Sister trajectories in string theory


Material Information

Sister trajectories in string theory
Physical Description:
iv, 85 leaves : ill. ; 29 cm.
Carbon, Steven L., 1958-
Publication Date:


Subjects / Keywords:
String models   ( lcsh )
Physics thesis Ph. D   ( lcsh )
Dissertations, Academic -- Physics -- UF   ( lcsh )
bibliography   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1993.
Includes bibliographical references (leaves 82-84).
Statement of Responsibility:
by Steven L. Carbon.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001897880
oclc - 29809574
notis - AJX3177
System ID:

Full Text








I owe many thanks to Professor Charles Thorn, my thesis advisor, for many

enlightening discussions, and for his encouragement. I would also like to thank

the University of Florida particle theory group for presenting a stimulating

atmosphere in which to work.




ABSTRACT . . . iv

1 INTRODUCTION ....................... 1

2 SISTER TRAJECTORIES ................... 5
Tree Level Six-Particle Scattering . . 9
Tree Level Eight-Particle Scattering . 14




6 CONCLUDING REMARKS ................. 50

Review of Weight Diagrams . . 53
Standard Construction of Lax Operators . 58
Diagrammatic Construction of Lax Operators . 61
Proof of Diagrammatic Scheme. . .. 70
Supersymmetric Lax Operators . . 75
Discussion . . . 80

REFERENCES . .. . .82

BIOGRAPHICAL SKETCH .................... 85

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy




May 1993
Chairman: Charles B. Thorn
Major Department: Physics

It is shown that by using higher-order corrections that include sister trajec-

tories, it may be possible to restore the Cerulus-Martiii bound in string theory,

which would resolve an objection to locality. In the single-Regge limit, we

review the existence of the first sister trajectory in the six-point function and

then exhibit the second sister in the eight-point function. New work includes

demonstrating that the first sister enters the four-point function at two loops,

and that it can be seen across the intermediate open string propagator which

appears as a higher order correction to closed string four-point scattering.

We then introduce a procedure for determining the state representation

of the sister trajectories, am(t), for space-like momentum transfer squared t.

These sister states are obtained by analytically continuing from the physical

states, and involve reciprocal oscillators signalling the sisters unplihyical nature.

We consider both tree level and higher order scattering.


A major hope of string theory has been that it would describe physics at

the Planck scale. This entails understanding its short distance behavior and

considering the related issue of locality. The latter is important because if the

theory is nonlocal at a fundamental level then acausality may result, which is

probably unacceptable. The high energy behavior at tree level seems to suggest

that it is not local, although Gross and Mende1 claim that it still obeys causal-

ity because the interaction of strings is local. The tree level objection to locality

in string theory is that in the limit of high center of mass energy, s --+ 00, for

fixed scattering angle sin2(0/2) a -_, it does not behave as do theories for

nonextended objects. Cerulus and Martin2(CM) found that for general theo-

ries, under certain assumptions which include locality, the scattering amplitude

obeys the rigorous lower bound | A(s, t) I> e-vlIn sc(9). String theory, on the

other hand, has the tree level fixed-angle behavior I A(s, t) I-- e-sf(0), which

was pointed out even in Veneziano's original paper.3 Gross and Mende1 and

later Mende and Ooguri,4 attempted to determine if stringy perturbative cor-

rections could restore the CM bound, but were unable to control the higher

order corrections to reach a decisive conclusion.

The CM bound was derived using the assumptions of unitarity, existence

of a finite mass gap, and polynomial boundedness. Off-shell, covariant string

theory must introduce Fadeev-Popov ghosts to be unitary. However, unitarity

can be established on-shell in the critical dimension, for instance, by breaking


manifest Lorentz invariance and fixing to the light-cone gauge. String theory

also does not have a finite mass gap since it has massless particles. But this

probably does not lead to the violation of the CM bound. The final assump-

tion of uniform polynomial boundedness states that, for fixed t, the amplitude

I A(s, t) I is bounded by sN where N does not depend on s or t. String theory

does not obey this condition either for t > N since its fixed t behavior goes as

st. Nor does this behavior obey even the weaker condition where N ~ 0(1 t 12)

which Martin showed also gives the CM bound.5 Furthermore, in quantum field

theory, polynomial boundedness is a consequence of locality. As a result, it is

generally thought this power behavior of s leads to the CM bound violation.1'6

Although restoring the CM bound is necessary for a local string theory, it is

not sufficient. Nevertheless, resolving this issue may be important for future

development of string theory.

In this thesis we will show how stringy corrections can slow the exponen-

tial fall-off of the scattering amplitude for t -- -oo. However, unlike Gross

and Mende who examined the high energy behavior that dominates over the

entire moduli space, we will focus on particular processes that dominate just

a small region. Our analysis will expose an overall subdominant behavior that

is consistent with the lower limit of the CM bound. This suggests that the

more dominant behavior considered by Gross and Mende may actually exceed

the lower bound. However, we have yet to sum the perturbative series to ver-

ify if in fact the bound is obeyed. By considering subdominant behavior we

hope that in performing the perturbative sum it may be possible to avoid the

uncontrollable corrections which obstructed Gross and Mende.

The plan is as follows. In Chap. 2 we review the emergence of the lin-

ear Regge trajectories, called sisters, which have more gradual slopes than the


standard a(t) Regge trajectories at large negative transfer momentum squared

t. Rather than working directly in the fixed angle limit, we find it more conve-

nient to take s --* oc with t held fixed. We show that the complete set of sister

trajectories is consistent with the CM bound. We then present a slightly new

approach for obtaining the sister contributions and discuss six-point and eight-

point tree level scattering. The generalization to N-point scattering should

then be apparent. Since the sister trajectories occur in the bosonic sector, our

results equally apply to the Superstring and Heterotic string.

In Chap. 3 we will show that sisters first appear in the open string four-

point scattering amplitude at the double-loop level. We focus on this process

because four-point scattering is the simplest case which must be shown to obey

the CM bound. An interesting result suggested from our analysis is that the

sisters also occur in the non-planar case defined when twists are placed on both

loops while they are separated by the intermediate sister propagator. Thus,

in Chap. 4 we are led to consider the case of four interacting closed strings

with an intermediate open string propagator. We will find that the open string

propagator supports the sister trajectory, which is degenerate with the dilaton

trajectory that may appear on the connecting closed string propagators.

In Chap. 5 the focus is shifted to the the state interpretation of the sisters.

We will determine the oscillator representation of the sisters by isolating the

appropriate propagator, and then saturating it with string oscillators. We

then analytically continue to the unphysical sister state. This procedure also

affords us the opportunity to confirm the amplitudes derived in the high energy

analysis. In Chap. 6 we give some concluding remarks and discuss a possible

physical interpretation of the sisters.


In the appendix we discuss the weight diagram construction of Lax oper-

ators, which have no connection with sister trajectories. Lax operators have

gained recent popularity in their application to the theory of matrix models,

which have been shown to be related to low dimensional string theories.7,8,9

Each Lax operator can be associated with a particular representation of an

affine Lie algebra, and generates a corresponding KdV equation. These KdV

equations, in turn, generate integrable systems which can reproduce matrix

model results.

Our notation is as follows. The standard Regge trajectory is given by

a(t) = a't + ao, where we choose the open string slope a' = 1 and intercept

a0 = 1. This leads to a tachyon mass of m2 = -1. In the same units, for the

closed string we have a(t) = t + 2. The trajectory a(ti) is associated with the

momentum transfer squared ti across the propagator zi. Finally, the trajectory

t(sij) is defined with respect to the energy sij = -(pi + Pi+1 + + Pj)2.


A central feature of the dual resonance theory3 was that in the high energy

limit s --+ co the scattering amplitude scales as A(s, t) oc sa(t), for fixed t. Until

the mid-70's, it was thought that the Regge trajectories a(t) were linear and

parallel, i.e., a(t) = a't+a0, differing only in their intercept ao. Then, in 1976,

Hoyer, T6rnqvist and Webber10 discovered that the theory also predicted a new

"sister" trajectory with a slope half that of the leading Regge trajectory. They

were led to this result by a careful examination of the six particle scattering

tree amplitude of Fig. 1.

Hoyer et al. argued as follows. In the limit s --+* o, the six-point amplitude

factorizes as follows:

A6 D(aa)V(aa, ab)D(ab)V(ab, ac)D(ac). (2.1)

Here, the propagators D(a) have zeros for a = -1,-2,.... On the other

hand, the vertices V(aa, ab) have unphysical poles for aa, ab = -1,-2,...,

which have the undesirable properties of negative spin(nonsense) and wrong-

signature. Now, for ab there is only one zero coming from the central propaga-

tor, while there are two poles coming from the adjacent vertices. This leaves an

unphysical pole, which does not appear in the exact expression for the scatter-

ing amplitude. For the theory to be consistent there must be some mechanism

to cancel this unwanted pole.

2 3 4 5

X 1zl z2 z3

1 1 6

Figure 1. The first sister f(t) requires twists on both zj and z3.

At that time, most people worked with four-dimensional models. Although

string theories are simplest in critical space-time dimensions of 10 and 26, we

can still adopt this restriction in the high energy limit by permitting momentum

to grow large in only four dimensions. In this case, for six particle scattering,

there are eight kinematic degrees of freedom yet nine free parameters. Thus,
one must apply a four-dimensionality constraint.11 For the higher dimension

string theories, one would have to consider a higher N-point function with

corresponding dimensionality condition. The following discussion holds in any


To examine the high energy behavior of the amplitudes in a way that

makes sense requires that one first analytically continue the energy into the

complex plane, e.g., s -- ioc.12 What Hoyer et al. observed was that previous

analyses had imposed the dimensionality condition only after the energy had
been analytically continued back to the physical plane. The effect of not main-

taining the constraint throughout the calculation is that some critical point

remains hidden. Fixing this oversight and imposing the constraint before the

back continuation, allows a factorization to occur in the amplitude, which then


exposes the critical point. Integration about this point then leads to the be-

havior I A(s,t) 1--+ (t) where /3(t) = -a(t) 1 is the first sister trajectory,

for 03(t) = -1,0, 1,... or equivalently a(t) = -1, 1,3,.... Furthermore, at

a(t) = -1, it was explicitly shown that the pole due to the sister trajectory

03(t) precisely cancels the remaining unphysical pole coming from the a(t) tra-

jectory. In addition, the sister trajectory has associated daughters which cancel

the poles at a(t) = -2, -3,....

Ref. 10 also noted that the sister trajectory had been elusive in the past

because at each vertex it can not couple to more than one on-shell state. This

decoupling can easily be understood by considering the factorization (2.1). In

the high energy limit, the end vertices that couple an intermediate propagator

to two on-shell states are represented by factors of unity, which obviously do

not have poles.

Shortly after the discovery of the first sister in the six-point amplitude,

working in the helicity-pole limit, Hoyer13 showed that the eight-point tree

amplitude predicts a second sister trajectory -y(t) = a(t) 1. The purpose

of this second sister is to cancel unphysical poles occurring on the /(t) sister

trajectory for 03(t) = -2, -3,.... He then proposed the generalization,

1 1
am(t) = -a(t) -(m -1), (2.2)
m 2

where the mth sister first appears at tree level for 2m + 4 interacting particles.

Hoyer et al. then showed that the first sister 3(t) could be obtained using the

more general single-Regge limit,14 as opposed to the helicity-pole limit used in

Ref. 10. Other work followed which examined the sister trajectories under the,

even less general, multi-Regge limit.15'16

Sisters were subsequently found in the Neveu-Schwarz sector of the NSR

superstring,17 and related phenomenological implications were discussed.18



Figure 2. Plot showing the leading a(t) Regge trajectory and the first two
sisters, 3(t) and 7(t).

Quir6s showed that the sister 3(t) appears in the single-loop six-point diagram,

and that it renormalizes the corresponding tree level sister.19 Further, several

papers also considered the closed bosonic string which found that, as in the

open string, sisters appear at tree level when there are at least six interacting


The first two sister trajectories are shown in Fig. 2 along with the leading

Regge trajectory. Due to successively more gradual slopes, the net behavior

of the sisters is clearly not linear as t -+ -oo. We can find the asymptotic

behavior by considering the intersection of two neighboring curves am(t) and

am+i(t) and then letting m -- oo. Equating these using (2.2), we easily find
that to lowest order a(t) -m2. Comparing this with (2.2) gives

am(t) -m = -V-2a(t) (2-.)t.



For fixed angle scattering we conclude sam(t) -* s-v', which is the CM bound.

This short calculation also demonstrates why locality may be responsible for

the CM bound violation. In string theory, the fundamental length scale is

defined by I = VaV. Further, the limit m -* oo is completely equivalent to

a' -- 0. Thus, the length scale I associated with each sister approaches zero as

the order of the sister increase. This means that each successive sister appears

more local than the previous one, and in the asymptotic limit we reach point-

like behavior.

The possibility for restoring the CM bound in four-point scattering exists

if we can show that the entire set of sisters, am(t), is present. Since, at each

vertex, sisters do not couple to more then one on-shell state, we must consider

higher order corrections. We make an initial step in this direction by showing

that the first sister 0(t) couples at the double-loop level. Because t is to be

held fixed, and in order to work under the most general conditions, we apply

the single-Regge limit s --* oo. However, since the original approach found in

Appendix B of Ref. 14 requires a priori knowledge of any twists, we modify

the calculation to remove the need for their explicit presence in the initial

expression of the amplitude. This has the advantage of allowing us to consider

many cases simultaneously, which significantly reduces the amount of work

evaluating higher order functions. Note that in the multi-Regge limit, one can

only determine the need for twists by first inserting them, and then computing

the final result to see if sisters appear.

Tree Level Six-Particle Scattering

The sister trajectory /(t) is seen in the high energy limit of the six-point

function only if twists are placed on both of the adjoining propagators as shown

in Fig 1. We begin, however, with the corresponding untwisted amplitude. This

is easily calculated in the Fubini-Veneziano formalism from

A6 = (0,P1 I V(p2)AV(p3)AV(p4)AV(p5) I 0,P6). (2.4)

In general, passing the vertex operators through each other produces factors

of the form

exp [-2pi pZ (1 z)2piPj, (2.5)
where z is the product of coordinates zi which are associated with the propa-

gators connecting the vertices. Final expressions for the amplitude are usually

written in terms of the right-hand factors. The left-hand form is more conve-

nient, however, for locating critical points in the high energy limit s -+ ioo.

