Ground Ring of Two Dimensional String Theory
Ground Ring of Two Dimensional String Theory
Ground Ring of Two Dimensional String Theory
August, 1991
⋆
Edward Witten
ABSTRACT
String theories with two dimensional space-time target spaces are characterized
by the existence of a “ground ring” of operators of spin (0, 0). By understanding
this ring, one can understand the symmetries of the theory and illuminate the
relation of the critical string theory to matrix models. The symmetry groups
that arise are, roughly, the area preserving diffeomorphisms of a two dimensional
phase space that preserve the fermi surface (of the matrix model) and the volume
preserving diffeomorphisms of a three dimensional cone. The three dimensions in
question are the matrix eigenvalue, its canonical momentum, and the time of the
matrix model.
Critical string theory with two dimensional target space time has been the
subject of much recent interest. The simplest target space is a flat space-time with a
linear dilaton field, and perhaps an additional “Liouville” term. The corresponding
world-sheet Lagrangian is
1
Z √ 1
Z √
2
d x h h ∂i X∂j X + h ∂i φ∂j φ − √ ′ d2 x h · φR(2)
ij ij
L=
4πα′ 2π α
(1.1)
Z √ √
+ µ d2 x h exp(−2φ/ α′ ).
(h is the world-sheet metric, R(2) is the Ricci scalar, and µ is the cosmological
constant.) The “matter” field X is decoupled from the “Liouville” field φ (and
from the ghosts that appear upon quantization), so some vertex operators can be
constructed as products of X operators and φ operators. The obvious conformal
fields constructed from X are standard vertex operators eipX . These depend on
the continuously variable parameter p. In addition, at particular values of the
momentum, additional “discrete” primary fields appear. The first of these, at zero
momentum, is the current ∂X (or, when left and right movers are combined, the
¯
gravitional vertex operator ∂X∂X). These discrete states have long been known
[1]. In the context of two dimensional string theory they were discussed at an
early stage by N. Seiberg (unpublished) and have showed up in matrix model
calculations [2]. It has been argued that they should be interpreted in terms of a
“stringy” target space topological field theory [3]; this indicates that understanding
these modes is essential for extracting the real physical lessons of two dimensional
string theory. These modes may be important in understanding two dimensional
black holes [4,5]. They have been the subject of several recent studies [6,7,8].
In addition to the discrete vertex operators that have usually been considered,
which occur (if one considers (1, 1) operators) at ghost number zero, there are
also discrete vertex operators which occur at the same values of the momenta at
adjacent values of the ghost number. This is obvious in the BRST formalism. If
2
one has a family of states |α(p)i, of unit norm, parametrized by a momentum
variable p, and BRST invariant only at, say, p = p0 , then one has
where |β(p)i is of unit norm and f (p) vanishes precisely at p = p0 . Then |α(p0 )i
and |β(p0)i are a pair of discrete states at adjacent values of the ghost number.
The whole pattern of discrete states of various ghost number has been thor-
oughly described by Lian and Zuckerman [9]. However, the physical consequences
of the existence of discrete states of non-standard ghost number have not yet been
considered. This is the problem that we will consider in the present paper.
To begin with, we consider the holomorphic part of the current algebra. Thus,
X is a free field with
The stress tensor is Tzz = −(∂z X)2 /2. The tachyon operators Vp = eipX are
conformal fields of dimension p2 /2.
For reasons that are standard and will become clear, it is convenient to first
consider the theory compactified at the SU(2) radius and with zero cosmological
√
constant. At the SU(2) radius, the allowed values of the momenta are p = n/ 2,
with n ∈ ZZ. There is an SU(2) current algebra with (T+ , T3 , T− ) represented by
√ √ √
(eiX 2 , i 2∂X, e−iX 2 ). For s ≥ 0,there is no singularity in the short distance
product
√ √
eiX 2
(z) · eisX/ 2
(w) (2.2)
√
for z → w, so the operator eisX/ 2 transforms as the highest weight of an SU(2)
representation. In view of the J3 value, this is a representation of spin s. By
3
repeatedly acting with the lowering operators J− , we fill out the SU(2) multiplets
√
with operators Vs,n , n ∈ {s, s − 1, . . . , −s}, such that Vs,s = eisX/ 2, Vs,−s =
√
⋆
e−isX/ 2 . The Vs,n are all Virasoro primaries, since the Vs,s are standard primary
fields and the SU(2) current algebra intertwines with Virasoro. For |n| = s, the Vs,n
are standard tachyon operators at particular momenta, but for |n| < s, they are
new objects, “discrete primaries.” The conformal dimension of Vs,n is s2 /4. More
detailed investigation [1,9] shows that regardless of the radius of compactification
of the X field, the only primary fields of the X system are the tachyon vertex
operators and the Vs,n . At an arbitrary radius, the discrete primaries are whatever
combinations of Vs,n ’s with anti-holomorphic counterparts are permitted (by the
standard momentum restrictions) at that particular radius.
Now, let us couple the Vs,n ’s to the rest of the system. First, we consider the
√
Liouville field φ. Its stress tensor is Tzz = −(1/2)(∂φ)2 + 2φ. The operator eiwφ
√
has conformal dimension w 2 /2 + iw 2. Looking for a spin one primary field of the
form Vs,n eiwφ , we see that there are two possibilities:
√ √
± 2φ∓sφ 2
Ws,n = Vs,n · e . (2.3)
√
± has momentum (n, i(−1 ± s)) · 2. The importance of the spin one
Note that Ws,n
condition is that (1, 1) vertex operators corresponding to infinitesimal deformations
of the theory can be constructed in the form
± ±
Ws,n (z) · W s,n′ (z) (2.4)
where W s,n are the antiholomorphic operators analogous to the Ws,n . Notice
that as the Liouville field is noncompact, the Liouville momenta must be matched
⋆ The latter holds since a rotation about the J1 axis gives X → −X and reverses the sign of
J3 .
