(AST 461) LECTURE 1: Basics of Radiation Transfer

Almost all we know about the astronomical universe comes from from
radiation emanating from faint sources. We need to know how to interpret
this radiation.
Electromagnetic spectrum, wavelength and frequency:

νλ = c, (1)
ν is frequency, λ is wavelength. The speed of light depends on the index of
refraction.
Justification for Macroscopic Treatment of Radiation
Scale of system >> λ of the radiation, the radiation can been considered
to travel in “straight lines” =rays. dA >> λ2
Amount of energy passing through source dA is in time dt is F dAdt where
F is the flux. Units are (erg/cm.s2 ) in CGS.
Flux from an isotropic source
Isotropic means energy emitted in all directions. Consider 2 different
spherical surfaces S1 and S2 . (fig 1.)

F1 dA1 dt1 = F2 dA2 dt2 (2)

but dt1 = dt2 so that

F1 /F2 = dA2 /dA1 = r22 /r12 . (3)

This is the inverse square law.

Intensity
Flux measures energy from all rays in given area. A more detailed ap-
proach is to consider energy carried from individual rays, or sets of rays
differing infinitesimally from the initial ray. (fig 2.)

1
 dE = (k̂ · n̂)Iν dAdtdΩdν, (4)
where Iν is the Specific Intensity and represents Energy /(time × area × sold
angle × frequency).
Now suppose we have an isotropic radiation field. The differential flux at
a given frequency is

dFν = (k̂ · n̂)Iν dΩ = Iν cosθdΩ. (5)

But if Iν is isotropic, then Fν = 0.

Constancy of Intensity in Free Space:
Specific intensity is constant along a ray in free space.
Here’s why: Consider all rays passing through both dA1 and dA2 and use
conservation of energy: (fig 3)

dE1 = Iν,1 dA1 dt1 dΩ1 dν1 = dE2 = Iν,2 dA2 dt2 dΩ2 dν2 (6)
Note that dt1 = dt2 , dν1 = dν2 . Note also that dΩ1 is the solid angle
subtended by dA2 at dA1 and dΩ2 is the solid angle subtended by dA1 at
dA2 . Thus we have
dΩ1 = dA2 /r2 (7)
and
dΩ2 = dA1 /r2 . (8)
Thus
Iν,1 = Iν,2 . (9)
The intensity is constant along free space.

2
Application to a Telescope (fig 3.)

Assume large distance between object and lens r >> f , where f is the
focal length. Determine image intensity Ii given object intensity I0 . Infinites-
imal amount of surface area of object dA0 has intensity I0 . We have

I0 dA0 dΩT,0 = I0 AT dA0 /r2 . (10)

where dΩT,0 is the solid angle of the telescope as measured from the dA0 of
the object. All photons from dA0 must strike dA1 on focal plane.
Thus

I0 dA0 dΩT,0 = I0 dA0 AT /r2 = Ii dAi dΩT,i = Ii dAi AT /f 2 , (11)

but solid angle sub-tending image from telescope equals solid angle sub-
tending object from telescope so

dA0 /r2 = dAi /f 2 , (12)

and as expected, Ii = Io .
Do telescopes measure flux or intensity?
If resolution of telescope is crude, and we cannot resolve object, then we
measure flux. If telescope can resolve object, then we measure intensity.
Why? Consider the case when the source is unresolved. Now imagine
pushing the source to farther distance. As the distance increases, the number
of photons falls as r2 . The flux is measured.
If instead the the source is resolved, then as the source is pushed far-
ther away, more area of the source would be included with the solid angle,
which compensates for the the increased distance and the collected number
of photons remain the same.
Flux from a uniformly bright sphere: read in text.
Relationship of Intensity to Stat Mech Quantities:

3
 Consider photon distribution function f such that fα (x, p, t)d3 xd3 p is the
number of photons in x, p space, with spin index α.

p = hk = (hν/c)k̂ (13)
so
2
hνfα (x, p, t)d3 xd3 p.
X
dE = (14)
α=1

But photons traveling in direction k̂ for time dt, through an element of

area dA whose normal = n̂, occupy

d3 x = cdt(k̂ · n̂)dA (15)

and
d3 p = p2 dΩdp = (E 2 /c2 )dΩdE/c = (h3 ν 2 /c3 )dΩdν, (16)
where we have used E = pc and E = hν. Thus,
2
(h4 ν 3 /c3 )fα (x, p, t)dAdtdΩdν
X
dE = (k̂ · n̂) (17)
α=1

so
2
(h4 ν 3 /c3 )fα (x, p, t).
X
Iν = (18)
α=1

In stat mech, the occupation number for each photon is

Nα = h3 fα , (19)

which is dimensionless so
2
(hν 3 /c2 )Nα (x, p, t).
X
Iν = (20)
α=1

Later we will see for example that Iν corresponds to Bν , the Planck func-
tion for b-body radiation. The occupation number corresponds to Nα =
(ehνα /kT −µα − 1)−1 for Bose-Einstein stats.
Radiation Pressure
photon carries momentum pν = Eν /c. (fig. 4)

4
 Can compute pressure by considering incident photons reflecting off of a
wall: The angle of incidence equals the angle of reflection. The change in
the z component of momentum of photon between frequency ν and ν + dν
reflected in time dt from area dA is
dpν dν = [(pν )f − (pν )i ]dν = (1/c)(Eν cosθ − (−Eν cosθ))dν
  = (2/c)Eν cosθdν
= 2c Iν dνdtdAcos2 θdΩ,
(21)
where we used (4). Now the change in momentum per unit area per unit
time is a differential force/area= differential pressure. Integrating over solid
angle gives the total pressure:
Z 2π Z π/2
dpν Z
Prad,ν = = (2/c) Iν cos2 θdΩhem = (2/c) Iν cos2 θsinθdθdφ.
dtdA φ=0 θ=0
(22)
Now imagine removing the reflecting surface. Thus instead of the factor of 2
in momentum, we would have photons coming in from the other side. Thus
we can remove the factor of 2, and integrate over the full sphere.
Z Z 2π Z π
Prad,ν = dpν /dtdA = (1/c) Iν cos2 θdΩ = (1/c) Iν cos2 θsinθdθdφ.
φ=0 θ=0
(23)
Energy Density
Consider a cylinder of cross section dA. (fig 5)

dE = uν (Ω)dV dνdΩ = uν (Ω)cdtdνdAdΩ = Iν dAdΩdtdν (24)

so the energy density is
uν (Ω) = (Iν /c). (25)
Then Z Z
uν = uν (Ω)dΩ = (1/c) Iν dΩ = 4πI ν /c (26)

5
For isotropic radiation field
I ν = Iν (27)
Note also that the pressure for an isotropic radiation field from (23) and (26)
is
Z
2
Z
2 1Z
P = (1/c) Iν cos θdΩdν = (I ν /c) cos θdΩdν = uν dν. (28)
3

6
 LECTURE 2
Equation of Radiative Transfer
Condition that Iν is constant along rays means that
dIν /dt = 0 = ∂t Iν + ck · ∇Iν , (29)
where k · ∇ = dIν /ds is the ray-path derivative. This is equation is the
statement that there are no sources or sinks. But real systems may have
sources and sinks: emission (source) and absorption (sink).
If light travels past a system much faster than the time scale for which
the system evolves then ∂t Iν ' 0 and we have then
dIν /dt = 0 = ck · ∇Iν = dIν /ds = sources − sinks, (30)
where ds is the path. We now construct the source and sink terms in this
radiative transfer equation.
Emission coefficient (source)
Spontaneous emission coefficient jν is defined as the energy emitted per
unit time per unit solid angle, per unit volume per unit frequency
dE = jν dV dΩdtdν, (31)
where jν has units erg/cm3 · steradian · sec · hz. The emissivity is defined
as ν = 4πjν /ρ, where ρ is the mass density. Since dV = dAds, the “source”
contribution to the radiative transfer equation is given by
dIν = jν ds. (32)
Absorption coefficient (sink)
Consider the propagation of a beam through a cylinder (again) of cross
section dA and length ds, populated with a number density n of absorbers
(particles), with absorption cross section σν . (fig 6)

The loss of intensity, or the energy absorbed out of the beam is given by
dEν = (Ii,ν − If,ν )dAdΩdtdν = Iν (nσν dAds)dΩdtdν. (33)

7
Thus the “sink” contribution to the radiative transfer equation is

dIν = −nσν Iν ds. (34)

The absorption coefficient αν is defined such that

dIν = −αν Iν ds, (35)

so αν = nσν and αν is units of cm−1 . Sometimes it’s written αν = ρκν , where

ρ is the mass density and κν is the mass absorption or “opacity” coefficient.
Note that the emission and absorption coefficients have different units. Also,
note that absorbing particles are assumed to be randomly distributed and
much smaller than inter-particle spacing.
The term “absorption” is used somewhat loosely: Stimulated emission is
also included in the absorption coefficient because it is proportional to the
incident intensity.
We have thus derived the radiative transfer equation which we will soon
solve:

dIν /ds = −αν Iν + jν . (36)

Solutions of Equation of Radiative Transfer
A primary goal is to see what the values of αν and jν are for different
absorption and emission processes. Scattering complicates things since radi-
ation from the initial solid angle is scattered into a different solid angle for
which the scatter depends on the initial solid angle, so we need numerics.
Consider some simple cases here.
Emission Only:

dIν /ds = jν , (37)

where the solution is
Z s
Iν (s) = Iν (s0 ) + jν (s0 )ds0 . (38)
s0

Intensity increase is equal to the integrated emission coefficient along the

propagation path.
Absorbtion Only: Then jν = 0. In this case
Z s
Iν (s) = Iν (s0 )Exp[− α(s0 )ds0 ]. (39)
s0

8
The intensity decreases exponentially along the ray path.
Optical Depth
Define
dτν = αν ds, (40)
or Z s
τν (s) = α(s0 )ds0 . (41)
s0

The optical depth is measure of how transparent a medium is to radiation.

For τν ≥ 1 the medium is optically thick, and for τν < 1 the medium is
optically thin. We rewrite the transfer equation (36) as

dIν /dτν = −Iν + Sν , (42)

where we define the source function Sν ≡ jν /αν as the ratio of the emission to
the absorption coefficient. Optical depth is a physically meaningful variable
because it contains the information about transparency along the path.
Now solve (42) by integrating factor: Multiplying by eτν , we obtain:

d(Iν eτν )/dτν = eτν Sν , (43)

so the solution is Z τν
0
Iν eτν = Iν (0) + Sν (τν0 )eτν dτν0 (44)
0
or Z τν
−τν 0
Iν = Iν (0)e + Sν (τν0 )e−(τν −τν ) dτν0 . (45)
0
The right side is the sum of the initial intensity attenuated by absorption,
and the emission source attenuated by absorption. For a constant source
function we have

Iν = Iν (0)e−τν + Sν (1 − e−τν ) = Sν + e−τν (Iν (0) − Sν ). (46)

At large optical depth, Iν ' Sν . Why? because the source function is the
local input of radiation–little is contributed to the to Iν from matter far from
the location of interest when τν >> 1.
Note also that there is a relaxation process (evidenced by (42) where Iν
tends toward the source function: If Iν > Sν , then dIν /dτν < 0. If Iν < Sν ,
then dIν /dτν > 0.
Mean Free Path

9
 Interpreting the exponential in(39) (which equals e−τν ) as the probability
of traveling an optical depth τν , the mean optical depth traveled
Z ∞
hτν i = τν e−τν dτν = 1. (47)
0

The associated mean free path (MFP) is

lν = hτν i/αν = 1/nσν . (48)

MFP is important concept for particles as well as radiation.

Thermal Radiation and Blackbody Radiation
Thermal radiation is radiation emitted from matter in thermal equilib-
rium.
A blackbody absorbs all incident radiation.
Blackbody radiation is radiation which itself is in thermal equilibrium
and in thermal equilibrium with surrounding matter.
Consider a thermally insulated box kept at temp T . Iν = Iν (T ) = Bν , the
Planck function for such a black body box. Dependence only on temperature
can be seen by considering two connected boxes at same T . Energy cannot
flow without violating second law. Thus Iν is same in both boxes.
Iν is isotropic because otherwise there would be a net energy flow from
one part of the box to another which is not possible when entire system is at
uniform T .
For a blackbody,
2hν 3 /c2
Iν = Bν (T ) = , (49)
Exp[hν/kT ] − 1
the Planck function, and is only a function of the temperature and frequency.
Kirchoff ’s Law
Put thermally emitting material at temperature T with source function
Sν (T ) in the box.
If Sν > Bν then the intensity Iν will tend to increase such that Iν > Bν
and if Sν < Bν then the intensity will decrease to make Iν < Bν . But in
equilibrium, the new box is still a blackbody, and so Sν = Bν or

jν = αν Bν , (50)

which is Kirchoff’s law for a thermal emitter. In general, Kirchoffs law relates
emission to absorption.

10
 Consider again the radiative transfer equation (42) with Sν = Bν :

dIν /ds = −αν Iν + αν Bν (T ). (51)

For a black body (or for free space) the left side is zero, and thus Iν = Bν .
For thermal radiation in a not necessarily optically thick system, the left
side need not be zero for example if the source function temperature changes
rapidly with position and Iν cannot keep up. Thermal radiation is Sν = Bν
whilst blackbody radiation is Iν = Bν .
Blackbody Thermodynamics
The text derives some results: The Stefan-Boltzmann law for the energy
density:
u(T ) = aT 4 , (52)
where a is a constant. Recall from (26) that u = 4π
R
c
Bν (T )dν for Iν =
Bν . The integral of (49) over frequency is B(T ) = (ac/4π)T 4 , where a =
8π 5 k 4 /15c2 h3 .
Now the flux from an uniformly bright, isotropically emitting sphere of
brightness Bν is given by
Z Z 1
Fν = Bν cosθdΩ = 2πBν cosθd(cosθ) = πBν (1−cos2 θc ) = πBν sin2 θc = πBν (R2 /r2 ),
cosθc
(53)
where θc = arcsin(R/r), R is radius of source, and r is distance to the source.
At r = R, Fν = πBν , Then
Z
F = Fν dν = πB(T ), (54)

and the total flux for a blackbody

F = σT 4 , (55)

where σ = 0.25ac = 5.67 × 10−5 erg cm−2 o K −4 sec−1 .

Properties of the Planck Law
1. Rayleigh-Jeans (RJ) limit
Taking the hν << kT limit of the Planck law of (49) we expand the
exponential and obtain

BνRJ (T ) ∼ 2ν 2 kT /c2 , (56)

11
which is the Rayleigh-Jeans limit. This usually applies in the radio regime. It
is the classical regime. It cannot apply to all frequencies, because integrating
would imply total energy divergence.
2. Wien Limit
If we take the limit hν >> kT we have
BνRJ (T ) ∼ 2(hν 3 /c2 )e−hν/kT , (57)
which is the Wien regime.
3. Temperature derivative
A blackbody curve with the higher temperature is always completely
above a curve with a lower temperature. This follows from the temperature
derivative of Bν (T ).
4. Wien Displacement Law
Peak frequency can be found by maximizing Bν by setting ∂Bν /∂ν = 0,
which leads to
νmax /T = 2.82k/h = 5.9 × 1010 Hz/o K (58)
Peak frequency shifts linearly with the temperature. This is the Wien Dis-
placement Law. Same can be done for wavelength to find
λmax T = 0.290cmo K. (59)
Note that λmax νmax 6= c. Can see that the Rayleigh-Jeans law is valid for
ν << νmax .
Other temperatures commonly used
1. Brightness Temperature
This is defined by the temperature of the blackbody having the same
brightness (=intensity) at that frequency. We define it by
Iν = Bν (Tb ). (60)
In radio astronomy one often sees this formula in the Rayleigh-Jeans regime
(hν << kT ). One has there
Tb = (c2 /2ν 2 k)Iν (61)
and for thermal emission Sν = Bν ∼ 2ν 2 kT /c2 . In the RJ regime we can
then set Sν = Bν in (46) and divide that equation by 2ν 2 k/c2 to obtain
Tb = Tb (0)e−τν + T (1 − e−τν ). (62)

12
Tb ∼ T for τν >> 1.
Only if source is a blackbody is the brightness temperature the same at
all frequencies.
For optically thin thermal emission, brightness temperature gives lower
temp than that of the actual source; this is because a blackbody gives max-
imum intensity of thermal emitter at temperature T . (to see this use (60)
with Tb (0) = 0).
Why is this?: A body which is not optically thick can be thought of as
a system that is not fully “black” and absorbs less than a blackbody. If
this system were in thermal equilibrium, then its emission must equal its
absorption. But if it absorbs less than a blackbody it must also emit less
then a blackbody to remain in equilibrium.
2. Color Temperature
The color temperature is the temperature obtained by fitting a blackbody
shape to an observed, approximate blackbody flux.
3. Effective Temperature
This is obtained by setting the observed flux to that of blackbody form
from the Stefan-Boltzmann law. That is

Tef f = (F/σ)1/4 . (63)

13
 LECTURE 3
Einstein Coefficients
Kirchoff’s law relating emission to absorption for a thermal emitter must
involve microscopic physics. Consider system with two energy states with
statistical weights g1 and g2 respectively. Transition from 2 to 1 is by emission
and from 1-2 by absorption. State 1 has energy E and state 2 has energy
E + hν.
1. Define the Einstein A21 coefficient as the probability per unit time for
spontaneous emission.
2. The absorption of a photon ∝ density of photons or the mean intensity
at frequency ν0 . Energy difference between two levels is not sharp, but
broadened, and so we also need to consider the line profile function φ obeying
R∞
0 φ(ν)dν = 1. This is usually quite narrow. The transition probability per
unit time for absorption is

¯ B12 Z Z ∞
B12 J = Iν φ(ν)dνdΩ, (64)
4π 0

1 R
where B12 is the Einstein B coefficientR and Jν = 4π Iν dΩ, the intensity
averaged over solid angle. Here J¯ = 0∞ Jν0 φ(ν 0 )dν 0 ∼ Jν0 for narrow line
profiles.
Derivation of Planck’s law led Einstein to include stimulated emission.
It can be thought of as negative absorption and has a coefficient B21 such
that B21 J¯ is the transition probability per time for stimulated emission, also
proportional to the intensity.
Relation between Coefficients
Micro-physical relations that are independent of temperature will hold
regardless of whether processes are in thermodynamic equilibrium or not,
but we can use case of thermodynamic equilibrium to get at them when
possible.
In thermodynamic equilibrium the number of transitions per unit time
per volume into state 1 are equal to the transitions out. If n1 and n2 are
number densities of atoms in states 1 and 2 we have

n1 B12 J¯ = n2 A21 + n2 B21 J.

¯ (65)

This gives
A21 /B21
J¯ = , (66)
(n1 /n2 )(B12 /B21 ) − 1

14
where
n1 /n2 = (g1 /g2 )ehν/kT (67)
so
A21 /B21
J¯ = . (68)
(g1 /g2 )(B12 /B21 )ehν/kT − 1
But Jν ∼ Bν for an emitter whose matter is in thermodynamic equilibrium,
we have the Einstein detailed balance relations which relate atomic properties

g1 B12 = g2 B21 (69)

and
A21 = 2(hν 3 /c2 )B21 . (70)
These do not depend on temperature and must hold independently of thermo-
dynamic equilibrium. If we can determine any one of the Einstein coefficients
we get the other two.
Wien’s law follows if you don’t include stimulated emission. Recall that
was the hν >> kT regime, so n2 << n1 and the stimulated emission term is
small. (Multiply top and bottom of (68) times g2 /g1 to see this limit.)
Absorption and Emission Coefficients in Terms of Einstein Co-
efficients
Assume that line profile for spontaneous emission is same as that for
absorption, and is φ(ν) (which has units of ν1 ). Then matching units using
the definition of the emission coefficient jν

dE = jν dV dΩdνdt = (hν0 /4π)φ(ν)n2 A21 dV dΩdνdt, (71)

so the emission coefficient is

jν = (hν0 /4π)n2 A21 φ(ν). (72)

For the absorption coefficient the energy absorbed in dV and in dt is

1 Z
 
¯ dt = hν0 n1 B12 dV dt
dE = hν0 n1 B12 JdV Iν φ(ν)dνdΩ , (73)
4π
R∞ 0
where we take dV = dsdA (the cylinder). Then using 0I φ(ν 0 )dν 0
ν = I¯ν
and dI ν = αν I ν ds for the absorprtion, we have

αν,abs = (hν0 /4π)n1 B12 φ(ν). (74)

15
Stimulated emission can be thought of as inverse absorption so the full ab-
sorpion coefficient is then

αν = (hν/4π)(n1 B12 − n2 B21 )φ(ν). (75)

Transfer Equation in terms of the Einstein Coefficients

dIν /ds = −(hν/4π)(n1 B12 − n2 B21 )φ(ν)Iν + (hν/4π)n2 A21 φ(ν). (76)

The source function is jν /αν and so

Sν = n2 A21 /(n2 B12 − n2 B21 ). (77)

Using the Einstein relations (69) and (70) gives

αν = (hν/4π)n1 B12 (1 − g1 n2 /g2 n1 )φ(ν) (78)

and
Sν = (2hν 3 /c2 )(g2 n1 /g1 n2 − 1)−1 . (79)
Masers/Lasers:
For a system in thermal equilibrium we have

n1 /n2 = (g1 /g2 )ehν/kT (80)

which implies
n1 /g1 > n2 /g2 (81)
But it is possible to pump atoms into the upper state to reverse this relation.
Then (75) (or 78) is negative, implying negative absorption. The intensity
actually increases along the ray path exponentially with the optical depth.
Many interesting maser sources in astrophysics. NGC4258 is a galaxy for
which H2 0 masers (22.235, 321..GHz) in the central region trace a Keplerian
warped accretion disk, which has provided some of the best evidence for black
holes in nature.
Also, OH/IR sources: large cool giant star or supergiant star losing mass
rapidly in winds and detected only in IR or masers in OH (1.665 Ghz).
Can use OH masers in these sources to estimate their distances. The maser
emission line from the expanding wind shell has a red and blue component
because the near side part of the wind is moving at us, and the far side

16
is moving away (centered at the star). But variability in the stellar wind
(i.e. turning on and off rather than being steady) will show up with a delay
between the red and blue parts of the line due to the cross time across the
source. Since radio telescopes can resolve the angular diameter, we can get
the source diameter and thus the distance. More explicitly, d = r/θ = ct/θ,
where d is distance, r is the “wind span” t is the measured time delay and θ
is the measured angular diameter.
Scattering
So far we have ignored scattering. Scattering can be considered an emis-
sion process that depends on the amount of radiation incident upon the
medium doing the scattering. (Contrast: Thermal radiation does not de-
pend on incident radiation.)
Consider electron scattering (scattering of photons by electrons). Here
we assume isotropic, coherent (elastic) scattering. More elaborate coverage
of these assumptions later.
The scattering emission coefficient is found by equating the power “ab-
sorbed”(think of scattering as absorption and immediate re-emission) per
unit volume and frequency to the power emitted isotropically

jν = σν Jν . (82)

Where here σν is a scattering coefficient with units of 1/length like the ab-
sorption coefficient (and here σν should not to be confused with the cross
section or Stefan Boltzmann constant!) Note that Jν is angle averaged ver-
sion of Iν . So by analogy to jν = αν Sν discussed earlier, here we have Sν = Jν
and radiative transfer equation has scattering coefficient in both the emission
and ”absorption” terms:

dIν /ds = −σν (Iν − Jν ). (83)

The solution cannot be easily extracted from (45) since Iν is not known
initially and Jν depends on the angular integral of Iν .
To make progress, instead think of the scattering, absorption and emission
processes in probabilistic terms for a single photon. Consider a photon in an
infinite homogeneous scattering region. Displacement of photon after N free
paths, where the ith path is of length ri
N
X
R= r(a) . (84)
a=1

17
where the sum is over free displacements, not vector indices. Mean square
photon displacement is then given by
N
l∗2 = hR2 i = h
X
r(a) · r(a) i, (85)
a=1

as the cross terms vanish for isotropic scattering when there are a large num-
ber of scatterings. (Book does not mention this requirement.) Statistically,
each term of the sum on the right of (85) contributes the same average mean
free path l squared, so we have

l∗2 = N l2 , (86)

which indicates the mean square displacement of the photon For a finite
medium, we can determine the number of scatterings N . For large optical
depths, N is found by setting l∗ = L, the typical size of the medium. so
that L/l = τ , the optical depth to scattering and N ' τ 2 . This is a large N
result only, because in deriving it we assumed that each contribution to the
sum on the right of (85) contributes the same amount (and that cross terms
vanish). This is true only within an error of ±1/N 1/2 , so the error is small
for large N and large for small N .
For small optical depths the probability for scattering is 1−e−τ ∼ τ << 1,
which is equal to the expected number of scatterings in transversing the
medium. The reason is that e−τ is probablity for a photon NOT to scatter in
trasversing an optical depth τ so 1 − e−τ is the probability to scatter. Think
e.g. if τ = 1/5 then this implies 0.2 scatterings per system size.

