One-Loop Divergencies in The Theory of Gravitation: G. 'T HOOFT ( ) and M. VELTMAN ( )

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Ann Inst Henn Pomcaré, Section A

Vol XX, n° 1, 1974, p 69 94 Physique theonque

One-loop divergencies
in the theory of gravitation
par

G. 't HOOFT (*) and M. VELTMAN (*)


C E R N , Geneva

ABSTRACT. — All one-loop divergencies of pure gravity and all those


of gravitation interacting with a scalar particle are calculated. In the case
of pure gravity, no physically relevant divergencies remain; they can all
be absorbed in a field renormalization. In case of gravitation interacting
with scalar particles, divergencies in physical quantities remain, even
when employing the socalled improved energy-momentum tensor.

1. INTRODUCTION

The recent advances in the understanding of gauge theories make a


fresh approach to the quantum theory of gravitation possible. First, we
now know precisely how to obtain Feynman rules for a gauge theory [1];
secondly, the dimensional regularization scheme provides a powerful tool
to handle divergencies [2]. In fact, several authors have already published
work using these methods [3], [4].
One may ask why one would be interested in quantum gravity. The
foremost reason is that gravitation undeniably exists; but in addition
we may hope that study of this gauge theory, apparantly realized in nature,
gives insight that can be useful in other areas of field theory. Of course,
one may entertain all kinds of speculative ideas about the role of gravi-
tation in elementary particle physics, and several authors have amused
themselves imagining elementary particles as little black holes etc. It
may well be true that gravitation functions as a cut-off for other interac-
tions; in view of the fact that it seems possible to formulate all known

(*) On leave from the University of Utrecht, Netherlands


70 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

interactions in terms of field-theoretical models that show only logarith-


mic divergencies, the smallness of the gravitational coupling constant need
not be an obstacle. For the time being no reasonable or convincing ana-
lysis of this type of possibilities has been presented, and in this paper we
have no ambitions in that direction. Mainly, we consider the present
work as a kind of finger exercise without really any further underlying
motive.
Our starting point is the linearized theory ofgravitation. Ofcourse, much
work has been reported already in the literature [5], in particular we men-
tion the work of B. S. Dewitt [6]. For the sake of clarity and completeness
we will rederive several equations that can be found in his work. It may
be noted that he also arrives at the conclusion that for pure gravitation
the counterterms for one closed loop are of the form R 2 or this
really follows from invariance considerations and an identity derived by
him. This latter identity is demonstrated in a somewhat easier way in appen-
dix B of this paper.
Within the formalism of gauge theory developed in ref. 7, we must first
establish a gauge that shows clearly the unitarity of the theory. This is
done in section 2. The work of ref. 7, that on purpose has been formulated
such as to encompass quantum gravity, assures us that the S-matrix remains
invariant under a change of gauge.
In section 3 we consider the one loop divergencies when the gravitational
field is treated as an external field. This calculation necessitates a slight
generalization of the algorithms recently reported by one of us [8]. From
the result one may read off the known fact that there are fewer divergencies
if one employs the so-called improved energy-momentum tensor [9].
Symanzik's criticism [10] applies to higher order results, see ref. 11. In the
one loop approximation we indeed find the results of Callan et al. [9].
Next we consider the quantum theory of gravity using the method of
the background field [6], [12]. In ref. 8 it has already been shown how this
method can be used fruitfully within this context. In sections 4, 5 and 6
we apply this to the case of gravitation interacting with a scalar field in
the conventional way. The counter Lagrangian for pure gravity can be
deduced immediately.
In section 7 finally we consider the use of the « improved » energy-
momentum tensor. Some results are quoted, the full answer being
unprintable.
Appendix A quotes notations and conventions. Appendix B gives the
derivation of a well-known [6] but for us very important result. An (also
well-known) side results is the fact that the Einstein-Hilbert Lagrangian
is meaningless in two dimensions. This shows up in the form of fac-
tors 1/(n — 2) in the graviton propagator, as noted by Neveu [13] and
Capper et al. [4].
Annales de l'Institut Henri Poincare-Section A
G 'T HOOFT AND M VELTMAN 71

