Professional Documents
Culture Documents
Michio Kaku
Physics Department, City College of the City University of New York, New York, New York I003I
(Received 21 April 1988)
We present an entirely new approach to closed-string field theory, called geometric string
Iield theory, which avoids the complications found in Becchi-Rouet-Stora-Tyutin string field theory
(e.g. , ghost counting, infinite overcounting of diagrams, midpoints, lack of modular invariance).
Following the analogy with general relativity and Yang-Mills theory, we define a new infinite-
dimensional local gauge group, called the unified string group, which uniquely specifies the connec-
tion fields, the curvature tensor, the measure and tensor calculus, and finally the action itself.
Geometric field theory, when gauge fixed, yields an entirely new class of gauges called the interpolat-
ing gauge which allows us to smoothly interpolate between the midpoint gauge and the end-point
gauge ("covariantized light-cone gauge"). We can show that geometric string field theory repro-
duces one copy of the Shapiro-Virasoro model. Surprisingly, after the gauge is broken, a new closed
four-string interaction emerges as the counterpart of the instantaneous four-fermion Coulomb term
in QED. This term restores modular invariance and precisely fills the missing region of the complex
plane.
'
C~C,
USG: (2. 1)
C~C .
We will devote the next two sections to defining an expli-
cit form for the generators of the USG and writing down
its irreducible representations. Our claim is that the
string field theory can be uniquely specified by gauging FIG. 2. Three unparametrized open strings in space-time
the universal string group when we follow our basic stra- which appear in the Jacobi identity. Notice that C& is conjugate
tegy (1.1). to C„which in turn is conjugate to C3, but C& is not conjugate
For each open-string C, we associate an abstract group to C3. Thus, the Jacobi identity is satisfied because 1 —1+0=0.
generator I.c. Notice that C is now an infinite index, la- Notice that there is no such thing as a parametrization mid-
beling an point.
Ro
Ri-2
transfinite, nondenumerable number of states. cannot simply generalize to closed strings.
For each open triplet, we can define the number: (2) Antisymmetric c number -structure constants do not
exist for the closed strin-g case On. ly symmetric c-number
+1 for triplets, structure constants are possible. If one tries to construct
a constant tensor which is antisymmetric in two strings,
fc c c = '
I 2 3
1 for antitriplets, (2.2)
then one can always smoothly deform the strings into the
0 otherwise . opposite order. Thus, the constant tensor must be sym-
metric, or zero.
Let C represent the string C with reversed orientation. (3) The Jacobi identities are not satisfied. For example,
We now define the structure constants of the (open-) in Fig. 3 we have a candidate for a triplet of three closed
string grogp as strings with equal parametrization length. However, a
C3 straightforward application of the Jacobi equation for
(2.3)
2 three closed strings based on Fig. 3 yields a new diagram
as in Fig. 4. This diagram contributes the following to the
fc, c, = fc, c,
C3 C3
. (2.4) Jacobi identity:
The algebra for the (open-) string group is therefore
1+1+1~0 (2.9)
C3
SG:lLc, Lc, l=fc, c,Ls (2.5)
which is not acceptable. The Jacobi identity based on
It is easy to show that the Jacobi identity is satisfied for Fig. 3 will always produce tetrahedron configurations
this algebra: which are symmetrical in all three strings. Thus, no
matter how we might tamper with the algebra, me can
(Lc, lLc, Lc, ]]+(perm)=0, (2.6) never satisfy the Jacobi identities with Fig. 3 as our vertex
What is needed, obviously, is a rather radical departure
C4
fc, c,fc,sc, +fc, c,fc,sc, +fc,4c,fc,sc, =0 .
4
(2.7) from the open-string case. In order to surmount these
fundamental diSculties, we must make a fundamental
To show this explicitly, take the configurations for the
strings 1, 2, and 3 as shown in Fig. 2. A careful applica-
tion of the identities shows that (2.7) can be written as
1 —1+0=0 (2.8)
proving the Jacobi identities for Fig. 2. We emphasize
that the Jacobi identities close independent of any param-
etrization or any background gravitational metric.
We call (2.5) a transfinite Lie algebra
Let us now generalize this discussion to the case of
closed strings. Immed&ately, we see severe problems
which will require nontrivial modifications for the
closed-string case. There are three reasons why a naive
generalization of (2.5) is not possible.
