Chapter 15

Variational Principle and Einsteins

equations
15.1 An useful formula
There exists an useful equation relating g

, g

and g = det(g

) :
g
x

= gg

. (15.1)
The proof is the following. The determinant g is a polynomial in g

g = g(g

) . (15.2)
Let us rst compute the derivatives
g
g

. (15.3)
We remind that g can be expressed by xing one row, i.e. , and computing the expansion
g =

(1)
+
(no sum over ) (15.4)
where M

is the minor , , i.e. the determinant of the matrix obtained by cutting the row
and the column from the matrix g

.
From (15.4) we nd that
g
g

= (1)
+
M

(15.5)
where, again, we are not summing on the indices , .
Furthermore, the components of g

, the matrix inverse of (g

), are given by
g

=
1
g
M

(1)
+
. (15.6)
Therefore,
g
g

= gg

, (15.7)
208
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 209
then
g
x

=
g
g

= gg

(15.8)
and (15.1) is proved.
The same rule applies to variations:
g =
g
g

(15.9)
thus
g = gg

. (15.10)
Furthermore, since
(g

) = (

) = 0
= g

+ g

(15.11)
multiplying with g

, we nd
g

= g

. (15.12)
Therefore, equation (15.10) gives
g = gg

. (15.13)
15.2 Gauss Theorem in curved space
First of all we give a preliminary denition.
- Given a manifold M described by coordinates {x

}, and a metric g

on M.
- Given a submanifold N M described by coordinates {y
i
}, such that on N x

= x

(y
i
).
We dene the metric induced on N from M as

ij

x

y
i
x

y
j
g

. (15.14)
We can now generalize Gauss theorem to curved space:
- Be an n-dimensional volume described by coordinates {x

}
=0,...,n1
, and g

the metric
on .
- Be the boundary of , described by coordinates {y
j
}
j=0,...,n2
with normal vector n

(having |n

| = 1 for a timelike or spacelike surface); be

ij
the metric induced on from
g

.
- Given a vector eld V

dened in , then
_

d
4
x

g V

;
=
_

d
3
y

V

n

. (15.15)
If we dene the surface integration element as
dS

d
3
y , (15.16)
the Gauss theorem can also be written as
_

d
4
x

g V

;
=
_

V

dS

. (15.17)
In particular, if one considers an innite volume, and if V

vanishes asymptotically, then
the integral of its covariant divergence is zero.
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 210
15.3 Dieomorphisms on a manifold and Lie derivative
Given an n-dimensional dierentiable manifold M, let us consider a mapping of the man-
ifold M into itself
: MM, (15.18)
in which we map any point P M to another point Q of the same manifold M:
P (P) = Q. (15.19)
If such a map is invertible and regular, as we will always assume in the following, the mapping
is named dieomorphism.
In a local coordinate frame that, by denition, is a mapping of an open subset of M to
IR
n
(cfr. Chapter 2 section 2.5)
P {x

(P)}
=1,...,n
(15.20)
a dieomorphism can be expressed as a set of n real functions on IR
n
, {

}, that acting
on the coordinates of the point P, {x

}, produce the coordinates of the point Q, {x

} in
the same coordinate frame i.e.
: x

(x) . (15.21)
The functions

(x

) are smooth and invertible.

Notice that the transformation (15.21) has the same form of a general coordinate trans-
formation, but in the present context its meaning is very dierent: in a coordinate transfor-
mation we assign to the same point P of the manifold two dierent n-ples of IR
n
; thus, in
this case the mapping denotes, as shown in Fig. 15.1, the set of functions which allows to
transform from the coordinates {x

} to the coordinates {x

}, i.e. it is a mapping of IR
n
to
IR
n
.
M
P
x
x

IR
n
IR
n
Figure 15.1: A coordinate transformation: and are the maps from M to IR
n
that associate
dierents sets of coordinates (i.e. dierent n-ples of IR
n
) to the same point P.
Conversely, given a dieomorphism which maps P to Q, where P and Q are two
dierent points of the same manifold M, we use the same coordinate frame to label both
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 211
P and Q as shown in Fig. 15.2. To hereafter we will use the following convention: when
dealing with a coordinate transformation we shall write, as usual
: x

(x) , (15.22)
while,a dieomorphism : MM expressed in a coordinate frame will be indicated as
: x

(x) . (15.23)
i.e. x

indicate the coordinates of the point Q.

