Introduction To Relativistic Quantum Fields

Introduction to Relativistic Quantum Fields
Jan Smit
Institute for Theoretical Physics
University of Amsterdam
Valckenierstraat 65, 1018XE Amsterdam
The Netherlands
[email protected]
Lecture notes 2007
Contents
0.1
0.2
0.3
0.4
Why Field Theory? . . . .

These notes . . . . . . . .
Books . . . . . . . . . . .
Conventions and notation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 Classical fields
1.1 Maxwell field . . . . . . . . . . . . . . . .
1.2 Einstein field . . . . . . . . . . . . . . . .
1.3 Scalar field . . . . . . . . . . . . . . . . . .
1.4 Poincare group . . . . . . . . . . . . . . .
1.5 Action . . . . . . . . . . . . . . . . . . . .
1.6 Canonical formalism . . . . . . . . . . . .
1.7 Symmetries and Noethers theorem . . . .
1.8 Summary . . . . . . . . . . . . . . . . . .
1.9 Appendix: Functional derivative . . . . . .
1.10 Appendix: Rudiments of representations of
1.11 Problems . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
continuous groups
. . . . . . . . . . .
2 Quantized scalar fields

2.1 Canonical quantization . . . . . . . . . . . . .
2.2 Free field . . . . . . . . . . . . . . . . . . . . .
2.3 Renormalization of the cosmological constant
2.4 Perturbation theory . . . . . . . . . . . . . . .
2.5 Scattering . . . . . . . . . . . . . . . . . . . .
2.6 Decay . . . . . . . . . . . . . . . . . . . . . .
2.7 Representation of the Poincare group . . . . .
2.8 Where is the particle? . . . . . . . . . . . . .
2.9 Causality and locality . . . . . . . . . . . . . .
2.10 Classical field . . . . . . . . . . . . . . . . . .
2.11 Summary . . . . . . . . . . . . . . . . . . . .
2.12 Problems . . . . . . . . . . . . . . . . . . . . .
3
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
7
7
0
.
.
.
.
.
.
.
.
.
.
.
1
1
2
2
4
9
12
13
17
17
18
19
.
.
.
.
.
.
.
.
.
.
.
.
27
27
28
31
33
35
39
41
43
45
47
49
50
4
3 Path integral methods
3.1 Path integral for quantum mechanics
3.2 Regularization by discretization . . .
3.3 Imaginary time . . . . . . . . . . . .
3.4 External force technique . . . . . . .
3.5 Ground state expectation values . . .
3.6 Harmonic oscillator . . . . . . . . . .
3.7 Scalar field and Ising model . . . . .
3.8 Free scalar field . . . . . . . . . . . .
3.9 Summary . . . . . . . . . . . . . . .
3.10 Problems . . . . . . . . . . . . . . . .
CONTENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
59
59
60
63
65
68
70
73
75
77
77
4 Perturbation theory, diagrams and renormalization

4.1 Preparation . . . . . . . . . . . . . . . . . . . . . . .
4.2 Diagrams for 4 theory. . . . . . . . . . . . . . . . .
4.3 Diagrams in momentum space . . . . . . . . . . . . .
4.4 One-loop vertex functions . . . . . . . . . . . . . . .
4.5 Renormalization and counterterms . . . . . . . . . .
4.6 Recipe for scattering and decay amplitudes . . . . . .
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . .
4.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . .
4.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
81
81
83
89
91
96
100
102
103
104
.
.
.
.
.
.
.
.
107
107
113
116
118
119
121
124
125
.
.
.
.
.
.
.
.
.
127
127
129
134
141
145
147
152
153
155
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 Spinor fields
5.1 Spinors . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Action, Dirac equation and Noether charges . . . . . . .
5.3 Solutions of the Dirac equation . . . . . . . . . . . . . .
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 Appendix: Charge conjugation matrix C . . . . . . . . .
5.6 Appendix: Polarization spinors . . . . . . . . . . . . . .
5.7 Appendix: Traces of gamma matrices and other identities
5.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . .
6 Quantized spinor fields: fermions
6.1 Canonical quantization: wrong . . . . . . .
6.2 Canonical quantization: right . . . . . . .
6.3 Path integral quantization . . . . . . . . .
6.4 Non-relativistic fields and antiparticles . .
6.5 Yukawa models . . . . . . . . . . . . . . .
6.6 Scattering in the tree graph approximation
6.7 Summary . . . . . . . . . . . . . . . . . .
6.8 Appendix: More on spin and statistics . .
6.9 Appendix: Anticommuting variables . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
6.10 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

7 Quantized electromagnetic field
7.1 Quantization in the Coulomb gauge
7.2 Energy-momentum eigenstates . . .
7.3 Lorentz invariance . . . . . . . . .
7.4 Photons . . . . . . . . . . . . . . .
7.5 Path integral for the Maxwell field
7.6 Generalized covariant gauge . . . .
7.7 Ghosts . . . . . . . . . . . . . . . .
7.8 Summary . . . . . . . . . . . . . .
7.9 Appendix: Locality in the Coulomb
7.10 Problems . . . . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
gauge
. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
163
163
167
171
173
175
179
181
184
184
186
8 QED
8.1 Gauge invariance . . . . . . . . . . . . . .
8.2 Spinor electrodynamics . . . . . . . . . . .
8.3 Example: e + e+ + + scattering .
8.4 Magnetic moment of the electron . . . . .
8.5 Gauge-invariant non-relativistic reduction
8.6 Summary . . . . . . . . . . . . . . . . . .
8.7 Problems . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
189
189
193
196
201
203
205
206
.
.
.
.
.
.
.
211
211
213
217
223
228
230
234
.
.
.
.
.
239
239
242
243
246
247
9 Scattering
9.1 Cross section . . . . . . . . . . . . . . . . . . . . . .
9.2 Scattering amplitude . . . . . . . . . . . . . . . . . .
9.3 Fields, particles and poles . . . . . . . . . . . . . . .
9.4 Scattering amplitudes from correlation functions . . .
9.5 Decay revisited . . . . . . . . . . . . . . . . . . . . .
9.6 Example: the decays e + e and +
9.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . .
10 C, P, and T
10.1 Charge conjugation
10.2 Parity . . . . . . .
10.3 Time reversal . . .
10.4 CPT . . . . . . . .
10.5 Problems . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
0.1
Why Field Theory?
Quantum Field Theory is our basic framework for the description of particles
and their interactions. To the shortest distances which can be explored with current accelerators, which is about 1025 cm, a theory called the Extended Standard Model (ESM)1 provides an accurate description of hadrons, leptons, gauge
bosons. Field theory is also used in cosmology, e.g. combining general relativity
(geometrodynamics) with the ESM, with a scalar field added to incorporate
inflation.
The world is evidently quantal, but why fields? Classical field theory is local.
Interactions are described by differential equations at a point in space and time.
They only refer to the immediate neighborhood of that point through derivatives
of finite order, usually only up to second order. The equations referring to spacetime points in Amsterdam do not refer to what goes on in Paris. A typical
example is given by the the electromagnetic field interacting with electrons, which
are described by the Maxwell equations and the equations for the Lorentz forces
acting on the electrons. The particles create propagating electromagnetic fields
which influence in turn the particles. In the quantum theory the electromagnetic
field describes also particles, the photons, which can be created or annihilated
by the electrons. Action at a distance (in space and time) can be avoided in this
description. Locality is the space-time version of causality. The word causality
suggests also an temporal order of cause and effect, and this is impletemented by
retarded boundary conditions.
These ideas have been questioned from time to time, but alternative descriptions (such Feynman and Wheelers absorber theory) have not been able to elicit
the same intuitive appeal as field theory and have not been pursued very much.
As we will see, quantising fields leads to a description of arbitrarily many
identical particles, bosons or fermions (in three spatial dimensions). This description is elegant and practical, which is another reason for the success of field
theory. Moreover, there are phenomena which cannot be captured in terms of
particles, for which the field formulation is essential. These are typically situations where strong fields prevail, which occur in classical electrodynamics but
also in geometrodynamics (e.g. near black holes). Another example is quantum
chromodynamcs, where quarks and gluons are confined into hadrons, the protons,
neutrons, pions, glueballs, etc.
Field theory is based on the existence of space-time. The latter may perhaps
be explainable in terms of an underlying theory, such as M theory. It may take
some time before such an extension can be tested by experimental results.
1
By this we mean the renormalizable extension of the Standard Model (SM) to allow for
non-zero neutrino masses; a.k.a. the MSM.
0.2. THESE NOTES
0.2
These notes
These lecture notes present an introduction to relativistic quantum field theory,

leading to quantum electrodynamics in covariant gauges and opening the road to
the Standard Model. Our approach is based on the idea that fields are the basic
variables, which upon quantization lead (or may not lead) to particles. Other
authors (notably Veltman, Weinberg) follow an opposite line of thought: particles
are basic and fields are to be constructed in accordance with their properties and
general principles. This latter approach appears to be essentially perturbative.
We will briefly touch upon renormalization and the non-perturbative lattice
formulation in the case the scalar field. For gauge fields and fermions our presentation of the path integral stays at the formal level,2 which is sufficient for developing perturbation theory. In case of fermions, the complications one encounters
in truly non-perturbative (lattice) formulations are remarkable and expose the
cheating implied by staying at the formal level. Progress here is steady and can
be traced in the proceedings of the yearly International Symposium on Lattice
Field Theory.
0.3
Books
Lecture notes are no substitute for a book. The following books are refered to in
the text by name of authors:
Books on mathematical methods:
Jon Mathews and R.L. Walker, Mathematical Methods of Physics, Benjamin 1970.
R.B. Dingle, Asymptotic expansions, their derivation and interpretation,
Academic Press 1973.
M.J. Lighthill, Introduction to Fourier analysis and generalised functions,
Cambridge University Press 1958.
H.F. Jones, Groups, Representations and Physics (2nd edition), Institute
of Physics 2003.
Books on classical electrodynamics and general relativity:

J.D. Jackson, Classical Electrodynamics, Wiley 1975/1998.
2
Formal = jargon for pretty but mathematically imprecise.
CONTENTS
C.W. Misner, K.S. Thorn and J.A. Wheeler, Gravitation, Freeman 1973.
S. Weinberg, General Relativity and Cosmology, John Wiley and Sons 1972.
Books on relativistic QFT:

J.D. Bjorken and S.D. Drell, I: Relativistic Quantum Mechanics, McGrawHill (1964).
J.D. Bjorken and S.D. Drell, II: Relativistic Quantum Fields, McGraw-Hill
(1965).
L.S. Brown, Quantum Field Theory, Cambridge University Press 1992.
Ta-Pei Cheng and Ling-Fong Li, Gauge Theory of Elementary Particle
Physics, Oxford University Press (1984).
See also ibid, Problems and Solutions (2000).
R.P. Feynman, The reason for antiparticles, S. Weinberg, Towards the final
laws of physics, in Elementary particles and the Laws of Physics, The 1986
Dirac Memorial Lectures, Cambridge University Press 1987.
C. Itzykson and J.-B. Zuber, Quantum Field Theory, McGraw-Hill (1980).
M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field Theory, Perseus 1995.
Stefan Pokorski, Gauge Field Theories, 2nd edition, Cambridge University
Press 2000
P. Ramond, Field Theory: A Modern Primer (second edition), Addison
Wesley (1989).
L. Ryder, Quantum Field Theory, Cambridge University Press 1996.
G. Sterman, Introduction to Quantum field Theory, Cambridge University
Press 1993.
M. Veltman, Diagrammatica, Cambridge Lecture Notes in Physics, 1994.
S. Weinberg, The Quantum Theory of Fields, I: Fundamentals, Cambridge
University Press 1995.
S. Weinberg, The Quantum Theory of Fields, II: Modern Applications,
Cambridge University Press 1996.
0.3. BOOKS
S. Weinberg, The Quantum Theory of Fields, III, Cambridge University

Press 1999.
B. de Wit and J. Smith, Field Theory in Particle Physics I, North-Holland
(1986).
Books on quantum field theory and critical phenomena:

M. Le Bellac and G. Barton, Quantum and Statistical Field Theory, Clarendon 1992.
C. Itzykson and J-M. Drouffe, Statistical Field Theory I & II, Cambridge
University Press 1989.
G. Parisi, Statistical Field Theory, Perseus 1998.
J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon
1996.
The following books are specifically on lattice field theory:

M. Creutz, Quarks, Gluons and Lattices, Cambridge University Press (1983).
G. M
unster, I. Montvay, Quantum Fields on a Lattice, Cambridge University Press (1994).
H.J. Rothe, Introduction to Lattice Gauge Theories, World Scientific 1992
or later.
J. Smit, Introduction to Quantum Fields on a Lattice, Cambridge University Press, 2002.
See also the Proceedings of the yearly meetings Lattice XX.
Books on (specialized topics in) particle physics:

D.H. Perkins, Introduction to High Energy Physics, CUP (2000).
I.I. Bigi and A.I. Sanda, CP Violation, CUP (2000).
G. Branco, L. Lavoura and J. Silva, CP Violation, Oxford University Press
(1999).
10
CONTENTS
A book on the Casimir effect:

K.A. Milton, The Casimir effect, World Scientific 2001.
There are a number of useful lecture notes which are easily available, e.g. via the
internet, for example:
P. van Baal, A Course in Field Theory, Instituut-Lorentz,
http://www-lorentz.leidenuniv.nl/vanbaal/FTcourse.html
P.J. Mulders, Quantum Field Theory, ITP, Vrije Universiteit,
http://www.nat.vu.nl/ mulders/lectures.html
J. Smit, Introduction to Quantum Field Theory 1994/95, ITFA 1995,
http://staff.science.uva.nl/ jsmit/
(The approach taken here differs substantially from the present notes.)
CONTENTS
0.4
Conventions and notation
The following conventions will be used:

- h
= c = 1. The dimensions of various quantities are like [mass] = [energy]
= [momentum] = [(x)] = [length1 ] = [time1 ]. Actions are dimensionless. To convert to ordinary units use appropriate powers of h
and c. A
particularly useful combination is h
c = 197.3 MeV fm, where fm (femto
meter or Fermi) denotes the unit of length 1013 cm. For example a mass
m of 200 MeV corresponds to a length 1/m of about 1 fm.
- Rationalized Gauss (Lorentz-Heaviside) units for electromgnetism. The
unit of electromagnetic charge e 0.30 ( = e2 /4 1/137). The charge
of the electron is e.
- Minkowski metric
x = x , 11 = 22 = 33 = 00 = 1,
x0 = x0 , xk = xk , k = 1, 2, 3,
=
,
x
x2 = x x = x2 x20 , 2 = = 2 02 ,
px = pk xk + p0 x0 = pk xk p0 x0 = px p0 x0 ,
0123 = +1,
d4 x = dx0 dx1 dx2 dx3 , d4 p = dp0 dp1 dp2 dp3 .
(1)
(2)
(3)
(4)
(5)
(6)
(7)
The same metric is used in the books by Brown, Weinberg, and MisnerThorn-Wheeler. Many other authors (e.g. Bjorken and Drell, Peskin and
Schroeder) use the metric with 00 = +1.
- Dirac matrices:
{ , } = 2 ,
= i ,
= i 0 ,
5 = i 0 1 2 3 .
(8)
(i = [Bjorken&Drell], 5 = 5 [Bjorken&Drell].) Left and right handed

chiral projectors PL = (1 5 )/2, PR = (1 + 5 )/2.
- Lorentz invariant volume element for mass m,
dp =
d3 p
,
(2)3 2p0
p0 = E(p) =
p2 + m 2 .
(9)
Depending Ron the context, p0 can be an arbitrary variable (e.g. a dummy

variable in d4 p), or on the energy shell as in dp .
Chapter 1
Classical fields
We recall here some familiar classical fields, introduce the Lorentz group, the
canonical formalism and the action functional, its symmetries and the corresponding conserved quantities such as energy and angular momentum.
1.1
Maxwell field
The electric and magnetic fields E and B constitute the Lorentz covariant antisymmetric tensor field F (x), F = F , such that
Ek (x) = F0k = F 0k ,
1
Bk = klm Flm .
2
(1.1)
In terms of F , Maxwells equations can be written in Lorentz invariant form,

F = 0,
F = j ,
(1.2)
(1.3)
where j (x) is the electromagnetic four-current density, or current for short, which
has to satisfy
j = 0,
(1.4)
for consistency with (1.3) (check: j = F = 0 because F = F
whereas = + ). In jargon we say that the current is conserved.1 The
homogeneous equations (1.2) can be satisfied identically by expressing F (x) in
terms of the (four-)vector potential A (x),
F = A A .
(1.5)
In terms of A the inhomogenous Maxwell equations (1.3) take the form

A A = j .
1
(1.6)
The terminology: Ra current j (x) is conserved, simply means: j (x) = 0. It is of course

the total charge Q = d3 x j 0 (x) which is conserved.
CHAPTER 1. CLASSICAL FIELDS
An important aspect of this description is gauge invariance: F is invariant under

the gauge transformations
A0 (x) = A (x) + (x),
0
F
= F (x).
(1.7)
(1.8)
The equations for A are gauge invariant, which implies that their solution is not
unique. To obtain a unique solution of (1.6) one may impose a gauge condition.
Two well-known gauge conditions are
A = 0,
k Ak = 0,
Lorentz gauge (also called Landau gauge),

Coulomb gauge (also called radiation gauge).
(1.9)
(1.10)
In the Lorentz gauge (1.6) reduces to

2 A = j ,
(1.11)
which are four hyperbolic wave equations for A (recall 2 = 2 02 ).

One has learned in the course of time that the A are the basic variables
for the description of the electromagnetic field. On the other hand, physically
observable quantities have to be gauge invariant. However, this does not mean
that everything physical is expressable in terms of the field strength F . In the
quantum theory this is spectacularly illustrated by the Aharonov-Bohm effect.
1.2
Einstein field
In General Relativity the gravitational field is described in terms of the metric

tensor field g (x) while matter gives rise to an energy-momentum tensor field
T (x). These have to satisfy the Einstein equations
1
(1.12)
R g R + g = 8GT ,
2
together with dynamical equations for matter. Here R and R are the Ricci
tensor and scalar constructed out of g , G is Newtons constant and the
cosmological constant. We shall not go into detail here as our working arena
will be Minkowski space within special relativity. But note that the energymomentum tensor will also play an important role in the following. It is good to
keep in mind its role as a source of the gravitional field.
1.3
Scalar field
For illustrative purpose we will often use a scalar field (x), which carries no
vector or tensor indices. A typical equation for such a field is the Klein-Gordon
equation with source J(x),
( 2 + m2 ) = J.
(1.13)
1.3. SCALAR FIELD
This equation is similar to (1.11), except for the parameter m, which has dimension of (length)1 or frequency (recall that c = 1 in our units, h
does not enter
in classical field theory). To get a feeling for its meaning, consider a plane wave
(x) = eikx + c.c. = eikxik
0 x0
+ c.c.
(1.14)
(c.c. denotes complex conjugate). This is a solution of (1.13) for the case J = 0,
provided that
k 2 + m2 = 0, k 0 = k2 + m2 .
(1.15)
So m is the frequency k 0 for zero wave vector, k = 0. Another example is the

solution for a static point source at the origin,
em|x|
,
4|x|
(2 + m2 )(x) = (x).
(x) =
(1.16)
(1.17)
The solution (1.16) is called the Yukawa potential. We see that it decays exponentially fast as |x| with the scale set by 1/m, the range of the potential.
As a classical field, the scalar field is not so familiar as the electromagnetic or
or gravitational fields. The reason is that in applications to relativistic physics
the particles described by quantized scalar fields are usually unstable with very
short life times. Furthermore, we shall see that m is the mass of such particles;
1/m = h
/mc their Compton wavelength, is typically a very short distance, such
that decays rapidly to zero away from its source. The minimal frequencies are
also very large. For example, for pions m1 = 1.4 1013 cm or m = 2 1023
s1 . In addition, wave packet solutions made out of superpositions
of plane waves
0
tend to spread rapidly because of the dispersion relation k = m2 + k2 (see
e.g. the discussion in Jackson sect. 7.9).
However, in the nonrelativistic domain such scalar fields do occur classically
in systems showing superfluidity or (normal) superconductivity. Suppose is
slowly varying in space compared to m1 . Then it is useful to derive a nonrelativistic form of the Klein-Gordon equation by separating out the high frequency
oscillations, writing
1
1 imt
e
(x, t) + eimt (x, t),
(1.18)
(x, t) =
2
2
im imt
im imt
t (x, t) =
e
(x, t) +
e (x, t),
(1.19)
2
2

i
(x, t) = eimt (x, t) + t (x, t) ,
(1.20)
m
with the complex conjugate equation for . This is a change of variables in
which and t are represented by two new independent fields, the real and
imaginary parts of . The equation ( 2 m2 ) = 0 is equivalent to

2
+ e2imt .
it =
2m
(1.21)
j
j(x)
j(x) = j(xa)
x
a
Figure 1.1: Translation of a field.
Assuming that measurements involve time scales much larger than m we can neglect the rapidly oscillating term exp(i2mt). The resulting equation is identical
to the Schrodinger equation, but this should not confuse us. The above has
nothing to do with this quantum mechanical wave function: it is simply a classical field which happens to be complex. Usually there are also additional source
terms like 2 in the nonrelativistic field equations (the field providing its own
source), in which case one sometimes uses the even more confusing terminology
nonlinear Schrodinger equation. Observable quantities should of course be real.
Simple examples are given by and ( )/2im.
In the application to superfluid 4 He, m is of the order of the mass of the
helium atom. In the application to superconductivity m is the mass of a Cooper
pair (about twice the electron mass) and is called the Landau-Ginsberg field
(the Cooper-pair field).
1.4
Poincar
e group
The field equations of the previous sections are invariant under translations and
Lorentz transformations. Such transformations form a group, the Poincare group,
which contains translations and Lorentz transformations as subgroups. This will
now be discussed in more detail2 .
We start with translations. Let j (x) be the electromagnetic current of the
system. Suppose we translate the system over a distance a in spacetime.3 Then
the corresponding current is j 0 (x) = j (x a), see figure 1.1. So under transla2
Our review is brief, see e.g. Jones or courses on group theory for more information.
This is a so-called active transformation. In the equivalent passive viewpoint the system
is unchanged but we make a coordinate transformation x x0 = x + a.
3
GROUP
1.4. POINCARE
tions:
x0 = x + a,
0 (x0 ) = (x), or 0 (x) = (x a),
A0 (x0 ) = A (x), or A0 (x) = A (x a),
(1.22)
(1.23)
(1.24)
and similar for other fields. The field equations are obviously invariant, e.g.
0 F 0 (x0 ) + j 0 (x0 ) = 0 implies F (x) + j (x) = 0, because 0 /x0 =
/x = .
For many derivations it is useful to consider infinitesimal transformations,
written as 1 ia P + O(a2 ). Here 1 denotes the abstract unit element (no
translation) and the coefficients P of the infinitesimal parameters a are the
generators of the translation group. A finite translation can be abstractly written
as exp(ia P ). For a representation of P , consider an infinitesimal translation
of a scalar field:
0 (x) = (x a) = (x) a (x) + .
(1.25)
(1 ia P + )(x),
(1.26)
P i
(1.27)
Writing this as
we see that
is a representation of the generators P .

Next we turn to Lorentz transformations.
of the form
1 0 0
0 1 0
=
0 0 1
0 0 0
Lorentz transformations
Recall that we use a metric tensor

0
0
0
1
(1.28)
x0 = ` x ,
(1.29)
leave invariant the quadratic form x2 = x x = x x ,

x02 = x2 ,
(1.30)
which implies
` ` = ,
or
` ` = .
(1.31)
We can write the transformations also in matrix form,

x0
x1
x = 2 , x0 = `x,
x
x3
(x0 ) = (`) (x) , (x) x , (`) ` .
(1.32)
(1.33)
Eq. (1.31) reads in matrix notation

`T ` = ,
or `1 = `T ,
(`1 ) = ` ,
(1.34)
where T denotes transposition. Note the subtlety in the connection between

matrix and tensor notation: () = = , but (`) = ` .
For a special Lorentz transformation (boost) in the 3-direction,
`=
cosh
0
0
sinh
0
1
0
0
0 sinh
0
0
1
0
0 cosh
(1.35)
where cosh = , = 1/ 1 v 2 , sinh = v, with v the velocity of the

transformation in units c = 1. For a rotation about the three axis over an angle
,
1
0
0
0
0 cos sin 0
`=
.
(1.36)
0 sin
cos 0
0
0
0
1
The transformations form a group, which includes the rotation group, parity (P )
and timereversal (T ). The latter two are defined by
(`P x)0 = x0 , (`P x)k = xk ,
(`T x)0 = x0 , (`T x)k = xk ,
(1.37)
(1.38)
where k = 1, 2, 3. The rotations and boosts have det ` = 1, while `P and `T have
determinant 1. The transformations with det ` = +1 form a subgroup, the
proper Lorentz group. In the following we shall mean this group when referring to
Lorentz transformations; parity and time reversal will be mentioned separately.
We denote matrix representations of the Lorentz group by D,
` D(`).
(1.39)
0 (x0 ) = D(x).
(1.40)
Fields (x) transform as

Simple examples are given by
(D),
D = 1, 0 (x0 ) = (x),
(D) = ` , 0 (x0 ) = ` (x),
= ` ` , 0 (x0 ) = ` ` (x),
scalar field,
vector field,
tensor field.
(1.41)
(1.42)
(1.43)
See figure 1.2 for an illustration of the transformation rule of the vector field for
the case of rotations in the active formulation.
GROUP
1.4. POINCARE
V(x)
V(x)
Figure 1.2: Rotation of a vector field in space: the vector V is rotated into V 0
and its base point x is transformed into x0 , Vk0 (x0 ) = `kl Vl (x), x0k = `kl xl , just as
would follow from a rotation of the coordinate frame.
The vector representation is the defining representation of the Lorentz group.
An example of a vector field is the electromagnetic current j (x). Under rotations
it has a spin zero component, the scalar (under rotations) j 0 (x), and a spin one
component, the vector (under rotations) j(x). An example of a tensor field is
the electromagnetic field F (x). It is an antisymmetric tensor under Lorentz
transformations, while under rotations it consists of two vectors E(x) and B(x),
E k = F 0k , B k = (1/2)klm F lm , k, l, m = 1, 2, 3.
To verify the invariance of field equations under Lorentz tranformations we
check that transforms as a (covariant) vector. For example
0 0
(x
)
=
(x)
=
(x) = (`1 ) (x)
x0
x0
x0 x
= ` (x),
(1.44)
0 0 (x0 ) =
where we used (1.34). It is now straightforward to check the Lorentz invariance

of field equations, e.g. 0 F 0 (x0 ) + j 0 (x0 ) = 0 implies F (x) + j (x) = 0.
Poincare transformations
x0 = `x + a
(1.45)
leave invariant (x y)2 and form therefore also a group. The field equations are
of course also invariant under this combined group. For more information see
Weinberg I or Ryder.
Consider next infinitesimal Lorentz transformations,
` = + ,
(1.46)
where is infinitesimal (note that = ). The relation (1.31) is satisfied if

=
(1.47)
(also the indices of are raised and lowered with the metric tensor). For a
rotation in the 1-2 plane about the 3-axis over a positive angle we have
12 = 12 = 21 = ,
(1.48)
(cf. (1.36) and Problem 10). Similarly, for special Lorentz transformation in the
0-3 plane of Minkowski space with hyperbolic angle , a boost in the positive
3-direction with velocity v,
03 = 03 = 30 = = tanh1 v
(1.49)
(cf. (1.35) and Problem 10; infinitesimally = v).

For a representation D we write
1
D = 1 + i S + O( 2 ),
2
S = S ,
(1.50)
where the S are matrices specifying the representation. They are called the
generators in the representation D. In the defining (vector) representation,
1
(D) = ` = + i (S ) .
2
(1.51)
It follows from this by comparison with (1.46), that in the defining representation
S is given by
(S ) = i( ), defining representation
(1.52)
(just substitute and check). It is now straightforward to verify that the generators
satisfy the commutation relations
[S , S ] = i( S S S + S ).
(1.53)
A finite transformation can be written as

1
D = exp i S .
2

(1.54)
If we have a set of matrices S satisfying the commutation relations (1.53)

then we have a representation of the Lorentz group. This follows from the BakerCambell-Haussdorff theorem: if D = exp(M ) = exp(i S /2) and similar for
D 0 = exp(M 0 ), then D 00 DD 0 can be expressed as the exponential of a series
in multiple commutators of M and M 0 , D 00 = exp(M + M 0 + [M, M 0 ]/2 + ),
which is completely determined by the commutation relations (1.53).
Note that the boosts do not form a group. This can be seen from the fact
that the set of boost generators S0k is not closed under commutation. For example [S02 , S03 ] = iS23 , which generates rotations about the three axis. So the
rotations are needed to complement the boosts into the (proper) Lorentz group.
1.5. ACTION
We end this introduction with the commutation relations of the generators

of the Poincare group. Let us denote for the moment the abstract generators of
the Lorentz group by J , so S is a representation of J . A representation is
provided by the transformation rule for the spacetime argument of fields; e.g. for
an infinitesimal transformation of a scalar field:
0 (x) = (`1 x) = (x) x (x)
1
= (x) + i (ix + ix )(x).
2
(1.55)
Writing this as

1
1 + i J + (x),
2

(1.56)
we see that we have the representation

J L ix + ix .
(1.57)
For a general field (x) transforming as 0 (x) = D(`)(`1 x) we have the representation
J L + S .
(1.58)
The commutation relations of the generators of the Poincare group now follow
straightforwardly from (1.27) and (1.57):
[J , J ] = i( J J J + J ),
[J , P ] = i( P P ),
[P , P ] = 0.
1.5
(1.59)
Action
A powerful tool in our considerations will be the action S, from which the equations of motion can be derived by requiring it to be stationary under small variations of the dynamical variables. Symmetries of S lead to symmetries of the
equations of motion and to quantities which are conserved in time: Noethers
theorem. Furthermore, the action plays a crucial role in the path integral formulation of quantum theory.
Consider first an action for the scalar field,
S=
1
1
d x m2 2 + J ,
2
2
4
(1.60)
where the integration is over some compact domain M of spacetime. The dynamical variables are here (x), whereas J(x) is considered to be a given function of
10
x, it is an external source. Consider the variation of S under a small variation

of which vanishes on the boundary M of the domain M :
S S[ + ] S[] =
=
d4 x ( m2 + J)
d4 x ( m2 + J),
(1.61)
where we made a partial integration in the second line and used the fact that
= 0 on M . Requiring S = 0 for arbitrary gives the Klein-Gordon
equation with source J,
( 2 m2 ) + J = 0,
(1.62)
in the interior of M . (Allowing not to vanish on M would in addition lead
to boundary conditions for .)
A suitable action for the nonrelativistic scalar field (cf. eq. (1.21)) is given by
S=
d x dt
1
1
1
it it
+ + ,
2
2
2m

(1.63)
where is the nonrelativistic analog of J.

Consider next the following action for the electromagnetic field,
S=
1
d4 x F F + J A ,
4

(1.64)
where J is an external current and the integral is again over M . Under a

variation A of A we have
S S[A + A] S[A]

Z
1
4
d x F F + J A ,
=
4
F = (A + A ) (A + A ) F
= A A ,
(F F ) = 2F F = 4F A ,
S =
=
(1.65)
(1.66)
(1.67)
d4 x (F A + J A )
d4 x ( F + J )A .
(1.68)
We made a partial integration in the last step and assumed that the surface term
is zero, which is correct if we impose that A vanishes on M . Requiring S = 0
for arbitrary variations in M gives Maxwells equations
F + J = 0.
(1.69)
The action for the gravitational field without matter is given by

1 Z 4 q
Sg =
d x det g (R 2).
16G
(1.70)
11
1.5. ACTION
Under a variation g we have4
Sg =
1
16G
d4 x
g R +
1
g R g g .
2

(1.71)
The demonstration of this is quite involved, see for example Weinbergs book on
general relativity. Setting Sg to zero gives the Einstein equations without the
T term. To get the full Einstein equations including T we have to add to Sg
a term representing the contribution of matter, which we denote by Sm . With
the total action
S = Sg + Sm ,
(1.72)
and
1
g T g ,
(1.73)
2
we get the full Einstein equations including T by setting S = 0. Note that
matter is just a name for anything not composed of g only. For example, it
could be a bunch of point particles, but is may also be the electromagnetic field
or a scalar field. We shall now derive the form of T for the scalar field and the
electromagnetic field.
Sm =
d4 x
The minimal prescription for constructing an action that is invariant under general coordinate transformations from a Lorentz invariant action is:
a) make the volume element invariant under general coordinate transformations
d4 x d 4 x
g(x);
(1.74)
b) replace derivatives by covariant derivatives.

With this prescription the scalar field action becomes
Sm [, g] =
dx
1
1
g + m2 2
2
2
(1.75)
(for a scalar field the covariant derivative is just the ordinary derivative). Similarly the generally covariant generalization of the Maxwell action is
Sm [A, g] =

1
g g F F .
d x g
4
4
(1.76)
The calculation of the variation of these matter actions with respect to g is not
difficult (Problem 5) and leads to
T = g g g
4
1
1
g + m2 2
2
2
(1.77)
Acccording to standard notation, g det g (the determinant of the matrix g), and g
(g ) , so g g = .
1
12
for the scalar field and

1
T = F F g F F ,
4
(1.78)
for the electromagnetic field, where F = g g F , etc. Specializing to Minkowski

space is easy: g .
1.6
Canonical formalism
Up to now we have used a manifestly covariant formalism for Lorentz invariant

theories. For the relation to quantum theory, the canonical formalism in terms
of a hamiltonian, which is a function of ps and qs, is appealing because of
the correspondence between Poisson brackets and commutators (qk , pl ) = kl
[
qk , pl ]/i
h = kl . Because time is singled out as special (e.g. pk = t qk ) this formalism breaks manifest covariance and it is complicated for gauge fields. However,
it is important for the proper statement of the initial value problem and useful
for an introduction to quantum field theory. We give the basics here for a scalar
field theory described by the action
S =
d4 x
1
+ V () ,
2
1 2 1 4
+ .
2
4
V () =
(1.79)
(1.80)
We have added a 4 term to the action, which makes theory more interesting
because the field equation is now nonlinear.5 Its importance is parametrized by
. The action can be written as (x0 = t, = 0 )
S =
dt L[, ],
1 2
U [],
2

Z
1
3
U [] =
d x V () + k k ,
2
L =
d3 x
(1.81)
(1.82)
(1.83)
where the dot denotes /t. This looks like the lagrangian for a sum of systems,
one for each x, which are coupled by the spatial gradient term. The canonical
momentum conjugate to is defined as6
(x) =
5
L
= (x),
(x)
(1.84)
This so-called 4 theory is a very useful model for illustration.

The time dependence is implicit in the following. See the appendix for the definition of the
functional derivative.
6
13
1.7. SYMMETRIES AND NOETHERS THEOREM
where we have used a functional derivative. As discussed in the appendix this

is just a generalization of a partial derivative from discrete indices to continuous
indices (here x). In the same generalization the hamiltonian is given by the
Legendre transform
H[, ] =
=
d3 x L[, ],
(1.85)
d3 x
(1.86)
1 2
+ U [].
2
The Poisson bracket (A, B) of A[, ] and B[, ] generalizes to

(A, B) =
"
B
B
A
A
.
dx
(x) (x) (x) (x)
3
(1.87)
With the Poisson brackets

((x), (y)) = (x y),
((x), (y)) = ((x, (y)) = 0,
(1.88)
we now expect the equations of motion to follow from the Hamilton equations:
(x)
= ((x), H) = (x),
(x)
= ((x), H) =
U.
(x)
(1.89)
This is the case indeed (cf. Problem 6). The canonical form of the action principle
is
S =
S =
S =
dt
Z
d x H[, ] ,
d4 x ( ) = 0 = ,
d4 x + 2
= 0 =
(1.90)
(1.91)
U
.
(1.92)
We finally note that the hamiltonian has the nice form of an integral over an
energy density H. This H is equal to T 00 obtained from the variation (1.73) of
the matter action Sm [, g], in the limit of Minkowski spacetime g . In
more general cases thereR may be aRdifference between H and T 00 , but the total
energy is unambiguous, d3 x H = d3 x T 00 .
1.7
Symmetries and Noethers theorem
The invariances of a theory are summarized by the symmetries of its action. This
is a nice idea, although it turns out that there are exceptions, so-called anomalies
where the classical action has more symmetry than the corresponding quantum
theory. For now this does not concern us and we shall explore the spacetime
14
symmetries of the action and their consequences according to Noethers theorem:

To each one parameter symmetry group of the action corresponds a conserved
quantity.
We shall illustrate how these conserved quantities can be found in a few examples
which will be useful later. Consider the action for the scalar field
S=
1
1
L = m2 2 .
2
2
d4 x L,
(1.93)
The lagrangian (density) L transforms as a scalar field under Poincare transformations. If the integration domain M is taken infinite, the action is symmetric
(invariant), S[0 ] = S[], with 0 (x) = (`1 x a). This follows from a simple
change of integration variables, the jacobian is 1. However, we prefer to keep the
integration domain finite so as not to have to discuss convergence of the action
integral.
Consider translations. According to Noether there should be four conserved
quantities, one for each parameter a , = 0, 1, 2, 3. A convenient way to find
these is to perform infinitesimal spacetime dependent translations (x) which
vanish on the boundary M .7 Then 0 (x) = (x (x)) = (x) (x) (x),
or
(x) 0 (x) (x) = (x) (x).
(1.94)
Because depends on x we no longer expect S to be invariant: the derivatives
act nontrivially on . So we may expect the form
S =
d4 x T (x) (x).
(1.95)
Now, when the equations of motion (field equations) for are satisfied, S = 0
for any vanishing on M , hence, also for the variation (1.94). Making a partial
integration and using the arbitrariness of (x), we conclude that
0 = S =
d4 x T ,
T = 0,
(1.96)
in the interior of M . So we expect four conserved currents, T , = 0, , 3, to

which correspond four conserved charges,
P =
d3 x T 0 .
(1.97)
As will be verified below, T is the energy-momentum tensor of the scalar field

and P is its total energy-momentum four-vector. Note that Noethers conserved
7
This method has the advantage that it can be taken over directly in the quantum theory
in the path integral formulation.
1.7. SYMMETRIES AND NOETHERS THEOREM
15
quantities are not necessarily normalized in standard ways: the way we identified
T depends on the normalization of the infinitesimal parameters . Note furthermore that for P to transform as a four-vector it is crucial that T = 0,
see Problem 8.
We now calculate T :
S =
=
=
=
d4 x ( + m2 )
d4 x [ ( ) + m2 ]
d4 x ( L + )
d4 x [ ( L) + ( + L) ] .
(1.98)
The integral over the total divergence in (1.98) vanishes because = 0 on M

and comparing with (1.95) we find
T = + L.
(1.99)
This is identical to (1.77) obtained from the coupling to the gravitational field,
in the limit g .
The Noether form for the energy-momentum tensor of the electromagnetic
field turns out not to be equal to the standard form following from (1.73). The
integrated P are however identical. The T from (1.73) is more physical as it
plays a dynamical role in gravitation (but its derivation is not so easy for spinor
fields). The Noether form can be modified by adding a divergence-less quantity
which integrates to zero in the total P , such that the energy-momentum tensor
satisfies standard criteria such as symmetry,
T = T ,
(1.100)
and gauge invariance. For more information, see Ryder, Itzykson & Zuber, Weinberg I, and Problem 9.
The Noether consequence of Lorentz invariance now follows. The scalar field
action is invariant under the infinitesimal Lorentz transformation ` = + ,
0 (x) = (`1 x) = (x) x (x). Making the transformation spacetime
dependent (and vanishing on the spacetime boundary) we get
(x) = (x)x (x).
(1.101)
This has the form (1.94) with (x) = (x)x . So we can take over eq. (1.95)
to get
S =
=
d4 x T ( x )
d4 x (T x + T )
1Z 4
=
d x (T x T x ) .
2
(1.102)
16
In the last line we used the antisymmetry of and the symmetry of T found
in (1.99). Making a partial integration, acts on the expression in parenthesis
and because the are arbitrary, the currents
J = x T x T
(1.103)
J = 0,
(1.104)
d3 x (x T 0 x T 0 )
(1.105)
are divergence-free,
and the Noether invariants
J
are time-independent. These are the generalized (from rotations to Lorentz transformations) angular momenta of the field. The angular momentum density of the
field, J 0ab , is easiest to interpret: it has the form of an outer product of xa with
the momentum density T 0a , as in lab = xa pb xb pa , or la 21 abc lab = abc xb pc for
a single particle.
R
Note that J 0k = x0 P k d3 x xk T 00 depends explicitly on time. This is
just right to ensure time-independence in the canonical formalism, according to
dJ 0k /dt = J 0k /t+(J 0k , H) = P k +(J 0k , H), where J 0k /t denotes the explicit
time derivative.
A straightforward calculation (check!) shows that the Poisson brackets of the
Noether invariants P and J with the canonical variables and are given by
((x), Pk ) = k (x),
((x), Pk ) = k (x),
((x), P0 ) = (x),
(1.106)
V
((x))
((x), P0 ) = 2 (x) +
, (1.107)
(x)
which can also be written as

((x), P ) = (x),
((x), P ) = (x).
(1.108)
Similarly on can write

((x), J ) = (x x )(x),
(1.109)
etc. So the P and J generate Poincare transformations via the Poisson algebra,
and indeed, their Poisson brackets are just given by (1.59) with a commutator
replaced by i times the Poisson bracket, or i[ , ] ( , ):
(J , J ) = J J J + J ,
(J , P ) = P P ,
(P , P ) = 0.
(1.110)
17
1.8. SUMMARY
1.8
Summary
In field theory the interactions are local. The electromagnetic and gravitational
fields are familiar classical fields, the somewhat less familiar relativistic scalar field
is often used for simplicity of presentation. We can formulate an action, which is
stationary (extremal) when the equations of motion (field equations) are satisfied.
The canonical formalism leads to field equations in Hamilton form. Lorentz transformations (including rotations) and translations form a group called the Poincare
group. These transformations are symmetries of the action. The Noether invariants corresponding
to Poincare transformations are the R total four-momentum
R 3
0
P = d x T and generalized angular momenta J = d3 x (x T 0 x T 0 ),
with T the energy-momentum tensor. The conservation laws have a local expression in the form of divergence equations, T = 0, (x T x T ) = 0.
The Noether invariants in turn act as generators of the symmetry group and their
Poisson brackets satisfy the corresponding Poincare algebra.
1.9
Appendix: Functional derivative
Consider the action for the scalar field (1.60). It is a function of infinitely many
variables, namely (x) for each x M . This is the continuum analog of a
function of many variables labeled by a discrete index k, say f () = f ({k }). In
the case of continuous indices like x we speak of a functional.8 The functional
derivative is the analog of the partial derivative. In the function case we can
write the change of f () under an infinitesimal variation k as
f () =
gk ()k ,
(1.111)
with
f ()
(1.112)
k
the partial derivative. Similarly, in the functional case we can work the variation
in the form
Z
F [] = d4 x G[x, ] (x)
(1.113)
gk () =
(we have seen how to do this in the case of the various actions using partial
integration), and the functional derivative is defined as
F []
= G[x, ].
(x)
(1.114)
For example, according to (1.61) the functional derivative of the scalar field action
is given by
S
= ( 2 m2 )(x) + J(x),
(1.115)
(x)
8
We indicate functionals with square brackets.
18
while the variation (1.68) of the Maxwell field action tells us that
S
= F (x) + J (x).
A (x)
(1.116)
Let us give some more examples for the case of one dimensional indices x and k:
FUNCTIONAL F []
R
F =
a
+
dx b(x)(x)
R
+ dx dy c(x, y)(x)(y) +
R
F
(x)
= b(x) + 2 dy c(x, y)(y) +
F
F = (x) (y)
= (x y)
(x)n
= n(x)n1 (x y)
(y) h
i
(x)
= x
(x y)
(y) x
FUNCTION f ()
P
f = a + k bk
P
+ kl ckl k l +
P
f

= bk + 2 l ckl l +
k
f
f = k
= kl
l
n
k
= nkn1 kl
l
(k+1 k ) =
l
k+1,l k,l ,
where (x y) and kl are the Dirac and Kronecker delta functions.
1.10
Appendix: Rudiments of representations

of continuous groups
A group G is a collection of elements g G with the properties

- there is a multiplication rule: g1 G, g2 G g1 g2 G
- there is a unit element e (often denoted by 1): ge = eg = g
- there is a inverse g 1 : gg 1 = g 1 g = e
Examples:
discrete groups of 90 rotations, groups of continuous rotations in 1,2,. . . dimensions
abelian group (named after Abel) satisfy g1 g2 = g2 g1 (for all g1,2 G)
non-abelian group g1 g2 6= g2 g1 (not necessarily for all g1,2 G)
Elements of continuous groups, such as the rotation group in three dimensions,
or the Poincare group, can be written in exponential form
g = eip Sp e +
1
(ip Sp )n
n!
n=1
(1.117)
(summation over repeated indices), where p are parameters (real numbers) and
Sp are called generators of the group. For example, for rotations, p , p = 1, 2, 3
19
1.11. PROBLEMS
are angles of rotation and Sp are hermitian 3 3 matrices. The commutator of

two generators is a superposition of generators
[Sp , Sq ] = ifpqr Sr ,
(1.118)
in which the coefficients (real numbers) fpqr are called the structure constants of
the group.
A representation of the group is a mapping g D(g) which is itself a group,
with
g1 g2 = g3 D(g1 )D(g2 ) = D(g3 ).
(1.119)
Elements D (j) (g) of a representation j of G can also be written in exponential

form,
(j)
g = eip Sp D (j) (g) = eip Sp ,
(1.120)
in which the Sp(j) are called the generators in the representation j. They satisfy
the same commutation relations as the original Sp :
[Sp(j) , Sq(j) ] = ifpqr Sr(j) .
(1.121)
Conversely, if we are able to construct generators Sp(j) satisfying (1.121), the we

have also constructed a representation of the group G:
00
00
(j)
g = eip Sp , g 0 = eip Sp , g 00 = gg 0 = eip Sp D (j) (g 00 ) = eip Sp .
(1.122)
This follows from (1.117), (1.120), (1.118), (1.121) and the Baker-CampbellHaussdorff formula
0
0
0
eM eM = eM +M +(1/2)[M,M ]+...,
where the . . . consist of multiple commutators of M and M 0 .
1.11
Problems
1. Mathematical methods
Asymptotic expansions occur frequently.
a. Familiarize your selve with the stationary phase approximation and
the saddle point approximation. See for example Mathews and Walker, or
Dingle.
b. We will freely consider Fourier integrals which do not converge. Such
integrals have meaning within the theory of generalized functions (distributions). See for example Lighthill. As an important example, consider
(x)
=
dt eixt (t),
(1.123)
20

where (t) is the Heavyside step function,
(t) = 1,
= 0,
t > 0,
t < 0.
(1.124)
This Fourier integral may be evaluated by introducing a convergence factor

exp(t), where is arbitrarily small postive. Then
(x)
=
i
.
x + i
(1.125)
We have the following identity between distributions

1
1
= P i(x).
x + i
x
(1.126)
Here (x) is the Dirac delta function and P denotes the Cauchy principal
value: if f (x) is a test function, then
Z
1
dx P f (x) lim
0
x
"Z
f (x)
+
dx
x
f (x)
.
dx
x
(1.127)
For finite , make a plot of the real and imaginary parts of 1/(x + i) and
convince yourself that (1.126) is correct in the limit 0.
Using the identity (t) + (t) = 1, verify that
1(x)
dt eixt = 2(x)
c. Consider the integral

I(t) =
dx f (x)eixt ,
(1.128)
which we assume to be convergent. Let C be the closed contour in the

complex z plane from (R, 0) to (+R, 0) on the real axis and then back
along a semicircle in the upper half plane with radius R.
Show that for t > 0 the above integral can be evaluated as
I(t) = lim
R C
dz f (z)eizt
(1.129)
(assume for example that f (z) 0 like 1/|z| as |z| ). Verify that
for t < 0 the integral along the semicircle in the upper half plane does
not converge, and that the analog contour integral for this case is along a
semicircle in the lower half plane.
21
1.11. PROBLEMS
2. Fourier and Cauchy methods
To prove (1.17) we may use the Fourier representation
(x) =
d3 k ikx
e (k),
(2)3
(1.130)
d3 k ikx
e .
(2)3
(1.131)
together with
(x) =
Then has to satisfy

(k2 + m2 )(k)
= 1 (x) =
d3 k
eikx
.
(2)3 m2 + k2
(1.132)
This integral may be done in spherical coordinates with kx = kr cos ,

integrating first over angles,
2 sin kr
1
k2
,
(x) =
dk
2
2
2
(2) 0
m +k
kr
Z
dk 2k eikr
1
=
Re
,
4r
2i m2 + k 2
(1.133)
(1.134)
and then over k using contour integration, by closing the contour in the
upper half of the complex k-plane and picking up the pole at k = im (cf.
Problem 1.c).
Verify this.
3. Green functions
Solutions of the Klein-Gordon equation with source can be obtained in
terms of Green functions,
(x) = in (x) +
(x) = out (x) +
d4 x0 GR (x x0 )J(x0 ),
d4 x0 GA (x x0 )J(x0 ),
where in and out are solutions of the homogenous equation ( 2 +m2 )in,out =
0, and GR and GA are retarded and advanced Green functions, which satisfy
( 2 + m2 )GR,A (x x0 ) = 4 (x x0 ).
(1.135)
We assume here infinite spacetime.

The names retarted and advanced indicate that
GR (x x0 ) = 0,
GA (x x0 ) = 0,
x0 < x00 ,
x0 > x00 .
(1.136)
22

In fact, GR/A (x x0 ) is nonzero only in the future/past light cone of x0 (cf.
Problem 4). Assuming that the source J(x) is nonzero only in a bounded
region of spacetime, it follows that (x) out (x) for x0 .
in
The Green functions may be represented as Fourier transforms

A
GR (x) =
d4 p ipx
1
e
,
(2)4
m2 + p2 (p0 i)2
(1.137)
where > 0 is infinitesimal, i.e. it is arbitrarily small, positive, and will be

set to zero where its presence is not needed anymore. Inthe representation
above it is needed to prescribe how the poles at p0 = m2 + p2 are to be
avoided in the integral.
Given (1.137), verify eqs. (1.135).
By evaluating the integral over p0 in (1.137), verify that
A
GR (x) = (x0 )(x),
(1.138)
where (x0 ) is the step function (1.124) and

(x) = i

d3 p
ipx
ipx
e
e
,
(2)3 2p0
p0 =
p2 + m2 .
(1.139)
Eqs. (1.136) follow from (1.138).

4. Lorentz invariance of GA/R
The advanced and retarted Green functions are Lorentz invariant. This
is suggested by the representation (1.137) if we ignore the infinitesimal .
More precisely it follows from:
i) the Lorentz invariance (x) in (1.139),
ii) the fact that this function vanishes for spacelike separations,
(x) = 0,
x2 > 0.
(1.140)
Under these circumstances the distinction x0 > 0 or < 0 in (1.138) is a

Lorentz invariant property. The advanced/retarted Green functions are
nonzero only inside the past/future light cones.
Property i) relies on the Lorentz invariance of the integration volume element
d3 p
dp
,
(1.141)
(2)2 2p0
i.e.
d`p = dp .
(1.142)
23
1.11. PROBLEMS
Verify this for a Lorentz tranformation along the 3-axis with velocity v < 1:
p00 = p0 + vp3 , p03 = p3 + vp0 , p01 = p1 , p02 = p2 ,
where = 1/ 1 v 2 is the relativistic dilatation factor.
(1.143)
Property ii) follows from the fact that a spacelike interval (x0 , x) can be
transformed into one with time separation zero, (0, x0 ), by a Lorentz transformation, and the fact that (0, x) = 0.
Verify.
5. T from coupling to g
Verify the expressions (1.77) and (1.78) for the energy-momentum tensors
of the scalar and electromagnetic fields, using (1.73) as the definition of
T .
For the variations the folowing formulas are useful. Let g denote the matrix
with matrix elements g . Then (g (
g 1 ) ):
gg1 = 1
g g1 + g
g 1 = 0
g 1 =
g 1
gg1 .
det g =
1

g g g g ,
4!
(1.144)
(1.145)
so
1

g g g g
3!
= det g g g .
det g =
(1.146)
(We implicitly derived Cramers rule for the inverse of a matrix.)

6. Hamilton equations
Verify the statement following eq. (1.89) in steps:
a) Derive the field equations (equations of motion) form the action principle
for the 4 theory given by (1.80).
b) Calculate the Poisson brackets of (x) and (x) with the hamiltonian
H[, ] and verify in this way that Hamiltons equations are equivalent to
the field equations found under a).
7. Functional derivatives may give distributions
From eq. (1.61) follows that
S
= ( 2 m2 )(x) + J(x).
(x)
(1.147)
24
x0
Figure 1.3: The constant-time hyperplane 0 and an arbitrary spacelike hypersurface in Minskowski space.
Verify that differentiating this once again gives
2S
= ( 2 m2 ) 4 (x x0 ).
(x)(x0 )
(1.148)
This can be viewed as a matrix with continuous indices x and x0 . Calculate

the action of this matrix on the vector (x0 ), i.e.
R 4 0 2
d x [ S/(x)(x0 )] (x0 ).
8. Conserved charges and Lorentz invariance
Let j (x) be a conserved current (jargon for divergence-free current density), i.e. j = 0. Assume it drops to zero faster than 1/x2 in spacelike
directions. Show that the corresponding charge
Q=
d3 x j 0 (x)
(1.149)
is conserved (time independent).

By integrating j over the four-volume between the spacelike hypersurfaces and 0 and using Gausss law, verify that
Q=
d (x) j (x).
(1.150)
Show that Q is Lorentz invariant, both from the active and passive point
of view.
0
Hint for the active case: Let j (x) = ` jR (`1 x) be theR current of the
0
transformed system. To be shown is Q0 = d3 x j 0 (x) = d3 x j 0 (x) = Q.
Make a transformation of variables x = `y, and use the identity
0
`0 ` 0 ` 0 = ` 0 0 0 0 det(`)
(1.151)
25
1.11. PROBLEMS
(verify!). Then d (x) = ` d (y), and it follows that
Q0 =
d (x)j (x) =
=
1
0
d (y)j (y)
d (y)j (y) = Q,
(1.152)
where 1 is the hypersurface for the integral over the (dummy) y-variables
corresponding to the hypersurface x0 = constant. In the last step we used
the previous result (1.150). Make a sketch of 0 and 1 similar to figure
1.3, for the case of a boost in the 3-direction.
Similarly, show that P =
d3 x T 0 transforms as a four-vector.
9. Noether and gravitational energy-momentum tensor

Consider the action (1.75) for a scalar field coupled minimally to the Einstein field, S = S[, g]. Let T be the energy-momentum tensor that is
the source in the Einstein field equations, which is given by (1.73), and let
TN the one found by the Noether method. We have seen that T = TN in
the Minkowski limit g = . In this problem we want to shed some light
on this equality. The action S is invariant under the general coordinate
transformations
0
x = f (x),
(x ) = (x),
0
g
(x0 )
x x
g (x),
=
x x0
(1.153)
i.e. S[0 , g 0 ] = S[, g]. Consider an infinitesimal transformation

0
x = x + (x).
(1.154)
Then, to first order in ,

(x) 0 (x) (x) = (x) (x),
(1.155)
0
g(x) g
(x) g (x) = (x) g (x) g (x) (x) g (x) (x).
The invariance of S has the consequence
0 = S =
dx
S
S
g +
.
g
(1.156)
The first term on the right hand side of (1.156) gives

Z
dx
in the Minkowski limit.
1
g T g =
2
d4 x T ( ),
(1.157)
26

The infinitesimal coordinate transformation looks like a spacetime dependent translation, and (1.155) is identical to the variation (1.94) used in the
derivation of TN . So the second term on the right hand side of (1.156)
produces, as in (1.95),
Z
d4 x TN .
(1.158)
Since the sum of the two terms is zero we conclude that

T = TN .
(1.159)
Note that we have not assumed that the equations of motion are valid: the
above identity holds for arbitrary .
For our minimally coupled action we actually found the stronger identity
T = TN , but this does not have to be the case. For example, adding to
the minimally coupled action the term
d4 x
1
g R2 ,
2
(1.160)
leads to a different T . In the Minkowski limit it is given by

T = T ( 2 )2 ,
(1.161)
where T is the old tensor corresponding to = 0.