(I) Antitriplets do not exist for closed strings. Any trip-
let, in fact, is its own antitriplet. By rotating a triplet of FIG. 3. The arrangement of three parametrized closed
closed strings, we can always smoothly deform it into its "
strings in the "midpoint gauge. The Jacobi identity ca@not be
antitriplet. Thus, the formalism invented for open strings satisfied for this definition of a triplet of strings.
38 GEOMETRIC DERIVATION OF STRING FIELD THEORY FROM. .. 3055
where fc c c 2 3
is symmetric in 1 and 2. If C is the loop When written out explicitly, this identity becomes
space string C with reversed orientation, then we define AC4 AC5 AC4 AC5 AC4 AC5
C3
fc, c,fc, c, +fc, c,fc, c, +fc, c,fc, c, =0 (2. 17)
fcic2 =fcic2C3 . (2. 13)
The proof of this identity is not hard, but requires a care-
Finally, our structure constant is defined as ful analysis of the location of the fermionic c number i, 23.
In Fig. 6, we see several configurations for which we must
f =fC|C2
C3
Ci C2 Ci n C2 rl C3 (2. 14) calculate the Jacobi identity. Let the intersection points
be represented as i, and ib. Then Fig. 6 contributes the
where i is a Grassmann number defined at the end (join- following to the Jacobi identity:
ing splitting) point of three strings:
6(a): i, ib+ibi, +0=0,
end point:—C, AC2 flC3 (2. 15)
6(b): i, ( ib)+0+— ( i, )( ib)=0, —
—
Notice that our closed-string structure constant is now
fermionic, as desired. 6(c): i, (ib)+(i— b )( i, )+0=0, —
The whole point of introducing this Grassrnann-type (2. 18)
construction is to satisfy the Jacobi identities. Let us
A
6(d): ( ib )( ib )+( i, )(—
A
—
i, )+—0=0,
now calculate the Jacobi identity for this structure con-
6(e): ( i, )( ib ) + (—i b )(—ig )+ 0 =0— , —
stant. The key identity we must show is
6(f): ( —~. )( —ib )+ 0+ (™i,)( —ib ) = 0 .
[LC, I LC, LC, I ]+ I.LC, I LC, Lc, l i
Thus, all the Jacobi identities are satisfied.
+lLC ILc Lc )i=0. (2. 16) In summary, we have now shown that a closed-string
3056 MICHIO KAKU 38
(a)
f™,
E. ~3
(c)
(e)
FIG. 6. %'hen calculating the Jacobi identity for closed triplet configurations given by Fig. 5, we find considerably more graphs
than for the case of open strings, given by Fig. 2. All these contributions cancel identically to zero, but only if an anticommuting c
"
number is placed at the "end point, i.e., the joining-splitting point where three strings meet. In physical space, the string algebra is
therefore not a true Lie algebra. Only when we introduce parametrizations does the algebra become a traditional Lie algebra.
generalization of the string group exists, but the algebra III. INVARIANTS FOR THE STRING GROUP
is based on anticommutators and the structure constants
are Grassmann numbers. Now that we have defined the string group in loop
It is essential to notice that the open and closed strings space, let us now define invariants. Let us first discuss
are, in some sense, opposites. The open-string algebra representations of the open-string case. Let the contra-
(2.5) closed with triplet configurations given in Fig. 1, so variant states e & represent an abstract set of basis
I
that the midpoint gauge has vierbein measure det e„"~ states for the algebra. Let its covariant dual be represent-
ed as (ec . We define the scalar product as
I I
In this representation, we can write an explicit form for the string algebra A is now also fermionic. For example,
the generators: the variation of covariant and contravariant fields is
given by
(3.4)
50 '=A 'f '
3 2
Let us now define the group element as (3. 14)
~C3
54c, =
C2
O=e
A Lc
(3.5)
A fc c 4c,
and its matrix elements in this basis as
where Pc and P are independent fields.