4
between left and right movers, which is the reason that the left-movers and right-
movers in (2.4) share the same value of s and of the sign ±. At the SU(2) radius,
+ −
there is no similar matching of n. Note that W0,0 = W0,0 is the cosmological
√
constant operator eφ 2.
Now we wish to include the ghosts, which are fermi fields b and c of spins 2
and −1, with Tzz = −2bzz ∂z cz − ∂z bzz · cz . One of the important merits of the
ghosts is that in their presence, the physical state condition can be incorporated
by considering the cohomology of the BRST operator Q. In the standard fashion,
± are related to spin 0, BRST invariant fields
the spin one fields Ws,n
± ±
Ys,n = cWs,n (2.5)
of ghost number 1.
This cannot be the whole story, for a reason indicated in the introduction. The
± of |n| = s are not truly discrete (but are special tachyon states). However, the
Ys,n
± of |n| < s must have “partners” at an
ghost number one discrete operators Ys,n
+ have partners
adjoining value of the ghost number, namely 0 or 2. In fact, the Ys,n
− have partners at ghost number 2 [9]. The ghost
at ghost number 0, and the Ys,n
+ will be called in this paper O
number 0 partner of Ys,n u,n , with u = s − 1. The
√
momentum of Ou,n is thus (n, iu) · 2. Since |n| < s, we have |n| ≤ u, so the
allowed values of u and n are u = 0, 1/2, 1, . . . and n = u, u − 1, . . . , −u.
+ with |n| < s is Y + = c∂X, a discrete
For instance, the first case of a Ys,n 1,0
primary field of ghost number 1 and momentum (0, 0). The corresponding discrete
primary field of ghost number 0 and momentum (0, 0) is the identity operator,
+ +
O0,0 = 1. At the next level, there are two discrete primaries, Y3/2,1/2 and Y3/2,−1/2 ,
of spin 0 and ghost number 1. The corresponding spin 0, ghost number 0, operators
5
are
√
i
x = O1/2,1/2 = cb + √ (∂X − i∂φ) · ei(X+iφ)/ 2
2
√ (2.6)
i
y = O1/2,−1/2 = cb − √ (∂X + i∂φ) · e−i(X−iφ)/ 2 .
2
Some details about their construction can be found in appendix 1.
There are two immediate reasons that the existence of spin zero, ghost number
zero operators is of vital importance:
(1) It leads to the existence of symmetries of the conformal field theory under
study.
(2) It leads to the existence of a ring, which we will call the “ground ring,”
which largely controls the properties of the theory, as we will see.
To understand the first point, note that we already know of the existence of
±
infinitely many spin (1, 0) and (0, 1) operators of zero ghost number, namely Ws,n
±
and W s,n . Because of the Liouville momentum that they carry, which must be
matched between left and right movers, to construct true quantum field operators
as opposed to the holomorphic or antiholomorphic chiral vertex operators of the
above discussion, W (or W ) should not be considered by itself but must be paired
with antiholomorphic (or holomorphic) fields of the same Liouville momentum.
One way to do this is to pair W · W , as in (2.4), to make fields of spin (1, 1),
corresponding to infinitesimal moduli. To make quantum field operators of spin
(1, 0), however, we must combine a W with an antiholomorphic spin zero field of
the same Liouville momentum, and conversely for W . The spin (1, 0) and spin
(0, 1) currents of the theory are consequently
+
Js,n,n′ = Ws,n Os−1,n′
+ (2.7)
J s,n,n′ = Os−1,n W s,n′
6
we will presently determined to be the Lie algebra of volume preserving diffeomor-
phisms of a certain three manifold.
By combining left and right movers in the usual way we can form spin (0, 0)
quantum field operators
which can be multiplied in just the same way. We will call the ring of the V’s
the ground ring of the theory, and we will call the ring of the chiral O’s the chiral
ground ring.
Since the ground ring is naturally associated with the conformal field theory,
any symmetries of the theory must be symmetries of the ground ring. This will
make it easy to determine the symmetry algebra. In fact, if J is a spin one current
generating a symmetry, then the action of J on a spin (0, 0) operator O is defined
by
1
I
J(O(P )) = J(z) · O(P ) (2.10)
2πi
C
7
(modulo BRST commutators) where C is any contour surrounding the point P .
Standard contour manipulations show that if J and J ′ are two currents and [J, J ′ ]
is their commutator then
so that the O(P )’s form a representation of the symmetry algebra. What is more
unusual is that since a contour enclosing two points P and Q is homologous to the
sum of a contour surrounding P and a contour surrounding Q, we get
Thus, the J’s act as derivations of the ring of O’s. These considerations can be
± act as derivations of the
applied in two different ways: the chiral currents Ws,n
chiral ground ring, and the quantum currents J and J of (2.7) are derivations of
the full quantum ground ring.
The ground ring is easily determined. First, we consider the chiral ground ring.
We note that the product xn y m has the quantum numbers of O(n+m)/2,(n−m)/2 ,
so Ou,n is a multiple of xu+n y u−n if the latter is not zero. We will prove that
xn y m 6= 0 for all n, m in the next subsection. Accepting this for the moment, the
chiral ground ring of the Ou,n ’s is just the ring of polynomial functions in x and y.
