ASTR 4003/8003, Class 11: Compton scattering I 3 April

Thus far we have mostly dealt with emission and absorption of radiation by electrons, and we
have treated scattering only briefly. In this week of the class we turn our attention back to
scattering processes. We will see that scattering can by itself produce important and observable
effects, including the production of high energy photons.

I. Scattering kinematics
In our discussion of scattering of electromagnetic radiation by electrons, we considered an
electron interacting with a plane wave, and calculated the power radiated by the electron
as it oscillated in the plane wave’s electric field. In doing so, we neglected magnetic forces
as small in comparison to electric ones for determining the electron’s kinematics. As we
showed on the first problem set, this is equivalent to neglecting the momentum or pressure
associated with the plane wave, because it is precisely that magnetic force that, in the
classical limit, transfers momentum to the matter.

A. Scattering in the electron rest frame

Our first step in revisiting scattering is to properly account for the momentum of
the radiation field, and the corresponding recoil of the electron. Scattering properly
accounting for this effect is known as Compton scattering. This is easiest if we view
radiation quantum mechanically, as consisting of photons. Consider an electron
initially at rest, which scatters a single photon that is initially travelling in the x
direction. The four-momenta of the electron and photon prior to the interaction are
   
me c 1
 0  i  1 

p~e,i = 
 0 
 p
~γ,i = , (1)
c 0 
0 0
where me is the electron rest mass and i is the photon energy prior to the scattering
event.
After the photon scatters, the photon is sent off at an angle θ relative to its initial
direction, and the photon energy is f . The electron recoils with some final three-
momentum pf . The four-momenta of the electron and photon after scattering are
therefore    
E/c 1
 px,f  f 
 cos θ  ,

p~e,f =  
 py,f  p
~γ,f = (2)
c  sin θ 
0 0
where E is the electron energy after scattering, and without loss of generality we
have set up our coordinate system so that the electron and photon trajectories after
the scattering event lie purely in the xy plane.
Conservation of momentum requires that p~e,f = p~e,i + p~γ,i − p~γ,f , so
   
E/c me c2 + i − f
 px,f  1  i − f cos θ 
 py,f  = c  (3)
   
−f sin θ 
0 0

1
 Taking the magnitude of both sides,

me c2 = me c2 + 2me (i − f ) − 2f i (1 − cos θ) (4)

i
f = i . (5)
1 + me c2 (1 − cos θ)

This gives us the final photon energy in terms of the initial energy and the scattering
angle. We see that, unless the initial photon energy i is comparable to or larger
than the electron rest energy me c2 = 511 keV, the recoil of the electron is negligible,
and we can treat the scattering as nearly elastic.
Although we will not make use of this result for the most part here, we pause to
note that when i & me c2 , the cross section is modified from the Thomson value as
well. The full energy-dependent cross section for scattering of photons of arbitrary
energy by a free electron, which we will quote but not prove, is
 2 
r02 f

dσ i f 2
= + − sin θ , (6)
dΩ 2 i f i
where r0 is the classical electron radius. This expression is known as the Klein-
Nishina cross section. For low-energy photons, i  me c2 , we have f ≈ i , and
dσ/dΩ ≈ (1/2)r02 (1 + cos2 θ), which is the Thomson cross section we derived earlier
in the class. The true scattering cross section at i & me c2 is reduced compared to
the Thomson value.

B. Scattering in other frames

Now let us consider the scattering process as viewed in a frame where the electron
is not initially at rest. Suppose that in the observer frame the scattering electron is
moving at velocity v at an angle θi relative to the direction of the incoming photon.
The photon energy is i . How are the incoming and outgoing photon energies related?
To answer this question, we can transform into the electron rest frame, use the result
we have already computed there, and then transform back. Viewed in the electron
rest frame, the photon frequency transforms following the relativistic Doppler for-
mula; energy, which is proportional to frequency, transforms in the same way. Thus
in the electron rest frame, the photon energy is

