You are on page 1of 16

PHYSICAL REVIEW D VOLUME 27, NUMBER 12 15 JUNE 1983

Strong-coupling approach to the quantization of conformal gravity

Michio Kaku
Physics Department, City College of the City University of New York, New York, New York 10031
(Received 14 October 1982)
Unlike the Einstein-Hilbert action, conformal gravity is a formally renormalizable theory.
However, conformal gravity has nonunitary dipole ghosts when quantized in the weak-
coupling approximation. We conjecture that the weak-coupling limit is not a valid one, be-
cause of the presence of infrared divergences and the existence of nontrivial Dirac con-
straints on the physical eigenstates of the theory. We show that the simplest, realistic ana-
log for conformal gravity is higher-derivative Yang-Mills theory. We study its strong-
coupling limit and show that its eigenstates are stringlike and that the dipole ghosts are
completely eliminated by the gauge constraints. We formulate conformal gravity in the
strong-coupling approximation, both in Hamiltonian and Lagrangian forms. Again, there
seem to be powerful constraints which eliminate dipole ghosts from the eigenstates of the
Hamiltonian and which allow stringlike states. We find "ghost confinement." One inter-
pretation, in this picture, is to identify the graviton as a stringlike solution to the Dirac con-
straints of the conformal group.

theory eliminate the dipole ghosts from the eigen-


I. INTRODUCTION states of the Hamiltonian. The ghosts are "con-
fined." Although higher-derivative Yang-Mills
The Einstein-Hilbert action probably is not a re- theory is probably not physical (due to positivity
normalizable theory because of the presence of a di- problems in the Hamiltonian), it does serve as a tool
mensional coupling constant (Newton's constant). with which to begin the quantization of conformal
The direction of research in gravitation has been to gravity.
take this unitary theory and to show that it is renor- Although there are many ways in which to inves-
malizable in the weak-coupling limit. tigate departures from weak-coupling perturbation
In this paper, we take the opposite approach and theory, the one we choose for this paper is the
take a formally renormalizable theory, conformal strong-coupling expansion of the wilson4 lattice,
gravity, and try to show that it is unitary in the patterning ourselves around the Hamiltonian for-
strong-coupling limit. malism of Kogut and susskind5 and the Lagrangian
Conformal gravity is usually ruled out as a physi- strong-coupling methods of Balian, Drouffe, and
cal theory because, in weak-coupling perturbation ~tz~kson.~
theory, the free action generates zero-norm states There are difficulties in actually calculating the
(dipole ghosts) associated with any free theory with free energy in the strong-coupling expansion of con-
1/p4 propagators.1 Our conjecture is that confor- formal gravity, but general features of the strong-
mal gravity may still be a unitary theory if we use coupling expansion are apparent. We find that there
nonperturbative t e ~ h n i ~ u e s .The
~ , ~ presence of di- are strong constraints placed on the eigenstates of
pole ghosts may simply be an artifact of the weak- the Hamiltonian due to local Lorentz invariance,
coupling expansion. and that one natural solution of these constraints is
We feel that the free higher-derivative theory is the closed string with zero mass (at least in lowest
simply too primitive to approximate the complex energy 1.
features that are found in conformal gravity, such as Proving the positivity of the Hamiltonian is a
Dirac constraints associated with local gauge invari- problem, however, because of the presence of com-
ance. We feel that the simplest realistic analogy to plicated operators in the theory. Although we have
conformal gravity should be a higher-derivative apparently eliminated the problem of dipole ghosts,
Yang-Mills theory. We show that the quantization we have another problem of proving positivity to all
of this theory creates constraints on the eigenstates orders.
of the Hamiltonian. The solutions of these con- One question remains even if the mathematics
straints are given by strings, not by free dipole ghost works out correctly, and that is how to physically
states. Apparently, infrared divergences in this reproduce the physics of the Einstein-Hilbert action

27
- 2819 @ 1983 The American Physical Society
2820 MICHIO KAKU 27
-

at large distances. One approach is to use especially because of the constraints. In the same
spontaneous-symmetry-breaking effects to generate way that free colored states are not solutions to the
an R term which will dominate the physics at dis- constraints of gauge theories, we take the view that
tances large compared to the Planck length. the dipole ghosts found in (2.2) are an artifact of
Another approach is to identify the graviton as a weak-coupling perturbation and they are not solu-
closed-string state. If a massless spin-2 state can be tions to the constraints of conformal gravity.
shown to exist in the theory, then we have one can- Thus, we conclude that the simplest model which
didate for the graviton itself. The theory of gravity should serve as a guide to the quantization of con-
is renormalizable, therefore, because the graviton formal gravity is a higher-derivative gauge theory,
has structure. which preserves the original constraints of SU(N)
local gauge invariance, while exhibiting 1/ p 4
11. HIGHER-DERIVATIVE GAUGE THEORY behavior in the weak-coupling propagators. The
question we ask in this section is the following: Do
Conformal gravity is usually ruled out as a physi- the gauge constraints found in higher-derivative
cal theory because, in lowest-order perturbation gauge theory forbid the presence of dipole ghosts
theory, it is not unitary. Conformal gravity in per- found in weak-coupling perturbation theory? Are the
turbation theory appears very much like an ordinary ghosts "confined"?
higher-derivative free-particle theory, which is In this section, we will set up the strong-coupling
known to have dipole ghosts. For example, the free expansion of higher-derivative gauge theory to see if
theory this Kogut-Susskind quantization is sufficient to re-
move the dipole ghosts of weak-coupling perturba-
tion theory.
We begin our discussion of higher-derivative
when quantized canonically has the Hamiltonian gauge theory by first considering a gauge-invariant
H= d3k(afbk+b:ak+ha:ak), action which has higher derivatives:

We see immediately that single-particle states have


zero norm, and that there are problems in diagonal-
izing the Hamiltonian.
In this paper, however, we take the view that
weak-coupling perturbation theory is not a valid for- Notice that this gauge theory has higher deriva-
malism in which to study conformal gravity. Be- tives, and that we must be careful in handling the
cause of the presence of infrared divergences in a time components of the theory. We expand out the
massless theory and because of nontrivial Dirac con- action into its components:
straints, perhaps weak-coupling perturbation theory
does not correctly give the true Milbert space of
asymptotic states.
We wish to explore analogies to ordinary gauge
theories, where free states found in perturbation
theory (colored states) are actually ruled out because
of infrared divergences and because the spectrum of
free states does not satisfy the Dirac constraints of
the theory. We realize that pushing analogies be-
tween gauge theories and conformal gravity may be
dangerous, but the question of the renormalization At this point, we see that there are too many time
of gravity is sufficiently important to warrant a full derivatives to put the theory into standard canonical
exploration of nonperturbative approaches to the form. We use a trick: inserting an auxiliary field ki
quantization of conformal gravity. which will soak up the extra time derivatives. We
In this section, we therefore make the assumption can introduce this auxiliary field by using Lagrange
that the Hamiltonian (2.2) is too simplified to exhib- multipliers, i.e., by inserting the following term into
it the complex features found in conformal gravity, the action:
27
- STRONG-COUPLING APPROACH TO THE QUANTIZATION OF ... 2821

where ri is the Lagrange multiplier. By integration


by parts and simple reordering, we can rewrite this
Lagrange multiplier term into a more transparent Putting everything together, we now obtain the com-
form: plete canonical reduction of the higher-derivative
gauge action,

Notice that the Lagrange multiplier has now become


the canonical conjugate to the ki Geld, and that A.
emerges as the Lagrange multiplier for gauge Although we originally started with a higher-
transformations, and that the generator of gauge derivative gauge theory defined in terms of the field
transformations includes the covariant derivative of A @ ,we now have four fields, arranged into two sets
ri. Using these two auxiliary fields, it is now a sim- of canonical conjugates: ( fi,k,) and (ri,A, ). The
ple task to reduce out the total action. generator of gauge transformations is simply the co-
The first term in the action can be rearranged by variant derivative of the r field along with i (f,k)
using the familiar Bianchi identities for gauge term.
theories: We can now introduce canonical commutation re-
lations among the canonical fields,

This allows us to define the generator of gauge


The second term in the action can be reduced by transformations
introducing yet another auxiliary field f i (the field
f i can be eliminated by functional integration using
Gaussian integrals):
which introduces the following transformations
among the fields:
[ T , A : ] = i ( a , ~ ~ + g f abcAibT c ) ,

(The equality means that, by functional integration,


[ T ,fF]= -igfakvbf:
the elimination of the f i field reproduces the left-
hand side.) This term can also be rearranged to
make it more transparent, It is important to point out that the Hamiltonian
I,= J d 4 x Tr( -+ f12+ f,kr-igAo[ki,fi]) -
contains terms which may not be positive definite.
Some of the terms occur with an incorrect sign, and
(2.9) the n term appears linearly in the Hamiltonian,
which may occur with either sign. Although posi-
Notice that the f i field is nothing but the canoni- tivity probably cannot be ruled out to any finite or-
cal conjugate of the ki field, and that the Lagrange der in perturbation theory, we must point out that
multiplier A. now introduces another term into the this Hamiltonian is probably not a physical one. We
generator of gauge transformations. The addition of use it only as a guide to examining the Hamiltonian
the ( f , k ) term into the generator simply means that of conformal gravity, which is considerably more
the f i and ki fields separately transform as the ad- difficult.
joint representation of the group S U ( N ) . A t this point, we are now in a position to examine
The third term will remain the same. The fourth the states of the theory. We can rescale the fields as
term can now be rewritten as follows: follows:
2822 MICHIO KAKU 27
-

Notice that the commutation relations are not af-


(2.15) fected, but the net effect is to alter the Hamiltonian:

I
In the strong-coupling limit, only the f term dom- As a -0,
inates (which resembles the case of strong-coupling
ia 'gFp,( n )
gauge theory, where the term E~ dominates). P(n)pv-e
Like in gauge theory, the theory cannot be defined
because the eigenvalues of f are not normalizable. and
In strong-coupling gauge theory, we define a nor- P(~),,u,,,P-'(~+8),,~-',,,
malizable Hilbert space by introducing a small pa-
rameter a (the lattice spacing), and then altering the
canonical commutation relations for E f so that the
electric field becomes the generator of rotations for
a rigid body. In principle, however, there are an in- -ia3g~,~p,(n)
finite number of ways in which the strong-coupling -e
action may be modified to reproduce a set of nor-
This implies that
malizable states.
The most compelling method, of course, is the in-
troduction of lattice gauge theory. Transforming
from the Wilson Lagrangian form to the Kogut-
Susskind strong-coupling Hamiltonian form gives us
a unique way in which to modify the canonical com-
mutation relations (2.12).
By analogy, we now introduce a higher-derivative
lattice theory in Lagrangian form and then carefully
convert the theory to Hamiltonian form in order to
find a generalization of the canonical commutation
relations which allow for normalizable states. We
now start to generalize (2.12).
We begin by defining the usual string bit,

We define a plaquette as the complete circuit


around a square in the lattice space.

In order to introduce higher derivatives into the


lattice gauge theory, we must necessarily perform
complete loops over paths which look like Figs. 1
and 2. In these paths, we are essentially taking the
covariant difference between two plaquettes (i.e., the
covariant derivative of a curvature). We now define
our action as
FIG. 1. These paths and those in Fig. 2 represent the
orientation of the plaquettes necessary to generate a
higher-derivative Yang-Mills theory. Notice that we are
essentially taking the difference between two plaquettes
separated by a distance, which ultimately will become the
derivative of a curvature.
STRONG-COUPLING APPROACH TO THE QUANTIZATION OF ... 2823

In ordinary quantum mechanics, we begin our


discussion of functional methods by defining the
usual eigenstates of the position operator,
4^19)=9/9) ,
19) J d q ( 4 1 = I 9 (2.22)

Similarly, we can generalize these equations to the


FIG. 2. See caption for Fig. 1. case of string states,' where we introduce the string
position operator U defined by

Following the example of gauge theories, we take


the gauge Ao=O, where the string bits in the time As in ordinary quantum mechanics, we can intro-
direction all reduce out to unity. duce the operator p^ which induces translations in the
As before, we can now reduce our four different eigenstates
types of terms in the action

which implies
/ q + Q ) = e - i ~ e / q .)
In the same way, we can introduce the operator
R^(g)which has the properties
h

R^ -'n,iig)9n,iR^,,i(g)= g u n , j , g ESU(N) ,
fin,ign,i(g)I u)=gun,iR^n,i(g)1 u ) 9

which implies that


h

Rn,i(g) 1 U ) = / g U ) = / U') ,
where

UAjZi=gU",I. if n =n', i = j ,
(2.25)
U,',8,i= UnBj if nf n', i#j .
Thus, if we now introduce the following
parametrization:

where

then it necessarily follows that


,
R^n,i(g)R^n,i(gl)=R^n,i(ggt)
In order to make sense out of each of these com-
plicated terms, we must first introduce the function- [f;(n), fib(n)]=ifabcfic(n), (2.27)
h h
al formalism defined on the lattice. We begin our [f;(n),Un,i]= -.PUn,i .
discussion of functional methods by first pursuing
an analogy with ordinary one-dimensional quantum The whole point of this digression was to find
mechanics. these commutation relations.
2824 MICHIO KAKU 27
-

These are the generalizations of the canonical In quantum mechanics, we make the transition
commutation relations (2.12) for ordinary gauge between the Lagrangian formalism to the Hamil-
theories. We now seek to generalize them to higher tonian formalism by making the following function-
derivatives. al manipulations:
J

F(q,t;q',tf)= J [+ [c ]
exp i
1
~ ( q , c j ) d t [L = i c j 2 - ~ ( q ) ~

Using

-i$.e.
s(ei-9i+9i+1)=(9i-ei 1 q i + 1 ) = ( Ie
~~ 1qi+]) ,

we get

I
which implies that reduce out the first term in the lattice action. Since
we want to express everything in canonical coordi-
nates, it is necessary to introduce auxiliary variables,
In the same way, we can also make the same func- just as in the continuum case. This is achieved by
tional manipulations on the lattice to derive the introducing the following term into the action:
Hamiltonian of the theory.
The manipulations of the lattice functionals, of
course, are more involved than the previous one-
dimensional example, but the same techniques ap-
ply. Let us examine each term in the lattice action
separately.
The first term involves the time displacement of a
one-plaquette state. In the continuum example
given earlier, this term was reduced to canonical
form by use of the Bianchi identities, converting
terms like DoFij into DiFoj, etc. The Bianchi identi-
ties on the lattice are more complicated, but can still
be rigorously defined.'
Consider the six distinct paths given in Fig. 3.
Let U(1) represent the SU(N) matrix corresponding
to the path along route number 1. It is then easy to
see that
FIG. 3. These paths represent the six directions neces-
sary to generate the Bianchi identities on the lattice.
These relations are just identities, and hence yield no new
because the combined paths of the six routes taken information on the lattice. We will use these Bianchi
together yield unity. identities to reduce our higher-derivative action to canoni-
Now we wish to use the above Bianchi identity to cal form.
27
- STRONG-COUPLING APPROACH TO THE QUANTIZATION OF ... 2825

be expressed entirely in terms of canonical string


variables,

It is easy to see that the V matrix represents the A

"time derivative" of the U matrix (as in the usual U-'(4)= Un,jP(n+ j ) o r ~ - ' n , j 7

gauge formalism on the lattice), but now we must ~-'(5)=P(n)~o,


introduce a higher-order auxiliary variable to soak
up the extra time derivative. W then represents the U ( 3 ) U ( 6 ) =~ ( 3 ) ~ - ' ( 5 ) ~ - ' ( 4 )
time derivative of the V matrix, or the "second time u-'(3)~-'(2)u-'(l),
derivative" of the U matrix.
It is now straightforward to show that Il can now which implies that

