Skip to main content

Full text of "On the Exact Operator Formalism of Two-Dimensional Liouville Quantum Gravity in Minkowski Spacetime"

See other formats


On the Exact Operator Formalism of 
Two-Dimensional Liouville Quantum Gravity 
in Minkowski Spacetime 



YoiCHi Kazama[| 

Institute of Physics, University of Tokyo, 
Komaba, Tokyo 153 Japan 

and 

Hermann Nicolai[] 
n. Institut fiir Theoretische Physik, Universitdt Hamburg 
Luruper Chaussee 149, 2000 Hamburg 50, F.R.G. 



Abstract 

A detailed reexamination is made of the exact operator formalism of two-dimensional 
Liouville quantum gravity in Minkowski spacetime with the cosmological term fully taken 
into account. Making use of the canonical mapping from the interacting Liouville field 
into a free field, we focus on the problem of how the Liouville exponential operator should 
be properly defined. In particular, the condition of mutual locality among the exponential 
operators is carefully analyzed, and a new solution, which is neither smoothly connected 
nor relatively local to the existing solution, is found. Our analysis indicates that, in 
Minkowski spacetime, coupling gravity to matter with central charge d < 1 is problem- 
atical. For d = 1, our new solution appears to be the appropriate one; for this value of 
d, we demonstrate that the operator equation of motion is satisfied to all orders in the 
cosmological constant with a certain regularization. As an application of the formalism, 
an attempt is made to study how the basic generators of the ground ring get modified 
due to the inclusion of the cosmological term. Our investigation, although incomplete, 
suggests that in terms of the canonically mapped free field the ground ring is not modified. 

^e-mail address: kazama@tkyvax.phys.s.u-tokyo. ac.jp 
■^e-mail address: i02nic@dhhdesy3.bitnet 



1 Introduction 



Beginning with the discovery |T[] that the matrix model techniques together with the idea 
of the double scaling limit can be used to define and solve models of two-dimensional Eu- 
clidean gravity (or random surfaces), our understanding of the subject has been advanced 
considerably. (For review see for example |0 and references cited therein.) Despite these 
successes, however, there still remain a number of important issues yet to be clarified. 
Notable among them is the problem of space-time interpretation. Due to its formulation 
without the explicit appearance of the metric, the matrix model offers only a modest 
insight into how the gravitational and the matter degrees of freedom interact. 

One possible way to clarify this problem is to study how one can make contact with 
the continuum formulation, for example in the conformal gauge, where the metric degree 
of freedom appears explicitly as the Liouville field. Here the problem is how to treat the 
notoriously tricky dynamics generated by the cosmological term /i^e*^. 

Recently there have been attempts to understand Liouville dynamics via path-integral 
formalism with certain amount of success In this method one first performs the 

integration over the Liouville zero mode and observes that if the total "momentum" , which 
is in general fractional, were a positive integer the remaining integral becomes effectively 
that for a free field with vertex operator insertions (although with a formally divergent 
factor). Upon performing this free- field integral one "analytically continues" back to the 
original fractional momentum. The method appears to be non-rigorous (see however |^ 
for justification) but it reproduces the matrix model results [10| for up to 3 point functions 
on a sphere and a torus. 

Although it has captured an important aspect of the Liouville theory, the formalism 
above which integrates out the Liouville zero mode in the beginning appears to be awk- 
ward for unravelling the deep symmetry structure of the theory. For this purpose, the 
operator formalism would appear to be more suitable. Indeed, the existence of discrete 
physical states [|ll|] -[[l^] and the associated Woo algebra |]18| first found in the ma- 
trix model approach have been analyzed in the operator formalism and such intriguing 
structures as the "ground ring" [^|-|^| have been discovered. These developments have 
clearly demonstrated the utility of the operator formalism but there is a caveat: in these 
works the Liouville field is treated as a free field and hence the role of the Liouville dy- 
namics remains unclear. Also it can be suspected that the naive perturbative treatment 
of the exponential interaction may not be justified. In view of this situation, it is an 
appropriate time to focus on the exact operator formalism for the Liouville theory and 
examine how it can be applied to the clarification of the aforementioned problems. 



2 



Exact operator quantization for the Liouville theory has been developed over the past 
ten years |^ (for a review see In apphcations of the operator formahsm to 
two-dimensional quantum gravity coupled to matter are discussed from a modern point of 
view, where special emphasis is placed on the quantum group structure of Liouville theory, 
and various results previously obtained by matrix model techniques are rederived (see also 
| 36| for a recent review). Despite the fact that many intriguing results have been obtained 
in these works, we feel that some subtle issues have been left unsettled (or overlooked). 
A major problem, and one that has still not been satisfactorily resolved in our opinion, 
concerns the rigorous construction of exponential Liouville operators of arbitrary weight 
having the right conformal properties and satisfying locality. The purpose of this article is 
to try to point out these and other subtleties as clearly as possible and propose solutions 
to some of these problems. Because of the technicalities involved we shall not explain the 
nature of the subtleties in detail here, but nevertheless let us mention one novel result of 
our investigation which may have serious consequences. As in the traditional treatment, 
we shall perform a canonical operator quantization in Minkowski space by a non-linear and 
non-local canonical transformation which maps the interacting Liouville field into a free 
field. In this approach, the requirement of mutual locality among exponential operators 
plays a crucial role. The locality condition was recently reanalyzed by Otto and Weigt in 
|3^ . They proposed a new definition of the exponential operator that is obtained from 
the usual exponential series by a kind of quantum deformation, and checked the locality 
of these operators in a perturbation expansion in terms of the cosmological constant /i^ 
up to third ordeiQ. We have carefully reexamined this locality equation and found an 
additional solution (valid to all orders in /i^) which is not smoothly connected to the 
solution of |32|. Moreover, our analysis casts some doubt (at least in Minkowski space) 
on the validity of the conventional treatment of the theory coupled with d < 1 matter. 
If we postulate that the operator e*^, which appears in the Liouville equation of motion, 
should belong to the set of admissible (i.e. mutually local) operators, we find that (i) 
(i < 1 is not allowed and (ii) for = 1 our solution, not the one constructed in |Q, is 
relevant. Since the notion of locality (or micro-causality) does not enter in the Euclidean 
treatment, our result is not necessarily in direct conflict with the matrix model, but it 
certainly calls for further investigation. 

With the formalism developed for the d = 1 case, we then made an attempt to study 
how the structure of the ground ring gets modified when the exponential interaction is 
turned on. The results we have obtained so far are unfortunately difficult to interpret. 



^ J. Schnittger has informed us that he has now verified the locahty condition for the deformed quantum 
Liouville operator proposed in to much higher orders. 