Verify that T and T lead to the same total energy-momentum, d3 x T0 =
R 3
d x T 0 .
R
For = 1/6 this new energy-momentum tensor is sometimes called the

improved energy-momentum tensor. Verify, using the equations of motion,
that the improved tensor satisfies
T1/6
= m2 2 .
(1.162)
So the trace of T1/6

vanishes for m2 = 0. This property is relevant for the
description of scale invariance.
10. The exponential parametrization

Let S be the generators (1.52) in the defining representation. Verify by
expansion that ` = exp(iS12 ) is equal to the rotation matrix (1.36).
Verify that ` = exp(iS03 ) is the boost matrix (1.35).
11. Vector-matrices
Let R = exp(ik Sk ) be a general rotation matrix, where Sk (1/2)klm Slm .
The boost generators S0k are vector-matrices (vectors for short) under rotations in the sense that R1 S0k R = Rkl S0l . Verify this for infinitesimal rotations (you need to check the commutation relations [Sk , S0l ] = iklm S0m ).
Chapter 2
Quantized scalar fields
This chapter introduces the basics of the theory of the canonically quantized
scalar field. We introduce the particle interpretation, touch briefly on applications
to scattering and decay processes, and discuss Lorentz symmetry and locality in
the quantum theory.
2.1
Canonical quantization
Consider the 4 model described by the action

1
S = d x + V () ,
2
1 2 1 4
V () = + + .
2
4
Z
(2.1)
(2.2)
For later use we have added a constant to (1.80). It may be interpreted as

representing the cosmological constant. In a world in which the above action
would describe all existing matter, shifting from the graviational field action
Sg to the matter action Sm would give = /8G (cf. (1.70)).
For interpretation of the quantized field the expressions for the energy, momentum etc. (the Noether invariants) are important, in particular
P
Pk
1
1
=
d x 2 + ()2 + V () = H(, ),
2
2
Z
= d3 x k ,
Z
Jk =
d3 x (klm xl m ).
(2.3)
(2.4)
(2.5)
We now quantize the theory by replacing the classical fields and by operators
and in Hilbert space, such that their commutators go over into their Poisson
brackets in the formal classical correspondence limit h
0: [ , ]/i
h ( , ). So
27
28
CHAPTER 2. QUANTIZED SCALAR FIELDS
at say, t = 0, we put (keeping Plancks constant h

explicit for the moment)
[(x), (y)] = i
h 3 (x y),
[(x), (y)] = [(x), (y)] = 0.
(2.6)
These are called the canonical commutation relations. In the Heisenberg picture
(where the operators are time dependent) they are supposed to hold at equal
times. The above relations are a straightforward generalization of the case of
discretely many variables. One realization of the commutation relations is the
coordinate representation:
(x) multiplication by (x),
(x)
,
i(x)
(2.7)
acting on Schrodinger wave functionals []. This realization is basic to the

path integral approach which will be introduced later. In this chapter we follow
another approach which is geared to the particle interpretation of the quantized
field.
2.2
Free field
For = 0 the hamiltonian is at most quadratic in the canonical variables. For

reasons that will become clear later on we change the notation in the quantum
theory by adding a subscript zero to the parameters in the action, so the hamiltonian is given by
H=
1
1
1
d x 2 + ()2 + 0 2 + 0 .
2
2
2
3
(2.8)
This model with hamiltonian quadratic in the fields is called the free theory,
because it is equivalent to a collection of uncoupled harmonic oscillators, as will
now be shown by going over to momentum space. To simplify the presentation
we first assume only one spatial dimension. Afterwards, we can easily generalize
to three spatial dimensions. We furthermore assume space to be a circle with
circumference
L, i.e. 0 x L with periodic boundary conditions at 0, L and
R
RL
dx = 0 dx. We expand the fields at time t = 0 in Fourier modes,
1 X ipx
1 X ipx
e p , (x) =
e
p ,
(x) =
L p
L p
1 ZL
1 ZL
p =
dx eipx (x),
p =
dx eipx (x),
L 0
L 0
(2.9)
(2.10)
where p = 2n/L, n = 0, 1, 2, . The modes are eigenfunctions of the

gradient operator /x with periodic boundary conditions. Since the fields are
hermitian, (x) = (x), the Fourier components satisfy the relations
p = p ,
p = p .
(2.11)
29
2.2. FREE FIELD
The hamiltonian and the momentum operator are diagonal in this representation:
X1
[
p
p + (p2 + 0 )pp ] + 0 L,
(2.12)
H =
2
p
P =
p p ip.
(2.13)
The hamiltonian looks like that of a sum of harmonic oscillators with frequencies
p =
p 2 + m2 ,
m2 = 0 ,
(2.14)
where we have chosen 0 > 0. As in the case of the harmonic oscillator, it is very
useful to introduce creation and annihilation operators, ap and ap , one for each
mode:
1
1
(pp + i
p ), ap =
(p p i
p ),
(2.15)
ap =
2p
2p
1
1
p =
(ap + ap ),
p =
(ip ap + ip ap ),
(2.16)
2p
2p
where we used (2.11). The creation and annihilation operators satisfy the commutation relations
[ap , aq ] = pq ,
[ap , aq ] = [ap , aq ] = 0.
(2.17)
The hamiltonian and the momentum operator can now be written in the form

X
1
p + 0 L,
(2.18)
H =
ap ap +
2
p
P =
X
p
ap ap p.
(2.19)
We see that the hamiltonian is just a sum of independent harmonic oscillators.

The ground state, i.e. the state with lowest energy, is given by (as usual, up to a
phase factor)
ap |0i = 0, for all p, h0|0i = 1.
(2.20)
It is also an eigenstate of P with eigenvalue zero. The other simultaneous eigenstates of H and P are obtained from the ground state |0i by application of the
creation operators,
Y (ap )np
q
|{np }i =
|0i,
(2.21)
np !
p
with occupation numbers np = 0, 1, . All eigenstates are normalized to 1. The

eigenvalues of H and P are given by
H|{np}i =
E0 +
P |{np}i =
X
p
np p |{np }i,
np p |{np }i.
E 0 = 0 L +
X
p
1
p ,
2
(2.22)
(2.23)
30
Consider now the ground state energy density:

1 X1
E0
= 0 +
p
L
L p 2
Z
dp 1 q 2
0 +
p + m2 ,
2 2
(2.24)
L .
(2.25)
The integral in the last line is the limit of a Riemann sum for L :
Z
1X
1 X
dp
F (p),
F (p) =
p F (p)
L p
2 p
2
2
.
L
p =
(2.26)
The ground state energy as written is infinite, because the integral diverges at
large p. The reason is that we are dealing with an infinite number of degrees of
freedom. However, we can absorb this infinity in 0 , such that is finite. We
come back to this shortly.
We now generalize to three spatial dimensions. Let us choose 0 such that
= 0. Then we can summarize as follows:
eipx
d3 x (x), etc.,
L3
1
=
(p p + i
p ) ,
2p
p =
ap
(x) =
X
p
(x) =
X
p
ap q
P =
2p
L3
e
,
+ ap q
2p L3
(2.29)
eipx
eipx
ip ap q
+ ip ap q
,
2p L3
2p L3
[ap , aq ] = [ap , aq ] = 0,
ap ap p ,
P 0 = H,
P |0i = 0,
(2.28)
ipx
ipx
[ap , aq ] = p,q ,
X
(2.27)
P |pi = p |pi,
P |p1 p2 i = (p1 + p2 )|p1 p2 i,
p 0 = p =
p2 + m2 ,
|pi ap |0i = |1p i,
|p1 p2 i ap1 ap2 |0i,
(2.30)
(2.31)
(2.32)
(2.33)
(2.34)
etc. In (2.33) we used the convention that only non-zero occupation numbers are
shown in the ket.
The interpretation of the scalar field model in terms of a collection of free
particles is very suggestive. The ground state |0i is interpreted as representing
the vacuum. The one particle state |pi is the state with np = 1 and all other
nq = 0, q 6= p. The mass of the particles is m = 0 . Their spin is zero since

their is no further index besides p to indicate a spin degree of freedom. More
formally, it follows from (2.119) below that a particle state at rest (p = 0) is
2.3. RENORMALIZATION OF THE COSMOLOGICAL CONSTANT
31
invariant under rotations, so its total angular momentum is identically zero and
the particles are spinless.
The two particle state1 |p1 p2 i is symmetric in the interchange of the labels
p1 and p2 : the particles are bosons.
2.3
Renormalization of the cosmological constant
We now return to the energy density in the groundstate, . It is the vacuum

expectation value of T 00 . Calculating the expectation value of the full energymomentum tensor gives in the infinite volume limit
h0|T (x)|0i = 0
+I ,
dp p p
(2.35)
(independent of x in accordance to translation invariance), where we introduced

the notation
d3 p
.
(2.36)
dp
(2)3 2p0
Apart from conveniently absorbing numerical factors, this volume element of
integration has the important property that it is Lorentz invariant (cf. Problem
1.4):
d`p = dp .
(2.37)
It follows that the integral is an invariant tensor under Lorentz transformations,
` ` I
` `
dp p p =
d`p (`p) (`p) =
dp p p = I .
(2.38)
There are only two independent Lorentz-invariant tensors, and (check!).

Hence, the vacuum expectation value of T is proportional to :
h0|T (x)|0i = .
(2.39)
We can now interpret as the true contribution to the cosmological constant,

while 0 is just a parameter in the action. In standard jargon, 8G is the
renormalized (or dressed) cosmological constant, and 8G0 the bare cosmological
constant.
However, the integral (2.35) is badly divergent at large momenta. To make
sense of it we should regularize it. Even better, we can start with a regularized
formulation of the theory such that at every stage we have well defined expressions. This can be done, e.g. by replacing the spacetime continuum by a lattice,
but it is cumbersome and we have learned that in many cases it is sufficient to
1
This state can also be written as |1p1 1p2 i, or
2|2p i if p1 = p2 = p.
32
deal with the problem on the fly, by regulating divergent integrals in a consistent manner. We could simply cut off the momentum integration at |p| = .
Using spherical coordinates this gives for T 00
4
h0|T |0i = = 0 +
2(2)3
00
dp p
p 2 + m2 .
(2.40)
For T kl , rotational invariance tells us that it has the form

T kl = p kl ,
(2.41)
since kl is the only relevant invariant tensor under rotations. In fact, p is the
pressure of the ground state. It follows that 3p = kl Tkl , and
1 4
p = 0 +
3 2(2)3
dp
p4
.
p2 + m2
(2.42)
The problem with this regularization is, that it is not consistent with Lorentz invariance: we are treating space and time differently and h0|T |0i will not be proportional to this way and p 6= . Inserting the identity 1 = (p2 /3)p3 /p
into (2.40) and making a partial integration gives
= p +
1 4
3
m2 + 2 ,
3 (2)3
(2.43)
where the second term on the r.h.s. is the surface term. This also shows that if
the unregularized integral were convergent and the surface term absent, p =
would follow indeed.
There are Lorentz covariant regularizations, for example dimensional regularization or Pauli-Villars regularization. The latter is simplest here to present and
is as follows. Define h0|T |0i as
h0|T |0i = 0
q
d3 p 1 X pi pi
ci 0 ,
(2)3 2 i
pi
(2.44)
where p0i = m2i + p2 , pi = p, and the coefficients ci and the masses mi are
chosen such that the integral converges, whith c1 1 and m1 m. This regularization is Lorentz invariant because the ci and mi are invariant. When the
masses mi , i > 1 are sent to infinity the result diverges again, but we cancel this
by a suitable choice of 0 . See Problem 9 for more details.
Having set the vacuum energy density equal to zero we can now ask meaningful
questions about the energy of the ground state in a finite volume. A famous
example is the Casimir effect. This was originally discovered in QED but it
applies also to our scalar field mutatis mutandis (two free massless scalar fields
to represent the two spin states of the photon, Dirichlet boundary conditions).
33
2.4. PERTURBATION THEORY
However, let us use the language of QED anyway as it is more intuitive. Consider
two parallel plates of a conductor a distance a apart, with a much smaller than
the linear size L of the plates. The presence of the plates is taken into account by
imposing boundary conditions corresponding to a perfect conductor. This shifts
the ground state energy inside and outside the plates relative to the vacuum, and
the result is (see e.g. the book by Milton, Itzykson and Zuber sect. 3-2-4, Van
Baal sect. 2)
h 2 L2
.
(2.45)
E =
720a3
It corresponds to a tiny attractive force which has been verified by experiment.
2.4
Perturbation theory
In the general case that the action is of higher than second order in the fields
the theory is said to be interacting, because there is then no Fourier or other
representation in which the harmonic oscillators are uncoupled. In our scalar
field model with hamiltonian2
H=
d3 x
1 2 1
1
1
0 + (0 )2 + 0 20 + 0 40 + 0 ,
2
2
2
4

(2.46)
the strength of the anharmonic 40 term is monitored by 0 , the coupling constant.

Its presence changes the eigenvalues and eigenvectors of P , and we have to
recalculate the ground state and the single and multiparticle states. A useful
tool is perturbation theory, e.g. making an expansion in powers of 0 .
However, it is much better to make an expansion in a renormalized coupling
constant that is closely related to experimental observation. In QED the corresponding coupling constant is the elementary charge e or the fine-structure
constant = e2 /4. The starting parameters in the quantum theory are not
simply related to familiar quantities like charge and mass of a particle, and we
make this explicit by giving them the subscript 0: 0 , 0 , 0 . They are called
the unrenormalized, or bare, parameters. The analogue physical quantities are
denoted by , and , and are called the renormalized, or dressed, parameters.
Furthermore, the strength of the field turns out to be changed by the interaction
and one introduces a renormalized field which differs from 0 by a factor that
is traditionally written as
(2.47)
0 = Z .
The constant Z (Z > 0) is called the wave function renormalization constant.
The fact that fields represent an infinite number of degrees of freedom easily
leads to divergent integrals in perturbation theory. It turns out that such divergencies can be absorbed in the bare parameters, such that the renormalized
2
The reason for the subscript 0 will be explained shortly.
34
ones come out finite. We have seen an example of this in 0 and its relation to .
Similarly, expectation values of products of 0 may contain divergencies that are
absorbed in Z such that the expectation values of are finite, when expressed
in terms of the renormalized parameters. In perturbation theory, formulated as
an expansion in , one finds
Z = 1 + O(2 ),
0 = + O(2 ),
0 = + O(),
0 = + O(1).
(2.48)
This is in accordance with the results in the previous section, where we found
for 0 = 0 that 0 = + constant, and 0 = m2 . We shall assume the
ground-state energy to be zero, = 0.
Consider now the time-evolution operator
U (t, 0) = eiHt .
(2.49)
We want to expand this is powers of . At t = 0, the hamiltonian can be written

as the sum of a free part3
H0 =
=
1
1
1
d x 2 + ()2 + m2 2 + free
0
2q
2
2
X
p
ap ap m2 + p2 ,
(2.50)
(2.51)
and an interaction part H1 = H H0 . Note that in H0 , is defined to have the

usual commutation relation with , hence
= Z 0 .
(2.52)
Furthermore, m is the physical mass and free
is chosen such that the vacuum
0
energy is zero in the free theory. The interaction hamiltonian has the form
1
(2.53)
d3 x 4 + H1 ,
4"
#
Z
2
()2
2
4
=
d3 x (Z 1 1) + (Z 1)
+ m2
+ + (2.54)
,
2
2
2
4
H1 =
H1
m2 = Z0 m2 ,
= Z 2 0 ,
= 0 free
0 .
(2.55)
The terms in H1 are called counterterms. In lowest nontrivial order of perturbation theory the counterterms are usually zero, and in these lecture notes we
shall usually ignore them. The relation between bare and renormalized parameter
will be studied in more detail in chapter 4. See, e.g. Peskin & Schroeder ch. 10
for more information on the counterterm method (also known as renormalized
perturbation theory).
3
The subscript 0 in H0 means free here, not bare.
35
2.5. SCATTERING
We want to expand the evolution operator in powers of H1 . It is wrong to

simply expand the exponential because H1 and H0 do not commute. This is
a standard problem in time-dependent perturbation theory, which is solved by
introducing4
U1 (t, 0) = eiH0 t eiHt ,
(2.56)
differentiating this with respect to t,
it U1 (t, 0) = eiH0 t H1 eiH0 t U1 (t, 0) = H1 (t) U1 (t, 0),
H1 (t) = eiH0 t H1 eiH0 t ,
(2.57)
(2.58)
and integrating this from 0 to t. The result can be written as a time-ordered

product, for example by discretizing time in small steps a = t/N , tn = na,
U1 (t, 0) =
lim eiH1 (tN 1 )a eiH1 (tN 2 )a eiH1 (0)a
T ei
= 1i
Rt
0
dt0 H1 (t0 )
t
(2.59)
(2.60)
dt0 H1 (t0 ) +
(i)
2!
t
0
00
dt0 dt T H1 (t0 )H1 (t00 ) + (2.61)
where T is the time-ordering operator, defined as an instruction to order operators by increasing time from right to left:
T H1 (t1 )H1 (t2 ) = (t1 t2 )H1 (t1 )H1 (t2 ) + (t2 t1 )H1 (t2 )H1 (t1 ),
(2.62)
T H1 (t1 ) H1 (tk ) = H1 (ti1 )H1 (ti2 ) H1 (tik ), ti1 > ti2 > > tik (. 2.63)
A derivation of (2.60) by iteration is given in most text books on quantum field
theory. Multiplying the expansion for U1 (t, 0) by eiH0 t from the left gives the
expansion for U (t, 0):
eiHt = eiH0 t ieiH0 t
2.5
t
0
dt0 eiH0 t H1 eiH0 t + O(H12 ).
(2.64)
Scattering
One of the most interesting new possible effects due to the interaction is scattering
of particles. Fortunately, to lowest non-trivial order we only need to know the
particle states in zeroth order, i.e. the free states, and we shall not need to
renormalize and .
4
This is the evolution operator in the interaction picture, the formalism in which states
evolve in time according to |, ti = U1 (t, 0)|, 0i, and operators O(t) = eiH0 t O(0) eiH0 t . In
the Heisenberg picture the states are time-independent and the operators evolve as O(t) =
eiHt O(0) eiHt ; in the Schr
odinger picture the operators are time-independent and the states
evolve according to |, ti = U (t, 0)|, 0i. Arbitrary matrix elements of operators are identical
in the three pictures. Note that in general the splitting H = H0 + H1 depends on time (here
chosen t = 0).
36
Consider the scattering 1 + 2 3 + 4. We start with a free two-particle

state |p1 p2 i at time t = 0 and wish to calculate the probability amplitude for the
transition to another such state |p3 p4 i at a later time t,
U (t, 0) = eiHt ,
hp3 p4 |U (t, 0)|p1 p2 i,
(2.65)
where U (t, 0) is the evolution operator. The hamiltonian H has the form
H = H0 + H1 ,
H1 =
1
d3 x 4 + H1 .
4
(2.66)
with H0 the free hamiltonian of the previous sections:

H0 |p1 p2 i = (E1 + E2 )|p1 p2 i,
E1 = E(p1 ) =
p21 + m2 ,
(2.67)
etc, and the H1 is given in (2.55). For non-trivial scattering the final state is
different from the initial state and the result would then be zero if H1 were zero.
Hence the scattering amplitude is at least of order H1 (order ). The expansion
of the evolution operator in powers of H1 is given in the previous section. We
only need the first order expression (2.64), which leads to
1 eiEt
hp3 p4 |H1 |p1 p2 i + O(2 ),(2.68)
E
E = E3 + E4 E1 E2 ,
(2.69)
hp3 p4 |U (t, 0)|p1 p2 i = ei(E3 +E4 )t

and
|hp3 p4 |U (t, 0)|p1 p2 i|2 =
2 2 cos(Et)
|hp3 p4 |H1 |p1 p2 i|2 .
(E)2
(2.70)
We now turn to the matrix element of H1 . Using

(x) =
X
q
eiqx
2E(q)L3
aq + q
eiqx
2E(q)L3
aq ,
(2.71)
and using the fact that only terms contribute which do not change the number of
particles (i.e. same number of annihilation and creation operators), we get terms
of the form
hp3 p4 |aq3 aq4 aq1 aq2 |p1 p2 i =
(q1 ,p1 q2 ,p2 + q1 ,p2 q2 ,p1 )hp3 p4 |q3 q4 i(2.72)

2q1 ,p1 q2 ,p2 hp3 p4 |q3 q4 i
2q1 ,p1 q2 ,p2 (q3 ,p3 q4 ,p4 + q3 ,p4 q4 ,p3 )
4q1 ,p1 q2 ,p2 q3 ,p3 q4 ,p4 ,
(2.73)
where the arrows indicate equivalence under relabeling of the dummy qs which
are to be summed over. (In (2.72) we worked the aq s to the right using the
37
2.5. SCATTERING
commutation relations until we got aq |0i = 0.) There are five more such contributions, differering in the order of the operators (aa a a, . . . , aaa a ), which
each give equivalent results (terms like qi ,qj do not contribute because the initial
and final states differ). The result is then
Z
6
6L3
3
i(p3 p4 +p1 +p2 )x
hp3 p4 |H1 |p1 p2 i = Q

d
x
e
=
p +p ,p +p .
Q
2Ei L3
2Ei L3 3 4 1 2
i
i
(2.74)
This gives for the probability
|hp3 p4 |U (t, 0)|p1 p2 i|2 =
(6)2 L6 2 2 cos(Et)
p3 +p4 ,p1 +p2 .
Q
L12 i 2Ei
(E)2
(2.75)
We are interested in scattering into a domain of final momenta,

X
(p3 ,p4 )
|hp3 p4 |U (t, 0)|p1 p2 i|2
(2.76)
2 2 cos(Et)
L3 (6)2
d3 d4
(2)3 3 (p3 + p4 p1 p2 ),(2.77)
4E1 E2
(E)2
d3 pi
,
(2.78)
di =
(2)3 2Ei
Z
where the arrow indicates the infinite volume limit (2.26), which also implies
L3 p,q (2)3 3 (p q).
(2.79)
For large times t (on the scale of the typical inverse energies E 1 ) we have the
identity
2 2 cos(Et)
= t 2(E) + O(1/t).
(2.80)
(E)2
This can be shown by integration with a test function F (E):
Z
dE F (E)
2 2 cos Et
= t
E2
2 2 cos u
(2.81)
u2

Z
2 2 cos u
2
= t F (0)
du
+ O(t ) (2.82)
u2
= tF (0) 2 + O(t1 ),
(2.83)
Z
du F
u
t

where we used F (u/t) = F (0) + F 0 (0)u/t + O(t2 ); the F 0 (0) term drops out by
symmetry.
Summarizing, we have the following result for the probability rate:
=
=
|hp3 p4 |U (t, 0)|p1 p2 i|2

t (p3 ,p4 )
L3 Z
d3 d4 (2)4 4 (p3 + p4 p1 p2 ) (6)2 .
4E1 E2
(2.84)
(2.85)
38
The probablity rate implies an event rate, which is expected to be proportional

R
to the overlap of particle densities d3 x n1 n2 . As will be discussed in greater
detail in chapter 9, the results of scattering experiments are expressed in terms
of the cross section . In a reference frame where the initial particle momenta
are aligned it is defined by
event
= v12
d 3 x n 1 n2 ,
(2.86)
with
v12 = |p1 /E1 p2 /E2 |
(2.87)
the relative velocity.

Realizing that event
= if we normalize to unit initial
R 3
particle number, d x n1,2 = 1, and that the density of our initial particles is
n1,2 = 1/L3 , we have the result
1
=
4E1 E2 v12
d3 d4 (2)4 4 (p3 + p4 p1 p2 ) |T |2 .
(2.88)
where T is the scattering amplitude (also called the invariant amplitude M) for
this case,
|T |2 = (6)2 .
(2.89)
The prefactor can be expressed as a Lorentz scalar,
E1 E2 v12 =
(p1 p2 )2 m21 m22 ,
(2.90)
and we see that if the integration domain is invariantly specified, the cross
section is a Lorentz scalar. For example, integrating over all momenta gives the
total cross section (cf. Problem 3)
=
1
(6)2 ,
32s
s (p1 + p2 )2 ,
(2.91)
where the Lorentz invariant s is equal to the total energy squared in the center
of mass frame. In a more detailed specification of we can fix the invariant
momentum transfer t. The corresponding cross section is conventially written
d/dt (cf. Problem 3):
d
1
=
(6)2 ,
dt
16s(s 4m2 )
t (p1 p3 )2 .
(2.92)
In the center of mass frame defined by p1 +p2 = 0, we have t = 2|p1 |2 (1cos ),

with the scattering angle (cf. Fig. 2.1), the angle between p1 and p3 , and
|p1 |2 = (s 4m2 )/4. So we see that the differential cross section
"
d
d
=
cm
1
(6)2
2
64 s
(2.93)
39
2.6. DECAY
p3
p2
p1
p4
Figure 2.1: Three-momenta for scattering in the center of mass frame.
is isotropic. This is special to the 4 theory, later we will encounter more interesting differential cross sections.
The above derivation of the scattering amplitude has the benefit that it is
short. In higher orders it gets complicated because it lacks manifest Lorentz
covariance. Only the end results are covariant or invariant. Later we will develop more sophisticated calculational techniques which are manifestly covariant.
Conceptually the above derivation can be improved by considering wave packet
states which are localized in space (unlike the plane wave states used here which
correspond to uniform density). This we will do in chapter 9.
2.6
Decay
Apart from leading to scattering, interactions may cause particles to be unstable,

transforming them into two or more particles of a different species. For example,
neutral pions are unstable and decay predominantly into two photons, 0 +,
with a mean life time = 8.8 1017 sec. The mean life time is the inverse of
the decay rate .
The possibility of decay can be illustrated by the following simple model
involving only spinless particles. The model is specified by the action
!
1
1
1
1
g
S[, ] = d x
+ M 2 2 + + m2 2 + 4 + 2 ,
2
2
2
2
4
2
(2.94)
which describes two types of particles and , with masses M and m,
respectively. There are two interaction terms,
Z
Hint =
d3 x
1 2
1
g + 4 ,
2
4

(2.95)
40
with strengths parametrized by the coupling constants g and (g has dimension

of mass). Apart from new types of scattering, the g2 term also allows for
transitions + .
Suppose at time zero the initial state contains only one -particle with fourmomentum p. The probability at a later time t for the decay into two -particles
with momenta q1 and q2 is then |hq1 ()q2 ()|U (t, 0)|p()i|2 (we use the same notation as in the scattering case). Going through similar steps as in the derivation
of the rate for scattering, gives for the decay rate
=
1 X
1
=
|hq1 ()q2 ()|U (t, 0)|p()i|2
t 2 q1 q2
Z
1 1
=
d1 d2 (2)4 (q1 + q2 p) g 2 ,
2p0 2
(2.96)
(2.97)
where we used
g
hq1 ()q2 ()|Hint |p()i =
2
d3 x
p0 q01 q02
1
q
8q100 q200 p00 L9
ei(p q1 q2 )x
hq1 ()q2 ()|aq0 ()aq0 ()ap0 ()|p()i (2.98)

1
1
1
= gq
q1 +q2 ,p .
(2.99)
8q10 q20 p0 L3
The explicit factor 1/2 in (2.96) avoids double counting the two identical particles
in the final state.
This example illustrates that the transition at relatively large times on the
scale of m1 , M 1 (i.e. the decay), is only possible if energy-momentum is
conserved: q1 + q2 = p. Examining this for the case of a -particle at rest one
finds that this leads to the condition that there has to be enough energy to create
the two particles in the final state,
M 2m.
(2.100)
The integral in (2.97) is Lorentz invariant. It depends only on g 2 , M and m (cf.

Problem 3),
1 2
|q|
2
M 4m2 .
(2.101)
g
,
|q|
=
=
16M p0
2
For a moving -particle the factor 1/p0 in (2.97) expresses the expected time
dilatation.
The unstable particles can be produced in scattering, e.g. (q1 ) + (q2 )
(p), which is just the inverse of the decay process.
GROUP
2.7. REPRESENTATION OF THE POINCARE
2.7
41
Representation of the Poincar

e group
We have seen in a few examples that Lorentz invariance emerges in the infinite
volume limit. We now go into somewhat more detail of the formal aspects of
symmetry in the quantum theory. In addition to the momentum operators P ,
also the angular momentum operators and generators of special Lorentz transformations
Z
J = d3 x (x T 0 x T 0 )
(2.102)
are time independent. The commutation relations with the field operators can
be calculated from the canonical commutation relation (Problem 4),
[(x), P ] = i (x),
[(x), J ] = i(x x )(x),
(2.103)
(2.104)
and these lead to the Poincare algebra (1.59).

The transformations are represented in Hilbert space by unitary operators
U (a), U (`). For translations we have
| 0 i = U (a)|i,
U (a) = eia
(2.105)
Here |i represents a state of the system and | 0 i represents this state actively
translated over a spacetime distance a . To see that (2.105) is correct we calculate
the expectation value of (x):
h|(x)|i = f (x) h 0 |(x)| 0 i = f (x a),
(2.106)
where we used (cf. (1.25), (2.103) and Problem 6)

U (a) (x)U (a) = (x) ia [(x), P ] +
= (x a).
(2.107)
Note that in the quantum theory (x) corresponds to the observables, whereas
classically its transformation properties were treated as representing the state of
the system. Similarly we have for Lorentz transformations5
| 0 i = U (`)|i,
U (`) = ei 2
` = ei 2
(2.108)
and (cf. (1.55), (2.104))

1
U (`) (x)U (`) = (x) + i [(x), J ] +
2
1
= (` x).
5
(2.109)
The sign in the exponent is here + by convention, but recall (cf. (1.48)) that U (`) =
exp(iJ3 ) for an active rotation about the 3-axis over an angle .
42
The fact that the unitary operators exist guarantees invariance of transition amplitudes, h10 |20 i = h1 |2 i. If Lorentz invariance is broken, then U (`) does not
exist or is time dependent. Despite its non-covariant features, canonical quantization gives a Lorentz invariant theory.
In the following we shall use a convenient covariant normalization of particle
states in the infinite volume limit
hp0 |pi = 2p0 (2)3 (p0 p),
p0 =
This has the property (cf. (2.37))

Z
p2 + m2 .
(2.110)
dp f (p) hp0 |pi = f (p0 ).
(2.111)
For the argument of ket and bra we use the four-momentum p, but note that here
p0 is not an independent variable. Comparing with our finite volume normalization we have
q
(2.112)
|pi = 2p0 L3 |pi
(recall (2.79)). In infinite volume we expand the free scalar field in terms of
covariant a(p) and a (p),
(x) =
dp a(p)eipx + a (p)eipx .
(2.113)
Comparison with the previous finite volume expansion at time zero

(x) =
X
p
"
eipx
eipx
0 3 ap + 0 3 ap
2p L
2p L
(2.114)
shows that6
a(p) =
2p0 L3 ap ,
(2.115)
[a(p), a (p0 )] = 2p0 (2)3 (p0 p),

|pi = a (p)|0i, etc.,
P
(2.116)
(2.117)
dp a (p)a(p) p .
(2.118)
From the transformation behavior (2.109) for the scalar field we infer that under
Lorentz transformations the covariantly normalized objects transform simply,
U (`)a(p)U (`) = a(`p),
U (`)|pi = |`pi.
Note that the non-covariantly normalized ap and |pi have a factor

their transformation rules.
6
(2.119)
q
(`p)0 /p0 in
For a free field a(p) is time-independent, it is the value of the Heisenberg operator at time
zero: a(p, t) = a(p) exp(ip0 t).
43
2.8. WHERE IS THE PARTICLE?
2.8
Where is the particle?
In the following we assume the scalar field to be free. Consider a normalized one
particle state
Z
Z
|i = dp f (p)|pi,
dp |f (p)|2 = 1.
(2.120)
Where is the particle? The answer to this question seems simply: measure its
position. However, there is no natural particle position operator in quantum field
theory. All operators are supposed to be made of the canonical variables (x)
and (x), for which x is just a label enumerating degrees of freedom. We can
of course use our intuition from non-relativistic quantum mechanics and define
localized states by
Z
d3 p ipx 1
0 |pi,
|xi =
(2.121)
e
(2)3
2p
which satisfy
hx|yi = (x y),
d3 x |xihx| = 11 ,
(2.122)
where 11 is the unit operator in the one particle subspace. The 2p0 is needed
to convert to non-relativistic normalization. In terms of these localized states
(x, t) hx|, ti = hx|e
iHt
|i =
dp f (p) 2p0 eipxip
0t
(2.123)
is the candidate probability amplitude and

|(x, t)|2
(2.124)
the probability density for finding the particle at x at time t. It is non-negative

and the total probability is conserved in time,
Z
d3 x |(x, t)|2 = 1.
(2.125)
Yet, there are some puzzling aspects to this interpretation: |(x, t)|2 is not the
time component of a conserved probability current. Going through the usual
steps we have
|(x, t)|2 = ih, t|H|xi(x, t) i(x, t) hx|H|, ti.

t
(2.126)
Now the one particle hamiltonian is given by

hx|H|, ti =
dp hx|pihp|H|, ti =
dp
m2 + p2 hx|pihp|, ti
m2 2 dp hx|pihp|, ti
Z
2
2
=
m (x, t) d3 y h(x y)(y, t).
(2.127)
44
This is a non-local operator. One way to see this is by expansion of the square
root in powers of 2 . This produces an infinite number of derivatives, which
result in shifting (x) around the point x, as in a Taylor series. The function
h(x y) is singular at x = y (it is a distribution), but at large distances it decays
like exp(m|x y|) (cf. Problem 7). So the right hand side of the equation of
probability conservation (2.126) does not look like the divergence of a current. In
fact, there is no such current that satisfies the requirements one would associate
with a probability current.
There is a density which does have an associated current which is sometimes
used as a probability density instead of (x, t). The function7
f (x) = h0|(x)|i =
dp f (p)eipx
(2.128)
satisfies the Klein-Gordon equation because (x) does so (Heisenberg picture).

In general, f (x) 6= f (x) and then the current
j (x) = f (x)i f (x) + i f (x) f (x)
(2.129)
does not vanish. It is easy to verify that it is a conserved current,

( 2 m2 )f (x) = 0 j (x) = 0.
Furthermore,
d3 x j 0 (x) = 1
(2.130)
(2.131)
(verify). Because of these attractive properties j 0 has been used as a probability

density. However, as a probability density j 0 should be non-negative, but there is
no guarantee that this is the case (not even for purely positive frequency solutions
of the Klein-Gordon equation). So we are really stuck with the non-covariant
|(x, t)|2 .
Consider wave packet functions f (p) that are sufficiently concentrated in a
, such that the variation
narrow region
of size p around an average momentum p
of p0 in 2p0 and dp under the integrals in (2.120), (2.121), (2.123) and (2.128)
can be neglected. In these circumstances
j (x) 2
p |f (x)|2 ,
|(x, t)|2 j 0 (x, t),
(2.132)
and we may think of j(x) as the probability current for |(x, t)|2 . The condition
|p| p0 corresponds in position space to |x| >> 1/
p0 1/m, i.e. to scales
much larger than the Compton wavelength.
We have to accept that position of particles is not a natural concept in relativistic field theory, and that the momentum basis is favoured. However, nonrelativistic intuition applies as long as we do not wish to localize particles on
7
For simplicity we use the same symbol f for f (x) as for f (p), but the two functions (f (p)
and f (x)) are very different and should not be confused.
45
2.9. CAUSALITY AND LOCALITY
the scale of the Compton wavelength 1/m. This is a very small scale for the
usual particles, except for neutrinos which are nearly massless, and of course, the
massless photon.
At a deeper level we ought to take into account how a position measurement
is performed. Typically, this is through electromagnetic or gravitational interactions, for which we do have a relativistic formulation the one we are developing
here in terms of quantum fields. For example, as will become clearer after introducing QED, the electromagnetic current of a spinless charged particle has the
form of (2.129) in the limit of vanishing charge: it is the current associated with
the complex scalar field considered in Problem 8. The problem of the absence
of a covariant position-probability current appears to be a red herring.
Another way to measure the position of a particle is through its gravitational
interaction. The energy momentum tensor is the source of the gravitational field,
and for our one particle state it can be shown that8
h|T (x)|i = f (x) f (x) + f (x) f (x)
h
f (x) f (x) + m2 f (x)f (x) ,
(2.133)
which resembles the classical expression. Similar expressions hold for massless
particles.
2.9
Causality and locality
According to standard lore nothing can go faster than the speed of light. Yet, for
a free relativistic particle, the amplitude
0
hx|eiH(tt ) |x0 i A(x, t; x0 , t0 )
(2.134)
is nonzero for spacelike separations (x x0 )2 (t t0 )2 > 0. To see this, use the

representation (2.121) to get
0
A(x, t; x , t ) =
=
d3 p ip(xx0 )ip0 (tt0 )

e
,
(2)3
p =
m2 + p 2 ,
dp 2p0 eip(xx ) ,
= 20 (+) (x x0 ),
(2.135)
dp e
(2.136)
(+) (x x0 ) i
ip(xx0 )
The Lorentz invariant function (+) is nonzero for spacelike separations. A saddle point evaluation9 for x0 = x00 and large |x x0 | shows that it drops off
8
The calculation can be done by expressing the fields in creation and annihilation operators,
but it is simpler to use the Green function techniques to be developed in the next chapter.
9
The calculation is similar to that in Problem 7.
46
exponentially fast on the scale of the Compton wavelength 1/m,

q
(+) (x x0 ) exp(m (x x0 )2 ),
(2.137)
where we have re-expressed the result in Lorentz invariant form. The amplitude
A(x, t; x0 , t0 ) has the same behavior; it is even nonzero for t < t0 !
So it seems that there is a problem with causality. The way out in field theory
is again10 the fact that coordinates x are not among the basic observables, which
are to be constructed from the fields. There is no natural position operator representing a position measurement, of which |xi is an eigenvector with eigenvalues
x. Field theory is causal in the sense that local measurements in spacetime commute if they are performed at spacelike separations. Such measurements cannot
influence each other, which is an expression of causality.
To check this for free fields we evaluate the commutator of two scalar fields,
[(x), (y)]. Using the representation of in terms of creation and annihilation
operators (2.113) and the commutation relations (2.116) we get
[(x), (y)] =
dp eip(xy) eip(xy) = i(x y).
(2.138)
The function (x y) was studied earlier in Problem 1.4: it vanishes for spacelike separations. It follows that local observables constructed out of (x) and its
derivatives (e.g. the energy-momentum tensor) also commute at spacelike separations.
In an interacting Lorentz invariant field theory the commutator of two fields
is no longer a c-number as in (2.138), and in gauge theories quantized in noncovariant gaugees they may not even commute at spacelike separations. However,
commutators of local gauge invariant observables constructed out of the fields
still vanish for spacelike separations. This property is sometimes called local
commutativity, or simply locality. In mathematical approaches to field theory
it is taken as an axiom in the formulation of the theory.
The function (+) (z) in (2.136) is the so-called positive frequency part of
(z), i.e. it is a superposition of exponentials exp(ip0 z 0 ) with only positive
frequencies (p0 > 0). Its nonvanishing at spacelike separations is canceled in
(z) by a negative frequency part:
= (+) + () ,
() (z) = i
dp eipx = i
(2.139)
Z
dp e
+ip0 x0 ipx
(2.140)
This cancelation leading to causality is seen by important authors as a deep

reason for the necessity of antiparticles, traveling backwards in time, with, for
the present case of the real scalar field, particles being their own antiparticles
10
Cf. the discussion in sect. 2.8.
47
2.10. CLASSICAL FIELD
(cf. Weinberg I, Peskin; see also Feynman). To the author of these lecture notes
such remarks are mystifying. Thus far we have not seen any need to introduce the
concept of antiparticles (i.e. particles with opposite charge(s) and the same mass
and spin). Antiparticles will appear naturally in the context of QED and more
general gauge field theory. Instead, we see the locality of field theory (differential
equations of motion, action and energy etc. written as integrals over space(time)),
as the fundamental property leading to causality.
2.10
Classical field
Having familiarized ourselves with a quantum field we would like to understand

its relation to the classical field. Classical behavior in a quantum theory may be
expected only for states with special properties, involving large quantum numbers
and coherence. Here we shall illustrate this with a simple example, the free scalar
field coupled to an external source. The operator field equation is
( 2 + m2 )(x) = J(x),
(2.141)
where J(x) is the external source which we assume to vanish outside a compact
domain in spacetime, in particular
J(x) = 0,
x0 > t+ or x0 < t .
(2.142)
The classical field will be an expectation value of the quantum field. Such expectation values vanish in any state with a definite number of particles because
(x), being linear in the creation and annihilation operators, changes the number
of particles. An interesting state with classical properties is the ground state of
the system before the source is acting, the so-called in-vacuum |0 ini. We shall
calculate the expectation value
c (x) = hin 0|(x)|0 ini,
(2.143)
and find that it has the properties of the classical field (hence the subscript c).
Let us solve the field equation with the help of the retarded Green function
GR (x y) (cf. Problem 1.3),
(x) = in (x) +
d4 y GR (x y)J(y),
(2.144)
where in (x) is the so-called incoming field, which is free, ( 2 + m2 )in (x) =
0. The incoming field is the unique solution of the free Klein-Gordon equation
(defined for all times) which is equal to the field (x) at times before the source
is acting, (x) = in (x) for x0 < t . Similarly, there is an outgoing field, which
is free and defined for all times, with out (x) = (x) for x0 > t+ . Note that
48
J(x) is a c-number while , in and out are q-numbers. The outgoing field can
be expressed in terms of the incoming field by choosing x0 > t+ and using (cf.
Problem 1.3)
GR (x y) = i
dp (eip(xy) eip(xy) ),
Then
out (x) = in (x) +
where
x0 > y 0 .
(2.145)
i
eipx iJ(p)
,
dp eipx iJ(p)
(2.146)
(2.147)
=
J(p)
d4 y eipy J(y)
is the Fourier transform of the source.

We can now identify the incoming and outgoing vacua from the creation and
annihilation operators of (x):
(x) =
dp a(p, x0 )eipx + a (p, x0 )eipx .
(2.148)
By choosing x0 > t+ or x0 < t we get just the a and a of the in and outgoing
fields:
a(p, x0 ) = ain (p), x0 < t ,
= aout (p), x0 > t+ ,
in,out (x) =
(2.149)
i
dp ain,out (p)eipx + ain,out (p)eipx .
(2.150)
From (2.146) (2.150) we see that
aout (p) = ain (p) + iJ(p).
(2.151)
The incoming vacuum satisfies

ain (p)|0 ini = 0,
(2.152)
since it is the ground state at times x0 < t . So, using (2.144), the expectation
value (2.143) is simply given by
c (x) =
d4 y GR (x y)J(y).
(2.153)
It is indeed just the solution of the classical field equation with retarted boundary
conditions c (x) = 0, x < t .
On the other hand, if we analyze the in-vacuum in terms of particles at late
times, it is a coherent state:
aout (p)|0 ini = iJ(p)|0 ini,
(2.154)
49
2.11. SUMMARY
as follows from (2.151). Such states are minimum uncertainty states. They are
superpositions of states with different particle numbers with well defined phase
relations. This follows from the theory of coherent states, which tells us that the
in-vacuum has the form
|0 ini exp
Z
dp iJ(p)aout (p) |0 outi.
(2.155)
Here |0 outi is the out-vacuum, the state with no particles at times x0 > t+ ,
which satisfies
aout (p)|0 outi = 0.
(2.156)
Information on coherent states can be found in Brown sects. (1.7), (1.8) and
books on quantum mechanics.
Summarizing, the classical field can be viewed as the expectation value of the
quantum field in a coherent state. The free field is equivalent to a collection of
harmonic oscillators, which are known to possess states for which the classical
approximation is exact. For interacting theories additional conditions need to be
satisfied, such as effectively weak interactions and large quantum numbers.
2.11
Summary
In the quantum theory the canonical variables (x) and (x) are operators in
Hilbert space with canonical commutation relations. The conserved Noether
charges P and J (which can be expressed in terms of and ) are operators
with the commutation relations of the Poincare algebra. They are the generators
of a unitary representation U (a, `) of Poincare transformations in Hilbert space.
In the Heisenberg picture the equations of motion are the Heisenberg equations
for the operators and . In principle these equations can differ from the classical
field equations, but often they have the same form.
The free field is equivalent to a collection of harmonic oscillators and the
energy-momentum operators P can be simultanously diagonalized using creation and annihilation operators. The ground state is interpreted as the vacuum
and its excitations are interpreted as particles. These particles have identical
mass and zero spin, they are bosons. The (formally infinite) energy of the vacuum is normalized to zero by adjusting a parameter which plays the role of the
cosmological constant in the theory with gravity.
The momentum basis gives the natural description of relativistic particles,
while the position basis is not covariant. Poincare invariance is however guaranteed by the existence of the U (a, `).
Adding simple interaction terms to the free field action, the particles also
interact and they can scattering amongst themselves. Particles with differing
masses can be described by introducing a field for each particle type with corresponding mass parameter in the action. Decay processes are then a natural
possibility.
50
The classical field can be understood as an expectation value of the quantum

field in a quantum state with classical properties.
2.12
Problems
1. Fourier modes
Given (2.9) verify (2.10) by doing the integrals; verify also the statement
about the eigenfunctions of /x.
2. Fock space
The Hilbert space for a system of arbitrarily many particles such as the free
scalar field is called Fock space. A basis is given by |0i, |pi, |p1 p2 i, etc. The
states are normalized as
q
hp|qi = 2p0 (2)3 (p q), p0 = m2 + p2 ,

hp1 p2 |q1 q2 i = hp1 |q1 ihp2 |q2 i + hp1 |q2 ihp2 |q1 i,
etc. In general we get a sum over all permutations of 1, , n (the value
of n will be clear from the context),
hp1 pm |q1 qn i = mn
hp1 |q1 i hpn |qn i.
(2.157)
The completeness relation can be written as

1 = 10 + 11 + 12 + ,
(2.158)
where 10 = |0ih0| and 1n is the unit operator in the n-particle subspace,

1n = 1
n!
dp1 dpn |p1 pn ihp1 pn |.
(2.159)
Verify this by taking matrix elements with |q1 qm i.

3. Phase space integral and scattering
In this problem we evaluate the remaining integrals encountered in twoparticle scattering and decay.
The integral (called a phase space integral)
I(p) =
dqi =
dq1 dq2 (2)4 4 (q1 + q2 p),
(2.160)
(2.161)
d 3 qi
q
(2)3 2 q2i + m2i
i = 1, 2,
51
2.12. PROBLEMS
is Lorentz invariant, I(p) = I(`p). It is convenient to evaluate it in the

center of mass frame defined by p = 0, in the following steps:
a) integrate over q2 using the momentum conserving delta functions,
b) choose spherical coordinates
q1 (|q|, , ),
d3 q1 = |q|2 d|q| d,
d = d(cos ) d
(2.162)
c) for the |q| integral use the energy conserving delta function and the
general formula
Z
b
a
dx (f (x))g(x) =
X
j
1
g(xj ),
|f 0 (xj )|
(2.163)
where the summation is over the zero(s) xj of f (x) in the interval

(a, b).
q
2
In the present case the argument of the delta function, m1 + |q|2 +
q
m22 + |q|2 p0 , has only one zero. We use s p2 , which is p20 in the
center of mass.
Verify that the result is given by

I=
|q|
|q| Z
,
d
=
16 2 s
4 s
with
|q|2 =
s = p2 ,
s2 + (m21 m22 )2 m21 + m22
.
4s
2
(2.164)
(2.165)
The application of I(p) to two-particle decay is straightforward.

In the application to scattering 1 + 2 3 + 4, q1 p3 , q2 p4 , p
is the total incoming momentum, p = p1 + p2 , and may be the angle
between p1 and p3 . The invariant amplitude |T |2 is a Lorentz invariant
function of the momenta, so a function of the two independent invariants
s = (p1 + p2 )2 = (p3 + p4 )2 , and t = (p1 p3 )2 = (p2 p4 )2 . The
other invariant11 u = (p1 p4 )2 = (p2 p3 )2 is not independent, because
s + t + u = m21 + m22 + m23 + m24 (verify).
The flux factor is given by
q
4E1 E2 v12 = 4 (p1 p2 )2 m21 m22 = 4|p1 | s,
with |p1 | given by (2.165) with |q| |p1 |.
(2.166)
For the total cross section we need to multiply by 1/2 if the two particles
in the final state are identical, to avoid double counting. This is the same
factor 1/2! as in (2.159), n = 2.
11
The invariants s, t and u are called Mandelstam variables.
52

For the differential cross section we do not integrate over and . In the
center of mass frame
1 |p3 |
|T |2 .
(2.167)
d = d
64 2 s |p1 |
Verify.
Alternatively, we can specify the invariant momentum transfer t = (p3

p1 )2 . It is linearly related to cos , dt = 2|p1 | |p3 | d cos , so d/dt can
be simply be read off from d/d cos = 2d/d. We can also insert a
constraining delta function (t + (p1 p3 )2 ) in the integral I(p1 + p2 ) which
gives the same result (see e.g. Brown section 3.4):
d
1
=
|T |2 ,
dt
64s p21
which holds also in the unequal mass case.
(2.168)
4. Commuting with generators

Verify the commutation relations (2.103), (2.104). Hint: Write out explicitly P0 , Pm , Jmn , J0n in terms of the canonical and . Since the generators
are constant in time, choose the time in P and J equal to the argument
x0 of (x), and use the canonical equal time commutation relations (2.6),
which are valid at any time t:
[(x, t), (y, t)] = i 3 (x y),
[(x, t), (y, t)] = [(x, t), (y, t)] = 0.

(2.169)
5. Commutators
Consider operators qa , pa , with the canonical commutation relations [qa , pb ] =
iab , [qa , qb ] = 0, [pa , pb ] = 0. In terms of these, and (c-number) matrices
Mab , which are zero on the diagonal, Maa = 0 (no summation), define operators A(M ) = iMab pa qb , or in matrix notation A(M ) = ipT M q. Calculate the commutators [qa , A(M )], [pa , A(M )], and show that [A(M1 ), A(M2 )] =
A([M1 , M2 ]).
As an application, let a be a continuous index a x, ab (x y).
Obtain the commutator [Jk , Jl ], where Jk = 21 klm Jlm is the generator for
rotations (e.g. for Jk , Mab iklm xl m (x y)). Likewise, obtain the
commutator of Jk with the translation generators, [Jk , Pl ] =?
6. Finite transformations
Let U = exp(iF ) be unitary, F = F . Use the identity
U A U = A + i[A, F ] +
= A+
i2
[[A, F ], F ] +
2
in
[ [A, F ], , F ]
n=1 n!
(2.170)
53
2.12. PROBLEMS
to prove (2.107).
7. Relativistic one-particle hamiltonian
Consider (2.127). Expanding the square root to finite order, h(xy) would
consist of derivatives of delta functions. For example in the non-relativistic
approximation
!
2
h(x y) = m +
(x y).
(2.171)
2m
Integration over y gives the usual free particle hamiltonian acting on (x, t)
as a differential operator, in addition to the rest-energy contribution m.
Using Fourier transformation we can express the full h in the form
h(x) =
d3 p ipx q 2
m + p2 .
e
(2)3
(2.172)
The integral does not converge but this merely reflects that h(x) is a distribution, singular around the origin. For large x we can use the saddle point
method to get the dominant behavior of h(x), e.g. by writing x = r
x,
h(x) =
1
d3 p
exp
ip
x
r
+
ln(m2 + p2 ) ,
3
(2)
2

(2.173)
and expanding about the correct stationary point of the exponent for r
. Verify that the stationary point is given by p im
x and that h(x)
exp(mr) for large mr.
8. O(2) model, complex scalar field
Consider a model with two scalar fields 1 and 2 with action12
S=
1
1
1
d x k k + m2 2 + (2 )2 .
2
2
4
4
(2.174)
We use the summation convention also for indices such as k: k k

P2
2
k=1 k k . The above action is invariant under rotations in internal
space
01 = cos 1 sin 2 ,
02 = sin 1 + cos 2 ,
S[0 ] = S[]. (2.175)
Infinitesimal rotations can be written in the form

k = kl l ,
12 = 21 = 1, 11 = 22 = 0,
(2.176)
with infinitesimal rotation angle . To a continuous symmetry corresponds

a conserved quantity, usually called charge (Noethers theorem). This can
12
All objects are bare, we ignore renormalization.
54

be seen as follows. Consider infinitesimal rotation angles depending on
space-time: (x). The action is now in general not invariant anymore
because depends on x; for an infinitesimal rotation,
S =
d x j =
d4 x j .
(2.177)
However, if the k satisfy the field equations (equations of motion), then

S = 0 and we have a local balance equation (a conserved current),
j = 0,
(2.178)
with a corresponding conserved charge

Q=
d3 x j 0 .
(2.179)
a. Show that the current is given by

j = kl k l .
(2.180)
b. Derive the field equations from the stationary action principle.

c. Verify using the field equations that j = 0.
d. In the quantum theory Q is an operator, which can be expressed in the
creation and annihilation operators at time zero. Show that
Q=
apk (ikl )apl .
(2.181)
e. Consider the free theory with = 0. Choosing the vacuum energy to be

zero, the energy-momentum operator is given by
P =
apk apk p .
(2.182)
Since Q is time independent it should be possible to diagonalize Q and P 0

simultaneously. This can be done as follows. Define
1
ap = (ap1 iap2 ).
2
(2.183)
Show that in terms of the new creation and annihilation operators

Q =
X
p
P =
X
p
(ap+ ap+ ap ap ),
(2.184)
(ap+ ap+ + ap ap ) p .
(2.185)
55
2.12. PROBLEMS
f. Verify that Q ap+ |0i = +ap+ |0i
The usual words that go with this model is as follows: ap+ is the creation
operator for particles, ap is the creation operator for antiparticles. The
particles have charge +1, the antiparticles have charge 1 and Q counts
the number of particles minus the number of antiparticles. Particles and
antiparticles have the same mass.
In the generalization to n fields with k = 1, 2, . . . , n (n > 2) the internal
symmetry group is larger (O(2) O(n)), with n(n 1)/2 Noether charges,
and no such natural division into particles and antiparticles.
Returning to n = 2, an often used notation is in terms of complex fields
1
(1 i2 ).
2
(2.186)
In this case and are treated as independent variables.

g. Show that the action can be rewritten as
S=
d4 x + m2 + ( )2 .
(2.187)
Rederive Noethers theorem and its consequences in this notation. Go

through the canonical formalism, quantization, etc. Express and its
canonical conjugate in terms of the creation and annihilation operators.
Calculate the matrix element h|j (x)|i for a wave packet state of a single
particle as in (2.120) and compare with (2.129).
9. Pauli-Villars regularization of h0|T |0i
The integral (cf. (2.40)) in
4
h0|T |0i = 0 +
2(2)3
00
4 Z
= 0 +
2(2)3
dp p2
ci m2i + p2
ci
dp p
(2.188)
m4i
1 m2i
3 + ,
p+
2 p
8p
!
converges at the upper limit provided that

X
i
ci = 0,
X
i
ci m2i = 0,
ci m4i = 0.
(2.189)
where c1 = 1 and m1 = m. Verify that the above conditions can be satisfied

with four cs and ms. Hint: make a suitable choice of the ratios ri = mi /m,
i = 2, 3, 4, and solve for c2 , c3 and c4 .
Verify that the regularized surface term in (2.43)
ishes in the limit .
i ci
m2i + 2 van-
56

Verify that h0|T kl |0i is also finite.
One can add the requirement that the integrand of the originally divergent
integral should be recovered when the regulator masses are sent to infinity.
To implement this an additional c5 and m5 would be needed such that in
addition to the requirements (2.189) also
X
ci mi = 0.
(2.190)
i2
Calculate in terms of the ci and mi . Hint 1: calculate h0|T |0i = 4.

Hint 2: check the identity
1
=
2
dp4
1
.
2
2 + p24
(2.191)
Hint 3: write the integral for h0|T |0i as a four-dimensional integral in

euclidean momentum-space and evaluate this with the help of a spherical
cutoff, using spherical coordinates,
4( 0 ) =
d4 p X
1
ci m2i 2
4
(2) i
mi + p2 + p24
X
1
2 2 Z
3
ci m2i 2
dp
p
,
=
4
(2) 0
mi + p 2
i
Z
X
X
1
=
dp p3
,
ci m2i
ci m2i 2
mi + p 2
0
i
i
. (2.192)
In the second step we used that in four dimensions the integral over angles
equals 2 2 .
The remaining integral is elementary and the limit can be taken
without a problem because of the properties of the ci and mi . The result
is (verify)
m4
m2 X m4i
m2i
4 = 40 +
c
ln
+
ln
,
(2.193)
i
2
16 2
2
2
i2 16
where 2 is some arbitrary mass scale introduced for convenient separation
of the dependence (the result does not depend on ).
If we now let the regulator masses approach infinity, mi , i = 2, 3,
4, 5, the regulator part appears to diverge quartically, m4i , which has to
be compensated by 0 such that remains finite. In common parlor: 0 is
renormalized to .
Example, using Mathematica with m2 = 2rm, m3 = 3rm, m4 = 4rm,
m5 = 5rm, gives

4 = 40 + m4 a4 r 4 + a2 r 2 + a0 ln r + a00
(2.194)
57
2.12. PROBLEMS
plus terms that vanish as r , with
a4 =
etc.
30
(32 ln 4 279 ln 9 + 640 ln 16 375 ln 25),
7
(2.195)
58
Chapter 3
Path integral methods
The path integral approach to quantization has resulted in a powerful language
for quantum field theory. Here we shall try to be brief and concentrate on
applications to perturbation theory. We have seen in simple cases that calculations need answers for the vacuum expectation value of products of field,
h0|(x
1 ) (x
n )|0i, and we shall develop the calculational tools for this.
3.1
Path integral for quantum mechanics
Consider a simple system described by the Lagrange function L = L(q, q,

t) or
the corresponding Hamilton function H = H(p, q, t),
1
L = mq2 V (q, t),
2
H=
p2
+ V (q, t),
2m
(3.1)
where p and q are related by p = L/ q = mq.

In the quantum theory p and q
1
become operators p, q with [
q , p] = i
h. The coordinate basis |qi is characterized
by
q|qi = q|qi,
(3.2)
hq 0 |qi = (q 0 q),
dq |qihq| = 1,
(3.3)
(t0 , t00 ). In case the potential V

The time evolution is described by the operator U
is time independent and
does not depend explicitly on time, the hamiltonian H
the evolution operator is given by
(t0 , t00 ) = exp[iH(t
0 t00 )/
U
h],
(3.4)
where we have temporarily made h

explicit. Path integral quantization will give
us a representation of the matrix element
1
When helpful, we indicate operators in Hilbert space by a.
59
60
CHAPTER 3. PATH INTEGRAL METHODS
(t0 , t00 )|q 00 i,

hq 0 |U
(t0 , t00 )|q 00 i =

hq 0 |U
d[q] exp
Here S is the action functional of the system,

S[q] =
and
t0
t00
i
S[q] .
h
dt L(q(t), q(t)),
(3.5)
(3.6)
d[q] symbolizes an integration over all functions q(t) such that

q(t0 ) = q 0 ,
q(t00 ) = q 00 .
(3.7)
Note that h
appears explicitly in (3.5), all other symbols are classical. Instead
of working with q-numbers (operators) p and q, the path integral allows us to
work with time dependent c-numbers (commuting numbers) q(t).
The path integral is a summation over all paths (trajectories, histories)
q(t), with given end points q 0 , q 00 . The classical path, which satisfies the equation
of motion
L
L
S
=
= 0,
(3.8)
q(t)
q(t) t q(t)
is only one out of infinitely many possible paths. Each path has a weight
exp(iS/
h). If h
is relatively small such that the phase exp(iS/
h) varies rapidly
over the paths, then a stationary phase approximation will be good in which the
clasical path and its small neighborhood gives the dominant contribution. The
other extreme is where the variation of S/
h is of order 1 and the system is in the
quantum regime. In the following
we
shall
use units again in which h
= 1.
R
A formal definition of d[q] is given by
Z
d[q] =
t00 <t<t0
dq(t),
(3.9)
i.e. for every t (t00 , t0 ) we integrate over the domain of q, e.g. < q < .
Q
The definition is formal because the continuous product t still has to be defined.
We shall give such a definition with the help of a discretization procedure.
3.2
Regularization by discretization
To define the path integral properly we discretize time in small units a, writing
t0 t00 = N a, tn = na, n = 0, 1, . . . , N,
q(tn ) = qn , q0 = q 00 , qN = q 0 . (3.10)
For a smooth function q(t) the time derivative q(t)
can be approximated by
q(t
n ) (qn+1 qn )/a, such that the discretized Lagrange function may be written
as
m
1
1
Ldiscr
(qn+1 , qn ) = 2 (qn+1 qn )2 Vn+1 (qn+1 ) Vn (qn ).
(3.11)
n
2a
2
2
61
3.2. REGULARIZATION BY DISCRETIZATION
We have devided the potential term equally between times n and n + 1, and
Vn (q) V (q, tn ). Except at the end points, the two halves add up to one in the
discretized action defined by
Sdiscr [q] = a
N
1
X
Ldiscr
(qn+1 , qn )
n
n=0
m
a
= VN (q 0 ) + (q 0 qN 1 )2 aVN 1 (qN 1 )
2
2a
m
+ (qN 1 qN 2 )2 aVN 2 (qN 2 )
2a
a
m
+ + (q1 q 00 )2 V0 (q 00 ).
2a
2
(3.12)
A tentative definition of the path integral is now

Z
d[q] exp{iS[q]} = lim cN

N
NY
1
dq) exp{iSdiscr [q]},
(3.13)
n=1
where the constant c is still to be specified.

