Lectures on black hole information and spacetime

wormholes
Henry Maxfield∗

February 1, 2021

Abstract
Lectures.

Contents
1 Introduction 2

2 Quantum black holes 3

2.1 The geometry of horizons . . . . . . . . . . . . . . . . . . . . 3
2.2 Conformal matter: energy dynamics . . . . . . . . . . . . . . 6
2.3 Hawking radiation . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 A reservoir for Hawking radiation. . . . . . . . . . . . . . . . 9
2.5 Black hole thermodynamics . . . . . . . . . . . . . . . . . . . 10
2.6 JT gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.7 Black holes in JT gravity . . . . . . . . . . . . . . . . . . . . 12
2.8 The Schwarzian description . . . . . . . . . . . . . . . . . . . 13
2.9 Symmetry origin of the Schwarzian . . . . . . . . . . . . . . . 15
2.10 The semiclassical approximation . . . . . . . . . . . . . . . . 16
2.11 Evaporation in JT . . . . . . . . . . . . . . . . . . . . . . . . 17
2.12 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Entropy and information in quantum systems 20

3.1 Fine-grained and coarse-grained entropy . . . . . . . . . . . . 20
3.2 The Page curve . . . . . . . . . . . . . . . . . . . . . . . . . . 22

∗
hmaxfield@physics.ucsb.edu

1
2 Quantum black holes
2.1 The geometry of horizons
We begin with a discussion of the geometry of black hole horizons in classical
general relativity, pointing out the key qualitative features for the physics
of Hawking radiation.
After a black hole forms, if it is isolated (not rapidly accreting matter,
for example), it will rapidly settle into a stationary state. Stationary solu-
tions to Einstein’s equations are typically characterised only by conserved
quantities: the mass, the angular momentum, and the charge under any
gauge fields (‘black holes have no hair’ [1]). Assuming spherical symmetry
for simplicity (no angular momentum), we can write any stationary metric
using ingoing coordinates r, v as

ds2 = −f (r)dv 2 + 2drdv + (transverse) (2.1)

for some function f (r). The ‘transverse’ piece describes the metric on the
orbits of spherical symmetry at constant v and r, and is given by the metric
on a sphere of some radius (a function of r). We will henceforth drop this
spherical part of the metric, concentrating on the two-dimensional geometry
at constant angle on the sphere.
The coordinate v is an ingoing time, constant on radial null geodesics
falling into the black hole. In general, we will define v to match the proper
time of a distant observer, at fixed but large r (see the discussion of asymp-
totics in a moment). For a black hole formed from collapse, (2.1) will be a
good approximation for ingoing times v sufficiently long after the black hole
has formed.
A black hole corresponds to a region r < r+ from which light cannot
dr
escape. From (2.1), outgoing null geodesics satisfy dv = 12 f (r), so they will
fail to escape to infinity if f (r) < 0. The event horizon of the black hole
is at r = rh , where rh is the largest r satisfying f (rh ) = 0. Generically, f
will vanish linearly at the horizon, so the surface gravity κ = 12 f ′ (rh ) will
be positive, and near to the horizon we can approximate the metric with
f (r) ∼ 2κ(r − rh ).
This immediately tells us something important: near to the horizon, the
outgoing null geodesics diverge exponentially, satisfying r − rh ∼ Aeκv . This
simple fact is at the root of much of the interesting physics of black holes.
More generally, outgoing null geodesics outside the horizon are at con-
stant values of the outgoing time u, defined analogously to v with respect to
the proper time of a distant observer. Concretely, it is given for the static

3
 1
metric (2.1) by u = v − 2r∗ (r), where r∗′ (r) = f (r) (often called the ‘tortoise
coordinate’). In terms of these lightcone coordinates u, v, we can then write
the metric in the exterior of the black hole as
ds2 = −f (r)dudv, (2.2)
where r is defined implicitly via r∗ (r) = v−u
2 . Close to the horizon, the
exponential divergence of outgoing null geodesics tells us that
A κ(v−u)
r − rh ∼ e (2.3)
2κ
for some A (the factor of 2κ inserted for later convenience). In particular,
the horizon r = rh itself lies at u → ∞, so our coordinates u, v cover only
the region r > rh .
At this point, let us be more specific about the asymptotic metric far
from the black hole, and the definitions of u and v. We will be interested in
two possibilities. First, for asymptotically flat spacetimes, we have f (r) → 1,
so
ds2 ∼ −dv 2 + 2drdv = −dudv,
(asymptotically flat) (2.4)
r ∼ v−u
2 → ∞.
2
In asymptotically AdS spacetimes, we have f (r) ∼ rℓ2 , and we define u, v so
that
r2 ! "2
ds2 ∼ − 2 dv 2 + 2drdv = − u−v
2ℓ
dudv,
ℓ (asymptotically AdS) (2.5)
2ℓ 2
r ∼ u−v → ∞.
In particular, the asymptotic boundary is timelike, at u = v = t, where t is
a ‘renormalised’ proper time (in AdS/CFT, t would correspond to the time
in the boundary dual theory).
Using our near-horizon coordinate change (2.3), we can write our near-
horizon metric in lightcone coordinates as
ds2 ∼ −Aeκ(v−u) dudv (r → rh , u ≫ κ−1 ). (2.6)
But, like any metric in a region much smaller than the curvature scale,
this is just two-dimensional Minkowski space in disguise. To see this very
explicitly, define new ‘Kruskal’ coordinates U, V , which in the near-horizon
region u → ∞ behave as
U ∼ − κ1 e−κu , V ∼ A κv
κe (2.7)
=⇒ ds2 ∼ −dU dV. (2.8)

