Chapter 6

Dust scattering off particles

6.1 Overview
The theory of scattering off dust particles is a quite advanced topic. There are entire
books devoted to this topic. The most famous of these books are:

• H.C. van de Hulst, “Light Scattering by Small Particles”, Dover Books on

Physics
• C.F. Bohren and D.R. Huffman, “Absorption and Scattering of Light by Small
Particles”, John Wiley and Sons

The main objective is to calculate the scattering and absorption cross sections at a
given wavelength, for a particle of a given size and shape, made from a given type of
material. In addition to this, the scattering cross section is angle-dependent, and tends
to polarize the light.
The calculation of these quantities requires us to solve the classical equations of elec-
trodynamics inside and outside of the dust grain. The dust grain is assumed to be made
of dielectric material, so the material reacts to the presence of an electromagnetic field
by becoming polarized. In other words: we consider the material to be made up of an
infinite number of infinitely small electric and magnetic dipoles who’s dipole strengths
are proportional to the imposed field strengths. The induced dipoles create their own
field in return. If the imposed field is an electromagnetic wave, the induced field is
also an electromagnetic wave, i.e. the dust partical emits its own waves in reaction to
the imposed waves. This is the process that is called scattering.
The properties of the material (how strongly the electric and magnetic dipoles react to
the imposed field) are fully given by the complex index of refraction

m(λ) = n(λ) − ik(λ) (6.1)

where n(λ) and k(λ) are the real and the imaginary part of the complex index of re-
fraction. They have to be given as a function of λ. Typically they are measured in the
laboratory for a given material, and then provided to the community as a table for a
finely sampled set of wavelengths. For common astrophysical materials you can find
such tables at the Jena Optical Constants database1 .
Roughly one can interpret n and k as the scattering and absorption part of the refractive
index. A particle with k = 0 (at some wavelength) will not absorb light, but it will
scatter light. In reality k will never be exactly 0, because all materials absorb a bit.
But k can easily be of order 10−8 or less for e.g. water or glass at visual wavelengths.
1 http://www.astro.uni-jena.de/Laboratory/Database/databases.html

92
Material with n = 1 and k = 0 would be invisible, since it would have the same optical
properties as vacuum. Typically n has values not much smaller than 1 and not much
bigger than a few.
Another fundamental parameter is the ratio of the size of the particle over the wave-
length:
2πa
x= (6.2)
λ
where a is the radius of the particle. Of course, this parameter is only well-defined for
spherical particles, and in general particles can have very irregular shapes. Neverthe-
less, it gives a rough idea of how the particle size and the wavelength compare. We
can distinguish three regimes:

• x " 1 (the Rayleigh regime): The particle is much smaller than the wavelength.
The scattering process in this regime is called Rayleigh scattering, for which
there is a simple theory.
• x # 1: The particle size is similar to the wavelength. This is the most complex
regime, and requires the full solution of the Mawell equations.
• x $ 1 (geometric optics regime): The particle is much larger than the wave-
length, so that it can be regarded in the geometric optics regime. This does not
mean that its scattering is simple: reflection on the surface and refraction in the
interior can still be quite complex (e.g. light passing through a rain drop), but it
can be calculated using e.g. ray-tracing through the particle and off the particle’s
surface.

Calculating the scattering and absorption cross sections for all wavelength regimes is
complicated. For homogeneous spheres this was first done in its full complexity by
Gustav Mie. This calculational method is called Mie theory. For the x " 1 regime Mie
theory produces the same scattering cross sections as Rayleigh theory, as it should.
For x $ 1 the theory becomes, however, more and more difficult to apply, since it
involves a series expansion that requires ever more terms. Therefore, for x $ 1 it, at
some point, becomes more practical to use geometric optics.
Mie Theory is only applicable to spherical particles. The T-matrix method is its ex-
tension to non-spherical particles, though this method works best for still relatively
simple shapes such as ellipsoids. A book that describes this method in detail is:

• Mishchenko, Travis & Lacis “Scattering, absorption and emission of light by

small particles”, Cambridge Univ Press2

For complex-shaped particles a fully numerical method has to be employed to solve

the Maxwell equations. One such method is the Discrete Dipole Approximation
(DDA). For a list of publicly available codes see e.g. the Wikipedia page on DDA
codes3 .
To give a complete overview of scattering theory, even of the basics, would require far
too much time for this lecture. We will therefore keep this chapter very summary.

