You are on page 1of 15

PHYSICAL REVIEW D VOLUME 41, NUMBER 12 15 JUNE 1990

Nonpolynomial closed-string field theory


Michio Kaku*
Institute for Advanced Study, Princeton, New Jersey 08540
(Received 4 May 1989; revised manuscript received 2 February 1990)
Recently, we announced the construction of a new nonpolynomial closed-string field theory
which successfully reproduced all N-point tree amplitudes, thus solving a long-standing problem.
This action generalized the four-string tetrahedron interaction that we introduced earlier, which
was required to derive the Shapiro-Virasoro model. However, we gave no derivation of the nonpo-
lynomial action. In this paper we present the detailed analysis of its derivation. We calculate all
nonpolynomial graphs up to eighth order and their coefficients, and even the coefficients of several
infinite series of polyhedra. We use gauge invariance to calculate all coefficients. Because of an
enormous redundancy created by a web of overconstrained equations, we have multiple checks that
our coefficients are correctly calculated.

I. INTRODUCTION The problem of constructing the nonpolynomial theory


is complicated by the fact that there are one or more po-
Closed-superstring theory1 gives us the hope of unify- lyhedra at each level. For example, we must include two
ing all known interactions. String field theory,* in turn, polyhedra with 6 faces, 5 polyhedra with 7 faces, and 14
gives us the best hope of implementing a fully nonpertur- polyhedra with 8 faces (modulo rotations and reflections).
bative approach to superstring theory. However, it is The number of polyhedra at each level in fact grows quite
quite embarrassing that a successful covariant closed- rapidly.
string field theory does not exist, because the fundamen- If we have a series of n-faced polyhedra, then let us in-
tal region of moduli space is either overcounted an dex them by i. Thus, the ith vertex function with the to-
infinite number of times or undercounted. pology of an n-faced polyhedra, which we call ni, is
Recently, we proposed3 that the solution to the prob- represented by ( Y n) i. Each of these n-faced polyhedra,
lem of undercounting moduli space is that a previously indexed by i, occurs with a distinct coefficient c ( n Ii, i.e.,
overlooked tetrahedron graph needs to be added to fill
the missing integration region for the Shapiro-Virasoro
amplitude. We then announced4 the complete closed-
T h i action can be shown to have the form
string action, which was nonpolynomial, where closed
strings of equal parametrization length interacted
through polyhedra. (A slightly different version of the
nonpolynomial action was also independently and simul-
taneously announced by Kugo and co-workers.' Howev- which is invariant under
er, in neither paper were any of the details of the calcula-
tion presented.)
In this paper, we rectify this situation and present the
detailed derivation of the action, giving all polyhedra to In this paper, we present only the first derivation of ac-
the eighth level and calculating all coefficients to that lev- tion, presenting the measure, the integration region, ver-
el, as well as the coefficients of several infinite series of tices, the coefficients c ( n ), , a,,and 8, to all orders, etc.
polyhedra. Because the system is highly overconstrained, However, we caution the reader that the first proof, using
there are many checks of the correctness of our calcula- brute force, requires a key identity that has not been
tion. proven to all orders in perturbation theory. We present a
We note that there is also a second derivation of the heuristic proof, but a proof to all orders is still being in-
nonpolynomial action, independent of Becchi-Rouet- vestigated. The second proof, using geometric string field
Stora-Tyutin (BRST) methods, using geometric string- theory directly, does not require this identity.
field theory.6 In fact, this is how we discovered the non-
polynomial action in the first place. However, this 11. WHY NONPOLYNOMIAL?
second proof will be presented in another paper.7 Be-
cause readers may be unfamiliar with the unorthodox In Fig. 1, we see the structure of the geometric theory.
methods introduced in geometric string field theory, we However, for our purposes all we need to know is that
have carefully written this paper using only familiar the geometric theory predicts the existence of a four-
BRST methods and not the techniques of the full string tetrahedron graph (Fig. 2) when we fix the parame-
geometric string field theory. Readers interested in this trization length of all closed strings to be equal.
second proof are advised to consult Refs. 6 and 7. Even without geometric string field theory, however,

41 3734 @ 1990 The American Physical Society


NONPOLYNOMIAL CLOSED-STRING FIELD THEORY

Unified String

Geometric String
Field Theory

Interpolating Gauge

FIG. 3. The equipotential lines created by N = 6 charges (one


charge at infinity) with the same absolute magnitude placed on
Midpoint Gauge Endpoint Gauge
the complex plane. The topology of the equipotential lines is
precisely the topology of an N-sided polygon, i.e., N strings col-
liding at 710. Riemann cuts lie along the equipotential lines
FIG. 1. The gauging of the universal string group uniquely and link pairs of strings. Notice that this diagram does not cor-
specifies geometric string field theory. (When we include super- respond to a naive midpoint collision of N strings, i.e., inter-
symmetry, the USG becomes the unified string group.) By mediate lines can have lengths larger than the length of an
fixing the gauge (i.e., fixing the vierbein), we parametrize the external string. Diagrams such as this intuitively show why
strings so that they break at their midpoints, end points, or any there are missing regions of the Koba-Nielsen space. Because
point in between which interpolates between the end-point and they are instantaneous Coulomb terms, the N-faced polyhedra
midpoint gauge. Thus, geometric string field theory contains all are products of N - 2 interpolating vertices.
known consistent string field theories as gauge choices.