4
The key result here is not that the metric is locally flat (since that is always
true), but rather the exponential relationship between the flat spacetime co-
ordinates U, V in the vicinity of the horizon and the coordinates u, v adapted
to the asymptotic region far from the black hole. The region of the near
horizon visible to an asymptotic observer for an unbounded range of outgo-
ing times u is exponentially compressed to a finite region. This is essentially
the same point that we emphasised before: outgoing geodesics diverge ex-
ponentially near the event horizon.
This fact is responsible for the phenomenon of Hawking radiation. Namely,
for any state of quantum fields that is regular at the event horizon, an ob-
server far from the black hole will see a flux of energy radiating from the
black hole.
To get some intuition for this, let’s see how time evolution acts. On
the coordinates u, v adapted to the asymptotic region, time translation by
a time t simply acts by addition of t. But in terms of our coordinates U, V
adapted to the near-horizon region, this becomes a boost:

u '→ u + t, v '→ v + t (2.9)

−κt +κt
U '→ e U, V '→ e V. (2.10)

So the Hamiltonian H = ∂u + ∂v for matter fields on the black hole back-

ground acts near the horizon like a boost generator K = V ∂V − U ∂U , with
proportionality constant given by the surface gravity: H ∼ κK.
Now, the vacuum state of any relativistic QFT on a half of two-dimensional
Minkowski space times any transverse manifold looks like a thermal state
with respect to the boost generator: ρhalf−space ∝ e−2πK . The underlying
reason is that the boost generator becomes a rotation in Euclidean signa-
ture, so we can think of ρhalf−space as enacting a rotation through angle
2π. More precisely, suppose we would like to compute the expectation value
Tr(Oe−2πK ) of some local operator O acting on the half space. The operator
e−2πK rotates the half-space by an angle 2π around its boundary, sweeping
out the whole of Euclidean space, so this expectation value is computed by
the path integral on flat Euclidean space. But (up to normalisation) this is
precisely the vacuum expectation value 〈0|O|0〉.
This is relevant because any nonsingular state in QFT looks like the
vacuum state at short distances, so the state of fields just outside the horizon
looks thermal in terms of the local boost generator. Identifying the near-
horizon boost with the asymptotic Hamiltonian, we see that the vacuum in
the near-horizon region appears to the asymptotic observer like a thermal

5
 2π
state at inverse temperature β = κ ,
giving us the Hawking temperature
κ
TH = . (2.11)
2π
This heuristic argument suggests that for any state of quantum fields that
looks like the vacuum near to the horizon, it will be interpreted as thermal
to the asymptotic observer.
Indeed, this conclusion is is borne out by more careful calculations, and
results in the physical effect of Hawking radiation: a flux of thermal radiation
escaping to infinity. To illustrate this, we now restrict to a class of theories
where the argument becomes extremely simple.

2.2 Conformal matter: energy dynamics

As a simple illustrative model, we will take the matter to be described
by a conformally invariant theory in two dimensions. If you would like
a concrete model, take some number of free massless bosons or fermions,
though everything we say will also apply to interacting matter. We do not
require any background in two-dimensional conformal field theory, and will
introduce the facts we need along the way (though feel free to consult [2, 3, 4]
for a more systematic and detailed account).
The dynamics of energy-momentum is extremely simple in these theories,
since it is determined completely by the symmetries. The energy-momentum
tensor has three independent components in two dimensions, and we have
two equations constraining their evolution from conservation. Adding con-
formal invariance provides us with a third equation fixing the trace, so we
can have a complete set of equations governing its evolution.
The defining property of conformal field theories is that there is a sym-
metry under changing the spacetime metric by a Weyl transformation: g '→
Ω2 g, where Ω is a local rescaling, which can be any function of spacetime.
This means that we change the notion of distance, but leave angles invariant.
In Lorentzian spacetime, this means that the causal structure (the space of
null vectors) remains invariant.
The energy-momentum tensor is defined (classically) through the varia-
tion of the CFT action ICFT as
2 δICFT
T ab (x) = − √ (2.12)
−g δgab (x)
(and in the quantum theory by metric variations of correlation functions).1
In particular, Weyl transformations (with δgab ∝ gab ) are generated by the
1
Note that this is the usual definition in most contexts (in particular, with this nor-

6
trace Tr T . To leave the correlation functions invariant, this means that
Tr T should vanish, and indeed it does in flat spacetime.
However, things are slightly more complicated in curved spacetime, since
the conformal symmetry is anomalous: that is, correlation functions are not
left invariant, but instead change in a predictable way. The anomaly is
determined only by an anomaly coefficient, in this case the ‘central charge’
c, some fixed positive constant (for a unitary theory). The upshot is that
the trace of T depends on the spacetime curvature:
c
Tr T = R, (2.13)
24π
where R is the Ricci scalar. For a full account of this anomaly, see §3.4
of [3] for example. Note that this is an operator equation, meaning that
Tr T is proportional to the identity operator in CFTs (except for contact
terms where Tr T becomes coincident with other operator insertions), and
in particular it holds in any state of the theory.
Writing a general metric in terms of lightcone coordinates u, v as

ds2 = −e2ω(u,v) dudv, (2.14)

the equation for the trace (2.13) becomes

c
Tuv = − ∂u ∂v ω. (2.15)
12π
Using this, the conservation equations ∇a T ab = 0 become equations for
∂v Tuu and ∂u Tvv with solutions
c # $
Tuu = ∂u2 ω − (∂u ω)2 + Fu (u), (2.16)
12π
c # 2 $
Tvv = ∂v ω − (∂v ω)2 + Fv (v), (2.17)
12π
where Fu , Fv are functions only of u, v respectively.
Now, in either asymptotically flat or asymptotically AdS spacetimes, the
inhomogeneous terms (depending on u, v) vanish at large r, so Fu and Fv
tell us the asymptotic value of the stress-energy:

Fu (u) = asymptotic outgoing energy flux,

(2.18)
Fv (v) = asymptotic incoming energy flux.
malisation T00 is interpreted as the local energy density), but differs from a common 2D
CFT convention by a factor of −2π.