To show that this definition of the path integral gives us the evolution operator, we introduce a minimal step evolution operator T by giving its matrix
elements
hq1 |Tn |q2 i = c exp[iaLdiscr
(q1 , q2 )].
(3.14)
n
The operator T is called the transfer operator, its matrix elements the transfer
matrix. In terms of the transfer operator we have
c
NY
1
n=1
dq) e
iSdiscr [q]
dq1 dqN 1 hq 0 |TN 1 |qN 1 i hq2 |T1 |q1 ihq1 |T0 |q 00 i
= hq 0 |TN 1 TN 2 T0 |q 00 i.
(3.15)
Now we are almost done. First, with a suitable choice of the constant c the
transfer operator can be written in the form
2
Tn = eiaVn+1 (q)/2 eiap /2m eiaVn (q)/2 .
(3.16)
Taking matrix elements between hq1 | and |q2 i and comparing with (3.11), we see
that this formula is correct if
2 /2m
hq1 |eiap
|q2 i = c eim(q1 q2 )
2 /2a
(3.17)
Inserting eigenstates |pi of the momentum operator p using

ipq
hq|pi = e ,
dp
|pihp| = 1,
2
(3.18)
62
q"
t"
Figure 3.1: Typical path contributing to the discretized path integral.

and the gaussian integral2
Z
1
dp exp (ia/m)p2 + irp =
2
2
1 (ir)2
exp
ia/m
2 ia/m
(3.19)
we find that (3.17) is true provided that we choose

c=
m
=
2ia
m i/4
e
.
2a
(3.20)
Second, using the Baker-Cambell-Haussdorff formula it is easy to show that (cf.

Problem 1)
n + O(a3 )] = U
(tn+1 , tn ) + O(a3 ).
Tn = exp[iaH
(3.21)
Hence,
TN 1 TN 2 T0 = U (t0 , t00 ) + O(a2 ),
(3.22)
and together with (3.13), (3.15) we get the desired result

Z
d[q] exp{iS[q]} = hq 0 |U (t0 , t00 )|q 00 i.
(3.23)
For finite N the errors are of order a2 . This is important in practical computations
and is the reason why we have devided the potential term in (3.11) over two time
slices.
In case V (q, t) does not depend explicitly on time, the subscript n in Ln , Vn ,
0 t00 ) + O(a2 )].
n can be dropped, and Qn Tn = TN = exp[iH(t
Tn and H
2
This is a convergent integral. It is more appropriately called a Fresnel integral, but we shall
generically call such integrals gaussian as if the exponential were real.
63
3.3. IMAGINARY TIME
In the limit N , the discretized action Sdiscr [q] becomes equal to the continuum action S[q] when we substitute smooth functions q(t). However, because
the qn are integrated over on every time slice n, such smoothness is not present
typically in the integrand of the path integral (typical paths qn have highly discontinuous first derivatives cf. Fig. 3.1). Hence, a continuum limit Sdiscr [q] S[q]
cannot be taken in the integrand of the path integral. We regard the path integral
as defined by the discretization procedure, with the limit a 0 (N ) taken
only at suitable points in calculations.3 In the following we shall mostly work
formally, in a notation in which the limit has already been taken (even under the
integral), which gives nice intuitive formulas. However, we may have to return
sometimes to the original definition to avoid mathematical ambiguities. In the
lattice field theory method such ambiguities are avoided by working only with
the well-defined discretized forms.
3.3
Imaginary time
It is interesting to make an analytic continuation to imaginary time according to

the substitution t it. This can be justified if the potential V (q) is bounded
from below. For example, this is the case for the anharmonic oscillator
1
1
V (q) = m 2 q 2 + q 4 ,
2
4
0,
(3.24)
where q (, ).
Consider the discretized path integral (3.15). The integrations over the variables qn converge at large qn , and continue to converge if we rotate a in the
complex plane according to
a = |a|ei , : 0 .
(3.25)
2
The reason is that for all (0, ) the real part of the exponent in (3.15) is
negative:
i
i
1
=
( sin + i cos ),
=
i
a
|a|e
|a|
ia = i|a|ei = |a|( sin i cos ).
(3.26)
The result of this analytic continuation in a is that the discretized path integral
takes the form
hq |T N |q 00 i = |c|N
0
SIdiscr
3
= |a|
Z Y
N
1
X
n=0
"
dqn ) exp[SIdiscr ],
(3.27)
#
1
1
m
2
(q
q
)
+
V
(q
)
+
V (qn ) .
n+1
n
n+1
2|a|2
2
2
There are also other definitions possible, see e.g. Problem 2. The discretization method
has the advantage that it can be applied also to gauge theories.
64
Here the subscript I denotes the imaginary time version of S. After transformation to imaginary time the transfer operator takes the hermitian form
2
T = ebV (q)/2 ebp /2m ebV (q)/2 ,
b = |a|.
(3.28)
This is a positive operator, i.e. all its expectation values and hence also all its
eigenvalues are positive. We may therefore define a discretized hermitian Hamil discr according to
ton operator H
discr ],
T = exp[bH
discr = H
+ O(b2 ),
H
(3.29)
and interpret this as the hamiltonian of the discrete system.

A natural object in the imaginary time formalism is the partition function
Z = Tr e
+ t )
H(t
dqhq|eH(t+ t ) |qi = lim Tr T N ,

N
(3.30)
where we think of t+ (t ) as the largest (smallest) time under consideration, with

t+ t = N a. From quantum statistical mechanics we recognize that Z is the
canonical partition function corresponding to the temperature
T = (t+ t )1 ,
(3.31)
in units where Boltzmanns constant kB = 1. The path integral representation

of Z is obtained by setting in (3.27) qN = q0 q (q 0 = q 00 q) and integrating
over q:
Z
Z=
d[q] eSI .
(3.32)
pbc
Here pbc indicates the fact that the integration is now over all discretized functions q(t), t < t < t+ , with periodic boundary conditions, q(t+ ) = q(t ).
The integrand exp(SI ) in the imaginary time path integral is real and
bounded from above. This makes numerical calculations and theoretical analysis very much easier. Furthermore, in the generalization to field theory to be
given later there is a direct connection to Statistical Physics, which has led to
many fruitful developments. The imaginary time formulation can be sufficient
to extract the relevant physical information, without the need to continue back
to real time. A prime example is the mass spectrum of QCD, where the eigenstates of the hamiltonian are bound states of quarks and gluons. The analogy of
exp(SI ) with a Boltzmann factor makes it possible to address this complicated
problem by Monte Carlo computations.
In the following we shall continue with ordinary but slightly modified real
time: we assume the rotation angle > 0 to be infinitesimal, i.e. very small and
going to zero at suitable stages of the calculations. This has several advantages.
For now we note that convergence is improved: integrals like
Z
dq exp
i1
1
m(q q 0 )2 ia m 2 q 2 q n
a2
2

(3.33)
65
3.4. EXTERNAL FORCE TECHNIQUE
are absolutely convergent for > 0 for n = 1, 2, and not only (conditionally
convergent) for n = 0 (which is the case for = 0). We also note at this point
that it does not hurt to add another convergence device, which will be useful
later: replace
2 2 i,
(3.34)
with > 0 infinitesimal. This produces the convergence factor exp(a 12 mq 2 ) (we
are assuming m > 0). We shall suppress and in our notation in the following,
unless explicitely needed.
3.4
External force technique
An important technique in field theory is probing the response of the system to

an external source, the analogue of which is here an external force. The potential
is taken to be of the form
V (q, t) = V (q) f (t)q,
(3.35)
where f (t) is an arbitrary function of time. We redefine the notation and use S
and L for the action and lagrangian of the unperturbed system,
S=
dt L,
1
L = mq2 V (q).
2
(3.36)
From the equations of motion

m
q=
V
+ f.
q
(3.37)
follows that f (t) has the interpretation of an external force. The path integral
for the evolution matrix is
hq 0 |U (t0 , t00 )|q 00 i =
d[q] eiS[q]+i
dt f (t)q(t)
(3.38)
We are going to differentiate this expression with respect to f (probing the

system).
Since we shall be mostly using the Heisenberg picture in which the operators
are time dependent,
q(t) = U (t, 0) q U (t, 0),
(3.39)
it will be convenient to use corresponding coordinate basis vectors |q, ti in Hilbert
space with the properties
q(t)|q, ti = q|q, ti,
hq, t|q 0 , ti = (q q 0 ),
(3.40)
Z
dq |q, tihq, t| = 1.
(3.41)
66
This basis can be chosen as

(0, t)|qi.
|q, ti = U (t, 0) |qi = U
(3.42)
Note that |q, ti is not the Schrodinger state at time t which developed out of the
initial |qi at time zero: this would be U (t, 0)|qi.
For the differentiation with respect to f we use again the method of small
time-steps a, a = (t0 t00 )/N , tk = ka, k = 0, , N , N . We write
(t0 , t00 )|q 00 i = hq 0 |
hq 0 , t0 |q 00 , t00 i = hq 0 |U
=
dq1 dqN 1
NY
1
k=0
NY
1
k=0
(tk+1 , tk )|q 00 i
U
hqk+1 , tk+1 |qk , tk i.
(3.43)
The one-step transition amplitudes are approximated by the matrix elements of

the transfer operator,
(tk + a, tk )|qk i hq1 |Tk |q2 i
hqk+1 , tk+1 |qk , tk i = hqk+1 |U

1
= c exp iaL(q1 , q2 ) + ia [f (tk+1 )q1 + f (tk )q2 ] .(3.44)
2
Assuming tj t (t0 , t00 ) we then have
hq 0 , t0 |q 00 , t00 i
if (t)
=
iaf (tj )
Z
dq1 dqN 1
dq1 dqN 1
NY
1
NY
1
k=0
hqk+1 , tk+1 |qk , tk i
k=j+1
hqj+1 , tj+1 |qj , tj ihqj , tj |qj1 , tj1 i

iaf (tj )
j2
Y
k=0
dq1 dqN 1
NY
1
k=j+1
hqj+1 , tj+1 |qj , tj i qj hqj , tj |qj1 , tj1 i

j2
Y
k=0
dq hq 0 , t0 |q, ti q hq, t|q 00 , t00 i,
(3.45)
as N . Note that q above is just a dummy integration variable. The

combination
Z
dq |q, ti q hq, t| = q(t),
(3.46)
67
3.4. EXTERNAL FORCE TECHNIQUE

is the Heisenberg operator at time t. So we have the result
hq 0 , t0 |q 00 , t00 i = hq 0 , t0 |
q (t)|q 00 , t00 i.
if (t)
(3.47)
Differentiating again, using the chain rule in (3.45) we get two contributions,
hq 0 , t0 |q 00 , t00 i =
if (t2 ) if (t1 )
if (t2 )
=
dq1 hq 0 , t0 |q1 , t1 i q1 hq1 , t1 |q 00 , t00 i
(hq 0 , t0 |q1 , t1 i) q1 hq1 , t1 |q 00 , t00 i

if (t2 )
Z
+ dq1 hq 0 , t0 |q1 , t1 i q1
(hq1 , t1 |q 00 , t00 i) .
if (t2 )
dq1
The first term only contributes for t2 > t1 , the second term only contributes for
t2 < t1 (recall that we only consider times t00 < t1,2 < t0 ). We can distinguish
these two cases with the help of the Heavyside step function
(t) = 1,
= 0,
t>0
t < 0,
(3.48)
and the result of the differentiation can be expressed in the form (3.47),
Z
0 0 00 00
hq , t |q , t i =
dq1 [(t2 t1 )hq 0 , t0 |
q (t2 )|q1 , t1 i q1 hq1 , t1 |q 00 , t00 i
if (t2 ) if (t1 )
+ (t1 t2 )hq 0 , t0 |q1 , t1 i q1 hq1 , t1 |
q (t2 )|q 00 , t00 i]
= (t2 t1 )hq 0 , t0 |
q (t2 )
q (t1 )|q 00 , t00 i
+ (t1 t2 )hq 0 , t0 |
q (t1 )
q (t2 )|q 00 , t00 i.
(3.49)
This can be compactly written with the help of the time ordering instruction
symbol T :
hq 0 , t0 |q 00 , t00 i = hq 0 , t0 |T q(t1 )
q (t2 )|q 00 , t00 i,
if (t2 ) if (t1 )
(3.50)
which is defined as
T q(t1 )
q (t2 ) = (t1 t2 )
q (t1 )
q (t2 ) + (t2 t1 )
q (t2 )
q (t1 ).
(3.51)
The above process evidently generalizes to
hq 0 , t0 |q 00 , t00 i = hq 0 , t0 |T q(t1 ) q(tn )|q 00 , t00 i,

if (t1 )
if (tn )
(3.52)
with T the instruction to write the operators in order of decreasing times (writing
this in terms of functions gets cumbersom for n > 2). Note that these so-called
time ordered products are symmetric under interchanging labels ti tj .
68
Setting f = 0 after differentiation we get the expansion coefficients of a

Taylor series in powers of f (called a Volterra series in case of functionals). So,
summing the series we have
dt1 dtn
hq , t |q , t i[f ] =
hq 0 , t0 |q 00 , t00 i [0]
f (t1 )
f (tn )
n=0 n!
f (t1 ) f (tn )
n Z
X
i
dt1 dtn hq 0 , t0 |T q(t1 ) q(tn )|q 00 , t00 i[0] f (t1 ) f (tn )
=
n!
n=0
Z
n
X
1 0 0
=
hq , t |T i dt f (t)
q (t) |q 00 , t00 i[0]
n=0 n!
0
00
00
= hq 0 , t0 |T ei
dt f (t)
q (t)
|q 00 , t00 i[0],
(3.53)
where we indicated the functional dependence on f (t) in square brackets. Note

that on the right hand side q(t) evolves in time according to the hamiltonian with
the external force set to zero.
On the other hand, we have path integral representations for various expressions:
Z
R
hq 0 , t0 |q 00 , t00 i[f ] = d[q] eiS[q]+i dt f (t)q(t) ,
(3.54)
and
00
00
d[q] eiS[q] q(t1 ) q(tn ),
(3.55)
R
R
ei dt f (t)q(t) = q(t1 ) q(tn ) ei dt f (t)q(t) .

if (t1 )
if (tn )
(3.56)
hq , t |T q(t1 ) q(tn )|q , t i[0] =

because
3.5
Ground state expectation values
In field theory we are especially interested in expectation values in the ground

state. We now illustrate how these can be calculated in our quantum mechanical
is the hamiltonian of the model with f = 0. The
model. In the following H
is bounded from below, i.e.
crucial assumption will be that the spectrum of H
there is a ground state with minimal energy:
n i = En |n i, H|
0 i = E0 |0 i En6=0 > E0 .
H|
(3.57)
We have made the further simplifying assumption that there is an energy gap
between the ground state energy E0 and the first excited state energy E1 .
Suppose now that f (t) is non-zero only in a bounded region t < t < t+ .
for t < t and t > t+ , and we
Then the hamiltonian is the time-independent H
may write
(t0 , t00 )|q 00 i = hq 0 |U
(t0 , t+ )U
(t+ , t )U
(t , t00 )|q 00 i
hq 0 |U
00
= hq 0 |eiH(t t+ ) U (t+ , t )eiH(t t ) |q 00 i.
(3.58)
69
3.5. GROUND STATE EXPECTATION VALUES

We now use completeness,
X
n
|n ihn | = 1,
(3.59)
and insert intermediate states to obtain for the right hand side
X
mn
0
(t+ , t )|n ihn |q 00 ieiEn (t t00 ) .
eiEm (t t+ ) hq 0 |m ihm |U
(3.60)
Consider now letting t0 , t00 , keeping in mind the fact that our times
are rotated over a small negative angle away from the real axis, as motivated
in Sect. 3.3. Let us make this temporarily explicit by writing
t = (1 i),
> 0.
(3.61)
With > 0, 0 + and 00 the energy exponentials such as

0
eiEm (t t+ ) = e(
+ )Em
e(
+ )iEm
(3.62)
suppress the excited states relative to the ground state. So the ground state
contribution dominates,
hq 0 , t0 |q 00 , t00 i = hq 0 |U (t0 , t00 )|q 00 i
(3.63)
00 )
iE0 (t0 t+ )
0
00
iE
(t
t
e
hq |0 ih0 |U (t+ , t )|0 ih0 |q ie 0
.
The neglected terms are suppressed by factors exp[ 0 (Em E0 )] and exp[ 00 (Em
E0 )], m 6= 0, relative to the ground state. Note the appearence of the groundstate wave function
hq|0 i = 0 (q),
(3.64)
which provides a way to calculate 0 from the path integral.
In a similar fashion we can derive, using (3.53), (3.42) and U (0, t) = eiHt ,
hq 0 , t0 |q 00 , t00 i[f ] = hq 0 , t0 |T ei
0
dt f (t)
q (t)
|q 00 , t00 i[0]
eiE0 t hq 0 |0 ih0 |T ei
dt f (t)
q (t)
00
|0 i[0] h0 |q 00 ieiE0 t .
We can cancel the dependence on the boundaries in time by deviding by the same
expression with f = 0, and arrive at the neat formula
h0 |T ei
dt f (t)
q (t)
|0 i =
=
hq 0 , |q 00 , i[f ]
hq 0 , |q 00 , i[0]
R
d[q] eiS[q]+i dt f (t)q(t)

R
,
d[q] eiS[q]
(3.65)
where we used h0 |0 i = 1. Since the boundary effects drop out in the ratio
(3.65), we may as well choose boundary values q 0 = 0, q 00 = 0, since this allows us
70
to do partial integrations without having to worry about boundary terms. Thus

we define the path integral for zero boundary values at infinite times,
Z[f ] h0, |0, i =
d[q] eiS[q]+i
in terms of which
h0 |T e
dt f (t)
q (t)
|0 i =
dt f (t)q(t)
(3.66)
Z[f ]
.
Z[0]
(3.67)
In field theory the analogous expressions will turn out to be very useful.
3.6
Harmonic oscillator
The results so far will now be illustrated by explicit calculations for the important
case of the harmonic oscillator. We shall be brief and only give a formal evaluation
of the ratio of path integrals with and without external force. Our first task is to
evaluate Z[f ] in (3.66), or rather the ratio Z[f ]/Z[0]. The action is given by
S[q] =
dt
1 2 1
1
mq m 2 q 2 =
2
2
2

2
dt q m 2 + m 2 q.
t
!
(3.68)
On the r.h.s. we made a partial integration, for which the boundary terms vanish
in the case (3.66). We now write this in suggestive matrix form,
1
S = q T G1 q,
2
(3.69)
with continuous matrix index t and

2
G (t, t ) = m 2 + m 2 (t t0 ).
t
1
(3.70)
For example,

G q (t) =
2
dt G (t, t )q(t ) = m 2 + m 2 q(t).
t
0
(3.71)
Note that G1 is symmetric:

G1 (t, t0 ) = G1 (t0 , t).
(3.72)
The path integral now looks like a multiple gaussian integral4

Z[f ] =
4
1
d[q] exp iq T G1 q + if T q ,
2

(3.73)
In discretized form it is an ordinary multiple gaussian integral, and G1 (tn , tn0 ) is just an
ordinary matrix.
71
3.6. HARMONIC OSCILLATOR

and we shall evaluate the dependence on f by a shift of integration variables,
q = q 0 + Gf,
(3.74)
where G is the inverse of G1 under the given boundary conditions,

G1 G = 1.
(3.75)
This inverse is also symmetric, GT = G, which will be verified below. The shift
makes the exponent purely quadratic in q 0 ,
1
1
1
q T G1 q + f T q = q 0T G1 q 0 + f T Gf.
2
2
2
(3.76)
The integration measure is invariant under shifts,

d[q] = d[q 0 ],
or
So we get
dq(t) F [q] =
dq 0 (t) F [q 0 + Gf ].
1
Z[f ] = Z[0] exp i f T Gf ,
2

(3.77)
(3.78)
or more explicitly
1
Z[f ] = Z[0] exp i
2

dt dt f (t)G(t, t )f (t ) .
(3.79)
From the explicit form of (3.75),

Z
2
dt G (t, t )G(t , t ) = m 2 + m 2 G(t, t0 ) = (t t0 ),
t
00
00
00
(3.80)
we see that G is the Green function corresponding to the differential operator

G1 . It can be found by Fourier transformation. Assuming G(t, t0 ) to depend on
the difference t t0 we try
G(t, t0 ) =
dp
0
G(p)eip(tt ) .
2
(3.81)
Since the Fourier transform of the Dirac delta function is 1 we have

m(p2 + 2 )G(p) = 1, or G(p) =
1
.
p2 )
m( 2
(3.82)
A prescription is needed to deal with the pole in the integrand of (3.81). The prescription for the present situation follows from the discussion about convergence
given at the end of Sect. 3.3: we give 2 a small negative imaginary part:
2 2 i.
(3.83)
72
x
x
Figure 3.2: Position of the poles of ( 2 p2 i)1 in the complex p plane.

Evaluating the integral by contour integration leads to (cf. figure 3.2)
i
i(tt )
dp 1 eip(tt )
ei(tt )
0 e
0
=
(t
t
)
+
(t
t)
.
2 m 2 p2 i
2m
2m
Hence,
(3.84)
ei|tt |
iG(t, t ) =
.
(3.85)
2m
It is symmetric in t t0 . It is complex. It is the analogue of the Feynman
propagator for a scalar field.
The remaining path integral Z[0] is formally just a multiple gaussian integral.
It can be evaluated by expanding q(t) in terms of a complete set of orthonormal
eigenfunctions of G1 :
0
G1 u = u , uT u = ,
X
X
q2 .
q u (t), q T G1 q =
q(t) =
The orthonormal eigenfunctions form a unitary matrix u (t), and

d[q]
Z[0]
dq ,
dq e
i 12
q2

"
#1/2
= [det G]1/2 .
(3.86)
The various factors of proportionality can be replaced by explicit factors in a given

regularization. We shall not need Z[0] in our introductory course. However, note
that in the imaginary time formulation it contains all the physics of the quantum
canonical partition function (cf. Problem 2).
3.7. SCALAR FIELD AND ISING MODEL
73
A few remarks are here in order.

The above derivation was elegant but formal. The results must and have
been checked in various ways. For instance, one can evaluate the path integral
carefully in discretized form and study the continuum limit in various quantities.
It is instructive to evaluate the path integral for generic boundary conditions at
finite time. Then the results (also G) do depend on the boundary values (which
can be used to calculate 0 [q]), and one can show that this dependence disappears
in the limit of infinite extent in time, as derived in the previous section. See e.g.
Brown sect. 1.5.
From (3.67) and (3.79) we now have an explicit formula for the generating
functional of ground state expectation values of time ordered products of the
coordinates:
R
R
1
0
0
0
(3.87)
h0 |T ei dt f (t)q (t) |0 i = ei 2 dt dt f (t)G(t,t )f (t ) .
Differentiating twice with respect to f and setting f = 0 afterwards gives

h0 |T q(t)
q (t0 )|0 i = iG(t, t0 ).
(3.88)
There are some instructive checks. First, apply the differential operator G1 to
the left hand side using the identity
(t t0 ) = (t t0 ),
t
(3.89)
the canonical commutation relations and the Heisenberg equations of motion: the
result is i(t t0 ), as it should according to (3.80) (cf. Problem 3). This shows
that the left hand side of (3.88) is a Green function, but it does not check the
boundary conditions. Second, a complete check can be made by evaluating the
left hand side of (3.88) explicitly using creation and annihilation operators:
q(t) = eiHt q eiHt =

[
a, a
] = 1, [
a, a
] = 0,

1
a
eit + a eit ,
2m
a
|0 i = 0, etc.
(3.90)
The result is (3.88) with (3.85).
3.7
Scalar field and Ising model
We now generalize to the case of infinitely many coordinates (x) labeled by

the continuous index x, i.e. the scalar field. We write |i for the basis vectors in
the coordinate representation (2.7). Then
(x)|i
= (x)|i
(3.91)
74
and formally
|i =
h|0 i =
Z
Y
x
Y
x
|(x)i,
[(x) 0 (x)],
d(x) |ih| = 1.
(3.92)
The path integral representation for the evolution kernel has the form
(t0 , t00 )|00 i =
h0 , t0 |00 , t00 i = h0 |U
d[] eiS[],
(3.93)
with formally
d[] =
d(x).
(3.94)
These formal equations become concrete upon discretization. Since we not

only have continuous time but also space this naturally leads to a lattice in spacetime. We illustrate this here briefly for the case of imaginary time, x0 ix4 .
Then time is equivalent to space, spacetime is replaced by four dimensional space
and it is natural to use a simple hypercubic lattice for discretization, with lattice
spacing b equal in all directions. The finite volume system at inverse temperature
L4 and spatial volume L3 is then approximated by a system on an L3 L4 lattice.
Furthermore it becomes natural to use lattice units, i.e. units in which the lattice
spacing is b = 1. With these conventions the partition function
Z =
SIdiscr =
discr []
d[] eSI
4
XX
x
X
1
V (x ),
(x+ x )2 +
x
=1 2
is well defined.
Models defined in this way have been studied by various analytical and numerical (Monte Carlo) methods. This has led to a fairly complete (and nonperturbative) understanding of four dimensional scalar field theory. For more information
on this Lattice Field Theory approach see Creutz, M
unster and Montvay, Rothe,
Smit, Le Bellac and Barton, and the proceedings of the yearly Lattice XX meetings.
There is a close connection with the spin models studied in statistical physics,
in particular the Ising model. Consider the 4 model
1
1
V = 0 2 + 0 4 .
2
4
Using the change of variables, generalizing to d euclidean dimensions,
1 2
2d, 0 = 2 ,
= 2 , 0 =
(3.95)
(3.96)
75
3.8. FREE SCALAR FIELD
the partition function can be written in the form (dropping an overal constant
factor)
Z =
Z Y
d(x ) exp(2
x x+ ),
(3.97)
d() = d exp[2 (2 1)2 ].
(3.98)
We see that parametrizes the coupling between neighboring spins . The

spins have a single site probability distribution given by (3.98). For the
spins are restricted to values 1:
R
1
d()f ()
R
[f (1) + f (1)],
2
d()
(3.99)
and the model goes over into the Ising model in d dimensions. It turns out that
the Ising model is indeed a very good formulation of the relativistic 4 theory in
four dimensions! (For more information see the books by Le Bellac and Barton,
M
unster and Montvay, Smit, and more generally on the field theoretic description
of critical phenomena: Zinn-Justin, Parisi, Drouffe and Itzykson.)
3.8
Free scalar field
We continue with real time and set 0 = 00 = 0, so that partial integrations

do not involve boundary terms at t0 , t00 . For space the same effect is obtained
by choosing periodic boundary conditions. To obtain ground state expectation
values we take the limit t0 (1 i) and t00 (1 i), as explained in
sect. 3.5. Introducing the external source J(x), the analogue of the external force
f (t), we can immediately write down the analogue of (3.66) and (3.67),
h0, |0, i[J] Z[J] =
h0 |T e
d4 x J(x)(x)
Z[J]
|0 i =
=
Z[0]
Z
R
d[] eiS[]+i
d4 x J(x)(x)
(3.100)
d[] eiS[]+i d x J(x)(x)

R
.
d[] eiS[]
(3.101)
For the free field these formulas can be worked out elegantly, assuming for simplicity infinite space. The action is
1
1
d4 x ( + m2 2 ) =
d4 x ( 2 + m2 )
2Z
2
1
=
d4 x d4 x0 (x)G1 (x, x0 )(x0 ),
2
1
(3.102)
T G1 ,
2
in matrix notation, with
S =
G1 (x, x0 ) = ( 2 + m2 ) 4 (x x0 ).
(3.103)
76
Then (3.79) generalizes to

1
Z[J] = Z[0] exp i
2
1
The inverse of G is given by

G(x, x ) =
with
G(p) =
p2
and
0
iG(x, x ) =
=
=
d x d x J(x)G(x, x )J(x ) .
d4 p ip(xx0 )
e
G(p)
(2)4
4 0
(3.104)
1
1
,
2
2
2
+m
p0 + p + m2 i
0
d3 p eip(xx )iE(p)|x
(2)3
2E(p)
0
dp eip(xx ) ,
0 x00 |
E(p) =
(3.105)
m2 + p 2 ,
x0 > x00 ,
dp eip(xx ) ,
x0 < x00 .
This is analogous to (3.82 3.85) for the harmonic oscillator, with HO E(p)
and mHO 1. In the last line we replaced the dummy p by p. The Green
function G(x, x0 ) is called the Feynman propagator.
Differentiating twice with respect to the external source and setting it to zero
afterwards gives the basic vacuum expectation value of the time ordered product
of two fields, the two-point function,
h0 |T (x
1 )(x
2 )|0 i = iG(x1 , x2 ).
(3.106)
For the generalization to more fields in the time ordered product we just differentiate more times. It is convenient to introduce a suggestive notation for such
n-point functions:
d[] eiS[] (x1 ) (xn )
R
d[] eiS[]
h(x1 ) (xn )i
h0 |T (x
1 ) (x
n )|0 i =
(3.107)
Note that the brackets in this notation do not mean an ordinary statistical average because exp(iS) is complex, and neither an ordinary quantum mechanical
expectation value because of the time ordering involved. Using repeated differentiation with respect to the external source, setting it to zero afterwars, one finds
(Prob. 3d)
h(x1 )i
h(x1 )(x2 )i
h(x1 )(x2 )(x3 )i
h(x1 )(x2 )(x3 )(x4 )i
=
=
=
=
0,
iG(x1 , x2 ),
0,
(i)2 [G(x1 , x2 )G(x3 , x4 ) + G(x1 , x3 )G(x2 , x4 )
+ G(x1 , x4 )G(x2 , x3 )],
(3.108)
77
3.9. SUMMARY
etc. The general formula goes under the name Wicks theorem,
h0 |T (x1 ) (xn )|0 i h(x1 ) (xn )i

= h(x1 )(x2 )i h(xn1 )(xn )i + permutations,
where every distinct permutation (partition into pairs (xi , xj )) occurs once. In
practise we often abbreviate the notation further for clarity, e.g.
h1 2 3 4 i = (i)2 [G12 G34 + G13 G24 + G14 G23 ].
3.9
(3.109)
Summary
In the path integral formalism the action takes a prominent role by specifying
the weight factor exp(iS). In imaginary time this has the form of a Boltzmann
factor exp(SI ), and for a field theory in d spatial dimensions, the imaginary
time path integral (with periodic boundary conditions in time) has the form of a
classical partition function in d + 1 dimensions, while it represents the quantum
partition function in d dimensions.
For free fields the path integral can be explicitly evaluated, because it is
just a multiple gaussian integral. The external source technique is useful for
dealing with correlation functions, which are vacuum expectation values of time
ordered products of field operators. The two-point correlation function, also
called Feynman propagator, is a Green function, similar to the retarded and
advanced Green functions. In contrast to the latter two, the Feynman propagator
G(x, x0 ) is complex, and it does not vanish outside the light cone (i.e. for (x
x0 )2 > 0).
The path integral has great intuitive appeal and properly defined its formalism is equivalent to the canonical operator formulation (provided the latter is also
properly defined!). This can be done with the lattice discretization, which is very
useful for non-perturbative computations. However, a common practise in perturbation theory is to go ahead formally and ignore possible infinities temporarily,
expecting to deal with them later on the fly.
3.10
Problems
1. Relation T H
Consider the Cambell-Baker-Haussdorff formula
1
eA eB = eA+B+ 2 [A,B]+,
(3.110)
where the consist of multiple commutators of higher order in A and B

([A, [A, B]], [B, [A, B]], etc.). Let
1
q ),
A = ia V (
2
p2
B = ia
.
2m
(3.111)
78

Verify that = O(a3 ), and
p
2
2
= p + V (
H
q ).
2m
eia 2 V (q) eia 2m eia 2 V (q) = eiaH+O(a ) ,
(3.112)
2. Partition function of the harmonic oscillator

Consider the harmonic oscillator with unit mass given by the hamiltonian
1
1
H = p2 + 2 q 2 .
2
2
(3.113)
We calculate the partition function
Z = Tr e H
(3.114)
in several ways.
a. Using the energy representation to evaluate the trace and the well known
energy spectrum verify that
1
ln Z = ln(1 e ).
2
(3.115)
We can also calculate Z from the path integral by continuation to imaginary

time, t i ,
Z
dq+
dq+ hq+ |eH |q+ i
pbc
dq+ hq+ |eiHt+ |q+ i

Z
[dq] exp i
[dq] exp
t+
0
(3.116)
1
dt
2
1
d
2
dq
d
dq
dt
!2
1
(3.117)
2q2
(3.118)
!2
1 2 2
q
,
(3.119)
where = + . The path integral is now over functions q( ) with q(0) =

q() = q+ and we also integrate over q+ ; this is indicated by the subscript
pbc, which means periodic boundary conditions. Introducing
d2
G (, ) = 2 + 2 ( 0 ),
d
1
"
(3.120)
be can write the partition function as

Z = [det G1 ]1/2 ,
(3.121)
79
3.10. PROBLEMS
up to a constant.
To evaluate this determinant we introduce a suitable complete set of periodic functions in the interval [0, ],
1
gn ( ) = exp(in ),
n =
2n
, n = 0, 1, 2, .
(3.122)
b. Obtain the matrix elements

G1
mn =
d d 0 gm
( )G1 (, 0 )gn ( 0 ).
(3.123)
c. The determinant det G1 as given by the product over its eigenvalues

diverges badly. We shall regulate the determinant by deviding it by a
similar determinant with a different frequency 0 . Using the infinite product
formula
!
Y
sinh x
x2
,
(3.124)
1+ 2 2 =
n
x
n=1
verify that
sinh( 0 /2)
Z()
=
.
Z( 0 )
sinh(/2)
(3.125)
If we now let 0 approach infinity, then in the trace representation (3.114)

for Z( 0 ) the ground state contribution dominates: Z( 0 ) exp( 0 /2).
Hence, verify that this leads again to the partition function (3.115) obtained
earlier.
Using the discrete time slicing definition of the path integral the divergence
of det G1 is avoided and it leads of course to the right answer, but we
shall not go into details. It is instructive anyway to play with divergent
continuum expressions.
3. Feynman propagator
Eq. (3.85) is the Green function introduced by Feynman (usually called the
propagator) for the case of the harmonic oscillator. We summarize:
Z[f ]
Z
[dq] exp i
1
= Z[0] exp i
2

1
1
dt ( q2 2 q 2 + f q) , (3.126)
2
2
dt dt f (t)G(t t )f (t ) ,
Z[f ]
= h0|T exp i dt f (t)
q (t) |0i,
Z[0]
0
i(tt0 )
e+i(tt )
0
0 e
0
iG(t t ) = (t t )
+ (t t)
.
2
2
Z
(3.127)
(3.128)
(3.129)
80

We have set the mass m = 1. Furthermore t (, ) with the understanding that we supply convergence for large times in a suitable way when
necessary.
a. Using d(t)/dt = (t) verify that
d2
+ 2 G(t t0 ) = (t t0 ).
dt2
!
(3.130)
Hint: after differentiating once with respect to t, simplify by using the

identity F (t)(t) = 0, which is valid for functions F (t) which vanish at
t = 0.
b. Calculate
d2
+ 2 [(t t0 )h0|
q (t)
q (t0 )|0i + (t0 t)h0|
q (t0 )
q (t)|0i]
dt2
!
(3.131)
by using the equation of motion for q(t) and the canonical commutation
relations [q (t), q(t)] = i. Compare with
h0|T q(t)
q (t0 )|0i = iG(t t0 ).
(3.132)
c. By supplying convergence factors exp(t) for large times t ,

with infinitesimal > 0, verify that the Fourier transform of G is given by
Z
dt eikt G(t) =
1
.
k 2 i
(3.133)
The idea is to let +0 in a later stage of calculations involving G, so

and e.g. 5 have the same meaning.
d. Express
h0|T q(t1 )
q (t2 )
q (t3 )
q (t4 )|0i
(3.134)
in terms of products of Feynman propagators by functionally differentiating

Z[f ] and setting f = 0 afterwards.
Chapter 4
Perturbation theory, diagrams
and renormalization
In the previous chapter we have seen, in the example of the free scalar field,
how the path integral can be used to calculate expectation values of time-ordered
products of field operators in the ground (vacuum) state:
h0 |T (x
1 ) (x
n )|0 i h(x1 ) (xn )i.
From these so-called n-point functions we can obtain many quantities of physical
interest. Important for us are scattering and decay amplitudes, and a simple
recipe for obtaining these will be given in section 4.6, deferring its derivation to
chapter 9.
We are now ready for the perturbative calculation of h(x1 ) (xn )i in interacting theories. Using the 4 theory as the simplest example, we introduce
Feynman diagrams which are an invaluable tool for representing the occurring
mathematical expressions and vertex functions, and evaluate the first few of
these in order to clarify renormalization and counterterms.
4.1
Preparation
We recall that the generating functional for the n-point functions has the pathintegral representation (3.101), i.e.
h0 |T e
d4 x J(x)(x)
Z[J]
=
|0 i =
Z[0]
d[] eiS[]+i d x J(x)(x)

R
d[] eiS[]
(4.1)
The action is written as the sum of a free part S0 and an interacting part S1 ,
S = S0 + S1 ,
(4.2)
where S0 is quadratic in the fields and S1 is of higher order. Linear terms can be
absorbed in the external source J, if needed, so we will not include such terms in
81
82CHAPTER 4. PERTURBATION THEORY, DIAGRAMS AND RENORMALIZATION

S. Note that a cosmological constant drops out in the ratio Z[J]/Z[0], and we
will often ignore such a term. In 4 theory
Z
1
S0 = d4 x (x)( 2 + m2 )(x),
(4.3)
2
Z
d4 x (x)4 + S.
(4.4)
S1 =
4
Here m and are renormalized parameters and S contains so-called counterterms: it is the difference between the terms shown explicitly and the original
action formulated in terms of bare parameters 0 , 0 and field 0 :

Z
1
1
1
2
2
4
4
(4.5)
S = d x (0 ) + 0 0 + 0 0 + 0
2
2
4

Z
1
1
1
= d4 x Z()2 + Z0 2 + Z 2 0 4 + 0 , 0 = Z , (4.6)
2
2
4

Z
1
1
1
(4.7)
= d4 x ()2 + m2 2 + 4 + S,
2
2
4

Z
1
1
4
S =
d x (1 Z)()2 + (m2 Z0 )2
2
2

1
2
4
+ ( Z 0 ) 0
(4.8)
4

Z
1
1
1
d4 x Z()2 + m2 2 + 4 + .
(4.9)
2
2
4
The so-called renormalized field isrelated to the original variable 0 , by the
multiplicative renormalization 0 = Z . In fact, we will calculate in renormalized perturbation theory the n-point functions of , not of 0 . The counterterms
will be specified later on, after we have calculated we few vertex functions. As
usual in minkowkian field theory, we have set renormalized vacuum energy to
zero, = 0.
The renormalized perturbation expansion is an expansion in powers of ,
not 0 . The basic idea of counterterms is to choose the starting point of the
perturbation expansion close to the final answer. For example, we may choose m
to be the exact mass of the particles and the exact coupling constant (possible
definitions of will be given later). So the effect of the interactions on m and
is exactly cancelled by the counterterms. We may then hope that more general
interaction effects will be small when is small. In particular, infinities, which
which pop up because of our cavalier dealing with infinitely many degrees of
freedom, are to be cancelled by the counterterms. It is hard to explain this
properly at this point so we shall ignore the counterterms for the moment (S
0) and return to them later.
To further prepare for the perturbative expansion we write, in condensed
notation,
h1 n i =
d[] eiS0 +iS1 1 n

h1 n eiS1 i0
R
=
,
d[] eiS0 +iS1
heiS1 i0
(4.10)
4.2. DIAGRAMS FOR 4 THEORY.
83
where the subscript 0 denotes the free field average

hF []i0 =
d[] eiS0 F []
R
d[] eiS0
(we have devided numerator and denominator by
4.2
(4.11)
d[q] exp(iS0 )).
Diagrams for 4 theory.
Ignoring the counterterms, expanding in means expanding the exponential

exp(iS1 ). Consider first the denominator heiS1 i0 in (4.10):
d4 x h(x)4 i0 + O(2 )
4
Z
= 1 i 3 d4 x (i)2 G(x, x)2 + O(2 ),

4
Z
1
= 1 + (i6) d4 x (i)2 G(x, x)2 + O(2 ),
8
heiS1 i0 = 1 i
(4.12)
(4.13)
where we used (3.108). The three terms in (3.108) give identical contributions.
Consider next the numerator of the two point function (n = 2):
h1 2 eiS1 i0 = h1 2 i0 i
d4 x h1 2 (x)4 i0 +
= iG(x1 , x2 ) i 3 d4 x (i)3 G(x1 , x2 ) G(x, x)2 (4.14)

4
Z
4
i 12 d x (i)3 G(x1 , x)G(x2 , x)G(x, x) + (4.15)
4
Z
A pairing of fields is called a contraction. The factor 3 in (4.14) has the same
origin as in (4.12), namely three possible contractions of four fields giving identical
contributions, and indeed, we recognize the denominator as a factor in (4.14),

iG(x1 , x2 ) 1 +
1
(i6)
8
d4 x (i)2 G(x, x)2 + O(2 ) .
(4.16)
The factor 12 in (4.15) comes from first contracting, say, 1 with one of the (x),
which gives 4 identical contributions, and then 2 with one of the remaining three
(x), which gives 3 identical contributions; then there remain two (x)s which
simply give G(x, x).
The numerator of the four point function,
h1 2 3 4 e
iS1
Z 4
i0 = h1 2 3 4 i0 i
d x h1 2 3 4 (x)4 i0 + O(2 ), (4.17)
4
= - i G(x,y)
x
y
Figure 4.1: The propagator.
gives rise to more contributions. The first term is just (3.108). Applying Wicks
theorem to the second, term there are contractions of only the 1 , . . . , 4 among
themselves. This gives
h1 2 3 4 i0
1
1 + (i6)
8
d x G(x, x) + O( ) ,
(4.18)
which has again the zero point contribution (4.13) found in the evaluation of the
denominator. Then there are contributions in which only two of the 1 , . . . , 4
contract with (x)4 ,
!
(i)G(x1 , x2 ) i
12
4
d4 x (i)3 G(x3 , x)G(x, x)G(x, x4 ) + 5 permutations.
(4.19)
The term shown, with {x1 , x2 , x3 , x4 } {x3 , x4 , x1 , x2 }, contains (4.15) as a
factor. There is one new contribution in which all of the 1 , . . . , 4 contract
with (x)4 :
h1 2 3 4 iS1 iconn
0
= i 24
4
d4 x (i)4 G(x1 , x)G(x2 , x)G(x3 , x)G(x4 , x),
(4.20)
where the meaning of the label conn (=connected) will get clear in the following.
At this point the expressions call out loudly for a representation in terms of
diagrams. The propagator G(x, y) times the factor (i) is represented by a line
joining points labeled x and y. See Fig. 4.1. The little circles at the end of the line
indicate that the line represents a propagator and not a two-point vertex function
(to be defined below). Where arguments coincide, such as in G(x, x), lines come
together in the same point we have a vertex. This is most clear in the expression
(4.20), see Fig. 4.2. RThe vertex represents the factor 24i/4 = i6, as well as
an integration over d4 x . The vertex is represented in Fig. 4.3, where the four
lines indicate that it is a four point vertex, i.e. four propagators can be attached.
Note that the lines of the vertex do not have the little circles at the end, whereas
the circles in Fig. 4.2 indicate the presence of the propagators.1 The diagrams
for the denominator, the numerator of the two-point function and the numerator
of the four-point functions are shown in Figs. 4.4, 4.5, 4.6, respectively.
1
This convention is taken over from the books by Bjorken and Drell.
85
Figure 4.2: Lowest order diagram for the connected four point function
h1 4 iconn .
w
= i S(w,x,y,z)
Figure 4.3: The four point vertex.
1 +
+...
Figure 4.4: Expansion of heiS1 i0 to order .
+ . . .
Figure 4.5: Expansion of h1 2 eiS1 i0 to order .

1
+
3
4
1
+ two perm. +
4
+...
3
Figure 4.6: Expansion of h1 4 eiS1 i0 to order .

More precisely, we associate with the four-point vertex i a vertex function
S(x1 , x2 , x3 , x4 ), defined by the 4-th functional derivative of the action:
Z
4S
= 6 d4 x 4 (x1 x) 4 (x4 x)(4.21)
S(x1 , . . . , x4 )
(x1 ) (x4 )
= 6 4 (x1 x2 ) 4 (x1 x3 ) 4 (x1 x4 ).
The first form maintains the symmetry under permutations of the xs. There is
also a two-point vertex function (cf. (1.148)),
2S
S(x, y) =
(x)(y)
"
=0
= ( 2 m2 ) 4 (x y).
(4.22)
A propagator-line emerging from a vertex now simply means integrating over the
common argument of the vertex function and the propagator. For instance, the
contribution (4.15) to the two-point function can be written as
1
2
d4 y1 d4 y4 (i)G(x1 , y1 ) iS(y1 , y2 , y3 , y4 )(i)G(y2 , x2 )(i)G(y3 , y4 ).
(4.23)
The diagrams greatly clarify the formulas. There is a one-to-one correspondence between diagrams and the mathematical expressions, provided we keep
in mind the accompanying combinatorical factors such as 1/8 in the zero-point
function, and 1/2 in the two-point function. Such factors are related to the symmetry of a diagram and there are rules for determining them (see e.g. Peskin and
Schroeder sect. 4.4), but it is often easy enough to get them by explicit counting
of identical contributions, as we have done above.
One furthermore observes:
- A diagram may be disconnected, i.e. consist of several connected parts that
are not connected by any line. The connected parts represent factors in the
mathematical expression. We usually concentrate on the connected parts.
87
+ . . .
Figure 4.7: Expansion of h1 2 i to order .

- The diagrams for the denominator in (4.10) (often called vacuum diagrams),
which do not have any external lines, cancel against similar disconnected
parts in the numerator. This is true to all orders. Hence, for the evaluation of n-point functions with n > 0, we can forget about disconnected
contributions containing diagrams without external lines. For example, the
diagrams for the two-point function are given in Fig. 4.7.
- The diagrams can be decomposed into one particle irreducible (1PI) components and one particle reducible (1PR) components. A 1PR diagram
becomes disconnected when cutting a single line, whereas a 1PI diagram
remains connected. The 1PI components are called dressed vertex functions, and denoted by (x1 , x2 , ). To lowest order they are equal to the
bare vertex functions. There is a special notation for the two-point vertex
function:
(x1 , x2 ) = S(x1 , x2 ) (x1 , x2 ).
(4.24)
and is called the selfenergy.
The perturbative expansion of the 2- and 4-point dressed vertex functions

is given in Fig. 4.8 and 4.9.
The typical 1PR component is the exact two-point function, which is called
the full, or fully dressed, propagator. Its expansion in terms of 1PI parts is
given in Fig. 4.10. Denoting the full propagator by G0 , Fig. 4.10 represents
the expansion
iG0 (x1 , x2 ) = iG(x1 , x2 ) +
(i)G(y2 , x2 ) +
d4 y1 d4 y2 (i)G(x1 , y1 )(i)(y1 , y2 )
Z
d4 y1 d4 y4 (i)G(x1 , y1 )
(i)(y1 , y2 )(i)G(y2 , y3 )(i)(y3 , y4 )(i)G(y4 , x2 )

+ ,
(4.25)
or in matrix notation
G
"
= G GG + GGG + = G 1 +
= G(1 + G)1 = (G1 + )1 .
n=1
(G)
(4.26)
In the last line we used the identity A1 B 1 = (BA)1 . We assumed that

the geometric series converged, which may be true for (some, not all) matrix
2

=
+ ...
Figure 4.8: Expansion of the selfenergy i(x1 , x2 ) to O(2 ).

1

3
=
4
+ 2 perm. + . . .
3
Figure 4.9: Expansion of the four point vertex function i(x1 , , x4 ) to lowest
non-trivial order.
elements, e.g. obtained by Fourier transformation, as will be done in the
next section.
- The four-point function, dressed to all orders in , can now be summarized
in terms of the full propagators and 4-point vertex function as in Fig. 4.11.
- More generally, an arbitrary diagram may be expressed in terms of full
propagators and vertex functions. Expanding the full propagators and vertex functions in powers of , and then re-expanding the total, gives the
correct expansion of the diagram.
- The perturbative expansion of vertex functions is a loop expansion: increasing the order in corresponds to increasing the number of loops in the
diagrams. This is true to all orders.
Keeping track of h
it can be shown that with each additional loop the
accompanying power of h
increases. For this reason the loop expansion
is also called a semiclassical expansion. The lowest order vertex functions
are classical in the sense that the bare vertex functions are equal to the
functional derivatives of the classical action.

+
+

+ . . .
Figure 4.10: Structure of the two-point function h1 2 i in terms of 1PI two-point

vertex functions.
89
4.3. DIAGRAMS IN MOMENTUM SPACE

+ 2 perm. +

Figure 4.11: Structure of the four point function h1 4 i.
4.3
Diagrams in momentum space
Because of translation invariance the n-point functions h(x1 ) (xn )i depend

only on differences of coordinates. We can then simplify expressions by taking
their Fourier transform, or in the jargon: going into momentum space. The
Fourier transform is defined as in the following example of the full n-point vertex
function,2
Z
d4 x1 d4 xn ei(p1 x1 ++pn xn ) (x1 , , xn ) = (2)4 (p1 + +pn ) (p1 , , pn ).
(4.27)
We have anticipated the presence of a momentum-conserving delta function.3 For
example, for the bare vertex function we have, using (4.21),
Z
d4 x1 d4 x4 ei(p1 x1 ++p4 x4 ) S(x1 , , x4 ) = 6
d4 x ei(p1 +p2 +p3 +p4 )x
S(p1 , p2 , p3 , p4 ) = 6.
(4.28)
So the bare vertex in momentum space is just the constant 6. There is a special
notation for two point functions since these depend only on one independent
momentum. For the bare two-point vertex function we have
S(p, p) G(p)1 = (m2 + p2 i),
(4.29)
where we have added the infinitesimal regulator term i (cf. (3.34)), while the
propagator is given by
G(p, p) G(p) =
m2
1
.
+ p2 i
(4.30)
For the full two-point vertex function we have according to (4.24),

(p, p) = [m2 + p2 + (p)]
(4.31)
(including the i in (p)). See Figs. 4.12 4.14 for the corresponding diagrams.
2
For simplicity the same symbol (here ) is used for the Fourier transform.
In line with ubiquitous jargon, we call the Fourier variables: momenta. The reason is that
in applications to scattering these variables become the four-momenta of (real or virtual)
particles.
3

An easy way to obtain the mathematical expression for the diagrams in momentum space is to represent everywhere the propagator by its Fourier transform
G(x, y) =
d4 p ip(xy)
e
G(p),
(2)4
(4.32)
carry out the integrations over the coordinates of the vertex functions, and take
the Fourier transform with respect to the external line variables. This leaves the
momentum integrations associated with the propagators as in (4.32), with a fourmomentum conservation delta function at every vertex. The result of integrating
out as much as possible the propagator momenta using the delta functions, is
one (2)4 4 ( ) expressing overall conservation of the momenta at the external
lines (which is to be extracted according to (4.27)), plus a number of remaining
integrations associated with closed loops in the diagrams.
So the rules for obtaining the formula from a diagram in momentum space
are:
- lines carry momenta p and a propagator iG(p);
- there is momentum conservation at each vertex, with a factor
iS(p1 , , p4 ) = i6;
- integrate over the free momenta.
The selfenergy diagram in Fig. 4.14 is equal to
i(p) =
Z
d4 q
1
i
(i6)
.
4
2
2
(2) m + q 2 i
(4.33)
The full propagator structure in Fig. 4.10 is just a geometric series in momentum
space:
iG0 (p) iG(p) + (i)G(p)(i)(p)(i)G(p)
+ (i)G(p)(i)(p)(i)G(p)(i)(p)(i)G(p) +
1
i
= iG(p)
= 2
.
(4.34)
1 + (p)G(p)
m + p2 + (p)
Comparing with (4.31) we see that the full propagator G0 (p) is the inverse of the
two-point vertex function, as for free fields.
Fig. 4.15 shows the four point function to order 2 . The corresponding formula
is
i(p1 , p2 , p3 , p4 ) = i6
Z
1
d4 q
i
i
+ (i6)2
4
2
2
2
2
(2) m + q i m + (q + p1 + p2 )2 i
+ 2 permutations
(4.35)
91
4.4. ONE-LOOP VERTEX FUNCTIONS
= - i G(p)
Figure 4.12: Diagram representing iG(p).
= -i6
p
3
p4
Figure 4.13: Vertex in momentum space.

