11.

1 Static spherically symmetric spacetimes

A metric is said to be stationary if it admits a timelike Killing vector K. It is said to be static if it
is stationary and there exists a surface Σ of codimension one which is everywhere orthogonal to K. In a
neighborhood of Σ every point x belongs to a unique flow line of K. This line passes through a point x̄ ∈ Σ,
and x = φ(x̄, t), where φ is the flow of K (section 3.6). We can use t and the coordinates of x̄ as coordinates
∂
for x. In these coordinates K = ∂t and the metric is

ds2 = −V 2 (x1 , x2 , x3 )dt2 + hij (x1 , x2 , x3 )dxi dxj , (11.1.1)

where V 2 = −g(K, K) and the condition of being static implies the absence of mixed terms dt dxi .
A metric has spherical symmetry if it admits three spacelike Killing vectors satisfying the algebra of
SO(3), whose orbits are two dimensional spheres. The induced metric in the orbits is the standard metric in
S 2 , up to rescalings. The three Killing vectors are given explicitly in (5.4.2.1). Let A be the area of an orbit
computed with the induced metric. Neighboring q orbits will generally have different areas. We can define a
A
coordinate r labelling different orbits by r = 4π . It is called “area radius” because by definition, the area
of an orbit is equal to 4πr2 . One should not assume however that r is a “radius” in the usual sense, nor that
the topology of space is the topology of R3 .
If a spacetime is both static and spherically symmetric, the timelike Killing vector K must commute
with the generators of rotations. In S 2 there does not exist any vectorfield commuting with all generators
of rotations, so K must have zero components in the orbits of SO(3). This implies that the orbits lie in the
surfaces t = const. Then the metric must have the form

ds2 = −f (r)dt2 + h(r)dr2 + r2 (dθ2 + sin2 θdϕ2 ) . (11.1.2)

This is as far as one can get in general. We have reduced the ten independent components of the metric,
which in general are functions of all the coordinates, to just two functions of a single coordinate. In order
to determine these functions we have to insert the ansatz (11.1.2) into Einstein’s equations.
11.2 The Schwarzschild solution
We are interested in the gravitational field of a point particle. We tentatively assume that space is R3
and that the particle lies at the origin of the coordinates, r = 0. Clearly for r 6= 0 there is no matter, so we
have to solve Einstein’s equations in vacuum

Rµν = 0 . (11.2.1)

We thus need the components of the Ricci tensor of the metric (11.1.2). One way to compute them is to
work out all the Christoffel symbols (9.2.5) and then all the components of the Riemann tensor (9.3.1). A
contraction then yields the Ricci tensor (9.3.5). This is generally a rather tedious, albeit fully straightforward,
procedure. It is sometimes convenient to use instead the vierbein gauge and to obtain the connection by
solving the equations (9.2.1). In the vierbein gauge, from (9.1.3) and (9.1.12) we see that the components
of the soldering form can be regarded as the components of an orthonormal frame for gµν . In this gauge it
is natural to decompose all tensor components in the orthornormal basis, but one can also choose to refer
some components to an orthonormal basis and others to a natural basis. Recall that in section 6.5 we found
it convenient to hide the Lie algebra indices and treat Aµ as a matrix–valued form. We are now going to
do the opposite: we will hide the form indices and keep the algebra indices in sight. Thus, the Levi–Civita
connection is a one form Γ a b = Γµ a b dxµ and its curvature is Ra b = 21 Rµν a b dxµ ∧ dxν . The soldering form
is θa = θa µ dxµ . Because of the metricity condition (9.2.1b), Γab = −Γba (and Rab = −Rba ). Then the
condition of vanishing torsion (9.2.1a) can be written

dθa + Γ a b ∧ θb = 0 (11.2.2)

and the definition of the curvature, equation (6.5.4), becomes

Ra b = dΓ a b + Γ a c ∧ Γ c b . (11.2.3)

