Three lectures on variational

principles

David Kubizňák
Institute of Theoretical Physics

January 10, 2024

Contents

Contents i

1 Lecture 1: Variational principles in GR 1

1.1 Matter in a curved spacetime . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Einstein–Hilbert action . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Higher-curvature gravities . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Lecture 2: Submanifolds & York–Gibbons–Hawking term 9

2.1 Gauss–Codazzi formalism . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 York–Gibbons–Hawking boundary term . . . . . . . . . . . . . . . . . . 13

3 Lecture 3: Black hole thermodynamics 17

3.1 Motivational foreplay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Euclidean Trick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Euclidean action calculation . . . . . . . . . . . . . . . . . . . . . . . . 21

Bibliography 24

i
Chapter 1: Lecture 1: Variational
principles in GR

1.1 Matter in a curved spacetime

ˆ Coupling to gravity. We have seen that the action for a particle in curved space-
time can be obtained by simply replacing the Minkowski metric by the general
metric. This is an example of the minimal coupling principle. The recipe is to
write the new laws in the tensorial form so that they reduce to the special rela-
tivistic laws in a local inertial frame. In practice this amounts to replacing the
Minkowski metric with a general metric, and partial derivatives with covariant
derivatives:
ηµν → gµν , ∂µ → ∇µ . (1.1)
Of course, this procedure is vague and not unique (as one can always add for
example the curvature terms) and the final say about whether the theory is right
or wrong is decided (as always in physics) by experiment.
ˆ Action principle. To write the action for the matter fields we have to use the
invariant volume element, together with the principle of minimal coupling, to
write
√
Z
Sm = dd x −gLm (φ, ∇φ, g) , (1.2)

where Lm is the scalar Lagrangian density and φ stands for various fields. The
variation gives
√  ∂Lm
Z Z
d δSm ∂Lm 
δSm = d x δφ = dd x −g δφ + δ∇µ φ . (1.3)
δφ ∂φ ∂∇µ φ
As always, we now
R interchange the Rcovariant derivative with δ, and, using the
µ
√ √
Stokes theorem, Σ ∇µ V −gd x = ∂Σ γV µ dΣµ , integrate by parts as follows:
d

√ √
Z Z
A (∇µ B) −gd x = − B(∇µ Aµ ) −gdd x + boundary term ,
µ d
(1.4)

to have
√ h ∂Lm
Z  ∂L i
m
δSm = dd x −g − ∇µ δφ . (1.5)
∂φ ∂(∇µ φ)
Thus we have derived the generalized Euler–Lagrange field equations:
δSm ∂Lm  ∂L 
m
=0 ⇔ − ∇µ = 0. (1.6)
δφ ∂φ ∂(∇µ φ)

1
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 2

ˆ Energy–momentum tensor. We can also define the following object:

√
Z
1
δg Sm = − dd x −gTµν δg µν . (1.7)
2

We call the object Tµν the (Rosenfeld’s) energy momentum tensor. It is symmetric
by construction.
For diffeomorphism invariant Lagrangian densities Lm , such an energy momentum
is conserved in the following sense: If the equations of motion for matter are
satisfied, the we have
∇µ T µν = 0 . (1.8)
(In fact, as we shall see, the conservation of energy–momentum is in many cases
equivalent to the equations of motion for the matter.)
The argument for this goes as follows. First, let us remind that an infininitesimal
diffeomorphism, generated by a vector field ξ,

xµ → x µ − ξ µ , (1.9)

induces the following variation of the fields and the components of the metric:

δφ = Lξ φ , δg µν = Lξ g µν = 2∇(µ ξ ν) . (1.10)

Second, consider a general variation of the matter action:

√
Z Z Z Z
δSm µν δSm 1 d µν δSm
δSm = δg + δφ = − d x −gTµν δg + δφ , (1.11)
δg µν δφ 2 δφ
using the above definition of Tµν . If this variation is induced by a diffeomorphism,
and since the action is diffeo-invariant, we have
√
Z Z
1 d µν δSm
0 = δξ Sm = − d x −gTµν Lξ g + Lξ φ
2 δφ
√
Z Z
d (µ ν) δSm
= − d x −g Tµν ∇ ξ + Lξ φ
| {z } δφ
Tµν ∇µ ξ ν
√
Z Z
d µ ν δSm
= d x −g∇ Tµν ξ + Lξ φ , (1.12)
δφ
where in the last step we have integrated by parts and thrown away the boundary
term. Obviously, and since ξ ν (in the bulk) is arbitrary, we have

δSm
∇µ T µν = 0 ⇔ Lξ φ = 0 , (1.13)
δφ
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 3

up to total derivatives (boundary terms). In particular, if the equation of mo-

tion for the matter are satisfied, δSδφm = 0, the energy-momentum tensor is con-
served. Typically, the converse is also true, provided the ‘field is sufficiently
non-degenerate’, see below for explicit examples.
To calculate Tµν explicitly, we can use the following two tricks:

√ 1√
δ −g = − −ggµν δg µν , δgµν = −gµρ gνσ δg ρσ . (1.14)
2

ˆ Example 1: Scalar field. Using the minimal coupling principle, the Lagrangian
reads
1
L = − g µν (∇µ φ)(∇ν φ) − V (φ) . (1.15)
2
Using (1.6), the corresponding field equation is
dV
∇µ ∇µ φ − = 0. (1.16)
dφ
To calculate the energy momentum tensor, we simply vary the corresponding
action:
Z h √ √ i 1
Z
√ h i
d
δg S = d Lδ −g + −gδL = − dd x −g gµν L + ∇µ φ∇ν φ δg µν (. 1.17)
2 | {z }
Tµν