we can see the necessity of higher-point contact terms. In fact, the topology of the complex p plane corresponds
First, consider N-point scattering of N closed strings of to the topology of N closed-string scattering.
equal parametrization length. We know from the dual The lines surrounding the various charges in Fig. 3
model that the region of integration must be the entire correspond to a single equipotential line. In Ref. 2, we
conformal plane. However, consider Fig. 3, where we showed that equipotential lines have a simple but impor-
have N = 6 charges (of equal absolute magnitude) placed tant interpretation: they map out the topology of the in-
on this complex z plane (with one point at infinity). In teracting strings in physical space. (Notice that this set
of equipotential lines is only possible because there are
the complex p plane, each charge is mapped to infinity.
Riemann cuts separating the equipotential lines which
encircle two charges.) Notice that the equipotential line
in Fig. 3 in the z plane is mapped to a series of vertical
lines at the slice T = O in the p plane [and not over an ex-
tended region of the ( a , ~complex
) plane].
In p space, it is easy to see that we cannot have all
strings touching each other at their midpoints. In fact, it
is easy to see that strings can overlap with each other
over continuous regions over the length. Thus, this ar-
rangement of charges cannot be explained if closed
strings only interact at their midpoints. In fact this ar-
rangement of charges corresponds to the missing region
of the complex plane.
Notice that the topology of the equipotential lines form
the edges of a polyhedron (a cube). These equipotential
lines can split into two (and not three o r higher) equipo-
tential lines when strings interact. This is because the
FIG. 2. The tetrahedron graph, representing the collision of merging of four equipotential lines only occupies zero
four strings. This graph is necessary to reproduce the Shapiro- measure in Koba-Nielsen space; i.e., by infinitessimally
Virasoro model. In general, graphs of this type cannot be pro- moving the charges we can separate four strings momen-
duced if strings interact purely at their midpoints. For these po- tarily colliding at one point into a sequence of three
lyhedra, strings can overlap over continuous regions of their pa- strings colliding at different times. Thus, we d o not have
rametrization length: they are simply products of the interpo- arbitrary polyhedra, but only ones in which no more than
lating vertex. three edges can meet at once; i.e., the number of corners
MICHIO KAKU

closed strings have equal length. Thus, near y ,,

This, in turn, gives us 2N constraints (for real and imagi-


nary parts):

FIG. 4. A triplet of three unparametrized, oriented strings.


Notice that these strings are defined in physical space-time. s+r
There is no such thing as midpoints for these unparametrized, we now must also fix the way in two strings
physical strings. When parametrized, they can have arbitrary overlap. There are two independent ways in this
length. can be done.
(a) First, we can describe the scattering of N external
strings where the interaction times are all different and
where strings interact strictly through their midpoints.
equals 2(N - 2 ) for N-string scattering.
These rules generate the familiar Feynman graphs.
In summary, there is a finite region of Koba-Nielsen
(b) Second, we can describe the situation when all in-
space where the topology of the graph remains the same
teraction times are equal to each other and where closed
as we move the charges, but where the overlap distances
strings interact not at their midpoints but interact con-
continuously change. Thus, there is a finite missing re-
tinuously over regions less than .rr in length. Thus, the
gion. In this sense, the n-faced polyhedra can be con-
missing region has the topology of a polyhedron and con-
structed as the ( n -2)nd power of the interpolating ver-
sists of the product of interpolating vertices, much as in
tex, in the same way that the open four-string vertex can
the open-string case.5
be written as the square of the interpolating vertex (Fig.
The sum of both integration regions yields the entire
4).6 The origin of this fact is that all these terms are in-
Koba-Nielsen plane, as expected.
stantaneous Coulomb interactions formed when breaking
Let us set
the gauge invariance of geometric string field theory.
T o give some detail to our intuitive result, let us gen- GiVj+, =Im[p(z, )-p(z, ) ] ( i odd) ,
eralize the open-string mapping for the closed-string case.
We begin by constructing an Abelian integral w=w,dz where we set the imaginary part to zero and 6i is a real
which contains only simple poles and zeros in the com- +
number which we set to be T . Thus, the ith and ( i 1)th
plex plane, such that string collide and overlap by a parametrization distance
.rr. (If we set 6, to be an arbitrary number, we are in the
interpolating gauge, which smoothly interpolates between
the end-point and midpoint gauges.) Because there are
f C complex equations, and two real equations for each
complex one, we have C - 2 constraints (we subtract two
because the last two conditions are redundant due to
overall conservation of charge, keeping in mind that the
Riemann cuts have charge). Let us now summarize the
counting of unknowns and constraints in the conformal
maD:
(This map can be derived easily from the Schwarz- 2N - 6 unknowns:( N,z, ,
Christoffel transformation, mapping the flat complex Midpoint scattering:
plane to a Riemann surface describing the collision of N ) .
2N - 6 con~traints:(a,,6~,,+,
closed strings.) We define w, to be an Abelian differential
whose square yields a metric gab on the space describing
a conformal surface which is flat everywhere, except at This map describes N closed strings colliding via their
the points zi, where there are singularities in the curva- midpoints when the interaction times are all different.
ture. Also, C = 2 ( N -2). The complex points y i in the z However, as we have seen, this does not fill up the miss-
plane will be sent to infinity in the p plane, and corre- ing region. Let us now explicitly display a second set of
spond to N external closed strings. The complex points zi constraints, which correspond to the missing region.
are the zeros of the Abelian differential, i.e., the interac- Let us relax the previous constraints, letting the 6,,, be-
tion or turning points of the N strings in p space. Alto- come unequal real numbers. We now set
gether, there are 4 N - 8 unknowns in the map, coming
from the zi and the N.
Now let us impose constraints to fix these unknowns, N
and 2,. We first demand that, at infinity, the external
41
- NONPOLYNOMIAL CLOSED-STRING FIELD THEORY 3737