1
With some practice, using the properties of differential forms can be much more efficient than working out
the Christoffel symbols. We illustrate this by computing the Levi–Civita connection. The vierbein for the
metric (11.1.2) is p √
θt = f dt ; θr = h dr ; θθ = r dθ ; θϕ = r sin θ dϕ .
Then (11.2.2) becomes

f′ √
√ dr ∧ dt + Γt r ∧ h dr + Γt θ ∧ r dθ + Γt ϕ ∧ r sin θdϕ =0 ; (11.2.4a)
2 f
Γr t ∧ f dt + Γr θ ∧ r dθ + Γr ϕ ∧ r sin θdϕ =0
p
; (11.2.4b)
p √
dr ∧ dθ + Γθ t ∧ f dt + Γθ r ∧ h dr + Γθ ϕ ∧ r sin θdϕ =0 ; (11.2.4c)
p √
dr ∧ sin θdϕ + r cos θdθ ∧ dϕ + Γϕ t ∧ f dt + Γϕ r ∧ h dr + Γϕ θ ∧ r dθ =0 ; (11.2.4d)

Since we know that the solution will be unique, we are free to make guesses about the coefficients Γ a b and
check them a posteriori. One reasonable guess is that the spherical coordinates do not mix with time, in the
sense that Γtθ = 0 and Γtϕ = 0. Inserting in (11.2.4a) we find that

f′
Γtr == −Γ t r = −Adr − √ dt
2 hf

where A is a function of r. Using this in (11.2.4b), the latter becomes

p
f A dr ∧ dt + Γrθ ∧ rdθ + Γrϕ ∧ r sin θdϕ = 0 (11.2.5)

Since the first term does not contain dθ or dϕ, we must have A = 0. Next we assume that

Γrθ = F dθ + Gdϕ ; Γrϕ = Hdθ + Ldϕ .

which inserted in (11.2.5) yields G = H sin θ. Inserting in (11.2.4c) gives

√
dr ∧ dθ − (F dθ + Gdϕ) ∧ hdr + Γθϕ ∧ r sin θdϕ = 0 ;

The coefficient of dr ∧ dθ must vanish, and this implies that F = − √1h . The remaining terms tell us that
√
G h
Γθϕ =− dr + M dϕ
r sin θ
for some function M . Inserting all the above in (11.2.4d) gives
√ !
√
 
G G h
sin θ dr ∧ dϕ + r cos θdθ ∧ dϕ − dθ + Ldϕ ∧ h dr + dr − M dϕ ∧ r dθ = 0
sin θ r sin θ

√ θ . The coefficient of dr∧dθ must vanish,

The coefficient of dr∧dϕ must vanish, and this implies that L = − sinh
and this implies that G = 0. The coefficient of dθ ∧ dϕ must vanish, and this implies that M = − cos θ.
Altogether we have found, with remarkably little calculational effort,

f′ 1 sin θ
Γtr = − √ dt ; Γtθ = 0 ; Γtϕ = 0 ; Γrθ = − √ dθ ; Γrϕ = − √ dϕ ; Γθϕ = − cos θdϕ .
2 hf h h
(11.2.6)
From here by using (11.2.3) and then extracting the coefficient of θa ∧ θb one reads off the independent
vierbein components of the Riemann tensor (one per generator of the Lorentz algebra):

1 d f′ f′ h′
 
1 1
Rtrtr = √ √ ; Rtθtθ = Rtϕtϕ = ; Rrθrθ = Rrϕrϕ = ; Rθϕθϕ = 2 1 − .
2 hf dr hf 2rhf 2rh2 r h
(11.2.7)

2
Contracting the first and third index of Rabcd with η ac one gets the vierbein components of the Ricci tensor.
Only the diagonal components are nonzero, and the θθ and ϕϕ are equal. Thus the independent Einstein
equations in vacuum reduce to

1 d f′ f′
Rtt = √ √ + =0, (11.2.8a)
2 hf dr hf rhf
1 d f′ h′
Rrr = − √ √ + 2 =0, (11.2.8b)
2 hf dr hf rh
f′ h′
 