3 



To be sure, it is straightforward to write down candidates for the basic generators of the 
ground ring in terms of the interacting fields with the expected conformal tranformation 
properties. However, the BRST transformation properties are not completely dictated by 
the conformal properties alone, and these candidates, unlike their free field counterparts, 
turn out not to be BRST invariants. This seems to indicate that we have exactly the 
same ground ring generators as for vanishing cosmological constant. Nonetheless, there is 
a large difference with the the fi^ = case. The free field in our case is not the Liouville 
field itself, but the canonically mapped field which is difficult to interpret physically. To 
put it differently, a simple generator in terms of the free field is actually a very complicated 
one in terms of the Liouville field. It is not entirely clear to us how our results are related 



to those of |38|,[3£], where the ground ring with non- vanishing cosmological constant is 



also discussed. In |^ , the basic idea is to determine the deformations of the ground ring 
by "fusion", i.e. by evaluating operator products of the free field ground ring operators 
in a perturbation expansion in /i^; in this way one arrives at the conclusion that there 
appear extra operators besides the ones already present in the free field case. Since we have 
not computed products of (interacting) ground ring operators, we cannot ascertain the 
compatibility of the two approaches at this point. We note the fact that the expansion 
in /i^ of the exact ground ring operators, on which our analysis is based, breaks off 
after finitely many terms precisely for the special weights which appear in the associated 
Liouville exponentials. It would be interesting to see whether this observation is related 
to the fact that only finitely many terms remain in the perturbative expansion of [^| due 
to charge screening. Obviously, more work is needed to clarify these issues. 

Despite a long history, pedagogical expositions of the operator approach to Liouville 
theory seem to be scarce. For this reason, we have decided to organize our paper in such 
a way that it can (hopefully) also serve as a relatively self-contained review. We start in 
section 2 with a review of the classical properties of the Liouville theory. In particular, 
we provide a rather detailed exposition of the non-linear and non-local canonical trans- 
formation which maps the interacting Liouville field into a free field. With the use of this 
canonical tranformation, we proceed to the operator quantization of the theory in section 
3. The central issue will be the question of the proper definition of exponential opera- 
tors and their locality properties. We will also demonstrate that the operator Liouville 
equation of motion is satisfied to all orders in /i^ for certain set of values for the matter 
central charge including d = 1. In section 4 we shall describe an attempt to construct the 
basic generators of the ground ring including the cosmological term. Finally in section 5 
we will give discussions on the issues which are left unsolved in this work. An appendix 
is provided to discuss the conversion between Minkowski and Euclidean formulations. 



4 



2 Classical Properties of Liouville Theory 
2.1 General Remarks 

In this section we describe some classical properties of Liouville theory |2^-||3^; as for 



our conventions and notation we will mostly follow Let us briefly recall how the 

Liouville action arises in string theory, and why it plays an important role there. Any 
conformal field theory can be formulated as a two-dimensional field theory of matter fields 
coupled to a gravitational background, which is described by a metric tensor g^iy. In two 
dimensions, (/^i, carries no propagating but only topological degrees of freedom. More 
specifically, it can always be brought to the form 

^?M- = e^^^. (2.1.1) 

by means of two-dimensional diffeomorphisms. Here, the background metric ^^j^ locally 
can be chosen equal to the fiat Minkowski metric (i.e. in each coordinate patch, we 
can set Qf^^, = rj^y). Globally, it will depend on the moduli characterizing the conformal 
equivalence class of the world sheet; we will, however, ignore these topological degrees 
of freedom in most of the remainder. The factor multiplying the background metric 
g^y will be referred to as the "conformal factor" in the sequel, and ip is the Liouville 
field. In a conformally invariant theory, the classical Lagrangian is not only invariant 
under two-dimensional diffeomorphisms, but in addition invariant under Weyl rescalings 
of the metric. This invariance allows us to gauge the conformal factor to unity, and the 
Liouville field therefore decouples from the matter sector. However, as is well known P0| , 



this decoupling in general cannot be maintained at the quantum level, because integration 
over the matter fields induces the quantum action 

S = ^l drda^g [^g^^^d^ifd^cp - fih^ - QcpR) (2.1.2) 

or, with g^^ = r^^^,, 

S = ^J drda - /iV) , (2.1.3) 

where /x^ is the cosmological constant, Q the background charge. Furthermore, we have 



1 = ^ (214) 
72 - 487r ■ ^ ' 

d is the number of scalar matter fields or, more generally, the total central charge of 
the matter sector (which need not even be described by a Lagrangian). Consequently, 
^For definiteness, we consider only bosonic theories. 



5 



at the quantum level, the Liouville field decouples only for d = 26, which is the critical 
dimension for a bosonic target space. For d < 26, on the other hand, the Liouville 
field (fi must be taken into account as an extra degree of freedom, whose dynamics is 
governed by the non-polynomial action ( 2.1.2| ) [|. In order to make sense out of such a 



"subcritical" theory, one is thus forced to study the Liouville action in the context of 
two-dimensional quantum field theory. The hope, then, is that through its non-linear 
dynamics, the Liouville field adjusts its contribution to the conformal anomaly in such 
a way that conformal invariance is restored in the full theory, and thereby a consistent 
theory can be formulated also outside the critical dimension. We remark that, in many 
approaches, the (renormalized) cosmological constant is set to zero, so the complicated 
dynamics of the Liouville field is masked by an effective free theory. However, we will not 
adopt this point of view and assume /i^ 7^ in this paper. 

One of the central problems of Liouville theory is the construction of quantum oper- 
ators corresponding to exponentials of the Liouville field, i.e. e'^'^ for arbitrary weights 
A. We will discuss this problem in much detail in section 3.2, but let us emphasize al- 
ready at this point that these operators are indispensable for the construction of physical 
vertex operators. The main reason is that ordinary string vertex operators built out of 
matter fields only will usually fail to have the correct conformal dimension, and must be 
"dressed up" by suitable powers of the worldsheet metric and its determinant before they 
can be consistently integrated over the world sheet. Since the conformal dimension of e'^'^ 
depends upon d, the conformal dimensions of the matter field vertex operators will also 
have to be modified away from the critical dimension. 

Before moving on to describe the classical aspects of Liouville theory, we note another 
important point. Due to quantum effects, we will have to replace the coupling constant 
multiplying the action ( 2.1.2| ) by the renormalized value |27 



(-) ^^ = ^ (215) 
V72 i 487r IQTTh ■ ^ ■ ■ ^ 

N ' / ren 

The shift from 26 to 25 is a quantum effect and will be explained below. In the above 
equation, we have also introduced "Planck's constant" h = 25^^- This terminology has 
the advantage that we can think of the limit — — 00 as some kind of semi-classical limit 
in which the theory simplifies considerably. 



■^The case d > 26 is not of interest as there are always negative norm states in the physical Hilbert 
space. 



6 



2.2 General Solution 

The classical equation of motion following from the action ( p.l.3|) reads 



4a+9_¥? + /iV = 0, (2.2.1) 
where, for later convenience, we have introduced the light-cone coordinates 

i^ = T±a d± = \{dr±d,) dl-dl = Ad+d-. (2.2.2) 



( p.2.1|) is a non-linear partial differential equation, whose general solution has been known 



for a long time (and was, in fact, known to Liouville ^^). It is most conveniently 
expressed in terms of left and right moving waves, i.e. two arbitrary functions A = A{C,~^) 
and B = B{^-): 

8 d+A{e)d.B{n 
^fi^l + A{^^)B{^-)y 

Observe that this representation is not unique and differs from the one used in earlier 



^(r..) = log|^ ";-';^'— t 'J - (2.2.3) 



papers [27|. The above parametrization was chosen by |31][32], and will turn out to be the 
most convenient for our subsequent considerations. The fact that the above equation can 
be solved exactly may be regarded as a consequence of the integrabiliy of Liouville theory, 
which follows from the existence of a Lax pair |^T|. Periodicity in the space coordinate 
requires the boundary condition ip{T,a + 27i) = ip{T,a). This does not imply that the 
"constituent functions" A, B are themselves periodic; rather, the periodicity constraint 
on the Liouville field is compatible with SL{2, R) (i.e. Mobius) transformations on A and 
B, viz. 

where a6 — (3'y = 1. Note that B transforms "contragrediently" to A; this is the reason 
why some authors prefer to work with the field —B{C,~)~^ instead, which transforms like 
A. In choosing boundary conditions below, we will not make use of the full Mobius group, 
but rather put /3 = 7 = 0. The solution ( p.2.3|) can be equivalently expressed as 



(/.(r,cr) = log9+A(e+)+log(^-|a_i?(r)j -21og(l + r(r,a)) , (2.2.5) 

where 

y(r,a)^A(e+)5(r) (2.2.6) 
is the only part of mixing left and right movers. 



