Tqq = Dpq
1
(αβγ) Dp1 q (αβγ)∗ Tpp ,
pp
where α, β, γ are the Euler angles of the rotation which brings the old reference
system into the new one.
208 CHAPTER 5
We can easily construct the associated irreducible tensors through the standard
procedure (see Eq. (5.124)). By so doing, the following irreducible tensors are
obtained
EQ
K
(α, β, Ω), TQ
K
(i, Ω), IQ
K
(ν, Ω), JQK
(ν)
corresponding to
Eqq (α, β, Ω),
Tqq (i, Ω),
Iqq (ν, Ω), Jqq (ν) ,
respectively. The relations between the irreducible tensors and the corresponding
reducible ones are repeated here for the sake of completeness
K 1+q
1 1 K
TQ = (−1) 3(2K + 1) Tqq , (5.155)
q −q −Q
qq
with the inverse transformation
.
2K + 1 1 1 K
1+q
Tqq = (−1) TQK . (5.156)
3 q −q −Q
KQ
Obviously, the various irreducible tensors are connected with each other by the
same relations which connect the corresponding reducible tensors, namely (cf.
Eqs. (5.131), (5.150), and (5.153))
1
TQ K =
(i, Ω)
(σ̂ ) E K (β, α, Ω)
2 i αβ Q
αβ=±1
3
IQ
K
(ν, Ω) = EQ
K I (ν, Ω)
(α, β, Ω) = TQ
K S (ν, Ω)
(i, Ω)
βα i
αβ=±1 i=0
5 5
dΩ K
3
dΩ K .
K
JQ (ν) = I (ν, Ω) = T (i, Ω) Si (ν, Ω) (5.157)
4π Q 4π i=0 Q
Their conjugation properties can be deduced from the corresponding properties of
the reducible tensors, and are found to be
∗ = (−1)Q E K (β, α, Ω)
E K (α, β, Ω)
Q −Q
TQ
K ∗ = (−1)Q T K (i, Ω)
(i, Ω)
−Q
IQ
K ∗ = (−1)Q I K (ν, Ω)
(ν, Ω)
−Q
K
JQ (ν)∗ = (−1)Q J−Q
K
(ν) . (5.158)
We rewrite here, for the sake of clarity, the expressions of the tensors EQ
K
(α, β, Ω)
in terms of rotation matrices (see Eqs. (5.126), (5.134), and (5.135))
and T Q (i, Ω)
K
1 1 K
K
EQ (α, β, Ω) = 3(2K + 1) DQ K
Q (R) (α, β = ±1) ;
α −β −Q
TQ =
P (i) DP Q (R)
K
(i, Ω) tK K
(5.159)
P
MATTER-RADIATION INTERACTION (CLASSICAL) 209
Ω Ω
Ω
Ω Ω
Ω
Fig.5.15. Reference direction unit vector (ea ) and associated unit vector (eb ) relative to the
and −Ω
directions Ω is parallel to e (−Ω).
. e (Ω)
a a
with
1
1 1 K
tK
P (i) = (σ̂i )αβ 3(2K + 1) , (5.160)
2 β −α −P
αβ=±1
e (Ω),
where R is the rotation bringing the reference system (ea (Ω), Ω) into the
b
system (xyz).
One can easily deduce some remarkable properties of the coefficients tKP (i):
a) The only non-zero components of tKP (i) are those with P = 0, ±2, as the indices
α and β can only take the values ±1.
b) From Eq. (5.147) and from the properties of the 3-j symbols it follows the
conjugation relation
∗
tK K
P (i) = t−P (i) .
c) Defining the quantities
ξi = (1, 1, 1, −1)
τi = (1, 1, −1, 1)
ζi = (1, 1, −1, −1) = ξi τi (5.161)
with i = 0, . . . , 3, one gets
(σ̂i )−α−β = ξi (σ̂i )βα , (σ̂i )αβ = τi (σ̂i )βα , (σ̂i )−α−β = ζi (σ̂i )αβ ,
and hence
tK K K
P (i) = ξi (−1) tP (i)
tK K
P (i) = τi t−P (i)
tK K K
P (i) = ζi (−1) t−P (i) . (5.162)
These properties allow a simple derivation of the relation between the tensors
TQK and T K (i, −Ω),
(i, Ω) which will be needed in the following.
Q
Let us consider two opposite directions, Ω and −Ω,
and the corresponding ref-
erence direction unit vectors ea (Ω), ea (−Ω). We assume for simplicity that the
and e (−Ω)
directions ea (Ω) coincide (see Fig. 5.15). From the definition (5.159)
a
we have
210 CHAPTER 5
TABLE 5.5
Values of the symbol tK
P (i)
i K=0 K=1 K=2
0 δP,0 0 √1 δP,0
2
√
1 0 0 − 23 δP,−2 + δP,2
√
2 0 0 i 23 δP,−2 − δP,2
3
3 0 2
δP,0 0
=
TQ
K
(i, −Ω) P (i) DP Q (R ) ,
tK K
where R is the rotation bringing the system (ea (−Ω),
e (−Ω),
b
−Ω) into the system
(xyz). The rotation R can be executed in two steps: first a rotation R0 which
e (−Ω),
brings the system (ea (−Ω), −Ω) into the system (e (Ω),
e (Ω),
Ω),
and then
b a b
a rotation R which brings this last system into (xyz). Thus from the closure
property of the rotation matrices (Eq. (2.74)) we have
TQ K =
(i, −Ω) tK DPKQ (R0 ) DQK
P (i) Q (R) .
P Q
On the other hand, the rotation R0 is given by
R0 ≡ (π, π, 0) ,
and since (see Eq. (2.70))
−iπP −iπP
DPKQ (π π 0) = e dK
P Q (π) = e (−1)K+P δP,−Q ,
we obtain
= −iπP
TQ
K
(i, −Ω) tK
P (i) e (−1)K+P D−P
K
Q (R) .
P
Using property a) of the coefficients tK
P (i) (see above) and the last of Eqs. (5.162)
we finally get
TQ
K = ζ T K (i, Ω)
(i, −Ω) . (5.163)
i Q
It should be remarked that this relation is valid only when the directions ea (Ω)
coincide.
and ea (−Ω)
Let us turn to the evaluation of the irreducible tensor T Q K
(i, Ω). From Eqs.
(5.160), (5.128), and (2.26) we can easily obtain the values of the coefficients tKP (i),
which are collected in Table 5.5. The explicit expressions for T Q K can then
(i, Ω)
be derived from the definition (5.159) by direct evaluation of the rotation matrices
(Eq. (2.68) and Table 2.1). The results are contained in Table 5.6.
MATTER-RADIATION INTERACTION (CLASSICAL) 211
TABLE 5.6
Explicit expression for the tensor T Q The components with negative Q-value can be ob-
K (i, Ω).
tained from the relation T−Q ∗ . The angles χ, θ, γ are shown in Fig.5.14.
= (−1)Q T K (i, Ω)
K (i, Ω)
Q
Note that the rotation carrying the reference system (a, b,c ) into (xyz) is R ≡ (−γ, −θ, −χ).
T 00 (0, Ω)
=1
T 01 (0, Ω)
=0
T 11 (0, Ω)
=0
T 02 (0, Ω)
= 1
√ (3 cos2 θ − 1)
2 2
√
T 12 (0, Ω)
= − 23 sin θ cos θ eiχ
√
T 22 (0, Ω)
=
4
3
sin2 θ e2iχ
T 00 (1, Ω)
=0
T 01 (1, Ω)
=0
T 11 (1, Ω)
=0
T 02 (1, Ω)
=− √3
cos 2γ sin2 θ
2 2
√
T 12 (1, Ω)
= − 23 (cos 2γ cos θ + i sin 2γ) sin θ eiχ
√
T 22 (1, Ω)
= − 43 cos 2γ (1 + cos2 θ) + 2i sin 2γ cos θ e2iχ
T 00 (2, Ω)
=0
T 01 (2, Ω)
=0
T 11 (2, Ω)
=0
T 02 (2, Ω)
= 3
√ sin 2γ sin2 θ
2 2
√
T 12 (2, Ω)
=
2
3
(sin 2γ cos θ − i cos 2γ) sin θ eiχ
√
T 22 (2, Ω)
=
4
3
sin 2γ(1 + cos2 θ) − 2i cos 2γ cos θ e2iχ
T 00 (3, Ω)
=0
3
T 01 (3, Ω)
=
2
cos θ
√
T 11 (3, Ω)
= − 23 sin θ eiχ
T 02 (3, Ω)
=0
T 12 (3, Ω)
=0
T 22 (3, Ω)
=0
212 CHAPTER 5
TABLE 5.7
Explicit expression for the tensor IQ
K (ν, Ω). The components with negative Q-value can be
obtained from the relation I−Q ∗.
= (−1)Q I K (ν, Ω)
K (ν, Ω)
Q
I00 (ν, Ω)
= I(ν, Ω)
3
I01 (ν, Ω)
=
2
cos θ V (ν, Ω)
√
I11 (ν, Ω)
= − 23 eiχ
sin θ V (ν, Ω)
I02 (ν, Ω)
= 1
√ (3 cos2 θ − 1) I(ν, Ω)
− 3 sin2 θ Q̃(ν, Ω)
2 2
√
I12 (ν, Ω)
= − 23 sin θ cos θ I(ν, Ω)
+ cos θ Q̃(ν, Ω)
+ i Ũ (ν, Ω)
eiχ
√
I22 (ν, Ω)
=
4
3 − (1 + cos2 θ) Q̃(ν, Ω)
sin2 θ I(ν, Ω) − 2i cos θ Ũ(ν, Ω)
e2iχ
where Q̃(ν, Ω) − sin 2γ U (ν, Ω)
= cos 2γ Q(ν, Ω)
= cos 2γ U (ν, Ω)
Ũ (ν, Ω) + sin 2γ Q(ν, Ω)
The derivation of the explicit expression for the tensor IQ K defined in Eq.
(ν, Ω)
(5.157) is straightforward (see Table 5.7). Note that the expressions in Table 5.7
– like those of Table 5.4 – contain only the linear combinations Q̃ and Ũ , which
are independent of the reference direction chosen to define the Stokes parameters
Q and U .
K
Finally, the irreducible tensor of the radiation field JQ (ν) is obtained by averag-
ing IQ (ν, Ω) over the solid angle. In the particular case of an unpolarized radiation
K
field having cylindrical symmetry about the z-axis only two components are not
zero, namely
5
dΩ
J00 (ν) = I(ν, θ)
4π
5
1 dΩ
J02 (ν) = √ (3 cos2 θ − 1) I(ν, θ) . (5.164)
2 2 4π
Obviously, the component J02 (ν) is also zero if the radiation field is isotropic.
5.12. Further Properties of the Scattering Phase Matrix
The expression given in Sect. 5.10 for the scattering phase matrix and the various
relations proved in Sect. 5.11 allow a further extension of the remarkable properties
of this matrix.
Starting from Eq. (5.133) and using Eq. (5.163) one can directly prove the fol-
lowing relations (which are however valid only when the same reference directions
MATTER-RADIATION INTERACTION (CLASSICAL) 213
and −Ω,
are chosen for Ω and −Ω
and for Ω )
Ω
Rij (−Ω, ; B
) = ζ R (Ω, ; B
Ω )
i ij
Rij (Ω, ; B
−Ω ) = ζ R (Ω, ; B
Ω )
j ij
Rij (−Ω, ; B
−Ω ) = ζ ζ R (Ω, ; B
Ω ).
i j ij
The first relation means that if (I, Q, U, V ) is the Stokes vector scattered along the
then, irrespective of the polarization of the incident beam and of the
direction Ω,
intensity and direction of the magnetic field, the Stokes vector scattered along the
direction −Ω is (I, Q, −U, −V ). The two other relations have analogous meanings.
Considering next Eq. (5.136), we can write the expression of the scattering phase
matrix for the most general geometry of Fig. 5.9. The two relevant rotations are
(see Eq. (5.122))
R ≡ (−γ, −θ, −χ) × (χB , θB , γB )
R ≡ (−γ , −θ , −χ ) × (χB , θB , γB )
where γB is an arbitrary angle, and hence from Eq. (2.74) we obtain
Ω
Rij (Ω, ; B
)=
∗
= tK K
P (i) tP (j) DPKQ (−γ −θ −χ) MK
QQ (B ) DQ P (χ θ γ ) , (5.165)
K
KP P QQ
where we have introduced the magnetic kernel defined by
1
MK DQQ
K
(χB θB γB ) DQ Q (−γB −θB −χB )
K
QQ (B ) = . (5.166)
1 + i Q H
Q
In terms of reduced rotation matrices (Eq. (2.68)) we have
i(Q −Q)χB 1
MK dK K
QQ (B ) = e QQ (θB ) dQ Q (−θB ) , (5.167)
1 + i Q H
Q
which shows explicitly that the magnetic kernel is independent of the angle γB ,
that can be set equal to zero.
The magnetic kernel satisfies several important properties which can be easily
derived from Eqs. (5.166) and (5.167) using some properties of the rotation matri-
ces:
i) Limiting case of zero magnetic field
Recalling Eqs. (5.84) and (2.74) we have
lim MK DQQ
K
(χB θB 0) DQ Q (0 −θB −χB )
K
QQ (B ) =
B→0
Q
= DQQ
K
(0 0 0) = δQQ ,
214 CHAPTER 5
and hence
Ω ; 0) = ∗
P (i) tP (j) DP P (R) ,
tK K K
Rij (Ω, (5.168)
KP P
e (Ω),
where R is the rotation bringing the reference system (ea (Ω), Ω) into the
b
system (ea (Ω ), eb (Ω ), Ω ).
ii) Limiting case of strong field
lim MK
QQ (B ) = DQ0 (χB θB 0) D0Q (0 −θB −χB ) .
K K
B→∞
iii) Conjugation property
∗ Q−Q
MK
QQ (B ) = (−1) MK
−Q−Q (B ) .
iv) Magnetic field symmetries
QQ (B, π − θB , −χB ) = M−Q−Q (B, θB , χB )
MK K
QQ (B, π − θB , π − χB ) = (−1)
MK Q−Q
MK
−Q−Q (B, θB , χB )
MK
QQ (B, θB , π + χB ) = (−1)
Q−Q
MK
QQ (B, θB , χB ) ,
a set of formulae giving the transformation of the magnetic kernel under a 180◦
rotation of the magnetic field vector about the axes x, y, and z of Fig. 5.9, respec-
tively.
v) Magnetic field reversal
MK
QQ (−B ) =
1
= DQQ
K
(π+χB π−θB 0) DQ Q (0 −π+θB −π−χB )
K
1 + i Q H
Q
1
= DQQ
K
(χB θB 0) DQ Q (0 −θB −χB )
K
, (5.169)
1 − i Q H
Q
which shows that the reversal of the magnetic field can also be obtained by the
formal substitution H → −H.
vi) Average over an isotropic distribution of magnetic fields
For an isotropic distribution of magnetic field vectors we can find the average value
of the magnetic kernel. Defining
2π π
K
) = 1
MQQ (B dχB sin θB MK
QQ (B ) dθB (5.170)
4π
0 0
MATTER-RADIATION INTERACTION (CLASSICAL) 215
we easily obtain, with the help of Eqs. (2.71) and (2.76)
MK
QQ (B ) = µK δQQ , (5.171)
where the quantities µK are given by
1 1
µK = (5.172)
2K + 1 1 + i Q H
Q
or, explicitly
µ0 = 1
1 1 2
µ1 = 1 + 2 cos2 α1 = 1+
3 3 1 + H2
1 1 2 2
µ2 = 1 + 2 cos2 α1 + 2 cos2 α2 = 1+ 2
+ , (5.173)
5 5 1+H 1 + 4H 2
the angles α1 and α2 being defined in Eq. (5.98).
This result shows that the average of the scattering phase matrix in an isotropi-
cally distributed magnetic field can be obtained directly from the scattering phase
matrix of the non-magnetic case by multiplication of the various multipole compo-
nents by the quantities µK ,
Ω
Rij (Ω, ; B
) = (K)
µK Rij (Ω, Ω ; 0) . (5.174)
K
In particular, in the limiting case of strong magnetic fields the matrix R(1) is
reduced by a factor 3, while the matrix R(2) is reduced by a factor 5.
5.13. Understanding Scattering Experiments
through Oscillator Models
In the previous sections we have derived the expression for the scattering phase
matrix in the presence of a magnetic field both in terms of direction cosines and
in terms of rotation matrices. However, these derivations are rather involved and
there is a danger that the physical meaning of the various results may be hidden
by the mathematical formalism. Thus we think it is worthwhile to present some
qualitative arguments – based on the atomic oscillator model – aimed at clarifying
the underlying physics. At the same time we will draw some interesting analogies
between the classical and quantum-mechanical descriptions of scattering phenom-
ena.
Let us consider a simple scattering event like that illustrated in Fig. 5.16a, left.
The incident radiation beam is unpolarized (conventionally, this is represented by
216 CHAPTER 5
unpolarized linear
incident beam oscillators
scattered beam
circular
oscillators
weak magnetic field
strong magnetic field
Fig.5.16. Qualitative description of a 90 ◦ scattering event in the absence of magnetic fields and
in the presence of a magnetic field parallel to the line of sight. See text for explanation.
MATTER-RADIATION INTERACTION (CLASSICAL) 217
two double-ended, perpendicular arrows which should be regarded as incoherent
electric field oscillations), and there is no magnetic field. We schematize the scat-
tering atom as a collection of three linear oscillators of frequency ν0 that will be
called the x, y, and z-oscillator, respectively. The scattering experiment can be
described in the following way.
i) The electric field of the incident radiation can be decomposed into its x and
z components; owing to the unpolarized character of the radiation, the two com-
ponents are incoherent, or, in other words, there are no phase relations between
them.
ii) The x component of the electric field excites the x-oscillator, while the z compo-
nent excites the z-oscillator (the latter is represented by a double-ended arrow in
the right part of the figure). The incoherent character of the electric field is simply
transferred to the oscillators.
iii) The oscillators decay with a damped motion and each of them emits, in any
given direction, a radiation beam polarized according to the well-known laws of
classical electrodynamics (see Eq. (3.29)). If we consider the beam scattered in
the x direction, it is easily seen that the x-oscillator is ineffective (being pole-on),
while the z-oscillator produces a radiation beam that is linearly polarized along the
z direction. As a consequence, the radiation scattered in the x direction is 100%
linearly polarized perpendicularly to the scattering plane, which is consistent with
the results derived in Sect. 5.9 (see Eq. (5.103)).
To investigate the effect of the magnetic field it is convenient to describe the
model atom by a different set of oscillators, namely the linear x-oscillator plus two
circular oscillators with opposite directions (called σ+ and σ− ) laying in the y-z
plane as shown in Fig. 5.16b, left. If the magnetic field is zero, the frequency of
the three oscillators is still ν0 , and we must find the same results as above.
Following the same line of reasoning, we see at step ii) that while the x component
of the electric field still excites the x-oscillator, the z component excites the σ+ and
σ− oscillators in such a way as to produce a well-defined phase relation between
them, because the resulting motion of the electric charge takes place along the
z direction (Fig. 5.16b, right). For the scattered radiation we obviously have the
same result as before, but we must bear in mind that this new description of
the atomic system has forced us to introduce the concept of phase relation (or
coherence) between the two circular oscillators.
In the presence of a magnetic field directed along the x-axis (see Fig. 5.16c)
the circular oscillators vibrate at different frequencies ((ν0 + νL ) and (ν0 − νL ),
respectively, with νL the Larmor frequency). Thus the phase relation produced by
the exciting electric field is lost little by little during the damped decay process. The
electric charge describes in the y-z plane a complicated pattern (sometimes called
a ‘rosette’ – Fig. 5.16c, right), and thus the polarization of the scattered radiation
– which reflects the weighted average of the pattern – is decreased and rotated
from the direction of the non-magnetic regime: in Fig. 5.16c (left) the scattered
beam is the superposition of a linearly polarized beam plus an unpolarized beam.
The shape of the rosette is controlled by the parameter H = 2πνL /γ defined in
Eq. (5.84). When the value of H is very large (H 1) the rosette degenerates
218 CHAPTER 5
into a different curve (that we will rather call a ‘daisy’, see Fig. 5.16d right) and
the scattered radiation is totally unpolarized.
This simple example shows that the effect of the magnetic field is to cause a
relaxation of the phase relations (or coherences) between the different oscillators:
the coherence between the σ+ and σ− oscillators is maximum for zero magnetic field
(Fig. 5.16b) and gradually decreases with increasing field strength (Fig. 5.16c,d).
The Hanle effect is just the consequence of this relaxation process in resonance
scattering.
Different geometrical configurations can be envisaged to get a deeper insight into
the phenomena of resonance polarization and the Hanle effect. Figure 5.17 presents
an example of forward scattering; here the scattered beam is unpolarized for zero
magnetic field and partially linearly polarized along the field’s direction when the
magnetic field is present. The discussion of this case, that can be carried out along
the same lines as before, is left to the reader as an exercise (cf. Eqs. (5.106)).
The classical model that represents the atom as a collection of one linear os-
cillator and two circular oscillators of opposite directions has a simple quantum
analogue. This is the two-level atom having a lower level of angular momentum
J = 0 and an upper level of angular momentum J = 1. The excitation of one of
the classical oscillators induced by the electric field of the incident beam has its
quantum equivalent in the excitation of one of the Zeeman sublevels of the upper
level. In this analogy, the linear oscillator corresponds to the sublevel M = 0, while
the circular oscillators of frequency (ν0 ± νL ) correspond to the sublevels M = ±1,
respectively.
In scattering experiments, apart from special geometrical configurations, the var-
ious oscillators are excited unevenly by the electric field of the incident beam, and
well-defined phase relations arise between different oscillators. This means – in the
language of Quantum Mechanics – that the upper level of the atom is polarized
(see Sect. 3.6); in other words, the diagonal density-matrix elements ρJ (M , M )
are different from each other and the off-diagonal elements ρJ (M , M ) are non-
zero. In particular, the amount of excitation present in a given oscillator has its
quantum analogue in the corresponding diagonal element of the density matrix,
while the combined excitation of two different oscillators with a well-defined phase
relation has its analogue in the off-diagonal element of the density matrix between
the two corresponding sublevels. For example, in the case of panel (b) of Fig. 5.16,
the non-zero density-matrix elements are ρ (0, 0), ρ (1, 1), ρ (−1, −1), ρ (1, −1), and
ρ (−1, 1), the last two elements being related to the coherence between the circular
oscillators. Passing to panels (c) and (d) of the same Figure, the diagonal ele-
ments remain unchanged, while the off-diagonal ones are reduced by the presence
of the magnetic field. In the following of this book we will see that these qualita-
tive concepts are fully confirmed by a more rigorous treatment based on Quantum
Mechanics.
MATTER-RADIATION INTERACTION (CLASSICAL) 219
unpolarized linear scattered
incident beam oscillators beam
circular
oscillators
weak magnetic field
strong magnetic field
Fig.5.17. Same as Fig.5.16 for a forward scattering event. The trajectories of the oscillating
electron in the y-z plane are the same as in Fig.5.16 right.
220 CHAPTER 5
5.14. The Role of Collisions
We present in this section a simple theory to describe the effect of collisions on
polarization phenomena. Both the atom and the colliding particles will be described
in terms of classical physics; in particular, the atom will be schematized as a three-
dimensional oscillator, like in the previous sections of this chapter.
Our first assumption is that the interactions with the closest perturbing particle
(binary interactions) play the principal role. In other words, we assume that triple
and multi-particle interactions can be neglected, which implies a limitation on the
density n of perturbing particles in the medium,
3
σ2 n 1
where σ is the typical collisional cross-section. This inequality is always well-
satisfied both in laboratory plasmas and in stellar atmospheres.
Next we distinguish between two different kinds of collisions, exciting collisions
and perturbing collisions, defined according to the following simple model.
a) Exciting collisions are inelastic, which implies an energy transfer from the col-
liding particle to the atom. When an exciting collision takes place, the oscillator is
abruptly forced to interrupt its previous motion, described by Eq. (3.23), and starts
oscillating again with a new set of initial conditions. These initial conditions de-
pend on the efficiency of the collision (which sets the amplitude of the oscillation),
and on the geometry of the impact event (which sets the polarization characteris-
tics of the oscillation). The geometry of the event is specified by the direction of
the colliding particle, the direction of the vector joining the center of the oscillator
with the center of the colliding particle at the moment of closest approach, and on
the orientation of the colliding particle if this is not spherically symmetrical.
If the colliding particles have an isotropic distribution of velocities, and – in the
case of oriented particles – if they have random orientations, the initial conditions
set up in the oscillator will be isotropic as well. Employing the same notations as
in Sect. 3.2, the initial values of the amplitudes of the oscillations will be given by
∗
Aα Aβ = δαβ |A|2 , (5.175)
where the symbol . . . means a statistical average over all the possible collisions.
This is just the physical situation described in Sect. 3.2. We remind the reader that
exciting collisions, as defined here, have their quantum counterpart in collisions able
to induce transitions between atomic levels.
In different cases, like for instance when the atom is bombarded by a collimated
and/or oriented beam of particles, Eq. (5.175) does not hold and, as a result, the
atom emits polarized radiation in the decay process. This phenomenon is known
under the name of impact polarization, a subject that will not be deepened here.
b) Perturbing collisions, on the contrary, are elastic collisions which are not able to
interrupt the oscillatory motion of the atom and to make it restart from scratch,
MATTER-RADIATION INTERACTION (CLASSICAL) 221
but are effective in shifting the phase of the oscillation during a time interval tc
that is called the collision time.
Following a model that was proposed by Lorentz (1915) and refined by Weisskopf
(1932) and Lenz (1933), we suppose that the collision time tc is much smaller than
the other typical times involved in the physical process, namely the decay time
of the oscillator 1/γ and the mean time between collisions 1/f , where f is the
frequency of perturbing collisions1
1 1
tc , tc . (5.176)
γ f
This hypothesis allows to treat the collisions as instantaneous (impact approxima-
tion). We further assume that each perturbing collision produces a large phase
shift in the atomic oscillation, so that the oscillations before and after the collision
are totally uncorrelated (strong collision hypothesis). These two assumptions are
at the basis of the usual treatment of collisional broadening in spectral lines (see
e.g. Sobel’man, 1972) and can also be used to derive the effect of collisions on
polarization phenomena.
To this aim, however, the second assumption must be suitably specified to de-
scribe with greater accuracy the geometrical details of the collisional event. We
suppose that each collision, characterized by its own geometry, induces an indepen-
dent phase shift in each of the three components of the oscillatory motion along a
triplet of real, orthogonal unit vectors whose orientation is fixed by the direction of
the velocity of the colliding particle (and by its orientation, in case of an oriented
particle). These unit vectors will be denoted in the following by the symbol ti
(i = 1, 2, 3), and the corresponding phase shifts by φi . Now we are going to derive
the consequences of this simplified model on the results obtained in the previous
sections of this chapter.
i) Radiative Transfer Equations.
In the derivation of the radiative transfer equations for polarized radiation in the
presence of a magnetic field (Sect. 5.3) we found in Eqs. (5.29) and (5.30) the
relation between the amplitude of the oscillation and the corresponding component
of the forcing electric field. This relation is based on a particular solution of the non-
homogeneous equation (5.28) which disregards the transient solution depending
on the initial conditions. In the presence of collisions the transient solution is
important, and we take it into account in the following way.
1
Obviously the frequency f can be written in the form
f = Np v̄r σ ,
where Np is the number density of perturbers, v̄r is the average velocity of the perturbers relative
to atoms, and σ is the cross-section. Expressing Np in cm−3 , v̄r in km s−1 , and σ in units of a20
(a0 being the Bohr radius), we have
f = 2.80 × 10−12 Np v̄r σ s−1 .
222 CHAPTER 5
Let t = 0 indicate the time at which the first collision takes place and suppose
the previous collision was sufficiently far in time for its transient to be already
damped (statistically, there will always be such a collision and nothing prevents us
from calling it the ‘first’ collision). Employing the same notations as in Sect. 5.3,
and making the same approximation (ν02 − ν 2 ) ≈ 2ν (ν0 − ν), we can write
e0 1
x (t = 0− ) = − E u . (5.177)
8π 2 mν α (ν0 − ανL − ν) − iΓ α α
Expansion on the unit vectors ti gives
e0 1
x (t = 0− ) = − E D t ,
8π mν αi (ν0 − ανL − ν) − iΓ α αi i
2
where the direction cosines Dαi are defined by1
Dαi = uα · ti∗ . (5.178)
According to the above assumptions, the effect of the collision is such that
e0 1 iφ
x (t = 0+ ) = − E D e i ti ,
8π mν αi (ν0 − ανL − ν) − iΓ α αi
2
and returning to the basis uα
e0 1 iφ ∗
x (t = 0+ ) = − Eα Dαi e i Dβi uβ . (5.179)
2
8π mν (ν0 − ανL − ν) − iΓ i
αβ
To obtain the motion of the electron after the collision we must solve Eq. (5.28).
Its general solution is the following
e0 1 −2πiνt
x (t) = − Eβ e uβ
2
8π mν (ν0 − βνL − ν) − iΓ
β
γ
−2πi (ν0 −βνL ) t − 2 t
+ Cβ e e uβ ,
β
where the constants Cβ must be determined in such a way as to satisfy the boundary
condition (5.179). Hence we obtain
e Eα iφ ∗
Cβ = − 20 Dαi e i Dβi
8π mν α
(ν0 − ανL − ν) − iΓ i
!
Eβ
− .
(ν0 − βνL − ν) − iΓ
1
uα · ti as we have supposed the unit vectors ti to
The quantity Dαi could also be written as
be real. However, we adopt the definition (5.178) to make easier a possible generalization of our
hypotheses on the effects of collisions.
MATTER-RADIATION INTERACTION (CLASSICAL) 223
Now we perform a statistical average over collisions. Since the phases φi are arbi-
trary large numbers (strong collision hypothesis), we have
iφi
e =0 (i = 1, 2, 3) .
The first term in curly brackets vanishes and we obtain
−2πiνt
e0 γ
−2πi (ν0 −βνL −ν) t − 2 t Eβ e
x (t) = − 1 − e e u ,
8π 2 mν (ν0 − βνL − ν) − iΓ β
β
so that (cf. Eq. (5.30))
γ
−2πi (ν −βν −ν) t −2t
e 1−e 0 L
e
χβ (t) = − 20
8π mν (ν0 − βνL − ν) − iΓ
or, in other words, an electric susceptibility variable with time. The average value of
the electric susceptibility can then be obtained by multiplying its value at time t by
the probability that no collision occurred in the interval (0, t). If f is the frequency
of collisions, we have1
∞
−f t e0 1
χβ (t) = f e χβ (t) dt = − ,
8π mν (ν0 − βνL − ν) − iΓ
2
0
where
γ f
Γ = Γ + 2Γc = + , (5.180)
4π 2π
with2
f
Γc = . (5.181)
4π
Comparison with Eq. (5.30) shows that the effect of perturbing collisions consists
in a broadening of both the absorption profile and the associated dispersion profile.
The same result could be formally obtained by the addition of a supplementary
friction force in the equation of motion of the oscillator.
1
The probability that a collision occurs in the interval (t, t + dt) is given by
dp = f dt .
Dividing the interval (0, t) into a large number N of equal parts, the probability that the first
collision occurs in the interval (t, t + dt) is given by (see Eq. (2.62) for a similar derivation)
t N
dP = lim 1 − f f dt = f e−f t dt .
N→∞ N
2
The definition Γc = f /4π (instead of f /2π) leads to simpler forms for the expressions that
will be encountered in the following.
224 CHAPTER 5
ii) Zeeman Effect, Resonance Polarization and Hanle Effect.
By a similar reasoning we can find the effect of perturbing collisions on the polar-
ization of the radiation emitted by an atom in a decay process.
We suppose that our model oscillator is excited at time t = t0 = 0 either by an
exciting collision or by the radiation field, and we suppose that the first collision
takes place at time t1 , the second at time t2 , and so on. In the interval (t0 , t1 ) the
motion of the oscillator is described by Eq. (3.23), which has the solution
γ
−2πi (ν0 −ανL ) t −2t
xα (t) = Aα e e (0 = t0 < t < t1 ) , (5.182)
where Aα are the initial amplitudes set up by the exciting mechanism. At time t1
the oscillator undergoes the first collision, which is characterized by its particular
geometry (specified by the three unit vectors ti(1) and by the corresponding phase
shifts φ(1)
i ). Along the same lines leading from Eq. (5.177) to Eq. (5.179) one can
write for the time interval (t1 , t2 )
(1) −2πi (ν0 −ανL ) t
γ
−2t
xα (t) = Aδ Kδα e e (t1 < t < t2 ) ,
δ
(1)
where Kδα , the collisional kernel due to the first collision, is given by
(1) −2πi (α−δ) νL t1
3
(1) iφi
(1)
(1) ∗
Kδα =e Dδi e Dαi , (5.183)
i=1
(1)
with the direction cosines Dδi defined as in Eq. (5.178). Similarly, for the time
interval (tn , tn+1 ) one has
−2πi (ν0 −ανL ) t
γ
−2t
xα (t) = Aδ Kδ(1) K (2) (n)
ϕ . . . Kρα e e (tn < t < tn+1 ) .
δ ϕ...ρ
To shorten notations we define the cumulative collisional kernel due to the first
n collisions as (0)
Kδα
(n)
= Kδ K (1) (n)
ϕ . . . Kρα (5.184)
ϕ...ρ
with
Kαβ
(0) (0)
= Kαβ = δαβ , (5.185)
and we can write
−2πi (ν0 −ανL ) t
γ
−2t
xα (t) = Aδ Kδα
(n)
e e (tn < t < tn+1 ) ,
δ
a general formula which includes Eq. (5.182).
MATTER-RADIATION INTERACTION (CLASSICAL) 225
The function xα (t) can be expanded in Fourier series,
∞
−2πiνt
xα (t) = Fα (ν) e dν ,
−∞
where1
∞ ∞
tn+1
γ
2πiνt −[2πi (ν0 −ανL −ν) + 2 ]t
Fα (ν) = xα (t) e dt = Aδ Kδα
(n)
e dt
0 n=0 δ tn
∞
−Nα tn −Nα tn+1
e −e
= Aδ Kδα
(n)
, (5.186)
n=0
Nα
δ
with
γ
Nα = 2πi (ν0 − ανL − ν) + = 2πi (ν0 − ανL − ν) − iΓ . (5.187)
2
To investigate how the Zeeman effect and resonance polarization are affected
by perturbing collisions we must start from Eqs. (3.34) and (5.75), respectively.
Owing to the different definition of the Fourier transforms Fα (ν) and Fα (ν), these
two equations must be rewritten in the form
∞
16π 4 e20 ∗ 1
Jij (r) = 2 4 Cαi Cβj ν 4 Fα (ν)∗ Fβ (ν) dν (5.188)
r c τ
αβ −∞
and
∞
2π 3 e20 ∗ 1
˜
dIij (Ω) = dΩ Cαi Cβj ν 4 Fα (ν)∗ Fβ (ν) dν , (5.189)
c3 τ
αβ −∞
respectively.
From Eq. (5.186) we have
Fα (ν)∗ Fβ (ν) =
∞
∞
= A∗δ A Kδα
(n) ∗
K(nβ )
n=0 n =0 δ
∗ ∗
−Nα tn −Nα tn+1 −Nβ tn −Nβ tn +1
e −e e −e
× . (5.190)
Nα∗ Nβ
1
The Fourier transform Fα (ν) differs from Fα (ν) defined in Eq. (3.32) for the presence of the
amplitudes Aδ . In the special case of no collisions (φ(n)
i = 0, K(n)
αβ
= δαβ ) we have
Fα (ν) = Aα Fα (ν) .
226 CHAPTER 5
This expression refers to a particular sequence of collisional events. Now we must
average over all the possible collision directions and over all the possible collision
times t1 , t2 ,. . . tn ,. . .
First of all, we observe that the phase shifts φ(n)i relative to the n-th collision
(n )
are totally uncorrelated with the phase shifts φj relative to the n -th collision.
This reduces the double sum in Eq. (5.190) to a single sum (n = n). Then we
consider the average of the collisional kernels over an isotropic velocity distribution
of perturbing particles. These statistical averages are performed in some detail in
the following section. Borrowing from there Eq. (5.200), we have
Fα (ν)∗ Fβ (ν) =
∞
1 1 K 1 1 K
= A∗δ A (−1)δ−α n
rK (2K + 1)
−δ Q β −α Q
n=0 δ KQ
∗ ∗
−Nα tn −Nα tn+1 −Nβ tn −Nβ tn+1
e −e e −e
× ,
Nα∗ Nβ
where the symbol rK is defined in Eq. (5.199).
Finally, we average over all the possible collision times. To this aim we must
evaluate the multiple integral (see footnote 1 on p. 223)
∞ ∞ ∞
−f t
f dt1 f dt2 · · · f dtn+1 e n+1 Fα (ν)∗ Fβ (ν) ,
0 t1 tn
where f is the frequency of perturbing collisions. The evaluation of the integral is
elementary and leads to
Fα (ν)∗ Fβ (ν) =
∞
1 1 K 1 1 K
= A∗δ A (−1)δ−α n
rK (2K + 1)
−δ Q β −α Q
n=0 δ KQ
f (Nα∗ + Nβ + 2f )
n
× .
(Nα∗ + Nβ + f )n+1 (Nα∗ + f ) (Nβ + f )
Performing the sum over n, and using the expressions of Nα∗ and Nβ given in
Eq. (5.187), one finally obtains
Fα (ν)∗ Fβ (ν) =
Γ
1 ∗ δ−α Γ + i (α − β)H
= A A (−1)
4π 2
δ
δ
(ν0 − ανL − ν) + iΓ (ν0 − βνL − ν) − iΓ
1 1 K 1 1 K
× (2K + 1)
−δ Q β −α Q
KQ
−1
Γ + Γc (1 − rK )
× + i (α − β)H , (5.191)
Γ
MATTER-RADIATION INTERACTION (CLASSICAL) 227
where the various damping coefficients are given by
γ f 2πνL e B
Γ = , Γc = , Γ = Γ + 2Γc , H= = 0 .
4π 4π γ 2mcγ
As far as the Zeeman effect is concerned, if we restrict attention – as in Sect. 3.2 –
to the case of natural excitation, we must replace A∗δ A by |A|2 δδ in Eq. (5.191).
Using Eq. (2.26a) we can write
1 1 0 1 1 0
δδ (−1)δ−α = 3 .
−δ 0 α −α 0
Application of the orthogonality relation (2.23a) shows that the only non-zero
contribution originates from the K = 0 term. Thus we obtain, with the help of
Eqs. (2.26a) and (5.199)
1 Γ 1
Fα (ν)∗ Fβ (ν) = δαβ |A|2 . (5.192)
4π 2 Γ (ν0 − ανL − ν)2 + Γ 2
Since Eq. (3.35) – which relates the amplitude |A|2 to the average energy Ē con-
tained in each oscillator – is still valid, from Eqs. (3.36), (5.188), and (5.192) we
obtain for the emission coefficient in tensorial form the expression (3.37) with Γ
replaced by Γ . In other words, the effect of perturbing collisions on the Zeeman
effect is nothing but a broadening of the profile. This is consistent with what we
found for the radiative transfer of polarized radiation.
As far as resonance polarization and the Hanle effect are concerned, the situation
is very different. Since the Fourier transform Fα (ν) is substantially non-zero only for
ν ≈ ν0 , we can extract the factor ν 4 from the integral over frequency in Eq. (5.189).
Using the residue theorem, we then obtain from Eq. (5.191)
∞
Fα (ν)∗ Fβ (ν) dν =
−∞
1 ∗ 1 1 K 1 1 K
= Aδ A (−1)δ−α (2K + 1)
4πΓ −δ Q β −α Q
δ KQ
−1
Γ + Γc (1 − rK )
× + i (α − β)H
Γ
(which is consistent, under the limit f → 0, with Eq. (5.83)). Repeating the same
arguments leading to Eq. (5.87), one finds that this equation must be replaced by
the following
Ω
Tij,kl (Ω, f) = 3
; B, ∗
(−1)δ−α Cαi
Cβj Cδk C ∗l
2
αβδ
1 1 K 1 1 K
× (2K + 1)
−δ Q β −α Q
KQ
−1
Γ + Γc (1 − rK )
× + i (α − β)H ,
Γ
228 CHAPTER 5
which is the general expression for the scattering phase matrix of the polarization
tensor in the presence of perturbing collisions. It can be easily seen that properties
(5.88) and (5.89) are still satisfied.
The scattering phase matrix for the Stokes parameters in the presence of collisions
is given by the analogue of Eq. (5.92), namely
Ω ; B,
f) = 1 Ω ; B,
f) .
Rij (Ω, (σi )lk (σj )mn Tkl,mn (Ω,
2
klmn
Again, properties (5.93) and (5.94) still hold.
The effect of perturbing collisions can be better understood if irreducible ten-
sors are introduced (see Sect. 5.10). Repeating the same arguments leading to
Eq. (5.121) one finds that this equation must be replaced by
Ω ; B,
f) = 3 )
E (σ, ρ, Ω
Tµν,ρσ (Ω, (−1)p−q Eqq (µ, ν, Ω) pp
2
qq pp
1 1 K 1 1 K
× (2K + 1)
p −p Q q −q Q
KQ
−1
Γ + Γc (1 − rK )
× + i (q − q )H .
Γ
Using then Eqs. (5.125) and (2.23a) one obtains
Ω
Tµν,ρσ (Ω, ; B,
f) =
1
−1
= (−1)Q EQ
K
(µ, ν, Ω) ) Γ + Γc (1 − rK ) + i QH
E K (σ, ρ, Ω ,
−Q
2 Γ
KQ
which is the same as Eq. (5.127) with the only substitution
−1
−1 Γ + Γc (1 − rK )
1 + i QH → + i QH .
Γ
Similarly, one finds that Eq. (5.133), which gives the scattering phase matrix for
the Stokes parameters, must be replaced by
−1
Ω
Rij (Ω, ; B,
f) = (−1)Q T Q
K
(i, Ω) ) Γ + Γc (1 − rK ) + i QH
T K (j, Ω .
−Q
Γ
KQ
These expressions show the substantial difference between magnetic field and
collisions as depolarizing agents in resonance scattering. Collisions affect all the
K-pole components (except that with K = 0, as it will be seen shortly), while the
magnetic field does not affect the Q = 0 components. It has been mentioned in
MATTER-RADIATION INTERACTION (CLASSICAL) 229
Sect. 5.13 – and will be shown in detail in Chap. 10 – that this feature is related to
the fact that the magnetic field can alter the coherences but not the populations
of the atomic levels, whereas collisions can alter both. In terms of classical dipoles
this means that collisions affect both phase correlations and amplitude differences
between oscillations, while the magnetic field affects only the former.
To discuss the results just obtained, we can write the depolarization factor in the
form
−1
Γ + Γc (1 − rK ) 1 1 1
+ i QH = = , (5.193)
Γ 1 + δK + i QH 1 + δK 1 + i QHK
where
Γc f
δK = (1 − rK ) = (1 − rK )
Γ γ
H 2πνL
HK = = . (5.194)
1 + δK γ + f (1 − rK )
This expression shows that the effect of collisions is twofold. First, each K-pole
contribution to the scattering phase matrix is reduced by the factor (1 + δK ), where
δK is a parameter proportional to the frequency of perturbing collisions or, in other
words, to the density of perturbers. It will be shown in the next section that the
values of δK depend on the specific assumptions which are made about the phase
shifts induced by perturbing collisions. The model considered here leads to the
values (see Eq. (5.199) of next section)
f 3 f
δ0 = 0 , δ1 = , δ2 = .
γ 5 γ
Thus
δ1 5
= , (5.195)
δ2 3
a result deduced through a quantum-mechanical calculation by Omont (1965) for
the case of Van der Waals, dipole-dipole interactions.
Moreover, perturbing collisions reduce the efficiency of the magnetic field in de-
polarizing (or in polarizing) the scattered radiation, the reduction being larger for
K = 1 than for K = 2.
As a specific example, let us consider the scattering geometry illustrated in
Fig. 5.10. In the absence of collisions the multipole components of the scatter-
ing phase matrix are given by Eqs. (5.139). In the presence of collisions, one finds
that the matrix R(1) must be divided by the factor (1 + δ1 ), and the quantity C1
appearing in its expression must be replaced by
C1 = cos α1 cos α1 + χ − χ ,
with α1 defined by
tan α1 = H1 .
230 CHAPTER 5
Similarly, the matrix R(2) must be divided by the factor (1 + δ2 ), and the quantities
C1 , S1 , C2 , S2 must be replaced by
C1 = cos α1 cos α1 + χ − χ S1 = cos α1 sin α1 + χ − χ
C2 = cos α2 cos α2 + 2(χ − χ) S2 = cos α2 sin α2 + 2(χ − χ) ,
with
tan α1 = H2 , tan α2 = 2H2 .
Finally, we want to recall that the theory presented in this section is based on
rather restrictive assumptions, and that it could in principle be generalized by
considering the effects of weak collisions, the existence of correlations between
different phase shifts, and so on.
5.15. Some Properties of the Collisional Kernels
We evaluate in this section – under the assumption of an isotropic velocity distribu-
tion of colliding particles – the average over all collision directions of the quantity
Kδα
(n) ∗
K(n)
β ,
where the kernels are defined in Eq. (5.184).
Starting from the kernel relative to the first collision, we have from Eqs. (5.185)
and (5.183)
Kδα
(1) ∗
K(1) (1) ∗
β = Kδα K β
(1)
2πi (α−δ−β+ ) νL t1
3
(1) ∗
(1)
−iφi (1)
3 (1)
iφj
= e Dδi e Dαi D(1)
j e
(1) ∗
Dβj . (5.196)
i=1 j=1
According to our hypotheses on perturbing collisions, the phase shifts relative to
the i-direction are totally uncorrelated with those relative to the j-direction. Thus
the double sum reduces to a single sum of the form
3
∗ ∗
S= D i Dδi Dαi Dβi ,
i=1
where
Dαi = uα · ti∗ = uα · ti .
Since we must average over all the possible orientations of the unit vectors ti , and
since, on the other hand, these three unit vectors are equivalent, we have
∗ ∗
S = 3 D i Dδi Dαi Dβi , (5.197)
MATTER-RADIATION INTERACTION (CLASSICAL) 231
where i = 1, or 2, or 3.
The average is easily evaluated using spherical tensors. In the reference system
of the magnetic field (ur , us , u0 ) we have, with the help of Eqs. (5.111), (5.112),
and (5.120)
Di= (u )q (ti )∗q = δ q (ti )∗q = (ti )∗ .
q q
From Eq. (5.116), identifying the unit vectors t1 , t2 , t3 with a, b, c, respectively,
we have
(tα )q = Dαq
1
(R) ,
where R is the rotation bringing (t1 , t2 , t3 ) into (ur , us , u0 ). Since in Eq. (5.197)
the index i can be chosen arbitrarily, taking i = 3 one has
∗ ∗
S = 3 D 3 Dδ3 Dα3 Dβ3 = 3 (t0 )∗ (t0 )δ (t0 )∗α (t0 )β
= 3 D01 (R)∗ D0δ
1
(R) D0α 1
(R)∗ D0β1
(R) .
To evaluate this expression we use twice Eq. (2.77) and then Eq. (2.73) and the
Weyl theorem (2.76) to get
δ−α 1 1 K 1 1 K
S = (−1) rK (2K + 1) , (5.198)
−δ Q β −α Q
KQ
where ⎧
2 ⎨1 for K = 0
1 1 K
rK = 3 = 0 for K = 1 (5.199)
0 0 0 ⎩
2/5 for K = 2 .
Substitution into Eq. (5.196) gives
(1) ∗ (1) 1 1 K 1 1 K
Kδα K β = (−1)δ−α rK (2K + 1) .
−δ Q β −α Q
KQ
The collisional kernels of higher degree can be evaluated by a recursive procedure.
Assuming the complete independence of successive collisions one obtains
(n) ∗ (n)
Kδα K β =
δ−α n 1 1 K 1 1 K
= (−1) rK (2K + 1) , (5.200)
−δ Q β −α Q
KQ
which – recalling the orthogonality relations (2.23b) – is valid also for n = 0.
It should be remarked that the expression for rK given in Eq. (5.199) is strictly
related to the assumptions on the phase shifts φi that we have made at the begin-
ning of Sect. 5.14. The motion of the atomic oscillator has been expanded into three
oscillations along a triplet of real , orthogonal unit vectors ti , and the phase shifts φi
232 CHAPTER 5
induced by a perturbing collision have been assumed to refer to such oscillations.
If we choose a different triplet of unit vectors, and retain the assumption of the
complete independence of the phase shifts along each of them, the final results will
generally be different. This is no wonder, as the choice of a specific triplet of unit
vectors is equivalent to a definite physical assumption on the effects of perturbing
collisions.
For example, let us choose the triplet of unit vectors
t−1 = √1 t1 + i t2
2
t0 = t3
t+1 = √1 −t1 + i t2 , (5.201)
2
with (t1 , t2 , t3 ) the usual triplet of real, orthogonal unit vectors. One finds the
same results as before, except that Eq. (5.197) must be replaced by the following
∗ ∗
S = D µ Dδµ Dαµ Dβµ
µ
(obviously the three unit vectors of Eq. (5.201) cannot be considered equivalent).
In terms of rotation matrices we have
1
S = Dµ (R)∗ Dµδ1
(R) Dµα
1
(R)∗ Dµβ
1
(R) ,
µ
and we obtain the same equation as Eq. (5.198) with
⎧
1 ⎨1 for K = 0
rK = = 1/3 for K = 1
2K + 1 ⎩ 1/5 for K = 2 .
Thus Eq. (5.195) becomes
δ1 5
= ,
δ2 6
which is just half the value obtained from the previous model.
5.16. Classification of the Physical Regimes
We have seen in this chapter how it is possible to describe in classical terms some
of the different phenomena which are able to induce polarization in spectral lines
or to affect its characteristics. In particular, we have seen that polarization can be
originated either by the presence of a magnetic field – which induces a frequency
splitting between the different classical dipoles – or by the presence of an intrinsic
MATTER-RADIATION INTERACTION (CLASSICAL) 233
ν ν
I II III
V Intense field regime
Collisional depolarization regime
∆ν IV Magnetograph regime ∆ν
Collision-dominated regime
Collision-free regime
III Strong field regime
γ II Hanle effect regime γ
I Zero field regime
γ γ
Fig.5.18. A classification scheme for polarization phenomena in spectral lines (the meaning of the
symbols is explained in the text).
anisotropy in the radiation field which is illuminating the atom. The anisotropy of
the radiation field, that is in general related both to its angular distribution and
to its polarization state, can induce amplitude differences between the oscillations
of the dipoles and well-defined phase correlations (or coherences) between such
oscillations. We have also seen that a magnetic field has the further effect of
reducing such phase correlations, and that collisions with an isotropic distribution
of perturbers produce a relaxation of the amplitude differences and of the phase
correlations.
All these phenomena, that will in general act simultaneously in a magnetized
plasma, can be characterized by four different parameters all having the dimensions
of frequency. These parameters are the following:
a) νL , the Larmor frequency, proportional to the magnetic field magnitude, which
is defined in Eq. (3.10);
b) ∆νD , the Doppler width, which depends on the temperature of the plasma (and
on its microturbulent velocity), defined in Eqs. (5.43) and (5.48);
c) γ, the oscillator damping constant, which has its quantum analogue in the
Einstein A coefficient for spontaneous emission from the upper level;
d) f , the frequency of depolarizing collisions, which is proportional to the number
density of perturbers.
According to the relative values of these four parameters we have different phys-
ical regimes that can be suitably classified in a two-dimensional diagram (Landi
Degl’Innocenti, 1983). Such diagram, shown in Fig. 5.18, bears a magnetic field
indicator (the Larmor frequency) on the vertical axis and a density indicator (the
frequency of collisions) on the horizontal axis.
In drawing the diagram we have assumed ∆νD γ. In terms of the thermal
velocity wT we have indeed
∆νD w
= T ,
γ λγ
234 CHAPTER 5
where λ is the line wavelength. Substituting typical values relative to a spectral line
of optical wavelength formed in a stellar atmosphere (wT = 2 km s−1 , λ = 5000 Å,
γ = 107 s−1 ) we get
∆νD
= 4.0 × 102 .
γ
The various physical regimes are sketched in Fig. 5.18; each of them can be
assigned a specific denomination.1 As far as the magnetic field is concerned, we
have the following five regimes:
Ia) Zero field regime (νL γ): the magnetic field is so weak as to induce a
negligible Zeeman splitting and a negligible relaxation effect on phase correlations.
In this regime the magnetic field can just be ignored.
IIa) Hanle effect regime (γ ≈ 2πνL ∆νD ): the magnetic field is strong enough
to reduce the phase correlations between different oscillators. By contrast, the
Zeeman splitting is still very small and can be neglected.
IIIa) Strong field regime (γ νL < ∆νD ): the relaxation of coherences is now
complete, so that the different oscillators can be treated as independent. The
Zeeman splitting is a small fraction of ∆νD and is able to produce an observable
(although weak) polarization signal.
IVa) Magnetograph (or intermediate) regime (γ νL ≈ ∆νD ): the Zeeman split-
ting is comparable with the Doppler width, which results in a strong polarization
signal; the different Zeeman components are, however, still unresolved. The obser-
vations performed by solar or stellar magnetographs usually fall in this regime.
Va) Intense field regime (νL > ∆νD ): the Zeeman splitting is now so large that the
different components are well-separated in the spectrum.
To sum up, we have that the Zeeman splitting can be neglected in regimes Ia and
IIa, while in regimes IIIa, IVa, and Va coherences are completely relaxed and can
be ignored. As far as the amplitude differences between oscillators are concerned,
the magnetic field has no effect on them.
Turning now to the influence of collisions, we have three different regimes:
Ib) Collision-free regime (f γ): depolarizing collisions have a negligible influence
on atomic polarization and can be disregarded.
IIb) Collisional depolarization regime (f ≈ γ): depolarizing collisions are important
in reducing atomic polarization (both coherences and amplitude differences between
oscillators) but not so strong as to produce a complete relaxation. In this regime
a detailed knowledge of the depolarizing rates is essential for a correct description
of the physical situation.
IIIb) Collision-dominated regime (f γ): the effect of depolarizing collisions is
so strong that atomic polarization is totally destroyed. In this regime there are no
coherences between different dipoles and the oscillation amplitudes are equal; in
other words, the classical dipoles are thermalized.
1
The denominations used here are slightly different from those in Landi Degl’Innocenti
(1983).
MATTER-RADIATION INTERACTION (CLASSICAL) 235
Zeeman
effect
∆ν
γ Hanle
effect Thermal
Resonance radiation
polarization
γ
Fig.5.19. The (νL - f ) diagram is divided into fifteen different regions resulting from the intersec-
tion of the physical regimes sketched in Fig.5.18.
By combining the five magnetic regimes and the three collisional regimes, one
finds out that there are 15 characteristic regimes for polarization in spectral lines,1
each occupying a well-defined region in the (νL - f ) diagram. Figure 5.19 points out
the regions of the diagram where conventional laboratory experiments are usually
performed.
It is important to remark that the classical description given above can be trans-
posed into quantum-mechanical terms. As outlined at the end of Sect. 5.13, the
phase correlations between different dipoles have their quantum analogue in the
off-diagonal density-matrix elements, while the oscillation amplitudes of the single
dipoles have their analogue in the diagonal matrix elements. Finally, the oscillator
damping constant γ has its quantum equivalent in the Einstein A coefficient for
spontaneous emission.
1
Indeed, regime (Ia-IIIb) is equivalent to (IIa-IIIb).
CHAPTER 6
INTERACTION OF MATERIAL SYSTEMS WITH POLARIZED
RADIATION (THE QUANTUM APPROACH)
We derive in this chapter the basic equations describing the interaction of a material
system with a polarized radiation field. These equations consist essentially of two
distinct sets, one for the material system (statistical equilibrium equations) and the
other for the radiation field (radiative transfer equations). But the two sets are
coupled, as the properties of the material system turn out to be affected by the
radiation field, and vice versa. From this point of view, the theory presented in
the following can be regarded as the basis for a generalized theory of Non Local
Thermodynamic Equilibrium which takes the polarization characteristics of the
material system and of radiation into account.
The two sets of equations are deduced by strictly similar procedures, which start
from the very principles of Quantum Mechanics and are based on the use of the
density operator. The radiation field is described with the formalism of (non-
relativistic) second quantization, while the material system is characterized by its
own Hamiltonian which does not need to be specified in detail. The interaction
is also treated under the non-relativistic approximation, and explicit formulae for
electric-dipole and magnetic-dipole transitions are derived.
The formalism used in this chapter is rather complicated, but the preceding
outlines of density operator and second quantization theories (Chaps. 3 and 4,
respectively) should enable the reader to follow all the developments. We also tried
to point out, whenever possible, the physical meaning of the various equations and
the analogies with the classical treatment presented in Chap. 5.
The results that will be obtained in this chapter suffer from some basic limita-
tions that arise both from the formalism itself and from the different approxima-
tions introduced in the derivation. Indeed, such results cannot be applied when
coherences between non-degenerate levels are present, unless the spectrum of the
radiation incident on the material system is flat across a range wider than the
frequency separation of those levels. Similarly, if coherences between degenerate
levels are present, the results provide an exact description of scattering phenomena
only if the spectrum of the incident radiation is flat across a frequency range wider
than the inverse lifetime of the levels.
These limitations are analogous – in the polarized case – to those contained in
the ‘complete frequency redistribution’ approach of the standard non-LTE theory
(see e.g. Mihalas, 1978). They are ultimately related to the fact that two consec-
utive interactions of the material system with the radiation field are considered as
independent: this means, for instance, that when an emission event takes place, no
memory is kept of the frequency of the photon that induced the previous absorption
event.
238 CHAPTER 6
Notwithstanding its basic limitations, the theory has a wide range of physical and
astrophysical applications because, in many cases, the spectrum of the incident
radiation is in fact flat across the relevant frequency range. Usually coherences
are important as long as the separation between energy levels is less than about
108 s−1 (a typical value of the spontaneous de-excitation Einstein coefficient), and
such frequency interval – which corresponds to 1 mÅ in the visible – is much smaller
than the typical Doppler widths of spectral lines.
However, there are some phenomena that cannot be fully described by the formal-
ism presented in this chapter, like the quantum-interference polarization observed
in the CaII H and K lines at the solar limb (Stenflo, 1980). In that case, coherences
between energy levels separated by a very large interval (about 50 Å) are found
to play a major role, and the incident radiation field is definitely not flat across
such interval. In these cases a more general formalism, able to encompass fre-
quency redistribution effects, is required. Although some results have indeed been
obtained on this subject (Stenflo, 1994; Bommier, 1997a,b; Landi Degl’Innocenti et
al., 1997), the underlying theory is still in a preliminary phase and remains outside
the scope of this book.
This chapter is based on previous works by the authors and collaborators (Landi
Degl’Innocenti and Landi Degl’Innocenti, 1972, 1975; Landi Degl’Innocenti et al.,
1976; Landi Degl’Innocenti, 1983). The same method has been reconsidered and
illustrated in some detail by Cannon (1985), Trujillo Bueno (1990), and Stenflo
(1994).
6.1. Equations of Motion
According to the principles of Quantum Mechanics (see e.g. Messiah, 1961) the
state of a physical system at time t is described by a vector in the Hilbert space,
| ψ(t) , which satisfies the normalization condition
ψ(t) | ψ(t) = 1 .
The time evolution of the system is governed by the well-known Schrödinger equa-
tion
h d
i | ψ(t) = H | ψ(t) ,
2π dt
where i is the imaginary unit, h is the Planck constant, and H is the Hamiltonian
operator that we suppose here independent of time.
On the other hand, any observable of the system is described by a linear, Her-
mitian operator acting on the same Hilbert space. The expectation value of the
observable (i.e. the average value of the measurement of the observable ideally
performed on an infinite number of ‘copies’ of the same physical system) is given
by
O(t) = ψ(t) | Ô | ψ(t) ,
where Ô is the quantum operator corresponding to the observable O.
MATTER-RADIATION INTERACTION (QUANTUM) 239
However, there are many physical situations where the state of the system cannot
be described by a pure state | ψ(t) (nor by a superposition of pure states), but a
detailed quantum-statistical description is necessary. This can be achieved in a very
natural way using the concept of density (or statistical) operator (see Sect. 3.6).
We recall that the expectation value of the observable O is related to the density
operator ρ by the expression (see Eq. (3.86))
6 7
O(t) = Tr Ôρ . (6.1)
The time evolution of O(t) can be easily obtained from Eq. (6.1). We have
! !
d dÔ dρ
O(t) = Tr ρ + Tr Ô , (6.2)
dt dt dt
where the first term is non-zero only when the operator Ô depends explicitly on
time. This equation can be rewritten, with the help of Eq. (3.88), in the form
! !
d dÔ 2πi
O(t) = Tr ρ − Tr Ô H, ρ ,
dt dt h
or, using the cyclic property of the trace, in the more convenient form
! !
d dÔ 2πi
O(t) = Tr ρ − Tr Ô, H ρ . (6.3)
dt dt h
In many cases the total Hamiltonian H of the system can be expressed as the
sum of two terms,
H = H0 + V , (6.4)
where H0 is the so-called unperturbed Hamiltonian and V is the interaction Hamil-
tonian. In such cases it is in many respects useful to introduce a different rep-
resentation for the quantum phenomena, called the interaction picture. This is
obtained from the usual Schrödinger picture – which has been employed so far –
by the following unitary transformation
a) for the state vectors:
i 2π H0 t
| ψ(t) I = e h | ψ(t)
b) for the operators:
i 2π H0 t −i 2π H0 t
ÔI (t) = e h Ô e h , (6.5)
where the index I means that the corresponding symbol (operator, or state vector)
is defined in the interaction picture.
The equation of motion for the density operator expressed in the interaction
picture can be easily obtained from Eq. (6.5). Using Eqs. (3.88) and (6.4) we get
d 2πi
ρI (t) = − VI (t), ρI (t) , (6.6)
dt h
240 CHAPTER 6
where
i 2π H0 t −i 2π H0 t
VI (t) = e h V e h (6.7)
is the interaction Hamiltonian expressed in the interaction picture.
Substitution of Eq. (6.6) into Eq. (6.2) leads to
! !
d d 2πi
O(t) = Tr Ô (t) ρI (t) − Tr ÔI (t), VI (t) ρI (t) . (6.8)
dt dt I h
Comparison of Eqs. (6.8) and (6.3) shows that when the interaction picture is
used, the time evolution of the expectation value O(t) is determined by the inter-
action Hamiltonian only (rather than the total Hamiltonian, as in the Schrödinger
picture). When the interaction Hamiltonian is small in comparison with the un-
perturbed Hamiltonian,1 Eq. (6.8) can be usefully expanded as follows. Integration
of Eq. (6.6) between 0 and t gives
t
2πi
ρI (t) = ρI (0) − VI (t ), ρI (t ) dt ,
h
0
which can be substituted into Eq. (6.8) to obtain, using the cyclic property of the
trace
! !
d d 2πi
O(t) = Tr ÔI (t) ρI (t) − Tr ÔI (t), VI (t) ρI (0)
dt dt h
t !
4π 2
− 2 Tr ÔI (t), VI (t) , VI (t ) ρI (t ) dt . (6.9)
h
0
This is an exact equation describing the time evolution of the observable O(t)
produced by the interaction Hamiltonian. In the following sections it will be applied
to describe the interaction between a material system and the radiation field.
6.2. The Interaction Hamiltonian
We consider the interaction between an arbitrary material system and the radiation
field. The material system is here schematized as the collection of N independent
subsystems (that will be called ‘atoms’) occupying a definite volume V in the
ordinary three-dimensional space. The total Hamiltonian of the coupled system
composed of one atom and the radiation field can be written, in the Schrödinger
picture, in the form
H = H0 + V ,
1
‘Small’ means here that the typical matrix elements of V are much smaller than the matrix
elements of H0 .
MATTER-RADIATION INTERACTION (QUANTUM) 241
where H0 – the unperturbed Hamiltonian – is the sum of the Hamiltonians of the
free radiation field and of the atom,
H0 = HR + HA . (6.10)
The expression for HR has already been deduced in Chap. 4, where the formalism
of second quantization has been introduced. From Eq. (4.34) we have
HR = hν a† (ν, Ω,
λ) a(ν, Ω,
λ) , (6.11)
ν Ωλ
λ) and a† (ν, Ω,
where a(ν, Ω, λ) are, respectively, the destruction and creation oper-
ators associated with the mode of frequency ν, direction Ω, and polarization state
characterized by the unit vector eλ (λ = 1, 2), and where the sum runs over all the
possible modes of the radiation field.
The expression of the atomic Hamiltonian HA depends on the particular atomic
system and on the possible presence of external agents (like, for instance, a static,
electric or magnetic field). For the time being we make no specific assumption
on HA . However, we suppose to know its normalized eigenstates | n and energy
eigenvalues En , solution of the stationary Schrödinger equation
HA | n = En | n , (6.12)
with
n | m = δnm (6.13)
and
|n n| = 1 . (6.14)
n
The interaction Hamiltonian V can be deduced using the standard methods of
Quantum Electrodynamics. According to Cohen-Tannoudji et al. (1977, 1987) the
expression for V is given, in the non-relativistic approximation, by the expression
e 2 , -2
V = 0 r ) + e0
pi · A(
s · r ) + e0
B( r) ,
A( (6.15)
i
mc i mc i i i
2mc2 i i
where ri , pi , and si are the position, momentum, and spin operators of the i-th
optical electron of the atom, and where A( r ) and B(
r ) are the operators corre-
sponding to the vector potential and to the magnetic field vector of the radiation
field, respectively.
Writing the different contributions to V in the form
V = V1 + V2 + V3 ,
we can easily find the order of magnitude of the various terms. We have just to
recall that, as an order of magnitude
h c h
p≈ , A≈ E, s≈ , B≈E,
2πa0 2πν 2π
242 CHAPTER 6
where E and ν are the electric field and the frequency of the radiation field, re-
spectively, and where a0 is the Bohr radius. Moreover, for the atomic Hamiltonian
HA we have
p2 e2
HA ≈ ≈ 0 ,
m a0
so that we obtain
V1 V3 e0 Ea0 V 2πa0
≈ ≈ , 2 ≈ . (6.16)
HA V1 hν V1 λ
At optical wavelengths the quantity hν is comparable with the atomic energy, so
that we can also write
V V e Ea E
1 ≈ 3 ≈ 0 0 = ,
HA V1 e0 Eat a0 Eat
where Eat is a typical interatomic electric field. Thus when the ratio between
the radiation electric field and the typical interatomic electric fields is small, V1
can be considered a perturbation in comparison with HA , and V3 a perturbation
(of the same order) in comparison with V1 . On the other hand, Eq. (6.16) shows
that V2 , at optical wavelengths, is also much smaller (by about four orders of
magnitude) than V1 . It follows that if we restrict attention to typical astrophysical
or laboratory plasmas, where the inequality E/Eat 1 is always well-satisfied,1
and to wavelengths λ such that λ a0 , we can simply write the interaction
Hamiltonian in the form
e r).
V = 0 pi · A(i (6.17)
mc i
This is the form that will be actually used in the following to describe the interaction
between the atomic system and the radiation field.
Substituting the expression of the operator A( r ) in terms of destruction and
creation operators (Eq. (4.30)) into Eq. (6.17), we obtain
& '
V = Q(ν, Ω, λ) + Q† (ν, Ω,
λ) a(ν, Ω, λ) a† (ν, Ω,
λ) , (6.18)
ν Ωλ
λ) and its adjoint Q† (ν, Ω,
where the operator Q(ν, Ω, λ) are given by2
λ) = d
Q(ν, Ω, e ik·ri
pi · eλ (Ω)
ν
i
† λ) = d
Q (ν, Ω, ∗ e−ik·ri ,
pi · eλ (Ω)
ν
i
1
This inequality is not satisfied in optical experiments performed with high-power lasers,
giving rise to non-linear optics phenomena.
2
To prove that the operator multiplying a† (ν, Ω, λ) in Eq. (6.18) is just the adjoint of
Q(ν, Ω, λ) one needs the commutation rule
[ p · eλ (Ω)
, eik·r ] = 0 ,
which follows from the commutation rule [rj , pk ] = ih̄ δjk and from Eq. (4.31).
MATTER-RADIATION INTERACTION (QUANTUM) 243
with .
e h
dν = 0 . (6.19)
m 2πνV
The interaction Hamiltonian given in Eq. (6.18) is expressed in the Schrödinger
picture. To derive its expression in the interaction picture, it should be noticed
λ) and Q† (ν, Ω,
that the operators Q(ν, Ω, λ) act only on the atomic system, while
†
the operators a(ν, Ω, λ) and a (ν, Ω, λ) act only on the radiation field, which implies
λ) = H , a(ν, Ω,
HR , Q(ν, Ω, λ) = H , H = 0.
A R A
We thus obtain from Eq. (6.7)
i 2π H0 t −i 2π H0 t
VI (t) = e h V e h = B(t) + B † (t) , (6.20)
where
i 2π H t 2π
!
i 2π H t 2π
!
B(t) = λ) e−i h HA t
e h A Q(ν, Ω, λ) e−i h HR t . (6.21)
e h R a(ν, Ω,
ν Ωλ
Denoting by P (t) the operator resulting from the second curly bracket, we can
easily prove that
−2πiνt
P (t) = e λ) .
a(ν, Ω, (6.22)
The formal proof is as follows. For t = 0 we have
λ) ,
P (0) = a(ν, Ω, (6.23)
and the time derivative of P (t) is
d 2πi i 2π H t 2π
λ) e−i h HR t .
P (t) = e h R HR , a(ν, Ω,
dt h
On the other hand, taking into account the expression for HR (Eq. (6.11)) and the
commutation properties of the operators a and a† (Eqs. (4.32)), one gets
λ) = −hν a(ν, Ω,
HR , a(ν, Ω, λ) ,
so that
d
P (t) = −2πi ν P (t) .
dt
Integration of this differential equation with the initial condition (6.23) leads to
expression (6.22).
244 CHAPTER 6
As far as the first curly bracket is concerned, we can express it in a more con-
venient form by introducing the complete set of eigenvectors {| n } of the atomic
Hamiltonian defined in Eqs. (6.12)-(6.14). We have
i 2π HA t λ) e −i 2π HA t
e h Q(ν, Ω, h =
i 2π HA t −i 2π HA t
= e h |n λ) | m
n | Q(ν, Ω, m| e h
nm
2πi νnm t
= λ)
dν q(ν, Ω, e |n m| , (6.24)
nm
nm
where νnm are the Bohr frequencies
En − Em
νnm = , (6.25)
h
λ)]
and where the matrix elements [q(ν, Ω, nm are given by
λ)
q(ν, Ω, = n| e i k·ri | m .
pi · eλ (Ω) (6.26)
nm
i
Substituting Eqs. (6.24) and (6.22) into Eq. (6.21) we get
B(t) = λ)
dν q(ν, Ω, λ) e 2πi (νnm − ν) t
| n m | a(ν, Ω, (6.27)
nm
ν Ωλ nm
and, similarly
B † (t) = dν q † (ν, Ω,
λ) |n λ) e 2πi (νnm + ν) t ,
m | a† (ν, Ω, (6.28)
nm
ν Ωλ nm
where
†
∗ e−i k·ri | m = q(ν, Ω,
λ)
q (ν, Ω, = n | pi · eλ (Ω) λ) ∗ .
nm mn
i
6.3. The Dipole Approximation
λ)]
The matrix element [q(ν, Ω, nm defined in Eq. (6.26) can be somewhat simplified
by a series of transformations that are discussed here in some detail.
First we observe that the coordinate ri of the i-th electron is measured from an
arbitrary origin. Introducing the relative coordinate xi , measured from the center
of mass of the atomic system, we have
+ x ,
ri = X i
MATTER-RADIATION INTERACTION (QUANTUM) 245
is the coordinate of the center of mass. With this position, the exponential
where X
λ)]
appearing in the definition of [q(ν, Ω, nm can be written as
i
k·
ri i
k·X i
k·
xi
e = e e ,
and expanding the last exponential in power series
i
k·
ri i
k·X 1 2
e = e 1 + i k · xi − k · xi + · · · . (6.29)
2
Since at optical wavelengths
k · x ≈ 2πa0 1 ,
i
λ
we can simply write (long-wavelength approximation)
i
k·
ri i
k·X
e = e , (6.30)
which is consistent with our previous approximation of neglecting V2 in the general
expression for the interaction Hamiltonian (see Eq. (6.16)).
To avoid the introduction of a heavier formalism, we suppose the center of mass
of the atomic system to be at rest, so that, taking X = 0, we get
λ)
q(ν, Ω, = n| |m .
pi · eλ (Ω)
(6.31)
nm
i
It should be remarked, however, that the theory presented in the following cannot
give – because of the assumption X = const. – a consistent description of the
phenomena associated with the motion of the atomic system (Doppler effect). We
observe that if the coordinate X is considered as an ordinary variable describing
the physical state of the atomic system (like the relative coordinates xi of the
optical electrons) the formalism of Quantum Electrodynamics is able to provide
an elegant derivation of the fundamental formulae of the Doppler effect – which
basically result from the laws of conservation of momentum and energy. The reader
is referred to Louisell (1973) for an example of such derivation; the original idea
of deducing the Doppler effect from the momentum-energy conservation is due to
Fermi (1932).
Equation (6.31) can be further simplified by considering the expression of the
atomic Hamiltonian HA . Neglecting relativistic corrections, we have
p2
i
HA = + VCoul ,
i
2m
where VCoul represents the electrostatic energy of the electrons which depends on
the coordinates xi . Using the commutation rule of the operators pi and xi , and
defining
x = xi ,
i
246 CHAPTER 6
we have
h
HA , x = −i pi ,
2πm i
thus Eq. (6.31) can be written in the form
λ)
q(ν, Ω, m ,
= 2πm i νnm n | x · eλ (Ω)| (6.32)
nm
where νnm is the Bohr frequency defined in Eq. (6.25). This new expression involves
the matrix elements of the operator x , which coincides, apart from the factor -e0 ,
with the dipole operator
d = −e0 x .
This justifies the alternative name of (electric) dipole approximation that is given
to the long-wavelength approximation (Eq. (6.30)).
An important point concerning the dipole approximation should be explicitly
stressed. Substitution of Eq. (6.32) into Eq. (6.27) gives
.
2πh
B(t) e.d. = i e0 νnm m
n | x · eλ (Ω)|
νV
ν Ωλ nm
× |n λ) e 2πi (νnm − ν) t ,
m | a(ν, Ω, (6.33)
where Eq. (6.19) has been used. This expression can be compared with a similar
expression that is obtained by assuming a priori that the interaction Hamiltonian
has the form
V = r)
e0 xi · E(i
i
r ), passing to the inter-
instead of the form (6.17). Substituting Eq. (4.33) for E(i
action picture, and adopting again the long-wavelength approximation, we get
i 2π H0 t −i 2π H0 t
VI (t) = e h V e h = B (t) + B † (t) ,
where
.
2πνh
m
B (t) = i e0 n | x · eλ (Ω)|
V
ν Ωλ nm
2πi (νnm − ν) t
× |n λ) e
m | a(ν, Ω, . (6.34)
Comparison of Eqs. (6.33) and (6.34) shows that B (t) can be obtained from
[B(t)]e.d. by replacing the factor νnm with ν. In most physical applications the
distinction between these two forms of the interaction Hamiltonian is insignifi-
cant. Roughly speaking, this is due to the presence of the rapidly oscillating factor
exp [2πi (νnm − ν) t] which, for large values of t, is zero unless νnm = ν. A full
MATTER-RADIATION INTERACTION (QUANTUM) 247
discussion on this subject, together with a formal proof on how the two differ-
ent Hamiltonians can be connected by a similarity transformation (the so-called
Göppert-Mayer transformation) can be found in Cohen-Tannoudji et al. (1987).
In this book we will take for B(t) the more compact expression of Eq. (6.34). The
final form of the interaction Hamiltonian is therefore
VI (t) = B(t) e.d. + B † (t) e.d. ,
where
2πi (νnm − ν) t
B(t) e.d. = −i cν d · eλ (Ω)
nm
|n λ) e
m | a(ν, Ω,
ν Ωλ nm
B † (t) e.d. = i ∗
cν d · eλ (Ω) |n λ) e 2πi (νnm + ν) t
m | a† (ν, Ω, (6.35)
nm
ν Ωλ nm
with
.
2πνh
cν =
V
d = −e0 xj = −e0 x
j
d · eλ (Ω)
nm
= n | d · eλ (Ω)|
m = n | d | m · e (Ω)
λ
= (d )nm · eλ (Ω)
∗
d · eλ (Ω) ∗ | m = n | d | m · e (Ω)
= n | d · eλ (Ω) ∗
nm λ
∗
= (d )nm · eλ (Ω) (6.36)
and with
∗
d · eλ (Ω) ∗ .
= d · eλ (Ω) (6.37)
nm mn
In some cases, the dipole matrix elements between two particular levels | n and
| m may well be zero. This means that the electric dipole approximation gives no
contribution to the transition between such levels, and one must go back to those
parts of the interaction Hamiltonian that have been neglected so far. The various
terms neglected in the Taylor expansion (6.29) give rise to the electric quadrupole,
electric octupole, etc., Hamiltonians and to those parts of the magnetic-multipole
Hamiltonian related to the orbital angular momentum of the electrons, while the
second term in the right-hand side of Eq. (6.15) gives rise to those parts of the
magnetic-multipole Hamiltonian related to the electron spin.
In particular, let us consider the so-called magnetic-dipole transitions. Setting
again X = 0, the contribution of the second term of the expansion (6.29) to the
interaction Hamiltonian can still be written in the form of Eq. (6.18) with the
248 CHAPTER 6
substitutions1
λ) → Q (ν, Ω,
Q(ν, Ω, λ) =
1
i
= dν k · x + k · x p · e (Ω)
pi · eλ (Ω)
i i i λ
2 i
λ) → Q† (ν, Ω,
Q† (ν, Ω, λ) =
1
i
= − dν ∗ k · x + k · x p · e (Ω)
pi · eλ (Ω) ∗ . (6.38)
i i i λ
2 i
Introducing the orbital angular momentum of the i-th electron,
l = x ×
pi ,
i i
it can be easily shown that
k · x + k · x p · e (Ω)
pi · eλ (Ω) = l · k × e (Ω)
i i i λ i λ
+ k · p x · e (Ω)
+ k · xi pi · eλ (Ω) i i λ
.
Recalling that k = 2πν Ω/c
, we obtain
λ) = i c e0
Q1 (ν, Ω, l · Ω
× e (Ω)
ν i λ
2mc i
!
+
Ω · xi
pi · eλ (Ω) + Ω · pi xi · eλ (Ω)
i
λ) = −i c e0
Q†1 (ν, Ω, l · Ω ∗
× e (Ω)
ν i λ
2mc i
!
+ Ω ∗+ Ω
· x p · e (Ω) ∗
· p x · e (Ω) , (6.39)
i i λ i i λ
i
where the definitions of dν and cν (Eqs. (6.19), (6.36)) have been used.
Consider now the second term of the interaction Hamiltonian in Eq. (6.15). Using
r ), and adopting the long-wavelength approximation (6.30)
expression (4.33) for B(
with X = 0, we see that this term can also be written in the form of Eq. (6.18)
with the substitutions
λ) → Q (ν, Ω,
Q(ν, Ω, λ) = i c e0 × e (Ω)
si · Ω
2 ν λ
mc i
λ) = −i c e0
λ) → Q† (ν, Ω,
Q† (ν, Ω, si · Ω ∗.
× e (Ω) (6.40)
2 ν λ
mc i
1
It can be easily shown that the operators pi ·eλ (Ω)
and k·x commute (cf. footnote 2 on p.242).
i
However, the symmetrized forms in Eqs. (6.38) lead in the most straightforward way to the correct
decomposition of the interaction Hamiltonian into the electric-quadrupole and magnetic-dipole
terms (see e.g. Shu, 1991).
MATTER-RADIATION INTERACTION (QUANTUM) 249
The terms in square brackets of Eqs. (6.39) are responsible for electric-quadrupole
transitions, while the first terms, combined with those in Eqs. (6.40), are responsible
for magnetic-dipole transitions. Passing to the interaction picture, the magnetic-
dipole Hamiltonian can thus be written in the form
VI (t) m.d. = B(t) m.d. + B † (t) m.d. ,
where
B(t) m.d. = −i cν × e (Ω)
µ·Ω
λ nm
ν Ωλ nm
2πi (νnm − ν) t
× |n λ) e
m | a(ν, Ω,
†
B (t) m.d. = i cν µ·Ω ∗
× e (Ω)
λ nm
ν Ωλ nm
× |n λ) e 2πi (νnm + ν) t .
m | a† (ν, Ω, (6.41)
In these expressions the quantity
e0
µ =−
li + 2si
2mc i
is the magnetic moment associated with the electrons, and the matrix elements are
defined in strict analogy with Eqs. (6.36).
Comparison of Eqs. (6.41) and (6.35) shows that the magnetic-dipole Hamilto-
nian can be simply obtained from the electric-dipole Hamiltonian by the formal
substitutions
d →
µ, →Ω
eλ (Ω) × e (Ω)
λ
. (6.42)
6.4. Approximate Equations of Motion
Substituting Eq. (6.20) into Eq. (6.9), the time variation of any physical observable
can be written as
! !
d d 2πi
O(t) = Tr ÔI (t) ρI (t) − Tr ÔI (t), B(t) ρI (0)
dt dt h
!
2πi †
− Tr ÔI (t), B (t) ρI (0)
h
t !
4π 2
− 2 Tr ÔI (t), B(t) , B(t ) ρI (t ) dt +
h
0
250 CHAPTER 6
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt
h
0
t !
4π 2
− Tr ÔI (t), B † (t) , B(t ) ρI (t ) dt
h2
0
t !
4π 2
− Tr ÔI (t), B † (t) , B † (t ) ρI (t ) dt . (6.43)
h2
0
It should be emphasized that this is a very general, exact equation that follows
directly from the principles of Quantum Mechanics. Now we are going to introduce
a number of assumptions which will enable us to derive a self-contained set of
evolution equations for the physical quantities characterizing the interaction of an
atomic system with the radiation field.
First of all we suppose that the density operator of the full system is the direct
product of the density operators of the individual systems,
I (t) ⊗ ρI (t)
ρI (t) = ρ(R) t≥0.
(A)
for (6.44)
In other words, if Ô(R) is an operator acting only on the radiation field, and Ô(A)
is an operator acting only on the atomic system, we assume that the expectation
value of their product is equal to the product of their expectation values,
& ' & ' & '
Tr ÔI(R) ÔI(A) ρI = Tr(R) ÔI(R) ρ(R)
I Tr (A)
ÔI
(A) (A)
ρ I ,
where Tr(R) and Tr(A) denote tracing over the states of the radiation field and of
the atomic system, respectively.
The physical meaning of this approximation is that the radiation field and the
atomic system are supposed to be uncorrelated. At time t = 0 (when the interac-
tion starts) this is indeed a reasonable approximation to describe the interaction
between light and matter in an environment like a stellar atmosphere. In such an
environment an atom interacts with radiation quanta that have been emitted at
typical distances of the order of many kilometers, so that atom-radiation correla-
tions can be safely neglected. The above approximation is also valid in laboratory
experiments, provided the light source is physically separated from the cell (or
analogous device) containing the atoms. By contrast, the approximation breaks
down in other physical conditions (like for instance in lasers, where atom-radiation
correlations play a major role).
Equation (6.44) postulates, however, that the uncorrelation between the radia-
tion field and the atomic system persists for t > 0. When Eq. (6.44) is substituted
into Eq. (6.43), we obtain an equation which can be considered as the lowest-order
approximation to the exact equation of motion. This approximation is justified
when the relaxation time of the density operator, tr , is much longer than the typ-
ical temporal factors which appear in Eq. (6.43) when the explicit expressions of
MATTER-RADIATION INTERACTION (QUANTUM) 251
B(t) and B † (t) are used (Eqs. (6.27)-(6.28), or (6.35)-(6.36) in the dipole approxi-
mation).
Our second assumption concerns the properties of the radiation field, and can be
expressed by the set of conditions
& ' & '
λ) ρ(R) (t) = Tr(R) a† (ν, Ω,
Tr(R) a(ν, Ω, λ) ρ(R) (t) = 0 (6.45)
I I
& '
λ) a(ν , Ω
Tr(R) a(ν, Ω, , λ ) ρ(R) (t) = 0 (6.46a)
I
& '
Tr(R) a† (ν, Ω,
λ) a† (ν , Ω
, λ ) ρ(R) (t) = 0
I (6.46b)
& '
Tr(R) a† (ν, Ω,
λ) a(ν , Ω
, λ ) ρ(R) (t) = 0 unless ν = ν , Ω
I
.
=Ω (6.47)
Equation (6.45) means that the expectation value of any observable which is linear
either in the destruction or the creation operator of the radiation field is zero. If
we consider, for instance, the expectation value of the electric or magnetic field
(whose associated operators are given, in the Schrödinger picture, by Eqs. (4.33))
we have & 2π '
2π
r ) e−i h H0 t ρ(R) (t) = 0
r, t) = Tr(R) e i h H0 t E(
E( I
and, similarly
r, t) = 0 .
B(
We thus obtain the analogue of the classical result that the statistical average of
such quantities is zero as long as the electromagnetic field can be regarded as the
incoherent superposition of different wavetrains with random phases.
Equation (6.45) is, however, somewhat restrictive in that it prevents the consider-
ation of processes involving coherent states of the radiation field (see e.g. Glauber,
1964, for an introduction to these concepts). Nevertheless, this approximation is
quite reliable to describe the radiation field typical of a stellar atmosphere, where
optical-coherent phenomena are believed to play a totally insignificant role. The
same is true for a quite large variety of laboratory experiments.
The meaning of Eqs. (6.46) is similar; indeed, the left-hand side of Eq. (6.46a)
represents, up to a factor, the statistical average of the quantity ckλ (t) ck λ (t),
where ckλ (t) is the amplitude of the Fourier component of the electric field vector
associated with a particular mode (see Eq. (4.20)). Similarly, the left-hand side
of Eq. (6.46b) represents the statistical average of the quantity ckλ (t)∗ ck λ (t)∗ .
Again, the approximation of Eqs. (6.46) is fully justified for electromagnetic fields
that can be described as the incoherent superposition of different wavetrains with
random phases.
Finally, assumption (6.47) means that no correlations exist between different
modes of the radiation field, except for correlations between polarization states
within each mode.
Using approximations (6.44)-(6.46) we can considerably simplify Eq. (6.43) for
two important classes of operators ÔI (t), namely the operators acting only on the
252 CHAPTER 6
atomic system and the operators acting only on the radiation field and having the
form a† (ν, Ω,
λ) a(ν , Ω
, λ ). It can be easily shown that for such operators, owing
to the form of B(t) and B † (t) (Eqs. (6.27)-(6.28)), Eq. (6.43) reduces to
!
d d
O(t) = Tr Ô (t) ρI (t)
dt dt I
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt
h
0
t !
4π 2
− Tr ÔI (t), B † (t) , B(t ) ρI (t ) dt .
h2
0
Because of the Hermitian character of the density operator and of the cyclic prop-
erty of the trace, this equation can be rewritten as
!
d d
O(t) = Tr ÔI (t) ρI (t)
dt dt
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt
h
0
& '
+ c.c. ÔI (t) → ÔI† (t) , (6.48)
where the last symbol denotes the quantity that is obtained by taking the complex
conjugate of the previous term and by replacing the operator ÔI (t) with its adjoint
ÔI† (t).
This is the basic equation that will be used in the following to derive the time
evolution of the radiation field and of the material system resulting from their
mutual interaction.
6.5. Evolution Equations for the Atomic System
A complete description of the atomic system can be obtained through the knowl-
edge of the instantaneous value of the matrix elements of the density operator
ρ(A) (t), & '
ρmm (t) ≡ m | ρ(A) (t) | m = Tr(A) | m m | ρ(A) (t) , (6.49)
where | m and | m are two arbitrary eigenvectors of the atomic Hamiltonian,
and ρ(A) (t) is the time-dependent density operator of the atomic system expressed
in the Schrödinger picture.
From Eq. (3.87) we have
& '
ρ(A) (t) = Tr(R) ρ(t) ,
MATTER-RADIATION INTERACTION (QUANTUM) 253
where ρ(t) is the density operator of the whole system, so that we can also write
& '
ρmm (t) = Tr | m m | ρ(t) .
If we now pass to the interaction picture, we get
& '
ρmm (t) = Tr ÔI (t) ρI (t) , (6.50)
where ÔI (t) is given by (see Eqs. (6.5) and (6.10))
i 2π HA t −i 2π HA t 2πi νm m t
ÔI (t) = e h | m m| e h = | m m| e . (6.51)
Thus the density-matrix element ρmm (t) represents the expectation value of an
operator given by Ô = | m m | in the Schrödinger picture, and by Eq. (6.51) in
the interaction picture.
To derive the time evolution of ρmm (t) we will now apply Eq. (6.48), taking for
B(t) and B † (t) their expressions in the electric-dipole approximation (Eqs. (6.35)).
The first term is easily evaluated; from Eqs. (6.51) and (6.50) we get
! & '
d 2πi νm m t
Tr ÔI (t) ρI (t) = 2π i νm m Tr | m m| e ρI (t)
dt
= 2π i νm m ρmm (t) . (6.52)
For the second term we have, on the contrary, a rather involved expression
ÔI (t), B(t) , B † (t ) =
= cν cν d · eλ (Ω)
)∗
d · eλ (Ω
nn rr
νΩ
ν Ωλ λ nn rr
2πi (νm m + νnn − ν) t + 2πi (νrr + ν ) t
× e C, (6.53)
where C is the double commutator
C= | m m|, |n n | a(ν, Ω,
λ) , | r r | a† (ν , Ω
, λ ) .
C can be easily evaluated using the commutation rule of the creation and destruc-
tion operators (Eq. (4.32)), and the commutation rule of the projection operators
|a b|,|c d | = δbc | a d | − δad | c b| .
254 CHAPTER 6
After some algebra, rearranging the final result in such a way that a† -operators
precede a-operators, we get
, -
C = δmn δn r | m r | a† (ν , Ω
, λ ) a(ν, Ω,
λ) + δ δ δ
νν ΩΩ λλ
− δmn δm r | r n | a† (ν , Ω
, λ ) a(ν, Ω,
λ)
, -
− δm n δmr | n r | a† (ν , Ω , λ ) a(ν, Ω,
λ) + δ δ δ
νν ΩΩ λλ
+ δm n δnr | r m | a† (ν , Ω
, λ ) a(ν, Ω,
λ) . (6.54)
Substituting this expression into Eq. (6.53) and renaming the summation indices
in the first line of Eq. (6.54) according to
n = r → n , r → m ,
we obtain
ÔI (t), B(t) , B † (t ) =
= cν c ν d · eλ (Ω)
)∗
d · eλ (Ω
mn nm
νΩ
ν Ωλ λ nm
2πi (ν − ν) t + 2πi (ν + ν ) t
× e m n nm
| m m |
, -
× a† (ν Ω
λ ) a(ν, Ω,
λ) + δ δ δ
νν ΩΩ λλ
+ (other terms) , (6.55)
where ‘(other terms)’ denotes the contribution arising from the other lines of
Eq. (6.54).
Now we multiply Eq. (6.55) by ρI (t ) and evaluate the trace. Using the approxi-
mations of Eqs. (6.44) and (6.47), and observing that
νnm = νm m − νm n ,
we get
!
Tr ÔI (t), B(t) , B † (t ) ρI (t ) =
2πi (νm n − ν)(t−t )
= c2ν d · eλ (Ω)
∗
d · eλ (Ω) e
mn nm
ν Ωλλ nm
& 2πi νm m t
'
× Tr(A) | m m | e ρ(A)
I (t
)
& '
× Tr(R) a† (ν, Ω,
λ ) a(ν, Ω,
λ) + δ ρ(R) (t )
λλ I
+ (other terms) . (6.56)
MATTER-RADIATION INTERACTION (QUANTUM) 255
The two separate traces appearing in this equation can be easily identified in terms
of physical observables. The trace over (A) is nothing but the density-matrix
element ρm m (t ) (see Eqs. (6.49), (6.50) and (6.51)), while from Eq. (4.37) we
have1
&, - ' 2
Tr(R) a† (ν, Ω, λ) + δ ρ(R) (t ) = c
λ ) a(ν, Ω, +δ ,
Iλ λ (ν, Ω)
λλ I 3 t λλ
hν
is the polarization tensor of the radiation of frequency ν prop-
where [Iλ λ (ν, Ω)]t
agating in the direction Ω at time t .
Now we have to perform the integral in dt , between 0 and t, of Eq. (6.56).
A possible way to evaluate the integral is to assume that the relaxation time tr
characterizing the evolution of the physical quantities of the coupled system is
much longer than the typical time-scale over which the oscillating factor
2πi (νm n − ν)(t−t )
e
varies. Under this assumption we can extract the quantities ρm m and Iλ λ from
the integral (and drop their explicit time dependence) to get
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt =
h
0
8π 3 c2
=−
d · eλ (Ω) ∗
d · eλ (Ω) ρm m
h2 ν 2 V mn nm
ν Ωλλ nm
3 t
+ hν δ
× Iλ λ (ν, Ω) e
2πi (νm n − ν)(t−t )
dt
c2 λλ
0
+ (other terms) ,
where the expression of c2ν has been used (see Eqs. (6.36)).
The last integral can be evaluated by standard methods; moreover, because of
the above assumption, it is natural to take its asymptotic value for t → ∞. We
have
t
2πi (νab − ν)(t−t ) 1 i 1
lim e dt = δ(νab − ν) + P ,
t→∞ 2 2π νab − ν
0
where δ(x) is the Dirac delta-function and P(x) means the principal part in the
sense of distribution theory.
If we define the complex profile Φ(ν0 − ν) to be
i 1
Φ(ν0 − ν) = φ(ν0 − ν) + i ψ(ν0 − ν) = δ(ν0 − ν) + P , (6.57)
π ν0 − ν
1
It follows from Eq. (6.22) that the operator a† (ν, Ω,
λ ) a(ν, Ω,
λ) has the same form in the
Schrödinger picture and in the interaction picture.
256 CHAPTER 6
and transform the summation over the various modes of the radiation field into a
double integral via Eq. (4.35), we obtain
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt =
h
0
5 ∞
16 π
4
dΩ
∗
=− 2 ρm m dν d · eλ (Ω)
mn
d · eλ (Ω) nm
h c
4π
λλ nm 0
+ hν 3
× Iλ λ (ν, Ω) δ Φ(νm n − ν)
c2 λλ
+ (other terms) . (6.58)
This procedure can however be criticized from several points of view. First of
all, by assuming ρm m (t ) to be independent of t , we give up the possibility of
describing those aspects of the phenomena which depend on the ‘history’ of the
density-matrix elements. This is the so-called secular approximation (cf. Cohen-
Tannoudji et al., 1988), a special case of the Markov approximation, which makes
impossible the treatment of correlation effects in successive interactions, e.g., fre-
quency redistribution effects in the absorption and re-emission process.
Furthermore, we should realize that the exact time evolution of the density-
matrix element ρmm is affected by several complicated phenomena that cannot be
accounted for by our approach, which is based on a lowest-order expansion in the
framework of Quantum Electrodynamics: the finite width of energy levels, their
energy shift due to interactions with real and virtual photons, and – when the
levels m and m are non-degenerate – the so-called quantum beats. The last phe-
nomenon consists in oscillations of ρmm whose frequencies depend on the spectral
characteristics of the incident radiation.
The first two phenomena (finite width and shift of the energy levels) can be
taken into account, in a phenomenological way, by replacing the complex profile of
Eq. (6.57) with the following
Φ(νab − ν) = φ(νab − ν) + i ψ(νab − ν)
1 Γab i νab + ∆ab − ν
= + , (6.59a)
2 + (ν
π Γab ab + ∆ab − ν)2 π Γ 2 + (ν
ab ab + ∆ab − ν)
2
where
γab γ + γb
Γab = = a , ∆ab = ∆a − ∆b , (6.59b)
4π 4π
with γa , γb the probabilities per unit time that the atom leaves level | a or | b ,
respectively, via spontaneous or stimulated transitions, and ∆a , ∆b the frequency
shifts of the two levels. Note that the real part of the profile satisfies the normal-
ization
∞
φ(νab − ν) dν = 1 . (6.59c)
−∞
MATTER-RADIATION INTERACTION (QUANTUM) 257
The existence of quantum beats affects Eq. (6.58) in such a way that, when off-
diagonal density-matrix elements connecting non-degenerate levels are present, the
central frequency νm n of the Φ profile is in fact an oscillating quantity. Therefore,
the equations obtained from our simple treatment remain valid when the precise
frequency position of the Φ profile is not crucial. This requires that the incident
radiation field should be flat, i.e. independent of frequency, across a spectral in-
terval ∆ν larger than the Bohr frequencies connecting the different m levels and,
moreover, larger than the inverse lifetime of each m level. Under this assumption
(flat-spectrum approximation), the integral over frequency of the first term in the
square bracket of Eq. (6.58) is seen to be proportional to
,
Iλ λ (νm∗ n , Ω)
where m∗ stands equivalently for m, m , or m since Iλ λ (νmn , Ω)
= I (ν , Ω)
λ λ mn
=
Iλ λ (νm n , Ω).
Two special cases are worth to be treated separately because, in principle, they
do not require the introduction of the flat-spectrum approximation. These are the
cases where:
i) all the m levels involved are degenerate;
ii) all the off-diagonal density-matrix elements are zero.
In both cases quantum beats are ineffective, and the first term in the square bracket
of Eq. (6.58) leads to an integral of the form
∞
Φ(ν − ν) dν .
Iλ λ (ν, Ω) mn
0
In order to describe both the general case and these two special cases by a unique
equation, we write the result of the integration over frequency in the form
,
Iλ λ (νmn , Ω)
where the ‘mean’ Bohr frequency νmn is given by
1
νmn = νmn + νm n + νm n . (6.60)
3
Obviously, the introduction of the ‘mean’ Bohr frequency is just a formal artifice.
In the general case (non-zero density-matrix elements connecting non-degenerate
levels), νmn is a ‘label’ which reminds that the radiation field at any of the Bohr
frequencies νmn , νm n , νm n should be considered; in the special cases i) and ii),
νmn reduces to the true Bohr frequency νmn .
Let’s now consider the second term in the square bracket of Eq. (6.58). This
term leads to an integral over frequency whose real part is easily evaluated. Taking
258 CHAPTER 6
into account that the real part of the Φ profile is practically a Dirac delta-function
centered at the frequency νmn , we obtain
⎧
⎨ h2 νmn
3
if νmn > 0
c
⎩
0 if νmn < 0 .
The imaginary part of the Φ profile leads, on the contrary, to a diverging integral.
Integrals of such kind are commonly encountered in Quantum Electrodynamics, and
describe the finite widths and the shifts of the energy levels due to the interaction of
the atom with virtual photons. The actual calculation of the shift of a given level
(Lamb shift) involves the removal of the divergence of the integral by standard
procedures based on the concept of mass renormalization. An example of such
calculation for the simple case of the hydrogen atom can be found in Bethe and
Salpeter (1957). In the following, these terms related to the Lamb shift will be
simply neglected in our equations. Obviously, they can be reintroduced a posteriori
in the theory by supposing that the eigenvalues En of the atomic Hamiltonian
represent the exact energies (including the Lamb shift) of the atomic levels.
The above reasoning allows Eq. (6.58) to be rewritten in the form
t !
4π 2
− 2 Tr ÔI (t), B(t) , B † (t ) ρI (t ) dt =
h
0
5
16 π 4 dΩ
∗
=− 2 ρm m
d · eλ (Ω) mn
d · eλ (Ω) nm
h c
4π
λλ nm
× Iλ λ (νmn , Ω)
16 π 4
− 3
ρm m 3
νmn Θ(νmn )
hc
λ m n
5
dΩ
∗
×
d · eλ (Ω) mn
d · eλ (Ω) nm
4π
+ (other terms) , (6.61)
where Θ(x) is the step-function, which is equal to 1 for positive values of the
argument and to 0 otherwise.
The ‘(other terms)’ appearing in this equation are those arising from the sec-
ond, third, and fourth line of Eq. (6.54). The evaluation of these terms requires
calculations quite similar to those developed above. Once these terms have been
evaluated (by introducing the relevant ‘mean’ Bohr frequencies as in Eq. (6.60)),
we substitute Eqs. (6.61) and (6.52) into Eq. (6.48) and we add the term result-
ing from the complex conjugate of Eq. (6.61) with the interchange m ← → m (this
†
interchange transforms the operator ÔI (t) defined in Eq. (6.51) into ÔI (t)).
After some algebra, which involves several operations of index renaming, we
obtain the evolution equations for the density-matrix elements ρmm (usually called
MATTER-RADIATION INTERACTION (QUANTUM) 259
the statistical equilibrium equations) in the form1
d
ρ = −2π i νmm ρmm
dt mm
+ ρnn TA (m, m , n, n )
nn
+ ρpp TE (m, m , p, p )
pp
+ ρpp TS (m, m , p, p )
pp
− ρmm RA (m, m , m ) + ρm m RA (m , m , m)
m
− ρmm RE (m , m, m ) + ρm m RE (m, m , m )
m
− ρmm RS (m , m, m ) + ρm m RS (m, m , m ) , (6.62)
m
where
5
32π 4 dΩ
TA (m, m , n, n ) = 2
d · eλ (Ω) mn
∗
d · eλ (Ω) nm
h c
4π
λλ
× Iλ λ (νmn , Ω)
5
32π 4 dΩ
TE (m, m , p, p ) = 3 p m λ
∗
d · e (Ω)
d · eλ (Ω) mp
hc 4π
λ
× νpm
3
Θ(νpm )
5
32π 4 dΩ
TS (m, m , p, p ) = 2 p m λ
∗
d · e (Ω)
d · eλ (Ω) mp
h c
4π
λλ
× Iλ λ (νpm , Ω)
16π 4 5 dΩ
RA (m, m , m ) = 2
d · eλ (Ω)
pm d · λ
∗
e (Ω)
m p
p
h c
4π
λλ
× Iλ λ (νpm , Ω)
1
It should be recalled that from Eqs. (1.37), (3.80), and (6.37) one has the relations
∗ = I (ν, Ω)
Iλλ (ν, Ω) , ρ∗ab = ρba , ∗ = [ d · e (Ω)
[ d · eλ (Ω)] ∗] .
λ λ ab λ ba
260 CHAPTER 6
16π 4 5 dΩ
∗
RE (m, m , m ) = 3
d · eλ (Ω)
mn
d · eλ (Ω) nm
n
hc 4π
λ
× νmn
3
Θ(νmn )
16π 4 5 dΩ
RS (m, m , m ) = 2
d · eλ (Ω)
mn
∗
d · eλ (Ω) nm
n
h c
4π
λλ
,
× Iλ λ (νmn , Ω) (6.63)
with
1
νrs = ν + νrs + νr s + νr s
4 rs
1
νrs = νrs + νrs + νrs
3
1
νrs = νrs + νr s + νr s . (6.64)
3
The physical meaning of the various terms in Eq. (6.62) can be easily understood.
The first term in the right-hand side describes a relaxation of the coherence ρmm
due to the energy difference between levels | m and | m . This term,which is
obviously zero for the coherences between degenerate levels (and in particular for
the diagonal elements ρmm representing the populations), produces in general a
relaxation of the coherence ρmm which is the stronger, the larger the energy sepa-
ration between the corresponding levels. Roughly speaking, the coherence between
levels | m and | m turns out to be practically zero when the corresponding Bohr
frequency 2π νmm is much larger than the typical rates T and R appearing in the
right-hand side of Eq. (6.62). As these rates are usually of order 109 s−1 or smaller,
we obtain the following criterion for the disappearance of coherences between levels
| m and | m
| νmm | 1.6 × 108 Hz
or, in other words
| Em − Em | 7 × 10−7 eV . (6.65)
The other terms in the right-hand side of Eq. (6.62) represent the rates at which
the coherence ρmm either increases as a result of coherence-transfer from different
levels or decreases as a result of coherence-relaxation to different levels. Coherence-
transfer rates are denoted by the symbol T and all bear a positive sign. These terms
are due to absorption from lower levels (TA ), to spontaneous emission from upper
levels (TE ), and to stimulated emission from upper levels (TS ). Similarly, coherence-
relaxation rates – denoted by the symbol R – are due to absorption towards upper
levels (RA ), to spontaneous emission towards lower levels (RE ), and to stimulated
emission towards lower levels (RS ). All these rates bear a negative sign. The rates
RS and RE originate from the first line of Eq. (6.54); the rate TA from the second,
MATTER-RADIATION INTERACTION (QUANTUM) 261
Fig.6.1. The radiative processes contributing to the time evolution of the coherence ρmm are
schematized in this simplified Grotrian diagram. Straight lines refer to spontaneous emission
while wavy lines refer to absorption (arrows pointing upward) or stimulated emission (arrows
pointing downward). Coherence-transfer rates are represented by transitions ending in the couple
of levels (m, m ) while coherence-relaxation rates are represented by transitions originating from
these levels. Each process is labelled with the symbol of the corresponding rate.
the rates TS and TE from the third, and the rate RA from the fourth line. The
processes described by the different rates are schematically shown in Fig. 6.1.
From the explicit expressions of the various rates (Eqs. (6.63)) it is possible
to prove some important properties. With easy transformations, one obtains the
conjugation relations
TA (m, m , n, n )∗ = TA (m , m, n , n)
TE (m, m , p, p )∗ = TE (m , m, p , p)
TS (m, m , p, p )∗ = TS (m , m, p , p)
RA (m, m , m )∗ = RA (m, m , m )
RE (m, m , m )∗ = RE (m , m , m)
RS (m, m , m )∗ = RS (m , m , m) .
Another property concerns the time evolution of the trace of the density operator.
Setting m = m in Eq. (6.62) and summing over m one finds, after some algebra
which involves several operations of index renaming, that the term arising from TA
exactly cancels out with the terms arising from RA , and, similarly, TE cancels out
with RE , and TS cancels out with RS , so that
d
ρmm =0.
dt m
This property – which should be expected on the basis of Eq. (3.83) – implies that
the trace of the density operator of the atomic system can be fixed once and for
262 CHAPTER 6
all. In many cases the density operator is normalized through the equation
ρmm = 1 ,
m
but different normalizations can be safely employed. Equation (6.62), being lin-
ear in the density-matrix elements, is independent of the particular normalization
condition employed.
The expressions for the various rates given in Eqs. (6.63) contain matrix elements
e (Ω)]
of the form [d·
λ mn . The evaluation of these matrix elements is generally easier
using the spherical components of the dipole operator (rather than its Cartesian
components).1 From Eq. (2.83) we have
d · eλ (Ω)
=
(−1)q d−q eλ (Ω) q
,
q
whence
d · eλ (Ω)
mn
=
(−1)q (d−q )mn eλ (Ω) q
(6.66)
q
and, similarly
∗
∗
d · eλ (Ω) ∗ =
= d · eλ (Ω)
(−1)q (d−q )nm eλ (Ω) .
mn nm q
q
To shorten notations we introduce the symbols
∗ ∗
(d−q )∗nm ≡ (d−q )nm , ∗ ≡ e (Ω)
eλ (Ω) ,
q λ q
so that
∗
d · eλ (Ω) = ∗.
(−1)q (d−q )∗nm eλ (Ω) (6.67)
mn q
q
which up to
At this point we choose a specific form for the unit vectors eλ (Ω),
now were just the unit vectors – satisfying the relations (4.31) – used to expand
the vector potential. From now on we require that these vectors have the form of
Eq. (5.141). This choice provides the simplest expression for their spherical com-
ponents and, moreover, it allows us to use the spherical tensors already introduced
in Sect. 5.11.
As a first consequence, we can easily evaluate the angular integrals appearing
in the expressions for the rates TE and RE related to spontaneous emission (see
Eqs. (6.63)). From Eqs. (6.66), (6.67), (5.140) and (5.143) we have
1
This is basically related to the Wigner-Eckart theorem, see Sect. 2.8.
MATTER-RADIATION INTERACTION (QUANTUM) 263
5 dΩ
d · eλ (Ω)
∗
d · eλ (Ω) =
4π ab cd
λ
5 dΩ
=
(−1)q+q (d−q )ab (d−q )∗dc eα (Ω) ∗
eα (Ω)
4π q q
α=±1 qq
5
dΩ
= (−1)q+q (d−q )ab (d−q )∗dc
Eqq (α, α, Ω)
α=±1
4π
qq
2
= (d−q )ab (d−q )∗dc . (6.68)
3 q
Next, recalling the definition of the radiation field tensor Jqq (ν) (see Eqs. (5.150)
and (5.153)), we can rewrite the rates appearing in the statistical equilibrium equa-
tions in the more compact form
32π 4
TA (m, m , n, n ) = 2 (−1)q+q (d−q )mn (d−q )∗m n Jqq (νmn )
h c qq
64π 4
TE (m, m , p, p ) = (d−q )p m (d−q )∗pm νpm
3
Θ(νpm )
3hc3 q
32π 4
TS (m, m , p, p ) = (−1)q+q (d−q )p m (d−q )∗pm Jqq (νpm )
h2 c qq
16π 4
RA (m, m , m ) = 2c
(−1)q+q (d−q )pm (d−q )∗pm Jqq (νpm )
p
h qq
32π 4
RE (m, m , m ) = 3
(d−q )mn (d−q )∗m n νmn
3
Θ(νmn )
n
3hc q
16π 4
RS (m, m , m ) = 2
(−1)q+q (d−q )mn (d−q )∗m n Jqq (νmn ) , (6.69)
n
h c qq
where the various ‘mean’ Bohr frequencies are defined in Eqs. (6.64).
An important remark should be made about the limit of the statistical equilib-
rium equations under the hypothesis of neglecting polarization phenomena in the
atomic system and of neglecting polarization and anisotropy effects in the radiation
field. Let us assume that
ρab = ρa δab , = 1 I(ν) δ ,
Iλ λ (ν, Ω) λλ
2
where ρa is the population of level | a and I(ν) is the radiation field intensity.
From Eq. (6.62) we have
d
ρm = ρn TA (m, m, n, n) + ρp TE (m, m, p, p) + TS (m, m, p, p)
dt n p
− 2 ρm RA (m, m, m) + RE (m, m, m) + RS (m, m, m) .
264 CHAPTER 6
From Eqs. (5.150), (5.153) and (5.143) the radiation field tensor is found to be
1
Jqq (ν) = I(ν) δqq .
3
According to Eqs. (6.69) the coherence-transfer rate TA becomes
32π 4 2
TA (m, m, n, n) = |d | I(νmn ) ,
3h2 c mn
where (see Eq. (2.82))
|dmn |2 = |dnm |2 = dmn · dnm = |(dq )mn |2 . (6.70)
q
Introducing the Einstein Bnm coefficient for the transition from the initial state
| n to the final state | m by the standard definition
32 π 4 2
Bnm = |d | , (6.71)
3 h2 c nm
one obtains
TA (m, m, n, n) = Bnm I(νmn ) .
Performing the same kind of calculations on the other rates one finally gets
d
ρm = ρn Bnm I(νmn )
dt n
+ ρp Apm + Bpm I(νpm )
p
− ρm Bmp I(νpm )
p
− ρm Amn + Bmn I(νmn ) , (6.72)
n
where
64π 4 3 2 3
2hνmn
Amn = νmn |dmn | = Bmn (6.73)
3hc3 c2
is the Einstein coefficient for spontaneous emission from the upper level | m to
the lower level | n .
Equations (6.72) are the standard statistical equilibrium equations whose deriva-
tion, based on heuristic arguments, can be found in various textbooks (see e.g.
Mihalas, 1978). They are just a special case of a more general set of equations
(Eqs. (6.62)) which have been derived here directly from the principles of Quan-
tum Electrodynamics. Equations (6.62) reduce to Eqs. (6.72) when polarization
and anisotropy phenomena are neglected.
MATTER-RADIATION INTERACTION (QUANTUM) 265
6.6. Evolution Equations for the Radiation Field
We derive here the evolution equations for the radiation field by essentially the same
method just employed for the atomic system. In particular, we are interested in the
related to a given frequency ν
time evolution of the polarization tensor Iαβ (ν, Ω)
whose associated quantum operator is given by Eq. (4.37).
and a given direction Ω,
As this operator remains unchanged in the interaction picture (see footnote on
p. 255), we have just to apply Eq. (6.48) with the operator ÔI (t) given by
hν 3 † β) .
ÔI (t) = a (ν, Ω, α) a(ν, Ω,
c2
Since dÔI (t)/dt = 0, and since the adjoint operator ÔI† (t) is obtained from ÔI (t)
by exchange of the indices α and β, we get
d =
I (ν, Ω)
dt αβ
t !
4π 2 ν 3 †
=− Tr a (ν, Ω, β) , B(t) , B † (t ) ρ (t ) dt
α) a(ν, Ω,
I
hc2
0
6 7
+ c.c. α →
←β . (6.74)
We start by evaluating the inner commutator, with B(t) given by Eq. (6.35). We
have
†
α) a(ν, Ω,
a (ν, Ω, β) , B(t) =
= −i )
cν d · eλ (Ω mn
νΩ
λ mn
× |m n | a† (ν, Ω, , λ ) e 2πi (νmn − ν ) t ,
β) , a(ν , Ω
α) a(ν, Ω,
and using the commutation properties of the operators a and a† (Eq. (4.32)) we
get
†
α) a(ν, Ω,
a (ν, Ω, β) , B(t) = i c
ν d · eα (Ω)
mn
mn
× |m β) e 2πi (νmn − ν) t .
n | a(ν, Ω,
Thus the double commutator can be written as
†
β) , B(t) , B † (t ) =
α) a(ν, Ω,
a (ν, Ω,
=− cν cν d · eα (Ω)
)∗
d · eλ (Ω
mn rs
νΩ
λ mn rs
2πi (νmn − ν) t + 2πi (νrs + ν ) t
×e C, (6.75)
266 CHAPTER 6
where C is the commutator
C = |m β) , | r
n | a(ν, Ω, s | a† (ν , Ω
, λ ) .
With some easy algebra, and rearranging the final result in such a way that a† -
operators precede a-operators, we obtain
, -
C = δnr | m s | a† (ν , Ω
, λ ) a(ν, Ω,
β) + δ δ δ
νν ΩΩ λ β
− δsm | r n | a† (ν , Ω
, λ ) a(ν, Ω,
β) . (6.76)
Now we substitute Eq. (6.76) into Eq. (6.75), multiply by ρI (t ) and take the trace
of the resulting operator. With the help of Eqs. (6.44) and (6.47) we obtain
!
†
†
Tr
a (ν, Ω, α) a(ν, Ω, β) , B(t) , B (t ) ρI (t ) =
2πi (νmn − ν)(t−t )
=− c2ν d · eα (Ω)
d ∗
· e (Ω) e
mn λ ns
λ mns
& 2πi νms t
'
× Tr(A) | m s| e ρ(A)
I (t )
& (R) '
× Tr(R) a† (ν, Ω,
λ) a(ν, Ω,
β) + δ
λβ ρI (t )
2πi (νmn − ν)(t−t )
+ c2ν d · eα (Ω)
∗
d · eλ (Ω) e
mn rm
λ mnr
& 2πi νrn t
'
× Tr(A) | r n| e ρ(A)
I (t )
& '
× Tr(R) a† (ν, Ω, β) ρ(R) (t ) .
λ) a(ν, Ω,
I
Similarly to the previous section, we identify the various traces in this equation
with the corresponding physical observables. Next we perform the integral in dt
between 0 and t (for t → ∞) and we substitute into Eq. (6.74) to obtain, after
some index renaming
d =
I (ν, Ω)
dt αβ
3
4π ν
= d · eα (Ω)
∗
d · eγ (Ω) ρm m
hV γ mn nm
mm n
× Φ(νmn − ν) Iγβ (ν, Ω)
4π 3 ν 4
∗
+
d · eα (Ω) d · eβ (Ω) nm
ρm m Φ(νmn − ν) +
c V
2
mn
mm n
MATTER-RADIATION INTERACTION (QUANTUM) 267
4π 3 ν
∗ ρ
−
d · eα (Ω) mn
d · eγ (Ω) n m nn
hV γ
mnn
!
× Φ(νmn − ν) Iγβ (ν, Ω)
6 7
→β ,
+ c.c. α ← (6.77)
where the complex profile Φ is given by Eq. (6.57).
However, the above derivation is not fully justified, because it neglects the fi-
nite widths and the shifts of the atomic energy levels due to interactions with real
and virtual photons, and quantum beats (see the analogous discussion following
Eq. (6.58)). The first two phenomena can be taken into account, in a phenomeno-
logical way, by assuming the Φ profile in Eq. (6.77) to have the form of Eq. (6.59a)
rather than Eq. (6.57). Even with this substitution, the validity of Eq. (6.77) is
subjected to a basic restriction related to the existence of quantum beats: in order
that our formalism can be consistent, the density-matrix elements appearing in
the equation must be deduced from the statistical equilibrium equations under the
secular approximation, which implies the flat-spectrum approximation illustrated
in the preceding section. Obviously, the formalism presented here cannot describe
frequency-redistribution phenomena.
Equation (6.77) must be interpreted as an evolution equation for the polarization
tensor in the phase space of the photon occupation numbers. To express it as an
evolution equation in the ordinary three-dimensional space, we attach an explicit
time and space dependence to the polarization tensor. In other words, we replace
by I (ν, Ω
Iαβ (ν, Ω) ; t, x ) and we regard it as the polarization tensor that is
αβ
propagating at time t through the point P of coordinate x in the direction Ω, at
frequency ν. The time derivative calculated in Eq. (6.77) is nothing but
; t + ∆t, x + Ω
Iαβ (ν, Ω c ∆t) − I (ν, Ω ; t, x )
αβ
lim =
∆t→0 ∆t
∂ ; t, x ) + c Ω
· grad I (ν, Ω
; t, x ) .
= I (ν, Ω
∂t αβ αβ
we can thus replace the
Denoting by s the spatial coordinate along the direction Ω,
time derivative in Eq. (6.77) with
d ∂ d
→ +c .
dt ∂t ds
Next we observe that Eq. (6.77) has been obtained under the assumption that the
radiation field, defined in the normalization box of volume V, interacts with a single
atomic system. For an ensemble of uncorrelated atomic systems we have just to
multiply the right-hand side by the number of such systems, with the result that
the factor 1/V is replaced by the number density N . Equation (6.77) can then be
268 CHAPTER 6
cast into the form of a radiative transfer equation
1 ∂ d =− + g ∗ I (ν, Ω)
+ Iαβ (ν, Ω) gαγ Iγβ (ν, Ω) βγ αγ
c ∂t ds γ
+ + h∗ I (ν, Ω)
hαγ Iγβ (ν, Ω)
βγ αγ
γ
1 ∗
+ f + fβα , (6.78)
2 αβ
where
4π 3 ν
gαβ = N d · eα (Ω)
∗ ρ Φ(ν − ν)
d · eβ (Ω)
ch mn n m nn mn
mnn
3
4π ν
hαβ = N d · eα (Ω)
∗
d · eβ (Ω) ρm m Φ(νmn − ν)
ch mn nm
mm n
3
2hν
fαβ = hαβ . (6.79)
c2
The physical interpretation of the terms in the right-hand side of Eq. (6.78) is
straightforward. The first term, which bears a negative sign and which results
from the third line in Eq. (6.77), describes the transfer effects produced by the
absorption of radiation. It is proportional to the coherence ρnn between two levels
| n and | n which are connected to the upper level | m by dipole transitions.
The imaginary part of the complex Φ profile describes the associated anomalous
dispersion effects.
In Chap. 5 we already found – using the formalism of classical electrodynamics –
a transfer equation for the polarization tensor (cf. Eq. (5.24)).1 Comparison with
Eq. (6.78) shows that the tensor g is just the quantum analogue of the propagation
tensor G defined in Eq. (5.22) (with G → g ∗ ). This analogy provides an interesting
relation between the refractive index and the matrix elements of the dipole operator
between eigenstates of the atomic system.
The second term in Eq. (6.78), which bears a positive sign and which results from
the first line in Eq. (6.77), describes the transfer effects produced by stimulated
emission of radiation. This term is proportional to the coherence ρm m between
two levels | m and | m which are connected to the lower level | n by dipole
transitions. We see the appearance, via the imaginary part of the complex Φ profile,
of anomalous dispersion effects associated with stimulated emission of radiation.
This term has no analogue in the classical equation, an obvious consequence of the
quantum nature of stimulated emission.
Finally, the third term in Eq. (6.78), which again bears a positive sign and which
results from the second line in Eq. (6.77), describes the transfer effects produced
1
Equation (5.24) involves the tensor J, while Eq. (6.78) involves the tensor I. We recall that
I and J are proportional (see Eq. (1.36)).
MATTER-RADIATION INTERACTION (QUANTUM) 269
by spontaneous emission of radiation. The classical analogue of this term has been
deduced for the special case of an atom in the presence of a magnetic field (see
Eq. (3.37)).
As in the preceding section, it is convenient to express the dipole matrix elements
appearing in Eqs. (6.79) in terms of spherical tensors, assuming at the same time
that the polarization unit vectors have the form of Eq. (5.141). Using Eqs. (6.66),
(6.67), and (5.140) we can easily rewrite the tensors g, h, f in the form
4π 3 ν
gαβ = N (−1)q+q (d−q )mn (d−q )∗mn
ch
mnn qq
ρ Φ(ν − ν)
× Eqq (α, β, Ω) nn mn
4π 3 ν
hαβ = N (−1)q+q (d−q )mn (d−q )∗m n
ch
mm n qq
ρ Φ(ν − ν)
× Eqq (α, β, Ω) mm mn
2hν 3
fαβ = hαβ . (6.80)
c2
An important remark on the radiative transfer equations concerns the special
case where polarization is neglected both in the radiation field and in the atomic
system. If we suppose that
= 1 δ ,
Iαβ (ν, Ω) I(ν, Ω) αβ ρnn = ρn δnn ,
2
we get
1 ∂ d = 1 ∂ + d
+ I(ν, Ω) I (ν, Ω)
c ∂t ds c ∂t ds α αα
1
∗
=− gαα + gαα I(ν, Ω)
α
2
1
+ hαα + h∗αα I(ν, Ω)
α
2
1
∗
+ fαα + fαα , (6.81)
α
2
where
4π 3 ν
gαα = N (−1)q+q (d−q )mn (d−q )∗mn
α
ch mn α qq
ρ Φ(ν − ν) ,
× Eqq (α, α, Ω) n mn
270 CHAPTER 6
etc. Now we ‘average over all the possible orientations of the atom’, which is
directions entering
equivalent to keeping the atom fixed and averaging over all the Ω
the tensor Eqq . Using Eq. (6.68) we obtain1
8π 3 ν
gαα = N |dmn |2 ρn Φ(νmn − ν) , (6.82)
α
3ch mn
where |dmn |2 is defined in Eq. (6.70). Performing similar calculations for the other
terms of Eq. (6.81) and bearing in mind the definition of the Einstein coefficients
(Eqs. (6.71) and (6.73)) we easily obtain
1 ∂ d +ε ,
= −k I(ν, Ω)
+ I(ν, Ω) ν ν
c ∂t ds
where2
hν
kν = B ρ − ρm φ(νmn − ν) N
4π nm nm n
hν
εν = A ρ φ(νmn − ν) N .
4π nm mn m
This is the usual radiative transfer equation whose derivation, generally based on
heuristic arguments, can be found in several textbooks (see for instance Mihalas,
1978). The quantities kν and εν are the line absorption coefficient (corrected for
stimulated emission) and the emission coefficient, respectively. The special case
now considered shows that the tensors g, h, and f defined in Eqs. (6.79) are just the
generalization of the absorption coefficient, the negative absorption coefficient due
to stimulated emission, and the emission coefficient, respectively. These quantities,
and the structure itself of the transfer equations, have been derived in this section
directly from the principles of Quantum Electrodynamics.
6.7. Evolution Equations for the Stokes Parameters
Equations (6.78) describe the radiative transfer of polarized radiation in terms of
the polarization tensor Iαβ (ν, Ω). To turn them into transfer equations for the
1
Equation (6.82) is correct only when the atomic system is evenly populated in all the
degenerate sublevels of a given energy level. Even in that case, it does not hold in the presence
of any kind of anisotropy (e.g., in the presence of a magnetic field).
2
The above expression for εν is obtained by noticing that, because of the presence of the
profile φ(νmn − ν), the Einstein coefficient Amn can also be written as
2hν 3
Amn = Bmn .
c2
MATTER-RADIATION INTERACTION (QUANTUM) 271
we have just to fix the form of the polarization unit
Stokes parameters Si (ν, Ω)
vectors. If we choose the unit vectors of Eqs. (5.141) – so that expressions (6.80)
can be used for the tensors f , g, h – the polarization tensor is connected with the
Stokes parameters by the transformation (5.129).1 Thus we can write
1 ∂ d = 1 ∂ d (α, β = ± 1)
+ Si (ν, Ω) (σ̂i )αβ + I (ν, Ω)
c ∂t ds c ∂t ds βα (i = 0, 1, 2, 3) ,
αβ
where σ̂i are the Pauli spin matrices of Eqs. (5.128). Substituting Eqs. (6.78), and
using the conjugation relation (5.147) and the cyclic property of the trace, one
obtains after some algebra
3 3
1 ∂ d =− A + S +ε ,
+ Si (ν, Ω) Kij Sj (ν, Ω) Kij Sj (ν, Ω) i (6.83)
c ∂t ds j=0 j=0
where2
A
Kij = Re Tr (σ̂j σ̂i g)
S
Kij = Re Tr (σ̂j σ̂i h)
εi = Re Tr (σ̂i f ) . (6.84)
This equation can be rewritten in the explicit matrix form
⎛ ⎞ ⎛ A A A
⎞⎛ ⎞
I ηI ηQ ηU ηVA I
⎜ ⎟ ⎜
∂ d ⎜Q⎟ A
⎜ ηQ ηIA ρAV A ⎟⎜
−ρU ⎟⎜ Q ⎟ ⎟
c + ⎜ ⎟ = −⎜ A ⎟⎜ ⎟
∂t ds ⎝ U ⎠ ⎝ ηU −ρAV ηIA ρQ ⎠ ⎝ U ⎠
A
V ηVA ρAU −ρAQ ηIA V
⎛ ⎞⎛ ⎞ ⎛ ⎞
ηIS ηQS S
ηU ηVS I εI
⎜ ηS ηIS ρSV −ρU ⎟⎜ Q ⎟ ⎜ εQ ⎟
S ⎟⎜ ⎟ ⎜
⎜ ⎟
+⎜ Q ⎟⎜ ⎟ + ⎜ ⎟ , (6.85)
⎝ ηUS
−ρSV ηIS ρSQ ⎠⎝ U ⎠ ⎝ εU ⎠
ηVS ρSU −ρSQ ηIS V εV
where
ηIA = K00
A A
= K11 A
= K22 = K33A
= Re Tr (g) = Re g++ + g−−
A
ηQ A
= K01 A
= K10 = Re Tr (σ̂1 g) = −Re g+− + g−+
1
Note that a similar transformation on the transfer equations has been performed in Sect. 5.2,
although using a different basis of polarization unit vectors.
2
It is understood that the components of the tensors f, g, h are labelled in the same way as
the spin matrices σ̂i (see the remark following Eq. (5.128)).
272 CHAPTER 6
ηUA A
= K02 = K20A
= Re Tr (σ̂2 g) = Im g+− − g−+
ηVA = K03
A
= K30A
= Re Tr (σ̂3 g) = Re g++ − g−−
ρAQ = K23
A
= −K32 A
= Im Tr (σ̂1 g) = −Im g+− + g−+
ρAU = K31
A
= −K13 A
= Im Tr (σ̂2 g) = −Re g+− − g−+
ρAV = K12
A
= −K21 A
= Im Tr (σ̂3 g) = Im g++ − g−−
εI = ε0 = Re Tr (f ) = Re f++ + f−−
εQ = ε1 = Re Tr (σ̂1 f ) = −Re f+− + f−+
εU = ε2 = Re Tr (σ̂2 f ) = Im f+− − f−+
εV = ε3 = Re Tr (σ̂3 f ) = Re f++ − f−− , (6.86)
and where ηIS , . . . , ρSV have the same expressions as ηIA , . . . , ρAV with hαβ substi-
tuted for gαβ . All these quantities can be easily written in terms of the reducible
spherical tensor Tqq (i, Ω) introduced in Chap. 5 (Eq. (5.146)). Defining the formal
vectors
ηiA = (ηIA , ηQ
A A
, ηU , ηVA ) (i = 0, 1, 2, 3)
ηiS = (ηIS , ηQ
S S
, ηU , ηVS ) (i = 0, 1, 2, 3)
A A A A
ρi = (ρQ , ρU , ρV ) (i = 1, 2, 3)
ρSi = (ρSQ , ρSU , ρSV ) (i = 1, 2, 3) , (6.87)
we have from Eqs. (6.80)
3
= 8π ν N
ηiA (ν, Ω) (−1)q+q
ch
mnn qq
× Re (d−q )mn (d−q )∗mn Tqq (i, Ω)
ρ Φ(ν − ν)
nn mn
8π 3 ν
=
ηiS (ν, Ω) N (−1)q+q
ch
mm n qq
× Re (d−q )mn (d−q )∗m n Tqq (i, Ω)
ρ Φ(ν − ν)
mm mn
6 7
= η A (ν, Ω)
ρAi (ν, Ω) Re → Im
i
6 7
= η S (ν, Ω)
ρSi (ν, Ω) Re → Im
i
3
= 2hν η S (ν, Ω)
εi (ν, Ω) , (6.88)
i
c2
where the symbol {Re → Im} means that ρi is obtained from the corresponding ηi
by substitution of the imaginary part for the real part.
MATTER-RADIATION INTERACTION (QUANTUM) 273
The radiative transfer equations become much simpler in the special case of
Local Thermodynamic Equilibrium, since the density-matrix elements of the atomic
system reduce to the expression (Boltzmann law)
En
−
e kB T
ρnn = δnn ,
Z(T )
where kB is the Boltzmann constant, T the absolute temperature of the medium,
and Z(T ) the partition function. With the use of the conjugation relation (5.148),
we have from Eqs. (6.88)
3
= 8π ν N
ηiA (ν, Ω) (−1)q+q
ch mn qq En
−
kB T
e
× (d−q )mn (d−q )∗mn Tqq (i, Ω)
φ(νmn − ν)
Z(T )
= 8π 3 ν
ηiS (ν, Ω) N (−1)q+q
ch mn qq Em
−
kB T
× (d−q )mn (d−q )∗mn e
Tqq (i, Ω) φ(νmn − ν) ,
Z(T )
and since the profiles φ(νmn − ν) are practically Dirac delta-functions
− hν
= e
ηiS (ν, Ω) kB T
ηiA (ν, Ω)
3 hν
= 2hν e− kB T η A (ν, Ω)
εi (ν, Ω) = η A (ν, Ω)
− η S (ν, Ω)
B , (6.89)
2 i i i P
c
where
2hν 3 1
BP = hν
c2
e −1 kB T
is the Planck function. Thus the radiative transfer equations can be written in the
compact form
1 ∂ d
3
+ Si = − Kij Sj − BP Uj ,
c ∂t ds j=0
where1
A
Kij = Kij − Kij
S
1
Note that the relation between the quantities ρA S
i and ρi analogous to the first of Eqs. (6.89),
hν
= e− kB T ρA (ν, Ω)
ρSi (ν, Ω) ,
i
is only approximately satisfied. Under such approximation we can write
, -
A − khνT
Kij = Kij 1− e B .
274 CHAPTER 6
is the propagation matrix corrected for stimulated emission, and
Uj = δj0
is the unity vector U = (1, 0, 0, 0)† . In matrix form the equation reads
⎛ ⎞ ⎛ ⎞⎛ ⎞
I ηI ηQ ηU ηV I − BP
1 ∂ d ⎜Q⎟ ⎜ ηQ ηI ρV −ρU ⎟⎜ Q ⎟
+ ⎝ ⎠ = −⎝ ⎠⎝ ⎠,
c ∂t ds U ηU −ρV ηI ρQ U
V ηV ρU −ρQ ηI V
with ηI = ηIA − ηIS , ηQ = ηQ
A
− ηQ
S
, etc.
6.8. Magnetic Dipole Transitions
The statistical equilibrium equations and the radiative transfer equations have been
derived in the preceding sections under the assumption of electric-dipole transitions,
but they can be easily rewritten for magnetic-dipole transitions. As shown at the
end of Sect. 6.3, we have just to change the interaction Hamiltonian according to
Eqs. (6.42).
It immediately follows that the basic equations (6.62) and (6.78) remain un-
changed, while the matrix elements in the expressions of the radiative rates (Eqs.
(6.63)) and in the expressions of the tensors f , g, h (Eqs. (6.79)) must be replaced
according to
d · eλ (Ω)
nm
→ × e (Ω)
µ·Ω λ
nm
.
When spherical tensors are introduced, it can be easily seen that the matrix ele-
ments of the form (dq )nm must be replaced by (µq )nm , and the tensor Eqq (α, β, Ω)
by
E (α, β, Ω) = Ω × e (Ω)
∗
× e (Ω)
Ω (α, β = ± 1) ,
qq m.d. α q β q
which can also be written in the form
Eqq (α, β, Ω)
= α β Eqq (α, β, Ω)
m.d.
since the unit vectors eα satisfy the relation
× e (Ω)
Ω = i α e (Ω)
.
α α
defined in Eq. (5.146) must be replaced by
It follows that the tensor Tqq (i, Ω)
Tqq (0, Ω)
= Tqq (0, Ω)
m.d.
Tqq (1, Ω) m.d.
= −Tqq (1, Ω)
Tqq (2, Ω)
= −Tqq (2, Ω)
m.d.
T (3, Ω) .
= Tqq (3, Ω)
qq m.d.
MATTER-RADIATION INTERACTION (QUANTUM) 275
As far as the statistical equilibrium equations are concerned, we have there-
fore that the radiation field tensor Jqq (ν) (defined in Eq. (5.153)) to be used in
Eqs. (6.69) for magnetic-dipole transitions is obtained from the corresponding ten-
sor for electric-dipole transitions by changing the sign of the linear polarization
Stokes parameters Q and U .
As to the radiative transfer equations, it appears from Eqs. (6.80) that the quan-
tities gαβ , hαβ , fαβ must be replaced by
[ gαβ ]m.d. = α β gαβ {d →
µ}
[ hαβ ]m.d. = α β hαβ {d →
µ}
[ fαβ ]m.d. = α β fαβ {d →
µ} (α, β = ± 1) ,
where the symbol in curly bracket means the substitution of the matrix elements
(µq )nm for (dq )nm . If we consider the matrix form (6.85) of the transfer equations,
we have therefore – besides the just mentioned substitution – a sign change of the
A A
quantities ηQ , ηU , ρAQ , ρAU , ηQ
S S
, ηU , ρSQ , ρSU , εQ , εU .
CHAPTER 7
STATISTICAL EQUILIBRIUM EQUATIONS AND RADIATIVE
TRANSFER COEFFICIENTS FOR ATOMIC SYSTEMS
In this chapter we will apply the basic equations derived in Chap. 6 (statistical
equilibrium and radiative transfer equations) to the special case where the material
system interacting with the radiation field is an atom – or, more properly, an
ensemble of uncorrelated identical atoms. We will also assume the atoms to be
embedded in an external, static magnetic field, on account of the great importance
of this topic in astrophysical plasmas. Obviously, the results that will be derived
in this chapter are subjected to the limitations outlined in the introduction to
Chap. 6.
The two sets of equations depend on the atomic system through the quantities:
i) ρnm = n | ρ | m , the matrix elements of the atomic density operator between
energy eigenstates, describing populations (n = m) and coherences (n = m);
ii) νnm = (En − Em )/h, the corresponding Bohr frequencies;
iii) n | d | m (or n | µ | m ), the matrix elements of the electric (or magnetic)
dipole operator between energy eigenstates. In these expressions | n and En are
the eigenvectors and eigenvalues of the Hamiltonian of the unperturbed atom, plus
the Hamiltonian describing the interaction with the external magnetic field.
The quantities | n and En depend on the spectral properties of the atom and
on the intensity of the magnetic field (L-S or different coupling schemes, possi-
ble presence of hyperfine structure, Zeeman or Paschen-Back effect regime, etc.).
Moreover, one should in principle consider the coherences between any couple of
levels | n , | m , which would make the problem extremely involved. However,
depending on the specific atomic structure and magnetic field value, certain coher-
ences can be disregarded. This is related to a general property of the statistical
equilibrium equations, according to which the coherence ρnm is the smaller, the
larger the energy difference between levels | n and | m .
We consider in the following three schematic situations, which cover most cases
of interest for practical applications:
i) atom with no hyperfine structure, Zeeman effect regime, coherences between
magnetic sublevels (M, M ) of each (αJ)-level (neglecting those between different
levels): Sects. 7.1-7.4;
ii) atom with no hyperfine structure, L-S coupling scheme, incomplete Paschen-
Back effect regime, coherences between magnetic sublevels (M, M ) of each (βLSJ)
level and between magnetic sublevels of different levels (βLSJ), (βLSJ ) belonging
to the same term (neglecting coherences between different terms): Sects. 7.5-7.8;
iii) atom with hyperfine structure, incomplete Paschen-Back effect regime (of hy-
perfine structure), coherences between magnetic sublevels (f, f ) of each hyperfine
level (αJIF ) and between magnetic sublevels of different hyperfine levels (αJIF ),
278 CHAPTER 7
(αJIF ) belonging to the same level (neglecting coherences between different lev-
els): Sect. 7.9.
The statistical equilibrium and the radiative transfer equations will be written
both in the standard representation of the atomic density operator (which is easier
to work with, starting from the formalism of Chap. 6) and in the spherical statis-
tical tensor representation, which leads to more compact and physically significant
expressions.
At the end of the chapter we will also study the effect of collisions on the statistical
equilibrium equations, under the assumption of an isotropic distribution of colliding
particles.
7.1. The Multi-Level Atom in the Standard Representation
We consider an atomic system devoid of hyperfine structure, whose energy levels
are characterized by the quantum numbers α and J, where J is the total angular
momentum and α is a set of quantum numbers related to different physical proper-
ties of the energy level (in particular, for an atom described in the Russel-Saunders
coupling scheme, α could represent the set of quantum numbers (β, L, S) which
describe the electronic configuration, the total orbital angular momentum, and the
total electronic spin, respectively).
Assuming that the magnetic field is sufficiently weak for the Zeeman effect regime
to hold (see the discussion in Sects. 3.1 and 3.4), and aligning the z-direction of
the reference system along the magnetic field, we have for the eigenvectors and
eigenvalues of the atomic Hamiltonian
HA |αJM = EαJ + µ0 gαJ B M |αJM ,
where all the symbols have the same meaning as in Sect. 3.1.
The formalism presented in Chap. 6 can now be applied to this particular system.
To begin with the simplest case, we will neglect here all coherences between mag-
netic sublevels that do not belong to the same (αJ)-level. In other words, we sup-
pose that all the atomic density-matrix elements of the form1 αJM | ρ |α J M
are zero except for (αJ) = (α J ),
αJM | ρ |α J M = δαα δJJ αJM | ρ |αJM . (7.1)
This approximation is usually justified when the energy separation between two
different levels – say (αJ) and (α J ) – is sufficiently large that the corresponding
Bohr frequency
E − Eα J
ναJ, α J = αJ (7.2)
h
1
In this chapter we denote by ρ the atomic density operator ρ(A) .
EQUATIONS FOR ATOMIC SYSTEMS 279
is much larger (in absolute value) than the Einstein A coefficient for spontaneous
emission.1 Under such approximation all the different atomic levels are decoupled
and the atomic system is referred to as a multi-level atom. For this system, the
flat-spectrum approximation discussed in Sect. 6.5 requires that the radiation field
incident on the atom should be constant across frequency intervals wider than
the Zeeman separations and than the inverse lifetimes of the relevant magnetic
sublevels. Therefore, it is sufficient to specify the incident radiation field at the
Bohr frequencies (7.2) relating different levels.
We can apply the formalism of Chap. 6 to the multi-level atom in the presence
of a magnetic field by performing the following substitutions
i) for the energy eigenvectors:
| n → |αJM ;
ii) for the corresponding energy eigenvalues:
En → EαJ + µ0 gαJ B M ,
whence
iii) for the Bohr frequencies:
En − Em
νnm = → ναJM, α J M = ναJ, α J + νL (gαJ M − gα J M ) , (7.3)
h
where ναJ,α J is defined in Eq. (7.2) and νL is the Larmor frequency (cf. Eq. (3.10));
iv) for the matrix elements of the spherical components of the dipole operator:
(dq )nm = n | dq | m → (dq )αJM, α J M = αJM | dq |α J M =
J +M+1
√ J J 1
= (−1) 2J + 1 αJ d α J , (7.4)
−M M q
where we have applied the Wigner-Eckart theorem (Eq. (2.96));
v) finally, for the atomic density-matrix elements:
ρnm = n | ρ | m → αJM | ρ |α J M =
= δαα δJJ αJM | ρ |αJM = δαα δJJ ραJ (M, M ) ,
where we have used the assumption (7.1) and we have introduced the notation of
Eq. (3.91a).
1
We will see however in the following (see Sect. 10.17) that the coherences between
different J-levels are fundamental to explain certain polarization phenomena in multiplets even
when ναJ, α J A.
280 CHAPTER 7
By means of substitutions i) to v) the statistical equilibrium equations and the
radiative transfer equations for the multi-level atom can be directly deduced from
the corresponding equations of Chap. 6.
7.1.a Statistical Equilibrium Equations
In Eqs. (6.62) and (6.69) we substitute (see also Fig. 6.1)
m → αJM
n → α J M p → αu Ju Mu
m → αJM
n → α J M p → αu Ju Mu ,
m → αJM
where the indices (for ‘lower’) and u (for ‘upper’) denote any level of energy lower
(or higher, respectively) than EαJ . We obtain
d
ρ (M, M ) = −2πi νL gαJ (M − M ) ραJ (M, M )
dt αJ
+ ρα J (M , M ) TA (αJM M , α J M M )
α J M M
+ ρα J (Mu , Mu ) TE (αJM M , αu Ju Mu Mu )
u u
αu Ju Mu Mu
+ TS (αJM M , αu Ju Mu Mu )
&
− ραJ (M, M ) RA (αJM M ) + RE (αJM M )
M
+ RS (αJM M )
+ ραJ (M , M ) RA (αJM M ) + RE (αJM M )
'
+ RS (αJM M ) , (7.5)
where the various rates (for which we have introduced shorthand notations) are
given by
TA (αJM M , α J M M ) =
32π 4
= 2 (−1)q+q (d−q )αJM, α J M (d−q )∗αJM , α J M Jqq (ναJ, α J )
h c
qq
TE (αJM M , αu Ju Mu Mu ) =
64π 4
= (d−q )α J M , αJM (d−q )∗α J M , αJM να3 J , αJ
3hc3 q u u u u u u u u
EQUATIONS FOR ATOMIC SYSTEMS 281
TS (αJM M , αu Ju Mu Mu ) =
32π 4
= 2 (−1)q+q (d−q )α J M , αJM (d−q )∗α J M , αJM Jqq (να J , αJ )
h c
u u u u u u u u
qq
RA (αJM M ) =
16π 4
= 2
(−1)q+q (d−q )α J M , αJM (d−q )∗α J M , αJM Jqq (να J , αJ )
h c
u u u u u u u u
αu Ju Mu qq
RE (αJM M ) =
32π 4
= (d−q )αJM, α J M (d−q )∗αJM , α J M ναJ,
3
α J
3hc3 q
α J M
RS (αJM M ) =
16π 4
= (−1)q+q (d−q )αJM, α J M (d−q )∗αJM , α J M Jqq (ναJ, α J ) .
h2 c
α J M qq
Because of Eq. (7.4), the expressions for the rates contain the square modulus of
the reduced matrix element of the dipole operator between the (αJ)-level and one
of the lower or upper levels,
| αJ d α J |2 or | αu Ju d αJ |2 .
These quantities can be easily expressed in terms of the Einstein coefficients for
transitions between different levels. Given two levels (α J ) and (αu Ju ), with
Eα J > Eα J , the line strength of the transition is defined by the symmetrical
u u
expression
S(α J , αu Ju ) = S(αu Ju , α J ) = (2J + 1) | α J d αu Ju |2
= (2Ju + 1) | αu Ju d α J |2 , (7.6)
where the last equality follows from Eq. (2.101). The Einstein coefficients for
spontaneous and stimulated emission and for absorption are connected with the
line strength by the relations1
64π 4 3
(2Ju + 1) A(αu Ju → α J ) = ν S(αu Ju , α J )
3hc3 αu Ju , α J
32π 4
(2Ju + 1) B(αu Ju → α J ) = S(αu Ju , α J )
3h2 c
32π 4
(2J + 1) B(α J → αu Ju ) = S(α J , αu Ju ) , (7.7)
3h2 c
1
Note that from Eqs. (6.71) and (6.73) the Einstein coefficients for the transition between
282 CHAPTER 7
with
2h 3
A(αu Ju → α J ) = ν B(αu Ju → α J )
c2 αu Ju , α J
(2J + 1) B(α J → αu Ju ) = (2Ju + 1) B(αu Ju → α J ) . (7.8)
Using these relations, we can rewrite the expressions of the rates in the form
TA (αJM M , α J M M ) = (2J + 1) B(α J → αJ)
J J 1 J J 1
× 3 (−1)M −M
−M M −q −M M −q
× Jqq (ναJ, α J ) (7.9a)
TE (αJM M , αu Ju Mu Mu ) = (2Ju + 1) A(αu Ju → αJ)
Mu −Mu Ju J 1 Ju J 1
× (−1) (7.9b)
−Mu M −q −Mu M −q
q
TS (αJM M , αu Ju Mu Mu ) = (2Ju + 1) B(αu Ju → αJ)
Ju J 1 Ju J 1
× 3 (−1)M −M
−M u M −q −M u M −q
× Jqq (να J , αJ ) (7.9c)
u u
the individual magnetic sublevels (α J M ) and (αu Ju Mu ) are given by
64π 4 3
Aα J Mu , α J M = ν | αu Ju Mu | d |α J M |2
u u 3hc3 αu Ju Mu , α J M
32π 4
Bα J Mu , α J M = | αu Ju Mu | d |α J M |2
u u 3h2 c
32π 4
Bα J M , αu Ju Mu = | α J M | d |αu Ju Mu |2 .
3h2 c
Using Eqs. (6.70), (7.4), (2.23a) and (7.6) we get
%
M
Aα J Mu , α J M = A(αu Ju → α J )
u u
% %
M
Bα J Mu , α J M = B(αu Ju → α J ) , Mu
Bα J M , αu Ju Mu = B(α J → αu Ju )
u u
(the first relation implies indeed the approximation
3 3
να J Mu , α J M = να J , α J ).
u u u u
EQUATIONS FOR ATOMIC SYSTEMS 283
1
RA (αJM M ) = (2J + 1) B(αJ → αu Ju )
2
αu Ju
M−M Ju J 1 Ju J 1
× 3 (−1)
−M u M −q −Mu M −q
qq Mu
× Jqq (να J , αJ ) (7.9d)
u u
1
RE (αJM M ) = (2J + 1) A(αJ → α J )
2
α J
J J 1 J J 1
× (−1)M−M
−M M −q −M M −q
qM
1
= δMM A(αJ → α J ) (7.9e)
2
α J
1
RS (αJM M ) = (2J + 1) B(αJ → α J )
2
α J
J J 1
J J 1
× 3
−M M −q −M M −q
qq M
× Jqq (ναJ, α J ). (7.9f)
7.1.b Radiative Transfer Coefficients
Quite similar transformations can be applied to the different quantities entering
the radiative transfer equations – which will be simply referred to as the radiative
transfer coefficients. Substituting in Eqs. (6.88)
m → αu Ju Mu n → α J M
m → αu Ju Mu n → α J M ,
we obtain1
hν
=
ηiA (ν, Ω) N (2J + 1) B(α J → αu Ju )
4π
α J αu Ju
M −M Ju J 1 Ju J 1
× (−1) 3
−M u M −q −M u M −q
M M Mu qq
ρ
× Re Tqq (i, Ω) α J (M , M ) Φ(να J Mu , α J M − ν) (7.10a)
u u
1 should be multiplied by the factor ν 3 /ν 3
The second expression for εi (ν, Ω) αu Ju , α J , which
is however very close to unity because of the presence of the Φ profile.
284 CHAPTER 7
hν
=
ηiS (ν, Ω) N (2Ju + 1) B(αu Ju → α J )
4π
α J αu Ju
Ju J 1 Ju J 1
× 3
−M u M −q −M u M −q
Mu Mu M qq
ρ
× Re Tqq (i, Ω) α J (Mu , Mu ) Φ(να J Mu , α J M − ν) (7.10b)
u u u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) Re → Im (7.10c)
i
& '
= η S (ν, Ω)
ρSi (ν, Ω) Re → Im (7.10d)
i
3
= 2hν η S (ν, Ω)
εi (ν, Ω)
i
c2
hν
= N (2Ju + 1) A(αu Ju → α J )
4π
α J αu Ju
Ju J 1 Ju J 1
× 3
−M u M −q −M u M −q
Mu Mu M qq
ρ
× Re Tqq (i, Ω) α J (Mu , Mu ) Φ(να J Mu , α J M − ν) . (7.10e)
u u u u
7.2. The Multi-Level Atom in the Spherical Statistical
Tensor Representation
We will now convert the equations of the previous section into the spherical statis-
tical tensor representation.
7.2.a Statistical Equilibrium Equations
Multiplying both sides of Eq. (7.5) by
J−M
√ J J K
(−1) 2K + 1
M −M −Q
and carrying out the summation over M and M we obtain, with the use of
Eqs. (3.97) and (3.99)
d K
ρ (αJ) = −2πi νL gαJ Q ρKQ (αJ)
dt Q
K
+ ρQ (α J ) TA (αJKQ, α J K Q ) +
α J K Q
EQUATIONS FOR ATOMIC SYSTEMS 285
K
+ ρQu (αu Ju ) TE (αJKQ, αu Ju Ku Qu )
u
αu Ju Ku Qu
+ TS (αJKQ, αu Ju Ku Qu )
− Q (αJ) RA (αJKQK Q ) + RE (αJKQK Q )
ρK
K Q
+ RS (αJKQK Q ) , (7.11)
where
TA (αJKQ, α J K Q ) =
(2K + 1)(2K + 1)
J−M+J −M J J K J J K
× (−1)
M −M −Q M −M −Q
MM M M
× TA (αJM M , α J M M )
TE (αJKQ, αu Ju Ku Qu ) =
(2K + 1)(2Ku + 1)
J J K Ju Ju Ku
× (−1)J−M+Ju −Mu
M −M −Q Mu −Mu −Qu
MM Mu Mu
× TE (αJM M , αu Ju Mu Mu )
TS (αJKQ, αu Ju Ku Qu ) =
(2K + 1)(2Ku + 1)
J J K Ju Ju Ku
× (−1)J−M+Ju −Mu
M −M −Q Mu −Mu −Qu
MM Mu Mu
× TS (αJM M , αu Ju Mu Mu )
RA (αJKQK Q ) = (2K + 1)(2K + 1)
J J K
×
M −M −Q
MM M
J J K
× RA (αJM M )
M −M −Q
!
Q −Q J J K
+ (−1) RA (αJM M )
M −M −Q
286 CHAPTER 7
RE (αJKQK Q ) = (2K + 1)(2K + 1)
J J K
×
M −M −Q
MM M
J J K
× RE (αJM M )
M −M −Q
!
J J K
+ (−1)Q −Q R (αJM M
)
M −M −Q E
RS (αJKQK Q ) = (2K + 1)(2K + 1)
J J K
×
M −M −Q
MM M
J J K
× RS (αJM M )
M −M −Q
!
J J K
+ (−1)Q −Q R (αJM M
) .
M −M −Q S
These expressions can be further developed by substitution of Eqs. (7.9). For the
rate TA we have
TA (αJKQ, α J K Q ) = 3 (2J + 1) B(α J → αJ) (2K + 1)(2K + 1)
J J K J J K
× (−1)J−M+J −M
M −M −Q M −M −Q
MM M M qq
J J 1 J J 1
×
−M M −q −M M −q
× Jqq (ναJ, α J ). (7.12)
K
We can now introduce the irreducible radiation field tensor JQ (ν) using Eq. (5.156),
which gives
.
2Kr + 1 1 1 Kr K
1+q
Jqq (ν) = (−1) JQ r (ν) . (7.13)
3 q −q −Qr r
Kr Qr
Substitution into Eq. (7.12) yields1
1
To arrange the sign factor in this expression it should be kept in mind that – as apparent
from the various 3-j symbols – the indices K, Q, K , Q , Kr , Qr are integers.
EQUATIONS FOR ATOMIC SYSTEMS 287
TA (αJKQ, α J K Q ) = (2J + 1) B(α J → αJ)
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
J J K J J K
× (−1)1+J−J +Q
M −M −Q M −M −Q
MM M M qq
J J 1 J J 1 1 1 Kr
×
−M M −q −M M −q q −q −Qr
K
× JQ r (ναJ, α J ).
r
The sum of the product of the five 3-j symbols can be evaluated via Eq. (2.52).
After some manipulations similar to those of App. 2 we obtain1
TA (αJKQ, α J K Q ) = (2J + 1) B(α J → αJ)
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ J J 1 ⎬
K K Kr K
× (−1)K +Q J J 1 JQ r (ναJ, α ). (7.14a)
⎩ ⎭ −Q Q −Qr r
J
K K Kr
With analogous procedures involving the use of Eq. (2.34) for TE , Eq. (2.52) for
TS , Eq. (2.23a) for RE , and Eq. (2.42) – applied twice – for RA and RS , we get
TE (αJKQ, αu Ju Ku Qu ) = δKK δQQ (2Ju + 1) A(αu Ju → αJ)
u u
!
Ju Ju K
× (−1)1+J+Ju +K (7.14b)
J J 1
TS (αJKQ, αu Ju Ku Qu ) = (2Ju + 1) B(αu Ju → αJ)
× 3(2K + 1)(2Ku + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ J Ju 1 ⎬
K Ku Kr K
× (−1)Kr +Ku +Qu J Ju 1 JQ r (να J , αJ ) (7.14c)
⎩ ⎭ −Q Qu −Qr r u u
K Ku Kr
1
Note that the sign factor appearing in this formula, (−1)K +Q , can be written in various
different ways. As Q is an integer, we can also write (−1)K −Q . As the 9-j symbol is zero unless
(K + K + Kr ) is an even integer (because of the symmetry properties following Eq. (2.48)), the
sign factor can also be written (−1)K+Kr +Q ; etc.
288 CHAPTER 7
RA (αJKQK Q ) = (2J + 1) B(αJ → αu Ju )
αu Ju
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
! !
K K Kr 1 1 Kr K K Kr
× (−1)1+Ju −J+Kr +Q
J J J J J Ju Q −Q Qr
K
× ζ+ JQ r (να J , αJ ) (7.14d)
r u u
RE (αJKQK Q ) = δKK δQQ A(αJ → α J ) (7.14e)
α J
RS (αJKQK Q ) = (2J + 1) B(αJ → α J )
α J
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
! !
K K Kr 1 1 Kr K K Kr
× (−1)1+J −J+Q
J J J J J J Q −Q Qr
K
× ζ+ JQ r (ναJ, α J ) , (7.14f)
r
where1
1
1 + (−1)K+K +Kr .ζ+ =
2
The equations above were first deduced by Bommier and Sahal-Bréchot (1978).
Bommier (1977) had formerly derived the corresponding equations in the standard
representation.
7.2.b Radiative Transfer Coefficients
The radiative transfer coefficients can be easily expressed in the spherical statistical
in terms of the
tensor representation. Using again Eq. (3.99) and writing Tqq (i, Ω)
irreducible spherical tensor T Q
K via Eq. (5.156), we obtain from Eqs. (7.10)
(i, Ω)
= hν N
ηiA (ν, Ω) (2J + 1) B(α J → αu Ju )
4π
α J αu Ju
× 3(2K + 1)(2K + 1) ×
KQK Q
1
Note that the quantity ζ+ is 1 or 0 according as the integer (K + K + Kr ) is even or odd,
respectively.
EQUATIONS FOR ATOMIC SYSTEMS 289
Ju J 1 Ju J 1
× (−1)1+J −M +q
−Mu M −q −Mu M −q
M M Mu qq
1 1 K J J K
×
q −q −Q M −M −Q
× Re T Q
K ρK (α J ) Φ(ν
(i, Ω) − ν) (7.15a)
Q
α J Mu , α J M
u u
hν
=
ηiS (ν, Ω) N (2Ju + 1) B(αu Ju → α J )
4π
α J αu Ju
× 3(2K + 1)(2Ku + 1)
KQKu Qu
1+Ju −Mu +q Ju J 1 Ju J 1
× (−1)
−Mu M −q −Mu M −q
Mu Mu M qq
1 1 K Ju Ju Ku
×
q −q −Q Mu −Mu −Qu
× Re T Q
K ρKu (α J ) Φ(ν
(i, Ω) − ν) (7.15b)
Q u u u
α J Mu , α J M
u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) Re → Im (7.15c)
i
& '
= η S (ν, Ω)
ρSi (ν, Ω) Re → Im (7.15d)
i
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.15e)
i
c2
These expressions simplify considerably when the dependence of the Φ profiles on
the magnetic quantum numbers can be neglected. This occurs, for instance, when
the line width is much larger than the Zeeman splitting (the physical regimes Ia
and IIa of the classification scheme presented in Sect. 5.16), or when low-resolution
observations of the radiation emitted by a thin plasma are to be interpreted. Sub-
stituting in Eqs. (7.15)
Φ(να J Mu , α J M − ν) → Φ(να J , α J − ν) ,
u u u u
the summations over the magnetic quantum numbers can % be Kcarried out using
Q T Q (i, Ω) ρQ (αJ) is
K
Eq. (2.34). Taking also into account that the quantity
real – as apparent from the conjugation properties (3.102) and (5.158) – one gets
hν
=
ηiA (ν, Ω) N (2J + 1) B(α J → αu Ju ) ×
4π
α J αu Ju
290 CHAPTER 7
√ !
1 1 K ρK (α J )
× 3 (−1)1+J +Ju +K TQ
K
(i, Ω) Q
J J Ju
KQ
× φ(να J , α J − ν) (7.16a)
u u
= hν N
ηiS (ν, Ω) (2Ju + 1) B(αu Ju → α J )
4π
α J αu Ju
√ !
1 1 K ρK (α J )
× 3 (−1)1+J +Ju TQ K
(i, Ω) Q u u
Ju Ju J
KQ
× φ(να J , α J − ν) (7.16b)
u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) φ(ν − ν) → ψ(να − ν) (7.16c)
i α J , α J
u u
J , α J
u u
& '
= η S (ν, Ω)
ρSi (ν, Ω) φ(ν − ν) → ψ(να − ν) (7.16d)
i α J , α J
u u
J , α J
u u
3
= 2hν η S (ν, Ω)
εi (ν, Ω) , (7.16e)
2 i
c
where the profiles φ and ψ are defined in Eq. (6.59a).
7.3. Conjugation Properties of the Rates
The physical meaning of the various rates defined in Sect. 7.1.a is strictly analogous
to that of the more general rates introduced in Chap. 6 (see the discussion follow-
ing Eq. (6.65)). For instance, the quantity TA (αJM M , α J M M ) represents the
transfer rate, due to absorption, from the ‘lower-level coherence’ ρα J (M , M )
to the coherence ραJ (M, M ). Similarly, the quantity RA (αJM M ) represents
the relaxation rate, due to absorption to upper levels, connecting the coherence
ραJ (M, M ) with the coherence ραJ (M, M ). The rates introduced in Sect. 7.2.a
have a similar physical interpretation. The quantity TA (αJKQ, α J K Q ), for in-
stance, represents the transfer rate, due to absorption, from the spherical statistical
K
tensor of the lower level ρQ (α J ) to the spherical statistical tensor ρK
Q (αJ).
We can easily find the conjugation properties of the different rates. Bearing in
mind Eq. (5.154), we have from Eqs. (7.9)
TA (αJM M , α J M M )∗ = TA (αJM M, α J M M )
TE (αJM M , αu Ju Mu Mu )∗ = TE (αJM M, αu Ju Mu Mu )
= TE (αJM M , αu Ju Mu Mu )
TS (αJM M , αu Ju Mu Mu )∗ = TS (αJM M, αu Ju Mu Mu )
EQUATIONS FOR ATOMIC SYSTEMS 291
RA (αJM M )∗ = RA (αJM M )
RE (αJM M )∗ = RE (αJM M ) = RE (αJM M )
RS (αJM M )∗ = RS (αJM M ) .
As to the rates in the spherical statistical tensor representation, we obtain from
Eqs. (7.14), with the help of Eqs. (5.158)
TA (αJKQ, α J K Q )∗ = (−1)Q+Q TA (αJK −Q, α J K −Q)
TE (αJKQ, αu Ju Ku Qu )∗ = (−1)Q+Qu TE (αJK −Q, αuJu Ku −Qu)
= TE (αJKQ, αu Ju Ku Qu )
TS (αJKQ, αu Ju Ku Qu )∗ = (−1)Q+Qu TS (αJK −Q, αu Ju Ku −Qu )
RA (αJKQK Q )∗ = (−1)Q+Q RA (αJK −QK −Q )
RE (αJKQK Q )∗ = (−1)Q+Q RE (αJK −QK −Q ) = RE (αJKQK Q )
RS (αJKQK Q )∗ = (−1)Q+Q RS (αJK −QK −Q ) .
7.4. The No-Coherence Case
The equations derived in Sects. 7.1 and 7.2 are rather involved, which is quite
natural as they describe a complex physical situation. It is interesting to consider
in detail the special case where atomic polarization can be disregarded; obviously,
the equations become considerably simpler in this case.
Let us assume that coherences between Zeeman sublevels can be neglected or,
in other words, that the atomic density matrix is diagonal. This case – which
will be referred to as no-coherence case – covers the union (in the sense of the
mathematical set theory) of the physical regimes IIIa, IVa, Va, IIIb outlined in
Sect. 5.16. The statistical equilibrium equations in the standard representation
might be deduced directly from Eq. (7.5). However, the resulting equations would
be of limited use because of the severe assumption on the spectrum of the incident
radiation. Since no coherences between magnetic sublevels are present, we can now
make – still consistently with the flat-spectrum approximation – the less restrictive
assumption that the incident radiation field is flat across frequency intervals wider
than the inverse lifetimes of the magnetic sublevels, allowing for its dependence on
the individual Bohr frequencies ναJM, α J M . Performing in Eqs. (6.62), (6.69) the
same substitutions as in Sect. 7.1.a and introducing shorthand notations for the
292 CHAPTER 7
rates, we have1
d
ραJ (M ) = ρα J (M ) tA (αJM, α J M )
dt
α J M
+ ρα J (Mu ) tE (αJM, αu Ju Mu )
u u
αu Ju Mu
+ tS (αJM, αu Ju Mu )
− ραJ (M ) rA (αJM ) + rE (αJM ) + rS (αJM ) , (7.17)
where
tA (αJM, α J M ) = (2J + 1) B(α J → αJ)
J J 1
2
× 3 Jqq (ναJM, α J M ) (7.18a)
−M M −q
q
J J 1
2
tE (αJM, αu Ju Mu ) = (2Ju + 1) A(αu Ju → αJ) u (7.18b)
−Mu M −q
q
tS (αJM, αu Ju Mu ) = (2Ju + 1) B(αu Ju → αJ)
J J 1
2
× 3 u Jqq (να J M , αJM ) (7.18c)
−Mu M −q u u u
q
rA (αJM ) = (2J + 1) B(αJ → αu Ju )
αu Ju
2
Ju J 1
× 3 Jqq (να J Mu , αJM ) (7.18d)
−Mu M −q u u
qMu
rE (αJM ) = A(αJ → α J ) (7.18e)
α J
rS (αJM ) = (2J + 1) B(αJ → α J )
α J
2
J J 1
× 3 Jqq (ναJM, α J M ) , (7.18f)
−M M −q
qM
where Eq. (3.91b) has been used.
1
The expressions for tE (αJM, αu Ju Mu ) and rE (αJM ) imply, respectively, the approximations
3 3 3 3
να J Mu , αJ M = να J , αJ , ναJ M, α J M = ναJ, α J .
u u u u
EQUATIONS FOR ATOMIC SYSTEMS 293
For the radiative transfer coefficients we obtain from Eqs. (7.10), with the help
of Eq. (5.148)
= hν N
ηiA (ν, Ω) (2J + 1) B(α J → αu Ju )
4π
α J αu Ju
2
Ju J 1 ρ
× 3 Tqq (i, Ω) (M )
−Mu M −q α J
M Mu q
× φ(να J Mu , α J M − ν) (7.19a)
u u
= hν N
ηiS (ν, Ω) (2Ju + 1) B(αu Ju → α J )
4π
α J αu Ju
2
Ju J 1 ρ
× 3 Tqq (i, Ω) (Mu )
−Mu M −q α J
u u
M Mu q
× φ(να J M , α J M − ν) (7.19b)
u u u
& '
= η (ν, Ω) φ(ν
αu Ju Mu , α J M − ν) → ψ(ναu Ju Mu , α J M − ν)
A A
ρi (ν, Ω) i (7.19c)
& '
= η S (ν, Ω)
φ(ν
ρSi (ν, Ω) i α J M , α J M − ν) → ψ(να J M , α J M − ν)
u u u u u u
(7.19d)
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.19e)
2 i
c
Similar simplifications occur in the equations written in the spherical statistical
tensor representation. The no-coherence hypothesis results here in the fact that all
the components of the multipole moments ρK Q vanish except those having Q = 0,
as apparent from the definition (3.101). Multiplying both sides of Eq. (7.17) by
√ J J K
(−1)J−M 2K + 1 ,
M −M 0
carrying out the summation over M , and using Eqs. (3.99), (3.101) and (7.13), we
obtain the following simplified statistical equilibrium equations1
d K K
ρ0 (αJ) = ρ0 (α J ) tA (αJK, α J K )
dt
α J K
K
+ ρ0 u (αu Ju ) tE (αJK, αu Ju Ku ) + tS (αJK, αu Ju Ku )
αu Ju Ku
− ρK
0 (αJ) rA (αJKK ) + rE (αJKK ) + rS (αJKK ) ,
K
1
The expressions for the rates tE (αJK, αu Ju Ku ) and rE (αJKK ) are derived with the help
of Eqs. (2.34) and (2.23a), respectively.
294 CHAPTER 7
where
tA (αJK, α J K ) = (2J + 1) B(α J → αJ) 3(2K + 1)(2K + 1)(2Kr + 1)
Kr
J J K J J K
× (−1)1+J−J
M −M 0 M −M 0
MM q
2
J J 1 1 1 Kr K
× J0 r (ναJM, α J M ) (7.20a)
−M M −q q −q 0
tE (αJK, αu Ju Ku ) = δKK (2Ju + 1) A(αu Ju → αJ)
u
!
Ju Ju K
× (−1)1+J+Ju +K
(7.20b)
J J 1
tS (αJK, αu Ju Ku ) = (2Ju + 1) B(αu Ju → αJ) 3(2K + 1)(2Ku + 1)(2Kr + 1)
Kr
J J K Ju Ju Ku
× (−1) 1+J−Ju
M −M 0 Mu −Mu 0
MMu q
2
Ju J 1 1 1 Kr K
× J0 r (να J Mu , αJM ) (7.20c)
−Mu M −q q −q 0 u u
rA (αJKK ) = (2J + 1) B(αJ → αu Ju ) 3(2K + 1)(2K + 1)(2Kr + 1)
αu Ju Kr
J J K J J K
× (−1) 1+q
M −M 0 M −M 0
MMu q
2
Ju J 1 1 1 Kr K
× J0 r (να J Mu , αJM ) (7.20d)
−Mu M −q q −q 0 u u
rE (αJKK ) = δKK A(αJ → α J ) (7.20e)
α J
rS (αJKK ) = (2J + 1) B(αJ → α J ) 3(2K + 1)(2K + 1)(2Kr + 1)
α J Kr
J J K J J K
× (−1) 1+q
M −M 0 M −M 0
MM q
2
J J 1 1 1 Kr K
× J0 r (ναJM, α J M ) . (7.20f)
−M M −q q −q 0
EQUATIONS FOR ATOMIC SYSTEMS 295
For the radiative transfer coefficients we have from Eqs. (7.15), with the use of
the conjugation relations (3.102) and (5.158)
hν
=
ηiA (ν, Ω) N (2J + 1) B(α J → αu Ju ) 3(2K + 1)(2K + 1)
4π
α J αu Ju KK
2
Ju J 1 1 1 K J J K
× (−1)1+J −M +q
−Mu M −q q −q 0 M −M 0
M Mu q
ρK (α J ) φ(ν
× T 0K (i, Ω) − ν) (7.21a)
0 α J Mu , α J M
u u
= hν N
ηiS (ν, Ω) (2Ju + 1) B(αu Ju → α J ) 3(2K + 1)(2Ku + 1)
4π
α J αu Ju KKu
2
1+Ju −Mu +q Ju J 1 1 1 K Ju Ju Ku
× (−1)
−Mu M −q q −q 0 Mu −Mu 0
M Mu q
K
ρ u (α J ) φ(ν
× T 0K (i, Ω) − ν) (7.21b)
0 u u α J Mu , α J M
u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) φ(ν − ν) → ψ(ν − ν) (7.21c)
i αu Ju Mu , α J M αu Ju Mu , α J M
& '
= η S (ν, Ω)
ρSi (ν, Ω) φ(ν − ν) → ψ(ν − ν) (7.21d)
i α J M ,α J M
u u u
α J M ,α J M u u u
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.21e)
2 i
c
7.5. The Multi-Term Atom in the Energy-Eigenvector Representation
We will now consider the interaction of an atomic system with a polarized radiation
field in more general terms, dropping some of the basic restrictions adopted so far.
First, we will allow for coherences between different J-levels, besides those between
magnetic sublevels of any individual J-level. Second, we will drop the limitation
on the magnetic field intensity,1 so that the validity of the new equations is not
restricted to the Zeeman effect regime. The only restrictions that will be retained
are the following:
a) the atom is devoid of hyperfine structure;
b) the atom is described by the L-S coupling scheme (cf. Sect. 3.1), so that
the different J-levels are grouped in terms, each term being characterized by the
quantum numbers β, L, and S;
c) coherences between J-levels pertaining to different terms are negligible.
1
The magnetic field must be however sufficiently weak for the diamagnetic term in the atomic
Hamiltonian to be negligible. This implies B 1010 G (see the discussion in Sect. 3.1).
296 CHAPTER 7
The model atom described by these approximations will be named in the following
the multi-term atom. For such system, the flat-spectrum approximation discussed
in Sect. 6.5 requires that the radiation field incident on the atom should be constant
across frequency intervals larger than the frequency separation between different
levels belonging to the same term. The radiation field can thus be characterized
by specifying its value at the Bohr frequencies
E(βLS) − E(β L S )
νβLS, β L S = , (7.22)
h
where E(βLS), E(β L S ) are the energies of the relevant terms (which disregard
both the spin-orbit and the magnetic Hamiltonian). This approximation is indeed
rather restrictive, but it can only be released within the framework of more general
theories able to encompass frequency redistribution effects (see the introductory
discussion to Chap. 6).
We now apply the equations derived in Chap. 6 to the multi-term atom embedded
in an arbitrary magnetic field. To this aim we must go back to the results obtained
in Sect. 3.4 for the Paschen-Back effect regime – which obviously describe, as a
particular case, also the Zeeman effect regime. Using the notations of Sect. 3.4
(which implies that the z-axis of the reference system is in the magnetic field
direction) we have for the eigenvectors and eigenvalues of the atomic Hamiltonian
HA |βLSjM = E(βLS) + λj (βLS, M ) |βLSjM ,
where the eigenvectors are given by (see Eq. (3.58))
|βLSjM = CJj (βLS, M ) |βLSJM . (7.23)
J
Therefore, we have to modify the formalism of Chap. 6 according to the following
substitutions
i) for the energy eigenvectors:
| n → |βLSjM ; (7.24)
ii) for the corresponding energy eigenvalues:
En → E(βLS) + λj (βLS, M ) , (7.25)
whence
iii) for the Bohr frequencies:
En − Em
νnm = → νβLSjM, β L S j M =
h
λj (βLS, M ) − λj (β L S , M )
= νβLS, β L S + ; (7.26)
h
EQUATIONS FOR ATOMIC SYSTEMS 297
iv) for the matrix elements of the spherical components of the dipole operator:
(dq )nm = n | dq | m → (dq )βLSjM, β L S j M = βLSjM | dq |β L S j M =
= CJj (βLS, M ) CJj (β L S , M ) βLSJM | dq |β L S J M
JJ
= CJj (βLS, M ) CJj (β L S , M )
JJ
J +M+1
√ J J 1
× (−1) 2J + 1 βLSJ d β L S J
−M M q
= CJj (βLS, M ) CJj (β L S , M )
JJ
× (−1)L+S−M (2L + 1)(2J + 1)(2J + 1)
!
J J 1 L L 1
× βL d β L δSS , (7.27)
−M M q J J S
where we have used the reality of the CJj coefficients (cf. Sect. 3.4), the Wigner-
Eckart theorem (Eq. (2.96)), and Eq. (2.108);
v) finally, for the atomic density-matrix elements:
ρnm = n | ρ | m → βLSjM | ρ |β L S j M =
= δββ δLL δSS βLSjM | ρ |βLSj M
= δββ δLL δSS ρβLS (jM, j M ) , (7.28)
where we have used assumption c).
By means of substitutions i) to v) the statistical equilibrium equations and the
radiative transfer equations for the multi-term atom in an arbitrary magnetic field
can be directly deduced from the corresponding equations of Chap. 6.
7.5.a Statistical Equilibrium Equations
Observing that dipole transitions are only allowed between terms having the same
S-value (as apparent from the factor δSS in Eq. (7.27)), we replace in Eqs. (6.62)
and (6.69) (see also Fig. 6.1)
m → βLSjM
n → β L Sj M p → βu Lu Sju Mu
m → βLSj M
n → β L Sj M p → βu Lu Sju Mu ,
m → βLSj M
where the quantum numbers (β L S) and (βu Lu S) refer to any term of energy lower
or higher, respectively, than E(βLS). We obtain
298 CHAPTER 7
d
ρ (jM, j M ) = −2πi νβLS (jM, j M ) ρβLS (jM, j M )
dt βLS
+ ρβ L S (j M , j M ) TA (βLSjM j M , β L Sj M j M )
β L j M j M
+ ρβ L S (ju Mu , ju Mu ) TE (βLSjM j M , βu Lu Sju Mu ju Mu )
u u
βu Lu ju Mu ju Mu
+ TS (βLSjM j M , βu Lu Sju Mu ju Mu )
&
− ρβLS (jM, j M ) RA (βLSj M j M )
j M
+ RE (βLSj M j M ) + RS (βLSj M j M )
+ ρβLS (j M , j M ) RA (βLSj M jM )
'
+ RE (βLSjM j M ) + RS (βLSjM j M ) , (7.29)
where
λj (βLS, M ) − λj (βLS, M )
νβLS (jM, j M ) = . (7.30)
h
Similarly to Sect. 7.1.a, we now define the line strength of the (electric-dipole)
transition between two terms (β L S) and (βu Lu S) – with E(βu Lu S) > E(β L S) –
by the symmetrical expression
S(β L S, βu Lu S) = S(βu Lu S, β L S) = (2L + 1) | β L d βu Lu |2
= (2Lu + 1) | βu Lu d β L |2 . (7.31)
The Einstein coefficients for spontaneous and stimulated emission and for absorp-
tion are connected with the line strength by the relations1
1
Starting from Eqs. (6.71) and (6.73), it can be proved that the Einstein coefficients for the
transition between the individual levels (β L Sj M ) and (βu Lu Sju Mu ) are connected with the
Einstein coefficients defined here by the relations
%
Aβ L Sju Mu , β L Sj M = A(βu Lu S → β L S)
%
j M u u
j M
Bβ L Sju Mu , β L Sj M = B(βu Lu S → β L S)
% u u
ju Mu
Bβ L Sj M , βu Lu Sju Mu = B(β L S → βu Lu S)
(the first relation holds only under the limit
νβ3 L Sju Mu , β L Sj M = νβ3 L S, β L S ).
u u u u
The proof can be carried out with the help of Eqs. (7.27), (3.62), (2.23a) and (2.39).
EQUATIONS FOR ATOMIC SYSTEMS 299
64π 4 3
(2Lu + 1) A(βu Lu S → β L S) = ν S(βu Lu S, β L S)
3hc3 βu Lu S, β L S
32π 4
(2Lu + 1) B(βu Lu S → β L S) = S(βu Lu S, β L S)
3h2 c
32π 4
(2L + 1) B(β L S → βu Lu S) = S(β L S, βu Lu S) , (7.32)
3h2 c
with
2h 3
A(βu Lu S → β L S) = ν B(βu Lu S → β L S)
c2 βu Lu S, β L S
(2L + 1) B(β L S → βu Lu S) = (2Lu + 1) B(βu Lu S → β L S) . (7.33)
Performing the substitutions outlined above, we obtain after some algebra
TA (βLSjM j M , β L Sj M j M ) = (2L + 1) B(β L S → βLS)
× 3 (−1)M −M CJj (βLS, M ) CJj (βLS, M )
JJ J J qq
j j
× CJ (β L S, M ) CJ (β L S, M ) (2J + 1)(2J + 1)(2J + 1)(2J + 1)
! !