2. UNITARITY

One of the remarkable aspects of gravitation is the freedom of choice


in the fundamental fields. In conventional renormalizable field theory
the choice of the fields is usually such as to produce the smoothest possible
Green's functions. In the case of quantum gravity there seems to be no
clear choice based on such a criterium. For instance, one may use as basic
field the metric tensor or its inverse or any other function of the
involving, say, the Riemann tensor. From the point of view of the S-matrix
many choices give the same result, the Jacobian of the transformation
being one (provided dimensional regularization is applied).
We chose as basic fields the related to the by
This is of course the conventional choice. The Lagrangian that we start
from is the Einstein-Hilbert Lagrangian, viz:
(2.1)

(see appendix A for symbols and notations). This Lagrangian is invariant


under the infinitesimal gauge transformation (or rather it changes by a
total derivative)
(2.2)
or
(2.3)

In here the are four independent infinitesimal functions of space-time.


The are the usual covariant derivatives.
In order to define Feynman rules we must supplement the Lagran-
gian (2.1) with a gauge breaking term and a Faddeev-Popov
ghost Lagrangian (historically, the name Feynman-DeWitt ghost Lagran-
gian would be more correct) In order to check unitarity and positivity of
the theory we first consider the (non-covariant) Prentki gauge which is
much like the Coulomb gauge in quantum-electrodynamics:

(2.4)

In the language of ref. 7 we take correspondingly:

(2.5)

Vol XX, n° 1 - 1974


72 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

With this choice for C the part quadratic in the is (comma denotes
differentiation):

(2.6)

This can be written as The Fourier transform of V is:

(2.7)

Calculations in the theory of gravitation are as a rule cumbersome, and


the calculation of the graviton propagator gives a foretaste of what is to
come. In principle things may be done as follows. First, symmetrize V with
respect to and interchange. Then find the propa-
gator P from the equation where

(2.8)

Subsequently the limit must be taken. It is of advantage to go


in the coordinate system where Alternatively, write

Then V can be rewritten as a rather simple 10 x 10


matrix, that subsequently must be symmetrized. In the limit
one may in a row or column containing a neglect all elements, except
ofcourse the term itself. Inversion ofthat matrix is trivial, and providing
a minus sign the result is the propagator in the Prentki gauge:

(2.9)

In here n is the dimensionality of space-time. Further

(2.10)
The first part of eq. (2.9) corresponds in 4 dimensions (i. e. n = 4) to
the propagation of two polarization states of a mass zero spin 2 particle
(see ref. 14, in particular section 3). The second and third part have no pole,

Annales de l'Insititut Henri Pomcare-Section A


G 'T HOOFT AND M VELTMAN 73

they are non-local in space but simultaneous (or « local ») in time. They
describe the l/r behaviour of the potential in this gauge. In any case, they
do not contribute to the absorptive part of the S-matrix.
In addition to the above we must also consider the Faddeev-Popov
ghost. Subjecting of eq. (2.5) to the gauge transformation (2.3) and
working in zero'th order of the field we obtain the quadratic part of
the F-P ghost Lagrangian:
(2.11)
where the arrow indicates the 3-dimensional derivative. The propagator
resulting from this has no pole, therefore does not contribute to the absorp-
tive part of the S-matrix.
The above may be formulated in a somewhat neater way by means
of the introduction of a fixed vector with zero space components. For
instance:
(2.12)

The continuous dimension regularization method can now be applied


without further difficulty.

