Kevin Zhou Physics Olympiad Handouts

Relativity III: Fields

Relativity in electromagnetism is covered in chapter 5 of Purcell and then sprinkled in throughout
the rest of the book, notably in sections 6.7 and 9.7, and appendix H. For a more advanced
discussion, see chapter 3 of Schutz for tensors, and chapter 12 of Griffiths and chapters I-34, II-13,
and II-25 through II-28 of the Feynman lectures for relativistic electromagnetism. For a brief taste
of general relativity, see chapter 14 of Morin, and chapter II-42 of the Feynman lectures. For a
great, accessible introduction to tests of general relativity, see Was Einstein Right? by Will. There
is a total of 77 points.

1 Electromagnetic Field Transformations

Idea 1: Field Transformations
If the electromagnetic field is (E, B) in one reference frame, then in a reference frame moving
with velocity v with respect to this frame, the components of the field parallel to v are

E∥′ = E∥ , B∥′ = B∥

while the components perpendicular are

 v 
E′⊥ = γ(E⊥ + v × B⊥ ), B′⊥ = γ B⊥ − 2 × E⊥ .
c
As alluded to in R2, this is the transformation rule for the components of a rank 2
antisymmetric tensor.

Under these transformations, Maxwell’s equations remain true in all inertial frames, and the
Lorentz force transforms properly as well. Furthermore, a Lorentz transformation does not
change the total amount of charge in a system, where total charge is defined by Gauss’s law
via the electric flux through a surface containing the system.

Remark
There are many ways of deriving the field transformations. The tensor method alluded to
above is the mathematically cleanest, but the conceptually clearest is to think about how
some simple setups must Lorentz transform, if Maxwell’s equations are to remain true. For
example, boosting a capacitor increases the charge density on the plates because of length
contraction, which is why E′⊥ contains γE⊥ . (Further examples are given in chapter 5 of
Purcell, which is essential reading for this section.) Another method is to demand that the
Lorentz force obeys the transformation of three-force derived in R2.

[4] Problem 1. Basic facts about the electric and magnetic fields of a moving charge.
(a) Show that the field of a point charge q at the origin moving with constant velocity v is
q 1 − v2
E= r̂
4πϵ0 r2 (1 − v 2 sin2 θ)3/2
in units where c = 1, and θ is the angle from v. In particular, the field is still radial.

1
Kevin Zhou Physics Olympiad Handouts

(b) Verify that the charge of this moving charge is still q. It may be useful to consult the integral
table in appendix K of Purcell.

(c) Argue that the magnetic field of this point charge must be exactly
v
B= × E.
c2

(d) Verify that the previous result is correct in nonrelativistic electromagnetism (i.e. using Coulomb’s
law and the Biot–Savart law).
The result of part (a), first found by Heaviside in 1888, implies that the field lines of a moving
charge “length contract” like they were rigid rods. In fact, since this result came first historiaclly, it
was one of the inspirations for length contraction in the first place! Since it’s very hard to measure
the Coulomb field of a relativistic electron, this prediction was first directly verified in 2022.
Solution. For concreteness, let the charge be moving along the z direction, and let the primed
frame be the rest frame of the charge.
(a) By rotational symmetry, it suffices to show this holds in the xz plane. Consider the point

(x, z) = (r sin θ, r cos θ)

at time t = 0. Lorentz transforming to the primed frame, this point corresponds to

(x′ , z ′ ) = (r sin θ, γr cos θ).

By Coulomb’s law, the electric field at that point is

q x′ q z′
Ex′ = , Ez′ = .
4πϵ0 ((x′ )2 + (z ′ )2 )3/2 4πϵ0 ((x′ )2 + (z ′ )2 )3/2
Applying the field transformations, the field in the original frame is
q γr sin θ q γr cos θ
Ex = γEx′ = 2 , Ez = Ez′ = 2 .
4πϵ0 r (sin θ + γ 2 cos2 θ)3/2
3 4πϵ0 r (sin θ + γ 2 cos2 θ)3/2
3

This is a radial field, as desired, and swapping the γ’s for v’s gives the desired result.
There’s a simple explanation for why the field is radial. The fact that the field is radial in
the charge’s rest frame just means z/x = Ez /Ex . When we Lorentz transform, z is Lorentz
contracted by a factor of γ, while Ex is enhanced by a factor of γ, so we still have the equality
of ratios z ′ /x′ = Ez′ /Ex′ . This should make it clear that the fact that the field stays radial
here is a fortuitous coincidence, since xµ and (E, B) transform totally differently in general.

(b) The electric flux through a unit sphere surrounding the charge must still be q/ϵ0 , so we need
to show that
1 − v2
Z
1
1= dΩ.
4π (1 − v 2 sin2 θ)3/2
Setting up spherical coordinates as usual, the dϕ integral gives 2π, leaving
π
1 − v2 π 1 − v2 − cos θ
Z
sin θ dθ
2 2 3/2
= p
2 0 (1 − v sin θ) 2 (1 − v 2 ) 1 − v 2 sin2 θ 0

where we used equation (K.15) from Purcell. Plugging in the limits gives 1 as desired.

2
 Kevin Zhou Physics Olympiad Handouts

(c) We know B′ = 0, and using the field transformations immediately gives the result.

(d) Applying Coulomb’s law, we have

1 qv µ0 qv
2
v × E = (µ0 ϵ0 ) 2
ẑ × r̂ = θ̂
c 4πϵ0 r 4π r2
which is precisely what the Biot–Savart law predicts.

[3] Problem 2 (Purcell 5.29). Two protons are moving antiparallel to each other, along lines separated
by a distance r, with the same speed v in the lab frame, as shown.

Consider the moment the protons are a distance r apart.

