Professional Documents
Culture Documents
Subcritical Closed String Field Theory in Less Than 26 Dimensions (1994)
Subcritical Closed String Field Theory in Less Than 26 Dimensions (1994)
Michio Kaku
Physics Department, City College of the City University of New York, New York, New York 10031
(Received 15 July 1983)
There has been a fundamental flaw in all previous attempts to put supersymmetry on the lattice.
Because supersymmetry closes on the Poincare group, and since the Wilson lattice breaks Poincare
invariance, the standard lattice must necessarily break supersymmetry. We solve this difficulty by
putting supersymmetry on a random super lattice, where each random site is a point in super space
( x i , < i ) . We construct the action out of unitary S U ( N superfields connecting two such super lattice
sites and sum the traces over simplexes. The theory is manifestly supersymmetric because there is
no preferred direction in either real or Grassmann space. This supersymmetric lattice gauge theory
on a random lattice reduces to the usual supersymmetric gauge theory in the continuum limit. We
discuss applications, such as calculating nonperturbative corrections to the vacuum energy, which
may yield a large enough supersymmetry breaking to explain the hierarchy problem. We discuss us-
ing random lattices to describe a lattice version of gravity and supergravity.
31
- 1992 @ 1985 The American Physical Society
-
31 SUPER LATTICES AND GAUGE THEORY 1993
straints and because of infrared divergences. In effect, strong-coupling expansion. In order to find this crucial
nonperturbative effects produce ghost confinement, where measure, let us review a few elementary facts about in-
the true Hilbert space may be unitary. The way in which tegration and lattices.
we hope to calculate these nonperturbative effects is to In ordinary calculus, we construct a translation-invar-
put general covariance on a random lattice. The invari- iant integral by summing a function over a set of regular-
ance under reparametrization of a manifold is replaced by ly spaced lattice sites
the equivalent summation over all random lattice sites. In
this way, we put gravity and supergravity on a random
lattice which preserves reparametrization invarianceS8
This definition of supersymmetry, of course, is closely (The variable y i , which for now is equal to an integral,
linked to the definition of the integral over a Grassmann will play an important role in defining the Grassmann lat-
variable. Berezin's integration rules are used so frequently tice.)
that one often forgets how these rules were derived in the For now, yi is chosen to label regularly spaced sites
first place. Berezin originally derived the definition of the only for pedagogical reasons. Of course, later on we must
Grassmann integral by demanding that it be invariant make yi label random sites. We will treat yi as regularly
under a translation. In other words, he demanded that the spaced for the next few examples, keeping in mind that yi
integral over a Grassmann variable obey will eventually be a covariant label for random sites on the
slattice.
This sum, although formally invariant under transla-
tions, is not well defined, so let us introduce the following
measure for convergence:
N
A', M
= lim ~ ~ , , v +[ b( aak c a + c k 2 6 ' c 2 ) + ( b 1 - c k c 2 ) ( n c ' + m e 2 )
N,M--tm n = - ~
where
'-1
N ,M we must now integrate over the set of all random configu-
M.v3~= lim
N,M-+eo
2
=-N
(n2+m2) PElE2 . (2.10) rations of the lattice sites in order to achieve a mapping
m=-M which is invariant under arbitrary Grassmann displace-
ments, not just fixed Grassmann displacements.
Again, we could equally well have achieved this result if Our next task is to generalize the previous discussion to
we had let k 2 equal the sum over the squares of n and m, the four-dimensional case. We begin with a complex
and if we had used the projection operator P Grassmann spinor p,(ga)+ = '. We can, of course, con-
f If2'
We now have constructed an alternative formulation struct a real vector superfield with 16 components out of
which is equivalent to Berezin's integration rules. Finally, these Grassmann variables .rr(C,f ) = a +
. . . +dc2f 2. In
-
31 SUPER LATTICES AND GAUGE THEORY 1995
order to write down a Grassmann lattice in four dimen- vector space and the corresponding Grassmann space are
sions, let us parametrize our lattice in the following way: remarkably similar, with rotations and scale transforma-
tions in one space reflected as rotations and scale transfor-
mations in the other. (In the case of k#O, we must take
-
,a,aa -6 6
CYBUBo,,a differences between Grassmann states in order to clarify
this topology. Rotations are to be studied when we com-
(2.11) pare yj,i with ~ k , ~If . yj,i and yk,i lie on the same hyper-
E ~ ~ = E1,~ ~ = - 1, E ~ ~ = ,
E ~ ~ = ~
sphere, then their corresponding Grassmann variables will
differ by a pure rotation in spinor space.)