7 



Having exhibited the exphcit solution in terms of left and right movers, the next task 
is to construct a free field out of them. In view of the different parametrizations alluded 
to before, the identification of the free field is not unambiguous. We will here follow 
and 1^ and define ■?/' by 

^(r, a) = + nn = log9+A(e+) + log (^^d.BiC)^ (2.2.7) 

so that 

ip = ip-2log{l + Y). (2.2.8) 

These equations will serve as our basis for all subsequent considerations, in particular 
those concerning the quantum theory. 

Integration of the differential equations d^A = e^^ and d^B = (yU^/8)e^ is straight- 
forward. Assuming the simplified boundary conditions A{^^ + 27r) = aA{^^) and B{^^ — 
27r) = a'^B{^-) we obtain |S2[ 

r2Tr 



A{e) = c{a) I ^ d^' E^{t - O^^^^^'^ , (2.2.9) 
Jo 

Bin = ^c{a) r dCE^iC (2.2.10) 
8 Jo 



where 



c{a)=(^-^ , (2.2.11) 



E«(0 = exp (ilogae(O) = , (2.2.12) 



and 



with e(^) the usual staircase function, i.e. e(.^) = 2?7, + 1 for 2r?,7r < < 2[n + l)7r. When 
written out, ( p.2.9|) reads 



A{t) = c{a) ( [^^ d^'e^^°^''e^^^^'^ + ^ d^'e-'^^^^'^e^^^^'n . (2.2.13) 



10 

The function (which is actually a distribution) obeys 

|^^»K) = J^^tt). P.2.14) 

where S{^) is the 27r-periodic delta- function. Note that we cannot choose the functions 
A, B to be simply periodic because c{a) diverges at a = 1. Solving for F(r, a), we obtain 

Y{T,a) = ^c{af r da' ^ da" E^{a - a')Ei{a - a")e'^'+(^+'^')e^"("-'^") . (2.2.15) 
8 Jo JO 



8 



It is now easy to check that F is a periodic function of cr, ahhough A and B are not 
periodic separately (note that with the more general boundary condition (|2.2.4|) Y would 



not have been periodic). Since, for the canonical formalism we will be mostly concerned 
with equal time expressions, we have explicitly displayed the dependence on the time 
and space variables in ( p^.2.15| ); this will permit us to put r = wherever appropriate. 
Although we will not make use of this fact, we note that Y{t, cr) can also be represented 
directly in terms of the free field ip and its time derivative at time r. The resulting 
expression is local in r, but non-local in the space variable a. 

Given the result for the Liouville field as a non-local function of the free field ip, we 
can now define exponentials with arbitrary weights A through the expansion 

(We have displayed the expression appropriate for positive A ) Observe that this is actually 
an expansion in powers of the cosmological constant, as Y is proportional to fi"^ by ( p. 2. 15 ). 



Note also that when — 2A is a non-negative integer the expansion contains only a finite 
number of terms. 

In the remainder, we will also make use of the Fourier expansion of the free field ip 
in terms of oscillators. For the left and right moving fields ip"^ this expansion takes the 
form 

^H^^) = bQ + + ^ E -4^^e-^<" , (2.2.17) 

where the extra factor of 7 has been inserted so as to obtain the canonically normalized 
Poisson bracket relations for Q, P and the oscillators a^j^'' ■ Comparison with ( p.2.5| ) leads 
to the identification 

loga = i7P. (2.2.18) 

We have already remarked above that we cannot put a = 1 in the expressions for the 
interacting Liouville field. Now we can understand this result in a more physical way: 
a 7^ 1 is equivalent to P 7^ 0, i.e. Liouville theory does not possess a translationally 
invariant groundstate; of course, this problem becomes apparent only for 7^ 0. There is 
another, and related, difficulty here. Any shift of the Liouville field by a constant can be 
absorbed into the cosmological constant, which is therefore an essentially undetermined 
parameter. This, in turn, means that there is no way in which ( |2.2.1(j| ) can be thought of 



as an expansion in a small parameter. In the absence of a better method, we will proceed 
in spite of this difficulty. 



9 



2.3 Canonical Transformation 

The Poisson brackets are determined from the action (2.1.3). The equal time brackets are 
given by 

{(p{T, a) , (p{T, a')} = {(p{T, a) , (fir, a')} = for a 7^ a , (2.3.1) 

{^{t, a) , \^{r, a')} = 6{a - a') , (2.3.2) 

7 

where the dot denotes the derivative with respect to r. The factor in front of ip is just 
the couphng constant multiplying the action ( p.l.3| ). 

It is quite remarkable that the non-local transformation connecting the Liouville field 
(p and the free field ip is canonical |2^,|^. In the present formulation, this means that 
the above brackets are equivalent to 

{^{T,a),^{T,a')} = {^{T,a)J{T,a')} = for a ^ a' , (2.3.3) 

{^(r, a) , \i;iT, a')} = 5(a - a') . (2.3.4) 

7 

The proof of this assertion requires a rather tedious computation. Since our setup, and 
in particular, our definition of the free field in (2.2.7), is different from the one in p7| . 



p8| , we present the details of this proof here. We have found that the simplest method is 



to proceed "backwards" by showing that the free field Poisson brackets imply the ones in 
terms of (p above. We now give some useful intermediate relations. Since all of these are 
equal time commutators, we will omit the time coordinate in the formulas listed below; 
however, we alert the reader that in all relations involving time derivates, we must keep 
r as a variable at the intermediate stages of the calculation and can put r = only 
at the very end. Also one must not forget that a contains the zero mode P and hence 
has non-trivial bracket with ip, which contains Q. A prime ' denotes the derivative with 
respect to the spatial coordinate a. 