(a) Show that the three-force experienced by each proton due to the electric field of the other is

γe2
F = .
4πϵ0 r2

(b) Compute the three-force experienced by one of the protons by transforming to its rest frame,
computing the force there, then transforming back to the lab frame. In particular, show that
this is not equal to the result of part (a).

(c) Show that the discrepancy is resolved if the magnetic three-force is also included.

Recall from R2 that the Lorentz three-force is F = q(E + v × B). You will also have to use the
three-force transformation laws you derived there.

Solution. (a) Using part (a) of problem 1 gives

e2 1 − v2 γe2
eE = =
4πϵ0 r2 (1 − v 2 )3/2 4πϵ0 r2

as desired.

(b) In the rest frame of one proton, we have

e2 1 − v ′2 2v
F ′ = eE ′ = , v′ = .
4πϵ0 r2 (1 − v ′2 )3/2 1 + v2

Simplifying a bit yields

e2 1 + v 2
F′ = .
4πϵ0 r2 1 − v 2

3
 Kevin Zhou Physics Olympiad Handouts

Using our result from R2 that transverse forces are redshifted,

F′ γe2
F = = (1 + v 2 ).
γ 4πϵ0 r2
This is more than the result we found in part (a).

(c) The extra three-force due to the magnetic field adds to the electric force,

γe2
evB = ev 2 E = v2
4πϵ0 r2

01mƒ
where we used part (c) of problem 1. This is exactly the missing piece.

[3] Problem 3. USAPhO 2014, problem B2. This isn’t the clearest of problems, but it introduces
and justifies the Galilean field transformations we first saw in E4.

[3] Problem 4 (Purcell 5.30). Consider an infinite wire oriented along x̂ with linear charge density λ
and current I. Show that under a Lorentz boost along x̂, (λ, I) transforms like (ct, x).

Solution. This can get kind of complicated if you think about a completely general set of charges
in the wire. On the other hand, we already know that the way electromagnetic fields transform
doesn’t depend on what makes the fields, and we know that Maxwell’s equations relate the fields
to the charge density and current. Thus, the transformation of (λ, I) shouldn’t depend on precisely
what’s responsible for the λ and I, so we can take a concrete choice that’s easy to work with.
Specifically, let’s suppose the wire is built entirely out of point charges of linear number density
n and charge q moving with velocity ux̂. Then we have

λ = nq, I = nqu.

Now we move to a frame moving to the right with speed v. In this frame, the charges are moving
with speed
u−v
u′ =
1 − uv
and, by applying length contraction, the linear number density is
s
γu′ 1 − u2
n′ = n= n = (1 − uv)γv n.
γu 1 − u′ 2

Putting things together, we have

λ′ = n′ q = (1 − uv)γv nq, I ′ = n′ qu′ = (u − v)γv nq.

Writing this in terms of the unprimed quantities, we conclude

λ′ = γv (λ − vI), I ′ = γv (I − vλ)

which is precisely the desired result.

[2] Problem 5 (Purcell 6.22). A neutral wire carries current I. A stationary charge q is nearby; the
Lorentz force on this charge is zero. Verify this remains true in a frame moving parallel to the wire
with velocity v, by using the Lorentz transformations of the fields.

4
 Kevin Zhou Physics Olympiad Handouts

Solution. Applying the transformations of problem 4, in that frame we have

λ′ = −γv vI, I ′ = γv I.

The electric and magnetic forces are

γv vIq µ0 γv vIq
qE = , qvB =
2πrϵ0 2πr
which balance because µ0 ϵ0 = 1, in our units where c = 1.

[3] Problem 6 (Purcell 6.69). Two very long sticks each have uniform linear proper charge density λ.
One stick is stationary in the lab frame, while the other moves to the left with speed v, as shown.

They are 2r apart, and a stationary point charge q lies midway between them. Find the Lorentz
three-force on the charge in the lab frame, and also in the frame of the bottom stick, and verify the
forces relate properly.

Solution. Let upward-pointing forces be denoted with a positive sign. In the lab frame, we only
have an electric force. The charge on the bottom stick is length contracted, so
 
γλ λ qλ
F =q − = (γ − 1).
2πϵ0 r 2πϵ0 r 2πϵ0 r

In the frame of the bottom stick, the bottom stick has charge density λ, while the top stick has

λ′ = γλ, I ′ = γvλ

where I ′ is directed to the right. Now the charge experiences both an electric and a magnetic force.
The electric force is
qλ
FE′ = − (γ − 1)
2πϵ0 r
by the same logic as in the lab frame. The magnetic force is

µ0 γv 2 qλ qλ
FB′ = qvB = = (γv 2 ).
2πr 2πϵ0 r
The sum of the two is
 
′ qλ qλ 1 F
F = (γv 2 − γ + 1) = 1− =
2πϵ0 r 2πϵ0 r γ γ

exactly as expected.

[3] Problem 7. The vectors E and B cannot go into four-vectors, as they transform among each other,
but rather fit together into an antisymmetric rank two tensor. As a result, there is a different set
of associated invariant quantities.

5
Kevin Zhou Physics Olympiad Handouts

(a) Show that under the relativistic field transformations, the quantities E · B and E 2 − B 2 are
both invariant. (Hint: this can be done using vector notation, using E⊥ · E∥ = B⊥ · B∥ = 0.)
These are the two basic invariants, out of which all other invariants can be constructed.

(b) Suppose that in an inertial frame, E is zero at a given point and B is nonzero. Is it possible
to find an inertial frame where B is zero at that point?

(c) Recall from E7 that, in units where ϵ0 = µ0 = 1, the energy density of the electromagnetic
field is E = E 2 /2 + B 2 /2, and the Poynting vector is S = E × B. Show that E 2 − |S|2 is
invariant. (Hint: don’t use the field transformations for this part.)

