4 General Theory of Relativity

Physical theories in mathematical form are complete, if mathematical structure is identified with physical observables.
In general relativity, the spacetime is a four-dimensional manifold with a metric, which can be measured by local exper-
iments, and the metric can be locally made Minkowski by a choice of coordinate. Furthermore, freely falling particles
move on a time-like geodesic (massless particles on a null geodesic). In addition to this identification, we need to derive
the Einstein equation, describing how physical objects and a curved spacetime are related.

4.1 Einstein Equations

4.1.1 Principle of General Covariance
The principle of general covariance states that the physical laws should take the same (tensorial) form in all coordinate
systems:

• The equations in general should reduce to those in special relativity (in the absence of gravity), when gµ⌫ ! ⌘ab
and ⇢µ⌫ ! 0,

• the equations should be in a covariant form.

This implies that we can derive the general tensorial equations for most physical laws that are valid in the presence of
gravity, simply by taking those in special relativity (free-falling system) without gravity and making them covariant:
dua Duµ
⌘ab ! gµ⌫ , m = fa ! m = fµ , (4.1)
d⌧ d⌧
where the covariant derivative of the four velocity is
Duµ duµ µ
:= + ⇢ u⇢ u = u⌫ uµ ;⌫ . (4.2)
d⌧ d⌧
In the same way, the Maxwell’s equations (1.82) and (1.83) in special relativity becomes

@⌫ F µ⌫ = 4⇡J µ ! r⌫ F µ⌫ = 4⇡J µ , 0 = "⇢ µ⌫

@ Fµ⌫ ! 0 = "⇢ µ⌫
r Fµ⌫ . (4.3)

4.1.2 Energy-Momentum Tensor for Macroscopic Fluids

For our purposes, we are not interested in the microscopic states of the systems, but interested in their macroscopic states,
often described by the density, the pressure, the temperature, and so on. Since the energy-momentum tensor was already
derived in special relativity, we can use the principle of general covariance to generalize the energy-momentum tensor in
a curved spacetime.
The energy-momentum tensor for a fluid can be expressed in terms of the fluid quantities measured by an observer
with four velocity uµ as (the most general decomposition)

Tµ⌫ := ⇢uµ u⌫ + pHµ⌫ + qµ u⌫ + q⌫ uµ + ⇡µ⌫ , 0 = Hµ⌫ u⌫ , (4.4)

where Hµ⌫ is the projection tensor and

Hµ⌫ = gµ⌫ + uµ u⌫ , Hµµ = 3 , uµ qµ = 0 = uµ ⇡µ⌫ , ⇡µ⌫ = ⇡⌫µ , ⇡µµ = 0 . (4.5)

The variables ⇢, p, qµ and ⇡µ⌫ are the energy density, the isotropic pressure (including the entropic one), the (spatial)
energy flux and the anisotropic pressure measured by the observer with uµ , respectively, i.e.,
1
⇢ = Tµ⌫ uµ u⌫ , p = Tµ⌫ Hµ⌫ , qµ = T⇢ u⇢ H µ , ⇡µ⌫ = T⇢ Hµ⇢ H⌫ pHµ⌫ . (4.6)
3
Remember that these fluid quantities are observer-dependent.
AST511 General Relativity JAIYUL YOO