Let us first list the brackets between the free field ip and the quantities A, B; to make 
the formulas less cumbersome, we put 7^ = in eqs. (2.3.5-21). So we obtain 

{ij{a^) , A{a,)} = ^A{a,)-^A{a^)E^ia,-a2) 

47r 2 a 

~A{a2Hai - a2) , (2.3.5) 

{^(ai) , A'ia^)} = ^A'ia^) - ^A'^aM^i " ^2) , (2.3.6) 
47r 4 



10 



{V^(ai) , A{a2)] = (ai - a^) , (2.3.7) 



{^(ai) , A'{a2)} = -^A'{a2)S{ai - cxa) , (2.3.8) 
47r Z « 

+ ^e(ai-a2)5(a2), (2.3.9) 

{^{ai) , B'{a2)} = -^B'{a2) - \e{ai - a2)B'{a2) , (2.3.10) 

ii^ia,) , B{a2)} = +^E^{a2 - a,)B'{a^) , (2.3.11) 

{^(ai) , B'{a2)} = ~B'{a,)6{ai - ^2) . (2.3.12) 
Prom these brackets, we derive 

{^(ai), r(a2)} = ~cia)(A{ai)Er{a,-a2)B{a2) + A{a2)Ei{a2-a,)B{a,)) 

= {V^((72) , r(ai)} , (2.3.13) 

{V'(ai) , y((72)} = ~c{a) {A'iai)E^{ai - (72)5(^2) - Aia2)E^ia2 - (T,)B'{a^)) 

= {^(a2),r(ai)}, (2.3.14) 

{iP{ai),Y{a2)} = ^c{a)(A{ai)E,{ai-a2)B'{a2)-A'{a2)Ei{a2-ai)B{ai 

= {V'(a2),y((Ji)}, (2.3.15) 

{V'(ai) , y(a2)} = -^5(ai - (72)^(^2) (2.3.16) 



^c(a) (^'(ai)£;i((7i - a2)B'{a2) + A\a2)E^{a2 - ai)B'{a,)) 



+ 2 

The equal time relations involving commutators of Y and Y read 

{y(aO , y(<72)} = liYia,)-Yia2))ma^),Yia2)} 

= ^(y(ai)-n^2))W^2),n^i)}, (2.3.17) 

11 



{Y{a,),Y{a,)}= - ]^{^{a,) ,Y{a^)}Y{a,) 



+ i(F(ai)-rM)Mai),F(a2)}, (2.3.18) 



{y(ai),r(a2)}= - \{i^{a^),Y{a{)}Y{a,) + \{i^{a{),Y{a,)}Y{a{) 

+ (2.3.19) 



These results can now be inserted to rewrite ( 2.3.1 ) and (|2.3.2| ) in terms of free field 
commutators. For instance, we get 

Men), V^(o^2)} = - ^^l, M {^i).Y{a,)}+ I Ma2),y(a0} 

1 + 1(0-2) 1 + J'^(cri) 

The last term can be rewritten by use of (2.3.17) 
r(ai) - F(a2) 



[l + Y{a,m + Y{a,)) 



{^(ai),F(a2)} + {^(a2),r(ai)}) 



= TTv^^^^^i) ' - i^w^^ i^(^^) ' • (2.3.21) 

l + y(cr2j l + y(o-i) 

and thus precisely cancels the first two terms. The proof of the remaining Poisson brackets 
along these lines is now completely analogous. 

By standard arguments it can be shown that the free field Poisson brackets are equiv- 
alent to the following brackets for the oscillators 

{g,P} = l , i{a^\a^^^} = i{a^J,ai^} = m5m+nfi. (2.3.22) 

(The brackets that have not been listed vanish). 

Although the proof of the canonical nature of the transformation is already complete, 
it is instructive to display explicitly how the Hamiltonian gets transformed. The Hamil- 
tonian following from the action ( |2.1.3| ) is given by 



= ^Jdcr (^0' + 1^'' + f^'e^) , (2.3.23) 



12 



where 11^ = is the canonical momentum. We now substitute the expression ( p^.2.8| ) and 



7- 

make use of identities such as 

d+^l)d+Y = dlY, (2.3.24) 

^_^p^_Y = dlY , (2.3.25) 

d+d-Y = ^e^, (2.3.26) 
8 

d+Yd^Y -Yd+d^Y = 0, (2.3.27) 
which easily follow from the definitions of Y and ip. After some calculations we obtain 
H = lj^'''da[^{^' + i;")-AdlHl + Y) 

= :^r^(^' + ^")- (2.3.28) 
This is nothing but a Hamiltonian for a free field ip. So there are two ways to look 
at the Liouville dynamics: one can say that the Liouville field evolves in a complicated 
way because of the exponential interaction. Alternatively one may say that the Liouvile 
dynamics is non-trivial because the Liouville field is a complicated combination of a free 
field which evolves trivially. 

2.4 Energy-Momentum Tensor and Conformal Transformations 

The energy momentum tensor is defined by 

T,. = 4.^. (2.4.1) 

It is not difficult to see that the trace of T^jy is proportional to the Liouville equation of 
motion and therefore vanishes on-shell. It is in this sense that we can regard Liouville 
theory as a conformal field theory, although the basic Liouville field does not decompose 
into a sum of left and right movers. Switching to light-cone notation, we find for the 
remaining traceless components 

Note that we must keep the fiducial metric g^i, in order to get the improvement term, 
and can equate it with the fiat Minkowski metric only at the very end of the calculation. 
From conservation of the energy monentum tensor (i.e. d^T^^ + 9_|_T__|_ = 0) and its 

tracelessness (i.e. T+_ = 0), we expect (T ) to depend only on ^+ This is 

borne out by an explicit computation, which yields 

T++ = ^(5±^9±^ - 2dlij) . (2.4.3) 
T 



13 



Therefore, despite the comphcated dependence of the Liouville field on r and a, the ++ 

and components of the energy momentum tensor take a rather simple form in terms of 

free fields, being equivalent to those of a free field theory, apart from the improvement term 
containing second derivatives. In other words, as far as its energy momentum tensor is 
concerned, Liouville theory behaves very much like a free field theory (with a background 
charge) . 

The Virasoro generators L^-*, defined from the expansion 

T^±i^^)=Y.&-'"'^^ (2.4.4) 

m 

are given by (cf. ( p.2.17|) ) 



^^^^ = I E 4^^«^-n + ^^rnal:t^ ■ (2-4.5) 

n T 

In the sequel, we will omit the superscript (±) if there is no danger of confusion. By use of 
the canonical brackets given in the preceding section, one recovers the Virasoro algebra 

4:71 

i{Lm, Ln} = (m - n)Lm+n + -^rn?5m+nfi ■ (2.4.6) 

7 

Consequently, there is a central term already at the classical level. Note the the anomaly 
term proportional to m? can be converted into the more usual form m{m? — 1) through 
the replacement of ma^^^ by (m + 1)0^-* in ( p.4.5|) . 

At this point, we can explain the shift from 26 to 25 mentioned in section 2.1. As is 
well known, in the quantum theory, the expression for the Virasoro generators must be 
normal ordered if it is to be remain well defined. This results in an extra contribution of 
^ to the central charge. Since we want the total central charge to be given by 26, we 
have to readjust its classical value in such a way that 



c _ 1 /47r\ _ 26-rf 
12 ~ 12 ^ J ~ ~[2 



+ (-1 =^7^' (2.4.7) 



which leads to the renormalized value given in (|2.1.5D . We will henceforth drop the 
subscript "ren", and always assume the renormalized value for the coupling constant. 



3 Quantization of Liouville Theory 
3.1 Free Field Quantization 

The main result of the foregoing chapter was the demonstration that the interacting 
Liouville field can be canonically reexpressed in terms of the free field ip defined in (2.2.7). 



14 



In quantizing Liouville theory, one tries to exploit this equivalence by performing the 
quantization in terms of this free field. In this subsection, we set up our conventions and 
notation and collect some basic (and well known) formulas for free fields. 