The crucial point is that two invariants are possible un-
&ec, ~0~e '&=Oc', (3.6) der the (closed-) string group:
bination
(3.7) IV. UNIVERSAL STRING GROUP
We have now shown that P and P are invariants under X —Y if Y„(cr)=X„(cr)+e(cr,X)X„'(o ) . (4. 1)
the (open )string gr-oup, independent of any parametriza
[Notice that e(o, X) is a function of the string state X,
tion of the string, parametrization midpoints, or indepen
meaning that our gauge group is a local one. Thus, the
dent of any background classical space time geomet-ry
reparametrization of a string C depends on its particular
Let us now generalize these results for the closed-string parametrization X. Each string therefore has its own in-
case. We begin by defining the action of the group gen- dependent Virasoro group of reparametrizations. ]
erator on the basis states for closed strings: Let us now associate Lx with L& in the following
LC, e
' = 'c, f e fashion:
I c,
& I
the universal string group, let us now define how to make mation convention:
a transformation on the parametrized string:
cr~o+e(o ) . (4.3)
A 8 = f
0
do A(cr)8(o)= g A "8 (4.5)
Let us now introduce Fourier components (keeping in where A has weight n and 8 has weight 1 — n, such that
mind that this is a bit dangerous; the Fourier decomposi- the product has weight 1, so that A 8 is invariant un-
tion of a string is actually a gauge-dependent choice in a der global reparametrizations.
theory which is independent of parametrizations; howev- Now let us make this reparametrization on a covariant
er, we will give the Fourier decomposition of our fields in string functional:
order to make contact with "rigid" BRST theories once
we fix the gauge). We define fiyx= f '"dae(a)X
nor no =& ~ -4x (4.6)
e(0)=e = g e'"cos +e""sin
tin
n=1 a a
where
inrr—/a
&
y
ll
&ne (4.4)
a. =X~ a„. (4.7)
where we emphasize with the symbol e that 0. labels a (and there is no summation or integration over cr). To be
specific representation of Diff(S, ). We will use the sum- specific, let us choose the following parametrization:
g &,
n~O
r (4.8}
5
+ B„„sin no
1 ncr r, . 1 —ina/a
e me/a+
1
Bo„+ 2n B„„cos OP P a P
e
2 n~O
2&n
L =
2&a
g(L„L„).
n
—
f =5(p —o )5'(co —p)+n5(co —o )5'(p —o ), vanish at the end points and then make
dependent variables. This corresponds
a and 0 into in-
to making the
where the weight n =2. The second line of (4. 14) de- decomposition:
scribes how the generators of the string group transform
under reparametrizations. It shows that the universal Diff(S, ) =D S, Diff(S, } (4. 16)
string group is the semidirect product of the reparametri-
where D is the dilatation subgroup which rescales the pa-
zation group and the string group. The third line of (4. 14)
rarnetrization length, S, is the angular rotation group,
describes the string group itself, where we will define the
and Diff(S, ) is the subgroup which preserves the pa-
structure constant shortly.
In general, the irreducible representations of the USG rametrization length but changes the internal parametriz-
are unknown. However, we have been able to identify ation of o. Then the reparametrization e and the
three. reparametrization operator 8 can be Fourier decom-
(1} The adjoint representation of the string group, posed into components as follows:
represented as Px and P . Thus, the field functional e = I5a, e"),
/[X] is not a scalar, but a vector transforming as the ad-
joint representation of the string group. 5
(2) The set U of Verma modules, represented by the set a cr— a', 8„—3
of states which transform like the products of raising
operators in the universal enveloping algebra: where we have explicitly isolated the dilatation part of
internal parametrization as well as rescale the overall ferred one in order to close the algebra. Thus, we do not
string length 2m. a. This is the formalism we have used in have the freedom of choosing all parametrization lengths
(4. 13}. However, this is at times not the most convenient equal at the very beginning of gauge fixing. As a conse-
parametrization. If, for example, we set e(0)=e(2m. a) quence, the second formalism based on (4. 16) is preferred
=0, then the rescaling o. ~o. +o5a/a is incompatible over the first for the closed-string case.
with this boundary condition unless we take the boun- Let us now write down an explicit construction of the
daries an infinitesimal distance away from 2m. a. Also, the structure constants of the universal string group. We be-
Jacobian of the transformation from one X to another pa- gin with the pararnetrized vertex of Fig. 3 where the a,
rarnetrization is infinite under rescalings. Similarly, the are arbitrary. Let a, be the pararnetrization length of the
Jacobian from one parametrization X to another which is line which is common to the ith and jth string. Since the
rotated by an angle L9 is also infinite. parametrization lengths of each of the closed strings can
The second parametrization we can use is to set E(cr ) to be arbitrary, we have
3060 MICHIO KAKU 38
where
~]2 @ o]) Written out in detail, this transformation rule (4.24) can
]2
be expressed as
~]3=~(2~ ~3 I o] )@o]—a]2 }
=g f do e(o)f
~
5$ ~Pl~
}Pp. (4.26)
~23 @a23 o2) IPl
(4.20)
b12
—2mu2 M ~
lengths are all fixed. The structure constant for the where ~ tangent„t by definition satisfies
Valapa3 &„„s,
group can now be written as the gauge-fixed vertex with
the insertion of an extra vector field which allows us to P nj 0&o,n&2~a, (L a. ~ijL].. —a } I Va]a2a3 &
tansent =0
change the parametrization at will: i &
anticornmuting factor i in our algebra, we will find it con- L „, and then annihilate them on the other Verma
venient to generalize our algebra so that the string gen- modules. After a sufficiently large number of reflections,
erators are Verma modules themselves. We wish to re- all L„disappear, leaving the Clebsch-Gordan coefficient
place Lz with Lz. To do this, we must define how Ver- as the only numerical factor.