J J 1 J J 1 L L 1 L L 1
×
−M M −q −M M −q J J S J J S
× Jqq (νβLS, β L S ) (7.34a)
TE (βLSjM j M , βu Lu Sju Mu ju Mu ) = (2Lu + 1) A(βu Lu S → βLS)
× (−1)Mu −Mu CJj (βLS, M ) CJj (βLS, M )
JJ Ju Ju q
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
! !
Ju J 1 Ju J 1 Lu L 1 Lu L 1
× (7.34b)
−Mu M −q −Mu M −q J Ju S J Ju S
TS (βLSjM j M , βu Lu Sju Mu ju Mu ) = (2Lu + 1) B(βu Lu S → βLS)
× 3 (−1)M−M CJj (βLS, M ) CJj (βLS, M ) ×
JJ Ju Ju qq
300 CHAPTER 7
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
! !
Ju J 1 Ju J 1 Lu L 1 Lu L 1
×
−Mu M −q −Mu M −q J Ju S J Ju S
× Jqq (νβ L S, βLS ) (7.34c)
u u
1
RA (βLSjM j M ) = (2L + 1) B(βLS → βu Lu S)
2
βu Lu
× 3 (−1)q+q CJj (βLS, M ) CJj (βLS, M )
JJ Ju qq Mu
Ju J 1 Ju J 1
× (2J + 1)(2J + 1) (2Ju + 1)
−Mu M −q −Mu M −q
! !
Lu L 1 Lu L 1
× Jqq (νβ L S, βLS ) (7.34d)
J Ju S J Ju S u u
1
RE (βLSjM j M ) = δjj δMM A(βLS → β L S) (7.34e)
2
β L
1
RS (βLSjM j M ) = (2L + 1) B(βLS → β L S)
2
β L
× 3 CJj (βLS, M ) CJj (βLS, M )
JJ J qq M
J J 1 J J 1
× (2J + 1)(2J + 1) (2J + 1)
−M M −q −M M −q
! !
L L 1 L L 1
× Jqq (νβLS, β L S ) . (7.34f)
J J S J J S
The expressions for RA and RS have been derived using Eq. (3.62b); the expression
for RE using Eqs. (3.62b), (2.23a), (2.39), and (3.62a).
7.5.b Radiative Transfer Coefficients
Substituting in Eqs. (6.88)
m → βu Lu Sju Mu n → β L Sj M
m → βu Lu Sju Mu n → β L Sj M ,
we obtain by similar transformations
EQUATIONS FOR ATOMIC SYSTEMS 301
= hν N
ηiA (ν, Ω) (2L + 1) B(β L S → βu Lu S)
4π
β L Sβu Lu
j j
× 3 (−1)q+q CJ (β L S, M ) CJ (β L S, M )
j j J J ju Ju Ju M M Mu qq
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
! !
Ju J 1 Ju J 1 Lu L 1 Lu L 1
×
−Mu M −q −Mu M −q J Ju S J Ju S
ρ
× Re Tqq (i, Ω) β L S (j M , j M ) Φ(νβ L Sju Mu , β L Sj M − ν) (7.35a)
u u
= hν N
ηiS (ν, Ω) (2Lu + 1) B(βu Lu S → β L S)
4π
βu Lu Sβ L
j j
× 3 CJ (β L S, M ) CJ (β L S, M )
ju ju Ju Ju j J J Mu Mu M qq
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
! !
Ju J 1 Ju J 1 Lu L 1 Lu L 1
×
−Mu M −q −Mu M −q J Ju S J Ju S
ρ
× Re Tqq (i, Ω) β L S (ju Mu , ju Mu ) Φ(νβ L Sju Mu , β L Sj M − ν) (7.35b)
u u u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) Re → Im (7.35c)
i
& '
= η S (ν, Ω)
ρSi (ν, Ω) Re → Im (7.35d)
i
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.35e)
i
c2
7.6. The Multi-Term Atom in the Spherical Statistical
Tensor Representation
It is worth pointing out an important difference between the case of the multi-level
atom considered in Sect. 7.1 and the case of the multi-term atom of Sect. 7.5. In
the former case, because of the Zeeman effect regime assumption, the energy eigen-
vectors coincide with the eigenvectors of angular momentum, so that there is no
302 CHAPTER 7
difference between energy-eigenvector representation and standard representation
of the atomic density operator. In the latter case, on the contrary – where the lim-
itation on the magnetic field intensity is dropped (Paschen-Back effect regime) –
the two representations are distinct, which leads to the appearance of the CJj coeffi-
cients of Eq. (7.23) in the expressions of the rates (Eqs. (7.34)) and of the radiative
transfer coefficients (Eqs. (7.35)). This fact, together with the consideration of co-
herences between different J-levels, is responsible for the much greater complexity
of the equations of Sect. 7.5 compared with those of Sect. 7.1.
It is a simple task to rewrite the equations for the multi-term atom either in the
standard representation or in the spherical statistical tensor representation. We
recall that the various representations are connected by the linear relations (see
Eqs. (7.23) and (3.99))
ρβLS (jM, j M ) = CJj (βLS, M ) CJj (βLS, M ) ρβLS (JM, J M )
JJ
ρβLS (jM, j M ) = CJj (βLS, M ) CJj (βLS, M )
JJ
√ J J K
× (−1) J−M
2K + 1 ρQ (J, J )
βLS K
(7.36)
M −M −Q
KQ
with the inverse formulae, that can be deduced using Eqs. (3.62b) and (3.97)
ρβLS (JM, J M ) = CJj (βLS, M ) CJj (βLS, M ) ρβLS (jM, j M )
jj
ρQ (J, J )
βLS K
= CJj (βLS, M ) CJj (βLS, M )
jMj M
√ J J K
× (−1)J−M 2K + 1 ρβLS (jM, j M ) . (7.37)
M −M −Q
In the following we derive the statistical equilibrium equations and the radiative
transfer coefficients directly in the spherical statistical tensor representation.
7.6.a Statistical Equilibrium Equations
Multiplication of both sides of Eq. (7.29) by
√ J J K
CJj (βLS, M ) CJj (βLS, M ) (−1)J−M 2K + 1
M −M −Q
followed by summation over j, M , j , M yields, with the use of Eqs. (7.36)-(7.37)
EQUATIONS FOR ATOMIC SYSTEMS 303
d
βLS K
ρQ (J, J )
βLS K
= −2πi NβLS (KQJJ , K Q J J ) ρQ (J , J )
dt
K Q J J
β L S K
+ ρQ (J , J ) TA (βLSKQJJ , β L SK Q J J )
β L K Q J J
βu Lu S Ku
+ ρQ (Ju , Ju ) TE (βLSKQJJ , βu Lu SKu Qu Ju Ju )
u
βu Lu Ku Qu Ju Ju
+ TS (βLSKQJJ , βu Lu SKu Qu Ju Ju )
βLS K
− ρQ (J , J ) RA (βLSKQJJ K Q J J )
K Q J J
+ RE (βLSKQJJ K Q J J ) + RS (βLSKQJJ K Q J J ) , (7.38)
where
NβLS (KQJJ , K Q J J ) =
= CJj (βLS, M ) CJj (βLS, M ) CJj (βLS, M ) CJj (βLS, M )
jMj M
× (−1)J−J (2K + 1)(2K + 1)
J J K J J K
× νβLS (jM, j M ) (7.39)
M −M −Q M −M −Q
TA (βLSKQJJ , β L SK Q J J ) =
j j
= CJj (βLS, M ) CJj (βLS, M ) CJ (β L S, M ) CJ (β L S, M )
jMj M j M j M
J J K
× (−1)J−M+J −M (2K + 1)(2K + 1)
M −M −Q
J J K
× TA (βLSjM j M , β L Sj M j M ) (7.40a)
M −M −Q
TE (βLSKQJJ , βu Lu SKu Qu Ju Ju ) =
j j
= CJj (βLS, M ) CJj (βLS, M ) CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu )
u u
jMj M ju Mu ju Mu
J J K
× (−1)J−M+Ju −Mu (2K + 1)(2Ku + 1)
M −M −Q
Ju Ju Ku
× TE (βLSjM j M , βu Lu Sju Mu ju Mu ) (7.40b)
Mu −Mu −Qu
304 CHAPTER 7
RA (βLSKQJJ K Q J J ) =
= CJj (βLS, M ) CJj (βLS, M )
jMj M j M
J−J
J J K
× (−1) (2K + 1)(2K + 1)
M −M −Q
× CJj (βLS, M ) CJj (βLS, M )
J J K
× RA (βLSj M j M )
M −M −Q
+ (−1)M −M CJj (βLS, M ) CJj (βLS, M )
J J K
× RA (βLSj M jM ) (7.40c)
M −M −Q
RE (βLSKQJJ K Q J J ) =
= CJj (βLS, M ) CJj (βLS, M )
jMj M j M
J J K
× (−1)J−J (2K + 1)(2K + 1)
M −M −Q
× CJj (βLS, M ) CJj (βLS, M )
J J K
× RE (βLSj M j M )
M −M −Q
+ (−1)M −M CJj (βLS, M ) CJj (βLS, M )
J J K
× RE (βLSjM j M ) (7.40d)
M −M −Q
(the relation between TS and TS is identical to the relation between TE and TE , as
obvious from the structure of Eq. (7.29); the same holds for the relation between
RS and RS , which is identical to the relation between RE and RE ).
The term N defined in Eq. (7.39) can be evaluated in greater detail. Using
Eqs. (7.30) and (3.62b,c) we have
NβLS (KQJJ , K Q J J ) = δQQ (−1)J−J (2K + 1)(2K + 1)
1 J J K
J J K
× δ
h JJ
M −M −Q M −M −Q
MM
× βLSJM |Hso + HB |βLSJ M +
EQUATIONS FOR ATOMIC SYSTEMS 305
J J K
J J K
− δJJ
M −M −Q M −M −Q
MM
× βLSJ M |Hso + HB |βLSJ M ,
where Hso is the spin-orbit Hamiltonian and HB the magnetic Hamiltonian. The
matrix elements in the right-hand side can be easily evaluated. Taking into account
that Hso is diagonal with respect to J, expressing the matrix elements of HB via
Eq. (3.57), and performing the summation over M and M by means of Eqs. (2.23a),
(3.42) and (2.42), we obtain after some algebra
NβLS (KQJJ , K Q J J ) = δKK δQQ δJJ δJ J νβLSJ, βLSJ
K K 1
+ δQQ νL (−1)J+J −Q (2K + 1)(2K + 1)
−Q Q 0
!
K K 1
× δJ J ΓLS (J, J )
J J J
!
K−K K K 1
+ δJJ (−1) ΓLS (J , J ) , (7.41)
J J J
where
ΓLS (J, J ) = (−1)J−J ΓLS (J , J)
= δJJ J(J + 1)(2J + 1)
!
J J 1
+ (−1) 1+L+S+J
(2J + 1)(2J + 1)S(S + 1)(2S + 1) (7.42)
S S L
and where
EβLS (J) − EβLS (J )
νβLSJ, βLSJ = , (7.43)
h
with EβLS (J) defined in Eq. (3.60). The quantity ΓLS (J, J ) is a sort of generalized
Landé factor. Use of Eq. (2.36d) shows that
ΓLS (J, J) = J(J + 1)(2J + 1) gLS (J) , (7.44)
where gLS (J) is the usual Landé factor introduced in Eq. (3.8).
The expressions for the different rates can be further developed by substituting
Eqs. (7.34) into Eqs. (7.40). Performing the summations over the indices j, j , j ,
j etc. with the use of Eqs. (3.62b), and substituting Eq. (7.13), the calculations
reduce to evaluating the sums over angular momentum components of products of
several 3-j symbols. This can be done using Eq. (2.52) for the rates TA and TS ,
306 CHAPTER 7
Eq. (2.34) for TE , Eq. (2.42) applied twice for RA and RS , and Eq. (2.23a), also
applied twice, for RE . The final results are the following
TA (βLSKQJJ , β L SK Q J J ) = (2L + 1) B(β L S → βLS)
× 3(2J + 1)(2J + 1)(2J + 1)(2J + 1)(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ J J 1 ⎬ !
K +Q +J −J L L 1
× (−1) J J 1
⎩ ⎭ J J S
K K Kr
!
L L 1 K K Kr K
× JQ r (νβLS, β L S ) (7.45a)
J J S −Q Q −Qr r
TE (βLSKQJJ , βu Lu SKu Qu Ju Ju ) = δKK δQQ (2Lu + 1) A(βu Lu S → βLS)
u u
× (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
! ! !
1+K+J +Ju J J K Lu L 1 Lu L 1
× (−1) (7.45b)
Ju Ju 1 J Ju S J Ju S
TS (βLSKQJJ , βu Lu SKu Qu Ju Ju ) = (2Lu + 1) B(βu Lu S → βLS)
× 3(2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)(2K + 1)(2Ku + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ J Ju 1 ⎬ !
Lu L 1
× (−1)Kr +Ku +Qu +Ju −Ju J Ju 1
⎩ ⎭ J Ju S
K Ku Kr
!
Lu L 1 K Ku Kr K
× JQ r (νβ L S, βLS ) (7.45c)
J Ju S −Q Qu −Qr r u u
RA (βLSKQJJ K Q J J ) = (2L + 1) (2Ju + 1) B(βLS → βu Lu S)
βu Lu Ju
+Kr +Q K K Kr
× 3(2K + 1)(2K + 1)(2Kr + 1) (−1)1+Ju −J
Q −Q Qr
Kr Qr
1
× δJJ (2J + 1)(2J + 1)
2 ! !
J−J Lu L 1 Lu L 1
× (−1)
J Ju S J Ju S
! !
K K Kr 1 1 Kr
× +
J J J J J Ju
EQUATIONS FOR ATOMIC SYSTEMS 307
+ δJ J (2J + 1)(2J + 1)
! !
Lu L 1 Lu L 1
× (−1)K+K +Kr
J Ju S J Ju S
! !
K K Kr 1 1 Kr K
× JQ r (νβ L S, βLS ) (7.45d)
J J J J J Ju r u u
RE (βLSKQJJ K Q J J ) = δKK δQQ δJJ δJ J
× (2L + 1) A(βLS → β L S) (2J + 1)
β L J
!2 !2
1 L L 1 L L 1
× + (7.45e)
2 J J S J J S
RS (βLSKQJJ K Q J J ) = (2L + 1) (2J + 1) B(βLS → β L S)
β L J
1+J −J +Q K K Kr
× 3(2K + 1)(2K + 1)(2Kr + 1) (−1)
Q −Q Qr
Kr Qr
1
× δJJ (2J + 1)(2J + 1)
2 ! !
J−J L L 1 L L 1
× (−1)
J J S J J S
! !
K K Kr 1 1 Kr
×
J J J J J J
+ δJ J (2J + 1)(2J + 1)
! !
L L 1 L L 1
× (−1)K+K +Kr
J J S J J S
! !
K K Kr 1 1 Kr K
× JQ r (νβLS, β L S ) . (7.45f)
J J J J J J r
The above expressions for the relaxation rates can be further simplified by carrying
out the summations on the J quantum numbers. Using Eq. (2.41) for RA and RS ,
and Eq. (2.39) for RE we obtain
RA (βLSKQJJ K Q J J ) = (2L + 1) B(βLS → βu Lu S)
βu Lu
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
!
1+Lu −S+J+Q L L Kr K K Kr
× (−1) ×
1 1 Lu Q −Q Qr
308 CHAPTER 7
1
× δJJ (2J + 1)(2J + 1)
2
! !
L L Kr K K Kr
×
J J S J J J
+ δJ J (2J + 1)(2J + 1) (−1)J −J +K+K +Kr
! !
L L Kr K K Kr K
× JQ r (νβ L S, βLS ) (7.46a)
J J S J J J r u u
RE (βLSKQJJ K Q J J ) = δKK δQQ δJJ δJ J
× A(βLS → β L S) (7.46b)
β L
RS (βLSKQJJ K Q J J ) = (2L + 1) B(βLS → β L S)
β L
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
!
1+L −S+J+Kr +Q L L Kr K K Kr
× (−1)
1 1 L Q −Q Qr
1
× δJJ (2J + 1)(2J + 1)
2
! !
L L Kr K K Kr
×
J J S J J J
+ δJ J (2J + 1)(2J + 1) (−1)J −J +K+K +Kr
! !
L L Kr K K Kr K
× JQ r (νβLS, β L S ) . (7.46c)
J J S J J J r
Expressions (7.45a-c) and (7.46a-c) have been given by Landi Degl’Innocenti
(1982b).1 The statistical equilibrium equations for the multi-term atom in the
standard representation of the atomic density operator have been derived by Bom-
mier (1980).
7.6.b Radiative Transfer Coefficients
Substitution of Eqs. (7.36) and (5.156) – applied to the tensor Tqq (i, Ω) – into
Eqs. (7.35) leads, with the help of Eq. (3.62b), to the following expressions
1
The expressions appearing in that paper contain indeed some additional terms, due to
K
the tensor FQ r , in the rates RA and RS . These terms are of very limited importance and their
r
presence is not justified in the flat-spectrum approximation.
EQUATIONS FOR ATOMIC SYSTEMS 309
= hν N
ηiA (ν, Ω) (2L + 1) B(β L S → βu Lu S)
4π
β L Sβu Lu
× 3(2K + 1)(2K + 1)
KQK Q
j j
× (−1)1+J −M +q CJ (β L S, M ) CJ (β L S, M )
j J J J ju Ju Ju M M Mu qq
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
Ju J 1 Ju J 1 1 1 K
×
−Mu M −q −Mu M −q q −q −Q
! !
J J K Lu L 1 Lu L 1
×
M −M −Q J Ju S J Ju S
β L S K
× Re T Q
K
(i, Ω) ρQ (J , J ) Φ(νβ L Sju Mu , β L Sj M − ν) (7.47a)
u u
= hν N
ηiS (ν, Ω) (2Lu + 1) B(βu Lu S → β L S)
4π
βu Lu Sβ L
× 3(2K + 1)(2Ku + 1)
KQKu Qu
j j
× (−1)1+Ju −Mu +q CJ (β L S, M ) CJ (β L S, M )
ju Ju Ju Ju j J J Mu Mu M qq
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
Ju J 1 Ju J 1 1 1 K
×
−Mu M −q −Mu M −q q −q −Q
! !
Ju Ju Ku Lu L 1 Lu L 1
×
Mu −Mu −Qu J Ju S J Ju S
βu Lu S Ku
× Re T Q
K
(i, Ω) ρQ (Ju , Ju ) Φ(νβ L Sju Mu , β L Sj M − ν) (7.47b)
u u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) Re → Im (7.47c)
i
& '
= η S (ν, Ω)
ρSi (ν, Ω) Re → Im (7.47d)
i
= 2hν 3 S
εi (ν, Ω) ηi (ν, Ω) . (7.47e)
c2
310 CHAPTER 7
These expressions become much simpler when the frequency splittings – due to
fine structure and to the magnetic field – of the lines of each multiplet can be
neglected (see the analogous case discussed at the end of Sect. 7.2.b). Replacing
the profiles
Φ(νβ L Sj M , β L Sj M − ν)
u u u u
in Eqs. (7.47) by a single profile
Φ(νβ L S, β L S − ν) ,
u u
and using Eqs. (3.62b), (2.34), (3.100) and (5.158), one obtains
= hν N
ηiA (ν, Ω) (2L + 1) B(β L S → βu Lu S)
4π
β L Sβu Lu
× (−1)1+Ju +J +K (2Ju + 1) 3(2J + 1)(2J + 1)
KQ J J Ju
! ! !
Lu L 1 Lu L 1 1 1 K
×
J Ju S J Ju S J J Ju
× TQ
K
(i, Ω) ρQ (J , J )
β L S K
φ(νβ L S, β L S − ν) (7.48a)
u u
or, performing the summation over Ju via Eq. (2.41)
hν
=
ηiA (ν, Ω) N (2L + 1) B(β L S → βu Lu S)
4π
β L Sβu Lu
× (−1)1−Lu +S+J 3(2J + 1)(2J + 1)
KQ J J
! !
L L K 1 1 K
×
J J S L L Lu
× TQ
K
(i, Ω) ρQ (J , J )
β L S K
φ(νβ L S, β L S − ν) . (7.48b)
u u
Similarly
= hν N
ηiS (ν, Ω) (2Lu + 1) B(βu Lu S → β L S)
4π
βu Lu Sβ L
× (−1)1+J +Ju (2J + 1) 3(2Ju + 1)(2Ju + 1)
KQ Ju Ju J
! ! !
Lu L 1 Lu L 1 1 1 K
×
J Ju S J Ju S Ju Ju J
× TQ
K
(i, Ω) ρQ (Ju , Ju )
βu Lu S K
φ(νβ L S, β L S − ν) (7.48c)
u u
EQUATIONS FOR ATOMIC SYSTEMS 311
or, performing the summation over J via Eq. (2.41)
= hν N
ηiS (ν, Ω) (2Lu + 1) B(βu Lu S → β L S)
4π
βu Lu Sβ L
× (−1)1−L +S+Ju +K 3(2Ju + 1)(2Ju + 1)
KQ Ju Ju
! !
Lu Lu K 1 1 K
×
Ju Ju S Lu Lu L
× TQ
K
(i, Ω) ρQ (Ju , Ju )
βu Lu S K
φ(νβ L S, β L S − ν) . (7.48d)
u u
The same calculations show that
& '
= η A (ν, Ω)
φ(ν
ρAi (ν, Ω) i βu Lu S, β L S − ν) → ψ(νβu Lu S, β L S − ν) (7.48e)
& '
= η S (ν, Ω)
ρSi (ν, Ω) φ(ν − ν) → ψ(ν − ν) (7.48f)
i β L S, β L S
u u
β L S, β L S u u
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.48g)
i
c2
7.7. Conjugation Properties of the Rates
The radiative rates appearing in the statistical equilibrium equations for the multi-
term atom have simple conjugation properties. In the energy-eigenvector repre-
sentation, we have from Eqs. (7.34), using Eq. (5.154) and the reality of the CJj
coefficients (see Sect. 3.4)
TA (βLSjM j M , β L Sj M j M )∗ = TA (βLSj M jM, β L Sj M j M )
TE (βLSjM j M , βu Lu Sju Mu ju Mu )∗ = TE (βLSj M jM, βu Lu Sju Mu ju Mu ) ,
with an analogous relation for TS ;
RA (βLSjM j M )∗ = RA (βLSj M jM ) ,
with analogous relations for RE and RS ; moreover, the rates TE and RE are real.
In the spherical statistical tensor representation, we have from Eqs. (7.45a-c) and
(7.46a-c), with the help of Eqs. (5.158)1
TA (βLSKQJJ , β L SK Q J J )∗ =
+Q+J −J +Q
= (−1)J−J TA (βLSK −QJ J, β L SK −Q J J )
1
To arrange the sign factors in these equations, one should bear in mind that the quantum
numbers J, J , J , J are all integers or all half-integers, and that the indices K, Q, K , Q K ,
Q , Ku , Qu are integers (see footnote on p.123).
312 CHAPTER 7
TE (βLSKQJJ , βu Lu SKu Qu Ju Ju )∗ =
+Q+Ju −Ju +Qu
= (−1)J−J TE (βLSK −QJ J, βu Lu SKu −Qu Ju Ju ) ,
with an analogous relation for TS ;
RA (βLSKQJJ K Q J J )∗ =
+Q+J −J +Q
= (−1)J−J RA (βLSK −QJ JK −Q J J ) ,
with analogous relations for RE and RS ; moreover, the rates TE and RE are real.
7.8. The Multi-Level Atom as a Special Case of the Multi-Term Atom
As apparent from Sects. 7.5 and 7.6, the description of the multi-term atom requires
much more complicated equations than the multi-level atom treated previously:
this is basically related to the fact that two types of coherences (between different
J-levels and between magnetic sublevels of each J-level) are considered in this
model. It can be easily realized that if we neglect coherences between J-levels,
retaining only those between magnetic sublevels of individual J-levels, the case of
the multi-term atom becomes very similar to the case of the multi-level atom. In
fact, after the formal identifications
α ≡ (βLS) , α ≡ (β L S) , αu ≡ (βu Lu S) , (7.49)
the statistical equilibrium equations for the two cases must indeed coincide provided
the incident radiation field is such that
K
JQ (να J , α J ) ≡ JQ
K
(νβ L SJu , β L SJ )
K
= JQ (νβ L S, β L S ) (7.50)
u u u u u u
for each multiplet. The radiative transfer coefficients for the two cases must also
coincide provided the frequency splittings of the Φ profiles due to fine structure are
neglected,
Φ(να J , α J − ν) ≡ Φ(νβ L SJu , β L SJ − ν) = Φ(νβ L S, β L S − ν) . (7.51)
u u u u u u
To check this property – which is a proof of consistency of the formalism – we
substitute into Eq. (7.38)
ρQ (J, J )
βLS K
= δJJ βLS K
ρQ (J, J) = δJJ βLS K
ρQ (J) , (7.52)
and we get
EQUATIONS FOR ATOMIC SYSTEMS 313
d
βLS K
βLS K
ρQ (J) = −2πi NβLS (KQJJ, K Q J J ) ρQ (J )
dt
K Q J
β L S K
+ ρQ (J ) TA (βLSKQJJ, β L SK Q J J )
β L K Q J
βu Lu S Ku
+ ρQ (Ju ) TE (βLSKQJJ, βu Lu SKu Qu Ju Ju )
u
βu Lu Ku Qu Ju
+ TS (βLSKQJJ, βu Lu SKu Qu Ju Ju )
βLS K
− ρQ (J ) RA (βLSKQJJK Q J J ) + RE (βLSKQJJK Q J J )
K Q J
+ RS (βLSKQJJK Q J J ) . (7.53)
From Eqs. (7.41) and (7.44) we have
NβLS (KQJJ, K Q J J ) =
K K 1
= δQQ νL (−1) 2J−Q
(2K + 1)(2K + 1)
−Q Q 0
!
K−K K K 1
× δJJ J(J + 1)(2J + 1) 1 + (−1) gLS (J) ,
J J J
which can be transformed using some Racah algebra. Because of the triangular
condition (K, K , 1) we can write
1 + (−1)K−K = 2 δKK ,
and using Eqs. (2.26d) and (2.36d) we obtain
NβLS (KQJJ, K Q J J ) = δKK δQQ δJJ νL Q gLS (J) , (7.54)
which makes the first term in the right-hand side of Eq. (7.53) equal to the corre-
sponding term in Eq. (7.11).1
As far as the other terms are concerned, it is readily seen that Eq. (7.53) – with
the identifications (7.49) – has the same form as Eq. (7.11), thus we have just to
compare the expressions for the rates. From Eq. (7.45a) we have
TA (βLSKQJJ, βL SK Q J J ) = (2L + 1) B(β L S → βLS)
× (2J + 1) (2J + 1) 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ J J 1 ⎬ !2
L L 1 K K Kr
× (−1)K +Q J J 1
⎩ ⎭ J J S −Q Q −Qr
K K Kr
K
× JQ r (νβLS, β L S ) , (7.55)
r
1
Of course, Eq. (7.54) contains the Landé factor evaluated in the L-S coupling scheme, while
Eq. (7.11) contains the more general Landé factor defined in Eq. (3.7).
314 CHAPTER 7
which should be compared with Eq. (7.14a). To this end we need the relation be-
tween the Einstein coefficient for the transition between two terms and the Einstein
coefficients for the transition between two individual J-levels.
Equation (7.6) with substitutions (7.49) gives
S(β L SJ , βu Lu SJu ) = (2J + 1) | β L SJ d βu Lu SJu |2 ,
and using Eqs. (2.108) and (7.31) we obtain the relation connecting the strengths1
!2
Lu L 1
S(β L SJ , βu Lu SJu ) = (2J + 1) (2Ju + 1)
J Ju S
× S(β L S, βu Lu S) . (7.56)
Finally, from Eqs. (7.7) and (7.32) we get
!2
Lu L 1
B(β L SJ → βu Lu SJu ) = (2L + 1) (2Ju + 1)
J Ju S
× B(β L S → βu Lu S) . (7.57a)
Substitution into Eq. (7.55) shows that the expressions in the right-hand side of
Eqs. (7.55) and (7.14a) are identical when approximation (7.50) is satisfied.
Quite similar calculations can be performed for the other rates. Using the further
relations between Einstein coefficients2
!2
Lu L 1
B(βu Lu SJu → β L SJ ) = (2Lu + 1) (2J + 1)
J Ju S
× B(βu Lu S → β L S) (7.57b)
1
From Eq. (7.56) we have, with the help of Eq. (2.39)
%
J Ju
S(β L SJ , βu Lu SJu ) = (2S + 1) S(β L S, βu Lu S) ,
therefore the relative strengths of the multiplet, normalized to unity, are given by
!2
(2J + 1)(2Ju + 1) Lu L 1
S J ,Ju = ,
2S + 1 J Ju S
which is just Eq. (3.65).
2
Note that Eq. (7.57c) implies the approximation
νβ3 L SJu , β L SJ / νβ3 L S, β L S =1.
u u u u
Note also that the Einstein coefficients in Eqs. (7.57) are related by the equations
%
J
A(βu Lu SJu → β L SJ ) = A(βu Lu S → β L S)
%
B(βu Lu SJu → β L SJ ) = B(βu Lu S → β L S)
%
J
Ju
B(β L SJ → βu Lu SJu ) = B(β L S → βu Lu S) ,
which are easily derived with the use of Eq. (2.39).
EQUATIONS FOR ATOMIC SYSTEMS 315
!2
Lu L 1
A(βu Lu SJu → β L SJ ) = (2Lu + 1) (2J + 1)
J Ju S
× A(βu Lu S → β L S) , (7.57c)
one finds that all the rates in Eq. (7.53) coincide with the corresponding rates of
Eqs. (7.14), which proves that the statistical equilibrium equations for the multi-
term atom reduce to the equations for the multi-level atom under the assumptions
mentioned above.
The same check can be repeated for the radiative transfer coefficients. Substitu-
tion of Eq. (7.52) into Eqs. (7.48) and use of approximation (7.51) yields exactly
the same expressions derived for the multi-level atom (Eqs. (7.16)).
7.9. The Multi-Level Atom with Hyperfine Structure
In the preceding sections of this chapter we have developed the basic equations for
a standard atom devoid of hyperfine structure. Now we are going to generalize the
description of the atomic system by including this further degree of freedom.
If I is the nuclear spin of a given atomic species, the energy eigenvectors – in the
absence of magnetic fields – have the form |αJIF f (cf. Sect. 3.5 for the meaning
of symbols). In principle, one could develop the statistical equilibrium equations
and the radiative transfer equations for the general case implying coherences of the
form
αJIF f | ρ |αJ IF f .
However, with the exception of the hydrogen atom (which presents, moreover, an
additional degeneracy of the eigenvalues with respect to the azimuthal quantum
number l, and therefore requires a more involved formalism), it is in most cases
sufficient to restrict the description of the atom with hyperfine structure to the
J-diagonal density-matrix elements
αJIF f | ρ |αJIF f = ραJI (F f, F f ) .
The model atom described by this approximation will be called in the following
the multi-level atom with hyperfine structure. For such system, the flat-spectrum
approximation (cf. Sect. 6.5) implies that the radiation field incident on the atom
should be constant across a frequency interval wider than the frequency shifts and
inverse lifetimes of the hyperfine-structure sublevels involved. This means that the
radiation field can be characterized by specifying its value at the Bohr frequencies
EαJ − Eα J
ναJ, α J = ,
h
where EαJ , Eα J are the energies of the relevant levels (which disregard both the
hyperfine-structure and the magnetic Hamiltonian). This approximation is usually
well-satisfied in astrophysical plasmas.
316 CHAPTER 7
We now apply the equations derived in Chap. 6 to such an atom embedded in
a magnetic field. If we choose a reference system with the z-axis pointing in the
magnetic field direction, we have for the eigenvectors and eigenvalues of the atomic
Hamiltonian
HA |αJIif = EαJ + λi (αJI, f ) |αJIif ,
where (see Eq. (3.74))
|αJIif = CFi (αJI, f ) |αJIF f . (7.58)
F
Thus we modify the formalism of Chap. 6 according to the following substitutions
i) for the energy eigenvectors:
| n → |αJIif ; (7.59)
ii) for the corresponding energy eigenvalues:
En → EαJ + λi (αJI, f ) , (7.60)
whence
iii) for the Bohr frequencies:
En − Em
νnm = → ναJIif, α J Ii f =
h
λ (αJI, f ) − λi (α J I, f )
= ναJ, α J + i ; (7.61)
h
iv) for the matrix elements of the spherical components of the dipole operator (see
the analogous derivation of Eq. (7.27)):
(dq )nm = n | dq | m → (dq )αJIif, α J Ii f = αJIif | dq |α J Ii f =
= CFi (αJI, f ) CFi (α J I, f )
FF
× (−1)J+I−f (2J + 1)(2F + 1)(2F + 1)
!
F F 1 J J 1
× αJ d α J ; (7.62)
−f f q F F I
v) finally, for the atomic density-matrix elements:
ρnm = n | ρ | m → αJIif | ρ |α J Ii f =
= δαα δJJ αJIif | ρ |αJIi f = δαα δJJ ραJI (if, i f ) . (7.63)
EQUATIONS FOR ATOMIC SYSTEMS 317
If we now compare Eqs. (7.59)-(7.63) with Eqs. (7.24)-(7.28), we see that the
equations for the multi-level atom with hyperfine structure can be deduced from
the corresponding equations of the multi-term atom by means of the formal sub-
stitutions
β→α S→I j→i
L→J J →F M →f . (7.64)
Thus the statistical equilibrium equations and the radiative transfer coefficients for
the multi-level atom with hyperfine structure, expressed in the energy-eigenvector
representation, are simply given by Eqs. (7.29),1 (7.34) and (7.35) with the formal
substitutions (7.64).
The conversion to the spherical statistical tensor representation is also straight-
forward. Using Eqs. (7.58) and (3.103) we obtain the relation
ραJI (if, i f ) = CFi (αJI, f ) CFi (αJI, f ) ραJI (F f, F f )
FF
= CFi (αJI, f ) CFi (αJI, f )
FF
F −f
√ F F K
× (−1) 2K + 1 ρQ (F, F )
αJI K
,
f −f −Q
KQ
which is just the same as Eq. (7.36) with substitutions (7.64). It follows that the
equations for the multi-level atom with hyperfine structure, expressed in the spher-
ical statistical tensor representation, can be easily deduced from the corresponding
equations for the multi-term atom by performing the same substitutions.
For the statistical equilibrium equations we have (cf. Eq. (7.38))
d
ρQ (F, F )
αJI K
= −2πi ÑαJI (KQF F , K Q F F ) αJIρK
Q (F , F )
dt
K Q F F
α J I K
+ ρQ (F , F ) TA (αJIKQF F , α J IK Q F F )
α J K Q F F
αu Ju I Ku
+ ρQ (Fu , Fu ) TE (αJIKQF F , αu Ju IKu Qu Fu Fu )
u
αu Ju Ku Qu Fu Fu
+ TS (αJIKQF F , αu Ju IKu Qu Fu Fu )
αJI K
− ρQ (F , F ) RA (αJIKQF F K Q F F )
K Q F F
+ RE (αJIKQF F K Q F F ) + RS (αJIKQF F K Q F F ) . (7.65)
1
The term in Eq. (7.30) is discussed later in greater detail.
318 CHAPTER 7
The only term which cannot be deduced by direct substitution of Eqs. (7.64) is
the first term in the right-hand side, as expected because of the intrinsic difference
between the fine-structure and the hyperfine-structure Hamiltonians. To evaluate
this term we must go back to Eqs. (7.39) and (7.30). Applying substitutions (7.64)
and using Eqs. (3.76b,c) we get
ÑαJI (KQF F , K Q F F ) = δQQ (−1)F −F (2K + 1)(2K + 1)
1 F F K
F F K
× δF F
h
f −f −Q f −f −Q
ff
× αJIF f |Hhfs
(1) (2)
+ Hhfs + HB |αJIF f
F F K
F F K
− δF F
f −f −Q f −f −Q
ff
× αJIF f |Hhfs
(1) (2)
+ Hhfs + HB |αJIF f ,
(1) (2)
with the Hamiltonians Hhfs and Hhfs given by Eqs. (3.70) and (3.71). The matrix
elements in the right-hand side can be evaluated using Eqs. (3.70) and (3.72).
Carrying out the summations over f and f via Eqs. (2.23a) and (2.42) one obtains
ÑαJI (KQF F , K Q F F ) = δKK δQQ δF F δF F ναJIF, αJIF
F +F +Q
K K 1
+ δQQ νL gαJ (−1)
(2K + 1)(2K + 1)
−Q Q 0
!
K K 1
× δF F ΓJI (F, F )
F F F
!
K K 1
+ δF F (−1)K−K ΓJI (F , F ) , (7.66)
F F F
where
EαJI (F ) − EαJI (F )
ναJIF, αJIF =
h
1
= αJIF f |Hhfs
(1) (2)
+ Hhfs |αJIF f − αJIF f |Hhfs
(1) (2)
+ Hhfs |αJIF f (7.67)
h
and where we have introduced a generalized Landé factor for hyperfine structure
through the position1
1
A slightly different generalized Landé factor has been introduced by Landolfi and Landi
Degl’Innocenti (1985). Note the similarity of Eq. (7.68) with Eq. (7.42). Note also that
ΓJ I (F, F ) = F (F + 1)(2F + 1) ghfs (F ) ,
where ghfs (F ) is given in Eq. (3.73).
EQUATIONS FOR ATOMIC SYSTEMS 319
ΓJI (F, F ) = (−1)F −F ΓJI (F , F )
!
F F 1
= (−1)1+J+I+F J(J + 1)(2J + 1)(2F + 1)(2F + 1) . (7.68)
J J I
All the other terms in Eq. (7.65) can be derived by carrying out the substitu-
tions (7.64). We write down here, for future reference, the expressions for the
radiative rates (cf. Eqs. (7.45a-c) and (7.46a-c))
TA (αJIKQF F , α J IK Q F F ) = (2J + 1) B(α J → αJ)
× 3(2F + 1)(2F + 1)(2F + 1)(2F + 1)(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ F F 1 ⎬ !
K +Q +F −F J J 1
× (−1) F F 1
⎩ ⎭ F F I
K K Kr
!
J J 1 K K Kr K
× JQ r (ναJ, α J ) (7.69a)
F F I −Q Q −Qr r
TE (αJIKQF F , αu Ju IKu Qu Fu Fu ) = δKK δQQ (2Ju + 1) A(αu Ju → αJ)
u u
× (2F + 1)(2F + 1)(2Fu + 1)(2Fu + 1)
! ! !
1+K+F +Fu F F K Ju J 1 Ju J 1
× (−1) (7.69b)
Fu Fu 1 F Fu I F Fu I
TS (αJIKQF F , αu Ju IKu Qu Fu Fu ) = (2Ju + 1) B(αu Ju → αJ)
× 3(2F + 1)(2F + 1)(2Fu + 1)(2Fu + 1)(2K + 1)(2Ku + 1)(2Kr + 1)
Kr Qr ⎧ ⎫
⎨ F Fu 1 ⎬ !
Kr +Ku +Qu +Fu −Fu Ju J 1
× (−1) F Fu 1
⎩ ⎭ F Fu I
K Ku Kr
!
Ju J 1 K Ku Kr K
× JQ r (να J , αJ ) (7.69c)
F Fu I −Q Qu −Qr r u u
RA (αJIKQF F K Q F F ) = (2J + 1) B(αJ → αu Ju )
αu Ju
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
!
1+Ju −I+F +Q J J Kr K K Kr
× (−1) ×
1 1 Ju Q −Q Qr
320 CHAPTER 7
1
× δF F (2F + 1)(2F + 1)
2
! !
J J Kr K K Kr
×
F F I F F F
+ δF F (2F + 1)(2F + 1) (−1)F −F +K+K +Kr
! !
J J Kr K K Kr K
× JQ r (να J , αJ ) (7.69d)
F F I F F F r u u
RE (αJIKQF F K Q F F ) = δKK δQQ δF F δF F
× A(αJ → α J ) (7.69e)
α J
RS (αJIKQF F K Q F F ) = (2J + 1) B(αJ → α J )
α J
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
!
1+J −I+F +Kr +Q J J Kr K K Kr
× (−1)
1 1 J Q −Q Qr
1
× δF F (2F + 1)(2F + 1)
2
! !
J J Kr K K Kr
×
F F I F F F
+ δF F (2F + 1)(2F + 1) (−1)F −F +K+K +Kr
! !
J J Kr K K Kr K
× JQ r (ναJ, α ). (7.69f)
F F I F F F r
J
Finally, the radiative transfer coefficients can be obtained by applying the same
substitutions to Eqs. (7.47). In the special case where all the profiles of the various
hyperfine components can be considered coincident, we have from Eqs. (7.48)
hν
=
ηiA (ν, Ω) N (2J + 1) B(α J → αu Ju )
4π
α J αu Ju
× (−1)1−Ju +I+F 3(2F + 1)(2F + 1) ×
KQ F F
EQUATIONS FOR ATOMIC SYSTEMS 321
! !
J J K 1 1 K
×
F F I J J Ju
× TQ
K
(i, Ω) ρQ (F , F )
α J I K
φ(να J , α J − ν) (7.70a)
u u
= hν N
ηiS (ν, Ω) (2Ju + 1) B(αu Ju → α J )
4π
αu Ju α J
× (−1)1−J +I+Fu +K 3(2Fu + 1)(2Fu + 1)
KQ Fu Fu
! !
Ju Ju K 1 1 K
×
Fu Fu I Ju Ju J
× TQ
K
(i, Ω) ρQ (Fu , Fu )
αu Ju I K
φ(να J , α J − ν) (7.70b)
u u
& '
= η A (ν, Ω)
ρAi (ν, Ω) φ(ν − ν) → ψ(να − ν) (7.70c)
i α J , α J
u u
J , α J
u u
& '
= η S (ν, Ω)
ρSi (ν, Ω) φ(ν − ν) → ψ(να − ν) (7.70d)
i α J , α J
u u
J , α J
u u
3
= 2hν η S (ν, Ω)
εi (ν, Ω) . (7.70e)
i
c2
7.10. The Principle of Spectroscopic Stability
The equations obtained in the previous section for the multi-level atom with hy-
perfine structure must be compatible with the corresponding equations derived in
Sect. 7.2 for the multi-level atom without hyperfine structure. By this we mean
that when hyperfine structure has a negligible effect (strictly speaking, when the
two constants A and B in Eqs. (3.70) are zero), it should be possible to get back
the equations of Sect. 7.2 from those of Sect. 7.9. This is what we are going to
prove here, thus obtaining – in a particular case – a confirmation of the principle
of spectroscopic stability. Among the several possible formulations of this princi-
ple, the most satisfactory is the following: ‘If two different descriptions are used
to characterize a quantum system – a detailed description which takes an inner
quantum number into account and a simplified description which disregards it –
the predicted results must be the same in all physical experiments where the struc-
ture described by the inner quantum number is unimportant’. In the particular
case that we are considering here, the inner quantum numbers are the nuclear spin
quantum numbers (I, MI ).
We will now derive from Eq. (7.65) the corresponding equation for the multipole
moments ρK Q (αJ) describing the atom irrespective of its hyperfine structure.
Equation (3.104) shows that the time evolution of ρK Q (αJ) can be obtained by
322 CHAPTER 7
multiplication of Eq. (7.65) by the factor
!
J+I+F +K
F F K
(−1) (2F + 1)(2F + 1)
J J I
followed by summation over F and F . We thus obtain
!
d K J+I+F +K
F F K
ρQ (αJ) = (−1)
(2F + 1)(2F + 1)
dt
J J I
FF
× Right-Hand Side of Eq. (7.65) . (7.71)
Now we perform the summation over F and F for the various terms in the right-
hand side of Eq. (7.65). For the term containing TA we obtain from Eq. (7.69a)
the following expression
!
F F K
(−1)J+I+F +K (2F + 1)(2F + 1)
J J I
FF
α J I K
× ρQ (F , F ) (2J + 1) B(α J → αJ)
α J K Q F F
× 3(2F + 1)(2F + 1)(2F + 1)(2F + 1)(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
⎧ ⎫
⎨ F F 1 ⎬ ! !
K +Q +F −F J J 1 J J 1
× (−1) F F 1
⎩ ⎭ F F I F F I
K K Kr
K K Kr K
× JQ r (ναJ, α J ) .
−Q Q −Qr r
The summation over F and F can be performed with the help of Eq. (2.56), and
the expression above reduces to
!
F F K
(−1)J +I+F +K (2F + 1)(2F + 1)
J J I
α J K Q F F
α J I K
× ρQ (F , F ) (2J + 1) B(α J → αJ)
⎧ ⎫
⎨J J 1 ⎬
× 3(2K + 1)(2K + 1)(2Kr + 1) (−1)K +Q J J 1
⎩ ⎭
Kr Qr K K Kr
K K Kr K
× JQ r (ναJ, α J )
−Q Q −Qr r
which, using again Eq. (3.104), can be cast into the form
K
ρQ (α J ) TA (αJKQ, α J K Q ) ,
α J K Q
EQUATIONS FOR ATOMIC SYSTEMS 323
where TA (αJKQ, α J K Q ) is the rate defined in Eq. (7.14a).
By similar calculations – which involve the use of Eqs. (2.48), (2.55), (2.56) and
(2.41) – we obtain from Eq. (7.71)
!
d K F F K
ρQ (αJ) = (−1)J+I+F +K (2F + 1)(2F + 1)
dt J J I
FF
×(−2πi) ÑαJI (KQF F , K Q F F ) αJIρK
Q (F , F )
K Q F F
K
+ ρQ (α J ) TA (αJKQ, α J K Q )
α J K Q
K
+ ρQu (αu Ju ) TE (αJKQ, αu Ju Ku Qu ) + TS (αJKQ, αu Ju Ku Qu )
u
αu Ju Ku Qu
− Q (αJ) RA (αJKQK Q ) + RE (αJKQK Q ) + RS (αJKQK Q ) ,
ρK
K Q
where all the rates are exactly the same as those of Eqs. (7.14).
Finally, we evaluate the first term in the right-hand side using the expression
for Ñ given in Eq. (7.66). With the help of Eqs. (2.41), (2.26d) and (2.36d) we get
!
d K F F K
ρQ (αJ) = −2πi (−1)J+I+F +K (2F + 1)(2F + 1)
dt
J J I
FF
× ναJIF, αJIF ρQ (F, F )
αJI K
− 2πi νL gαJ Q ρK
Q (αJ)
K
+ ρQ (α J ) TA (αJKQ, α J K Q )
α J K Q
K
+ ρQu (αu Ju ) TE (αJKQ, αu Ju Ku Qu ) + TS (αJKQ, αu Ju Ku Qu )
u
αu Ju Ku Qu
− Q (αJ) RA (αJKQK Q ) + RE (αJKQK Q )
ρK
K Q
+ RS (αJKQK Q ) . (7.72)
When hyperfine structure is negligible, the Bohr frequencies ναJIF, αJIF are zero
(cf. Eq. (7.67)), and Eqs. (7.72) reduce to the statistical equilibrium equations
for the multi-level atom without hyperfine structure (Eq. (7.11)) – which confirms
the principle of spectroscopic stability. Equation (7.72) also shows that the ‘in-
ner’ description of the atomic system involving hyperfine structure can be avoided
whenever the frequency shifts among the F -sublevels of each (αJ)-level are negli-
gible in comparison with the relaxation rates of the same level (roughly speaking,
324 CHAPTER 7
when they are much smaller than the inverse lifetime of the level). This provides
a quantitative criterion to ascertain whether hyperfine structure can be neglected
in a real physical problem.
As far as the radiative transfer coefficients are concerned, a direct application of
Eq. (3.104) shows that Eqs. (7.70), which have just been obtained by neglecting
the frequency shifts between hyperfine components, are nothing but Eqs. (7.16)
in
written in a different form. Referring for instance to the expression for ηiA (ν, Ω)
Eq. (7.70a), and recalling that, according to Eq. (3.104)
ρK
Q (α J ) =
!
J +I+F +K F F K
= (−1) (2F + 1)(2F + 1) ρQ (F , F )
α J I K
,
J J I
F F
we obtain
hν
=
ηiA (ν, Ω) N (2J + 1) B(α J → αu Ju )
4π
α J αu Ju
√ !
1−J −Ju −K 1 1 K ρK (α J )
× 3 (−1) TQ
K
(i, Ω) Q
J J Ju
KQ
× φ(να J , α J − ν) ,
u u
which coincides with the corresponding expression in Eq. (7.16a). This is a further
aspect of the principle of spectroscopic stability.
The same reasoning followed about hyperfine structure can be repeated for fine
structure, which is – in terms of the principle of spectroscopic stability – the ‘inner’
structure of the atom due to the electronic spin. According to this principle, the
equations describing the multi-term atom should exactly reduce to the equations
for the ‘spinless atom’ when the effects due to the presence of spin are negligible.
To check this property we start from Eq. (7.38), with N given by Eqs. (7.41)-
(7.43) and the rates by Eqs. (7.45a-c) and (7.46a-c), and – bearing in mind Eq.
(3.105) – we multiply both sides by the factor
!
L+S+J +K
J J K
(−1)
(2J + 1)(2J + 1)
L L S
and sum over J and J . Neglecting the fine-structure shifts νβLSJ, βLSJ in the
final result, we obtain the equation1
1
The first term in the right-hand side can be deduced using the Racah-algebra relation
! !
a b 1 a b 1
a−b
c (c + 1)(2c + 1) + (−1) d (d + 1)(2d + 1) =
c c d d d c
= δab (−1)a+c+d+1 a (a + 1) 2a + 1 ,
which follows from Eqs. (2.36d) and (2.36f).
EQUATIONS FOR ATOMIC SYSTEMS 325
d K
ρ (βL) = −2πi νL Q ρK Q (βL)
dt Q
K
+ ρQ (β L ) TA (βLKQ, β L K Q )
β L K Q
K
+ ρQu (βu Lu ) TE (βLKQ, βu Lu Ku Qu ) + TS (βLKQ, βu Lu Ku Qu )
u
βu Lu Ku Qu
− Q (βL) RA (βLKQK Q ) + RE (βLKQK Q )
ρK
K Q
+ RS (βLKQK Q ) , (7.73)
where all the rates are the same as those in Eqs. (7.14) with the formal substitutions
α→β α → β αu → βu
J →L J → L Ju → Lu . (7.74)
It is easily seen that, performing these substitutions, Eq. (7.73) is identical to
Eq. (7.11) except for the presence, in the latter equation, of the Landé factor gαJ ,
which is however equal to unity when spin is neglected.
Similarly, it can be seen using Eq. (3.105) that the expressions (7.48) for the ra-
diative transfer coefficients (which have been obtained by neglecting the frequency
shifts due to fine structure) are identical to Eqs. (7.16) with substitutions (7.74).
7.11. Selection Rules
All the expressions derived in the preceding sections for the radiative rates and for
the radiative transfer coefficients involve some 3-j, or 6-j, or 9-j symbols which
are responsible for the appearance of several selection rules. These rules originate
from the electric-dipole approximation and can be considered as an obvious gener-
alization of the selection rules that are usually met in conventional spectroscopy.
In the following we list the selection rules for the radiative rates written in the
spherical statistical tensor representation. The list is divided into three parts,
which refer to the different atomic models considered previously. We recall that
– as apparent from Eqs. (5.153) and (5.155) – the only non-zero components of the
K
radiation field tensor JQ r are those with Kr = 0, 1, 2.
r
326 CHAPTER 7
1) Multi-Level Atom (Eqs. (7.14))1
a) Transfer rate due to absorption from lower levels,
TA (αJKQ, α J K Q ) :
∆J = J − J = 0, ±1; 00
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±2; 00 for Kr = 2
b) Transfer rate due to spontaneous emission from upper levels,
TE (αJKQ, αu Ju Ku Qu ) :
∆J = Ju − J = 0, ±1; 00
∆K = Ku − K = 0
c) Transfer rate due to stimulated emission from upper levels,
TS (αJKQ, αu Ju Ku Qu ) :
∆J = Ju − J = 0, ±1; 00
⎧
⎨0 for Kr = 0
∆K = Ku − K = 0, ±1; 00 for Kr = 1
⎩
0, ±2; 00 for Kr = 2
d) Relaxation rate due to absorption towards upper levels,
RA (αJKQK Q ) :
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±2; 00 for Kr = 2
e) Relaxation rate due to spontaneous emission towards lower levels,
RE (αJKQK Q ) :
∆K = K − K = 0
f) Relaxation rate due to stimulated emission towards lower levels,
RS (αJKQK Q ) :
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±2; 00 for Kr = 2
1
Note that, for Kr = 2, the values ∆K = ±1 are forbidden. For TA and TS , this property
follows from the presence of the 9-j symbol (cf. footnote on p.287); for RA and RS , it follows
from the presence of the factor ζ+ .
EQUATIONS FOR ATOMIC SYSTEMS 327
2) Multi-Term Atom (Eqs. (7.45a-c) and (7.46a-c))
a) Transfer rate due to absorption from lower terms,
TA (βLSKQJJ , β L SK Q J J ) :
∆L = L − L = 0, ±1; 00
∆J = J − J = 0, ±1; 00
∆J = J − J = 0, ±1; 00
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
b) Transfer rate due to spontaneous emission from upper terms,
TE (βLSKQJJ , βu Lu SKu Qu Ju Ju ) :
∆L = Lu − L = 0, ±1; 00
∆J = Ju − J = 0, ±1; 00
∆J = Ju − J = 0, ±1; 00
∆K = Ku − K = 0
c) Transfer rate due to stimulated emission from upper terms,
TS (βLSKQJJ , βu Lu SKu Qu Ju Ju ) :
∆L = Lu − L = 0, ±1; 00
∆J = Ju − J = 0, ±1; 00
∆J = Ju − J = 0, ±1; 00
⎧
⎨0 for Kr = 0
∆K = Ku − K = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
d) Relaxation rate due to absorption towards upper terms,
RA (βLSKQJJ K Q J J ) :
⎧
⎨0 for Kr = 0
∆J = J − J = 0 and ∆J = J − J = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
or
⎧
⎨0 for Kr = 0
∆J = J − J = 0 and ∆J = J − J = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2;
328 CHAPTER 7
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
e) Relaxation rate due to spontaneous emission towards lower terms,
RE (βLSKQJJ K Q J J ) :
∆J = J − J = 0
∆J = J − J = 0
∆K = K − K = 0
f) Relaxation rate due to stimulated emission towards lower terms,
RS (βLSKQJJ K Q J J ) :
⎧
⎨0 for Kr = 0
∆J = J − J = 0 and ∆J = J − J = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
or
⎧
⎨0 for Kr = 0
∆J = J − J = 0 and ∆J = J − J = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2;
⎧
⎨0 for Kr = 0
∆K = K − K = 0, ±1; 00 for Kr = 1
⎩
0, ±1, ±2; 00, 01, 10 for Kr = 2
3) Multi-Level Atom with Hyperfine Structure (Eqs. (7.69))
The selection rules are here strictly analogous to those for the multi-term atom;
they can be simply derived by substituting L → J, J → F .
7.12. Changing the Reference System
All the equations of the former sections have been derived in a reference system
having its z-axis (i.e. the quantization axis for angular momentum) in the magnetic
field direction – obviously, the z-axis direction is arbitrary for zero magnetic field.
It is now important to discuss how these equations are modified when a different
reference system is chosen.
EQUATIONS FOR ATOMIC SYSTEMS 329
Consider first the statistical equilibrium equations. The procedure to obtain
these equations in a reference system other than the ‘magnetic’ reference system
(the one with the z-axis parallel to the magnetic field) is straightforward. One
has just to consider the equations written in the ‘old’ system and to express the
density-matrix elements as functions of the density-matrix elements in the ‘new’
system.
This procedure can be formalized as follows: we introduce a formal vector ρ ,
having dimension N , whose N components are the density-matrix elements of the
model atom considered, and we observe that the statistical equilibrium equations
can be written in the condensed form
d
ρ = Aρ ,
dt
where A is an N × N matrix that will be called in the following the rate ma-
trix . Since the density-matrix elements change under rotation according to linear
relations (see e.g. Eqs. (3.95) or (3.98)), we can write
ρ(new) = S ρ(old) (7.75)
with the inverse formula
ρ(old) = S −1 ρ(new) ,
where S is the N × N transformation matrix that depends on the rotation. It
follows that
d (new)
ρ = A(new) ρ(new) ,
dt
where
A(new) = S A(old) S −1 .
This is the transformation law for the rate matrix under rotation of the reference
system.
To give an example, let us consider the statistical equilibrium equations for the
multi-level atom in the spherical statistical tensor representation (Eq. (7.11)). If
R is the rotation which carries the old reference system into the new one, we have
from Eq. (3.98)
∗
ρK
Q (αJ) = ρK
Q (αJ) DQ
K
Q (R) , (7.76)
new old
Q
with the inverse transformation (that can be easily deduced using Eqs. (2.72))
ρK
Q (αJ) = ρK
Q (αJ) DQQ
K
(R) . (7.77)
old new
Q
Note that these transformations imply that the S matrix defined above is block-
diagonal, with its non-zero elements connecting only statistical tensors of the same
level and of the same rank.
330 CHAPTER 7
With the help of some algebra it is easily shown that the statistical equilibrium
equations in the new reference system are the following
d K
ρ (αJ) = −2πi νL gαJ KQQ
K
ρK
Q (αJ)
dt Q new new
Q
K
+ ρQ (α J ) TA (αJKQ, α J K Q )
new new
α J K Q
K
&
+ ρQu (αu Ju ) TE (αJKQ, αu Ju Ku Qu )
u new new
αu Ju Ku Qu
'
+ TS (αJKQ, αu Ju Ku Qu )
new
&
− ρK
Q (αJ) RA (αJKQK Q ) + RE (αJKQK Q )
new new new
K Q
'
+ RS (αJKQK Q ) , (7.78)
new
where the kernel KQQ
K
is given by
∗
KQQ
K
= DQ
K
Q (R) Q DQ
K
Q (R) (7.79)
Q
and where
TA (αJKQ, α J K Q ) =
new
K
∗
= DQK
Q (R) TA (αJKQ , α J K Q ) DQQ (R)
old
Q Q
TE,S (αJKQ, αu Ju Ku Qu ) =
new
K
∗
= DQ K
Q (R) TE,S (αJKQ , αu Ju Ku Qu ) DQu Q (R)
old u u
Q Qu
RA,E,S(αJKQK Q ) =
new
∗
= DQ
K
Q (R) RA,E,S(αJKQ K Q ) DQ
K
Q (R) . (7.80)
old
Q Q
It can be proved that the expressions for the radiative rates are formally invariant.
Consider for instance Eq. (7.14a); from Eqs. (7.80) we have
EQUATIONS FOR ATOMIC SYSTEMS 331
TA (αJKQ, α J K Q ) =
new
K
∗
= DQ
K
Q (R) D
Q Q (R) (2J + 1) B(α J → αJ)
Q Q
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
⎧ ⎫
⎨J J 1 ⎬
K +Q K K Kr K
×(−1) J J 1 JQ r (ναJ, α ) ,
⎩ ⎭ −Q Q −Qr r
J
old
K K Kr
K
where JQ r (ναJ, α J ) old is the radiation field tensor evaluated in the old reference
r
system. After some algebra, which involves the use of Eqs. (2.77), (2.23a) and
(2.73), one obtains the expression
TA (αJKQ, α J K Q ) = (2J + 1) B(α J → αJ)
new
× 3(2K + 1)(2K + 1)(2Kr + 1)
Kr Qr
⎧ ⎫
⎨ J J 1 ⎬
K K Kr K K
× (−1)K +Q
J J 1 DQ rQ (R) JQ r (ναJ, α J ) ,
⎩ ⎭ −Q Q −Q r r r r old
K K Kr
and since from Eq. (2.78)
K K K
DQ rQ (R) JQ r (ναJ, α J ) = JQr (ναJ, α J ) , (7.81)
r r r old r new
Qr
we see that the rate TA has the same expression in the old and new reference
systems.
It can be easily proved that this property holds for the other rates as well. Thus
we can state that all the radiative rates entering the statistical equilibrium equations
are invariant under rotation of the reference system.1
This result – which has been deduced here for the multi-level atom – is also valid
for the other model atoms considered in this chapter, namely the multi-term and
the multi-level atom with hyperfine structure. This means that the radiative rates
of Eqs. (7.45a-c), (7.46a-c) and (7.69) are formally invariant under rotation of the
reference system.
1
This is of course a formal invariance. The actual value of the rates TA , TS , RA , RS for an
assigned incident radiation field does depend on the reference system because the radiation field
K changes according to Eq. (7.81).
tensor JQ
332 CHAPTER 7
Moreover, this result is valid irrespective of the representation used for the den-
sity operator. We have proved it here for the equations written in the spherical
statistical tensor representation, but it is also valid in the standard representation.
The proof is of course more involved, since the transformation law of the density-
matrix elements under rotation of the reference system is more complicated than
the transformation law of multipole moments (cf. Eqs. (3.96) and (3.98)).1
It should be remarked that, in spite of the formal invariance of the radiative rates,
the statistical equilibrium equations in a reference system other than the ‘magnetic’
system are in general more involved because of the presence of the magnetic term
(Eq. (7.79)).2 This term has its simplest expression in the reference system having
the z-axis in the magnetic field direction.
The radiative transfer coefficients, like the radiative rates, have always been
expressed in the ‘magnetic’ reference system in the former sections; thus we are
again faced with the problem of finding their expressions in a different system.
Referring, for instance, to the emission coefficient εi (ν, Ω), its expression in the
‘magnetic’ reference system can be formally written as the scalar product
= C† · ρ ,
ε (ν, Ω)
i i
where Ci is a vector having dimension N . In the new reference system, one has
from Eq. (7.75)
= C † (new) · ρ(new) ,
εi (ν, Ω) i
where
† (new) † (old)
Ci = Ci S −1 .
This is the general transformation law for the vector Ci† .
To give an example, let us consider the expression of εi (ν, Ω) for the multi-level
atom in the approximation of neglecting the dependence of the Φ profiles on the
magnetic quantum numbers (Eq. (7.16e)). Recalling Eqs. (7.8) and (7.77) we can
write
= hν N
εi (ν, Ω) (2Ju + 1) A(αu Ju → α J )
4π
α J αu Ju
√ !
1 1 K
× 3 (−1) 1+J +Ju
TQ K
(i, Ω)
Ju Ju J old
KQ
× ρK
Q (αu Ju ) DQQ
K
(R) φ(να J , α J − ν) ,
new u u
Q
1
The proof is however straightforward. Considering for instance the multi-level atom, one
should start from Eq. (7.5), multiply both sides by the factor DM ∗
N (R) DM N (R) and sum over
J J
M and M to get, using Eq. (3.96) and its inverse, the statistical equilibrium equations in the
new reference system. The expressions for the rates in the old system are given by Eqs. (7.9).
The corresponding expressions in the new system can then be derived via several Racah-algebra
calculations.
2
An exception occurs when the radiation field is cylindrically symmetrical about an axis.
In that case it may be simpler to write the statistical equilibrium equations in a reference system
having the z-axis directed along this symmetry axis (see e.g. Sect. 10.8).
EQUATIONS FOR ATOMIC SYSTEMS 333
where R is the rotation which carries the ‘magnetic’ reference system into the new
one. Since from Eq. (2.78)
TQ
K
(i, Ω) DQQ
K
(R) = TQ
K
(i, Ω) ,
old new
Q
has the same expression in the old and new reference systems.
we see that εi (ν, Ω)
As this property holds for the other radiative transfer coefficients too, we conclude
that when the dependence of the Φ profiles on the magnetic quantum numbers can
be neglected, all the radiative transfer coefficients are invariant under rotation of
the reference system.1
Similar invariance properties can be proved for the multi-term atom and for the
multi-level atom with hyperfine structure (Eqs. (7.48) and (7.70)).
7.13. Collisional Rates
In the previous sections we have derived the statistical equilibrium equations for the
atomic density-matrix elements by considering only the interaction of the atomic
system with the radiation field. In real plasmas atoms are also subjected, in general,
to a wide variety of collisions both with particles of several species (like electrons,
ions, and different atoms) and – in the case of laboratory experiments – with the
walls of the cell containing the plasma.
A detailed theory of collisions is beyond the aim of this book. Here we want just
to deduce some general results about the effect of collisions on the physical state
of the atomic system under the assumption of an isotropic distribution of colliding
particles.
In Chap. 5 (see Sects. 5.14 and 5.15) we presented a thorough discussion, based on
classical physics, of the role of collisions in polarization phenomena. The distinction
there introduced between ‘exciting’ and ‘perturbing’ collisions can be transferred
to the quantum description of collisional interactions.
Let us consider a multi-level atom with energy eigenvectors of the form |αJM ,
and let us suppose, for the time being, that no magnetic field (nor any other
anisotropic agent) is present in the physical environment. The energy eigenvectors
will then be degenerate with respect to the magnetic quantum number M . We can
divide collisions into inelastic and elastic. Inelastic collisions induce transitions
between the level |αJM and an upper level |αu Ju Mu with a corresponding energy
loss of the colliding particle. Such collisions, which are the quantum analogue
1
Contrary to the radiative rates, this is a real – not just a formal – invariance. It is clear
that once the multipole moments in an assigned reference system are given, the radiative transfer
coefficients are univocally determined. If the reference system is rotated, both the multipole
moments and the components of the tensor T Q are changed, but the quantity
K (i, Ω)
%
Q
TQ ρK (αJ)
K (i, Ω)
Q
is invariant.
334 CHAPTER 7
of the (classical) ‘exciting’ collisions, have their inverse process in the so-called
superelastic collisions, which induce transitions between the level |αJM and a
lower level |α J M with a corresponding energy increase of the colliding particle.
Elastic collisions induce transitions between the level |αJM and levels of the form
|αJM or, in other words, between magnetic sublevels belonging to the same
(αJ)-level. In these collisions, which are the quantum analogue of the (classical)
‘perturbing’ collisions, the energy of the colliding particle does not change during
the interaction.
The effect of all these collisions on the physical state of the atomic system can
be described by suitable rates (which will be divided into transfer and relaxation
collisional rates, with the same meaning as the radiative rates) appearing in the
statistical equilibrium equations for the density-matrix elements of the atomic sys-
tem. We will now derive some properties of these rates which follow from general
physical principles under the hypothesis of an isotropic distribution of colliding
particles.
7.13.a Collisional Rates due to Inelastic and Superelastic Collisions
Consider first the transfer rate due to inelastic collisions which contribute to the
time evolution of a particular density-matrix element. In an assigned – although
arbitrary – reference system we can write, with self-evident notations
d
ρ (M, M ) = CI (αJM M , α J M M ) ρα (M , M ) , (7.82)
dt αJ J
α J M M
where the quantum numbers (α J ) denote any atomic level having energy lower
than the level (αJ).
We now consider a new reference system obtained from the old one by the rota-
tion R. Using Eq. (3.96) and its inverse, we have
d
ρ (M, M ) =
dt αJ new
J J
∗ ∗
= DN
J
M (R) DN M (R) DN M (R) DN M (R)
J
α J M M N N N N
!
× CI (αJN N , α J N N ) ρα J (M , M ) . (7.83)
new
On the other hand, the assumption of isotropic collisions implies that all quantiza-
tion directions are equivalent, therefore Eqs. (7.82) and (7.83) have to be identical.
It follows that the collisional rates must satisfy the relation
J J
CI (αJM M , α J M M ) = DN
J ∗
M (R) DN M (R) DN M (R) DN M (R)
J ∗
N N N N
× CI (αJN N , α J N N ) (7.84)
EQUATIONS FOR ATOMIC SYSTEMS 335
whatever the rotation R.
The right-hand side of this equation can be transformed using Eq. (2.77) and the
complex conjugate of Eq. (2.75). Writing
∗
DN
J
M (R) DN M (R) =
J
J J K J J K
= (−1)N −M (2K + 1) DPKQ (R)∗ ,
N −N P M −M Q
K
J J
DN M (R) DN M (R)∗ =
N −M J J K J J K
= (−1) (2K + 1) DPK Q (R)∗ ,
N −N P M −M Q
K
DPKQ (R)∗ DPK Q (R)∗ =
K K K K K K
= (2K + 1) DPK Q (R) , (7.85)
P P P Q Q Q
K
one gets
CI (αJM M , α J M M ) = CI (αJN N , α J N N )
N N N N
× (2K + 1)(2K + 1)(2K + 1) (−1)N −M+N −M
KK K
J J K J J K J J K
×
N −N P M −M Q N −N P
J J K K K K K K K
× DPK Q (R) .
M −M Q P P P Q Q Q
As the right-hand side must be independent of the rotation R, the index K can
only take the value K = 0. This implies K = K , P = −P , Q = −Q . We thus
obtain, with the help of Eq. (2.26a)1
.
J−M +J −M 2J + 1
CI (αJM M , α J M M ) = (−1)
2J + 1
J J K J J K
× (2K + 1) CI(K) (αJ, α J ) , (7.86)
M −M Q M −M Q
K
1
The factor
2J + 1 2J + 1
is introduced in Eq. (7.86) to get simpler relations between these rates and the usual collisional
rates connecting atomic populations (see below).
336 CHAPTER 7
where we have defined the multipole components of the inelastic collisional rates
by the equation
2J + 1
CI(K) (αJ, α J ) =
2J + 1
+J −N J J K J J K
× (−1)J−N
N −N P N −N P
N N N N P
× CI (αJN N , α J N N ) . (7.87)
Equation (7.86) shows that the transfer collisional rates due to inelastic collisions
take a quite simple form as a result of the isotropy of the colliding particles.
Some properties of the multipole components CI(K) (αJ, α J ) follow directly from
Eq. (7.87). Taking the complex conjugate of Eq. (7.82) we have, with the use of
Eq. (3.80)
CI (αJM M , α J M M )∗ = CI (αJM M, α J M M ) ,
so that, recalling the symmetry properties of the 3-j symbols
CI(K) (αJ, α J )∗ = CI(K) (αJ, α J ) ,
which shows the reality of the multipole components. Moreover, setting K = 0 in
Eq. (7.87) we obtain via Eq. (2.26a)
1
CI(0) (αJ, α J ) = CI (αJN N, α J N N ) .
2J + 1
N N
The rate CI (αJN N, α J N N ) represents, by definition of transfer rate, the proba-
bility that a collisional transition occurs from the level |α J N to the level |αJN ,
therefore it is non-negative. It follows that the zero-rank multipole component is
also non-negative,
CI(0) (αJ, α J ) ≥ 0 .
A strictly similar line of reasoning can be followed for the transfer rates due to
superelastic collisions. These rates contribute to the time evolution of a particular
density-matrix element via the equation (cf. Eq. (7.82))
d
ρ (M, M ) = CS (αJM M , αu Ju Mu Mu ) ρα (Mu , Mu ) , (7.88)
dt αJ J
u u
αu Ju Mu Mu
and can be written, under the assumption of isotropic collisions, in the form
.
J−M +Ju −Mu 2Ju + 1
CS (αJM M , αu Ju Mu Mu ) = (−1)
2J + 1
J J K Ju Ju K
× (2K + 1) CS(K) (αJ, αu Ju ) ,
M −M Q Mu −Mu Q
K
EQUATIONS FOR ATOMIC SYSTEMS 337
where
2J + 1
CS(K) (αJ, αu Ju ) =
2Ju + 1
J−N +Ju −Nu J J K Ju Ju K
× (−1)
N −N P Nu −Nu P
N N Nu Nu P
× CS (αJN N , αu Ju Nu Nu ) (7.89)
are the multipole components of the superelastic collisional rates. These quantities
are real and CS(0) (αJ, αu Ju ) is non-negative.
Let’s now discuss the relaxation rates due to inelastic and superelastic collisions.
In a given reference system, these rates contribute to the time evolution of a par-
ticular density-matrix element via an equation of the form
d
ραJ (M, M ) = − f (αJM M M ) ραJ (M, M )
dt M
+ g(αJM M M ) ραJ (M , M ) .
The conjugation properties of the density-matrix elements (Eq. (3.80)) require that
g(αJM M M ) = f (αJM M M )∗ ,
thus we can write
d 1
ραJ (M, M ) = − S(αJM M M ) ραJ (M, M )
dt
2
M
1
+ S(αJM M M )∗ ραJ (M , M ) . (7.90)
2
As above, we consider a new reference system obtained from the former by an
arbitrary rotation R. In the new system Eq. (7.90) becomes
d
ρ (M, M ) =
dt αJ new
1
∗
=− DN
J
M (R) DN M (R) S(αJN N N )
J
2
M M NN N
∗
× DN
J
M (R) DN M (R) ραJ (M , M )
J
new
1
∗ ∗
+ DN
J
M (R) DN M (R) S(αJN N N )
J
2
N N N !
∗
× DN
J
M (R) DN
J
M (R) ραJ (M , M ) . (7.91)
new
338 CHAPTER 7
Owing to the isotropy of collisions, Eqs. (7.90) and (7.91) must be identical, there-
fore
S(αJM M M ) δMM =
∗ ∗
= DN J
M (R) DN M (R) S(αJN N N ) DN M (R) DN M (R)
J J J
(7.92)
N N N
whatever the rotation R. This requires the rate S(αJN N N ) to be independent
of the quantum number N . As a consequence, we can carry out the summation
over N via Eqs. (2.72) to get
S(αJM M ) = S(αJN N ) DN
J
M (R) DN M (R)
J ∗
N N
or, using Eq. (2.77)
−M
S(αJM M ) = S(αJN N ) (−1)N
N N
J J K J J K ∗
× (2K + 1) DQQ
K
(R) .
N −N Q M −M Q
K
Since the right-hand side must be independent of the rotation R, the index K can
only take the value K = 0. This implies M = M , N = N , and using Eq. (2.26a)
one gets
1
S(αJM M ) = δM M S(αJN N ) .
2J + 1
N
Substitution into Eq. (7.90) gives
d
ρ (M, M ) = − S0 (αJ) ραJ (M, M ) ,
dt αJ
where the relaxation collisional rate
1
S0 (αJ) = Re S(αJM M )
2J + 1
M
is real and independent of the magnetic quantum numbers.
Collecting together transfer and relaxation rates, the statistical equilibrium equa-
tions for the density-matrix elements turn out to be
.
d 2J + 1
J−M +J −M
ραJ (M, M ) = (−1)
dt 2J + 1
α J M M K
J J K J J K
× (2K + 1)
M −M Q M −M Q
× CI(K) (αJ, α J ) ρα J (M , M ) +
EQUATIONS FOR ATOMIC SYSTEMS 339
.
2Ju + 1
J−M +Ju −Mu
+ (−1)
2J + 1
αu Ju Mu Mu K
J J K Ju Ju K
× (2K + 1)
M −M Q Mu −Mu Q
× CS(K) (αJ, αu Ju ) ρα J (Mu , Mu )
u u
− S0 (αJ) ραJ (M, M ) . (7.93)
The conservation of the total number of atoms implies that the quantities CI(K) ,
(K)
CS and S0 are not independent. Denoting by nαJ the population of the level
(αJ), we require that (cf. Eq. (3.89))
d d
nαJ = ραJ (M, M ) = 0 . (7.94)
dt dt
αJ αJ M
Using Eq. (7.93) and the relation (see Eqs. (2.23a) and (2.26a))
J−M J J K
(−1) =
M −M Q
M
√ J J K
J J 0
√
= 2J + 1 = 2J + 1 δK0 δQ0 , (7.95)
M −M Q M −M 0
M
we obtain
d (0) (0)
nαJ = CI (αJ, α J ) nα J + CS (αJ, αu Ju ) nα J
dt u u
α J αu Ju
− S0 (αJ) nαJ , (7.96)
thus the condition (7.94) becomes
CI(0) (αJ, α J ) nα J + CS(0) (αJ, αu Ju ) nα J =
u u
αJ α J < αJ αJ αu Ju > αJ
= S0 (αJ) nαJ .
αJ
In the first term we replace (αJ) by (αu Ju ) and (α J ) by (αJ), and in the second
term we replace (αJ) by (α J ) and (αu Ju ) by (αJ). This yields
S0 (αJ) = CI(0) (αu Ju , αJ) + CS(0) (α J , αJ) , (7.97)
αu Ju α J
340 CHAPTER 7
which is the above mentioned relation between CI(0) , CS(0) and S0 .
Substitution of Eq. (7.97) into Eq. (7.96) gives
d (0) (0)
nαJ = CI (αJ, α J ) nα J + CS (αJ, αu Ju ) nα J
dt u u
α J αu Ju
− CI(0) (αu Ju , αJ) + CS(0) (α J , αJ) nαJ .
αu Ju α J
This equation can be compared with the standard statistical equilibrium equation
that is found in many classical textbooks, like for instance in Mihalas (1978). One
finds an exact correspondence between the two formulations, which allows a direct
comparison of the different rates. In the notation used by Mihalas, Cu denotes
the inelastic collisional rate for the transition from the lower level to the upper
level u, while Cu denotes the superelastic collisional rate for the transition from
the level u to the level . The relation with our quantities is the following
Cu = CI(0) (αu Ju , α J ) , Cu = CS(0) (α J , αu Ju ) .
Because of this simple relation, we can directly apply the results derived in Mi-
halas (1978) to the zero-rank multipole components of the collisional rates. In
particular, if the colliders are electrons with a velocity distribution f (v), such that
∞
f (v) dv = 1 ,
0
we have
∞
(0)
CI (αu Ju , α J ) = Ne σα J , αu Ju (v) f (v) v dv ,
v0
where Ne is the electron density, σα J , α J (v) is the cross-section for electron
u u
excitation from the level (α J ) to the level (αu Ju ), and v0 is the threshold velocity
defined, for non-relativistic electrons, by the equation
1
m v02 = Eα J − Eα J ,
2 u u
with m the electron mass and Eα J , Eα J the energies of the two levels.
u u
If the colliding particles have a Maxwellian distribution of velocities, a simple
relation can be established between the zero-rank multipole components CI(0) and
CS(0) . From detailed-balance arguments one gets the following Einstein-Milne rela-
tion
Eα − Eα
u Ju J
(0) 2J + 1
CS (α J , αu Ju ) = e kB Tc
CI(0) (αu Ju , α J ) , (7.98)
2Ju + 1
EQUATIONS FOR ATOMIC SYSTEMS 341
where kB is the Boltzmann constant and Tc is the absolute temperature character-
izing the Maxwellian distribution of the colliders.
Collisional rates for atom-electron interactions have been (and still are) widely
investigated both from the theoretical and from the experimental point of view, and
an extended collection of results is now available. These results can be directly ap-
plied to the study of polarization phenomena but, in general, they provide only the
K = 0 multipole component of the collisional rates. To find the other components
one must go back to a detailed study of the atom-collider interaction. In App. 4
we show how it is possible to relate, in some particular cases, the components CI(K)
and CS(K) with the corresponding components CI(0) and CS(0) .
7.13.b Collisional Rates due to Elastic Collisions
Elastic collisions induce transitions between degenerate energy sublevels of the form
|αJM , |αJM . The contribution of these collisions to the time evolution of the
density-matrix element ραJ (M, M ) can be written, by analogy with Eqs. (7.82),
(7.88) and (7.90), as
d
ραJ (M, M ) = CE (αJM M , αJM M ) ραJ (M , M )
dt
M M
1 1
− S(αJM M M ) ραJ (M, M ) + S(αJM M M )∗ ραJ (M , M ) .
2 2
M
If collisions are isotropic one finds, similarly to the preceding subsection, that this
equation can be cast into the form
d J J K
ραJ (M, M ) = (−1)M −M (2K + 1)
dt M −M Q
M M K
J J K
× CE(K) (αJ) ραJ (M , M )
M −M Q
− CE(0) (αJ) ραJ (M, M ) , (7.99)
where
(K) N −N J J K J J K
CE (αJ) = (−1)
N −N P N −N P
N N N N P
× CE (αJN N , αJN N ) (7.100)
are the multipole components of the elastic collisional rates, which satisfy the prop-
erties
CE(K) (αJ)∗ = CE(K) (αJ) , CE(0) (αJ) ≥ 0 .
342 CHAPTER 7
From Eq. (7.99) one can prove, using Eq. (7.95), that
d d
nαJ = ραJ (M, M ) = 0 ,
dt dt
M
as expected because the overall population of the (αJ)-level is not affected by
elastic collisions.
The collisional rates due to elastic collisions are discussed in greater detail in the
following subsection.
7.13.c Collisional Rates in the Spherical Statistical Tensor Representation
Adding together the contributions of inelastic, superelastic, and elastic collisions,
we can write the statistical equilibrium equations in the form (see Eqs. (7.93),
(7.97), and (7.99))
.
d 2J + 1
J−M +J −M
ρ (M, M ) = (−1)
dt αJ 2J + 1
α J M M K
J J K J J K
× (2K + 1)
M −M Q M −M Q
× CI(K) (αJ, α J ) ρα J (M , M )
.
2Ju + 1
J−M +Ju −Mu
+ (−1)
2J + 1
αu Ju Mu Mu K
J J K Ju Ju K
× (2K + 1)
M −M Q Mu −Mu Q
× CS(K) (αJ, αu Ju ) ρα J (Mu , Mu )
u u
−M
+ (−1)M
M M K
J J K J J K
× (2K + 1)
M −M Q M −M Q
× CE(K) (αJ) ραJ (M , M )
− CI(0) (αu Ju , αJ) + CS(0) (α J , αJ) + CE(0) (αJ)
αu Ju α J
× ραJ (M, M ) .
The conversion to the spherical statistical tensor representation is easily performed.
Use of Eqs. (3.101) and (2.23a) gives
EQUATIONS FOR ATOMIC SYSTEMS 343
.
d K 2J + 1 (K)
ρQ (αJ) = C (αJ, α J ) ρK
Q (α J )
dt 2J + 1 I
α J
.
2Ju + 1 (K)
+ C (αJ, αu Ju ) ρK Q (αu Ju )
2J + 1 S
αu Ju
− CI(0) (αu Ju , αJ) + CS(0) (α J , αJ)
αu Ju α J
+ D(K) (αJ) ρK
Q (αJ) , (7.101)
where we have introduced the depolarizing rates due to elastic collisions defined
by1
D(K) (αJ) = CE(0) (αJ) − CE(K) (αJ) . (7.102)
Equation (7.101) shows that, under the assumption of isotropic collisions, only
spherical statistical tensors with the same K and Q value are coupled by collisional
transitions.
Depolarizing rates have received little attention both from the experimental and
from the theoretical point of view.2 This is obviously due to the fact that the actual
values of depolarizing rates are needed only for the interpretation of polarization
phenomena. A detailed discussion of depolarizing rates can be found in Lamb and
ter Haar (1971). According to these authors, the main contribution to depolar-
izing rates in astrophysical plasmas arises from long-range collisions with neutral
perturbers (generally hydrogen atoms). In such collisions, the interaction between
atom and perturber is described by a Van der Waals potential, which depends on
the atom-perturber distance R as
VVdW = h C6 (αJ) R−6 ,
where h is the Planck constant and C6 (αJ) is the Van der Waals constant for the
level in question.
Let us consider a particular collision having impact parameter b. If v̄r is the
average relative velocity between atom and collider, the interaction energy varies
with time according to
1
VVdW (t) = h C6 (αJ) ,
(b2 + v̄r2 t2 )3
1
It follows that D (0) (αJ) = 0, which shows at once that the population of the (αJ)-level is
unaffected by elastic collisions (see Eq. (3.107)).
2
Some theoretical results on depolarizing rates by isotropic collisions with hydrogen atoms have
recently appeared in the literature. Applications to NaI and other atoms of the same isoelectronic
sequence can be found in Kerkeni (2002) and Kerkeni et al. (2003). Applications to p states of
neutral atoms have been presented by Derouich et al. (2003).
344 CHAPTER 7
where t = 0 is the time of closest approach.
The typical collision time tc is of order b/v̄r , and for astrophysical plasmas it is
much shorter than the other typical times characterizing the evolution of the atomic
system. This means that the collisional interaction produces a phase shift φ in the
eigenfunction of the atomic system whose order of magnitude is given by
∞
2π 3π 2 C6 (αJ)
φ= VVdW (t) dt = . (7.103)
h 4 b5 v̄r
−∞
The cumulative effect of these phase shifts is responsible for the depolarizing rates
via a mechanism similar to that illustrated in Sect. 5.14 for classical oscillators.
We can give a rough estimate of the depolarizing rate by writing D(K) in the form
D(K) ≈ f = σ v̄r nH ,
where f is the collision frequency, nH the number density of hydrogen atoms, and
σ – the depolarizing cross-section – is given by
σ = π b2c ,
bc being the critical impact parameter which produces a phase shift equal to some
large value φ0 1. This approximation is equivalent to assuming that all collisions
having b < bc are totally effective in depolarizing the atom, while the others can
be neglected. Taking for φ0 the value 0.61,1 one gets from Eq. (7.103)
0.2
C6 (αJ)
bc = 1.65 , (7.104)
v̄r
whence
0.4 0.6
f = 8.5 C6 (αJ) v̄r nH . (7.105)
To get a rough estimate of C6 (αJ) we can use the expression given by Aller
(1963)
e2
C6 (αJ) = 0 p r2 (αJ) ,
h
where e0 is the electron
charge,
h the Planck constant, p the polarizability of the
hydrogen atom, and r2 (αJ) the mean square radius of the electronic cloud in the
level (αJ). Since p 6.7 × 10−25 cm3 (Allen, 1973), one gets, expressing r2 (αJ)
in atomic units
C6 (αJ) 6.5 × 10−34 r2 (αJ) cm6 s−1 . (7.106)
1
This value is discussed in some detail by Aller (1963) in connection with the frequency of
collisions responsible for the broadening of spectral lines.
EQUATIONS FOR ATOMIC SYSTEMS 345
Values for r2 (αJ) can be deduced by means of scaled Fermi-Thomas potentials
(see e.g. Warner, 1969); otherwise one can refer to an approximate expression
which, for neutral atoms, reduces to (Unsöld, 1955)
2
2
5 13.6
r (αJ) = ,
2 I − EαJ
where I is the ionization potential of the atom and EαJ is the excitation energy of
the (αJ)-level, both expressed in eV.
On the other hand, assuming a Maxwellian distribution of velocities at temper-
ature T , the average relative velocity is
8kB T 1
v̄r = 1+ , (7.107)
π mH µ
where kB is the Boltzmann constant, mH is the mass of the hydrogen atom, and
µ is the atomic weight of the atom.1 Substitution of Eqs. (7.106) and (7.107) into
Eq. (7.105) gives
−10
2 0.4 1
0.3
f 1.4 × 10 r (αJ) T 1+ nH s−1 , (7.108)
µ
where T is in K and nH in cm−3 .
Particularly interesting is the ratio D(1) (αJ)/D(2) (αJ), that is the ratio of depo-
larizing rates for orientation and alignment (see Sect. 3.7), because it is strongly
dependent on the type of interaction between atom and collider. We have shown
in App. 4 that – for isotropic collisions – the interaction can be expanded into a
series of tensor operators of different rank. Equations (7.102), (A4.3) and (2.36a)
give
D(K) (αJ) =
2J ! !
J J 0 J J K
= (2J + 1) (−1)2J+K − (−1)K ΓE(K ) (αJ) ,
J J K J J K
K =1
which shows that depolarizing rates are unaffected by a zero-rank tensor operator.
If the interaction is described by a single tensor operator of rank 1 (or by a tensor
operator of rank 1 plus a tensor operator of rank 0) we have, using Eq. (2.36d)
D(1) (αJ) 1 1
= J
= 0, ,
D(2) (αJ) 3 2
1
It is worth noticing that Eqs. (7.104), (7.106) and (7.107) give for the collision time
0.2 , - −0.6
1
tc = bc / v̄r 3.8 × 10−12 r 2 (αJ) T 1+ s,
µ
thus tc 10−13 - 10−14 s for a typical stellar atmosphere.
346 CHAPTER 7
whereas for a single tensor operator of rank 2 (or a tensor operator of rank 2 plus
a tensor operator of rank 0) we have, using Eq. (2.36h)
D(1) (αJ) 1 4J 2 + 4J − 3 1
= J
= 0, . (7.109)
D(2) (αJ) 3 4J 2 + 4J − 7 2
In particular, the Van der Waals interaction depends on the atomic variables
through the quadrupolar momentum of the atomic cloud, which is a symmetric
Cartesian tensor that can be decomposed into the sum of a spherical tensor of
rank 0 plus a spherical tensor of rank 2, so that the ratio D(1) (αJ)/D(2) (αJ) is
given by Eq. (7.109). We can compare this expression with the results obtained in
Sect. 5.14, where the atom was schematized as a collection of classical oscillators.
The corresponding ratio (δ1 /δ2 ) was found to be (see Eq. (5.195))
δ1 5
= ,
δ2 3
which is just the value given by Eq. (7.109) for J = 1 (the only quantum-mechanical
atomic level for which the classical analogy works).
7.13.d Collisional Rates in the Presence of a Magnetic Field
Collisional rates have been discussed in this section under the assumption that
colliders have an isotropic velocity distribution, and that no physical agent able
to break the spherical symmetry of the problem is present (otherwise Eqs. (7.84)
and (7.92) could not be deduced). In the presence of a magnetic field the latter
requirement is not satisfied, and we can wonder whether the previous results are
still valid or not.
In typical astrophysical plasmas the inelastic and superelastic collisional rates are
due to thermal electrons. Therefore they depend only on the velocity distribution of
electrons and on the eigenfunctions of the atomic levels involved in the (collisional)
transition. If we restrict attention to the Zeeman effect regime, the eigenfunctions
are not altered by the presence of a magnetic field, and the collisional rates remain
the same as in the zero-field case provided
h νL kB T
(with νL the Larmor frequency and kB the Boltzmann constant), or, numerically
B 1.5 × 104 T , (7.110)
where B is in G and T in K. This inequality is well-satisfied both for the sun and
for magnetic stars.
As far as depolarizing rates are concerned, the situation is more involved. Besides
the condition of Eq. (7.110) (which still must be satisfied), there is the further
constraint that the typical collision time tc should be much smaller than the Larmor
time (2πνL )−1 . Thus
B 1.1 × 10−7 t−1c ,
EQUATIONS FOR ATOMIC SYSTEMS 347
where B is in G and tc in s. This condition is also well-satisfied both for the sun
and for magnetic stars (see footnote on p. 345).
7.13.e Concluding Remarks
The treatment of collisional rates presented in this section has been restricted to the
multi-level atom, but it could be generalized to the multi-term atom. Considering
for instance the inelastic collisional transitions between two terms (β L S ) and
(βLS), one could write an equation of the form
d
ρ(JM, J M ) =
βLS
CI (βLSJM J M , β L S J M J M )
dt
β L S J M J M
× β L S
ρ(J M , J M ) .
Assuming that collisions are isotropic and that no magnetic field is present (or
that the magnetic field is sufficiently weak to satisfy the constraints discussed in
Sect. 7.13.d), one could derive, along the same lines of the former subsections, the
general properties of the rates CI (βLSJM J M , β L S J M J M ). However, this
more complicated type of collisional rates will not be studied in this book.
A major point concerning collisional rates is that, under a wide variety of astro-
physical and laboratory conditions, they can be simply added to the corresponding
radiative rates in the statistical equilibrium equations; symbolically
d
Density-Matrix Element =
dt
= Radiative Rates + Collisional Rates × Density-Matrix Element .
Although this equation may seem physically evident, it is indeed correct only when
the so-called impact approximation is valid (see Lamb and ter Haar, 1971, for a
thorough discussion of this subject). The impact approximation – that we have
already encountered in Sect. 5.14, see Eq. (5.176) – requires that the collision time
is much smaller than the relaxation time due to radiative rates, a condition that is
generally well-satisfied both in laboratory plasmas and in stellar atmospheres.
CHAPTER 8
RADIATIVE TRANSFER FOR POLARIZED RADIATION
In previous sections of this book we have derived the radiative transfer equations for
polarized radiation from general principles. In Sect. 5.2 the derivation was based
on a classical approach involving the familiar concept of index of refraction, while
in Sect. 6.7 it was based on a more powerful quantum-electrodynamical approach.
As already remarked, the structure of the resulting equations is obviously the same.
In Sect. 5.5 we have indeed outlined how this structure is intimately connected with
very general physical principles.
This chapter is devoted to the solution of the radiative transfer equations for
polarized radiation in their most general formulation. We will also consider some
particular cases where the solution can be expressed in purely analytical form.
Most practical applications of radiative transfer for polarized radiation concern
the problem of line formation in a magnetic field, and the consequent analysis of
polarimetric observations of solar active regions or of magnetic stars. These special
aspects of the general problem of radiative transfer for polarized radiation will be
discussed in the following chapters.
8.1. Generalities
The radiative transfer equations for polarized radiation have been derived from the
principles of Quantum Electrodynamics in Chap. 6. For a stationary medium (i.e.
a medium whose properties do not vary over time) we have from Eq. (6.83)
d 3 3
=−
Si (ν, Ω) A
Kij +
Sj (ν, Ω) S
Kij +ε
Sj (ν, Ω) (i = 0, 1, 2, 3) , (8.1)
i
ds j=0 j=0
is the Stokes vector
where s is the coordinate measured along the ray path, Si (ν, Ω)
K A and K S are
of the radiation of frequency ν propagating along the direction Ω, ij ij
the propagation matrices due to absorption and stimulated emission, respectively,
and εi is the emission vector.
For our present purposes it is convenient to cast Eq. (8.1) into matrix form.
Introducing the propagation matrix tout court
A
Kij = Kij − Kij
S
,
and stressing the explicit dependence of the various quantities on the coordinate s ,
we have
d
I(s) = −K(s) I(s) + ε(s) , (8.2)
ds
350 CHAPTER 8
where1 ⎛ ⎞ ⎛ ⎞
S0 (s) I(s)
⎜ S (s) ⎟ ⎜ Q(s) ⎟
⎜ ⎟ ⎜ ⎟
I(s) = ⎜ 1 ⎟ = ⎜ ⎟ .
⎝ S2 (s) ⎠ ⎝ U (s) ⎠
S3 (s) V (s)
The matrix K, as deduced both from classical and from quantum physics, has the
form (see Eqs. (5.25) and (6.85))
⎛ ⎞
ηI ηQ ηU ηV
⎜η −ρU ⎟
⎜ ηI ρV ⎟
K=⎜ Q ⎟ , (8.3)
⎝ ηU −ρV ηI ρQ ⎠
ηV ρU −ρQ ηI
where the seven independent quantities ηI , ηQ , . . . , ρV are related to the (direction-
dependent) properties of the medium by Eqs. (5.26) (for the classical case) or
Eqs. (6.86) (for the quantum-mechanical case). The general expression for the
emission vector ε, derived from Quantum Mechanics, is given in Eqs. (6.86). An-
other expression, obtained from classical physics for the special case of an atomic
vapor embedded in a magnetic field, is given in Eqs. (5.38).
It is quite interesting to compare Eq. (8.2) with the usual transfer equation for
‘unpolarized’ radiation (see for instance Mihalas, 1978)
d
I(s) = −k(s) I(s) + ε(s) , (8.4)
ds
where I is the intensity of the beam and where k and ε are the absorption and
emission coefficients, respectively. It is seen that the two equations have a similar
structure, with the obvious difference that the scalar quantities I and ε of Eq. (8.4)
are replaced, in the ‘polarized’ case, by the four-component vectors I and ε, while
the scalar absorption coefficient k is replaced by the 4 × 4 propagation matrix K.
The peculiarity of the transfer of polarized radiation originates just from the matrix
character of Eq. (8.2), which produces a typical effect of ordering that will be
explained in the following section.
8.2. Formal Solution of the Radiative Transfer Equations:
the Evolution Operator
To find a formal solution to Eq. (8.2), we start by considering the homogeneous
equation
d
I(s) = −K(s) I(s) , (8.5)
ds
1
In this chapter we denote the Stokes vector by the symbol I (instead of S) to avoid confusion
with the source-function vector that will be introduced later.
RADIATIVE TRANSFER FOR POLARIZED RADIATION 351
which describes the transfer of radiation through a purely absorbing medium.
We define the evolution operator 1 O(s, s ) as the 4 × 4 real matrix which, when
applied to the Stokes vector at point s , yields the Stokes vector at point s (with
s ≥ s )
I(s) = O(s, s ) I(s ) . (8.6)
Obviously, the evolution operator obeys the condition
O(s, s) = 1 , (8.7)
1 being the 4 × 4 identity matrix, and the composition law
O(s, s ) = O(s, s ) O(s , s ) (s ≥ s ≥ s ) . (8.8)
Derivation of Eq. (8.6) with respect to s and use of Eq. (8.5) show that the
evolution operator satisfies the differential equation
d
O(s, s ) = −K(s) O(s, s ) . (8.9)
ds
In a similar way we get
d
O(s, s ) = O(s, s ) K(s ) . (8.10)
ds
Equation (8.9) can be formally integrated to give
s
O(s, s ) = 1 − K(s1 ) O(s1 , s ) ds1 . (8.11)
s
The operator O(s1 , s ) in the right-hand side can in turn be expressed via Eq. (8.11),
and iterating the procedure we obtain
∞
s s1 sn−1
O(s, s ) = 1 + (−1)n
ds1 ds2 · · · dsn K(s1 ) K(s2 ) · · · K(sn ) .
n=1 s s s
This equation can be rewritten in the form
O(s, s ) =
∞ s s s & '
(−1)n
=1+ ds1 ds2 · · · dsn P K(s1 ) K(s2 ) · · · K(sn ) , (8.12)
n=1
n!
s s s
1
The concept of evolution operator expressed as a 4 × 4 real matrix was introduced by Landi
Degl’Innocenti and Landi Degl’Innocenti (1985). A similar concept in terms of 2 × 2 complex
matrices is due to Van Ballegooijen (1985, 1987) and is developed in Sects. 8.8 and 8.9. The
presentation given here follows rather closely the derivation by Landi Degl’Innocenti (1987). It
appears that the general concept of evolution operator has been known for a long time in matrix
theory (see e.g. Gantmacher, 1966, where it is referred to as matricant).
352 CHAPTER 8
where we have introduced the Dyson chronological product of the n matrices K(s1 ),
K(s2 ), . . . , K(sn ), defined by
& '
P K(s1 ) K(s2 ) · · · K(sn ) = K(si ) K(sj ) · · · K(sk ) , (8.13)
where (i, j, . . . , k) is a permutation of the integers (1, 2, . . . , n) such that
si ≥ sj ≥ . . . ≥ s k .
In other words, the chronological product orders the matrices in the bracket in such
a way that larger values of the argument s stand to the left of smaller values.
The formal solution of the non-homogeneous transfer equation (Eq. (8.2)) can
be easily expressed in terms of the evolution operator O(s, s ). Bearing in mind
Eqs. (8.7) and (8.9), it is seen that the expression
s
I(s) = O(s, s ) ε(s ) ds + O(s, s ) I(s ) (8.14)
s
satisfies Eq. (8.2). The physical interpretation of this equation is straightforward.
Let us consider a slab extending between points s and s along the ray path. The
Stokes vector at the exit of the slab, I(s), results from the addition of two terms.
The latter is just the Stokes vector entering the slab, I(s ), transformed by the
evolution operator associated with the whole slab, O(s, s ). The former results
from the contribution of the emission in each infinitesimal interval of the slab,
ε(s ) ds , transformed by the evolution operator associated with the fraction of
the slab crossed by the radiation, O(s, s ).
Equation (8.14) can be directly applied to the case of a slab extending to infinity
in one direction (semi-infinite atmosphere). Provided the emission vector does not
grow above reasonable limits when s → −∞, or, more properly, provided
lim O(s, s ) ε(s ) = 0 ,
s →−∞
the formal solution of Eq. (8.2) is simply given by
s
I(s) = O(s, s ) ε(s ) ds . (8.15)
−∞
The operator O(s, s ) introduced in this section is nothing but the generalization
to the radiative transfer problem for polarized radiation of the usual attenuation
operator : s
− k(s ) ds
o(s, s ) = e s (8.16)
RADIATIVE TRANSFER FOR POLARIZED RADIATION 353
that solves the transfer equation for ‘unpolarized’ radiation (Eq. (8.4)). It can be
easily proved that the homogeneous equation
d
I(s) = −k(s) I(s)
ds
has the solution
I(s) = o(s, s ) I(s ) , (8.17)
and that the non-homogeneous equation has the solution
s
I(s) = o(s, s ) ε(s ) ds + o(s, s ) I(s ) . (8.18)
s
Comparison of Eqs. (8.17)-(8.18) with Eqs. (8.6)-(8.14) shows the strict correspon-
dence between the operators O(s, s ) and o(s, s ). However, Eq. (8.12) – contrary
to Eq. (8.16) – contains a chronological product. This peculiarity is intimately
connected with the physical fact that two slabs, A and B, acting differently on the
polarization properties of a radiation beam, do not ‘commute’: in other words, the
emerging polarization is in general different if slab A is located in front of slab B
or vice versa. On the contrary, the ordering of a train of absorbing slabs does not
matter for ‘unpolarized’ radiation.
8.3. Analytical Expressions for the Evolution Operator
Owing to the presence of a chronological product in Eq. (8.12), the expression for
the evolution operator O(s, s ) cannot be reduced, in general, to simpler forms.
An important exception occurs when the propagation matrix K(s) at a particular
point s1 commutes with the same matrix at any other point s2 ,
K(s1 ), K(s2 ) = K(s1 ) K(s2 ) − K(s2 ) K(s1 ) = 0. (8.19)
In such case the chronological product reduces to an ordinary product, so that
Eq. (8.12) becomes
∞ s n
(−1)n
O(s, s ) = 1 + K(s ) ds . (8.20)
n=1
n!
s
Let’s now recall the definition of function of a matrix. If A is a square matrix,
and f (z) an arbitrary function of the variable z with Taylor expansion
∞ ∞
1 dn f n
f (z) = n
z = fn z n ,
n=0
n! dz z=0 n=0
354 CHAPTER 8
the matrix F = f (A) is defined to be
∞
F = f (A) = fn An ,
n=0
with A0 ≡ 1 . In particular
∞
−A (−1)n n
e = A , (8.21)
n=0
n!
so that Eq. (8.20) can be rewritten in the form
:s
− K(s ) ds
O(s, s ) = e s . (8.22)
There are three remarkable cases where Eq. (8.19) holds and, therefore, Eq. (8.22)
can be applied. These are the following:
a) the matrix K is independent of s ;
b) the matrix K can be written in the form
K(s) = c1 (s) 1 + c2 (s) H
with H independent of s ;1
c) the matrix K has an arbitrary dependence on s but it has a particular structure,
with several elements identically zero.
8.3.a Propagation matrix independent of s
The integral appearing in Eq. (8.22) can be immediately performed to give
−(s−s ) K
O(s, s ) = e .
This expression can be transformed, with some algebra, in the linear combination
of four matrices, each multiplied by a suitable exponential factor. The relevant
calculations are carried out in App. 5 and allow the evolution operator to be written
in the form
−(s−s ) η
O(s, s ) = e I
1
× cosh (s − s ) Λ+ ( ) + cos (s − s ) Λ− (η , ρ
η, ρ ) M1 (η , ρ
)
2
− sin (s − s ) Λ− ( ) M2 (
η, ρ η, ρ
)
− sinh (s − s ) Λ+ ( ) M3 (
η, ρ η, ρ)
!
1
+ cosh (s − s ) Λ+ ( η, ρ ) − cos (s − s ) Λ− (η , ρ
) M4 (η , ρ
) (8.23)
2
1
Obviously a) is a special case of b). The two cases will be however treated separately, as
both of them have important applications.
RADIATIVE TRANSFER FOR POLARIZED RADIATION 355
or, alternatively
−(s−s ) [ ηI + Λ+ ( ρ)]
η , −(s−s ) [ ηI − Λ+ ( ρ)]
η ,
O(s, s ) = e N1 (
η, ρ
) + e N2 (η , ρ
)
−(s−s ) [ ηI + i Λ− ( ρ)]
η , −(s−s ) [ ηI − i Λ− ( ρ)]
η ,
+e N3 (
η, ρ
) + e N4 (η , ρ
) ,
where the quantities Λ± ( ) and the matrices Mi (η , ρ
η, ρ ) and Ni (η , ρ
) are given
in App. 5. In the special case
Λ+ (
η, ρ
) = Λ− (
η, ρ
) = 0
the expressions above should be replaced by the following
−(s−s ) ηI 1
O(s, s ) = e 1 − (s − s ) G(
η, ρ
) + (s − s )2 G(η , ρ
)2 ,
2
with G given by Eq. (A5.25).
8.3.b Propagation matrix of the form [c1 (s) 1 + c2 (s) H ] , with H independent of s
Equation (8.22) gives, with the help of Eq. (A5.6)
−C1 (s,s ) 1 −C2 (s,s ) H −C1 (s,s ) −C2 (s,s ) H
O(s, s ) = e e = e e ,
where
s s
C1 (s, s ) = c1 (s ) ds , C2 (s, s ) = c2 (s ) ds .
s s
Obviously the matrix H has the same structure as the matrix K. Writing
⎛ ⎞
hI hQ hU hV
⎜h −rU ⎟
⎜ hI rV ⎟
H =⎜ Q ⎟ ,
⎝ hU −rV hI rQ ⎠
hV rU −rQ hI
we obtain from Eq. (A5.20)
−[C1 (s,s ) + C2 (s,s ) hI ]
O(s, s ) = e
1
× cosh C2 (s, s ) Λ+ (h, r ) + cos C2 (s, s ) Λ− (h, r ) M1 (h, r )
2
− sin C2 (s, s ) Λ− (h, r ) M2 (h, r )
− sinh C2 (s, s ) Λ+ (h, r ) M3 (h, r )
!
1
+ cosh C2 (s, s ) Λ+ (h, r ) − cos C2 (s, s ) Λ− (h, r ) M4 (h, r )
2
356 CHAPTER 8
or, from Eq. (A5.23)
&
−C1 (s,s ) −C2 (s,s ) [ hI + Λ+ ( r)]
h,
O(s, s ) = e e N1 (h, r )
−C2 (s,s ) [ hI − Λ+ ( r)]
h,
+e N2 (h, r )
−C2 (s,s ) [ hI + i Λ− ( r)]
h,
+e N3 (h, r )
−C2 (s,s ) [ hI − i Λ− ( r)]
h,
'
+e N4 (h, r ) .
If Λ+ (h, r ) = Λ− (h, r ) = 0, these expressions must be replaced by
−[C1 (s,s ) + C2 (s,s ) hI ] 1
O(s, s ) = e 1 − C2 (s, s ) G(h, r ) + C (s, s )2 G(h, r )2 ,
2 2
with G given by Eq. (A5.25).
8.3.c Propagation matrix with peculiar form
There are some particular cases where the matrix K(s) has only few non-zero
elements, so that, even if these elements are arbitrary functions of s, Eq. (8.19) is
satisfied. As an example, let us consider the following particular case
⎛ ⎞
ηI (s) 0 0 ηV (s)
⎜ 0 0 ⎟
⎜ ηI (s) ρV (s) ⎟
K(s) = ⎜ ⎟ . (8.24)
⎝ 0 −ρV (s) ηI (s) 0 ⎠
ηV (s) 0 0 ηI (s)
Using the matrices A3 and B3 defined in App. 5 (Eq. (A5.1c)) we can write
K(s) = ηI (s) 1 + a3 (s) A3 + b3 (s) B3 , (8.25)
where
1 1
a3 (s) = η (s) + i ρV (s) , b3 (s) = a3 (s)∗ = η (s) − i ρV (s) .
2 V 2 V
As the three matrices in the right-hand side of Eq. (8.25) are commuting matrices
(see Eq. (A5.3)), the matrix K(s) satisfies Eq. (8.19), thus the evolution operator
is given by Eq. (8.22). Using Eq. (A5.6) we get
−HI 1 −1
2 (HV + i RV ) A3 −1
2 (HV − i RV ) B3
O(s, s ) = e e e ,
where
s s s
HI = ηI (s ) ds , HV = ηV (s ) ds , RV = ρV (s ) ds . (8.26)
s s s
RADIATIVE TRANSFER FOR POLARIZED RADIATION 357
Expanding the exponentials as in App. 5, we obtain after some algebra
⎛ ⎞
cosh HV 0 0 − sinh HV
⎜ − sin RV ⎟
−HI ⎜ 0 cos RV 0 ⎟
O(s, s ) = e ⎜ ⎟ . (8.27)
⎝ 0 sin RV cos RV 0 ⎠
− sinh HV 0 0 cosh HV
In a similar way it can be shown that a propagation matrix of the form
⎛ ⎞
ηI (s) ηQ (s) 0 0
⎜ η (s) η (s) 0 ⎟
⎜ 0 ⎟
K(s) = ⎜ Q I
⎟ (8.28)
⎝ 0 0 ηI (s) ρQ (s) ⎠
0 0 −ρQ (s) ηI (s)
leads to the evolution operator
⎛ ⎞
cosh HQ − sinh HQ 0 0
⎜ − sinh H ⎟
−HI ⎜ cosh HQ 0 0 ⎟
O(s, s ) = e ⎜
Q
⎟ ,
⎝ 0 0 cos RQ − sin RQ ⎠
0 0 sin RQ cos RQ
with HI given by Eq. (8.26) and
s s
HQ = ηQ (s ) ds , RQ = ρQ (s ) ds .
s s
Finally, if ⎛ ⎞
ηI (s) 0 ηU (s) 0
⎜ 0 −ρU (s) ⎟
⎜ η (s) 0 ⎟
K(s) = ⎜ I
⎟ (8.29)
⎝ ηU (s) 0 ηI (s) 0 ⎠
0 ρU (s) 0 ηI (s)
one gets
⎛ ⎞
cosh HU 0 − sinh HU 0
⎜ sin RU ⎟
−HI ⎜ 0 cos RU 0 ⎟
O(s, s ) = e ⎜ ⎟ ,
⎝ − sinh HU 0 cosh HU 0 ⎠
0 − sin RU 0 cos RU
where
s s
HU = ηU (s ) ds , RU = ρU (s ) ds .
s s
358 CHAPTER 8
8.4. Solution of the Radiative Transfer Equations by Diagonalization
A different way to find a formal solution to the radiative transfer equations is
provided by the diagonalization method. This method was first introduced by
Kjeldseth-Moe (1968) for the case of a simplified propagation matrix (ρQ = ρU =
ρV = 0) and was later generalized by Katz (1971) and Šidlichovsky (1976).
Let us consider the eigenvalue equation for the propagation matrix K at point s,
K u(i) = λi u(i) (i = 1, 2, 3, 4) .
The eigenvalues λi are obtained by solution of the equation
det (K − λ 1 ) = 0. (8.30)
Introducing the formal vectors
η = (ηQ , ηU , ηV ) ,
ρ
= (ρQ , ρU , ρV ) , (8.31)
Eq. (8.30) leads to the fourth-degree algebraic equation in λ
(ηI − λ)4 − (η 2 − ρ2 ) (ηI − λ)2 − (η · ρ
)2 = 0 , (8.32)
whose solution is
λ1 = ηI + Λ+ (
η, ρ
)
λ2 = ηI − Λ+ (
η, ρ
)
λ3 = ηI + i Λ− (
η, ρ
)
λ4 = ηI − i Λ− (
η, ρ
) , (8.33)
where Λ+ (η, ρ
) and Λ− (
η, ρ
) are the quantities – which now acquire a more precise
significance – defined in Eqs. (A5.18).
If both Λ+ (η, ρ
) and Λ− (η, ρ
) are non-zero (so that the four eigenvalues λi are
distinct), we can consider the matrix X such that
X −1 K X = K , (8.34)
where K is the diagonal matrix1
Kij = λi δij .
1
If Λ+ (η , ρ ) = 0 and/or Λ− (η , ρ ) = 0, the matrix K cannot be reduced, in general, to
diagonal form. In such cases the method described in this section cannot be applied. A similar
method, based on a different reduction for K, could indeed be employed (see e.g. Heading, 1958,
p.63), but it will not be considered here. Note that the case ρQ = ρU = ρV = 0, which implies
Λ− (η , ρ ) = 0, is a special one: K becomes a symmetric matrix and can be diagonalized.
RADIATIVE TRANSFER FOR POLARIZED RADIATION 359
Multiplying Eq. (8.2) by X −1 from the left, and introducing the identity XX −1 = 1
between K(s) and I(s), we get
d
X −1 I(s) = −K (s) X −1 I(s) + X −1 ε(s) . (8.35)
ds
Since K depends in general on s, the matrix X depends also on s, thus it does
not commute with the operator d/ds. In the special case where X is independent
of s, Eq. (8.35) becomes
d
I (s) = −K (s) I (s) + ε (s) , (8.36)
ds
where
I (s) = X −1 I(s) (8.37a)
−1
ε (s) = X ε(s) . (8.37b)
As the matrix K (s) is diagonal, Eq. (8.36) represents a set of four decoupled
differential equations (in other words, the linear combinations of Stokes parameters
given in Eq. (8.37a) are the eigenvectors of the radiative transfer equations). Thus
they have the solution (cf. Eqs. (8.4), (8.18) and (8.16))
s
I (s) = O (s, s ) ε (s ) ds + O (s, s ) I (s ) ,
s
where O (s, s ) is the diagonal matrix
⎛ ⎞
e1 (s, s ) 0 0 0
⎜ e2 (s, s ) ⎟
⎜ 0 0 0 ⎟
O (s, s ) = ⎜ ⎟ (8.38)
⎝ 0 0 e3 (s, s ) 0 ⎠
0 0 0 e4 (s, s )
with :s
− λi (s ) ds
ei (s, s ) = e s (i = 1, 2, 3, 4) .
Inversion of Eq. (8.37a) yields the solution of the radiative transfer equations in
the form
s
I(s) = X I (s) = X O (s, s ) ε (s ) ds + X O (s, s ) I (s ) . (8.39)
s
It is easily seen that this expression, although different from the solution based on
the evolution operator (Eq. (8.14)), is however strictly connected with it. Compar-
ison of Eqs. (8.39) and (8.14) shows, with the help of Eqs. (8.37), that the evolution
360 CHAPTER 8
operator O(s, s ) and the matrix O (s, s ) defined in Eq. (8.38) are connected by
the relation
O(s, s ) = X O (s, s ) X −1 . (8.40)
It should be emphasized that the assumption X independent of s, which is at the
basis of the diagonalization method, implies that condition (8.19) is satisfied. This
is easily seen by inversion of Eq. (8.34), which gives
K(s1 ), K(s2 ) = X K (s1 ) X −1 X K (s2 ) X −1 − X K (s2 ) X −1 X K (s1 ) X −1
= X K (s1 ), K (s2 ) X −1 = 0 ,
because the matrix K is diagonal. It follows that the solution of the radiative
transfer equations based on the diagonalization of the propagation matrix is equiv-
alent to the solution based on the evolution operator with the evolution operator
given by Eq. (8.22).
The actual calculation of the matrices X and X −1 can be performed by a stan-
dard method that is presented in App. 6. We give here the final results.
Defining the formal vectors
ζ = η × ρ
≡ (ηU ρV − ηV ρU , ηV ρQ − ηQ ρV , ηQ ρU − ηU ρQ ) (8.41)
η − Λ+ ζ + (
f (1) = Λ2+ η · ρ ) ρ
f (2) = Λ2+
η + Λ+ ζ + (
η · ρ ) ρ
f (3) = −Λ2− η − i Λ− ζ + (
η·ρ
) ρ
f (4) = −Λ2− η + i Λ− ζ + (
η·ρ
) ρ
and the scalar quantities
1 1
g+ =
η·ρ
2 Λ4+ η 2 + 2 Λ2+ ( )2 + ρ2 (η · ρ
)2
1 1
g− =
2 Λ4− η 2 − 2 Λ2− (
η·ρ )2 + ρ2 (η · ρ
)2
l+ = Λ+ ρ2 + Λ2+
l− = Λ− ρ2 − Λ2− ,
one has (under the restrictions specified at the end of App. 6)
⎛ ⎞
l+ f1(1) f2(1) f3(1)
⎜ ⎟
⎜ −l+ f1(2) f2(2) f3(2) ⎟
X −1
=⎜
⎜
⎟
⎟ (8.42a)
⎝ i l− f1(3) f2(3) f3(3) ⎠
−i l− f1(4) f2(4) f3(4)
RADIATIVE TRANSFER FOR POLARIZED RADIATION 361
⎛ ⎞
g + l+ −g+ l+ i g − l− −i g− l−
⎜ (2) ⎟
⎜ g+ f1 g+ f1(1) g− f1(4) g− f1(3) ⎟
X =⎜
⎜ (2)
⎟.
⎟ (8.42b)
⎝ g+ f2 g+ f2(1) g− f2(4) g− f2(3) ⎠
g+ f3(2) g+ f3(1) g− f3(4) g− f3(3)
Substitution of Eqs. (8.42) into Eq. (8.37a) yields explicit expressions for the eigen-
vectors of the radiative transfer equations. These are the linear combinations (two
real and two complex)
I1 = l+ I + f1(1) Q + f2(1) U + f3(1) V with eigenvalue ηI + Λ+
I2 = −l+ I + f1(2) Q + f2(2) U + f3(2) V with eigenvalue ηI − Λ+
I3 = i l− I + f1(3) Q + f2(3) U + f3(3) V with eigenvalue ηI + i Λ−
I4 = −i l− I + f1(4) Q + f2(4) U + f3(4) V with eigenvalue ηI − i Λ− .
If the propagation matrix is constant, so that the elements of the matrix O (s, s )
defined in Eq. (8.38) reduce to
−(s−s ) λi
ei (s, s ) = e
(with λi independent of s), the expression for the evolution operator obtained by
substitution of Eqs. (8.42) into Eq. (8.40) must coincide with Eq. (8.23). The
formal proof is however rather involved; moreover, it should be kept in mind that
Eqs. (8.42) are ill-defined in several particular cases (see App. 6).
Finally, we want to remark that the diagonalization method is particularly suit-
able to solve the radiative transfer equations in the special cases that have been
collected in Sect. 8.3.c. Let us suppose, for instance, that the propagation matrix
has the form of Eq. (8.24). It is easily shown that the radiative transfer equations
for the linear combinations of Stokes parameters
1 1
I1 = I +V I3 = Q + iU
2 2
1 1
I2 = I −V I4 = Q − iU (8.43)
2 2
are
d 1
I = − ηI + ηV I1 + εI + εV
ds 1 2
d 1
I = − ηI − ηV I2 + εI − εV
ds 2 2
d 1
I = − ηI − i ρV I3 + ε + i εU
ds 3 2 Q
d 1
I = − ηI + i ρV I4 + εQ − i εU , (8.44)
ds 4 2
362 CHAPTER 8
which is a set of four decoupled equations. The solution to these equations is trivial
(cf. Eqs. (8.4), (8.18) and (8.16)), and the solution for the Stokes parameters is
readily obtained by inversion of Eqs. (8.43).1 This is of course the same solution
that is obtained from Eq. (8.14) with the evolution operator of Eq. (8.27). It is
important to stress that the eigenvectors of the radiative transfer equations have in
this case a simple physical meaning: the combinations I1 and I2 in Eq. (8.43) are
just the amount of right and left circular polarization contained in the radiation
beam.
Analogous considerations can be repeated for the propagation matrices described
by Eqs. (8.28) and (8.29). In these cases the eigenvectors of the transfer equations
represent the amount of linear polarization along different directions.
8.5. Evolution Operator for Purely Dichroic
or Purely Dispersive Media
The analytical expression for the evolution operator derived in Sect. 8.3.a takes a
simpler form when the medium where the radiation is propagated is either purely
dichroic (ρQ = ρU = ρV = 0) or purely dispersive (ηI = ηQ = ηU = ηV = 0).
The purely dichroic case is seldom encountered in practice. Its analysis is however
important because, in some cases, it can be useful to tackle a transfer problem
neglecting – as a first-order approximation – the influence of anomalous dispersion
(this leads in general to considerable simplifications). The opposite case of a purely
dispersive medium is met in some astrophysical applications (the propagation of
radio waves in a magnetized plasma is a typical example). Further examples of
purely dichroic or purely dispersive media can be found in laboratory optics: an
ideal (dichroic) polarizer is composed of a purely dichroic medium, while an ideal
retarder is composed of a purely dispersive medium.
a) purely dichroic media
From Eqs. (A5.18) one has
Λ+ (
η , 0) = η2 = 2 + η2 + η2 = η
ηQ U V
Λ− (
η , 0) = 0 ,
and the evolution operator can be written, using Eqs. (8.23) and (A5.21), in the
form
1
Note that in this case, being
η2 ρ2 = ηV ρV = (η · ρ )2 ,
2 2
the condition of Eq. (A6.14) is not satisfied, thus the matrices X, X−1 in Eqs. (8.42) are ill-
defined (actually the denominator of the quantity g− vanishes).
RADIATIVE TRANSFER FOR POLARIZED RADIATION 363
⎧ ⎛ ⎞
⎪ 0 0 0 0
⎪
⎪
⎨1 ⎜0 2
ηU + ηV2 −ηQ ηU −ηV ηQ ⎟
−x ηI ⎜ ⎟
O(s, s ) = e ⎜ ⎟
⎪
⎪ η2 ⎝0 −ηQ ηU ηV2 + ηQ
2
−ηU ηV ⎠
⎪
⎩
0 −ηV ηQ −ηU ηV 2
ηQ 2
+ ηU
⎛ ⎞
η2 0 0 0
1 ⎜ ηV ηQ ⎟
2
⎜ 0 ηQ ηQ ηU ⎟
+ cosh(xη) 2 ⎜ ⎟
η ⎝ 0 ηQ ηU 2
ηU ηU ηV ⎠
0 ηV ηQ ηU ηV ηV2
⎛ ⎞⎫
0 ηQ ηU ηV ⎪
⎪
⎜
1 ⎜η ⎟⎪⎬
0 0 0 ⎟
− sinh(xη) ⎜ Q ⎟ , (8.45)
η ⎝ ηU 0 0 0 ⎠⎪⎪
⎪
⎭
ηV 0 0 0
where x = (s − s ).
b) purely dispersive media
In this case one has
Λ+ (0, ρ
) = 0
) = ρ2 =
Λ− (0, ρ ρ2Q + ρ2U + ρ2V = ρ ,
and the evolution operator is
⎛ 2 ⎞
ρ 0 0 0
⎜
1 ⎜ 0 ρ2Q ρQ ρU ρV ρQ ⎟
⎟
O(s, s ) = 2 ⎜ ⎟
ρ ⎝ 0 ρQ ρU ρ2U ρU ρV ⎠
0 ρV ρQ ρU ρV ρ2V
⎛ ⎞
0 0 0 0
⎜
1 ⎜ 0 ρ 2
+ ρ2V −ρQ ρU −ρV ρQ ⎟
U ⎟
+ cos(xρ) 2 ⎜ ⎟
ρ ⎝ 0 −ρQ ρU ρ2V + ρ2Q −ρU ρV ⎠
0 −ρV ρQ −ρU ρV ρ2Q + ρ2U
⎛ ⎞
0 0 0 0
⎜
1 ⎜0 −ρU ⎟
0 ρV ⎟
− sin(xρ) ⎜ ⎟, (8.46)
ρ ⎝0 −ρV 0 ρQ ⎠
0 ρU −ρQ 0
where x = (s − s ).
364 CHAPTER 8
8.6. Evolution Operator and Mueller Matrices
The concept of Mueller matrix is a very useful tool that is often used to characterize,
from a mathematical point of view, the properties of an optical device with respect
to the polarization of the light that is crossing the device itself. If I is the Stokes
vector of the light beam that is entering the device, and I the Stokes vector at the
exit of the device, the Mueller matrix is defined by
I = M I . (8.47)
This concept, introduced by Mueller (1948), provides a matrix-algebraic method
– the so-called Mueller calculus – for computing the effect of a train of several
devices on a light beam. If MA , MB , . . . , MN are the Mueller matrices of the
single optical devices, and if the light beam crosses the train in the order A, B,
. . . , N, we have, with evident notations
I = MN · · · MB MA I . (8.48)
Comparison of Eq. (8.47) or (8.48) with Eq. (8.6) shows that the Mueller matrix of
an optical device is nothing but the evolution operator describing the propagation of
radiation inside the material medium of which the device is composed.1 Therefore,
the expression of the Mueller matrix can be deduced once the geometrical and
optical properties of the medium are specified.
As an example, let us consider the case of a dichroic linear polarizer of geometrical
thickness l. The medium composing the polarizer can be characterized by a triplet
of real, orthogonal unit vectors (ut , ua , up ) – parallel to the transmission axis,
the ‘absorption’ axis and the propagation direction, respectively – such that the
corresponding indices of refraction are
nt = 1 + i kt , na = 1 + i ka
(the index np is unimportant). Using the notations of Sect. 5.1, the unit vectors
uα are given by2
uα = uα (α = t, a, p) .
Aligning the reference direction with the unit vector ut , we obtain for the propa-
gation tensor (see Eq. (5.22))
ν ν
G11 = −2π i n , G22 = −2π i n , G12 = G21 = 0 ,
c t c a
1
Actually, the concept of Mueller matrix is broader, because it can also be used to characterize
optical devices where light is partially reflected or deviated laterally. Of course the identification
of Mueller matrix and evolution operator does not hold in these cases.
2
Note that the medium composing a dichroic polarizer is a very special case of an absorbing
anisotropic medium.
RADIATIVE TRANSFER FOR POLARIZED RADIATION 365
so that Eqs. (5.26) give
ν ν
ηI = 2π kt + ka , ηQ = 2π kt − ka
c c
ηU = ηV = ρQ = ρU = ρV = 0 .
The Mueller matrix of the polarizer, MP , can be deduced directly from Eq. (8.45)
by substituting l for x . We obtain
⎧⎛ ⎞ ⎛ ⎞
⎪ 0 0 0 0 1 0 0 0
⎨
−l ηI ⎜0 0 0 0⎟ ⎜0 1 0 0⎟
MP = e ⎝ ⎠ + cosh(l ηQ ) ⎝ ⎠
⎪
⎩ 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
⎛ ⎞⎫
0 1 0 0 ⎪
⎬
⎜1 0 0 0⎟
− sinh(l ηQ ) ⎝ ⎠
0 0 0 0 ⎪ ⎭
0 0 0 0
⎛ ⎞
Kt + Ka Kt − Ka 0 0
⎜
1 ⎜ t − Ka
K Kt + Ka 0 0 ⎟
⎟
= ⎜ ⎟,
2⎝ 0 0 2 Kt Ka 0 ⎠
0 0 0 2 Kt Ka
where
−4π ν
c kt l −4π ν
c ka l
Kt = e , Ka = e .
For an ideal polarizer, one has Kt = 1 (no absorption along the transmission axis)
and Ka = 0 (complete absorption along the orthogonal axis), thus
⎛ ⎞
1 1 0 0
1 ⎜1 1 0 0⎟
MP = ⎝ ⎠,
2 0 0 0 0
0 0 0 0
which is the correct Mueller matrix for an ideal polarizer with its transmission axis
parallel to the reference direction (see e.g. Clarke and Grainger, 1971).
As a further example, let us consider the Mueller matrix of a retarder. In this case
the medium is characterized by a triplet of real, orthogonal unit vectors (uf , us , up )
– parallel to the fast axis, the slow axis and the propagation direction – such that
the corresponding indices of refraction are given by
n f = 1 + δf , ns = 1 + δ s
with δs > δf (again the index np is unimportant). Aligning the reference direction
with the unit vector uf , we have from Eq. (5.22)
ν ν
G11 = −2π i n , G22 = −2π i n , G12 = G21 = 0 ,
c f c s
366 CHAPTER 8
and from Eqs. (5.26)
ν
ρQ = − 2π δ − δf
c s
ηI = ηQ = ηU = ηV = ρU = ρV = 0 .
The Mueller matrix of the retarder, MR , is derived from Eq. (8.46). After some
easy algebra we get
⎛ ⎞
1 0 0 0
⎜0 1 0 0 ⎟
MR = ⎝ ⎠,
0 0 cos δ sin δ
0 0 − sin δ cos δ
where δ is the retardance defined in Eq. (1.28)
ν l
δ = −ρQ l = 2π δs − δf l = 2π ns − nf
c λ
(l is the thickness of the retarder).
Using the method outlined above, it is possible to obtain the Mueller matrix of
any optical device once the geometrical and optical properties of its constituents
are known.
8.7. Perturbative Solution of the Radiative Transfer Equations
In many practical applications, it is convenient to express the propagation matrix K
and the emission vector ε as the sum of a leading (or zero-order) term plus a
perturbation,
K(s) = K0 (s) + δK(s)
ε(s) = ε0 (s) + δε(s) .
Equation (8.2) takes then the form
d
I(s) = − K0 (s) + δK(s) I(s) + ε0 (s) + δε(s) . (8.49)
ds
We look for a solution to this equation of the form
I(s) = I0 (s) + δI(s) ,
where I0 (s) is the solution of the ‘zero-order’ equation
d
I (s) = −K0 (s) I0 (s) + ε0 (s) (8.50)
ds 0
RADIATIVE TRANSFER FOR POLARIZED RADIATION 367
and where δI(s) satisfies the equation (resulting from Eqs. (8.49) and (8.50))
d
δI(s) = − K0 (s) δI(s) − δK(s) I0 (s) − δK(s) δI(s) + δε(s) . (8.51)
ds
If the perturbation is ‘everywhere small’, or, in other words, if
δK(s) K0 (s) and δε(s) ε0 (s)
for any s , we should expect that
δI(s) I0 (s) ,
so that the third term in the right-hand side of Eq. (8.51), being a ‘second-order’
term, can be neglected. We have therefore
d
δI(s) = − K0 (s) δI(s) + δε(s) − δK(s) I0 (s) . (8.52)
ds
Equations (8.50) and (8.52) can be solved using the evolution operator O0 (s, s ) as-
sociated with the unperturbed propagation matrix K0 (s). For a medium extending
in s from −∞ to s , one gets from Eq. (8.15)
s
I0 (s) = O0 (s, s ) ε0 (s ) ds (8.53)
−∞
s
δI(s) = O0 (s, s ) δε(s ) − δK(s ) I0 (s ) ds
−∞
s
= O0 (s, s ) δε(s ) ds
−∞
s s
− ds O0 (s, s ) δK(s ) ds O0 (s , s ) ε0 (s ) . (8.54)
−∞ −∞
Equations (8.53) and (8.54) solve, in a perturbative way, the transfer problem.
This method of solution is especially useful when the operator O0 can be written
in analytical form. Moreover, through an adaptation of this same method, it is
possible to introduce the concept of response function, which will be done in the
following chapter (see Sect. 9.16) in connection with the problem of line formation
in a magnetic field.
368 CHAPTER 8
8.8. An Alternative Form of the Radiative Transfer Equations
The transfer equation (8.2) can be cast into a different form by introducing, in the
place of the Stokes vector I , a 2 × 2 matrix which is strictly related to the polar-
ization tensor. Indeed, both the classical derivation of Sect. 5.2 and the quantum-
mechanical derivation of Sect. 6.6 lead quite naturally to a transfer equation for the
polarization tensor (cf. Eqs. (5.24) and (6.78)). These equations were later con-
verted into a transfer equation for the Stokes vector, which in both cases resulted
in being of the form of Eq. (8.2).
In this section we will proceed in the opposite direction, showing in full generality
how it is possible to transform Eq. (8.2) into a transfer equation having the form
of Eq. (5.24) or (6.78).
From the Stokes vector I we define the tensor Ĩ via the equation (cf. Eqs. (1.40)
and (5.129))
1
3
I˜αβ = (τ ) I (α, β = 1, 2) , (8.55)
2 i=0 i αβ i
where the matrices τi are any representation of the Pauli spin matrices. In particu-
lar, we can take τi = σi or τi = σ̂i (where σi and σ̂i are given by Eqs. (1.17) and
(5.128), respectively) or, more generally, τi = x σi x−1 , where x is any unitary
2 × 2 matrix (x−1 = x† ).1
Whatever representation is chosen, the matrices τi are Hermitian,
† †
τi† = x σi x−1 = x−1 σi† x† = x σi x−1 = τi ,
thus the tensor Ĩ is also Hermitian
1 1
3 3
∗
I˜αβ = (τi )∗αβ Ii = (τ ) I = I˜βα .
2 i=0 2 i=0 i βα i
Moreover, from Eq. (1.18) we have
τj τk = x σj x−1 x σk x−1 = x δjk σ0 + i jkl σl x−1
l
= δjk τ0 + i jkl τl (j, k, l = 1, 2, 3) , (8.56)
l
where τ0 is the 2 × 2 identity matrix, and from Eq. (1.19)
Tr τj τk = Tr x σj x−1 x σk x−1 = Tr x−1 x σj σk
= 2 δjk (j, k = 0, 1, 2, 3) , (8.57)
1
The arbitrariness of the matrices τi is strictly connected with the arbitrariness of the choice
of the unit vectors defining the polarization tensor (cf. Sect. 1.8).
RADIATIVE TRANSFER FOR POLARIZED RADIATION 369
which allows to find the inverse relation of Eq. (8.55)
Ii = (τi )αβ I˜βα . (8.58)
αβ
Deriving Eq. (8.55) with respect to s , and using Eqs. (8.2) and (8.58), one obtains
the transfer equation for I˜αβ
1
3 3
d ˜
Iαβ (s) = − (τi )αβ (τj )γδ Kij (s) I˜δγ (s) + eαβ (s) , (8.59)
ds 2 i=0 j=0
γδ
where
1
3
eαβ (s) = (τ ) ε (s) .
2 i=0 i αβ i
We now observe that, because of its symmetry (and reality) properties, the ma-
trix K can be written in the following form, suggested by Eq. (6.84)
Kij = Re Tr (τj τi q) , (8.60)
where q is a 2 × 2 matrix implicitly defined by this equation. The expression of q
can be obtained by inversion of Eq. (8.60). This can be done in the easiest way by
noticing that any representation of the Pauli spin matrices forms a complete basis
for the 2 × 2 matrices. Writing then
3
q= ak τk (8.61)
k=0
one gets
3
Kij = Re ak Tr (τj τi τk ) .
k=0
The trace is easily evaluated from Eqs. (8.56) and (8.57), which give
Tr (τm τn τp ) = 2 i mnp (m, n, p = 1, 2, 3) (8.62)
(if one of the indices is zero, the trace is given directly from Eq. (8.57)). We obtain
ηI = K00 = K11 = K22 = K33 = 2 Re (a0 )
ηQ = K01 = K10 = 2 Re (a1 )
ηU = K02 = K20 = 2 Re (a2 )
ηV = K03 = K30 = 2 Re (a3 )
ρQ = K23 = −K32 = 2 Im (a1 )
ρU = K31 = −K13 = 2 Im (a2 )
ρV = K12 = −K21 = 2 Im (a3 ) ,
370 CHAPTER 8
or1
1 1
a0 = η a2 = ηU + i ρU
2 I 2
1 1
a1 = ηQ + i ρQ a3 = ηV + i ρV . (8.63)
2 2
Substituting Eq. (8.60) into Eq. (8.59), and using the remarkable property of the
Pauli spin matrices
3
(τi )αβ (τi )µν = 2 δαν δβµ , (8.64)
i=0
one finds after some algebra
d ˜
Iαβ (s) = − qαγ (s) I˜γβ (s) + qβγ (s)∗ I˜αγ (s) + eαβ (s) ,
ds γ
or, in matrix form
d
Ĩ(s) = − q(s) Ĩ(s) − Ĩ(s) q † (s) + e(s) , (8.65)
ds
†
where q † , the transpose conjugate of q ( qαβ ∗
= qβα ), is
3
†
q = a∗k τk .
k=0
The above derivation shows that the radiative transfer equation (8.65) is fully
equivalent to Eq. (8.2).
8.9. Solution of the Alternative Form of the Radiative
Transfer Equations
Equation (8.65) can be formally solved by the introduction of an evolution operator,
similarly to what done in Sect. 8.2 for the solution of Eq. (8.2). This operator
– which will be referred to in the following as reduced evolution operator – has
been introduced by Van Ballegooijen (1985, 1987) for a particular representation
of the Pauli spin matrices.
1
The inversion of Eq. (8.60) leaves unspecified the imaginary part of a0 . From the mathematical
point of view this indetermination is not surprising, because the complex 2 × 2 matrix q is defined
by eight real numbers, while the matrix K contains just seven independent (real) quantities. From
the physical point of view, the indetermination is related to the fact that the transfer equations
do not depend on the absolute phases of the components of the electric field. In Eqs. (8.63) we
have set Im (a0 ) = 0, but the following results are unaffected by the value of Im (a0 ).
RADIATIVE TRANSFER FOR POLARIZED RADIATION 371
Considering first the homogeneous equation
d
Ĩ(s) = − q(s) Ĩ(s) − Ĩ(s) q † (s) , (8.66)
ds
we define the reduced evolution operator E(s, s ) as the 2×2 complex matrix which,
acting on Ĩ(s ) , gives Ĩ(s) (with s ≥ s ) through the expression
Ĩ(s) = E(s, s ) Ĩ(s ) E † (s, s ) , (8.67)
with
E(s, s) = 1 (8.68)
1 being the 2 × 2 identity matrix) and
(1
E(s, s ) = E(s, s ) E(s , s ) (s ≥ s ≥ s ) .
Derivation of Eq. (8.67) with respect to s gives
d d d †
Ĩ(s) = E(s, s ) Ĩ(s ) E † (s, s ) + E(s, s ) Ĩ(s ) E (s, s ) ,
ds ds ds
while substitution of Eq. (8.67) into the right-hand side of Eq. (8.66) gives
d
Ĩ(s) = − q(s) E(s, s ) Ĩ(s ) E † (s, s ) − E(s, s ) Ĩ(s ) E † (s, s ) q † (s) .
ds
Comparison of these two equations yields
d
E(s, s ) = − q(s) E(s, s ) . (8.69)
ds
In a similar way one finds
d
E(s, s ) = E(s, s ) q(s ) . (8.70)
ds
Because of the strict analogy between Eqs. (8.69)-(8.70) and Eqs. (8.9)-(8.10),
we can expand the operator E(s, s ) in the form (cf. Eq. (8.12))
∞ s s s & '
(−1)n
E(s, s ) = 1 + ds1 ds2 · · · dsn P q(s1 ) q(s2 ) · · · q(sn ) ,
n=1
n!
s s s
where P is the Dyson chronological product defined in Eq. (8.13). Moreover, it can
be easily proved, with the help of Eqs. (8.68) and (8.69), that the formal solution
of the non-homogeneous equation (Eq. (8.65)) is
s
Ĩ(s) = E(s, s ) e(s ) E † (s, s ) ds + E(s, s ) Ĩ(s ) E † (s, s ) .
s
372 CHAPTER 8
Analytical forms of the reduced evolution operator E(s, s ) can be found when-
ever
q(s1 ), q(s2 ) = 0 (8.71)
for any couple of points s1 , s2 in the interval (s , s). In this case one gets (cf.
Eq. (8.22)) : s
− q(s ) ds
E(s, s ) = e s . (8.72)
Similarly to Sect. 8.3, one can distinguish three different cases where Eq. (8.71)
is satisfied. As an example, let us consider the case q = const., where Eq. (8.72)
reduces to
−(s−s ) q
E(s, s ) = e .
Use of Eqs. (8.61) and (A5.6) gives
−(s−s ) a0 −(s−s )
a·
τ
E(s, s ) = e e , (8.73)
where
a = (a1 , a2 , a3 ) , τ = (τ1 , τ2 , τ3 ) ,
with ai given by Eqs. (8.63). The exponential in the right-hand side of Eq. (8.73)
can be expanded by a method similar to that of App. 5. The result is1
−1
2 x ηI
sinh(xa)
E(s, s ) = e cosh(xa) 1 − a · τ , (8.74)
a
where
x = (s − s )
√ 1
a = a2 = Λ (
η, ρ
) + i σ Λ− (η , ρ
)
2 +
with Λ± (
η, ρ
) and σ defined in Eqs. (A5.18) and (A5.19), respectively.
The reduced evolution operator E(s, s ) is of course related to the evolution
operator O(s, s ) defined in Sect. 8.2. To find this relation we start from Eq. (8.58),
written at point s , and substitute for Ĩ(s) its expression in terms of the reduced
evolution operator (Eq. (8.67)). Next we use Eq. (8.55), written at point s , and
compare with the definition (8.6) to obtain
1
Oij (s, s ) = Tr τi E(s, s ) τj E † (s, s ) (i, j = 0, 1, 2, 3) . (8.75)
2
This formula shows that the 16 real elements Oij are bilinear combinations of the
4 complex elements Eαβ . The explicit expressions are rather cumbersome and
depend on the representation used for the matrices τi .
1
If (η2 − ρ2 ) = η ·ρ = 0 (so that Λ+ (η , ρ ) = Λ− (η , ρ ) = 0), Eq. (8.74) should be replaced by
1
E(s, s ) = e − 2 x ηI
1 − x a ·
τ .
RADIATIVE TRANSFER FOR POLARIZED RADIATION 373
On the other hand, E(s, s ) depends also on the representation because of its
dependence on the matrix q (see Eqs. (8.69) and (8.61)). It is however possible
to obtain an ‘intrinsic’ relation (i.e. a relation independent of the representation)
between the operators O(s, s ) and E(s, s ) by expanding the latter on the ba-
sis {τi },1
3
E(s, s ) = E (k) (s, s ) τk . (8.76)
k=0
Substitution into Eq. (8.75) gives
1 (k)
3 3
Oij (s, s ) = E (s, s ) E (l) (s, s )∗ Tr τi τk τj τl , (8.77)
2
k=0 l =0
which clearly does not depend on the representation.
Equation (8.76) suggests a further important remark. Using Eq. (8.57), the
coefficients E (k) (s, s ) can be written in the form
1
E (k) (s, s ) = Tr τk E(s, s ) .
2
Derivation with respect to s leads, with the help of Eqs. (8.69), (8.61) and (8.76),
to the set of differential equations (independent of the representation)
1
3 3
d (k)
E (s, s ) = − a E (j) (s, s ) Tr τi τj τk . (8.78)
ds 2 i=0 j=0 i
It follows that the general problem of finding the evolution operator is equivalent
to solving a system of 8 linear, homogeneous, real differential equations in the
8 unknowns
Re E (k) (s, s ) , Im E (k) (s, s ) (k = 0, 1, 2, 3) ,
subjected to the boundary condition (see Eqs. (8.68) and (8.76))
Re E (k) (s , s ) = δk0 , Im E (k) (s , s ) = 0 .
Once the system is solved, the reduced evolution operator E(s, s ) can be computed
from Eq. (8.76), and the evolution operator O(s, s ) from Eq. (8.77).
Equations (8.78) and (8.77) are developed in detail in App. 7. Similar equations
have been obtained, with a different formalism, by Sánchez Almeida (1992).
1
In the case q = const. considered above, E(s, s ) is already expressed in this form (see
Eq. (8.74)).
CHAPTER 9
LINE FORMATION IN A MAGNETIC FIELD
The problem of line formation in a magnetic field is a classical one in the theory of
stellar atmospheres.1 In its most general formulation it consists in the following.
Consider a wavelength interval of the electromagnetic spectrum where one or sev-
eral spectral lines are present. The radiation contained in the interval flows through
a stellar atmosphere permeated by a magnetic field and interacts with the atoms
(responsible for those spectral lines) via the usual processes of absorption, emis-
sion (spontaneous and stimulated) and scattering. The result is, quite generally,
the formation of absorption (or emission) lines with typical polarization features.
The problem is to relate the polarization features of the emerging radiation, de-
scribed by its Stokes parameters, to the physical properties of the atmosphere and
to the magnetic field vector.
The overall problem is extremely complicated and ultimately requires a full
non-equilibrium (or non-LTE) approach, whose basic physics will be outlined in
Chap. 14. However, in most astrophysical applications only the deepest layers of
stellar atmospheres are involved. In that case, depolarizing collisions can be as-
sumed to be sufficiently strong to destroy any atomic polarization that might be
induced by the (anisotropic and polarized) radiation field. As we will see in the
following, this assumption greatly simplifies the general problem and, in particular,
allows the emission term in the radiative transfer equation to be described by a
scalar source function (which reduces to the Planck function in LTE). The present
chapter is devoted to the subject of line formation in a magnetic field under the
assumption of complete atomic depolarization.
9.1. Transfer Equation for Polarized Radiation in a
Magnetized Stellar Atmosphere
Consider a radiation beam of frequency ν that is flowing in a stellar atmosphere
along the direction Ω. If the frequency ν is close to the frequency ν of a spectral
0
line (produced by an atomic species present in the atmosphere), the radiation beam
undergoes two different kinds of interactions with the ambient medium: absorption
and emission processes due to the continuum, and the corresponding processes due
to the spectral line.
1
In this chapter we use the term ‘stellar atmosphere’ in a rather broad sense. Many of
the results that will be derived in the following sections can as well be applied to laboratory or
astrophysical plasmas of arbitrary geometrical shape, like slabs, cylinders, etc. In many cases,
these can indeed be regarded as ‘pieces’ of a stellar atmosphere.
376 CHAPTER 9
The theory of the continuum absorption coefficient in stellar atmospheres is de-
scribed in detail in several books and will not be repeated here (see e.g. Mihalas,
1978). Although such theory has been developed for non-magnetic atmospheres, it
can be shown that the presence of a magnetic field does not alter the frequency de-
pendence of the continuum absorption coefficient and does not introduce dichroism
or anomalous dispersion effects.1 Since this is also true for the emission coefficient,
the contribution of continuum processes to the radiative transfer equation for po-
larized radiation is of the form
⎛ ⎞ ⎛ ⎞⎛ ⎞
I kc 0 0 0 I − Sc
d ⎜⎜Q⎟
⎟ ⎜0 k
⎜ c 0 0⎟ ⎜
⎟⎜ Q ⎟
⎟
⎜ ⎟ = −⎜ ⎟⎜ ⎟, (9.1)
ds ⎝ U ⎠ ⎝ 0 0 kc 0 ⎠⎝ U ⎠
V 0 0 0 kc V
where I, Q, U , and V are the Stokes parameters of the radiation of frequency ν
s is the coordinate measured along the ray path, k
flowing along the direction Ω, c
is the local continuum absorption coefficient (corrected for stimulated emission) at
frequency ν and, finally, Sc is the continuum source function that, in most cases,
can be equated to the local Planck function BP .
We must now add to the transfer equation the contribution of the absorption
and emission processes in the spectral line. For electric-dipole transitions, this
contribution is given by Eq. (6.85), with the substitution
∂ d d
c + →
∂t ds ds
which applies to a stationary radiation field. The coefficients of this equation
depend – as illustrated in Chap. 7 – both on the physical characteristics of the line
and on the value of the magnetic field. The line can be ‘unstructured’ or formed by
the superposition of several fine-structure (or hyperfine-structure) components. On
the other hand, the splitting of the atomic levels will be described by the Zeeman or
Paschen-Back effect regime, depending on the magnetic field value (see Chap. 3).
Throughout this chapter – except Sect. 9.23, dealing with ‘structured’ lines – we
shall be concerned with the simplest situation, where absorption and emission at
frequency ν involve an isolated spectral line originating in the transition between
two levels (α J ) and (αu Ju ) of an atom devoid of hyperfine structure. We will
further assume that the magnetic field is weak enough for the Zeeman effect regime
to hold, and that no atomic polarization is present in the two levels. The latter
assumption means that the atomic density matrix in the subspace of the magnetic
quantum numbers is diagonal and proportional to the unit matrix.
This is a special case of the model that has been denoted in Chap. 7 as ‘multi-
level atom’. The radiative transfer coefficients in the standard representation are
1
This is indeed valid for the magnetic fields that are typically found in the solar atmosphere
or in magnetic, non-degenerate stars. The case of magnetic white dwarfs (where B ≥ 107 G) is
quite different (cf. Kemp, 1970; Lamb and Sutherland, 1974).
LINE FORMATION IN A MAGNETIC FIELD 377
θ
Ω
is the propagation
Fig.9.1. The magnetic field direction is specified by the angles θ and χ. Ω
direction of the radiation while ea is the reference unit vector.
given by Eqs. (7.19), where (because of the isolated spectral line assumption) the
summations over the atomic levels are limited to one lower level (α J ) and one
upper level (αu Ju ), and where (because of the no-atomic-polarization assumption)
we have to substitute
N (2J + 1) ρα J (M ) → N
N (2Ju + 1) ρα J (Mu ) → Nu ,
u u
N and Nu being the number of atoms per unit volume in the lower and upper
level, respectively. Equation (7.19a) reduces to
= hν N B(α J → α J )
ηiA (ν, Ω)
4π u u
J J 1
2
× φ(ν
Tqq (i, Ω) αu Ju Mu , α J M − ν) , (9.2)
3 u
−Mu M −q
M Mu q
where, according to Eq. (7.3)
να J Mu ,α J M = ν0 + νL (gu Mu − g M ) , (9.3)
u u
with ν0 the unperturbed frequency of the transition, νL the Larmor frequency, gu
and g the Landé factors of the upper and lower level, respectively.
The quantities Tqq (i, Ω) depend on the geometrical relations between the (local)
magnetic field vector, the propagation direction Ω, and the reference direction
chosen to define the Stokes parameters. Let us denote by θ and χ the inclination
and azimuth angles of the magnetic field (see Fig. 9.1). Recalling the definitions in
Sect. 5.11, we have to find the Euler angles of the rotation that brings the triplet
into a right-handed coordinate system (xyz) having the
of unit vectors (ea , eb , Ω)
z-axis along the magnetic field direction (our quantization axis). From Fig. 9.1 we
get
R ≡ (χ, θ, γ) ,
378 CHAPTER 9
where γ is an arbitrary angle (specifying the orientation of the x and y axes) which
These expressions can be deduced from
does not enter the expressions of Tqq (i, Ω).
Table 5.3 which, however, refers to the rotation R ≡ (−γ, −θ, −χ). Performing the
substitutions (γ → −χ , θ → −θ, χ → −γ), we obtain
= 1 1 + cos2 θ
= T (0, Ω)
T−1−1 (0, Ω) 11
4
= − 1 sin2 θ cos 2χ
= T (1, Ω)
T−1−1 (1, Ω) 11
4
= T (2, Ω) 1
= − sin2 θ sin 2χ
T−1−1 (2, Ω) 11
4
= −T (3, Ω)
T−1−1 (3, Ω) = − 1 cos θ
11
2
1
= sin2 θ
T00 (0, Ω)
2
= 1 sin2 θ cos 2χ
T00 (1, Ω)
2
= 1 sin2 θ sin 2χ
T00 (2, Ω)
2
T (3, Ω) =0.
00
Equation (9.2) can then be cast into the form (see also Eqs. (6.87))
= k A 1 φ sin2 θ + φ−1 + φ1 (1 + cos2 θ)
= η A (ν, Ω)
η0A (ν, Ω) I L
2 0 2
= k A 1 φ − φ−1 + φ1 sin2 θ cos 2χ
= η A (ν, Ω)
η1A (ν, Ω) Q L
2 0 2
= k A 1 φ − φ−1 + φ1 sin2 θ sin 2χ
= η A (ν, Ω)
η2A (ν, Ω) U L
2 0 2
= kA 1 φ − φ
= η A (ν, Ω)
η3A (ν, Ω) V L −1 cos θ , (9.4)
2 1
where
hν
kLA = N B(α J → αu Ju ) (9.5)
4π
is the frequency-integrated absorption coefficient in the line1 and where
2
Ju J 1
φq = 3 φ(να − ν) (q = −1, 0, +1) . (9.6)
−Mu M −q J Mu , α J M
u u
M Mu
1
Since the φq profiles are normalized to unity in frequency (cf. Eqs. (6.59c) and (2.23a)), we
have
ηIA (ν, Ω)
dν = k A .
L
line
LINE FORMATION IN A MAGNETIC FIELD 379
With strictly analogous substitutions one finds from Eq. (7.19c)
= k A 1 ψ − ψ−1 + ψ1 sin2 θ cos 2χ
= ρA (ν, Ω)
ρA1 (ν, Ω) Q L
2 0 2
= ρA (ν, Ω)
ρA2 (ν, Ω)
1 ψ
= k A ψ − −1 + ψ1 sin2 θ sin 2χ
U L
2 0 2
= ρA (ν, Ω) 1
= kA ψ − ψ
ρA3 (ν, Ω) V L −1 cos θ , (9.7)
2 1
where
2
Ju J 1
ψq = 3 ψ(να − ν) (q = −1, 0, +1). (9.8)
−Mu M −q u u
J Mu , α J M
M Mu
Equations (9.4) and (9.7) can be conveniently rewritten in the form
= k A φ (ν, Ω)
ηIA (ν, Ω)
L I
A
ηQ = k A φ (ν, Ω)
(ν, Ω) = k A ψ (ν, Ω)
ρAQ (ν, Ω)
L Q L Q
A
ηU = k A φ (ν, Ω)
(ν, Ω) = k A ψ (ν, Ω)
ρAU (ν, Ω)
L U L U
= k A φ (ν, Ω)
ηVA (ν, Ω) = k A ψ (ν, Ω)
ρAV (ν, Ω) ,
L V L V
with
= 1 φ sin2 θ + φ−1 + φ1 (1 + cos2 θ)
φI (ν, Ω)
2 0 2
= 1 φ − φ−1 + φ1 sin2 θ cos 2χ
φQ (ν, Ω)
2 0 2
= 1 φ − φ−1 + φ1 sin2 θ sin 2χ
φU (ν, Ω)
2 0 2
= 1 φ −φ
φV (ν, Ω) −1 cos θ
2 1
= 1 ψ − ψ−1 + ψ1 sin2 θ cos 2χ
ψQ (ν, Ω)
2 0 2
= 1 ψ − ψ−1 + ψ1 sin2 θ sin 2χ
ψU (ν, Ω)
2 0 2
= 1 ψ −ψ
ψV (ν, Ω) −1 cos θ . (9.9)
2 1
Using these notations, one gets for the stimulated emission coefficients (Eqs. (7.19b)
and (7.19d))
380 CHAPTER 9
= η S (ν, Ω)
η0S (ν, Ω) = k S φ (ν, Ω)
I L I
= η S (ν, Ω)
η1S (ν, Ω) = k S φ (ν, Ω)
Q L Q
= η S (ν, Ω)
η2S (ν, Ω) = k S φ (ν, Ω)
U L U
= η S (ν, Ω)
η3S (ν, Ω) = k S φ (ν, Ω)
V L V
= ρS (ν, Ω)
ρS1 (ν, Ω) = k S ψ (ν, Ω)
Q L Q
= ρS (ν, Ω)
ρS2 (ν, Ω) = k S ψ (ν, Ω)
U L U
= ρS (ν, Ω)
ρS3 (ν, Ω) = k S ψ (ν, Ω)
,
V L V
where
hν
kLS = N B(αu Ju → α J ) ,
4π u
and for the spontaneous emission coefficients (Eq. (7.19e); see also Eqs. (7.8))
= ε (ν, Ω)
ε0 (ν, Ω) = ε φ (ν, Ω)
I L I
= ε (ν, Ω)
ε1 (ν, Ω) = ε φ (ν, Ω)
Q L Q
= ε (ν, Ω)
ε2 (ν, Ω) = ε φ (ν, Ω)
U L U
= ε (ν, Ω)
ε3 (ν, Ω) = ε φ (ν, Ω)
, (9.10)
V L V
where
hν
εL = N A(αu Ju → α J ) . (9.11)
4π u
Collecting all the various terms, the contribution of the line to the radiative transfer
equation for polarized radiation takes the form
⎛ ⎞ ⎛ ⎞⎛ ⎞
I φI φQ φU φV I − SL
d ⎜⎜Q⎟
⎟ ⎜φ
⎜ Q φI ψV −ψU ⎟ ⎜
⎟⎜ Q ⎟
⎟
⎜ ⎟ = − kL ⎜ ⎟⎜ ⎟, (9.12)
ds ⎝ U ⎠ ⎝ φU −ψV φI ψQ ⎠⎝ U ⎠
V φV ψU −ψQ φI V
where
kL = kLA − kLS
is the (frequency-integrated) line absorption coefficient corrected for stimulated
emission, and where
ε
SL = A L S (9.13)
kL − kL
is the line source function. Using the well-known relations among the Einstein
coefficients (see Eqs. (7.8)), these two quantities can be rewritten as
−1
Nu g̃ 2hν03 N g̃u
kL = kL 1 −
A
, SL = 2 −1 , (9.14)
N g̃u c Nu g̃
LINE FORMATION IN A MAGNETIC FIELD 381
where g̃ = (2J + 1) and g̃u = (2Ju + 1) are the degeneracies of the lower and upper
level, respectively. In the particular case of LTE, one simply gets
hν0 hν0 −1
− 2hν03
kL = kLA 1 − e kB T , SL = e kB T − 1 = BP (ν0 ) .
c2
Adding the contributions of continuum processes and line processes, one obtains
from Eqs. (9.1) and (9.12)
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 0 0 0 I − Sc
⎜
d ⎜ ⎟ ⎟ ⎜ 0⎟ ⎜ ⎟
Q ⎜0 1 0 ⎟⎜ Q ⎟
⎜ ⎟ = − kc ⎜ ⎟⎜ ⎟
ds ⎝ U ⎠ ⎝0 0 1 0 ⎠⎝ U ⎠
V 0 0 0 1 V
⎛ ⎞⎛ ⎞
φI φQ φU φV I − SL
⎜φ −ψU ⎟ ⎜ ⎟
⎜ φI ψV ⎟⎜ Q ⎟
− kL ⎜ Q ⎟⎜ ⎟. (9.15)
⎝ φU −ψV φI ψQ ⎠⎝ U ⎠
φV ψU −ψQ φI V
It should be noticed that the equations derived in this section prove some prop-
erties of the Zeeman patterns that were anticipated in Sect. 3.1. First of all, the
φq and ψq profiles in Eqs. (9.6) and (9.8) can be written in the form
J J
φq = Sq u (M , Mu ) φ(να J Mu , α J M − ν)
u u
M Mu
J J
ψq = Sq u (M , Mu ) ψ(να J Mu , α J M − ν) , (9.16)
u u
M Mu
J J
where Eq. (3.16) has been used. Thus the quantity Sq u (M , Mu ) is indeed the
relative strength of the Zeeman component (σb , π or σr according as q = −1, 0,
+1, respectively) connecting the sublevels (J M ) and (Ju Mu ). The strengths are
normalized as in Eq. (3.17) and satisfy the relations (3.18)-(3.19). Furthermore,
the polarization features of the radiation emitted by each type of component can
be deduced from Eq. (9.10), and it is easily seen that they are just as illustrated
in Sect. 3.1 (cf. Figs. 9.1 and 3.1).
9.2. Comparison with the Classical Theory
We are now in a position to compare the line transfer equation derived from Quan-
tum Mechanics with the corresponding equation derived from classical physics in
Chap. 5. Direct comparison of Eqs. (9.12) and (5.39) shows that the structure
of the two equations is exactly the same and allows us to relate the line source
382 CHAPTER 9
function SL , as given by Eq. (9.14), with the ‘classical’ source function defined
in Eq. (5.40). Equating the two expressions (and neglecting in SL the contribu-
tion due to stimulated emission, a phenomenon that has no classical analogue)
one finds a relationship between the average energy per degree of freedom of the
classical oscillator, Ē, and the number density of atoms in the lower and upper
level
N g̃
Ē = hν0 u u .
N g̃
This is just the expression that is obtained for a two-level atom having a proba-
bility Nu /g̃u of being found in the upper level (of energy hν0 ) and a much larger
probability N /g̃ of being found in the lower level (of zero energy).
Comparison of Eqs. (9.12) and (5.39) also shows that the elements of the propa-
gation matrix coincide provided the following transformations are performed (see
Eqs. (9.9) and (5.36), and Figs. 9.1 and 5.1)
πe20
N → kL (9.17)
mc
and
φb → φ−1 ψb → ψ−1
φp → φ0 ψp → ψ0
φr → φ1 ψr → ψ1 . (9.18)
Transformation (9.17) can be easily interpreted if we neglect again stimulated emis-
sion and write the frequency-integrated line absorption coefficient in the form
πe20
kL → kLA = N f (α J → αu Ju ) , (9.19)
mc
where f (α J → αu Ju ) is the oscillator strength of the transition. The dimension-
less parameter f can be regarded as an efficiency parameter for the transition:
multiplication of the number of atoms in the lower level by f yields the equivalent
number of classical oscillators. Comparison of Eqs. (9.19) and (9.5) shows that the
oscillator strength is given by
hν mc
f (α J → αu Ju ) = B(α J → αu Ju ) ,
4π 2 e20
or, using Eqs. (7.6) and (7.7)
8π 2 mν
f (α J → αu Ju ) = | α J d αu Ju |2
3 h e20
1 mc3 2Ju + 1
= 2 A(αu Ju → α J ) . (9.20)
8π e20 ν 2 2J + 1
LINE FORMATION IN A MAGNETIC FIELD 383
The oscillator strength is generally a quite small number which can attain a value
of order unity only for the strongest spectral lines.
The correspondence between classical oscillators and atomic transitions also ap-
pears from transformation (9.18). Some qualitative aspects of this correspondence
have been anticipated in Sect. 3.2. More quantitatively, recalling Eqs. (9.3) and
(6.59a), and neglecting the frequency shifts, we have from Eqs. (9.16)
J J 1 Γ
φq = Sq u (M , Mu )
π Γ 2 + [ ν0 + νL (gu Mu − g M ) − ν ]2
M Mu
J J 1 ν0 + νL (gu Mu − g M ) − ν
ψq = Sq u (M , Mu ) , (9.21)
π Γ 2 + [ ν0 + νL (gu Mu − g M ) − ν ]2
M Mu
where Γ , defined in Eq. (6.59b), is the natural width of the line
γ + γu
Γ = ,
4π
with γ and γu the inverse lifetimes of the lower and upper level, respectively. In
particular, for a normal Zeeman triplet having unit Landé factor (g = 1 if Ju = 0,
or gu = 1 if J = 0), we simply have
ν0 + νL (gu Mu − g M ) − ν = ν0 + νL (Mu − M ) − ν = ν0 − νL q − ν ,
and performing the sum over M and Mu via Eq. (3.17)
1 Γ
φq =
π Γ 2 + (ν0 − νL q − ν)2
1 ν0 − νL q − ν
ψq = .
π Γ 2 + (ν0 − νL q − ν)2
Since the components σb , π, σr correspond to q = −1, 0, +1 respectively, these
profiles are just the same as the ‘classical’ profiles given by Eqs. (5.37).
Therefore, as far as the absorption and anomalous dispersion profiles are con-
cerned, the theory developed from the principles of Quantum Electrodynamics
leads, for the case of a normal Zeeman triplet with unit Landé factor, to the same
results as the classical theory. The latter is however unable to describe the more
general case of anomalous Zeeman patterns. This can only be done by consider-
ing the strengths and splittings of the different components, as derived from the
quantum-mechanical theory.
9.3. Collisional and Doppler Broadening
The absorption and anomalous dispersion profiles given by Eqs. (9.21) are still
inadequate to describe the physical situation of a stellar atmosphere. This is be-
cause the quantum theory presented in Chap. 6 is based on the assumption that
384 CHAPTER 9
the material system (the atom in this case) interacting with the radiation field is
isolated and at rest. The obvious consequence is that the absorption and anoma-
lous dispersion profiles are broadened by a single mechanism, namely natural (or
radiation) damping.
Collisional broadening can be carried into our formalism in a rather straightfor-
ward way. The effect of collisions is simply described by replacing Γ , the natural
line width appearing in Eqs. (9.21), with the sum Γ of the natural and collisional
width, defined by
Γ = Γ + 2Γc . (9.22)
This result has been proved in Sect. 5.14 within the framework of the classical
theory (see Eq. (5.180)). It still holds, however, when more refined, quantum-
mechanical models of the atom-radiation interaction are considered (see for instance
Loudon, 1983).
To get an estimate of the collisional width, we recall that Γc is proportional to
the frequency f of perturbing collisions (Γc = f /4π, see Sect. 5.14), and that f
has already been estimated in Sect. 7.13.c in connection with our discussion of
depolarizing rates.1 If we assume that perturbing collisions are mainly due to
long-range interactions with neutral perturbers (hydrogen atoms),2 we get from
Eqs. (5.181) and (7.105)3
0.4 0.6
Γc = 0.68 C6 (αu Ju ) v̄r nH ,
where C6 (αu Ju ) is the Van der Waals constant for the upper level involved in the
transition, v̄r the average relative velocity atom-perturber, and nH the number
density of hydrogen atoms. Using Eq. (7.108) we can also write
−11
2 0.4 1
0.3
Γc 1.1 × 10 r (αu Ju ) T 1+ nH s−1 ,
µ
where r2 (αu Ju ) is the mean square radius of the electronic cloud in the level
(αu Ju ) expressed in atomic units, T is the temperature in K, µ is the atomic weight
of the atom, and nH is expressed in cm−3 .
When collisions are taken into account, the profiles φq and ψq in Eqs. (9.21)
become
1
It should be remarked that both the phenomena of collisional broadening and atomic-level
depolarization are basically due to the same physical mechanism, as shown in Sect. 5.14.
2
By so doing, we neglect collisions with electrons and ions. In typical stellar atmospheres,
their contribution to collisional line broadening is quite modest (with the exception of hydrogen
and helium lines). An excellent discussion of collisional broadening due to charged particles can
be found in Mihalas (1978) or in Griem (1974).
3
This result agrees with the often quoted expression of Unsöld (1955)
2/5
ΓU 17 C6 v̄r3/5 nH .
The definition of ΓU is such that ΓU = 8π Γc .
LINE FORMATION IN A MAGNETIC FIELD 385
J J 1 Γ
φq = Sq u (M , Mu )
π Γ + [ ν0 + νL (gu Mu − g M ) − ν ]2
2
M Mu
J J 1 ν0 + νL (gu Mu − g M ) − ν
ψq = Sq u (M , Mu )
,
π Γ + [ ν0 + νL (gu Mu − g M ) − ν ]2
2
M Mu
where
Γ = Γ + 2Γc .
As far as the motion of the atomic system is concerned, we already noticed (see
the discussion following Eq. (6.31)) that the quantum theory presented in Chap. 6
cannot account for the Doppler effect. However, this effect can be reintroduced in
our formalism by the same arguments outlined in Sect. 5.4. By so doing we obtain
J J 1 1
φq = Sq u (M , Mu ) √ H v − vA + vB (gu Mu − g M ), a
π ∆νD
M Mu
J J 1 1
ψq = Sq u (M , Mu ) √ L v − vA + vB (gu Mu − g M ), a , (9.23)
π ∆νD
M Mu
where the functions H(v, a) and L(v, a) are defined in Eqs. (5.45), and where
wT Γ
∆νD = ν0 , a=
c ∆νD
νL wA ν w ν0 − ν
vB = , vA = = 0 A , v= . (9.24)
∆νD wT c ∆νD ∆νD
Note that these equations are the same as Eqs. (5.43) except for the definition of the
damping constant a, where Γ is substituted for Γ in order to allow for collisional
broadening. We recall that wA and wT represent the line-of-sight component of
the bulk velocity of the ambient medium and the thermal velocity, respectively.
As mentioned in Sect. 5.4, the various quantities entering the definitions of the
dimensionless parameters v, vA , vB , a can be expressed in wavelength units rather
than in frequency units. Introducing the Doppler width in wavelength units
λ20 w
∆λD = ∆νD = λ0 T , (9.25)
c c
we can write (cf. Eqs. (5.47))
λ − λ0 λ20 Γ λ0 wA
v= , a= , vA =
∆λD c ∆λD c ∆λD
λ20 νL λ20 e0 B ∆λB
vB = = 2
= . (9.26)
c ∆λD 4πmc ∆λD ∆λD
386 CHAPTER 9
A numerical expression for vB is given in Eq. (5.49).
9.4. Different Forms of the Transfer Equation
For practical applications, and especially for those concerning the diagnostics of
magnetic fields in the solar and stellar atmospheres, it may be convenient to write
the transfer equation (9.15) in other forms, using different definitions for the ab-
sorption and anomalous dispersion profiles and/or substituting optical depth for
geometrical depth.
Consider first the propagation matrix multiplying kL in the right-hand side of
Eq. (9.15). This matrix contains the seven independent quantities φI , φQ , φU , φV ,
ψQ , ψU , ψV defined in Eqs. (9.9), which are linear combinations of the profiles φq
or ψq defined in Eqs. (9.23). It can be easily shown via Eqs. (3.17) and (5.60) that
the profiles φq are normalized to unity in frequency,
∞
φq dν = 1 .
−∞
It is customary to introduce different profiles, ηq , normalized to unity in v, the
reduced frequency (or wavelength) defined in Eqs. (9.24)-(9.26). Setting
ηq = ∆νD φq
J J 1
= Sq u (M , Mu ) √ H v − vA + vB (gu Mu − g M ), a (9.27a)
π
M Mu
and, by analogy
ρq = ∆νD ψq
J J 1
= Sq u (M , Mu ) √ L v − vA + vB (gu Mu − g M ), a , (9.27b)
π
M Mu
we obviously have
∞
ηq dv = 1 , (9.28)
−∞
and introducing the notation
ηb = η−1 ρb = ρ−1
ηp = η0 ρp = ρ0
ηr = η1 ρr = ρ1 , (9.29)
LINE FORMATION IN A MAGNETIC FIELD 387
the transfer equation for polarized radiation can be written in the form
⎛ ⎞ ⎧⎛ ⎞⎛ ⎞
I ⎪
⎪ 1 0 0 0 I − Sc
⎪ ⎜
⎜
d ⎜Q⎟ ⎟ ⎨
⎜0 1 0 0⎟ ⎜
⎟⎜ Q ⎟
⎟
⎜ ⎟ = − kc ⎜ ⎟⎜ ⎟
ds ⎝ U ⎠ ⎪
⎪ ⎝0 0 1 0 ⎠⎝ U ⎠
⎪
⎩
V 0 0 0 1 V
⎛ ⎞⎛ ⎞⎫
hI hQ hU hV I − SL ⎪
⎪
⎜h −rU ⎟ ⎜ ⎟⎪
⎬
⎜ hI rV ⎟⎜ Q ⎟
+ κL ⎜ Q ⎟⎜ ⎟ , (9.30)
⎝ hU −rV hI rQ ⎠⎝ U ⎠⎪ ⎪
⎪
⎭
hV rU −rQ hI V
where κL , the ratio between the frequency-integrated line absorption coefficient
(corrected for stimulated emission and expressed in Doppler width units) and the
continuum absorption coefficient at the line wavelength, is given by (see Eqs. (9.14)
and (9.19))
kL πe2 N f (α J → αu Ju ) Nu g̃
κL = = 0 1− , (9.31)
kc ∆νD mc kc ∆νD N g̃u
and where
1 ηb + ηr
hI = ηp sin2 θ + 1 + cos2 θ
2 2
1 ηb + ηr
hQ = ηp − sin2 θ cos 2χ
2 2
1 η + ηr
hU = ηp − b sin2 θ sin 2χ
2 2
1
hV = ηr − ηb cos θ
2
1 ρb + ρr
rQ = ρp − sin2 θ cos 2χ
2 2
1 ρ + ρr
rU = ρp − b sin2 θ sin 2χ
2 2
1
rV = ρr − ρb cos θ . (9.32)
2
With the present definitions,1 the ratio between the line absorption coefficient at
line center and the continuum absorption coefficient is given, for zero magnetic field
1
Note that κL , as well as hI , hQ , hU , hV , rQ , rU , rV , are dimensionless quantities. For
the sake of brevity, the parameter κL will be sometimes referred to in the following as the ‘ratio
between line and continuum absorption coefficients’.
388 CHAPTER 9
ray path
top
α
stellar
atmosphere
bottom
Fig.9.2. The coordinates s and z generally used for radiative transfer in stellar atmospheres.
√
and no damping, by κL / π (see Eqs. (5.55), (9.27), (3.17) and (9.32)). Different
definitions can be found in the√literature, where the symbol √ η0 is often used to
represent the ratio kL /(kc ∆νD π ) (thus resulting a factor π smaller than κL ),
and where the profiles ηb , ηp , ηr , ρb , ρp , ρr are defined as in Eqs. (9.29) and
√ √
(9.27) without the factor 1 / π (thus resulting a factor π larger than ours).
When dealing with radiative transfer in astrophysical plasmas, it is generally
convenient to refer to the familiar concept of optical depth. The coordinate s mea-
sured along the ray path can be replaced by the optical depth τc in the continuum
adjacent to the line, or by the optical depth τr in the continuum at a given reference
frequency νr , or by the optical depth τL in the line. These quantities are defined
by
kL
dτc = −kc (ν0 ) ds , dτr = −kc (νr ) ds , dτL = − ds , (9.33)
∆νD
where the minus sign means that the increasing direction for optical depth is op-
posite to the propagation direction. For plane-parallel atmospheres, optical depths
can be referred to the coordinate z measured outward along the vertical to the
atmosphere. If we denote by t such optical depths and if α is the angle defined in
Fig. 9.2, we obviously have
dtc = −kc (ν0 ) dz = µ dτc
dtr = −kc (νr ) dz = µ dτr
k
dtL = − L dz = µ dτL , (9.34)
∆νD
where µ = cos α.
Introducing, for instance, the optical depth τc , Eq. (9.30) becomes
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 0 0 0 I − Sc
d ⎜ ⎟ ⎜
⎜ Q ⎟ ⎜ 0 1 0 0 ⎟⎜ Q ⎟
⎟⎜ ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟ +
dτc ⎝ U ⎠ ⎝ 0 0 1 0 ⎠⎝ U ⎠
V 0 0 0 1 V
LINE FORMATION IN A MAGNETIC FIELD 389
⎛ ⎞⎛ ⎞
hI hQ hU hV I − SL
⎜h −rU ⎟ ⎜ ⎟
⎜ hI rV ⎟⎜ Q ⎟
+ κL ⎜ Q ⎟⎜ ⎟ , (9.35)
⎝ hU −rV hI rQ ⎠⎝ U ⎠
hV rU −rQ hI V
while introducing the optical depth τL we have
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 0 0 0 I − Sc
d ⎜⎜ ⎟
Q ⎟ ⎜0 1
⎜ 0 0⎟ ⎜
⎟⎜ Q ⎟
⎟
⎜ ⎟ = κc ⎜ ⎟⎜ ⎟
dτL ⎝ U ⎠ ⎝0 0 1 0 ⎠⎝ U ⎠
V 0 0 0 1 V
⎛ ⎞⎛ ⎞
hI hQ hU hV I − SL
⎜h −rU ⎟ ⎜ ⎟
⎜ hI rV ⎟⎜ Q ⎟
+⎜ Q ⎟⎜ ⎟ , (9.36)
⎝ hU −rV hI rQ ⎠⎝ U ⎠
hV rU −rQ hI V
where
1 k (ν ) ∆νD
κc = = c 0 . (9.37)
κL kL
Particularly important is the case of LTE, where both the source functions Sc and
SL reduce to the Planck function BP ; in this case Eq. (9.35) can be written in the
form
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 + kI kQ kU kV I − BP
d ⎜ ⎟ ⎜
⎜ Q ⎟ ⎜ kQ 1 + kI fV −fU ⎟ ⎜
⎟⎜ Q ⎟
⎟
⎜ ⎟=⎜ ⎟⎜ ⎟ , (9.38)
dτc ⎝ U ⎠ ⎝ kU −fV 1 + kI fQ ⎠⎝ U ⎠
V kV fU −fQ 1 + kI V
where
kI = κL hI
kQ = κL hQ fQ = κL rQ
kU = κL hU fU = κL rU
kV = κL hV fV = κL rV , (9.39)
with hI , hQ , hU , hV , rQ , rU , rV given by Eqs. (9.32).
9.5. Generalities and Symmetry Properties of the Transfer Equation
Let us consider the transfer equation in the form (9.35). For any wavelength λ,
Eq. (9.35) is a system of four linear, first-order, ordinary differential equations in
390 CHAPTER 9
the Stokes parameters I(τc ), Q(τc ), U (τc ), V (τc ) characterizing a radiation beam
flowing along a given direction Ω. The solution to the system requires a set of
boundary conditions, which depend of course on the specific physical situation.
Here we restrict attention to the case of a semi-infinite stellar atmosphere.
If the atmosphere is not illuminated from outside (obviously this is not the case
for a star in a binary system), the boundary conditions at the outermost layer are
simply
I(τc = 0) = Q(τc = 0) = U (τc = 0) = V (τc = 0) = 0 (9.40)
for any λ and any inward propagation direction ( Ω · k < 0, with k the outward-
directed unit vector perpendicular to the stellar surface). The boundary conditions
in the deep layers are more delicate and require a previous discussion of the source
functions Sc and SL . It is reasonable to assume that the atmospheric plasma at
large optical depths is in thermodynamic equilibrium, so that
lim Sc (τc ) = lim SL (τc ) = BP T (τc ) . (9.41)
τc →∞ τc →∞
The boundary conditions for the Stokes parameters are then
lim I(τc ) = BP T (τc )
τc →∞
lim Q(τc ) = lim U (τc ) = lim V (τc ) = 0 (9.42)
τc →∞ τc →∞ τc →∞
for any λ and any Ω. Equations (9.40)-(9.42) still hold if the optical depth τ is
c
replaced by any of the optical depths considered in Sect. 9.4.