3. EXTERNAL GRAVITATIONAL FIELD

In this section we assume that the reader is acquainted with the work
ofref. 8. The principal result is the following. Let there be given a Lagrangian

(3.1)

where N and M are functions of external fields etc., but do not depend
on the quantum fields The counter-Lagrangian that eliminates
all one loop divergencies is

(3.2)

The trace is with respect to the indices i,j, ... Further

(3.3)

It is assumed that is antisymmetric and M symmetric in the indices i,j.

Vol XX, n° 1 - 1974


74 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

For our purposes it is somewhat easier to work with complex fields.


Given the Lagrangian
(3.4)
one easily derives

(3.5)

Writing with real fields A and B, it is seen that (3.2)


is contained twice in (3.5) provided and are symmetric
and antisymmetric respectively. Eq. (3.5) is valid independently of these
symmetry properties.
There is now a little theorem that says that the counter-Lagrangian
remains unchanged if in the original Lagrangian, eq. (3.4), everywhere
is replaced by where Z is a (possibly space-time dependent) matrix.
To see that we consider the following simplified case:
(3.6)
The Feynman rules are
propagator,

Z-vertex,

vertex.

Any vertex may be followed by 0, 1, 2, . . . vertices of the Z-type. Now


it is seen that the Z-vertex contains a factor that precisely cancels the pro-
pagator attached at one side. So one obtains a geometric series of the form

Clearly, the results are identical in case we had started with the Lagrangian

(3.7)

which is related to the above Lagrangian by the replacement

Annales de l'Institut Henri Poincare - Section A


G 'T HOOFT AND M VELTMAN 75

In the following we need the generalization of the above to the case that
the scalar product contains the metric tensor This tensor, in the appli-
cations to come, is a function of space-time, but not of the fields Thus
we are interested in the Lagrangian:

(3.8)

This Lagrangian is invariant under general coordinate transformations,


and so will be our counter-Lagrangian. This counter-Lagrangian is given
in eq. (3.35), and we will sketch the derivation.
Taking into account that will contain terms of a certain dimen-
sionality only, we find as most general form:

(3.9)
The term need not be considered; see Appendix B. As usual,
etc. Several coefficients can readily be determined by compa-
rison with the special case

(3.10)

The remaining coefficients are determined in two steps. First, we take the
special case
(3.11)

In here is an arbitrary function of space-time. The Lagrangian (3.8)


becomes:
(3.12)

The replacement leaves the counter-Lagrangian unchanged;


we have thus the equivalent Lagrangian

(3.13)

Vol XX, n° 1 - 1974


76 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

This is precisely of the form studied before; the answer is as given in


eq. (3.5), but now with the replacements

(3.14)

To proceed, it is necessary to compute a number of quantities for this


choice of Using the notations

we have (on the right hand side there is no difference between upper and
lower indices):
(3.15)

For any contravariant vector

(3.16)

(3.17)

(3.18)

(3.19)

(3.20)
We leave it to the reader to verify that
= total derivative. (3.21)
In this particular case, unfortunately, also another identity holds:

= total derivative. (3.22)

Annales de l'Institut Henri Poincaré-Section A


G 'T HOOFT AND M VELTMAN 77

We must now try to write the counter-Lagrangian in terms of covariant


objects. First:

(3.23)
Similarly:

(3.24)
This equation looks more complicated then it is; one has

Now is symmetrical in the two lower indices, and therefore

The various F dependent terms all cancel out.


The result for the special case is:

(3.25)

Inspecting the general form eq. (3.9) and remembering the identity (3.22)
we see that we have determined up to a term

(3.26)

To determine the coefficient we consider the special case


(3.27)
With we need to expand up to first order in h

(3.28)

Vol XX, nº 1 - 1974


78 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

With we need only compute the selfenergy type of


graph with two s vertices. Note that the terms of second order in h
in the Lagrangian do not contribute because they give rise to a vertex
with two h, and they can only contribute to in order by closing
that vertex into itself:

These tadpole type diagrams give integrals of the form

and these are zero in the continuous dimension method.


