22.51 Problem Set 7 (due Fri, Nov.

9)
1

Form Factor of Ellipsoid

Question: We showed that the form factor of a dielectric sphere is,

k02 (m2 1) 4a3 3j1 (|Q|a)

.
4
3
|Q|a

F (Q) =

where Q k0 k00 is the scattering vector. Based on this, is it straightforward to get the
form factor of an ellipsoid with principal axis lengths a,b,c?
Answer: Let us first rotate to a frame where x is along the a-axis of the ellipsoid, y is along
the b-axis, and z is along the c-axis, e.g., define,

Q0 Q0x , Q0y , Q0z ,

Q0x Q

a
b
c
, Q0y Q
, Q0z Q ,
|a|
|b|
|c|

(1)

from any initial frame. Then, we make the observation that,

Z
ellipsoid

=
=
=

ellipsoid

exp (iQ0 x) d3 x

exp i(Q0x x + Q0y y + Q0z z) dxdydz

a
a
bc Z
0
0 b a
0 c a
exp
i(Q
x
+
Q

y
+
Q

z)
dx
dy
dz
x
y
z
a2 ellipsoid
a b
a c
b
c
Z

bc
exp i(Q00x x + Q00y y + Q00z z) dxdydz,
a2 sphere

(2)

where the sphere has radius a, and,

00

Q0x ,

Q0y

b
c
, Q0z
.
a
a

(3)

However, the last expression of (2) is identical to the form factor of a sphere, except by a
constant Jacobian factor bc/a2 . Therefore, the form factor for an ellipsoid is,
F (Q) =

k02 (m2 1) 4abc 3j1 (|Q00 |a)

,
4
3
|Q00 |a

where Q00 is a simple function of Q via (1) and (3).

The point of this exercise is not to re-work the math but to illustrate the power of an invariant
transformation on Q x. Suppose someone has worked out the integral of,
Z
sphere

tanh3 (Q x) sin(Q x)(Q x)5 d3 x,

it is as easy to get the result for an ellipsoid as above. All depends on the symmetry properties
of Q x.

Static Polarizability of Dielectric Sphere

Question: Prove Equation (4.180) and explain its significance in connecting radiation with
scattering phenomena.
Answer: The governing equations for E(x) in a linear-response dielectric medium, when
there is no B field involved, are shown in class to be,
(E) = 0,

E = 0,

where the dielectric constant may depend on x. In our problem,

(x) =

, |x| < a,
1
, |x| a,
0

(4)

Since E = 0, using existence theorem we know there exists scalar field (x) such that,
E(x) = (x),

(5)

and (x) should be continuous across the interface (even though its first-order derivative may
not). When we plug (4) and (5) into (E) = 0, the condition turns out to be 2 = 0
both inside and outside of the sphere, and 1 n inside should match 0 n outside across
the interface. Furthermore, if the external field is E0 ez , then the leading x asymptote
of (x) should be E0 r cos .
Because there is no azimuthal variation in the setup, (x) should take the form,
(x) = g(r)f (cos ),

both inside and outside of the sphere. Since,

r2
= 2
r r
r

+ 2
(1 cos2 )
r cos
cos

1
2
,
+ 2 2
r sin 2

and let us furthermore assume the angular dependence is simply,

f (cos ) = cos ,
then the equation for g(r) would be,
d 2 0
r g (r) 2g(r) = 0,
dr
whose possible solutions are,
g(r) = r, r2 .
For (x) inside the sphere, the r2 solution is impossible, therefore,
(x) = Ar cos ,

|x| < a.

For (x) outside of the sphere, the solution may look like,
(x) = E0 r cos + Br2 cos ,

|x| a.

To satisfy the continuity condition, we have,

A = E0 + Ba3 .
To satisfy the D(x) divergence-free condition, we must have,
1 A = 0 (E0 2Ba3 ).
So there must be,
m2 A = E0 2(A + E0 ),
therefore,
A =

3E0
,
2 + m2

(x) =

3E0
r cos ,
2 + m2

|x| < a.

therefore the electric field inside the sphere is,

E =

3E0
,
2 + m2

and the polarization density is,

P =

DE
1 1
1 1 3E0
=
E =
,
4
4
4 2 + m2

so the total induced dipole of the sphere is,

p =

4a3
1 1
P = a3
E0 .
3
2 + m2

When the background is vacuum, 0 = 1, 1 = m2 , equation (4.180) is correct. When the

background is not vacuum, one must subtract off the background polarization density from
P, because there still would be no scattering in a homogeneous dielectric medium 1 = 0 > 1,
even when there is finite and time-varying polarization. In other words, the scattering is
really caused by the extra polarization density inside the sphere. When one works out
the details, equation (4.180) would still be correct. But instead, we again have a simpler
approach using a transformation, where we map the background to vacuum!
The idea is the following. Suppose we want to solve,
E =

1 B
B
1 (x)E
,
=
, B = 0,
c t
(x)
c t

(x)E = 0,

(6)

in the case of (x) 1, and (x) may vary with space but with baseline value 0 , then by
redefining space and magnetic field, we can map to a world spatially larger than ours, with
background dielectric constant 1! Specifically, Let us define x0 such that,

0 0 ,

0 B0 ,

E0 E,

then we can convince ourselves that,

1 B0
E =
,
c t
0

B0
1 (x0 )/0 E0

=
, 0 B0 = 0,
0
(x )
c
t
0

0 (x)/0 E0 = 0.