Finally Fig. 4.16 shows a general n > 4 point vertex function in the one-loop
approximation. Note that these are zero classically: there are no bare vertex
functions with n > 4 in the 4 model (they would be represented by a tree-like
vertex with more than four external lines).
4.4
One-loop vertex functions
We shall now evaluate the one loop diagrams for the two- and four-point vertex
functions. The integrals in the expression for the two-point function (selfenergy)
(4.33), and four-point function (4.35), diverge, so we first need to regularize
them. This can be done in a variety of ways, leading to a variety of answers.
This arbitrariness is to be absorbed in the bare parameters, similar to the case
of the cosmological constant in sect. 2.3, such that the final physical answer is
unambiguous. To be able to discuss this we first need to evaluate the diagrams.
Here we shall use a simple method, which consists of making a Wick rotation,
such that the integrals take a euclidean four-dimensional form, and then imposing
Figure 4.14: Order contribution to the selfenergy i(p).
p1
p3
+ 2 perm.
+
p
p4
q+p+p
1
2
Figure 4.15: The 4-point vertex function i(p1 , p2 , p3 , p4 ) to order 2 .
1
q
2
q1
3
4
n-1
.
q2 5
. .
Figure 4.16: Contribution to the n-point vertex function (p1 , , pn ) in one loop
order, q1 = q p1 p2 , q2 = q p1 p2 p3 p4 , etc.
a spherical cutoff q 2 < 2 .
Consider the integral in the self energy (4.33),
I = i
dq 0
1
= i
2
2
m + q (q 0 )2 i
dq0
m2
q2
1
,
q02 i
(4.36)
where we changed variables to q0 = q 0 . The position of the poles of the integrand are illustrated in Fig. 4.17. Rotating the integration contour counter-clock
wise over an angle /2 (the Wick rotation) it is easy to verify that the above
integral is unchanged. During the rotation we do not cross any singularity.
(Adding arcs at infinity, which do not contribute, the rotated contour is just a
deformation of the original one along the real axis, see the figure on the right
hand side.) Hence, setting q0 = iq4 , with real q4 , we get
I=
dq4
m2
1
,
+ q2 + q42
q4 = iq0 ,
(4.37)
where we have set to zero because it is not needed anymore as the denominator
does not vanish. It follows that the selfenergy can be written in euclidean form
(i.e. a form where the metric is euclidean and q4 enters equivalently to q1 , . . . q3 ):
= 3
d4 q
1
,
4
2
(2) m + q 2
(4.38)
where now q 2 = q12+ + q42 , d4 q = dq1 dq4 . We regularize this integral by

introduce a cutoff q 2 < . Introducing four dimensional spherical coordinates,
93
q0
Figure 4.17: Left: Integration contours in the complex q0 plane

2 for 2the integral
(4.36). The crosses indicate the position of the poles at ( m + q i). The
original contour along the real axis is rotated to the contour along the imaginary
axis. Right: Integration contour giving a zero result, which can be used to prove
that the two contours on the left give the same answer.
the integral over angles is4 2 2 and we are left with
1
3
2
dq q 3 2
2
(4.39)
4
(2)
m + q2
0
"
!#
"
#
2
3
2
3
2
2
2
2
2
m ln 1 + 2
=
m ln 2 + O( ) .
=
16 2
m
16 2
m
The above evaluation was particularly easy because the one loop (p) is independent of p in 4 theory. The four-point vertex function is more interesting.
The typical integral here is
J(p) = i
1
d4 q
,
4
2
2
(2) (m + q i)(m2 + (q + p)2 i)
(4.40)
in terms of which
1
(p1 , p2 , p3 , p4 ) = 6 + (6)2 [J(p1 + p2 ) + J(p1 + p3 ) + J(p1 + p4 )] , (4.41)
2
to order 2 . First we combine the denominators by using the identity (cf. Appendix)
Z 1
1
1
=
dx
.
(4.42)
AB
[xA + (1 x)B]2
0
Applying this identity for A = m2 + q 2 i, B = m2 + (q + p)2 i gives
J(p) i
4
1
d4 q Z 1
dx
.
(2)4 0
[x(m2 + q 2 i) + (1 x)(m2 + (q + p)2 i)]2
(4.43)
The surface of the unit ball in n dimensions is 2 n/2 /(n/2).
q0
Figure 4.18: Integration contours in the complex q0 plane for J(p) for the case
p2 < 4m2 . The crosses indicate again the position of the poles and the original
contour along the real axis can also in this case be rotated into the contour along
the imaginary axis.
We now complete the square in the denominator by shifting the integrations over
q,
q q (1 x)p,
(4.44)
while ignoring for the moment the cutoff we are imposing on the integrations (we
come back to this later). Then,
J(p) = i
d4 q
(2)4
1
0
dx
[m2
1
.
+ x(1 x)p2 + q 2 i]2
(4.45)
Doing the integration over q 0 first, we see similar pole positions as in the case of
the selfenergy (here these are double poles), provided that m2 + x(1 x)p2 > 0,
which is true for all x if p2 > 4m2 . In case p2 < 4m2 the poles are situated
as in Fig. 4.18 and the Wick rotation can still be done. Performing the Wick
rotation and the q integrations with the euclidean cutoff q 2 < 2 leads to
1
J(p) =
16 2
1
0
2
dx ln 2
1 + O(2 )
2
m + x(1 x)p i
"
(4.46)
The remaining integration over the Feynman parameter x is elementary,

!
2
1
2
4m + p2 i + p2 i
2
J(p) =
+ ln 2 + O( ) , (4.47)
1 ln 2
16 2
m
4m + p2 i p2 i
but this explicit answer does improve the clarity of the analytic structure of J(p).
The integral representation (4.46) is sufficient for our purpose.
We end this section with more comments:
- The maximum power of in the expression for the vertex functions is the
same as their dimension in mass units, i.e. 2 for the two-point function and
95
0 ( ln + const.) for the four-point function. The one-loop six- and

higher-point functions (cf. Fig. 4.16) are finite.
- The regularization procedure used above has some arbitrariness, because
the answer may depend on the moment we introduce the cutoff . It
was introduced after the Wick rotation because then space and time are
completely equivalent and we get Lorentz invariant answers. But even then
can be introduced either before or after shifting integration variables as
in (4.44), whereas we do not shift the integration region accordingly. In
addition there is more than one way to assign momenta to the propagators
in the loop integrals (such that momentum conservation at the vertices is
satisfied), which are related by transformations of the variables of the loop
integrals. It can be shown5 (see also Prob. 1) that in the limit the
arbitrariness is confined to a polynomial in the external momenta, of degree
equal to the mass dimension of the vertex function. For dimension zero (i.e.
logarithmically divergent integrals), shifts of the integration variable (as in
(4.44)) do not to influence the momentum dependence of the regularized
answer as , because each power of external momentum would imply
a factor 1 .
- It is evidently very advantageous to employ a regularization in which one
can freely make shifts and other transformations of variables without having to worry about boundary effects. Important examples are dimensional
regularization, Pauli-Villars regularization and the lattice. Dimensional
regularization is very convenient and described in many books, e.g. De Wit
& Smith, Veltman, Peskin & Schroeder.
- It is instructive to compare results with the lattice, where one finds (see
e.g. Smit)

= 3 C0 a
1
C2 m +
m2 ln(a2 m2 ) + O(a2 ) ,
2
16
2
(4.48)
h
i
1
1
2
2
2
dx
ln{a
[m
+
x(1
x)p
i]}
+
1
16 2 0
+ O(a2 ),
(4.49)
= 0.154933..., C2 = 0.0303457...,
(4.50)
J(p) = C2
C0
where a is the lattice spacing. Note the correspondence a1 . Apart

from a, all the lattice details reside in the constants C0 and C2 (neglecting
terms vanishing in the continuum limit).
5
The possible differences in the regularized expressions come from the integration region
near the cutoff. In this region we can expand the propagator denominators in powers of the
external momenta, with accompanying inverse powers of .
p
2
2
= i m + i Z p
= i 6
Figure 4.19: Counterterms in momentum space.
The arbitrariness in regularization method reflects itself in polynomials in
p, m and or a1 , of degree equal to the dimension of the diagram, except
for degree-zero terms where we have in addition logarithms (e.g. ln or
ln a1 ) accompanying 0 or a0 .
4.5
Renormalization and counterterms
In this section we will incorporate the effect of the counterterms and make a
convenient choice for the constants Z, m2 and in (4.8,4.9). The counterterms
lead to additional vertices indicated by a cross in Fig. 4.19. To lowest order they
contribute only to the two and four point vertex functions. For the selfenergy
this means adding m2 Z p2 to (p),
3
2
2
2
(p) =
m
ln
16 2
m2
m2 Z p2 + O(2 ),
(4.51)
or for the complete two point function:

2
3
2
2
(p, p) = m m + (1 Z)p +
m
ln
16 2
m2
2
+ O(2 ). (4.52)
The counterterms can now be chosen such that (p, p) is finite as . This
determines the dependence of m2 and we see that Z is not needed to cancel
infinities at this stage, since there is no divergent p2 term (it is, however, needed
at two-loop order). The possibility of finite parts in the counterterms induces a
polynomial arbitrariness in (p, p), of the form + p2 . We now choose the
counterterms such that (p, p) satisfies so-called renormalization conditions.
We need two conditions to fix the polynomial. A convenient choice is determined
by an expansion about p2 = m2 :
(p, p) = (m2 + p2 ) + O((m2 + p2 )2 ),
(4.53)
4.5. RENORMALIZATION AND COUNTERTERMS
97
in which the constant term is chosen zero and the coefficient of m2 + p2 is chosen
one. It follows that
!
6
2
2
2
2
m =
m ln 2 + O(2 ),
(4.54)
2
32
m
Z = O(2 ).
(4.55)
Similarly, the four point vertex function gets an additional contribution 6,
1
(p1 , p2 , p3 , p4 ) = 6( ) + (6)2 [J(p1 + p2 ) + J(p1 + p3 ) + J(p1 + p4 )]
2
+ O(3 ),
(4.56)
with the function J(p) given by (4.46). Requiring this to be finite determines
the dependence of and leaves an ambiguity of the form of a constant, which
is eliminated by imposing a renormalization condition on the four-point vertex
function. A convenient condition is
(0, 0, 0, 0) = 6,
for which
3 (6)2
2
6 =
ln
1 + O(3 ).
2 16 2
m2
We now have the following results for the vertex functions:
!
(4.57)
(4.58)
(p, p) = m2 p2 + O(2 ),
(4.59)
2 Z 1
2
2
(6)
m + x(1 x)(p1 + p2 ) i
(p1 , p2 , p3 , p4 ) = 6
dx ln
2
32 0
m2
+ 2 permutations.
(4.60)
Keeping in mind that the n > 4 point vertex functions are finite, the approximation at one-loop order is now determined and unambigous. All regularization dependence has disappeared (it is easy to check for example that the lattice
method leads to the same final result under the same renormalization conditions).
We end this section with some more remarks:
- As can be seen from (4.60), for < 1 the perturbative O(2 ) correction to
the leading order four-point vertex function (i.e. 6) is small for momenta
of order m.
- We can now understand better the meaning of the renormalized parameter
: it is simply the value of the four point vertex function at zero momentum.
We could make another choice for , say 1 is the value of (p1 , p2 , p3 , p4 ) at
some other standard set of momenta p1 , . . . , p4 . Then 1 = + O(2 ). So
other renormalization conventions imply a re-ordering of the perturbation
series. This freedom in choice of expansion parameter is more fully exploited
in the method of the Renormalization Group.

- With the above choice of renormalization condition for (p, p), the parameter m is the mass of the particles. This follows from the fact that the
pole in the
full propagator G0 (p) = [m2 + p2 + (p)]1 as a function of p0 is
0
at p = ( m2 + p2 i). The relation,
position of pole in G0 (p) particle mass,
(4.61)
will be explained more fully in chapter 9. Other renormalization conditions

are possible in which m is not the particle mass, but just a parameter with
dimension of mass, related to the particle mass by a relation of the form:
mpart = m + O(). To see this, consider the equation
m2 + p2 + (p) = 0.
(4.62)
Let p2 = m2part be the zero of this equation (using p2 as variable instead of

p0 , which is simpler). Expanding the selfenergy about the zero p2 = m2part ,
(p) = 0 + 1 (m2part + p2 ) + O((m2part + p2 )2 ),
(4.63)
m2part = m2 + 0 = m2 + O().
(4.64)
gives
Then the pole behavior is
G0 (p) =
1
m2
p2
+ (p)
Zpole
2
mpart +
p2
Zpole =
1
.
1 + 1
(4.65)
With our previous renormalization conditions (4.53),

m = mpart ,
- The rescaling of the field 0 =
tion.
0 = 1 = 0,
Zpole = 1.
(4.66)
Z is called a wave function renormaliza-
The counterterm Z is chosen to cancel divergencies in (p, p) of the form

p2 , which occur in higher loop order. Above we assumed its finite part to
be fixed such that the residue of the pole in G0 (p) is 1. So in general Z 6= 1.
As we have seen above, other renormalization conditions may lead to a
different residue Zpole . The value of Zpole is not a physical parameter, it just
specifies the scale of . After all, also the scale of 0 , the starting field, does
not matter. It is a coordinate in configuration space and physical results
should not depend on the choice (scale) of coordinates. This will be evident
in chapter 9, where we make the connection between h(x1 ) (xn )i and
the scattering cross-section.
99
4.5. RENORMALIZATION AND COUNTERTERMS
- It has been shown that the infinities can be cancelled by polynomial counterterms to arbitrary order in perturbation theory. For the 4 theory these
have the form used in this chapter, with the interpretation of an adjustment
of the bare parameters of the theory. Such theories are called renormalizable. For non-renormalizable theories the number of different counterterms
that are needed increases with the order of the loop expansion, and the number of free parameters ( renormalization conditions) increases accordingly,
thus limiting the predictive power of the theory. The Standard Model is
a renormalizable theory. Einstein quantum gravity is non-renormalizable.
The proofs of these statements are complicated.
- The nonperturbative regularization provided by the lattice gives a different view on renormalization. The Compton wavelength in lattice units,
1/am, is similar to the correlation length in spin models. The continuum
region am 1 corresponds to the critical region with large correlation
lengths. The arbitrariness in the regularization procedure (different formulations giving the same physics) corresponds to the universality of critical
phenomena. Universality is a nonperturbative version of renormalizability.
- Last, but not least:
It is instructive to express the bare parameters in terms of the renormalized
ones (cf. eqs. (4.8,4.9):
0
0
2
6
2
2
= Z (m m ) = m
m
ln
+ O(2 ),(4.67)
2
2
32
m
!
2
2
1 (6)
ln 2 1 + O(3 ).
(4.68)
= Z 2 ( ) = +
2
4 16
m
1
where we used Z = 1 + O(2 ). We see that 0 has the tendency to get

negative, and 0 has the tendenct to grow (more positive), if we let grow.
This is indeed what happens in a nonperturbative analysis on the lattice.
One finds that 0 goes to infinity as the lattice spacing diminishes. Recalling
sect. 3.7, this means that 4 theory is the 4D Ising model in disguise. With
no bare coupling to adjust anymore (0 being infinite) the renormalized
coupling turns out still to depend on the lattice spacing, but very weakly,
while continuum behavior, Lorentz invariance, etc. can be excellent. One
finds that in the strict continuum limit the renormalized coupling vanishes
1/ ln(am). So we should not take this limit.
This at first sight shocking phenomenon is called triviality (the theory
being trivial at zero renormalized coupling). It is not a property of the
regularization but of the model itself. The interpretation is that the model
breaks down at the regularization scale (cutoff). As we know now, this is
also a property of our best phenomenological theory, the Standard Model.

However, in practise there is no problem, since the cutoff can be quite large,
e.g. larger than the Planck scale mPlanck = G1/2 = 1.2 1019 GeV, depending
on the renormalized couplings, some of which are not yet known.
Note that, for /m = mPlanck /1 GeV, the perturbative correction between
0 and is actually smaller than 1 if < 1.
4.6
Recipe for scattering and decay amplitudes
The question, how to obtain scattering and decay amplitudes from n-point functions will be answered later in chapter 9. The discussion there is somewhat
lengthy and a simplified derivation suitable for the tree graph approximation will
be given also later in sect. 6.6. Here we just give some rules which allow us to go
ahead and apply the diagrammatic techniques to calculate the physically measurable cross sections and decay rates:
1. Let Zpole be the residue of the pole in the two-point function,
G0 (p)
Zpole
2
mpart +
p2
(4.69)
where mpart is the particle mass. For simplicity we shall assume to have
arranged things such that mpart = m and Zpole = 1. In the tree graph
approximation this is the standard convention, in higher orders it implies
a suitable choice of counterterms with a suitable rescaling of the fields.
2. Let G0 (p1 , , pn ) be the Fourier transform (as in (4.27)) of the correlation
function G0 (x1 , , xn ) defined by
h1 n iconn (i)n1 G0 (x1 , , xn )
(4.70)
This function has the structure (cf. Fig. 4.11)

G0 (p1 , , pn ) = G0 (p1 ) G0 (pn ) H(p1 , , pn ),
(4.71)
i.e. one can always identify full propagators at the external lines. Removing
these leaves H, which is sometimes called the amputated n-point correlation function. Note that H is a symmetric function of its arguments.
3. The scattering amplitude for the process 1 + 2 3 + n is simply given
by
T (p3 , , pn ; p1 , p2 ) = H(p3 , , pn , p1 , p2 ),
(4.72)
q
where now p0j = m2j + p2j , j = 1, , n (for the 4 model the mj s are all
equal). The overall minus sign is a convention.
101
4.6. RECIPE FOR SCATTERING AND DECAY AMPLITUDES
= - i G (p)
k
p
1
= i S (p ,p ,k) = - i g
1
Figure 4.20: Propagator and vertex of .

4. For the decay amplitude 1 2 + + n we have a similar formula, in
the approximation which treats the decaying particle initially as a stable
particle with mass m1 :
T (p2 , , pn ; p1 ) = H(p2 , , pn , p1 ).
(4.73)
We now give some applications. The scattering amplitude of the 4 model is

given by
T (p3 , p4 ; p1 , p2 ) = (p3 , p4 , p1 , p2 ),
(4.74)
and from (4.60) we see that T = 6 + O(2), in accordance with the lowest order
result (2.89) obtained earlier. Note that the O(2 ) contribution is in general
complex, because (p1 + p2 )2 4m2 , with an imaginary part determined by
the i in the logarithm in (4.60). (N.B. In the broken phase of the 4 model
(corresponding to the magnetized phase of the Ising model) there is also a threepoint vertex and T is not just equal to .)
As a different application, consider the two-field model described by the classical action
!
1
1
1
1
g
S[, ] = d x
()2 + M 2 2 + ()2 + m2 2 + 4 + 2 ,
2
2
2
2
4
2
(4.75)
which was used in sect. 2.6 to illustrate the possibility of particle decay. Here we
assume the particle to be stable, M < 2m. The free part of the action leads to
an additional propagator for the field,
Z
G (p) G (p, p) =
M2
1
,
+ p2 i
(4.76)
To distinguish it from the propagator G (p) we use a dashed line in diagrams.

The interaction S1 leads to an additional bare three point vertex as in Fig. 4.20.
The following will be the subject of Problem 2:
3
k3
k4
k3
k4
+
p1
p1
p2
Figure 4.21: Tree graph contribution to h1 2 3 4 iconn .

3
4
p4
p3
+
1
+ 2 perm.
p1
p2
Figure 4.22: Tree graph contribution to h1 2 3 4 iconn .

- The tree graphs in Fig. 4.21 represent the zero loop contribution to h1 2 3 4 iconn .
- The tree graph contribution to h1 4 iconn is shown in Fig. 4.22. The
correspondig scattering amplitude is given by
T (p3 , p4 ; p1 , p2 ) = 6 g 2
1
1
1
+
+
,
M2 s M2 t M2 u

(4.77)
with s = (p1 + p2 )2 , t = (p1 p3 )2 and u = (p1 p4 )2 in standard

notation. The differential cross section in the center of mass frame is given
by
"
d
d
cm
"
1
1
1
=
6 g 2
+ 2
2
2
2
64 s
M s M + 2|p| (1 + cos )
1
+ 2
M + 2|p|2 (1 cos )
!#2
(4.78)
Note the symmetry under , + corresponding to the

identical particles.
4.7
Summary
Expanding exp(iS1 ) in the path integral we can express the full correlation functions in terms of free field propagators and vertex functions. Diagrams are a great
103
4.8. APPENDIX
help in clarifying these expressions and we can establish a 1-1 correspondence between the diagrams and the mathematical expressions. The correlation functions
can be decomposed in terms of 1PI vertex functions and full propagators (i.e.
two-point correlation functions).
The cavalier treatment of the infinite number of degrees of freedom in field
theory leads to infinities in loop diagrams. These infinities can be cancelled by
counterterms and in a renormalizable theory the counterterms can be absorbed
in the parameters appearing in the action. The parameters in the action are
called bare parameters (in the 4 model these are 0 , Z, m20 and 0 ), while the
physical parameters are defined in terms of the renormalized vertex functions
(dressed by loop diagrams). The arbitrariness in the finite parts of diagrams
and counterterms is fixed by imposing some standard normalization condition on
the renormalized vertex functions.
A simple recipe can be followed to obtain scattering or decay amplitudes from
the correlation functions.
4.8
Appendix
The identity (4.42) can be derived as follows. First use the exponential representation
Z
1
1
dt t1 etA ,
(4.79)
=
A
() 0
to get
Z
1
1
=
dt du t1 u1 etAuB .
(4.80)
A B
()() 0
Next, make the transformation of variables
t = xr,
and use
which gives
u = (1 x)r,
r (0, ),
dr r +1 er[xA+(1x)B] =
x (0, 1),
(4.81)
( + )
,
[xA + (1 x)B]+
(4.82)
1
( + ) 1
x1 (1 x)1
=
dx
.
(4.83)
A B
()() 0
[xA + (1 x)B]+
The derivation is valid for postive A, B and , , and furthermore anywhere else
that can be reached by analytic continuation such that the integral over x remains
convergent. The generalization to
Z
(1 + + n ) 1
1
=
dx1 dxn (1 x1 xn )
A1 1 Ann
(1 ) (n ) 0
x1 1 1 xnn 1
(4.84)
[x1 A1 + + xn An ]1 ++n
is straightforward.
Z
4.9
Problems
1. Shifting integration variables in divergent integrals

Consider the quadratically divergent integral
I(p) =
1
d4 q
,
4
2
(2) m + (q + p)2
(4.85)
in euclidean momentum-space. Shifting the integration variable q q p

one would conclude that I(p) = I(0). This is indeed true for dimensional,
Pauli-Villars, and lattice regularization. However, suppose we regulate the
integral by imposing a spherical cutoff:
I(p)
q 2 <2
d4 q
1
.
4
2
(2) m + (q + p)2
(4.86)
The dependence on p can be studied by makin a Taylor expansion about

p = 0:
h
I(p) = I(0) + p p I(p)
h
i
1
+ p p p p I(p)
+
p=0
p=0
2
(4.87)
Show that for large ,

1 2
p + O(2 ).
2
8
(4.88)
1
d4 q
.
4
2
(2) [m + (q + p)2 ]2
(4.89)
I(p) = I(0) +
Let K(p) be given by
K(p) =
q 2 <2
Neglecting terms that vanish as

K(p) and K(0)?
d4 y , what is the relation between
Hint: use symmetry arguments to conclude that,

Z
Z
q 2 <2
q 2 <2
d4 q
q
= 0,
4
2
(2) [m + q 2 ]n
Z
d4 q
q q
d4 q
q2
=
.
(2)4 [m2 + q 2 ]n
4 q2 <2 (2)4 [m2 + q 2 ]n
(4.90)
(4.91)
The above examples I and K may seem a little academic, but in gauge
theories the self-energy diagrams typically lead to quadratically divergent
integrals of the type
Z
d4 q
(2q + p) (2q + p)
,
4
2
(2) [m + q 2 ][m2 + (q + p)2 ]
(4.92)
105
4.9. PROBLEMS
which one would like to evaluate with the method (4.42), with subsequent
shifting of integration variable as in (4.44). For such integrals the ambiguities of sharp momentum cutoffs have to be well understood and absorbed
into the renormalization prcedure. It is however, much simpler to use a
regularization that is invariant under shifts of integration variables. A convenient method is dimensional regularization, which is described in many
text books on relativistic quantum fields.
2. Revisiting the two-scalar-field model
a. Verify Fig. 4.20, i.e. derive the bare vertex function in position space
and Fourier transform this to momentum space.
b. Verify that the tree graphs in Fig. 4.21 represent the zero loop contribution to h1 2 3 4 iconn and determine the corresponding mathematical
expression (in momentum space) using the path integral.
c. Similarly, verify the tree graph contribution to h1 4 iconn shown in
Fig. 4.22 and write down the corresponding mathematical expression.
d. Verify the expression (4.78) for the differential cross section (cf. Prob.
2.3).
e. Assume that in the quantum theory the action of the two scalar field
model (4.75) is the total action without counterterms. Introduce the counterterms corresponding to mass, coupling and wave function renormalization and write down S0 , S1 and S for this model.
3. Nonzero vacuum expectation value
The 4 model and also the model (4.75) has a discrete symmetry .
From this one may conclude that hi = 0. However, the symmetry may be
broken spontaneously, as in the Ising model. This does not happen for free
fields, but with interactions it is possible, depending on the value of the parameters of the model. Here we assume no spontaneous symmetry breaking
(we have implicitly made this assumption throughout this chapter).
On the other hand, there is no symmetry in the model (4.75).
Therefore we can expect that h(x)i 6= 0. Under homogeneous circumstances it will be independent of x. The full propagator is defined as a
correlation function,
G0 (x, y) = h(x)(y)i h(x)ih(y)i.
(4.93)
a. Verify that in one-loop order hi is given by the diagram in Fig. 4.23

and determine the corresponding mathematical expression.
b. Verify that the selfenergy vertex function for the field, (p), is given
in one loop order by the diagram in Fig. 4.24
Figure 4.23: One loop contribution to hi.
Figure 4.24: One loop contribution to i (p).

c. Evaluate (p) with the cutoff method.
d. Determine the counterterms to make (p) finite, such that
G0 (p)
near its pole.
M2
1
,
+ p2
(4.94)
Chapter 5
Spinor fields
Up to now we have only considered in detail fields that have integer spin under
rotations. There are also representations of the rotation group with half integer
spin which are embedded into representations of the Lorentz group.1 The simplest
have spin 1/2, these are called spinor representations. Majorana fields and Dirac
fields are real and complex spinor fields, respectively. This chapter provides an
introduction to the classical fields, the next chapter deals with the quantum case.
5.1
Spinors
To construct spinor representations we introduce Dirac matrices. These are four

4 4 matrices which satisfy the anticommutation relations2
{ , } = 2 1.
(5.1)
= ,
02 = 1, k2 = 1, k = 1, 2, 3.
(5.2)
(5.3)
Hence,
Below we shall give explicit examples of matrices satisfying the above relations.
Let us now first show that generators S of Lorentz transformations can be
constructed in terms of them. Consider
1
(5.4)
S = S = i [ , ], spinor representation
4
= 0, =
(5.5)
1
= i , 6= .
(5.6)
2
1
Recall the discussion below eq. (1.43).

This relation reads more explicitly ( )ab ( )bc + ( )ab ( )bc = 2 ac , where a, b, c =
1, . . . , 4 are the matrix indices. We usually denote the identity matrix (11)ac = ac simply by 1,
and sometimes even omit this symbol as well in compact notation.
2
107
108
CHAPTER 5. SPINOR FIELDS
Using the anticommutation relations3 (5.1) we have for 6= ,

1
[S , ] = i [ , ]
2
1
1
= i { , } + i { , }
2
2
= i( ),
(5.7)
(5.8)
(5.9)
and in addition for 6= ,

1
[S , S ] = i [S , ]
2
1
= i ([S , ] + [S , ])
2
1
= ( + )
2
= i( S + S + S S ).
(5.10)
(5.11)
(5.12)
(5.13)
In the last line we used

= 2iS + 1,
(5.14)
etc. Hence, the matrices S satisfy the commutation relations (1.53) and therefore they provide a representation of the Lorentz group.
We now go into more detail on the properties of the Dirac matrices. The
gamma matrices are unitary,
= 1, no sum over ,
(5.15)
which implies with (5.3),

0 = 0 ,
k = k , k = 1, 2, 3.
(5.16)
As a special notation for i 0 we introduce a matrix

= i 0 = i0 = ,
(5.17)
in terms of which we can write

= .
(5.18)
The notation
= i ,
(0 = 0 = 1)
(5.19)
also occurs. An important combination is furthermore the matrix 5 ,

5 5 = i 0 1 2 3 = 5 ,
3
(5.20)
Basic identities such as [ab, c] = a[b, c] + [a, c]b and [ab, c] = a{b, c} {a, c}b are useful here.
109
5.1. SPINORS
which anticommutes with all the ,
5 = 5 ,
= 0, , 3.
(5.21)
There is an infinite number of realizations of the Dirac matrices, which differ

by unitary transformations. We shall give two: a so-called chiral representation
in which 5 is diagonal, and a Majorana representation in which the are real.
It is convenient to use a tensor product notation in terms of Pauli matrices k ,
k = 1, 2, 3. We also introduce 4 4 matrices k and k which are essentially
just the Pauli matrices. Introducing a two-index notation for the Dirac index a,
a , with , = 1, 2 (so a = 1, . . . , 4 = 11, 21, 12, 22), they are given
by
(5.22)
(k ),0 0 = (k )0 0 , (k ),0 0 = (k ) 0 0 ,
where the k on the right hand side are the 2 2 Pauli matrices. Their action
on a spinor is given by
(k l ) = (k )0 (l ) 0 0 0 ,
(k ) = (k ) 0 0 , (k ) = (k )0 0 ,
(5.23)
(5.24)
with the usual summation over repeated indices. For simplicity we use the same
symbol for the 4 4 k as for the 2 2 Pauli matrices and the unit matrix is
often omitted. A chiral representation is now given by
= i 0 =
5 =
0 11
11 0
0
ik
ik 0
11 0
0 11
k
0
0 k
= 1 ,
(5.25)
= 2 k ,
= 3 ,
!
= 3 k ,
(5.26)
(5.27)
0 = 1.
(5.28)
where we have also indicated the matrices in 2 2 block form. A Majorana

representation is given by
1 = 3 , 2 = 2 2 , 3 = 1 , 0 = i3 2 ,
(5.29)
which are real, = , and imply

5 = 1 2 = 5 .
(5.30)
The anticommutation relations (5.1) can be easily checked from these representations, since [k , l ] = 0. For example, in the chiral representation,
{k , l } = 22 {k , l } = 2kl , {0 , k } = i{1 , 2 }k = 0, 02 = 21 = 1.
(5.31)
110
A spinor field (x) is now a field transforming under Lorentz transformations

x = `x as
0
0 (x0 ) = D(x),
(a0 (x0 ) = Dab b (x), a, b = 1, . . . , 4),
(5.32)
with
1
def
` = exp(i S
),
2
1
1
D(`) = exp(i S ) = exp( ),
2
4
(5.33)
def
where S
are the generators of the defining representation (1.52) and S the
generators in a spinor representation.4 We see that D is real in a Majorana
representation of the gamma matrices. Accordingly, it can be convenient use
such representations for Majorana fields:
(x) = (x), D = D ,
Majorana.
(5.34)
From two Majorana fields 1,2 we can construct a complex field, a Dirac field ,
1
= (1 i2 ).
2
(5.35)
It is then generally convenient to allow the to be complex. When we make

a unitary transformation from a Majorana representation to a representation
where the are complex, the in (5.35) stays of course complex but the new
1,2 become also complex. In general then, Majorana fields are complex but
satisfy a reality constraint, cf. (5.120) in Appendix 5.5. We shall continue with
the Dirac case, from which it is straightforward to specialize to the Majorana
case.
Let us now see how the Dirac field transforms under rotations and confirm
that the spin of the representation is 1/2. Rotations only involve kl , k, l = 1, 2, 3.
Now in the chiral representation (5.28) we have explicitly
1
1
Skl = m =
2
2
and
D=
m 0
0 m
0
ei 2
i 12
0
e
(k, l, m = 1, 2, 3 cycl.)
1
, k klm lm ,
2
rotation,
(5.36)
(5.37)
which shows that the representation has spin 1/2. For a special Lorentz transformation we have similarly
S0k
4
1
1
= i 3 k =
2
2
ik 0
0
ik
spin
For notational convenience we suppress the label spin on S
and Dspin .
(5.38)
111
5.1. SPINORS
and
1
D=
e 2
0
12
0
e
k 0k ,
special Lorentz transformation. (5.39)
The 2 2 block diagonal form shows that D is a reducible representation of the

proper Lorentz group. The representation in the upper 2 2 block is inequivalent
to that in the lower block. In standard notation these representations are denoted
by (1/2,0) and (0,1/2). Another notation is R and L, D (1/2,0) = DR , D (0,1/2) =
DL . The fields R and L transforming under the irreducible DR and DL can be
obtained from by applyting the chiral projectors PR,L defined by
1
PR = (1 + 5 ),
2
1
PL = (1 5 ,
2
(5.40)
with the properties

PR2 = PR ,
PL2 = PL ,
PR + PL = 1,
PR PL = 0.
(5.41)
In the chiral representation
R = P R
1
2
3
4
1
2
0
0
L = P L
1
2
3
4
0
0
3
4
(5.42)
L = R ,
(5.43)
In the Majorana representation (5.29)
R = P R
1
2
3
4
1 i4
2 + i3
i(2 + i3 )
i(1 i4 )
where the latter relation is true for real . We see that a single real Majorana
field is equivalent to a field R with two independent complex components. We
also see that R is equivalent to a L .
The representation D = DR + DL is irreducible if parity is included: we
shall see below that the parity transformation is represented by DP = 0 , which
interchanges the two blocks of the chiral representation.
In Nature, parity is violated, so why does not one just use one of these irreducible representations? The reason is that we need both representations for
making Lorentz tensors, in particular scalars such as the action and, in the quantum theory, Lorentz invariant amplitudes. It is a curious fact that the 4 4
D in the Majorana representation is irreducible within the real numbers. So in
Majorana language the question posed above does seem to not come up.
112
We now turn to the construction of Lorentz invariant or covariant combinations of and . Instead of we shall use = T , which is more convenient.
As in (5.32) we may consider to be column vector on which the Dirac matrices can act from the left, and a row vector on which the matrices can act
from the right. Now transforms as 0 (x0 ) = (x) D . However, 0 0 6=
since D is not unitary in general: D 6= D 1 (it is known that finite dimensional
representations of the Lorentz group cannot be unitary). Using (5.18) we have
instead
1
1
D = exp( ) = exp( )
4
4
1
= exp( )
4
= D 1 ,
(5.44)
(5.45)
(5.46)
or using 2 = 1,
D = D 1 .
(5.47)
This shows that is a scalar:

0 (x0 ) 0 (x0 ) = (x)D D(x) = (x)(x).
(5.48)
We see that the matrix plays the role of a metric in spinor space. It is customary
to use a special notation for the combination ,
which transforms as
(5.49)
0 (x0 ) = (x)D
.
(5.50)
As the notation anticipated, is a vector matrix, in the sense that

D 1 D = ` .
(5.51)
This follows by writing the left hand side in infinitesimal form,

1
1
1
(1 i S ) (1 + i S ) = + i [ , S ] + . . .
2
2
2
(5.52)
and using (5.9), in which we recognize on the right hand side the generators in
the vector representation (1.52),
[ , S ] = i( ) .
(5.53)
It follows that (x)

(x) is a vector field,
0 (x0 ) 0 (x0 ) = ` (x)

(x).
(5.54)
5.2. ACTION, DIRAC EQUATION AND NOETHER CHARGES
113
Similarly, is a tensor matrix; the antisymmetric part is proportional to S ,

while the symmetric part is proportional to .
Let us introduce at this point suitable representatives of parity and time
reversal,
DP = 0 ,
DT = i 0 5
(5.55)
(5.56)
(which are both real in a Majorana representation). These are reasonable definitions as can be seen from
DP1 DP = (`P ) , 0 (x0 ) 0 (x0 ) = (`P ) (x)

(x),
DT1 DT = (`T ) , 0 (x0 ) 0 (x0 ) = (`T ) (x)

(x).
(5.57)
(5.58)
The matrix 5 is a Lorentz scalar,

D 1 5 D = 5 ,
(5.59)
since it commutes with the generators S . It is in fact a pseudoscalar under

parity and time reversal:
DP1 5 DP = 5 ,
DT1 5 DT = 5 .
(5.60)
The antisymmetrized product of three Dirac matrices

1
= i 5
3!
(5.61)
is a pseudovector.
The 16 matrices 1, , 2iS , i 5 and 5 are linearly independent and form
a basis for the 4 4 matrices: an arbitrary 4 4 matrix can be written as a linear
superposition of them. We shall not go into details here.
5.2
Action, Dirac equation and Noether charges
Consider the problem of writing down an action S for spinor fields. To get linear
field equations from the stationary action principle, S has to be quadratic or
bilinear in the fields. We also want it to be local, Lorentz and parity invariant,
i.e. of the form
Z
S = d4 x L(x),
(5.62)
with a real Lagrange density L which is a scalar under Lorentz transformations as
well as under the parity transformation P . The reality condition will be relevant
for Dirac fields, which are complex, so consider first this case. Candidate terms
114

+ c.c,
5 + c.c.,
and so on. A convenient
in L are ,
rule is
[1 M 2 ] = 2 M 1 ,
(5.63)
= is real, while
where M is any matrix. So
] = [ ] = =
,
[
(5.64)
where we used (5.18). Keeping a minimal (> 0) number of derivatives and avoiding parity violation (5 leads to a sign change under P ), the minimal form for L
is given by
+ 1
1
(m
+ ),
L = m
2
2
Dirac.
(5.65)
We introduced the convenient shorthand
1
f1 f2 = (f1 f2 f1 f2 ).
2
(5.66)
The parameter m is real and the coefficient of the derivative term has been chosen
unity, which is a normalization convention for the fields. The need for the overall
minus sign will be verified later.
Requiring S to be stationary leads to the Dirac equation. In the Dirac case
we treat and as independent, since the real and imaginary parts 1,2 of
are independent variables. The relation of 1,2 to and can be written as a
linear transformation of variables
a
a
1
=
2
1 i
1 i
1a
2a
(5.67)
(a = 1, , 4 is the spinor index). The variation with respect to gives

S =
+ ) =
d x (m
4
+ ),
d4 x (m
(5.68)
where we made a partial integration (with = 0 at the boundary of the integration volume, as usual). Requiring the action to be stationary gives the Dirac
equation
(m + ) = 0.
(5.69)
The conjugate equation
0 = S =
+ ) m
= 0,
d4 x (m
can also be obtained by hermition conjugation of (5.69).
(5.70)
5.2. ACTION, DIRAC EQUATION AND NOETHER CHARGES
115
Symmetries of the action lead to Noether currents and conserved quantities.

These can be found by the methods used in sect. 1.7 for the scalar field. Infinitesimal spacetime dependent translations
= ,
= ,
(5.71)
lead to the energy-momentum tensor
+ L,
T =
T = 0.
(5.72)
Similarly, the infinitesimal spacetime dependent Lorentz transformations

1
= (x i S ),
2
1
= (x + i S
),
2
(5.73)
lead to the generalized angular momentum currents

i
J = x T x T + {
, S },
2
J = 0.
(5.74)
The conserved energy-momentum and generalized angular momenta are as usual

given by
Z
Z
P = d3 x T 0 , J = d3 x J 0 .
(5.75)
Notice that the above T is not symmetric: it cannot be used in the Einstein
equations. The route of finding a better energy-momentum tensor via a Dirac
action in general relativity is too involved to be exposed here. There are other
ways to improve the above T into a symmetric one, see for instance Weinberg
I sect. 7.4. Such improvement does not affect the total P and J .
The action has more symmetries. It is invariant under phase transformations
0 (x) = ei (x).
0 (x) = ei (x),
(5.76)
Using spacetime dependent infinitesimal transformations
= i,
= i,
(5.77)
we find a conserved current from

S =
d4 x j = 0 j = 0.
(5.78)
The current and corresponding charge are given by

,
j = i
Q=
d xj =
d3 x .
(5.79)
If the parameter m vanishes we have one more continuous symmetry:

0 (x) = ei5 (x),
i5
,
0 (x) = (x)e
(5.80)
116
with corresponding conserved current and charge

5 ,
j5 = i
Q5 =
d3 x j50 =
d3 x 5 .
(5.81)
We end this section with an at first sight puzzling phenomenon. For the
Majorana case we assume to be real and choose a Majorana representation
of the . Then with the conventional factor 1/2 for real fields, the Lagrange
density becomes
1
L = T (m + ),
2
Majorana.
(5.82)
Now, since is hermitian and imaginary (in a Majorana representation) it is

antisymmetric, = T . Therefore the term T is zero if is an ordinary real
number. The derivative term has a similar property, as follows from i =
T (in a Majorana representation) and the fact that is antisymmetric. The

Majorana action is identically zero for a real field! This raises the question how to
use the action in a path integral. Or is it not true that a Dirac field is equivalent
to two Majorana fields? The surprising resolution of this puzzle is, that spinor
fields in a path integral have to be anticommuting, i.e. a (x)b (y) = b (y)a (x).
For such objects the Majorana action is not identically zero.
5.3
Solutions of the Dirac equation
To get familiar with spinors and also for later use, we shall now study plane wave
solutions of the Dirac equation
(m + ) = 0.
(5.83)
Multiplying from the left by m we encounter

1
m2 ()2 = m2 { , } = m2 ,
2
(5.84)
so a solution of the Dirac equation is also a solution of the Klein-Gordon equation.

The Dirac equation can be put into hamiltonian form5 by multiplication from the
left by :
i0 = (m i) = HD , HD m i,
(5.85)
where k = k = ik . The Dirac hamiltonian HD is a hermitian differential
operator in the usual inner product,
Z
5
d3 x 1 HD 2 =
Z
d3 x 2 HD 1
(5.86)
The similarity with the Schr

odinger wave equation will be commented upon at the end of
sect. 6.2
117
5.3. SOLUTIONS OF THE DIRAC EQUATION
assuming infinite volume or appropriate boundary conditions allowing partial

intgration without surface terms. For simplicity we shall assume for a while
periodic boundary conditions in an L3 box. Because of its hermiticity we may
expect HD to possess a complete set of eigenfunctions, which can be found among
the eigenfunctions exp(ipx) of :
HD wp eipx = eipx HD (p) wp ,
HD (p) m + p.
(5.87)
Here HD (p) is just a hermitian 4 4 matrix which we expect to have four independent eigenvectors. Since k and satisfy the anticommutation relations
{k , l } = 2kl ,
{k , } = 0,
(5.88)
we have the identity

1
(5.89)
HD (p)2 = m2 + mpk {, k } + pk pl {k , l } = m2 + p2 .
2
Hence, the eigenvalues of HD (p) are equal to m2 + p2 , which are presumably

two-fold degenerate:
HD (p) wp = Ep wp ,
= ,
= ,
Ep =
m2 + p 2 .
(5.90)
Here distinguishes the two eigenvectors with the same eigenvalue of HD . To

specify this further we can look for another matrix which commutes with HD and
use its eigenvalues to pin down the ws with the same . A suitable candidate is
the hermitian matrix pS (Sk = klm Slm /2 is the spin matrix and p = p/|p| is the
unit vector in the direction p). This commutes with HD (p) and has eigenvalues
1/2:
[m + k pk , pl Sl ] = 0, (
pk Sk )2 = 1/4
(5.91)
(which follow from the anticommutation relations (5.88), or by using the chiral
representation of the gamma matrices). So we may choose the wp to satisfy
1
p S wp = wp .
2
(5.92)
Such spinors are called helicity spinors. In the zero mass case [HD , 5 ] = 0, so
then the helicity spinors can be chosen to be also eigenvectors of 5 :
5 wp = wp ,
m = 0 only
(5.93)
(it can be seen e.g. in the chiral representation that = 25 S).

Another possibility for specifying the meaning of is to start with p = 0 and
take the ws to be basis vectors with the usual properties under the action of S:
1
0 w00 ,
2
1
w0.
=
2
S w0 =
S3 w0
(5.94)
(5.95)
118
Then the p 6= 0 ws can be obtained by boosting w0 with the special Lorentz
transformation taking (m, 0) into (Ep , p) (cf. Appendix 5.6).
In a Majorana representation where is real,
HD (p) = HD (p),
S = S,
(5.96)
so we can choose to relate the positive and negative energy spinors by
wp,, = wp,,+
.
(5.97)
In a general representation there is a similar relation involving the so-called charge

conjugation matrix, which is explained in the appendix 5.5.
The orthonormality and completeness relations can be chosen to have the
(non-covariant) form
wp
wp0 0 = 0 0 ,
wp wp
= 11.
(5.98)
(5.99)
Putting things together we have for the time-dependent equation

i0 wp eipx = p0 wp eipx
= HD (p) wp eipx = Ep wp eipx ,
(5.100)
so the frequencies are given by p0 = Ep .

For infinite volume we shall use covariantly normalized spinors defined by
u(p, ) =
v(p, ) =
2Ep wp,,+ ,
2Ep wp,, ,
(5.101)
(5.102)
Note the minus signs as in (5.97). In a general representation u and v are related by the charge conjugation transformation (5.119). Their properties are the
subject of Appendix 5.6 and Problems 1, 2.
5.4
Summary
The 4 4 spinor representation of the Lorentz group is conveniently constructed

with the help of Dirac matrices. This representation is reducible into 22 blocks,
from which it can be seen that rotations are represented in the same way as the
spin 1/2 representation in non-relativistic quantum mechanics. Including parity,
the 4 4 representation is irreducible.
Spinor fields are fields transforming in the spinor representation. It is possible to choose a real representation of the gamma matrices, called the Majorana
5.5. APPENDIX: CHARGE CONJUGATION MATRIX C
119
representation, with corresponding real spinor fields. Two real Majorana fields
make one complex Dirac field.
The action for a free spinor field can be constructed such that it is Poincareand parity-invariant. It leads to a field equation called the Dirac equation. The
continuous symmetries of the action give rise to Noether charges P and J
in the usual way, which are conserved once the field equations are satisfied. The
Dirac action has an additional U(1) symmetry, with conserved charge Q and, in
case of zero mass parameter, a second U(1) symmetry with charge Q5 . The latter
(but not the former) is also present in case of a single Majorana field.
The Dirac equation can be rewritten in a form resembling the Schrodinger
equation, which was the motivation for Dirac for inventing this equation. However, in the present context it is just a classical field equation which has nothing
to do with quantum mechanics yet.
A plane wave ansatz for solutions to the Dirac equation leads to a (4 4)
matrix-eigenvalue equation. The solutions are called polarization spinors (by
analogy to polarization vectors in the Maxwell case), which are labelled by the
sign of the energy (positive or negative frequency), the momentum (wave vector), and the sign of the helicity (the eigenvalue of the z-component of the spin
matrix in the rest frame).
5.5
Appendix: Charge conjugation matrix C
In general, hermiticity properties are preserved under a change of representation.

In a real representation these become symmetry properties under transposition.
It is possible to express symmetry and reality properties of Dirac matrices in a
representation-independent form. For this we need the so-called charge conjugation matrix C.
In any representation there is a unitary antisymmetric matrix C,
C C = 1, C T = C,
(5.103)
relating and ( )T according to

T = C C.
(5.104)
In the Majorana representation T = and (cf. (5.18))

C = = i 0
(5.105)
(= 3 2 ). In any other representation obtained by a unitary transformation

(indicating the Majorana representation by thefor the moment):
= U U , U U = 1,
U T
T = (U U )T = U T U T = U
= U U U U T ,
(5.106)
(5.107)
(5.108)
120
and we obtain C in the form
C = U U T C,
C = U U T .
(5.109)
We then also have in any representation

= ( )T = ( )T = C C
= C C.
(5.110)
(5.111)
In the Majorana representation C = 1. In our chiral representation a suitable U

is given by
1
U = ei 4 2 2 = (1 i2 2 )
(5.112)
2
(e.g. U 1 U = U 3 U = U 2 2 = i2 2 3 = 2 1 ), and
C = U U T = ei 2 2 2 = i2 2 ,
C = 3 2 = 0 2 .
(5.113)
The charge conjugation matrix is useful for relating D T to D 1 ,

D T = C D 1 C,
(5.114)
D = D T = C D C.
(5.115)
and D to D,
Cab
a b
For example,
= T C is a scalar.
For general complex spinors and = the so-called charge conjugate
spinors (c) and (c) are defined as
T = C = C ,
(c) = (C)
(c) = (C )T
(5.116)
(where the formula for (c) follows from (c) ). Under Lorentz transformations
(c) transforms like . In particular the antiparticle spinors v(p, ) are defined
as the charge conjugates of u(p, ). Denoting again the spinors in the Majorana
representation by a, we have
v(p, ) = u
(p, ),
(Majorana representation)
(5.117)
and in a general representation

u(p, ) = U u
(p, ), v(p, ) = U v(p, ),
(5.118)
v(p, ) = U v(p, ) = U u
(p, ) = U U u (p, ) = C u (p, )
(c)
= u (p, ), arbitrary representation.
(5.119)
The Majorana property of a spinor field (i.e. is real in a Majorana representation) can be expressed in representation independent form: the Majorana
field is self (charge) conjugate,
(c) = ,
(c) = .
(5.120)
121
5.6. APPENDIX: POLARIZATION SPINORS
5.6
Appendix: Polarization spinors
The main text introduced polarization spinors as eigenvectors of the Dirac hamiltonian. We make a fresh start here and present a construction based solely on
the spinor representation of the Lorentz group. (See also Problem 2.)
A particle at rest transforms under rotations like a two component spinor ,
+ =
1
0
0
1
(5.121)
From these two-component spinors we make a four-component spinor for a particle at rest in the chiral representation,
!
,
(5.122)
u(
p, ) = 2m , =
2
where m is the particle mass and we use the notation
= 0, p0 = m.
p
(5.123)
The curious normalization factor 2m is put in for later convenience. We can

characterize u(
p, ) by the eigenvalues of the two commuting matrices 3 and ,
3 u(
p, ) = u(
p, ), 3 =
u(
p, ) = u(
p, ), = .
(5.124)
(5.125)
These relations together with u(

p, ) = 21 (1 i2 ) u(
p, +) serve to characterize
u(
p, ) also in a general representation.
Polarization spinors u(p, ) for arbitrary momentum p now follow by applying
a standard boost `p which takes p into p (cf. (5.39) and Problem 1.10, see also
Prob. 2):
`p p = p,
p0 =
m2 + p 2
Dp D(`p ) = e5 /2 ,
, p
=
= p
Applying this standard boost to u(

p, ) we get
(5.126)
p
,
|p|
tanh =
u(p, ) = Dp u(
p, )
5 )u(
= (cosh + sinh p
p, )
2
q
q2
5 ) .
= ( p0 + m + p0 m p
|p|
. (5.127)
p0
(5.128)
(5.129)
(5.130)
We shall also need conjugate spinors related to u(p, ) by charge conjugation (cf.
(5.116)),
v(p, ) u(c) (p, )
= C u(p, ) = [
u(p, )C]T ,
v(p, ) = u
(c) (p, ) = [C u(p, )]T .
(5.131)
(5.132)
(5.133)
122
In the Majorana representation C = , giving simply

v(p, ) = u(p, ) ,
Majorana rep.
(5.134)
Since charge conjugate spinors transform under Lorentz transformations like ordinary spinors we have
v(p, ) = Dp v(
p, )
q
= ( p0 + m +
(c)
= C .
(5.135)
q
(c)
5 ) ,
p0 m p
(5.136)
(5.137)
Furthermore, at rest
3 v(
p, ) = v(
p, ),
v(
p, ) = v(
p, ),
(c)
(c)
3 = ,
(c)
(c)
,
(5.138)
(5.139)
and
u
(
p, ) i u(
p, 0 )
v(
p, ) i v(
p, 0 )
u
(
p, ) u(
p, 0 )
v(
p, ) v(
p, 0 )
u
(
p, ) v(
p, 0 )
=
=
=
=
=
2
p 0 ,
2
p 0 ,
2m 0 ,
2m 0 ,
v(
p, ) u(
p, 0 ) = 0.
(5.140)
(5.141)
(5.142)
(5.143)
(5.144)
The orthogonality of a u(
p, ) and a v(
p, 0 ) follow from the fact that they are
eigenvectors of with different eigenvalues. From the above follow the relations
for general p:
u
(p, ) i u(p, 0 )
v(p, ) i v(p, 0 )
u
(p, ) u(p, 0)
v(p, ) v(p, 0)
u
(p, ) v(p, 0)
=
=
=
=
=
2p 0 ,
2p 0 ,
2m 0 ,
2m 0 ,
v(p, ) u(p, 0) = 0.
(5.145)
(5.146)
(5.147)
(5.148)
(5.149)
For example,
u
(p, ) i u(p, 0 ) = u
(
p, )Dp1 i Dp u(
p, ) = `p 2
p 0
= 2p 0 .
(5.150)
Since u
(p, )i 0 = u(p, ) we can interprete (5.145) and (5.146) for = 0 as
orthogonality relations. The us are orthogonal to the vs in the sense
u(p, ) v(
p, 0 ) = u(
p, ) Dp Dp v(
p, 0 )
0
= 0, p (p, p ),
(5.151)
(5.152)
123
5.6. APPENDIX: POLARIZATION SPINORS

where we used Dp = Dp = Dp1
.
Similarly, we have completeness type relations at rest,
X
u(
p, ) u
(
p, ) = 2m
= m(1 + )
= m i p ,
(5.153)
and for general momentum

X
u(p, ) u
(p, ) = m i p ,
(5.154)
v(p, ) v(p, ) = m i p .
(5.155)
The second relation follows from the first and the definition of v(p, ),
X
va (p, ) vb (p, ) =
[
u(p, )C]a [C u(p, )]b = [C (m i p ) C]ba
= [C (m i p ) C]Tab = [C (m i p )T C ]ab
= (m + i p )ab .
(5.156)
In the Majorana representation these relations follow more easily from the reality
of the and v = u .
Because of the orthogonality relations (5.145), (5.146) and (5.152) the completeness relation in four dimensional spinor space reads
X
[u(p, )u(p, ) + v(
p, )v(
p, ) ] = 2p0 .
(5.157)
Eqs. (5.125) and (5.139) generalize to arbitrary p as,

i p u(p, ) = mu(p, ), i p v(p, ) = mv(p, ),
(5.158)
which turn out to be the free Dirac equation in momentum space.

We conclude this appendix with the zero-mass limit of the polarization spinors
(which is also their approximate form for high energies). From eqs. (5.130) and
(5.136) we see that for m 0,
u(p, )
v(p, )
5 ) ,
|p| (1 + p
(5.159)
5 ) .
|p| (1 + p
(5.160)
(c)
Note that the quantity within parenthesis is essentially a projector: [(1 + p

5 )/2. Let us change the specification of the such that
5 )/2]2 = (1 + p
they become eigenvectors of the helicity matrix
1
,
p
2
(5.161)
124
with eigenvectors /2. This can be done by a standard rotation which brings the
,
three axis along p
1
(, ) = ei 2 3 ei 2 2 ,
= (sin cos , sin sin , cos ).
p
(5.162)
(5.163)
Then is the sign of the helicity,

(c)
(c)
(, ) = (, ), p
(, ) = (, ),
p
(5.164)
and the helicity is tied to 5 ,
u(|p|, , , ) =
v(|p|, , , ) =
|p| (1 + 5 ) (, ),
(5.165)
|p| (1 5 ) (, ).
(5.166)
(c)
We recognize here chiral projectors PL,R = (1 5 )/2. Since 5 commutes with

we can choose the helicity s to be eigenvectors of 5 . The eigenvalue of 5 ,
which takes values 1, is called the chirality (handedness). We see that for the
u-spinors = , whereas for the v-spinors = . Then a right handed spinor
uR = PR u, which in the chiral representation has only the upper two components
nonzero, has positive helicity, while a left handed spinor uL = PL u, which in the
chiral representation has only the lower two components nonzero, has negative
helicity, and vice versa for vR,L .
5.7
Appendix: Traces of gamma matrices and

other identities
Calculations of unpolarized cross sections or decay rates of processes involving

fermions can be carried out in terms of traces over products of gamma matrices.
The following identities can be derived (see for example Bjorken & Drell I sect.
7.2 and De Wit & Smith sect. E.4):
Tr 1 n
Tr 1
Tr
Tr
Tr 5 1 n

=
=
=
=
=
=
=
=
=
=
0, n = odd,
4,
4 ,
4( + ),
0,
n = 0, 1, 2, 3,
1 4
, n = 4,
4i
4,
2 ,
4 ,
2 .
(5.167)
(5.168)
(5.169)
(5.170)
(5.171)
(5.172)
(5.173)
(5.174)
(5.175)
(5.176)
125
5.8. PROBLEMS
5.8
Problems
1. Projectors
A projector P satisfies the equation P 2 = P ; its eigenvalues are 0 and 1.
Polarization sums over outer products of Dirac spinors produce projectors
onto the positive and negative energy subspace of the Dirac Hamiltonian
HD (p). Verify
X
wp wp
=
Ep HD (p)
P ,
2Ep
P2 = P ,
P+ P = 0,
(5.177)
P + + P = 1
(5.178)
A covariant version of these relations is given by

X
u(p, )
u(p, ) = m ip = (m ip)2 /2m,
(5.179)
v(p, )
v (p, ) = (m + ip) = (m + ip)2 /2m.
(5.180)
2. Polarization spinors
(See also Appendix 5.6.)
In the following the derivations can be carried out using only algebraic
properties; explicit representations are not required (but e.g. the chiral
representation can be useful if one gets stuck).
Let `p be the special Lorentz
that transforms p (m, 0)
2 transformation

0
2
into p: ` p = p , p = m + p . In matrix notation we have
`p = ei
kS
0k
k =
pk
,
|p|
tanh =
|p|
,
p0
defining representation.
(5.181)
This can be seen as follows. First choose p in the 3-direction. Then the
boost B exp(iS03 ) reproduces (1.35) (cf. Problem 1.10). To get p to
point in an arbitrary direction, apply a suitable rotation R: pk = Rk3 |p|.
Then p = RB p = RBR1 p, since p is invariant under rotations. Hence,
`p = RBR1 is a candidate for `p . That this is correct indeed follows from
the fact that the boost generator is a 3-vector under rotations, R S0k R1 =
Rl k S0l (cf. Problem 1.11).
From now on let S be the generators in the spinor representation. Then
Dp D(`p ) = ei
kS
0k
spinor representation.
(5.182)
We can use the special boost Dp to define rest-frame specified polarization

spinors, als follows.
126

Let u(
p, ) be a spinor at rest, characterized by (recall Sk 21 klm Slm )
u(
p, ) = u(
p, ),
1
u(
p, ),
S3 u(
p, ) =
2
p0 = m).
u (
p, )u(
p, 0 ) = 2
p0 0 (
(5.183)
(5.184)
(5.185)
Note that [, S3 ] = 0. Define

u(p, ) = Dp u(
p, ).
(5.186)
u
(p, ) u(p, 0) = 2m 0 ,
u
(p, ) i u(p, 0 ) = 2p 0 ,
X
u(
p, )
u(
p, ) = m(1 + ) = m i p ,
(5.187)
(5.188)
(5.189)
Verify that
u(p, )
u(p, ) = m i p ,
(5.190)
i p u(p, ) = m u(p, )
(5.191)
This last equation shows that u(p, ) exp(ipx) satisfies the Dirac equation
and the u are positive-energy polarization spinors.
In a Majorana representation = , and the negative-energy spinors v
may simply be defined by v(p, ) = u(p, ) (cf. (5.97)). We then have
(verify)
1
S3 u(
p, ) = u(
p, ) ,(5.192)
2
v(p, ) u(p, ) = Dp u(
p, ) = Dp u(
p, ) ,
(5.193)
u(
p, ) = u(
p, ) ,
X
v(p, )
v (p, ) = (m + i p ).
(5.194)
Verify also that

u (p, )u(
p, 0 ) = 0,
p = (p0 , p).
(5.195)
In an arbitrary representation of the this becomes (cf. (5.119))

u (p, )v(
p, 0 ) = 0,
p = (p0 , p).
(5.196)
This is in accordance with the fact that the positive-energy eigenfunctions of

the Dirac hamiltonian HD are orthogonal to the negative-energy eigenfunctions, since they correspond to different eigenvalues of a hermitian operator:
Z
d3 x [u(p, )eipx ] v(p0 , 0 )eip x = 0.
(5.197)
Chapter 6
Quantized spinor fields: fermions
In this chapter we quantize the spinor fields. We shall find that their quanta have
the (to be expected) interpretation as spin 1/2 particles. We shall go through the
essentials of the arguments leading to the spin-statistics theorem, which states
that the spin 1/2 particles described by a relativistic spinor field have to be
fermions. Functional techniques and the path integral lead to the introduction
of anticommuting numbers, resulting in elegant analogues of many formulas of
the bosonic (commuting) case.
6.1
Canonical quantization: wrong
The real (Majorana) field has presumably half the number of degrees of freedom
of the complex (Dirac) field and would therefore be a good starting point for
quantization. But below (5.82) we deduced that the action for the real field
vanishes identically, so how to proceed? It seems best to face this problem headon: make the logical jump by assuming that the spinor field is anticommuting.
However, let us first try to evade the problem, by noting that it seems to be
absent in the case of complex fields. Proceeding this way we are led by the
canonical method into a cul-de-sac, from which we will recover by going back
and changing the quantization rules in accordance with the statistics expected
for spin 1/2 particles.
Since the action is linear in the time derivative it is natural to assume it has
the pq form (1.90). Let us rewrite the Dirac action in terms of the real and
imaginary parts of ,
1
= (1 i2 ),
2
(6.1)
and use a Majorana representation for , such that the symmetries of various
127
128
CHAPTER 6. QUANTIZED SPINOR FIELDS: FERMIONS
objects are explicit. This gives1

S =
H =
dt L,
L=
d3 x 1T 2 H,
d 3 x HD =
d3 x i2T HD 1 ,
HD = m i ,
(6.2)
(6.3)
(6.4)
Recall that = = T and k = kT = k in a Majorana representation,

so L and H are real. The functional H is evidently the hamiltonian (cf. (1.90),
which is in accordance with (5.75),
H=
d3 x T 00 .
(6.5)
The canonical variables are 2a and 2a = L/ 2a = 1a .