The previous expression, in the limit of zero lattice spacing, simply reduces to terms like (Dikj12.
We are now in a position to reduce I2into canonical form, using the functional techniques developed for the
simple one-dimensional case. It is necessary to have two sets of independent string operators U and V (because
we are deali~gwith: higher-derivative field theory), and to have two sets of momentum operators associated
with them, R and R ', respectively, which have the following properties:

which implies that

I
where where the lattice spacing in the time direction is set
-I~,&T'
continuously to zero, keeping the lattice spacing in
V ~ ,=Ie , the space directions the same. [If we let a. be the
wn , ~
-el~a7a
time spacing and a be the space spacing on the lat-
- (2.33) tice, then the I,'s, of course, must be modified. In-
h h

[.-",,,,Un,11=-.p~n,1 9
stead of occurring as order ( l / a 2 ) , I, occurs with
(ao/a3), I2with ( l/aoa), I3with (ao/a3), and I4
[ftl,qn.fI = - ~ f n , l with (ao/a3). We now let a. go to zero while keep-
ing the other a fixed.]
We can therefore rewrite I 2 as
In the limit a. goes to zero, we can perform the
Tr (2.34) integration over qa, ipaf;,i
12=-- 2 2 2 (Wn,i+ w-'n,i) .
2ag , i J [ d W ( ~ ) l ( V n +Ige~ I Vn,i)
Inserting I2into the S matrix, we can now function-
ally integrate out over the qa variable, which is a
simple Gaussian integral, because the integral is
dominated by pa when W becomes a maximum.
Tr a
1
- (Vn+6,i1 e ~ ~ - [ ( i g ~ / 2 a ) a ~ (( fVn,i)
~i)~ .l
This occurs because we are now taking the limit (2.35)
2826 MICHIO KAKU 27
-

We see immediately that this reproduces the strong- like before, is unchanged. And I4becomes
coupling Hamiltonian. The great advantage in this
Tr ao
rather long derivation of the Hamiltonian is that we I4= 7~ (~n,iun,/~-'n+/,iu-~n,/-l)
now have the exact commutation relations between 28 a n,i/
the f field and all other fields in the lattice approxi-
mation, which generates normalizable states. The +H.c. (2.36)
whole point is that (2.33) replaces (2.12).
All other terms in the action can be readily ex- We can now reassemble all terms in the action, and
pressed in terms of canonical string variables. 13, we are left with the final lattice Hamiltonian

From the first principles, we might not have give a mass to stringlike states by constructing se-
guessed that the commutator of the 7~ field with the quences like UUUVVVUUUVVV. . .. The U 's do
Ai must be modified as in (2.33). The advantage of not change under the action of the Hamiltonian f ',
going to the Lagrangian formulation of the lattice to but the Hamiltonian does count the number of V's
the Hamiltonian is that we get the complete commu- in any such chain. ( U and V, however, transform
tation relations of all fields immediately. This, in differently under the gauge transformation.)
turn, leads to a normalizable set of eigenstates of the The important thing to realize, of course, is that
Hamiltonian. the dipole states found in weak-coupling perturba-
In the limit that the spacing a goes to zero, we tion theory are not part of the lowest-order strong-
rederive the old continuum action given earlier, coupling expansion, in analogy with gauge theories.
(2.11). The ghosts are "confined."
The purpose of this digression into lattice gauge
theory is to give a new structure to the continuum 111. HAMILTONIAN STRONG-COUPLING
commutation relations given earlier, which led to EXPANSION FOR CONFORMAL GRAVITY
non-normalizable states. We now see that f actually
has commutation relations identical with the genera- The purpose of the previous section was to prove
tors of the algebra itself, meaning that the spectrum that nontrivial complications are introduced when
of states resembles those generated by a spinning we quantize a higher-derivative theory with gauge
rigid rotor, much like in Kogut-Susskind lattice invariance. In particular, we find that the naive set
gauge theory. of dipole ghosts states derived from the weak-
The great advantage of going to lattices is that we coupling perturbation theory look nothing like the
can now construct the Hilbert space of the theory to actual eigenstates of the Hamiltonian. The Dirac
all orders in the strong-coupling expansion because constraints are sufficient to rule out, to lowest order
(2.33) replaces (2.12). in the inverse coupling constant, the presence of di-
The first observation we make is that the strong- pole states. (This does not mean that higher-
coupling Hamiltonian f 2 does not depend on the derivative Yang-Mills theory is a physical theory be-
string variables $ and U . This means that the cause of the question of positivity of the Hamiltoni-
asymptotic set of states includes, to lowest order in an.) We also note the presence of zero-mass string
the inverse coupling constant, zero-energy string states in the theory (which may or may not remain
states composed of U's which are gauge invariant. massless to all orders in the inverse coupling con-
Thus, any state of the form stant). In conclusion, the simple free theory of a
higher-derivative particle action with zero-norm
states cannot serve as the "minimal" model in which
necessarily has zero energy to lowest order. We can to study conformal gravity.
27
- STRONG-COUPLING APPROACH TO THE QUANTIZATION OF . . . 2827