4.1.3 Einstein’s Equation

While most of the physical laws can be deduced by using the principle of general covariance, the physical laws for gravity
cannot be deduced, as there is no gravity in a free-falling system, where special relativity is valid. In the non-relativistic
limit, the Newtonian mechanics works, and the gravitational field equation should reduce to the Poisson equation
r2 = 4⇡G⇢ . (4.7)
Given that the metric and the energy momentum tensor in the Newtonian limit are
✓ ◆
2
g00 ' 1+ 2 , T00 ' ⇢c2 , (4.8)
c
the gravitational field equation in the Newtonian limit is expected to be
8⇡G
r2 g00 = T00 , (4.9)
c4
where the proportionality constant was set by the Poisson equation = 4⇡G⇢. Note that the left-hand side is a function
of metric, and the right hand-side is a matter component.
Using the principle of general covariance, we can deduce the gravitational field equation in any coordinate to take a
tensorial form:
8⇡G µ ⌫ 2
Gµ⌫ = 4 Tµ⌫ , lim Gµ⌫ = 0 0 r gµ⌫ , (4.10)
c nr
where the limit represents the non-relativistic case and Gµ⌫ is a function of metric. The requirement for the LHS is as
follows: Gµ⌫ should be tensorial, and it should be symmetric and conserved:
Gµ⌫ = G(µ⌫) , Gµ⌫;µ = 0 , (4.11)
as Tµ⌫ is symmetric and conserved. Furthermore, the dimension of Gµ⌫ should be two in mass dimension:
1
[Tµ⌫ ] = L 4
, 8⇡G = 2
2 =L , ) [Gµ⌫ ] = L 2
. (4.12)
Mpl
A tensor Gµ⌫ that is a function of metric tensor gµ⌫ and of mass dimension two is uniquely determined up to two arbitrary
dimensionless constants as
Gµ⌫ = c1 Rµ⌫ + c2 R gµ⌫ , (4.13)
which is already symmetric. The conservation condition Gµ⌫;µ = 0 imposes
✓ ◆
1 1 8⇡G
c2 = c1 , Gµ⌫ ⌘ c1 Rµ⌫ R gµ⌫ = 4 Tµ⌫ . (4.14)
2 2 c
The remaining constant c1 can be determined by taking the non-relativistic limit:
✓ ◆
1
G00 = c1 R00 R ⌘00 G↵ , (4.15)
2
where the inequality arises from the non-relativistic conditions in Tµ⌫ . Noting that in this limit
g µ⌫ Gµ⌫ = c1 R ' ⌘ 00 G00 = G00 , (4.16)
we just need to compute only one component R00 of the Riemann tensor or just the Ricci scalar R:
G00 ' 2c1 R00 ' c1 R , R ' 2R00 . (4.17)
The Ricci tensor in the non-relativistic limit is
1
Rµ⌫ = R⇢ µ⇢⌫ ' ⇢
µ⌫,⇢
⇢
µ⇢,⌫ , R00 ' i
00,i '
h00 , (4.18)
2
where we kept only the spatial derivative terms. Therefore, we obtain the Einstein tensor G00 and fix the coefficient c1 :
G00 ' c1 r2 h00 , c1 ⌘ 1 . (4.19)
The gravitational field equation (or Einstein’s equations) is
1 8⇡G 8⇡G
Gµ⌫ := Rµ⌫ R gµ⌫ = 4 Tµ⌫ , R= T. (4.20)
2 c c4
Given the energy-momentum tensor, the Ricci scalar is algebraically determined.

33
AST511 General Relativity JAIYUL YOO

4.1.4 Structure of the Einstein Field Equation

The Einstein equations Gµ⌫ = 8⇡G Tµ⌫ constrain the relation between the metric and the energy momentum tensor. In
a Machian view, the existence of matter determines the spacetime geometry, and the matter moves along the straightest
path in a given spacetime geometry. This is summarized in a well-known statement by John Wheeler: “matter tells how
to curve, and spacetime tells matter how to move.”

• Degrees of freedom.— It is clear that any symmetric tensor gµ⌫ plugged in the Einstein equations will result in un-
physical energy-momentum tensor. In general, the field equation is considered as “constraints” for twenty independent
variables in gµ⌫ and Tµ⌫ . Given ten dofs in Tµ⌫ , ten Einstein equations provide ten constraints for gµ⌫ . Due to the Bianchi
identity, however, four of the Einstein equations are redundant. This situation of under-determination is saved by general
covariance or gauge transformation, which takes away four dofs in gµ⌫ or Tµ⌫ .

• Nonlinearity.— The Einstein equations are non-linear, such that one cannot add two independent solutions to derive
another solution. In other words, one cannot analyze a complex system by breaking it into several simpler systems and
adding the solutions of the simpler systems to obtain the solution of the complex system. Consider a matter source, and
it produces gravitational fields, which contain energy. This energy of the gravitational fields is then equivalent to mass,
which in turn produces gravitational fields. Due to this complexity in the field equation, Einstein anticipated that no one
would find an exact solution to the nonlinear equations, and he was greatly surprised to see Schwarzschild solution, less
than a year after general relativity was published.

