CHAPTER 9

GENERAL RELATIVITY: THE FIELD THEORY APPROACH

We move now to the modern approach to General Relativity: field theory. The chief advantage of
this formulation is that it is simple and easy; the only thing to specify is the so-called Lagrangian
density. We start by presenting a simple introduction to classical field theory in flat spacetime which
we later generalize to curved spacetime. The last part of the Chapter is devoted to the action for the
gravitational field and the recovery of Einstein equations from it.

9.1 Classical mechanics

The fundamental problem of classical mechanics is to determine the way particles move given a
potential. Dynamical systems can be described by equations of motion or by action functionals
Z
S = dt L (qj (t), q̇j (t)) (9.1)

depending on generalized coordinates and velocities {qj (t), q̇j (t)}. The classical trajectory is defined
as the unique path that extremizes the action functional (δS = 0) for all variations qj → qj + δqj with
fixed initial qj (t0 ) and final values qj (tf ). An explicit variation of the action gives
Z tf  
∂L ∂L
δS = dt δqj + δ q̇j . (9.2)
t0 ∂qj ∂ q̇j

Integrating the last term by parts to flip the temporal derivative onto ∂L/∂ q̇j we get
Z tf   
∂L d ∂L
δS = dt − δqj , (9.3)
t0 ∂qj dt ∂ q̇i

where we have omitted a total derivative that vanishes because of the boundary conditions δqj (t0 ) =
δqj (tf ) = 0. Since δqj is arbitrary, the extremization of the action translates into the so-called Euler-
Lagrange equations  
d ∂L ∂L
− = 0. (9.4)
dt ∂ q̇i ∂qj
This variational formulation has several advantages:
9.2 From Classical Mechanics to Field theory 130

i) The properties of the system are compactly summarized in one function, the Lagrangian.
ii) There is a direct connection between invariances of the Lagrangian and constants of motion1 .
iii) There is a close relation between the Lagrangian formulation of classical mechanics and quantum
mechanics.

9.2 From Classical Mechanics to Field theory

The Lagrangian formalism presented above can be extended to continuous systems involving an infinite
number of degrees of freedom. This is achieved by taking the appropriate limit of a system with a
finite number of degrees of freedom. Consider a one dimensional chain of length l made of N equal
masses m. The masses are separated by a distance a and connected by identical massless springs
with force constant k. The total length of the system is l = (N + 1)a. The displacement of the
particles with respect to its equilibrium position x̄j = ja is described by a generalized coordinate
φ(xj , t) ≡ xj (t) − x̄j with j = 1, . . . , N and φ0 = φN +1 = 0. The Lagrangian of the full system
includes the kinetic energy of the particles and the energy stored into the springs, i.e.
N N
1 X 2 1 X 2
L=T −U = m φ̇j (t) − k (φj+1 (t) − φj (t)) . (9.5)
2 j=1 2 j=0

The continuous limit of the previous expression can be taken by sending N → ∞ and a → 0 in such
a way that the total length of the chain, l = (N + 1)a, remains fixed. To keep the total mass of the
system and the force between particles finite we require m/a and ka to go to some finite values µ and
Y playing the role of the mass density and the Young modulus in the continuous theory. We have
N N 2
1 l h 2

φj+1 (t) − φj (t)
Z
1 X m 2 1X 2
i
L= a φ̇j (t) − a (ka) −→ L = dx µφ̇ − Y (∂x φ) ,
2 j=1 a 2 j=0 a 2 0

with the finite number of generalized coordinates φj replaced by a continuous function φ(x, t). The
antisymmetric dependence of Eq.(9.6)
p on the derivatives suggests the introduction of a set of coordi-
nates xµ = (cs t, x)T with cs = Y /µ and a Lorentzian metric ηµν = diag(−1, 1). This allows us to
write Z
S = d(cs t)dx L (9.6)

with
µcs µν
L=−
η ∂µ φ∂ν φ , (9.7)
2
the so-called Lagrangian density. The jump from fields existing within a physical medium to fields in
vacuum is now straightforward: we must simply replace cs by the speed of light c. Generalizing the
metric ηµν to arbitrary dimensions , we can write generically write the action for relativistic fields as
Z
S = dn x L (φ, ∂µ φ) , (9.8)

where we have allowed for a dependence of the Lagrangian density on the fields.