Solution. (a) Setting c = 1 for convenience as usual, we have

E′ · B′ = E′⊥ · B′⊥ + E∥′ B∥′ = γ 2 (E⊥ + v × B⊥ ) · (B⊥ − v × E⊥ ) + E∥ B∥ .

The cross terms vanish (try an explicit example if you don’t see why), which leaves

E′ · B′ = γ 2 (1 − v 2 )E⊥ · E⊥ + E∥ B∥ = E · B

as desired. As for the other quantity, we have

2 2
E ′2 − B ′2 = E ′ ∥ + E′⊥ · E′⊥ − B ′ ∥ − B′⊥ · B′⊥
= E∥2 + γ 2 (E⊥ + v × B⊥ )2 − B∥2 − γ 2 (B⊥ − v × E⊥ )2
= E∥2 + γ 2 (E⊥ · E⊥ + v 2 B⊥ · B⊥ ) − B∥2 − γ 2 (B⊥ · B⊥ + v 2 E⊥ · E⊥ )
= E∥2 + E⊥ · E⊥ − B∥2 − B⊥ · B⊥
= E2 − B2

as desired, where in the third equality the cross-terms canceled.

(b) This is impossible, because then E 2 − B 2 would have different signs in the two frames.

(c) If we plugged in the field transformations, the algebra would get extremely messy. Instead,
we use the hint that this invariant can be constructed out of the ones found in part (a), so

1 2 1 2 2
 
2 2
E − |S| = E + B − |E × B|2
2 2
1 2 1 2 2
 
= E + B + (E · B)2 − E 2 B 2
2 2
1 2 1 2 2
 
= E − B + (E · B)2 .
2 2

This is constructed out of the invariants in part (a), so it is invariant as well.

Remark: Is Magnetism Real?

Purcell’s electromagnetism textbook is exceptional because it shows that a force like

magnetism must exist, if one believes Coulomb’s law and relativity. The idea is simple. We
know how forces transform between frames, and given some reasonable assumptions, can
also deduce how electric fields transform between frames. If electric fields were all there

6
Kevin Zhou Physics Olympiad Handouts

were, then electric forces would have to transform just like three-forces, but they don’t. So
there must be some other force to make up the difference, and it turns out to be precisely
the magnetic force. We saw an example of this in problem 2.

It is important not to misunderstand this beautiful idea. Many people, upon reading Purcell,
believe that magnetism “doesn’t exist” because it’s all “just electric fields”. Sometimes
people even say that magnetic forces are a “mistake” caused by “forgetting about” relativistic
corrections. This is all totally backwards. Sometimes time dilation in one frame can be
explained in terms of length contraction in another, but that doesn’t mean that length
contraction doesn’t exist, or is a mistake – it’s perfectly real in that particular frame. (Not
to mention that there are plenty of situations where you can’t get rid of the magnetic field
in any frame, as we saw in problem 7!)

The real lesson of relativity isn’t that magnetic fields are a mistake, it’s that electric and
magnetic fields are as intertwined as space and time, as you can see from their transformation
properties. Just as space and time combine into a four-vector, electric and magnetic fields
combine, in an equal footing, into the electromagnetic field tensor.

Remark: Electromagnetism in Covariant Form

Problem 4 is a first step to showing that J µ = (ρ, J) is a four-vector, where ρ is the charge
density and J is the current density. Note that the continuity equation for charge, as
mentioned in T2, can be simply written in four-vector notation as

∂µ J µ = 0.

As another example, you can show that the four-current of a single charged particle q is
J µ = quµ . We can go even further and write the whole of electromagnetism in terms of
four-vectors and tensors. Maxwell’s equations can be written as

∂µ F µν = J ν .

The invariant quantities found in problem 7 can be written in terms of the field strength
tensor as Fµν F µν and ϵµνρσ F µν F ρσ where ϵµνρσ is the Levi–Civita symbol. These are the
only two ways to “contract all the indices” to get a scalar.

Remark: Elegant Notation

Sometimes people dislike the index notation above because of all the little Greek letters
floating around. If you only want to deal with vectors, vector notation is often better. It
hides all the indices, at the cost of requiring you to introduce special symbols like · and × to
specify the vector operations you want to do. The reason we don’t use a vector-like notation
for tensors is because there are too many operations you can do with them (e.g. “contract the
3rd index of a rank 4 tensor with the 1st index of a rank 2 tensor”) to define separate symbols
for each one; indices are just more efficient. On the other hand, if you only work with totally
antisymmetric tensors, then there are only a few possible operations, and one can use the

7
 Kevin Zhou Physics Olympiad Handouts

elegant, index-free “differential form” notation. In this notation, Maxwell’s equations are

d⋆F =J

where d is called the exterior derivative, ⋆ is the Hodge dual, and the fact that the electro-
magnetic fields are derivatives of potentials is expressed as

F = dA.

So is this the best, most true formulation of Maxwell’s equations? Well, as Feynman once
pointed out, you can easily do better. For example, you can define the “unworldliness”

U = |F − ma|2 + (∇ · E − ρ/ϵ0 )2 + . . . .

Then all physical laws can be expressed in terms of the amazingly simple equation

U = 0.

But this doesn’t actually help, because to use the equation for anything, you need to plug in
the definition of U , and then you’re back to where you were before. In general, more elegant
notation is often more brittle: it only works well in a smaller set of situations. (For example,
with differential form notation, you just can’t write down the stress-energy tensor of the
electromagnetic field, because that’s symmetric rather than antisymmetric.) Index notation
is great because it works as long as indices are contracted in pairs, which holds as long as
you’re dealing with laws that are independent of coordinate system. In general, there’s no
need to be ideological about notation; it’s just a tool, and we should use the best tool for each
job. If anyone tells you that their preferred notation for vectors or tensors will revolutionize
physics, keep your hand on your wallet.