If yi rotates into -y, , then the vector has rotated 180
where E is a constant spinor and is a Pauli spin matrix degrees. We suspect that its corresponding Grassmann
( a = 1, . . . , 4 ) . Notice that the vector yi takes the place variables will have rotated a full 360 degrees in order to
of the integers we introduced in our earlier discussion. In pick up an overall minus sign. T o prove this, we must
y space, the vector yi connects the origin with a lattice now calculate precisely how ci differs from c,. If k = 0 ,
point i, where these lattice points are, for the moment, all then we can calculate c in terms of 6 and therefore calcu-
integrally spaced in a rectangular array as in the Wilson late Ci in terms of cj:
lattice. This allows us quite generally to parametrize the
Grassmann lattice by a simple vector field, not by abstract
Grassmann variables. In much the same way, we can now
introduce a summation over lattice sites which preserves
translation invariance and maps vector fields into real
numbers:
( a ( & , E ) ) = lim CM,a(fi)=d,
lul+miEu We see immediately that cidiffers from Cj by a scale
transformation and a Lorentz rotation. By allowing yi to
sweep out all of real space, therefore, the Grassmann vari-
ables sweep out all possible fermionic variables that can be
reached by a scale and Lorentz transformation. As we
where i labels all points in a rectangular volume v and we noticed earlier, the square of ti is equal to 2 y i 2 ~ 2SO ,
let this volume smoothly approach infinity. As before, therefore any pure Lorentz transformation on the y's
PE2z2extracts out the coefficient of c2z2,which is equal to translates into a pure Lorentz transformation on the e's.
the square of the sum over yi2. Alternatively, we could Also, any pure scale transformation on the y's converts
have set k equal to the square of the sum over yZ2,in into a pure scale transformation on the Grassmann vari-
which case we would have gotten the same result if had ables. (In Euclidean space, this means for all yi and y,
used the projection operator PgzE2.Notice this definition which lie on the same hyperspace, fi and cj
are related by
a pure four-dimensional rotation.)
of the mapping yields exactly the same integration formu-
Let us now examine the case of a pure rotation. Let yja
las of Berezin:
be a vector pointing along the x l axis in Euclidean space
and now rotate this vector into yib, which has the same
length as y; but lies in the x l - x 2 plane and is rotated
It is instructive at this point to elaborate how the vector from y;. Under this rotation, we want to calculate how
yi parametrizes the Grassmann variables. As we men- the corresponding Grassmann variables transform:
tioned earlier, the distribution of points yi determines the
distribution of Grassmann variables on the super lattice.
When yi sweeps out real vector space, the corresponding = (cosw/2 +2a2'sinw /2 . (2.15)
ci sweeps out Grassmann space, so it is important to clari-
fy how the topology of the real vector space differs from Comparing this expression to the previous one for scale
the complex Grassmann space. Let us assume that k = 0 , and Lorentz transformations, we see immediately that we
so we can easily calculate how ci
differs from cj
when yi can write down a correspondence between rotations in y
is close to yj. Notice that ei2
= 2 ( y p ) 2 ~ 2Since
. the square space and rotations in Grassmann space, which are
of a Grassmann variable is proportional to the square of parametrized by the angle w :
its real vector counterpart, then ci2=cj2 if the square of
their real vectors are equal to each other (i.e., if yi and y i
lie on the same four-dimensional hypersphere). h hi$ If the length of the vector yi is equal to that of yj, then
means that when yi and yj differ by a pure rotation, we see that the angle between these two vectors is equal to
which preserves the lengths of their vectors, then their one-half the angle w.
corresponding Grassmann variables also differ by a pure We see, therefore, that as the vector y (in Euclidean
rotation (in a spinor representation). Notice also that if yi space) rotates around by 180 degrees along the surface of
and y j differ by a pure scale transformation, then their a sphere, the Grassmann variables rotate 360 degrees.
corresponding Grassmann variables also differ by a pure This means, of course, that the Grassmann variables pick
scale transformation. Therefore, the topology of the real up a factor of ( - 1 ) when y rotates 180 degrees or when
1996 MICHIO KAKU 31
-
where i labels the lattice points in a rectangular region v Notice that two supersymmetric displacements do not
in this eight-dimensional space with volume v 1 , while anticommute, but instead yield a net torsion. On the oth-
c, is the appropriate volume element so that we recover er hand, Q and D anticommute with each other (we let E
the integration over four-dimensional real space. PEZZ2 ex- parametrize the Q displacement and 7 parametrize the D
displacement):
tracts out the coefficient of c2g2. We could equally well
have set k equal to the square of the sum over yi2, in
which case we could also have chosen the projection
operator to extract out the coefficient of c2Z2.