where 6,,, is a real variable. The first constraint allows polyhedra forms at an instantaneous ~ = 0slice of the
the (real) overlap distance between any two closed strings complex plane. We thus have C -2 constraints emerging
to change continuously. The second constraint places all from the second constraint. In summary, we now have
interaction points directly on top of each other, so the the following counting for the missing region:

Missing region:
i 2N - 6 unknowns:(N,zi ) ,
2N -6 constraints:(ai,Rep(zl) the same) .

(The missing region can also be understood via group for each j. (Notice that this is true because the a,, are
theory. For open- [closed-]string vertices based on Fig. equal to zero when i and j do not appear in the same
5(a) [6(a)]the algebra does not close, and the missing re- face.)
gion is filled by the four-string interaction in Fig. 5(b) This, however, is not the only constraint that must be
[6(b)]. However, closed-string vertices based on Fig. 7 imposed. In Ref. 3, we found the curious constraint
close properly with no missing region. See Refs. 3, 6, and
7 for details. ) a,, i r (3.4)
which gave the correct missing region. We now give a
generalization of this strange constraint, and give its ori-
111. MEASURE O F INTEGRATION gin.
Let us have an arbitrary polyhedron with N sides, and
Let us define our integration measure. Let N equal the then bisect it into two separate clusters of polygons, i.e.,
number of faces of a polyhedra, C equal the number of left and right pieces L and R as in Fig. 8. Let the number
corners, and E equal the number of edges. Then of polygons in the left and right half equal NL and N R ,
such that they sum to N and they are each greater than 1.
N = N o . faces , Let the perimeter of the line which divides the left and
right halves equal P. Then
E=3(N-2)=No. edges, (3.1)
C = 2 ( N - 2 ) = N o . corners .
This has precisely the number of dimensions to fill the i.e., the perimeter between the left and right halves equals
missing region in Koba-Nielsen space, because the number of edges along the perimeter. The perimeter,
in turn, can be written as
No. Koba-Nielsen variables= E - N = 2 N - 6 (3.2)
which is the dimension of moduli space (i.e., two vari-
ables for each vertex minus six variables which are fixed
when we fix projective invariance).
Let a,,, equal the parametrization distance between the
ith and jth string. a,,, equals zero if the ith and jth string Now the mysterious constraint (3.4), generalized to all
do not overlap in the polyhedra. The total number of a's polyhedra, can be written as
equals E = 3 ( N -2 ). The next constraint we must im-
pose is that the total length of each external face is equal
to 2 r :
N

FIG. 6. If we choose (a) as our basic triplet, then the Jacobi


identities generate the unwanted four-string tetrahedron graph,
FIG. 5. If we choose (a) as our basic triplet, then the Jacobi shown in (b). The nonclosure of the Jacobi identities requires
identities generate the unwanted four-string graph in (b). Thus, the addition of the tetrahedron and an infinite number of po-
(a) is an incorrect definition of a triplet, unless we use four- lyhedra. Thus, in certain gauges, geometric string field theory
string interactions. will be nonpolynomial.
MICHIO KAKU fi

bein in the midpoint gauge. aiirepresents the reduced set


of independent variables and'z, represents the indepen-
dent Koba-Nielsen variables. When we reduce out the
vierbein, as in Refs. 6 and 7 , we find that its elimination
gives us precisely the correct measure necessary to con-
vert a,] space into the correct Koba-Nielsen space. With
this measure, we can precisely reproduce the dual model
for N-point scattering of closed strings.
These constraints give us a complete description of the
missing region. We will now show this for N = 4 and
N = 5.
For the tetrahedron, we have the constraints

Since there are 2N -6= 2 independent variables, we will


,,
choose a and a 1 3 . It is now a simple matter to solve for
the region of integration:
FIG. 7 . A correct definition of a triplet of closed strings,
which satisfies the Jacobi identities without four-string and
higher graphs. In the end-point gauge, geometric string field
theory is thus cubic.
Missing region:
I
a 1 2 , a I 35 7 ,
a12+a135:7.
In the a ,,-a plane, this missing region corresponds to a
right triangle (see Fig. 9). This, in turn, can be mapped
where I labels all possible perimeters one can draw by onto the northern hemisphere of the Riemann sphere and
cutting a polyhedra into two distinct clusters of polygons. we find that the missing region actually occupies most of
Notice that this is a nontrivial constraint, which elimi- the complex sphere (see Fig. 10).
nates a vast number of possible polyhedra from the start. Notice that, inside the missing region, closed strings do
[This origin of this constraint comes from the nonclosure not interact at their midpoints. In fact, the overlap of
of the universal string group (USG) in the midpoint closed strings is less than 7 ,i.e., the missing region is sim-
gauge.] ply the square of the interpolating vertex.
Let us now write down the measure of integration for For N=5, we now have 2 N - 6 = 4 independent vari-
the arbitrary case. We find ables (see Fig. 11). The aij now satisfy