7
Exercise 1. Check these results. For what spacetimes can we ignore the
inhomogeneous terms, so Tuu is just a function of u and Tvv is a function
of v? (Hint: compare the conservation equations to the derivatives of the
curvature.)
In the quantum theory, the trace equation (2.13) and the conservation
equations apply to correlation functions when stress-energy tensor is inserted
away from any other operators or matter sources. In particular, our results
(2.16), (2.17) and (2.15) hold for the one-point function 〈Tab 〉 in any state.

2.3 Hawking radiation

Let’s now apply this result to conformal matter in a fixed static black hole
background, looking at the stress-energy in the near-horizon region. From
(2.6) we have ω ∼ κ2 (v − u), so
c 2
Tuu ∼ − κ + Fu (u), (2.19)
48π
c 2
Tvv ∼− κ + Fv (v). (2.20)
48π
Now, u is not a good coordinate at the horizon (where it goes to infinity).
To get an idea of the physical energy density near the horizon, we can use
instead the r coordinate. We then have
c
Fu (u) − 48π κ2
Trr ∼ . (2.21)
κ2 (r − rh )2
But for any quantum state which is nonsingular at the horizon, Trr must
c
remain finite there. This requires that Fu must approach 48π κ2 as u → ∞
to cancel the inhomogeneous term in (2.19), plus terms that go to zero at
least as fast as (r − rh )2 . We therefore have
c 2
Fu (u) = 2π T + O(e−2κu ), (2.22)
24 H
where we used (2.11) to rewrite the surface gravity κ in terms of the Hawking
temperature TH . This is precisely the expectation value of Tuu for outgoing
thermal radiation at temperature TH . The conclusion is that for any state
that is nonsingular at the horizon, an asymptotic observer will see a thermal
flux of radiation at the Hawking temperature!
The exponential correction to (2.22) at late times corresponds to the
energy density at the horizon in the immediate aftermath of the formation
of the black hole. The coefficient of the correction term is proportional to

8
e2κv0 if the black hole is formed at time v0 , times the energy density at the
horizon at that time. This exponentially decaying transient energy can be
thought of as a simple example of a quasinormal mode, which describes the
black hole’s approach to equilibrium.
Now we look at the near-horizon value of Tvv , which gives the flux of
energy falling through the horizon into the black hole. From (2.20), this
flux is given by the energy coming in from infinity, minus a constant which
matches the outgoing flux from Hawking radiation. This is in accord with
conservation of energy: the energy absorbed by the black hole is given by
the incoming energy flux minus the outgoing energy flux as measured at
infinity.
If there is no energy incoming from infinity (Fv = 0), the negative flux
c
Tvv = − 48π κ2 of energy across the horizon will provide a source for Einstein’s
equations, altering the metric. Specifically, it will cause the black hole to
shrink as it loses energy via Hawking radiation. This is the process of black
hole evaporation, which we will describe explicitly in the next section, after
we introduce a model for the gravitational dynamics.
Alternatively, we can keep the black hole in equilibrium by producing
c
an ingoing flux of energy Fv = + 48π κ2 from infinity, balancing the outgoing
Hawking radiation. One way to do this is by ‘putting the black hole in a
box’: rather than letting the Hawking radiation escape, we reflect it back so
that the black hole attains equilibrium with its own Hawking radiation. This
is the usual situation in asymptotically AdS spacetimes: energy-conserving
boundary conditions require Fu (t) = Fv (t), so black holes do not evaporate.2

2.4 A reservoir for Hawking radiation.

Our simple model will describe black holes in AdS2 , but would like them
to be able to evaporate. To this end, we will not use the usual reflecting
boundary conditions, but instead couple the matter theory to an auxiliary
‘reservoir’ system at the boundary. This reservoir will simply be described
by the same matter CFT, living on a fixed flat spacetime, given by half of
Minkowski space ds2 = −dudv, with v > u. The boundary of this space at
u = v = t will be joined to the boundary of our dynamical asymptotically
! "2
2ℓ
AdS spacetime, with asymptotic metric ds2 ∼ − u−v dudv for u > v as
2
This need not apply to very small black holes in more than two spacetime dimensions.
The energy takes a time of order the AdS time ℓ to travel to infinity and reflect back,
which can be longer than the black hole’s lifetime if it is sufficiently small.

9
in (2.5).

ds2 = −e2ω(u,v) dudv (2.23)

Reservoir: v > u, fixed background ω = 0.
(2.24)
Black hole: u > v, ω dynamical.

This metric is not continuous at the boundary where we couple the two sys-
tems: indeed, it is singular as we approach it from the AdS side! But because
our matter is conformally invariant, we may first define the theory on any
′
smooth, flat metric related by a Weyl transformation, ds2 = −e2ω (u,v) dudv.
Correlation functions in the physical metric are then determined by applying
an appropriate Weyl transformation (including the anomaly (2.13)). Indeed,
we have already seen one example of this Weyl transformation in (2.16),
(2.17), where the inhomogeneous terms are the contribution of the Weyl
anomaly transforming from the flat metric −dudv to the physical black hole
metric −e2ω dudv. This allows matter excitations to pass freely between the
gravitational black hole spacetime and the reservoir. In particular, Hawking
radiation can escape from the black hole into the reservoir, where it will
propagate freely away to future null infinity (v → ∞ at fixed u).