1 1
Rθθ = Rϕϕ =− + + 1 − =0, (11.2.8c)
2rhf 2rh2 r2 h

Summing the first two equations one obtains

f′ h′
+ =0
f h
√
whose solution is f = K
h for some integration constant K. One can use the freedom of rescaling t → t/ K
to set K = 1. Reinserting in (11.2.8c) one obtains a single differential equation for f , which reduces to

d
(rf ) = 1 .
dr
The solution of this equation is f = 1 + rc , where C is a constant of integration. This constant can be
determined by comparison with (9.2.2):
 
C 2M G
− 1+ → −1 +
r r

which implies C = 2M G. To summarize, the static spherically solution of Einstein’s equations in vacuum
can be written in the form
   −1
2 2M G 2 2M G
ds = − 1 − dt + 1 − dr2 + r2 (dθ2 + sin2 θ dϕ2 ) . (11.2.9)
r r

This is called the Schwarzschild solution in “standard” form. It has an singularity at r = 0, that one would
have expected on the basis of our Newtonian understanding, but also a singularity at r = 2M G that one
could not have anticipated. We shall discuss the meaning if this (apparent) singularity later on.

2
Exercise 11.2.1. For r > 2M G define a new radial coordinate r′ by r = r′ 1 + MG 2r ′ (the
transformation becomes singular at r = 2M G). Show that the metric assumes the “isotropic” form
MG

4
1−

M G  ′2
ds2 = − 2r ′
2
dr + r′2 (dθ2 + sin2 θ dϕ2 )

MG
dt + 1+ ′
1+ 2r ′
2r

The coordinate singularity now occurs at r′ = M G/2.

Exercise 11.2.2. Construct harmonic coordinates for the Schwarzschild metric. (See Weinberg
equation (8.2.15).)
Exercise 11.2.3. From (11.2.7), using the explicit solution for f and h. compute the tetrad
components of the Riemann tensor. Compare with the calculation in exercise 8.3.1. The components
that were called R0x0x and R0y0y correspond to Rtθtθ and Rtϕtϕ , while R0z0z corresponds to Rtrtr .
In the preceding derivation we have assumed that the body producing the gravitational field is a point
particle. The gravitational fields that we can observe are all produced by macroscopic bodies such as planets
or stars, which have a finite spatial extension. In Newton’s theory, Birkhoff’s theorem guarantees that the
gravitational field outside a spherical body is the same as that produced by a point particle located in the

3
center and having a mass equal to the total mass of the body. This theorem holds true also in Einstein’s
theory. Therefore the gravitational fields outside the Earth or the Sun is described by the metric (11.2.9),
with M equal to the mass of the Earth or the Sun. The gravitational field inside these bodies is not described
by the metric (11.2.9), because one would have to solve Einstein’s equations in the presence of a nontrivial
Tµν . The singularity at r = 0 does therefore not occur in these cases. The singularity at r = 2M G does
not occur either, because the “Schwarzschild radius” rS = 2M G is much smaller than the actual radius
of ordinary objects. For the Earth, rS ≈ 1cm while for the Sun rS ≈ 3km. In fact one can find regular
solutions for the interior of planets and stars and match them continuously to the Schwarzschild solution at
the surface.
11.4 The gravitational red shift
We are now in a position to check that the theory reproduces the prediction of gravitational red-
shift/blueshift, that we made in section 1.4 based just on the equivalence principle. To this effect, imagine
two observers located at two fixed radii r1 and r2 in the Schwarzschild metric. The first observer sends a
photon of frequency ω1 towards the second observer. What is the frequency of the photon measured by the
second observer? Let us call k µ the four-momentum of the photon. We exploit the fact that the metric has
a timelike Killing vector K = ∂t and the gravitational Noether theorem discussed in section 12.4, and saying
that kµ K µ is constant along the null geodesic traced by the photon. The four-velocities of the two p
observers
µ µ
are parallel to K butp have different norm: g(u,µu) = −1 whereas g(K, K) = g00 , therefore K = |g00 |u .
µ
Then we have kµ u |g00 | =constant. But kµ u is the frequency of the photon as measured by the observer
with four–velocity u (section 10.6). Thus
p p
ω1 |g00 (r1 )| = ω2 |g00 (r2 )|