The first delta function simply fixes all external string Let us carefully choose our independent variables. We
lengths to be 2 7 . The next term enforces the constraint choose a12,a24,a13,a35.(A naive choice will actually in-
that all possible perimeters one can draw which divide clude unwanted dependent variables because of hidden
the polyhedra in half have a length greater than 2a. The
factor K comes directly from the elimination of the vier-

A
FIG. 9. The missing region for the tetrahedron graph. No-
tice that it has three edges, each boundary corresponding to the
configuration where strings interact purely at their midpoints.
The missing region thus smoothly connects the three Mandel-
FIG. 8. The splitting of a N-faced polyhedron into a left and stam channels. In general, the boundary of the n-faced polyhe-
a right polygon with NL and NR faces, respectively. The line dra is formed by the interface with the various Mandelstam
separating these two polygons is called the perimeter. channels of the n-string interaction.
NONPOLYNOMIAL CLOSED-STRING FIELD THEORY

FIG. 12. The labeling of a cube.


FIG. 10. The missing region mapped onto the Riemann
sphere. The missing region occupies most of the northern and
southern hemispheres, including the poles. The midpoint ables for the 6-point function. The choice of these 6 in-
scattering (where strings interact purely at their midpoints at dependent variables, however, is a bit tricky. For exam-
different interaction times) is given by the three disks which sur- ple, if we choose four of these variables to be the four
round the missing region. parallel lines that form a cube, we find that the sum of
these four edges is equal to a constant, and hence there is
a hidden relationship among them. In fact, most symme-
trical choices have hidden relationships among them.
identities.) Let us now solve for the dependent variables: We thus choose an unsymmetric set of independent
,,,
variables a ,,,a a 34,a36,a4,,a16. The resulting six equa-
tions can be solved, giving us

a26=2~-a16-a36-a46 ,

a23=a16+a36+a46-a12-a25

a 3 s = 2 ~ - 2 a 3 6 - a 3 4 + a 1 2 + a , 5 - a 1 6 - a 4 6,
(3.14)
a , , = 2 ~ + a- ~a ,~, - - ~ ~ , - a ~ ~ - a ~, ~
Setting the fact that all aij 1T , we now find the complete
description of the missing region: a 1 5 = - a 1 2 - - a 2 5 - a 4 6 + a 3 4 + a 3 6,
a,, = -a25 + a 3 6 + a 1 6 .

a,, i a 3 , 1a12+a,, < a 3 , + ~ ,


Missing region: (3.13) To calculate the missing region, we have to impose
a I 31 a 2 , I a 1 3+a3, <_a,,+.n , several constraints. First, we must set all aij L T, for both
independent and dependent variables. This constrains
the perimeters of all polygons with 2 faces (and simul-
Obviously, with little effort, we can now calculate the taneously, 4 faces). However, we also have the condition
missing region for all possible polyhedra. that all perimeters of polygons with 3 faces must also be
To illustrate the complications that may emerge at constrained. This give us yet another set of constraints.
higher levels, let us work out the missing region for the Restricting the perimeter of 3-faced polygons is
symmetric N = 6 case, i.e., the cube. In Fig. 12, we have equivalent to restricting the interior edges which make
the labeling of the cube. (See also Fig. 13.) We have 12 up the polygon. Thus, we must impose the constraints
variables aij, and 6 constraints, leaving us with 6 indepen-
dent variables, the correct number of Koba-Nielsen vari-

FIG. 13. In addition to the cube, this 6-faced polyhedron


FIG. 11. The labeling of a 5-faced polygon must also be added to the theory.
3740 MICHIO KAKU 41