Consequently, we use the left-hand side of (2.5) if one of the connecting propa-

gators sees the energy s, and the right-hand side for non-overlapping quantities.

In the particular case of Fig. 1, the complete exponential factor is then easily

found to be
oo n
exp [2 z(-P2z P3) (P4 + P5z) (2.6)
Substituting in the momentum scalar products, the full amplitude becomes

A 1 dz dz2dzz --a(t) z-1-a(t2) -1-a(t3)
A6 ,- dzadz2az3z1 z2 23

x (1 zl)-1-a(s23)(1 z3)-1-a(s45) (2.7)

x exp z nz(s24 34) + zzs +s 34 + z( (335 34) )
where we have defined

s= 34 + S61 s24 s35. (2.8)

In writing (2.7), we have also dropped terms in the exponential which can be

safely neglected in the high energy limit.

We are now in a position to impose the four-dimensionality constraint,

which, in the high energy limit, reduces to

s35s24 1. (2.9)

Applying this constraint to (2.7) allows the argument of the exponential to be

factorized, giving

A6 01dzldz2dz3z-1-a(ti) -1-a(t)3 -1-a(t3)(1 l)-1-a(23)
0 0 n )] (2.10)

x (1 z3)-1-a(45) exp[ E 2 1 -- Xl)(Z2 -


1 = l )- 1, X3 -_ 1 (2.11)
s35 S24
To discuss the high energy limit we must let s -+ ocei6 where the real part of s

is held fixed, and 6 is such that the real part of s is in the strip of convergence.14

The result is a Fourier integral whose asymptotic behavior is dominated by its

critical points.12 For (xl,x3) to be a useful critical point it must fall within

the integration region, 0 < z1, z3 < 1. Critical points taken at the boundaries

do not produce sisters. Since the boundary of the integration region is not

included, the factors in (2.7), other than the exponential, can be ignored during

integration. To recover the proper limit s -+ ooe i, we obtain a double critical

point by choosing the phases

s34, 61 -* ooei,
s24, 35 -+ -cei
This is completely equivalent to twisting the propagators corresponding to zi

and z3 since energies that overlap an odd number of twisted propagators change

sign. In other words, the role of the twists here is to place the critical point

inside the integration region.

To obtain the leading sister trajectory by evaluating (2.7) about the critical
points, we keep only the lowest order terms in the exponential and integrate

Iz x1 5e, I z3 x3 1< e, (2.13)

for e small. Choosing higher powers of z1 and z3 would lead to daughter
trajectories. After shifting z1 and z3, we obtain

A6 -~ l 1-a(ti) x1-a(t6)(1 l)-1-'a(s2)(1 x3)-1-0a(s45)6, (2.14)

16= dz2z2 )ez dz1dz3 exp [z2z1z31, (2.15)

c = zX3(x1 1)(x3 1). (2.16)

Setting y = -iez2z3s, gives

6 = i()-1 dz22-a(t2)e f d 0 dy exp yz), (2.17)

where yo = -iE2z2s. Integrating over z1 we easily find

6 -1 1 -2-(t2) zjc y exp(iy) exp(-iy)
I6 = -1 dz2 z2 e P )f0dy (2.18)
o Y-0 y
where the y integral is symmetric. If z3 were not critical, taking the limit e -- 0
now would give L6 = 0 (use dy ~ e). This demonstrates the need for a double
critical point.
Now, define z = -sz2c, which gives

I6 = _(-c)a(t)+-1 1 dzz--a(t2)e-z 0Y dexp(iy) -exp(-iy)

For us to consistently write z = y2e-4g-1c we must have 1 > e > g-4 to

reach the lower limit z -+ 0 for fixed y. Consquently, in the high energy limit

s -t ioo, yo -+ 00 and so the integration over y gives iwr. Next, the z integral

gives r(-1 a(t2)) which is valid only for a(t2) < -1. Thus, the complete

amplitude is
.1+ )1 X 1 1 \ -1-a(ti) -1-a(ta)
A6 -c 2 1 3 (2.20)

x (1 x1)-1-a(s23)(1 x3)-1-a(45).
We now analytically continue the energy back by making the replacement -A-
e-i'. Defining f3(t2) = a(t2) 1, which corresponds to the first sister

trajectory, and simplifying, we finally arrive at

A6 ~Zire-i7r(t2)2-l(t2)-10(t2)F(-/3(t2) 1)X(t2)-0a(1) 4(t2)-a(t3)
3 (2.21)
x (1 x1),(t2)-a(S23)(l x3)(t2)-a(s4).
Since each of the energies comprising 9 overlaps with s34, the Regge behavior
)(Wt2) shows that the central propagator in Fig. 1 sees the sister. Using the

four-dimensionality constraint (2.9) we can write

s = 834 + 61 s24 35 = s61(s35 861)(s24 s61), (2.22)

and easily recover Eq. B.19 of Ref. 14.

Examining the F function in (2.21), we see that the poles of the sister
trajectory are for 3(t2) = -1,0,1,.... Our approach makes it particularly easy

to determine the signature r of these poles. Twisting the sister propagator t2

changes the sign of all overlapping energies. Although both the numerator and

denominator of x1 and x3 change sign in Eq. (2.11), the signs of the energy
ratios remain unchanged. Thus, the twisted and untwisted diagrams can be

added together giving an overall factor r + 1. Therefore, the poles of 0(t2)

have pure positive signature. Since these poles correspond to odd values of

spin, i.e., a(t2) = -1,1,3,..., they have unphysical wrong-signature.
For the existence of the sister it was necessary that the argument of the

exponential factorize, producing a 2-tuple critical point. Integrating over both

2 3 4 5 6 7

\P(t) y(t) 0(t) /

z1 z2 z3 z4 z5

1 8

Figure 3. At tree level, the second sister 7(t) first appears in eight-point
scattering. Concurrently, z2 and z4 see #3(t).

coordinates, in effect, removed the linear power of the propagator variable 22

from the exponential. In general, integrals of the form

I= 1 -a-1-n exp(-czrn) (2.23)

in the limit c -+ oo integrate to

_I (- -- )- -~ for a < -n. (2.24)
m mm

Thus, sisters do not appear in 4- or 5-pt scattering since both retain the linear

power of z. Furthermore, to produce the second sister "7(t), both the linear and

quadratic powers of z must be integrated away, leaving the cubic power. This

occurs when the critical point is a 4-tuple, which first arises in the eight-point

scattering amplitude.

Tree Level Eight-Particle Scattering

In this section we will expose the second sister, -"(t), in the open string tree

diagram of Fig. 3, where the sister appears across the propagator with z3.16

In the corresponding amplitude, we isolate the relevant terms by including,

in the exponential only quantities which overlap the central propagator. We

gather the other terms into a function f(zi, z2, z4, zs), whose exact form can

be ignored since, as shown in the last section, the sisters depend only on the

exponential factor. The advantage of using the single-Regge limit over past
approaches becomes more apparent in this example.

From Fig. 3(without twists), we immediately write down

1 5
AS = 0 j dzif(zl,z2,Z4,z5)z3at3)
0i= 0 (2.25)

x exp [2 n(-P2zz p3z2 p4) (P5 + P6z + P7z z ).
Substituting in the high energy limit values of the momentum scalar products

gives the eight-tachyon amplitude
1 5
A8 f- 0 dzif(z,'z2,z4,z5)z31
00 n
x exp[ znz(s25 -- s35) + z1zz4 (s26 s25 + s35 s36)
n=1 (2.26)
+ 45 + 2z z2 z4 + 5 2(s35 845) + z4 (s46 S45)

+ z (s47 s46) + 4zz (836 s35 + S45 s46)

+ z2z4z5(s37 s36 + s46 S47))

where, now,

= (s81 826 + s36 s37). (2.27)

Applying the four-dimensionality constraints

s81836 8135 -= 1, 81-46 = 1, 81845 = 1, (2.28)
s26s37 s25s37 s26847 s25847

factorizes the argument of the exponential yielding
1 51-a(t3)
AS 0 dzif(zl,z2, z4, z5)z3 (
i= l (2.29)
x exp [ (z X)(Z x2)(z x4)(z x5)],

s36 s37 + s47 s46 s25 35
l -4=
s s26 s25 + s35 s36 (2.30)
S47 s46 25 s26 + s36 s35
X2 = 5 5
s37 s36 + s46 47'
For the critical point (xl, X2, X4, x5) to be inside the integration region, we

must place twist on each of the associated propagators, and apply an additional

four-dimensionality constraint:

3645 1. (2.31)
Consequently, due to the twists we have the sign changes

826, 35, 37, s46 --+ -ooe. (2.32)

To remove the first two powers of z3 in the exponential in (2.29), and to obtain

a leading trajectory, we will integrate around
z1 x1 1e, z2- x2 15 E,
z4 4 |< E, I25 Vx5 1 E-
Clearly, this is just one of many critical points that we could have chosen. By

writing z x4 = (z4 V/4)(z4 + v'4) ~ 2/f-4 (z4 ViT-), etc., and shifting

the z's, we find

A8 ~ f(x1, x2, vx4, /i) dz3z31-O(t3)esz3c

x dzl dz2 exp z3zlz24(v/4 -4)(v5 x5) (2.34)

x dz4 dz5 exp[2z2z4z5s. (x2 X1)(12 x2)],

13 3 3
c = 3( l)(X2 x2)(x4 x4)( x5). (2.35)

The last four integrals in (2.34) can be done in pairs, resulting in

A8 ~ 47r2(2.i2x,(x2 l1)(2 x2)(1 V )(1 i5))-1
x f(x1, x2, V/-5, v/i dz3-4-Q))ea3 c

Using (2.24) then gives

A8 ~ f(x1,x 2,v)2(-sc) (- ) 1)
3-1 (2.37)
x (22x4x5(x al)(a2 x2)(1 v/4)(1 V/- ))

Again, we analytically continue back by replacing -9 with e-"9. Thus, we
find the Regge behavior A8 oc 9 a(ta)-1 g(t3), which corresponds to the
second sister trajectory. The first pole at q-(t3) = -2 cancels the pole of the
first sister trajectory at /(t3) = -2. The daughters of 7(t3) cancel the other

poles at 3(t3) = -3, -4, ..
When the central propagator in Fig. 3 carries the second sister 'Y(t), the
adjacent propagators, z2 and z4, see the first sister 3(t). Each of these sisters
can easily be computed by constructing the exponential term in (2.25) from the

appropriate overlap quantities, taking the corresponding high energy limit and
then integrating over a 2-tuple. Finally, if we had chosen to initially integrate
over a 2-tuple critical point for z3, then we would have found the first sister
trajectory, /(t).


We now adopt our procedure to handle loop corrections. As in the tree level

case, we must first isolate terms in the corresponding amplitude which overlap

the appropriate propagator. In particular, to search for sisters in the double-
loop four-point amplitude, we consider the limiting situation where the two
loops are sufficiently separated such that they and the connecting propagator
can be treated as individual objects. Two such topologies are shown in Fig.

4. Both may be constructed by sewing together two single-loop diagrams. For
this, we use the formalism from appendix D of Di Vecchia et al. 23 where the

open string N-point multi-loop vertex has the form

V(N;g) oc dDp exp [i7Tp r- 1 + p B + C], (3.1)

and where 7r is the period matrix. Completing the square and integrating over
the loop momentum p gives

V(N;g) c(det D/2 exp [-rB 7-1 B + Cl. (3.2)

The factorized four-tachyon double-loop amplitude is then

1 4 1 2
A(4;2) = dz dzi f d(0,p,p2 exp 2 + CL
i1 (3.3)
L(<'-2 < r 1 (BR')2 + t
x -2 exp +[ ]0,p3,P4),

where the subscripts L and R refer to the left and right loop, resp., and the
superscript on Lc) labels the leg connecting the loops. The period matrix


2 3
\ / 2 3
S (t) j3(t)

1 4
(a) (b)
Figure 4. Two distinct topologies for producing the 3/(t) sister in double-
loop four-point scattering.

has been reduced to the single-loop case r = 2ri ln k, where k will be defined
below. The details of the measure dtz, which is a function of k1 and k2, may

be suppressed in the analysis below as long as we avoid the boundaries of the
integration region.
In the multi-loop case the coefficient BY in (3.1) is given by(with a' = 1)

--/ qm( n Tt (z)) ^ Ta (zo)
B 'a= \/2 I) In (3.4)
i=1 m=O P Ta(Vi(z)) P Ta(zo)) z=O' (3.4)
where zo, irp, and p are fixed points, and a product of Schottky group elements
is defined by

Ta=S SS S r=1,2,...,g; nieZ/{O}; pi Ipi+1, (3.5)

where g is the genus number. Also, (P) Ec, means that the sum is over all

elements of the Schottky group except that the leftmost element in Ta can
not be S,. In the single-loop case Ta = S' and S'(y) = kny, where k is
the multiplier and related to the radius of the loop. Here, however, the sum
restriction leaves just the identity. Finally, for one loop o1 -- oo, i71 -+ 0. Thus,
dropping the loop index,

B = V2 1 m In (3.6)
i=1 m=O MzO

where the projective transformation is explicitly given by

T(z) = zi-l(zi zi+l)z + zi(zi+l zi-1) alz + a2
(zi zi+l)z + (zi+1 zi-1_l) a3z +4 (3.7)

To reduce (3.3) we will need the commutator
[B. a()] = 12 ( lnVc(z)) (3.8)
[ c -(m 1)! z= (
Partial derivatives of the projective transformation can easily be taken giving
ala4 a2a3 (zi zi+)(zi-1 zi)
(a3z + a4)2 z=0 (zi+1 zi_) (3.9)
or, more generally,
m!(-a3)mn-(ala4 a2a3)
(a3z + a4)m+l z=
=_ m-lm! (zi zi+l)m(zi-1 zi)
(Zi+1 zi-1)
The single-loop three-point diagram is constructed by sewing together two legs

of a five-point diagram, and then fixing three of the projective coordinates.