Applying these same techniques to conformal


gravity meets considerable problems, as we shall see,
but we can make considerable progress in defining
the strong-coupling expansion of the theory.
We first start with the group structure of the con-
formal group 0(4,2) [locally isomorphic to SU(2,2)]
with six Lorentz rotations Mob,four translations Pa,
four conformal transformations KO, and one dilata-
tion D,'

We are using the usual convention adopted in


canonical vierbein theories where the raised index
i,j, k,. . . refer to a lower index which has been raised
by 3g'j, which is the three-dimensional inverse of
3gij,i.e.,
In the usual way, we form curvatures

(see Refs. 2 and 3 for more conventions).


where
The kf has 12 independent fields. We further
decompose this mixed tensor into a six-component
antisymmetric tensor kab and a six-component sym-
metric tensor kij as follows:

- 2 ( e ~ ~ - e ~ ; - e ~ ~ + e etc.
~ ; ) ,

In the usual theory, the b, field drops out of the In much the same way, we now take the 16 com-
action. Here, however, we want to keep this field ponents within the f tensor and decompose it into a
because it will eventually become one of the dynami- three-vector fin, another three-vector fni, two scalars
cal fields of the canonical theory.
~ e f o r e , we
~ , ~showed that the conformal action
f,,, and f:, a symmetric traceless tensor x,,
antisymmetric tensor fij with three components:
and an

can be put into canonical form by first identifying


which fields are physical and which ones are not.
It turns out that the theory cannot be placed into
canonical form unless we introduce auxiliary fields.
We introduce the field kf (the generalized second
fundamental form) which is related to the time
derivative of the ef field. We introduce the kf field
into the action by using a Lagrange multiplier

We now can rearrange the action into the familiar


canonical form of a Hamiltonian plus Dirac con-
straints.
The Lagrange multipliers for the first-class Dirac
where constraints are (composing the 15-parameter confor-
2828 MICHIO KAKU

ma1 group):
(ep,+") ,
e t =multiplier for general covariance, (&,,Ti') ,
kab=multiplier for local Lorentz transformations,
fin, f,,=multipliers for proper conformal (bi,fni),
transformations, (k,f;) .
b =n 'ba =multiplier for scale transformations.

Furthermore, we can now identify the canonical We can now put the action into the usual canoni-
pairs of coordinates cal form

where
Ka=fieai(4f3 S k , i - ~ ; & ~f
- bik)
,-
+6na(--Tk;jk'J+ f k 2 + 3 b k b k - 2 ~ ; b k - 4 f : - ~ 3 ~ ) ,

I
1 = DiA j. +gijbkAk-bjA;
D r* Aj .- , terms. We actually find that, by carefully including
surface effects, the form of the two generators Ho
and and Hi, given as

We notice immediately that, like the case with the


Einstein-Hilbert action, the Hamiltonian is actually
a total derivative.'' However, this surface term is
nonvanishing and can be calculated. actually creates surface terms in the Hamilton's
In the case of the Einstein-Hilbert action, we no- equations. The correct equations, including total
tice that the full Hamiltonian formalism is not com- derivative terms, actually reads as follows:
plete unless we include the surface term. In the
metric form of gravity, where we have the canonical
field gij and its conjugate d j , we find that the equa-
tion

is actually not satisfied unless we add in surface If we take boundary conditions equivalent to a
27
- STRONG-COUPLING APPROACH TO THE QUANTIZATION OF ... 2829

Schwarzschild metric at large distances, we find that constraints by the canonical fields does not repro-
the only term which survives is duce the correct equations of motions because sur-
face terms were deleted.
To begin this reduction, we find it convenient to
first eliminate some of the invariances. We will ex-
plicitly eliminate proper conformal transformations
by setting Ka=O, and then by choosing four more
Thus, the Hamiltonian is actually not zero in the constraints: bi =O and k =O. Of course, we still
Einstein-Hilbert example. have the usual Poincari group and dilatations. In
We find an exact parallel in the conformal case. this reduced form, we can write down an equivalent
We also find that the simple-minded variation of the theory based on the following constraints:

I
and M~~is unchanged. straint yields two types of terms, one corresponding
Once again, it is easily checked that the equation to surface terms and the other corresponding to nor-
mal terms which make up the A,B,C,D found above.
The sum of these two terms is zero. But the surface
terms make up the real Harniltonian. By setting the
P constraints to zero, then, we can use the nonsur-
(3.17) face terms instead of the surface terms as our Ham-
does not represent the correct variation of the con- iltonian (since they sum to zero).
straints unless we add in an extra surface term S. Second, we can use the strong-coupling limit
The actual calculation of the surface terms is very (scaling f by a factor of g and scaling k down by a
straightforward, but rather tedious, and certainly factor of g-'), in which case only the f term dom-
not enlightening. However, we can take a short cut inates. (The other terms can also be ruled out by
to the final answer. asymptotic arguments concerning reasonable as-
First, if we set Pa to be zero (thereby breaking sumptions about the large-scale behavior of the
general covariance), then the variation of the con- fields.)
straint also is zero. But the variation of the P con- The expression for S is
MICHIO KAKU

where

If we now take the limit as the coupling gets Hilbert space of the theory.
strong, take into account reasonable asymptotic ex- There may be an infinite number of ways in
pressions for the fields at infinity, and use the fact which to generalize the commutation relations so
that Pa=O, we find that the Hamiltonian reduces to that a normalizable set of eigenstates emerges. The
lattice, however, should give us the most convenient
set of generalized commutation relations, so we now
Like the higher-derivative gauge theory found in turn to a lattice description of conformal gravity.
Sec. 11, we find that the conformal Hamiltonian is Unfortunately, we will find that the lattice formula-
nothing but the square of the f field in the strong- tion of conformal gravity is not sufficiently
coupling limit. developed to give a definite answer concerning the
The crucial difference, however, between the con- spectrum of states of the theory.
formal case and the higher-derivative gauge theory
is that the commutation relations are much more in- IV. LATTICE CONFORMAL GRAVITY
volved, and so are the higher-order interactions. Be-
cause we have explicitly broken general covariance In the previous section, much like the case of con-
by taking Pa=O, we must necessarily choose four tinuum higher-derivative gauge theory, we found
more constraints for the theory. Fortunately, a sim- that the commutation relations were not sufficient
ple examination of the Dirac brackets shows that to define a normalizable set of eigenstates. This
there is a wide class of constraints which can be ar- problem was solved by going to the lattice approxi-
bitrarily imposed which preserve nice properties of mation.
the theory. We begin a discussion of the lattice formulation
For example, by choosing our constraints so that of conformal gravity by warning ahead of time that
the present formulation is incomplete.
Starting with a string bit [using the Dirac ma-
we find that trices as 0(4,2) generators],

so that, to lowest order in the inverse coupling con-


stant, we find that a string state composed of
Lorentz string bits has zero mass and satisfies all
constraints, It is easy to show that the following action has the
T r P exp [ $~ ' ~ o ? ~] d1 Ox) . desired properties:

We find, therefore, that string loops defined over


the Lorentz group have zero energy and satisfy all
constraints. (Unfortunately, we also find that any We see that in the limit of zero lattice spacing, we
combination of metric tensors gij, because they have reproduce the original continuum action.
zero Poisson brackets with the Lorentz generator In the continuum case, however, we had to add
M'~, also have the property of having zero mass. one more constraint into the theory,
Apparently, higher-order corrections will give them
mass.)
The great difficulty, however, in actually calculat-
ing with this system is that we do not know exactly
how to generalize the commutation relations among
f, k, e, and a so that we have normalizable eigen- (4.3)
states. In the case of higher-derivative gauge theory, This constraint allows us to make the usual iden-
we appealed to the lattice to give us a prescription tification with the connection on the group. We
which generalized the commutation relations found must impose this constraint, because the equation of
in the continuum theory. Using the quantum motion for the connection field for Lorentz transfor-
mechanics of rigid rotors, we then can construct the mations does not reproduce the usual result.
27
- STRONG-COUPLING APPROACH TO THE QUANTIZATION OF ..