Once the propagation direction and the boundary conditions are specified, the
Stokes parameters at any assigned optical depth depend on the coefficients of
Eq. (9.35), which, in general, are themselves functions of optical depth. Inspection
of Eqs. (9.35), (9.32), (9.29), (9.27) and (9.26) shows that the value of the Stokes
parameters for a specific spectral line at a given distance (λ − λ0 ) from line center
is determined by the following parameters:
– the continuum source function Sc , which is generally well-approximated by the
local Planck function BP ;
– the line source function SL , which reduces to the Planck function only when the
LTE approximation is valid;
– the ratio κL between line and continuum absorption coefficient;
– the magnetic field intensity B, entering the definition of vB ;
– the inclination angle θ of the magnetic field vector defined in Fig. 9.1;
– the azimuth angle χ of the magnetic field vector, also defined in Fig. 9.1;
– the Doppler width of the line ∆λD , entering the definitions of the reduced wave-
length v, the damping constant a, and the quantities vA and vB ;
– the natural plus collisional width Γ , defined in Eq. (9.22), entering the damping
constant a ;
– the line-of-sight component of the macroscopic velocity of the ambient medium
wA , entering the definition of vA .
LINE FORMATION IN A MAGNETIC FIELD 391
In addition, the Stokes parameters depend on the Zeeman pattern of the line and
on the wavelength λ0 which affects the parameters vB , vA and a.
In the special case where the parameter wA is independent of optical depth, its
effect on the Stokes parameters profiles is just a global wavelength shift of the
spectrum. It may then be convenient to neglect the term vA in Eqs. (9.27) and use
a different definition for the reduced wavelength,
λ − λ0
v ≡ v − vA = , (9.43)
∆λD
where , w -
λ0 = λ0 1 + A . (9.44)
c
This case will be referred to in the following as the constant velocity case.1 The
opposite case, where wA is variable with optical depth, will be referred to as the
velocity gradient case, although, more properly, one should speak of ‘the case where
the line-of-sight component of the bulk velocity of the ambient medium is variable
with optical depth’.
The wavelength dependence of the Stokes parameters emerging from a stellar
atmosphere reflects, broadly speaking, the wavelength dependence of the quantities
hI , hQ , hU , hV , rQ , rU , rV defined in Eqs. (9.32). As an illustrative example,
Fig. 9.3 shows this dependence for different Zeeman patterns and for assigned values
of the relevant parameters. The quantities hU and rU , not shown in the figure,
vanish identically since the azimuth angle χ has been set to zero.
The existence of wavelength symmetries about line center appears clearly from
Fig. 9.3, and a general proof of this fact can be easily obtained. From Eqs. (9.27)
we get, with the help of Eqs. (3.18), (5.59), and (9.26)
ηq (−∆λ) = η−q (∆λ) , ρq (−∆λ) = − ρ−q (∆λ) ,
where , w -
∆λ = λ − λ0 1 + A (9.45)
c
is the wavelength distance from (local) line center. Thus from Eqs. (9.29)
ηb (−∆λ) = ηr (∆λ) ηp (−∆λ) = ηp (∆λ)
ρb (−∆λ) = − ρr (∆λ) ρp (−∆λ) = − ρp (∆λ) , (9.46)
and substituting into Eqs. (9.32) we obtain
hI (−∆λ) = hI (∆λ)
hQ (−∆λ) = hQ (∆λ) rQ (−∆λ) = − rQ (∆λ)
hU (−∆λ) = hU (∆λ) rU (−∆λ) = − rU (∆λ)
hV (−∆λ) = −hV (∆λ) rV (−∆λ) = rV (∆λ) , (9.47)
1
Obviously, it includes the more restrictive case of a static atmosphere (wA = 0).
392 CHAPTER 9
Fig.9.3. The elements of the propagation matrix are plotted as functions of the reduced wavelength
for the following transitions (cf. Fig.3.2 on p. 83): (a) 1 S0 − 1 P1 (normal Zeeman triplet); (b)
5 D − 7 D ; (c) 5 P − 5 F ; (d) 5 D − 5 F ; (e) 5 P − 5 D ; (f) 5 D − 5 G . The relevant parameters
2 3 1 1 3 3 3 4 1 2
are: vA = 0, vB = 1.5, θ = 60◦ , χ = 0◦ , a = 0.05. Note that cases (c) and (f) show characteristic
reversals in hQ , hV , rQ , rV relative to the other cases. Transition (c) corresponds to ḡ > 0 and
Ḡ < 0 (see Sect. 9.6), while transition (f) corresponds to ḡ < 0 and Ḡ < 0.
which are just the symmetry properties shown by Fig. 9.3.
These properties are very important because they entail – under certain assump-
tions – definite symmetry characteristics of the solutions to the transfer equation.1
1
A different proof of the properties expressed by Eqs. (9.54) and (9.56), based on the analogy
between the transfer equation and the motion of the representative point in the Poincaré sphere,
has been presented by Landi Degl’Innocenti and Landi Degl’Innocenti (1981).
LINE FORMATION IN A MAGNETIC FIELD 393
Consider again Eq. (9.35), which we rewrite in the compact matrix form
dI
= I − Sc U + κ L H I − SL U , (9.48)
dτc
where
I = (I, Q, U, V )†
U = (1, 0, 0, 0)† (9.49)
and ⎛ ⎞
hI hQ hU hV
⎜h −rU ⎟
⎜ hI rV ⎟
H =⎜ Q ⎟. (9.50)
⎝ hU −rV hI rQ ⎠
hV rU −rQ hI
Equations (9.47) show that wavelength inversion about (local) line center corre-
sponds to a unitary similarity transformation of the matrix H
H(−∆λ) = X −1 H(∆λ) X , (9.51)
where X is the diagonal matrix
⎛ ⎞
1 0 0 0
⎜0 1 0 0 ⎟
X = X −1 =⎝ ⎠ . (9.52)
0 0 1 0
0 0 0 −1
Let us now consider the constant velocity case. The quantity ∆λ defined in
Eq. (9.45) is now independent of optical depth, thus it can be used in the place
of λ to describe the wavelength dependence of the Stokes profiles. The transfer
equations for I(∆λ) and I(−∆λ) are, from Eq. (9.48)
d
I(∆λ) = I(∆λ) − Sc U + κL H(∆λ) I(∆λ) − SL U (9.53a)
dτc
d
I(−∆λ) = I(−∆λ) − Sc U + κL H(−∆λ) I(−∆λ) − SL U . (9.53b)
dτc
In the latter equation we substitute Eq. (9.51) and multiply both sides by X from
the left. Since XU = U , we get
d
X I(−∆λ) = X I(−∆λ) − Sc U + κL H(∆λ) X I(−∆λ) − SL U .
dτc
Comparison with Eq. (9.53a) shows that the two vectors X I(−∆λ) and I(∆λ)
obey the same differential equation. If the boundary values I (b) are such that1
1
This is always the case for the radiation emerging from a semi-infinite stellar atmosphere
(see Eqs. (9.42)).
394 CHAPTER 9
X I (b) (−∆λ) = I (b) (∆λ), we can conclude that the two vectors coincide, or, in
other words
I(−∆λ) = I(∆λ)
Q(−∆λ) = Q(∆λ)
U (−∆λ) = U (∆λ)
V (−∆λ) = −V (∆λ) (9.54)
at any optical depth.
Another general property of the solutions to the transfer equation can be proved
under certain assumptions. It concerns the local inversion of the magnetic field
direction at each point along the ray path,
)→B
B(τ (τ ) = −B(τ
).
c c c
This transformation implies, at any optical depth (cf. Fig. 9.1)
θ = π − θ , χ = π + χ ,
hence from Eqs. (9.32)
hI = hI
hQ = hQ
rQ = rQ
hU = hU
rU = rU
hV = − hV
rV = − rV .
The relation between the propagation matrices H and H can be expressed in
terms of a unitary similarity transformation
H = Y −1 H Y . (9.55)
It can be shown with some matrix algebra that the most general matrix satisfying
Eq. (9.55) has the form
⎛ ⎞
1 0 0 0
⎜ 0 cos 4χ 0 ⎟
⎜ sin 4χ ⎟
Y = Y −1 = ⎜ ⎟ .
⎝ 0 sin 4χ − cos 4χ 0 ⎠
0 0 0 −1
If we now assume this matrix (that is, the χ angle) to be independent of optical
depth, we can follow the same line of reasoning which leads from Eqs. (9.53) to
Eqs. (9.54). We deduce that inversion of the magnetic field direction changes the
Stokes parameters into I = Y I, or, explicitly
I = I
Q = cos 4χ Q + sin 4χ U
U = sin 4χ Q − cos 4χ U
V = −V , (9.56)
LINE FORMATION IN A MAGNETIC FIELD 395
χ
θ
−Ω Ω
Fig.9.4. The angles θ and χ specifying the magnetic field direction are referred to the right-handed
The unit vector e is parallel to e .
orthogonal system (ea ,eb , Ω). a a
provided the boundary values I (b) , I (b) are themselves connected by these rela-
tions. In the preferred reference frame (Sect. 5.5) we simply have a sign switch of
the Stokes parameters U and V . It should be remarked that, while the magnetic
field strength and inclination (as well as the other parameters affecting the trans-
fer equation) are allowed to vary with optical depth, the azimuth angle must be
constant in order for Eqs. (9.56) to hold.
A further symmetry property of the solutions to the transfer equation concerns
the 180◦ rotation of the magnetic field vector about the propagation direction, at
each point along the ray path. This transformation, corresponding to
χ(τc ) → π + χ(τc ) , (9.57)
leaves the propagation matrix unchanged, so that the Stokes parameters are also
unchanged. This is an obvious consequence of the invariance of the Stokes param-
eters under a 180◦ rotation of the reference frame about the propagation direction.
Finally, we point out a symmetry property relating the transfer equations of two
radiation beams flowing in opposite directions. Let (ea , eb ) and (ea , eb ) denote the
unit vectors defining the Stokes parameters for the directions Ω and −Ω, respec-
tively. We choose the reference directions ea and ea to be the same. If the magnetic
field vector at a given point is specified in the reference system (ea , eb , Ω) by the
angles θ and χ (see Fig. 9.4), the corresponding angles in the system (ea , eb , −Ω)
are given by
θ = π − θ , χ = −χ . (9.58)
If, in addition, a macroscopic velocity field is present, having a component wA
its component along the opposite direction is w = −w .
along the direction Ω, A A
The relation between the propagation matrices H and H (referring to −Ω and Ω,
respectively) can be easily deduced. From Eqs. (9.27) we obtain, with the help of
Eqs. (5.59) and (3.18)
ηq (∆λ) = η−q (−∆λ) , ρq (∆λ) = − ρ−q (−∆λ) ,
where
∆λ = λ − λ0 (9.59)
396 CHAPTER 9
is the wavelength distance from the rest line center. Bearing in mind Eqs. (9.29)
and (9.58), we have from Eqs. (9.32)
hI (∆λ) = hI (−∆λ)
hQ (∆λ) = hQ (−∆λ)
rQ (∆λ) = −rQ (−∆λ)
hU (∆λ) = −hU (−∆λ)
rU (∆λ) = rU (−∆λ)
hV (∆λ) = hV (−∆λ) rV (∆λ) = −rV (−∆λ) .
It can be shown with some matrix algebra that this transformation can be written
as a unitary similarity transformation
H (∆λ) = Z −1 H(−∆λ) Z , (9.60)
where
⎛ ⎞
1 0 0 0
⎜0 1 0 0⎟
Z = Z −1 =⎝ ⎠ .
0 0 −1 0
0 0 0 1
Denoting by I and I the Stokes vectors of the radiation beams flowing in the
and Ω
directions −Ω respectively, we have from Eq. (9.48)
d
I (∆λ) = I (∆λ) − Sc U + κL H (∆λ) I (∆λ) − SL U ,
dτc
where τc is the continuum optical depth measured along −Ω.
Left-multiplication
by Z gives, with the use of Eq. (9.60)
d
Z I (∆λ) = Z I (∆λ) − Sc U + κL H(−∆λ) Z I (∆λ) − SL U ,
dτc
since ZU = U . Defining now the formal vector
Ĩ = Z I = (I , Q , −U , V )† ,
and noting that dτc = − dτc , we have
d & '
Ĩ (∆λ) = − Ĩ (∆λ) − Sc U + κL H(−∆λ) Ĩ (∆λ) − SL U . (9.61)
dτc
It follows that the transfer equations for Ĩ (∆λ) and I(−∆λ) are the same except
for a sign switch.
LINE FORMATION IN A MAGNETIC FIELD 397
9.6. The Weak Field Approximation
When the magnetic field is weak or, more precisely, when the Zeeman splitting is
much smaller than the typical width of the profiles ηq and ρq defined in Eqs. (9.27),
it is possible to deduce some properties of the solutions to the transfer equation
without actually solving it. The above condition means
∆λB
ḡ vB = ḡ 1,
∆λD
where ḡ is the effective Landé factor. This sets an upper bound to the magnetic
field strength, which depends (see Eq. (5.49)) both on the spectral line (through
its wavelength and the atomic weight of the element) and on the ambient medium
(through its kinetic temperature and microturbulent velocity). For instance, for
an iron line of optical wavelength formed in the solar atmosphere, we have from
Eq. (5.49)
ḡ B 2500 G .
The following derivation is based on an original, perturbative scheme presented
by Landi Degl’Innocenti and Landi Degl’Innocenti (1973) and later extended by
Jefferies et al. (1989).
Let us consider a radiation beam flowing along a given direction in a stellar
) the magnetic field vector at optical depth τ
atmosphere, and let us denote by B(τ c c
along the ray path. The transfer equation is given by Eq. (9.35). Now suppose this
reference (or ‘true’) atmosphere is replaced by a ‘fictitious’ atmosphere identical to
the former except for the substitution
) → α B(τ
B(τ ), (9.62)
c c
which entails
vB (τc ) → α vB (τc ) ,
where α is a dimensionless, real parameter which at the end of the calculation will
be set to 1. All the quantities in Eq. (9.35) – except the source functions and
the ratio κL between line and continuum absorption coefficient – become functions
of α, hence they can be expanded in power series of α. For the Stokes parameters
we can write1
I = I0 + I1 α + I2 α2 + I3 α3 + · · ·
Q = Q0 + Q1 α + Q2 α2 + Q3 α3 + · · · (9.63)
with similar equations for U and V . To obtain the series expansions of the elements
of the propagation matrix, we write the ηq and ρq profiles in the form
∞ ∞
∂ n ηq αn ∂ n ρq αn
ηq = , ρq = . (9.64)
n=0
∂αn α=0 n! n=0
∂αn α=0 n!
1
In this section we denote by I0 , I1 , I2 . . . the coefficients of the power series expansion
of the intensity I (first Stokes parameter). These should not be confused with the symbols
(I0 , I1 , I2 , I3 ) used in other sections to denote the full Stokes vector (I, Q, U, V ).
398 CHAPTER 9
The zero-order coefficients are given by (see Eqs. (9.27) and (3.17))
1
ηq α=0 = √ H(v − vA , a) ≡ η
π
1
ρq α=0 = √ L(v − vA , a) ≡ ρ , (9.65)
π
thus they are simply the absorption profile and the associated dispersion profile
for zero magnetic field. The higher-order coefficients can be expressed in terms of
derivatives of η and ρ with respect to wavelength. From Eqs. (9.27) and (9.26) we
have
∂ n ηq J J ∂nη
= Sq u (M , Mu ) (gu Mu − g M )n ∆λnB
∂αn α=0 ∂λn
M Mu
n
∂ ρq J J ∂nρ
= Sq u (M , Mu ) (gu Mu − g M )n ∆λnB , (9.66)
∂αn α=0 ∂λn
M Mu
where
∂nη 1 ∂n
= n
√ H(v − vA , a)
∂λn ∆λD π ∂v n
∂nρ 1 ∂n
n
= n
√ L(v − vA , a) . (9.67)
∂λ ∆λD π ∂v n
The sum over M and Mu appearing in Eqs. (9.66) has already been calculated in
Sect. 3.3. Using Eqs. (3.53), (3.20) and (3.54) we get
∂ n ηq ∂ nη
= (−1)n G(n) n
q (J , Ju ) ∆λB
∂αn α=0 ∂λn
n
∂ ρq ∂nρ
= (−1)n G(n) n
q (J , Ju ) ∆λB , (9.68)
∂αn α=0 ∂λn
(n)
where the symbols Gq , defined in Eq. (3.55), are expressed analytically (up to
n = 4) in Table 3.4. Substituting into Eqs. (9.64) and bearing in mind Eqs. (9.29),
we obtain from Eqs. (9.32) the following expansions up to third-order terms in α
1 2
hI = η + α ∆λ2B η G(2)
0 sin θ + G1 (1 + cos θ) + · · ·
2 (2) 2
4
1 2
hQ = α ∆λ2B η G(2)
0 − G1
(2)
sin2 θ cos 2χ + · · ·
4
1
hU = α2 ∆λ2B η G(2)0 − G1
(2)
sin2 θ sin 2χ + · · ·
4
1 3
hV = − α ∆λB η G(1)
1 cos θ − α ∆λ3B η G(3)
1 cos θ + · · ·
6
LINE FORMATION IN A MAGNETIC FIELD 399
1 2
rQ = α ∆λ2B ρ G(2)
0 − G1
(2)
sin2 θ cos 2χ + · · ·
4
1
rU = α2 ∆λ2B ρ G(2)0 − G1
(2)
sin2 θ sin 2χ + · · ·
4
1 3
rV = − α ∆λB ρ G(1)
1 cos θ − α ∆λ3B ρ G(3)1 cos θ + · · · , (9.69)
6
where
∂η ∂ρ ∂2η
η = , ρ = , η = , etc. (9.70)
∂λ ∂λ ∂λ2
Substitution of the series expansions (9.69) and (9.63) into the transfer equation
(9.35) leads to four equations each representing an equality between two power
series in α. Since α is arbitrary, the coefficients of each power must separately be
equal. To the lowest perturbative order we have
dI0 dU0
= I0 − Sc + κL η (I0 − SL ) = (1 + κL η) U0
dτc dτc
dQ0 dV0
= (1 + κL η) Q0 = (1 + κL η) V0 . (9.71)
dτc dτc
Boundary conditions should now be considered. For a semi-infinite atmosphere
– the case to which we restrict attention from now on – the equations resulting
from the substitution of Eqs. (9.63) into Eqs. (9.42) must hold for any value of α.
Therefore, the boundary conditions at the bottom of the atmosphere are1
lim I (τ ) = BP T (τc )
τc →∞ 0 c
lim I (τ ) = lim I2 (τc ) = . . . = lim Q0 (τc ) = lim Q1 (τc ) = . . .
τc →∞ 1 c τc →∞ τc →∞ τc →∞
= lim U0 (τc ) = lim U1 (τc ) = . . . = lim V0 (τc ) = lim V1 (τc ) =
τc →∞ τc →∞ τc →∞ τc →∞
... = 0 . (9.72)
It follows from Eqs. (9.71)-(9.72) that the zero-order Stokes parameters Q0 , U0 ,
and V0 , obeying a linear homogeneous differential equation (without source terms)
and being identically zero at the boundary of the atmosphere, are identically zero
everywhere. As far as I0 is concerned, we observe that it obeys the same differ-
ential equation as the intensity I¯ of a radiation beam travelling through the same
atmosphere with no magnetic field.2 We conclude that, to the lowest perturbative
order, the polarization is zero and the intensity equals, at any optical depth and for
1
For the radiation flowing inward, one has to consider the boundary conditions at the top of
the atmosphere. For an isolated star, all terms in Eqs. (9.63) are identically zero at τc = 0.
2
By ‘same atmosphere’ we mean that the line source function SL is also the same. This
remark is not trivial for non-LTE lines, since the presence of a magnetic field may alter the line
source function.
400 CHAPTER 9
any direction, the intensity that would be present in the same atmosphere without
magnetic field. This was to be expected since, according to Eqs. (9.63) and (9.62),
the zero-order terms correspond to a non-magnetic atmosphere.
Proceeding along the same lines, we obtain a system of differential equations for
each perturbative order. The results up to the third order are the following:
– Zero order
I = I¯
0
Q0 = U0 = V0 = 0 .
– First order
dV1
= (1 + κL η) V1 − ∆λB G(1)
1 cos θ κL η (I0 − SL )
dτc
I1 = Q1 = U1 = 0 .
– Second order
dI2 1
0 sin θ + G1 (1 + cos θ) κL η (I0 − SL )
= (1 + κL η) I2 + ∆λ2B G(2) 2 (2) 2
dτc 4
− ∆λB G(1)
1 cos θ κL η V1
dQ2 1
0 − G1
= (1 + κL η) Q2 + ∆λ2B G(2) (2)
sin2 θ cos 2χ κL η (I0 − SL )
dτc 4
dU2 1
0 − G1
= (1 + κL η) U2 + ∆λ2B G(2) (2)
sin2 θ sin 2χ κL η (I0 − SL )
dτc 4
V2 = 0 .
– Third order
I3 = 0
dQ3
= (1 + κL η) Q3 − ∆λB G(1)
1 cos θ κL ρ U2
dτc
1
− ∆λ2B G(2) 0 − G1
(2)
sin2 θ sin 2χ κL ρ V1
4
dU3
= (1 + κL η) U3 + ∆λB G(1)
1 cos θ κL ρ Q2
dτc
1
+ ∆λ2B G(2) 0 − G1
(2)
sin2 θ cos 2χ κL ρ V1
4
dV3 1
= (1 + κL η) V3 − ∆λ3B G(3)
1 cos θ κL η (I0 − SL )
dτc 6
− ∆λB G(1)
1 cos θ κL η I2
1
+ ∆λ2B G(2) 2 (2) 2
0 sin θ + G1 (1 + cos θ) κL η V1 .
4
LINE FORMATION IN A MAGNETIC FIELD 401
These equations suggest two general remarks:
a) Each of the coefficients of the power series expansions (9.63), if non-zero, obeys
a differential equation of the form
dCn
= (1 + κL η) Cn + (source term) ,
dτc
n
where the source term is proportional (see Eqs. (9.67)) to vB = (∆λB /∆λD )n .
Thus, I0 is of order 0 in vB ; V1 is proportional to vB ; I2 , Q2 and U2 are propor-
2
tional to vB , and so on.
b) Anomalous dispersion effects enter the perturbative equations only from the
third order on. This means that for weak magnetic field they are negligible.
The perturbative expansion considered in this section is obviously meaningful
only in the limiting case of weak magnetic field (vB → 0). If, for each Stokes
parameter, we retain only the lowest-order non-zero term in Eqs. (9.63), we have,
setting α = 1
I = I0 , Q = Q2 , U = U2 , V = V1 . (9.73)
Thus the transfer equation for the Stokes parameters reduces, under the limit of
weak magnetic field, to the following system
dI
= (1 + κL η) I − (Sc + κL η SL ) (9.74a)
dτc
dV
= (1 + κL η) V − ∆λB ḡ cos θ κL η (I − SL ) (9.74b)
dτc
dQ 1
= (1 + κL η) Q − ∆λ2B Ḡ sin2 θ cos 2χ κL η (I − SL ) (9.74c)
dτc 4
dU 1
= (1 + κL η) U − ∆λ2B Ḡ sin2 θ sin 2χ κL η (I − SL ) , (9.74d)
dτc 4
where we have taken into account (see Table 3.4) that
G(1)
1 = ḡ , (9.75)
with ḡ the effective Landé factor of the line defined in Eq. (3.44), and where
(2)
1 − G0 .
Ḡ = G(2) (9.76)
Note that the quantity Ḡ can be written in the form
Ḡ = ḡ 2 − δ , (9.77)
where, using the notations of Table 3.4
1 2
δ = g (16s − 7d2 − 4) . (9.78)
80 d
402 CHAPTER 9
A number of interesting results can be derived from Eqs. (9.74):
i) Since Sc , SL and κL are practically wavelength-independent, derivation of Eq.
(9.74a) with respect to wavelength yields
d ∂I ∂I
= (1 + κL η) + κL η (I − SL ) , (9.79)
dτc ∂λ ∂λ
where the notation of Eq. (9.70) has been used. Multiplying both sides by the
factor −∆λB ḡ cos θ, and assuming this factor (i.e. the line-of-sight component of
the magnetic field) to be independent of optical depth, we see that the resulting
equation is formally identical to Eq. (9.74b), which means that the functions
∂I
V and − ∆λB ḡ cos θ
∂λ
obey the same differential equation. On the other hand, since from Eqs. (9.73)
and (9.72)
∂I ∂
lim V (τc ) = 0 , lim = B T (τc ) = 0 ,
τc →∞ τc →∞ ∂λ ∂λ P
they also satisfy the same boundary conditions. We conclude that, at any optical
depth
∂I
V (λ) = − ∆λB ḡ cos θ . (9.80)
∂λ
ii) Next consider Eqs. (9.74c) and (9.74d). Multiplying the former by sin 2χ , the
latter by cos 2χ , subtracting the resulting equations and assuming χ to be inde-
pendent of optical depth, we get
d
sin 2χ Q − cos 2χ U = (1 + κL η) sin 2χ Q − cos 2χ U ,
dτc
and since from Eqs. (9.73) and (9.72)
lim Q(τc ) = lim U (τc ) = 0 ,
τc →∞ τc →∞
we deduce that, at any optical depth
sin 2χ Q − cos 2χ U = 0 ,
or, for Q = 0
U (λ)
= tan 2χ . (9.81)
Q(λ)
LINE FORMATION IN A MAGNETIC FIELD 403
iii) If we retain the assumption χ = const., we can choose to measure the Stokes
parameters in the preferred frame (so that χ = 0; see Sect. 5.5). Denoting by Q̃
and Ũ the Stokes parameters in this frame, we have from Eqs. (9.74c) and (9.81)
dQ̃ 1
= (1 + κL η) Q̃ − ∆λ2B Ḡ sin2 θ κL η (I − SL )
dτc 4
Ũ = 0 . (9.82)
On the other hand, derivation of Eq. (9.79) with respect to wavelength yields
2
d ∂ I ∂2I ∂I
= 1 + κ L η + κL η (I − SL ) + 2 κL η .
dτc ∂λ2 ∂λ2 ∂λ
In the constant velocity case, the last term vanishes at the wavelength λ0 cor-
responding to line center (see Eqs. (9.44) and (9.67)). Multiplying the resulting
equation by the factor −∆λ2B Ḡ sin2 θ/4 and assuming this factor (hence the trans-
verse component of the magnetic field) to be independent of optical depth, we
obtain by comparison with the first of Eqs. (9.82) that the two functions
1 ∂2I
Q̃(λ0 ) and − ∆λ2B Ḡ sin2 θ
4 ∂λ2 λ=λ0
satisfy the same differential equation. As the boundary conditions are also the
same, we deduce that, at any optical depth
1 ∂2I
Q̃(λ0 ) = − ∆λ2B Ḡ sin2 θ . (9.83)
4 ∂λ2 λ=λ0
iv) Another expression for Q̃, valid for any λ , can be obtained under a set of more
restrictive assumptions. Consider Eq. (9.79) and multiply both sides by the factor
1 η
− ∆λ2B Ḡ sin2 θ .
4 η
If this factor is independent of optical depth, which requires the τc -independence
of:
a) the transverse component of the magnetic field;
b) the line-of-sight velocity wA ;
c) the parameters ∆λD and Γ controlling the shape of the η profile,
then we see by comparison with the first of Eqs. (9.82) that, at any optical depth1
1 η ∂I
Q̃(λ) = − ∆λ2B Ḡ sin2 θ . (9.84)
4 η ∂λ
1
Note that Eq. (9.84) is meaningless at line center, since Eqs. (9.67) and (9.54) yield
∂η ∂I
= = 0.
∂λ λ=λ ∂λ λ=λ
0 0
The value of Q̃(λ0 ) is however given by the (more general) equation (9.83).
404 CHAPTER 9
TABLE 9.1
Expressions for the Stokes parameters in the weak field limit. The expressions are valid provided
the physical parameters marked by a star are independent of optical depth. B and B⊥ stand
for B cos θ and B sin θ, respectively.
Expression Eq. B B⊥ χ wA ∆λD Γ Domain
∂I
V (λ) = −∆λB ḡ cos θ ∂λ
(9.80) any λ
U (λ)
Q(λ)
= tan 2χ (9.81) any λ
∂2I
Q̃(λ0 ) = − 14 ∆λ2B Ḡ sin2 θ ∂λ2 λ
(9.83) line center
0
η
∂I
Q̃(λ) = − 14 ∆λ2B Ḡ sin2 θ η ∂λ
(9.84) any λ
3 1
∂I
Q̃(λw ) = 4
∆λ2B Ḡ sin2 θ λw −λ0 ∂λ λ
(9.85) line wings
w
v) If we restrict ourselves to the line wings, assumption c) above is unnecessary in
order for Eq. (9.84) to hold. In fact, denoting by λw a particular wavelength far
away in the wing, or, more precisely, such that
3 3 3 3
3 λ − λ 3 ∆λ and 3 λ − λ 3 a ∆λ ,
w 0 D w 0 D
the profile η can be written in the form (see Eqs. (9.65), (9.43), and (5.57))
1 a ∆λ2D
η= ,
π (λw − λ0 )2
so that
η 3
=− .
η λw − λ0
This ratio is independent of ∆λD and Γ , therefore assumption c) can be dropped
and Eq. (9.84) becomes
3 1 ∂I
Q̃(λw ) = ∆λ2B Ḡ sin2 θ . (9.85)
4 λw − λ0 ∂λ λ=λw
The results just derived are summarized in Table 9.1, together with the assump-
tions required – in addition to the weak field hypothesis – for the various expressions
to hold. In particular, it should be emphasized that Eq. (9.80) is rather general
and that it can also be applied in the ‘velocity gradient case’ (see Sect. 9.5).
Some general features of the Stokes parameters profiles arise from the analytical
expressions in Table 9.1. First we consider Eq. (9.80), and we refer to a typical
stellar atmosphere where both line and continuum source functions are monotoni-
cally increasing functions of optical depth. Such atmosphere produces absorption
LINE FORMATION IN A MAGNETIC FIELD 405
lines and, for any line, the quantity ∂I/∂λ is negative in the blue wing and positive
in the red wing. For any propagation direction making an angle θ < π/2 with the
magnetic field direction, it follows that the V Stokes parameter is positive in the
blue wing and negative in the red wing, while the opposite holds for θ > π/2.1 It
also follows that there is a particular wavelength (more precisely, an odd number
of wavelengths) where the V profile vanishes.
Another consequence of Eq. (9.80) concerns the dependence of circular polar-
ization on the characteristics of spectral lines. Let us suppose that the intensity
profile emerging from a static stellar atmosphere can be written in the form2
η(λ)
I(λ) = Ic 1 − dc , (9.86)
η(λ0 )
where Ic is the intensity of the continuum adjacent to the line and where
Ic − I(λ0 )
dc =
Ic
is the line central depression in units of Ic . Assuming a depth-independent Doppler
width, from Eqs. (9.86), (9.67) and (9.80) we have, as an order of magnitude
V ∆λB
≈ cos θ ḡ dc . (9.87)
Ic ∆λD
This formula shows that V /Ic is inversely proportional to ∆λD , which means that
circular polarization in narrow lines is larger than in broad lines. Furthermore, it
is larger in lines of heavy elements compared with light elements, because ∆λD
decreases with increasing atomic weight (see Eqs. (9.25) and (5.48)). For a given
element, the largest circular polarization is produced by lines having the largest
effective Landé factor ḡ , the largest central depression dc , and the largest wave-
length λ0 . This last effect stems from the fact that ∆λB scales as λ20 (Eqs. (9.26))
while ∆λD scales as λ0 (Eq. (9.25)).
The above properties suggest the introduction of a dimensionless parameter, the
circular polarization sensitivity index of a spectral line, having the form
λ0
sV = ḡ dc , (9.88)
λref
where λref is a reference wavelength, like for instance 5000 Å. A search for the
most sensitive lines in the FeI solar spectrum (restricted to the list of about 400
1
We consider here the most frequent case of ḡ > 0. For the very few lines having ḡ < 0 the
situation is reversed.
2
This is generally an acceptable approximation for unsaturated absorption lines. The following
remarks also apply, as a rule, to the more general case where η(λ) in Eq. (9.86) is replaced by a
bell-shaped profile having a typical width ∆λp . For saturated lines one usually has ∆λp > ∆λD .
406 CHAPTER 9
TABLE 9.2
The lines of the FeI solar spectrum are listed in order of decreasing circular polarization sensitivity
index (computed from Eq.(9.88) with λref = 5000 Å). The value of ḡ is calculated according to
L-S coupling. Zeeman triplets are identified by a star in the last column.
λ0 (Å) Multiplet Transition ḡ sV
5250.21 1 5D
0 − 7 D1 3.000 2.249
6302.50 816 5P
1 − 5 D0 2.500 2.058
6173.34 62 5P
1 − 5 D0 2.500 1.923
5506.78 15 5F
2 − 5D
3 2.000 1.808
6336.83 816 5P
1 − 5 D1 2.000 1.769
5225.53 1 5D
1 − 7 D1 2.250 1.764
4690.14 820 5P
1 − 7 D1 2.750 1.708
5501.47 15 5F
3 − 5 D4 1.875 1.669
6213.43 62 5P
1 − 5 D1 2.000 1.665
4938.82 318 7F
2 − 7 D3 2.000 1.661
unblended lines in the interval 4365-6859 Å given by Stenflo and Lindegren, 1977)
yields the list in Table 9.2.
Let us now turn to the discussion of linear polarization, described by Eqs. (9.83),
(9.84), and (9.85). For an absorption line the quantity [∂ 2 I/∂λ2 ]λ=λ is positive,
0
so that, according to Eq. (9.83), Q̃(λ0 ) is negative when Ḡ is positive and vice
versa. Equation (9.85) shows that the sign of Q̃ in the wings is opposite to the sign
at line center. It follows that the Q̃ profile vanishes at two particular wavelengths
(more precisely, at an even number of wavelengths). These wavelengths can be
easily found only when Eq. (9.84) holds (which implies however more restrictive
assumptions – see Table 9.1). In that case they are the solutions to the equation
η = 0, which, using Eqs. (9.65) and (5.58), can be written in the form1
2a
√ − (1 + 2a2 − 2v 2 ) H(v, a) − 4av L(v, a) = 0 .
π
√
For zero damping (a = 0) this equation has the two solutions v0 = ± 1/ 2. For
a = 0 one can find an analytical, approximate solution if a 1. The calculation,
that is left as an exercise to the reader, leads to the following result
1
v0 = ± √ + α a ,
2
where, denoting by D(x) the Dawson function defined in Eq. (5.56),
.
e √ 1
α = 2 2 D √ − 1 0.2957 .
2π 2
1
We assume here, to avoid formal complications, that the ambient medium velocity wA is
zero. In any case, wA must be constant for Eq. (9.84) to hold.
LINE FORMATION IN A MAGNETIC FIELD 407
Another consequence of Eq. (9.84) concerns the special case where the intensity
profile I(λ) has the form of Eq. (9.86). Since in that case
η ∂ 2 I ; ∂I
= ,
η ∂λ2 ∂λ
Eq. (9.84) becomes
1 ∂2I
Q̃(λ) = − ∆λ2B Ḡ sin2 θ ,
4 ∂λ2
which ‘extends’ the validity of expression (9.83) to all wavelengths (of course this
holds only for weak, unsaturated lines). Use of Eq. (9.67) yields, as an order of
magnitude (cf. Eq. (9.87))
Q̃ ∆λ2B
≈ sin2 θ Ḡ dc ,
Ic ∆λ2D
which suggests the introduction of the (dimensionless) linear polarization sensitivity
index of a spectral line
2
λ0
sQ = Ḡ dc . (9.89)
λref
The parameter Ḡ clearly plays for linear polarization the same role as the effective
Landé factor ḡ does for circular polarization. As Ḡ is connected with the second-
order moments of Zeeman components, it can quite naturally be called the second
order effective Landé factor . Its main properties can be deduced from Eqs. (9.77)
and (9.78). Since for electric-dipole transitions the quantity δ defined in Eq. (9.78)
is non-negative for any Zeeman pattern and zero for triplets (normal or anomalous),
one can draw the conclusion that Ḡ = ḡ 2 for triplets, while Ḡ < ḡ 2 for any other
Zeeman pattern.
Although most lines have Ḡ > 0, there are several with Ḡ < 0. Analysis of the
list of FeI lines quoted above shows1 that about 8% have Ḡ ≤ 0, and that the
average value is Ḡ = 1.60. A search in the same list for which lines produce the
largest linear polarization yields the results shown in Table 9.3. Note that the first
three spectral lines in Tables 9.3 and 9.2 are the same.
For certain applications, it may be useful to know which transitions produce
– because of their intrinsic structure – small polarization, or no polarization at all.
Table 9.4 shows a list of transitions with ḡ = 0 or Ḡ = 0 (the former are the same
as in Table 3.2). The list collects all the transitions between L-S levels having
S ≤ 7/2 and L ≤ 5, and satisfying the conditions
∆J = 0, ±1 , ∆S = 0, ±1 , ∆L = 0, ±1, ±2 .
It appears from Table 9.4 that both ḡ and Ḡ are zero in some cases. It can be easily
shown that this property, combined with the selection rule ∆J = 0, ±1, implies
1
The analysis has been performed using L-S Landé factors; the average Ḡ was obtained
by weighting the Ḡ factors of individual lines by their central depression.
408 CHAPTER 9
TABLE 9.3
The lines of the FeI solar spectrum are listed in order of decreasing linear polarization sensitivity
index (computed from Eq.(9.89) with λref = 5000 Å ; upper panel). The lower panel lists the
most sensitive lines with Ḡ < 0. Ḡ is calculated in L-S coupling. A star in the last column
identifies Zeeman triplets.
λ0 (Å) Multiplet Transition Ḡ sQ
5250.21 1 5D
0 − 7D
1 9.000 7.085
6302.50 816 5P
1 − 5 D0 6.250 6.485
6173.34 62 5P
1 − 5 D0 6.250 5.936
4690.14 820 5P
1 − 7 D1 7.375 4.296
6232.64 816 5P
2 − 5 D1 3.983 3.974
6842.69 1197 5P
1 − 5 P1 6.250 3.921
4704.95 821 5P
1 − 5D
0 6.250 3.885
5506.78 15 5F
2 − 5 D3 3.900 3.884
6136.99 62 5P
2 − 5 D1 3.983 3.673
6336.83 816 5P
1 − 5 D1 3.250 3.644
5197.93 1091 5F − 5P −3.125 −1.351
1 1
4598.12 554 5D − 5F −1.125 −0.740
1 1
5705.46 1087 5F − 5D −1.125 −0.583
1 1
5432.95 1143 5G − 5F −0.600 −0.453
2 2
5916.25 170 3H − 3F −0.559 −0.422
4 4
TABLE 9.4
A star identifies the transitions having ḡ = 0 or Ḡ = 0. The Landé factors are calculated in L-S
coupling. The list includes all the transitions obeying the constraints specified in the text.
Transition ḡ = 0 Ḡ = 0 Transition ḡ = 0 Ḡ = 0 Transition ḡ = 0 Ḡ = 0
2S
1 − 4D 1 5D
2 − 3 G3 7F
2 − 7 H3
2 2
2P
1 − 4D
1 6D
1 − 6G
3 7F
3 − 5H
4
2 2 2 2
3P
0 − 5F
1 7D
1 − 5F
2 8F
1 − 6G
3
2 2
4P
1 − 4D
1 8D
5 − 6G
7 8F
3 − 6G
5
2 2 2 2 2 2
6P
3 − 4F
5 4F
7 − 6H
7 8F
5 − 6G
7
2 2 2 2 2 2
4D
1 − 4D
1 5F
1 − 5F
1 6G
3 − 6G
3
2 2 2 2
4D
1 − 6D
1 5F
1 − 7F
0 6G
3 − 8G
1
2 2 2 2
4D
1 − 6F
1 5F
1 − 7H
2 7H
2 − 7H
2
2 2
4D
1 − 6G
3 5F
2 − 5H
3
2 2
5D
0 − 5F
1 6F
1 − 6G
3
2 2
LINE FORMATION IN A MAGNETIC FIELD 409
that the Landé factors of both the upper and lower level are zero. Therefore, these
transitions produce no polarization whatever the magnetic field strength.
As for the transitions with ḡ = 0 and Ḡ = 0, it can be proved that the Landé
factors of both levels are non-zero. According to Eq. (9.80), such transitions pro-
duce no circular polarization under the weak field limit. They can produce circular
polarization if the magnetic field is not weak.
According to Eq. (9.84), the transitions with Ḡ = 0 and ḡ = 0 produce no linear
polarization in the weak field limit. They can in principle give rise to linear polar-
ization when the magnetic field is strong. However, as shown by Sánchez Almeida
and Vela Villahoz (1993), all the transitions in Table 9.4 involving the terms 4 D1/2
or 6 G3/2 are characterized by peculiar Zeeman patterns where a compensation oc-
curs between σ and π components, which eventually causes the elements hQ , hU ,
rQ , rU of the propagation matrix to be identically zero. It follows (cf. Eq. (9.35))
that the transfer equations for the Stokes parameters Q and U are decoupled from
those for I and V ; if the boundary values for Q and U are also zero (Eqs. (9.42)), the
linear polarization for such transitions is zero whatever the magnetic field strength.
As a concluding remark, we emphasize that most of the results contained in
this section, and especially those related to the shape of the Stokes profiles and
to the dependence of polarization on the characteristics of spectral lines, remain
qualitatively valid even when the weak field assumption is dropped, that is when
∆λB ≈ ∆λD .
9.7. Formal Solution through the Evolution Operator
Consider again the transfer equation in the form (9.35). A formal solution to this
equation is easily obtained by adjusting the formalism of the evolution operator
developed in Sect. 8.2. For this purpose it is convenient to cast Eq. (9.35) in the
matrix form1
d
I(τc ) = C(τc ) I(τc ) − j(τc ) , (9.90)
dτc
where
⎛
⎞
I
⎜Q⎟
I=⎝ ⎠
U
V
⎛ ⎞
1 + kI kQ kU kV
⎜ k −fU ⎟
⎜ 1 + kI fV ⎟
C = 1 + κL H = ⎜ Q ⎟ (9.91)
⎝ kU −fV 1 + kI fQ ⎠
kV fU −fQ 1 + kI
1
The notation used here is rather uncommon: the matrix C appearing in this equation is
often denoted in the literature with the symbol K. We use this notation to avoid confusion with
the matrix K introduced in Chap. 8 (see Eq. (8.2)).
410 CHAPTER 9
⎞ ⎛
Sc + kI SL
⎜ k S ⎟
⎜ Q L ⎟
j = Sc 1 + κ L SL H U = ⎜ ⎟ , (9.92)
⎝ kU SL ⎠
kV SL
with U and H given by Eqs. (9.49) and (9.50), and the coefficients kI , kQ , kU ,
kV , fQ , fU , fV by Eqs. (9.39).
Equation (9.90) is formally identical to Eq. (8.2), except for a sign switch due to
the use of the optical depth (which increases in the direction opposite to propaga-
tion) as independent variable. Thus we can define an evolution operator O(τc , τc )
via the equation (cf. Eq. (8.6))
I(τc ) = O(τc , τc ) I(τc ) (τc ≤ τc ) , (9.93)
which has the properties (cf. Eqs. (8.7)-(8.10))
O(τc , τc ) = 1 (9.94)
O(τc , τc ) = O(τc , τc ) O(τc , τc ) (τc ≤ τc ≤ τc ) (9.95)
d
O(τc , τc ) = C(τc ) O(τc , τc ) (9.96)
dτc
d
O(τc , τc ) = − O(τc , τc ) C(τc ) , (9.97)
dτc
and which can be written in the form (cf. Eq. (8.12))
∞ τc τc τc & '
(−1)n
O(τc , τc ) = 1 + dτ1 dτ2 · · · dτn P C(τ1 ) C(τ2 ) · · · C(τn ) ,
n=1
n!
τc τc τc
where P is the Dyson chronological product
& '
P C(τ1 ) C(τ2 ) · · · C(τn ) = C(τi ) C(τj ) · · · C(τk ) , (9.98)
(i, j, . . . , k) being a permutation of the integers (1, 2, . . . , n) such that
τi ≤ τj ≤ . . . ≤ τk . (9.99)
The formal solution to Eq. (9.90) is given by (cf. Eq. (8.14))
τc
I(τc ) = O(τc , τc ) j(τc ) dτc + O(τc , τc ) I(τc ) , (9.100)
τc
LINE FORMATION IN A MAGNETIC FIELD 411
which for a semi-infinite atmosphere reduces to
∞
I(τc ) = O(τc , τc ) j(τc ) dτc (9.101)
τc
provided
lim O(τc , τc ) j(τc ) = 0 . (9.102)
τc →∞
It can be easily proved, by the same arguments used in Sect. 8.3, that the evolution
operator corresponding to a C matrix independent of optical depth is
−(τc − τc ) C
O(τc , τc ) = e , (9.103)
while for a matrix of the form C = c1 (τc ) 1 + c2 (τc ) H, with H independent of
optical depth, one has
: τc : τc
− c1 (τc ) dτc − c2 (τc ) dτc H
O(τc , τc ) = e τc e τc . (9.104)
9.8. The Unno-Rachkovsky Solution
The formal solution derived in the previous section allows us to express the Stokes
parameters of the radiation propagating along any direction at any specified point
of a magnetized stellar atmosphere in terms of the physical characteristics of the
atmosphere itself. To specify such characteristics means to give a model of the at-
mosphere or, in other words, a set of mathematical functions which fix, at any point
in the atmosphere, the value of all the physical quantities affecting the propagation
matrix and the emission vector.
Probably the simplest atmospheric model1 is the so-called Milne-Eddington model
which results from the following set of assumptions:
a) the atmosphere is plane-parallel, semi-infinite, and in LTE (Sc = SL = BP );
b) all the quantities affecting the propagation matrix C, namely κL , B, θ, χ ,
∆λD , Γ , wA are depth-independent, so that, for any propagation direction, C is
constant;
c) the Planck function is linear in tc , the continuum optical depth measured along
the vertical
BP = B0 + B1 tc = B0 (1 + β tc ) . (9.105)
Under these assumptions we can easily find, using the formulae of the previous
section, an analytical solution to the transfer equation.
1
The even simpler reversing-layer model that leads to the Seares formulae (see Sect. 9.13)
should be considered, more properly, a slab model.
412 CHAPTER 9
Owing to assumptions a) and c), Eq. (9.92) reduces to (see also Eq. (9.91))
j(tc ) = B0 (1 + β tc ) C U ,
and using assumption b) we obtain from Eqs. (9.103), (9.101) and (9.100)1
∞ tc − tc
− C dtc
I(tc , µ) = B0 e µ
(1 + β tc ) C U (µ > 0)
µ
tc
tc tc − tc
− C dtc
I(tc , µ) = B0 e |µ| (1 + β tc ) C U (µ < 0) ,
|µ|
0
with µ = cos α and α defined in Fig. 9.2. Both integrals can be performed by
means of the substitution (tc − tc )/µ = x. Bearing in mind the definition of the
exponential of a matrix (Eq. (8.21)), it can be easily proved that
b
−x C −a C −b C
e dx = e − e C −1 (9.106)
a
b
−x C −a C −b C
e x dx = a 1 + C −1 e − b 1 + C −1 e C −1 , (9.107)
a
where C −1 is the inverse of the matrix C,
C C −1 = C −1 C = 1 .
Use of Eqs. (9.106) and (9.107) leads to the expressions
I(tc , µ) = B0 (1 + β tc ) 1 + βµ C −1 U (µ > 0)
I(tc , µ) = B0 (1 + β tc ) 1 − β |µ| C −1
− tc C
− 1 − β |µ| C −1 e |µ| U (µ < 0) . (9.108)
In particular, the radiation emerging from the atmosphere (tc = 0 , µ > 0) is given
by
I(0, µ) = B0 1 + βµ C −1 U . (9.109)
1
It is understood that the boundary condition at tc = 0 is I (µ < 0) = 0; using Eqs. (A5.23)
and (9.111) it can be shown that condition (9.102) is also satisfied.
LINE FORMATION IN A MAGNETIC FIELD 413
TABLE 9.5
Analytical expressions for the elements of the matrix C−1
(C −1 )00 = ∆−1 (1 + kI ) (1 + kI )2 + fQ
2 + f2 + f2
U V
(C −1 )01 = −∆−1 (1 + kI )2 kQ + (1 + kI )(kU fV − kV fU ) + fQ (kQ fQ + kU fU + kV fV )
(C −1 )02 = −∆−1 (1 + kI )2 kU + (1 + kI )(kV fQ − kQ fV ) + fU (kQ fQ + kU fU + kV fV )
(C −1 )03 = −∆−1 (1 + kI )2 kV + (1 + kI )(kQ fU − kU fQ ) + fV (kQ fQ + kU fU + kV fV )
(C −1 )10 = −∆−1 (1 + kI )2 kQ − (1 + kI )(kU fV − kV fU ) + fQ (kQ fQ + kU fU + kV fV )
(C −1 )11 = ∆−1 (1 + kI ) (1 + kI )2 + fQ
2 − k2 − k2
U V
(C −1 )12 = −∆−1 (1 + kI )2 fV − (1 + kI )(kQ kU + fQ fU ) − kV (kQ fQ + kU fU + kV fV )
(C −1 )13 = ∆−1 (1 + kI )2 fU + (1 + kI )(kV kQ + fV fQ ) − kU (kQ fQ + kU fU + kV fV )
(C −1 )20 = −∆−1 (1 + kI )2 kU − (1 + kI )(kV fQ − kQ fV ) + fU (kQ fQ + kU fU + kV fV )
(C −1 )21 = ∆−1 (1 + kI )2 fV + (1 + kI )(kQ kU + fQ fU ) − kV (kQ fQ + kU fU + kV fV )
(C −1 )22 = ∆−1 (1 + kI ) (1 + kI )2 + fU
2 − k2 − k2
V Q
(C −1 )23 = −∆−1 (1 + kI )2 fQ − (1 + kI )(kU kV + fU fV ) − kQ (kQ fQ + kU fU + kV fV )
(C −1 )30 = −∆−1 (1 + kI )2 kV − (1 + kI )(kQ fU − kU fQ ) + fV (kQ fQ + kU fU + kV fV )
(C −1 )31 = −∆−1 (1 + kI )2 fU − (1 + kI )(kV kQ + fV fQ ) − kU (kQ fQ + kU fU + kV fV )
(C −1 )32 = ∆−1 (1 + kI )2 fQ + (1 + kI )(kU kV + fU fV ) − kQ (kQ fQ + kU fU + kV fV )
(C −1 )33 = ∆−1 (1 + kI ) (1 + kI )2 + fV2 − kQ
2 − k2
U
where
2 + f 2 + f 2 − k 2 − k 2 − k 2 ) − (k f + k f + k f )2
∆ = (1 + kI )4 + (1 + kI )2 (fQ U V Q U V Q Q U U V V
The matrix C −1 can be calculated analytically from the expression of the ma-
trix C given in Eq. (9.91). A straightforward calculation yields the expressions
contained in Table 9.5, whence one gets the so-called Unno-Rachkovsky solutions 1
& '
I(0, µ) = B0 1 + βµ ∆−1 (1 + kI ) (1 + kI )2 + fQ2
+ fU2 + fV2
&
Q(0, µ) = −B0 βµ ∆−1 (1 + kI )2 kQ − (1 + kI ) (kU fV − kV fU )
'
+ fQ (kQ fQ + kU fU + kV fV )
1
Note that the quantity (kQ fU − kU fQ ) appearing in the expression of V (0, µ) – and of
the matrix elements (C −1 )03 and (C −1 )30 – has been written down just to point out the ‘cyclic’
character of the formulae for Q , U , V ; however, as obvious from Eqs. (9.39) and (9.32), it is
identically zero.
414 CHAPTER 9
&
U (0, µ) = −B0 βµ ∆−1 (1 + kI )2 kU − (1 + kI ) (kV fQ − kQ fV )
'
+ fU (kQ fQ + kU fU + kV fV )
&
V (0, µ) = −B0 βµ ∆−1 (1 + kI )2 kV − (1 + kI ) (kQ fU − kU fQ )
'
+ fV (kQ fQ + kU fU + kV fV ) . (9.110)
As far as the quantity ∆ (which is the determinant of the matrix C) is concerned,
it is worth noticing that the inequality
(kQ fQ + kU fU + kV fV )2 ≤ (kQ
2 2
+ kU + kV2 ) (fQ
2
+ fU2 + fV2 ) (9.111)
implies
∆ ≥ (1 + kI )2 + (fQ
2
+ fU2 + fV2 ) (1 + kI )2 − (kQ
2 2
+ kU + kV2 ) .
On the other hand, it can be shown via Eqs. (9.39) and (9.32) that the term in
the second bracket is positive. It follows that, whatever the values of the various
parameters, one always has ∆ > 0.
Equations (9.110) are particularly important because they provide simple ana-
lytical expressions for the Stokes parameters profiles of the line radiation emerging
from a magnetized stellar atmosphere. Although the Milne-Eddington model (sum-
marized by approximations a), b) and c) at the beginning of this section) is indeed
the simplest schematization of a real stellar atmosphere, the Unno-Rachkovsky
formulae contain the basic physics of radiative transfer for polarized radiation
and have been widely used in astrophysical applications. They were first de-
rived by Unno (1956) who neglected, however, magneto-optical effects. By setting
fQ = fU = fV = 0 in Eqs. (9.110), one obtains the simplified Unno solutions
I(0, µ) = B0 1 + βµ ∆−1
0 (1 + kI )
Q(0, µ) = −B0 βµ ∆−1
0 kQ
U (0, µ) = −B0 βµ ∆−1
0 kU
V (0, µ) = −B0 βµ ∆−1
0 kV ,
where
∆0 = (1 + kI )2 − kQ
2
− kU
2
− kV2 .
The equations acquired their standard form with the introduction of magneto-
optical effects by Rachkovsky (1962b).
The Stokes parameters profiles given by the Unno-Rachkovsky solutions depend
on 9 different parameters: the seven quoted under point b) above plus the coeffi-
cients B0 and β representing the surface value and the slope of the Planck function.
Obviously there is a further dependence on the Zeeman pattern and on the rest
LINE FORMATION IN A MAGNETIC FIELD 415
wavelength of the spectral line. The dependence on the parameter B0 can be
dropped by normalizing the profiles to the intensity Ic (0, µ) of the nearby contin-
uum. Since in the continuum all the coefficients k and f vanish, we have from
Eqs. (9.110)
Ic (0, µ) = B0 (1 + βµ) .
A frequently used normalization is the following
Ic (0, µ) − I(0, µ)
RI (0, µ) =
Ic (0, µ)
βµ & '
= 1 − ∆−1 (1 + kI ) (1 + kI )2 + fQ
2
+ fU2 + fV2
1 + βµ
Q(0, µ) βµ
RQ (0, µ) = =− ∆−1
Ic (0, µ) 1 + βµ
& '
× (1 + kI )2 kQ − (1 + kI )(kU fV − kV fU ) + fQ (kQ fQ + kU fU + kV fV )
U (0, µ) βµ
RU (0, µ) = =− ∆−1
Ic (0, µ) 1 + βµ
& '
× (1 + kI )2 kU − (1 + kI )(kV fQ − kQ fV ) + fU (kQ fQ + kU fU + kV fV )
V (0, µ)
RV (0, µ) =
Ic (0, µ)
βµ & '
=− ∆−1 (1 + kI )2 kV + fV (kQ fQ + kU fU + kV fV ) . (9.112)
1 + βµ
Because of the large number of parameters on which the Unno-Rachkovsky so-
lutions depend, we show in Figs. 9.5 and 9.6 just a few illustrative examples.1
Figure 9.5 shows the Stokes profiles for six different Zeeman patterns (the same as
in Figs. 3.2 and 9.3) and for a fixed set of parameters’ values (which will be referred
to as ‘standard case’). Figure 9.6 refers to a normal Zeeman triplet (panel (a) in
Fig. 9.5) and shows how the Stokes profiles change when the different parameters
are varied one at a time.
As apparent from panel (d) of Fig. 9.6, the I and V profiles are unaffected by a
variation of the azimuth angle χ, while Q and U tend to change into each other.
Since χ is independent of optical depth, such behavior is implied by the definition
itself of the Stokes parameters: a variation ∆χ is equivalent to a rotation of the
reference direction through an angle α = −∆χ . This property is in fact contained
in Eqs. (9.110): bearing in mind Eqs. (9.39) and (9.32), it can be easily seen that
I and V are independent of χ, while Q and U change according to the law (cf.
1
An atlas of Stokes profiles, based on the Unno-Rachkovsky formulae, has been presented by
Arena and Landi Degl’Innocenti (1982).
416 CHAPTER 9
Fig.9.5. The Stokes parameters profiles, as given by the Unno-Rachkovsky formulae (9.110), are
plotted as functions of the reduced wavelength for the same transitions considered in Fig.9.3.
The profiles are normalized to B0 , and the values of the relevant parameters are the following
(‘standard case’): βµ = 5, κL = 20, vA = 0, vB = 1.5, θ = 60◦ , χ = 0◦ , a = 0.05 (the last
5 values are the same as in Fig.9.3).
Eqs. (1.45))
Q(χ + ∆χ) = cos(2 ∆χ) Q(χ) − sin(2 ∆χ) U (χ)
U (χ + ∆χ) = sin(2 ∆χ) Q(χ) + cos(2 ∆χ) U (χ) .
It should be remarked that Eqs. (9.110) provide analytical expressions for the
LINE FORMATION IN A MAGNETIC FIELD 417
Fig.9.6. Influence of the different parameters, varied one at a time, on the Stokes profiles. The
profiles refer to a Zeeman triplet and are calculated as in Fig.9.5; the full line corresponds to the
‘standard case’. (a) κL = 5 (dotted), κL = 40 (dashed); (b) vB = 0.5 (dotted), vB = 3 (dashed);
(c) θ = 30◦ (dotted), θ = 80◦ (dashed); (d) χ = 30◦ (dotted), χ = 60◦ (dashed); (e) a = 0.0
(dotted), a = 0.2 (dashed).
emerging Stokes parameters profiles (tc = 0, µ > 0). To obtain analytical expres-
sions valid for any optical depth and direction we must resort to Eqs. (9.108). The
result for µ > 0 is trivial
I(tc , µ) = I(0, µ) + B0 β tc
Q(tc , µ) = Q(0, µ)
U (tc , µ) = U (0, µ)
V (tc , µ) = V (0, µ) ,
while for µ < 0 one needs the expressions for the matrix elements of C −1 contained
418 CHAPTER 9
in Table 9.5 and those for the exponential of a matrix deduced in App. 5. The
resulting formulae are rather involved and will not be given here.
9.9. More General Analytical Solutions
The Milne-Eddington atmosphere, which is at the basis of the Unno-Rachkovsky
solution to the radiative transfer equation, is in many cases a too rough approx-
imation for a real stellar atmosphere. The main limitations of the model are the
LTE assumption, the linearity of the Planck function with optical depth, and the
assumption of a depth-independent κL (the ratio between line and continuum opac-
ity).
If we drop these approximations, but retain all the other assumptions of the
Milne-Eddington model, it is still possible to obtain a fully analytical solution to
the transfer equation. To be definite, let us make the following (somewhat more
realistic) assumptions:
a ) the atmosphere where the line is formed is plane-parallel and semi-infinite;
b ) the quantities B, θ, χ , ∆λD , Γ , wA are depth-independent throughout the
whole atmosphere.
Assumption b ) means that the magnetic field vector and the absorption and
anomalous dispersion profiles are independent of optical depth. It follows that the
propagation matrix C defined in Eq. (9.91) has the form
C(tc ) = 1 + κL (tc ) H ,
with H = const. According to Eqs. (9.104), (9.101), (9.102) and (9.92), the emerg-
ing Stokes parameters are given by
∞ t
− µc −
KL (tc )
H dt
I(0, µ) = e e µ
Sc (tc ) 1 + κL (tc ) SL (tc ) H U c , (9.113)
µ
0
where
tc
KL (tc ) = κL (tc ) dtc ,
0
provided
t
− µc −
KL (tc )
H
lim e e µ
Sc (tc ) 1 + κL (tc ) SL (tc ) H U = 0 . (9.114)
tc →∞
Using Eq. (9.113), explicit expressions for I(0, µ) can be obtained once the functions
Sc (tc ), SL (tc ), and κL (tc ) are specified.
LINE FORMATION IN A MAGNETIC FIELD 419
For example, let us consider a simple generalization of the Milne-Eddington model
which is suitable to describe a chromospheric temperature rise at low optical depths.
Assuming
κL = const.
−α1 tc
Sc (tc ) = B0 (1 + β tc ) + A1 e
−α1 tc −α2 tc
SL (tc ) = B0 (1 + β tc ) + A1 e − A2 e (9.115)
with α1 and α2 positive, so that Eq. (9.114) is satisfied, one has the possibility of
adjusting the various parameters so as to simulate a variety of different behaviors
for the continuum and line source functions. In particular, the last term in SL (tc )
allows for a drop of the line source function below the continuum source function
at optical depths tc ≤ 1/α2 .
The integral resulting from the substitution of Eqs. (9.115) into Eq. (9.113) can
be carried out with the help of Eqs. (9.106) and (9.107). After some algebra we
obtain
&
I(0, µ) = B0 1 + βµ C −1 + A1 1 − α1 µ (α1 µ 1 + C)−1
'
− A2 1 − (1 + α2 µ) (α2 µ 1 + C)−1 U , (9.116)
which of course reduces to the Unno-Rachkovsky solution (9.109) for A1 = A2 = 0.
Figure 9.7 shows the Stokes parameters profiles deduced from Eq. (9.116) in some
particular cases. The profiles are normalized to the intensity of the nearby contin-
uum, which is given by
A1
Ic (0, µ) = B0 (1 + βµ) + .
1 + α1 µ
A remarkable expression for the emerging Stokes parameters can be obtained if,
in addition to assumptions a ) and b ) specified above, we adopt the LTE hypoth-
esis. To derive this expression we use the line optical depth tL (instead of tc ) as
independent variable.
Using Eqs. (9.36) and (9.34), we can write the transfer equation in the form
dI
µ = (κc 1 + H) I − (κc Sc 1 + SL H ) U ,
dtL
where κc is given by Eq. (9.37) and H by Eq. (9.50). According to Eqs. (9.104),
(9.101) and (9.102), the emerging Stokes parameters are given by
∞
−
Kc (tL ) t
− µL H dt
I(0, µ) = e µ
e κc (tL ) Sc (tL ) 1 + SL (tL ) H U L , (9.117)
µ
0
420 CHAPTER 9
Fig.9.7. Stokes parameters profiles as functions of reduced wavelength, computed from Eq.(9.116)
for a normal Zeeman triplet and the source functions Sc and SL plotted in the lower part of the
figure. The profiles are normalized to the continuum intensity, the source functions to B0 . Full
line: A1 = A2 = 0 (model (a), i.e. a Milne-Eddington atmosphere); dotted line: A1 /B0 = 4,
α1 = 5, A2 = 0 (model (b)); dashed line: A1 /B0 = A2 /B0 = 4, α1 = 5, α2 = 10 (model (c)).
The parameters β and µ are set to 5 and 1 respectively. The remaining parameters have the same
values as in Fig.9.5.
with
tL
Kc (tL ) = κc (tL ) dtL ,
0
provided
−
Kc (tL ) t
− µL H
lim e µ
e κc (tL ) Sc (tL ) 1 + SL (tL ) H U = 0 .
tL →∞
Equation (9.117), which is basically equivalent to Eq. (9.113), can be further de-
veloped in the LTE case, where
Sc (tL ) = SL (tL ) = BP (tL ) .
Since
K (t ) !
−
Kc (tL ) t
− µL H d − cµ L
t
− L H
e µ
e κc (tL ) 1 + H = − µ e e µ ,
dtL
integration by parts of Eq. (9.117) yields
∞ K (t ) t !
− cµ L − L H dBP
I(0, µ) = BP (0) 1 + e e µ dtL U, (9.118)
dtL
0
LINE FORMATION IN A MAGNETIC FIELD 421
where BP (0) = BP (tL = 0). Using Eq. (A5.23), the matrix exp(−tL H/µ) can be
written in the form
t
− µL H
4 t
− µL λi ( r)
h,
e = e Ni ( h, r ) , (9.119)
i=1
where the matrices Ni are given by Eqs. (A5.24) and λi are the eigenvalues of the
matrix H (cf. Sect. 8.4)
λ1 ( h, r ) = hI + Λ+ ( h, r ) λ3 ( h, r ) = hI + i Λ− ( h, r )
λ2 ( h, r ) = hI − Λ+ ( h, r ) λ4 ( h, r ) = hI − i Λ− ( h, r ) ,
with Λ± ( h, r ) given by Eqs. (A5.18). Equation (9.119) enables the emerging Stokes
parameters to be expressed in terms of Laplace transforms. We recall that the
Laplace transform of the function f (x ; µ) is defined, in the half plane Re p > 0, by
∞
−px
L(p ; µ) = e f (x ; µ) dx .
0
Substitution of Eq. (9.119) into Eq. (9.118) yields1
4 !
λ ( h, r )
I(0, µ) = BP (0) 1 + L i ; µ Ni ( h, r ) U , (9.120)
i=1
µ
where L is the Laplace transform of the function
Kc (x)
− µ dBP
f (x ; µ) = e .
dx
Equation (9.120) is rather general and provides analytical expressions for the emerg-
ing Stokes parameters for a wide variety of model atmospheres. As observed in
App. 5, the matrices Ni are ill-defined when Λ+ = Λ− = 0. In that case Eq. (9.120)
should be replaced by the following
!
hI hI 1 hI
I(0, µ) = BP (0)+L ; µ 1 − L1 ; µ G( h, r ) + L2 ; µ G( h, r ) U ,
2
µ µ 2 µ
where the matrix G is given by Eq. (A5.25), and where L1 (p ; µ) and L2 (p ; µ) are
the Laplace transforms of the functions
2
x − Kcµ(x) dBP x K (x)
− cµ dBP
f1 (x ; µ) = e , f2 (x ; µ) = e ,
µ dx µ dx
respectively.
1
The condition Re p > 0 is clearly satisfied for the eigenvalues λ1 , λ3 and λ4 . Using the
inequality
( h · r )2 ≤ h2 r 2 ,
it can be easily proved via Eqs. (A5.18) and (9.32) that it is also satisfied for λ2 .
422 CHAPTER 9
9.10. Solutions for Special Magnetic Field Orientations
As outlined in Sect. 8.3.c, there are some special cases where, because certain
elements of the propagation matrix are zero, the transfer equation can be solved
in closed form without any additional assumption. Two such cases occur when, at
each point along the ray path, the magnetic field vector is parallel or perpendicular
to the propagation direction.
Let us consider the former case (θ = 0◦ or 180◦ in Fig. 9.1). From Eqs. (9.32) we
have
1
hI = ηb + ηr
2
hQ = hU = 0 rQ = rU = 0
1 1
hV = ± ηr − ηb rV = ± ρr − ρb , (9.121)
2 2
where the plus sign in the expressions of hV and rV refers to θ = 0◦ and the minus
sign to θ = 180◦ . Substitution into the transfer equation (9.35) yields
dI
= (1 + κL hI ) I + κL hV V − (Sc + κL hI SL )
dτc
dQ
= (1 + κL hI ) Q + κL rV U
dτc
dU
= (1 + κL hI ) U − κL rV Q
dτc
dV
= (1 + κL hI ) V + κL hV I − κL hV SL . (9.122)
dτc
These equations can be solved either by the evolution operator technique (cf.
Eq. (8.27) which refers to a similar case) or by diagonalization. We apply here
the second method, which is very simple in this case (cf. Eqs. (8.43) and (8.44)).
As apparent from Eqs. (9.122), the Stokes parameters Q and U are only coupled
with each other: since the equations for Q and U do not contain source terms, we
have, at any optical depth1
Q=U =0
provided the boundary values of Q and U are also zero. As far as I and V are
concerned, summation and subtraction of the first and fourth of Eqs. (9.122) gives,
1
Note that Eqs. (8.44) do not imply the solution Q = U = 0, although the propagation matrix
has the same form. This is because all atomic coherences are neglected in the present chapter, so
that the condition hQ = hU = 0 implies that the Q and U components of the emission vector are
zero (see Eq. (9.92)).
LINE FORMATION IN A MAGNETIC FIELD 423
in the θ = 0◦ case1
d
(I + V ) = (1 + κL ηr ) (I + V ) − (Sc + κL ηr SL )
dτc
d
(I − V ) = (1 + κL ηb ) (I − V ) − (Sc + κL ηb SL ) , (9.123)
dτc
where Eqs. (9.121) have been used. These equations are uncoupled and have the
formal solutions (cf. Eqs. (8.4) and (8.18))
∞ : τc
− 1 + κL (τc ) ηr (τc ) dτc
(I + V )τ =0 = e 0
c
0
× Sc (τc ) + κL (τc ) ηr (τc ) SL (τc ) dτc
∞ : τc
− 1 + κL (τc ) ηb (τc ) dτc
(I − V )τ =0 = e 0
c
0
× Sc (τc ) + κL (τc ) ηb (τc ) SL (τc ) dτc , (9.124)
which allow the emerging I and V Stokes parameters to be written in the form
1
I(0) = (I + V )τ =0 + (I − V )τ =0
2 c c
1
V (0) = (I + V )τ =0 − (I − V )τ =0 . (9.125)
2 c c
Equations (9.123) show that magneto-optical effects do not influence the transfer
of radiation: this is indeed one of the few cases where these effects can a priori
be neglected (cf. Sect. 9.22). It also appears that the quantity (I + V ) – which
represents twice the amount of right circular polarization contained in the radiation
beam, see Sect. 1.6) – obeys the same transfer equation as the intensity of an
(unpolarized) beam propagating through a non-magnetic atmosphere, with the
only difference that the profile ηr would be replaced, in the latter case, be the
profile η defined in Eq. (9.65). The same holds for (I − V ), with ηb playing the role
of ηr . If we restrict attention to a Zeeman triplet (normal or anomalous), the ηr
and ηb profiles are nothing but the η profile shifted in wavelength by the amount
g ∆λB to the red or to the blue, respectively2 (see Eqs. (9.26), (9.27) and (9.29)).
Therefore, if the amplitude (besides the direction) of the magnetic field is constant
along the ray path, and if suitable boundary conditions (like those in Eqs. (9.42))
1
The case θ = 180◦ leads to the same equations with the interchange of ηr and ηb .
2
Obviously, the opposite shifts occur if the Landé factor is negative.
424 CHAPTER 9
hold, we have, at any optical depth1
1 ¯ ¯ + g ∆λ )
I(λ) = I(λ − g ∆λB ) + I(λ B
2
1 ¯ ¯ + g ∆λ ) ,
V (λ) = I(λ − g ∆λB ) − I(λ B (9.126)
2
¯
where I(λ) is the intensity that would be observed if the magnetic field were
switched off leaving unchanged all the other physical parameters (including the
source functions).
Let’s now turn to the case where the magnetic field vector is perpendicular to
the propagation direction. For θ = 90◦ , Eqs. (9.32) reduce to
1 η + ηr
hI = η + b
2 p 2
1 η + ηr 1 ρ + ρr
hQ = η − b cos 2χ rQ = ρ − b cos 2χ
2 p 2 2 p 2
1 η + ηr 1 ρ + ρr
hU = η − b sin 2χ rU = ρ − b sin 2χ
2 p 2 2 p 2
hV = 0 rV = 0 ,
and Eqs. (9.35) now read
dI
= (1 + κL hI ) I + κL hQ Q + κL hU U − (Sc + κL hI SL )
dτc
dQ
= (1 + κL hI ) Q + κL hQ I − κL rU V − κL hQ SL
dτc
dU
= (1 + κL hI ) U + κL hU I + κL rQ V − κL hU SL
dτc
dV
= (1 + κL hI ) V + κL rU Q − κL rQ U .
dτc
These equations can be solved by diagonalization provided the azimuth angle χ is
constant along the ray path. Introducing the linear combinations2
Q̃ = Q cos 2χ + U sin 2χ
Ũ = U cos 2χ − Q sin 2χ , (9.127)
1
The case θ = 180◦ yields a sign switch in the expression of V (λ).
2
These combinations are nothing but the Stokes parameters defined in the preferred frame
(see Sect. 5.5).
LINE FORMATION IN A MAGNETIC FIELD 425
and setting
1 η + ηr
hW = hQ (θ = 90◦ , χ = 0◦ ) = η − b
2 p 2
1 ρ + ρr
rW = rQ (θ = 90◦ , χ = 0◦ ) = ρ − b ,
2 p 2
one gets
dI
= (1 + κL hI ) I + κL hW Q̃ − (Sc + κL hI SL )
dτc
dQ̃
= (1 + κL hI ) Q̃ + κL hW I − κL hW SL
dτc
dŨ
= (1 + κL hI ) Ũ + κL rW V
dτc
dV
= (1 + κL hI ) V − κL rW Ũ .
dτc
These are strictly similar to Eqs. (9.122) and can be solved by the same method.
Since Ũ and V are only coupled with each other and no source term is present in
their equations, one has
Ũ = V = 0 .
On the other hand, summation and subtraction of the first two equations yields
the decoupled equations
d , - , -
(I + Q̃) = 1 + κL ηp (I + Q̃) − Sc + κL ηp SL
dτc
d , η + ηr - , η + ηr -
(I − Q̃) = 1 + κL b (I − Q̃) − Sc + κL b SL ,
dτc 2 2
whose solutions are
∞ : τc
− 1 + κL (τc ) ηp (τc ) dτc
(I + Q̃)τ =0 = e 0
c
0
× Sc (τc ) + κL (τc ) ηp (τc ) SL (τc ) dτc
∞ : τc ηb (τc ) + ηr (τc )
− 1 + κL (τc ) dτc
(I − Q̃)τ =0 = e 0 2
c
0
ηb (τc ) + ηr (τc )
× Sc (τc ) + κL (τc ) SL (τc ) dτc . (9.128)
2
426 CHAPTER 9
Finally, inversion of Eqs. (9.127) leads to the following expressions for the emerging
Stokes parameters
1
I(0) = (I + Q̃)τ =0 + (I − Q̃)τ =0
2 c c
1
Q(0) = (I + Q̃)τ =0 − (I − Q̃)τ =0 cos 2χ
2 c c
1
U (0) = (I + Q̃)τ =0 − (I − Q̃)τ =0 sin 2χ
2 c c
V (0) = 0 .
The above equations show that magneto-optical effects do not affect the Stokes
parameters in this case either.
9.11. The Stepanov Solution
In the early days of polarized radiative transfer, the importance of magneto-optical
effects was not completely realized, and the transfer equations were often considered
in the simplified form resulting from the drastic approximation rQ = rU = rV = 0.
Under this approximation, Eqs. (9.35) reduce to
dI
= I − Sc + κL hI (I − SL ) + hQ Q + hU U + hV V
dτc
dQ
= Q + κL hQ (I − SL ) + hI Q
dτc
dU
= U + κL hU (I − SL ) + hI U
dτc
dV
= V + κL hV (I − SL ) + hI V . (9.129)
dτc
Bearing in mind Eqs. (9.32), we can write
hQ = a cos 2χ , hU = a sin 2χ , hV = b ,
with
1 η + ηr 1
a= η − b sin2 θ , b= η − ηb cos θ . (9.130)
2 p 2 2 r
Setting
a b
√ = cos δ , √ = sin δ , (9.131)
a + b2
2 a + b2
2
LINE FORMATION IN A MAGNETIC FIELD 427
we have
hQ = cos δ cos 2χ h2Q + h2U + h2V
hU = cos δ sin 2χ h2Q + h2U + h2V
hV = sin δ h2Q + h2U + h2V . (9.132)
The Stepanov solution derives from the assumption that both χ and δ are inde-
pendent of τc : this assumption enables Eqs. (9.129) to be cast into diagonal form.
Contrary to the angle χ (the azimuth of the magnetic field vector defined in
Fig. 9.1), the angle δ, which depends on wavelength, has no precise physical mean-
ing. However, as apparent from Eqs. (9.130) and (9.131), the assumption δ = const.
implies that the parameters B, θ, ∆λD , Γ , wA should be independent of optical
depth. Thus the approximations underlying the Stepanov solution are equivalent
to approximation b ) considered in Sect. 9.9 plus the additional approximation of
neglecting magneto-optical effects.
Let us substitute Eqs. (9.132) into Eqs. (9.129), and let us replace the last
three equations by the linear combinations obtained by summation of the sec-
ond, third, and fourth equation multiplied, respectively, by the factors cos δ cos 2χ ,
cos δ sin 2χ , sin δ ; then by the factors sin δ cos 2χ , sin δ sin 2χ , − cos δ ; and finally
by the factors sin 2χ , − cos 2χ , 0. The following set of equations is obtained
d
I = I − Sc + κL hI (I − SL ) + h2Q + h2U + h2V PA
dτc
d
P = PA + κL h2Q + h2U + h2V (I − SL ) + hI PA
dτc A
d
P = (1 + κL hI ) PB
dτc B
d
P = (1 + κL hI ) PC , (9.133)
dτc C
where
PA = cos δ cos 2χ Q + cos δ sin 2χ U + sin δ V
PB = sin δ cos 2χ Q + sin δ sin 2χ U − cos δ V
PC = sin 2χ Q − cos 2χ U . (9.134)
The last two equations in the set (9.133) show that, provided Q , U , and V are
zero at the boundary,
PB = PC = 0 . (9.135)
The solution to the algebraic system (9.134) with the condition (9.135) is
Q = cos δ cos 2χ PA , U = cos δ sin 2χ PA , V = sin δ PA .
428 CHAPTER 9
Finally, by summing and subtracting the first two equations in the set (9.133) one
gets the decoupled equations
dI (+)
= 1 + κL h(+) I (+) − Sc + κL h(+) SL
dτc
dI (−)
= 1 + κL h(−) I (−) − Sc + κL h(−) SL , (9.136)
dτc
where
I (+) = I + PA h(+) = hI + h2Q + h2U + h2V
I (−) = I − PA h(−) = hI − h2Q + h2U + h2V .
Equations (9.136) are the so-called Stepanov equations (Stepanov, 1958), and can
be easily integrated (cf. the analogous solutions (9.124) or (9.128)). The Stepanov
equations are seldom used in modern research. They have been illustrated mainly
because of their historical importance in the problem of line formation in a magnetic
field.
9.12. The Intense Field Solution
In the limiting case of very strong magnetic field, the various Zeeman components
spread out considerably in wavelength and the spectral line consists of a num-
ber of independent components, each associated with the corresponding Zeeman
component.1
This physical situation, which will be referred to as ‘intense field’ and which is
characterized by the inequality (see Eqs. (9.26))
∆λB
vB = 1,
∆λD
is easily achieved in laboratory plasmas where it is currently used to illustrate the
Zeeman effect and to measure Landé factors. In astrophysical plasmas, on the
contrary, the Doppler broadening is generally quite large, so that really strong
fields are needed for the above inequality to hold. According to the discussion
at the beginning of Sect. 9.6, the magnetic field strength should be of order 104 -
105 G for optical lines; in the far infrared, however, considerably weaker fields are
sufficient, because vB is proportional to the line wavelength (see Eqs. (9.25) and
(9.26)).
The problem of line formation in an intense magnetic field can be dealt with
by introducing a drastic approximation. This consists in the assumption that, for
1
According to the basic hypotheses of this chapter (Sect. 9.1), the magnetic splitting
– although large – is assumed to follow the Zeeman effect regime, and the spectral line is assumed
to be ‘unstructured’.
LINE FORMATION IN A MAGNETIC FIELD 429
any assigned wavelength, only one Zeeman component contributes to the elements
of the propagation matrix. In other words, we assume that the expressions of the
profiles ηq and ρq in a wavelength interval about the Zeeman component connecting
two specific sublevels (J M ) and (Ju Mu ) are (cf. Eqs. (9.27))
J J 1
ηq = Sq u (M , Mu ) √ H v − vA + vB (gu Mu − g M ), a
π
J J 1
ρq = Sq u (M , Mu ) √ L v − vA + vB (gu Mu − g M ), a , (9.137)
π
with q = (M − Mu ) = −1, 0, +1 for σb , π, and σr components respectively.
This approximation requires some comments. First of all, there are Zeeman
patterns where two or more components overlap irrespective of the magnetic field
strength. An example is given by the transition 5 P1 − 5 F1 (panel (c) in Fig. 3.2),
where overlappings of σ and π components and of two σ components occur. These
special cases would deserve a separate discussion but, being rather uncommon,
they will not be considered in the following. Second, it should be kept in mind
that the dispersion profiles have very broad wings, as apparent from Eqs. (5.57)
(see also Fig. 5.3). Since the distance between Zeeman components is roughly
∆λB , our present approximation introduces in ρq an error of order ∆λD /∆λB (the
corresponding error in ηq is much smaller being of order a ∆λ2D /∆λ2B ). Finally, the
approximation can hold at any optical depth only if there are no ‘contaminations’
between different Zeeman components: these could be caused either by a magnetic
field gradient or by a line-of-sight velocity gradient. If δ(∆λB ) represents the
variation of ∆λB between – say – the atmospheric layers τc = 0 and τc = 1, and if
δ(wA ) has the same meaning for wA , we should thus require
λ0
δ(∆λB ) ∆λB , δ(wA ) ∆λB . (9.138)
c
Let us consider the transfer equation in the form of Eq. (9.35), and let us restrict
attention to a wavelength interval about the Zeeman component connecting two
particular sublevels (J M ) and (Ju Mu ). This component – σb , π or σr according to
the value of (M − Mu ) – will be referred to as the i-component . Using assumption
(9.137), the elements of the propagation matrix can be written in the form (see
Eqs. (9.29) and (9.32))
(i)
hI = αI si η (i)
(i) (i)
hQ = αQ si η (i) rQ = αQ si ρ(i)
(i) (i)
hU = αU si η (i) rU = αU si ρ(i)
(i) (i)
hV = αV si η (i) rV = αV si ρ(i) , (9.139)
430 CHAPTER 9
TABLE 9.6
Expressions of the angular factors for different types of Zeeman components
(i) (i) (i) (i)
αI αQ αU αV
σb 1
4
(1 + cos2 θ) − 14 sin2 θ cos 2χ − 14 sin2 θ sin 2χ − 12 cos θ
1 1 1
π 2
sin2 θ 2
sin2 θ cos 2χ 2
sin2 θ sin 2χ 0
σr 1
4
(1 + cos2 θ) − 14 sin2 θ cos 2χ − 14 sin2 θ sin 2χ 1
2
cos θ
where
J J
si = Sq u (M , Mu )
1
η (i) = √ H v − vA + vB (gu Mu − g M ), a
π
1
ρ(i) = √ L v − vA + vB (gu Mu − g M ), a , (9.140)
π
(i) (i) (i) (i)
and where the angular factors αI , αQ , αU , αV are defined in Table 9.6. We will
now show that the transfer equation can be cast into diagonal form if the ratios
(i) (i) (i)
αQ αU αV
(i)
, (i)
, (i)
αI αI αI
are assumed to be independent of optical depth. As apparent from Table 9.6, this
means that the direction of the magnetic field vector is assumed to be constant
along the ray path.
Setting
(i) (i) (i)
αQ α αV
Q = (i) P , U= U (i)
P , V = (i)
P , (9.141)
αI αI αI
and noticing that, both for σ and π components
(i) 2 (i) 2 (i) 2 (i) 2
αQ + αU + αV = αI ,
we obtain from Eqs. (9.35)
dI (i)
= I − Sc + κL αI si η (i) (I + P − SL )
dτc
dP (i)
= P + κL αI si η (i) (I + P − SL ) .
dτc
Summation and subtraction of these two equations yields the decoupled equations
d (i) (i)
(I + P ) = 1 + 2κL αI si η (i) (I + P ) − Sc + 2 κL αI si η (i) SL
dτc
d
(I − P ) = (I − P ) − Sc . (9.142)
dτc
LINE FORMATION IN A MAGNETIC FIELD 431
It is interesting to note that the latter of Eqs. (9.142) has the same form as the
transfer equation for the continuum intensity Ic , while the former is just the trans-
(i) (i)
fer equation for the intensity If of a fictitious, non-magnetic line having a ratio κf
between line and continuum absorption coefficient given by
(i) (i)
κf = 2 κL αI si (9.143)
and line source function SL . Setting
(i)
I − P = Ic , (I + P ) = If ,
and using Eqs. (9.141), we can write for the Stokes parameters in a wavelength
interval about the i-component
1 (i)
I (i) (τc ) = I (τ ) + If (τc )
2 c c
(i)
1 αQ (i)
Q(i) (τc ) = − I (τ ) − If (τc )
2 α(i) c c
I
(i)
1 αU (i)
U (i) (τc ) = − I (τ ) − If (τc )
2 α(i) c c
I
(i)
1 αV (i)
V (i) (τc ) = − I (τ ) − If (τc ) , (9.144)
2 α(i) c c
I
where
∞
−(τc − τc )
Ic (τc ) = Sc (τc ) e dτc
τc
∞ : τc
−
(i)
1 + κf η (i) dτc
dτc .
(i) (i) (i) τc τc
If (τc ) = Sc + κ f η SL τ e (9.145)
c
τc
The equations above show that magneto-optical effects do not affect the Stokes
parameters. This is a direct consequence of the two basic assumptions of this
section, namely that the magnetic field is very strong and has a fixed direction.
Bearing in mind the geometrical interpretation of the transfer equation presented in
Sect. 5.6, we see from Eqs. (9.139) that the formal vectors η and ρ
point in the same
(or the opposite) direction in the Poincaré space (Q/I, U/I, V /I); furthermore,
since the angles θ and χ are assumed to be constant, this direction is fixed. It
follows that the formal vector p is also parallel to the same direction, so that the
term ρ × p in Eq. (5.74b) – the only term due to magneto-optical effects – is zero.
432 CHAPTER 9
In the special case of the Milne-Eddington model atmosphere, the expressions for
the emerging Stokes parameters in a wavelength interval about the i-component
become very simple. We write them down because they will be needed in subse-
quent applications (see Sects. 9.19, 9.20 and 9.21). According to assumptions a),
b) and c) of Sect. 9.8, we have
(i)
Sc = SL = B0 (1 + β tc ) , κf = const. , η (i) = const. ,
and Eqs. (9.145) reduce to
Ic (0, µ) = B0 (1 + βµ)
(i) 1
If (0, µ) = B0 1 + βµ (i)
.
1 + κf η (i)
Substituting into Eqs. (9.144) and adopting the normalization of Eqs. (9.112), we
obtain
(i)
(i) Ic (0, µ) − I(0, µ) 1 βµ κf η (i)
RI (0, µ) = =
Ic (0, µ) 2 1 + βµ 1 + κ(i) η (i)
f
(i)
(i)
(i) Q(0, µ) 1 βµ αQ κf η (i)
RQ (0, µ) = =−
Ic (0, µ) 2 1 + βµ α(i) 1 + κ(i) η (i)
I f
(i) (i)
(i) U (0, µ) 1 βµ αU κf η (i)
RU (0, µ) = =−
Ic (0, µ) 2 1 + βµ α(i) 1 + κ(i) η (i)
I f
(i) (i)
(i) V (0, µ) 1 βµ αV κf η (i)
RV (0, µ) = =− (i) (i)
. (9.146)
Ic (0, µ) 2 1 + βµ α 1 + κf η (i)
I
These formulae can also be obtained by direct substitution into Eqs. (9.112) of
the following expressions (resulting from Eqs. (9.39), (9.139) and (9.143)) for the
elements of the propagation matrix
1 (i) (i)
kI = κ η
2 f
(i) (i)
1 αQ (i) (i) 1 αQ (i) (i)
kQ = κ η fQ = κ ρ
2 α(i) f 2 α(i) f
I I
(i) (i)
1 αU (i) (i) 1 αU (i) (i)
kU = κ η fU = κ ρ
2 α(i) f 2 α(i) f
I I
(i) (i)
1 αV (i) (i) 1 αV (i) (i)
kV = κ η fV = κ ρ . (9.147)
2 α(i) f 2 α(i) f
I I
LINE FORMATION IN A MAGNETIC FIELD 433
9.13. The Seares Formulae
The solution to the radiative transfer equation is particularly simple when dealing
with optically thin, homogeneous slabs rather than semi-infinite atmospheres. To
be definite, let us consider a cold, absorbing slab superposed on a deeper layer which
emits continuum, unpolarized radiation. The slab is assumed to absorb only in the
line (not in the continuum), and to have zero emission and small optical thickness.
This schematic situation (the reversing layer model) was used in the early decades
of this century to represent a stellar atmosphere. Nowadays, it should be regarded
as a slab model rather than a real model atmosphere.
In the reversing layer – that we assume magnetized – the transfer of radiation can
be described using Eq. (9.36) with κc = 0 (no continuum absorption) and SL = 0
(b)
(no emission). Let ∆τL and Ic denote, respectively, the optical thickness of the
slab and the intensity of the continuum radiation (at the line wavelength) entering
the slab from the lower boundary. Since ∆τL 1, we can replace the Stokes
vector in the right-hand side of Eq. (9.36) by the Stokes vector at the boundary
(Ic , 0, 0, 0)† . Hence we obtain
(b)
(b)
∆I Ic − I ∆U U
≡ = hI Ic(b) ≡− = hU Ic(b)
∆τL ∆τL ∆τL ∆τL
∆Q Q ∆V V
≡− = hQ Ic(b) ≡− = hV Ic(b) ,
∆τL ∆τL ∆τL ∆τL
where I, Q, U , V are the emerging Stokes parameters. Use of Eqs. (9.32) yields
!
1 η + ηr
I = Ic(b) 1 − ∆τL ηp sin2 θ + b 1 + cos2 θ
2 2
1 η + ηr
Q = −Ic(b) ∆τL ηp − b sin2 θ cos 2χ
2 2
1 η + ηr
U = −Ic(b) ∆τL η − b sin2 θ sin 2χ
2 p 2
1
V = −Ic(b) ∆τL η − ηb cos θ . (9.148)
2 r
These are the so-called Seares formulae (Seares, 1913). They give Stokes parame-
ters profiles which match exactly the corresponding dichroic profiles.
It should be remarked that the same result is predicted by the Unno-Rachkovsky
solutions under the limit of weak spectral lines (κL 1). Recalling Eqs. (9.39),
we obtain by series expansion of Eqs. (9.110) up to first order in κL
I(0, µ) = B0 (1 + βµ) − βµ κL hI
Q(0, µ) = −B0 βµ κL hQ
U (0, µ) = −B0 βµ κL hU
V (0, µ) = −B0 βµ κL hV , (9.149)
434 CHAPTER 9
which coincide with Eqs. (9.148) after the formal substitutions
βµ
B0 (1 + βµ) → Ic(b) , κ → ∆τL .
1 + βµ L
The Seares formulae have a very restricted domain of applicability. They are
mainly important from a historical point of view, being at the basis of the first
methods used for the diagnostics of magnetic fields in the solar atmosphere.
For weak spectral lines, we can find an approximate solution to the radiative
transfer equation which is valid for a semi-infinite atmosphere instead of a slab.
This solution can be regarded as a generalization of the Seares formulae.
Let us consider the transfer equation in the form (9.38).1 Using Eqs. (9.39) we
can write
dI
= 1 + κL H I − BP 1 + κL H U , (9.150)
dτc
with H and U given by Eqs. (9.50) and (9.49), respectively. Since κL 1, we
can apply a perturbative technique (cf. Sect. 8.7) and look for a solution to this
equation of the form
I(τc ) = I0 (τc ) + δI(τc ) ,
where I0 corresponds to κL = 0 (no spectral line) and δI is a first-order correction.
We thus replace Eq. (9.150) by the zero-order and first-order equations
d
I = I0 − BP U
dτc 0
d
δI = δI − κL H BP U − I0 .
dτc
The evolution operator corresponding to these equations is (cf. Eqs. (9.90) and
(9.103))
−(τc − τc )
O(τc , τc ) = e 1 ,
so that according to Eq. (9.101) we obtain the solutions
∞
−(τc − τc )
I0 (τc ) = e BP (τc ) U dτc = Ic (τc ) U
τc
∞
−(τc − τc )
δI(τc ) = e κL (τc ) H(τc ) BP (τc ) − Ic (τc ) U dτc ,
τc
where Ic is the continuum intensity at the line wavelength. The emerging Stokes
parameters are therefore given by
I(0) = I0 (0) + δI(0)
1
We restrict attention to the LTE transfer equation rather than the more general equation (9.35).
This is justified by the fact that LTE is generally a very good approximation for weak lines.
LINE FORMATION IN A MAGNETIC FIELD 435
with
∞
−τc
I0 (0) = e BP (τc ) U dτc = Ic (0) U
0
∞
−τc
δI(0) = e κL (τc ) H(τc )
0
∞
−(τc − τc )
× BP (τc ) − e BP (τc ) dτc U dτc . (9.151)
τc
These expressions will be referred to as the generalized Seares formulae. They
reduce to Eqs. (9.149) for the Milne-Eddington model atmosphere, where κL =
const. , H = const. , BP (τc ) = B0 (1 + βµ τc ). It should be noticed that magneto-
optical effects are unimportant when the spectral line is weak.
9.14. The Magnetic Field with Variable Azimuth
The problem of line formation in a magnetic field whose azimuth is variable with
optical depth can be reduced to the corresponding problem in a magnetic field with
constant azimuth via a simple transformation on the propagation matrix.
To illustrate this statement, we consider a radiation beam flowing in a stellar
atmosphere along the direction Ω, and take the transfer equation in the form (9.35).
We denote by I the Stokes vector in a fixed reference system – say, the system
of Fig. 9.1 – and by I the Stokes vector in the depth-dependent system
(ea , eb , Ω)
whose reference unit vector ea points in the direction of the transverse component
of the local magnetic field vector. It follows that the azimuth angle of the magnetic
field is χ(τc ) in the fixed system and χ = 0◦ in the depth-dependent system.
The Stokes vectors I and I are related by the transformation (see Eqs. (1.45))
I = R I , (9.152)
where ⎛ ⎞
1 0 0 0
⎜0
⎜ cos 2χ(τc ) sin 2χ(τc ) 0⎟⎟
R=⎜ ⎟ .
⎝0 − sin 2χ(τc ) cos 2χ(τc ) 0⎠
0 0 0 1
Writing Eq. (9.35) in the matrix form
dI
= 1 + κ L H I − Sc 1 + κ L SL H U (9.153)
dτc
436 CHAPTER 9
(with H and U given by Eqs. (9.50) and (9.49)), and deriving Eq. (9.152) with
respect to τc , we obtain
dI dR dI dR
= I +R = I + R 1 + κ L H I − R Sc 1 + κ L SL H U .
dτc dτc dτc dτc
The Stokes vector I in the right-hand side can be expressed in terms of I via the
inverse transformation of Eq. (9.152)
I = R−1 I , (9.154)
which leads to
dI dR −1
= R + 1 + κL R H R−1 I − Sc 1 + κL SL R H R−1 R U . (9.155)
dτc dτc
A direct calculation shows that
⎛ ⎞
1 0 0 0
⎜ 0 cos 2χ(τ )
− sin 2χ(τc )
0⎟
⎜ ⎟
R−1 = ⎜
c
⎟ ,
⎝ 0 sin 2χ(τc ) cos 2χ(τc ) 0⎠
0 0 0 1
whence we easily obtain
dR −1 dχ
R =2 Z,
dτc dτc
with ⎛ ⎞
0 0 0 0
⎜0 0 1 0⎟
Z=⎝ ⎠ ,
0 −1 0 0
0 0 0 0
and, using Eqs. (9.32)
⎛ ⎞
hI hQ̃ 0 hV
⎜h 0 ⎟
⎜ hI rV ⎟
R H R−1 = ⎜ Q̃ ⎟ , (9.156)
⎝ 0 −rV hI rQ̃ ⎠
hV 0 −rQ̃ hI
where1
1 η + ηr
hQ̃ = hQ (χ = 0◦ ) = η − b sin2 θ
2 p 2
1 ρ + ρr
rQ̃ = rQ (χ = 0◦ ) = ρ − b sin2 θ . (9.157)
2 p 2
1
Note that the matrix in Eq. (9.156) is nothing but the propagation matrix for a magnetic
field having χ = 0◦ .
LINE FORMATION IN A MAGNETIC FIELD 437
Since, moreover
RU = U , ZU =0,
Eq. (9.155) can be rewritten in the form
dI
= 1 + κ L H I − Sc 1 + κ L SL H U , (9.158)
dτc
where
⎛ ⎞
hI hQ̃ 0 hV
⎜ rV 0 ⎟
2 dχ ⎜h hI ⎟
H = R H R−1 + Z = ⎜ Q̃ ⎟ , (9.159)
κL dτc ⎝ 0 −rV hI rQ̃ ⎠
hV 0 −rQ̃ hI
with
2 dχ dχ
rV = rV + = rV + 2 (9.160)
κL dτc dτL
(Eqs. (9.31) and (9.33) have been used to derive the last expression).
Comparison of Eqs. (9.158) and (9.153) shows that the transfer equation for I
is obtained from the transfer equation for I by substituting the matrix H for the
matrix H. As apparent from Eqs. (9.159) and (9.160), this means that the depth
dependence of the azimuth of the magnetic field can be taken into account by setting
χ = 0◦ at any optical depth and by adding the (wavelength-independent) term
2 (dχ /dτL ) to the Faraday rotation coefficient rV . This result was first derived
by Staude (1969). Obviously, once Eq. (9.158) is solved, the Stokes vector I is
obtained by applying the transformation (9.154).
The influence of the depth dependence of the azimuth angle on the Stokes pa-
rameters profiles is illustrated in Fig. 9.8, in the form of Unno-Rachkovsky solution
to the radiative transfer equation. Of course, such solution can only be used under
the assumption that (dχ /dτL ) is independent of optical depth (in addition to the
assumptions listed at the beginning of Sect. 9.8). As expected, the main effect of
azimuth variability is a decrease of linear polarization.
9.15. Numerical Solutions
In the preceding sections of this chapter we have considered several cases where
the transfer equation can be solved in closed form thanks to certain assumptions
about the structure of the medium where propagation occurs. In a real stellar
atmosphere, or in detailed model atmospheres (like for instance the solar models
given by Gingerich et al., 1971, or Vernazza et al., 1981), these assumptions are
generally not satisfied, so that the approximate solutions presented so far cannot
be used reliably. This does not mean, however, that such solutions are pointless,
since it should be borne in mind that the sophisticated models which are nowa-
days available are themselves based on specific assumptions about the physics and
438 CHAPTER 9
Fig.9.8. Stokes parameters profiles for the normal Zeeman triplet 1 S0 − 1 P1 emerging from
an atmosphere with depth-dependent magnetic field’s azimuth. The profiles are computed using
Eqs.(9.110) with fV = κL rV and r given by Eq.(9.160). They are normalized to B and plotted
V 0
against the reduced wavelength. The reference direction is parallel to the transverse component
of the magnetic field at τc = 0. The values of the relevant parameters are those of the ‘standard
case’ (see Fig.9.5). The full line, drawn for reference, corresponds to dχ /dτL = 0. The dotted
line corresponds to the following values of dχ /dτL : −0.1 (a); +0.1 (b); −0.5 (c); +0.5 (d).
geometry of real atmospheres. The interpretation of an observed phenomenon
may result clearer, and in some cases even more precise, if a rough model like the
Milne-Eddington is used instead of a detailed model. This is particularly true for
magnetic stellar atmospheres because the theoretical construction of fully consis-
tent magneto-hydrodynamic models is still in a preliminary phase and, on the other
hand, the deduction of empirical models is hampered by the inadequate knowledge
of the magnetic structure.
Detailed models for plane-parallel atmospheres generally consist of a table of
numbers giving the run of several physical quantities (such as temperature, partial
pressures, microturbulent velocity, magnetic field vector, etc.) with depth. Mod-
els can also be constructed for more complicated (non plane-parallel) atmospheres.
One could consider for instance a composite atmosphere formed by magnetic struc-
tures (like sunspots or flux-tubes) embedded in a plane-parallel atmosphere. This
would be a three-dimensional model and the actual run of the physical quantities
LINE FORMATION IN A MAGNETIC FIELD 439
with depth would depend on the individual ray path.
Different depth indicators are used in different models, the most common being
the geometrical depth, the column mass, and the continuum optical depth at a
reference frequency νr (typically the frequency corresponding to 5000 Å). Standard
calculations enable the conversion between different indicators. Such calculations
involve the equation of state of the atmospheric plasma and require the knowledge
of its chemical composition and of the processes which contribute to the continuum
opacity (see e.g. Unsöld, 1955; Aller, 1963; Mihalas, 1978).
Let us consider the radiative transfer problem in the case where the proper-
ties of the medium are specified through a numerical model atmosphere. Without
loosing in generality, we assume that the model is given in terms of τr , the contin-
uum optical depth at the reference frequency νr measured along the ray path (see
Eqs. (9.33)). The transfer equation (9.30) reads
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 0 0 0 I − Sc
⎜
d ⎜Q⎟ ⎟ ⎜ 0⎟ ⎜ ⎟
⎜0 1 0 ⎟⎜ Q ⎟
⎜ ⎟ = κr ⎜ ⎟⎜ ⎟
dτr ⎝ U ⎠ ⎝0 0 1 0 ⎠⎝ U ⎠
V 0 0 0 1 V
⎛ ⎞⎛ ⎞
hI hQ hU hV I − SL
⎜h −rU ⎟ ⎜ ⎟
⎜ hI rV ⎟⎜ Q ⎟
+ κR ⎜ Q ⎟⎜ ⎟ , (9.161)
⎝ hU −rV hI rQ ⎠⎝ U ⎠
hV rU −rQ hI V
κr and κR being the dimensionless quantities (see Eq. (9.31))
kc (ν0 ) kL
κr = , κR = , (9.162)
kc (νr ) kc (νr ) ∆νD
or, in matrix form1
d
I(τr ) = A(τr ) I(τr ) − b(τr ) , (9.163)
dτr
where
A = κr 1 + κR H
b = κ r Sc 1 + κ R SL H U , (9.164)
with H and U given by Eqs. (9.50) and (9.49), respectively.
The quantity κr and the elements of the matrix H can be evaluated numeri-
cally, at each optical depth and for any wavelength, via the standard computations
1
Obviously, Eq. (9.163) is identical, from the mathematical point of view, to Eq. (9.90).
The difference between the matrices A and C and the vectors b and j is only due to the use of
different depth indicators.
440 CHAPTER 9
mentioned above. In the case of LTE, this is also true for the source functions
(Sc = SL = BP ) and for the quantity κR . For non-LTE situations, we assume here
that Sc , SL and κR (or the ratio Nu /N , see Eqs. (9.14)) are known, i.e. given
together with the model. (The problem of non-LTE will be considered in more
generality in Chap. 14).
Equation (9.163) can be formally solved by the evolution operator method. Sim-
ilarly to Eq. (9.100), we have
τr
I(τr ) = O(τr , τr ) b(τr ) dτr + O(τr , τr ) I(τr ) , (9.165)
τr
and, for the Stokes parameters emerging from a semi-infinite atmosphere
∞
I(0) = O(0, τr ) b(τr ) dτr , (9.166)
0
provided
lim O(0, τr ) b(τr ) = 0 .
τr →∞
The evolution operator O(τr , τr ) satisfies the analogues of Eqs. (9.94)-(9.97) and
can be written in the form
O(τr , τr ) =
∞ τr τr τr & '
(−1)n
=1+ dτ1 dτ2 · · · dτn P A(τ1 ) A(τ2 ) · · · A(τn ) , (9.167)
n=1
n!
τr τr τr
P being the chronological ordering operator defined in Eqs. (9.98)-(9.99).
Unfortunately, Eq. (9.167) is too involved from the mathematical point of view
to provide a convenient method for the numerical computation of the evolution
operator. Alternative algorithms have been proposed for the solution of the trans-
fer equation (9.163). In this section we give a brief account of these methods and
of their main advantages and disadvantages. We point out, however, that an ex-
haustive discussion of the numerical accuracy and computing time typical of each
method is beyond the aim of this book.
A numerical technique that has been widely used for the solution of Eq. (9.163)
is the so-called 4-th order Runge-Kutta method (Beckers, 1969a; Wittmann, 1974;
Landi Degl’Innocenti, 1976). This method, whose mathematical description can
be found in Collatz (1966), requires – for each wavelength at which the equation is
to be solved – the division of the integration domain into a number N of intervals,
not necessarily equal, from τr = τmin (the smallest τr -value where the model is
specified or suitably extrapolated) to τr = τb (an appropriate boundary value):
this defines a grid of points τ0 = τmin , τ1 , τ2 , . . . , τN = τb . Boundary values for
LINE FORMATION IN A MAGNETIC FIELD 441
the Stokes parameters at the optical depth τb are then chosen.1 The integration
starts at τb and proceeds outwards, step by step, the Stokes parameters at the k-th
grid-point being related to those at the (k + 1)-th point by the equations
1
I(τk ) = I(τk+1 ) − ∆I (1) + 2 ∆I (2) + 2 ∆I (3) + ∆I (4) , (9.168)
6
where
& '
∆I (1) = (τk+1 − τk ) A(τk+1 ) I(τk+1 ) − b(τk+1 )
& 1 '
∆I (2) = (τk+1 − τk ) A(τ̃k ) I(τk+1 ) − ∆I (1) − b(τ̃k )
2
& 1 '
∆I (3) = (τk+1 − τk ) A(τ̃k ) I(τk+1 ) − ∆I (2) − b(τ̃k )
2
& '
∆I (4) = (τk+1 − τk ) A(τk ) I(τk+1 ) − ∆I (3) − b(τk ) , (9.169)
with τ̃k the midpoint of the (k + 1)-th interval
τk + τk+1
τ̃k = .
2
The accuracy of the Runge-Kutta integration is strictly related to the length of
the intervals. There are two distinct reasons why the intervals should be kept small.
The first is that the variation of A and b in a ‘large’ interval could be too large:
in other words, the three-point sampling of A and b performed in Eqs. (9.169)
– at the ends and midpoint of the interval – might be insufficient. This is usually
a minor drawback, because the grid of a well-constructed numerical model is in
general sufficiently close for the physical quantities to show very smooth point-to-
point variations. In such cases an integration grid everywhere equal to (or finer
than) the model grid may be adequate.2
The second reason for keeping the integration intervals small becomes evident by
fully expanding Eq. (9.168) and by writing it in the form
Ik = pk + Πk Ik+1 , (9.170)
where
Ik = I(τk ) , Ik+1 = I(τk+1 ) .
1
One generally takes the Unno-Rachkovsky solutions resulting from the assumption that, for
τr > τb , the propagation matrix is constant and the source functions (coincident with the Planck
function at large optical depth) are linear in τr .
2
Note that this is a very qualitative statement. To the authors’ knowledge, a systematic
analysis of the influence of the τr -variation of A and b on the numerical accuracy of the Runge-
Kutta integration has never been carried out.
442 CHAPTER 9
We obtain
1
pk = δ b(τk+1 ) + 4 b(τ̃k ) + b(τk )
6 k
1
− δk2 A(τ̃k ) b(τk+1 ) + A(τ̃k ) b(τ̃k ) + A(τk ) b(τ̃k )
6
1 3
+ δ A2 (τ̃k ) b(τk+1 ) + A(τk ) A(τ̃k ) b(τ̃k )
12 k
1 4
− δ A(τk ) A2 (τ̃k ) b(τk+1 ) ,
24 k
1
Πk = 1 − δ A(τk+1 ) + 4 A(τ̃k ) + A(τk )
6 k
1 2
+ δ A(τ̃k ) A(τk+1 ) + A2 (τ̃k ) + A(τk ) A(τ̃k )
6 k
1 3
− δ A2 (τ̃k ) A(τk+1 ) + A(τk ) A2 (τ̃k )
12 k
1 4
+ δ A(τk ) A2 (τ̃k ) A(τk+1 ) , (9.171)
24 k
where δk is the length of the (k + 1)-th interval
δk = τk+1 − τk .
Comparison of Eqs. (9.170) and (9.165) shows that the matrix Πk provides a
numerical representation of the evolution operator associated with the interval
(τk , τk+1 ),
Πk ≈ O(τk , τk+1 ) .
To test the accuracy of the Runge-Kutta integration against the step length, we
can thus adopt the following criterion: assume the propagation matrix A to be
constant in the interval (τk , τk+1 ) and compare the exact analytical expression of
the evolution operator with its numerical representation Πk . Equations (9.103)
and (9.171) yield
−δk A
O(τk , τk+1 ) = e (9.172)
1 2 2 1 3 3 1 4 4
Π k = 1 − δk A + δ A − δ A + δ A . (9.173)
2! k 3! k 4! k
It follows that the 4-th order Runge-Kutta method gives a Taylor expansion of the
analytical result which is correct up to fourth-order terms in the quantity δk A .
Consider now the eigenvalues of the matrix A. From Eqs. (8.33) and (9.164) we
have
λ1 = κr + κR hI + Λ+ (h, r )
λ2 = κr + κR hI − Λ+ (h, r )
λ3 = κr + κR hI + i Λ− (h, r )
λ4 = κr + κR hI − i Λ− (h, r ) , (9.174)
LINE FORMATION IN A MAGNETIC FIELD 443
where Λ+ (h, r ) and Λ− (h, r ) are given by Eqs. (A5.18). Since each of the quantities
κr , κR , hI , Λ+ and Λ− is non-negative, the modulus of the eigenvalue having the
largest modulus (that will be simply referred to as maximum eigenvalue) is
λmax = max λ1 , |λ3 | = max λ1 , |λ4 |
& 2 2 '
= max κr + κR hI + Λ+ (h, r ) , κr + κR hI + κR Λ− (h, r ) . (9.175)
Similarly, the minimum eigenvalue is
λmin = λ2 = κr + κR hI − Λ+ (h, r ) . (9.176)
Comparison of Eqs. (9.172) and (9.173) suggests the following criterion to choose
the length of the integration step. The error made in substituting the series ex-
pansion (9.173) for the exact expression (9.172) is of order (δk λmax )5 / 120 . If we
want to keep this error less than an assigned small number , we must require
(120 )1/5
δk ≤ , (9.177)
λmax
which shows how the step length should be controlled by the local value of the
maximum eigenvalue.
As far as the choice of τb is concerned, we must make sure that the solution
does not ‘feel’ possible inexactitudes in the boundary condition. Bearing in mind
Eq. (9.165), this leads to a constraint on the evolution operator O(τmin , τb ). This
constraint can be expressed in the form1
: τb
− λ (τ ) dτr
e τmin min r ≤
or
τb
λmin (τr ) dτr ≥ − ln , (9.178)
τmin
where is a given small number.
It follows from Eqs. (9.177) and (9.178) that the total number of steps, N , is
determined essentially by the ratio
λmax
= .
λmin
Using the Cauchy mean-value theorem we have in fact
τb τb τb
1 λmax (τr )
N≈ dτ ≥ dτ ≈ λmin (τr ) dτr
δk r (120 )1/5 r (120 )1/5
τmin τmin τmin
ln
≥ − 1/5
. (9.179)
(120 )
1
The quantity λmin is always positive (see footnote on p.421).
444 CHAPTER 9
Setting for instance = = 10−3 , we have
N ≥ 10.6 ,
and it is easily seen that this number may be very large at certain wavelengths,
especially for strong lines ( κR κr ). This is the main disadvantage of the Runge-
Kutta method.
Beside being used to solve the radiative transfer equation, this method can also
be used to compute numerically the evolution operator. In particular, Grossmann-
Doerth et al. (1988a) and Sánchez Almeida (1992) have applied the 4-th order
Runge-Kutta technique to evaluate the reduced evolution operator (see Sect. 8.9
and App. 7).
Another numerical technique that has been used to solve Eq. (9.163) is the so-
called Feautrier method. This is the direct generalization to the polarized case of
the method proposed by Feautrier (1964) to solve the ‘scalar’ transfer equation
for the intensity. It has been developed by Auer et al. (1977a) in the simplified
case where magneto-optical effects are neglected and later generalized by Rees
and Murphy (1987) to include these effects. The approach consists in writing the
transfer equation as a second-order differential equation with boundary conditions
at both ends of the integration interval.
Let us suppose the model atmosphere is given, and let us recall the property of
the radiative transfer equation considered at the end of Sect. 9.5 and summarized
by Eq. (9.61). Denoting by Ω an arbitrary outward direction, and by −Ω the
opposite (inward ) direction, we have from Eq. (9.163)
d
I(τr , ∆λ) = A(τr , ∆λ) I(τr , ∆λ) − b(τr , ∆λ)
dτr
d
− Ĩ (τr , −∆λ) = A(τr , ∆λ) Ĩ (τr , −∆λ) − b(τr , ∆λ) , (9.180)
dτr
where the vectors I = (I, Q, U, V )† and Ĩ = (I , Q , −U , V )† refer to the radi-
and −Ω
ation travelling in the directions Ω respectively, ∆λ is given by Eq. (9.59),
and the reference direction unit vectors are as in Fig. 9.4. Introducing the formal
vectors
1
F (τr , ∆λ) = I(τr , ∆λ) − Ĩ (τr , −∆λ)
2
1
G(τr , ∆λ) = I(τr , ∆λ) + Ĩ (τr , −∆λ) , (9.181)
2
and omitting the arguments of the various functions for conciseness, we have using
Eqs. (9.180)
dF dG
= AG − b , = AF , (9.182)
dτr dτr
LINE FORMATION IN A MAGNETIC FIELD 445
whence the following second-order differential equation for G is obtained
d dG
A−1 = AG − b . (9.183)
dτr dτr
We must now specify the boundary conditions at τr = 0 and τr = τb . At τr = 0, for
an atmosphere not illuminated from above (the case to which we restrict attention),
we have
I = Q = U = V = 0 , (9.184)
so that Eqs. (9.181) yield F = G . Using the second of Eqs. (9.182) we get
dG
A−1 = G (τr = 0) . (9.185)
dτr
At τr = τb we take I = I (b) , where I (b) is for instance the Unno-Rachkovsky
solution obtained as specified in footnote 1 on p. 441. Use of Eqs. (9.182) and
(9.181) yields
dG
A−1 = − G + I (b) (τr = τb ) . (9.186)
dτr
The numerical solution of Eq. (9.183), subjected to the boundary conditions
(9.185) and (9.186), can be carried out by standard methods of numerical anal-
ysis which imply the solution of a tridiagonal system of linear equations. The
mathematical details are developed in App. 8 for the case – considered by Auer et
al., 1977a, and Rees and Murphy, 1987 – where derivatives are replaced by differ-
ence formulae having second-order accuracy. The results of App. 8 show that the
Feautrier solution (obtained with such difference formulae) is equivalent, from the
point of view of numerical accuracy, to a second-order Runge-Kutta. This obvi-
ously implies that the length of the integration step should be controlled by the
local value of the ‘maximum eigenvalue’.
The same arguments developed for the Runge-Kutta solution show that Eqs.
(9.177) and (9.179) must be replaced by
(6 )1/3 ln
δk ≤ , N≥ − ,
λmax (6 )1/3
respectively. Note that for = = 10−3 the latter gives N ≥ 38.0 , or a
number of steps almost 4 times larger than in the 4-th order Runge-Kutta solution.
Difference formulae having higher-order accuracy can in principle be used within
the Feautrier method. References for the solution of the ‘scalar’ transfer equation
are given by Mihalas (1978, p.155).
Another method that has been proposed for the numerical solution of Eq. (9.163)
is the so-called diagonal element lambda-operator (or DELO ) method (Rees and
Murphy, 1987; Rees et al., 1989). The transfer equation is reformulated by taking
446 CHAPTER 9
advantage of the fact that the four diagonal elements of the propagation matrix A
are equal. This enables Eq. (9.163) to be cast into the form
d & '
I(τr ) = aI (τr ) 1 + A (τr ) I(τr ) − b (τr ) , (9.187)
dτr
where
aI = κ r + κ R h I
A(τr ) κ [ H − hI 1 ]
A (τr ) = −1 = R
aI (τr ) κr + κR hI
b(τr ) [ κ r Sc 1 + κ R SL H ] U
b (τr ) = = . (9.188)
aI (τr ) κr + κR hI
Equation (9.187) can be rewritten in the form
d
I(τr ) = aI (τr ) I(τr ) − s(τr ) , (9.189)
dτr
where
s(τr ) = b (τr ) − A (τr ) I(τr ) (9.190)
is a ‘modified source vector’ which is a function of the Stokes vector itself.
The formal solution to Eq. (9.189)
∞ : τr
− aI (τr ) dτr
I(τr ) = e τr aI (τr ) s(τr ) dτr (9.191)
τr
is an integral Volterra equation for I(τr ) . It can be solved via an iterative proce-
dure, called lambda-iteration,1 as done by Staude (1969). The numerical method
proposed by Rees and Murphy (1987), whose details are developed in App. 9, has
the advantage of avoiding the slow convergence of the lambda-iteration.
The results contained in App. 9 show that the integration grid must be locally
controlled also for the DELO method. The criterion to choose the step length
1
The concepts of lambda-operator and lambda-iteration are basic tools of the theory of
(scalar) radiative transfer (see e.g. Mihalas, 1978). Recalling Eq. (9.190), Eq. (9.191) can be
written in the form
I(τr ) = Λ( I ; τr ) + Λ (b ; τr ) ,
where Λ and Λ are integral operators acting, respectively, on the vectors I (unknown) and b
(known). The usual lambda-iteration consists in starting with the zero-order solution I (0) (τr ) = 0
and evaluating repeatedly the left-hand side of the equation by substituting in the right-hand side
the value I (k) (τr ) obtained in the previous iteration. The procedure is stopped when convergence
is reached. Lambda-iterations generally converge very slowly.
LINE FORMATION IN A MAGNETIC FIELD 447
is suggested by Eq. (A9.9). The eigenvalues of the matrix (A − aI 1 ) are (cf.
Eqs. (9.164), (9.188) and (8.33))
λ1 = κR Λ+ (h, r ) λ3 = i κR Λ− (h, r )
λ2 = −κR Λ+ (h, r ) λ4 = −i κR Λ− (h, r ) .
Defining
λmax = max κR Λ+ (h, r ) , κR Λ− (h, r ) , (9.192)
the error introduced at each step by the DELO integration is of order
2
δk3 3 a
λmax 1+ I .
12 λmax
If we want to keep this error less than , we must require
(12 )1/3
δk ≤ 2/3 . (9.193)
a
λmax 1 + I
λmax
The number of steps necessary for the integration is therefore (cf. the derivation
of Eq. (9.179))
ln
N ≥ − ,
(12 )1/3
where 2/3
λ aI
= max 1+ ,
λmin λmax
with λmin given by Eq. (9.176).
The performance of the DELO method against the Runge-Kutta method can
be examined by comparing the corresponding step lengths. Taking for instance
= 10−3 , we have from Eqs. (9.177) and (9.193)
1/3 2/3
(δk )R.K. λmax aI + λmax
≈ 2.9 ,
(δk )DELO λmax
with λmax given by Eq. (9.175). If the propagation matrix A is ‘almost diagonal’,
we have from Eqs. (9.174), (9.188) and (9.192)
λmax aI , λmax ≈ aI ,
thus the DELO method is more efficient. When λmax and aI are comparable (which
is often the case in most practical applications), it is difficult to establish which
method is more efficient.
448 CHAPTER 9
The last integration method that we discuss here is based on the direct application
of the concept of evolution operator. The method, already presented in a slightly
different form in Landi Degl’Innocenti (1987), will be referred to as the piecemeal
evolution operator (or PEO ) method . It consists in the following.
We introduce a grid of N points (τ1 , τ2 , . . . , τN ) not necessarily evenly spaced.
In each interval we assume that the matrix H is constant and equal to the average
of the values at the end points of the interval,
1
Hk = H(τk ) + H(τk+1 ) . (9.194)
2
According to Eqs. (9.164) and (9.104), the evolution operator associated with the
interval (τk , τk+1 ) is
−Kr (τk ,τk+1 ) −KR (τk ,τk+1 ) H k
O(τk , τk+1 ) ≡ Ok = e e (k = 1, 2, . . . , N − 1) ,
where
τ τ
Kr (τ, τ ) = κr (τr ) dτr , KR (τ, τ ) = κR (τr ) dτr , (9.195)
τ τ
and where the exponential of the matrix can be conveniently expressed via Eqs.
(A5.20) or (A5.23).
Iterative application of Eq. (9.165) enables the emerging Stokes vector to be
written in the form
I(τ1 ) = W1 + O1 W2 + O2 W3 + O3 . . . WN −1 + ON −1 WN . . .
or
N
I(τ1 ) = W1 + O1 O2 · · · Ok−1 Wk ,
k=2
where, using Eqs. (9.164)
τk+1
Wk = O(τk , τr ) b(τr ) dτr
τk
τk+1
−Kr (τk ,τr ) −KR (τk ,τr ) H k
= e e
τk
× κr (τr ) Sc (τr ) 1 + κR (τr ) SL (τr ) H k U dτr (9.196)
for k = 1, 2, . . . , N − 1 , and
WN = I(τN ) = I (b) ,
LINE FORMATION IN A MAGNETIC FIELD 449
with I (b) the boundary Stokes vector (see footnote 1 on p. 441).
The run with τr of the quantities κr , κR , Sc and SL is deduced from the at-
mospheric model at the grid points where the model is specified. Values in inter-
mediate points can be computed by usual interpolation methods. The integrals in
Eqs. (9.195) and (9.196) can then be evaluated by standard numerical techniques.
Obviously, for a hypothetic atmosphere where the matrix H is constant, the
PEO method provides the exact solution to the transfer equation irrespective of
the number and length of the integration steps. It is also obvious that the criterion
used to evaluate the accuracy of the preceding methods (comparison between the
numerical and analytical expressions of the evolution operator) is meaningless for
the PEO method. What is important for this method is not the local value of the
maximum eigenvalue, but the rate of variation of the matrix H with optical depth.
For instance, if H has a smooth variation in a certain interval, a relatively large
integration step can be chosen even if the maximum eigenvalue of H is ‘large’.
It should be pointed out that the accuracy of the PEO method can be improved
through an algorithm based on the concept of perturbative solution introduced in
Sect. 8.7. The solution previously obtained – with the matrix H assumed constant
in each interval – can be considered as the ‘zero-order solution’. The ‘perturbation’
consists in assuming the matrix H linear in τr in the interval (τk , τk+1 ) ,
2τr − τk+1 − τk
δH(τr ) = H(τr )−H k = H(τk+1 )−H(τk ) (τk ≤ τr ≤ τk+1 ) ,
2 (τk+1 − τk )
where Eq. (9.194) has been used. Calculations similar to those in Sect. 8.7 show
that the correction to the zero-order solution can be written in the form
∞
δI(τ1 ) = O0 (τ1 , τr ) κR (τr ) δH(τr ) SL (τr ) U − I0 (τr ) dτr ,
τ1
where O0 and I0 are the evolution operator and the solution to the zero-order
equation, respectively. The integral in the right-hand side is rather involved, but
it can be easily performed under suitable assumptions. Since δI(τ1 ) is in general
a small correction to I0 (τ1 ) , even a crude approximation is justified. An adequate
approximation is to consider the ‘effective source vector’ [SL (τr )U − I0 (τr )] as a
linear function of τr in each interval (τk , τk+1 ) . This allows the integral to be
evaluated by standard numerical techniques.
It should be mentioned that a numerical method quite similar to the PEO method
has been presented by Van Ballegooijen (1987) to solve the transfer equation in the
form of Eq. (8.65).
9.16. Response Functions
The concept of response function is particularly important in the problem of line
formation in a magnetic field since it clarifies the role played by the different phys-
450 CHAPTER 9
ical parameters in determining the characteristics of the Stokes parameters pro-
files emerging from a magnetized stellar atmosphere. The concept was introduced
by Beckers and Milkey (1975) – see also Mein, 1971 – for non-magnetic lines,
and was later generalized to magnetic lines by Landi Degl’Innocenti and Landi
Degl’Innocenti (1977). Although response functions may in principle be defined
also for non-LTE lines, the following discussion will be restricted to the simpler
case of LTE.
Let us consider the transfer equation in the form (9.161) or (9.163). Because of
the LTE assumption, the emission vector defined in Eqs. (9.164) reduces to
b(τr ) = A(τr ) BP (τr ) U ,
with BP (τr ) the Planck function, and the transfer equation becomes
d
I(τr ) = A(τr ) I(τr ) − BP (τr ) U , (9.197)
dτr
where A(τr ) is given by Eqs. (9.164). For a semi-infinite atmosphere, the formal
solution to this equation is (cf. Eq. (9.165))
∞
I(τr ) = O(τr , τr ) A(τr ) BP (τr ) U dτr , (9.198)
τr
with the evolution operator O(τr , τr ) given by Eq. (9.167).
We now observe that the quantities A(τr ) and BP (τr ) depend on the local1 value
of a set of physical parameters {ζi }, where i = 1, 2, . . . , p . These parameters
might in principle be divided in two groups, the independent and the dependent
parameters. If the chemical composition of the atmosphere is fixed, the independent
parameters are only the temperature, the gas pressure, the magnetic field vector
and the line-of-sight component of the velocity field, the last resulting both from
the microturbulent velocity and from the bulk velocity of the ambient medium.
Otherwise, the element abundances play also the role of independent parameters.
The dependent parameters are quantities (like the Doppler width, the damping
constant, the continuum and line absorption coefficients) which depend on the
parameters of the first group. In other words, they can be evaluated (by a simple
formula or a series of calculations) once the independent parameters are known.
In the following we just assume that the set {ζi } contains a number p of physical
parameters that are independent of each other . A set composed of the magnetic
field intensity and the Doppler width is a particular example of a set with p = 2 .
Given a stellar atmosphere where the p physical parameters ζi vary with optical
depth, we consider a ‘perturbed’ atmosphere characterized by the parameters
ζi (τr ) = ζi (τr ) + δζi (τr ) (i = 1, 2, . . . , p) .
1
In the non-LTE case, the source functions Sc (τr ) and SL (τr ) have a non-local dependence
on the physical parameters. This makes difficult the generalization of the concept of response
function to that case.
LINE FORMATION IN A MAGNETIC FIELD 451
If the ‘increments’ δζi (τr ) are small, we can express the propagation matrix and
the Planck function by a first-order expansion
p
∂A
A (τr ) = A(τr ) + δA(τr ) = A(τr ) + δζ (τ )
i=1
∂ζi τ i r
r
p
∂BP
BP (τr ) = BP (τr ) + δBP (τr ) = BP (τr ) + δζi (τr ) . (9.199)
i=1
∂ζi τr
Some remarks are necessary here. A variation of the physical parameters entails, in
general, a variation of the continuum absorption coefficient. Therefore, the optical-
depth scales in the unperturbed and perturbed atmosphere are different. In the
following, the optical-depth scale of the unperturbed atmosphere will always be
used. This implies that the partial derivatives in Eqs. (9.199) should be performed
by keeping constant the value of the absorption coefficient kc (νr ) which enters
the definitions of κr and κR (see Eqs. (9.162)). The second remark concerns the
expression of δBP in Eqs. (9.199). Since BP depends only on temperature, δBP is
identically zero if the temperature is not contained in the set {ζi } .
Substituting Eqs. (9.199) into Eq. (9.197), writing the Stokes parameters for the
perturbed atmosphere in the form
I (τr ) = I(τr ) + δI(τr ) , (9.200)
and considering only the first-order terms in the resulting equation, we get for δI
the transfer equation
p
d ∂BP
δI(τr ) = A(τr ) δI(τr ) − A(τr ) δζ (τ ) U
dτr i=1
∂ζi τ i r
r
p
!
∂A
− δζi (τr ) I(τr ) − BP (τr ) U .
i=1
∂ζi τr
The correction to the Stokes parameters emerging from the perturbed atmosphere
can thus be written in the form
p ∞
∂BP
δI(0) = O(0, τr ) A(τr ) U
i=1
∂ζi τ
0 r
!
∂A
− I(τr ) − BP (τr ) U δζi (τr ) dτr , (9.201)
∂ζi τr
where I(τr ) is given by Eq. (9.198).
The concept of response function is suggested quite naturally by Eq. (9.201): the
response function for the physical parameter ζi , RF (ζi ; τr ), is defined to be the
i-th term in the integrand of Eq. (9.201),
!
∂BP ∂A
RF (ζi ; τr ) = O(0, τr ) A(τr ) U − I(τr ) − BP (τr ) U . (9.202)
∂ζi τ ∂ζi τ
r r
452 CHAPTER 9
Obviously, it depends on wavelength and on the model atmosphere. Using this
definition, Eq. (9.201) reads
p ∞
δI(0) = RF (ζi ; τr ) δζi (τr ) dτr . (9.203)
i=1 0
The physical meaning of response functions follows directly from Eq. (9.203). If
∆τr is a small optical-depth interval centered at τr , and δζi is a small variation of
the physical parameter ζi in that interval, one gets
δI(0)
RF (ζi ; τr ) = ,
δζi ∆τr
which shows that response functions can be regarded as a sort of partial derivatives
of the emerging Stokes parameters with respect to the physical parameters (Ruiz
Cobo and del Toro Iniesta, 1994). Equation (9.203) also gives an ‘operational def-
inition’ of response functions: in principle, they could be measured by changing
the value of a physical parameter in a small region of a stellar atmosphere and
by observing the corresponding variation of the emerging Stokes parameters. Al-
though this is obviously a Gedankenexperiment , it shows that response functions
are correctly defined from the physical point of view.
Response functions satisfy the important integral property
∞
∂I(0)
RF (ζi ; τr ) dτr = ,
∂ζi
0
which follows directly from Eq. (9.203) by considering an increment δζi indepen-
dent of optical depth.
Equation (9.202) shows that response functions can be expressed in analytical
form whenever the same property holds for the evolution operator. As an exam-
ple we can consider the case of the Milne-Eddington atmosphere that has been
developed in Sect. 9.8.
First of all, it is convenient to rewrite Eq. (9.202) in terms of the optical depth τc
rather than τr , so that the expressions derived in Sect. 9.8 can be applied. Starting
at Eq. (9.203) we obtain, with the help of Eqs. (9.33), (9.162), (9.164) and (9.31)1
!
∂BP ∂C
RF (ζi ; τc ) = O(0, τc ) C(τc ) U − I(τc ) − BP (τc ) U ,
∂ζi τ ∂ζi τ
c c
where C(τc ) is defined in Eq. (9.91). In the Milne-Eddington atmosphere C is
constant, the evolution operator is given by Eqs. (9.103), BP by Eq. (9.105) and
1
Use of τc instead of τr as independent variable makes more complicated the formal treatment
of response functions for physical variables affecting the continuum absorption coefficient, since
in that case the ratio κr defined in Eqs. (9.162) is unity by definition. This is the reason why our
general discussion of response functions is based on τr .
LINE FORMATION IN A MAGNETIC FIELD 453
I(tc , µ), the Stokes parameters of the radiation flowing along the µ-direction at
optical depth tc = µ τc , by Eqs. (9.108). The response function for any physical
parameter ζi except temperature1 is therefore
t
− µc C ∂C
RF (ζi ; tc , µ) = −B0 βµ e C −1 U ,
∂ζi
and the correction to the emerging Stokes parameters due to the ‘perturbations’
δζi (tc ) is
p ∞
dt
δI(0, µ) = RF (ζi ; tc , µ) δζi (tc ) c .
i=1
µ
0
Explicit expressions for δI(0, µ) can be derived for different forms of the func-
tions δζi (tc ). Assuming, for instance, a linear behavior of the perturbation (Lan-
dolfi, 1987)
δζi (tc ) = ai + bi tc ,
we obtain with the help of Eqs. (9.106) and (9.107)
p
∂C −1 2 ∂C −1
δI(0, µ) = −B0 βµ ai C −1 C + µ bi C −1 C U,
i=1
∂ζi ∂ζi
or, observing that
∂ C −1 ∂C −1
= − C −1 C , (9.204)
∂ζi ∂ζi
p
∂ C −1
δI(0, µ) = B0 βµ ai 1 + µ bi C −1 U. (9.205)
i=1
∂ζi
Bearing in mind the expression of the ‘unperturbed’ emerging Stokes parameters
(Eqs. (9.108)), we can also write
p
∂
δI(0, µ) = ai 1 + µ bi C −1 I(0, µ) .
i=1
∂ζi
Similarly, assuming an exponential behavior
−αi tc
δζi (tc ) = Ai e ,
we get
p
−1 ∂C −1
δI(0, µ) = −B0 βµ Ai C + µ αi 1 C U,
i=1
∂ζi
1
In a Milne-Eddington atmosphere the partial derivative ∂BP / ∂T is meaningless because
of the parameterization adopted for BP .
454 CHAPTER 9
or, in a different form
p
−1 ∂ C −1
δI(0, µ) = B0 βµ Ai 1 − µ αi C + µ αi 1 U
i=1
∂ζi
p
−1 ∂
= Ai 1 − µ αi C + µ αi 1 I(0, µ) .
i=1
∂ζi
If the unperturbed atmosphere is static – or, more generally, in the ‘constant
velocity case’ defined in Sect. 9.5 – the response functions for a given physical pa-
rameter evaluated at two wavelengths symmetrical about line center are related by
interesting symmetry properties, which follow directly from Eqs. (9.202). Defining
∆λ as in Eq. (9.45), the propagation matrix A and the Stokes vector I satisfy, at
any optical depth, the relations (cf. Eqs. (9.51) and (9.54))
A(τr ; −∆λ) = X A(τr ; ∆λ) X −1
I(τr ; −∆λ) = X I(τr ; ∆λ) , (9.206)
where the matrix X is defined in Eq. (9.52). On the other hand, the series expan-
sion of the evolution operator given in Eq. (9.167) shows that
O(0, τr ; −∆λ) = X O(0, τr ; ∆λ) X −1 . (9.207)
The relation between the matrices (∂A / ∂ζi )τr evaluated at ∆λ and −∆λ depends
on the physical parameter ζi . It is necessary to divide the physical parameters in
two groups, (a) and (b). Group (a) includes temperature, pressure, magnetic field
(B, θ, χ), Doppler width and damping constant: in other words, all the parameters
except the line-of-sight component wA of the bulk velocity. Group (b) contains wA
alone. It can be proved from the explicit expression of A (see Eqs. (9.164), (9.162),
(9.50), (9.32), (9.27), (9.29) and (9.26)) that
∂ ∂
A(−∆λ) = X A(∆λ) X −1
∂ζa τr ∂ζa τr
∂ ∂
A(−∆λ) = −X A(∆λ) X −1 , (9.208)
∂ζb τr ∂ζb τr
where ζa is any physical parameter of group (a), and ζb = wA . Using Eqs. (9.206)-
(9.208), and observing that
∂BP
= 0, X −1 U = U ,
∂ζb τ
r
one easily obtains from Eq. (9.202)
RF (ζa ; τr , −∆λ) = X RF (ζa ; τr , ∆λ)
RF (ζb ; τr , −∆λ) = − X RF (ζb ; τr , ∆λ) . (9.209)
LINE FORMATION IN A MAGNETIC FIELD 455
According to Eq. (9.203), the corrections δI(0, ∆λ) and δI(0, −∆λ) satisfy sym-
metry relations of the same form as Eqs. (9.209). It follows that only a variation
of the physical parameter wA with optical depth can alter the symmetry char-
acteristics of the Stokes parameters expressed by Eqs. (9.54).1 The (first-order)
corrections produced by such variation have the opposite symmetry of the corre-
sponding Stokes parameters,
δI(0, −∆λ) = − δI(0, ∆λ)
δQ(0, −∆λ) = − δQ(0, ∆λ)
δU (0, −∆λ) = − δU (0, ∆λ)
δV (0, −∆λ) = δV (0, ∆λ) .
Finally, we want to remark that the concept of response function can be easily
generalized to any observable quantity obtained by a mathematical transformation
on the emerging Stokes parameters I(0, λ). As an example, let us consider the
line-integrated Stokes parameters defined by2
I = (I, Q , U, V)† = I(0, λ) − Ic (0) U dλ , (9.210)
line
where Ic (0) is the emerging intensity in the continuum adjacent to the line ( Ic (0) U
= I(0, λc ) , with λc a wavelength sufficiently far in the line wing), and the integral
is over the line profile. Following our perturbative scheme, we have from Eq. (9.203)
p ∞ !
δI = RF (ζi ; τr , λ) − RF (ζi ; τr , λc ) δζi (τr ) dτr dλ .
line i=1 0
Inverting the order of the integrals, this equation can be cast into the form
p ∞
δI = RF (ζi ; τr ) δζi (τr ) dτr ,
i=1 0
where the line-integrated response function is defined by
RF (ζi ; τr ) = RF (ζi ; τr , λ) − RF (ζi ; τr , λc ) dλ .
line
Further examples of ‘generalized’ response functions can be found in Landi Degl’In-
nocenti and Landolfi (1983), Grossmann-Doerth et al. (1988a), Ruiz Cobo and
1
These characteristics can also be altered by fine or hyperfine structure effects (see Sect. 9.23),
or by atomic polarization effects (see Sect. 10.5).
2
Note that I is proportional to the line equivalent width W . In fact I = −Ic (0) W .
456 CHAPTER 9
del Toro Iniesta (1994). In this last paper the concept is precisely stated from the
mathematical point of view.
9.17. Contribution Functions
Contribution functions are a rather controversial subject in the theory of radiative
transfer, both in the ‘classical’ formulation that neglects polarization phenomena
and, even more, in the case of polarized radiation. The underlying idea is to find an
expression, defined in terms of opacities and source functions, whose dependence on
optical depth should represent the contribution of the different atmospheric layers
to the emerging intensity, or to the line depression, or to some other observable
spectral feature.
Although this idea seems rather simple and intuitive, difficulties usually arise
when trying to express it into a quantitative form. This is demonstrated by the
large number of papers dealing with the subject and by the even larger number
of contribution functions proposed in the literature (see e.g. Caccin et al., 1977;
Magain, 1986; Achmad et al., 1991, and references therein). Moreover, most of the
work has been devoted to contribution functions for the scalar case (no magnetic
field), where physical intuition can be more helpful in checking the soundness of
the definitions: contribution functions for polarized radiation are usually obtained
by direct generalization of the corresponding scalar expressions. Without entering
into the details of the controversy, we limit ourselves to give a few examples of
contribution functions for the Stokes parameters.
Equation (9.166), rewritten in the form
∞
I(0) = CF (τr ) dτr , (9.211)
0
provides the possibility of defining a contribution function for the Stokes parameters
in the form (Van Ballegooijen, 1985)
CF (τr ) = O(0, τr ) κr Sc 1 + κR SL H U .
This contribution function has a very intuitive physical meaning, but has the draw-
back of mixing the contributions of the continuum and of the spectral line. To over-
come this difficulty one can define the contribution function for the line depression
[Ic (0) U − I(0)], or the contribution function for the normalized line depression 1
defined by
I (0) U − I(0) I(0)
D(0) = c =U − . (9.212)
Ic (0) Ic (0)
1
The term ‘line depression’ is applied here to the four Stokes parameters, although, more
properly, it should only be applied to the intensity.
LINE FORMATION IN A MAGNETIC FIELD 457
Let us consider the second possibility. According to Magain (1986), a correct
definition of the contribution function requires the following procedure: one should
write a transfer equation for the quantity D(τr ) , find a formal solution to the
equation, and identify the integrand with the contribution function itself. From
Eq. (9.163) and from the transfer equation for the continuum intensity
dIc
= κr Ic − Sc (9.213)
dτr
one obtains, with the use of Eqs. (9.164), the transfer equation
dD S S
= κr c 1 + κR H D − κR 1 − L H U .
dτr Ic Ic
This is formally similar to any of the transfer equations considered previously,
with the only difference that, in the present case, the propagation matrix and the
emission vector depend also on the local value of the continuum intensity. Defining
Sc Sc
AD = κr 1 + κR H = A − κr 1 − 1 ,
Ic Ic
with A given by Eqs. (9.164), the value of AD at any optical depth can be evaluated
using the solution to Eq. (9.213). We can then calculate the evolution operator OD
corresponding to the propagation matrix AD , and finally write
∞
D(0) = CF (τr ) dτr , (9.214)
0
where CF (τr ) , the contribution function to the normalized line depression, is defined
by (Grossmann-Doerth et al., 1988a; Rees et al., 1989)
S
CF (τr ) = OD (0, τr ) κR 1 − L H U .
Ic
The main drawback of contribution functions is indeed implicit in their definition.
As apparent from Eqs. (9.211) or (9.214), these functions are invariant under a kind
of ‘gauge transformation’ of the form
CF (τr ) → CF (τr ) + g(τr ) ,
where g(τr ) is an arbitrary function such that
∞
g(τr ) dτr = 0 .
0
458 CHAPTER 9
The choice of the gauge is more or less arbitrary, although in some cases it can
be suggested by physical considerations. We believe that no real progress can be
achieved in this subject until a clear, operational definition of contribution function
is given.
A possible way out could be to replace the concept of contribution function by
an appropriate response function, like the response function to the logarithm of
the line oscillator strength, defined by (see Eq. (9.203))1
∞
δI(0) = RF ( lnf ; τr ) δ lnf (τr ) dτr .
0
According to Eq. (9.212), the correction to the normalized line depression produced
by the ‘perturbation’ δ lnf (τr ) is given by
δI(0)
δD(0) = − ,
Ic (0)
because the continuum intensity is unaffected by a variation of the line opacity. If
we introduce a (generalized) response function RF ( lnf ; τr ) such that
∞
δD(0) = RF ( lnf ; τr ) δ lnf (τr ) dτr ,
0
we have from the above equations
RF ( lnf ; τr )
RF ( lnf ; τr ) = − . (9.215)
Ic (0)
Since
∂BP
= 0
∂(lnf ) τ
r
and, using Eqs. (9.164), (9.162) and (9.19)
∂A
= κR (τr ) H(τr ) ,
∂(ln f ) τ
r
we obtain from Eqs. (9.202) and (9.215) the remarkable expression
1 & '
RF ( lnf ; τr ) = κR (τr ) O(0, τr ) H(τr ) I(τr ) − BP (τr ) U .
Ic (0)
1
We restrict ourselves to the LTE case since our definition of response function does not
apply to non-LTE lines – see the discussion in the previous section.
LINE FORMATION IN A MAGNETIC FIELD 459
Finally, it should be noticed that all the contribution functions defined in this
section, as well as the response functions defined in the former section, are referred
to the optical-depth scale τr . When a different depth indicator x is used, the new
contribution function is obtained through the obvious transformation
dτr
CF (x) = CF (τr ) ,
dx
with a similar relation for RF (x) . Assuming for instance x = Log τr , one has
CF (x) = ln(10) τr CF (τr ) .
The scale Log τr is generally convenient for plotting contribution and response
functions against optical depth.
9.18. Blends
Blends are a common feature of stellar spectra. They simply result from the fact
that, in a given section of the spectrum, two or more lines, generally belonging to
different elements, may overlap to give rise to a complicated spectral feature. The
most common case is the presence of a weak line in the blue or in the red wing of
a stronger line.
When the individual lines are independent , which means that they belong either
to different elements or to different multiplets of the same element, their cumulative
effect can be simply described by adding in the transfer equation the contributions
of each line. When the blending lines are the fine-structure or hyperfine-structure
components of a single ‘parent’ line, the situation is more complex and requires a
different approach (see Sect. 9.23).
(i)
For an ensemble of N independent lines, each characterized by its own κL (the
ratio between line and continuum absorption coefficient defined in Eq. (9.31)) and
(i)
SL (the line source function defined in Eq. (9.14)), Eq. (9.35) generalizes into the
following
⎛ ⎞ ⎛ ⎞⎛ ⎞
I 1 0 0 0 I − Sc
⎜ ⎟ ⎜
d ⎜Q⎟ ⎜0 1 0⎟ ⎜ ⎟
0 ⎟⎜ Q ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟
dτc ⎝ U ⎠ ⎝ 0 0 1 0 ⎠⎝ U ⎠
V 0 0 0 1 V
⎛ (i) ⎞⎛ (i) ⎞
hI hQ
(i) (i)
hU hV
(i)
I − SL
⎜ (i) ⎟⎜ ⎟
N ⎜ (i) (i)
−rU ⎟
(i) ⎜ Q ⎟
(i) ⎜ hQ hI rV ⎟⎜ ⎟
+ κL ⎜ ⎟ ⎜ ⎟ , (9.216)
⎜ h(i) −rV
(i) (i)
hI rQ ⎟
(i) ⎜ U ⎟
i=1 ⎝ U ⎠⎝ ⎠
(i) (i) (i) (i)
hV rU −rQ hI V
460 CHAPTER 9
or, in matrix form
dI N
(i)
N
(i) (i)
= 1 + κL H (i) I − Sc 1 + κL SL H (i) U , (9.217)
dτc i=1 i=1
(i) (i) (i) (i) (i) (i) (i)
where the elements hI , hQ , hU , hV , rQ , rU , rV of the matrix H (i) are defined
as in Eqs. (9.32). Obviously each line has its own central wavelength, Zeeman
pattern, Doppler width and damping constant.
Many of the results derived in this chapter for single spectral lines can be easily
generalized to blends. In particular, the Stokes parameters emerging from a Milne-
Eddington atmosphere are still given by Eq. (9.109), with the matrix C given by
N
(i)
C=1 + κL H (i) .
i=1
Figure 9.9 shows the emerging Stokes parameters, calculated according to Eq.
(9.109), for a blend of two lines. The line on the left is the normal Zeeman triplet
of Fig. 9.5 (a), the other one is weaker but has a larger sensitivity to the magnetic
field (normal triplet with g = 2). Panels (a) to (d) are obtained by reducing
the wavelength separation of the lines. The Stokes parameters are normalized to
B0 and are plotted against the reduced wavelength v = (λ − λ̄0 ) / ∆λD , where
λ̄0 = (λ(1) (2)
0 + λ0 ) / 2 and ∆λD is the (common) Doppler width of the two lines.
The figure shows quite clearly that blends break the symmetry characteristics of
Stokes profiles about line center. The ‘distortion’ is particularly evident in the Q
profile, panels (a) and (b).
In the quite common case of a weak line lying on the wing of a stronger line,
the emerging Stokes parameters can be obtained by a perturbative solution of the
transfer equation (9.217). If the index 1 refers to the stronger line and the index 2
to the weaker, one can first solve the equation
dI0
L H
= 1 + κ(1) I0 − L SL H
Sc 1 + κ(1) U
(1) (1) (1)
dτc
to obtain the ‘zero-order’ Stokes vector I0 (τc ) . The correction δI is then found by
solving the approximate equation
d
L H
δI = 1 + κ(1) δI − κ(2)
L H SL(2) U − I0 .
(1) (2)
dτc
L Hij κL Hij .
Obviously, such perturbative solution is justified only if κ(2) (2) (1) (1)
9.19. The Magnetic Intensification Mechanism
Let us consider a spectral line formed in the radiation flowing along a particular
direction in a given stellar atmosphere. If no magnetic field is present in the
LINE FORMATION IN A MAGNETIC FIELD 461
Fig.9.9. Stokes parameters profiles for a blend of two lines, computed for a Milne-Eddington
atmosphere. The profiles are normalized to B0 and are plotted against the reduced wavelength
defined in the text. The lines have different strength (κ(1) L
= 20, κ(2)
L
= 10) and magnetic
(1) (2)
sensitivity (normal Zeeman triplets with g = 1, g = 2 ). The values of the remaining
parameters are those of the ‘standard case’ (see Fig.9.5). The four panels correspond to different
wavelength separation of the lines, parameterized through ∆v = (λ(2) (1)
0 − λ0 ) /∆λD with ∆λD
the (common) Doppler width. Panel (a) ∆v = 3 ; (b) ∆v = 1.5 ; (c) ∆v = 0.5 ; (d) ∆v = 0.
atmosphere, the line will be characterized by a zero-field equivalent width W0
defined by
1
W0 = ¯
Ic (0) − I(0) dλ , (9.218)
Ic (0)
line
where Ic (0) is the emerging intensity in the nearby continuum and I(0) ¯ is the
zero-field, wavelength-dependent emerging intensity.
In a sort of Gedankenexperiment , let us now suppose a constant magnetic field
is switched on throughout the whole atmosphere, all the other physical parameters
remaining unchanged. The new atmosphere will be characterized by the same
continuum intensity1 but a different line intensity, and this will result in a different
1
See footnote on p.376.
462 CHAPTER 9
equivalent width for the line considered. Defining the new equivalent width as
1
W = Ic (0) − I(0) dλ , (9.219)
Ic (0)
line
where I(0) is the intensity emerging from the magnetic atmosphere, the influence of
the magnetic field on the equivalent width can be characterized by the dimensionless
parameter
W
M= , (9.220)
W0
that we will call the magnetic intensification parameter . In most stellar atmo-
spheres and for most spectral lines, this parameter is larger than 1 : in other
words, the equivalent width in a magnetic atmosphere is usually larger than the
equivalent width in the corresponding (in the sense specified above) non-magnetic
atmosphere.
This fact has a very simple physical explanation, which can be understood with
the help of the classical theory of the equivalent width developed, for instance, in
Aller (1963) or in Mihalas (1978). Consider a line in the flat part of the curve
of growth, where saturation phenomena are important. In the presence of a mag-
netic field the line splits into its Zeeman components and, as soon as the Zeeman
splitting becomes of the order of the Doppler width (∆λB ≈ ∆λD ), it broadens
– thus increasing its equivalent width – and desaturates. This desaturation process
increases with increasing magnetic field strength and reaches its maximum when
all the Zeeman components are well-separated (∆λB ∆λD ). In this process the
line intensifies, which justifies the name of magnetic intensification given to this
physical mechanism.
Ten Bruggencate and von Klüber (1939) were the first to suggest that such mech-
anism may be effective in sunspots. Further contributions to this subject were given
by Babcock (1949) and Boyarchuk et al. (1961).
Even in the simple case of an atmosphere with constant magnetic field, the mag-
netic intensification parameter defined in Eq. (9.220) depends on many quantities.
Besides the magnetic field intensity and direction, it depends on the spectral line
(the structure of the Zeeman pattern being of crucial importance), on the full
set of physical quantities specifying the properties of the atmosphere, and on the
direction of the emerging radiation.
We can obtain an approximate expression for M by assuming the atmosphere to
be described by the Milne-Eddington model. Although this may seem a very rough
approximation, it should be remarked that the equivalent width of a spectral line
has a rather weak dependence on the detailed structure of the atmosphere, and
this holds especially for the magnetic intensification parameter which is the ratio
of two equivalent widths. Recalling Eqs. (9.218)-(9.220) and Eq. (9.112), we obtain
∞&
'
1 − ∆−1 (1 + kI ) (1 + kI )2 + fQ 2
+ fU2 + fV2 dλ
M = −∞ ∞ & ' , (9.221)
−1
1 − (1 + κL η) dλ
−∞
LINE FORMATION IN A MAGNETIC FIELD 463
where ∆ is defined in Table 9.5, η is the normalized absorption profile for zero
magnetic field
1 λ − λ0
η = √ H ,a ,
π ∆λD
and the remaining quantities are given by Eqs. (9.31) and (9.39).
Despite the substantial simplification introduced by the Milne-Eddington model,
M still remains a complicated function of the Zeeman pattern and of the four
parameters1 vB , θ, κL , a . Only a few properties of M can be deduced analytically
from Eq. (9.221). These are the following:
a) Under the limit of weak, unsaturated lines, the magnetic intensification param-
eter is equal to 1, ∞
κL hI dλ
−∞
lim M = ∞ = 1.
κL →0
κL η dλ
−∞
In other words, the line-broadening mechanism due to the magnetic field is ineffec-
tive in this particular case. The proof is easily obtained by series expansion of the
integrands in Eq. (9.221) and use of the normalization (9.28).
b) Under the limit of weak magnetic field (∆λB ∆λD ), the numerator of
Eq. (9.221) can be expanded in power series of ∆λB with the help of Eqs. (9.69).
After some algebra we obtain, up to second-order terms
2
0 sin θ + G1 (1 + cos θ) I2 (κL , a) − G1
G(2) cos2 θ I3 (κL , a)
1 2 (2) 2 (1)
4
M = 1 + vB
2
,
I1 (κL , a)
where the integrals I1 , I2 , I3 are given by
∞
κL η
I1 (κL , a) = dv
1 + κL η
−∞
∞
κL η̈
I2 (κL , a) = dv
(1 + κL η)2
−∞
∞
κ2L η̇ 2
I3 (κL , a) = dv , (9.222)
(1 + κL η)3
−∞
with v the reduced wavelength defined in Eqs. (9.26) and
1
The dependence on χ , the azimuth angle of the magnetic field vector, disappears for obvious
symmetry reasons.
464 CHAPTER 9
∂η ∂η ∂2η ∂2η
η̇ = = ∆λD , η̈ = 2
= ∆λ2D . (9.223)
∂v ∂λ ∂v ∂λ2
Integration by parts of I2 (κL , a) shows that
I2 (κL , a) = 2 I3 (κL , a) , (9.224)
hence
I3 (κL , a)
M = 1 + vB
2
Γ (θ) , (9.225)
I1 (κL , a)
where
1 2
0 sin θ + G1 (1 + cos θ) − G1
G(2) 2 (2) 2 (1)
Γ (θ) = cos2 θ .
2
The function Γ (θ) can be transformed by substituting the expressions for the quan-
(n)
tities Gq contained in Table 3.4. With some easy algebra one gets
1 2
Γ (θ) = ḡ + δ sin2 θ + δ , (9.226)
2
where ḡ is the effective Landé factor defined in Eq. (3.44), δ is defined in Eq. (9.78)
and δ is given by
1 2
δ = g (8s − d2 − 12) . (9.227)
80 d
For electric-dipole transitions, the quantity δ is – like δ, see Sect. 9.6 – a non-
negative quantity that vanishes only for Zeeman triplets (and for transitions J =
1/2 → J = 1/2).
Since both integrals I1 and I3 are positive, substitution of Eq. (9.226) into
Eq. (9.225) shows that the following conclusions can be drawn about the mag-
netic intensification parameter in the weak field limit: i) it increases quadrati-
cally with the magnetic field strength; ii) it increases linearly with sin2 θ , reach-
ing its maximum value for transverse fields (θ = 90◦ ); iii) in a longitudinal field
(θ = 0◦ or 180◦ ), it is always equal to 1 for Zeeman triplets and for transitions
J = 1/2 → J = 1/2;1 iv) its dependence on κL and a is neither affected by the
magnetic field nor by the Zeeman pattern of the spectral line, and is fully described
by the function I3 / I1 . The functions I1 , I3 and the ratio I3 / I1 are plotted in
Fig. 9.10. Note that I1 is proportional to the equivalent width of a non-magnetic
line expressed in the Milne-Eddington approximation. From Eqs. (9.218), (9.112)
and (9.222) we have in fact
βµ
W0 (µ) = ∆λD I (κ , a) . (9.228)
1 + βµ 1 L
1
This property holds even if the magnetic field is not weak, as it can be easily realized from
the discussion presented in Sect. 9.10 (cf. Eqs. (9.126)).
LINE FORMATION IN A MAGNETIC FIELD 465
Fig.9.10. The functions I1 (κL , a) and I3 (κL , a) are plotted against κL on a Log-Log scale, for
different values of the damping constant: a = 10−3 (full line), a = 10−2 (short dash), a = 10−1
(long dash). Note that I1 (κL , a) represents the standard curve of growth for non-magnetic lines.
According to Eq.(9.225), the ratio I3 / I1 describes the dependence on κL and a of the magnetic
intensification parameter M.
c) Under the limit of very large magnetic field (∆λB ∆λD ), when the profiles
of the different Zeeman components are completely separated and do not overlap,
each component can be treated independently. Its contribution is easily obtained
with the help of the results derived in Sect. 9.12. Equations (9.220), (9.219), (9.228)
466 CHAPTER 9
and (9.146) yield, for the Milne-Eddington atmosphere
N N ∞ (i) (i)
1 (i) 1 κf η
M= RI (0, µ) dλ = dv ,
W0 (µ) i=1 2 I1 (κL , a) i=1 (i) (i)
1 + κf η
line −∞
(i)
where κf is given by Eq. (9.143), η (i) by Eqs. (9.140), and N is the total number
of Zeeman components. Recalling the definition of I1 (Eqs. (9.222)) and dividing
the Zeeman components into π and σ, we obtain, with the use of the expressions
in Table 9.6
Nπ
, - Nσ
, 2 -
I1 κL si sin2 θ, a + I1 κL si 1 + cos
2
θ ,a
1 i=1 i=1
M= , (9.229)
2 I1 (κL , a)
where si is the strength of the i-component defined in Eqs. (9.140) and the two
summations refer to π and σ components, respectively (Nπ + Nσ = N ). It follows
from Eq. (9.229) that, when κL is increased, the magnetic intensification parameter
tends to a maximum asymptotic value given by1 (Babcock, 1949)
Nπ + Nσ N
M= = ,
2 2
which clearly shows the importance of the Zeeman pattern in determining the value
of the magnetic intensification parameter.
All the properties discussed above are confirmed by detailed numerical calcula-
tions. Sample results are shown in Fig. 9.11.
9.20. Net Linear Polarization in Spectral Lines: the
Differential Saturation Mechanism
Saturation effects in magnetic spectral lines not only increase their equivalent width
(as shown in the former section) but also produce a net amount of linear polariza-
tion in the line.
It is easily seen that weak, unsaturated magnetic lines cannot produce net linear
polarization. According to the generalized Seares formulae (9.151), which are valid
for weak lines in an arbitrary atmosphere, the emerging Q Stokes parameter is
given by
∞ ∞
−τc −(τ − τ )
Q(0) = e κL (τc ) hQ (τc ) BP (τc ) − e c c BP (τc ) dτc dτc ,
0 τc
1
If θ is close to 0◦ or 180◦ , the ‘fictitious opacity’ in the π components (κL si sin2 θ) is small.
For such inclination angles the asymptotic value for M is Nσ / 2 instead of N / 2 . This shows
again that M = 1 for Zeeman triplets when the magnetic field is parallel or antiparallel to the
propagation direction (the magnetic intensification mechanism is ineffective).
LINE FORMATION IN A MAGNETIC FIELD 467
Fig.9.11. Numerical results for the magnetic intensification parameter M computed according to
Eq.(9.221). The solid line refers to a normal Zeeman triplet ( 1 S0 − 1 P1 ), while the dashed line
refers to a more complicated Zeeman pattern ( 5 P3 − 5 D4 , see Fig.3.2). Panel (a): M versus vB
for θ = 60◦ , κL = 20, a = 0.05. Panel (b): M versus θ for vB = 1.5, κL = 20, a = 0.05. Panel (c):
M versus Log κL for vB = 1.5, θ = 60◦ , a = 0.05.
with a similar expression for U (0). Since from Eqs. (9.32), (9.29) and (9.28)
hQ dλ = hU dλ = 0 ,
line line
468 CHAPTER 9
the line-integrated Stokes parameters Q and U defined in Eq. (9.210)
Q= Q(0, λ) dλ , U= U (0, λ) dλ
line line
are both zero.
This property breaks down when saturation effects are important. In that case,
the contributions of π and σ components (which affect linear polarization in op-
posite ways) no longer compensate each other, because saturation effects are in
general different for either type. It follows that Q = 0 and U = 0. This is the
so-called mechanism of differential saturation (of π and σ components). Being due
to saturation effects, it is strictly related to the magnetic intensification mechanism
studied in the previous section. In the first papers dealing with net linear polar-
ization in magnetic lines (Leroy, 1962; Calamai et al., 1975) the term ‘magnetic
intensification’ was indeed used to denote both mechanisms.
In order to describe the net amount of linear polarization in a spectral line, it
is convenient to introduce dimensionless quantities by normalizing Q and U to the
net amount of continuum radiation subtracted by the line. Thus we introduce the
net linear polarization parameters
Q Q U U
q =− = , u =− = , (9.230)
I Ic (0) W I Ic (0) W
where I is defined in Eq. (9.210) and W is the equivalent width given by Eq. (9.219).
As in the case of the magnetic intensification parameter M, a good approximation
for q and u is obtained by assuming the Milne-Eddington model atmosphere. It is
also convenient to express the net linear polarization parameters in the preferred
reference frame (see Sect. 5.5). Setting
q̃ = q (χ = 0◦ ) , ũ = u (χ = 0◦ ) ,
we have from Eqs. (9.230), (9.219) and (9.112)
∞
∆−1 (1 + kI )2 kQ̃ + fQ̃ (kQ̃ fQ̃ + kV fV ) dλ
q̃ = − −∞
∞ & '
1 − ∆−1 (1 + kI ) (1 + kI )2 + fQ̃
2
+ fV2 dλ
−∞
∞
∆−1 (1 + kI ) (kV fQ̃ − kQ̃ fV ) dλ
−∞
ũ = ∞ & ' , (9.231)
1 − ∆−1 (1 + kI ) (1 + kI )2 + fQ̃
2
+ fV2 dλ
−∞
where
∆ = (1 + kI )4 + (1 + kI )2 (fQ̃
2
+ fV2 − kQ̃
2
− kV2 ) − (kQ̃ fQ̃ + kV fV )2
LINE FORMATION IN A MAGNETIC FIELD 469
and (see Eqs. (9.39) and (9.157))
kQ̃ = κL hQ̃ , fQ̃ = κL rQ̃ . (9.232)
The quantities Π and Φ, implicitly defined by
q̃ = Π cos 2Φ , ũ = Π sin 2Φ
with Φ ranging from 0 to π, are sometimes used in the place of q̃ and ũ . Inversion
of these relations gives (cf. Eqs. (1.8))
Π= q̃2 + ũ2
and, for q̃ = 0
1 ũ
Φ= arctan + Φ0
2 q̃
with ⎧
⎨0 if q̃ > 0 and ũ ≥ 0
Φ0 = π if q̃ > 0 and ũ < 0
⎩
π/2 if q̃ < 0 ,
while for q̃ = 0 1
4π if ũ > 0
Φ= 3
4π if ũ < 0 .
The relation between the net linear polarization parameters expressed in an arbi-
trary frame (rather than the preferred frame) and the set (Π, Φ) is easily obtained
from the transformation law of the Stokes parameters under rotation of the refer-
ence direction. Equations (1.45) yield
q = Π cos 2(Φ + χ) , u = Π sin 2(Φ + χ) .
The physical meaning of Π and Φ is obvious: Π is the net linear polarization
degree, Φ is the angle between the direction of maximum net linear polarization
and the transverse component of the magnetic field.
The net linear polarization parameters q̃ and ũ (or the quantities Π and Φ)
depend on the four parameters vB , θ, κL , a and on the structure of the Zeeman
pattern. Because of the involved form of Eqs. (9.231), only a few analytical results
can be obtained. These are the following:
a) Symmetry properties: if we consider the transformation θ → π − θ, we have
from Eqs. (9.32)
q̃ (π − θ) = q̃ (θ) Π(π − θ) = Π(θ)
ũ (π − θ) = − ũ (θ) Φ(π − θ) = π − Φ(θ) .
470 CHAPTER 9
b) Under the limit of weak, unsaturated lines (κL 1), expansion of Eqs. (9.231)
into power series of κL yields, with the help of Eqs. (9.32), (9.29) and (9.28)
∞ ∞
q̃ = 2 κL hQ̃ hI dv , ũ = κL hV rQ̃ − hQ̃ rV dv ,
−∞ −∞
where v is the reduced wavelength defined in Eqs. (9.26). These equations show
that Π grows linearly with κL and that Φ attains a finite value for κL → 0.
c) Under the limit of weak magnetic field (∆λB ∆λD ) we can expand the
integrands in Eqs. (9.231) by using Eqs. (9.69). To the lowest order in vB we
obtain, with the help of Eqs. (9.76), (9.224) and Table 3.4
1 2 I (κ , a)
q̃ = vB Ḡ sin2 θ 3 L
2 I1 (κL , a)
1 3 I (κ , a)
ũ = vB ḡ Ḡ sin2 θ cos θ 4 L , (9.233)
4 I1 (κL , a)
where I1 and I3 are defined in Eqs. (9.222) and where
∞
κ2L (η̇ ρ̈ − η̈ ρ̇)
I4 (κL , a) = dv ,
(1 + κL η)3
−∞
with η̇, η̈ given by Eqs. (9.223) and, similarly
∂ρ ∂ρ ∂2ρ ∂2ρ
ρ̇ = = ∆λD , ρ̈ = 2
= ∆λ2D .
∂v ∂λ ∂v ∂λ2
Equations (9.233) show that, for weak fields, the net linear polarization grows
quadratically with the magnetic field strength and linearly with sin2 θ. The direc-
tion of maximum polarization is determined by the sign of q̃ , hence (since both I1
and I3 are positive) by the sign of Ḡ. Most spectral lines have Ḡ > 0, which im-
plies that the net linear polarization is parallel to the transverse component of the
magnetic field. The opposite holds for the lines with Ḡ < 0 (net linear polarization
perpendicular to the transverse component of the magnetic field). Comparison of
the expression for q̃ given by Eqs. (9.233) with Eqs. (9.83) and (9.85) shows that
the sign of q̃ is the same as the sign of the Q̃ profile in the wings and is opposite
to the sign of Q̃ in the line core. This means that, for weak fields, σ components
are always less saturated than π components.
d) Under the limit of very large magnetic field (∆λB ∆λD ), we can use the results
developed in Sect. 9.12. The following expressions are obtained (see Eq. (9.229)
LINE FORMATION IN A MAGNETIC FIELD 471
for a similar derivation)
Nσ
, 1 + cos2 θ -
Nπ
, -
sin2 θ I1 κL si ,a − I1 κL si sin2 θ, a
1 + cos2 θ 2
i=1 i=1
q̃ = Nσ
, 1 + cos2 θ -
Nπ
, -
I1 κL si ,a + I1 κL si sin2 θ, a
i=1
2 i=1
ũ = 0 ,
where Nσ and Nπ are the total number of σ and π components, respectively. It
follows that, for saturated lines lying on the flat part of the curve of growth (and
excluding θ values close to 0◦ or 180◦ ), q̃ is roughly given by the expression
sin2 θ Nσ N
q̃ = − π , (9.234)
1 + cos2 θ N N
with N = Nπ + Nσ . Therefore, q̃ is negative for small values of sin2 θ (net lin-
ear polarization perpendicular to the transverse component of the magnetic field,
π components more saturated than σ components), while it is positive for large
values of sin2 θ (polarization parallel to the transverse component of the magnetic
field, π components less saturated than σ components). Obviously, there are two
values of θ such that the net linear polarization is zero,
.
Nσ − Nπ
θz = arccos and θz = 180◦ − θz .
N
Note that in the most common case where Nσ = 2 Nπ ,1
.
1
θz = arccos = θV ,
3
where θV is the Van Vleck angle defined in Eqs. (5.100)-(5.101). Equation (9.234)
also shows that for θ = 90◦ (transverse field), q̃ reaches its maximum value
Nσ − Nπ
q̃ max = ,
N
or, for most Zeeman patterns
1
q̃ max = .
3
1
For electric-dipole transitions, the relation Nσ = 2Nπ is always valid except when the lower
and upper level have the same half-integral J quantum number. In that case Nπ = 2J + 1 and
Nσ = 4J, hence
.
2J − 1
θz = arccos .
6J + 1
472 CHAPTER 9
Fig.9.12. Numerical results for the differential saturation mechanism. The net linear polarization
parameters q̃ and ũ are calculated from Eqs.(9.231) for the normal Zeeman triplet 1 S0 − 1 P1 .
The big dot corresponds to the ‘standard case’ values vB = 1.5, θ = 60◦ , κL = 20, a = 0.05. The
four curves are obtained by varying one parameter at a time, namely: vB (a); θ (b); κL (c);
a (d). Values of the varying parameter are marked by small dots.
The analytical results presented above are only valid within a limited region of
the hyperspace of the four parameters vB , θ, κL , a. In general, the calculation
of the net linear polarization requires (under the Milne-Eddington assumptions)
a numerical evaluation of the integrals in Eqs. (9.231). Sample results are shown
in Fig. 9.12. It is seen that, except in the limiting cases of very weak and very
strong magnetic field, the parameter ũ makes an important contribution to the net
linear polarization. Since this parameter vanishes when magneto-optical effects are
neglected (see Eqs. (9.231)), it follows that these effects play a major role in the
mechanism of differential saturation. Numerical results on the net linear polar-
ization have been presented by Calamai et al. (1975), Landi Degl’Innocenti and
Calamai (1982), Calamai and Landi Degl’Innocenti (1983), and Bagnulo (1993). In
these works the line strength is parameterized
√ through the quantity η0 which is con-
nected with κL by the relation η0 = κL / π (see the remarks following Eqs. (9.32)).
Finally, it should be pointed out that differential saturation of π and σ compo-
LINE FORMATION IN A MAGNETIC FIELD 473
nents is the only mechanism that can introduce net linear polarization in spectral
lines when atomic polarization effects are neglected. We will see in the following
(Sect. 10.5) that if the alignment components of the atomic density matrix are
non-zero, there is an extra-contribution to net linear polarization. The definition
of alignment and its physical meaning are discussed in Sect. 3.7.
9.21. Net Circular Polarization in Spectral Lines
In the two previous sections we have seen that saturation effects in magnetic lines
produce an increase of equivalent width and the appearance of net linear polariza-
tion. For circular polarization the situation is quite different, because saturation
effects are necessary but not sufficient for the appearance of a net contribution
from a given spectral line.
The proof of this statement is straightforward. For the first part (saturation
effects are necessary) it is enough to consider the generalized Seares formulae
(Eqs. (9.151)) that we rewrite here for the emerging V Stokes parameter
∞ ∞
−τc −(τc − τc )
V (0) = e κL (τc ) hV (τc ) BP (τc ) − e BP (τc ) dτc dτc .
0 τc
Since from Eqs. (9.32), (9.29) and (9.28)
hV dλ = 0 ,
line
the line-integrated V Stokes parameter in a weak (unsaturated) line is zero,
V= V (0, λ) dλ = 0 .
line
The second part of the statement (saturation effects are not sufficient) is directly
proved by the wavelength symmetry properties of the Stokes parameters outlined
in Sect. 9.5 and summarized in Eqs. (9.54): for a static (or constant-velocity)
atmosphere we find again V = 0.
These arguments suggest that, in order to obtain net circular polarization in
magnetic spectral lines, it is necessary to have saturation effects combined with a
velocity gradient.
Similarly to Eqs. (9.230), we define the (dimensionless) net circular polarization
parameter to be
V V
v =− = , (9.235)
I Ic (0) W
with I and W given by Eqs. (9.210) and (9.219) respectively, and Ic (0) the emerg-
ing continuum intensity. This parameter has a very complicated dependence on
474 CHAPTER 9
the physical quantities specifying the stellar atmosphere (including the velocity
gradient) and on the Zeeman pattern of the line. This makes extremely difficult
to derive its general properties except in some special cases based on drastic ap-
proximations. Obviously, it is impossible to use the Milne-Eddington solution to
the transfer equation (as done in the former section for the net linear polarization)
because the Milne-Eddington model atmosphere is static.
In the following we consider three special cases where analytical expressions for
the parameter v can be derived:
i) a zero-order Milne-Eddington atmosphere is ‘perturbed’ by a (small) velocity
gradient;
ii) an arbitrary atmosphere with vertical magnetic field presents a velocity gradient
associated with a gradient of the magnetic field intensity;
iii) the atmosphere is composed of two superposed Milne-Eddington atmospheres
with a discontinuity of all the physical parameters (including velocity) at the tran-
sition layer.
In case i) the velocity gradient is small and continuous and it is not associated
with other gradients. In case ii) the velocity gradient is continuous and not nec-
essarily small, and it is associated with a gradient of the magnetic field intensity.
Finally, in case iii) the velocity gradient is discontinuous (being mathematically de-
scribed by a Dirac delta-function) and it is associated with discontinuous gradients
of the other physical quantities.
Before we analyze these special cases, some general remarks about the line-of-
sight velocity gradient are necessary. Given a stationary stellar atmosphere where
a macroscopic velocity field w A is present, we introduce the terms1
expanding atmosphere if A>0
div w
compressing atmosphere if A<0.
div w
The reason for these terms is the following. Denoting by ρ the mass density of the
medium, its total (Eulerian) time derivative is given by
dρ
A · grad ρ ,
= w
dt
since the atmosphere is stationary (∂ρ/∂t = 0). Combining this relation with the
mass conservation equation
div (ρ w A · grad ρ + ρ div w
A) = w A=0,
we get
dρ
= − ρ div w
A ,
dt
1 A has the same sign throughout the whole atmosphere.
For simplicity, we assume that div w
Note that a similar terminology was first introduced by Skumanich and Lites (1987).
LINE FORMATION IN A MAGNETIC FIELD 475
so that dρ/dt is negative for an ‘expanding’ atmosphere and positive for a ‘com-
pressing’ atmosphere. It follows that each plasma element goes across regions of
decreasing density in the former case and of increasing density in the latter.
In the particular case of a plane-parallel atmosphere with a vertical (upward or
A , depend only on
downward) velocity field, all the physical quantities, including w
the geometrical depth z defined in Fig. 9.2. For a radiation beam travelling along
the line-of-sight component of the velocity w
the direction Ω, A – which we keep on
denoting by wA – is, according to the sign astrophysical convention (see Sect. 5.4)
·w
wA = − Ω A .
If s is the coordinate measured along the ray path and µ = cos α, with α defined
in Fig. 9.2, we get from the two equations above
dwA
· dw A = − µ Ω
= −Ω · ∂w A = − µ2 div w
A .
ds ds ∂z
In terms of the optical depths τc or tc defined in Eqs. (9.33) and (9.34) we have
dwA µ2 dwA µ
= A ,
div w = A .
div w
dτc kc dtc kc
Therefore, under the conditions specified above (plane-parallel atmosphere, vertical
velocity field, astrophysical convention for wA ), equivalent definitions of ‘expand-
ing’ atmosphere are provided by the inequalities
dwA
>0
dτc
or
dwA
> 0 for radiation travelling outward .
dtc
It should be remarked that the distinction between ‘expanding’ and ‘compressing’
atmosphere only involves the sign of the velocity gradient, not the sign of the
velocity itself. In fact, as recalled at the beginning of this section, the addition of
a constant velocity field simply produces an overall wavelength shift of the Stokes
profiles and has no effect on the net circular polarization parameter v.
Let’s now turn to the analysis of the three schematic models listed above.1
Case i) We consider a Milne-Eddington atmosphere, characterized by the constant
propagation matrix C defined in Eq. (9.91). The atmosphere is perturbed by a
vertical velocity field which varies linearly with optical depth. For a radiation beam
flowing along an assigned µ-direction, we describe the line-of-sight velocity by the
equation
(0) (1)
wA (tc ) = wA + wA tc ,
1
A similar analysis has been presented in Landolfi and Landi Degl’Innocenti (1996).
476 CHAPTER 9
where tc is the optical depth measured along the vertical (Eqs. (9.34)).
We now make the further assumption that the perturbation is everywhere small,
that is (cf. Eqs. (9.26))
wA (tc ) λ0
vA (tc ) = 1
c ∆λD
or
(0)
wA (1)
, wA wT , (9.236)
where wT is the thermal velocity defined in Eqs. (5.42) and (5.46). This assump-
tion enables the first-order correction to the emerging V Stokes parameter to be
evaluated using the appropriate response function. Similarly to Eq. (9.200), we can
write
V (0, µ) = V (0, µ) + δV (0, µ) , (9.237)
where the zero-order term V (0, µ), corresponding to the static atmosphere, makes
no contribution to the net circular polarization parameter, which thus reduces to
(see Eqs. (9.235) and (9.210))
∞
1
v = δV (0, µ) dλ . (9.238)
Ic (0) W
−∞
On the other hand, Eqs. (9.39), (9.32), (9.29), (9.27) and (9.26) show that
∂ C −1 λ0 ∂ C −1 λ0 ∂ C −1 λ ∂ C −1
= =− =− 0 ,
∂wA c ∆λD ∂vA c ∆λD ∂v c ∂λ
thus we can apply Eq. (9.205) to obtain
λ0
δV (0, µ) = −B0 βµ
c
∂ −1 3
−1 ∂ −1
× wA
(0)
C 30
+ µ wA
(1)
C 3i
C i0
. (9.239)
∂λ i=0
∂λ
Since from Table 9.5
−1
lim C ij
=0 for i = j ,
λ→±∞
we have1
∞
∂ −1
C 30
dλ = 0 . (9.240)
∂λ
−∞
1 (0)
Equation (9.240) shows that the term wA does not affect the net circular polarization. The
(0)
condition wA wT in Eq. (9.236) is necessary only if we want to regard V (0, µ) in Eq. (9.237) as
the solution corresponding to the static atmosphere. However, V (0, µ) could as well be regarded
(0)
as the solution corresponding to a constant-velocity atmosphere with line-of-sight velocity wA ,
(0)
which again proves that the condition wA wT is indeed unnecessary for the following results
(in particular Eq. (9.241)) to hold.
LINE FORMATION IN A MAGNETIC FIELD 477
Integrating by parts the last term in Eq. (9.239) we have
∞ ∞
3
−1 ∂ −1
3
∂ −1 −1
C 3i ∂λ
C i0
dλ = − C 3i
C i0
dλ ,
i=0 i=0
∂λ
−∞ −∞
whence
∞ ∞
3
−1 ∂ −1 1 ∂ C −1
C 3i ∂λ
C i0
dλ = − , C −1 dλ ,
i=0
2 ∂λ 30
−∞ −∞
where the integrand in the right-hand side is the element (3, 0) of the commutator
of the two matrices. The denominator of Eq. (9.238) is easily evaluated using
Eqs. (9.219) and (9.109). To the lowest order we obtain
∞
Ic (0) W = B0 βµ 1 − C −1 00 dλ .