For that case, following Di Vecchia et al. 24 we sew together legs 3 and 4

and then choose z3 = k, z4 = oo, and z5 = 1. In the present case, we will

associate the coordinate z5 with the connecting leg coordinate zc. This gives

9zVc(z = 0) = 21 1, along with Oz'Vc(z = 0) = 0 for m > 2. Thus, the
commutator (3.8) becomes
[B, = c) (-)m E (z))m z= = 2 E(1 zl)m. (3.11)
m= 1 m= 1
Next, the coefficient C in (3.1) is given by
3 00
C i) (i) 1 ln[i'(z)]z=0
i=1 m=0
3 o0 (i) (j)
+ 2 E > n! rm! ln[(y) Vj(z)]y= z=0 (3.12)

i,j=n O Vm=O) y=z=O

where the prime form is defined by

wi \z T/ (w) w T (z)
E(z,w) = (z w) z T() w T,(w) (3.13)
Sz To,(z) w Tc,(w)'

and the indicates the the identity is not included. For a single loop the prime

form reduces to
00 nw w kz
E(z,w) = (z w) I z k w k (3.14)
z knz w knw"
Below, we will need the commutator
(0 [Cc)] O,pi) = ( 1)! mln[Ve(z)]zo
m=l 1
+2 (Pi mln[zi Vc(z)]z=O
i c m= 1

+2E Pi 9 zi knVc(z) Vc(z) knz =
Sm=1 = (m 1)!z n zi knzi Vc(z) k Vc(z) z=0'

where pc = p. Due to momentum conservation we can neglect the second

denominator in the last term. Further, in the high energy limit s -- oo, we

have Pl P3 -+ s/2, Pl P4 --* -s/2, P2 P3 -- -s/2 and P2 P4 -+ s/2. These


P Pi = (P1 +P2)Pi --+ 0. (3.16)

Consequently, some of the terms in (3.15) do not survive the high energy limit

in (3.3). This permits us to drop the entire first term, and the i = c term in

the last sum. Rearrangement then yields
(0 I[C, aMIc)] O,p) =2 E (m -)!
i c m=1
00 00
x [in [zi knV (z)] [Vc(z) krz] z=
n=O r=l

Taking the derivatives gives

(0 [C, c)] 0,pi) =2 pi[
ic m=1

(zi knVc(z))m


+ (_)+l(Vi(z))m
S(V(z) knz1)m =0'

which simplifies to

(0 l [C, a,)] lO, Pi)

kmn(1 z)m +
(k --Y


(- Z1)m


We also need the single-loop result

exp 1-2 + C 0, P10,P2) = ,' 2 2 I 0, p,p2),


where f arises in planar loop amplitudes and can be expressed in terms of the

Jacobi theta function. Substituting (3.11) and (3.19) into (3.3) then gives

A(4;2)" 1j dzz-1-a(t) Jd 1j dazi122p P2V,2'p4
i= 1
ep2 zm(1 zl)m(1 z3)m Pi .j In zi In zj
x exp 2 n
m=1 ij

1 k kmn
I:P (k 2 zj)m
n=0 2
1- kmn
in Z1 [E _I
PInk2 L (k ) -zi)m
n=0 1

i j5c

i#c j

+ E pi Pj

X 2
= (kr z)m
r -- 2 O

n= 1 1

S (1- kfzi)m

00 1
+E(1 knz)m)

00 1
Z=1(1 kyJ)m))


where i and j correspond to the different loops, and we have dropped a mo-

mentum independent factor which can be ignored in the high energy limit.

=-2Ep [E
ic m=l1

(0 k"mn
= (k" zi)m
n=0 1


Replacing the momentum scalar products by their high energy limits allows us

to factorize the argument of the exponential to get

A(4;2) dzz-1-(t) dp dz( ','12'34)-1-ra(t)
JO zm(1 zi)m( z
x exp s z zl)( z3)mgm(zl, z2, kl)gm(z3, z4, k2)

where we have defined

9in y, k) = (k-m (1 )m -< x --+ y >. (3.23)
n=O n=1
The function gm(x, y, k) is for orientable planar loops and is essentially the

mth derivative of in Thus, we can immediately write down the expression

in the non-orientable case:

o In x (-k)mn 00 1
= Ink ((-k)n )m- (1- (-k)nx)m < >
n=0 n=1

and for the non-planar case:

SIn (_ )mkmn 00 1
gmn(x, y, k) = ( (1+ knx)m- . (3.25)
n=0 n=1
Now, we search for critical points which do not reside on the boundary of

the integration region. Unfortunately, due to its complicated form, one must

numerically search for zeros in gm(x, y, k). It is found that gm(x, y, k), for all

m, does indeed possess zeros that are exclusively within the integration range.

These zeros generate the critical-point curve x = P(y, k), for some function

P(y, k) which satisfies gm(P(y, k), y, k) = 0. In addition, numerically analysis

indicates that both non-orientable and non-planar cases also possess critical-

point curves. In all these cases the zeros do not seem to be confined to any

particular region of integration space.

This case differs from the tree calculation in two respects. First, to factorize
Eq. (3.21) it was not necessary to impose a dimensionality constraint. Clearly,
this is due to the fact that there are only four interacting particles, and not due
to the loops. Second, unlike the tree amplitudes, the presence of twists is not
significant. In the former case, the twists were necessary to change the sign
of some of the energies to place critical points inside the integration region.
In the loop amplitudes, the signs change as a result of the periodicity of the
Jacobi theta function.
Continuing with the calculation, in the limit s -* ooei6 (3.22) becomes

A(4 2) ~ dzz-1-a() d dzi(0 12 34)1-(t)esZh22
) J O i=1 (3.26)
x exp [sz(1 z1)(1 z3)g1(zl, z2, kl)gl(z33, z4, k2)]
h2 = (1 21)2(1 z3)2g2(zl, z2, kl)92(z3, z4, k2) (3.27)
We will evaluate about the critical curve

I 21 P(z2, k) 1 e, z3 P(z4, k2) 1< (3.28)

Expanding the gl's about this curve, and then shifting z21 and 23, gives

A(4;2) ~ 1 dzz-1-a(t) / 1 dz2dz4(12 034)-1-a(t)esz h2
Jo J Jo (3.29)
x dz1dz3 exp[szzlz3h ],
hl =(1 P(z2, kl))2(1 P(z4, k2))2
x gj(zl = P(z2, kl), z2, kl)g(z3 = P(z4,k2),z4, k2),
and h2, '12, and 034 are now evaluated on the critical curve. The integration
of zi and z3 proceeds as before, giving

A(4;2) irs-1 dzz-2-a(t) f d (3.31)
O (3.31)
x j dz2dz4(12 34)-l-a(t)h-lesz2h2.

Similarly, the z integration is also easily done giving

A(4;2) ~ iTre-ir(t) (t) n(-f(t) 1) I dy
11 (3.32)
x dz2dz4(12034)-1-(t)h1h 0(t)+1,

which exhibits the first sister trajectory /(t). Since the integrands involve

derivatives of the Jacobi theta functions, we are unable to complete the cal-

culation showing explicitly that the sister does not decouple. For the planar

diagram, however, in the special case 0(t) = -1, it can easily be shown that

the signs of each of the integrand factors are the same over the entire integra-

tion region. On the other hand, to show that decoupling does not occur in the

non-orientable and non-planar cases is more difficult, although the results of

the next chapter indicate that the sister survives the latter case.

The existence of the the second sister requires that two of the g's share the

same critical point. Using

(1 + x)-(r+l) (1 + X)-r = -xe, (3.33)

it follows that
gr(x, y, k) gr+1(x, y, k) = (knxe-knz knye-kY)
n=O O (3.34)

SE -x/k /kn
n0 e- /" kne
Since the difference is independent of the index r, for any given critical point

either one gr vanishes, resulting in a single sister, or they all vanish simultane-

ously. In the latter case, (3.22) results in the form

A(4;2) j1dzz-1-a(t) dl dz2dz4(012'34)-1-a(t)
Sf (3.35)

x dzdz3 exp sziz3 Z zmem].

2 3

1 4

2 3

1 4

Figure 5. General 2m-loop four-point diagrams for generating the mth
sister am(t).

Integrating over z1 and z3, we obtain

A(4;2) -i7rs-1 1 dzz-1-a(t) fdp
1 0 1 (3.36)
x dz2dz4(012034)-1-a(t) zmem)
)J m=1

The right factor gives a z-1 in leading order. Consequently, the z integral

generates a leading pole at a(t) = -1, whereas the second sister requires
a(t) = -2.

Presumably, the 7*(t) trajectory is present if there are at least two loops on

both sides of the propagator. We suspect that, in this case, there would be a

factorization of the form

Gmn(x, y, k, k2) = gm(x,y, kl)gm(x,y, k2) (3.37)

where k1 and k2 correspond to same-side loops. In Fig. 5. we display two

distinct possible multi-loop topologies for producing the higher order sisters.

s s

(a) (b)

Figure 6. The Regge cut behavior is across the dotted lines.

In both cases the central propagator may allow up to the mth sister if there

are at least m loops on either side. However, evaluating Fig. 5a is not practical

since the Schottky representation of the prime form (3.13) is much too formal

when two or more unfactorized loops are present. On the other hand, since

Fig. 5b completely factorizes the loops it requires no more than the techniques

presented in this chapter.

The sister trajectories may also appear across propagators which are em-

bedded in an irreducible diagram. An example is the double-loop diagram

displayed in Fig. 6. The sister here may be across one of the horizontal propa-

gators. Such diagrams are, however, dominated by the behavior of Regge cuts.

In the present case, the cut in Fig. 6a gives

'(t In s
A (~ (3.38)
(ln s)P '

for some p at fixed t. The cut has the same Regge slope as the first sister,

yet its a(t)-intercept is higher. In general, the nth cut occurs at the same

order as that of the n4h sister, but with a trajectory lying above the sister.

This implies that the collective behavior of the cuts would actually exceed the


CM bound. A high energy analysis of the entire moduli space, such as that

of Gross and Mende, would be dominated by the cuts. This is supported, in

part, by their proposal that the fixed t behavior have the form

(t Ins
A ~ +1- (3.39)
(In S)12g

where g is the genus number. The single-loop amplitude, computed first in the

fixed angle limit, was shown explicitly to reduce to (3.39), for p = 1, in the

fixed t limit 0 -- 0.


An unexpected result of the last chapter is uncovered by considering the

non-planar diagram in Fig. 7. The central propagator that carries the sister

f(t) is that of the open string, while the non-planar loops on either side contain

closed string poles. This raises the interesting possibility of open string sisters

coupling to closed string propagators as in the diagram shown in Fig. 8. Below

we show that this is in fact the case. In the case of the Heterotic string,

however, the diagram in Fig. 8 decouples since the open string propagator

can not accommodate the achiral boundary conditions required by the closed

string propagators.

The amplitude for four-tachyon closed string scattering with an intermedi-

ate open string propagator, takes the form

AG2 =(2--) d2zld2z2cl(O,p4 V't(p3,z l,2-11) x o(O I T(At,a) I0)c
(41r), 1J|<1
x Ao c(0 I T(A, a) I 0)o x V(p2, z2,2) I 0, p)c.

Among the many expressions appearing in the literature for the transition op-

erator T between the open and closed string state, we will use that of Shapiro

and Thorn.25 We will ignore here the ghosts terms given in their explicit expres-

sion for T. These give a non-trivial contribution only if loops are present. Even

then, the ghosts can be ignored since they have no bearing on the calculation

which focuses on the exponential contributions away from the integration

Figure 7. Non-planar double-loop four-point diagram. The loops contain
closed string poles.


Figure 8. Four-point
string propagator.

closed string interaction with an intermediate open

boundary region. The transition operator is then given by


E C)As As -
2 nm n Am



(1) (_)n+m
nm -2m 2n + 1


721 I_2I
-2 -
n m

+ (2)
+ nm-2n-I -2m-1
1 0-0 a-2c


(2) 1 (-)n+m 1 11 ,
nm4n+m+l[n m

(3) (-)n+m _
n+m n m




The sine and cosine oscillators of the closed string are given by

1 i
Ac = -(Ar + Ar), As = -(Ar Ar), (4.6)

where Ar and Ar correspond to the left and right movers, respectively. In

terms of the sine and cosine modes, the closed string vertex operator is written

V(P2,z,z) 0, Pl)cl =ei(p+P)x z 1-2-a( exp[p (-- n(zn + -)
-As (zn )n 0)c,,
n /i
where the first two factors are the zero modes.

In Eq. (4.7) we have written the closed string trajectory a(t) acl(t). To

eliminate confusing notation, we will write the open string trajectory also in

terms of a(t), i.e., ao(t) = 2a(t) 3. Then, we have

zLo-2 0)o = zp2-2 0)o = z-t-2 0)o = z2-2a(t) 0)o. (4.8)

Pushing T to the right, Tt to the left, moving the propagator to the right, and

then using momentum conservation to eliminate part of the zero modes, Eq.

(4.2) becomes

AG2 =()2 dzz2-2a(t) < d2z1d2z2 zlz2 -2-a(t
<47-r ) o 1 |zj|<1
1 0 + i
x o(0 I exp [/- P3 (-a2n( 1)
+ i Cn a2m+l( -
x exp [ C $ma2n+1 '2m+1l (4.9)

x exp [ (a-2n-l a -2m-z2(n+m+1)

x exp [P2 (E-- 2nz2n (z + 42 )
V'_ n=1

o nV2 -Z-(1) 2m+l n _z2 n \)
+ j CnmaQ-2m-1z2m1(4 2 0)o.
We can easily move the even oscillators through to the vacuum states since

they only appear at the far left and far right. This produces the factor

exp [-P2 P3 z 2nzl + 2i)(' + (.)
n=1 (4.10)

= (1 z2zlz2)(1 z2zl12) p2-P3,

which can be pulled outside the vacuum states.

Next, pushing the quadratic terms past each other produces the factor
exp 4(2(k + )C (Ca42n-a 2m+2(n+k+l)], (4.11)
enk km -2n-lOa2m+1

and an oscillator independent exponential which we can neglect since it will

not survive in the high energy limit. Moving the quadratic terms to the vac-

uum states will produce no other permanent effect as they pass by the vertex

operators. Pushing the factor (4.11) past the right vertex then results in

AG2 )2 j1 dzz2-2a(t) jz 1 d d2z22 zlz2-2-a(t)

2 2 2p2IP3
x (1 z2zlz2)(1-_ z2z12) 2p2 p3
x o(0 exp[-iVzp3 Cnma2m+1i(z -
x exp[ivP2 E C( -2m-2z2m+1(z2 i)

(x + 4(2m + 1) (2k + I)a-2j-1 .C2)Cz2(j+k+1) 0)o.