Exercise
Consider again the chain of masses connected by springs. Modify the system to give rise to an
explicit dependence of the Lagrangian on φ. Hint: Eq. (9.7) is shift-invariant.
1 For instance, if the Lagrangian is invariant under rotations, angular momentum is conserved.
9.3 Principles of Lagrangian construction 131

Gauge freedom in the Lagrangian

The Lagrangian of a physical system is not unique. To see this, consider a transformation of
the form
L −→ L0 = L + ∂σ g σ , (9.9)
and its effect on the action (9.8)
Z Z Z
S 0 = dn xL0 = S + dn x ∂σ g σ = S + g σ dSσ . (9.10)
R ∂R

The term associated to ∂σ g σ turns out to be a boundary term, which does not contribute to
the equations of motion. Lagrangians differing by a contribution ∂σ g σ give rise to the same
equations of motion.

The equations of motion for the field φ(x, t) can be obtained by considering the change of the action
under an infinitesimal change φ(x, t) −→ φ(x, t) + δφ(x, t). The only requirement to be satisfied by
the variations δφ is to be differentiable and to vanish outside some bounded region of spacetime (to
allow an integration by parts). Performing this variation we get
Z   Z   
n ∂L ∂L n ∂L ∂L
δS = d x δφ + δφ,µ δφ = d x − ∂µ δφ . (9.11)
∂φ ∂φ,µ ∂φ ∂(∂µ φ)

Requiring the action to be stationary (δS = 0) and taking into account that δφ is completely arbitrary,
we obtain the continuous version of the Euler-Lagrange equations
 
∂L ∂L
∂µ − = 0. (9.12)
∂(∂µ φ) ∂φ

A worked-out example
As a direct application of Eq. (9.12), let me consider the action (9.7)

1 ∂2φ ∂2φ
 
∂L
∂µ =0 −→ ∂µ (η µν ∂ν φ) = 0 −→ − 2 2 + = 0. (9.13)
∂(∂µ φ) cs ∂t ∂x2

As expected, we get a wave equation.

9.3 Principles of Lagrangian construction

What kind of Lagrangian density should we choose? To be in the safe side, the Lagrangian density of
a relativistic theory is required to satisfy the following requirements:

1. L must be a real-valued function, since it enters into expressions of physical significance, like
the Hamiltonian.
2. L must have dimension 4 in units of energy, since in natural units the action is dimensionless
and [d4 x] = −4.
3. L must be a linear combination of Lorentz invariant quantities constructed from the fields, their
first partial derivatives and the universally available objects ηµν and µνρσ .
9.3 Principles of Lagrangian construction 132

4. The coefficients of this linear combination can be restricted by the symmetries of the problem
(internal symmetries/ gauge symmetries).
5. L should be bounded from below.

The power of the previous program is made most vividly evident by considering some examples.

9.3.1 A complex scalar field with U (1) symmetry

Consider a complex scalar field

Φ = φ1 + iφ2 , Φ∗ = φ1 − iφ2 . (9.14)

The quadratic Lorentz invariants which can be constructed from Φ, Φ∗ , ∂µ Φ and ∂µ Φ∗ lead to a
Lagrangian density of the form
1 µν 1
L= η [a∂µ Φ∂ν Φ + a∗ ∂µ Φ∗ ∂ν Φ∗ + 2a0 ∂µ Φ∗ ∂ν Φ] + [bΦΦ + b∗ ΦΦ∗ + 2b0 Φ∗ Φ] . (9.15)
2 2
where the reality condition L = L∗ imposes the appearance of the pairs a, a∗ and b, b∗ and requires the
coefficients a0 and b0 to be real. The previous Lagrangian density can be written in a more compact
way by introducing the arrays
T
Φ∗
  