8
Now let us combine left and right movers to make the full quantum ground
′ ′
ring. It is generated by operators xn y m x′n y ′m with n + m = n′ + m′ (to balance
the Liouville momenta). Such operators are multiplied in the obvious way, by
multiplying the left and right moving parts separately. At n + m = 0 there is only
the identity operator. At n + m = 1 there are four operators,
a1 = xx′
a2 = yy ′
(2.14)
a3 = xy ′
a4 = yx′ .
′ ′
The ground ring is generated by the ai since any monomial xn y m x′n y ′m with
n + m = n′ + m′ can obviously be written as a monomial in the ai . The ai obey
the one obvious relation
a1 a2 − a3 a4 = 0 (2.15)
and are otherwise independent. (In algebraic geometry, one says that the left hand
side of (2.15) is a homogeneous polynomial whose zeros form a quadric in P3 , and
the equations (2.14) are an isomorphism of this quadric with P1 × P1 .)
Thus, the ground ring of the theory, at the SU(2) point, is the ring of polyno-
mial functions on the three dimensional quadric cone Q defined by a1 a2 −a3 a4 = 0.
In comparing with matrix models, we will learn that the three dimensions of this
cone correspond to the matrix eigenvalue, its canonical momentum, and the “time”
of the matrix model Schodinger equation. At the SU(2) point, there is a complete
symmetry among these variables, which is partly lost upon decompactification.
9
2.4. Determination Of The Chiral Symmetry Algebra
Now we will determine the symmetry algebra of the theory, by using the fact
that it acts as an algebra of derivations of the ground ring. First we consider the
chiral currents.
To determine how a chiral current J acts on the chiral ground ring, it is enough,
from (2.13), to know what it does to x and y. Indeed if
then it follows from (2.13) that for an arbitrary element w(x, y) of the ground ring,
we get
∂ ∂
J(w) = f +g · w. (2.17)
∂x ∂y
In particular, J can be identified with the vector field f ∂x + g∂y on the (x, y)
plane, and so the chiral currents generate an algebra of diffeomorphisms of the
(x, y) plane. (And similarly, the quantum currents which combine left and right
movers act by diffeomorphisms of the quadric cone Q.)
± . First
Let us determine the vector fields corresponding to some particular Ws,n
− are trivially disposed of. They have momenta (n, i(−1 − s)) 2,
√
of all, the Ws,n
and would map x or y to spin (0, 0) operators of ghost number 0 and Liouville
√
momentum i(−1/2 − s) 2. As there is no BRST cohomology with those quantum
− annihilate x and y and therefore the whole ground ring. Sim-
numbers, the Ws,n
+
ilarly, the cosmological constant operator W0,0 (which in any case is the same as
− + . The first non-trivial case is
W0,0 ) annihilates x and y. Now we consider the Ws,n
+
W1/2,±1/2 . These are tachyon operators,
√
+
W1/2,±1/2 = e(±iX+φ)/ 2
. (2.18)
10
+
The action of W1/2,1/2 on y is given by the residue of the pole, which is 1, so
+
W1/2,1/2 (y) = 1. (2.20)
+
On the other hand, W1/2,1/2 annihilates x (there is no short distance singularity in
+ +
the operator product W1/2,1/2 · x). Hence we have determined that W1/2,1/2 acts
on the chiral ground ring as
+ ∂
W1/2,1/2 = . (2.21)
∂y
+
Similarly, W1/2,−1/2 acts as
+ ∂
W1/2,−1/2 = . (2.22)
∂x
At this point, we can fill a gap in our determination of the chiral ground ring
by proving that xn y m 6= 0. Indeed, an n-fold repeated commutator of xn y m with
+ +
W1/2,−1/2 followed by an m-fold repeated commutator with W1/2,1/2 gives 1, which
does not vanish, so xn y m must not be zero.
+ +
From (2.21) and (2.22), we see that W1/2,1/2 and W1/2,−1/2 commute in their
action on the x − y plane. One might be inclined to think that they commute
altogether. In fact
√ √ 1 √ 1
+ +
W1/2,1/2 (z)W1/2,−1/2 (w) = e(iX+φ)/ 2
(z)e(−iX+φ)/ 2
(w) ∼ eφ 2 (w) = W + (w).
z−w z − w 0,0
(2.23)
+
The object W0,0 (w) is “central” in the sense that it commutes with all of the
currents, as we will see, and acts trivially on the x−y plane. (2.23) is the beginning
of the discovery that what is going on in the problem is quantum mechanics on the
x − y plane, as the commutator of (the operators representing) ∂x and ∂y is not
zero but is a central object.
+
Now, let us determine the rest of the algebra. The operators W1,n are simply
the SU(2) currents. x and y have T3 = ±1/2, respectively (as we see from their
11
momentum values), and as they are the only cohomology classes with their values
of the ghost number and Liouville momentum, they form an SU(2) representation
which must have spin 1/2. Thus, without further ado we have
+ ∂
W1,1 =x
∂y
+ 1 ∂ ∂
W1,0 = x −y (2.24)
2 ∂x ∂y
+ ∂
W1,−1 =y .
∂x
(These formulas have been written as if x and y might be real variables, with
symmetry group SL(2, IR) rather than the SU(2) that is natural in the Euclidean
theory. Upon complexification, (2.24) is equivalent to the standard SU(2) genera-
tors.)
