Problem Set 1.

General Relativity, HT18

THE SIGN OF FOUR: COORDINATES, 4-VECTORS and 4-TENSORS

“That’s its business,” said Lambert. “If Balbus says it’s the same bulk, why, it’s the same bulk,
you know.”
“Well, I don’t believe it,” said Hugh.
“You needn’t,” said Lambert. “Besides, it’s dinner-time. Come along.” They found Balbus
waiting dinner for them, and to him Hugh at once propounded his difficulty.
“Let’s get you helped first,” said Balbus, briskly cutting away at the joint. “You know the old
proverb, ‘Mutton first, mechanics afterwards’ ?”
The boys did not know the proverb, but they accepted it in perfect good faith, as they did every
piece of information, however startling, that came from so infallible an authority as their tutor.
They ate on steadily in silence, and, when dinner was over, Hugh set out the usual array of pens,
ink, and paper, while Balbus repeated to them the problem he had prepared for their afternoon’s
task.
— Excerpt from A Tangled Tale, by Lewis Carroll

1.) Consider the following thought:

“Special relativity holds for frames moving at constant relative velocity, but of course
acceleration requires general relativity because the frames are noninertial.”
Ineffable twaddle. Special relativity certainly doesn’t cower before simple kinematical ac-
cleration. On the other hand, acceleration, even just uniform accleration in one dimension,
is not without its connections with general relativity. We shall explore some of them here.
For ease of notation, we set c = 1. In part (d) we’ll put c back.

1a.) Let us first ask what we mean by “uniform acceleration.” After all, a rocket approaching
the speed of light c can’t change its velocity at a uniform rate forever without exceeding c at
some point. Go into the frame moving instantaneously at velocity v with the rocket relative
to the “lab.” The instantaneous rocket velocity, v 0 , vanishes in this frame. Wait a time dt0
later, as measured in this frame. The rocket now has velocity dv 0 in this same frame. What
we mean by constant acceleration is dv 0 /dt0 ≡ a0 is constant. The acceleration measured in
the lab is certainly not constant! The question is, how is the lab acceleration a = dv/dt
related to the constant a0 ?
√
To answer this, let V = v/ 1 − v 2 , the spatial part of the 4-vector V α associated with
the ordinary velocity v. The same relation holds for V 0 and v 0 . Assume for the moment
that the primed and unprimed frames differ by some arbitrary velocity w. The 4-velocity
differentials are given by: √
dV 0 = (dV − w dV 0 )/ 1 − w2
√
where V 0 = 1/ 1 − v 2 . Explain.
This is the standard Lorentz transformation applied to the 4-vector differential dV α , with
dV the space conponent and dV0 the time part.

1b.) Now, set w = v. We thereby go into the frame in which v 0 = 0; the rocket is instanta-
neously at rest. Prove that dv = dv 0 (1 − v 2 ). (Remember, v and v 0 are ordinary velocities.)

1
From here, prove that
dv
= a0 (1 − v 2 )3/2 .
dt
The equation of part (a) reads

v0
! " ! !#
1 v 1
d √ =√ d √ − vd √
1 − v 02 1 − v2 1 − v2 1 − v2

Carrying through the differentials and setting v 0 = 0 leads directly to dv 0 = dv/(1

√ − v 2 ).
Finally, the invariance of dt2 − dx2 between rocket and lab frames implies dt0 = dt 1 − v 2 :
the rocket frame is defined by dx0 = 0, while in the lab frame, dx = vdt. Hence
dv dv 0
a= = 0 (1 − v 2 )3/2 = a0 (1 − v 2 )3/2
dt dt

1c.) Show that, starting from rest at t = t0 = 0,

a0 t
v=√ , a0 t = sinh(a0 t0 ),
1 + a02 t2
and hence show that (for x = 0 at t = t0 = 0):
1
v = tanh(a0 t0 ), x= [cosh(a0 t0 ) − 1]
a0
The integrals are not difficult; do them yourselves.
The differential equation becomes
Z
dv
= a0 t
(1 − v 2 )3/2
If you are resourceful, or even had just peaked one line further down in the Problem, you
would be inspired to try v = tanh u as a new integration variable. This leads to :
Z
dv Z
0 sinh u a0 t
= cosh u du = sinh u = a t → v = tanh u = q = √
(1 − v 2 )3/2 1 + sinh2 u 1 + a02 t2