Φ
Φ̃ ≡ and Φ̃† ≡ = (Φ∗ Φ) . (9.16)
Φ∗ Φ

to get2
a∗ b∗
   
1 a0 1 b0
L = η µν Φ̃†,µ Φ̃,ν + Φ̃∗ Φ̃ . (9.17)
2 a a0 2 b b0
The number of terms appearing in this Lagrangian can be reduced in cases in which we have symmetries
on top of Lorentz invariance. As an illustration of this, imagine the field Φ to possess an internal
symmetry
Φ → eiω Φ , Φ∗ → e−iω Φ∗ . (9.18)
In this case, we necessarily have a = b = 0 and the matrices in (9.17) become diagonal. This leaves
us with a simpler Lagrangian, that with some notational adjustments, can be written as
1  µν
K −η ∂µ Φ∗ ∂ν Φ − κ2 Φ∗ Φ .

L= (9.19)
2
A direct application of the Euler-Lagrange equations (9.12) provides two uncoupled equations for Φ
and Φ∗ , namely
2 − κ2 Φ = 0 , 2 − κ2 Φ∗ = 0 ,



(9.20)
µ
that we can use to provide a physical interpretation for the parameter κ2 . Indeed, setting Φ = eikµ x
with pµ = (E, p) in any of these two equations, we get a dispersion relation E 2 − p2 = κ2 , which
makes it natural to identify the parameter κ2 with the mass m2 of the field.
2 In this notation, the reality condition L = L∗ results from the hermiticity of the 2 × 2 matrices.
9.3 Principles of Lagrangian construction 133

9.3.2 A worked-out example: Vector and tensor fields

The previous analysis can be easily extended to other kinds of fields. Consider for instance a vector
field Aµ and a tensor field Bµν . Which is the most general Lagrangian L(A, ∂A, B, ∂B) that can be
constructed out of these two fields? To clarify the procedure, let me split the problem into several
pieces
L(A, B, ∂A, ∂B) = LA (A, ∂A) + LB (B, ∂B) + Lint (A, B, ∂A, ∂B) . (9.21)
The lagrangian LA is a free lagrangian for Aµ , i.e. a linear combination of Lorentz invariant terms
quadratic in Aµ 3 ,
LA = a1 ∂µ Aµ ∂ν Aν + a2 ∂µ Aν ∂ µ Aν + a3 ∂µ Aν ∂ ν Aµ + a4 Aµ Aµ + a5 µνρσ (∂ ν Aµ )(∂ σ Aρ ) . (9.22)
The lagrangian LB is a free lagrangian for Bµν
LB = b1 ∂µ B νν ∂ µ B κκ + b2 ∂µ (B κν + B νκ ) ∂ µ (Bκν + Bνκ ) + b3 ∂µ (B µν + B νµ ) ∂ν B κκ + (9.23)
2
+ b4 ∂µ (B κµ + B µκ ) ∂ ν (Bκν + Bνκ ) + b5 µνρσ B µν B ρσ + b6 Bµν B µν + b7 Bµν B νµ + b8 (B µ µ ) .
The interaction Lagrangian Lint follows the same logic, but now involving the interactions between
the two fields,
Lint = i1 ∂µ Aν (B µν + B νµ ) + i2 Aµ ∂ν (B µν + B νµ ) + i3 µνρσ (∂ µ Aν )B ρσ +
+ i4 µνρσ Aµ (∂ ν B ρσ ) + i5 Aµ Aν B µν . (9.24)
I maybe missed some terms, but I think you get the idea. As in the scalar case, the list of operators
and independent coefficients can be shortened cases in which there are extra symmetries on top on
Lorentz invariance. Let me illustrate this with a simpler example.