On the lattice, this constraint can be reproduced


by imposing

Now, let us test to see that our lattice action is actu-


ally invariant under conformal transformations.
First, let us discuss local Lorentz transformations.
If Z is a matrix created by exponentiating the
Lorentz operator M'~, then we find that the
transformation
This means that the action and the constraint are
each invariant under proper conformal transforma-
keeps the action invariant because y5 commutes with tions.
z. The theory, however, is not invariant under ordi-
The constraint can similarly be treated. Under
nary translations Pa. This is because, in the contin-
the action of Z, we note that the matrix associated uum theory, the Weyl action is only invariant under
with P a transformations simply rotates into itself, a modified P transformation, and this modified P
transformation can be shown to be exactly a general
covariant transformation. Thus, the P transforma-
tion in the fiber space becomes, after modification,
the general covariant transformation. This is very
elegant, but for our purposes it means that we can-
not treat the P transformation like the others. Be-
The entire action, including the constraint, is there- cause our lattice theory explicitly breaks general co-
fore invariant under local Lorentz transformations. variance (which is only restored in the continuum
The variation of the action under dilatations is limit), we find that our lattice action is actually in-
also easy to calculate, because D commutes with the variant under the conformal subgroup of the Poin-
y5 found in the action, and D simply transforms card transformations and dilatations, not the entire
ya( 1+ys) into itself. Thus, both the action and the 0(4,2).
constraint are again invariant. At this point, one may proceed in two directions,
Under the action of proper conformal transforma- both of which will receive more attention in further
tions, the situation is a bit more difficult. First, no- work.
tice that the constraint does not change under the (1) We can try to use the radiation gauge to
action of proper conformal transformations, because transform the lattice into a Kogut-Susskind Hamil-
tonian form, in which case we may be able to derive
the correct commutation relations which will permit
a normalizable set of eigenstates for the strong-
coupling limit. The difficulty here is that the gauge
And second, notice that the action of proper confor- Un,,= 1 cannot be chosen, because we do not have
mal transformations induces enough invariances to completely eliminate the
yU(1+y5)c'I --~~(l+y~)€~ fields corresponding to the P transformations (the
e Y5e =y5+2ya(l+y5ka+ . . . vierbein). Because general covariance is only re-
stored in the continuum, we cannot impose the
gauge Ungo=1 from the start. This means that any
Therefore, we now must check that generalization of the commutation relations must, at
some point, include the effects of general covariance
(which is explicitly broken on the lattice).
(2) One may proceed with the strong-coupling ex-
However, this term is also zero because we can in- pansion of the lattice in the Lagrangian formalism
sert a complete set of states between the two pla- without imposing the radiation gauge. As in the
quettes, and each term in this expansion is separate- usual gauge theory on the lattice, we can power ex-
ly zero, pand in powers of the inverse coupling,
2832 MICHIO KAKU 27
-

We must be careful in integrating over noncom- Similarly, conformal gravity also possesses a
pact groups such as 0(4,2) [or SU(2,2)], but other- strong-coupling expansion which allows zero-mass
wise all integrations in the previous equation can be closed string states (the strings are defined over the
performed to any order in the inverse coupling. Lorentz group). Like higher-derivative gauge
The problem, however, is that even the integration theory, we find that the Hamiltonian to lowest order
without any powers of the inverse coupling is just as is simply f 2 . There may be many ways in which to
hard as solving the original gauge theory because of modify the commutation relations so that we obtain
the presence of the Lagrange multiplier term. a renormalizable set of eigenstates.
In principle, it should be possible, perhaps with The simplest nontrivial generalization of the com-
Monte Carlo techniques, to calculate the space mutation relations should come out of the lattice.
dependence of terms like However, because general covariance is not a sym-
metry of the lattice version of conformal gravity, we
have a certain amount of difficulty in making the
where C1 and C2 correspond to closed loops in con- transition from the Lagrangian to the Hamiltonian
formal space. approach. Meanwhile, the lattice itself has a
Analytical methods of actually calculating the strong-coupling expansion in the Lagrangian for-
strong-coupling expansion are currently being stud- malism. We find that all integrations can formally
ied. be performed, but that it is difficult to obtain this
In summary, our lattice techniques are not yet strong-coupling power expansion analytically.
developed enough to give a definitive answer on how Eventually, we hope to be able to establish posi-
to modify the commutation relations so that the tivity of the Hamiltonian of conformal gravity. Be-
strong-coupling Hamiltonian f can generate a nor- cause conformal gravity is renormalizable, proving
malizable set of states. that it is also unitary would render the theory a
leading candidate for a quantum theory of gravity.
V. DISCUSSION One remaining question may be asked: where is
the graviton in this theory? One approach may be
We have proven what we set out to prove, namely, to use symmetry breaking1' to derive an R term in
that the free higher-derivative action with dipole the action, which would then dominate interactions
ghosts is simply too elementary to serve as a model at distances large compared to the Planck length.
for the quantization of conformal gravity. There is another possibility, however. We men-
We have further shown that the dipole ghost tioned that the constraints to the theory are very re-
states are actually forbidden (to lowest order in in- strictive, and that we want conformal dipoles to be
verse coupling) because they do not satisfy the gauge eliminated by these constraints. The solutions to
constraints of the higher-derivative gauge theory. these constraints would be Lorentz-invariant states.
We thus make the case that higher-derivative One possibility is to identify the graviton as a string-
gauge theory is the simplest model where nontrivial like state of the theory. The graviton would have
complications due to gauge constraints begin to structure, which would disappear at distances large
emerge. We suspect that, due to infrared diver- compared to the Planck length. In order to show
gences, the naive weak-coupling expansion is not this, we must first establish that the free energy of
adequate to describe the asymptotic states of the our lattice is compatible with zero-mass states. (Be-
theory. The ghosts are "confined." cause higher-derivative gauge theory seems to have
Although higher-derivative gauge theory is not a zero-mass strings, at least in lowest order, it may be
physical theory (because there are terms which may possible that there are massless particles in the con-
eventually violate positivity in the Hamiltonian), we formal case as well given by closed strings.) Then
do see novel features, such as the presence of closed we must establish that there are spin-2 states in the
string states with zero mass (to lowest order). theory (most likely given by a linear combination of
Because the continuum theory in the strong- several plaquettes, each one rotated with respect to
coupling limit did not yield a normalizable set of the other). This is an unorthodox conjecture, postu-
states, we had to go to the lattice to make a modifi- lating that the graviton emerges as a string in the
cation of the commutation relations compatible with strong-coupling limit, but it is a conjecture that we
normalizable states. wish to explore. W e have developed a generally co-
27 STRONG-COUPLING APPROACH TO THE QUANTIZATION O F .. . 2833