In other words, {E0 (x0 ), B0 (x0 )} satisfy Maxwells equations as if the background is vacuum.
Therefore, in this parallel world, the solution is equation (4.177) but with Es E0s , k

k 0 / 0 , r r0 = 0 r, a a0 = 0 a. We find that because the magnitude of E is not

4

influenced before and after the transformation, and because k, r, a are influenced but their
combined effect cancel, equation (4.177) is still correct.

Differential and Total Scattering Cross-Sections

Question:
a. Fill in the missing details about the far-field magnetic field B(x, t) and the Poynting
vector S(x, t), that enables one to derive Equation (4.184).
Answer: At far-field, the scattered wave locally is like a plane-wave with wave-vector k0 er ,
therefore from equation (4.150),
Bs (x, t) =

c
c
er Es (x, t) = n0 er Es (x, t),
(ik0 er ) Es (x, t) =
i
v0

and so,

c
cn0
(Es Bs ) =
|Es (x, t)|2 er .
4
4

Ss (x, t) =
The incoming energy flux is,

S(x, t) =

cn0
|E0 (x, t)|2 ek0 .
4

Therefore the differential scattering cross-section is indeed,

d
=
d

|Ss (x, t)|r2

= h|f ()|2 i,
r |S(x, t)|
lim

as in equation (4.183).
A paradox can be posed as the following. Why should not
Bs (x, t) = g()

ei(krt)
B0 ,
r

g() k02 (I er er )

B0
,
B0

where g() is clearly different from f (), since B appears to be symmetric with E in that
they both satisfy,
2 W n2 kv2 W = 0,
in a homogeneous medium, so we can use the same perturbation theory?
The answer is: it is true that B and E have symmetric positions in a homogeneous medium.
5

But when = (x), = (x), their positions are not necessarily symmetric. A careful
examination of equation (4.153) derived from (6) shows that it is only true for E when
(x) 1, and not true for B. Therefore, B is puppet to E when (x) 1 everywhere, and
vice versa.
b. Equation (4.185) is clearly wrong. Explain why and derive the correct result. Does it
change the final conclusion, though?
Answer: Let,
E0 = E0 ez = E0 (er cos e sin ),
therefore,
(I er er )

E0
= e sin ,
E0

and sin = 1 only if = /2, in the VV plane. Therefore, the (4.185) estimate should be
scaled by a factor,
1 Z
1 Z
2
d sin2 =
2d cos (1 cos2 ) =
.
4
4
3
But this would not change the conclusion that a small dielectric particle is much less reflective
than its size suggests when kd 1.

Why is the Sky Blue?

Question: Let f ()d be the probability that a measurement of suns radiation on moon
yields a photon sample within frequency range (, + d). Show why our sky is blue - or,
to be more precise, bluer than the sun.
Answer: On moon, when one looks directly at the sun, the probability distribution of
photon frequency is,
P1 () = f ().
On earth, when we look at the sky (away from the suns direction), the photons that come into
our eyes must have undergone scattering by the air molecules in the atmosphere. Assuming
just single scattering events occur, the probability distribution is,
P2 () =

R
0

f () 4
,
f () 4 d

since the scattering cross-section is 4 in any direction. P2 () distribution is clearly going

to have a larger mean than the P1 () distribution. Therefore the sky looks blue.
When we do look directly at the sun on earth, the probability distribution is,
P3 () =

P1 () rP2 ()
,
1r

where r is the number ratio of scattered photons when the sunlight pierce the atmosphere.
P3 () is going to look even redder than P1 (). The larger r is, the more it is so. Therefore
the sun looks especially red at sunrise and sunset, since when the sun is near the horizon it
is separated from our eyes by the thickest layer of air!
Could this redshift spell trouble for astronomers when they, say, measure the real redshift of
stars, since anything that depends on air thickness is going to terribly tricky?
When we think about it, the answer is no. The spectral lines themselves do not shift due
to the scattering loss, but only their relative weights change, causing the overall redshift.
Therefore this scattering redshift is of a nature entirely different from the Doppler redshift.