• Hidden boundary conditions?.— Einstein was puzzled by the existence of a Minkowski solution for an empty universe,

gµ⌫ ⌘ ⌘µ⌫ , Rµ⌫ = Tµ⌫ = 0 , (4.21)

which would be in conflict with Mach’s principle. Some unknown boundary conditions might be needed to exclude such
solutions. Einstein in the end added a cosmological constant ⇤ to the field equation:
1
Rµ⌫ gµ⌫ R + ⇤ gµ⌫ = 8⇡G Tµ⌫ , (4.22)
2
to remove the Minkowski solution and to make the Universe static. After the introduction of his cosmological constant in
1917, Edwin Hubble found in 1929 that the Universe was expanding and Einstein allegedly said “it was my biggest blun-
der!” A cosmological constant makes the Universe unstable and indeed expand in an accelerating manner. Furthermore,
de Sitter also found a solution in an empty universe with non-zero cosmological constant.

• Geodesic equation without field equation.— The particle motion or the geodesic equation is expected to follow from
the Einstein field equation, as the matter determines the spacetime and the spacetime affects the trajectory. It turns out,
however, that the geodesic equation is rather generic and follows only from the conservation equation, not from the field
equation. Consider dust particles ⇢ moving with a time-like vector uµ , and the conservation equation yields

Tµ⌫ = ⇢uµ u⌫ , 0 = r⌫ T µ⌫ = ⇢u⌫ r⌫ uµ + uµ r⌫ (⇢u⌫ ) . (4.23)

Given the normalization 1 = uµ uµ , and 0 = uµ r⌫ uµ , the conservation equation yields

0 = r⌫ (⇢u⌫ ) , 0 = u ⌫ r⌫ u µ . (4.24)

The latter is the geodesic equation.

4.1.5 Cauchy Problem: Initial Value Problem

The Cauchy problem refers to the case of whether the physical system given the governing equation can be solved from
the initial conditions. In general relativity, the initial conditions are the metric tensor gµ⌫ at some initial time t (or hyper-
surface) and its time derivative gµ⌫,0 . Mind that given the information on the initial hypersurface, the spatial derivative of
the metric tensor is also available, i.e., all first derivative gµ⌫,⇢ information at the initial hypersurface.
For simplicity, we consider a vacuum solution to the Einstein equation Rµ⌫ = 0. If we obtain the condition for gµ⌫,00
from the Einstein equation and the initial conditions, we obtain all the derivatives of the metric tensor by subsequently

34
AST511 General Relativity JAIYUL YOO

taking derivatives of the Einstein equation and we derive the solution for gµ⌫ everywhere all the time, provided that the
metric tensor is differentiable and analytic.
In fact, computation of the Einstein equation shows
1 ↵ 1 1 00
R00 = g g↵ ,00 + I00 = 0 , R0↵ = g 0 g↵ ,00 + I0↵ = 0 , R↵ =+ I↵ = 0 , g g↵ ,00
2 2 2
(4.25)
where Iµ⌫ are the tensors written only in terms of the initial conditions. The dynamical equations contain no information
about g00,00 or g0↵,00 (under-determination), or ten dynamical equations over-determine six g↵ ,00 . The first issue can be
removed by a specific coordinate transformation
1
x̃µ = xµ + t3 F µ (x) , (4.26)
6
to set g̃00,00 = g̃0↵,00 = 0, while keeping the initial conditions unchanged:

g̃µ⌫ = gµ⌫ , g̃µ⌫,⇢ = gµ⌫,⇢ , g̃µ⌫,↵⇢ = gµ⌫,↵⇢ , (4.27)

but
g̃00,00 = g00,00 2g0µ F µ , g̃0↵,00 = g0↵,00 g↵µ F µ , g̃↵ ,00 = g↵ ,00 . (4.28)
The dynamical equations are indeed not over-determined. Provided 6= 0, the last dynamical equation R↵ = 0
g 00
can be used to solve for g↵ ,00 , then two remaining dynamical equations are then the constraint equations on the initial
conditions. Re-arranging the dynamical equations, we obtain

g 00 R00 g ↵ R↵ = g 00 I00 g ↵ I↵ = 0 , g 00 R0↵ + g 0 R↵ = g 00 I0↵ + g 0 I↵ = 0 , (4.29)

which are also equivalent to

G↵ 0 = 0 . (4.30)
As long as the main dynamical equation R↵ = 0 is satisfied, R00 = 0 and R0↵ = 0 are satisfied only in terms of initial
conditions. Hence, the Einstein equations are now composed of six dynamical equations R↵ = 0 and four constraint
equations G↵ 0 = 0. Once the constraint equations are satisfied on the initial hypersurface, they are satisfied all the time
due to the Bianchi identity:

0 = r ⌫ Gµ ⌫ = r 0 Gµ 0 + r↵ Gµ ↵ , r 0 Gµ 0 = r ↵ Gµ ↵ = 0 , (4.31)

where the second equality can be proved by showing that r↵ Gµ ↵ is a linear combination of Gµ 0 and Gµ 0 ,↵ , both of
which vanish on the initial hypersurface.