We now quantize the theory by requiring the fields to be operators in Hilbert
space with the canonical commutation relations (2a = 1a )
[2a (x), 1b (y)] = iab (x y),
[2a (x), 2b (y)] = [1a (x), 1b (y)] = 0. (6.6)
Going back to the original Dirac field we get the commutation relations
[a (x), b (y)] = ab (x y),
[a (x), b (y)] = [a (x), b (y)] = 0.
(6.7)
Having derived these relations in a Majorana representation for , they remain

valid under unitary transformations and from now on we shall allow arbitrary
representations, unless otherwise stated.
Of immediate interest are the eigenstates and eigenvalues of the hamiltonian
and of the momentum operator
P =
d3 x (i).
(6.8)
Note that, although and do not commute, there is no ambiguity in the

ordering of these operators since (x y) is an antisymmetric matrix. For
the hamiltonian the order also does not matter because k and are traceless
matrices. We shall start out again with a finite L3 box with periodic boundary
conditions, such that we can freely do partial integration and do not encounter
Dirac delta functions with vanishing argument. We expand the fields in a complete set of eigenfunctions of HD and i, say at time zero,
(x) =
ap =
1
eipx
ap wp ,
L3
p
X
Z
eipx
(x),
d3 x wp
L3
(6.9)
(6.10)
We made
a partial integration to convert all time derivatives to 2 . Otherwise we would
R
get L = d3 x 21 (1 2 2 1 ) H, which would lead to canonical momenta 1 = L/ 1 =
2 /2, 2 = 1 /2, giving conflicting canonical commutation relations like 0 = [1 (x), 2 (y)] =
[1 (x), 21 (y)] = 2i(x y).
129
6.2. CANONICAL QUANTIZATION: RIGHT
with the conjugate expressions obtained by hermitian conjugation. The expansion coefficients have the commutation relations of the creation and annihilation
operators of a harmonic oscillator for each mode,
[ap , ap0 0 0 ] = pp0 0 0 ,
(6.11)
[ap , ap0 0 0 ] = [ap , ap0 0 0 ] = 0.
(6.12)
They are represented in Hilbert space as creation and annihilation operators

indeed, by assuming an empty state |i with the property
ap |i = 0,
(6.13)
and letting the ap act on |i any number of times. The hamiltonian and momentum operator take the following form upon inserting the expansions of (x)
and (x),
H =
Xh
p
P =
ap+ ap+ ap ap Ep ,
ap ap p,
(6.14)
(6.15)
p
where Ep = m2 + p2 is the eigenvalue of the Dirac hamiltonian found in (5.90).

We now see a disastrous problem. By application of arbitrarily many negativeenergy creation operators ap on the empty state |i, we get a state with arbitrarily negative energy: the hamiltonian has no state with lowest energy, it
has no ground state, no vacuum state! This also excludes Statistical Physics in
terms of a canonical partition function Z = Tr eH/T , with positive temperature
T . Lacking a ground state, the theory developed so far cannot have a physical
interpretation and has to be abandoned.
The reason is not the infinite number of degrees of freedom in field theory.
Imposing a cutoff |p| < leads to the same problem (even keeping only a single
momentum mode). The basic reason is the fact that an operator like HD has
eigenvalues coming in pairs E. Even the classical hamiltonian is unbounded
from below if the and are ordinary complex numbers.
6.2
Canonical quantization: right
The problem was found and resolved by Dirac, not in present quantum fieldtheoretic setting, but in a formulation starting from a single particle. We shall
re-phrase his resolution in the present field theory context.
First we note that the states
|p+i = ap+ |i
(6.16)
130
are like particle states with energy Ep = m2 + p2 and momentum p, hence

their mass is m.2 We expect these particles to have spin 1/2. To show this,
consider the angular momentum operator Jk = klm Jlm /2:
J=
d3 x (ix + S)
(6.17)
(there is no operator ordering ambiguity because Tr S = 0). Its commutator

with is given by
[J, (x)] = +ix (x) + (x)S,
(6.18)
from which we infer, using the conjugate of (6.10),

[J, ap ]
d x (x) (ix + S) wp
eipx
.
L3
(6.19)
Since acting on the exponential gives ip is it is simplest here to consider p = 0

and the spinor specification (5.94), which gives
1
[J, a0 ] = 0 a00 .
2
(6.20)
Using this result we calculate

1
J|0+i = [J, a0+ ]|i = 0 |00 +i,
2
(6.21)
and we conclude that the particles have spin 1/2. Aternatively, we can concentrate on the helicity operator p J, use the helicity spinors (5.92) and reach the
same conclusion.
Because the particles have spin 1/2 we expect them to be fermions and the
states to be antisymmetric under interchange of labels,
|p1 1 +, p2 2 +i = |p2 2 +, p1 1 +i.
(6.22)
But this requires ap1 1 + ap2 2 + = ap2 2 + ap1 1 + i.e. the creation operators have
to anticommute instead of commute. We therefore discard the canonical commutation relations and assume instead the so-called canonical anticommutation
relations
{a (x), b (y)} = ab (x y),
{a (x), b (y)} = {a (x), b (y)} = 0,
(6.23)
which imply
{ap , ap0 0 0 } = pp0 0 0 ,
{ap , ap0 0 0 } = {ap , ap0 0 0 } = 0.

2
(6.24)
(6.25)
Strickly speaking, the mass is |m|. We shall assume m > 0, the case m < 0 can be shown
to be equivalent.
131
Let us pause for a moment and consider a single mode p, , . Suppressing the
labels we have
{a, a } = 1, a2 = a2 = 0.
(6.26)
This can be represented as

a|i = 0,
a |i = |1i,
a |1i = 0,
a|1i = |i,
(6.27)
and we see that we have a two-dimensional Hilbert space. The operators are
nilpotent (in accordance with the Pauli principle) and their matrix elements are
given by a00 = a10 = a11 = 0, a01 = 1, and similar for a , or
0 1
0 0
0 0
1 0
(6.28)
Generalizing to N different a and a , we get the tensor product of N two dimensional Hilbert spaces, which has dimension 2N . However, we can re-interpret the
state |1i as the empty state. Then |i = a|1i is re-interpreted as the filled state
and the role of the creation and annihilation operators are interchanged. This
can be made explicit by writing
d = a , d = a,
{d, d } = 1,
d|1i = 0, d |1i = |i.
(6.29)
Consider again the hamiltonian (6.14), this time with the anticommutation
rules for the as and a s. Assuming for the moment a cutoff |p| < on the
number of modes we have a well defined ground state |0i given by
|0i =
Y
p
ap |i,
ap+ |0i = 0,
(6.30)
ap |0i = 0,
H|0i = E0 |0i,
E0 =
(6.31)
Ep .
(6.32)
All the negative energy states are filled: this state is called the Dirac sea. A hole
in the sea, ap |0i, means an excitation of the ground state with positive energy
compared to the ground state. This is interpreted as a different type of particle.
We can represent the situation clearer by a change of notation, writing
bp = ap+ ,
bp = ap+ ,
dp = ap,, ,
{bp , bp0 0 } = pp0 0 ,
dp = ap,, ,
{dp , dp0 0 } = pp0 0 ,
{b, b0 } = {d, d0 } = {b, d0 } = {b, d0 } = . . . = 0.
(6.33)
(6.34)
(6.35)
Then the energy and momentum operators are given by

H =
X
p
P =
X
p
(6.36)
(6.37)
bp bp + dp dp Ep + E0 ,
bp bp + dp dp p.
132
Note the d d order with positive coefficients, for H as well as for P. We see
that bp and dp each create (different) particles with the same energy Ep and
momentum p out of the vacuum |0i, which satisfies
bp |0i = dp |0i = 0.
(6.38)
It is also straightforward to check along the lines leading to (6.21) that the b and
d particles have identical spin properties (thanks to the in (6.34).
As the above expressions for H and P show, the particle states are now
obtained by application of b and d to the ground state. e.g.
|pi = bp |0i, ;
|pi = dp |0i,
|p1 1 , p2 2 i = |p2 1 , p1 1 i =
bp1 1 bp2 2
(6.39)
|0i,
(6.40)
etc. We distinguish the d states with a bar over its labels.

Consider next the Noether charge Q found in (5.79). Its operator form has
an ordering ambiguity which we resolve by (anti)symmetrization:
Q
d3 x

1
a a a a .
2
(6.41)
In terms of the creation and annihilation operators Q is given by

Q=
X
p
bp bp dp dp .
(6.42)
It just counts the number of b-particles minus the number of d-particles, with the
same momentum and spin.
In the context of QED, the Noether current j turns out to be the electromagnetic current and Q is the charge in units of the elementary charge e. The
b- and d-particles have opposite charge. For this reason the d-particles are called
antiparticles (and the bs just particles).
In the zero mass case the operator Q5 (cf. (5.81)) is also conserved. Using
helicity spinors we have
Q5 =
d3 x 5 =
X
p
bp bp dp dp .
(6.43)
So Q5 counts the number right handed ( = +) particles minus the number of

left handed ( = ) particles, plus the number of left handed antiparticles minus
the number of right handed antiparticles.
In terms of the bs and ds the expansion of the Dirac field takes the form
(x) =
X
p
bp wp+
eipx
eipx
+ dp wp,
.
L3
L3
!
(6.44)
133

For infinite volume we use the covariantly normalized spinors defined by
q
u(p, ) =
v(p, ) =
2p0
wp,,+ ,
p =
2p0 wp,, ,
m2 + p 2 ,
(6.45)
(6.46)
and corresponding creation and annihilation operators

q
(6.47)
{b(p, ), b (p0 , 0 )} = 2p0 (2)3 (p p0 ) 0 ,
(6.48)
etc., such that,
2p0 L3 bp ,
2p0 L3 dp ,
b(p, ) =
d(p, ) =
etc. Including time dependence we have for the free field3

(x) =
XZ
(x)
=
XZ
(6.49)
(6.50)
dp b(p, ) u(p, ) eipx + d (p, ) v(p, ) eipx ,

dp b (p, ) u
(p, ) eipx + d(p, ) v(p, ) eipx .
Subtracting the vacuum energy, we get for the energy-momentum operators

P =
XZ
dp b (p, )b(p, ) + d (p, )d(p, ) p ,
(6.51)
and the charge

Q=
XZ
dp b (p, )b(p, ) d (p, )d(p, ) .
(6.52)
The anticommutation relations have led to a sound physical interpretation of

the Dirac field. Only if the particles described by the field are fermions did we
get a satisfactory situation. Another approach to the relation between spin and
statistics, that stresses the importance of locality, is described in the appendix 6.8,
whereas in the next section we also get the same results from the path integral.
It is instructive to rewrite these results
in terms of hermitian Majorana fields
1,2 related to by = (1 i2 )/ 2 (using real ). We shall be very brief.

A Dirac field is simply equivalent to two Majorana fields. For a single Majorana
field = , and we note:
v(p, ) = u (p, ),
(x) =
XZ
j5
3
(cf. (5.97,5.102)),
h
(6.53)
i
dp b(p, ) u(p, ) eipx + b (p, ) u (p, ) eipx , (6.54)
= 0,
1 T
i 5 6= 0.
=
2
(6.55)
(6.56)
Because of the analogy with the plane wave expansion of the free electromagnetic field the
us and vs are called polarization spinors.
134
We end this section with some comments. Dirac formulated his equation for
a Schrodinger quantum-mechanical wave function of a single electron: his was
not a classical field. Interpreting in eq. (5.85) as a Schrodinger wave function
means conceptually something very different from a field equation, and it led
to several difficulties, one of which was how to interpret the negative energy
solutions. Diracs solution in postulating a filled sea of fermionic states led to
the prediction of antiparticles as holes in the sea. However, this gave additional
difficulties since it brought the theory outside the single-particle framework. The
quantum field theory version does not have these difficulties. Since it looked to
the field theory pioneers like a quantization of a single particle Schrodinger wave
function , which was already supposed to describe a quantum theory, the name
second quantization got into use as a name for field theory.
We like to mention again that antiparticles are not predicted by mere relativistic field theory. There is no reason for the doubling of degrees of freedom
embodied in the Dirac field, a single Majorana field is the simplest possibility and
(as for a single scalar field) it implies no antiparticles. However, we shall see later
in chapter 8 that the coupling to the electromagnetic field does require the Dirac
field (or two Majorana fields), and hence predicts the existence of antiparticles.
But in non-relativistic quantum electrodynamics antiparticles are not needed, as
we review in section 6.4.
6.3
Path integral quantization
We shall now follow an independent approach to the quantization of spinor fields,

using the notion of a path integral. As mentioned below (5.82), the action vanishes identically for a single real-valued Majorana field. So if a path integral with
integrand exp(iS) is going to make sense, the Majorana field a (x) cannot consist
of ordinary real numbers. For the action not to vanish, the variable appearing
in the path integral has to be anticommuting:
a (x)b (y) = b (y)a (x),
[a (x)]2 = 0.
(6.57)
This is what we going to assume from now on. We enlarge the system of complex numbers with such anticommuting objects. They are called Grassmann
variables, somewhat inappropriately, because they do not vary at all like ordinary variables. Instead they can be given meaning as generators of a Grassmann
algebra. They are also called anticommuting numbers, also somewhat inappropriately, because they are not numbers. We shall gradually develop a reasonable
mathematical framework for them. To start with, we assume the rule
[a (x)b (y)] = b (y)a (x)
(6.58)
which we used in (5.63) to get a real Lagrange density. For ordinary complex
numbers the above relation is of course trivial, but for noncommuting objects we
135
6.3. PATH INTEGRAL QUANTIZATION
have to define it. The rule above is similar to the operator relation (AB) = B A .
We shall also assume that the anticommuting numbers commute with ordirary
numbers, and hence also with bosonic field operators, while they anticommute
with fermionic operators:4
a (x)b (y) = b (y)a (x),
(6.59)
As a first consequence we see that the Dirac

field is equivalent to
two Majorana
fields indeed, since, writing = (1 i2 )/ 2, = (1 + i2 )/ 2, and using

(6.58), the Dirac action is just the sum of two identical Majorana actions:
S =
Z
+ )
d4 x (m
1
1
d x 1T (m + )1 + 2T (m + )2
2
2
4
(6.60)
and similar for the hamiltonian. Note that this is very different from (6.2,6.3).
It would now be logically clearest to continue with the case of a single Majorana
field, but for conciseness we shall instead go directly to the doubled case of the
Dirac field.
We proceed heuristically and assume the following relations by analogy to the
scalar field,
Z[, ] =
[dd]
eiS[,]+i
d4 x (
+)
R 4
Z[, ]

= h0|T ei d x ( +) |0i,
Z[0, 0]
(6.61)
(6.62)
in which S is the Dirac action, and are anticommuting external sources, and
R
|0i is the ground state (vacuum). The meaning of the symbols [dd]
has to
be specified further. For now we make the crucial assumption that the measure
[dd]
is translation invariant, such that
Z
=
[dd]F
[, ]
[dd]F
[ + , + ],
(6.63)
for any anticommuting , and functional F . Then, using the shorthand notation5
Z
4
+ ) G
1 ,
d4 x (m
G1 (x, y) = (m + ) 4 (x y),
(6.64)
(6.65)
In this section (and whenever needed for clarity) we denote operators in Hilbert space by
a caret. So b (x) denotes the quantized fermion-operator field and a (x) denotes an anticommuting number.
5
Notice that we are freely making partial integrations without picking up boundary terms,
implying appropriate boundary conditions in space and time (as in the bosonic case).
136
and making the transformation of variables

= 0 + G,
such that
we have the result
= 0 + G,
G1 G = 1,
(6.66)
1 + +
= 0 G1 0 + G,
G
Z[, ] = Z[0, 0] eiG = Z[0, 0] ei
d4 x d4 y (x)G(x,y)(y)
(6.67)
(6.68)
Comparing coefficients of multinomials in and we will get the Wick formula

for fermions, relating vacuum expectation values of time ordered products of field
operators to sums of products of Green functions G. From the properties of the
latter we can then infer once more the interpretation in terms of fermions.
Consider the first few terms in and ,
Z[, ]
= 1+i
Z[0, 0]
d4 x d4 y (x)G(x, y)(y) +
= 1 + ih0|
i2
+ h0|
2Z
+ i2
d4 x (x)(x)|0i
+ ih0|
d4 x (x)(x)|0i
i2
d4 x (x)(x)
|0i + h0|
2
Z
4
(y)(y)|0i
d x d4 y h0|T (x)(x)
+ ,
Z
d x (x)(x)
4
2
Z
2
|0i
(6.69)
in first instance together since, as

where we kept the combinations and
a product of two anticommuting objects, they are of the commuting type. Note
can be freely interchanged
that this implies that even products such as and
(y)(y)
under the time-ordering operator, i.e. T (x)(x)

= T (y)(y)
(x)(x),
as for the scalar-field case. Comparing coefficients of , we conclude that
= 0.
= h0||0i
h0||0i
(6.70)
Comparing the bilinear (x)(y) terms we have to be careful about signs: for
x0 > y 0 ,
(y) (y) = (x) (x)
(y) (y) = (x) (y) (x)
(y) (6.71)
T a (x)a (x)
b
a
a
b
a
b
a
b
b
b
whereas for x0 < y 0 ,
(y) (y) =
(y) (y) (x) (x) =
(y) (x).
T a (x)a (x)
a (x)b (y)
b
b
a
a
a
b
b
b
(6.72)
Since the product a (x)b (y) is commuting and can be taken in front of the T
symbol we conclude that for fermion fields
(y) = (x)
(y), x0 > y 0
T a (x)
a
b
b
= b (y)a (x), x0 < y 0 .
(6.73)
137

Comparison of the bilinear terms in , gives
(y)|0i = iG (x, y).
h0|T a (x)
ab
b
(6.74)
The generalization to more fields is given by, in shorthand notation,

h0|T 1
1 2 2 n n |0i h1 1 2 2 n n i
R
[dd]
eiS 1 1 n n
R
=
[dd]
eiS
= (i)n
(1) G1 1 Gn n ,
(6.75)
(6.76)
(6.77)
where
G1 1 = Ga1 b1 (x1 , y1 ),
etc.,
(6.78)
and the summation is over all permutations of 1, . . . , n, with signature (1) .

Instead of comparing coefficients of , we can also use repeated differentation
with respect and ; see appendix 6.9 for fermionic differentiation rules. So in
general the fermion field operators behave as anticommuting inside a time ordered
product, and in a given time ordering we get a sign that is equal to the signature
of the permutation needed to order the operators according to decreasing times
from left to right.
To get a nonvanishing result there have to be an equal number of s and
s. This is a consequence of the U(1) phase-symmetry (5.76). Making the

transformation of variables
= ei 0 ,
= ei 0
(6.79)
and assuming the fermionic integration measure [dd]

to be invariant it follows
that
h1 2 m 1 n i = ei(mn) h1 2 m 1 n i,
(6.80)
so the result vanishes if m 6= n. For a single Majorana field this U(1) symmetry is
absent and the analogue of (6.77) would be similar to the scalar-field case, except
for signs.
Next we study the fermion Green function G, which has to satisfy G1 G = 1,
or
(m + ) G(x, y) = 4 (x y).
(6.81)
Writing G as a Fourier transform
G(x, y) =
d4 p ip(xy)
e
G(p),
(2)4
(6.82)
we get the equation

(m + ip) G(p) = 1.
(6.83)
138
Using (cf. (5.84))

(m + ip)(m ip) = m2 + (p)2 = m2 + p2 ,
(6.84)
we see that
m ip
,
(6.85)
m2 + p2 i
is a possible solution, where we have used the same i prescription for dealing
with the poles at p2 = m2 as for the scalar field. We shall see shortly that this
leads to consistent expressions. Thus the spinor field Green function is a linear
combination of the scalar field propagator and its derivatives,
G(p) =
d4 p ip(xy) m ip
e
(2)4
m2 + p2 i
Z
d4 p ip(xy)
1
= (m )
e
.
(2)4
m2 + p2 i
Z
G(x, y) =
(6.86)
(6.87)
We shall now show that the results above that followed from the path-integral
approach to quantization lead to the same particle interpretation as the canonical
quantization in the previous section. Choosing x0 > y 0 we have
(y)|0i
h0|(x)
= (m )
=
=
dp eip(xy)
dp eip(xy) (m ip),
XZ
dp eip(xy) u(p, )
u(p, ),
(6.88)
(6.89)
(6.90)
where we used the i prescription in reducing the four-dimensional Fourier integral

to a three-dimensional one, and also (5.179) in Problem 5.1. We insert a complete
set of states between the field operators on the left-hand side, assuming this
can be written in terms of a zero-particle (vacuum) contribution, one-particle
contributions, two-particle contributions, etc. (cf. (2.158):
- The vacuum does not contribute because h0||0i

= h0||0i
= 0, as we have
learned from the coefficients linear in , in (6.69);
- The one particle contribution is
XZ
dp h0|(x)|pihp|
(y)|0i.
(6.91)
Using translation invariance

(x)
= eiP x (0)
e iP x ,
eiP y ,
(y)
= eiP y (0)
(6.92)
this can be written as

XZ
dp eip(xy) h0|(0)|pihp|
(0)|0i,
(6.93)
139
which has the same exponentials as (6.90).

-The multiple particle contributions would give exponentials exp[ip(x y)] (by
the same translation invariance argument), with, since p is in this case the total
four-momentum of a multiparticle state, p2 4m2 . Hence, these contributions
have to zero because the time-dependence cannot fit that of (6.90).
We conclude that only one particle states contribute, with
h0|(x)|pi
= u(p, ) eipx,
hp, |(y)|0i
= u
(p, ) eipy .
(6.94)
(6.95)
Repeating the reasoning for x0 < y 0 gives
=
(y) (x)|0i =
= h0|
a
b
dp eip(xy)
(6.96)
dp eip(xy) (m + ip)ab
(6.97)
iGab (x, y) = (m )ab

XZ
dp eip(xy) va (p, )
vb (p, )
(6.98)
XZ
,
dp h0|b (y)|p, ihp, |a (x)|0i + . . .(6.99)
where we assumed that there are different particles (indicated by the bar) with
identical momentum and spin properties, because of the Dirac field being complex. Comparing the first and last lines we conclude that again the multiple
particle contributions vanish and that
= v(p, ) eipy ,
h0|(y)|pi
hp, |(x)|0i
= v(p, ) eipx.
(6.100)
(6.101)
The i specification of the fermion Green function is evidently the correct one for
being able to match the left- and right-hand side of (6.74).
Taking limits of equal time, we easily derive the vacuum expectation value
of canonical anticommutation relations between the field operators (see also appendix 6.8),
h0|{a (x, t), b (y, t)}|0i = ab (x y).
(6.102)
These can presumably be extended to operator relations (i.e. holding true for arbitrary matrix elements) by the Wick formula (6.77), thus reconstructing fermionic
Fock space. Note that we could easily have got a minus sign on the right hand
side above, if G1 had the opposite sign. This would be unacceptable as it would
imply a negative metric in state vector space (it would not be a Hilbert space),
leading to negative probabilities. Obtaining the correct sign is the reason for
the sign choice L = +i 0 + in the lagrange density (5.65).
140
So the path integral formula for the vacuum expectation value of time-ordered
products (6.62), together with some assumed basic properties of fermionic integration, provides an alternative formulation of the theory. The exposition so far
is sufficient for us to proceed with perturbation theory for interacting theories
with fermions. Appendix 6.9 gives further results on anticommuting numbers
and fermionic integration.
Let us sketch the evaluation of the remaining Z[0, 0]. Expanding the fields in
terms of eigenfunction of the hermitian operator G1 , where G1 is the Dirac
operator m + ,
G
f = f ,
(x) =
d4 x f (x)f (x) = ,
f (x),
(x) =
f (x),
(6.103)
(6.104)
brings the action into the form

S=
(6.105)
The transformation of variables (x) is formally unitary, so its determinant
is a phase factor which is expected to cancel in the combination dd,

and using
also det = 1 we have formally
[d d] =
da (x) da (x) =
da (x) da (x) =
d d ,
(6.106)
xa
xa
Then, with the rules derived in appendix 6.9,

Z
[d d] e
iS
Z "Y
d ei
det[G1 ] = det G1 .
(6.107)
(6.108)
Note that det G1 is in the numerator, and not in the denominator as for Bose
fields. There is no square root because the Dirac field is complex. This formula
has been checked (e.g. for the partition function) also by other means.
We warn the reader that in case of fermions the formal character of the above
derivations is known to easily give rise to problems6 , especially with gauge-field
interactions involving 5 , and careful regularizations of the otherwise ambigouous
infinities are all the more important.
We have not yet given a thorough definition of the time-evolution operator
U (t1 , t2 ) in terms of the fermionic path integral, using spacetime discretization.

This turns out to be remarkably involved, because of the so-called fermion doubling phenomenon. Problems arise with the implementation of chiral symmetries
6
One then speaks of anomalies, which turn out to have important physical implications.
6.4. NON-RELATIVISTIC FIELDS AND ANTIPARTICLES
141
like (5.80). Because of the importance for a proper mathematical definition of

quantum field theory as well as for applications to nonperturbative computations, much effort has been spent on solving these problems, which involve deep
properties of gauge theories and fermions, and research into this area of lattice
fermions is still in progress. For an introduction, see the books on lattice field
theory.
6.4
Non-relativistic fields and antiparticles
In the non-relativistic regime it is better to use a reduced description that measures kinetic energies of the particles relative to their energy at rest: E = p2 /2m.
For example, this is customary in the field-theoretic formulation of condensedmatter physics.
For free fields the reduction can be conveniently done in terms of the explicit
form for the generating functional for the n-point functions. Consider the case
of a single Majorana field,

Z

Z
1 T
4
Z[] =
[d] exp i d x (m + ) + , (6.109)
2
Z[0] exp {iW []} ,
(6.110)
Z
1
W [] =
d4 x d4 y (x)T G(x, y)(y)
2
Z
d4 p
1
m i p
T
(p),
(6.111)
(p) 2
=
2 (2)4
m + p2 i
where
(p) = (p) =
d4 x eipx (x).
(6.112)
The sources can be thought of as mimicking the effects of other particles in

an interacting theory. In the non-relativistic regime (p) is non-zero only for
frequencies near the rest-energy, p0 m, and momenta such that the nonrelativistic kinetic energy p2 /2m m. This means that in the integral in (6.111),
only values p0 m and momenta |p| m contribute. To implement this, let
us first rewrite (6.111) using
(m i p )
(p)
m2 + p2 i
Z
Z
(m i p )
=
dp0 d3 p (p)T 2
(p),
m + p2 i
0
Z
dp0
d3 p (p)T
(6.113)
where we changed variables p p , interchanged the sources and used the

symmetry properties T = , ( )T = that hold in a Majorana representation. Using the above identity (6.111) can be written as
W [] =
dp0 Z d3 p
m i p
(p).
(p)
2
(2)3
m2 + p2 i
(6.114)
142
Measuring frequencies relative to the rest energy is expressed by writing

p0 = m + p00 .
(6.115)
The non-relativistic reduction in the denominator in (6.114) is then implemented

by
m2 + p2 i = m2 + p2 (m + p00 )2 i p2 2mp00 i,
(6.116)
where we neglected (p00 )2 relative to m2 . Note that we also assumed p00 to be of

the same order as the kinetic energy p2 /2m. For the numerator we get
m i p = m + i 0 p0 i k pk m(1 + i 0 ) = 2m(1 + )/2,
(6.117)
where we neglected p 0 and |p| compared to m. The result of this approximation

is
W []
dp00
2
d3 p
(1 + )/2
00
(m+p
,
p)
(m+p00 , p). (6.118)
(2)3
p2 /2m (p00 + i)
Now
1+
P+ = P+2
(6.119)
2
is a projector, it projects onto the subspace of eigenvalue +1 of . This means
that only two of the four Dirac-components of contribute in the non-relativistic
form. To make this explicit, we note that (p) and (p) are for p0 > 0 not
constraint by the reality condition (p) = (p). So we can consider the real
and imaginary parts of (p), or equivalently, (p) and (p) , to be independent
Grassmann variables in (6.118), and we can change the representation of the
Dirac matrices by a unitary transformation,
(p) U (p), (p) (p) U,
(6.120)
to one in which is diagonal, the so-called Dirac representation:

U U = 3 ,
U k U = 1 k ,
Dirac representation
(6.121)
1
0
0
0
(6.122)
In this representation
P+ =
1 + 3
=
0
1
0
0
0
0
0
0
0
0
0
0
and the two-component nature of P+ (p) and (p) P+ is manifest. We can clarify
the situation further by introducing two-component sources and ,
a (p00 , p) = a (m + p00 , p),
a (p00 , p) = a (m + p00 , p) ,
a = 1, 2.
(6.123)
6.4. NON-RELATIVISTIC FIELDS AND ANTIPARTICLES
143
We can also extend the lower limit of integral over p00 in (6.118) from m to
, since m is already out of the regime were the sources are practically nonvanishing. Dropping now the prime on the dummy-integration variable p00 , we
can rewrite (6.118) in the form
W [, ] =
=
d4 p
1
a (p) 2
a (p)
4
(2)
p /2m (p0 + i)
d4 x
d4 y a (x)GR
nr (x, y)a (y),
(6.124)
(6.125)
where
(x) =
d4 p ipx
e (p),
(2)4
and
GR
nr (x, y)
(x) =
d4 p ipx
e
(p),
(2)4
d4 p
eip(xy)
(2)4 p2 /2m (p0 + i)
(6.126)
(6.127)
is a the retarded Green function of the differential operator i0 2 /2m. It is

not difficult to find the path integral that reproduces the non-relativistic n-point
functions specified by (6.125),
=
Z[, ]
Snr =
[d d ] eiSnr +i

d4 x a i0 a
d4 x (a a +a a )
(6.128)
1
k a k a ,
2m

(6.129)
where a , a = 1, 2, is a two-component non-relativistic spinor field. The index a

is the spin index on which the Pauli matrices k can act.
The above action is the same as that of the scalar-field version (1.63), apart
form the doubling by the spin index a and the fact that here we are dealing with
Grassmann variables. We end this section with the following observations:
- It is straightforward to derive, from the n-point functions obtained from
the generating functional Z, the (anti)commutation relations of the field
operators in the operator version of the theory (x0 = y 0 ):
[(x),
(y)]
[(x),
(y)]
{a (x), b (y)}
{a (x), b (y)}
= (x y),
= [ (x), (y)] = 0, Bose,
= ab (x y),
= {a (x), b (y)} = 0, Fermi.
(6.130)
(6.131)
- The fact that a retarded Green function appears in the expression for twopoint function implies that annihilates the ground state (unlike the relativistic field):
h0|T a (x)b (y)|0i = iGR

nr (x, y) h0|b (y)a (x)|0i = 0.
(6.132)
144

The stronger statement a (x)|0i = 0 follows from taking all n-point functions into consideration. It follows from the commutation relations that
(x) acts like a creation operator; it creates a particle at position x.
- The non-relativistic field is complex, in contrast to the real Majorana field.
The action (6.129) has an extra U (1) symmetry that the Majorana action
in (6.109) does not possess:
(x) ei (x),
(x) ei (x).
(6.133)
The corresponding Noether-current is given by

j 0 = ,
jk =
1
( ik + ik ),
2m
(6.134)
and the conserved charge simply counts the number of particles. This
complex-number property of non-relativistic fields is just right for representations of the Gallilei group, the group of translations, rotations and
non-relativistic boosts; see e.g. Weinberg I.
The action (6.129) even has an SU (2) symmetry corresponding to transformations on the spin index, independent of spatial rotations: a (x) ab b ,
a (x) b (x)ba , SU (2). This symmetry is an artifact of the free
field and it is usually broken by interactions. For example, a magnetic moment interaction (x)k (x)Bk (x) to an external magnetic field B(x)
is invariant under spatial rotations that affect both the spin index a and
the spatial index x in the usual way.
- Suppose the non-relativistic field describes electrons. As a consequence of
the field being complex, we can couple it gauge-invariantly to the electromagnetic field by the minimal substitution rule + ieA , adding
also a magnetic-moment coupling,
S[, , A] =
d x (i0 eA0 )
ge
k B k ,
4m
1
(k ieAk ) (k + ieAk )
2m
(6.135)
where g is the Lande g factor (gyromagnetic ratio).

Such a procedure is not possible for the real Majorana field. Instead (cf.
chapter 8), the complex Dirac field is needed to be able to formulate a gauge
invariant coupling to the Maxwell field. The relativistic theory also gives a
precise value for the coupling constant g, which is a free parameter in the
non-relativistic theory.
As we have seen, the Dirac field describes two kinds of particles with identical masses and opposite charges it turns out that the eigenvalue 1 of
the Noether charge is indeed the charge of a particle.
145
6.5. YUKAWA MODELS
Thus, whereas antiparticles are not needed in non-relativistic electrodynamics, the relativistic quantum field theory of charged particles predicts
the existence of antiparticles.
For more information on non-relativistic quantum fields see, e.g. the book by
Brown, and specialized books on application to condensed matter physics.
6.5
Yukawa models
As an example for illustration of perturbative methods we shall use a so-called

Yukawa model described by the action7
S=
d4 x
1
1
+ M
+ g
+ m2 2 +
+ S, (6.136)
2
2

where either = 1 or = i5 , and S denotes the counterterms . For = i5

this action is still parity invariant, but with a minus sign in the transformation
of the scalar field:
1 x) D 1 , 0 (x) = (`1 x),
0 (x) = DP (`1
0 (x) = (`
P x),
P
P
P
1 5 0 =
5 (DP = 0 ).
0 5 0 0 =
(6.137)
0
This phase factor in the transformation rule for (+ for = 1, for = i5 )
is called the intrinsic parity of the scalar field. In case of the plus sign the field
and the corresponding particles are called scalars, in case of the minus sign
and as giving a
pseudoscalars. In the pseudoscalar case we can think of , ,
crude description of the interactions of protons (p), antiprotons (
p) and neutral
pions ( 0 ).
Writing
S = S0 + S1 ,
S1 =
(6.138)
d4 x g
+ S,
(6.139)
the perturbative expansion of h1 2 3 i leads to expressions which are again

well represented by diagrams. To distinguish the bosonic and fermionic propagators we shall change the notation of the latter to S:
Sab (x, y) =
d4 p ip(xy)
e
Sab (p),
(2)4
Sab (p) =
M ab ip ab
.
M 2 + p2 i
(6.140)
The fermion propagator Sab (x, y) is not a symmetric function of its arguments,
Sab (x, y) 6= Sba (y, x) or Sab (p) 6= Sba (p), and to be able to distinguish its arguments we draw an arrow in its line in the diagrams, as in Fig. 6.1: the arrow
146
= -iS ab (p)
= -ig ab
Figure 6.1: Fermion propagator Sab (p) = (M ip)ab /(M 2 + p2 i) and bare
vertex function igab ( = 1 or i5 ) of the Yukawa models. The fermion line
is drawn, the scalar line is dashed. The arrow in the fermion line points to the
row-index a.
+...
Figure 6.2: First two nonzero terms in the perturbative expansion of h(x1 )(x2 )i
in a Yukawa model (without bare 4 interaction).
points in the direction of the row index of S. The interaction S1 leads to a bare
vertex function in position space
Sa b (u, v, w) = gab
d4 x (x u) (x v) (x w),
(6.141)
or in momentum space
Sa b (p, q, k) = gab ,
(6.142)
as illustrated in Fig. 6.1. The calculation of n-point functions goes along the
same lines as in scalar field theory. We write
h1 2 3 i =
h1 2 3 eiS1 i0
,
heiS1 i0
(6.143)
with h i0 the free field average, and expand the exponential. Disconnected
vacuum diagrams again cancel between numerator and denominator. For example
the first few diagrams for the scalar two point function are given in Fig. 6.2. In
position space these represent
(ig)2 Z 4 4
d x d y h1 2 a (x)ab b (x)(x)
2!
c (y)cd d (y)(y)iconn
+
0
h1 2 i = h1 2 i0 +
7
There is no special reason why we switched using the notation in stead of .
(6.144)
6.6. SCATTERING IN THE TREE GRAPH APPROXIMATION
147
(ig)2
= iG(x1 , x2 ) 2
d4 x d4 y (i)4 G(x1 , x)G(x2 , y)
2!
ab Sbc (x, y)cd Sda (y, x) + ,
(6.145)
Z
where the factor 2 comes from similar contractions with x y which give an identical contribution, and the minus sign in front of it comes from putting the fermion
fields in standard order as for the propagator (a b c d = b c d a ). Note
that such minus signs are typical for closed fermion loops. In momentum space
we get
G0 (p) = G(p) G(p)(p)G(p) +
(6.146)
Z
4
dq
Tr (ig)(i)S(q)(ig)(i)S(p + q). (6.147)
i(p) =
(2)4
6.6
Scattering in the tree graph approximation
The calculation of scattering amplitudes and cross sections will here be illustrated
in the Yukawa model. We concentrate on the leading contributions which are
represented by tree graphs. First we derive some convenient formulas involving
creation and annihilation operators.
We define for the scalar field
()
(p, x ) =
d3 x eipx (i 0 i 0 )(x),
p0 =
m2 + p 2 ,
(6.148)
(6.149)
Recalling the expansion

free (x) =
dp a(p) eipx + a (p) eipx
of the fee scalar field, it is easy to check that if (x) satisfies the free Klein-Gordon
equation with mass m, then () (p, x0 ) are actually time independent and equal
to the usual covariantly normalized annihilation and creation operators:
(+) (p, x0 ) a(p, x0 ) = a(p) if (m2 2 ) = 0,
() (p, x0 ) a (p, x0 ) = a (p) if (m2 2 ) = 0.
(6.150)
(6.151)
The notation () (p, x0 ) stresses that we are just taking linear combinations of
the otherwise arbitray scalar field.8 Because they are linear combinations we can
make contractions with them just as for the fields themselves. For instance,
h(+) (p, x0 )(y)i0 =
8
d3 x eipx (i 0 i 0 )(i)G(x, y),
(6.152)
For the free case the may be thought of indicating the positive and negative frequency
components of the field.
148
where the subscript 0 on the left-hand side denotes the free case as usual. For
clarity in applications we shall use the a(p, x0 ) notation in the following. Using
the above expression in terms of G(x, y), or the interpretation as creation or
annihilation operators, we now have the convenient formulas9
ha(p, ) (x)i0
ha(p, ) (x)i0
ha (p, ) (x)i0
ha (p, ) (x)i0
=
=
=
=
h0|a(p) (x)|0i0 = hp|(x)|0i0 = eipx ,

h0|(x)a(p)|0i0 = 0,
h0|a (p)(x)|0i0 = 0,
h0|(x) a (p)|0i0 = h0|(x)|pi0 = eipx .
(6.153)
(6.154)
(6.155)
(6.156)
Here means that the time has been sent to .

Similarly, we define for the fermion field
b(p, , x0 ) = (+) (p, , x0 ) =
b (p, , x0 ) = () (p, , x0 ) =
d (p, , x0 ) = () (p, , x0 ) =
d(p, , x0 ) = (+) (p, , x0 ) =
where
p0 =
Recalling the expansions
free (x) =
XZ
free (x) =
d3 x eipx u
(p, ) i 0 (x),
(6.157)
i 0 u(p, ) eipx,
d3 x (x)
(6.158)
d3 x eipx v(p, ) i 0 (x),
(6.159)
i 0 v(p, ) eipx,
d3 x (x)
(6.160)
M 2 + p2 .
(6.161)
dp b(p, ) u(p, ) eipx + d (p, ) v(p, ) eipx , (6.162)
XZ
dp b (p, ) u
(p, ) eipx + d(p, ) v(p, ) eipx , (6.163)
we see that if the fermion field satisfies the free Dirac equation with mass M ,
then b(p, , x0 ), . . . , d (p, , x0 ) are actually time-independent and equal to the
annihilation, . . . , creation operators of particles, . . . , antiparticles. We have in
the free case
hb(p, ; x )(y)i
0 =
0
d3 x eipx u(p, ) i 0 (i)S(x, y),
(6.164)
etc., and using such expressions, or the operator versions, the following formulas
can be derived which are convenient for later use:
hb(p, , ) (x)i
(p, ) eipx ,
0 = h0|b(p, ) (x)|0i0 = u
hb(p, , ) (x)i0 = 0,
9
(6.165)
(6.166)
Inside h i we write a in stead of a because in the path integral these are not operators.
149

h(x) b (p, , )i0
b (p, , )i0
h(x)
hd(p, , ) (x)i0
hd(p, , ) (x)i
0
h(x) d (p, , )i0

h(x) d (p, , )i0
h0|(x) b (p, )|0i0 = u(p, ) eipx ,

0,
h0|d(p, ) (x)|0i0 = v(p, ) eipx,
0,
d (p, )|0i0 = v(p, ) eipx ,
h0|(x)
0,
=
=
=
=
=
=
(6.167)
(6.168)
(6.169)
(6.170)
(6.171)
(6.172)
with the remaining time-specifications vanishing.

These formulas will now be used for the calculation of scattering amplitudes.
The formulation and calculation of scattering processes in field theory will be
discussed in more detail in chapter 9, but in the tree graph approximation the
following reasoning is plausibly correct. Consider the proton-pion scattering process (p, ) + k (p0 , 0 ) + k 0 . The probability amplitude hp0 0 , k 0 out|p, k ini for
finding the outgoing state in the incoming state can be calculated as
hp0 0 , k 0 out|p, k ini = ha(k 0 , t0 )b(p0 0 , t0 )b (p, , t)a (k, t)i
ha(k 0 , t0 )b(p0 0 , t0 )b (p, , t)a (k, t)eiS1 i0
=
, (6.173)
heiS1 i0
where t , t0 + and h i0 denotes the free field average
h i0 =
as in (4.10). Expanding
eiS1 = 1 + +
(ig)2
2!
[dd][d]
eiS0
R
,
[dd][d]
eiS0
(6.174)
0 )(x0 )(x0 ) + (6.175)

d4 x d4 x0 (x)(x)(x)
(x
we encounter (assuming the final state to be different from the initial state, such
that the disconnected contributions leading to hp0 0 |pihk 0 |ki vanish):
0 )(x0 )(x0 ) b (p, t)a (k, t)i0

ha(k 0 , t0 )b(p0 0 , t0 )(x)(x)(x)
(x
h
i
0
0
0
= hb(p0 0 , )(x)i
h(x)
(x
)i
h(x
)b
(p,
)i
+
x
x
0
0
0
[ha(k 0 , )(x)i0 h(x0 )a (k, )i0 + x x0 ]

+ contractions leading to loop diagrams
0
0
0
0
0 0
= eip x u
(p0 , 0 )(i)S(x, x0 )u(p, )eipx [eik x eikx + eik x eikx ]
+ x x0 + contractions leading to loop diagrams.
(6.176)
Note the condensed notation in which the summation over Dirac indices is suppressed. For example, we have more explicitly
0
0
hb(p0 0 , )(x)i
(6.177)
0 h(x)(x )i0 h(x )b (p, )i0
0 0
0
0
= hb(p , )a (x)i0 ab hb (x)c (x )i0 cd hd (x )b (p, )i0 ,
150
k
+
p
,
p
,
Figure 6.3: Tree diagrams for fermion-(pseudo)scalar scattering in the Yukawa

model; q = p + k, r = p k 0 .
with the usual summation convention for a, b, c, d. Substituting
S(x, x0 ) =
d4 s is(xx0 ) M is
e
(2)4
M 2 + s2
(6.178)
and integrating over x and x0 we encounter

4 (p0 k 0 + s) 4 (p + k s) = 4 (p + k p0 k 0 ) 4 (p + k s)
(6.179)
for the first contribution and

4 (p0 + k + s) 4 (p k 0 s) = 4 (p + k p0 k 0 ) 4 (p k 0 s),
(6.180)
for the second. The x x0 terms give the same contribution and we find the
form
hp0 0 , k 0 out|p, k ini = i(2)4 4 (p0 + k 0 p k) T (p0 , 0 , k 0 ; p, , k),
(6.181)
with the result for the scattering amplitude (invariant amplitude) T :

M i(p + k) M i(p k 0 )
+ 2
(ig)u(p, ), (6.182)
T =u
(p , )(ig)
M 2 + (p + k)2
M + (p k 0 )2
0
"
The two diagrams corresponding to this process are given in Fig. 6.3.
Defining
1X 2
|T |2 =
|T |
4 0
(6.183)
as the average of |T |2 over initial and final spins, the differential cross section
takes the same form as for spinless particles
"
d
d
=
cm
1
2|T |2 ,
64 2 s
(6.184)
where the factor 2 comes from the summation over final spin states. The further
calculation of |T |2 (which is a Lorentz-invariant function of the momenta) will
not be done here (similar calculations will be carried out later in QED).
151
However, we like to draw the attention to the angular dependence coming

from the
denominator
of the second fermion propagator: since M 2 + (p k 0 )2 =
m2 + 2 M 2 + p2 m2 + p2 + 2|p|2 cos in the center of mass frame ( being

the angle between p and p0 ), we get a backward peak at high energies, where the
masses may be neglected:
M2
1
1
,
0
2
2
+ (p k )
|p| ( )2
(6.185)
Consider next antifermion-(pseudo)scalar scattering. Here (6.173) gets replaced by

ha(k 0 , )d(p0 0 , )d (p, , )a (k, )eiS1 i0
.
heiS1 i0
(6.186)
To make the contractions slightly more transparant we now first put the expectation value in standard order (recall d and d ):
hp0 0 , k 0 out|p, k ini =
ha(k 0 , )d(p0 0 , )d (p, , )a (k, )eiS1 i0

= ha(k 0 , )d (p, , )d(p0 , 0 , )a (k, )eiS1 i0 .
(6.187)
We now encounter
0 )(x0 )(x0 ) d(p0 0 , )a (k, )i0

ha(k 0 , )d (p, )(x)(x)(x)
(x
0
0 0
0
= hd (p, )(x)i
0 h(x)(x )i0 h(x )d(p , )i0
[ha(k 0 , )(x)i0 h(x0 )a (k, )i0
+ ha(k 0 , )(x0 )i0 h(x)a (k, )i0 ]]
+ x x0
+ contractions leading to loop diagrams
0 0
0
0
0 0
= eipx v(p, )(i)S(x, x0 )v(p0 , 0 )eip x [eik x eikx + eik x eikx ]
+ x x0 + contractions leading to loop diagrams.
(6.188)
The integrations over x and x0 now produce

4 (p k 0 + s) 4 (p0 + k s) = 4 (p + k p0 k 0 ) 4 (p k 0 + s)
(6.189)
for the first contribution and

4 (p + k + s) 4 (p0 k 0 s) = 4 (p + k p0 k 0 ) 4 (p + k + s),
(6.190)
for the second, which leads to

M + i(p k 0 )
M + i(p + k)
T =
v (p, )(ig)
(ig)v(p0 , 0 ),
+ 2
2
0
2
2
M + (p k )
M + (p + k)
(6.191)
"
152
k
+
p
,
p
,
Figure 6.4: Tree diagrams for antifermion-(pseudo)scalar scattering; q = p + k,

r = p k0.
with the identification s = k 0 p for the first contribution and s = p k for
the second. We see an overall minus sign, and v and v describe the initial state
and final state, respectively. The diagrams for this process are shown in Fig. 6.4.
Notice that the momenta q and r flow in this case against the direction of the
arrow in the fermion line.
6.7
Summary
Canonical quantization of the Dirac field is not straightforward. Physical requirements (the existence of a ground state, locality, positivity of Hilbert space metric)
led to anticommutation relations for the field operators. These can be realized
in Hilbert space with creation and annihilation operators of the anticommuting
type, such that occupation numbers obey the Pauli principle and basis vectors
are antisymmetric under simultaneous exchange of labels. The excitations of the
ground state are particles with spin 1/2 obeying Fermi-Dirac statistics.
The Dirac field describes two distinct types of particles with the same mass
and spin. The particle states can be eigenstates of Q (which commutes with
the operator P of total four-momentum), and those with Q = +1 ( 1) are
conventionally called particles (antiparticles). Q itself equals the total number of
particles minus antiparticles.
A single Majorana field has only one type of particle of a given spin (and
correspondingly, no conserved Q). The concept of antiparticles does not arise
in this case, except that for zero mass Q5 is conserved and it can take the role
of Q in the Dirac case. In the non-relativistic limit of a single Majorana field a
new U(1) symmetry emerges with a Noether charge corresponding to conservation of particles. Relativistically only the number of particles minus the number
of antiparticles is conserved. Non-relativistic quantum electrodynamics does not
require the existence of antiparticles, but, as we shall see in chapter 8, the relativistic formulation does predict antiparticles indeed.
The path-integral description of fermions requires the introduction of anticommuting numbers, which have nice properties and allow for an easy evaluation
6.8. APPENDIX: MORE ON SPIN AND STATISTICS
153
of the correlation functions in the free case. Perturbation theory leads again to
diagrams, here with Dirac propagators as well. A heuristic derivation of scattering amplitudes shows how spin dependence is represented by the polarization
spinors.
6.8
Appendix: More on spin and statistics
We give here another reasoning that a description of spin 1/2 particles in terms
of spinor fields requires the particles to obey Fermi-Dirac statistics.
Assume a theory of free spin 1/2 particles, in which there is a vacuum state |0i
with zero energy-momentum, and one particle states |pi with energy-momentum
p ,
P |0i = 0, P |pi = p |pi
(6.192)
(as before P is the energy-momentum operator and = is a spin index). The
conventions are such that these states are obtained by the action of standard
boosts `p (as in Appendix 5.5.6) on a particle state at rest,
|pi = U (`p )|
pi,
= 0, p0 = m,
p
(6.193)
where U (`p ) is the unitary operator representing `p in Hilbert space. The index
= labels the eigenvalue of the third component of angular momentum J3 in
the rest frame of the particle,
1
J3 |
pi = |
pi.
2
(6.194)
Let (x) now be a Majorana spinor-field operator. In the Majorana representation for the gamma matrices this is a hermitian spinor-field
a (x) = a (x).
(6.195)
By analogy to the scalar field we assume (x) to annihilate spin 1/2 particles to
the vacuum according to
h0|a (x)|p, i = ua (p, ) eipx .
(6.196)
The form of this equation is dictated by translation invariance (the factor exp(ipx))
and Lorentz invariance (the factor ua (p, ), because |p, i and u(p, ) are constructed in exactly the same way with the boost `p ). The remaining factor (= 1)
is a q
normalization condition for (x). In general we may have an additional factor Z , which we choose to be 1 in case of no interactions. Taking the complex
conjugate of (6.196) and multiplying by by , gives
h0|(x)|p,
i = v(p, ) eipx .
(6.197)
154
We have seen that free fields create only single particle states out of the vacuum.
If we assume this to be the case of our free spinor field as well, we can derive the
vacuum expectation value of equal time commutator or anticommutator relations
as follows. Using completeness we have
h0|(x) (y)|0i
=
=
XZ
XZ
dp h0|(x)|p, ihp, |(y)|0i
(6.198)
dp eip(xy) u(p, ) u
(p, )
(6.199)
dp eip(xy) (m ip ).
(6.200)
Similarly, we have
h0|b (y) a (x)|0i =
=
XZ
XZ
dp h0|b (y)|p, ihp, |a(x)|0i
(6.201)
dp eip(yx) va (p, ) vb (p, )
(6.202)
dp eip(yx) (m + ip )ab .
(6.203)
From these relations now follow the vacuum expectation values of equal time
commutators or anticommutators:
h0|[a (x) b (y) b (y) a (x)]|0ix0 =y0
=
=
(6.204)
dp [eip(xy) (m ip ) [eip(xy) (m + ip )]ab
(6.205)
dp eip(xy) [(m ipk k )(1 1) + ip0 0 (1 1)]ab .
(6.206)
It follows that the vacuum expectation value of the commutator [a (x), b (y)] is
given by
h0|[a (x), b (y)]|0ix0=y0 = 2(m k k )ab
dp eip(xy) .
(6.207)
This does not vanish for x 6= y: the commutator is not local in the sense that
it vanishes for spacelike distances. On the other hand, the vacuum expectation
value of the anticommutator is simple and local:
0
h0|{a (x), b (y)}|0ix0=y0 = iab
(x y),
h0|{a (x), b (y)}|0ix0=y0
= ab (x y).
(6.208)
(6.209)
The spinor operators at different points in space evidently do not commute at

equal times: they anticommute! This then leads to the conclusion that multiparticle states created by repeated application of the spinor field on the vacuum state
are antisymmetric in interchange of labels: the particles have to be fermions.
Let us list the important ingredients which went into this celebrated spinstatistics connection:
6.9. APPENDIX: ANTICOMMUTING VARIABLES
155
- Hilbert space (of course with positive metric);

- a vacuum state |0i and one particle states |p, i with the expected energy
momentum eigenvalues (6.192);
- translation invariance and Lorentz invariance;
- locality.
We stress here the relevance of the locality principle. Imagine constructing local
observables O(x) out of the spinor field. We want these to be local, i.e. they
should commute for spacelike separations,
[O(x), O(y)] = 0, (x y)2 > 0.
(6.210)
The spinor fields are not local in this sense, because anticommutators are not
commutators, and apparently spinor fields are not observables. However, bilinears of the type ( is some combination of Dirac matrices)
O(x, ) = (x)(x),
(6.211)
(and generalizations thereoff, e.g. involving derivatives), do satisfy locality. This

follows from the fact that we can express commutators of field bilinears in terms
of anticommutators of fields. The anticommutators satisfy locality, and therefore
also the commutators of the bilinears,
[O(x, 1 ), O(y, 2)] = 0, (x y)2 > 0.
(6.212)
Familiar observables like currents and the energy momentum tensor can indeed
be expressed as bilinears. Had we insisted on commutation relations for (x),
we would have had to assume a nonlocal commutator [a (x), b (y)], as follows
from its vacuum expectation value (6.207), and we could not have satisfied the
locality principle.
6.9
Appendix: Anticommuting variables
Because fermion variables anticommute, the variation of the action can be written
in two equivalent but different ways
S =
Z
S
d x a (x)
= d4 x S
a (x),
a (x)
a (x)
4
(6.213)
and correspondingly we have to distinguish between left and right derivatives. To

see this in more detail let us write the Majorana action in a condensed notation,
S=
d4 x
1
1 T
(m + ) Skl k l ,
2
2
(6.214)
156
where Skl = Slk . Then

S = S( + ) S()
1
Skl [( k + k )( l + l ) k l ]
=
2
1
=
Skl ( k l + k l )
2
= Skl k l = Skl l k .
Hence
(6.215)
S = Skl l ,
k
= Skl l .
k
(6.216)
The differentiations also behave like anticommuting variables, e.g.
l = kl ,
k
( l m ) = kl m l km .
k
(6.217)
(6.218)
We shall always use left derivatives. Notice that (2) S/ l k (i.e. first / k
and then / l ) equals Skl .
To define integration, consider two anticommuting variable and . Since
2
= 2 = 0, the most general function of and has the form
f (, ) = f0 + f1 + f2 + f12 ,
(6.219)
with the f s ordinary real or complex numbers. In the path integral is was crucial
to assume translation invariance of the measure, so we require the integral over
to satisfy
Z
Z
d f (, ) = d f ( + , ).
(6.220)
Suppose f12 = 0. Comparing the coefficients of we get

Z
from which we conclude

The remaining term
d f2 =
Z
Z
d (f1 + f2 ),
(6.221)
d = 0.
(6.222)
d f1
(6.223)
needs to be given some value. The simplest choice is

Z
d = 1,
(6.224)
157
6.9. APPENDIX: ANTICOMMUTING VARIABLES

such that
d f1 = f1 . Then, if f12 is also non-zero,

Z
d f (, ) = f1 + f12 .
(6.225)
Symbols like d are also treated as anticommuting. The odd looking integration rules (6.222,6.224) are known as Berezin integration. Note that fermionic
integration is just differentiation!
Consider now a finite set of anticommuting numbers k , k = 1, , n. These
generate a so-called Grassmann algebra: indeed, anticommuting variables are
nothing but generators of such an algebra. We enlarge the set by new generators
k and stipulate that under complex conjugation
(ck ) = c k ,
(ck l ) = c l k ,
(6.226)
where c is a complex number. The above rule has already been used in the
discussion of the reality of the Dirac action.
We now have the interesting formula
Z
d d e
= det M,
d d
n
Y
dk dk ,
(6.227)
k=1
where M is any square matrix. The derivation is straightforward. We have

Z
d d
k1 l1
kn ln
=
=
d1 dn dn d1 k1 kn ln l1
d1 dn dn d1 k1 kn l1 ln 1 n n 1
= k1 kn l1 ln =
(1) k1 ,l1 kn ,ln .
(6.228)
Then
Z
d d el Mlk k =
d d
(+1)n
k1 l1 kn ln Ml1 k1 Mln kn
n!
1
k k l l Ml k Mln kn = k1 kn M1k1 Mnkn
n! 1 n 1 n 1 1
= det M.
(6.229)
Using translation invariance10 we can also derive

Z
d d e
M + +
= det M e
M 1
(6.230)
These formulas need of course modification if M does not have an inverse. However, note that the expression on the left-hand side is defined for arbitrary M
10
We shall always assume a sufficiently large set of anticommuting generators to be able to

introduce sources , , etc.
158
Writing an arbitray function f (, ) as

f (, ) =
n
X
1
k1 kp lq l1 fk1 kp ,l1 lq ,
p!q!
p,q=0
(6.231)
with coefficients completely antisymmetric in permutations on the ks or ls, it is

easy to derive along similar lines that
Z
d d f (, ) = f1n,1n
=
(6.232)
d d f ( + , + ),
d d f (A, B ) = det A det B
d d f (, ),
(6.233)
(6.234)
which we may interprete as a property of the fermionic measure under linear

transformations of variables:
d(A + ) = (det A)1 d, d(B + ) = (det B)1 d .
(6.235)
Note that the effect of the matrices A and B is opposite to the bosonic case, for
which dn (A) = det A dn .
Corresponding formulas can be derived for the real case and antisymmetric
M , M = M T . Since the complex case can be viewed as a doubling of the real
case we get (up to a possible sign depending on the ordering in the definition of
d)
Z
1 T
(6.236)
d e 2 M = det M ,
which is a monomial in the matrix elements Mkl = Mlk . It is called a pfaffian.
Last, but not least, we can formulate quantum mechanics of fermions in terms
of wave functions ( ) = h|i representing a state |i and matrix elements
2 i representing operators O
in Hilbert space. This is usually
O(1 , 2 ) = h1 |O|
done in terms of fermionic coherent states. See for example the book by this
author, appendix C.
6.10
Problems
1. Exercize
Verify eq. (6.37).
2. Consistency check
In the correct quantization with anticommutators, verify the analogue of
(6.21) for antiparticle states, for J3 :
pi =
J3 |
1
|
pi,
2
p (m, 0)
(6.237)
159
6.10. PROBLEMS
(where |pi = d (p, ) |0i). Assume the vacuum to be rotationally invariant,

J|0i = 0.
3. Matrix elements of currents
a. For a free Dirac field, verify the following matrix elements of the current
:
j = i
h0|j (x)|0i
hp0 0 |j (x)|pi
hp0 0 |j (x)|pi
hp0 0 , p|j (x)|0i
h0|j (x)|p, p0 0 i
=
=
=
=
=
0,
0
u
0 i u ei(pp )x ,
0
v i v 0 ei(pp )x ,
0
u
0 i v ei(p+p )x ,
0
v0 i u ei(p+p )x ,
(6.238)
(6.239)
(6.240)
(6.241)
(6.242)
where u = u(p, ), u
0 = u
(p0 , 0 ), etc.
Hint:11 convert the matrix elements to path integral averages, e.g.
hp0 0 |j (x)|pi = hb(p0 , 0 , ) (x)i

(x) b (p, , )i.
(6.243)
b. Using the charge conjugation matrix C introduced in Appendix 5.5 to

relate the us with the vs, verify from the explicit answers obtained above
that
(6.244)
hp0 0 |j (x)|pi = hp0 0 |j (x)|pi.
c. Verify j = 0 in the above matrix elements of the current j .
d. For the explicit expressions obtained above for the matrix elements of
j verify that
hp0 0 |Q|pi = hp0 0 |pi,
(6.245)
etc., where Q = d3 x j 0 (x).
R
e. Gordon decomposition
Recall that for general and ,
S =
i
[ , ].
4
(6.246)
Let u = u(p, ), u
0 = u
(p0 , 0 ). Verify that
2(p p0 ) u
0 S u = 2m u
0 u + i(p + p0 ) u
0 u,
(6.247)
and the Gordon decomposition

u
0 i u =
11
1
[(p + p0 ) u
0 u + 2i(p p0 ) u
0 S u].
2m
(6.248)
Note that this technique can also be used for other matrix elements, for example, the matrix
element (2.133) of the energy-momentum tensor of the scalar field.
160
4. Commutators revisited
This problem is the analog of Problem 2.5.
Consider fermion operators A and A with the anticommutation relations
{A , B } = 0, {A , B } = 0, {A , B } = AB .
(6.249)
Let M be a matrix with c-number elements MAB , with vanishing trace,

tr M = 0, and define O(M ) = MAB A B (summation over repeated indices). Verify that
[O(M1 ), O(M2 )] = O([M1 , M2 ]).
(6.250)
Hint, for arbitray operators X, Y and Z we have (check) [XY, Z] = X{Y, Z}

{X, Z}Y , and also [XY, Z] = X[Y, Z] + [X, Z]Y .
As an application, let A be the pair of indices (a, x) and A = a (x) a
Dirac field. Obtain the commutation relations for the angular-momentum
operators Jk = klm Jlm /2, for which MAB iklm xl m (x y)ab + (x
y)(Sk )ab .
5. Fermion-(anti)fermion scattering
Derive, along the lines in sect. 6.5, the diagrams and scattering amplitudes
for the processes (p1 1 ) + (p2 2 ) (p01 01 ) + (p02 02 ) and (p1 1 ) + (p2 2 )
(p01 01 ) + (p02 02 ). Put all relevant labels in the diagrams, as in Figs. 6.3 and
6.4.
Answers for these (anti)fermion scattering amplitudes:
i
m2 + (p1 p01 )2
i
u
02 (ig)u1 u
01 (ig)u2 2
m + (p1 p02 )2
(6.251)
iT (p01 01 , p02 02 ; p1 1 , p2 2 ) = u
01 (ig)u1 u
02 (ig)u2
and
i
m2 + (p1 p01 )2
i
+u
01 (ig)v20 v2 (ig)u1 2
,
m + (p1 + p2 )2
(6.252)
u01 (ig)u1 v2 (ig)v20

iT (p01 01 , p02 02 ; p1 1 , p2 2 ) =
(p02 , 02 ) etc.
where u
02 = u
161
6.10. PROBLEMS
6. Fermion-antifermion annihilation and creation
Give the diagrams (with labels) and scattering amplitudes for the processes
(p) + (p0 0 ) k + k 0 and k + k 0 (p) + (p0 0 ). In these two problems
do not give the derivation but make sure you are able to do so.
Answers for these fermion-antifermion annihilation or creation amplitudes:
iT (k, k 0 ; p, p0 0 ) = v(p0 , 0 )(ig)(i)
+ k k0,
M i(p k)
(ig)u(p, )
M 2 + (p k)2
(6.253)
and
iT (p, p0 0 ; k, k 0 ) = u(p, )(ig)(i)
+ k k0.
M i(p k)
(ig)v(p0 , 0 )
2
2
M + (p k)
(6.254)
162
Chapter 7
Quantized electromagnetic field
The quantization of the electromagnetic field poses special problems due to gauge
invariance. We first quantize canonically in the Coulomb gauge. The resulting
formulation is logically simple, but its lack of manifest covariance can be cumbersome in practical calculations. This problem is remedied by using the path
integral in transforming to a generalized covariant gauge. In the resulting formulation feature unphysical particles, named ghosts, and Hilbert space has to
be extended into a vector space with indefinite metric. However, the connection with the Coulomb gauge should guarantee correct physics, whereas manifest
covariance will make the calculation of perturbative corrections in interacting
gauge-field theory a lot simpler.
7.1
Quantization in the Coulomb gauge
We rewrite the Maxwell action with external current in the form

S =
dt L,
(7.1)
1
d3 x ( F F + J A )
4

Z
1
3
=
dx
Am Am A m m A0
2
1
1
1
+ n Am m An m An m An + m A0 m A0
2
2
2
0
m
+ J A0 + J Am
L =
(7.2)
(7.3)
There is now a complication in the canonical formalism which is typical for gauge
theories: A 0 is lacking in L, so the canonical conjugate to A0 vanishes, 0
L/ A 0 = 0. One way to deal with this is to eliminate A0 itself as a dynamical
variable by choice of gauge condition. Consider the equation of motion that
163
164
CHAPTER 7. QUANTIZED ELECTROMAGNETIC FIELD
follows from varying the action with respect to A0 ,

0 = S =
d4 x ( F 0 + J 0 )A0 .
(7.4)
This is Gausss law, or Coulombs law,

0=
S
= m F m0 + J 0 = E + J 0
A0
= m (m A0 0 Am ) + J 0 .
(7.5)
Now impose the Coulomb gauge condition

m Am = 0.
(7.6)
This has the effect that the time derivative drops out of (7.5), such that (7.5)
takes the form
A0 = J 0 , 2 .
(7.7)
Since this equation does not contain time derivatives, it is not a dynamical equation anymore, but an equation of constraint at every instant in time. With
suitable spatial boundary conditions the potential A0 is completely determined
in terms of J 0 . For infinite space
A0 (x, t) =
d3 y
1
J 0 (y, t),
4|x y|
(7.8)
where we used the fact that the Coulomb potential is a Green function for the
laplacian :
1
= (x y).
(7.9)
4|x y|
Hence, A0 is not a dynamical variable in the Coulomb gauge.