Again, the quadratic term will not leave any permanent imprint after moving

to the left-hand side. Finally,

AG2 )211 dzz2-2a(t) j d2z1d2z2 ZlZ2 -2-a(t)

x (1 z2zlz2)(1 z2zl2) 2p2.p3
xexp [2p2 P3 E(2m + 1)z2m+1kmnm(zk kl)(z -2

To perform the sums in the second exponential, we go to the limit s -- oceib

and keep only the term linear in z. The sums can now be done by noting

2 2.7 (4.14)
n --1 2n (4.14)

00 00 1
L znO E 1- 2n = (1 z)I 1. (4.15)
n=1 n=1

Eq. (4.13) then becomes

AG2 (1)2 1 dzz2-2a(t)z<1 d2zd2z2 I zlz2-2-a(t)

x (1 z2zlz2)(1 Z2Z12) -s-8 (4.16)

x exp [-sz((1 z 1) (1 2i) (1 z2)5 -(1- z2))]
As in the double-loop case, since there are only four interacting particles, the
argument of the exponential appears in a factorized form without resorting to
a dimensionality constraint.
Examining the second exponential term in (4.16), we see that there is a
critical point when z1 = -\ or z2 = 52. Writing z = pei', implies 0 = 0 or 7r.
To integrate (4.16) about these points, we return to the Taylor series expansion
in (4.13) and (4.10), i.e.,

AG2 ~ dzz2-2a(t) dpdP2(P1P2)-1-a(t) 2 d01 d02
Jo Jo Jo 47r 47r
x exp [2sz2P1p2 cos 01 cos 02 (4.17)
x exp [4sz E CkkC)p ^sin(k91) sin(n02)]
We expand by setting sin(nO) w nO for 0 = 0, and sin(nO) ; (-)nnO for 0 = 7r.
The p sums for 0 = 0 can easily be carried out as follows:
00 00 (_pn 1 1 ,
P n-'nCO( P 2n = --p(1 p)-. (4.18)
n=l n=l1
With a similar expression for 0 = r, eq. (4.17) becomes

AG2 ~2 dzz2-2a(t) dpdp2(P1P2)-1-(t) fd0 d02
O JO J- 47r 47r
x (exp [sz90192P1P2(1 PI)(1 P2) + 2sp1p2z2 (4.19)

+ exp -sz9Ol2PlP2(1 + Pl) (1 + P2) 2splP2z2),
where the first exponential is for 91 and 02 expanded around the same value,
and the second for the converse case.

Integration over 01 yields

AG2 8 dzzl-2a(t) dpidP2(PIP2 )-2-a(t) dy [eY e-'

x ((1 pi)(1 p2)2 exp [2splP2 21

+ (1 + pl)1(1 + p2)1 exp -2spiP2z2]),
where the exact expression for yo is not needed. The integration over y gives
2i7r. Unlike the previous examples, the sister is not necessarily the dominant
behavior. This requires that we extend our considerations to higher orders.
By Taylor expanding (1 p)2 and (1 + p)2, we obtain many terms which may
indicate the presence of the open string sister, o. To make a firm determina-
tion requires some care since the O3o trajectory is degenerate with the dilaton
trajectories that may appear across the adjacent closed string propagators.
There is no doubt, however, when (4.20) generates a triple pole. The form of
the required solution is suggested by the partial wave analysis term

/-+ sat In2 s, (4.21)
C (ao(t) t)(acl() t)(acl(t) t) sa ln2 21)
for the case ao(t) = acl(t). Eq. (4.20) yields this result if we select the p2
expansion terms for both pl and P2. This gives,

AG2 2T 1 dzzl-2(t) dpidp2(plP2)-a(t)
28s7r 0Jo 0 (4.22)
x (exp [2splp2z2 + exp 2sp1P2z21.

Let w = P1P2. Then

AG2 -2- 1 dzzl-2a'(t) dww-a(t) e2swz+-2swz2) dP2P 1
,2- 7 1Q Jo (p2P


The P2 integral easily gives In w. Next, defining w = z-2y, we integrate over

z to obtain
S [1 (22sy e2sy)
AG2 210s 1 dyy-a(t) In2 y e2sy -2sy

i d2 dyy-a(t) e 2sy +e -2sy

Finally, the end result is

AG2 ir-1 (1 e-irflo(t) )2-9+ (t)I(-1 /o(t))so(t) In2 s, (4.25)

where the open string sister 03o(t) = a(t) 2.

One possible concern that may arise in the above calculation is that in

writing Eq. (4.17) we have discarded the term

[(1 z2zlz2)(1 z2zlz2)(1 z2 lz2)(1 z2z12)] 2 (4.26)

When tachyons are present this may diverge at the critical points in the neigh-

borhood z = z = I z2 = 1. Fortunately, the sister trajectory emerges

from the other end of the integration region where these quantities approach


In place of Fig. 8b, we could also represent the open string propagator

as a disk that is cut out of a plane which parametrizes the world-sheet.26 In

principle, we can recover the situation discussed in this chapter if we impose

Neumann boundary conditions on the hole, and then factorize by restricting the

locations of the vertices. An alternative case is when the hole obeys Dirichlet

boundary conditions. In this case, the open string propagator is physical only

when there is zero momentum across it. Since the sister trajectories occur in

the limit of large s momentum transfer squared, we can rule out their existence

in the Dirichlet theory.


In this chapter we will determine the state representation of the sister tra-

jectories. The basic nature of sister states will differ from the states associated

with the standard Regge trajectory a(t) since the corresponding sister poles are

not physical. In the space-like t region, the poles have the manifest unphysical

characteristic of nonsense, i.e., negative spin J. The time-like resonances are

not physical either. In exact expressions, the residues associated with physi-

cal resonances can always be written as polynomials in the energies.10 At tree

level, we can see from the final expressions (2.21) and (2.37) that the energies

overlapping the sister propagator are not in this form. However, this is not the

case in the double-loop expression (3.32) where only a single energy appears.

Here, repeating the argument given for the six-point case in Chap. 2, we rely

on the fact that the sister has wrong-signature. Constructing the sister states

will give us another means for uniformly displaying the unphysical nature of

the poles in all regions of t.

There exists, yet, a second motivation for being interested in the state

representations of the sisters. At about the same time sisters were found in

the high energy analysis, they were noticed in an entirely different context by

Goldstone27 who was investigating the problem of counting physical states at

each mass level.28,29 This is a nontrivial problem because in D dimensions the

physical states transform under the group O(D 1), whereas, due to gauge

invariance, string states fill multiplets of the transverse group O(D 2). For

the case of four space-time dimensions, the counting problem was solved in

1976 when Goldstone presented the generating function
00 00
x(x, J) = [E(1 n2] XrJ+r(r-l)/2()r(-l(l xr)2. (5.1)
n=1 r=1
This has since been generalized to higher dimensions, and for the Superstring

and Heterotic string.29'30'31 Expanding out (5.1), the coefficient of x counts

the number of 0(3) representations of spin J, while the exponent is the corre-

sponding mass level. The connection to sisters can be made if in (5.1) one sets

the x exponent in the second factor equal to M2, i.e.,
M2 1
Jr(M) = (r 1). (5.2)
r 2
Since poles in the a(t) plane are labeled by (M2, J), we can identify (5.2) with


The state analysis of Goldstone is in the time-like t region, while the high

energy analyses exposes dominant behavior in the space-like region. Determin-

ing the state representation of the sister trajectories will provide a more direct

link between these two approaches. The unifying feature of pole cancellation

can be seen in Fig. 9 which, for a(t) > 0, displays the lowest mass levels

obtained from Goldstone's formula (5.1). The figure shows how the various

trajectories conspire to form the physical states(solid dots) and remove some

of the pure gauge states(crosses), and that the 3(t) trajectory in both regimes

enters with the opposite sign to the a(t) and *y(t) trajectories.

The state representations of the physical states, defined at the poles of

the standard Regge a(t) trajectories and its daughters, are well known. The

first three states of the leading trajectory displayed in Fig. 9 are given by the

tachyon | 0), the "photon" a_ | 0), and the massive spin two symmetric state

a1[i-ai 0), where the transverse index i = 0, D 2. Suppressing the

space-time index, the general leading state is given by an 0), n > 0. It is

J a(t)


cc X

a"02a-2 (t)

Figure 9. Lower mass states in the open bosonic string. Dots denote phys-
ical states, and crosses denote pure gauge states. The coefficients indicate con-
tributions from the various leading and daughter trajectories. Nonsense poles
are from high energy scattering analysis, and sense poles are from Goldstone's

important to note that the mode number of the states along the a(t) trajectory

differ by one. This implies that by varying nr in the general open string state

-na n-- a nr--- 0), (5.3)

we move along a path in Fig. 9 that parallels the rth sister trajectory. Al-

though the poles of the rA sister trajectory do differ by mode number r, the

corresponding sister states can not be represented by the physical states (5.3).

Instead, by analogy with the high energy analysis, we must analytically

continue away from the states defined by (5.3). To proceed, we will work

directly with the factorized scattering amplitude. This isolates the appropriate

propagator which allows us project onto it all possible classes of physical states.

We are then free to select the states which lead to the sisters. For the six-point

diagram of Fig. 1, again ignoring the twists, we project the physical states
onto the central propagator. To preserve unitarity, we insert the corresponding
identity operator on adjacent sides of the propagator, i.e.,

A6 = dzld30,p I V(p2,z-l)V(p3,)Io 1IV(p4,1)V(p5,z3) 0,P6),
JQ zlz3 L 1


n00= 00
I = E-- ansla) 0) E 2, -1 2 '(
n=0 .n2=0
The normalizations in (5.5) are fixed by the projector condition 12 = I, and
the commutation relations

[a4a, 4] = mrnm+nrl7'. (5.6)

When we project I onto the central propagator, we easily obtain
1 00 00 1 1
,I I I- n, J0) ( anr r (57)
S 1 rn nr! 1 n ljy a(t), (5.7)
r=l nr=0
where t t2. Substituting this into (5.4), using the four-dimensionality con-
straint (2.9), and then taking the high energy limit s34 -+ oo, yields

Ag j dz1dz3zz 1-a(tl) z 1-(t3)(1 zl)-1-a(s23)( z3)-1-a(s4s)

r=1 n,=

where x 1, x3 and 9 are as before. Note, by using the integral representation

jnj_ -a(t) jdzzr1=,-(1)-I (5.9)
we can replace the sums in (5.8) by exponential functions to get

A6 1 dzdz3dzz1-(t)Z 1-a()(l zl)-l-a(s23)(1 z)-l-Q(S4)
x z-1-a(t)exs
X ~ exp I X 3- x3

Thus, we have completely recovered Eq. (2.10). In fact, at virtually each step

of our computations below, there is a parallel step using the exponentiated
form. This provides a useful check on our results, and allows us to be brief in

much of the derivation.

As in Chap. 2, to obtain the leading /3(t) trajectory we require that (xl, x3)
be a critical point. In this case, not only do we twist the propagators zl and

z3, but the index n1 must be analytically continued to a negative value. The
standard procedure is to replace the infinite sum by a Sommerfeld-Watson

contour integral, and then push back the contour exposing the poles on the
negative real axis. To generate the necessary pole in n1 we first evaluate (5.8)

at the critical point. About (xl, x3), the amplitude (5.8) is approximately

A6 ~ x -c(txl)X3 1-a(t3)(1 Xl)-l+a(s23)(1 x3)-+a(s45)I6, (5.11)

where, after shifting z1 and z3, the integral becomes
r00 00 r

Or=2 n=O (5.12)

x 1 1 dzldz3(gzlz3)nl
nl=0 n1 j a(t)

The double integral is easily performed, giving

16 =2N o1 11 1 [K X) (r x1]3 r
r=2nr=O (5.13)

x y-+ 1 1 (-1)n+1 (
n= nl! = l =1jnj a(t) (nl + 1)2
where yo = e2.. As we will see shortly, it is crucial for the sister that n1 appear

as a double pole.

As a prelude to replacing the discrete variable ni by a continuous one, we

must replace n1! by its gamma function representation F(nI + 1). To analyt-

ically continue to the pole at nI = -1 we must be careful since Eq. (5.13)

vanishes for odd n1 because of the numerator in the last factor. Using the
Sommerfeld-Watson transformation to convert the sum over n into a contour
integral, we get

1 1/E \llr
r=2 n.=0 (5.14)

S dx yz+1 1 1 (-1)x+1
x sinxr(x+l)x+E=2jnj-a(t) ( + 1)2 '
where we have separated out, and displayed, the odd n contributions. To
continue back to the pole, we must first signaturize the last factor by setting
-1 = e". Pushing back the contour then exposes the double pole with residue

16 =29-1 J00 00 -n X1)Xr -X3)]
r=2 nr=0 (5.15)
d[ 1 y+ 1 eir(x+l)
dx sinrxr(x+1)x + jnj -a(t) x=-1
Since this expression vanishes when we set x = -1 in the last factor, we only
need to differentiate this term. The result is -ir. The reason why we require
a double critical point should now be clear. If only z\ or z3 were critical, a
single nl pole would result whose residue vanishes.
Next, in the limit x -+ -1,

sin x7rx(x + 1) -- 1. (5.16)

00 00 1 r X Xr n
I6 =- 2i7r-1 11 E rnrnr! ) 3 r
r=2 n,=0 (5.17)
1 + Ej-=2 IJnj a (t)
In the special case nr = 0 for r = 2,3,..., we have the nonsense pole at
a(t) = -1, i.e..,
6 = -2i 1 a(t) (5.18)

Since the analytical continuation was along the curve described by an"1 0), it
is clears that this pole is generated by the leading Regge trajectory a(t). The
analysis above shows that this pole is given by "(a_1)-1 | 0)". The inverse
oscillator indicates the unphysical nature of this state.
Now, we must explicitly show that the pole (5.18) is canceled by a cor-
responding pole on the leading first sister trajectory, /3(t). To analytically
continue to this pole along /3(t), we must convert the n2 sum in (5.13) to a
contour integral

1 -10 n+ 1 (-l)n+1
16 = g E c
rZ =O nl! (ni + 1)2
n=O "(5.19)
f dx 1 1+2 2 A _
C sin rx 2r(x + 1) 13 n + 2x a(t)'

where we have set nr = 0 for r = 3,4,.... Picking up the pole at x = 2a(t) -
nn1 gives

2-1 Y 1+1 1 (-1)n1+1 1
0 n1! (nl + 1)2 sin r(a(t)- ) (l)
nl=0 (5.20)

rF(a(t)- 1ni + 1)) x3)]

Repeating the steps for n and evaluating at the pole nl = -1, we finally

16 = _ir25/(t) 1 1 2 Xl (2- x3)](t)+x (5.21)
sin 7r(f(t)+ 1) r(f(t)+ 2) X1)X3x3) (5.21)

That the first pole at f3(t) = -1 cancels the amplitude (5.18) can be seen by
writing P(fo(t) + 2) = (f/(t) + 1)F(/3(t) + 1), and then canceling the pole coming
from the sine function against the zero in r-1(/(t) + 1). The remaining poles
are represented by the states (a_)-la on 0) and cancel unphysical poles
generated by the daughters of the a(t) trajectory.

