Compton Scattering

Last time we talked about scattering in the limit where the photon energy is much
smaller than the mass-energy of an electron. However, when X-rays and gamma-rays are
considered, this approximation is no longer so good. Therefore, today we’ll talk about the
more general process of Compton scattering. We’ll start with the basic process, then discuss
how it has an effect on spectra and the energetics of electrons.
Think about a photon scattering off an electron that is initially at rest. Let’s treat both
the photon (of initial energy h̄ω0 ) and the electron as if they were particles. Suppose that
after the photon scatters it moves off into a direction an angle θ from its original direction.
Ask class: how would we figure out the new energy of the photon? We do this by energy
and momentum conservation. Ask class: will the entire interaction take place in a single
plane? Yes, it will, so that simplifies things. Before we calculate, Ask class: will the final
photon energy be less than, equal to, or greater than the initial photon energy? Less than,
since electron recoil will take some energy. This is the difference between full Compton
scattering and our Thomson approximation, in which the photon energy was unchanged
(thus producing coherent scattering).
We can rephrase our conservation laws a la special relativity by saying that the total four-
momentum is conserved. The initial four-momentum of the photon is P~γi = (h̄ω0 /c)(1, ni ),
where ni is the initial direction of the photon. Similarly, the final four-momentum of the
photon is P~γf = (h̄ω1 /c)(1, nf ). The initial four-momentum of the electron is P~ei = (mc, 0),
since it isn’t moving, and the final is P~ef = (E/c, p). Setting P~γi + P~ei = P~γf + P~ef and
manipulating a bit gives
h i
h̄ω1 = h̄ω0 / 1 + (h̄ω0 /mc2 )(1 − cos θ) . (1)
In terms of wavelength it’s even easier:
λ1 − λ0 = λc (1 − cos θ) (2)
where λc ≡ h/mc = 0.02426Å(me /m) is the Compton wavelength.
Does this make sense to us? Ask class: what limits can we check? Indeed the
wavelength only increases, as we guessed beforehand. If θ = 0 there is no deflection and no
change, as is reasonable. The change in wavelength is independent of λ, meaning that for
smaller λ (larger photon energy) the fractional change is greater, which makes sense. Also,
for a more massive particle (e.g., a proton instead of an electron) the wavelength change is
less, meaning that the photon energy has to be larger for there to be a significant deviation
from Thomson.
 Something not as obvious is that the cross section is also modified by this effect. The
differential cross section is
h i
dσ/dΩ = (r02 /2)(ω1 /ω0 )2 ω0 /ω1 + ω1 /ω0 − sin2 θ . (3)
The main effect is to reduce the cross section from the Thomson value, but not very rapidly.
Rybicki and Lightman give the expression for the total angle-integrated cross section, but
it’s complicated and the most important parts are the limits at low and high energy:
σ ≈ σT (1 − 2x + . . .), x ≪ 1,
(4)
σ ≈ 83 σT x−1 (ln 2x + 1/2 + . . .), x≫1.
Here we define x ≡ h̄ω/mc2 . As before, we’re usually thinking of photon-electron scattering,
but it’s more general than that (e.g., a proton will do if we let m = mp ). In any case, we see
that for low energies the cross section is close to (but less than) the Thomson cross section,
whereas at high energies the cross section scales as ∼ (ln x)/x, ignoring smaller terms.
Ultimately, this means that very high energy gamma rays have a substantially longer mean
free path to scattering (or absorption) than lower energy photons do.
Fine, so we’ve done scattering off of a stationary electron. We found that in such
a scattering the photon always loses energy. Ask class: is this still true in a frame in
which the electron was moving initially? No! Think of a photon incident from the left,
and a rapidly-moving electron incident from the right. If the photon doesn’t have too
overwhelming an energy, the result will be that the electron moves more slowly after
the collision (think of an elastic collision of two balls with different masses). Energy
conservation then demands that the photon gain energy. This is still Compton scattering,
but because the energy is transferred from the electron to the photon it is usually called
inverse Compton scattering. Note, by the way, that if the energy of the photon is large
enough, the electron will gain energy as before.
As always in special relativity, the key to analyzing this is to pick a simple reference
frame for your analysis, then transform to the original frame. In this case, the simple frame
is the one in which the electron is originally at rest, which we already did. If the electrons
are relativistic (γ 2 − 1 ≫ h̄ω/mc2 ) then, if in the e− rest frame we have h̄ω ≪ mc2 , the
energy of the photon before scattering, the energy of the photon in the electron rest frame,
and the energy of the photon after scattering are in the rough ratio 1 : γ : γ 2 . That means
that a gain in energy in a single scattering can be a factor of about γ 2 , which can be
enormous. Of course, there are limits. Ask class: if the electron has energy γme c2 , what
is the maximum gain of energy by the photon? It can only gain (γ − 1)me c2 . Ask class:
suppose that in the electron rest frame the condition h̄ω ≪ mc2 does not hold. What are
the qualitative effects? One is that the electron will experience recoil, so the photon will
either lose energy or not gain as much energy. Another is that the cross section will drop,
so interactions will be less frequent.
 Now let’s ask the following question: if (as measured in a particular reference frame)
one has an isotropic distribution of electrons all of which have the same speed βc, and
one has an isotropic distribution of photons, what will be the net power transferred from
electrons to photons by inverse Compton scattering?
Suppose that the number density of photons with energy in a range dǫ is (dnγ /dǫ)dǫ.
Rybicki and Lightman use υ instead of dnγ /dǫ, but that looks too much like velocity to me
:). If we ignore any change of energy of the photon in the rest frame, then Rybicki and
Lightman show that the scattered energy in a direction θ with respect to the motion of the
electron is Z
dE1 /dt = cσT γ 2 (1 − β cos θ)2 ǫ(dnγ /dǫ)dǫ . (5)
This makes sense: in a direction θ, the Doppler shift is γ(1 − β cos θ), so by the 1 : γ : γ 2 rule
one expects that the final energy of the photon in the lab frame will be its original energy
times the square of the Doppler shift. To get the average over an isotropic distribution of
photons, we need h(1 − β cos θ)2 i, which is 1 + β 2 /3 (since hcos θi = 0). Note also that this
average has the right symmetry: the direction of motion of the electrons can’t matter (i.e.,
β → −β is irrelevant), so the power must go like an even power of β. The scattered energy
rate is then
dE1 /dt = cσT γ 2 (1 + β 2 /3)Uph , (6)
where Uph ≡ ǫ(dnγ /dǫ)dǫ is the initial photon energy density. Subtracting from this the
R