M
P
Q
x
x

R I
n

Figure 15.2: A dieomorphism on the manifold, and its representation in a coordinate

frame, . The map of M to the coordinate frame is denoted by .
15.3.1 The action of dieomorphisms on functions and tensors
Be f a real function dened on a manifold M. A dieomorphism changes f into a dierent
function, which is called the pull-back of f, and is usually denoted by the symbol

f:
f

f (15.24)
such that
(

f)(P) = f(Q) (

f)(P) = f( (P)) (15.25)

i.e.

f f . (15.26)
In a coordinate frame, where P has coordinates x

and Q has coordinates x

, eq. (15.25)
takes the form
(

f)(x) = f(x

) (

f)(x) = f((x)) . (15.27)

The function

f(P), which is equal to f(Q), will be dierent from f(P); the dierence
is
f(P) = (

f)(P) f(P) = f(Q) f(P) = f( (P)) f(P) (15.28)

or, in a coordinate frame,
f(x) = (

f)(x) f(x) = f((x)) f(x) . (15.29)

CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 212
M
P
Q
f
f(P)
R I
f(Q)=
f
f
*

*
f(P)
Figure 15.3: The mapping f from the manifold M to IR, and its pull-back

f.
In a similar way, the dieomorphism also induces a change on tensors: due to the
action of , a tensor eld T changes to a new tensor eld called the pull-back of T, denoted
by

T, such that
(

T)(P) = T(Q) . (15.30)

The dierence between the pull-back and the original tensor, evaluated in a given point P
of the manifold, is
T(P) = (

T)(P) T(P) = T( (P)) T(P) . (15.31)

An important point to stress
As discussed in Chapter 3, a tensor eld depends on the point of the manifold, but also on
vectors and one-forms dened in the tangent space (and its dual) at that point of the manifold.
In eq. (15.30) we have explicitly written the dependence on the point of the manifold, but
we have left implicit the dependence on vectors and one-forms. This dependence can be
made explicit as follows:
T acts on vectors and one-forms dened in the tangent space (and in its dual) in Q, while

T acts on vectors and one-forms which are dened in the tangent space (and in its dual)
in P. The complete denition of the pull-back

T therefore is
_
(

T)(

V , . . . ,

q, . . .)
_
(P) =
_
T(

V , . . . , q, . . .)
_
(Q) . (15.32)
In this denition there are the pull-backs of vectors and one-forms,

V and

q, which we
still have to dene.
To this purpose, we remind that vectors are in one-to-one correspondence with directional
derivatives (see Chapter 3); indeed a vector

V {
dx

d
} tangent to a given curve with
parameter , associates to any function f the directional derivative

V (f)
df
d
=
f
x

dx

d
= V

f
x

.
Thus, we dene the pull-back of a vector

V as follows: for any function f

_
(

V )(

f)
_
(P) =
_

V (f)
_
(Q) . (15.33)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 213
Let us now consider one-forms. By denition a one-form q associates to any vector

V a real
number q(

V ).
The pull-back of a one-form,

q is dened as follows: for any vector

V ,
_
(

q)(

V )
_
(P) =
_
q(

V )
_
(Q) . (15.34)
In particular, the pull-backs of the coordinate basis vectors and of the coordinate basis
one-forms are dened as follows: the pull-back maps the coordinate basis (of vectors or
one-forms) in Q in the coordinate basis in P:

=
x

(15.35)

dx

=

dx

=
x

dx

(15.36)
where
x

, (15.37)
and

(x) are the functions dened in eq. (15.21), whereas x

/x

are the derivatives of

the inverse functions.
15.3.2 The action of dieomorphisms on the metric tensor
The pull-back of the metric tensor is, by denition,
(

g)(P) = g(Q). (15.38)