Then with
dt0 √ 1 0 0
Z
a0 dt
= 1 − v2 = √ → a t = √
dt 1 + a02 t2 1 + a02 t2
Then substituting a0 t = sinh u we obtain immediately a0 t0 = u, or equivalently
a0 t = sinh(a0 t0 ) → v = tanh(a0 t0 )
Finally, a direct elementary integration gives
Z t
a0 t dt 1 √ 1
x= √
02
= 0
( 1 + a 02 t2 − 1) =
0
(cosh a0 t0 − 1)
0 1+a t 2 a a

2
1d.) Let’s use these results to construct a full coordinate transformation from the lab frame
x, t to the accelerating x0 , t0 frame. A good start is to guess a transform of the form
t = A(x0 ) sinh(a0 t0 ) + B(x0 ), x = A(x0 ) cosh(a0 t0 ) + C(x0 )
where A, B, and C depend only upon x0 . Then on x0 = constant surfaces, dx/dt =
tanh(a0 t0 ) = v, which is indeed what we need.
We can make t0 a true time variable by demanding that constant t0 surfaces are true
constant time surfaces in the (x0 , t0 ) frame that moves instantaneously with velocity v =
tanh(a0 t0 ) with respect to the (x, t) frame. Within such a constant t0 surface, dt0 /dx0 = 0.
We zero our clock and fix our origin by demanding that t → t0 , x → x0 as t0 → 0. Show that
these constraints force B and C to be constant, and that B in particular must vanish.
Finally put the speed of light c back into the equations, demand that x goes to x0 at
0
t = 0, and show that
c2 c2 c2
! !
ct = + x0 sinh(a0 t0 /c), x= + x 0
cosh(a 0 0
t /c) −
a0 a0 a0

Within a constant t0 surface, for a change dx0 , we must have dt0 = 0 = γ(dt − vdx) (by
Lorentz). With x0 derivatives denoted by Dx0 , at constant t0 we must have
v
z }| {
0 = (Dx0 A sinh(a0 t0 ) + Dx0 B) − tanh(a0 t0 ) (Dx0 A cosh(a0 t0 ) + Dx0 C) → Dx0 B = Dx0 C tanh(a0 t0 )
| {z } | {z }
dt dx

which is impossible to satisfy if B and C depend only on x0 , unless they are both actually
constant. Next, at early times, the t equation gives t = Aa0 t0 + B. But as this must be true
for all early times, and A depends only x0 , B=0, and at small x0 , A = 1/a0 . But the equation
for x0 tells us that at early times, x0 = A + C. However A is a function only of x0 and
knows nothing of time, so A = x0 − C in general. Taking the limit x0 → 0, we must have
C = −A(x0 =0) = −1/a0 . Hence

A = x0 + 1/a0 , B = 0, C = −1/a0
Since only the combination a0 t0 /c is dimensionless and c2 /a0 has dimensions of length, this
is the desired result when we put c back in.
1e.) Show that the invariant Minkowski line element may be written in x0 , t0 coordinates as:
!2
a0 x 0
2 2 2 2
c dτ = c dt − dx = 1 + 2 2
c2 dt02 − dx02 .
c
Provide a physical interpretation of your result in terms of a gravitational redshift.
We have
c2
!
cdt = dx0 sinh(a0 t0 /c) + + x0 dt0 (a0 /c) cosh(a0 t0 /c)
a0
c2
!
0 0 0
dx = dx cosh(a t /c) + + x0 dt0 (a0 /c) sinh(a0 t0 /c)
a0

3
Evaluating c2 dt2 − dx2 leads directly to the provided expression. For a local dx0 = 0 acceler-
ating observer, the relationship dt = dt0 (1 + ax0 /c2 ) is recognisable as the kinematic redshift
arising from the a photon covering the distance x0 , and thus the effect of a local gravitational
imparting an acceleration a0 to freely falling objects.

2.) Recognising tensors. One way to prove that something is a vector or tensor is to show
explicitly that it satisfies the coordinate transformation laws. This can be a long and arduous
procedure if the tensor is complicated. There is another way.
Show that if Vν is an arbitrary covariant vector and the combination T µν Vν is known to be
a contravariant vector (note the free index µ), then
0µ
∂x0ν
!
0µν λσ ∂x
T −T Vν0 = 0
∂xλ ∂xσ
Why does this prove that T µν is a tensor? Does your proof actually depend on the rank of
the tensors involved?