Armed with the above construction in eight-dimen-
sional space, we are now in a position to begin a discus-
sion of physical models, such as the Wess-Zumino model. Now notice that successive applications of Q (or D)
We will find that there are two nontrivial complications adds Grassmann bilinears to the variable x, which is no
to our super lattice when we introduce these new models: longer strictly real. In fact, because this is a torsion term,
torsion and constraints. these bilinears do not vanish when one makes a complete
U p to now, we have not introduced torsion into the circuit in Grassmann space. Instead, we find that these
theory, which arises because the anticommutator between bilinears are proportional to the area enclosed by any such
supersymmetric translations yields a Poincare translation. closed path in Grassmann space. For example, let us take
All our summations have been invariant under separate two points zi and zj and connect them via n small succes-
translations in x and c.
When we introduce supersym- sive displacements generated by Q. We can label these
metric operators, however, we find that the variable x is discrete displacements starting from zi and ending up
no longer simply a real number, but contains bilinears in with zi+, . The new variables are represented as
the Grassmann variables. Most of these Grassmann bilin-
ears are proportional to the squares of infinitesimal dis-
placements, and hence can be dropped. Furthermore,
these bilinears will all drop out when we go to the contin-
uum limit, where they can be eliminated as total deriva-
tives. However, in the lattice we will find that we must
take into account these bilinears.
In this eight-dimensional space, let us now define a lat-
tice action which will eventually reduce to the Wess-
Zumino model in the continuum limit. We first begin Notice that, if we now take the limit of infinitesimal
with defining finite supersymmetric displacements Q in displacements in Grassmann space, our new variables can
this eight-dimensional space, which will reduce to the usu- be represented by the following integral, if we let zi and
al supersymmetric operators only in the continuum limit zf represent the initial and final points:
(any two random slattice sites can be connected by the
proper combination of supersymmetric translation opera-
tors P and Q. But the converse is not true. Arbitrary ap-
plications of the displacement operators P and Q to a slat- + ~ . c . , i ~ + n , i , E f + nJ ,.i (2.23)
tice site d o not, in general, generate another slattice site.
This will be important when we define chiral fields): Notice that, if we make a complete closed path in
-
31 SUPER LATTICES AND GAUGE THEORY 1997
Grassmann space, then the bilinear term is proportional to generated by repeated displacements by the operators Q
the area of that closed path. and P. However, because there is a torsion defined when
In the old Wilson lattice, we construct string elements making repeated operations by Q, a naive summation
out of the displacement vector xi -xj, which we notice is based on repeated displacements by Q will not yield a
invariant under translations. In the same way, for the unique description of each lattice site. The ambiguity is
super lattice, we also want to construct tensors which are reflected in the choice of the vectorial index x , , which will
invariant under supersymmetric translations. We notice in general contain bilinear products of Grassmann states
that the following tensors are invariant under both in addition to the usual real numbers. Therefore, we must
Lorentz and supersymmetric translations: modify the summation in the previous expression to in-
clude all possible summations over bilinear Grassmann
states in the variable x,. In the continuum limit, however,
this additional summation will have no effect on the final
result. This is because the modification of the variable xi
Notice that we can now construct fully Lorentz invari- to include a bilinear Grassmann component disappears
ant and supersymmetric invariants by simply contracting like a total derivative.
any of these tensors on each other to form scalars. Next, we can add interactions into our lattice version of
We are now in a position to define an action on this the Wess-Zumino model. If we simply take powers of the
eight-dimensional space. If we take an arbitrary super- chiral superfield and sum them over the lattice, we find
field .rr(x,<,g ), we notice that we have a large number of that the action vanishes in the continuum. In order to
unwanted fields. As in the continuum theory, we want to find nonzero actions, we must therefore make modifica-
base our action on the smallest irreducible representation tions in our Lagrangian. As with the case in the continu-
of the group, i.e., we need a constraint on the superfield um theory, there are actually several ways in which in-
T ( Z ) which preserves supersymmetry. (We will find the teractions can be defined in superspace, depending on
problem of constraints to be a crucial aspect of the con- which formalism we use: whether we integrate over two-
struction of supersymmetric gauge theories on the lattice.) or four-dimensional Grassmann space and whether or not
Because Q and D displacements anticommute with each we use the usual Q's or a chirally shifted version of the
other, the simplest constraint is to impose Q's. Each of these various formalisms has a direct lattice
analog. For example, instead of the usual choice of the
Q's defined on the lattice, we could equally well have con-
structed the lattice points out of displacements generated
It is simple to check that this constraint on the super- by the translation operators:
field does not change its invariance properties. If z
represents a point on our super lattice, then let
z,=z,+z, represent another point separated by a dis-
placement Q,+Q,. We now check that the above con-
straint is maintained if we displace our superfield by a su-
persymmetric translation generated by Q, Out of the several choices one can make for a lattice
version of interactions for the Wess-Zumino model, we
will use the usual Hermitian choice. In the continuous
case, we can write down an interaction term by integrat-
ing the reduced chiral term .rr(x,()* over two Grassmann
The first step is possible because D and Q anticom- variables, or by integrating T ( Z + , ~over
) ~ the same two
mute. The second step is possible because it is nothing Grassmann variables and letting the integration over x
but a reexpression of (2.25) with z replaced by z,. Thus, eliminate the bilinear term in z + (because this bilinear
we see that the constraint on the superfield is maintained term appears in a total derivative). In the same way, we
even when we make a translation via Q on our coordi- will adopt this latter choice in order to construct a super
nates. [In the continuum limit, the constraint (2.25) lattice analog:
reduces to Bn-=0.1
Our constraint can be satisfied if we reexpress our
superfield in terms of new variables. Let z + = ( x
+irE',ga). Then d z + ) automatically satisfies the
above constraint. In other words, to every point ( z i ) on
the eight-dimensional lattice, we can assign a function
T(Z+ 1. The projection operator P F 2 extracts out the coefficient
Now let us define our propagator by the lattice summa- of c 2 . If k equals the sum over yi2, then the projection
tion: operator P-, can be used to select out the coefficient of
E
lim 2 M"?i(z-i).ir(z+i) . <2, with the same result.
