General Relativity From An Action: 1 The Einstein-Hilbert Action

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

General relativity from an action

1 The Einstein–Hilbert action


We will demonstrate that the vacuum Einstein equations Gab = 0 can be derived using the
Einstein–Hilbert action

Z
1
IEH = d4 x gR, (1)
16π M
where g = | det gab |, by varying with respect to the metric. Since gab g bc = δac , we have
δ(gab g bc ) = 0, so
δg ab = −g ac g bd δgcd . (2)
By diagonalizing, we see that log g = tr log g, where we have denoted the metric by g to avoid
confusion with its determinant. Upon varying, we have
δg
= δ(tr log g) = tr(δ log g) = tr(g−1 δg) = g ab δgab . (3)
g
We used the cyclic property of the trace in the third step; this can be understood in more detail
by letting g = eM and expanding the exponentials as power series to show that tr(δM) =
tr(e−M δeM ). It follows that we have the useful result

δg = gg ab δgab , (4)

from which we obtain


√ √
δ g = 12 gg ab δgab , (5)
which implies that
√ √
∂a g = gΓb ab . (6)
This last result has useful applications, for example it gives for a vector
1 √
∇a V a = ∂a V a + Γa ab V b = √ ∂a ( gV a ), (7)
g
for a scalar φ
1 √
φ := ∇a ∇a φ = √ ∂a ( gg ab ∂b φ), (8)
g
for any antisymmetric F ab
1 √
∇a F ab = ∂a F ab + Γa ac F cb = √ ∂a ( −gF ab ), (9)
g

and more generally for any totally antisymmetric F a1 a2 ...an = F [a1 a2 ...an ] , i.e. the components
of an n-form, we have
1 √
∇a1 F a1 a2 ...an = √ ∂a1 ( gF a1 a2 ...an ). (10)
g

1
Varying the Riemann tensor Ra bcd := ∂c Γa bd −∂d Γa bc +Γa ce Γe bd −Γa de Γe bc and using normal
coordinates, denoted by a star above an equals sign, we have
∗ ∗
δRa bcd = ∂c (δΓa bd ) − ∂d (δΓa bc ) = ∇c (δΓa bd ) − ∇d (δΓa bc ), (11)

and since the last expression is a tensor, we get

δRa bcd = ∇c (δΓa bd ) − ∇d (δΓa bc ). (12)

Therefore variation of the Ricci tensor gives

δRab = ∇c (δΓc ab ) − ∇b (δΓc ac ). (13)

Later we will need to use the variation of the Christoffel symbols Γa bc := 21 g ad (∂c gbd + ∂b gcd −
∂d gbc ). Since the difference of two connections is a tensor, such a quantity is tensorial, so by
using normal coordinates we obtain

δΓa bc = 21 g ad [∇b (δgcd ) + ∇c (δgbd ) − ∇d (δgbc )]. (14)

By using (1), (2), (5) and (13), we have



Z
1
δIEH = d4 xδ( gg ab Rab )
16π M

Z
1
= d4 x g[ 12 g ab Rδgab − g ac g bd Rcd δgab + ∇c (g ab δΓc ab ) − ∇b (g ab δΓc ac )]
16π M

Z
1
=− d4 x gGab δgab , (15)
16π M
after removing a total divergence in the last line. We will see later that this divergence, which
after the use of the divergence theorem can be expressed as an integral over the boundary of
the manifold, can be considered in more detail. From (5), it is straightforward to include a
cosmological constant in the Einstein–Hilbert action, which is simply altered to

Z
1
IEH = d4 x g(R − 2Λ). (16)
16π M
The other side of the Einstein equations is determined by the matter in the spacetime. If

the matter Lagrangian density is Lmatter = gLmatter , then the contribution to the action is
Z
Imatter = d4 xLmatter , (17)
M

and the energy-momentum tensor is defined as


2 δImatter
T ab := √ . (18)
g δgab
A particularly important energy-momentum tensor is that of the electromagnetic field, for
which
1 ab
LEM = − F Fab , (19)
16π
with corresponding action contribution