In the flat space limit, this is the canonical energy momentum tensor for the scalar
field derived from the Noether’s procedure due spacetime translation invariance.
Let us next look at what imposing ∇µ T µν = 0 yields. We have

∇µ T µν = g µν ∇µ L +∇2 φ∇ν φ + ∇µ φ∇µ ∇ν φ (1.18)

| {z }
−∇ν ∇µ φ∇µ φ− dV
dφ
∇ν φ
   dV 
= ∇µ φ ∇µ ∇ν φ − ∇ν ∇ν φ + ∇ν φ ∇2 φ − = 0. (1.19)
| {z } dϕ
0

So, provided ∇φ 6= 0 we recover the equations of motion (1.16).

ˆ Example 2: Electromagnetic field. The Lagrangian reads

1 µν 1 µα νβ
L[Aµ , g αβ ] = − F Fµν = − g g Fµν Fαβ , Fµν = 2∇[µ Aν] . (1.20)
16π 16π
It gives rise to the Maxwell equations in curved spacetime:

∇µ F µν = 0 , dF = 0 , (1.21)
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 4

where the latter is automatically satisfied from the definition of the field strength.
Varying the action w.r.t. the metric, we have
√ √
Z
1 h i
δg S = − dd x F 2 δ −g + 2 −gFµδ Fν δ δg µν . (1.22)
16π
This immediately yields the (automatically symmetric and gauge invariant) energy–
momentum tensor:
1 α 1 2

Tµν = Fµα Fν − gµν F . (1.23)
4π 4
It is this electromagnetic energy momentum tensor that couples to gravity (not
the canonical one, which is not symmetric and not gauge invariant). Moreover,
we find
1
4π∇µ T µ ν = ∇µ F µα Fνα + F µα ∇µ Fνα − ∇ν Fαβ F αβ = ∇µ F µα Fνα , (1.24)
| {z } 2
1 βα
2
F (∇β Fνα +∇α Fβν )

where the last three terms vanished upon using the second set of Maxwell’s equa-
tions.1 Obviously, if Fνα has an inverse, then conservation of the energy momen-
tum tensor implies the first set of Maxwell’s equations.

1.2 Einstein–Hilbert action

ˆ Action. Let us now think about how to construct the variational principle for
the gravitational field itself. To get the 2nd-order equations of motion for gµν ,
we want a scalar invariant which depends on the metric and its first deriva-
tives, I = I(g, ∂g). Unfortunately there is no such thing—why? So we have
to give up and take an invariant that depends also on the second derivatives of
the metric. The simplest one is the Ricci scalar. This leads to the following
Einstein–Hilbert action:
√
Z
1
SEH [g] = −gR(g, ∂g, ∂ 2 g) . (1.26)
16πG

ˆ Variation. Varying this action, and using the dirty trick that R = Rµν g µν , we get
√ √ √ √
Z Z
1 αβ 1
δSEH = δ( −gRαβ g ) = (Rδ −g+ −gRµν δg µν + −gg µν δRµν ) .
16πG 16πG
(1.27)
1
We could have derived this directly using the result (1.13). Namely, using the Cartan’s lemma
Lξ ω = ξ · (dω) + d(ξ · ω) , valid for any p-form ω, we have
δSm √ √ √
Lξ Aµ ∝ −g∇ν F νµ Lξ Aµ = −g∇ν F νµ (dA)µα ξ α + −g∇ν F νµ ∇µ (ξ · A) . (1.25)
δAµ
However, by integrating the second term by parts (we are under an integral really), we turn it into
∇µ ∇ν F µν (ξ · A) = 0. Thus we arrive at the same conclusion as above.
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 5

The first two terms are easy, they give

√ √ √  1  √
Rδ −g + −gRµν δg µν = −g Rµν − Rgµν δg µν = −gGµν δg µν . (1.28)
2
upon using the first identity (1.14). On the other hand, the last term, g µν δRµν ,
seems horrible. Fortunately it can be show that it only gives a boundary term:2

g µν δRµν = ∇µ V µ , Vµ = ∇β (δgµβ ) − g αβ ∇µ (δgαβ ) , (1.38)

2
If you want to know that bloody details, here they are, see also [1]. We have Palatini lemma:

δRµν = (δΓλ µλ );ν − (δΓλ µν );λ . (1.29)

To prove this lemma, we shall use ‘local inertial frame’ equipped with Riemann normal coordinates,
that is Γµ νλ = 0 at a point, but ∂Γ non-vanishing (gµν,λ = 0 but gµν,λδ is not) together with the fact
that if the tensor identity is proved in one frame, it is valid in any frame. So,
.
δRαβ = δRγ αγβ = δ[Γγ αβ,γ − Γγ αγ,β + ΓΓ − ΓΓ] = δ[Γγ αβ,γ − Γγ αγ,β ]
= ∇γ (δΓ)γ αβ − ∇β (δΓ)γ αγ = δΓγ αβ;γ − δΓγ αγ;β , (1.30)

where we have used the fact that in normal coordinates ∂ and ∇ are the same and that δΓ is a tensor.
We also have
1 µσ 1 µσ
δΓµ νλ = g (δgσν,λ + δgσλ,ν − δgνλ,σ ) + δg (gσν,λ + gσλ,ν − gνλ,σ )
2 2
1 µσ 1 µσ
= g (δgσν;λ + δgσλ;ν − δgνλ;σ ) + δg (gσν;λ + gσλ;ν − gνλ;σ )
2 2
1 µσ
= g (δgσν;λ + δgσλ;ν − δgνλ;σ ) , (1.31)
2
which also yields
1 ασ 1
δΓα αβ = g (∇β δgσα + ∇α δgσβ − ∇σ δgαβ ) = g ασ ∇β δgασ . (1.32)
2 2
Hence we found
1
∇µ ∇λ δgλν + ∇ν ∇λ δgλµ − ∇2 δgµν − ∇µ ∇ν (g λσ δgλσ ) .