2.5 Black hole thermodynamics

We saw above that a black hole will remain stationary if we put it in contact
with a system of temperature TH . But, essentially by definition (the zeroth
law of thermodynamics), this means that the black hole itself has tempera-
ture TH ! We can then use the first law (along with the energy of the black
hole, determined from the metric far away) to assign an entropy,
%
dE
SBH (E) = . (2.25)
TH (E)

This is the Bekenstein-Hawking entropy, and (as we will see later) it can
always be described in terms of the geometry of the event horizon. For
Einstein gravity, it is given by the famous formula
A
SBH = , (2.26)
4GN
where A is the area of the event horizon.
Deriving the entropy from the temperature is somewhat backwards from
a historical point of view. Bekenstein proposed that the area of a black
hole should be assigned an entropy, by considerations of the second law [5].

10
Hawking set out to prove him wrong by showing that black holes didn’t have
a temperature, so couldn’t possibly be assigned an entropy: of course, he
instead discovered that phenomenon of Hawking radiation we just described
above [6].
We usually think of this sort of equilibrium thermodynamics as arising
from a quantum statistical description: namely, an entropy is the logarithm
of the number of quantum states in a small window of energy. Does the
Bekenstein-Hawking entropy count states in this way? This is the key ques-
tion that we will be testing later.

2.6 JT gravity
So far, we have been discussing only the fixed geometry of a static black
hole, and the physics of quantum fields on that background. It is time now
to move beyond this, and to incorporate gravitational dynamics. We will
make our lives as easy as we can by discussing perhaps the simplest theory
possible: Jackiw-Teitelboim (JT) gravity [7, 8, 9].
At first you may like to study pure Einstein gravity, with action
&% % '
1 2 √
IEH = d x −g R + 2 ds K . (2.27)
16πGN M ∂M

But in two dimensions this is purely topological: the variation with respect
to the metric (the Einstein tensor) is identically zero. This does not give
sensible and interesting dynamics, and cannot be coupled to matter (clas-
sically, at least: Einstein’s equations set the matter stress-energy to zero,
Tab = 0).
We must therefore add another term to the action. The simplest possi-
bility is to introduce a scalar ‘dilaton’ field φ, and add the term
% %
1 2 √
IJT = d x −g φ (R + 2) + ds φ (K − 1) . (2.28)
2 M ∂M

This may look like an arbitrary choice, but in fact this theory emerges
very naturally from studying near-extremal (low temperature) black holes
in higher dimensions, for example Reissner-Nordstrom black holes in four
dimensions. In such cases, the Einstein term is proportional to the area of
the event horizon of the extremal black hole, and φ describes deviations of
the area of the transverse sphere from its extremal value. The JT action
is then the leading order approximation (linear in φ) when these deviations
are small. This has a description in terms of a weakly broken symmetry,
which we will briefly describe later.

11
 We can now begin to understand why this theory is simple. Since φ
appears linearly, its equation of motion simply becomes the constraint

R = −2, (2.29)

so our solutions have constant negative curvature. We have chosen units

to set the curvature scale ℓ to unity. In fact, something stronger is true,
since in the quantum theory we can integrate out φ and impose (2.29)
as a delta-function, constraining the path integral only to such constant-
curvature metrics.
The total action we will study is given by

I = IEH + IJT + Imatter + Icounterterm . (2.30)

Imatter is the action of our (conformal) matter theory, which propagates on

the metric g: in particular, it does not couple directly to the dilaton, so
we do not spoil the property above. We have allowed for the possibility of
counterterms (which will not be too important for us, though we will have
occasional cause to mention them).
Lastly, we discuss the boundary conditions, which we impose asymp-
totically. Take large dilaton, φ = γ' where we will take * → 0. Writing
the intrinsic metric on the boundary as ds2 = − '12 dt2 defines the physical
boundary time t. This time defines our lightcone coordinates u, v at infinity
(u = v = t), and our coupling to the reservoir as described in section 2.4.

2.7 Black holes in JT gravity

Writing our metric in ingoing coordinates r, v as above, the curvature is
given by R = −∂r2 f . We can therefore write the most general constant
curvature metric as

ds2 = −(r2 − rh (v)2 )dv 2 + 2drdv (2.31)

for some function rh (v), where we have used the freedom to shift r by a
function of v alone to remove a term linear in r. The boundary is then at
large r, r = 1' + O(*).
Now, on constant curvature metrics R = −2, the bulk part of the JT ac-
tion vanishes, and the action is purely a boundary term. Using the boundary
condition φ∂ = γ' , we can evaluate this on the metric (2.31), finding
% t
γ
IJT = − dv rh (v)2 (2.32)
2 t0

12
in the limit * → 0. In particular, from this we can read off the energy: this
is given by the Hamilton-Jacobi equation E = − ∂I ∂t , where t is the endpoint
of the integral computing the action I, so
E = 12 γrh2 . (2.33)
Let us look now at the classical solutions, first without matter. In addi-
tion to the constant curvature condition, we need to satisfy the equation of
motion from varying the metric,
∇a ∇b φ − gab ∇2 φ + gab φ = 0. (2.34)
The rr component and rv component, along with the boundary conditions,
uniquely specify a simple solution:
φ = γr. (2.35)
The vv component then fixes ∂v rh = 0, so the energy (2.33) is independent
of time as we would expect, giving us a static black hole solution.
The surface gravity is given by κ = rh , so in the presence of of matter
we expect the black hole to radiate at the Hawking temperature
rh
T = . (2.36)
2π
Now that we have the energy and temperature, we can use the first law
dE = T dS to find the entropy:
S = S0 + 2πγrh = S0 + 2πφh = S0 + γ(2π)2 T. (2.37)
We have here included an integration constant S0 , which is not fixed by the
first law. We will later find a natural way to fix this using the Einstein-
Hilbert term IEH in the action, but for now it is arbitrary.