• also add § 8 in Wald (1984).

4.2 Einstein-Hilbert Action

4.2.1 Gauge Transformation
The general covariance of general relativity guarantees that any coordinate system can be used to describe the physics
and it has to be independent of coordinate systems. This is known as the diffeomorphism symmetry in general relativity.
However, when we split the metric into the background and the perturbations around it by choosing a coordinate system,
we explicitly change the correspondence of the physical Universe to the background homogeneous and isotropic Universe.
Hence, the metric perturbations transform non-trivially (or gauge transform), and the diffeomorphism invariance implies
that the physics should be gauge-invariant.
The gauge group of general relativity is the group of diffeomorphisms. A diffeomorphism corresponds to a differen-
tiable coordinate transformation. The coordinate transformation on the manifold M can be considered as one generated
by a smooth vector field ⇣ µ . Given the vector field ⇣ µ , consider the solution of the differential equation

d µ( ) d
µ
= ⇣µ [ ⌫
P( )] , P( = 0) = xµP , = ⇣ µ @µ , (4.32)
d P d

35
AST511 General Relativity JAIYUL YOO

defines the parametrized integral curve xµ ( ) = µP ( ) with the tangent vector ⇣ µ (xP ) at P . Therefore, given the vector
field ⇣ µ on M we can define an associated coordinate transformation on M as xµP ! x̃µP = µP ( = 1) for any given P .
Assuming that ⇣ µ is small one can use the perturbative expansion for the solution of equation to obtain

d 1 d2
x̃µP = µ
P( = 1) = µ
P( = 0) + µ
P + µ
P + ··· (4.33)
d =0 2d 2 =0
1 ⌫
= xµP + ⇣ µ (xP ) + ⇣ µ ,⌫ ⇣ ⌫ + O(⇣ 3 ) = e⇣ @⌫ xµ .
2
This parametrization corresponds to the gauge-transformation with ⇣ µ .
In general, any gauge-transformation of tensor T for an infinitesimal change ⇣ can be expressed in terms of the Lie
derivative (valid to all orders of T)
⇣ T := T̃ T = £⇣ T + O(⇣ 2 ) , (4.34)
To all orders in ⇣, we have
1
£⇣ T + £2⇣ T + · · · = exp [ £⇣ ] T .
T̃(x) = T(x) (4.35)
2
Therefore, the gauge-transformation in perturbation theory is simply
(1) (n)
⇣ T̄ =0, ⇣T = £⇣ T̄ , ⇣T = £⇣ T(n 1)
, (4.36)

where we used that ⇣ is also a perturbation. At the linear order, the Lie derivative is trivial, and the the most general
coordinate transformation in Eq. (4.34) becomes

x̃µ = xµ + ⇠ µ , (4.37)

where we now use ⇠ µ = ⇣ µ . The transformation of the metric tensor at the leading order in ⇠ is then

⇠ gµ⌫ (x) := g̃µ⌫ (x) gµ⌫ (x) = £⇠ gµ⌫ = (⇠µ;⌫ + ⇠⌫;µ ) , (4.38)

where the semi-colon represents the covariant derivative with respect to the full metric gµ⌫ .

4.2.2 Variational Method and Equation of Motion

Physical theories can be concisely expressed in terms of its action S, and the principle of least action states that the
equation of motion in a given theory follows the path that minimizes the change in S. The action S is a general and
non-local function of its dynamic variables such as , , Aµ , gµ⌫ , but in most cases the actions can be expressed in terms
of local functions, called Lagrangian: Z Z
S =: dt L =: d4 x L , (4.39)