Let us write eq. (15.38) in components. The tensor

g, evaluated in P, must be expanded

in terms of the coordinate basis of one-forms in P, i.e.

dx

, while the tensor g must be

expanded in terms of the coordinate basis of one-forms in Q, which is

dx

, therefore
(

g)

(x)

dx

dx

= g

(x

)

dx

dx

(15.39)
and using eq. (15.36) we nd
(

g)

(x) =
x

((x)). (15.40)
The change induced by on the metric tensor therefore is
g

(x) = (

g)

(x) g

(x) =
x

((x)) g

(x) . (15.41)
15.3.3 Lie derivative
Let us consider an innitesimal dieomorphism , which in a coordinate frame is

(x) = x

(x), (15.42)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 214
where

is a vector eld and is a small parameter. To hereafter we shall neglect O(

2
)
terms. From eq. (15.42) it follows that
x

= x

. (15.43)
According to eq. (15.29) the change induced by on a function f is
f(x) = f(x + ) f(x) =

f
x

. (15.44)
We dene the Lie derivative of a scalar function f with respect to the innitesimal dieo-
morphism associated to the vector eld

, as
L

(f) lim
0

f(P) f(P)

= lim
0
f( (P)) f(P)

, (15.45)
or, in coordinates,
L

(f) lim
0
f(x + ) f(x)

f
x

. (15.46)
Therefore, eq. (15.44) can also be written as
f = L

f . (15.47)
In a similar way we can dene the Lie derivative of a tensor eld T with respect to the
vector eld

as
L

T lim
0

T(P) T(P)

= lim
0
T( (P)) T(P)

. (15.48)
Therefore, from eq. (15.31) we see that the change induced on a tensor by an innitesimal
dieomorphism is given by the Lie derivative:
T = L

T. (15.49)
In the case of the metric tensor, since from eq.(15.42)
x

, (15.50)
equation (15.41) gives
g

(x) =
_

__

_
g

(x + ) g

(x)
=
_

_
, (15.51)
therefore
L

. (15.52)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 215
This expression has already been derived in Chapter 7 when we studied how the metric
tensor changes under innitesimal translations along a given curve of tangent vector

(x)d (15.53)
Indeed, an innitesimal translation is an innitesimal dieomorphism; thus the derivation of
eqs. (15.51) and (15.52) repeats, in a more rigorous way, the derivation presented in Chapter
8.4. Furthermore, as shown in Chapter 8.4, the expression (15.52) can be rewritten in terms
of covariant derivatives of

=
;
+
;
. (15.54)
15.3.4 General Covariance and the role of dieomorphisms
Since the equation
: x

(x) (15.55)
allows a double interpretation as a general coordinate transformation and as a dieomor-
phism on the spacetime manifold, it follows that to any coordinate transformation we can
associate a dieomorphism, and viceversa.
The principle of general covariance states that since the laws of physics are expressed by
tensorial equations, they retain the same form in any coordinate frame, i.e. they are invariant
under a general coordinate transformation; since a coordinate transformation corresponds,
in the sense explained above, to a dieomorphism, we can restate the principle of general
covariance as follows: all physical laws are invariant for dieomorphisms, i.e. we
restate general covariance as symmetry principle dened on the manifold.
15.4 Variational approach to General Relativity
In the variational approach to eld theory, the dynamics of elds is described by the action
functional.
15.4.1 Action principle in special relativity
Let us consider a collection of tensor (and, eventually, spinor) elds in special relativity
_

(A)
(x)
_
A=1,...
, (15.56)
where x denotes the point of coordinates {x

}. We shall use symbols in boldface to denote

a generic tensorial or spinorial object. For instance, for a vector eld we shall write
V = (V

), (15.57)
and summation over the tensor indices will be left implicit:
L
V
V
L
V

V

. (15.58)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 216
It should be noted that here we use a convention dierent from that used for vector elds in
Chapter 2.6.
Typically, the elds involved are scalars, rank-one tensors (vectors and one-forms) and
spinors. The action is a functional of these elds and of their rst derivatives, written as an
integral of a Lagrangian density over the 4-dimensional volume:
I =
_
d
4
x L
_