δRµν = (1.33)
2
Everywhere above “δg” means variation of gµν . We also need variation w.r.t. g µν , which is given by

δg µν = −g σµ g λν δgλσ . (1.34)

So we have

g µν δRµν = −∇µ ∇ν δg µν + ∇2 δg = ∇µ V µ , V µ = −∇ν δg µν + ∇µ δg −1 , (1.35)

where
δg −1 = gµν δg µν , (1.36)
Note also a useful identity
1
δΓα αβ = − ∇β δg −1 . (1.37)
2
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 6

which does not contribute to the equations of motion. Thus we have the following
Einstein equations in the absence of matter:

Gµν = 0 . (1.39)

Note that, by contracting the equation with g µν , we find that in d-dimensions

we have (1 − 21 d)R = 0 and thence for d 6= 2 we must also have R = 0 and the
vacuum equations can be written as: Rµν = 0. Do you know what happens in
2d?
ˆ Let us make 4 remarks.

– Remark 1. Obviously, the Einstein-Hilbert action is diffeomorphism invari-

ant. We can thus repeat the argument leading to the conservation of the
energy-momentum tensor, to arrive at the conclusion that even off-shell (that
is for any metric) we have the following Bianchi identity:

∇µ Gµν = 0 . (1.40)

The same conclusion will be true for ‘generalized Einstein tensor’ of any
diffeomorphism invariant (higher-curvature) theory of gravity.
– Remark 2. The obtained equation is second order PDE for the metric, Gµν =
Gµν (g, ∂g, ∂ 2 g). How is this possible? We started from the action, given by
R, which already is of second order in derivatives. We should thus have
received 4th order equations of motion. However, this is not the case and
the reason is simple. It can be shown that the Lagrangian density can be
split to a piece that depends only on the first derivatives, plus a piece that
is a total derivative (none of which is a tensor):
√ √
−gR(g, ∂g, ∂ 2 g) = −g R̃(g, ∂g) + ∂µ R̂µ (g, ∂g) , (1.41)
√
where δ R̂µ = −gV µ .3 The latter term does not contribute to the equa-
tions of motion, provided we impose the corresponding boundary conditions.
These, however, do not give rise to the standard Dirichlet problem. To
achieve that, one needs to add the so called York–Gibbons–Hawking term,
as we shall see in the next lecture.
– Remark 3. In the above variation we used the so called second-order formalism:
the action was treated as a function of the metric gµν and contained its first
and second derivatives.
3
In fact, one has
√
R̃ = g ικ (Γλ µκ Γµ λι − Γλ ικ Γµ λµ ) , R̂µ = −g(g ικ Γµ ικ − g ιµ Γκ ικ ) . (1.42)

Moreover, there is a ‘holographic relation’ between the bulk and surface part of the Lagrangian, with
the surface terms obtainable from the bulk on by differentiation, [2].
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 7

Perhaps more useful is the first-order (Palatini) formalism where the action
is treated as action for two fields: the metric gµν and the connection Γα βγ :
√
Z
1
SPalatini [g, Γ] = −gg µν Rµν (Γ) , (1.43)
16πG
where we used the fact that Rµν can be entirely written in terms of the
connection and its derivatives, Rµν = Rµν (Γ). Thus the variation w.r.t.
the metric yields immediately the Einstein equations (1.39). It can then be
shown that the variation w.r.t. the connection establishes that the connec-
tion is given by Christoffel symbols (this was an input in the second order
formalism.)
Note that in this spirit, one can also write down the purely connection
dependent action for gravity with cosmological constant Λ, known as the
Eddington’s action:
Z
1
q
4
SEddington [Γ] = d x − det Rµν (Γ) . (1.44)
8πGΛ
If you want to, please show the equivalence with the Einstein-Hilbert action
in the presence of Λ.
– Remark 4: Einstein equations. Recovering now the full Einstein equations is
easy. We simply add the corresponding matter Lagrangian density:
√
Z
S = SEH [g] + d4 x −gLm . (1.45)

The variation w.r.t. the metric and throwing away the boundary terms then
yields
√  1
Z
4 µν 1 µν

δS = d x −g Gµν δg − Tµν δg (1.46)
16πG 2
Thus we recover the famous 1915 Einstein’s field equations:4

Gµν = 8πGT µν . (1.47)

1.3 Higher-curvature gravities

ˆ Other curvature invariants. One might think that the Ricci scalar we chose is
simply one possibility for the action, but that we have other choices, such as:

R2 , Rµν Rµν , Rµνκλ Rµνκλ , R3 , ∇µ R∇µ R , . . . . (1.48)