rna modules transform under the USG. Consider the Our final form for the algebra of the universal string
identity group is
L„~ e &=f p eP&, (4.22) [L,L ]=f" L
..Lrl=fxvf
~
where
shove
fL„directly
is a constant. For positive
into the Verma
or negative n, we can
module. Because the
USG: [L. ]=~ a.
L. L. +~ fP. ~ (4.29)
commutator of the L„'s produces more L„'s and con- [Lx 'L5 .
38 GEOMETRIC DERIVATION OF STRING FIELD THEORY FROM. . . 3061
Q2= 7
2al 2a2
2 2 —2 V. TANGENT SPACE AND CONNECTIONS
a&+a2 a3
a3= —a3, bl ——
The second step in our basic strategy (1.1) is to intro-
2al
(4.30) duce connection fields, but first, let us say a few words
al2 —a22 —a33 about the necessity of a tangent space in our theory. For
—2a3 the case of four-dimensional Dirac spinors interacting
with gravity, it is well known that a tangent space is re-
b =at2 —ap —
2
2
a =a3, 2
a3, c =a22 . quired because there are no finite-dimensional spinor rep-
Notice that this map has the same structure as the con- resentations of GL(4). Thus, the Dirac spinor
forms under local Lorentz transformations as
trans- f
formal map found in the open-string case in the interpo-
lating gauge. The difference is now we must consider the 5$ =[a""(x)tr„]$P, (5. 1)
entire complex z plane, not just the upper half plane.
To show this mapping has the correct analytic struc- where cr„„ is a spinorial representation of the Lorentz
ture, we can show by a direct but tedious calculation that generators. However, the spinor transforms under gen-
eral covariance as
dp (z —z ) (z —z )
—1) (4.31) 5$ =e"(x)B„Q' . (5.2)
dz z(z
where zp and zp are the roots of az +bz+c=0. This The total Lorentz generator is actually the sum of the x-
shows that the points z = 0, 1,
(x) correspond to the dependent Lorentz generator and the Lorentz generator
three closed strings at infinity in the p plane, while zp and for the tangent space:
zp correspond to the points in the p plane where the three
Mpv =X+v X vpp+ (trpv)p (5.3)
strings join.
Given the conformal map, it is possible to construct where o.„„are Dirac matrices which transform like the
the coefficients of the Neumann function N „which ap- generators of the Lorentz group.
pears in the vertex function. We can always write them as Thus, the spinor g transforms differently under two
— &[ ] —"&['] local gauge groups:
.. 2
mn
mn(2@i ) O(3, 1) spinor,
general covariant scalar . (5.4)
(4.32)
These Neumann coefficients N„"' satisfy a large number In addition, we can introduce the Dirac matrix which
of identities which are tabulated in Ref. 8. They can be operates in the tangent space. Let e ) be a spinor state I
written explicitly in terms of a power expansion with within the universal enveloping algebra of O(3, 1). Let y"
known coefficients. The final form of the vertex function be some vector operator. Then
is then just given by the product of left- and right-moving
oscillators: (r").it= e. & I
1'" I
ep& . (5.5)
3062 MICHIO KAKU 38
If S represents the set of spinor states, and V represents dependent part and the tangent space contribution. Let
the set of vector states, then the Dirac matrix can be M represent the generators of the full Diff(Si ), not just
represented as the Clebsch-Gordan coefficient for the fol- its reparametrization subgroup. Then
lowing tensor product decomposition:
M =8+L (5. 11)
Ug S~S . (5.6)
The commutator of M with itself must reproduce the
A similar situation applies for the string theory. The usual Virasoro algebra:
field Px cannot transform in X space under the full
Diff(S, ), as required by our first principle, because it is [M, M j=f M +(D —26) (5. 12)
impossible to form a representation of DifPSi) with a (When we choose a specific representation for the tangent
function of X„(o ) alone. For example, it is well known space generators, we will find that the total Virasoro gen-
that the combinations erator closes only in 26 dimensions. D comes from the
1 base space, while — 26 comes from the tangent space. )
4m Finally, let us introduce a Dirac-type matrix which
operates within the tangent space. We define the Dirac-
1 cine la type matrix as follows:
2m. a
(5.7) (e ~ep)=(y ) p (5. 13)
~y
. 1
— t
( )
i 8„—
4a X&~ which is the Clebsch-Gordan coefficient for the tensor
oo
product decomposition:
&incr/a
27Ta
V(3) S~V . (5. 14)
form a representation of the full Diff(S, ), but not X by it-
self. In general, functions of X„(tr) can only transform Notice that (y ) p is totally specified as soon as we define
under the smaller o reparametrization groups Diff(S, ) the transformation properties of y . For example, if we
generated by
and L„separately.