First replace Poisson brackets by commutators in the usual fashion so that 

[Q.P]=i , [aW,aW] = [a(„-),ai-)]=m<5^+„,o (3.1.1) 



(again, commutators that have not been listed vanish). This can be inserted into ( |2.2.17| ) 
for the free field ip to derive the quantum commutators 

[V'(r,a),^(r,a')] = [^(r,a),^(r,a')] = for a ^ a' , (3.1.2) 

[^[r,a),^i^{T,a')]=t5{a-a'). (3.1.3) 

For the derivation of short distance expansions and the computation of vacuum expecta- 
tion values it will be convenient to split the field if) into creation and annihilation parts 
by defining 

V n=l 

and 

^c^(e^)--^£-a^V^"^', (3.1.5) 

where zero mode terms involving Q and P have been left out. 

All relevant formulas can now be derived from the reordering relation (for the non-zero 
modes) 

and the corresponding formula for exponentials of '?/'~(^~), which reads exactly the same. 
The relation for the zero modes is 

e'^«/(P) = /(P + za)e"^. (3.1.7) 

To establish these relations, we need the formula 



-ln|l-e-^|-z^e(^^) + z^, (3.1.8) 



15 



which is obtained by carefully looking at the phase of the logarithm in the region where 9 
is small. In deriving the reordering relations above, one encounters infinite sums which fail 
to converge without regularization. The sums are regularized by adding a small imaginary 
part to the argument 9; of course, e must be taken to zero at the end of the calculation 
(this shift is very similar to the well known ie-prescription needed to make Feynman 
integrals well defined). The phase factors will be especially important in our analysis of 
the locality condition in section 3.3. 

Normal ordering is now defined in the usual fashion by putting all annihilation op- 
erators to the right, except that in the zero mode sector we use the symmetric normal 
ordering given by 

: e"'2/(P) := e5"'2/(p)e5»Q . (3.I.9) 
In accordance with these remarks, the fully normal ordered exponential is now given by 

: exp {aip{T, a)) : = exp (|a7Q)) exp j^-P^^ exp (^^07(5) exp (aip^{T + a)) 

• exp (aip+{T + (j) exp (aV'-(T - cr)) exp (aV'^lr - cr)) . 

(3.1.10) 

By means of the reordering relation (3.1.6), it is straightforward to prove that 



(3.1.11) 



and 



(3.1.12) 

so the product of two normal ordered exponentials equals the fully normal ordered product 
up to a "short-distance factor". Multiplying these two expressions and noticing that, at 
equal times, ^i" — = ~C2) — ^\~'^2i we see that the phase factor cancels between 
the left and the right moving sectors. The final result is therefore symmetrical, so that 
for arbitrary weights a and 6, we deduce 

. gaV'Cr.ai) .. ^ij.a^) ._. ^H{r,a2) .. ^a^{T,a,) . (Ti 7^ (T2 . (3.1.13) 



16 



This shows that two free-field exponentials are indeed mutually local, although this prop- 
erty does not hold for the left- and right-moving sectors separately because of the extra 
phase factors. 

It is also well known that the normal ordered free field exponentials behave properly 
under conformal transformations. With the normal ordered Virasoro generators 

L'^' = \Y.-- -'rt'a'^'-n : +^ma^^' , (3.1.14) 

n 1 

one obtains 

[Lm, Ln] = (m - n)Lm+n + —m^Sm+n,o , (3.1.15) 

where the central charge is now given by ( |2.4.7| ); as we explained there, the contribution 
from the renormalized coupling constant, which appears already in the classical algebra 
( p.4.6| ), is augmented by ^ from the normal ordering of the oscillators above. A standard 
calculation shows that 

'Li , : e'^^^"'") : ] = e*"^«* (^-id± + m{a - ^a^)^ : e"'^^"'") : . (3.1.16) 

The customary Euclidean formulation of conformal field theory makes use of operator 
product expansions. These are most conveniently derived by the use of Wick's theorem. 
For convenience, we define the variables 

= e*("^") (3.1.17) 

(or just z, w, ... if there is no danger of confusion). As is well known, these variables can be 
analytically continued to imaginary times by the replacement r —ir (in the appendix, 
we will provide further details of the transcription between Minkowski and Euclidean 
conventions). It is now elementary to show that 

2 

'y 111 z 

d+ij{z)d+^{w) =: d+,lj{z)d+^{w) : . (3.1.18) 

47r [z — wY 

Here, the ze-prescription introduced above is understood; in the Euclidean formulation, 
it becomes unnecessary as the singularity at coincident points is avoided by radial order- 
ing. Now using Wick's theorem to evaluate such products, we can derive the important 
operator product expansion 

^-(^)^-(-) = i + (;^^-(») + (i^^^- w • ("-i^* 

where d now denotes the partial derivative with respect (and not ^^); as is customary 
in conformal field theory, we have dropped the non-singular terms on the right hand side. 
From the formulas given in the appendix, one can easily verify that the somewhat unusual 
factors of zw disappear upon conversion of these formulas into Euclidean language. 



17 



3.2 Quantum Definition of exp{X(p) 

Our aim in this section is to define the quantized exponential of the (interacting) Liouville 
field if, which will provide the necessary "gravitational dressing" of string vertex operators 
in a non-trivial gravitational background. For lack of a better notation, we will continue 
to designate the renormalized exponential operators by : e'^'^ :, although the semicolons 
now mean something different from and should not be confused with ordinary free field 
normal ordering. The obvious point of departure for this construction is the expansion 
( p.2.16|) of the Liouville exponential in terms of the free field ip. Since the quantization of 



free fields is well understood (see foregoing section), one hopes that the interacting theory 
can also be exactly quantized by exploiting this relation. However, matters are not so 
easy, and although considerable progress can be made, there remain some unresolved 
problems, most notably the question of whether the expansion in /i^ makes sense at all. 
The main difficulty is the complicated non-local and non-linear character of the relation 
between the Liouville field and ip. So, for instance, it will become apparent that free 
field normal ordering by itself is not sufficient to achieve consistency with the two main 
requirements, which we expect the exponential operators to satisfy. These are: 

(i) The exponential operator : e'^'^ : should transform properly under the conformal group, 
i.e. 

, : e^^("'") : ] = e''"^^ {-id± + mA±(A)) : e^^("'") : . (3.2.1) 

Here A-i-(A) denotes the renormalized ("dressed") conformal dimension of the exponential 
operator which, in general, will be different from the classical dimension as we already 
pointed out. 

(ii) Two exponential operators at spacelike distances should commute, i.e. they should 
be mutually local: 



for or ^ a'. (3.2.2) 



To evaluate condition (i), it is crucial that we use the free field energy momentum 
tensor to define since there is no way to make sense out of T±±(lp) before the Liouville 
field itself has been defined as a quantum operator. It is an open question, whether 
T±±{(f) can be defined as a quantum operator and whether this operator coincides with 
the free field energy momentum tensor. 

As we have shown in the preceding section, both requirements are met for normal 
ordered exponentials of free fields, but they are far from trivial to satisfy for the full 



18 



Liouville operator and will lead to stringent and unexpected constraints. For instance, 
the obvious idea of defining : e'^'^ : by simply putting (free field) normal ordering symbols 
around the full sum which defines e'^'^ in terms of the free field at the classical level, fails 
because it clashes with the first condition (as will become clear in a moment). The locality 
condition is not considered in the customary Euclidean formulation. This omission could 
be justified if one had an Osterwalder-Schrader type reconstruction theorem for Liouville 
theory, which is, however, not available at this point. This is one of our main reasons 
for sticking with a Minkowskian world-sheet: the locality condition will lead to further 
restrictions on the theory, which are "invisible" in the Euclidean formulation. These will 
be discussed in the following section. 