[4] Problem 8. Consider an electromagnetic wave of the form

E(z, t) = E0 cos(kz − ωt)x̂, B(z, t) = B0 cos(kz − ωt)ŷ.

As usual, you may work in units where c = 1.

(a) What do Maxwell’s equations imply about the relation between E0 and B0 , and k and ω?

(b) Now consider a frame moving with velocity v along the ẑ direction. Show that the electromag-
netic wave continues to have the same basic functional form for E′ (z ′ , t′ ) and B′ (z ′ , t′ ), but
with new parameters E0′ , B0′ , k ′ , and ω ′ . Using these results, show that the energy density of
the wave is smaller by a factor of (1 − v)/(1 + v).

(c) The energy of a photon in an electromagnetic wave of frequency ω is E = ℏω. Show that for
a finite-sized electromagnetic wave, the initial and boosted frames agree on the number of
photons. This was one of the hints Einstein used to conclude light was made of photons.

(d) Now consider another question Einstein pondered: what does the light wave look like if we
try to “catch up” with it, taking v → c? Is this consistent with the invariants of problem 7?
Solution. (a) From E7, we know that E0 = B0 and k = ω.

8
Kevin Zhou Physics Olympiad Handouts

(b) The electromagnetic field only has perpendicular components. Using the field transformations,

E′ (z ′ , t′ ) = γ(E0 cos(kz − ωt) − vB0 cos(kz − ωt))x̂.

In units where c = 1, we have E0 = B0 for an electromagnetic wave, so this simplifies to

E′ (z ′ , t′ ) = γ(1 − v)E0 cos(kz − ωt)x̂.

Repeating the reasoning for the magnetic field, we conclude

E0′ = γ(1 − v)E0 , B0′ = γ(1 − v)B0

which still obeys E0′ = B0′ as expected. Thus, the energy density is reduced by a factor of

1−v
γ 2 (1 − v)2 =
1+v
as stated. To find k ′ and ω ′ , we can simply apply the Lorentz transformations,

kz − ωt = k(γ(z ′ + vt′ )) − ω(γ(t′ + vz ′ )) = γ(k − ωv)z ′ − γ(ω − kv)t′ .

This indicates that E′ (z ′ , t′ ) is still a plane wave proportional to cos(k ′ z ′ − ω ′ t′ ), where

k ′ = γ(k − ωv), ω ′ = γ(ω − kv)

which is of course just the statement that (ω, k) is a four-vector, derived in R1. Using the
fact that ω = k, we conclude
r
′ ′ 1−v
k = ω = γ(1 − v)ω = ω
1+v
which is of course just the usual Doppler shift.

(c) The number of photons is the ratio of the total energy in the wave
pto the energy of each
photon. Since the energy of each photon is reduced by a factor of (1 − v)/(1 + v) in the
boosted frame, we need to show that the total energy of the wave is reduced by the same
factor. This results from the combination of two effects.
First, we know the energy density
p is reduced by the factor (1−v)/(1+v). Second, the wavenum-
ber
p is reduced by a factor of (1 − v)/(1 + v), which means the wavelength is increased by
(1 + v)/(1 − v). Since the number of wavelengths contained in the wave p is the same in
every reference frame, this means the volume of the wave is increased by (1 + v)/(1 − v).
Multiplying these factors gives the desired result.

(d) In this case we have E0′ , B0′ , ω ′ , k ′ → 0, so the light wave disappears! That is, you can never
“catch up” to a light wave. This result is completely compatible with the invariants from part
(a) of problem 7, which both vanish for a plane electromagnetic wave. The invariant in part
(c) vanishes as well, since E = |S| for a plane wave.

9
 Kevin Zhou Physics Olympiad Handouts

Idea 2
If a uniformly moving point charge suddenly stops moving, then the field outside a spherical
shell, centered at the charge when it stopped moving, expanding at speed c, is precisely that
calculated in problem 1. The same occurs if the point charge suddenly changes its velocity;
information about the change only propagates at c.

[1] Problem 9 (Purcell 5.18). In the figure below, you see an electron at time t = 0 and the associated
electric field at that instant.

(a) Describe what has been going on, as quantitatively as you can.
(b) Where was the electron at the time t = −0.75 ns?

Solution. (a) Clearly, the particle isn’t moving at t = 0. Since there’s a kink in the field lines
at r = 15 cm, it must have quickly stopped at t = −r/c = −0.5 ns, since the speed of light is
c = 30 cm/ns. We also see that the field lines outside this shell are straight, and point towards
the location x = 12 cm. This implies that shortly before the charge stopped, it was moving
with constant velocity v = |x/t| = 24 cm/ns = 0.8c.
(b) By combining the results from (a), it must have been at x = −(24 cm/ns)(0.25 ns) = −6 cm.
[2] Problem 10 (Purcell 5.19). The figure below shows two highly relativistic particles with opposite
charge approaching the origin.

10
 Kevin Zhou Physics Olympiad Handouts

They collide at the origin at time t = 0 and remain there as a neutral entity. Sketch the field lines
at some time t > 0.
Solution. There are a lot of wackos out there that misunderstand relativity and then conclude
that physics must be all wrong. For example, on multiple occasions I have seen claims that particle
annihilation is impossible, because that would imply the electric field has to “instantly vanish”.
Purcell presumably ran into these people too, which is why this is a problem in his textbook.