Repeating the six-point computation for a general critical point defined by

z4 = x\ and z = x3 produces the (r 1)h- daughter of the /(t) sister. The

complete set of states corresponding to /(t) and all its daughters is given by

(a-1)-ma" 2 I 0), for m > 0 and n2 > 0. The set can only exist in totality,
and results from a complete saturation of the propagator with the oscillator

a-1. In the present context, we see that the 0/(t) sister does not appear in

either the four- or five-point function, or on the zi and z3 propagators of the

six-point function, because coupling the propagator to two on-shell states at

any vertex prevents total saturation.

This analysis suggests that to obtain the second sister trajectory, Y(t),

we must first saturate the appropriate propagator with the oscillator a-i,

permitting 03(t) to exist, and then with a-2. To verify this, we again consider

the eight-point diagram of of Fig. 3. Inserting the identity operator on adjacent

sides of the propagator z3, and using the four-dimensionality constraints, yields

the form
A8g dzldz2dz4dz5f(zl, z2, z4, z5)
oo oo 1 1 (5.22)
Z1 rnrnr! an (lj (t)(Hr
r=1 nr=0
where t t- 3, and

Hr = (z xl)(z x2)(z4 x4)(zr x5). (5.23)

and where the x's are as before.

Recall, we can obtain the leading 7(t) trajectory if we twist all the noncen-

tral propagators and expand about the point z1 = xl, z2 = x2, z4 = VI--, and

z5 = V/px. For (xi, x2) and (/4, v/i-) to represent double critical points will

also require we continue both nl and n2 to negative values. Assuming this to

be case, we have

A8 f(x1,x2,' /,vV')0 0 E 1 (IHr)nr dzldz2dz4dz5
r=3 nr=0r r -

x [gzlz2(v4 X4)(x5 x5)]

0 1 x 2(x _)( 2 )]n 1
n2=0 2n2

Each pair of integrals is the same form as in (5.12). The first pair generates
a double pole at nl = -1 and moves us onto a 3(t) trajectory. Subsequently,
the second pair gives a double pole at n2 = -1 which now transfers us to a

7(t) trajectory.

Integrating and analytically continuing to nl = -1 we immediately find

Ag ~ 4i7r (22x4x5(x x)(2 X2)(1 V/f4)(1 V 1))
00 00 1
x f(xl1,x2, vX, v ) Hr)n (5.25)
r=3 nr=0
00 2+1 1 (-1)n2+1 1
Sn! (n+ 1)2 -1 + -
where the exact form of vo is not important. We can approach the final state

(a_la_2)-1 10) by moving along either the leading /(t) or 7(t) trajectories.
The nonsense poles obtained in two cases must cancel. For the first path, we

set nr = 0 for r = 2,3,..., and perform a Sommerfeld-Watson transformation

on n2, to get
A8 47r2 22x445(x2- x1)(}x x2)(1 v/)(1 1))-
1 / (5.26)
x f (xl, x 2 x4, V x- 3 t) t '
Similarly, for n3 : 0, sliding down the -(t) trajectory we obtain the result
4 2 x o \-1
A8 7r 2(2x4x5(x2 x)(x2 x2)(1 Vi)(1 V/)) 1
S 1 (5.27)
x f(x1, x2, V 4, ) H)+ )()2 1 1 3)
3 sin r(7(t)t + 2) r(7(t) + 3)

where, for -7(t) = -2, this cancels the pole (5.26).

Extending these results to the most general case, suggests that the rth

sister trajectory forms when the propagator becomes successively saturated by

the oscillators a-m, starting with m = 1, and eventually reaching m = r-1. In
another words, to get to the trajectory am(t) we begin by moving down either

a leading or daughter a(t) trajectory curve to either a leading or daughter 0(t)

trajectory, which we reach by analytical continuation, etc. The resultant sister

and its daughter trajectories are represented by open string states of the form

(a_1)-ml ... (ar+1)-m (a-r)"r 10), for m1, -,mr-1 > 0,nr > 0,

where the leading trajectory is given by mr = ... mr1 = 1, nr = 0.

By analogy we can immediately write down the corresponding closed string

sister states by replacing the open string oscillator a_- with the closed string

oscillators a_i&_i everywhere.

An important point that needs to be stressed here for applying the proce-

dure we have presented, is that it be possible to completely isolate the sister

propagator. In the case of the double-loop four-point interaction, this crite-

rion adds justification to our approach in Chap. 3 where we factorized the

amplitude so that we could treat as individual objects the two loops and the

connecting propagator. In the state analysis approach, projecting the physical

states onto the connecting propagator gives
1 1 01 4
A(4;2) = H E rnrnr! ,nj a(t) d dzi
r=1 n=0 i=1

x (0| exp[ + CL a(c n, ,(c n 0) (5.29)

x (0 (a ) p- CI) ,o>

After simplifying, the calculation leading to the sister proceeds exactly as in
the six-point case given earlier in this chapter, and reproduces the results of
Chap. 3.
Exposing the open string sister in the four-point closed string diagram of
Fig. 8 presents a new difficultly, however, since we must look for a triple Regge
pole that also is not a leading order term. The expression

AG2 =( 2 ) 1d2zld2z2cl(0,p4 Vf(p3,z l,1 ) x o(O I T(At,a) i )c,

x IAoI x c(0 T(A,at) | 0)o x V(p2,z2, 2) 0,Pi)ci.

reduces, after some algebra, to
00 00
AG2 12 nr 1 d2z1d2z2
47 rnnr! = jnj + 3 2a(t) ,1<
r=1 nr= 0
I zlz2 I-2-(t) [-2p2 p3(z/2 + /2)( r/2 + r/2)ven (5.31)

x [2P2 P3r2 Cr n z- )( z- 2
We used the fact here that the even and odd oscillator parts can be treated
separately. To obtain the open string /3o(t) trajectory we set nr = 0 for
r = 3, 4, ..., which allows the sums to carried out. In the high energy
limit, we find

1 2 1 1 1 d2zld2z2
AG2 ) ( n1 n! 2n2n2 n1 + 2n2 + 3 2a(t) J< d1d

x I z12 -2-a(t) [-s(z1 + )(z2 + z2)] n2

x [s ((1 zl) (1 )) ((1 z2) (1 -2))]n

Of course, using the integral representation for the propagator we can easily
recover the corresponding expression of Chap. 4.

Now set z = pei'. Expanding about the critical points at 0 = 0 and r gives
Z 1 2n2 1 1
AG2 =2 + 2n + 3 (t) dpd2(P1P2)n+n2-1-at
ni=O n2=0
sni+n2 E d01 d02(9) (1 P 2)( 2 ]

+ [(1 + pI)0(1 + p2)-1()n1n2),
where the first term in the last factor is for 01 and 02 being expanded about

the same value, while the second term is the converse case. The 0 integrals are
executed as before, giving (1 + (-)nl)262nl+2/(n1 + 1)2. Each p integration

produces a factor B(n1 +n2 -at, 1- n1). Combining the terms then produces

the factor (1 + (-)n1+n2). This leads to the result

Ak=2 = 1 1 2n2 1 sni+n2
47r2 n2 n1 + 2n2 + 3 2a(t)
n=0o n2=0
1 2 e2n"+2
x B(n + n2 (t), 1 l)2 + (-n2)(1 + (-)nl) E 12,
2 (n1 + 1)2'
where we have used (1 + (-)nl)2 = 2(1 + (-)nl).

Utilizing the Sommerfeld-Watson transformation, the residue due to the

double pole at n1 = -1 yields the result

Ak=2 1r) 2n2 1 + (-)n2 B(- 1 + n2 a(),3/2)2-1+n2
472 =(-i2! 2 + 2n2 2(t)

Writing the Beta function in terms of F functions, and pulling out the first

three poles from one of the F's gives

Ak=2 =-i 2n 1+(-)fn -i+n2
87 n 2!(1 + n2 a(t))3
n2=0 (5.36)
x [(-1 r(2 + n2 a(t))r(3/2) ] 2
(-1 + n2 a(t))(n2 a(t))F( + n2 a(t))J

Thus, we have recovered the triple pole at n2 = a(t) 1. Computing the
residue by taking the second derivative of the energy factor, gives the final

i9 1_ -_+__8_ _
Ak=2 = 9+(t 1 e- 1 Io(t) in2 s. (5.37)
7r sin 7r(,3o(t) + 1) LF(lo(t) + 2)

To show that these poles cancel, we again start with (5.34) but now pick
up the single pole at n1 = 2a(t) 3 2n2 to get

k=2 1 0 2n2 F(-3 n2 + a(t))r(5/2 a(t) + n2) 2
4 n--2 2 r(-1/2)
S2a(t)-3-n2( (-)n2)(1 (_)2a()-3-2n) 4a(t)-4-4n2
(2a(t)- 2- 2n2)2

We must take the residue of the quadruple pole at n2 = a(t) 1, which will
give a factor of 1 For this, we take two derivatives of the energy factor and,
to get a nonvanishing result, one derivative of the factor (1 + ()2a(t)-3-2n2),
which can be done 3 ways. This leads to
Ak=2 31 1 1 + e-i(a()-) 1 r F(3/2) 2n2 s

3! 8 7r2 sin7r(a(t) -1) r(a(t)) [(-1/2)J (2s)a(-2(2)ln2s'

which reduces to the negative of (5.37).


In this thesis we have indicated that the Cerulus-Martin bound may not

be violated in string theory if one includes higher order corrections to the tree

diagram. To complete the proof requires that the entire perturbation series

be summed to determine if the coefficients of the amplitudes have any effect

upon the result. It is not clear, however, how to take the fixed angle limit in

the high energy analysis employed above. Instead, the proper approach may

be to adopt the techniques used by Gross and Mende.4 Basically, for closed

strings, this means searching for saddle points on an N sheeted Riemann sur-

face defined by an appropriate algebraic curve. To have the sisters produce the

dominant behavior would require that we consider the limiting situation where

the Riemann surface is divided in two, separated by a thread representing the

sister propagator. The hope is that the uncontrollable higher order corrections

which plagued the work of Ref. 4 would now be absent.

Although this may remove one of the objections to locality, it should not

imply that in any way have we shown that string theory is, or can be, a local

theory. There still remains serious objections which may be more difficult,

if not impossible, to overcome. For example, in string field theory, Eliezer

and Woodard6 note that the cubic formulation of the field theory produces an

infinite number of Abelian solutions.32 This causes a breakdown of the initial

value problem since it requires an infinite amount of initial data. They show

that attempts to restore this loss of predictability result in acausal behavior,

which, again, leads to nonlocality. Another problem in string field theory, is

that the individual elements of the perturbative S-matrix still violate the CM

bound. Possibly, with the help of the sisters, one can find a local field theory

where strings are produced non-pertubartively and appear as bound states.

This would probably resolve most of the locality problems in the field theory.

We should point out, though, that it is not even clear if field theory should be

the fundamental formulation for strings.

Finally, let us comment on a physical interpretation for the sister states.

Recall, the Regge slope a' is related to the string tension T, or energy per unit

length, by

a (6.1)

This shows that, for instance, the first sister 03(t) has twice the tension of a(t).

We visualize this occurring by bending over the string once to create a double

strand, giving a 'folded' string. This picture is in accord with a reduction of

the fundamental length scale I = vW1. The notion of folded strings originally

dates back to the early 1970's where it was noted that pure states of the form

anr 1 0) have Regge slope reduced by a factor of 1. Thus, the state analysis of

Chap. 5 furthers the identification of the sister trajectories with folded strings.

In conclusion, the motivation for studying sister trajectories is that they

may eventually lead to a useful description of the short distance behavior of

string theory. Presently, there is a growing belief that the current version

of string theory is nonlocal at a fundamental level. The sister trajectories,

and their interpretation as folded strings, may be the necessary ingredient for

reformulating string theory to produce a local theory.


Recently, matrix models have received a great deal of attention as nonper-

turbative descriptions of string theory.7,8,9 Since initial advancements, progress

has proceeded in many different directions. In particular, Douglas33 has shown

that the limited number of known matrix model solutions can be derived from

the Lax pair formalism usually associated with the KdV equations. This iden-

tification with integrable systems greatly increases the number of classifiable

matrix models since it was shown a long time ago34 that Lax operators are

associated with affine Lie algebras. For example, the models discussed by

Douglas are related to the canonical representations of A(1). More recently, Di

Francesco and Kutasov35'36 have discussed D(1) based matrix models which

the standard matrix techniques37,38,39 have yet to solve. Thus, it may be

worthwhile to focus on the integrable systems approach.

Several approaches to constructing the Lax operators have been developed.

The matrix procedure discussed by Drinfel'd and Sokolov34 defines first a ma-

trix eigenvalue equation. The system incorporates knowledge of the Cartan

subalgebra and root system of some embedding affine Lie algebra g. Starting

with an affine Lie algebra facilitates the construction of an integrable system

from the resulting Lax pair operators. To fix the gauge invariance in the matrix

system, the gradation conventions of Drinfel'd and Sokolov require that one of

the simple roots, say the mth must be removed from the affine system. The

resulting system is denoted by (, cm,). This is equivalent to deleting the mth

Dynkin vertex. For the most part, Drinfel'd and Sokolov choose the "canoni-

cal" gauge in which to express the coordinate dependent terms. In this gauge,

Lax operators generate the regular KdV hierarchy equations.