where B ( s ,t , u ) represents the Shapiro-Virasoro model,


M ( s , t , u ) represents the contribution from the missing
region (where strings interact continuously along their
parametrization length), and A ( s ) , etc., represent the
naive contribution from midpoint interactions (where
strings can only interact via their midpoints).
Thus, we now have an algorithm by which to construct
the missing region: the boundary of the missing region pre-
By plugging in the value of the dependent variables into cisely connects with the d~jierentMandelstam channels.
these equations, we obtain the restrictions due to the con- We can summarize this discussion with the simple ob-
straint on 3-faced polygons which appear inside the cube. servation
Many of these constraints are actually redundant
These sets of constraints (restricting perimeters of 2-, aM = ( P I= 2.rr for all I ; (3.18)
4-, and 3-faced polygons appearing inside the cube) deter- i.e., the boundary of the missing region is determined by
mine the missing region totally for the cubic interaction. the constraint that all internal perimeters which bisect
The missing region is therefore the polygon equal 2 ~ .
Now let us draw the various polyhedra which satisfy
a,, 5 T , these constraints. In principle, there are a large number
a36 a 12 +025 of equivalent ways we can write down the same polyhe-
Missing region: . a 3 , +a 34 + I 2~ , (3.16)
dra, so let us standardize our results. The following rules
give us a simple way to eliminate redundancies and gen-
5a34+a36 7 erate higher polyhedra for arbitrary N.
(1) We will draw all polygons as flat two-dimensional
a 12 '46 7
objects. These flat diagrams are topologically equivalent
to the equipotential lines we found from the conformal
where we keep in mind that 6 of the 12 all's are depen-
map in two dimensions.
dent variables, and where the last four set of constraints (2) We will always take the largest polygon within the
come from restricting the perimeters of 3-faced polygons. polyhedra as the outer polygon, and inscribe all equipo-
In summary, the missing region is determined totally
tential lines within this polygon. This will eliminate two
by the constraints (3.4) and (3.7) on the perimeters which polyhedra which may look different but actually are the
bisect the polygon. The origin of this important con- same.
straint can be seen from several angles. (3) For simplicity, we will then delete the largest po-
First, it arises naturally from geometric string field
lygon completely.
theory when we make the transition from the end-point
(4) T o eliminate the remaining ambiguity caused by
gauge to the midpoint gauge. Since the end-point gauge
two or more polyhedra which have the same largest po-
is modular invariant to all orders in perturbation theory
lygon, we can distinguish between them by decomposing
(since it reduces to the light-cone theory), it gives us a
each polyhedron into the basic constituent polygons
convenient triangulation of moduli space. When we
which are contained within it.
change the vierbein and slowly go to the midpoint gauge, The advantages of these simple, standardized rules is
the triangulation of moduli space slowly changes, and that we quickly eliminate redundant polyhedra. More-
missing regions gradually open up. (This has been seen over, the resulting diagrams are exactly equivalent to the
explicitly using the Vax computer for the Shapiro-
Virasoro model, where the north pole is just a point in
the end-point gauge, gradually opens up in the interpolat-
ing gauge, and finally occupies most of the northern hem-
isphere in the midpoint gauge.) Since the end-point
gauge was modular invariant, the midpoint gauge gives
us a new triangulation of moduli space, previously un-
known to the mathematicians.
Second, the boundaries of the missing region can be
defined by looking at the intersection between the s-, t-,
and u-channel graphs for the 4-point function and the in-
tersection between the various s,] channels in the N-point
function. As first pointed out in Ref. 3, the s-, t-, and u-
channel graphs cannot be smoothly topologically de-
formed into each other if strings interact at their mid-
FIG. 14. The polyhedra for N=6. O u r rules are (a) the dia-
points. The boundary of the missing region thus con-
gram is two dimensional, (b) the largest polygon appears o n the
nects the three s-, t-, and u-channel graphs, where exter-
outside, and all polygons appear in the interior (c) we delete the
nal strings d o interact at their midpoints:
largest polygon. The advantages of this are (1) it gives us an
easy way to delete redundant graphs and (2) each graph corre-
sponds to a familiar Feynman graph from 4' theory.
NONPOLYNOMIAL CLOSED-STRING FIELD THEORY

FIG. 16. The polyhedra for N = 8 .

and where 0, and u i overlap with each other from


u i = c i j to u i = h i j (see Fig. 17).
Notice that the origins do not necessarily have to
match for the ith and jth strings. We will multiply the
vertex function with the usual projection operator at each
FIG. 15. The polyhedra for N = 7 . external leg,

usual, familiar Feynman graphs found in 4~~theory (see to ensure the independence of our result on the choice of
Figs. 14- 16). origin. (This constraint arises naturally from the U S G . ~ )
With these rules, one can fairly rapidly write down the T o calculate the tangent-space contribution to the
polyhedra for N = 9 and N = 10. structure constant (i.e., the ghost sector vertex I V,, ) ),
we simply replace the X with the usual b and c ghost
IV. MULTIPLICATION RULES fields, with small corrections.
Notice that we cannot use the usual definition of the
Now let us introduce the multiplication rules for fields. multiplication of fields. If we did this, then (YE) =0.
For brevity, we will simply postulate what they are, but However, we can take Y to be Grassmann odd, and A to
we emphasize that they can all be derived from first prin- be Grassmann even. But we must use unorthodox sym-
ciples from geometric string field theory. They are, in metry rules for the vertices to achieve a nonzero answer.
fact, just Clebsch-Gordan coefficients for the USG. Let us now write down the rules for multiplication of
These postulates thus arise from a unifying principle. fields:
Let us first define the base-space structure constant.
Let the edge of the ith polygon within a polyhedra (now
defined in real physical space-time) be labeled by X i ( u i )
where 0 < u i < 2 ~ r .Then let the structure constant for the
USG (in the midpoint gauge), be represented by
N

i=l j =1

where
Q11 . . = 7?I7 . . Q ( u iLJ- b . . ) Q ( ~ i j,- ~ i )
where
+1
r]..=
lJ
I if a i , f 0
0 otherwise ,
,
FIG. 17. The parametrization for two strings which share a
common boundary along the edges of the polyhedra.
3742 MICHIO KAKU 41

perimeter is 2a. These double vertical barred vertices, we


will see, lie precisely on the boundary of the missing re-
gion as given by (3.18).
Now let us take the variation of the action. We find

where by the angular brackets we mean taking all possi-


ble permutations of the various indices. Notice that, with
the last definition, the product of several Y fields does not
vanish. (See Refs. 6 and 7 for the ghost factors.)
Finally, we would like to introduce the notation 1 1 .
The double vertical bars are formed when we join two
clusters of polygons such that the mutual perimeter
equals 2 ~ i.e.,
; we are at the boundary of the missing re- Notice that this variation is equal to zero if
gion. Thus, the measure for the double vertical bars, in-
stead of being (3.8), is now defined with