Before the charges collide, we have a kind of distorted dipole field. After they collide, we still
have a dipole field at r > ct, with the positive charge on the right and the negative charge on the
left. For r < ct the field is zero, and at the shell r ≈ ct there is a thin shell that connects up the
field lines. This transverse pulse of radiation steadily moves outward over time.
[3] Problem 11. Work through the derivation of the Larmor formula in Appendix H of Purcell.
[3] Problem 12 (Purcell H.4). The Larmor formula only applies to particles moving nonrelativistically.
To get a result valid for faster particles, we can simply transform into an inertial frame F ′ where
the particle is nonrelativistic, apply the Larmor formula, then transformed back to the lab frame.
(a) Consider an relativistic electron moving perpendicularly to a magnetic field B. Defining the
′ , the power in a frame instantaneously comoving
radiation power as Prad = dE/dt, find Prad
with the electron.
′ , and conclude that
(b) Argue that in this context, Prad = Prad

γ 2 v 2 e4 B 2
Prad = .
6πϵ0 m2 c3
Thus, the power increases rapidly as v → c. Incidentally, a “relativistic” way to write the
general result is
q2 1 dpµ 2
 
Prad =
6ϵ0 c3 m dτ
which clearly reduces to the Larmor formula in the nonrelativistic limit.

(c) This radiation is also called synchrotron radiation. Qualitatively, how does its angular distri-
bution differ from radiation from an accelerating nonrelativistic charge?

11
Kevin Zhou Physics Olympiad Handouts

Solution. (a) In the frame comoving with the electron, it’s not relativistic, so we can just apply
the Larmor formula,
′ e2 a′ 2
Prad = .
6πϵ0 c3
In this frame, the only force is the electric force, so
eE ′ eγvB
a′ = = .
m m
Putting it together, we conclude
′ e4 γ 2 v 2 B 2
Prad = .
6πϵ0 m2 c3
(b) We have Prad′ = dE ′ /dt′ . Now, in the primed frame, the electron is just accelerating trans-
versely, with no component along the unprimed frame’s v. Thus, when we Lorentz transform
back to the unprimed frame, we simply get dE = γdE ′ and dt = γdt′ . The γ factors cancel
out, giving the desired result.

(c) In the primed frame, the radiation power comes out with a wide angular distribution, but
none of it comes out along the direction of motion of the charge, and most of it comes out
roughly transverse to the motion. But when we boost back to the original frame, where
the charge is moving very quickly, the radiation’s direction gets a big component along the
charge’s direction of motion. Thus, almost all the radiation is “beamed” in a narrow cone
along the charge’s motion (as we saw in R1), though there still is zero radiation intensity
exactly along the charge’s direction.

Remark: Gravitoelectromagnetism

As mentioned in E1, there’s a close analogy between electrostatic fields, which are sourced by
charge density ρe , and gravitational fields, which are sourced by energy density ρ. Therefore,
if you apply the analogy and run the same arguments as in Purcell, you would expect there
to be a “gravitomagnetic” field, which is sourced by momentum density J = ρv. That’s
indeed correct! In the theory of gravitoelectromagnetism, the force on a point mass is

F = m(E + 4v × B)

where the gravitoelectric and gravitomagnetic fields E and B satisfy

∇ · E = 4πGρ, ∇ · B = 0, ∇ × E = −Ḃ, ∇ × B = 4πGJ + Ė.

From this you can draw some interesting conclusions. For example:

• Two masses moving parallel to each other will have an extra attraction due to the
gravitomagnetic force.

• A rotating object will produce a gravitomagnetic field which can cause gyroscopes to
precess; this is called the Lense–Thirring, or frame dragging effect, which has been
measured by satellites such as Gravity Probe B. (There is also a significantly larger
“geodetic” effect caused by the curvature of spacetime around the Earth, but this isn’t
captured within gravitoelectromagnetism.)

12
Kevin Zhou Physics Olympiad Handouts

• A mass at rest, inside a cylinder which suddenly starts to rotate, will pick up a small
angular velocity in the same direction due to the induction of a gravitoelectric field.

• Gravitational waves are generated by accelerating masses and carry energy, just like
electromagnetic radiation.

Now you might be puzzled by two things: first, how does gravitoelectromagnetism relate
to general relativity, and second, why is there an extra 4 in one of the equations above?
Well, the truth is that Purcell’s arguments
R don’t really work for gravity. These arguments
crucially depend on electric charge Q = ρe dx being Lorentz invariant, which in our more
sophisticated languageR was necessary to ensure j µ = (ρe , ρe v) is a four-vector. However,
the total energy E = ρ dx is not Lorentz invariant – instead it’s itself a component of a
four-vector. Thus, (ρ, ρv) isn’t a four-vector, so none of the arguments really work: the
theory of gravitoelectromagnetism is just not Lorentz invariant at all.

Instead, gravitoelectromagnetism is properly derived as a limiting case of general relativity,

valid when all the masses involved are moving slowly, v ≪ c. The fact that general relativity
is a theory of a rank 2 tensor field, the metric gµν , is responsible for the extra factors of 2
above. Even though it’s only approximately true, gravitoelectromagnetism is a very useful
tool for analyzing precision tests of general relativity, since it’s much easier to calculate with.

On the other hand, there’s also a lot of nonsense written about it by people who don’t
understand it. For example, a lot of internet luminaries are certain that it can be used to
replace dark matter, even though, using just the basic equations above, you can see that the
gravitomagnetic force is (v/c)2 times smaller than the usual gravitational force. That makes
it about 106 times too small to explain the anomalous rotation of galaxies.

In fact, now is a good time to issue a warning. There’s a concept called Lizardman’s constant,
which is the fact that in any survey, no matter how it’s designed, about 3% of the answers
will be complete nonsense. 3% of people will enthusiastically tell you that they were born
on Mars, that the Moon landing was faked, or that the Earth is run by lizardmen. That’s
because there’s an irreducible fraction of people that are mistaken, crazy, or just plain trolling.