The modified KdV(mKdV) equations can be generated by expressing the

coordinate term q(x) in the "diagonal" gauge. The canonical Lax operators can

then be recovered using the well-known Miura transformations. The diagonal

gauge is technically simpler than the canonical gauge. Furthermore, the final

Lax operator is in a factorized form which has been used to quantize the


In this appendix our focus will be on the explicit construction of the Lax

(pseudo)differential operators in the diagonal gauge using a simple diagram-

matic technique.41 In most cases this technique arrives at these operators much

quicker than a direct application of the scheme of Drinfel'd and Sokolov. Fur-

thermore, the scheme also applies to higher representations of the embedding

affine Lie algebra. In the first section we briefly review the construction of

weight diagrams corresponding to representations of affine and non-affine Lie

algebras. From there we review the matrix method of Drinfel'd and Sokolov

for building Lax operators. Next is a presentation of our method, which re-

places the matrix procedure with a scheme utilizing cyclic weight diagrams

of representations of affine Lie algebras. We then present a proof that the

diagrammatic algorithm produces the correct Lax operator. Finally, we dis-

cusses the generalization to Lax operators based on supersymmetric affine Lie


Review of Weight Diagrams

As noted in the introduction, each Lax operator can be associated with

a representation of some affine Lie algebra. Thus, in this section, we give a

brief review for constructing weight diagrams corresponding to these represen-


Recall, one can associate uniquely to every irreducible representation of a

basic Lie algebra a highest weight vector.42 For each highest weight one can

construct a weight diagram which encodes all relevant information concerning

the particular representation studied, e.g., from it one can build explicit matrix

representations of the generators of the Cartan subalgebra as well as the various

raising and lowering operators. The level of a weight is the number of lowering

operators applied to the highest weight which produces that weight. Finally,

the height 7 of the weight diagram is the level of the lowest weight.

Weight diagrams are generated by subtracting rows of the Cartan matrix

initially from the highest weight vector written in the Dynkin basis. Rules of

construction can be summed up as follows:

1. Subtract the iii row of the Cartan matrix n times from a weight vector

whose iAh component has a positive value n.

2. When weight vectors have more than one positive component, subtract all

possible permutations of the appropriate Cartan rows.

A theorem due to Dynkin43 states that the final weight diagram is always

"spindle shaped". In other words: i) the number of weight vectors at the level

k is equal to the number at level -y7 k, ii) the number of weights at level k + 1

is greater than or equal to the number at level k for k < i.

For an explicit example consider the algebra A2. Though this is almost a

trivial case, the results will be useful for the next section. The Dynkin diagram

is given by
1 0


where, recall, the single bar represents 1200. The Cartan matrix is then easily

found to be
A=(2 -1). (A.1)

The highest weight vector of the fundamental representation is (10). 44 Since a

positive one appears in the first place, we subtract the first row of the Cartan

matrix one time. This gives the weight (-11). Now, due to the one in the

second place, we subtract the second row of the Cartan matrix once to get

(0 -1). This completes the process since no positive components remain. The

result is the height two weight diagram
(-11)2 (A.2)


where the subscripts on the weight vectors indicate a counting of the vectors.

The ones adjacent to the arrows represent the normalization factors of the cor-

responding negative simple roots. These values are fixed by the commutation

relations of the Lie algebra. To simplify our diagrams, we will not display val-

ues of unity. Later, we will see that the procedure for building weight diagrams

is slightly modified in the supersymmetric case.

For an affine Lie algebra,45 since there exists a linear combination among

simple roots, weight diagrams of affine representations generally have infinite

extent. However, some affine representations give cyclic weight diagrams of

finite extent. In fact, it is these cyclic cases that are crucial to the scheme

below. To produce a cyclic weight diagram, the affine component which is

appended to the highest weight vector of the underlying non-affine Lie algebra,

unlike the non-affine weight components, may have to be assigned a negative



2 -1 -1

A -1 -1 2

(a) (b)

(-1 1 0 ) -,

(0 -1 1 )
4, I
( 1 0 -1)--


Figure 10. The (-110) representation of A'). (a) Dynkin diagram; (b)
Cartan matrix; (c) Cyclic weight diagram. The dashed arrow is the deleted

As an explicit example, consider the non-twisted affine algebra A1). To

generate the cyclic weight diagram corresponding to the canonical representa-

tion, start with the weight vector (-110), where -1 corresponds to the affine

root. Figure 10 gives the Dynkin diagram and subsequent Cartan matrix which

then generates the displayed resultant weight diagram.

This particular cyclic weight diagram can further be thought of as the affine

extension of the highest weight diagram based on the fundamental representa-

tion of the basic Lie algebra A2. This is easy to see by removing everywhere

the component due to the affine root. However, this is not always the case.

For example, Fig. 11 displays the cyclic weight diagram constructed with the

weight vector (-211), where now the affine component is -2. Although (11)

(-2 1 1)

/I \
(-1-1 2) (-1 2 -)

\ L4- /
L4 ----- ------
(1 1 -2) (1 -2 1)

(2 -1-1)

Figure 11. Cyclic weight diagram of A(') from the weight (-211). The
dashed arrows are the deleted root.

(1 1)

(-1 2 ) ( 2 -1)

(0 ) (0 0o)
4, ,4,
( 1 -2) (-2 1 )


Figure 12. Highest weight diagram of the adjoint representation of A2,

generates the highest weight diagram of the adjoint representation of A2, we
see by comparing with Fig. 12 that the affine extension contains an extra zero
weight (000).

In general, the affine component in the affinely extended vector, associated

with the highest weight vector of a basic Lie algebra, will always be negative.

However, we will give an example below showing that some supersymmetric

cases require positive affine components.

Standard Construction of Lax Operators

The Lax operator46 L(x, t) is defined to be linear and Hermitian. Further-

more, it satisfies the characteristic equation

L(x,t)o(x,t) = yo(x,t), (A.3)

where the eigenvalue p is required to be constant under nonlinear evolution. In

other words, the nonlinear behavior of the eigenfunctions O(x, t) are governed

by some operator A(t), which may be nonlinear, via the equation

t A(t)(x,t). (A.4)

Futhermore, A(t) enters into the differential scalar Lax equation

OL( t) = [A(t),L(x, t)], (A.5)

which generates the integrable KdV equations.

The matrix construction of Lax operators utilizing generators of some em-

bedding affine Lie algebra ?(k), reviewed by Drinfel'd and Sokolov,34 begins

with a matrix operator of the form

L = I + A + q(x), (A.6)

where I denotes the N x N dimensional unit matrix, and to simplify notation

we have suppressed the argument t. The third term is discussed below. The

second term is generated by the negative simple roots Ei of the embedding

affine Lie algebra. In the gradation conventions of Drinfel'd and Sokolov, we

have the circulant matrix
A = ciEi (A.7)
The procedure is then to reduce the system of linear equations given by

the kernel matrix equation

L(x) = 0, (A.8)

where O = (N',... N), to the linear differential eigenvalue equation (A.3),

where the vacuum solution 4 is a function of the components of the eigen-

function b. Drinfel'd and Sokolov show that such reduction is possible if one

removes a simple root, say the mth from the affine root system. They denote

this situation (O(k), cm), which is in the homogeneous or standard gradation.47

The coefficients in (A.7) are then assigned the values ciom = 1 and Cm = A,

where A is a constant function of the spectral parameter z.

Removing an element from the simple root system is equivalent to deleting

the corresponding vertex from the Dynkin diagram. Thus, when an extremal

vertex is deleted, the system (G(k), Cm) represents a single residual basic Lie

algebra. Removing the affine vertex obviously gives G. On the other hand,

deleting internal vertices splits the Dynkin diagram into two sections, corre-

sponding to a pair of basic Lie algebras. For example, splitting A(2) at the

mrh vertex gives Lax operators in the Bn-m and Cm series. Furthermore,

the (pseudo)differential operator associated with the Dn series is derived using

the embedding algebra D() In both situations, a pair of (pseudo)differential

operators is found whose product gives the Lax operator L of Eq.(A.5).

To determine the exact form of the vacuum solution 4 recall that negative

simple roots are lowering operators on system eigenstates. Further, removal

of a root in the affine system produces the simple root system of a non-affine

Lie algebra. Thus, due to the linear combination among the roots of the affine

system, one root must be singled out to act as a conventional state raising

operator. This role is given to the removed root. Thus, the vacuum eigenstate

will be annihilated by a vacuum projection operator A- defined by
A = Eiom + AEm = A- + AEm. (A.9)
This requirement fixes the scalar vacuum solution 0 by setting it equal to a

linear combination of the components of such that

A-= 0, (A.10)

is satisfied. A direct relation between the scalar operator L and the matrix

operator will be given in the next section.

For the kernel equation (A.8) to produce a unique solution, we require that

the number of independent degrees of freedom equal the rank of the embedding

affine Lie algebra O(k), or equivalently the residual system ( (k), Cm). The extra

degrees of freedom generate gauge invariance. To fix the gauge invariance, one

must find a matrix operator S(x) that enforces the gauge transformation

0 = ead S, (A.11)

where ad denotes the adjoint mapping. The gauge freedom in Eq.(A.8) al-

lows one freedom in determining the form of the coordinate dependent term

qo(x),i. e.,

C O = I + A + qo(x). (A.12)

Drinfel'd and Sokolov find the sufficient condition that S E CX(R rl), where

rT is generated by the positive simple roots Fi, i /- m.

Many authors, including Drinfel'd and Sokolov, work most frequently in

the "canonical" gauge. However, in this paper we choose to work in their

"diagonal" gauge which has the form
/q 0 ... 0 0
0 q2 ... 0 0
qdiag(x) = : .. : (A.13)
0 0 ... qN-1 0
0 0 ... 0 qN

This gauge leads to the convenient form
qdiag = E vi( q2",., qN)Hi, (A.14)

which is in the canonical or principal gradation.47 Here, Hi are the generators

of the Cartan subalgebra and the functions vi are linear combinations of the

elements qi. In this gauge, the gauge term qdiag associated with (g(k), Cm)

is the special case where the sum excludes i = m. The Lax operator Ldiag

generates the mKdV equations, and is related to Lcan via the well-known

Miura transformations.

Diagrammatic Construction of Lax Operators

To exploit gauge invariance of the Lax operators, one should choose a q(x)

gauge most suited to ones needs. Here, we are interested in developing a

diagrammatic scheme for constructing L. In this regard, the diagonal gauge

proves more useful than the other choices. In this section, we will demonstrate

how the diagonal gauge allows one to build Lax operators directly from cyclic

weight diagrams of representations of affine Lie algebras.

To motivate the algorithm, we first review the construction of L by solving

the matrix system L = 0. For the present discussion, it will be sufficient to

consider embedding algebras of the form (g(k), co) where the affine vertex is

deleted. Thus, the diagonal gauge simply reduces to the form
qdiag(x) = i(x)Hi, (A.15)

where we have excluded H0 from the sum.

Consider again the canonical representation of the embedding affine Lie

algebra (A21, co) presented in Fig. 10. Matrix representations of the Cartan

matrix can be read-off from the cyclic weight diagram. The matrix element

(Hi)jj is extracted from the ith element of the j1th weight vector, while the off-

diagonal elements are set to zero. The matrix entry of the negative simple root

(Ei)jk is assigned its normalization factor if the kth weight vector branches

into the jth weight vector as a result of subtracting the iA row of the Cartan

matrix in the process. The other entries are by default zero. Thus, the matrix

representations of the simple roots are easily found to give

A = (1 0 (A.16)
0 1 0

where the effect of the affine root, indicated in Fig. 10 by the dashed arrow

line, is assigned the value A. Plugging these values into the kernel equation

(A.8) produces the system of equations

[09+ vl]b1 = -A3,

[0- v + v.'L, = -i01, (A.17)

[0 V.] 3 = -2

Here, on the right-hand side we have placed the terms due to the matrix A.

The vacuum condition (A.10) determines the scalar function to be 0 = ,'3.

Thus, we must solve by starting with the last equation. First, we multiply

this equation through by [9 v1 + v2], and then eliminate ('.? using the second

equation. Then, multiplying through by [0 + vi] and using the top equation

gives the scalar Lax eigenvalue equation

L(A2 o) = [0 + [ -V1 + v21][ 2]3 (A.18)
= -A3,

where the spectral parameter is given by p = -A. Imposing the field redefini-


qI = vi, q2 = v2 v1, (A.19)

gives the standard form

L(A(,co) = [9 + q1][9 + q2][( 1 q2]. (A.20)

This example exhibits a common feature relevant for our scheme below.

When the vacuum condition (A.10) requires the scalar eigenfunction to be

given by a single component of the eigenfunction, say = ,, then the resulting

characteristic equation satisfies

Ly, = p ,. (A.21)

Consequently, the system reduction must start with the ith equation in the

matrix system, and proceed upward till the top equation is reached. If i < N

the process continues with the bottom equation and moves upward until the

ith equation is reached again. We shall refer to this case as trivial since the cor-

responding cyclic affine weight diagram is linear, containing no branch points.

A second feature brought out in this example, is that the number of factors

in the resultant Lax operator (A.20) is equal to the number of weights in the

weight diagram. Unfortunately, this is valid only for trivial cases. Nevertheless,

this last observation is key to our scheme.