original energy of the photons, we have

dErad /dt = cσT Uph [γ 2 (1 + β 2 /3) − 1]
(7)
= 34 σT cγ 2 β 2 Uph .
This is the net power transferred by inverse Compton scattering.
Let’s take stock. Ask class: does it make sense that we only had to specify that the
photons are distributed isotropically, rather than deciding on their spectrum? Yes, because
in the rest frame Thomson approximation, all photons have the same fractional energy
boost, so a spectrum isn’t necessary. Ask class: how else can we test this expression? In
the limit β → 0 the power goes to zero (quadratically), as expected. In the limit β → 1 we
have γ → ∞, so the power becomes large, as expected. The power also scales linearly with
cross section and photon energy density, which makes sense.
We did, however, ignore electron recoil. Ask class: qualitatively, what difference
would that make? If there is enough electron recoil, the photons can transfer energy to the
electrons, so the Compton power could be negative. When this is accounted for, we get
instead
63 γhǫ2 i
" #
4 2 2
Pcompt = σT cγ β Uph 1 − . (8)
3 10 mc2 hǫi
Here the angle brackets around ǫ2 and ǫ are mean values over the photon spectrum. Ask
class: why should the spectrum matter now? It’s because electron recoil definitely does
depend on the energy of the photon!
Since we have the power for electrons of a single speed, if we’re given a distribution
N (γ)dγ of electrons per unit volume with Lorentz factors between γ and γ + dγ we can just
integrate over the distribution to get the total power per volume:
Z
Ptot = Pcompt N (γ)dγ . (9)

Again, this is a power per volume, or erg s−1 cm−3 . For example, for a power law distribution
N (γ) = Cγ −p for γ between γmin and γmax and zero otherwise, the total power per volume is
4 3−p 3−p
Ptot = σT cUph C(3 − p)−1 (γmax − γmin ) (10)
3
assuming the electrons are nonrelativistic. Similarly, if the electrons are in a thermal
distribution with temperature T , so that hβ 2 i = 3kT /mc2 , then
Ptot = (4kT /mc2 )cσT ne Uph . (11)