−∞
Substitution of the above equations into Eq. (9.238) yields
∞
∂ C −1
, C −1 dλ
1 λ0 (1) −∞ ∂λ 30
v = µ wA ∞ . (9.241)
2 c
1 − C −1 00 dλ
−∞
The analytical calculation can be further developed by substituting the expressions
for the elements of the matrix C −1 given in Table 9.5. If the calculation is per-
formed in the preferred frame (the Stokes parameter V is independent of the choice
of the reference direction), the following expression is obtained
∞
(1) ∆−2 A dv
wA λ0 −∞
v = µ ∞&
'
, (9.242)
c ∆λD −1
1 − ∆ (1 + kI ) (1 + kI )2 + fQ̃ 2
+ fV2 dv
−∞
where ∆λD is the Doppler width, v the reduced wavelength defined in Eqs. (9.26),
and
&
A = (1 + kI )3 (kQ̃ kV + fQ̃ fV )
2 2 ' ∂kQ̃
+ (1 + kI ) −kQ̃ kV fQ̃ − fV2 + fQ̃ fV kQ̃ − kV2
∂v
∂k
− (1 + kI )3 kQ̃2 2
+ fQ̃ V
∂v
&
+ (1 + kI )3 (kV fQ̃ − kQ̃ fV )
2 ' ∂fQ̃
+ (1 + kI ) kQ̃ fV kV2 + fQ̃ 2
+ kV fQ̃ kQ̃ + fV2
∂v
2
2 ∂fV
− (1 + kI ) (kQ̃ fQ̃ + kV fV ) kQ̃ + fQ̃ ; (9.243)
∂v
478 CHAPTER 9
the quantities kQ̃ and fQ̃ are given by (cf. Eqs. (9.232))
kQ̃ = kQ (χ = 0◦ ) = κL hQ̃ , fQ̃ = fQ (χ = 0◦ ) = κL rQ̃ .
Equations (9.242)-(9.243) show that the net circular polarization parameter de-
pends on the four quantities vB , θ, κL , a and on the structure of the Zeeman
(1)
pattern. Moreover, it is proportional to the line-of-sight velocity gradient µ wA ,
so that its sign changes if the velocity gradient is inverted (wA → − wA ). Be-
(1) (1)
cause of the involved structure of the expression, only a few results can be derived
analytically. These are the following:
a) It is easily seen, with the help of Eqs. (9.32), that v = 0 for θ = 0◦ and θ = 180◦ ,
and that
v (180◦ − θ) = − v (θ) ,
which implies in particular v (θ = 90◦ ) = 0. In other words, under the assumption
of weak velocity gradient considered here, the net circular polarization vanishes for
longitudinal and transverse magnetic fields.
b) Under the limit of weak magnetic field (∆λB ∆λD ), the expansions in
Eqs. (9.69) can be used. Recalling Eqs. (9.75) and (9.76) one obtains, to the
lowest order in vB
1 w(1) λ0 I (κ , a)
v =− µ A ḡ Ḡ2 vB
5
sin4 θ cos θ 5 L , (9.244)
16 c ∆λD I1 (κL , a)
where I1 (κL , a) is defined in Eqs. (9.222) and
∞ ... ...
κ5L (η̈ η̇ + ρ̈ ρ̇) η − (η̈ 2 + ρ̈ 2 ) η̈ + (η̇ ρ̈ − η̈ ρ̇) ρ
I5 (κL , a) = dv ,
(1 + κL η)5
−∞
with η and ρ given by Eqs. (9.65) and with the dots denoting derivatives with
respect to the reduced wavelength (cf. Eqs. (9.223)).
A numerical analysis of the integral I5 shows that, excluding weak lines (κL
less than about 1.5), it is a positive quantity. This implies that the net circular
polarization parameter has the same sign as the red wing of the V Stokes profile1
(1)
for ‘expanding’ atmospheres (wA > 0), while it has the sign of the blue wing
(1)
for ‘compressing’ atmospheres (wA < 0). It must be pointed out that magneto-
optical terms play a crucial role in determining the value of v : neglect of these
terms results in a sign switch of the integral I5 for practically all values of the
parameters κL and a .
Another important consequence of Eq. (9.244) is the explicit dependence of v on
the inclination angle θ. The net circular polarization is maximum for
1
θ1 = arccos √ 63◦. 4 , θ2 = 180◦ − θ1 116◦. 6 .
5
1
According to Eq. (9.80), the sign of the red wing of the V profile is the sign of the quantity
(− ḡ cos θ).
LINE FORMATION IN A MAGNETIC FIELD 479
Finally, the proportionality to the factor ḡ Ḡ2 shows that, for electric-dipole tran-
sitions of assigned effective Landé factor, the net circular polarization is maximum
for Zeeman triplets (see Sect. 9.6).
c) Under the limit of very large magnetic field (∆λB ∆λD ), direct substitution
of Eqs. (9.147) into Eq. (9.242) shows that v = 0.1
When the magnetic field is neither weak nor strong, one must resort to a numer-
ical evaluation of Eq. (9.242). Some results are shown in Fig. 9.13 for the quantity
v/v0 , where v0 = µ wA (1)
λ0 /(c ∆λD ) is the perturbative parameter (v0 1 for
Eq. (9.242) to hold).
Case ii) Given an arbitrary stellar atmosphere, let us consider the radiation flowing
along a given outward direction, and let us suppose that the magnetic field vector
at each point along the ray path is parallel to the propagation direction (θ = 0◦ )
or antiparallel (θ = 180◦). The results obtained in Sect. 9.10 and summarized by
Eqs. (9.124) and (9.125) allow the net circular polarization parameter to be written
in the form (see Eqs. (9.235), (9.210) and (9.219))
Wb − Wr
v = σ , (9.245)
Wb + Wr
where Wb and Wr are the equivalent widths of two ‘fictitious’ lines formed in the
same atmosphere, and having the same κL as the real line, and absorption profiles
ηb and ηr , respectively; σ is a sign factor given by
+1 for θ = 0◦
σ= (9.246)
−1 for θ = 180◦ .
To analyze the behavior of the net circular polarization parameter, we consider
the simplest case of a Zeeman triplet. In this case the expressions for ηb and ηr are
particularly simple. From Eqs. (9.29) and (9.27) we have
1
ηb = √ H(v − vA + g vB , a)
π
1
ηr = √ H(v − vA − g vB , a) ,
π
which show that ηb and ηr are identical profiles centered, respectively, at the wave-
lengths λb and λr given by (see Eqs. (9.26))
λb = λ0 + ∆λA − g ∆λB
λr = λ0 + ∆λA + g ∆λB ,
1
Note that the two conditions in Eqs. (9.138) are satisfied since, by assumption, the magnetic
field is constant and large and the velocity gradient is small.
480 CHAPTER 9
Fig.9.13. The net circular polarization computed from Eq.(9.242) is plotted versus different
parameters. Panels (a), (b) and (c) refer to the normal Zeeman triplet 1 S0 − 1 P1 . Panel (d)
refers to three different transitions having ḡ = 1 : 1 S0 − 1 P1 (Ḡ = 1, full line), 3 P1 − 3 D1
(Ḡ = 0.250, short dash), 3 D2 − 3 F3 (Ḡ = 0.997, long dash).
where ∆λA = λ0 wA /c .
Let us now suppose that both the magnetic field intensity and the line-of-sight
component of the macroscopic velocity of the medium vary with optical depth. It
follows that the wavelengths λb and λr are themselves functions of optical depth,
LINE FORMATION IN A MAGNETIC FIELD 481
η η
τ λ λ λ λ λ
Fig.9.14. The combined effect of a velocity gradient and a magnetic field gradient on the absorp-
tion profiles ηb and ηr . For an ‘expanding’ atmosphere (∆λ(1)A
> 0), at each τc the profiles would
be centered at the wavelengths λb and λr (dashed lines) if the magnetic field were constant. If,
on the contrary, the intensity of the magnetic field increases with τc (∆λ(1) B
> 0), the profiles
are centered at the wavelengths λb and λr (solid lines). The ηr profiles at different τc are less
overlapped than the ηb profiles, thus Wr > Wb .
and this affects the equivalent widths of the two ‘fictitious’ lines. The effect will in
general be different for Wb and Wr , which eventually results in a net amount of
circular polarization.
General properties of v cannot be established without a knowledge of the depen-
dence of ∆λA and ∆λB on optical depth. Here we restrict attention to the simplest
case of linear variations1
∆λA (τc ) = ∆λ(0) (1)
A + ∆λA τc
∆λB (τc ) = ∆λ(0) (1)
B + ∆λB τc , (9.247)
whence
(1)
b + ∆λA − g ∆λB
λb = λ(0) (1)
τc
(1) (1)
λr = λ(0)
r + ∆λA + g ∆λB τc , (9.248)
(0) (0)
where λb and λr are the values of λb and λr at τc = 0.
If the atmosphere is ‘standard’ (i.e. if temperature is a monotonic increasing
function of optical depth, so that absorption lines are formed), the equivalent width
should be expected to be larger, the larger is the wavelength shift of the absorption
profile between two assigned optical depths, because saturation effects are reduced
(see Fig. 9.14). It follows that
1
Note that ∆λB is, by definition, a positive quantity. If ∆λ(1)
B
is negative, Eqs. (9.247) lead
to negative values of ∆λB for large τc . Obviously, we assume that these atmospheric layers do
not contribute to the emerging radiation.
482 CHAPTER 9
⎧
⎪ Wr > Wb if | ∆λ(1)
A + g ∆λB | > | ∆λA − g ∆λB |
(1) (1) (1)
⎪
⎨
Wr = Wb if | ∆λ(1)
A + g ∆λB | = | ∆λA − g ∆λB |
(1) (1) (1)
(9.249)
⎪
⎪
⎩
Wr < Wb if | ∆λ(1)
A + g ∆λB | < | ∆λA − g ∆λB | .
(1) (1) (1)
These relations show that: (a) Wr = Wb if either ∆λ(1)
A = 0 (no velocity gradient)
or ∆λ(1) (1)
B = 0 (no magnetic field gradient); (b) Wr > Wb if ∆λA and g ∆λB
(1)
(1) (1)
have the same sign; (c) Wr < Wb if ∆λA and g ∆λB have opposite signs.
Equations (9.249) can be rewritten in the form1
sign Wr − Wb = sign g ∆λ(1) (1)
A ∆λB ,
or, in terms of v
sign (v ) = sign − σ g ∆λ(1) (1)
A ∆λB , (9.250)
with σ defined in Eq. (9.246).
A quantitative confirmation of this result can be obtained by introducing further
approximations. Following Sánchez Almeida et al. (1989), we assume both the
velocity gradient and the magnetic field gradient to be small (∆λ(1) A and ∆λB
(1)
much smaller than the Doppler width ∆λD ), so that the transfer equation can
be solved by a perturbative expansion. By further assuming a Milne-Eddington
model atmosphere, one obtains a closed expression for the equivalent width. These
calculations are carried out in App. 10 and show that, up to the second order, the
equivalent widths Wr and Wb can be written in the form (see Eqs. (A10.8), (9.248)
and (A10.7))
(1) 2
β ∆λA ± g ∆λ(1)
Wr,b = ∆λD I1 (κL , a) + B
I6 (κL , a) ,
1+β ∆λD
where the plus and minus signs refer to the red and blue component respectively,
I1 (κL , a) is given by Eqs. (9.222) and I6 (κL , a) by Eq. (A10.9). Substitution into
Eq. (9.245) yields, to the lowest order
∆λ(1)
A ∆λB
(1)
I6 (κL , a)
v = −2 σg ,
∆λ2D I1 (κL , a)
which confirms Eq. (9.250) since both I1 (κL , a) and I6 (κL , a) are positive. Note
that this expression is valid irrespective of the magnetic field strength: only the
gradients (∆λ(1) (1)
A and ∆λB ) are required to be small. A similar expression has
been given by Pavlov and Shibanov (1978).
Case iii) Let us consider the following discontinuous model atmosphere. A thin
slab with a vertical velocity field is superposed on an arbitrary, static atmosphere;
1
A similar expression has been derived by Solanki and Pahlke (1988).
LINE FORMATION IN A MAGNETIC FIELD 483
both the slab and the underlying atmosphere are magnetized, the magnetic field
vectors being in general different.1 We restrict attention to the radiation travelling
in the vertical, outward direction, and denote by ts the continuum optical thickness
(s)
of the slab (ts 1), by wA the line-of-sight component of the velocity field (de-
≡ (B , θ , χ )
fined, as usual, according to the astrophysical convention), and by B s s s s
and B ≡ (B, θ, χ), respectively, the magnetic field vectors in the slab and in the
underlying atmosphere.
The transfer of radiation across the slab is described by Eq. (9.90), where the
propagation matrix C (s) and the emission vector j (s) can be considered constant
because ts 1. The emerging Stokes parameters can be easily calculated by
integration of Eq. (9.90)
I(0) = 1 − ts C (s) I (b) + ts j (s) , (9.251)
where I (b) is the boundary Stokes vector resulting from the lower atmosphere.
As far as the V Stokes parameter is concerned, the first and third term in
Eq. (9.251) give rise to antisymmetrical profiles that make no contribution to the
line-integrated circular polarization, which reduces to
∞
3
(s) (b)
V = − ts C3i Ii dλ .
−∞ i=0
To the lowest order in ts , the net circular polarization parameter is therefore (see
Eqs. (9.235), (9.219), (9.91) and (9.39))
∞
(s) (s) (s) (s)
hV I (b) + rU Q(b) − rQ U (b) + hI V (b) dλ
(s) −∞
v = − ts κ L ∞ , (9.252)
Ic(b) − I (b) dλ
−∞
(b)
where Ic is the continuum intensity emerging from the underlying atmosphere.
Note that the Stokes profiles I (b) , Q(b) , U (b) , V (b) are centered at the rest wave-
(s) (s) (s) (s)
length λ0 , while the profiles hV , rU , rQ , hI are centered at the wavelength
(s)
(s) wA
λ0 + ∆λA = λ0 1 + .
c
(s)
The symmetry properties (9.47) and (9.54) show that v is zero both when ∆λA is
(s) (s)
zero and when it is so large that the profiles I (b) , . . . , V (b) and hV , . . . , hI do
1
This model was proposed by Illing et al. (1975) to explain the broad-band circular polarization
observed in sunspots, and it is often referred to as the Illing (or Illing, Landman and Mickey)
model. Some quantitative estimates of the net circular polarization produced by the model have
been given by Auer and Heasley (1978).
484 CHAPTER 9
not overlap. It is therefore to be expected that the maximum value of v is attained
(s)
when ∆λA and the widths of the profiles are comparable.
The integrand in the numerator of Eq. (9.252) is rather involved and it is practi-
cally impossible to obtain any quantitative conclusion about v without introducing
further approximations. Even if the Unno-Rachkovsky solution is used to express
the boundary Stokes vector, the integrand remains too involved. Therefore, we
make the drastic approximation that the boundary Stokes profiles are described by
the Seares formulae (we recall that the Unno-Rachkovsky formulae reduce to the
Seares formulae for weak spectral lines, cf. Sect. 9.13). Substituting Eqs. (9.149)
– with µ = 1 – into Eq. (9.252), we obtain
(s) ∞
t κ (s) (s) (s) (s)
v = s L hV hI + rU hQ − rQ hU + hI hV dλ
∆λD
−∞
where the quantities hI , . . . , hV and the Doppler width ∆λD refer to the lower
atmosphere. Use of Eqs. (9.32) yields, with evident notations
(s)
1 ts κ L
v =
4 ∆λD
∞
1 (s)
× cos θs (1 + cos2 θ) + cos θ (1 + cos2 θs ) ηr(s) ηr − ηb ηb dλ
2
−∞
∞ ∞
(s)
+ cos θs sin θ 2
ηr(s) − ηb ηp dλ + cos θ sin θs ηp(s) ηr − ηb dλ
2
−∞ −∞
∞
1 (s)
+ cos θs (1 + cos2 θ) − cos θ (1 + cos2 θs ) ηr(s) ηb − ηb ηr dλ
2
−∞
∞ (s) (s) !
ρb + ρr ηb + ηr
− sin θs sin θ sin 2(χ − χs )
2 2
ρ(s)
p − ηp − dλ . (9.253)
2 2
−∞
Two general properties of v follow from this expression. The first is that magneto-
optical effects are important only if the azimuth angles of the magnetic field in the
slab and in the underlying atmosphere are different. The second, which can be
easily deduced with the help of Eqs. (9.46), is that v is zero if the magnetic field
gradient is zero (B =B ), even in the presence of a velocity gradient (w(s) = 0).
s A
This could seem inconsistent with the results of Case i) considered above, but is
really quite logical because saturation effects are neglected in the present model:
the slab is by assumption optically thin and the Seares formulae describe a weak,
unsaturated spectral line.
In the following we analyze separately the effects of the velocity gradient com-
bined with a discontinuity at the boundary layer of: (a) the modulus of the
LINE FORMATION IN A MAGNETIC FIELD 485
magnetic field (∆B-effect ); (b) the inclination angle (∆θ-effect ); (c) the azimuth
angle (∆χ-effect ).
In case (a), being θs = θ and χs = χ , only the first two lines of Eq. (9.253) are
non-zero. Thus we obtain
(s) ∞
1 ts κ L (s)
v = cos θ 2
(1 + cos θ) ηr(s) ηr − ηb ηb dλ
4 ∆λD
−∞
∞ !
(s)
+ sin2 θ ηr(s) ηp + ηp(s) ηr − ηb ηp − ηp(s) ηb dλ . (9.254)
−∞
For simplicity, we restrict attention to Zeeman triplets and we assume that the
Doppler width and the damping constant in the slab and in the lower atmosphere
are the same. The integrals in Eq. (9.254) can be performed analytically via
(s)
Eqs. (5.64). Using Eqs. (9.29) and (9.27), and denoting by ∆λB and ∆λB the
Zeeman splittings in the slab and in the atmosphere, respectively, we get
1 (s)
v = √ ts κL cos θ
4 2π
&
× (1 + cos2 θ) H(v̂rr , â) − H(v̂bb , â)
'
+ sin2 θ H(v̂rp , â) + H(v̂pr , â) − H(v̂bp , â) − H(v̂pb , â) , (9.255)
where (some additional symbols are defined for future reference)
1 (s) (s)
v̂rr = √ ∆λA + g ∆λB − ∆λB
2 ∆λD
1 (s) (s)
v̂bb =√ ∆λA − g ∆λB − ∆λB
2 ∆λD
1 (s) (s)
v̂rp = √ ∆λA + g ∆λB
2 ∆λD
1 (s) (s)
v̂bp = √ ∆λA − g ∆λB
2 ∆λD
1 (s)
v̂pr = √ ∆λA − g ∆λB
2 ∆λD
1 (s)
v̂pb = √ ∆λA + g ∆λB
2 ∆λD
1 (s) (s)
v̂rb =√ ∆λA + g ∆λB + ∆λB
2 ∆λD
486 CHAPTER 9
1 (s) (s)
v̂br = √ ∆λA − g ∆λB + ∆λB
2 ∆λD
1 (s)
v̂pp = √ ∆λA
2 ∆λD
√
â = 2 a .
If we separate the contributions of the first and second line of Eq. (9.255) by
writing
v = (1 + cos2 θ) v(1) + sin2 θ v(2) , (9.256)
the sign of v(1) can be easily deduced. Since H(v, a) is a monotonically decreasing
function of |v| , we have
(s) (s)
sign v(1) = sign cos θ g ∆λA ∆λB − ∆λB , (9.257)
which agrees with the result obtained previously for continuous gradients and
vertical magnetic field (cf. Eq. (9.250)).1 For slightly inclined magnetic field
(sin2 θ 1), Eq. (9.257) gives the sign of the net circular polarization. By contrast,
(s)
no general rule can be established for the sign of v(2) . When either ∆λB = 0 (non-
magnetic slab superposed on a magnetic atmosphere) or ∆λB = 0 (magnetic slab
superposed on a non-magnetic atmosphere), it can be easily shown that v(2) = v(1) .
However, there are cases where the signs of v(2) and v(1) are opposite.
Let us now consider the ∆θ-effect. Being now Bs = B and χs = χ , only the
second and third line of Eq. (9.253) are non-zero. Under the same assumptions as
before (Zeeman triplet, same Doppler width and damping constant in the slab and
in the atmosphere), one obtains
1 (s)
v = √ ts κ L
4 2π
&
× cos θs sin2 θ − cos θ sin2 θs H(v̂rp , â) − H(v̂bp , â)
1 '
+ cos θs (1 + cos2 θ) − cos θ (1 + cos2 θs ) H(v̂rb , â) − H(v̂br , â) .
2
It can be easily shown that both the factors containing θs and θ are positive when
θ > θs and negative when θ < θs , and that both the factors containing the Voigt
(s) (s)
functions are positive when g ∆λA < 0 and negative when g ∆λA > 0. Therefore,
we have the sign rule
(s)
sign (v ) = sign − g ∆λA θ − θs . (9.258)
1
Note that positive values of ∆λ(1) and ∆λ(1)
(s)
B A
in Eq. (9.250) are equivalent to ∆λB > ∆λB
(s)
and ∆λA < 0 in Eq. (9.257).
LINE FORMATION IN A MAGNETIC FIELD 487
Fig.9.15. Contour plot of the function in square brackets of Eq.(9.259) in the plane (vA , vB ) =
(s)
(∆λA /∆λD , ∆λB /∆λD ). The plus and minus signs denote the regions where the function is
positive or negative. The value of the damping constant is a = 0 (the contour plot depends very
slightly on a).
It is interesting to compare this relation with Eq. (9.257). It appears that a mag-
netic field of constant orientation whose modulus increases with optical depth, and
a magnetic field of constant modulus whose longitudinal component increases (in
absolute value) with optical depth, produce the same effect on the net circular
polarization parameter. This result has been deduced by Solanki and Montavon
(1993) with numerical computations. Sánchez Almeida and Lites (1992), who pro-
posed for the θ-gradient effect the name ∆θ-mechanism, pointed out that it is more
effective than the ∆B-effect when the magnetic field is highly inclined, while the
opposite holds when the field is slightly inclined.
Finally, we consider the ∆χ-effect.1 In this case, being Bs = B and θs = θ, only
the fourth line of Eq. (9.253) is non-zero. Under the same assumptions as before,
and using Eqs. (5.64) to evaluate the convolutions of Voigt functions and dispersion
profiles, we obtain the expression
1 (s)
v = √ ts κL sin4 θ sin 2(χ − χs )
4 2π
3 1 1
× L(v̂pp , â) − L(v̂rp , â) − L(v̂bp , â) + L(v̂rb , â) + L(v̂br , â) . (9.259)
2 4 4
It is important to remark that the presence of the factor sin4 θ in this equation
makes the ∆χ-effect particularly sensitive to transverse magnetic fields.
A numerical analysis of the term in square brackets of Eq. (9.259) is presented
in Fig. 9.15, which shows that no simple rule can be established for the sign of this
1
The possibility that this mechanism might be effective in sunspots was first suggested by
Makita (1986).
488 CHAPTER 9
(s)
term. However, if we exclude values of |∆λA | larger than approximately 1.5 ∆λD ,
we get the rule
(s)
sign (v ) = sign ∆λA χ − χs . (9.260)
Contrary to Eqs. (9.257) and (9.258), the sign rule (9.260) is independent of the
Landé factor. This property enables in principle the ∆χ-effect to be disentangled
from the two previous effects by using a couple of spectral lines having a positive
and a negative Landé factor respectively. According to Eq. (9.260), the sign of v
depends on the sign of the velocity gradient and on the twisting direction of the
magnetic field. Thus we have the following rule for the ∆χ-effect: an expanding
(s)
atmosphere (∆λA < 0) and a magnetic field winding as the thread of a right-
handed screw (χ < χs ) give rise to positive net circular polarization.
The results presented in this section show that the phenomenon of net circular
polarization in spectral lines is extremely complex. Furthermore, it should be
stressed that the net circular polarization discussed above is due to velocity and/or
magnetic field gradients only, while blends and atomic orientation have not been
considered.
Blends between independent lines, or between different components of a multi-
plet, can indeed produce net circular polarization (cf. Figs. 9.9 and 9.18). Further
remarks on this topic are presented in App. 11. The effect of atomic orientation
will be discussed in Sect. 10.5. We just point out that atomic orientation can
be set up both in isolated atomic levels (through a mechanism which involves
the simultaneous presence of magnetic fields and velocity gradients; see Landi
Degl’Innocenti, 1985b) and in fine-structured or hyperfine-structured atomic levels
(through a mechanism of ‘alignment-to-orientation conversion’ due to anisotropic
illumination; see Sect. 10.20). Net circular polarization in spectral lines can also
be produced by velocity-magnetic field correlations in stochastic media. This phe-
nomenon will be discussed in Sect. 9.24.
9.22. The Importance of Magneto-Optical Effects
In the early days of polarized radiative transfer, the importance of magneto-optical
effects was basically underestimated and, in many applications, these effects were
simply neglected. Now we know that they are a fundamental ingredient of polarized
radiative transfer, and we have seen that they arise quite naturally both in the
classical derivation (Sects. 5.1-5.2) and in the quantum derivation (Sects. 6.6-6.7)
of the transfer equation.
However, since this equation considerably simplifies when magneto-optical ef-
fects are ignored, it is important to point out those physical situations where such
approximation is justified. These are the following:
a) Weak field limit ( ∆λB ∆λD ).
Extension of the perturbative scheme described in Sect. 9.6 up to the fifth order
LINE FORMATION IN A MAGNETIC FIELD 489
shows that magneto-optical effects yield, in general, corrections of order
5 4
∆λB ∆λB ∆λB
, ,
∆λD ∆λD ∆λD
on the Stokes parameters I, V , and Q-U , respectively.
b) Intense field limit ( ∆λB ∆λD ).
In Sect. 9.12 it has been shown that under this limit, and under the supplementary
condition of a magnetic field having the same direction at any optical depth (θ =
const., χ = const.), magneto-optical effects can be neglected.
c) Longitudinal field ( θ = 0◦ or 180◦ ) or transverse field ( θ = 90◦ ).
In these special cases magneto-optical effects do not influence the Stokes parame-
ters, as shown in Sect. 9.10. When θ = 90◦ , however, the supplementary condition
of constant azimuth (χ = const.) is also necessary.
Except for the particular cases listed above, magneto-optical effects must be
taken into account and give rise to characteristic signatures in the Stokes pa-
rameters profiles. A comparison between typical profiles emerging from a Milne-
Eddington atmosphere, computed with and without magneto-optical effects, is
shown in Fig. 9.16. From the figure, and from similar results obtained for dif-
ferent values of the relevant parameters, the following qualitative conclusions can
be drawn:
i) As pointed out by Wittmann (1971), magneto-optical effects can introduce a
reversal in the V profile around line center (magneto-optical undulation). This
feature is rather sensitive to the intensity and inclination of the magnetic field,
and also to the structure of the Zeeman pattern: in the case of triplets, it appears
only for g ∆λB of the order or larger than 2 ∆λD and for rather inclined fields
(approximately 40◦ ≤ θ ≤ 140◦ – see e.g. panel (b) of Fig. 9.5).
ii) The Stokes parameters which are relatively more affected by magneto-optical
effects are the linear polarization parameters Q and U . This point is better il-
lustrated by considering (under the assumption that the azimuth of the magnetic
field is constant with optical depth) Q̃ and Ũ , the Stokes parameters defined in
the ‘preferred frame’. The profile Ũ vanishes at all wavelengths if magneto-optical
effects are neglected (because the transfer equation for Ũ is decoupled from the
others, see Eq. (9.35)). As a rule, the sign of Ũ is the same at all wavelengths
and depends on the magnetic field polarity, being positive for 0◦ ≤ θ ≤ 90◦ and
negative for 90◦ ≤ θ ≤ 180◦ .1 Magneto-optical effects also change the shape of Q̃
by reducing, in general, the amplitude of its central depression and by moving the
lateral lobes closer to line center.
The influence of magneto-optical effects on the linear polarization profiles can also
be illustrated by diagrams like the one shown in Fig. 9.17. Proceeding from the line
center to the line wing, the representative point draws in the Q-U plane a curve
that is typical of magneto-optical effects. If these were neglected, the curve would
1
Exceptions to this simple rule can be found, especially for ‘exotic’ Zeeman patterns. In
some cases Ũ can even change sign across the line profile.
490 CHAPTER 9
Fig.9.16. The Stokes parameters profiles, as given by the Unno-Rachkovsky solutions (9.110),
are plotted against reduced wavelength for the normal Zeeman triplet 1 S0 − 1 P1 . The profiles
are normalized to B0 and the relevant parameters have the same values as in the ‘standard case’
(see Fig.9.5) except for vB = 2. Full line: with magneto-optical effects; dashed line: without
magneto-optical effects (U is identically zero).
reduce – for χ = const. – to a straight line through the origin, partially covered
twice, having angular coefficient tan 2χ. These diagrams, introduced by Kawakami
(1983), clearly show that the importance of magneto-optical effects decreases in
the wings.
iii) Finally, magneto-optical effects have also a (small) influence on the intensity
profile. By scrambling the polarization among Q, U , and V , they reduce the
importance of saturation, which eventually leads to a deeper profile and to a slight
increase of the equivalent width.
The influence of magneto-optical effects is even more pronounced on the net (line-
integrated) polarization. As evident from our thorough discussion in Sect. 9.20, net
linear polarization is dramatically affected by magneto-optical effects. Referring
for instance to Fig. 9.12, all the representative points would lie on the q̃-axis if
magneto-optical effects were neglected. Similar conclusions can be drawn for the
net circular polarization. In the absence of magneto-optical effects, the net circular
LINE FORMATION IN A MAGNETIC FIELD 491
Fig.9.17. Polarization diagram ( U (λ) vs. Q(λ) ) for the normal Zeeman triplet 1 S0 − 1 P1 . The
Stokes parameters are computed from the Unno-Rachkovsky solutions (9.110) with the ‘standard’
values of the relevant parameters (see Fig.9.5). The dots indicate distance from line center in
Doppler width units.
polarization obtained for small velocity gradient and weak magnetic field would
have the opposite sign (see the comments following Eq. (9.244)), and the ∆χ-effect
discussed at the end of Sect. 9.21 would simply disappear.
9.23. Transfer Equation for Fine-Structured and
Hyperfine-Structured Lines
The transfer equation that we have considered throughout this chapter refers to
an isolated spectral line originating in the transition between two atomic levels
whose magnetic splitting is described by the Zeeman effect regime. For a line
multiplet composed of different fine-structure or hyperfine-structure components,
the equation must be modified. The modifications do not concern, of course, the
form of the transfer equation – which is still given by Eq. (6.85) – but only its
coefficients: in other words, we have to find the new expressions for the elements
of the propagation matrix and of the emission vector.
Let us consider the multi-term model atom discussed in Sect. 7.5 embedded
in an arbitrary magnetic field. The radiative transfer coefficients, expressed in
the energy-eigenvector representation of the atomic density operator, are given by
Eqs. (7.35). Similarly to Sect. 9.1, we assume that only two atomic terms contribute
to absorption and emission at frequency ν (the analogue of the isolated spectral
line hypothesis), and that no atomic polarization is present in either term. The
former assumption implies that the summation over the terms in Eqs. (7.35) is
limited to one lower term (β L S) and one upper term (βu Lu S); the latter implies
492 CHAPTER 9
the substitutions
N
N (2L + 1) ρβ L S (j M , j M ) → δ δ
2S + 1 j j M M
Nu
N (2Lu + 1) ρβ L S (ju Mu , ju Mu ) → δ δ ,
u u 2S + 1 ju ju Mu Mu
where N and Nu are the overall number of atoms per unit volume in the lower and
upper term, respectively.
Using these assumptions, and recalling Eq. (5.148), Eq. (7.35a) reduces to the
following
= hν N
ηiA (ν, Ω) B(β L S → βu Lu S)
4π 2S + 1
j j
× 3 CJ (β L S, M ) CJ (β L S, M )
j J J ju Ju Ju M Mu q
j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1)
u u
! !
Ju J 1 Ju J 1 Lu L 1 Lu L 1
×
−Mu M −q −Mu M −q J Ju S J Ju S
φ(ν
× Tqq (i, Ω) − ν) ,
β L Sju Mu , β L Sj M
u u
which can be cast into a simpler form by introducing the normalized strengths of
the multiplet components defined in Eq. (3.64). Taking into account the symmetry
properties of the 3-j and 6-j coefficients, we have
= hν
ηiA (ν, Ω) N B(β L S → βu Lu S)
4π
j M ,j M
× φ(ν
Sq u u Tqq (i, Ω) − ν) .
β L Sju Mu , β L Sj M
u u
j ju M Mu q
Comparison with Eq. (9.2) – see also Eq. (3.16) – shows a remarkable similarity.
Since the same similarities are also found for the other radiative transfer coefficients,
we conclude that the transfer equation for polarized radiation in a fine-structure
multiplet can be obtained from the corresponding equation for an isolated spectral
line by performing the following formal substitutions
i) for the absorption coefficient:
kL , the frequency-integrated line absorption coefficient defined in Eqs. (9.14) and
(9.5), must be replaced by kM , the frequency-integrated multiplet absorption co-
efficient defined by
hν hν
A
kM = kM − kM
S
= N B(β L S → βu Lu S) − N B(βu Lu S → β L S)
4π 4π u
hν Nu ω̃
= N B(β L S → βu Lu S) 1 − ,
4π N ω̃u
LINE FORMATION IN A MAGNETIC FIELD 493
where ω̃ = (2L + 1)(2S + 1) is the degeneracy of the lower term and ω̃u =
(2Lu + 1)(2S + 1) the degeneracy of the upper term.
ii) for the source function:
SL , the line source function defined in Eqs. (9.14), must be replaced by SM , the
multiplet source function defined by
−1
2hν03 N ω̃u
SM = −1 ,
c2 Nu ω̃
where ν0 = νβ L S, β L S is the ‘central frequency’ of the multiplet defined in
u u
Eq. (7.22).
iii) for the strengths of the components:
J J
Sq u (M , Mu ), the normalized strength of the Zeeman component connecting
the lower level |α J M with the upper level |αu Ju Mu , must be replaced by
j M ,j M
Sq u u , the normalized strength of the multiplet component connecting the
lower level |β L Sj M with the upper level |βu Lu Sju Mu . This strength is given
by Eq. (3.64) and its explicit calculation requires the numerical diagonalization of
the spin-orbit plus magnetic Hamiltonians, as explained in Sect. 3.4.
iv) for the splitting of the components:
the splitting νL (gu Mu −g M ), associated with the transition |α J M → |αu Ju Mu ,
must be replaced by the splitting (νβ L Sj M , β L Sj M − ν0 ) associated with the
u u u u
transition |β L Sj M → |βu Lu Sju Mu . This quantity is defined in Eq. (3.68)
(see also Eq. (7.26)); its value is also deduced by diagonalization of the above
Hamiltonians.
Once these substitutions are carried out, any fine-structure multiplet can be
treated as a usual Zeeman multiplet, with the only difference that the structure of
the ‘equivalent Zeeman pattern’ is generally more complicated.
Hyperfine-structure multiplets can be handled in a strictly similar way. As no-
ticed in Sects. 3.5 and 7.9, there is indeed a very strict analogy – both physical and
formal – between fine-structure and hyperfine-structure multiplets. If we refer to
an atom having nuclear spin I and consider the transition between the lower level
(α J ) and the upper level (αu Ju ), the assumption of complete depolarization of
both levels implies
N
N (2J + 1) ρα J I (i f , i f ) → δ δ
2I + 1 i i f f
Nu
N (2Ju + 1) ρα J I (iu fu , iu fu ) → δ δ , (9.261)
u u 2I + 1 iu iu fu fu
where N and Nu are the overall number densities of atoms in the lower and upper
level, respectively, and where the density-matrix elements are defined in Eq. (7.63).
Using Eqs. (9.261), one easily finds that the transfer equation for polarized radi-
ation in a hyperfine-structure multiplet can be obtained from the corresponding
494 CHAPTER 9
equation for a non-structured spectral line by simple substitution of the appropri-
ate ‘equivalent Zeeman pattern’. The normalized strengths and the splittings of
the hyperfine components are given by Eqs. (3.75) and (3.78), respectively.
Most of the results obtained in this chapter can be applied both to fine-structure
and to hyperfine-structure multiplets. The most important difference is that the
‘equivalent Zeeman pattern’ of a multiplet does not show the characteristic sym-
metry properties of usual Zeeman patterns expressed by Eqs. (3.18) and (3.21).
This means that even in the constant velocity case, the symmetry properties (9.54)
of the Stokes parameters are no longer valid for multiplets, which eventually leads
to the appearance of net circular polarization, as noticed at the end of Sect. 9.21.
A remarkable property of the ‘equivalent Zeeman patterns’, concerning the fre-
quency shifts of the σr , σb , and π components, has been proved in Chap. 3.
It is expressed by Eq. (3.66) for fine-structure multiplets, and by Eq. (3.77) for
hyperfine-structure multiplets. An important consequence of this property can be
derived via a perturbative expansion similar to that of Sect. 9.6. Let us consider
a spectral line resulting from a multiplet whose components, in the absence of
magnetic field, have wavelength separations much smaller than the Doppler width
∆λD . This means, for fine-structure multiplets
hc
ζ ∆λD , (9.262)
λ20
and for hyperfine-structure multiplets
hc hc
A ∆λD , B ∆λD , (9.263)
λ20 λ20
where λ0 is the central wavelength of the multiplet and where ζ and A , B are
defined in Eqs. (3.59) and (3.70), respectively. In the presence of a weak magnetic
field (∆λB ∆λD ) one obtains, using the above-mentioned property, a relation
between the Stokes parameters V and I of the form of Eq. (9.80), namely
∂I
V = − ∆λB cos θ
∂λ
for fine-structure multiplets, and
∂I
V = − ∆λB ḡ cos θ ,
∂λ
with ḡ defined in Eq. (3.44), for hyperfine-structure multiplets. It follows that when
inequality (9.262) or (9.263) is satisfied, the presence of fine or hyperfine structure
does not change – relative to the case where the ‘inner structure’ is neglected – the
relation existing between V and I in the weak field regime.1
1
Note that this result is non-trivial and is not a direct consequence of the principle of spec-
troscopic stability. For this principle to be applicable, one should also assume ζ hc ∆λB /λ20
for fine-structure multiplets and A , B hc ∆λB /λ20 for hyperfine-structure multiplets. Were
these last inequalities satisfied, Eqs. (9.83), (9.84) and (9.85) would also be valid (with Ḡ = 1 for
fine-structure multiplets).
LINE FORMATION IN A MAGNETIC FIELD 495
Fig.9.18. The Stokes parameters profiles of the line λ 5420.362 of MnI are plotted against wave-
length distance from line center in m Å (solid line). They are computed according to Eqs.(9.110)
and are normalized to B0 . The relevant parameters are the following: ∆λD = 40 mÅ, B = 3000 G
(whence vB = 1.029), θ = 60◦ , χ = 0◦ , κL = 7, a = 0.05, βµ = 5. The ‘equivalent Zee-
man pattern’ is computed using the values of the hyperfine-structure constants given by Brix and
Kopfermann (1952). The dashed line refers to the Stokes parameters profiles computed neglecting
hyperfine structure.
Figure 9.18 shows the Stokes parameters profiles, as deduced from the Unno-
Rachkovsky solution, for a typical hyperfine-structure multiplet. Note the charac-
teristic broadening of the profiles and the ‘asymmetries’ introduced by hyperfine
structure.
9.24. Line Formation in Stochastic Media
In this chapter we have treated several aspects of the problem of line formation in a
magnetic field. In all cases, it was understood that the physical parameters affect-
ing the transfer of radiation through the ambient medium were deterministic: in
other words, they were assumed to take well-defined values in any volume element
496 CHAPTER 9
of the medium. Here we assume, on the contrary, that the medium is stochastic.
By this we mean that the physical parameters are random variables having, locally,
a certain probability distribution (with possible correlations between different pa-
rameters) and, non-locally, a typical autocorrelation length.
The idea of a non-deterministic ambient medium is indeed realistic. For instance,
the analysis of spectral lines in stellar atmospheres has shown for a long time
the existence of turbulent velocity fields: this suggests quite naturally that the
magnetic field (and other physical variables) may also be present in a turbulent
form. Furthermore, one can easily imagine the existence of correlation effects in
this turbulent medium. As an example, one might consider the possibility that
the magnetic field is larger-than-average when the turbulent velocity is directed
outward and smaller-than-average when it is directed inward (or vice versa).
Writing the transfer equation in its general form (9.163)
dI
= A(τr ) I − b(τr ) , (9.264)
dτr
one is faced with the problem of solving a linear stochastic equation, that is a linear
equation with random coefficients.1 The general problem is extremely complex;
however, it is possible to obtain a quite simple analytical solution under a set of
restrictive assumptions. Following Auvergne et al. (1973), who solved the simpler
problem of the transfer of unpolarized radiation in a turbulent velocity field, we
make the following hypotheses:
a) We assume the line is formed in LTE, with a Planck function linear in τr . This
allows us to write the emission coefficient b(τr ), defined in Eqs. (9.164), in the
form2
b(τr ) = BP (τr ) A(τr ) U = B0 (1 + β τr ) A(τr ) U . (9.265)
b) We assume the physical parameters {ζi } (with i = 1, . . . , p), on which the prop-
agation matrix A(τr ) depends – see the discussion at the beginning of Sect. 9.16 –
to have a stochastic distribution described by the function f (ζ1 , ζ2 , . . . , ζp ). The
probability dP of finding the first parameter in the interval (ζ1 , ζ1 +dζ1 ), the second
in the interval (ζ2 , ζ2 + dζ2 ), etc., is given by
dP = f (ζ1 , ζ2 , . . . , ζp ) dζ1 dζ2 · · · dζp , (9.266)
with
dζ1 dζ2 · · · dζp f (ζ1 , ζ2 , . . . , ζp ) = 1 .
1
The optical depth τr in Eq. (9.264) should be considered a deterministic variable.
For a stochastic medium where the continuum absorption coefficient at the reference frequency,
kc (νr ), is also a stochastic variable, the definition of τr given in Eqs. (9.33) must be replaced by
dτr = − kc (νr ) ds, where the meaning of the symbol · · · is explained below (see Eq. (9.271)).
2
The slope of the Planck function is here denoted by β to distinguish it from the
quantity β defined in Eq. (9.105). For a plane-parallel atmosphere with a constant ratio of the
absorption coefficients kc (νr ) and kc (ν0 ), the two quantities are related by the simple equation
β = βµ kc (ν0 ) / kc (νr ).
LINE FORMATION IN A MAGNETIC FIELD 497
c) We assume that each given ray path crosses a sequence of independent ‘eddies’.
The n-th eddy covers the optical depth interval (τr = τn−1 , τr = τn ), with τ0 = 0.
In each interval the physical parameters are constant and their values are drawn
at random from the probability distribution f (ζ1 , ζ2 , . . . , ζp ). Accordingly, the
propagation matrix in the n-th interval is constant and will be denoted by An .
Obviously there is no correlation between the matrix Ai in the i-th interval and
the matrix Aj in the j-th interval.
d) We finally assume the ‘jumping points’ τ1 , τ2 , . . . , τn , . . . to be distributed
according to a Poisson law characterized by the frequency 1/τe or, in other words,
by the mean length of the eddies τe . The probability dQ that the first jumping
point is found in the interval (τ1 , τ1 + dτ1 ), the second in the interval (τ2 , τ2 + dτ2 ),
etc., is given by
τ1 − τ0 τ2 − τ1 τn − τn−1
− − − dτ1 dτ2 dτ
dQ = e τe
e τe
··· e τe
··· ··· n ··· , (9.267)
τe τe τe
with
∞ ∞ ∞ τ −τ τ −τ τ −τ
dτ1 dτ2 dτn − 1τ 0 − 2τ 1 − n τ n−1
dQ = ··· ··· e e e e ··· e e ··· = 1 .
τe τe τe
τ0 τ1 τn−1
Hypotheses b), c), and d) characterize what is often referred to as a Poisson-step
or Kubo-Anderson process (see Brissaud and Frisch, 1971).
The transfer equation can be easily solved with the help of the assumptions above
(see Landi Degl’Innocenti, 1994). Let us first go back to Eq. (9.166), which in LTE
reduces to (see Eqs. (9.164))
∞
I(0) = O(0, τr ) A(τr ) BP (τr ) U dτr . (9.268)
0
Using the analogues of Eqs. (9.97)-(9.94) for the evolution operator O(τr , τr ) and
integrating by parts, we get
∞
dBP (τr )
I(0) = BP (0) 1 + O(0, τr ) dτr U,
dτr
0
and for a linear Planck function (assumption (a) )
∞
I(0) = B0 1 + β O(0, τr ) dτr U. (9.269)
0
498 CHAPTER 9
Consider now a particular realization of the jumping points (τ1 , τ2 , . . . , τn , . . .) and
of the propagation matrices in the various intervals (A1 , A2 , . . . , An , . . .). Using
the analogues of Eqs. (9.95) and (9.103), we have
∞ ∞
τn
O(0, τr ) dτr = O(0, τr ) dτr
0 n=1 τ
n−1
∞
τn
−(τ1 − τ0 ) A1 −(τ2 − τ1 ) A2 −(τn−1 − τn−2 ) An−1 −(τr − τn−1 ) An
= e e ··· e e dτr .
n=1 τn−1
The integral in the right-hand side can be evaluated via Eq. (9.106). Substitution
into Eq. (9.269) yields
∞
−(τ − τ ) A −(τ − τ ) A −(τ −τ )A
I(0) = B0 1 + β e 1 0 1 e 2 1 2 · · · e n−1 n−2 n−1
n=1
!
−(τn − τn−1 ) An
× 1 − e A−1
n U. (9.270)
We now multiply both sides by the probability dQ defined in Eq. (9.267) and inte-
grate over all possible realizations of the jumping points. Using again Eq. (9.106),
we obtain
∞
I(0) dQ = B0 1 + β 1 + τe A1 )−1 (1
(1 1 + τe A2 )−1 · · · (1
1 + τe An−1 )−1
n=1
!
1 + τe An )−1 A−1
× 1 − (1 n U.
Finally, we average over the distribution of the physical parameters {ζi }. Let
us denote by g the mean value of any given function g (ζ1 , ζ2 , . . . , ζp ) over the
distribution function f (ζ1 , ζ2 , . . . , ζp ) defined in Eq. (9.266),
g = dζ1 dζ2 · · · dζp g (ζ1 , ζ2 , . . . , ζp ) f (ζ1 , ζ2 , . . . , ζp ) . (9.271)
Since by assumption c) the values of the physical parameters in different eddies are
uncorrelated, we obtain for the statistical average of the emerging Stokes parame-
ters1
∞
n−1
I(0) av = B0 1 + β 1 + τe A)−1
(1
n=1
!
× A−1 − (1
1 + τe A)−1 A−1 U. (9.272)
1
Note that the term corresponding to n = 1 in the sum over n is the identity matrix 1 .
LINE FORMATION IN A MAGNETIC FIELD 499
This expression can be further developed as follows. Given an arbitrary matrix B,
1 − B)−1 into power series of B and use of the relation
expansion of (1
1 − B)−1 (1
(1 1 − B) = 1
yields
∞
1 − B)−1 =
(1 Bk .
k=0
Hence the sum over n in Eq. (9.272) can be written in the form
∞
n−1 −1
1 + τe A)−1
(1 = 1− 1 + τe A)−1
(1
n=1
−1 1 −1
1 + τe A)−1
= 1 − (1 = 1 + τe A)−1
A (1 . (9.273)
τe
On the other hand, we have
−1
A − (11 + τe A)−1 A−1 = A−1 − A−1 (1
1 + τe A)−1
= A−1 1 − (1
1 + τe A)−1 1 + τe A)−1 .
= τe (1
Substitution of these relations into Eq. (9.272) leads to the expression
−1
I(0) av
= B0 1 + β A (1
1 + τe A)−1 1 + τe A)−1
(1 U. (9.274)
This formula contains, as particular cases, some interesting limits:
i) If the physical parameters are deterministic, the symbols · · · can be removed
and one obtains
I(0) av = B0 1 + β A−1 U , (9.275)
which is nothing but the Unno-Rachkovsky solution.1
ii) Microturbulent limit: under the limit τe → 0 (microscopic eddies) one gets
−1
I(0) av
= B0 1 + β A U, (9.276)
or, in other words, the Unno-Rachkovsky solution of the deterministic case with
the matrix A replaced by A.
iii) Macroturbulent limit: under the limit τe → ∞ (very large eddies) one obtains
I(0) av
= B0 1 + β A−1 U , (9.277)
1
The connection with the Unno-Rachkovsky solution will be discussed more precisely in the
following.
500 CHAPTER 9
which is again the Unno-Rachkovsky solution with the matrix A−1 replaced by
A−1 .
iv) Finally, for non-magnetic media the matrix A is proportional to the unit matrix,
A = (κr + κR η) 1 = κ 1 , (9.278)
with η defined in Eqs. (9.65). The intensity (the other Stokes parameters are
obviously zero) is given by
8 9<8 9
1 κ
I(0) av = B0 1 + β , (9.279)
1 + τe κ 1 + τe κ
a formula derived by Auvergne et al. (1973) for the case of turbulent velocity fields.
For a plane-parallel atmosphere, we can rewrite Eq. (9.274) in a form more similar
to the Unno-Rachkovsky solution (9.109) by identifying the reference frequency νr
with the line frequency ν0 . Let us denote by te the mean length of the eddies
measured along the vertical to the atmosphere. Bearing in mind footnote 2 on
p. 496, the emerging Stokes profiles for a radiation beam travelling in the µ-direction
are given by
8 , -−1 9−1 8, -−1 9
t t
I(0) av
= B0 1 + βµ C 1 + e C 1 + e C U , (9.280)
µ µ
where β and C are defined in Eqs. (9.105) and (9.91), respectively.1
Figures 9.19 and 9.20 show some numerical results for the case where all the
physical parameters are deterministic except for the magnetic field intensity and the
line-of-sight component of the velocity field, which are characterized by a bimodal
distribution of the form (Martin, 1971)
−α
e
f (vB , vA ) = , (9.281)
2π 1 − ρ2 ∆vB ∆vA
where
2 2
(0) !
1 1 vB − vB
(0)
1 vA − vA
(0)
vB − vB
(0)
vA − vA
α= + −ρ .
1 − ρ2 2 ∆vB 2 ∆vA ∆vB ∆vA
In these equations, vB and vA are defined by Eqs. (9.26), the Doppler width ∆λD
being due to the thermal velocity only. The quantity ρ is the correlation parameter
1
Note, however, that stochastic variations of the continuum absorption coefficient cannot be
considered if the matrix C has the form of Eq. (9.91). Such variations can be allowed for if C in
Eq. (9.280) is defined to be kc (ν0 )
C = 1 + κL H .
kc (ν0 )
LINE FORMATION IN A MAGNETIC FIELD 501
Fig.9.19. Stokes parameters profiles computed from Eq.(9.280) for the Zeeman triplet 1 S0 − 1 P1
and the bimodal distribution (9.281). The profiles are normalized to B0 . The values of the
(0) (0)
parameters specifying the distribution are: ρ = 0, vB = 1.5, ∆vB = 0.5, vA = 0, ∆vA = 0.5.
◦ ◦
The values of the remaining parameters are: κL = 20, θ = 60 , χ = 0 , a = 0.05, βµ = 5. Full
line: microturbulent limit (te = 0); dashed line: macroturbulent limit (te = 1000).
(−1 < ρ < 1). When ρ = 0 the distribution function reduces to the product of two
(0)
Gaussian distributions, one for vB (centered at vB and having standard deviation
(0)
∆vB ) and the other for vA (centered at vA and having standard deviation ∆vA ).
By simple algebra it can be proved that the distribution (9.281) is normalized to
unity,
∞ ∞
dvB dvA f (vB , vA ) = 1 .
−∞ −∞
Both figures refer to the normal Zeeman triplet 1 S0 − 1 P1 . Figure 9.19 gives the
emerging Stokes profiles computed for ρ = 0 (absence of correlations). Comparison
with panel (a) of Fig. 9.5 clearly shows the broadening of the profiles introduced
by the Gaussian distributions of magnetic field intensity and of turbulent velocity.
The remarkable effects of correlations are shown in Fig. 9.20, computed for ρ = 0.7.
Correlations between magnetic field intensity and line-of-sight velocity are able to
502 CHAPTER 9
Fig.9.20. Same as Fig.9.19 with ρ = 0.7.
produce asymmetric profiles and a definite amount of net circular polarization
(except in the macroturbulent limit te → ∞).
According to Eq. (9.265), the Planck function – hence the temperature of the
medium – has been assumed to be a deterministic variable: therefore, correlations
involving temperature (like temperature-velocity or temperature-magnetic field)
cannot be treated via Eqs. (9.274) or (9.280). A simple generalization of these
equations, which allows for such correlations, can be obtained by writing the Planck
function in the form
BP (τr ) = B0 (1 + β τr ) + ∆BP , (9.282)
where ∆BP satisfies assumptions b), c) and d) above – in other words, is itself a
Kubo-Anderson process.
Let us denote by (∆B1 , ∆B2 , . . . , ∆Bn , . . .) a particular realization of ∆BP in
the intervals (τ0 , τ1 ), (τ1 , τ2 ), . . . , (τn−1 , τn ), . . . . Following the same line of
reasoning, it is easily seen that the additional term ∆BP in Eq. (9.282) makes an
LINE FORMATION IN A MAGNETIC FIELD 503
additional contribution to Eq. (9.268) of the form
∞
τn
Ia.c. (0) = O(0, τr ) A(τr ) ∆Bn U dτr .
n=1 τ
n−1
Using Eqs. (9.97), (9.95) and (9.103) we obtain
∞
−(τ1 − τ0 ) A1 −(τ2 − τ1 ) A2 −(τn−1 − τn−2 ) An−1
Ia.c. (0) = ∆Bn e e ··· e
n=1
−(τn − τn−1 ) An
× 1 − e U,
which is the additional contribution to Eq. (9.270). Integration over the Poisson
distribution of the jumping points and over the distribution of the physical param-
eters yields
∞
n−1
Ia.c. (0) av
= ∆B (11 + τe A)−1
n=1
n−1
1 + τe A)−1
− (1 1 + τe A)−1 U .
∆B (1
Since
1 + τe A)−1 = τe ∆B A (1
∆B 1 − ∆B (1 1 + τe A)−1 ,
we finally obtain, with the help of Eq. (9.273)
−1
Ia.c. (0) av
1 + τe A)−1
= A (1 1 + τe A)−1 U .
∆B A (1 (9.283)
This formula contains the following special cases (cf. Eqs. (9.275)-(9.279)):
i) Deterministic limit
Ia.c. (0) av
= ∆B U .
ii) Microturbulent limit
−1
lim Ia.c. (0) av
= A ∆B A U .
τe →0
iii) Macroturbulent limit
lim Ia.c. (0) av
= ∆B U .
τe →∞
504 CHAPTER 9
iv) Scalar case
8 9<8 9
∆B κ κ
Ia.c. (0) av
= U,
1 + τe κ 1 + τe κ
where κ is defined in Eq. (9.278).
Adding Eqs. (9.274) and (9.283) we obtain the expression
−1
I(0) av
= B0 1 + 1 + τe A)−1
A (1
× (B0 β 1 + ∆B A) (1
1 + τe A)−1 U, (9.284)
which generalizes Eq. (9.274) by allowing for the stochastic behavior of the Planck
function.1
The procedure followed in this section to find an analytical solution to the transfer
equation (9.264) for a Kubo-Anderson process can also be applied to solve the
homogeneous transfer equation
dI
= A(τr ) I . (9.285)
dτr
The calculations are developed in App. 12 and lead to a closed analytical form for
the stochastic average of the evolution operator.
It should be mentioned that the problem of line formation in a stochastic magnetic
field has been tackled, with a completely different approach, by Faulstich (1980).
This approach is a generalization of the method developed by Gail et al. (1980)
for the non-polarized case and is based on the assumption that the intensity of the
magnetic field is constant while its direction changes along the ray path according
to a Markov process. The stochastic average of the emerging Stokes parameters is
I) of finding,
found by solving a Fokker-Planck equation for the probability P (x, Ω,
at point x along the ray path, the magnetic field in the solid angle dΩ about the
and the Stokes parameters in the interval (I, I + dI).
direction Ω
9.25. Isotropic, Microturbulent Magnetic Field
In the former section and in App. 12 we have found that, under the microturbulent
limit, radiative transfer for polarized radiation can be simply described by replac-
ing the propagation matrix A with its stochastic average A (see Eqs. (9.276) and
(A12.8)). This result is valid well beyond the set of assumptions outlined in the
1
Note that a stochastic behavior of the Planck function implies a stochastic behavior of
the temperature, which obviously affects the propagation matrix A. In other words, the symbols
· · · in Eq. (9.284) mean an average over a distribution function f (ζ1 , ζ2 , . . . , ζp ) which includes
temperature among its arguments.
LINE FORMATION IN A MAGNETIC FIELD 505
former section. In fact, it holds whenever the correlation lengths of the stochas-
tic physical parameters are much smaller than the typical scales of the radiative
transfer problem (determined by the eigenvalues of the propagation matrix), irre-
spective of the type of process – Kubo-Anderson, Markov, etc. – characterizing the
stochastic behavior (see e.g. Stenflo, 1971).
Here we want to consider the effect of a microturbulent, isotropic magnetic field
on the propagation matrix, under the assumption that the field modulus B, char-
acterized by a normalized distribution f (B), is the only stochastic variable (except
for microturbulent velocity, assumed uncorrelated with it).1
Since the continuum and line absorption coefficients are not affected by the mag-
netic field, the average propagation matrix reduces to (see Eqs. (9.164))
A = κr 1 + κR H .
On the other hand, the off-diagonal elements of the matrix H vanish because of
the isotropy of the magnetic field.2 From Eqs. (9.50) and (9.32), being
cos θ = sin2 θ cos 2χ = sin2 θ sin 2χ = 0
and
2 2 2 1
sin θ = , cos θ = ,
3 3
we get
1
H = hI 1 = ηb + ηp + ηr 1 , (9.286)
3
where
∞
ηb,p,r = ηb,p,r (B) f (B) dB . (9.287)
0
If we now take for f (B) a Gaussian distribution characterized by the r.m.s. value Bt
(the ‘turbulent’ field),
. 2
2 B 2 − 21 B
Bt
f (B) = e , (9.288)
π Bt3
we can easily find an analytical expression for the matrix H.
Let us first assume the turbulent field is weak, that is (cf. Eqs. (9.26))
(t) λ20 e0 Bt
∆λB ≡ ∆λD , (9.289)
4π mc2
1
More complicated magnetic field distributions have been considered by Dolginov and Pavlov
(1972).
2
It follows that the radiation emerging from a semi-infinite atmosphere where a microturbulent,
isotropic magnetic field is present, is unpolarized.
506 CHAPTER 9
with ∆λD the Doppler width. From Eqs. (9.64) and (9.68) we have, up to second-
order terms in ∆λB
1 (2)
q η − Gq ∆λB η +
ηq = G(0) (1)
G ∆λ2B η (q = −1, 0, +1) , (9.290)
2 q
where η is defined in Eqs. (9.65), η and η in Eqs. (9.70), and
λ20 e0 B
∆λB = ∆λD .
4π mc2
Recalling Eqs. (9.29), and using Eqs. (9.290) and (9.288), the integrals in Eq.
(n)
(9.287) can be performed. Substituting the expressions for Gq given in Table 3.4
we finally obtain
(t) 2
hI = η + Ḡt ∆λB η , (9.291)
where Ḡt , the second-order effective Landé factor for turbulent fields, is given by
1 (2)
Ḡt = G + G(2) 2
1 = ḡ + δ ,
2 0
with1
1 2
δ = g (4s − d2 − 4) .
16 d
It can be easily shown that δ is a non-negative quantity that vanishes only for
Zeeman triplets (normal or anomalous). Consequently, Ḡt is also a non-negative
quantity.
Equation (9.291) can be cast into a more significant form. Let us define a ‘total
Doppler width’ ∆λT as
(t) 2
∆λT = ∆λ2D + 4 Ḡt ∆λB ,
(t)
where ∆λB satisfies the condition (9.289). Using the relation
∂ 2 H(v, a) ∂H(v, a) ∂H(v, a)
= − 2 H(v, a) + v +a ,
∂v 2 ∂v ∂a
which follows directly from Eqs. (5.58), it can be easily proved that, up to second-
(t)
order terms in ∆λB , we can write
∆λD 1 λ − λ0 a ∆λD
hI = √ H , ,
∆λT π ∆λT ∆λT
1
Note that δ = (δ + 3 δ ) / 2, where δ and δ are defined in Eqs. (9.78) and (9.227),
respectively.
LINE FORMATION IN A MAGNETIC FIELD 507
where λ0 is defined in Eq. (9.44). We conclude that a weak, turbulent magnetic
field described by the distribution (9.288) produces a broadening of the absorption
profile which is the same as the broadening produced by a velocity field of amplitude
λ0 e0 Bt
weq = Ḡt ,
2π mc
or, numerically
−3 λ0
weq = 1.4 × 10 Ḡt Bt ,
5000
where λ0 is expressed in Å, Bt in G, and weq in km s−1 .
For arbitrary values of Bt , when inequality (9.289) is no longer satisfied, it is
still possible to find an analytical expression for hI . Let us consider the case of
a Zeeman triplet. Bearing in mind Eqs. (9.29), (9.27a) and (9.26), we have from
Eqs. (9.286)-(9.288)
1 λ − λ0
hI = √ H ,a
3 π ∆λD
√ ∞
2 λ − λ0 + g ∆λB λ − λ0 − g ∆λB
+ H , a +H ,a
3π ∆λD ∆λD
0
2
(t) 2
∆λB −1
2 ∆λB / ∆λB
d (∆λB )
× (t)
e (t)
. (9.292)
∆λB ∆λB
Since the integrand is an even function of ∆λB , the integral can be evaluated by
extending it from −∞ to +∞ and dividing by 2. Then we make use of the relation
y
2 1 y 2
−2 y ∂2 y−x
e y0
= H √ 2
, 0 + y0 lim H √ ,0 ,
y0 2 y0 x→0 ∂x2 2 y0
which can be proved with the help of Eqs. (5.55). The evaluation of the integral
in Eq. (9.292) is thus reduced to the evaluation of convolutions of Voigt functions
with different widths, which can be easily done – using the symmetry property
(5.59) – via Eqs. (5.65). The final result is
1 λ − λ0 2 ∆λD λ − λ0 a ∆λD
hI = √ H ,a + √ H ,
3 π ∆λD 3 π ∆λT ∆λT ∆λT
2 ∆λD 2 (t) 2 ∂
2
λ − λ0 a ∆λD
+ √ g ∆λ H , ,
3 π ∆λT B
∂λ2 ∆λT ∆λT
where
(t) 2
∆λT = ∆λ2D + 2 g 2 ∆λB .
This formula could be generalized to more complicated Zeeman patterns, but we
will not develop the relevant calculations.
CHAPTER 10
NON-EQUILIBRIUM ATOMIC PHYSICS
Let us consider a typical astrophysical plasma (like, for instance, a stellar atmo-
sphere) and let us focus our attention on a particular atom belonging to a given
volume element of the plasma. The atom experiences a continuous series of in-
teractions with the photons propagating inside the plasma, with the surrounding
particles and, if the plasma is magnetized, with the magnetic field. Except for a
few very special cases, the atom will be in a typical non-equilibrium situation: the
distribution of level populations will differ from the Saha-Boltzmann distribution,
and each individual level will have a definite amount of atomic polarization. The
aim of this chapter is to describe the main aspects of these non-equilibrium phe-
nomena, by showing how the most important atomic properties – like the emissivity
and the absorptivity – depend in a critical way on the physical characteristics of
the environment.
Once such physical characteristics are specified, the state of the atom is obtained
by solving the statistical equilibrium equations. In stationary situations, these
equations form a system of linear, algebraic equations whose unknowns are the
atomic density-matrix elements (or those linear combinations of density-matrix
elements representing the multipole moments or spherical statistical tensors). The
number of equations (and of unknowns) depends on the model atom considered,
the coefficients depend both on the model atom and on the physical environment
(characteristics of the incident radiation field, presence or absence of a magnetic
field and/or of collisions). In practice, one is faced with very complicated equations
even for relatively simple model atoms and physical situations. Therefore, although
the calculation of the atomic state under given physical conditions is in principle a
trivial task (it just requires the solution of a system of linear, algebraic equations),
it is in fact extremely difficult to understand the role played by the different factors
in determining such state.
In the general case, there is no way out of this situation: once the model atom
and the physical environment are given, the statistical equilibrium equations have
to be solved numerically. This is however of little help to understand the physics
of the different phenomena. Therefore, rather than illustrating a variety of sample
cases obtained by numerical solutions to the equations, we consider in this chapter
a number of simplified physical situations, by introducing certain approximations
which, in many cases, lead to analytical expressions for the atomic density-matrix
elements.
The basic approximation used throughout this chapter is to consider model atoms
having just two energy states, an upper and a lower state, between which transi-
tions take place. All the three model atoms described in Chap. 7 (multi-level,
multi-term, and multi-level with hyperfine structure) will be studied, but they will
510 CHAPTER 10
be ‘restricted’ to two-level or two-term model atoms. The only exceptions are
Sects. 10.10, 10.11, and 10.12, where three-level model atoms are considered in
order to point out some peculiar phenomena.
The second, frequently used approximation is to neglect the atomic polarization
of the lower level (or term). In general, this is a reasonable assumption provided
depolarizing collisions are present and the radiation field is sufficiently weak (see
Sect. 10.7). Accordingly, stimulation effects are systematically neglected whenever
this approximation is made.
Two further assumptions introduced in some sections concern the radiation field.
Instead of an arbitrary field, we sometimes consider either an unpolarized field
having cylindrical symmetry about a given direction, or a weakly anisotropic field.
Both cases lead to major simplifications in the statistical equilibrium equations;
on the other hand, such radiation fields are rather common both in laboratory and
astrophysical situations.
As far as collisions (either inelastic/superelastic or depolarizing) and the magnetic
field are concerned, we will treat a number of different situations, in order to
illustrate their specific effects on the atomic density matrix. However, collisions
will only be considered when dealing with the multi-level atom, because the theory
of collisions presented in Chap. 7 was not explicitly developed for the multi-term
model atom.
In order to make the study of individual sections easier, the various assumptions
are briefly recalled by subheadings or ‘keywords’ at the beginning of each section.
Whenever analytical expressions for the atomic density matrix are worked out,
they can be substituted into the expressions of the radiative transfer coefficients.
This yields explicit formulae relating, for instance, the atomic emissivity in the four
Stokes parameters to the radiation field impinging on the atom and to the other
physical properties of the environment (collisions, magnetic field). In several cases
such relations turn out to be linear, in the sense that the emission coefficient in
the i-th Stokes parameter for the direction Ω is linearly related to the j-th Stokes
parameter of the radiation impinging on the atom from the direction Ω . This
provides the possibility of defining suitable redistribution matrices for scattering
processes. Such procedure was first introduced by Landi Degl’Innocenti (1984).
The validity of the results derived in this chapter is obviously subjected to the
limitation outlined in the introduction to Chap. 6 and thoroughly discussed in
Sect. 6.5. This limitation concerns the spectral characteristics of the radiation
field that is illuminating the atom, which must satisfy the so-called flat-spectrum
approximation.
10.1. The Two-Level Atom: Generalities
We consider a simple atomic model consisting of two levels: a lower and an upper
level characterized by the quantum numbers1 (α , J ) and (αu , Ju ), respectively.
1
The spectroscopic notation is the same as that used throughout the whole book. J is the
NON-EQUILIBRIUM ATOMIC PHYSICS 511
The atom is interacting with an anisotropic, polarized radiation field character-
and any frequency ν, by the Stokes vector I (ν, Ω).
ized, for any given direction Ω 1
i
It is also interacting with a collection of perturbers (or colliders) whose veloc-
ity distribution is assumed to be isotropic and Maxwellian, and which produce
elastic, inelastic, and superelastic collisions (see Sect. 7.13). A magnetic field is
present, with associated Larmor frequency νL (see Eq. (3.10)) and direction Ω .
B
We adopt the flat-spectrum approximation by assuming that the radiation field has
no spectral structure across a frequency interval ∆ν centered at the frequency ν0
(corresponding to the energy separation between the two levels) and larger both
than νL and than the width (in frequency units) of the levels.
We fix a reference system having the z-axis in the magnetic field direction, and
we describe the atom by the multipole moments of the density matrix. Collecting
the results on the radiative and collisional rates derived in Chap. 7 (see Eqs. (7.11)
and (7.101), and the discussion in Sects. 7.13d and 7.13e), one gets for the time
evolution of the multipole moments of the upper level
d K
ρ (α J ) = −2πi νL gα J Q ρK Q (αu Ju )
dt Q u u u u
+ TA (αu Ju KQ, α J K Q ) ρK
Q (α J )
K Q
− RE (αu Ju KQK Q ) + RS (αu Ju KQK Q ) ρK
Q (αu Ju )
K Q
2J + 1 (K)
+ C (αu Ju , α J ) ρK
Q (α J )
2Ju + 1 I
− CS(0) (α J , αu Ju ) + D(K) (αu Ju ) ρK
Q (αu Ju ) , (10.1)
and for the lower level
d K
ρ (α J ) = −2πi νL gα J Q ρK Q (α J )
dt Q
+ TE (α J KQ, αu Ju K Q ) + TS (α J KQ, αu Ju K Q ) ρK
Q (αu Ju )
K Q
− RA (α J KQK Q ) ρK
Q (α J )
K Q
2Ju + 1 (K)
+ C (α J , αu Ju ) ρK
Q (αu Ju )
2J + 1 S
− CI(0) (αu Ju , α J ) + D(K) (α J ) ρK
Q (α J ) . (10.2)
total (orbital + spin) angular momentum quantum number of the electronic cloud, while α is a
collection of inner quantum numbers (see Sect. 3.1).
1
Throughout this chapter the Stokes vector will be denoted by the symbol Ii (with i = 0, 1, 2, 3)
rather than Si used in preceding chapters.
512 CHAPTER 10
All the rates appearing in these equations have been introduced and discussed
in Chap. 7 (see Sect. 7.2 for the radiative rates and Sect. 7.13 for the collisional
rates). Here we just recall that TA , TE , and TS are the radiative transfer rates due
to absorption, spontaneous emission and stimulated emission, respectively, while
RA , RE , and RS are the corresponding relaxation rates. CI and CS are the rates due
to inelastic and superelastic collisions, respectively. Finally, D are the relaxation
rates due to depolarizing elastic collisions.
Equations (10.1) and (10.2) form a system of linear, homogeneous, differential
equations in the unknowns ρK K
Q (α J ) and ρQ (αu Ju ). In stationary situations, the
left-hand sides of both equations are zero and the system reduces to a set of linear,
homogeneous, algebraic equations. The number of equations (and of unknowns) is
given by
Neq = (2J + 1)2 + (2Ju + 1)2 . (10.3)
On the other hand, the determinant of the system is zero,1 so that the multipole
moments can be determined up to a multiplicative factor. This factor can be
found by considering the trace equation, which for the two-level atom reads (see
Eqs. (3.84) and (3.108))
2J + 1 ρ00 (α J ) + 2Ju + 1 ρ00 (αu Ju ) = 1 , (10.4)
or, introducing the number density of atoms N
N + Nu = N , (10.5)
where
N = N 2J + 1 ρ00 (α J ) , Nu = N 2Ju + 1 ρ00 (αu Ju ) . (10.6)
Once the statistical equilibrium equations are solved, the radiative transfer coeffi-
cients can be found from Eqs. (7.15).
In general, it is impossible to obtain a closed analytical expression for the solution
to Eqs. (10.1)-(10.2). An important exception is provided by the case where the
lower level is unpolarized (or naturally populated) so that its multipole moments
are given by
ρK 0
Q (α J ) = δK0 δQ0 ρ0 (α J ) . (10.7)
This relation is approximately satisfied when the lower-level radiative rates are
much smaller than the lower-level collisional rates, which is often true in the pres-
ence of a very weak (or ‘diluted’) radiation field (see the discussion in Sect. 10.7).
Obviously, Eq. (10.7) is identically satisfied in the special case where the lower level
1
It can easily be shown, with the help of Eqs. (7.14), (2.36a) and (2.49), that the equations
for the time derivatives of
2Ju + 1 ρ00 (αu Ju ) and − 2J + 1 ρ00 (α J )
are identical.
NON-EQUILIBRIUM ATOMIC PHYSICS 513
has J = 0, since in that case ρ00 (α J ) is the only non-zero multipole moment (cf.
Eq. (3.101)).
Consider the statistical equilibrium equations for the upper level (Eq. (10.1))
under the assumption that Eq. (10.7) – the unpolarized lower level hypothesis –
is valid. In stationary situations, and neglecting the rate for stimulated emission,
we immediately get an analytical expression for ρK Q (αu Ju ). Bearing in mind the
expression for RE (Eq. (7.14e)) we obtain
2J +1
TA (αu Ju KQ, α J 0 0) + 2Ju +1 CI(0) (αu Ju , α J ) δK0 δQ0
ρK
Q (αu Ju ) =
2πi νL gα J Q + A(αu Ju → α J ) + CS(0) (α J , αu Ju ) + D(K) (αu Ju )
u u
× ρ00 (α J ) , (10.8)
where, from Eq. (7.14a) and using Eqs. (2.26a) and (2.49)
TA (αu Ju KQ, α J 0 0) = 3(2J + 1) B(α J → αu Ju )
!
1 1 K
× (−1)1+J +Ju +Q K
J−Q (ν0 ) . (10.9)
Ju Ju J
In the following we will apply Eq. (10.8) to different physical regimes, characterized
by the relative order of magnitude of the different parameters.
10.2. The Two-Level Atom: Resonance Polarization
(unpolarized lower level - no magnetic field - no collisions)
The ideal case of resonance polarization is obtained by neglecting in Eq. (10.8) both
the effect of the magnetic field and the effect of collisions. Setting in Eq. (10.8)
νL = CI(0) (αu Ju , α J ) = CS(0) (α J , αu Ju ) = D(K) (αu Ju ) = 0 ,
and substituting Eq. (10.9), we have1
B(α J → αu Ju )
ρK
Q (αu Ju ) = 3(2J + 1)
A(αu Ju → α J )
!
1 1 K
× (−1)1+J +Ju +Q K
J−Q (ν0 ) ρ00 (α J ) . (10.10)
Ju Ju J
It is convenient to introduce the symbol wJ(K)J defined by2
u
!
(K) 1+J +Ju 1 1 K
wJ J = (−1) 3(2Ju + 1) , (10.11)
u Ju Ju J
1
Note that Eq. (10.10) is valid in an arbitrary reference system, since the magnetic field is
zero.
2
The symbol was first introduced by Landi Degl’Innocenti (1984).
514 CHAPTER 10
or, equivalently (see Eq. (2.36a))
!
1 1 K
Ju Ju J
wJ(K)J = ! . (10.12)
u 1 1 0
Ju Ju J
With this new symbol, Eq. (10.10) takes the more compact form
2J + 1 B(α J → αu Ju ) (K)
ρK
Q (αu Ju ) = w (−1)Q J−Q
K
(ν0 ) ρ00 (α J ) . (10.13)
2Ju + 1 A(αu Ju → α J ) Ju J
This expression shows that the upper-level multipole moment ρK Q (αu Ju ) is propor-
tional to the corresponding multipole component of the radiation field tensor.1 As
the latter has a limited number of components (K ≤ 2, see Eq. (5.155)), it follows
that only the multipole moments ρK Q (αu Ju ) with K ≤ 2 can be excited by the
radiation field.
Equation (10.13) also points out the meaning of the symbol wJ(K)J . It is a sort
u
of ‘efficiency factor’ characterizing the transfer of the K-th order multipole from
the radiation field to the atomic density matrix in the absorption process from the
lower level (α J ) to the upper level (αu Ju ). The symbol wJ(K)J is defined for any
u
dipole transition (Ju − J = 0, ±1 ; 00) with K ranging from 0 to 2. For K = 0
one obviously has, from Eq. (10.12)
wJ(0)J = 1 . (10.14)
u
Numerical values of wJ(K)J for K = 1 and K = 2 are given in Table 10.1, together
u
with the values of other symbols that will be introduced later on.
Using the expression of the upper-level statistical tensors given in Eq. (10.13),
the emission coefficient can be easily deduced. Since by assumption the magnetic
field is zero, the emission coefficient is given by Eq. (7.16e), which can be rewritten
in the form2
= hν N 2J + 1 A(α J → α J )
εi (ν, Ω) u u u
4π
(K)
× wJ J T Q
K ρK (α J ) φ(ν − ν) ,
(i, Ω) (10.15)
u
Q u u 0
KQ
1
The sign difference in the subscript Q and the factor (−1)Q arise from the fact that JQ K
is a true irreducible tensor (which, under rotation of the reference system, changes according to
Eq. (2.78)), while ρK K ∗
Q is not (it changes according to Eq. (3.98)). The conjugate (ρQ ) – and
not ρKQ itself – is an irreducible tensor. The conjugation property (3.102) explains the above
differences.
2
The right-hand side should indeed be multiplied by the factor ν 3 /ν03 (cf. footnote on p.283).
NON-EQUILIBRIUM ATOMIC PHYSICS 515
TABLE 10.1
Values of the quantities wJ(K)
J
, WK (J , Ju ), (pQ )max , |pV |max for different transitions
u
3 3
J Ju wJ(1)J W1 (J , Ju ) wJ(2)J W2 (J , Ju ) pQ 3p V 3
u u max max
0 1 −1. 1. 1. 1. 1. 1.
1/2 1/2 −0.816 0.667 0. 0. 0. 1.
1/2 3/2 −0.913 0.833 0.707 0.5 0.429 1.
1 0 0. 0. 0. 0. 0. 0.
1 1 −0.5 0.25 −0.5 0.25 0.2 0.333
1 2 −0.866 0.75 0.592 0.35 0.288 0.957
3/2 1/2 0.408 0.167 0. 0. 0. 0.25
3/2 3/2 −0.365 0.133 −0.566 0.32 0.261 0.172
3/2 5/2 −0.837 0.7 0.529 0.28 0.226 0.921
2 1 0.5 0.25 0.1 0.01 0.008 0.373
2 2 −0.289 0.083 −0.592 0.35 0.288 0.106
2 3 −0.816 0.667 0.490 0.24 0.191 0.893
5/2 3/2 0.548 0.3 0.141 0.02 0.015 0.446
5/2 5/2 −0.239 0.057 −0.605 0.366 0.302 0.072
5/2 7/2 −0.802 0.643 0.463 0.214 0.170 0.871
3 2 0.577 0.333 0.169 0.029 0.022 0.493
3 3 −0.204 0.042 −0.612 0.375 0.310 0.053
3 4 −0.791 0.625 0.443 0.196 0.155 0.854
7/2 5/2 0.598 0.357 0.189 0.036 0.027 0.526
7/2 7/2 −0.178 0.032 −0.617 0.381 0.316 0.04
7/2 9/2 −0.782 0.611 0.428 0.183 0.144 0.840
4 3 0.612 0.375 0.204 0.042 0.032 0.551
4 4 −0.158 0.025 −0.620 0.385 0.320 0.031
4 5 −0.775 0.6 0.416 0.173 0.136 0.828
9/2 7/2 0.624 0.389 0.216 0.047 0.035 0.570
9/2 9/2 −0.142 0.020 −0.623 0.388 0.322 0.025
9/2 11/2 −0.769 0.591 0.407 0.165 0.129 0.819
5 4 0.632 0.4 0.226 0.051 0.039 0.585
5 5 −0.129 0.017 −0.624 0.39 0.324 0.021
5 6 −0.764 0.583 0.399 0.159 0.124 0.811
11/2 9/2 0.640 0.409 0.234 0.055 0.041 0.597
11/2 11/2 −0.118 0.014 −0.626 0.392 0.326 0.018
11/2 13/2 −0.760 0.577 0.392 0.154 0.12 0.804
6 5 0.645 0.417 0.240 0.058 0.044 0.607
6 6 −0.109 0.012 −0.627 0.393 0.327 0.015
6 7 −0.756 0.571 0.387 0.149 0.116 0.798
516 CHAPTER 10
TABLE 10.1
(continued)
3 3
J Ju wJ(1)J W1 (J , Ju ) wJ(2)J W2 (J , Ju ) pQ 3p V 3
u u max max
13/2 11/2 0.650 0.423 0.246 0.060 0.046 0.616
13/2 13/2 −0.101 0.010 −0.628 0.394 0.328 0.013
13/2 15/2 −0.753 0.567 0.382 0.146 0.113 0.792
7 6 0.655 0.429 0.251 0.063 0.048 0.623
7 7 −0.094 0.009 −0.628 0.395 0.328 0.011
7 8 −0.75 0.562 0.377 0.142 0.111 0.788
15/2 13/2 0.658 0.433 0.255 0.065 0.050 0.630
15/2 15/2 −0.089 0.008 −0.629 0.395 0.329 0.010
8 7 0.661 0.437 0.259 0.067 0.051 0.635
8 8 −0.083 0.007 −0.629 0.396 0.329 0.009
where Eqs. (10.11) and (7.8) have been used. We recall that N is the number
density of atoms, and that the tensor T Q K is defined in Eq. (5.159) – see
(i, Ω)
Table 5.6 for the explicit expressions of its components. Substitution of Eq. (10.13)
into Eq. (10.15) leads, with the help of Eqs. (10.6), to the expression
= hν N B(α J → α J ) φ(ν − ν)
εi (ν, Ω)
4π u u 0
× WK (J , Ju ) (−1)Q T Q
K J K (ν ) ,
(i, Ω) (10.16)
−Q 0
KQ
where the new symbol WK (J , Ju ) is defined to be (see Eq. (2.36a))
, -2
WK (J , Ju ) = wJ(K)J
u !2
1 1 K
!2
1 1 K Ju Ju J
= 3(2Ju + 1) = !2 . (10.17)
Ju Ju J 1 1 0
Ju Ju J
Numerical values of WK (J , Ju ) for K = 1 and K = 2 are given in Table 10.1. For
K = 0, we simply have
W0 (J , Ju ) = 1 . (10.18)
Equation (10.16) gives, for the two-level atom under the approximations consid-
ered in this section, the Stokes parameters of the radiation emitted in a scattering
process as a function of the incident radiation field. A still more significant expres-
sion can be obtained by writing the radiation field tensor in terms of the Stokes
parameters of the incoming radiation. Bearing in mind the definition of kLA , the
NON-EQUILIBRIUM ATOMIC PHYSICS 517
frequency-integrated line absorption coefficient defined in Eq. (9.5), we obtain via
Eqs. (5.157)
5
dΩ
3
= k A φ(ν − ν)
εi (ν, Ω) ) I (ν , Ω
Ω
P (Ω, ) , (10.19)
L 0
4π j=0 ij j 0
Ω
where Pij (Ω, ), the (quantum-mechanical) scattering phase matrix, is given by
Ω
Pij (Ω, ) = WK (J , Ju ) (−1)Q T Q
K
(i, Ω) ) .
T K (j, Ω (10.20)
−Q
KQ
This result, first derived by Hamilton (1947) – see also Chandrasekhar (1950)1 –
generalizes to Quantum Mechanics the expression for the scattering phase matrix
Ω
Rij (Ω, ) that we have obtained by a ‘classical’ derivation (see Sects. 5.7 and
5.10). The relation between Pij (Ω, ) and R (Ω,
Ω ) can easily be established.
Ω
ij
From Eqs. (5.137)-(5.138) we have
(K)
Ω
Rij (Ω, ) ≡ R (Ω, ; 0) =
Ω Rij (Ω, Ω ; 0)
ij
K
= (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω (10.21)
−Q
K Q
hence
Ω
Pij (Ω, ) = (K)
WK (J , Ju ) Rij (Ω, Ω ; 0) . (10.22)
K
In other words, the quantum-mechanical scattering phase matrix is obtained by
summation of the multipole components of the classical matrix multiplied by
the corresponding WK factors. It is interesting to note that for the transition
(J = 0, Ju = 1) all the WK are unity, and that such a property is valid only for
this transition (see Table 10.1). We thus find a further confirmation of the fact that
(J = 0, Ju = 1) is the only transition properly described by the classical theory.
) have been proved in Chap. 5. These
Ω
Various properties of the matrix Rij (Ω,
properties can be easily extended to the matrix Pij (Ω, Ω ). First of all, P (Ω, )
Ω
ij
is real: this is obvious from Eq. (10.19) – where all the other quantities are real –
but can directly be checked from the definition in Eq. (10.20) with the help of
Ω
Eqs. (5.158). From Eq. (10.20) it follows also that Pij (Ω, ) satisfies the Helmoltz
principle of reciprocity for a scattering process,
Ω
Pij (Ω, ) = P (Ω , Ω)
,
ji
1
Chandrasekhar introduces the symbols E1 , E2 , and E3 instead of W1 and W2 . The
relationship is the following
E1 = W 2 , E2 = 1 − W 2 , E3 = W 1 .
518 CHAPTER 10
Ω Θ
scattering plane
Fig.10.1. Geometry of a scattering process. The unit vectors ea and ea , which define the positive Q
direction for the incident and the scattered beam, respectively, are perpendicular to the scattering
plane (Ω , Ω).
which generalizes Eq. (5.95). Finally, from Eq. (10.22), using the second expression
in Eq. (5.138) – with H = 0 – and Eq. (2.74), we obtain
Ω ) = ∗
P (i) tP (j) DP P (R) ,
WK (J , Ju ) tK K K
Pij (Ω, (10.23)
KP P
where the symbol tK
is defined in Eq. (5.160) and R is the rotation carrying
P
e (Ω),
the reference frame (ea (Ω), Ω) into the frame (e (Ω
), e (Ω
), Ω
). Equation
b a b
(10.23) is the quantum-mechanical generalization of Eq. (5.168).
The physical meaning of the symbol WK (J , Ju ) can be better illustrated by
considering the polarization of the emitted radiation in a scattering process. Let
us consider the scattering geometry of Fig. 10.1. The phase matrix can be obtained
from Eqs. (10.22) and (5.139) by setting in the latter1 µ = µ = 0, C1 = cos Θ,
S1 = − sin Θ, C2 = cos 2Θ, S2 = − sin 2Θ. One gets, with the use of Eq. (10.18)
Ω
P (Ω, ) =
⎛ ⎞
1 − 14 W2 (1 − 3 cos2 Θ) 3
4 W2 sin2 Θ 0 0
⎜ 3
sin2 Θ 3 2 ⎟
⎜ 4 W2 4 W2 (1 + cos Θ) 0 0 ⎟
=⎜ ⎟. (10.24)
⎝ 0 0 3
2 W2 cos Θ 0 ⎠
3
0 0 0 2 W1 cos Θ
For an unpolarized incident beam, the scattered radiation is found to be linearly
polarized with fractional polarization given by
εQ (ν, Ω) P (Ω, Ω ) 3 W2 sin2 Θ
pQ ≡ = 10 =
εI (ν, Ω) Ω
P00 (Ω, ) 4 − W2 + 3 W2 cos2 Θ
εU (ν, Ω) P (Ω, Ω )
pU ≡ = 20 =0. (10.25)
εI (ν, Ω) Ω
P00 (Ω, )
1
We recall that Eqs. (5.139) refer to the geometry of Fig. 5.10. The geometry of Fig. 10.1
is recovered by setting θ = θ = 90◦ , χ − χ = Θ. The corresponding values of µ, µ , C1 , S1 , C2 ,
S2 are then obtained from Eqs. (5.84), (5.98) and (5.97).
NON-EQUILIBRIUM ATOMIC PHYSICS 519
TABLE 10.2
Analytical expressions of the quantities WK (J , Ju ), (pQ )max , |pV |max
3 3
W1 (J , Ju ) W2 (J , Ju ) pQ 3p V 3
max max
J +2 (2J + 5)(J + 2) (2J + 5)(J + 2) 5 (J + 2)(2J + 1)
Ju = J + 1
2 (J + 1) 10 (J + 1)(2J + 1) 26J 2 + 37J + 10 14J 2 + 23J + 10
1 (2J − 1)(2J + 3) (2J − 1)(2J + 3) 5
Ju = J
2 J (J + 1) 10 J (J + 1) 12J 2 + 12J + 1 8J 2 + 8J − 1
J −1 (2J − 3)(J − 1) (J − 1)(2J − 3) 5 (J − 1)(2J + 1)
Ju = J − 1
2J 10 J (2J + 1) 26J 2 + 15J − 1 14J 2 + 5J + 1
Since W2 values range from 0 to 1, pQ is always positive, which means that the
polarization is perpendicular to the scattering plane. Its maximum value is found
for Θ = 90◦ and is given by
3 W2
pQ = . (10.26)
max 4 − W2
This is an increasing function of W2 . The factor W2 can thus be regarded as an
efficiency factor for the generation of linear polarization from unpolarized radiation
in scattering processes.
The parameter W1 , on the contrary, is only related to circular polarization. In
the geometry of Fig. 10.1, assuming that the incident radiation is totally circu-
larly polarized (say positively), the fractional circular polarization of the scattered
radiation is given by
εV (ν, Ω) P (Ω, Ω ) 6 W1 cos Θ
pV ≡ = 33 = .
εI (ν, Ω)
P00 (Ω, Ω ) 4 − W2 + 3 W2 cos2 Θ
Since W1 is non-negative, the handedness of the scattered circular polarization is
the same as that of the incident radiation for Θ < 90◦ , the opposite for Θ > 90◦
( pV vanishes for Θ = 90◦ ). The maximum absolute value of pV is found for Θ = 0◦
(forward scattering) and for Θ = 180◦ (backward scattering), and is given by
3 3 3 W1
3p 3 = .
V max
2 + W2
This formula shows that W1 can be considered as an efficiency factor for the gen-
eration of circular (from circular) polarization in scattering processes.
Numerical values of (pQ )max and |pV |max for different transitions are given in
Table 10.1. It is important to notice that W2 – hence (pQ )max – vanishes for any
transition having Ju = 0 or Ju = 1/2. On the contrary, W1 – hence |pV |max –
vanishes only for the transition (J = 1, Ju = 0).
Table 10.2 gives the analytical expressions of W1 , W2 , (pQ )max and |pV |max for
dipole transitions. Such expressions can be obtained from Eq. (10.17) with the
help of Eqs. (2.36d) and (2.36h).
520 CHAPTER 10
10.3. The Two-Level Atom: the Hanle Effect
(unpolarized lower level - no collisions)
The basic characteristics of the Hanle effect are obtained from Eq. (10.8) by retain-
ing the ‘magnetic’ term (first term in the denominator, which was set to zero in
the preceding section).1 In order to further simplify our treatment, we still neglect
– as in the preceding section – the effects of collisions (this matter is deferred until
Sect. 10.6). Using Eqs. (10.9) and (10.11), we can rewrite Eq. (10.8) in the form
2J + 1 B(α J → αu Ju )
ρK
Q (αu Ju ) =
2Ju + 1 A(αu Ju → α J ) + 2πi νL gα J Q
u u
× wJ(K)J (−1)Q J−Q
K
(ν0 ) ρ00 (α J ) . (10.27)
u
We recall that the statistical tensors are defined in a reference system having the
z-axis in the magnetic field direction. Comparison with the corresponding expres-
sion obtained for the non-magnetic regime (Eq. (10.13)) shows that the effect of the
magnetic field is to reduce and dephase, via the imaginary term in the denominator,
all the multipole moments with Q = 0. The efficiency of this relaxation process is
controlled by the dimensionless parameter Hu defined by
2πνL gα J
u u
Hu = . (10.28)
A(αu Ju → α J )
This parameter is the quantum-mechanical analogue of the parameter H introduced
in the classical derivation of the Hanle effect (see Eq. (5.84)).2 Numerically, one
has (cf. Eq. (5.85))
0.879 gα J B
u u
Hu = , (10.29)
A(αu Ju → α J )
where B is in G and A(αu Ju → α J ) in 107 s−1 .
Using Hu , Eq. (10.27) can be expressed in the form
1
ρK
Q (αu Ju ) = ρK (α J ) , (10.30)
1 + i QHu Q u u B=0
where [ρKQ (αu Ju )]B=0 are the multipole moments for the non-magnetic case (in
the presence of the same radiation field and defined in the same reference sys-
tem). Equation (10.30) summarizes the ‘essence’ of the Hanle effect. It shows
that coherences with Q = 0 (those connected with the populations of the Zeeman
sublevels, see Eq. (3.101)) are unaffected by the magnetic field, while those with
1
Note that we are still using the unpolarized lower level assumption and we are neglecting,
accordingly, stimulated emission effects.
2
The classical damping constant γ, defined in Eq. (5.28), has its obvious quantum-mechanical
counterpart in the Einstein coefficient A.
NON-EQUILIBRIUM ATOMIC PHYSICS 521
Q = 0 are reduced, with respect to the non-magnetic case, by a factor 1 + Q2 Hu2 .
The magnetic field has the additional effect of introducing a phase factor, given
by arctan(QHu ), in the latter. Such conclusions were anticipated in Sect. 5.13
where an intuitive description of the Hanle effect, based on classical physics, was
presented.
The radiation emitted by the atom along a given direction is obtained by substi-
tution of Eq. (10.27) into Eq. (7.15e) – adapted to the case of our two-level atom.
The resulting expression is quite involved, and will be studied in detail in the next
section. Here we disregard the spectral details of the emitted radiation, and we
restrict attention to the frequency-integrated emission coefficient, defined by
=
ε̃i (Ω) dν ,
εi (ν, Ω)
∆ν
where the interval ∆ν is sufficiently broad to fully cover all the Zeeman components
of the line. Taking into account that all the profiles Φ(να J M , α J M − ν) are
u u u
normalized to unity in frequency (see Eqs. (6.59a-c)), we obtain, along the same
lines leading from Eq. (7.15e) to Eq. (7.16e)
2 4
= h ν N (2J + 1) B(α J → α J )
ε̃i (Ω) u u u
2πc2
√ !
1 1 K ρK (α J ) ,
× 3 (−1)1+J +Ju
TQ
K
(i, Ω) Q u u
Ju Ju J
KQ
or, with the use of Eqs. (7.8) and (10.11)
hν (K)
=
ε̃i (Ω) N 2Ju + 1 A(αu Ju → α J ) wJ J T Q
K ρK (α J ) .
(i, Ω) (10.31)
Q u u
4π u
KQ
Substitution of Eq. (10.27) into Eq. (10.31) yields, with the help of Eqs. (5.157),
(9.5), (10.6), (10.17) and (10.28)
5
dΩ
3
= kA
ε̃i (Ω) P (Ω, ; B)
Ω ) ,
I (ν , Ω (10.32)
L
4π j=0 ij j 0
Ω
where Pij (Ω, ; B),
the quantum-mechanical scattering phase matrix in the pres-
ence of a magnetic field, is given by1
1
Ω
Pij (Ω, ; B)
= WK (J , Ju ) (−1)Q T Q
K
(i, Ω) )
T K (j, Ω . (10.33)
−Q
1 + i QHu
KQ
1
The expression in Eq. (10.33) was first derived by Landi Degl’Innocenti (1985c).
522 CHAPTER 10
This expression generalizes to Quantum Mechanics the result that we have derived
in Chap. 5 from the classical theory of the electron (Eq. (5.133)). The difference
from the classical result is twofold: the factor WK comes into play, and the classical
parameter H is replaced by Hu (note that both WK and Hu depend on the atomic
transition). As expected, the quantum-mechanical and the classical phase matrix
coincide for the transition (J = 0, Ju = 1) with gα J = 1 (normal Zeeman triplet),
u u
provided the classical damping constant γ is identified with the Einstein coefficient
A(αu Ju → α J ) – we recall that WK (0, 1) = 1 for K = 0, 1, 2 (see Table 10.1).
In general, the quantum-mechanical scattering matrix is formally related to the
multipole components of the classical matrix by the expression
Ω
Pij (Ω, ; B)
= (K)
WK (J , Ju ) Rij (Ω, Ω ; B) , (10.34)
K
(K)
where Rij (Ω, Ω ; B) is defined as in Eq. (5.138) with H replaced by Hu .
Several properties proved in Chap. 5 for the classical scattering phase matrix
are valid for the quantum-mechanical matrix as well. It can easily be shown that
Ω
Pij (Ω, ; B)
is real and satisfies the relation1
Ω
Pij (Ω, ; −B)
= P (Ω , Ω;
B)
, (10.35)
ji
analogous to Eq. (5.94). It can also be shown, using Eq. (10.34), that the value of
the Van Vleck angle (Eq. (5.100)) is independent of the atomic transition – except
of course for Ju = 0 or Ju = 1/2 since in those cases the linear polarization of
the scattered radiation is zero. This property is basically due to the fact that the
values of W2 are limited between 0 and 1. The other properties proved in Sect. 5.12
for the classical matrix are easily extended to the quantum-mechanical matrix. In
particular, we can consider the average of the scattering matrix over an isotropic
distribution of magnetic fields, defined by
2π π
Ω
Pij (Ω, = 1
; B) Ω
dχB sin θB Pij (Ω, ; B)
dθ ,
B
4π
0 0
where θB and χB are the polar and azimuth angle, respectively, of the magnetic
field vector in a given reference frame (see Fig. 5.9). Along the same lines of
Sect. 5.12 we obtain
Ω
Pij (Ω, ; B)
= (K)
WK (J , Ju ) µK Rij (Ω, Ω ; 0) , (10.36)
K
(K)
where the multipole components Rij (Ω, Ω ; 0) are defined in Eq. (10.21) and the
quantities µK are given by Eq. (5.172) with the substitution H → Hu . Equation
(10.36) is the quantum analogue of Eq. (5.174); as expected, the two expressions
1
Equation (10.35) can be proved with the help of Eqs. (5.137), (5.138), (5.165) and (5.169).
NON-EQUILIBRIUM ATOMIC PHYSICS 523
Fig.10.2. Polarization (or Hanle) diagram for the transition (J = 1, Ju = 2), relative to the scat-
tering process illustrated in Fig.5.11. Full lines correspond to β = const., dashed lines to constant
magnetic field strength, parameterized through the angle α2 (see Eqs.(10.28) and (10.38)). Note
the scale difference from the classical Hanle diagram of Fig.5.12.
coincide for the transition (J = 0, Ju = 1) with gα J = 1 (see Table 10.1 and
u u
footnote 2 on p. 520).
In order to illustrate the Hanle effect in a specific case, let us consider the scat-
tering geometry of Fig. 5.11. The polarization of the scattered radiation is given
by Eq. (10.32) with the phase matrix of Eq. (10.34). The multipole components
(K)
Rij (Ω, Ω ; B) are given by Eqs. (5.139).1 The resulting fractional (frequency-
integrated) linear polarization is found to be
ε̃Q (Ω) 3 W2 sin2 β + (1 + cos2 β) cos2 α2
p̃Q ≡ =
ε̃ (Ω)
I
8 + W2 (1 − 3 cos2 β − 3 sin2 β cos2 α2 )
ε̃U (Ω) 6 W2 cos β sin α2 cos α2
p̃U ≡ = , (10.37)
ε̃I (Ω) 8 + W2 (1 − 3 cos2 β − 3 sin2 β cos2 α2 )
where
tan α2 = 2Hu . (10.38)
Figure 10.2 shows the polarization diagram predicted by Eqs. (10.37) for the tran-
sition (J = 1, Ju = 2), which has a W2 value of 0.35 (see Table 10.1). Note that
1
Note that the scattering geometry of Fig. 5.11 is a special case of the geometry of Fig. 5.10
(which is implied by Eqs. (5.139)) corresponding to µ = cos β, µ = 0, χ − χ = 90◦ . The values
of µ, µ , C1 , S1 , C2 , S2 are given by Eqs. (5.97) and (5.98) – in the latter, H is replaced by Hu .
524 CHAPTER 10
the diagram is quite similar to the classical Hanle diagram shown in Fig. 5.12 (cf.
Eqs. (5.102)), apart from a scale factor due to the smaller polarizing efficiency of
the transition (the maximum fractional polarization, corresponding to zero mag-
netic field, is 21/73 0.288 – see Table 10.2 – instead of 1). The shape of the Hanle
diagrams can be appropriately characterized by the relative height, r, of the ‘cen-
ter’ of the diagram (defined as the point corresponding to α2 = 90◦ , β = 90◦ ). The
parameter r is just the ratio of the scattered polarization for infinitely large mag-
netic field to the polarization for zero field, in the 90◦ scattering of an unpolarized
beam with the magnetic field perpendicular to the scattering plane,
p̃Q (α2 = 90◦ , β = 90◦ )
r= .
p̃Q (α2 = 0◦ )
From Eqs. (10.37) we have
4 − W2
r= ,
8 + W2
thus r varies from 1/3 for W2 = 1 to 1/2 for W2 → 0 (minimum polarizing effi-
ciency). For the transition (J = 1, Ju = 2) we get r 0.437.
10.4. The Two-Level Atom: Spectral Details of the Hanle Effect
(unpolarized lower level - no collisions)
In this section we analyze in some detail the frequency dependence of the emitted
radiation in the Hanle effect. It is important to recall that the results presented
in the following are valid under the flat-spectrum approximation for the incident
radiation field. As in the preceding section, we assume that the lower level is
unpolarized and we neglect collisions.
The emission coefficient in the four Stokes parameters, as a function of frequency,
is obtained by substitution of Eq. (10.27) into Eq. (7.15e). Before performing the
substitution, it is convenient to rewrite the latter in the form
= hν N 2J + 1 A(α J → α J )
εi (ν, Ω) u u u
4π
× TQ
K ρK (α J ) ΦKK (J , J ; ν) ,
(i, Ω) (10.39)
Q u u Q u
KK Q
where the generalized profile ΦKK
Q (J , Ju ; ν) is given by1
1
The symbol was introduced by Landi Degl’Innocenti et al. (1991).
NON-EQUILIBRIUM ATOMIC PHYSICS 525
ΦKK
Q (J , Ju ; ν) = 3(2Ju + 1)(2K + 1)(2K + 1)
1+Ju −Mu + q Ju J 1 Ju J 1
× (−1)
−M u M −q −M u M −q
Mu Mu M qq
Ju Ju K 1 1 K
×
Mu −Mu −Q q −q −Q
1
× Φ(να J M , α J M − ν) + Φ(να J M , α J M − ν)∗ . (10.40)
2 u u u u u u
Equation (10.39) shows that the generalized profile is a frequency-dependent cou-
pling coefficient connecting the (K, Q) multipole moment of the atomic density
matrix with the (K , Q) multipole component of the emitted radiation. Its main
properties, including the proof of Eq. (10.39), are given in App. 13.
Substitution of Eq. (10.27) into Eq. (10.39) yields, with the help of Eqs. (5.157),
(9.5), (10.6) and (10.28)
= kA
εi (ν, Ω) ΦKK (J , Ju ; ν)
L Q
KK Q
5
dΩ (K)
3
K )
T K (j, Ω 1 ) . (10.41)
× w (−1)Q T Q (i, Ω) −Q I (ν , Ω
4π j=0 Ju J 1 + i QHu j 0
Obviously, Eq. (10.41) reduces to Eq. (10.32) after integration over frequency (see
Eqs. (A13.3) and (10.17)).
A remarkable result can be deduced from Eq. (10.41): provided the width of
the lower level is much smaller than the width of the upper level (which requires a
sufficiently weak incident radiation field), the Hanle effect vanishes in the far wings
of the line. In other words, resonance scattering in the line wings is unaffected by
the presence of the magnetic field.
To prove this property, we go back to Eq. (6.59a). Neglecting frequency shifts,
we can write
1 1
Φ(να J M , α J M − ν) = ,
u u u π Γ − i (να J M , α J M − ν)
u u u
where (see Eq. (7.3))
να J Mu , α J M = ν0 + νL (gα J Mu − gα J M ) .
u u u u
It follows that
1
Φ(να J M , α J M − ν) + Φ(να J M , α J M − ν)∗ =
2 u u u u u u
1 2Γ + i νL gα J (Mu − Mu )
=
u u
. (10.42)
2π Γ − i (να J Mu ,α J M − ν) Γ + i (να J M , α J M − ν)
u u u u u
526 CHAPTER 10
In the far wings ( |ν0 − ν| Γ and |ν0 − ν| νL ) we obtain
1
Φ(να J M , α J M − ν) + Φ(να J M , α J M − ν)∗ ≈
2 u u u u u u
νL gα J 1 Γ
≈ 1+i u u
(Mu − Mu ) .
2Γ π Γ 2 + (ν0 − ν)2
Substitution into Eq. (10.40) yields, owing to the presence of the third 3-j symbol
νL gα J 1 Γ
ΦKK (J , Ju ; ν) ≈ 1 + i Q u u
Q
2Γ π Γ 2 + (ν0 − ν)2
× 3(2Ju + 1)(2K + 1)(2K + 1)
1+Ju −Mu + q Ju J 1 Ju J 1
× (−1)
−M u M −q −M u M −q
Mu Mu M qq
Ju Ju K 1 1 K
× .
Mu −Mu −Q q −q −Q
The summation can easily be evaluated (see the derivation of Eq. (A13.3)). If the
width of the lower level is negligible compared to the width of the upper level, we
have from Eqs. (6.59b) – see footnote 2 on p. 520
A(αu Ju → α J )
Γ = , (10.43)
4π
and recalling Eq. (10.28) we finally obtain
ΦKK
Q (J , Ju ; ν) ≈ δKK wJ(K)J (1 + i QHu ) φ(ν0 − ν) .
u
Substitution into Eq. (10.41) shows that the dependence on the magnetic field
disappears – in fact, Eq. (10.41) reduces to Eq. (10.19). This proves that the Hanle
effect vanishes in the line wings.
To get a deeper insight into Eq. (10.41), let us consider a specific scattering
process. We refer again to the geometry of Fig. 5.11, and we set β = 0◦ (the
magnetic field points in the direction of the outgoing radiation) and denote by I
and ∆Ω , respectively, the intensity and the (infinitesimal) angular spread of the
incoming, unpolarized beam. The tensors T Q K
can be evaluated from Table 5.6.1
For the sake of simplicity we consider the transition (J = 0, Ju = 1) with gα J = 1.
u u
1
A possible choice for the angles appearing in Table 5.6 is: θ = 0◦ , χ, γ = 90◦ for the emitted
radiation; θ = 90◦ , χ = χ for the incident radiation. An equivalent choice – cf. footnote on
p.523 – is: θ = 0◦ , χ, γ = 0◦ ; θ = 90◦ , χ = χ − 90◦ .
NON-EQUILIBRIUM ATOMIC PHYSICS 527
Fig.10.3. The emission coefficients of Eqs.(10.44), normalized to (3/8) kL A
∆Ω I , are plotted
in units of 108 s−1 against the ‘reduced’ frequency (ν0 − ν)/Γ . The relevant parameters are:
A(αu Ju → α J ) = 5 × 107 s−1 , Hu = 1 (hence Γ = 3.98 × 106 s−1 , B = 5.69 G). The dashed
lines represent the values obtained from frequency-integrated observations. Note that the full
frequency range corresponds to 1 mÅ for a line at 5000 Å.
After some algebra we obtain, with the help of Eqs. (A13.11) and Table 10.1
= 3 k A ∆Ω I φ + φ
ε0 (ν, Ω) −1
8 L 1
= 3 k A ∆Ω I
ε1 (ν, Ω)
1
(φ + φ1 ) −
2Hu
(ψ − ψ1 )
8 L 1 + 4Hu2 −1 1 + 4Hu2 −1
= 3 k A ∆Ω I
ε2 (ν, Ω)
2Hu
(φ + φ1 ) +
1
(ψ − ψ1 )
8 L 1 + 4Hu2 −1 1 + 4Hu2 −1
= − 3 k A ∆Ω I φ − φ ,
ε3 (ν, Ω) (10.44)
−1
8 L 1
where the profiles φ−1 , φ1 , ψ−1 , ψ1 are defined in Eqs. (A13.10).
The emission coefficients in the four Stokes parameters predicted by Eqs. (10.44)
are shown in Fig. 10.3. One can note the presence of circular polarization – due
528 CHAPTER 10
Fig.10.4. The linear polarization degree pL and the rotation angle α (in degrees) corresponding
to the emission coefficients of Fig.10.3. The dashed lines have the same meaning as in Fig.10.3.
to the Zeeman effect – and the complex shape of the linear polarization profiles.
From Eqs. (10.44) it can easily be shown that the scattered radiation is totally
polarized at every single frequency. Some features of the linear polarization are
better appreciated in Fig. 10.4, where the polarization degree pL and the rotation
angle α, given by
,ε -
ε21 + ε22 1
pL = , α = arctan 2 + α0
ε0 2 ε1
(with α0 defined as in Eqs. (1.8b,c)) are plotted against frequency. At line center
the polarization degree is unity while the direction of maximum polarization is
rotated through a large angle ( 63◦ ) from the zero-field direction. In the far
wings, pL tends again to unity while α tends to zero, according to the property
proved earlier that the Hanle effect vanishes. In fact, the asymptotic values of the
fractional emission coefficients derived from Eqs. (10.44) for |ν0 − ν| → ∞ are
pQ = 1 , pU = p V = 0 .
10.5. The Two-Level Atom: Resonance Polarization
for Strong Magnetic Fields
(unpolarized lower level - no collisions)
In this section we consider the interaction of a two-level atom with the radiation
field in the presence of a strong magnetic field, in order to illustrate an important,
general concept: atomic polarization can be produced not only by the anisotropy
and/or polarization of the incident radiation, but also by the spectral structure of
the radiation itself. For the sake of simplicity we assume (as in the former sections)
the lower level to be unpolarized and we neglect collisions.
NON-EQUILIBRIUM ATOMIC PHYSICS 529
In the strong field regime, the Zeeman splitting is much larger than the natural
line width. This implies Hu 1 (see Eq. (10.28)) so that, according to Eq. (10.30),
the upper-level statistical tensors with Q = 0 can be neglected.1 In other words, all
coherences are zero and atomic polarization can only involve differences between
the populations of Zeeman sublevels (cf. Eq. (3.101)). Under this approximation
Eq. (10.27) reduces to
2J + 1 B(α J → αu Ju ) (K) K
ρK
Q (αu Ju ) = δQ0 w J (ν ) ρ0 (α J ) .
2Ju + 1 A(αu Ju → α J ) Ju J 0 0 0
Substitution into Eq. (10.39) gives the corresponding expression for the emission
coefficient. Using Eqs. (5.157), (9.5) and (10.6) we obtain
= kA
εi (ν, Ω) ΦKK (J , Ju ; ν)
L 0
KK
5
dΩ (K)
3
× ) I (ν , Ω
T K (j, Ω
wJ J T 0K (i, Ω) ) ,
0 j 0
4π j=0 u
which is a special case of Eq. (10.41).
The above derivation assumes the incident radiation field to be flat across the
whole line width, i.e., across a frequency interval ∆ν larger than the overall Zeeman
splitting of the line. Since in the strong-field regime coherences between Zeeman
sublevels are negligible, we can investigate, still within the flat-spectrum approxi-
mation, the more general case where the incident radiation is different in different
Zeeman components, yet being practically constant (over frequency intervals of
the order of the natural line width) across each component. Such case has been
analyzed in Sect. 7.4.
To handle this physical situation, it is convenient to use the standard (rather than
the statistical tensor) representation of the density matrix. In this representation
the atom is described by the diagonal density-matrix elements ραJ (M ) giving the
population of the magnetic sublevel M of level (αJ) – cf. Eq. (3.91b).
The relevant statistical equilibrium equations are given by Eq. (7.17). We write
such equations for the upper level of a two-level atom neglecting stimulated emis-
sion. Taking into account the expressions for the rates tA and rE (Eqs. (7.18a,e))
we obtain, in stationary situations
2
(2J + 1) B(α J → αu Ju ) Ju J 1
ρα (Mu ) = 3
u u
J
A(αu Ju → α J ) −Mu M −q
M q
× Jqq (να J Mu , α J M ) ρα J (M ) . (10.45)
u u
1
Obviously, this statement holds only in a reference system having the z-axis along the magnetic
field direction. It should be remarked that neglecting the statistical tensors with Q = 0 is, in any
case, an approximation which makes impossible to account for certain phenomena typical of line
wings (like the disappearance of the Hanle effect described in the previous section).
530 CHAPTER 10
Introducing the number density of atoms N , the assumption of unpolarized lower
level gives (see Eqs. (3.84), (3.89) and (10.5))
N 1
ρα (M ) = .
J
N 2J + 1
Substitution into Eq. (10.45) yields, with the use of Eqs. (7.8) and (5.153)
2
2Ju + 1 N c2 Ju J 1
ρα J (Mu ) = 3
u u 2J + 1 N 2hν03 −Mu M −q
M q
5
dΩ
3
× ) I (ν
T (j, Ω
j αu Ju Mu , α J M , Ω ) . (10.46)
4π j=0 qq
This expression gives the populations of the Zeeman sublevels of the upper level as
determined by a radiation field which can be anisotropic, polarized, and frequency-
dependent (within the limitations specified above). Its implications are best illus-
trated by comparison with the extreme case where the radiation field is isotropic,
unpolarized, and flat across the whole frequency interval spanned by the spectral
line,
) = δ J ,
Ij (να J M , α J M , Ω
u u u
j0
where J is the field intensity. In such case the integral over the solid angle can be
evaluated with the help of Eqs. (5.146), (5.128) and (5.143),
5
dΩ ) = 1 ,
T (0, Ω
4π qq 3
and the summation over M and q can be performed via Eq. (2.23a). The result is
N 1 c2
ρα (Mu ) = J,
u u
J
N 2J + 1 2hν03
which shows that all the Zeeman sublevels of the upper level are evenly populated
(unpolarized upper level).
If any of the above conditions is not met, a certain amount of atomic polariza-
tion (population unbalances) will form in the upper level, because of a mechanism
of selective pumping of the different Zeeman sublevels. There may exist selec-
tive pumping due to anisotropy (and/or polarization) of the incident radiation,
or selective pumping due to spectral structure of the incident radiation, or both.
Obviously, the second mechanism is effective only when the Zeeman sublevels are
split because of the presence of a magnetic field. In the preceding sections of this
chapter only the first mechanism has been considered.1
1
It should be remarked that the strong-field regime analyzed in this section is one of the
few cases where a frequency-structured radiation field can be treated within the flat-spectrum
approximation (such possibility is related to the fact that the atomic density matrix is diagonal).
If the Zeeman separation is comparable with the natural line width and the radiation field is
significantly structured across that width, the theory developed in this book cannot be applied
(see Sect. 6.5).
NON-EQUILIBRIUM ATOMIC PHYSICS 531
χ Ω
θ Ω
θ scattered
Ω radiation
Ω
incident
radiation
, is scattered into the direction Ω.
Fig.10.5. An unpolarized radiation beam, having direction Ω
The direction of the magnetic field B is specified by the angles θ and χ in the reference frame
e (Ω),
(ea (Ω), Ω).
The angle between Ω and B
is θ .
b
Let us apply Eq. (10.46) to a specific case. A two-level atom with J = 0,
Ju = 1, and gα J = 1 is illuminated by an unpolarized radiation beam having
u u
intensity I (ν), angular spread ∆Ω , and making an angle θ with the direction
of the magnetic field (see Fig. 10.5). Equations (10.46) yield, with the help of
Eq. (2.26a) and Table 5.3
3 N c2 ∆Ω
ρ∓1 ≡ ρα J (∓1) = (1 + cos2 θ ) I (ν0 ∓ νL )
u u 4 N 2hν03 4π
3 N c2 ∆Ω
ρ0 ≡ ρα J (0) = 2 sin2 θ I (ν0 ) . (10.47)
u u 4 N 2hν03 4π
These formulae show that by suitably choosing the angle θ and the spectrum of the
incident radiation, the three Zeeman sublevels can be arbitrarily populated. Note,
however, that if the spectrum of the incident radiation is symmetrical about line
center – in particular, if it is flat across the whole spectral line – the populations
of the sublevels Mu = 1 and Mu = −1 are the same (the atom is ‘aligned’ but
not ‘oriented’, see Sect. 3.7 and Table 3.6): atomic orientation can arise only if
I (ν0 − νL ) and I (ν0 + νL ) are different.
Next we evaluate the radiation scattered by the atom in the direction Ω (see
Fig. 10.5). The emission coefficient can be obtained from Eq. (7.19e), with the
given by Table 5.3 (with θ → θ, γ → 180◦ − χ). Using Eqs. (7.8),
tensors Tqq (i, Ω)
(3.107) and (10.6), and the shorthand notations for ρα J (Mu ) of Eqs. (10.47), we
u u
obtain
3 1 ρ φ + ρ−1 φ1
= ε
ε0 (ν, Ω) L ρ0 φ0 sin2 θ + 1 −1 1 + cos2 θ
ρ−1 + ρ0 + ρ1 2 2
3 1 ρ φ + ρ−1 φ1
= ε
ε1 (ν, Ω) L ρ φ − 1 −1 sin2 θ cos 2χ
ρ−1 + ρ0 + ρ1 2 0 0 2
3 1 ρ φ + ρ−1 φ1
= ε
ε2 (ν, Ω) L ρ φ − 1 −1 sin2 θ sin 2χ
ρ−1 + ρ0 + ρ1 2 0 0 2
532 CHAPTER 10
= ε 3 1
ε3 (ν, Ω) ρ φ − ρ1 φ−1 cos θ , (10.48)
L
ρ−1 + ρ0 + ρ1 2 −1 1
where εL is defined in Eq. (9.11) and the profiles φ−1 , φ0 , φ1 in Eq. (A13.10).
If the atom is unpolarized (ρ−1 = ρ0 = ρ1 ), Eqs. (10.48) reduce to Eqs. (9.10)
written for the transition (J = 0, Ju = 1). In such case two remarkable spectral
properties hold: a) the frequency profile of ε0 , ε1 , ε2 is symmetrical about line cen-
ter, the frequency profile of ε3 is antisymmetrical; b) the integral over frequency of
ε1 , ε2 and ε3 is zero. As obvious from Eqs. (10.48), none of these properties is any
longer valid if atomic polarization is present. In particular, the symmetry character-
istics of the profiles about line center are broken if the atom is oriented (ρ−1 = ρ1 ).
The frequency-integrated linear polarization turns out to be proportional to the
amount of atomic alignment, the frequency-integrated circular polarization to the
amount of atomic orientation (cf. Table 3.6).
It should be remarked that the formal invariance of Eqs. (10.48) under the trans-
formation χ → π + χ (rotation through 180◦ of the magnetic field vector about the
direction of the scattered radiation) does not mean that the scattered radiation is
unaffected by such transformation – this is only true when the atom is unpolarized.
In general, the populations ρ−1 , ρ0 , ρ1 depend on the angle θ (see Eqs. (10.47))
which is not invariant for χ → π + χ.
10.6. The Two-Level Atom: the Role of Collisions
(unpolarized lower level)
Resonance polarization and the Hanle effect are deeply modified by the presence
of collisions. In this section we illustrate their effects for the two-level atom, under
the restrictive assumption that the lower level is unpolarized. Thus we refer again
to Eq. (10.8), and we recall that the collisional rates for inelastic and superelastic
collisions are connected by the Einstein-Milne relation (Eq. (7.98))
hν
(0) 2J + 1 − k T0
CI (αu Ju , α J ) = u e B c CS(0) (α J , αu Ju ) , (10.49)
2J + 1
since by assumption the colliding particles are characterized by a Maxwellian dis-
tribution of velocities with temperature Tc .
Substitution of Eq. (10.9) into Eq. (10.8) and use of Eqs. (7.8), (10.11) and
(10.28) yields
2J + 1 B(α J → αu Ju )
ρK
Q (αu Ju ) =
2Ju + 1 A(αu Ju → α J )
BP (Tc ) δK0 δQ0 + wJ(K)J (−1)Q J−Q
K
(ν0 )
× u
ρ00 (α J ) , (10.50)
1 + i QHu + + δu(K)
NON-EQUILIBRIUM ATOMIC PHYSICS 533
where
CS(0) (α J , αu Ju ) D(K) (αu Ju )
= , δu(K) = , (10.51)
A(αu Ju → α J ) A(αu Ju → α J )
and where
hν
2hν03 − k T0
BP (Tc ) = e B c
c2
is the Planck function (in the Wien limit, where stimulated emission is neglected)
at the temperature Tc .
The physical meaning of the quantities introduced in Eqs. (10.51) is quite obvious.
represents the ratio of collisional to radiative de-excitation rates of the upper level,
whereas δu(K) – with δu(0) = 0, see Eq. (7.102) – represents the effective number of
depolarizing collisions (for the statistical tensors of rank K) taking place during the
lifetime of the excited level. It is worth noticing that, under the limit δu(K) → ∞,
Eq. (10.50) predicts that all the ρK Q (αu Ju ) with K = 0 vanish. The only non-zero
statistical tensor is ρ00 (αu Ju ), and Eq. (10.50) can be cast into the form
2Ju + 1 A(αu Ju → α J ) ρ00 (αu Ju ) 1
= B (T ) + J 0 (ν ) . (10.52)
2J + 1 B(α J → αu Ju ) ρ00 (α J ) 1+ P c 1+ 0 0
This relation is easily recognized as the usual statistical equilibrium equation of the
‘scalar’ (unpolarized) case, once the left-hand side is interpreted as the line source
function (see Eqs. (9.5), (9.11), (9.13) and (10.6)).1
To obtain the radiation emitted by the atom along a given direction Ω, we have to
substitute Eq. (10.50) into Eq. (10.39) or, if one is only interested in the frequency-
integrated radiation, into Eq. (10.31). Let us consider the latter case first. Use of
Eqs. (9.5), (10.6), (10.17), (5.157) and Table 5.6 yields
= kA
ε̃i (Ω) L B (T ) δ
1 + P c i0
5
dΩ
3
+ Ω
P (Ω, ; B, ) ,
, δ (K) ) I (ν , Ω (10.53)
4π j=0 ij u j 0
where the generalized scattering phase matrix (i.e. the phase matrix in the presence
of a magnetic field and collisions) is given by
Ω
Pij (Ω, ; B,
, δ (K) ) = WK (J , Ju ) (−1)Q T Q
K )
T K (j, Ω
(i, Ω)
u −Q
KQ
1
× . (10.54)
1 + i QHu + + δu(K)
1
The right-hand side is often written in the slightly different form
BP (Tc ) + (1 − ) J00 (ν0 ) ,
where = /(1 + ).
534 CHAPTER 10
Equations (10.53) and (10.54) show the double role played by collisions in resonance
scattering. On the one hand, inelastic (and superelastic) collisions introduce an
additional term in the emission coefficient. This term affects only intensity, not
polarization, and is proportional to /(1 + ) or, in other words, to the branching
ratio
CS(0) (α J , αu Ju )
,
A(αu Ju → α J ) + CS(0) (α J , αu Ju )
which represents the collisional de-excitation probability divided by the total de-
excitation probability of the upper level. The unpolarized character of the emitted
radiation is a consequence of the fact that we have assumed an isotropic distribution
of colliders, which prevents the possibility of describing impact polarization.
On the other hand, collisional rates affect the scattering phase matrix via the
factor
1
, (10.55)
1 + i QHu + + δu(K)
which involves both inelastic and elastic collisions. This factor describes the so-
called quenching of resonance polarization (see Mitchell and Zemanski, 1934). Its
effects are easily understood by rewriting Eq. (10.55) in the form (cf. Eqs. (5.193),
(5.194) for a similar ‘classical’ expression)
1 1 1
= , (10.56)
1 + i QHu + + δu(K) 1 + + δu(K) 1 + i QHu (K)
where, using Eqs. (10.28) and (10.51)
Hu 2πνL gα J
Hu (K) = = u u
.
(K)
1 + + δu A(αu Ju → α J ) + CS(0) (α J , αu Ju ) + D(K) (αu Ju )
Equations (10.54) and (10.56) show that the generalized phase matrix can be for-
mally deduced from the Hanle phase matrix (Eq. (10.33)) by the substitutions
WK (J , Ju )
WK (J , Ju ) → , Hu → Hu (K) .
1 + + δu(K)
Thus the quenching of resonance polarization consists both in a net reduction of
the scattered polarization and in a reduction of the efficiency of the magnetic field
as depolarizing agent in the Hanle effect.
If the dependence of the scattered radiation on frequency is considered, a further
effect of collisions comes into play, i.e., the broadening of the individual profiles
Φ(να J M , α J M − ν). Neglecting frequency shifts, these profiles are given by
u u u
1 Γ
Φ(να − ν) =
J M , α J M
u u u π Γ 2 + (να J Mu , α J M − ν)2
u u
i να J M , α J M − ν
u u u
+
,
π Γ + (να J M , α J M − ν)2
2
u u u
NON-EQUILIBRIUM ATOMIC PHYSICS 535
where (see Eqs. (6.59b) and (5.180))
γu + γ f
Γ = Γ + 2Γc = + ,
4π 2π
γu and γ being the inverse lifetimes of the upper and lower level, respectively, and
f the frequency of elastic collisions. For the two-level atom, neglecting the width
of the lower level, we have
A(αu Ju → α J ) + CS(0) (α J , αu Ju ) f
Γ = + .
4π 2π
The form of the profiles Φ(να J M , α J M − ν) is the same as in the collisionless
u u u
regime except for the replacement Γ → Γ . Thus, in the presence of collisions,
we can still define the generalized profiles as in Eq. (10.40), and the expression
of the emission coefficient is formally identical with Eq. (10.39). Substitution of
Eq. (10.50) into Eq. (10.39) leads, using again Eqs. (9.5), (10.6) and (5.157), to the
following expression
= kA
0 (J , Ju ; ν) T 0 (i, Ω)
Φ0K K
εi (ν, Ω) L BP (Tc )
1+
K
5
dΩ (K)
3
KK K )
T K (j, Ω
+ ΦQ (J , Ju ; ν) wJ J (−1)Q T Q (i, Ω) −Q
4π j=0
u
KK Q
1 ) .
× I (ν , Ω (10.57)
1 + i QHu + + δu(K) j 0
The broadening of the profiles due to collisions has important consequences and,
in particular, it breaks down a property proved in Sect. 10.4 for the collisionless
regime: while in the latter resonance scattering in the far wings of a spectral line is
unaffected by the magnetic field, this is no longer true in the presence of collisions.
For → ∞ only the first term of Eq. (10.57) is non-zero. It can easily be shown
that under this limit Eq. (10.57) reduces to Eqs. (9.10) written for LTE.
10.7. The Two-Level Atom: the Role of Lower-Level Polarization
(cylindrical symmetry - no magnetic field - no stimulated
emission - no inelastic collisions)
Up to now we have systematically ignored the possibility that some atomic polar-
ization may form in the lower level of a two-level atom under the action of the
incident radiation field: all the results derived in Sects. 10.2 to 10.6 are based on
the unpolarized lower level approximation. Here we release this approximation,
in order to illustrate how the former results are modified when lower-level atomic
polarization is allowed for.
536 CHAPTER 10
As already mentioned in Sect. 10.1, the statistical equilibrium equations become
remarkably more complicated as soon as lower-level polarization is taken into ac-
count, and closed analytical solutions to these equations can hardly be found even
for the simplest cases. If we want to get such solutions (at least for a few sample
cases), so as to be able to understand in detail the role of lower-level polarization,
we are therefore forced to introduce some other approximation.
An interesting possibility is offered by the so-called weak anisotropy approxima-
tion: this will be discussed in Sects. 10.13 and 10.14. However, as obvious from the
denomination, such approximation is not applicable when the incoming radiation
field is markedly anisotropic – in particular, when it is just a beam. In the present
section we adopt an assumption on the incident radiation field which in a sense is
complementary to the weak anisotropy approximation: we assume the radiation to
be unpolarized and cylindrically symmetrical about a fixed direction. For the sake of
simplicity, we also adopt the following additional assumptions: absence of inelastic
and superelastic collisions (whereas depolarizing collisions are taken into account);
absence of the magnetic field (this will be considered in the next section); absence
of stimulated emission (which will be considered in Sect. 10.9).
The simplification introduced by the above assumptions (in particular, those con-
cerning the incoming radiation and the magnetic field) can easily be understood.
If the symmetry axis of the radiation field is taken as the z-axis of our refer-
ence system, and if the radiation is frequency-independent across the line width
(flat-spectrum approximation), two single quantities are sufficient to fully specify
the incoming radiation, i.e., the multipole components J00 (ν0 ) and J02 (ν0 ) – see
Eqs. (5.164). This implies that, in stationary situations, the only non-zero statis-
tical tensors are those with K even and Q = 0.
This property can be proved directly from the statistical equilibrium equations
(10.1) and (10.2) by keeping in mind the expressions of the radiative rates given
K
in Eqs. (7.14).1 Since the multipole components JQ r (ν0 ) with Qr = 0 are zero, all
r
the radiative rates in Eqs. (10.1) and (10.2) are zero unless Q = Q : this is due to
the presence of a 3-j symbol of the form
K K Kr
(10.58)
Q −Q Qr
in the expressions of the rates TA and RA – on the other hand, the rates TE and RE
are proportional to the Kronecker symbol δQQ . Thus the overall set of equations
is decoupled into separate subsets characterized by a fixed Q value. As the trace
equation (10.4) – the only non-homogeneous equation – involves only statistical
tensors with Q = 0, all the other ones are identically zero. Note, incidentally, that
this means that all coherences between magnetic sublevels are zero (see Eq. (3.101)):
under the assumptions adopted in this section, atomic polarization can only consist
in population differences of the magnetic sublevels. On the other side, since the
1
As obvious from the following argument, the property is valid even if inelastic and superelastic
collisions and/or stimulated emission are taken into account. It also holds for the multi-level (not
just the two-level) model atom.
NON-EQUILIBRIUM ATOMIC PHYSICS 537
K
multipole components JQ r (ν0 ) with Kr = 1 are zero, all the radiative rates vanish
r
unless (K − K ) = 0 or (K − K ) = ±2 : as far as TA and RA are concerned, this
is again due to the 3-j symbol in Eq. (10.58) which, for Q = Q = Qr = 0, is
zero unless (K + K + Kr ) is even (see Eq. (2.25)); as for TE and RE , both are
proportional to the Kronecker symbol δKK . It follows that the remaining set of
equations (involving the statistical tensors with Q = 0 ) is in turn decoupled into
two separate subsets relating statistical tensors of even and odd rank, respectively.
As the latter do not appear in the trace equation, the corresponding subset is
identically solved by the null solution.
Because of the above property, the number of equations to be solved is largely
reduced. It can be shown that such number is
J + Ju + 2 for J , Ju integers
Neq = (10.59)
J + Ju + 1 for J , Ju half-integers ,
to be compared with Eq. (10.3).
Let us rewrite Eqs. (10.1) and (10.2) for the physical situation considered in this
section. Using Eqs. (7.14b) and (7.14e) we get
d K
ρ (α J ) = − A(αu Ju → α J ) + D(K) (αu Ju ) ρK 0 (αu Ju )
dt 0 u u
+ TA (αu Ju K 0, α J K 0) ρK
0 (α J )
K even
!
d K Ju Ju K
ρ0 (α J ) = (2Ju + 1) A(αu Ju → α J ) (−1)1+J +Ju ρK
0 (αu Ju )
dt J J 1
− RA (α J K 0 K 0) + δKK D(K) (α J ) ρK
0 (α J ) , (10.60)
K even
where K is even. The rates TA and RA , given by Eqs. (7.14a) and (7.14d) respec-
tively, reduce, with the help of Eqs. (2.26a), (2.36a) and (2.49), to the following
TA (αu Ju K 0, α J K 0) = (2J + 1) B(α J → αu Ju )
⎡
!
Ju Ju K
× ⎣ δKK (−1)1+J +Ju J00 (ν0 )
J J 1
⎧ ⎫ ⎤
⎨ Ju J 1 ⎬
K K 2
+ 15(2K + 1)(2K + 1) J J 1 J02 (ν0 )⎦
⎩ u ⎭ 0 0 0
K K 2
538 CHAPTER 10
RA (α J K 0 K 0) = (2J + 1) B(α J → αu Ju )
⎡
1
× ⎣ δKK J00 (ν0 ) + (−1)1−J +Ju 15(2K + 1)(2K + 1)
2J + 1
⎤
! !
K K 2 1 1 2 K K 2
× J02 (ν0 )⎦. (10.61)
J J J J J Ju 0 0 0
In view of practical applications, it is convenient to cast Eqs. (10.60) in dimension-
less form, by dividing both sides by the Einstein coefficient A(αu Ju → α J ) and
by introducing the parameters
(2J + 1) B(α J → αu Ju ) 0 c2
J0 (ν0 ) = J 0 (ν ) = n̄
(2Ju + 1) A(αu Ju → α J ) 2hν03 0 0
J02 (ν0 ) w
=√
J00 (ν0 ) 2
D(K) (αu Ju )
= δu(K)
A(αu Ju → α J )
D(K) (α J )
= δ(K) (10.62)
B(α J → αu Ju ) J00 (ν0 )
(Eqs. (7.8) have been used in the first relation). Their meaning is the following. The
parameter n̄ represents the solid-angle average of the number of photons per mode
at frequency ν0 (see Eqs. (4.36), (5.157) and Table 5.7); in the present case it is
much less than unity, since we have assumed stimulated emission to be negligible.
The parameter w, satisfying the condition −1/2 ≤ w ≤ 1 under the conditions
considered in this section, is the so-called anisotropy factor . Its value is unity for a
unidirectional (unpolarized) radiation beam, and zero for an isotropic (unpolarized)
radiation field – see Eqs. (5.164). Finally, δu(K) , already introduced in Eqs. (10.51),
represents the effective number of depolarizing collisions (for the statistical tensors
of rank K) taking place during the lifetime of the upper level, while δ(K) is the
corresponding quantity for the lower level. Note that, by combining the first and
the last of Eqs. (10.62), we can write
2J + 1 D(K) (α J )
δ(K) = ; (10.63)
2Ju + 1 n̄ A(αu Ju → α J )
since D(K) (α J ) and D(K) (αu Ju ) are of the same order of magnitude and n̄ 1,
it follows that
δ(K) δu(K) .
In the following we apply Eqs. (10.60) to a few simple cases. We begin with the
transition (J = 1, Ju = 0) which is the simplest transition where lower-level polar-
ization plays a role. After some calculations, involving the numerical evaluation of
NON-EQUILIBRIUM ATOMIC PHYSICS 539
several 3-j, 6-j, and 9-j symbols, one gets the following independent equations1
1 d 0 1 1
ρ0 (αu Ju ) = − ρ00 (αu Ju ) + √ n̄ ρ00 (α J ) + √ n̄w ρ20 (α J )
A(αu Ju → α J ) dt 3 6
1 d 2 1 1 1
ρ (α J ) = − n̄ 1 − w + δ(2) ρ20 (α J ) − √ n̄w ρ00 (α J ) ,
A(αu Ju → α J ) dt 0 3 2 3 2
where the parameters defined in Eqs. (10.62) have been introduced. In stationary
situations, the solution can be written in the form2
ρ20 (α J ) √ w
σ02 (α J ) ≡ =− 2
ρ00 (α J ) 2 − w + 2 δ(2)
ρ00 (αu Ju ) 1 w2
= √ n̄ 1 − . (10.64)
ρ00 (α J ) 3 2 − w + 2 δ(2)
These expressions exhibit several interesting aspects. First of all, atomic polar-
ization, as well as the ratio of the upper to the lower level population, depend
in a non-linear way on the anisotropy factor w: this is generally the case when
lower-level polarization is allowed for.
Secondly, atomic polarization is zero if w = 0, which is quite obvious because in
that case the physical environment is completely isotropic (we recall that in our
treatment collisions are isotropic by assumption). Apart from this special case,
atomic polarization tends to zero for δ(2) → ∞: in general, the approximation
of the unpolarized lower level is strictly justified only under such limit. On the
other hand, according to Eq. (10.63), δ(2) is inversely proportional to n̄, or to
the intensity of the radiation field. Thus we can also say that lower-level atomic
polarization increases as the intensity of the radiation field is increased. This is
easily understood if we consider that the radiative rates (which are responsible for
atomic polarization) are proportional to the intensity of the radiation field, while
the collisional rates are independent of it (they only depend on the density of the
colliders).
Another peculiar aspect of Eqs. (10.64) is that for w = 1 and δ(2) = 0 (unidirec-
tional, unpolarized radiation beam and no collisions) we have
√
σ02 (α J ) = − 2 , ρ00 (αu Ju ) = 0 . (10.65)
The second result means that the upper level is completely depopulated (the atom
is perfectly transparent to the incident radiation), which looks rather surprising
1
We recall that the equations for the time derivatives of ρ00 (α J ) and ρ00 (αu Ju ) are not
independent: see footnote on p.512.
2 K (αJ) = ρK (αJ)/ρ0 (αJ), sometimes referred to as reduced statistical tensor,
The symbol σQ Q 0
will be often used in the following.
540 CHAPTER 10
at first sight. However, one should bear in mind that a unidirectional, unpo-
larized radiation beam induces only transitions with ∆M = ±1.1 The Zeeman
sublevels M = ±1 of the lower level are therefore depopulated by the incident
radiation, while the sublevel M = 0 is not. On the other hand, the upper level
(Ju = 0, Mu = 0) de-excites with the same probability to each of the three sub-
levels of the lower level. The result of this cycle is that all the atoms are eventually
pumped
√ into the sublevel M = 0 of the lower level, which explains both the factor
− 2 for the relative amount of alignment (cf. Table 3.6) and the total absence of
population in the upper level.
Finally, Eqs. (10.64) show that depolarizing collisions affect the ratio of the upper
level to the lower level population, which may look surprising because such colli-
sions are elastic (see Sect. 7.13). The reason for this is that depolarizing collisions
tend to mix up the populations of the Zeeman sublevels of each (αJ)-level, thus
affecting (indirectly) the absorption and emission of radiation. Such effect is clearly
illustrated by the example just considered of a unidirectional, unpolarized incident
beam: in stationary situations and under the limit δ(2) = 0, all the atoms are in the
sublevel M = 0 of the lower level whence absorption can no longer take place, so
that the upper level is completely depopulated (Eqs. (10.65)). But if depolarizing
collisions are present, some of the atoms are forced to move from the M = 0 to
the M = ±1 sublevels whence absorption is possible, which eventually leads to a
non-null value for the population of the upper level.
The transition (J = 1, Ju = 0), though interesting under several aspects, is not
so illuminating under the point of view of resonance polarization: since Ju = 0, the
emitted radiation is unpolarized irrespective of the presence of lower-level atomic
polarization. The latter can only affect the absorption properties of the atom, an
argument that we are not going to pursue here but that might be easily handled
by substituting the expressions of the statistical tensors of the lower level into
Eqs. (7.15a,c) applied to the two-level atom.
We now analyze the simplest transition where atomic polarization can be present
both in the lower and the upper level, namely the transition (J = 1, Ju = 1). Appli-
cation of Eqs. (10.60) to this case yields the following set of independent equations
1 d 0 1
ρ (α J ) = − ρ00 (αu Ju ) + n̄ ρ00 (α J ) − √ n̄w ρ20 (α J )
A(αu Ju → α J ) dt 0 u u 2 2
1 d 2 1
ρ (α J ) = − 1 + δu(2) ρ20 (αu Ju ) − √ n̄w ρ00 (α J )
A(αu Ju → α J ) dt 0 u u 2 2
1
− n̄ 1 + w ρ20 (α J )
2
1
This fact can be easily verified by considering the statistical equilibrium equations in the
standard representation and the corresponding expressions of the radiative rates (Eqs. (7.5) and
(7.9), respectively). It is sufficient to recall that for a unidirectional, unpolarized radiation beam
the only non-zero components of the (reducible) radiation field tensor are J11 (ν0 ) and J−1−1 (ν0 )
– cf. Eqs. (5.153) and Table 5.4.
NON-EQUILIBRIUM ATOMIC PHYSICS 541
1 d 2 1 1
ρ (α J ) = − ρ20 (αu Ju ) + √ n̄w ρ00 (α J )
A(αu Ju → α J ) dt 0 2 2 2
1
− n̄ 1 + w + δ(2) ρ20 (α J ) ,
4
where the notations of Eqs. (10.62) have been used. In stationary situations, the
solution can be written in the form
√ 6 + 3w + 4 δ(2)
σ02 (αu Ju ) = − 2 w
4 (1 + δu )(4 + w + 4 δ(2) ) − (1 + w) − w2 (3 + 2 δu(2) )
(2)
√ 3 + 2 δu(2)
σ02 (α J ) = 2w
2 (1 + δu )(4 + w + 4 δ(2) ) − (1 + w)
(2)
ρ00 (αu Ju ) 4 (1 + δu(2) )(4 + w + 4 δ(2) ) − (1 + w) − w2 (3 + 2 δu(2) )
= n̄ . (10.66)
ρ00 (α J ) 4 (1 + δu(2) )(4 + w + 4 δ(2) ) − (1 + w)
These expressions are rather involved and clearly show the intricate dependence of
atomic polarization on the different parameters. However, it is easily seen that the
basic characteristics found for the transition (J = 1, Ju = 0) are confirmed: atomic
polarization (of both the upper and the lower level) depends non-linearly on the
anisotropy factor w, and vanishes for w = 0; the ratio of the upper to the lower
level population also depends non-linearly on w, and is affected by the presence of
depolarizing collisions.
Consider now the limit δ(2) → ∞. Equations (10.66) reduce to
√
2 w ρ00 (αu Ju )
σ02 (αu Ju ) =− , σ02 (α J ) = 0 , = n̄ . (10.67)
4 1 + δu(2) ρ00 (α J )
These can be compared with the corresponding expressions derived in Sect. 10.6
for the two-level atom under the unpolarized lower level approximation. Setting
= Hu = 0 (the conditions considered in the present section) into Eq. (10.50) we
obtain, for an arbitrary transition (J , Ju )
2 wJ(2)J ρ00 (αu Ju ) 2Ju + 1
σ0 (αu Ju ) u.l.l. = √ u (2) w , = n̄ , (10.68)
2 1 + δu ρ00 (α J ) u.l.l. 2J + 1
where ‘u.l.l.’ stands for ‘unpolarized lower level’, and where we have used the
notations of Eqs. (10.62). For the transition (J = 1, Ju = 1), Eqs. (10.68) give
the same results as Eqs. (10.67) – see Table 10.1. Once again we see that the
unpolarized lower level approximation is correct under the limit δ(K) → ∞ – a
statement which, on the other hand, is suggested by the structure itself of the
statistical equilibrium equations. It is important to remark that the condition
δ(K) 1 occurs not only when D(K) A(αu Ju → α J ) – in which case δu(K) 1, so
that the polarization of the upper level tends to zero as well – but also when n̄ 1
(see Eq. (10.63)). Therefore the unpolarized lower level approximation is especially
542 CHAPTER 10
appropriate in the presence of depolarizing collisions and a radiation field of low
intensity. It should also be noticed that, according to Eqs. (10.68), the atomic
polarization of the upper level depends linearly on w, contrary to Eqs. (10.66).
Finally, it is instructive to consider the special case δ(2) = δu(2) = 0, w = 1 (absence
of depolarizing collisions, atom illuminated by a unidirectional, unpolarized beam).
Equations (10.66) yield
√ 1 ρ00 (αu Ju ) 3
σ02 (αu Ju ) = − 2 , σ02 (α J ) = √ , = n̄ .
2 ρ00 (α J ) 4
According to Table 3.6, this means that the Zeeman sublevels of the upper level
with Mu = ±1, and the Zeeman sublevel of the lower level with M = 0 are
completely depopulated. The populations of the remaining sublevels, normalized
to unity, are
3 n̄ 2
ρα J (0) = , ρα J (±1) = .
u u 4 + 3 n̄ 4 + 3 n̄
The analytical results obtained for the two simple cases (J = 1, Ju = 0) and
(J = 1, Ju = 1) might in principle be extended to transitions involving larger J
values. However, the algebraic system to be solved becomes of course more and
more complicated; at the same time, the dependence of atomic polarization on the
various parameters becomes so intricate that no real insight into the physics of the
scattering process can be gained. In order to give an idea of the atomic polarization
arising in different transitions, we collect in Table 10.3 the values of the reduced
statistical tensors
ρK
0 (αu Ju ) ρK
0 (α J )
σ0K (αu Ju ) = , σ0K (α J ) =
ρ00 (αu Ju ) ρ00 (α J )
obtained by a numerical code for the collisionless, maximum anisotropy regime
(δ(K) = δu(K) = 0, w = 1, corresponding to the irradiation of the atom by a unidi-
rectional, unpolarized beam). For comparison, we also report the value of
2 wJ(2)J
σ0 (αu Ju ) u.l.l. = √u
2
obtained for the same physical conditions plus the assumption of the unpolarized
lower level (cf. Eqs. (10.68)). The last two columns of Table 10.3 show, respectively,
the fractional linear polarization pQ expected in a 90◦ scattering process, and the
same quantity evaluated under the additional assumption of the unpolarized lower
level, [pQ ]u.l.l.. The expression of the former can be deduced from Eq. (10.15).
Defining the positive Q direction as the perpendicular to the scattering plane, we
obtain, with the help of Table 5.6 (with θ = 90◦ , γ = 90◦ ) and Eq. (10.14)
εQ (90◦ ) 3 wJ(2)J σ02 (αu Ju )
pQ = = √ u
. (10.69)
εI (90◦ ) 2 2 − wJ(2)J σ02 (αu Ju )
u
NON-EQUILIBRIUM ATOMIC PHYSICS 543
TABLE 10.3
Atomic polarization for different transitions in the collisionless, maximum anisotropy regime.
The letters u and denote the upper and lower level, respectively. The last two columns give
the fractional linear polarization in a 90 ◦ scattering. The label u.l.l. means ‘unpolarized lower
level’.
σ02 (u) σ04 (u) σ06 (u) σ08 (u) σ010 (u) [σ02 (u)]u.l.l.
J Ju pQ [pQ ]u.l.l.
σ02 ( ) σ04 ( ) σ06 ( ) σ08 ( ) σ010 ( )
0.707 − − − − 0.707
0 1 1. 1.
− − − − −
− − − − − −
1/2 1/2 0. 0.
− − − − −
0.5 − − − − 0.5
1/2 3/2 0.429 0.429
− − − − −
− − − − − −
1 0 0. 0.
−1.414 − − − −
−1.414 − − − − −0.354
1 1 1. 0.2
0.707 − − − −
0.622 0.214 − − − 0.418
1 2 0.448 0.288
0.321 − − − −
− − − − − −
3/2 1/2 0. 0.
−0.5 − − − −
−0.526 − − − − −0.4
3/2 3/2 0.353 0.261
0.308 − − − −
0.764 0.265 − − − 0.374
3/2 5/2 0.5 0.226
0.5 − − − −
−0.202 − − − − 0.071
2 1 −0.021 0.008
−0.489 0.146 − − −
−0.460 −0.247 − − − −0.418
2 2 0.319 0.288
0.272 0.146 − − −
0.903 0.354 0.089 − − 0.346
2 3 0.556 0.191
0.659 0.137 − − −
−0.357 − − − − 0.1
5/2 3/2 −0.053 0.015
−0.541 0.141 − − −
−0.408 −0.109 − − − −0.428
5/2 5/2 0.287 0.302
0.244 0.065 − − −
1.035 0.458 0.132 − − 0.327
5/2 7/2 0.611 0.170
0.809 0.242 − − −
544 CHAPTER 10
TABLE 10.3
(continued)
σ02 (u) σ04 (u) σ06 (u) σ08 (u) σ010 (u) [σ02 (u)]u.l.l.
J Ju pQ [pQ ]u.l.l.
σ02 ( ) σ04 ( ) σ06 ( ) σ08 ( ) σ010 ( )
−0.452 0.054 − − − 0.120
3 2 −0.079 0.022
−0.590 0.164 −0.027 − −
−0.388 −0.092 −0.047 − − −0.433
3 3 0.275 0.310
0.233 0.055 0.028 − −
1.156 0.569 0.186 0.037 − 0.313
3 4 0.664 0.155
0.948 0.352 0.060 − −
−0.522 0.100 − − − 0.134
7/2 5/2 −0.101 0.027
−0.632 0.196 −0.036 − −
−0.376 −0.080 −0.021 − − −0.436
7/2 7/2 0.268 0.316
0.226 0.048 0.013 − −
1.266 0.683 0.248 0.062 − 0.303
7/2 9/2 0.711 0.144
1.077 0.466 0.114 − −
−0.579 0.143 −0.013 − − 0.144
4 3 −0.120 0.032
−0.670 0.230 −0.047 0.005 −
−0.369 −0.075 −0.018 −0.009 − −0.439
4 4 0.264 0.320
0.222 0.045 0.011 0.005 −
1.363 0.797 0.318 0.091 0.015 0.294
4 5 0.753 0.136
1.195 0.584 0.175 0.025 −
−0.628 0.183 −0.026 − − 0.153
9/2 7/2 −0.137 0.035
−0.704 0.263 −0.061 0.009 −
−0.364 −0.071 −0.015 −0.004 − −0.440
9/2 9/2 0.261 0.322
0.219 0.043 0.009 0.002 −
1.449 0.908 0.392 0.125 0.027 0.288
9/2 11/2 0.790 0.129
1.300 0.701 0.242 0.051 −
−0.669 0.222 −0.040 0.003 − 0.160
5 4 −0.152 0.039
−0.733 0.296 −0.076 0.012 −0.001
−0.360 −0.069 −0.014 −0.003 −0.002 −0.442
5 5 0.259 0.324
0.217 0.042 0.008 0.002 0.001
−0.704 0.259 −0.056 0.006 − 0.165
11/2 9/2 −0.165 0.041
−0.759 0.327 −0.093 0.017 −0.002
−0.358 −0.068 −0.013 −0.003 −0.001 −0.442
11/2 11/2 0.258 0.326
0.216 0.041 0.008 0.002 0.000
NON-EQUILIBRIUM ATOMIC PHYSICS 545
In the special case of unpolarized lower level, this expression reduces to
3 W2 (J , Ju )
pQ u.l.l. = , (10.70)
4 − W2 (J , Ju )
where Eqs. (10.68) and (10.17) have been used. Note that expression (10.70) has
already been derived in Sect. 10.2 (Eq. (10.26)).
It is interesting to observe that, with some exceptions, the linear polarization of
the scattered radiation is larger (in absolute value) when the atomic polarization
of the lower level is taken into account. The exceptions are provided by transitions
having ∆J = Ju − J = 0 (and J > 2), for which the difference between the last
two columns of Table 10.3 is, however, rather small. It is also interesting to notice
that for transitions with ∆J = −1 the polarization of the scattered radiation is
negative (polarization direction parallel to the scattering plane) when the atomic
polarization of the lower level is taken into account.
The results in Table 10.3 can also be used to reconstruct the relative popu-
lations of the Zeeman sublevels via Eq. (3.99). Referring for instance to the
transition (J = 1, Ju = 2), one can easily find that the relative populations are
9/22 (M = ±1) and 4/22 (M = 0) for the lower level, and 9/25 (Mu = ±2), 2/25
(Mu = ±1), 3/25 (Mu = 0) for the upper level.1
10.8. The Two-Level Atom: the Hanle Effect with
Lower-Level Polarization
(cylindrical symmetry - no stimulated emission - no inelastic collisions)
The results of the previous section can be generalized by allowing for the presence
of a magnetic field: this leads to an investigation of the Hanle effect in more gen-
eral terms than in Sect. 10.3, since lower-level atomic polarization is now taken into
account. In order to simplify the problem, we adopt exactly the same assumptions
as in the preceding section (except of course for the presence of the magnetic field):
we neglect stimulated emission as well as inelastic and superelastic collisions, and
we assume the incoming radiation to be unpolarized and cylindrically symmetri-
cal about a given direction. The flat-spectrum approximation requires now the
radiation to be unstructured across a frequency interval centered at the transition
frequency ν0 and larger than the Zeeman splitting of the line.
Since there are two preferred directions (the magnetic field vector and the symme-
try axis of the radiation field), it is convenient to write the statistical equilibrium
equations in a reference system whose z-axis is parallel to one of them. Either
choice has its own advantages and disadvantages. In the frame with the z-axis
1
In terms of rational numbers,
√ one has √ √
2 5 2 13 70 2 14
σ0 (α J ) = , σ02 (αu Ju ) = , σ04 (αu Ju ) = .
22 175 35
546 CHAPTER 10
along the magnetic field direction, one can apply directly Eq. (7.11). The magnetic
contribution to the equations is very simple (being diagonal), but the expressions
of the transfer and relaxation rates are rather complicated because the radiation
field has no special symmetry characteristics about the magnetic field’s direction.
In the frame with the z-axis parallel to the symmetry axis of the radiation field,
the expressions of the rates are much simpler (because the radiation field tensor
has only two non-zero components, J00 (ν0 ) and J02 (ν0 )), whereas the magnetic term
involves a kernel connecting different statistical tensors (see Eqs. (7.78) and (7.79)).
In the following we choose the latter reference frame. It can easily be proved
that, in stationary situations, the only non-null statistical tensors are those with
K even (see Sect. 10.7). Owing to the presence of the magnetic kernel, which mixes
up statistical tensors of the same rank but with different Q values, the statistical
tensors with Q = 0 are in general non-zero. Therefore, contrary to the previous
section, atomic polarization involves both population unbalances and coherences
between Zeeman sublevels. The number of equations (and of unknowns) is now
given by (cf. Eq. (10.59))
(J + 1)(2J + 1) + (Ju + 1)(2Ju + 1) for J , Ju integers
Neq = (10.71)
J (2J + 1) + Ju (2Ju + 1) for J , Ju half-integers .
The statistical equilibrium equations for the two-level atom, under the physical
conditions specified above, can be obtained from Eq. (7.78) by adding the contribu-
tion of depolarizing collisions (Eq. (7.101)). Bearing in mind the formal invariance
of the radiative rates proved in Sect. 7.12 – which implies that such rates are still
given by Eqs. (7.14) – we get, with the help of Eqs. (2.26a), (2.36a) and (2.49)1
d K
ρQ (αu Ju ) = −2πi νL gα J KQQ
K K
ρQ (αu Ju )
dt u u
Q
− A(αu Ju → α J ) + D(K) (αu Ju ) ρK
Q (αu Ju )
+ TA (αu Ju KQ, α J K Q) ρK
Q (α J )
K even
d K
ρQ (α J ) = −2πi νL gα J KQQ
K K
ρQ (α J )
dt
Q
!
Ju Ju K
+ (2Ju + 1) A(αu Ju → α J ) (−1) 1+J +Ju
ρK
Q (αu Ju )
J J 1
− RA (α J KQK Q) + δKK D(K) (α J ) ρK
Q (α J ) , (10.72)
K even
where K is even, and where
1
The rates D (K) (αJ) are obviously invariant, because of the isotropic distribution of the
colliding particles. Such invariance can be formally proved with the help of Eqs. (2.72).
NON-EQUILIBRIUM ATOMIC PHYSICS 547
θ
γ
symmetry axis
of the incident radiation
Fig.10.6. The rotation R appearing in Eqs.(10.73) and (10.74) carries the reference system (XY Z)
into (xyz).
TA (αu Ju KQ, α J K Q) = (2J + 1) B(α J → αu Ju )
⎡
!
⎣ Ju Ju K
× δKK (−1) 1+J +Ju
J00 (ν0 )
J J 1
⎧ ⎫ ⎤
⎨ Ju J 1 ⎬
K K 2
+ (−1)Q 15(2K + 1)(2K + 1) Ju J 1 J02 (ν0 )⎦
⎩ ⎭ −Q Q 0
K K 2
RA (α J KQK Q) = (2J + 1) B(α J → αu Ju )
⎡
1
× ⎣ δKK J00 (ν0 ) + (−1)1−J +Ju + Q 15(2K + 1)(2K + 1)
2J + 1
⎤
! !
K K 2 1 1 2 K K 2
× J02 (ν0 )⎦ ,
J J J J J Ju Q −Q 0
and
∗
KQQ
K
= DQ
K
Q (R) Q DQ
K
Q (R) , (10.73)
Q
R being the rotation which carries the reference system with the z-axis directed
along the magnetic field into the reference system with the z-axis directed along
the symmetry axis of the radiation field.
The expression of the kernel KQQ K
can be somewhat simplified. Referring to
Fig. 10.6, the rotation R is given, in terms of Euler angles, by
548 CHAPTER 10
R ≡ (−γ, −θ, −χ) .
Expression (10.73) is independent of the angle γ, as obvious from Eq. (2.68). The
summation over Q can be performed by expressing Q via Eq. (2.26d),
K K 1
Q = K(K + 1)(2K + 1) (−1)K−Q ,
Q −Q 0
and the product of two rotation matrices via Eq. (2.77). With the help of Eqs.
(2.23a) and (2.73) we obtain the simpler expression
K K 1
KQQ
K
= K(K + 1)(2K + 1) (−1)K−Q D0q
1
(R) . (10.74)
−Q
Q q
Even for the simplest transitions, Eqs. (10.72) are so involved that it isn’t worth
trying to solve them (in stationary situations) analytically: for instance, according
to Eq. (10.71), the transition (J = 1, Ju = 1) is described by a system of 12 equa-
tions in 12 unknowns. It is therefore necessary to resort to a numerical solution.
Once the solution is found, the radiation emitted in a given direction can be
evaluated via Eq. (10.39), or via Eq. (10.31) if only the frequency-integrated radi-
ation is required. To give an example, we refer to the geometry of Fig. 10.6 and we
evaluate the frequency-integrated fractional polarization of the radiation emitted
in the x-direction (i.e., at 90◦ from the symmetry axis of the incoming radiation),
taking the y-axis as the positive Q direction. Using the expressions of T Q K in
(i, Ω)
◦ ◦ ◦
Table 5.6 (with θ = 90 , χ = 0 , γ = 90 ) we obtain from Eq. (10.31), with the
help of Eqs. (3.102) and (10.14)
√
ε̃Q wJ(2)J 3 σ02 + 6 Re(σ22 )
p̃Q = = √ u
√
ε̃I 2 2 − wJ(2)J σ02 − 6 Re(σ22 )
u
√
ε̃ 2 6 wJ(2)J Im(σ12 )
p̃U = U = − √ u √ , (10.75)
ε̃I 2 2 − wJ(2)J σ02 − 6 Re(σ22 )
u
K
where σQ are the reduced statistical tensors of the upper level.
Figures 10.7 and 10.8 show the Hanle diagrams for the transitions (J = 1, Ju = 1)
and (J = 1, Ju = 2), respectively.1 They are obtained from Eqs. (10.75), with
the values of the statistical tensors derived by solving numerically Eqs. (10.72)
– written for stationary situations – together with the trace equation (10.4), under
the following conditions: the magnetic field vector lies in the x-y plane (θ = 90◦ in
Fig. 10.6); the incoming radiation is a unidirectional, unpolarized beam travelling
in the z-direction; its intensity is very weak; depolarizing collisions are negligible.
1
Hanle diagrams for other transitions have been presented by Landolfi and Landi Degl’Innocenti
(1986).
NON-EQUILIBRIUM ATOMIC PHYSICS 549
Fig.10.7. Polarization diagram for the transition (J = 1, Ju = 1) corresponding to the geometry
of Fig.10.6 with θ = 90◦ . The scattering direction is the x-axis. Lower-level atomic polarization
is taken into account. The relevant parameters are: w = 1, n̄ = 10−6 , δu(K) = δ(K) = 0 (see
Eqs.(10.62)). Full lines correspond to χ = const., dashed lines to constant magnetic field strength
(Hu = const., see Eq.(10.28); the Landé factors of the upper and lower level are assumed to be
the same).
In terms of the parameters defined in Eqs. (10.62) the last three conditions read:
w = 1, n̄ 1, δu(K) = δ(K) = 0.
If the diagrams in Fig. 10.7 or 10.8 are compared with a ‘usual’ Hanle diagram
(see e.g. Fig. 10.2), an obvious difference comes out: the latter consists of a unique
‘lobe’, while the former are split into two distinct lobes. Such lobes correspond to
the relaxation of the lower-level coherences (lower-level Hanle effect) and of the
upper-level coherences (‘usual’ Hanle effect), respectively. The order of magnitude
of the magnetic field strength in either lobe is different, because – as suggested by
Eqs. (10.72) – the typical field which depolarizes the lower level is such that
2πνL gα J ≈ B(α J → αu Ju ) J00 (ν0 ) ,
whereas the typical field which depolarizes the upper level is such that
2πνL gα J ≈ A(αu Ju → α J ) .
u u
Since from Eqs. (10.62)
B(α J → αu Ju ) J00 (ν0 ) 2J + 1
= u n̄ ,
A(αu Ju → α J ) 2J + 1
550 CHAPTER 10
Fig.10.8. Same as Fig.10.7 for the transition (J = 1, Ju = 2).
it follows that the former is approximately n̄ times smaller than the latter. The
lower-level and the ‘usual’ Hanle effect are clearly separated in Figs. 10.7 and 10.8
because the incident radiation is extremely diluted (n̄ = 10−6 ). The shape of the
diagrams will however remain the same provided n̄ 1. For n̄ 1 the two regimes
tend to overlap; but a correct treatment of this case requires that the stimulated
emission rates be included in the statistical equilibrium equations (see the next
section).
Let us compare the diagrams in Figs. 10.8 and 10.2 more closely. They refer to
the same transition and the same geometrical and physical conditions: the only
difference is the atomic polarization of the lower level, which is allowed for in
Fig. 10.8 and neglected in Fig. 10.2 (unpolarized lower level approximation). For
zero magnetic field, the value of p̃Q in Fig. 10.8 is larger than the value in Fig. 10.2
(approximately 0.45 and 0.29, respectively: see Table 10.3), consistently with the
results presented at the end of Sect. 10.7. The value of p̃Q corresponding to the
relaxation of the lower-level coherences (the ‘transition region’ between the two
lobes of Fig. 10.8) depends on the value of the angle χ, ranging approximately
from 0.30 for χ = 0◦ or 180◦ to 0.33 for χ = 90◦ . In any case, this value is larger
than p̃Q 0.29 of Fig. 10.2, where lower-level atomic polarization is neglected a
priori. The same characteristic appears from Fig. 10.7, where the value of p̃Q cor-
responding to the transition between the two lobes (independent of χ in this case)
is about 0.33, to be compared with p̃Q = 0.2 given in Table 10.3 for the transition
(J = 1, Ju = 1) under the unpolarized lower level approximation. These examples
NON-EQUILIBRIUM ATOMIC PHYSICS 551
show that atomic polarization of the lower level is reduced, but not completely re-
moved, by the magnetic field . The complete cancellation of lower-level polarization
can only occur in the presence of depolarizing collisions.
10.9. The Two-Level Atom: the Role of Stimulated Emission
(cylindrical symmetry - no inelastic collisions)
In this section we further enlarge the treatment of resonance polarization in a two-
level atom by taking stimulated emission (besides lower-level atomic polarization)
into account. Apart from this generalization, the physical conditions under which
the phenomenon is studied are the same as in the two previous sections: the incom-
ing radiation is assumed to be unpolarized and cylindrically symmetrical about a
fixed direction, and to satisfy the flat-spectrum approximation; inelastic and su-
perelastic collisions are neglected. As far as the magnetic field is concerned, we will
consider separately the (simpler) non-magnetic and the magnetic case.1
We begin with the non-magnetic case, and we choose a reference system with the
z-axis parallel to the symmetry axis of the radiation field. As shown in Sect. 10.7,
the only non-zero components of the radiation field tensor are J00 (ν0 ) and J02 (ν0 ),
and the only non-zero statistical tensors are those with K even and Q = 0. The
statistical equilibrium equations (identical with Eqs. (10.60) except for the presence
of the stimulated emission rates) are easily deduced from Eqs. (10.1) and (10.2)
d K
ρ (α J ) = − A(αu Ju → α J ) + D(K) (αu Ju ) ρK 0 (αu Ju )
dt 0 u u
+ TA (αu Ju K 0, α J K 0) ρK
0 (α J )
K even
− RS (αu Ju K 0 K 0) ρK
0 (αu Ju )
K even
!
d K Ju Ju K
ρ (α J ) = (2Ju + 1) A(αu Ju → α J ) (−1) 1+J +Ju
ρK0 (αu Ju )
dt 0 J J 1
− RA (α J K 0 K 0) + δKK D(K) (α J ) ρK
0 (α J )
K even
+ TS (α J K 0, αu Ju K 0) ρK
0 (αu Ju ) , (10.76)
K even
where K is even and the rates TA and RA are given by Eqs. (10.61), while the
rates TS and RS can be obtained from Eqs. (7.14c,f) with the help of Eqs. (2.26a),
(2.36a) and (2.49)
1
The precise meaning of the flat-spectrum approximation is different in the two cases: see
Sects. 10.7 and 10.8.
552 CHAPTER 10
TS (α J K 0, αu Ju K 0) = (2Ju + 1) B(αu Ju → α J )
⎡
!
⎣ J J K
× δKK (−1) 1+J +Ju
J00 (ν0 )
Ju Ju 1
⎧ ⎫ ⎤
⎨ J Ju 1⎬
K K 2
+ 15(2K + 1)(2K + 1) J Ju 1 J02 (ν0 )⎦
⎩ ⎭ 0 0 0
K K 2
RS (αu Ju K 0 K 0) = (2Ju + 1) B(αu Ju → α J )
⎡
1
× ⎣ δKK J00 (ν0 ) + (−1)1+J −Ju 15(2K + 1)(2K + 1)
2Ju + 1
⎤
! !
K K 2 1 1 2 K K 2
× J02 (ν0 )⎦ .
Ju Ju Ju Ju Ju J 0 0 0
In order to illustrate the effects of stimulated emission, we apply Eqs. (10.76)
to a few sample transitions, beginning with the simple case (J = 0, Ju = 1). The
obvious advantage of this transition is that lower-level atomic polarization is zero
by definition.1 It should be emphasized that, in general, lower-level polarization
should not be neglected when stimulated emission starts becoming important . This
is because the radiative lifetime of the lower level becomes comparable with that
of the upper level and, even invoking depolarizing collisions, it is impossible to find
a physical regime where the upper level is polarized while the lower level is unpo-
larized. For the transition (J = 0, Ju = 1) we obtain from the first of Eqs. (10.76),
recalling Eqs. (7.8)
1 d 0 √
ρ0 (αu Ju ) = − 1 + n̄ ρ00 (αu Ju ) + 3 n̄ ρ00 (α J )
A(αu Ju → α J ) dt
1
− √ n̄w ρ20 (αu Ju )
2
1 d 2 1
ρ0 (αu Ju ) = − 1 + n̄ − n̄w + δu(2) ρ20 (αu Ju )
A(αu Ju → α J ) dt 2
.
1 3
− √ n̄w ρ00 (αu Ju ) + n̄w ρ00 (α J ) ,
2 2
1
The same simplification occurs for the transitions (J = 1/2, Ju = 1/2) and (J = 1/2, Ju = 3/2),
where lower-level polarization is zero in the presence of an unpolarized, cylindrically symmetrical
radiation field. In fact, the only statistical tensors that can be defined for J = 1/2 are ρ00 and
ρ1Q , and the latter are zero because K is odd.
NON-EQUILIBRIUM ATOMIC PHYSICS 553
where the notations of Eqs. (10.62) have been used. In stationary situations, the
solution can be written in the form
1 w
σ02 (αu Ju ) = √
2 1 + n̄ 1 − w − 12 w2 + δu(2)
1
2
ρ00 (αu Ju ) √ 1 + n̄ 1 − 12 w − 12 w2 + δu(2)
= 3 n̄ . (10.77)
ρ00 (α J ) 1 + n̄ 1 + n̄ − 12 n̄w + δu(2) − 12 n̄2 w2
Obviously, we get back previous results if we neglect stimulated emission: for n̄ 1
we have
1 w ρ00 (αu Ju ) √
σ02 (αu Ju ) = √ (2)
, = 3 n̄ ,
2 1 + δu ρ00 (α J )
which are the same expressions obtained from Eqs. (10.50) under the physical
K
conditions considered here ( = 0, Hu = 0, JQ = 0 except for J00 and J02 ).
In general, the first of Eqs. (10.77) shows that stimulated emission reduces the
atomic polarization of the upper level, except in the special case w = 1 (unidi-
rectional, unpolarized
√ beam) where it has no effect. The second equation yields
the classical value 3 n̄ /(1 + n̄) either for an isotropic, unpolarized radiation field
(w = 0), or under the limit of extremely strong collisional rates (δu(2) → ∞). In
general, it shows that – even for the simple transition (J = 0, Ju = 1) and for
a quantity that may be expected to be rather insensitive to anisotropy phenom-
ena (the ratio of the upper to the lower level population) – the interplay of the
anisotropy itself, the intensity of the radiation field and depolarizing collisions is
very complicated and can produce large differences from the case where polarization
phenomena are neglected.
Next we consider the simplest transition where lower-level atomic polarization
plays a role, i.e., (J = 1, Ju = 0). From Eqs. (10.76) we obtain, using again
Eqs. (7.8) and (10.62)
1 d 0 1
ρ0 (αu Ju ) = − 1 + n̄ ρ00 (αu Ju ) + √ n̄ ρ00 (α J )
A(αu Ju → α J ) dt 3
1
+ √ n̄w ρ20 (α J )
6
1 d 2 1 1
ρ0 (α J ) = − n̄ 1 − w + δ(2) ρ20 (α J )
A(αu Ju → α J ) dt 3 2
1 1
− √ n̄w ρ00 (α J ) + √ n̄w ρ00 (αu Ju ) .
3 2 6
In stationary situations, the solution can be written in the form
√ w
σ02 (α J ) = − 2
2 − w + 2 δ(2) + n̄ 2 − w − w2 + 2 δ(2)
ρ00 (αu Ju ) 1 2 − w − w2 + 2 δ(2)
= √ n̄ . (10.78)
0
ρ0 (α J ) 3 2 − w + 2 δ(2) + n̄ 2 − w − w2 + 2 δ(2)
554 CHAPTER 10
Fig.10.9. Fractional polarization of the radiation scattered at 90 ◦ from the incident (unpolarized)
beam as a function of the beam intensity, for different atomic transitions (J , Ju ) involving integral
J values.
Comparison with Eqs. (10.64) shows that stimulated emission reduces the polariza-
tion of the lower level (except in the extreme case w = 1 and δ(2) = 0 ). The ratio
of the upper to the lower level population depends again in a complicated way on
the different parameters (anisotropy factor, intensity, depolarizing collisions). Only
for an isotropic, unpolarized radiation field (w = 0), or for strong collisional
√ rates
(δ(2) → ∞), the second of Eqs. (10.78) yields the classical value (1/ 3) n̄ /(1 + n̄).
For more complicated transitions, the influence of stimulation effects on atomic
polarization can be evaluated numerically, by solving Eqs. (10.76) – in stationary
situations – together with the trace equation (10.4). Rather than atomic polar-
ization itself, it is probably more instructive to examine the polarization of the
radiation scattered by the atom in a given direction. If we refer for instance to the
geometry of Fig. 10.6 (with B = 0) and consider the frequency-integrated fractional
polarization scattered in the x-direction (i.e., at 90◦ from the symmetry axis of the
incident radiation), taking the y-axis as reference direction for Q, we have from
Eqs. (10.75)
3 wJ(2)J σ02 (αu Ju )
p̃Q = √ u
, p̃U = 0 .
2 2 − wJ(2)J σ02 (αu Ju )
u
Figures 10.9 and 10.10 show the results of such calculations for several atomic
NON-EQUILIBRIUM ATOMIC PHYSICS 555
Fig.10.10. Same as Fig.10.9 for transitions involving half-integral J values.
transitions. Both figures refer to the collisionless, maximum anisotropy regime
(δu(K) = δ(K) = 0, w = 1: see Eqs. (10.62)). Except for a few special cases, the
polarization of the scattered radiation decreases (in absolute value) with increasing
n̄. However, an outstanding difference comes out between transitions involving
integral and half-integral J values: for n̄ → ∞ the polarization tends to a non-zero
value (dependent on the individual transition) in the former case, while it tends to
zero for all transitions in the latter.
The decrease of atomic polarization due to stimulated emission is not surpris-
ing. Atomic polarization is due to the fact that, in an anisotropic environment,
the transitions between certain magnetic sublevels of the upper and lower level
(characterized by their own quantum numbers Mu , M ) are more efficient than the
others in absorbing radiation, which results in a population unbalance. When stim-
ulated emission is important, however, just the same transitions are more efficient
in emitting radiation, and the population unbalance is thus reduced.
Let’s now examine how the above results are modified by the presence of a mag-
netic field. To obtain the statistical equilibrium equations in a reference system
with the z-axis parallel to the symmetry axis of the incoming radiation, we have
just to add in the right-hand sides of Eqs. (10.72) the contributions of stimulated
emission, i.e., the term (see Eq. (10.1))
− RS (αu Ju KQK Q) ρK
Q (αu Ju )
K even
556 CHAPTER 10
in the first equation, and the term (see Eq. (10.2))
+ TS (α J KQ, αu Ju K Q) ρK
Q (αu Ju )
K even
in the second. The expressions of the rates TS and RS can be derived from
Eqs. (7.14c,f) with the help of Eqs. (2.26a), (2.36a) and (2.49), bearing in mind
that the index K is an even integer and that the only non-zero components of the
radiation field tensor are J00 (ν0 ) and J02 (ν0 )
RS (αu Ju KQK Q) = (2Ju + 1) B(αu Ju → α J )
⎡
1
× ⎣ δKK J00 (ν0 ) + (−1)1+J −Ju + Q 15(2K + 1)(2K + 1)
2Ju + 1
⎤
! !
K K 2 1 1 2 K K 2
× J02 (ν0 )⎦
Ju Ju Ju Ju Ju J Q −Q 0
TS (α J KQ, αu Ju K Q) = (2Ju + 1) B(αu Ju → α J )
⎡
!
⎣ J J K
× δKK (−1) 1+J +Ju
J00 (ν0 )
Ju Ju 1
⎧ ⎫ ⎤
⎨ J Ju 1 ⎬
K K 2
+ (−1)Q 15(2K + 1)(2K + 1) J Ju 1 J02 (ν0 )⎦ .
⎩ ⎭ −Q Q 0
K K 2
In stationary situations, the statistical equilibrium equations can be solved numer-
ically for any given transition (J , Ju ), and the polarization of the scattered radi-
ation can then be evaluated. As an example, we present in Fig. 10.11 the results
of such calculations for the transition (J = 1, Ju = 2). The scattering geometry is
the same as in Figs. 10.7-10.8, and the polarization of the scattered radiation has
been computed from Eqs. (10.75). The figure is directly comparable with Fig. 10.8
(collisionless, maximum anisotropy regime; same Landé factors for the upper and
lower level), with the only difference that the intensity of the incident (unpolar-
ized) beam is now ‘large’ (n̄ = 0.5), so that stimulation effects are important. One
can notice the overall reduction of the polarization (consistent with the results of
Fig. 10.9) and the disappearance of the double-lobe shape of the diagram. The
lower-level Hanle effect mixes with the ‘usual’ Hanle effect because the lifetimes of
the upper and lower level are now of the same order of magnitude.
NON-EQUILIBRIUM ATOMIC PHYSICS 557
Fig.10.11. Same as Fig.10.8 for n̄ = 0.5 (intense irradiation of the model atom).
10.10. Three-Level Atoms: the Stepwise Excitation
For model atoms composed of more than two levels, the statistical equilibrium
equations become more and more involved and depend on a growing number of
physical parameters. However, according to the specific model atom and to the
characteristics of the incoming radiation, it is sometimes possible to simplify con-
siderably the problem. One of such cases, which can even be treated analytically,
is illustrated in this section.
Consider the three-level model atom of Fig. 10.12, where, by assumption, tran-
sitions between levels a,b and between levels b,c are allowed, while the transition
between levels a,c is forbidden. We assume the incident radiation field at each of
the Bohr frequencies νba , νcb to be very diluted,
n̄ba 1 , n̄cb 1 , (10.79)
where n̄ba , n̄cb are the average numbers of photons per mode at frequencies νba ,
νcb respectively. This situation is often referred to as stepwise excitation. For
simplicity, we assume that there is no magnetic field and we neglect inelastic and
superelastic collisions, while we retain depolarizing collisions.
An order-of-magnitude estimate shows that the statistical tensors of levels a, b,
c are of order
1, n̄ba , n̄ba n̄cb ,
558 CHAPTER 10
Fig.10.12. Three-level model atom for stepwise excitation.
respectively. Let us consider the statistical equilibrium equation for the statistical
tensors of the intermediate level (level b). From Eqs. (7.11) and (7.101) we have,
using shorthand notations
d
ρ = TA ρa + TE + TS ρc − RA + RE + RS ρb − D ρb . (10.80)
dt b
The order of magnitude of the different terms is the following (see Eqs. (7.14) and
(10.62))
TA ρa ≈ Aba n̄ba RA ρb ≈ Acb n̄ba n̄cb
TE ρc ≈ Acb n̄ba n̄cb RE ρb ≈ Aba n̄ba
TS ρc ≈ Acb n̄ba n̄2cb RS ρb ≈ Aba n̄2ba D ρb ≈ D n̄ba , (10.81)
where Aba and Acb are the Einstein coefficients for spontaneous emission of the
transitions (a, b) and (b, c), respectively. Equations (10.80)-(10.81) show that, to
first order in the number of photons per mode, only the rates TA , RE and D make
a contribution to the time derivative of ρb : it follows that level b is only coupled
with the lower level a.1 The same reasoning shows that stimulated emission can
be disregarded in both transitions (a, b) and (b, c). The three-level model atom
considered can therefore be regarded as the combination of two ‘superposed’ two-
level atoms. In stationary situations, the statistical tensors ρK
Q (b) are obtained as
the solution to the statistical equilibrium equations for the two-level atom (a, b).
The statistical tensors ρKQ (c) are then obtained as the solution to the statistical
equilibrium equations for the two-level atom (b, c), with ρK Q (b) known from the
previous solution.
As an illustration, we consider a three-level model atom with Ja = 0, Jb = 1,
Jc = 2. We assume for simplicity that the incoming radiation (frequency-inde-
pendent around the Bohr frequencies νba and νcb ) is unpolarized and cylindrically
1
This approximation implies indeed the further condition Acb n̄cb Aba , which follows from
Eq. (10.79) provided the order of magnitude of the two Einstein coefficients is the same.