Thus, the notation


In this form, this identity may not be very transparent, so
('4'111'42) let us rewrite it in the form

means that the mutual perimeter between Y , and Y2


(which represent clusters of polygons) is taken to be 2a.
Finally, we note that there is more than one polyhedra
at each value of N which satisfy our rules. In fact, we
have 2 different ones at the 6th level, 5 different ones at Notice that everything is encoded within the numbers C,"
the 7th level, and 14 different ones at the 8th level (sub- and c(n )i. In this form, it is easy to see that the vertex is
tracting out those which can be reached by trivial rota- not BRST invariant:
tions). N
2 Q,IV)#O. (5.7)
i =1

Instead, it generates all possible vertices with double vert-


Now, we have carefully made all the definitions and ical bars, i.e., all possible slicings of a polyhedra into left
possess all the tools to write down the action and right clusters of polygons such that the perimeter is
2T.
Now it is clear why we had to impose the rather
strange-looking constraint, that all perimeters must be
greater than 2 ~ .
Partially solving these equations, we find
which we want invariant under

Notice that this already gives us the simple constraint, to


We must stress again that each ( Y n )i represents a sum all orders:
over all possible n polyhedra. Thus, this power series is
actually more than a power series in n. It is also a series
within each n, i.e.,
For practice, let us now work out some of the lower
values of the coefficients. For the cubic term, we have

Now let us take the variation of a polyhedra by a field


with parametrization length 2a: If we let g be the coupling constant, then we can set

Notice that we have used the double vertical bars to


denote the fact that 6Y (which represents polygon) as expected.
merges with the rest of the vertex such that the mutual For the tetrahedron graph, we have
NONPOLYNOMIAL CLOSED-STRING FIELD THEORY 3743

2 cated, because we have five possible polyhedra at this lev-


9' p=-* c;=l. (5.12) el. Once again, the system is highly overconstrained,
a4=7-9
cI c:, yielding
For the N = 5, we have
c ,-
,-- C(7Il - -
5 ---- ~ ( 7-)-
3 -- ~-
30 ~ ( 7 1-
-, -
- 6-~(714
~ ~ ( 6 ) 7~ ~ ( 6 ) 7 ~~ ( 6 ) ~
14 ~ ( 6 ) 28
a5=- c;=+c;. (5.13)
- 9 c(7), - 6 ~ ( 7 1 ,
By now, it is easy to see the overall pattern that is devel- 7 ~ ( 6 ) 7 ~~ ( 6 ' ) ~
oping. Notice that all a, and fin- 2 can be written entire- (5.21)
ly in terms of C;. Even the higher C: can be written in
,- 25
-c,------ c ( 7 ) 1- 75 ~ ( 7 1 1
terms of C:. 28 ~ ( 6 ) 7~ ~ ( 6 ) ~
We now present the general cse.
Let us define a modified factorial consisting totally of
the C;:
This, in turn, yields

where we set C: =2. Then we find Notice that this overconstrained system again confirms
the correct relationship between the C s :

The calculation of all the coefficients at the N = 8 level


(see Fig. 16) is not complete; however, we can also
confirm the fact that these overconstrained equations are
Notice that we have now determined all possible self-consistent and uniquely yield the relative weights for
coefficients in terms of C;. Determining the value of this c ( n I i for 81,82,811:
set of coefficients is the most difficult step in the entire
calculation, because it also requires a complete descrip-
tion of the coefficients c ( n ) i multiplying each n-faced po-
lyhedron. and
The interesting feature of the calculation of the C; and
the c ( n Ii is that the system of constraints is highly over-
determined, creating an intricate web of interlocking con-
ditions. Thus, there are a large number of checks to tell Thus, we find
us when we have made a small error in our calculations.
For the case of N = 5, we find

It is now a simple matter to find relations among cer-


tain infinite sequences of polyhedra and deduce what
Notice that this result obeys the redundant constraint must be the final answer.
Let us define the polyhedra n , to be the N-faced po-
lyhedra which generalizes 7, and 8,, i.e., two ( n -2)-
sided polygons which face each other connected by n -2
For N = 6 , the situation is more complicated because squares. Let us consider the relationship between the n
we have two polyhedra at this level, which we call 6, (the
cube) and 62 (see Figs. 12 and 13). We have
+
and n 1 , polyhedra. We find

Notice that this system is overconstrained, giving us


This gives us

and, as an added check on our results,


We also have a reconfirmation that

For N = 7 , the situation becomes much more compli- C,"-,=(-l)"-'+(n -1)C; .


3744 MICHIO KAKU -
41

Similarly, other infinite classes of coefficients can also Finally, we can now extract out the coefficient which
be calculated in this straightforward way. Thus, it is not will correctly normalize the overall perturbation series.
hard to deduce the general rule for all possible polyhed- As we have said, dividing out the symmetry factor s ( n ),
ron. for each polyhedron n, conveniently divides out all the
For example, let us rewrite the equation for the other symmetry factors as well. This symmetry factor is
coefficients at the 7th level in a more transparent fashion: 2( n - 2 ) (except for the case of n = 4 and n = 6, because of
their unusually high symmetry). We thus define

We immediately recognize the pattern: each numerical


coefficient is simply the total number ways we can rotate
the polyhedron around various axes and obtain the same
polyhedra. This is also verified at selected values for
N = 8 and also for the general case. Let us call this sym-
metry factor s ( n Ii for each polyhedron. Then we have