Using the Schwarzschild solution we find,

q
2MG
ω2 1− r1
=q . (11.4.1)
ω1 1− 2MG
r2

We see that a red shift occurs when the photon travels outwards and a blueshift occurs when the photon
travels inwards. In particular, when the second observer is at r2 = ∞ the frequency is red-shifted by a factor
q
1 − 2MG
r1 . This factor increases as r1 decreases and becomes infinite when r1 = 2M G. Therefore r = 2M G
is a surface of infinite redshift. As discussed in section 1.4, one can see the redshift as the energy loss of the
photon as it climbs out from the potential well; then the infinite redshift means that a photon generated
just outside r = 2M G must spend all its energy to climb the potential well. This suggests that a photon
generated just inside r = 2M G will not be able to climb the potential well at all. We shall see later that
this is indeed the case: anything that is inside the surface r = 2M G cannot escape, and for this reason this
surface will appear to an external observer as a black hole.
When both r1 and r2 are much larger than the Schwarzschild radius 2M G, we can approximate
ω2 MG MG
≈1− + . (11.4.2)
ω1 r1 r2
and the fractional frequency shift is
ω2 − ω1 r1 − r2
≈ MG . (11.4.3)
ω1 r1 r2
It would be possible to measure this effect for photons coming from the sun, were it not for the fact that the
thermal widening of the spectral lines is much greater than the predicted frequency shift. As mentioned in
section 1.4, it has been possible to observe this effect directly in the laboratory using gamma ray photons.
11.3 Geodesics
The first experimental tests of Einstein’s theory came from accurate measurements of the motion of
planets or light rays in the gravitational field of the Sun. This requires understanding the geodesics of the
Schwarzschild metric.

4
 The solution is greatly simplified by exploiting the symmetries of the problem. Due to the existence of
four Killing vectors, there are four conserved quantities along any geodesic, which can be roughly associated
to energy and angular momentum. As in Newtonian mechanics, two of the three conserved quantities
associated to spherical symmetry imply that the motion occurs in a plane. Without loss of generality we
can assume this to be the plane θ = π/2. The remaining two Killing vectors are then ∂t and ∂ϕ . Consider
first the case of a massive particle. If we call u the vector tangent to the geodesic (the four-velocity) we
define g(∂t , u) = −E and g(∂ϕ , u) = L. The four-momentum p is related to u by p = mu, so we find that
the covariant components p0 = −mE and pϕ = mL are constant. The other components of the momentum
dr
are pθ = 0 (because the motion is in a plane) and pr = m dτ . The equation g(p, p) = −m2 then tells us that
 2
dr
= E 2 − Vef f (r) , (11.3.1)
dτ

where
L2
  
2M G
Vef f (r) = 1 − 1+ 2 . (11.3.2)
r r
For a massless particle p0 = −E and pϕ = L are constant. The previous argument leads again to equation
(11.3.1) but now with
2M G L2
 
Vef f (r) = 1 − . (11.3.3)
r r2
In both cases the qualitative features of the orbits can be gleaned from these equations, considered as the
energy conservation for a particle with mass 2 and total energy E 2 , moving in the potential Vef f .
The potential for a massive particle is plotted in the following figure for L = 8, 6, 4, 2 (from top to
bottom), on two different scales
V V
3 1.1