NON-EQUILIBRIUM ATOMIC PHYSICS 559
symmetrical about a fixed direction: this implies, as shown in Sect. 10.7, that only
the statistical tensors with K even and Q = 0 are non-zero. Following Eqs. (10.62),
we introduce the anisotropy factors in the two lines
wba J 2 (ν ) wcb J 2 (ν )
√ = 00 ba , √ = 00 cb .
2 J0 (νba ) 2 J0 (νcb )
In stationary situations, the statistical tensors of level b – considered as the upper
level of the two-level atom (a, b) – are obtained from Eqs. (10.68)
ρ20 (αb Jb ) w ρ00 (αb Jb ) √
0 = √ ba (2) , = 3 n̄ba , (10.82)
ρ0 (αb Jb ) 2 1 + δb ρ00 (αa Ja )
(2)
where we have used the value w10 = 1 given by Table 10.1 and where, according
to Eqs. (10.62)
D(2) (αb Jb )
δb(2) = .
Aba
The statistical equilibrium equations for level c – considered as the upper level of
the two-level atom (b, c) – are given by the first of Eqs. (10.60). Using Eqs. (10.61)
and (10.62) we obtain, after some algebra
.
1 d 0 5 1
ρ (α J ) = − ρ00 (αc Jc ) + n̄cb ρ00 (αb Jb ) + √ n̄cb wcb ρ20 (αb Jb )
Acb dt 0 c c 3 2 30
.
1 d 2 1 7
ρ (α J ) = − 1 + δc(2) ρ20 (αc Jc ) + n̄ w ρ0 (α J )
Acb dt 0 c c 2 6 cb cb 0 b b
.
1 7 1
+ n̄cb 1 + wcb ρ20 (αb Jb )
2 3 7
.
1 d 4 3
ρ0 (αc Jc ) = − 1 + δc ρ0 (αc Jc ) + 3
(4) 4
n̄ w ρ2 (α J ) ,
Acb dt 35 cb cb 0 b b
where
D(K) (αc Jc )
δc(K) = .
Acb
In stationary situations, eliminating ρ00 (αb Jb ) and ρ20 (αb Jb ) via Eqs. (10.82), one
finally gets
. (2)
2 1 7 wcb 1 + δb + wba 1 + 17 wcb 1
σ0 (αc Jc ) = (2)
2 10 1
1 + δb + 20 wcb wba 1 + δc(2)
9 wcb wba 1
σ04 (αc Jc ) = √ (2)
5 14 1 + δ b + 1
w w
20 cb ba 1 + δc(4)
ρ00 (αc Jc ) √ 1 wcb wba
= 5 n̄cb n̄ba 1 + . (10.83)
0
ρ0 (αa Ja ) 20 1 + δb(2)
560 CHAPTER 10
These expressions show that the amount of atomic polarization of level c depends
on the anisotropy factors in both transitions and on the depolarizing rates of levels
b and c, while it is independent of the intensity of the radiation field in either tran-
sition (this is due to the basic characteristics of stepwise excitation n̄ba , n̄cb 1,
Acb n̄cb Aba ). Note that σ02 (αc Jc ) – contrary to σ04 (αc Jc ) – is non-zero even if the
radiation field at the frequency νcb of the upper transition is isotropic (wcb = 0).
In the extreme case of maximum anisotropy (wba = wcb = 1) and no depolarizing
collisions (δb(2) = δc(2) = δc(4) = 0), Eqs. (10.83) give
. .
5 10 6 2
σ02 (αc Jc ) = , σ04 (αc Jc ) = . (10.84)
7 7 7 7
The populations of the Zeeman sublevels of level c can be found by substitut-
ing Eqs. (10.84) into Eq. (3.99). The relative populations turn out to be 3/7
(Mc = ±2), 1/7 (Mc = 0), and 0 (Mc = ±1). This result is easily understood if
we recall that for w = 1 only transitions with ∆M = ±1 can be excited by the
incident radiation (cf. footnote on p. 540). Since wba = 1, in level b only the
sublevels Mb = ±1 are populated;1 and since wcb = 1, in level c only the sublevels
Mc = 0, ±2 are populated. The relative populations are determined by the strength
of the Zeeman transitions. As the transition from the (lower) sublevel Mb = 1 to
the (upper) sublevel Mc = 2 is six times stronger than the transition from Mb = 1
to Mc = 0 (see Eq. (3.16)), the above result is straightforward.
The fractional polarization of the radiation emitted in the upper transition at
90◦ from the symmetry axis of the radiation field is obtained by substituting the
value of σ02 (αc Jc ) into Eq. (10.69). In particular, for the maximum anisotropy,
collisionless case we get, with the help of Eqs. (10.84) and Table 10.1
pQ 0.652 .
This figure can be compared with the value 0.448, corresponding to the two-level
atom (J = 1, Ju = 2), and with the value 0.288 corresponding to the same model
atom with unpolarized lower level (see Table 10.3). The basic result of stepwise
excitation for the three-level atom considered is to increase the polarization of the
radiation scattered in the upper transition.
10.11. Three-Level Atoms: the Raman Effect
Another special case of three-level model atom leading to simple and interesting
results is illustrated in Fig. 10.13. There are two ‘lower’ levels, a and b, but
absorption processes can only take place between one of them (say, level a) and the
upper level c. Level c can then decay to both levels a and b, giving rise to resonance
1 √
This is consistent with the value σ02 (αb Jb ) = 1/ 2 deduced from Eqs. (10.82) for wba = 1
(2)
and δb = 0. According to Table 3.6, such value means that the sublevel Mb = 0 is depopulated.
NON-EQUILIBRIUM ATOMIC PHYSICS 561
Fig.10.13. Three-level model atom for Raman scattering.
scattering radiation and to Raman scattering radiation, respectively. In practice,
the atom is pumped at frequency νca and two lines are scattered, at frequencies
νca and νcb respectively. The Raman line (at frequency νcb ) is called a Stokes line
if νcb < νca or an anti-Stokes line if νcb > νca .
It should be remarked that this phenomenon can be easily produced in the lab-
oratory by selecting the spectral range of the pumping radiation. In astrophysical
plasmas, the situation where level c is pumped by the radiation field in both lines
at the same time is much more common.
The model atom of Fig. 10.13 can be handled at different levels of sophistica-
tion. Here we wish to deduce the main properties of Raman scattering under the
simplest physical conditions. Thus we make the following assumptions: there is no
magnetic field; the incoming radiation (frequency-independent about νca ) is suffi-
ciently diluted to induce negligible stimulation effects; level a is unpolarized; the
collisional rates in the statistical equilibrium equations for the upper level c can be
neglected. Under these assumptions, we have from Eq. (7.11)
d K
ρ (α J ) = TA (αc Jc KQ, αa Ja 0 0) ρ00 (αa Ja )
dt Q c c
− RE (αc Jc KQK Q ) ρKQ (αc Jc ) .
K Q
In stationary situations we obtain, with the use of Eqs. (7.14e), (10.9) and (10.11)1
2Ja + 1 B(αa Ja → αc Jc )
ρK
Q (αc Jc ) = w(K)
2Jc + 1 A(αc Jc → αa Ja ) + A(αc Jc → αb Jb ) Jc Ja
× (−1)Q J−Q
K
(νca ) ρ00 (αa Ja ) . (10.85)
1
Obviously, the schematic model of Fig. 10.13 is not self-consistent: in stationary situations,
all the atoms would be in level b. Here it is understood that level b is coupled (for instance
collisionally) to level a.
562 CHAPTER 10
The emission coefficient in the Raman line is given by Eq. (10.15), which now reads
= hν N 2J + 1 A(α J → α J )
εi (ν, Ω) c c c b b
4π
(K)
× wJ J T Q
K ρK (α J ) φ(ν − ν) .
(i, Ω) (10.86)
c b
Q c c cb
KQ
Substitution of Eq. (10.85) into Eq. (10.86) yields, with the help of Eq. (10.6)
= hν N B(α J → α J ) b φ(ν − ν)
εi (ν, Ω)
4π a a a c c r cb
× WK (Ja , Jc , Jb ) (−1)Q T Q
K J K (ν ) ,
(i, Ω) (10.87)
−Q ca
KQ
where Na is the overall population of level a, br is the branching ratio
A(αc Jc → αb Jb )
br = ,
A(αc Jc → αa Ja ) + A(αc Jc → αb Jb )
and the new symbol WK (J , Ju , J ) is defined by (see Eq. (10.11))1
WK (J , Ju , J ) = wJ(K)J wJ(K)J
u u
! !
J −J 1 1 K 1 1 K
= (−1) 3(2Ju + 1) . (10.88)
Ju Ju J Ju Ju J
Comparison of Eqs. (10.87) and (10.16) shows that, apart from the obvious presence
of the branching ratio and the frequency change of the emitted radiation, Raman
scattering is described by an expression strictly similar to that for resonance scat-
tering. The only difference is that the symbol WK (J , Ju ) of resonance scattering
is replaced by the symbol WK (J , Ju , J ). This difference is non-trivial because for
certain combinations of the three J quantum numbers, W2 (J , Ju , J ) turns out to
be negative. In such cases, the direction of Raman scattering polarization produced
by unpolarized incident radiation is parallel to the scattering plane (although the
lower level a is by assumption unpolarized).2
Numerical values of the symbol WK (J , Ju , J ) can be obtained from Eq. (10.88)
and Table 10.1. It is worth noticing that wJ(2)J is negative for J = Ju (except the
u
special case J = Ju = 1/2 for which wJ(2)J = 0) and positive for J = Ju : therefore
u
W2 (J , Ju , J ) is negative when J = J and either Ju = J or Ju = J . From
Eqs. (10.14) and (10.17) it follows that
W0 (J , Ju , J ) = 1 , WK (J , Ju , J ) = WK (J , Ju ) .
1
The symbol WK (J , Ju , J ) was first introduced by Landi Degl’Innocenti (1984).
2
We recall that resonance scattering polarization, for unpolarized lower level and unpolarized
incident radiation, is always perpendicular to the scattering plane (see Sect 10.2).
NON-EQUILIBRIUM ATOMIC PHYSICS 563
ν
ν
Fig.10.14. Three-level model atom leading to population inversion.
10.12. Three-Level Atoms: an Example Leading to
Population Inversion
A particularly interesting three-level model atom is illustrated in Fig. 10.14. There
are two lower levels (a and b), both connected to the upper level c by radiative
transitions at the frequencies νca and νcb , respectively. The Einstein coefficient for
the transition between a and b is assumed to be very small, so that such transition
can be disregarded in the statistical equilibrium equations for levels a and b.
We study this model atom under the following set of assumptions: no mag-
netic field; no inelastic and superelastic collisions; the radiation field is frequency-
independent about νca and νcb , although its intensity at the two frequencies may be
different; it is unpolarized and cylindrically symmetrical about a fixed direction,
although the amount of anisotropy at either frequency may be different; finally,
it is sufficiently diluted to produce negligible stimulation effects. We recall from
Sect. 10.7 that under such conditions, if we choose a reference system with the
z-axis parallel to the symmetry axis of the radiation field, the only non-zero com-
ponents of the radiation field tensor are J00 and J02 , and that in stationary situations
the only non-zero statistical tensors are those with K even and Q = 0.
The statistical equilibrium equations are easily deduced from Eqs. (7.11) and
(7.101). With the help of Eqs. (7.14b,e) we obtain
d K
ρ (α J ) = − A(αc Jc → αa Ja ) + A(αc Jc → αb Jb ) + D(K) (αc Jc ) ρK 0 (αc Jc )
dt 0 c c
+ TA (αc Jc K 0, αa Ja K 0) ρK
0 (αa Ja )
K even
+ TA (αc Jc K 0, αb Jb K 0) ρK
0 (αb Jb )
K even
!
d K Jc Jc K
ρ0 (αa Ja ) = (−1)1+Ja +Jc (2Jc + 1) A(αc Jc → αa Ja ) ρK
0 (αc Jc )
dt Ja Ja 1
− RA (αa Ja K 0 K 0) + δKK D(K) (αa Ja ) ρK
0 (αa Ja ) ,
K even
564 CHAPTER 10
where K is even and where the rates TA and RA are given by Eqs. (10.61); the equa-
tion for dρK K
0 (αb Jb )/dt is the same as that for dρ0 (αa Ja )/dt with the substitutions
αa → αb , Ja → Jb .
We wish to illustrate the typical results arising from the solution of the statis-
tical equilibrium equations on a particularly simple example. Assuming Ja = 1/2,
Jb = 3/2, Jc = 1/2, the atom is described by the four statistical tensors ρ00 (Ja ),
ρ00 (Jb ), ρ20 (Jb ), ρ00 (Jc ). After some algebra we obtain the following equations
1 d 0
ρ (α J ) = ρ00 (αc Jc ) − n̄ca ρ00 (αa Ja )
A(αc Jc → αa Ja ) dt 0 a a
1 d 0 1 1
ρ (α J ) = √ ρ00 (αc Jc ) − n̄cb ρ00 (αb Jb )
A(αc Jc → αb Jb ) dt 0 b b 2 2
1
− n̄cb wcb ρ20 (αb Jb )
4
1 d 2 1 1
ρ (α J ) = − n̄cb wcb ρ00 (αb Jb ) − n̄cb 1 + δb(2) ρ20 (αb Jb ) ,
A(αc Jc → αb Jb ) dt 0 b b 4 2
where, according to Eqs. (10.62) and (10.63), we have introduced the anisotropy
factor
wcb J 2 (ν )
√ = 00 cb ,
2 J0 (νcb )
the average numbers of photons per mode (much less than unity by assumption)
B(αa Ja → αc Jc ) 0 B(αb Jb → αc Jc ) 0
n̄ca = J (ν ) , n̄cb = 2 J (ν ) ,
A(αc Jc → αa Ja ) 0 ca A(αc Jc → αb Jb ) 0 cb
and the quantity
D(2) (αb Jb )
δb(2) = 2 .
n̄cb A(αc Jc → αb Jb )
In stationary situations, the solution to these equations can be written in the form
√
ρ00 (αb Jb ) 4 2 1 + δb(2) n̄ca
= (10.89a)
0
ρ0 (αa Ja ) 4 1 + δb − wcb n̄cb
(2) 2
w
σ02 (αb Jb ) = − cb (2) (10.89b)
2 1 + δb
ρ00 (αc Jc )
= n̄ca . (10.89c)
ρ00 (αa Ja )
These expressions are very interesting since they predict that, depending on the
physical situation, different ‘kinds’ of population inversion between the lower levels
a and b may occur.
From Eqs. (10.89a) and (3.108), the ratio of the overall populations of levels a
and b is given by
nα J 8 1 + δb(2) n̄ca
b b
= . (10.90)
nα J
a a
4 1 + δb − wcb n̄cb
(2) 2
NON-EQUILIBRIUM ATOMIC PHYSICS 565
If the incident radiation is isotropic (wcb = 0), the populations of the two levels
are proportional to the corresponding statistical weights when n̄ca = n̄cb ,
nα Jb
b
=2;
nα Ja
a
for n̄ca > n̄cb there is an overall population inversion.
However, in astrophysical situations n̄ca is usually smaller than n̄cb . Consider
for instance the case where the atom is irradiated, within a solid angle Ω, by a
black-body radiation at temperature T . In that case we have
hν hν
Ω − k ca Ω − k cb
n̄ca = e BT , n̄cb = e BT ,
4π 4π
where the Wien limit has been taken consistently with the assumption of neglecting
stimulated emission in the radiative rates. Thus
hν
n̄ca − ba
= e kB T ,
n̄cb
or
n̄ca hν
≈ 1 − ba
n̄cb kB T
when the energy difference between levels a and b is very small.
On the other hand, Eq. (10.90) shows that, in the presence of anisotropic incident
radiation (wcb > 0), population inversion between levels a and b can take place
even if n̄ca < n̄cb . This occurs when
., -, n̄ -
wcb > 2 1 + δb(2) 1 − ca .
n̄cb
In a low-density medium where depolarizing collisions are negligible, this condition
becomes
hνba
wcb > 2 , (10.91)
kB T
which shows that a weak anisotropy of the incident radiation can easily induce
population inversion of levels a and b, provided the two levels are sufficiently close
in energy.
It is interesting to consider the relative populations of the Zeeman sublevels of
levels a and b for given values of wcb , δb(2) and n̄ca /n̄cb. From Eqs. (3.99) and
(10.89a,b) we obtain
ρ(αb Jb , Mb = ± 1/2) wcb 4 1 + δb(2) n̄ca
= 1+
ρ(αa Ja , Ma = ± 1/2) 2 1 + δb (2)
4 1 + δb − wcb n̄cb
(2) 2
ρ(αb Jb , Mb = ± 3/2) w 4 1 + δb(2) n̄ca
= 1 − cb (2) . (10.92)
ρ(αa Ja , Ma = ± 1/2) 2 1 + δb 4 1 + δb(2) − wcb2 n̄
cb
566 CHAPTER 10
Fig.10.15. The Zeeman sublevels (artificially split) corresponding to levels a and b of the model
atom of Fig.10.14, for Ja = 1/2, Jb = 3/2, Jc = 1/2. Under the conditions specified in the text,
their populations are proportional to the number of dots drawn on each of them.
In the extreme case wcb = 1, δb(2) = 0, n̄ca = n̄cb we have
ρ(αa Ja , Ma = ± 1/2) ρ(αb Jb , Mb = ± 1/2) ρ(αb Jb , Mb = ± 3/2)
= = .
3 6 2
A pictorial representation of this situation is given in Fig. 10.15, which clearly
shows that, in this case, it is more appropriate to speak about selective popula-
tion inversion rather than population inversion tout court. In fact, the sublevels
Mb = ± 1/2 of level b are overpopulated compared to those of the lower level a,
while the sublevels Mb = ± 3/2 are not.
In the case illustrated in Fig. 10.15, selective population inversion occurs together
with an overall population inversion of levels a and b (nα J /nα J = 8/3 > 2).
b b a a
However, under suitable physical conditions, it is possible to have selective popula-
tion inversion without overall population inversion. According to Eqs. (10.92) and
(10.90), this occurs when
, n̄ -2 , -, .,
n̄ - n̄ -, n̄ -
ca
+ 4 1 + δb(2)
1 − ca − ca < wcb < 2 1 + δb(2) 1 − ca ,
n̄cb n̄cb n̄cb n̄cb
which for low-density media (δb(2) = 0) reduces to (cf. Eq. (10.91))
hνba hνba
2 < wcb < 2 .
kB T kB T
The presence of selective population inversion may have important consequences
on the polarization of the radiation, which can be amplified through maser action
in the transition between levels a and b.
The mechanism for selective population inversion described in this section is
very robust. Inspection of Eqs. (10.89a,b) shows that, if depolarizing collisions
NON-EQUILIBRIUM ATOMIC PHYSICS 567
are negligible, the results are independent of the values of the Einstein coefficients
of the transitions connecting the lower levels a,b with the upper level c. Under
the same condition, the results are not affected by the radiation intensity at the
individual frequencies νca and νcb , they only depend on their ratio. It follows that,
provided this ratio is not too far from unity, selective population inversion can take
place irrespective of the distance between the atom and the radiating source.
It should be remarked that selective population inversion (without overall popu-
lation inversion) can easily arise also in the ‘inverted’ three-level model atom having
Ja = 3/2, Jb = 1/2, Jc = 1/2. In the case of maximum anisotropy, neglecting de-
polarizing collisions and assuming n̄ca /n̄cb = 1, one obtains the same result as in
Fig. 10.15 except for the exchanged energy position of the two levels. The sublevels
M = ± 1/2 of the (upper) level J = 1/2 show population inversion with respect to
the levels M = ± 3/2 of the (lower) level J = 3/2.
10.13. The Weak Anisotropy Approximation
In some environments of astrophysical interest, like for instance in stellar atmo-
spheres, the anisotropy of the radiation field interacting with the atomic systems
is rather weak. This means that the ‘off-diagonal’ components of the radiation
field tensor are much smaller (in absolute value) than the mean intensity J00 (ν). In
formulae, defining
K
JQ (ν)
sK
Q (ν) = , (10.93)
J00 (ν)
we have 3 K 3
3 s (ν) 3 1 (K = 1, 2) (10.94)
Q
at all frequencies of interest.
In this situation it is possible to simplify considerably the statistical equilibrium
equations (written in the spherical statistical tensor representation) via a pertur-
bative expansion. This can formally be obtained by multiplying the components
1 2
JQ (ν) and JQ (ν) by the real parameter λ (which at the end of the calculations will
be set to 1) and by expanding the statistical tensors as a function of λ. Referring
for instance to the equations for the multi-level atom written in the ‘magnetic’
frame (Eq. (7.11) supplemented in the right-hand side with the collisional rates of
Eq. (7.101)), we set
∞
K (n) n
ρK
Q (αJ) = ρQ (αJ) λ , (10.95)
n=0
where [ρK
Q (αJ)]
(n)
is the n-th order contribution to the statistical tensor ρK Q (αJ).
Substitution of Eq. (10.95) into the statistical equilibrium equations leads to an
equality between two power series in λ. Since λ is arbitrary, the coefficients of each
power must separately be equal.
To the lowest order, the radiative rates TA , TS , RA , RS (the only rates connecting
statistical tensors with different K, Q values) contain only the J00 component of
568 CHAPTER 10
the radiation field tensor. Bearing in mind their expressions (Eqs. (7.14)), it is
easily seen that the statistical equilibrium equations split into separate systems of
equations, each one involving statistical tensors with the same value for K and Q.
Since the trace equation involves only populations (proportional to ρ00 ), in station-
ary situations all the statistical tensors with K = 0 are zero. For K = 0 we obtain,
with the use of Eqs. (7.14), (2.26a), (2.36a) and (2.49)
d √ (0) (0)
2J + 1 ρ00 (αJ) = B(α J → αJ) J00 (ναJ, α J ) 2J + 1 ρ00 (α J )
dt
α J
(0)
+ A(αu Ju → αJ) + B(αu Ju → αJ) J00 (να J , αJ ) 2Ju + 1 ρ00 (αu Ju )
u u
αu Ju
− B(αJ → αu Ju ) J00 (να J , αJ ) + A(αJ → α J )
u u
αu Ju α J
√ (0)
+ B(αJ → α J ) J00 (ναJ, α J ) 2J + 1 ρ00 (αJ)
α J
(0)
+ CI(0) (αJ, α J ) 2J + 1 ρ00 (α J )
α J
(0)
+ CS(0) (αJ, αu Ju ) 2Ju + 1 ρ00 (αu Ju )
αu Ju
√ (0)
− CI(0) (αu Ju , αJ) + CS(0) (α J , αJ) 2J + 1 ρ00 (αJ) , (10.96)
αu Ju α J
and the trace equation reads (see Eqs. (3.84) and (3.108))
(0) √ (0)
nαJ = 2J + 1 ρ00 (αJ) =1.
αJ αJ
Thus the zero-order equations coincide with the usual statistical equilibrium equa-
tions for the level populations of the scalar (unpolarized) case – see e.g. Mihalas,
1978).
Passing to the first order, it is convenient to consider separately the equations
Q (αJ)] /dt with K = 0. The former turn out to
for d [ρ00 (αJ)](1) /dt and for d [ρK (1)
be formally identical to Eqs. (10.96), but the trace equation is now
√ (1)
2J + 1 ρ00 (αJ) =0.
αJ
This means that, in stationary situations, the first-order correction to ρ00 (αJ) is
identically zero. For K = 0 we obtain, using again Eqs. (7.14), (2.26a), (2.36a) and
(2.49)
NON-EQUILIBRIUM ATOMIC PHYSICS 569
d K (1) (1)
ρQ (αJ) = −2πi νL gαJ Q ρKQ (αJ)
dt .
2J + 1 & (1)
+
B(α J → αJ) yJJ (K)
J00 (ναJ, α J ) ρK Q (α J )
2J + 1
α J
(K)
(0) '
+ wJJ (−1)Q J−Q K
(ναJ, α J ) ρ00 (α J )
.
2J + 1 (1)
+ u
A(αu Ju → αJ) yJ(K)J ρK Q (αu Ju )
2J + 1 u
αu Ju
. &
2J + 1 (1)
+ u
B(αu Ju → αJ) yJJ (K)
J00 (να J , αJ ) ρK Q (αu Ju )
2J + 1 u u u
αu Ju
(K)
(0) '
+ wJJ (−1)K+Q J−Q K
(να J , αJ ) ρ00 (αu Ju )
u u u
& K (1)
− B(αJ → αu Ju ) J0 (να J , αJ ) ρQ (αJ)
0
u u
αu Ju
(K)
(0) '
+ wJJ (−1)K+Q J−Q
K
(να J , αJ ) ρ00 (αJ)
u u u
K (1)
− A(αJ → α J ) ρQ (αJ)
α J
& (1)
− B(αJ → α J ) J00 (ναJ, α J ) ρK
Q (αJ)
α J
(K)
(0) '
+ wJJ (−1)Q J−Q
K
(ναJ, α J ) ρ00 (αJ)
.
2J + 1 (K) (1)
+ C (αJ, α J ) ρK
Q (α J )
2J + 1 I
α J
.
2Ju + 1 (K) (1)
+ C (αJ, αu Ju ) ρK
Q (αu Ju )
2J + 1 S
αu Ju
& (0) ' (1)
− CI(0) (αu Ju , αJ) + CS (α J , αJ) + D(K) (αJ) ρKQ (αJ) . (10.97)
αu Ju α J
(K)
In this equation, the symbol wJJ is defined as in Eq. (10.11),
!
(K) 1+J+J 1 1 K
wJJ = (−1) 3(2J + 1) , (10.98)
J J J
(K)
while the symbol yJJ , invariant under interchange of J and J , is defined by
!
1+J+J +K
J J K
y (K)
JJ =y (K)
JJ = (−1) (2J + 1)(2J + 1) , (10.99)
J J 1
570 CHAPTER 10
or, using Eq. (2.36a) !
J J K
J J 1
(K)
yJJ = (−1)
K ! ,
J J 0
J J 1
(0)
which shows that yJJ = 1.
Equation (10.97) has a very peculiar structure. In compact notation it can be
written in the form
d K (1) (1)
ρQ (αi Ji ) = aij ρKQ (αj Jj )
dt j
(0)
+ bij ρ00 (αj Jj ) (K = 0) , (10.100)
j
where the indices i and j run over all the atomic levels (excluding, obviously, those
for which the multipole moment ρK Q is undefined). The coefficients aij contain the
contributions of the magnetic field, of collisions, and of the radiative rates: the
last one, however, is restricted to terms involving the only component J00 of the
radiation field tensor (evaluated at the Bohr frequency connecting levels i and j).
The coefficients bij contain only contributions of the radiative rates (absorption
K
and stimulated emission) restricted to terms involving the components J−Q of the
radiation field tensor with K = 0.
The structure of Eq. (10.100) immediately suggests the method to solve (in sta-
tionary situations, and up to the first order) the statistical equilibrium equations
under the weak anisotropy approximation. First, the values of [ρ00 (αj Jj )](0) are
obtained by solving Eq. (10.96) – the statistical equilibrium equations for the
level populations. These values are then substituted into the right-hand side of
Eq. (10.100) which reduces to an algebraic, inhomogeneous system in the unknowns
[ρK
Q (αi Ji )]
(1)
with K = 0. The substantial simplification introduced by the weak
anisotropy approximation is that this system is formed by a set of uncoupled sub-
systems, each characterized by its own K, Q values.
As apparent from the structure of Eq. (10.100), the solutions [ρK
Q (αi Ji )]
(1)
will be
K
linear combinations of the multipole components J−Q (evaluated at the different
Bohr frequencies of the model atom considered) with the same K, Q values. An
obvious consequence of this fact is that in the weak anisotropy approximation only
statistical tensors with K ≤ 2 can be non-zero (to first order).
10.14. The Two-Level Atom in the Weak Anisotropy Approximation
In this section we go back to the two-level model atom, already analyzed from
several points of view, and we solve the statistical equilibrium equations under the
assumption of weak anisotropy of the radiation field. Following the method outlined
NON-EQUILIBRIUM ATOMIC PHYSICS 571
in the former section, we will first determine (from the zero-order equations) the
statistical tensors ρ00 , then (from the first-order equations) the statistical tensors
Q with K = 0, both for the upper and the lower level.
ρK
From Eqs. (10.96) we obtain, in stationary situations
ρ00 (αu Ju ) 2J + 1
=
ρ00 (α J ) 2Ju + 1
B(α J → αu Ju ) J00 (ν0 ) + CI(0) (αu Ju , α J )
× ,
A(αu Ju → α J ) + B(αu Ju → α J ) J00 (ν0 ) + CS(0) (α J , αu Ju )
where ν0 is the transition frequency and where, for conciseness, the index (0) of
the statistical tensors has been omitted. This expression can be simplified by
dividing numerator and denominator by the Einstein coefficient A(αu Ju → α J ).
Using Eqs. (7.8) and the Einstein-Milne relation between inelastic and superelastic
collisional rates (Eq. (7.98), valid for a Maxwellian distribution of velocities), we
get
ρ00 (αu Ju ) 2Ju + 1 n̄ + b(Tc)
0 = , (10.101)
ρ0 (α J ) 2J + 1 1 + n̄ +
where the dimensionless parameters n̄ and are given by Eqs. (10.62) and (10.51),
respectively, and where b(Tc ) is the Boltzmann factor
hν0
−
kB Tc
b(Tc ) = e ,
with Tc the temperature characterizing the distribution of the colliding particles.
Next we write down Eq. (10.97) for the upper and lower atomic level, and we make
use of Eq. (10.101). This leads, in stationary situations, to an algebraic system
of two equations in the unknowns [ρK Q (αu Ju )]
(1)
and [ρK (1)
Q (α J )] . Omitting the
K
index (1) and introducing the reduced statistical tensors σQ = ρK 0
Q /ρ0 , the solution
1
can be written in the compact form
n̄ 1 + − b(Tc) 1
K
σQ (αu Ju ) = wJ(K)J n̄ 1 + δ(K) + i H Q + b(Tc )
n̄ + b(Tc) ∆ u
!
− (−1) wJ J n̄ yJ J +
K (K) (K) (K)
(−1)Q sK −Q (ν0 )
u u
n̄ 1 + − b(Tc) 1
K
σQ (α J ) = wJ(K)J yJ(K)J 1 + n̄ + (K)
u
1 + n̄ + ∆ u u
!
− (−1)K wJ(K)J 1 + n̄ + + δu(K) + i Hu Q (−1)Q sK
−Q (ν0 ) , (10.102)
u
1 (K) (K)
We recall that yJ J = yJ J (see Eq. (10.99)).
u u
572 CHAPTER 10
where K = 1, 2 and
∆ = 1 + n̄ + + δu(K) + i Hu Q n̄ 1 + δ(K) + i H Q + b(Tc )
− n̄ yJ(K)J 1 + n̄ + (K) u yJ(K)J + (K)
.
u u
In these expressions, all the symbols denote dimensionless quantities: sK Q (ν0 ) is
(K) (K)
defined in Eq. (10.93); wJJ and y JJ in Eqs. (10.98) and (10.99), respectively;
Hu , δu(K) and δ(K) in Eqs. (10.28), (10.51) and (10.63), respectively. The remaining
symbols are defined as follows
CS(K) (α J , αu Ju )
(K)
u =
A(αu Ju → α J )
CI(K) (αu Ju , α J ) 2J + 1 CI(K) (αu Ju , α J )
(K) = =
B(α J → αu Ju ) J0 (ν0 )
0 2Ju + 1 n̄ A(αu Ju → α J )
2πνL gα J 2J + 1 2πνL gα J
H = = . (10.103)
B(α J → αu Ju ) J00 (ν0 ) 2Ju + 1 n̄ A(αu Ju → α J )
The physical meaning of these new quantities is immediate. (K) u represents the
number of superelastic collisions (for the statistical tensors of rank K) taking place
during the lifetime of the upper level and, similarly, (K) the number of inelastic
collisions (for the statistical tensors of rank K) taking place during the lifetime of
the lower level; H (the analogue of Hu ) quantifies the efficiency of the relaxation
process of the lower level due to the magnetic field. Note that the quantities (K) u
and (K)
are related to by the expressions (see Eq. (10.49))
CS(K) (α J , αu Ju ) CI(K) (αu Ju , α J ) b(Tc )
(K)
u = , (K) = .
CS(0) (α J , αu Ju )
CI(0) (αu Ju , α J ) n̄
In the following we study in detail the solutions (10.101) and (10.102), consid-
ering separately the effects of depolarizing collisions, of the magnetic field and of
stimulated emission, and pointing out the analogies (and differences) with previous
results concerning the two-level model atom.
10.14.a No magnetic field - no collisions - no stimulated emission
(K) (K)
This is the simplest case that can be considered. Setting , (K) (K)
u , , δu , δ , Hu
and H to zero, and assuming n̄ 1, Eq. (10.101) reduces to
ρ00 (αu Ju ) 2Ju + 1
= n̄ , (10.104)
ρ00 (α J ) 2J + 1
and Eqs. (10.102) yield the remarkably symmetrical expressions
K
σQ (αu Ju ) = x(K) Q K
J J (−1) s−Q (ν0 )
u
K
σQ (α J ) = (−1)1+K x(K) Q K
J J (−1) s−Q (ν0 ) , (10.105)
u
NON-EQUILIBRIUM ATOMIC PHYSICS 573
where the new symbol x(K)
JJ is given by
1
− (−1)
(K) K (K) (K)
wJJ wJ J yJJ
x(K)
JJ = (K) 2 = (−1)1+J+J 3(2J + 1)
1 − yJJ
! ! !
1 1 K J+J 1 1 K J J K
+ (−1) (2J + 1)
J J J J J J J J 1
× !2 . (10.106)
J J K
1 − (2J + 1)(2J + 1)
J J 1
Using Eqs. (10.104), (10.62) and (10.93), the first of Eqs. (10.105) can be rewritten
in the form
2J + 1 B(α J → αu Ju ) (K)
ρK
Q (αu Ju ) = x (−1)Q J−Q
K
(ν0 ) ρ00 (α J ) .
2Ju + 1 A(αu Ju → α J ) Ju J
This is identical with Eq. (10.13) except for the presence of the quantity x(K)
J J in the u
place of wJ(K)J . It follows that for the two-level atom under the weak anisotropy
u
approximation, resonance scattering is still described by Eq. (10.19), where the
scattering phase matrix is now given by
Pij (Ω, ) =
Ω XK (J , Ju ) (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω (10.107)
−Q
KQ
with
XK (J, J ) = wJ(K) (K)
J xJ J = 3(2J + 1)
!2 ! ! !
1 1 K 1 1 K 1 1 K J J K
+ (−1)J+J (2J + 1)
J J J J J J J J J J J 1
× !2
J J K
1 − (2J + 1)(2J + 1)
J J 1
for K = 1, 2, and
X0 (J, J ) ≡ 1 . (10.108)
Obviously, it should be kept in mind that the incident radiation field in the right-
hand side of Eq. (10.19) has now to satisfy the weak anisotropy approximation
described by Eq. (10.94). For instance, we cannot use Eq. (10.19) with the phase
matrix of Eq. (10.107) to describe the scattering of a radiation beam, because the
weak anisotropy approximation does not hold in that case.
1
Note that for K = 0 the symbol is indeterminate (being of the form 0/0). This is
not surprising, since Eqs. (10.105) are a special case of Eqs. (10.102), which hold for K = 1, 2.
(0)
Equations (10.105) are correct also for K = 0 if we set by definition xJ J = 1 in the former and
(0) u
xJ J = −1 in the latter.
u
574 CHAPTER 10
Numerical values of the symbols yJ(K) (K)
J , xJ J and XK (J, J ) are given, for different
transitions, in Table 10.4. Table 10.5 collects the analytical expressions of X1 (J, J )
and X2 (J, J ) obtained via Eqs. (2.36d) and (2.36h). Note that, contrary to the
symbols W1 and W2 , which satisfy the inequality 0 ≤ WK ≤ 1, the symbols X1 and
X2 can be negative as well as larger than unity. In particular, X2 is negative for
all transitions having Ju = J − 1 (except (J = 1, Ju = 0) and (J = 3/2, Ju = 1/2)
for which X2 is zero). As illustrated at the end of Sect. 10.7, this is due to the
presence of lower-level atomic polarization.
10.14.b No magnetic field - no inelastic collisions - no stimulated emission
Consider now the effect of depolarizing collisions. Neglecting inelastic and supere-
lastic collisions ( = (K)
u = (K)
= 0), the magnetic field (Hu = H = 0) and
stimulated emission (n̄ 1), Eq. (10.101) reduces again to Eq. (10.104), while
Eqs. (10.102) take the form
wJ(K)J 1 + δ(K) − (−1)K wJ(K)J yJ(K)J
K
σQ (αu Ju ) = u
u
2
u
(−1)Q sK
−Q (ν0 )
1 + δu(K) 1 + δ(K) − yJ(K)J
u
K
σQ (α J ) = (−1)1+K
wJ(K)J 1 + δu(K) − (−1)K wJ(K)J yJ(K)J
× u
u
(K) 2
u
(−1)Q sK
−Q (ν0 ) . (10.109)
(K)
1 + δu (K)
1 + δ − yJ J
u
Substitution of Eqs. (10.104), (10.62) and (10.93) into the first of Eqs. (10.109)
yields, for K = 0
2J + 1 B(α J → αu Ju )
ρK
Q (αu Ju ) =
2Ju + 1 A(αu Ju → α J )
wJ(K)J 1 + δ(K) − (−1)K wJ(K)J yJ(K)J
× u
u
2
u
(−1)Q J−Q
K
(ν0 ) ρ00 (α J ) .
1 + δu(K) 1 + δ(K) − yJ(K)J
u
Comparison with Eq. (10.13) shows that resonance scattering is still described by
Eq. (10.19), with a phase matrix given by
Ω
Pij (Ω, ) = XK (J , Ju ) d.c. (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω
−Q
KQ
where the new symbol (the symbol XK (J , Ju ) ‘modified by depolarizing collisions’)
is defined to be
(K) 2
wJ J 1 + δ(K) − (−1)K wJ(K)J wJ(K)J yJ(K)J
u u u u
XK (J , Ju ) d.c. = 2 (K = 1, 2)
1 + δu(K) 1 + δ(K) − yJ(K)J
u
X0 (J , Ju ) d.c. ≡ 1 . (10.110)
NON-EQUILIBRIUM ATOMIC PHYSICS 575
TABLE 10.4
Values of the quantities yJ(K) (K)
J , xJ J , XK (J, J ) for different transitions
J J yJ(1)
J x(1)
J J
X1 (J, J ) yJ(2)
J x(2)
J J
X2 (J, J )
0 1 0. −1. 1. 0. 1. 1.
1/2 1/2 −0.333 −0.612 0.5 0. 0. 0.
1/2 3/2 0.745 −1.369 1.25 0. 0.707 0.5
1 0 0. 0. 0. 0. 0. 0.
1 1 0.5 −1. 0.5 −0.5 −1. 0.5
1 2 0.866 −1.732 1.5 0.592 0.819 0.485
3/2 1/2 0.745 −0.612 −0.25 0. 0. 0.
3/2 3/2 0.733 −1.369 0.5 0.2 −0.471 0.267
3/2 5/2 0.917 −2.092 1.75 0.748 0.962 0.509
2 1 0.866 −1. −0.5 0.592 −0.385 −0.038
2 2 0.833 −1.732 0.5 0.5 −0.394 0.233
2 3 0.943 −2.449 2. 0.828 1.113 0.545
5/2 3/2 0.917 −1.369 −0.75 0.748 −0.579 −0.082
5/2 5/2 0.886 −2.092 0.5 0.657 −0.365 0.221
5/2 7/2 0.958 −2.806 2.25 0.875 1.268 0.587
3 2 0.943 −1.732 −1. 0.828 −0.753 −0.127
3 3 0.917 −2.449 0.5 0.75 −0.350 0.214
3 4 0.968 −3.162 2.5 0.905 1.424 0.631
7/2 5/2 0.958 −2.092 −1.25 0.875 −0.920 −0.174
7/2 7/2 0.937 −2.806 0.5 0.810 −0.341 0.211
7/2 9/2 0.975 −3.518 2.75 0.925 1.581 0.677
4 3 0.968 −2.449 −1.5 0.905 −1.084 −0.221
4 4 0.95 −3.162 0.5 0.85 −0.335 0.208
4 5 0.980 −3.873 3. 0.939 1.738 0.724
9/2 7/2 0.975 −2.806 −1.75 0.925 −1.246 −0.269
9/2 9/2 0.960 −3.518 0.5 0.879 −0.331 0.206
9/2 11/2 0.983 −4.228 3.25 0.950 1.896 0.771
5 4 0.980 −3.162 −2. 0.939 −1.407 −0.318
5 5 0.967 −3.873 0.5 0.9 −0.329 0.205
5 6 0.986 −4.583 3.5 0.958 2.054 0.819
11/2 9/2 0.983 −3.518 −2.25 0.950 −1.568 −0.366
11/2 11/2 0.972 −4.228 0.5 0.916 −0.327 0.204
11/2 13/2 0.988 −4.937 3.75 0.964 2.212 0.867
6 5 0.986 −3.873 −2.5 0.958 −1.727 −0.415
6 6 0.976 −4.583 0.5 0.929 −0.325 0.204
6 7 0.990 −5.292 4. 0.969 2.370 0.916
576 CHAPTER 10
TABLE 10.4
(continued)
(1) (1) (2) (2)
J J yJ J xJ J X1 (J, J ) yJ J xJ J X2 (J, J )
13/2 11/2 0.988 −4.228 −2.75 0.964 −1.887 −0.464
13/2 13/2 0.979 −4.937 0.5 0.938 −0.324 0.203
13/2 15/2 0.991 −5.646 4.25 0.973 2.528 0.965
7 6 0.990 −4.583 −3. 0.969 −2.046 −0.513
7 7 0.982 −5.292 0.5 0.946 −0.323 0.203
7 8 0.992 −6. 4.5 0.976 2.686 1.014
15/2 13/2 0.991 −4.937 −3.25 0.973 −2.205 −0.562
15/2 15/2 0.984 −5.646 0.5 0.953 −0.322 0.202
8 7 0.992 −5.292 −3.5 0.976 −2.364 −0.611
8 8 0.986 −6. 0.5 0.958 −0.321 0.202
TABLE 10.5
Analytical expressions of the quantities XK (J, J )
X1 (J, J ) X2 (J, J )
J +2 (J + 2)(2J + 1)(2J + 5)
J = J + 1 2 10 (4J 2 + 8J + 1)
1 (2J − 1)(2J + 3)
J = J 2 10 (2J 2 + 2J − 3)
(J − 1)(2J + 1)(2J − 3)
J = J − 1 −J − 1 −
2 10 (4J 2 − 3)
This symbol satisfies the obvious relations (cf. Eqs. (10.106), (10.108), (10.17),
(10.54))
lim XK (J , Ju ) d.c. = XK (J , Ju )
(K)
δ →0
(K)
δu →0
lim XK (J , Ju ) d.c. = WK (J , Ju )
(K)
δ →∞
(K)
δu →0
W (J , J )
lim XK (J , Ju ) d.c. = K (K)u .
(K)
δ →∞ 1 + δu
It is worth noticing that analytical results similar to those of Eqs. (10.104)
and (10.109) have already been obtained, for specific transitions, in Sect. 10.7.
Consider for instance Eqs. (10.64) or Eqs. (10.66), which refer to the transitions
(J = 1, Ju = 0) and (J = 1, Ju = 1), respectively. They give the atomic polariza-
tion of a two-level atom under the same physical conditions considered here (no
NON-EQUILIBRIUM ATOMIC PHYSICS 577
inelastic and superelastic collisions, no magnetic field, no stimulated emission) ex-
cept for the incident radiation field: in Sect. 10.7√the only non-zero components of
the radiation field tensor are J00 and J02 = J00 w/ 2, whereas in the present section
any component of the radiation field tensor can be non-zero but it must satisfy the
weak anisotropy approximation |JQ K
| J00 . It is obvious that Eqs. (10.64) and
(10.66) should give the same results as Eqs. (10.104) and (10.109) for w 1.1 As
an example, consider the first of Eqs. (10.66). To first order in w, it reduces to
w 3 + 2 δ(2)
σ02 (αu Ju ) = − √ ,
2 4 1 + δu(2) 1 + δ(2) − 1
which is consistent with√the first of Eqs. (10.109) since (see Eq. (10.93) and Ta-
(2)
bles 10.1, 10.4) s20 = w/ 2 , w11 (2)
= y11 = −1/2 .
10.14.c No inelastic collisions - no stimulated emission
Consider now the effect of the magnetic field. Neglecting inelastic and superelastic
collisions and stimulation effects ( = (K) (K)
u = = 0 ; n̄ 1), Eq. (10.101) reduces
to Eq. (10.104) while Eqs. (10.102) become
wJ(K)J 1 + δ(K) + i H Q − (−1)K wJ(K)J yJ(K)J
K Q K
σQ (αu Ju ) = u (K)
u u
2 (−1) s−Q (ν0 )
1 + δu + i Hu Q 1 + δ(K) + i H Q − yJ(K)J
u
K
σQ (α J ) = (−1)1+K
wJ(K)J 1 + δu(K) + i Hu Q − (−1)K wJ(K)J yJ(K)J
× u
u
u Q K
2 (−1) s−Q (ν0 ) . (10.111)
1 + δu(K) + i Hu Q 1 + δ(K) + i H Q − yJ(K)J
u
Using Eqs. (10.104), (10.62) and (10.93), the first of Eqs. (10.111) can be rewritten
in the form
2J + 1 B(α J → αu Ju )
ρK
Q (αu Ju ) =
2Ju + 1 A(αu Ju → α J )
wJ(K)J 1 + δ(K) + i H Q − (−1)K wJ(K)J yJ(K)J
× u (K)
u
u Q K 0
2 (−1) J−Q (ν0 ) ρ0 (α J ) .
1 + δu + i Hu Q 1 + δ(K) + i H Q − yJ(K)J
u
Comparison with Eq. (10.27) shows that the frequency-integrated radiation emitted
in a scattering process is still given by Eq. (10.32), where the phase matrix is now
Ω
Pij (Ω, ; B
)= XKQ (J , Ju ; B) d.c. (−1)Q T Q
K ) , (10.112)
T K (j, Ω
(i, Ω) −Q
KQ
1
The comparison must of course be restricted to statistical tensors of the form ρ00 or ρ20 .
We recall that, when the radiation field is characterized by the only components J00 and J02 and
the magnetic field is zero, the only non-null statistical tensors are those with K even and Q = 0;
by contrast, under the weak anisotropy approximation, the only non-null statistical tensors are
those with K ≤ 2 and any Q.
578 CHAPTER 10
with the symbol [XKQ (J , Ju ; B)]d.c. – which, contrary to the symbols XK (J , Ju )
and [XK (J , Ju )]d.c. , depends also on the quantum number Q – defined by
XKQ (J , Ju ; B) d.c. =
(K) 2
wJ J 1 + δ(K) + i H Q − (−1)K wJ(K)J wJ(K)J yJ(K)J
u u u u
= 2 (K = 1, 2)
1 + δu(K) + i Hu Q 1 + δ(K) + i H Q − yJ(K)J
u
X00 (J , Ju ; B) d.c. ≡ 1 . (10.113)
As an application of Eq. (10.112), we can evaluate the average of the scattering
phase matrix over an isotropic distribution of magnetic fields. Along the lines of
Sect. 5.12 we obtain (cf. Eq. (10.36))
Ω
Pij (Ω, ; B
) = (K)
MK (B) Rij (Ω, Ω ; 0) ,
K
(K)
where the matrix Rij (Ω, Ω ; 0) is defined in Eq. (10.21) and the quantity MK (B)
– the quantum analogue of the factor µK defined in Eq. (5.172) – is the average of
[XKQ (J , Ju ; B)]d.c. over all possible values of Q
1
MK (B) = XKQ (J , Ju ; B) d.c. .
2K + 1
Q
Let us rewrite the first of Eqs. (10.113) in the form
a + ibQ
XKQ (J , Ju ; B) d.c. = ,
c + i d Q + e Q2
where1
2
a = wJ(K)J 1 + δ(K) − (−1)K wJ(K)J wJ(K)J yJ(K)J
u u u u
(K) 2
b = wJ J H
u
2
c = 1 + δu(K) 1 + δ(K) − yJ(K)J
u
d = 1 + δu(K) H + 1 + δ(K) Hu
e = −H Hu .
After some algebra we obtain
M0 (B) = 1
1 a a(c + e) + bd
M1 (B) = +2
3 c (c + e)2 + d2
1 a a(c + e) + bd a(c + 4e) + 4bd
M2 (B) = +2 +2 . (10.114)
5 c (c + e)2 + d2 (c + 4e)2 + 4d2
1
Note that the quantities a, b, c, d depend on K.
NON-EQUILIBRIUM ATOMIC PHYSICS 579
We can consider some interesting limits of these expressions. For the transition
(J = 0, Ju = 1), neglecting depolarizing collisions, Eqs. (10.114) give the ‘classical’
result of Eqs. (5.173) with Hu in the place of H (see Eqs. (10.103) and Tables 10.1
and 10.4). For zero magnetic field, Eqs. (10.114) obviously reduce to Eqs. (10.110),
lim MK (B) = XK (J , Ju ) d.c. .
B→0
For H → ∞ (complete relaxation of lower-level coherences) we obtain, with the
use of Eq. (10.17)
!
1 W1 (J , Ju ) 1 + δu(1)
lim M1 (B) = X1 (J , Ju ) d.c. + 2 2
H →∞ 3 1 + δu(1) + Hu2
1 W2 (J , Ju ) 1 + δu(2)
lim M2 (B) = X2 (J , Ju ) d.c. + 2 2
H →∞ 5 1 + δu(2) + Hu2
!
W2 (J , Ju ) 1 + δu(2)
+2 2 .
1 + δu(2) + 4Hu2
For Hu → ∞ (complete relaxation of both upper and lower level coherences), we
have1
1 1
lim M1 (B) = X1 (J , Ju ) d.c. , lim M2 (B) = X2 (J , Ju ) d.c. .
Hu →∞ 3 Hu →∞ 5
10.14.d No inelastic collisions
Finally, we examine how atomic polarization is affected by stimulated emission.
(K)
Neglecting inelastic and superelastic collisions ( = (K)
u = = 0), Eq. (10.101)
reads
ρ00 (αu Ju ) 2Ju + 1 n̄
= ,
ρ00 (α J ) 2J + 1 1 + n̄
and Eqs. (10.102) can be written in the form
wJ(K)J 1 + δ(K) + i H Q − (−1)K wJ(K)J yJ(K)J (−1)Q sK
−Q (ν0 )
K u u u
σQ (αu Ju ) = (K) 2
1 + δu + i Hu Q 1 + δ + i H Q − yJ J
(K) (K) 1 + n̄
u
K
σQ (α J ) = (−1)1+K
wJ(K)J 1 + δu (K) + i Hu Q − (−1)K wJ(K)J yJ(K)J (−1)Q sK
−Q (ν0 )
× u
u
u
,
1 + δu (K) + i Hu Q 1 + δ(K) + i H Q − yJ(K)J
2 1 + n̄
u
1
We recall that Eq. (10.112) is valid for n̄ 1, which implies H Hu .
580 CHAPTER 10
where
δu(K) Hu
δu (K) = , Hu = .
1 + n̄ 1 + n̄
Comparison with Eqs. (10.111) shows that the effect of stimulated emission is
twofold. On the one hand, the efficiency of collisions and of the magnetic field in
depolarizing the upper level is reduced:1 this is a consequence of the fact that the
lifetime of the upper level is now shorter. On the other hand, atomic polarization
is globally decreased by a factor (1 + n̄). This is consistent with the results derived
in Sect. 10.9.
10.15. The Two-Term Atom: Generalities
In the previous sections of this chapter we have studied the interaction of an atomic
system with the radiation field by referring to the simplest model atom, that we
denoted as multi-level. In that model the atomic energy states are ‘uncorrelated’
J-levels, and atomic transitions take place between two such levels. Now we extend
this study to the so-called multi-term atom considered in Sects. 7.5-7.7, which
describes the atomic structure by taking into account (under the assumption of
L-S coupling) the spin-orbit interaction. In this model, transitions involve an
upper and a lower term, i.e., two ‘groups’ of J-levels, each characterized by the
same values of the quantum numbers L and S. Transitions are now associated with
a set of spectral lines (the components of the multiplet) rather than with a single
line. Atomic polarization involves coherences both between magnetic sublevels of
each J-level and between magnetic sublevels of the different J-levels of each term.
Coherences of the latter kind (which do not exist, by definition, in the multi-level
atom) are responsible for several remarkable phenomena, both in the presence and
in the absence of a magnetic field.
The following treatment of the multi-term atom is strictly analogous to our pre-
vious treatment of the multi-level atom. In particular, we restrict attention to
a simplified atomic model composed of just two terms. However, two important
points should be emphasized. The first one concerns collisions, which – contrary
to the previous case – will not be considered. This is because the theory of col-
lisions developed in Chap. 7 is only valid for the multi-level atom. Extension of
such theory to the multi-term atom is rather complicated and is outside the aims
of this book. The second point concerns the flat-spectrum approximation. We still
assume that the radiation impinging on the atomic system satisfies such approx-
imation – this is indeed required by the general equations (statistical equilibrium
and radiative transfer) established in Chap. 6: from this point of view, there is no
difference from the multi-level case. However, the range over which the radiation
field must be frequency-independent is much larger in the present case: neglecting
the magnetic field, it basically coincides with the entire frequency range spanned by
1 (K)
A similar effect is also present in the lower level, since the quantities δ and H are
inversely proportional to n̄ (cf. Eqs. (10.63) and (10.103)).
NON-EQUILIBRIUM ATOMIC PHYSICS 581
the multiplet, to be compared with the (natural plus collisional) frequency width
of a single line in the multi-level case. For instance, the relevant wavelength range
for the CaII doublet is about 100 Å (see Sect. 10.17). These two points (exclusion
of collisions and flat-spectrum approximation) obviously set severe constraints to
the applicability of the following results to real cases.
As in the multi-level case, we will often introduce certain approximations (con-
cerning either the atomic system or the radiation field) in order to simplify the
equations and to obtain analytical expressions for the atomic polarization and for
the radiation emitted in scattering processes. These approximations are briefly
recalled by subheadings at the beginning of each section.
Let us consider a two-term model atom, and let us denote by (β L S) and (βu Lu S)
the quantum numbers characterizing the lower and upper term, respectively (both
in L-S coupling).1 The atom is interacting with an anisotropic, polarized radiation
field specified by the Stokes vector Ii (ν, Ω). A magnetic field is present, with asso-
ciated Larmor frequency νL (see Eq. (3.10)) and direction Ω . The flat-spectrum
B
approximation requires that the radiation field is frequency-independent across an
interval ∆ν centered at νβ L S, β L S (the Bohr frequency corresponding to the
u u
transition between the two terms, see Eq. (7.22)) and encompassing the Bohr fre-
quencies of all the transitions between the different magnetic sublevels originating,
in the Paschen-Back effect regime, from the two terms.2 We fix a reference sys-
tem with the z-axis along the magnetic field direction, and we describe the atom
through the multipole moments of the density matrix. From Eq. (7.38) we get, for
the upper term
d βL SK
u u ρ (J ,
Q u u J ) = −2πi Nβ L S (KQJu Ju , K Q Ju Ju ) βu Lu SρK
Q (Ju , Ju )
dt
u u
K Q Ju Ju
+ TA (βu Lu SKQJu Ju , β L SK Q J J ) β L SρK
Q (J , J )
K Q J J
− RE (βu Lu SKQJu Ju K Q Ju Ju )
K Q Ju Ju
βu Lu S K
+ RS (βu Lu SKQJu Ju K Q Ju Ju ) ρQ (Ju , Ju ) , (10.115)
and for the lower term
d βL SK
ρ
Q (J , J ) = −2πi Nβ L S (KQJ J , K Q J J ) β L SρK
Q
(J , J )
dt
K Q J J
+ TE (β L SKQJ J , βu Lu SK Q Ju Ju ) +
K Q Ju Ju
1
The spectroscopic notation is the same as that employed in the whole book: L is the orbital
angular momentum quantum number of the electronic cloud and S is the spin, while β describes
the electronic configuration. The spin of the lower and upper term is the same, so that a dipole
transition is allowed between them.
2
Such frequencies, denoted by νβ L Sju Mu , β L Sj M , are defined in Eq. (7.26).
u u
582 CHAPTER 10
βu Lu S K
+ TS (β L SKQJ J , βu Lu SK Q Ju Ju ) ρQ (Ju , Ju )
− RA (β L SKQJ J K Q J J ) β L SρK
Q (J , J ) . (10.116)
K Q J J
The expressions of the rates are given in Sect. 7.6. We remind the reader that TA ,
TE , and TS are the radiative transfer rates due to absorption, spontaneous emission
and stimulated emission, respectively, while RA , RE , and RS are the corresponding
relaxation rates. N is the kernel containing the contributions of fine structure and
of the magnetic field.
Equations (10.115) and (10.116) form a system of linear, differential equations in
the unknowns β L SρKQ (J , J ) and ρQ (Ju , Ju ). In stationary situations the left-
βu Lu S K
hand sides are zero, and the equations reduce to an algebraic, linear, homogeneous
system. The number of equations (and of unknowns) is given by
Neq = (2S + 1)2 (2L + 1)2 + (2Lu + 1)2 , (10.117)
and the determinant of the system is zero.1 The normalization condition is provided
by the trace equation, which reads (see Eqs. (3.84), (3.101), (3.108))
β L S 0 βu Lu S 0
2J + 1 ρ0 (J , J ) + 2Ju + 1 ρ0 (Ju , Ju ) =1, (10.118)
J Ju
or, introducing the number density of atoms N ,
N + Nu = N ,
with
N = N 2J + 1 β L S 0
ρ0 (J , J )
J
Nu = N 2Ju + 1 βu Lu S 0
ρ0 (Ju , Ju ) . (10.119)
Ju
Once the statistical equilibrium equations are solved, the radiative transfer coeffi-
cients can be found from Eqs. (7.47).
As obvious from Eq. (10.117), the system contains a large number of equations
even for the simplest transitions. Analytical solutions are therefore impracticable
unless some approximation is introduced. One of the most natural and fruitful
1
Using the expressions of the rates and of the kernel N , it can be shown directly that the
equations for the time derivatives of the linear combinations
% β L S 0
% βu Lu S 0
J
2J + 1 ρ0 (J ,J ) and Ju
2Ju + 1 ρ0 (Ju , Ju )
(the populations of the lower and upper term) are identical, except for the opposite sign.
NON-EQUILIBRIUM ATOMIC PHYSICS 583
approximations, that will often be used in the following sections, is to neglect the
atomic polarization of the lower term. Although collisions are not included in
Eqs. (10.115)-(10.116), it is reasonable to believe, by analogy with the two-level
case (cf. Sect. 10.7), that such unpolarized lower term approximation is roughly
satisfied in the presence of depolarizing collisions and a weak radiation field (which
also implies that stimulation effects should be neglected).
Let us examine how the statistical equilibrium equations are modified by this
approximation. The condition that all the sublevels of the lower term are evenly
populated and that no coherence is present between them can be written in the
form
ρβ L S (j M , j M ) = δj j δM M C , (10.120)
where ρβ L S (j M , j M ) is the density-matrix element in the energy-eigenvector
representation (see Sect. 7.5) and C is a constant (independent of j and M ).
The statistical tensors of the lower term can be easily deduced by substituting
Eq. (10.120) into Eq. (7.37) and by using Eqs. (3.62b), (2.26a) and (2.23a). One
gets
ρQ (J , J ) = δK0 δQ0 δJ J 2J + 1 C .
β L S K
(10.121)
The constant C is proportional to the overall population of the lower term. Sub-
stitution of Eq. (10.121) into the first of Eqs. (10.119) yields
1 N
C= . (10.122)
(2S + 1)(2L + 1) N
The equations for the statistical tensors of the upper term can now be derived
from Eq. (10.115) by neglecting the relaxation rate due to stimulated emission and
by making use of Eqs. (10.121) and (10.122). Recalling Eq. (7.46b) we obtain, in
stationary situations
A(βu Lu S → β L S) βu Lu SρK
Q (Ju , Ju ) +
βu Lu S K
+ 2πi Nβ L S (KQJu Ju , K Q Ju Ju ) ρQ (Ju , Ju ) =
u u
K Q Ju Ju
2J + 1 N
= TA (βu Lu SKQJu Ju , β L S 0 0J J ) , (10.123)
(2S + 1)(2L + 1) N
J
where the kernel N is given by Eq. (7.41) and the transfer rate TA by Eq. (7.45a).
Taking into account Eqs. (2.26a) and (2.49), the latter can be rewritten in the form
TA (βu Lu SKQJu Ju , β L S 0 0J J ) = (2L + 1) B(β L S → βu Lu S)
! ! !
1 1 K Lu L 1 Lu L 1
× 3(2J + 1)(2Ju + 1)(2Ju + 1)
Ju Ju J J Ju S J Ju S
× (−1)1+J +Ju +Q J−Q
K
(νβ L S, β L S ) . (10.124)
u u
584 CHAPTER 10
In the following sections we will analyze the consequences of Eq. (10.123) both in
the non-magnetic and in the magnetic regime.
10.16. The Two-Term Atom: Resonance Polarization
(unpolarized lower term - no magnetic field)
To describe resonance polarization, we neglect in Eq. (10.123) the contribution of
the magnetic field. Setting νL = 0 in Eq. (7.41), the kernel N takes the simpler
form
Nβ L S (KQJu Ju , K Q Ju Ju ) = δKK δQQ δJ J δJ J
u u u u u u
× νβ L SJu , βu Lu SJu , (10.125)
u u
where the Bohr frequency νβ L SJ , β L SJ is given by Eq. (7.43). Equation (10.123)
u u u u u u
can now be solved for the statistical tensors of the upper term. Substitution of
Eqs. (10.124) and (10.125) yields
βu Lu S K N 3(2Ju + 1)(2Ju + 1)
ρQ (Ju , Ju ) =
N 2S + 1
B(β L S → βu Lu S)
×
A(βu Lu S → β L S) + 2πi νβ L SJ , βu Lu SJu
u u u
× (−1)1+J +Ju +Q (2J + 1)
J
! ! !
1 1 K Lu L 1 Lu L 1
× K
J−Q (νβ L S, β L S ) ,
Ju Ju J J Ju S J Ju S u u
or, performing the summation over J via Eq. (2.41)
βu Lu S K N 3(2Ju + 1)(2Ju + 1)
ρQ (Ju , Ju ) =
N 2S + 1
B(β L S → βu Lu S)
× (−1)1−L +S+Ju +K+Q
A(βu Lu S → β L S) + 2πi νβ L SJ , β L SJ
u u u u u u
! !
1 1 K Lu Lu K
× K
J−Q (νβ L S, β L S ) . (10.126)
Lu Lu L Ju Ju S u u
To obtain the radiation emitted by the atom along a given direction Ω, we have to
substitute Eq. (10.126) into the expression of the emission coefficient for the multi-
term atom. This is given by Eq. (7.47e) for the general case where a magnetic field
is present. In the non-magnetic case it becomes much simpler, since the following
substitutions are to be performed
CJj (βLS, M ) → δjJ , Φ(νβ L Sju Mu , β L Sj M − ν) → Φ(νβ L SJu , β L SJ − ν) ,
u u u u
NON-EQUILIBRIUM ATOMIC PHYSICS 585
where
Eβ L S (Ju ) − Eβ L S (J )
u u
νβL SJu , β L SJ = ,
u u h
with EβLS (J) the eigenvalue of the spin-orbit Hamiltonian corresponding to the
eigenvector |βLSJM (see Eq. (3.61a)). With the help of Eqs. (2.34), (3.100) and
(5.158) we obtain1
2 4
= h ν N (2L + 1) B(β L S → β L S)
εi (ν, Ω) u u u
2πc2
× (−1)1+J +Ju (2J + 1) 3(2Ju + 1)(2Ju + 1)
KQ Ju Ju J
! ! !
Lu L 1 Lu L 1 1 1 K
×
J Ju S J Ju S Ju Ju J
× TQ
K
(i, Ω) ρQ (Ju , Ju )
βu Lu S K
1
× Φ(νβ L SJ , β L SJ − ν) + Φ(νβ L SJ , β L SJ − ν)∗ . (10.127)
2 u u u u u u
Now we substitute in this expression the upper-term statistical tensors given by
Eq. (10.126). Bearing in mind the relations between the Einstein coefficients
(Eqs. (7.33)), and introducing the frequency-integrated absorption coefficient in
the multiplet2
h νβ L S, β L S
A
kM = u u
N B(β L S → βu Lu S) , (10.128)
4π
we obtain
= k A 2Lu + 1
εi (ν, Ω)
(−1)S−L +Ju +Ju +J +K+Q
M
2S + 1
KQ Ju Ju J
! !
Lu L 1 Lu L 1
× 3(2Ju + 1)(2Ju + 1)(2J + 1)
J Ju S J Ju S
! ! !
1 1 K 1 1 K Lu Lu K J K (ν
× TQ
K
(i, Ω) −Q βu Lu S, β L S )
Ju Ju J Lu Lu L Ju Ju S
∗
1 Φ(νβu Lu SJu , β L SJ − ν) + Φ(νβu Lu SJu , β L SJ − ν)
× . (10.129)
2 1 + 2πi νβ L SJ , β L SJ A(βu Lu S → β L S)
u u u u u u
1
Similar calculations have been performed to derive Eq. (7.48g) from Eq. (7.47e).
2 A A
The quantity kM – the analogue of kL of Eq. (9.5) for a multiplet – is defined as the integral
over frequency of the absorption coefficient η0A (ν, Ω)
for a two-term atom with unpolarized lower
term. The integral is easily evaluated using Eq. (7.47a) – or equivalently Eq. (7.48b) – and taking
into account Eqs. (10.121) and (10.122). The expression in Eq. (10.128) implies the approximation
ν ≈ νβ L S, β L S .
u u
586 CHAPTER 10
contained in the last term of Eq. (10.129),
The frequency dependence of εi (ν, Ω),
will be analyzed in the next section. Here we restrict attention to the frequency-
defined by
integrated emission coefficient, ε̃i (Ω),
= ε (ν, Ω)
ε̃i (Ω) dν , (10.130)
i
∆ν
where the interval ∆ν is sufficiently broad to fully cover all the lines of the multiplet.
Since the Φ profiles are normalized to unity in frequency (Eqs. (6.59a-c)), we can
perform the summation over J via Eq. (2.41). Writing the radiation field tensor in
terms of the Stokes parameters via Eqs. (5.157), we obtain the expression (see the
analogous Eq. (10.19) valid for the two-level atom with unpolarized lower level)
5
dΩ
3
= kA
ε̃i (Ω) Ω
Pij (Ω, ) I (ν ) ,
,Ω (10.131)
M fs j βu Lu S, β L S
4π j=0
Ω
where [Pij (Ω, )] , the fine-structure scattering phase matrix, is given by
fs
Ω
Pij (Ω, ) = WK (β L Sβu Lu ) fs (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω (10.132)
fs −Q
KQ
with
3(2Lu + 1)
WK (β L Sβu Lu ) fs =
2S + 1
!2 !2
1 1 K Lu Lu K
× (2Ju + 1)(2Ju + 1)
Lu Lu L Ju Ju S
Ju Ju
1
× . (10.133)
1 + 2πi νβ L SJu , βu Lu SJu A(βu Lu S → β L S)
u u
Comparison of Eqs. (10.132) and (10.20) shows that the new coefficient is the direct
equivalent, for fine-structure multiplets, of the coefficient WK (J , Ju ) introduced for
the two-level atom. It can easily be proved that, like WK (J , Ju ), it is real
∗
WK (β L Sβu Lu ) fs = WK (β L Sβu Lu ) fs , (10.134)
and satisfies (see Eq. (2.36a))
W0 (β L Sβu Lu ) fs = 1 .
Equation (10.133) can be cast in a more significant form. Recalling Eq. (10.17)
one has
WK (β L Sβu Lu ) fs = WK (L , Lu ) DK (βu Lu S) fs , (10.135)
NON-EQUILIBRIUM ATOMIC PHYSICS 587
where [DK (βLS)]fs , the depolarizing factor due to fine structure,1 is given by
!2
1 L L K
DK (βLS) fs = (2J + 1)(2J + 1)
2S + 1
J J S
JJ
1
× . (10.136)
1 + 2πi νβLSJ , βLSJ A(βLS → β L S)
The depolarizing factor satisfies some important properties:
a) When the fine-structure intervals of the term are negligible in comparison with
the de-excitation Einstein coefficient, the last factor in Eq. (10.136) reduces to
unity and one obtains, with the use of Eq. (2.39)
DK (βLS) fs = 1 . (10.137)
In this case Eq. (10.135) becomes
WK (β L Sβu Lu ) fs = WK (L , Lu ) , (10.138)
which shows that the two-term atom behaves as a two-level atom having J = L ,
Ju = Lu . This is an obvious manifestation of the principle of spectroscopic stability:
if the spin induces negligible splittings, it can just be disregarded.
b) In the opposite case where the fine-structure intervals are much larger than the
Einstein coefficient, all the terms with J = J make a negligible contribution to
the double summation in Eq. (10.136), so that
!2
∞ 1 L L K
DK (βLS) fs = DK (LS) fs = (2J + 1)2 . (10.139)
2S + 1 J J S
J
∞
Values of DK are collected in Table 10.6 for K = 1, 2 2 and for all terms having
S ≤ 9/2, L ≤ 7. Terms with S = 0 or L = 0 are not included since for such terms
∞
the definition of DK is meaningless.
c) When the J-levels satisfy the Landé interval rule (see Eq. (3.60)), one has
1
νβLSJ , βLSJ = ζ(βLS) J (J + 1) − J(J + 1) , (10.140)
2h
and the depolarizing factor can be written in the form
!2
1 L L K
DK (βLS) fs = (2J + 1)(2J + 1)
2S + 1
J J S
JJ
1
× 2 , (10.141)
1+ x2
J (J + 1) − J(J + 1)
1
It is important to note that the depolarizing factor depends only on the fine-structure
properties and on the Einstein coefficient for spontaneous de-excitation of the term.
2
Use of Eq. (2.36a) shows that [D0∞ (LS)]fs = [D0 (βLS)]fs = 1.
588 CHAPTER 10
TABLE 10.6
∞ (LS)] for different terms
Values of the depolarizing factors [DK fs
Term D1∞ D2∞ Term D1∞ D2∞ Term D1∞ D2∞
2P 0.778 0.333 5P 0.389 0.231 8P 0.354 0.212
2D 0.92 0.76 5D 0.5 0.265 8D 0.391 0.227
2F 0.959 0.878 5F 0.705 0.349 8F 0.453 0.248
2G 0.975 0.926 5G 0.813 0.529 8G 0.573 0.255
2H 0.983 0.950 5H 0.873 0.658 8H 0.691 0.342
2I 0.988 0.964 5I 0.908 0.744 8I 0.770 0.456
2K 0.991 0.973 5K 0.930 0.803 8K 0.824 0.555
3P 0.5 0.278 6P 0.371 0.222 9P 0.35 0.210
3D 0.796 0.473 6D 0.441 0.247 9D 0.378 0.221
3F 0.894 0.702 6F 0.597 0.261 9F 0.425 0.238
3G 0.935 0.813 6G 0.737 0.399 9G 0.5 0.257
3H 0.956 0.872 6H 0.818 0.541 9H 0.623 0.280
3I 0.969 0.908 6I 0.867 0.648 9I 0.716 0.374
3K 0.976 0.930 6K 0.899 0.724 9K 0.780 0.474
4P 0.422 0.247 7P 0.361 0.216 10 P 0.347 0.208
4D 0.646 0.276 7D 0.41 0.235 10 D 0.369 0.217
4F 0.807 0.510 7F 0.5 0.260 10 F 0.406 0.231
4G 0.880 0.674 7G 0.655 0.302 10 G 0.461 0.248
4H 0.919 0.772 7H 0.757 0.432 10 H 0.558 0.253
4I 0.942 0.833 7I 0.821 0.549 10 I 0.659 0.310
4K 0.956 0.872 7K 0.864 0.640 10 K 0.734 0.400
where
π ζ(βLS)
x= . (10.142)
h A(βLS → β L S)
The dimensionless parameter x represents the ratio between a typical fine-structure
splitting of the J-levels of the term and the natural width of the same levels.
Obviously, Eq. (10.141) reduces to Eq. (10.137) for x = 0, and to Eq. (10.139) for
x → ∞. As an illustration of Eq. (10.141) we plot in Fig. 10.16, as functions of x,
the depolarizing factors [D1 ]fs and [D2 ]fs for the term 3 P.
Consider now the polarization of the radiation emitted in a scattering process.
Because of the close similarity of Eqs. (10.131) and (10.19), the phase matrix cor-
responding to the geometry of Fig. 10.1 is given by Eq. (10.24) with the factor
[WK (β L Sβu Lu )]fs in the place of WK (J , Ju ). For the 90◦ scattering of an unpo-
larized radiation beam, the fractional frequency-integrated polarization is therefore
given by (cf. Eqs. (10.25))
3 W2 (L , Lu ) D2 (βu Lu S) fs
p̃Q =
4 − W2 (L , Lu ) D2 (βu Lu S) fs
p̃U = 0 . (10.143)
NON-EQUILIBRIUM ATOMIC PHYSICS 589
Fig.10.16. The depolarizing factors [D1 (βLS)]fs and [D2 (βLS)]fs for the term 3 P against the
parameter x defined in Eq.(10.142). The asymptotic values for x → ∞ (0.5 for D1 and 0.278 for
D2 ) are also shown.
If we consider, for instance, the transition 3 S − 3 P (L = 0, Lu = 1, S = 1) and
suppose the upper term to be described by the Landé interval rule, the polarization
varies from p̃Q = 1 for x = 0 to p̃Q = 0.224 for x → ∞.1 This example clearly
shows the depolarizing effect of fine structure in scattering processes. Such effect
is the larger, the smaller the value of the Einstein A coefficient.
10.17. The Two-Term Atom: Spectral Details of
Resonance Polarization
(unpolarized lower term - no magnetic field)
The frequency dependence of resonance polarization in a two-term atom with un-
polarized lower term is fully described by Eq. (10.129).2 That expression is very
complicated and we will try to illustrate its consequences by studying a number of
special cases. In the previous section we have already pointed out the depolarizing
effect of fine structure on the frequency-integrated emission coefficient.
is contained in the last factor of the right-
The frequency dependence of εi (ν, Ω)
hand side of Eq. (10.129). If the width of the J-levels of the lower term is much
smaller than the width of the J-levels of the upper term, this factor can be cast
into a simpler form (cf. the derivation of Eq. (10.42)). From Eqs. (6.59a,b) we
1
We recall that a positive value for p̃Q means that the polarization direction is perpendicular
to the scattering plane.
2
Note, however, that Eq. (10.129) is valid under the flat-spectrum approximation.
590 CHAPTER 10
have, neglecting frequency shifts
1 1
Φ(νβ L SJu , β L SJ − ν) = ,
u u π Γ − i νβ L SJu , β L SJ −ν
u u
where (see footnote 2 on p. 520 and footnote 2 on p. 314)
1 A(βu Lu S → β L S)
Γ = A(βu Lu SJu → β L SJ ) = .
4π 4π
J
Recalling Eq. (7.43), we easily obtain
∗
1 Φ(νβu Lu SJu , β L SJ − ν) + Φ(νβu Lu SJu , β L SJ − ν)
=
2 1 + 2πi νβ L SJ , β L SJ A(βu Lu S → β L S)
u u u u u u
1 Γ
= . (10.144)
π Γ − i (νβ L SJu , β L SJ − ν) Γ + i (νβ L SJu , β L SJ − ν)
u u u u
In most fine-structure multiplets the different lines are well-separated in fre-
quency, lying at distances that are much larger than their natural width.1 To
make an example, consider the H and K lines of the CaII doublet at 3950 Å. The
frequency interval between the two lines is about 6.7 × 1012 s−1 , while the Einstein
A coefficient is about 1.5 × 108 s−1 , i.e., 4 × 104 times smaller. For such multiplets,
the emission coefficient in the neighborhood of a single line – say the line corre-
sponding to Ju = J¯u and J = J¯ – can therefore be evaluated by restricting the
summation over Ju , Ju and J in Eq. (10.129) to the values
Ju = Ju = J¯u , J = J¯ .
Hence we can write, in the vicinity of νβ L SJu , β L SJ
u u
!2
≈ kA 2Lu + 1 Lu L 1
εi (ν, Ω) M 3(2Ju + 1)2 (2J + 1)
2S + 1 J Ju S
! ! !
S−L −J +K+Q 1 1 K 1 1 K Lu Lu K
× (−1)
Ju Ju J Lu Lu L Ju Ju S
KQ
× TQ
K J K (ν
(i, Ω) − ν) , (10.145)
−Q β L S, β L S )
u u
φ(νβ
u u
L SJu , β L SJ
where
1 Γ
φ(νβ − ν) = .
L SJu , β L SJ
u u π Γ 2 + (νβ L SJu , β L SJ − ν)2
u u
1
This is no longer true for hyperfine-structure multiplets. See Sect. 10.22.
NON-EQUILIBRIUM ATOMIC PHYSICS 591
If we introduce the relative strengths of the lines in the multiplet defined in
Eq. (3.65) – see also footnote 1 on p. 314 – we can rewrite Eq. (10.145) in the
form
≈ k A S J ,Ju φ(ν
εi (ν, Ω) M βu Lu SJu , β L SJ − ν)
× WK (L Lu S, J Ju ) (−1)Q T Q K J K (ν
(i, Ω) −Q β L S, β L S )
u u
, (10.146)
KQ
where the symbol WK (L Lu S, J Ju ) is given by
WK (L Lu S, J Ju ) = (−1)S−L −J +K 3(2Lu + 1)(2Ju + 1)
! ! !
1 1 K 1 1 K Lu Lu K
× ,
Ju Ju J Lu Lu L Ju Ju S
or, using Eq. (2.36a)
! ! !
1 1 K 1 1 K Lu Lu K
J J J L L L J Ju S
WK (L Lu S, J Ju ) = (−1)K u u ! u u
! u ! . (10.147)
1 1 0 1 1 0 Lu Lu 0
Ju Ju J Lu Lu L Ju Ju S
Equation (10.146), when compared with Eq. (10.16), shows that a line belonging
to a multiplet behaves in resonance scattering as an isolated line, with the only
difference that the symbol WK (J , Ju ) defined in Eq. (10.17) is replaced by the more
general symbol WK (L Lu S, J Ju ) defined above. This new symbol satisfies several
important properties:
a) When S = 0 the multiplet reduces to a single line. This implies Lu = Ju and
L = J . Using Eq. (2.36a) one obtains the obvious result
WK (J Ju 0, J Ju ) = WK (J , Ju ) .
b) For K = 0, we have from Eq. (10.147)
W0 (L Lu S, J Ju ) = 1 .
c) Apart from the trivial case S = 0, the only multiplet for which the symbols
WK (L Lu S, J Ju ) and WK (J , Ju ) coincide is the simplest of the multiplets, namely
2
S − 2 P (L = 0, Lu = 1, S = 1/2). Using Eq. (2.36a) we have in fact
WK 0 1 12 , 12 Ju = WK 12 , Ju .
d) Owing to the properties of the 6-j symbols, all lines originating from an S upper
term (Lu = 0) have W1 and W2 symbols equal to zero. The same holds for all lines
having Ju = 0, whereas all lines with Ju = 1/2 have W2 = 0, but W1 = 0.
592 CHAPTER 10
e) For several lines the value of W2 is negative, which means that the direction
of the scattered linear polarization produced by an unpolarized radiation beam is
parallel to the scattering plane (see comments about Eq. (10.143)).
Table 10.7 collects the numerical values of WK (L Lu S, J Ju ) for all multiplets (ex-
cept singlets) satisfying the electric-dipole selection rules (cf. Sect. 7.11), up to
S = 5/2, L = 2.
For frequencies that are not in the neighborhood of single spectral lines, the
approximation of restricting the summation over Ju , Ju and J in Eq. (10.129) to
specific values is not justified. All the terms contribute, and the influence of the
‘off-diagonal’ terms (those having Ju = Ju ) is fundamental.
Let us consider a frequency ν which is well outside the frequency range of the
multiplet. The last term in Eq. (10.129) can approximately be written in the form
(see Eq. (10.144))
1 Γ
φ(ν0 − ν) ≈ ,
π (ν0 − ν)2
where ν0 is a sort of center of gravity of the whole multiplet. As this term does not
depend on the quantum numbers Ju , Ju , J , the summation over these numbers
can be performed via Eqs. (2.41) and (2.39). One gets
≈ k A φ(ν − ν)
εi (ν, Ω) M 0
× WK (L , Lu ) (−1)Q T Q
K J K (ν
(i, Ω) −Q β L S, β L S ) ,
u u
(10.148)
KQ
where WK (L , Lu ) is the symbol introduced for the two-level atom (Eq. (10.17)).
This is an interesting result: at large distance from the ‘center of gravity’, the mul-
tiplet behaves in resonance scattering as a simple transition between two spinless
levels L and Lu .
To find the frequency dependence of the scattered radiation outside the limited
ranges considered above (neighborhood of single lines, ‘far wings’ of the multiplet)
one should resort to numerical calculations, by specifying the full set of physical
parameters entering Eq. (10.129). These include – in addition to the scattering
geometry – the quantum numbers of the lower and upper term, the energies of the
different J-levels, and the value of the Einstein A coefficient.
Two examples are shown in Figs. 10.17 and 10.18, where the fractional linear po-
larization pQ = εQ /εI resulting from the 90◦ scattering of an unpolarized radiation
beam is plotted as a function of wavelength for two different multiplets, multiplet
n. 1 of CaII (2 S − 2 P) and multiplet UV n. 1 of FeII (6 D − 6 D). The curves have
been computed from Eq. (10.129) by neglecting the width of the lower levels, so
that the frequency-dependent term is given by Eq. (10.144). Both figures refer to
the scattering geometry of Fig. 10.1 (with Θ = 90◦ ),1 therefore a positive value for
1
We recall that, for the scattering conditions√specified above, the only non-zero components
of the radiation field tensor are J00 and J02 = J00 / 2 (cf. Eqs. (5.164)), and that the components
K are obtained from Table 5.6 setting θ = γ = 90◦ .
of the tensor T Q
NON-EQUILIBRIUM ATOMIC PHYSICS 593
TABLE 10.7
Values of the quantities WK (L Lu S, J Ju ) for different multiplets
Term uTerm J Ju W1 W2 Term uTerm J Ju W1 W2
2S 2P 1/2 1/2 0.667 0. 3P 3D 0 1 0.75 0.35
1/2 3/2 0.833 0.5 1 1 0.375 −0.175
2P 2S 1/2 1/2 0. 0. 1 2 0.625 0.175
3/2 1/2 0. 0. 2 1 −0.375 0.035
2P 2P 1/2 1/2 0.333 0. 2 2 0.208 −0.175
1/2 3/2 0.417 −0.25 2 3 0.667 0.24
3/2 1/2 −0.167 0. 3D 3P 1 0 0. 0.
3/2 3/2 0.167 0.2 1 1 −0.125 0.025
2P 2D 1/2 3/2 0.75 0.35 1 2 −0.375 0.035
3/2 3/2 0.3 −0.28 2 1 0.125 −0.005
3/2 5/2 0.7 0.28 2 2 −0.125 −0.035
2D 2P 3/2 1/2 0.167 0. 3 2 0.25 0.01
3/2 3/2 −0.167 −0.04 3D 3D 1 1 0.125 0.175
5/2 3/2 0.25 0.01 1 2 0.208 −0.175
2D 2D 3/2 3/2 0.1 0.28 2 1 −0.125 −0.035
3/2 5/2 0.233 −0.28 2 2 0.069 0.175
5/2 3/2 −0.15 −0.07 2 3 0.222 −0.24
5/2 5/2 0.067 0.32 3 2 −0.139 −0.05
3S 3P 1 0 0. 0. 3 3 0.056 0.3
1 1 0.25 0.25 4S 4P 3/2 1/2 0.167 0.
1 2 0.75 0.35 3/2 3/2 0.133 0.32
3P 3S 0 1 0. 0. 3/2 5/2 0.7 0.28
1 1 0. 0. 4P 4S 1/2 3/2 0. 0.
2 1 0. 0. 3/2 3/2 0. 0.
3P 3P 0 1 0.25 0.25 5/2 3/2 0. 0.
1 0 0. 0. 4P 4P 1/2 1/2 −0.167 0.
1 1 0.125 −0.125 1/2 3/2 0.167 0.2
1 2 0.375 −0.175 3/2 1/2 0.083 0.
2 1 −0.125 0.025 3/2 3/2 0.067 −0.16
2 2 0.125 0.175 3/2 5/2 0.35 −0.14
5/2 3/2 −0.1 0.04
5/2 5/2 0.1 0.16
594 CHAPTER 10
TABLE 10.7
(continued)
Term uTerm J Ju W1 W2 Term uTerm J Ju W1 W2
4P 4D 1/2 1/2 0.5 0. 5P 5S 1 2 0. 0.
1/2 3/2 0.5 0. 2 2 0. 0.
3/2 1/2 −0.25 0. 3 2 0. 0.
3/2 3/2 0.2 0. 5P 5P 1 1 −0.125 0.025
3/2 5/2 0.55 0.1 1 2 0.125 0.175
5/2 3/2 −0.3 0. 2 1 0.125 −0.005
5/2 5/2 0.157 −0.114 2 2 0.042 −0.175
5/2 7/2 0.643 0.214 2 3 0.333 −0.12
4D 4P 1/2 1/2 0.167 0. 3 2 −0.083 0.05
1/2 3/2 −0.167 −0.04 3 3 0.083 0.15
3/2 1/2 −0.083 0. 5P 5D 1 0 0. 0.
3/2 3/2 −0.067 0.032 1 1 0.125 0.175
3/2 5/2 −0.35 0.028 1 2 0.375 −0.075
5/2 3/2 0.1 −0.008 2 1 −0.125 −0.035
5/2 5/2 −0.1 −0.032 2 2 0.125 0.075
7/2 5/2 0.25 0.01 2 3 0.5 0.06
4D 4D 1/2 1/2 0.167 0. 3 2 −0.25 −0.021
1/2 3/2 0.167 0. 3 3 0.125 −0.075
3/2 1/2 −0.083 0. 3 4 0.625 0.196
3/2 3/2 0.067 0. 5D 5P 0 1 0.25 0.01
3/2 5/2 0.183 −0.1 1 1 0.125 −0.005
5/2 3/2 −0.1 0. 1 2 −0.125 −0.035
5/2 5/2 0.052 0.114 2 1 −0.125 0.001
5/2 7/2 0.214 −0.214 2 2 −0.042 0.035
7/2 5/2 −0.131 −0.036 2 3 −0.333 0.024
7/2 7/2 0.048 0.286 3 2 0.083 −0.01
5S 5P 2 1 0.25 0.01 3 3 −0.083 −0.03
2 2 0.083 0.35 4 3 0.25 0.01
2 3 0.667 0.24
NON-EQUILIBRIUM ATOMIC PHYSICS 595
TABLE 10.7
(continued)
Term uTerm J Ju W1 W2 Term uTerm J Ju W1 W2
5D 5D 0 1 0.083 0.35 6P 6D 3/2 1/2 0.167 0.
1 0 0. 0. 3/2 3/2 0.033 0.2
1 1 0.042 −0.175 3/2 5/2 0.3 −0.1
1 2 0.125 0.075 5/2 3/2 −0.05 −0.05
2 1 −0.042 0.035 5/2 5/2 0.086 0.114
2 2 0.042 −0.075 5/2 7/2 0.464 0.036
2 3 0.167 −0.06 7/2 5/2 −0.214 −0.036
3 2 −0.083 0.021 7/2 7/2 0.103 −0.048
3 3 0.042 0.075 7/2 9/2 0.611 0.183
3 4 0.208 −0.196 6D 6P 1/2 3/2 0.25 0.01
4 3 −0.125 −0.025 3/2 3/2 0.1 −0.008
4 4 0.042 0.275 3/2 5/2 −0.1 −0.032
6S 6P 5/2 3/2 0.3 0.02 5/2 3/2 −0.15 0.002
5/2 5/2 0.057 0.366 5/2 5/2 −0.029 0.037
5/2 7/2 0.643 0.214 5/2 7/2 −0.321 0.021
6P 6S 3/2 5/2 0. 0. 7/2 5/2 0.071 −0.011
5/2 5/2 0. 0. 7/2 7/2 −0.071 −0.029
7/2 5/2 0. 0. 9/2 7/2 0.25 0.01
6P 6P 3/2 3/2 −0.1 0.04 6D 6D 1/2 1/2 −0.111 0.
3/2 5/2 0.1 0.16 1/2 3/2 0.028 0.25
5/2 3/2 0.15 −0.01 3/2 1/2 0.056 0.
5/2 5/2 0.029 −0.183 3/2 3/2 0.011 −0.2
5/2 7/2 0.321 −0.107 3/2 5/2 0.1 0.1
7/2 5/2 −0.071 0.057 5/2 3/2 −0.017 0.05
7/2 7/2 0.071 0.143 5/2 5/2 0.029 −0.114
5/2 7/2 0.155 −0.036
7/2 5/2 −0.071 0.036
7/2 7/2 0.034 0.048
7/2 9/2 0.204 −0.183
9/2 7/2 −0.120 −0.017
9/2 9/2 0.037 0.267
596 CHAPTER 10
Fig.10.17. Fractional polarization pQ against vacuum wavelength λv (in Å) in the 90◦ scattering
of an unpolarized radiation beam, for multiplet n.1 of CaII. The vertical lines mark the wavelength
positions of the two lines of the multiplet according to the code: 1) J = 1/2, Ju = 3/2 (K line);
2) J = 1/2, Ju = 1/2 (H line). Atomic data are from Moore (1949). The value of the Einstein
coefficient is A = 1.48 × 108 s−1 .
pQ means that the polarization direction is perpendicular to the scattering plane:
note that pQ is negative (polarization direction parallel to the scattering plane)
within certain frequency intervals.
The value of pQ at the wavelengths of the individual lines, as well as its asymptotic
value in the far wings, are easily understood in the light of our preceding results.
Consider for instance the line marked with number 3 in Fig. 10.18 (λ = 2600.2 Å,
J = Ju = 9/2). From Eq. (10.146) we obtain, with the use of Table 10.7
3 W2 2 2 52 , 92 92
pQ = 0.214 .
4 − W2 2 2 52 , 92 29
On the other hand, the asymptotic value for the multiplet of Fig. 10.18 can be
derived from Eq. (10.148) with the use of Table 10.1,
3 W2 (2, 2)
pQ = 0.288 .
4 − W2 (2, 2)
It should be realized that the εI profile, as given by Eq. (10.129), basically consists
of a set of extremely narrow (∆λ < 0.1 mÅ) Lorentzian profiles centered at the
wavelengths of the individual lines (J , Ju ). The effect of interferences on the ratio
NON-EQUILIBRIUM ATOMIC PHYSICS 597
Fig.10.18. Same as Fig.10.17 for multiplet UV n.1 of FeII. The J , Ju values of the lines are the
following: 1) 9/2, 7/2; 2) 7/2, 5/2; 3) 9/2, 9/2; 4) 5/2, 3/2; 5) 7/2, 7/2; 6) 3/2, 1/2; 7) 5/2, 5/2;
8) 3/2, 3/2; 9) 1/2, 1/2; 10) 7/2, 9/2; 11) 1/2, 3/2; 12) 3/2, 5/2; 13) 5/2, 7/2. Atomic data are
from Moore (1952). The value of the Einstein coefficient is A = 3.1 × 108 s−1 .
pQ shows up just across those wavelength intervals where the emissivity is extremely
weak . This is the reason why it is difficult to detect such interference phenomena
in laboratory experiments. The situation may drastically change in astrophysical
plasmas, where optical thickness effects can compensate for the low emissivity.
These phenomena have indeed been observed, for the first time, in the radiation
scattered by the CaII doublet at the extreme limb of the sun (Stenflo, 1980).
The ‘fragile’ nature of interference phenomena can also be realized by assuming
that the plasma, in addition to the resonantly scattered radiation in a given mul-
tiplet, also emits a certain amount of continuum radiation (which, for the sake of
simplicity, we suppose to be unpolarized). Denoting by ε(c) the continuum emis-
sivity, the total emission coefficient is given by
= ε(M) (ν, Ω)
εi (ν, Ω) + ε(c) δ , (10.149)
i i0
where ε(M)
i (ν, Ω), the emissivity in the multiplet, is still given by Eq. (10.129).
Figure 10.19 illustrates, in a specific case, the strong reduction of interference
effects due to the presence of some continuum emissivity.
598 CHAPTER 10
Fig.10.19. Solid line: same as Fig.10.18 in the presence of a given amount of continuum, unpo-
larized radiation (see Eq.(10.149)). The value of ε(c) is 3 × 10−9 times the maximum emissivity
in the multiplet (at the center of line 3 of Fig.10.18). The dotted curve (identical with the curve
of Fig.10.18) corresponds to ε(c) = 0.
10.18. The Two-Term Atom: the Hanle Effect
(unpolarized lower term)
In this section we study the effect of a magnetic field of arbitrary strength1 on
the atomic polarization of a two-term atom, by keeping the assumption that the
lower term is unpolarized. Obviously, the flat-spectrum approximation requires the
incident radiation to be frequency-independent across a range which encompasses
all the multiplet components connecting the magnetic sublevels in the Paschen-
Back effect regime.
The set of equations for the statistical tensors of the upper term, βu Lu SρK
Q (Ju , Ju ),
has already been deduced (Eq. (10.123)). However, while adequate to derive a
numerical solution, this set is not suitable to obtain an analytical solution, because
of the presence of the kernel N which mixes up different statistical tensors. Such
difficulty is intrinsic to the two-term atom: in the two-level atom the magnetic
term is ‘diagonal’, which immediately yields (under the unpolarized lower level
assumption) the solution to the statistical equilibrium equations (Eq. (10.8)). By
contrast, in the present case one is faced with a large algebraic system even for the
simplest terms (see Eq. (10.117)).
1
See, however, the discussion following Eq. (3.2).
NON-EQUILIBRIUM ATOMIC PHYSICS 599
A closed expression for the statistical tensors of the upper term can be found by
solving (in stationary situations) the statistical equilibrium equations in the energy-
eigenvector representation, so to obtain the density-matrix elements of the upper
term; these can then be combined, by the appropriate transformation, in order to
get the statistical tensors. This procedure does not avoid the need for numerical
calculations: the problem of solving a linear algebraic system is changed into the
problem of diagonalizing the spin-orbit plus magnetic Hamiltonian (see Sect. 3.4),
which in most cases must be performed numerically. However, the latter is a much
easier task, because the matrix involved is of lower rank and block-diagonal.
The statistical equilibrium equations in the energy-eigenvector representation
are given by Eq. (7.29). We are interested in the statistical tensors of the upper
term of a two-term atom whose lower term is, by assumption, unpolarized. Using
Eqs. (7.34e), (10.120) and (10.122), and neglecting stimulated emission, we obtain,
in stationary situations
N 1
ρβ L S (ju Mu , ju Mu ) =
u u N (2S + 1)(2L + 1)
TA (βu Lu Sju Mu ju Mu , β L Sj M j M )
× ,
A(βu Lu S → β L S) + 2πi νβ L S (ju Mu , ju Mu )
j M u u
where the rate TA is given by Eq. (7.34a) and the Bohr frequency νβ L S by
u u
Eq. (7.30). The statistical tensors of the upper term can now be obtained by
applying the transformation (7.37). Via a long but direct calculation involving the
use of Eqs. (3.62b), (5.156), (2.42), and (2.41), one obtains
N 3(2Ju + 1)(2Ju + 1)
βu Lu S K
ρQ (Ju , Ju ) = K
N 2S + 1
Ju Ju K
! !
1+L −S+Ju +Ju +Ju +K +Q 1 1 K Lu Lu K
× (−1)
Lu Lu L Ju Ju S
× J−Q
K
(νβ L S, β L S ) , (10.150)
u u
where the kernel K is given by
Ju Ju K Ju Ju K
K= (2K + 1)(2K + 1)
Mu −Mu −Q Mu −Mu −Q
Mu Mu
j j j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu )
u u u u
ju ju
B(β L S → βu Lu S)
× .
A(βu Lu S → β L S) + 2πi νβ L S (ju Mu , ju Mu )
u u
Obviously, for zero magnetic field we get back previous results. For B = 0 we have
in fact
j
CJu → δj J , νβ L S (ju Mu , ju Mu ) → νβ L SJ , β L SJ ,
u u u u u u u u u u u
600 CHAPTER 10
where the last Bohr frequency is given by Eq. (7.43). Use of Eq. (2.23a) yields
B(β L S → βu Lu S)
lim K = δKK δJ J δJ J ,
B→0 u u u u A(βu Lu S → β L S) + 2πi νβ L SJ , βu Lu SJu
u u u
hence Eq. (10.150) reduces to Eq. (10.126).
To obtain the radiation emitted by the atom along a given direction Ω, we have
to substitute Eq. (10.150) into Eq. (7.47e). The resulting expression is very com-
plicated, and will not be written down explicitly. We rather restrict attention to
defined in Eq. (10.130). Since
the frequency-integrated emission coefficient, ε̃i (Ω),
the profiles Φ(νβ L Sj M , β L Sj M − ν) are normalized to unity in frequency, the
u u u u
turns out to be identical to Eq. (7.48g) except for the disap-
expression for ε̃i (Ω)
pearance of the profile φ(νβ L S, β L S − ν). Substitution of Eq. (10.150) in this
u u
expression yields, with the use of Eqs. (5.157), (7.33) and (10.128)
5
dΩ
3
= kA
ε̃i (Ω) Ω
Pij (Ω, ; B)
I (ν ) ,
,Ω (10.151)
M fs j βu Lu S, β L S
4π j=0
Ω
where [Pij (Ω, ; B)]
, the scattering phase matrix for a fine-structure multiplet in
fs
the presence of a magnetic field, is given by1
Pij (Ω, Ω ; B)
= WKK Q (β L Sβu Lu ; B) fs
fs
KK Q
× (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω (10.152)
−Q
with
! !
3(2Lu + 1) 1 1 K 1 1 K
WKK Q (β L Sβu Lu ; B) fs =
2S + 1 Lu Lu L Lu Lu L
× (2K + 1)(2K + 1)(2Ju + 1)(2Ju + 1)(2Ju + 1)(2Ju + 1)
Ju Ju Ju Ju Mu Mu
! !
Lu Lu K Lu Lu K Ju Ju K Ju Ju K
×
Ju Ju S Ju Ju S −Mu Mu −Q −Mu Mu −Q
j j j j
× CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu ) CJu (βu Lu S, Mu )
u u u u
ju ju
1
× . (10.153)
S (ju Mu , ju Mu ) A(βu Lu S → β L S)
1 + 2πi νβ
L
u u
The symbol now introduced depends on the quantum numbers specifying the mul-
tiplet (L , Lu , S), on the fine-structure intervals of the upper term, on the Einstein
1
The following expression was first derived by Landi Degl’Innocenti (1990).
NON-EQUILIBRIUM ATOMIC PHYSICS 601
coefficient A(βu Lu S → β L S), and on the magnetic field modulus B. Its main
properties are collected in App. 14.
In the following we apply Eq. (10.151) to a particularly simple case, i.e., the
transition 3 S − 3 P (L = 0, Lu = 1, S = 1). We assume that the upper term
satisfies the Landé interval rule, so that it can be fully characterized by the quantity
ζ(βu Lu S) – see Eqs. (3.60), (10.140). Bearing in mind Eqs. (10.153) and (3.61), it
is immediately seen that the Stokes parameters of the scattered radiation depend
on A(βu Lu S → β L S), ζ(βu Lu S) and B via the two ratios
ζ(βu Lu S) µ0 B
and .
A(βu Lu S → β L S) A(βu Lu S → β L S)
Equivalently, we can use the dimensionless parameters
π ζ(βu Lu S) µ0 B
x= , γ= , (10.154)
h A(βu Lu S → β L S) ζ(βu Lu S)
already introduced in Eqs. (10.142) and (3.63), respectively.
Let us assume that the incident radiation is a unidirectional, unpolarized beam,
and let us refer – as in previous sections of this chapter – to the scattering geometry
of Fig. 5.11. The relevant components of the tensors T Q K
for the incident and
scattered radiation can be derived from Table 5.6 (the simplest choice is θ = 90◦ ,
χ = 0◦ ; θ = β, χ = 90◦ , γ = 0◦ ). The frequency-integrated fractional polarization
of the scattered radiation can be derived from Eqs. (10.151)-(10.152). Taking into
account the properties of the symbol WKK Q given in App. 14, we obtain
3 sin2 β W220 + (1 + cos2 β) Re(W222 )
p̃Q =
8 + (1 − 3 cos2 β) W220 − 3 sin2 β Re(W222 )
−6 cos β Im(W222 )
p̃U =
8 + (1 − 3 cos2 β) W220 − 3 sin2 β Re(W222 )
√
−2 3 cos β W120
p̃V = . (10.155)
8 + (1 − 3 cos2 β) W220 − 3 sin2 β Re(W222 )
The last expression shows that the scattered radiation contains, in general, a def-
inite amount of circular polarization. This phenomenon is closely associated with
the existence of fine structure (it has no counterpart in the multi-level atom) and
will be discussed extensively in Sect. 10.20. The expressions for the linear po-
larization still bear a clear resemblance to the corresponding expressions for the
two-level atom (Eqs. (10.37)). For zero magnetic field the circular polarization van-
ishes and the expressions for the linear polarization reduce to Eqs. (10.143) – see
Eqs. (A14.6), (10.134) and (10.135).
602 CHAPTER 10
Fig.10.20. Polarization (or Hanle) diagrams for the transition between the lower term 3 S and the
upper term 3 P, corresponding to the scattering geometry of Fig.5.11 (only the curves for β = 0◦
and β = 180◦ are shown). The left panel is obtained for x = 0.25, the right panel for x = 0.5.
The curve β = 0◦ is labelled by the values of γ.
Fig.10.21. Same as Fig.10.20 for x = 1 (left panel) and x = 3 (right panel). The curve corre-
sponding to β = 180◦ (symmetrical to the curve β = 0◦ about the line p̃U = 0 ) is not drawn.
Note the scale difference with Fig.10.20.
Figures 10.20-10.21 show the Hanle diagrams obtained from Eqs. (10.155)1 by
assigning different values to the parameter x. The first thing to notice is the overall
1
Note that the evaluation of the symbols WKK Q requires the numerical calculation, for
each value of x and γ , of the eigenvalues and eigenvectors of the matrix defined in Eqs. (3.61).
Alternatively, one can determine (for each value of x and γ ) the upper-level statistical tensors by
solving numerically Eqs. (10.123); the frequency-integrated scattered polarization is then obtained
from Eq. (7.48g) – with the profile φ replaced by unity.
NON-EQUILIBRIUM ATOMIC PHYSICS 603
Fig.10.22. Absolute value of the quantity W222 for the transition 3 S − 3 P as a function of the
magnetic field strength (parameterized through γ), for different values of x. Note the appearance
of sharp peaks around the critical γ values 0.616, 0.707, and 1.217 for large x values.
decrease of the scattered polarization with increasing x (i.e., with decreasing the
Einstein A coefficient): this effect has already been pointed out in our discussion
of resonance polarization in the absence of magnetic fields (see Sect. 10.16).
The second phenomenon is the appearance of characteristic curly shapes which
become more and more prominent as x is increased. This phenomenon is due
to level-crossing interferences (see Bommier, 1980), and is associated with the
overlapping of different M -sublevels in the (incomplete) Paschen-Back effect – see
Sect. 3.4, and in particular Fig. 3.8 which refers to the term 3 P. Whereas in the
Zeeman effect of an ‘isolated’ J-level the M -sublevels spread out linearly in energy
with increasing magnetic field – with a monotonic decrease of the corresponding
coherences – the M -sublevels of a given term may get closer, or even overlap, as
the field is increased. Table 3.5 shows that for γ = 0.616, 0.707, and 1.217 there are
overlappings between pairs of M -sublevels of the term 3 P corresponding to ∆M = 2
(these are the only overlappings that affect scattering polarization in the particular
geometry considered). Around such critical γ values the coherence between the two
overlapping sublevels gets larger (in absolute value) and the scattering polarization
approaches its zero-field value, which explains the presence of curls in the diagrams.
On the other hand, the Einstein A coefficient controls the natural width of the
sublevels. Thus the larger is A, the larger is the γ-interval (around each critical
value) in which the sublevels effectively overlap. This explains why the curls in
the diagrams clearly show up for small values of A (large values of x) while they
practically disappear for larger values of A (smaller values of x).
As a further illustration of these concepts, we plot in Fig. 10.22 the absolute value
of W222 as a function of γ for three different values of x. It is clearly seen that the
604 CHAPTER 10
three peaks are almost ‘unresolved’ for small x values, while they get sharper and
sharper as x is increased.
Apart from the loopy structure of the diagrams, the Hanle effect in two-term
atoms involves another interesting phenomenon which shows up most clearly when
the magnetic field is perpendicular to the scattering plane. For β = 90◦ or 270◦ ,
Eqs. (10.155) yield
3 W220 + Re(W222 )
p̃Q =
8 + W220 − 3 Re(W222 )
p̃U = p̃V = 0 .
Let us compare the values of p̃Q corresponding to zero magnetic field and to very
strong magnetic field (in the sense specified in App. 14). From Eqs. (A14.6),
(10.135), and (A14.8) we have
3 W2 (L , Lu ) D2 (βu Lu S) fs
lim p̃Q =
B→0 4 − W2 (L , Lu ) D2 (βu Lu S) fs
3 W2 (L , Lu )
lim p̃Q = ,
B→∞ 8 + W2 (L , Lu )
and it may well happen that the second value is larger than the first. Consider
for instance the transition 3 S − 3 P. If the fine-structure separations of the upper
term are much larger than the Einstein A coefficient (x → ∞), one obtains from
Tables 10.1 and 10.6
lim p̃Q 0.224 , lim p̃Q 0.333 .
B→0 B→∞
This is an extreme example of a phenomenon that can be called the anti-level-
crossing effect.1 It is due to the growth of the element W220 with increasing
magnetic field. As shown in point g) of App. 14, the phenomenon is related to the
basis transformation of the energy eigenstates which takes place in the regime of
complete Paschen-Back effect.
Finally, it is interesting to evaluate the average of the Hanle phase matrix for
two-term atoms over an isotropic distribution of magnetic fields. Along the same
lines of Sect. 5.12 (see also Sect. 10.3), we obtain from Eq. (10.152)
Ω
Pij (Ω, ; B)
= MK (B) fs Rij(K)
(Ω, Ω ; 0) , (10.156a)
fs
K
where
1
MK (B) fs = WKKQ (β L Sβu Lu ; B) fs . (10.156b)
2K + 1
Q
1
This name has been suggested by Bommier (1980) who discusses the importance of the
effect on the polarization of the HeI D3 line at 5876 Å.
NON-EQUILIBRIUM ATOMIC PHYSICS 605
The quantity [MK (B)]fs generally decreases with increasing magnetic field. In
particular we have, using Eqs. (A14.6), (10.135), and (A14.8)
MK (B → ∞) fs 1
= .
MK (B = 0) fs (2K + 1) DK (βu Lu S) fs
If the fine-structure intervals of the upper term are much smaller than the Einstein
A coefficient, the depolarizing factor is unity (see Eq. (10.137)), thus the above
ratio is 1/3 for K = 1 and 1/5 for K = 2. This is the classical result predicted
by Eqs. (5.173). On the other hand, if the fine-structure intervals are comparable
with A, the depolarizing factor is less than unity, hence the above ratio is larger
than 1/3 and 1/5, respectively. For instance, for an upper term 3 P in the limit of
very large fine-structure intervals (x → ∞), one gets from Table 10.6
M1 (B → ∞) fs M2 (B → ∞) fs
0.667 , 0.719 .
M1 (B = 0) fs M2 (B = 0) fs
This is a further manifestation of the anti-level-crossing effect discussed above.
10.19. The Two-Term Atom: the Franken Effect
(unpolarized lower term)
Besides the characteristic ‘curls’ in the Hanle diagram, level-crossing interferences
also produce, in many cases, sharp variations of the intensity of the scattered
radiation: this phenomenon can be used in laboratory spectroscopy to measure
the fine-structure (and hyperfine-structure)1 energy separations of excited atomic
terms. The technique consists in pumping the excited term by a spectroscopic
lamp and in recording the intensity variation of the resonantly scattered radiation
as a function of the magnetic field strength. The knowledge of the magnetic field
values corresponding to the intensity peaks allows one to recover the value of the
fine-structure constants of the excited term.
Many different geometries can be envisaged to perform the experiment and, in
some cases, it can be convenient to introduce polarizers in the path of either the in-
coming or the outgoing beam. The simplest geometry is that of forward scattering,
schematically depicted in Fig. 10.23.
The signal SD recorded by the detector can be easily obtained from Eqs. (10.151)-
(10.152), using the values of the tensors T Q
K
given in Table 5.6 (with θ = δ, χ = 0◦ ,
◦
γ = 0 both for the incident and for the scattered beam). Bearing in mind the
properties of the symbol WKK Q (see App. 14), we get
1 2 3
SD = k 1 + 3 cos2 δ − 1 W220 + sin2 δ cos2 δ Re(W221 )
8 2
3
+ sin4 δ Re(W222 ) , (10.157)
8
1
Hyperfine structure is discussed in Sect. 10.22.
606 CHAPTER 10
unpolarized scattered
beam δ beam
lamp detector
cell
Fig.10.23. Schematic experimental set-up for measuring fine-structure constants via level-crossing
interferences (Franken effect). The cell contains a vapor of the atomic species to be investigated.
In practice, it is necessary to introduce a small tilt angle between the scattered and the incident
beam to avoid contamination of the signal by direct incoming light.
Fig.10.24. The signal SD at the photodetector (in units of k) as a function of the magnetic
field strength, for the scattering geometry of Fig.10.23. The resonant transition is 3 S − 3 P. The
relevant parameters are x = 10, δ = 45◦ .
where k is a constant proportional to the intensity of the incident radiation.
As an illustration, we show in Fig. 10.24 the signal SD as a function of the
magnetic field strength for the transition 3 S − 3 P already considered in the previous
section. The field strength is expressed in terms of γ, the parameter x is set to 10
(see Eqs. (10.154)), and the angle δ to 45◦ (or 135◦ ). According to Eq. (10.157),
both the interferences between sublevels with ∆M = 1 (contained in W221 ) and
those between sublevels with ∆M = 2 (contained in W222 ) play a role.
Figure 10.24 shows the appearance of six distinct peaks (in addition to the ‘stan-
dard’ peak at γ = 0). Each of them corresponds to the crossing of a specific pair
of M -sublevels with ∆M = 1 or ∆M = 2, in strict agreement with the results of
Table 3.5. Notice that there is no peak at γ = 0.45 (at this γ value there is a
crossing of two sublevels with ∆M = 3), and that the peak at γ 1.2 corresponds
in fact to two distinct crossings (with ∆M = 1 and ∆M = 2, respectively).
NON-EQUILIBRIUM ATOMIC PHYSICS 607
It is important to remark that the determination of fine-structure constants by
the method described above requires that the intensity peaks corresponding to level
crossings can be easily identified on a plot like that of Fig. 10.24. This means that
the Einstein A coefficient of the transition must be sufficiently small compared to
the fine-structure intervals (h A ζ, or x 1, for a term satisfying the Landé
interval rule).
The first application of this method dates back to 1959 (Colegrove et al., 1959)
and led to the measurement of the fine structure of the term 2 3 P of HeI. In lab-
oratory spectroscopy, the modulation of resonantly scattered radiation by level-
crossing interferences is referred to as the Franken effect (Franken, 1961).
10.20. The Two-Term Atom: the Alignment-to-Orientation
Conversion Mechanism
(unpolarized lower term)
The Hanle phase matrix for the two-term atom (Eq. (10.152)) differs radically from
the corresponding matrix for the two-level atom (Eq. (10.33)) because of the double
summation over K and K . The presence of this double summation, together with
the fact that the only non-zero components of the tensor T Q K are those with
(3, Ω)
K = 1 (see Table 5.6), has important consequences on the circular polarization of
the scattered radiation.
In a two-level atom, the only way to get frequency-integrated, circularly polarized
scattered radiation is to irradiate the atom with circularly polarized light. This
property is already contained in the classical expression of the scattering phase
matrix (Eqs. (5.99)) and passes directly to the quantum-mechanical expression
(Eq. (10.34)). It can be understood by observing that frequency-integrated circular
polarization can be emitted in a given transition only if some orientation is present
in the upper level (see Eq. (10.31)); on the other hand, orientation in the upper level
of a two-level atom can only be produced by illuminating the atom with circularly
polarized radiation.1
These properties are no longer valid for the two-term atom. Equation (10.150)
shows, because of the presence of the summation over K , that orientation can be
produced in the upper term even if no circular polarization is present in the incident
radiation. This phenomenon has properly been called the alignment-to-orientation
conversion mechanism 2 since the upper term has, so to say, the possibility of
converting some ‘alignment’ present in the radiation field (the K = 2 components
of the radiation field tensor) into atomic orientation (the K = 1 components of the
statistical tensors).3
1
An exception to this rule occurs in the presence of strong magnetic fields, provided
the spectrum of the incident radiation satisfies suitable conditions (see Sect. 10.5 and Landi
Degl’Innocenti, 1985b). Such peculiar situation is not considered in this section.
2
This name was proposed by Kemp et al. (1984).
3
Obviously, the component J00 makes no contribution to the atomic orientation of the upper
608 CHAPTER 10
scattered
radiation
unpolarized
radiation
Fig.10.25. Scattering geometry for illustrating the alignment-to-orientation conversion mecha-
nism. The magnetic field vector lies in the scattering plane.
In order to illustrate the phenomenon in a specific case, we consider the scattering
of an unpolarized radiation beam for the transition 3 S − 3 P (already used in the for-
mer sections) and the geometry of Fig. 10.25.1 The frequency-integrated fractional
circular polarization can be obtained from Eqs. (10.151)-(10.152) by substituting
the values of the tensors T Q K
given in Table 5.6 (with θ = δ, χ = 180◦ , γ = 0◦ ;
◦ ◦
θ = 90 − δ, χ = γ = 0 ) and by using the properties of the symbols WKK Q (see
App. 14). We get
√
2 sin δ 3 (3 cos2 δ − 1) W120 − 6 cos2 δ Re(W121 )
p̃V = .
8 + (3 sin2 δ − 1)(3 cos2 δ − 1) W220 − 3 sin2 δ cos2 δ 4 Re(W221 ) − Re(W222 )
Figure 10.26 shows p̃V as a function of γ, for x = 10 (see Eqs. (10.154)) and
δ = 40◦ . It is clear from the figure that the conversion mechanism is quite efficient
especially at those magnetic field values corresponding to level crossings between
pairs of M -sublevels with ∆M = 1 (cf. Table 3.5).
The importance of this mechanism for broad-band circular polarization in stellar
atmospheres has been discussed by Kemp et al. (1984). According to these au-
thors, Lehmann (1964) was apparently the first to realize that excitation of atoms
in a magnetic field by non-circularly polarized light could lead to atomic orien-
tation. Some circular polarization profiles observed in the radiation from solar
prominences have been interpreted by Landi Degl’Innocenti (1982b) on the basis
of this mechanism (to obtain the line profile of the scattered radiation in the four
term (see Eq. (A14.5)).
1
The scattered circular polarization p̃V for this transition has already been calculated for
a slightly different geometry (see Eqs. (10.155)). The geometry of Fig. 10.25, with the magnetic
field vector lying in the scattering plane, is more appropriate to illustrate the variation of p̃V with
the field strength, since level crossings with ∆M = 1 (contained in the factors WKK 1 ) also come
into play.
NON-EQUILIBRIUM ATOMIC PHYSICS 609
Fig.10.26. The frequency-integrated fractional circular polarization for the transition 3 S − 3 P
(with x = 10) and the scattering geometry of Fig.10.25 is plotted as a function of the magnetic
field strength. The value of the angle δ is 40◦ .
Stokes parameters one should substitute the statistical tensors of Eq. (10.150) into
the frequency-dependent expression of the emission coefficient, Eq. (7.47e)).
It should be remarked that the ‘opposite’ of the alignment-to-orientation conver-
sion mechanism exists as well. Such mechanism, that can be named orientation-
to-alignment conversion, is responsible for the appearance of an additional contri-
bution to the linear polarization of scattered radiation when the atom is irradiated
by circularly polarized light.
10.21. The Two-Term Atom: the Role of Lower-Term Polarization
(cylindrical symmetry - no magnetic field - no stimulated emission)
The results obtained in Sects. 10.16-10.20 are based on the hypothesis that no
atomic polarization is present in the lower term. If we want to release this hy-
pothesis, we have to go back to the statistical equilibrium equations (10.115) and
(10.116) established in Sect. 10.15. In the general case (arbitrary radiation field,
presence of a magnetic field) these equations are rather involved. In this section
we confine attention to a particularly simple, yet illuminating case.
Similarly to Sect. 10.7, we assume that the incident radiation field – besides
satisfying the flat-spectrum approximation, see Sect. 10.15 – is unpolarized and
cylindrically symmetrical about a fixed direction, which we take as the z-axis of our
reference system: this means that the only non-zero components of the radiation
field tensor are J00 (νβ L S, β L S ) and J02 (νβ L S, β L S ). We also assume that the
u u u u
610 CHAPTER 10
magnetic field is zero and that stimulation effects are negligible.
The first two assumptions (concerning the incident radiation and the magnetic
field) lead to the considerable simplification that, in stationary situations, the only
non-vanishing statistical tensors ρK
Q (J, J ) are those with K even and Q = 0. The
proof of this property is quite similar to that given in Sect. 10.7, because the
spontaneous emission rates TE , RE (Eqs. (7.45b) and (7.46b)) are proportional to
δKK δQQ , and the expressions of the absorption rates TA , RA (Eqs. (7.45a) and
(7.46a)) contain a 3-j symbol of the form (10.58); on the other hand, the kernel N
defined in Eq. (7.41) is also proportional to δKK δQQ for zero magnetic field.
An obvious consequence of this property is that the number of equations (and of
unknowns) is largely reduced. Such number is given by1
Neq = n(Lu , S) + n(L , S) , (10.158)
where n(L, S), the number of statistical tensors of the form ρQ (J, J )
βLS K
with K
even and Q = 0, can be written in the form
1 2
n(L, S) = J − J< 2J> + 2J> J< − 4J<
2
+ 6J> + 6J< + 1 + J> + ξ ,
6 >
where
1 if S is an integer
J> = L + S , J< = |L − S| , ξ= 1
2 if S is a half-integer .
Under the above assumptions, the statistical equilibrium equations (10.115) and
(10.116) reduce to the following (see Eqs. (7.41), (7.45b) and (7.46b))
d
ρ0 (Ju , Ju )
βu Lu S K
= − A(βu Lu S → β L S) + 2πi νβ L SJu , βu Lu SJu ρ0 (Ju , Ju )
βu Lu S K
dt u u
+ TA (βu Lu SK0Ju Ju , β L SK 0J J ) β L SρK
0 (J , J )
K even J J
d
ρ0 (J , J )
β L S K
= −2πi νβ L SJ , β L SJ ρ0 (J , J )
β L S K
dt
+ TE (β L SK0J J , βu Lu SK0Ju Ju ) βu Lu SρK
0 (Ju , Ju )
Ju Ju
− RA (β L SK0J J K 0J J ) β L SρK
0 (J , J ) , (10.159)
K even J J
where K is even and the Bohr frequencies are defined in Eq. (7.43). The expressions
of the rates TA , TE , RA are easily derived from Eqs. (7.45a,b) and (7.46a) with the
1
Because of the conjugation property (3.100), Neq also represents the number of real unknowns
of the algebraic system.
NON-EQUILIBRIUM ATOMIC PHYSICS 611
help of Eqs. (2.26a), (2.36a) and (2.49)
TA (βu Lu SK0Ju Ju , β L SK 0J J ) = (2L + 1) B(β L S → βu Lu S)
! !
Lu L 1 Lu L 1
× (2Ju + 1)(2Ju + 1)(2J + 1)(2J + 1)
J Ju S J Ju S
!
Ju Ju K
× δKK (−1)1+J +Ju J00 (νβ L S, β L S )
J J 1 u u
⎧ ⎫
⎨ Ju J 1 ⎬
J −J K K 2
+ 15(2K + 1)(2K + 1) (−1) J J 1
⎩ u ⎭ 0 0 0
K K 2
× J02 (νβ L S, β L S )
u u
TE (β L SK0JJ , βu Lu SK0Ju Ju ) = (2Lu + 1) A(βu Lu S → β L S)
× (2J + 1)(2J + 1)(2Ju + 1)(2Ju + 1) (−1)1+J +Ju
! ! !
J J K Lu L 1 Lu L 1
×
Ju Ju 1 J Ju S J Ju S
RA (β L SK0J J K 0J J ) = (2L + 1) B(β L S → βu Lu S)
1
× δKK δJ J δJ J J 0 (ν )
2L + 1 0 βu Lu S, β L S
!
1+Lu −S+J L L 2 K K 2
+ 15(2K + 1)(2K + 1) (−1)
1 1 Lu 0 0 0
! !
1 L L 2 K K 2
× δJ J (2J + 1)(2J + 1)
2 J J S J J J
! !
J −J L L 2 K K 2
+ δJ J (2J + 1)(2J + 1) (−1)
J J S J J J
!
× J02 (νβ L S, β L S ) .
u u
Equation (10.159) can be cast in dimensionless form by dividing both sides by the
Einstein coefficient A(βu Lu S → β L S) and by introducing the parameters
(2L + 1) B(β L S → βu Lu S) 0
J (ν )=
(2Lu + 1) A(βu Lu S → β L S) 0 βu Lu S, β L S
c2
= J00 (νβ L S, β L S ) = n̄ ,
2h νβ3 L S, β L S
u u
u u
612 CHAPTER 10
J02 (νβL S, β L S ) w
0
u u
=√ (10.160)
J0 (νβ L S, β L S ) 2
u u
(Eqs. (7.33) have been used in the first relation). The quantities n̄ and w – already
introduced in Sect. 10.7 – represent the solid-angle average of the number of photons
per mode and the anisotropy factor at the transition frequency, respectively. If both
the upper and lower term satisfy the Landé interval rule (Eq. (10.140)), we can
also introduce the parameters
π ζ(βu Lu S)
= xu ,
h A(βu Lu S → β L S)
π ζ(β L S)
=
h B(β L S → βu Lu S) J00 (νβ L S, β L S )
u u
2L + 1 π ζ(β L S)
= = x , (10.161)
2Lu + 1 h n̄ A(βu Lu S → β L S)
which represent the ratios of the typical fine-structure splitting of the upper and
lower term, respectively, to the corresponding inverse lifetime. The parameter xu
has already been used in former sections (see Eq. (10.142)).
Let us apply Eqs. (10.159) to the simplest transition where lower-term atomic
polarization (including coherences between different J-levels) plays a role: this is
the transition between a lower term 2 P and an upper term 2 S. The statistical
tensors involved are six; introducing shorthand notations, they can be written in
the form
, - , - , -
u 0 1 1 0 3 3 2 1 3
ρ0 , =x ρ0 , =z ρ0 , = u + iv
2 2 2 2 2 2
, - , - , -
0 1 1 2 3 3 2 3 1
ρ0 , =y ρ0 , =t ρ0 , = −u + i v ,
2 2 2 2 2 2
where x, y, z, t, u, v are real. We assume that the lower term satisfies the Landé
interval rule, so that the quantity x defined in Eq. (10.161) can be used. After
some calculations, involving the numerical evaluation of several 3-j, 6-j, and 9-j
symbols, we obtain the set of equations
√ √ √
1 d 1 2 2 2
x = −x + n̄ y + n̄ z + n̄w t − n̄w u
A(u → ) dt 3 3 6 3
√
1 d 1 1 2
y = x − n̄ y + n̄w u
A(u → ) dt 3 3 6
√
1 d 2 1 1 1
z= x − n̄ z − n̄w t + n̄w u
A(u → ) dt 3 3 6 6
1 d 1 1 1
t = − n̄w z − n̄ t − n̄w u
A(u → ) dt 6 3 6
NON-EQUILIBRIUM ATOMIC PHYSICS 613
√
1 d 2 1 1 1
u= n̄w y + n̄w z − n̄w t − n̄(4 − w) u − n̄ x v
A(u → ) dt 12 12 12 12
1 d 1
v = n̄ x u − n̄(4 − w) v , (10.162)
A(u → ) dt 12
and the trace equation (see Eq. (10.118))
√ √
2x+ 2y+2z=1.
The expressions of the statistical tensors corresponding to the stationary situation
are found to be1
,1 1- 1
0
ρ0 , = √ (4 − w2 ) (4 − w)2 + 144 x2 ∆
2 2 2
,3 3-
0
ρ0 , = (4 − w)2 (4 − w2 ) + 576 x2 ∆
2 2
, -
2 3 3
ρ0 , = −w (4 − w)2 (2 + w) + 288 x2 ∆
2 2
,1 3-
2
ρ0 , = w(4 − w)(2 + w) 4 − w + 12 i x ∆
2 2
,1 1- 1
u 0
ρ0 , = √ n̄(2 + w) (4 − w)2 (2 + w)(1 − w) + 144(2 − w) x2 ∆ , (10.163)
2 2 2
where
∆ = (4 − w)2 (2 + w) 3(2 − w) + n̄(1 − w)(2 + w)
+ 144 12 − w2 + n̄(4 − w2 ) x2 .
These equations show that the interference term, ρ20 (1/2, 3/2), decreases with in-
creasing x and tends to zero for x → ∞. Under this limit, the two-term atom con-
sidered here reduces to a special case of the three-level atom examined in Sect. 10.12
(two lower ‘uncorrelated’ levels with J = 1/2 and J = 3/2 respectively, one up-
per level with J = 1/2). It can be easily seen that the results there obtained
(Eqs. (10.89a-c)), once adapted to the present case by the substitutions
n̄ca → n̄ , n̄cb → n̄ , wcb → w , δb(2) → 0 ,
coincide with those derived from Eqs. (10.163) for x → ∞.
1
It can easily be shown that the determinant of the (algebraic) system (10.162) is zero – cf.
footnote on p.582.
614 CHAPTER 10
On the other hand, for x → 0 the interference term is comparable with the other
statistical tensors. Under this limit, the J-levels of the lower term are degenerate
and one can apply the principle of spectroscopic stability (see Sect. 7.10). The
statistical tensors of the lower and upper term (now considered as spinless terms)
can be derived from Eq. (3.105). Bearing in mind Eq. (3.100), we obtain
√
0 3 (2 − w)
ρ0 (L = 1) =
3(2 − w) + n̄(1 − w)(2 + w)
√
6w
ρ0 (L = 1) = −
2
3(2 − w) + n̄(1 − w)(2 + w)
u 0 n̄(1 − w)(2 + w)
ρ0 (L = 0) = .
3(2 − w) + n̄(1 − w)(2 + w)
These results are in agreement with those obtained in Sect. 10.7 for the two-level
atom with J = 1, Ju = 0 (cf. Eqs. (10.64) with δ(2) = 0).
Equations (10.163) show that the statistical tensors of both the lower and the
upper term are strongly sensitive to the value of x . It should be remarked, how-
ever, that the quantity x is fairly large for most fine-structure multiplets under
typical astrophysical or laboratory conditions (where the radiation field is very di-
luted, n̄ 1). As an example, we can consider the MgI triplet at λ 5170 Å.
The lower term is a 3 P whereas the upper term is a 3 S. The three lines of the
multiplet – named b1 , b2 , and b4 , respectively – lie at 5183.6 Å (J = 2, Ju = 1),
5172.7 Å (J = 1, Ju = 1), and 5167.3 Å (J = 0, Ju = 1). The quantity ζ/h for the
lower term is approximately 6 × 1011 s−1 (see Eq. (10.140)), and the Einstein A
coefficient is 1.04 × 108 s−1 . Assuming n̄ 10−2 (typical of the solar atmosphere),
we get x 5 × 106 , which means that coherences between different J-levels of the
lower term are 6 to 7 orders of magnitude smaller than the ‘diagonal’ statistical
tensors. In particular, the atomic polarization of the upper term (hence scattering
polarization) is practically unaffected by lower-term interferences.
The simple transition 2 P−2 S discussed above requires the solution of an algebraic
system of 6 equations. For more complicated transitions, it is almost mandatory to
resort to a numerical solution of the statistical equilibrium equations (10.159). The
Stokes parameters of the scattered radiation can then be obtained by substituting
the values of the statistical tensors of the upper term into Eq. (10.127).
In the following we illustrate the results of such numerical calculations for two
different multiplets, multiplet UV n. 1 of FeII (6 D − 6 D) – already considered in
Sect. 10.17 – and multiplet n. 2 of MgI (3 P − 3 S). According to Eq. (10.158),
the corresponding algebraic systems contain 110 and 12 equations, respectively. In
both cases we refer to the 90◦ scattering of an unpolarized radiation beam (Fig. 10.1
with Θ = 90◦ ): the anisotropy factor w is unity, and the fractional polarization of
the scattered radiation is independent of the average number of photons n̄ provided
n̄ 1 (which is implicit in Eqs. (10.159) since stimulated emission is neglected).1
1
In fact it can be shown, bearing in mind the structure of Eqs. (10.159) and the expressions of
NON-EQUILIBRIUM ATOMIC PHYSICS 615
Fig.10.27. Fractional polarization pQ against vacuum wavelength λv (in Å) in the 90◦ scattering
of an unpolarized radiation beam, for multiplet UV n.1 of FeII. The reference direction for linear
polarization is the perpendicular to the scattering plane. The solid line is obtained by taking
lower-term atomic polarization into account, the dashed line (identical to the curve in Fig.10.18)
by neglecting it.
Figures 10.27 and 10.28 show the fractional linear polarization pQ as a function
of wavelength in the two multiplets. For the multiplet of FeII the results are
not very different from those obtained in the absence of lower-term polarization.
On the contrary, there is a dramatic effect for the multiplet of MgI. The fractional
polarization has a complicated wavelength dependence, with a peak value of almost
40% (in absolute value): if lower-term atomic polarization were neglected, pQ would
be zero at all wavelengths.1
the radiative rates, that the statistical tensors of the upper term are proportional to n̄ provided n̄
is much smaller than the ratio between the typical fine-structure splitting of the lower term and
the Einstein A coefficient.
1
This is easily verified from Eq. (10.129). The only non-zero components of the tensor
TQ have K = 2 (see Table 5.6), and the 6-j symbol
K (1, Ω)
!
1 1 K
Lu Lu L
is zero for Lu = 0 and K = 2. Note that the same symbol appears in Eq. (10.126), which means
that there is no atomic polarization in the upper term 3 S.
616 CHAPTER 10
Fig.10.28. Same as Fig.10.27 for multiplet n.2 of MgI. The vertical lines mark the wavelength
positions of the three lines of the multiplet according to the code: 1) J = 0, Ju = 1 (b4 line);
2) J = 1, Ju = 1 (b2 line); 3) J = 2, Ju = 1 (b1 line). Atomic data are from Moore (1949). The
value of the Einstein A coefficient is 1.04 × 108 s−1 .
10.22. The Two-Level Atom with Hyperfine Structure
As remarked in Sect. 7.9, the statistical equilibrium equations for the multi-level
atom with hyperfine structure bear a very strong resemblance to the corresponding
equations for the multi-term atom, and the same holds for the expressions of the
radiative transfer coefficients. Most of the results obtained in this chapter for the
two-term atom can thus be extended to the two-level atom with hyperfine structure
by means of simple, formal transformations.
Let us consider a simplified atomic model consisting of two levels, a lower level
having quantum numbers (α , J ) and an upper level having quantum numbers
(αu , Ju ). If the atom has nuclear spin I, both levels are split into a collection
of hyperfine components characterized by the total angular momentum F (for
the lower level) and Fu (for the upper level). When a magnetic field is present,
the F -levels are further split into their magnetic components as fully explained in
Sect. 3.5.
For this model atom, we can study the problem of resonance scattering under
the flat-spectrum approximation and the unpolarized lower level assumption. The
relevant equations are readily obtained by carrying out the formal substitutions
NON-EQUILIBRIUM ATOMIC PHYSICS 617
β → α βu → αu S→I
L → J Lu → Ju
J → F Ju → Fu (10.164)
on the corresponding equations of Sect. 10.16.
Consider for instance the expression of the frequency-integrated absorption co-
A
efficient. According to Eqs. (10.164), the quantity kM defined in Eq. (10.128)
1
transforms into
h να J , α J
u u
N B(α J → αu Ju ) ,
4π
which is just the frequency-integrated absorption coefficient in the line, kLA , defined
in Eq. (9.5).
Similarly, from Eqs. (10.131)-(10.133) we obtain the expression of the frequency-
integrated emission coefficient
5
dΩ
3
= kA
ε̃i (Ω) Ω
P (Ω, ) I (ν ) ,
,Ω
L
4π j=0 ij hfs j αu Ju , α J
Ω
where [Pij (Ω, )] , the hyperfine-structure scattering matrix, is given by
hfs
Ω
Pij (Ω, ) = WK (α J Iαu Ju ) hfs (−1)Q T Q
K
(i, Ω) ) ,
T K (j, Ω
hfs −Q
KQ
with
3(2Ju + 1)
WK (α J Iαu Ju ) hfs =
2I + 1
!2 !2
1 1 K Ju Ju K
× (2Fu + 1)(2Fu + 1)
Ju Ju J Fu Fu I
Fu Fu
1
× ,
1 + 2πi να J IFu , αu Ju IFu A(αu Ju → α J )
u u
where the Bohr frequency να J IF , α J IF is given by Eq. (7.67).
u u u u u u
It is also possible to define the depolarizing factors due to hyperfine structure.
From Eqs. (10.135)-(10.136) we have
WK (α J Iαu Ju ) hfs = WK (J , Ju ) DK (αu Ju I) hfs ,
1
Following the formalism of Sect. 7.9, we omit the index I when inessential.
618 CHAPTER 10
where
!2
1 J J K
DK (αJI) hfs = (2F + 1)(2F + 1)
2I + 1
F F I
FF
1
× .
1 + 2πi ναJIF , αJIF A(αJ → α J )
The properties of the depolarizing factors are strictly similar to those of their fine-
structure analogues (cf. Eqs. (10.137)-(10.142)):
a) If the hyperfine-structure intervals of the upper level are much smaller than the
Einstein A coefficient, the depolarizing factors reduce to unity.
b) If such intervals are much larger than A, they tend to the asymptotic values
!2
∞
1 2 J J K
DK (JI) hfs = (2F + 1) .
2I + 1 F F I
F
Numerical values of [D1∞ ]hfs and [D2∞ ]hfs for half-integral values of J are given
in Table 10.8 (the table is restricted to I ≤ 9/2, J ≤ 15/2). For integral values
of J one can directly use Table 10.6, taking into account the formal identity of
∞ ∞
[DK (JI)]hfs and [DK (LS)]fs under the substitution J → L, I → S.
c) If the hyperfine-structure intervals of the upper level are simply described by
the magnetic-dipole term, one has (see Eqs. (3.70))
1
ναJIF , αJIF = A(α, J, I) F (F + 1) − F (F + 1) ,
2h
and the depolarizing factor can be written in the form
!2
1 J J K
DK (αJI) hfs = (2F + 1)(2F + 1)
2I + 1
F F I
FF
1
× 2 ,
1+ x2 F (F + 1) − F (F + 1)
where x is given by
π A(α, J, I)
x= . (10.165)
h A(αJ → α J )
The full frequency-dependent emission coefficient is given by the following expres-
sion, obtained by carrying out the formal transformations (10.164) on Eq. (10.129)
NON-EQUILIBRIUM ATOMIC PHYSICS 619
TABLE 10.8
∞ (JI)]
Values of the depolarizing factors [DK hfs for levels with half-integral J quantum number
I J D1∞ D2∞ I J D1∞ D2∞ I J D1∞ D2∞
1/2 1/2 0.5 0. 2 1/2 0.36 0. 7/2 1/2 0.344 0.
3/2 0.875 0.625 3/2 0.433 0.247 3/2 0.37 0.219
5/2 0.944 0.833 5/2 0.617 0.266 5/2 0.418 0.237
7/2 0.969 0.906 7/2 0.768 0.444 7/2 0.5 0.258
9/2 0.98 0.94 9/2 0.847 0.600 9/2 0.637 0.290
11/2 0.986 0.958 11/2 0.892 0.705 11/2 0.735 0.399
13/2 0.990 0.969 13/2 0.920 0.776 13/2 0.800 0.508
15/2 0.992 0.977 15/2 0.939 0.825 15/2 0.844 0.597
1 1/2 0.407 0. 5/2 1/2 0.352 0. 4 1/2 0.342 0.
3/2 0.692 0.293 3/2 0.400 0.233 3/2 0.362 0.215
5/2 0.856 0.609 5/2 0.5 0.262 5/2 0.399 0.229
7/2 0.918 0.766 7/2 0.677 0.321 7/2 0.457 0.248
9/2 0.947 0.847 9/2 0.783 0.474 9/2 0.564 0.254
11/2 0.963 0.892 11/2 0.846 0.599 11/2 0.674 0.324
13/2 0.973 0.920 13/2 0.885 0.689 13/2 0.751 0.425
15/2 0.979 0.938 15/2 0.911 0.754 15/2 0.805 0.519
3/2 1/2 0.375 0. 3 1/2 0.347 0. 9/2 1/2 0.34 0.
3/2 0.5 0.27 3/2 0.382 0.224 3/2 0.356 0.212
5/2 0.743 0.395 5/2 0.448 0.247 5/2 0.385 0.224
7/2 0.85 0.603 7/2 0.583 0.258 7/2 0.430 0.239
9/2 0.902 0.729 9/2 0.712 0.366 9/2 0.5 0.256
11/2 0.932 0.806 11/2 0.793 0.494 11/2 0.612 0.274
13/2 0.950 0.854 13/2 0.845 0.598 13/2 0.700 0.354
15/2 0.961 0.887 15/2 0.880 0.677 15/2 0.763 0.445
= k A 2Ju + 1
εi (ν, Ω)
(−1)I−J +Fu +Fu +F +K+Q
L
2I + 1
KQ Fu Fu F
! !
Ju J 1 Ju J 1
× 3(2Fu + 1)(2Fu + 1)(2F + 1)
F Fu I F Fu I
! ! !
1 1 K 1 1 K Ju Ju K J K (ν
× TQ
K
(i, Ω) −Q αu Ju , α J )
Fu Fu F Ju Ju J Fu Fu I
∗
1 Φ(ναu Ju IFu , α J IF − ν) + Φ(ναu Ju IFu , α J IF − ν)
× . (10.166)
2 1 + 2πi να J IF , α J IF A(αu Ju → α J )
u u u u u u
If the width of the F -sublevels of the lower level is negligible compared to the width
of the F -sublevels of the upper level, the last term can be written in the simpler
620 CHAPTER 10
form (cf. Eq. (10.144))
1 Γ
,
π Γ − i (να J IFu , α J IF − ν) Γ + i (να J IFu , α J IF − ν)
u u u u
where
A(αu Ju → α J )
Γ = .
4π
It should be remarked that the frequency distance between the individual lines
of a hyperfine-structure multiplet is usually comparable with their natural width.
Therefore, the value of the emission coefficient at the frequency of a specific line
(F̄u , F̄ ) cannot be deduced by restricting the summation in Eq. (10.166) to the
values Fu = Fu = F̄u , F = F̄ : in other words, the analogue of Eq. (10.145) for
hyperfine-structure multiplets is not valid. On the contrary, the asymptotic ex-
pression given in Eq. (10.148) has its direct hyperfine-structure counterpart. At
large distance from the single lines we have
1 Γ
≈ kA
εi (ν, Ω) WK (J , Ju ) (−1)Q T Q
K J K (ν
(i, Ω) −Q αu Ju , α J ) ,
L
π (ν0 − ν)2
KQ
where ν0 is the ‘center of gravity’ of the multiplet.
To make an example, let us consider the NaI D2 line at 5890 Å (2 S1/2 − 2 P3/2 ).
The nuclear spin is I = 3/2, the Einstein coefficient A = 6.3 × 107 s−1 , and the
hyperfine-structure constant for the upper level is approximately A/h = 1.6 ×
107 s−1 . From Eq. (10.165) we obtain x = 0.80, and the depolarizing factors
[D1 ]hfs and [D2 ]hfs turn out to be 0.56 and 0.30, respectively. The line consists
of 6 components, with a natural width of the order of 1 mÅ. Owing to the large
hyperfine splitting of the lower level, they are divided in two groups of 3 components
each. The typical wavelength distance between the components of either group is
1 mÅ (comparable with the natural width), whereas the separation between the
two groups is about 20 mÅ.
Figure 10.29 shows the fractional linear polarization to be expected in the 90◦
scattering of an unpolarized radiation beam for this line. One can notice the
strong depolarizing effect of hyperfine structure and the rapid approach of pQ to
its asymptotic value (pQ = 3/7, typical of the transition (J = 1/2, Ju = 3/2)) at
a distance of a few mÅ from either group of components. This ‘spiky’ behavior of
the fractional polarization with wavelength is due to the fact that the hyperfine-
structure intervals of the upper level are of the same order as the Einstein A
coefficient.
The Hanle effect in hyperfine-structured lines can be deduced by applying trans-
formations (10.164) to the relevant equations derived in Sect. 10.18. In this case,
however, one has to apply the additional formal substitutions (see Eqs. (7.64))
ju → iu , Mu → fu ,
NON-EQUILIBRIUM ATOMIC PHYSICS 621
Fig.10.29. Fractional polarization pQ against wavelength distance from line center (in m Å) in
the 90◦ scattering of an unpolarized radiation beam, for the NaI D 2 line. Hyperfine-structure
constants are from Fuller and Cohen (1969), Hartmann (1970), Figger and Walther (1974).
connected with the presence of the magnetic field. Neglecting stimulation ef-
fects, we can write the frequency-integrated emission coefficient in the form (cf.
Eqs. (10.151)-(10.153))
5
dΩ
3
= kA
ε̃i (Ω) Ω
Pij (Ω, ; B)
I (ν ) ,
,Ω
L hfs j αu Ju , α J
4π j=0
where [Pij (Ω, Ω ; B)]
hfs , the Hanle phase matrix for a hyperfine-structure multiplet,
is given by
Ω
Pij (Ω, ; B)
= WKK Q (α J Iαu Ju ; B) hfs (−1)Q T Q
K T K (j, Ω
(i, Ω) ) ,
hfs −Q
KK Q
with
! !
3(2Ju + 1) 1 1 K 1 1 K
WKK Q (α J Iαu Ju ; B) hfs =
2I + 1 Ju Ju J Ju Ju J
× (2K + 1)(2K + 1)(2Fu + 1)(2Fu + 1)(2Fu + 1)(2Fu + 1)
Fu Fu Fu Fu fu fu
! !
Ju Ju K Ju Ju K Fu Fu K Fu Fu K
× ×
Fu Fu I Fu Fu I −fu fu −Q −fu fu −Q
622 CHAPTER 10
Fig.10.30. Polarization (or Hanle) diagram for the D 2 line of NaI, corresponding to the scattering
process illustrated in Fig.5.11. The curve β = 0◦ is labelled by the field intensity in G.
i i i i
× CFu (αu Ju I, fu ) CFu (αu Ju I, fu ) CFu (αu Ju I, fu ) CFu (αu Ju I, fu )
u u u u
iu iu
1
× . (10.167)
1 + 2πi να J
u u
I (iu fu , iu fu ) A(αu Ju → α J )
The Bohr frequency occurring in the last term is defined by (cf. the analogous
definition (7.30) for fine structure)
λi (αJI, f ) − λi (αJI, f )
ναJI (if, i f ) = .
h
The eigenvalues λi (αJI, f ) and the coefficients CFi (αJI, f ) are obtained by diag-
onalization of the hyperfine-structure and magnetic Hamiltonians, as described in
Sect. 3.5.
Figure 10.30 shows the Hanle diagram for the NaI D2 line and the scattering
geometry of Fig. 5.11, with evident loops due to level-crossing interferences. In
the geometry considered, only crossings between levels with ∆f = 2 are effective.
From Fig. 3.11 one can readily see that four such crossings occur, for B values of
the order of 12, 22, 25, and 43 G, respectively. This is consistent with the curves
in Fig. 10.30.
Finally, it should be remarked that the phenomena due to the anti-level-crossing
effect, the Franken effect, and the alignment-to-orientation conversion mechanism
NON-EQUILIBRIUM ATOMIC PHYSICS 623
Fig.10.31. Depolarizing effect due to an isotropic distribution of magnetic fields on the NaI D 2
line. The quantities [MK (B)]hfs are defined in Eq.(10.168).
– described in Sects. 10.18, 10.19 and 10.20, respectively – also take place in
hyperfine-structure multiplets. All these phenomena are conveniently described
in terms of the symbol WKK Q defined in Eq. (10.167). For instance, the Hanle
phase matrix averaged over an isotropic distribution of magnetic fields is given by
(cf. Eqs. (10.156a,b))
Ω
Pij (Ω, ; B)
= (K)
MK (B) hfs Rij (Ω, Ω ; 0) ,
hfs
K
with
1
MK (B) hfs = WKKQ (α J Iαu Ju ; B) hfs . (10.168)
2K + 1
Q
Figure 10.31 shows the behavior of the quantities [MK (B)]hfs with magnetic field
intensity for the NaI D2 line. Note the conspicuous anti-level-crossing effect on this
particular line. The ratio [MK (B → ∞)]hfs /[MK (B = 0)]hfs is found to be 0.60 for
K = 1 and 0.67 for K = 2, rather than 1/3 and 1/5, respectively, as it would if
hyperfine separations were negligible compared to the Einstein A coefficient.
CHAPTER 11
ASTROPHYSICAL APPLICATIONS:
SOLAR MAGNETOMETRY
Solar magnetometry, or the ‘art’ of measuring magnetic fields in the solar atmo-
sphere, started almost one century ago with the first application of polarimetry
to astrophysical research. By means of a Fresnel prism (acting as a quarter-wave
plate) and a Nicol prism (acting as a polarizer), in 1908 Hale succeeded in taking
two separate spectra of a sunspot in right and left circular polarization. The line
shifts between the two spectra, due to the Zeeman effect, led to the first measure-
ment of a solar magnetic field (Hale, 1908).
Solar magnetometry has enormously evolved since the pioneering work of Hale
and has now become a mature science which employs sophisticated technologies
and modern astronomical instrumentation. The photographic plate used by Hale
has been replaced, during the years, by photomultipliers, diode arrays, and CCD
cameras. At the same time, new polarimeters have been conceived, capable of
attaining higher and higher polarimetric accuracy. The advent of techniques of
image stabilization and the construction of new instruments in proper sites (or
operating from space) has also opened the possibility of observing the sun at higher
and higher spatial and temporal resolution.
All these technological improvements have led to a better understanding of solar
magnetism but, at the same time, have opened new questions about the meaning
itself of the process of measuring magnetic fields in a highly structured plasma such
as the solar atmosphere. There is today strong evidence that the solar magnetic
field outside sunspots is often concentrated in structures whose lateral dimensions
are at the limit of the resolving power that can be attained with present instru-
mentation, which is of the order of a few tenths of arcsec1 (see for instance Stenflo,
1994). Even more complex is the picture emerging from some recent ideas, which
point to the existence of magnetic fields organized in fibrils of some few km in di-
ameter (the MISMA hypothesis, Sánchez Almeida et al., 1996), and of ‘turbulent’
fields of a few G permeating the full solar atmosphere outside active regions (Sten-
flo, 1994). Although these ideas have not yet been fully confirmed by observations,
they clearly suggest the difficulties inherent in the measurement of a quantity, the
magnetic field, that may show sharp variations over length-scales smaller than the
resolution element of the instrument.
Most of the methods and the techniques that have been developed for solar
magnetometry ignore, however, this kind of difficulties, and are based on the simple
assumption that the magnetic field is uniform across the resolution element. In
some cases a less restrictive assumption is made, i.e., that the magnetic field is
uniform but covers only a fraction f – the so-called filling factor – of the observed
1
One arcsec corresponds to about 730 km on the solar surface.
626 CHAPTER 11
solar area, while the remaining fraction (1 − f ) is field-free or non-magnetic. In
recent times, more sophisticated methods which allow for line-of-sight gradients or
for more complicated structuring of the magnetic field have also been developed.
In this chapter we will describe in some detail only the most traditional methods,
whereas the other ones will be briefly reviewed in a single section (Sect. 11.6).
A particular problem of solar magnetometry is the ambiguity of the azimuth of
the magnetic field vector, which is related to a symmetry property of the solution
to the transfer equation (see Eq. (9.57)). To solve this problem, some specific
‘disambiguation’ techniques have been developed. These are described in Sect. 11.7.
11.1. The Longitudinal Magnetograph
First introduced in solar research in the early 1950s (Babcock, 1953), the lon-
gitudinal magnetograph is probably the most renowned and certainly the most
widespread instrument for the study of solar magnetism.
A typical longitudinal magnetograph performs a polarimetric and spectral analy-
sis of the solar radiation and records, for each resolution element, a signal SV given
by
V (λ) p(λ) dλ
SV = , (11.1)
I(λ) p(λ) dλ
where I(λ), V (λ) are the Stokes parameters and p(λ) is a narrow profile centered
over one of the wings (the blue or the red) of a magnetically sensitive line. The
signal is then encoded on a grey-scale to produce a map of the full sun or of an
active region called a magnetogram.
The spectral analysis can be performed by a Lyot-type birefringent filter or by an
interference filter or by a combination of both. In this case p(λ) is nothing but the
filter profile. Alternatively, one can use a spectrograph and feed the detector by
means of a slit in its focal plane. In this case p(λ) is (nominally) a rectangle function
that differs from zero in the wavelength interval corresponding to the slit. The
polarimetric analysis, which must take place before the spectral analysis, can be
carried out in a variety of different ways. In the original Babcock’s magnetograph,
for instance, it was performed by the combination of an ADP crystal (see Sect. 1.5)
and a polarizer. In some magnetographs two signals are recorded, one in the blue
wing and the other (of opposite sign) in the red wing. The two individual signals
are then combined to obtain a composite signal with less noise.
The signal SV can be related in a simple way to the longitudinal component of
the magnetic field vector in the observed area under a number of assumptions, the
most important of which is the weak field approximation. We have seen in Sect. 9.6
that, under this approximation, the V Stokes parameter is given by the expression1
1
We recall here that the only condition for this formula to hold – besides the weak field
SOLAR MAGNETOMETRY 627
(Eq. (9.80))
∂I(λ)
V (λ) = −∆λB ḡ cos θ , (11.2)
∂λ
where ∆λB is the Zeeman splitting defined in Eq. (3.13), ḡ is the effective Landé
factor of the selected line (defined in Eq. (3.44)), θ is the angle between the magnetic
field direction and the line of sight (see Fig. 9.1), and I(λ) is the intensity profile.
Since V (λ) is linear in the product B cos θ and I(λ) is unaffected by the magnetic
field to lowest order, under the weak field approximation we have
SV = C B , (11.3)
where B = B cos θ is the longitudinal component of the magnetic field and where
C is the so-called calibration constant of the longitudinal magnetograph, that can
be deduced via more or less complicated procedures involving the knowledge of the
filter profile p(λ) and of the line profile I(λ).
As a first-order approximation, one can assume that the line profile has a Gaus-
sian shape of the form (cf. Eq. (9.86))
λ−λ 2
− 0
∆λp
I(λ) = Ic 1 − dc e , (11.4)
where Ic is the continuum intensity, dc is the line central depression in units of Ic ,
λ0 is the line center wavelength, and ∆λp is a quantity proportional to the line
width.1 If the filter profile has also a Gaussian shape, which is typical of filter-type
magnetographs, the integrals in Eq. (11.1) can be performed analytically and the
calibration constant can readily be recovered. Let us set
λ−λ 2
− F
∆λF
p(λ) = k e , (11.5)
where k is an arbitrary constant, λF is the central wavelength of the filter, and
∆λF is a quantity proportional to its width. Evaluation of elementary integrals
yields
SV = −∆λB ḡ cos θ X ,
where
(λF −λ0 )2
(λF −λ0 ) ∆λp −
∆λ2p + ∆λ2F
2 dc e
(∆λ2p + ∆λ2F )3
X= .
(λF −λ0 )2
−
∆λp ∆λ2p + ∆λ2F
1 − dc e
∆λ2p + ∆λ2F
assumption – is the constancy with depth of the longitudinal component of the magnetic field;
see Table 9.1.
2
1
In a profile of the form e−(x/a) √, the quantity a is connected with the full width at half
maximum, ∆x, by the relation ∆x = 2 ln 2 a 1.665 a.
628 CHAPTER 11
Recalling Eqs. (11.3) and (3.14), the calibration constant of the longitudinal mag-
netograph can thus be written in the form
C = −4.67 × 10−10 λ20 ḡ X ,
−1
where C is in G−1 , λ0 in Å and X in mÅ .
It is often in the observer’s interest to increase the signal SV (for an assigned line
and a given magnetic field) as much as possible. This can be obtained by carefully
selecting the filter position λF so as to maximize the absolute value of the quantity
X. As an example, consider the FeI line λ 6302.5, a Zeeman triplet having ḡ = 2.5.1
From the atlas of the solar spectrum by Delbouille et al. (1973) one can estimate
the quantities dc and ∆λp of Eq. (11.4). Their values, which refer to disk center,
are found to be dc 0.66 and ∆λp 60 mÅ. If the filter passband is ∆λF = 75 mÅ
(a typical value for filter-type magnetographs), the extrema of X are located at
−1
(λF − λ0 ) = ± 57.3 mÅ, where X = ± 5.05 × 10−3 mÅ . The corresponding value
of the calibration constant (when the filter is positioned in the blue wing) is
C = 2.3 × 10−4 G−1 .
The linear relationship between polarization signal and longitudinal component
of the magnetic field is indeed valid only in the limit of weak field. When the field
intensity increases, the signal SV turns out to be much smaller than what expected
from the linear relationship and the magnetograph is said to enter the saturation
regime.
To analyze this phenomenon in some detail, let us assume the process of line for-
mation in the solar atmosphere to be described by the Milne-Eddington model. The
Stokes profiles of the emergent radiation are then given by the Unno-Rachkovsky
solution (Eqs. (9.110)). Once the atmospheric and line parameters, and the width
and position of the filter are specified, the magnetograph signal SV can be com-
puted from Eq. (11.1). Figure 11.1 shows the behavior of SV as a function of
B for a collection of parameters’ values simulating the λ 6302.5 FeI line, namely
κL = 6.75, ∆λD = 40 mÅ, a = 0.04, βµ = 5. The signal SV was computed by
taking for p(λ) a Gaussian function centered in the blue wing of the line at 60 mÅ
from line center and having a full width at half maximum of 125 mÅ (∆λF = 75 mÅ
in Eq. (11.5)).
Apart from saturation, the longitudinal magnetograph suffers from other causes
of inaccuracy that are due to the calibration procedure. In practice the calibra-
tion is performed by selecting a fixed intensity profile I(λ) to be substituted into
Eqs. (11.1) and (11.2), whereas such profile generally shows non-negligible varia-
tions over the solar disk. These may be due either to limb-darkening effects, or to
1
Owing to its high sensitivity to magnetic fields (see Tables 9.2 and 9.3), this line is widely
used for solar magnetometry, notwithstanding the presence of a blend due to a telluric O 2 line in
its red wing. The presence of the blend is often rather useful in observations because it provides
an absolute reference wavelength for measuring Doppler shifts.
SOLAR MAGNETOMETRY 629
Fig.11.1. The magnetograph signal SV as a function of the longitudinal component of the mag-
netic field B for a specific line – a Zeeman triplet with ḡ = 2.5 – formed in a Milne-Eddington
atmosphere (see text for details). The curves are labelled by the value of the inclination angle θ
defined in Fig.9.1, and are drawn up to a field intensity of 3500 G. The dotted line represents the
linear relation implicit in the magnetograph calibration.
velocity fields, or to differences – with respect to the average sun – of the thermo-
dynamic parameters characterizing the various structures of the solar atmosphere
(granules, intergranular lanes, faculae, pores, sunspots’ umbrae and penumbrae,
etc.).
The saturation effect and the inaccuracies just mentioned are important lim-
itations which prevent one from regarding the longitudinal magnetograph as a
fully quantitative instrument for the measurement of solar magnetic fields. Yet its
importance in solar research can hardly be overestimated. It is just through mag-
netographs that the general properties of solar magnetism have been determined
(large-scale topology, fine structure, cyclic variations, etc.) and that several cor-
relations between magnetic fields and other indicators of solar activity have been
established. It is also important to mention that the results obtained by magne-
tographs are currently used for the reconstruction of coronal magnetic fields by
means of suitable numerical techniques.
11.2. The Vector Magnetograph
The vector magnetograph is a natural generalization of the longitudinal magneto-
graph. Besides the circular polarization signal SV of Eq. (11.1), two more signals
related to linear polarization are recorded
630 CHAPTER 11
Q(λ) p (λ) dλ U (λ) p (λ) dλ
SQ = , SU = , (11.6)
I(λ) p (λ) dλ I(λ) p (λ) dλ
where p (λ) is a profile that may differ from p(λ). The three signals SQ , SU , SV may
be recorded either simultaneously or sequentially. The first alternative implies the
use of a sophisticated polarimeter, capable of measuring the four Stokes parameters
at the same time. In this case, the spectral analysis is the same for the various
Stokes parameters and p (λ) is identical with p(λ). In the second alternative it
is convenient to use different profiles in order to maximize the signals. Since the
extrema of Q and U do not occur at the same wavelength positions as those of V ,
one can adapt the spectral analysis to the observed Stokes parameter by tuning
the filter (in filter-type magnetographs) or by moving the slit (in spectrograph-type
magnetographs).
Similarly to the preceding section, one can easily relate the polarization signals
SQ , SU to the modulus of the transverse component of the magnetic field and to
its orientation in the plane perpendicular to the line of sight provided a number
of conditions – basically the weak field approximation – are met. We found in
Sect. 9.6 that, under this approximation, the U Stokes parameter defined in the
preferred frame is given by Ũ(λ) = 0, which implies
Ũ (λ) p (λ) dλ
SŨ = =0, (11.7)
I(λ) p (λ) dλ
whereas for the Q Stokes parameter, defined in the same frame, we obtained various
expressions (see Eqs. (9.83), (9.84) and (9.85)) each valid in different wavelength
intervals and under different assumptions on the depth dependence of the physical
parameters (see Table 9.1).1 For our purposes it is convenient to use Eq. (9.84)
1 η ∂I(λ)
Q̃(λ) = − ∆λ2B Ḡ sin2 θ , (11.8)
4 η ∂λ
which is valid at any λ, even if it implies rather restrictive assumptions on the
process of line formation. The meaning of the symbols ∆λB , θ, and I(λ) has
already been recalled after Eq. (11.2). Ḡ is defined in Eq. (9.76), while η and η
are defined in Eqs. (9.65) and (9.70).
Since Q̃(λ) is proportional to the product B 2 sin2 θ, and I(λ) is unaffected by the
magnetic field to lowest order, under the weak field approximation we have
Q̃(λ) p (λ) dλ
SQ̃ = = C⊥ B⊥ 2
, (11.9)
I(λ) p (λ) dλ
1
A common assumption is the constancy with depth of the azimuth of the magnetic field
vector, which enables a preferred frame to be defined.
SOLAR MAGNETOMETRY 631
where B⊥ 2
= B 2 sin2 θ is the square modulus of the transverse component of the
magnetic field vector and C⊥ is the second calibration constant of the vector mag-
netograph. To derive an expression for C⊥ , we follow a procedure quite similar
to that followed for C in the previous section. Using Eqs. (11.4) and (11.5) for
the line and filter profile, respectively, and assuming that the profile of the line
absorption coefficient for zero magnetic field, η, has also a Gaussian shape
λ−λ 2
0
1 −
η(λ) = √ e ∆λD
,
π
one finds, by evaluating elementary integrals
1
SQ̃ = − ∆λ2B Ḡ sin2 θ Y ,
4
where
(λF −λ0 )2
− ∆λ2p ∆λ2F (λ −λ )2 ∆λ4
∆λ2p + ∆λ2F
2 dc 1 e 1− (∆λ2p + ∆λ2F ) ∆λ2D
− 2 (∆λ2F+ ∆λ
0 p
2 )2 ∆λ2
∆λp ∆λ2p + ∆λ2F p F D
Y= .
(λF −λ0 )2
−
∆λp ∆λ2p + ∆λ2F
1 − dc e
∆λ2p + ∆λ2F
From Eqs. (11.9) and (3.14), the second calibration constant of the vector magne-
tograph can thus be written in the form
C⊥ = −5.45 × 10−20 λ40 Ḡ Y , (11.10)
−2
where C⊥ is in G−2 , λ0 in Å and Y in mÅ .
To give an example, let us refer to the line FeI λ 6302.5, already considered in
the former section, with ∆λp = 60 mÅ, dc = 0.66, ∆λD = 40 mÅ, and ∆λF =
75 mÅ. It can easily be seen that the quantity Y is negative for all values of
(λF − λ0 ), and that its extrema are located at (λF − λ0 ) = ± 71.3 mÅ, where
−2
Y = −2.32 × 10−4 mÅ .This would be the optimum filter position for measuring
weak transverse fields. However, in actual measurements, it is convenient to shift
the filter position to larger distances from line center in order to avoid, as much as
possible, the contamination due to magneto-optical effects (that will be analyzed
below),1 whose importance strongly decreases in the line wings (see Sect. 9.22 and,
in particular, Fig. 9.17). In the present example, a good compromise is to position
the filter at (λF − λ0 ) = ± 120 mÅ (three Doppler widths), where the quantity Y
−2
is −1.64 × 10−4 mÅ , which corresponds to a reduction of sensitivity of only 30%
1
Note that Eq. (11.8), which is obtained from an expansion of the radiative transfer equations
to the second order in ∆λB /∆λD , does not contain terms related to magneto-optical effects. Such
terms appear in the expansion from the third order on (see Sect. 9.6).
632 CHAPTER 11
relative to its maximum value. Since for the FeI λ 6302.5 line is Ḡ = 6.25 (see
Table 9.3), we obtain from Eq. (11.10)
C⊥ = 8.8 × 10−8 G−2 .
The calibration procedure described so far is obviously incomplete, as Eqs. (11.7)
and (11.9) are valid in the preferred reference frame which in real observations is
a priori unknown. The quantities actually available to the observer are the signals
SQ and SU of Eq. (11.6), where Q(λ) and U (λ) refer to the observer’s reference
frame. This frame is determined by the experimental set-up of the instrument,
and defines a physical direction on the solar disk which may be used as the zero
for reckoning the azimuth χ of the magnetic field vector (see Fig. 9.1). The value
of χ – undetermined up to multiples of π because of an intrinsic symmetry of the
process of LTE line formation in a magnetic field, see Sect. 9.5 – can be recovered
via the following line of reasoning. Performing a rotation through an angle α = χ
of the reference direction as in Fig. 1.10, the observer’s frame is changed into the
preferred frame. From Eqs. (1.45) one has
SQ̃ = cos 2χ SQ + sin 2χ SU
SŨ = − sin 2χ SQ + cos 2χ SU . (11.11)
On the other hand, Eq. (11.7) yields SŨ = 0. Assuming SQ̃ > 0,1 the angle χ,
defined in the interval (−π/2, π/2), can be found from the following expressions
(cf. Eqs. (1.8)): for SQ = 0
1 S
χ= arctan U + χ0 (11.12a)
2 SQ
where ⎧
⎪
⎨0 if SQ > 0
π
χ0 = if SQ < 0 and SU > 0 (11.12b)
⎪
⎩−π
2
2 if SQ < 0 and SU < 0 ,
and for SQ = 0
1
χ= 4π if SU > 0
(11.12c)
− 14 π if SU < 0 .
Finally, since from Eqs. (11.11) we have SQ̃ = 2 + S 2 , Eq. (11.9) can be
SQ U
rewritten in the form
SL ≡ SQ 2 + S2 = C B2 .
U ⊥ ⊥ (11.13)
1
This is the case for the example just mentioned of the FeI λ 6302.5 line, where the calibration
constant C⊥ was found to be positive. In the (less frequent) opposite case, all the > signs in
Eqs. (11.12) should be changed into < signs, and vice versa.
SOLAR MAGNETOMETRY 633
Fig.11.2. The magnetograph signal SL as a function of the transverse component of the magnetic
field B⊥ for the same line and atmospheric parameters as in Fig.11.1. The curves are labelled
by the value of the inclination angle θ defined in Fig.9.1, and are drawn up to a field intensity of
3500 G. The dotted line represents the quadratic relation implicit in the magnetograph calibration.
The expressions just derived allow one to recover the direction and the modulus
of the transverse component of the magnetic field vector from the linear polariza-
tion signals SQ and SU . Such expressions are however valid only in the limit of
weak magnetic fields. When the field intensity increases, saturation effects come
into play, similarly to the case of the longitudinal magnetograph discussed in the
previous section.
For linear polarization there are two distinct effects due to saturation: on the
one hand, the signal SL turns out to be much smaller than what predicted by the
quadratic relationship of Eq. (11.13); on the other hand, the azimuth χ is found to
be affected by an ‘error’, ∆χ, due to magneto-optical effects.
To give a quantitative idea of these two effects, we assume – as in the former
section – the process of line formation in the solar atmosphere to be described by
the Milne-Eddington model. Once the atmospheric and line parameters, and the
width and position of the filter are specified, the magnetograph signals SQ and SU
can be computed from Eqs. (11.6) using the Unno-Rachkovsky expressions (9.110)
for I(λ), Q(λ) and U (λ). The behavior of SL and ∆χ against B⊥ for a set of
parameters (the same as in Fig. 11.1) simulating the λ 6302.5 FeI line is shown in
Figs. 11.2 and 11.3, respectively. The profile p (λ) in Eqs. (11.6) was assumed to be
a Gaussian function centered at 120 mÅ from line center and having ∆λF = 75 mÅ.
The sign convention for ∆χ in Fig. 11.3 is such that it has to be added to the χ
value of Eqs. (11.12) in order to retrieve the true azimuth of the magnetic field
vector.
Apart from saturation effects, the vector magnetograph is subjected to the same
634 CHAPTER 11
Fig.11.3. The ‘error’ angle ∆χ due to magneto-optical effects is plotted against B⊥ for the same
line and atmospheric parameters as in Figs.11.1 and 11.2 and the same filter profile as in Fig.11.2.
The curves are labelled by the values of θ. For the sign convention of ∆χ refer to the text.
limitations pointed out for the longitudinal magnetograph at the end of Sect. 11.1.
Notwithstanding such limitations, this instrument remains one of the main tools of
solar physics. A particularly long lasting and successful example is the tower vector
magnetograph of the University of Alabama in Huntsville and of the Marshall Space
Flight Center that has been operating since the mid-1970s (Hagyard et al., 1982)
and that has provided important results on the shear of magnetic fields in flare sites.
It has to be remarked that in the early period of operation of this instrument, the
error ∆χ introduced by saturation and magneto-optical effects was not taken into
account in the data-reduction analysis, which led Hagyard et al. (1977) to interpret
their observations of isolated sunspots in terms of a twisted magnetic field. It
was shown later (Landi Degl’Innocenti, 1979a) that such observations were fully
consistent with a pure radial magnetic configuration once the ∆χ correction was
properly included.
11.3. The Unno-fit Technique
To overcome the limitations of magnetographs, it was early realized that the mea-
surement of solar magnetic fields could be substantially improved by recording the
full profiles of the four Stokes parameters across a wavelength interval encompass-
ing one or more suitable spectral lines. An instrument capable of such recordings
is currently called a Stokesmeter , though the name of spectropolarimeter is used
as well. The data obtained with a Stokesmeter contain a wealth of information,
whereas a typical vector magnetograph ‘compresses’ all this information into just
SOLAR MAGNETOMETRY 635
three quantities (the signals SV , SQ , and SU defined in Eqs. (11.1) and (11.6)).
Hence the obvious superiority of Stokesmeters over magnetographs.1
Several techniques have been developed in the past for recovering the magnetic
field vector from the Stokes parameters’ profiles recorded, as a function of wave-
length, by a Stokesmeter. In this section we describe in some detail one of such
techniques, the so-called Unno-fit technique. Pioneered by Harvey et al. (1972) and
Auer et al. (1977b), and improved by Landolfi and Landi Degl’Innocenti (1982)
and Landolfi et al. (1984) to allow for magneto-optical and damping effects, this
technique has been brought to a high level of sophistication by the group of the
High Altitude Observatory working on the Advanced Stokes Polarimeter data (see
for instance Skumanich and Lites, 1987; Lites and Skumanich, 1990).
The technique consists in comparing the observed Stokes profiles with the Unno-
Rachkovsky solution to the radiative transfer equation for polarized radiation (see
Sect. 9.8) and in varying the set of parameters on which such solution depends
until a best fit is obtained. In practice, a kind of ‘merit function’ is constructed,
typically of the form
3
N
(obs) (thr) 2
M {ζi } = Wjα Ijα − Ijα {ζi } , (11.14)
j=0 α=1
(obs) (thr)
where Ijα is the j-th Stokes parameter measured at wavelength λα ,2 Ijα is
the corresponding theoretical value dependent on a set {ζi } of physical parameters
(thr)
(i = 1, 2, . . . , p), and Wjα is a matrix of weights. The quantities Ijα can be
computed from Eq. (9.109) by taking into account the spectral smearing profile of
the instrument, which depends on the resolving power of the spectrograph, on the
geometrical width of the entrance slit, on the pixel width and, in many cases, on
the data reduction procedure. If P (∆λ) is such a profile (normalized to unity in
wavelength), one has from Eq. (9.109)
!
−1
Ijα = B0 δj0 + β
(thr)
(C )j0 λ=λ −∆λ P (∆λ) d(∆λ) , (11.15)
α
where C is the matrix defined in Eq. (9.91) and β = βµ is the slope of the Planck
function along the propagation direction in the solar atmosphere.
The dependence of Ijα on the two physical parameters B0 and β is explicitly
(thr)
expressed in Eq. (11.15). The dependence on the other seven physical parameters
(κL , B, θ, χ, ∆λD , Γ , wA – see Sect. 9.5 for their definition) is implicitly con-
tained in the matrix C. The merit function M ({ζi }) thus depends, in total, on
1
It should not be forgotten that magnetographs may be preferred to Stokesmeters for scientific
purposes which do not require high precision magnetometry. Just because they involve a much
smaller number of observational data, magnetographs are generally faster and more agile than
Stokesmeters. For this reason they are often preferred in space research.
2
We assume that the Stokes parameters are sampled at different wavelength points (pixels)
along the dispersion axis, as typical of modern observations. N represents the number of pixels
covered by the observation.
636 CHAPTER 11
nine physical parameters which, for later reference, can be conveniently divided
into magnetic parameters (B, θ, χ), velocity parameter (wA ), and thermodynamic
parameters (B0 , β , κL , ∆λD , Γ ).1 As to the weights Wjα , they can be used for
two different purposes. On one side they allow one to remove from the fit – by
setting to zero the corresponding weight – certain wavelength intervals where the
observed Stokes profiles are contaminated either by experimental errors or by the
presence of a blending line (like, for instance, the telluric line in the red wing of
FeI λ 6302.5, see footnote on p. 628). On the other side, they can be used to in-
crease the importance of the fit to one (or more) Stokes parameter(s) relative to
the remaining ones. In many cases, for instance, the intensity profile is one to two
orders of magnitude larger (in absolute value) than the other Stokes profiles. A fit
with equal weights will eventually favor just that profile which is less sensitive to
the magnetic parameters. Since in general the intensity profile is also affected by
other sources of contamination – like stray light in the spectrograph – it turns out
that a good choice is to assign weight unity to the Stokes profiles Q, U , and V and
a weight ranging from 0.01 to 0.1 to the I profile.2
In order to obtain the best fit to the observed data, one needs to find the absolute
minimum of the merit function in the hyperspace of the parameters ζi . This is not
a trivial problem because the dependence of the merit function on the parameters
is non-linear and because the merit function has, in general, several secondary
minima. An appropriate mathematical procedure that can be followed in such cases
is due to Marquardt (1963) and is fully described in Bevington (1969). Starting
from some ‘guess values’ of the parameters ζi , the Marquardt algorithm looks for
the minima of the merit function by following a path in the parameters’ hyperspace.
The path consists in a succession of steps whose length and direction is determined
by evaluating the quantities βk (k = 1, . . . , p) and αkm (k, m = 1, . . . , p) defined
by
∂M {ζi } ∂ 2 M {ζi } ∂βk ∂βm
βk = , αkm = = = .
∂ζk ∂ζk ∂ζm ∂ζm ∂ζk
From Eq. (11.14) one has
3
N
(obs) (thr)
(thr)
∂Ijα
βk = −2 Wjα Ijα − Ijα
j=0 α=1
∂ζk
3
N (thr) (thr) 2 (thr) !
∂Ijα ∂Ijα (obs) (thr) ∂ Ijα
αkm = 2 Wjα − Ijα − Ijα ,
j=0 α=1
∂ζk ∂ζm ∂ζk ∂ζm
1
The number of parameters can be reduced to 8 by defining the merit function in terms of the
Stokes parameters normalized as in Eqs. (9.112). By so doing, the dependence on B0 disappears
and β is replaced by β /(1 + β ).
2
This statement is based on the experience gathered mostly with the data of the Advanced
Stokes Polarimeter of the High Altitude Observatory. For an ideal Stokesmeter, with observed
Stokes profiles affected by Gaussian errors with variance σjα , the weights should indeed be given
by Wjα = 1/σjα . In that case the merit function of Eq. (11.14) would reduce to a standard
chi-square function (see e.g. Bevington, 1969).
SOLAR MAGNETOMETRY 637
the last expression being usually approximated by neglecting the second term in
braces which gets much smaller that the first when approaching the minimum
of the merit function. The Marquardt algorithm thus requires the evaluation of
(thr)
derivatives of the form ∂Ijα /∂ζk . As apparent from Eq. (11.15), the derivatives
with respect to B0 and β are trivial. The derivatives with respect to the other
parameters (implicitly contained in the matrix C) can be obtained with the help
of Eq. (9.204). One has
,
∂C −1 -
(thr)
∂Ijα
= −B0 β C −1 C P (∆λ) d(∆λ) ,
∂ζk ∂ζk j0 λ=λα−∆λ
and the derivatives of the matrix C with respect to the various parameters are
easily evaluated using Eqs. (9.91), (9.39), (9.32), (9.29), (9.27), (9.26) and (5.58).
As stated above, the Marquardt algorithm eventually finds a minimum of the
merit function in the hyperspace of the parameters, but there is no guarantee that
this is the absolute minimum. To overcome such problem, one is forced to iterate
the procedure starting from a different set of guess values of the parameters, and
only after a sufficient number of iterations one can be reasonably confident that
the absolute minimum has indeed been reached. To make the search easier, it is
advisable to start the Marquardt algorithm with ‘plausible’ initial values (as close
as possible to those corresponding to the absolute minimum).1 By so doing, the
possibility of finding a secondary minimum is largely reduced. In the case of raster-
type observations, the best fit values of the parameters found for a given point of
the solar surface can serve as initial values for the surrounding points.
When applied to data showing a reasonable
amount of polarization (roughly
speaking, when the maximum value of Q2 + U 2 + V 2 / I across the line profile
is larger than 1%), the Unno-fit technique provides reliable values of the magnetic
parameters. Such values are practically independent of the weights Wjα introduced
in Eq. (11.14). The same is true for the velocity parameter, whereas the thermo-
dynamic parameters may show larger variations when varying the weights and,
in some cases, even somewhat unrealistic values may be found for them. This is
due to the fact that there is a strong ‘trade-off’ in the Unno-Rachkovsky solutions
among the parameters B0 , β , κL , ∆λD , and Γ , in the sense that a given set
of Stokes profiles can be reproduced, approximately with the same accuracy, by
several different combinations of these parameters. However, this drawback of the
Unno-fit technique does not represent a serious limitation for the measurement of
the magnetic field – which is the main object of the method.
Two examples of the application of the Unno-fit technique are shown in Figs. 11.4
and 11.5. The data were collected at the Advanced Stokes Polarimeter on June
19, 1992 in the penumbra of the large round sunspot of Active Region NOAA 7201
(see Skumanich et al., 1997, for a description of the observations). One can notice
1
The thermodynamic parameters of a line are known fairly well and do not show very
large variations over the solar surface. The velocity parameter can be simply set to zero. The
real problem is with the magnetic parameters. Suitable first-guess values can be obtained, for
instance, from a preliminary data analysis based on the weak field approximation.
638 CHAPTER 11
Fig.11.4. Unno-fit to ASP data for the FeI λ 6302.5 line observed in the penumbra of a sunspot.
The Stokes parameters are normalized to the continuum intensity and are plotted against pixel
number. The dispersion is 12.74 mÅ/pixel. The data points drawn as open circles are due to a
telluric line and were excluded from the fit. Note the reversal in the V profile due to magneto-
optical effects (see Sect. 9.22). The magnetic parameters are B = 2130 G, θ = 128◦ , χ = 22◦ (see
Fig.9.1 for the definitions of θ and χ ).
the quite remarkable fit to the observed Q, U , and V profiles in Fig. 11.4 (the fit
to the intensity profile is worse because, in these illustrative examples, we have not
taken into account the spectral smearing profile of the instrument, P (∆λ) ). On
the contrary, the observed polarization profiles shown in Fig. 11.5 are reproduced
only marginally by the fit. This is probably due to the presence of strong velocity
gradients (Evershed flow) which introduce considerable asymmetries in the profiles
(note, for instance, the presence of three distinct lobes in the V profile). Obviously,
such asymmetries cannot be reproduced by the fit, based on a model which does
not allow for any depth dependence of the physical parameters.