+ with s > 1. These operators transform under SU(2)
Now we consider the Ws,n
with spin s. By considering the Liouville momenta, we see that they map x or y to
polynomials in x and y that are homogeneous of degree 2s − 1. Such a polynomial
has spin s − 1/2. On the other hand, we know that x and y have spin 1/2. In
combining spin s with spin 1/2 to make spin s − 1/2, there is only one invariant
coupling. If these couplings are not zero (which we will prove presently), they can
+ by factors depending only
be set to any desired non-zero value by scaling the Ws,n
on s. Therefore, if we can find any Lie algebra in which these couplings are all
non-zero, and in which the Ws,n of s ≤ 1 act correctly, it is the one we want.
ω = dx dy (2.25)
12
that generates an infinitesimal area-preserving transformation of the x − y plane.
+ with the transformation determined by the vector field that
If we identify Ws,n
corresponds to h(x, y) = xs+n y s−n , then all of the conditions are obeyed since (i)
these W ’s are non-zero; (ii) this assignment is SU(2) (or SL(2, IR)) covariant; (iii)
these W ’s obviously generate a Lie algebra. The area preserving diffeomorphisms
of the x − y plane are therefore the Lie algebra that we are looking for.
√ √ 1 √ 1
e(−iX+φ)/ 2
(z)·e(isX+(1−s)φ) 2
(w) ∼ e(i(s−1/2)X+(3/2−s)φ) 2 (w) = W (w),
z−w z − w s−1/2,s−1/2
(2.27)
one has
+ + +
[W1/2,−1/2 , Ws,s ] = Ws−1/2,s−1/2 . (2.28)
It is interesting to note that the “energy” operator of the theory, that is, the
+
operator that generates translations of X, is ∂X which is none other than W1,0
in our present notation. According to the above, this operator generates on the
x − y plane the motion derived from the Hamiltonian function h(x, y) = xy. With
p = x + y, q = x − y, this is none other than the inverted harmonic oscillator
Hamiltonian h = p2 − q 2 familiar in the c = 1 matrix model. We will make a more
extensive comparison with the matrix model in §3, after combining left and right
movers. The conclusion will be much the same.
13
The Other Operators
s. Since spin s + s′ + 1 does not appear in the tensor product of spin s and spin
s′ , we must have
−
[Ws,n , Ws−′ ,n′ ] = 0, for all s, n, s′ , n′ . (2.29)
+
[Ws,n , Ws−′ ,n′ ] = const · Ws−′′ ,n+n′ , (2.30)
− ∂ (s−n)/2 ∂ (s+n)/2 2
Ws,n ∼ (s−n)/2 (s+n)/2 δ (x, y). (2.31)
∂x ∂y
I will just briefly indicate how this can be proved. The proposed formula is obvi-
+
ously SU(2) covariant, so the W1,n act correctly. For s = 1/2, the commutators
+ − ] are all non-zero. This is proved by explicitly evaluating them; the
[W1/2,1/2 , Ws,s
14
+ + ]. The commu-
calculation is just analogous to the one above for [W1/2,−1/2 , Ws,s
+ − −
tators [W1/2,n , Ws,n′] ∼ W
s+1/2,n+n′ are all determined by SU(2) in terms on one
unknown s dependent coefficient, which is nonzero for all s in view of what was
− can be uniquely normalized so as to transform
just said. This being so, the Ws,n
+
under W1/2,n as claimed in (2.31). The derivatives of δ 2 (x, y) appearing in (2.31)
give a representation of the algebra of polynomial area-preserving vector fields, so
− transform as claimed, it is enough to show that there is
to prove that the Ws,n
+ for s ≤ 1 act as claimed. This
only one possible representation in which the Ws,n
+ + +
is an immediate consequence of the formula [W1/2,n , Ws,n′] ∼ W
s−1/2,n+n′
which
+ once
combined with SU(2) invariance uniquely determines the action of the Ws,n
+
the action of the Ws−1/2,n is known.
− act trivially on the x − y plane; this can be demonstrated by consid-
The Ws,n
ering the Liouville momentum and SU(2) transformations, as above.
Now we will combine the left and right movers and determine the quantum
symmetry algebra, that is the algebra of the operators introduced earlier in (2.7):
+
Js,n,n′ = Ws,n Os−1,n′
+ (2.32)
J s,n,n′ = Os−1,n W s,n′
I will again proceed indirectly, using the fact that the J’s and J’s must correspond
to vector fields on the quadric surface Q defined by
a1 a2 − a3 a4 = 0. (2.33)
15
H such that on H
Λ = −df · Θ (2.35)
where Λ = da1 da2 da3 da4 is (up to a constant multiple) the unique translation and
SL(4) invariant volume form on the ambient affine space. The latter description
makes it possible to define Θ without singling out one of the coordinates. For the
quadric Q, f = a1 a2 − a3 a4 , so
If we write
The polynomial vector fields on Q transform under SO(2, 2), whose complexi-
fication is the same as that of SU(2) × SU(2), as
M
((n + 1, n) ⊕ (n, n + 1) ⊕ (n + 1, n + 1)) . (2.39)
n=0,1/2,1,...
M
((n + 1, n) ⊕ (n, n + 1)) . (2.40)
n=0,1/2,1,...
These facts will be demonstrated in appendix 2. I claim that the Lie algebra of the
J’s and J’s is the algebra of volume preserving polynomial vector fields on Q, and
in fact that the J’s and J’s correspond respectively to the pieces of spin (n + 1, n)
and (n, n + 1). To demonstrate this, it is enough to note that Js,n,n′ and J s,n,n′
16
have spin (s, s − 1) and (s − 1, s), for s = 1, 3/2, 2, . . .. Moreover, the vector fields
corresponding to the Js,n,n′ and J s,n,n′ are nonzero because of arguments similar
to those that we gave in the chiral case. The SU(2) × SU(2) content therefore
ensures that the J’s and J’s must exactly fill up the volume preserving polynomial
vector fields on Q.