where L is the Lagrangian and L is the Lagrangian (density). The dimension

M · L2 M · L2 M
[S] = ⇠h, [L] = ⇠E, [L] = ⇠ ⇢E , (4.40)
T T2 L · T2
where h is the Planck constant whose dimension is the same as the action as the angular momentum (times the angle) or
the energy times the time.
Since our interest is a theory of gravity, the Lagrangian should be a function of metric gµ⌫ and its spacetime derivatives
as
Lg = Lg (gµ⌫ , gµ⌫,⇢ , gµ⌫,⇢ , · · · ) . (4.41)
When we make a variation to the dynamical variable g̃µ⌫ = gµ⌫ + gµ⌫ , the action must remain stationary:
Z
Lg Lg
0 = S̃ S = d4 x gµ⌫ , Lµ⌫
g := =0, (4.42)
gµ⌫ gµ⌫

and we have derived the equation of motion for Lg . Note that the tensor Lµ⌫ is symmetric and of density 1, as gµ⌫ is
symmetric and the volume factor is of weight +1. Furthermore, the general covariance guarantees that we can use any

36
AST511 General Relativity JAIYUL YOO

coordinates, and under a general coordinate transformation the metric tensor transforms as in Eq. (4.38). In this case, the
least action condition yields
Z Z 
gµ⌫ = 2 ⇠(µ;⌫) , 0 = S = 2 d x Lg r⌫ ⇠µ = 2 d x r⌫ (Lµ⌫ ⇠µ ) ⇠µ r⌫ Lµ⌫ .
4 µ⌫ 4
(4.43)

The first term in the square bracket vanishes, because a divergence of a tensor of weight 1 can be replaced with a
coordinate derivative and integration by part yields a surface term, where assume that the metric variation gµ⌫ vanishes
on the boundary of the manifold. Hence, we obtain another condition for our gravity theory Lg

r⌫ Lµ⌫ = 0 . (4.44)

In fact, these conditions are valid for any gravity theories, as long as it is a local function of metric tensor and it contains
the general covariance.

4.2.3 Einstein-Hilbert Action

Having discussed the general properties of Lg , we look for a specific Lagrangian for general relativity. First, note that Lg
should be also a scalar of weight 1. The simplest possibility we can construct out of gµ⌫ is
p
Lg ⌘ g, (4.45)

and we can readily derive

p 1p 1p
( g) = g g µ⌫ gµ⌫ , Lµ⌫ = g g µ⌫ , r⌫ Lµ⌫ = 0 . (4.46)
2 2
This theory of gravity is not dynamic. The next simplest possibility as a function of gµ⌫ is called the Einstein-Hilbert
action: Z
p p c4
LEH := gR, SEH := d4 x g R, (4.47)
16⇡G
where we restored the dimensionful constants in the action. The variation of the Einstein-Hilbert action gives two terms
p 1p p
( g R) = g g µ⌫ gµ⌫ R + g R, (4.48)
2
and we need to compute the variation of the Ricci tensor

R = g µ⌫ Rµ⌫ , R = g µ⌫ Rµ⌫ + g µ⌫ Rµ⌫ , (4.49)

µ
where the first term can be manipulated by using ⌫ = g µ⇢ g⇢⌫ as

g µ⌫ gµ⌫ = gµ⌫ g µ⌫ , g µ⌫ = g µ⇢ g ⌫ g⇢ , gµ⌫ = gµ⇢ g⌫ g⇢ , (4.50)

and the second term can be from the Palatini identity in Eq. (3.118):
⇢ ⇢
Rµ⌫ = µ⌫ ;⇢ µ⇢ ;⌫ . (4.51)

Foremost, we deal with the second term with Rµ⌫ and show that it is just a surface term in the action. With r⇢ gµ⌫ =
0 , we first compute

g µ⌫ Rµ⌫ = g µ⌫ ⇢
µ⌫ ;⇢ g µ⌫ ⇢
µ⇢ ;⌫ = rµ g ⇢ µ
⇢ g ⇢µ ⇢ =: rµ V µ , (4.52)

where we re-arranged the dummy indicies and defined a vector V µ . The contribution of this term in the action is then
p @ p
g rµ V µ = gVµ , (4.53)
@xµ

37
AST511 General Relativity JAIYUL YOO

integrated by part and removed. The remaining terms in the action is now
Z  Z
c4 4 p 1 µ⌫
SEH = d x g gµ⌫ g R g ⇢µ g ⌫ R⇢ =: d4 x gµ⌫ Lµ⌫
EH , (4.54)
16⇡G 2
and the Einstein (vacuum) field equation is
1
0= Lµ⌫
EH = R
µ⌫
R g µ⌫ =: Gµ⌫ , (4.55)
2
where we defined the Einstein tensor Gµ⌫ . Given the general covariance, we derived the differential identity:

0 = r⌫ Lµ⌫
EH = r⌫ G
µ⌫
, (4.56)

which is known as the Bianchi identity.