L„L„, whi—
le Diff(S, ) is generated by L„ define y to have weight — 1, then
~(+)=V(+)V+) we can define the measure and make contact with the
PO'
CJ PCT
tangent space. We need the vierbein to define a measure
'2
because the differential and derivative transform as
—
i d„+
1
X„'
4m
„eincr /a ax~
(6.4}
(6. 1)
PO d~ VP
a'-'= v'-'v'-' BX"
CT PO' PO'
The vierbein must have double the number indices ap-
—i(}„—1 X„' pearing in the coordinate. Since the coordinate has two
4m indices X", then the vierbein must have four indices:
e„"i'. The lower o index transforms under local Diff(S) )
e ina/a repararnetrizations. If we describe the space of all
(2ma) parametrized strings as "base space, then the lower e "
where we sum over the Lorentz index but not a. . Let us index transforms under reparametrizations of the base
now define space.
However, the upper p index transforms according to
V(+) g(+)+uL (+) local Diff(S) ) in the tangent space. To summarize show
the two spaces transform in the vierbein, let us write
We wish the cornrnutator between two covariant deriva-
tives to vanish: Base space: Diff(S) )
Z= DA D exp i X
~cr — 55 ~
5
5
p~n where the functional integration over P sums over all
possible parametrizations of the loop space strings.
such that Now that we have written down the unique action
which satisfies our two original principles, our next task
&=y v: , y—
5
+y 5
5
+yy"v „, (6.8) is to fix the gauge. Notice that our action
variant under
is actually in-
(7.8)
Written out in full, this equation reads
where 1F p&=F p1e &. However, this must be ruled (7. 10)
out because 5 is not a constant tensor of the theory. where we have omitted the integration over all a, .
This may seem strange, but this can be easily shown by Decomposing the action to And the on-shell conditions is
calculating the variation of not difficult, and resembles that found in Ref. 11, which
also introduced proper time. The on-shell conditions be-
gL„L„, gL „L„ (7.2)
come
and various combinations under Diff(S, ). None of them
14' & physical
are invariant; they all have the incorrect weight. In fact, 7
the only constant tensor that we can use is the Dirac-type 5
matrix y . Let us define a 0 &physicai=0
I ~
end-point gauge. Our Feynman rules then require that VIII. CONCLUSION
we use the vertex ~
V
1 2 3
& where g„a, &0. At first,
In conclusion, the geometric string field theory pro-
this seems to be a disaster, because the use of this vertex vides a modular-invariant, gauge-covariant string field
function is known to undercount moduli space in the
theory which can derive the Shapiro-Virasoro amplitude.
midpoint gauge. However, the fundamental difference
Specifically, we note the following.
between our gauge-fixed formalism and the work of Refs.
(1) The action (7.5) can be derived by gauging the
4 and 5 is that we have an extra measure term coming universal string group, the group which maps loop space
from the factor det e„"i' . ~ ~
strings into themselves and all their conjugates. In order
The removal of this measure factor can be accom- to satisfy the Jacobi identities, the structure constants are
plished as follows. We can always extract the presence of based on triplet configurations in loop space for Fig. 5,
the vierbein in any vertex function or field by using a not for Fig. 3.