4
Poor Einstein was scooped by Hilbert by a couple of months, after giving a seminar to him about
what he is trying to do. Well, perhaps an appropriate punishment for leaving Prague before finishing
his theory of gravitation :).
CHAPTER 1. LECTURE 1: VARIATIONAL PRINCIPLES IN GR 8

However, one can show that this is not the case, as we have the following [3]:
Lovelock theorem. In four dimensions, the Einstein–Hilbert action is the only local
action (apart from the cosmological constant and topological terms) that leads to
the second order differential equations for the metric.
In other words, in four dimensions Einstein’s theory is the unique theory we
can obtain from the action principle that yields 2nd-order EOM for the metric.
In higher dimensions this is no longer true – we have a possibility of the so
called Lovelock gravities. The corresponding Lagrangian is given in terms of the
Euler densities, a certain combination of curvature invariants constructed from
the powers of the Riemann tensor. They lead to the second order equations of
motion for the metric, naturally generalizing the Einstein equations to higher
dimensions.
ˆ Gauss–Bonnet gravity. To give and example, let us consider the so called 2nd-
order Lovelock gravity, also know as the Gauss–Bonnet gravity. The correspond-
ing Lagrangian reads:
√
Z
1
S=− dd x −g (R + αG) , G = R2 − 4Rαβ Rαβ + Rαβγδ Rαβγδ , (1.49)
16πGN
where α is a coupling constant with the dimension [α] = L2 . This extension of
General Relativity naturally appears in the low energy effective action of heterotic
string theory. It yields the following equations of motion:

Gαβ + αHαβ = 0 , (1.50)

where the Gauss–Bonnet modification amounts to

1
Hαβ = − gαβ G + 2RRαβ − 4Rαγ Rβ γ + 4Rγαβδ Rγδ + 2Rα γδκ Rβγδκ . (1.51)
2
As argued above, it has to satisfy ∇µ H µν = 0. Note that, despite the appearance
of higher derivative curvature terms in the action, the equations of motion for
the metric remain of the second order. Surprisingly, the modification (1.51) is
non-trivial only in d ≥ 5 dimensions; G is topological (a total derivative) in d = 4,
and vanishes for d < 4.
ˆ Quite recently, people have been thinking about taking a d → 4 limit of the above
theory. This leads to the following scalar-tensor theory with a peculiar kinetic
term [4, 5, 6]:
√ h
Z
1  i
S=− d4 x −g R + α φG + 4Gµν ∂µ φ∂ν φ − 4(∂φ)2 φ + 2(∇φ)4 ,
16πGN
(1.52)
which is a special case of Horndeski theory [7] (the most general scalar-tensor
theory whose equations of motion remain second-order).
Chapter 2: Lecture 2: Submanifolds
& York–Gibbons–Hawking term

2.1 Gauss–Codazzi formalism

ˆ Motivation. i) To describe extended low-dimensional objects such as domain walls,
cosmic strings, and branes. ii) To make sense of the variational principle for
gravity.
A boundary is a hypersurface in spacetime, that is, a surface with 1 dim less than
M . Lower-dimensional surfaces in spacetime can be submanifolds.

ˆ A submanifold Σ ∈ M is a subset of M that is also a manifold in its own right,

see picture:

Here and in what follows we use the following notation: we assume

dim(M ) = D , dim(Σ) = d = D − n , (2.1)

and denote indices on Σ by A, B, . . . , and indices on M by µ, ν, . . . ; coordinates

on Σ are σ A and coordinates on M are xµ .
The co-dimension of Σ is n = D − d, that is, there are n linearly independent
normal directions to Σ, that is vectors ni in Tp M (p ∈ Σ, M ) such that

ni (σ A ) = 0 i = 1, . . . , n , (2.2)

9
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 10
for all coordinate functions σ A on Σ:

We will take
ni · nj = ±δij , (2.3)
where − is for timelike n and + for spacelike n.
We now have two ways to describe things. i) spacetime point of view (in terms
of spacetime objects) and ii) submanifold point of view (in terms of objects on a
submanifold). Let us start with the first:
ˆ Spacetime point of view. In this description we define the first fundamental form,
or induced metric of Σ as
n
X
hµν = gµν + j njµ njν , (2.4)
j=1

where j = + for timelike and − for spacelike nj . hµν is the metric Σ inherits
from M (but lies in T ∗ M ⊗ T ∗ M ) At the same time

hµ ν (2.5)

is a projector onto Σ: obeys hµ ν hν κ = hµ κ and hµ ν nνi = 0. It follows that hµν is

degenerate from the spacetime point of view.
We also define the second fundamental form or extrinsic curvature of Σ as

Kiµν = h(µ σ hν) λ ∇σ niλ . (2.6)

This measures how Σ curves in M .

CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 11
In what follows we focus on a co-dimension 1 hypersurface. We also define the
corresponding extrinsic curvature scalar as

K = g µν Kµν = ∇µ nµ . (2.7)

Indeed, we have

K = g αβ Kαβ = g αβ hγ α hδ β ∇γ nδ = hγβ (δβδ + nδ nβ )∇γ nδ

= hγβ ∇γ nβ = (g γβ + nγ nβ )∇γ nβ = ∇β nβ + nβ ∇n nβ = ∇β nβ . (2.8)

ˆ Manifold point of view. Since Σ is a manifold in its own right, we can also consider
quantities intrinsic to Σ. To this purpose, can think of Σ as a map Σ → M , given
by:
Σ : xµ (σ A ) . (2.9)

The corresponding pull-back allows one to define the following projection to Σ:

∂xµ
eµ A = : Tp∗ (M ) → Tp∗ (Σ) . (2.10)
∂σ A

For example,
ωµ → eµ A ωµ = ωA . (2.11)
In particular, we define the intrinsic (pull-back) metric (element of T ∗ Σ ⊗ T ∗ Σ):

hAB = eµ A eν B gµν . (2.12)

One can verify the following ‘completeness relation’:

g αβ = hAB eαA eβB − nα nβ . (2.13)

The simplest example is the intrinsic metric for a particle

 ds 2
hτ τ = ẋµ ẋν gµν = , (2.14)
dτ
where xµ (τ ) is the particle’s worldline.
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 12
Note also that we can think of eA = eµ A ∂µ as tangent vectors to Σ:

∂xµ
eA · ni = niµ = 0, i = 1, . . . , n . (2.15)
∂σ A

Can also define the corresponding extrinsic curvature (from the intrinsic point of
view)
KAB = eµ A eν B ∇µ nν = −nµ DA eµ B , (2.16)
where D is the intrinsic covariant derivative (again Levi-Civita) inherited from
the covariant derivative ∇:

Dµ Vν = hµ λ hν σ ∇λ Vσ , (2.17)

for any V tangent to Σ (V · ni = 0). We also have

K = hAB KAB (2.18)

That this is the same as before can be seen as follows:

K = hAB eαA eβB ∇α nβ = (g αβ + nα nβ )∇α nβ = ∇α nα + nβ ∇n nβ = ∇α nα = K .

(2.19)

ˆ For example: consider a cylinder {x2 + y 2 = a2 } ⊂ R3 , using the Cartesian,

xα = (x, y, z), and cylindrical, xα = (r, θ, z), coordinates:

ds2 = gµν dxµ dxν = dx2 + dy 2 + dz 2 = dr2 + r2 dθ2 + dz 2 . (2.20)

We have nµ ∝ ∂µ Σ, and upon normalizing, we find

   
cos θ 1
n=  sin θ  =  0 , (2.21)
0 0
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 13
with the first expression valid in Cartesian and the latter in cylindrical coordi-
nates. We then have
sin2 θ
   
− sin θ cos θ 0 0 0 0
hµν = gµν −nµ nν =  − sin θ cos θ cos2 θ 0  =  0 a2 0  , (2.22)
0 0 1 0 0 1

The extrinsic curvature in polar coordinates is then

   
0 0 0 0 0 0
Kµν = −Γrµν = rδµθ δνθ =  0 a 0  , K µ ν =  0 1/a 0  , (2.23)
0 0 0 0 0 0

and K = a1 . Intrinsic picture involves coordinates σ A = (θ, z). This yields

 2   
a 0 a 0
γAB = , KAB = . (2.24)
0 1 0 0

ˆ Curvature of Σ is related to curvature of M and extrinsic curvature through the

Gauss equations:

(d) 0 0 0 0
Rα βγδ = (D) Rα β 0 γ 0 δ0 hα α0 hβ β hγ γ hδ δ ∓ (K α γ Kβδ − K α δ Kβγ ) , (2.25)

where the minus sign applies to spacelike normal.

Cylinder example: because (3) Rα βγδ = 0 (we are in R3 ) and Kαβ has only one
component we have K α γ Kβδ − K α δ Kβγ = 0 and therefore intrinsic Riemann,
(2) α
R βγδ = 0.

2.2 York–Gibbons–Hawking boundary term

ˆ In the previous section we have shown that
√ √
Z Z
1 4 µν 1
δSEH = − d x −g Gµν δg − d4 x −g∇µ V µ , (2.26)
16πG M 16πG M
| {z
√
}
R
∂M d3 x hV µ nµ

where
V = V µ nµ = ∇n δg −1 − nα ∇β δg αβ , δg −1 = gαβ δg αβ . (2.27)
The second term does not yield the standard Dirichlet boundary conditions. In
what follows we would like to eliminate it, by adding a proper boundary term to
the Einstein–Hilbert action, to get the Dirichlet variational principle for gravity.
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 14
ˆ To this purpose we use the Gauss–Codazzi formalism introduced above, thinking
of ∂M as Σ. That is, we consider co-dimension one Σ – a “wall”. To simplify
the calculations, we take ∂M to be a timelike boundary, that is nµ spacelike, and
(w.l.o.g.) extend nµ geodesically into the bulk. Thus we have

n2 = 1 , hµν = gµν − nµ nν , ∇n nµ = 0 . (2.28)

Lemma: We can show that

1 1
δΓα αβ = − ∇β δg −1 , δnµ = nν δg µν . (2.29)
2 2
The first one was shown in a Footnote 2 of the previous lecture. The second one
is proved here.1

ˆ Consider next

δK = δ(∇µ nµ ) = δ(∂α nα + Γα αβ nβ ) = ∇α δnα + δΓα αβ nβ

| {z }
− 12 ∇β δg −1

1 1 1
= ∇α δnα − ∇n δg −1 = ∇α ( nβ δg αβ ) − ∇n δg −1
2 2 2
1 1
nβ ∇α δg αβ − ∇n δg −1 + ∇α nβ δg αβ .