18
LECTURE 4

Scattering + Absorption
Consider thermal material with absorption coefficient αν describing ther-
mal absorption and emission and a scattering coefficient σν describing coher-
ent isotropic scattering. The transfer equation is then

dIν /ds = −αν (Iν − Bν ) + σν (Jν − Iν ). (87)

The source function

Sν = (αν Bν + σν Jν )/(αν + σν ), (88)

is a weighted average of the thermal and scattering source functions. We can

define dτν = (σν + αν )ds as the differential optical depth to both scattering
and absorption, with extinction coefficient (σν + αν ).
Note limits that Sν = Bν for Jν = Bν (e.g. at large optical depths) and
Sν < Bν for regions where Jν → 0. (Note that both Jν and σν depend on
the presence of matter but not necessarily to the same functional powers).
Random Walk for Combined Scattering and Absorption
Mean free path to both scattering and absorption

lν = 1/(σν + αν ). (89)

The probability that a free photon will be absorbed rather than scattered is

ν = αν /(αν + σν ), (90)

and the probability for scatter is the single scattering albedo, and equals
1 − ν .
Consider again a homogeneous infinite medium: For large number of
scatterings N per absorption, N also equals the mean number of free paths
before absporption and the inverse of the probability of absorption. The
RMS net displacement of the photon is then (recall (85)

l∗2 = N lν2 = lν2 /ν = 1/(αν (αν + σν )). (91)

Here l∗ is the net displacement for a photon from creation to absorption.

It is the “effective diffusion length.” because the photon takes a stochastic
path to move this net distance before being absorbed. If we think of this

19
diffusion length as a kind of “modified mean free path” then The “effective
optical depth”
τ∗ = R/l∗ = [τa (τa + τs )]1/2 , (92)
where R is scale of system. When τ∗ << 1 we have a transluscent medium.
Why not transparent?: Because there can still be a lot of scattering. A
transparent medium would also be translucent by this definition, the former
being a stronger constraint in that addition to not being absorbed, there
would be little scattering. Transparent is both τs << 1 and τa << 1.
Note however that the analysis that led to (92) above really assumes
multiple scatterings and is most applicable when σs >> αs in which case
τ∗ ∼ (τa τ1s )1/2 .
For τ∗ << 1 the monochromatic luminosity from a thermal source would
just be determined by the emission coefficient:

L̃ν = 4πjν V = 4παν Bν V. (93)

For τ∗ >> 1 thermal equilibrium between matter and radiation is reached,

and in addition, the emission comes only from a thin layer of order l∗ , on
surface of area A. Thus

Lν ' 4παν Bν l∗ A = 4πν1/2 Bν A. (94)

This does not quite reduce to a black-body for zero scattering (i.e. ν = 1),
so numerical factor is not quite accurate. One part of the reason is that
l∗2 = lx2 + ly2 + lz2 ' 3lx2 , and since the radial direction in a star for example,
corresponds to only 1 of the three directions, rather than use l∗ in (94) we
should use l∗ /31/2 .
Note that at τ∗ > 1 matter and radiation approach thermal equilibrium,
thus l∗ is also called the “effective thermalization length.”
Radiative Diffusion Equation
Source function becomes black body at large effective optical depths for
homogeneous systems. Global homogeneity is not realistic though local ho-
mogeneity may be OK, as in the Sun or stars.
Consider “plane parallel” assumption: properties of an atmosphere de-
pend only on depth. Assume planar geometry. Intensity then depends on
angle from normal θ such that ds = dz/cosθ ≡ dz/µ and

µ∂Iν (z, µ)/∂z = −(αν + σν )(Iν − Sν ). (95)

20
If we consider a region deep in the atmosphere, the intensity varies slowly on
scale of mean free path. Intensity approximates a blackbody but with small
variation. We describe this as follows: To zeroth order in cosθ we have

Iν(0) = Jν(0) = Sν(0) = Bν (T ). (96)

Then we iterate and plug the next higher term back into (95), that is

µ ∂B(T )
Iν(1) ' Bν (T ) − . (97)
αν + σν ∂z
The flux is then
Z 1
Z
4π ∂Bν (T ) ∂T
Fν (z) ' Iν(1) (z, µ)cosθdΩ = 2π Iν(1) (z, µ)µdµ = − ,
−1 3(αν + σν ) ∂T ∂z
(98)
where the first term on the right of (97) does not contribute due to isotropy.
Integrating over all frequencies, we have
4π ∂T Z ∞ ∂Bν
F (z) ' − (αν + σν )−1 dν (99)
3 ∂z 0 ∂T
Note that
R∞
Z ∞
∂Bν (αν + σν )−1 ∂B
0
ν
dν Z ∞ ∂Bν 1 Z ∞ ∂Bν
(αν + σν )−1 dν = R ∞ ∂Bν ∂T
dν ≡ dν,
0 ∂T 0 ∂T
dν 0 ∂T αR 0 ∂T
(100)
where αR is the Rosseland mean absorption coefficient αR . But
Z ∞
∂Bν
dν = ∂B(T )/∂T = 4σT 3 /π, (101)
0 ∂T
using the Stefan-Boltzmann law. Thus

−16σT 3 ∂T
F (z) = . (102)
3αR ∂z
Key properties: 1) Flux increases with decreasing temperature gradient.
Makes sense since isotropy gets progressively violated at larger radii (nearer
to the stellar surface). 2) Radiative energy transport is similar to heat con-
duction.
Absorption Lines vs. Emission lines, & Limb Darkening

21
 From (88) and (90) above we have the source function

Sν = (1 − ν )Jν + ν Bν . (103)

Ignore scattering for the moment, and assume thermal emission. We are back
to
Sν = Bν . (104)
Recall solution of radiative transfer equation in this case: we have from (46)
for Iν (0) = 0
Iν = Bν (1 − e−τ
ν ).
ν
(105)
For small optical depths, this goes as

Iν ' τν Bν , (106)

so the intensity is larger in lines where the optical depth is larger.

But for a source with τ ≥ 1, we do not see the whole source, only the
layers which are optically thin. The temperature decreases outward in a
medium whose optical depth decreases with radius. We are then looking at
layers of lower temperature for lines, since the lines’ optical depth is larger
for a given distance of ray travel when compared to the continuum. Thus
we would see absorption lines for large optical depths in a system with radi-
ally decreasing temperature because the intensity for a higher temperature
blackbody exceeds that of a lower temperature blackbody at all wavelengths.
It is important to emphasize that the minimum trough of an absorption line
does not go to zero in intensity—just to a value lower than the continuum
that would be present in the absence of the line.
For a point source emitting I(0), located behind a emitting and absorbing
region in local thermal equilibrium we would have

Iν = Iν (0)e−τν + Bν (1 − e−τν ), (107)

When Iν (0) is brighter than the foreground (assuming that τν here increases
toward the observer) (Iν (0) − Bν (T ))e−τν decreases with increasing optical
depth. Since we see into the star less deeply for the line frequencies, the line
emission is dominated by outer regions whose local black body has a lower
flux.
A related issue explains limb darkening. White light image of Sun shows
edges darker than center because the edge view penetrates less deeply into

22
Sun for given optical depth, and thus sees emission coming from regions of
lower temperature than the face on view.
Eddington Approximation
Rosseland approximation employed the fact that system approached black
body at large effective optical depths. Eddington only assumes isotropy, thus
is good at smaller optical depths. We write a power series in µ = cosθ

Iν (τ, µ) = aν (τ ) + bν (τ )µ. (108)

If we take moments of this equation we obtain

1Z 1
Jν = Iν dµ = a (109)
2 −1
1Z 1
Hν = µIν dµ = b/3 (110)
2 −1
1Z 1 2
Kν = µ Iν dµ = a/3. (111)
2 −1
The fact that Kν = Jν /3 is the Eddington approximation, and is equivalent
to pν = uν /3 found in first lecture. The relation is found here for a more
general system with slight anisotropy.
Writing dτ (z) = −(αν + σν )dz (note sign here) we then have

µdIν /dτ = Iν − Sν , (112)

where dz = µds. Here ds is along the ray path and dzis the vertical direction
through the slab. Then taking moments in µ and integrating from −1 ≤ µ ≤
1 we get
dHν /dτ = Jν − Sν . (113)
and
1
dKν /dτ = H = dJν /dτ. (114)
3
Combining these last two equations we obtain

d2 Jν /dτ 2 = 3(Jν − Sν ) = 3ν (Jν − Bν ), (115)

where the latter eq. follows from (88) and (90). If ν does not depend on z,
we can write this
d2 Jν /dτ∗2 = (Jν − Bν ). (116)

23
 √ q
using the effective optical depth τ∗ = τ 3 = 3τa (τa + τs ) keeping in mind
the discussion below (94).
These are forms of the radiative transfer equation. If we have Bν (τ ), we
can solve for Jν and get Sν from (88) if the matter properties of the medium
are known (density, cross sections). Then we have what we need to solve
(112).
Text goes onto discuss two-stream approximation, for obtaining the two
boundary conditions. Assume the radiation propagates at two angles Iν± =
Iν (τ, µ = ±1/31/2 ). Then plug into the moment equations (109) and (110)
to get J = 21 (I + + I − ), H = 2√1 3 (I + − I − ), K = J/3, and use (114) to relate
eliminate H. This then gives two equations for the boundary conditions
1
Iν+ = Jν + √ ∂Jν /∂τ (117)
3
and
1
Iν− = Jν − √ ∂Jν /∂τ. (118)
3
The reason for this odd choice of µ is that it reproduces exactly Kν = Jν /3,
the Eddington approximation.
More on Moments
Consider integrating the intensity over various vector multiples of k. We
have Z
cuν = Iν dΩ (119)
and Z
Fν = kIν dΩ (120)
Z
cPν = kkIν dΩ. (121)
From fluid mechanics or E&M, different components of stress tensor refer
to rate of momentum transfer across surfaces of different orientations. Rate
of momentum transfer per photon across surface of unit normal k is is ckp.
Thus the total rate of momentum transfer is given locally by the radiation
stress tensor
2 Z 2 Z Z
3 4 3 3
X X
P= ck̂pfα d p = ck̂k̂(h ν /c )fα dΩdν = (1/c) k̂k̂Iν dΩdν
α=1 α=1
(122)

24
 The total derivative of Iν obeys
Z
∂t Iν + (1/c)k · ∇Iν = −αν Iν + jν − σν Iν + σν Φ(k̂, k̂0 )Iν (k0 )dΩ0 (123)

Compare this more general form to (87): In our previous discussions we have
ignored the 1st term on the left and replaced the last term on the right with
σν J ν . If we integrate over all solid angles of photon paths, we have, for “front
back symmetric scattering”

∂t uν + ∇ · Fν = −cαν uν + 4πjν . (124)

The next highest moment is obtained by multiplying (123) by k and

integrating over solid angle (assuming jν is independent of k) to obtain

∂t Fν + c∇ · Pν = −(αν + σν )Fν . (125)

Note that right side represents a radiation force on matter. (Sink for radiation
energy/momentum appears as source for matter energy/momentum and vice
versa. Removal of radiation momentum means gain in momentum of matter.)
Note that to fully solve the system of equations, one needs an infinite set of
equation because each equation depends on a higher moment. This is the
closure problem. (Also found in turbulence and MHD).
The Eddington approximation, combined with the approximation that
the systems evolve on time scales slower than the travel time associated
with a mean free path (e.g. ratio is 10−28 ) for sun, allows ignoring the time
derivative terms.
Note the general form of conservation law:

∂t (quantity density+∇·(quantity f lux) = (quantity sources−quantity sinks)

(126)

25
 LECTURE 5
The first four lectures discussed general properties of radiation transfer
and the some aspects of the specific case of a blackbody. The rest of the
course builds up to trying to understand the physics behind the emission,
absorption, and scattering coefficients of real systems, and how this relates
to the material constituents.
Basic E&M
The electromagnetic force on a particle moving with velocity v and charge
q is
F = q(E + v × B/c), (127)
where E and B are the electric and magnetic fields. The rate of work done
is
v · F = dt (mv 2 /2) = qv · E. (128)
For many particles the force density is

f = ρc E + j × B/c, (129)
P
where ρc = a qa na is the total charge density (sum of charge times number
P
density over particle type) and j = a qa na va . Rate of work per unit volume
is then X X
f a · va = qa na va · E = j · E. (130)
a a

The rate of work done per unit volume is the rate of change of mechanical
energy density:
dt umech = j · E. (131)
Maxwell’s Equations

∇ · D = 4πρc (132)

∇·B=0 (133)

∇ × E + c−1 ∂B/∂t = 0 (134)

∇ × H − c−1 ∂D/∂t = 4πj/c. (135)

26
Here D = E. This comes from

D = E + 4πP (136)

where P = a na hpa i is the electric polarization= dipole

P
moment density
0 3 0
R
averaged over spatial region. (Note dipole moment is ρc x d x ). For linear
response, we take P ∝ E or P = χE so

D = (1 + 4πχ)E ≡ E, (137)

where  is the permittivity=dielectric constant.

Similarly, for B = µH. Here H is the applied field and we have the
magnetization M = Σa na hma i, where the dipole moment is m = x ×jd3 x0 .
R 0

For linear response materials M = χm H. This leads to B = µH, where µ =

1+4πχm is the magnetic permeability. and χm is the magnetic susceptibility.
For the vacuum, B = H and D = E
Conservation laws of E& M
The law for charge conservation is

∂t ρ + ∇ · j = 0 (138)

.
Let us now derive the Poynting theorem of electromagnetic energy con-
servation. Use Maxwell’s equation (135) and dot with E. This gives

j · E = (1/4π)[c(∇ × H) · E − E · ∂t D]. (139)

Now use a vector identity on the first term on the right:

E · ∇ × H = H · ∇ × E − ∇ · (E × H). (140)

(prove using index tensor notation and ijk tensor). Use Maxwell’s, equation
(134), and assume the dielectric and magnetic permeability are independent
of time to get

(1/4π)∂t (E 2 + B 2 /µ) = −j · E − (c/4π)∇ · (E × H). (141)

This is Poyntings theorem. If we integrate we can see its true meaning:

Z Z Z
(1/4π)∂t (E 2 + B 2 /µ)d3 x = − j · Ed3 x − (c/4π) (E × H) · dA, (142)

27
where we use the divergence theorem on the last term. This shows that the
change in electromagnetic energy equals that lost to work plus that flowing
out the boundary. S ≡ cE × H/4π is the Poynting flux. It is the energy flux
vector (energy per unit time per area).
Note that it is possible to put the material currents into a redefinition of
j and ρc so that the constants do not appear in the time derivative term of
(142).
Definition of Radiation Field
For charges moving at a finite constant speed, the electromagnetic fields
E and B both fall as 1/r2 . Thus at large distances S · dA goes as 1/r2 .
R

However for the time varying fields of accelerated charges, we will find
that there are components of E, B that are ∝ 1/r, and so the Poynting flux
from these terms is finite are large distances. This is the definition of the
radiation field.
Plane Waves
Maxwell’s equation in vacuum

∇·E=0 (143)

∇·B=0 (144)

∇ × E + c−1 ∂t B = 0 (145)

∇ × B − c−1 ∂t E = 0. (146)
Then taking curl of (145), using (146) and the vector identity ∇ × (∇ × E) =
∇(∇ · E) − ∇2 E we get

∇2 E − (1/c2 )∂ 2 E/∂t2 = 0. (147)

Assume
E = â1 E0 ei(k·r−ωt) (148)
and
B = â2 B0 ei(k·r−ωt) , (149)
where â1 and â2 are unit vectors and k is the wave vector. These are trav-
eling wave solutions because the argument of the exponential is constant for
surfaces which propagate in time along the wave direciton.

28
 Plugging back into Maxwell’s equations we have

ik · â1 E0 = 0, (150)

ik · â2 B0 = 0 (151)
and
ik × â1 E0 = iωâ2 B0 /c, (152)

ik × â2 B0 = −iωâ1 E0 /c. (153)

These imply that k ⊥ â1 , â2 and â2 ⊥ â1 . They also imply

E0 = B0 ω/kc, B0 = E0 ω/kc, (154)

so
ω 2 = c2 k 2 (155)
and E0 = B0 . The phase velocity vph ≡ w/k = c. Group velocity is also c
because there is no medium and the group velocity depends on the index of
refraction n:
c
vg = , (156)
n(ω) + ω(∂n/∂ω)
but for the vacuum n = 1 and is independent of ω so vg = vph = c.
Wave Energy Flux and Density
Note that E and B vary sinusoidally in time, but in astro, we measure
time averages for measurements of interest. (The measuring time (observing
time) assumed to be longer than wave period.)
We are going to calculate products of complex quantities (the electro-
magnetic fields) but we measure only the real part. So we are interested in
the products of the real parts. Take two complex quantities Q1 = E0 eiwt and
Q2 = B0 eiwt . Then
1 1
Re(Q1 ) = (Q1 + Q∗1 ) = (E0 eiωt + E0∗ e−iωt ), (157)
2 2
where ∗ indicates complex conjugate. Similarly,
1
Re(Q2 ) = (B0 eiωt + B0∗ e−iωt ). (158)
2

29
(These follow from recognizing any complex number can be written as a + bi
with a and b real.) Then

Re(Q1 )Re(Q2 ) = (1/4)(E0 B0 e2iωt + E0 B0∗ + E0∗ B0 + E0∗ B0∗ e−2iωt ) (159)

Take time average, then the terms with exponentials vanish when the av-
erages are taken over sufficiently long time compared to the ω −1 . Thus we
have
1
hRe(Q1 )Re(Q2 )i = hE0 B0∗ + E0∗ B0 i. (160)
4
But only the real part of each term contributes. Thus
1 1 1
hRe(Q1 )Re(Q2 )i = hRe(E0 B0∗ + E0∗ B0 )i = hRe(E0∗ B0 )i = hRe(E0 B0∗ )i.
4 2 2
(161)
Applying this to the time averaged Poynting flux for waves gives
c c c c c
hSi = hRe(E)×Re(B)i = k̂h|Re(E)||Re(B)|i = k̂Re(E0 B0∗ ) = k̂|E0 |2 = |B0 |2 ,
4π 4π 8π 8π 8π
(162)
where we have again eliminated e±2iωt terms by averaging them a way. For
the electromagnetic energy density, we find
1 1 1 1
uem = hRe(E0 E0∗ + B0 B0∗ )i = |E0 |2 + |B0 |2 = |B0 |2 , (163)
16π 16π 16π 8π

so huem ik̂ = hSi/c, implying energy flows at speed c for vacuum plane waves
(not surprising).