We shall use the Coulomb gauge for the canonical formalism and continue to
write A0 , for simplicity, keeping in mind that it is a given function of J 0 . In this
gauge we can rewrite the lagrangian in the form
1
1
L =
d x A m A m Am ()Am + Jm Am EC ,
2
2

Z
Z
1
1
3
0
0
0 0
EC =
d x m A m A + J A = d3 x J 0 A0 .
2
2
Z
(7.10)
(7.11)
We used m Am = 0, A0 = J 0 and made partial integrations of m assuming

boundary conditions such that surface terms vanish. The quantity EC is the
Coulomb energy. Using (7.8) it can be written as
1
1Z 3 3 0
J 0 (y, t).
d x d y J (x, t)
EC =
2
4|x y|
(7.12)
7.1. QUANTIZATION IN THE COULOMB GAUGE
165
The lagrangian is now in the form L(q, q)

with qk (t) Am (x, t).
Next we have to deal with the continuous nature of the index x and the constraint m Am (x, t) = 0. This can be done by expanding the potentials into a
discrete set of basis functions fm (x) satisfying m fm (x) = 0. Let us enclose the
system in a cubic box L/2 xm L/2 with periodic boundary conditions. For
a large enough box its finiteness and the type of boundary conditions should not
matter. Periodic boundary conditions are convenient because with it all boundary terms in partial integrations vanish (the box has no boundary) and they do
not spoil translation invariance. We can use the discrete set of eigenfunctions
of the laplacian to construct the fm (x). The real eigenfunctions of the laplacian correspond to products of the standing waves cos(k1 x1 ) cos(k2 x2 ) cos(k3 x3 ),
sin(k1 x1 ) cos(k2 x2 ) cos(k3 x3 ), . . . , sin(k1 x1 ) sin(k2 x2 ) sin(k3 x3 ), with km = 2nm /L,
nm = 0, 1, 2, . . . , and the eigenvalues are given by 2 = k2 . Out of these
eigenfunctions the real fm (x) can be constructed satisfying m fm (x) = 0. The
details of this are tedious and not needed in the following and we shall just record
their properties:
fm (x) = 2 fm (x), m fm (x) = 0,
d3 x fm (x) fm (x) = ,
X
(7.13)
(7.14)
T
fm (x)fn (y) = Pmn
(x, y).
(7.15)
We have written these equations in general complex form, because we usually use
the plane waves that are eigenfunctions of the gradient operator n , which are
complex. In the next section we shall give an explicit construction of these basis
functions. For the moment have to keep in mind that the fm (x) are real. The
T
object Pmn
(x, y) is a projector on the space of transverse vector functions, i.e.
a projector, P 2 = P , or
Z
T
T
d3 y PklT (x, y) Plm
(y, z) = Pkm
(x, z),
(7.16)
T
that is transverse, m Pmn
(x, y) = 0. It is the identity operator for transverse
vector functions (satisfying m Am (x) = 0),
T
d3 y Pmn
(x, y)An (y) = Am (x),
if m Am = 0.
(7.17)
An explicit expression for P T will be given in the next section (cf. (7.42)).
P
In the summation we exclude the zero mode k = (0, 0, 0) (this would be
automatic with Dirichlet boundary conditions). This means that we exclude
here potentials Am which are constant in space. Such potentials complicate the
(otherwise interesting) mathematics and we usually do not need them in physical
applications.
166
In terms of these basis functions we can now expand the potentials in normal
modes,
X
Am (x, t) =
q (t)fm (x),
(7.18)
q (t) =
d3 x fm (x)Am (x, t),
(7.19)
and in terms of the new coordinates q the lagrangian takes the form, for J = 0,
L=
X 1
1
q q 2 q q .
2
2

(7.20)
This shows that the electromagnetic field is equivalent to an infinite set of harmonic oscillators, with unit mass and frequencies . The canonical description
is now an obvious generalization of the case of one harmonic oscillator,
p = L/ q = q ,

X 1
1 2
H =
p p + q q ,
2
2
(q , p ) = , (q , q ) = (p , p ) = 0,
p = (p , H), q = (q , H) = p ,
(7.21)
(7.22)
(7.23)
(7.24)
where (A, B) is the Poisson bracket. Evidently, the canonical conjugate to the
field Am (x) is
m (x) = (Am (x), H) = A m (x)
X
=
p fm (x).
(7.25)
The system is quantized by imposing canonical commutation relations between the ps and qs,
[q , p ] = i ,
[q , q ] = [p , p ] = 0.
(7.26)
The quantized electromagnetic field is now an operator in Hilbert space. The

commutation relations between the p and q imply the following relations between Am and m ,
T
[Am (x), n (y)] = iPmn
(x, y), [Am (x), An (y)] = [m (x), n (y)] = 0.
(7.27)
For example,
[Am (x), n (y)] =
[q , p ] fm (x)fn (y) = i
T
iPmn
(x.y).
fm (x)fn (y)
(7.28)
167
7.2. ENERGY-MOMENTUM EIGENSTATES
7.2
Energy-momentum eigenstates
To guide our physical interpretation we shall use the energy momentum P of the
field, which is now also an operator, and determine its eigenstates and eigenvalues.
In the Coulomb gauge A0 vanishes when J = 0, cf. (7.8). Then (cf. (1.78))
1
(Em Em + Bm Bm )
2
1
=
(Am Am + n Am n Am n Am m An ),
2
= nmp Em Bp = A m n Am + A m m An ,
T 00 =
T 0n
(7.29)
(7.30)
giving
P0 =
Pn =
d3 x T 00 =
d3 x T 0n =
d3 x
1
1
m m + Am ()Am ,
2
2

d3 x (m n Am ).
(7.31)
(7.32)
We used the Coulomb gauge condition m Am = 0 and A m = m . Notice that

there is no operator-ordering ambiguity in Pm : we can also write m to the right of
Am , the difference involves the derivative of the commutator, m (xy)|x=y = 0.
Using the normal mode expansion we find
P
1
p p + 2 q q
2
2
X 1
= H.
(7.33)
The momentum operator is less easy to express in terms of the normal modes
because the real mode functions fm (x) are not eigenfunctions of n . Therefore
we now introduce a different set fm (x) which are eigenfunctions of the hermitian
differential operators in and , and satisfy m fm (x) = 0. They are complex
and have the form1 ( (k, ))
m
fk,
(x) = em (k, )eikx , km = nm 2/L, nm = 0, 1, 2, . . . ,
(7.34)
These are clearly eigenfunctions of n and . Recall that the nm have to be

integers to satisfy periodic boundary conditions in a box of size L3 . To satisfy
m f m = 0, the em (k, ) have to be orthogonal to k (hence the terminology
transverse), as illustrated in fig. 7.1,
k e(k, ) = 0.
(7.35)
For example for k = (0, 0, |k|) = |k|3, e(k, 1) = (1, 0, 0) = 1, e(k, 2) = (0, 1, 0) =
1
We use here a normalization of the Fourier modes that treats the volume-factors L 3 differently from the scalar and fermion cases (for no good reason).
168
e(k,1)
e(k,3) = k/|k|
e(k,2)
Figure 7.1: Real transverse polarization vectors e(k, 1) and e(k, 2), and the longitudinal unit vector e(k, 3).
2. In general the e(k, ) may be obtained from this by a rotation, a standard
rotation that takes (0, 0, |k|) into k. Another set, which is also well-known from
classical electrodynamics, consists of the right- and left-handed polarization vectors
1
e(k, ) = [e(k, 1) ie(k, 2)].
(7.36)
2
The polarization vectors satisfy
em (k, ) em (k, 0 ) = 0 ,
X
em (k, )en (k, ) =
mn
(7.37)
km kn
k2
T
Pmn
(k).
(7.38)
The basis functions are orthogonal and complete in the sense (7.15), with
(k, ),
0 0 L3 k,k0 ,
X
1 X
,
L3 k,
T
Pmn
(x, y)
(7.39)
(7.40)
(7.41)
!
1 X ikx+iky
km kn
e
mn
,.
3
L k
k2
(7.42)
where the zero mode k = 0 is absent again.

We now expand the Am and m in terms of these basis functions as follows,
Am (x) =
m (x) =
1 X 1 ikx m
[e e (k, )a(k, ) + eikx em (k, ) a (k, )],
L3 k, 2k 0
(7.43)
1 X 1
[ik 0 eikx em (k, )a(k, ) + ik 0 eikx em (k, ) a (k, )],
L3 k, 2k 0
where
k 0 = |k|
(7.44)
7.2. ENERGY-MOMENTUM EIGENSTATES
169
and the factor 1/2k 0 anticipates covariant normalization. The form of (7.43) is
guided by the harmonic oscillator forms,
1
2 (a + a ),
2
1
2 (i a + i a ).
=
2
q =
(7.45)
(7.46)
The relations (7.43) define a(k, ) and a (k, ), and may be inverted as follows.
We write
am (k) =
em (k, )a(k, ),
(7.47)
a(k, ) = em (k, ) am (k).
(7.48)
Then
1
[am (k) + am (k)],
0
2k
Z
1
3
ikx
[iam (k) + iam (k)],
d xe
m (x) =
2
d3 x eikx Am (x) =
(7.49)
(7.50)
giving
am (k) =
am (k) =
d3 x eikx [k 0 Am (x) + im (x)],
(7.51)
d3 x eikx [k 0 Am (x) im (x)].
(7.52)
The commutation relations between am (k) and am (k) can now be calculated from
(7.27) to be
T
[am (k), an (l)] = Pmn
(k) 2k 0 L3 k,l ,
[am (k), an (l)] = [am (k), an (l)] = 0.
(7.53)
For example,
[am (k), an (l)]
d3 xd3 y eikx+ily [k 0 Am (x) + im (x), l0 Am (y) im (y)]
= (k 0 + l0 )
0
= (k + l
T
d3 xd3 y eikx+ily Pmn
(x, y)
T
)Pmn
(l)
d3 y ei(lk)y
T
= 2k 0 Pmn
(k) L3 k,l .
(7.54)
[a(k, ), a (k0 , 0 )] = 2k 0 L3 k,k0 ,0 ,

[a(k, ), a(l, 0 )] = [a (k, ), a (l, 0 )] = 0.
(7.55)
(7.56)
It follows that
170
Hence, the new a and a satisfy the commutation relations of creation and annihilation operators of an infinite set of harmonic oscillators labelled by (k, ).
Expressing the hamiltonian (7.31) and momentum operator (7.32) in terms
of the creation and annihilation operators we find (cf. Problem 1)
P0 =
Pm =
E0 =
1 X 1
a (k, )a(k, )k 0 + E0 ,
L3 k, 2k 0
1 X 1
a (k, )a(k, )km ,
L3 k, 2k 0
X
k,
1 0
k .
2
(7.57)
(7.58)
(7.59)
By analogy to the ordinary harmonic oscillator we recognize the number operator

a (k, )a(k, ) for each mode (k, ). The ground state (state with lowest energy)
is the no-quantum state |0i defined by
a(k, )|0i = 0,
(7.60)
P 0 |0i = E0 |0i, P|0i = 0.
(7.61)
with
The excited states are given by
|k, i = a (k, )|0i,
|k1 1 , k2 2 i = a (k1 , 1 )a (k2 , 2 )|0i,
|k1 1 , k2 2 , k3 3 i = a (k1 , 1 )a (k2 , 2 )a (k3 , 3 )|0i,
(7.62)
(7.63)
(7.64)
[P ,0 E0 ]|k1 1 . . . kn n i = (k1 + . . . + kn )|k1 1 . . . kn n i,
(7.65)
etc., with
The four-momenta k correspond to zero mass, k k = 0. The excited states

are the photons, which are massless particles. The symmetry of the basis vectors
|k1 1 kn n i under interchange of labels ((ki i ) (kj j ) has the consequence
that photons follow Bose-Einstein statistics.
In the infinite volume limit the ground state represents the vacuum. In this
limit the wave vectors become practically continuous, in the sense that for a
continuous function F (k),
Z
1 X
d3 k
F
(k)
F (k).
L3 k
(2)3
(7.66)
Furthermore, in the sense of generalized functions

L3 k,l (2)3 (k l).
(7.67)
171
7.3. LORENTZ INVARIANCE

Hence
hk, |k 0 , 0 i 2k 0 (2)3 (k k0 ),0 ,
(7.68)
and the energy density of the ground state takes the form
X
E0
L3
d3 k 1
(k 0 )2 ,
3
0
(2) 2k
(7.69)
where we recognize again the Lorentz invariant volume element.

dk
d3 k 1
,
(2)3 2k 0
k 0 = |k|.
(7.70)
The energy density is twice ( = 2) that of a massless scalar field. We shall

absorb it in a renormalization of the cosmological constant such that
P
E0 = 0,
(7.71)
similar to the discussion for the scalar field in sect. 2.3.
7.3
Lorentz invariance
The quantization in Coulomb gauge is straightforward, but not manifestly Lorentz

covariant. One might expect covariance to be manifest in a covariant gauge like
the Lorentz gauge A = 0, but quantization is not so easy in this gauge. Later
we will make contact with the Lorentz gauge via the path integral. However, the
theory in Coulomb gauge should still be Lorentz invariant, because after quantization the generators of Lorentz transformation J exist as operators and they
provide a representation of the Lorentz group in Hilbert space, in the usual form
U = exp( 21 i J ).
Let us take a closer look at this. The Noether form of the energy-momentum
tensor turns out to be
TNoether
= F A + L,
(7.72)
which is not symmetric and also not gauge invariant. This tensor can be improved
by adding the current
j = F A ,
(7.73)
which is chargeless when the free Maxwell equations are satisfied,
Z
d3 x j 0 =
d3 x F 0r r A =
d3 x (r F 0r ) A = 0,
(7.74)
and conserved
j = F A + F A = 0.
(7.75)
172
The improved tensor
T TNoether
+ j = F F + L
(7.76)
is gauge invariant, symmetric, conserved, with the same charges P as the

Noether charges. It is the one entering in Einsteins equations. Because it is
symmetric, the currents
J = x T x T ,
(7.77)
are also conserved, J = 0, and we take it for granted that they are satisfactory generalized angular momentum currents of conserved Lorentz generators
J
d3 x J 0 .
(7.78)
For example, in Coulomb gauge the angular momenta of the free field (for which
A0 = 0) take the canonical form
1
Jk = klm Jlm =
2
d3 x (ir ) [iklm xl m rs + (Sk )rs ] As ,
(7.79)
with Sk the spin matrices in the vector representation,2

(Sk )lm = iklm
(7.80)
Rotations of the operator fields are generated by taking the commutator [Ar , Jk ].
With the transverse projector P T in the commutation relations, this commutator
takes the form
[Ar (x), Jk ] =
T
0
0 0
0
d3 x0 Prr
(7.81)
0 (x, x ) [iklm xl m r 0 s + (Sk )r 0 s ] As (x ),
= [iklm xl m rs + (Sk )rs ] As (x) + gauge transf.,
(7.82)
where the gauge transformation, i.e. a term of the form r , comes from the r
T
term in Prr
0:
1
T
0
0
0
.
(7.83)
Prr
0 (x, x ) = rr 0 (x x ) + r r 0
4|x x0 |
This gauge transformation takes care that A stays in the Coulomb gauge. In fact,
for rotations it does not contribute (make a partial integration for r0 0 ), because
A = 0 is a rotationally invariant equation. However, in general one finds
that under Lorentz transformations A transforms like a four-vector field, up to
a gauge transformation that keeps it in Coulomb gauge:
U (`)A (x)U (`) = ` A (`1 x) + (x),
(7.84)
with such that such that A(x) = 0, A0 (x) = 0.

2
With S in the defining representation (1.52) of the Lorentz group, Sk = 21 kab Sab .
173
7.4. PHOTONS
7.4
Photons
We have seen that the mass of the photon is zero,

P P |k, i = (k2 k02 )|k, i = 0.
(7.85)
We shall now determine its possible helicities. The helicity is defined as the
eigenvalue of the angular momentum operator J in the direction of motion,
k
k =
.
|k|
(7.86)
k J|k, i = [k J, a (k, )] |0i,
(7.87)
k J |k, i = |k, i,
To determine the helicities we use
so we first have to calculate the above commutator. The calculation of a(k, ) in

terms of Al and l is conveniently summarized by the pair of equations
Ar (x) =
XZ
a(k, ) =
dk a(k, ) er (k, ) eikx + a (k, ) er (k, ) eikx , (7.88)
d3 x eikx er (k, ) (i 0 i 0 ) Ar (x),
(7.89)
which is valid for the free Maxwell field (kx = kx k 0 x0 = kx |k|x0 ). We then
have using (7.82),
[a(k, ), Jl ] =
d3 x eikx er (k, ) [ilmn xm n rs + (Sl )rs ] (i 0 i 0 ) As (x).
(7.90)
Making a partial integration the orbital part of the angular momentum operator,
ilmn xm n lmn xm kn . This is transverse to k, so in k J = kl Jl this term does
not contribute. Thus we have
[a(k, ), k J] =
d3 x eikx (i 0 i 0 ) er (k, ) kl (Sl )rs As (x),
(7.91)
and conjugating this expression,

[k J, a (k, )] =
d3 x eikx (i 0 i 0 ) Ar (x) k Srs es (k, ).
(7.92)
The problem is reduced to calculating the action of the spin matrix S on the polarization vectors e(k, ). Recall that in the vector representation the spin matrices
S1 , S2 and S3 are represented by (Sl )mn = ilmn , which satisfy [Sk , Sl ] = iklm Sm
and S2 = s(s + 1) = 1(1 + 1) = 2. The right and left handed polarization vectors (7.36) were constructed such that they are eigenvectors of S3 for the special
momentum k = |k|3 (in which case em (k, j) = mj , j = 1, 2, 3):
(S3 )mn en (k, ) = em (k, ),
k = |k|3,
(7.93)
174
with the usual phase relations (k = |k|3)
(S1 + iS2 )mn en (k, ) =

2 e (k, 0),
m
(S1 + iS2 )mn en (k, 0) =
2 em (k, +),
en (k, 0) en (k, 3),
(7.94)
where en (k, 0) is the eigenvector of S3 with eigenvalue 0. For general k the
polarization vectors may be defined by some standard rotation taking 3 into k.

so
This does not affect the eigenvalues of the rotation invariant combination kS,
finally:
k Smn en (k, ) = em (k, ) k J |k, i = |k, i,
(7.95)
where we used (7.87), (7.92) and the conjugate of (7.89). The eigenvector e(k, 3)
with eigenvalue 0 does not occur among the polarization vectors.
k of kS
So the photons have helicity 1, and there is no helicity-zero state, as might
be expected from the vector representation in which the eigenvalues of S3 are
+1, 0, 1. The helicity zero polarization vector would be the longitudinal mode
e(k, 3) k, which is equivalent to a gauge transformation and therefore unphysical. It was eliminated by the Coulomb gauge condition.
In general, massless particles have only two independent spin-states. This can
be understood directly from the representation theory of the Poincare group, see
e.g. Weinberg I, Ryder. Under a general Lorentz transformation one finds that
the polarization vectors
e (k, ) (0, e(k, )),
k 0 = |k|,
(7.96)
transform covariantly up to a gauge transformation

` e (k, ) =
X
0
C0 (`, k)e (`k, 0 ) + terms k ,
(7.97)
and correspondingly3 for the states in Hilbert space:

U (`)|ki =
X
0
C0 (`, k)|k0 i.
(7.98)
The states are gauge invariant, a(k, ) is gauge invariant. The matrix C0 0
in case of the helicity polarization vectors: for massless particles the helicity is
Lorentz invariant. This is not true for massive particles, for example, massive
spin-1 particles. Loosely speaking, a massive particle has a rest frame, therefore
we can make a special Lorentz transformation along its momentum vector k
to bring it first to rest, and then boost it again in direction k, e.g. giving it
momentum k. This transformation does not affect the transverse polarization
3
This correspondence is also P

true for the spin 1/2 case, where the transformation prop0
erty of the spinor
L(`)
u(p,
)
=
0 C0 (`, p)u(`p, ), implies for the state in Hilbert space
P
0
U (`) |p, i = 0 C0 (p, )|`p, i (as follows from the expansion of (x) in terms of creation
and annihilation operators).
175
7.5. PATH INTEGRAL FOR THE MAXWELL FIELD
vectors, hence also not the eigenvalue of k S, but since the momentum flips sign,
so does the helicity.
In conclusion, in Coulomb gauge the quantized Maxwell field has a sound
physical interpretation. Furthermore, local observables satisfy locality, see Appendix 7.9). For perturbative calculations this gauge is however rather awkward,
and we shall transform to a manifestly covariant formalism using the path integral, to be introduced next.
7.5
Path integral for the Maxwell field
It is straightforward to write down a path integral representation for the electromagnetic field in the oscillator representation (7.20). However, we shall start
afresh and make the Ansatz
?
Z[J] h0, |0, i[J] =
[dA] eiS[A]+i
d4 x J A
(7.99)
where S[A] is the free Maxwell action and the boundary conditions are taken to
be A (x) = 0 at x0 = , similar to the case of the scalar field. We have put a
question mark above the equality sign because it is not clear at this stage if the
above formula makes sense. In fact, if we assume the measure [dA] to be formally
given by
Z
Z Y
[dA] =
dA (x),
(7.100)
x
it does not make sense. The reason is that the above implies for each A , also
integration over all gauge transforms A + , for which the integrand of the
path integral does not vary because of its gauge invariance. So we expect an
Q R
d(x). RWe assume in this argument that the external
infinity of the form x
current J is conserved such that d4 x J A is gauge invariant.
To see the problem in another way, suppose we want to separate the J dependence in the usual way by completing the square in the exponent and making a
change of variables,
A (x) = A (x) +
d4 y G (x, y)J (y),
(7.101)
such that
1
Z[J]
= exp[i
d4 x d4 y J (x)G (x, y)J (y)].
(7.102)
Z[0]
2
Here G (x, y) is supposed to be a Green function of the differential operator
Z
K = 2 + ,
(7.103)
which enters in the action as (making partial integrations)

S=
1
d4 x A K A ,
2
(7.104)
176
such that
( 2 + )
d4 y G (x, y) J (y) = J (x).
(7.105)
However, without gauge fixing such a Green function does not exist! The differential operator K has an infinite number of zero modes, i.e. eigenfunctions
with zero eigenvalue. These are the pure-gauge configurations
Apg = ,
K Apg = 0,
(7.106)
where is arbitrary. This is infinitely more general than the usual zero modes
that can be formed from plane wave solutions exp(ikx), k 2 = 0, and which are
dealt with by the i prescription , e.g. for the Green function of 2 (propagator
of a massless scalar field), G(k) = 1/(k 2 i).
So we have4 to fix the gauge in the path integral (7.99). One way to do this
is the FaddeevPopov procedure, which also generalizes also to nonabelian gauge
field theories. We shall begin with the Coulomb gauge. Consider the functional
C [A] defined by
1 = C [A]
[d] [ A ],
A A + .
(7.107)
Here [A] is a delta functional enforcing the Coulomb gauge, formally given by
[ A] =
(k Ak (x)).
(7.108)
The functional C [A] is gauge invariant,

C [A1 ] = C [A],
(7.109)
which follows easily by shifting the dummy integration variable 1 in

(7.107). Explicitly we have5
1
=
C [A]
[d] [ A + 2 ] =
1
,
det[2 ]
(7.110)
A way to see this is by making the transformation of variables 0 = 2 , [d] =

[d 0 ]/| det[2 ]|. The argument of the delta functions (2 + A) is assumed
to vanish only once, which is correct because the solution = (2 )1 A is
unique, barring zero modes of the laplacian.
4
There exist lattice regularizations which do not require gauge fixing and which are currently
used in solving gauge theories like QCD nonperturbatively by numerical simulation. However,
for the usual perturbation expansion gauge fixing is necessary.
5
As mentioned earlier we exclude zero-momentum modes, so 2 (4) (x x0 ) is a positive
matrix.
177
7.5. PATH INTEGRAL FOR THE MAXWELL FIELD
gauge orbits
hypersurface satisfying gauge condition
Figure 7.2: Gauge fixing in the path integral. Shown is the space of all gauge
field configurations. The directed lines indicate orbits of gauge transformations.
Each orbit is supposed to cross the hypersurface of functions satisfying A = 0
only once.
So C [A] is just a constant, det[2 ]. We shall keep it in full glory to better
present the following reasoning. Inserting the identity (7.107) into the integrand
of the Ansatz (7.99) we get
Z
Z[J] =
[dA] [d] [ A ] C [A] eiS[A]+i
[dA ] [d] [ A] C [A ] eiS[A

Z
[d] [dA] [ A] C [A] eiS[A]+i
JA
(7.111)
]+i
JA
JA
(7.112)
(7.113)
In the second line we made the changeR of variables A A and in the

third
R
line we use the gauge invariance of S, JA, C and [dA]. The factor [d] in
front in (7.113) is the divergence anticipated
earlier, whereas the remaining path
R
integral over A looks sound. The infinite [d] is the volume of the gauge group,
a constant, which we drop. So the corrected Ansatz is now
Z[J] =
[dA] [ A] C eiS[A]+i
d4 x J A
(7.114)
Fig. 7.2 illustrates what we have done.

We shall now perform the path integral successively over A0 and A. Because we now only integrate over gauge potentials satisfying the Coulomb gauge
condition we can use the form (7.10,7.11) of the action:
S=
d4 x
i
1h 0 2 0
A A + Ak (02 2 )Ak .
2
(7.115)
The integral over A0 gives the factor

"
1
exp i
2
(x0 y 0 ) 0
d x d y J (x)
J (y) = exp i
4|x y|
4
dt EC (t) ,
(7.116)
where EC is the Coulomb energy.

Since it is a term in the total hamiltonian, we
R
expect indeed a factor exp[i dt EC ] to be present in the matrix element Z[J] =
178
h0, |0, i[J] of the evolution operator U (, ) = T exp[i dt H(t)]. The

integral over A gives
R
1
exp
2

d xd y
Jk (x)Gkl
C (x, y) Jl (y)
(7.117)
with Gkl
C the Coulomb-gauge propagator given by
Gkl
C (x, y)
PklT (k)
d4 k ik(xy)
e
.
(2)4
k2 k02 i
(7.118)
There is the usual k 2 i in the denominator. The transverse projector appears

because we only integrate over transverse A. The answer cannot depend on
the longitudinal part of J. Defining a complete Coulomb-gauge Green function
G
C (x, y) by supplying the other , combinations as
(x0 y 0 )
=
4|x y|
k0
G0l
C (x, y) = GC (x, y) = 0,
G00
C (x, y) =
d4 k ik(xy) 1
e
,
(2)4
k2
(7.119)
(7.120)
we can summarize the result so far as

Z
1
4
4
d x d y J (x)GC (x, y) J (y) .
(7.121)
Z[J] = Z[0] exp i
2
On the other hand, making the connection with vacuum expectation values
of time ordered products of field operators, we should have
R
R 4
h0, |0, i[J]
Z[J]
=
= ei dt EC (t) h0|T ei d x JA |0i.
Z[0]
h0, |0, i[0]
(7.122)
As a check we can compare the coefficients quadratic in J with the results obtained by canonical quantization. One finds that (Problem 3)
h0|T Ak (x)Al (y)|0i = iGkl
C (x, y),
(7.123)
indeed. The FaddeevPopov method has given the same result as canonical
quantization. In this check we have to relax the condition J = 0, otherwise
we cannot compare coefficients of arbitrary Js. This means that we define Z[J]
for arbitrary J by (7.114).
We shall now show that in the physical case in which external current is
conserved, the amplitude ratio Z[J]/Z[0] is Lorentz invariant, despite its noncovariant looks. The Coulomb gauge propagator can be written in the form
d4 k ik(xy)
e
GC (k),
(2)4
NC (k)
,
G
(k)
=
C
k 2 i
k k + (kn)(k n + n k )
NC (k) =
,
k 2 + (kn)2
G
C (x, y) =
(7.124)
(7.125)
n = 0 .
(7.126)
179
7.6. GENERALIZED COVARIANT GAUGE

We see that
NC (k) = + terms k , k and k k .
(7.127)
Using k J (k) = 0 (J (k) is the Fourier transform of J (x)), it follows that

Z
d xd
y J (x)G
C (x, y) J (y)
d4 k
=
(2)4
Z
d4 k
=
(2)4
Z
J (k) G
C (k) J (k) (7.128)

J (k), (7.129)
J (k) 2
k i
which is Lorentz invariant.
7.6
Generalized covariant gauge
Although the Coulomb gauge leads to Lorentz invariant results for gauge-invariant
quantities, it is very awkward in calculations in higher orders of perturbation theory. We shall therefore make a transformation of variables in the path integral
such that we get manifestly covariant expressions.
We start with the gauge-invariant quantity [A] defined by
1 = [A]
[d] eiSgf [A ] ,
(7.130)
where Sgf [A] is a gauge fixing action given by

1 Z 4
Sgf [A] =
d x ( A )2 .
2
(7.131)
Here is a so-called gauge parameter. Rewriting the above expression in the

form
Z
R 4 2 Z
1
1
i 2
d x
[d] [ A ]
(7.132)
= [d] e
[A]
we see that here we are not really fixing the gauge but htaking an average
over coi
1 R 4
2
variant gauge conditions A = , with weight exp i 2 d x . Evaluating
the integral over in the second form (7.132) we get
Z
[d] [ A + 2 ] =
1
,
| det[ 2 ]|
(7.133)
which is independent of . So we find

[A] = | det[ 2 ]|
Z
[d] ei 2
d4 x 2
1
(7.134)
180
which is also a constant independent of A, like C . Inserting the identity (7.130)

into the integrand of the path integral for Z[J] and making a transformation of
variables A A gives
Z[J] =
=
=
[dA] [d] [ A] C eiS[A]+iSgf [A
]+i
[dA] [d] [ A ] C eiS[A]+iSgf [A]+i

[dA] eiS[A]+iSgf [A]+i
JA
JA
JA
(7.135)
(7.136)
(7.137)
We used the gauge invariance of [dA], C , , S and JA, as well as (7.107).

The action plus gauge-fixing term can be written as
S + Sgf =
1
1
d x A 2 +
2
A .
(7.138)
The differential operator in parenthesis has a Green function (for 6= ), which

we shall denote by D :
D (x, y) =
=
D (k) =
d4 k ik(xy)
e
D (k),
(2)4
!
1
2
k k k + k k D (k),
!
k k 1
k k
2
+
.
k
k2
(k 2 )2
(7.139)
(7.140)
(7.141)
This is easy to check with the help of four-dimensional transverse and longitudinal
projectors
k k
k k
PT (k) = 2 , PL (k) = 2 ,
(7.142)
k
k
with the properties
PT2 = PT ,
PL2 = PL ,
PT PL = 0,
PT + PL = 1.
(7.143)
For = 1 we get the Feynman gauge propagator

DF (k) =

,
k2
(7.144)
which is very convenient for calculations, whereas for 0 we get the Landau
gauge (or Lorentz gauge) propagator
DL (k)
k k
2
k
1
.
k2
(7.145)
181
7.7. GHOSTS
We shall assume the i prescription
k 2 k 2 i
(7.146)
in the denominators of D (k). It amounts to adding the term d4 x A A

in the exponent of the integrand in the path integral. Since 00 = 1, the
A0 A0 term has a positive sign, which is wrong for improving convergence in
the path integral (recall the motivation for introducing at the end of sect.
3.3). For convergence reasons we could change the sign of in this A0 A0 term,
but this would to spoil manifest covariance. Within perturbation theory the
prescription (7.146) leads to satisfactory results. Nonperturbatively, the lattice
regularization has been used mostly within the imaginary time formulation (where
A0 A0 A4 A4 ), and sometimes in real time in the Coulomb gauge, where the
A20 problem does not arise.
R
7.7
Ghosts
The equivalence between the Coulomb gauge and covariant gauges holds for
gauge-invariant quantities, and accordingly we assumed the current J to be
conserved. Still, having arrived at a manifestly covariant formalism we can relax
the condition J = 0 and study the correlation functions hA (x)A (y) i obtained by differentiating with respect to J and setting J = 0 afterwards. These
correlation functions are not themselves gauge invariant, but we they can be used
to construct gauge invariant observables.
We can now ask the question: is there a representation in terms of operators
A (x) such that

hA (x)A (y) i = h0|T A (x)A (y) |0i?
(7.147)
The answer is yes, but the operators act in an extended space with indefinite
metric (hence, not a Hilbert space). Consider the case = 1, for which, after
some partial integrations,
S+
=1
Sgf
1
d4 x A A .
2
(7.148)
This looks just Rlike the action of four massless scalar fields, except that the
contribution + 21 d4 x A0 A0 has the wrong sign. For a discussion of canonical
quantization with such an action see, e.g. Itzykson and Zuber. Here we shall cut
the discussion short by appealing to the following plausible outcome.
If we want to reproduce the correlator
h0|T A (x)A (y)|0i = iD (x, y) =
dk eik(xy) ,
x0 > y 0 ,
(7.149)
182
by inserting intermediate states,

Z
dk eik(xy) = h0|A (x)A (y)|0i =
XZ
dk h0|A (x)|kihk|A (y)|0i,
(7.150)
we find that in
h0|A (x)|ki = e (k, ) eikx ,
= 0, 1, 2, 3,
(7.151)
we have to allow for states with polarization = 0 and 3. This is needed to

reproduce the polarization sum
e (k, 0)e (k, 0) +
3
X
e (k, ) e (k, ) = .
(7.152)
=1
Notice that we have used a minus sign in front of the = 0 term in (7.152).
Without this sign the left-hand side would constitute a positive-definite matrix,
in contradiction with the indefinite on the right hand side. An explicit realization of the e (k, ) is given by
e (k, ) = (0, e(k, )),
e (k, 3) = (0, k),
e (k, 0) = (1, 0).
= 1, 2,
(7.153)
(7.154)
(7.155)
The sum over intermediate states in the right of (7.150) has to be re-written as
Z
dk
3
X
=0
h0|A (x)|k, ihk, |A (y)|0i.
(7.156)
Together with (7.151), hk|A (x)|0i = h0|A (x)|ki and (7.152), this reproduces
the expression on the left of (7.150). The minus sign in the decomposition of unity
1 = |0ih0| +
dk |k, 0ihk, 0| +
3
X
=1
|k, ihk, | +
(7.157)
reflects an indefinite metric in ket space (it should not be called Hilbert space
as this has positive metric),
such that
hk, |k 0 , 0 i = 0 2k 0 (2)3 (k k0 ),
(7.158)
1 |k, i = |k, i
(7.159)
(with 1 given by (7.157)), also for = 0.

The extra particle-like degrees of freedom with = 0 and 3 are called respectively timelike and longitudinal photons. These unphysical particles are
183
7.7. GHOSTS
called ghosts. The timelike photon states have negative norm. Further analysis
reveals that their contributions to gauge-invariant quantities cancel against the
longitudinal photons, and only the physical photons corresponding to = 1, 2
remain. The physical-photon states are have positive norm, as needed for the
probability interpretation of quantum mechanics. The physical Hilbert space is
a subspace of a ket space that has indefinite metric.
There are also other ghosts, named after Faddeev and Popov. They correspond to the determinant in [A] = | det[ 2 ]|. We may ignore the absolute
value symbols, since is just an A-independent constant that cannot change
sign. Using fermionic integration this determinant can be written (up to a factor) as
Z
det[ 2 ] =
[d d] ei
d4 x
(7.160)
which reveals that these ghosts are scalar fermions! For the electromagnetic field
does not depend on A and the FaddeevPopov ghosts do not interact. But in
nonabelian gauge theories [A] does depend on A and these ghosts have to be
explicitly taken into account. However, even in the electromagnetic case we have
to keep them in mind when calculating the partition function Z in the imaginarytime formalism. The FaddeevPopov determinant does contribute to Z and
compensates, for example, for the unphysical contribution of the longitudinal and
timelike photons to the specific heat.
We shall not use the operator language for gauge theories in covariant gauges.
Transition amplitudes hout|ini and matrix elements of operators will be obtained
directly from the covariant correlation functions. For the free electromagnetic
field these are given by the Wick formula
hA1 (x1 )A2 (x2 ) An (xn )i = (i)n/2 D 1 2 (x1 , x2 ) D n1 n (xn1 , xn )
+ permutations.
(7.161)
To get amplitudes out of these we define the creation and annihilation symbols
0
a(k, , x ) = A
(+)
(k, , x ) =
a(k, , x0 ) = A() (k, , x0 ) =
d3 x eikx e (k, ) (i 0 i 0 ) A (x),
d3 x A (x) (i 0 i 0 ) e (k, ) eikx . (7.162)
where k 0 = |k|, = 1, 2, or = . These are gauge invariant because they are

constructed out of the physical polarization vectors which satify6
k e (k, ) = 0,
6
k 0 = |k|,
= 1, 2, or = ,
(7.163)
In fact, k could be arbitrary here (not necessarily on shell k 2 = 0). This would do some
violence to the notation because originally the k in e (k, ) was supposed to satisfy k 0 = |k|.
But the explicit form e (k, ) = (0, e(k, )) (which really depends only on k/|k|) shows that
the extension to arbitrary k is possible. However, for A() (k, , x0 ) we only need k 0 = |k|.
184
So we can interpret correlation functions of them (possibly with other gauge

invariant quantities) as being expectation values of operators in Coulomb gauge.
However, being gauge invariant they can be directly used in a covariant gauge.
A simple example in the free theory is (cf. Problem 4)
ha(k, , x0 ) a (k 0 , 0 , x00 )i =
d3 x d3 x0 eikx e (k, ) (i 0 i 0 ) hA (x)A (x0 )i
0 0
(i 00 i 00 ) e (k 0 , 0 ) eik x
= 2k 0 (2)3 (k k0 ) 0 , x0 > x00 ,
= hk|k 0 0 i.
(7.164)
The answer is independent of the gauge parameter , at it should be.
7.8
Summary
The gauge invariance of the Maxwell action leads to complications with canonical
quantization, which can be solved by choosing the Coulomb gauge. In this gauge
the time component A0 of the gauge field is not a dynamical variable and one
more degree of freedom is lost by the vector potential satisfying the condition
A = 0. The resulting interpretation for the quantized free Maxwell field leads
to photons: massless spin-one particles with only two spin polarization degrees
of freedom. The formulation is not manifestly Lorentz covariant, but Lorentz
invariance is guaranteed by the existence of the unitary representation U (`) in
Hilbert space, with conserved Noether generators J .
Path-integral quantization is also complicated by gauge invariance, which is
taken care of with the FaddeevPopov method. Choosing the Coulomb gauge,
the path integral is found to produce the same correlation functions as found
with the canonical operator formalism.
The FaddeevPopov method allows for a very convenient transformation from
the non-covariant Coulomb gauge to a manifestly covariant gauge. A corresponding operator description is possible, at the expense of introducing ghosts, unphysical particles, and a space of state vectors with indefinite (positive and negative)
metric of which the physical Hilbert space is a subspace. The equivalence to the
Coulomb gauge with its (positive metric) Hilbert space guarantees that physical
probability is still conserved.
7.9
Appendix: Locality in the Coulomb gauge
We started from an Raction S which has nice invariance properties and is local:
it has the form S = d4 x L(x) where L(x) is a Lorentz scalar which depends on
the fields at x and in the immediate neighborhood of x (through the derivatives).
7.9. APPENDIX: LOCALITY IN THE COULOMB GAUGE
185
This leads to covariant and local classical equations of motion. No signals can
travel faster than the velocity of light. Upon quantization in the Coulomb gauge
we have ended up with non-Lorentz and nongauge invariant expressions which
furthermore look terribly nonlocal: the projector
T
Pmn
(x y) = mn (x y) + m n
1
4|x y|
(7.165)
drops off only like a power of |x y| for large separations.

An important expression of locality and Lorentz invariance is the following.
Two observables O1,2 associated with compact spacetime regions R1,2 (local observables) commute, when all points x1 R1 are spacelike to all points x2 R2 .
In the standard lore of quantum mechanics observables correspond to measurements, and measurements in spacelike separated regions should not be able to
influence each other. Observables have to be gauge invariant. An example is
given by the field strength F (x). Locality is expressed by
[F (x), F (y)] = 0, (x y)2 > 0.
(7.166)
This is indeed the case as will now be shown for the case of vanishing external
current.
Using the expansion
A (x, t) =
XZ
dk [eikx e (k, )a(k, ) + h.c.]
(7.167)
(recall that in Coulomb gauge e (k, ) has no time component and is restricted
to the values 1,2), gives
[A (x), A (y)] =
dk (eik(xy) eik(xy) )PC (k),
(7.168)
where (recall n = ,0 )
PC (k) =
e (k, )e (k, )
k k + (kn)(k n + n k )
(kn)2
2
= NC (k), k = 0.
(7.169)
(7.170)
Working out the derivatives in F = A A we get

[F (x), F (y)] =
dk {eik(xy) [k k PC (k) k k PC (k)
k k PC (k) + k k PC (k)] (k k)}. (7.171)
186
Now the operation of the curl in F projects to zero any 4D longitudinal part
k in PC , such that only the part of PC contributes. In position space
we can then write
[F (x), F (y)] = (
+ ) i(x y),
(x y) = i
dk (eik(xy) eik(xy) ).
(7.172)
(7.173)
The (generalized) function (x) (which we have met before in the scalar field
case) has the following properties:
- (x) is Lorentz invariant, (`x) = (x),
- (x) = 0 for x0 = 0, x 6= 0.
Since x = (x, 0) is spacelike and (x) is Lorentz invariant it follows that it
vanishes for general spacelike distances,
(x y) = 0,
(x y)2 > 0.
(7.174)
(It is also interesting to note that (x) is the solution of 2 (x) = 0 with initial
conditions (x) = 0, 0 (x) = (x) at x0 = 0.)
Consequently the field strengths and all local observables that can be made
out of these have the locality property (7.166).
7.10
Problems
1. Expressing P in terms of as and a s.

To obtain the expressions (7.57) for the hamiltonian, we insert (7.43) into
(7.31), using (7.47):
Z
1 X 1
1
H =
d3 x [
2
0
0
V k,l 4k l
2
(ik 0 eikx am (k) + ik 0 eikx am (k) )(il0 eilx am (l) + il0 eilx am (l) )
+ (eikx am (k) + eikx am (k) ) l2 (eilx am (l) + eilx am (l) )]. (7.175)
The integration sets l = k and the aa and a a terms cancel (k 0 = |k|),

leaving
H =
1 X 1 01
k [am (k)am (k) + am (k) am (k)].
V k 2k 0
2
(7.176)
187
7.10. PROBLEMS
We can now convert to a(k, ) or use the commutation relation (7.53) diP
P
T
(k) = 2 = to put a to the left of a,
rectly with m Pmm
am (k)am (k) = am (k) am (k) + 2k 0 V
(7.177)
This gives (7.31) after converting to a(k, ). This calculation of the hamiltonian is basically the same as for the one dimensional harmonic oscillator.
The calculation of the momentum operator (7.32) proceeds in similar fashion,
Z
1 X 1
P = 2
d3 x [
0
0
V k,l 4k l
(ik 0 eikx am (k) + ik 0 eikx am (k) ) il (eilx am (l) eilx am (l) ]

1 X 1 1
=
k [am (k)am (k) + am (k) am (k)
0
V k 2k 2
+ am (k)am (k) + am (k) am (k)],

1 X 1
=
k am (k) am (k)
V k 2k 0
(7.178)
(7.179)
(k is odd under k k, such that accompanying factors even under

k k do not contribute). Expressing the result in terms of a(k, ) gives
(7.32).
2. Coulomb gauge Green function
Verify that G
C (k) in the form (7.126) satisfies the Coulomb gauge condition
m
km GC (k) = 0, and for a conserved current
K (k)G
C (k) J (k) = J (k).
(7.180)
Hence, eq. (7.105) is satisfied as well and although G

C (x, y) is not the
inverse of K it is a good Green function in this restricted sense.
3. Checking (7.123)
Verify (7.123) by expanding the left and right hand sides up to terms
quadratic in J. Choose time orderings and evaluate the right hand side
by contour integration over k 0 . On the left hand side insert the resolution
of unity
XZ
1 = |0ih0| +
dk |kihk| + ,
(7.181)
and use the formulas obtained by canonical quantization.
4. Amplitudes from correlation functions in covariant gauge.

Using the Wick formula (7.161), verify (7.164).
188
Chapter 8
QED
Two concepts have proved very successful in the development of quantum field
theory of elementary particles (in addition to locality and Poincare invariance):
gauge invariance and renormalizability.1 With gauge invariance we mean local
symmetries, i.e. symmetries in which the transformations may vary from point to
point in space-time. Renormalizability the property that ultraviolet infinities
can be absorbed in the bare parameters of the model gives strong restrictions
on the possible actions: they have to be polynomial in the fields,2 with coupling
constants that can only have dimension 0 in units of mass. Quantum electrodynamics (QED) is the simplest example of a nontrivial gauge theory. In a wide
sense it is the theory of electromagnetic interactions of all things, but we understand it here to be spinor electrodynamics (an even simpler version is scalar
electrodynamics).
8.1
Gauge invariance
We have seen that the Lagrange density for the Dirac field,
L(x) = (x)(m
+ )(x),
(8.1)
has a U(1) invariance,

0 (x) = ei (x),
0 (x) = ei (x).
(8.2)
This does not mean that nonrenormalizable models are not useful. Such models have a
regulatorization scale that cannot be removed in perturbation theory, indicating the need of
new physical input at that scale. When we increase the regulatorization scale, its influence on
the physics diminishes and only renormalizable interactions remain important. Renormalizable
models come in universality classes much like critical phenomena.
2
This holds in the formal continuum presentation. With an explicit regularization such as
the lattice the action can be an infinite series with increasing powers of the fields.
189
190
CHAPTER 8. QED
In terms of the Majorana fields 1,2 defined by

1
= (1 i2 ),
2
(8.3)
this is an SO(2) rotation

10 (x)
20 (x)
cos sin
sin cos
1 (x)
2 (x)
(8.4)
according to the equivalence U(1)'SO(2). We can interpret these transformations also as passive transformations of the coordinate system in an internal
two-dimensional space.
The transformations are global in the sense that the angle does not depend
on the spacetime coordinate x. It is natural to ask if the reference system that
picks out the real and imaginary parts of , or equivalently its components 1
and 2 , has to be globally defined. For example do we have to choose a reference
frame3 in Amsterdam now the same as somewhere in a far away galaxy five years
later? This seems unnatural, and so requiring the parameter of a symmetry
transformation to be the same all over the universe, for all times, seems unphysical. So one is led to consider transformations with depending on the spacetime
point x. How rapidly (x) may vary as a function of x is not clear at this point.
A natural scale of variation is the Compton wavelength 1/m, which is very small
in human perception. At this point it is simplest to assume that (x) should be
allowed to vary without restriction.
Hence, we want to construct an action invariant under U(1) transformations
in which the angle (x) depends on spacetime. To achieve this we need to compensate the noninvariance of the derivative terms in the Lagrange density (8.1).
Under a local transformation
0 (x) = ei(x) (x),
0 (x) = ei(x) (x),
(8.5)
the term m(x)(x)

is invariant, but the derivative transforms in an inhomogeneous and noncovariant way
0 (x) = [ei(x) (x)] = ei(x) [ (x) + i (x)(x)].
(8.6)
Instead, a covariant derivative D transforming as

D0 (x)0 (x) = ei(x) D (x)(x),
D0 (x) 0 (x) = ei(x) D (x)(x),
(8.7)
would allow for the construction of an invariant Lagrange density

L(x) = (x)
m + D (x),
3
D =
1
1
D D .
2
2
The reasoning here is familiar from in the theory of gravity.
(8.8)
191
8.1. GAUGE INVARIANCE
The wellknown construction of such a covariant derivative uses the invariance of

the electromagnetic field system under the gauge transformation
1
A0 (x) = A (x) + (x)
e
(8.9)
(e is here an arbitrary constant). For the form

D (x) = ieA (x),
(8.10)
has the required property: under the combined gauge transformation (8.5), (8.9),
D0 (x) 0 (x) = [ ieA0 (x)] 0 (x)
= [ ieA (x) i (x)] [ei(x) (x)]

= ei(x) [ ieA (x)] (x)
= ei(x) D (x) (x),
(8.11)
For we have by conjugation
D = + ieA D ,
D0 0 = ei D .
(8.12)
A derivative involves the comparison of fields at infinitesimally close points in

spacetime. The electromagnetic potentials play the role of a connection, which is
used in comparing (connecting) the orientations of the internal spaces at these
infinitesimally close points.
The classical action for the combined electromagnetic and Dirac field system
is now given by
1
+ D )
S =
d x F F (m
4

Z
1
4
=
d x F F (m + ) + ei A .
4
(8.13)
(8.14)
We see in (8.14) an interaction term of the form d4 x ej A familiar from classical

electrodynamics, with electromagnetic current e times the Noether current (5.79):
R
.
j = i
(8.15)
This current is also a Noether current of the action (8.14) corresponding to its
global U(1) invariance. The construction of a gauge invariant action has led to a
specific interaction between the electromagnetic and spinor fields. The parameter
e parametrizes the strength of an interaction term trilinear in the fields. It is
evidently a coupling constant, and it is dimensionless, [e] = 0.
As suggested by the notation, e is also the elementary unit of charge. This
can be seen by the following argument. Suppose A is not dynamical but a
192
CHAPTER 8. QED
given external electromagnetic field. Then the action is bilinear in the dynamical
variables (the fermion fields), so the system is like a free field and its quantum
version is relatively simple. Suppose furthermore that the fermions are protons.4
Consider now the operator for electric charge
Z
d3 x ej 0 (x) = eQ,
Q =
XZ
(8.16)
h
dp b (p, , t)b(p, , t) d (p, , t)d(p, , t) , (8.17)
where we used (6.42). Because A may depend on time we have kept the time
dependence in the creation and annihilation operators (this similar to the case
of the scalar field in sect. 2.10). On the other hand the Noether argument still
shows that Q is time independent and it is reasonable to interpret it as counting
the number of particles minus antiparticles at time t. It follows that the electric
charge of an eigenstate of Q is e times an integer, so e is the elementary charge
unit, and e > 0.
For electrons and positrons we replace e e, because conventionally the
electrons are the particles and the positrons the antiparticles, such that the
and the electric charge operator is eQ.
electromagnetic current is ei
Denoting the charge of the particles (proton, electron, . . . ) in units of e by q,
we have in general
D = ( ieqA ) , D = ( + ieqA ) ,
1
+ D )
L = F F (m
4
1
+ ) + eq i
A .
= F F (m
4
(8.18)
(8.19)
(8.20)
So q = 1 for the protons, q = 1 for electrons, q = 2/3 for the up-quark, etc.
It is of course possible to write down other gauge invariant interactions, for
example
F ,
La = S
(8.21)
which corresponds to an explicit anomalous magnetic moment interaction (S =
i[ , ]/4, cf. sect. 8.4). The coupling constant has dimension 1 in mass
units, and this interaction is not renormalizable. Assuming renormalizability,
we discard it. The theory of quantum electrodynamics described by (8.20) is the
most general form (conserving parity) with coupling constants of mass dimensions
0 and it is renormalizable. We take it as the best possible starting point for a
fundamental theory of only photons and electrons. The theory then predicts the
magnetic moment of the electron (as well as a host of other physical properties).
4
This is an approximate description, we know of course that in a more fundamental theory

protons are to be described in terms of quarks and gluons.
193
8.2. SPINOR ELECTRODYNAMICS
Let us also note at this point that the gauge group here is not U(1) but
the group of real numbers R: in A A + 1e can be any real number.
An important consequence of the group being R is that q does not have to be
an integer: it can be any real number. If the gauge group were U (1), then
U = exp(i) U q = exp(iq) would only be a representation if q where integer,5
which would imply charge quantization.6
8.2
Spinor electrodynamics
We continue with the theory of photons, electrons and positrons specified by

the lagrange density (8.20) (spinor electrodynamics) and quantize it covariantly
using the path integral methods described in the previous chapter. The previous
reasoning goes through with minor modifications. Let us present it here again
be a gauge invariant functional made
in a slightly modified form. Let F [A, , ]
out of the fields and hF i be its path integral average (e.g. F could be the field
strength or the electromagnetic current, or hF i could be a transition amplitude
hout|ini). The line of argument is now given by
1 = [A]
[d] e
iSgf [A ]
Sgf [A] =
d4 x
1
( A )2 ,
2
[dA dd]
eiS[A,,] F [A, , ]
R
[dA dd]
eiS[A,,]
R
[dA dd]
[d] [A] eiSgf [A ]+iS[A,,] F [A, , ]
=
R
[dA dd]
[d] [A] eiSgf [A ]+iS[A,,]]
?
hF i =
[d] [dA dd]

[A] eiSgf [A]+iS[A,,] F [A, , ]
R
=
R
[d]
[dA dd]
[A] eiSgf [A]+iS[A,,]
R
[dA dd]
[A] eiSgf [A]+iS[A,,] F [A, , ]
.
hF i
R
[dA dd]
[A] eiSgf [A]+iS[A,,]
R
(8.22)
(8.23)
(8.24)
(8.25)
(8.26)
We made a transformation of variables looking like the gauge transformation
A = A0 , = 0 , = 0 , and used the gauge invariance of S, F , [dA dd],

and . The last line is the motivation for the path integral with sources, which
is not gauge invariant, but out of which gauge invariant hF is can be constructed,
Z[J, , ] =
[dA dd]
eiS+iSgf +i
d4 x (JA+
+)
(8.27)
We recall that is a constant in QED.