2
1.05

1
r
€€€€€€€
50 100 150 200 MG
r
€€€€€€€
5 10 15 20 25 30 MG

0.95
-1

-2 0.9

The effective potential has a maximum and a minimum at

r !
L2 12(M G)2
r± = 1± 1− . (11.3.4)
2M G L2

If E 2 is greater than the maximum, an incoming particle reaches r = 0, if it is less than the maximum, r
reaches a minimum value at the turning point. There are two circular orbit corresponding to the stationary
points r+ and r− . The one at r+ is stable, the one at r− unstable.
The stable orbit is given by r(τ ) = r+ . The function ϕ(τ ) can be determined as follows:
dϕ 1 L
= uϕ = g ϕϕ pϕ = 2 . (11.3.5)
dτ m r
L
R 2πr 2
Therefore ϕ(τ ) = 2
r+
τ and the orbital period is T = dτ = L + . Let us consider a small perturbation of
this stable orbit. It will be given by a small radial oscillation with frequency

1 d2 Vef f  M G(r+ − 6M G)
ωr2 =  = 3 . (11.3.6)
2 dr2 r+ r+ (r+ − 3M G)

5
On the other hand the frequency of the angular motion remains the same to first order. Using the identity
2 L2
r± = MG (r± − 3M G), it can be written
2
L2

2π MG
ωϕ2 = = 4 = 2 . (11.3.7)
T r+ r+ (r+ − 3M G)

If the two frequencies were the same, then the orbits would closed curves, as happens in Newtonian gravity.
But they are not the same: the ratio is
ωr2 6M G
=1− . (11.3.8)
ωϕ2 r+
The difference ωP = ωϕ − ωr is called the precession. To first order it is given by

(M G)3/2
ωP = 3 5/2
. (11.3.9)
r+

The precession of the perihelion of Mercury has been measured and is in accordance with the prediction.
Let us now consider null geodesics, the orbits of photons in the geometric optics approximation. The
following figure shows the effective potential for L=8,6,4,2, from top to bottom.
V
3

2.5

1.5

0.5

r
€€€€€€€
5 10 15 20 25 30 MG

-0.5

-1

It has a maximum at r = 3M G and the value of the potential at the maximum is L2 /27M G. It has
a zero at r = 2M G. There is only one circular orbit at r = 3M G and it is unstable. Every other null
geodesic falls in one of the following two classes. If E 2 > L2 /27M G, it comes in from r = ∞ and ends at
the singularity r = 0. If E 2 < L2 /27M G, it comes in from r = ∞, reaches a turning point and returns to
r = ∞. In both cases, the motion of the photon in the gravitational field of the particle in the origin can
be viewed as a scattering problem. At large distances the geometry becomes Minkowskian so we can apply
the standard notions of special relativity. The angular momentum L = pb where p = |~ p| and b is the impact
parameter, and E √ = p. Thus the distinction between the two classes of orbits depends on whether b is larger
or smaller than 27M G ≈ 5.196M G. If it is smaller it falls inside the horizon at r = 2M G and eventually
reaches the singularity at r = 0. We can therefore say that the cross section for the capture of photons by
the black hole is 27πM 2 G2 .
Consider now a photon that does escape to infinity. To calculate the scattering angle without integrating
the equations of the geodesic, we can integrate dφ ϕ̇
dr = ṙ , where a dot indicates derivative with respect to an
affine parameter. Using (11.3.1) and (11.3.5) we have
−1/2
2M G L2
  