Now we will briefly discuss how some of the structures change in the com-
pactified theory under a small departure from the SU(2) point. Some of the most
interesting points will be speculative.
The infinitesimal moduli of the theory, for small departure from the SU(2)
point, are
± ± ±
Zs,n,n′ = Ws,n W s,n′ . (2.41)
I will focus on deformations in which only the Z + are excited. (However, the Z −
moduli are also important, for instance in two dimensional black holes.) The Z +
moduli transform under SU(2) × SU(2) as
M
(n, n). (2.42)
n=0,1/2,1,...
As we will see in appendix (2), this is precisely the SU(2) × SU(2) content of
a general polynomial function φ on the hypersurface Q. Thus, the Z + moduli
naturally combine into such a function.
Now let us try to guess the geometric meaning of the function φ. Such a
function gives precisely the data we need to deform the quadric Q to a general
nearby hypersurface described by an equation
a1 a2 − a3 a4 = φ(ai ). (2.43)
I conjecture that the first order Z + corrections to the operator algebra deform
the relation a1 a2 − a3 a4 = 0 into (2.43). In particular, the cosmological constant
17
corresponds to the case φ = µ (a constant). Thus, according to our ansatz, the
ground ring at µ 6= 0 but still at the SU(2) radius is the ring of functions on the
smooth quadric
a1 a2 − a3 a4 = µ. (2.44)
We can easily implement this prescription in the present situation. Since the
structure constants for left and right movers are just the Poisson brackets in x − y
and x′ − y ′ , respectively, the beta function is
′ ′ ∂2φ ∂2φ
β = ǫij ǫi j ′. (2.47)
∂xi ∂x′i ∂xj ∂x′j
′
After a relatively long but straightforward calculation, one finds that this can be
P
written in terms of the ai as follows. Let Z be the scaling operator Z = i ai ·∂/∂ai .
Let ηij be the metric such that a1 a2 − a3 a4 = ηij ai aj /2. Let ∆ = η ij ∂i ∂j be the
corresponding Laplacian. Then
1 ij
η (∂i Zφ)(∂j Zφ) − (Z 2 φ) · ∆φ .
β= (2.48)
2
18
and φ′ , as they coincide on Q. As will be explained in the appendix, there is always
a unique choice of f such that ∆φ′ = 0. It is easy to see that if φ is a homogenous
polynomial (so Zφ = λφ for some λ) such that ∆φ = 0 and φ is a solution of β = 0,
then w(φ) is also a solution of β = 0 for any function w. The simplest nonconstant
function obeying the stated conditions is φ = (a1 a2 + a3 a4 )/2. Consequently (since
in fact φ = a1 a2 on Q)
∞
X
φ=− ǫn (a1 a2 )n (2.49)
n=0
is a perturbation with vanishing second order beta function, for any values of the
constants ǫn .
According to our ansatz for the meaning of φ, the marginal operator (2.49)
corresponds to deforming the quadric Q to a more general hypersurface
∞
X
a3 a4 = a1 a2 + ǫn (a1 a2 )n . (2.50)
n=0
In comparing to matrix models, as we will see, the eigenvalue phase space will
correspond to the a1 − a2 plane, and the curve a3 a4 = 0 in the a1 − a2 plane will
correspond to the fermi surface. The function a1 a2 will correspond to the standard
inverted harmonic oscillator Hamiltonian H = p2 − q 2 , and the function on the
right hand side of (2.50) is the Hamiltonian of a perturbed matrix model.
19
3. The Uncompactified Theory; Comparison To The Matrix Model
Of the operators that we have found in §2, the ones that survive in the uncom-
pactified theory are the ones that have pL − pR = 0, where pL and pR are the left
and right moving momenta. Recalling the definitions
a1 = xx′
a2 = yy ′
(3.1)
a3 = xy ′
a4 = yx′
we see that the allowed operators are precisely the ones that are invariant under
∂ ∂
K = a3 − a4 . (3.3)
∂a3 ∂a4
∂
K= . (3.4)
∂ψ
20
Now, let us determine the symmetry group of the uncompactified model. This
is simply the subgroup of the group of volume-preserving diffeomorphisms of the
quadric cone Q which are also K invariant. We recall from (2.38) that the volume
form in these coordinates is
∂ ∂ ∂
f (a1 , a2 ) + g(a1 , a2 ) + u(a1 , a2 ) . (3.6)
∂a1 ∂a2 ∂ψ
∂ ∂
f +g (3.7)
∂a1 ∂a2
must generate a diffeomorphism of the a1 − a2 plane that preserves the area form
da1 da2 . Thus area-preserving diffeomorphisms of the a1 − a2 plane appear in the
uncompactified theory. The operators
∂
u(a1 , a2 ) (3.8)
∂ψ
21
immediately sees that
∂ ∂ ∂ a3 f g
f +g +u a3 = · + + ua3
∂a1 ∂a2 ∂ψ 2 a1 a2
(3.9)
∂ ∂ ∂ a4 f g
f +g +u a4 = · + + ua4 .