To highlight one more property of the general procedure we turn to a non-

trivial example. For this, we require a representation of an affine Lie algebra

whose cyclic weight diagram has at least one branching point. Thus, consider

the canonical representation of the affine algebra (D1), co). Fig. 13 presents

the Dynkin diagram, Cartan matrix and corresponding cyclic weight diagram

( 2 0 -1
0 2 -1
-1 -1 2
0 0 -1
0 0 -1

0 0
0 0
-1 -1 i;
2 0
0 2

-----"- (-1 1 0 0 0)
I 4,
S(-1 -1 1 0 0) ------
(0 0 -1 1 1)

(0 0 0 -1 1) (0 0 0 1 -1)

(0 0 1 -1 -1)

-----------(1 1-10 0)
-1 0 0 ---------
(1 -1 0 0 O)-------


Figure 13. The (-11000) representation of D4. (a) Dynkin diagram; (b)
Cartan matrix; (c) Cyclic weight diagram. The dashed arrow is the deleted

which has two branch points. Reading off from the weight diagram gives

0 0 0

0 0 0


The branch points have manifested themselves by placing more than one non-

zero entry in the second and sixth rows. Now, further reading off the elements

of the Cartan matrices gives the system of equations
[8+ v+] 0 = -A07,

[8 v1 + V2 ]2 = -01 A08,

[8 V2 + V3 + v4]03 = -02,

[8 v3 + v4]4 = 3,
[8 + v3 V4]s = -03,

[8 + v2 6 = 05,

[8 + v1 v2]07 = -06,

[a Vl]0,8 = -b7.
The vacuum condition (A.10) produces two distinct solutions, '8 and the
linear combination 04 bs. Here, we consider the first case. Proceeding as
before, we eliminate 46 and 07 in the last two equations to get

[8 + v2 va v4][8 + vi v2][8 vi]08 = -04 05. (A.24)

Now, we encounter a well-known technical problem not found in the trivial
case. The components 04 and 05 can not both be simultaneously eliminated
since the expressions [8 v3 + v4] and [8 + v3 v4] do not commute. This
dilemma is directly linked to the fact the corresponding cyclic weight diagram

has a branch point connecting the fourth and fifth weights to a single weight
located below them.
To overcome this obstacle the pseudo-differential operator p-1 must be
introduced. Its operation on any function f(x) is given by the expansion

81lf(x) = (-1)if(i)(x)8l1-i. (A.25)
Utilizing the pseudo-differential operator, we rewrite the fourth equation in
(A.23) as

04 = -[0 -V3 + V41-1


Thus, the combined effect of the bottom five equations is

03 ={[9 v3 + v4-1 + [ + v3 v4]-1}[ + v2 v3 V41
x [0 + v1 v21[Q v1]8-
A helpful identity we use repeatedly is

{A-1 + B-}-1= {A-[A + B]B-l}-1 = B[A + B]-'A. (A.28)

When applied to Eq.(A.27), a cancellation occurs among the via's appearing in
the curly brackets. This simplifies the expression to

03 = -[--v3 +v4]-1 [O +v3 v4] [ +2 v3 4][+VI v2][-v]s. (A.29)

Continuing, incorporating the next two equations in (A.23) requires a second
application of the relation (A.28). Finally, the Lax operator based on (D1), co)
with vacuum V)8 is

L =-1[9 + V1 ][9 V1 + V2[19 V2 + V3 + V4][O V3 + V4]
4 (A.30)
4 -1[10 + V3 V4][9 + V2 V3 V41[+V V1 ][- V].

Using the field redefinitions

qi = vi, q2 = V2 v, (A.31)
q3 = -V2 + V3 4, q4 = -v3 + v4,

we get

L= 1 -l[+q ][9+q2 [O+q3[a+q4-l [9-q4][O-q3]['-q2][9-ql]. (A.32)

which is proportional to the standard result.
We have chosen these two examples because they introduce the techniques
needed to generate Lax operators associated with even the most complicated
algebraic systems. Furthermore, they show how closely the structure of cyclic
affine weight diagrams is linked with the construction of general Lax operators.


As a result, we propose a set of four steps which allows one to construct Lax

operators associated with cyclic representations of affine Lie algebras.

First, we propose that to every weight vector of an affine cyclic weight

diagram one can associate an operator as follows:

Step 1: (aoala2...) --) [Ox + alvi(x) + a2v2(x) +-..]. (A.33)

The coefficient ao does not appear on the right-hand side as it corresponds to

the deleted vertex. Next, we introduce a step which is designed to facilitate the

construction of Lax operators when branch points exist in the corresponding

weight diagram. Essentially, this step reduces more complicated non-trivial

cases to a sum of manageable trivial cases by reducing the branched weight

diagram to a sum a linear subdiagrams.

Step 2: Replace branching weight diagrams by the sum

of linear subdiagrams, each representing a vertical route (A.34)

beginning with, and ending on, the vacuum weightss.

For example, Fig. 14 presents the four linear subgraphs associated with the

canonical representation of D(1)

In drawing cyclic weight diagrams, it is important that the arrows gener-

ated by the deleted vertex are distinguished from the others. Our convention

is to use dashed lines. Furthermore, the direction of the arrows must also

be noted. The Lax operators associated with each subdiagram are then con-

structed as follows:
Step 3: Circulate around the loop beginning with the vacuum solution, such

that the flow is opposite most of the arrows. If a weight vector is

approached by an arrow's

a) tail, append its weight factor to the operator's left side,

b) head, append the weight factor's inverse to the operator's left side

For weights at the tail end of both connecting arrows, do nothing.

Multiply by the product of the corresponding normalization factors.

The loop is to be circulated in a direction opposite most of the arrows so that

the leading term of the Lax operator L = (n + ... has positive exponent. i.e.,

n > 0.

For trivial cases, this completes the computation of L. However, for non-

trivial cases with branching weight diagrams we can not naively build the final

Lax operator from a sum of its constituent linear subgraphs.Instead, as we

shall prove in the next section, they are added together analogously to how

one computes total resistance of resistors in parallel.
Step 4: The Lax operator is given by the inverse of the sum of
the inverses, of the operators resulting from step three.

For example, in the non-trivial case (D4 co). with 0 = 5 s as before, we build

four operators corresponding to the linear subdiagrams in Fig. 14.,
L1 =[9- v 1[(9 + V1] [- V1 + V][(9 t"'. + V3 + V4][9 V3 + V41
X [O + v2 v3 v4][O + v1 v2][O v1],
L2 =[9- v1]-l[a + v1][O v1 + v2[][ v2 + v3 + v4][a + v3 v4]
x [Q + vi v3 v\][o + v1 V,][Q V1],
L3 =[a v, + V_][o v2 + V3 + v4][o V3 + v4][9 + v2 V3 v31
x [o + v, v]}[9 v,],

r ------ (-1 1 0 0 0)

(-1 -1 1 0 0) <-------

(0 0 -1 1 1)

(0 0 0 -1 1) (0 0 0 1 -1)

(0 0 1 -1 -1 )

--------(1 1-1 0 0)

(1 -1 0 0 0)--------

(-1 1 0 0 0)
(-1 -1 1 0 0) <--------,

(0 0 -1 1 1)
= (0 0 0 -1 1)
(0 0 1 -1 -1 )

(1 1 -1 0 0)

(1 -1 0 0 0) <- (1 1 -1 0 0)

(-1 1 0 0 0)
(-1 -1 0 0) (-1 -1 1 0 0) <- (-1 -1 1 0 0)(-I
(-1-1 1 0 0)<-- -4, ...

(0 0-1 1 1) (0 0 -1 1 1) (0 0 -1 1 1)
+ (0 0 1 -1) + (0 0 0-1 1) + (0 0 0 1 -1)

(0 0 1 -1 -1) (0 0 1 -1 -1) (0 0 1 -1 -1)

(1 1 -1 0 ) (1 1 -1 0 0) (1 1 -1 0 0)

(1 -1 0 0 0) -(1 1 -1 0 0) (1-1 0 0 0)- (1-1 0 0 0)-

Figure 14. Subdiagram of the cyclic weight diagram of D 1). The dashed
arrows are the deleted root.

L4 =[0 v1 + v2][( v2 + V3 + v4][O + v3 v4][O + v2 va v3]

x [9 +v1 v2][a- Vi].
Factoring out common terms, we find
L-1 = (Li)-1
= [- v11-1[0 + V1 v21-] [a+ v- V3 V4]-1

X {[O9 v3 + v4]-1 + [ + v3 v41-1 }[ V2 + V3 + V]-1

x [O-v,1+v2]- [+-v1l]-l{1 +[-v -1['+vV,]}.



By taking the reciprocal, and simplifying, we reproduce the previous result


To end this section we consider the alternative vacuum choice 13 04. It

should be obvious that since we are dealing with cyclic weight diagrams, Lax

operators associated with other vacuum states can be achieved by cyclically

permuting factors in the primary Lax operator. Thus, this second vacuum

choice immediately gives the Lax operator

L = 1[o-q3][O-q2][O-q]9-l[O+q1][O+q][9+q93][Oq4]-1[O-q4]. (A.42)

Proof of Diagrammatic Scheme.

To prove the equivalence between the matrix system C0 = 0 and the

diagrammatic algorithm, we begin by rewriting the former as

7D(x) = -A (x), (A.43)

where, to simplify notation, we have defined

D A = I + q(x). (A.44)

The structure of the associated cyclic weight diagram is encoded entirely in

the matrix A. Specifically, recall that the general matrix element Aij is pro-

portional to A if the difference between the ith weight and the connecting jth

weight equals the eliminated root of the embedding affine simple root system.

All other connecting weights Ai are proportional to 1. Otherwise, the matrix

element is assigned the value 0. In all cases the proportionality constant is the

normalization factor of the connecting root.

We construct the proof in stages. For the first stage we consider the trivial

case, i.e., a single entry in each row and column of A. Removing a root produces


only a single vacuum state. This stage corresponds to weight diagrams with no

branch points and only one arrow associated with the eliminated root. Clearly,

we can rearrange the matrix equations in L = 0 such that the vacuum state

equation appears last. Furthermore, it can be arranged such that A is lower

triangular with ones located along a diagonal once removed from the main

diagonal, except for the eliminated root whose coefficient A appears in the

upper right-hand corner, i.e., A is a circulant matrix. Thus, the Lax eigenvalue

equation becomes

LV4N(x) = pON(X). (A.45)

Since D9 is diagonal, the kth equation in (A.43) can be written
Dkk Akii, k = 2,3,... ,N. (A.46)
Clearly, since A is a circulant matrix as specified above, the inequality i < k

holds for k 0 1. Next, by repeatedly replacing the function 0j, appearing

on the right-hand side, with the it- matrix equation we eventually reach the

expression Ak1 1, i.e.,
NN = N ANb,.
i=l 1
= D, AN i A j (A.47)
i=1 j=1 (A.47)

= (_)7 ... AT1iAZ ... D k Akll
i=1 k=l1
where 7 is the height of the cyclic weight diagram. Due to the successive

applications of the state lowering operators A1y with i > j, this equation is

interpreted as taking the highest state 71P and lowering it to the vacuum state


Replacing 01 through

S0, = -ANo 4,,, (A.48)

where we have used A1N = .A0 which excites the level of the state since 1 < N,

DVNN = -(-)IA'A ".. ANMD A, A DA AklD-I 'N. (A.49)
i=1 k=1
where No is the normalization factor of the affine root. Finally, moving terms

to the left-hand side we recover (A.45) where

L = {N0o ... D 1ANiD 1 ... Ak,) }-1, (A.50)
i=1 k=1
and the spectral parameter is given by

P = -(-)A. (A.51)

Since each row and column of A contain only one entry the sum will generate

a single term, i.e.,

L-1 = {NDV 1D_ D-.. }-1, (A.52)

where N is the product of the normalization factors. Now, each D is a weight

factor as defined in step 1. Thus, there is a direct mapping between the order

of the weight factors and their location in the corresponding weight diagram.

Now, suppose we permit multiple row entries in A, i.e., branch points in the

weight diagram. First consider the case where such multiple entries occur above

the Nth row. As before, there is a single vacuum state, and the constant A is

located in the upper right-hand corner of A. Therefore, the constraint i < k, for

k : 1, remains in effect for Eq.(A.46). Hence, the derivation leading to (A.50)

follows through unchanged. Now, each new entry in A causes an additional

final term in (A.50). Clearly, per step 4 of the diagrammatic algorithm, the


final Lax operator is obtained by taking the reciprocal of the sum of terms

generated by (A.50).

Next, suppose the multiple row entries in A, due to the branch point, occur

in the NLh row. The vacuum condition (A.10) shows that this is equivalent

to a degenerate vacuum state with, say, degeneracy d. Subsequently, this row

will be associated with the eliminated root, and the d integers will be assigned

the value A. In fact, A appears only in this row. Clearly, in the weight diagram

the d weights share the same level.

Let us first discuss the case where the coefficients A occur in the first row.

Thus, Eq.(A.46) remains valid, keeping intact the constraint i < k, for k 4 1.

Furthermore, the scalar eigenfunction O(x) is now a linear combination of the

components ON, wN-1,... v-. ,, and the equation for 1 becomes
Di1 = E An, ,. (A.53)

Consequently, Eq.(A.49) is modified to

-E E -'A.,-'
i=1 k=lj=N-d+l

where f is the number of field replacements performed. The characteristic

equation is obtained by multiplying both sides by Ai, and then summing over

s, i.e.,

SA,. = -(-) ... A,D,-'Ai ...
s=N-d+l s=1 k=l j=N-d+l (A.55)
x )1Akl D1 AljPj.

Note, the sum over s on the right-hand side has been extended to the entire

range for convenience.

Each term in A1, contains the factor A, which can then be factored out. As

a result, the scalar eigenfunction is found to be
A= A A, L,, (A.56)
and the Lax operator,
L = A{5 ... Y A,, TA, 1 ...AklD1}-1. (A.57)
s=1 k=1
Clearly, this has the same interpretation as the non-degenerate branching case.

For the last stage of the proof, we relax the condition that multiple occur-

rences of A must all be in the first row of A. In the weight diagram this means

not all the arrows associated with the eliminated root point to the bottom

level. Recall from the discussion surrounding Eq.(A.9), the eliminated root

with coefficient A acts as a state raising operator. Thus, every occurrence of

A will appear in the upper triangular portion of A, and the unit coefficients of

the state lowering roots are in the lower triangular portion.

For A in the kth row of A, k : 1, Eq.(A.46) is modified to
Dj 4' -= 1 A-o,'-,, (A.58)
where, since A corresponds to the state raising operator, k > j. Consider the

case where j is the largest such index to satisfy this equation. Then, allowing

degenerate vacuum states, we have
: .. ED-(--)- "1 A,-i)l. Aj1 LA '

i=1 j=1 (A.59)
= (-) ... )7 DIAiEI -.AjEI[ ,.\(.'. + E A-j,(.],
i=l j=l n k
where, again, j is the number of field replacements performed. The effect of

the factor in front of C" is to first, due to A, raise this state to ij and then

to lower it till the vacuum state 0i, is reached. Now, since the corresponding

weight diagram is cyclic, there must exist some factor that will circulate ',

back to gO.

First, as was the case with (D(1), co), consider the situation where bk is an

intermediate state in (A.59), i.e.,

= (_-) -... D E AsDi-1 Ajkk. (A.60)
i=1 k=l1

This gives
-. = (-)a{-j ... )1A -I 1 }-,,. (A.61)
i=1 k=1
Thus, per step 3 of the diagrammatic algorithm, the factor Dk, associated with

the weight vector at the tail end of both connecting arrows does not appear.

Further, proceeding from higher weights to lower weights in the weight diagram

contributes factors of )-1 in the operator defined in step 2 for the linear


Finally, if ik does not appear as an intermediate state of the vacuum state

, in (A.59), then it must occur as an intermediate state for one of the other

vacuum states. Again, since the weight diagram is cyclic, there is some closed

path going from ', to each of these other vacua. However, to write down a final

expression is too unweilding. Nevertheless, it should be clear that the general

rules of the diagrammatic algorithm are complete and provide an accurate

mapping between weight diagrams and the scalar Lax equation.