(1)
(x), . . . ,
(A)
(x), . . . ,

(1)
(x), . . . ,

(A)
(x), . . .
_
. (15.59)
To hereafter

. All elds are assumed to vanish on the boundary of the integration

volume or asymptotically, if the volume is innite.
Let us consider the variation of the action with respect to a given eld
(A)
I =
_
d
4
x

A
L

(A)

(A)
=
_
d
4
x

A
_
L

(A)

(A)
+
L
(

(A)
)

(A)
_
=
_
d
4
x

A
_
L

(A)

(A)
+
L
(

(A)
)

(A)
_
=
_
d
4
x

A
_
L

(A)

L
(

(A)
)
_

(A)
.
(15.60)
Here we have used the general property that the operations of variation and dierentiation
commute, and then we have integrated by parts. The stationarity of I with respect to the
considered eld gives the equation of motion for that eld
I = 0,
(A)
,
and since the integral (15.60) has to vanish for every
(A)
(x), it follows that
L

(A)

L
(

(A)
)
= 0 (15.61)
which are the Euler-Lagrange equations for the eld
(A)
.
15.4.2 Action principle in general relativity
In general relativity, besides the elds
_

(A)
_
, which are the matter and gauge elds, there
is the metric eld
g(x) = (g

(x)) (15.62)
which describes the gravitational eld whose action is the Einstein-Hilbert action
I
EH
=
c
3
16G
_
d
4
x

g R. (15.63)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 217
Due to the strong equivalence principle, in a locally inertial frame the dynamics of all elds
_

(A)
_
except gravity is described by the action (15.59). Therefore, according to the prin-
cipal of general covariance in a general frame the action, which is a scalar, retains the same
form provided

, the partial derivatives

are replaced by covariant derivatives

, and the integration volume element d

4
x is replaced by the covariant volume element

gd
4
x.
With these replacements, we shall now show that the results of the previous section (in
particular, the derivation of Euler-Lagrange equations) remain valid.
The total action is
I = I
EH
+ I
FIELDS
(15.64)
with
I
FIELDS
=
_
d
4
x

gL
FIELDS
_

(1)
(x), , . . . ,
(A)
(x), . . . ,

(1)
(x), . . . ,

(A)
(x), . . . , g
_
.
(15.65)
Notice that now the Lagrangian density L
FIELDS
depends explicitely on g because we have
replaced

by g

and

by

.
As in special relativity, the equations for a eld
(A)
are found by varying the action
with respect to that eld, and since the Einstein-Hilbert action does not depend on
I I
FIELDS
=
_
d
4
x

A
L
FIELDS

(A)

(A)
=
_
d
4
x

A
_
L

(A)

(A)
+
L
(

(A)
)

(A)
_
=
_
d
4
x

A
_
L

(A)

(A)
+
L
(

(r)
)

(A)
_
=
_
d
4
x

A
_
L

(A)

L
(

(A)
)
_

(A)
= 0,
(A)
(15.66)
where we have used the property

. To obtain the last row of eq. (15.66) we

have integrated by parts using the generalization of Gauss theorem in curved space (see
Section 15.2) which assures that the integral of a covariant divergence is zero, provided all
elds vanish at the boundary, or asymptotically if the volume is innite, i.e.
_
d
4
x

g V

;
= 0 . (15.67)
Thus, the equations of motion for the eld
(A)
are the Euler-Lagrange equations generalized
in curved space:
L

(A)

L
(

(A)
)
= 0 . (15.68)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 218
The eld equations for the gravitational eld are obtained by varying the action (15.64) with
respect to g. In section 15.4.3 we shall show that the variation of the Einstein-Hilbert action
with respect to g gives the Einstein tensor; here we just write the result:
I
EH
=
c
3
16G
_
d
4
x (

gR) =
c
3
16G
_
d
4
x

g
_
R

1
2
g

R
_
g

. (15.69)
The variation of I
FIELDS
with respect to g it is easy to nd if we use the property (15.13),
from which
(g)
1/2
=
1
2
(g)
1/2
g