2.8 The Schwarzian description

Before adding matter, we will give another description for the metric (2.31),
which relates JT gravity to the ‘Schwarzian theory’ [10, 11, 12].
To arrive at this, we can observe that the constant curvature constraint
R = −2 uniquely fixes the metric to be AdS2 , since the scalar curvature
completely fixes the geometry in two dimensions (up to topology). There
must therefore be some other coordinates U, V such that the metric can be
written in the usual ‘Poincaré’ form
( )2
2 2
ds = − dU dV. (2.38)
U −V

13
The asymptotic boundary is at U = V = T , but T will not coincide with
our physical time t: instead, we have some diffeomorphism F relating them
as T = F (t). This diffeomorphism determines U, V in terms of our usual
lightcone coordinates as U = F (u), V = F (v), so we can write the metric as
( )2
2
2
ds = − F ′ (u)F ′ (v)dudv. (2.39)
F (u) − F (v)

Finally, we can recover the form of our ingoing coordinates (2.31) by defining

F ′′ (v) F ′ (v)
r= + 2 . (2.40)
F ′ (v) F (u) − F (v)

The final step is to identify the horizon radius rh in terms of the diffeomor-
phism F , which gives us

rh (t)2 = −2 Schw(F, t), (2.41)

where Schw is the ‘Schwarzian derivative’:

( )2
F ′′′ (t) 3 F ′′ (t)
Schw(F, t) = ′ − . (2.42)
F (t) 2 F ′ (t)

Putting this expression for rh2 into our action (2.32) gives us the Schwarzian
action %
ISchw = γ dt Schw(F, t), (2.43)

and the energy is

E = −γ Schw(F, t). (2.44)
The constant γ is not a dimensionless coupling constant (it has units of
time), so we should think of it as setting a scale in the theory. Specifically,
γ −1 gives an energy or temperature scale below which the theory becomes
strongly coupled. For large energies γE ≫ 1, the action will be large, which
suppresses quantum fluctuations so we may treat the theory classically. We
will study such large energy black holes, so gravitational quantum fluctu-
ations are a small effect. Since we have fixed units in AdS by setting the
curvature scale ℓ to unity, we take rh to be of order one and γ ≫ 1 in these
AdS units.
We can now describe the gravitational dynamics in terms of this diffeo-
morphism F with the Schwarzian action. However, importantly, not every

14
diffeomorphism gives rise to a different metric: the physical configuration is
invariant under the fractional linear transformations
aF + b
F '→ , a, b, c, d ∈ R, ad − bc = 1, (2.45)
cF + d
which form the group P SL(2, R). This redundancy arises from the symme-
tries of AdS2 . In particular, this means that the Schwarzian action is not in
fact a local functional of time, despite appearances: the apparent locality is
spoiled by the nonlocal action of the P SL(2, R) gauge redundancy.
For avoidance of possible confusion, we should note that the Lorentzian
signature description in terms of diffeomorphisms modulo P SL(2, R) is valid
only locally, since the range of F will generically not be R, and fractional
linear transformations introduce poles. Things are slightly more straight-
forward in the Euclidean description, discussed briefly in a moment.
We can now write our static classical solutions without matter in terms
of the Schwarzian variable. Up to a fractional linear transformation, the
solution is
F (t) = − exp (−rh t) . (2.46)
Note that F describes the change of variables U = F (u) between the outgo-
ing coordinate u and an outgoing ‘Kruskal’ coordinate U which is smooth
at the horizon. The exponential form of F is a signature of the exponential
divergence of outgoing null geodesics near the horizon.

2.9 Symmetry origin of the Schwarzian

This section provides a little background about the Schwarzian theory: it is
not necessary for what follows.
The Schwarzian theory described above makes an appearance in other
contexts, most notably in the SYK model, a quantum mechanical model of
N ≫ 1 Majorana fermions with random interactions. The fact that it crops
up in somewhat generic circumstances can be understood from symmetry:
the Schwarzian can be thought of as a pseudo-Nambu-Goldstone boson for
a particular pattern of symmetry breaking.
To do this, it will be slightly simpler to discuss the Euclidean theory, with
periodic Euclidean time tE ∼ tE + β describing the theory at temperature
T = β −1 . The metric is then Euclidean AdS2 , which we think of as the
Poincaré disk. The Schwarzian diffeomorphism gives the map from physical
time tE to an angular coordinate φ on the boundary, where F = tan φ2 ,
and φ is 2π periodic. This means that we have a map from the tE circle

15
to the φ circle, so the configurations of the theory (the gauge orbits of the
diffeomorphism) take values in the coset

[φ] ∈ Diff(S 1 )/P SL(2, R). (2.47)

If we were interested in locally AdS spacetimes with only the Einstein-

Hilbert action, this coset describes spontaneous symmetry breaking. Since
the Einstein theory is topological, it has zero Hamiltonian, and can only
describe degenerate states at zero energy.3 All diffeomorphisms of time
them become symmetries. But any given solution does not respect this full
symmetry group, but only the isometry group of AdS: in other words, it
spontaneously breaks Diff(S 1 ) to P SL(2, R), and F is a Goldstone boson
for that breaking.
To get an interesting theory with dynamics, we must include a leading
order correction which explicitly breaks the symmetry. That comes from
writing down an action on the coset with the lowest dimension possible,
which is precisely the Schwarzian action. In JT gravity, this explicit breaking
comes from the dilaton. F is then only a ‘pseudo-Goldstone’.
This can be compared to chiral symmetry breaking in QCD. With NF
massless quarks, there is a chiral symmetry SU (NF )×SU (NF ), which is bro-
ken spontaneously to the diagonal SU (NF ) subgroup. We then have Gold-
stone bosons — pions — living on the coset SU (NF ) × SU (NF )/SU (NF ).
Unlike the Schwarzian, this is an interesting theory because we can write
down an action with kinetic terms on the coset (the chiral Lagrangian).
Giving the quarks small masses (compared to the scale of the symmetry
breaking) provides a small explicit symmetry breaking, which can be de-
scribed by adding a term which does not respect the full symmetry of the
coset. The pions are then described as pseudo-Goldstone bosons.