One of the additional elements needed in the construction of the full theory is a finite 
multiplicative (i.e. wave function) renormalization of the free field, whose necessity was 
first pointed in ||2^. We thus rescale ip according to 

V'-^r/V', (3.2.3) 

where the constant t] will be determined shortly. The rescaling of ip must be performed 
wherever ip appears, i.e. in particular inside the expression for Y. The definition of c{a) 
(cf. ( |2.2.18| ) must be modified accordingly, viz. 

loga = ir77P. (3.2.4) 



As another example, consider formula (|3.1.18|), which is replaced by 



w z 

d+^lj{z)d+^{w) =: d+^{z)d+^{w) : +2g- , (3.2.5) 

[z — wy 

where, for later convenience (and in order to remain in unison with the literature), we 
have introduced the constant g 

g = !^ = 2hr]\ (3.2.6) 



The renormalization implied by the rescaling ( p.2.3|) will entail a shift in the canonical 



conformal dimension of vertex operators and is therefore at the origin of "gravitational 
dressing" . 

The necessity of introducing t] can be seen as follows. The expansion ( p^.2.16D contains 



arbitrary powers of Y. Clearly, a definite conformal dimension can be assigned to the 
infinite sum if and only if Y has conformal dimension zero, and the full conformal weight 
is carried by the prefactor, which is local in the free field ip. So, we first make Y well 
defined, replacing it by : F :, where semicolons denote symmetric free field normal order- 
ing. As a consequence, the conformal weight of the full expression is determined by the 



19 



prefactor alone, which we now replace by : e'^^''' :, taking into account the multiplicative 
renormalization introduced above. To compute the conformal dimension of : F :, we first 
rewrite ( 3.1.1(j| ) in the form 



~id± + ■m{Xr] 



,,2^2 



: e 



(3.2.7) 



In order to arrange for : y : to have conformal dimension zero, we demand that 



r] — 2hr]'^ = 1] — g = 1 



Putting A = 1 in ( p.2.7| ) and assuming ( |3.2.8| ) to hold, we get 



^?yi/){T,cr) . 



id+ (( 



(3.2.8) 



(3.2.9) 



(and the analogous result for i^^^). Although the exponentials of ip~^ and appear 
with different spatial arguments under the integral, this is already enough to compute the 
commutator of L^-* with : Y : because L^-* commutes with the non-zero mode part of 
'?/'". As for the zero modes, the momentum dependence of the integrand can be ignored 
because is independent of Q. Furthermore, the dependence of the integrand on the 
center of mass coordinate Q is the same as in : e^'^*^'^''^-* :. After a little algebra, we obtain 

,2 



[LW , : r(r,a) :] = ^ : \c{af r da' ^ da" E^{a - a')Ei{a - a" 

8 I Jo Jo 



d_ 



(3.2.10) 



This can now be integrated by parts. The boundary terms cancel, and since the derivative 
on Ea is proportional to the 5-function by (|2.2.14| ), we can replace e*"*('^+'^') in the integrand 
by e*'"(^+'"), which can be pulled out of the integral. In this way, we arrive at the desired 
result 



[L+,:F(r,a) :] = -^e 



_d_ 



: Y(T,a) : . 



(3.2.11) 



Let us now return to equation ( |3.2.8| ); it is solved by 

1 



V 



-(25-rf±V(25-rf)(l-ci)) , 



(3.2.12) 



where we have inserted the (renormalized) value for the coupling constant in terms of d. 
This result shows not only that rj depends on d (the central charge of the matter system), 
but, more significantly, that the solution imposes a rather stringent constraint on d itself, 
and thus on the dimension of the target space in which the string moves. Since rj cannot 



20 



be complex, we must restrict d to the ranges d < 1 or d > 25. Especially, for d = 1 
and d = 25, one obtains rj = 2 {g = 1) and rj = {g = —1), respectively; in the latter 
case, however, the renormalized coupling constant diverges in such a way that the product 
r/7 = 2\/27ii stays finite. In the remainder, we will be mostly interested in the borderline 
case (i = 10. 

The renormalized dimension of the exponential, which now comes entirely from the 
factor : e'*"'^ :, is given by 

2 2 

A{X) = T]X-%^X^ = r]X- gX\ (3.2.13) 

When coupling the exponential of the Liouville field to a string vertex operator $ of di- 
mension Ao, conformal invariance requires that A(A) + Aq = 1, so that the total vertex 
operator can be integrated over the world sheet. Classically, A(A) = A, so Aq = 1 — A. 
When A (A) is deformed to the expression above, Aq must be modified accordingly. De- 
noting the renormalized ( "gravitationally dressed") dimension of $ by A, a little algebra 
shows that, with the above result, we must demand 

A- Ao = 2hri'^Ao{l- Ao) (3.2.14) 



in order to maintain total conformal dimension one. Formula ( p. 2. 141 ) is nothing but the 
famous KPZ condition [E3 . 



In summary, requiring correct behavior of the exponential operator with respect to 
conformal transformations, we have been led to the result 

: :=: e^'''^ : £ (-^)"r(2A + n) ^ ^^ ^.15) 



n=0 



n\ r(2A) 



with T) given by ( |3.2.12|) . The renormalized ("dressed") dimension of this operator is 



given by (|3.2.13|) . Let us stress once more that this formula defines what we mean by 
the semicolons on the left hand side of this equation. If instead, we had defined the 
renormalized operator by free field normal ordering, the result would have differed by 
certain short distance factors (see also the discussion below), which would have led to 
additional and anomalous contributions in the commutator of with the exponential 
operator, and thus would have destroyed the nice behavior of the operator under conformal 
transformations. 

It must be emphasized at this point that the solution as written down in (|3.2.15|) 

is not yet unique. Since the Virasoro generators do not depend on Q, and the desired 

"^We mention that for the unitary minimal models with d = 1 r^-prr, one obtains two solutions 

9+^-^ and g_ = 



21 



commutation relation with works order by order, we are free to replace ( |3.2.15| ) by 
the modified expansion 



■ t ^-^'-^Un (: y 0" . (3.2.16) 



n=0 



n\ r(2A) 



where fn{P) are arbitrary functions of the center of mass momentum P; the modified 
exponential operator behaves properly under the conformal group. One can exploit (and 
eliminate) this remaining freedom by imposing locality. 