The internet is a wonderful place to learn introductory physics, because it’s relatively straight-
forward, so the sincere and competent outnumber the crazy. But as you go to more advanced
topics, the fraction of people who know what’s going on, and who have the time and energy
to tell you, rapidly drops, while the 3% stays just as large. Now that you’re at the end of
this curriculum, you’re also at the point where the majority of internet commentators on the
topics you’re learning are completely wrong. Fortunately, you’re also learning what sources
are good, and developing the knowledge needed to check things for yourself. As you continue
learning tougher subjects, these skills will keep you on the right track.

2 Charges in Fields
Now we consider some problems in the spirit of E4, using more advanced tools.

13
 Kevin Zhou Physics Olympiad Handouts

[3] Problem 13 (Purcell 5.24). In the rest frame of a particle with charge q, another particle with
charge q is approaching with relativistic velocity v. Assume that both particles are extremely
massive, and hence their velocities are nearly constant. The second particle passes a minimum
distance b from the first.

(a) Show that the impulse acquired by each particle is perpendicular to v with magnitude
q 2 /2πϵ0 vb. (Hint: you can avoid doing a nasty integral by using Gauss’s law.)

(b) If the particles have mass m, roughly how large does m have to be for the above result to be
a good approximation?

Solution. (a) Let the stationary particle be at the origin, and let the velocity of the moving
particle be along the x-axis. Then the impulse experienced by the moving particle is

q ∞
Z Z
J = F dt = Ez (x, b) dx.
v −∞

On the other hand, suppose we consider the electric flux through an infinite cylinder of radius
b, oriented along the x-axis. Then
Z Z ∞
Φ = E · dS = 2πb Ez (x, b) dx.
−∞

Therefore, we conclude
q Φ q2
J= = .
v 2πb 2πϵ0 vb
The great thing about this derivation is that exactly the same reasoning applies to the impulse
experienced by the stationary charge. It can be written as a similar integral, except that
Ez (x, b) is now the electric field of a moving charge. But the impulse is the exactly the same in
magnitude because the Gauss’s law argument still works. That’s good to know, as it ensures
momentum is conserved.

(b) First off, we’ve ignored magnetic forces, even though the particles will pick up transverse
velocity and hence begin to feel them. Since magnetic forces are small compared to electric
forces when the (tranverse) speeds of the charges are nonrelativistic, we should require

J ≪ mc

to be safe. Now, given this assumption, we can focus on the electric forces. Here we have
assumed the charges don’t move transversely during the whole process. The final transverse
velocity is of order J/m, and the total time the interaction takes is of order b/v, so we need

J b
≪b
mv
which simplifies to
J ≪ mv.
Since this is strictly stronger than the other condition, this is the only one we really need. In
other words, this kind of calculation only works if the transverse speed J/m the charges pick
up is small compared to the original speed, i.e. if the angular deflection is small.

14
 Kevin Zhou Physics Olympiad Handouts

[2] Problem 14. When we consider conservation of energy for a particle of charge q, we always
include, along with the kinetic energy mv 2 /2, the potential energy qϕ. Similarly, when we consider
conservation of momentum, we must consider not only the ordinary momentum M v, but also the
“potential momentum” qA. The sum of the two is called the canonical momentum.
Conservation laws are associated with symmetries. When the ϕ and A a particle moves in are
time-independent, the energy M v 2 /2 + qϕ is conserved. And when they are time-independent and
space-independent, the canonical momentum M v + qA is conserved. Unfortunately, this is an
extremely restrictive condition which forces the electric and magnetic fields to be zero, rendering
the full conservation law basically useless in practice. However, in some cases a single component
of this momentum can be conserved, which can be useful.

(a) Check that a vector potential appropriate for a constant field B in the −ẑ direction is

B
A= (yx̂ − xŷ).
2

(b) Now consider a charge tied to the end of a string, executing horizontal uniform circular motion
about the origin with radius r. Suppose a magnetic field −Bẑ is turned on. Show that

L = r × (M v + qA)

is conserved. This is the conserved quantity due to rotational symmetry about the z-axis.

Solution. (a) By definition, we need B = ∇ × A, and (letting ∂i = ∂/∂xi )

B
∇×A= (−∂y y − ∂x x)ẑ = −Bẑ
2
as desired.

(b) When the field is turned on, a tangential electric field is produced,

dΦB dB
(2πr)E = − = −πr2
dt dt
which implies an impulse Z
qrB
∆(mv) = q E dt = −
2
directed tangentially. On the other hand, we have qA = qBr/2 also directed tangentially, so
the tangential component of qA decreases by the same amount the tangential component of
M v increases. Thus, L is conserved.

[3] Problem 15 (Cahn). A charged particle is orbiting in a uniform magnetic field of magnitude B0
in a circular orbit of radius R0 . Assume the particle moves much slower than the speed of light, and
that all fields are cylindrically symmetric about the original axis of rotation of the particle. (For
concreteness, the particle could originally be orbiting about the axis of symmetry of a big solenoid.)

(a) The field is slowly changed to B1 . What is the new radius R1 of the orbit? (Hint: you can
solve this problem in many ways. For example, you can directly use the Lorentz force, or you
can use conservation of the quantity in problem 14, or you can use the adiabatic theorem,
where the momentum is mv + qA.)

15
Kevin Zhou Physics Olympiad Handouts

(b) The field is suddenly changed back to B0 . What is the final radius R2 ?
For a more challenging problem along these lines, see Physics Cup 2017, problem 3.
Solution. (a) First we’ll solve it directly. Let v be the speed of the particle. We know that
qB
v = ωr, ω= .
m
The only thing that changes the speed is the tangential electric field, and by Faraday’s law,
1
E= πr2 Ḃ
2πr
which implies that
q r
v̇ = Ḃ.
m2
Using the product rule to evaluate v̇ and simplifying gives
ṙ 1 Ḃ
+ =0 ⇒ Br2 = const.
r 2B
p
The magnetic flux is conserved, and the final radius of the orbit is R1 = R0 B0 /B1 .
We can also solve it by conserving Lz , where by our work in problem 14,
qBr2
Lz = mvr − .
2
Since mv 2 /r = qvB, the two terms here are actually proportional, so it is equivalent to say
that qBr2 is conserved, i.e. the magnetic flux is conserved, just as in the first solution.
Finally, we can use the adiabatic theorem. The adiabatic invariant is
I I
I = m v · dx + q A · dx = m(2πr)v + qΦB = 2πrmv − πqr2 B.