lo l k m i ~ ~ At this point, because we are summing the Lagrangian
over a random series of supersymmetric points, it is easy
The lattice summation here is slightly different from to see that the product of superfluids or chiral superfields
the summations introduced earlier. The lattice sites are transforms on the super lattice in the same way as the
1998 MICHIB
superfields themselves. This is because the transforma- Q and P transformations, but the converse is not true.
tion of superfields or products of superfields is given by This problem is remedied by noticing that we can extend
the difference of these quantities at two nearby lattice the parameter Tj until zgi approaches within the average
sites. Usually, in field theory we use the Leibnitz rule to lattice spacing of a real'lattice site. Because supersym-
construct group generators whose action on products of metry cannot be defined on the lattice in our approxima-
fields is derived from their action on individual fields. In tion to less than the average lattice spacing, the point zlii
the super-lattice case, however, we see that these group will come close enough to a real lattice site to preserve our
generators act identically on products of superfields as definition of supersymmetry, which is recovered only as
well as on the superfields themselves. (As shown in the an average over all lattice configurations. Therefore,
first reference, the alternative to this is to define the group without loss of generality or loss of invariance, we can
generators acting on individual fields only, and then to de- maintain the definition of (2.25) by comparing two real
fine nonlocal products of superfields which transform ac- lattice sites, not fictitious ones.
cording to the Leibnitz rule. This leads to the unpleasant Next, notice that the measure in (2.29) and the measure
necessity of defining nonlocal exponentials of superfields, in (2.17) differ from each other. This is because the defi-
such as those found-in super-lattice gauge theory.) nition of a chiral integration and a normal integration,
This feature is perfectly acceptable and is a natural out- even in the continuum limit, are quite different. In par-
growth of defining discrete space-time symmetries on lat- ticular, notice that the measure in (2.29) does, in fact, like
tices, whether they be discrete Poincare transformations the measure in (2.17) select out the largest divergent factor
defined on the Wilson lattice or continuous Poincare in the integrand. This is because factors which are pro-
transformations defined on the random lattice. Even in portional to cubic terms in the Grassmann variable be-
the Wilson lattice, a lattice field a ( x i ) transforms under come total divergences in the continuum limit and can be
discrete translations in the same way as r r ( x i 1". disregarded. Thus, notice that the measure in (2.17)
This is the way in which the random lattice field theory selects out quartic terms in the Grassmann variable and
evades the Leibnitz problem that plagues other formula- (2.29) selects out quadratic terms in the Grassmann vari-
tions of super lattices, which are defined on regular (rath- ables. but both of them select out the term in the in-
er than random) lattices. In these hypercubical formula- tegrand which has the largest number of Grassmann vari-
tions, we evaluate the transformation of the fields them- ables.
selves under supersymmetry SUSY transformations, and Lastly, one final comment must be made concerning
thus have a Leibnitz problem. In the random-lattice ap- the measure of integration that we have used throughout
proach, however, zi is treated as a dynamical variable. our discussion.