∂a1 ∂a2 ∂ψ 2 a1 a2
The latter requirement means that the symmetries of the theory include not
all of the area preserving polynomial vector fields on the a1 − a2 plane, but only
those that preserve the locus a1 a2 = 0. I claim that the a1 − a2 plane corresponds
to the eigenvalue phase space of the matrix model. a1 and a2 correspond to p + q
and p − q, where q is the matrix eigenvalue and p is its canonical momentum.
The locus a1 a2 = 0 corresponds to the fermi surface of the matrix model, that
is, it is the surface on which the inverted harmonic oscillator Hamiltonian H =
p2 − q 2 vanishes. When we study the matrix model presently, we will see why area
preserving diffeormorphisms that preserve the fermi surface arise naturally.
a3 a4 = h(a1 , a2 ), (3.10)
with h an arbitrary Hamiltonian function. If so, then the “branch locus” would be
the curve h = 0 in the a1 − a2 plane. K invariant volume preserving transforma-
22
tions of the three-manifold (3.10) correspond to area preserving diffeomorphisms
of the a1 − a2 plane that map the locus h = 0 to itself, plus the abelian gauge
transformations u · ∂/∂ψ. This would be expected if h corresponds to the one body
Hamiltonian of the matrix model. In particular, according to the ansatz, when the
cosmological constant is turned on, the hypersurface a1 a2 − a3 a4 = 0 is deformed
to a hypersurface Qµ given by a1 a2 − a3 a4 = µ. A K invariant volume preserving
diffeomorphism of Qµ is an area preserving diffeomorphism of the a1 − a2 plane
that leaves fixed the curve a1 a2 − µ = 0. This curve corresponds to the fermi
surface of the matrix model at chemical potential µ.
Let us now analyze the symmetries of the matrix model version of the c = 1
theory. Our discussion will be fairly similar to that of [7,13, 14], but we will
consider a somewhat larger class of symmetries. The extra symmetries might be
considered “trivial” in the matrix model but are important in comparing to the
conformal field theory.
α = p dq − H dt. (3.13)
that leaves the Lagrangian invariant. To this end, it is not necessary to leave α
23
invariant. A transformation under which α → α + dβ is also a symmetry. Instead
of the one form α, we should consider the two form ω = dα = dp dq − dH dt. A
symmetry is a transformation that leaves ω invariant.
Let us consider the matrix model with the standard inverted harmonic oscilla-
tor Hamiltonian, H(p, q) = (p2 − q 2 )/2. Then ω = dp dq − (p dp − q dq)dt. If we
set
p′ = p cosh t − q sinh t
(3.15)
q ′ = −p sinh t + q cosh t
then we find
ω = dp′ dq ′ . (3.16)
It is therefore easy to identify the vector fields that preserve ω: they are of the
form
∂g(p′ , q ′ ) ∂ ∂g(p′ , q ′ ) ∂
∂
− + u(p′ , q, t) . (3.17)
∂q ′ ∂p′ ∂p′ ∂q ′ ∂t
Note that g is a function of p′ , q ′ only but u may also depend on t. This is very
similar to the answer that we found in the conformal field theory, with a1 , a2 ↔
p′ , q ′ and ψ ↔ t. There are two discrepancies: (i) in the conformal field theory u
was a function of a1 and a2 only, and not ψ; (ii) in the conformal field theory g was
required to obey a certain restriction, such that the surface a1 a2 = 0 is invariant
under the symmetry generated by g.
I cannot explain the first discrepancy, but there is a very nice way to
parametrize it. Let Θ be the volume form in (p, q, t) space
24
As for that last restriction, which is our second apparent discrepancy, it has
a very nice explanation in the matrix model. The (1, 0) and (0, 1) currents that
we found in the conformal field theory are unbroken symmetries of a particular
ground state. They should be compared to the symmetries of the matrix model
that preserve the ground state. The ground state of the matrix model is the state
in which all the single particle levels of H(p, q) < 0 are filled and the others are
empty. An area preserving transformation that maps filled states to filled states
and empty states to empty states necessarily maps the fermi surface to itself. As
H(p, q) = (p2 − q 2 )/2 = (p′2 − q ′2 )/2, the area preserving transformations of the
p′ −q ′ plane that leave fixed the matrix model ground state are those that leave fixed
the fermi surface p′2 − q ′2 = 0. (It must be left fixed as a set, not pointwise.) The
fermi surface is obviously isomorphic to the fixed locus a1 a2 = 0 of the conformal
field theory, with a1 = p′ + q ′ , a2 = p′ − q ′ .
Thus, we find the dictionary for comparing the conformal field theory to the
matrix model: (a1 , a2 , ψ) corresponds to (p′ + q ′ , p′ − q ′ , t).
Let us give an example of the use of this dictionary. The difference between the
left and right moving X momenta, which corresponds to the difference ∂X − ∂X, ¯
annihilates all states in the (uncompactified) conformal field theory. It is mapped
under our correspondence to
∂
, (3.19)
∂t p′ ,q′
which annihilates all matrix model observables since the single particle equation
of motion of the matrix model equation of motion is dp′ /dt = dq ′ /dt = 0. The
time translation operator of the conformal field theory corresponds to the current
¯
∂X + ∂X. This is mapped to the canonical transformation of the p′ − q ′ plane
generated by the vector field derived from the Hamiltonian a1 a2 = p′2 − q ′2 – the
usual inverted harmonic oscillator, in other words.
We have carried out this discussion at zero cosmological constant, but if the
ansatz in §2.6 is correct, then the generalization to nonzero cosmological constant
25
– and arbitrary perturbation of the matrix model Hamiltonian – is immediate.