5
Otherwise U q would not be single valued.

We do indeed see something similar to charge quantization in Nature, which can be understood if the electromagnetic gauge group is a subgroup of a semisimple nonabelian group, e.g.
in a Grand Unified Theory.
6
194
CHAPTER 8. QED
= -eq ( )
ab
= -i S ab (p)
= -i D
(k)
k
Figure 8.1: Propagators an vertexfunctions in spinor electrodynamics.
Correlation functions can now be defined by differentiation with respect to
the sources and they can be calculated in perturbation theory in much the same
fashion as in the Yukawa model introduced in section 6.5. The interaction is
S1 =
A + S,
d4 x eq i
(8.28)
where S contains counterterms corresponding to mass, charge and wave function renormalization. This will not concern us here as we shall limit ourselves to
the tree graph approximation, so S 0 in the following.
The ingredients of the diagrams are shown in Fig. 8.1. There is only one bare
vertex function, which is in full glory given by
Sa b A (u, v, w) =
d4 x ieq( )ab 4 (u x) 4 (v x) 4 (w x),
(8.29)
such that
S1 =
d4 u d4 v d4 w a (u)Sa b A (u, v, w)b (v)A (w),
(8.30)
and a momentum space version which is momentum independent,

Sa b A (p, q, k) = ieq( )ab ,
(8.31)
195
8.2. SPINOR ELECTRODYNAMICS

The propagators are given by
m ab ip ab
,
Sab (p) =
m2 + p2 i

k k
D (k) = 2
+ ( 1) 2
.
k i
(k i)2
(8.32)
(8.33)
We now turn to the calculation of scattering processes in the tree graph approximation. Convenient formulas for obtaining the objects (such as u(p, ), etc.)
associated with external fermion or scalar lines in scattering diagrams have already been given in section 6.6. These are also needed for external photon lines.
We repeat the expressions for creation and annihilation of photons introduced
earlier in (7.162),
0
a(k, , x ) =
a (k, , x0 ) =
d3 x eikx e (k, ) (i 0 i 0 ) A (x),
d3 x A (x) (i 0 i 0 ) e (k, ) eikx .
(8.34)
where k 0 = |k|, = 1, 2, or . In terms of these we have in the free theory (cf.

Prob. 7.4)
hA (x) a (k, , )i0
hA (x) a (k, , +)i0
ha(k, , ) A (x)i0
ha(k, , +) A (x)i0
=
=
=
=
e (k, ) eikx ,
0,
0,
e (k, ) eikx.
(8.35)
(8.36)
(8.37)
(8.38)
The steps given in sect. 6.6 can be followed almost verbatim here as well. Problem
2 deals with photon-electron scattering (called Compton scattering). Note the
remarks made in Problem 2.d. Electron-electron scattering is the subject of
Problem 3.
Another example is electron-positron scattering, the diagrams of which are
shown in Fig. 8.2. The scattering amplitude T is defined in terms of the transition
amplitude (for unequal initial and final states) as
hp3 3 , p4 4 out|p1 1 , p2 2 ini
i(2)4 4 (p1 + p2 p3 p4 ) T (p3 3 , p4 4 ; p1 1 , p2 2 )
= hd(p4 , 4 , ) b(p3 , 3 , ) b (p1 , 2 , ) d (p2 , 2 , )i.
(8.39)
Note the convention of not changing the order of the labels in the bra, compared
the ket, and likewise in T , whereas upon conjugation the actual order of bs
and ds is interchanged in (8.39). Since we are dealing with fermions, this order
matters in getting the right sign. The scattering amplitude is given by
iT (p3 3 , p4 4 ; p1 1 , p2 2 ) =
u3 e u1 v2 e v4 (i)D (p1 p3 ) (8.40)
+u
3 e v4 v2 e u1 (i)D (p1 + p2 ),
196
CHAPTER 8. QED
p4 4
p2
p 4
p 2
p -p 3
1
+
p
p +p
p 3
p
1 1
p3
1 1
+
Figure 8.2: Diagram for the process e
1 + e2 e3 + e4 .
where u
3 = u
(p3 , 3 ), etc. The spin averaged squared magnitude |T |2 determines
the unpolarized cross section. The interference of the two terms makes this
calculation less simple than one might wish. In the next section we shall do the
calculation for a simpler situation which does not have such interference.
8.3
Example: e + e+ + + scattering
A simple example for working out the unpolarized cross section is the process
e + e+ + + . The muon mass M 106 MeV, much larger than the electron mass m 0.51 MeV. For the process to take place, the center-of-mass energy
has to be sufficiently high to be able to create the muon-antimuon pair. Similar processes in which quark-antiquarks pairs are produced (which subsequently
transform into hadrons) have played an important role in establishing QCD as
the theory of the strong interactions, and have led further to the construction of
the Standard Model.
We introduce fermion fields for the muon as well as for the electron and the
action is given by
S =
1
d x F 2 (e) (m + ) (e) () (M + ) ()
4
i
(e) (e)
e i A e() i () A .
(8.41)
Because the electron and muon fields are independent there is only one relevant
diagram, shown in Fig. 8.3. Then the invariant scattering amplitude is given by
T ( (3), + (4); e (1), e+ (2))
= e2 u
(p3 , 3 ) v(p4 , 4 ) v(p2 , 2 ) u(p1 , 1 )
(8.42)
"
k k
+ ( 1) 2 2 ,
2
k
(k )
with k = p3 + p4 = p1 + p2 . The gauge terms k k in the photon propagator

do not contribute because of current conservation (cf. Prob. 3 in chapter 6). For
8.3. EXAMPLE: E + E + + + SCATTERING
p 4
197
p 2
p +p
1
p3
1 1
+
Figure 8.3: Diagram for the process e
1 + e 2 3 + 4 .
example,
u
(p3 , 3 )i v(p4 , 4 ) (p3 + p4 ) = 0,
(8.43)
where we used the fact that the polarization spinors satisfy the (momentum space
version of the) Dirac equation,
u
(p3 , ) ip3 = M u
(p3 , 3 ),
ip4 v(p4 , 4 ) = M v(p4 , 4 ).
(8.44)
Thus we reach the important conclusion that the scattering amplitude is independent of the gauge parameter , as it should be.
To calculate the cross section we need T , which leads to
v 4 u3 ,
(
u3 v4 ) = v4 u3 =
(
v 2 u1 )
u1 v2
=
u 1 v2 .
(8.45)
(8.46)
Averaging over initial and final spins gives

|T |2 = e4
1 X

u
3 v4 v2 u1 v4 u3 u
1 v2
,
16 3 1 2 4
s2
(8.47)
where s = (p3 + p4 )2 = (p1 + p2 )2 is one of the Mandelstam variables (equal

to the total cm energy squared). To evaluate the polarization sums we order the
spinor factors in a suggestive way, interpreting ua u
b and va vb as matrices and
using, for example,
u
3 v4 v4 u3 = Tr [ v4 v4 u3 u
3 ].
(8.48)
Then
|T |2 = e4
X
X
1
u3 u
3 )]
v
v
)
(
Tr
[
(
4
4
s2 16
3
4
Tr [ (
X
1
u1 u
1 ) (
X
2
v2 v2 )].
(8.49)
198
CHAPTER 8. QED
We now use the properties

X
4
X
1
v(p4 , 4 )
v (p4 , 4 ) = (M + ip4 ),
(8.50)
u(p1 , 1 )
u(p1 , 1 ) = m ip1 ,
(8.51)
etc. and obtain the form

1
Tr [ (M + ip4 ) (M ip3 )]
s2 16
Tr [ (m ip1 ) (m + ip2 )].
|T |2 = e4
(8.52)
To evaluate this we use the trace formulas

Tr = 4 ,
Tr = 0,
Tr = 4( g + ).
(8.53)
(8.54)
(8.55)
These follow from the fact that (1) the trace of a product of gamma matrices
vanishes unless each 0 , . . . , 3 appears an even number of times, (2) 02 = 1,
32 = 12 = 22 = 1, (2) the gammas anticommute and (4) Tr 1 = 4. More
identities are in Appendix 5.7. The two traces in (8.52) are given by
4(M 2 + p4 p3 p3 p4 + p3 p4 )
(8.56)
4(m2 + p1 p2 p1 p2 + p2 p1 ).
(8.57)
and
The evaluation of |T |2 is now straightforward and results in a large number of
scalar products of the momenta. Using the Mandelstam variables
s = (p3 + p4 )2 = 2M 2 2p3 p4 = (p1 + p2 )2 = 2m2 2p1 p2 ,
(8.58)
t = (p3 p1 )2 = m2 M 2 + 2p3 p1 = (p2 p4 )2 = m2 M 2 + 2p2 p4 , (8.59)
u = (p3 p2 )2 = m2 M 2 + 2p3 p2 = (p1 p4 )2 = m2 M 2 + 2p1 p4 , (8.60)
the result simplifies to

|T |2 =
e4
[4m2 M 2 + M 2 (s 2m2 ) + m2 (s 2M 2 )
s2
1
1
+ (t + m2 + M 2 )2 + (u + m2 + M 2 )2 ].
2
2
(8.61)
We recall that u can be eliminated in favor of s and t by the relation s + t + u =

2m2 +2M 2 . At high energies we can neglect the electron and muon masses. Then
|T |2
e4 2
2 (t + u2 ).
2s
(8.62)
8.3. EXAMPLE: E + E + + + SCATTERING
199
Under these cricumstances t and u are related to the scattering angle in the centre
of mass frame by
1
1
t s (1 cos ), u s (1 + cos ),
2
2
(8.63)
and we get for the differential crossection at high energies

"
d
d
1 |p3 |
4|T |2
64 2 s |p1 |
2
e2
(1 + cos2 ), =
,
4s
4
=
cm
(8.64)
(8.65)
where is the finestructure constant. The factor 4 in front of |T |2 represents the

summation over final spins. The total cross section is given by
2
= 2
4s
1
1
d cos (1 + cos2 ) =
42
.
3s
(8.66)
It is instructive to rederive these formulas by evaluating first the high energy

form of T for given polarization combinations using helicity spinors, and from
this |T |2 . See e.g. Peskin & Schroeder chapter 5.
We shall now briefly review the relation with hadron production. The total
cross section (e+ + e hadrons) goes via the intermediate production of
quark-antiquark pairs, which subsequently transform into hadrons. The quarks
come in three colors for every flavor f = u, d, s, c, b, t (i.e. up, down, strange,
charm, bottom, top). Their effective masses are of the order of mu md 0.3
GeV, ms 0.4 GeV, mc 1.4 GeV, mb 4.1 GeV and mt 175 GeV. At
energies much higher than the thresholds for production of the quark-antiquark
bound states and resonances, the ratio of the cross section into hadrons is simply
given by
X
(e+ + e hadrons)
=
3
qf2 + corrections,
(8.67)
R
+
+
(e + e + )
f
where the sum is over the contributing flavors and the factor 3 is the number of
colors. The electric charges of the quarks are given by
2
qu = q c = q t = ,
3
1
qd = q s = q b = .
3
(8.68)
The corrections (mainly due to QCD interactions) are reasonably small at energies
above a few GeV; at higher energies electroweak radiative corrections become also
important.
Experimentally one finds (cf. Fig. 8.4) in the region leading up to the charmanticharm threshold at 3 GeV, modulo the corrections,
R3
4 1 1
+ +
= 2,
9 9 9

200
CHAPTER 8. QED

!"#%$&
"$(')*"!++
,.-/10*230%46587:99-<;=-7%/>;
6
5
I
{MARK
MARK I/LGW
MEA
4
R
3
2
J/(1S)
(2S)
3
(nS)
n= 1 234
10
AMY
CELLO
CLEO
15
CRYSTAL BALL
CUSB
DASP II
20
25
30
JADE
LENA
MAC
35
TASSO
TOPAZ
VENUS
MARK J
MD1
PLUTO
40
45
50
55
60
Ecm (GeV)
?@ ABC<DEFG HJIKMLONP NQ+RSNTVU:NWOXSYJZSNU:N[\R]X_â`_bdcfe=g hai_hOjlknmWOToZ]Ô[X]pSq<egrhai_hOjlknsi_s=j p tu mJNZ]NvRSmJNwWa[J[Jx mJx P WaRSx Ô[:x [yRSmJN*[oYJU%NZWaR ÔZ{zJZS^JQ+N+NTJX{|\x W
Ô[JNvzJmJÔRSâ[}ÔZ{|\x W*RSmJN%~{NWOXSYJZSNU:N+[oR]Xx [}RSmJN|ox Q+x [Jx Rrw^<`=RSmJN%~U}WOXSX WaZSNvXSmJ^<u [}x [}R mN` ÔP P ^<u x [JOYJZ]N<mJN*ToN[JÔU%x [WaR ÔZ_x X_RSmJN*Q+WaP QYJP WaR NTy _
XSx [JOP N+rzJmJÔRSâ[}zJZS^JQ+NX]X+XSN+N*RSmJN*XSNQ+RSx Ô[VÔ[Z]ÔX]X SLONQ+RSx Ô[%ÔZSUwYJP WaN{` ÔZLOzNQ+x QZ]ôQ+NXSXSNX_vWaTox WaR x |ONQ+ÔZ]ZSNQ+RSx Ô[XWa[Tu mJNZ]Nx U%zÔZSR]Wa[\R+QÔZ]Z]NQ+RSx Ô[X
` ÔZ%Rruâ zJmJÔRSÔ[&zJZ]ôQ+NXSXSNXwW<[T&zJZ]ôToYQRSx â[&mW|ONVNN[U:WOToNa âRSNVRSmWaR%RSmJN w TJWaRSWg >Wa[T&y vp*x X!` ÔZy3>mWOToZSÔ[X+{mJNyzÔx [\R]X
x [RSmJN}{groa<op{Z]NOx Ô[WaZSN%` ZSÔURSmJN:> r=NWOT&P WOX]XvW<P PN+ozNZSx U%N[\R+{^zJZ]NXSNZS|ON:Q+P WaZ]x RrÔ[JP WyZ]N+zJZ]NX N[oR]W<RSx |aN%XSYJXSN+R!â`{RSmJNVW|<Wax P WaJP N
U:NWOXSYJZ]N+U:N[\R]X x XXSmJ^<u{[oZSN+` NZ]N+[QNX{RS^VWOTJTox RSx Ô[WaPTJWaR]W}WaZ]N!x [Q+P YToNTyNP âu{{P XS^:` ÔZvQP WaZSx RraX ÔU:N*zÔx [\R]XmW|ON*NN+[Q+ÔUwJx [JNTVÔZXSmJx ` R NT>X P x Om\RSP
g]lJwpx [ !_Wa[TX ÔU%N:zâx [\R]Xvu{x RSmP ^<uXSRSWaRSx XSRSx QWaPXSx O[Jx Q+Wa[Q+NwmW|ON%NN[ÔU:x RSRSNTLOJX RSNU:WaRSx Q![JÔZ]U:WaP x WaRSx Ô[NZSZ]ÔZ]XvWaZSN:[JÔR*x [Q+P YToNTRSmJN
ZWa[JON{` ZSÔUJ<o%JToNzN[Tox [J!Ô[VN+ozNZ]x U%N[oRN!Q+WaYJRSx Ô[}R mWaR NX zNQ+x WaP P !RSmJNvÔP ToNZNJzN+Z]x U:N[\R]X_R N[T}RS^%mW|ONP WaZSON{[JÔZ]U:WaP x W<RSx â[wYJ[Q+NZ]RSWax [\RSx NX+
ÔRSN:RSmJN}XSYJzJzJZ]NX]XSNTNZS^J{mJN}mJÔZSx +Ô[oRSWaP#N+oRSN[\R*â` RSmJN:zJP âR!X oUwÔP XmWOXv[J^X x a[Jx QWa[QNa mJN:z^\X x R x Ô[Xâ`{R mNyq{gSOpS={g \pSW<[TRSmJN%` ÔYJZ
P ^<uNXSRl|aNQ+RSÔZ rU%NXSÔ[:ZSNXSÔ[Wa[Q+NX{WaZSNvx [Tox Q+WaRSNT{u^:QYJZ]|ONX WaZSNv^<|ONZ]P Wax Tw` ÔZ 3(O%N+*X mJ^<u{x [JwRSmJN*RSmJNÔZSNRSx QWaPzJZ]NTox Q+RSx Ô[:` ÔZb!x [Q+P YTox [J
mJx amN+ZÔZToN+Z!*#1 vx [JN%Wa[T#L\WazJx ZX RSNx [mooX+N+|=NRSR+E=Oag <o<op Wa[TNP NQ+RSZS^<uNWaVQ+ÔZSZ]NQRSx â[X mNwl|aW<P YJNXWaZ]N!` ÔZ*%W|OÔZX{x [
RSmJN X]Q]mJNU:NWa[TyWaZSNv {] f O*>N+g P ^<uNZ{QYJZ]|ONapWa[TV {] f Oa*>N+g YJzJzN+ZQYZS|ONapSgSÔYJZ]RSNXS:â`=J_ÔZSRSNZ+<O\\ p!N+` NZ]N[QNXg x [Q+P YTox [J
XSN|aNZWaPZ]N+` NZSN[Q+NX_RS^}TJW<R]Ww[JÔR WazJzNWaZSx [J!x [VR mNOYJZ]NWa[TyXSâU:NZSN+` NZ]N+[Q+NXR ^:zJZ]N+P x U%x [WaZSwTJW<R]WopS
VlJÔZ]x_a Jm\JXo=NR R+%HoJ\O:gS<aOopS
w_ J! o{NmJZ]N+[Ta Jm\JXJ=N+RSR+Ho%ao%gSOaJpS
Wa[T:! rJ NmJZ]N+[T_a omooX+o=NR R+HJE%J\O:gS<O\OpS
w_ J!{x P NXa omooX+oN|Vaa\wgS<OaJp
Wa[T:!{NXSXSÔ[+< Jm\oX+oN+|J=NR R+JOowg <O\\pS
wV %Jox Q+N%a om\JX+oN|J=NRSR+OO:gS<OO\pS
w ! _ vXSRSNZSmJNP T>_a La## ro<Oo
Wa[ToW<oYJ^<u{XS\x+_a Jm\JX \:g <OOopS
w J!J ZWa[ToNP x y_a Jm\JXo=NR R+FIwO\%gS<oopS
w 3 JmooX+\=NRSR+HHoI%OO:g <O\\pS
v v^\XSU%N%a oYQ+P omooX+:H\a%gS<\<opS
Vy3oo{ôQ]>_a g La vWaU!YJZSa N+x ToNP NZSa
y_YJ [QmJN[>_âP P Wa p JmooX+ Iaa%gS<OO\pS
# v {WOQQ+x#_a Jm\JXo=N+RSR+Iwaa}gS<\<opS
V owJ{N[ToNZ*_a Jm\JXoN|VEHJwgS<O\\p
y J{WaZ]RSN+P_+< Jm\oX+o=NRSR+HJ%owgS<OOop
Wa[T}lW<Z]RSNP#_a Jm\JXo=NR R+HJI%OowgS<a\\pS
_ yJvox Q+ozÔZ]YJ_a omooX+ HJaO:gSO\\pS
d JoNZ][Wa[ToN%_a om\JXJN|VEHaa\wgS<O\OpS
} #JToN|aWV_a Jm\JXoN|J=NRSR+O:gS<OaopS
d
Wa[T:vToN|<W+a JmooX+oN|VEJOowg <OOopS
d } _LOx NOZSx X R_a om\JXJN|yIOO%gS<O\\p
d }w D\J: Jv3o r# vJvW<zx Tox Xa
mooX+oN|o=NR R+EOa%gS<oOOp Wa[T: vJWazJx Tox X+oRSmJNXSx X+
La## NzÔZ]R Oa%gSo<opS
d }< JJ_WaRSZSx Q]\m wRSmJNXSx X+\ ]oaOwgS<O\\p
3%Hv J{P x [J^<|y+_a Jm\JX FowgS<OO\pS
yo X z^\X x R ^V_a J=NRSRJY^<|O^:_x U:N+[oRS^HJ\wgS<oOOp
*_}w vJ*WO Q]ONZoRSmJNXSx X{NXSWaUwRSmJôQ]mXSQ]moYJP NvLOx NON+[
LadO<]aO<OwgS<oaOpS{{NZ]ONaJRSmJNX x XovWaUwJYJZ]*[Jx |gS<o<\pS
_m {NZSONZva JmooX+\=NRSRHJ%JJ<%gS<o<op
Wa[T}=WOQ<WOXJRSmJNX x Xo=f_WOQ]mJN[JLaP YJRS^< oOOa*gSOoOpS
== J*J{Z]Wa[ToNP x ya JmooX+o=NR R+HHJE%OO:gS<OO\pS
Wa[T}{P RSmJâ+_a m\oX+o=NRSR+HoE%JOJ!gS<OaJp
%: y_+TJWOQ]mJx_a Jm\JXJN|o=NRSR+IowgS<OaopSWa[T
y J!J_^\X mJx TJWV+_< omooX+o=NRSR=How\:gS<OoapS
Figure 8.4: Experimental results for R (from the web site of the Particle Data
Group, see http://pdg.lbl.gov/2000/contents plots.html).
8.4. MAGNETIC MOMENT OF THE ELECTRON
201
then, after passing through the c

c resonance region leading up to the bottomantibottom threshold at 9 GeV,
10
4
R 2+3 = .
9
3
Beyond the upsilon resonance region R settles at about
R
1
11
10
+3 = ,
3
9
3
presumably untill the top-antitop The ratio R has given important confirmation
for the color degree of freedom of the hadron constituents (i.e. the quarks): without it theory and experiment would disagree by about a factor 3. For further
discussion see Peskin and Schroeder chapter 5, De Wit & Smith ch. 6, Brown
sect. 8.2,
8.4
Magnetic moment of the electron
In non-relativistic quantum mechanics, an electron in an external static electromagnetic potential is described by the hamiltonian
2 + e[
] + e2 A2 (
p
p A(
x) + A(
x) p
x)
eg
x),
H=
eA0 (
x) +
S B(
2m
2m
(8.69)
the spin operator,

is the position operator, p
the momentum operator, S
where x
g is the Lande g-factor (gyromagnetic ratio) and B = A is the magnetic field.
A(
+ e2 A2 (
The terms p
x) + A(
x) p
x) come from the minimal substitution
2 (
p
p + eA(
x))2 . It will be shown in this section that in the approximation
where (8.69) is valid, spinor electrodynamics predicts g = 2.
We first derive the form of the non-relativistic H in the momentum representation and then identify the corresponding form in spinor electrodynamics. Using
momentum states with the normalization
hp0 0 |pi = (2)3 (p0 p) 0 ,
(8.70)
hx0 |pi = 0 eipx .
(8.71)
we have
The momentum representation of H is given by
hp0 0 |H|pi
=
p2
e
0 p) 0
(2)3 (p0 p) 0 +
(p0 + p) A(p
2m
2m
e2 0 0 2
+
hp |A (
x)|pi
2m
eg
0 p),
eA0 (p0 p) 0 +
0 B(p
(8.72)
4m
202
CHAPTER 8. QED
where = (1 , 2 , 3 ) is the vector of Pauli matrices and we used

0 0
hp |A (
x)|pi =
XZ
00
d3 x hp0 0 |x00 i hx00 |pi A (x)
= 0 A (p0 p),
A (k) =
(8.73)
(8.74)
d3 x eikx A (x).
(8.75)
The Fourier transform of the magnetic field is related to that of the vector potential by
k (k) = ilmn km An (k).
B
(8.76)
In spinor electrodynamics, the approximation where (8.72) is valid, is the
non-relativistic approximation in which radiation effects corresponding to the
quantized photon field are neglected. So we consider the electron field in an
external static electromagnetic potential A (x). The hamiltonian of this system
can be derived by the Noether argument and is given by
=
H
d3 x (x) m eA0 (x) + (i + ieA(x)) (x)
0
= H
d3 x (e)j (x) A (x),
(8.77)
(8.78)
0 the free fermion hamiltonian. Using our usual relativistic normalization,

with H
the one-particle eigenstates of the free hamiltonian satisfy
0 |pi = p0 |pi,
H
p0 =
m2 + p 2 ,
hp0 0 |pi = 2p0 (2)3 (p0 p)0 . (8.79)
in the one-particle subspace are given by

The matrix elements of H
hp0 0 |H|pi = p0 hp0 0 |pi + e
d3 x hp0 0 |j (x)|pi A (x).
(8.80)
Using the results derived in Problem 6.3 we have

0
hp0 0 |j (x)|pi = u
(p0 , 0 ) i u(p, ) ei(pp )x ,
(8.81)
1
u
(p0 , 0 ) [(p + p0 ) + 2i(p p0 ) S ] u(p, ). (8.82)
2m
In the non-relativistic approximation the spinor-matrix elements become
u
(p0 , 0 ) i u(p, ) =
u
(p0 , 0 ) u(p, ) = 2m [0 + O(p2 /m2 )],
u(p0 , 0 ) S 0n u(p, ) = 2m [O(|p|/m)],
u
(p0 , 0 ) S mn u(p, ) = 2m [mnl (l /2)0 + O(|p|/m)].
(8.83)
(8.84)
(8.85)
In the chiral representation S 0n = 5 n /2 and 5 anticommutes with . So at

zero momentum, u = u, u
= u
, and it follows that u0 S 0n u = 0. Substitution
8.5. GAUGE-INVARIANT NON-RELATIVISTIC REDUCTION
203
in (8.80) and deviding by the relativistic normalization factor 1/2p0 1/2m

to get the same normalization as in (8.70), reproduces the magnetic moment
coupling in (8.72) with g = 2. The rest energy m is omitted in the usual nonrelativistic expressions. The A2 term in (8.72) is not reproduced this way; this
will be clarified in section 8.5.
The g factor has been calculated in higher orders in perturbation theory (including the quantized photon field) to great accuracy, for the electron and the
muon. The theoretical calculations agree with stunningly precise experimental
results,
(ge 2)/2 = (1159.652193 0.000010) 106 ,
(g 2)/2 = (1165.9230 0.0084) 106 ,
(8.86)
(8.87)
to eight significant digits in the case of the electron. For more comments on
precision results in QED, see for example Peskin & Schroeder, end of section 6.3.
8.5
Gauge-invariant non-relativistic reduction
In the non-relativistic regime the energies are so small that particle creation is
not possible becasue of energy conservation. We may then expect the number
of particles and antiparticles to be separately conserved to a very good approximation, and indeed, the non-relativistic field theory derived in section 6.4 has
a corresponding U(1) symmetry. Similarly, a one-particle wave function in the
momentum representation,
(p, t) = hp|, ti,
|pi |pi/ 2p0 ,
(8.88)
may be expected to satisfy a Schrodinger equation

X
d
i (p, t) =
dt
0
d3 p0
0 0 i 0 (p0 , t) + ,
hp|H|p
3
(2)
(8.89)
for which the contribution of other intermediate states (e.g. multi-particle

states) is negligible in the non-relativistic limit.
0 0 i is linear in A (see
However, the one-particle matrix element hp|H|p
(8.80)), and the A2 term of (8.72) is lacking. This quadratic term is needed
for gauge-covariance of the non-relativistic Schrodinger equation.7 Evidently, the
contribution in (8.89) must contain the A2 term somehow, but bringing this
to the fore could be cumbersome. Instead, let us give a heuristic derivation of
the non-relativistic theory along the lines of section 6.4.
is gauge invariant, its matrix elements hp|H|p
0 0 i
Although the original hamiltonian H
0 and its eigenstates |pi are not gauge
do not have this property, because free hamiltonian H
invariant.
7
204
CHAPTER 8. QED
Consider the path integral for the electron field coupled to an external electromagnetic potential, with sources and ,
Z[, ] =
Z
[dd]
exp i
Z
= Z[0, 0] exp i
d x [(D
+ m) + + ]
4
d x d y (x)G(x, y, A)(y) ,
(8.90)
where D = + ieA and G(x, y, A) is the fermion propagator in the given A

field. It satisfies
(m + D )ab Gbc (x, y, A) = ac 4 (x y),
(8.91)
with appropriate boundary conditions, or in condensed notation (suppressing the

i),
G = (m + D)1 .
(8.92)
We now first try to write G in a form similar to that of the free propagator:
(m + )1 = (m )(m2 2 )1 , which in momentum space is the familiar
(m + ip)1 = (m ip)(m2 + p2 )1 . Using = + 2iS and [D , D ] =
ieF , with F = A A as usual, we have
(m + D)(m D) = m2 D D = m2 D D + eS F
m2 D 2 + eSF.
(8.93)
Hence,
G = (m D)(m2 D 2 + eSF )1 = (m2 D 2 + eSF )1 (m D).
(8.94)
We now mimic the approximations made in section 6.4, treating eA0 of the order
of the kinetic energy p2 /2m and eA of the order of the momentum |p|. We start
with the analog of (6.117):
m D m 0 0 = m + i0 2mP+ ,
P+ = (1 + )/2,
(8.95)
which leads again to the conclusion that the non-relativistic sources (containing
only frequencies near m) have only components in the = +1 subspace. In the
present Dirac case (instead of the Majorana case of sect. 6.4), this is implemented
by
0
(x) eimx (x), (x) eimx (x),
= P+ , = P+ ,
(8.96)
where and possess only frequencies with magnitude m, and wave vectors
|p| m.
205
8.6. SUMMARY
The time components S 0n of the matrices S appearing in the denominator

in (8.94) may give transitions between the subspaces with = +1 and 1, but
this is a higher order effect: we have8
P+ (m2 D 2 + eSF )1 P+
= P+ (m2 D 2 )1 (m2 D 2 )1 P+ eSF P+ (m2 D 2 )1 +
i
= P+ (m2 D 2 )1 (m2 D 2 )1 2eS B(m2 D 2 )1 +
P+ (m2 B 2 + 2eS B)1 ,
(8.97)
where we used the fact that S 0n anticommutes with , whereas the spin matrices
Smn = mnp Sp commute with , so P+ S 0n P+ = 0, and
P+ SF P+ = P+ (2S 0n F0n + S mn Fmn )P+ = 2S B P+ .
(8.98)
The analog of (6.116), combined with (8.94) (8.97) now leads to

G = (m D)(m2 D 2 + eSF )1

D2 /2m + 2eS B/2m 2m(iD0 + i)
1
(8.99)
The corresponding path integral in terms of two-component non-relativistic spinor

fields is given by
D2
2e
Z[, ] = [d d] exp i d x iD0
S B + + .
2m
2m
(8.100)
We see again the magnetic-moment coupling with g-factor 2, and the gauge invariance is manifest.
8.6
( Z
"
Summary
Gauge invariance and renormalizability are guiding principles in constructing

covariant actions for interacting fields. QED is a prime example, which predicts
a wealth of phenomena which have been verified experimentally.
Scattering amplitudes can be shown to be independent of the gauge parameter specifying a covariant gauge. Spin-averaged amplitudes are Lorentz invariant and can be evaluated with the trace technique. The Lande g-factor, which
8
A useful expansion formula is

(A + B)
= [(1 + BA
)A]
=A
(1 + BA
1 1
=A
= A1 A1 BA1 + A1 BA1 BA1 + .
"
1+
n=1
(BA
1 n
206
CHAPTER 8. QED
parametrizes the non-relativistic magnetic-moment coupling of the electron to an

external magnetic field, is 2, with calculable corrections that have been verified
by experiment to extremely high precision.
The road is now open for generalization to theories with non-abelian gauge
fields, such as the Standard Model.
8.7
Problems
1. Gauge invariant measure

Verify formally (using the ill defined
[dd].
x)
the gauge invariance of [dA] and
2. Compton scattering
The elastic scattering of a photon and an electron is an example of Compton
scattering. Consider the process (k)+e (p) (k 0 0 )+e (p0 0 ), where
k, . . . , p0 denote the four momenta and , . . . , 0 the spin polarizations of
the particles. The amplitude can be calculated by expanding9
hk 0 0 , p0 0 out|k, p ini
= hA(+) (k 0 0 , ) (+) (p0 0 , )() (p, )A() (k, )i
hA(+) (k 0 0 , ) (+) (p0 0 , )() (p, )A() (k, ) eiS1 i0
=
heiS1 i0
in terms of
S1 =
A + S.
d4 x eq i
(8.101)
Recall that q = 1 for the electron field; the counterterms S can be

neglected in the tree-graph approximation. The first nontrivial scattering
occurs at second order and given by
hk 0 0 , p0 0 out|k, p ini =
(ieq)2
d4 x d4 x0 hA(+) (k 0 0 , ) (+) (p0 0 , )
2!
0 )i (x0 )A (x0 )
(x)i
(x)A (x)(x
() (p, )A() (k, )i0 + (8.102)
Z
In short-hand, we evaluate the contractions

hA(+) A A0 A() i0 = hA(+) A i0 hA0 A() i0 + hA(+) A0 i0 hA A() i0
+ hA(+) A() i0 hA A0 i0
(8.103)
9
For a change, we use the notation A(+) (k, , t) = a(k, , t), etc.
207
8.7. PROBLEMS
(the last contraction leads to loop diagrams) and
0 i 0 () i0 = h (+) i
0 i h 0 i0 i h 0 () i0
h (+) i
+ x x0 ,
(8.104)
plus contractions leading to loop diagrams.
a. Verify that the above leads to the following scattering contribution to
(8.102):
(ieq)2 Z 4 4 0 ip0 x
0
d xd x e
u
(p0 0 )i (i)S(x x0 )i u(p)eipx
2
i
h 2!0
0
0 0
(8.105)
eik x e (k 0 0 )e (k)eikx + eik x e (k 0 0 )e (k)eikx
= i(2)4 (p0 + k 0 p k)T,
(8.106)
with
iT =
(8.107)
m i(p + k)
(eq )u(p)e (k) +
2
2
m + (p + k)
m i(p k 0 )
(eq )u(p)e (k)
e (k 0 0 )
u(p0 0 )(eq )(i) 2
m + (p k 0 )2
e (k 0 0 )
u(p0 0 )(eq )(i)
b. Write down the corresponding diagrams, include all the relevant labels
and establish a 1-1 correspondence between the diagrams and the expression
for T .
c. Verify that the other contributions lead to loop diagrams
d. The scattering amplitude has the form
T = e (k 0 0 )T (p0 , k 0 , p, k)e (k),
(8.108)
which defines T (p0 , k 0 , p, k) (of course only three four-momenta are independent because p0 + k 0 = p + k). Using the Ward-Takahasi identities
ik = S 1 (p + k) S 1 (p) = S 1 (p0 ) S 1 (p0 k)
(8.109)
and the Dirac equation for the spinors u(p) and u

(p0 0 ), verify that
k T (p0 , k 0 , p, k) = 0,
(8.110)
and similarly k0 T (p0 , k 0 , p, k) = 0. Using this result we know that on

calculating the unpolarized cross section we may use eq. (7.169) and replace
X
e (k)e (k) .
(8.111)
208
CHAPTER 8. QED
The property (8.110) can be understood as implementing invariance with respect to gauge tranformations on the photon polarization vectors, e (k, )
e (k, )+ k , which occur in Lorentz transformations in Coulomb gauge.
The replacement (8.111) implements Lorentz invariance in this respect.
Note that T has the form T = u0 T u, with T transforming as a
P
tensor-matrix, and that the polarization sum spins u
0 T u (
u0 T u) results
in a Lorentz tensor.
e. Further evaluation of the differential cross section is described for example in the book by Weinberg (part I sec. 8.7), or by Bjorken & Drell (part
I), Sterman, etc. We quote the Klein-Nishina formula which holds in the
laboratory frame:
"
d
d
e2
4
=
lab
!2
k 00
k0
1
2m2
!2
k0
k 00
+
sin2 .
k 00
k0
!
(8.112)
Express k 00 in terms of m, k 0 and .

3. Electron-electron scattering
In this problem we shall work out the unpolarized cross section for the
process e + e e + e . Consider the amplitude for e

1 + e2 e3 + e4 ,
T (34; 12) = e2 [
u 3 u1 u
4 u2 D (k) u
3 u2 u
4 u1 D (l)] , (8.113)
where k = p3 p1 and l = p4 p1 .
a. Derive this.
b. Show using the Dirac equation in momentum space that u3 u1 k = 0,

u
3 u2 l = 0. This corresponds to current conservation. Consequently the
amplitude can be simplified to
T = e
1
1
u
3 u1 u
4 u2 2 u
3 u2 u
4 u1 2 .
k
l
(8.114)
c. When the incoming electrons are unpolarized and we do not measure

the spins of the outgoing electrons, we have to average |T |2 over initial spin
polarizations and sum over final spin polarizations. We use the shorthand
notation
1 X
|T |2 =
|T |2 .
(8.115)
16 1 4
Derive along similar lines as in sect. 8.3, that
e4
|T |2 =
16
"
T1
T2
+ (p3 p4 ) ,
2
2
((p1 p3 ) )
(p1 p3 )2 (p1 p4 )2
(8.116)
209
8.7. PROBLEMS
where, using the convenient slash notation /p = p ,
T1 = Tr [ (m ip
/1 ) (m ip
/3 )] Tr [ (m ip
/2 ) (m ip
/4 )],
(8.117)
T2 = Tr [ (m ip
/1 ) (m ip
/4 ) (m ip
/2 ) (m ip
/3 )].
(8.118)
d. Using the trace identities and of course momentum conservation p1 +

p2 = p3 + p4 and p2i = m2 , show that
T1 = 32[2m4 + 2m2 p1 p3 + (p1 p2 )2 + (p1 p4 )2 ],
T2 = 32[2m2 p1 p2 + (p1 p2 )2 ].
(8.119)
(8.120)
e. These expressions can be evaluated in the center of mass frame. Let be

the scattering angle between particles 1 and 3, p1 p3 = m2 |p|2 (1cos ).
From now on we use the notation p |p|. Verify that
T1
T2
|T |2
"
= 64 m + 4m p cos + 2p4 (1 + cos4 ) ,

2
2
= 32(m4 + 4p4 ),
#
"
e4
T2
T1
=
+ ( ) .
256p4 sin4 2
cos2 2 sin2 2
4
2 2
(8.121)
(8.122)
(8.123)
f. Under ultra-relativistic conditions we may neglect the electron mass m.

Verify that in the center of mass frame
"
d
2
= 2
d cm, ur 8p
#
"
1 + cos4
sin4 2
2
+
2
cos 2 sin2
1 + sin4
+
cos4 2
(8.124)
g. Under non-relativistic conditions we may neglect p compared to m.

Verify that
"
d
2 m2
=
d cm, nr
16p4
#
"
1
sin4
2
cos 2 sin2
1
+
cos4
(8.125)
The middle term is due to the interference of the two diagrams contributing
to the amplitude. The first term goes over in the Rutherford formula for
Coulomb scattering off a heavy target, upon expressing it in terms of the
reduced mass mred = mm/(m + m) = m/2.
The total cross section is infinite because the integration over angles diverges at = 0. This can be attributed to the infinite range of the Coulomb
potential.
210
CHAPTER 8. QED
Chapter 9
Scattering
Having seen how vacuum expectation values of time-ordered products of fields
can be calculated in perturbation theory, we derive here how these can be used for
the construction of scattering amplitudes. The discussion will be quite general,
without recourse to perturbation theory, such that it also applies to composite
fields and bound states as found in QCD. Results of scattering experiments are
expressed in terms of cross sections. We start with the relation between cross
sections, transition amplitudes hout|ini and invariant-scattering amplitudes T
(often denoted by M). After deriving so-called spectral representations for twopoint functions, which exhibit the connection between the occurrence of poles in
the complex frequency plane and the existence of particles, we derive a relation
between connected n-point functions and scattering amplitudes. This relation
is often called the LSZ formula, after the work of Lehmann, Symanzik and
Zimmermann. We shall not follow their derivation, but use an intuitive reasoning
inspired by Schwingers Source Theory. We end with a heuristic derivation of
formulas for unstable particle decay.
9.1
Cross section
In a colliding beam experiment, two bunches of particles approach each other

such that they overlap for a short period, during which the particles scatter. The
(x). We assume that

bunches 1,2 may be described by a four-current density j1,2
these currents are slowly varying on the scale of the Compton wavelengths m1
1,2 ,
and that the spread in momentum of the particles is small, such that to a good
approximation the current is just the average velocity times the density,
ji (x) = vi ji0 (x),
vi = pi /p0i ,
i = 1, 2.
(9.1)
Here pi is the average four-momentum of the particles in bunch i. The number

of scattering events N of a kind specified by , in a frame in which the bunch
211
212
CHAPTER 9. SCATTERING
velocities are aligned, v1 v2 , is given by

N = v12
d4 x j10 (x) j20 (x)
(9.2)
where is the corresponding cross section and v12 = |v1 v2 | is the relative
velocity. Multiple scattering is neglected. The number of events is Lorentz invariant, and the cross section is defined to be invariant, by generalizing the above
formula to
N =
d 4 x
[j1 (x)j 2 (x)]2 j1 (x)j 1 (x) j2 (x)j 2 (x)
(p1 p2 )2 p21 p22 Z 4 0

d x j1 (x) j20 (x)
p01 p02
= (v1 v2 )2 (v1 v2 )2
d4 x j10 (x) j20 (x).
(9.3)
(9.4)
(9.5)
Specializing to the case of aligned momenta p1 p2 , for which

q
(p1 p2 )2 m21 m22 = p01 p02 v12 ,
(9.6)
we get back (9.2).

To make contact with the elementary discussion in chapter 2, in which we
used the concept of event rate, consider very big uniform bunches. Then the
integration over time in (9.5) will be proportional to the total overlap time of the
bunches, and we identify the event rate
dN /dt = L(t),
with
L(x0 ) =
(v1 v2 )2 (v1 v2 )2
(9.7)
d3 x j10 (x) j20 (x),
(9.8)
the luminosity. However, in the following we shall use (9.5) to identify the scattering amplitude.
If we redefine the currents by dividing by the total number of particles N1,2 :
1
j1,2 (x) j1,2

(x),
N1,2
0
d3 x j1,2
(x) = 1,
(9.9)
then
N P ,
(9.10)
with P the probability for events of type . Possible specifications of have

been discussed in Problem 3 in chapter 2. Spin is ignored above; in case the
particles have spin, we may assume that it is not specified in the initial state and
not observed in the final state.
213
9.2. SCATTERING AMPLITUDE
9.2
Scattering amplitude
We now consider the scattering from the quantum point of view. The particles
in the bunches are assumed to be effectively noninteracting, so their specification
is like single particles. Let
|pi,
hp|p0 0 i = 0 2p0 (2)3 (p p0 ),
(9.11)
denote a covariant basis state for a single massive particle with four-momentum
p and spin index . The index may be taken to be the eigenvalue of J3 in a rest
frame of the particle, where J3 is the third component of the angular momentum.
Then = s, s + 1, , s, with s the spin of the particle and s(s + 1) the
eigenvalue of J2 in the rest frame. Another possibility is to use the helicity, i.e.
the eigenvalue of J in the direction of momentum, which also takes 2s + 1 values.
For massless particles there is no rest frame, and we shall assume to be the
helicity, which takes only two values in case it is non-zero. We have seen this to
be true for the photon and the same holds for the graviton; a general discussion
is in Weinberg I, and in Ryder. Massless particles require special treatment in
scattering theory, but in a first approach it is useful to simply generalize formulas
derived for massive particles to the massless case.
For spinless particles we have seen in (2.119) how |pi transforms under Lorentz
transformations. In case of spin we have
U (`)|p, i =
X
0
C0 (`, p)|`p, 0i,
(9.12)
with C0 (`, p) a unitary matrix depending on p and `,

X
C0 (`, p) C0 00 (`, p) = 00 ,
C 1 (`, p) = C (`, p) = C(`1 , p).
(9.13)
This unitarity is a consequence of the unitarity of U (`). The explicit form of C

is complicated, see e.g. Weinberg I or Ryder, but we do not need this.
In terms of these basis states we define wave-packet states
|f i =
XZ
dq f (q, ) |qi,
hf |f i =
and wave functions

f (x, ) =
XZ
dq |f (q, )|2 = 1,
dq f (q, ) eiqx ,
(9.14)
(9.15)
similar to the spinless case discussed in Sect. 2.8. Under Lorentz transformations
f (q, ) transforms as
f 0 (q, ) =
X
0
C0 (`, q) f (`1 q, 0 ).
(9.16)
214
v5
f5
v3
v4
f4
f3
v1
f1
v2
f2
Figure 9.1: In and outgoing wave packet states. The ingoing states are specified
in the far past, the outgoing in the far future
Generalizing the discussion in section 2.8 to particles with spin, let f (q, ) be
suitably sharply peaked about a mean momentum p. Then
j (x)
[f (x, ) i f (x, ) + i f (x, ) f (x, )]
' 2p
|f (x, )|2
(9.17)
is a good probability current for finding the particle at spacetime point x. Note
P
P
that f 0 (x, ) ' 0 C0 (`, `p)f (`1 x, 0 ) and so |f (x, )|2 is a scalar field
under Lorentz transformations, because of the unitarity of the matrix C. We
identify the current (9.17) with the normalized classical current (9.9). More
generally, one can consider a spin density matrix j 0 (x; , 0 ),
j 0 (x; , 0 ) = 2p0 f (x, )f (x, 0 ),
(9.18)
such that, for instance, the expectation value of the spin in the particle rest frame
is given by Tr j 0 (x)S, where S 0 is the spin matrix (e.g. for spin 1/2, S = /2,
with the Pauli matrices).
Consider now an incoming state of two widely separated particles converging
to a scattering region as in Fig. 9.1:
|f1 , f2 ini.
(9.19)
We are using the Heisenberg picture in which the states in Hilbert space are
time-independent. Quantum numbers other than momentum and spin (e.g. distinguishing an electron from a proton) are suppressed. How to make such a
state in a quantum field theory with interactions will be discussed later in this
chapter. Similarly, consider states representing widely separated particles at late
215
9.2. SCATTERING AMPLITUDE
times, which may or may not have emerged from the scattering region, as in Fig.
9.1:
|f3 , , fn outi.
(9.20)
The probability amplitude for the process f1 + f2 f3 + + fn is
hout f3 , , fn |f1 , f2 ini.
(9.21)
It is assumed that there are momentum-basis states |q1 1 , q2 2 ini, in terms of

which we can write
|f1 , f2 ini =
XZ
1 2
dq1 dq2 f1 (q1 , 1 ) f2 (q2 , 2 ) |q1 1 , q2 2 ini,
(9.22)
and similar for the out states. Note that these sharp momentum-states are certainly overlapping in position space (being plane-wave states). They are therefore
not merely tensor products of free-particle states, and the labels in and out are
important. Only single-particle states are free states1 in the sense that the labels
in and out may be omitted: |p, ini = |p, outi = |p, i. The probability for
ending up in the final state specified by a momentum region is now
P =
X Z
3 n
dp3 dpn |hout p3 3 pn n |f1 , f2 ini|2 .
(9.23)
We are going to rewrite this expression into the form (9.2).

First, making the wave functions explicit we have
|hout p3 3 pn n |f1 , f2 ini|2 =
1 2 01 02
dq1 dq2 dq10 dq20
f1 (q1 , 1 ) f2 (q2 , 2 ) f1 (q10 , 01 ) f2 (q20 , 02 )

hout p3 3 pn n |q1 1 , q2 2 ini
hout p3 3 pn n |q10 01 , q20 02 ini .
(9.24)
Second, the in and out states are specified in the far past and far future. Therefore we expect energy-momentum conservation to hold, such that we may write
schematically,2
hout | ini = , i(2)4 4 (p p ) T (, ).
(9.25)
Here the first term on the right hand side represents the possibility of no scattering, while the second term represents scattering, with energy-momentum conservation made explicit. The occurrence of the Dirac delta function 4 (p p )
1
Here these states are assumed to be eigenstates of the complete hamiltonian, including
interaction terms.
2
The matrix S, hout | ini is called the scattering matrix, or S-matrix for short.
216
will be evident later on (we have seen it emerge in lowest order calculations); it
is also implicitly present in the -term. More explicitly, (9.25) reads, in case
the final state is different from the initial,
hout p3 3 pn n |q1 1 , q2 2 ini
(9.26)
4 4
= i(2) (p3 + + pn q1 q2 ) T (p3 3 , , pn n ; q1 1 , q2 2 ).(9.27)
The function T (p3 3 , ) is the called the invariant amplitude, which is often
denoted by M. Assuming the final states contributing in to be different from
the initial state, only the T -terms contribute.
Third, we have the identity
4 (P q1 q2 ) 4 (P q10 q20 ) = 4 (P q1 q2 ) 4 (q1 + q2 q10 q20 ). (9.28)
Fourth, the delta functions occur in integrals with functions f1,2 (q, ) which are
sharply peaked about p1,2 , so we may make the replacement
4 (P q1 q2 ) 4 (P p1 p2 ).
(9.29)
Fifth, using
(2)4 4 (q1 + q2 q10 q20 ) =
d4 x ei(q1 +q2 q1 q2 )x ,
(9.30)
the factor in (9.24)

Z
dq1 dq2 dq10 dq20 (2)4 4 (q1 + q2 q10 q20 )
f1 (q1 , 1 ) f2 (q2 , 2 ) f1 (q10 , 01 ) f2 (q20 , 02 )
(9.31)
equals approximately
1
4p01 p02
d4 x j10 (x; 1 , 01 ) j20 (x; 2 , 02 ),
(9.32)
with
ji0 (x; , 0 ) = 2p0i fi (x, )fi (x, 0 )
(9.33)
the density matrix for particle i = 1, 2, as in (9.18).

Putting things together we can rewrite (9.23) in the form
P =
1 02
(01 02 , 1 2 )v12
d4 x j10 (x; 1 , 01 ) j20 (x; 2 , 02 )
(9.34)
with
(01 02 , 1 2 ) =
X Z
1
dp3 dpn (2)4 4 (p1 + p2 p3 pn )
4p01 p02 v12 3 n
T (p3 3 , ; p1 1 , p2 2 ) T (p3 3 , ; p1 01 , p2 02 ) .
(9.35)
217
9.3. FIELDS, PARTICLES AND POLES
The initial density matrices have here a pure-state form. In an experimental

setup they have to be replaced by the appropriate density matrices. In case one
is interested in measuring spin dependence these would correspond to a spinpolarized initial state and the spin dependence of the final state would also be
analyzed in more detail. Here we shall assume that spin is not monitored, unless
mentioned otherwise. This means that we should replace the pure-state forms by
ji0 (x; , 0 ) ji0 (x)
0
,
2si + 1
i = 1, 2,
(9.36)
with s1,2 the spin of the particle and j 0 (x) defined in (9.17). Correspondingly we
introduce the short-hand notation
X
1
|T (p3 3 , ; p1 1 , p2 2 )|2 ,
i=1 (2si + 1) 1 n
|T |2 = Qn
(9.37)
which is a Lorentz-invariant function of the momenta (as follows from (9.12) and
(9.26)). Recalling (9.6) gives finally the formula for the unpolarized cross section
=
(2s3 + 1) (2sn + 1)
q
4 (p1 p2 )2
m21 m22
dp3 dpn (2)4 4 (p1 + p2 p3 pn ) |T |2 .
(9.38)
Problem 3 in chapter 2 elaborates on how to use this formula for the case 1 + 2
3 + 4.
9.3
Fields, particles and poles
In free field theory the eigenstates of the energy-momentum operator P are

found by application of the creation operators a (p) on the vacuum state |0i. In
an interacting field theory the creation and annihilation operators can be defined
similarly at some time t, with a corresponding empty state |, ti, a(p, t)|, ti = 0,
as we did e.g. in sect. 2.5. However the states a (p1 , t) a (pn , t)|, ti are not
eigenstates of P anymore; in perturbation theory they are only eigenstates at
lowest order. They are called bare particle states. The true, dressed, eigenstates
are complicated linear superposition of the bare states. We can try to calculate
these in perturbation theory, but this is cumbersome, and it would not work in
case of bound states, which are non-perturbative.3 We have learned to focus
on the probability amplitudes hout|ini, rather than on states themselves. These
amplitudes can be expressed in terms of vacuum expectation values of field operators. We shall keep the discussion general, such that it applies also to the case
3
For example, the ground state of hydrogen cannot be found by perturbing a state with a
free electron and a free proton. In QCD all eigenstates of the hamiltonian are bound states of
quarks and gluons.
218
of bound states, proceed heuristically and appeal to intuition in avoiding difficult

proofs.
Suppose a (x) is a hermitian operator field transforming in a representation
L of the Lorentz group,
U (`) a (x)U (`) = Dab (`)b (`1 x).
(9.39)
It may be an elementary field, i.e. a basic variable appearing in the action, or

a composite field contructed out of products of elementary fields, e.g. a current
j (x). Suppose furthermore that Ja (x) is a function with support in a compact
spacetime region, with the help of which we smear out the field a ,
(J)
d4 x Ja (x)a (x).
(9.40)
Intuitively one expects (J) to create a particle out of the vacuum |0i in the
spacetime region J, i.e. the amplitude
hp|(J)|0i
(9.41)
is nonzero.4
Quantum numbers other than p, (such as electric charge, lepton number,
. . . ) have to correspond to those of (J), otherwise the above amplitude would
be zero. However, because of interactions the state (J)|0i will have many more
components in Hilbert space, multiparticle states and possibly even the vacuum
itself: for a scalar field h0|(x)|0i may be nonzero. In the following we shall
assume that the vacuum expectation value of a has been subtracted from the
field:
h0|a (x)|0i = 0.
(9.42)
To see how to construct transition amplitudes with a given number of particles
we start with the simplest of all: hp|p0 0 i, which can be made out of the two
point function
h0|a (x)b (y)|0i.
(9.43)
First we derive a spectral representation.5 We are going to insert a complete set of
intermediate states between the two fields above. To contribute, the intermediate
states must have the same quantum numbers as the field, because of Schurs
lemma (we suppress these quantum numbers in our notation). For the eigenvalues
of the energy-momentum operator we make some reasonable assumptions:
4
The annihilation and creation operators a(p, t) etc., defined in sect. 6.6 etc., can also be
written in the general form (J), with suitable J(x) containing (x0 t) and its derivative.
The notation J suggests it to be a source for and this aspect is indeed and essential element
in the methods employed by Schwinger in his Source Theory .
5
In this context also called a K
allen-Lehmann representation.
219
p0

p
Figure 9.2: Energy-momentum spectrum of contributing states. The single curve
represents the mass shell p2 = m2 , p0 > 0 of a single particle state. The hatched
area represents p2 (m + m0 )2 of the multiparticle states.
- The groundstate (vacuum state) |0i is a non-degenerate Lorentz-invariant
eigenvector of P and J with eigenvalue zero,
P |0i = 0,
J |0i = 0.
(9.44)
- The contributing single-particle states are stable. The single-particle nature

implies that p is on a mass shell as illustrated in Fig. 9.2,
P |pi = p |pi,
p0 =
m2 + p 2 .
(9.45)
- The infinite remainder of the basis states are multi-particle states (e.g. ingoing or out-going),
P |p, i = p |p, i,
p2 > m2 .
(9.46)
Here p is the total four-momentum of the states. Other quantum numbers

(such as relative momentum, spins) are lumped into the symbol . Unstable
particles decay eventually in terms of stable particles, so unstable particles
are to be somehow represented by the states |p, i.
In writing p2 > m2 we have assumed that there are no massless particles. For
example, suppose the smallest mass other than m is m0 . Then, for a two-particle
state with four-momenta p and p0 , p2 = m2 , p02 = m02 , the
invariant
total
02 mass
0 2
2
2
2
is given in the center of mass frame by (p + p ) = m + 2 m + p m + p2 +
m02 (m+m0 )2 , which is larger than m2 if m0 6= 0. The spectrum of contributing
states is illustrated in Fig. 9.2.
220
Inserting the complete set of contributing states we write6

a
h0| (x) (y)|0i =
XZ
XZ
dp =
dp h0|a (x)|pihp|b (y)|0i
m+m0
dM
d3 p
,
(2)3 2 m2 + p2
(9.47)
dp (M ) h0|a (x)|pihp|b (y)|0i,

dp (M )
d3 p
, (9.48)
(2)3 2 M 2 + p2
where we have explicitly indicated the invariant mass p2 M 2 of the multiP

particle states. Note that the shorthand includes also integrations over the
continuous variables buried in . The above expression can be simplified further
using Poincare symmetry. Translation invariance gives
h0|a (x)|p, i = h0|eiP x a (0)eiP x |p, i = h0|a (0)|p, i eipx.
(9.49)
Lorentz invariance gives

h0|U (`) a (0)U (`)|p, i = Dab (`)h0|b (0)|p, i
= C0 (`, p)h0|a (0)|`p, 0 i,
(9.50)
(9.51)
where we used (9.12) and (9.39) (and used the summation convention
for the
7
0
repeated ). This shows that Ua (p, ), defined up to a constant Z by
Z Ua (p, ) = h0|a (0)|p, i,

(9.52)
transforms just like the basis vector |p, i:
Dab (`)Ub (p, ) = C0 (`, p) Ua (`p, 0 )
(9.53)
The Ua (p, ) are polarization tensors or spinors, depending on the representation
L. They can be normalized in some convenient covariant way. The constant Z

depends on a . It can we any complex number, including zero, but we assume it
here to be nonzero. In case a is an elementary
field, the phases of the basis states
|p, i are usually chosen such that Z is positive, and Z is the wave function
renormalization constant first introduced in chapter 4. For a scalar field and a
spin zero state, eq. (9.53) reads U (p) = U (`p), so U
(p)2 is a2 Lorentz invariant
0
function of p. For the on-shell case here (i.e. p = m + p ) this means it is
just a constant, which we choose to be unity, i.e. U (p) = 1 for a scalar field.
6
We have written this as if there is only one type of particle contributing to the one-particle
intermediate states. If there are more, all should be included. For instance, both particles and
antiparticles may contribute if a a hermitian combination of charged currents. In such a
case one often uses non-hermitian a for which only one type of particle contribute.
7
This Ua (p, ) should of course not be confused with the representation U (a, `) of Poincare
transformations in Hilbert space.
221

For a spinor
(recall a is
and massive
later.
In (9.47)
field Ua (p, ) = ua (p, ), the spinor in the Majorana representation

hermitian). Problem 1 deals the case of the vector representation
spin-one particles. The photon case is special and will be discussed
we need the summation over spins
Pab (p) =
Ua (p, ) Ub (p, ).
(9.54)
Because of the unitarity of C0 (`, p), this is an invariant tensor or spinor matrix,
Daa0 (`)Dbb0 (`)Pa0 b0 (p) = Pab (`p).
(9.55)
For example, in the spinor case, using a Majorana representation for the matrices, Pab (p) = [(m ip)]ab . For the vector case Pab (p) P (p) = +
p p /m2 , see Problem 1.
We conclude that translation and Lorentz invariance restrict the form of the
single particle contribution to
h
i

X
h0|a (x)|p, ihp, |b (y)|0i = Z Pab (p) eip(xy) 0 2 2 , Z Z
Z .
p +m
p =
(9.56)
The constant Z above is evidently positive. This is a direct consequence of our implicit assumption that ket space is a Hilbert space with positive metric. Within
the same assumption Pab (p) is a positive matrix (i.e. for any ca , ca Pab (p)cb 0).
In a perturbative treatment of gauge theories in covariant gauges, the metric
is not positive, and the summation over intermediate states has to be modified
accordingly, as in (7.156,7.157). Then we cannot conclude anymore that Z is positive. For simplicity we shall continue under the assumption of a positive metric
Hilbert space. (This is actually the situation in nonperturbative treatments of
QCD with the lattice regularization.)
Similarly, we can write for the combination appearing in the multiple particle
contribution,
X
h0|a (x)|p, ihp, |b(y)|0i = (M ) Pab (p) eip(xy)
p0 =
p2 +M 2
(9.57)
We have separated a kinematical part in the form of Pab (p) exp[ip(x y)]. The
dynamical (interaction dependent) part is in the positive function (M ), which
is the analogue of Z for the multiple particle contribution.
Summarizing, we have derived the spectral representation
h0|a (x)b (y)|0i = Z
dp Pab (p) eip(xy)
m+m0
dM
p0 =
(9.58)
p2 +m2
dp (M ) (M ) Pab (p) eip(xy)
p0 =
p2 +M 2
222
Calculations are usually carried out in terms of time-ordered products of

fields. The ordinary T product, when defined in terms of Heaviside step functions (x0 y 0 ), is not necessarily covariant, except for scalar and spinor fields.
Manifestly covariant path-integral techniques lead to expressions ha (x)b (y) i
which represent covariant time ordered products. These are denoted by T , and
still satisfy the usual relations
ha (x)b (y)i = h0|T a (x)b (y)|0i = h0|a (x)b (y)|0i, x0 > y 0 (9.59)
= h0|b (y)a (x)|0i, x0 < y 0 ,
with the upper/lower sign for Bose/Fermi fields. The difference between the
T product and the T product is concentrated at x0 = y 0 , i.e. it contains the
Dirac delta function (x0 y 0 ) and possibly also its derivatives. The spectral
representation for the covariant time ordered product can be found as follows.
First one observes that the invariant tensor/spinor Pab (p) has an extension off
the mass shell. This means that there is a quantity
apolynomial in p
N2ab (p, m),
0
2
(hence defined for all p, not only those with p = p + m or p2 + M 2 ), such
that on-shell
Nab (p, m) = Pab (p),
Nab (p, m) = Pba (p),
p0 =
p0 =
p2 + m2 ,
(9.60)
(9.61)
p2 + m2 ,
where we may also read M instead of m. Examples are the Majorana spinor case
Nab (p, m) = [(m i p )]ab ,
and the vector case
N (p, m) = +
p p
m2
(9.62)
(9.63)
(cf. in Prob. 1).