dφ L 2 1
= 2 E − 1− = q .
dr r r r2 2
r rb2 − 1 + 2MG
r

Therefore the scattering angle is

∞
dr 1
Z
∆ϕ = 2 q , (11.3.10)
r0 r r r −1+
2 2MG
b2 r

6
where r0 is the turning point, defined by E 2 = Vef f (r0 ). Using (11.3.3) it satisfies r03 = b2 (r0 − 2M G). Once
again, observations are in agreement with the predicted deflection.
11.4 Resolving coordinate singularities
The theory of general relativity has certain peculiar features originating from its gauge symmetry which,
in the metric gauge that we are using here, consists of coordinate transformations (or diffeomorphisms, when
one uses the active point of view). A first consequence of this gauge symmetry is that when one finds a new
solution, it is always possible that it is an already known solution written in a different coordinate system.
There is no simple criterion that allows one to decide whether the new solution is really new, and one has
to resort to detailed analyses on a case by case basis. A related consequence is that when one encounters
a singularity it is not a priori clear whether it is a true physical singularity or the result of a bad choice of
coordinates. A related point is this: when one set out to solve Einstein’s equations one does not know a
priori what spacetime manifold it will describe. This will only emerge in the end, once one has understood
the global properties of the solution.
All these features have analogs in Yang-Mills theories. When one finds a new solution, it may be an
already known solution written in a different gauge. A singularity in the gauge potential may be due to a
bad choice of local gauge and disappear when one chooses a different local gauge. And finally, when one
sets out to solve the Yang-Mills equations, one does not know a priori what bundle they will describe (the
base space and the fiber are given, but the geometry and even the topology of the totalspace are not given
a priori).
Thus, having found the Schwarzschild solution (11.2.9) is not the end of the story. We still have to
understand its global structure, and the nature of the singularities at r = 0 and r = 2M G. But before
addressing these issues in the next section, it will be convenient to study some simpler examples in two
dimensions.
The first example is the following metric which a silly physicist may find solving Einstein’s equations in
vacuum:
1
ds2 = − 4 dt2 + dx2 . (11.4.1)
t
It is singular at t = 0 and at t = ∞; what is the nature of this singularity? A hint comes from the following
observation. Consider the distance between two points located at (t1 , x) and (t2 , x), with the same value of
x. It is given by Z t2 ′
dt 1 1
Z
ds = ′2
= − .
t1 t t2 t1
Thus, the distance from t1 = 1 to t2 = 0 is infinite, whereas the distance from t1 = 1 to t2 = ∞ is one.
Clearly the coordinate system does not reflect at all the metric relations in this spacetime. To make this
more physical: curves of constant x are geodesics, and if one follows such a geodesic from t = 1 to t = ∞ one
arrives there in finite proper time. Where does one go next? When this happens one says that the geodesics
are incomplete. Clearly the spacetime must continue beyond t = ∞. The natural thing that comes to mind
is to define t′ = 1/t, and indeed one has
ds2 = −dt′2 + dx2 . (11.4.2)
A somewhat less trivial example is the Rindler metric
ds2 = −x2 dt2 + dx2 , (11.4.3)
with −∞ < t < ∞ and 0 < x < ∞. It is singular at x = 0. If one computes the curvature tensor one finds
that it is actually zero, so this cannot be a physical singularity. Actually we know from section 1.4 that
there must exist a coordinate system where the metric reduces to the Minkwskian form. But how do we
find such a coordinate system? As in the previous example, a hint comes from studying geodesics. It will be
sufficient to study null geodesics, which indicate the causal structure. In fact, we do not even need to know
the parametrized form of the curves (t(λ), x(λ)): it will be enough to know x(t) or t(x). This is very simple:
one just sets ds2 = 0 and obtains the equation
dt 1
=± ,
dx x
7
whose solutions are t = ± log x+constant. They form a sort of grid, which suggests using the integrations
constants as new coordinates. Define u = t − log x and v = t + log x; u is constant on outgoing geodesics, v
on incoming geodesics. The inverse coordinate transformation is t = (u + v)/2 and x = exp((v − u)/2), and
in the new coordinates
ds2 = −ev−u du dv . (11.4.4)
The new coordinates range from −∞ to +∞ but still only cover the region x > 0. In fact the singularity at
x = 0 corresponds to u → ∞ or v → −∞.
Let us now ask ourselves if the null geodesics are complete or not. Instead of directly integrating the
equation for the geodesics we use some shortcuts. The Rindler metric (11.4.4) has a timelike Killing vector
∂t . Calling k µ = ẋµ the tangent vector to the geodesics, the quantity E = g(k, ∂t ) = x2 ṫ is constant along
the geodesics, by Noether’s theorem. From here we see that dλ = x2 dt/E, and integrating on an outgoing
null geodesic (u=constant) we find