We must sum over all possible orientations of slattice At first. the measure that we have chosen, which is a
sites. Therefore, the slattice gauge theory is SUSY invari- function of yi2, appears to break supersymmetry because
ant because the action consists of an infinite sum over all the x and y parts have "factorized." In particular, be-
wossible orientations of slattice sites. A SUSY transfor- cause it is not a function symmetric in x and y , at first it
mation simply rotates this infinite collection of terms into seems that the measure does not rotate properly under su-
itself: The action is invariant under SUSY only when we persymmetry transformations.
sum over all possible orientations. This is totally different Our measure, of course, is supersymmetric. Under a
from the usual hypercubical lattice formalism, where we supersymmetry transformation, notice that the
want the lattice action defined over one specific orienta- Grassmann variables do not rotate into x variables. Su-
tion to mav into itself under SUSY transformations. The persymmetry is not a symmetry which rotates the
important feature is the invariance of the action, not Grassmann and x variables into each other; in fact, the
necessarily the fields, under SUSY. Of course, we can Grassmann variables simply undergo a displacement. In
still extract out the variation of the fields from the varia- y space, this means that y is simply displaced by a con-
tion of the action, but this transformation does not have stant amount. This is the reason why the measure does
to satisfy the Leibnitz rule. not have to be symmetric in x and y.
Given the fact that the action (which consists of an in- Second, under a constant displacement in y, the mea-
finite sum of terms each defined over a particular random sure does transform, but notice that the change in the
orientation of sites) maps into itself under a SUSY measure is infinitesimally small. The measure diverges
transformation, we easily find that the measure of each like yi2, while the change in the measure diverges as y i .
orientation must be unity. In our final result (3.31), we This means that, in the limit of infinite y space, the
functionally integrate over all possible zi as well as all change in the measure goes rapidly to zero. The measure,
possible unitary string bits Ui,with a measure equal to therefore, is indeed supersymmetric (i.e., invariant under a
unity for each orientation. constant displacementin the Grassmann variables).
We see that our action by construction is guaranteed to We have chosen our measure so that, in the limit of in-
preserve invariance under SUSY rotations. finite lattice size, we recover the usual Berezin rules. Be-
We close this section by making a few clarifying state- cause of this, the measure is not manifestly supersym-
ments concerning our action. First, notice that the defini- metric under variations in the x variables, although it is
tion of a chiral field (2.25) in general may involve com- fully supersymmetric in the limit of infinite lattice size.
paring the value of a scalar superfield at a fictitious point Of course, it must be possible to rewrite the measure in a
not defined on the random lattice. This is because every form which is manifestly supersymmetric even before we
random site can be generated by repeated applications of take the infinite limit. This manifestly supersymmetric
-
31 SUPER LATTICES AND GAUGE THEORY 1999
measure can be written down as a function of supersym- rameter where we can always reparametrize the dis-
metric quantities, which are given in (2.24). Any function placement by a and P translations:
constructed out of scalar products of these functions are
guaranteed to be manifestly supersymmetric. In particu-
lar, the most important function that we can construct We can parametrize this unitary superfield by introduc-
from the invariant tensors in (2.24) is the supervolume of ing superfields which have vectorial and spinorial indices
a super polygon. defined over displacements on the lattice space:
The idea of a supervolume comes from the usual
Lorentz (or Euclidean) random lattice. When construct-
ing the random lattice, we must assign a unique simplexi-
fication of the lattice into closed nonoverlapping sim-
plexes for each particular randomization of points. In
Sec. 111, we will divide up the random lattice into clusters
+
of points, consisting of D 1 lattice sites connected by all
possible links in a D-dimensional lattice. In Lorentz
space, each cluster has a volume which can be written as a Let zi,zJ,zk be three points which represent a three-
function of the various Lorentz-invariant products of the simplex in a cluster. (We will define how to calculate dis-
intervals xi-xi Likewise, in random superspace, it is tances between lattice sites, and therefore how to con-
also possible to write down an expression entirely in terms struct clusters, later.)
of the tensors given in (2.24) which describe the super- As a first guess, we can formally construct the action as
volume of a supercluster. The measure of integration is follows (we will modify this action later on when we in-
then manifestly supersymmetric even before taking the troduce constraints on our string elements):
limit of infinite lattice size.
Unfortunately, the formalism involved in constructing I= kijkTr( Uij Ujk Uki) . (3.4)
Ar,k
supervolumes from supersymmetric tensors is beyond the
scope of this article. We will present the results of the We will find that the evaluation of the measure kijk
measure defined as a supervolume, which are quite plays a key role in defining the theory on the super lattice.
mathematical, in a later paper. Notice that the definition of the action allows a gauge
As far as this paper is concerned, the measure that we invariance on the string element U:
constructed is perfectly adequate to describe supersym-
metric integration in the limit of infinite lattice size.
