Born approximations and the blue sky

April 10, 2016

1 Perturbation theory of scattering

1.1 General formalism
Scattering can occur in a medium with spatially varying or time varying properties. If these variations are
small, they may be treated perturbatively.
The analysis starts by assuming that
D 6= 0 E
1
H 6= B
µ0
so we treat all four fields as independent. Let 0 , µ0 be the unperturbed values of the dielectric constant and
permeability (and not the vacuum values). Then we can rewrite Maxwell’s equations in terms of the small
differences
D − 0 E
B − µ0 H
Starting from
∇·D = 0
∇·B = 0
∂B
∇×E+ = 0
∂t
∂D
∇×H− = 0
∂t
we have
∂B
0 = ∇×E+
∂t
1 1 ∂B
= ∇ × D − ∇ × (D − 0 E) +
0 0 ∂t
and taking the curl,
 
1 1 ∂B
0 = ∇× ∇ × D − ∇ × (D − 0 E) +
0 0 ∂t
1 1 ∂B
= ∇ × (∇ × D) − ∇ × [∇ × (D − 0 E)] + ∇ ×
0 0 ∂t
1 1 2 1 ∂H ∂
= ∇ (∇ · D) − ∇ D − ∇ × (∇ × (D − 0 E)) + µ0 ∇ × +∇× (B − µ0 H)
0 0 0 ∂t ∂t
1 ∂2D 1 ∂
= − ∇2 D + µ0 2 − ∇ × (∇ × (D − 0 E)) + ∇ × (B − µ0 H)
0 ∂t 0 ∂t

1
so that
∂2D ∂
∇2 D − µ0 0 2
= −∇ × (∇ × (D − 0 E)) + 0 [∇ × (B − µ0 H)]
∂t ∂t
where the terms on the right are small. With harmonic time dependence, and setting µ0 0 ω 2 = k 2 , this
becomes
∇2 + k 2 D = −∇ × (∇ × (D − 0 E)) − iω0 [∇ × (B − µ0 H)]

1.2 Born approximation

We can now perform a systematic perturbation theory, allowing small corrections to the permittivity and
permeability,

 = (0) + (1) (x) + (2) (x) + . . .

µ = µ(0) + µ(1) (x) + µ(2) (x) + . . .

where (0) = 0 , µ(0) = µ0 , and setting the fields equal to

D = D(0) + D(1) + . . .
E = E(0) + E(1) + . . .
D = E
  
= (0) + (1) + (2) + . . . E(0) + E(1) + . . .
h i h i
= (0) E(0) + (0) E(1) + (1) E(0) + (1) E(1) + (2) E(0) + (0) E(2) + . . .
B = B(0) + B(1) + . . .
H = H(0) + H(1) + H(2) + . . .
B = µH
  
= µ(0) + µ(1) + µ(2) + . . . H(0) + H(1) + H(2) + . . .
h i h i
= µ(0) H(0) + µ(1) H(0) + µ(0) H(1) + µ(2) H(0) + µ(1) H(1) + µ(0) H(2) + . . .

This means that the differences on the right side of the wave equation are of higher order than the field on
the left.
h i
D − 0 E = D − 0 E(0) + E(1) + E(2) + . . .
h i h i
= (1) E(0) + (1) E(1) + (2) E(0) + . . .
h i
B − µ0 H = B − µ0 H(0) + H(1) + H(2) + . . .
h i h i
= µ(0) H(0) + µ(1) H(0) + µ(0) H(1) + µ(2) H(0) + µ(1) H(1) + µ(0) H(2) + . . .
h i
−µ0 H(0) + H(1) + H(2) + . . .
h i h i
= µ(1) H(0) + µ(2) H(0) + µ(1) H(1) + . . .

Notice that the small source terms only depend on lower order fields.
As a first approximation, we put in the zeroth order approximations for D and B,
  h i
(0) (0)
∇2 + k 2 D(0) = −∇ × ∇ × (D − 0 E)


− iω0 ∇ × (B − µ0 H)
= 0

2
and we see that D(0) is the plane wave solution. At the next order, we have
   h  i
∇2 + k 2 D(1) = −∇ × ∇ × (1) E(0) − iω0 ∇ × µ(1) H(0)

The solution to this equation is the first Born approximation, and as we show below it gives the first
approximation to the scattered wave. Notice again that the wave equation for D(1) has sources that only
depend on the lowest order solutions for the fields, E(0) and H(0) .
We may continue to arbitrary order, to find corrections. Thus, at second order, we have
 h i  h i
∇2 + k 2 D(2) = −∇ × ∇ × (1) E(1) + (2) E(0) − iω0 ∇ × µ(2) H(0) + µ(1) H(1)

As long as we have a sensible perturbative series for the permittivity and permeability, we may continue this
process to arbitrary order.