The action for the electromagnetic field

Consider the action (9.22) alone. The first thing that can simplify our life is the gauge freedom in
the choice of the Lagrangian density. In particular notice that choosing gσ = µνρσ (∂ ν Aµ )Aρ in (9.9)
allows us to eliminate the term a5 in (9.22), since
∂ σ gσ = µνρσ (∂ ν Aµ ) (∂ σ Aρ ) + µνρσ (∂ ν ∂ σ Aµ ) Aρ . (9.25)
| {z }
0 by symmetry

Taking this into account we are left with an action containing 4 pieces
Z
S = d4 x [a1 ∂µ Aµ ∂ν Aν + a2 ∂µ Aν ∂ µ Aν + a3 ∂µ Aν ∂ ν Aµ + a4 Aµ Aµ ] . (9.26)

Imagine now that Aµ is the field of a gauge theory. In that case the field configurations Aµ and
A0µ = Aµ + ∂µ χ , (9.27)
with arbitrary scalar function χ give rise to the same physical observables4 . This automatically forbids
the a4 term in (9.26)5 and puts some restrictions on the other coefficients. To see this, let me split
∂µ Aν into its symmetric Sµν ≡ ∂(µ Aν) and antisymmetric Fµν ≡ ∂[µ Aν] parts
∂µ Aν = Sµν + Fµν , (9.28)
3 Quadratic actions give rise to linear equations of motion, where the superposition principle can be applied.
4 In the same way that physicality cannot be attributed to L, we cannot make any claim about the physicality of Aµ .
Physicality might be attributed to the set {Aµ } of gauge-equivalent 4-potentials or to any gauge invariant attribute of
that set, but not to its individual elements.
5 It cannot be compensated by the transformation of the other (derivative) terms.
9.4 The action for the graviton 134

and rewrite the action (9.26) as

Z
S = d4 x a1 S µµ S νν + (a2 + a3 )Sµν S µν + (a2 − a3 )Fµν F µν .
 
(9.29)

The invariance of the action under the gauge transformation Aµ → Aµ + ∂µ χ requires a1 = 0 and
a3 = −a2 . This restriction leaves us with an action
Z
S = d4 xFµν F µν , (9.30)

where we have omitted an overall normalization factor that can be determined by choosing the coupling
of the gauge field Aµ to matter and the units of that coupling. The equations of motion associated
with this action can be computed via the Euler-Lagrange equations (9.12) or by varying the action
with respect to Aµ . We follow the second procedure to get
Z Z
δS = d x [F δFµν + Fµν δF ] = 2 d4 xF µν δFµν
4 µν µν

Z Z
= 2 d4 xF µν (∂µ δAν − ∂ν δAµ ) = 4 d4 xF µν ∂µ δAν (9.31)
Z
= −4 d4 x∂µ F µν δAν + boundary terms ,

where we used the symmetry properties of Fµν and performed an integration by parts. Imposing
finally the condition δS = 0 for arbitrary δAν , we arrive to the very familiar result

∂µ F µν = 0. (9.32)

The Maxwell equations in vacuum are recovered from an action (9.30) constructed with very limited
principles, namely, quadraticity in the fields, Lorentz invariance and gauge invariance.

9.4 The action for the graviton

The procedure outlined in the previous section is quite powerful. As an interesting application for
General Relativity, let me consider the action for a second rank symmetric and massless tensor field
hµν . As in the vector field case, the kinetic term is constructed out of scalars that are quadratic in
the derivatives ∂ρ hµν . The most general expression will be the sum of the different scalars obtained
by contracting pairs of indices in all possible ways. The resulting action takes the form
Z
S = d4 x [c1 ∂µ hνν ∂ µ hκκ + c2 ∂µ hκν ∂ µ hκν + c3 ∂µ hµν ∂ν hκκ + c4 ∂µ hκµ ∂ ν hκν ] , (9.33)

with c1 , c2 , c3 , c4 some undetermined constants. These constants can be determined up to an overall

factor by requiring the action to be invariant under the gauge transformations

hµν → hµν − ∂µ ξν − ∂ν ξµ . (9.34)