In the Einstein-Hilbert action, we can add a cosmological constant term in terms of the simplest Lagrangian Lg we
considered first in Eq. (4.45)
Z ✓ ◆
⇤ p 4 p R ⇤
L⇤ := g, Sg = d x g , (4.57)
8⇡G 16⇡G 8⇡G
and this additional term yields
1
0= Lµ⌫ = Rµ⌫ R g µ⌫ + ⇤ g µ⌫ . (4.58)
2

4.2.4 Matter Action and Energy-Momentum Tensor

Now we consider the matter Lagrangian Lm in particle physics, which includes scalar fields, vector fields (E&M), and
other fluids. With the matter action, the full action including gravity becomes
Z ✓ ◆
p R ⇤
S = Sg + Sm = d 4 x g + Lm , (4.59)
16⇡G 8⇡G

where we kept the notation for the matter Lagrangian Lm .1 The variation of Sm with respect to the metric yields
Z ✓ ◆
4 p 1 µ⌫ Lm
Sm = d x g gµ⌫ g Lm + , (4.60)
2 gµ⌫
and the full Einstein field equations becomes
✓ ◆
16⇡G 1 µ⌫ Lm 8⇡G µ⌫
Gµ⌫ = g Lm + =: T , (4.61)
c4 2 gµ⌫ c4
where we defined the energy-momentum tensor in a formal way as
Z
2 Sm
T µ⌫ := p = g µ⌫ Lm + 2 d4 x L m . (4.62)
g gµ⌫ gµ⌫
Given that we can define Z
1p
Sm = d4 x gµ⌫ Lµ⌫
m , Lµ⌫
g T µ⌫ ,
m = (4.63)
2
the differential identity condition gives the energy-momentum conservation

0 = r⌫ Lµ⌫
m = r⌫ T
µ⌫
. (4.64)

For the same computation, we obtain

Z
2 Sm
Tµ⌫ := p = gµ⌫ Lm 2 d4 x L m . (4.65)
g g µ⌫ g µ⌫
1
Mind that the Lagrangian density we defined is a scalar of weight 1, while in particle physics Lm is derived in the Minkowski spacetime
p p
⌘ ⌘ 1. So we kept Lm but add g to the Lagrangian in general relativity.

38
AST511 General Relativity JAIYUL YOO

Mind the subtlety, for example, for the scalar field action,
1 µ⌫ 1 µ⌫ 1
L = g @µ @⌫ , L = g @µ @⌫ = gµ⌫ @ µ @ ⌫ . (4.66)
2 2 2

• E&M Action.— The action for the E&M is

Z ✓ ◆
4 p 1 µ⌫ µ
SE&M = d x g Fµ⌫ F + 4⇡ J Aµ . (4.67)
4

Using
Fµ⌫ F µ⌫ = g µ⇢ g ⌫ Fµ⌫ F⇢ , (@⇢ A ) Fµ⌫ = @⇢ (A Fµ⌫ ) A @⇢ Fµ⌫ , (4.68)
we vary the action with respect to A and perform the integration by part to derive the equation of motion

r⌫ F µ⌫ = 4⇡J µ . (4.69)

Mind that the integration by part is performed by a coordinate derivative but the metric tensor commutes only with a
covariant derivative, which is possible due to the density weight of the Lagrangian. By varying the action with respect to
the metric tensor, we then obtain the energy-momentum tensor for E&M:
✓ ◆
1
Tµ⌫ = Fµ⇢ F⌫ ⇢ + gµ⌫ F⇢ F ⇢ + 4⇡J ⇢ A⇢ . (4.70)
4

4.2.5 Structure of the Einstein-Hilbert Action

The Lagrangian for the Einstein-Hilbert action is a local function of the metric tensor at a given spacetime, and it contains
the second derivatives of gµ⌫ in the Ricci scalar (first derivative in the Christoffel symbol). While the full action is a
functional of the metric tensor field gµ⌫ , the (local) Lagrangian is just a function of gµ⌫ , first derivative gµ⌫,⇢ , and second
derivative gµ⌫,⇢ . Hence, the variation with respect to gµ⌫ yields