6.2 Rayleigh scattering

For small particles (2πa " λ) we can calculate the scattering cross sections according
to Rayleigh scattering theory. We will not derive it, but instead quote the results from
the books.
2 http://www.giss.nasa.gov/staff/mmishchenko/books.html
3 http://en.wikipedia.org/wiki/Discrete_dipole_approximation_codes

93
The scattering efficiency factor for Rayleigh scattering is
"4 # #2
8 2πa ### m2 − 1 ###
!
Qsca,Ray = ## 2 (6.3)
3 λ m + 2#
#

where m is the complex index of refraction. This means that the scattering cross
section of a single particle is
# #2
2 5 (2a)6 ### m2 − 1 ###
σsca,Ray = π (6.4)
3 λ4 # m2 + 2 #
# #

The scattering opacity of small particles in the Rayleigh regime therefore goes roughly
as 1/λ4 , modulo the wavelength-dependence of m.
The scattering phase function (see Section 5.1.5) is given by
3
Φ(cos Θ) = (1 + cos2 Θ) (6.5)
4
$ +1
(normalized such that 21 −1 Φ(µ)dµ = 1. Note that this is the same scattering phase
function as for electron scattering.
The absorption efficiency factor is
! " ! 2 "
2πa m −1
Qabs,Ray = 4 Im 2 (6.6)
λ m +2
meaning that the absorption cross section of a single particle is
(2a)3
! " ! 2 "
m −1
σabs,Ray = π Im 2 (6.7)
λ m +2
The absorption opacity of small particles in the Rayleigh regime therefore goes
roughly as 1/λ, modulo the wavelength-dependence of m.

6.3 Mie Theory

Understanding Mie theory is not an easy task, and the derivation of the equations is
somewhat elaborate. Here I merely discuss the basic idea behind it, and I refer to the
book by Bohren & Huffman (see above) for details.
We consider a plane electromagnetic wave in vacuum with a wavelength λ, incident on
a sphere of radius a with a complex refractive index m. The interaction of this wave
with the sphere causes the sphere to radiate electromagnetic waves of itself. These
waves are not isotropic in general. We can expand this outgoing wave using vector
spherical harmonics. The idea is the same as for expanding a scalar field in spherical
harmonics, but now for a vector field. Like with spherical harmonics, this involves
Legendre polynomials and Bessel functions. The expansion coefficients are written as
an and bn , where n is the expansion index going from n = 1 to n → ∞. These an and
bn are called the scattering coefficients.
The next step is to also expand the incoming plane wave into the same set of vector
spherical harmonics. These expansion coefficients are usually written as cn and dn .
The goal is now to link an and bn (the outgoing wave) to cn and dn (the incoming
wave). This can be done by applying the usual boundary conditions for the electric
and magnetic fields at the edge of the sphere, with which we are familiar from our
lectures on electromagnetism. This can be done for each value of n separately. The
result is:
mψn (mx)ψ'n (x) − ψn (x)ψ'n (mx)
an = (6.8)
mψn (mx)ξn' (x) − ξn (x)ψ'n (mx)
ψn (mx)ψ'n (x) − mψn (x)ψ'n (mx)
bn = (6.9)
ψn (mx)ξn' (x) − mξn (x)ψ'n (mx)

94
where ψ and ξ are the Riccati-Bessel functions4 .
Now that we know the full expansion of the outgoing wave, we can calculate the
scattering cross section, and find:
∞
2π % & '
σscat = 2
(2n + 1) |an |2 + |bn |2 (6.10)
k n=1

where k = 2π/λ is the wave number of the incident wave. The full extinction coeffi-
cient (scattering + absorption) is:
∞
2π %
σext = (2n + 1)Re {an + bn } (6.11)
k2 n=1

From this you can calculate the absorption cross section as

σabs = σext − σsca (6.12)

The larger the particle is compared to the wavelength, the more terms have to be
included in the sum. Typically, for x ! 1, one needs n # x terms. For large particles
(x $ 1) this may require a lot of computing time. Good Mie codes take particular care
of various possible numerical problems arising due to finite precision of the computer,
which could lead to wrong results for large sums of terms. A good code is the BHMIE
code of Bohren & Huffman, a version of which can be downloaded from the website
of B. Draine5 .