30
 LECTURE 6 Radiation Spectrum

The radiation spectrum (histogram of energy as a function of frequency)

cannot be measured at an instant of time because of the “uncertainty” rela-
tion δωδt >
∼ 1. True for waves in general. Instead we will eventually consider
radiation received from a fixed location averaged over a “long” time.
Consider finite radiation pulse and consider the scalar E(t) = â1 ·E pulling
out one component of E. We then have, using Fourier transforms,
1 Z∞
Ẽ(ω) = E(t)eiωt dt, (164)
2π −∞
so that Z ∞
E(t) = Ẽ(ω)e−iωt dω, (165)
−∞

also note
Ẽ(−ω) = Ẽ ∗ (ω), (166)
since E is real. The Ẽ has the power spectrum info. Write the energy/(time·
Area) as
dW
= cE(t)2 /4π (167)
dtdA
from the result immediately above for the Poynting flux. For the total energy
per area we have, integrating (167)
Z ∞
dW
=c E 2 (t)dt/4π. (168)
dA −∞

But Z ∞ Z ∞
=c E(t)2 dt/4π = (c/2) |Ẽ(ω)|2 dω (169)
−∞ −∞

by Parsevals theorem relating the Fourier transform of a product to the

product of the Fourier transforms. Further, since |Ẽ(ω)|2 = |Ẽ(−ω)|2 , we
take the integral from 0 and multiply by 2 to obtain:
Z ∞
dW/dA = c |Ẽ(w)|2 dω, (170)
0

and
dW
= c|Ẽ(w)|2 . (171)
dAdω

31
But this is the total energy per frequency per area for the entire pulse. If pulse
repeats over the time T or if it is more or less stationary for long times, we
can determine the spectrum for some finite part of an “infinitely” long pulse,
based on the results for finite pulse. (The finite pulse issue enters to ensure
that the integrals above are finite). Spectrum is not useful if the characteristic
time changes T ≤ 1/δω, where δω is the characteristic frequency resolution.
We can write
dW dW
' = c|E(w)|2 /T, (172)
dAdωT dAdωdt
which is a power spectrum per unit area for the duration T , assuming that
the signal has approximately same properties over that duration. (fig 7)

Elliptical Polarization, Stokes Parameters for Monochromatic

Waves
Plane waves are linearly polarized. Waves of arbitrary polarization can
be generated from a linear combination of two orthogonal linearly polarized
waves. Consider the electric field

E = (E1 x̂ + E2 ŷ)e−iwt . (173)

We can write the complex amplitudes

E1 = ε1 eiφ1 (174)

and
E2 = ε2 eiφ2 , (175)
where ε1 , φ1 and ε2 , φ2 are real. so

E = ε1 x̂ei(φ1 −wt) + ε2 ŷei(φ2 −wt) . (176)

The real part is thus
Ex = ε1 cos(ωt − φ1 ) (177)

32
 Ey = ε2 cos(ωt − φ2 ). (178)
(fig 8)

These equations show that the electric field traces out an ellipse, as we
will now see. The general equation for an ellipse can be written as

(X/a)2 + (Y /b)2 = 1. (179)

We write
Ex0 = ε0 cosβcoswt (180)
and
Ey0 = −ε0 sinβsinwt, (181)
for −π/2 < β < π/2. These satisfy (179) using X = Ex0 , Y = Ey0 , a =
ε0 cosβ and b = ε0 sinβ. When the wave is propagating toward us, the ellipse
is traced out clockwise for 0 < β < π/2 and counterclockwise for −π/2 < β <
0. These represent left-handed and right-handed polarization respectively
(negative and positive helicity). When β = ±π/4 the ellipse becomes a
circle and we have circular polarization. For β = 0, ±π/2 we have linear
polarization.
To show that (177) and (178) also form an ellipse, perform a clockwise
rotation on the x0 , y 0 components (180) and (181) by angle χ (multiply by
the rotation matrix for clockwise rotation) so that we have

Ex = ε0 (cosβcosχcosωt + sinβsinχsinωt) (182)

and
Ey = ε0 (cosβsinχcosωt − sinβcosχsinωt). (183)
Then these match (177) and (178) if we set

ε1 cosφ1 = ε0 cosβcosχ (184)

33
 ε1 sinφ1 = ε0 sinβsinχ (185)
ε2 cosφ2 = ε0 cosβsinχ (186)
ε2 sinφ2 = −ε0 sinβcosχ. (187)
We can solve for ε0 , β, χ given ε1 , φ1 , ε2 , φ2 . This is normally done with
the monochromatic Stokes Parameters:

I = ε20 = ε21 + ε22 = E12 + E22 (188)

Q = ε20 cos2βcos2χ = ε21 − ε22 = E12 − E22 (189)

U = ε20 cos2βsin2χ = 2ε1 ε2 cos(φ1 − φ2 ) = E1 E2∗ + E2 E1∗ (190)
V = ε20 sin2β = 2ε1 ε2 sin(φ1 − φ2 ) = −i(E1 E2∗ − E2 E1∗ ) (191)
To prove (188-191) start with the left side of (184-187) and construct
combinations that give the right sides of (188-191): For example to prove
(188) we add the squares of (184-187) and use sin2 φ1 + cos2 φ1 = sin2 φ2 +
cos2 φ2 = 1.
To prove (189) we add the squares of (184) (185) and subtract from the
sum of (186) and (187), and then use the identities cos2 χ − sin2 χ = cos2χ
and cos2 β − sin2 β = cos2β.
To prove (190) first note that cos(φ1 − φ2 ) = cosφ1 cosφ2 + sinφ1 sinφ2 )
then add the product of (184) times (186) to the product of (185) times (187)
and use the identities 2sinχcosχ = sin2χ and cos2 β − sin2 β = cos2β.
Finally, to prove (191) first note that sin(φ1 − φ2 ) = sinφ1 cosφ2 −
sinφ2 cosφ1 then subtract the product of (184) times (187) from the prod-
uct of (185) times (186) and use the identities 2sinβcosβ = sin2β and
cos2 χ + sin2 χ = 1.

(192)
Pure elliptical polarization (polarization of monochromatic waves) is con-
trolled by the 3 parameters (ε0 , χ, β) but there are 4 Stokes parameters. This
is because they are related by

I 2 = Q2 + U 2 + V 2 . (193)

That
I = ε20 , (194)

34
highlights how I is related to the intensity of the wave. (Recall our calculation
of the energy density of a plane wave above). We have

sin2β = V /I (195)

and thus V measures the circularity of the wave polarization (see discussion
below (181), and is related to the ratio of the power of the linearly indepen-
dent components. If β = 0, then the wave is linearly polarized and V = 0.
If β = π/4, V /I = 1 implying circular polarization. Finally,

tan 2χ = U/Q, (196)

so that U or Q relate to the orientation of the ellipse that the electric field
vector traces with respect to a fixed set of axes. However for circular polar-
ization Q = U = 0 and I = V from (195), or (193).
Stokes Parameters for Quasi Monochromatic Waves
In general, the observed emission is a superposition of wave components
with different polarizations and different frequencies. The waves are generally
not monochromatic. The amplitude and phase for a given frequency can
change with time. We have

E1 (t) = ε1 (t)eiφ1 (t) (197)

and
E2 (t) = ε2 (t)eiφ2 (t) . (198)
Assume that the time variation is relatively slow, such that over a time
period of 1/ω the wave looks fully polarized, but such that for δt >> 1/ω
the amplitude and phases can vary. Then δωδt ∼ 1 implies δω << ω and so
we call this limit a quasi-monochromatic wave of bandwidth δω and coherence
time δt.
Most waves are measured in the time averaged sense by a detector. In
addition, measuring devices often split waves into a linear combinations of
two independent electric field polarization components. Assuming any time
for the wave to pass through measurement apparatus is short compared to
the coherence time δt, then the splitting of waves into the two components
is a linear transformation of the form

Ex0 ≡ E10 = λ11 E1 + λ12 E2 (199)

35
 Ey0 ≡ E20 = λ21 E1 + λ22 E2 (200)
Where the λs represent the effect of the measuring apparatus. The time
average sum of the squares of the x0 and y 0 components are measured. Using
the identity (161) we have for the x0 component
hRe(E∗x0 Ex0 )i = 2h[ReE10 e−iwt ]2 i = |λ211 |hE1 E1∗ i+|λ212 |hE2 E2∗ i+|λ11 λ∗12 |hE1 E2∗ i+|λ12 λ∗11 |hE2 E1∗ i,
(201)
where the terms with the residual e−iwt have been dropped assuming the
quasi-monochromatic approximation, and we only consider averages over
longer times.
The λs are known features of the apparatus. The four different terms
which represent averages of the radiation field can be written in terms of
Stokes parameters
I = hE1 E1∗ i + hE2 E2∗ i = hε21 + ε22 i (202)

Q = hE1 E1∗ i − hE2 E2∗ i = hε21 − ε22 i (203)

U = hE1 E2∗ i + hE2 E1∗ i = h2ε1 ε2 cos(φ1 − φ2 )i (204)

V = −i(hE1 E2∗ i − hE2 E1∗ i) = h2ε1 ε2 sin(φ1 − φ2 )i. (205)

These are the generalized Stokes parameters which reduce to the monochro-
matic ones when the amplitudes and phases are time independent. We cannot
distinguish waves with the same Stokes parameters. But here,
I 2 ≥ Q2 + U 2 + V 2 . (206)
Physically this makes sense when the wave is not completely elliptically po-
larized: If part of the radiation is unpolarized, the right side (206), which
is the intensity squared of the polarized contribution, is an undersetimate of
the total I 2 . The result follows from the Schwartz inequality
hE1 E2∗ ihE2 E1∗ i ≤ hE1 E1∗ ihE2 E2∗ i, (207)
where the equality holds when the ratio of E1 /E2 is independent of time.
(I leave as an exercise to show this from (197) and (198)). A completely
unpolarized wave has no preferred orientation so hε21 i = hε22 i and Q = U =
V = 0.

36
 LECTURE 7
Degree of Polarization
Stokes parameters are additive for waves of uncorrelated phase. That is,
if we take the net electric field vector for the sum of waves, no cross terms
P
contribute to the time averaging, that is for Ei = n Ei,n we have

hE1 E2∗ i = ∗ ∗
XX X
hE1,n E2,l i= hE1,n E2,n i. (208)
n l n
P P P P
Thus I = n In , Q = n Qn , U = n Un , and V = n Vn .
We can then decompose the total Stokes parameters for an arbitrary col-
lection of quasi-monochromatic waves into contributions from the polarized
part and the unpolarized part:

I, Q, U, V = [I −(Q2 +U 2 +V 2 )1/2 , 0, 0, 0]+[(Q2 +U 2 +V 2 )1/2 , Q, U, V ]. (209)

The former is the unpolarized part. The degree of polarization is the ratio of
the polarized part of the intensity to the total. (Note that we may be missing
constants to make I exactly equal to the total intensity but intensity ratios
are independent of this constant).

Π = (Q2 + U 2 + V 2 )1/2 /I = Ipol /I. (210)

As an example, consider partial linear polarization. Then V = 0 and

we take the plane of the polarized part to be the x0 plane. One can the
rotate a polarizing filter which will allow the maximum intensity when the
filter is aligned with the electric field. The maximum intensity is the sum
of the polarized part + 1/2 the unpolarized part because the intensity for
the unpolarized part is shared equally between any two orthogonal directions
and the polarized part only contributes when the filter is oriented properly.
Thus
Imax = Iunpol /2 + Ipol . (211)
The minimum intensity is

Imin = Iunpol /2. (212)

But Iunpol = I − (Q2 + U 2 )1/2 and Ipol = (Q2 + U 2 )1/2 . Thus

Imax − Imin
Π= , (213)
Imax + Imin

37
when V = 0, i.e. partial linear polarization.
When does the Macroscopic Transfer theory of “rays” apply?
When size of the area we are interested in, through which radiation passes,
approaches the wavelength of the radiation, then the macroscopic classi-
cal theory is not applicable. That is, the validity is for (dpx dx)(dpy dy) ∼
p2 dAdΩ > 2
∼ h or 2
dAdΩ > ∼λ . (214)
In addition we require
dνdt >
∼ 1, (215)
from the energy uncertainty relation. Thus when λ is greater than the scale
of interest, classical transfer theory of rays, as we have studied, fails (e.g. on
the scale of atoms).
But now that we have studied waves, we can be more precise. Define rays
as the curves whose tangents point along the direction of wave propagation.
Thus, rays are well defined only if the amplitude and wave direction are
nearly constant over a wavelength. That is the geometric optics limit.
To see the specific quantitative relations for the validity of this limit
assume that a wave (electric field vector) is represented by

g(r, t) = a(r, t)eiψ(r,t) . (216)

For a wave, g satisfies

c−2 ∂t2 g − ∇2 g = 0. (217)
For constant a, k = ∇ψ is direction of propagation and ω = −∂ψ/∂t is
frequency. These will be important later because we will show that the
approximations which lead to a plane wave approximation using these cor-
respondences to wavevector and frequency.
If we substitute (216) into (217) we get
1 2 1 1 a
∇2 a− 2
∂t a+ia(∇2 − 2 ∂t2 )ψ+2i(∇a·∇ψ− 2 ∂t ψ∂t a)−a(∇ψ)2 + 2 (∂t ψ)2 = 0.
c c c c
(218)
Now in the limit that the following relations are satisfied
1 1
|∇a| << |∇ψ|; |∂t a| << |∂t ψ| (219)
a a

1 2
|∇ a| << |∇ψ|2 (220)
a
38
 |∇2 ψ| << |∇ψ|2 ; |∂t2 ψ| << |∂t ψ|2 , (221)
(218) reduces to
1
(∇ψ)2 − 2
(∂t ψ)2 = 0, (222)
c
which represents the Eikonal equation describing the geometric optics limit.
The above limits represent the case for which the the amplitude is slowly
varying and the phase is rapidly varying. For nearly constant a, the direction
of propagation which is perpendicular to surfaces of constant phase, is

k = ∇ψ. (223)

and the frequency would be given by

ω = −∂t ψ. (224)

Using these in (222) then reproduces the usual plane wave relation of k 2 =
w2 /c2 . Thus Eikonal equation expresses the geometric optics limit, and that
limit leads to the relation between wave number and frequency we found
earlier for plane waves. This is the relationship that applies in the regime of
the classical transfer theory.
Electromagnetic Potentials
E and B can be expressed in terms of the electromagnetic 4-vector po-
tential Aµ = (φ, A), where φ is the scalar potential and A is the vector
potential. We have
1
E = −∇φ − ∂t A. (225)
c

B = ∇ × A. (226)
Then, incorporating the bound and free charges into ρ and j, we have
1 2 1 1
−∇ · E = ∇2 φ − ∂t φ + ∂t (∇ · A + ∂t φ) = −4πρ (227)
c2 c c
and
1 1
∇ × B + ∂t (∇φ + ∂t A) = 4πj/c, (228)
c c
2
and ∇ × B = ∇(∇ · A) − ∇ A Gauge Invariance

39
 E and B are invariant under gauge transformations, that is under the
change
1
A → A + ∇χ; φ → φ − ∂t χ. (229)
c
Only E and B are measured and so the choice of gauges is merely a choice
to make the equations easier to solve. Different gauges are more or less
convenient in different contexts.
The Coulomb gauge is such that ∇ · A = 0. The Lorenz gauge employs
∂t φ + ∇ · A = 0. For the latter case we obtain for the potentials
1 2
∇2 φ − ∂ φ = −4πρ (230)
c2 t
and
1
∇2 A − ∂t A = −4πj. (231)
c2
The formal solutions are
Z
[ρ] 3 0
Z Z
ρ(r0 , t0 )δ(t0 − t + |r − r0 (t0 )|/c) 0 3 0
φ(r, t) = dr = dt d r (232)
|r − r0 | |r − r0 |

and
1Z [j] 3 0
Z Z
j(r0 , t0 )δ(t0 − t + |r − r0 (t0 )|/c) 0 3 0
A(r, t) = d r = dt d r ,
c |r − r0 | |r − r0 |
(233)
where the brackets indicate evaluation at the retarded time. Thus these are
the retarded time potentials. The retarded time means that the quantity is
evaluated at a time tret = t − |r − r0 |/c (where t is the present time) due to
finite speed of light travel. Thus e.g. [ρ] = [ρ(r0 , tret )].
Formally, given these potentials, we can solve for E and B.
Potential of Moving Charges
Consider charge q following path r = r0 (t), u = ṙ0 . Then the localized
charge and current densities are

ρ(r0 , t0 ) = qδ(r0 − r0 (t0 )) (234)

j(r0 , t0 ) = qu(t0 )δ(r0 − r0 (t0 )). (235)

The volume integrals give the total charge q and current qu, for a single
particle.

40
 Using the formal solutions (232) and (233), we can calculate retarded
potentials for this single particle, these are
Z
δ(t0 − t + R(t0 )/c) 0
φ(r, t) = q dt (236)
R(t0 )
and Z
δ(t0 − t + R(t0 )/c) 0
A(r, t) = q u(r0 ) dt , (237)
R(t0 )
where R(t0 ) = |R(t0 )| and

R(t0 ) = r − r0 (t0 ). (238)

If we now use tret = t − R(t0 )/c or c(t − tret ) = R(t0 ). We then can change
variables. Let
t00 = t0 − tret (239)
so that
0 2 0 2 0
dt00 = dt0 − dtret = dt0 + dt0 Ṙ(tc ) = dt0 + dt0 2cR(t
1
0)
dR (t )
dt0
= dt0 + dt0 2cR(t
1
0)
dR (t )
dt0
0 0 u(t0 )
= dt (1 − n̂(t ) · c ),
(240)
0
where n̂ = R/R and u(t ) = ṙ0 = −Ṙ. (Note that we consider t and r
fixed in this differentiation ; we are not integrating or differentiating over t
or r, there are just the time and position at which the quantities are being
measured.) Thus using (240) in (236) and (237) gives
Z
1 00 00 q
φ(r, t) = q δ(t )dt =
R(t0 )(1 − n̂(t0 ) · u(t0 )/c) R(tret )(1 − n̂(tret ) · u(tret )/c)
(241)
and
Z
1 qu(tret )
A(r, t) = q u(t0 )/c δ(t00 )dt00 = ,
R(t0 )(1
0 0
− n̂(t ) · u(t )/c) R(tret )(1 − n̂(tret ) · u(tret )/c)
(242)
00 0
where we have integrated δ(t ) by setting t = tret usingR (239). Recall that
integrating a function over a δ function works like this: δ(q − q0 )f (q)dq =
f (q0 ).
Notice the factor appearing on the bottom which “concentrates” the po-
tentials along the direction of motion for strongly relativistic flows. This is

41
related to the concept of relativistic beaming of radiation into narrow cones
along the direction of motion that we will derive later. These potentials are
also evaluated at the retarded time–again important for very relativistic mo-
tion. This retarded time is also the key to getting the radiation part of the
electromagnetic fields, that is the part that falls off as 1/r rather than 1/r2 .
The potentials above are the Liénard-Wiechart potentials.

42
 LECTURE 8
Velocity and Radiation Components of the Electromagnetic Fields
To get E and B at r and t note that they depend on the retarded position
and time rret and tret . At the retarded time, the particle has u = ṙ0 (tret ) and
acceleration u̇ = r̈0 (tret ). There is some messy algebra to obtain, from the
potentials presented at the end of last lecture, the fields (see Jackson)
~˙
   
(n̂ − ~
β)(1 − β 2
) q n̂ × ((n̂ − ~ × β)
β)
E(r, t) = q  +   (243)
κ3 R2 c κ3 R

and
B(r, t) = [n̂ × E], (244)
where β~ = u/c and κ = (1 − n̂ · β). ~ Where the brackets again indicate
retarded time, and so only the conditions at tret matter even though particle
is currently at r, t.(fig3.1).

The first term in the electric field is just the relativistic generalization of
Coulomb’s law for moving particles. It falls off as 1/R2 . This is the “velocity
field.” It points in the line toward the current position of the particle.
The second term is the radiation field. It falls off as 1/R. Alas we see
why the Poynting flux associated with radiation from an accelerating particle
can be finite at large distances.
The effect can be seen from the Figure below. (fig 3.2 of R&L).

Suppose the particle is originally moving at constant speed and then it

43
stops. This information is not immediately communicated to the exterior.
Only for distances less than c(t1 − t0 ) is the field informed that the particle
decelerated, but outside this field is that of the particle had it not decelerated
(that is, as if it were at x=1). Thus there is the kink in the field if the field
is to match together for the two regions.
The thickness of this the interface region is constant and equal to the
light travel distance during the period of deceleration or acceleration. Now
take a ring whose central axis is the line of particle motion and the plane of
the ring lies perpendicular to the field lines from the interface region, such
that they penetrate the ring. Since Gauss’ law implies the volume integrated
electric flux is constant,
Z
E · dS = constant (245)

Wth time, the ring keeps same thickness, but its radius increases as R if it
is to keep up with the interface region field lines. To conserve flux the field
must fall of as 1/R.
Energy/(Freq· Solid angle) for single particle
From Lecture 6 we have that
dW c 2
= E (t). (246)
dAdt 4π
Then using dA = R2 dΩ we have
dW c c
= (R(t)E(t))2 = (Q(t))2 = (247)
dΩdt 4π 4π
where Q(t) = R(t)E(t), so that (using the radiation part of the electric field)

q 2 Z ∞
Z ∞ 2 2
dW c
 
= cQ̃(ω)2 = 2

 iwt 
[RE(t)]e dt

= 2  ~ × β)κ
[n̂ × ((n̂ − β) ~˙ −3 ]eiwt dt .
dΩdω 4π 
−∞ 4π c −∞ 
(248)
The brackets again mean evaluated at retarded time t0 = t − R(t0 )/c. Chang-
ing time variables using dt =qκdt0 and using the Taylor expansion of R(t0 )
in r0 for small r0 /r (R(t0 ) = (r − r0 )2 ∼ (r2 − 2r · r0 )1/2 = |r| − (rr·r
2 )1/2 =
0

|r| − n̂ · r0 ) gives

dW q 2 Z ∞ 2
 
= 2  n̂ × ((n̂ − β) ~˙ −2 eiw(t0 −n̂·r0 (t0 )/c) dt0  .
~ × β)κ (249)
dΩdω 4π c −∞ 

44
Then using the identity

n̂ × ((n̂ − β) ~˙ −2 = dt0 (κ−1 n̂ × (n̂ × β)

~ × β)κ ~ (250)

we can integrate (249) by parts to obtain

dW q 2 ω 2 Z ∞ 2
 
= 2  ~ iw(t0 −n̂·r0 (t0 )/c) dt0  ,
n̂ × (n̂ × β)e (251)
dΩdω 4π c −∞ 

where a factor |eiω|r|/c |2 = 1 has been extracted from the integrand.

The integration by parts (really just the chain rule for differentiation)
led to two terms, one of which was of the form dt0 (U V )dt0 , which vanished
R

under the assumption that E falls off rapidly at t → ±∞.

Radiation from Non-relativistic systems
In the non-relativistic limit β << 1. Also the ratio of the second term on
the right of (243) to the first term on the right is Erad /Evel = Ru̇/c2 . For a
particular frequency of oscillation ν

Erad /Evel = uRν/c2 = (u/c)(R/λ). (252)

For the near zone, R ≤ λ, the velocity field dominates, whilst for the far zone
R ≥ λc/u, the radiation field dominates.
When β << 1,
~˙
Erad = (q/c)[n̂ × (n̂ × β)/R] (253)
and
Brad = [n̂ × Erad ]. (254)
Then
q β̇
Erad = Brad = sinΘ, (255)
Rc
˙
where Θ is the angle between β~ and n̂. The Poynting flux is given by

dW c 2 c q 2 β̇ 2
=S= Erad = sin2 Θ. (256)
dAdt 4π 4π R2 c2
The vector Poynting flux S ∝ E × B is directed along n̂. Using dA = R2 dΩ
we have
dW q 2 β̇ 2
= sin2 Θ. (257)
dtdΩ 4πc

45
The total power emitted is then

q 2 β̇ 2 Z 1 2 2q 2 β̇ 2
dW/dt = (1 − cos Θ)d(cosΘ)) = , (258)
2c −1 3c
which is Larmor’s formula for emission from a single particle. (fig3.3)

Note that as a function of angle, there is a dipole pattern: no radiation

emitted along the direction of acceleration. The maximum is emitted per-
pendicular to the acceleration. Also, the instantaneous direction of Erad is
˙
determined by β~ and n and 1-D acceleration means linear polarization in the
˙
β~ and n plane.
Dipole Approximation
For a system of particles, differences in the retarded times can be ignored
if the time scales for a system to change are τ >> L/c, the light crossing time.
Since this also characterizes an inverse frequency of radiation, the equivalent
condition for ignoring retarded time differences is if L << c/ω = λ Since
the time for variation also characterizes the time for change in velocity of a
particle, a long time for change in velocity over a particle orbit of radius l
implies τ ∼ l/u >> L/c. But l ≤ L, so u << c is also an expression of the
same limit.
We can use the non-relativistic radiation fields in these regimes (general-
izing (253) for many particles)
X qa
Erad = ~˙
[n̂ × (n̂ × β)/Ra ]. (259)
a c

For large distances from the source, Ra ∼ R0 , where R0 can be defined as the
distance from the geometric centroid of the particle system to the observer.
Then
1
Erad = [ 2 n̂ × (n̂ × d̈)], (260)
c R0

46
where d = a qa ra ' ρc rd3 r, where the latter similarity follows for a nearly
P R

continuous charge distribution. We have

dP d̈2
= sin2 Θ (261)
dΩ 4πc3
and
2d̈2
P = 3 (262)
3c
fig (3.5)

Using Fourier transforms

Z ∞
d(t) = ˜
e−iωt d(ω)dω (263)
−∞
so Z ∞
¨ =−
d(t) ˜
ω 2 e−iωt d(ω)dω (264)
−∞

and
1 4 ˜
dW/dωdΩ = ω |d(ω)|2 sin2 Θ (265)
c3
and
8π 4 ˜
dW/dω = ω |d(ω)|2 (266)
3c3
It can be shown that the dipole term represents the first term in the
small expansion parameter L/λ. (or Lω/c ∼ (L/l)(u/c) << 1), where l is the
particle orbit scale and u is the particle velocity. As the system gets more and
more relativistic, the lowest order expansion is not sufficient. The frequencies
which show up in the radiation spectrum are not only that associated with
the dipole oscillation of the particles, but higher harmonics, where the dipole
produces the primary frequency ω0 but and the quadrupole etc. produce the
higher integer harmonics.