2.10 The semiclassical approximation

We would now like to incorporate matter. Before getting to the specific de-
tails, we first outline the general strategy. Ideally, we would like to compute
the full path integral over gravitational variables (the metric g and dilaton
φ) and matter fields X. But even for such a simple model, that’s not so
3
This is related to the fact that there is no nontrivial scale-invariant theory in one
dimension, so we must break that symmetry [13]. Without introducing a scale, the density
of states can only be of the form eS0 δ(E) + EC
. The second term is unphysical, describing
a continuum of infinitely many states, so we only have the first: a space of degenerate
zero-energy states. This is avoided in higher dimensions, since the volume of space sets a
scale.

16
easy, so we will make use of a semiclassical limit, looking for saddle-points
where the action is stationary. But simply looking for saddle-points of the
classical action will not quite be sufficient, because over a long period of
time, quantum effects from the matter (namely, production of Hawking ra-
diation) build up to become important. We therefore use an intermediate
‘semiclassical’ approach, treating the matter fully quantum mechanically,
but looking for saddle-points in the integral over gravitational variables.
The full quantum mechanical matter theory is captured by the quantum
effective action Ieff . This is a functional only of the metric, obtained by
integrating out the CFT fields X:
%
eiIeff [g] = DX eiICFT [X,g] . (2.48)

This implicitly depends on our boundary conditions, in particular the state

of the matter and any operators we are inserting to compute an expectation
value. We then look for stationary points of the action IJT for the metric
and dilaton plus the quantum effective action Ieff ,

δIJT + δIeff = 0, (2.49)

for any metric variation.

By definition, varying the effective action with respect to the metric g
gives us the expectation value of the stress tensor (inserted in whatever
correlation function we want to compute). Saddle-points therefore corre-
spond to solutions to the metric equation of motion (2.34) sourced by the
expectation value of the CFT stress tensor:

∇a ∇b φ − gab ∇2 φ + gab φ + 〈Tab 〉 = 0. (2.50)

Our task is to find solutions to this equation, giving us the quantum-

corrected geometry.

2.11 Evaporation in JT
Now, to determine the expectation value 〈Tab 〉 we have to choose some state.
However, as we saw from studying the static black hole spacetime, the dy-
namics quickly becomes independent of this choice. Specifically, any state
which is nonsingular at the horizon at some time v0 will have exponentially
small 〈Trr 〉 a little later, decaying as e−2κ(v−v0 ) . The timescale κ−1 for this
decay is much shorter than the time over which appreciable evaporation

17
takes place, so our static near-horizon analysis from earlier remains appro-
priate. We will therefore choose a state with 〈Trr 〉 = 0, as an excellent
approximation to any state with a smooth horizon at earlier times.
It will also be convenient to add a counterterm to cancel the source from
the trace of the stress tensor, given by the anomaly (2.13). Because the met-
ric is constant curvature, the anomaly gives just a constant source: this can
be
* cancelled with a finite cosmological constant counterterm, proportional to
√
d2 x −g. This amounts only to a finite redefinition of parameters, which
can be absorbed by shifting the dilaton. The upshot is that we will simply
set 〈Tr T 〉 = 0.
That leaves the only nonzero component of 〈Tab 〉 as the ingoing energy
c
〈Tvv 〉 = Fv (v) − rh (v)2 , (2.51)
48π
where we have !used the conservation
" equation (2.17) with conformal factor
1 4F ′ (u)F ′ (v)
ω(u, v) = 2 log (F (u)−F (v))2 to write it in terms of the flux from infinity.
With this simple expression for the stress tensor expectation value, it
is simple to solve the equation of motion (2.50) as before. In fact, we still
have the same result as (2.35) relating the dilaton to r (φ = γr), but now
we have interesting dynamics for rh , or equivalently the energy E:
c
Ė(t) = Fv (t) − E(t). (2.52)
24πγ
This simply tells us that the change in the energy equals the ingoing flux,
minus the outgoing flux of Hawking radiation.
Let us now suppose that there is no incoming energy, Fv = 0. The energy
then decays exponentially:

E(t) = E(0)e−2kt =⇒ rh (t) = rh (0)e−kt , (2.53)

where the decay rate is given by

c
k= ≪ 1. (2.54)
48πγ
Now, the exact solution for the diffeomorphism F is slightly compli-
cated, involving Bessel functions. But there is a very simple approximate
solution, which is extremely close to the exact solution whenever the black
hole remains in the semiclassical regime γE ≫ 1:
! " # $
F (t) ∼ − exp k1 rh (0)e−kt = − exp k1 rh (t) . (2.55)

18
 c
In fact, this solves (2.52) with a very weak constant incoming flux Fv = 48π k2
k
of radiation at temperature 2π ≪ 1, which is negligible until extremely late
times. Note that over any short period of time, this is well approximated
by our static solution (2.46). Specifically, write t = t0 + ∆t where t0 can be
of order k −1 , but ∆t ≪ k −1 . Then rh (t) ∼ rh (t0 )(1 − k∆t), so
# $
F (t) ∼ − exp k1 rh (t0 ) e−rh (t0 )∆t . (2.56)