@L @L @L
LEH = gµ⌫ + gµ⌫,⇢ + gµ⌫,⇢ + O( g 2 ) , (4.71)
@gµ⌫ @gµ⌫,⇢ @gµ⌫,⇢

and a few integrations by part then give the equation of motion

@L @ @L @2 @L
Lµ⌫
EH = + . (4.72)
@gµ⌫ @x⇢ @gµ⌫,⇢ @x⇢ @x @gµ⌫,⇢

Given the last term with four derivatives, it is surprising that the Einstein equation in the end contains only the second
derivatives of gµ⌫ . Indeed, an explicit computation yields

@L p 1 µ⇢ ⌫
= g (g g + g µ g ⌫⇢ ) g µ⌫ g ⇢ . (4.73)
@gµ⌫,⇢ 2

This feature is unique to general relativity.

Ostrogradski (1850) showed that a non-degenerate Lagrangian with finite higher-order derivative variables (than ordi-
nary one derivative) leads to a unbounded Hamiltonian, i.e., the theory includes a dynamical variable with negative kinetic
energy (or Ostrogradski ghost). The Lagrangian LEH is indeed degenerate:

@ 2 LEH
Hij := , det Hij = 0 , (4.74)
@ q̇ i @ q̇ j

or the lapse N and the shift N i are Lagrange multipliers, where i, j here indicate the dynamical variables.

39
AST511 General Relativity JAIYUL YOO

4.2.6 Palatini Method

In a curved spacetime, there are two essential ingredients that determine the inner product and the parallel transport, i.e.,
the metric tensor gµ⌫ and the affine connection ⇢µ⌫ . One can think of these two objects as two independent spacetime
fields, while they are related to each other in general relativity. In fact, the Einstein-Hilbert action can be exclusively
written in terms of these two elements:
p p
LEH = g g µ⌫ Rµ⌫ = g g µ⌫ ✏µ⌫,✏ ✏ ✏
µ✏,⌫ + µ⌫ ⇢✏
⇢ ✏ ⇢ ⇢ ⇢
µ⇢ ⌫✏ = LEH (gµ⌫ , µ⌫ , µ⌫, ) , (4.75)

arising from the fact that the Ricci tensor is just a function of ⇢µ⌫ and its first derivatives.
Treating the affine connection as an independent dynamic variable, we apply the variational method with respect
to ⇢µ⌫ and derive p p
LEH = g g µ⌫ Rµ⌫ = g g µ⌫ ⇢
µ⌫ ;⇢
⇢
µ⇢ ;⌫ , (4.76)

where we again used the Palatini identity. Mind that gµ⌫ is not varied, as it is independent from ⇢µ⌫ in this approach. An
integration by part yields
p p ⇥ ⌫ p p ⇤ ⇢
LEH = r⇢ g g µ⌫ ⇢
µ⌫ + r⌫ g g µ⌫ ⇢
µ⇢ = ⇢r g gµ r⇢ g g µ⌫ µ⌫ , (4.77)

where in the second equality we re-arranged the dummy indicies. The least action principle states that the square bracket
vanishes against the variation ⇢µ⌫ . Note, however, that since ⇢µ⌫ is symmetric over µ, ⌫, only the symmetric part of
the square bracket should vanish:
p p p
0 = ⇢⌫ r g g µ + ⇢µ r g g⌫ 2r⇢ g g µ⌫ , (4.78)

where we made the first term in LEH symmetric over µ, ⌫.

The variation with respect to the affine connection gives a differential equation for the metric tensor, and we can show
that this equation is satisfied, only if the metricity condition is satisfied
✏ ✏
0 = r⇢ gµ⌫ = @⇢ gµ⌫ ⇢µ g✏⌫ ⇢⌫ gµ✏ . (4.79)

In other words, the Palatini approach treats the affine connection as an independent variable, and the equation of motion
for the connection gives the metricity condition, which then leads to the relation to the metric tensor, i.e., Christoffel
symbol in general relativity. Note that the metricity condition in the Palatini approach is a consequence, not a choice.
It is also clear from the above derivation that in any other theories of gravity the relation between the affine connection
and the metric tensor will not be the same, when the Palatini approach is applied.

40