Supersymmetric Lax Operators

In this section we generalize the diagrammatic scheme to supersymmetric

affine Lie algebras. The classification of all possible supersymmetric exten-

sions of the basic Lie algebras has been given by Kac.48 In addition to the

bosonic simple roots of the basic Lie algebra, the simple root system of the

supersymmetric algebras contains two distinct kinds of fermionic roots. The

Dynkin symbol of the first type is sometimes given by a shaded vertex repre-

senting a non-zero norm. The second fermionic root type has zero norm whose

Dynkin symbol is given correspondingly by a crossed out vertex. As always,

the bosonic root is denoted by a white vertex.

A new feature occurring in the supersymmetric Lie algebras is that they

may have several non-equivalent simple root systems, corresponding to differ-

ent Dynkin diagrams and Cartan matrices. In other words, the different root

systems can not be transformed into each other through standard Weyl rota-

tions. Instead, they are obtained by performing the "Weyl" transformation

with respect to the nilpotent fermionic root. For more details, see Frappat et

al. 49 which also presents a large collection of Dynkin diagrams associated with

all of the classical contragradient supersymmetric cases, those of the affine and

twisted affine supersymmetric algebras.

Non-equivalent simple root systems which represent the same supersym-

metric Lie algebra differ in the distribution of bosonic and fermionic roots.

However, here we are interested in considering a natural extension of the

Drinfel'd-Sokolov procedure to the supersymmetric case. This restricts the

possible choices for the simple root system used for building the supers-ym-

metric Lax operators.50 Recall, in the bosonic case the mKdV Lax operator

constructed with the gradation choice of Drinfel'd and Sokolov generates Toda

lattice models.34,51,52 For supersymmetric algebras it has been shown53 that

Toda lattices are possible only for simple root systems composed purely of

fermionic roots. Supersymmetric Lie algebras with purely fermionic root sys-


teams have been given by Leites et al. 54

SL(n + 1 I n), OSp(m I 2n) (m = 2n, 2n + 2, 2n 1), D(2 11; a). (A.62)

Furthermore, the infinite-dimensional affine supersymmetric Lie algebras with

purely fermionic simple root systems are

SL(n I n)(1), OSp(2n + 2 | 2n)(1), D(2 | 1; a)(1), (A.63)

while the infinite-dimensional twisted affine cases are

SQ(2n + 1)(2), SL(n I n)(2), OSp(2n | 2n)(2). (A.64)

The supersymmetric extension of the KdV equations was first discussed in

Manin and Radul. 55 They suggested replacing the bosonic derivative 0x by

its supersymmetric analog, i.e.,

x -- D = + x (A.65)

Note that D2 = '. The system of matrix equations of Drinfel'd and Sokolov

can then be generalized to50

Cb(x, ) =[D + Q(x, 0) + A]h(x, 0) = 0, (A.66)

where A is generated by the purely negative fermionic roots, and Q(x, 0) is a

Grassmann odd fermionic superfield which can be expanded as
Q(x,9) = H iJi(x, 0), (A.67)
where now Hi are elements of the Cartan-Kac subalgebra. The vacuum con-

dition is as before,

A-O(X, 0) = 0. (.8


Since the second type of fermionic root is nilpotent, they deserve special

treatment when constructing cyclic weight diagrams. To illustrate how this

comes about, consider the fundamental representation of the supersymmetric

algebra OSp(2 I 2). The Dynkin diagram of the purely fermionic root system

is given by
1 1

where both fermionic roots are denoted as having zero norm, and where we

have indicated the choice (11) for a highest weight vector. The Cartan matrix

is then easily found to be

A=(02 2). (A.69)

To construct the highest weight diagram we proceed as before. Since a positive

one appears both in the first and second places we have two permutations of

subtraction to perform. In particular, we can start by subtracting the first

row of the Cartan matrix giving (13), and then subtracting the second row

resulting in (33). However, unlike the bosonic case, we may not subtract the

first row of the Cartan another time from the weight (13). This is because

here the fermionic weight vectors are nilpotent and subtracting any Cartan

row twice gives a decoupled state. Similarly, we can start by subtracting the

second Cartan row once(and only once) and then the first row giving (33).

Thus, we find the weight diagram with height two:

(13)2 (31)3 (A.70)

The decoupling which occurs when constructing a cyclic weight diagram for

an affine supersymmetric algebra is almost as straightforward. For example, in

Fig. 15 we display the partially decoupled weight diagram of (SL(2 I 2)(1), co)

(1 -1 0 2) 2 ---
b bb

(0 -1 1 2) 4 ---

(1 0 0 0) <-I

(0 1 1 0)3


(0 1 1 0)5

(1 1 0 0)6 -

Figure 15. The (1001) representation of SL(2 I2)(1). (a) Dynkin diagram;
(b) Cartan matrix; (c) Cyclic weight diagram. The dashed arrows are the
deleted root.

where states were decoupled as we went from top to bottom. There are several

ways to decouple the remaining weights since the lowering operators b0, b2 and

b3 still appear more than once. The only way for a cyclic weight diagram to

emerge is by decoupling the weights outside the box. To see that this is also

consistent, note that all paths leading from weight 5 to weights 2 or 4 require

two applications of b0.

To construct a super-Lax operator let us take the vacuum solution 4 = -4.

We easily find the super-Lax operator to be


As in the non-supersymmetric case, Lax operators corresponding to the other

three vacua are obtained through cyclic permutations of the above operator.


In this appendix we have shown how one can read off from cyclic weight

diagrams, associated with representations of affine Lie algebras, Lax operators

in the diagonal gauge. This method is most useful when tables of matrix

representations are not at hand and must be generated by weight diagrams

anyway. Furthermore, this procedure can easily be implemented on computer

by virtue of the fact that computer generated algorithms currently exist for

building highest weight diagrams.56 With minor modifications, these programs

can be adapted for cyclic weight diagrams.

It remains to be seen whether higher representations lead to any new

physics. If so then a program of categorizing these results might be pursued

to identify redundant solutions. This might be easier to answer for supersym-

metric algebras since nilpotency projects out decoupled weight vectors. What

is clear though, at least for the non-supersymmetric cases is that these higher

representations lead to integrable systems. Recall, to prove the integrability

of KdV systems Drinfel'd and Sokolov found the necessary infinite set of con-

served currents to be given by the coefficients of the Laurent expansion of C

in the affine parameter A. Our conclusion follows from the fact that every

representation of a basic Lie algebra has an affine extension, and that defining

properties of affine Lie algebras are representation independent.

Finally, it would be interesting to see if our procedure could be modified

to directly generate Lax operators in other gauges. Furthermore, in light of

recent work47 on generalizations of the Drinfel'd and Sokolov scheme, one may


also consider different gradations of the affine Lie algebra from which to obtain

the matrix A and the form of q(x).


1. D. J. Gross and P. F. Mende, Phys. Lett. 197B (1987) 129; Nucl. Phys. B303
(1988) 407

2. F. Cerulus and A. Martin, Phys. Lett. 8 (1964) 80

3. G. Veneziano, Nuovo Cim. 57A (1968) 190

4. P. F. Mende and H. Ooguri, Nucl. Phys. B339 (1990) 641

5. A. Martin, Nuovo Cimento 37 (1965) 671

6. D. A. Eliezer and R. P. Woodard, Nucl. Phys. B325 (1989) 389

7. M. Douglas and S. Shenker, Nucl. Phys. B335 (1990) 635

8. E. Brezin and V.A. Kazakov, Phys. Lett. B236 (1990) 144

9. D. Gross and A.A. Migdal, Nucl. Phys. B340 (1990) 333

10. P. Hoyer, N. A. T6rnqvist and B. R. Webber, Phys. Lett. 61B (1976) 191

11. Chan Hong-Mo, P. Hoyer and P. V. Ruuskanen, Nucl. Phys. B38 (1974) 125

12. V. Alessandrini, D. Amati, and B. Morel, Nuovo Cim. 7A (1972) 797

13. P. Hoyer, Phys. Lett. 63B (1976) 50

14. P. Hoyer, N. A. T6rnqvist and B. R. Webber, Nucl. Phys. B115 (1976) 429

15. W. J. Zakrzewski, Nucl. Phys. B130 (1977) 164

16. M. Sarbishaei and W. J. Zakrzewski, Nucl. Phys. B132 (1977) 268; Nucl. Phys.
B132 (1977) 294

17. C. Barratt, Nucl. Phys. B120 (1977) 147

18. M. Sarbishaei, W. J. Zakrzewski and C. Barratt, Nucl. Phys. B132 (1978) 478

19. M. Quir6s, Ni-l. Phys. B155 (1979) 509



20. C. Barratt, Nucl. Phys. B126 (1977) 133

21. P. Hoyer and J. Kwiecinski, Nucl. Phys. 145 (1970) 409

22. M. Quir6s, Nucl. Phys. B160 (1979) 269

23. P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda and S. Scuito, Nucl.
Phys. B322 (1989) 317

24. P. Di Vecchia, M. Frau, A. Lerda and S. Scuito, Nucl. Phys. B298 (1988) 526

25. J. A. Shapiro and C. B. Thorn, Phys. Rev. D 36 (1987) 432

26. M. B. Green, Phys. Lett. 266B (1991) 325

27. J. Goldstone, unpublished (1976).

28. W. Nahm, Nucl. Phys. B120 (1977) 125.

29. T. L. Curtright and C. B. Thorn, Nucl. Phys. B274 (1986) 520.

30. T. L. Curtright, C. B. Thorn and J. Goldstone, Phys. Lett. 175B (1986) 47.

31. T. L. Curtright, G. I. Ghandour and C. B. Thorn, Phys. Lett. 187B (1986) 45.

32. G. T. Horowitz, J. Morrow-Jones, S. P. Matrin and R. P. Woodard, Phys. Lett.
60B (1088) 261

33. M. Douglas, Phys. Lett. B238 (1990) 176

34. V. Drinfeld and V. Sokolov, J. Sov. Math. 30 (1984) 1975

35. P. Di Francesco and D. Kutasov, Nucl. Phys. B342 (1990) 589

36. P. Di Francesco and D. Kutasov, Princeton preprint PUPT-1206 (1990)

37. D. Bessis, C. Itzykson and J.-B. Zuber, Adv. Appl. Math. 1 (1980) 109

38. M. L. Mehta, Commun. Math. Phys. 79 (1981) 327

39. S. Chadha, G. Mahoux and M. L. Mehta, J. Phys A14 (1981) 579

40. V. A. Fateev and S. L. Luk'yanov, Int. J. Mod. Phys. A3 (1988) 507

41. S. L. Carbon and E. J. Piard, J. Math. Phys. 33 (1992) 2664

42. R. Slansky, Phys. Rep. 79 (1981) 1

43. E. B. Dynkin, Amer. Math. Soc. Trans. Ser. 2,6 (1957) 111 and 246


44. M. R. Bremner, R. V. Moody and J. Patera, Tables of Dominant Weight Mul-
tiplicities for Representations of Simple Lie Algebras (Dekker, New York,
1985) p. 1

45. P. Goddard and D. Olive, Int. J. Mod. Phys. Al (1986) 303

46. A. Das, Integrable models (World Scientific,Singapore,1989) p. 1

47. M. F. De Groot, T. J. Hollowood and J. L. Miramontes, Princeton preprint
IASSNS-HEP-91/19, PUPT-1251 (1991)

48. V. G. Kac, Adv. in Math. 26 (1977) 8

49. L. Frappat, A. Sciarrino and P. Sorba, Comm. Math. Phys. 121 (1989) 457

50. T. Inami and H. Kanno, Comm. Math. Phys. 136 (1991) 519

51. A. Bilal and J.-L. Gervais, Phys. Lett. B206 (1988) 412; Nucl. Phys. B314
(1989) 646

52. P. Forgacs, A. Wipf, J. Balog, L. Feher and L. O'Raifeartaigh, Phys. Lett. B227
(1989) 214; J. Balog, L. Feher, L. O'Raifeartaigh, P. Forgacs and A. Wipf,
Ann. Phys. 203 (1990) 76

53. M. A. Olchanetsky, Comm. Math. Phys. 88 (1983) 63

54. D. A. Leites, M. V. Saveliev and V. V. Serganova, Serpukhov preprint, IHEP
85-81 (1985)

55. Yu. I. Manin and A. 0. Radul, Comm. Math. Phys. 98 (1985) 65

56. W. G. McKay, J. Patera, and D. W. Rand, SimpLie User's Manual and Software
(CRM, Montreal, 1990) p. 1


Steve Carbon was born in Evanston, Illinois, on May 21, 1958. It was

while was attending high school near Chicago that he was fascinated by his

first physics course. Later, he graduated from the Miami Public School system

and enrolled at the University of Central Florida in Orlando. He obtained B.S.

degrees in both math and physics while at UCF before obtaining a job with

RCA at the Kennedy Space Center as a software analyst. While with RCA,

he earned an M.S. degree in math at UCF and started on an M.S. degree in

physics. He then left his job with RCA to complete his physics masters. Af-

terwards, he came to the University of Florida and subsequently began doing

research under the supervision of Professor Charles Thorn. His research inter-

ests have included Skyrmions, matrix models, covariant quantization of string

string, and the high energy behavior of string theory.

I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope
and quality, as a dissertation for the degree of Doctor of Philosophy.

Charles B. Thorn, Chairman
Professor of Physics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate. in scope
and quality, as a dissertation for the degree of Doctor of Philosophy.

re Ramond
Professor of Physics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope
and quality, as a dissertation for the degree of Doctor of Philosophy.

Pierre Sikivie
Professor of Physics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope
and quality, as a dissertation for the degree of Doctor of Philosophy.

Richard D. Field
Professor of Physics

I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope
and quality, as a dissertation for the degree of Doctor of Philosophy.

Christopher Stark
Associate Professor of Mathematics

This dissertation was submitted to the Graduate Faculty of the Department
of Physics in the College of Liberal Arts and Sciences and to the Graduate
School and was accepted as partial fulfillment of the requirements for the degree
of Doctor of Philosophy.

May 1993
Dean, Graduate school

3 I1262 08553 9244
3 1262 08553 9244