Our final result is therefore


Notice that the symmetry factor s ( n 1, is 2(n - 2 ) for n
greater than 6 . [We set s ( n ), = 1.]
Notice that we can also write a key identity
~(7)~~(7)~=~(7)~~(7), (5.32)
for i,j = 1,2, . . . , 5. Thus, if we divide out by the symme- This is our final result.
try factor for the n , polyhedra, we have a simple normal- Lastly, let us show that the algebra closes when we
ization for the series: take two distinct variations in succession (if we assume
that the equations of motions are satisfied). We find

Now take a second variation

5
=c(n),s(n), s ( n ) , 7 ' ( ~ " ). ~ (5.33)
i =1
Using this, let us now rewrite the result for the entire Working out the details, we find
series. We first write down the values for the various a , :

where

For arbitrary n, we have Finally, let us give a heuristic "proof' of (5.6) which
will make basic structure of the identity intuitively obvi-
ous. However, more work has to be done to actually
prove it to all orders in perturbation theory.
Let us now write down the final power series, dividing First, we begin with the observation that the BRST
out by the overall symmetry factor of the n , polyhedra operator Q, for the rth string acting on the n-faced po-
(which, as we have seen, conveniently divides out by all lyhedra vertex can always be expressed as a line integral
the symmetry factors of all other polyhedra). We find from 0 to 27r:

Furthermore, we can shift the value of u by a small imag-


inary quantity i7 to become the complex variable p. The
value of Q is not changed by this equation. Thus, the
operator expression
NONPOLYNOMIAL CLOSED-STRING FIELD THEORY

Then we calculate

can be rewritten as the sum of line integrals over each


face:
Fortunately, we know all Jacobians required to make the
transition from aij as independent variables to PI as in-
dependent variables.
Notice that alas,, makes contribution to a / a P I even if
Because the vertex obeys the usual continuity conditions the perimeter is quite distant from the ith and jth edge.
(4.1) across the various boundaries of the polyhedra, we Because of the measure (3.8), the vertex must be in-
can use a trick used by Mandelstam: cancel all contribu- tegrated over all independent variables within the aij,
tions from the line integral except for the regions sur- which can be converted to integrals over various P I . This
rounding each interaction point, where there are poten- leaves us with terms such as
tial infinities. All other contributions outside the vertices
vanish. Thus, we now have line integrals which just sur-
round the corners o r vertices of the polyhedron, labeled
by the index i:
But this total integral leaves us with precisely the value of
the integrand over the boundary of the missing region,
which satisfies PI = 2a.
T o evaluate the residue at the interaction point, we now Thus, in the measure (3.8), the terms
make the transition from the p plane to the z plane by
making a conformal transformation. Calculating the
residue of the poles due to the map (2.1) is a bit tricky, now converts to the term
because of the presence of Riemann cuts, singularities,
discontinuities, and because yP and V, each transform
under conformal transformations according to their con- i.e., surface integral over the boundary of the missing re-
formal weight. gion.
Although the results are rather complicated, we can Now compare this with the definition of the "double
use another trick. Notice that we sum over the rtcom- vertical bars" that we introduced in Sec. IV. Notice that
ponents of the p index. When this summation is carefully the definition of the double vertical bar was precisely the
carried out, taking the ghosts into account, only the insertion of 6(PI - 2 a ) into the measure.
imaginary part of L, survives. We notice that the value In summary, the entire effect of x Q r acting on the ver-
of imaginary part of V, at the interaction point is just tex function was to bisect the vertex function into two
halves, such that the perimeter of the boundary is 277.
This is precisely what happens in (5.6).
As a n independent check on the correctness of this ap-
But this is precisely the reparametrization operator. proach, let us use this method on a simpler problem, the
Thus, we can easily show that the dominant terms which BRST invariance of the four-string open-string interac-
contribute to the poles can be reduced to simple sums of tion. As before, following Mandelstam, we can convert
reparametrizations defined at the interaction points. We the sum of Qr on the vertex function into sums of line in-
can therefore make the substitution tegrals over the joining point u , . However, because the
line integral for open strings defined over 0 < u < a is ac-
tually taken over twice that distance, i.e., 0 < u < 2 a , we
find two contributions to the line integral around the
when the reparametrization operator acts on the ijth joining point, one for a, and the other for the complex-
boundary. (There are other ghost contributions, but conjugate point o f . These two terms will add to give us
these go toward making the measure come out correctly.) precisely L, -L -,, which is the reparametrization
Thus, the action Q on the vertex function will produce operator. T o see this, let us first define
terms such as

(The ghost field c is roughly the same as our field y.)


Then
However, this is not the correct form for the vertex.
We want a term which is proportional to the derivative
with respect to all perimeters P I .
However, as we have seen, the polyhedron only has
2n - 6 independent variables. Because the measure is ex-
plicitly known in (3.81, we can always change our basic
set of independent variables so that PI is one of them.
3746 MICHIO KAKU !!