However, there is a price to pay. Even ignoring questions concerning the convergence 
of the expansion, one may wonder whether the individual terms in this expansion are 
actually well defined in view of the appearance of arbitrary powers of : F : in it. Naively 
and from one's experience with ordinary conformal field theory, one would not anticipate 
difficulties, as : y : has dimension zero by construction, and therefore the product of 
two such operators should be non-singular at coincident points if there are no negative 
dimension operators in the theory. However, trouble is caused by the non-local dependence 
of : y : on ip"^, which will lead to non-integrable singularities inside the integrals defining 
:¥ :. To make these explicit, consider the product : Y{T,ai) :: Y{t,o'2) :. Making use of 
(3.1.11) and ( 3.1.12| ) to rewrite this as a fully normal ordered expression, we get additional 



short distance factors inside the integral. Concentrating on the "critical region", where 
the integration variables coincide, we can approximate the integral by 

= I ^[ I < J /<'(■■■) (g _ g _ (g, _ g _ ■ (3.2.1T) 

where we have only exhibited the singular factor, and the dots stand for the harmless 
(and normal ordered) part of the integrand. We repeat that we could have avoided this 
problem by defining the exponential operator in terms of fully normal ordered products 
right away, but at the expense of spoiling the conformal properties, since : (y(r, cr))" : is 
not a conformal field in the sense of ( p.l.l6|) . Remarkably, the singularities thus occur only 
in the integration variables, whereas the dependence of the integrand on the "external" 
variables = t + ai and = r + o"2 is completely regular. Performing two of the 
(indefinite) integrals, we end up with 

/ < -^2- ^e)'-'' J del iei - ei - ^ef-'' (...) (3.2.18) 

These integrals exist for (7 < 1; at (7 = 1, they are still well defined because of the ie 
prescription. This means that at low orders in the expansion, the singularities are still 
integrable for a suitable range of values of g\ yet, at higher orders, the number of poles 



22 



at coinciding arguments will increase faster (namely as n{n — l)/2 for (: Y :)") than the 
number of integrations (of which there are only n). Consequently, the singularities will 
eventually become non-integrable for sufficiently large n, no matter how we choose g. 
The problem disappears only at special values of the parameter A (namely when 2A is a 
negative integer), for which the series terminates after a finite number of terms. 

3.3 Locality Condition 

Having constructed the quantum Liouville operator, we now proceed to exploit the con- 
sequences of the locality requirement stated in (p.2.2|) . We will find two (mutually incom- 
patible) ways to satisfy it. The first severely restricts the allowed values of A, but works 
to all orders. The second solution, proposed in [0, is based on a modification of the 
exponential operator ( 3.2.15| ), which exploits (and eliminates) the remaining freedom of 



choosing the P dependence of the expansion coefficients, cf. ( |3.2.16| ). This proposal has 
the advantage that it salvages locality for arbitrary values of A, but has so far only been 
shown to work to cubic order in /i^ in fact, we will present the proof to lowest non- 

trivial order only. To simplify the notation, we will drop the normal ordering symbols, 
and set r = for convenience. 



For ease of comparison with |32|, we shall trade the field Y for another one, called S, 



which differs in that the zero mode dependence has been pulled out. So we write 



e 



A</3 _ „Xv>P 



$:C^(P,A)5'", (3.3.1) 



where we only indicate the dependence on the center of mass momentum and adopt the 
normalization 

Co(P,A) = l. (3.3.2) 
For convenience, we have also defined 

Q = -fVQ, P = \lvP, (3.3.3) 
in terms of which the canonical commutator reads 

[Q,P]=27Tzg. (3.3.4) 
The quantity S is related to Y and defined by 

8(0, a) = |da'da"e'5/2e™(-)e^('^'-'^")e«/2 

.g'?V'^{o-')g'?V'^{o-')g^'/'r{-o-")g'?V'a ^ (3.3.5) 



^See, however, the footnote in the introduction. 



23 



where 



(a: a\ a") 



e((T — a') — da — a") 



(3.3.6) 



It should be clear that this is essentially the same expression as the one that we derived 
for Y in section 2.1. The integrand has been written in a slightly different fashion by 
means of the above definition (|3.3.3|) and the identity 



Ea{o - a')Eiia{a - a") = exp (^P {e{a - a) - e{a - a")) 



(3.3.7) 



Evidently, the operators Y and S differ only in their dependence on the momentum P. 
The precise relation is 



Y 

C{P) 



C{P + tTTg) 
1 



S. 



(3.3.8) 



2 sinh(P) ' 

where the shift in the argument of C{P) comes from pulling the P-dependence out of the 
integral and the normal ordering symbols. Expanding both sides and comparing term by 
term, one arrives at the result 



C™(P,A) 



-l)™r(2A + m) 



C{P + iTxg) 



2m 



(3.3.9) 



ml r(2A) 8 

It will be important to note that this expression is invariant under the shift 

P — > P + iirn neZ. (3.3.10) 

For convenience, we here repeat the exchange relation ( ^.1.6 ) in a form suitable for 
the analysis of the locality condition. Not forgetting to take into account the rescaling 
( ^.2.3| ), we have 

I ]_ _ g«kl-o"2| \-2abg 



^ar]ipa (o-i)gb»?V'c (0-2) 



_^-iTTabge{ai-a-2) ^iabg{cri-a2) 
. Qbrjtjjt (ct2 ) ^arjipi (cri ) 
1 _ o«ki-'^2| \-2abg 



_ ^iTrabge{ai — (72) Q—iabg{(Ti —a2) 



(3.3.11) 



where ip"^ and ip"^ were already defined in (|3.1.4|) and (|3.1.5|) . The corresponding formulas 
for the zero modes read 



eaQ^(p) = /(P + za27r^7)e'^«, 
/(P)e'^'3 = e''^f{P-ia2TTg). 



(3.3.12) 



24 



This means 



e^^faiPV^^^hiP)^^ = ^^^h{P + ia47rg)e'"5e"^/„(P - ihAi^g^^ . (3.3.13) 

We are now ready to analyze the consequences of the locahty condition. For this 
purpose, we expand 



= for cr^p, (3.3.14) 



and require the result to vanish order by order. At lowest non-trivial order, we obtain 
the following condition (the 0(1) contribution is trivial), which must hold for all values 
of the arguments a and p 

= e^'"^(")e'^'"^(^)Ci(P,z/)5(p) 

_gZ.^V(p)eA'?V'WC^(p^ X)s{a) . (3.3.15) 

Since the non-zero mode contributions coincide in all four terms, we must only look at 
the zero mode contributions. It is now straightforward to show that this leads to the 
following conditions on the coefficients: 

= Ci(P,z/)e(^+^'^»)^('') 

+Ci{P - in2ug, A)e(-P+^'^5(i-2-))^(<^)+^'^2.5e(p) 

-Ci(P,A)e(^+''^^)^('^). (3.3.16) 

This equation must be satisfied in all the regions of (cr, p) space. From its definition, 9{a) 
is easily seen to take the values 







if 


a>a' , a" 


or a', a" > a 




\i 


if 


a" >a> a' 








if 


a' >a> a" 





Therefore, Qia') and Q{p) can take separately the values 0, ±2, except for the combination 
Q{a) — ±2, ^(p) = ^2. It is not difficult to see that many of the regions give identical 
conditions and there are only three independent equations (plus the corresponding ones 
in which A and v are interchanged): 

= Ci (P, v) -Ci{P- iTT2gX, u) 



25 



+C^{P-ni2gu, X)-Ci{P, A), 

+Ci(P - i7c2gu, A)e2"*2^3 - Ci{P, A) , 
= Ci(P,z/)e-2(^^+*"^) -Ci(P-z7r2^A,z/)e-2(^^+*"^(i-2^)) 
+Ci(P - Z7r2^z/, X)e-^^'^''3 - Ci(P, A) 



(3.3.17) 
(3.3.18) 
(3.3.19) 