In terms of Nab the spectral representation for the covariant time ordered
product of the vacuum expectation value of a is given by8
d4 p ip(xy) ab
e
G (p),
(9.64)
(2)4
Z
Nab (p, m)
Nab (p, M )
Gab (p) = Z 2
+
.
dM
(M
)
m + p2 i
M 2 + p2 i
m+m0
h0|T a (x)b (y)|0i = i
It is straightforward to check that (9.64) leads to (9.58), by performing the integral over p0 for x0 > y 0 or x0 < y 0 by contour integration in the usual way,
provided that one ignores possible nonconvergence of the integral at p0 = .
This happens with massive particles of spin > 1/2. The distribution (generalised
8
In this chapter we do not distinguish dressed propators (G0 ) from bare ones (G).
9.4. SCATTERING AMPLITUDES FROM CORRELATION FUNCTIONS223

function) aspects come more to the forefront in the higher spin case. Such matters
are to be resolved in actual calculations using the path integral.
The analytic structure
full a propagator Gab (p) is transparant: it
2 of the
0
2
has poles at p = p + m , the positions of which are determined by the
9
0
mass
q of the particle created out of the vacuum, and branch points at p =
(m + m0 )2 + p2 . Accordingly, the function G(p) is rendered unique by cutting
the complex p0 plane along the intervals p0 = (m + m0 , ) and (, m m0 ).
If there is no particle with the quantum numbers of a , then Z = 0 and there is
no pole but only a branch cut.
This is the particle-pole connection. It gives us a very important method
for calculating particle masses: construct a field a with well defined quantum
numbers and look for the position of a pole in its full propagator. If there is a
pole, the particle exists; if there is no pole, it does not.
9.4
Scattering amplitudes from correlation functions
We now continue our derivation of scattering amplitudes from correlation functions (also called connected Green functions) ha (x)b (y) iconn . First we want
to recover the free particle amplitude hp|p0 0 i from the fully interacting two
point function.
Consider
h0|(J1 )(J2 )|0i =
= Z
d4 x d4 y J1a (x)J2b (y)h0|a(x)b (y)|0i

XZ
dp J1 (p, )J2 (p, )
(9.65)
(9.66)
m+m0
dM (M )
dp (M )J1a
(p)Pab (p)J2b (p).
where we used the spectral representation (9.58), with (9.54) and the definition
J(p, ) = Ja (p) Ua (p, ),
in which Ja (p) is the Fourier transform
Ja (p) =
p0 =
m2 + p 2 ,
d4 x eipx Ja (x).
(9.67)
(9.68)
Note that this Fourier transform is defined for arbitrary p whereas in (9.67)
only on-shell values of p are used. The first term in (9.66) is the single particle
9
A continous superposition of poles leads to a discontinuity along a branch cut, ending at

R1
branch points, as in the example 0 dx 1/(x + y i) = ln(1 + y i/y i), where the cut runs
from 0 to 1.
224
contribution. It has the desired form of an inner product of wave packet states,
XZ
dp J1 (p, )J2 (p, ) = hJ1 |J2 i.
(9.69)
The second term is the multi-particle contribution which we do not want. To get
rid of it we take J1 to the far future and J2 to the far past, i.e. the support (the
region where it is non-zero) of J1 (x) moves to the far future of that of J2 (x). Intuitively, the multi-particles then disperse and the overlap h0|(J1 )|multiihmulti|(J2 )|0i
goes to zero. Mathematically, the product of Fourier transforms J1 (p)J2 (p) picks
up a phase factor exp(ip0 t), where t is the time difference between the supports
of J1 and J2 , causing the integral in (9.66) to die out as t because of the
Riemann-Lebesque lemma, In doing so we do not want to change the on-shell
(p2 = m2 ) values of Ja (p), since these determine the wave functions J(p, ) of
our particles.10
Of course, in a practical experimental setup we do not actually take limits. At
finite but sufficiently large separation the multiparticle contribution is negligible.
Sufficiently large means much larger than the typical relevant time scale, which is
here set by m1 . This is generally a very short time compared to the experimental
situation, e.g. of order of 1010 cm for m = 1 MeV. For an estimate of how fast
the multiparticle contribution vanishes, see e.g. Brown sect. 6.2.
So we know how to make hJ1 |J2 i from the two point function:
1
hJ1 |J2 i =
lim h0|(J1 )(J2 )|0i
(9.70)
Z
Z
1
lim d4 x d4 y J1a (x) (i)Gab (x, y) J2b (y),
(9.71)
=
Z
where we have replaced h0|a (x)b (y)|0i by the time ordered product iGab (x, y),
as the time ordering is just right for the limit we are taking. As a corollary we
note the following useful formulas which follow along the same lines,
Z
XZ
1
4
ab
Z
dp J (p, ) Ub (p, ) eipy
lim
d x Ja (x)(i)G (x, y) =
Jfuture
Z
1

Z
= hJ|b (y)|0i,
(9.72)
Z
Z
X
Z
dp J(p, ) Ua (p, ) eipx
lim
d4 y (i)Gab (x, y)Jb (y) =
Jpast
= h0| (x)|Ji.
10
(9.73)
To see this more explicitly, consider a translation in a time-like direction v , v 0 > 0,

v = 1 (v can be taken along the average four-velocity of the wave packet): J(x) J(x, t) =
t) = J(p)
exp(it pv). We can keep the on-shell
J(x vt). In Fourier space this means J(p,
p
values of J fixed
sign(pv) m2 + p2 ], p = p + v pv, because
p by multiplication 2with exp[it
pv sign(pv) m2 + p2 = 0 for p = m2 . h (This
n factor causespJ(x, t) tooispread as t gets
large.) Using now the form J(p, t) = J(p) exp it pv sign(pv) m2 + p2 , for J1 and J2 ,
R
R
the multiparticle contribution m+m0 dM (M ) dp (M )J1a

(p, t1 )Pab (p)J2b (p, t2 ) goes to zero
for t1 , t2 by the Riemann-Lebesque lemma.
2

Consider now two sources (smearing functions) J1 , J2 in the past and an
arbitrary number J3 , . . . , Jn in the future as in Fig. 9.1. If the initial sources do
not overlap and are outside each others light cone it is plausible that the particles
are created independently. Similarly when the final sources are space-like with
respect to each other will particles be absorbed independently . Thus, for large
separations of the sources,11
1
lim h0|T (Jn ) (J3 )(J1 )(J2 )|0iconn
Zn
Z
1
=
lim d4 x1 d4 xn J1a1 (x1 ) Jnan (xn )
Z1 Zn
h0|T a1 (x1 ) an (xn )|0iconn .
(9.74)
hJ3 Jn out|J1 J2 iniconn =
Z1
The limit is J1,2 far past, J3n far future, e.g. along the average fourvelocities of the wave packets, again without changing the on-shell values of the
This gives us the desired probability amplitude for scattering. Taking the
J.
connected part eliminates the no scattering contribution.
We now continue with case that the fields a are the dynamical variables that
appear in the lagrangian model. Then we can proceed using the fact that in a
diagrammatical analysis, the connected n-point functions (also called connected
Green functions) can decomposed in two-point functions and 1PI vertex functions,
and there is always a two-point function G0 at each external leg. This implies
that they can be written in the form
h0|T a1 (x1 ) an (xn )|0iconn (i)n1 Ga1 an (x1 , , xn )
=
(9.75)
d4 y1 d4 yn (i)Ga1 b1 (x1 , y1 )
(i)Gan bn (xn .yn ) iHb1 bn (y1 , , yn ),

which defines the function H. Using the first lines in (9.72) and (9.73), eq. (9.74)
can be written in the form
hJ3 Jn out|J1 J2 ini
conn
Z1 Z2 Z3
Zn
X Z
1 n
dp1 dpn
(2)4 (p1 + p2 p3 pn )
J3 (p3 , 3 ) Jn (pn , n ) Ua3 (p3 , 3 ) Uan (pn , n )
iHan a3 a1 a2 (pn , , p3 , p1 , p2 )
Ua1 (p1 , 1 ) Ua2 (p2 , 2 ) J1 (p1 , 1 ) J2 (p2 , 2 ) (9.76)
Here the momentum-conserving delta function is due to translation invariance;
recall that in our notation such delta functions are extracted from the definition (4.27) of the Fourier transform. Because the Js are arbitray (within the
11
For notational simplicity we assume

Z = Z .
226
constraint of describing reasonable wave packet states) we conclude that for the
sharp-momentum basis states
hp3 2 pn n out|p1 1 , p2 2 iniconn = i(2)4 (p1 + p2 p3 pn ) T,
q
T (p3 3 pn n ; p1 1 , p2 2 ) = Z1 Z2 Z3 Zn
Ua3 (p3 , 3 ) Uan (pn , n )
Han a3 a1 a2 (pn , , p3 , p1 , p2 )
Ua1 (p1 , 1 ) Ua2 (p2 , 2 ),
(9.77)
where T is the scattering amplitude.

So the recipe for obtaining the scattering amplitude is quite simple:
1. take the full connected n-point function in momentum space and make
explicit the two-point functions on the external lines:
Ga1 an (p1 , , pn ) = Ga1 b1 (p1 ) Gan bn (pn ) Hb1 bn (p1 , , pn );
(9.78)
2. omit the external line G0 s so as to obtain

Ha1 an (p1 , , pn ),
(9.79)
which is sometimes called the amputated Green function;

3. put H on-shell and contract with the polarization tensors/spinors,
Ua3 (p3 , 3 ) Uan (pn , n ) Han a3 a1 a2 (pn , , p3 , p1 , p2 )
Ua1 (p1 , 1 ) Ua2 (p2 , 2 );
(9.80)
4. multiply by the external-line wave-function renormalization factors to obtain T , as in (9.77).
The minus sign can be avoided by changing the sign in the definition of T ,
a convention that is also followed in the literature.
This recipe generalizes straightforwardly to the case in which (some of) the
are composite fields, as is typically the case for hadrons in QCD (see section
9.6). The off-shell form of H is then somewhat arbitrary, since it depends on the
choice of field a used to create and absorb a particle (usually there is more than
one
natural candidate). However, this arbitrariness is compensated for by the
Z factors and the contraction with the U s, and the scattering amplitude T is
unique. The same is true in the case in which the a are dynamical variables: then
T is invariant under a transformations of variables. This uniqueness property of
T is sometimes called the arbitrariness of the interpolating field.
a

For simplicity of notation we emphasized only the properties of a under
Lorentz transformations and suppressed the type of particle, e.g. is it a W boson
or a proton? This information can be added to the case at hand. Alternatively,
it is sometimes useful to think of a as running over a set of fields and use the
U a select a particle.
In practise we often do not use real fields but complex fields, for example
Dirac fields. With the experience obtained in the previous chapters it is easy to
adapt the formulas above for this case. Alternatively, we can consistently work
with real fields, which leads to quite a transparent formalism (see e.g. the books
by J. Schwinger, Particles, Sources and Fields), but which is not generally used.
Let us now comment on the effect of massless particles. the typical example
being the photon. Fundamentally, their effect invalidates our derivation, since
the pole in spectral representations is not separated from the cut if m0 = 0. Single particle contributions involving massless particles cannot be separated from
multiparticle contributions. However, we are partially saved by the fact that with
each photon emission or absorption comes a factor of 1/137. It is customary
to talk of hard and soft photons, the first having a sizable energy and the second having arbitrarily small energy. The hard photons are limited in number by
energy conservation, and since is so small, we can treat them perturbatively.
For example, in electron-electron scattering we can neglect them altogether in a
first approximation. But the soft photons can be unlimited in number, and when
there are more than 1/ of them their contribution falls outside perturbation
theory. The difficulty is resolved by a careful analysis that takes into account the
fact that we cannot distinguish experimentally a particle from a particle with a
cloud of soft photons (Bloch-Nordsieck resolution of the infrared catastrophe).
Massless fermions are less dangerous because the Pauli principle forbids them
piling up into nonperturbative numbers, and their effect can usually be treated
in perturbation theory. Still, also in this case the analysis of what is measurable
has to be refined. In practise, one can sometimes ignore these complications.
Finally, some comments on gauge invariance and using the Coulomb gauge.
The wave function renormalization constants Z depend on the gauge. In a general
covariant gauge they depend on the gauge parameter . However, this dependence
is just what is needed to make the transition amplitudes gauge in-dependent. In
the Coulomb gauge ket space has positive metric, but we have lost manifest
Lorentz covariance. This means, for example, that the wave function renormalization constants Z depend on |p| (there is still manifest rotation invariance),
which causes explicit demonstrations of equivalence to covariant gauges to be
quite involved.
228
9.5
Decay revisited
The reasoning in the previous sections can also be extended heuristically to the
decay of unstable particles. Let us briefly present a derivation for Rthe example
of the simple scalar field model of sect. 2.6, with interaction S1 = d4 x 12 g2 .
This new derivation complements that given in sect. 2.6 by expressing the decay amplitude in terms of the full three point function h(x1 )(x2 )(y)i. For
example, it allows for arbitrarily strong interactions of the particles in the final
state.
We start from a normalized wave packet state for the unstable particle,
|f i =
dq f (q) |qi,
dq =
d3 q
,
(2)3 2q 0
q0 =
M 2 + q2 ,
(9.81)
with M the mass of . We assume that f (q) is sharply peaked about q = P. The
probability amplitude for decay into two particles described by wave functions
f1 (p1 ) and f2 (p2 ) is, following the same heuristic reasoning as in the previous
sections, given by
hf1 f2 out|f i = lim h0|(J1 ) (J2 ) (J)|0i,
= lim
d4 x1 d4 x2 d4 y J1 (x1 )J2 (x2 )J(y)
h0|T (x1 )(x2 )(y)|0i,

q
f (q) = J(q),
q 0 = M 2 + q2 ,
f1,2 (p1,2 ) = J1,2

(p1,2 ),
p01,2 =
m2 + p21,2 ,
(9.82)
(9.83)
(9.84)
where the limit consists of taking J1,2 to the far future; the source J for the
particle stays put. The wave packet state |f i is supposed to be given, somehow,
at some time, say t = 0. It is not made by a limiting process in which the source
J is moved to the far past. This would not make sense because of the instability
of the particle. (Imagine the time reversed amplitude hf |J1 J2 ini: it would not
make sense to move the source J for the decaying particle to the far future.)
The total probability P for decay is tentatively given by
Z
1
dp1 dp2 |hp1 p2 out|f i|2 ,
(9.85)
P=
2
where the factor 1/2 takes care of the fact that the two particles in the final
state are identical. Note that the mass hidden in the dp s is the -mass m and
not M .
Using the reasoning in the previous sections, the above formula can be expressed in terms of an invariant decay-amplitude T , which can be computed in
terms of the connected three-point function with external lines removed, H,
hp1 p2 out|qi = i(2)4 (p1 + p2 q) T (p1 , p2 ; q),
1
T (p1 , p2 ; q) = q
H(p1 , p2 , q).
Z Z2
(9.86)
(9.87)
229
9.5. DECAY REVISITED
The amplitude T (p1 , p2 ; q) is Lorentz invariant, so it may depend on scalar products of the momenta, which leaves apparently one non-trivial invariant, (p1 p2 )q.
However, this vanishes in the rest frame of the decaying particle. So T is just a
constant.
Using the same type of approximations as in the derivation of the scattering
cross section we have
1Z
dp1 dp2 dq dqf (q)f (
q ) (2)4 (p1 + p2 P )(2)4 (q q) |T |2
2
Z
1
(9.88)
|T |2 I(P ) d4 x |f (x)|2 ,
2
Z
I(P ) =
f (x) =
dp1 dp2 (2)4 (p1 + p2 P ),
(9.89)
dq eiqx f (q).
(9.90)
For a wave packet f (q) sharply peaked about P we have

Z
d x |f (x)| =
dx
1
1
dq 0 |f (q)|2
2q
2P 0
dx0 ,
P0 =
M 2 + P2 ,
(9.91)
and we discover that we have obtained a result that diverges at large times:
R
dx0 .
This divergence is due to the fact that we have neglected higher order contributions. These will be such that, if we start with a large number N of particles,
this number will decay exponentially,
N (t) = N (0)et ,
(9.92)
with the decay rate. Furthermore, the number of particles will be given by
(assuming no other decay modes)
N (t) = 2N (0)(1 et ),
(9.93)
or, interpreting this as a probability:

P(t) =
N (t)/2
= 1 et .
N (0)
(9.94)
Now if we expand P(t) in , P(t) = t + , we get nonsense for large t if

we throw away the higher order terms.
We can make sense out of our result
R
(9.88)(9.91) for P by identifying dx0 in (9.91) with t:
P = t +
with
dx0 ,
|T |2
1
= I(P )
.
2
2P 0
(9.95)
(9.96)
230
We expect that the result to all orders will look like

P=
dx0 ex = 1,
(9.97)
where we have replaced t0 by 0. It is to be expected that the particle decays with

probability 1 after infinite time.
A more rigorous treatment of decaying particles can be given, which is somewhat involved, since it should deal with infinite orders in perturbation theory.
Some of this is covered in Brown section 6.3. It turns out that the exponential
decay law itself is an approximation. However, eq. (9.96) is a useful and correct
result if the rate is not too large, or equivalently, if the lifetime 1 is not
too small, on the relevant time scales involved.
For a particle at rest (P = 0) we have already verified in Problem 2.3 that
p
, p |p1 |,
(9.98)
I(P ) =
4M
hence,
p
=
|T |2 ,
(9.99)
2
16M
where we can express p in terms of m and M . Recall that for a moving particle
the rate behaves M/P 0 , in accordance with relativistic time dilatation.
In a similar way it is possible to derive a formula for the total rate for a decay
1 2 + + n:
X
S
=
2p01 (2s1 + 1) 1 n
dp2 dpn (2)4 (p1 p2 pn )
|T (p2 2 , , pn n ; p1 1 )|2 ,
(9.100)
where S is a statistical factor taking care of identical particles in the final state.
The invariant decay amplitude is then given by the analog of (9.77),
q
T (p2 2 pn n ; p1 1 ) = Z1 Z2 Zn Ua2 (p2 , 2 ) Uan (pn , n )

Han a2 a1 (pn , , p2 , p1 ) Ua1 (p1 , 1 ),
(9.101)
Unstable particles really require a much more involved discussion, especially
in case the decay rate is a sizable fraction of the mass of the unstable particle.
See the books for more information, e.g. Brown, sect. 6.3, Peskin and Schroeder,
sect. 7.3, De Wit and Smith, sect. 3.6.
9.6
Example: the decays e + e and

+
In the following we consider the decay process + to illustrate how to

the formalism of this chapter can be used to deal with bound states.
9.6. EXAMPLE: THE DECAYS E + E AND +
231
QCD possesses the remarkable property of confinement: quarks and gluons

cannot occur as free particles, they form bound states, called hadrons, which
do occur as free particles (to a good approximation). Well-known hadrons of
low mass are the proton, neutron, the charged pions , and the neutral pion
0 . When the electroweak interactions are neglected, these particles are stable,
but in the Standard Model only the proton is stable (up to a negligible decay
probability related to so-called sphaleron processes).
For the description the we use composite pseudo-scalar fields,
(x) d (x)i5 u (x),
(x) = u (x)i5 d (x),
(9.102)
where the u,d fields are the up and down quark fields. The computation of
correlation functions such as
h0|T (x) (y)|0i h(x) (y)i G (x, y)
(9.103)
is a complicated problem in non-perturbative QCD, which can be done using

computers and the lattice regularization (lattice QCD). The Fourier transform
of this correlator has a pole at p2 = m2+ = m2 , corresponding to the matrix
elements
ipx
hp, + | (x)|0i =
Z e , hp, | (x)|0i = 0,
(9.104)
ipx
+
hp, |(x)|0i =
Z e , hp, |(x)|0i = 0
(9.105)
(the zeros follow e.g. from
conservation of electric charge, Qu = 2/3, Qd = 1/3,
12
Q+ = 1) . We assumed Z to be positive. Hence,
Z
x0 >y 0
h0|T (x) (y)|0i
x0 <y 0
G (p)
dp h0|(x)|p, +ihp, + | (y)|0i + m.p.(9.106)

dp h0| (y)|p, ihp, |(x)|0i + m.p.(9.107)
Z
d4 p ip(xy)
e
+ m.p., (9.108)
4
2
(2)
m + p2 i
Z
+ m.p.,
(9.109)
2
m + p2 i
where m.p. denotes the contribution from multi-particle intermediate states. The
value of Z can be obtained with the methods of lattice QCD, but we shall not
need this in the following.
For the decays, let us use an effective action that can be derived as an approximation to the Standard Model. It is given by
S =
12
d4 x LQCD + Lfree
lepton + L1 = S0 + S1 ,
(9.110)
Formally: [Q, ] = Q|0i = [Q, ]|0i = |0i, hence |0i has charge Q = 1.
232
L0 = LQCD + Lfree
lepton ,

Lfree
lepton = e ( + me )e e e + e ,
GF cos
e i (1 5 )e j + h.c. + e ,
L1 =
2
j = u i (1 5 )d .
(9.111)
(9.112)
(9.113)
(9.114)
Here LQCD is the QCD lagrangian in terms of quark and gluon fields, which will
not be specified further; Lfree
lepton is the free lagrangian for the electron and the
muon and the corresponding neutrinos, which are taken to be massless; L1 is
the relevant part of an effective lagrangian for the weak interactions, the fourFermi interaction, with GF the Fermi constant and the Cabibbo angle. The
electromagnetic interactions are neglected.
For the decay (p) e(k) + e (k 0 ) we need the correlation function
h0|T e (x) e (y) (z)|0i he (x) e (y) (z)i (i)2 Gee (x, y, z),
(9.115)
Perturbing in S1 we have
heiS1 e (x) e (y) (z)i0
(9.116)
he (x) e (y) (z)i =
heiS1 i0
Z
= i d4 u hL1 (u) e (x) e (y) (z)i0 + . . . (9.117)
where the . . . denote terms of higher order in GF and hi0 is the path-integral
average over the full QCD variables and the free lepton variables,
R
eiS0 F
.
iS0
QCD,lepton e
R
hF i0 = QCD,lepton
(9.118)
Performing the contractions of the lepton fields gives

iGF cos
ihL1 (u)e (x)e (y)(z)i0 =

(i)Ge (x, u)i (1 5 )(i)Ge (u, y)
2
hj (u)(z)i0
(9.119)
The second factor above is a correlator within QCD:
hj (u) (z)i0 = hj (u) (z)iQCD iGj (u, z),
(9.120)
since the path integral over the lepton variables is just a factor that cancels between numerator and denominator when F in (9.118) contains only QCD fields.
Putting things together and taking the Fourier transform, the three-point correlator becomes
(i)2 Gee (k, k 0 , p) =
iGF cos
(i)Ge (k) i (15 )(i)Ge (k 0 ) (i)Gj (p),

2
(9.121)
9.6. EXAMPLE: THE DECAYS E + E AND +
k
p

233

p

Figure 9.3: Left: Diagram for the three-point

function (i)2 Gee (k, k 0 , p). It
consists of the vertex function (iGF cos / 2)i (1 5 ), the electron propagator
iGe (k), the neutrino propagator iGe (k 0 ), and the correlator iGj (p). In
the approximation used here, the four-Fermi interaction, the wiggly line merely
indicates a vector index. In the Standard Model it would represent the Wpropagator. Right: the correlator iGj (p).
whith the diagrammatical representation in figure 9.3.
Consider next the j correlator. Choosing u0 > z 0 and inserting intermediate
states gives
u0 >z 0
h0|T j (u) (z)|0i =
dp h0|j (u)|p, ihp, |(z)|0i + m.p.,
(9.122)
where we exhibited the one-pion intermediate-state contribution. The matrix

element of the current can be parametrized, on grounds of Poincare invariance,
in terms of only one constant, f , which has the dimension of mass,
(9.123)
h0|j (u)(u)|p, i = ip f 2 eipu
(only the pseudo-vector part of j contributes, cf. problem 10.2). Hence,
u0 >z 0
h0|T j (u) (z)|0i =
dp ip f 2Z eip(uz) + m.p..
(9.124)
dp (ip )f 2Z eip(uz) + m.p.,
(9.125)
The other time-ordering gives

u0 <z 0
h0|T j (u) (z)|0i =
and so the correlator has the form
d4 p ip(uz) ip f 2Z
e
+ m.p., (9.126)
= i
(2)4
m2 + p2 i
ip f 2Z
+ m.p..
(9.127)
Gj (p) =
m2 + p2 i
h0|T j (u) (z)|0i
234
To obtain the H function, we need to strip off the external-line propagators

(correlators) from the three-point e
u correlator. There is no explicit correlator in (9.121), so we simply multiply by the inverse correlator:
iHee (k, k 0 , p) =
iGF cos
i (1 5 )(i)Gj (p) iG (p)1

2
(9.128)
Going on-shell,
G (p)1
p2 m2
Z 1 (m2 + p2 ),
(9.129)
which selects the pole part of Gj (p). Hence,
ip f 2
GF cos
i (1 5 )
,
H(k , k , p) =
2
Z
0
o.s.
and the decay amplitude becomes
T (k, ; k 0 0 ; p) = Z u
(k, )H(k, k 0 , p)v(k 0 , 0 )
= GF cos f p u
(k, ) (1 5 )v(k 0 , 0 ).
(9.130)
(9.131)
(9.132)
Alternatively, we could have used the divergence of an axial current to obtain

a pseudoscalar field
h
i
0 (x) = d (x)i 5 u (x) .
(9.133)
In this case the corresponding Z 0 in known in terms of f and m . The resulting

expression for T is the same.
9.7
Problems
1. Spin-one polarization vectors

Let a V be a vector field and |p, i basis vectors for spin-one particles.
Then the e (p, ) in
h0|V (x)|p, i = Z e (p, ) eipx

(9.134)
are the polarization vectors for this case. For a particle at rest we may
assume
e0 (
p, ) = 0, ek (
p, ) = k , = 1, 2, 3.
(9.135)
where
p = (m, 0).
(9.136)
For general p the polarization vectors can be defined by applying the special
Lorentz boost `p which transforms p into p:
e (p, ) = e (`p p, ) = `p e (
p, ).
(9.137)
235
9.7. PROBLEMS
a. Derive for a boost in the 3-direction that
|p|
,
e (p, 3) =
m
0
p0
e (p, 3) = k,3 ,
k
k
(9.138)
where p0 = p2 + m2 . With an eye on completing the two vectors e(p, )

used for the photon we may generalize this to
p0
|p|
, e(p, 3) = p ,
m
m
0
e (p, ) = 0, e(p, ) = as for photon, = 1, 2
(9.140)
e (p, )e (p, 0 ) = 0 .
(9.141)
e0 (p, 3) =
(9.139)
b. Verify
c. Verify
p e (p, ) = 0,
= 1, 2, 3.
d. Verify
P (p)
e (p, ) e (p, ) = +
(9.142)
p p
.
m2
(9.143)
(This can also be obtained directly by transformation of the polarization

sum at rest.)
e. Verify that
N (p, m) = +
p p
m2
(9.144)
satisfies (9.61).
2. Effective action for massive spin-one field
In this problem we look for an effective action for the field theoretic description of massive spin-one particles. We choose to describe the particles
with a vector field V . Suppose the vector field of the previous problem
is normalized such that Z = 1. If the field is free, then the multiparticle
contribution in its spectral representation wil vanish, and the propagator
for the free system is given by
G (p) =
+ p p /m2
N (p, m)
=
.
m2 + p2 i
m2 + p2 i
(9.145)
The inverse G1
(p) is hopefully a polynomial in p, such that we get a
differential operator in position space (p i ). This is indeed the
case:
a. Verify
[ (m2 2 ) + ]G (x y) = 4 (x y),
(9.146)
236
with G (x y) the Fourier transform of G (p).
b. Show using path integral quantization that the effective action

Z
1
1 2
4
S= d x
V V + m V V , V = V V , (9.147)
4
2
leads to the above propagator.
The above effective action should in first instance only be used for tree
diagram calculations. If we go beyond this and use it as a starting point for
a fundamental theory of interacting spin-one particles, then the fact that the
above propagator G (p) does not go to zero nicely for p causes a lot of
trouble with divergencies in loop diagrams, especially in non-abelian gauge
theories. These problems are resolved by invoking the Higgs mechanism as
the underlying reason for the mass of spin-one particles like W and Z in the
Standard Model. For their contribution to the theory of renormalization of
non-abelian gauge theories, t Hooft and Veltman received the Nobel price
in 1999. See, e.g. Veltmans book for more information on the relation
between gauge invariance, renormalizability and unitarity.
3. Calculation of matrix elements
The construction of transition amplitudes given in this chapter can be extended to matrix elements of local operators, e.g. the single particle matrix
element hf |T (x)|f i. In free scalar field theory, derive (2.133) using the
Wick formula.
4. The decays + and e + e
The charged pions are unstable and decay mainly into muons and
muon neutrinos (
) , with a life time of 2.60 108 s, or 1 = 780 cm.
There is a corresponding decay into electrons e and electron neutrinos
(
e )e , with a much smaller rate. These processes can be described by
an effective action of the form S = S0 + S1 , where S0 is the sum of the
actions for the free pions,13 muons, electrons, muon neutrinos and electron
neutrinos,
S0 = S + S + S e + S + S e ,
S =
S =
(9.148)
d4 x ( + m2 ),
(9.149)
d4 x ( + m ) ,
(9.150)
and similar for e, and e with m = me = 0, and S1 is the interaction

S1 = c
d4 x [ i (1 5 ) + i (1 5 )
+ ( e)].
13
The are described by the complex scalar field .
(9.151)
237
9.7. PROBLEMS
The constant c is given by
c = f GF cos C ,
(9.152)
with f the pion decay constant, GF the Fermi weak interaction constant
and C the Cabibbo angle.
Notice that the interaction S1 does not conserve parity P , as it is the sum
of terms odd and even under parity.
a. Draw the diagram for the decay (p) (k, ) + (k 0 , 0 ) and verify
that the decay amplitude is given by
hk, k 0 0 out|p ini = i(2)4 (p k k 0 ) T (k, , k 0 , 0 ; p), (9.153)
T (k, , k 0 , 0 ; p) = c
u(k, )p(1 5 )v(k 0 , 0 ).
(9.154)
b. Verify the polarization sum
1
|T |2 = c2 Tr [p(1 5 )(ik 0 )p(1 5 )(m ik)].
2
(9.155)
(The neutrino has only one spin polarization.)

c. Using the anticommutation relations of the gamma matrices, the properties of the right and lefthanded projectors PR,L = (1 5 )/2) and the
trace identities, show that
|T |2 = 4c2 [2(pk)(pk 0 ) p2 kk 0 ].
(9.156)
d. In the rest frame of the pion, verify

m2 m2
|k| =
,
2m
m2
k |k| =
,
m
0
(9.157)
and
|T |2 = 2c2 (m2 m2 )m2 ,
and
( + ) =
c2 (m2 m2 )2 m2
.
4
m3
(9.158)
(9.159)
e. The masses of the particles are given by m = 139.6 MeV, m = 105.7

MeV, me = 0.5110 Mev (the neutrino masses were already assumed to be
zero). Using GF 1.17 105 GeV2 , C 13 , and the fact that
decays for 99.988% into + , verify that f 93 MeV from the rate
= (780cm)1 .
238
f. Calculate the branching ratio
( e + e )
( + )
(9.160)
and compare this with the experimental value 1.22 104 .
The striking smallness of the above branching ratio is a consequence of

the combination (1 5 ) in the interaction S1 . The interaction conserves
chirality: 15 projects on to chirality 1, in the neutrino fields as well as in
the electron or muon fields (recall that contains = i 0 and 5 commutes
with i 0 ). For the massless antineutrinos, chirality 1 means helicity
+1/2. For the electron and muon, chirality 1 would mean helicity 1/2
if these particles were massless. However, angular momentum conservation
requires that the muon or electron have the same helicity (+1/2) as the
antineutrino, since the pion at rest has angular momentum zero. Hence, if
m and me would be zero, the decay amplitude would vanish (1 5 acting
on a massless helicity +1/2 particle spinor gives zero). So we may expect
that the decay amplitude is proportional to m,e as m,e goes to zero. In
fact, it can be shown using helicity spinors that the decay amplitude is
given by
q
T = 2icm |k| ( k 0 + m k 0 m) ,0 0 ,+ ,
(9.161)
with k 0 = k2 + m2 and m = m or me . In this way we can understand

why the above branching ratio m2e /m2 is so small.
Chapter 10
C, P, and T
In addition to Lorentz invariance, the discrete symmetry transformations C (charge
conjugation), P (parity) and T (time reversal) are of crucial importance in particle physics and cosmology. We shall give a brief introduction here which can
only scratch the surface, as a proper presentation of the important experimental
implications and results would lead much too far. For more information, see e.g.
the books by Weinberg, Perkins, Branco et al., Sanda & Bigi.
10.1
Charge conjugation
We have seen that the QED action is invariant under global (and of course local)
U(1) transformations. In Majorana language this means SO(2) invariance, in
which the real and imaginary parts 1,2 of the Dirac field get rotated into each
other, as in (8.3). In fact, the free action is invariant under the full O(2) group,
obtained by adding reflections,
1 1 ,
2 2 ,
(10.1)
which are orthogonal transformations in internal space with determinant minus

one. For the Dirac field this means , in a Majorana representation of the
-matrices. In a general representation we have to be a bit careful, because
does not transform like under Lorentz transformations. However, as seen in
Appendix 5.5, the charge-conjugate spinors
T,
(c) = C = (C)
(c) = (C )T
(10.2)
do transform in the same way. Recall that C is unitary and antisymmetric with
T = C C,
C C = ,
and recall also our convention C = 1 in a Majorana representation.

239
(10.3)
240
CHAPTER 10. C, P, AND T
It is straightforward to check that the transformation C defined by

T,
(C)
C:
(C )T ,
(10.4)
leaves the free Dirac action invariant. We can extend this symmetry to QED if
we adopt as the transformation rule for the electromagnetic field:
C:
A A .
(10.5)
Let us check this for the interaction term in the Lagrange density:
,
L1 = ej A , j = i
T
T = Ci
= j ,
j (C )T i (C)
C = i
L1 e(j )(A ) = L1 .
(10.6)
(10.7)
(10.8)
which gives a minus sign by their

In the second line we interchanged and ,
anticommuting property, and then used (10.3). This symmetry of the action
under C implies, via the path integral, a unitary operator UC in the quantum
Hilbert space (or in the extended negative metric space, in case of a covariant
gauge). The derivation of this is cumbersome (we also have not discussed yet
fermionic coherent states) and it is simpler to guess the answer directly. Let the
unitary operator C have the properties
T,
UC UC = (C)
UC A UC = A ,
= (C )T ,
UC U
C
(10.9)
(10.10)
It is straightforward to check that UC leaves the canonical (anti)commutation

relations invariant (otherwise it could not be unitary), as well as the total hamiltonian,
UC HUC = H,
H =
[UC , H] = 0,

1 2 1 2
0 0
3
E + B + HD + ej A ejA ,
d x
2
2
(10.11)
(10.12)
where HD is the Dirac hamiltonian (5.85). So UC is time-independent and we

may assume that it leaves the vacuum invariant,
UC |0i = |0i.
(10.13)
We end this section with miscellaneous remarks:

- Products of fermion fields in operators should be antisymmetrized, such
that they may be freely interchanged as anticommuting (the delta functions
from the canonical anticommutation relations cancel). An example is the
current operator
1
1
(10.14)
j = i
T i T T ,
2
2
241
10.1. CHARGE CONJUGATION

which satisfies
UC j UC = j
(10.15)
also in operator language.

- C interchanges particles and antiparticles:
UC b(p, )UC = d(p, ),
UC d(p, )UC = b(p, ).
(10.16)
- Neutral states have a definite C-parity:

UC a (k, )|0i = a (k, )|0i,
UC b (p, )d (p0 , 0 )|0i = b (p, )d (p0 , 0 )|0i.
(10.17)
(10.18)
The photon has negative negative C-parity.

- The C-parity has to be the same in initial and final states.1
- The Yukawa model with pseudoscalar coupling is an example of an effective
theory for neutral pions and protons. The interaction lagrangian
5
L1 = g i
(10.19)
UC UC = +.
(10.20)
is invariant under C if
The neutral pion described by the field has positive C-parity. It decays
primarily electromagnetically into two photons and cannot decay into an
odd number of photons, because of C-parity conservation.
- QED and QCD are C-invariant, but the full Standard Model violates C.
- For nonabelian gauge fields, as in QCD, the coupling to fermions can be
written in the form
A ,
g i
(10.21)
where g is a coupling constant and A is a matrix acting on additional
(color or flavor) indices of the fermion fields. The charge conjugation
transformation for A is now given by
UC A UC = AT .
1
Neglecting the C-violation caused by the weak interactions.
(10.22)
242

- We may make an additional global U(1) transformation in the definition of
UC ,
T , UC U
= (C )T ,
UC UC = (C)
(10.23)
C
with || = 1. This puts phase factors in (10.16) but does not affect the
minus sign in (10.18).
As a generalization of this, it is always possible to change the definition of a
discrete transformation like UC by making an additional transformation of
variables, which may or may not be a symmetry transformation (the same
holds for P and T). This may turn an apparently noninvariant model into an
invariant one. It is clearest to make such transformations of variables first,
putting the action into some convenient form. This is commmon practise
in the Standard Model, e.g. for the Cabibbo-Kobayashi-Maskawa matrix.
10.2
Parity
We have encountered parity already several times. It represents the improper

Lorentz transformation (x0 , x) (x0 , x),
x0 = `P x, `P = diag(1, 1, 1, 1),
DP = D(`P ) = 0 , DP1 = 0 = DP ,
(10.24)
(10.25)
where DP is a spinor representation. In terms of a unitary operator UP the

transformation rules read in operator language
UP (x)UP = DP (`P x),
UP A (x)UP = `P A (`P x)
UP (x)U
P = (`P x)DP ,
(10.26)
(10.27)
(with identical transformation rules for Majorana fields.) Note that `P = `1

P
(but for our choice, DP 6= DP1 ).
Miscellaneous remarks:
- The QED hamiltonian (10.12) is parity invariant. Experimentally, QED
and QCD are parity invariant, e.g. there is no indication for non-invariant
terms like F F in the lagrangian.
- It follows from the group properites of ` and `P that momentum and angular
momentum are respectively odd and even under parity:
UP PUP = P,
UP JUP = +J.
(10.28)
- Hence, a particle at rest can be in an eigenstate of UP . The eigenvalue

is called the (intrinsic) parity of the particle. For single fermions with a
243
10.3. TIME REVERSAL

charge, we can choose
UP b (
p, )|0i = ib (
p, )|0i,
UP d (
p, )|0i = id (
p, )|0i,
p = (m, ~0)
(10.29)
(10.30)
this is convention dependent, because we can always put an additional phase

in the parity transformation, similar to (10.23) for C. However, these phases
cancel in a fermion-antifermion state, for which the resulting minus sign is
convention independent. A fermion-antifermion in a spatially symmetric
bound state with total momentum zero has negative parity. The neutral
pion, a quark-antiquark state, is an example. Its effective field transforms
as (x) (`P x) under parity. The effective Yukawa model (10.19)
illustrates the properties of 0 under both C and P.
A neutral fermion (describable by a Majorana field) can have parity i.
- The following interaction cannot be invariant under C or P,
(gV + gA 5 ) A ,
L1 = i
(10.31)
if the coupling constants gV and gA are both non-zero, but it is invariant

under the product CP. This illustrates the situation in the Standard Model.
10.3
Time reversal
The time reflection is defined by (x0 , x) (x0 , x), or
x0 = `T x, `T = diag(1, 1, 1, 1),
DT = D(`T ) = i 0 5 , DP1 = i 0 5 = DP ,
(10.32)
(10.33)
where DT is a spinor representation. It turns out that in Hilbert space this

symmetry is represented by an antiunitary operator UT , i.e. a unitary operator
that is antilinear.2 To understand this we first turn to non-relativistic quantum
mechanics.
Consider a lagrangian of the conventional type L(q, q)
= 12 mq2 V (q). It
is invariant under the transformation q 0 (t) = q(t). The equations of motion
are invariant under time reversal. If we have a solution q(t) satisfying q(t1 ) = q1 ,
q(t2 ) = q2 , then q 0 (t) = q(t) is also a solution, with boundary values q 0 (t1 ) = q1
and q 0 (t2 ) = q2 . If the velocity q happens to be positive throughout the whole
trajectory, then q0 is negative.3 The symmetry of L has consequences in the path
integral. To cut the discussion short, consider the semiclassical approximation
Z
hq1 , t1 |q2 , t2 i exp i

2
3
t1
t2
dt L[q, q]
,
(10.34)
Recall that for an antilinear operator A and complex number , A|i = A|i.
A time-honoured German name for T is Bewegungsumkehr (motion-reversal).
244
where q(t) is a classical solution as mentioned earlier. Recall that |q, ti =

U (t, 0)|qi is the eigenvector of the Heisenberg operator q(t) with eigenvalue q.
Substituting q(t) = q 0 (t) we see that the right-hand side of (10.34) turns into
Z
exp i
t2
t1
dt L[q , q ] hq2 , t2 |q1 , t1 i.
(10.35)
The consequence of T-invariance is evidently

hq1 , t1 |q2 , t2 i = hq2 , t2 |q1 , t1 i.
(10.36)
The interchange of initial and final states is intuitively appealing.

Next we introduce UT by
UT |q, ti = |q, ti,
UT eiHt |qi = eiHt |qi.
(10.37)
We want it to represent the property (10.36), i.e.

hq1 , t1 |q2 , t2 i = (hq2 , t2 |UT ) (UT |q1 , t1 i).
(10.38)
Using h|i = h|i on the right-hand side this gives the relation
hq1 , t1 |q2 , t2 i = [(hq1 , t1 |UT ) (UT |q2 , t2 i)] .
(10.39)
Note the parenthesis, which are needed here in the bra-ket notation. Usually we
are dealing with linear operators, for which it does not matter if we interpret
the action of an operator in the bra-ket notation to act to the right or the left:
h| (O|i) = (h|O) |i. However, this cannot be true for UT . We shall now show
that it has to be antiunitary.
First, choosing t1 = t2 = 0 the amplitude hq1 , 0|q2 , 0i = hq1 |q2 i is real. For
R
any state |i = dq (q)|qi, the norm h|i = h|i is real, and we get
(h|UT )(UT |i) = h|(UT UT |i) = h|(UT UT )|i, where it does not matter if
UT is linear or antilinear. Since |i is arbitrary this means that UT is unitary,
UT UT = 1.
(10.40)
Note that the product UT UT is linear. Secondly, for general t1,2 , the amplitude
(10.39) is complex, and letting UT act to the right and using UT UT = 1 has to
produce a complex conjugate, such that it equals the left hand side of (10.39).
So UT should satisfy
(h|UT ) |i = [h| (UT |i)] ,
(10.41)
for arbitrary and (from which (h|UT ) |i = [h| (UT |i)] follows by conjugation). From (10.41) follows that UT is antilinear: let be a complex number
h|(UT |i) = [(h|UT ) |i] = [(h|UT ) |i] = h|(UT |i),
(10.42)
245
10.3. TIME REVERSAL

so
UT |i = UT |i,
U T = UT .
(10.43)
For t = 0 we see from (10.37) that UT acts as the identity on the coordinate-basis
|qi,
UT |qi = |qi.
(10.44)
So on this basis it is just the operator of complex conjugation. It follows that in
the momentum basis
UT |pi = UT
dq eipq |qi =
dq eipq |qi = | pi.
(10.45)
For general t the relation (10.37) and the unitarity of UT gives

UT exp[iHt] |qi = exp[it UT HUT ] |qi = exp[iHt] |qi,
(10.46)
so the hamiltonian is invariant,

UT HUT = H,
[UT , H] = 0.
(10.47)
For the canonical Heisenberg operators we get

UT q(t)UT = q(t),
UT p(t)UT = p(t).
(10.48)
Having familiarized ourselves with the antilinear nature of UT we turn to QED,

using operator notation. For Majorana fields with a Majorana representation of
the Dirac matrices it is natural to define the T transformation as
UT (x)UT = DT (`T x),
UT T (x)UT = T DTT = T DT1 .
(10.49)
For the complex

Dirac field build out of two Majorana fields 1,2 according to
= (1 i2 )/ 2, this UT would imply interchanging and , as in charge
conjugation. This looks intruiging, reminding us of the idea that charged particles
running backwards in time are just antiparticles running forward (cf. Feynman).
However, the conventional definition of T for Dirac fields does not interchange
and , so let us undo this by following the transformation (10.49) by charge
conjugation. Thus defining T we get
UT (x)UT = C DT (`T x),
UT A (x)UT = `T A ,
UT (x)U
T = (`T x)DT C (10.50)
(10.51)
where we also included the transformation for the electromagnetic field. To see
what this does, consider the transformation of the current (suppressing the antisymmetrization as in (10.14)),
T x)D 1 C (i ) C DT (`T x)
UT j (x)UT = UT (x)i
(x)UT = (`
T
= `T j (`T x),
(10.52)
246
where we used (10.3). With this new definition of T the charge does not flip sign,
but the current gets inverted: j 0 (x, t) j 0 (x, t), j(x, t) j(x, t). The total
QED hamiltonian (10.12) is T-invariant, UT H(t)UT = H(t) = H(t), because it
is time-independent.
Miscellaneous remarks:
- UT changes the signs of both momentum and angular-momentum operators.
- Eigenvalues of UT have no special significance, their phase can be changed
by a phase transformation of the state vector. However, UT2 is linear and its
eigenvalues have physical significance. For a fermion state |p, i it is 1.
- UT changes in-states into out-states (cf. chapter 9),
1 , , pn ,
n outi,
UT |p1 , 1 , , pn , n ini |
p1 ,
1 , , pn ,
n ini,
UT |p1 , 1 , , pn , n outi |
p1 ,
(10.53)
(10.54)
where p = (p0 , p), and the constant of proportionality is a phase factor.

- The experimental detection of T-violation is difficult, because the product
CPT seems to be an exact symmetry of Nature (see next section) and CPviolation turns out to be effectively very small.
10.4
CPT
The CPT theorem states that in a local, Lorentz invariant and unitary (hermitian
hamiltonian) theory, all interactions are invariant under the product transformation CPT. This is just our original UT for Majorana fields in (10.49) times UP .
We shall demonstrate CPT invariance here by showing that any lagrange density L satisfying the assumptions, keeping for simplicity only scalar, spinor and
vector fields, transforms as L(x) L(x). This then leads to a CPT invariant
hamiltonian.
Let us use Majorana language (Majorana fermion fields and a Majorana representation for the Dirac matrices, and real bosonic fields), which is most general, as
any Dirac structure can be translated into this form. It also helps avoiding overlooking less familiar possibilities, such as so-called Majorana interactions which
do not allow any charge conservation. Consider a theory with K real Majorana
fields , L real scalar fields, k , and M real vector fields Ar , = 1, . . . , K,
k = 1, . . . , L, r = 1, . . . , M . Under U UP UT these fields transform simply as
U k (x)U = k (x),
U Ar (x)U = Ar (x),
U (x)U = i5 (x),
(10.55)
247
10.5. PROBLEMS
where we used `T `P = 1, DT DP = i5 and did not introduce any other phase

(sign) factors. For the fermion bilinears it follows that
U T (x) 1 n 0 (x)U = (1)n T (x) 1 n 0 (x).
(10.56)
Hence, these transform just like tensors build out of the Bose fields, e.g.
U Ar (x)U = (1)2 [/(x )]Ar (x). The fermion bilinears are to be antisymmetrized, a b (a b b a )/2, if necessary, such that the field
operators can be freely anticommuted as if they were Grassmann numbers. Then
they are hermitian, and either symmetric or antisymmetric under exchange of
the labels and 0 . Corresponding coupling constants (or mass matrices) have
to have the same symmetry. They have to be real in order that L be hermitian. Hence, they are not affected by the antilinearity of U . It now follows that
any Lorentz invariant multinomial in the fields with real coupling constants is
CPT-covariant. For example in

1 T
1 T
0
i5 0 Ar
0 T0 T T Ar + g
0r
4
2
1
1
+ h0 k T 0 k + h00 k T i5 0 k .
(10.57)
2
2
L1 = g0 r
the couplings have to satisfy
g0 r = g0 r = g
0r,
h0 k = h0 k = h0 k ,
0
0
0
g
0 r = g0 r = g0 r ,
h00 k = h00 k = h0
0 k ,
(10.58)
(10.59)
and L1 is hermitian and covariant under CPT, U L1 (x)U = L1 (x). Note that
terms involving the epsilon tensor are also invariant, because det(`P `T ) = 1,
similarly the presence of 5 = i /4! does not affect CPT invariance.
Finally, we note that the somewhat mysterious presence of C in the usual
treatments of the CPT transformation, which is essentially just x `P `T x = x,
has disappeared in the Majorana formulation used here: there is no trace of C in
the natural U = UP UT .
An important consequence of CPT-invariance is:
- particles and antiparticles have the same mass and lifetime, even when C,
P and T are all violated (as they are in Nature).
10.5
Problems
1. Decay of the neutral pion

The neutral pion decays primarily electromagnetically into two photons:
0 (p) (k, ) + (k 0 , 0 ). This process can be described in terms of an
effective action, the form of which depends on the parity of 0 . In the
248

following we assume not to know the parity of 0 and intend to investigate
how it might be determined. Candidate gauge invariant interactions are
given by
Z
C
S1 =
d4 x (x) F (x)F (x),
(10.60)
m
and
Z
C
S2 =
d4 x (x)F (x)F (x).
(10.61)
m
Here = e2 /4 is the fine structure constant, m the 0 mass, C a constant
and the Levi-Civita tensor with 0123 = 1.
a. How should transform for S1 to be invariant under C, P, and T?
b. How should transform for S2 to be invariant under C, P, and T?
c. Compare your results for a,b with the transformation properties of the
and i
5 .
composite fields
d. Calculate the vertex function
S
3S
(u, v, v )
,
(u)A (v)A0 (v 0 )
0
(10.62)
for S1 as well as for S2 .

e. Determine the corresponding vertex functions in momentum space,
S0 (p, k, k 0 ).
f. Draw the Feyman diagrams for the decay process 0 (p) (k, ) +
(k 0 , 0 ) in the tree-graph approximation.
g. Write down the corresponding decay amplitude T (k, k 0 0 ; p) for both
cases 1 and 2.
h. Suppose you can observe correlations between the photon polarizations
and their momenta. Discuss how you might determine the pion parity from
the decay distribution of the photons.
i. Calculate |T |2 in both cases. For 1 the identity = 2(
) is useful.
j. Calculate the unpolarized distribution d/d and the total rate in the
pion rest frame.
2. The pion decay constant f
The positively charged pion decays primarily through the process
+ + + . Its decay amplitude depends on the matrix element of the
current
j = d i u d i 5 u V A
(10.63)
249
10.5. PROBLEMS
between the vacuum state and the one-pion state in QCD; V is called the
(polar-)vector current, A the axial-vector current.
a. Verify that the following parametrization of these matrix elements has
the general form compatible with Poincare invariance:
h0|A (x)|p, + i = ip f 2 eipx ,

(10.64)
ipx
+

h0|V (x)|p, i = ip f 2 e .
(10.65)
Hint: use the Poincare in variance of the vacuum, (9.12) for the scalar
case and (9.39) for the vector case; insert U (`) U (`) = 1 into the matrix
elements, and similar for translations.
Verify that the constants f and f have the dimension of mass.
b. Parity is a good symmetry in QCD, which implies UP |0i = |0i. The
pions are known to be pseudo-scalars, i.e.
UP |p, + i = |`P p, + i.
(10.66)
Verify that because of parity, f = f , hence f = 0, but that there is no

such constraint on f .
Hint: insert UP UP = 1 into the matrix elements.
c. Charge conjugation is a good symmetry in QCD. Use C to obtain the
relations for with the corresponding currents.

Introduction To Relativistic Quantum Fields

Uploaded by

Copyright:

Available Formats

Introduction To Relativistic Quantum Fields

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Relativistic Quantum Fields

Uploaded by

Copyright:

Available Formats

Introduction to Relativistic Quantum Fields

Lecture notes 2007

Why Field Theory? . . . .

2 Quantized scalar fields

4 Perturbation theory, diagrams and renormalization

6.10 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

Why Field Theory?

0.2. THESE NOTES

These lecture notes present an introduction to relativistic quantum field theory,

Books on classical electrodynamics and general relativity:

Formal = jargon for pretty but mathematically imprecise.

Books on relativistic QFT:

S. Weinberg, The Quantum Theory of Fields, III, Cambridge University

Books on quantum field theory and critical phenomena:

The following books are specifically on lattice field theory:

Books on (specialized topics in) particle physics:

A book on the Casimir effect:

Conventions and notation

The following conventions will be used:

(i = [Bjorken&Drell], 5 = 5 [Bjorken&Drell].) Left and right handed

Depending Ron the context, p0 can be an arbitrary variable (e.g. a dummy

In terms of F , Maxwells equations can be written in Lorentz invariant form,

In terms of A the inhomogenous Maxwell equations (1.3) take the form

The terminology: Ra current j (x) is conserved, simply means: j (x) = 0. It is of course

CHAPTER 1. CLASSICAL FIELDS

An important aspect of this description is gauge invariance: F is invariant under

Lorentz gauge (also called Landau gauge),

In the Lorentz gauge (1.6) reduces to

which are four hyperbolic wave equations for A (recall 2 = 2 02 ).

In General Relativity the gravitational field is described in terms of the metric

1.3. SCALAR FIELD

So m is the frequency k 0 for zero wave vector, k = 0. Another example is the

CHAPTER 1. CLASSICAL FIELDS

is a representation of the generators P .

Recall that we use a metric tensor

leave invariant the quadratic form x2 = x x = x x ,

We can write the transformations also in matrix form,

CHAPTER 1. CLASSICAL FIELDS

Eq. (1.31) reads in matrix notation

where T denotes transposition. Note the subtlety in the connection between

where cosh = , = 1/ 1 v 2 , sinh = v, with v the velocity of the

Fields (x) transform as

where we used (1.34). It is now straightforward to check the Lorentz invariance

where is infinitesimal (note that = ). The relation (1.31) is satisfied if

CHAPTER 1. CLASSICAL FIELDS

(cf. (1.35) and Problem 10; infinitesimally = v).

A finite transformation can be written as

If we have a set of matrices S satisfying the commutation relations (1.53)

We end this introduction with the commutation relations of the generators

we see that we have the representation

CHAPTER 1. CLASSICAL FIELDS

x, it is an external source. Consider the variation of S under a small variation

where is the nonrelativistic analog of J.

where J is an external current and the integral is again over M . Under a

The action for the gravitational field without matter is given by

b) replace derivatives by covariant derivatives.

CHAPTER 1. CLASSICAL FIELDS

for the scalar field and

for the electromagnetic field, where F = g g F , etc. Specializing to Minkowski

The integral over the total divergence in (1.98) vanishes because = 0 on M

where > 0 is infinitesimal, i.e. it is arbitrarily small, positive, and will be

Then, to first order in ,

For later use we have added a constant to (1.80). It may be interpreted as