We notice that the above unitary transformation can be
111. SLATTICE GAUGE THEORY reexpressed, in the continuum limit, as a transformation
We can now begin our discussion of the gauge theory on the gauge fields given by
defined on this super lattice. Let each slattice site be la-
beled by eight (or more) real numbers ( xi,yi ), where x and
y are vector fields, or equivalently, by the point
zi =(xi,Ci,Ei). We want to connect the lattice points by
links so that D-dimensional space is divided up into non- It is clear from our discussion that the difference be-
overlapping polyhedra which completely fill up the space. tween the points zi and zj can be rewritten as displace-
Unfortunately, there are many ways in which we may ments generated by D and P (rather than by Q and P ),
connect these points in order to create a random lattice. with coefficients given by the 6's. The next question we
T o achieve a unique prescription, we first group sets of must ask is whether or not this construction is invariant
points into c l ~ s t e r s . ~ +
A cluster is a set of D 1 lattice under supersymmetric displacements on the lattice. In the
sites such that the hypersphere that goes through all these old Wilson lattice, we notice that the coefficients of dis-
sites contains no other sites in the interior. If D =3. for placement are trivially invariant under Poincare transla-
example, the entire space can be simplexified by dividing tions, because they are nothing but differences between
it up into nonoverlapping tetrahedrons. The set of links is the two lattice sites. In our case, however, we must check
defined to be the set of all lines that one can draw between that our string element is defined in a supersymmetric
the points in a cluster. way.
Now let us define a unitary element of S U ( N defined First of all, the ea's are supersymmetric because they
between zi and zj: are nothing but differences between two Grassmann sites,
which are supersymmetric. But it is not at all clear that
the vector 6' is supersymmetric. To check supersym-
metry, let us rewrite as a function of the lattice sites:
ment is not changed when we make supersymmetric Q A t this point, notice that the above expression for the
displacements on the lattice, in the same way that the usu- action does not appear to be gauge invariant because of
al Wilson string element is invariant under Poincare the piece involving the summation over the three lattice
translations. We can define distances between sites in the displacements. In reality, this expression is gauge invari-
Wilson lattice because xj,: is invariant under translations ant because we must remember that there is a torsion de-
in each orthogonal direction. In the same way, we can use fined with respect to the anticommutator of two super-
the invariance of (3.7) to define*the invariant distance be- symmetric translations. We can rewrite the action in a
tween two super lattice sites, which is essential in decom- more familiar form if we realize that the summation over
posing our super lattice into sets of clusters of sites. the displacements is not zero, but in fact is
Whether or not two sites are to be connected by unitary
SU(N) supermatrices depends crucially on whether or not
they are part of the same cluster.
We can now redefine our curvatures to agree with the
Given any point in superspace, it is possible to associate
ones found in continuum supersymmetric field theories:
a real number with it. We simply decompose this point
into its various Grassmann components, and then take the FAB =RAE except for FaB. =R US. - 2iAab .
sum of the absolute value (or square) of each of the real
coefficients of each Grassmann c ~ m p o n e n t .For
~ the rec- We are now in a position to determine the value of the
tangular lattice considered in Sec. 11, the invariant dis- measure kijk, using the criteria that we wish to obtain the
tance between zi and zj would simply be the sum of (yi,q)2 usual supersymmetric gauge theory in the continuum lim-
and (xi,q12. In our case, because we have torsion, we must it. We need to show that the summations over the lattice
also include the real coefficients of the bilinears contained sites can eventually be given in terms of 6 functions in or-
in xi,:, which is no longer strictly a real number. Our der to recover the usual gauge action. For this purpose, it
original eight-dimensional space, composed of (xi,yi), is is essential that we use the parametrization of Grassmann
actually higher dimensional if we include the space space given earlier in terms of the vector field y. There is
spanned by the real coefficients of the bilinear Grassmann a theorem3 that we will need to reduce out this expression.
terms contained in xi. The actual dimensionality of our First, however, we need some preliminary definitions.
space, therefore, depends crucially on how we define our The cell corresponding to lattice site i is the space of all
parametrization of the Grassmann space. From now on, points that are closer to the site i than to any other site
we will treat our superspace as if it were eight dimension- (see Fig. 1). Notice that the boundaries of this cell, which
al for convenience, but we must realize that the complete is a convex polyhedra, bisect the links which connect i
theory must include the other terms arising from the real with the other sites in the cluster. We see, therefore, that
coefficients of those terms generated by the torsion. The we can either decompose D-dimensional space into the set
modification introduced by these extra terms, it turns out, of all nonoverlapping clusters or cells. The ( D - 1)-
is trivial. dimensional surface sij is the set of points which bisects
Let us now begin a discussion of how our lattice action the link connecting i and j. The boundary of the cell cor-
can be reduced down to the usual supersymmetric gauge responding to the site i are made up by all sij, where j la-
theory in the continuum limit. The key to this construc- bels all sites which connect with the site i. The surface sij
tion, as we shall see, is the weight function kijk. We will in turn, has a ( D - 2)-dimensional boundary given by T ~ ~ ~ .
calculate this weight function by taking the continuum In three-space, rijk corresponds to a line, while in super-
limit and requiring that we reproduce the usual supersym- symmetric space, Tijk is a six- (or higher) dimensional
metric gauge theory. volume.