The computation of the pole part of the graph

is not particularly difficult. The result is:

(3.29)

Working to second order in h one has

(3.30)

(3.31)
(3-32)

(3 • 33)
The result is:
(3.34)

The first term has been found before (see eq. 3.25).

Annales de l'Institut Henri Poincare-Section A


G T HOOFT AND M VELTMAN 79

All coefficients have now been determined and we can write the final
result. To the Lagrangian (3.8) corresponds the counter-Lagrangian:

(3.35)

Note that a trace is to be taken; the last term has as factor the unit matrix.
As before
(3.36)
To obtain the result for real fields, write
(3.37)
and substitute
(3.38)

The result is then as follows. To the Lagrangian

(3.39)

corresponds the counter-Lagrangian

(3.40)

Eq. (3.40) contains a well known result. The gravitational field enters
through and the Lagrangian (3.39) describes the interaction of bosons
with gravitation, whereby gravity is treated in the tree approximation.
If one adds now to the Lagrangian (3.39) the term

then the « unrenormalizable » counterterms of the form MR and


disappear. Indeed, the energy-momentum tensor of the expression

6
80 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

is precisely the « improved » energy-momentum tensor of ref. 9. We see


also that closed loops of bosons introduce nasty divergencies quadratic
in the Riemann tensor. This unpleasant fact remains if we allow also for
closed loops of gravitons. This is the subject of the following sections.

4. CLOSED LOOPS INCLUDING GRAVITONS

We now undertake the rather formidable task of computing the diver-


gencies of one loop graphs including gravitons. The starting point is the
Lagrangian
(4.1)

is the Riemann scalar constructed from In section 7 we will include


other terms, such as
Again using the background field method [6], [72] we write

(4.2)
If we take the c-number quantities and such that they obey the clas-
sical equations of motion then the part of linear in the quantum fields
and is zero. The part quadratic in these quantities determines the one
loop diagrams. We have:
(4.3)

and are linear and quatratic in the quantum fields and h respec-
tively. The higher order terms contained in play a role only in mul-
tiloop diagrams.
At this point we may perhaps clarify our notations. In the following
we will meet quantities like the Riemann tensor This is then the tensor
made up from the classical field In the end we will use the classical
equations of motion for this tensor. All divergencies that are physically
irrelevant will then disappear. In fact, using these classical equations of
motion is like putting the external lines of the one loop diagrams on mass-
shell, with physical polarizations. Note that we still allow for trees connected
to the loop. Only the very last branches of the trees must be physical.
To obtain and we must expand the various quantities in eq. (4.1)
up to second order in the quantum fields. We list here a number ofsubresults.
Note that
(4.4)
Thus indices are raised and lowered by means of the classical field
G 'T HOOFT AND M VELTMAN 81

The following equations hold up to terms of third and higher order in h


and
(4.3)
(4.4)
Using

we find
(4.5)
Further
(4.6)

(4.7)

(4.8)

(4.9)

We used here the fact that the co- or contra-variant derivative of is zero;
therefore
(4.10)

Note that we employ the standard notation to denote the co- and contra-
variant derivatives:
(4.11)
Observe that the order of differentiation is relevant. The D symbol involves
the Christoffel symbol made up from the classical field
(4.12)
82 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

In the derivation we used the identity


(4.14)
Further:
(4.15)
(4.16)

(4.17)

(4.18)
(4.19)
(4.20)

(4.21)
Inserting the various quantities in eq. (4.1), we find

(4.22)

This expression will eventually supply us with the equations.of motion


for the classical fields and Allowing partial (co- and contra-variant)
integration and omitting total derivatives:

(4.23)
G 'T HOOFT AND M VELTMAN 83

Further:

(4.24)
Performing partial integration and omitting total derivatives:

(4.25)