47
 LECTURE 9
Thomson Scattering
We want to apply the dipole formula to the case for a charge radiating
in response to an impinging wave. Ignore β~ × B force for non-relativistic
velocities. The electric force is the main contributor and we have the equation
of motion for forcing from a linear polarized wave

mr̈ = F = e~E0 sinω0 t, (267)

or
d̈ = (e2 /m)~E0 sinω0 t, (268)
where e is the charge and ~ is the direction of E0 . The solution is an oscillating
dipole
d = er = −(e2 E0 /mω02 )sinω0 t. (269)
fig (3.6)

Time averaged power is then

d̈2 2 e4 E02
dP/dΩ = 3
sin Θ = 2 3
sin2 Θ, (270)
4πc 8πm c
where we used LimT →∞ (1/T ) 0T sin2 (ωt)dt = 1/2. We also have upon inte-
R

gration over solid angle P = e4 E02 /3m2 c3 .

Define the differential cross section dσ for scattering an initial electro-
magnetic wave into solid angle dΩ from interaction with the particle. Then

dP dσ cE02 dσ
= hSi = , (271)
dΩ dΩ 8π dΩ
therefore from (270)
dσ
= ro2 sin2 Θ, (272)
dΩ polarized

48
where
e4
r02
= 2 4. (273)
mc
Note here that hSi is the incident flux.
For an electron, r = 2.8 × 10−13 cm, and is called the classical electron
radius. Integrating over solid angle gives

σ = 8πr02 /3 (274)

which is σT = 6.6 × 10−25 cm2 , called the Thomson cross section for an elec-
tron. This remains valid when hν < 0.5M eV , the electron rest mass. Above
this we are in the quantum regime, and the “Klein Nishina” cross section
must be used (discussed later). Dipole approximation is also invalid for rel-
ativistic motions.
Polarization of Electron scattered radiation
The scattered radiation is polarized in the plane spanned by initial po-
larization vector ~ and n̂.
Differential cross section (as a function of angle into which radiation is
scattered) for scattering of initially unpolarized radiation is found by writing
the incident radiation as the sum of two orthogonal linearly polarized beams.
Take ~1 to be in plane of incident and scattered beam.
Take ~2 to be ⊥ to this plane.
Take Θ as the angle between ~1 and n̂. ~2 ⊥ n̂, and n is the direction of
the scattered wave.
(fig3.7)

Then using (272) the differential cross section for scattering into dΩ from
an initially unpolarized beam is

ro2 1 1
dσ/dΩ|unpo = [dσ(Θ)/dΩ + dσ(π/2)/dΩ] = (1 + sin2 Θ) = (1 + cos2 θ),
2 2 2
(275)

49
where θ = π/2 − Θ. This depends only on the angle between the direction of
the incident and scattered radiation. Note that integrated cross section for
polarized radiation is the same as for the unpolarized case (because electron
has no intrinsic direction and energy is conserved).
Recall that n, the outgoing wave direction, is perpendicular to the polar-
ization vector of the scattered radiation. The two differential cross sections
on the right of (275) refer to the intensities of these two planes of polarization.
The ratio of the differential cross section for scattering into the plane and
perpendicular to the plane of scattering is cos2 θ, so we have, from (213)
Imax − Imin 1 − cos2 θ
Π= = . (276)
Imax + Imin 1 + cos2 θ
This means that initially unpolarized radiation can be polarized upon scat-
tering. This makes sense. If we look along incident direction, then we get
no polarization since there is azimuthal symmetry around the initial wave
vector axis and all directions are the same. If we look ⊥ to the initial wave
vector axis θ = π/2, then we see 100% polarization. This is because the
particle moves in the plane ⊥ to the initial direction.
Radiation Damping Force
The radiation damping force can be assumed to be perturbation on the
particles motion when Trad >> tp , i.e. the time over which the particles
energy is changed significantly by radiation is much longer than the time for
the particle to change its position (e.g. orbit time).
3mc3
 2
u
Trad ∼ mv 2 /Prad = . (277)
2e2 u̇
This then requires that tp >> τ = r0 /c = 10−23 s, the light crossing time of
a classical electron radius.
For scattering problems, an example of this holds when the distance elec-
trons wander in an atom (typically of order 1 Angstrom) is << than the
wavelength of the radiation—that is the wavelength on which the electric
field varies (e.g. >> 1 Angstrom.)
The radiative loss of energy, time averaged is
hdW/dti = P = (2e2 /3c3 )h|ẍ|2 i, (278)
where
1 Z t+tp /2 2
h|ẍ|2 i = ẍ dt, (279)
tp t−tp /2

50
where tp = 2π/ω. Since we are considering x = x0 eiωt , upon integrating by
parts we note that products of odd and even numbers of derivatives produce
a real part that cancels out, the only term that survives is

1 Z t+tp /2 ˙
h|ẍ|2 i = − ẍ · ẋdt, (280)
tp t−tp /2

We can then write

hdW/dti = −hFrad · ẋi, (281)
so
2e2 ˙
Frad = ẍ. (282)
3c3
Radiation from Harmonically Bound Particles, Classical Line
Profile
Assume that particle is bound to a center of force. Then its motion
represents oscillation and its equation of motion

F = −kr̈ = −mω02 r, (283)

such that the oscillation is with frequency ω0 . A classical oscillator (Thomson

atom) is not ultimately applicable, though quantum results are quoted in
relation to the classical results.
The radiation reaction force for a radiating particle damps the oscillator.
We assume that the damping is a perturbative effect. From (277), the validity
of this classical non-relativistic regime is when τ /tp << 1 so that, τ , the light
crossing time over a classical electron radius is short compared to any particle
orbit period. This means the time scale to lose significant energy to radiation
is long compared to particle orbit time. For oscillation along the x − axis:

−τ ẍ˙ + ẍ + ω02 x = 0. (284)

If the 3rd derivative term is small we can assume the motion is harmonic to
lowest order. Then we assume x = cos(ω0 t + φ0 ) to this order. We then have
τ ẍ˙ = −ω02 ẋ so
ẍ + ω02 τ ẋ + ω02 x = 0. (285)
To solve assume x(t) = x0 eαt so that

α2 + ω02 τ α + ω02 = 0. (286)

51
The solution is
1
α = ±iω0 − ω02 τ + O(τ 2 ω03 ), (287)
2
obtained by solving quadratic and expanding for ωτ << 1. At t = 0 take
x(0) = x0 and ẋ(0) ' 0 [i.e. ẋ(0) ∼ ω0 x0 (ω0 τ ) << ω0 x0 ]. We thus have
1
x(t) = x0 e−Γt/2 cosω0 t = x0 (e−Γt/2+iω0 t + e−Γt/2−iω0 t ). (288)
2
where Γ = ω02 τ .
If we Fourier transform we can find the power spectrum of the emitted
radiation in the dipole approximation. We have
" # " #
1 Z ∞ 0 iwt0 0 x0 1 1 x0 1
x̃(ω) = x(t )e dt = + ∼ ,
2π 0 4π Γ/2 − i(ω + ω0 ) Γ/2 − i(ω − ω0 ) 4π Γ/2 − i(ω − ω0 )
(289)
where we ignore the first term on the right as we are working for the regime
of ω near ω0 . The energy radiated per unit frequency is then
8πω 4 ˜ 8πω 4 e2
! " #
dW 1
= 3
|d(ω)|2 = 3 2
x20 . (290)
dω 3c 3c 16π (Γ/2) + (ω0 − ω)2
2

The power spectrum is then a Lorentz profile. (fig 3.8)

The power spectrum for a particle damped by radiation–classical line.

There is a sharp max near ω = ω0 , and the width is δω ∼ Γ = ω02 τ << 1.
The quantity in curved parentheses can be thought of as a spring constant,
characterizing the potential energy of the particle. Integrating (290) over
frequency gives the total energy of 21 kx20 . This is the energy that is being
radiated away.
Driven Harmonically Bound Particles
In this case we have a forcing from beam of radiation, so the equation of
motion is
ẍ = −ω02 x + τ ẍ˙ + (eE0 /m)cosωt. (291)

52
Now take x to be complex so we have

ẍ + ω02 x − τ ẍ˙ = (eE0 /m)eiωt . (292)

The solution is
x = x0 eiωt ≡ |x0 |ei(ωt+δ) , (293)
where
eE0 /m
x0 = (294)
w2 − ω02 − iω03 τ
and
ω3τ
" #
δ = tan−1 . (295)
w2 − ω02
(We find the latter by writing x0 in the form x0 = a/(b − ic) = a(b+ic) a2 +b2
=
ab ac
qcosδ + iqsinδ, with a, b, c, q, δ all real. Then qcosδ = b2 +c2 and qsinδ = b2 +c2
so tanδ = c/b.)
The damping term from the radiation reaction provides a phase shift. For
w > w0 , δ > 0 and for ω < ω0 , δ < 0. This represents an oscillating dipole
with frequency ω and amplitude |x0 |.
The time averaged power radiated is then

e4 E02 ω4
P (ω) = e2 |x0 |2 ω 4 /3c3 = . (296)
3m2 c2 (ω 2 − ω02 )2 + (ω03 τ )2

Dividing by the time averaged Poynting flux hSi = cE02 /8π we “project out”
the cross section for scattering

ω4
σ = σT . (297)
(ω 2 − ω02 )2 + (ω 3 τ )2

For ω >> ω0 σ(ω) → σT . Basically this is because at high incident

energies, the frequency of the particle motion is not seen by the radiation, so
it appears free.
For ω << ω0 , we have σ ∝ ω 4 . This is the Rayleigh-Scattering regime—
bluer frequencies scattered more (thus looking at horizon means we the Sun
to be redder). Inertial term in equation of motion is small, so the dipole
moment of the particle is directly proportional to the incident radiation.
kx = eE. (e.g. (291) with ẍ and ẍ˙ terms dropped.)

53
 For ω ∼ ω0 we have ω 2 − ω02 ∼ 2ω(ω − ω0 ) so

σT Γ 2π 2 e2 Γ
σ(ω) = = , (298)
τ 4(ω − ω0 )2 + Γ2 mc 4(ω − ω0 )2 + Γ2

recalling that Γ = ωo2 τ . This is the Lorentz profile. This result arises because
free oscillations of the unforced oscillator can be excited by a pulse of radi-
ation E(t) = δ(t). The spectrum for this pulse is independent of frequency.
(White spectrum) and so the emission is just scattering of a white spectrum.
The emission is then proportional to the cross section for this scattering.
P = hSiσ
If we integrate σ(ω) over ω we find that
Z ∞
2π 2 e2
σ(ω)dω = ∼ σT /τ. (299)
0 mc
But this is not valid because at large ω the classical approximation breaks
down (classical requires ωτ << 1, where τ is light crossing time across clas-
sical electron radius). Actually we have
Z ωmax
σ(ω)dω = σT ωmax , (300)
0

because the range of validity of the classical radiation reaction is, ωmax <<
1/τ = c/r0 .
For quantum regime however, we write
Z ∞
σ(ω)dω = (σT /τ )fmm0 ∼ σT /τ, (301)
0

where fmm0 is the oscillator strength for m → m0 transitions. (fig 3.9)

Correspondence between emission and Einstein Coeff.

54
 Note the power radiated in the Larmor formula time averaged over 1 cycle
for a harmonic oscillator is
e2 ω04
P = |x0 |2 . (302)
3c3
But this should equal A21 hν21 using the form of the Einstein coefficients,
where ν21 is the resonance line frequency. We can then say that |x0 |/2 =
h2|x|1i is 1/2 the displacement of the oscillation representing the expectation
value between the transition from state 2 to state 1. Then we would have
8πe2 ω21
3
A21 = |h2|x|1i|2 , (303)
3c3 h
which turns out to be not a bad approximation for specific cases in in the
quantum regime (e.g. when the statistical weights of the two states are non-
degenerate).

55
 LECTURE 10
Basic Special Relativity
We review some concepts in special relativity very quickly.
First note the basic postulates:
1) laws of physics take the same form in frames in relative uniform motion.
2) speed of light is the same in free space for these frames, independent
of their relative velocity.
The appropriate transformations of the coordinates x, t that preserve
these relations between two frames K and K 0 such that K 0 is moving with
positive velocity v on the x axis for a fixed observer in K are

x0 = γ(x − vt); y = y 0 ; z = z 0 (304)

and
t0 = γ(t − vx/c2 ), (305)
where γ = 1/(1 − v 2 /c2 )1/2 We talk about a 4-dimensional space-time coor-
dinate taking place at (t, x).
Note that a consequence of the constant speed of light is that for a pulse
of light emitted at t = 0 where we assume the origins coincide is

x2 + y 2 + z 2 − c2 t2 = x02 + y 02 + z 02 − c2 t02 = 0. (306)

Length Contraction
Consider a rigid rod of length L0 = x02 −x01 carried in rest frame K 0 . What
is the length in the unprimed frame? The length is given by L = x2 − x1
where x2 and x1 are the rod positions as measured at fixed time in L. We
have
L0 = x02 − x01 = γ(x2 − x1 ) = γL. (307)
Thus the length we measure at a fixed time in L is smaller than the proper
length measured in L0 . The length of the moving rod is contracted.
Time Dilation
Assume a clock at rest in K 0 measure a time interval T0 = t02 − t01 at fixed
position x0 = 0. This implies that

T = t2 − t1 = γ(t02 − t01 ) = γT 0 . (308)

Thus we see that more time elapses in the unprimed (=lab) frame per unit
time in the source (moving) frame. Thus “moving clocks run slow.” This is
the time dilation.

56
 The point is that clocks at two different positions in K are not simu-
latenously synchronized as seen by K’. When x02 = x01 in K’, x2 6= x1 . At
these two locations in K, clocks in K do not appear to be synchronized as
measured by an observer in K 0 even if they are synchronized as seen by an
observer in K.
Transformation of Velocities and Beaming of Radiation
Write the transformations in differential form for boost along x:
dx = γ(dx0 + vdt0 ); dy = dy 0 ; dz = dz 0 (309)
and
dt = γ(dt0 + vx0 /c2 ), (310)
where we have also taken advantage of the symmetry of relative frames and
let v → −v and t0 , x0 → t, x.
Then we have
γ(dx0 + vdt0 ) u0x + v
ux = dx/dt = = (311)
γ(dt0 + vdx0 /c2 ) 1 + u0x v/c2
and
u0y
uy = (312)
γ(1 + u0x v/c2 )
and
u0z
uz = . (313)
γ(1 + u0x v/c2 )
fig 4.2

In general,
u0|| + v
u|| = (314)
1 + u0|| v/c2
and
u0⊥
u⊥ = . (315)
γ(1 + u0⊥ v/c2 )

57
Thus we can take the ratio of the perpendicular to the parallel velocities as
measured in the lab frame (K). We have

u0⊥ 1 + u0|| v/c2 u0⊥ u0 sinθ0

tanθ = u⊥ /u|| = = = .
γ(1 + u0⊥ v/c2 ) u0|| + v γ(u0|| + v) γ(u0 cosθ0 + v)
(316)
0
For u = c, corresponding to photons emitted in the source frame, we
have
sinθ0
tanθ = (317)
γ(cosθ0 + v/c)
and
cosθ0 + v/c sinθ0
cosθ = ; sinθ = (318)
(1 + cosθ0 v/c) γ(1 + cosθ0 v/c)
At θ0 = π/2, we have
1
tanθ = ; cosθ = v/c, (319)
γv/c

so that sinθ = 1/γ.

This means that if a source is emitting isotropically in its rest frame (K’
frame) then in the K frame, 50% of the emission is beamed into a cone of
angle 1/2γ.
This has many consequences in astrophysics. Both in the microphysics of
radiation from moving particles, and also for relativistic bulk outflows such
as Active Galactic Nuclei and Gamma-Ray Bursts. It means for example,
that a for sufficiently relativistic and radiating bipolar jet, we would only see
emission from the flow moving toward our line of sight.
Doppler Effect
It is particularly important in astrophysics to distinguish between local
observers (those omnipresent entities that have a rod and clock at each point
in space time) and actual observers (those like us fixed at a point). There is
no Doppler effect for local observers.
In the rest frame of K consider the moving source to move from point 1
to point 2 and emit one radiation cycle. The period as measured in the lab
frame (K) is given by the time dilation effect

∆t = γ∆t0 = 2πγ/ω 0 , (320)

58
but difference in arrival times must be considered because the source is mov-
ing relative to the observer. Consider the distance d to be positive if the
source moves toward the observer. (fig 4.4)

Then accounting for the difference in arrival times gives

∆tobs = ∆t − d/c = ∆t(1 − (v/c)cosθ) = γ∆t0 (1 − (v/c)cosθ), (321)

where θ is now the angle between the direction of motion and the line of sight
as measured in the lab frame (K). Rearranging we then have

ω0
ωobs = 2π/∆tobs = (322)
γ(1 − (v/c)cosθ)

or as the inverse
ωobs = ω 0 γ(1 + (v/c)cosθ0 ), (323)
where θ0 is measured in the K 0 frame. Note that the γ is a purely relativistic
effect whereas the (1 + (v/c)cosθ0 ) is also a non-relativistic effect.
Combined Doppler Effect and Special Relativity
Rearranging (322) and assuming the case for which θ ∼ 0, and v ∼ c we
have
(1 − v/c)(1 + v/c)
ω 0 ' ωobs γ(1 − (v/c)cosθ) = γ ∼ ωobs /2γ. (324)
(1 + v/c)
or
∆tobs ∼ ∆t0 /2γ. (325)
Compare this to (320): the γ is in the opposite place! Thus there is no time
dilation but rather time contraction for a source moving at the observer (see
E. Blackman 1998, Eur. J. Phys. 19 195 for more discussion.) “A clock
moving at the observer runs fast!” This highlights that for astrophysics,
where we the observer are at a fixed location, that the standard time dilation

59
and Lorentz contraction effects must be considered carefully. (See Rees 1966,
Nature 211 468 for superluminal motion effects; and ).
Proper time
Note that the quantity dτ , the proper time, is in an invariant:

c2 dτ 2 = c2 dt2 − (dx2 + dy 2 + dz 2 ) = c2 dτ 02 . (326)

It is called the proper time because it measures the time interval between
events occurring at the same spatial location. Dividing by c2 this equation
implies dt2 (1 − v 2 /c2 ) = dτ 02 or dt2 = γ 2 dτ 02 which is the time dilation effect.
Notes on 4 vectors
The quantity xµ = (ct, x, y, z) is a four-vector, in that it obeys the trans-
formation properties previously described. The Greek indices indicate 4 com-
ponents. Note also that
xµ = ηµν xν (327)
where summation is over repeated indices and ηµν = 0 for µ 6= ν, −1 for
µ = ν = 0 and 1 for µ = ν > 0. The Lorentz transformations can be written
0
xµ0 = Λµν xν (328)
0
where Λµν is the transformation matrix, which, for a boost along x satisfies
0
Λµν = γ for µ = ν = 0, 1; 1 for µ = ν = 2, 3; −βγ for µ = 0, ν = 1,
and −βγ for µ = 1, ν = 0. General Lorentz transformations have more
complicated matrices. We restrict ourselves to isochronous (Λ00 ≥ 0), and
proper transformations detΛ = +1.
The norm of xµ is given by

xµ xµ = ηµν xµ xν . (329)

Since this is a Lorentz invariant, we have

0 0
Λτν Λ̃νσ0 = δστ 0 (330)
0
where δστ 0 is the Kronecker delta.
All 4-vectors transform in the same way as xµ under a Lorentz transfor-
mation. Some other 4-vectors include k µ = (ω/c, k) for a electromagnetic
wave, the 4-velocity U µ = (γc, γv), the current density 4-vector j µ = (ρc, j).
the momentum 4-vector (γE/c, γp).

60
 The transformation laws also hold for tensor indices. Thus two Λ matrices
are required for a two index tensor. That is for example
0 0 0 0
Aµ ν = Λµσ Λνδ Aσδ . (331)

It turns out that the electric and magnetic field are not separately 4-
vectors but comprise part of of an electromagnetic tensor

Fµν = ∂µ Aν − ∂ν Aµ (332)

where Aµ = (−φ, Ai ) is the 4-vector potential with components (Latin indices

mean only 1,2,3 components.) F0i = −Ei , Fi0 = Ei , F00 = 0, F12 = Bz ,
F21 = −Bz , F13 = −By , F31 = By , F23 = Bx , F32 = −Bx , and the rest zeros.
Maxwell’s equations can then be written

∂ ν Fµν = 4πjν /c (333)

which includes ∇ · E = 4πρ and ∇ × B = c−1 ∂t E + 4πj/c. and

∂σ Fµν + ∂ν Fσµ + ∂µ Fνσ = 0, (334)

which includes ∇ · B = 0 and ∇ × E + c−1 ∂t B = 0. Note that Fµν (E) =

ηµσ ηνλ F σλ (E) = F µν (−E).
Lorentz invariants are F µν Fµν = 2(E2 − B2 ) and Det(F ) = (E · B)2 .
Lorentz Transformations of the Fields
For a boost with velocity β~ we have

E0|| = E|| , (335)

B0|| = B|| (336)

and
E0⊥ = γ(E⊥ + β~ × B) (337)
and
B0⊥ = γ(B⊥ − β~ × E). (338)
For a physical idea of the transformations, consider a charged capacitor
with surface charge density σ 0 in its rest frame and plate separation d. (fig)

61
 Place the capacitor plates perpendicular to the direction of motion (x).
In the lab frame K the capacitor moves with speed vx . Since E||0 = 4πσ 0 , and
the area does not contract with this motion as seen in K, E|| = E||0 . There is
no surface current either.
Now rotate the plates 90 degrees so that they are parallel to the direction
of motion, parallel to the x-z plane. Now the perpendicular electric field
Ey = Ey0 γ because the surface charge density as seen in K increases from the
Lorentz contraction along the direction of motion (length of plate appears
smaller). Now since the current per area and charge density transform as a
4-vector we have
j = γ(j0 + vx x̂σ 0 /d) (339)
and since j0 = 0 we have the only component jx = γvx σ 0 /d. Now the current
is also given by ∇ × B. Since the only direction of variation is perpendicular
to the plates, (the y direction in the K frame) we have

(∇ × B)x = ∂y Bz = 4πjx /c, (340)

so that
Bz ' 4πjx d/c = 4πγvx σ 0 . (341)
We can see that this result also follows directly from (338) which gives

Bz = γ(Bz0 + βx Ey0 ) = 4πγvx σ 0 , (342)

since Bz0 = 0. This exemplifies one restricted case of the transformations

above.
Fields of Uniformly Moving Charge
We want to apply the transformation relations for the fields to a charge
moving with constant velocity along x. In the primed (rest) frame we have

Ex0 = qx0 /r03 ; Bx0 = 0 (343)

Ey0 = qy 0 /r03 ; By0 = 0 (344)

Ez0 = qz 0 /r03 ; Bz0 = 0, (345)
with r0 = (x02 + y 02 + z 02 )1/2 . We use the inverse transform for the relations
(335-338), (β → −β, and switch primed and unprimed quantities) to find

Ex = qx0 /r03 ; Bx = 0 (346)

62
 Ey = qγy 0 /r03 ; By = −qγβz 0 /r03 (347)
Ez = qz 0 /r03 ; Bz = qγβy 0 /r03 . (348)
But these are in the primed coordinates. We can transform to unprimed
coordinates by using the simple coordinate transformations to get

Ex = qγ(x − vt)/r03 ; Bx = 0 (349)

Ey = qγy/r03 ; By = −qγβz/r03 (350)

03 03
Ez = qz/r ; Bz = qγβy/r , (351)
2
with now r0 = (γ 2 (x − vt) + y 2 + z 2 )1/2 .
Now let’s see that these fields are exactly the same as those derived from
retarded time and the Lienard-Wichart potentials. In particular, we show
that the electric fields just calculated are the same as the first term on the
right of (243) which is the constant velocity (including zero velocity) contri-
bution to the electric field. Let tret = t − R/c, z = 0. fig 4.5

Then we have
R2 = y 2 + (x − vt + vR/c)2 (352)
so solving for R gives

R = γ 2 β(x − vt) + γ(y 2 + γ 2 (x − vt)2 )1/2 . (353)

The unit vector to the field point from the retarded time position is then
yŷ + (x − vt + vR/c)x̂)
n̂ = (354)
R
and
κ = 1 − n · β~ = (y 2 + γ 2 (x − vt)2 )1/2 /γR. (355)
Thus
q γRq
= . (356)
γ 2 R2 κ3 (y 2 + γ 2 (x
− vt)2 )3/2

63
If we combine the last 3 equations with (349) (350), and (351) we have
h i
~
E = q(n̂ − β)(1 − β 2 )/κ3 R2 (357)

which is the same result we get for the velocity field of a moving charge from
the Liénard-Wiechart potentials.