0i = i γ(1 − β cos θi ), (7)

where β = v/c and γ = (1 − β 2 )−1/2 . Suppose that the photon scatters through an
angle ψ 0 as viewed in the electron rest frame. Using the results we derived previously,
the photon energy after scattering as measured in the electron rest frame is
0i
0f = 0i
. (8)
1+ m e c2
(1 − cos ψ 0 )

If the photon energy as viewed in the co-moving frame is small compared to the
electron rest mass, then this is approximately
0i
 
0 0 0
f ≈ i 1 − (1 − cos ψ ) . (9)
m e c2
This is a relatively common case, even for pretty relativistic electrons: for example
even if the electron has γ = 104 , which is very relativistic, an optical photon, i ≈ 3

2
 eV, only has an energy of at most 30 keV as viewed in the electron rest frame, much
less than the 511 keV electron rest mass.
Finally, we can boost back into the observer frame to find the final electron energy
in that frame:
0i
 
0 0 2 0 0
f = f γ(1+β cos θf ) ≈ i γ (1−β cos θi )(1+β cos θf ) 1 − (1 − cos ψ ) , (10)
m e c2
where θf0 is the angle between the photon and the boost as viewed in the electron
rest frame. The important thing to notice about this expression is that the various
trigonometric factors in parentheses are all of order unity for most scattering angles,
so the final energy is typically larger than the initial energy by a factor of order
γ 2 . This can be large: our hypothetical γ = 104 electron that scattered an optical
photon with an initial energy of i = 3 eV would yield a γ-ray photon with an energy
of order 100 MeV!
It is worth pausing to note that the phenomenon we’re identifying here is very closely
analogous to gravity boosting in orbital dynamics. In gravity boosting, a small body
such as a spacecraft makes a close pass by a more massive body such as a planet,
scattering off it gravitationally. In the reference frame of the planet, the scattering is
elastic, so the magnitude of the spacecraft velocity is not altered by the interaction;
only the direction is changed. However, in the frame of the Sun in which the planet
is moving, the spacecraft gets a boost of order twice the orbital speed of the planet,
because the frame transformations are asymmetric between when the spacecraft is
on its way inward toward the planet and when it is on its way outward. Here we
have basically the same situation.

II. Inverse Compton scattering in the isotropic, single-scattering limit

Now that we have understood the microphysics of how moving electrons can scatter
photons to high energy, we will next attempt to calculate the rate at which this process
occurs for a population of electrons and photons with isotropic velocities. We will first
compute the total power in high energy photons produced by this process, and then
compute the spectrum of the resulting radiation. For now we will limit ourselves to
media where the scattering probability is low enough that any given photon is unlikely to
scatter more than once before it escapes from the region occupied by the hot electrons.
We will generalise to the case of multiple scatterings in the next class.

A. Inverse Compton power

Begin by considering the case of a population of photons that all have the same
frequency ν0 as measured in the lab frame, and that are travelling isotropically. The
intensity of this radiation field is
Iν (n̂) = Kδ(ν − ν0 ) (11)
where K is some constant; isotropy of the radiation field means that Iν is independent
of direction n̂.
Now consider an electron travelling at speed v = βc. What radiation intensity does
it see in its rest frame? The answer is given by the Thomas transformations we
derived earlier in the class:
 0 3  0 3
0 0 ν ν
Iν 0 (n̂ ) = Iν (n̂) = Kδ(ν − ν0 ) , (12)
ν ν

3
where the frequencies in the electron rest frame (denoted by primes) and the lab
frame (unprimed) are related by the relativistic Doppler formula
ν
0
= γ(1 + βµ0 ) (13)
ν
where θ0 is the angle between the photon’s direction and the electron’s direction
as viewed in the electron rest frame, and we have written µ0 = cos θ0 for compact-
ness. We want to have everything written in the electron rest frame, so we need to
transform the δ function:
ν0
δ(ν − ν0 ) = δ(ν 0 − ν00 ). (14)
ν
Note thatR one can see that the leading factor of ν 0 /ν is required because 1 = δ(ν −
R