= (2.30)
2 | {z } 2
−V

However, we have (using geodesicity of n)

Kαβ = hγ α hδ β ∇γ nδ = ∇α nβ , δg αβ ∇α nβ = δhαβ Kαβ , (2.31)

and so
V = −2δK + Kαβ δhαβ . (2.32)

ˆ The boundary term in (2.26) is thus

1
Z √ 
d3 x h 2δK − Kµν δhµν

δS∂M =
16πG
1
Z
3
√ 1
Z √
= d xδ( hK) − d3 x h(Kµν − Khµν )δhµν . (2.33)
8πG 16πG
The first term is a (minus) variation of the famous York–Gibbons–Hawking bound-
ary term:
1
Z √
SGH = − d3 x hK , (2.34)
8πG
whose variation cancels the unwanted boundary term in (2.26).
1
Under construction – or up to you :)
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 15
ˆ To summarize, we should consider the following gravitational action:

1
Z
4 √ 1
Z √
Sg = SEH + SYGH =− d x gR − d3 x hK . (2.35)
16πG 8πG

The first terms gives the E-L equations plus the boundary derivative term. The
latter is cancelled by the second term. However, we still get the “Kµν − Khµν ”
term:
1
Z
4 √ µν 1
Z √
δSg = − d x gGµν δg − d3 x h(Kµν − Khµν )δhµν . (2.36)
16πG M 16πG
For fixed boundary and Dirichlet boundary conditions we have δhµν = 0, which
yields a well defined variational principle.

ˆ Dynamical boundary. However, we could also have a “dynamical boundary”. If

hµν is no longer fixed on ∂M , we have

δSGH Kµν − Khµν

µν
= . (2.37)
δh −16πG
In particular, consider a thin wall/brane-world. We can think about it as a
combination of 2 manifolds (bulks) with boundary on each side:

Σ = ∂M+ = ∂M− sourced by wall energy-momentum tensor

Tµν ∼ Sµν δ xµ − xµ (σ A ) .


(2.38)

Boundary description gives

hµν |∂M+ = hµν |∂M− , hKµν − Khµν i = 8πGSµν , (2.39)

where h · i stands for averaging of “ +00 +“−00 quantities, e.g. K(n+ ) + K(n− ).
These are the Israel junction conditions.
CHAPTER 2. LECTURE 2: SUBMANIFOLDS & YORK–GIBBONS–HAWKING
TERM 16
ˆ To derive these, Israel [8] took the limit δ → 0 of the Einstein equations, Gµν =
8πGTµν , in the following setup (note that in this case we have a continuous nor-
mal across Σ):

He then used these to study the gravitational collapse of thin matter shells.

ˆ Alternatively, one could consider a certain kind of the Neumann boundary con-
ditions, see e.g. [9].
Chapter 3: Lecture 3: Black hole
thermodynamics

3.1 Motivational foreplay

ˆ Charged AdS black hole. Let us consider the following solution:

dr2 2M Q2 r2 Q
ds2 = −f dt2 + + r2 dΩ2 , f =1− + 2 + 2, A = − dt , (3.1)
f r r l r

where dΩ2 = dθ2 +sin2 θdϕ2 . This is a solution of Einstein Maxwell-AdS equations
(EM )
Gµν = −Λgµν + 8πTµν , ∇µ F µν = 0 , F = dA , (3.2)

where the cosmological constant is given by Λ = −3/l2 . Identifying its contribu-

tion in Einstein equations with the energy-momentum tensor of a perfect fluid:
 
−Λgµν = 8π (ρ + P )uµ uν + P gµν , (3.3)

we may identify the corresponding ‘cosmological pressure’ as

Λ 3
P =− = = −ρ . (3.4)
8π 8πl2

ˆ Basic properties. The above solution is characterized by its mass M and charge
Q. For given M, Q, l, its horizon is located at the horizon radius r+ , given by the
largest root of
f (r+ , M, Q, l) = 0 . (3.5)
It is a Killing horizon: a null surface generated by Killing field k = ∂t . Associated
with it is the concept of surface gravity κ:

17
CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 18

It can be shown that for a spherically symmetric metric characterized by a single

metric function f , as above, we have
1 ∂f
κ= . (3.6)
2 ∂r r=r+

We may also calculate the horizon area. Taking dt = 0 = dr, the induced spatial
metric ‘on the horizon’ is dγ 2 = r+
2
dΩ2 . The area then reads
Z p Z
2 2
A= det γdθdϕ = r+ sin θdθdϕ = 4πr+ . (3.7)

ˆ First law of black hole mechanics. Let’s now consider a physical process under
which the black hole spacetime is perturbed and eventually settles to a new
charged AdS black hole spacetimes with modified spacetime parameters, {M +
δM, Q + δQ, l + δl}. Consequently, the horizon radius also modifies to r+ + δr+ ,
determined from f (r+ + δr+ , M + δM, Q + δQ, l + δl) = 0. Using the Taylor
expansion to linear order in perturbation, together with (3.5), we thus have

∂f ∂f ∂f ∂f
0= δM + δr+ + δQ + δl . (3.8)
∂M ∂r ∂Q ∂l r=r+

Re-arranging this equation, we thus have

 ∂f −1  ∂f ∂f ∂f 
δM = − δr+ + δQ +
∂M ∂r ∂Q ∂l r=r+
2
r+  2Q 2r 
= 2κδr+ + 2 δQ − 3+ δl , (3.9)
2 r+ l

where in the second line we used (3.6) and the specific form of f for the charged
AdS black hole. We thus find
κ δA Q 4 3
δM = + δQ + πr+ δP , (3.10)
2π 4 r+ 3

treating M = M (A, Q, P ). Let’s now make the following definitions of electrostatic potential
and black hole volume:
 ∂M  Q  ∂M  4 3
φ= = , V = = πr+ . (3.11)
∂Q A,P r+ ∂P A,Q 3

The first is indeed the electrostatic potential on the horizon, φ = −k · A|r=r+ .