Plugging (9.34) into (9.33) and performing some simple manipulations we get
Z
S → S + d4 x[−2(2c1 + c3 )∂µ hκκ ∂ µ ∂λ ξ λ − 2(2c2 + c4 )∂ µ hµν ξ ν − 2(c3 + c4 )∂ µ hµν ∂ ν ∂κ ξ κ

+ (4c1 + 2c2 + 4c3 + 3c4 )∂µ ∂κ ξ κ ∂ µ ∂λ ξ λ + (2c2 + c4 )ξµ ξ µ ] , (9.35)

9.5 Field theory in curved spacetime 135

which imposing δS = 0, provides the constraints

c2 = −c1 , c4 = −c3 = 2c1 . (9.36)

Taking this into account, the action (9.33) takes the form
Z
S = d4 x [∂µ hκν ∂ µ hκν + 2∂µ hµν ∂ν hκκ − 2∂µ hκµ ∂ ν hκν − ∂µ hνν ∂ µ hκκ ] , (9.37)

where we have omitted an overall normalization factor that can be determined by specifying the
coupling to matter and setting the units of the coupling. The associated equations of motion can be
obtained by varying the action with respect to the field. This leads to
Z
δS = d4 x 2∂ µ hκν ∂µ δhκν + 2∂ν hκκ ∂µ δhµν + 2ηκλ ∂µ hµν ∂ν δhκλ


−2∂ ν hκν ∂µ δhκµ − 2∂ µ hκµ ∂ν δhκν − 2ηκλ ∂ µ hνν ∂µ δhκλ ,


(9.38)

which, integrating by parts, dropping boundary terms and renaming indices can be written as
Z
δS = 2 d4 x −hκν δhκν − ∂µ ∂ν hκκ δhµν − ηκλ ∂µ ∂ν hµν δhκλ +∂µ ∂ ν hκν δhκµ + ∂ µ ∂ν hκµ δhκν + ηκλ hνν δhκλ
 

Z
= 2 d4 x [−hµν − ∂µ ∂ν hκκ − ηµν ∂κ ∂λ hκλ + ∂µ ∂ κ hκν +∂ν ∂ κ hκµ + ηµν hκκ ] δhµν . (9.39)

A simple inspection reveals that the quantity inside the square brackets is nothing else than the
linearized version of the Einstein tensor Gµν
Z
δS ∝ d4 x Gµν δhµν . −→ Gµν = 0 . (9.40)

The linearized version of Einstein equations in vacuum are recovered from an action (9.39) constructed
with very limited principles, namely, quadraticity in the fields, Lorentz invariance and gauge invari-
ance.

9.5 Field theory in curved spacetime

The variational approach for fields presented in the previous section can be generalized to include the
interaction with gravity. We guess the form of the action in this case with the help of the Equivalence
Principle:

• Replace the Minkowski metric ηµν by gµν .

• Replace partial derivatives by covariant derivatives (colon-goes-to semicolon rule).
p
• Replace the Lorentz invariant volume element dn x by the covariant volume element dn x |g|.
p
Since L̃ and dn x |g| are scalars under general coordinate transformations, the resulting action
Z p
S = dn x |g| L̃(φ, ∇µ φ, gµν ) (9.41)
| {z }
L
p
is guaranteed to provide covariant equations of motion. Note that the untilded quantity L ≡ |g|L̃
is a scalar density of weight 1.
9.6 The energy-momentum tensor 136

9.6 The energy-momentum tensor

Consider now an arbitrary infinitesimal coordinate transformation

xµ → x̄µ = xµ + ξ µ (x) . (9.42)

This transformation generates a perturbation to both the fields and the metric in such a way that the
Lagrangian density L (no tilde) becomes