6.4 Describing polarized light with a Stokes vector

Scattering tends to polarize light. Also, the scattering cross sections may be different
for differently polarized light. So we must, for a more accurate study of scattering,
introduce a method for describing polarized light. To do this we introduce the Stokes
vector. Let us do this step by step, starting from the electromagnetic wave description
of light.
Perfectly coherent light propagating into a direction n can be described by its fre-
quency ν, the amplitudes of the left- and right- polarized components (or equivalently
the horizontal- and vertical- polarized components) and their phase difference. The
absolute phase (which varies on a time scale 1/ν) is for our purpose irrelevant. To
make this quantitative we must introduce e x' and ey' basis vectors perpendicular to
n. In doing so we have various choices. Most importantly: we choose globally a
right-handed coordinate system. We choose e x' and ey' such that

e x ' × ey ' = n (6.13)

This means that if we look into the beam of radiation (the radiation coming toward
us) we can define the x' -direction to be “right” and the y' -direction to be “up”. The n
vector can also be written as ez' .
At some fixed point P in space the electric field components for a perfectly coherent
plane wave can be written as

E x' (P, t) = E x' ,0 cos(ωt − ∆ x' ) (6.14)

Ey' (P, t) = Ey' ,0 cos(ωt − ∆y' ) (6.15)

where ∆ x' and ∆y' are phase lags. The phase lag between the y' and x' -fields is ∆ =
∆y' − ∆ x' , meaning that for positive ∆ the y' -field lags behind the x' -field. The full
4 The Riccati-Bessel function ψ is also often written as S , and is given by ψ (x) = x j (x) in terms of
n n n n
the spherical bessel function jn (x).
5 http://www.astro.princeton.edu/ draine/scattering.html
˜

95
plane wave can be written as:

E x' (x, t) = E x' ,0 cos(ωt − k · x − ∆ x' ) (6.16)

Ey' (x, t) = Ey' ,0 cos(ωt − k · x − ∆y' ) (6.17)

where x is the position vector, k is the wave vector such that c = ω/|k| (with c the light
speed). The full electric field vector is then

E(x, t) = E x' (x, t)e x' + Ey' (x, t)ey' (6.18)

The magnetic field H is then

H(x, t) = n × E(x, t) (6.19)

The average Poynting vector of this plane wave is equal to the flux vector F

F = )E(x, t) × H(x, t)* (6.20)

The intensity I(n' ) is

I(n' ) = |F|δ(n' − n) (6.21)
' '
meaning it is infinity in the direction n = n and zero for n ! n. This is because a
perfectly plane wave corresponds to radiation from an infinitely small solid angle Ω,
yet with finite total power, meaning that the power per solid angle diverges.
A perfectly coherent plane wave is fully described by its direction n, its frequency ν,
its amplitudes E x' ,0 , Ey' ,0 and the phase difference ∆ ≡ ∆y' − ∆ x' .
In many applications radiation is not perfectly coherent. In those cases just n, ν, E x' ,0 ,
Ey' ,0 and ∆ are not enough to describe the radiation fully. A complete description can,
however, be given by the Stokes parameters, which, for coherent radiation read:

I = E 2x' ,0 + Ey2' ,0 (6.22)

Q = E 2x' ,0
− Ey2' ,0 (6.23)
U = 2E x' ,0 Ey' ,0 cos ∆ (6.24)
V = 2E x' ,0 Ey' ,0 sin ∆ (6.25)

but they are also applicable to non-coherent radiation. These Stokes parameters are
often conveniently written in terms of a column vector:

 I 
 
 Q
  (6.26)
U 
V

Note that the Stokes parameters can be used to describe the flux, as is done here, but
can also be used for intensity. This can lead to some confusion, in particular when we
write I for the 1st Stokes parameter of the flux rather than the intensity. We will stick
to this notation, however, because it is convention. The meaning will be made clear in
the context.
The meaning of the Stokes parameters is as follows. I is the total flux (or intensity,
depending on the context). The Q, U and V have the same dimension as I but only
describe the polarization state. If Q = U = V = 0 the light is unpolarized, meaning
it is not coherent. Perfectly coherent light has Q2 + U 2 + V 2 = I 2 . In general we
have 0 ≤ Q2 + U 2 + V 2 ≤ I 2 . For V = I (∆ = π/2, i.e. the Ey' lags π/2 behind
E x' ) we have right-handed circularly polarized light, meaning that the tip of the E
field at a fixed point in space, when looking into the light (the propagation of light is
toward the reader) rotates counter-clockwise (when the x-coordinate points right, and
the y' -coordinate points up). The 3-D helix of his field will be left-handed (when the