Similarly, there exists a geometric definition for V [10]. This yields the first law
of black hole mechanics:
κ δA
δM = + φδQ + V δP . (3.12)
2π 4
CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 19

However, this looks a lot like a 1st law of (black hole) thermodynamics (especially
because of the work terms), provided we identify M with enthalpy and

~κ A
T = , S= . (3.13)
2πkB 4~GN

While this seems strange for classical black holes, as shown by Hawking in 1974,
[11, 12], when quantum effects are taken into account, black holes radiate away
as black body with these characteristics. Derivation used QFT in curved space.
Hawking basically showed “stimulated emission”. The problem with his deriva-
tion is that due to the bluehift near the horizon, the test field approximation
breaks down and we cannot really trust the result. However, since then the same
result has been reproduced by many other approaches, e.g: Euclidean path in-
tegral, tunneling, string theory, LQG. Let’s exploit our knowledge of variational
principles to ‘derive’ these results.

3.2 Euclidean Trick

ˆ Let us sue the following fact. Thermal Green functions have periodicity in imag-
inary Euclidean time τ = it :

G(τ ) = G(τ + β) , β = 1/T . (3.14)

Conversely, periodicity of G defines a thermal state. Green functions of quantum

fields in the vicinity of black holes have this property (as seen by a static observer).
What about gravitational field itself?
ˆ The Euclideanized spherical black hole solution (τ = it) is

dr2
ds2 = f dτ 2 + + r2 dΩ2 . (3.15)
f
Near the horizon we may expand

f = f (r+ ) + (r − r+ ) f 0 (r+ ) + · · · = 2κ∆r . (3.16)

| {z } | {z } | {z }
0 ∆r 2κ

Therefore, the near horizon limit of the Euclidean solution takes the following
form:
dr2
ds2 = 2κ∆rdτ 2 + 2
+ r+ dΩ2 . (3.17)
2κ∆r
We can now introduce a new coordinate ρ by

dr2 dr κ 2
dρ2 = ⇔ dρ = √ ⇔ ∆r = ρ , (3.18)
2κ∆r 2κ∆r 2
CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 20

getting
ds2 = κ2 ρ2 dτ 2 + dρ2 + r+
2
dΩ2 = ρ2 dϕ2 + dρ2 + r+
2
dΩ2 , (3.19)
upon introducing a new angle coordinate, ϕ = κτ . This looks like a flat space
written in polar coordinates, provided the angle ϕ has a period 2π, otherwise
there is a conical singularity at ρ = 0, which corresponds to the original black
hole horizon. The reasoning now goes as follows: since the original black hole was
originally non-singular (there is no matter there), we expect it to be non-singular
again. This is achieved by setting (we want to avoid conical singularity)
κ
ϕ ∼ ϕ + 2π ⇔ τ ∼ τ + 2π/κ ⇔ T = , (3.20)
| {z } 2π
β

which is the famous Hawking temperature. In particular, for Schwarzschild black

hole we recover
1 ~c3 M
T = → ∼ 6 × 10−8 K. (3.21)
8πGM kB 8πGN M kB M

Temperature for Kerr*

ˆ The Euclidean trick described above can also be used to determine the tempera-
ture of the Kerr black hole. In order to get real Euclidean geometry, we have to
rotate both the time and the rotation parameter:
t → iτ , a → ib , (3.22)
which yields
∆ Σ 2 sin2 θ 2
2
ds2E 2 2 2
= (dτ − b sin θdϕ) + dr + Σdθ + (r − b2 )dϕ + bdτ , (3.23)
Σ ∆ Σ
where ∆ = r2 − b2 − 2M r and Σ = r2 − b2 cos2 θ. When zooming on the horizon
r → r+ given by the largest root of ∆(r+ ) = 0, we have to eliminate the last term
in the metric, setting
bdτ
dϕ = − 2 . (3.24)
r+ − b2
When this is plugged back, we have dτ − b sin2 θdϕ ≈ Σ+ /(r+ 2
− b2 )dτ . Thus we
can write
2 Σ+  2 dr2  ∆
dsE ≈ 2 f dτ + + . . . , f = 2
. (3.25)
r + − b2 f r+ − b2
So, up to an overall constant conformal factor (which does not matter) we are
back to the spherical case and derive the following temperature:
∆0 (r+ ) r+ − M
T = 2 2
= 2
, (3.26)
4π(r+ − b ) 2π(r+ + a2 )
where in the last step we have Wick-rotated back the rotation parameter b → −ia.
CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 21

3.3 Euclidean action calculation

ˆ Consider the partition function of a system at temperature 1/β:

Z = Tre−βH = e−βF ∼ e−βH[gc ] , (3.27)

using the WKB approximation, in which the sum is dominated by classical (sta-
tionary) solutions gc . For those we can then write:
Z Z
3 1 IE
H ∼ d xH = d4 xLE = . (3.28)
β β
Thus, we can identify the Euclidean action with the free energy