∂x0µ λσ ∂x0µ ∂x0ν λσ 0

T 0µν Vν0 = T V σ = T Vν
∂xλ ∂xλ ∂xσ
The first equality holds because it is the just the vector transformation law for the vector
T µν Vν . The second equality holds because it is just the transformation law for the vector Vσ .
The above is then
∂x0µ ∂x0ν λσ
!
0µν
T − T Vν0 = 0,
∂xλ ∂xσ
the desired equation. It proves that T µν must be a tensor because Vν0 is arbitrary so that the
left side can be identically zero only if the terms in the braces is identically zero.
The proof is identical in construction regardless of the actual rank of the vectors and
tensors involved. If you know that an inner product of general tensors is a tensor, and you
know that all the individual scripted objects within this general form are tensors—except for
one—the remaining quantity will be a tensor.

3.) What about d2 xµ /dτ 2 ? The geodesic equation in standard form gives us an expresssion
for d2 xµ /dτ 2 in terms of the affine connection, Γµνλ . For the covariant coordinate xµ , show
that
d2 x µ 1 dxν dxρ ∂gνρ
=
dτ 2 2 dτ dτ ∂xµ
(Hint: start with the standard geodesic equation for d2 xµ /dτ 2 , call it dV µ /dτ , multiply by
Vµ ≡ dxµ /dτ , and take it from there.) Under what conditions is V0 ≡ Vt a constant of the
motion?
Following instructions, we have:
dV µ
Vµ + Vµ Γµνρ V ν V ρ = 0
dτ
Integrating by parts and remembering that Vµ V µ = −c2 is constant,
dVµ
−V µ + Vµ Γµνρ V ν V ρ = 0
dτ
4
In the last term, interchange dummy indices µ and ρ
dVµ
−V µ + Vρ Γρνµ V ν V µ = 0
dτ
or !
µ dVµ
−V − Γρνµ Vρ V ν =0
dτ
The terms in the brackets comprise a vector that vanishes in local Cartesian coordinates
(dVρ /dτ = 0), and so for general coorindate systems,
dVµ
− Γρνµ Vρ V ν = 0
dτ
(We could have actually just begun with this, based on the results of §4.7!) Now

gτ ρ
! !
∂gντ ∂gµτ ∂gµν 1 ∂gντ ∂gµτ ∂gµν
Γρνµ Vρ V ν = µ
+ ν
− Vρ V ν = + − V τV ν
2 ∂x ∂x ∂xτ 2 ∂x µ ∂x ν ∂xτ

But the last two g partial derivatives vanish by ντ antisymmetry. This leaves
dVµ 1 ∂gντ τ ν
= V V
dτ 2 ∂xµ
Since τ is a dummy variable, this equation is equivalent to

d2 x µ 1 ∂gνρ dxρ dxν

=
dτ 2 2 ∂xµ dτ dτ
as desired. With µ = t, we see immediately that if metric is time independent, Vt is a
constant.

4a.) Practise with the Ricci Tensor. Consider the 2D surface given by

z 2 = x2 + y 2
where x, y, z are Cartesian coordinates in 3D Euclidian space. This represents a pair of cones
centred on the origin, one cone opening upward, the other opening downward. The opening
angle is 45◦ measured from the z axis. Justify this description.
The easiest way to see this is to use cylindrical coordinates with R2 = x2 + y 2 . Then the
equation is z = ±R which means in any zR plane the 2D surface is just the straight line
with a slope of unity, which is indeed an angle of 45◦ from the z axis.
4b.) A point in the 2D conic surface can be determined by R, the cylindrical radius of the
point measured from the z-axis, and φ, the usual azimuthal angle. Show that the metric for
the 2D surface in these coordinates is
ds2 = 2dR2 + R2 dφ2
(Hint: Start with a standard metric in good old 3D Euclidian space, then enforce the con-
straint that z 2 = x2 + y 2 = R2 . This is known as “embedding.” The Nash Embedding

5
Theorem states that pretty much any Riemannian hypersurface can always be embedded in
some higher dimensional Euclidian space. )
Start with
ds2 = dx2 + dy 2 + dz 2 = dR2 + R2 dφ2 + dz 2
in good old cylindrical coordinates. But the surface is just z 2 = R2 . So
R
dz = dR = ±dR
z
and
ds2 = 2(dR)2 + R2 (dφ)2