The Lagrangian is invariant to the gauge transformation of and


(4.26)
To have Feynman rules we must supplement the Lagrangian eq. (4.25)
with a gauge fixing part and a Faddeev-Popov ghost Lagran-
gian. In section 2 we have shown that there exists a gauge that allows
easy verification of unitarity; the work of ref. 7 tells us that other choices
of C are physically equivalent, and describe therefore also a unitary theory.
We will employ the following C:

(4.27)

The quantity is the root of the tensor


(4.28)
84 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

It has what one could call « mid-indices », but it will not play any substan-
tial role. Using (4.27) we find:

(4.29)

With this choice for C we obtain:

(4.30)
with

(4.31)

(4.32)

(4.33)
The Lagrangian eq. (4.30) is formally of the same forms as considered
in the previous section, with fields written for the Even if the result
of the previous section was very simple it still takes a considerable amount
of work to evaluate the counter Lagrangian. This will be done in the next
section. There also the ghost Lagrangian will be written down.

5. EVALUATION OF THE COUNTER LAGRANGIAN

To evaluate the counter Lagrangian we employ first the doubling trick.


In addition to the fields h and we introduce fields h' and that interact
with one another in the identical way as the h and That is, to the expres-
sion (4.30) we add the identical expression but with h' and instead of h
and Obviously our counter Lagrangian will double, because in addition
to any closed loop with h and particles we will have the same closed loop
but with h' and particles.
G. 'T HOOFT AND M VELTMAN 85

After doubling ofeq. (4.30) and some trivial manipulations we obtain:

(5.1)

(5.2)

The counter Lagrangian is invariant to the replacement


(5.3)
with (compare eq. 2.8):
(5.4)
It is to be noted that the replacement (5.3) is not a covanant replacement.
But at this point the transformation properties of the are no more rele-
vant, they are treated simply as certain fields as occurring in the equa-
tions of section 3.
We so arrive at a Lagrangian of which the part containing two deriva-
tives (with respect to the fields h and is of the form
(5.5)
Note that is not the same thing as treating the as
scalars. We must rewrite in terms of derivatives that do not
work on the indices We have:

(5.6)
In this way one obtains
(5.7)
with
(5.8)
We have written for symplicity only one term, the subscript « symm »
denotes that only the part symmetrical with respect to exchange, as
well as exchange is to be taken. Further:

or symbolically
(5.10)

where the covariant derivative « sees » only the index explicity written.
86 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

We can now apply the equations of section 3. The fields etc.


may be renamed the remaining fields as 11 etc. The
matrix of section 3 can be identified with eq. (5.8), that is non-zero
only in the first 10 x 10 submatrix. The matrix is of the form

(5.11)
Here

(5.13)
with X, Y, Z given in eqs. (4.31-33) and in eq. (5.4). The counter Lagran-
gian due to all this is given in eq. (3.35). Note that the in eq. (5.11) cancels
out (see eq. 5.10). A calculation of a few lines gives:

(5.14)
See appendix B for the last equality (apart from total derivatives). From
eq. (3.35) we see that we must evaluate

The various pieces contributing to this are


(5.15)

(5.16)

In evaluating terms like Tr(R 2 ) remember that one takes the trace of a
10 x 10 matrix.
(5.17)

(5.18)

As a final step we must compute the contribution due to the Faddeev-


Popov ghost. The Faddeev-Popov ghost Lagrangian is obtained by sub-
jecting to a gauge transformation. Without any difficulty we find
(note
(5.19)
Terms containing h or can be dropped, because we are not splitting
up in classical and quantum part. The Faddeev-Popov ghost is never
external. In deriving (5.19) we used an equation like (4.14), and trans-
G 'T HOOFT AND M. VELTMAN 87

formed a factor away by means of an substitution. Again, it is neces-


sary to work out the covariant derivatives; the contribution due to that is:
(5.20)
The evaluation of the rest is not particularly difficult; the total result is:

(5.21)

Notice the minus sign that is to be associated with F-P ghost loops.
Adding all pieces together, not forgetting the factor 2 to undo the doubl-
ing of the non-ghost part, gives the total result (remember also the last
part of eq. 3.35 for the non-ghost part; one must add 11 times that part):