SOLAR MAGNETOMETRY 639
Fig.11.5. Same as Fig.11.4 for a different region of the sunspot. The magnetic parameters are
B = 1070 G, θ = 100◦ , χ = 31◦ . Note that the quality of the fit is much worse than in Fig.11.4.
This is probably due to the Evershed effect (see text).
11.4. The Bisector and the Center-of-Gravity Techniques
The bisector technique and the center-of-gravity technique were proposed in the
late 1960s to overcome the saturation problems inherent in the magnetographic
measurements of the longitudinal component of the magnetic field vector. The two
techniques were developed almost simultaneously by Rayrole (1967) and Semel
(1967), respectively.
Both techniques are based on the observation of the right and left (positive and
negative) circular polarization in a magnetically sensitive spectral line. Such quan-
tities are connected with the Stokes parameters I and V by the simple relation
1
I± (λ) = I(λ) ± V (λ) . (11.16)
2
In the bisector technique the two profiles I+ (λ) and I− (λ) are observed simultane-
(b) (b)
ously, and their ‘central’ wavelengths or bisectors, λ+ and λ− respectively, are
640 CHAPTER 11
recovered via a procedure described in more detail below. The longitudinal com-
ponent of the magnetic field is then obtained from the equation (see Eq. (3.13) for
the definition of ∆λB )
(b) (b) ḡ λ20 e0
λ+ − λ− = 2 ḡ ∆λB cos θ = B cos θ , (11.17)
2π mc2
where ḡ is the effective Landé factor of the line. Numerically one gets
1.071 × 109 (b) (b)
B = B cos θ = 2 λ+ − λ− , (11.18)
ḡ λ0
(b) (b)
where λ0 is in Å, (λ+ − λ− ) in mÅ, and B in G.
The center-of-gravity technique is based, rather than on the profiles I± (λ), on
the direct measurement of their first-order moments defined by1
1
I − I± (λ) λ dλ
2 c
λ± =
(g)
, (11.19)
1
Ic − I± (λ) dλ
2
where Ic is the intensity of the continuum (assumed unpolarized), and where the
integrals are extended to the full line profile. The longitudinal component of the
magnetic field vector is obtained from equations strictly similar to Eqs. (11.17)-
(11.18)
(g) (g)
λ+ − λ− = 2 ḡ ∆λB cos θ , (11.20)
or, numerically
1.071 × 109 (g) (g)
B = B cos θ = 2 λ+ − λ− . (11.21)
ḡ λ0
Equations (11.17) and (11.20) are indeed exact for any Zeeman triplet (normal
or anomalous) formed in an arbitrary atmosphere with a constant magnetic field
parallel or antiparallel to the line of sight (θ = 0◦ or 180◦ referring to the geometry
of Fig. 9.1). This particular case has been analyzed in Sect. 9.10. From Eqs. (9.126)
we have, for θ = 0◦
1 ¯
I± (λ) = I(λ ∓ g ∆λB ) ,
2
¯
where I(λ) is the intensity that would be observed if the magnetic field were
switched off leaving all the other physical parameters (including the source func-
tions) unchanged. This equation shows that the profiles I+ (λ) and I− (λ) have
exactly the same shape, and are just shifted in wavelength by 2g∆λB . It follows
(b) (b)
that, irrespective of the procedure followed to deduce λ+ and λ− from the profiles
1 (g)
An instrumental set-up for measuring the quantities λ± , sometimes called a lambdameter,
is described in Semel (1967).
SOLAR MAGNETOMETRY 641
(b) (b) (g) (g)
λ+ − λ− = λ+ − λ− = 2 g ∆λB . (11.22)
This equation coincides with Eqs. (11.17) and (11.20) evaluated for θ = 0◦ and for
a Zeeman triplet (ḡ = g). The case θ = 180◦ can be treated in a strictly similar
way and shows again the correctness of Eqs. (11.17) and (11.20). Moreover, the
same line of reasoning leads to the conclusion that the restriction about the Zeeman
pattern can be partly released. Still under the assumption of a constant magnetic
field with θ = 0◦ or 180◦ , the two equations are also valid for Zeeman patterns
classified as Type III in Sect. 3.3 (σ components symmetrical about their center of
gravity).
On the other hand, Eqs. (11.17) and (11.20) are correct for any value of θ and any
Zeeman pattern provided the magnetic field is weak and its longitudinal component
is independent of depth. In this case we have from Eq. (9.80) – see also Table 9.1
∂I
V (λ) = − ḡ ∆λB cos θ ,
∂λ
so that Eq. (11.16) becomes
1 ∂I 1
I± (λ) = I(λ) ∓ ḡ ∆λB cos θ ≈ I(λ ∓ ḡ ∆λB cos θ) .
2 ∂λ 2
This expression shows that the right and left circular polarization profiles are again
the same except for an overall shift, hence1
(b) (b) (g) (g)
λ+ − λ− = λ+ − λ− = 2 ḡ ∆λB cos θ ,
which coincides with Eqs. (11.17) and (11.20).
Apart from the special cases analyzed above, it is expected that Eqs. (11.17)
and (11.20) are only approximately valid. However, to put this statement on a
more quantitative basis, it is necessary to specify the procedure for deriving the
(b)
quantities λ± from the observed right and left circular polarization profiles (this
(g)
is unnecessary for the quantities λ± because Eq. (11.19) is itself an operational
definition). As suggested by Rayrole (1967), the profiles I± (λ) have first to be
convolved with a rectangle function having width ∆λR . This yields the new profiles
I± (λ) defined by
∆λR /2
1
I± (λ) = I± (λ − λ ) dλ .
∆λR
−∆λR /2
Such convolution has the double purpose of reducing the noise contained in the
data and of transforming the original profiles I± (λ), often characterized by two or
1
Note that this result is independent of the actual shape of the profile and is also independent
(b) (b)
of the procedure followed to deduce λ+ and λ− from the profile itself.
642 CHAPTER 11
Fig.11.6. The relative error introduced in the measurement of B by the bisector technique is
plotted as a function of |B | for different values of the inclination angle θ, defined as in Fig.9.1.
The curves are drawn up to a field strength of 3500 G. The figure refers to the line FeI λ 5250.21
formed in a Milne-Eddington atmosphere (see text for further details).
three minima,1 into profiles having just one minimum. Referring to one of the two
profiles, for instance I+ (λ), one then looks for the two solutions, λ1 and λ2 , of the
equation
I+ (λ) = Im + r Iw − Im ,
where Im is the value of I+ corresponding to the minimum of the profile, Iw is
its value in the far wings, and r is a parameter ranging from 0 to 1. From λ1
(b) (b)
and λ2 one finally obtains the bisector λ+ as (λ1 + λ2 )/2. The bisector λ− is
obtained by applying the same procedure to the profile I− (λ). Obviously, the two
parameters ∆λR and r are in principle arbitrary, and the value of B derived from
Eq. (11.18) depends on them. This dependence is weak if the magnetic field is
only mildly inclined with respect to the line of sight but can drastically increase
for large inclinations.
We have checked the reliability of Eqs. (11.18) and (11.21) by selecting a specific
spectral line and by varying the modulus and inclination of the magnetic field. The
line is FeI λ 5250.21, a Zeeman triplet having ḡ = 3 that is currently used in solar
magnetometry for its high sensitivity (see Tables 9.2 and 9.3). In our simulation
the line is formed in a Milne-Eddington model atmosphere with a suitable choice
of atmospheric and line parameters (κL = 6.1, ∆λD = 32 mÅ, a = 0.04, βµ = 12).
(b)
Figure 11.6 shows the relative error, δB , introduced by the bisector technique
1
When the magnetic field is sufficiently large and inclined with respect to the line of sight,
the profiles I± (λ) show three minima, corresponding to the Zeeman components σb , π, and σr .
SOLAR MAGNETOMETRY 643
Fig.11.7. Same as Fig.11.6 for the relative error introduced in the measurement of B by the
center-of-gravity technique.
when the two parameters ∆λR and r are set to 140 mÅ and 0.65, respectively. The
(b)
error δB is defined by
(b)
(b)
B −B
δB = ,
B
(b)
where B is evaluated via Eq. (11.18) and B is the value of the longitudinal
magnetic field employed in the simulation. The figure shows that the relative error
is always less than about 10% for magnetic field’s inclinations smaller than 60◦ .
(g)
Similarly, Fig. 11.7 gives the relative error δB introduced by the center-of-gravity
technique
(g)
(g)
B −B
δB = ,
B
(g)
where B is obtained from Eq. (11.21). In this case the relative error is always
less than about 10%, with a definite tendency to underestimate the actual value of
B .
The bisector technique and the center-of-gravity technique provide only the lon-
gitudinal component of the magnetic field. They have to be complemented with
other techniques in order to obtain the full measurement of the magnetic field
vector (see e.g. Rayrole, 1967, for the possibility of recovering the transverse com-
ponent from the maximum amplitude of the V Stokes parameter). Moreover, both
techniques are limited by systematic errors which, however, can be fairly well es-
timated by means of correction tables adapted to the specific observation, and
constructed via procedures similar to that followed above for obtaining Figs. 11.6
644 CHAPTER 11
and 11.7. The main advantage of these techniques is the possibility of obtaining a
fairly accurate estimate of the longitudinal component of the magnetic field (not
affected by saturation effects) either by a simple procedure of data handling on the
profiles I± (λ) (bisector technique) or by a direct measurement of the quantities
(g)
λ± (center-of-gravity technique).
11.5. Unresolved Fields
One of the major problems of solar magnetometry is the fact that, in many cases,
the magnetic field is not uniform across the resolution element of the instrument
employed in the observations. A second problem, intimately related to the first,
is the fact that the local values of the thermodynamic parameters of the solar
atmosphere, being strongly correlated with the amplitude and direction of the
magnetic field vector, may also show large spatial variations over the same area.
In this situation, any method or technique that can be envisaged to deduce the
magnetic field vector – or one of its components – from observations is obviously
bound to yield a sort of ill-defined ‘average’ value.
In this section we will analyze in some detail this particular aspect of solar mag-
netometry, by illustrating how the different techniques that have been described in
the previous sections are affected by the lack of resolution in the observations.
11.5.a Longitudinal and Vector Magnetographs
Denoting by I(λ; P) the intensity profile emerging from an element of the solar
surface dS centered at the point P, and introducing similar notations for the
other Stokes parameters, the signal of a longitudinal magnetograph is given by
(cf. Eq. (11.1))
dS V (λ; P) p(λ) dλ
SV Σ = Σ , (11.23)
dS I(λ; P) p(λ) dλ
Σ
where the integrals in dS are extended to the solar region Σ covered by the in-
strument. A more significant expression can be obtained by introducing the local
value of the magnetograph signal, SV (P), defined by
V (λ; P) p(λ) dλ
SV (P) = ,
I(λ; P) p(λ) dλ
and the local value of the filtered intensity
I(P) = I(λ; P) p(λ) dλ .
SOLAR MAGNETOMETRY 645
With these notations Eq. (11.23) can be rewritten in the form
SV (P) I(P) dS
Σ
SV Σ = ,
I(P) dS
Σ
which shows that the ‘unresolved’ magnetograph signal (SV )Σ is an average over
the observed area of the ‘resolved’ magnetograph signal SV (P) weighted by the
filtered intensity.
As shown in Sect. 11.1, the magnetograph signal is proportional to the longitu-
dinal component of the magnetic field vector under the weak field approximation.
Therefore, if the magnetic field is everywhere weak across the surface Σ, one has,
recalling Eq. (11.3)
C (P) B (P) I(P) dS
SV Σ = Σ ,
I(P) dS
Σ
where C (P) is the calibration constant corresponding to the intensity profile
emerging from point P. On the other hand, the magnetograph is calibrated with
a fixed intensity profile, usually the profile corresponding to the quiet (or non-
magnetic) average photosphere (see Sect. 11.1). Let us denote by (C )q the actual
calibration constant. When the magnetograph signal is interpreted in terms of a
linear relation of the form of Eq. (11.3), what is in fact obtained is an apparent
longitudinal component given by
C (P) B (P) I(P) dS
(B )app = Σ . (11.24)
(C )q I(P) dS
Σ
Two special cases of this equation deserve some attention. The first is the case
where the intensity profile I(λ; P) is the same as the profile corresponding to the
quiet photosphere. This implies that the quantity I(P) is independent of P and
that C (P) = (C )q , thus Eq. (11.24) reduces to
1
(B )app = B = B (P) dS , (11.25)
A Σ
where A is the area of the surface Σ. This result is at the basis of the state-
ment – often quoted in solar magnetometry – according to which the longitudinal
magnetograph measures the flux of the magnetic field vector across the observed
area. Apart from an obvious dimensional factor (needed to transform the average
longitudinal field into the magnetic flux), and from an equally obvious geometrical
646 CHAPTER 11
distinction (the longitudinal component coincides with the vertical component only
at disk center), the above derivation shows that this statement is only valid under
a heavy set of assumptions (field everywhere weak, intensity profile independent of
point P and coincident with the average photospheric profile).
The second special case of Eq. (11.24) occurs when one assumes that a fraction
f of the observed area is covered by a uniform magnetic field having longitudinal
component (B )m , whereas the remaining fraction (1 − f ) is just the quiet (non-
magnetic) solar atmosphere. The quantity f is often referred to as the filling
factor . The intensity profile is assumed to be constant throughout the ‘magnetic
fraction’ of the observed area and may differ from the profile corresponding to the
quiet atmosphere. This implies different values for the filtered intensities of the
‘magnetic’ and ‘quiet’ fractions (Im and Iq , respectively) as well as different values
for the calibration constants (C )m and (C )q . Under these assumptions we obtain
from Eq. (11.24)
f (C )m Im
(B )app = (B )m . (11.26)
(C )q (1 − f ) Iq + f Im
If, for instance, the intensity profile in the magnetic region is simply scaled by a
factor Θ (a sort of thermodynamic parameter) with respect to the profile corre-
sponding to the quiet atmosphere, one has
Im = Θ Iq , (C )m = (C )q , (11.27)
so that
fΘ
(B )app = (B )m . (11.28)
1−f +fΘ
Equations (11.26) and (11.28), though valid under a number of simplifying as-
sumptions, are able to account for a phenomenon known since a long time in solar
magnetometry, namely the differences that are commonly found when comparing
the values of B derived from longitudinal magnetographs operating simultane-
ously on different spectral lines (Harvey and Livingston, 1969). Such differences
cannot be interpreted as due only to magnetic saturation effects (see Sect. 11.1)
and imply that each spectral line has its own peculiar behavior in magnetic regions
(its own Θ value, for instance, in the simplified model leading to Eq. (11.28)). In
other words, since the thermodynamic structure of magnetic regions is different
from that of the quiet atmosphere, each line is characterized by a different amount
of weakening (or strengthening) which eventually affects the apparent value of B
measured by a longitudinal magnetograph.
In order to analyze the influence of unresolved fields on the results obtained by
vector magnetographs, Eq. (11.23) has to be complemented with the corresponding
equations for the Stokes parameters Q and U (see Eqs. (11.6))
dS Q(λ; P) p (λ) dλ
SQ Σ = Σ
dS I(λ; P) p (λ) dλ
Σ
SOLAR MAGNETOMETRY 647
dS U (λ; P) p (λ) dλ
SU Σ = Σ . (11.29)
dS I(λ; P) p (λ) dλ
Σ
Similarly to the case of the longitudinal magnetograph, these expressions can be
cast into a more significant form by introducing the local values of the linear po-
larization signals (defined in the preferred frame)
Q̃(λ; P) p (λ) dλ Ũ (λ; P) p (λ) dλ
SQ̃ (P) = , SŨ (P) = ,
I(λ; P) p (λ) dλ I(λ; P) p (λ) dλ
and the local value of the filtered intensity
I (P) = I(λ; P) p (λ) dλ .
Denoting by χ(P) the azimuth of the magnetic field at point P reckoned in the ob-
server’s frame, Eqs. (11.29) can be rewritten in the following form (cf. Eqs. (11.11))
SQ̃ (P) cos 2χ(P) − SŨ (P) sin 2χ(P) I (P) dS
SQ Σ = Σ
I (P) dS
Σ
SQ̃ (P) sin 2χ(P) + SŨ (P) cos 2χ(P) I (P) dS
SU Σ = Σ .
I (P) dS
Σ
As shown in Sect. 11.2, when the magnetic field is weak, the signal SŨ is zero
(Eq. (11.7)), whereas the signal SQ̃ is proportional to the square of the transverse
component of the magnetic field vector (Eq. (11.9)). Thus, if the magnetic field is
everywhere weak across the surface Σ, one has
2
C ⊥ (P) B⊥ (P) cos 2χ(P) I (P) dS
SQ Σ = Σ
I (P) dS
Σ
2
C ⊥ (P) B ⊥ (P) sin 2χ(P) I (P) dS
SU Σ = Σ , (11.30)
I (P) dS
Σ
where C⊥ (P) is the calibration constant corresponding to the intensity profile
emerging from point P. Denoting by (C⊥ )q the calibration constant correspond-
ing to the intensity profile of the quiet solar atmosphere, the signals of Eqs. (11.30)
648 CHAPTER 11
will then be interpreted in terms of ‘apparent’ quantities: the apparent azimuth
χapp implicitly defined by1
2
C⊥ (P) B⊥ (P) sin 2 χ(P) − χapp I (P) dS = 0 , (11.31)
Σ
and the apparent transverse component
A
B 2
B
B C (P) B⊥ (P) cos 2 χ(P) − χapp I (P) dS
B Σ ⊥
(B⊥ )app = B . (11.32)
C
(C⊥ )q I (P) dS
Σ
Similarly to the former discussion concerning the longitudinal magnetograph, we
consider two special cases. The first is the case where the intensity profile does
not depend on point P and can be considered the same as the profile of the quiet
atmosphere. Here the angle χapp is implicitly defined by
2
B⊥ (P) sin 2 χ(P) − χapp = 0 ,
where the symbol · · · means an average over the surface Σ as in Eq. (11.25), and
the apparent transverse component is given by
2
(B⊥ )app = B⊥ (P) cos 2 χ(P) − χapp .
The second is the case of the filling factor hypothesis. Here, denoting by the
index ‘m’ all quantities referring to the magnetic region, and by the index ‘q’ all
quantities referring to the quiet atmosphere, one has
χapp = χm
and
f (C⊥ )m Im
(B⊥ )app = (B⊥ )m , (11.33)
(C⊥ )q (1 − f ) Iq + f Im
or, introducing the scale factor Θ as in Eqs. (11.27)
fΘ
(B⊥ )app = (B⊥ )m . (11.34)
1−f +fΘ
It is also interesting to combine these last expressions with Eqs. (11.26) or (11.28)
in order to find the influence of the lack of resolution on the inclination angle θ of
1
Note that Eq. (11.31) leaves an ambiguity of 90 ◦ in the value of χapp . This ambiguity can
be removed by requiring that the argument of the square root in Eq. (11.32) be positive.
SOLAR MAGNETOMETRY 649
Fig.11.8. The ‘apparent’ inclination of the magnetic field vector is plotted as a function of the
true inclination angle of the field in an unresolved magnetic region. The parameter x is, as an
order of magnitude, equal to the filling factor f (see text).
the magnetic field vector with respect to the line of sight. The ‘apparent’ value of
the inclination angle, θapp , is connected with the value in the unresolved magnetic
region, θm , by the relation
(B⊥ )app tan θ
tan θapp = = √ m ,
(B )app x
where1
fΘ
x= .
1−f +fΘ
The relation between θapp and θm is shown in Fig. 11.8 for different values of the
parameter x. The figure clearly shows that when the filling factor f is very small
(hence x is also very small), the magnetic field recovered by a vector magnetograph
tends to appear as a highly inclined field (θ close to 90◦ ), no matter what the true
inclination of the field in the unresolved region is.
Up to now we have basically restricted our discussion to weak fields. When strong,
unresolved fields are present, things get more complicated because saturation effects
come into play and it is difficult, in general, to disentangle such effects from those
1
This expression for x is obtained by combining Eqs. (11.28) and (11.34). The combination
of Eqs. (11.26) and (11.33) leads to a more complicated expression for x which, however, is still
proportional to the filling factor f under the limit f → 0.
650 CHAPTER 11
due to the lack of resolution. Exceptions can however be found, as demonstrated by
a diagnostic technique which is just based on saturation phenomena and is known
under the name of line-ratio technique (Stenflo, 1973).
The line-ratio technique is based on the comparison of the (longitudinal) mag-
netograph signals observed simultaneously in two spectral lines having the same
behavior with respect to the thermodynamic parameters but different Landé fac-
tors. Denoting the two lines by the indices 1 and 2, respectively, and introducing
the filling factor hypothesis, the magnetograph signal in either line is given by
f Vm(i) (λ) pi (λ) dλ
SV i = (i = 1, 2) , (11.35)
(1 − f ) Iq(i) (λ) + f Im(i)
(λ) pi (λ) dλ
(i)
where Iq (λ) is the intensity of the solar radiation coming from the quiet atmo-
(i) (i)
sphere, Im (λ) and Vm (λ) are the intensity and circular polarization emerging from
the magnetic region, and p1 (λ), p2 (λ) are two identical profiles, the first centered
in a wing of line 1 and the second in the corresponding wing of line 2, at the same
distance from line center. If the two lines are identical from the point of view of
their thermodynamic behavior, this implies that
Iq(1) (λ) p1 (λ) dλ = Iq(2) (λ) p2 (λ) dλ , (11.36)
and if the magnetic field in the magnetic region is weak we also have
(1) (2)
Im (λ) p1 (λ) dλ = Im (λ) p2 (λ) dλ
and, from Eq. (11.2)
ḡ1
Vm(1) (λ) p1 (λ) dλ = Vm(2) (λ) p2 (λ) dλ ,
ḡ2
where ḡ1 and ḡ2 are the effective Landé factors of the two lines. Substituting these
relations into Eq. (11.35) and evaluating the ratio R of the two signals one has,
irrespective of the filling factor and of any possible weakening (or enhancement) of
the lines in the magnetic region
SV 1 ḡ
R= = 1 .
SV 2 ḡ2
If, on the contrary, the magnetic field falls outside the weak field regime, it can be
expected that the line having the larger Landé factor (line 1, say) saturates more
than the other, so that
ḡ
R< 1 .
ḡ2
SOLAR MAGNETOMETRY 651
Fig.11.9. A simulation of the ratio R of the circular polarization signals observed by a longitudinal
magnetograph in the two lines λ 5250.21 and λ 5247.05, plotted as a function of the magnetic field
intensity for different values of the inclination angle θ (full line: θ = 0◦ or 180◦ ; dotted line:
θ = 30◦ or 150◦ ; dashed line: θ = 60◦ or 120◦ ).
Since the ratio R is a function of the magnetic field vector that can be evaluated
through model calculations, it follows that from the observed value of R it is possi-
ble to obtain an order-of-magnitude estimate of the magnetic field intensity in the
unresolved structure.
By a fortunate coincidence, the solar spectrum does indeed present a pair of
lines particularly suitable for the application of the line-ratio technique. These are
the two FeI lines at 5250.21 Å (line 1) and 5247.05 Å (line 2). Both lines belong to
multiplet n. 1 of FeI 1 and happen to have practically the same value for the quantity
g̃ f (α J → αu Ju ), g̃ being the degeneracy of the lower level of the transition
and f (α J → αu Ju ) the oscillator strength defined in Eq. (9.20). Laboratory
measurements by Blackwell et al. (1979) give in fact
Log(g̃f ) line 1 = −4.938 , Log(g̃f ) line 2 = −4.946 ,
with a relative difference less than 2%. This property, together with the fact that
the lower levels of the two lines are very close in energy (the difference is 0.034 eV)
and with the further property of the closeness in the spectrum (which implies
the same Doppler broadening) assures that the two lines have, to a very good
approximation, the same absorption coefficient and hence the same behavior for
any stratification of the thermodynamic parameters with depth.
Figure 11.9 shows the behavior of the ratio R for the two lines mentioned as
a function of the magnetic field intensity, for different field inclinations. R was
1
Line 1 corresponds to the transition a5 D0 − z 7 Do1 . It is a normal Zeeman triplet having
ḡ = 3. Line 2 corresponds to the transition a5 D2 − z 7 Do3 . It is a Zeeman multiplet with ḡ = 2.
652 CHAPTER 11
computed under the limit of very small filling factor which yields, taking into
account Eqs. (11.35) and (11.36)
Vm(1) (λ) p1 (λ) dλ
R= .
Vm(2) (λ) p2 (λ) dλ
The V Stokes profiles were computed assuming the two lines to be formed in a
Milne-Eddington atmosphere with the same set of atmospheric and line parameters
(the same values as in Fig. 11.6 – which refers to the λ 5250.21 line – were used,
namely κL = 6.1, ∆λD = 32 mÅ, a = 0.04, βµ = 12). The profiles p1 (λ) and p2 (λ)
were supposed to have a Gaussian shape as in Eq. (11.5) with ∆λF = 75 mÅ, and
to be positioned at 60 mÅ from line center for both lines. As expected from our
previous discussion, the ratio R approaches asymptotically the value 1.5 (= ḡ1 /ḡ2 )
under the limit B → 0, and decreases monotonically with increasing B.
Through the line-ratio technique applied to this specific pair of lines it has been
possible to give a quantitative estimate of the dimensions and of the intrinsic field
intensity of magnetic regions observed at low spatial resolution by magnetographic
techniques (Howard and Stenflo, 1972; Frazier and Stenflo, 1972; Stenflo, 1973).
The magnetic field outside sunspots was found to be concentrated in small areas,
with typical dimensions of the order of 100-300 km, mostly located in the super-
granular network and having intrinsic magnetic field intensities in the kG range.
It should be remarked, however, that in low resolution observations weak fields
tend to be substantially underestimated, because they have a larger probability of
reconnecting at short distances, i.e., within the resolution element of the instru-
ment. This implies that the contribution of weak fields to both the observed signals
(SV )1 and (SV )2 , and hence to their ratio R, is likely to cancel out in many cases.
11.5.b Unno-fit Technique
The Unno-fit technique described in Sect. 11.3 can be suitably adapted to measure
the intensity and direction of the magnetic field vector in an unresolved magnetic
structure under the ‘filling factor’ approximation. To this aim, the theoretical
Stokes profiles entering the merit function of Eq. (11.14) are modified according to
the transformation
(thr) (thr)
Ijα → f Ijα + (1 − f ) δj0 (Iα )q ,
(thr)
where Ijα (in the right-hand side) is defined in Eq. (11.15) and where (Iα )q , the
intensity of the quiet atmosphere at wavelength point α, is assumed to be known
a priori.1 The filling factor f can now be considered as an additional parameter
that can be deduced, together with the other ones, from the observational data.
1
In raster-type observations, (Iα )q can be obtained by taking the average of the intensity
spectra over all the points of the raster (excluding sunspots and active regions, if necessary).
SOLAR MAGNETOMETRY 653
This procedure, when applied to active region observations, generally shows that
the filling factor f recovered from the fit is much less than unity outside sunspots, a
further indication of the existence of unresolved fields (Lites and Skumanich, 1990).
11.5.c Center-of-Gravity Technique
In this subsection we analyze the results of the application of the center-of-gravity
technique to observations involving unresolved fields. The corresponding analysis
on the bisector technique is omitted because too complicated and dependent on
too many parameters.
Using notations similar to those of Eq. (11.23), the first-order moments of the
right and left circular polarization profiles are given by (cf. Eq. (11.19))
1
, - dS I (P) − I± (λ; P) λ dλ
2 c
= Σ
(g)
λ± , (11.37)
Σ 1
dS Ic (P) − I± (λ; P) dλ
Σ 2
where the integrals in dS are extended to the solar region Σ covered by the in-
(g)
strument. It is convenient to introduce the local value of the moments, λ± (P),
defined by
1
I (P) − I± (λ; P) λ dλ
2 c
λ± (P) =
(g)
,
1
I (P) − I± (λ; P) dλ
2 c
and the local value of the integrated line depression
1
J± (P) = I (P) − I± (λ; P) dλ .
2 c
With these notations, Eq. (11.37) reads
(g)
, - λ± (P) J± (P) dS
(g) Σ
λ± = ,
Σ
J± (P) dS
Σ
which shows that the ‘unresolved’ center of gravity of the right (left) circular po-
larization profile is an average over the observed region of the ‘resolved’ center of
gravity of the same profile weighted by the corresponding integrated line depression.
As shown in Sect. 11.4, for any Zeeman triplet formed in a constant magnetic
field parallel to the line of sight (θ = 0◦ in the geometry of Fig. 9.1), one has (see
Eq. (11.22))
(g) (g)
λ+ (P) − λ− (P) = 2 g ∆λB (P)
and
J+ (P) = J− (P) = J (P) ,
654 CHAPTER 11
so that one obtains, in terms of apparent fields1
B (P) J (P) dS
(B )app = Σ . (11.38)
J (P) dS
Σ
This expression bears a strong resemblance to the corresponding expression de-
rived for the longitudinal magnetograph (Eq. (11.24)). By analogy, we consider
two particular cases. The first is the case where the right and left circular polar-
ization profiles are independent of the point P (apart from the shift due to the
magnetic field), and the continuum intensity is also independent of P. In this case
the integrated line depression J is independent of P and one obtains
1
(B )app = B = B (P) dS ,
A Σ
where A is the area of the surface Σ. This equation shows that, in the case of
unresolved fields, the center-of-gravity technique measures the flux (see comments
after Eq. (11.25)) of the magnetic field vector across the observed area, though
it should be stressed that this statement is only valid under the restrictive set of
assumptions specified above (Zeeman triplet, field parallel or antiparallel to the
line of sight, integrated line depression independent of P).
The second particular case is that of the filling factor hypothesis; one gets (cf.
del Toro Iniesta et al., 1990)
f Jm
(B )app = (B )m ,
(1 − f ) Jq + f Jm
where the indices ‘q’ and ‘m’ refer to the quiet atmosphere and to the magnetic
atmosphere, respectively.
11.6. Other Inversion Techniques
In the previous sections of this chapter we have analyzed in some detail the most
traditional inversion techniques of solar magnetometry, with the aim of providing
the reader with a firm background on the physical phenomena that underlie the
problem of measuring magnetic fields in the solar atmosphere. In this section we
give a brief review and some bibliographical references on more recent techniques,
without however pretending to be exhaustive. It should be realized that the amount
1
The same line of reasoning can be applied to θ = 180◦ and leads to the same expression for
(B )app . Equation (11.38) is thus valid for an arbitrary mixture of fields parallel or antiparallel
to the line of sight.
SOLAR MAGNETOMETRY 655
of work carried out on this subject during almost one century is very large, and a
full treatment is outside the aims of this book.
11.6.a Stokes Inversion based on Response Functions (SIR)
One of the most sophisticated methods that have been developed in the last few
years for recovering the magnetic field vector from Stokes profiles (observed in a
wavelength range covering one or more spectral lines) is the so-called SIR (‘Stokes
Inversion based on Response functions’: Ruiz Cobo and del Toro Iniesta, 1992).
This method can be considered as the natural evolution of two distinct techniques:
on one side, the Unno-fit technique (described in Sect. 11.3); on the other side,
an empirical approach which consists in fitting numerical solutions of the radia-
tive transfer equation based on detailed atmospheric models (quiet atmosphere,
sunspot, plage, etc.) to the observed Stokes profiles. The Unno-fit technique suf-
fers from the disadvantage of being based on a rather crude approximation (the
Milne-Eddington model atmosphere, described in Sect. 9.8), which does not allow
for the depth dependence of important physical parameters such as the absorption
coefficients, the slope of the source function, the line Doppler width, etc. On the
other hand, in the empirical approach outlined above the description of the depth
dependence of the thermodynamic parameters is certainly more accurate, but the
fit to the observed data is obtained by a trial-and-error procedure which suffers
from all the limitations typical of its kind.1
In the SIR method one constructs a merit function similar to that introduced for
the Unno-fit technique (see Eq. (11.14)), with the difference that the theoretical
Stokes profiles are now obtained via a numerical solution to the radiative transfer
equation for polarized radiation. The set of parameters {ζi } includes the magnetic
field (intensity, inclination and azimuth), the line-of-sight velocity of the atmo-
spheric plasma, and the thermodynamic parameters like temperature, gas pressure
and microturbulent velocity, all these quantities being parameterized at a few depth
points in the atmosphere. The procedure to find the best fit is still based on the
Marquardt algorithm and makes a heavy use of response functions and of general-
ized response functions (see Sect. 9.16) in order to calculate the derivatives of the
theoretical Stokes profiles with respect to the different parameters ζi .
The SIR technique was developed for the interpretation of spectral lines formed
under LTE conditions in plane-parallel atmospheres. The method has been success-
fully applied to solar spectropolarimetric observations in photospheric lines, e.g.,
for recovering the depth dependence of the atmospheric parameters in sunspots
(Westendorp Plaza et al., 1997) and the magnetic and thermodynamic parameters
of unresolved magnetic structures embedded in the quiet solar photosphere (Bellot
Rubio et al., 1997). More recently, the combination of efficient iterative methods
for the solution of multilevel transfer problems (Trujillo Bueno and Fabiani Ben-
dicho, 1995; Socas Navarro and Trujillo Bueno, 1997) with the SIR method has
allowed the development of a non-LTE inversion technique of Stokes profiles (Socas
1
The most important limitation is the ‘cultural bias’ introduced by the operator in determining
which of the parameters has to be varied to obtain a better fit.
656 CHAPTER 11
Fig.11.10. The peak asymmetry and the area asymmetry are defined in terms of the quantities
ab , ar , Ab , and Ar (see Eqs.(11.39)). Ab and Ar (where b and r stand, as customary, for ‘blue’
and ‘red’) are the absolute values of the areas contained between the V profile and the wavelength
axis, ab and ar the absolute values of the peak heights. The figure refers to a positive polarity
observation. For negative polarity it would be reversed about the wavelength axis.
Navarro et al., 2000).
11.6.b Inversion of FTS Data
The use of the Fourier Transform Spectrometer at the McMath-Pierce solar facility
of the National Solar Observatory (Kitt Peak) made possible, in the late 1970s, the
observation of the full solar spectrum in two directions of polarization (typically
I + V and I − V ) with unprecedented spectral resolution though with very low
spatial resolution (Stenflo et al., 1983a, 1983b, 1984). Such observations, usually
performed on facular regions, revealed a number of interesting phenomena and, in
particular, pointed out the existence of prominent asymmetries in the V profiles
of most spectral lines.1 These asymmetries have been characterized by Solanki
and Stenflo (1985) by means of two parameters: the peak asymmetry and the area
asymmetry defined, respectively, as
ab − ar Ab − Ar
δa = , δA = , (11.39)
ab + ar Ab + Ar
where the meaning of the different symbols is specified in Fig. 11.10. It turns
out that both the peak asymmetry and the area asymmetry are generally positive
(irrespective of the prevailing polarity of the observed active region), and widely
vary from line to line. Typical values at disk center are of the order of 10% for the
peak asymmetry and 5% for the area asymmetry. Such values generally decrease
with increasing distance from disk center and become negative for many lines at
the extreme limb.
1
Stokes profiles asymmetries had already been observed by the so-called Stokes-I and Stokes-II
polarimeters of the High Altitude Observatory (Baur et al., 1981).
SOLAR MAGNETOMETRY 657
ray path
walls of the
magnetic flux-tube
velocity flow
Fig.11.11. In a funnel-shaped magnetic flux-tube a ray path like (a) gives rise to a V profile with
positive peak asymmetry and positive area asymmetry.
The interpretation of these results has basically been given in terms of the filling
factor hypothesis, by supposing that the polarization signal contained in the obser-
vations (the spectrum of V , or of Q and U ) is due to a collection of unresolved and
ubiquitous magnetic structures, the so-called flux-tubes. In an extensive series of
papers, the geometrical and thermodynamic properties of a kind of ‘average’ flux-
tube have been established by comparing the observed polarization profiles with
numerical solutions to the radiative transfer equations for polarized radiation (see
Solanki, 1993, for a detailed review of the subject). The model that has emerged
from these investigations is that of a slender, funnel-shaped, static magnetic tube
having typical horizontal dimensions of a few hundreds of km, in hydrostatic equi-
librium with the surrounding unperturbed atmosphere. The asymmetries of the V
profiles are interpreted in this model as due to the presence of a macroscopic flow of
material directed downward in the non-magnetic atmosphere surrounding the tube
(Grossman-Doerth et al., 1988b). It can be proved that the ray paths crossing
the walls of the flux-tube, such as that schematized in Fig. 11.11, lead to positive
values for the peak and area asymmetries defined in Eqs. (11.39).1 It should be
remarked that the empirical flux-tube model described above is consistent with a
dynamical model obtained by two-dimensional magnetohydrodynamic simulations
of a collapsing magnetic structure (Knölker et al., 1988).
Notwithstanding the success obtained by the flux-tube hypothesis in explaining
the FTS data, the intrinsic lack of spatial resolution of the data themselves in-
evitably raises some doubts on whether the model described above may indeed
1
The positive sign of the area asymmetry is in fact predicted by the simple slab-model
developed as Case iii of Sect. 9.21. That model can be adapted to the present case by setting
(s) (s)
∆λA < 0 and (∆λB − ∆λB ) < 0. For a line with g > 0, Eqs. (9.256) and (9.257) yield v > 0
for θ = 0 and v < 0 for θ = 180◦ , which are consistent with a positive area asymmetry (cf.
◦
Eq. (9.238)).
658 CHAPTER 11
give a faithful representation of the physical reality. A definite breakthrough on
this subject may only be obtained when high spatial resolution observations, of the
same spectral quality as the FTS data, will be available.
11.6.c Infrared Lines
Infrared lines are of great importance in solar magnetometry. This is due to the
fact that, with increasing the wavelength λ, the Zeeman splitting of spectral lines
increases as λ2 (see Eq. (3.14)), whereas the typical line width, due essentially to
the Doppler effect, increases as λ (see Eq. (5.46)). As a consequence, when moving
towards the infrared, the different Zeeman components are more and more spaced
in wavelength for a given value of the magnetic field intensity. To give an example,
a magnetic field of 300 G is sufficient to produce a splitting as large as three Doppler
widths in a Zeeman triplet at 12 µm formed in the solar atmosphere.1 Although
this looks very promising, it should be taken into account that the resolving power
of a telescope scales as λ−1 , which is a rather serious drawback in solar magnetom-
etry. A 1m-aperture telescope, for instance, has a theoretical spatial resolution of
approximately 0”.1 at optical wavelengths, but the resolution degrades to 2”.5 at
12 µm.
In spite of these limitations, infrared solar magnetometry is a field which is rapidly
evolving both on the instrumental and the theoretical side. Lists of infrared lines
especially suitable for solar magnetometry can be found in Solanki et al. (1990),
Ramsauer et al. (1995), Rüedi et al. (1995).
In general, any of the techniques described earlier in this chapter – like the Unno-
fit or the center-of-gravity technique – can be applied to recover the magnetic field
from infrared observations. There is however a particular case which deserves
special attention, because it leads to a very simple inversion method. This is
the case where the magnetic field is sufficiently strong to separate completely the
different Zeeman components, a case discussed in detail in Sect. 9.12. It was shown
there that the solutions to the radiative transfer equation can be written in the form
of Eqs. (9.144). Here we restrict attention to the case of a normal Zeeman triplet
and we also assume the magnetic field vector to be constant and the atmosphere
to be static.2 Under these approximations the emerging Stokes parameters can be
cast into the form:
- in the σb component
I(λ) = Ic 1 − pb (λ)
sin2 θ cos 2χ
Q(λ) = Ic pb (λ)
1 + cos2 θ
1
This result is obtained from Eq. (5.49) with T = 5800 K, ξ = 1 km s−1 , µ = 55.8 (atomic
weight of iron).
2
It should be recalled that Eqs. (9.144) are valid under the restriction of a constant magnetic
field direction. Gradients of the magnetic field intensity and of the line-of-sight velocity are in
principle allowed, provided that no overlapping of the different Zeeman components occurs (cf.
Eqs. (9.138)).
SOLAR MAGNETOMETRY 659
sin2 θ sin 2χ
U (λ) = Ic pb (λ)
1 + cos2 θ
cos θ
V (λ) = 2Ic pb (λ) , (11.40)
1 + cos2 θ
- in the π component
I(λ) = Ic 1 − pp (λ)
Q(λ) = −Ic pp (λ) cos 2χ
U (λ) = −Ic pp (λ) sin 2χ
V (λ) = 0 , (11.41)
- in the σr component
I(λ) = Ic 1 − pr (λ)
sin2 θ cos 2χ
Q(λ) = Ic pr (λ)
1 + cos2 θ
sin2 θ sin 2χ
U (λ) = Ic pr (λ)
1 + cos2 θ
cos θ
V (λ) = −2Ic pr (λ) , (11.42)
1 + cos2 θ
where Ic is the continuum intensity, θ and χ are the inclination and azimuth an-
gles defined in Fig. 9.1, and pb (λ), pp (λ), pr (λ) are three profiles centered at the
wavelengths (λ0 − g ∆λB ), λ0 , and (λ0 + g ∆λB ) respectively, g being the Landé
factor and ∆λB the Zeeman splitting defined in Eq. (3.14). The two profiles pb (λ)
and pr (λ) have the same shape and coincide with the profile of a fictitious, non-
magnetic line formed in the same atmosphere and having a ratio between line and
continuum absorption coefficient given by
1
κ= 1 + cos2 θ κL ,
2
where κL is the corresponding quantity for the full line. The profile pp (λ) has
a different shape and coincides with the profile of a fictitious, non-magnetic line
having
κ = sin2 θ κL .
Equations (11.40)-(11.42) can be used to derive the magnetic field vector from
the observed Stokes profiles of an infrared triplet with well-separated σ and π
components. The field intensity is directly obtained by measuring the wavelength
separation between any two components (σ or π) in any of the four Stokes param-
eters and by using Eq. (3.14).1 The azimuth χ is obtained by considering the ratio
1
Note that to perform this operation there is no need for polarimetric observations, it
660 CHAPTER 11
between Q and U in any of the components, σ or π. One has
U (λ)
tan 2χ = .
Q(λ)
Finally, the inclination θ is obtained from the ratio between V and Q2 + U 2 in
either the σb or the σr component. For instance, in the σb component
2 cos θ V (λ)
= ,
sin2 θ Q (λ) + U 2 (λ)
2
whence √
1 + x2 − 1
θ = arccos ,
x
where
V (λ)
x= .
Q2 (λ) + U 2 (λ)
Obviously, in the case of noisy data this simple procedure can be replaced by a
more sophisticated one involving the minimization of an appropriate merit function,
similarly to what discussed in Sect. 11.3 about the Unno-fit technique.
11.6.d The Inversion Technique under the MISMA Hypothesis
As illustrated in Subsect. 11.6.b, the Stokes profiles asymmetries observed in facular
regions with the Fourier Transform Spectrometer have been interpreted in terms of
flux-tube models. An alternative explanation of such asymmetries and of similar
phenomena observed in the penumbrae of sunspots (Sánchez Almeida and Lites,
1992) and in the quiet sun (Lites et al., 1996) is based on the MISMA model,
where the acronym stands for ‘Micro-Structured Magnetic Atmosphere’ (Sánchez
Almeida et al., 1996). This model is based on the idea that the magnetic structures
pervading the solar atmosphere have typical sizes smaller than the mean free path
of a photon. In such an environment, the most natural way to treat the process of
line formation is the one that we have analyzed in detail in Sect. 9.24 by introducing
a stochastic approach. We found there that, under the so-called microturbulent
limit (corresponding to the case where the dimensions of the magnetic structures
are much smaller than the photon’s mean free path), one obtains for the Stokes
parameters emerging from a Milne-Eddington model atmosphere (cf. Eq. (9.276))
−1
I = B0 1 + β A U, (11.43)
is sufficient to have the intensity spectrum. Moreover, this operation is so simple as to justify
the statement, often heard about infrared lines, ‘A ruler is all that is needed to measure the
magnetic field’. However, Eqs. (11.40)-(11.42) show that the intensity spectrum carries very little
information about θ and no information at all about χ .
SOLAR MAGNETOMETRY 661
where B0 and β are the parameters specifying the source function, A is the prop-
agation matrix defined in Eqs. (9.164), and where the meaning of the symbol · · ·
is defined in Eq. (9.271).
In the MISMA hypothesis, the stochastic average of the propagation matrix, A,
is assumed to be represented by an expression of the form
n
A = fi Ai . (11.44)
i=1
This is a special case of Eq. (9.271) corresponding to the assumption that the
atmosphere consists of n independent components, each one characterized by its
own propagation matrix, Ai , and its own probability (or occupation fraction), fi .
By means of Eqs. (11.43) and (11.44) it is possible to reproduce rather well the
typical asymmetries of the Stokes profiles observed either in sunspots, or in facular
regions or on the quiet sun, even if this requires, in some cases, the introduction
of up to three components (Sánchez Almeida et al., 1996). For explaining facular
observations, for instance, one needs one prevailing, non-magnetic component with
an inclined flow, a static magnetic component with a field slightly inclined with
respect to the vertical, and a supplementary magnetic component having a field
again slightly inclined with respect to the vertical and a flow of material antiparallel
to the magnetic field.
The MISMA hypothesis has also been generalized to allow for a depth depen-
dence of the matrices Ai of Eq. (11.44) and an arbitrary behavior with depth of
the source function BP . An inversion technique for this case has been proposed by
Sánchez Almeida (1997). Broadly speaking, this technique resembles, in its math-
ematical details, the one that we have already described in Subsect. 11.6.a, with
the difference that the radiative transfer equation that is solved to find a fit to the
observations is the ‘stochastic’ equation in the microturbulent limit, namely
dI
= A I − BP U ,
dτ
where the matrix A is given by Eq. (11.44) and BP is the local value of the Planck
function. This generalized version of the MISMA hypothesis can lead to a very large
number of degrees of freedom. To narrow down the number of possibilities, the
inversion code uses several restrictive assumptions based on magnetohydrodynamic
considerations. This inversion procedure seems very promising and has proved
particularly successful for interpreting the asymmetries observed in the V profile
in quiet sun observations (Sánchez Almeida and Lites, 2000).
11.7. Disambiguation
We have already remarked in Sect. 9.5 that the radiative transfer equation for po-
larized radiation is invariant under a rotation of 180◦ of the magnetic field vector
662 CHAPTER 11
about the line of sight, provided atomic polarization phenomena are neglected (see
the introduction to Chap. 9). In most of the spectral lines that are commonly
employed in solar magnetometry, atomic polarization is believed to play an in-
significant role. It follows that the magnetic field vector recovered from standard
observations suffers from an intrinsic 180◦ ambiguity in its azimuth. This raises the
fundamental problem – often referred to as the disambiguation problem – of find-
ing the true orientation of the transverse component of the magnetic field vector
between the two alternatives which are left undecided by the observations.
In the case of an isolated observation the problem cannot be solved. On the
contrary, in the case of raster-type observations, several different strategies can be
followed. Important contributions to this matter have been given by Cuperman et
al. (1990), Canfield et al. (1993), Lites et al. (1995), and Semel and Skumanich
(1998). Just to give an example, we describe in some detail the procedure followed
by Lites et al. (1995): one starts by considering the sunspots as ‘azimuth centers’
from which the transverse component of the magnetic field must diverge for positive
polarity or to which must converge for negative polarity. This criterion solves the
ambiguity for the regions surrounding the sunspots. In plage areas, and away from
regions of newly emerging flux, one selects the azimuth which gives a field direction
closer to the local vertical.1 Finally, when the previous criteria are not effective,
the selection is made by minimizing the point-by-point discontinuity of the azimuth
in the observer’s frame. By so doing, a ‘map’ of the magnetic field is recovered and
this also allows one to determine one component of the electric current density jz
(with z the vertical axis) through the equation
∂By ∂Bx
4π jz = − .
∂x ∂y
It should be remarked that a purely spectroscopic disambiguation method has
also been devised (Landi Degl’Innocenti and Bommier, 1993). The method is based
on the observation of the Stokes parameters profiles in resonance lines originating
from transitions between polarized atomic levels. In this case, the vertical to the
solar atmosphere introduces a ‘symmetry breaking’ in the physical scenario which
allows one, in many cases, to recover the magnetic field vector without ambiguity.
1
This criterion is based on the idea that in such regions the magnetic field is nearly always
oriented close to the vertical direction. Obviously, the criterion does not work for observations
performed at disk center.
CHAPTER 12
ASTROPHYSICAL APPLICATIONS:
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES
As shown in Chaps. 5, 7, and 10, the anisotropy of the radiation field, suitably de-
K
scribed by the irreducible tensor JQ (ν) defined in Eqs. (5.157), is the basic physical
quantity controlling the amount of atomic polarization in atomic energy levels. Re-
ferring for instance to the idealized case of a two-level atom with unpolarized lower
level, Eq. (10.10) shows that in the non-magnetic, collisionless regime, the statisti-
cal tensors of the upper level are proportional to the corresponding components of
the radiation field tensor. In the following chapters we will discuss the phenomena
of resonance polarization and the Hanle effect – which are intimately connected
with the presence of atomic polarization – in the astrophysical context. In view of
this analysis, it is useful to study in some detail the properties of the radiation field
anisotropy in some of the most common astrophysical scenarios, like the interior
or the outer layers of a ‘classical’ stellar atmosphere.1
The study presented in this chapter is restricted to schematic situations where
the atmospheric structure is described by analytical, plane-parallel models. In
particular, we will consider the Milne-Eddington and the grey model atmospheres,
and we will find the behavior of the radiation field anisotropy both ‘inside’ the
atmosphere (as a function of optical depth) and ‘outside’ the atmosphere (as a
function of height over the stellar surface).
It should be remarked from the very beginning that a plane-parallel atmosphere
is a cylindrically symmetrical environment, the symmetry axis being the vertical
to the atmosphere itself. It is readily seen from Eqs. (5.157) and Table 5.6 that,
in such an environment, the only non-zero components of the irreducible radiation
field tensor (in a reference system with the z-axis directed as the vertical) are J00 and
J02 . A single parameter is therefore sufficient to fully characterize the anisotropy
of the radiation field. This parameter is the so-called anisotropy factor, already
defined in Chap. 10 (cf. Eqs. (10.62) and (10.160))
√ J02 (ν)
wν = 2 , (12.1)
J00 (ν)
where
√ the expressions of J00 (ν) and J02 (ν) are given by Eqs. (5.164) and the factor
2 is introduced for convenience. We remind the reader (cf. Sect. 10.7) that the
value of the anisotropy factor is always contained in the range (−1/2, 1), being 0 for
an isotropic radiation field, 1 for an infinitely sharp beam parallel or antiparallel to
the z-axis, and −1/2 for an azimuth-independent radiation field in the x-y plane.
1
By ‘classical’ we mean an atmosphere where the radiation field is unpolarized and therefore
described by the only intensity.
664 CHAPTER 12
The main body of this chapter is devoted to the analysis of the anisotropy factor
in stellar atmospheres. In Sect. 12.4 we will consider some typical phenomena
of ‘symmetry breaking’ which entail the appearance of further components of the
K
tensor JQ (ν).
12.1. The Milne-Eddington Model Atmosphere
We consider a plane-parallel atmosphere such as the one schematized in Fig. 9.2
and we denote by tν the optical depth at frequency ν measured along the z-axis in
the inward direction
dtν = −kν dz ,
where kν is the absorption coefficient (corrected for stimulated emission) at fre-
quency ν.1 The transfer equation for the specific intensity Iν (tν , µ) of the radia-
tion propagating at optical depth tν along a direction forming an angle θ with the
vertical is
d
µ I (t , µ) = Iν (tν , µ) − Sν (tν ) , (12.2)
dtν ν ν
where µ = cos θ and Sν , the source function at frequency ν, is the ratio between
the emission and the absorption coefficient (Sν = εν /kν ).
Equation (12.2) can easily be solved to give, for the radiation propagating out-
wards
∞ t −t
− ν µ ν dtν
Iν (tν , µ) = Sν (tν ) e (µ > 0) , (12.3)
µ
tν
and, for the radiation propagating inwards
tν t −t
− ν ν dtν
Iν (tν , µ) = Sν (tν ) e |µ| (µ < 0) , (12.4)
|µ|
0
where we have assumed that the atmosphere is not irradiated by any external
source.
We now evaluate the moments of the intensity, defined by (cf. Chandrasekhar,
1950)
1
1
Jν (tν ) = Iν (tν , µ) dµ
2
−1
1
The absorption coefficient considered here is the ‘scalar’ absorption coefficient of the
traditional (non polarized) theory of stellar atmospheres. For our purposes it is not necessary to
specify the physical processes contributing to kν .
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 665
1
Fν (tν ) = 2 Iν (tν , µ) µ dµ
−1
1
1
Kν (tν ) = Iν (tν , µ) µ2 dµ . (12.5)
2
−1
Jν is the solid-angle average of the specific intensity, Fν is (apart from a factor π)
the specific flux,1 and Kν is the so-called K-integral . Substituting Eqs. (12.3)-
(12.4) into Eqs. (12.5), and recalling the definition of the exponential integrals
∞ −xy 1
e x
−µ
En (x) = dy = e µn−2 dµ (x > 0 ; n = 0, 1, 2, . . .) , (12.6)
yn
1 0
one obtains with some easy algebra
∞
1
Jν (tν ) = E1 |tν − tν | Sν (tν ) dtν (12.7a)
2
0
∞
Fν (tν ) = 2 E2 |tν − tν | σ(tν − tν ) Sν (tν ) dtν (12.7b)
0
∞
1
Kν (tν ) = E3 |tν − tν | Sν (tν ) dtν , (12.7c)
2
0
where
tν − tν
σ(tν − tν ) = sign (tν − tν ) = . (12.8)
|tν − tν |
The anisotropy factor at optical depth tν can be expressed in terms of the mo-
ments Jν (tν ) and Kν (tν ). From Eqs. (12.1), (5.164) and (12.5) we have
3Kν (tν ) − Jν (tν )
wν (tν ) = , (12.9)
2 Jν (tν )
or, using Eqs. (12.7a,c)
∞
3E3 |tν − tν | − E1 |tν − tν | Sν (tν ) dtν
0
wν (tν ) = ∞ . (12.10)
2 E1 |tν − tν | Sν (tν ) dtν
0
1
The flux is indeed π Fν .
666 CHAPTER 12
This expression is valid under the only assumption that the atmosphere is plane-
parallel (hence cylindrically symmetrical).
Now we consider the special case of the Milne-Eddington model, which implies
that the source function coincides with the local Planck function (LTE hypothesis)
and that the latter varies linearly with the optical depth tν ,1
Sν (tν ) ≡ BP (ν, tν ) = B0 (1 + βν tν ) . (12.11)
Substituting into Eq. (12.10), and performing elementary integrals,2 one finds the
following expression
1 E2 (tν ) − βν E3 (tν ) − 3E4 (tν ) + 3βν E5 (tν )
wν (tν ) = . (12.12)
2 2 + 2βν tν − E2 (tν ) + βν E3 (tν )
The surface value of the anisotropy factor is
1 βν
wν (0) = . (12.13)
4 βν + 2
In the special case of an isothermal atmosphere (βν = 0) one has
1 E2 (tν ) − 3E4 (tν )
wν (tν ) isoth = .
2 2 − E2 (tν )
Figure 12.1 shows the behavior of the anisotropy factor with optical depth as de-
rived from Eq. (12.12), for several values of βν (only positive values are considered,
which implies a source function increasing with optical depth). Such behavior can
be understood via the following arguments. First of all it should be noticed that,
as apparent from Eqs. (5.164), the component J02 (ν) – hence the anisotropy fac-
tor – tends to be positive when the radiation flowing ‘almost vertically’ (θ 0◦ or
θ 180◦ ) is more intense than the radiation flowing ‘almost horizontally’ (θ 90◦ ),
and negative in the opposite case.3 Next we write down the expressions of the inten-
sity Iν (tν , µ) for the Milne-Eddington model atmosphere. From Eqs. (12.3)-(12.4)
we obtain, using Eq. (12.11)4
1
Apart from slightly different notations, these assumptions are the same as those introduced
in Sect. 9.8 when discussing the Unno-Rachkovsky solution.
2
It is useful to recall that the exponential integrals satisfy the relations
d 1
E (x) = −En−1 (x) (n > 0) , En (0) = (n > 1) .
dx n n−1
3 √
The boundary between the two cases involves the Van Vleck angle θV = arccos (1/ 3), cf.
Eq. (5.100).
4
Equations (12.14) can also be derived as a special case of the Unno-Rachkovsky solutions
(Eqs. (9.108)) corresponding to no magnetic field. It should be kept in mind, however, that tc is
the continuum optical depth, while tν is the total (line plus continuum) optical depth.
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 667
Fig.12.1. The anisotropy factor in a Milne-Eddington model atmosphere is plotted as a function
of optical depth for different values of the parameter βν defined in Eq.(12.11).
Iν (tν , µ) = B0 1 + βν µ + βν tν (µ > 0)
−
tν , t -
− ν
Iν (tν , µ) = B0 1 + βν tν − e |µ| − βν |µ| 1 − e |µ| (µ < 0) . (12.14)
These expressions show that, for the outward-directed radiation (µ > 0), the in-
tensity increases – provided βν > 0 – with increasing µ, whatever the value of tν .
The opposite occurs for the radiation flowing inward (µ < 0): at any given tν , the
intensity decreases as |µ| is increased.
The former effect is responsible for limb darkening and will be referred to as the
source-function gradient effect: it gives a positive contribution to the anisotropy
factor. The latter, which may be called the surface effect , gives a negative con-
tribution to the anisotropy factor. For βν = 0 the radiation flowing outward is
isotropic, thus only the surface effect is at work and, as shown in Fig. 12.1, the
anisotropy factor is always negative. By contrast, for large values of βν the source-
function gradient effect takes over and the anisotropy factor is always positive. For
intermediate values of βν , the anisotropy factor is positive at tν = 0 (as shown by
Eq. (12.13)) and then decreases with increasing tν . If βν is smaller than a thresh-
old value (that is numerically found to be 1.1098) the anisotropy factor becomes
negative beyond a certain value of tν , otherwise it remains positive at any tν . In
any case, the anisotropy factor tends to zero for tν → ∞, because both the outward
and the inward-directed radiation become isotropic.
The maximum value of the anisotropy factor in the Milne-Eddington model at-
mosphere is 0.25. Such value is found at tν = 0 in a high-gradient atmosphere
668 CHAPTER 12
(βν → ∞). The minimum value is −0.0611, found at tν = 0.2097 in an isothermal
atmosphere (βν = 0).
12.2. The Grey Atmosphere
A plane-parallel atmosphere in local thermodynamic equilibrium and in radiative
equilibrium, and having a frequency-independent absorption coefficient, is referred
to as a grey atmosphere. This model atmosphere has been extensively studied
in the past and its main properties have been fully established by means of very
elegant analytical and numerical techniques (Chandrasekhar, 1950).
As the absorption coefficient is ν-independent, the grey atmosphere allows a
‘unique’ optical depth t to be defined. Bearing in mind the LTE assumption, the
transfer equation for the specific intensity at frequency ν reads
d
µ I (t, µ) = Iν (t, µ) − BP ν, T (t) , (12.15)
dt ν
where BP ν, T (t) is the Planck function at temperature T (t) and frequency ν.
The formal solution for Iν (t, µ), as well as the expressions for the moments and for
the anisotropy factor, are readily obtained from Eqs. (12.3)-(12.9) by performing
the substitutions
tν → t , Sν (tν ) → BP ν, T (t) . (12.16)
The radiative equilibrium assumption means that the frequency-integrated flux is
independent of optical depth. It is therefore convenient to introduce the frequency-
integrated intensity
∞
I(t, µ) = Iν (t, µ) dν ,
0
and its moments (cf. Eqs. (12.5))
1
1
J(t) = I(t, µ) dµ
2
−1
1
F (t) = 2 I(t, µ) µ dµ
−1
1
1
K(t) = I(t, µ) µ2 dµ . (12.17)
2
−1
Integration of Eq. (12.15) over frequency yields the transfer equation for I(t, µ)
d
µ I(t, µ) = I(t, µ) − BP (t) , (12.18)
dt
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 669
where
∞
σ 4
BP (t) = BP ν, T (t) dν = T (t) , (12.19)
π
0
with σ the Stefan-Boltzmann constant and T the local temperature. By averaging
over the solid angle Eq. (12.18), and the same equation multiplied by µ, one obtains
1 d
F (t) = J(t) − BP (t)
4 dt
d 1
K(t) = F (t) ,
dt 4
where Eqs. (12.17) have been used. The assumption of radiative equilibrium gives
J(t) = BP (t) (12.20a)
F (t) = F0 (12.20b)
1
K(t) = F0 t + Q , (12.20c)
4
where Q is an integration constant. Equations (12.20a) and (12.18) allow us to
write the zero-order moment of the frequency-integrated intensity in the form (cf.
Eq. (12.7a))
∞
1
J(t) = E1 |t − t| J(t ) dt , (12.21)
2
0
while for the first-order moment we have, considering the flux at the surface (cf.
Eq. (12.7b))
∞
F0 = 2 E2 (t) J(t) dt . (12.22)
0
In order to solve the grey-atmosphere problem, we need to find the dependence on
optical depth of the temperature T , or equivalently – because of Eq. (12.20a) – of
the mean intensity J. The latter obeys the homogeneous integral equation (12.21)
and the normalization required by the boundary condition at the stellar surface
(Eq. (12.22)).
Consideration of the boundary conditions in the deep interior of the atmosphere
suggests a convenient algebraic manipulation to be performed on Eq. (12.21). Since
under the limit t → ∞ the radiation field eventually becomes isotropic, and since
for an isotropic field one has (see Eqs. (12.17))
J(t) = 3 K(t) ,
it follows from Eq. (12.20c) that for t 1
3
J(t) ≈ F0 t + Q .
4
670 CHAPTER 12
Following Hopf (1934), the function J(t) is written, for an arbitrary value of t, in
the form
3
J(t) = F0 t + q(t) , (12.23)
4
where q(t) is the so-called Hopf function which obviously satisfies the property
lim q(t) = Q .
t→∞
Substitution of Eq. (12.23) into Eq. (12.21), and evaluation of elementary integrals,1
leads to the following non-homogeneous integral equation for q(t)
∞
1 1
q(t) = E3 (t) + E1 |t − t| q(t ) dt . (12.24)
2 2
0
On the other hand, substitution of Eq. (12.23) into Eq. (12.22) shows that the Hopf
function obeys the relation
∞
1
E2 (t) q(t) dt = . (12.25)
3
0
The Hopf function has been widely studied by means of analytical, as well as
numerical techniques (Chandrasekhar, 1950; Kourganoff and Busbridge, 1952). In
particular, it can be proved that q(0), the surface value of the Hopf function, is
given by (see App. 15)
1
q(0) = √ . (12.26)
3
With modern computing techniques, the evaluation of the Hopf function is rather
straightforward. A simple numerical algorithm is described in App. 16, and a graph
of q(t) is given in Fig. 12.2. The figure√shows that q(t) is a monotonic function
which grows from the surface value 1/ 3 = 0.57735 to the boundary value at
t → ∞, Q = 0.71044.
The anisotropy factor for the frequency-integrated intensity in a grey atmosphere
can easily be expressed in terms of the Hopf function. Considering the grey ana-
logue of Eq. (12.9) one obtains, with the use of Eqs. (12.20c) and (12.23)
3K(t) − J(t) Q − q(t)
w(t) = = . (12.27)
2 J(t) 2 t + q(t)
In particular, for t = 0 one has
√
3Q − 1
w(0) = = 0.11526 .
2
1
See footnote 2 on p.666.
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 671
Fig.12.2. The Hopf function, solution of the integral equation (12.24), is plotted against optical
depth.
Fig.12.3. The anisotropy factor for the frequency-integrated intensity in the grey model atmo-
sphere is plotted as a function of optical depth. Its surface value is 0.11526.
Figure 12.3 shows a graph of the anisotropy factor w versus Log t. The function w(t)
decreases monotonically with increasing t – as obvious from Eq. (12.27). According
to the discussion presented at the end of the former section this means that, in the
grey model atmosphere, the source-function gradient effect prevails over the surface
effect at all optical depths.
In the grey atmosphere it is also possible to find the behavior with optical depth
of the anisotropy factor at a specific frequency (or wavelength). The expression of
the ‘monochromatic’ anisotropy factor, wν (t), is obtained from Eqs. (12.9), (12.7a)
and (12.7c) by performing the substitutions in Eq. (12.16). It is therefore necessary
to determine the dependence of the temperature T on optical depth, and then
to evaluate numerically the integrals in Eqs. (12.7a,c). Introducing the effective
672 CHAPTER 12
Fig.12.4. The frequency-dependent anisotropy factor of a grey atmosphere is plotted against
optical depth for different frequencies. The curves are labelled by the value of the normalized
frequency x defined in Eq.(12.28).
temperature Teff via the definition
σ 4
F0 = T ,
π eff
we obtain, with the use of Eqs. (12.19), (12.20a) and (12.23)
.
4
3
T (t) = Teff t + q(t) ,
4
thus the source function can be written in the form
< . −1
2hν 3 4
3
Sν (t) = BP ν, T (t) = 2 exp x t + q(t) −1 ,
c 4
where
hν
x= . (12.28)
kB Teff
Figure 12.4 shows the anisotropy factor wν as a function of Log t for four different
x values. For the solar case (Teff = 5800 K) the four curves correspond approxi-
mately to the wavelengths (from top to bottom) 3500, 5000, 8000, and 16000 Å,
respectively. The increase of the anisotropy factor with increasing frequency is
simply understood in terms of the source-function gradient effect illustrated in the
previous section. The relative gradient of the source function is given by
1 dSν (t) d
βν = = ln Sν (t) ,
Sν (t) dt dt
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 673
Fig.12.5. The surface value of the anisotropy factor of a grey atmosphere is plotted as a function
of the normalized frequency x.
and, for the LTE case,
d − hν −1 h ν dT
βν = ln BP ν, T (t) = 1 − e kB T ,
dt kB T 2 dt
which, at any optical depth, is a monotonically increasing function of frequency.
As a further illustration of the same effect, we plot in Fig. 12.5 the surface value of
the anisotropy factor wν as a function of x. The limiting value in the far infrared
(x → 0) is 0.02571.
12.3. Outer Atmospheres
In the previous sections we have analyzed the anisotropy factor in the interior of a
stellar atmosphere. Here we extend such analysis to the ‘outer’ environment of a
star. Referring to Fig. 12.6, let us consider a point P located at a height h over the
stellar surface. Neglecting the possible presence of starspots over the spherical cap
seen from point P, the radiation field is cylindrically symmetrical about the vertical
direction, and its properties are fully described by the two components J00 (ν) and
J02 (ν) of the radiation field tensor or, alternatively, by J00 (ν) and the anisotropy
factor wν .
Similarly to Sect. 12.1, we introduce the two integrals
1 1
1 1
Jν = Iν (µ) dµ , Kν = Iν (µ) µ2 dµ , (12.29)
2 2
−1 −1
where Iν (µ) is the specific intensity of the radiation propagating through point P
along the direction specified by the angle θ, with µ = cos θ. The anisotropy factor
674 CHAPTER 12
Fig.12.6. Geometry for the calculation of the anisotropy factor in the outer layers of a star. R∗
is the stellar radius.
at point P is given by Eq. (12.9),
3Kν − Jν
wν = . (12.30)
2 Jν
Obviously the anisotropy factor is affected by the limb-darkening law, which is
usually given by an expression of the form
N
Iν (α) = Iν (0) 1 − ui (ν) 1 − cosi α , (12.31)
i=1
where α is the angle between the propagation direction of the radiation and the ver-
tical to the atmosphere, and where ui (ν) are empirically determined coefficients.1
As apparent from Fig. 12.6, the angle α is related to the angle θ. Since
R∗
sin γ = , (12.32)
R∗ + h
1
The direct determination of these coefficients from observations is possible only for the
sun. In the stellar case, they can be determined from a fit to theoretical results based on model
atmospheres.
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 675
one obtains from the sine-formula
R∗ + h sin θ
sin α = sin θ = ,
R∗ sin γ
whence
cos2 θ − cos2 γ
cos α = . (12.33)
sin γ
Substituting Eq. (12.31) into Eq. (12.29), and taking into account Eq. (12.33), one
gets
1
1
N
(µ2 − cos2 γ)i
Jν = Iν (0) u0 (ν) + ui (ν) dµ
2 i=1
sini γ
cos γ
1
1
N
(µ2 − cos2 γ)i
Kν = Iν (0) u0 (ν) + ui (ν) µ2 dµ ,
2 i=1
sini γ
cos γ
where
N
u0 (ν) = 1 − ui (ν) .
i=1
We now restrict our analysis to the case of a quadratic expansion of the limb-
darkening law (N = 2 in Eq. (12.31)). After some algebra, which involves the
evaluation of elementary integrals, one gets
1
Jν = I (0) a0 + a1 u1 (ν) + a2 u2 (ν)
2 ν
1
Kν = Iν (0) b0 + b1 u1 (ν) + b2 u2 (ν) , (12.34)
2
where
a0 = 1 − Cγ
1 1 Cγ2 1 + Sγ
a1 = Cγ − − ln
2 2 Sγ Cγ
(Cγ + 2)(Cγ − 1)
a2 = (12.35)
3(Cγ + 1)
and
1
b0 = 1 − Cγ3
3
1 3 1 Cγ4 1 + Sγ
b1 = 8Cγ − 3Cγ − 2 −
2
ln
24 8 Sγ Cγ
(Cγ − 1)(3Cγ3 + 6Cγ2 + 4Cγ + 2)
b2 = ,
15(Cγ + 1)
676 CHAPTER 12
with
Cγ = cos γ , Sγ = sin γ .
Substitution into Eq. (12.30) yields
1 c0 + c1 u1 (ν) + c2 u2 (ν)
wν = , (12.36)
2 a0 + a1 u1 (ν) + a2 u2 (ν)
where
c0 = 3b0 − a0 = Cγ Sγ2
1 Cγ2 1 + Sγ
c1 = 3b1 − a1 = 8Cγ − 3Cγ − 8Cγ + 2 + (4 − 3Cγ )
3 2 2
ln
8 Sγ Cγ
Cγ − 1
c2 = 3b2 − a2 = 9Cγ3 + 18Cγ2 + 7Cγ − 4 . (12.37)
15(Cγ + 1)
It is instructive to consider some limiting cases of Eq. (12.36). The first is the case
of a star without limb darkening. Setting u1 (ν) = u2 (ν) = 0, Eq. (12.36) reduces
to
1 c0 1
wν = = Cγ (1 + Cγ ) .
2 a0 2
The anisotropy factor is independent of frequency and is simply due to geometri-
cal effects. Recalling Eq. (12.32), wν can explicitly be expressed in terms of the
height h . We have √
2x + x2 + (1 + x) 2x + x2
wν = ,
2(1 + x)2
where
h
x= .
R∗
In particular, for small heights (h R∗ ) we obtain the asymptotic behavior
.
x h
wν ≈ = ,
2 2R∗
while for large heights we find the obvious result (see comments after Eq. (12.1))
lim wν = 1 .
h→∞
Next we consider the behavior of wν for small and large values of h/R∗ in the
general case where u1 (ν) and u2 (ν) are non-zero. For h → 0 (or γ → π/2) one
obtains from Eq. (12.36), with the help of Eqs. (12.35) and (12.37)
1 15 u1 (ν) + 16 u2 (ν)
lim wν = wν 0 = .
h→0 20 6 − 3 u1 (ν) − 4 u2 (ν)
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 677
Fig.12.7. The anisotropy factor in the outer layers of the solar atmosphere is plotted against the
height h normalized to the solar radius R . The full lines correspond, from top to bottom, to
the wavelengths 0.37, 0.50, 0.80, 2.0, and 10.0 µm, respectively. The dashed line is obtained by
neglecting limb darkening. Data for limb darkening are from Allen (1973).
By means of a series expansion it can also be proved that for h R∗ one has
9 1 − u1 (ν) − u2 (ν) 20 − 5 u1 (ν) − 8 u2 (ν) h
wν ≈ wν 0 + 2 .
5 6 − 3 u (ν) − 4 u (ν) 2R∗
1 2
Finally, it can be proved by series expansion that for large values of h (h R∗ )
one gets the asymptotic behavior
3 15 − 7 u1 (ν) − 10 u2 (ν) 1
wν ≈ 1 − .
10 6 − 2 u1 (ν) − 3 u2 (ν) (1 + h/R∗ )2
For the solar case, the quantities u1 (ν) and u2 (ν) are tabulated by Allen (1973),
and the anisotropy factor can be computed via Eq. (12.36). Figure 12.7 shows the
behavior of wν as a function of height over the solar surface expressed in solar radii.
For a given height, the anisotropy factor is larger in the ultraviolet, consistently
with the results obtained in the previous section (cf. Fig. 12.4).
12.4. Symmetry-Breaking Effects
In the former sections of this chapter we have considered the idealized situation of
a plane-parallel, cylindrically symmetrical stellar atmosphere. The assumption of
cylindrical symmetry implies that the only non-zero components of the radiation
678 CHAPTER 12
field tensor are J00 (ν) and J02 (ν). However, such assumption is only approximately
satisfied in real stellar atmospheres, and there are many different phenomena that
may be responsible for symmetry breaking. In these cases, the radiation field can
no longer be described in terms of the mean intensity and anisotropy factor alone.
Further components of the radiation field tensor come into play, as obvious from
Eqs. (5.157) and Table 5.6. Restricting our analysis to ‘classical’ (or unpolarized,
see footnote on p. 663) atmospheres, the non-zero components of the radiation field
tensor are given by 5
dΩ K ,
K
JQ (ν) = T (0, Ω) I(ν, Ω)
4π Q
or, more explicitly, by
5
dΩ
J00 (ν) = I(ν, Ω)
4π
5
1 dΩ
J02 (ν) = √
(3 cos2 θ − 1) I(ν, Ω)
2 2 4π
√ 5
3 dΩ ± iχ
J±1 (ν) = ∓
2
sin θ cos θ e I(ν, Ω)
2 4π
√ 5
3 dΩ ± 2iχ
.
2
J±2 (ν) = sin2 θ e I(ν, Ω)
4 4π
Consider for instance the solar atmosphere. The phenomena of granulation and
supergranulation, or the presence of any structure typical of solar activity, such
as sunspots, faculae, etc.,1 introduce local symmetry breakings. Obviously, the
quantitative evaluation of the radiation field tensor cannot be carried out without
a detailed three-dimensional model of the structure considered and of its interface
with the surrounding atmosphere.
It is possible, however, to write down simple expressions for the radiation field
tensor at a given point P of the outer atmosphere under the assumption of a
‘point-like’ structure. Let us refer to Fig. 12.8, where the z-axis of the right-
handed, orthogonal frame (xyz) is directed along the vertical through P and the
x-axis points in an arbitrary direction. We suppose the spot to be centered at a
point of the solar surface characterized by the colatitude δ and the longitude ϕ in
the spherical coordinate system illustrated in the same figure. We also suppose
that the spot is seen under a small solid angle ∆Ω from point P and that ∆Iν is
the contrast, as seen from point P, of its intensity with respect to the unperturbed
atmosphere. Under these assumptions, the radiation field tensor at point P is given
by2
1
We recall again that we are concerned here with ‘classical’ atmospheres. In this context
typical symmetry-breaking agents, like magnetic fields, have no direct effect. An ‘active region’
is simply characterized by an intensity contrast with respect to the nearby regions.
2
Equations (12.38) are valid provided the spot is visible from point P, which implies
δ < arccos [R /(h + R )] .
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 679
Fig.12.8. Geometry for the evaluation of the radiation field tensor at point P in the presence of
a spot over the solar surface.
∆Ω
J00 (ν) = J00 (ν) cyl.s. + ∆Iν
4π
1 ∆Ω
J02 (ν) = J02 (ν) cyl.s. + √ (3 cos2 θ0 − 1) ∆Iν
2 2 4π
√
3 ± iχ0 ∆Ω
2
J±1 (ν) = ∓ sin θ0 cos θ0 e ∆Iν
√ 2 4π
3 ± 2iχ0 ∆Ω
2
J±2 (ν) = sin2 θ0 e ∆Iν , (12.38)
4 4π
where [J00 (ν)]cyl.s. and [J02 (ν)]cyl.s. are the components of the radiation field tensor
as computed in the cylindrically symmetrical case (see former sections), and where
the angles θ0 and χ0 are connected with the spot’s coordinates by the simple
relations
sin δ
χ0 = ϕ − π , θ0 = arctan .
1 − cos δ + h/R
Equations (12.38) can easily be generalized to the case where several structures
are present over the solar surface, and clearly show the role played by activity in
determining the radiation field tensor in the outer solar atmosphere.
680 CHAPTER 12
Fig.12.9. Schematic illustration of the radiation field symmetry breaking at point P due to the
presence of a prominence between P and the solar surface.
Another cause of symmetry breaking may be the presence of absorbing material,
like for instance a prominence, between the solar surface and the point P where the
radiation field tensor is to be evaluated. This physical situation is schematically
of the radiation propagating through
illustrated in Fig. 12.9. The intensity I(ν, Ω)
point P along the direction Ω is obviously modified by the presence of the interposed
material, but the quantitative evaluation of the radiation field tensor requires a
detailed physical and geometrical model of the material itself. An example of how
such calculation can be carried out when the absorbing material (the prominence)
is schematized as a homogeneous vertical slab of finite thickness can be found in
Landi Degl’Innocenti et al. (1987).
All the above-mentioned phenomena produce a loss of the symmetry characteris-
tics of the radiation field at a given point of the outer stellar atmosphere. Besides
them, there is an important symmetry-breaking phenomenon which is of a com-
pletely different nature, being connected with the Doppler effect. We go back to
the case of a cylindrically symmetrical stellar atmosphere, but we now consider
the radiation field as seen in a reference frame located at a point P of the outer
atmosphere and moving with velocity v with respect to the star. Such frame will
be referred to in the following as the comoving frame. Since the radiation coming
from different points of the stellar surface is differently affected by the Doppler
effect, if the spectrum presents an absorption (or emission) line there is an obvious
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 681
symmetry-breaking effect on the radiation field tensor. In quantitative terms, if
I(ν , Ω)
is the intensity of the radiation propagating through point P along the di-
rection Ω, as seen in a reference frame at rest with respect to the star, an observer
where, to first order in
moving with velocity v will experience an intensity I(ν, Ω),
1
v/c
, -
v · Ω , -
v · Ω
ν = ν 1 − , ν = ν 1 + . (12.39)
c c
The radiation field tensor in the comoving frame is thus given by
5
dΩ K
K
JQ (ν) = T (0, Ω) I(ν , Ω)
4π Q
5 , -
dΩ K v · Ω .
= T Q (0, Ω) I ν 1 + ,Ω (12.40)
4π c
Its symmetry properties are summarized in App. 17.
As an application of Eq. (12.40), we consider the case where the radiation coming
from the star is cylindrically symmetrical about the vertical and the stellar spec-
trum presents an absorption (or emission) line with a Gaussian profile. For the sake
of simplicity we also suppose that limb darkening is frequency-independent and de-
scribed by a linear function (N = 1 in Eq. (12.31)). Under these assumptions, the
intensity entering the integral in Eq. (12.40) can be written in the form
ν −ν 2
− 0
=I
I(ν , Ω) c
(0)
1 − dc e ∆νp
1 − u1 (1 − cos α) , (12.41)
where Ic(0) is the continuum intensity at disk center, dc is the central depression of
the line (dc < 1), ν0 and ∆νp are the line central frequency and width, respectively,
u1 is the limb-darkening coefficient, and α is the angle between the direction Ω and
the vertical to the atmosphere. The geometry is illustrated in Fig. 12.10: (xyz) is
the reference system (the comoving frame) which is moving with velocity v with
respect to the star; the velocity direction is specified by the angles θv and χv , the
direction Ω by the angles θ and χ ; the angle α is related to θ and to the relative
height h/R∗ of point P by Eqs. (12.33) and (12.32). Introducing these notations,
and recalling Eq. (12.39), the argument of the exponential in Eq. (12.41) takes the
form
1
The transformation between the two reference systems entails indeed further effects of order
v/c. One is aberration, which implies that a ray propagating along the direction Ω in the star’s
frame propagates, in the comoving frame, along a direction Ω given by Ω (1 + v · Ω
/c) − v /c
(to first order in v/c). The second is the fact that the intensity is multiplied by the factor
(ν/ν )3 ≈ 1 − 3(v · Ω)/c
when passing from the first to the second frame. Both effects are however
rather small, and negligible in most applications. Even considering a velocity v 100 km s−1 ,
aberration implies an angular distance between Ω and Ω of the order of 1 arcmin, whereas the
second effect implies a correction to the intensity of order 10 −3 .
682 CHAPTER 12
θ θ
χ χ
Fig.12.10. Geometry for the evaluation of the comoving-frame radiation field tensor at point P.
The reference system (xyz) – the comoving frame – moves with velocity v relative to the star.
The direction of v is specified by the angles θv and χv .
2 2
ν − ν0 ν − ν0 + ν (v/c) cos Θ
− =− , (12.42)
∆νp ∆νp
where, via the cosine theorem
v · Ω
cos Θ = = cos θ cos θv + sin θ sin θv cos(χ − χv ) . (12.43)
v
Substitution of Eqs. (12.41) and (12.42) into Eq. (12.40) leads to the following
expression for the comoving-frame radiation field tensor at frequency ν0
2π γ
I (0) 2 2
1 − d e−ω cos Θ
K
JQ (ν0 ) = c dχ dθ sin θ T Q
K
(0, Ω) c
4π
0 0
× 1 − u1 (1 − cos α) , (12.44)
where
ν0 v
ω= (12.45)
∆νp c
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 683
Fig.12.11. The radiation field tensor in the comoving frame, computed according to Eq.(12.44),
is plotted against the parameter ω defined in Eq.(12.45). Curves are labelled by the value of θv ,
while χv is set to 0◦ . The average intensity J00 (ν0 ) – upper panel – is normalized√
to the continuum
intensity at disk center, while J02 (ν0 ) – lower panel – is normalized to J00 (ν0 )/ 2. The relevant
parameters are: dc = +0.8 (absorption line), u1 = 0.2, h = 0.01 R∗ . Note in the upper panel the
Doppler brightening effect.
is the ratio between the Doppler shift induced by the velocity v and the width of
the spectral line (both expressed in frequency units).
It should be borne in mind that the key parameter in the Doppler effect is the
v · Ω
component of the velocity v along the propagation direction Ω, = v cos Θ. For
an assigned v , the angle Θ obviously depends both on θ and on χ (see Fig. 12.10
or Eq. (12.43)), and this entails the symmetry breaking: the radiation field at
point P is cylindrically symmetrical (χ-independent) in the star’s frame, but not in
the comoving frame. At the same time, Eqs. (12.44)-(12.45) show that this effect
vanishes for v → ∞ : under this limit the radiation field becomes cylindrically
symmetrical also in the comoving frame.
The integral in Eq. (12.44) can be evaluated numerically. Figures 12.11 and
12.12 show the various components of the radiation field tensor as functions of the
684 CHAPTER 12
Fig.12.12. Same as Fig.12.11 for the components J12 (ν0 ) and J22 (ν0 ). The remaining components
2 (ν ) = (−1)Q J 2 (ν ).
are obtained from J−Q 0 Q 0
parameter ω for χv = 0◦ and for different values of θv (with 0◦ ≤ θv ≤ 90◦ ). The
components of the radiation field tensor corresponding to χv = 0◦ and/or θv > 90◦
can easily be deduced from the symmetry properties proved in App. 17. Figures
12.11 and 12.12 are obtained by assuming for the central depression of the line the
value dc = 0.8, for the limb-darkening coefficient the value u1 = 0.2, and for the
height of point P over the stellar surface the value h = 0.01 R∗ . In the solar case,
these values are representative of a strong absorption line in the visible and a point
P at a height of approximately 10 arcsec over the solar limb. The two figures show
the strong sensitivity of the comoving-frame radiation field tensor to the velocity v .
Hereafter we briefly comment on the main features of the various curves.
The upper panel of Fig. 12.11 shows that the average intensity, J00 (ν0 ), increases
monotonically with increasing velocity. This is the well-known phenomenon of
Doppler brightening, which is an obvious consequence of the presence of an absorp-
tion line in the star’s spectrum: an observer moving with sufficiently large velocity
‘sees’, at frequency ν0 , the radiation emitted by the star in the far wings of the
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 685
Fig.12.13. Same as Fig.12.11 except for the line depression dc , which is set to −4 (emission line).
Note in the upper panel the Doppler dimming effect.
line, where the intensity is larger than in the line core.
The interpretation of the lower panel of Fig. 12.11, showing the behavior of the
anisotropy factor (see Eq. (12.1)), follows from similar arguments. Consider first
an observer moving in the upward direction (θv = 0◦ ). The radiation coming from
‘below’ (θ 0◦ ) is redshifted and therefore enhanced, while the same effect is
weaker for the radiation coming from the sides (θ γ). Thus the Doppler effect
acts in the same sense as geometrical and limb-darkening effects (see the former
sections). As a result, the anisotropy factor increases when v is increased. On the
contrary, for an observer moving horizontally (θv = 90◦ ) the maximum Doppler
shift (hence the maximum intensity enhancement) occurs for the radiation coming
from the sides – especially from the directions (θ ≈ γ, χ ≈ χv ) and (θ ≈ γ,
χ ≈ χv + 180◦ ).1 Now the roles of the Doppler effect and of geometrical/limb-
darkening effects are opposite, and the anisotropy factor is reduced compared to
1
Obviously, there is a blueshift for the former direction and a redshift for the latter, but this
makes no difference in the present context.
686 CHAPTER 12
Fig.12.14. Same as Fig.12.12 for an emission line with dc = −4.
the zero-velocity case. As expected, the Doppler effect becomes ineffective for very
large velocity, since in that case the observer sees, at frequency ν0 , the radiation
emitted by the star in the far wings of the line irrespective of the velocity direction.
Thus for ω → ∞ the anisotropy factor tends to its ω = 0 value,1 and this implies
that the curves in the lower panel of Fig. 12.11 have to go through a maximum or
a minimum. Actually, such extrema are found at ω values of about 1.5 or slightly
less.
Figure 12.12 shows the behavior of the other non-vanishing components of the
comoving-frame radiation field tensor. The interpretation of the different curves
follows from arguments similar to those developed above. In particular, it should
be remarked that J12 and J22 are zero both for ω = 0 and for ω → ∞, because in
either case the radiation field in the comoving frame is cylindrically symmetrical.
The situation is completely reversed for an emission line, as shown in Figs. 12.13
1
The coincidence of these two values is a consequence of the assumption of a frequency-
independent limb-darkening coefficient.
RADIATION ANISOTROPY IN STELLAR ATMOSPHERES 687
and 12.14. In this case one speaks of Doppler dimming effect (upper panel of
Fig. 12.13). The anisotropy factor turns out to be lowered – relative to its zero-
velocity value – for radial velocities (θv 0◦ ), and enhanced for transversal veloc-
ities (θv 90◦ ). Similarly, there is a sign switch for the components J12 andJ22 in
comparison with the absorption line case.
CHAPTER 13
ASTROPHYSICAL APPLICATIONS:
THE OUTER LAYERS OF STELLAR ATMOSPHERES
The theoretical results on atomic polarization derived in Chap. 10, and the study
performed in Chap. 12 on the properties of the radiation field flowing from a stellar
atmosphere allow one to deduce in a rather straightforward way the polarization
signatures of spectral lines formed in an optically thin plasma (like a prominence
or a coronal condensation) located at a certain height over the stellar surface. In
this chapter we present a number of applications of such results with the purpose,
on one side, of discussing the limits of applicability of the theory developed in
this book, and, on the other side, of showing the important role often played by
the Doppler effect in determining the shape of polarimetric profiles. The methods
illustrated in this chapter can be considered as the basic tools for the interpretation
of spectropolarimetric observations in solar prominences or in the solar corona – as
long as the hypothesis of optical thinness is verified – and for the diagnostics of
prominence and/or coronal magnetic fields.
13.1. The Flat-Spectrum Approximation
The flat-spectrum approximation has been introduced and widely discussed in
Chap. 6. This approximation is at the basis of the statistical equilibrium equa-
tions that have been derived in Chap. 7 and solved in Chap. 10 for atomic systems
of increasing complexity. The astrophysical applications that we are going to per-
form in the present chapter heavily rely on the results obtained in Chaps. 7 and
10, so that it is now necessary to discuss in some detail the limitations imposed by
the flat-spectrum approximation on the description of resonance polarization and
the Hanle effect.
Suppose we wish to investigate the polarization properties of a spectral line of
a given atom, located at a point P in the outer atmosphere of a star and moving
with velocity v with respect to the star. The spectral line can be a resonance line,
originating from the transition between the ground level and an excited level, or,
more generally, a subordinate line due to the transition between two excited levels.
The atom is illuminated by the radiation field flowing from the stellar atmosphere
which affects, together with collisions and with a magnetic field, possibly present at
point P, its physical state. Such state can be determined by introducing a suitable
model, capable of describing the atomic structure and the radiative and collisional
connections among the various atomic levels, and by solving the statistical equilib-
rium equations for the different levels (or terms). The model involves a number of
electric-dipole transitions between the various levels and, for each transition, the
690 CHAPTER 13
flat-spectrum approximation has to be satisfied.
As far as the radiation field is concerned, there are two possibilities: either the
stellar spectrum presents an absorption (or emission) line at (or in the immediate
neighborhood of) the transition frequency, or such frequency falls into a continuum
window of the stellar spectrum. In the former case, denoting by ∆νp the line width,
the spectrum can be considered flat only within a frequency interval ∆ν ∆νp . In
the latter, if ∆νw is the width of the continuum window, the spectrum is actually
flat within an interval ∆ν ≤ ∆νw . In order for the flat-spectrum approximation
to hold, the interval ∆ν must be larger than the natural width of the levels, and,
when coherences between non-degenerate levels are involved, it must be larger than
the corresponding Bohr frequencies (cf. Chap. 6).
As to the natural width requirement, it should be remarked that stellar spectra
have, as a rule, rather broad absorption (or emission) lines, the typical widths being
two or more orders of magnitude larger than the typical natural width of a level.
To make an example, a medium-strength line of the solar spectrum at 5000 Å has
a typical full width at half maximum of the order of 100 mÅ, which corresponds to
∆νp 1.2 × 1010 s−1 . This is to be compared to a characteristic natural width,
which is of the order or less than 108 s−1 .
The only limitation to the applicability of the flat-spectrum approximation thus
comes from coherences between non-degenerate levels, whose Bohr frequencies have
to be sufficiently small. In the case of a multi-level atom without hyperfine struc-
ture, such coherences are those between magnetic sublevels split by a magnetic
field. Since the frequency separation between two different sublevels is of the or-
der of νL , the flat-spectrum approximation is verified when νL ∆νp in the first
of the two cases outlined above (presence of an absorption or emission line), and
when νL ∆νw in the second case (transition falling into a continuum window).
Referring to the previous example of a line with ∆νp 1.2 × 1010 s−1 , and recalling
the definition of νL (see Eq. (3.11)), it follows that the magnetic field has to satisfy
the condition B 0.9 × 104 G; such value can be much larger in the second case.
Similar limitations apply to fine-structured or hyperfine-structured atomic sys-
tems. In these cases coherences may be present between J-levels belonging to
the same term, or between F -levels belonging to the same level, respectively. Re-
calling the definitions of the quantities ζ and A given in Eqs. (3.59) and (3.70),
respectively, we see that the following inequalities must be satisfied in order for the
flat-spectrum approximation to hold
ζ A
∆νp , ∆νp (13.1)
h h
in the case of an absorption or emission line, and
ζ A
∆νw , ∆νw (13.2)
h h
in the case of a continuum window.
It should also be considered that, because of a general property of the statistical
equilibrium equations (see the introduction to Chap. 7), coherences between non-
degenerate levels are the smaller, the larger the energy separation between the
THE OUTER LAYERS OF STELLAR ATMOSPHERES 691
levels. The ratio of such coherences to level populations is roughly the same as
the ratio of the natural width of the level to the Bohr frequency of the coherence.
Therefore, if the Bohr frequency of one of the coherences is ‘large’ (comparable with
∆νp or ∆νw , so that the flat-spectrum approximation is not valid), the coherence
itself is expected to be much smaller that the diagonal elements of the density
matrix (by two orders of magnitude in the example above). Moreover, as illustrated
in Sect. 10.17, such coherences are important in affecting the polarization properties
of the emissivity only in the far wings of the spectral line, where the emissivity itself
is very low in all four Stokes parameters and is often masked by other processes,
like, e.g., the continuum emissivity. These arguments show that, if we exclude some
few special cases involving particularly strong lines with well-developed damping
wings formed in optically thick media, it is usually justified to consider a simplified
atomic model where such coherences are set to zero ab initio. This approximation is
particularly suited for the applications that we are going to develop in this chapter,
which refer to optically thin plasmas.
To conclude, we want to stress that the flat-spectrum approximation in polarized
radiative transfer is analogous to the complete redistribution approximation (in the
atomic frame) in standard radiative transfer. In the latter, only few lines need to
be treated with the more sophisticated theory of partial redistribution; similarly,
in polarized radiative transfer most of the applications to concrete problems can be
handled with the simple approach of the flat-spectrum approximation. Only in few
cases more sophisticated (and more complicated) formalisms have to be invoked
(Bommier, 1997a,b; Landi Degl’Innocenti et al., 1997).
13.2. Velocity/Density-Matrix Correlations
and the Approximation of Complete Redistribution on Velocities
As shown in Sect. 12.4, the radiation field tensor experienced by an observer moving
with velocity v in the outer layers of a stellar atmosphere may strongly depend,
because of the Doppler effect, on v . As a consequence, it has to be expected that
the density matrix of an atomic system moving in the same environment may be
strongly correlated with its velocity.
Given this situation, a statistical description of the atomic system has to be given
in terms of the product of two functions: the velocity distribution function f (v ),
defined such that f (v ) d3v , with
f (v ) d3v = 1 , (13.3)
is the probability of finding an atom of the given species with velocity contained in
the infinitesimal volume d3v of the velocity space, times ρnm (v ), the v -dependent
density matrix, here defined on the basis of the energy eigenvectors. With these
notations, the quantity
dP = f (v ) ρnn (v ) d3v
692 CHAPTER 13
gives, for instance, the probability of finding the atom with velocity v and in the
internal state specified by the eigenvector | n . In the following, we will refer to
products of the form f (v ) ρnm (v ), or – when using the spherical tensor represen-
tation – f (v ) ρK v ) or f (v ) βLS ρK
Q (αJ; Q (J, J ;
v ), with the name of velocity-space
density matrix .
We will now derive a statistical equilibrium equation for the velocity-space density
matrix. To this aim, it should be taken into account that all the processes that we
have considered so far in establishing the time evolution of the density matrix, such
as absorption and emission of photons, inelastic and superelastic collisions with
electrons, and depolarizing collisions with perturbers (mainly neutral hydrogen
atoms), are practically ineffective in changing the velocity of the atom. Consider,
for instance, the emission or the absorption of a photon of visible wavelength by
an atom which is moving with velocity v . The conservation of momentum implies
that the atom experiences a recoil corresponding to a velocity variation ∆v given
by
h
∆v ≈ ,
λ µ mH
where λ is the photon’s wavelength and µ is the atomic weight of the atom. For
example, assuming λ = 5000 Å and µ = 56 (the atomic weight of iron), one gets
∆v 1.4 cm s−1 ,
which should be compared with velocities of the order of some km s−1 that are
typically found in stellar atmospheres. Somewhat more efficient from this point of
view are collisions with thermal electrons, either elastic, inelastic, or superelastic.
If ve is the typical velocity of an electron, the recoil suffered by the atom is given,
in terms of velocity variation, by
m ve
∆v ≈ .
µ mH
Assuming for v and ve the r.m.s. values corresponding to the same temperature,
one has .
ve µ mH
= ,
v m
so that .
∆v m
≈ .
v µ mH
For µ = 56 one gets, for instance
∆v
3.1 × 10−3 .
v
Finally, depolarizing collisions are elastic collisions due to long-range interactions,
and they are also ineffective in changing appreciably the velocity of the atom.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 693
The only processes that are important, under this point of view, are close col-
lisions with atoms. Such collisions, that can be referred to as velocity-changing
collisions, are characterized by very small impact parameters and by a relatively
large exchange of kinetic energy between the colliding atoms. It has then to be
expected that such collisions will also induce transitions between the energy levels
(or sublevels) of the atom, thus affecting its density matrix.
Referring for simplicity to the case of a multi-level atom in the statistical tensor
representation, and taking into account the foregoing considerations, we can write
the statistical equilibrium equation for the velocity-space density matrix in the
form
d d K δ
f (v ) ρK
Q (αJ;
v ) = f (
v ) ρ (αJ;
v ) + f (
v ) ρ K
(αJ;
v ) . (13.4)
dt dt Q 0 δt Q
v.c.c.
The first term in the right-hand side is due to the ‘ordinary’ processes mentioned
above – which are ineffective, to a very good approximation, in changing the ve-
locity of the atom. In particular, the quantity
d K
ρ (αJ; v )
dt Q 0
is the sum of the right-hand side of Eq. (7.11) – radiative processes – and the
right-hand side of Eq. (7.101) – collisional processes. It should be remarked that
the radiative rates TA , TS , RA , and RS (which are proportional to the radiation
field tensor) depend in general on the velocity v .
The second term in the right-hand side of Eq. (13.4) is due to velocity-changing
collisions, and can be regarded as a generalization of the Boltzmann term which is
met in the usual kinetic theory of gases (see e.g. Oxenius, 1986). Neglecting the
internal state of the colliders, such term – that will be referred to in the following
as the generalized Boltzmann term – can be written in the form
δ K
f (v ) ρQ (αJ; v ) =
δt v.c.c.
= f (v ) ρK
v ) Fi (u )
Q (α J ;
i α J K Q
× qi (v , u → v , u ; α J K Q → αJKQ) w d3 u
− f (v ) ρK
Q (αJ;
v)
× Fi (u ) qi (v , u → v , u ; αJKQ → α J K Q ) w d3 u . (13.5)
i α J K Q
In this equation, Fi (u ) is the distribution function of particles of species i, normal-
ized to the number density ni of the same particles,
ni = Fi (u ) d3 u ;
694 CHAPTER 13
qi (v , u → v , u ; αJKQ → α J K Q ) is the cross section for the collision in which
the atom goes from velocity v to velocity v (changing at the same time its internal
state from level αJ to level α J and from multipole KQ to multipole K Q )
whereas the collider – of species i – goes from velocity u to velocity u ; w and w
are the moduli of the relative velocities defined by w = v − u .1
= v − u, w
Owing to the presence of the generalized Boltzmann term, Eq. (13.4) is in fact
extremely complicated. Moreover, little is known, both from the experimental
and from the theoretical side, on the cross sections qi entering Eq. (13.5). For
these reasons, two different approximations (or working hypotheses) are generally
introduced to bring back the problem to a more tractable form.
The first one, that can be referred to as the velocity-coherence approximation,
consists in simply neglecting the generalized Boltzmann term in Eq. (13.4). This is
justified only when the number density of perturbers (typically hydrogen atoms or
ions) is sufficiently low, say, less than a critical value nc whose order of magnitude
can be estimated through the following considerations.
The order of magnitude of the rates for velocity-changing collisions is given by
n q v, where n is the number density of perturbers, q is the cross section for velocity-
changing collisions, and v is the average velocity of perturbers relative to the atom.
Such a rate has to be compared with the other rates appearing in the first term
of the right-hand side of Eq. (13.4), and, more specifically,2 with the Einstein
coefficient for spontaneous de-excitation, A. This leads to the following equation
for the critical density
nc q v ≈ A .
Let us take for v the average relative velocity of hydrogen atoms corresponding to
a Maxwellian distribution at temperature T ,
8kB T 1
v = 1+ .
π mH µ
Neglecting the factor 1/µ, and expressing q in units of πa20 (a0 being the Bohr
radius), A in units of 107 s−1 , and T in units of 104 K, one gets, numerically
A
nc 7.8 × 1016 √ cm−3 . (13.6)
q T
This expression, joined with the fact that values of q (in units of πa20 ) are rarely
larger that 101 -102 , leads to the conclusion that in the outer layers of stellar at-
mospheres the approximation of neglecting the generalized Boltzmann term in
Eq. (13.4) is fully justified.3
1
Equation (13.5) is a heuristic equation based on the possibility of defining a cross section
such as qi (v , u → v ,
u ; αJKQ → α J K Q ).
2
Excluding particular cases involving anisotropic velocity distributions, velocity-changing
collisions are responsible for a relaxation mechanism of the density matrix. For this reason, the
comparison has to be made with the largest of the relaxation rates, which, in many cases, is the
spontaneous de-excitation rate.
3
In solar prominences, for instance, the typical number density of hydrogen is of the order
of 1011 cm−3 .
THE OUTER LAYERS OF STELLAR ATMOSPHERES 695
Under this approximation, Eq. (13.4) reduces to
d K d K
f (v ) ρQ (αJ; v ) = f (v ) ρ (αJ; v ) ,
dt dt Q 0
which is solved by
d
f (v) = 0
dt
d K d K
ρ (αJ; v ) = ρ (αJ; v ) .
dt Q dt Q 0
These equations show that the density matrix of an atom moving with velocity v
is completely decoupled from the density matrix of atoms moving with different
velocities, and that its value can be found by solving the statistical equilibrium
equations of Chap. 7: obviously, the Doppler effect must be taken into account in
the expressions for the radiative rates involving the radiation field tensor. On the
other hand, the velocity distribution f (v ) remains undetermined, and can only be
established by means of different physical considerations. In many cases, f (v ) can
be simply assumed to be a Maxwellian distribution, possibly centered at a non-zero
velocity, like in the case of the solar wind.1
It should be remarked that, when the velocity-coherence approximation is valid,
the v -dependence of the density matrix is only due to the fact that atoms moving
with different velocities may experience, because of the Doppler effect, different
values of the radiation field tensor. Obviously, when the incident radiation field
presents no spectral structure across each of the frequency intervals centered at
the transition frequencies of the model atom and sufficiently wide to encompass all
possible Doppler shifts, the v -dependence of the density matrix disappears and one
is left with a ‘unique’ density matrix ρK Q (αJ).
The second approximation that can be introduced to bring back Eq. (13.4) to
a more tractable form is the so-called approximation of complete redistribution on
velocities. Here one assumes a priori that velocity-changing collisions are so efficient
in reshuffling the atomic velocities that any velocity/density-matrix correlation
is lost. This implies that the velocity-space density matrix can be written as
f (v ) ρK
Q (αJ), the second factor being independent of
v , which in turn implies that
the term
d K
ρ (αJ; v )
dt Q 0
appearing in Eq. (13.4) reduces to the sum of products of velocity-dependent
rates times velocity-independent density-matrix elements. It is then convenient
to integrate Eq. (13.4) in d3v . Disregarding, for the time being, the generalized
Boltzmann term, and recalling the normalization of the velocity distribution (see
Eq. (13.3)), one finds for ρKQ (αJ) the ‘ordinary’ statistical equilibrium equation
1
In certain situations, bi-modal or three-modal distributions can be more appropriate.
696 CHAPTER 13
with the velocity-dependent rates replaced by their averages over the velocity dis-
tribution f (v ). Synthetically,
Rate (v ) → f (v ) Rate (v ) d3v . (13.7)
Since the rates depend on v only through the radiation field tensor, it is easily seen,
recalling Eqs. (5.157) and the formulae for the non-relativistic Doppler effect, that
substitution (13.7) is equivalent to
K
K
JQ (ν) → J¯Q
K
(ν) = d3v f (v ) JQ (ν) v , (13.8)
K
where [JQ (ν)]v , the radiation field tensor in the comoving frame, is given by (cf.
Eq. (12.40))
5 , -
dΩ K
3
v · Ω .
K
JQ (ν) v = T Q (i, Ω) Ii ν 1 + ,Ω (13.9)
4π i=0 c
In particular, for an isotropic distribution of velocities, denoting by fc (v) the dis-
tribution (normalized to unity) of the component of the velocity along an arbitrary
direction, one has
∞ 5 ,
dΩ K v-
3
J¯Q
K
(ν) = dv fc (v) T Q (i, Ω) Ii ν 1 + ,Ω ,
4π i=0 c
−∞
or, via the new variable ν = ν (1 + v/c),
∞ , ν − ν - 5 dΩ 3
c I (ν , Ω)
J¯Q
K
(ν) = dν fc c T K (i, Ω)
ν ν 4π i=0 Q i
−∞
∞
= dν p(ν − ν) JQ
K
(ν ) , (13.10)
−∞
where
c , ν − ν -
p(ν − ν) = f c .
ν c ν
For an isotropic distribution of velocities, substitution (13.7) is thus equivalent
to the substitution of the radiation field tensor with a suitable convolution over
frequencies. In the case of a Maxwellian distribution, where fc (v) is given by (cf.
Eq. (5.42)) 2
1 1 − vv
fc (v) = √ e T ,
π vT
THE OUTER LAYERS OF STELLAR ATMOSPHERES 697
with vT the thermal velocity, the convolution profile is given by
ν −ν 2
1 1 −
p(ν − ν) = √ e ∆νD
, (13.11)
π ∆νD
where
vT
∆νD = ν .
c
Up to now, the integral in d3v of the generalized Boltzmann term has been
disregarded. An inspection of the right-hand side of Eq. (13.5), joined with the
hypothesis of the velocity independence of the density matrix, shows that such term
leads to the appearance in the statistical equilibrium equations of two additional
contributions that can be written in the form
+ BT (α J K Q → αJKQ) ρK
Q (α J )
α J K Q
− BR (αJKQ → α J K Q ) ρK
Q (αJ) , (13.12)
α J K Q
where
BT (α J K Q → αJKQ) =
= d3v d3 u f (v ) Fi (u ) qi (v , u → v , u ; α J K Q → αJKQ) w
i
BR (αJKQ → α J K Q ) =
= d3v d3 u f (v ) Fi (u ) qi (v , u → v , u ; αJKQ → α J K Q ) w .
i
In particular, when the velocity distribution of each species of colliders is isotropic,
the considerations developed in Sect. 7.13 can be applied to these terms, so that
the contribution of Eq. (13.12) can be cast into a form similar to the right-hand
side of Eq. (7.101).
It should be pointed out that the approximation of complete redistribution on
velocities can in principle be applied only when the generalized Boltzmann term
predominates over the other terms appearing in Eq. (13.4), or, in other words, when
the number density of colliders is much larger that the critical value nc defined in
Eq. (13.6). A large number density of colliders implies, however, large depolarizing
rates, due both to long-range interactions and to velocity-changing collisions: as
a consequence, when the approximation of complete redistribution on velocities
is fully justified, polarization phenomena turn out, in general, to be of marginal
importance. Actually, an ‘intermediate’ approach is often used where the gener-
alized Boltzmann term is neglected, still assuming a velocity-independent density
matrix and using averaged rates according to Eq. (13.7). This obviously leads to
698 CHAPTER 13
major simplifications, because the atomic system is described by a ‘unique’ density
matrix and because the poorly known cross sections appearing in the generalized
Boltzmann term are no longer involved. Such approach, though incorrect in prin-
ciple, is adopted in order to get an approximate solution to a problem that would
otherwise be insoluble in practice. Hence its importance in radiative transfer for
polarized radiation, particularly for the solution of non-LTE problems.
13.3. Resonance Polarization and the Hanle Effect
in the Absence of Velocity/Density-Matrix Correlations
As shown in the former section, there is a special case where the density matrix
does not depend on the velocity of the atom. This happens when the radiation field
impinging on the atom is spectrally unstructured at all the transition frequencies
that contribute to establish its statistical equilibrium.
Whereas this situation is common in laboratory experiments, it is much less
frequent in astrophysical plasmas. However, some of the spectral lines that are
currently used for the diagnostics of the outer layers of the solar atmosphere fall
within this category. A typical example is the so-called coronal green line of FeXIV
at 5304 Å. The solar spectrum shows no prominent lines in the neighborhood of
this wavelength, which implies that the excitation of the ions is due to the solar
continuum and is practically independent of the Doppler effect.
In this section we illustrate the main features of the phenomena of resonance
polarization and the Hanle effect for this simple case where it is fully justified to
neglect velocity/density-matrix correlations, so that a unique, velocity-independent
density matrix can be defined.
Referring to Fig. 13.1, we consider an atom (described by a suitable model as
outlined in Sect. 13.1) located at point P in the outer solar atmosphere. We assume
the incident radiation field at P to be unpolarized, cylindrically symmetrical about
the vertical through P, and spectrally unstructured in the neighborhood of each
of the frequencies νi corresponding to the transitions involved in the model atom.
The observer, receiving the radiation along the direction Ω , sees the atom at an
0
apparent height h over the solar limb, the true height h over the solar surface
being related to h and to the aspect angle1 δ by the simple equation
(h + R ) cos δ = h + R ,
where R is the solar radius. From the knowledge of h and of the limb-darkening
coefficients of the solar spectrum, the radiation field tensor can easily be derived at
each of the frequencies νi . Assuming a quadratic limb-darkening law, and drawing
on the results derived in Sect. 12.3, we can write the non-zero components of
1
The determination of the angle δ is often non-trivial and, in some cases, even impossible.
For prominences, one can obtain an estimate of δ by measuring the position of the corresponding
filament over the disk during the few days preceding (or following) the observation.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 699
χ
θ
Fig.13.1. Geometry of a scattering process in the outer layers of the solar atmosphere. The
points to the observer, while the direction e – the reference direction for positive
direction Ω 0 a
Q – is parallel to the solar limb (as seen by the observer). θB and χB are the polar angles of the
magnetic field vector relative to the local vertical.
the radiation field tensor in a reference system with the z-axis directed along the
vertical through P in the form
1
J00 (νi ) = Jν , J02 (νi ) = √ wν Jν , (13.13)
i i i
2
where Jν and wν are given in Eqs. (12.34) and (12.36), respectively.
i i
The radiation field tensor is just one of the factors that have to be specified to find
the density-matrix components of the atom and, ultimately, the Stokes parameters
of the radiation emitted along the direction Ω . The remaining factors are the
0
magnetic field, that can be parameterized through its modulus, B, and the angles
θB and χB as in Fig. 13.1, and the collisional cross sections. Here we are mainly
interested in the basic features of resonance polarization and the Hanle effect in
the outer layers of the solar atmosphere, so we will just suppose that, owing to
the low density of the plasma, collisions can be neglected. Obviously, when using
a specific spectral line as a diagnostic tool for inferring – for instance – the value
of the magnetic field, this approximation has to be carefully tested, case by case.
For example, detailed analyses have shown that such approximation is justified for
the HeI λ 10830 and λ 5876 lines, whereas it is not for the hydrogen lines of the
Balmer series and for the magnetic-dipole forbidden lines that are commonly used
for the diagnostics of coronal fields (FeXIV λ 5304, FeXIII λ 10747 and λ 10798,
FeX λ 6374).1
1
For the hydrogen lines, collisions with electrons (and protons) turn out to be important
700 CHAPTER 13
Having specified all the geometrical and physical parameters, one can turn to the
problem of finding the density matrix of the radiating atom. This is obtained by
solving (for stationary situations) the statistical equilibrium equations that have
been derived in Chap. 7. The relevant equations are Eq. (7.11) for the case of a
multi-level atom, Eq. (7.38) for a multi-term atom, or Eq. (7.65) for a multi-level
atom with hyperfine structure. It should be remarked that all three equations are
valid in a reference system with the z-axis directed along the magnetic field. The
radiative rates appearing in these equations obviously contain the radiation field
tensor evaluated in that reference system, while Eqs. (13.13) refer to a system with
the z-axis in the vertical direction. The two systems are related by a suitable
rotation. Taking into account Eq. (2.78) one has, with evident notations1
K
JQ (νi ) mag.field = D0Q
K
(R) J0K (νi ) vert. , (13.14)
where (cf. Fig. 13.1)
R ≡ χB , θB , γB . (13.15)
Equation (13.14) shows that the radiation field tensor in the magnetic frame is
unaffected by the angle χB : this is due to the assumption of an unpolarized and
cylindrically symmetrical radiation field. The angle γB can be chosen arbitrarily,
in the sense that all the reference frames corresponding to different γB values have
the z-axis in the magnetic field direction. However, by fixing a γB value, a specific
frame is chosen, so that the atomic density matrix is defined in that frame. This
should be kept in mind when expressing the emissivity in the observer’s frame (see
later). In any case, the polarization of the emitted radiation is independent of γB .
The solution to the statistical equilibrium equations yields the statistical tensors
of all the levels (or terms) involved in the model atom considered, and, in particular,
those of the upper level (or term) of the transition under study. This allows one
to evaluate the emission coefficient, εi (ν, Ω ), which is given by Eq. (7.15e) – or
0
Eq. (10.39) – for the multi-level atom, by Eq. (7.47e) for the multi-term atom, or
by Eq. (7.70e) for the multi-level atom with hyperfine structure.2 It should be
recalled, however, that such expressions are valid in the rest frame of the atom.
For a collection of atoms having a velocity distribution f (v ), the expression for
) must be transformed to take the Doppler effect into account. This is
εi (ν, Ω 0
easily done by substituting each of the complex profiles of the form Φ(ν0 − ν) with
the convolution
, -
v · Ω
Φ(ν0 − ν) → Φ̂(ν0 − ν) = d3v f (v ) Φ ν0 1 + 0
−ν , (13.16)
c
because of the quasi-degeneracy of the energy levels with respect to the quantum number l. For
coronal lines, collisional rates are comparable to radiative rates because of the low values of the
Einstein coefficients (A 101 -102 s−1 ).
1
Alternatively, one can use the equations written in the vertical frame, with the difference
that the magnetic term takes a more complicated form, as explained in Sect. 7.12.
2
Note that this last equation is valid under the limit where all the hyperfine components can
be considered coincident.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 701
which, in the case of a Maxwellian distribution of velocities, implies the appear-
ance of Voigt functions and of the associated dispersion profiles (or Faraday-Voigt
functions; cf. Sect. 5.4).
Next, we have to evaluate the geometrical tensor T Q K ) appearing in the
(i, Ω 0
expression of the emission coefficient for the geometry of Fig. 13.1. Recalling
Eq. (5.159), one has
TQ )=
P (i) DP Q (R) ,
K
(i, Ω 0 tK K
(13.17)
P
) into the system
where R, the rotation carrying the reference system (ea , eb , Ω 0
having the z-axis directed along the magnetic field, is given by
R ≡ −90◦ , −90◦ + δ , 0◦ × χB , θB , γB , (13.18)
where the value of the angle γB is the same as in Eq. (13.15). We recall that the
rotation matrix DPKQ (R) for the composite rotation of Eq. (13.18) can be computed
with the help of Eq. (2.74).
The Stokes parameters of the radiation flowing along the direction Ω can be
0
finally obtained by integration of the radiative transfer equations. Under the hy-
pothesis of an optically thin plasma, one has
) = ε (ν, Ω
Ii (ν, Ω ) ds , (13.19)
0 i 0
where s is the coordinate measured along the direction Ω .
0
The procedure that we have outlined above can be followed for any model atom,
provided the flat-spectrum approximation is verified. According to our discussion
of Sect. 13.1, since we are considering the case where all the transition frequencies
of the model atom fall within windows of the continuous solar spectrum, the only
limitations are those imposed by inequalities (13.2). Obviously, the validity of such
conditions has to be carefully checked before applying the present theory to specific
model atoms.
We will now illustrate the basic features of resonance polarization and the Hanle
effect in the outer solar atmosphere for a particularly simple model atom, i.e., a
two-level atom with J = 0, Ju = 1. This simple model avoids the need of finding a
numerical solution to the statistical equilibrium equations since the lower level is,
by definition, unpolarized. We will also neglect stimulated emission, which is a good
approximation for lines at optical wavelengths. The general problem of resonance
scattering and the Hanle effect for two-level atoms has already been considered in
Sects. 10.2-10.4, so we have just to adapt the equations there obtained to the case
of the solar atmosphere. In particular, we can draw the expression for the emission
coefficient directly from Eq. (10.41), that we can here rewrite, via Eqs. (5.157), in
the form1
1
Note that the contribution to Eq. (13.20) from K = 1 is zero (see Eqs. (13.13)-(13.14)), and
(0) (2)
that w10 = w10 = 1 (see Eq. (10.14) and Table 10.1).
702 CHAPTER 13
) = kA
εi (ν, Ω Φ̂KK (J = 0, Ju = 1 ; ν)
0 L Q
KK Q
) 1
× (−1)Q T Q
K
(i, Ω J K (ν ) , (13.20)
0
1 + i QHu −Q 0
where kLA is the frequency-integrated line absorption coefficient defined in Eq. (9.5),
Φ̂KK
Q is the generalized profile defined in Eq. (10.40)1 – with the individual profiles
Φ(νi − ν) replaced by their Doppler convolutions as in Eq. (13.16) – , Hu is given
by Eq. (10.29), T Q K ) by Eq. (13.17), and J K (ν ), the radiation field tensor
(i, Ω 0 Q 0
at the transition frequency ν0 defined in the magnetic field frame, is given by
Eq. (13.14). Similarly, for the frequency-integrated emission coefficient one can
rewrite Eq. (10.32) in the form
1
) = kA
ε̃i (Ω (−1)Q T Q
K )
(i, Ω J K (ν ) . (13.21)
0 L 0
1 + i QHu −Q 0
KQ
Substituting Eq. (13.20) into Eq. (13.19), and assuming a small-size coronal conden-
sation permeated by a uniform magnetic field, one obtains for the Stokes parameters
of the radiation reaching the observer the same expression as the right-hand side
of Eq. (13.20) with the quantity kLA replaced by the frequency-integrated optical
thickness τL defined by
τL = kLA ds . (13.22)
Similarly, for the frequency-integrated Stokes parameters one obtains the same
expression as the right-hand side of Eq. (13.21) with kLA replaced by τL .
In spite of the simplicity of the model atom considered, the expressions just
derived implicitly depend on several parameters (limb-darkening coefficients of the
continuous solar radiation, height and aspect angle of the coronal condensation,
intensity and direction of the magnetic field, Einstein coefficient of the line, velocity
distribution of the scattering atoms). A full analysis of the interplay of the different
parameters is well outside the purposes of this book. In the following we limit
ourselves to present a number of illustrative results for a few special cases.
a) Resonance polarization
In the absence of magnetic fields, the expressions of the emitted Stokes parameters
considerably simplify, since Hu = 0 and it is not necessary to perform the rotation
(13.15) to change the reference system. On the other hand, the generalized profiles
simplify according to Eq. (A13.5). With the help of Table 5.6 one obtains for the
non-zero Stokes parameters
) = τ J 0 (ν ) + √ 1
I(ν, Ω 0 L 0 0 3 sin2 δ − 1 J02 (ν0 ) φ̂(ν)
2 2
3
) = τ √ cos δ J 2 (ν ) φ̂(ν) ,
2
Q(ν, Ω 0 L 0 0 (13.23)
2 2
1
The generalized profiles for the special case J = 0, Ju = 1 are explicitly given in App. 13
(see Eqs. (A13.11)).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 703
) is plotted as a function of the projected height
Fig.13.2. The fractional linear polarization pQ (Ω 0
h (normalized to the solar radius) for different values of the angle δ in the geometry of Fig.13.1.
The figure refers to a two-level atom (J = 0, Ju = 1) illuminated by the continuous solar radiation
at 5000 Å in the absence of magnetic fields.
K
where the tensor JQ (ν0 ) is evaluated in the vertical frame, and where the real
profile φ̂(ν), which takes the Doppler effect due to the velocity distribution of
atoms into account, is given by
, -
v · Ω
φ̂(ν) = d3v f (v ) φ ν0 1 + 0
−ν . (13.24)
c
Equations (13.23) show that the ratio pQ = Q(ν, Ω )/I(ν, Ω
) is frequency-inde-
0 0
pendent and equal to the ratio between the frequency-integrated Stokes profiles.
Figure 13.2 shows the behavior of pQ as a function of the projected height h for
several values of δ. The limb-darkening coefficients are assumed to be u1 (ν0 ) = 0.95,
u2 (ν0 ) = −0.20, corresponding to the wavelength λ = 5000 Å (Allen, 1973). Such
curves, when computed for realistic model atoms, can be very useful for inferring, by
comparison with the observed fractional linear polarization, the possible presence
of depolarizing mechanisms and, in particular, the presence of magnetic fields (a
typical example, referring to the D3 line of HeI, can be found in Leroy et al., 1977).
b) The Hanle effect (polarization diagrams)
In the presence of a magnetic field, the polarization of the radiation scattered along
is deeply modified by the Hanle effect. This phenomenon can be
the direction Ω 0
illustrated by drawing suitable polarization diagrams for the quantities p̃Q and p̃U
704 CHAPTER 13
Fig.13.3. Polarization diagram for the radiation emitted along the direction Ω in the geometry
0
of Fig.13.1. The graph refers to a two-level atom (J = 0, Ju = 1) located in the plane of the sky
(δ = 0◦ ) and illuminated by the continuous solar radiation at 5000 Å in the presence of a horizontal
magnetic field (θB = 90◦ ). Full lines correspond to χB = const., while broken lines correspond
to constant magnetic field strength, parameterized through the quantity Hu (see Eq.(10.29)).
defined by
) dν
Q(ν, Ω ) dν
U (ν, Ω
0 0
p̃Q = , p̃U = .
) dν
I(ν, Ω ) dν
I(ν, Ω
0 0
As apparent from Eq. (13.21), the corresponding quantity p̃V vanishes identically,
K
because the only non-zero components of JQ (ν0 ) are those with K = 0 or K = 2,
and the components T 0 (3, Ω0 ) and T Q (3, Ω0 ) are zero (see Table 5.6).
0 2
Traditionally, such polarization diagrams are obtained by fixing the value of θB
to 90◦ and by drawing in the plane p̃U - p̃Q the curves corresponding to χB = const.
(for variable Hu ) and to Hu = const. (for variable χB ).1 Two examples (drawn on
the same scale) are shown in Figs. 13.3 and 13.4. Both figures are obtained for
a projected height h /R = 0.073 (corresponding to about 70 arcsec) and for the
same limb-darkening coefficients as in point a). The diagram of Fig. 13.3, which
1
It should be recalled that such diagrams have been introduced for the diagnostics of magnetic
fields in prominences (Bommier, 1977), and that it is generally believed that in these objects the
magnetic field is horizontal in order to sustain the prominence material against gravity.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 705
Fig.13.4. Same as Fig.13.3 for a different aspect angle (δ = 30◦ ). The solid lines labelled with 1,
2, . . . , 12 correspond to χB = 0◦ , 30◦ , . . . , 330◦ , respectively. The broken lines correspond to
the same Hu values as in Fig.13.3.
refers to a coronal condensation observed in the plane of the sky (δ = 0◦ ), shows
an obvious symmetry. A given point in the diagram corresponds to a single value
of the magnetic field intensity and to two opposite values of the angle χB , whose
sign is left undetermined. This simple symmetry is clearly lost in the diagram of
Fig. 13.4, obtained for an aspect angle δ = 30◦ . Again, a single point of the diagram
generally corresponds to two determinations of the parameters (Hu , χB ), but the
connection between the two sets cannot be expressed in simple terms. Note also
in Fig. 13.4, that the maximum fractional polarization p̃Q is not obtained for zero
magnetic field but for the set (Hu 0.3, χB = −90◦ ) – which means that the Hanle
effect acts, in this case, as a polarizing (instead of depolarizing) mechanism – and
that for large values of Hu , p̃Q can even become negative.
If the angle θB is allowed to differ from 90◦ , the polarization diagrams take
different forms. As an example, we show in Fig. 13.5 the polarization diagram
obtained for θB = 45◦ , δ = 30◦ .
c) The Hanle effect (polarization profiles)
can be computed
The Stokes parameters profiles of the radiation directed along Ω 0
from Eq. (13.20) once the velocity distribution f (v ) of the scattering atoms is
specified. For our illustrative purposes, we can simply assume that the velocity
distribution is Maxwellian and is characterized by the thermal velocity vT , which
706 CHAPTER 13
Fig.13.5. Same as Fig.13.3 for θB = 45◦ and δ = 30◦ . The labelling of the curves is the same as
in Fig.13.4.
implies on the profiles a Doppler broadening in frequency units, ∆νD , given by
∆νD = ν0 vT /c, ν0 being the line frequency.
Figure 13.6 shows the Stokes profiles computed for a line at 5000 Å having Ein-
stein coefficient A = 5 × 107 s−1 . The limb-darkening coefficients are the same as
in points a) and b), while the projected height and aspect angle are the same as
in Fig. 13.3 (h = 0.073 R , δ = 0◦ ). The magnetic field points to the observer
(θB = 90◦ , χB = 0◦ ) and is characterized by Hu = 1 (which implies B = 5.69 G
for a line with gα J = 1; see Eq. (10.29)). Finally, the Doppler broadening is
u u
∆νD = 4 × 109 s−1 (corresponding to vT = 2 km s−1 ), and the damping constant
is Γ = A/(4π) – see Eq. (10.43). The Stokes parameters √ turn out to be propor-
tional to the combination of parameters τL Iν (0)/( π ∆νD ), where τL is defined
0
in Eq. (13.22) and Iν (0) is the disk-center intensity of the continuous solar radia-
0
tion at the frequency ν0 . In Fig. 13.6 the Stokes parameters are normalized to this
quantity.
In practice, the Stokes parameters I(ν, Ω ), Q(ν, Ω
), and U (ν, Ω
) show almost
0 0 0
pure Gaussian profiles characterized by a common width ∆νD , whereas V (ν, Ω )
0
shows a typical antisymmetrical profile of very small amplitude.1 Considering
1
Were not for the presence of atomic polarization, the V profile would be well-represented
by Eq. (11.2), because the magnetic field is very weak (νL /∆νD 2 × 10−3 ). Since, in the
case considered, atomic polarization is about 10% (as shown by the values of Q/I and U/I in
THE OUTER LAYERS OF STELLAR ATMOSPHERES 707
Fig.13.6. The Stokes parameters, computed from Eq.(13.20) for the geometry of Fig.13.1, are
plotted against the reduced frequency (ν0 − ν)/∆νD . The main geometrical parameters are:
δ = 0◦ , θB = 90◦ , χB = 0◦ . The remaining parameters, as well as the normalization of the
Stokes parameters, are specified in the text. Note the different scale of the panels.
however the ratios pQ , pU , and pV , defined by
)
Q(ν, Ω )
U (ν, Ω )
V (ν, Ω
)=
pQ (ν, Ω 0
, )=
pU (ν, Ω 0
, )=
pV (ν, Ω 0
,
0 ) 0 ) 0 )
I(ν, Ω 0 I(ν, Ω 0 I(ν, Ω 0
some departures from this idealized situation clearly appear (see Fig. 13.7). In the
line core, up to a distance of approximately two Doppler widths from line center,
pQ and pU are practically constant and equal, with very good approximation, to
the corresponding values p̃Q and p̃U relative to the integrated profiles. Beyond two
Doppler widths the Hanle effect disappears – consistently with our discussion of
Sect. 10.4 – so that pU goes asymptotically to zero, whereas pQ approaches the
limiting value corresponding to resonance polarization in the absence of magnetic
Fig. 13.6), it has to be expected that Eq. (11.2) represents the V profile with a precision of the
same order.
708 CHAPTER 13
Fig.13.7. The fractional polarization in the three Stokes parameters Q, U , and V of Fig.13.6 is
plotted against the reduced frequency (ν0 − ν)/∆νD (solid lines). The dashed lines represent the
values obtained from frequency-integrated profiles.
fields (pQ 0.28, cf. Fig. 13.3). It should be remarked that the frequency distance
from line center where the curves pQ and pU start to inflect (about two Doppler
widths in the example of Fig. 13.7) is a function of the reduced damping constant,
Γ A
a= = ,
∆νD 4π ∆νD
and that such distance increases for decreasing a, and vice versa. It should also be
remarked that the behavior illustrated in Fig. 13.7 may substantially be altered by
any source of (unpolarized) continuous emission that may add to the line emission.
As apparent from Fig. 13.6, at two Doppler widths from line center the intensity
is reduced by a factor 102 with respect to its peak value.
It is interesting to compare the Stokes profiles in Fig. 13.6 with those in Fig. 10.3.
The two figures refer to the same transition (J = 0, Ju = 1) and to the same ge-
ometry (a 90◦ scattering with the magnetic field pointing to the observer); the
Einstein coefficient A, the Landé factor, and the magnetic field intensity are the
same; stimulated emission is neglected in both cases. The main difference is that
Fig. 10.3 refers to the rest frame of the atom, while Fig. 13.6 refers to the observer’s
frame, the emitting atoms being characterized by an assigned velocity distribution.
The second difference concerns the anisotropy of the incident radiation field, which
is much smaller in the case of Fig. 13.6 (in Fig. 10.3 we had a radiation beam).
The profiles in Fig. 13.6 are convolutions of profiles like those in Fig. 10.3 with
the velocity distribution of atoms. As a result, the ‘structure’ visible in Fig. 10.3
– characterized by typical widths of order Γ = A/(4π) – is completely lost in
Fig. 13.6, where the typical width is the Doppler broadening ∆νD (numerically,
THE OUTER LAYERS OF STELLAR ATMOSPHERES 709
Fig.13.8. Same as Fig.13.6 for a magnetic field lying in the plane of the sky (δ = 0◦ , θB = 45◦ ,
χB = −90◦ ). Note the presence of a weak circular polarization signal.
∆νD 103 × Γ ). In Fig. 13.6 the polarization in the far wings is reduced because
of the smaller anisotropy (pQ 0.28, to be compared with pQ = 1). Moreover,
as shown by Fig. 13.7, the polarization within the line is further reduced by the
Doppler effect (in Fig. 10.3 the radiation is totally polarized at every single fre-
quency).
The polarization profiles in the Hanle effect regime may show, in some cases,
rather peculiar features. A striking example is given in Fig. 13.8, which corresponds
to the same set of parameters as Fig. 13.6 except for the magnetic field’s direction,
which is now characterized by the angles θB = 45◦ , χB = −90◦ . In spite of the fact
that the magnetic field is now lying in the plane of the sky, and is thus perpendicular
to the line of sight, the V Stokes parameter shows a typical antisymmetrical profile
that might erroneously be interpreted (according to Eq. (11.2)) as due to a very
weak magnetic field having a non-zero component along the line of sight. In fact,
this circular polarization profile originates from the terms with K = 1, K = 2,
Q = ±1 in Eq. (13.20), whose contribution does not vanish, in general, even for
magnetic fields perpendicular to the line of sight.
710 CHAPTER 13
13.4. Diagnostics of Magnetic Fields in Solar Prominences
It is nowadays believed (on the basis of several observations) that magnetic fields
in solar prominences typically range from 0 to 100 G, with most of reliable mea-
surements falling in a more restricted interval centered at about 20 G. On the other
hand, as we have seen on several examples developed in this book (cf. Sects. 5.9,
10.3, and 13.3), the Hanle effect, as observed in a given spectral line, turns out to be
particularly sensitive to magnetic field values such that the dimensionless parame-
ter Hu , defined in Eq. (10.29), is contained in an interval that we can – somewhat
arbitrarily – set to (0.1–10). As shown by Eq. (10.29), the range of sensitivity
of the Hanle effect in terms of magnetic field’s intensity depends on the Einstein
coefficient of the line and on the Landé factor of the upper level involved in the
transition. Assuming for these quantities the typical values A = 5 × 107 s−1 and
gα J = 1, we obtain a range that extends between 0.6 G and 60 G. Therefore, the
u u
Hanle effect turns out to be a particularly suitable diagnostic tool for the magnetic
fields typically found in solar prominences.1
Among the many spectral lines that are observed in solar prominences, one may
ask which would be the most suitable for this type of diagnostics. The answer is
that the ‘ideal’ spectral line should have the following properties:
a) It should be sufficiently bright in the prominence spectrum to allow high precision
polarimetric measurements. Typically, an error less than at least 10−3 in the
observed quantities p̃Q and p̃U is required.
b) The spectral line itself, and all the other lines involved in the model atom
needed to treat the problem, should be optically thin. When this requirement is
K
fulfilled, the radiation field tensor JQ (νi ) of Eqs. (13.13) can be simply computed
from the photospheric intensity, whose spectral and center-to-limb properties are
K
well-known. For optically thick lines, the tensor JQ (νi ) can only be computed by
introducing a model for the thermodynamic structure of the prominence and its ge-
ometry (vertical slab, horizontal slab, cylinder, arch, etc.). This obviously leads to
a strongly model-dependent diagnostics (for an example, see Landi Degl’Innocenti
et al., 1987).
c) Since high velocities are often observed in prominences,2 all the lines involved
in the model atom should fall within continuum windows of the photospheric solar
spectrum, in order to avoid possible contaminations of the radiation field tensor
K
JQ (νi ) due to the Doppler effect. If this is not the case, the theory presented in the
former section – which implies the absence of velocity/density-matrix correlations –
cannot be applied, and one has to resort to a more complicated formalism (see
Sect. 13.7). It should be noticed that this is a rather severe requirement because
1
Early suggestions for using the Hanle effect for the measurement of magnetic fields in
prominences were given by Öhman (1929).
2
According to Tandberg-Hanssen (1995), non-thermal velocities up to 10 km s −1 are often
observed in the outer edges of quiescent prominences. In active prominences velocities can be
even higher.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 711
λ
λ
λ λ
Fig.13.9. The model atom of the triplet system of HeI (orthohelium) used for the diagnostics
of magnetic fields in prominences. The dashed lines refer to two infrared transitions that can
be omitted in the theoretical calculations. Note that each of the terms 3 P and 3 D is actually
composed of three different J-levels.
it implies that all the lines involved in the model atom are not present in the
photospheric spectrum.
d) Since depolarizing collisions are often poorly known, it may be convenient to
avoid lines whose interpretation requires atomic models involving transitions with
low values of the A Einstein coefficient. By so doing, one can avoid the introduc-
tion of depolarizing rates into the statistical equilibrium equations and the ensuing
dependence of the results on a further parameter (the number density of particles).
On the other hand, when the cross sections for depolarizing collisions are accurately
known (either from laboratory experiments or from theoretical calculations), ob-
servations of linear polarization may also be used as a diagnostic tool for inferring
the particle density – although the disentangling process generally requires the use
of simultaneous observations in several lines. An example of such procedure has
been given by Bommier et al. (1986a,b).
One line which meets to a large extent the above requirements is the D3 line of
HeI at 5876 Å. According to Bommier (1977), the polarization characteristics of
this line can be described by means of the atomic model of Fig. 13.9. The model
involves 5 terms (11 levels) and 6 transitions, though the two infrared transitions
drawn in broken lines in the figure can safely be omitted. The two visible lines
at 5875.6 Å and 7065.3 Å fall into broad continuum windows. The infrared line at
10830 Å presents in the photospheric spectrum a very shallow depression (of the
order of 5% of the continuum) that can safely be neglected. The only problem is
712 CHAPTER 13
Fig.13.10. Theoretical polarization diagram of the frequency-integrated linear polarization Stokes
parameters of the D3 line observed in a solar prominence. The diagram refers to a point located in
the plane of the sky at a height of 70 . The magnetic field is assumed to be horizontal (θB = 90◦ ,
see Fig.13.1). Full lines correspond to χB = const., while broken lines correspond to constant
field intensity (expressed in G).
caused by the UV line at 3888.6 Å, which falls into a complicated region of the
spectrum covered by several absorption lines of FeI. Since, however, this is a line
of secondary importance in establishing the statistical equilibrium of the atomic
model, it is reasonable to neglect the exact wavelength dependence of the solar
spectrum in the neighborhood of this line and to smooth it through a convolution
with a rectangular profile having a width of the order of 1 Å.
Concerning optical thickness, it can be stated with a good degree of confidence
that the four lines involved in the calculations are indeed optically thin in quiescent
prominences, except for the infrared line λ 10830 which, in some cases, may reach
non-negligible values of optical thickness.1
Taking into account these remarks, and considering that the four lines involved
in the model have pretty large values of the A Einstein coefficient (so that collisions
can be neglected), one can apply to this model atom the theory developed in the
former section.
Figure 13.10 shows the polarization diagram for the frequency-integrated radia-
1
Polarimetric observations of this line, performed by Lin et al. (1998) on filaments over the
disk, provide a clear indication of non-negligible optical thickness along the vertical direction in
the observed objects.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 713
tion emitted in the D3 line. The diagram refers to a prominence observed in the
plane of the sky (δ = 0◦ ) at a height h = 0.073 R (corresponding to 70 arcsec).
The atomic data for the fine-structure intervals of the terms 3 P and 3 D of the
model atom are from Wieder and Lamb (1957), Pichanik et al. (1968), Kponou et
al. (1971), and Tam (1975), whereas the Einstein coefficients of the transitions are
from Wiese et al. (1966). The limb-darkening coefficients and the intensities of the
continuum radiation (needed to compute the radiation field tensor components J00
and J02 at the frequencies of the transitions) were obtained from the tables of Allen
(1973) by interpolation. For the UV line at 3888.6 Å, the continuum intensity is
reduced by a factor 5 (see our previous discussion).
The shape of the diagram presents many of the features illustrated on simpler
model atoms in Chap. 10. In particular, the feature present in the upper part of
the diagram is due to the partial relaxation of coherences in the levels 2s 3 S and
2p 3 P (lower-level Hanle effect, cf. Sects. 10.8-10.9), while the loops appearing for
field intensities 10 G are due to level-crossing interferences in the upper term of
the D3 line 3d 3 D (cf. Sect. 10.18).
Diagrams like that of Fig. 13.10 have been extensively used by Bommier and
collaborators for the diagnostics of magnetic fields in prominences (Bommier et al.,
1994). It should be remarked that the intrinsic symmetry of the diagram leaves
a fundamental ambiguity in the determination of the magnetic field vector. Such
ambiguity can be removed either with the help of simultaneous measurements in
optically thick lines or by taking advantage of the rotation of the sun under the
hypothesis of a rigid or quasi-rigid rotation of the prominence (Bommier et al.,
1981, 1994).
The diagnostics of prominence magnetic fields can also be performed by inter-
preting directly the Stokes parameters profiles observed in the D3 line. This line
is in fact composed of six fine-structure components, five of which are very close
in wavelength while the sixth, originating from the transition 2p 3 P0 − 3d 3 D1 , is
well-separated, lying at a distance of approximately 350 mÅ to the red from the
center of gravity of the others. In practice, it is like observing two distinct lines for
which two separate polarization diagrams can be drawn.1 Full theoretical details
of this diagnostic procedure are given in Landi Degl’Innocenti (1982b). Its appli-
cations to the interpretation of Stokes profiles from prominences can be found in
Athay et al. (1983) and Querfeld et al. (1985).
An interesting feature of the D3 line is the appearance of a complicated circular
polarization profile. This is because the usual Zeeman effect mixes with the effect
of atomic orientation due to the alignment-to-orientation conversion mechanism
discussed in Sect. 10.20. Figures 13.11 and 13.12 show the Stokes profiles computed
according to Eq. (7.47e). One can notice that the signature of the Zeeman effect is
largely masked by the presence of atomic orientation which produces the prominent
negative lobe.
1
The polarization diagrams for the frequency-integrated radiation in the two components
can be computed from Eqs. (7.48g) and (7.48c). In the latter, one should omit the profile φ and
restrict the summation over J to the values J = 1, 2 for the blue component and J = 0 for the
red component.
714 CHAPTER 13
Fig.13.11. The theoretical Stokes profiles of the D 3 line, normalized to the maximum intensity, are
plotted as a function of the wavelength distance from the center of the red component (expressed
in Å). The profiles refer to a point located in the plane of the sky at a height of 70 . The magnetic
field vector is specified by B = 35 G, θB = 90◦ , χB = 0◦ . The Doppler width is ∆λD = 137 mÅ,
corresponding to a thermal velocity of 7 km s −1 .
Apart from the Hanle effect, the diagnostics of magnetic fields in prominences
can also be carried out with the traditional technique of the longitudinal magne-
tograph.1 As discussed in Sect. 11.1, such technique is based on Eq. (11.2), which
implies both the weak field approximation and the absence of atomic polarization
in the levels involved in the transition. Whereas the weak field approximation is
definitely justified in prominences, the existence of atomic polarization implies that
the magnetograph technique fails in general to give reliable measurements of B in
these objects. This is especially true for fine-structured and hyperfine-structured
lines, whose circular polarization profiles are deeply contaminated by the presence
of a ‘spurious’ signal due to the atomic orientation induced in the upper level by the
alignment-to-orientation conversion mechanism. A typical example is the V pro-
1
Early measurements of the longitudinal component of the magnetic field in prominences
were reported by Zirin and Severny (1961) and by Rust (1966).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 715
Fig.13.12. Same as Fig.13.11 for the fractional polarization in the three Stokes parameters Q, U ,
and V . Note the disappearance of the Hanle effect in the far wings.
file of the D3 line shown in Fig. 13.11. There is no doubt that the magnetograph
technique applied to that profile would produce an erroneous result.
For ‘spectroscopically simple’ lines, on the other hand, Eq. (11.2) gives indeed a
zero-order approximation, that might be improved by solving the statistical equi-
librium equations and by computing, via Eqs. (10.39) and (A13.6), the first-order
correction to the V Stokes parameter due to the presence of atomic polarization.
In general, this correction is expected to be rather small (typically of the order
of 10%–20%), so that Eq. (11.2) should give a fairly good representation of the V
profile.1 Unfortunately, among the many lines that have extensively been used for
the diagnostics of magnetic fields in prominences through the magnetograph tech-
nique (Hα, Hβ, HeI λ 4771, HeI D3 , NaI D1 and D2 , Mg b), only the magnesium
lines fall into this category.
13.5. Diagnostics of Magnetic Fields from Coronal Forbidden Lines
The procedure that has been outlined in Sect. 13.3 for deducing the density matrix
of an atom (or ion) located in the outer layers of a stellar atmosphere, and hence
the polarization properties of the radiation emitted in one of its spectral lines, can
be directly applied to the interpretation of coronal observations in the well-known
forbidden lines of iron. These are the so-called green line of FeXIV at 5304 Å, the
1
Note, however, that the zero-order approximation predicts V = 0 when the magnetic field is
perpendicular to the line of sight. In that case, the only contribution to the circular polarization
profile comes from the first-order correction, and Eq. (11.2) cannot be applied (see Fig. 13.8).
716 CHAPTER 13
so-called red line of FeX at 6374 Å, and the two infrared lines of FeXIII at 10747 Å
and 10798 Å. All the transitions involved in the model atoms needed to interpret
these lines do not match significant features of the solar spectrum, so that the flat-
spectrum approximation is well-justified and velocity/density-matrix correlations
can be neglected.1
Notwithstanding the similarities of the theoretical approach, the interpretation of
polarimetric observations in coronal forbidden lines involves some differences with
respect to the ‘prominence case’ that have to be discussed in some detail.
The first difference concerns the atomic models. Because of the large ionization
degree of the iron ions, fine-structure intervals turn out to be very large, which
entails that such ions can be treated as multi-level atoms, rather than multi-term
atoms.
The second difference arises from the small values of the Einstein coefficient. The
four lines mentioned above, as well as the other lines involved in the model atoms,
are magnetic-dipole forbidden lines with values of A(αu Ju → α J ) ranging from
10 s−1 to 102 s−1 . This means a difference of the order of 106 with respect to the
Einstein coefficient of typical prominence lines, and implies that collisional rates
have to be included in the statistical equilibrium equations, as they are in fact
comparable to the radiative rates, or even larger (Sahal-Bréchot, 1974a,b; 1977).
This is a serious drawback because collisional cross sections are, in general, rather
poorly known; a major effort is still needed on this point in order to attain fully
reliable diagnostic techniques.
The third difference is due to the magnetic-dipole character of the transitions.
As shown in Sect. 6.8, this implies the following transformation on the geometrical
tensor T QK – see Eq. (5.155)
(i, Ω)
TQ
K → γ T K (i, Ω)
(i, Ω) , (13.25)
i Q
where γi (i = 0, . . . , 3) is the formal vector defined by
γi ≡ (1, −1, −1, 1) . (13.26)
For the case we are considering here, transformation (13.25) has no effect on the sta-
K
tistical equilibrium equations because the radiation field tensor JQ (νi ) appearing in
the radiative rates involves only the unpolarized solar radiation (cf. Eqs. (5.157)).
On the contrary, such transformation is fundamental in determining the Stokes pa-
rameters of the radiation emitted in a magnetic-dipole transition: it entails a sign
switch of the emission coefficients εQ and εU with respect to the usual electric-
dipole case.
The fourth difference is also connected with the low values of the Einstein coef-
ficient: even for magnetic fields as weak as 10−4 G, the parameter Hu defined in
1
The only exceptions may be provided by some UV transitions needed for the interpretation
of the green line. However, such transitions are dominated by collisional rates, thus the solar
UV spectrum plays no significant role in the statistical equilibrium equations (Casini and Judge,
1999).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 717
Eq. (10.29) turns out to be much larger than unity. This means that, in the ref-
erence system of the magnetic field, all the density-matrix elements ρK Q (αJ) with
Q = 0 can simply be set to zero. From the diagnostic point of view, this also
implies that there is no possibility of measuring the magnetic field intensity from
linear polarization observations in these lines.
Finally, the fifth difference is related to the fact that the corona, as observed
in the forbidden lines mentioned above, does not present, in general, remarkable
condensations. Therefore, the diagnostic content of observations in these lines is
non-local, and one must ultimately rely either on coronal models of the density and
magnetic field, or on tomographic techniques based on the hypothesis of a rigid or
quasi-rigid rotation.
The peculiarities outlined above have some consequences that are worth to be
discussed in detail. The first thing to notice is that the statistical equilibrium
equations for the density matrix take, in this case, a relatively simple form. Writing
these equations in the reference system of the magnetic field, it turns out that
the only non-vanishing statistical tensors are those of the form ρK Q (αJ) with K
even and Q = 0. The condition on Q has been discussed above, whereas the
condition on K follows from arguments similar to those presented in Sect. 10.7,1
which remain valid even considering the presence of collisions and/or stimulation
effects. It follows that for an atomic model composed of N levels with quantum
numbers (αi Ji ), with i = 1, . . . , N , the total number of equations (and unknowns)
is ⎧
⎪
⎪ %
N
⎪
⎨ (Ji + 1) for Ji integers
i=1
Neq =
⎪
⎪ %
N
⎪
⎩ (Ji + 12 ) for Ji half-integers .
i=1
Next we observe that the emission coefficient in a magnetic-dipole forbidden line
is given by Eq. (10.39) with the tensor T Q
K
transformed according to Eq. (13.25).
one has
For the radiation emitted along the direction Ω 0
hν
)=
εi (ν, Ω N 2Ju + 1 A(αu Ju → α J )
0
4π
× ) ρK (α J ) Φ̂KK (J , J ; ν) ,
γi T 0K (i, Ω 0 0 u u 0 u
K even K
where γi is defined in Eq. (13.26), the tensor ρK 0 is defined in the magnetic frame,
and the symbol Φ̂KK
Q (J ,
uJ ; ν) has been introduced in Eq. (13.20). This expres-
sion considerably simplifies by introducing the angle Θ between the magnetic field
vector and the direction Ω , and by defining the reference direction for the Stokes
0
parameters in such a way that the positive Q direction coincides with the projec-
onto the plane perpendicular to Ω
tion of B . By so doing, the components of the
0
1 K in the vertical frame are by assumption
Note that, since the only non-zero components of JQ
J00 and J02 , the components JQ1 in the magnetic frame are zero (see Eq. (13.14)).
718 CHAPTER 13
tensor T 0K can be simply deduced from Table 5.6 with the substitutions θ = Θ,
) vanish. As a
γ = 0, which imply, in particular, that all the components T 0K (2, Ω 0
consequence, the emission coefficient in the U Stokes parameter is zero, so that the
linear polarization of the radiation emitted along Ω is either parallel or perpen-
0
dicular to the projection of the magnetic field vector onto the plane perpendicular
to Ω .1
0
An analytical expression for the non-vanishing Stokes parameters can be found
with the help of Eq. (A13.6), which is well-verified in the corona given the weak
intensity of the magnetic field. After some Racah algebra we obtain for the emission
coefficient in the four Stokes parameters
) = C ρ0 (α J ) + √1
ε0 (ν, Ω 0 0 u u 3 cos2 Θ − 1 wJ(2)J ρ20 (αu Ju ) φ̂(ν)
2 2 u
3
) = C √ sin2 Θ w(2) ρ2 (α J ) φ̂(ν)
ε1 (ν, Ω 0 Ju J 0 u u
2 2
ε (ν, Ω )=0
2 0
) = C cos Θ ν ḡ ρ0 (α J ) + ∆ ρ2 (α J ) ∂ φ̂(ν)
ε3 (ν, Ω 0 L 0 u u 0 u u , (13.27)
∂ν
where
hν
C= N 2Ju + 1 A(αu Ju → α J ) ,
4π
and where wJ(2)J is the symbol defined in Eq. (10.12) – see also Table 10.1 – , φ̂(ν)
u
is given by Eq. (13.24), νL is the Larmor frequency given by Eq. (3.10), ḡ is the
effective Landé factor of Eq. (3.44), and ∆ is given by
⎡
∆ = −3 2Ju + 1 ⎣ gα J (−1)1+J −Ju Ju (Ju + 1)(2Ju + 1)
u u
! !
2 1 1 1 1 1
×
Ju Ju Ju J Ju Ju
⎧ ⎫⎤
⎨ J Ju 1⎬
+ gα J (J + 1)(2J + 1) J Ju 1 ⎦. (13.28)
J
⎩ ⎭
1 2 1
The sign of ε1 (ν, Ω ) determines whether the linear polarization is parallel or
0
perpendicular to the projection of B. In general, this can only be established
through the solution of the statistical equilibrium equations, which determines the
sign of ρ20 (αu Ju ). For a two-level atom with unpolarized lower level, however, the
1
Note that this property is valid locally; line-of-sight integration of the emission coefficient
can obviously invalidate this statement.
THE OUTER LAYERS OF STELLAR ATMOSPHERES 719
value of ρ20 (αu Ju ) can be obtained from Eq. (10.50). One has
(2) 2
2J + 1 B(α J → αu Ju ) wJu J J0 (ν0 ) 0
ρ20 (αu Ju ) = ρ (α J ) , (13.29)
2Ju + 1 A(αu Ju → α J ) 1 + + δu(2) 0
where the quantities and δu(2) are defined in Eq. (10.51), and where J02 (ν0 ) is
the radiation field tensor evaluated in the magnetic frame. Taking into account
Eqs. (13.14)-(13.15) and the expression of the rotation matrix D00 2
(see Table 2.1),
it is easily found, by substitution of Eq. (13.29) into Eq. (13.27), that the sign of
) is the same as the sign of the quantity (3 cos2 θ −1), where θ is the angle
ε1 (ν, Ω 0 B B
between the magnetic field vector and the vertical. Bearing in mind the definition
of the Van Vleck angle θV (see Eq. (5.100)), one has1
)>0
ε1 (ν, Ω )
(linear polarization parallel to B for 0 < θB < θV
0
or π − θV < θB < π
)<0
ε1 (ν, Ω ) for θ < θ < π − θ .
(linear polarization perpendicular to B
0 V B V
It should be mentioned that the diagnostics of coronal magnetic fields based on ob-
servations of linear polarization in forbidden lines was first suggested by Charvin
(1965), and that important theoretical contributions where brought by House
(1970a,b; 1971), Sahal-Bréchot (1974a,b; 1977) and, more recently, by Casini and
Judge (1999). From the observational side, successful results were obtained by
Mickey (1973), Arnaud (1982), Querfeld and Smartt (1984), and Arnaud and
Newkirk (1987).
Obviously, the diagnostics of magnetic fields in the corona can also be performed
via the more traditional techniques based on circular polarization. For such ob-
servations, the magnetograph relation (Eq. (11.2)) is still valid, although the effec-
tive Landé factor, ḡ, should be corrected to allow for atomic polarization. From
Eqs. (13.27) we easily obtain
∂I
V (λ) = −∆λB ḡ κ cos Θ ,
∂λ
where the ‘correction factor’ κ is given by (cf. Casini and Judge, 1999)
∆
1+ ḡ σ02 (αu Ju )
κ= ,
1+ 1
√
2 2
(3 cos2 Θ − 1) wJ(2)J σ02 (αu Ju )
u
with ∆ given by Eq. (13.28) and σ02 (αu Ju ) = ρ20 (αu Ju )/ρ00 (αu Ju ).
Early attempts of measuring the longitudinal component of the coronal magnetic
field through this technique were made by Harvey (1969), who could only give
1
Note that this conclusion is just the opposite of that obtained in Sect. 5.8 for the resonance
scattering of a classical oscillator. The difference is due to the fact that now we are dealing with
a magnetic-dipole transition.
720 CHAPTER 13
an upper limit of about 40 G. Recently Lin et al. (2000) have detected clear
antisymmetrical circular polarization signals in the FeXIII λ 10747 line from a
coronal condensation located over an active region. Two observations, interpreted
according to Eq. (11.2), give magnetic field components of 33 G at 0.15 R , and
10 G at 0.12 R , respectively.
13.6. Resonance Polarization in the Presence of
Velocity/Density-Matrix Correlations
We will now consider a physical situation more complicated than that of Sects. 13.3-
13.5. We still assume that the particle density is much smaller than the critical
value nc of Eq. (13.6), so that the generalized Boltzmann term can be neglected;
but we suppose that the transition frequencies involved in the model atom may
correspond to significant spectral features of the incident radiation field, so that
– contrary to the previous sections – the density matrix of atoms moving with ve-
locity v will depend on v . Under such circumstances, the density matrices of atoms
moving with different velocities are completely decoupled (see Sect. 13.2). The
v -dependent density matrix can be determined by solving the ‘ordinary’ statistical
equilibrium equations, which are written in the atomic rest frame (the comoving
frame introduced in Sect. 12.4). Once the equations are solved, the emissivity in
that frame can be calculated from the expressions derived in Chap. 7. To ob-
tain the emissivity in the observer’s frame, we must take into account the Doppler
shifts corresponding to the velocities of the individual atoms and integrate over the
velocity distribution.
In this section we illustrate these concepts by referring to a simple case where
an approximate analytical expression can be found for the v -dependent density
matrix. The problem that we are going to address consists in finding the radiation
scattered by a collection of two-level atoms (with unpolarized lower level) having an
arbitrary distribution of velocities and illuminated by an anisotropic and frequency-
dependent radiation field. The basic features of this problem have already been
analyzed by Sahal-Bréchot et al. (1998), who also carried out an application to
the OVI coronal line at 1032 Å, and by de Kertanguy (1998) who studied the
polarization of Hα in solar spicules.
Let us consider an atom located at a point P in the outer layers of a stellar atmo-
sphere. The two-level model atom with unpolarized lower level has been analyzed
in Sects. 10.1-10.2, where it was found that, neglecting the effects of magnetic fields
and collisions, the upper-level statistical tensors are given by Eq. (10.13). That ex-
pression is valid in the comoving frame. In the observer’s frame, the upper-level
statistical tensors can be written in the form
2J + 1 B(α J → αu Ju ) (K) K
ρK
Q (αu Ju ;
v) = wJ J (−1)Q J−Q (ν0 ) v
2Ju + 1 A(αu Ju → α J ) u
× ρ00 (α J ) , (13.30)
THE OUTER LAYERS OF STELLAR ATMOSPHERES 721
K
where [JQ (ν0 )]v is given by Eq. (13.9),
5 , -
K dΩ K
3
v · Ω .
JQ (ν0 ) v = T (i, Ω) Ii ν0 1 + ,Ω (13.31)
4π i=0 Q c
In Eq. (13.30) we have introduced the approximation of neglecting the v -depen-
dence of ρ00 (α J ): since the incident radiation field is weak (as implied by the
unpolarized lower level approximation – see the introduction to Chap. 10), the
population of the lower level is always very large compared to the statistical tensors
of the upper level.
On the other side, we have to remind that Eq. (10.13) is valid under the flat-
spectrum approximation, which is not rigorously verified for the problem we are
considering here because, by assumption, the incident radiation field has a spectral
structure around the transition frequency ν0 – typically, an absorption (or emission)
line characterized by a width ∆νp . It has then to be expected that Eq. (13.30) is
only approximately valid, the error related to the flat-spectrum approximation
being of the order of Γ/∆νp , where Γ is the natural width of the upper level. We
assume Γ/∆νp 1, a condition that is satisfied in many cases of interest (see
Sect. 13.1).1
The emission coefficient of the atom is given by Eq. (10.15) – but, again, we have
to remind that such expression is valid in the comoving frame; in the observer’s
frame, the frequency ν0 must be replaced by ν0 (1 + v · Ω/c). Moreover, the width
of the emission profile corresponding to any realistic velocity distribution is much
larger than Γ , the width of the φ profile appearing in Eq. (10.15), so that such
profile may in practice be replaced by a Dirac delta-function.2 It follows that the
emission coefficient for an ensemble of atoms having a velocity distribution f (v )
can be written in the form
hν
εi (ν, Ω) = d3v f (v ) N 2Ju + 1 A(αu Ju → α J )
4π
(K) , -
× wJ J T Q K ρK (α J ; v ) δ ν 1 + v · Ω − ν . (13.32)
(i, Ω) Q u u 0
u c
KQ
Substitution of Eq. (13.30) into Eq. (13.32) leads, with the help of Eqs. (10.6),
(9.5), and (10.17), to the expression
= kA
εi (ν, Ω) WK (J , Ju ) (−1)Q T Q
K
(i, Ω)
L
KQ
, -
v · Ω K
× 3
d v f (v ) δ ν0 1 + − ν J−Q (ν0 ) v . (13.33)
c
1
A rigorous treatment of this problem requires more sophisticated theories of frequency-
redistribution effects.
2
This substitution, joined with the approximation contained in Eq. (13.30), is often referred
to as the coherent-scattering approximation. It introduces an error of the order of Γ/∆νe , where
∆νe is the width of the emission profile.
722 CHAPTER 13
This expression can be rewritten in a compact form by introducing a suitable
redistribution function. Substituting Eq. (13.31) and recalling the definition of the
quantum-mechanical scattering phase matrix Pij (Ω, ) – see Eq. (10.20) – , one
Ω
obtains
5
dΩ
3
= kA
εi (ν, Ω) Ω
P (Ω, )
L
4π j=0 ij
, - , -
v · Ω v · Ω ,
× d v f (v ) δ ν0 1 +
3
− ν Ij ν0 1 + ,Ω
c c
or
5
3
dΩ
= kA
εi (ν, Ω) L dν ; ν, Ω
Rij (ν, Ω ) I (ν , Ω
j
) , (13.34)
4π j=0
; ν, Ω
where the redistribution function Rij (ν, Ω ) is given by1
; ν, Ω
Rij (ν, Ω ) =
v · Ω v · Ω
= Pij (Ω, Ω ) d v f (v ) δ ν − ν0 − ν0
3
δ ν0 − ν + ν0 . (13.35)
c c
An expression for Rij (ν, Ω ; ν, Ω
) valid for a Maxwellian distribution of velocities
is given in App. 18.
It is interesting to compare Eq. (13.33) with the expression that would be de-
duced for the emission coefficient under the hypothesis of complete redistribution
on velocities. Following the discussion of Sect. 13.2 one obtains
εi (ν, Ω) = kLA WK (J , Ju ) (−1)Q T Q
K
(i, Ω)
c.r.v.
KQ
/ , 0
-
v · Ω
× 3
d v f (v ) δ ν0 1 + −ν J¯−Q
K
(ν0 ) , (13.36)
c
where J¯QK
(ν0 ) is given by Eq. (13.8).
Equations (13.33) and (13.36) are deeply different. The former involves a sin-
gle integral over the velocity space, while the latter involves the product of two
independent integrals. Only when the radiation field presents no variation across
a spectral interval centered at ν0 and encompassing all the possible Doppler shifts
induced by the velocity distribution f (v ), do the two equations coincide – because,
in that case, [J−QK
(ν0 )]v is independent of v and equal to J¯−Q
K
(ν0 ). Obviously, it
1
The notation employed here for the redistribution function differs from that employed in
the classical book of Mihalas (1978). Apart from the obvious difference that we have here a 4 × 4
matrix-type redistribution function instead of a scalar one, there is an ordering inversion in the
arguments. What we write as R00 (ν, Ω ; ν, Ω
) is written by Mihalas as R(ν ,
n ; ν, n).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 723
Fig.13.13. Geometry of a scattering process in the outer layers of the solar atmosphere, in the
) is the same as in Fig.13.1.
presence of a velocity distribution. The observer’s frame (ea ,eb , Ω 0
The direction of the atomic velocity v is specified by the angles θv and χv .
is Eq. (13.33) that gives, in general, the correct physical description of scattering
phenomena in very diluted plasmas.
It is also interesting to observe that the two equations give the same result for
the frequency-integrated emission coefficient. Bearing in mind Eq. (13.3), one gets
≡ dν ε (ν, Ω)
ε̃ (Ω) = ε̃ (Ω)
≡ dν ε (ν, Ω) =
i i i c.r.v. i c.r.v.
= kLA WK (J , Ju ) (−1)Q T Q
K J¯K (ν ) .
(i, Ω) (13.37)
−Q 0
KQ
This means that velocity/density-matrix correlations act, in resonance scattering,
as a redistribution mechanism within each Stokes profile. The profiles are altered,
but their frequency-integrals are the same as those resulting from the approxima-
tion of complete redistribution on velocities.
As an application of the results just obtained, we refer to the geometry of
Fig. 13.13 and we study the polarization properties of the radiation scattered along
the direction Ω by a collection of two-level atoms located at the point P and hav-
0
ing a velocity distribution f (v ). We assume that the radiation field illuminating
the atoms, due to the star, is unpolarized and cylindrically symmetrical around
the vertical, and that its spectrum contains an absorption (or emission) line with
Gaussian shape centered at the transition frequency ν0 . As in Sect. 12.4, we denote
by Ic(0) the continuum intensity at disk center, by dc the central depression of the
spectral line, by ∆νp its width in frequency units, and by u1 the limb-darkening
724 CHAPTER 13
coefficient (supposed independent of frequency). Under these assumptions, the
radiation field tensor in the comoving frame is given by Eq. (12.44).
Let us apply Eq. (13.33) to the case of a coronal condensation lying in the plane
of the sky (δ = 0◦ in Fig. 13.13). The components of the tensor T QK ) can be
(i, Ω 0
◦ ◦ ◦
obtained from Table 5.6 by setting χ = 0 , θ = 90 , γ = 90 . Taking into account
properties (A17.1) and (A17.6), one obtains
) = kA v
εi (ν, Ω 0 L d3v f (v ) δ ν0 − ν + ν0 x Li (v ) , (13.38)
c
where
vx = v sin θv cos χv ,
and where
√
1 3
L0 (v ) = J00 (ν0 ; v, θv , 0) − √ W2 J02 (ν0 ; v, θv , 0) + W2 cos 2χv J22 (ν0 ; v, θv , 0)
2 2 2
√
3 3
L1 (v ) = √ W2 J0 (ν0 ; v, θv , 0) +
2
W2 cos 2χv J22 (ν0 ; v, θv , 0)
2 2 2
√
L2 (v ) = 3 W2 sin χv J12 (ν0 ; v, θv , 0)
L3 (v ) = 0 . (13.39)
In particular, for the frequency-integrated Stokes parameters one simply has
) = kA
ε̃i (Ω d3v f (v ) Li (v ) . (13.40)
0 L
The expressions just derived show the existence of an important physical phe-
nomenon induced by velocity on the polarization of the scattered radiation. As
shown in Sect. 12.4, the component J12 (ν0 ; v, θv , 0) of the comoving frame radiation
field tensor is generally non-zero. Therefore, as apparent from Eqs. (13.40) and
(13.39), the frequency-integrated U Stokes parameter of the scattered radiation
does not vanish in general (unless the velocity distribution is isotropic or at least
symmetrical about the vertical). This implies a rotation of the polarization direc-
tion with respect to the tangent to the solar limb. From this point of view, an
anisotropic distribution of velocities acts similarly to a magnetic field, and much
care has to be taken, when interpreting observations, to disentangle the rotation
induced by velocities from that due to the Hanle effect.
As an application of Eq. (13.40) to the diagnostics of velocities, we consider a
stream of atoms moving with a bulk velocity w and we suppose, for the sake of
simplicity, that there is no dispersion around this value, so that the function f (v )
is given by
f (v ) = δ (3) (v − w
).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 725
Fig.13.14. Velocity diagram for the scattering by a two-level atom in the geometry of Fig.13.13.
The scattering point is characterized by δ = 0◦ and h = 0.01 stellar radii. The stellar spectrum
presents an absorption line, characterized by dc = 0.8 and u1 = 0.2 (see Eq.(12.41)), at the
transition frequency. The atom moves relative to the star with a velocity w of fixed modulus.
The ratio between the Doppler shift induced by w and the width of the absorption line is ω = 1.4
(see Eq.(12.45)). Full lines correspond to χw = const., broken lines to θw = const. The parameter
W2 in Eqs.(13.39) is set to unity, corresponding to the transition (J = 0, Ju = 1).
In this case, one simply has
) = k A L (w
ε̃i (Ω 0 L i ),
where Li (w ) is given by Eqs. (13.39) with the substitutions v → w, θv → θw ,
χv → χw , θw and χw being the polar and azimuth angles of w defined according
to Fig. 13.13.
Figure 13.14 shows a typical ‘velocity diagram’, obtained by fixing the velocity
modulus w and by plotting the fractional polarization parameters
)
ε̃Q (Ω )
ε̃U (Ω
0 0
p̃Q = , p̃U =
ε̃ (Ω )
ε̃ (Ω )
I 0 I 0
against each other, while allowing the angles θw and χw to vary in the intervals
0◦ ≤ θw ≤ 90◦ , −90◦ ≤ χw ≤ 90◦ .
726 CHAPTER 13
Fig.13.15. Velocity diagram for the same physical situation considered in Fig.13.14, obtained by
varying the velocity modulus. Only the ‘external borders’, corresponding to χw = −90◦ and
+90◦ (right and left side respectively, cf. Fig.13.14) are shown. Curves are labelled by the value
of ω (see Eq.(12.45)).
The velocity diagram of Fig. 13.14 presents a high degree of degeneracy. From
Eqs. (13.39), and from the properties of the comoving-frame radiation field tensor,1
it can easily be seen that both the transformation χw → π − χw and the transfor-
mation w → −w (i.e., θw → π − θw , χw → π + χw ) leave the quantities p̃Q and
p̃U unchanged. Therefore, each point of the diagram corresponds to four possible
determinations: the nominal values (θw , χw ) and the combinations (θw , π − χw ),
(π − θw , π + χw ), (π − θw , −χw ). A further reason of degeneracy is the velocity
modulus, kept fixed to a specific value in the diagram. Obviously, for w = 0 the
diagram reduces to a single point, representing the fractional linear polarization
due to the scattering on a static atom. When w is increased the diagram grows,
reaching its maximum dimensions for a definite w value, and then decreases, as
illustrated in Fig. 13.15. It can be shown that for w → ∞ the diagram reduces
again to a single point, which coincides with the point corresponding to w = 0
because of the assumption of a frequency-independent limb-darkening coefficient.
1
Note that Eqs. (A17.1) and (A17.9) yield the relation
K
JQ (ν0 ; v, π − θv , 0) = e iQπ JQ
K
(ν0 ; v, θv , 0) .
THE OUTER LAYERS OF STELLAR ATMOSPHERES 727
Fig.13.16. Observer’s view of resonance polarization due to an atom moving with velocity w
off the solar limb. The velocity vector lies in the plane of the sky. The direction of linear
polarization (solid line) is intermediate between the parallel to the solar limb (dashed line) and
the perpendicular to the velocity (dotted line).
The degeneracy contained in the velocity diagram can be considerably reduced
by means of the further information coming from the frequency position of the
scattered line. According to Eq. (13.38), this line is centered at the frequency
ν = ν0 + ν0 wx /c, which allows one to retrieve the combination of parameters
w sin θw cos χw directly from observations. In particular, the four-fold degeneracy
in the plane θw -χw of Fig. 13.14 is reduced to a two-fold degeneracy: (θw , χw ),
(π − θw , −χw ) when wx > 0, or (θw , π − χw ), (π − θw , π + χw ) when wx < 0.
The shape of the diagram of Fig. 13.14 can be qualitatively understood via argu-
ments similar to those presented in Sect. 12.4. An illustration is given in Fig. 13.16,
which is a ‘front view’ of the scattering geometry of Fig. 13.13 with δ = 0◦ (ea points
to the right, eb upward; the velocity w is characterized by θw 40◦ , χw = +90◦).
Owing to limb-darkening and geometrical effects, the radiation scattered by a static
atom would be linearly polarized along the parallel to the solar limb (dashed line:
p̃Q > 0, p̃U = 0). The presence of an absorption line in the solar spectrum, com-
bined with the motion of the scattering atom, induces the phenomenon of Doppler
brightening: the radiation flowing ‘almost parallel’ to the velocity w is more red-
shifted (hence more intense) than the radiation coming ‘from the sides’, so that, if
limb-darkening and geometrical effects were absent, the radiation scattered by the
moving atom would be linearly polarized perpendicularly to w (dotted line). The
combined result of the two effects is a skewed polarization represented by the solid
line in the figure. Note that this implies a clockwise rotation of the polarization
direction relative to the horizontal direction, thus p̃U < 0 (left side of the diagram
of Fig. 13.14).
As expected on general physical grounds, the velocity diagram depends critically
both on the spectral characteristics of the stellar radiation and on the location of
the observed point over the stellar surface (height h, aspect angle δ). As an exam-
ple of the former effect, we show in Fig. 13.17 the velocity diagram corresponding
to an emission line. Comparison of Figs. 13.14 and 13.17 shows that the velocity
diagrams for absorption and emission lines are completely reversed; this is consis-
tent with the results of Sect. 12.4 and with the intuitive considerations illustrated
in Fig. 13.16. For an emission line, the phenomenon of Doppler dimming leads
728 CHAPTER 13
Fig.13.17. Velocity diagram for an emission line. All the parameters are the same as in Fig.13.14
except for dc and ω, now set to −4 and 2.5, respectively.
to the opposite orientation (parallel to w ) of the dotted line in Fig. 13.16. The
resulting polarization will thus be rotated in the counterclockwise direction with
respect to the parallel to the solar limb.
Consider now the effect of velocity/density-matrix correlations on the polariza-
tion profiles of the scattered radiation. The frequency-dependent emission coef-
ficient is given by Eq. (13.38), and strongly depends on the velocity distribution
f (v ) of the atoms. We consider here the case of a displaced Maxwellian, namely
v −wx 2 vy −wy 2 v −wz 2
1 − x
vT − vT − z
vT
f (v ) = 3
e e e , (13.41)
π 3/2 vT
where vT is the thermal velocity and w ≡ (wx , wy , wz ) is the bulk velocity of
the ensemble of atoms. Substituting Eq. (13.41) into Eq. (13.38), the integral in
dvx is immediately performed owing to the presence of the Dirac’s delta. The
integrals in dvy and dvz can then be performed numerically by means of Hermite
integrations. The emission profiles depend on a large number of parameters, and
a detailed investigation of their shapes is well outside the aims of this book. As an
example, we show in Fig. 13.18 the profiles obtained for a specific combination of the
parameters. Such profiles (solid line) are plotted together with those resulting from
the approximation of complete redistribution on velocities (broken line). Whereas
the latter peak all at the same distance from line center (the distance corresponding
THE OUTER LAYERS OF STELLAR ATMOSPHERES 729
Fig.13.18. Stokes profiles of the radiation scattered in the direction Ω (Fig.13.13 with δ = 0◦ )
0
by an ensemble of atoms characterized by the velocity distribution of Eq.(13.41). The profiles are
normalized to the maximum intensity and are plotted against the reduced frequency (ν0 −ν)/∆νD .
The solid line is obtained by taking velocity/density-matrix correlations into account, the broken
line corresponds to complete redistribution on velocities. The combination of parameters is:
dc = 0.8, u1 = 0.2, h = 0.01 R , ω = 1, w = vT , θw = 60◦ , χw = −45◦ , W2 = 1 (see text
for explanation of the symbols). Note the blue-shift due to the negative component of the bulk
velocity w
along the line of sight.
to the Doppler shift due to the bulk velocity w ) and
), the maxima of εI (ν, Ω 0
) for the former case are shifted to the blue, and the maximum of ε (ν, Ω
εQ (ν, Ω )
0 U 0
is shifted to the red. The profile of εU (ν, Ω0 ) seems to be the most affected by
velocity/density-matrix correlations. This is better illustrated in Fig. 13.19, where
the fractional Stokes parameters
)
εQ (ν, Ω )
εU (ν, Ω
)= 0 )= 0
pQ (ν, Ω 0 , pU (ν, Ω 0
ε (ν, Ω ) ε (ν, Ω )
I 0 I 0
are compared with the corresponding quantities obtained under the assumption of
complete redistribution on velocities. The figure clearly shows the exact symmetry
730 CHAPTER 13
Fig.13.19. Same as Fig.13.18 for the fractional Stokes parameters.
of these profiles about line center (which reflects the symmetry of the intensity
)
profile of the photospheric absorption line) and the substantial decrease of pU (ν, Ω 0
in the wings. This is a further similarity between the Hanle effect and the effects
of a velocity distribution on resonance polarization.
13.7. The Hanle Effect in the Presence of
Velocity/Density-Matrix Correlations
The results obtained in the former section can be easily generalized to the case
where a weak magnetic field is present. By ‘weak’ we mean here a field such that
νL ∆νp ,
so that the flat-spectrum approximation is well-verified (see Sect. 13.1) and the
splittings between the different Zeeman components of the transition can be ne-
glected. The results obtained by this approach are obviously in error by a quantity
of order νL /∆νp – or Γ/∆νp , where Γ is the natural width in frequency units of
the upper level of the transition, if this ratio is larger than the former. We recall
again that more sophisticated theories of frequency redistribution are needed to
overcome such limitation.
The generalization of the equations of the former section to the case of the Hanle
effect is obtained starting from the expression of the upper-level statistical tensors
in the presence of a magnetic field (i.e., from Eq. (10.27) instead of Eq. (10.13)).
THE OUTER LAYERS OF STELLAR ATMOSPHERES 731
Equation (13.30) is therefore replaced by the following
2J + 1 B(α J → αu Ju ) K
ρK
Q (αu Ju ;
v) = wJ(K)J (−1)Q J−Q (ν0 ) v
2Ju + 1 A(αu Ju → α J ) + 2πi νL gα J Q u
u u
× ρ00 (α J ) , (13.42)
where all the tensorial quantities are defined in a reference system having the z-axis
in the magnetic field direction.1 On the other hand, the expression for the emission
coefficient given in Eq. (13.32) is still valid. Substitution of Eq. (13.42) yields
1
= kA
εi (ν, Ω) WK (J , Ju ) (−1)Q T Q
K
(i, Ω)
L
1 + i QHu
KQ
, -
v · Ω K
× d3v f (v ) δ ν0 1 + − ν J−Q (ν0 ) v , (13.43)
c
where Hu is given by Eq. (10.28), or, similarly to Eq. (13.34)
5
3
dΩ
= kA
εi (ν, Ω) L dν ; ν, Ω
Rij (ν, Ω ; B)
I (ν , Ω
j
) ,
4π j=0
where
; ν, Ω
Rij (ν, Ω ; B)
= P (Ω, ; B)
Ω
ij
v · Ω v · Ω
× d v f (v ) δ ν − ν0 − ν0
3
δ ν0 − ν + ν0 , (13.44)
c c
Ω
Pij (Ω, ; B)
being the quantum-mechanical scattering phase matrix in the pres-
ence of a magnetic field of Eq. (10.33). It is easily seen that Eq. (13.36) generalizes
into
1
εi (ν, Ω) = kLA WK (J , Ju ) (−1)Q T Q
K
(i, Ω)
c.r.v. 1 + i QHu
KQ
/ , 0
-
v · Ω
× 3
d v f (v ) δ ν0 1 + −ν J¯−Q
K
(ν0 ) , (13.45)
c
1 K (ν )] appearing in Eq. (13.42) is defined in a
Note that the radiation field tensor [JQ 0 v
Effects of order
reference frame which moves with velocity v and has its z-axis directed along B.
v/c on B implied by the Lorentz transformations are neglected.
732 CHAPTER 13
Fig.13.20. Velocity-Hanle diagram corresponding to the physical situation of Fig. 13.14 with
the additional presence of a magnetic field of constant modulus (Hu = 2) parallel to w
. All the
remaining parameters have the same values as in Fig.13.14.
and Eq. (13.37) into
= ε̃ (Ω)
ε̃i (Ω) =
i c.r.v.
1
= kLA WK (J , Ju ) (−1)Q T Q
K
(i, Ω) J¯K (ν ) . (13.46)
1 + i QHu −Q 0
KQ
The expressions now derived depend on a very large number of parameters. A few
applications are described in the following.
First of all, we will examine how the frequency-integrated linear polarization
of the scattered radiation is affected by the simultaneous presence of a velocity
distribution and a magnetic field. To make an example, we consider the physical
situation of Fig. 13.14, where the atoms move with a unimodal velocity w and the
incident radiation presents an absorption line at the transition frequency, and we
‘add’ a magnetic field B of constant modulus parallel to w.
The resulting ‘velocity-
Hanle diagram’ is shown in Fig. 13.20. The mirror-like diagram of Fig. 13.21 refers
to the same situation with B antiparallel to w.