1.3 First Born approximation for light scattering

For light scattering in gasses, higher order Born approximations are not generally appropriate because we
do not really know  or µ. Instead, we can make use of the first Born approximation by supposing

 = 0 + δ (x)
µ = µ0 + δµ (x)
where 0 and µ0 are constant (but not necessarily the vacuum values) and that there are small, position
and/or time dependent fluctuations in addition. First Born approximation immediately gives
∇2 + k 2 D(0) = 0

for the incoming plane wave, and the first order correction
∇2 + k 2 D(1) = −µ0 0 J


   h  i
µ0 0 J = ∇ × ∇ × δE(0) + iω0 ∇ × δµH(0)

This has solution

ˆ 0
(1) 1 eik|x−x | J
3
D = d x
4π |x − x0 |
We now assume that the fluctuations in  and µ are localized in space and look at the scattered wave D(1)
in the radiation zone,

D = D(0) + D(1)
ˆ
eikr 0
= D(0) + d3 x0 e−ikn·x J
4πr
The integral of J may be simplified by integration by parts by recalling the identity
∇ × (f v) = ∇f × v + f ∇ × (v)
Applying this to the curl terms in the electric part of J ,
ˆ
(1) eikr 0
  
D1 = d3 x0 e−ikn·x ∇0 × ∇0 × δE(0)
4πr
ˆ
eikr h 0
  i
= d3 x0 ∇0 × e−ikn·x ∇0 × δE(0)
4πr
ˆ
eikr h 0
  i
− d3 x0 ∇0 e−ikn·x × ∇0 × δE(0)
4πr

3
The first term on the right may be integrated, giving a result that depends on the fields at large distance,
where the perturbation, δ, vanishes. Then
ˆ
(1) eikr h 0
  i
D1 = − d3 x0 −ike−ikn·x n × ∇0 × δE(0)
4πr
ˆ
ikeikr h 0
 i
= n × d3 x0 e−ikn·x ∇0 × δE(0)
4πr
ˆ
ikeikr h 0
  i
= n × d3 x0 −∇0 e−ikn·x × δE(0)
4πr
ˆ
k 2 eikr
 h i
0
= − n × n × d3 x0 e−ikn·x δE(0)
4π r
where we have done a second integration by parts and again dropped the surface term. The electric dipole
0
term is the lowest order term when we expand e−ikn·x = 1 − ikn · x0 + . . ., so the electric dipole field is
ˆ
(1) k 2 eikr h   i
D1 = − d3 x0 n × n × δE(0)
4π r
Comparing with the general electric dipole expression,
k 2 eikr
Dsc = [(n × p) × n]
4π r
we see that we have an effective electric dipole source,
ˆ
p = d3 x0 δ (x0 ) E(0) (x0 )

where E(0) is the incident wave.

The second term in J works the same way,
ˆ
(1) iω0 eikr 0
h  i
D2 = d3 x0 e−ikn·x ∇0 × δµH(0)
4π r
ˆ
iω0 eikr 0
h  i
= − d3 x0 e−ikn·x −ikn × δµH(0)
4π r
ˆ
k 2 0 eikr 0
h  i
= − d3 x0 e−ikn·x n × δµH(0)
4πc r
and this expression has the same form as the magnetic dipole field,
k 2 0 eikr
 
1
Dsc = − n×m
4π r c
if we identify ˆ  
m= d3 x0 δµH(0)
(1) (1)
The total scattering amplitude is Dsc = D(1) = D1 + D2 , and the form is exactly the same as the
form of the electric field for electric and magnetic dipole radiation,
k 2 0 eikr
 
1
Dsc = (n × p) × n − n × m
4π r c
If we define
eikr
Dsc = Asc
r

4
then the differential cross-section follows immediately as
2
dσ |ε∗ · Asc |
= 
dΩ D(0) 2


If the initial wave has the form

D(0) = ε0 D0 eikn0 ·x
r
µ0
B(0) = n0 × D(0)
0

then the general multipole expression for the scattered wave becomes
ˆ i k 2  ˆ
k2
 h h  i
0 0 0
Asc = − n × n × d3 x0 e−ikn·x δE(0) − d3 x0 e−ikn·x n × δµH(0)
4π 4πc
2 ˆ  
k 0 δ δµ
= D0 d3 x0 eikq·x (n × ε0 ) × n − [n × (n0 × ε0 )]
4π 0 µ0
dσ
so that dΩ is the square of
ˆ
ε∗ · Asc k2
 