1 1 e−u v
Z Z
λ= x2 (t)dt = ev−u dv = e ,
E E E

where we have set an integration constant to zero. So we see that ev is an affine parameter on the outgoing
geodesics, and by a similar argument −e− u is an affine parameter on the incoming geodesics. Clearly this
space is geodesically incomplete: an outgoing geodesic comes from v → −∞, but its affine parameter there
was finite, namely zero. One can continue such geodesics further in the past. Similarly an incoming geodesics
arrives at u = ∞ when the affine parameter is zero, and can be continued further into the future. These
arguments show that the coordinates (t, x) cannot cover all the spacetime.
The tranformation to flat coordinates is suggested by looking at (11.4.4): U = −e−u and V = ev . In
such coordinates ds2 = −dU dV , which is the Minkowski metric in null coordinates. In fact, defining further
T = (U + V )/2 and X = (V − U )/2, the line element becomes ds2 = −dT 2 + dX 2 . The main point to stress
is that the coordinate system (u, v) covers only the region U < 0 and V > 0, but there is no singularity in
the metric at either U = 0 or V = 0, which correspond to the location of the apparent singularity in the
(u, v) coordinates. Therefore spacetime can be extended beyond the original coordinate patch.
The complete transformation from the original Rindler coordinates to the Minlowskian coordinates is
√
x = T 2 − X 2 and t = arctanh(T /X). Thus the lines t =constant correspond to straight lines through the
origin in Minkowski space, and thelines x =constant correspond to hyperbolas. The following figure shows
the Rindler coordinate patch and coordinate grid.
y

x
-2 -1 1 2

-1

-2

The following facts should be remarked. The lines x =constant are the worldlines of uniformly ac-
celerated observers in Minkowski space, and the proper time of these observers is xt. Therefore Rindler
coordinates are somehow naturally well suited for such uniformly accelerated observers, that we shall call
“Rindler observers”. A Rindler observer cannot be influenced by any event that occurs at U > 0 and cannot
influence any event that occurs at V < 0. Therefore the boundary of the Rindler wedge (the line x = 0) is
an event horizon for the Rindler observer. The line V = 0, U < 0 is a past event horizon, meaning a surface
across which the Rindler observer can see but can have no influence on. The line U = 0, V > 0 is a future

8
event horizon, meaning a surface across which the Rindler observer can never see but can have influence on.
All the above is only true for an eternally accelerated observer: if the acceleration started at some finite time
an lasted forever, then there would only be a future horizon and no past horizon, and if the acceleration had
lasted forever in the past and stopped at some finite time, then there would only be a past horizon and no
future horizon.
11.4 Kruskal coordinates
In order to appreciate the similarities between the Rindler metric and the Schwarzschild metric, let us
discuss the acceleration four–vector along a worldline a = ∇u u, where u is the four-velocity. The acceleration
by definition vanishes on a geodesic, but here we will be interested in worldlines that are not geodesics.
In order to know a along some curve, in general one has to know the Christoffel symbols. However,
we will only be interested in the case when the spacetime is static with a timelike Killing vector K, and
the four-velocity is everywhere parallel to the Killing vector. In other words we are interested in the case
when the curve describes an observer at rest in the coordinates of (11.1.1). Then there is a simpler way √ to
calculate a. Let us define the positive scalar function V = −Kλ K λ . Since u is unit-normalized, u = K/ V .
Therefore
K ν ∇ν K µ 1
aµ = − K µ K ν ∇ν V .
V 2V 2
Using the Killing equation, it is easy to see that K ν ∇ν V = 0. Therefore

∇µ V
aµ = . (11.4.1)
2V

In the case of a Rindler observer, V = −g00 = x2 and therefore a = 1/x. This confirms what we already
knew, namely that the acceleration is constant on the lines x =constant, and that it is inversely proportional
to the distance of the observer from its own horizon. In particular, an observer at x = 0 would have infinite
acceleration.
Compare with an observer at rest in Schwarzschild coordinates (r, θ, ϕ). Now V = 1 − 2MG r , and the
only component of the acceleration is ar = g rr ∂r V /2V = M G/r2 . The modulus of the acceleration is
p
−1/2 MG
a = grr ar2 = 1 − 2MG r r 2 . It becomes infinite at the surface r = 2M G. It is not possible to remain
at r =constant when r < 2M G, because the coordinate r is timelike there.