64
 LECTURE 11
Field from a strongly relativistic charge
fig 4.6

For γ >> 1 at field point at x = 0, y = b we have

qvγt
Ex = − ; Bx = 0 (358)
(γ 2 v 2 t2 + b2 )3/2
qγb
Ey = ; By = 0 (359)
(γ 2 v 2 t2
+ b2 )3/2
Ez = 0; Bz = βEy ∼ Ey . (360)
Fields are are concentrated mainly in plane transverse to motion (Ey direction)
in cone of angle 1/γ (found from ratio of (Ex /Ey )max ). Book calculates spec-
trum for this uniformly moving particle. The key point is that the spectrum
has peaks at frequency given by 1/∆t = γv/b which is the time period for
which the electric field is significant–very short–thus a very broad peak in
spectrum. Ẽ 2 (ω)max ∼ Ey,max
2
(∆t)2 ∼ (qγ/b2 )2 (b/γv)2 ∼ q 2 /(b2 v 2 ). Spec-
trum nearly flat for ∆ω ∼ 1/∆t ∼ γv/b.
fig 4.7

Relativistic Mechanics
The four-momentum of a particle is given by

pµ = muµ = muµ = m(cγ, γv) = (E/c, γmv). (361)

65
Expanding the zeroth component in the non-relativistic regime

p0 c = mc2 (1 − β 2 )−1/2 = m(c2 + v 2 /2) + ... (362)

which is the rest energy plus the kinetic energy. The norm is

pµ pµ = −m2 c2 = −E 2 /c2 + p2 (363)

or
E 2 = m2 c4 + c2 p2 = (γmc2 )2 . (364)
For a photon we have
h
pµ = (ω/c, k), (365)
2π
and pµ pµ = 0.
If we take the derivative with respect to the proper time ( a Lorentz
invariant scalar) we have the acceleration 4-vector.

aµ = duµ /dτ. (366)

The generalization of Newton’s law is

F µ = maµ . (367)

Note that
m
F µ uµ = d(uµ uµ )/dτ = 0. (368)
2
This implies that 4-surfaces of uµ and F µ are orthogonal, so that as you move
along F µ as a function of its parameters, you are moving along different 4-
planes in uµ . This means that the four force is a function of uµ so 4-force
has a dependence on 4-velocity.
For the case of the electromagnetic field, the covariant generalization of
the Lorentz force
F = qE + v × B/c (369)
can be written in terms of the electromagnetic tensor as
q
F µ = Fνµ uν = maµ . (370)
c
This indeed has a 0th component which reduces to

dW/dt = qE · v (371)

66
and the ith components

dpi /dt = qEi + (v × B)i /c. (372)

Total Emission from Relativistic Particles

Move into “local Lorentz frame.” Even if particle is accelerating, we
can move into a frame for which particle moves non-relativistically for short
times away from the initial time. Then we use the dipole formula, and then
transform back to lab frame.
The radiation has zero momentum for “front-back symmetric” emission
in this source frame. Using the lorentz transformation for the momentum
and positon 4-vectors only the zeroth components then matters in the primed
(co-moving) frame and we have and

dW = γdW 0 ; dt = γdt0 . (373)

Thus
dW 0 /dt0 = dW/dt. (374)
The non-relativistic dipole formula is

2q 2 0 2 2q 2
P = P0 = a = 3 aµ aµ (375)
3c3 3c
where the last equality follows because in the instantaneous rest frame of the
0
emitting particle, aµ = (0, a).
We can write this in terms of the 3-vector acceleration by quoting the
result that
a0|| = γ 3 a|| (376)
and
a0⊥ = γ 2 a⊥ (377)
so
2q 2 0 2 2q 2 4 2
P = a = 3 γ (a⊥ + γ 2 a2|| ). (378)
3c3 3c

67
 LECTURE 12
Angular Distribution of Emitted and Received Power
fig 4.9

Now consider dW 0 emitted into angle dΩ0 = sinθ0 dθ0 dφ0 , where θ0 is to the
x-axis. Let µ = cosθ and µ0 = cosθ0 . We do not assume forward backward
symmetry here so the momentum is not zero. Then

dW = γ(dW 0 + vdp0x ) = γ(1 + βµ0 )dW 0 , (379)

using the relation between photon energy and momentum. From (318) we
have
µ0 + β
µ= (380)
1 + βµ0
so that differentiating and using dφ = dφ0 we get

dΩ0
dΩ = . (381)
γ 2 (1 + βµ0 )2

Thus
dW/dΩ = γ 3 (1 + βµ0 )3 (dW 0 /dΩ0 ) (382)
To get the differential emitted power in the lab frame we divide by the
time as determined from time dilation dt = γdt0 .
To get an observed differential power we must include the Doppler effect
so that dtobs = γ(1 − βµ)dt0 from (321).
We then have

dPe /dΩ = γ 2 (1 + βµ0 )3 dP 0 /dΩ0 = γ −4 (1 − βµ)−3 dP 0 /dΩ0 , (383)

without taking into account Doppler but just time dilation and

dPobs /dΩ = γ 4 (1 + βµ0 )4 dP 0 /dΩ0 = γ −4 (1 − βµ)−4 dP 0 /dΩ0 (384)

68
when we do take into account Doppler effect. I have used the inverse of (323)
and the principle that we can interchange primed with un-primed variables
and change the sign of v to obtain the inverse transformation. The factor
γ −4 (1−βµ)−4 is peaked near θ = 0 for small θ which is highlights the beaming
effect described earlier.
(fig 4.10)

Apply to a moving particle. Recall that in the non-relativistic regime

(and thus the proper frame here)

0 q 2 a02
0
dP /dΩ = 3
sin2 Θ0 (385)
4πc
Thus

q 2 (γ 2 a2|| + a2⊥ ) 2 0
dP/dΩ = sin Θ , (386)
4πc3 (1 − βµ)4
where we have used (376) and (377).
acceleration || velocity: Then Θ0 = θ0 so

sin2 θ
sin2 Θ0 = sin2 θ0 = , (387)
γ 2 (1 − βµ)2

using the inverse of (318). Then

q 2 2 sin2 θ
dP|| /dΩ = a . (388)
4πc3 || (1 − βµ)6

for acceleration ⊥ velocity Then CosΘ0 = cosφ0 sinθ0 (where φ0 = 0 is

aligned or anti-aligned with a⊥ = a) measured from acceleration so

sin2 θcos2 φ
sin2 Θ0 = 1 − . (389)
γ 2 (1 − βµ)2

69
Then
q2 2 sin2 θcos2 φ
!
1
dP⊥ /dΩ = a ⊥ 1 − . (390)
4πc3 (1 − βµ)4 γ 2 (1 − βµ)2
Extreme relativistic limit
Since
1
β = (1 − 1/γ 2 )1/2 ∼ 1 − 2 (391)
2γ
and
µ = cosθ = 1 − θ2 /2 (392)
we have
1 + γ 2 θ2
(1 − βµ) ' . (393)
2γ 2
and (388) becomes
16q 2 a2|| γ 12 θ2
dP|| /dΩ = (394)
πc3 (1 + γ 2 θ2 )6
while (390) becomes

4q 2 a2⊥ 8 (1 − 2γ 2 θ2 cos2φ + γ 4 θ4 )
dP⊥ /dΩ = γ . (395)
πc3 (1 + γ 2 θ2 )6
Notice the dependence on φ, the angle traced in the plane perpendicular to
v.
(fig 4.11)

The beaming as a result of the θγ factors is highlighted in the figures.

Notice that the maximum powers occur for θ ∼ 1/γ in the case of parallel
acceleration, and θ < 1/γ for perpendicular acceleration. Note also the non-
azimuthal symmetry for the
Specific Intensity and Invariants
Consider group of particles moving along x in source (comoving) frame
with phase space measure d3 x0 d3 p0 . There is slight spread in momentum but

70
not in energy: this is because the energy of a single particle in this frame is
E 0 = mc2 + mu02 /2 + .. ∼ mc2 , which follows because u0 ∼ 0 in the comoving
frame, so deviations from u0 = 0 are quadratic in u0 , and we ignore second
order term in the co-moving frame. This implies that phase space measure
is invariant because d3 x → d3 x0 /γ, but dpx = γ(dp0x + βdE/c), and since
dE = 0 the result of invariant phase-space measure follows.
So d3 xd3 p is an invariant and thus dN/d3 xd3 p = fα (x, p, t) is also an
invariant. But we related the latter to the intensity in equation (20):
2
(h4 ν 3 /c3 )fα (x, p, t).
X
Iν = (396)
α=1

so Iν /ν 3 is an invariant, as is Sν /ν 3 .
Now let us find the transformation property of the absorption coefficient
αν . Consider thickness l of material of material moving along x with β. The
optical depth as seen by an observer at angle π/2−θ to the vertical is invariant
since e−τ measures the fraction of photons passing through material.
fig 4.12

Then
τν = αν (l/sinθ) = ναν (l/νsinθ) (397)
is invariant, but νsinθ is the vertical momentum component and l is perpen-
dicular to the direction of motion, so these quantities are not affected by the
motion along the x − axis. Thus ναν must be an invariant.
Since the emission coefficient jν = Sν αν = ndP/dΩdν, jν /ν 2 is also an
invariant.

71
 LECTURE 13
Bremsstrahlung =Free-Free process
Radiation from acceleration of charge in Coulomb field of another charge.
Quantum treatment required, as process can produce photons comparable in
energy to that of emitting particle, but do classical treatment first.
Bremsstrahlung is zero in dipole approximation for two of the same type of
particles: think of diplole moment for like particles: its zero.
Electrons provide most of the radiation since acceleration is inversely pro-
portional to mass.
Emission from Single Speed Electrons
Consider electron moving in field of nearly stationary ion (at origin).
Consider small-angle scattering regime.
Consider electron of charge −e moving past ion of charge Ze with impact
parameter b.
fig 5.1

Dipole moment is d = −eR where R is the separation, and so

d̈ = −ev̇, (398)

so Fourier transform gives

2 e Z ∞ iwt
−ω d̃(ω) = − v̇e dt. (399)
2π −∞
Consider asymptotic limits of large and small frequencies. Since τ ∼ b/v
is the interaction time, then the integral mainly contributes over this time.
When ωτ >> 1 exponential oscillates rapidly and integral is small.
When ωτ << 1 exponential is unity so
e
d̃ ∼ ∆v, (400)
2πω 2
with ∆v the change in velocity over the interaction time.

72
 Using
dW 8πω 4 ˜
= |d(ω)|2 (401)
dω 3c3
we have
dW 2e2
= |∆v|2 . (402)
dω 3πc3
The acceleration is mainly perpendicular to the electron path in the small
angle approximation. The change in velocity is then found by integrating the
perpendicular component of acceleration:
Z Z ∞
∆v = a⊥ dt = (e2 Z/me ) b/(b2 + v 2 t2 )3/2 dt = 2e2 Z/me bv, (403)
−∞

where we are in the ωτ << 1 limit, and extend the integral to infinity.
For small angle scatterings then

dW 8e2 Z 2
= (404)
dω 3πc3 m2e v 2 b2

For ωτ << 1 and 0 otherwise.

To get the total energy per unit frequency for a number density of ions
ni and electrons ne with velocity v, we note that the flux of electrons per ion
is ne v. Then, the number of electrons per unit time per ion that interact is
ne v2πbdb. Then multiplying by the number density of ions and integrating,
we have Z ∞
dW
= 2πni ne v (dW (b)/dω)bdb. (405)
dωdV dt bmin

The main contribution to the integral, by assumption of our regime, is

for ωτ << 1 or b << v/ω and so substitute (404) for the integrand

dW 16Z 2 e6
= 3 2 ni ne ln(bmax /bmin ) (406)
dωdV dt 3c me v

bmax = v/ω. There are two possibilities for bmin . One is classical, when
∆v = v (i.e. the limit defined by when small angle approximation becomes
invalid). From (403) this can be approximated (for convenience 2 is replaced
by π4 )
4Ze2
bmin ∼ . (407)
πme v 2

73
Second is the quantum limit ∆x∆p = h/2π or bmin,q = h/(2πmv), so
bmin,q = h/mv. (408)
When bmin,q < bmin , then classical treatment is ok. When bmin,q > bmin then
mv 2 > Z 2 Ry ∼ Z 2 13.6eV. Here the interaction time is short, the scale is
small, and need quantum treatment.
Convention is to write the intergal with either limit as a Gaunt factor
gf f (v, ω) so
dW 16πZ 2 e6
= √ 3 2 ni ne gf f . (409)
dωdV dt 3 3c me v
The Gaunt factor is of order 1 usually.
Thermal Bremsstrahlung Emission = Thermal Free-Free Emis-
sion
We can integrate single speed expression above over thermal distribution
of electrons. The probability for electron to have velocity in range d3 v for
an isotropic distribution of velocities is
2 /2kT
dQ = e−E/kt d3 v = v 2 e−mv dv (410)
A subtlety is that the lower bound for the velocity must satisfy 2hν <
2
mv , so that photon can be produced (remember we are in non-relativistc
limit so relevant energy is mv 2 /2). Then
R∞ 2 −mv 2 /2kT
dW (T, ω) vmin [dW (v, ω)/dV dωdt]v e dv
= R∞ 2
2 −mv /2kT dv
(411)
dV dtdω 0 v e
The emission coefficient is then given by:
dW (T, ω)
4πjνf f = f f,ν = = 6.8×10−38 Z 2 ne ni T −1/2 e−hν/kT ḡf f (T, ν). (412)
dV dtdω
Note the dependence on the number density squared for ne = ni .
The Gaunt factors tend to be of order a few for a wide range of velocity.
(problem in book fig 5.3 with the Gaunt factor X-axis graph. u = hν/kT )
Note that the exponential tail is due to the cutoff in the velocity. Also,
the spectrum is more or less flat until hν/kT ∼ 1.
To get the total power per volume, one integrates over frequency. This
ultimately gives
!1/2
32πe6
!
dW 2πkT
= Z 2 ne ni gB (T ) = 10−27 T 1/2 ne ni Z 2 gB (T ).
dtdV 3 3hm3/2 c3
(413)

74
where gB (T ) is frequency averaged Gaunt factor.
Thermal Bremsstrahlung Absorption = Thermal Free-Free Ab-
sorption
Consider electron absorbing radiation while moving in field of an ion.
(Free-free absorption).
For thermal processes we have Kirchoff’s law

jνf f = ανf f Bν (T ), (414)

where ανf f is the free-free abs. coefficient and jνf f is the associated emission
coefficient as above. Plugging in for the Planck function and using fν f above
we have
In the Rayleigh-Jeans regime hν/kT << 1, expand exponential and then:

ανf f = 0.018T −3/2 Z 2 ne ni ν −2 ḡf f (415)

In the limit hν >> kT , the Wien regime, we have instead

ανf f = 3.7 × 108 T −1/2 Z 2 ni ne ν −3 ḡf f . (416)

Recall that the Rosseland mean absorption coefficient was defined when
we considered plane-parallel atmospheres in the limit of Kirchoffs law at large
optical depths leading to
16σT 3 ∂T
F (z) = − . (417)
3αR ∂z
For Bremsstrahlung,
R ∞ 1 ∂Bν
1 0 dν
αf f ∂T 1
ff = R ∞ ν∂Bν = . (418)
αR 0 ∂T
dν 1.7 × 10−25 T −7/2 Z 2 n e ni ḡR

Relativistic Bremsstrahlung
One can use the method of virtual quanta to get a relativistic correction
to Bremsstrahlung emission. (Electrons moving relativistically). Ultimately,
we want a correction in the lab frame, in which the electron is moving. Here
is the procedure
1) First move to rest frame of electron.
2) The electron sees the ion’s field as pulse of electromagnetic radiation
emitted by the ion (the virtual quanta). This pulse is scattered by the elec-
tron.

75
 3) Then transform back to the rest frame of the ion to get the emission.
Roughly, consider the pulse of virtual quanta in the electron frame dW 0 /dA0 dω 0 .
Then project out the scattered part by multiplying by the electron scattering
cross section. (σT for low freq hν << mc2 ). This gives

(dW 0 /dω 0 )sc = σT dW 0 /dA0 dω 0 . (419)

Then note that dW 0 and dω 0 transform the same way under a Lorentz trans-
formation. So that
(dW/dω)sc = (dW 0 /dω 0 )sc . (420)
Then appeal to forward backward symmetry of the scattering, and write
primed frame quantities such as ω 0 and b0 in terms of quantities in the un-
primed frame by a Lorentz transformation.
For a thermal distribution of electrons, one must integrate over the ther-
mal distribution function. Then frequency integrating, one obtains
Z
dW/dV dt = 4π jν dν = 1.4 × 10−27 Z 2 ḡB T 1/2 ni ne (1 + 4.4 × 10−10 T ). (421)

The second term on the right is the relativistic correction. Important for
quite large temperatures.

76
 LECTURE 14
Synchrotron Radiation
Radiation of particles (e.g. electrons) gyrating around a magnetic field is
called cyclotron radiation when the electrons are non-relativistic and syn-
chrotron radiation when the electrons are relativistic. The spectrum of
cyclotron emission frequency is a peak at the frequency of gyration. Syn-
chrotron frequency picks up higher harmonics and is not as simple.
Total emitted synchrotron power:
Motion of a particle in a magnetic field is governed by
e µ ν
aµ = F u (422)
mc ν
with components
qv × B
dt (γv) = (423)
mc
and
qv · E
dt γ = = 0. (424)
mc2
Thus γ and |v|2 are constants so

qv × B
γdt v = (425)
mc
and
γdt v|| = 0 (426)
and
q
dt v ⊥ = v⊥ × B. (427)
γmc
fig 6.1

The magnitude of v is a constant, and the parallel component is un-

changed, the motion represents helical motion along a magnetic field. This

77
is circular motion in the frame where v|| = 0. The gyration frequency is
ωB = qB/γmc. The acceleration (a⊥ = ωB v⊥ ) is perpendicular to the veloc-
ity and so the power emitted is

2q 2 0 2 2q 2 4 2 2 2 2q 2 4 2 2 2 2 2 2
P = a = γ (a ⊥ + γ a || ) = γ a ⊥ = r cβ γ B , (428)
3c3 3c3 3c3 3 0 ⊥
where we are now assuming that the electrons are raditating so r0 = e2 /me c2
is the classical elecron radius as defined earlier.
If we average over pitch angles 0 ≤ ψ ≤ π then

β2 Z 2β 2
hβ⊥2 i = sin2 ψdΩ = . (429)
4π 3
Then
4 4 4
P = cr02 β 2 γ 2 B 2 = σT cβ 2 γ 2 Um ' σT cγ 2 Um . (430)
9 3 3
2
where σT = 8πr0 /3.
Spectrum of Synchrotron Radiation
fig 6.2

To estimate this, note that observer sees emission during the time the
particle moves from point 1 to 2 below. Consider the extremely relativistic
regime γ >> 1.
The emission from synchrotron radiation will be spread over a broader
range of frequency than just the gyro-frequency because the time interval
over which observer sees emission is much smaller than the gyro-frequency.
The uncertainty relation then implies the broader spectrum.
Radius of curvature is a = ∆s/∆θ. We see that ∆θ = 2/γ so ∆s = 2a γ
.
The radius of curvature is also given by asinα = v/ωB , which follows from

∆v ∆θ
γm = (q/c)v × B = γmv 2 , (431)
∆t ∆s

78
where |∆v| = v∆θ and ∆s = v∆t.
Thus
a = ∆s/∆θ = γmcv/qBsinα = v/ωB sinα, (432)
where α is the pitch angle. Therefore
∆s = 2v/γωB sinα. (433)
The particle passes between points 1 and 2 in a time t2 − t1 so that
2
(t2 − t1 ) ∼ ∆s/v = . (434)
γωB sinα
Defining tA1 and tA2 as the arrival times of radiation at the observation point
from points 1 and 2, the difference tA2 − tA1 is less than t2 − t1 by an amount
∆s/c, which is the time for radiation to move ∆s. Thus
2
(tA2 − tA1 ) ∼ (1 − v/c). (435)
γωB sinα
For v ∼ c this becomes
1
(tA2 − tA1 ) ∼ . (436)
γ 3 ωB sinα
Thus for a finite pitch angle, observed pulse duration is γ 3 times smaller
than gyro-period. Pulse frequency cutoff will be something like
ωc ∼ 1/∆tA = γ 3 ωB sinα. (437)
fig 6.3

Now since the electric field, as reflected in (??) and (395) is only a function
of the product θγ, we can write E(t) = F (γθ), with t in the lab frame. Let
s = t = 0 when pulse is along the axis of the observer. Then
t = ∆tA = (s/v)(1 − v/c). (438)

79
and
vt
θ ∼ (s/a) = , (439)
a(1 − v/c)
so
γvt
θγ = = 2γ 3 tωB sinα ∼ ωc t. (440)
(1 − v/c)a
Thus
E(t) ∝ g(ωc t). (441)
The Fourier transform is then
Z ∞
Ẽ(ω) ∝ g(ζ)eiωζ/ωc dζ, (442)
−∞

where ζ = ωc t.
The spectrum dW/dωdΩ ∝ Ẽ 2 . Integrating over solid angle and dividing
by the orbital period gives the time averaged power per frequency

dW/dtdω = (ωB /2π)dW/dω = P (ω) = KF (ω/ωc ), (443)

where K is a constant which can be fixed by integrating and comparing to

the total power derived earlier. Since the total power emitted for β ∼ 1 is
given by

2q 4 B 2 γ 2 sin2 α Z ∞ Z ∞
Ptot ' =K F (ω/ωc )dω = ωc K F (ξ)dξ, (444)
3m2 c3 0 0

where ξ ≡ ω/ωc . Then using

3γ 2 qBsinα
ωc = (445)
2mc
√
3 3
P (ω) = KF (ω/ωc ) = ωc KF (ξ)dξ/dω = q BsinαF (ω/ωc ), (446)
2πmc2
where the numerical integral factor is arbitrary, and comes from choice of
normalization of the dimensionless integral.
Spectral Index for power law electron distribution
Often the emission spectrum can be described by a power law in frequency
for some range. That is,
P (ω) ∝ ω −s (447)

80
where s is the spectral index. Electrons also often follow a power law distri-
bution when they have been accelerated. For relativistic electrons, consider
an energy distribution
N (γ)dγ = Cγ −p dγ. (448)
Then the total power radiated is given by
Z γ2 Z γ2
Ptot (ω) = C P (ω)γ −p dγ ∝ F (ω/ωc )γ −p dγ. (449)
γ1 γ1

If we change variables of integration to x ≡ ω/ωc and use ωc ∝ γ 3 ωB ∝ γ 2

we get Z x2
1−p p−3
Ptot (ω) ∝ ω 2 F (x)x 2 dx. (450)
x1

For wide enough frequency limits, take x1 ∼ 0 and x2 ∼ ∞. Then we have

Z ∞
1−p p−3
Ptot (ω) ∝ ω 2 F (x)x 2 dx. (451)
0

The spectral index is then related to the particle distribution index:

p−1
s= . (452)
2
For shocks, p is often 2 − 4 so spectral index is 0.5 − 1.5. Relevant for radio
jets in AGN (e.g. Blackman 1996).
Note that F (x) is something like a Gaunt factor, but it is narrowly peaked
helping to justify the assumption of wide limits of integration. fig 6.6

Summary of Results: (1) emission into 1/γ half angle, (2) emission up to
critical frequency ωc and dependence only on ω/ωc , and (3) spectral index
for power law electron distribution is s = (p − 1)/2.