(5.22)

The obtain the result for pure gravitation we note that contained in eq. (5.22)
are the contributions due to closed loops of particles. But this part is
already known, from our calculations concerning a scalar particle in an
external gravitational field. It is obtained from eq. (3.40) with M = N = 0:

Subtracting this from eq. (5.22) and setting equal to zero gives the counter
Lagrangian for the case of pure gravity:

(5.24)

6. EQUATIONS OF MOTION

From eq. (4.23) we can trivially read off the equations of motion that
the classical fields must obey in order that the first order part disappears:
(6.1)

(6.2)
88 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

Taking the trace of eq. (6.2) we find

Substituting this back into eq. (6.2) gives us the set:

(6.3)

For pure gravity we have simply


(6.4)
Inserting eq. (6.3) into eq. (5.22) and (5.24) gives:

(6.5)

(6.6)
If one were to approach the theory of gravitation just as any other field
theory, one recognizes that the counterterm eq. (6.5) is not ofa type present
in the original Lagrangian eq. (4.1), and is therefore of the non-renorma-
lizable type.
The question arises if the counterterm can be made to disappear by
modification of the original Lagrangian. This will be investigated in the
next section.

7. THE « IMPROVED » ENERGY-MOMENTUM TENSOR

The Lagrangian eq. (4.1) can be modified by inclusion oftwo extra terms:

(7.1)

The last term cannot improve the situation, because it has not the required
dimension. So, we have not considered the case b 0. Concerning the
coefficient a we know already that the choice reduces divergen-
cies of diagrams without internal gravitons. These are not present in
eq. (7.1) because the -field has no non-gravitational interactions of the
type say. Actually this same choice for a seems of some help in the
more general case, but it still leaves us with divergencies.
G 'T HOOFT AND M VELTMAN 89

The essential tool in the study of more complicated theories is the Weyl
transformation (see for example ref. 15). This concerns the behaviour under
the transformation
(7.2)
where is any function of space-time. By straightforward calculation one
establishes that under this transformation

with

From this:

and

(7.6)
Conversely

(7.7)

Consider now the Lagrangian

(7.8)

with Now perform the transformation (7.7). We obtain

(7.9)

This Lagrangian belongs to the general class

(7.10)
90 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

It is not too difficult to see the changes with respect to the treatment of
the previous sections. However, it becomes quite cumbersome to work
out the quantity and we have taken recourse to the computer and
the Schoonschip program [16]. Roughly speaking the following obtains.
The required Lagrangian (7.1) has = 1 — As is clear from eq. (7.9)
the Lagrangian written in the form eq. (7.10) becomes a power series in
the field with coefficients depending on a. The counter Lagrangian
becomes also a power series in with non-trivial coefficients. Putting
the coefficient a to is of little help, the final result seems not to be of
any simple form.

8. CONCLUSIONS

The one loop divergencies of pure gravitation have been shown to be


such that they can be transformed away by a field renormalization. This
depends crucially on the well-known identity (see appendix B)
= total derivative,
which is true in four-dimensional space only.
In case of gravity interacting with scalar particles divergencies of phy-
sically meaningful quantities remain. They cannot be absorbed in the para-
meters of the theory.
Modification of the gravitational interaction, such as would correspond
to the use of the improved energy-momentum tensor is of help only with
respect to a certain (important) class of divergencies, but unrenormali-
zable divergencies of second order in the gravitational coupling constant
remain. We do not feel that this is the last word on this subject, because
the situation as described in section 7 is so complicated that we feel less
than sure that there is no way out. A certain exhaustion however prevents
us from further investigation, for the time being.
G. 'T HOOFT AND M. VELTMAN 91