ν0 ) dν = δ(ν 0 − ν00 ) dν 0 , and it therefore follows that δ(ν − ν0 ) = δ(ν 0 − ν00 )(dν 0 /dν).
Thus we have
 4  
0 0 1 0 ν0
Iν 0 (n̂ ) = K δ ν − (15)
γ(1 + βµ0 ) γ(1 + βµ0 )

We are now in a position to write down the scattered power in the electron rest
frame, which is related to the incoming photon intensity in the usual way:
ZZ ∞ Z 1
dE 0 0 0 0 dµ0
= σ T Iν 0 dν dΩ = 2πσT K (16)
dt0 0
0 4
−1 γ(1 + βµ )

This expression just says that the power emitted by the scattering electron is equal
to the incoming intensity, integrated over all directions and frequencies, multiplied
by the Thomson cross section. The integral is straightforward to evaluate, and the
result is
dE 0 β2
 
2
= 4πσT Kγ 1 + . (17)
dt0 3
However, we now recall that we previously showed that radiated power is a Lorentz
invariant: Lorentz boosting changes the angular distribution of the radiated energy,
but it doesn’t change the total amount of energy radiated. Thus dE/dt = dE 0 /dt0 ,
and we therefore the radiated power as measured in the lab frame is

β2
 
dE 2
= 4πσT Kγ 1 + . (18)
dt 3

Of course the angular distribution is different in the laboratory and electron rest
frames, but this doesn’t matter because in the end we are concerned with isotropic
populations, so we know that, once we average over all the electrons that are scat-
tering photons, the result will be isotropic as well.
The quantity K is most conveniently expressed as follows. Note that the frequency
ν does not appear in Equation 18, so the power radiated just depends on how much
energy there is at a given photon frequency, and not on the frequency itself. Thus
scattering power for photons at different frequencies just adds, and for an arbitrary
frequency-dependent (but still isotropic) intensity, we must have

β2 β2
 Z  
dE 2 4 2
= 4πσT γ 1 + Iν dν = σT cγ 1 + Uγ , (19)
dt 3 3 3

4
 RR
where Uγ = (1/c) Iν dν dΩ is the photon energy density.
Finally, we must be careful to make a distinction between the radiated power and
the inverse Compton power. The radiated power is the total scattered energy, which
would be non-zero even if the electrons were at rest and had β = 0 and γ = 1. The
inverse Compton power is the extra energy added to the radiation by scattering off
the electrons, i.e., it is the difference between the actual power radiated and the
energy per unit time that the electrons scatter. This latter quantity is just
 
dE
= cσT Uγ , (20)
dt scat
since the (isotropic) flux of energy hitting the electrons is cUγ , and the scattering
cross section is σT . Thus the inverse Compton power is given by
β2
     
dE dE 2 4
PIC = − = cσT Uγ γ 1 + − 1 = σT cβ 2 γ 2 Uγ , (21)
dt dt scat 3 3
where in the last step we used the fact that γ 2 − 1 = γ 2 β 2 .
We have therefore derived the inverse Compton power emitted per electron. This
result has a few immediate applications. First, recall that we previously showed that
the synchotron power emitted per electron is
4
Psynch = σT cβ 2 γ 2 UB , (22)
3
where UB is the magnetic energy density. Thus we arrive at the relationship
PIC Uγ
= , (23)
Psync UB
i.e., the ratio of a source’s inverse Compton power to its synchotron power is given
by the ratio of its radiation and magnetic energy densities. In interstellar space
in normal galaxies like the Milky Way, magnetic and radiation energy densities
tend to be about comparable, and thus normal galaxies are about equally bright
in synchotron and inverse Compton radiation, though the two processes generally
populate very different parts of the electromagnetic spectrum.
A second application is that we can derive the inverse Compton power produced by
a population of non-relativistic but hot electrons in thermal equilibrium; an astro-
physical example where this is very relevant is emission produced by light passing
through the hot corona around a black hole’s accretion disc – the light emitted in
the cooler regions in the disc interior passes through the hot coronoa, and some of it
is up-scattered to much higher energies. For non-relativistic electrons, taking γ ≈ 1,
the inverse Compton power produced per unit volume by electrons with number
density ne is
4
PIC = σT cUγ β 2 ne . (24)
3
Since β = v/c, and the mean square velocity for a Boltzmann distribution is
3kB T /me , we therefore have
4kB T 4Pe
PIC = 2
cσT ne Uγ = cσT Uγ , (25)
me c me c2
where Pe is the electron pressure. Thus the inverse Compton power just scales with
the product of the pressure of the electrons and the photon energy density.