Conjectured Interpretation
26
APPENDIX (1): BRST Analysis Of Some Low-Lying States
Our goal in this appendix is to describe the lowest non-trivial examples of some
phenomena analyzed theoretically in [9]. First we recall the standard construction
of the BRST operator. It is
X 1X
Q= cn L−n − c−n c−r bn+r (n − r), (3.20)
n
2 n,r
with αn and φn (obeying [αn , αm ] = [φn , φm ] = nδn+m ) being the Fourier modes
of X and of the Liouville field φ. In particular, α0 and φ0 are the X and φ
⋆
momentum operators. The Fock vacuum will be labeled |p, ↑i or |p, ↓i, where
p = (α0 , φ0 ) labels the momentum eigenvalues, |p, ↓i is a state annihilated by b0 ,
and |p, ↑i = c0 |p, ↓i.
The above expressions for the L’s and Q should all be normal ordered. The
normal ordering constants are zero for this two dimensional system (the “tachyon”
is a massless particle in D = 2).
We will define the ghost number operator, which we will call G, such that |p, ↓i
has G = 0 (from some points of view G = −1/2 is more natural).
27
describe the level one situation in full detail. When we get to level two, we will
just focus on G = −1. The other peculiarities of level one are repeated at level
two (and at all higher levels, according to [9]), but the detailed description would
be somewhat long.
At level one and G = −1, there is only one state, b−1 |p, ↓i, with
√
Qb−1 |p, ↓i = α0 α−1 + (φ0 − i 2)φ−1 |p, ↓i. (3.22)
For G = 1, we get
√
Qφ−1 |p, ↑i = φ0 + i 2 c−1 |p, ↑i
Qα−1 |p, ↑i = α0 c−1 |p, ↑i (3.24)
Qc−1 |p, ↓i = 0.
Finally, at G = 2,
28
Something interesting happens at α0 = 0. At that value of α0 , the L0 condition
√
permits two values of φ0 , namely ±i 2. The behavior at those two values are
√
completely different. First we consider the value φ0 = i 2 (which is shifted to
φ0 = 0 when one goes from states to operators).
In this case there is a one dimensional cohomology at G = −1, since Qb−1 |p, ↓
i = 0 for such p. This state corresponds to the identity operator 1, which is the
first of the spin zero, ghost number zero operators that were the main focus of the
body of the paper. (G as we have defined it is increased by one in going from states
to operators, so states of G = −1 will correspond to operators of G = 0, and so
on.) At G = 0, the cohomology is actually two dimensional. For representatives
√
we can take α−1 |p, ↓i and φ−1 |p, ↓i + i 2b−1 |p, ↑i. The former corresponds to
the spin zero, G = 1 operator c∂X and to the spin one, G = 0 operator ∂X.
The latter is the first of the ghost number zero discrete currents constructed from
the bosons only (without the ghosts) and usually considered in discussions of the
√
discrete states. The latter corresponds to the spin zero, G = 1 operator c∂φ+ 2∂c
but does not correspond to any spin one, G = 0 current. The reason for this will
be explained presently. Finally, to complete the enumeration of the cohomology
at this value of the momenta, the G = 1 state α−1 |p, ↑i corresponds to the G = 2
spin zero operator c∂c∂X, but does not correspond to any current.
Let us now briefly explain why certain of the states just found do not correspond
to currents. First of all, given a BRST cohomology class, to find a corresponding
BRST invariant spin zero primary field we should find a representative |ψi of the
cohomology class that is of Virasoro highest weight, obeying Ln |ψi = 0, n ≥
0. In the cases at hand this is possible; the representatives given above satisfy
this condition. In particular, as the highest weight of these states is zero, they
correspond to primaries of spin zero. To get a current, we need a highest weight
state of spin one. The general strategy for doing this is to set
29
If this vanishes, we cannot proceed further; this is why the G = −1 state b−1 |p, ↓i
does not correspond to a current. Suppose |αi =
6 0. Then
so if |αi is a highest weight state, the highest weight is one, and the operator
corresponding to |αi will be a current. Moreover,
Although this is not zero, it will correspond to a total derivative (since acting on
operators L−1 ∼ ∂/∂z); this is good enough. The question that remains is whether
|αi is a highest weight vector, that is, whether Ln |αi is zero, or at least a BRST
commutator, for n > 0. We have
as |ψi is BRST invariant and highest weight. Therefore, if the BRST invariant
states bn−1 |ψi are BRST commutators, then |αi is of highest weight, at least
modulo BRST commutators.
For two of the cohomology classes found above, this last condition fails. The
√
obstructions arise for n = 1. For |ψi = φ−1 |p, ↓i + i 2b−1 |p, ↑i, we get b0 |ψi =
√
−i 2b−1 |p, ↓i, which is not a BRST commutator; indeed it represents the now
familiar G = −1 cohomology class. For |ψi = α−1 |p, ↑i, we get b0 |ψi = α−1 |p, ↓i,
the standard discrete state at G = 0.
Actually, the only possible obstruction is the one at n = 1 because Ln−1 kψi is
an L0 eigenstate with eigenvalue n − 1, and hence automatically a BRST commu-
tator unless n = 1.