Z
1
IEM = − d4 x gg ce g df Fcd Fef . (20)
16π M

2
Variation of this gives

Z
1
δIEM =− d4 x g( 12 g ab g ce g df − g ac g be g df − g ce g ad g bf )Fcd Fef δgab
16π M

Z
1
=− d4 x g( 12 g ab F cd Fcd − 2F ac F b c )δgab , (21)
16π M
so the energy-momentum tensor for electromagnetism is
1
T ab = (4F ac F b c − g ab F cd Fcd ). (22)
16π
The Einstein equations Gab + Λgab = 8πTab follow by variation of

I = IEH + Imatter . (23)

2 Three dimensional hypersurfaces


Consider a spacetime manifold M with metric g, connection ∇ and curvature tensor Rabcd .
A three dimensional (or more generally codimension 1) hypersurface Σ is defined within the
spacetime by
f (xa ) = 0, (24)
for some function f . We will be interested in the curvature of Σ and in particular how it
relates to the curvature of M.
The normal to the hypersurface is in the direction of ∂a f . We will exclude the more
complicated case in which the normal is null, and instead only consider timelike and spacelike
normals. A unit normal is given by
na = N ∂a f, (25)
with the normalization factor N chosen so that

na na = ±1, (26)

where the upper + sign corresponds to a spacelike normal and the lower − sign to a timelike
normal. We now define the first fundamental form to be

hab := gab ∓ na nb . (27)

There are three important basic properties of this:

1. We have

hab hbc = (gba ∓ na nb )(gcb ∓ nb nc ) = gca ∓ 2na nc + nb nb na nc = gca ∓ na nc = hac , (28)

so h is a projection operator.

2. We have
hab nb = (g ab ∓ na nb )nb = na − na = 0. (29)

3. We have, using the previous result,

hab hab = hab gab = (g ab ∓ na nb )gab = 4 ∓ (±1) = 3. (30)

3
The last of these properties can be a useful way of remembering the sign in (27). All three of
these properties will be used repeatedly later on.
h provides a way of decomposing a vector into a piece which is normal to Σ and a piece
orthogonal to it. Explicitly, the decomposition is given by

X a = hab X b ± na (nb X b ), (31)

since the first piece is orthogonal to n so lies in Σ, and the second piece is normal to Σ. We
can extend this to tensors with any number of indices, and so can project the metric tensor
onto the hypersurface Σ by
hac hbd gab = hbd hcb = hcd . (32)
hab is often called the induced metric on Σ.
We define the second fundamental form or extrinsic curvature to be

Kab := hca hdb ∇c nd = hca gbd ∇c nd , (33)

the projection of the covariant derivative of na onto Σ. Now we have

∇c nd = ∇c (N ∇d f )
= N ∇c ∇d f + ∇c N ∇d f
∇c N
= N ∇c ∇d f + nd by (25), (34)
N
so the second fundamental form can be written as

Kab = hca hdb N ∇c ∇d f. (35)

From the absence of torsion, the connection is symmetric, so it follows that ∇c ∇d f = ∇d ∇c f ,


implying that
Kab = Kba . (36)
Since na Kab = 0, it is obvious that Kab lies entirely in Σ.
Now it is useful to decompose ∇a nb into pieces which lie in Σ and are orthogonal to Σ by

∇a n b = gac gbd ∇c nd
= (hca ± na nc )(hdb ± nb nd )∇c nd
= (hca ± na nc )hdb ∇c nd
= Kab ± na (nc ∇c nb ), (37)

using 0 = 21 ∇c (nd nd ) = nd ∇c nd in the third step.