Taylor expansion. In (4.21}, for example, we can extract (2) The group determines the basic fields of the theory,
out the dependence on as follows: P which transform as connections. Also, there is a string
i=1
3
II U. "I V. .. &= V
vierbein e„ in the theory, which makes possible the
treatment of the parametrization length as a gauge degree
of freedom. Thus, there is no overcounting of the param-
where etrization length.
(3} Gauge fixing is performed by using the decomposi-
U
l
'=exp g e"(8„— „)
n
Q tion
USG
And e" is the remnant of the vierbein which changes the
parametrization length of a string from a; to o.';.
Diff(S) )
It would thus be totally incorrect to use the midpoint which demonstrates that the universal string group is the
or interpolating vertices blindly in constructing an N- semidirect product of the Virasoro group and the string
point amplitude. Only the end-point gauge has the prop- group.
erty that the measure equals one, and hence N-point (4) The ghost sector of the "rigid" BRST theory corre-
functions can be constructed without keeping track of the sponds to the tangent space of the geometric theory. The
vierbein. When using the midpoint and interpolating various ghost representations of the fields and vertices are
gauge, we must be careful to consider the vierbein mea- representation-dependent ways of expressing Clebsch-
sure term, which, by a X redefinition, can be recast in the Gordan coefficients in the geometric theory.
form of the U operator. Fortunately, there exists a large (5) When gauge fixed, we have a new vertex, called the
body of identities which will allow us to eliminate the U interpolating vertex, which allows us to smoothly go be-
matrix entirely. Using (4.28), we can reflect L„off the tween the end-point and the midpoint gauge.
vertex function and covert it to a L . Thus, the new (6) The measure equals one in the end-point gauge, but
vertex can be written totally in terms of the operator is not equal to one in the midpoint gauge (opposite of the
L „.For infinitesimal e, we have situation in the open-string case). When the measure
(vierbein) is removed, the closed four-string interaction
emerges as the counterpart of the instantaneous Coulomb
term. This provides the missing piece which completes
By using the above identity, we can write the proof of modular-in variance for the geometric
theory.
&=exp(L" „N„"' 5a' +L" „N„"' 5a' (7) Proper time is introduced through Diff(S, ), so that
V ) V
~
1 2 3
~
Interpolating gauge: three twists, two stretches, one We will show in a later paper that our theory is modu-
stretch/one expansion. lar invariant to all orders in perturbation theory.
In the interpolating gauge, there is a subtle interplay
between the stretch and expansion modes which will be ACKNOWLEDGMENTS
fully explored in a forthcoming paper.
By contrast, the work of Refs. 2 and 3 apparently has The author would like to thank B. Sakita for stimulat-
the following counting: three stretches, three twists, one ing conversations. This work was supported in part by
one expansion, adding up to seven Teichmiiller parame- Contracts Nos. NSF-PHY-86-15338 and CUNY-
ters per loop, which is undesirable. FRAP-RF-13873.
'M. Kaku and K. Kikkawa, Phys. Rev. D 10, 1110 (1974); 10, M. Kaku, Introduction to Superstring (Springer, New York,
1823 (1974). 1988); Int. J. Mod. Phys. A2, 1 (1987); Phys. Lett. B 200, 22
A. Neveu and P.C. West, Phys. Lett. 168B, 192 (1986). (1988); Reports Nos. CCNY-HEP-14-1986, CCNY-HEP-3-
H. Hata, K. Itoh, T. Kugo, H. Kunitomo, and K. Ogawa, 1987, and CCNY-HEP-7-1988 (unpublished).
Phys. Rev. D 34, 2360 (1987). M. Kaku and J. Lykken, following paper, Phys. Rev. D 38,
4J. Lykken and S. Raby, Nucl. Phys. B278, 256 (1986). 3067 (1988).
5S. Sen and R. Holman, Phys. Rev. Lett. 58, 1304 (1987); see ' S. Giddings and S. Wolpert, Commun. Math. Phys. 19, 177
also A. Strominger, Nucl. Phys. B294, 93 (1987); J. Feng, Re- (1987).
port No. CALT-68-1466, 1987 (unpublished); G. Siopsis, Re- "A. Neveu and P.C. West, Nucl. Phys. 8293, 266 (1987); S.
port No. CTP-TAMU-633-87, 1987 (unpublished). Uehara, Phys. Lett. B 190, 76 (1987); W. Siegel and B.
E. Witten, Nucl. Phys. B276, 291 (1986). Zwiebach, Nucl. Phys. B282, 125 (1987).
S. Giddings and E. Martinec, Nucl. Phys. B278, 91 (1986).