3 0 ikq·x0 δ ∗ δµ ∗
= d xe ε · [(n × ε0 ) × n] − ε · [n × (n0 × ε0 )]
D0 4π 0 µ0

Rewriting

ε∗ · [(n × ε0 ) × n] = ε∗ · ε0
ε∗ · [n × (n0 × ε0 )] = (n0 × ε0 ) · (ε∗ × n)
= − (n0 × ε0 ) · (n × ε∗ )

this becomes ˆ
ε∗ · Asc k2
 
0 δ ∗ δµ
= d3 x0 eiq·x ε · ε0 + (n0 × ε0 ) · (n × ε∗ )
D0 4π 0 µ0
where q = k (n0 − n). A simple approximation is to then take δ constant inside a sphere of radius a, and
set δµ = 0. Then the integral becomes
ˆ
ε∗ · Asc k2
 
0 δ ∗
= d3 x0 eiq·x ε · ε0
D0 4π 0
ˆ
k 2 δ ∗ 0
= (ε · ε0 ) d3 x0 eiq·x
4π 0
ˆ
k 2 δ ∗ 0
= (ε · ε0 ) r02 dr0 dϕ0 d (cos θ0 ) eiq·x
4π 0
Taking n0 in the z-direction,
ˆ ˆ
0 0 0
r02 dr0 dϕ0 d (cos θ0 ) eiq·x = r02 dr0 dϕ0 d (cos θ0 ) eiqr cos θ
ˆ
0 0
= 2π r02 dr0 d (cos θ0 ) eiqr cos θ

ˆa ˆ1
0
02 0
= 2π r dr dxeiqr x
0 −1

5
 ˆa " 0
#+1
02 0 eiqr x
= 2π r dr
iqr0
0 −1
ˆa " 0 0
#
02 0 eiqr − e−iqr
= 2π r dr
iqr0
0
ˆa
sin qr0
 
= 4π r02 dr0
qr0
0
ˆa
4π
= dr0 r0 sin qr0
q
0
ˆa
 
4π  d
= − dr0 cos qr0 
q dq
0
a 
d sin qr0
 
4π
= −
q dq q
  0
4π d sin qa
= −
q dq q
4π
= (sin qa − qa cos qa)
q3
In the long wavelength limit, k, q → 0, and this becomes
1 3 3

!
− qa 1 − 21 q 2 a2 + O q 4


qa −
 
sin qa − qa cos qa 3! q a
4π lim = 4π lim
q→0 q3 q→0 q3
1 3 3

!
− 3! q a + 12 q 3 a3 + O q 4
= 4π lim
q→0 q3
 
1 3 1 3 3


= 4π lim − a + a + O q
q→0 3! 2
4 3
= πa
3
The differential cross section is therefore
 2 2
dσ  k δ ∗ 4 3 
= 
 (ε · ε0 ) πa 
dΩ 4π 0 3
 2
 δ 
= k 4 a6   |(ε∗ · ε0 )|2
30 

This agrees, to first order in δ, with our result for scattering from a dielectric sphere,
 2
dσ  − 0
= k 4 a6 sin2 ϕ
dΩ  + 20

when we set  − 0 = δε and  + 20 = 30 + δ, since

 − 0 δ
=
 + 20 30 + δ

6
 δ
=  
δ
30 1 + 30
 
δ δ
= 1− + ...
30 30
δ
+ O (δ) 2


=
30

2 The blue sky

Jackson gives two treatments of scattering in the atmosphere. The first is to approximate the atmosphere
as a dilute gas with randomly distributed molecules. Then taking the molecules to have dipole moments

pi = 0 γmol E

where γmol is the molecular polarizability, the effective dielectric constant is

X
δ = 0 γmol δ 3 (x − xi )
i

Then
ˆ
ε∗ · Asc k2
 
0 δ ∗
= d3 x0 eiq·x ε · ε0
D0 4π 0
2 ˆ
k ∗ X 0
= ε · ε0 γmol d3 x0 eiq·x δ 3 (x − xi )
4π i
k2 ∗ X
= ε · ε0 γmol eiq·xi
4π i

so that
 ε · Asc 2
 ∗ 
dσ
= 
 
dΩ D0 
k4 2 ∗
X
2
2
iq·x0 
= |γ mol | |ε · ε 0 | e i
16π 2
 

When we sum over all particles, the structure function

0 2
X 
F = eiq·x

i 

will be the total number of particles because of the randomness of the gas. For a single particle, we may
therefore drop the phase – the average differential cross-section per particle is just

dσ k4 2 2
= |γmol | |ε∗ · ε0 |
dΩ 16π 2
Now, for a dilute gas,
r ≈ 1 + N γmol
where r is the relative dielectric constant and N is the number density of molecules. Therefore,
2
2 |r − 1|
|γmol | =
N2