4c.) Is this 2D surface curved, in the mathematical sense of having nonvanishing components
of the curvature tensor Rλκµν ? (We use R for the tensor, R for the radial coordinate.) Answer
the question by showing that the metric of part 4b) can be transformed to new coordinates
R0 , φ0 , for which
ds2 = dR02 + R02 dφ02
(The transformation law is extremely simple!) Why does this result alone answer the posed
question? Can you give a physical interpretation of your mathematical transformation?
√
The obvious thing to do is to define R02 = 2R2 .√Then the metric is dR02 + (R0 / 2)2 dφ2 .
Now the obvious thing to do is to define φ0 = φ/ 2. We are back to Euclidian space via a
coordinate transformation, so there can’t have been any curvature in this space: coordinates
don’t make curvature!
Imagine starting out with a piece of paper, Euclidian 2-space, described by R0 , φ0 coor-
dinates. Now fold the paper into a cone, or fool’s cap. You do this, without creasing the
paper, by shrinking the cylndrical radius while increasing the azimuthal angle in the process
of “wrapping”! All of the interspatial relationships (think of a drawing) on the original pa-
per remain undistorted. This is true whether we make either a cone or a cylinder from the
original.
4d.) Next, consider a different 2D surface: z = (α/2)(x2 + y 2 ) where α is an arbitrary
constant parameter. Show that this is a paraboloid of revolution, i.e. a parabola spun
around the z-axis. Prove that the metric within this surface is given by

ds2 = (1 + α2 R2 )dR2 + R2 dφ2 .

Embed, exactly as in part 4b). With dz = αR dR the result follows by inspection.

4e.) Prove that this surface is distorted by curvature. Calculate, for example, Rφφ and show
that it is not zero, but given by
α2 R2
Rφφ = −
(1 + α2 R2 )2
You should show en route that the only nonvanishing affine connection coefficients are
α2 R 1 R
ΓR
RR = , ΓφφR = ΓφRφ = , ΓR
φφ = −
1 + α2 R2 R 1 + α2 R2

6
The metric determinant is
g = R2 (1 + α2 R2 )
and for future reference
α2 R
!
1 ∂ ln g 1
= +
2 ∂R R 1 + α2 R2
From the Ricci Tensor formula on page 6 of the notes,

η ΓR
φφ
Rφφ = −∂R ΓR λ
φφ + Γφλ Γφη − ∂R ln g
2
Then
α2 R2 − 1
∂R ΓR
φφ =
(1 + α2 R2 )2
2
Γηφλ Γλφη = 2ΓR φ
φφ ΓφR = −
1 + α2 R2
ΓR α2 R2
!
φφ 1
∂R ln g = − 1 +
2 1 + α2 R2 1 + α2 R2
Adding it all up:
ΓR
Rφφ = −∂R ΓR
+ φφ − φφ ∂R ln g =
Γηφλ Γλφη
2
1 − α R − (2 + 2α R ) + 1 + α R + α2 R2
2 2 2 2 2 2
α2 R2
= −
(1 + α2 R2 )2 (1 + α2 R2 )2

5.) Hydrostatic Equilibrium in GR. Model a neutron star atmosphere with a simple equation
of state: P = Kργ , where P is pressure, ρ is mass density, γ is the adiabatic index and K
is a constant. Assume that g00 = −(1 − 2GM/rc2 ), where M is the mass of the star and r
is radius. If ρ = ρ0 at the surface r = R0 , solve the equation of hydrostatic equilibrium to
show that !α
1 + Kργ−1 /c2 1 − RS /r0
=
1 + Kργ−1
0 /c
2 1 − RS /r
where RS = 2GM/c2 is the so-called Schwarzschild radius, and 2αγ = γ − 1. (Hint: See §4.6
of the notes.) What is the Newtonian limit of the above equation? Express your answer in
terms of the speed of sound a, a2 = γP/ρ and the potential Φ(r) = −GM/r.
The equation of hydrostatic equlibrium (HSE) is
1/2
∂P 2 ∂ ln |g00 |
+ (P + ρc ) =0
∂xµ ∂xµ
When P = P (ρ) this separates nicely. Upon integration:
Z
P 0 (ρ) dρ
+ ln |g00 |1/2 = const.
P (ρ) + ρc2