81
 LECTURE 15
Synchrotron Radiation Polarization
Notice that for an electron, the emission is elliptically polarized as the
electron moves around the field line. Looking inside the cone of maximal
emission, the electron moves around counterclockwise and the emission is
left circularly polarized. Seen from outside the cone of maximal emission
the electron appears to move clockwise, and the emission is right circularly
polarized.
For a distribution of particles that varies relatively smoothly with Sinα,
the elliptical polarization should cancel and we would get a linear polariza-
tion, largely in plane perpendicular to the magnetic field. fig 6.5

The power perpendicular and parallel to the magnetic field as projected

onto the plane of the sky can then be used to calculate the polarization. For
particles of a single energy
P⊥ − P||
Π(ω) = . (453)
P⊥ + P||

The frequency integrated value is 0.75. For a power law distribution N (γ) ∝
γ −p , the polarization is
P⊥ − P|| p+1
Π= = . (454)
P⊥ + P|| p + 7/3

Note that the larger p the higher the polarization. Why?

Transformation from Cyclotron to Synchrotron Radiation
For low energies the electric field oscillates with the same frequency of
gyration of the electron around the magnetic field. This produces a single
line in the emission spectrum. fig 6.8 ab

82
 For slightly higher energies, the beaming kicks in and so the E-field is
stronger when the particles move toward the observer. As this effect increases
for larger and larger velocities, the spectrum picks up multiple harmonics as
shown in the expansion discussed earlier in the course. fig 6.10 ab

Eventually mapping out the F (ω/ωc ) function.

The spectrum becomes continuous for a system in which there are a distri-
bution of particle energies, or for which the field is not exactly uniform. The
electric field received by observer from a distribution of particles is the ran-
dom superposition of many pulses. The spectrum is the sum of the spectra
from the individual pulses as long as the average distance between particles
is larger than a gyro-radius.
Distinction Between Emitted and Received Power
Emitted power is not equal to received power. We need to consider the
Doppler effect as the particle moves toward us. fig 6.11

Thus for period of the projected motion of T = 2π/ωB , we have the

period measured by the received emission.
v 2π
TA = T (1 − v|| cosα/c) = T (1 − cos2 α) ∼ sin2 α (455)
c ωB
for v ' c.
The fundamental observed frequency is ωB /sin2 α. This implies that the
spacing between harmonics is larger, being ωB /sin2 α.

83
 Recall that we got the emitted power by dividing by ωB , so now we have
instead
Pe
Pr = . (456)
sin2 α
Should we include this in astrophysical situations?
In fact not usually. This is because in general the gammas of the particles
are large compared to the gamma of the bulk flow, and are often more or less
randomly moving. When the particle moves toward the observer, the power
is increased by sin2 α but the particle emission only arrives to the observer
per period for a time T /sin2 α. fig 6.13 fig S.8

We can see this as follows. If the time to travel the bottle half way is t/2
as measured by the observer with proper clocks, then the correction for the
Doppler effect is
t t
(1 − v|| cosα/c) = (1 − vcos2 α/c) ' (t/2)sin2 α. (457)
2 2
Suppose we arrange the bottle such that t/2 = π/ωB . Then the cancellation
with the factor for the observed power follows: That is

(Pe /sin2 α)(2πsin2 α/ωB )

Pr,ave ∼ = Pe (458)
2π/ωB

Synchrotron self-absorption
Photons emitted by synchrotron can be re-absorbed. Or such photons can
stimulate more emission in some component of phase space where photons
already exist. (Stimulated emission).
There is one state for each element of phase space h3 . We break up the
continuous volume in to discrete elements of size h3 and consider transitions
between states.
We must sum over all upper and lower states. We can use the Einstein
coefficient formalism. We then have for the absorption coefficient, assuming

84
a tangled magnetic field and isotropic particle distribution (thus assuming
isotropy)
XX
αν = (hν/4π) [n(E1 )B12 − n(E2 )B21 ]φ21 (ν). (459)
E1 E2

The φ(ν) is approximately a delta function restricting sum to states differing

by hν = E2 − E1 .
We want to relate the Einstein coefficients to (446), the microscopic com-
ponents of emission. We write

P (ν, E2 ) = 2πP (ω) (460)

where ν is the frequency and E2 is the energy of the radiating electron.

In terms of the Einstein coefficients and using Einstein relations:

A21 φ21 (ν) = (2hν 3 /c2 )hν

X X
P (ν, E2 ) = hν B21 φ21 (ν), (461)
E1 E1

where Einstein relations have been used. Note there are no statistical weights.
Thus for stimulated emission we have
hν X X c2 X
− n(E2 )B21 φ21 = − n(E2 )P (ν, E2 ) (462)
4π E1 E2 8πhν 3 E2

and for true absorption

hν X X c2 X
n(E1 )B12 φ21 = n(E2 − hν)P (ν, E2 ), (463)
4π E1 E2 8πhν 3 E2

where we use B21 = B12 . Note that the argument of n corresponds to those
particles which have radiated. Thus, plug into (459) to obtain

c2 X
αν = [n(E2 − hν) − n(E2 )]P (ν, E2 ). (464)
8πhν 3 E2

Now consider an isotropic electron distribution f (p) such that f (p)d3 p is

the number density of electrons within momentum range d3 p.
The number of quantum states per volume range d3 p is h−3 ξd3 p, where
ξ = 2 is the statistical weight for electrons. Electron density per quantum

85
state is then f (p)d3 p/h−3 ξd3 p = (h3 /ξ)f (p). Thus we can make the following
exchanges: number density is then

n(E2 ) → (h3 /ξ)f (p2 ), (465)

and sum over quantum states is then

Z
→ (ξ/h3 ) d3 p2 .
X
(466)
2

We then have
c2 Z 3
αν = d p2 [f (p∗2 ) − f (p2 )]P (ν, E2 ), (467)
8πhν 3
where the ∗ labels the momentum for the E2 − hν energy. The reason for
keeping the 2 is to remind us that it is the 2 which labeled the initial energy
state of particles that radiate.
Note that for a thermal distribution of particles

f (p) = Ce−E(p)/kT . (468)

We note that

f (p∗2 ) − f (p2 ) = Ce−(E2 −hν)/kT − Ce−yE2 /kT = f (p2 )(ehν/kT − 1) (469)

so
c2 hν/kT
Z
αν = (e − 1) d3 p2 f (p2 )P (ν, E2 ). (470)
8πhν 3
The integral is just the total power per frequency for isotropic thermal emis-
sion here, which is 4πjν . Thus
jν
αν,ther = (471)
Bν (T )
which is Kirchoff’s law as expected.
Using energy rather than momentum for the distribution function, we
have, assuming E = pc for extremely relativistic particles,

N (E)dE = 4πp2 f (p)dp (472)

so
c2 Z
!
N (E − hν) N (E)
αν = − P (ν, E)E 2 dE. (473)
8πhν 3 (E − hν)2 E2

86
Now assume hν << E as required for classical e&m. Then from the definition
of the derivative (e.g. expanding for small hν)

c2 Z
" #
2 ∂ N (E)
αν = − P (ν, E)E dE. (474)
8πν 2 ∂E E2

Then for thermal distribution of ultra-relativistic particles

N (E) = CE 2 e−E/kT . (475)

This leads to
c2 Z
jν c2
αν,ther = N (E)P (ν, E)dE = , (476)
8πν 2 kT 2ν 2 kT
which is Kirchoff’s law for the Rayleigh-Jeans regime.
For a power-law distribution, we have
!
d N (E) (p + 2)N (E)
−E 2
= (p + 2)CE −(p+1) = (477)
dE E2 E
so
(p + 2)c2 Z N (E)
αν = 2
dEP (ν, E) (478)
8πν E
Recall that in (450) we ignored ignored the hidden E dependence in the
bounds of of the integral. We do the same here. This allows us to avoid the
frequency dependence in P (ν, E) (the emission for a single particle, peaked
near ω/ωc ) since we will just leave it in the integrand as in (444). After
changing variables from E to γ to ωc /ω, the frequency dependence in (478)
then becomes
αν ∝ ν −(p+1)/2+1/2−4/2 = ν −(p+4)/2 , (479)
where we get the 4/2 from the ν −2 , the ν 1/2 from the measure, and the
ν −(p+1)/2 from changing N (E)/E to γ −(p+1) in (477).
The source function (using (451) for P (ν))

jν P (ν)
Sν = = ∝ ν −(p−1)/2+(p+4)/2 = ν 5/2 , (480)
αν 4παν
where here P (ν) is now the emission spectrum for a power-law distribution
calculated earlier.

87
 Can derive this result also like this:

Sν = jν /αν ∝ ν 2 Ē (481)

where Ē is some mean particle energy. This approximate relation comes

fromR noting that P (ν) = dEN (E)P (ν, E) and then noting that αν ∝
R

ν −2 dEN (E)P (ν, E)/E and then taking ratio to get Sν . Then note/recall
νc ∝ γ 2 ∝ Ē 2 . Thus each frequency in the spectrum corresponds to a νc
for a particle of a specific energy. Thus νc = ν when applied to the source
function which is integrated over all particles. Thus Sν ∝ ν 5/2 .
For optically thin synchrotron emission

Iν ∝ jν ∝ Pν ∝ ν −(p−1)/2 , (482)

since for optically thin plasmas, intensity is proportional to the emission

function. For optically thick plasmas, intensity is proportional to the source
function
Iν ∝ Sν ∝ ν 5/2 . (483)
fig6.12

This optically thin regime occurs at high frequencies, and the optically
thick regime occurs at low frequencies.
The absorption produces a cutoff in the frequency. Note that the electrons
are non-thermally distributed so even though the system is optically thick
to synchrotron emission in the ν 5/2 regime, the system is not in thermal
equilibrium, and thus shape is different from Rayleigh jeans B-Body which
is proportional to ν 2 .
No synchrotron masers in a vacuum
The absorption coefficient is positive for an arbitrary distribution of par-
ticle energies N (E). This means that if there were stimulated emission re-
sulting from making N (E) at a certain energy E0 larger so that emission
from E0 to E0 − hν was a maser ¡, then someplace else in the distribution,

88
there would be more positive absorption that would more than compensate.
This means that αν is positive since it is integrated over energy. Can show
this is also true for separate polarization states of synchrotron emission.

89
 Lecture 16
Compton Scattering
For low photon energies, the scattering of radiation from free charges
reduces to the classical case of Thomson scattering. For that case

i = f , (484)

1
dσT /dΩ = r02 (1 + cos2 θ) (485)
2
and
σT = 8πr02 /3 = 6.6 × 10−25 cm, (486)
where i and f are the incident and scattered photon energy, dσT /dΩ is the
differential cross section for scattering into Ω and σT is the Thomson cross
section, and θ is the angle between the incident and scattered direction. This
is “elastic scattering.”
In reality the scattering is not elastic because the charge recoils. fig 7.1

The initial and final 4-momenta of the photon are Piγ = (i /c)(1, ni ) and
Pf γ = (f /c)(1, nf ) respectively.
For the electron Pie = (mc, 0) and Pf e = (E/c, p) respectively. Con-
servation of energy and momentum can be written in terms of the energy
momentum 4-vectors for the electron and photons:

Pie + Piγ = Pf e + Pf γ , (487)

This leads to the two equations

mc2 + i = E + f (488)

and
i ni + 0 = f nf + pc. (489)
Rearranging we have

(mc2 + i − f )2 = E 2 = m2 c4 + p2 c2 (490)

90
and
(i ni − f nf )2 = p2 c2 . (491)
where I used E 2 = m2 c4 + p2 c2 . Then solving to eliminate p, we obtain
i
f = i . (492)
1+ mc2
(1 − cosθ)

If we write i = hν = hc/λ and similarly for f we have,

λf − λ = λc (1 − cosθ), (493)
h
where λc = mc = 0.024 angstroms for electrons is the Compton wavelength.
When λ >> λc or hν << mc2 , the scattering is elastic in the rest frame of
the electron.
2
When quantum effects are important, ie. i > ∼ mc , the cross section
becomes
r02 2f i
!
dσ f 2
= + − sin θ . (494)
dΩ 2 2i f i
For f ∼ i Eq. this reduces to the classical expression. Note that since
f ≤ i , the quantum regime produces a lower overall cross section than the
Thomson regime. Think about the role of the uncertainty principle here.
The total cross section in the non-relativistic regime (hν << mc2 ) is

σ ∼ σT (1 − 2x + ...), (495)

where x = hν/mc2 , and in the relativistic regime x >> 1,

3 1
σ ∼ σT x−1 (ln2x + ). (496)
8 2
Inverse Compton energy transfer
When the electron has more kinetic energy in the lab frame than the
photon energy, there can be energy transfer from the electron to the photon
during the scattering. This is inverse Compton scattering. The scattering in
the electron rest frame (K 0 ) and the lab frame (K), are shown below. fig 7.2

91
 In frame K’ (rest frame of the electron) we have

0i
0f = 0i
. (497)
1+ mc2
(1 − cosΘ0 )

We also have from the Doppler formulas,

0i = i γ(1 − βcosθi ) (498)

and
f = 0f γ(1 + βcosθf0 ). (499)
Since electron scattering in the rest frame of the electron is front back sym-
metric, a typical angle for scattering is θi ∼ π/2 or θf0 ∼ π/2. Thus we have
roughly, from the previous 3 equations

0f = 0i (500)

0i ∼ i γ (501)
and
f ∼ γ0f , (502)
so that f = γ 2 i . Thus there is a gain by a factor of γ 2 in the energy
from Compton scattering where γ is the Lorentz factor of the electron. We
assumed that γ << mc2 in the rest frame of the electron. If 0i = i γ is
too large, then we see from (497) that 0f << 0i so that the processes is less
efficient. Also, when 0i = i γ is too large then the cross section is reduced,
which lowers the scattering probability again making the process less efficient.
Note also that γ 2 is the proportional to the energy of the electron squared.
Inverse Compton power for isotropic photon distribution (single
scattering)
To get the emitted power from an isotropic distribution of photons scat-
tering off of an isotropic distribution of electrons incurring single scattering
per photon, we must average (497-499). To do this let q be density of pho-
tons of energy in range range d. Let g(p) be the phase space distribution
function. Then
qdi = g(p)d3 p. (503)
We have that γd3 p = d3 p0 as we derived earlier when discussing the phase
space invariants in relativity. Thus d3 p transforms as energy under Lorentz

92
transformations. Since g(p) is an invariant, we have also that qdi /i is an
invariant that is
qdi /i = q 0 d0i /0i . (504)
The total power scattered in the electron’s rest frame is

dEf0 Z 0 0
02 q d
= cσ T f . (505)
dt0 0f

We assume that the energy change of photon in rest frame is small compared
to that in lab frame: γ 2 − 1 >> i /mc2 , so that so 0f = 0i . We also have

dEf0 dEf
= (506)
dt0 dt
by Lorentz invariance for isotropic emission. Thus
0 0
dEf Z
02 q di
Z
qdi
= cσT i 0 = cσT 02
i , (507)
dt i i

using the 0f = 0i assumption. Using the Doppler formula

0i = i γ(1 − βcosθ) (508)

we have
dEf Z
= cσT γ (1 − βcosθ)2 qi di .
2
(509)
dt
Thus all quantities are now written in the K frame as desired. For isotropic
distribution of photons, using hcos2 θi = 1/3 we have
1
h(1 − βcosθ)2 i = 1 + β 2 . (510)
3
Thus
dEf
= cσT γ 2 Uph (1 + β 2 /3), (511)
dt
R
where Uph = i q(i )di which is the energy density of incident photons.
Now the norm of the rate of decrease of the initial photon energy is
 
 dE  Z
i

  = |cσT i qdi | = σT c2 Uph , (512)
 dt 

93
and so the energy actually radiated by the electron and converted into radi-
ation is the difference between the energy scattered and the energy lost by
the incoming photons. That is, it is the energy out minus the energy in. We
have  
dErad dEf  dEi  1 2
 
2
= −  = cσT Uph γ (1 + β ) − 1 . (513)
dt dt  dt  3
Using γ 2 − 1 = γ 2 β 2 we have
dErad 4
Pc = = σT cγ 2 β 2 Uph . (514)
dt 3
(When the energy transfer in the electron rest frame is not ignored, there is
a correction term. (Blumenthal & Gould (1970)).)
Note that the synchrotron power we calculated for a single electron was
4
σT cγ 2 β 2 UB . (515)
3
Thus the ratio of Compton to Synchrotron power is

Pc /Psyn = Uph /UB . (516)

The result also is true for arbitrarily small γ (i.e. the non-relativistic limit,
as long as γ << mc2 . This gives an important rough and ready tool to
determine which of the two emission processes are more important.

94
 LECTURE 17
Compton power for isotropic distributions of photons and relativis-
tic electrons (single scattering)
Let N (γ) be the number of electrons per volume between γ and γ + dγ.
Then Z
Ptot = Pc N (γ)dγ erg s−1 cm−3 (517)

Choose a power law so that N (γ) = Cγ −p for a range of γ. Then

3−p
4 γ 3−p − γmin
Ptot = σT cUph C max erg s−1 cm−3 (518)
3 3−p
for β ∼ 1.
We can also compute the total power from a distribution of non-relativistic
electrons. For γ ∼ 1 we have hβ 2 i = 3kTe /mc2 so that
4kTe
Ptot = cσT ne Uph mma (519)
mc2
Inverse Compton Spectra (isotropic distribution of photons and
electrons)
Spectrum depends on both incident distribution of photons and energy
distribution of the electrons. But we first calculate for scattering of mono-
energetic photons off of mono-energetic electrons assuming isotropy and sin-
gle scattering. This means we just have to later average over energy dis-
tributions of photons and electrons to get overall spectrum spectrum (that
averaging over distribution functions will be done in next section.).
Assume again the Thomson scattering regime 0i ∼ γi << mc2 . Let us
assume that the scattering in the rest frame is isotropic. We thus assume
dσ 0 σT 2
0
= = r02 . (520)
dΩ 4π 3
Following the text use In here to mean photon number intensity rather than
energy intensity. Then In ddAdΩdt is the number of photons passing through
area dA in time dt within solid angle dΩ and energy d. This intensity is just
the specific intensity divided by the energy (or Iν /h). The analagous applies
for source and emission functions.
Consider now isotropic incident photon field as mono-energetic

In (i ) = F0 δ(i − 0 ), (521)

95
where F0 is the number of photons per unit area, per time, per steradian.
(The flux per solid angle). Note that δ function always has units of 1 over
the argument. We scatter these photons off of a beam of electrons of number
density N and energy γmc2 moving along the x-axis.
fig 7.2

The incident intensity in the rest frame, from Lorentz transformation

properties of In , is
!2
0i
In0 (0i , µ0 ) = F0 δ(i − 0 ). (522)
0

Then using the Doppler formula  = γ0 (1 + βµ0 ) we have

!2 !2
0i γ0i − 0
0i
!
F0
In0 (0i , µ0 )
= F0 δ (γ0i (1 0
+ βµ ) − 0 ) = δ + µ0 .
0 γ0i β γ0i β
0
(523)
Here µ0 is the cosine of the angle between the electron travel direction (x-axis)
and the incident photon.
The emission function in the rest frame for isotropic distribution of pho-
tons is
1Z 1 0 0 0
jn0 (0f ) = N 0 σT I ( , µ )dµ0 (524)
2 −1 n f
which here is the number of emitted photons per volume per steradian per
time, and N 0 σT is the scattering coefficient. Note N = γN 0 . We replaced 0i
by 0f by using the rest frame Thomson scattering approximation. Then

N 0 σT 0f F0
jn0 (0f ) = (525)
220 γβ
for the range
0 0
< 0f < , (526)
γ(1 + β) γ(1 − β)

96
and zero otherwise. This range corresponds to the range for which −1 <
µ0 < 1 which has to be the case since µ0 is a cosine of the angle between
incident photon and electron velocity.
We really want jn (f , µf ), the value in the lab frame. Recalling that jν /ν 2
is an invariant, we thus have jn / is an invariant. Using this we have

f N σT f F0
jn (f , µf ) = jn0 (0f ) 0
= (527)
f 220 γ 2 β

when
0 0
< f < 2 , (528)
γ 2 (1 + β)(1 − βµf ) γ (1 − β)(1 − βµf )
using the Doppler shift formula for 0f , and N = γN 0 . Note also that for large
β and µf = 1 the inequality range can be written

0 < f < 40 γ 2 , (529)

while for large β and µf = −1 the inequality range can be written

0
< f < 0 . (530)
4γ 2
For an isotropic distribution of electrons we average over the angle be-
tween the electron and emitted photon:
1Z 1
jn (f ) = jn (f , µf )dµf . (531)
2 −1
Note that the limits in (531) can be written in terms of µf , that is the
non-zero region of jn occurs when
! !
0 0
β −1 1 − (1 + β) < µf < β −1 1 − (1 − β) . (532)
f f

For f /0 < (1 − β)/(1 + β) or for f /0 > (1 + β)/(1 − β), µf does not fall
in range −1 < µ < 1 so the integral must vanish. The finite contribution
comes from within the integral limits given by
!
1 0
−1 < µf < 1 − (1 − β) (533)
β f

97
for (1 − β)/(1 + β) < f /0 < 1 and
!
1 0
1 − (1 + β) < µf < 1 (534)
β f

for 1 < f /0 < (1 + β)/(1 − β). Since jn (f , µf ) is isotropic, the integrals
just amount to multiplication of the integrand of (531) by each of the bounds
and subtracting. When the dust clears, using (527), we have

N σ T F0
jn (f ) = Q (535)
40 γ 2 β 2
where
f
Q = (β − 1) + (β + 1) (536)
0
for (1 − β)/(1 + β) < f /0 < 1 and
f
Q= (β − 1) + (β + 1) (537)
0

for 1 < f /0 < (1 + β)/(1 − β). (Which is 1 < f /0 < 4γ 2 for large γ).
The integrals satisfy:
Z ∞
jn (f )df = N σT F0 (538)
0

and Z ∞
4
jn (f )(f − 0 )df = γ 2 β 2 N cσT (0 F0 /c). (539)
0 3
The first means that number of photons are conserved since it is the number
of photons scattering per volume per solid angle per unit time. The second
is just the average gain in energy per unit time measured as radiation per
electron where 0 F0 /c is the average incident photon energy density. That is
it is the POWER emitted per electron.
Notice that in contrast to Bremsstrahlung, this equation is only propor-
tional to the number density of electrons rather than the number density
squared.
The quantity Q/(4γ 2 β 2 ) = Q/[4(γ 2 − 1)] = 0 j(f )/N σT F0 is plotted
below. fig 7.3a

98
 One can see the shift with larger β as the energy is gained, highlighting
how the average energy of the scattered photons is increased with larger γ.
At very small γ there should be a peak around the initial photon energy
with no gain or loss. As the γ is increased, there will be some loss as well as
gain, because the photon distribution is isotropic and catch-up interactions
deplete photon energy. But with large γ the beaming means electron sees
most photons as head on, and the overall energy gain is produced.
When β ∼ 1, we are in the second regime of Q , Eq. (537), which actually
now corresponds to (529). Define x = f /4γ 2 0 , and then from (535),

1 3
jn (f ) = 2
N F0 σT (1 − x) = 2 N F0 σT f (x) (540)
2γ 0 4γ 0

where we have used f /0 = 4γ 2 x and 4γ 2 x(β − 1) + (β + 1) ∼ 2x + 2 for

β ∼ 1 (and thus γ >> 1). The result (540) applies only in range 0 < x < 1
because the limit below (537) can be written as 0 < x < 1 for large γ.
We have assumed everything to be isotropic in the electron rest frame.
Small corrections result when this assumption is not made. See fig 7.3b.