Notice that the last step was possible because this term We need the action of the BRST operator on the vertex
L,, -Lot has a Fourier expansion in terms of L, -L -,, function:
which is just the reparametrization operator. ( - 1 ) " ( Y " ~ ~ Q+
An) (QY~~Y"-'A)
But since we now must integrate over the measure d p n -2
of the four open-string interaction, the integral term sim-
ply creates a total derivative, which results in a boundary
+2
cp"(Y" -pIIYpA) =o ,
p=l
term where the double vertical bars signify that the boundary
of the perimeter separating the two clusters of polygons is
given by 27~.After an extended calculation, we find

The vertex function of the four-string interaction is not


BRST invariant, but instead produces total derivatives,
which are in turn equal to the vertex function defined at
the end points of the interaction uinitial
and a,,,,.
In sum, the simple reason why (5.6) is correct is that
the operator Q contains the reparametrization operator and, finally,
L, -z-, . When carefully calculating the residue of the
an=--
2,-1
gn-2 ,
poles at the interaction point and summing over the posi- ~(n)~c(nj,
tive and negatiue oscillators, only the imaginary part of
L, survives, so Q basically reduces down to this where the factorial with an overbar is defined in (5.14)
reparametrization operator. But this means that we have and where s ( n Ii is a symmetry factor, the number of
a total derivative, which in turn means that we must sit ways the ith n-faced polyhedra can be rotated and yield
on the boundary of the missing region. Thus, Q acting on the same polyhedra. It obeys the important result
the vertex function must equal a sum over all possible
boundary terms of the missing region.
The rest of the proof requires combinatorics, which we If we define
will present in another paper.
(Yn)sym=-- (Yn)
s(n),
VI. CONCLUSION then we find our final result for the nth level is
Let us now collect all our results. We have defined our
action as power expansion
In conclusion, let us make the following remarks.
(1) We have, using standard BRST methods, derived a
nonpolynomial closed-string field theory. We have calcu-
lated the coefficients to eighth order, where there are
which we force to be invariant under many consistency checks. Our coefficients seem to differ
a bit from those of Kugo and collaborators (Ref. 8).
(2) There is also a second derivation of the action, us-
ing geometric string field theory, which we present in
This expansion is complicated by the fact that there are another paper.7 This second derivation, in fact, is how
more than one n-faced polyhedra, which we label by the we first discovered the nonpolynomial action in the first
index i, i.e., place. In the geometric approach, the gauge-invariant
theory is actually cubic, but contains a vierbein. When
the parametrization length of the string is gauge fixed to
be 27~,then an infinite number of instantaneous Coulomb
where interactions are created, which precisely equal the po-
lyhedra found in this paper. Thus, a polynomial theory
can be converted to a nonpolynomial theory, and vice
versa, i.e.,
By setting the variation of the action equal to zero, we
wish to calculate the three sets of unknowns:
a, ,p,, c ( n );. This is complicated by the fact that the ver-
tex function is not BRST invariant, but in fact generates
a large number of terms which are total derivatives; i.e.,
they only exist on the boundary of moduli space of the
a,,. We thus find a intricate web of self-consistent condi-
( v3)geometric = [ ( Y3 )end point
(Y3+Y4+Y5+
9

. lmidpoint
tions, giving us many checks of our results at each stage The reader is advised to consult Refs. 6 and 7 for more
of the calculation. details of the geometric theory.
41 NONPOLYNOMIAL CLOSED-STRING FIELD THEORY 3747

(3) The question of higher loops is presently under in- organizers of the 25th Winter School of the University of
vestigation. Wroclaw, 1989, Feb. 20-Mar. 5, 1989, where these re-
sults were first presented. We wish to thank the
ACKNOWLEDGMENTS US/Japan Cooperative Science Program, Grant No.
NSF-INT-8715626 as well as Grants Nos. NSF-PHY-
The author wishes to thank J. Lykken and L. Hua for 86615338 cUNy-FRAP-6-669347, and DE-FG02-
extensive discussions. We also would like to thank the 90ER40542.

'permanent address: Physics Department, City College of the (1989). See also M. Saadi and B. Zwiebach, Ann. Phys. (N.Y.)
City University of New York, NY 10031. 192, 213 (1989).
'See M. Kaku, Introduction to Superstrings (Springer, New 6 ~ Kaku,
. Report No. CCNY-HEP-86-14, 1986 (unpublished);
York, 1988), and references therein. Report No. CCNY-HEP-87-3, 1987 (unpublished); M. Kaku,
2 ~Kaku . and K. Kikkawa, Phys. Rev. D 10, 1110 (1974); 10, Int. Mod. Phys. A 2, 1 (1987); Phys. Lett. B 200, 22 (1988);
1823 (1974). Int. J. Mod. Phys. A 5, 659 (1990);see Ref. 1, Chap. 8.
3 ~Kaku,. Phys. Rev. D 38, 3052 (1988); M. Kaku and J. Lykk- 'M. Kaku, Osaka Report No. OU-HET 121, 1989 (unpub-
en, ibid. 38, 3067 (1988). lished).
4M. Kaku, in Functional Integration, Geometry and Strings, 8 ~ Kugo
. and K. Suehiro, Report No. KUNS 988 H E (TH)
edited by Z . Haba and J. Sobczyk (Birkhauser, Boston, 1989). 89/08, 1989 (unpublished).
5 ~ Kugo,
. H. Kunitomo, and K. Suehiro, Phys. Lett. B 226, 48
FIG. 9. The missing region for the tetrahedron graph. No-
tice that it has three edges, each boundary corresponding to the
configuration where strings interact purely at their midpoints.
The missing region thus smoothly connects the three Mandel-
stam channels. In general, the boundary of the n-faced polyhe-
dra is formed by the interface with the various Mandelstam
channels of the n-string interaction.

You might also like