30
The Other Liouville Dressing
At level two, the mass shell condition, which we may as well impose, is (α0 2 +
φ0 2 )/2 = −2. The space of states at G = −1 is three dimensional, generated by
b−2 |p, ↓i, b−1 α−1 |p, ↓i, and b−1 φ−1 |p, ↓i. We get
√
1 2 1 2
Qb−2 |p, ↓i = α0 α−2 + α−1 + φ0 φ−2 + φ−1 − 2i 2φ−2 + 3c−1 b−1 |p, ↓i
2 2
2
√
Qb−1 α−1 |p, ↓i = α0 c−1 b−1 + α0 α−1 + α−2 + (φ0 − i 2)φ−1 α−1 |p, ↓i
√ 2 √
Qb−1 φ−1 |p, ↓i = α0 α−1 φ−1 + (φ0 − i 2)φ−1 + φ−2 + (φ0 + i 2)c−1 b−1 |p, ↓i.
(3.31)
√
The kernel of Q in this space vanishes unless α02 = 1/2, that is, α0 = ǫ/ 2
√
with ǫ = ±1. There are then two possible Liouville dressings, with φ0 = ±3i/ 2.
√
At φ0 = −3i/ 2, Q remains acyclic at G = −1, as the reader can easily verify
31
√
from (3.31). But at φ0 = 3i/ 2, the cohomology is one dimensional, generated by
ǫ i
b−2 − √ b−1 α−1 + √ b−1 φ−1 |p, ↓i. (3.32)
2 2
i
√
x = cb + √ (∂X − i∂φ) · ei(X+iφ)/ 2
2
√ (3.33)
i
y = cb − √ (∂X + i∂φ) · e−i(X−iφ)/ 2
2
that were discussed in the text and shown to generate multiplicatively the entire
ring of spin zero, G = 0 operators. They also correspond to the spin one, G = −1
√
currents b · exp(±i(X ± iφ)/ 2).
The other phenomena that we found at level one all recur at level two. In the
text, the important states at G ≥ 0 were G = 0 states that give rise to currents.
At level two, these are
√ 2
α−2 − ǫ 2α−1 |p, ↓i, (3.34)
√ √ √
±
W3/2,ǫ/2 = ∂ 2 X − iǫ 2(∂X)2 exp(iǫX/ 2 + φ(2 ∓ 3)/ 2). (3.35)
32
APPENDIX (2): Functions And Vector Fields On Certain Quadrics
in IR4 .
′ ′ ′
φ = wA1 A2 ...An ;A′1 A′2 ...A′n xA1 A1 xA2 A2 . . . xAn An , (3.37)
with w symmetric in all Ai and all A′j , transforms as spin (n/2, n/2). Any poly-
nomial function on the sphere is a linear combination of these, for the following
reason. If one were to antisymmetrize in a pair of indices, say A1 and A2 , to make
′ ′
ǫA1 A2 xA1 A1 xA2 A2 , then by bose statistics one may as well antisymmetrize in A′1
′ ′
and A′2 . But ǫA1 A2 ǫA′1 A′2 xA1 A1 xA2 A2 = i xi 2 . Such a factor can be discarded, as
P
it equals 1 on the sphere. This justifies the claim made in §2 that the polynomial
functions on the sphere transform as
M
(n, n). (3.38)
n=0,1/2,1,...
33
reason. ∆φ is a polynomial in the x’s of degree n − 2 that must transform with
spin (n/2, n/2); as x has spin (1/2, 1/2), this is impossible. This justifies the claim
in §2.6 that every polynomial on the sphere can be extended on IR4 as a solution of
the Laplace equation. Conversely, any polynomial on IR4 that obeys the Laplace
equation is a linear combination of the φ’s. To show this, it is enough, by the
SU(2) × SU(2) symmetry and the homogeneity of the Laplace operator, to show
P 2 n ′
that for n > 0, i xi · (x11 )m is not a solution of the Laplace equation. This
′ ′
is easy if one recalls that the Laplacian is ∆ = ǫAB ǫA B ∂AA′ ∂BB ′ .
4
X ∂
f i (xj ) (3.39)
∂xi
i=1
with each f i being a function. The ∂/∂xi transform as (1/2, 1/2). For each i, the
space of possible f i ’s, modulo functions that vanish on the sphere, transforms as
L L
n≥0 (n, n). By combining (1/2, 1/2) with n≥0 (n, n), we get
M M M
(n, n) (m, m) ((r + 1, r) ⊕ (r, r + 1)) . (3.40)
n≥0 m≥1/2 r≥0
We want to restrict this to vector fields on the sphere, that is vector fields that leave
the surface i x2i = 1 fixed, or in other words vector fields such that xi f i = 0 on the
P
M M
(m, m) ((r + 1, r) ⊕ (r, r + 1)) . (3.41)
m≥1/2 r≥0
34
of the vector fields of the type
f i = ∂iw (3.42)
for any function w. The w’s would transform like (3.38), but a constant term in
w is irrelevant. Removing the vector fields (3.42) thus amounts to removing the
first term in (3.41), and thus the volume preserving polynomial vector fields on the
sphere transform under SU(2) × SU(2) as
M
((r + 1, r) ⊕ (r, r + 1)) (3.43)
r≥0
as was claimed in §2. The various pieces can be described very explicitly. The
vector fields
′ ′ ′ ′ ′ ∂
wA1 ,...,An ;A′1 ...A′n+2 xA1 A1 . . . xAn An xBAn+1 ǫAn+2 B (3.44)
∂xBB
′
transform as spin (n/2, n/2 + 1). Volume preserving vector fields of spin (n/2 +
1, n/2) are constructed similarly.
REFERENCES
35
4. E. Witten, “On String Theory And Black Holes,” Phys. Rev. D44 (1991)
314.
36
13. J. Avan and A. Jevicki, “Classical Integrability And Higher Symmetries of
Collective String Field Theory,”BROWN-HET-801 (1991).
37