h can be used to define a connection (3) ∇ on Σ given by
(3) 0 0 0 0 0
∇e T ab... cd... := hee haa0 hbb0 . . . hcc hdd . . . ∇e0 T a b ... c0 d0 ... , (38)

that is by projecting each index of the usual covariant derivative onto Σ. Now
(3)
∇a (3) ∇b f = hca hdb ∇c (hed ∇e f )
= hca heb ∇c ∇e f + hca hdb (∇c hed )∇e f
= hca heb ∇c ∇e f ∓ hca hdb ne (∇c nd )∇e f
= hca hdb ∇c ∇e f ∓ hca hdb ne Kcd by (37), (39)

4
(3)
The above expression is symmetric under the interchange a ↔ b, so the connection ∇ is
torsion-free. In addition, (3) ∇ is a metric connection since
(3)
∇c hab = hfc hda heb ∇f (gde ∓ nd ne ), (40)
which vanishes from ∇ being a metric connection and using (29). Thus (3) ∇ is the unique
symmetric metric connection on Σ.
The curvature of the connection (3) ∇ can be defined by the Ricci identity
((3) ∇a (3) ∇b − (3) ∇b (3) ∇a )Vc = (3) Rabc d Vd , (41)
where V is an arbitrary vector field lying in Σ, that is with
V a na = 0. (42)
(3)
We seek a relation between Rabcd and Rabcd . Now we have
(3)
Rabcd V d = hda heb hfc ∇d (hge hhf ∇g Vh ) − (a ↔ b) by (41)
= hda hgb hhc ∇d ∇g Vh + hda heb hfc ∇d (geg gfh ∓ geg nf nh ∓ ne ng gfh + ne nf ng nh )∇g Vh
−(a ↔ b)
= hda hgb hhc ∇d ∇g Vh ∓ hda hgb hfc nh ∇d nf ∇g Vh ∓ hda heb hhc ng ∇d ne ∇g Vh − (a ↔ b)
= hda hgb hhc ∇d ∇g Vh ± hda hgb hfc V h ∇d nf ∇g nh ± hda heb hhc Vh ∇d ne ∇g ng − (a ↔ b)
by (42)
= hda hgb hhc ∇d ∇g Vh ± Kac hgb V h ∇g nh ± hhc Kab Vh ∇g ng − (a ↔ b)
= (hea hfb hgc hhd Ref gh ± Kac Kbd ∓ Kbc Kad )V d (43)
by Ricci identity, (36) and (42).
Since V was an arbitrary vector field lying in Σ, we obtain Gauss’s equation
(3)
Rabcd = hea hfb hgc hhd Ref gh ± Kac Kbd ∓ Kbc Kad . (44)
We can now form the three dimensional Ricci tensor and Ricci scalar, which are
(3)
Rab = hcd(3) Rcadb
= heg hfa hhb Ref gh ± KKab ∓ K a c Kbc , (45)
and
(3)
R = hac hbd(3) Rabcd
= (g ac ∓ na nc )(g bd ∓ nb nd )Rabcd ± K 2 ∓ K ab Kab
= R ∓ 2na nb Rab ± K 2 ∓ K ab Kab , (46)
where
K := hab Kab = g ab Kab = hab ∇a nb = ∇a na . (47)
These relations tell us about the curvature of a hypersurface embedded in a general spacetime.
Because
(3)
∇c K a c = (3) ∇c (hfa hcg ∇f ng ) = hfa hcg hdc hbf hge ∇d ∇b ne = hba hde ∇d ∇b ne , (48)
and
(3)
∇a K = (3) ∇a (hfg ∇f ng ) = hfg hba hdf hge ∇b ∇d ne = hba hde ∇b ∇d ne , (49)
we have
(3)
∇b K a b − (3) ∇a K = hba hde (∇d ∇b − ∇b ∇d )ne = hba hde Rdb ec nc = hab Rbc nc . (50)
Therefore we have derived the Gauss–Codazzi equation
(3)
∇b K a b = (3) ∇a K + hab Rbc nc . (51)