99
 LECTURE 18
Spectrum for arbitrary isotropic photon distribution (not nec-
essarily mono-energetic) off of a power law, isotopic distribution of
electrons
The q() = (4π/c)In () = Fn /c. We will use this below. Note also that
1Z 1
jn (f ) = N σT In (f , µ)dµ = N σT In (541)
2 −1
for isotropic In .
We have for the total scattered power per volume per energy
dE
= 4πf jn (f ) (542)
dV df dt

Then for N (γ) = Cγ −p , we have

dE 1 Z Z
= cσT C di q(i )(f /i ) dγγ −(p+2) (1 − f /4γ 2 i + ...) (543)
dV df dt 2

Change from γ to x = f /4γ 2 i . That implies dγ = −(21/2 /4)(f /i )1/2 x−3/2 dx
So we have:
Z x2
dE 1−p
Z p−1
∝ f 2
di i q(i )
2
(1 − x + ..)dx. (544)
dV df dt x1

with x1 = f /4γ12 i and x2 = f /4γ22 i . If γ2 >> γ1 and q(i ) peaks strongly

at ¯i then second integral is just a number that factors out (like for synch.).
We then have
dE 1−p
Z p−1
∝ f 2
di i 2 q(i ). (545)
dV df dt
Note this is restricted to the regime 4γ12 ¯i << f << 4γ22 ¯i .
Note that the power law is the same as for synchrotron!.
For an initial photon black body distribution, we have

8π2i 1
q(i ) = 3 3  /kT
(546)
h c ei −1
and
dE p+5 1−p
∝ (kT ) 2 f 2 . (547)
dV df dt

100
Note the strong dependence on temperature. Also note that the temperature
is that of the incident radiation (e.g. like a star radiating incident photons
into a corona of highly accelerated electrons).
Energy transfer for repeated scattering, Y parameter
When does scattering affect total photon energy? again assume Thomson
limit in rest frame of electron. Consider limit γi << mc2
Define y- parameter such then when y > 1, scattering changes spectrum
and energy wheras when y < 1, scattering does not affect energy but can still
change spectrum. The appropriate definition is the typical fractional energy
change per scattering times the mean number of scatterings:
∆i
y= × (mean number of scatterings). (548)
i
Consider a thermal electron distribution in the non-relativistic limit. Recall
that
0
!

0f ∼ 0i 1 − i
(1 − CosΘ) , (549)
mc2
when 0i << mc2 . If we average over Θ (the angle between incident and
scattered photons) we get
∆0 0i
= − , (550)
0i mc2
where this is a small term. In the lab frame, to lowest order in the two small
parameters /mc2 . and kT /mc2 we must have
∆ i
= − 2 + αkT /mc2 . (551)
i mc
We will determine α by first considering the case in which the electrons
and photons are in thermal equilibrium. but interact only through scattering.
Ignore stimulated emission. Photons thus have a Bose-Einstein distribution
in which particle number cannot change. We are in the classical regime,
where we ignore the stimulated emission. (This means that e−µ >> 1), This
gives
n = K2 e−/kT , (552)
where K includes eµ among the other constants. so that
R dN
 d d
hi = R dN = 3kT (553)
d
d

101
and
h2 i = 12(kT )2 . (554)
In this limit, no net energy can be exchanged between photons and electrons
so we have
αkT h2 i 3kT
h∆i = 0 = 2
hi − 2
= (α − 4)kT, (555)
mc mc mc2
thus α = 4. Thus for non-relativistic electrons in thermal equilibrium,

(∆)N R = (4kT − ). (556)
mc2
This shows that when  > 4kT , the energy is transferred from photons to
electrons. When  < 4kT energy is transferred from electrons to photons,
inverse Compton scattering.
Now consider the limit γ >> 1. Then ignoring the energy transfer in the
electron rest frame we recall that

0i = i γ(1 − βcosθi ) (557)

and
f = 0f γ(1 + βcosθf0 ). (558)
These give
f = i γ 2 (1 − βcosθi )(1 + βcosθf0 ). (559)
The average over angles gives

f − i = i (γ 2 (1 − β 2 h(cosθi )(cosθf0 )i) − 1). (560)

But since the average is over both angles, only when they are equal does
the average contribute and only when the cosines are equal and opposite is
energy gained. The first integral then gives a delta function and the second
then is an average over −(cosθi )2 . Thus we have, using γ 2 − 1 = γ 2 β 2

f − i = i (γ 2 (1 + β 2 /3) − 1) = . (561)

This can be written approximately, for γ > 1.

4
(∆)R ∼ γ 2 i (562)
3

102
Then for thermal distribution of relativistic electrons E ∝ γ and thus using
(554)
!2
2 hE 2 i kT
hγ i = = 12 . (563)
(mc2 )2 mc2
Thus !2
4 kT
(∆)R ∼ γ 2  = 16 . (564)
3 mc2
That is the first factor for the y parameter.
For the second factor (recall chap 1 of book) we recall that the mean
number of scatterings is just the size of the medium over the mean free path
all squared: Ns = (R/l)2 = τ 2 for τ >> 1, and Ns ∼ 1 − e−τ ∼ τ for τ << 1.
Thus we take
2
Ns = M ax(τes , τes ) (565)
and τes = ρκes R, where κes is the electron scattering opacity and is 0.4cm2 /g
for ionized hydrogen.
Thus for the non-relativistic regime we have for the y parameter using
(556) and (565) we have
! !
4kT 2 4kT 2
y= −  M ax(τes , τes ) ∼ M ax(τes , τes ), (566)
mc2 mc2

where the latter similarity follows for  << 4kT .

For the relativistic regime we have using (564) and (565)
!2
kT 2
y = 16 M ax(τes , τes ). (567)
mc2

In the non-relativistic regime, it can be shown that the gain in energy

of the photon as is scatters off non-relativistic electrons satisfies f /i = ey .
(problem 7.1).
In general, the Compton y parameter is frequency dependent. Then τes =
τes (ν). Then since absorption takes away photons before they can scatter
indefinitely, we determine τes as the value such that τ∗ (ν) = 1, where the
latter is the effective optical depth. Define l∗ (ν) as the effective path the
photon travels before absorption. When l∗ < R we have

τes (ν) = ρκes l∗ (ν), (568)

103
Since 1/2
αν + σν

l∗ = lN 1/2
∼l ∼ [αν (αν + σν )]−1/2 (569)
αν
since l = (αν +σν )−1 . If we then note that κa (ν) = ρ−1 αν and κes (ν) = ρ−1 σν
we have
!1/2
−1/2 κ2es
τes (ν) = ρκes l∗ (ν) = ρκes [αν (ν)(αν + σν )] = . (570)
κ2a + κes κa

Note that κa and κes are the absorption and scattering opacities.

104
 LECTURE 19
Inverse Compton Power for repeated scatterings: relativistic
electrons small optical depth
Interesting result: we will find that a photon power law spectrum can arise
even when electrons are not a power law, if there are repeated scatterings.
Define A as the amplification per scattering
4
A = f / = hγ 2 i = 16(kT /mc2 )2 , (571)
3
for a thermal distribution of electrons. Consider initial photon energy distri-
bution with mean photon energy i satisfying

i << mc2 hγ 2 i−1/2 , (572)

and number intensity In (i ). Then after k scatterings, the energy of a mean
initial photon is
k = i Ak . (573)
If the medium has small scattering optical depth then the probability of
k
a photon making k scatterings is pk (τes ) ∼ τes = (1 − e−τes )k . This is because
probability of a photon propagating through optical depth τes before single
scattering is e−τes .
k
The intensity is also proportional to τes quantity since the width of the
probability distribution (histogram) is of order the peak frequency. Thus
−s
k

k
In (k ) ∼ In (i )τes ∼ I(i ) (574)
i
where s = −ln(τes )/ln(A). To prove, take lnτ of (574) and get back (573).
This holds only when
k ≤ hγ 2 i1/2 mc2 , (575)
to ensure last scattering is still in Thomson type limit in rest frame. The
final photons are those which emerge therefore with energy of order kT for
thermal distribution of relativistic electrons. The total Compton power is
Z A1/2 mc2 "Z #
A1/2 mc2 /i
−s
P ∝ In (k )dk = In (i )i x dx , (576)
1

where the upper bound is ∼ kT as required for the upper limit on the allowed
energy range.

105
 You can see that the initial energy is amplified by the factor in brackets,
which is important for s < 1. When does this happen? When
−lnτes < lnA (577)
that is, when Aτes ≥ 1.
This is equivalent to the condition that yR > 1 for small τes , since A is
given by (571). Compare to (567). Thus energy amplifciation important in
this regime.
Repeated Scatterings by non-relativistic electrons
In general we need to solve an explicit equation for evolution of the pho-
ton distribution function due to repeated non-relativistic inverse Compton
scatterings (Kompaneets equation). For frequencies with y << 1 and y >> 1
we can get away with approximations. We we consider these regimes here.
We are interested in thermal media with both emission and absorption
take place by bremsstrahlung, and we want to see the effect of Compton
scattering in such a medium.
Note that bremsstrahlung absorption takes place more strongly at lower
frequencies.
There are three key frequencies of interest. First define the frequency ν0
such that the scattering and the absorption opacities are equal. We recall
that αν = ρκν,a , σ(ν) = ρκes (ν), and κes = σT /mp = 0.4cm2 g, we have
κes = κf f (ν) (578)
ανf f = 3.7 × 108 T −1/2 Z 2 ni ne ν −3 (1 − e−hν/kT )ḡf f . (579)
That is then
x30
−x
∼ 4 × 1025 T −7/2 ρḡf f (x0 ), (580)
1−e 0

or
x0 = 6 × 1012 T −7/4 ρ1/2 [ḡf f (x0 )]1/2 , (581)
for x0 << 1. where ḡf f (x) = 3π −1/2 ln(2.25/x). For x = hν/kT < x0
scattering is not important. We assume x0 << 1 below.
The second frequency νt is that for which the medium becomes “effectively
thin.” Using
(L/l∗ )2 = 1 = ρ2 κa (ν)(κa (ν) + κes )L2 . (582)
2
Then dividing by τes = ρ2 L2 κ2es , and ignoring the term (κa /κes )2 we have
2
κes = κa (ν)τes . (583)

106
This gives the condition

x3t
∼ 4 × 1025 T −7/2 ρḡf f (xt )τes
2
. (584)
1 − e−xt
The τes is the total optical depth to electron scattering. For values x >
xt the absorption is unimportant. For x0 < x < xt both scattering and
absorption are important.
Third, define the frequency νcoh where inverse Compton can be important.
This is defined such that y(νcoh ) = 1. That is for ν > νcoh , inverse Compton
is important. But this frequency is only of interest if the y parameter for the
full medium is greater than unity, implying that scattering is important for
some photons at least.
In this case we then have
4kT 2
1= τ (ν) (585)
mc2 es
and using
!1/2
κes /κa (ν)
τes = (586)
1 + κa (ν)/κes
gives
mc2
κes = κf f (νcoh ). (587)
4kT
Then again using κf f = αf f /ρ we have

xcoh = 2.4 × 1017 ρ1/2 T −9/4 [ḡf f (xcoh )]1/2 . (588)

Note that xcoh must be less than xt because for x > xt the medium is
effectively thin. Inverse Compton cannot be significantly important when
the medium is effectively thin.
Now let us use these frequencies to get info about various regimes.
Modified blackbody spectral regime y << 1
For y << 1 there is only coherent (elastic) scattering. You derived in
problem 1.10 that the formula for the intensity in a scattering and absorbing
medium is
2Bν
Iν = q . (589)
1 + (1 + κes /κf f )
For x << x0 , this reduces to blackbody spectrum.

107
 For x >> x0 this reduces to modified blackbody spectrum, that is
1/2
κf f

1/2
Iνmb = 2Bν = 8.4 × 10−4 T 5/4 ρ1/2 ḡf f x3/2 e−x/2 (ex − 1)−1/2 . (590)
κes

Note that for x0 << x << 1, Iνmb ∝ ν not Iνrj ∝ ν 2 as in Rayleigh Jeans.
(Refer to (581)). Less steep than RJ regime.
The total flux is

F mb = σT 4 (κR /κes )1/2 ∼ 2.3 × 107 T 9/4 ρ1/2 erg/s cm−2 , (591)

here using the Rosseland mean absorption coefficient, (defined as inverse

of expectation value of α1f f , weighted by ∂Bν /∂T .) where αR = 1.7 ×
10−25 T −7/2 Z 2 ne ni ḡR The flux is reduced from the bbody flux. This is be-
cause if you think of absorption and emission in equilibrium as “trapping”
radiation density, then the radiation density is higher where there is more
trapping.
** For finite media, we have to consider xt . When xt < x0 , emission is
blackbody for x < xt and optically thin bremsstrahlung for x > xt where
scattering is not important.
For x0 < xt < 1, the emission is (589) for x < xt , and is optically thin
brems. for x > xt .
(all this is because scattering is more important than abs for x > x0 , but
system is effectively thin for x > xt .
For xt > 1 (589) is good for all x.
Wien Spectra y >> 1
Inverse Compton may be important depending on whether xcoh << 1 or
xcoh >> 1 . In the latter case IC can be neglected since the majority of
photons x < xcoh undergo coherent scattering (previous discussion applies!)
Here consider xcoh << 1. In the limit x0 << 1, then we have

xcoh = (mc2 /4kT )1/2 x0 . (592)

For x < xcoh Eq. (589) is OK. For x > xcoh a saturated inverse Compton
spectra will result. This is a Wien spectrum, maximal scattering produces a
kind of thermal equilibrium. We have

2hν 3 2hν 3 −f −hν/kT

IνW = nγ = e e , (593)
c2 c2

108
where e−f is the rate at which photons are input into the system. Total flux
is then Z
12πe−f k 3 T 4
F W = π IνW dν = . (594)
c2 h3
The mean photon energy is 3kT .
The intensity is shown below fig 7.4

As you might expect, the flux is given by

F W ∼ RAf f , (595)

which is the bremsstrahlung initial photon energy density per unit time,
times an amplification factor A times a thickness over which the emission is
coming from. If xt << 1 the medium is effectively thin, so thickness is size
of medium. For xt >> 1, photons of x > xcoh amplify to kT . So we replace
R by R̄ where R̄ satisfies τ∗ (R̄, x = 1) = 1.

109
 LECTURE 20
Plasma Effects
Debye Length:
Consider plasma composed of charge Ze and electrons of charge e. For
charge neutrality we have
ne Zi = ni , (596)
summation implied. The ions attract the electrons. Focus on a single ion
and let us see the influence of the ambient plasma on shielding the potential
of this ion. The Coulomb law takes the form

∇2 φ = −4πρc (597)

where
ρc = Zi eδ(r) − ene eeφ/kT + eni Zi e−eZi φ/kT , (598)
assuming Ti = Te and ni and ne are the average mean densities, and assuming
Boltzmann spatial distributions. Let us assume for the moment that the
exponentials are small (justified later). The first terms in the expansion
cancel under the assumption of lowest order charge neutrality. Then we have
e2
∇2 φ = −4πZi eδ(r) + 4π (ne + ni Zi2 )φ = −4πZi eδ(r) + 4πφ/ld2 , (599)
kT
where ld = (kT /e2 (ne + ni Zi2 ))1/2 is the Debye length and is ∼ (kT /2e2 ne )1/2
for Zi = 1. Spherical symmetry means that the only derivative components
are r components, thus the laplacian becomes
1 d 2 dφ
r = −4πZi eδ(r) + 4πφ/ld2 . (600)
r2 dr dr
The appropriate solution which satisfies φ = eZi /r at r = 0 and 0 at r = ∞
is
Zi e −r/lD
φ= e . (601)
r
What this means is that the Coulomb potential is reduced at large r. The
statistical attraction of electrons and repulsion of ions shield the primary
charge.
Ions and electrons contribute equally to the shielding, because it depends
on the ratio of particle speed to mass through the temperature, and not
solely on the particle mass. The message is that for r > ld , the free charge
potential must be modified.

110
 Conditions for the above analysis require that there be enough particles
inside a volume ∼ ld3 for a statistical treatment, that is nd ld3 >> 1. (The
inverse of this quantity is called the plasma parameter.) Note that the mean
distance between electrons is ne−1/3 . Thus the mean electrostatic energy/kT
−2/3 −2
e2 n1/3
e /kT ∼ ne ld << 1, (602)
in the statistical regime, thus justifying the earlier expansion.
Plasma Frequency
The electron contribution to the inverse Debye length is
−2
ld,e ∼ 4πne e2 /kT = 4πne e2 /me ve2 = ωp,e
2
/ve2 , (603)
where ve is the electron thermal velocity and ωp,e = 4πn2e e2 /me is the electron
plasma frequency.
The meaning is as follows. Imagine displacing the electrons of a plasma
relative to the ions by an amount x. The resulting electric field is like that
of a two plate capacitor with surface charge density ±ene x, so that
E = 4πene x. (604)
This comes from integrating ∇ · E = 4πρc . The equation of motion for the
electron between the plates is then
ẍ = −eE/me = −e2 x4πne /me = −ωpe
2
x/me . (605)
Thus the electron plasma frequency is the natural oscillator frequency for
an electron from the charge separation. Ions oscillate at a much slower rate
and so one can consider them to be fixed. Numerically ωpe = 5.6 × 104 ne1/2 .
Larger frequency forcing or disturbances will set up faster oscillations. Oth-
erwise fast oscillations are set up to counteract the disturbances. No charge
separation occurs for disturbances on scales larger than the Debye length.
For such disturbances one is in the neutral fluid MHD regime.
Now lets come at this from a slightly different angle. Consider the lin-
earized equation of motion of an electron fluid (take, n, E, B, v equal to
zeroth order part + perturbation and consider only linear surviving terms).
We get
me n0 ∂t v1 = −en0 E1 (606)
From perturbed Maxwells equations we get
∇ × B1 = −(4π/c)no ev1 + (1/c)∂t E1 (607)

111
and
∇ × E1 = −(1/c)∂t B1 . (608)
Note that no n1 appears in these equations. Lets assume eiωt solutions for
the time dependence. Then for (606) we have

v1 = −(ie/ωme )E1 . (609)

Plugging this into (607) gives

∇ × B1 = −(i/c)ωE1 , (610)

where  = 1 − ωp2 /ω 2 . Note that in standard E+M, this represents the

dielectric constant since by replacing of the −iw in (610) with ∂t we would
recover
∇ × B1 = ∂t E1 . (611)
More on this in next section.
No E-M wave propagation at ω < ωp,e
Assuming wave propagation solutions to Maxwell’s equations of the form
ik·x−ωt
e . Maxwell’s equations for the plasma become

ik · E= 4πρc ; ik · B = 0. (612)

and
ik × E = iω/cB; ik × B = 4πj/c − iEω/c. (613)
We again ignore the magnetic force (non-relativistic) and then for the equa-
tion of motion we have
v = eE/iωme . (614)
The current density then satisfies

ine2
j= E = σE. (615)
ωme
Charge conservation gives

−iωρc + ik · j = 0, (616)

so that
ρc = k · j/ω = σk · E/ω. (617)

112
If we again define the dielectric constant
2
 = 1 − 4πσ/iω = 1 − ωp,e /ω 2 , (618)

we can then get “modified” vacuum equations:

ik · E = 0; ik · B = 0 (619)

and
ik × E = iωB/c; ik × B = −iωE/c. (620)
These are the source free Maxwell’s equations. We solved them earlier, now
we have a revised relation:

c2 k 2 = ω 2 = ω 2 [1 − (ωp,e /ω)2 ]. (621)

Thus we see that

kc = (ω 2 − ωp,e
2 1/2
) . (622)
Thus for ω < ωp,e the wave number is imaginary. This means that such waves
do not propagate. Plugging back into eik·x−ωt means that a wave propagating
in the direction x will decay on length scale ∼ 2πc/ωp , and instead reflects!
This allows probing the density of the ionosphere (90-150km above surface).
Send a pulse of radiation upward. Where n is larger than critical, ωp >
ω and the radiation is reflected. Electron density can be determined as a
function of height. (MHz range) when one uses also the time measured for
the signal to go up and come back down.
Dispersion Measure
For ω > ωp , the E-M radiation propagates with phase velocity

vph = ω/k = c/nr ≥ c (623)

where nr = 1/2 = (1 − ωp2 /ω 2 )1/2 is the index of refraction. The group

velocity travels at v ≤ c

vg = ∂ω/∂k = c(1 − ωp2 /ω 2 )1/2 . (624)

When electron density varies, and thus the index of refraction, radiation
travels along curved path. From Snell’s law, the curved trajectories satisfy

d(nr k̂)/dl = ∇nr , (625)

113
where l is the trajectory path length and k̂ is the ray direction.
One application of the spread in group velocities for waves of different
frequencies is for pulsars. Pulsars have pulses with a range of frequencies
per pulse. Interaction with ISM means that different frequency ranges will
travel at different group velocities and thus reach observer at different times.
If pulsar is a distance d away, then the time elapsed before the pulse reaches
earth is Z d
tp = ds/vg , (626)
0
where ds is line of sight distance differential. Interstellar space has kHz
plasma frequencies (Radio waves are in GHz). Thus assume ω >> ωp . Then
we have
2 −1/2 2
! !
1 ω p 1 1 ωp
vg−1 = 1− 2 ∼ 1+ . (627)
c ω c 2 ω2
Thus
d 1 Zd 2
tp ∼ + ω ds. (628)
c 2cω 0 p
The second term is the plasma correction. To get a convenient measured
quantity take the frequency derivative to obtain

4πe2
dtp /dω = − D, (629)
cmω 3
where Z d
D= ne ds (630)
0
is the dispersion measure. If we assume that the density is constant across
interstellar space between the observer and the pulsar, we get the pulsar
distance.