To begin a discussion of the continuum limit, let us It can be shown that the following summation over the
first Taylor expand the vector and spinor superfields in sites in the cell holds in D-dimensional space:
order to construct the various curvatures of the theory.
We will power expand each of the three string elements
around the average point Z= (z; +zj +zk )/3. For exam-
ple, if we let cij =cj
-&, then the power expansion of A
yields
In this case, A,B,C,D represent the purely vectorial in-
dices in D-dimensional space, and Aijk is the area of the
ijk triangle. The term Si is a small correction which sums
to zero when we sum over all cells in the lattice. Ltiis the
A simple power expansion of the trace yields D-dimensional volume of the ith cell.
-
Uij Ujk Uki exp ( i [ A ; ~ R/2~ ~
Applying these results to our eight-dimensional space
( xia,yp), we can prove the following relations by inserting
(2.11) into the previous expression where we introduce a
measure on the space (the indices A,B,C,D now represent
vectorial and spinorial indices):
( a , b ) = ( O , l , l ) for (&a,a;ct) .
SUPER LAWICES AND GAUGE THEORY 2001
We have now derived a supersymmetric gauge action. Whenever a unitary string element is represented as the
However, because we have not placed any constraints on product of two (one the inverse of the other) de-
the theory, we find that this theory contains an unaccept- fined only at the end points, then this string
ably large number of fields. xn order to make contact represents a pure gauge, i.e., the trace around a loop of a
with the usual supersymmetric gauge theory, we must im- series of contiguous string elements cancels out identical-
pose constraints on the string elements. ly. In our case, the theory does not reduce to the identity
In the usual theory, we impose the con- because the string element is no longer unitary.
straint FaB =Fdrb =Fab=O on the curvatures. From these Products of these complex vector fields can be used to
generate a string elemznt defined between two points
constraints, we can reexpress the entire theory in terms of
separated by a D and D displacement. In the continuum
two complex vector fields, instead of tensor fields. Thus,
limit, we find
any constraint on the string elements which make these
Grassmann curvatures vanish will, in the continuum lim- V(Z,,Z~)~(Z~,Z~)V(Z~,Z~)
it, reduce down to the usual supersymmetric gauge theory.
Unfortunately, there are many ways in which this con- - e x p ( i [ c a ~ , ( w,W )+zaK'( W, w )])
straint may be implemented in the continuum limit. In
the weak-coupling approximation, we are unable to dif- Z2 =" +DcZ1 12' z3 = z 2 Diz2 '
+ (3.19)
ferentiate which is the unique choice among the several z4 =Z3 +D ~ 1 2Z .~
possibilities which will yield the best reduction to the con-
tinuum theory. Ultimately, the correct choice of con- Similarly, it is possible to find a discrete version of the
straints on the string elements which will yield the unique string element defined between two points separated by a
2002 MICHIO KAKU
pure Poincare (not Grassmann) displacement using prod- Fag=R ,g -2iAag, where Rab is given by
ucts over exp( W). However, we find that, because - -
Grassmann displacements generated by products of D's Rair = - i D a [ e x ~ ( lD&exp(- F )I
cannot ever generate a pure real number Poincare dis-
-iD,[exp( - W)D,exp( W)]
placement, we find that we cannot reduce a string element
defined between two points separated by a Poincare -i { exp( - W)D,exp( W),exp( W )Daexp(- p ) ) + .
translation with a real-number coefficient. [Instead, we
will use a trick in order to rewrite A,,, -- in terms of (3.23)
e x ~W).]
( We know from the expansion of the action that the
We can satisfy FaB =Fkb= 0 by rewriting the usual ex-
term Fan occurs quadratically in the action. Therefore,
pression for Uij (where zi and zj are separated by both D
we can perform the functional integral over A,&, which is
and Poincare translations) in terms of the complex vector
field exp( W). We begin by rewriting the string element nothing more than a Gaussian functional integral. Of
Uij into a product of separate terms, each representing a course, we will find that the equation of motion for A,,
modification of the original path from zi to zj. Between is nothing more than A,, =( - i /2)Ra&(W, W ). (Actual-
these two points, let us define several intermediate points ly, there are more terms in this expansion, but all of them
and the corresponding string elements defined through are of lower order.) In this manner, we have now elim-
these intermediate points: inated the troublesome term A,, in terms of the complex
vector field W.