5
B. Inverse Compton spectra
Next let us attempt to compute the spectrum of the IC-scattered photons, again
in the limit where we assume that the medium is sufficiently optically thin that
we can neglect multiple scattering and where the photon energy is small compared
to the electron rest mass in the electron rest frame. To minimise mathematical
complication, we will also neglect the angle-dependence of Thomson scattering, and
treat it as isotropic in the electron rest frame. This is not quite correct, but it is
sufficient to recover the major physical characteristics of the result.

1. Scattering by electrons traveling in one direction

As with our calculation of IC power, we will start by considering a mono-
energetic photon distribution being scattered by a single electron with Lorentz
factor γ. As above, the intensity as seen in the electron rest frame is
 0 3
0 0 ν
Iν 0 (n̂ ) = Kδ(ν − ν0 ) , (26)
ν
where K is a constant. We will now make a slightly different change of variables
inside the δ function than we did above. Using the relativistic Doppler formula,
we can rewrite the frequency ν in the lab frame as
ν = γ(1 + βµ0 )ν 0 , (27)
and thus the condition that ν − ν0 = 0 expressed by the δ function can be
re-arragned as the condition
ν0 − γν 0
µ0 − =0 (28)
γβν 0
Physically, all we have done here is to note that, for any given frequency ν 0 as
seen in the electron rest frame, the condition that ν = ν0 is satisfied only at one
particular angle µ0 between the photon direction and the boost. As before, we
can find the relationship between the δ function in terms of ν and that in terms
of µ0 by noting that
Z ∞ Z 1 
ν0 − γν 0

0
1= δ(ν − ν0 ) dν = δ µ − 0
dµ0 (29)
0 −1 γβν
and thus
dµ0 ν0 − γν 0 ν0 − γν 0
   
0 1 0
δ(ν − ν0 ) = δ µ − = δ µ − , (30)
dν γβν 0 γβν 0 γβν 0
where the last step follows immediately from Equation 27. Thus the intensity
in the electron rest frame is
 0 3
ν0 − γν 0
 
0 0 ν 1 0
Iν 0 (n̂ ) = K δ µ − . (31)
ν γβν 0 γβν 0

Armed with this expression, we can now immediately write down the scattering
emissivity in the electron rest frame:
 0 3
1 1 ν0 − γν 0
Z Z  
0 0 σT 0 0 0 0 ν 1 0
jν 0 = ne Iν 0 (n̂ ) dΩ = ne σT K δ µ − dµ0 ,
4π 2 −1 ν γβν 0 γβν 0
(32)