∂L ∂L ∂L
L(φ + δφ, φ,µ + δφ,µ , gµν + δgµν ) ≈ L(φ, φ,µ , gµν ) + δφ + δφ,µ + δgµν . (9.43)
∂φ ∂φ,µ ∂gµν

Integrating by parts, we get two pieces

Z    Z
n ∂L ∂L ∂L
δS = d x − ∂µ δφ + dn x δgµν . (9.44)
∂φ ∂(∂µ φ) ∂gµν
| {z }
=0

The first one is associated to a particular variation δφ and vanishes when taking into account the
Euler-Lagrange equation for φ. The second term must be then equal to zero for S to remain unchanged.
The integrand ∂L/∂gµν is a scalar density. Let’s define a symmetric second-rank tensor out of such a
density
2 ∂L
T µν ≡ p , (9.45)
|g| ∂gµν
and write Z
1 p
δS = dn x |g|T µν δgµν . (9.46)
2
p
Although is tempting to simply set |g|Tµν = 0, this condition is overly restrictive, since δgµν refers
here to a specific type of variation, not to an arbitrary one. The variation δgµν can be however
expressed in terms of the arbitrary perturbation ξµ by taking into account that δgµν = −(ξµ;ν + ξν;µ )
(cf. Eq. (6.60) and notice (9.50)). This gives
Z Z
1 p p
δS = dn x |g|T µν δgµν = − dn x |g|T µν ξµ;ν
2
Z p Z p 
= d x |g|T ;ν ξµ − dn x
n µν
|g|T µν ξµ , (9.47)
;ν
| {z }
=0

where we have made use of the symmetry property of T µν and integrated by parts to get a total
derivative that vanishes by assumption on the boundary of integration. Since ξµ is arbitrary we must
have
∇ν T µν = 0 , (9.48)
which is a continuity equation suggesting that we can identify the tensor (9.45) with the energy-
momentum tensor of any physical system.
9.7 The Einstein-Hilbert action 137

A common sign mistake

You will find some books giving an alternative definition of the energy momentum tensor
2 ∂L
Tµν = − p , (9.49)
|g| ∂g µν

in terms of δg µν rather than δgµν . The difference in sign between these two equivalent expres-
sions comes from

gµν g νλ = δ µ λ δ gµν g νλ = 0 δg µν = −g µλ g νρ δgλρ .

→ → (9.50)

9.6.1 A particular case

p
When the Lagrangian L̃ on L = |g|L̃ depends only on the metric and not on the first derivatives
of the metric6 , i.e. L̃ = L̃(φ, ∂µ φ, gµν ), it is possible to derive an
√ alternative expression for the
∂ |g| 1
p µν
energy-momentum tensor. In particular, taking into account that ∂gµν = 2 |g|g , we can write
p ! !
µν 2 ∂L 2 ∂ |g| p ∂ L̃ 2 1p p ∂ L̃
T =p =p L̃ + |g| =p |g|g µν L̃ + |g| , (9.51)
|g| ∂gµν |g| ∂gµν ∂gµν |g| 2 ∂gµν

or equivalently
∂ L̃
T µν = g µν L̃ + 2 . (9.52)
∂gµν

Exercise
Compute the energy-momentum tensor for (9.7).

9.7 The Einstein-Hilbert action

In order to construct an action for the gravitational field we must define a Lagrangian. This Lagrangian
must transform as a scalar under general coordinate transformations and depend on the metric tensor
and its derivatives. Note however that gravity is completely different from all other fundamental
interactions, since non-trivial quantity can be constructed from the metric and its first derivatives
alone. This can be easily seen by considering an arbitrary scalar combination f (gµν , ∂ρ gµν ) of this
quantities around a small region in spacetime. According to the local flatness theorem, in such a
region, it is always possible to find coordinates such that gµν = ηµν and ∂ρ gµν = 0 and therefore
f = constant. But ,since we are dealing with a scalar quantity, this will be the case in any other
coordinate system. In other words, any covariant scalar function constructed just from the metric
tensor and its derivatives will be a trivial constant.
The simplest non-trivial quantity that can be constructed from the metric and its derivatives is the
Ricci scalar, which depends on the metric and its first and second order derivatives. The resulting
action reads Z Z Z
4 1 4
p 1 p
SEH = d xLEH = 2 d x |g|R = 2 d4 x |g|g µν Rµν (9.53)
2κ 2κ
6 Particular cases are the lagrangian for scalar fields or that for the electromagnetic field, where the covariant derivative

reduces to the standard derivative.