IE 1
F = = − log Z . (3.29)
β β
Our goal is thus to calculate the Euclidean action for the given black hole solution.
We then find the corresponding entropy by the standard thermodynamic relation:
∂F
S=− = β 2 ∂β F . (3.30)
∂T
ˆ Calculation for Schwarzschild. Recall that we need to calculate
1
Z
4 √ 1
Z √
IE = − d x gR − d3 x hK (3.31)
16πG M 8πG ∂M
for the Euclidean Schwarzschild solution
dr2 2M
ds2 = f dτ 2 + + r2 dΩ2 , f =1− . (3.32)
f r
The Euclidean Schwarzschild geometry corresponds to a “cigar”:

Even after the Wick rotation, this is a vacuum solution of Einstein equations,
Rµν = 0 and thence
SEH = 0 . (3.33)
CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 22

ˆ However, we have a boundary at large r = R and so the Gibbons–Hawking term

may contribute. The induced metric is

dγ 2 = f (R)dτ 2 + R2 dΩ2 . (3.34)

The corresponding normal in M is n ∝ ∂r , and after normalization reads

p
n= f ∂r . (3.35)
r=R

ˆ Extrinsic curvature is
1 √ 1 2p 1 f 0 (R)
K = ∇µ nµ = √ ∂µ ( gnµ ) = 2 (r2 nr )0 = f (R) + p . (3.36)
g r r=R R 2 f (R)

The volume factor is

√ 3 p
hd x = f (R)R2 sin θdτ dθdφ . (3.37)

Putting together, we have

Z √ 3 2 2
h 1 0 i
K hd x = 4πβ R f (R) + f (R) = 4πβ(2R − 3M ) . (3.38)
R
|{z} R 2
dτ dθdφ

This diverges as R → ∞! However, note that this is also divergent for M = 0,

that is flat spacetime has divergent action!

ˆ To deal with this divergence we consider the appropriate background that has the
same time periodicity β and matches our spacetime exactly at the boundary:

This ‘cylinder’ corresponds to ‘thermal flat space’ filled with radiation of T = 1/β.
Its metric and induced metric are

ds20 = f0 dτ 2 + dr2 + r2 dΩ2 , dγ02 = f0 dτ 2 + R2 dΩ2 . (3.39)

CHAPTER 3. LECTURE 3: BLACK HOLE THERMODYNAMICS 23

To match ∂M0 and ∂M , we take f0 = f (R) . Then we have

2 p p
n = ∂r , K0 = , h0 d3 x → f (R)R2 sin θdθdφdτ , (3.40)
R
and so
Z p p  M 
d3 x h0 K0 = 8πβR f (R) = 8πβR 1 − + O(1/R2 ) . (3.41)
R

ˆ We now use the ‘renormalization’ subtraction procedure and write the total
gravitational action as:

1
Z
4 √ 1
Z √
IE = − d x gR − d3 x h(K − K0 ) , (3.42)
16πG M 8πG ∂M

where the first boundary term, with K, is called the York term, and its purpose
is to yield a well posed Dirichlet variational principle, and the second term, with
K0 , is called the Gibbons-Hawking term, and its purpose is to ‘tune’ the value of
the action.

Thus we have
βM
ISch = . (3.43)
2
The corresponding free energy is thus
ISch M β
F = = = , (3.44)
β 2 16π
and the corresponding entropy reads:
β2 A
S = β 2 ∂β F = = 4πM 2 = πr+
2
= , (3.45)
16π 4
deriving the Bekenstein formula. At the same time, we can check that

F = M − TS , (3.46)

as it must for the (Gibbs) free energy.

Bibliography

[1] R. M. Wald, General relativity. University of Chicago press, 2010.

[2] A. Mukhopadhyay and T. Padmanabhan, Holography of gravitational action

functionals, Phys. Rev. D 74 (2006) 124023, [hep-th/0608120].

[3] D. Lovelock, The einstein tensor and its generalizations, Journal of Mathematical
Physics 12 (1971) 498–501.

[4] H. Lu and Y. Pang, Horndeski gravity as D → 4 limit of Gauss-Bonnet, Phys.

Lett. B 809 (2020) 135717, [2003.11552].

[5] P. G. S. Fernandes, P. Carrilho, T. Clifton and D. J. Mulryne, Derivation of

Regularized Field Equations for the Einstein-Gauss-Bonnet Theory in Four
Dimensions, Phys. Rev. D 102 (2020) 024025, [2004.08362].

[6] R. A. Hennigar, D. Kubizňák, R. B. Mann and C. Pollack, On taking the D 4

limit of Gauss-Bonnet gravity: theory and solutions, JHEP 07 (2020) 027,
[2004.09472].

[7] G. W. Horndeski, Second-order scalar-tensor field equations in a

four-dimensional space, Int. J. Theor. Phys. 10 (1974) 363–384.

[8] W. Israel, Singular hypersurfaces and thin shells in general relativity, Il Nuovo
Cimento B (1965-1970) 44 (1966) 1–14.

[9] C. Krishnan and A. Raju, A Neumann Boundary Term for Gravity, Mod. Phys.
Lett. A 32 (2017) 1750077, [1605.01603].

[10] D. Kastor, S. Ray and J. Traschen, Enthalpy and the Mechanics of AdS Black
Holes, Class. Quant. Grav. 26 (2009) 195011, [0904.2765].

[11] S. W. Hawking, Black hole explosions?, Nature 248 (1974) 30–31.

[12] S. W. Hawking, Particle creation by black holes, Communications in

mathematical physics 43 (1975) 199–220.

24