114
 Faraday Rotation
Consider a fixed magnetic field. This means the plasma is anisotropic
since the plasma properties will be different perpendicular to and parallel to
the magnetic field. We also can expect to find the cyclotron frequency enter-
ing the problem for non-relativistic plasmas. Ignoring the particle pressure
is called the cold plasma approximation, which we will do.
The dielectric constant is now a tensor, and also waves have different
polarizations so in general the coefficient multiplying the e(ik·x−ωt) in the
waves just discussed is not a constant.
Lets assume that the fixed field is stronger than the fields associated with
the propagating wave. Then
dv e
m = −eE − v × B (631)
dt c
Consider a circularly polarized wave
E(t) = Ee−iωt (ˆ1 ∓ iˆ2 ). (632)
Assume that the waves propagate along the uniform field, which we take to
be in the z-direction, ˆ3 . Using d/dt = −iω we then have
e
v(t) = E(t). (633)
im(ω ± ωB )
Comparing to (614) and (618), we have a new dielectric constant
ωp2
R,L = 1 − , (634)
ω(ω ± ωB )
where R, L corresponds to −, +. These travel with different velocities since
the dielectric constant enters the wave velocity. Thus the plane of a linearly
polarized wave, which is the superposition of two circularly polarized waves
rotates as the wave propagates. That is Faraday rotation. fig 8.1

For a constant wavenumber, the phase angle of rotation of the electric

field vector is k · d where d is the distance the wave travels. (This dot

115
product is = 2πd/λ where λ is the wavelength.) This can be computed more
generally, for a space-varying wavenumber, from
Z d
φR,L = kR,L ds, (635)
0

where
ω√
kR,L =  . (636)
c R,L
The plane polarized wave gets rotated by an angle ∆θ = 0.5(φR − φL ). If
we assume that ω >> ωp and ω >> ωB we have
ω2
" #
ω
kR,L = 1 − p2 (1 ∓ ωB /ω) . (637)
c 2ω
Thus
1Z d 1 Z d 2 −1
∆θ = (kR − kL )ds = (cω ) ωB ωp2 ds. (638)
2 0 2 0
Then substituting for ωp and ωB we have
2πe3 Z d 4 2
Z d
∆θ = 2 2 2 nB|| ds = 1.2 × 10 λ nB|| ds. (639)
mcω 0 0

By measuring the relative rotation at different frequencies we can get info

about the magnetic field.
Combining the pulsar dispersion measure and the Faraday rotation, we
can get an estimate of the mean magnetic field. Suppose for example that we
have a signal from a pulsar measured to have a time delay of dtp /dω = 10−5 s2
and a Faraday rotation that varies with frequency as d∆θ/dω = 2 × 10−4 sec.
Let us assume the measurements are made around frequency of 100MHz and
that we don’t know the distance of the pulsar. From this we can estimate
the mean magnetic field defined as
R
nB|| ds
hB|| i = R . (640)
nds
We simply take the derivative of (639) and divide by (629) to get
d∆θ/dω
∼ 2 × 107 B|| s−1 . (641)
dtp /dω
Substituting dtp /dω = 10−5 s2 and d∆θ/dω = 2 × 10−4 sec, we get B|| ∼
10−6 G. Which is typical for the ISM. Note that this is independent of the
frequency of emission.

116
 LECTURE 21
Effects of plasma on high energy radiation
The radiation from all of the processes that we have discussed is subject
to the influence of the plasma, e.g. little radiation will propagate below
the plasma frequency. Faraday polarization will also set up a depolarization
screen.
There are some effects on the high energy processes themselves which
change the radiative process fundamentally. Need to then consider the
induced motions and subsequent emission from particles in the ambient
medium responding to the radiation. We consider a dielectric constant of
the plasma which is independent of frequency, which is a cheat, but will
capture basic ideas.
Maxwell’s equations in the presence of a constant dielectric become

∇ · E = 4πρ/; ∇ · B = 0 (642)
1 ∂B 4πj  ∂E
∇×E= ; ∇×B= + (643)
c ∂t c c ∂t
These equations result from Maxwell’s equations with the substitution
√
E → E ; B → B, (644)

√
A → A; φ → φ (645)
and
e c
e→ √ ; c→ √ . (646)
 
Recall that these are solved by the Liénard-Wiechart potentials
√
√ −e/ 
φ = (647)
κ(tret )R(tret )
where √
κ̃ = 1 − n · β~  (648)
and
eu(tret )
A=− , (649)
cκ̃(tret )R(tret )
so that
√ −e n

˙

Erad (r, t) = 3
× {(n − β̃) × β̃} (650)
c κ̃ R

117
and √
Brad = [n × Erad ] (651)
Cherenkov effect
Charge moving uniformly in a vacuum cannot radiate: No acceleration, no
force, no energy change. Also true for a charge moving uniformly in dielectric
medium if the phase velocity exceeds particle velocity. This is because the
modified L-W potentials still then has only the term proportional to 1/R2 .
However, if medium has index of refraction larger than 1, then the particle
speed can be larger than the phase
√ velocity. Then κ̃ = 1−βnr cosθ can vanish
when cosθ = 1/nr β. (nr ≡ ). The potentials can become quite large at
places where this holds, and now there can be effective radiation from the
1/R2 term (consider Poynting flux argument, not necessarily zero at large
distances in this case).
Note also that for v > c/nr , the potentials can be determined by two
retarded positions. fig 8.2

Note the cone, for this regime. The resulting radiation is confined within
the cone of angle θ. The faster the particle the narrower the cone. Radiation
propagates at c/nr at the angle θ. Note that

c/nr < v < c (652)

for Cherenkov.
Note that the Cherenkov effect operates for weakly ionized plasmas, be-
cause for highly ionized plasmas, the conductivity is large and the conductiv-
ity dominates in determining the index of refraction. Sometimes when people
refer to a “plasma” they mean a “fully ionized plasma” so be careful. Actu-
ally, for that case, we have also been ignoring the temperature contribution
to the conductivity and dynamics (=cold plasma approximation).

118
 LECTURE 22
Razin effect
For the index of refraction nr < 1 as in a cold plasma, there is no
Cherenkov radiation.
In this case however, beaming can be reduced and thus synchrotron ra-
diation made less potent. Since the beaming is the result of the factor
κ̃ = 1 − βnr cosθ, for nr < 1 beaming is suppressed since no matter how
large β becomes, κ̃ is still significantly larger than zero.
The angle for the beaming in a vacuum was shown to be θ ∼ 1/γ ∼
(1 − β 2 )1/2 . But for a plasma we have instead

θ = (1 − n2r β 2 )1/2 . (653)

If nr ∼ 1 then this reduces to the approximate vacuum case. If nr << 1 but

β ∼ 1, then
θ = (1 − n2r )1/2 = ωp /ω, (654)
for a cold plasma. Thus at low frequencies, the medium will quench the
wave propagation and the plasma effects dominate the beaming effects. At
high frequencies, θ decreases until it gets down to 1/γ and then the vacuum
results hold. Thus the plasma medium effects are unimportant when

ω >> γωp . (655)

The suppression of the beaming effect has a large effect on synchrotron

emission. Below the frequency γωp the spectrum will be largely cut off. This
effect dominates the usual plasma cutoff since it enters for ω < γωp , so large
γ means the effect is strong.
Basically, the decrease in beaming occurs because the electron can no
longer “catch up” to the wave that it had just emitted to reinforce it with
another emission wave. The relativistic beaming enhancement due to this
effect is what would boost synchrotron emission effect over other emission
processes at a given frequency. (note that the enhancement is the same
effect as the Cherenkov only in the latter the particle actually catches up
completely and the particle is moving at a constant speed). Thus the Razin
effect is a relative effect. Since the particle speed v cannot approach ω/k for
sufficiently small nr (since ω/k >> c) the synchrotron is choked.
Selected Topics in Atomic and Molecular Physics
Note that macroscopic theory of radiative transfer requires that the scale
of the system be much larger than the wavelength of the emission. This is

119
not true on the atomic scale. Macroscopic theory OK on larger scales but
we have to compute the absoprtion and emission coefficients from the proper
electromagnetic or quantum processes.
Quantum calculations are needed when the De Broglie wavelengths of
particles are larger than the dimension L which determines the conditions of
a given problem, in this case the atomic scale. Note that if classical theory
applied, the electron orbiting the atom would radiate away its energy and
spiral in indefinitely.
Some special topics to address.
Distribution of energy levels and Ionization
Boltzmann Population of levels:
Relative populations of levels is not easy since it depends on detailed
processes of how a level becomes populated or depopulated. But in thermal
equilibrium things are simple and determined by temperature T . In any
atom collection, the number in a given level is proportional to ge−E/kT , with
g as a statistical weight describing the level degeneracy.
Let Ei be the energy of the ”ith” level with respect to the ground state.
P
Let Ni be the number density of the ith level and N = i Ni is the total
number density, we have Boltzmann’s law:
N −βEi
Ni = gi e , (656)
U
where β = 1/kT , U is the partition function, and is simply

gi e−βEi .
X
U= (657)
i

At small temperatures the only the lowest term matters, the g0 term. This
gives
U = g0 . (658)
Note that for finite temperature the sum diverges. Thus we have to think
about how to cutoff the sum. In fact for a gas, the distance between particles
is finite and outer electrons can be stripped by interactions between atoms.
The ionization potential is thus reduced. A characteristic cutoff results from
assuming the Bohr orbit corresponding to n = nmax is that of the inter-atomic
spacing.
n2max a0 Z −1 ∼ N −1/3 . (659)

120
or
Z 1/2 −1/6
 
nmax ∼ N . (660)
a0
The Bohr radius a0 ∼ .5 × 10−8 cm marks a characteristic scale such that
e2 /a0 ∼ 27 ev is a characteristic atomic binding energy.
For hydrogen at 1012 cm−3 , nmax = 100. However note that there is also
Debye screening and such which will complicate this calculation.
In general, the cutoff is not well understood, and people spend lots of
time writing thesis on this subject.
For many cases the value of the cutoff is not critical. Up to 104 K the U
is usually equal to g0 . Ok for Solar interior, ISM, AGN accretion disks.
Saha Equation
Above was for distribution of levels in a density of a single atom type
in thermal equilibrium. Now we investigate distribution of among various
stages of ionization of single atomic type. The Saha equation describes this.
Consider the tow states being the neutral atom and the first stage of ioniza-
tion.
Consider more general Boltzmann law:
dN0+ (v) g −(χI +1/2me v 2 )
= exp kT . (661)
N0 g0
Here dN0+ (v) is the differential number of ions in the ground state with free
electron in velocity range v + dv, χI is the ionization potential, N0 is the
number of atoms in the ground state and the statistical weight g. Here
g = g0+ ge , the product of statistical weights of the electron and the ground
state ion. The statistical weight ge is given by
2dx1 dx2 dx3 dp1 dp2 dp3
ge = , (662)
h3
where the 2 represents the two spin states and the rest of the factors define
the phase space. In particular Ne dx1 dx2 dx3 = 1, when we apply to volume
containing 1 electron. For the momentum space
dp1 dp2 dp3 = 4πm3e v 2 dv, (663)
for non-relativistic electrons. Then
dN0+ (v) 8πm3e g0+ (χI + 12 me v 2 ) 2
" #
= exp v dv. (664)
N0 h3 Ne g0 kT

121
To find the total number density of the ions we integrate over v to get
!3/2 Z
N0+ Ne 8πm3e g0+ −χI /kT 2kT ∞ 2
= e e−x x2 dx, (665)
N0 h3 g0 me 0

where x ≡ (me /2kT )1/2 v. Using the Gaussian integral result that
R ∞ −x2 2
0 e x dx =
π/4, we have
!3/2
N0+ Ne 2πme kT 2g0+ −χI /kT
= e . (666)
N0 h2 g0
Now if we want the number of atoms or ions in any state not only the
ground state, we use
N0 g0
= (667)
N U (T )
and
N0+ g0+
= . (668)
N+ U + (T )
so that N + /N = U + (T )/U (T ). Thus
!3/2
N + Ne U + (T ) 2πme kT
=2 e−χI /kT , (669)
N U (T ) h2
where N is number density of all neutral atoms and N + is that for ions. This
is the Saha equation.
A similar relationship holds for successive stages of ionization, that is
!3/2
Nj+1 Ne Uj+1 (T ) 2πme kT
=2 e−χj,j+1 /kT . (670)
Nj Uj (T ) h2
We can rewrite all this in terms of pressure rather than number density. The
ideal gas law is
P = nkT, (671)
so that 3/2
Pj+1 Pe 2Uj+1 (T ) 2πme

= (kT )5/2 e−χj,j+1 /kT . (672)
Pj Uj (T ) h2
When you have a mixture of various different species, one must have an
equation that gives the conservation of nuclei
(i)
Nj = N (i)
X
(673)

122
 (i)
where Nj is the jth ionization state of the ith species, so that N (i) is the
total number density of the ith species. In addition we would need charge
conservation XX (i)
Ne = Zj Nj . (674)
i j

In general, the problem must be done numerically.

Also note that trace elements e.g. of alkali metals with low ionization
potentials can make important difference in the equilibrium state because
they may strongly affect the electron number density. Particularly the alkali
metals, like Na, K, Rb, Cs. (Left most column on peroidic table.).

123
 LECTURE 23
Line Broadening
Transitions between atomic states are broadened by a number of effects.
Recall we used the line profile function φ(ν) in our analyses of the Einstein
coefficients.
Doppler Broadening:
This results from the thermal motion of the atom. Thus the emission
and absorption frequency in the frame of the atom represents a different
frequency for the observer.
The change in frequency is

ν − ν0 = ν0 vz /c (675)

where ν0 is the rest frame frequency. The number of atoms having velocities
in the range vz , vz + dvz is
2
e(−ma vz /2kT ) dvz . (676)

From the above, we get

c(ν − ν0 )
vz = (677)
ν0
and
dν
dvz = c . (678)
ν0
Thus the strength of emission in frequency range ν, ν + dν is
−ma c2 (ν−ν0 )2
h i
2ν 2 kT
e 0 dν, (679)

so that the profile function becomes

1 2 2
φ(ν) = 1/2
e−(ν−ν0 ) /(∆νd ) , (680)
∆νd π
where the Doppler width is
ν0
∆νd = (2kT /ma )1/2 . (681)
c
R
The constant 1/(∆νd ) is determined by φ(ν)dν = 1 assuming that ∆νd <<
ν0 .

124
 Let us see what this does to absorption cross sections. Recall that B12 J¯ν
is a transition probability per unit time per frequency (for absorption). This
means that B12 has units of (time)(steradian)(mass)−1 . In addition, we
have
hνul
αν = n1 B12 φ(ν) (682)
4π
which comes from relating the absorption coefficient to the Einstein coeffi-
cient. Note also that
αν = α(ω) = n1 σ(ω), (683)
which relates the cross section to absorption coefficient. Now recall that
in the classical limit we had that the absorption cross section for electron
scattering, based on radiation driven, harmonically bound particles, was

σT Γ 2π 2 e2 Γ
σ(ω) = 2 2
= , (684)
τ 4(ω − ω0 ) + Γ mc 4(ω − ω0 )2 + Γ2

where Γ = ω02 τ . τ is the time for radiation to cross a classical electron radius,
and ω0 is the frequency of the free oscillator. Integrating this, we have
Z
2π 2 e2
σ(ω)dω = (685)
mc
or Z
πe2
σ(ν)dν = = Blu,clas hνlu /4π. (686)
mc
or
4π 2 e2
Blu,clas = . (687)
hνul
The quantum correction to this classical result is given by the absorption
oscillator strength flu and is usually less than but of order 1. It is derived in
the text. Then
4π 2 e2
Blu,clas = flu . (688)
hνul
Getting back to the cross section, we note that from (682) that the line
center cross section for each atom, corrected for Doppler broadening, is
1/2
104
!
hνul 1 hνlu 1/2 1 A
σνlu = B12 φ(ν) = B12 = π r0 cf 12 = f12 cm2 .
4π ∆νD π 1/2 4π ∆νD νlu T
(689)

125
Note that the cross section for a given frequency is reduced for large tem-
peratures. This is because the absorption is “spread out” over a range of
frequencies.
There can be turbulent broadening in addition to the Doppler broadening.
When the scale of the turbulence is small in comparison to the mean free
path (micro-turbulence) the motions can be accounted for with an effective
Doppler width:
!1/2
νul 2kT 2
∆νD = +ξ (690)
c ma
assuming that the micro-turbulence is Gaussian. If the turbulence is on a
scale larger than a mean free path, than we have “hydrodynamic” or “magne-
tohydrodynamic” turbulence. Then, one would take whatever profile one has
with the line (including collisional effects as described below) and integrate
over the hydrodynamic turbulent distribution. One can also have systematic
effects on the shape of the line profiles.
Natural Broadening
Atomic level has a certain width due to the uncertainty relation. The
spread in energy ∆E∆t ∼ h/2π. The spontaneous decay of state n into
occurs at rate X
Γ= Ann0 (691)
n0

the sum is over the lower energy states. The energy for the nth state the
decays as e−Γt . The emitted spectrum is determined y the decaying sinu-
soid type of electric field (fig 2.3), or radiation from a harmonically bound,
undriven particles. This was given by the Lorentz profile

Γ/4π
φ(ν) = (692)
(ν − ν0 )2 + (Γ/4π)2

also called the natural profile. Actually, the above applies to cases when
the bottom state is the ground state, and the difference in energies really
correspond to a broadening of the upper state. If both upper and lower
states are broadened, then Γ = Γu + Γd , which means that the total width is
determined by the sum of the widths of the upper and lower states.
Collisional Broadening
If the emitted radiation suffers collisions, then the phase of the emission
can be lost. fig 10.3

126
 By assuming that collisions induce a time dependent random phase in the
electric field, and calculating the emitted power as a function of frequency
by Fourier transforming the square of E(t), then further assuming a Poisson
distribution for number of collisions with given mean, one can show that the
collision effect also gives Lorentz profile.
To see this, consider the electric field and let φc (t) be random phase
resulting from collisions. We have

E(t) = Aeiωo t−Γt/2+iφc (t) , (693)

(with A as a constant, Γ is the spontaneous decay rate and ω0 is the funda-

mental frequency). The averaged power spectrum is given by
Z
2
h|Ẽ(ω)|i = h| E(t)eiωt dt|2 i. (694)

Using E(t) above we have

Z ∞Z ∞
2
|Ẽ(ω)| = |A| 2
dt1 dt2 ei(ω−ω0 )(t1 −t2 )−Γ(t1 +t2 )/2 ei[φc (t1 )−φc (t2 )] . (695)
0 0

Only the phase function is random. Thus we get

Z ∞Z ∞
2 2
h|Ẽ(ω)| i = |A| dt1 dt2 K(t1 , t2 )hei[φc (t1 )−φc (t2 )] i (696)
0 0

where K(t1 , t2 ) = ei(ω−ω0 )(t1 −t2 )−Γ(t1 +t2 )/2 . Now write

φc (t1 ) − φc (t2 ) = ∆φc (∆t) (697)

where ∆φc (∆t) is the change of phase in time interval ∆t = t2 − t2 . The

randomness of the collisions means that if 1 or more collisions occur in time
interval (t1 − t2 ), then the average vanishes (average of a rapidly oscillating
function). If no collisions occur, then the quantity averages to 1.
Given that the mean rate of collisions is νcol , we have that for a Poisson
distribution of collisions, the probability for no collisions to occur in ∆t is

P (x = 0) = |e−λ λx /x!|x=0 = e−|t2 −t1 |νcol , (698)

where λ = νcol ∆t is the probability of collisions per unit time times the length
of time, and x is the number of collisions we are seeking the probability for.

127
 Thus
hei∆φ(∆t) i = e−|∆t|νcol , (699)
and we have
Z ∞Z ∞
h|Ẽ(ω)|2 i = |A|2 dt1 dt2 K(t1 , t2 )e−|∆t|νcol , (700)
0 0

Integration gives a combined collisional/natural Lorentz profile

Γ̃/4π
φ(ν) = , (701)
(ν − ν0 )2 + (Γ̃/4π)2

with Γ̃ = Γ + 2νcol , where νcol is mean number of collisions per unit time.
Combined Doppler Broadening and Lorentz Profiles
One can combine the Lorentz profile, plus the Doppler effect. We can
write the profile as an average of the Lorentz profile weighted by the velocity
states.
Γ̃ Z ∞
(m/2πkT )1/2 exp(−mvz2 /2kT )
φ(ν) = 2 = dvz (702)
4π −∞ (ν − ν0 − νvz /c)2 + (Γ̃/4π)2

where vz is the line of sight component of the velocity. This can be simplified
with the Voigt function
2
aZ∞ e−y dy
H(a, u) ≡ (703)
π −∞ a2 + (u − y)2

then
φ(ν) = (∆νD )−1 π −1/2 H(a, u) (704)
with a = Γ̃/4π∆νD and u = (ν − ν0 )/∆νD . For small values of a, the central
part of the line is dominated by the Doppler profile and the regions away
from ν = ν0 the “wings” are dominated by the Lorentz profile.

128