variant random lattice formalism to explore this pos- behavior of the correlation functions will indeed
sibility.12 show free ghosts, but that they are confined within
Of course, because the assumptions underlying the boundaries of a bag whose size emerges from a
this approach in this paper are speculative, serious dynamical breaking of scale symmetry (such as
objections can still be raised. We comment on two found, perhaps, in a lattice approach). These ghosts,
of them. in some crude sense, are confined, so that they are
(1) If an R term is nonperturbatively generated by never manifested as physical states of the theory, but
the R 2 action, then it seems that we can now use nevertheless the correlation functions will detect the
weak-coupling perturbation theory to expand around presence of such ghosts. If we can use these correla-
the R term and hence ghosts are once again intro- tion functions in the ultraviolet region to extract a
duced as a possibility. This most likely does not physically meaningful result which is nonunitary
happen in our scenario because Einstein's action is due to the presence of ghosts, then our approach to
only an effective van der Waals force that emerges the quantization of gravity may be shown to be in-
at large distances between real states of the Hamil- correct. Our conjecture is that, if the theory is
tonian. In field theory, it is possible that the effec- asymptotically free, then the pointlike ghosts may
tive, van der Waals force between bound states has show up in Green's functions of the theory, but not
interactions which are seemingly nonphysical (e.g., in any physically measurable fashion.
nonrenormalizable, nonunitary) because the "real" In conclusion, we would like to stress that many
dynamics does not take place at the effective scale. of these questions are being explored and that our
In our approach, the real states of the Hamiltonian approach is speculative, and that it represents a
are solutions of the Dirac constraints, which may be drastic departure from the current direction of
bound states quite unlike the ghost states found in research in the quantization/renormalization prob-
weak-coupling perturbation theory, while the long- lems of gravity. Although this new avenue of
range force of Einstein's R term only emerges as a research significantly deviates from the current phi-
van der Waals force, and hence should not be quan- losophy and conventional wisdom, we feel that the
tized in the usual way. question of the renormalization of gravity is a suffi-
(2) Because conformal gravity and superconfor- ciently important problem to warrant new ideas.
ma1 gravity might be asymptotically free, then per- Using the renormalization properties of conformal
turbation theory should be a valid tool for exploring gravity to explore the strong-coupling or nonpertur-
the ultraviolet behavior of correlation functions, bative properties of the theory, which may yield en-
which in turn should show nonunitary behavior. tirely new physical states or confined objects, may
We evade this problem in our approach with the fol- be a promising new approach to the question of the
lowing analogy. In QCD, the ultraviolet behavior of nonunitary structure of the theory in the weak-
correlation functions does manifest the pointlike coupling expansion and to the question of the renor-
structure of the quarks, but these quarks are only malization of gravity itself.
"free" within the confines of a "bag." At high ener- ACKNOWLEDGMENTS
gies, the quarks are like free particles as long as they
are within the effective bag, but quarks as physical We would like to thank B. Sakita for critical com-
objects are never seen. In a crude analogy, we con- ments and valuable discussions. M. K. was support-
jecture that, if these theories are found to be asymp- ed in part by Grants Nos. NSF PHY-78-24888 and
totically free, then once again the ultraviolet CUNY F R A P R F 13090.

'J. Kiskis, Phys. Rev. D l5, 2178 (1974); S. Ferrara and 3376 (1974);11, 2090 (1975);11,2104 (1975).
m,
B. Zumino, Nucl. Phys. 301 (1978). 'M. Creutz, Phys. Rev. D 15,1128 (1976).
2M. Kaku, Nucl. Phys. m, 285 (1982). 8J. Kiskis, Phys. Rev. D 26, 429 (1982).
3M. Kaku, preceding paper, Phys. Rev. D 27, 2809 (1983). 9M. Kaku, P. Townsend, and P. van Nieuwenhuizen,
4K. Wilson, Phys. Rev. D 10,2445 (1974). Phys. Lett.m, 304 (1977); Phys. Rev. D u, 3179
5J. Kogut and L. Susskind, Phys. Rev. D 15,395 (1974). (1978). For discussions of lattice gravity, see A. Das,
6R. Balian, J. Drouffe, and C . Itzykson, Phys. Rev. D 10, M . Kaku, and P. Townsend, Phys. Lett. m, 11
2834 MICHIO KAKU 27
-

( 1979); M. Kaku, in Superspace and Supergrauity, edit- (1974).


ed by M. Rocek and S. Hawking (Cambridge University "See S. Adler, Rev. Mod. Phys. 54, 729 (1982) for com-
Press, Cambridge, England, 1980). See also L. Smolin, plete references. M. Kaku and P. Townsend, Phys.
Nucl. Phys. w, 333 (1978). Lett.m, 54 (1978).
l q . Regge and C . Teitelboim, Ann. Phys. (N.Y.) 88, 286 12M. Kaku (unpublished).

You might also like