2.12 Horizons
To complete our discussion of the geometry of the evaporating black hole,
we can look at two notions of ‘horizon’. First, in this dynamical geometry,
r = rh is no longer the event horizon: outgoing light from there will escape
to infinity. Instead, it is an ‘apparent horizon’. In the JT context, the
apparent horizon is defined by the criterion that the dilaton + is stationary
∂φ +
to first order variation along outgoing null geodesics: ∂v + = 0. In higher
u
dimensions, we define an apparent horizon as a surface with stationary area
along outgoing null rays (a marginally trapped surface).
If the null energy condition is satisfied, an apparent horizon is destined
to lie inside an event horizon. The reason is that lightrays focus (since
gravity is attractive), so if the area is stationary to first order it can only
decrease. This focusing theorem is the key idea behind the Hawking-Penrose
singularity theorems.
The event horizon is defined by the surface where u → ∞. Using our
approximate solution (2.55) and (2.40), the event horizon lies at

rEH (v) ∼ rh (v) − k, (2.57)

just inside the apparent horizon. This is possible because quantum matter
does not obey the null energy condition. Indeed, we saw earlier that there is
a flux of negative energy through the horizon, a violation of the null energy
condition.
d
Exercise 2. Let N a = dλ be the tangent vector of a null geodesic with affine
d
parameter λ, and define the ‘divergence’ of the geodesic by θ = dλ φ. Using
the equation of motion (2.50), find the JT gravity version of the ‘Raychaud-
huri equation’:
dθ
= −N a N b Tab . (2.58)
dλ
Show that under the null energy condition N a N b Tab ≥ 0, a trapped surface
(θ < 0 for the outgoing direction) must lie inside an event horizon.

19
3 Entropy and information in quantum systems
We now take a brief detour, leaving black holes to one side and considering
entropy in quantum systems more generally. This will set the stage for
studying entropy and information for the evaporating black hole.

3.1 Fine-grained and coarse-grained entropy

First, there are several different notions of entropy, so we should be precise
about what we mean.
The main thing we will be interested in is the von Neumann entropy of
a quantum state with density matrix ρ:

S(ρ) = − Tr(ρ log ρ). (3.1)

In many cases we will be looking at the state of a subsystem, in which case we

use the reduced density matrix. In that context the von Neumann entropy is
often called the entanglement entropy, since if the full state is pure, the von
Neumann entropy of a subsystem is a measure of the quantum entanglement
between the subsystem and its complement.
The von Neumann entropy is distinct from the entropy we meet in ther-
modynamics. In particular, the second law tells us that the thermodynamic
entropy will typically increase over time, even if we have an isolated system.
But the von Neumann entropy of an isolated system does not change: the
density matrix evolves as ρ '→ U (t)ρ U (t)† for the unitary time-evolution
operator U (t) = e−iHt , which does not affect S(ρ). We need a different
notion of ‘coarse-grained’ entropy.
Coarse-graining means that we regard many states as indistinguishable
if they are the same for some set O of simple ‘macroscopic’ observables O:

O = {operators O corresponding to macroscopic observables} . (3.2)

This clearly involves a choice, and we will mention a couple of possibilities

in a moment. Having chosen O, we can then define a coarse-grained entropy
SO by maximising the von Neumann entropy S(ρ̃) over all states ρ̃ with the
specified expectation values for macroscopic observables O ∈ O:

SO(ρ) = max {S(ρ̃) : Tr(Oρ̃) = Tr(Oρ) for all O ∈ O} . (3.3)

We can define a coarse-grained density matrix ρO as the (unique) ρ̃ that

attains this maximum. A ‘coarse-graining’ means choosing a smaller set of
macroscopic observables.

20
Exercise 3. Show that the coarse-grained density matrix ρO (attaining the
maximum in (3.3)) is unique.
In a thermodynamic limit, for which there are many orthogonal states
N ≫ 1 with observables close to the specified values, we may think of the
entropy as providing us with a count of those states: SO ∼ log N .
The simplest example, most relevant to systems close to equilibrium, is
the canonical entropy, where O = {H}, so we fix only the energy Tr(ρH) =
E. A state maximising the entropy under such a constraint is given by the
canonical density matrix

e−βH
ρcan (β) = , Z(β) = Tr(e−βH ) (3.4)
Z(β)
1
for some inverse temperature β = T. The canonical entropy is then the von
Neumann entropy of this state:

Scan (β) = S(ρcan (β)). (3.5)

This can also be expressed in terms of the free energy F :

∂F
Scan = − , F = −T log Z. (3.6)
∂T
If there are other conserved quantities (like charge or angular momentum)
which do not equilibrate, we might like to add them to the list O, considering
a grand canonical ensemble.
If our system is not in equilibrium, we might like to use a slightly more
refined choice of coarse-graining. For example, we might specify the density
of energy and momentum (and perhaps other conserved quantities) as a
function of space, as in hydrodynamics where equilibrium is only attained
locally.
For us, coarse-grained entropy will be useful to provide a bound on fine-
grained (von Neumann) entropy. By definition, entropy increases under
coarse graining. That is, if O is more coarse-grained that O′ , meaning that
it contains fewer macroscopic observables O ⊂ O′ , then SO ≥ SO′ . This
follows simply because in (3.3), we are maximising with fewer constraints.
The extreme example is the von Neumann entropy, for which we take O′ to
contain all operators:
S(ρ) ≤ SO(ρ). (3.7)
We will suppress the choice of coarse-graining, from here on denoting SO
as a ‘thermal entropy’ Sthermal . We assume some choice which is sufficiently

21
detailed to capture the local properties and dynamics of the state, so that
our bound (3.7) approaches the best possible without considering detailed
correlations.
For a black hole, the ‘no hair theorem’ suggests that the canonical (or
grand canonical) entropy is a good choice of coarse-graining, and we will
work under the hypothesis that this is given by the Bekenstein-Hawking
formula obtained from the first law in section 2.5. For the Hawking radia-
tion, a better choice involves the local stress-tensor 〈Tab (x)〉, since radiation
emitted at different times does not equilibrate.