6
 where n0e is the number of scattering electrons per unit volume as seen in the
electron rest frame. The integral just integrates over all the directions of in-
coming electrons. Note that this expression makes use of our two assumptions:
first, that the scattering is isotropic, so that dσ 0 /dΩ0 = σT /4π, and second, that
the photon energy hν 0 is much smaller than the electron rest mass, so that we
neglect the energy difference between the incident and scattered photon in the
electron rest frame.
The integral is completely trivial: it vanishes for values of ν 0 such that the
argument of the δ-function is never zero for µ0 in the range [−1, 1], and for any
other value of µ0 it is just the integrand evaluated with µ0 set to the value that
makes the argument of the δ function zero, which is equivalent to setting ν = ν0 .
That is
n0e σT Kν 02 ν0 ν0
jν0 0 = 3
, if < ν0 <
2γβν0 γ(1 + β) γ(1 − β)
= 0, otherwise (33)
The physical meaning of this result is that there is minimum and maximum
frequency of IC emission, corresponding to the cases of scattering a photon
that is travelling exactly opposite the direction of the electron (which gives the
highest frequency scattered light) and scattering a photon that is travelling in
the same direction as the electron (which gives the lowest frequency). In between
these two, the spectrum just has a ν 02 dependence, which arises entirely from
the angle-dependence of the relativistic Doppler formula.
Finally, to obtain the emissivity in the lab frame, we note that the lab frame
and comoving frame number densities are related by ne = γn0e , and again make
use of the Thomas transformations
 ν 2
jν (µ) = jν0 0 (34)
ν0
ne σT Kν 2 ν0 ν0
= 2 3
, if 2 <ν<
2γ βν0 γ (1 + β)(1 − βµ) γ(1 − β)(1 − βµ)
= 0, otherwise (35)
Here µ is the angle between the electron velocity and the scattered photon
in the lab frame. We have therefore found the frequency-dependent emissivity
produced by electrons traveling in one direction scattering an isotropic radiation
field.
2. Scattering by isotropic electrons
The final step is to average the emissivity over µ, since we are interested in
an isotropic population of electrons, and thus the angular dependences of all
the electrons travelling in random directions will average out. Thus we need to
evaluate
1 1
Z
jν = jν (µ)dµ. (36)
2 −1
To aid in this evaluation, we can re-arrange the condition on the frequency in
Equation 35 into a condition on the angle µ: the emissivity at a given frequency
ν is non-zero only for angles µ that satisfy
1h ν0 i 1h ν0 i
1 − (1 + β) < µ < 1 − (1 − β) . (37)
β ν β ν

7
Physically, this just says that photons of frequency ν0 get scattered to frequency
ν0 only within a certain range of scattering angles. For example, producing
photons near the maximum possible energy requires that the electron scatter
the photon almost perfectly forward, along its direction of motion. The highest
frequency that is possible corresponds to µ = 1, which (using the expression
above) corresponds to ν/ν0 = (1 + β)/(1 − β). The lowest frequency possible is
for µ = −1, and corresponds to ν/ν0 = (1 − β)/(1 + β).
With this understood, the integration itself is trivial, since µ does not appear
explicitly in jν (µ) except through defining the limits where it is non-zero. We
just need to be careful about the bookkeeping on the limits of integration. Doing
so, the final result is
(1 + β) νν0 − (1 − β), 1−β < νν0 < 1

ne σT Kν  1+β
jν = 2 2 2 (1 + β) − νν0 (1 − β), 1 < νν0 < 1−β
1+β
(38)
4γ β ν0 
0, otherwise

Let us try to understand the physical meaning of this result. This is a bit easier
to understand if we work in terms of photons rather than in terms of energy.
The intensity of the background radiation field that is being scattered, measured
in photons per solid angle per frequency per area per time, is just K/ν0 , and the
scattered intensity measured in photons per unit volume per frequency per time
is jν /ν, so the rate at which individual photons are scattered from frequency ν0
to frequency ν is proportional to
 1−β
(1 + β)ξ − (1 − β), 1+β <ξ<1
1  1−β ν
g(ξ) = 2 2 (1 + β) − ξ(1 − β), 1 < ξ < 1+β ξ= (39)
γ β  ν0
0, otherwise
where ξ = ν/ν0 . Written this way, the emissivity is
ne σT Kν
jν = g(ξ). (40)
4ν02
The shape of g(ξ) is interesting (Figure 1). For β near zero, the function is
symmetric about ξ = 1, i.e., photon energy is equally likely to increase or
decrease. As β → 1, however, the function becomes highly asymmetric, with a
much greater probability of scattering to higher energy, and a probability that
is nearly flat up to the cutoff (1 + β)/(1 − β) ≈ 4γ 2 . Note that, since energy is
proportional to frequency, this means that the spectral energy distribution will
peak at frequencies just slightly below the cutoff, i.e., the spectrum produced
by a single photon energy and a single electron Lorentz factor γ carries most of
its energy in photons with energies just a bit less than 4γ 2 .
Now that we have obtained the inverse Compton spectrum in the single-scattering
limit, we can estimate the observable spectrum produced by scattering of isotropic
photons with any frequency distribution by isotropic electrons with any velocity
distribution just by integrating over the two distributions:
Z
σT ν dne
jν = 2
g(ξ)Iν0 dν0 dγ. (41)
4ν0 dγ