Note that a minus sign appears in the last step because the magnetic flux is negative, as can
be verified by drawing the setup and using the right-hand rule. Up to a constant, this is the
same thing as Lz , so we get the same conclusion.
Incidentally, since the two terms in Lz are proportional to each other, you could just as well
say that mvr is conserved. But this is a lucky coincidence specific to this setup. For a general
field profile B(r) ẑ which is still symmetric about the z-axis, Lz = mvr − qΦB /2 will remain
conserved, but mvr won’t be. If you recall from E4, this is the tricky conserved quantity you
used to solve IPhO 1996, problem 2.
(b) As we found above, we have
q r
v̇ = Ḃ
m2
but in this case the decrease is sudden, so we can treat r = R1 as a constant, so
qmR1
∆v = (B0 − B1 ).
2
Hence the final velocity is
qR1
v= (B0 + B1 ).
2m
Since we also have v = (qB0 /m)R2 , we find
B0 + B1 B0 + B1
R2 = R1 = R0 √ .
2B0 2 B0 B1

16
 Kevin Zhou Physics Olympiad Handouts

[5] Problem 16. 01T†

APhO 2001, problem 2. This tough, rather mathematical problem covers

01T†
generalized momentum with vector notation, extending the results of problem 14.

[5] Problem 17. IPhO 1991, problem 2. A problem on a subtle relativistic effect.

[5] Problem 18. Physics Cup 2021, problem 1. A really tough electromagnetism question.

Solution. See the official solutions here.

3 Gravitational Fields
Idea 3
In classical mechanics, you’ve seen that a uniform gravitational field behaves a lot like the
fictitious force due to a uniform acceleration. The equivalence principle states that the two
behave exactly identically, in all possible contexts; it was one of the key ideas that led to the
development of general relativity.

[4] Problem 19. In this problem, we given one of the classic justifications for gravitational redshift,
the fact that photons redshift when moving against a gravitational field. Suppose that point B is a
height h above point A, in a gravitational field g. A set of electrons and positrons with total rest
mass M are converted into photons of frequency f at point A. The photons fly upward to point B,
where they are converted back into electrons and positrons. Assume throughout that g is small.

(a) Find the total mass M ′ at point B.

(b) Find the frequency f ′ of the photons measured at point B.

(c) Since the frequencies of photons can be used as a clock, the result of part (b) shows that
gravitational fields cause time dilation, which applies to everything, not just photons. Show
that your result in part (b) is equivalent to the statement that times are dilated by a factor
of 1 + ϕ/c2 , where ϕ is the gravitational potential and ϕ/c2 ≪ 1.

We should also be able to understand part (b) using the equivalence principle. To confirm this,
suppose that two observers C and D begin at rest, with D a distance h to the right of C. At a
certain moment, both observers begin accelerating to the right with a small acceleration a.

(d) If C emits light of frequency f (in C’s rest frame), show that D observes light of frequency
f ′ , where f ′ matches your answer to part (b).

(e) The predicted frequency shift was observed in the 1959 Pound–Rebka experiment, where
gamma rays were transmitted from the top to the bottom of a tower of height h = 22.5 m.
What is the fractional change in energy of the photons?

(f) Gamma ray photons of energy 14 keV were used in the Pound–Rebka experiment. According
to the energy-time uncertainty principle, what is the minimum time needed to detect the
effect?

17
Kevin Zhou Physics Olympiad Handouts

Solution. (a) By conservation of energy,

M c2 = M ′ c2 + M ′ gh

from which we conclude, using the fact that g is small, that

 
′ gh
M =M 1− 2 .
c

(b) Each photon has its energy reduced by a factor of 1 − gh/c2 , and since E = hf ,
 
′ gh
f =f 1− 2 .
c

(c) If we measure time through the frequency of light, then

 
′ gh
∆t = ∆t 1 + 2
c

again to lowest order in g. This is the desired result, since ϕ = gh. It implies that higher
clocks tick faster.

(d) Since the acceleration is small, it takes about a time h/c for the light to arrive at D. By this
time, D has picked up a velocity ah/c, so by the Doppler shift,
s
1 − ah/c2
 
′ ′ ah
f =f =f 1− 2
1 + ah/c2 c

where we again work to lowest order in a.

(e) Plugging in the numbers, gh/c2 = 2.5 × 10−15 .

(f) The change in energy is ∆E = h∆f , and the uncertainty principle says we need time
h h
∆t ≳ ∼ −15
= 10−4 s.
∆E (2.5 × 10 )(14 keV)
In the real experiment, Pound and Rebka used two identical samples of iron as the emitted
and receiver, and vibrated one of them vertically at a few tens of Hz. Whenever the relative
velocity was just enough to cancel out the gravitational redshift effect, absorption occurred.
Since the ∆t required was substantially lower than the period of the vibration, the vibration
didn’t mess up the experiment.

Remark
You might be a little worried that the result of part (c) above does not seem to be invariant
under a large, constant shift of ϕ, even though in Newtonian mechanics we can always do
this. In fact, in that case the same analysis is essentially valid, but the “extra” gravita-
tional time dilation is canceled out by other effects, which unfortunately can’t be explained
without full general relativity. In other words, the analysis above is only valid when ϕ is small.