3.2 The Page curve

Let us now apply these ideas to learn about the state of a quantum system as
it loses energy to its surroundings. Specifically, prepare a many-body system
A in some excited pure state, and weakly couple it to a large reservoir (or
‘bath’) B prepared in its ground state. Then, energy will typically flow
from A to B, until (if B is sufficiently large) A cools to its own ground
state. Ultimately, we will be interested in an evaporating black hole, but
for now it might be helpful to think of a more ordinary system. Perhaps A
could be a lump of metal, and B a lab in which it sits: as the metal cools,
it fills the lab with radiation.
Now, since the coupled system evolves unitarily and the final state of A
is pure (the ground state), the final state of B must also be pure. So the
von Neumann entropy of A or B (i.e., their entanglement entropy) begins
at zero and ends at zero. But unitarity tells us more about the evolution of
the von Neumann entropy over time, since we have the bound (3.7) by the
coarse-grained thermodynamic entropy. This in fact gives us two separate
bounds, by the thermal entropy of either subsystem:
SvN ≤ min {Sth (A), Sth (B)} . (3.8)
At early times, the thermal entropy of B will be small and the thermal
entropy of A large, so the second bound will be relevant. At late times,
the first bound will be relevant instead. Moreover, we typically expect the
tighter bound to be extremely close to saturation (for a sensible choice of
coarse-graining), so that (3.8) becomes an approximate equality.
The resulting function
SPage (t) = min {Sth (ρA (t)), Sth (ρB (t))} (3.9)
is called the ‘Page curve’, since it was introduced by Don Page as a quan-
titative expectation of unitary black hole evaporation. A typical example

22
is shown in a figure. There is an initial steady increase (following Sth (B)),
until the ‘Page time’ tPage at which there is a sharp transition to a steady
decrease (following Sth (A)).

Exercise 4. Take system A to be a box containing weakly-interacting rel-

ativistic bosons (photons, perhaps) in d dimensions of space, so the energy
at temperature T is E = CT d+1 for some constant C (proportional to the
volume of the box). Allow the photons to escape through a small hole into
the outside world (system B), so some small fraction λ of them are emitted
per unit time. Assume that this process is slow compared to the equilibration
time of the system, so that the photons remain in equilibrium as the energy
leaks out. (Note that for d = 1, this is also describes our evaporating black
hole model in section 2.1!)
What is the rate of increase of thermal entropy for system B? What
proportion of the energy has escaped at the ‘Page time’ ? Compare the final
thermal entropy of B to the initial thermal entropy of A.

To provide some justification for why we might expect the bound (3.8) to
be close to saturation, we can make a crude model for the dynamics. Namely,
we model systems A and B as finite-dimensional Hilbert spaces HA , HB with
dimensions dim HA ≈ eSth (A) , dim HB ≈ eSth (B) given by their respective
thermal entropies, and take the state to be chosen at random from the tensor
product HA ⊗ HB . Roughly, this model says that the dynamics produce a
pure state on AB with the correct macroscopic observables, but which is
otherwise completely generic. In this model, the average entropy is very close
to maximal: the difference between the two sides of (3.8) is exponentially
small in the difference |Sth (A) − Sth (B)| between the two thermal entropies.
We expect this to be a reasonable approximation if the dynamics of A
is sufficiently chaotic. The time-evolution operator should be sufficiently
generic after some timescale — the ‘scrambling time’ tscr — on which local
information is spread over the whole system. As long as this timescale is
much shorter than the time over which evaporation is significant, we expect
the Page curve to be an excellent approximation. A reasonable estimate for
the scrambling time of a strongly-interacting system is tscr ∼ β log Sth (A).
The number of degrees of freedom effected by a small local perturbation will
tend to grow exponentially at a rate set by the timescale of local interactions,
which might typically be of order β. The effect of the perturbation will then
be spread over the whole system after a time which is logarithmic in the
number of excited degrees of freedom.

23
References
[1] C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation. W. H.
Freeman, San Francisco, 1973. 3

[2] P. Di Francesco, P. Mathieu, and D. Senechal, Conformal Field

Theory. Graduate Texts in Contemporary Physics. Springer-Verlag,
New York, 1997. 6

[3] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic

string. Cambridge Monographs on Mathematical Physics. Cambridge
University Press, 12, 2007. 6, 7

[4] X. Yin, “Aspects of Two-Dimensional Conformal Field Theories,” PoS

TASI2017 (2017) 003. 6

[5] J. D. Bekenstein, “Black holes and entropy,” Phys. Rev. D 7 (1973)

2333–2346. 10

[6] S. Hawking, “Particle Creation by Black Holes,” Commun. Math.

Phys. 43 (1975) 199–220. [Erratum: Commun.Math.Phys. 46, 206
(1976)]. 11

[7] R. Jackiw, “Lower Dimensional Gravity,” Nucl. Phys. B 252 (1985)

343–356. 11

[8] C. Teitelboim, “Gravitation and Hamiltonian Structure in Two

Space-Time Dimensions,” Phys. Lett. B 126 (1983) 41–45. 11

[9] A. Almheiri and J. Polchinski, “Models of AdS2 backreaction and

holography,” JHEP 11 (2015) 014, 1402.6334. 11

[10] J. Maldacena, D. Stanford, and Z. Yang, “Conformal symmetry and

its breaking in two dimensional Nearly Anti-de-Sitter space,” PTEP
2016 (2016), no. 12 12C104, 1606.01857. 13

[11] K. Jensen, “Chaos in AdS2 Holography,” Phys. Rev. Lett. 117 (2016),
no. 11 111601, 1605.06098. 13

[12] J. Engelsöy, T. G. Mertens, and H. Verlinde, “An investigation of

AdS2 backreaction and holography,” JHEP 07 (2016) 139,
1606.03438. 13

24
[13] K. Jensen, S. Kachru, A. Karch, J. Polchinski, and E. Silverstein,
“Towards a holographic marginal Fermi liquid,” Phys. Rev. D 84
(2011) 126002, 1105.1772. 16

25