, (15.70)
thus
I
FIELDS
=
_
d
4
x

g
_
L
FIELDS
g

1
2
L
FIELDS
g

_
g

. (15.71)
Combining eqs. (15.69) and (15.71), and dening the stress-energy tensor as
T

2c
_
L
FIELDS
g

1
2
L
FIELDS
g

_
, (15.72)
the variation of the total action (15.64) can be written as
I =
c
3
16G
_
d
4
x

g
_
G

8G
c
4
T

_
g

= 0 . (15.73)
Thus, with this denition of T

the action principle gives Einsteins equations

G

=
8G
c
4
T

. (15.74)
The advantage of deriving eld equations using a variational approach is that it makes
explicit the connection between symmetries and conservation laws: any symmetry of a theory
corresponds to a conservation law.
In Section 15.3 we have shown that the principle of general covariance implies that
equations expressing the laws of physics are invariant for dieomorphisms. If they can be
derived from an Action Principle, this is equivalent to impose that the action is invariant
for dieomorphisms.
As an example, let us consider the action I
FIELDS
. It is invariant for dieomorphisms,
because the equations of motion (15.68) for the elds
(A)
, arising from I
FIELDS
= 0, are
dieomorphism invariant. We will now show that the dieomorphism invariance of I
FIELDS
implies the divergenceless equation T

;
= 0, which generalizes the conservation law T

,
= 0
of at spacetime.
Let us consider an innitesimal dieomorphism (see eq. 15.42)
x

= x

. (15.75)
In Section 15.3 we have shown how tensor elds change under innitesimal dieomorphisms,
and in particular that the change of the metric tensor, written in components, is (eq. 15.54):
g

= [
;
+
;
] . (15.76)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 219
The change in g

can be derived by using the property (15.12):

g

= [
;
+
;
] . (15.77)
The dieomorphism (15.75) will produce a variation of the elds
(A)
s, and a variation of
the metric tensor given by (15.77); consequently I
FIELDS
, dened in eq. (15.65), will vary
as follows
I
FIELDS
=
_
d
4
x

g
_
L
FIELDS
g

1
2
L
FIELDS
g

_
g

+
_
d
4
x

A
L
FIELDS

(A)

(A)
. (15.78)
Since the action I
FIELDS
is invariant under dieomorphisms, I
FIELDS
must vanish. The
term in square brackets is the stress-energy tensor T

dened in eq. (15.72). Therefore,

using eq. (15.77), we nd
I
FIELDS
=

2c
_
d
4
x

gT

2
;
+
_
d
4
x

A
L
FIELDS

(A)

(A)
= 0
(15.79)
where we have used: T

(
;
+
;
) = 2T

;
, which follows from the symmetry of T

.
Since the elds
(A)
s satisfy their equations of motion (15.68) the last integral in eq.
(15.79) vanishes. Thus
I
FIELDS
=

c
_
d
4
x

g T

;
=

c
_
d
4
x

g T

= 0

(15.80)
and consequently
T

;
= 0 . (15.81)
We stress that eq. (15.81) is not satised for all eld congurations, but only for the eld
congurations which are solutions of the eld equations, i.e. of the Euler-Lagrange equations
(15.68).
The dieomorphism invariance of the Einstein-Hilbert action I
EH
, instead, implies the
Bianchi identities. Indeed, the variation of I
EH
is given by (15.69)
I
EH
=
c
3
16G
_
d
4
x

g G

(15.82)
and if the variation is due to an innitesimal dieomorphism, from (15.77)
g

= [
;
+
;
] (15.83)
thus, integrating by parts,
I
EH
=
c
3
16G
_
d
4
x

g G

2
;
=
c
3
8G
_
d
4
x

g G
;