Finally, we note that doing the full problem, properly keeping the angle depen-
dence of Thomson scattering in the electron rest frame, just mildly changes the

8
 Scattering rate
100

10

β=0.1
1
β=0.5
β=0.9
0.100
β=0.99

0.010

0.001 ν/ν0
0.5 5 50 500

Figure 1: Inverse Compton photon scattering rate as a function of ratio of outgoing to incoming
photon frequency ν/ν0 , for different electron β values.

shape of g(ξ), without altering the overall qualitative behaviour. In particular,

the high frequency cutoff, and the fact that the spectral energy distribution
peaks near it in the ultra-relativistic case, are unchanged, since they arise just
from consideration of the highest energy that can be achieved when a photon
scatters off an electron head-on.

C. Powerlaw electron distributions

One of the most important applications of this result is to the case of a powerlaw
distribution of relativistic particles, as expected for example for cosmic rays. This
emission is routinely observed from galaxies at X-ray and γ-ray wavelengths. As
usual, we will consider a powerlaw distribution dne /dγ ∝ γ −p . We will not bother
keeping track of normalisations from this point, because the proper normalisation
can always be found in the end by requiring that the total radiated power integrated
over all frequencies add up to the result given by Equation 21. Thus we will only
keep factors that affect the spectral shape.
To evaluate Equation 41, it is helpful to rewrite the function g(ξ) in the ultra-
relativistic limit, since that is what we are concerned with. We define
ν
x= , (42)
4γ 2 ν0
so that x = 1 corresponds to the maximum possible frequency in the ultra-relativistic
limit. In this case we can rewrite the emissivity for a particular combination of
photon energy ν0 and electron Lorentz factor γ as
3ne σT Kν
jν (ν0 , γ) = f (x) (43)
4γ 2 ν02

where for the isotropic case we solved and x > 1/(4γ 2 ) we have f (x) = 2/3(1 − x).
The result of properly including the Thomson scattering angle dependence is very
similar,
f (x) = 2x ln x + x + 1 − 2x2 . (44)

With this re-parameterisation, we can write the total emissivity as

ZZ ZZ
3
jν = jν (ν0 , γ) dν0 dγ = C νIν0 γ −p−2 f (x) dν0 dγ, (45)
4

9
where C is a constant that collects all the terms that do not depend on ν, ν0 , or γ.
We can evaluate this by making a change of variables from γ to x, yielding
Z ∞ Z x1
−(p−1)/2 (p−1)/2 (p−1)/2
jν = Cν ν0 x Iν0 f (x) dν0 dx, (46)
0 0

where x1 = ν/4ν0 . The remaining integrals over x and ν0 depend on frequency

ν only through the limit of integration x1 . If we choose to evaluate the spectrum
at a frequency such that x1  1, then we can approximate the integral over x as
running from zero to infinity, and the integrals are just numbers with no frequency
dependence. We are therefore left with

jν ∝ ν −(p−1)/2 , (47)

i.e., exactly the same spectrum as synchotron emission. This is not an accident: the
basic approximation we made here is the same as for the synchotron case, and for
much the same reason: it doesn’t really matter what the exact spectral shape is at
a given γ, as long as it is less broad than an infinite powerlaw. If it is, then the final
spectrum is going to be determined just by the powerlaw index of dne /dγ, and the
way that emissivity scales with γ, which is the same in the two cases. Details of the
spectral shape don’t matter.

10