18
 Kevin Zhou Physics Olympiad Handouts

If you find this confusing, you’re not alone. In 2018, there was some excitement as researchers
claimed to explain a long-standing anomaly in particle physics, making a mistake precisely
along these lines. (A rebuttal is given here.)

[3] Problem 20. In this problem we consider the effects of relativity on a clock on the surface of the
Earth, which has mass M and radius R. It rotates about its axis in time T , as measured by an
observer at infinity who is at rest relative to the center of the planet

(a) Consider a clock C that lies on the surface of the planet at a point on the equator. Compute
the time measured by the clock C after a single rotation of the planet, incorporating both
special relativity and gravitational time dilation. Which effect is bigger?

(b) Repeat part (a) for a clock C ′ on a satellite orbiting the planet, in a circular orbit a height h
above the equator.

(c) Using the numbers M = 5.97 × 1024 kg, R = 6.4 × 106 m, and h = 2 × 107 m, estimate
the difference in time elapsed per day for the two clocks, counting only the effect of special
relativity, or only the effect of gravitational time dilation.

Solution. (a) The clock C always has speed v = 2πR/T , so if we only counted time dilaton,
 2 !
p 1 2πR
TC = T 1 − (2πR/cT )2 ≈ T 1 − .
2 cT

It is also at a lower gravitational potential than a clock at infinity, so counting only gravitational
time dilation,
TC = T 1 + ∆ϕ/c2 = T 1 − GM/Rc2 .

Of course in reality both effects occur, and at leading order they just add,
 2 !
p 1 2πR GM
TC = T 1 − (2πR/cT )2 ≈ T 1 − − .
2 cT Rc2

The two effects are equal when v 2 = 2GM/R, which describes escape velocity. Since the Earth
is rotating a lot slower than that, the gravitational time dilation effect is much larger.

(b) We ′
p can just repeat the exercise, the only difference being that the clock C has speed v =
GM/(R + h), which implies
 
p
2
1 GM GM
TC = T 1 − (2πR/cT ) ≈ T 1 − − .
2 (R + h)c2 (R + h)c2

The gravitational time dilation effect is still larger, but only by a factor of 2.

(c) Plugging in the numbers, we have

 2
2πR GM GM
= 2.4 × 10−12 , = 7.0 × 10−10 , = 1.7 × 10−10 .
cT Rc2 (R + h)c2

19
 Kevin Zhou Physics Olympiad Handouts

If we just consider the time dilation effect, the time difference per day is
 !
2πR 2

T GM
− = 7 µs.
2 (R + h)c2 cT

If we consider just gravitational time dilation, the time difference per day is
 
GM GM
T − = −46 µs.
(R + h)c2 Rc2
Not only is the gravitational effect important, it’s more important than the time dilation

01hˆ
effect! Accounting for this was crucial to get GPS working.

[5] Problem 21. APhO 2014, problem 3. Gravitational fields bend light; this problem is about
the geometry of gravitational lensing. Print out the official answer sheets and record your answers

01hˆ
on them.
[5] Problem 22. IPhO 1995, problem 1. This problem is about the applications of gravitational

01@
redshift, and also serves as a nice review of R2.

[3] Problem 23. IPhO 2023, problem 2, parts C.1 through C.4. A neat problem on how the
Shapiro delay, a classic test of general relativity, can be used to measure the masses of neutron
stars.

Remark: Visualizing Relativity

You’ve probably heard that in general relativity, gravity is explained by the curvature of
spacetime. In other words, freely falling objects always move in straight lines through
spacetime; they only look like they’re accelerating downward because we are constantly being
accelerated upward. This is nicely illustrated here and explained in greater detail in this paper.

There is a common analogy for this involving picturing space as a distorted rubber sheet.
It’s a very bad analogy, because things will only accelerate towards the valleys in the sheets
if you have gravity pointing down the sheet. In other words, the analogy tries to explain
gravity by assuming you have spatial curvature and gravity. This misses the beautiful
key point of relativity, which is that the gravity can be explained by spacetime curvature alone.

The fact that freely falling objects move in straight lines means that an object sitting
on the surface of the Earth is actually being constantly accelerated. But this leads to a
common followup question: in this picture, the surfaces of America and India are constantly
accelerated in opposite directions, so why doesn’t the Earth tear itself apart? Indeed, in
special relativity this would make no sense. It’s only possible because of spacetime curvature.

This can be explained with a spatial curvature analogy. Consider two people walking
east, side by side, with one just north of the equator and the other south. In order to
stay a constant distance apart, the person walking on the north will constantly have to
bear to the right, while the person walking on the south will have to bear to the left,
because the Earth’s surface is spatially curved. Similarly, in a situation with spacetime cur-
vature, America and India need constant opposite accelerations to maintain the same distance.

20
Kevin Zhou Physics Olympiad Handouts

There’s a dramatically different way to visualize this situation called the “river model”, which
is illustrated here. The basic idea is that we think of space as a river that is constantly flowing
towards the center of the Earth. Observers in America and India constantly need to paddle
in opposite directions against the river to stay in place. This is also a good way to think
about the event horizon of a black hole, which is where the river starts to flow faster than light.

In this remark I’ve given three analogies about spacetime, so which of them is “correct”?
None, really. The analogies don’t tell us what spacetime is. They’re just different ways
of verbally describing what the equations of general relativity say. They each imperfectly
describe some aspects of the equations, and fail to capture others. (Any simple analogy
must fail to capture the content of a theory, because if it really were simpler and just as
valid, then that analogy would be the theory instead!) There is no actual spacetime rubber
or river; those are just stories we tell ourselves to make the mathematics more appealing to
our animal-descended minds. Of course, philosophers debate over whether the attitude I’ve
expressed in this paragraph is right. It’s called “anti-realism”, and I wrote about it here.

21