APPENDIX A

NOTATIONS AND CONVENTIONS

Our metric is that corresponding to a purely imaginary time coordinate. In flat


space Units are such that the gravitational coupling constant is one.
The point of view we take is that gravitation is a gauge theory. Under a gauge trans-
formation scalars, vectors, tensors are assigned the following behaviour under infini-
tesimal transformations
scalar
covariant vector
contravariant vector
Note that dot-products such as A"B,, are not invariant covariant
but behave as a scalar.
two-tensor

Let now be an arbitrary two-tensor. It can be established that under a gauge trans-
formation
(A.1)
A Lagrangian of the form

where is a scalar (in the sense defined above) is invariant under gauge transformations.
One finds:
(A.2)

The second term is a total derivative and the integral of that term vanishes (under proper
boundary conditions).
Further invariants may be constructed in the usual way:
(A.3)
(A.4)
with
(A.5)

In here may be any symmetric two tensor possessing an inverse but in practice
one encounters here only the metric tensor. The quantities do not transform under gauge
transformations as its indices indicate; in fact
(A.6)
The quantities behave under gauge transformations as a covariant two tensor.
Similarly
(A.7)
behaves as a mixed two tensor.
Let now be the tensor used in the definition of the covariant derivatives. Then it is
easy to show that
(A.8)
92 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

Another useful equation relates the Christoffel symbol and the determinant ofthe tensor
used in its construction:
(A.9)

This leads to the important equation


(A.10)
As a consequence one can perform partial differentiation much like in the usual theories:

(A.11)

Given any symmetric two tensor having an inverse one can construct the associated Rie-
mann tensor:
(A.12)
We use the convention which is of importance in connection with the raising
and lowering of indices.
The Riemann tensor has a number of symmetry properties. With, as usual
(A.13)
one has

(A.14)
The Bianchi identities are
(A.15)
The generalisation of the completely antisymmetric four-tensor is

(A.16)
Note that
(A.17)
It is easily shown that
(A.18)
In the derivation one uses the fact that in four dimensions a totally antisymmetric tensor
with five indices is necessarily zero. Thus:
(A.19)
Finally, the Bianchi identities lead directly to the following equation
(A.20)
G. T HOOFT AND M. VELTMAN 93

APPENDIX B

PRODUCTS OF RIEMANN TENSORS [6]


Let us consider the following Lagrangian

(B.l)
Subjecting to a small variation

we will show that the integral remains unchanged (the integrand changes by a total deri-
vative).
In section 4 the equations showing the variation of the various objects under a change
of the have been given. One finds:

The last term can be treated by means of the same identity as we used at the end of appen-
dix A, eq. (A. 19); i. e. this term is equal to the sum ofthe four terms obtained by interchang-
ing with and respectively. The last three terms are equal to minus the original
term, and one obtains, after some fiddling with indices:

(B.2)
The final result is
(B.3)
Using the Bianchi identities in the form (A.20), and exploiting the fact that the covariant
derivatives of and are zero we see that
(B.4)
which is the desired result.
The consequence ofthis work is that in a Lagrangian the expression (B. 1) can be omitted.
Now (B. 1) can be worked out by using the well known identity

+ (all permutations of the upper indices with the approprate sign). (B. 5)
One obtains
(B.6)
The above derivation can be generalized to an arbitrary number of dimensions. The
recipe is simple, take two totally antisymmetric objects and saturate them with the Rie-
mann four tensor. The resulting expression is such that its variation is a total derivative.
For instance in two dimensions:
(B.7)

Vol XX, n° 1 - 1974


94 ONE-LOOP DIVERGENCIES IN THE THEORY OF GRAVITATION

Clearly, the Einstein-Hilbert Lagrangian is meaningless in two dimensions. This fact shows
up as an n-dependence in the graviton propagator; in the Prentki gauge as shown in eq. (2.9).
Also in other gauges factors l/(n — 2) appear, as found by Neveu [13], and Capper et al. [4].

Annales de rinstitut Henri Poincaré - Section A

You might also like