7
Rewriting this in terms of the index of refraction,
r
µ 
n =
µ0 0
√
= r
√
n−1 = r − 1
p
n−1 = 1 + (r − 1) − 1
1
n−1 ≈ 1 + (r − 1) − 1
2
2 (n − 1) = r − 1
so that the differential cross section becomes
dσ k4 2 2
|γmol | |ε∗ · ε0 |
=
dΩ 16π 2
2
k 4 |r − 1| 2
= |ε∗ · ε0 |
16π 2 N 2
2
k 4 |n − 1| 2
= |ε∗ · ε0 |
4π 2 N 2
For unpolarized initial and final states, we average over initial and sum over final polarizations, so the
polarization factor becomes
  !
1 X X 2 1 X X 2
  |ε∗ · ε0 | = |ε∗ · ε0 |
2 ε=ε ,ε
2 ε=ε ,ε
ε0 =i,j 1 2 ε0 =i,j 1 2

which we can find by dropping the magnetic dipole part of our earlier calculation for the sphere. Averaging
over

1
εk (n0 × n) × n
=
sin θ
1
ε⊥ = n0 × n
sin θ
we first find the parallel and perpendicular cases separately:
!
2
dσk k 4 |n − 1| 1 X  ∗ 2
= · ε̂

k 0
dΩ 4π 2 N 2 2
ε 
ε̂0k ,ε̂0⊥
2  2
1 X  ∗ X  1 
εk · ε̂0  = ((n0 × n) × n) · ε̂0 
 
2  sin θ
ε̂0k ,ε̂0⊥ ε̂0k ,ε̂0⊥

1 1 X 2
= 2 |((n0 × n) × n) · ε̂0 |
2 sin θ
ε̂0k ,ε̂0⊥

1 1  2
= 2 ((n0 × n) × n) · ε̂0k 
2 sin θ
1
= cos2 θ
2
and
!
2
dσ⊥ k 4 |n − 1| 1 X 2
= |ε∗⊥ · ε̂0 |
dΩ 4π 2 N 2 2
ε̂0k ,ε̂0⊥

8
  2
1 X 2 1 X  1
|ε∗⊥ · ε̂0 |

=  sin θ (n0 × n) · ε̂0 

2 2
ε̂0k ,ε̂0⊥ ε̂0k ,ε̂0⊥
 2
1  1 
= (n0 × n) · ε̂0⊥ 
2  sin θ
1
=
2
If the final polarization is unmeasured, we sum these, giving
dσ dσk dσ⊥
= +
dΩ dΩ dΩ !
4 2
k |n − 1| 1
1 + cos2 θ


=
4π 2 N 2 2

The total cross section is now given by integrating over angles

2 ˆ
k 4 |n − 1|
1 + cos2 θ dΩ


σ = 2 2
8π N
2 ˆ1
k 4 |n − 1|
1 + x2 dx


= 2 2
2π
8π N
−1
4 2 1
k |n − 1| 1 3
= 2π x + x
8π 2 N 2 3 −1
4 2  
k |n − 1| 8
= 2π
8π 2 N 2 3
2
2k 4 |n − 1|
=
3π N 2

 molecule. Remembering that σ is the effective

This is the total scattered flux per incident flux from a single
area of scatterers, the total fraction of light scattered,  dI
I , in a travel distance dx is therefore N σdx.

Suppose the incident beam has intensity I0 . Then
dI
= −N σdx
I
I = I0 e−N σx

The absorption coefficient, α = N σ is therefore

2
2k 4 |n − 1|
α=
3π N
This result describes Rayleigh scattering.
And improved version, due to Einstein, starts with the density fluctuations ∆Nj in cells of volume v, as
given by the Clausius-Mossotti relation (eq.4.70),
(r − 1) (r + 2)
δj = ∆Nj
3N v
Carrying out the cross-section calculation again, and summing over all cells, gives the attenuation coefficient
as 2
k 4  (r − 1) (r + 2)  ∆NV2

α=
6πN  3  NV

9
where ∆NV2 is the mean-square number fluctuation per unit volume,
2
∆NV2 = Ni2 − hNi i



The final ratio may be expressed in terms of the isothermal compressibility, βT = − V1 dV

dP T ,

∆NV2
= N kT βT
NV
so that the absorption coefficient becomes
 2
kT βT 4  (r − 1) (r + 2) 
α= k 
6π 3 

This is called the Einstein-Smoluchowski formula.

With N kT βT = 1 and the approximation

(r + 2)
(r − 1) ≈ 2 (n − 1) · 1
3
as above, we recover the previous result.

10