Exercise. Compute the acceleration of a Schwarzschild observer by computing the covariant

derivative of u. Only the Christoffel symbol Γ0 r 0 is needed.
This suggests that the nature of the singularity at r = 2M G is very similar to that of the singularity at
x = 0 in Rindler coordinates. The null geodesics have ds2 = 0, so
 −1
dt 2M G
= ± 1− .
dr r

2MG −1


Let us define the coordinate r∗ by dr∗ = 1 − r dr. Then
 r 
r∗ = r + 2M G log −1 . (11.4.2)
2M G
The Schwarzschild line element can be written
   
2 2M G 2 2 2M G
ds = 1 − (−dt + dr∗ ) = − 1 − du dv , (11.4.3)
r r

where u = t − r∗ , v = t + r∗ . Unlike in the Rindler case, it is impossible to explicitly invert the relation
between r and r∗ , so in the following r has to be regarded as a function of r∗ , or u and v. From (11.4.2) one
has  
v−u r r 2M G
e 4M G =e 2M G 1−
2M G r

9
and therefore the metric (11.4.3) can be rewritten in the form
2M G − r v−u
ds2 = − e 2M G e 4M G du dv , (11.4.4)
r
The remarkable fact is that there is no singularity at r = 2M G anymore. As in the Rindler case we can
define new coordinates U = − exp(−u/4M G) < 0 and V = exp(v/4M G) > 0, such that
32M 3 G3 − r
ds2 = − e 2M G dU dV , (11.4.5)
r
and since there is no longer any singularity at r 6= 0, we can extend the solution to all values of U and V .
Finally, passing to coordinates T and X
32M 3G3 − r
ds2 = e 2M G (−dT 2 + dX 2 ) , (11.4.6)
r
Tracing back all the transformations,
 r  r t T
− 1 e 2M G = −U V = X 2 − T 2 , = 2arctanh . (11.4.7)
2M G 2M G X
Thus t is constant along lines through the origin and r is constant along hyperbolas, as in the Rindler case.
Of course, the geometry is not flat here, but in the plane (r, t) it is conformally flat, so the causal structure
is very simple to visualize: the light cones are the same as in flat space. The maximal analytic extension of
the Schwarzschild metric can now be represented by the following figure:
Y

X
-2 -1 1 2

-1

-2

The horizon at r = 2M G corresponds to the light cone through the origin. The singularity at r = 0
corresponds to the hyperbolas T 2 − X 2 = −1. The role of the surface r = 2M G as an event horizon for the
observers at r > 2M G is now evident, and exactly parallels the Rindler case. The original Schwarzschild
coordinates r and t with 2M G < r < ∞ cover only the right quadrant, and with 0 < r < 2M G the upper
quadrant. But the solution extends to a manifold which is actually double of the one originally described.
The existence of the left and lower quadrant could not have been guessed beforehand.
In the Rindler case the existence of a past and future horizon corresponds to an eternally accelerating
observer; likewise here it corresponds to the existence of an eternal black hole. As far as we know this is
an unphysical idealization. Ethernal black holes do not exist for the simple reason that the universe itself is
supposed to have emerged from a singularity a finite time ago. Any black holes that exist today must have
formed through astrophysical processes of accretion and gravitational collapse. For example in the case of a
star, the surface of the star is at constant r for a very long time. This corresponds to one of the hyperbolas
in the right quadrant, let’s say for T < 0. Suppose the star begins to collapse at T = 0. Then the surface
of the star describes a curve with decreasing r, eventually crossing the Schwarzschild radius and falling into
the singularity at r = 0. The point is that the interior solution, to the left of the surface, is not described
by the Schwarzschild metric. Therefore all that lies to the left of this line is unphysical. A real black hole
can only have a future horizon and a future singularity.

10