9.7 The Einstein-Hilbert action 138

and is known as the Einstein-Hilbert action. The constant of proportionality κ2 is included on dimen-
sional grounds and will be determined of the end of the computation.

Exercise
Which is the dimension of κ2 ?

Consider the variation of (9.53) resulting from the variation of the metric tensor

gµν → gµν + δgµν , (9.54)

where δgµν and its first derivative are assumed to vanish at the boundary of the integration region.
We obtain
p p p
δLEH ∝ |g|δg µν Rµν + δ |g|R + |g|g µν δRµν , (9.55)
| {z } | {z }
δL1 δL2

where we have defined two pieces, L1 and L2 . The first one can be easily evaluated by taking into
account the variation and the variation of the metric determinant
p 1p
δ |g| = − |g|gµν δg µν . (9.56)
2

Exercise
Prove Eq. (9.56).

Collecting the terms, we get  

p 1
δL1 = |g| Rµν − gµν R δg µν . (9.57)
2
The evaluation of δL2 is slightly more involved since it requires to perform the variation of the Ricci
tensor. The easiest way of doing this is to consider the variation of the Riemann tensor and perform
the required contractions at the end of the computation. Schematically, the variation of the Riemann
tensor has the structure

Riemann ∼ ∂Γ + ΓΓ −→ δ (Riemann) ∼ δ (∂Γ) + (δΓ) Γ , (9.58)

so the first thing that we have to compute is the variation of Christoffel symbols δΓρµν defined by

Γρ µν −−−−−−→ Γ̃ρ µν = Γρ µν + δΓρµν =⇒ δΓρ µν = Γ̃ρ µν − Γρ µν . (9.59)

gµν +δgµν

Be careful! We are just performing a variation of the metric, not transforming it.

It is easy to see that, even though a connection is not a tensor, the difference of two connections δΓρµν
transforms as a tensor, i.e.
∂ x̄ρ ∂xλ ∂xκ ρ
δ Γ̄ρ µν = δΓ λκ . (9.60)
∂xσ ∂ x̄µ ∂ x̄ν
9.7 The Einstein-Hilbert action 139

Exercise
Check it.

The property (9.60) extremely simplifies the computation of the variation of the Riemann tensor.
Indeed, we can always go to a local free fall reference frame in which Γρµν = 0 at some arbitrary point
P . In such a point the expression (9.58) becomes

δRµν = ∂ρ (δΓρ µν ) − ∂ν (δΓρ ρµ ) = ∇ρ (δΓρ µν ) − ∇ν (δΓρ ρµ ) , (9.61)

where in the last step used of the fact that the partial and covariant derivatives coincide when Γρ µν = 0.
The resulting Palatini equation

δRµν = ∇ρ (δΓρ µν ) − ∇ν (δΓρ ρµ ) (9.62)

is a tensorial equation (remember the property (9.60)) valid in any arbitrary coordinate system (and
not only in the free fall reference frame at P ).

Exercise
Prove that the second term in the previous expression is symmetric, as it should be.

A similar trick can be applied to get the explicit expression of δΓ, that takes the same form that the
definition of the Christoffel symbols, with the metric replaced by the metric variation and the partial
derivatives replaces by covariant derivatives, i.e.
1 µσ
δΓµ νρ = g (∇ν δgσρ + ∇ρ δgσν − ∇σ δgνρ ) . (9.63)
2

Exercise
Check the previous expression by explicit computation.