= 0

(15.84)
and consequently we nd the Bianchi identities
G

;
= 0 . (15.85)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 220
15.4.3 Variation of the Einstein-Hilbert action
In this section we shall prove that the variation of the Einstein-Hilbert action with respect
to the metric tensor gives the Einstein tensor (equation (15.69)):
_
(

gR) =
_

g
_
R

1
2
g

R
_
g

. (15.86)
As a rst step we derive the Palatini identity:
R

= (

)
;
(

)
;
. (15.87)
By varying the Ricci tensor:
R

,
+

, (15.88)
we nd
R

,
+

. (15.89)
To evaluate this expression, we need to compute

. To this purpose, we will use the

property (15.12)
g

= g

. (15.90)
If we dene

=
1
2
(g
,
+ g
,
g
,
) (15.91)
we can write the variation of Christoels symbols as follows

=
_
g

_
= g

+ g

= g

+ g

= g

+ g

1
2
[g
,
+ g
,
g
,
]
=
1
2
g

_
g
,
+ g
,
g
,
2

_
(15.92)
which can be recast in the form

=
1
2
g

__
g
,

_
+
_
g
,

_
g
,

__
=
1
2
g

[g
;
+ g
;
g
;
] . (15.93)
Eq. (15.93) shows that since g

is a tensor,

is also a tensor. Therefore, the expression

(15.87),
(

)
;
(

)
;
(15.94)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 221
can be evaluated with the usual rules of covariant dierentiation of tensors. Thus we nd
(

)
;
(

)
;
=

,
+

. (15.95)
A comparison of this equation with eq. (15.89)) shows that
R

= (

)
;
(

)
;
.
QED.
Let us now go back to the variation of the Einstein-Hilbert action
(

gR) = (

gg

) . (15.96)
Using eq.(15.70) we nd
(

gR) =

g
_
g

+ g

1
2
g

R
_
. (15.97)
Using the fact that g
;
= 0, the rst term in (15.97) becomes
g

= g

_
(

)
;
(

)
;
_
= (g

)
;
(g

)
;
=
_
g

_
;
(15.98)
which is the divergence of a vector; therefore by Gauss theorem such term vanishes when
integrated over the 4-volume

gd
4
x. The remaining two terms in (15.97) give the Einstein
tensor, therefore we can write
(

gR) =

g
_
R

1
2
g

R
_
g

+ surface terms (15.99)

which shows that eq. (15.86) is true.
15.4.4 An example: the electromagnetic eld
The Lagrangian density of the electromagnetic eld is
L =
1
4c
F

(15.100)
with
F

= A
;
A
;
(15.101)
(the last equality arises from the symmetry property of the Christoel symbols

).
The eld equations for the electromagnetic eld A

are
c
L
A

=
1
2
F

=
1
2
F

(A
;
A
;
)
= F

A
;
= F

;
+ surface terms ; (15.102)
CHAPTER 15. VARIATIONAL PRINCIPLE AND EINSTEINS EQUATIONS 222
as usual, we eliminate the surface terms on the assumption that the elds vanish at the
boundary of a given volume, or at innity. The eld equations then are
F

;
= 0 , (15.103)
and the stress-energy tensor is
T

= 2c
_
L
g

1
2
Lg

_
= 2
_

1
2
F

+
1
8
g

_
= F

1
4
g

. (15.104)
15.4.5 A comment on the denition of the stress-energy tensor
In special relativity, tipically one denes the canonical stress-energy tensor as
can
T

= c
_
L
(

L
_
. (15.105)
In principle, we can generalize this denition in curved space by appealing to the principle
of general covariance
can
T

= c
_
L
(

L
_
. (15.106)
However, the tensor (15.106) is not a good stress-energy tensor. Indeed, in general it is not
symmetric, and it does not satisfy the divergenceless condition
can
T

;
= 0 . (15.107)
The correct denition for the stress-energy tensor in general relativity is given by eq. (15.72)
T

= 2c
_
L
g

1
2
g

L
_
. (15.108)
Anyway, it can be shown that the dierence between (15.108) and the canonical stress-
energy tensor (15.106) is a total derivative, which disappears when integrated in the overall
spacetime; furthermore, it can be shown that in the Minkowskian limit, where g

and

the two tensors (15.106), (15.108) coincide.