5
3 The Gibbons–Hawking–York boundary term
Recall that in (15) we neglected an overall divergence in order to obtain the Einstein equa-
tions. If the spacetime has a boundary, then the neglected term gives rise to an integral over
the boundary which contributes to the Einstein–Hilbert action. The gravitational action is
incomplete as it stands and must be modified by a boundary term for the variational principle
to be well-defined.
To illustrate the point more clearly, consider the action
Z t1
I= dtL(x, ẋ), (52)
t0

so then t1
Z t1   
∂L d ∂L ∂L
δI = dt − δx + δx . (53)
t0 ∂x dt ∂ ẋ ∂ ẋ t0

Imposing δx = 0 at the endpoints t = t0 and t = t1 means that the equation of motion follows
from δI = 0.
Analogously for the graviational action we should only consider variations of the metric
which vanish on the endpoint of integration, the boundary ∂M, that is

δgab |∂M = 0. (54)

Let us now examine more closely the term we neglected,



Z
1
d4 x g∇a (g bc δΓa bc − g ab δΓc bc )
16π M
1
Z √
= d3 x hna (g bc δΓa bc − g ab δΓc bc ) by divergence theorem
16π ∂M
1
Z √
= d3 x hna {g bc g ad [∇b (δgcd ) + ∇c (δgbd ) − ∇d (δgbc )]
32π ∂M
−g ab g cd [∇b (δgcd ) + ∇c (δgbd ) − ∇d (δgbc )]} by (14)
1
Z √
= d3 x hna g bc [∇b (δgac ) − ∇a (δgbc )]
16π ∂M
1
Z √
= d3 x hna hbc [∇b (δgac ) − ∇a (δgbc )]
16π ∂M
1
Z √
=− d3 x hna hbc ∇a (δgbc ). (55)
16π ∂M

In the last step, we used hbc ∇b (δgac ) = 0, when evaluated on the boundary. This is because
δgab is constant (in fact zero) on the boundary and so its derivative must be normal to the
boundary. Regarding the indices a and b as fixed, we have the decomposition

∇a (δgab ) = hdc ∇d (δgab ) ± nc nd ∇d (δgab ), (56)

with, when evaluated on the boundary, the first term tangential and the second term normal
to the boundary, so
hdc ∇d (δgab ) = 0, (57)
and in particular
hbc ∇b (δgac ) = 0. (58)

6
However we have from (47) and varying with respect to the metric,

δK = δ(hbc ∇b nc ) = na hbc δΓc ac = 12 na hbc ∇a (δgbc ), (59)

again using hbc ∇b (δgac ) = 0. Note that the boundary ∂M, which can be expressed locally
by an equation of the form f (xa ) = 0, is fixed during the variation, i.e. it is still given by
f (xa ) = 0 when we vary the metric inside. The unit normal is
1
na = p ∂a f, (60)
|g bc (∂b f )(∂c f )|

evaluated at ∂M, and so, since the metric on ∂M is fixed, we have δna = 0 when the metric
inside is varied.
Therefore we should add to the action a Gibbons–Hawking–York boundary term, first
discovered by York [2] and used in the context of quantum gravity by Gibbons and Hawking
[1], given by
1
Z √
IGHY = d3 x hK. (61)
8π ∂M
It is often the case that the last integral diverges, in which case it is necessary to subtract
from K the trace of the second fundamental form of a background spacetime, K0 , so the
boundary term is then
1
Z √
IGHY = d3 x h(K − K0 ). (62)
8π ∂M
To summarize, the full action for general relativity is

I = IEH +
Z IGHY + Imatter
1 4 √ 1
Z
3
√ Z

= d x g(R − 2Λ) + d x h(K − K0 ) + d4 x gLmatter . (63)
16π M 8π ∂M M

References
[1] G. W. Gibbons and S. W. Hawking, “Action integrals and partition functions in quantum
gravity”, Phys. Rev. D 15, 2752 (1977).

[2] J. W. York, “Role of conformal three-geometry in the dynamics of gravitation”, Phys.


Rev. Lett. 28, 1082 (1972).

You might also like