7
Taking P = Kργ , the first integral is

γ Z K(γ − 1)ργ−2 γ
2 γ−1
= ln(c2 + Kργ−1 )
γ−1 c + Kρ γ−1
so that HSE becomes
γ−1
ln(c2 + Kργ−1 ) + ln(1 − RS /r) = const.
2γ
With boundary conditions ρ = ρ0 at r = r0 , this is easily rearranged as
!(γ−1)/2γ
1 + Kργ−1 /c2 1 − RS /r0
=
1 + Kργ−1
0 /c
2 1 − RS /r

In the Newtonian limit, we treat 1/c2 and RS as small quantities. A linear expansion in
these quantities gives to first order,
γ−1
Kργ−1 /c2 − Kργ−1 2
0 /c = (−RS /r0 + RS /r)
2γ
or cleaning up and putting it more neatly,

a2 a20
+ Φ(r) = + Φ(r0 )
γ−1 γ−1
This expresses “conservation of enthalpy.”

6.) Bondi Accretion: go with the flow. THIS IS AN OPTIONAL PROBLEM! To get some
more practise working with the equations of GR, consider the problem of Bondi Accretion,
the (exactly) spherical flow of gas into a black hole. Here we take the diagonal metric gµν
as known. Later in the course we will derive gµν for a simple black hole, and build on the
results we find here.
6a.) First, let us assume that particles are neither created or destroyed, just to keep things
simple. So particle number is conserved. We use the usual r, θ, φ spherical coordinates. If
n is the particle number density in the local rest frame of the flow, then the particle flux
is J µ = nU µ , where U µ is the flow 4-velocity. Justify this statement, and using §4.5 in the
notes, show that particle number conservation implies:
J µ;µ = 0.

With n a scalar, the only vector that reduces to a simple flux nv in the nonrelativistic limit
is J µ = nU µ . The equation J µ;µ = 0 is just the GR upgrade of ∂t n + ∂i (nvi ) = 0 in the
nonrelativistic limit, a statement of particle number conservation, or the familiar statement
of (rest) mass conservation.
If nothing depends upon time, show that this integrates to

nU r |g 0 |1/2 = constant,

8
where g 0 is the determinant of gµν divided by sin2 θ, and U r is...well, you tell me what U r is.

With only a radial component of the 4-velocity present, U r (this answers the last question),
§4.5 of the notes gives without further ado:

d(nU r |g|1/2 )
=0
dr
with |g| = r4 sin2 θ. Obviously the factor of sin2 θ can be dropped. Integrating this equation
is trivial and leads to stated result.
6b.) We move on to energy conservation, T tν;ν = 0. (Refer to §4.6 in the notes.) Show that
the only nonvanishing affine connection that we need to use is
1 ∂ ln |gtt |
Γttr = Γtrt =
2 ∂r

The condition T tν;ν = 0 is:

∂P 1 ∂ h 1/2 i
0 = g µt + |g| (ρ + P/c2
)U µ t
U + Γtµλ (ρ + P/c2 )U µ U λ
∂xµ |g|1/2 ∂xµ

The only Γtµλ that are relevant are those with the two bottom indices either t or r (corre-
sponding to the nonvanishing U µ U λ terms). The only one of these Γs that does not vanish
for time-steady flow is
1 ∂ ln |gtt |
Γttr = Γtrt =
2 ∂r
(the expression on the right is from the front pages of the notes.)
Derive and solve the energy equation. Show that its solution may be written

(P + ρc2 )U r Ut |g 0 |1/2 = constant

where Ut = gtµ U µ , and ρ is the total energy density of the fluid in its rest frame, including
any thermal energy. These two surprisingly simple equations will be solved in the next
problem set.
Our energy equation for time-steady spherical flow is therefore:
1 ∂ h 1/2 2 r t
i
0= |g| (ρ + P/c )U U + 2 × Γtrt (ρ + P/c2 )U r U t
|g|1/2 ∂r

the factor of 2 from Γtrt = Γttr . We may replace g by g 0 in the r derivative since the θ
dependence disappears, and this becomes
1 ∂ h 0 1/2 r t ∂ ln |gtt | 1 ∂ h 0 1/2
i i
2 r t 2 2 r
0= |g | (ρ + P/c )U U +(ρ+P/c )U U = |g | (ρ + P/c )U Ut
|g 0 |1/2 ∂r ∂r gtt |g 0 |1/2 ∂r

where we have used Ut = gtt U t . The r integration is now trivial, and leads to the desired
result.