We’d also like to figure out how a single Compton scattering affects the spectrum of
an isotropic field of radiation. In principle, this poses no fundamental difficulties: from the
fundamental Compton process we know the energy a photon will have if it comes in with
one direction and scatters off in another, so we “merely” have to do the proper integrals.
Rybicki and Lightman do this for a simplified case, in which in the electron’s rest frame the
scattering has the Thomson cross section but is isotropic instead of the dumbbell-shaped
Thomson angular distribution. Even then, the integrals end up involving things like gamma
and zeta functions. Feel free to look at Rybicki and Lightman, section 7.3, for details.
We’re not going to discuss the spectra to that level, but will instead mention a couple of
the main points.
First, Ask class: if the energy of a photon is initially h̄ω, then in the Thomson limit
(coherent scattering in the electron rest frame), what are limits on the energy of the photon
after scattering? The maximum Doppler blueshift is when the scattering is head-on, in
which case the photon energy changes by a factor [γ(1 + β)]2 . Similarly, the minimum factor
(maximum redshift) is [γ(1 − β)]2 . Thus, for a given input photon energy, the spectrum is
bounded. The second point is the form of the spectrum if the electron speeds have a power
law distribution. As before, say N (γ) = Cγ −p for γ between γmin and γmax , and 0 otherwise.
Then it turns out that, for a stretch of the spectrum, a blackbody distribution is scattered
into a power law distribution in which
dE/(dV dt dǫ) ∝ ǫ−(p−1)/2 (12)
where ǫ is the energy of the photon as measured in the lab frame. Ask class: can this power
law distribution continue indefinitely to arbitrarily low or high energies? No! Integrated
from 0 to ∞, all power laws diverge. This means that there must be a lower cutoff or an
upper cutoff, and usually both. This is an important principle. Another important principle
to mention in passing is that power laws arise a lot in problems like this, and the deep basic
reason is that a power law is what you get if there is no scale in the problem. I’ll just state
that, but you may want to think about it. If there is no “important” energy, for example,
then a power law is the most general thing that doesn’t bend or otherwise have a special
place in it. One example: think of a blackbody spectrum. Over the whole spectrum, the
thermal energy kT is important, so it isn’t a power law. But in the very low-energy portion,
the exact temperature doesn’t change the form of the spectrum, so in the Rayleigh-Jeans
limit the spectrum is a power law. Just some food for thought. This might suggest (and in
fact it’s true) that a very general spectrum produced by Comptonization of, say, a Planck
spectrum of temperature T1 by thermal electrons of temperature T2 is likely to be a power
law for photon energies kT1 ≪ h̄ω ≪ kT2 . This is one reason why often a high-energy
source will have a near power-law spectrum.
The slope of the power law depends on the details of the scattering interaction, which
brings us to our final point. Suppose you have photons in a medium involving high-energy
electrons, and you’d like to know what happens after repeated scatterings. One useful
quantity to know is the typical change in energy of a photon per scattering. If the electrons
are in a thermal distribution of temperature T and are nonrelativistic, then the change in
energy is
(∆ǫ)NR = (ǫ/mc2 )(4kT − ǫ) . (13)
This makes sense: if the photons are hotter than the electrons, the photons lose energy on
average; if cooler, they gain energy on average. In the relativistic limit,
(∆ǫ)R ∼ (4/3)γ 2 ǫ , (14)
where the 4/3 comes from angle averaging. If we think of a situation in which the electron
temperature is much larger than the temperature of the input photons, then we can define
the “Compton y parameter”, which is the typical fractional gain in energy per scattering
times the typical number of scatterings (in other words, it’s the typical fractional change
in energy by the time the photon leaves the region). From our work on random walks we
know that if the electron scattering optical depth is τes , an approximation to the average
2
number of scatterings is ∼ Max(τes , τes ), where the first one is appropriate for optically thin
and the second for optically thick. The y parameter in the nonrelativistic and relativistic
limits then becomes
yNR = (4kT /mc2 )Max(τes , τes 2
)
2 2 2 (15)
yR = (4kT /mc ) Max(τes , τes ) .
The slope of the power-law portion of the spectrum depends on y:
Iν ∝ ν 3+m q
(16)
m = −3/2 ± 9/4 + 4/y ,
where the + root is appropriate for y ≫ 1 and the - root for y ≪ 1; if y ∼ 1, a superposition
is required and no power law exists. Note that even if y is determined from the spectrum,
we don’t know T and τes independently, just a combination. To break the degeneracy, the
rollover in the high energy end of the spectrum must be measured.