Taking (9.62) into account, δL2 becomes

p p
δL2 = |g| g µν [∇ρ (δΓρ µν ) − ∇ν (δΓρ ρµ )] = |g| ∇σ [g µν δΓσ µν − g µσ δΓρ ρµ ] , (9.64)

where we have used the metric compatibility condition (4.68). We are left therefore with the covariant
divergence of a vector. Using the property
1 p 
V µ ;µ = p |g|V µ , (9.65)
|g| ,µ

we get the boundary term

p hp p i
δL2 = |g| ∇σ [g µν δΓσ µν − g µσ δΓρ ρµ ] = ∂σ |g|g µν (δΓσ µν ) − |g|g µσ (δΓρ ρµ ) . (9.66)
9.8 Einstein equations in the presence of matter 140

As I said before, gravity is a quite particular field theory. The existence of second derivatives
in the Einstein-Hilbert action gives rise to a contribution depending on the value of the first
derivatives on the boundary. To deal with these, we have two options:
• Extend the variational principle and require the fields and their derivatives to be fixed at
the boundary. This would give rise to reasonable field equations. A clear example from
classical mechanics illustrating this would be
Z tf 
1 tf 2
 Z
1 2
S= dt q̈ + q̇ = q̇(tf ) − q̇(t0 ) + q̇ , (9.67)
t0 2 2 t0

with the assumption that both q̇ and q are fixed at the boundary. This approach has
however some caveats. On the one hand, it does not obey a composition rule of the kind

S(0 → 1 → 2) = S(0 → 1) + S(1 → 2) , (9.68)

where the paths connecting (q0 , t0 ) and (q2 , t2 ) are decomposed at an intermediate time
t1 with t0 < t1 < t2 . Although the paths are expected to be continuous at t = t1 , they do
not need to be smooth at that point which requires leaving q̇1 free at t = t1 . On the other
hand, the action principle has its roots in quantum mechanics, where the simultaneous
fixing of q and q̇ is inappropriate.
• Add the so-called Gibbons-Hawking-York counterterm to the action
Z Z
1 p 1 p
S = SEH + SHGY = 2 d4 x |g|R + 2 d3 x |h|K , (9.69)
2κ R κ ∂R

with h the determinant of the induced metric on the boundary and K the trace of the
extrinsic curvature. The Gibbons-Hawking-York is constructed in such a way that its
variation cancels the unwanted term associated to the second derivatives of the metric,
keeping only the part associated to the quadratic part of the action. Proving this statement
is beyond the scope of this course. The interested reader is referred to the excellent
discussion in Padmanabhan’s book.
Forgetting about the boundary term, the Einstein-Hilbert action (9.53) becomes
Z  
1 p 1
δSEH = 2 d x |g| Rµν − gµν R δg µν ,
4
(9.70)
2κ 2

which, demanding it to vanish for arbitrary variations δg µν , gives us the Einstein’s equations in the
absence of matter
1
Gµν ≡ Rµν − Rgµν = 0 . (9.71)
2

9.8 Einstein equations in the presence of matter

Having obtained the Einstein equations in the vacuum, let us now derive its full form in the presence
of matter. Consider the action
S = SEH + SM , (9.72)
9.8 Einstein equations in the presence of matter 141

with SM containing all the non-gravitational fields. The variation of SM with respect to δg µν (upper
indices) gives Z
1 p
δSM = − d4 x |g|Tµν δg µν , (9.73)
2
where we have made use of the covariant definition (9.49). Putting everything together we get
Z  
1 p 1
δSEH + δSM = 2 d4 x |g| Rµν − R − κ2 Tµν δg µν . (9.74)
2κ 2

Since δg µν is arbitrary, we must have

Gµν = κ2 Tµν , (9.75)
which confirms the identification of (9.45), or (9.49), with the energy-momentum tensor and allows
us to identify the proportionality constant κ2 with 8πG.

Exercise
Modify the Einstein-Hilbert action to obtain the Einstein equations with cosmological constant.