96
z-coordinate points into the propagation direction of the light, i.e. toward the reader,
i.e. a right-handed coordinate system). For Q = I we have linearly polarized light in
which the E-field lies in the x-direction. For U = I we have linearly polarized light
in which E lies along the x' = y' line (when looking into the light). These definitions
are consistent with the IAU 1974 definitions (Hamaker & Bregman 1996, A&AS 117,
pp.161).
Important note: In order to be able to make sense of a Stokes vector in problems of
radiative transfer, we must always specify a reference coordinate system. Since light
may travel in any direction we cannot define this reference coordinate system apriori.
So far we are used to specifying the direction of propagation n if we talk about the
intensity. Let us call this the polarization reference vector. Now, if we talk about the
polarized intensity in terms of a Stokes vector, we must, in addition to specifying n,
also specify some additional unit vector S, which we define to point along the positive
y' direction, as defined above. Clearly it must hold that S · n = 0. The S vector gives
the direction of the E field for radiation polarized with Q = −I and U = V = 0.
If, for whatever reason, we need to do a rotation coordinate transformation from the
(x' , y' )-basis to an (x'' , y'' )-basis according to
! '' " ! "! "
x cos ψ sin ψ x'
= (6.27)
y'' − sin ψ cos ψ y'
(e.g. when rotating the direction of the polariation reference vector S), we must change
the Stokes vectors with the following transformation:
 ''  
 I  1 0 0 0  I ' 
 
 Q''  0 cos(2ψ) sin(2ψ) 0  Q' 
 ''  =  (6.28)
U  0 − sin(2ψ) cos(2ψ) 0 U ' 
 
V '' 0 0 0 1 V'
Matrices that manipulate Stokes vectors are called Müller matrices. This particular ex-
ample is the Müller matrix of rotation in the (x' , y' )-plane. But we will later encounter
scattering Müller matrices.

6.5 The scattering matrix

Using the Stokes vector description of radiation, let us look again at the scattering
problem. Let us describe the incoming plane wave as a flux Stokes vector F =
(F I , F Q , FU , FV ), and the outgoing semi-spherical wave as W = (WI , WQ , WU , WV ),
where W is a function of distance r from the particle, and roughly goes as 1/r2 . Let us
focus on the scattering into some direction θ and φ. We can now describe the relation
between W and F with a scattering matrix:

 WI  Z11 Z12 Z13 Z14   F I 

    
W  1 Z
 Q  =  21 Z22 Z23 Z24   F Q 
 
(6.29)
WU  r2 Z31 Z32 Z33 Z34   FU 
WV Z41 Z42 Z43 Z44 FV
which is written under the assumption r $ λ/2π. The elements Zi j are the elements of
the scattering matrix. Note that the scattering matrix is dependent on angle: S i j (θ, φ).
By definition the total scattering cross section is the angle-integrated cross section:
. π . 2π
σscat = sin θ dθ dφ Z11 (θ, φ) (6.30)
0 0

6.5.1 Randomly oriented and/or spherical particles

For spherical particles and/or for randomly oriented non-spherical particles the total
angle-integrated scattering cross section is independent of the polarization state of

97
the incoming photon. This means that the scattering angular dependence becomes
less complex. We can choose the polarization reference vector S such that φ = 0,
i.e. that the scattering occurs in the (x' , z' )-plane, i.e. the photon changes angle by
rotation around the S-vector, such that it moves, afterward, in positive x' -direction
(if we look into the light, the photon gets scattered to the right). Then the scattering
matrix becomes simpler:
Z11 Z12 0 0 
 
Z12 Z22 0 0 

(6.31)
 0 0 Z33 Z34 
 
0 0 −Z34 Z44
If we use Mie theory (i.e. for spherical particles) it reduces to an even simpler form:

Z11 Z12 0 0 
 
Z Z11 0 0 

 12 (6.32)
 0 0 Z33 Z34 

0 0 −Z34 Z33

In conclusion: Mie theory gives four independent scattering matrix elements for each
wavelength. They are:
& '
S 11 = 12 |S 2 |2 + |S 1 |2 (6.33)
& '
S 12 = 21 |S 2 |2 − |S 1 |2 (6.34)
1/ ∗ ∗0
S 33 = 2 S 2 S 1 + S 2 S 1 (6.35)
1 / ∗ ∗0
S 34 = i 2 S 2 S 1 − S 2 S 1 (6.36)

where
∞
% 2n + 1
S1 = (an πn + bn τn ) (6.37)
n=1
n(n + 1)
∞
% 2n + 1
S2 = (an τn + bn πn ) (6.38)
n=1
n(n + 1)

where
P1n (θ) dP1n (θ)
πn (θ) = and τn (θ) = (6.39)
sin θ dθ
with P1n (θ) the associated Legendre functions.
For unpolarized light the scattering phase function can now be calculated according
to:
1
Φ(cos θ) = S 11 (cos θ) (6.40)
σscat

6.6 Some examples of real phase functions

In the margin figures we show some scattering phase functions computed with Mie
theory. We show them for a water droplet and for a spherical olivine particle. The
wavelenght is always λ = 0.55 µm and we show the phase functions for particle radii
of a = 1 µm and a = 10 µm.

98