Comparison of Figs. 13.20-13.21 with Fig. 13.14 shows that the magnetic field has
an important effect on the frequency-integrated fractional polarization, especially
for large inclinations with respect to the vertical. Another obvious effect of the
THE OUTER LAYERS OF STELLAR ATMOSPHERES 733
Fig.13.21. Same as Fig.13.20 with the magnetic field antiparallel to w
.
magnetic field is the symmetry breaking of the diagram. Whereas in the zero-field
case the transformation
χw → −χw
produces a sign switch in p̃U (leading to a diagram symmetrical about the vertical
axis), this symmetry is lost in the magnetic case and is replaced by a more general
one,
p̃Q → p̃Q
for χw → −χw and B → −B ,
p̃U → −p̃U
where B → −B has to be understood in the sense that the magnetic field vector
changes from parallel to antiparallel to the velocity vector w.
Similarly to the non-magnetic case discussed in Sect. 13.6, the velocity-Hanle
diagrams of Figs. 13.20-13.21 present a high degree of degeneracy. It can indeed
be shown from Eq. (13.46), using the properties of the comoving-frame radiation
field tensor and those of the rotation matrices, that a single point in the diagram
of Fig. 13.20 corresponds to four determinations: the nominal set of parameters
θw , χw , ξ = +1
and the further combinations
734 CHAPTER 13
Fig.13.22. Velocity-Hanle diagram for a magnetic field of constant modulus (Hu = 0.4) and
direction (pointing to the observer). The other parameters have the same values as in Fig.13.14.
θw , π − χw , ξ = −1
π − θw , π + χw , ξ = −1
π − θw , −χw , ξ = +1 ,
where ξ = +1 or −1 according as B and w are parallel or antiparallel, respectively.
Further degeneracies are related to the velocity modulus (cf. Fig. 13.15) and to
the magnetic field modulus. Obviously, the degeneracy can be partly removed if
spectroscopic observations are available, as already discussed in Sect. 13.6.
A different kind of velocity-Hanle diagram is presented in Fig. 13.22. Here, the
combination of parameters is the same as in Fig. 13.20 except for the magnetic
field direction, which is kept fixed and parallel to the line of sight.
Consider now the combined effect of a magnetic field and a velocity distribution
on the Stokes profiles of the scattered radiation. We refer to the physical situa-
tion of Fig. 13.18 with the following differences: the bulk velocity of the atoms
(characterizing the ‘displaced’ Maxwellian distribution of Eq. (13.41)) points in
the vertical direction, and a magnetic field specified by Hu = 2 and pointing in the
same direction is present at the scattering point. The resulting profiles are shown
in Fig. 13.23. It is interesting to notice the presence of a purely antisymmetrical
U profile, a surprising result which is one of the most remarkable consequences of
THE OUTER LAYERS OF STELLAR ATMOSPHERES 735
Fig.13.23. Same as Fig.13.18 except for the direction of the bulk velocity w , here parallel to
the local vertical (θw = 0◦ ), and for the presence of a magnetic field pointing in the same
direction (θB = 0◦ , Hu = 2 ). Full line: taking velocity/density-matrix correlations into account
(Eq.(13.43)). Dashed line: assuming complete redistribution on velocities (Eq.(13.45)).
velocity/density-matrix correlations. The sign of U is positive in the red wing and
negative in the blue wing. It can be shown that the inversion of either w or B
produces a sign switch, so that, when w and B are antiparallel, the U profile is
negative in the red wing and positive in the blue wing.
CHAPTER 14
ASTROPHYSICAL APPLICATIONS:
STELLAR ATMOSPHERES
In the previous chapter we studied the polarization characteristics of spectral lines
formed in an optically thin plasma, where transfer effects are negligible. Here we
tackle the much more difficult subject of line formation in a (possibly magnetized)
optically thick plasma – e.g., a stellar atmosphere. This requires the simultaneous
consideration of the statistical equilibrium and the radiative transfer equations,
so that the theoretical framework of this chapter represents an extension to the
‘polarized case’ of the non-equilibrium theory usually referred to as non-LTE the-
ory of stellar atmospheres. The methods discussed in the following are the basic
tools for the interpretation of spectropolarimetric observations performed over the
solar disk. Special emphasis is devoted to resonance polarization in observations
performed close to the solar limb, both in the absence and in the presence of a
magnetic field.
14.1. The Non-LTE Problem
We have already analyzed some of the non-equilibrium phenomena that naturally
arise under laboratory conditions or in astrophysical plasmas when an atomic sys-
tem interacts with the radiation field. Both in the systematic investigation per-
formed in Chap. 10 and in the applications described in Chap. 13 we assumed that
the radiation field impinging on the atomic system is a known function of frequency
and direction. Obviously, this approach becomes totally inappropriate when one
aims at understanding the spectropolarimetric profiles of lines that are formed in
an optically thick plasma such as a stellar atmosphere. In this case one is forced, in
general, to look for a self-consistent solution of the statistical equilibrium equations
and of the radiative transfer equations, starting from a model – either theoretical
or empirical – of the medium where the process of line formation takes place.
Historically, it was in the late 1960s that solar physicists realized that the inter-
pretation of the intensity profiles of many lines of the solar spectrum required this
more sophisticated approach, for which the name of non-LTE was soon adopted.1
This first non-LTE theory, fully described in the classical monograph by Mihalas
(1978), considerably improved our understanding of spectral line profiles. However,
this theory has the intrinsic limitation of leaving aside polarization phenomena: the
atomic system is described in terms of the only level populations, and the radiation
1
The name ‘non-LTE’ reminds the upgrading of the older and simpler theory based on the
hypothesis of Local Thermodynamic Equilibrium (LTE).
738 CHAPTER 14
radiative
collisions transfer
sol
equations
te
wri
(atmospheric
ve
model)
compare with
ρ α νΩ
observed νΩ
magnetic
sol
field statistical
ve
te
wri
equilibrium
equations
Fig.14.1. Schematic representation of the non-LTE problem of the 2 nd kind. In principle, the
problem is solved by iterating the operations indicated in the loop until self-consistency is reached.
field in terms of the only intensity.
In the following we will refer to the ‘old’ non-LTE theory as non-LTE of the 1 st
kind , and we will use the expression non-LTE of the 2 nd kind for the more general
theory which takes polarization phenomena (both in the atomic system and in the
radiation field) into account. Needless to say that the two theories have to coincide
under the limit where polarization phenomena are neglected.
The physical scenario underlying the theory of non-LTE of the 2nd kind can be
synthetically described by a block diagram such as that of Fig. 14.1. The diagram
refers to the case of a collection of multi-level atoms having a local distribution
of velocities f (v ). It is assumed that velocity/density-matrix correlations can be
neglected, so that a unique, velocity-independent density matrix can be defined
(this is the approximation of complete redistribution on velocities discussed in
Sect. 13.2). It is understood that the flat-spectrum approximation, which is at the
basis of our theoretical development, is also satisfied (the limitations on non-LTE
theory implied by this approximation will be discussed at the end of this section).
Referring to Fig. 14.1, let us suppose that, for each level (αJ) of the model atom
considered, the statistical tensors ρK Q (αJ) are known at each point of the medium.
ρA (ν, Ω),
The transfer coefficients ηiA (ν, Ω), η S (ν, Ω),
ρS (ν, Ω),
and ε (ν, Ω)
are given
i i i i
– in the atomic rest frame – by Eqs. (7.15). Since the atoms are moving with a
local velocity distribution f (v ), the Φ profiles in Eqs. (7.15) should be changed
according to (cf. Eq. (13.16))1
1 for which the transfer coefficients
This substitution depends, in general, on the direction Ω
are computed. However, this is not true in the important case where the velocity distribution is
Maxwellian. In that case the real and imaginary parts of the Φ̂ profile reduce, respectively, to the
(suitably normalized) Voigt function and the associated dispersion profile (see Sect. 5.4).
STELLAR ATMOSPHERES 739
Φ(να J Mu , α J M − ν) → Φ̂(να J Mu , α J M − ν) =
u u u u
, -
v · Ω
= 3
d v f (v ) Φ να J Mu , α J M 1+ −ν . (14.1)
u u c
Substituting the values of ρK Q (αJ) at a given point into Eqs. (7.15), we can compute
the transfer coefficients at that point for any direction and frequency.
On the other hand, the knowledge of the transfer coefficients at each point of
the medium allows one to solve, via the methods developed in Chap. 8, the trans-
fer equation for the (polarized) radiation travelling along any given direction. In
other words, we can compute the Stokes parameters I(ν, Ω), Q(ν, Ω), U (ν, Ω),
and
V (ν, Ω) at any point.
This allows one to compute the radiation field tensor J¯Q K
(ν) of Eq. (13.8) at each
point, and then the radiative rates appearing in the statistical equilibrium equations
via Eqs. (7.14).1 Adding the collisional rates of Sect. 7.13 (see Eq. (7.101)), and
solving the statistical equilibrium equations, we obtain new values for the statistical
tensors ρKQ (αJ).
2
The loop of Fig. 14.1 is thus closed, and – at least in principle – it is sufficient to
repeat it as many times as needed until a self-consistent solution for the statistical
tensors and the radiation field is reached, which solves the non-LTE problem of the
2nd kind.
In the following sections we will present some practical applications of the scheme
just described. However, before proceeding, we feel necessary to add some general
remarks and considerations which may help to clarify the physical and mathemat-
ical aspects of the problem.
a) Under the limit of very large depolarizing rates, or, in other words, when the
quantities D(K) (αJ) defined in Eq. (7.102) are such that
D(K) (αJ) → ∞ (K ≥ 1)
for each (αJ)-level of the atomic model, all the statistical tensors – except those
of rank 0 – vanish. In this case there is no atomic polarization, and the physical
state of the atom is completely specified by the level populations (or the statistical
tensors ρ00 (αJ)). If, in addition, the magnetic field is zero throughout the medium,
the radiation field is unpolarized, so it is fully specified by its intensity (the Stokes
parameter I(ν, Ω)). As a result, the non-LTE problem of the 2nd kind reduces to
the non-LTE problem of the 1st kind. The latter is thus a special case of the former,
corresponding to the two conditions just specified.
1 K (ν) with J¯K (ν) in Eqs. (7.14) is the obvious counterpart of the
The replacement of JQ Q
substitution in Eq. (14.1).
2
In this chapter we adopt the assumption of complete redistribution on velocities in its more
restrictive meaning, where the generalized Boltzmann term is neglected (see comments at the end
of Sect. 13.2). Accordingly, the contributions of Eq. (13.12) are disregarded.
740 CHAPTER 14
This argument also shows that it is possible to envisage a kind of intermediate
non-LTE regime, that may be called non-LTE of order 1.5 , where the atomic sys-
tem is unpolarized whereas the radiation field, owing to the presence of a magnetic
field, is polarized. This intermediate regime, which also can be regarded as a special
case of non-LTE of the 2nd kind, has received considerable attention in the liter-
ature because it is considered appropriate to describe line formation in sunspots.
Its discussion is deferred until Sect. 14.4.
b) The second remark concerns the general role of the magnetic field and of colli-
sions in non-LTE of the 2nd kind. As pointed out in Fig. 14.1, the magnetic field
has a double effect: on the one side, it affects the radiative transfer coefficients
by splitting up the lines into several components; on the other side, it affects the
statistical equilibrium equations via a term (or a kernel) which controls the relax-
ation of coherences between different magnetic sublevels. Roughly speaking, one
can ascribe the first action to the Zeeman effect, and the second to the Hanle effect.
Collisions play also a double role: they affect the radiative transfer coefficients
by increasing the width of the individual profiles, and the statistical equilibrium
equations through different kinds of rates (depolarizing, inelastic, superelastic).
c) The third remark concerns the model atom that we have used as a prototype
to introduce the non-LTE problem. We referred to a multi-level atom described
in the statistical tensor representation, but the theoretical scheme is obviously
more general: it can be applied both to the multi-level atom described in differ-
ent representations (like the standard representation ραJ (M, M )) and to different
atomic models (like the multi-term atom, or the multi-level atom with hyperfine
structure: these models can in their turn be described in different representations).
Each atomic model and each representation obviously implies the use of the relevant
equations. For instance, to deal with the multi-term model atom in the statisti-
cal tensor representation one should use the statistical equilibrium equations of
Eq. (7.38) and the radiative transfer coefficients of Eqs. (7.47).
d) A fourth remark concerns the self-consistent solution of the non-LTE loop of
Fig. 14.1. The procedure implied in the figure, often referred to as Picard’s or
Λ-iteration method, has, in general, an extremely slow convergence rate under
typical non-LTE conditions. For this reason, much effort has been devoted since
the earliest years of the non-LTE theory to develop faster and faster numerical
methods for solving the coupled set of radiative transfer and statistical equilibrium
equations. The currently used techniques are based on accelerated iterative schemes
that evolved from earlier techniques developed for the non-LTE problem of the 1st
kind. A full discussion of this subject is well outside the purposes of this book. We
just refer the reader to a few papers by Trujillo Bueno and collaborators (Trujillo
Bueno and Manso Sainz, 1999; Trujillo Bueno, 1999) where the state-of-the-art
techniques are described in detail.
e) A further remark concerns the possibility of ‘cutting’, under particular physical
circumstances, the block diagram of Fig. 14.1, thus avoiding the need to solve the
coupled set of statistical equilibrium and radiative transfer equations.
STELLAR ATMOSPHERES 741
radiative solve compare with
transfer
collisions observed νΩ
equations
te
wri
(atmospheric
model)
ρ α
magnetic
field
Fig.14.2. The self-consistency loop of Fig.14.1 is ‘broken off’ by the LTE assumption, which sets
the density-matrix elements to fixed values, independent of the radiation field.
A possibility is provided by the LTE hypothesis, which implies that the statistical
tensors at any point of the medium are univocally determined by the local values
of the thermodynamic parameters. The statistical equilibrium equations are in
this case ‘implicitly solved’, so that – as schematically illustrated in Fig. 14.2 –
to obtain the Stokes parameters of the radiation travelling along any direction in
the medium one only needs to solve the radiative transfer equations. Most of the
results obtained in Chaps. 9 and 11 are based on this approach.
The second possibility takes place when the medium is optically thin at each of
the wavelengths corresponding to the different transitions involved in the model
atom considered, so that the radiation field impinging on the atom is fixed by the
boundary conditions. Knowing the radiation field and the thermodynamic parame-
ters of the medium, one can write down the statistical equilibrium equations. Their
solution provides the values of the statistical tensors, which allow one to determine
the emissivity and absorptivity properties of the medium and, consequently, the
spectropolarimetric profiles of the emerging radiation. This second possibility is
illustrated in Fig. 14.3. The results obtained in Chaps. 10 and 13 follow from an
approach of this kind.
f) The final remark concerns the limitations related to the flat-spectrum approx-
imation and to the approximation of complete redistribution on velocities. The
problems raised by the flat-spectrum approximation have already been discussed,
though in a different context, in Sect. 13.1. The conclusions that have been reached
there can be easily transferred to the non-LTE problem of the 2nd kind. In partic-
ular, one has to be especially careful with the limitations on the magnetic field’s
intensity (νL ∆νp ) when dealing with the multi-level atom, and with the lim-
itations of Eq. (13.1) when dealing with the multi-term atom or the multi-level
atom with hyperfine structure. If such limitations are not satisfied, one can still
treat the non-LTE problem of the 2nd kind under the flat-spectrum approximation
provided the interferences which violate the limitations themselves are set to zero
742 CHAPTER 14
radiative compare with
transfer
collisions observed νΩ
coefficients
te
wri
(atmospheric
model)
ρ α νΩ
magnetic
sol
field statistical
ve
te
wri
equilibrium
equations
Fig.14.3. In an optically thin plasma – for instance in a solar prominence – the radiation field
illuminating the atomic system is known a priori. This again ‘breaks off’ the self-consistency loop
of Fig.14.1.
ab initio. This approach fails to give an accurate description of the far wings of
the lines, but this is usually a minor drawback since the diagnostic content of such
spectral intervals is often limited. As remarked in Sect. 13.1, only for particularly
strong spectral lines this approach proves to be inadequate, and a theory including
partial frequency redistribution effects is required.
As far as the approximation of complete redistribution on velocities is concerned,
the situation is quite different. In principle, this approximation might be dropped
by introducing – as in Sect. 13.2 – velocity/density-matrix correlations, thus de-
scribing the atomic system in terms of velocity-dependent statistical tensors. This
implies, however, a dramatic increase in the number of unknowns, which has so far
discouraged any attempt to tackle the non-LTE problem of the 2nd kind in these
more general terms. Yet, the results on resonance scattering and the Hanle effect
obtained in Chap. 13 clearly show that the introduction of velocity/density-matrix
correlations into the non-LTE theory may be important for the correct interpre-
tation of spectropolarimetric profiles. For the time being, the approximation of
complete redistribution on velocities can simply be considered as an artifice used
to bring an extremely complex problem to a tractable form.
14.2. The Two-Level Atom: Non-LTE Theory for
Weak Magnetic Fields (Hanle Effect Regime)
As an application of the general non-LTE problem of the 2nd kind outlined in
Sect. 14.1, we consider – following Landi Degl’Innocenti et al., 1990 – the simplest
STELLAR ATMOSPHERES 743
possible atomic model, namely a two-level atom without hyperfine structure and
with unpolarized lower level. Consistently with the last assumption, we suppose
that the radiation field impinging on the atom is weak, in the sense that the average
number of photons per mode, n̄, is much smaller than unity, which allows one to
neglect stimulated emission.
We suppose that a collection of such atoms is distributed within a static medium
of arbitrary shape. In this medium the atoms interact with a magnetic field, B,
and with a population of colliding particles having a Maxwellian distribution of
velocities characterized by the temperature T . The atoms have also a Maxwellian
distribution of velocities, such that the corresponding Doppler width of the absorp-
tion profile is ∆νD (in frequency units).
No restriction is made on the spatial variation of the temperature T of the col-
liders and of the magnetic field vector B, as well as on the densities of the atoms
and of the colliders. However, we suppose for simplicity the Doppler width ∆νD
to be constant throughout the medium.
Furthermore, we suppose that the magnetic field is weak (in the sense that the
associated Larmor frequency νL is much smaller than ∆νD ) and that the inverse life-
time of the upper level, γu , is also much smaller than ∆νD , so that the flat-spectrum
approximation is satisfied.1 Finally, we adopt the approximation of complete redis-
tribution on velocities, which implies that at any point P of the medium the atom
can be described by a unique, velocity-independent density matrix that we denote
by K
ρQ (αJ) x ,
with (αJ) = (αu Ju ) for the upper level and (αJ) = (α J ) for the lower level, where
x is the coordinate of point P.
Now we write the statistical equilibrium equation for the multipole moments of
the upper level. Such equation has already been written down explicitly in Chap. 10
(see Eq. (10.1)): however, Eq. (10.1) holds in a reference system with the z-axis
parallel to the magnetic field. Since in the present case the magnetic field vector
is allowed to vary in the medium, it is convenient to introduce a fixed reference
system Σ and to characterize the direction of the field at each point through the
angles θB and χB defined in Fig. 14.4. Then we have to transform Eq. (10.1) as
explained in Sect. 7.12. This leads to the following equation2
d K K
ρQ (αu Ju ) x = −2πi νL gα J KQQ
K
ρQ (αu Ju )
x
dt u u
Q
+ TA (αu Ju KQ, α J K Q ) ρK
Q (α J ) x
+
K Q
1
The applicability of the flat-spectrum approximation follows from the two inequalities
γu ∆νD and νL ∆νD . The latter obviously implies an upper limit on the magnetic field
intensities that can be handled by this formalism.
2
Note that the last two lines of Eq. (10.1) are not affected by a rotation of the reference
system because of the assumed isotropy of collisions.
744 CHAPTER 14
θ Ω
θ
Ω γ
χ
χ
Fig.14.4. The fixed reference system Σ ≡ (xyz) used for the calculations of this section. At each
point P of the medium, the magnetic field vector is specified by the angles θB and χB . ea (Ω)
are the polarization unit vectors for the radiation flowing through P in the direction
and eb (Ω)
≡ (θ, χ).
Ω
− RE (αu Ju KQK Q ) + RS (αu Ju KQK Q ) ρK
Q (αu Ju ) x
K Q
2J + 1 (K)
+ CI (αu Ju , α J ) ρK
Q (α J ) x
2Ju + 1
− CS(0) (α J , αu Ju ) + D(K) (αu Ju ) ρKQ (αu Ju ) x
, (14.2)
where all the rates are evaluated at point x, νL is the Larmor frequency at the
same point, and the kernel KQQ
K
is explicitly given by Eq. (7.79),
∗
KQQ
K
= DQ
K
Q (RB ) Q DQ
K
Q (RB ) , (14.3)
Q
RB being the rotation that carries the local ‘magnetic’ reference system (having
the z-axis aligned with the magnetic field) into the fixed reference system Σ. In
terms of Euler angles one simply has
RB ≡ (−γB , −θB , −χB ) , (14.4)
where γB is an arbitrary angle that can be set to zero. The main properties and
the explicit expressions of the components of KQQ
K
are given in App. 19.
STELLAR ATMOSPHERES 745
The assumptions made at the beginning of this section yield two basic simplifi-
cations in Eq. (14.2):
- because stimulation effects are neglected, the relaxation rate RS is zero;
- because lower-level polarization is neglected, the statistical tensors of the lower
level reduce to ρK 0
Q (α J ) = ρ0 (α J ) δK0 δQ0 . This implies that the only radiative
rate TA needed in Eq. (14.2) is TA (αu Ju KQ, α J 0 0). Such rate is explicitly given
– in the atomic rest frame – by Eq. (10.9). Because of the assumption of complete
K
redistribution on velocities, the radiation field tensor JQ (ν0 ) in Eq. (10.9) has to be
¯K
replaced with the ‘average’ radiation field tensor JQ (ν0 ) given by Eq. (13.8). And
since the velocity distribution of the atoms is assumed to be Maxwellian, J¯Q K
(ν0 )
is given by (cf. Eqs. (13.10) and (13.11))
∞
J¯Q
K
(ν0 ) = dν p(ν0 − ν) JQ
K
(ν) , (14.5)
−∞
with ν
0
−ν 2
1 1 −
p(ν0 − ν) = √ e ∆νD
. (14.6)
π ∆νD
Recalling the expression of the radiative rate RE (Eq. (7.14e)), we can rewrite
Eq. (14.2) in the form
d K K
ρQ (αu Ju ) x = −2πi νL gα J KQQ
K
ρQ (αu Ju )
x
dt u u
Q
− A(αu Ju → α J ) + CS(0) (α J , αu Ju ) + D(K) (αu Ju ) ρKQ (αu Ju ) x
2J + 1
+ B(α J → αu Ju ) wJ(K)J (−1)Q J¯−QK
(ν0 )
2Ju + 1 u
+ δK0 δQ0 CI(0) (αu Ju , α J ) ρ00 (α J ) x ,
where the symbol wJ(K)J is given by Eq. (10.11).
u
Since the colliding particles have also a Maxwellian distribution of velocities, we
can apply the Einstein-Milne relation (Eq. (10.49)). Next we divide both members
by A(αu Ju → α J ) and introduce the usual notations (cf. Eqs. (10.51), (10.28))
CS(0) (α J , αu Ju ) D(K) (αu Ju ) 2πνL gα J
= , δu(K) = , Hu = u u
. (14.7)
A(αu Ju → α J ) A(αu Ju → α J ) A(αu Ju → α J )
Recalling Eqs. (7.8) we obtain, for stationary situations
K
1 + + δu(K) ρKQ (αu Ju ) x
+ iHu KQQK
ρQ (αu Ju )
x
=
Q
c2 2Ju + 1
= wJ(K)J (−1)Q J¯−Q
K
(ν0 ) + δK0 δQ0 BP (T ) ρ00 (α J ) x , (14.8)
2hν03 2J + 1 u
746 CHAPTER 14
where
hν
2hν03 − k T0
BP (T ) = e B (14.9)
c2
is the Planck function in the Wien limit (where stimulation effects are neglected).
In view of the following applications, it is convenient to rewrite Eq. (14.8) in
a more compact form by introducing suitable ‘source functions’ for the different
statistical tensors. Defining
2hν 3 2J + 1 ρKQ (αu Ju )
K
SQ (x ) = 20 x , (14.10)
c 2Ju + 1 ρ00 (α J ) x
Eq. (14.8) becomes
K
1 + + δu(K) SQ (x ) + iHu KQQ
K K
SQ (
x) =
Q
= wJ(K)J (−1)Q J¯−Q
K
(ν0 ) + δK0 δQ0 BP (T ) . (14.11)
u
K
The quantities SQ (x ) can be referred to as the irreducible components of the two-
level source function. They represent the obvious extension of the usual concept
of source function to the ‘polarized case’, as it is easily seen by considering the
(K = 0, Q = 0) component
2hν 3 2J + 1 ρ00 (αu Ju ) x
S00 (x ) = 20 ,
c 2Ju + 1 ρ00 (α J ) x
which, recalling Eq. (3.108), can be written as
2hν03 2J + 1 nαu Ju
S00 (x ) = , (14.12)
c2 2Ju + 1 nα J
where nα J and nα J are the overall populations at point x of the upper and
u u
lower level, respectively. The last expression shows that S00 (x ) is nothing but
the ‘classical’ source function for a two-level atom under the limit of negligible
stimulation effects. It should also be noticed that, according to Eq. (14.11), S00 (x )
obeys the equation (see Eqs. (7.102), (14.7), (10.14) and (A19.2))
(1 + ) S00 (x ) = J¯00 (ν0 ) + BP (T ) ,
or
S00 (x ) = (1 − ) J¯00 (ν0 ) + BP (T ) , (14.13)
where
CS(0) (α J , αu Ju )
= = .
1+ A(αu Ju → α J ) + CS(0) (α J , αu Ju )
STELLAR ATMOSPHERES 747
Equation (14.13) is the ‘classical’ expression relating the source function to the
angle and frequency averaged radiation field and to the Planck function.
We point out that the preceding equations (particularly Eqs. (14.8) and (14.13))
are basically the same as those derived in Sect. 10.6 (see Eqs. (10.50) and (10.52),
respectively). The only difference is that Eq. (10.50) holds in a reference system
where the atom is at rest and with the z-axis parallel to the magnetic field, whereas
Eq. (14.8) holds in the fixed reference system Σ of Fig. 14.4, the atoms being
characterized by an assigned velocity distribution and the magnetic field’s direction
by the angles θB , χB . This implies the appearance in Eq. (14.8) of the kernel KQQ
K
and – because of the assumption of complete redistribution on velocities – of the
tensor J¯Q
K K
(ν0 ) in the place of JQ (ν0 ).
Consider now the radiative transfer equation. From Eq. (6.83) we have, neglecting
stimulated emission
d 3
=−
Ii (ν, Ω) A
Kij +ε
Ij (ν, Ω) (i = 0, . . . , 3) , (14.14)
i
ds j=0
are the Stokes parameters of the radiation flowing through point x in
where Ii (ν, Ω)
the direction Ω, e (Ω)
defined with respect to the unit vectors e (Ω), of Fig. 14.4.
a b
The radiative transfer coefficients are given by Eqs. (6.86) and (6.87). Their explicit
expressions for the case we are concerned with can be derived from Eqs. (7.15) by
modifying the Φ profiles according to Eq. (14.1). However, since we have assumed
that the magnetic field is weak (νL ∆νD ), we can use Eqs. (7.16) in the place
of Eqs. (7.15). The φ profile in the expressions of ηiA and εi is a Voigt profile
centered at the transition frequency ν0 , because the velocity distribution of the
atoms has been assumed to be Maxwellian. But since the inverse lifetime of the
upper level has also been supposed much smaller than ∆νD , the Voigt profile can
be approximated by the Gaussian profile p(ν0 − ν) of Eq. (14.6).1 On the other
hand, the assumption of unpolarized lower level implies that the summation over
= δ (see
K and Q in Eq. (7.16a) is restricted to K = Q = 0. And since T 00 (i, Ω) i0
Table 5.6), we obtain
hν
ηiA = N (2J + 1) B(α J → αu Ju )
4π
!
√ 1 1 0
× 3 (−1)1+J +Ju ρ00 (α J ) p(ν0 − ν) δi0 ,
J J Ju
or, using Eqs. (2.36a), (10.6) and (9.5)
ηiA = kLA p(ν0 − ν) δi0 ,
1
It should be remarked that such substitution, based on the assumptions νL ∆νD and
γu ∆νD , implies the same order of approximation as the flat-spectrum approximation that we
have used to derive the statistical equilibrium equation.
748 CHAPTER 14
enters the medium at point x , where its Stokes parameters
Fig.14.5. A ray with direction Ω 0
(b) At point x, the Stokes parameters are given by Eq.(14.16).
are Ii (ν, Ω).
where kLA is the frequency-integrated absorption coefficient in the line,
hν
N B(α J → αu Ju ) ,
kLA = (14.15)
4π
N being the number density of atoms in the lower level. It can easily be seen
that the unpolarized lower level assumption also implies that the coefficients ρAi
are identically zero. It follows that the propagation matrix is diagonal,
A
Kij = kLA p(ν0 − ν) δij .
By similar arguments we obtain
εi = kLA p(ν0 − ν) wJ(K)J T Q
K S K (x ) ,
(i, Ω)
u
Q
KQ
where the quantities wJ(K)J and SQ K
(x ) are defined by Eqs. (10.11) and (14.10),
u
respectively.
Obviously, the preceding expressions imply that the only contribution to the
opacity and the emissivity of the medium comes from transitions between the two
levels of the model atom. The case where a source of continuum opacity (and
emissivity) is also present is formally more complicated and will not be treated
here.
Since the propagation matrix is diagonal, Eq. (14.14) reduces to a system of
four uncoupled equations whose solution can be expressed in terms of the scalar
attenuation operator of Eq. (8.16). Using Eq. (8.18) and referring to Fig. 14.5, we
can write the Stokes parameters at point x of the radiation at frequency ν flowing
in the form
along the direction Ω
x
−τν ( x )
x,
=
Ii (ν, Ω) p(ν0 − ν) kLA (x ) e wJ(K)J T Q
K S K (x ) ds
(i, Ω) Q
u
x0
KQ
−τν ( x0 )
x, (b)
+e ,
Ii (ν, Ω) (14.16)
STELLAR ATMOSPHERES 749
(b) is the Stokes vector of the radiation entering the medium at point
where Ii (ν, Ω)
x0 along the direction Ω, s is the coordinate of x reckoned along Ω
(s = |x −x |),
0
and τν (x, x ) is the optical depth at frequency ν between points x and x ,
x
τν (x, x ) = p(ν0 − ν) kLA (x ) ds . (14.17)
x
From Eq. (14.16) it is now possible to find the expression for the radiation field
tensor at point x. Using Eqs. (14.5) and (5.157) we obtain two contributions,
arising from the two terms in the right-hand side of Eq. (14.16). We can write
K K
J¯Q
K
(ν0 ) = J¯Q (ν0 ) I + J¯Q (ν0 ) E , (14.18)
where the ‘internal’ part [J¯Q
K
(ν0 )]I is given by
∞ 5
x
K dΩ K
3
J¯Q (ν0 ) I = dν p(ν0 − ν) T Q (i, Ω) ds p(ν0 − ν)
4π i=0
−∞
x0
−τν ( x)
x,
× kLA (x ) e wJ(KJ) T Q
K K
x ) ,
(i, Ω) SQ ( (14.19)
u
K Q
and the ‘external’ part [J¯Q
K
(ν0 )]E , originating from the boundary conditions, by
∞ 5
K dΩ K −τν (x,x0 ) (b)
3
J¯Q (ν0 ) E = dν p(ν0 − ν) T (i, Ω) e Ii (ν, Ω) . (14.20)
4π i=0 Q
−∞
Equation (14.19) can be cast in a simpler form by changing the double integral in
dΩ and ds into a volume integral. Since
d3 x = (x − x )2 dΩ ds , (14.21)
we get
∞ −τ ( x )
K k A (x ) e ν
x,
J¯Q (ν0 ) I = dν [p(ν0 − ν)] 2 3
d x L
4π(x − x )2
−∞
3
× TQ
K
(i, Ω) wJ(KJ) T Q
K K
x ) .
(i, Ω) SQ ( (14.22)
u
i=0 K Q
Finally, we can substitute the expression of the radiation field tensor at point x
into the statistical equilibrium equation. From Eqs. (14.11), (14.18), and (14.22)
750 CHAPTER 14
we obtain
K
1 + + δu(K) SQ (x ) + iHu KQQ
K K
SQ (
x) =
Q
K
= δK0 δQ0 BP (T ) + wJ (K)
J (−1)Q J¯−Q (ν0 ) E
u
kLA (x )
+ d3 x GKQ,K Q (x, x ) SQ
K
x ) ,
( (14.23)
4π(x − x )2
K Q
where
∞
2 −τ (x,x )
GKQ,K Q (x, x ) = dν p(ν0 − ν) e ν
−∞
3
× wJ(K)J wJ(KJ)
(−1)Q T−Q
K T K (i, Ω)
(i, Ω) . (14.24)
u u
Q
i=0
The quantities GKQ,K Q (x, x ) appearing in this equation can be expressed in
terms of Wigner symbols and of rotation matrices. Details are given in App. 20.
From the physical point of view, they represent a numerical factor which weights the
amount of coupling between the statistical tensor ρK Q at point x and the statistical
K
tensor ρQ at point x . For this reason, they can be referred to as multipole coupling
coefficients.
Equation (14.23) is a system of linear, non-homogeneous, integral equations in the
K
unknowns SQ (x ), the irreducible components of the source function, which can in
principle be solved once the properties of the medium and the boundary conditions
are specified. When the values of these components are known at each point, the
Stokes parameters of the radiation emerging from the medium can be computed
by applying Eq. (14.16). It should be remarked that, owing to property (A20.8)
of the multipole coupling coefficients, Eq. (14.23) decouples in two different sets
of equations involving, respectively, the components with K = 0, 2 and those with
K = 1. In the latter set, the only source term is [J¯−Q 1
(ν0 )]E , which vanishes unless
the boundary radiation field, once integrated in frequency according to Eq. (14.20),
has some contribution from circular polarization. Excluding this case of limited
1
interest, all the components SQ (x ) are everywhere zero in the medium.
It is especially interesting to study Eq. (14.23) in the particular case of a plane-
parallel, semi-infinite stellar atmosphere. In this case all the physical quantities of
the medium depend on a single coordinate, the height in the atmosphere, that we
assume as the z-axis of our reference system Σ. As a consequence, the irreducible
components of the source function depend only on the height z. Introducing the
line optical depth tL through the equation (cf. Eqs. (9.34))
kLA
dtL = − dz , (14.25)
∆νD
STELLAR ATMOSPHERES 751
and assuming that the stellar atmosphere is not illuminated by external sources of
radiation, Eq. (14.23) takes the form
K
1 + + δu(K) SQ (tL ) + iHu KQQ
K K
SQ (tL ) = δK0 δQ0 BP (T )
Q
∞ ∞ ∞
∆νD
+ dtL dx dy GKQ,K Q (x, x ) SQ
K
(tL ) , (14.26)
4π(x − x )2
0 −∞ −∞ K =0,2 Q
where the indices K and K are restricted to the values 0 and 2. The integral
over x and y can be performed by introducing cylindrical coordinates. These
calculations, that are developed in detail in App. 21, restrict the summation over
Q in the last term of Eq. (14.26) to the single value Q = Q and lead to the
following equation
K
1 + + δu(K) SQ (tL ) + iHu KQQ
K K
SQ (tL ) =
Q
∞
K
= δK0 δQ0 BP (T ) + GKQ,K Q |tL − tL | SQ (tL ) dtL , (14.27)
K =0,2 0
where we have introduced the multipolar kernels GKQ,K Q defined by
∞ ∞
∆νD
GKQ,K Q |tL − tL | =
dx dy G x, x ) .
( (14.28)
4π(x − x )2 KQ,K Q
−∞ −∞
These quantities are real and satisfy the symmetry properties (cf. Eqs. (A20.5)
and (A20.6))
GK Q,KQ (x) = GK −Q,K −Q (x) = GKQ,K Q (x) . (14.29)
The explicit expressions of the multipolar kernels appearing in Eq. (14.27) can be
derived from Eqs. (A21.8), (A21.7) and (A20.11). We obtain
∞
1 2
G00,00 (x) = dv ϕ(v) E1 x ϕ(v)
2
−∞
∞
1 2
G00,20 (x) = √ wJ(2)J dv ϕ(v) 3E3 x ϕ(v) − E1 x ϕ(v)
4 2 u
−∞
∞
1 2
G20,20 (x) = W2 (J , Ju ) dv ϕ(v) 5E1 x ϕ(v) −12E3 x ϕ(v) +9E5 x ϕ(v)
8
−∞
∞
3 2
G21,21 (x) = W (J , J ) dv ϕ(v) E1 x ϕ(v) + E3 x ϕ(v) − 2E5 x ϕ(v)
8 2 u
−∞
752 CHAPTER 14
∞
3 2
G22,22 (x) = W (J , J ) dv ϕ(v) E1 x ϕ(v) + 2E3 x ϕ(v)
16 2 u
−∞
+ E5 x ϕ(v) . (14.30)
The remaining components can be deduced from Eqs. (14.29). In these expressions,
En is the nth exponential integral defined in Eq. (12.6), the symbol W2 (J , Ju ) is
defined in Eq. (10.17), v and ϕ(v) are the reduced frequency and the Gaussian
profile of Eq. (A21.5),
ν0 − ν 1 −v2
v= , ϕ(v) = √ e . (14.31)
∆νD π
Recalling that
∞
1
En (x) dx = (n = 1, 2, . . .) ,
n
0
and that
∞
ϕ(v) dv = 1 ,
−∞
we obtain the integral properties
∞ ∞
1
G00,00 (x) dx = , G00,20 (x) dx = 0 ,
2
0 0
∞
7
G2Q,2Q (x) dx = W (J , J ) (Q = 0, ±1, ±2) . (14.32)
20 2 u
0
The multipolar kernels can be evaluated from Eqs. (14.30) via numerical computa-
tions. Figure 14.6 shows a plot of these quantities obtained by Gauss-Hermite inte-
gration. The plot refers to the transition (J = 0, Ju = 1) which implies wJ(2)J = 1
u
and W2 (J , Ju ) = 1.
Equation (14.27) can be solved by means of different numerical techniques. A
possible algorithm is similar to the one described in App. 16 for the solution of the
Hopf equation and is based on a discretization grid on tL and the transformation of
the system of integral equations into a linear system of algebraic equations. Details
can be found in Bommier et al. (1991). Other methods have also been devised,
among which we wish to quote, for their speed and accuracy, the operator-splitting,
iterative methods described by Trujillo Bueno and Manso Sainz (1999).
Once the irreducible components of the source function are known, the Stokes
parameters profiles of the radiation emerging from the atmosphere can be deduced
STELLAR ATMOSPHERES 753
Fig.14.6. The multipolar kernels defined in Eqs.(14.30) are plotted as a function of x for the
transition (J = 0, Ju = 1). The curves are labelled according to the following code: 1) G00,00 ;
2) G00,20 ; 3) G20,20 ; 4) G21,21 ; 5) G22,22 . Note the sign change of curve 2 which occurs at x =
0.75125.
by applying Eq. (14.16) to the case of a plane-parallel atmosphere. This gives,
setting µ = cos θ
∞ tL ϕ(v)
= −
S K (t ) dtL
Ii (ν, Ω) ϕ(v) e µ
wJ(K)J T Q
K
(i, Ω) Q L ,
u µ
0 K=0,2 Q
or, introducing the variable τν = tL ϕ(v)/µ
∞ , µτ -
−τν
=
Ii (ν, Ω) wJ J T Q
(K)
K
(i, Ω) e K
SQ ν
dτν . (14.33)
u ϕ(v)
K=0,2 Q 0
parallel to the stellar
In particular, for the radiation emerging along a direction Ω
surface one has, by considering the limit µ → 0
(K)
)=
Ii (ν, Ω wJ J T QK ) S K (0) .
(i, Ω (14.34)
u
Q
K=0,2 Q
Since under this limit the irreducible components of the source function appearing
in Eq. (14.33) become independent of optical depth, the Stokes parameters of the
emerging radiation turn out to be independent of frequency. This is not the case
when continuum opacity is taken into account (see Sect. 14.7).
Particularly important for the theory of radiative transfer of polarized radiation
is the special case of Eq. (14.27) which is obtained by setting Hu = δu(K) = 0 and by
754 CHAPTER 14
assuming and BP (T ) independent of tL . One is here faced with a homogeneous,
isothermal, non-magnetic, non-depolarizing atmosphere, and Eq. (14.27) takes the
form1
∞
(1 + ) S00 (tL ) = BP + G00,00 |tL − tL | S00 (tL ) dtL
0
∞
+ G00,20 |tL − tL | S02 (tL ) dtL
0
∞
(1 + ) S02 (tL ) = G20,00 |tL − tL | S00 (tL ) dtL
0
∞
+ G20,20 |tL − tL | S02 (tL ) dtL . (14.35)
0
This system of equations was established – though with different notations – by
Stenflo and Stenholm (1976) and by Rees (1978). It can be considered a typical
bench-mark problem in radiative transfer for polarized radiation because, as proved
in App. 22 (see Eq. (A22.13) with δu(2) = 0), the irreducible components of the source
function at the surface of the atmosphere obey the following √ relation, originally due
to Ivanov (1990) and often referred to as the generalized - law 2
.
2 2 √
S00 (0) + S02 (0) = BP = BP ,
1+
which can conveniently be used to test the accuracy of the numerical method
employed for solving the system.
Numerical results on the integration of the system of Eqs. (14.35) and on the en-
suing spectropolarimetric profiles have been published by different authors (Rees,
1978; Faurobert-Scholl and Frisch, 1989; Ivanov, 1990; Bommier et al., 1991; Bom-
mier and Landi Degl’Innocenti, 1996; Ivanov et al., 1997; Trujillo Bueno and Manso
Sainz, 1999). Other numerical results on the profiles have been obtained as partic-
ular cases of more general approaches involving partial redistribution effects (see
Sect. 14.8 and references therein). In Figs. 14.7-14.9 we show some examples for
the transition (J = 0, Ju = 1).
A particularly significant numerical result is the surface value of the ratio of the
two components of the source function, σ = S02 (0)/S00 (0). This ratio is related
1 2 with
In the system of Eq. (14.27) with Hu = 0, the equations involving the components SQ
Q = 0 are decoupled from the equation containing the source term BP (T ). As a consequence,
such components are zero.
2 √
In our notations, this law should better be referred to as the generalized - law.
STELLAR ATMOSPHERES 755
Fig.14.7. The logarithm of the component S00 (tL ) of the source function (normalized to BP ), and
the ratio S02 (tL )/S00 (tL ) – upper and lower panel, respectively – are plotted against the logarithm
of optical depth. The curves are obtained by solving Eqs.(14.35) – homogeneous, isothermal, non-
magnetic, non-depolarizing atmosphere – for the transition (J = 0, Ju = 1) and for two different
values of .
to the fractional linear polarization observed in the tangential direction by the
relation, easily obtained from Eq. (14.34)1
)
Q(ν, Ω 3 wJ(2)J σ
p ≡ = √ u
. (14.36)
)
I(ν, Ω 2 2 − wJ(2)J σ
u
For the transition (J = 0, Ju = 1) – which implies wJ(2)J = 1 – and for = 10−4 ,
u
the published values of σ are 0.08294 (Bommier and Landi Degl’Innocenti, 1996)
and 0.08292 (Trujillo Bueno and Manso Sainz, 1999), corresponding to the p values
1
In Eq. (14.36), the reference direction for positive Q is parallel to the stellar surface (θ = 90◦ ,
γ = 90◦ in Table 5.6).
756 CHAPTER 14
Fig.14.8. The emerging profiles of the intensity (normalized to BP ) and of the ratio Q/I – left
and right panel, respectively – corresponding to the irreducible components of the source function
shown in Fig.14.7. The profiles, obtained from Eq.(14.33) for a heliocentric angle θ = 84◦ .3
(µ = 0.1), are plotted against the reduced frequency (ν0 − ν)/∆νD . The positive Q direction is
parallel to the stellar surface.
Fig.14.9. Limb-darkening curve for the fractional polarization at line center, corresponding to the
irreducible components of the source function of Fig.14.7.
0.09063 and 0.09061, respectively. These values are in agreement with the result
of Ivanov et al. (1997), who give for p the asymptotic expression
√
p = 0.09443 − 0.3805 .
Apart from the special case of Eqs. (14.35), it is important to remark that
Eq. (14.27) holds without any limitation on the optical depth variation of the
different parameters. The quantities , δu(K) , Hu , BP (T ), θB , and χB , which appear
either explicitly or implicitly (through the kernel KQQ K
) in this equation, can be
arbitrary functions of optical depth. A full analysis of the solution of Eq. (14.27) for
different atomic transitions and for different values (and depth dependence) of the
STELLAR ATMOSPHERES 757
above-mentioned parameters is well outside the purposes of this book. Some results
can be found in Bommier et al. (1991) and in Bommier and Landi Degl’Innocenti
(1996).
For further applications involving turbulent magnetic fields, it is convenient to
rewrite Eq. (14.27) by introducing as unknowns the irreducible components of
the source function expressed in the local magnetic frame.1 Such components
K
will be denoted by the symbol [SQ (tL )]M and are still given by Eq. (14.10), the
statistical tensors in the right-hand side being now defined in the local magnetic
frame. Referring to the geometry of Fig. 14.4, and denoting by RB the rotation that
carries the magnetic frame at tL into the fixed reference frame Σ (see Eq. (14.4)),
we have from Eqs. (7.76) and (7.77)
∗
K
SQ (tL ) = SQK
(tL )
M
DQ
K
Q (RB ) (14.37)
Q
(tL ) DQQ (RB ) .
K K K
SQ (tL ) M = SQ (14.38)
Q
Multiplication of Eq. (14.27) by DQK
Q (RB ) and summation over Q yields, with the
use of Eqs. (14.37), (14.38), (14.3), and (2.72)
K
1 + + δu(K) + iHu Q SQ (tL ) M = δK0 δQ0 BP (T )
∞
K
+ ĜKQ,K Q |tL − tL | SQ (tL )
M
dtL , (14.39)
K =0,2 Q 0
where
ĜKQ,K Q |tL − tL | =
K ∗
= DQQ
K
(RB ) GKQ ,K Q |tL − tL | DQ Q (RB ) , (14.40)
Q
RB being the rotation that carries the magnetic frame at tL into the frame Σ.
The kernels ĜKQ,K Q are a generalization of the multipolar kernels GKQ,K Q
introduced in Eq. (14.28). Since these quantities depend on the magnetic field’s
direction at different optical depths, in general they do not satisfy specific symme-
try properties. However, if the magnetic field’s direction is independent of optical
depth, and if the x-axis of the magnetic frame is chosen to lie in the plane con-
taining the vertical to the atmosphere and the magnetic field vector itself (γB = 0
in Eq. (14.4)), it can be shown, using Eqs. (14.29) and (2.68), that the kernels
ĜKQ,K Q are real and obey the symmetry property
ĜK Q ,KQ (x) = ĜKQ,K Q (x) . (14.41)
1
The local magnetic frame has the z-axis directed along the local magnetic field.
758 CHAPTER 14
Moreover, they obey the integral properties
∞ ∞
1
Ĝ00,00 (x) dx = , Ĝ00,2Q (x) dx = 0 ,
2
0 0
∞
7
Ĝ2Q,2Q (x) dx = δQQ W (J , J ) , (14.42)
20 2 u
0
which follow from Eqs. (14.32) and (2.72).1 As shown
√ in App. 22, these properties
allow one to prove a further generalization of the - law (see Eq. (A22.14)), hold-
ing in the case of a homogeneous, isothermal atmosphere permeated by a depth-
independent magnetic field.
An important application of Eq. (14.39) concerns the case of an atmosphere
embedded in a turbulent magnetic field. To deal with this case, it is convenient to
cast the equation in the equivalent form
K
SQ (tL ) M = δK0 δQ0 BP (T )
ĜKQ,K Q |tL − tL |
∞
+ (K)
K
SQ (tL )
M
dtL ,
K =0,2 Q
1 + + δ u + iH u Q
0
and then to return to the fixed reference system Σ through Eqs. (14.37) and (14.38).
As a result of this back-transformation we obtain the further equivalent equation
K
SQ (tL ) = δK0 δQ0 BP (T )
∞
K
+ K
MQQ (tL ) GKQ ,K Q |tL − tL | SQ
(tL ) dtL , (14.43)
K =0,2 Q 0
where
1
∗
K
MQQ (tL ) = DQ
K
Q (RB ) DQ Q (RB )
K
.
Q
1 + + δu + i Q Hu
(K)
K
The quantity MQQ (tL ) now introduced bears a strong resemblance with the mag-
netic kernel MK
QQ B) defined in Eq. (5.166). Indeed, from Eqs. (14.4) and (2.71)
(
we have
1
K
MQQ (tL ) = DQQ
K
(χB θB γB ) DQ Q (−γB −θB −χB )
K
,
Q
1 + + δu + i Q Hu
(K)
1
Equations (14.42) are valid (provided the magnetic field’s direction is depth-independent)
irrespective of the value of γB .
STELLAR ATMOSPHERES 759
which is the same as Eq. (5.166) except for a slight difference in the denominator
of the last factor.
Equation (14.43) can conveniently be used to handle the case of an atmosphere
pervaded by a microturbulent magnetic field. By ‘microturbulent’ we mean here
that, at any optical depth, the magnetic field has an isotropic distribution of direc-
tions and that the direction at optical depth tL is uncorrelated with the direction
at optical depth tL , no matter how close the two points are in space. Under this
hypothesis, and under the further assumption that the quantities , δu(K) , BP , and
Hu are deterministic, one obtains, averaging over the distribution of magnetic fields
at each depth
K
SQ (tL ) = δK0 δQ0 BP (T )
∞
K
+ K
MQQ (tL ) GKQ ,K Q |tL − tL | SQ (tL ) dtL . (14.44)
K =0,2 Q 0
The components of the averaged magnetic kernel MQQ
K
(tL ) are easily obtained
by comparison with Eqs. (5.170)-(5.173). One gets
0 1 2
1
M00 (tL ) = , MQQ (tL ) = δQQ µ (t ) , (14.45)
1+ 1 + + δu(2) 2 L
where
1 2 2
µ2 (tL ) = 1+ + ,
5 1 + (Hu )2 1 + 4(Hu )2
with
Hu
Hu = .
1 + + δu(2)
Using Eqs. (14.45) and (14.30), it is easily seen that Eq. (14.44) reduces to a
system of coupled integral equations involving only the two quantities S00 (tL )
and S02 (tL ), namely
∞
(1 + ) S00 (tL ) = BP (T ) + G00,00 |tL − tL | S00 (tL ) dtL
0
∞
+ G00,20 |tL − tL | S02 (tL ) dtL
0
∞
(1 + + δu(2) ) S02 (tL ) = µ2 (tL ) G20,00 |tL − tL | S00 (tL ) dtL
0
∞
+ G20,20 |tL − tL | S02 (tL ) dtL . (14.46)
0
760 CHAPTER 14
Apart from the presence of the factor µ2 (tL ) in the right-hand side of the second
equation, this is the same as the system holding for S00 (tL ) and S02 (tL ) in a non-
magnetic atmosphere (cf. Eqs. (14.35)),1 and obviously reduces to it under the
limit Hu → 0. Formally, one can pass from the non-magnetic to the ‘turbulent’
atmosphere by the simple transformations
G20,00 |tL − tL | → µ2 (tL ) G20,00 |tL − tL |
G20,20 |tL − tL | → µ2 (tL ) G20,20 |tL − tL | .
It is important to remark that, even in the case of a depth-independent turbulent
field (µ2 (tL ) = const.), these transformations imply G00,20 (x) = G20,00 (x). As a
√
consequence, the generalized - law does not hold for the turbulent-field atmo-
sphere.
One can also consider the case of a horizontal magnetic field with random azimuth
distribution. Following the same line of reasoning, one still gets Eqs. (14.46) with
the depolarizing factor µ2 (tL ) replaced by
1 3
µ2 (tL ) = 1+ .
4 1 + 4(Hu )2
14.3. The Two-Level Atom: Non-LTE Theory for
Strong Magnetic Fields
The results obtained in the former section are restricted to a physical regime where
the magnetic field is either zero or, if present, is so weak that the Larmor frequency
νL is much smaller than the Doppler broadening ∆νD . When this inequality is not
satisfied, one enters a different regime that can be simply referred to as the ‘strong
field regime’.2 From one side, the Larmor frequency is much larger than the in-
verse lifetime of the upper level γu , and from the other side it is comparable with
or even larger than ∆νD . As explained in Sect. 13.1, the formalism developed in
this book – based on the flat-spectrum assumption – can still be applied in this
physical regime provided all interferences between different Zeeman sublevels are
neglected a priori. Under this approximation, the atom is described only in terms
of populations of Zeeman sublevels, i.e., only by the diagonal elements of the den-
sity matrix ραJ (M, M ) in the standard representation, or only by the components
ρK
0 (αJ) in the statistical tensor representation.
1 (2)
In Eqs. (14.35), δu was set to zero. Moreover, and BP (T ) were assumed to be
depth-independent, but this does not change the structure of the equations.
2
According to the classification proposed in Sect. 5.16, we deal here with regimes III, IV and
V of panel (a) in Fig. 5.18.
STELLAR ATMOSPHERES 761
In this section we present another example of the general non-LTE problem of
the 2nd kind by considering, similarly to the previous section, a two-level atom
without hyperfine structure and with unpolarized lower level. We suppose again
that a collection of such atoms is distributed within a static medium of arbitrary
shape, where it interacts with a strong magnetic field and with a population of
colliding particles having a Maxwellian distribution of velocities. We assume, as
in the former section, that the atoms themselves have a Maxwellian distribution
of velocities characterized by the parameter ∆νD , and that the radiation field is
weak so that stimulated emission can be neglected. Finally, we adopt again the
approximation of complete redistribution on velocities and, to avoid the introduc-
tion of a too heavy formalism, we suppose that the Doppler broadening ∆νD and
the magnetic field vector are constant throughout the medium, whereas we do not
impose any restriction on the spatial variation of the other physical parameters.
The derivation presented below follows closely the line of reasoning of the former
section. A similar derivation has been given by Landi Degl’Innocenti et al. (1991).
At a point P of the medium, having coordinate x, we write the statistical equi-
librium equation for the multipole moments of the density matrix, defined in the
reference system of the magnetic field. To this aim, we use for the radiative rates
the expressions developed in Sect. 7.4 and, in particular, Eqs. (7.20a) and (7.20e),
whereas for the collisional rates we adopt the same expressions as those of the for-
mer section (see Sect. 7.13.d). Recalling the hypothesis of the unpolarized lower
level, neglecting stimulated emission, taking into account the approximation of
complete redistribution on velocities through the ‘recipe’ of Eq. (13.8), and per-
forming some algebra along the lines of the former section, one obtains, in station-
ary situations, the following equation
c2 2Ju + 1
1 + + δu(K) ρK0 (αu Ju ) =
x 2hν03 2J + 1
× 3(2Ju + 1)(2K + 1)(2K + 1) (−1)1+Ju −M
K Mu M q
2
Ju J 1 Ju Ju K 1 1 K
×
−Mu M −q Mu −Mu 0 q −q 0
!
× J¯0K (να J Mu , α J M ) + δK0 BP (T ) ρ00 (α J ) x , (14.47)
u u
with J¯0K (να J M , α J M ) defined according to Eqs. (14.5) and (14.6).
u u u
The quantity in square brackets in the right-hand side of Eq. (14.47) can be ex-
pressed in terms of the convolution of the generalized profiles defined in Eq. (10.40)
with the velocity distribution of atoms. Such convolution has already been intro-
duced in Chap. 13 and denoted by the symbol Φ̂KK Q (J , Ju ; ν) – see Eq. (13.20).
Here we define a slightly different quantity by setting, for Q = 0
762 CHAPTER 14
Φ̃KK
0 (J , Ju ; ν) = 3(2Ju + 1)(2K + 1)(2K + 1)
2
1+Ju −Mu +q Ju J 1 Ju Ju K
× (−1)
−Mu M −q Mu −Mu 0
Mu M q
1 1 K
× p(να J M , α J M − ν) , (14.48)
q −q 0 u u u
where the profile p is given by Eq. (14.6).1 With this position, Eq. (14.47) takes
the form
c2 2Ju + 1
1 + + δu(K) ρK0 (αu Ju ) =
x 2hν03 2J + 1
∞ !
K
0
× dν Φ̃KK
0 (J ,
uJ ; ν) J0 (ν) + δ K0 BP (T ) ρ0 (α J ) x ,
K −∞
or, introducing the irreducible components of the two-level source function defined
in Eq. (14.10)2
1 + + δu(K) S0K (x ) M = δK0 BP (T )
∞
+ dν Φ̃KK
0 (J , Ju ; ν) J0K (ν) . (14.49)
K −∞
Next we turn to the radiative transfer equation. Neglecting stimulated emission,
the Stokes parameters of the radiation propagating through point P along the
(defined according to the reference direction e (Ω)
direction Ω of Fig. 14.4) obey
a
the transfer equation (14.14), with the propagation matrix K A explicitly given by
(see Eqs. (6.86), (6.87))
⎛ ⎞
η0A (ν, Ω)
η1A (ν, Ω)
η2A (ν, Ω)
η3A (ν, Ω)
⎜ A ⎟
⎜ η1 (ν, Ω) η0A (ν, Ω)
ρA (ν, Ω) −ρA2 (ν, Ω) ⎟
KA = ⎜
⎜ A
3 ⎟ . (14.50)
⎝ η2 (ν, Ω) −ρA3 (ν, Ω)
η0A (ν, Ω) ⎟
ρA1 (ν, Ω) ⎠
η3A (ν, Ω)
ρA2 (ν, Ω)
−ρ1 (ν, Ω) η0 (ν, Ω)
A A
1
The two quantities Φ̂KK0 and Φ̃KK
0 just differ in the fact that the first is a weighted sum of
Voigt profiles centered at the individual frequencies of the Zeeman components of the transition,
while in the second the Voigt profiles are replaced by Gaussian profiles.
2 K (
Consistently with the notations employed in the former section, we use the symbol [SQ x)]M to
recall that the irreducible components of the source function are defined in the magnetic reference
system.
STELLAR ATMOSPHERES 763
The radiative transfer coefficients ηiA , ρAi , and εi can be found from Eqs. (7.21).
Taking into account the hypothesis of complete redistribution on velocities, ne-
glecting for consistency the finite width of the levels, and recalling the definition
of the symbol Φ̃KK
0 of Eq. (14.48), one has (see Eq. (A13.9))
= kA
ηiA (ν, Ω) Φ̃0K (J , J ; ν) ,
T 0K (i, Ω) (14.51)
L 0 u
K
where kLA is given by Eq. (14.15). Similarly (see Eq. (A13.14))
= kA
ρAi (ν, Ω) Ψ̃ 0K (J , J ; ν) ,
T 0K (i, Ω) (14.52)
L 0 u
K
where Ψ̃00K (J , Ju ; ν) is the same as Φ̃0K
0 (J , Ju ; ν) except for the replacement of
the Gaussian profile p(ν0 − ν) with the associated dispersion profile q(ν0 − ν) given
by (see Eqs. (5.44) and (5.55))
2 ν0 − ν
q(ν0 − ν) = D , (14.53)
π ∆νD ∆νD
where D is the Dawson function defined in Eq. (5.56). Finally (see Eqs. (A13.8),
(7.8), (14.10), (10.6), and (14.15))
= kA
εi (ν, Ω) S K (x ) Φ̃KK (J , J ; ν) .
T 0K (i, Ω) (14.54)
L 0 M 0 u
KK
The radiative transfer equation can be solved by the methods developed in Chap. 8.
Substituting Eq. (14.54) into Eq. (8.14) and referring to Fig. 14.5, one obtains, for
the Stokes parameters of the radiation propagating through point P along the
direction Ω
x
3
Ii (ν, Ω) = Oij (x, x ; ν) kLA (x ) S K (x ) Φ̃KK (J , J ; ν) ds
T 0K (j, Ω) 0 M 0 u
x0 j=0
KK
3
(b)
+ ,
Oij (x, x0 ; ν) Ij (ν, Ω) (14.55)
j=0
where Oij (x, x ; ν) is the evolution operator between points x and x, s is the
coordinate of x reckoned along the direction Ω, and I (b) (ν, Ω)
j
are the Stokes
parameters of the radiation entering the medium at point x0 and directed along Ω.
Because of the assumption of constant magnetic field and Doppler broadening, the
propagation matrix K A of Eq. (14.50) can be written, according to Eqs. (14.51)
and (14.52), in the form
,
K A = kLA H(ν, Ω) (14.56)
764 CHAPTER 14
where kLA depends on x, while the matrix H depends on frequency and direction
but is independent of x. It follows that the evolution operator has the form of
Eq. (8.22),
−τ ( x ) H(ν,Ω)
x,
Oij (x, x ; ν) = e , (14.57)
ij
where
x
τ (x, x ) = kLA (x ) ds .
x
Using Eq. (14.55), it is now possible to find an expression for the radiation field
tensor J0K (ν) to be substituted into the statistical equilibrium equation (14.49).
From the definition in Eqs. (5.157) one gets
J0K (ν) = J0K (ν) I + J0K (ν) E , (14.58)
where the ‘internal’ part [J0K (ν)]I is given by
5 x
K dΩ K
3 3
J0 (ν) I = T 0 (i, Ω) ds Oij (x, x ; ν) kLA (x )
4π i=0 j=0
x0
× S K (x ) Φ̃K K (J , J ; ν) ,
T 0K (j, Ω) 0 M 0 u
K K
and the ‘external’ part [J0K (ν)]E by
5
K dΩ K
3 3
(b) .
J0 (ν) E = T 0 (i, Ω) Oij (x, x0 ; ν) Ij (ν, Ω)
4π i=0 j=0
Now we substitute Eq. (14.58) into Eq. (14.49) and transform the double integral
over dΩ and ds into an integral in d3 x (see Eq. (14.21)). This leads to the
following equation
∞
(K)
1 + + δu S0K (x ) M = δK0 BP (T ) + dν Φ̃KK
0 (J , Ju ; ν) J0K (ν) E
K −∞
kLA (x )
+ d3 x
G̃K0,K 0 (x, x ) S0K (x ) M , (14.59)
4π(x − x ) 2
K
STELLAR ATMOSPHERES 765
where
∞
K
G̃K0,K 0 (x, x ) = dν Φ̃KK
0 (J , Ju ; ν) Φ̃K
0 (J , Ju ; ν)
K K −∞
3
3
× O (x, x ; ν) T K (j, Ω)
T 0K (i, Ω) . (14.60)
ij 0
i=0 j=0
The quantities G̃K0,K 0 are a natural generalization (though restricted to the values
Q = Q = 0) of the multipole coupling coefficients GKQ,K Q defined in Eq. (14.24).
Their main properties are collected in App. 23.
Equation (14.59) is a system of linear, non-homogeneous, integral equations in the
unknowns [S0K (x )]M . When the local properties of the medium and the boundary
conditions are specified, the system can be solved, and the Stokes parameters of
the radiation emerging from the medium can be computed by means of Eq. (14.55).
Owing to property (A23.6), Eq. (14.59) decouples in two distinct sets of equations,
involving the irreducible components of the source function of even rank (K =
0, 2, . . .) and of odd rank (K = 1, 3, . . .), respectively. In the latter, the only
source terms may be provided by the radiation entering the medium through the
boundary, and it can be shown, using property (A13.4) of the generalized profiles,
that such terms are zero unless the boundary radiation field either contains some
net circular polarization, or is characterized by I, Q, and U Stokes parameters that
are not symmetrical about line center. Excluding these cases, all the components
[S0K (x )]M with K odd are everywhere zero in the medium, and one is left with the
K
unknowns [S00 (x )]M , [S02 (x )]M , . . . , [S0 max (x )]M , where
2Ju if Ju is integer
Kmax = (14.61)
2Ju − 1 if Ju is half-integer .
Similarly to the previous section, we can apply Eq. (14.59) to the special case of a
plane-parallel, semi-infinite stellar atmosphere. Introducing the line optical depth
of Eq. (14.25), and assuming that the atmosphere is not illuminated by external
sources of radiation, we obtain
1 + + δu(K) S0K (tL ) M = δK0 BP (T )
∞
+ G̃K0,K 0 |tL − tL | S0K (tL ) M dtL , (14.62)
K =0,2,...,Kmax 0
where the kernels G̃K0,K 0 are given by (cf. Eq. (14.28))
∞ ∞
∆νD
G̃K0,K 0 |tL − tL | =
dx dy G̃ x, x ) .
( (14.63)
4π(x − x )2 K0,K 0
−∞ −∞
766 CHAPTER 14
Performing some algebra similar to that developed in App. 21, Eq. (14.63) can be
cast in the form
π/2 2π
∆νD
G̃K0,K 0 (x) = dθ tan θ dχ
4π
0 0
3
3 ∞
K
× dν Φ̃KK
0 (J , Ju ; ν) Φ̃K
0 (J , Ju ; ν)
i=0 j=0 K K −∞
x ∆νD
− H(ν,Ω)
× T 0K (i, Ω) e cos θ .
T 0K (j, Ω) (14.64)
ij
The kernels G̃K0,K 0 (x) obey some important properties. The first is the symme-
try property
G̃K 0,K0 (x) = G̃K0,K 0 (x) , (14.65)
which follows directly from the analogous property of the generalized multipole
coupling coefficients (Eq. (A23.7)).
A second property concerns the integrals of some of the kernels G̃K0,K 0 . One
has
∞ ∞
1
G̃00,00 (x) dx = , G̃00,20 (x) dx = 0 . (14.66)
2
0 0
To prove these properties, we observe that, using Eq. (9.106)
∞
−
x ∆νD
H(ν,Ω)
cos θ −1
e cos θ dx = H (ν, Ω) ij ,
ij ∆νD
0
where H −1 is the inverse of the matrix H. On the other hand, from Eqs. (14.50),
(14.51), and (14.56), we have
Φ̃0K = H(ν, Ω)
(J , Ju ; ν) T 0K (i, Ω) , (14.67)
0 0i
K
so that
∞ π/2 2π
∞
1
K
.
G̃00,K 0 (x) dx = dθ sin θ dχ dν Φ̃K
0 (J , Ju ; ν) T 0K (0, Ω)
4π
0 0 0 K −∞
Taking finally into account Eq. (A13.3), and considering that (see Table 5.6)
π/2 2π π/2 2π
1 = 1 , 1 =0,
dθ sin θ dχ T 00 (0, Ω) dθ sin θ dχ T 02 (0, Ω)
4π 2 4π
0 0 0 0
STELLAR ATMOSPHERES 767
we obtain the results in Eq. (14.66).
The properties expressed by Eqs. (14.65) and (14.66) are particularly important
because they√allow one to prove that, for a homogeneous and isothermal atmo-
sphere, the - law can be √ extended to the strong field regime analyzed in this
section. Such generalized - law is given by Eq. (A22.15).
It should be remarked that Eq. (14.62) holds without any limitation on the
variation of the different parameters with optical depth. The quantities , δu(K)
and BP appearing in the equation can be arbitrary functions of optical depth
(contrary to the magnetic field vector which is assumed to be constant). A full
analysis of the solution of Eq. (14.62) for different atomic transitions, different
values and/or depth dependence of the above-mentioned parameters, and different
inclinations and strengths of the magnetic field vector is well outside the aims of this
book. Some numerical results can be found in Bommier and Landi Degl’Innocenti
(1996). It should also be mentioned that integral equations similar to Eq. (14.62)
and containing the basic physical aspects of the problem of line formation in a
(strong) magnetic field with population unbalances among Zeeman sublevels were
derived, using more standard formalisms, by Domke (1969, 1971) and by Landi
Degl’Innocenti (1978b).
14.4. The Non-LTE Regime of Order 1.5
The formalism developed in the former section is suitable to discuss in some detail
the special case of the non-LTE problem that has been classified in Sect. 14.1 as
non-LTE of order 1.5. This is an intermediate non-LTE regime where it is assumed
that the atomic system is not polarized whereas the radiation field, because of the
presence of a magnetic field, is polarized. As mentioned in Sect. 14.1, this physical
situation has received considerable attention in the literature because it is consid-
ered appropriate to describe the process of formation of weak or medium-strong
spectral lines in relatively dense and strongly magnetized plasmas, like sunspots
or the atmospheres of magnetic stars. Owing to the relatively high density of
the plasma, it is believed that depolarizing collisions are so efficient to completely
destroy atomic polarization, which considerably simplifies the description of the
atomic system.
Restricting our analysis to a two-level atom, the non-LTE problem of order 1.5
reduces to the solution of the following equation (obtained from Eq. (14.62) under
the limit δu(K) → ∞ for K = 2, 4, . . . , Kmax , with Kmax given by Eq. (14.61))1
∞
(1 + ) S00 (tL ) = BP (T ) + G̃00,00 |tL − tL | S00 (tL ) dtL , (14.68)
0
where we have omitted the bracket [. . .]M around the symbol S00 (tL ) because this
1
Equation (14.62) holds for a two-level atom with unpolarized lower level. In the present
regime, both levels are unpolarized by definition.
768 CHAPTER 14
component of the source function does not depend on the reference system (see
Eq. (14.38)).
The explicit expression of the kernel can be obtained from Eq. (14.64) with the
help of Eq. (14.67). One gets
π/2 2π ∞
∆νD
G̃00,00 (x) = dθ tan θ dχ dν H , (14.69)
4π
0 0 −∞
where the quantity H is defined by
x ∆νD
− H(ν,Ω)
e
H = H(ν, Ω) cos θ
H(ν, Ω) . (14.70)
00
In order to simplify Eq. (14.68), two different approximations have been proposed
in the literature, referred to as the field-free and the polarization-free approxima-
tion, respectively. The former (Rees, 1969) is based on the substitution of the
quantity H with its limiting value obtained by setting to zero the magnetic field.
In other words, with evident notations
Hf.f. = lim H .
B→0
From Eqs. (14.51)-(14.52), using Eqs. (A23.2) and (A23.3), it can easily be seen
that under this limit the matrix H reduces to
H(ν, Ω) = δij p(ν0 − ν) ,
ij
where p(ν0 − ν) is the profile defined in Eq. (14.6). Substitution into Eq. (14.70)
shows that the kernel G̃00,00 takes the form
∞
1 2
G̃00,00 (x) = dv ϕ(v) E1 xϕ(v) ,
f.f. 2
−∞
where v and ϕ(v) are defined in Eq. (14.31), and where E1 is the first exponential
integral defined in Eq. (12.6). Obviously, under this approximation Eq. (14.68)
reduces to the standard equation of the non-LTE problem of the 1st kind, and
S00 (tL ) is just the corresponding source function.
The polarization-free approximation (Trujillo Bueno and Landi Degl’Innocenti,
1996) is in principle less restrictive. It consists in retaining the diagonal elements
of the matrix H(ν, Ω) while setting to zero the off-diagonal ones. In this approxi-
mation one has
H(ν, Ω) ,
= δij h0 (ν, Ω)
ij
is given by (cf. Eqs. (14.51) and (14.56))
where h0 (ν, Ω)
STELLAR ATMOSPHERES 769
A
= η0 (ν, Ω) =
h0 (ν, Ω) Φ̃0K (J , J ; ν) .
T 0K (0, Ω) 0 u
kLA
K
One then gets, substituting into Eqs. (14.69)-(14.70) and performing the integration
over χ
π/2 ∞
∆νD x ∆νD
2 e− cos θ
h0 (ν,Ω)
G̃00,00 (x) = dθ tan θ dν h0 (ν, Ω) .
p.f. 2
0 −∞
It should be remarked that, although the field-free and the polarization-free ap-
proximations have been discussed here in the special case of a two-level atom, they
are indeed more general and can be applied to multi-level atoms as well. Given a
model atmosphere, the field-free approximation, for instance, tackles the non-LTE
problem of order 1.5 through the following strategy: a) the statistical equilibrium
equations for the level populations are solved as if the magnetic field were not
present; b) from such values, and from the value of the magnetic field vector, one
finds the expressions of the radiative transfer coefficients; c) the radiative transfer
equation is finally solved for any desired line of sight. This procedure implies the
use of a standard non-LTE numerical code for the solution of the statistical equi-
librium equations (point a). The complications arising from the presence of the
magnetic field and from polarization only intervene at points b) and c), when the
self-consistency loop has already been solved. Given its simplicity, the field-free
approximation has been widely used in the past, especially to compute theoretical
Stokes profiles of weak or medium-strong spectral lines originating from magnetized
atmospheres of relatively high density, such as sunspots or Ap stars.
14.5. The Non-LTE Problem for More Complicated Atomic Models
In Sects. 14.2 and 14.3 we have analyzed the non-LTE problem of the 2nd kind
for a two-level atom with unpolarized lower level. The hypothesis of the absence
of atomic polarization in the lower level, joined with the further hypothesis of ne-
glecting stimulated emission, has allowed us to reduce the general non-LTE problem
(summarized by the self-consistency loop of Fig. 14.1) to the solution of a linear
system of integral equations. However, this linearity property is strictly confined to
the particular case just mentioned. It is no longer valid both when a model atom
with more than two levels is considered and, even for a two-level model atom, when
either the approximation of the unpolarized lower level or the approximation of ne-
glecting stimulated emission cannot be applied. In such cases, the solution of the
self-consistency loop has to be performed by means of numerical algorithms, and
it is usually impossible to establish general analytical results.
As an illustration of the problems involved, we consider in this section the sim-
plest possible atomic model leading to a non-linear set of equations, namely a
770 CHAPTER 14
two-level atom having J = 1 and Ju = 0. Following Trujillo Bueno and Landi
Degl’Innocenti (1997), we introduce the same hypotheses as those specified at the
beginning of Sect. 14.21 for treating the case of the two-level atom with unpolar-
ized lower level. In addition, we assume that there is no magnetic field, and that
the medium is cylindrically symmetrical about an axis (that we take as the z-axis
of our reference system) and is not illuminated from the boundary by circularly
polarized radiation. Under these conditions, the only non-zero components of the
radiation field tensor are J00 and J02 , and the only non-zero statistical tensors are
ρ00 (α J ), ρ20 (α J ), and ρ00 (αu Ju ) – cf. Sect. 10.7.
The statistical equilibrium equations can be derived from Eqs. (10.1) and (10.2)
by setting νL = 0 and by evaluating the rates for the transition (J = 1, Ju = 0).
Taking into account Eqs. (7.14a), (7.14b), (7.14d) and (7.14e), and performing on
the radiation field tensor the substitution of Eq. (13.8) implied by the approxima-
tion of complete redistribution on velocities, one obtains, for the statistical tensors
at point x
d 0
ρ0 (αu Ju ) x = − A(αu Ju → α J ) + CS(0) (α J , αu Ju ) ρ00 (αu Ju ) x
dt
√ & '
+ 3 B(α J → αu Ju ) J¯00 (ν0 ) ρ00 (α J ) x + J¯02 (ν0 ) ρ20 (α J ) x
√
+ 3 CI(0) (αu Ju , α J ) ρ00 (α J ) x
d 2
ρ (α J ) = −B(α J → αu Ju )
dt 0 x
& 1 '
× J¯02 (ν0 ) ρ00 (α J ) x + J¯00 (ν0 ) ρ20 (α J ) x − √ J¯02 (ν0 ) ρ20 (α J ) x
2
− CI(0) (αu Ju , α J ) + D(2) (α J ) ρ20 (α J ) x ,
where J¯Q
K
(ν0 ) is given by Eq. (14.5).
These two equations can be cast in a simpler form by introducing a scalar source
function, S(x ), defined by (cf. Eq. (14.10))
2hν03 √ ρ00 (αu Ju ) x
S(x ) = 2 3 0 , (14.71)
c ρ0 (α J ) x
and a parameter, σ(x ), describing the amount of alignment in the lower level
2
ρ0 (α J ) x
σ(x ) = 0 . (14.72)
ρ0 (α J ) x
1
The hypotheses are the following: static medium, atoms and colliding particles with
Maxwellian velocity distributions, constant Doppler broadening, weak radiation field (negligible
stimulated emission), flat-spectrum approximation, approximation of complete redistribution on
velocities, negligible continuum opacity.
STELLAR ATMOSPHERES 771
Use of Eqs. (10.49) and (7.8) allows one to rewrite the two equations, for stationary
situations, in the form
1
(1 + ) S(x ) = J¯00 (ν0 ) 1 + √ w̄ σ(x ) + BP (T )
2
1 w̄
σ(x ) = − √ , (14.73)
2 1− 1
2 w̄ + β + δ
where w̄ is the average anisotropy factor (cf. Eqs. (10.62))
√ J¯02 (ν0 )
w̄ = 2 ¯0 ,
J0 (ν0 )
is defined in Eqs. (14.7), BP (T ) is the Planck function in the Wien limit (see
Eq. (14.9)), and where the quantities β and δ are given by
B (T ) D(2) (α J )
β = ¯0P , δ = .
J0 (ν0 ) B(α J → αu Ju ) J¯00 (ν0 )
Equations (14.73) depend, either explicitly or implicitly through the parameters
w̄, β, and δ , on the components of the radiation field tensor, J¯00 (ν0 ) and J¯02 (ν0 ),
expressed at point x. These quantities can be evaluated by solving the radiative
transfer equation which, owing to the cylindrical symmetry of the medium, takes a
to lie in the plane
particularly simple form. Defining the reference direction ea (Ω)
containing the z-axis and the direction Ω (see Fig. 14.4 with γ = 0), and taking
into account Eqs. (7.16a) and (7.16e) and the hypothesis of complete redistribution
on velocities, one obtains
/ 0
d
I(ν, Ω) h0 h1
I(ν, Ω) ε0
= −kLA p(ν0 − ν) − , (14.74)
ds
Q(ν, Ω) h1 h0
Q(ν, Ω) 0
where kLA and p(ν0 − ν) are defined, respectively, in Eqs. (14.15) and (14.6), and
where h0 , h1 , and ε0 are given by
1
h0 = 1 + √ 3 cos2 θ − 1 σ(x )
2 2
3
h1 = − √ sin2 θ σ(x )
2 2
ε0 = S(x ) ,
θ being the angle between the z-axis and the direction Ω (see Fig. 14.4).
The radiative transfer equation can be solved through the methods outlined in
Chap. 8 (see Eq. (8.28)). For a plane-parallel, semi-infinite atmosphere, introducing
772 CHAPTER 14
the optical depth tL of Eq. (14.25) and setting µ = cos θ, one has, for the radiation
at optical depth tL propagating outward (µ > 0)
∞ / 0
ϕ(v) H+ (tL ,tL ) ϕ(v) H− (tL ,tL )
= 1 − − dtL
I(ν, Ω) ϕ(v) S(tL ) e µ
+ e µ
2 µ
tL
∞ / 0
ϕ(v) H+ (tL ,tL ) ϕ(v) H− (tL ,tL )
= 1 − − dtL
Q(ν, Ω) ϕ(v) S(tL ) e µ
− e µ
, (14.75)
2 µ
tL
where v and ϕ(v) are the reduced frequency and the Gaussian profile of Eq. (14.31),
and where
tL
1
H+ (tL , tL ) = 1 + √ 3µ2 − 2 σ(tL ) dtL
2
tL
tL
1
H− (tL , tL ) = 1 + √ σ(tL ) dtL .
2
tL
Similarly, for the radiation propagating inward (µ < 0) one has
tL / ϕ(v) H+ (tL ,tL ) ϕ(v) H− (tL ,tL )
0
= 1 − − dtL
I(ν, Ω) ϕ(v) S(tL ) e |µ| + e |µ|
2 |µ|
0
tL / ϕ(v) H+ (tL ,tL ) ϕ(v) H− (tL ,tL )
0
= 1 − − dtL
Q(ν, Ω) ϕ(v) S(tL ) e |µ| − e |µ| . (14.76)
2 |µ|
0
Equations (14.75) and (14.76) can be used to obtain the two components of the
radiation field tensor, J¯00 and J¯02 , through the usual expressions (see Eqs. (5.157)
and (14.5))
∞ 1
1
J¯00 (ν0 ) = dv ϕ(v)
dµ I(ν, Ω)
2
−∞ −1
∞ 1
1 2
J¯02 (ν0 ) = √ dv ϕ(v) dµ + 3 µ2 − 1 Q(ν, Ω)
3µ − 1 I(ν, Ω) .
4 2
−∞ −1
Substitution of these expressions into Eqs. (14.73) closes the non-LTE loop of the
2nd kind, which – as anticipated – shows a complicated non-linear character.
STELLAR ATMOSPHERES 773
Fig.14.10. The logarithm of the source function S of Eq.(14.71) normalized to BP , and the reduced
statistical tensor σ of Eq.(14.72) – upper and lower panel, respectively – are plotted against the
logarithm of optical depth for two different values of . The plots refer to a homogeneous,
isothermal, non-magnetic, non-depolarizing atmosphere and to the transition (J = 1, Ju = 0).
As pointed out in Sect. 14.1, a solution of the non-LTE loop can be found only
via sophisticated numerical methods. Some results, obtained through accelerated
iterative schemes, are presented in Figs. 14.10 and 14.11. We point out that the
surface value of lower-level atomic polarization, σ(0), obtained for a homogeneous,
isothermal, non-depolarizing atmosphere with = 10−4 , turns out to be equal to
−0.09171. This ratio is related to the fractional linear polarization observed in the
tangential direction by the expression, easily derived from Eq. (14.74)1
)
Q(ν, Ω 3 σ(0)
p ≡ =− √ , (14.77)
I(ν, Ω ) 2 2 − σ(0)
1 )/ds = 0 because of the
Equation (14.77) can be obtained by observing that dQ(ν, Ω
translational invariance of the atmosphere under lateral displacements. Moreover, the reference
direction for positive Q in Eq. (14.77) is parallel to the stellar surface (while in Eq. (14.74) it is
perpendicular to the stellar surface).
774 CHAPTER 14
Fig.14.11. The emerging profiles of the intensity (normalized to BP ) and of the ratio Q/I corre-
sponding to the functions S(tL ) and σ(tL ) of Fig.14.10 are plotted against the reduced frequency
(ν0 − ν)/∆νD (solid line). The plots refer to a heliocentric angle θ = 84◦ .3 (µ = 0.1). The
positive Q direction is parallel to the stellar surface. The dotted curves in the right panel show,
for comparison, the Q/I profiles corresponding to the ‘reversed’ transition (J = 0, Ju = 1) – cf.
Fig.14.8.
which, for the σ(0) value quoted above, yields p = 0.09422. Note that both
|σ(0)| and p turn out to be larger than the corresponding values obtained for the
‘inverted’ transition (J = 0, Ju = 1) – see the discussion following Eq. (14.36).
This result shows that, in the absence of depolarizing collisions, dichroism is a very
efficient mechanism for introducing linear polarization in spectral lines.
It should be remarked that in recent times the non-LTE problem of the 2nd kind
has been solved for a variety of two-level model atoms with different values of J
and Ju , and also for more complicated atomic models involving up to 5 different
levels (Trujillo Bueno, 1999; Manso Sainz and Trujillo Bueno, 2001; Manso Sainz,
2002).
14.6. Applications to Realistic Atmospheres: Polarization
in the Continuum Spectrum
In the former sections we have shown how the self-consistency loop underlying the
non-LTE problem of the 2nd kind can be solved in some rather schematic physical
situations involving simple atomic models and highly idealized, unidimensional (or
plane-parallel) atmospheres. However, when trying to interpret real observations
from stellar atmospheres, and particularly solar observations, one has to be aware
of the presence of additional facts which tend to complicate, to some extent, the
physical scenario.
First of all, it should be kept in mind that the solar atmosphere is highly struc-
tured. Even considering the so-called ‘quiet’ atmosphere, the presence of con-
vective motions which manifest themselves through the well-known phenomena of
granulation and supergranulation raises serious doubts about the reliability of the
STELLAR ATMOSPHERES 775
plane-parallel approximation, especially when one tries to interpret polarimetric
observations at high heliocentric angles. Under this respect, it is important to
remind that even the interpretation of the usual solar spectrum at high heliocen-
tric angles has raised several problems, which still remain partially unsolved (see
Pecker, 1999, for a historical perspective of this subject).
Another problem is the polarization of the continuum spectrum and its interac-
tion with the line spectrum polarization. This is an important subject that we are
going to analyze here in some detail.
Historically, the continuum solar spectrum was discovered to be linearly polarized
by Lyot (1948). According to his observations, the polarization is perpendicular to
the solar radius and increases with decreasing limb distance. More recently, Kemp
et al. (1983) have shown that the same phenomenon is present in the eclipsing
variable Algol.1
Polarization is introduced in the continuum stellar spectrum by two different
kinds of mechanism: dichroism and scattering. The role of dichroism as a source of
continuum polarization has generally been neglected in the past. It is nevertheless
obvious that a photoionization process taking place from a polarized atomic level
has a cross section which depends on the polarization of the incident radiation.
This implies that the continuum absorption coefficient depends on polarization, or,
in other words, that the stellar atmosphere behaves as a dichroic medium.
Referring more specifically to the solar case, the major contribution to the con-
tinuum absorption coefficient at visible and near-infrared wavelengths comes from
the photoionization of the H− ion. Since this ion has a single bound state 1s2 1 S0 ,
and since this state cannot harbor atomic polarization – being a J = 0 state –
it follows that dichroism phenomena are of marginal importance in this spectral
range. However, at short wavelengths in the visible and especially in the near
UV, the major contributors to the continuum absorption coefficient are different
atoms, such as neutral hydrogen, silicon, carbon, magnesium, aluminum, and iron.
Since the levels of these atoms from which photoionization processes can take place
may be polarized, it follows that dichroism has to be taken into account at these
wavelengths.
The formalism to describe this phenomenon does not need to be derived through
ab initio calculations. We can just adapt the results derived in Sect. 7.2, in par-
ticular Eq. (7.16a), to the case of a transition where the upper level is a level of
the continuum. Denoting by α and J the quantum numbers of the atomic level
from which photoionization takes place, by α+ and J+ the quantum numbers of
the level in which the ion is left after photoionization, with , and j the energy
and quantum numbers of the electron released by photoionization, and finally with
J the total angular momentum of the final state (J = J+ + j ), one obtains from
1
For obvious symmetry reasons, it is impossible to detect this phenomenon in an isolated
star, because of the lack of spatial resolution. On the contrary, the phenomenon can show up
with a typical time-dependent signal when the star is eclipsed by a dark companion (see Landi
Degl’Innocenti et al., 1988, for an analysis of the time variability of the polarization signal during
the eclipse).
776 CHAPTER 14
Eq. (7.16a)1
hν
ηiA (ν, Ω) ph
= N h n() (2J + 1) B(αJ → α+ J+ , j, J )
4π
jJ
√ !
1 1 K ρK (αJ) . (14.78)
× 3 (−1)1+J+J +K TQ K
(i, Ω)
J J J Q
KQ
In this equation, N is the number density of atoms, ρK Q (αJ) is the statistical
tensor of the photoionized level, ν is the frequency of the absorbed photon, which
is connected to the energy of the free electron by the relation
1
ν= Eα J + − EαJ ,
h + +
n() d is the number of quantum states of the free electron with energy contained
in the interval (, + d), and, finally
32π 4
B(αJ → α+ J+ , j, J ) = | αJ d α+ J+ , j, J |2 . (14.79)
3h2 c
Substitution of Eq. (14.79) into Eq. (14.78) yields
8π 3 ν
ηiA (ν, Ω) ph
= N n() (2J + 1) | αJ d α+ J+ , j, J |2
3c
jJ
√ !
1 1 K ρK (αJ) . (14.80)
× 3 (−1)1+J+J +K TQ
K
(i, Ω)
J J J Q
KQ
It is interesting to connect this expression with the ‘classical’ expression (see e.g.
Mihalas, 1978) for the absorption coefficient due to photoionization. Such coeffi-
cient, denoted by kph (ν), is generally expressed in the form
kph (ν) = NαJ σph (ν) αJ , (14.81)
where NαJ is the number density of atoms in the photoionized level and [σph (ν)]αJ
is the photoionization cross section for the same level. From Eq. (14.80), assuming
the level (αJ) to be unpolarized ( ρK 0
Q (αJ) = δK0 δQ0 ρ0 (αJ) ), and taking into
account that
√
NαJ = 2J + 1 ρ00 (αJ) N , =1,
T 00 (0, Ω)
1
Equation (14.78) refers to a single photoionization process from a given level (αJ) to
a given level (α+ J+ ). Moreover, the sum over the final states also implies an integral over the
energy of the free electron. This integral is easily performed taking into account that the profile
φ contained in Eq. (7.16a) is practically a Dirac delta-function.
STELLAR ATMOSPHERES 777
and that !
1 1 0 1
= (−1)1+J+J ,
J J J 3(2J + 1)
one gets, by comparison with Eq. (14.81)
8π 3 ν
σph (ν) αJ = n() | αJ d α+ J+ , j, J |2 .
3c jJ
Recalling Eq. (10.11), this allows one to cast Eq. (14.80) in the more significant
form
A
ηi (ν, Ω) = NαJ σph (ν) αJ (−1)K WαJ (K)
() T Q
K σ K (αJ) , (14.82)
(i, Ω)
ph Q
KQ
Q (αJ)/ρ0 (αJ) and where the symbol WαJ (), which represents
K (K)
where σQ (αJ) = ρK 0
(K)
an extension of the symbol wJJ of Eq. (10.11) to photoionization, is defined by
(K)
wJJ | αJ d α+ J+ , j, J |2
jJ
WαJ () =
(K)
. (14.83)
| αJ d α+ J+ , j, J |2
jJ
The explicit evaluation of the reduced matrix elements appearing in Eq. (14.83)
is, in general, quite involved. In App. 24 we consider a particularly simple case
where the atomic states can be described in the L-S coupling scheme and where
the photoionized electron belongs, before photoionization, to an open shell.
We now turn to discuss scattering phenomena in the continuum, namely Thomson
scattering on free electrons and Rayleigh scattering on atoms or molecules. These
phenomena have been analyzed in great detail by Chandrasekhar (1950). Here we
just recall his conclusions and adapt his results to the formalism used in this book.
In the non-relativistic limit, i.e., at frequencies ν such that hν mc2 , m being
the electron mass, Thomson scattering is characterized by a frequency-independent,
non-dichroic cross section given by
8π e40
σT =
3 m 2 c4
(where e0 is the electron charge and c the speed of light), and by a scattering phase
matrix which, in our notations, has been denoted by the symbol Rij (Ω, Ω ) – or
Ω
Rij (Ω, ; 0) – and which is given by (cf. Eqs. (5.137), (5.138), and (10.21))
Ω
Rij (Ω, ) = (−1)Q T Q
K
(i, Ω) ) .
T K (j, Ω (14.84)
−Q
KQ
Moreover, it has to be taken into account that scattering is coherent in the reference
frame of the electron.
778 CHAPTER 14
From the point of view of radiative transfer, Thomson scattering is therefore
characterized by an absorption coefficient of the form
ηi (ν, Ω) = Ne σT δi0 , (14.85)
T.sc
Ne being the number density of electrons, and by an emission coefficient εi (ν, Ω)
which, taking into account the Doppler effect due to the motion of the electrons,
is given by1
5
dΩ
3
εi (ν, Ω) = Ne σT Ω
Rij (Ω, ) dν d3v f (v )
T.sc 4π j=0
v · Ω v · Ω
×δ ν −ν +ν − ν Ij (ν , Ω
) , (14.86)
c c
where f (v ) is the normalized velocity distribution of the electrons. In particular,
for the case of a Maxwellian distribution of velocities, the integral in d3v can be
performed to give (see the analogous calculation carried out in App. 18)
5
dΩ
3
εi (ν, Ω) = Ne σT Ω
Rij (Ω, ) dν
T.sc 4π j=0
(ν−ν )2
1 − 2
× e 2(1−cos Θ) ∆νe Ij (ν , Ω
) , (14.87)
2π(1 − cos Θ) ∆νe
where Θ is the angle between the directions Ω and Ω , and where ∆ν is the
e
Doppler broadening (in frequency units) corresponding to the thermal velocity of
the electrons, .
ν 2kB Te
∆νe = , (14.88)
c m
with kB the Boltzmann constant and Te the electronic temperature.
Equation (14.86) can be cast in a different form recalling Eq. (14.84) and the
definition of the radiation field tensor in the comoving frame. With easy transfor-
mations one gets
εi (ν, Ω) = Ne σT (−1)Q T Q
K
(i, Ω)
T.sc
KQ
, -
v · Ω
× d3v f (v ) J−Q
K
ν−ν , (14.89)
c
v
where the radiation field tensor in the comoving frame is given by Eq. (13.31).
1
In this formula, as well as in Eqs. (14.87) and (14.89), we suppose v/c 1.
STELLAR ATMOSPHERES 779
The preceding results on Thomson scattering can be simply modified to han-
dle Rayleigh scattering. Obviously, Rayleigh scattering may occur on any atom
(or molecule), but in the solar atmosphere it is generally sufficient to take into
account the contributions of neutral hydrogen and neutral helium. Neglecting
atomic polarization,1 Rayleigh scattering on hydrogen atoms is still described by
Eqs. (14.85)-(14.89) provided the following formal transformations are performed
Ne → NH , σT → σR
(H)
, ∆νe → ∆νH ,
(H)
where NH is the number density of hydrogen atoms, σR is the frequency-dependent
Rayleigh scattering cross section, and ∆νH is the Doppler broadening (in frequency
units) corresponding to the thermal velocity of the hydrogen atoms. The explicit
(H)
expression for σR is quite involved (Kramers, 1924), but a good approximation at
optical wavelengths is provided by the formula
(H) ν4
σR = σT , (14.90)
(ν 2 − ν02 )2
where ν0 is the frequency of Lyman α (ν0 = 2.466 × 1015 s−1 ).
Similar expressions hold for Rayleigh scattering on helium atoms. The main
difference from the hydrogen case, apart from the Doppler broadening which is now
smaller by a factor 2 (∆νHe 0.5 ∆νH ), comes from the cross section – denoted
(He)
in the following by σR – which can still be approximated as in Eq. (14.90) with
ν0 the frequency of the first resonance line of HeI, 1s2 1 S0 → 1s 2p 1 P1 (ν0 =
5.130 × 1015 s−1 ).
An important remark about Eq. (14.86) is that the integral over the distribution
of velocities can be trivially performed whenever the Stokes parameters Ij (ν , Ω
)
are constant over a frequency interval centered at the frequency ν and having a
width of the order of ∆νe for the case of Thomson scattering or ∆νH (∆νHe ) for the
case of Rayleigh scattering on hydrogen (helium) atoms. In such case, irrespective
of the velocity distribution f (v ), one gets, for Thomson scattering
5
dΩ
3
εi (ν, Ω) = Ne σT ) I (ν, Ω
Ω
R (Ω, )
T.sc 4π j=0 ij j
= Ne σT (−1)Q T Q
K J K (ν) ,
(i, Ω) (14.91)
−Q
KQ
with similar expressions holding for Rayleigh scattering on hydrogen or helium
atoms.
1
This approximation must not be taken for granted and might prove to be unreliable
to properly handle continuum polarization phenomena in the solar atmosphere. A consistent
theoretical treatment of Rayleigh scattering on polarized atoms is still missing and is beyond the
purposes of this book.
780 CHAPTER 14
The expressions now derived allow one to deduce the polarization properties of the
continuum spectrum of the solar atmosphere. In the plane-parallel approximation,
owing to the cylindrical symmetry of the environment, the transfer of polarized
radiation can be described by a simplified equation which, by a suitable choice
of the reference direction,1 only involves the two Stokes parameters I(ν, Ω) and
Q(ν, Ω). This follows from the fact that the only non-zero components of the
K K
tensor JQ (ν) – appearing in Eqs. (14.89) and (14.91) – and σQ (αJ) – appearing
in Eq. (14.82) – are those with (K = 0, Q = 0) and (K = 2, Q = 0).2 Taking
into account the expressions of the tensor T Q K – Table 5.6 – it can easily be
(i, Ω)
seen that the only non-vanishing components of [εi (ν, Ω)] A
T.sc and of [ηi (ν, Ω)]ph
are those with i = 0, 1, so that the transfer equation has the form
d
I(ν, Ω)
η0 (ν, Ω)
η1 (ν, Ω)
I(ν, Ω)
ε0 (ν, Ω)
=− + , (14.92)
ds
Q(ν, Ω)
η1 (ν, Ω)
η0 (ν, Ω)
Q(ν, Ω)
ε (ν, Ω)
1
where the transfer coefficients η0 , η1 , ε0 , and ε1 result from the sum of the contribu-
tions of all the processes responsible for absorption and emission in the continuum.
Since the degree of anisotropy in the solar atmosphere is everywhere weak, and
since the major source of continuum opacity at visible wavelengths (the H − ion)
contributes neither to dichroism nor to emission of polarized radiation, it follows
that
η (ν, Ω)
η (ν, Ω) , ε (ν, Ω)
ε (ν, Ω) ,
1 0 1 0
so that Eq. (14.92) can be solved by a perturbative approach. One thus gets the
zero-order equation
d = −η (ν, Ω)
I(ν, Ω)
+ ε (ν, Ω)
,
I(ν, Ω) 0 0 (14.93)
ds
and the first-order equation
d
= −η (ν, Ω)
Q(ν, Ω) Q(ν, Ω)
+ ε (ν, Ω)
− η (ν, Ω)
I(ν, Ω)
, (14.94)
0 1 1
ds
where, from Eqs. (14.82), (14.85), and (14.91)
1
This choice is illustrated in Fig. 14.4, where z is the vertical to the atmosphere and where
the angle γ has to be set to π/2 (or 0).
2
The tensor components with K = 1 can be excluded through general considerations involv-
ing the absence of circular polarization inside the atmosphere (see, for instance, the discussion
following Eq. (14.23)). On the contrary, the presence of a magnetic field inclined with respect
to the vertical may introduce non-vanishing σQ 2 (αJ) components with Q = 0. When dichroism
phenomena are important, Eq. (14.92) is valid only in the absence of such fields.
STELLAR ATMOSPHERES 781
=
η0 (ν, Ω) NαJ σph (ν) αJ + Ne σT + NH σR
(H) (He)
+ NHe σR
αJ
=
η1 (ν, Ω) NαJ σph (ν) αJ WαJ
(2) σ 2 (αJ)
() T 02 (1, Ω) 0
αJ
=B
ε0 (ν, Ω) NαJ σph (ν) αJ + Ne σT + NH σR
(H) (He)
+ NHe σR J00 (ν)
P
αJ
= N σ + N σ (H) + N σ (He) T 2 (1, Ω)
ε1 (ν, Ω) J 2 (ν) . (14.95)
e T H R He R 0 0
where B is the Planck function at the
The first term in the expression of ε0 (ν, Ω), P
electronic temperature Te , describes radiative recombination (the process inverse
to photoionization), which takes place under LTE conditions.
Equations (14.93)-(14.94) can be formally integrated by introducing the optical
depth in the continuum, tν , defined by
dz ,
dtν = −η0 (ν, Ω)
z being the vertical to the atmosphere (directed upwards), and the continuum
source function
ε (ν, Ω)
Sc (ν) = 0 .
η (ν, Ω)0
We easily obtain
Sc (ν) = (1 − β) BP + β J00 (ν) , (14.96)
where β is the fraction of continuum opacity due to scattering processes,
(H) (He)
Ne σT + NH σR + NHe σR
β= . (14.97)
NαJ σph (ν) αJ + Ne σT + NH σR (H) (He)
+ NHe σR
αJ
Integration of Eq. (14.93) gives, for the intensity of the radiation emerging from
the solar atmosphere
∞ t
− ν dtν
I(ν, Ω) = Sc (ν) e µ , (14.98)
µ
0
where µ = cos θ is the cosine of the heliocentric angle. Similarly, integration of
Eq. (14.94) yields the Q Stokes parameter of the emerging radiation. Two contribu-
tions arise, one due to scattering processes and the other to dichroism phenomena.
With self-evident notations
782 CHAPTER 14
= Q(ν, Ω)
Q(ν, Ω)
+ Q(ν, Ω) ,
sc ph
1
where, recalling Eqs. (14.95) and (14.98), and the expression of the tensor T 02 (1, Ω)
∞
3 t
− µν dtν
Q(ν, Ω) sc
= √ 1 − µ2
β J02 (ν) e , (14.99)
2 2 µ
0
and
∞ NαJ σph (ν) αJ WαJ
(2)
() σ02 (αJ)
3
Q(ν, Ω) ph
= − √ 1 − µ2 1−β αJ
2 2 NαJ σph (ν) αJ
0 αJ
∞ t
− µν dtν dtν
× Sc (ν) e . (14.100)
µ µ
tν
The emerging intensity and the scattering contribution to the Q Stokes parameter
can be computed by standard numerical techniques from a given model of the solar
atmosphere. Since in the solar atmosphere the contribution of scattering processes
to the continuum opacity is small (β 1), one can simply set Sc (ν) = BP (ν) – see
Eq. (14.96). The emerging intensity can then be evaluated using Eq. (14.98), and
the tensor J02 (ν) at any optical depth can be obtained by a numerical quadrature.
The scattering contribution to Q can finally be calculated from Eq. (14.99). De-
tailed theoretical results on the fractional polarization, obtained through numerical
procedures similar to the one just described, can be found in Débarbat et al. (1970)
and in Fluri and Stenflo (1999).2
On the contrary, the contribution to the continuum linear polarization arising
from dichroism is more difficult to compute because of the presence in Eq. (14.100)
of the atomic polarization factors, σ02 (αJ), relative to all the atomic levels of differ-
ent atomic species from which significant photoionization can take place. Obviously,
the evaluation of the factors σ02 (αJ) requires the solution of the non-LTE problem
of the 2nd kind for multi-level atoms, a difficult subject which is presently in a
phase of development and upon which we will not dwell any longer in this book.
1
The reference direction for the Stokes parameters is defined here as in Fig. 14.4 with γ = π/2.
Positive Q means linear polarization parallel to the solar limb.
2
Note that very little progress has been made in this field from the observational side.
To the authors’ knowledge, the most complete measurements of continuum linear polarization at
the solar limb are still those by Leroy (1972). Such observations, however, were performed with
broad-band filters and are therefore contaminated by the spurious signals due to spectral lines.
Observations of the true continuum polarization in selected spectral windows have been presented
by Mickey and Orrall (1974) and by Wiehr (1975, 1978).
STELLAR ATMOSPHERES 783
14.7. Applications to Realistic Atmospheres: Approximate Results
on the Polarization of the Fraunhofer Spectrum
Although a fully consistent description of the polarization properties of a spectral
line formed in a stellar atmosphere has eventually to rely on the solution of the self-
consistency loop schematized in Fig. 14.1, it is nevertheless possible to obtain some
results by means of suitable approximations performed directly on the radiative
transfer equation for polarized radiation. The approximate results that will be
deduced in this section can be considered, to some extent, as the generalization to
polarized transfer of the well-known Eddington-Barbier approximation extensively
used in the usual transfer theory.
We refer to a plane-parallel, non-magnetic atmosphere and we consider an iso-
lated (non-blended) spectral line originating from the transition between two levels
of a given atomic species. The two levels, which we suppose devoid of hyperfine
structure, are described by the sets of quantum numbers (α J ) for the lower level
and (αu Ju ) for the upper level. We assume the atoms to have a Maxwellian dis-
tribution of velocities characterized, at the frequency ν0 of the transition, by the
Doppler broadening ∆νD , and we adopt the approximation of complete redistribu-
tion on velocities, which implies that the atomic polarization of the two levels is
described by the (velocity-independent) statistical tensors ρK K
Q (α J ) and ρQ (αu Ju ),
defined in a reference system with the z-axis directed along the vertical. We assume
the atmosphere to be characterized by a non-dichroic continuum absorption coef-
ficient and by an emission coefficient partly due to scattering phenomena. Finally,
we neglect stimulated emission.
An analysis of the statistical equilibrium equations similar to that carried out in
Sect. 14.2 shows that under the preceding hypotheses only the statistical tensors
of even rank (K = 0, 2, 4, . . .) and Q = 0 are different from zero. This remark,
joined with the other hypotheses outlined above, allows one to write the radiative
transfer equation for polarized radiation in the same form as Eq. (14.92),1 where
the radiative transfer coefficients, resulting from the contributions of continuum
and line processes, are given by
1
= k (c) + k A 1 + √
η0 (ν, Ω) ν L 3 cos2 θ − 1 wJ(2)J σ02 (α J ) φ̂(ν0 − ν)
2 2 u
= kA √3
η1 (ν, Ω) L sin2 θ wJ(2)J σ02 (α J ) φ̂(ν0 − ν)
2 2 u
3 , -
= k (c) (1 − β) B + v · Ω
ε0 (ν, Ω) ν P β i d3vi f (vi ) J00 ν − ν i +
i=1
c
vi
1
The reference direction for the definition of the Stokes parameters is the same as for
of Fig. 14.4 with γ = π/2.
Eq. (14.92), namely the direction ea (Ω)
784 CHAPTER 14
, - !
1 3
v · Ω
+ √ 3 cos2 θ − 1 βi d3vi f (vi ) J02 ν − ν i
2 2 i=1
c
vi
1
+ kLA SL 1 + √ 3 cos2 θ − 1 wJ(2)J σ02 (αu Ju ) φ̂(ν0 − ν)
2 2 u
3 , -
3
= √ v · Ω
ε1 (ν, Ω) sin2 θ kν(c) βi d3vi f (vi ) J02 ν − ν i
2 2 i=1
c
vi
!
+ kLA SL wJ(2)J σ02 (αu Ju ) φ̂(ν0 − ν) .
u
(c)
In these expressions, kν is the continuum absorption coefficient due to photoion-
ization and scattering processes, kLA and SL are the line opacity and the line source
function defined, respectively, in Eqs. (14.15) and (14.12), σ02 (α J ) and σ02 (αu Ju )
are the reduced statistical tensors defined by
ρ20 (α J ) ρ20 (αu Ju )
σ02 (α J ) = , σ02 (αu Ju ) = ,
ρ00 (α J ) ρ00 (αu Ju )
(2)
wJJ is the symbol defined in Eq. (10.98), θ is the heliocentric angle defined as in
Fig. 14.4 (z being the vertical to the solar atmosphere), the profile φ̂(ν0 − ν) is
defined according to Eq. (14.1), BP (ν) is the Planck function, and βi (i = 1, 2, 3)
is the fractional contribution to the opacity coming from scattering by electrons,
H atoms, and He atoms, respectively
(H) (He)
Ne σT NH σR NHe σR
β1 = (c)
, β2 = (c)
, β3 = (c)
;
kν kν kν
finally (cf. Eq. (14.97))
3
β= βi .
i=1
In the solar atmosphere, the reduced statistical tensors of rank 2 are generally
very small ( σ02 (α J ), σ02 (αu Ju ) 1 ), as well as the scattering contribution to the
continuum opacity (β 1).1 This implies that the fractional linear polarization
Q/I is also small, so that the radiative transfer equation can be solved by a per-
turbative approach, similarly to the previous section. For the intensity one obtains
the zero-order equation
d = − k (c) + k A φ̂(ν − ν) I(ν, Ω)
+ k (c) B + k A S φ̂(ν − ν) , (14.101)
I(ν, Ω) ν L 0 ν P L L 0
ds
1
In the top layers of the atmosphere, and especially in the UV region of the spectrum, this
inequality may well be violated, β attaining values close to unity. This happens, however, in a
layer of very small optical thickness, that can be safely neglected for our considerations.
STELLAR ATMOSPHERES 785
whereas for the Q Stokes parameter one gets the first-order equation
d = − k (c) + k A φ̂(ν − ν) Q(ν, Ω)
Q(ν, Ω) ν L 0
ds
3
+ √ kLA sin2 θ wJ(2)J σ02 (αu Ju ) SL − wJ(2)J σ02 (α J ) I(ν, Ω)
φ̂(ν − ν)
0
2 2 u u
3 , -
3 v · Ω
+ √ kν(c) sin2 θ βi d3vi f (vi ) J02 ν − ν i . (14.102)
2 2 i=1
c
vi
, parallel to the atmo-
These equations can be formally solved for a direction, Ω
sphere. Taking into account the translational invariance of the atmosphere under
lateral displacements, one has
d ) = d Q(ν, Ω
)=0.
I(ν, Ω
ds ds
From Eq. (14.101) we thus obtain
(c)
)= kν BP + kLA SL φ̂(ν0 − ν)
I(ν, Ω (c)
, (14.103)
kν + kLA φ̂(ν0 − ν)
and from Eq. (14.102), using Eq. (14.103), we get
)
Q(ν, Ω )
Q(ν, Ω )
Q(ν, Ω
= + ,
)
I(ν, Ω )
I(ν, Ω line
)
I(ν, Ω sc
where the ‘line contribution’ is given by
/
)
Q(ν, Ω 3 wJ(2)J σ02 (αu Ju ) SL
= √ kLA u
)
I(ν, Ω line 2 2 (c)
kν BP + kLA SL φ̂(ν0 − ν)
0
wJ(2)J σ02 (α J )
− (c) u
φ̂(ν0 − ν) , (14.104)
kν + kLA φ̂(ν0 − ν)
and the ‘scattering contribution’ by
, -
3
vi · Ω
βi d3vi f (vi ) J02 ν − ν
)
Q(ν, Ω 3 c
vi
= √ kν(c) i=1
. (14.105)
)
I(ν, Ω sc 2 2 (c)
kν BP + kL SL φ̂(ν0 − ν)
A
All the quantities in the right-hand sides of Eqs. (14.103), (14.104) and (14.105)
have to be evaluated at the top layers of the atmosphere (formally at τ = 0).
786 CHAPTER 14
In the following we analyze separately the two contributions to the fractional
polarization. As far as the line contribution is concerned, we observe that the
polarization signal disappears in the line wings and that it is usually maximum (in
absolute value) at line center. At this frequency we obtain the following expressions,
valid in the limiting cases of strong and weak lines, respectively
(c)
a) strong line (kLA φ̂(0) kν )
)
Q(ν0 , Ω 3
= √ wJ(2)J σ02 (αu Ju ) − wJ(2)J σ02 (α J ) , (14.106)
)
I(ν0 , Ω line 2 2 u u
(c)
b) weak line (kLA φ̂(0) kν )
)
Q(ν0 , Ω 3 k A φ̂(0) S
= √ L (c) wJ(2)J σ02 (αu Ju ) L − wJ(2)J σ02 (α J ) . (14.107)
)
I(ν0 , Ω line 2 2 kν u BP u
Equation (14.106) has been proposed by Trujillo Bueno (1999) to give an estimate
of the polarization signal observed in spectral lines close to the solar limb, and to
point out the importance of lower-level atomic polarization in the solar atmosphere.
Considering a multiplet of lines sharing the same upper level, and neglecting lower-
level polarization, we see from Eq. (14.106) that the observed fractional polarization
in the different lines has to be proportional to the value of the corresponding symbol
wJ(2)J . Following Trujillo Bueno (1999), consider for instance the MgI triplet in the
u
green region of the solar spectrum. The three lines at 5167.3 Å, 5172.7 Å, and
5183.6 Å correspond to the transition between a common upper level 3s 4s 3 S1 and
a lower level 3s 3p 3 PoJ with J = 0, 1, and 2, respectively. In the absence of lower-
level polarization, the fractional linear polarization should therefore be proportional
(2)
to w10 (2)
, w11 (2)
, w12 respectively, or, according to Table 10.1, to 1.0, −0.5, 0.1. Since
the observed values of the limb fractional polarization in the three lines are very
far from satisfying these relations, it follows that lower-level polarization must play
a significant role.
It is also important to remark that, according both to Eq. (14.106) and to
Eq. (14.107), the line contribution to the fractional polarization may be either
positive or negative, because both expressions contain the difference between two
terms that may themselves be positive or negative.
Turning to the scattering contribution given by Eq. (14.105), it is important
to distinguish between Thomson scattering and Rayleigh scattering. In the first
case, we should keep in mind that the thermal velocities of electrons are at least
two orders of magnitude larger than the thermal velocities of most of the atoms
contributing to the Fraunhofer spectrum. The ratio between the thermal velocity of
an electron and that of a sodium atom, for instance, is 205, while for an iron atom
such ratio is 319. It follows that the integral over the electron velocities contained
in the numerator of the right-hand side of Eq. (14.105) completely washes out any
spectral detail of the tensor J02 (ν), yielding an expression of the form
STELLAR ATMOSPHERES 787
)
Q(ν, Ω 3
(c)
kν β1 J02 ν
= √ , (14.108)
)
I(ν, Ω T.sc 2 2 kν(c) BP + kLA SL φ̂(ν0 − ν)
where the quantity J02 ν , defined by
, -
2 v1 · Ω
J0 ν = d3v1 f (v1 ) J02 ν − ν , (14.109)
c
v1
is in practice the average of J02 (ν) over a broad frequency interval centered at the
frequency ν and having a width of the order of 2∆νe (cf. Eq. (14.87)), with ∆νe
given by Eq. (14.88).1 This phenomenon bears a strong resemblance with another
phenomenon that solar physicists have known for a long time: the photospheric
radiation scattered by electrons in the corona is indeed strongly polarized and has
a featureless spectrum – whence the name of K (or continuum) corona.
As Eq. (14.108) clearly shows, the Thomson scattering contribution to fractional
polarization is characterized by a frequency dependence typical of the absorption
lines of the usual intensity spectrum, with a positive wing value2 and a depression
at line center. Such depression is totally independent of the intrinsic polarizability
characteristics of the line, like atomic polarization or w(2) factors.
For the Rayleigh scattering contribution to fractional polarization similar argu-
ments can be repeated, and an equation strictly similar to Eq. (14.108) holds. The
only difference is that the factor β1 has to be replaced by β2 (for scattering on hy-
drogen atoms) or β3 (for scattering on helium atoms), whereas the average J02 ν is
still given by Eq. (14.109), the integral being performed on the hydrogen or helium
atom velocities, respectively. Since the hydrogen atoms have thermal velocities
that are several times (about 5 for sodium, 7 for iron, etc.) larger than the thermal
velocities of most of the atoms contributing to the Fraunhofer spectrum, it follows
that the frequency dependence of J02 (ν) will be substantially smoothed by the ve-
locities of the hydrogen atoms but not totally destroyed. A qualitative sketch of
the contribution to the fractional polarization arising from scattering processes is
shown in Fig. 14.12.
We want to remark that the results derived in this section are strongly dependent
on the approximation of plane-parallel atmosphere, thus they have to be used with
much care when interpreting polarimetric observations of the solar limb spectrum.
14.8. Alternative Methods for the Solution of the Non-LTE Problem
The non-LTE theory that we have introduced in this chapter, schematically sum-
marized in the self-consistency loop of Fig. 14.1 and usually referred to as ‘non-LTE
1
In terms of wavelength, and assuming an electronic temperature Te of 6000 K, the width of
the interval is about 10 Å at 5000 Å.
2
J02 (ν) is generally positive in the top layers of a stellar atmosphere (see Chap. 12).
788 CHAPTER 14
Fig.14.12. Schematic behavior, as a function of the reduced frequency (ν0 − ν)/∆νD , of the
fractional polarization introduced by Thomson and Rayleigh scattering (full and dotted line,
respectively) in the neighborhood of a spectral line of central frequency ν0 ; ∆νD is the Doppler
width of the profile φ̂(ν0 − ν) of Eq.(14.108). The plot refers to the radiation propagating along
the top layer of a plane-parallel atmosphere. The reference direction for positive Q is parallel
to the atmosphere, and the fractional polarization is normalized to the value of the continuum
spectrum.
of the 2nd kind’, rests, as a foundation stone, on the concept of atomic density ma-
trix. Such concept generalizes to polarized radiative transfer the usual concept of
level populations employed in the non-LTE theory of the 1st kind, and appears to
be well-suited to handle most transfer problems.
There are however some special cases where a different approach, not based on
the density matrix, can be followed. These are the cases where the basic interaction
process between radiation and atoms is a scattering process for which the emission
coefficient vector can be expressed by a linear relation of the form
5 ∞
3
dΩ
=k
εi (ν, Ω) s dν ; ν, Ω
Rij (ν, Ω ) I (ν , Ω
j
) , (14.110)
4π j=0
0
and the radiative transfer equation has the simple form
d = −k I (ν, Ω)
+ ε (ν, Ω)
.
I (ν, Ω) (14.111)
ds i s i i
In these equations, Ii (ν, Ω) is the Stokes vector of the radiation at frequency ν
propagating along the direction Ω, k is a scalar quantity proportional to the cross
s
section of the scattering process and to the number density of scatterers, and
; ν, Ω
Rij (ν, Ω ) is the so-called redistribution matrix.
STELLAR ATMOSPHERES 789
As shown several times in this book, a linear equation similar to Eq. (14.110)
can in fact be established under severe restrictive hypotheses which imply two-level
(or two-terms) atomic models, the absence of atomic polarization in the lower level
(or lower term), and negligible stimulated emission. Under these circumstances,
the density matrix can be factored out and the non-LTE problem of the 2nd kind
reduces to the self-consistent solution of Eqs. (14.110) and (14.111). In other words,
there is no need to solve the statistical equilibrium equations for the density-matrix
elements because such solution is implicitly contained in Eq. (14.110). Obviously,
in these cases the theory provides explicit expressions for the coefficient ks and for
the redistribution matrix.
In general, an approach based on Eqs. (14.110) and (14.111), with suitable expres-
; ν, Ω
sions for ks and Rij (ν, Ω ), avoiding the introduction of the density matrix,
can be more appropriate for the description of certain phenomena. An important
example is the non-LTE problem concerning the interpretation of the extended
wings of some particularly strong spectral lines formed in the solar atmosphere.
Even the sophisticated redistribution matrices deduced in Chap. 13, which take
velocity/density-matrix correlations into account (see Eqs. (13.35) and (A18.1)-
(A18.2) for the non-magnetic case, and Eq. (13.44) for the magnetic case) are
unable to explain such wings, and have to be replaced by redistribution matrices
which include partial redistribution effects (in the atomic rest frame). The density-
matrix theory developed in this book, being based on the Markov approximation,
cannot account for such effects.
Much work has been devoted in the past to find the correct expression for the
redistribution matrix capable of taking into account, at the same time, partial redis-
tribution effects, depolarizing collisions, and the presence of a magnetic field. Im-
portant contributions to this subject have been given by Omont et al. (1972), Sten-
flo (1994), Bommier (1997a,b), and Landi Degl’Innocenti et al. (1997). Further ef-
forts have been devoted to find numerical, self-consistent solutions to Eqs. (14.110)-
(14.111) in plane-parallel atmospheres, either non-magnetized or magnetized, using
various forms of the redistribution matrix and often introducing ad hoc approxima-
tions on the matrix itself.1 Remarkable results on this subject have been obtained
by Rees and Saliba (1982), Saliba (1986), Faurobert (1987, 1988), and Nagendra
(1988, 1994) for the non-magnetic case, and by Faurobert-Scholl (1991, 1992) and
Nagendra et al. (1998) for the case of the Hanle effect.
1
From a historical point of view, it should be recalled that the first self-consistent solution
of Eqs. (14.110)-(14.111) in a plane-parallel atmosphere has been given, through semi-analytical
methods, by Chandrasekhar and Breen (1947). The problem solved by Chandrasekhar is that of an
atmosphere composed of Thomson-scattering free electrons; in this case one has (see Eqs. (14.85)
; νΩ
and (14.91)) ks = Ne σT , Rij (ν, Ω ) = R (Ω, ) δ(ν − ν ).
Ω
ij
APPENDIX
A1. A Fortran Code for Computing 3-j, 6-j, and 9-j Symbols
The following code computes 3-j, 6-j, and 9-j symbols.1 3-j symbols are evaluated
from Eqs. (2.19) and (2.22), 6-j symbols from Eq. (2.35), and 9-j symbols from
Eq. (2.48). The variables appearing as dummy arguments of the three Fortran-
functions W3JS, W6JS, W9JS are integers representing the values of the corre-
sponding quantum numbers multiplied by 2. For instance
1 3
2 2 2
= W3JS (4, 1, 3, −2, −1, 3)
−1 − 21 3
2
3 1 !
1 2 2
3 = W6JS (2, 3, 1, 3, 2, 4)
2 1 2
⎧ 1 3 ⎫
⎨2 2 2 ⎬
1 1
1 = W9JS (4, 1, 3, 1, 1, 2, 5, 2, 5) .
⎩ 25 2
5
⎭
2 1 2
The various conditions that must be satisfied by the arguments of each symbol are
explicitly checked within the code. If any of these conditions is not satisfied, the
code returns the value 0 for the symbol. The instruction CALL FACTRL must
appear in the main program before any of these Fortran-functions is used.
FUNCTION W3JS(J1,J2,J3,M1,M2,M3)
INTEGER Z,ZMIN,ZMAX
COMMON/FACT/FACT(0:301)
W3JS=0.0
IF(M1+M2+M3.NE.0) GOTO 1000
IA=J1+J2
IF(J3.GT.IA) GOTO 1000
IB=J1-J2
IF(J3.LT.IABS(IB)) GOTO 1000
JSUM=J3+IA
IC=J1-M1
ID=J2-M2
IF(MOD(JSUM,2).NE.0) GOTO 1000
IF(MOD(IC,2).NE.0) GOTO 1000
IF(MOD(ID,2).NE.0) GOTO 1000
IF(IABS(M1).GT.J1) GOTO 1000
IF(IABS(M2).GT.J2) GOTO 1000
IF(IABS(M3).GT.J3) GOTO 1000
IE=J3-J2+M1
IF=J3-J1-M2
1
The code is due to Stephen Jackson, formerly at the National Center for Atmospheric
Research, Boulder, Colorado.
792 APPENDIX
ZMIN=MAX0(0,-IE,-IF)
IG=IA-J3
IH=J2+M2
ZMAX=MIN0(IG,IH,IC)
CC=0.0
DO 200 Z=ZMIN,ZMAX,2
DENOM=FACT(Z/2)*FACT((IG-Z)/2)*FACT((IC-Z)/2)
1 *FACT((IH-Z)/2)*FACT((IE+Z)/2)*FACT((IF+Z)/2)
IF(MOD(Z,4).NE.0) DENOM=-DENOM
CC=CC+1.0/DENOM
200 CONTINUE
CC1=FACT(IG/2)*FACT((J3+IB)/2)*FACT((J3-IB)/2)
1 /FACT((JSUM+2)/2)
CC2=FACT((J1+M1)/2)*FACT(IC/2)*FACT(IH/2)
1 *FACT(ID/2)*FACT((J3-M3)/2)*FACT((J3+M3)/2)
CC=CC*SQRT(CC1*CC2)
IF(MOD(IB-M3,4).NE.0) CC=-CC
W3JS=CC
1000 RETURN
END
FUNCTION W6JS(J1,J2,J3,L1,L2,L3)
INTEGER W,WMIN,WMAX
INTEGER SUM1,SUM2,SUM3,SUM4
COMMON/FACT/FACT(0:301)
W6JS=0.0
IA=J1+J2
IF(IA.LT.J3) GOTO 1000
IB=J1-J2
IF(IABS(IB).GT.J3) GOTO 1000
IC=J1+L2
IF(IC.LT.L3) GOTO 1000
ID=J1-L2
IF(IABS(ID).GT.L3) GOTO 1000
IE=L1+J2
IF(IE.LT.L3) GOTO 1000
IF=L1-J2
IF(IABS(IF).GT.L3) GOTO 1000
IG=L1+L2
IF(IG.LT.J3) GOTO 1000
IH=L1-L2
IF(IABS(IH).GT.J3) GOTO 1000
SUM1=IA+J3
SUM2=IC+L3
SUM3=IE+L3
SUM4=IG+J3
IF(MOD(SUM1,2).NE.0) GOTO 1000
IF(MOD(SUM2,2).NE.0) GOTO 1000
IF(MOD(SUM3,2).NE.0) GOTO 1000
IF(MOD(SUM4,2).NE.0) GOTO 1000
WMIN=MAX0(SUM1,SUM2,SUM3,SUM4)
II=IA+IG
IJ=J2+J3+L2+L3
IK=J3+J1+L3+L1
APPENDIX 793
WMAX=MIN0(II,IJ,IK)
OMEGA=0.0
DO 200 W=WMIN,WMAX,2
DENOM=FACT((W-SUM1)/2)*FACT((W-SUM2)/2)*FACT((W-SUM3)/2)
1 *FACT((W-SUM4)/2)*FACT((II-W)/2)*FACT((IJ-W)/2)
2 *FACT((IK-W)/2)
IF(MOD(W,4).NE.0) DENOM=-DENOM
OMEGA=OMEGA+FACT(W/2+1)/DENOM
200 CONTINUE
THETA1=FACT((IA-J3)/2)*FACT((J3+IB)/2)*FACT((J3-IB)/2)
1 /FACT(SUM1/2+1)
THETA2=FACT((IC-L3)/2)*FACT((L3+ID)/2)*FACT((L3-ID)/2)
1 /FACT(SUM2/2+1)
THETA3=FACT((IE-L3)/2)*FACT((L3+IF)/2)*FACT((L3-IF)/2)
1 /FACT(SUM3/2+1)
THETA4=FACT((IG-J3)/2)*FACT((J3+IH)/2)*FACT((J3-IH)/2)
1 /FACT(SUM4/2+1)
THETA=THETA1*THETA2*THETA3*THETA4
W6JS=OMEGA*SQRT(THETA)
1000 RETURN
END
FUNCTION W9JS(J1,J2,J3,J4,J5,J6,J7,J8,J9)
KMIN=ABS(J1-J9)
KMAX=J1+J9
I=ABS(J4-J8)
IF(I.GT.KMIN) KMIN=I
I=J4+J8
IF(I.LT.KMAX) KMAX=I
I=ABS(J2-J6)
IF(I.GT.KMIN) KMIN=I
I=J2+J6
IF(I.LT.KMAX) KMAX=I
X=0.
DO 1 K=KMIN,KMAX,2
S=1.
IF(MOD(K,2).NE.0) S=-1.
X1=W6JS(J1,J9,K,J8,J4,J7)
X2=W6JS(J2,J6,K,J4,J8,J5)
X3=W6JS(J1,J9,K,J6,J2,J3)
X=X+S*X1*X2*X3*FLOAT(K+1)
1 CONTINUE
W9JS=X
RETURN
END
SUBROUTINE FACTRL
COMMON/FACT/FACT(0:301)
DATA NFAC/31/
FACT(0)=1.0
DO 10 I=1,NFAC
10 FACT(I)=FACT(I-1)*FLOAT(I)
RETURN
END
794 APPENDIX
A2. Sample Evaluation of a Quantity Involving the Contraction
of 3-j Coefficients
Consider the quantity
+M +2k−K+Q
S= (−1)J
qq MM M
J J k J J k
×
−M M q −M M q
k k K J J K
× . (A2.1)
q q −Q −M M Q
This is simply the contraction of four 3-j symbols over five angular momentum
components, and closely resembles the right-hand side of Eq. (2.34) with e = e,
= . In Eq. (2.34), the only component which is not a summation index ()
appears in the first and fourth symbols, while the corresponding component (Q) in
Eq. (A2.1) appears in the third and fourth symbols. Thus we start, for instance,
from the third 3-j in Eq. (A2.1) and substitute k = c, k = d, K = e, q = γ, q = −δ,
Q = −. By so doing, the third 3-j symbol in Eq. (A2.1) takes the same form as
the first 3-j in Eq. (2.34).
Looking again at Eq. (2.34), we see that the second 3-j contains the column (d, δ)
which appears also (apart from a sign) in the first 3-j. According to our previous
substitution, d and δ correspond to k and −q , respectively, so that we must pick
the second 3-j in Eq. (A2.1) and reshape it, using the symmetry properties of the
3-j symbols, in one of the two equivalent forms
J J k k J J
→
−M M q −q −M M
k J J
→ (−1)k +J +J . (A2.2)
−q M −M
Of these two forms, it is the second one that brings the right-hand side of Eq. (A2.1)
into a form equivalent to the right-hand side of Eq. (2.34). Indeed, choosing the
second form, we substitute J = b, J = f , M = −β, M = −φ, and we can now
reshape the first 3-j in Eq. (A2.1) in the form
J J k k+J+J J k J
→ (−1)
−M M q −M −q M
to get, by the substitutions J = a, M = α and apart from a sign factor, the third
3-j in Eq. (2.34).1
1
Note that if the first form were chosen for the 3-j in Eq. (A2.2) (implying the substitutions
J = b, J = f , M = β, M = φ), it would be impossible to arrange the first 3-j in Eq. (A2.1)
in the form of either the third or the fourth 3-j in Eq. (2.34).
APPENDIX 795
Finally, reshaping the fourth 3-j in the form
J J K J J K
→ ,
−M M Q −M M −Q
we obtain, by the substitutions given above, the correct 3-j appearing at the fourth
place in Eq. (2.34).
Summarizing all the substitutions,
J =a M =α
J = f M = −φ
J = b M = −β
k=c q=γ
k = d q = −δ
K=e Q = − , (A2.3)
and recalling all the sign factors, we can write
S= (−1)b−β+2c−e− +d+f +b+c+a+b
αβγδφ
c d e d b f b c a f a e
× . (A2.4)
γ −δ δ −β φ β −γ α φ α
Now we must work on the sign factor to cast it in a form such that Eq. (2.34) can
be applied. In general, when performing this kind of calculations on sign factors
appearing in Racah-algebra expressions, a number of rules can be applied. If Σ is
an arbitrary linear combination of angular momentum and z-component quantum
numbers, these rules can be summarized as follows
I)
a b c a b c
(−1)a±α+Σ = (−1)−a∓α+Σ ,
α β γ α β γ
as (a ± α) is an integer;
II)
a b c a b c
(−1)α+β+γ+Σ = (−1)Σ ,
α β γ α β γ
as (α + β + γ) = 0 because of the presence of the 3-j symbol;
III)
a b c a b c
(−1)a±b±c+Σ = (−1)−a∓b∓c+Σ ,
α β γ α β γ
as the quantity (a ± b ± c) is an integer;
796 APPENDIX
IV)
(−1)3a+Σ = (−1)−a+Σ , (−1)4a+Σ = (−1)Σ ,
as 4a is an even integer.
With the help of these rules, the sign factor appearing in Eq. (A2.4) can be
transformed in the following way
(−1)a+3b+3c+d−e+f −β− =
= (−1)b+c+d+β+γ+δ (−1)a+3b+3c+d−e+f −β− −b−c−d−β−γ−δ
= S1 S2 .
S1 is the correct sign factor appearing in Eq. (2.34), while S2 can be transformed
by replacing, according to Rule I, (−b − β) by (b + β) and (−d − δ) by (d + δ).
Thus we obtain
S2 = (−1)a+4b+2c+2d−e+f −γ+δ− .
Now we can eliminate the terms 4b and (−γ + δ − ) according to Rules IV and II,
respectively, and we get
S2 = (−1)a+2c+2d−e+f .
Finally, according to Rule III, we can substitute (c + d − e) by (−c − d + e) to
obtain
S2 = (−1)a+e+f .
Since S2 does not depend on the summation indices in Eq. (A2.4), we can apply
Eq. (2.34) to get
!
1 a b c
S = (−1)a+e+f ,
2e + 1 d e f
and recalling the substitutions (A2.3) we obtain the final result
! !
J+J +K 1 J J k J+J +K 1 J J K
S = (−1) = (−1) .
2K + 1 k K J 2K + 1 k k J
Obviously, the calculation scheme presented here is nothing but an example, and
different procedures can be followed to obtain the same result.
A3. Momentum and Angular Momentum of the
Electromagnetic Field
In classical physics, the momentum of the electromagnetic field is given by (see e.g.
Jackson, 1962)
1
P = ×B
E dV .
4πc V
APPENDIX 797
= curl A,
Since B the j-th Cartesian component of P is
1 ∂Am ∂Aj
Pj = Em − Em dV . (A3.1)
4πc m V ∂xj ∂xm
= 0, the second term in the right-hand side can be written
Being in vacuum div E
in the form
∂Aj ∂
Em dV = (Em Aj ) dV .
m V ∂xm m V ∂xm
On the other hand, this integral can be transformed into a surface integral. As-
suming that the fields vanish at infinity, Eq. (A3.1) reduces to
1 ∂Am
Pj = E dV . (A3.2)
4πc m V m ∂xj
Similarly, writing the angular momentum of the electromagnetic field in the form
= 1
M × B)
r × (E dV ,
4πc V
one obtains for the j-th component of M
1 ∂Am ∂Al
Mj = jkl xk Em − Em dV .
4πc V ∂xl ∂xm
klm
For the second term we have
∂Al ∂
xk Em dV = (xk Em Al ) dV − Ek Al dV ,
m V ∂xm m V ∂xm V
and observing that the first integral in the right-hand side is zero we obtain
1 ∂Am 1
Mj = jkl xk Em dV + jkl Ek Al dV . (A3.3)
4πc V ∂xl 4πc V
klm kl
We now recall that in Quantum Mechanics the momentum operator p̂j and the
orbital angular momentum operator l̂j are given by
∂
p̂j = −ih̄
∂xj
∂
l̂j = −ih̄ jkl xk ,
∂xl
kl
798 APPENDIX
therefore Eqs. (A3.2) and (A3.3) can be written in the form
i
Pj = Em p̂j Am dV
4πh̄c V m
i 1
Mj = Em l̂j Am dV + E ×A
dV . (A3.4)
4πh̄c V m 4πc V j
These expressions suggest quite naturally that the first term in the right-hand side
of Eq. (A3.4) can be regarded as the orbital angular momentum, and the second
term as the intrinsic angular momentum (or spin) of the radiation.
Now we introduce the formalism of second quantization to derive the expressions
of the operators Pˆ and Mˆ according to the Correspondence Principle. For the
operator Pˆ
we start from Eq. (A3.2). Substituting for A and E
the expressions of
the corresponding operators given by Eqs. (4.30) and (4.33), and carrying out the
volume integration along the same lines leading to Eq. (4.24), one obtains
ˆ hν λ) a† (ν, Ω,
λ) + a† (ν, Ω,
P = Ω a(ν, Ω, λ) a(ν, Ω,
λ) .
2c
ν Ωλ
Owing to the commutation rule (4.32) this expression reduces to
ˆ hν †
P = λ) ,
Ω a (ν, Ω, λ) a(ν, Ω,
c
ν Ωλ
which, recalling the physical meaning of the operator a† (ν, Ω,
λ) a(ν, Ω,
λ), shows
that each photon of frequency ν and direction Ω carries a momentum hν Ω/c.
ˆ
For the operator M, we consider only the contribution of the intrinsic angular
momentum. Therefore, we calculate the operator
ˆ
1 ˆ × A
ˆ dV .
MS = E
4πc V
Evaluation of the volume integral via the same substitutions leads to
ih
Mˆ = λ) a† (ν, Ω,
a(ν, Ω, λ ) e (Ω) ∗
× e (Ω)
S λ λ
4π
ν Ωλλ
− a† (ν, Ω, λ ) e (Ω)
λ) a(ν, Ω,
λ
∗ × e (Ω)
λ
. (A3.5)
The expression in square bracket can be related to the Stokes parameter operator
defined in Sect. 4.4. Let us consider, for instance, the representation (4.38)
V̂ (ν, Ω)
The cross products are
for the unit vectors eλ (Ω).
e+1 (Ω) ∗ = iΩ
× e (Ω)
+1
e−1 (Ω) ∗ = −i Ω
× e (Ω)
−1
∗ = e (Ω)
× e (Ω)
e+1 (Ω) ∗=0,
× e (Ω)
−1 −1 +1
APPENDIX 799
thus Eq. (A3.5) becomes
h
ˆ = −
M
Ω +1) a† (ν, Ω,
a(ν, Ω, +1) + a† (ν, Ω,
+1) a(ν, Ω,
+1)
S
4π
νΩ
!
−1) a† (ν, Ω,
− a(ν, Ω, −1) + a† (ν, Ω,
−1) a(ν, Ω,
−1) ,
or, using Eq. (4.32)
h
ˆ = −
M a† (ν, Ω,
Ω +1) − a† (ν, Ω,
+1) a(ν, Ω, −1) a(ν, Ω,
−1) . (A3.6)
S
2π
νΩ
Recalling Eqs. (4.37) and (4.39) we finally obtain
c2
ˆ = −
M .
Ω V̂ (ν, Ω) (A3.7)
S
2πν 3
νΩ
Although this result has been obtained using the specific unit vectors of Eqs. (4.38),
it can be shown directly that it holds also for the more general unit vectors defined
in Eqs. (1.41).
Equations (A3.6) and (A3.7) have a simple physical interpretation. They show
that each circularly polarized photon carries an intrinsic angular momentum of
magnitude h̄, parallel or antiparallel to the propagation direction Ω. It is also
seen that a positive (or right-handed) circular polarization corresponds to a nega-
tive projection of the angular momentum on the propagation direction (negative
helicity), and vice versa.
The inconsistency between the sign of circular polarization and the sign of helicity
is obviously due to the convention adopted in this book for the sign of circular
polarization. The reasons of our choice are mainly of practical nature and are
explained in Sect. 1.2. In any case, it should be remarked that even the definition
of positive helicity is purely conventional.
A4. Multipole Components of Collisional Rates
In Sect. 7.13.a we have seen that the hypothesis of isotropic collisions implies that
the transfer collisional rates CI (αJM M , α J M M ) must satisfy Eq. (7.84) for
any arbitrary rotation R of the reference system. Instead of transforming the
right-hand side of Eq. (7.84) via Eqs. (7.85), we can couple the rotation matrices
in a different way, writing
J ∗
DN M (R) DNJ
M (R) =
N −M J J K J J K
= (−1) (2K + 1) DPKQ (R)∗ ,
N −N P M −M Q
K
800 APPENDIX
J ∗
DN
J
M (R) D
N M (R) =
N −M J J K J J K
= (−1) (2K + 1) DPK Q (R)∗ ,
N −N P M −M Q
K
DPKQ (R)∗ DPK Q (R)∗ =
K K K K K K
= (2K + 1) DPK Q (R) . (A4.1)
P P P Q Q Q
K
Substitution into Eq. (7.84) gives
CI (αJM M , α J M M ) = CI (αJN N , α J N N )
N N N N
× (2K + 1)(2K + 1)(2K + 1) (−1)N −M+N −M
KK K
J J K J J K J J K
×
N −N P M −M Q N −N P
J J K K K K K K K
× DPK Q (R) .
M −M Q P P P Q Q Q
Since the right-hand side must be independent of the rotation R, the index K can
only take the value K = 0. This implies K = K , P = −P , Q = −Q . We thus
obtain, using Eq. (2.26a)
CI (αJM M , α J M M ) = (−1)M −M (2J + 1)
J J K
J J K
× ΓI(K) (αJ, α J ) , (A4.2)
−M M Q −M M Q
K
where we have defined another set of multipole components of the collisional rates,
given by
2K + 1
ΓI(K) (αJ, α J ) =
2J + 1
N −N J J K J J K
× (−1)
−N N P −N N P
N N N N P
× CI (αJN N , α J N N ) .
The multipole components ΓI(K) are of course related to the multipole components
(K)
CI defined in Eq. (7.87). To find this relation we start from Eq. (7.87) and write
APPENDIX 801
the product of the two 3-j symbols via Eq. (2.43)
J J K J J K J J K J J K
=
N −N P N −N P N −N P N −N −P
= (−1)2J+2J +K −K−N −N
K !
J J K J J K J J K
× (2K + 1) .
J J K −N N P −N N P
We thus obtain, after some algebra
CI(K) (αJ, α J ) = (2J + 1)(2J + 1)
!
J+J +K −K J J K
× (−1) ΓI(K ) (αJ, α J ) ,
J J K
K
with the inverse formula (that can be deduced with the help of Eq. (2.39))
2K + 1
ΓI(K) (αJ, α J ) =
(2J + 1)(2J + 1)
!
J J K
× (−1)J+J −K +K
(2K + 1) CI(K ) (αJ, α J ) .
J J K
K
In particular, using Eq. (2.36a) we obtain
(K)
CI(0) (αJ, α J ) = ΓI (αJ, α J ) .
K
The coupling of rotation matrices in Eqs. (A4.1) is particularly interesting be-
cause of the decomposition of the transfer collisional rate expressed by Eq. (A4.2).
The structure of the 3-j symbols (cf. the expression of the transfer radiative rate
TA (αJM M , α J M M ) in Eq. (7.9a)) suggests for Eq. (A4.2) the following inter-
pretation: the interaction between the atomic system and the collider is described
by a sum of tensor operators of rank K acting on the state vectors of the atom. The
summation over K represents the contribution to the rate of the various operators.
The important fact to be stressed here is that, in many cases, the interaction is
suitably described by just one operator of rank K̃.1 In such cases one has
CI(K) (αJ, α J ) = (2J + 1)(2J + 1)
!
J J K
× (−1)J+J +K̃−K
ΓI(K̃) (αJ, α J ) ,
J J K̃
1
For instance, the electron-atom interaction can be treated under the Born approximation
provided the energy of the colliding electrons is much larger than the threshold energy. In this
case the interaction Hamiltonian depends on the dynamical variables of the atomic system only
through the dipole operator (which is a tensor of rank 1), so that K̃ = 1.
802 APPENDIX
so that the multipole component of rank K is related to the multipole component
of rank 0 by the equation
!
J J K
J J K̃
CI (αJ, α J ) = (−1)K
(K)
! CI(0) (αJ, α J ) .
J J 0
J J K̃
The same line of reasoning can be followed for the collisional rates due to supere-
lastic and to elastic collisions (considered in Sects. 7.13.a and 7.13.b, respectively).
As far as superelastic collisions are concerned, the expansion in multipole compo-
nents of the form ΓS(K) is given by
CS (αJM M ,αu Ju Mu Mu ) = (−1)Mu −Mu (2Ju + 1)
J Ju K
J Ju K
× ΓS(K) (αJ, αu Ju )
−M Mu Q −M Mu Q
K
with
2K + 1
ΓS(K) (αJ, αu Ju ) =
2Ju + 1
Nu −Nu J Ju K J Ju K
× (−1)
−N Nu P −N Nu P
N N Nu Nu P
× CS (αJN N , αu Ju Nu Nu ) .
The multipole components CS(K) defined in Eq. (7.89) are related to these new
components by
CS(K) (αJ, αu Ju ) = (2J + 1)(2Ju + 1)
!
J+Ju +K −K J J K
× (−1) ΓS(K ) (αJ, αu Ju ) .
Ju Ju K
K
The corresponding expansion for elastic collisional rates is
−M
CE (αJM M ,αJM M ) = (−1)M (2J + 1)
J J K
J J K
× ΓE(K) (αJ)
−M M Q −M M Q
K
with
2K + 1
ΓE(K) (αJ) =
2J + 1
N −N J J K J J K
× (−1)
−N N P −N N P
NN N N P
× CE (αJN N , αJN N ) ,
APPENDIX 803
and the relation with the multipole components CE(K) defined in Eq. (7.100) is
!
−K J J K
CE(K) (αJ) = (2J + 1) (−1)2J+K ΓE(K ) (αJ) . (A4.3)
J J K
K
When the interaction between the atomic system and the collider is described by
a single operator of rank K̃ we have
!
J J K
Ju Ju K̃
CS(K) (αJ, αu Ju ) = (−1)K ! CS(0) (αJ, αu Ju )
J J 0
Ju Ju K̃
and !
J J K
J J K̃
CE(K) (αJ) = (−1)K ! CE(0) (αJ) .
J J 0
J J K̃
A5. Explicit Expression for the Exponential of the
Propagation Matrix
We want to calculate the expression
−x Z
e ,
where x is a real variable and Z is a 4 × 4 real matrix of the form
⎛ ⎞
w0 w1 w2 w3
⎜ w1 w0 y3 −y2 ⎟
Z=⎝ ⎠.
w2 −y3 w0 y1
w3 y2 −y1 w0
Let us introduce the formal vectors
w
= (w1 , w2 , w3 ) , y = (y1 , y2 , y3 )
and their complex linear combinations
1 1
a = w
+ iy = w1 + i y1 , w2 + i y2 , w3 + i y3
2 2
∗ 1 1
b = a = w − i y = w1 − i y1 , w2 − i y2 , w3 − i y3 .
2 2
804 APPENDIX
Introducing also the six 4 × 4 matrices
= (A , A , A ) ,
A = (B , B , B )
B
1 2 3 1 2 3
defined by
⎛ ⎞
0 1 0 0
⎜ 1 0 0 0 ⎟
A1 = B1∗ = ⎝ ⎠ (A5.1a)
0 0 0 −i
0 0 i 0
⎛ ⎞
0 0 1 0
⎜ 0 0 0 i⎟
A2 = B2∗ = ⎝ ⎠ (A5.1b)
1 0 0 0
0 −i 0 0
⎛ ⎞
0 0 0 1
⎜ 0 0 −i 0⎟
A3 = B3∗ = ⎝ ⎠, (A5.1c)
0 i 0 0
1 0 0 0
the matrix Z can be written in the form
+ b · B
Z = w0 1 + a · A , (A5.2)
where 1 is the 4 × 4 identity matrix.
The matrices A and B satisfy the following algebra
Aj , Bk = Aj Bk − Bk Aj = 0 (A5.3)
Aj Ak = δjk 1 + i jkl Al (A5.4)
l
Bj Bk = δjk 1 − i jkl Bl , (A5.5)
l
where δjk is the Kronecker symbol and jkl is the antisymmetric tensor (defined
on p. 7). Since by a well-known theorem on matrices1
M1 + M2 M1 M2
e = e e if M1 , M2 = 0 , (A5.6)
we have from Eqs. (A5.2) and (A5.3)
−x Z −x w0 1 −x
a·A
−x b·B
e = e e e . (A5.7)
1
See e.g. Cohen-Tannoudji et al., 1977, p.170.
APPENDIX 805
Now we turn to the evaluation of the single factors appearing in this equation.
Recalling Eq. (8.21), and observing that 1 n = 1 , we have
∞
−x w0 1 (−1)n −x w0
e = (xw0 )n 1 = e 1 . (A5.8)
n=0
n!
On the other hand, using Eq. (A5.4) one gets
2=
a · A aj Aj ak Ak
jk
= aj ak δjk 1 + i jkl aj ak Al = a2 1 ,
jk jkl
where
1 2
a2 = a · a = w − y2 + 2 i w
· y (A5.9)
4
with
·w
w2 = w = w12 + w22 + w32
y 2 = y ·
y = y12 + y22 + y32 .
Recalling again Eq. (8.21) we thus obtain
−x
a·A
x2 a2 x4 a4 x x3 a2 x5 a4 ,
e = 1+ + + ··· 1 − + + + · · · a · A
2! 4! 1! 3! 5!
which can also be written in the form1
−x
a·A
sinh(xa) ,
e = cosh(xa) 1 − a · A (A5.10)
a
where a is either of the two possible determinations of the square root of the
complex number a2 (note that Eq. (A5.10) is invariant under interchange of a
and −a ). In a similar way one gets
−x b·B
sinh(xb)
e = cosh(xb) 1 − b·B , (A5.11)
b
where
1 2
b2 = b · b = (a2 )∗ = w − y2 − 2 i w
· y (A5.12)
4
and where b is either of the two possible determinations of the square root of b2 .
1
The special case a2 = 0 will be considered later.
806 APPENDIX
Substitution of Eqs. (A5.8), (A5.10), and (A5.11) into Eq. (A5.7) gives
−x Z −x w0 sinh(xa)
e = e cosh(xa) 1 − a · A
a
sinh(xb)
× cosh(xb) 1 − b·B . (A5.13)
b
Now we choose the determination of a and b setting, in terms of the real numbers α
and β
1 1
a= α + iβ , b = a∗ = α − iβ , (A5.14)
2 2
with the supplementary condition
α ≥ 0. (A5.15)
Substituting Eqs. (A5.14) into Eq. (A5.13) we obtain
−x Z −x w0 1
e = e cosh(xα) + cos(xβ) 1
2
a · A
− sinh(xα) + i sin(xβ)
α + iβ
b · B
− sinh(xα) − i sin(xβ)
α − iβ
)!
) (b · B
(a · A
+ 2 cosh(xα) − cos(xβ) . (A5.16)
α2 + β 2
The expressions for α and β can be found using Eqs. (A5.9) and (A5.14). One
gets the equation
w2 − y 2 + 2 i w
·
y = α2 − β 2 + 2 i αβ ,
which can be solved, adding the condition (A5.15), to give
α = Λ+ (w
, y ) β = σ Λ− (w
, y ) , (A5.17)
where1
Λ+ (w
, y ) = (w2 − y 2 )2 /4 + (w
· y )2 + (w2 − y 2 )/2
Λ− (w
, y ) = (w2 − y 2 )2 /4 + (w
· y )2 − (w2 − y 2 )/2 (A5.18)
· y
w
· y ) =
σ = sign (w . (A5.19)
|w
· y |
1
Note that when w · y = 0, σ is undetermined. In this case, however, being Λ− (
w , y ) = 0,
the final expression for exp(−xZ) is independent of σ (see Eqs. (A5.20)-(A5.21)).
APPENDIX 807
,
Finally, substituting Eq. (A5.17) into Eq. (A5.16) and evaluating the matrices a · A
b · B
, and their product, one gets the expression
−x Z −x w
e = e 0
1
× cosh x Λ+ (w
, y ) + cos x Λ− (w , y ) M1 (w , y )
2
− sin x Λ− (w , y ) M2 (w , y )
− sinh x Λ+ (w , y ) M3 (w , y )
!
1
+ cosh x Λ+ (w , y ) − cos x Λ− (w , y ) M4 (w , y ) , (A5.20)
2
where the four matrices Mi (w
, y ) are given by
M1 (w
, y ) = 1 (A5.21a)
M2 (w
, y ) = (A5.21b)
⎛ ⎞
0 Λ− w1 − σΛ+ y1 Λ− w2 − σΛ+ y2 Λ− w3 − σΛ+ y3
⎜ ⎟
1 ⎜ Λ− w1 − σΛ+ y1 0 σΛ+ w3 + Λ− y3 −σΛ+ w2 − Λ− y2 ⎟
= ⎜⎜
⎟
⎟
Θ ⎝ Λ− w2 − σΛ+ y2 −σΛ+ w3 − Λ− y3 0 σΛ+ w1 + Λ− y1 ⎠
Λ− w3 − σΛ+ y3 σΛ+ w2 + Λ− y2 −σΛ+ w1 − Λ− y1 0
M3 (w
, y ) = (A5.21c)
⎛ ⎞
0 Λ+ w1 + σΛ− y1 Λ+ w2 + σΛ− y2 Λ+ w3 + σΛ− y3
⎜ ⎟
1 ⎜ Λ+ w1 + σΛ− y1 0 −σΛ− w3 + Λ+ y3 σΛ− w2 − Λ+ y2 ⎟
= ⎜ ⎟
Θ⎜⎝ Λ+ w2 + σΛ− y2 σΛ− w3 − Λ+ y3 0
⎟
−σΛ− w1 + Λ+ y1 ⎠
Λ+ w3 + σΛ− y3 −σΛ− w2 + Λ+ y2 σΛ− w1 − Λ+ y1 0
M4 (w
, y ) =
⎛ ⎞
t2 w3 y2 − w2 y3 w1 y3 − w3 y1 w2 y1 − w1 y2
⎜ ⎟
2 ⎜ −w3 y2 + w2 y3 w12 + y12 − t2 w1 w2 + y1 y2 w3 w1 + y3 y1 ⎟
= ⎜ ⎟ (A5.21d)
Θ⎜⎝ −w1 y3 + w3 y1 w1 w2 + y1 y2 w22 + y22 − t2
⎟
w2 w3 + y2 y3 ⎠
−w2 y1 + w1 y2 w3 w1 + y3 y1 w2 w3 + y2 y3 w32 + y32 − t2
with t2 = (w2 + y 2 )/2 and
Θ = α2 + β 2 = Λ2+ + Λ2− = 2 (w2 − y 2 )2 /4 + (w
· y )2 . (A5.22)
808 APPENDIX
Equation (A5.20) can also be written in a different form, which involves only
exponential factors (although complex)
−x Z −x [ w0 + Λ+ (w,
y)] −x [ w0 − Λ+ (w,
y)]
e = e N1 (w
, y ) + e N2 (w
, y )
−x [ w0 + i Λ− (w,
y)] −x [ w0 − i Λ− (w,
y)]
+e N3 (w
, y ) + e N4 (w
, y ) , (A5.23)
where
1
N1 (w
, y ) = M1 (w
, y ) + 2 M3 (w
, y ) + M4 (w
, y )
4
1
N2 (w
, y ) = M1 (w
, y ) − 2 M3 (w
, y ) + M4 (w
, y )
4
1
N3 (w
, y ) = M1 (w
, y ) − 2 i M2 (w
, y ) − M4 (w
, y )
4
1
N4 (w
, y ) = M1 (w
, y ) + 2 i M2 (w
, y ) − M4 (w , y )∗ . (A5.24)
, y ) = N3 (w
4
The explicit expressions of the matrices Ni are rather cumbersome and will not
be given here.
It should be noticed that the matrices Mi (w , y ) defined in Eqs. (A5.21b,c,d)
– and hence the expression (A5.20) for exp(−xZ) – are ill-defined when Θ = 0,
that is
Λ+ (w , y ) = Λ− (w
, y ) = 0 ,
which implies (see Eq. (A5.22))
w2 = y 2 and · y ) = 0 .
(w
Since in this case we have from Eqs. (A5.9) and (A5.12)
a2 = b 2 = 0 ,
it is easily seen that expressions (A5.10) and (A5.11) reduce to
−x
a·A
−x b·B
e ,
= 1 − x a · A e = 1 − x b · B
.
Substituting into Eq. (A5.7) and applying the same procedure as before, one finds
that Eq. (A5.20) must be replaced, in this particular case, by the expression
−x Z −x w0 1 2
e = e 1 − x G(w
, y ) + x G(w
, y )2 ,
2
where ⎛ ⎞
0 w1 w2 w3
⎜w −y2 ⎟
⎜ 0 y3 ⎟
, y ) = ⎜ 1
G(w ⎟ . (A5.25)
⎝ w2 −y3 0 y1 ⎠
w3 y2 −y1 0
APPENDIX 809
This expression can also be obtained from Eqs. (A5.20)-(A5.21) performing the
, y ) → 0, Λ− (w
limit Λ+ (w , y ) → 0.
A6. Diagonalization of the Propagation Matrix
We want to determine the matrices X and X −1 such that
X −1 K X = K , (A6.1)
where K is the propagation matrix defined in Eq. (8.3) and K is the diagonal
matrix
Kij = λi δij (i, j = 1, 2, 3, 4) , (A6.2)
λi being the eigenvalues of K (given by Eqs. (8.33)), that we assume to be distinct
(see footnote on p. 358).
The matrices X and X −1 can be found as follows. First we determine the
eigenvectors u(i) of the matrix K , defined by
K u(i) = λi u(i) (i = 1, 2, 3, 4) . (A6.3)
Using these eigenvectors we construct a matrix Z whose columns are the four
eigenvectors u(i) each multiplied by an arbitrary constant ci ,
(j)
Zij = cj ui . (A6.4)
From Eqs. (A6.2) and (A6.3) it follows that the matrix Z satisfies the equation
K Z = Z K . (A6.5)
Then we find the ‘left’ eigenvectors v (i) of the matrix K , defined by1
(i) † †
v K = λi v (i) , (A6.6)
and construct a matrix Y whose rows are the four eigenvectors (v (i) )† each mul-
tiplied by an arbitrary constant di ,
(i)
Yij = di vj . (A6.7)
The matrix Y satisfies the equation
Y K = K Y . (A6.8)
1
It can be easily proved that the quantities λi in Eqs. (A6.6) and (A6.3) are the same.
810 APPENDIX
If we now consider the product Y K Z , we have from Eqs. (A6.5) and (A6.8)
Y K Z = Y Z K = K Y Z ,
which proves that the matrix Y Z commutes with the diagonal matrix K . Since
the four eigenvalues of K have been assumed to be non-degenerate, it follows
that Y Z is also diagonal. On the other hand, the diagonal elements of Y Z are
given by
4
4
(i) (i)
(Y Z)ii = Yik Zki = ci di vk uk (i = 1, 2, 3, 4) .
k=1 k=1
Thus we get
Y Z =1
if we choose the coefficients ci , di in such a way that
1
ci di = . (A6.9)
%
4
(i) (i)
vk uk
k=1
It follows that a solution to Eq. (A6.1) is
X = Z, X −1 = Y ,
with Z and Y given by Eqs. (A6.4) and (A6.7), respectively, and with the coeffi-
cients ci and di satisfying Eq. (A6.9).
Let’s now apply this method to find an explicit expression for the matrices X
and X −1 . The first step is to determine the eigenvectors of the propagation ma-
trix K . From Eq. (A6.3) we have
⎧
⎪ (ηI − λ) u1 + ηQ u2 + ηU u3 + ηV u4 = 0
⎪
⎪
⎨ η u + (η − λ) u + ρ u − ρ u = 0
Q 1 I 2 V 3 U 4
(A6.10)
⎪
⎪ η u − ρ u + (η − λ) u + ρ Q u4 = 0
⎪
⎩
U 1 V 2 I 3
ηV u1 + ρU u2 − ρQ u3 + (ηI − λ) u4 = 0 ,
which is a system of four homogeneous, linear equations in the four unknowns u1 ,
u2 , u3 , u4 . Solutions other than the zero solution (u1 = u2 = u3 = u4 = 0) exist
only if the determinant of the system vanishes: this leads to Eq. (8.30), whence the
eigenvalues λi are obtained (see Eqs. (8.33)). Substituting one of the eigenvalues
for λ , we can eliminate one of the equations and assign an arbitrary value to one
of the unknowns. For instance, we can reject the first equation and write
⎧ (i) (i) (i) (i)
⎪
⎪ (η − λi ) u2 + ρV u3 − ρU u4 = − ηQ u1
⎨ I
(i) (i) (i) (i)
−ρV u2 + (ηI − λi ) u3 + ρQ u4 = − ηU u1 (A6.11)
⎪
⎪
⎩ (i) (i) (i) (i)
ρU u2 − ρQ u3 + (ηI − λi ) u4 = − ηV u1 ,
APPENDIX 811
whose solution is1
(i)
u1 = (λi − ηI ) (λi − ηI )2 + ρ2Q + ρ2U + ρ2V
(i)
u2 = (λi − ηI )2 ηQ + (λi − ηI )(ηU ρV − ηV ρU )
+ (ηQ ρQ + ηU ρU + ηV ρV ) ρQ
(i)
u3 = (λi − ηI )2 ηU + (λi − ηI )(ηV ρQ − ηQ ρV )
+ (ηQ ρQ + ηU ρU + ηV ρV ) ρU
(i)
u4 = (λi − ηI )2 ηV + (λi − ηI )(ηQ ρU − ηU ρQ )
+ (ηQ ρQ + ηU ρU + ηV ρV ) ρV . (A6.12)
The left eigenvectors are found in a strictly similar way; their expression is obtained
by substituting ρQ → − ρQ , ρU → − ρU , ρV → − ρV in Eqs. (A6.12).
From the left eigenvectors we then construct the matrix X −1 without any mul-
tiplicative factor (the coefficients di in Eq. (A6.7) are set to 1). This is just the
matrix given in Eq. (8.42a). To construct the matrix X we must first evaluate the
coefficients ci . Using the compact notations of Eqs. (8.31) and (8.41) we have
4
(i) (i) 2
uk vk = (λi − ηI )2 (λi − ηI )2 + ρ2 + (λi − ηI )4 η 2
k=1
− (λi − ηI )2 ζ 2 + 2 (λi − ηI )2 (η · ρ
)2 + (η · ρ
)2 ρ2 ,
and since from Eq. (8.32)
(λi − ηI )2 (λi − ηI )2 + ρ2 = (λi − ηI )2 η 2 + (η · ρ )2
and from Eq. (8.41)
η 2 ρ2 − ζ 2 = (
η·ρ
)2 ,
we can write
4
(i) (i)
uk vk = 2 (λi − ηI )4 η 2 + 2 (λi − ηI )2 (η · ρ
)2 + (η · ρ
)2 ρ2 .
k=1
The coefficients ci then follow from Eq. (A6.9), and this leads to the matrix X as
given in Eq. (8.42b).
It should be emphasized that the expressions for X and X −1 given in Eqs. (8.42)
are valid provided two conditions are satisfied, namely: i) the four eigenvalues of
1
Expressions (A6.12) are correct only if the determinant of the system (A6.11) is non-
zero. However, the system (A6.10) can always be reduced – provided the four eigenvalues λi are
distinct – to a system of three equations in three unknowns having a non-zero determinant. In
other words, the matrix of the coefficients of the system (A6.10) is of rank 3.
812 APPENDIX
the propagation matrix are distinct, and ii) the determinant of the system (A6.11)
is non-zero. As apparent from Eqs. (8.33), the first condition means that both
Λ+ (
η ,ρ
) and Λ− (
η,ρ
) must be non-zero, thus it is equivalent to (see Eqs. (A5.18))
η·ρ
= 0 . (A6.13)
The second condition leads to the further constraint
ρ2 − Λ− ( )2 = 0 ,
η ,ρ
or, using again Eqs. (A5.18)
η 2 ρ2 = (
η·ρ
)2 . (A6.14)
Expressions (8.42) are correct only if conditions (A6.13) and (A6.14) are satisfied.
A7. Formulae for the Calculation of the Evolution Operator
To shorten notations we write the coefficients E (k) (s, s ), introduced in Eq. (8.76),
in the form
E (k) (s, s ) = Xk + i Yk (k = 0, 1, 2, 3) , (A7.1)
with Xk and Yk real. Using Eqs. (8.62) and (8.63), the transfer equation (8.78)
becomes
⎛ ⎞ ⎛ ⎞⎛ ⎞
X0 ηI ηQ ηU ηV 0 −ρQ −ρU −ρV X0
⎜X ⎟ ⎜η −ρ −ρ −ηU ⎟ ⎜ ⎟
⎜ 1⎟ ⎜ Q η I ρ V U Q 0 ηV ⎟ ⎜ X1 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ X2 ⎟ ⎜ ηU −ρV ηI ρQ −ρU −ηV 0 ηQ ⎟ ⎜X ⎟
⎜ ⎟ ⎜ ⎟⎜ 2 ⎟
⎜ X ⎟ ⎜ −ρQ −ρV −ηQ 0 ⎟ ⎜ ⎟
d ⎜ 3⎟ ⎜ ηV ρU ηI ηU ⎟ ⎜ X3 ⎟
2 ⎜ ⎟= − ⎜ ⎟⎜ ⎟.
ds ⎜ Y0 ⎟ ⎜ 0 ρQ ρU ρV ηI ηQ ηU ηV ⎟⎜ Y0 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜Y ⎟ ⎜ρ −ηV −ρU ⎟ ⎜ ⎟
⎜ 1⎟ ⎜ Q 0 ηU ηQ ηI ρV ⎟⎜ Y1 ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎝ 2⎠
Y ⎝ ρU ηV 0 −ηQ ηU −ρV ηI ρQ ⎠⎝ Y2 ⎠
Y3 ρV −ηU ηQ 0 ηV ρU −ρQ ηI Y3
It is interesting to note that the 8 × 8 matrix can be written in the form
K −K
,
K K
where K is the propagation matrix (Eq. (8.3)) and K is the matrix obtained
from K setting ηI = 0 and substituting
ηQ → ρQ ρQ → − ηQ
ηU → ρU ρU → − ηU
ηV → ρV ρV → − ηV .
APPENDIX 813
As far as Eq. (8.77) is concerned, the trace of the product of four Pauli spin
matrices is easily evaluated using Eqs. (8.56), (8.57), (8.62) and the relation
ijk lmk = δil δjm − δim δjl .
k
We get
Tr τi τj τk τl = 2 δij δkl − δik δjl + δil δjk (i, j, k, l = 1, 2, 3)
(if one of the indices is zero, the trace is given directly from Eq. (8.62)). Substitution
into Eq. (8.77) yields the expressions of the matrix elements Oij in terms of the
quantities Xk , Yk defined in Eq. (A7.1)
O00 = X02 + X12 + X22 + X32 + Y02 + Y12 + Y22 + Y32
O11 = X02 + X12 − X22 − X32 + Y02 + Y12 − Y22 − Y32
O22 = X02 − X12 + X22 − X32 + Y02 − Y12 + Y22 − Y32
O33 = X02 − X12 − X22 + X32 + Y02 − Y12 − Y22 + Y32
O01 = 2 (X0 X1 + Y0 Y1 − X2 Y3 + X3 Y2 )
O10 = 2 (X0 X1 + Y0 Y1 + X2 Y3 − X3 Y2 )
O02 = 2 (X0 X2 + Y0 Y2 − X3 Y1 + X1 Y3 )
O20 = 2 (X0 X2 + Y0 Y2 + X3 Y1 − X1 Y3 )
O03 = 2 (X0 X3 + Y0 Y3 − X1 Y2 + X2 Y1 )
O30 = 2 (X0 X3 + Y0 Y3 + X1 Y2 − X2 Y1 )
O12 = 2 (X1 X2 + Y1 Y2 + X0 Y3 − X3 Y0 )
O21 = 2 (X1 X2 + Y1 Y2 − X0 Y3 + X3 Y0 )
O23 = 2 (X2 X3 + Y2 Y3 + X0 Y1 − X1 Y0 )
O32 = 2 (X2 X3 + Y2 Y3 − X0 Y1 + X1 Y0 )
O31 = 2 (X1 X3 + Y1 Y3 + X0 Y2 − X2 Y0 )
O13 = 2 (X1 X3 + Y1 Y3 − X0 Y2 + X2 Y0 ) .
A8. The Feautrier Method: Numerical Details
To obtain a numerical solution of Eq. (9.183), subjected to the boundary conditions
(9.185) and (9.186), we introduce a grid of N points (τ1 , τ2 , . . . , τN ) not necessarily
evenly spaced.
814 APPENDIX
Let us consider two adjacent intervals (τk−1 , τk ) and (τk , τk+1 ), and their mid-
points
τk−1 + τk τk + τk+1
τ̃k−1 = , τ̃k = (k = 2, 3, . . . , N − 1) . (A8.1)
2 2
Using a standard finite-differences formula we can write
dG A−1 −1
k−1 + Ak Gk − Gk−1
A−1 =
dτr τr = τ̃k−1 2 τk − τk−1
dG A−1 −1
k + Ak+1 Gk+1 − Gk
A−1 = , (A8.2)
dτr τr = τ̃k 2 τk+1 − τk
where a subscript k means that the quantity is evaluated at τr = τk . Similarly, we
can replace Eq. (9.183) by the following
dG dG
A−1 − A−1
dτr τr = τ̃k dτr τr = τ̃k−1
= Ak Gk − bk . (A8.3)
τ̃k − τ̃k−1
Substitution of Eqs. (A8.1) and (A8.2) into Eq. (A8.3) leads to the linear system
− x(k) Gk−1 + y (k) Gk − z (k) Gk+1 = u(k) (k = 2, 3, . . . , N − 1) , (A8.4)
where
A−1
k−1 + Ak
−1
x(k) =
(τk − τk−1 ) (τk+1 − τk−1 )
A−1
k + Ak+1
−1
z (k) =
(τk+1 − τk ) (τk+1 − τk−1 )
y (k) = x(k) + z (k) + Ak
u(k) = bk . (A8.5)
Boundary conditions can also be written in the form of difference equations. For
the first interval we can write
dG A−1 + A−1 G2 − G1
A−1 = 1 2
.
dτr τr = τ̃1 2 τ2 − τ1
On the other hand, following a method suggested by Auer (1967) for the scalar
case, we have by a first-order expansion
!
dG dG d dG τ2 − τ1
A−1 = A−1 + A−1 .
dτr τ = τ̃ dτr τ = τ dτr dτr τ =τ 2
r 1 r 1 r 1
APPENDIX 815
Equating the expressions in the right-hand sides and applying the boundary con-
dition (9.185) at τr = τ1 we obtain, with the help of Eq. (9.183)
A−1 + A−1 G2 − G1 τ − τ1
1 2
= G1 + A1 G1 − b1 2 ,
2 τ2 − τ1 2
which can be cast into the form
y (1) G1 − z (1) G2 = u(1) , (A8.6)
where
A−1 + A−1
z (1) = 1 2
2 (τ2 − τ1 )2
1 1
y (1) = z (1) + 1 + A1
τ2 − τ1 2
1
u(1) = b . (A8.7)
2 1
Performing analogous calculations at τr = τb = τN , one obtains the equation
− x(N ) GN −1 + y (N ) GN = u(N ) , (A8.8)
where
A−1
N −1 + AN
−1
x(N ) =
2 (τN − τN −1 )2
1 1
y (N ) = x(N ) + 1 + AN
τN − τN −1 2
1 1
u(N ) = b + I (b) . (A8.9)
2 N τN − τN −1
Collecting together Eqs. (A8.4), (A8.6), and (A8.8), we obtain a tridiagonal
system of N equations (each being a vector equation of order 4). Such system can
be solved via a recursive algorithm.
Left-multiplication of Eq. (A8.6) by [ y (1) ]−1 gives
G1 = g1 + Γ1 G2 , (A8.10)
where
−1
g1 = y (1) u(1)
−1
Γ1 = y (1) z (1) . (A8.11)
816 APPENDIX
Substitution of Eq. (A8.10) into the first of Eqs. (A8.4) and left-multiplication by
[ y (2) − x(2) Γ1 ]−1 gives G2 as a function of G3 . Iteration of this procedure leads
to the equations
Gk = gk + Γk Gk+1 , (A8.12)
with
−1
gk = y (k) − x(k) Γk−1 u(k) + x(k) gk−1
−1
Γk = y (k) − x(k) Γk−1 z (k) , (A8.13)
and with k ranging from 2 to (N − 1). For k = 1, Eq. (A8.10) is incorporated in
Eq. (A8.12) if we define
g0 = Γ0 = 0 .
Finally, substitution of the expression for GN −1 given by Eq. (A8.12) into Eq.
(A8.8) allows the vector GN to be determined,
−1
GN = y (N ) − x(N ) ΓN −1 u(N ) + x(N ) gN −1 .
The vectors GN −1 , GN −2 , . . . , G1 can then be recursively obtained from Eqs.
(A8.12). In particular, 2 G1 is just the emerging Stokes vector (see Eqs. (9.181)
and (9.184)).
The formulae now derived can be conveniently used to test the accuracy of the
Feautrier solution. To this aim, let us consider the case of a purely absorbing
atmosphere (Sc = SL = 0). From Eqs. (9.164), (A8.5), (A8.7) and (A8.9) one
has that all the vectors u(k) are zero except u(N ) . This implies, according to
Eqs. (A8.11) and (A8.13), that all the gk (k = 1, 2, . . . , N − 1) are also zero.
Therefore, Eq. (A8.12) reduces to
Gk = Γk Gk+1 (k = 1, 2, . . . , N − 1) . (A8.14)
This formula gives a precise physical meaning to the matrix Γk . For the particular
atmosphere that we are considering, the emission vector b(τr ) is zero and – because
of the boundary condition (9.184) – the vector Ĩ is also zero at any optical depth.
Therefore, at any optical depth we have from Eqs. (9.181) G = I / 2, so that
Eq. (A8.14) can be rewritten in the form
Ik = Γk Ik+1 .
Comparison with Eq. (9.165) – written for the case b(τr ) = 0 – shows that Γk gives
a numerical representation of the evolution operator associated with the interval
(τk , τk+1 ), a result that was pointed out by Rees et al. (1989),
Γk ≈ O(τk , τk+1 ) .
APPENDIX 817
Let’s now assume, in addition to the pure absorption hypothesis, that the propa-
gation matrix A is independent of optical depth. It follows from Eqs. (A8.7) and
(A8.11) that
−1
1 2
Γ1 = A−1 + τ2 − τ1 1 + τ2 − τ1 A A−1 ,
2
whence
1 2
Γ1−1 = 1 + τ2 − τ1 A + τ2 − τ1 A2 . (A8.15)
2
Expansion of Γ1 into power series of A
∞
Γ1 = c A
=0
and use of the relation Γ1 Γ1−1 = 1 , with Γ1−1 given by Eq. (A8.15), yields
1 2 1 4
Γ1 = 1 − τ2 − τ1 A + τ2 − τ1 A2 − τ2 − τ1 A4 + · · · (A8.16)
2 4
On the other hand, the analytical expression of the evolution operator associated
with the interval (τ1 , τ2 ) is (cf. Eq. (9.103))
−(τ2 − τ1 ) A
O(τ1 , τ2 ) = e ,
thus the expression for Γ1 provided by the Feautrier method is correct up to second-
order terms.
It can be shown by recursion that this statement is valid for all the matrices
Γk (k = 1, 2, . . . , N − 1). Considering for instance the matrix Γ2 , we have from
Eqs. (A8.5) and (A8.13), with easy algebra
τ3 − τ1 1 τ − τ2
Γ2−1 = 1 + τ3 − τ2 τ3 − τ1 A2 − 3 Γ .
τ2 − τ1 2 τ2 − τ1 1
Substitution of Eq. (A8.16) and expansion of Γ2 into power series of A lead to the
expression
1 2
Γ2 = 1 − τ3 − τ2 A + τ3 − τ2 A2
2
1 3
− τ3 − τ2 τ2 − τ1 + τ3 − τ2 )3 A4 + · · · ,
4
which should be compared with the analytical result
−(τ3 − τ2 ) A
O(τ2 , τ3 ) = e .
818 APPENDIX
A9. The Diagonal Element Lambda-Operator (DELO) Method:
Numerical Details
Starting at Eq. (9.189), we introduce an optical-depth scale y defined by
dy = aI (τr ) dτr , (A9.1)
so that
d
I(y) = I(y) − s(y) . (A9.2)
dy
Let us introduce a grid of N points (y1 , y2 , . . . , yN ) not necessarily evenly spaced.
Integration of Eq. (A9.2) in the interval (yk , yk+1 ) gives
yk+1
−(y− yk ) −(yk+1 − yk )
I(yk ) = e s(y) dy + e I(yk+1 ) . (A9.3)
yk
Now we assume s(y) to be linear in y in the same interval,
(yk+1 − y) sk + (y − yk ) sk+1
s(y) = , (A9.4)
yk+1 − yk
with sk = s(yk ). Substituting Eq. (A9.4) into Eq. (A9.3) and evaluating analyti-
cally the integral, one gets, with the help of Eq. (9.190)
Ik = pk + Πk Ik+1 , (A9.5)
where Ik = I(yk ) and
−1
pk = 1 + mk Ak mk bk + nk bk+1
−1
Πk = 1 + mk Ak k 1 − nk Ak+1 , (A9.6)
with
∆k + k − 1
∆k = yk+1 − yk mk =
∆k
−∆k 1 − k (∆k + 1)
k = e nk = . (A9.7)
∆k
Comparison of Eqs. (A9.5) and (9.165) shows that Πk gives a numerical represen-
tation of the evolution operator associated with the interval (yk , yk+1 ),
Πk ≈ O(yk , yk+1 ) .
APPENDIX 819
Once the boundary condition IN = I (b) is specified (see footnote 1 on p. 441), the
emerging Stokes vector I1 can be determined by recursive application of Eq. (A9.5).
To test the accuracy of the DELO solution against the step length, we assume
the matrix A to be independent of y in the interval (yk , yk+1 ) and compare Πk
with the exact analytical expression of the evolution operator. From Eqs. (A9.6)
we have −1
Πk = 1 + mk A k 1 − nk A .
1 + mk A )−1 into power series of A yields
Expansion of the matrix (1
∞
−1
1 + mk A =1 + (−1)r mrk A r ,
r=1
whence ∞
nk
Πk = k 1 + (−1) mk +r
mr−1
k A r .
r=1
k
The quantities mk and (mk + nk / k ) can be expanded into power series of ∆k ;
up to third order we have
∆k ∆2 ∆3
mk = − k + k + O(∆4k )
2 6 24
∆k −∆k
nk e + e −2 ∆3k
mk + = = ∆k + + O(∆5k ) ,
k ∆k 12
so that, up to the same order of approximation, one gets
1 2 2 ∆3
Πk = k 1 − ∆k A + ∆k A − k A + 2A 2 + 3A 3 + O(∆4k ) . (A9.8)
2 12
On the other hand, the analytical expression of the evolution operator is (see
Eqs. (9.187), (9.104) and (A9.1))
−(yk+1 − yk ) −(yk+1 − yk ) A −∆k A
O(yk , yk+1 ) = e e = k e
1 2 2 1
= k 1 − ∆k A + ∆k A − ∆3k A 3 + O(∆4k ) .
2 6
Comparison with Eq. (A9.8) shows that the expression for Πk resulting from the
DELO method is correct up to second-order terms in ∆k A . The ‘error’ is
−∆k ∆3k
Ek = O(yk , yk+1 ) − Πk = e A + 2A 2 + A 3 + O(∆4k ) .
12
Since the series expansion in the right-hand side is meaningful only when ∆k 1,
we can replace the exponential by 1. The ‘error’ Ek can be easily expressed in
820 APPENDIX
terms of τr and A rather than y and A . Denoting by (τ1 , τ2 , . . . , τN ) the τr grid-
points corresponding to (y1 , y2 , . . . , yN ), integration of Eq. (A9.1) in the interval
(τk , τk+1 ) yields
yk+1 − yk = aI (τk+1 − τk ) ,
where aI is the mean value of aI (τr ) in the interval. Using Eqs. (A9.7) and (9.188)
we have
δk3
Ek ≈ a2 (A − aI 1 ) + 2aI (A − aI 1 )2 + (A − aI 1 )3 , (A9.9)
12 I
with
δk = τk+1 − τk .