The next lower-order terms in the continuum expansion
of (3.9) are F,,Faa and FabFablThese terms, now written
entirely in terms of W and W , can be further reduced.
After a certain amount of algebra, we find
symmetric gauge theory, which will require a different will be given in a later paper, but we will make a few re-
choice of measure. marks about the strong-coupling expansion in our formal-
We notice from (3.24) that the term we want, F,,,', is ism. The expansion of our theory can be carried out in
actually a chiral object, since Dw,=o. If the measure powers of ( 1/ g ) in the usual way:
contained terms proportional to the square of sums over
( yp)2, then our action would vanish, because the chiral ob-
ject is only proportional to t2,not ('5'. Thus, we actually
want to use a different measure in this case, and let our
projection operators screen out the higher-order
term. We choose
where
A~a kb-
cl _8 L z ( Eijcik
a b cil
c +cjlcjkcji
a b c
Pnbc
+ €&€6ki€Eki +clj
a b c
Eli Elk ) , (4.4)
where ( ] stands for cyclic permutations of the indices, so
where Pabcequals the sum over all cyclic permutations of that a negative sign is picked up upon interchanging two
abc. (We must be careful when handling the curvatures real indices or one real and one Grassmann, while there is
found in each separate path. Notice that the reduced ex- no sign change picked up by interchanging two
pression for each path is not a gauge-invariant quantity, Grassmann indices. The presence of the F ' s is due to tor-
so we must handle non-gauge-invariant factors until the sion terms that must be carefully included, e.g., because of
very end of the calculation.) (3.10).
The tensor quantity can be further reduced. No-
tice that if it is contracted on cabcd,the resulting vector is V. CONCLUSION
perpendicular to all the c's. Notice that this tensor is also
cyclic in all its indices. Thus we conclude that this tensor In summary, we have found that the usual Wilson lat-
must be expressed in terms of the totally antisymmetric tice is not a suitable form in which to introduce super-
tensor cabcd: symmetry, because the Wilson lattice manifestly breaks
Lorentz invariance, while supersymmetry is an extension
of Lorentz invariance. Thus, we are forced to formulate
the theory on a random supersymmetric lattice.
We have seen that the essence of a supersymmetric
transformation is a translation-invariant mapping which
where, of the six 6's that make up the tetrahedron, only takes an arbitrary superfield and maps it into a real num-
three are actually independent. Because of the identities ber. The Berezin integration rules are one consequence of
linking up the various c's, it is easy to arrive at a more this definition, and the superlattice is another.
symmetrical-looking expression for the above tensor quan- With this new approach to constructing lattice versions
tities. of the Berezin integration rules, we have shown that we
In the continuum limit, we see therefore that in four di- can also construct supersymmetric actions, both for the
mensions we have the following identity: Wess-Zumino model and for the supersymmetric gauge
theory. In theory, by adding isospin indices to our
Grassmann variables, we should also be able to build ex-
SUPER LATTICES A N D GAUGE THEORY 2005
'P. H. Dondi and H. Nicolai, Nuovo Cimento 41A, 1 (1977); E. (unpublished), but the previous references represent the most
Elitzur, E. Rabinovici, and A. Schwimmer, Phys. Lett. 119B, complete approach to this method.
165 (1982); V. Rittenberg and S. Yankielowicz, CERN Report 4M. Kaku, Phys. Rev. Lett. 50, 1893 (1983).
No. TH-3263 (unpublished); C. M. Bender, F. Cooper, and 5D. Z. Freedman, P. van Nieuwenhuizen, and S. Ferrara, Phys.
A. Das, Rochester Report No. UR-838 (unpublished). See also Rev. D 13, 3214 (1976); S. Deser and B. Zumino, Phys. Lett.
M. Kaku, in Superspace and Supergrauity, edited by S. W. 62B, 335 (1976).
Hawking and M. Rocek (Cambridge University Press, Cam- 6M. Kaku, P. K . Townsend, and P. van Nieuwenhuizen, Phys.
bridge, England, 1980). Rev. D 17, 3179 (1978).
2E. Witten, Nucl. Phys. B186, 513 (1981). 7M. Kaku, Nucl. Phys. B203, 285 (1982); Phys. Rev. D 27, 2809
3N. H. Christ, R. Freidberg, and T. D. Lee, Nucl. Phys. B202, (1983); 27, 2819 (1983).
89 (1982); Columbia Reports Nos. CU-TP-206 and 207 (un- 8M. Kaku, CCNY report (unpublished).
published). Random lattices, and, in particular, actions like 9A. Rogers, J. Math. Phys. 21, 1352 (1980).
(3.4), have been proposed before, for example by this author 1°J. Kiskis, Phys. Rev. D 26, 429 (1982).