Finally, it should be remarked that, for the particular case of the Milne-Eddington
atmosphere, the DELO method provides the exact solution to the transfer equa-
tion for any outward propagation direction. This fact, which may look surprising at
first sight, can be easily understood by the following argument. The basic assump-
tions of the Milne-Eddington model imply that the matrix A and the vector b
in Eq. (9.187) are independent of τr and linear in τr , respectively. On the other
hand, the solution I(τr ) given by Eqs. (9.108) is also linear in τr , thus the ‘modified
source vector’ s(τr ) defined in Eq. (9.190) is in turn linear in τr . Since the quantity
aI is also constant, it turns out that s is linear in y (cf. Eq. (A9.1)). Therefore,
Eq. (A9.4) – which is the only assumption of the DELO method – is exactly sat-
isfied under the Milne-Eddington approximations. It follows, incidentally, that it
doesn’t make sense to test the accuracy of the DELO method by comparing the
numerical result with the Milne-Eddington analytical solution.
A10. Equivalent Width in the Presence of Depth-Dependent
Line Shifts
Consider the transfer equation for the intensity of a non-magnetic spectral line
formed in a static atmosphere in LTE. From Eqs. (9.29), (9.27), (9.32) and (9.35)
=0
we have, under the limit B
dI
= (1 + k) (I − BP ) , (A10.1)
dτc
where
1
k = κL η = κL √ H(v, a) , (A10.2)
π
with v the reduced wavelength
λ − λ0
v= . (A10.3)
∆λD
APPENDIX 821
Following Sánchez Almeida et al. (1989), we assume that the absorption profile η
is shifted from the wavelength λ0 by a small amount δλ which depends on optical
depth, where small means
δλ(τc ) ∆λD (A10.4)
for any τc . We want to evaluate the effect of this wavelength shift on the equivalent
width of the line.
The variable v in Eq. (A10.2) becomes
λ − λ0 − δλ(τc )
v(τc ) = ,
∆λD
and owing to assumption (A10.4) we can expand k in power series of δλ(τc ) . Up
to the second order we have
1 2
k = κL η − η δλ(τc ) + η δλ(τc ) , (A10.5)
2
where η is the unperturbed profile defined in Eq. (A10.2) and where primes denote
derivatives with respect to λ.
Similarly to Sect. 9.6, we write I in the form
I = I0 + I1 + I2 , (A10.6)
n
where In is of order δλ . Substituting Eqs. (A10.5) and (A10.6) into Eq. (A10.1)
and equating the terms of the same order, we obtain the following set of equations
dI0
= (1 + κL η) (I0 − BP )
dτc
dI1
= (1 + κL η) I1 − κL η δλ (I0 − BP )
dτc
dI2 1
= (1 + κL η) I2 − κL η δλ I1 + κ η δλ2 (I0 − BP ) .
dτc 2 L
To solve these equations we make two further assumptions: i) the atmosphere is
described by the Milne-Eddington model (see Sect. 9.8), and we consider vertical
propagation (µ = 1, tc = τc ); ii) the wavelength shift δλ varies linearly with
optical depth,
δλ(τc ) = b τc . (A10.7)
A straightforward calculation shows that the contributions to the emerging inten-
sity are
β
I0 (0) = B0 1 +
1 + κL η
κL η
I1 (0) = B0 β b
(1 + κL η)3
κL η κ2L η 2
I2 (0) = − B0 β b2 − 3 .
(1 + κL η)4 (1 + κL η)5
822 APPENDIX
Using these expressions, we can calculate the equivalent width according to Eq.
(9.219). We have
∞
1
W = W0 − I1 (0) + I2 (0) dλ ,
B0 (1 + β)
−∞
where W0 , the equivalent width corresponding to δλ = 0, is given by
β
W0 = ∆λD I (κ , a) ,
1+β 1 L
with I1 defined in Eqs. (9.222). Since the profile η is symmetrical about λ0 , the
contribution of I1 (0) to the equivalent width is zero. The contribution of I2 (0) can
be simplified by taking into account that
∞ ∞
κL η κ2L η 2
dλ = 4 dλ .
(1 + κL η)4 (1 + κL η)5
−∞ −∞
The equivalent width can thus be written in the form
2
β b
W = ∆λD I1 (κL , a) + I6 (κL , a) , (A10.8)
1+β ∆λD
where
∞
κ2L η̇ 2
I6 (κL , a) = dv , (A10.9)
(1 + κL η)5
−∞
with v the reduced wavelength defined in Eq. (A10.3) and η̇ = ∂η / ∂v.
A11. Net Circular Polarization in Blends
We consider the simplest case where the blend is formed by two independent,
Zeeman-triplet spectral lines. We assume the atmosphere is plane-parallel and
static and the magnetic field is in the vertical (outward or inward) direction, and
we restrict attention to the radiation flowing in the vertical outward direction.
Use of Eqs. (9.216) and (9.32) allows us to write the net circular polarization
parameter – defined in Eq. (9.235) – in the form (cf. Eq. (9.245))
Wb − Wr
v = σ ,
Wb + Wr
where σ is the sign factor defined in Eq. (9.246) and where Wb and Wr are the
equivalent widths of two ‘fictitious’ lines formed in the same atmosphere (without
APPENDIX 823
magnetic field) and having absorption coefficients given by the blend of the σb and
σr components of the two lines respectively, that is by
κ(1) (1) (2) (2)
L ηb + κL ηb , κ(1) (1) (2)
L ηr + κL ηr
(2)
.
Denoting by g1 and g2 the Landé factors of the two lines and by λ1 and λ2 their
central wavelengths, the separation between the σb components is
| λ2 − λ1 − (g2 − g1 ) ∆λB | ,
while the separation between the σr components is
| λ2 − λ1 + (g2 − g1 ) ∆λB | .
We now observe that in ‘standard’ stellar atmospheres (where temperature is a
monotonically increasing function of optical depth) the equivalent width of a spec-
tral line formed by two components increases when the separation between the
components is increased (cf. Sect. 9.21, Case ii ). It follows that
⎧
⎪ W > Wb if | λ2 − λ1 + (g2 − g1 ) ∆λB | > | λ2 − λ1 − (g2 − g1 ) ∆λB |
⎪ r
⎨
Wr = Wb if | λ2 − λ1 + (g2 − g1 ) ∆λB | = | λ2 − λ1 − (g2 − g1 ) ∆λB |
⎪
⎪
⎩
Wr < Wb if | λ2 − λ1 + (g2 − g1 ) ∆λB | < | λ2 − λ1 − (g2 − g1 ) ∆λB | .
The sign rule for v is therefore
sign (v ) = sign σ (λ2 − λ1 ) (g1 − g2 ) .
Obviously the net circular polarization is zero if g1 = g2 . Bearing in mind
Eq. (9.80), we can express the above relation in the following form: in a blend
of two Zeeman triplets, the sign of the net circular polarization is the same as the
sign of the V Stokes parameter in the external wing of the line having the larger
Landé factor.1
It might be tempting to generalize this statement to more complicated physical
situations (different orientations of the magnetic field, anomalous Zeeman patterns,
etc.). Detailed numerical calculations show, however, that there are many excep-
tions to this simple rule. The exceptions occur mainly for large inclination angles
of the magnetic field, for line pairs having close values of g, and for Type I and
Type II anomalous Zeeman patterns.
One could wonder whether the simple mechanism described above (blends be-
tween magnetic lines) might produce a certain amount of net circular polarization
in a broad spectral interval containing many blends. Although the net circular po-
larization in a single blend can be rather conspicuous (it is easy to find values of v
1
‘External’ means here ‘at larger distance from the other line’. Note that the above statement
is valid provided the larger of the two Landé factors is positive.
824 APPENDIX
of the order or even larger than 0.2), there is indeed no reason why the blending
line having the larger Landé factor should preferably lie on the red rather than on
the blue side of the other line. A systematic effect could possibly exist for fine-
structure or hyperfine-structure multiplets, but such investigation is beyond the
aims of this book.
A12. Evolution Operator in Stochastic Media
We solve Eq. (9.285) considering the matrix A(τr ) as a Kubo-Anderson process; in
other words, we suppose that its stochastic properties are described by assumptions
b), c), and d) of Sect. 9.24.
The evolution operator is defined in Eq. (9.93),
I(τ0 ) = O(τ0 , τ ) I(τ ) (τ0 ≤ τ ) .
First of all we evaluate its expression for a particular realization of the jumping
points (τ1 , τ2 , . . . , τi , . . . ). If the number of such points in the interval (τ0 , τ ) is n,
we have from Eqs. (9.95) and (9.103)
−(τ1 − τ0 ) A1 −(τ2 − τ1 ) A2 −(τn − τn−1 ) An
O(τ0 , τ ) = e e ··· e
−(τ − τn ) An+1
×e , (A12.1)
where A1 is the (constant) value of A(τr ) in the interval (τ0 , τ1 ), and so on.
Next we integrate over all possible realizations of the jumping points. The com-
bined probability dQn that the jumping points in (τ0 , τ ) are n and that they lie in
the intervals (τ1 , τ1 + dτ1 ), (τ2 , τ2 + dτ2 ), . . . , (τn , τn + dτn ), respectively, is given
by
τ1 − τ0 τ2 − τ1 τn − τn−1 τ − τn
− − − − dτ1 dτ2 dτ
dQn = e τe
e τe
··· e τe
e τe
··· n , (A12.2)
τe τe τe
with
∞
∞ τ τ
dτ1 dτ2
dQn = ···
n=0 n=0
τe τe
τ0 τ1
τ τn − τn−1
dτn − τ1 − τ0 τ −τ
− 2τ 1 − −
τ − τn
··· e τe
e e ··· e τe
e τe
= 1.
τe
τn−1
From Eqs. (A12.1) and (A12.2) we obtain
APPENDIX 825
∞
τ − τ0 ∞ τ τ
− dτ1 dτ2
O(τ0 , τ ) dQn = e τe
···
n=0 n=0
τe τe
τ0 τ1
τ
dτn −(τ1 − τ0 ) A1 −(τ2 − τ1 ) A2 −(τ − τ )A −(τ − τn ) An+1
··· e e · · · e n n−1 n e .
τe
τn−1
To perform this integral we expand the exponentials into power series. Using the
formula1
τ τ τ
dτ1 dτ2 dτn
··· (τ1 − τ0 )i1 (τ2 − τ1 )i2 · · · (τn − τn−1 )in (τ − τn )in+1 =
τe τe τe
τ0 τ1 τn−1
n
τ − τ0 i1 ! i2 ! · · · in ! in+1 !
= (τ − τ0 )i1 + i2 + ··· + in + in+1 ,
τe (i1 + i2 + · · · + in + in+1 + n) !
we have
∞
τ − τ0 ∞ n ∞ ∞
− τ − τ0
O(τ0 , τ ) dQn = e τe
···
n=0 n=0
τe i =0 i =0
1 2
∞
[ − (τ − τ0 ) ]i1 + i2 + ··· + in+1 i i in+1
··· A11 A22 · · · An+1 .
in+1 =0
(i1 + i2 + · · · + in+1 + n) !
Finally, we average over the distribution of the physical parameters and obtain
the following expression for the statistical average of the evolution operator
τ − τ0 ∞ n ∞ ∞
− τ − τ0
O(τ0 , τ ) av
= e τe
···
n=0
τe i =0 i =0 1 2
∞
[ − (τ − τ0 ) ]i1 + i2 + ··· + in+1 i i
··· A 1 A 2 · · · Ain+1 .
in+1 =0
(i1 + i2 + · · · + in+1 + n) !
This equation can be rewritten in a more convenient form by grouping together all
the terms of the form
j j
A 1 A 2 · · · Ajm
1
See e.g. Gradshteyn and Ryzhik (1965), equation n. 4.634, p.620.
826 APPENDIX
with j1 ≥ 1, j2 ≥ 1, . . . , jm ≥ 1 . With the help of some algebra, we obtain the
expression
O(τ0 , τ ) av =
∞
k
τ − τ0
=1 + (−1)k τek fkm Aj1 Aj2 · · · Ajm , (A12.3)
m=1
τe j1 j2 ··· jm
k=1
where (j1 , j2 , . . . , jm ) are all the different solutions of the equation
j1 + j2 + · · · + jm = k , (A12.4)
and where the function fkm (x), defined only for k ≥ 1 and 1 ≤ m ≤ k, is given by
∞
−x k xn n+1
fkm (x) = e x . (A12.5)
n=m−1
(n + k) ! m
The function fkm (x) can be expressed in a variety of different forms. A tedious
calculation, which is left as an exercise to the reader, allows one to transform the
infinite sum into two finite sums. The result is the following1
f11 (x) = x
m
xn k+m−2−n
fkm (x) = (−1)m (−1)n
n=0
n! k−2
k−2 !
−x xn k+m−2−n
−e (k ≥ 2) . (A12.6)
n=0
n! m
The solution to the homogeneous transfer equation (9.285) for a Kubo-Anderson
process can thus be written in the form
I(τ0 ) av = O(τ0 , τ ) av I(τ ) ,
with [ O(τ0 , τ ) ]av given by Eqs. (A12.3) and (A12.6).
Equation (A12.3) contains, as particular cases, some interesting limits:
i) If the physical parameters are deterministic, the symbols · · · can be removed.
k−1
Since for a given k the number of different solutions of Eq. (A12.4) is m−1 , we
obtain
∞
k
k−1 τ − τ0
O(τ0 , τ ) av = 1 + (−1)k τek Ak fkm .
m=1
m−1 τe
k=1
1
To obtain Eqs. (A12.6), use has been made of Eq. 45 on p.166 and Eq. 4 on p.705 in
Prudnikov et al. (1986), and of Eq. 4 on p.8 and Eq. (ii) on p.1 in Riordan (1968). Equation 3
in the latter book is useful to prove Eq. (A12.7) using expression (A12.5).
APPENDIX 827
The sum over m can be evaluated and yields
k
1 τ − τ0
, (A12.7)
k! τe
so that
∞
(τ − τ0 )k k −(τ − τ0 ) A
O(τ0 , τ ) av
=1 + (−1)k A = e ,
k!
k=1
which is the correct expression for the evolution operator in the deterministic case.
ii) Microturbulent limit (τe → 0). According to Eqs. (A12.6), the asymptotic
expression of the function fkm (x) for x → ∞ is
xm
fkm (x) ∼ .
m!
Substitution into Eq. (A12.3) shows that, for an assigned k, the only term in the
summation over m which makes a non-zero contribution for τe → 0 is the one
corresponding to m = k. This implies j1 = j2 = · · · = jm = 1, hence Eq. (A12.3)
reduces to
∞
(τ − τ0 )k k −(τ − τ0 ) A
O(τ0 , τ ) av
=1 + (−1)k A = e . (A12.8)
k!
k=1
iii) Macroturbulent limit (τe → ∞). According to Eq. (A12.5), the asymptotic
expression of fkm (x) for x → 0 is
xk+m−1
fkm (x) ∼ .
(k + m − 1) !
For an assigned k, the only term in the summation over m in Eq. (A12.3) which
makes a non-zero contribution for τe → ∞ is the one with m = 1. This implies
j1 = k, so that Eq. (A12.3) yields
∞
(τ − τ0 )k k
O(τ0 , τ ) av
=1 + (−1)k A
k!
k=1
8 ∞
9 1 −(τ − τ ) A 2
k (τ − τ0 )
k
= 1 + (−1) Ak
= e 0
.
k!
k=1
iv) Scalar case. For non-magnetic media, A has the form A = κ 1 (with κ given
by Eq. (9.278)), thus all the matrices Aji in Eq. (A12.3) commute. Equation
828 APPENDIX
(A12.3) can be rewritten in the form
∞
k
τ − τ0
O(τ0 , τ ) av
= 1+ (−1)k τek fkm
m=1
τe
k=1
!
× N (p1 , p2 , . . . , pm ) κ p1 κ p2 · · · κ pm 1 ,
p1 p2 ··· pm
where
p1 + p2 + · · · + pm = k , 1 ≤ p1 ≤ p2 ≤ · · · ≤ pm , (A12.9)
and where N (p1 , p2 , . . . , pm ) is an integer defined in the following way. For each
set (p1 , p2 , . . . , pm ) satisfying Eqs. (A12.9) we denote by the number of distinct
pi values contained in the set, and we divide the set into groups by collecting
the pi ’s with the same value into one group. If the number of elements in the i-th
group is ni , then
m!
N (p1 , p2 , . . . , pm ) = .
n1 ! n2 ! · · · n !
As an example, let us consider the case k = 6, m = 3. The possible sets (p1 , p2 , p3 )
are (1, 1, 4), (1, 2, 3), and (2, 2, 2). The corresponding values of N (p1 , p2 , p3 ) are 3,
6, and 1, respectively.1
v) Explicit expression up to k = 3 terms. By evaluating the functions fkm (x) one
obtains
x2 −x 2
O(τ0 , τ ) av
= 1 − (τ − τ0 ) A + τe2 1 − x + −e A
2
−x 2
− τe2 1 − x − e A
x3 −x 3
+ τe3 4 − 3x + x2 − − (4 + x) e A
6
x2 −x 2 2
− τe3 3 − 2x + − (3 + x) e A A + A A
2
−x
+ τe3 2 − x − (2 + x) e A3 ,
where x = (τ − τ0 ) / τe .
1 % 5
Note that N (p1 , p2 , p3 ) = 10 = 2
. In general,
% k−1
N (p1 , p2 , · · · , pm ) = m−1
.
APPENDIX 829
A13. Properties of the Generalized Profiles
a) Definition
The generalized profile is defined by Eq. (10.40)
ΦKK
Q (J , Ju ; ν) = 3(2Ju + 1)(2K + 1)(2K + 1)
Ju J 1 Ju J 1
× (−1)1+Ju −Mu + q
−M u M −q −M u M −q
Mu Mu M qq
Ju Ju K 1 1 K
×
Mu −Mu −Q q −q −Q
1
× Φ(να J M , α J M − ν) + Φ(να J M , α J M − ν)∗ , (A13.1)
2 u u u u u u
where Φ(ν0 − ν) is given by Eq. (6.59a). In general, it is a complex quantity.
b) Limitation on indices
Owing to the presence of the 3-j symbols, the indices are restricted to the range
K = 0, 1, . . . , 2Ju
K = 0, 1, 2
Q = 0, ±1, ±2 with |Q| ≤ K , |Q| ≤ K .
Denoting by NQ the number of different generalized profiles with given Q, we have
N0 = 3(2Ju + 1) 0 if Ju = 0
N2 = N−2 =
N1 = N−1 = 4Ju , 2Ju − 1 0,
if Ju =
so that the total number of generalized profiles is1
2
3 if Ju = 0
N= NQ =
Q=−2
18Ju + 1 if Ju = 0 .
c) Conjugation
Taking the complex conjugate of Eq. (A13.1) and interchanging the summation
indices Mu ←→ Mu and q ←
→ q , one can easily prove, using the symmetry properties
of the 3-j symbols (Eqs. (2.24), (2.25)), that
ΦKK
Q (J , Ju ; ν)∗ = ΦKK
−Q (J , Ju ; ν) . (A13.2)
1
Because of the conjugation property proved below (Eq. (A13.2)), N represents also the
number of independent real quantities needed to fully specify all the complex generalized profiles.
830 APPENDIX
d) Integration over frequency
∞
ΦKK
Q (J , Ju ; ν) dν = δKK wJ(K)J , (A13.3)
u
−∞
where wJ(K)J is the symbol defined in Eq. (10.11). This result can easily be proved
u
with the help of Eqs. (6.59a-c), (2.42) and (2.23a).
e) Inversion about line center
Denoting by ν, ν two frequencies symmetrical about the central frequency ν0 of
the line,
ν − ν0 = ν0 − ν ,
we have
ΦKK
Q (J , Ju ; ν ) = (−1)K+K ΦKK
Q (J , Ju ; ν) ; (A13.4)
in other words, the generalized profile is symmetrical or antisymmetrical about line
center according as the integer (K + K ) is even or odd, respectively. To prove this
statement, we recall that from Eq. (7.3)
να J Mu , α J M = ν0 + νL (gα J Mu − gα J M ) ,
u u u u
and that the profile Φ(ν0 − ν) satisfies the conjugation property (cf. Eq. (6.59a))
Φ(x) = Φ(−x)∗ ,
thus
Φ(να J Mu , α J M − ν ) = Φ(−ν0 + νL (gα J Mu − gα J M ) + ν)
u u u u
= Φ(ν0 + νL [ gα J (−Mu ) − gα J (−M )] − ν)∗ .
u u
After some index renaming, Eq. (A13.4) is easily obtained from Eq. (A13.1) with
the help of Eqs. (2.24) and (2.25).
f) Limit for zero magnetic field
Under this limit all the Zeeman splittings vanish, so that
1
lim Φ(να J M , α J M − ν) + Φ(να J M , α J M − ν)∗ = φ(ν0 − ν) ,
νL →0 2 u u u u u u
φ being the real part of the complex profile Φ (see Eq. (6.59a)). Substituting into
Eq. (A13.1) and performing the same calculations as in point d), one obtains
lim ΦKK
Q (J , Ju ; ν) = δKK wJ(K)J φ(ν0 − ν) . (A13.5)
νL →0 u
APPENDIX 831
g) Expansion for weak field
If the magnetic field is weak, the generalized profile can be expanded into power
series of νL . To first order we can write
Φ(να J Mu , α J M − ν) ≈ Φ(ν0 − ν)
u u
∂ Φ(να J Mu , α J M − ν)
+ νL (gα J Mu − gα J M ) u u
u u ∂νL νL =0
∂ Φ(ν0 − ν)
= Φ(ν0 − ν) − νL (gα J Mu − gα J. M )
u u ∂ν
After substitution into Eq. (A13.1), we are left with a zero-order and a first-order
contribution. The former has already been calculated (Eq. (A13.5)); the latter
contains summations of the form
Ju J 1 Ju J 1
M (−1)1+Ju −Mu + q
−Mu M −q −Mu M −q
Mu Mu M qq
Ju Ju K 1 1 K
× ,
Mu −Mu −Q q −q −Q
where M is either Mu , or Mu , or M . These can be performed by writing M
in terms of a suitable 3-j symbol (as done, for instance, in Eqs. (3.42)). The
summation involving Mu or Mu can be evaluated by applying twice Eq. (2.42),
while the summation involving M can be evaluated via Eq. (2.52). The result is
the following
ΦKK
Q (J , Ju ; ν) = δKK wJ(K)J φ(ν0 − ν)
u
K K 1
−νL 3(2Ju + 1)(2K + 1)(2K + 1)
Q −Q 0
⎡
! !
K K 1 K 1 1
× ⎣ gα J Ju (Ju + 1)(2Ju + 1) (−1)1+J −Ju + Q
u u Ju Ju Ju J Ju Ju
∗
1 ∂ Φ(ν0 − ν) ∂ Φ(ν − ν)
× + (−1)1+K+K 0
2 ∂ν ∂ν
⎧ ⎫ ⎤
⎨ J Ju 1 ⎬ ∂φ(ν − ν)
+ gα J J (J + 1)(2J + 1) (−1)Q J Ju 1 0 ⎦. (A13.6)
⎩ ⎭ ∂ν
1 K K
h) Special case K = 0
For K = 0 one gets, with the help of Eq. (2.26a)
2
Ju J 1 1 1 K
Φ0K (J , J ; ν) = 3(2K + 1) (−1) 1+ q
0 u −Mu M −q q −q 0
Mu M q
× φ(να J Mu , α J M − ν) . (A13.7)
u u
832 APPENDIX
In particular, for K = 0
J J 1
2
Φ00
0 (J , Ju ; ν) = u φ(να − ν) .
−Mu M q J Mu , α J M
u u
Mu M q
i) Expression of the emission coefficient in terms of the generalized profile
Consider the expression of the emission coefficient (Eq. (7.15e)), written for the
two-level atom. Owing to the presence of the 3-j symbols, the indices Q and Qu
coincide. The last term can therefore be rewritten in the form
Re T Q
K ρKu (α J ) Φ(ν
(i, Ω) − ν) =
Q u u α u u
J Mu , α J M
1 K Ku
= T (i, Ω) ρQ (αu Ju ) Φ(να J M , α J M − ν)
2 Q u u u
K
+ T−Q
K ρ u (α J ) Φ(ν
(i, Ω) −Q u u α J Mu , α J M − ν)∗ ,
u u
where Eqs. (3.102) and (5.158) have been used. Bearing in mind the properties
of the 3-j symbols (Eqs. (2.24), (2.25)) and the relation between the Einstein
coefficients (Eqs. (7.8)), one easily obtains after some index renaming
hν
=
εi (ν, Ω) N 2Ju + 1 A(αu Ju → α J )
4π
K ρK (α J ) ΦKK (J , J ; ν) .
× TQ (i, Ω) Q u u Q u (A13.8)
KK Q
j) Expression of the absorption coefficient in terms of the generalized profile
We restrict attention to the two-level atom with unpolarized lower level. From
Eq. (7.15a) we obtain, with the help of Eqs. (10.7), (10.6), (2.26a) and (5.158)
= hν
ηiA (ν, Ω) N B(α J → αu Ju )
4π
2
Ju J 1 1 1 K
× 3(2K + 1) (−1)1+ q
−Mu M −q q −q 0
K Mu M q
φ(ν
× T 0K (i, Ω) − ν) ,
α u u
J Mu , α J M
or, via Eq. (A13.7)
hν
=
ηiA (ν, Ω) N B(α J → αu Ju ) Φ0K (J , J ; ν) ,
T 0K (i, Ω) (A13.9)
0 u
4π
K
where N is the overall population of the lower level.
APPENDIX 833
k) Special case J = 0, Ju = 1
Evaluating the relevant 3-j symbols via Eqs. (2.26), and introducing the compact
notation (consistent with Eqs. (9.6) and (9.8))
Φq = φq + i ψq = Φ(να 1 −q, α 00 − ν) (q = −1, 0, +1) , (A13.10)
u
one obtains from Eq. (A13.1)
1
Φ00
0 = φ + φ0 + φ−1
3 1
1
0 = −Φ0 = √
Φ01 φ1 − φ−1
10
6
1
0 = Φ0 = √
Φ02 φ1 − 2φ0 + φ−1
20
3 2
1
0 = −
Φ11 φ + φ−1
2 1
1
1 = −
Φ11 φ + i ψ1 + 2φ0 + φ−1 − i ψ−1
4 1
1
0 = −Φ0 = − √
Φ12 φ1 − φ−1
21
2 3
1
1 = −Φ1 = −
Φ12 φ + i ψ1 − 2i ψ0 − φ−1 + i ψ−1
21
4 1
1
Φ22
0 = φ + 4φ0 + φ−1
6 1
1
Φ22
1 = φ + i ψ1 + 2φ0 + φ−1 − i ψ−1
4 1
1
Φ22
2 = φ + i ψ1 + φ−1 − i ψ−1 . (A13.11)
2 1
The generalized profiles with negative Q are obtained from the expressions above
via the conjugation property (A13.2).
As shown above – points i) and j) – the generalized profile ΦKK
Q (J , Ju ; ν) is useful
to express the frequency dependence of the emission and absorption coefficients for
the two-level atom in the presence of a magnetic field. Its direct counterpart, related
to the anomalous dispersion coefficients ρAi (ν, Ω) and ρS (ν, Ω),
is the generalized
i
KK
dispersion profile ΨQ (J , Ju ; ν), which is defined by an expression strictly similar
to Eq. (A13.1)
KK
i ΨQ (J , Ju ; ν) = 3(2Ju + 1)(2K + 1)(2K + 1)
Ju J 1 Ju J 1
× (−1)1+Ju −Mu + q
−M u M −q −M u M −q
Mu Mu M qq
Ju Ju K 1 1 K
×
Mu −Mu −Q q −q −Q
1
× Φ(να J M , α J M − ν) − Φ(να J M , α J M − ν)∗ .
2 u u u u u u
834 APPENDIX
The expressions of ΦKK
Q
KK
and of i ΨQ are the same except for the opposite sign in
the last square bracket. Because of this close resemblance, the following properties
of the generalized dispersion profiles can be proved by strictly similar arguments.
b ) Limitation on indices
Same as b).
c ) Conjugation
KK
ΨQ (J , Ju ; ν)∗ = Ψ−Q
KK
(J , Ju ; ν) .
d ) Integration over frequency
∞
KK
ΨQ (J , Ju ; ν) dν = 0 .
−∞
e ) Inversion about line center
KK
ΨQ (J , Ju ; ν ) = (−1)1+K+K ΨQ
KK
(J , Ju ; ν) , (A13.12)
where the frequencies ν, ν are symmetrical about the central frequency of the line.
f ) Limit for zero magnetic field
KK
lim ΨQ (J , Ju ; ν) = δKK wJ(K)J ψ(ν0 − ν) , (A13.13)
νL →0 u
where ψ is the imaginary part of the profile Φ (see Eq. (6.59a)).
g ) Expansion for weak field
KK
ΨQ (J , Ju ; ν) = δKK wJ(K)J ψ(ν0 − ν)
u
K K 1
− νL 3(2Ju + 1)(2K + 1)(2K + 1)
Q −Q 0
⎡
! !
K K 1 K 1 1
× ⎣ gα J Ju (Ju + 1)(2Ju + 1) (−1) 1+J −Ju + Q
u u Ju Ju Ju J Ju Ju
∗
−i ∂ Φ(ν0 − ν) ∂ Φ(ν − ν)
× + (−1)K+K 0
2 ∂ν ∂ν
⎧ ⎫ ⎤
⎨ J Ju 1 ⎬
∂ψ(ν0 − ν) ⎦
+ gα J J (J + 1)(2J + 1) (−1)Q J Ju 1 .
⎩ ⎭ ∂ν
1 K K
APPENDIX 835
h ) Special case K = 0
2
Ju J 1 1 1 K
Ψ00K (J , Ju ; ν) = 3(2K + 1) (−1) 1+ q
−Mu M −q q −q 0
Mu M q
× ψ(να J Mu , α J M − ν) ,
u u
and for K = 0
J J 1
2
Ψ000 (J , Ju ; ν) = u ψ(να − ν) .
−Mu M q u u
J Mu , α J M
Mu M q
i ) Expression of the anomalous dispersion coefficient ρSi (ν, Ω)
For the two-level atom, Eq. (7.15d) can be written in the form
= hν N 2J + 1 B(α J → α J )
ρSi (ν, Ω) u u u
4π
× TQ
K ρK (α J ) Ψ KK (J , J ; ν) .
(i, Ω) Q u u Q u
KK Q
j ) Expression of the anomalous dispersion coefficient ρAi (ν, Ω)
For the two-level atom with unpolarized lower level, Eq. (7.15c) can be written in
the form
= hν N B(α J → α J )
ρAi (ν, Ω) Ψ 0K (J , J ; ν) .
T 0K (i, Ω) (A13.14)
u u 0 u
4π
K
A14. Properties of the Symbol WKK Q (β L Sβu Lu ; B) fs
a) Definition
The symbol, simply denoted by WKK Q in this appendix, is defined by Eq. (10.153)
! !
3(2Lu + 1) 1 1 K 1 1 K
WKK Q =
2S + 1 Lu Lu L Lu Lu L
× (2K + 1)(2K + 1)(2Ju + 1)(2Ju + 1)(2Ju + 1)(2Ju + 1)
Ju Ju Ju Ju Mu Mu
! !
Lu Lu K Lu Lu K Ju Ju K Ju Ju K
×
Ju Ju S Ju Ju S −Mu Mu −Q −Mu Mu −Q
j j j j 1
× CJu (Mu ) CJu (Mu ) CJu (Mu ) CJu (Mu ) , (A14.1)
ju ju
u u u u 1 + 2πi ν(ju Mu , ju Mu ) A
836 APPENDIX
where we have used shorthand notations for the Bohr frequencies νβLS (jM, j M )
and the coefficients CJj (βLS, M ) defined in Eqs. (7.30) and (3.58), respectively, and
for the Einstein coefficient A(βu Lu S → β L S).
b) Limitation on indices
Owing to the presence of the 6-j and 3-j symbols, the indices are confined to the
range
0 if Lu = 0
K, K =
0, 1, 2 if Lu = 0
Q = 0, ±1, ±2 with |Q| ≤ K, |Q| ≤ K . (A14.2)
c) Symmetry
Exchange of the dummy indices Ju →
← Ju , Ju →
← Ju shows that
WKK Q = WK KQ . (A14.3)
d) Conjugation
By exchanging the dummy indices ju → → Mu , Ju ←
← ju , Mu ← → Ju , Ju →
← Ju , and
j
bearing in mind the reality of the CJ coefficients (Sect. 3.4), it can easily be proved,
with the help of Eqs. (2.24) and (2.25), that
∗
WKK Q = WKK −Q . (A14.4)
e) Special cases
Use of Eqs. (2.26a), (2.36a), (3.62a,b) and (2.23a) shows that
WK00 = W0K0 = δK0 . (A14.5)
From Eqs. (A14.2)-(A14.5) it follows that there are only 8 independent symbols,
namely W000 (= 1), W110 , W210 , W220 , W111 , W211 , W221 , W222 . The first four
are real, the remaining are in general complex.
f) Limit for zero magnetic field
Under this limit we have
j
CJu (Mu ) → δj J , ν(ju Mu , ju Mu ) → νβ L SJu , βu Lu SJu ,
u u u u u
the last quantity being defined in Eq. (7.43). Performing the summation over Mu
and Mu via Eq. (2.23a), one finds
lim WKK Q = δKK WK (β L Sβu Lu ) fs , (A14.6)
B→0
where the symbol [WK (β L Sβu Lu )]fs is defined in Eq. (10.133).
APPENDIX 837
g) Limit for strong magnetic field
Let us assume that the field modulus B satisfies the inequalities
µ0 B
µ0 B ζ , A, (A14.7)
h
or, equivalently
γ1, γx1,
where γ and x are defined in Eqs. (10.154). It can be proved that under this limit
WKK Q = δKK δQ0 WK (L , Lu ) , (A14.8)
where the symbol WK (L , Lu ) is defined in Eq. (10.17). This property is at the
basis of the anti-level-crossing effect described in Sect. 10.18.
The proof of Eq. (A14.8) is rather complicated. First of all, we notice that
the first of inequalities (A14.7) implies that the upper term is in the complete
Paschen-Back effect regime. It follows (see Sect. 3.4) that the energy eigenvectors
j
of the upper term are of the form |βu Lu SML MS , hence the coefficients CJu (Mu )
u u
defined in Eq. (3.58) reduce to Clebsh-Gordan coefficients. The magnetic quantum
numbers are related by ML + MS = Mu , thus the index ju can be identified with
u
either ML or MS . Setting ju = MS we can write, with the use of Eq. (2.22)
u
j M
CJu (Mu ) ≡ CJ S (Mu ) = Lu SJu Mu |Lu SML MS =
u u u
Lu−S+Mu
Lu S Ju
= (−1) 2Ju + 1 .
ML MS −Mu
u
Substituting into Eq. (A14.1), and performing the summations over Ju and Ju via
Eq. (2.43), next the summations over Ju and Ju via Eq. (2.23b), one obtains
! !
3(2Lu + 1) 1 1 K 1 1 K
WKK Q = (2K + 1)(2K + 1)
2S + 1 Lu Lu L Lu Lu L
K Lu Lu K Lu Lu
× (−1)K−K −Q −ML ML −Q −ML ML
Mu Mu MS u u u u
1
× , (A14.9)
1 + 2πi ν(MS Mu , MS Mu ) A
where ML = Mu − MS , ML = Mu − MS .
u u
On the other hand, in the complete Paschen-Back effect regime the Bohr fre-
quency ν(MS Mu , MS Mu ) can be written in the form (see Sect. 3.4)
µ0 B
ν(MS Mu , MS Mu ) = (Mu − Mu ) + (Mu , Mu , MS ) ,
h
838 APPENDIX
where is a correction of order ζ/h due to fine structure. Taking into account
inequalities (A14.7), it follows that
1
= δM M ,
1 + 2πi ν(MS Mu , MS Mu ) A u u
which implies ML = ML and Q = 0. The triple summation in Eq. (A14.9) reduces
u u
to a double summation over Mu and MS , or equivalently over ML and MS . The
u
former can be performed via Eq. (2.23a), the latter yields a factor (2S+1). Equation
(A14.8) is thus proved.
A15. A Property of the Hopf Function
The Hopf function obeys the integral equation (cf. Eq. (12.24))
∞
1 1
q(τ ) = E3 (τ ) + E1 |t − τ | q(t) dt , (A15.1)
2 2
0
and satisfies the property (cf. Eq. (12.25))
∞
1
E2 (τ ) q(τ ) dτ = . (A15.2)
3
0
Equation (A15.1) can be rewritten in the more explicit form
τ ∞
1 1 1
q(τ ) = E3 (τ ) + E1 (τ − t) q(t) dt + E1 (t − τ ) q(t) dt ,
2 2 2
0 τ
or, performing the substitutions (τ − t) = u and (t − τ ) = v in the first and second
integral, respectively
τ ∞
1 1 1
q(τ ) = E3 (τ ) + E1 (u) q(τ − u) du + E1 (v) q(τ + v) dv .
2 2 2
0 0
Let us consider the derivative of q(τ ) with respect to τ . Recalling that
d
E (τ ) = −En−1 (τ ) ,
dτ n
and denoting derivatives by apices, we have
τ
1 1 1
q (τ ) = − E2 (τ ) + q(0) E1 (τ ) + E1 (u) q (τ − u) du
2 2 2
0
∞
1
+ E1 (v) q (τ + v) dv ,
2
0
APPENDIX 839
or, by means of the inverse transformations to those outlined above
∞
1 1 1
q (τ ) = − E2 (τ ) + q(0) E1 (τ ) + E1 |t − τ | q (t) dt . (A15.3)
2 2 2
0
We now multiply both sides by q(τ ) and integrate in dτ between 0 and ∞ ,
∞
1 1 1
q(τ ) q (τ ) dτ = − I1 + q(0) I2 + I3 , (A15.4)
2 2 2
0
where
∞
I1 = E2 (τ ) q(τ ) dτ
0
∞
I2 = E1 (τ ) q(τ ) dτ
0
∞ ∞
I3 = dτ q(τ ) E1 |t − τ | q (t) dt .
0 0
For the first integral we have from Eq. (A15.2)
1
I1 = . (A15.5)
3
For the second, evaluating Eq. (A15.1) at τ = 0 and taking into account that
E3 (0) = 1/2, we obtain
1
I2 = 2 q(0) − . (A15.6)
2
For the third integral, reversing the order of the integrations we have
∞ ∞ ∞
I3 = dt q (t) q(τ ) E1 |t − τ | dτ = q (t) 2q(t) − E3 (t) dt ,
0 0 0
where Eq. (A15.1) has been used. On the other hand, using again Eq. (A15.2) we
get
∞ ∞
1 1
q (t) E3 (t) dt = −q(0) E3 (0) + q(t) E2 (t) dt = − q(0) + ,
2 3
0 0
thus
∞
1 1
I3 = 2 q(t) q (t) dt + q(0) − . (A15.7)
2 3
0
840 APPENDIX
Substitution of Eqs. (A15.5), (A15.6), and (A15.7) into Eq. (A15.4) yields
∞ ∞
2 1
q(τ ) q (τ ) dτ = q(0) − + q(τ ) q (τ ) dτ ,
3
0 0
whence
2 1
q(0) = .
3
Excluding the negative root which is physically meaningless (leading to a negative
value for the surface source function, see Eqs. (12.20a) and (12.23)), one gets the
final result
1
q(0) = √ .
3
A16. A Numerical Algorithm for the Solution of the
Hopf Equation
Given the integral equation (see Eq. (12.24))
∞
1 1
q(τ ) = E3 (τ ) + E1 |t − τ | q(t) dt ,
2 2
0
we discretize it by selecting a grid of N τ -values τ1 , τ2 , . . . , τN , and we denote
by q1 , q2 , . . . , qN the unknown values taken by the function q(τ ) at the various
grid-points. We then calculate the integral appearing in the right-hand side by
dividing the integration domain into N +1 intervals (0, τ1 ), (τ1 , τ2 ), . . . , (τN −1 , τN ),
(τN , ∞), and by assuming the function q(τ ) to be linear in τ in each interval, and
to be constant (= qN ) in the last one. Taking into account the elementary formulae
for the integration of the exponential integrals
b
En (τ ) dτ = En+1 (a) − En+1 (b)
a
b
En (τ ) τ dτ = a En+1 (a) − b En+1 (b) + En+2 (a) − En+2 (b) ,
a
we obtain after some algebra the following set of equations
−1
N
qi+1 − qi
E3 |τi − τj | − E3 |τi+1 − τj | = q0 E2 (τj ) − E3 (τj ) ,
i=0
τi+1 − τi
APPENDIX 841
where j = 1, 2, . . . , N and where (see Eq. (12.26))
1
τ0 = 0 , q0 = √ .
3
This is a set of N linear equations in the N unknowns q1 , q2 , . . . , qN that can be
solved with standard numerical techniques – thus providing a numerical solution
for the Hopf function q(τ ).
The result presented in the text (Fig. 12.2) has been obtained with a grid of 1000
points equispaced in Log τ between τ = 10−4 and τ = 10.
A17. Symmetry Properties of the Comoving-Frame Radiation Field
Tensor for a Cylindrically Symmetrical Atmosphere
For a given star, and for a given point P in its outer atmosphere, the radiation field
tensor, defined in the comoving frame by Eq. (12.40), is a function of the velocity
v of the frame itself. In this appendix we make a slight change of notations and
K
explicitly attach such a functional dependence to the symbol JQ , that is now
written in the form
K
JQ (ν ; v, θv , χv ) ,
where θv and χv are the angles defining the direction of v according to the geometry
of Fig. 12.10.
If the radiation coming from the stellar surface is cylindrically symmetrical – in
the star’s frame – about the vertical axis through the point P, the radiation field
tensor in the comoving frame satisfies a number of properties that are illustrated
below.
a) The dependence on χv can be factored in the form
K iQχv K
JQ (ν ; v, θv , χv ) = e JQ (ν ; v, θv , 0) . (A17.1)
This relation can
be proved by considering
the χ-dependence of the two factors
TQK and I ν(1 + v · Ω/c),
(0, Ω) appearing in Eq. (12.40). From Table 5.6 it
Ω
follows that
TQ
K = f (θ) e iQχ ,
(0, Ω) (A17.2)
KQ
where fKQ is a real function satisfying
fK,−Q (θ) = (−1)Q fKQ (θ) . (A17.3)
Moreover, because of the assumed cylindrical symmetry of the radiation field in
the star’s frame, we can write (see Eq. (12.43))
, -
v · Ω
I ν 1+ = g ν, v, θ, θ , cos(χ − χ ) ,
,Ω (A17.4)
v v
c
842 APPENDIX
where g is another real function that doesn’t need to be specified. Substitution of
Eqs. (A17.2) and (A17.4) into Eq. (12.40) yields
γ
K 1
JQ (ν ; v, θv , χv ) = dθ sin θ fKQ (θ)
4π
0
2π
iQχ
× g ν, v, θ, θv , cos(χ − χv ) e dχ , (A17.5)
0
where the angle γ is defined in Eq. (12.32). Performing in the last integral the
substitution (χ − χv ) = χ , one easily obtains the property in Eq. (A17.1).
K
b) All the components of the tensor JQ (ν ; v, θv , 0) are real, and satisfy the relation
K
J−Q (ν ; v, θv , 0) = (−1)Q JQ
K
(ν ; v, θv , 0) . (A17.6)
K
In fact, Eq. (A17.5) shows that the imaginary components of JQ (ν ; v, θv , 0) are
zero, because both fKQ and g are real, and g is an even function of χ in the interval
K
(0, 2π) whereas sin(Qχ) is odd. The real components of JQ (ν ; v, θv , 0) have the
form
γ 2π
K 1
JQ (ν ; v, θv , 0) = dθ sin θ fKQ (θ) g(ν, v, θ, θv , cos χ) cos(Qχ) dχ ,
4π
0 0
which entails, because of Eq. (A17.3), the property in Eq. (A17.6).
c) If the spectrum of the radiation coming from the stellar surface is ‘locally’
symmetrical about a particular frequency ν0 in the star’s frame (as in the case of
an absorption or emission line), namely if
= I(ν + ∆ν, Ω)
I(ν0 − ∆ν, Ω) (A17.7)
0
then
for any direction Ω,
K
JQ (ν0 − ∆ν ; v, π − θv , π + χv ) = JQ
K
(ν0 + ∆ν ; v, θv , χv ) , (A17.8)
and, in particular
K
JQ (ν0 ; v, π − θv , π + χv ) = JQ
K
(ν0 ; v, θv , χv ) . (A17.9)
This property can easily be proved by observing that the transformation
θv → π − θv , χv → π + χv
APPENDIX 843
implies v → −v. From Eq. (12.40) one has
5 -
dΩ K , v · Ω
K
JQ (ν0 − ∆ν ; v, π − θv , π + χv ) = T Q (0, Ω) I ν0 − ∆ν 1 − .
,Ω
4π c
Neglecting the ‘second-order’ term ∆ν (v · Ω)/c and taking into account the as-
sumption in Eq. (A17.7), the integrand can be transformed via the substitution
- -
, v · Ω ,
I ν0 − ∆ν 1 − = I ν + ∆ν 1 + v · Ω , Ω
,Ω ,
0
c c
which shows, recalling again Eq. (12.40), the validity of Eq. (A17.8).
d) When the condition expressed by Eq. (A17.7) is satisfied, the comoving-frame
radiation field tensor obeys the further relation
K
JQ (ν0 ; v, π − θv , χv ) = JQ
K
(ν0 ; v, θv , χv − π) = (−1)Q JQ
K
(ν0 ; v, θv , χv ) ,
which follows from Eqs. (A17.9) and (A17.1).
A18. Redistribution Matrix for a Maxwellian Distribution
of Velocities
In Eq. (13.35) we suppose that the velocity distribution is Maxwellian, so that
f (v ) d3v = fc (vx ) fc (vy ) fc (vz ) dvx dvy dvz ,
where (xyz) is an arbitrary reference system and where
v 2
1 1 −
fc (v) = √ e vT
,
π vT
vT being the thermal velocity.
We now choose the reference system in such a way that the x-y plane coincides
with the plane containing the two unit vectors Ω and Ω , with the x-axis directed
. If Θ is the angle between the two unit vectors, we have
along Ω
= v
v · Ω x
= cos Θ v + sin Θ v .
v · Ω x y
Substituting into Eq. (13.35), the integrals over vz and vx can be easily performed.
Introducing the Doppler width in frequency units, ∆νD = ν0 vT /c (see Eqs. (5.43)),
one gets
844 APPENDIX
ν −ν 2
− 0
; ν, Ω
) = P (Ω, )
Ω 1 ∆νD
Rij (ν, Ω ij e
π vT ∆νD
∞ v 2 , -
− y ν
× dvy e vT δ ν0 − ν + (ν − ν0 ) cos Θ + 0 sin Θ vy ,
c
−∞
whence it immediately follows, for Θ = 0
; ν, Ω
Rij (ν, Ω ) =
ν −ν 2 ν−ν0 ν −ν0 2
− 0 − −
Ω ) 1 ∆νD ∆νD sin Θ ∆νD tan Θ
= Pij (Ω, 2 sin Θ e e , (A18.1)
π ∆νD
and for Θ = 0
ν −ν 2
− 0
; ν, Ω
Rij (ν, Ω ) = P (Ω, ) √1
Ω 1
e ∆νD
δ(ν − ν) . (A18.2)
ij
π ∆νD
This redistribution function is nothing but the generalization to polarized radiation
of the function RI introduced by Hummer (1962).
A19. Properties of the Kernel KQQ
K
(RB )
The kernel KQQ
K
(RB ) is defined in Eq. (14.3),
∗
KQQ
K
(RB ) = DQ
K
Q (RB ) Q DQ
K
Q (RB ) , (A19.1)
Q
where RB is the rotation of Eq. (14.4)
RB ≡ (−γB , −θB , −χB ) .
Use of Eq. (2.68) shows that KQQK
(RB ) is independent of the angle γB .
As proved in Sect. 10.8 (see Eq. (10.74)), Eq. (A19.1) can be rewritten in a
simpler form, involving a single rotation matrix
K K 1
KQQ
K
(RB ) = K(K + 1)(2K + 1) (−1)K−Q D0q
1
(RB ) . (A19.2)
Q −Q q
This shows that the kernel K00
0
(RB ) is zero, and that the only non-zero components
are those connecting Q-values such that ∆Q ≡ Q − Q = 0, ±1.
APPENDIX 845
It can easily be seen, with the help of Eqs. (2.24), (2.25), and (2.73), that
KQQ
K
(RB ) satisfies the symmetry properties
∗
KQ
K
Q (RB ) = KQQ (RB )
K
K−Q−Q
K
(RB ) = (−1)
2K+1+Q+Q
KQ
K
Q (RB ) .
Explicit expressions for the components with K = 1 and K = 2 can be obtained
from Eq. (A19.2) using Eqs. (2.68), (2.70), (2.26), and the algebraic formulae for
the reduced rotation matrices given in Table 2.1. The non-zero components are
the following
K−1−1
1
= −K11
1
= − cos θB
1 iχ
K−10
1 1
= (K0−1 )∗ = (K10
1 ∗
) = K01
1
= √ sin θB e B
2
K−2−2
2
= −K22
2
= −2 cos θB
iχB
K−2−1
2 2
= (K−1−2 )∗ = (K21
2 ∗
) = K12
2
= sin θB e
K−1−1
2
= −K11
2
= − cos θB
.
3 iχ
K−10
2
= 2
(K0−1 )∗ = 2 ∗
(K10 ) = K01
2
= sin θB e B .
2
A20. The Multipole Coupling Coefficients
The multipole coupling coefficients are defined by Eq. (14.24),
∞
2 −τ (x,x )
GKQ,K Q (x, x ) = dν p(ν0 − ν) e ν
−∞
3
× wJ(K)J wJ(KJ) (−1)Q T−Q
K T K (i, Ω)
(i, Ω) , (A20.1)
u u
Q
i=0
where the symbol wJ(K)J is given by Eq. (10.11), the tensor T Q K – given by
(i, Ω)
u
Eq. (5.134) – is defined in the system Σ of Fig. 14.4, the unit vector Ω is given by
(x − x )/|x − x |, p(ν0 − ν) is the Gaussian profile of Eq. (14.6), and τν (x, x ) is
the optical thickness at frequency ν between points x and x (see Eq. (14.17)).
Since T QK is defined for K = 0, 1, 2 and, for a fixed K, for Q = −K, . . . , K,
(i, Ω)
it follows that there are in principle 81 different multipole coupling coefficients
satisfying the limitations
K = 0, 1, 2 K = 0, 1, 2
Q = −K, . . . , K Q = −K , . . . , K .
846 APPENDIX
However, the number of non-zero, independent quantities is much smaller, as it
will be shown in the following. Moreover, for special values of Ju this number can
further decrease because of the presence of the symbol wJ(K)J . For Ju = 0, for
u
instance, only the symbol G00,00 is non-zero, whereas for Ju = 1/2 the indices K
and K run only on the values 0 and 1.
To study the properties of the multipole coupling coefficients it is convenient to
rewrite Eq. (A20.1) in the form
∞
2 −τ (x,x )
GKQ,K Q (x, x ) = dν p(ν0 − ν) e ν
−∞
,
× wJ(K)J wJ(KJ) ΓKQ,K Q (Ω) (A20.2)
u u
where
3
=
ΓKQ,K Q (Ω) (−1)Q T−Q
K T K (i, Ω)
(i, Ω) .
Q
i=0
From the conjugation property of the tensor T Q
K – see Eqs. (5.158) – it follows
(i, Ω)
=Γ
ΓK Q ,KQ (Ω) ∗
KQ,K Q (Ω) (A20.3)
and
= (−1)Q+Q Γ
ΓK−Q,K −Q (Ω) ∗
KQ,K Q (Ω) . (A20.4)
Obviously, similar relations hold for the multipole coupling coefficients,
GK Q ,KQ (x, x ) = GKQ,K Q (x, x )∗ (A20.5)
and
GK−Q,K −Q (x, x ) = (−1)Q+Q GKQ,K Q (x, x )∗ . (A20.6)
can be expressed in terms of rotation matrices using
The quantity ΓKQ,K Q (Ω)
Eq. (5.159)
3
= (−1)Q
P (i) DP −Q (R) tP (i) DP Q (R) ,
tK K K K
ΓKQ,K Q (Ω) (A20.7)
i=0 P P
e (Ω),
where R ≡ (−γ, −θ, −χ) is the rotation carrying the system (ea (Ω), Ω) into
b
the system Σ where the multipole coupling coefficients are defined (see Fig. 14.4).
On the other hand, Eqs. (5.162) and (5.161) yield
3
3
3
tK K
P (i) tP (i) = (−1)
K+K
ξi2 tK K
P (i) tP (i) = (−1)
K+K
tK K
P (i) tP (i) ,
i=0 i=0 i=0
APPENDIX 847
which implies
= (−1)K+K Γ
ΓKQ,K Q (Ω)
KQ,K Q (Ω)
and
GKQ,K Q (x, x ) = (−1)K+K GKQ,K Q (x, x ) . (A20.8)
and G
As a consequence, the quantities ΓKQ,K Q (Ω) x, x ) vanish unless K
KQ,K Q (
and K are both even or both odd. This reduces the number of non-zero components
from 81 to 45. Taking into account the symmetry properties (A20.3) and (A20.4),
it can easily be seen that the number of non-zero, independent components is 17.
Equation (A20.7) can be simplified via the following transformations. First we
express the coefficients tK P (i) in terms of the Pauli spin matrices σ̂i defined in
Eq. (5.128). Using Eq. (5.160) we have
3
3
tK K
P (i) tP (i) = (2K + 1)(2K + 1)
i=0
4
3
1 1 K 1 1 K
× (σ̂i )αβ (σ̂i )γδ ,
β −α −P δ −γ −P
i=0 αβγδ=±1
where the rows and columns of σ̂i are labelled as in Eqs. (5.129). Using Eq. (8.64)
we get1
3
3
tK K
P (i) tP (i) = (2K + 1)(2K + 1)
i=0
2
1 1 K
1 1 K
× . (A20.9)
β −α −P α −β −P
αβ=±1
This equation shows that P = −P and that the only allowed values for P and P
are 0 and ±2. Substitution of Eq. (A20.9) into Eq. (A20.7) and use of Eq. (2.75)
yields
= (−1)Q 3
ΓKQ,K Q (Ω) (2K + 1)(2K + 1)
2
1 1 K 1 1 K
× (2K + 1)
β −α −P α −β −P
αβ=±1 P P K
K K K K K K
∗
× DSS
K
(R) .
P P S −Q Q S
The sums over α and β cannot be performed by standard sum rules of Racah
algebra because the ‘components’ α and β run on the values ±1 but not on the
1
We recall that Eq. (8.64) is valid for any representation of the Pauli spin matrices.
848 APPENDIX
value 0. To overcome this difficulty, we multiply the right-hand side by the quantity
α2 β 2 which, using Eq. (3.52), can be written as
! !
1 1 f 1 1 f
α2 β 2 = (−1)α−β 36 (2f + 1)(2f + 1)
1 1 1 1 1 1
ff
1 1 f 1 1 f 1 1 f 1 1 f
× .
0 0 0 0 0 0 α −α 0 β −β 0
Now the sum can be extended to the values α = 0 and β = 0, and the summation
over the indices α, β, P , P can be carried out with the help of Eq. (2.52). Changing
the index K into f , and using Eq. (2.73), we obtain the expression
= (−1)K +Q 54
ΓKQ,K Q (Ω) (2K + 1)(2K + 1)
1 1 f 1 1 f
× (2f + 1)(2f + 1)(2f + 1)
0 0 0 0 0 0
ff f
!
f f f K K f 1 1 f
×
0 0 0 −Q Q Q − Q 1 1 1
⎧ ⎫
!⎨f f f ⎬
1 1 f f
× 1 1 K D0,Q −Q (R) . (A20.10)
1 1 1 ⎩ ⎭
1 1 K
It should be noticed that, owing to the symmetry properties of the 3-j symbols,
the sum over f , f , f is in fact restricted to even values, thus it contains 6 terms
at most, corresponding respectively to the following triplets (f, f , f ): (0, 0, 0),
(0, 2, 2), (2, 0, 2), (2, 2, 0), (2, 2, 2), and (2, 2, 4). According to the values of K and
K , some of these terms can however be zero. If, for instance, K and K are both
zero, it follows that f = 0 and the sum contains only two terms.
The explicit expressions of the non-zero Γ ’s can be found from Eq. (A20.10) by
evaluating the various 3-j, 6-j, and 9-j symbols and by using the analytical formulae
for the rotation matrices (Eqs. (2.68) and Table 2.1). The results – that depend
on the angles θ and χ specifying the direction Ω but are obviously independent of
the angle γ – are the following
=1
Γ00,00 (Ω) =
Γ20,20 (Ω) 1
(5 − 12 cos2 θ + 9 cos4 θ)
4
= = iχ
Γ00,20 (Ω) 1
√
2 2
(3 cos2 θ − 1) Γ20,21 (Ω) 1
2
3
2 sin θ cos θ (2 − 3 cos2 θ) e
√
=− iχ = 2iχ
Γ00,21 (Ω) 2
3
sin θ cos θ e Γ20,22 (Ω) 1
4
3
2 sin2 θ (1 + 3 cos2 θ) e
√
= 2iχ = − 3 sin3 θ cos θ e −3iχ
Γ00,22 (Ω) 4
3
sin2 θ e Γ21,2−2 (Ω) 4
= −2iχ
Γ21,2−1 (Ω) 3
4 sin2 θ (1 − 2 cos2 θ) e
APPENDIX 849
=
Γ10,10 (Ω) 3
cos2 θ =
Γ21,21 (Ω) 3
sin2 θ (1 + 2 cos2 θ)
2 4
=− √
3 iχ = 3 iχ
Γ10,11 (Ω) 2 2
sin θ cos θ e Γ21,22 (Ω) 4 sin θ cos θ (1 + cos2 θ) e
= − 3 sin2 θ e−2iχ
Γ11,1−1 (Ω) =
Γ22,2−2 (Ω) 3
sin4 θ e
−4iχ
4 8
=
Γ11,11 (Ω) 3
sin2 θ =
Γ22,22 (Ω) 3
(1 + cos2 θ)2 . (A20.11)
4 8
The remaining Γ ’s can easily be deduced from the symmetry properties (A20.3)
and (A20.4). The multipole coupling coefficients for a given transition (J , Ju ) are
then obtained from Eq. (A20.2).1
is their invariance under
A remarkable property of the quantities ΓKQ,K Q (Ω)
inversion of the direction Ω,
=Γ
ΓKQ,K Q (−Ω)
KQ,K Q (Ω) . (A20.12)
To prove this relation, we recall that the rotation appearing in Eq. (A20.10) carries
e (Ω),
the reference system (ea (Ω), Ω)
into the system Σ of Fig. 14.4,
b
R ≡ (−γ, −θ, −χ) .
e (−Ω),
The rotation carrying (ea (−Ω), −Ω)
into Σ is
b
R ≡ (−γ , −(π − θ), −(π + χ)) ,
where the angle γ specifies the direction of the unit vector ea (−Ω)
in the plane
perpendicular to Ω. Thus the rotation matrix appearing in the expression of
is
ΓKQ,K Q (−Ω)
f
i(π+χ)(Q −Q)
D0,Q −Q (R ) = e df0,Q −Q (θ − π) = (−1)f D0,Q
f
−Q (R) ,
where Eqs. (2.68) and (2.70) have been used. But the summation in Eq. (A20.10)
is limited to even values of the index f , so Eq. (A20.12) is proved. In fact, it
can easily be checked that the analytical expressions of Eqs. (A20.11) are invariant
under the transformation θ → π − θ, χ → π + χ.
Since the optical thickness τν (x, x ) is invariant under exchange of x and x ,
τν (x , x) = τν (x, x ) ,
it follows from Eqs. (A20.2) and (A20.12) that the multipole coupling coefficients
satisfy the important relation
GKQ,K Q (x , x) = GKQ,K Q (x, x ) .
1 and G
Note that Eq. (A20.10) implies that the components ΓKQ,K Q (Ω) x, x ) are
KQ,K Q (
real.
850 APPENDIX
Finally, from Eq. (A20.10) one can easily evaluate the solid-angle average of the
defined by
quantity ΓKQ,K Q (Ω),
5
1
ΓKQ,K Q (Ω) = .
dΩ ΓKQ,K Q (Ω)
4π
Since D00
0
(αβγ) = 1, we obtain from the Weyl theorem (Eq. (2.76))
2π π
dγ dβ sin β D0N
J
(αβγ) = 4π δJ0 δN 0 ,
0 0
whence, using Eqs. (2.68) and (2.70)
5
1 f
dΩ D0,Q −Q (−γ −θ −χ) = δf 0 δQQ .
4π
Substitution into Eq. (A20.10) yields, with the help of Eqs. (2.26a) and (2.49)
= δ δ (−1)K 54
ΓKQ,K Q (Ω) KK QQ
2 !2 !
1 1 f 1 1 f 1 1 K
× (2f + 1) ,
0 0 0 1 1 1 1 1 f
f =0,2
and evaluating the relevant 3-j and 6-j symbols (Eqs. (2.26a,f), (2.36a,f))
!
2 1 1 K
ΓKQ,K Q (Ω) = δKK δQQ + (−1) K
,
3 1 1 2
or, explicitly (Eqs. (2.36a,d,h))
=1
Γ00,00 (Ω)
=1
Γ1Q,1Q (Ω) (Q = 0, ±1)
2
= 7
Γ2Q,2Q (Ω) (Q = 0, ±1, ±2) .
10
A21. The Calculation of a Double Integral
The integral to be calculated is the following (see Eq. (14.26))
∞ ∞
∆νD
I= dx dy G x, x ) ,
( (A21.1)
4π(x − x )2 KQ,K Q
−∞ −∞
APPENDIX 851
Ω
θ
χ
α
Γ
Fig.A21.1. In the reference system Σ we consider a fixed point P located at height z corresponding
to line optical depth tL . The point P lies on the plane Γ parallel to the x-y plane; its height is
z corresponding to line optical depth tL . The cylindrical coordinates (r, α) define the position of
P in the plane Γ . The angles θ and χ specify the direction Ω.
where GKQ,K Q (x, x ) is defined in Eq. (14.24), ∆νD is a constant (representing
the Doppler broadening of the absorption profile), and the geometry is illustrated in
Fig. A21.1 (for the case tL > tL or z < z). Introducing the cylindrical coordinates
r and α of the point P , one has
dx dy = r dr dα .
On the other hand
(x − x )2 = (z − z )2 + r2 ,
and
z − z
r = (z − z ) tan θ , dr = dθ .
cos2 θ
From these relations we get
dx dy
= tan θ dθ dα ,
(x − x )2
and since χ = α + π, the double integral over x and y can be transformed into an
integral over the angles θ and χ specifying the direction Ω,
2π π/2
∆νD .
I= dχ dθ tan θ GKQ,K Q (Ω) (A21.2)
4π
0 0
852 APPENDIX
To make the dependence of GKQ,K Q (Ω) on θ and χ explicit, we express the
optical thickness τν (x, x ) defined in Eq. (14.17) in terms of the line optical depth
of Eq. (14.25)
∆νD p(ν0 − ν) (tL − tL )
τν (x, x ) = , (A21.3)
cos θ
and write the profile p(ν0 − ν) in the form (see Eq. (14.6))
1
p(ν0 − ν) = ϕ(v) , (A21.4)
∆νD
where
ν0 − ν 1 −v 2
v= , ϕ(v) = √ e , (A21.5)
∆νD π
with
∞
ϕ(v) dv = 1 .
−∞
Substitution of Eqs. (A21.3), (A21.4) into Eq. (A21.2) and use of Eq. (A20.2) leads
to the expression
∞ 2π π/2 (t −t ) ϕ(v)
1 2 − L L
I= dv ϕ(v) dχ dθ tan θ e cos θ
4π
−∞ 0 0
.
× wJ(K)J wJ(KJ) ΓKQ,K Q (Ω) (A21.6)
u u
on the angles θ and
Equation (A20.10) shows that the dependence of ΓKQ,K Q (Ω)
χ is contained in the factor
f iχ(Q −Q)
D0,Q −Q (−γ −θ −χ) = e df0,Q −Q (−θ) ,
hence the integration over χ yields 2π for Q = Q and zero for Q = Q . On the other
are linear combinations
side, Eqs. (A20.11) show that the quantities ΓKQ,K Q (Ω)
2n
of factors of the form [cos θ] with n = 0, 1, 2,
2
=
ΓKQ,K Q (Ω) cn (K, K , Q) [cos θ]2n . (A21.7)
n=0
It follows that Eq. (A21.6) can be rewritten as
∞
1 (K) (K ) 2
2
I = δQQ wJ J wJ J cn (K, K , Q) dv ϕ(v) In (tL − tL ) ϕ(v) ,
2 u u n=0
−∞
APPENDIX 853
where
π/2 (t −t ) ϕ(v)
− L L
In (tL − tL ) ϕ(v) = e cos θ [cos θ]2n tan θ dθ .
0
Introducing the variable y = 1/ cos θ, it can readily be seen that the last integral
is nothing but the exponential-integral function of order (2n + 1) – cf. Eq. (12.6)
∞ −(tL −tL ) ϕ(v) y
e
In (tL − tL ) ϕ(v) = dy ≡ E2n+1 (tL − tL ) ϕ(v) .
y 2n+1
1
In the above calculations we assumed tL > tL . The opposite case can be treated
in a strictly similar way. The final result is that, irrespective of the relative height
of the points x and x , the integral in Eq. (A21.1) can be written in the form
1 (K) (K )
I = δQQ w w
2 Ju J Ju J
∞
2
2
× cn (K, K , Q) dv ϕ(v) E2n+1 |tL − tL | ϕ(v) , (A21.8)
n=0 −∞
with the coefficients cn (K, K , Q) implicitly defined by Eq. (A21.7).
√
A22. The Generalization of the - Law
One of the few analytical results that can be established in the scalar theory of
radiative transfer concerns the integral equation
∞
(1 + ) S(τ ) = BP + K |τ − τ | S(τ ) dτ , (A22.1)
0
that can also be written in the form
∞
S(τ ) = BP + (1 − ) K |τ − τ | S(τ ) dτ ,
0
with
= . (A22.2)
1+
This equation is encountered in the non-LTE approach to the problem of line for-
mation for a two-level atom in plane-parallel atmospheres, and relates the source
function at optical depth τ to the same function at different optical depths. BP
854 APPENDIX
is the Planck function at the kinetic temperature of the colliding particles (char-
acterized by a Maxwellian velocity distribution), and is the ratio of collisional to
radiative de-excitation rates for the upper level – a quantity specifying the degree
of coupling between atoms and radiation.
Provided and BP are independent of τ , and the kernel K obeys the integral
property
∞
1
K(t) dt = , (A22.3)
2
0
it can be proved that the value of the source function at τ = 0 is given by
√
S(0) = BP .
√
This result, generally referred to as the - law , is discussed in classical papers
and textbooks (Avrett and Hummer, 1965; Jefferies, 1968; Mihalas, 1978). It is
already contained in the discrete-ordinate method for coherent scattering developed
by Chandrasekhar (1947), but, according to Avrett and Hummer (1965), it is also
implicit in the work of a number of earlier scientists. More recent derivations based
on different mathematical methods can be found in Ivanov (1973), Frisch and Frisch
(1975),√Rybicki (1977), and Landi Degl’Innocenti (1979b).
The - law can be generalized to a linear system of (N + 1) equations of the
form
∞
N
ai Si (τ ) = b δi0 + Kij |τ − τ | Sj (τ ) dτ (i = 0, . . . , N ) , (A22.4)
j=0 0
where:
- ai and b are non-vanishing quantities independent of τ , with
a0 > 1 , b>0;
- the kernels Kij are integrable functions of their argument and satisfy the sym-
metry property
Kij (t) = Kji (t) (i, j = 0, . . . , N ) , (A22.5)
and the integral property
∞
1
K0j (t) dt = δ (j = 0, . . . , N ) . (A22.6)
2 j0
0
Under these hypotheses, it can be proved (Landi Degl’Innocenti and Bommier,
1994) that the surface values of the ‘source functions’ Si (τ ) obey the relation
N
2 b2
ai Si (0) = . (A22.7)
i=0
a0 − 1
APPENDIX 855
To prove this relation, first we consider the asymptotic value of Si (τ ) for τ → ∞.
Since the kernels are integrable functions of their argument, the functions Si (τ )
tend to finite values, Si (∞), for τ → ∞. Such values can be found by solving the
system
N
∞ τ !
ai Si (∞) = b δi0 + lim Kij (τ − τ ) dτ + Kij (τ − τ ) dτ Sj (∞) ,
τ →∞
j=0 τ 0
where i = 0, . . . , N . Performing the substitutions (τ − τ ) = u and (τ − τ ) = v in
the first and second integral, respectively, we get
N
ai Si (∞) = b δi0 + 2 Aij Sj (∞) (i = 0, . . . , N ) , (A22.8)
j=0
where
∞
Aij = Kij (t) dt . (A22.9)
0
Owing to property (A22.6), the system (A22.8) decouples into
a0 S0 (∞) = b + S0 (∞)
N
ai Si (∞) = 2 Aij Sj (∞) (i = 1, . . . , N ) ,
j=1
whose solution is
b
Si (∞) = δ . (A22.10)
a0 − 1 i0
Next we consider the τ -derivative of Eq. (A22.4),
∞
dS (τ ) d
N
ai i = Kij |τ − τ | Sj (τ ) dτ .
dτ j=0
dτ
0
The derivative in the right-hand side can be evaluated by splitting the integral in
two parts (cf. the derivation of Eq. (A15.3)). One obtains
∞
dSi (τ ) dSj (τ )
N N
ai = Kij (τ ) Sj (0) + Kij |τ − τ |
dτ .
dτ j=0 j=0
dτ
0
√
Following a method developed by Frisch and Frisch (1975) to derive the - law
in the scalar case, we multiply both members by Si (τ ), integrate in dτ , and sum
over i,
856 APPENDIX
N ∞ N N ∞
dS (τ )
ai Si (τ ) i dτ = Sj (0) Kij (τ ) Si (τ ) dτ
i=0
dτ i=0 j=0
0 0
N ∞ ∞
N
dSj (τ )
+ dτ Si (τ ) dτ Kij |τ − τ | . (A22.11)
i=0 j=0
dτ
0 0
The term in the left-hand side is immediately evaluated via integration by parts,
and gives
1
N & 2 2 '
ai Si (∞) − Si (0) .
2 i=0
Using Eqs. (A22.5) and (A22.4), the first term in the right-hand side can be cast
into the form
N
2
ai Si (0) − b S0 (0) .
i=0
For the second term in the right-hand side we obtain, by interchanging the inte-
gration order and by using again Eqs. (A22.5) and (A22.4)
N ∞ ∞
dS (τ ) dS0 (τ )
ai Si (τ ) i dτ − b dτ ,
i=0
dτ dτ
0 0
or
1
N & 2 2 ' & '
ai Si (∞) − Si (0) − b S0 (∞) − S0 (0) .
2 i=0
Using these expressions, we obtain from Eq. (A22.11)
N
2
ai Si (0) = b S0 (∞) ,
i=0
√
which proves – taking into account Eq. (A22.10) – the generalized - law of
Eq. (A22.7). √
In the following we show that the ‘classical’ - law can be recovered as a special
√
case of Eq. (A22.7), and we deduce the explicit expression of the generalized -
law for some physical situations that are treated in the text.
a) Scalar case
Equation (A22.1) is a special case of Eq. (A22.4) corresponding to
N =0, S0 = S , a0 = 1 + , b = BP .
APPENDIX 857
Since the quantities and BP in Eq. (A22.1) are τ -independent and positive, and
the kernel K obeys Eq. (A22.3), it is easily seen that all the conditions for the
validity of Eq. (A22.7) are met. Performing the above substitutions, we obtain
from Eq. (A22.7)
2
(1 + ) S(0) = BP2 ,
or .
√
S(0) = BP = BP , (A22.12)
1+
√
where Eq. (A22.2) has been used. Equation (A22.12) is just the ‘classical’ - law.
b) Non-magnetic atmosphere
Consider the system of integral equations (14.27) under the assumption that ,
BP , and δu(K) are independent of tL and that Hu = 0: we obtain the system of
Eqs. (14.35) with the only difference that the factor in front of S02 (tL ) in the left-
hand side of the second equation is (1 + + δu(2) ) instead of (1 + ). This system
has the form of Eq. (A22.4) with
N =1, S0 = S00 , S1 = S02 , a0 = 1 + , a1 = 1 + + δu(2) , b = BP ,
and because the multipolar kernels GKQ,K Q (x) defined in Eq. (14.28) are integrable
functions of x satisfying the symmetry properties of Eq. (14.29) and the integral
properties of Eqs. (14.32), Eq. (A22.7) can be applied. We get
2 2
(1 + ) S00 (0) + (1 + + δu(2) ) S02 (0) = BP2 ,
or
0 2 2 √
S0 (0) + [ 1 + (1 − ) δu(2) ] S02 (0) = BP . (A22.13)
√
This is the explicit form of the generalized - law for the physical situation con-
sidered.
c) Magnetic atmosphere (Hanle effect regime)
The system in Eq. (14.39) has the form of Eq. (A22.4) with
2
N =5, S0 = S00 M , Si = Si−3 M
(i = 1, . . . , 5) ,
a0 = 1 + , ai = 1 + + δu(2) + iHu (i − 3) (i = 1, . . . , 5) , b = BP .
Let us assume that , BP , δu(2) , and the magnetic field vector are independent of tL ,
and that the x-axis of the ‘magnetic frame’ lies in the plane containing the magnetic
field and the perpendicular to the atmosphere (RB = RB ≡ (0, −θB , −χB ) in
Eq. (14.40)). Under these assumptions the kernels ĜKQ,K Q (x) obey the properties
858 APPENDIX
of Eqs. (14.41) and (14.42), and the conditions for the validity of Eq. (A22.7) are
satisfied. We get
A
B
B 2
√
C S 0 (0) 2 + [ 1 + (1 − ) (δ (2) + iH Q) ] S 2 (0) 2 =
u BP . (A22.14)
0 M u Q M
Q=−2
d) Magnetic atmosphere (strong field regime)
The system in Eq. (14.62) has the form of Eq. (A22.4) with
1
N= K , S0 = S00 M , Si = S02i M (i = 1, . . . , N ) ,
2 max
a0 = 1 + , ai = 1 + + δu(2i) (i = 1, . . . , N ) , b = BP .
The kernels G̃K0,K 0 (x) obey the properties in Eqs. (14.65) and (14.66). Provided ,
BP , and δu(K) are independent of tL , all the conditions for the validity of Eq. (A22.7)
are satisfied, so that
A
B
B 0 2 2 √
C S0 (0) + [ 1 + (1 − ) δu(K) ] S0K (0) M = BP . (A22.15)
M
K=2,4,...,Kmax
√
Finally, it should be remarked that a further generalization of the - law has
been proved by Frisch (1998). This generalization concerns the coupled system of
(N + 1) integral equations of the form
∞
N
ai Si (τ ) = bi + Kij |τ − τ | Sj (τ ) dτ (i = 0, . . . , N ) ,
j=0 0
where:
- the coefficients ai , bi are independent of τ , with
a0 > 1 , b0 > 0 ;
- the kernels Kij have the same properties as the kernels appearing in Eq. (A22.4).
Following a procedure similar to the former, it can be proved that
N
2 b20 N N
ai Si (0) = + bi (X −1 )ij bj ,
i=0
a0 − 1 i=1 j=1
where the matrix X is defined by
Xij = ai δij − 2 Aij (i, j = 1, . . . , N ) ,
with Aij given by Eq. (A22.9).
APPENDIX 859
A23. The Generalized Multipole Coupling Coefficients
The generalized multipole coupling coefficients are defined by Eq. (14.60),
∞
K
G̃K0,K 0 (x, x ) = dν Φ̃KK
0 (J , Ju ; ν) Φ̃K
0 (J , Ju ; ν)
K K −∞
3
3
× O (x, x ; ν) T K (j, Ω)
T 0K (i, Ω) , (A23.1)
ij 0
i=0 j=0
where the profiles Φ̃KK0 (J , Ju ; ν) are given by Eq. (14.48), Oij (x, x ; ν) is the
evolution operator defined in Eq. (14.57), Ω is the direction of the vector joining
point x with point x, and the tensors T 0 are defined in the reference system of the
K
magnetic field. They are real quantities (because the single factors appearing in the
right-hand side are real – see Eqs. (5.158) and (A13.2)),1 and obey the following
properties:
a) Limitation on indices
As apparent from Eq. (14.48), the indices K and K are restricted to the range
K, K = 0, 1, . . . , 2Ju .
The number of generalized multipole coupling coefficients is therefore (2Ju + 1)2 .
b) Limit for zero magnetic field
From Eq. (A13.5) we have
lim Φ̃KK
0 (J , Ju ; ν) = δKK wJ(K)J p(ν0 − ν) , (A23.2)
νL →0 u
and from Eq. (A13.13)
lim Ψ̃00K (J , Ju ; ν) = δK0 q(ν0 − ν) , (A23.3)
νL →0
where the profile q(ν0 − ν) is given by Eq. (14.53). As a consequence, Eqs. (14.51)
and (14.52) yield
= k A p(ν − ν) δ
lim ηiA (ν, Ω) (i = 0, 1, 2, 3)
νL →0 L 0 i0
=0
lim ρAi (ν, Ω) (i = 1, 2, 3) ,
νL →0
1
Equation (A13.2) refers in fact to the generalized profiles ΦKKQ , but it can easily be seen
that the same relation holds for Φ̃KK
Q . This remark also applies to other properties that will be
needed in the following.
860 APPENDIX
defined in Eq. (14.56) becomes diagonal,
so that the matrix H(ν, Ω)
= p(ν − ν) δ .
lim Hij (ν, Ω)
νL →0 0 ij
Thus from Eq. (14.57)
−τν ( x )
x,
lim Oij (x, x ; ν) = e δij , (A23.4)
νL →0
where τν (x, x ) is given by Eq. (14.17). Substitution of Eqs. (A23.2) and (A23.4)
into Eq. (A23.1) yields
∞
2 −τ (x,x ) (K) (K )
lim G̃K0,K 0 (x, x ) = dν p(ν0 − ν) e ν wJ J wJ J
νL →0 u u
−∞
3
× T K (i, Ω)
T 0K (i, Ω) .
0
i=0
This expression formally coincides with the definition of GK0,K 0 (x, x ) given in
Eq. (14.24), but one must bear in mind the difference between the reference systems
where the two quantities are defined.
c) Expression in terms of rotation matrices
Taking into account Eq. (5.159), the generalized multipole coupling coefficients can
be expressed in the form
∞
K
G̃K0,K 0 (x, x ) = dν Φ̃KK
0 (J , Ju ; ν) Φ̃K
0 (J , Ju ; ν)
K K P P −∞
3
3
× tK x, x ; ν) tK
P (i) Oij ( P (j) DP 0 (R) DP 0 (R) ,
K K
(A23.5)
i=0 j=0
e (Ω),
where R is the rotation carrying the reference system (ea (Ω), Ω) into the
b
system of the magnetic field.
d) Vanishing values
G̃K0,K 0 (x, x ) = (−1)K+K G̃K0,K 0 (x, x ) , (A23.6)
a relation showing that all the generalized multipole coupling coefficients connecting
irreducible components of the source function of different parity are identically zero.
To prove this property, we consider two frequencies ν, ν symmetrical about line
center (ν − ν0 = ν0 − ν). From Eqs. (A13.4) and (A13.12) we have
Φ̃0K K 0K
0 (J , Ju ; ν ) = (−1) Φ̃0 (J , Ju ; ν)
Ψ̃00K (J , Ju ; ν ) = (−1)1+K Ψ̃00K (J , Ju ; ν) .
APPENDIX 861
Using these relations and the expressions of the tensors T Q K – Table 5.6 – it is
(i, Ω)
easily seen from Eqs. (14.51) and (14.52) that the radiative transfer coefficients η0A ,
η1A , η2A , ρA3 are symmetrical about line center, while η3A , ρA1 , ρA2 are antisymmetrical.
Thus the matrix H of Eq. (14.56) satisfies the symmetry property1
Hij (ν ) = ξi ξj Hij (ν) (i, j = 0, . . . , 3) ,
where ξi is the formal vector defined in Eqs. (5.161). Since this property remains
valid for any arbitrary function of the matrix H, it holds also for the evolution
operator,
Oij (x, x ; ν ) = ξi ξj Oij (x, x ; ν) .
Using the first of Eqs. (5.162), it can be shown that the integrand in Eq. (A23.5)
has the parity of (K + K ), so that Eq. (A23.6) follows after integration.
e) Symmetry with respect to the indices
G̃K0,K 0 (x, x ) = G̃K 0,K0 (x, x ) . (A23.7)
To prove this property, one should bear in mind that the generalized multipole
coupling coefficients do not depend, as obvious on general physical grounds, on the
chosen to define the Stokes parameters of the radiation
reference direction ea (Ω)
flowing in the direction Ω. On the other hand, it can easily be shown that if the
and
reference direction is rotated through an angle α, the components T 02 (1, Ω)
transform according to the equations
T 0 (2, Ω)
2
2
T 0 (1, Ω)
= cos 2α T 02 (1, Ω)
+ sin 2α T 02 (2, Ω)
new old old
2
T 0 (2, Ω)
= − sin 2α T 02 (1, Ω)
+ cos 2α T 02 (2, Ω) .
new old old
van-
This implies that it is possible to find a reference direction such that T 02 (2, Ω)
A
ishes. With this choice, one obtains from Eqs. (14.51) and (14.52) that η2 (ν, Ω) =
A = 0, so that the matrix K defined in Eq. (14.50) takes the form
ρ2 (ν, Ω) A
⎛ ⎞
η0A (ν, Ω)
η1A (ν, Ω) 0
η3A (ν, Ω)
⎜ A ⎟
⎜ η1 (ν, Ω)
η A (ν, Ω)
ρA3 (ν, Ω) 0 ⎟
KA = ⎜
⎜
0 ⎟ .
⎝ 0
−ρA3 (ν, Ω)
η A (ν, Ω) ⎟
ρA1 (ν, Ω) ⎠
0
A
η (ν, Ω) 0
−ρA1 (ν, Ω) A
η (ν, Ω)
3 0
1
This is the well-known property of the absorption matrix deduced in Chap. 9, Eq. (9.51).
We recall that in the non-LTE problem considered in Sect. 14.3 the lower level is assumed to be
unpolarized.
862 APPENDIX
It follows that the matrix H of Eq. (14.56) obeys the symmetry property
Hji = τi τj Hij (i, j = 0, . . . , 3) ,
where τi is the formal vector defined in Eqs. (5.161). Since this property holds for
any function of the matrix H, it is also valid for the evolution operator,
Oji (x, x ; ν) = τi τj Oij (x, x ; ν) .
= 0, one also has
On the other hand, being T 0K (2, Ω)
= T K (i, Ω)
τi T 0K (i, Ω) ,
0
so that
O (x, x ; ν) T K (j, Ω)
T 0K (i, Ω) O (x, x ; ν) T K (j, Ω)
= T K (i, Ω) .
ji 0 0 ij 0
This means that the indices i and j of the evolution operator in Eq. (A23.1) can
be interchanged. This is sufficient to prove Eq. (A23.7) by exchange of the dummy
summation indices K and K in Eq. (A23.1).
f) Symmetry with respect to the arguments
G̃K0,K 0 (x, x ) = G̃K0,K 0 (x , x) . (A23.8)
To prove this property, we observe that the interchange of x and x implies the
interchange of the direction Ω with the opposite direction −Ω. Since, as already
noticed at point e), the generalized multipole coupling coefficients do not depend
on the reference direction, we choose for the direction −Ω a reference direction
unit vector ea (−Ω) = ea (Ω), as shown in Fig. 5.15. With this choice, one has from
Eq. (5.163)
= ζ T K (i, Ω)
T 0K (i, −Ω) (i = 0, . . . , 3) , (A23.9)
i 0
where ζi is the formal vector defined in Eqs. (5.161). Substitution of Eq. (A23.9)
into Eqs. (14.51) and (14.52) shows that the matrix H of Eq. (14.56) satisfies, at
each point of the medium, the property
= ζ ζ H (ν, Ω)
Hij (ν, −Ω) ,
i j ij
and because the optical depth τ (x, x ) is obviously invariant under interchange of
its arguments, it follows that
Oij (x , x ; ν) = ζi ζj Oij (x, x ; ν) . (A23.10)
Substituting Eqs. (A23.9) and (A23.10) into Eq. (A23.1), property (A23.8) is easily
proved.
APPENDIX 863
A24. Reduced Matrix Elements for Photoionization
Cross Sections
We consider a simple case where photoionization involves the absorption of a photon
by an optical electron belonging to an open shell (n b ). The photoionized electron
leaves the atom with kinetic energy , orbital angular momentum , and total
angular momentum (orbital + spin) j. Assuming the atom to be described in
the L-S coupling scheme, the reduced matrix element entering the photoionization
cross section is of the form (see Eq. (14.83))
M = | β L0 S0 , n b ; LSJ d β (L0 S0 )J0 , j; J |2 , (A24.1)
where, with obvious notations, (βL0 S0 ) is the set of quantum numbers describing
the parent term of the photoionized level and, at the same time, the term encom-
passing the level where the atom is left after photoionization (this final level being
further described by the quantum number J0 ).
Notwithstanding the simplifications introduced, the evaluation of the matrix el-
ement is rather complicated because of the different coupling schemes of the ‘bra’
and ‘ket’ vectors. A recoupling of the various angular momenta is necessary. With
standard techniques one has
|β L0 S0 , n b ; LSJ = (2L + 1)(2S + 1) (2J0 + 1)(2jb + 1)
J0 jb
⎧ ⎫
⎨ L0 S0 J0 ⎬
× b 1
jb |β (L0 S0 )J0 , n b jb ; J . (A24.2)
⎩ 2 ⎭
L S J
On the other hand, since the operator d only acts on the variables of the optical
electron, using Eq. (2.109) one has
β (L0 S0 )J0 , n b jb ; J d β (L0 S0 )J0 , j; J =
!
jb j 1
= (−1) J0 +j+J+1
(2J + 1)(2jb + 1) n b jb d j δJ J ,
J J J0 0 0
and taking into account that d only acts on the orbital variables of the optical
electron, from Eq. (2.108) one obtains
β (L0 S0 )J0 , n b jb ; J d β (L0 S0 )J0 , j; J =
1
= (−1)J0 +J−b − 2 (2J + 1)(2jb + 1)(2j + 1)(2 b + 1)
! !
jb j 1 b 1
× n b d . (A24.3)
J J J0 j jb 21
864 APPENDIX
The reduced matrix element n b d can be evaluated through standard meth-
ods. One has
√ b 1
n b d = e0 (−1) +1
2 + 1 I(n b , ) , (A24.4)
0 0 0
where I(n b , ) is the radial integral defined by
∞
I(n b , ) = Pn (r) r χ (r) dr , (A24.5)
b
0
Pn (r) being the reduced radial wavefunction of the bound electron and χ (r) be-
b
ing the same quantity for the free electron with kinetic energy at infinite distance
from the atomic nucleus.
Substituting Eq. (A24.4) into Eq. (A24.3), and taking into account Eq. (A24.2),
one obtains from Eq. (A24.1)
M = e20 (2L + 1)(2S + 1)(2J0 + 1)(2J + 1)(2j + 1)(2 b + 1)(2 + 1)
⎧ ⎫⎧ ⎫
⎨ L0 S0 J0 ⎬ ⎨ L0 S0 J0 ⎬
× (2jb + 1)(2jb + 1) b 12 jb 1
jb
⎩ ⎭⎩ b 2 ⎭
jb jb L S J L S J
! ! ! !
jb j 1 jb j 1 b 1 b 1
×
J J J0 J J J0 j jb 2 1
j jb 1
2
2
b 1 2
× I(n b , ) . (A24.6)
0 0 0
Equation (A24.6) can be substituted into Eq. (14.83) to find the dichroism coef-
ficient due to photoionization. The summation over , j, J in the denominator of
Eq. (14.83) can be evaluated by summing first over J and then over j. Applying
twice Eq. (2.39) one gets
⎧ ⎫
⎨ L0 S0 J0 ⎬2
M = e20 (2L + 1)(2S + 1)(2J0 + 1) (2jb + 1) b 1
jb
⎩ 2 ⎭
jJ jb L S J
2
b 1 2
× (2 + 1) I(n b , ) . (A24.7)
0 0 0
(K)
For the numerator, recalling the definition of wJJ (see Eq. (10.11)) and applying
twice Eq. (2.41), one gets
APPENDIX 865
M = e0 (2L + 1)(2S + 1)(2J0 + 1)(2 b + 1)
(K) 2
wJJ 3(2J + 1)
jJ
⎧ ⎫⎧ ⎫
⎨ L0 S0 J0 ⎬ ⎨ L0 S0 J0 ⎬
× (2jb + 1)(2jb + 1) b 12 jb 1
jb
⎩ ⎭⎩ b 2 ⎭
jb jb L S J L S J
! !
jb jb K jb jb K
×
b b 12 J J J0
! 2
1 1 1 K b 1
× (−1) 2 +J+J0 + (2 + 1)
b b 0 0 0
2
× I(n b , ) . (A24.8)
Equations (A24.7) and (A24.8), when substituted into Eq. (14.83), allow one to
express the dichroism coefficients due to photoionization as functions of the radial
integrals defined in Eq. (A24.5). Obviously, such integrals can be evaluated only
when the wavefunctions of the bound and free electronic states are known. To this
aim, one needs to solve the radial Schrödinger equation, which generally requires
the use of self-consistent Hartree-Fock methods. An approximate method that
can conveniently be used in many cases to express the radial integrals is the so-
called Quantum Defect Method developed by Seaton (1958) and Burgess and Seaton
(1960).
LIST OF TABLES
2.1 Algebraic formulae for d1MN (β) and d2MN (β) . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.1 Analytical expressions for the strengths of Zeeman components . . . . . . . 81
3.2 Transitions having large, null, or negative effective Landé factor . . . . . . 91
3.3 Barycentric moments of Zeeman components . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.4 Moments of Zeeman components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5 Level-crossings for the most common L-S terms . . . . . . . . . . . . . . . . . . . . . . 105
3.6 Multipole moments of the density matrix for atomic levels . . . . . . . . . . . . 125
3.7 Multipole moments of the density matrix for atomic terms . . . . . . . . . . . 127
5.1 Geometrical factors appearing in the propagation matrix . . . . . . . . . . . . . 159
. . . . . . . . . . . . . . . . . . . . . . . . 204
5.2 Explicit expression for the tensor Eqq (α, β, Ω)
. . . . . . . . . . . . . . . . . . . . . . . . . . . 206
5.3 Explicit expression for the tensor Tqq (i, Ω)
. . . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.4 Explicit expression for the tensor Iqq (ν, Ω)
5.5 Values of the symbol tK
P (i) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
. . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.6 Explicit expression for the tensor TQK (i, Ω)
5.7 Explicit expression for the tensor IQ
K . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
(ν, Ω)
9.1 Expressions for the Stokes parameters in the weak field limit . . . . . . . . . 404
9.2 Solar FeI lines having the largest circular polarization sensitivity index 406
9.3 Solar FeI lines having the largest linear polarization sensitivity index 408
9.4 Transitions having ḡ = 0 or Ḡ = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
9.5 Analytical expression for C−1 , the inverse of the propagation matrix 413
9.6 Angular factors relevant to the intense field solution . . . . . . . . . . . . . . . . . . 430
10.1 Values of wJ(K)J , WK (J , Ju ), (pQ )max , |pV |max for different transitions 515
u
10.2 Analytical expressions of the quantities WK (J , Ju ), (pQ )max , |pV |max 519
10.3 Atomic polarization induced by an unpolarized radiation beam . . . . . . . 543
10.4 Values of the quantities yJ(K) (K)
J , xJ J , XK (J, J ) for different transitions 575
10.5 Analytical expressions of the quantities XK (J, J ) . . . . . . . . . . . . . . . . . . . . 576
∞
10.6 Values of the factors [DK (LS)]fs for different terms . . . . . . . . . . . . . . . . . . 588
10.7 Values of the quantities WK (L Lu S, J Ju ) for different multiplets . . . . . . 593
∞
10.8 Values of the factors [DK (JI)]hfs for half-integral J values . . . . . . . . . . . 619
REFERENCES
Abramowitz, M., and Stegun, I.A.: 1965, Handbook of Mathematical Functions,
Dover Publications, New York.
Achmad, L., de Jager, C., and Nieuwenhuijzen, H.: 1991, Astron. Astrophys. 250,
445.
Akhiezer, A.I., and Berestetskii, V.B.: 1965, Quantum Electrodynamics, Wiley
Interscience, New York [etc.].
Alder, K.: 1952, Helv. Phys. Acta 25, 235.
Allen, C.W.: 1973, Astrophysical Quantities, 3rd ed., Athlone Press, University of
London, London.
Aller, L.H.: 1963, The Atmospheres of the Sun and Stars, 2nd ed., Ronald Press,
New York.
Angel, J.R.P., Borra, E.F., and Landstreet, J.D.: 1981, Astrophys. J. Suppl. 45,
457.
Arena, P., and Landi Degl’Innocenti, E.: 1982, Astron. Astrophys. Suppl. 48, 81.
Arnaud, J.: 1982, Astron. Astrophys. 112, 350.
Arnaud, J., and Newkirk, G.N. Jr.: 1987, Astron. Astrophys. 178, 263.
Athay, R.G., Querfeld, C.W., Smartt, R.N., Landi Degl’Innocenti, E., and Bom-
mier, V.: 1983, Solar Phys. 89, 3.
Auer, L.H.: 1967, Astrophys. J. 150, L53.
Auer, L.H., and Heasley, J.N.: 1978, Astron. Astrophys. 64, 67.
Auer, L.H., Heasley, J.N., and House, L.L.: 1977a, Astrophys. J. 216, 531.
Auer, L.H., Heasley, J.N., and House, L.L.: 1977b, Solar Phys. 55, 47.
Auvergne, M., Frisch, H., Frisch, U., Froeschlé, C., and Pouquet, A.: 1973, Astron.
Astrophys. 29, 93.
Avrett, E.H., and Hummer, D.G.: 1965, Monthly Not. R. A. S. 130, 295.
Babcock, H.W.: 1949, Astrophys. J. 110, 126.
Babcock, H.W.: 1953, Astrophys. J. 118, 387.
Back, E., and Landé, A.: 1925, Zeemaneffekt und Multiplettstruktur der Spek-
trallinien, Springer, Berlin.
Bagnulo, S.: 1993, Tesi di Laurea, Università di Firenze.
Baur, T.G., Elmore, D.E., Lee, R.H., Querfeld, C.W., and Rogers, S.R.: 1981,
Solar Phys. 70, 395.
Beckers, J.M.: 1969a, Solar Phys. 9, 372.
Beckers, J.M.: 1969b, A Table of Zeeman Multiplets, Air Force Cambridge Research
Laboratory.
Beckers, J.M., and Milkey, R.W.: 1975, Solar Phys. 43, 289.
Bellot Rubio, L.R., Ruiz Cobo, B., and Collados, M.: 1997, Astrophys. J. 478,
L45.
Bethe, H.A., and Salpeter, E.E.: 1957, Quantum Mechanics of One and Two Elec-
tron Atoms, Springer-Verlag, Berlin [etc.].
Bevington, P.R.: 1969, Data Reduction and Error Analysis for the Physical Sci-
ences, McGraw-Hill, New York [etc.].
870 REFERENCES
Biedenharn, L.C.: 1952, Oak Ridge National Laboratory, Report No.1098.
Biedenharn, L.C., Blatt, J.M., and Rose, M.E.: 1952, Rev. Mod. Phys. 24, 249.
Blackwell, D.E., Ibbetson, P.A., Petford, A.D., and Shallis, M.J.: 1979, Monthly
Not. R. A. S. 186, 633.
Blum, K.: 1981, Density Matrix Theory and Applications, Plenum Press, New York
[etc.].
Bommier, V.: 1977, Thèse de 3ème cycle, Université de Paris VI.
Bommier, V.: 1980, Astron. Astrophys. 87, 109.
Bommier, V.: 1997a, Astron. Astrophys. 328, 706.
Bommier, V.: 1997b, Astron. Astrophys. 328, 726.
Bommier, V., and Landi Degl’Innocenti, E.: 1996, Solar Phys. 164, 117.
Bommier, V., and Sahal-Bréchot, S.: 1978, Astron. Astrophys. 69, 57.
Bommier, V., Leroy, J.L., and Sahal-Bréchot, S.: 1981, Astron. Astrophys. 100,
231.
Bommier, V., Leroy, J.L., and Sahal-Bréchot, S.: 1986a, Astron. Astrophys. 156,
79.
Bommier, V., Leroy, J.L., and Sahal-Bréchot, S.: 1986b, Astron. Astrophys. 156,
90.
Bommier, V., Landi Degl’Innocenti, E., and Sahal-Bréchot, S.: 1991, Astron. As-
trophys. 244, 383.
Bommier, V., Landi Degl’Innocenti, E., Leroy, J.L., and Sahal-Bréchot, S.: 1994,
Solar Phys. 154, 231.
Born, M., and Wolf, E.: 1964, Principles of Optics, 2nd ed., Pergamon Press,
Oxford.
Boyarchuk, A.A., Efimov, Yu.S., and Stepanov, V.E.: 1961, Soviet Astron. 4, 766.
Brink, D.M., and Satchler, G.R.: 1968, Angular Momentum, 2nd ed., Clarendon
Press, Oxford.
Brissaud, A., and Frisch, U.: 1971, J. Quant. Spectrosc. Rad. Transfer 11, 1767.
Brix, P., and Kopfermann, H.: 1952, Hyperfeinstruktur der Atomterme und Atom-
linien, in Landolt-Börnstein, Zahlenwerte und Funktionen aus Physik, Chemie,
Astronomie, Geophysik und Technik, Sechste Auflage, I Band, 5 Teil, 1, Sprin-
ger-Verlag, Berlin.
Buckmaster, H.A.: 1964, Can. J. Phys. 42, 386.
Buckmaster, H.A.: 1966, Can. J. Phys. 44, 2525.
Burgess, A., and Seaton, M.J.: 1960, Monthly Not. R. A. S. 120, 121.
Caccin, B., Gomez, M.T., Marmolino, C., and Severino, G.: 1977, Astron. Astro-
phys. 54, 227.
Calamai, G., and Landi Degl’Innocenti, E.: 1983, Astron. Astrophys. Suppl. 53,
311.
Calamai, G., Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1975, As-
tron. Astrophys. 45, 297.
Canfield, R.C., de la Beaujardière, J.F., Fan, Y., Leka, K.D., McClimont, A.N.,
Metcalf, R.F., Mickey, D.L., Wülser, J.P., and Lites, B.W.: 1993, Astrophys.
J. 411, 362.
REFERENCES 871
Cannon, C.J.: 1985, The Transfer of Spectral Line Radiation, Cambridge Univer-
sity Press, Cambridge [etc.].
Casini, R., and Judge, P.G.: 1999, Astrophys. J. 522, 524.
Chandrasekhar, S.: 1947, Astrophys. J. 106, 145.
Chandrasekhar, S.: 1950, Radiative Transfer , Clarendon Press, Oxford.
Chandrasekhar, S., and Breen, F.H.: 1947, Astrophys. J. 105, 435.
Charvin, P.: 1965, Ann. Astrophysique 28, 877.
Clarke, D.: 1974, Appl. Opt. 13, 222.
Clarke, D., and Grainger, J.F.: 1971, Polarized Light and Optical Measurement ,
Pergamon Press, Oxford [etc.].
Cohen-Tannoudji, C., Diu, B., and Laloë, F.: 1977, Mécanique Quantique, Her-
mann, Paris.
Cohen-Tannoudji, C., Dupont-Roc, J., and Grynberg, G.: 1987, Photons et Ato-
mes, InterEditions, Editions du CNRS, Paris.
Cohen-Tannoudji, C., Dupont-Roc, J., and Grynberg, G.: 1988, Processus d’Inter-
action entre Photons et Atomes, InterEditions, Editions du CNRS, Paris.
Colegrove, F.D., Franken, P.A., Lewis, R.R., and Sands, H.R.: 1959, Phys. Rev.
Letters 3, 420.
Collatz, L.: 1966, The Numerical Treatment of Differential Equations, Springer-
Verlag, Berlin [etc.].
Condon, E.U., and Odabaşi, H.: 1980, Atomic Structure, Cambridge University
Press, Cambridge [etc.].
Condon, E.U., and Shortley, G.H.: 1935, The Theory of Atomic Spectra, Cambridge
University Press, Cambridge.
Corliss, C., and Sugar, J.: 1982, J. Phys. Chem. Ref. Data 11, No.1, 135.
Cuperman, S., Ofman, L., and Semel, M.: 1990, Astron. Astrophys. 227, 227.
Débarbat, S., Dumont, S., and Pecker, J.C.: 1970, Astron. Astrophys. 8, 231.
Delbouille, L., Neven, L., and Roland, G.: 1973, Photometric Atlas of the Solar
Spectrum from 3000 to 10000 Å, Institut d’Astrophysique de l’Université de
Liège.
Derouich, M., Sahal-Bréchot, S., Barklem, P.S., and O’Mara, B.J.: 2003, Astron.
Astrophys. 404, 763.
Dirac, P.A.M.: 1935, The Principles of Quantum Mechanics, Clarendon Press,
Oxford.
Dolginov, A.Z., and Pavlov, G.G.: 1972, Soviet Astron. 16, 450.
Domke, H.: 1969, Astrofizika 5, 525.
Domke, H.: 1971, Astrofizika 7, 39.
Edmonds, A.R.: 1957, Angular Momentum in Quantum Mechanics, Princeton
University Press, Princeton, New Jersey.
Elliott, J.P., and Flowers, B.H.: 1955, Proc. Roy. Soc. A 229, 536.
Faddeeva, V.N., and Terent’ev, N.M.: 1961, Tables of Values of the Function w(z)
for Complex Argument, Pergamon Press, New York.
Falkoff, D.L., Colladay, G.S., and Sells, R.E.: 1952, Can. J. Phys. 30, 253.
Fano, U.: 1949, J. Opt. Soc. Am. 39, 859.
Faulstich, R.: 1980, J. Quant. Spectrosc. Rad. Transfer 24, 229.
872 REFERENCES
Faurobert, M.: 1987, Astron. Astrophys. 178, 269.
Faurobert, M.: 1988, Astron. Astrophys. 194, 268.
Faurobert-Scholl, M.: 1991, Astron. Astrophys. 246, 469.
Faurobert-Scholl, M.: 1992, Astron. Astrophys. 258, 521.
Faurobert-Scholl, M., and Frisch, H.: 1989, Astron. Astrophys. 219, 338.
Feautrier, P.: 1964, Comptes Rendues Acad. Sci. Paris 258, 3189.
Fermi, E.: 1932, Rev. Mod. Phys. 4, 87.
Figger, H., and Walther, H.: 1974, Z. Phys. 267, 1.
Fluri, D.M., and Stenflo, J.O.: 1999, Astron. Astrophys. 341, 902.
Franken, P.A.: 1961, Phys. Rev. 121, 508.
Frazier, E.N., and Stenflo, J.O.: 1972, Solar Phys. 27, 330.
Frisch, H.: 1998, Astron. Astrophys. 338, 683.
Frisch, U., and Frisch, H.: 1975, Monthly Not. R. A. S. 173, 167.
Fuller, G.H., and Cohen, V.W.: 1969, in K. Way (ed.), Nuclear Data Tables,
Vol. A5, Academic Press, New York, p. 433.
Gail, H.P., Sedlmayr, E., and Traving, G.: 1980, J. Quant. Spectrosc. Rad. Trans-
fer 23, 267.
Gantmacher, F.R.: 1966, Théorie des Matrices, Dunod, Paris.
Gingerich, O., Noyes, R.W., Kalkofen, W., and Cuny, Y.: 1971, Solar Phys. 18,
347.
Glauber, R.J.: 1964, in C. de Witt, A. Blandin, and C. Cohen-Tannoudji (eds.),
Quantum Optics and Electronics, Gordon and Breach, New York, p. 63.
Gradshteyn, I.S., Ryzhik, I.M.: 1965, Tables of Integrals, Series, and Products,
Academic Press, New York [etc.].
Griem, H.: 1974, Spectral Line Broadening by Plasmas, Academic Press, New York
[etc].
Grossmann-Doerth, U., Larsson, B., and Solanki, S.K.: 1988a, Astron. Astrophys.
204, 266.
Grossmann-Doerth, U., Schüssler, M., and Solanki, S.K.: 1988b, Astron. Astro-
phys. 206, L37.
Hagyard, M.J., West, E.A., and Cumings, N.P.: 1977, Solar Phys. 53, 3.
Hagyard, M.J., Cumings, N.P., West, E.A., and Smith, J.E.: 1982, Solar Phys. 80,
33.
Hale, G.E.: 1908, Astrophys. J. 28, 315.
Hamilton, D.R.: 1947, Astrophys. J. 106, 457.
Hanle, W.: 1924, Z. Phys. 30, 93.
Hartmann, W.: 1970, Z. Phys. 240, 323.
Harvey, J.W.: 1969, Ph.D. Thesis, University of Colorado.
Harvey, J.W.: 1973, Solar Phys. 28, 9.
Harvey, J.W., and Livingston, W.: 1969, Solar Phys. 10, 283.
Harvey, J., Livingston, W., and Slaughter, C.: 1972, in Line Formation in the
Presence of Magnetic Fields, High Altitude Observatory, NCAR, Boulder, Col-
orado, p. 227.
Hauge, P.S.: 1976, in W.L. Hyde and R.M.A. Azzam (eds.), Polarized Light, Proc.
Soc. Photo-Opt. Instr. Eng. 88, p. 3.
REFERENCES 873
Heading, J.: 1958, Matrix Theory for Physicists, Longmans, Green and Co., Lon-
don [etc.].
Heitler, W.: 1954, The Quantum Theory of Radiation, 3rd ed., Clarendon Press,
Oxford.
Hopf, E.: 1934, Mathematical Problems of Radiative Equilibrium, Cambridge
Tracts, No.31.
House, L.L.: 1970a, J. Quant. Spectrosc. Rad. Transfer 10, 909.
House, L.L.: 1970b, J. Quant. Spectrosc. Rad. Transfer 10, 1171.
House, L.L.: 1971, J. Quant. Spectrosc. Rad. Transfer 11, 367.
Howard, R., and Stenflo, J.O.: 1972, Solar Phys. 22, 402.
Hui, A.K., Armstrong, B.H., and Wray, A.A.: 1978, J. Quant. Spectrosc. Rad.
Transfer 19, 509.
Humlicek, J.: 1982, J. Quant. Spectrosc. Rad. Transfer 27, 437.
Hummer, D.G.: 1962, Monthly Not. R. A. S. 125, 21.
Illing, R.M.E., Landman, D.A., and Mickey, D.L.: 1975, Astron. Astrophys. 41,
183.
Ivanov, V.V.: 1973, Transfer of Radiation in Spectral Lines, National Bureau of
Standards Spec. Publ. No.385, Govt. Printing Office, Whashington D.C.
Ivanov, V.V.: 1990, Soviet Astron. 34, 621.
Ivanov, V.V., Grachev, S.I., and Loskutov, V.M.: 1997, Astron. Astrophys. 321,
968.
Jackson, J.D.: 1962, Classical Electrodynamics, John Wiley & Sons, New York
[etc.].
Jahn, H.A., and Hope, J.: 1954, Phys. Rev. 93, 318.
Jefferies, J.T.: 1968, Spectral Line Formation, Blaisdell, Waltham, Mass. [etc.].
Jefferies, J.T., Lites, B.W., and Skumanich, A.: 1989, Astrophys. J. 343, 920.
Jones, R.C.: 1941, J. Opt. Soc. Am. 31, 488.
Katz, J.M.: 1971, Solar Phys. 20, 362.
Kawakami, H.: 1983, Publ. Astron. Soc. Japan 35, 459.
Kemp, J.C.: 1969, J. Opt. Soc. Am. 59, 950.
Kemp, J.C.: 1970, Astrophys. J. 162, 169.
Kemp, J.C., Henson, G.D., Barbour, M.S., Kraus, D.J., and Collins II, G.W.: 1983,
Astrophys. J. 273, L85.
Kemp, J.C., Macek, J.H., and Nehring, F.W.: 1984, Astrophys. J. 278, 863.
Kerkeni, B.: 2002, Astron. Astrophys. 390, 783.
Kerkeni, B., Spielfiedel, A., and Feautrier, N.: 2003, Astron. Astrophys. 402, 5.
de Kertanguy, A.: 1998, Astron. Astrophys. 333, 1130.
Kjeldseth-Moe, O.: 1968, Solar Phys. 4, 267.
Knölker, M., Schüssler, M., and Weisshaar, E.: 1988, Astron. Astrophys. 194, 257.
Kopfermann, H.: 1958, Nuclear Moments, Academic Press, New York.
Kourganoff, V., and Busbridge, I.W.: 1952, Basic Methods in Transfer Problems,
Clarendon Press, Oxford.
Kponou, A., Hughes, V.W., Johnson, C.E., Lewis, S.A., and Pichanik, F.M.J.:
1971, Phys. Rev. Letters 26, 1613.
Kramers, H.A.: 1924, Nature 113, 783.
874 REFERENCES
Krebs, K., and Winckler, R.: 1960, Z. Phys. 160, 320.
Lamb, F.K., and Sutherland, P.G.: 1974, in C. J. Hansen (ed.), Physics of Dense
Matter , IAU Symp. 53, Reidel, Dordrecht, p. 265.
Lamb, F.K., and ter Haar, D.: 1971, Physics Reports 2, 253.
Landi Degl’Innocenti, E.: 1975, Astron. Astrophys. 45, 269.
Landi Degl’Innocenti, E.: 1976, Astron. Astrophys. Suppl. 25, 379.
Landi Degl’Innocenti, E.: 1978a, Astron. Astrophys. Suppl. 33, 157.
Landi Degl’Innocenti, E.: 1978b, Astron. Astrophys. 66, 119.
Landi Degl’Innocenti, E.: 1979a, Solar Phys. 63, 237.
Landi Degl’Innocenti, E.: 1979b, Monthly Not. R. A. S. 186, 369.
Landi Degl’Innocenti, E.: 1982a, Solar Phys. 77, 285.
Landi Degl’Innocenti, E.: 1982b, Solar Phys. 79, 291.
Landi Degl’Innocenti, E.: 1983, Solar Phys. 85, 3.
Landi Degl’Innocenti, E.: 1984, Solar Phys. 91, 1.
Landi Degl’Innocenti, E.: 1985a, Solar Phys. 99, 1.
Landi Degl’Innocenti, E.: 1985b, in H.U. Schmidt (ed.), Theoretical Problems in
High-Resolution Solar Physics, Max-Planck-Institute für Physik und Astro-
physik, München, p. 162.
Landi Degl’Innocenti, E.: 1985c, Solar Phys. 102, 1.
Landi Degl’Innocenti, E.: 1987, in W. Kalkofen (ed.), Numerical Radiative Trans-
fer , Cambridge University Press, Cambridge, p. 265.
Landi Degl’Innocenti, E.: 1990, in V. Ruždjak and E. Tandberg-Hanssen (eds.),
Dynamics of Quiescent Prominences, IAU Coll. 117, Lecture Notes in Physics
363, 206.
Landi Degl’Innocenti, E.: 1992, in F. Sánchez, M. Collados, and M. Vázquez
(eds.), Solar Observations: Techniques and Interpretation, Cambridge Uni-
versity Press, Cambridge, p. 71.
Landi Degl’Innocenti, E.: 1994, in R.J. Rutten and C.J. Schrijver (eds.), Solar
Surface Magnetism, NATO ASI Ser. C 433, Kluwer, Dordrecht, p. 29.
Landi Degl’Innocenti, E., and Bommier, V.: 1993, Astrophys. J. 413, L49.
Landi Degl’Innocenti, E., and Bommier, V.: 1994, Astron. Astrophys. 284, 865.
Landi Degl’Innocenti, E., and Calamai, G.: 1982, Astron. Astrophys. Suppl. 49,
677.
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1972, Solar Phys. 27, 319
(see erratum in Solar Phys. 29, 528).
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1973, Solar Phys. 31,
299.
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1975, Il Nuovo Cimento
27 B, 134.
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1977, Astron. Astrophys.
56, 111.
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1981, Il Nuovo Cimento
62 B, 1.
Landi Degl’Innocenti, E., and Landi Degl’Innocenti, M.: 1985, Solar Phys. 97,
239.
REFERENCES 875
Landi Degl’Innocenti, M., and Landi Degl’Innocenti, E.: 1988, Astron. Astrophys.
192, 374.
Landi Degl’Innocenti, E., and Landolfi, M.: 1983, Solar Phys. 87, 221.
Landi Degl’Innocenti, M., Landolfi, M., and Landi Degl’Innocenti, E.: 1976, Il
Nuovo Cimento 35 B, 117.
Landi Degl’Innocenti, E., Bommier, V., and Sahal-Bréchot, S.: 1987, Astron. As-
trophys. 186, 335.
Landi Degl’Innocenti, E., Landi Degl’Innocenti, M., and Landolfi, M.: 1988, As-
tron. Astrophys. 204, 133.
Landi Degl’Innocenti, E., Bommier, V., and Sahal-Bréchot, S.: 1990, Astron. As-
trophys. 235, 459.
Landi Degl’Innocenti, E., Bommier, V., and Sahal-Bréchot, S.: 1991, Astron. As-
trophys. 244, 391.
Landi Degl’Innocenti, E., Landi Degl’Innocenti, M., and Landolfi, M.: 1997, in N.
Mein and S. Sahal-Bréchot (eds.), Forum THEMIS , Publications de l’Obser-
vatoire de Paris, p. 59.
Landolfi, M.: 1987, Solar Phys. 109, 287.
Landolfi, M., and Landi Degl’Innocenti, E.: 1982, Solar Phys. 78, 355.
Landolfi, M., and Landi Degl’Innocenti, E.: 1985, Solar Phys. 98, 53.
Landolfi, M., and Landi Degl’Innocenti, E.: 1986, Astron. Astrophys. 167, 200.
Landolfi, M., and Landi Degl’Innocenti, E.: 1996, Solar Phys. 164, 191.
Landolfi, M., Landi Degl’Innocenti, E., Arena, P.: 1984, Solar Phys. 93, 269.
Lehmann, J.C.: 1964, J. Physique 25, 809.
Lenz, W.: 1933, Z. Phys. 80, 423.
Leroy, J.L.: 1962, Ann. Astrophysique 25, 127.
Leroy, J.L.: 1972, Astron. Astrophys. 19, 287.
Leroy, J.L., Ratier, G., and Bommier, V.: 1977, Astron. Astrophys. 54, 811.
Lin, H., Penn, M.J., and Kuhn, J.R.: 1998, Astrophys. J. 493, 978.
Lin, H., Penn, M.J., and Tomczyk, S.: 2000, Astrophys. J. 541, L83.
Lites, B.W., and Skumanich, A.: 1990, Astrophys. J. 348, 747.
Lites, B.W., Low, B.C., Martı́nez Pillet, V., Seagraves, P., Skumanich, A., Frank,
Z.A., Shine, R.A., and Tsuneta, S.: 1995, Astrophys. J. 446, 887.
Lites, B.W., Leka, K.D., Skumanich, A., Martı́nez Pillet, V., and Shimizu, T.:
1996, Astrophys. J. 460, 1019.
Lorentz, H.A.: 1915, The Theory of Electrons, 2nd ed., Dover Publications, New
York.
Loudon, R.: 1983, The Quantum Theory of Light, 2nd ed., Clarendon Press, Ox-
ford.
Louisell, W.H.: 1973, Quantum Statistical Properties of Radiation, John Wiley
& Sons, New York [etc.].
Lyot, B.: 1948, Comptes Rendues Acad. Sci. Paris 226, 25.
Magain, P.: 1986, Astron. Astrophys. 163, 135.
Makita, M.: 1986, Solar Phys. 106, 269.
Manso Sainz, R.: 2002, Ph.D. Thesis, Universidad de La Laguna.
876 REFERENCES
Manso Sainz, R., and Trujillo Bueno, J.: 2001, in M. Sigwarth (ed.), Advanced
Solar Polarimetry. Theory, Observations and Interpretation, ASP Conf. Ser.
236, p. 213.
Marquardt, D.W.: 1963, J. Soc. Ind. Appl. Math. 11, No.2, p. 431.
Martin, B.R.: 1971, Statistics for Physicists, Academic Press, London [etc.].
Mathys, G., and Stenflo, J.O.: 1987a, Astron. Astrophys. 171, 368.
Mathys, G., and Stenflo, J.O.: 1987b, Astron. Astrophys. Suppl. 67, 557.
Maxwell, J.C.: 1873, A Treatise on Electricity and Magnetism, Oxford University
Press, Oxford.
Mein, P.: 1971, Solar Phys. 20, 3.
Messiah, A.: 1961, Quantum Mechanics, North-Holland, Amsterdam [etc.].
Mickey, D.L.: 1973, Astrophys. J. 181, L19.
Mickey, D.L., and Orrall, F.Q.: 1974, Astron. Astrophys. 31, 179.
Mihalas, D.: 1978, Stellar Atmospheres, 2nd ed., W.H. Freeman, San Francisco.
Mitchell, A.C.J., and Zemansky, M.W.: 1934, Resonance Radiation and Excited
Atoms, Cambridge University Press, Cambridge.
Moore, C.E.: 1949, Atomic Energy Levels, Vol. I, National Bureau of Standards.
Moore, C.E.: 1952, Atomic Energy Levels, Vol. II, National Bureau of Standards.
Moore, C.E.: 1958, Atomic Energy Levels, Vol. III, National Bureau of Standards.
Mueller, H.: 1948, J. Opt. Soc. Am. 38, 661.
Nagendra, K.N.: 1988, Astrophys. J. 335, 269.
Nagendra, K.N.: 1994, Astrophys. J. 432, 274.
Nagendra, K.N., Frisch, H., and Faurobert-Scholl, M.: 1998, Astron. Astrophys.
332, 610.
Öhman, Y.: 1929, Monthly Not. R. A. S. 89, 479.
Omont, A.: 1965, J. Physique 26, 26.
Omont, A.: 1977, Progr. Quant. Electron. 5, 69.
Omont, A., Smith, E.W., Cooper, J.: 1972, Astrophys. J. 175, 185.
Ord-Smith, R.J.: 1954, Phys. Rev. 94, 1227.
Oxenius, J.: 1986, Kinetic Theory of Particles and Photons, Springer-Verlag,
Berlin [etc].
Pauli, W.: 1927, Z. Phys. 43, 601.
Pavlov, G.G., and Shibanov, Yu.A.: 1978, Soviet Astron. 22, 43.
Pecker, J.C.: 1999, Solar Phys. 184, 1.
Perrin, F.: 1942, J. Chem. Phys. 10, 415.
Pichanik, F.M.J., Swift, R.D., Johnson, C.E., and Hughes, V.W.: 1968, Phys. Rev.
169, 55.
Poincaré, H.: 1892, Théorie Mathématique de la Lumière, Gauthiers-Villars, Paris.
Prudnikov, A.P., Brychkov, Yu.A., and Marichev, O.I.: 1986, Integrals and Series,
Vol. I, Gordon and Breach, New York.
Querfeld, C.W., and Smartt, R.N.: 1984, Solar Phys. 91, 299.
Querfeld, C.W., Smartt, R.N., Bommier, V., Landi Degl’Innocenti, E., and House,
L.L.: 1985, Solar Phys. 96, 277.
Racah, G.: 1942, Phys. Rev. 62, 438.
Rachkovsky, D.N.: 1962a, Izv. Krymsk. Astrofiz. Obs. 27, 148.
REFERENCES 877
Rachkovsky, D.N.: 1962b, Izv. Krymsk. Astrofiz. Obs. 28, 259.
Ramsauer, J., Solanki, S.K., and Biémont, E.: 1995, Astron. Astrophys. Suppl.
113, 71.
Rayrole, J.: 1967, Ann. Astrophysique 30, 257.
Reader, J., and Sugar, J.: 1975, J. Phys. Chem. Ref. Data 4, No.2, 353.
Rees, D.E.: 1969, Solar Phys. 10, 268.
Rees, D.E.: 1978, Publ. Astron. Soc. Japan 30, 455.
Rees, D.E.: 1987, in W. Kalkofen (ed.), Numerical Radiative Transfer , Cambridge
University Press, Cambridge, p. 213.
Rees, D.E., and Murphy, G.A.: 1987, in W. Kalkofen (ed.), Numerical Radiative
Transfer , Cambridge University Press, Cambridge, p. 241.
Rees, D.E., and Saliba, G.: 1982, Astron. Astrophys. 115, 1.
Rees, D.E., Murphy, G.A., and Durrant, C.J.: 1989, Astrophys. J. 339, 1093.
Reichel, A.: 1968, J. Quant. Spectrosc. Rad. Transfer 8, 1601.
Riordan, J.: 1968, Combinatorial Identities, John Wiley & Sons, New York [etc.].
Robinson, R.D. Jr.: 1980, Astrophys. J. 239, 961.
Rotenberg, M., Bivins, R., Metropolis, N., and Wooten, J.K. Jr.: 1959, The 3-j
and 6-j Symbols, MIT, Cambridge, Mass.
Rüedi, I., Solanki, S.K., Livingston, W., and Harvey, J.: 1995, Astron. Astrophys.
Suppl. 113, 91.
Ruiz Cobo, B., and del Toro Iniesta, J.C.: 1992, Astrophys. J. 398, 375.
Ruiz Cobo, B., and del Toro Iniesta, J.C.: 1994, Astron. Astrophys. 283, 129.
Rust, D.M.: 1966, Ph.D. Thesis, University of Colorado.
Rybicki, G.B.: 1977, Astrophys. J. 213, 165.
Sahal-Bréchot, S.: 1974a, Astron. Astrophys. 32, 147.
Sahal-Bréchot, S.: 1974b, Astron. Astrophys. 36, 355.
Sahal-Bréchot, S.: 1977, Astrophys. J. 213, 887.
Sahal-Bréchot, S., Bommier, V., and Feautrier, N.: 1998, Astron. Astrophys. 340,
579.
Saito, R., and Morita, M.: 1955, Progr. Theor. Phys. 13, 540.
Saliba, G.: 1986, Ph.D. Thesis, University of Sidney.
Sánchez Almeida, J.: 1992, Solar Phys. 137, 1.
Sánchez Almeida, J.: 1997, Astrophys. J. 491, 993.
Sánchez Almeida, J., and Lites, B.W.: 1992, Astrophys. J. 398, 359.
Sánchez Almeida, J., and Lites, B.W.: 2000, Astrophys. J. 532, 1215.
Sánchez Almeida, J., and Vela Villahoz, E.: 1993, Astron. Astrophys. 280, 688.
Sánchez Almeida, J., Collados, M., and del Toro Iniesta, J.C.: 1989, Astron. As-
trophys. 222, 311.
Sánchez Almeida, J., Landi Degl’Innocenti, E., Martı́nez Pillet, V., and Lites,
B.W.: 1996, Astrophys. J. 466, 537.
Schiff, L.I.: 1949, Quantum Mechanics, McGraw-Hill, New York [etc.].
Schmidt, G.D.: 1987, Mem. S. A. It. 58, 77.
Seares, F.H.: 1913, Astrophys. J. 38, 99.
Seaton, M.J.: 1958, Monthly Not. R. A. S. 118, 504.
878 REFERENCES
Segrè, E.: 1976, Personaggi e Scoperte nella Fisica Contemporanea, Mondadori,
Milano.
Semel, M.: 1967, Ann. Astrophysique 30, 513.
Semel, M., and Skumanich, A.: 1998, Astron. Astrophys. 331, 383.
Sharp, W.T., et al.: 1954, Tables of Coefficients for Angular Distribution Analysis,
CRT-556, AECL No.97, Atomic Energy of Canada Ltd.
Shenstone, A.G., and Blair, H.A.: 1929, Phil. Mag. 8, 765.
Shu, F.H.: 1991, The Physics of Astrophysics. Vol. I:Radiation, University Science
Books, Mill Valley, California.
Shurcliff, W.A.: 1962, Polarized Light, Harvard University Press, Cambridge,
Mass.
Šidlichovsky, M.: 1976, Bull. Astron. Inst. Czech. 27, 71.
Simon, A.: 1954, Oak Ridge National Laboratory, Report No.1718.
Simon, A., Vandersluis, V.H., and Biedenharn, L.C.: 1954, Oak Ridge National
Laboratory, Report No.1679.
Sistla, G., and Harvey, J.W.: 1970, Solar Phys. 12, 66.
Skumanich, A., and Lites, B.W.: 1987, Astrophys. J. 140, 1120.
Skumanich, A., Lites, B.W., Martı́nez Pillet, V., and Seagraves, P.: 1997, Astro-
phys. J. Suppl. 110, 357.
Smith, K.: 1958, Supplement to a Table of Wigner 9-j Coefficients for Integral and
Half-integral Values of the Parameters, Argonne National Laboratory, Report
No.5860 I-II.
Smith, K., and Stevenson, J.W.: 1957, A Table of Wigner 9-j Coefficients for Inte-
gral and Half-integral Values of the Parameters, Argonne National Laboratory,
Report No.5776.
Sobel’man, I.I.: 1972, An Introduction to the Theory of Atomic Spectra, Pergamon
Press, Oxford.
Socas Navarro, H., and Trujillo Bueno, J.: 1997, Astrophys. J. 490, 383.
Socas Navarro, H., Trujillo Bueno, J., and Ruiz Cobo, B.: 2000, Astrophys. J.
530, 977.
Solanki, S.K.: 1993, Space Science Rev. 63, 1.
Solanki, S.K., and Montavon, C.A.P.: 1993, Astron. Astrophys. 275, 283.
Solanki, S.K., and Pahlke, K.D.: 1988, Astron. Astrophys. 201, 143.
Solanki, S.K., and Stenflo, J.O.: 1985, Astron. Astrophys. 148, 123.
Solanki, S.K., Biémont, E., and Mürset, U.: 1990, Astron. Astrophys. Suppl. 83,
307.
Staude, J.: 1969, Solar Phys. 8, 264.
Stenflo, J.O.: 1971, in R. Howard (ed.), Solar Magnetic Fields, IAU Symp. 43,
p. 101.
Stenflo, J.O.: 1973, Solar Phys. 32, 41.
Stenflo, J.O.: 1980, Astron. Astrophys. 84, 68.
Stenflo, J.O.: 1994, Solar Magnetic Fields, Kluwer, Dordrecht [etc].
Stenflo, J.O., and Lindegren, L.: 1977, Astron. Astrophys. 59, 367.
Stenflo, J.O., and Stenholm, L.: 1976, Astron. Astrophys. 46, 69.
REFERENCES 879
Stenflo, J.O., Twerenbold, D., and Harvey, J.W.: 1983a, Astron. Astrophys. Suppl.
52, 161.
Stenflo, J.O., Twerenbold, D., Harvey, J.W., and Brault, J.W.: 1983b, Astron.
Astrophys. Suppl. 54, 505.
Stenflo, J.O., Harvey, J.W., Brault, J.W., and Solanki, S.K.: 1984, Astron. Astro-
phys. 131, 33.
Stepanov, V.E.: 1958, Izv. Krymsk. Astrofiz. Obs. 19, 20.
Stokes, G.G.: 1852, Trans. Cambridge Phil. Soc. 9, 399.
Tam, A.C.: 1975, Phys. Rev. A 12, 539.
Tandberg-Hanssen, E.: 1995, The Nature of Solar Prominences, Kluwer, Dordrecht
[etc.].
Ten Bruggencate, P., and von Klüber, H.: 1939, Z. Astrophys. 18, 284.
del Toro Iniesta, J.C., Semel, M., Collados, M., and Sánchez Almeida, J.: 1990,
Astron. Astrophys. 227, 591.
Trujillo Bueno, J.: 1990, in F. Sanchez and M. Vazquez (eds.), New Windows to the
Universe, XI th European Meeting of the IAU, Cambridge University Press,
Cambridge, p. 119.
Trujillo Bueno, J.: 1999, in K.N. Nagendra and J.O. Stenflo (eds.), Solar Polariza-
tion, Kluwer, Dordrecht, p. 73.
Trujillo Bueno, J., and Fabiani Bendicho, P.: 1995, Astrophys. J. 455, 646.
Trujillo Bueno, J., and Landi Degl’Innocenti, E.: 1996, Solar Phys. 164, 135.
Trujillo Bueno, J., and Landi Degl’Innocenti, E.: 1997, Astrophys. J. 482, L183.
Trujillo Bueno, J., and Manso Sainz, R.: 1999, Astrophys. J. 516, 436.
Unno, W.: 1956, Publ. Astron. Soc. Japan 8, 108.
Unsöld, A.: 1955, Physik der Sternatmosphären, 2nd ed., Springer, Berlin [etc.].
Van Ballegooijen, A.A.: 1985, in M.J. Hagyard (ed.), Measurements of Solar Vector
Magnetic Fields, NASA Conf. Publ. 2374, p. 322.
Van Ballegooijen, A.A.: 1987, in W. Kalkofen (ed.), Numerical Radiative Transfer ,
Cambridge University Press, Cambridge, p. 279.
Varshalovich, D.A., Moskalev, A.N., and Khersonskii, V.K.: 1988, Quantum The-
ory of Angular Momentum, World Scientific, Singapore [etc.].
Vernazza, J.E., Avrett, E.H., and Loeser, R.: 1981, Astrophys. J. Suppl. 45, 635.
Walker, M.J.: 1954, American J. Phys. 22, 170.
Warner, B.: 1969, Observatory 89, 11.
Weisskopf, V.F.: 1932, Z. Phys. 75, 296.
Westendorp Plaza, C., del Toro Iniesta, J.C., Ruiz Cobo, B., Martı́nez Pillet, V.,
Lites, B.W., and Skumanich, A.: 1997, Nature 389, 47.
Wieder, I., and Lamb, W.E. Jr.: 1957, Phys. Rev. 107, 125.
Wiehr, E.: 1975, Astron. Astrophys. 38, 303.
Wiehr, E.: 1978, Astron. Astrophys. 67, 257.
Wiese, W.L., Smith, M.W., and Glennon, B.M.: 1966, Atomic Transition Proba-
bilities, Vol. I (NSRD-NBS4), U.S. Department of Commerce.
Wittmann, A.: 1971, Solar Phys. 20, 365.
Wittmann, A.: 1974, Solar Phys. 35, 11.
Zirin, H., and Severny, A.B.: 1961, Observatory 81, 155.
AUTHOR INDEX
Italic numbers refer to ‘et al.’ quotations. Multiple occurrences of an author’s name on the same
page are not indicated.
Abramowitz, M., 165, 166 Burgess, A., 865
Achmad, L., 456 Busbridge, I.W., 670
Akhiezer, A.I., 131
Alder, K., 39 Caccin, B., 456
Allen, C.W., 344, 677, 703, 713 Calamai, G., 468, 472
Aller, L.H., 344, 439, 462 Canfield, R.C., 662
Angel, J.R.P., 74 Cannon, C.J., 238
Arena, P., 415, 635 Casini, R., 716, 719
Armstrong, B.H., 171 Chandrasekhar, S., 517, 664, 668, 670, 777,
Arnaud, J., 719 789, 854
Athay, R.G., 713 Charvin, P., 719
Auer, L.H., 444, 445, 483, 635, 814 Clarke, D., 5, 365
Auvergne, M., 496, 500 Cohen, V.W., 621
Avrett, E.H., 437, 854 Cohen-Tannoudji, C., 241, 247, 256, 804
Colegrove, F.D., 607
Babcock, H.W., 462, 466, 626 Colladay, G.S., 39
Back, E., 96 Collados, M., 482, 654, 655, 821
Bagnulo, S., 472 Collatz, L., 440
Barbour, M.S., 775 Collins II, G.W., 775
Barklem, P.S., 343 Condon, E.U., 73, 76, 77, 99
Baur, T.G., 656 Cooper, J., 789
de la Beaujardière, J.F., 662 Corliss, C., 78
Beckers, J.M., 82, 90, 162, 440, 450 Cumings, N.P., 634
Bellot Rubio, L.R., 655 Cuny, Y., 437
Berestetskii, V.B., 131 Cuperman, S., 662
Bethe, H.A., 258
Bevington, P.R., 636 Débarbat, S., 782
Biedenharn, L.C., 39, 44 Delbouille, L., 628
Biémont, E., 658 Derouich, M., 343
Bivins, R., 39, 44 Dirac, P.A.M., 131, 132
Blackwell, D.E., 651 Diu, B., 241, 804
Blair, H.A., 90 Dolginov, A.Z., 505
Blatt, J.M., 44 Domke, H., 767
Blum, K., 145 Dumont, S., 782
Bommier, V., 238, 288, 308, 524, 603, 604, Dupont-Roc, J., 241, 247, 256
662, 680, 691, 703, 704, 710, 711, 713, 720, Durrant, C.J., 445, 457, 816
742, 752, 754, 755, 757, 761, 767, 789, 854
Born, M., 149, 150 Edmonds, A.R., 29, 35, 42, 56, 67
Borra, E.F., 74 Efimov, Yu.S., 462
Boyarchuk, A.A., 462 Elliott, J.P., 49
Brault, J.W., 656 Elmore, D.E., 656
Breen, F.H., 789
Brink, D.M., 29, 35, 39, 42, 67 Fabiani Bendicho, P., 655
Brissaud, A., 497 Faddeeva, V.N., 171
Brix, P., 111, 495 Falkoff, D.L., 39
Brychkov, Yu.A., 826 Fan, Y., 662
Buckmaster, H.A., 56 Fano, U., 145
882 AUTHOR INDEX
Faulstich, R., 504 Jackson, J.D., 84, 796
Faurobert, M., 754, 789 Jackson, S., 791
Feautrier, N., 343, 720 de Jager, C., 456
Feautrier, P., 444 Jahn, H.A., 49
Fermi, E., 245 Jefferies, J.T., 152, 162, 397, 854
Figger, H., 112, 621 Johnson, C.E., 713
Flowers, B.H., 49 Jones, R.C., 16
Fluri, D.M., 782 Judge, P.G., 716, 719
Frank, Z.A., 662
Franken, P.A., 607 Kalkofen, W., 437
Frazier, E.N., 652 Katz, J.M., 358
Frisch, H., 496, 500, 754, 789, 854, 855, 858 Kawakami, H., 490
Frisch, U., 496, 497, 500, 854, 855 Kemp, J.C., 15, 376, 607, 608, 775
Froeschlé, C., 496, 500 Kerkeni, B., 343
Fuller, G.H., 621 de Kertanguy, A., 720
Khersonskii, V.K., 29, 35, 39, 44, 47
Gail, H.P., 504 Kjeldseth-Moe, O., 358
Gantmacher, F.R., 351 von Klüber, H., 462
Gingerich, O., 437 Knölker, M., 657
Glauber, R.J., 251 Kopfermann, H., 110, 111, 495
Glennon, B.M., 713 Kourganoff, V., 670
Gomez, M.T., 456 Kponou, A., 713
Grachev, S.I., 754, 756 Kramers, H.A., 779
Gradshteyn, I.S., 825 Kraus, D.J., 775
Grainger, J.F., 5, 365 Krebs, K., 115
Griem, H., 384 Kuhn, J.R., 712
Grossmann-Doerth, U., 444, 455, 457, 657
Grynberg, G., 241, 247, 256 Laloë, F., 241, 804
Lamb, F.K., 343, 347, 376
ter Haar, D., 343, 347 Lamb, W.E. Jr., 713
Hagyard, M.J., 634 Landé, A., 96
Hale, G.E., 625 Landi Degl’Innocenti, E., 90, 92, 115, 153,
Hamilton, D.R., 517 162, 172, 185, 233, 234, 238, 308, 318, 351,
Hanle, W., 184 392, 397, 415, 440, 448, 450, 455, 468, 472,
Hartmann, W., 621 475, 488, 497, 510, 513, 521, 524, 548, 562,
Harvey, J.W., 90, 635, 646, 656, 658, 719 600, 607, 608, 625, 634, 635, 660, 661, 662,
Hauge, P.S., 18 680, 691, 710, 713, 742, 752, 754, 755, 757,
Heading, J., 358 761, 767, 768, 770, 775, 789, 854
Heasley, J.N., 444, 445, 483, 635 Landi Degl’Innocenti, M., 162, 172, 185, 238,
Heitler, W., 131 351, 392, 397, 450, 468, 472, 691, 775, 789
Henson, G.D., 775 Landman, D.A., 483
Hope, J., 49 Landolfi, M., 238, 318, 453, 455, 475, 548,
Hopf, E., 670 635, 691, 775, 789
House, L.L., 444, 445, 635, 713, 719 Landstreet, J.D., 74
Howard, R., 652 Larsson, B., 444, 455, 457
Hughes, V.W., 713 Lee, R.H., 656
Hui, A.K., 171 Lehmann, J.C., 608
Humlicek, J., 171 Leka, K.D., 660, 662
Hummer, D.G., 844, 854 Lenz, W., 221
Leroy, J.L., 468, 703, 711, 713, 782
Ibbetson, P.A., 651 Lewis, R.R., 607
Illing, R.M.E., 483 Lewis, S.A., 713
Ivanov, V.V., 754, 756, 854 Lin, H., 712, 720
Lindegren, L., 90, 406
AUTHOR INDEX 883
Lites, B.W., 152, 162, 397, 474, 487, 625, Ord-Smith, R.J., 49
635, 637, 653, 655, 660, 661, 662 Orrall, F.Q., 782
Livingston, W., 635, 646, 658 Oxenius, J., 693
Loeser, R., 437
Lorentz, H.A., 221 Pahlke, K.D., 482
Loskutov, V.M., 754, 756 Pauli, W., 7
Loudon, R., 384 Pavlov, G.G., 482, 505
Louisell, W.H., 245 Pecker, J.C., 775, 782
Low, B.C., 662 Penn, M.J., 712, 720
Lyot, B., 775 Perrin, F., 16
Petford, A.D., 651
Macek, J.H., 607, 608 Pichanik, F.M.J., 713
Magain, P., 456, 457 Poincaré, H., 26
Makita, M., 487 Pouquet, A., 496, 500
Manso Sainz, R., 740, 752, 754, 755, 774 Prudnikov, A.P., 826
Marichev, O.I., 826
Marmolino, C., 456 Querfeld, C.W., 656, 713, 719
Marquardt, D.W., 636
Martin, B.R., 500 Racah, G., 35, 36, 42, 67
Martı́nez Pillet, V., 625, 637, 655, 660, 661, Rachkovsky, D.N., 162, 414
662 Ramsauer, J., 658
Mathys, G., 95 Ratier, G., 703
Maxwell, J.C., 1 Rayrole, J., 639, 641, 643
McClimont, A.N., 662 Reader, J., 78
Mein, P., 450 Rees, D.E., 162, 444, 445, 446, 457, 754, 768,
Messiah, A., 7, 29, 35, 74, 238 789, 816
Metcalf, R.F., 662 Reichel, A., 171
Metropolis, N., 39, 44 Riordan, J., 826
Mickey, D.L., 483, 662, 719, 782 Robinson, R.D. Jr., 90
Mihalas, D., 237, 264, 270, 340, 350, 376, Rogers, S.R., 656
384, 439, 445, 446, 462, 568, 722, 737, 776, Roland, G., 628
854 Rose, M.E., 44
Milkey, R.W., 450 Rotenberg, M., 39, 44
Mitchell, A.C.J., 192, 534 Rüedi, I., 658
Montavon, C.A.P., 487 Ruiz Cobo, B., 452, 455, 655, 656
Moore, C.E., 78, 104, 596, 597, 616 Rust, D.M., 714
Morita, M., 39 Rybicki, G.B., 854
Moskalev, A.N., 29, 35, 39, 44, 47 Ryzhik, I.M., 825
Mueller, H., 364
Murphy, G.A., 444, 445, 446, 457, 816 Sahal-Bréchot, S., 288, 343, 524, 680, 710,
Mürset, U., 658 711, 713, 716, 719, 720, 742, 752, 754,
757, 761
Nagendra, K.N., 789 Saito, R., 39
Nehring, F.W., 607, 608 Saliba, G., 789
Neven, L., 628 Salpeter, E.E., 258
Newkirk, G.N. Jr., 719 Sánchez Almeida, J., 373, 409, 444, 482, 487,
Nieuwenhuijzen, H., 456 625, 654, 660, 661, 821
Noyes, R.W., 437 Sands, H.R., 607
Satchler, G.R., 29, 35, 39, 42, 67
Odabaşi, H., 73 Schiff, L.I., 7, 73
Ofman, L., 662 Schmidt, G.D., 74
Öhman, Y., 710 Schüssler, M., 657
O’Mara, B.J., 343 Seagraves, P., 637, 662
Omont, A., 123, 229, 789 Seares, F.H., 433
884 AUTHOR INDEX
Seaton, M.J., 865 Tsuneta, S., 662
Sedlmayr, E., 504 Twerenbold, D., 656
Segrè, E., 88
Sells, R.E., 39 Unno, W., 162, 414
Semel, M., 639, 640, 654, 662 Unsöld, A., 345, 384, 439
Severino, G., 456
Severny, A.B., 714 Van Ballegooijen, A.A., 351, 370, 449, 456
Shallis, M.J., 651 Vandersluis, V.H., 39, 44
Sharp, W.T., 44 Varshalovich, D.A., 29, 35, 39, 44, 47
Shenstone, A.G., 90 Vela Villahoz, E., 409
Shibanov, Yu.A., 482 Vernazza, J.E., 437
Shimizu, T., 660
Shine, R.A., 662 Walker, M.J., 16
Shortley, G.H., 73, 76, 77, 99 Walther, H., 112, 621
Shu, F.H., 248 Warner, B., 345
Shurcliff, W.A., 5, 13 Weisshaar, E., 657
Šidlichovsky, M., 358 Weisskopf, V.F., 221
Simon, A., 39, 44 West, E.A., 634
Sistla, G., 90 Westendorp Plaza, C., 655
Skumanich, A., 152, 162, 397, 474, 635, 637, Wieder, I., 713
653, 655, 660, 662 Wiehr, E., 782
Slaughter, C., 635 Wiese, W.L., 713
Smartt, R.N., 713, 719 Winckler, R., 115
Smith, E.W., 789 Wittmann, A., 440, 489
Smith, J.E., 634 Wolf, E., 149, 150
Smith, K., 47 Wooten, J.K. Jr., 39, 44
Smith, M.W., 713 Wray, A.A., 171
Sobel’man, I.I., 221 Wülser, J.P., 662
Socas Navarro, H., 655
Solanki, S.K., 90, 444, 455, 457, 482, 487, Zemansky, M.W., 192, 534
656, 657, 658 Zirin, H., 714
Spielfiedel, A., 343
Staude, J., 437, 446
Stegun, I.A., 165, 166
Stenflo, J.O., 90, 95, 238, 406, 505, 597, 625,
650, 652, 656, 754, 782, 789
Stenholm, L., 754
Stepanov, V.E., 428, 462
Stevenson, J.W., 47
Stokes, G.G., 16
Sugar, J., 78
Sutherland, P.G., 376
Swift, R.D., 713
Tam, A.C., 713
Tandberg-Hanssen, E., 710
Ten Bruggencate, P., 462
Terent’ev, N.M., 171
Tomczyk, S., 720
del Toro Iniesta, J.C., 452, 456, 482, 654,
655, 821
Traving, G., 504
Trujillo Bueno, J., 238, 655, 656, 740, 752,
754, 755, 768, 770, 774, 786
SUBJECT INDEX
absorption and dispersion profiles, 255–256 elastic, 333–334
absorption coefficient: exciting, 220
continuum, 376 inelastic, 333–334
line, frequency-integrated, 378 perturbing, 220
multiplet, frequency-integrated, 585 superelastic, 334
absorption matrix, 155 velocity-changing, 693
alignment, atomic, 129 comoving frame, 680
alignment-to-orientation conversion, 607–609, complete redistribution on velocities, 695–698
622–623 compressing atmosphere, 474
angular momentum: contribution functions, 456–459
commutation rules, 29 coronal magnetic field, diagnostics, 715–720
eigenvalues and eigenvectors, 29–32 Correspondence Principle, 49, 132, 137, 138,
anisotropy factor, 538 142, 798
comoving frame, 683, 685 creation operator, 134, 138–141
grey atmosphere, 670
monochromatic, 671–673 damping constant, 82, 156, 526, 590
Milne-Eddington atmosphere, 666–668 Dawson integral, 167
outer atmosphere, 676–677 density matrix:
plane-parallel atmosphere, 665 multipole moments, 122–130
source-function gradient effect, 667 Schwarz inequality, 117
surface effect, 667 velocity-space, 692
annihilation operator, 134, 138–141 density operator, 115–122
anti-level-crossing effect, 604–605, 622–623 radiation field, 144–145
antisymmetric tensor, 7 spherical tensor representation, 122–130
standard representation, 119
bisector technique (see inversion technique) depolarizing cross section, 344
Bohr: depolarizing factor:
frequency, 119 fine structure, 587
magneton, 74 hyperfine structure, 618
branching ratio, 534, 562 destruction operator, 134, 138–141
dichroism matrix, 155
Cartesian tensor, 60 dielectric:
center-of-gravity technique (see inversion constant, 148
technique) principal axes, 149
circular polarization: principal constants, 149, 156
handedness, 5 tensor, 149
sensitivity index, 405–406 differential saturation mechanism, 466–468
Clebsh-Gordan coefficients, 33–38 Milne-Eddington atmosphere, 468–473
coherence-relaxation rates, 260 dipole approximation, 245
coherence-transfer rates, 260 disambiguation problem, 661–662
coherences, atomic, 119–122 dispersion:
collisional rates: matrix, 155
depolarizing, 343–346 profile, 163–171
Einstein-Milne relation, 340 Doppler:
elastic, 341–346 brightening, 684
inelastic and superelastic, 334–341 dimming, 687
in the presence of a magnetic field, 346 width, 163
multipole components, 336, 337, 341, Dyson chronological product, 352
799–803
collisions: Einstein coefficients, 264
886 SUBJECT INDEX
J levels, 281–282 effect of stimulated emission, 555–557
L-S terms, 299 spectral details, 524–528
relation between J levels and L-S terms, with lower-level polarization, 545–551
314–315 two-level atom with hyperfine structure,
Einstein-Milne relation, 340 620–622
electric: two-term atom, 598–605
displacement, 148, 156 with velocity/density-matrix correlations,
polarization, 148, 156 730–735
susceptibility, 156 harmonic oscillator, quantization, 131–134
electromagnetic field: Helmoltz principle of reciprocity, 184
Hamiltonian, 139 Hopf:
mode, 137 equation, 670, 840–841
momentum and angular momentum, function, 670, 838–840
796–799 hyperfine structure, 110
occupation number, 139
quantization, 137–141 Illing model, 483
vacuum state, 139 impact approximation, 221
equivalent width, 461 index of refraction, 148, 150, 153, 157
Euler angles, 52 interaction Hamiltonian, 241–244
evolution operator, 351–353 electric-dipole, 244–247
analytical expressions, 354–357 magnetic-dipole, 249
purely dichroic media, 363 interaction picture, 239
purely dispersive media, 363 inversion technique:
reduced, 370–373, 812–813 based on MISMA model, 660–661
expanding atmosphere, 474 based on response functions (SIR), 655
exponential integral, 665 bisector, 639–644
center-of-gravity, 639–644
Faraday: effect of unresolved fields, 653–654
pulsation, 178 infrared lines, 658–660
rotation, 178 line-ratio, 650–652
Faraday-Voigt function, 163 peak and area asymmetry, 656–657
field-free approximation, 768 Unno-fit, 634–639
filling factor, 625 effect of unresolved fields, 652–653
flat-spectrum approximation, 257 irreducible spherical tensor, 61
flux-tube, 657
Franken effect, 605–607, 622–623 K tensor, 208
JQ
generalized: Kronecker symbol, 7
Boltzmann term, 693 Kubo-Anderson process, 497
√
-law, 754, 758, 767, 853–858
profile, 524–525, 829–835 lambda-iteration, 446
grey model atmosphere, 668 lambdameter, 640
Landé factor, 75–78
half-wave plate, 13 effective, 89–91
Hanle diagram, 191 for hyperfine structure, 111
Hanle effect: generalized:
characteristic parameter, 181–182, 520 fine structure, 305
classical, 179–184, 190–193, 215–219 hyperfine structure, 318–319
with collisions, 228–230 j-j coupling, 76–77
lower-level, 549, 556 L-S coupling, 76
outer solar atmosphere, 703–709 second-order effective, 407
solar prominences, 710–715 for turbulent fields, 506
two-level atom, 520–524 Landé interval rule, 99
effect of collisions, 532–535 Larmor frequency, 78
SUBJECT INDEX 887
level-crossing: parameter, 473
fine structure, 104 net linear polarization, 466–473
hyperfine structure, 112 parameters, 468
interferences, 603, 622 nine-j symbols, 45–49, 791–793
limb darkening, 674 non-LTE problem:
line depression, 456 complex atomic models, 769–774
line-of-sight velocity, sign convention, 162 of order 1.5, 740, 767–769
line-ratio technique (see inversion technique) of the second kind, 738–742
line strength: strong magnetic field, 760–765
J levels, 281 plane-parallel atmosphere, 765–767
L-S terms, 298 weak magnetic field, 742–750
relation between J levels and L-S terms, plane-parallel atmosphere, 750–757
314 turbulent, 757–760
linear polarization:
position angle, 25 orientation, atomic, 128
sensitivity index, 407–408 orientation-to-alignment conversion, 609
total, 25 oscillator strength, 382
longitudinal magnetograph (see
magnetograph) Paschen-Back effect, 97–110
long-wavelength approximation, 245 complete, 101
for hyperfine structure, 110–115
magnetic Hamiltonian, 73 incomplete, 101
magnetic intensification: Paschen-Back pattern, 104–110
mechanism, 460–462 Pauli spin matrices, 7, 199, 370
Milne-Eddington atmosphere, 462–466 photoionization, absorption coefficient,
parameter, 462 775–777, 863–865
magnetogram, 626 Planck function, 161
magnetograph: Rayleigh-Jeans approximation, 161
longitudinal, 626–629 Wien limit, 533
calibration constant, 627–628 Planck law, 140
effect of unresolved fields, 644–646 Poincaré sphere, 26–27, 177, 431
saturation, 628 Poisson-step process, 497
vector, 629–634 polarization:
effect of unresolved fields, 644–649 basic spherical tensors, 202–212
saturation, 633 continuum, 775–782
second calibration constant, 631–632 degree, 10
magneto-optical: ellipse, 2–5, 9
effects, 414, 488–491 impact, 220
undulation, 489 limb Fraunhofer spectrum, 783–787
Markov approximation, 256 tensor, 6–10, 22–24
Maxwell equations, 147–148 quantum operator, 142
Milne-Eddington model atmosphere, 411 polarization-free approximation, 768
moments: polarizer, 11–13
of frequency-integrated intensity, 668 population inversion, 563–567
of intensity, 664–665 selective, 566
Mueller matrices, 364–366 projection theorem, 70
multipolar kernels, 751, 850–853 prominence magnetic field, diagnostics,
multipole coupling coefficients, 750, 845–850 710–715
generalized, 765, 859–862 propagation matrix, 271–272
classical, 154, 157–160
net circular polarization, 473–488, 822–824 eigenvalues, 358, 803–809
∆B effect, 485–486 microturbulent magnetic field, 504–507
∆θ effect, 486–487 propagation tensor of the electric field, 153
∆χ effect, 487–488
888 SUBJECT INDEX
quarter-wave plate, 13 by diagonalization, 358–362
by the evolution operator, 351–352
Racah coefficients, 41 perturbative, 366–367
radiation field tensor: symmetry properties, 172–176
comoving frame, 681–687, 841–843 radiative transfer equation in a magnetic field,
irreducible, 208 375–381
reducible, 207 absorption and dispersion profiles, 383–385
radiative rates, 259–260, 263 boundary conditions, 390
conjugation relations, 261 effect of blends, 459–460
multi-level atom: fine-structured lines, 491–495
conjugation relations, 290–291 hyperfine-structured lines, 493–495
irreducible tensor representation, solution:
285–288 DELO method, 445–447, 818–820
no-coherence case, 292, 294 evolution operator, 409–411
selection rules, 326 Feautrier method, 444–445, 813–817
standard representation, 280–283 intense field, 428–432
multi-level atom with hyperfine structure: PEO method, 448–449
energy-eigenvector representation, 317 Runge-Kutta method, 440–444
irreducible tensor representation, Seares, 433–435
319–320 special field orientations, 422–426
selection rules, 328 special model atmospheres, 418–421
multi-term atom: Stepanov, 426–428
conjugation relations, 311–312 Unno, 414
energy-eigenvector representation, Unno-Rachkovsky, 411–417, 460
299–300 variable azimuth, 435–437
irreducible tensor representation, stochastic media, 495–507, 824–828
303–308 symmetry properties, 391–396, 494, 502
selection rules, 327–328 using different depth indicators, 387–389
rotation of the reference system, 330–331 weak field approximation, 397–409
radiative transfer coefficients: Raman effect, 560–562
multi-level atom: Rayleigh scattering, 779
irreducible tensor representation, recoupling coefficients, 40–41
288–290 redistribution function, 722, 731, 788–789,
no-coherence case, 293, 295 843–844
standard representation, 283–284 reduced statistical tensors, 539
multi-level atom with hyperfine structure: reference direction:
energy-eigenvector representation, 317 for Stokes parameters, 15
irreducible tensor representation, preferred, 174
320–321 regimes for spectral line polarization, 232–235
multi-term atom: resonance polarization:
energy-eigenvector representation, 301 outer solar atmosphere, 698–703
irreducible tensor representation, quenching, 534
309–311 two-level atom, 513–519
rotation of the reference system, 332–333 effect of collisions, 532–535
radiative transfer equation: effect of stimulated emission, 551–555
alternative form, 368–373 strong magnetic field, 528–532
classical derivation, 153–161 with lower-level polarization, 535–545
with collisions, 221–223 two-level atom with hyperfine structure,
eigenvectors, 359, 809–812 616–620
geometrical interpretation, 176–179 two-term atom, 584–589
in LTE, 273–274 spectral details, 589–597
magnetic-dipole transitions, 274–275 with lower-term polarization, 609–616
quantum derivation, 265–272 with velocity/density-matrix correlations,
solution: 720–730
SUBJECT INDEX 889
resonance scattering, classical, 179–184 quantum derivation, 252–260
response function, 451, 458 rotation of the reference system, 330
examples of analytical, 453 scalar case, 263–264
generalized, 455 two-level atom, 511
line-integrated, 455 two-term atom, 581–582
retarder, 13–15 step-function, 258
reversing layer model, 433 stepwise excitation, 557–560
rotation matrices, 53–60 Stokesmeter, 634
reduced, 54–57 Stokes parameters:
rotation operator, 50–53 addition theorem, 26
and photons, 27–28
scattering phase matrix: definition, 15–18
for polarization tensor, 182, 187–188, line-integrated, 455
197–199, 227 measurement, 18–21
for Stokes parameters, 183, 188–189, quantum operators, 143
200–202, 228 reference direction, 15
magnetic kernel, 213–215 relation with polarization tensor, 22–24
multipole components, 201 rotation of the reference direction, 25
Rayleigh, 183
two-level atom, 517 tK
P symbol, 201, 210
weak anisotropy approximation, 573, TQK tensor, 200–201, 211
574, 577, 578 Thomson scattering, 777–778
with magnetic field, 521 three-j symbols, 38–39, 791–793
with magnetic field and collisions, 533
two-level atom with hyperfine structure, Unno-fit (see inversion technique)
617, 621 unpolarized lower level approximation, 513, 552
two-term atom, 586 unpolarized lower term approximation, 583
with magnetic field, 600, 835–838
SIR technique (see inversion technique) Van Vleck angle, 189
six-j symbols, 42–44, 791–793 vector-coupling coefficients, 33–38
source function: vector magnetograph (see magnetograph)
continuum, 376 velocity-coherence approximation, 694–695
line, 380 velocity/density-matrix correlations, 691–698
irreducible components, 746 velocity diagram, 725
Spectroscopic Stability Principle, 110, 114, velocity-Hanle diagram, 732
321–325, 494, 587, 614 Voigt function, 163–171
spherical statistical tensors, 122–130
spherical tensor operator, 62–65 w (K) factor, 513
reduced matrix element, 67 WK factor, 516
spherical tensors for polarimetry, 202–212 wave:
statistical equilibrium equations: partially polarized, 10
magnetic-dipole transitions, 274–275 totally polarized, 10
multi-level atom: weak anisotropy approximation, 567–570
collisional rates, 342–343 two-level atom, 570–580
irreducible tensor representation, Weyl theorem, 59
284–285 Wigner-Eckart theorem, 67
no-coherence case, 292, 293 Wigner symbols, 38–39, 42–44, 45–49, 791–793
standard representation, 280
multi-level atom with hyperfine structure: Zeeman effect, 73–82
energy-eigenvector representation, 317 anomalous, 81
irreducible tensor representation, 317 classical theory, 82–88
multi-term atom: for hyperfine structure, 111
energy-eigenvector representation, 298 normal, 80
irreducible tensor representation, 303 with collisions, 227
890 SUBJECT INDEX
Zeeman pattern, 78–82, 89–96
minimal equivalent, 97
moments, 94–95
barycentric, 92–94
Zeeman splitting, 78–79
normalized, 163–165
Astrophysics and Space Science Library
Volume 302:Stellar Collapse, edited by Chris L. Fryer
Hardbound, ISBN 1-4020-1992-0, April 2004
Volume 301: Multiwavelength Cosmology, edited by Manolis Plionis
Hardbound, ISBN 1-4020-1971-8, March 2004
Volume 300:Scientific Detectors for Astronomy, edited by Paola Amico, James W.
Beletic, Jenna E. Beletic
Hardbound, ISBN 1-4020-1788-X, February 2004
Volume 299: Open Issues in Local Star Fomation, edited by Jacques Lépine, Jane
Gregorio-Hetem
Hardbound, ISBN 1-4020-1755-3, December 2003
Volume 298: Stellar Astrophysics - A Tribute to Helmut A. Abt, edited by K.S.
Cheng, Kam Ching Leung, T.P. Li
Hardbound, ISBN 1-4020-1683-2, November 2003
Volume 297: Radiation Hazard in Space, by Leonty I. Miroshnichenko
Hardbound, ISBN 1-4020-1538-0, September 2003
Volume 296: Organizations and Strategies in Astronomy, volume 4, edited by André
Heck
Hardbound, ISBN 1-4020-1526-7, October 2003
Volume 295: Integrable Problems of Celestial Mechanics in Spaces of Constant
Curvature, by T.G. Vozmischeva
Hardbound, ISBN 1-4020-1521-6, October 2003
Volume 294: An Introduction to Plasma Astrophysics and Magnetohydrodynamics, by
Marcel Goossens
Hardbound, ISBN 1-4020-1429-5, August 2003
Paperback, ISBN 1-4020-1433-3, August 2003
Volume 293: Physics of the Solar System, by Bruno Bertotti, Paolo Farinella, David
Vokrouhlický
Hardbound, ISBN 1-4020-1428-7, August 2003
Paperback, ISBN 1-4020-1509-7, August 2003
Volume 292: Whatever Shines Should Be Observed, by Susan M.P. McKenna-Lawlor
Hardbound, ISBN 1-4020-1424-4, September 2003
Volume 291: Dynamical Systems and Cosmology, by Alan Coley
Hardbound, ISBN 1-4020-1403-1, November 2003
Volume 290: Astronomy Communication, edited by André Heck, Claus Madsen
Hardbound, ISBN 1-4020-1345-0, July 2003
Volume 287/8/9: The Future of Small Telescopes in the New Millennium, edited by
Terry D. Oswalt
Hardbound Set only of 3 volumes, ISBN 1-4020-0951-8, July 2003
Volume 286: Searching the Heavens and the Earth: The History of Jesuit
Observatories, by Agustín Udías
Hardbound, ISBN 1-4020-1189-X, October 2003
Volume 285: Information Handling in Astronomy - Historical Vistas, edited by André
Heck
Hardbound, ISBN 1-4020-1178-4, March 2003
Volume 284: Light Pollution: The Global View, edited by Hugo E. Schwarz
Hardbound, ISBN 1-4020-1174-1, April 2003
Volume 283: Mass-Losing Pulsating Stars and Their Circumstellar Matter, edited by
Y. Nakada, M. Honma, M. Seki
Hardbound, ISBN 1-4020-1162-8, March 2003
Volume 282: Radio Recombination Lines, by M.A. Gordon, R.L. Sorochenko
Hardbound, ISBN 1-4020-1016-8, November 2002
Volume 281: The IGM/Galaxy Connection, edited by Jessica L. Rosenberg, Mary E.
Putman
Hardbound, ISBN 1-4020-1289-6, April 2003
Volume 280: Organizations and Strategies in Astronomy III, edited by André Heck
Hardbound, ISBN 1-4020-0812-0, September 2002
Volume 279: Plasma Astrophysics , Second Edition, by Arnold O. Benz
Hardbound, ISBN 1-4020-0695-0, July 2002
Volume 278: Exploring the Secrets of the Aurora, by Syun-Ichi Akasofu
Hardbound, ISBN 1-4020-0685-3, August 2002
Volume 277: The Sun and Space Weather, by Arnold Hanslmeier
Hardbound, ISBN 1-4020-0684-5, July 2002
Volume 276: Modern Theoretical and Observational Cosmology, edited by Manolis
Plionis, Spiros Cotsakis
Hardbound, ISBN 1-4020-0808-2, September 2002
Volume 275: History of Oriental Astronomy, edited by S.M. Razaullah Ansari
Hardbound, ISBN 1-4020-0657-8, December 2002
Volume 274: New Quests in Stellar Astrophysics: The Link Between Stars and
Cosmology, edited by Miguel Chávez, Alessandro Bressan, Alberto Buzzoni,Divakara
Mayya
Hardbound, ISBN 1-4020-0644-6, June 2002
Volume 273: Lunar Gravimetry, by Rune Floberghagen
Hardbound, ISBN 1-4020-0544-X, May 2002
Volume 272:Merging Processes in Galaxy Clusters, edited by L. Feretti, I.M. Gioia, G.
Giovannini
Hardbound, ISBN 1-4020-0531-8, May 2002
Volume 271: Astronomy-inspired Atomic and Molecular Physics, by A.R.P. Rau
Hardbound, ISBN 1-4020-0467-2, March 2002
Volume 270: Dayside and Polar Cap Aurora, by Per Even Sandholt, Herbert C. Carlson,
Alv Egeland
Hardbound, ISBN 1-4020-0447-8, July 2002
Volume 269: Mechanics of Turbulence of Multicomponent Gases, by Mikhail Ya.
Marov, Aleksander V. Kolesnichenko
Hardbound, ISBN 1-4020-0103-7, December 2001
Volume 268: Multielement System Design in Astronomy and Radio Science, by Lazarus
E. Kopilovich, Leonid G. Sodin
Hardbound, ISBN 1-4020-0069-3, November 2001
Volume 267: The Nature of Unidentified Galactic High-Energy Gamma-Ray Sources,
edited by Alberto Carramiñana, Olaf Reimer, David J. Thompson
Hardbound, ISBN 1-4020-0010-3, October 2001
Volume 266: Organizations and Strategies in Astronomy II, edited by André Heck
Hardbound, ISBN 0-7923-7172-0, October 2001
Volume 265: Post-AGB Objects as a Phase of Stellar Evolution, edited by R. Szczerba,
S.K. Górny
Hardbound, ISBN 0-7923-7145-3, July 2001
Volume 264: The Influence of Binaries on Stellar Population Studies, edited by Dany
Vanbeveren
Hardbound, ISBN 0-7923-7104-6, July 2001
Volume 262: Whistler Phenomena - Short Impulse Propagation, by Csaba Ferencz,
Orsolya E. Ferencz, Dániel Hamar, János Lichtenberger
Hardbound, ISBN 0-7923-6995-5, June 2001
Volume 261: Collisional Processes in the Solar System, edited by Mikhail Ya. Marov,
Hans Rickman
Hardbound, ISBN 0-7923-6946-7, May 2001
Volume 260: Solar Cosmic Rays, by Leonty I. Miroshnichenko
Hardbound, ISBN 0-7923-6928-9, May 2001
Volume 259: The Dynamic Sun, edited by Arnold Hanslmeier, Mauro Messerotti, Astrid
Veronig
Hardbound, ISBN 0-7923-6915-7, May 2001
Volume 258: Electrohydrodynamics in Dusty and Dirty Plasmas- Gravito-
Electrodynamics and EHD, by Hiroshi Kikuchi
Hardbound, ISBN 0-7923-6822-3, June 2001
Volume 257: Stellar Pulsation - Nonlinear Studies, edited by Mine Takeuti, Dimitar D.
Sasselov
Hardbound, ISBN 0-7923-6818-5, March 2001
Volume 256: Organizations and Strategies in Astronomy, edited by André Heck
Hardbound, ISBN 0-7923-6671-9, November 2000
Volume 255: The Evolution of the Milky Way- Stars versus Clusters, edited by
Francesca Matteucci, Franco Giovannelli
Hardbound, ISBN 0-7923-6679-4, January 2001
Volume 254: Stellar Astrophysics, edited by K.S. Cheng, Hoi Fung Chau, Kwing Lam
Chan, Kam Ching Leung
Hardbound, ISBN 0-7923-6659-X, November 2000
Volume 253: The Chemical Evolution of the Galaxy, by Francesca Matteucci
Paperback, ISBN 1-4020-1652-2, October 2003
Hardbound, ISBN 0-7923-6552-6, June 2001
Volume 252: Optical Detectors for Astronomy II, edited by Paola Amico, James W.
Beletic
Hardbound, ISBN 0-7923-6536-4, December 2000
Volume 251: Cosmic Plasma Physics, by Boris V. Somov
Hardbound, ISBN 0-7923-6512-7, September 2000
Volume 250: Information Handling in Astronomy, edited by André Heck
Hardbound, ISBN 0-7923-6494-5, October 2000
Volume 249: The Neutral Upper Atmosphere, by S.N. Ghosh
Hardbound, ISBN 0-7923-6434-1, July 2002
Volume 247: Large Scale Structure Formation, edited by Reza Mansouri, Robert
Brandenberger
Hardbound, ISBN 0-7923-6411-2, August 2000
Volume 246: The Legacy of J.C. Kapteyn, edited by Piet C. van der Kruit, Klaas van
Berkel
Paperback, ISBN 1-4020-0374-9, November 2001
Hardbound, ISBN 0-7923-6393-0, August 2000
Volume 245: Waves in Dusty Space Plasmas, by Frank Verheest
Paperback, ISBN 1-4020-0373-0, November 2001
Hardbound, ISBN 0-7923-6232-2, April 2000
Volume 244: The Universe, edited by Naresh Dadhich, Ajit Kembhavi
Hardbound, ISBN 0-7923-6210-1, August 2000
Volume 243: Solar Polarization, edited by K.N. Nagendra, Jan Olof Stenflo
Hardbound, ISBN 0-7923-5814-7, July 1999
Volume 242: Cosmic Perspectives in Space Physics, by Sukumar Biswas
Hardbound, ISBN 0-7923-5813-9, June 2000
Volume 241: Millimeter-Wave Astronomy: Molecular Chemistry & Physics in Space,
edited by W.F. Wall, Alberto Carramiñana, Luis Carrasco, P.F. Goldsmith
Hardbound, ISBN 0-7923-5581-4, May 1999
Volume 240: Numerical Astrophysics, edited by Shoken M. Miyama, Kohji
Tomisaka,Tomoyuki Hanawa
Hardbound, ISBN 0-7923-5566-0, March 1999
Volume 239: Motions in the Solar Atmosphere, edited by Arnold Hanslmeier, Mauro
Messerotti
Hardbound, ISBN 0-7923-5507-5, February 1999
Volume 238: Substorms-4, edited by S. Kokubun, Y. Kamide
Hardbound, ISBN 0-7923-5465-6, March 1999
Volume 237: Post-Hipparcos Cosmic Candles, edited by André Heck, Filippina Caputo
Hardbound, ISBN 0-7923-5348-X, December 1998
Volume 236: Laboratory Astrophysics and Space Research, edited by P. Ehrenfreund,
C. Krafft, H. Kochan, Valerio Pirronello
Hardbound, ISBN 0-7923-5338-2, December 1998
Missing volume numbers have not yet been published.
For further information about this book series we refer you to the following web site:
http://www.wkap.nl/prod/s/ASSL
To contact the Publishing Editor for new book proposals:
Dr. Harry (J.J.) Blom: [email protected]
You might also like
- (Holden-Day Series in Time Series Analysis.) Jenkins, Gwilym M. - Watts, Donald G. - Spectral Analysis and Its applications-Holden-Day (1968) PDFNo ratings yet(Holden-Day Series in Time Series Analysis.) Jenkins, Gwilym M. - Watts, Donald G. - Spectral Analysis and Its applications-Holden-Day (1968) PDF542 pages
- Vitalii Dugaev, Vladimir Litvinov - Modern Semiconductor Physics and Device Applications-CRC Press (2021)0% (1)Vitalii Dugaev, Vladimir Litvinov - Modern Semiconductor Physics and Device Applications-CRC Press (2021)399 pages
- Gardiner (Crispin W.) - Handbook of Stochastic Methods For Physics, Chemistry and The Natural Sciences Springer Series in Synergetics 13 1994No ratings yetGardiner (Crispin W.) - Handbook of Stochastic Methods For Physics, Chemistry and The Natural Sciences Springer Series in Synergetics 13 1994410 pages
- Debasis Kundu, Swagata Nandi Statistical Signal Processing Frequency Estimation 2012No ratings yetDebasis Kundu, Swagata Nandi Statistical Signal Processing Frequency Estimation 2012140 pages
- ZXR10 M6000-S V5.00.10 Product Description - Gemnet-Core Node in IPRAN SolutionNo ratings yetZXR10 M6000-S V5.00.10 Product Description - Gemnet-Core Node in IPRAN Solution95 pages
- Jacques Vanier, Cipriana Tomescu - Universe Dynamics - The Least Action Principle and Lagrange's Equations (2019, CRC Press) - Libgen - LiNo ratings yetJacques Vanier, Cipriana Tomescu - Universe Dynamics - The Least Action Principle and Lagrange's Equations (2019, CRC Press) - Libgen - Li185 pages
- Database Systems Concepts, Design and Applications by S. K. SinghNo ratings yetDatabase Systems Concepts, Design and Applications by S. K. Singh1,541 pages
- GlobalData - IP Edge Routing Landscape - July 2023No ratings yetGlobalData - IP Edge Routing Landscape - July 20235 pages
- BRKMPL-2253 - EVPN Deep Dive With IOS XR Configuration Examples For Service Provider Metro and Data CenterNo ratings yetBRKMPL-2253 - EVPN Deep Dive With IOS XR Configuration Examples For Service Provider Metro and Data Center136 pages
- Pranab K Sen - Julio M Singer - Large Sample Methods in Statistics (1994) - An Introduction With Applications (2017, CRC Press) - Libgen - LiNo ratings yetPranab K Sen - Julio M Singer - Large Sample Methods in Statistics (1994) - An Introduction With Applications (2017, CRC Press) - Libgen - Li395 pages
- MA6451-Probability and Random ProcessesNo ratings yetMA6451-Probability and Random Processes19 pages
- Introduction To Probability and Random Signals100% (9)Introduction To Probability and Random Signals139 pages
- SJ-20101203114016-003-ZXR10 M6000 (V1.00.30) Carrier-Class Router Hardware Manual - 326967100% (1)SJ-20101203114016-003-ZXR10 M6000 (V1.00.30) Carrier-Class Router Hardware Manual - 326967239 pages
- Evžen Kočenda - Alexandr Černý - Elements of Time Series Econometrics - An Applied Approach-Karolinum Press, Charles University (2017)No ratings yetEvžen Kočenda - Alexandr Černý - Elements of Time Series Econometrics - An Applied Approach-Karolinum Press, Charles University (2017)220 pages
- A Blueprint of Technology, Applications and Market Drivers Towards The Year 2030 and Performance Target AnalysisNo ratings yetA Blueprint of Technology, Applications and Market Drivers Towards The Year 2030 and Performance Target Analysis45 pages
- A Comprehensive Survey On Pretrained Foundation ModelsNo ratings yetA Comprehensive Survey On Pretrained Foundation Models97 pages
- CH. A. Charalambides, M.V. Koutras, N. Balakrishnan - Probability and Statistical Models With Applications-Chapman and Hall - CRC (2000)No ratings yetCH. A. Charalambides, M.V. Koutras, N. Balakrishnan - Probability and Statistical Models With Applications-Chapman and Hall - CRC (2000)609 pages
- EANTC MPLSSDNInterop2024 TestReport v1.3No ratings yetEANTC MPLSSDNInterop2024 TestReport v1.364 pages
- Koranga Bs Padaliya SK Nautiyal VK Special Functions and The100% (1)Koranga Bs Padaliya SK Nautiyal VK Special Functions and The124 pages
- Statistical Methods For Stochastic Differential Equations: Monographs On Statistics and Applied Probability 124No ratings yetStatistical Methods For Stochastic Differential Equations: Monographs On Statistics and Applied Probability 124498 pages
- SJ-20140721105958-012-ZXCTN 9000-E (V3.00.10) Carrier Class Multi-Service Packet-Based Platform Configuration Guide (MPLS)No ratings yetSJ-20140721105958-012-ZXCTN 9000-E (V3.00.10) Carrier Class Multi-Service Packet-Based Platform Configuration Guide (MPLS)410 pages
- (Series in Mathematical Analysis and Applications 9) Leszek Gasinski, Nikolaos S. Papageorgiou - Nonlinear Analysis-Chapman & Hall - CRC (2006)100% (1)(Series in Mathematical Analysis and Applications 9) Leszek Gasinski, Nikolaos S. Papageorgiou - Nonlinear Analysis-Chapman & Hall - CRC (2006)960 pages
- MSC Thesis - Computational Study of Electronic Band Structure of Twisted Bilayer GrapheneNo ratings yetMSC Thesis - Computational Study of Electronic Band Structure of Twisted Bilayer Graphene73 pages
- Astronomy: Principles and Practice: A E Roy, PHD, Fras, Frse, FbisNo ratings yetAstronomy: Principles and Practice: A E Roy, PHD, Fras, Frse, Fbis10 pages
- Baidyanath Basu, Tanuka Chattopadhyay, Sudhindra Nath Biswas - An Introduction To Astrophysics-PHI Learning Private Limited (2010)100% (4)Baidyanath Basu, Tanuka Chattopadhyay, Sudhindra Nath Biswas - An Introduction To Astrophysics-PHI Learning Private Limited (2010)598 pages
- E/B Decomposition of Finite Pixelized CMB MapsNo ratings yetE/B Decomposition of Finite Pixelized CMB Maps23 pages
- The Variable Sources in The Gaia Archive: Ecole Evry Schatzman 2023: Stellar Physics With GaiaNo ratings yetThe Variable Sources in The Gaia Archive: Ecole Evry Schatzman 2023: Stellar Physics With Gaia15 pages
- Quantum Mechanics Course ZeemansplittingNo ratings yetQuantum Mechanics Course Zeemansplitting29 pages
- Physical Mathematics 2nd Edition Cahill K - Download The Entire Ebook Instantly and Explore Every DetailNo ratings yetPhysical Mathematics 2nd Edition Cahill K - Download The Entire Ebook Instantly and Explore Every Detail63 pages
- S2634: Physique de La Matière Condensée & Nano-Objets: SuperconductivityNo ratings yetS2634: Physique de La Matière Condensée & Nano-Objets: Superconductivity24 pages
- American International University-Bangladesh (AIUB) : Summer 2019-2020 SEMESTER I. Course Core and TitleNo ratings yetAmerican International University-Bangladesh (AIUB) : Summer 2019-2020 SEMESTER I. Course Core and Title11 pages
- 6.641 Electromagnetic Fields, Forces, and Motion: Mit OpencoursewareNo ratings yet6.641 Electromagnetic Fields, Forces, and Motion: Mit Opencourseware8 pages
- Group Theory Lecture 2 (Dinesh Khattar)No ratings yetGroup Theory Lecture 2 (Dinesh Khattar)18 pages
- Quantum Stress-Energy at Timelike Boundaries: Testing A New beyond-CDM Parameter With Cosmological DataNo ratings yetQuantum Stress-Energy at Timelike Boundaries: Testing A New beyond-CDM Parameter With Cosmological Data25 pages
- Lecture Notes - AKR Centennial Lectures - Sunil MukhiNo ratings yetLecture Notes - AKR Centennial Lectures - Sunil Mukhi72 pages
- FX e X X X FX NX FX X X: The Fourier Series For in The IntervalNo ratings yetFX e X X X FX NX FX X X: The Fourier Series For in The Interval3 pages
- Angew Chem Int Ed - 2009 - Wang - Icon of Chemistry The Periodic System of Chemical Elements in The New CenturyNo ratings yetAngew Chem Int Ed - 2009 - Wang - Icon of Chemistry The Periodic System of Chemical Elements in The New Century12 pages
- Problem Set #4 - Mathematics in Population BiologyNo ratings yetProblem Set #4 - Mathematics in Population Biology14 pages
- Good Morning Everybody!! I Am Arindam Sarkar, Bsc. (Phy.), Mtech. (Textile Engg. - Iit Delhi) Prof. Rndr. D Avid Lukáš, CSCNo ratings yetGood Morning Everybody!! I Am Arindam Sarkar, Bsc. (Phy.), Mtech. (Textile Engg. - Iit Delhi) Prof. Rndr. D Avid Lukáš, CSC18 pages
- Witten Index of BMN Matrix Quantum MechanicsNo ratings yetWitten Index of BMN Matrix Quantum Mechanics22 pages