By subtracting ( |3.3.18| ) and ( |3.3.19| ) from ( |3.3.17| ), we get 



-Ci(P - t7i2gX, u) (l - e2(^+-9(i-2A)) 
+Ci(P - 27r2^z/, A) (l - e^^'"^) = , 
Ci(P, z/) (l - e-2(^+''^9)) 

-Ci(P - t7T2gX, u) (l - e-2(^+-5(i-2A)) 
+Ci(P - «7r2^i^, A) (l - e~^"*^^) = 



(3.3.20) 



(3.3.21) 



Notice that in both of these equations a factor of the form 1 — e^'^'^^^s appears in the last 
term. Although the notations are somewhat different, effectively this factor was implicitly 
assumed to be non- vanishing in the existing analysis As we shall show, the case 

where this factor vanishes will yield a new solution which is neither smoothly connected 
nor relatively local to the solution obtained in [^. Thus we now solve ( pj.3.20D and 
( ^.3.211 ) for these two cases separately. Since our analysis will involve the consideration 
of special discrete values of A, we emphasize that, except for d = 1, 25, the equation for g 
admits two solutions. 



Case 1: Begin with the case where the factor above vanishes. We then have 



2vg = integer . 



(3.3.22) 



and from ( |3.3.20| ) and ( ^.3.21[ ) we easily obtain 

1 _ p2(P+i7rg) 

Ci{P -tn2gX,u) 
CiiP -in2gX,u) 



1 _ e2(P+i7r<;(l-2A)) 
1 _ Q-2(P+ing) 



1 _ g-2(P+j7rg(l-2A)) 

From the compatibility of these equations, we get 

2Xg = integer , 
C,{P-i7r2gX,u) = C,{P,u). 



Ci{P,u), 
Ci(P,z/) 



(3.3.23) 
(3.3.24) 



(3.3.25) 



26 



It is easy to see that the classical form of the coefficient ( ^.3.9| ) satisfies this requirement. 



Case 2: Now consider the case where 2vg ^ integer. After some calculation using (|3.3.20|) 
and (|3.3.21| ), we obtain 



^'^^'''^ ~ sinh(P + m(7)sinh(/' + m(7(l-2z/))' ^^'^-^^^ 

where a{v) is an unknown function only of v. It is determined by substituting (|3.3.26|) 
into (13.3.201 ). The result is 



a{v) = asm{2'Kug) {a = const.). (3.3.27) 

Thus we finally obtain 

ri ( t) \ ^ sin(27rz/(7) 

'''^ " sinh(P + iTTg) sinh(P + t7rg{l - 2z/)) ' (3.3.28) 

It can be checked by a rather tedious calculation that this form of the coefficient does 
satisfy the locality equations in all the regions without any additional conditions on the 
values of A and u. Apart from an inessential normalization, this is the form obtained 
in |32]. Clearly, this formula is very suggestive of a hidden quantum group structure of 



quantum Liouville theory, and this has led the authors of [32] to conjecture a formula for 
the quantum deformed Liouville operator to all orders^. 

Let us summarize the implications of the 0{fi^) analysis just performed : First, if 
2i'g and 2Xg are both not integers, then exp(z/(y9) and exp(A(/?) are mutually local with 
the choice of the coefficients given by Otto and Weigt. On the other hand, if 2i'g G Z, 
then exp{Xip) is mutually local with respect to exp(z/(y9) only if 2Xg G Z. This latter 
case is precisely the one which is relevant for the Liouville theory since the operator e"^ 
appearing in the equation of motion must certainly exist. One might wonder if Case 1 
is but a special case of Case 2. This is not so: In order for the solution for Case 2 to 



correctly reproduce the classical limit {i.e. g —> ), the constant a in ( 3.3.2S ) must be 
proportional to 1/ sin(7r5f), namely 

r{p,A ^ sin{27ii^g)/ sm{7rg) (o o oq\ 

oc ginh(P + ^7r^7)sinh(P + m(7(l-2z/))- ^^-^'^^^ 

( Strictly speaking, the factor sin(7r5f) can be replaced by any function going linearly to 

zero a.s g — > 0.) However, if we now take the limit u n/{2g) where n is a non-zero 

integer, then the coefficient Ci(P, z/) above vanishes. Therefore the cases 1 and 2 above 

are not smoothly connected. 

^ The quantum group structure of Liouville theory has also been extensively discussed in | |35| | , although 
from somewhat different point of view. 



27 



Having seen that we must take 2g( e Z for the Liouville theory to make sense, we will 
now show that the locahty condition is satisfied to all orders in /x^ by taking the classical 
form of the coefficients Cjn{P, A). 

Let us first consider the effect of commuting the free-field exponential exp(Xriijj) 
through various operators. When commuted through P, it produces the shift P — 
P + i7r2gX = P + m where n is an integer. Since the coefficient Cm{P, A) has the period- 
icity 

Cm{P + inn, A) = C^(P, A) , (3.3.30) 

it does not affect Cm{P)- Also the operator 

Ea(a - a')Ei/a{a - a") = exp (P(e(a - a') - e{a - a")) (3.3.31) 

is not affected, since e{a — a') — e{a — a") is always an even integer. 

One more change exp{Xr]ip) produces when commuted with the operator S{p) is a phase 
of the type exp(i(p' — p")a). Prom the exchange formula, we can check that while the non- 
zero mode exchange produces the phase exp(— i(p' — p")2Xg), it is precisely compensated 
by the opposite phase coming from the zero-mode exchange. Thus we conclude that, for 
the case under consideration, exp(A77'0) has no effect in the locality equation and hence 
can be ignored. It is therefore sufficient to prove 

Y,{CUP,X)S"\a)C,,{P,iy)S^{p) 

m,n 

-Cn{P, iy)S^{p)Cm{P, A)5"^((7)} = . (3.3.32) 

Since S"^{a) contains the zero mode part exp{mQ), the left hand side of the above equation 
becomes 

J2 {Cm{P, X)Cn{P + 2t7Tgm, z/)5'"(a)5"(p) 

m,n 

-Cn{P, u)C^{P + 2mgn, X)S^{p)S^{<t)} . (3.3.33) 
For each term in the sum, the exchange of 5'"(p) and S''"(cr) produces the phase 

5"(p)5™(cr) = 5'n(^)5«(p)e(2'riflm^e(pJ)-27ri5n^e((T,i)) _ (3.3.34) 

But since each sum in the exponent is an even integer, the phase factor is actually 1 for 
2g e Z. Also, in this case, we have C„(P -|- 2mgm,v) — Cn{P) etc.. This means that 
all the operators appearing in the locality equations commute and therefore the locality 
requirement is satisfied to all orders in p? with the classical expression for Cm{P, A). 



28 



3.4 Operator Equation of Motion 



The procedure by which we constructed the exponential operator (|3.2.16| ) rehed heavily 
on the existence of the associated free field ^p, which itself was built out of the solution 
of the classical Liouville equation in terms of the functions A, B. We can now ask under 
what circumstances this equation can remain valid in the quantized theory. The main 
question we have to address here is how to define the quantum Liouville field itself (and 
not its exponential). This is a rather subtle issue in view of the results of the preceding 



section. If the proposal of ||32[ could be shown to work to all orders, we could define the 
quantum field if by taking the derivative of : e'^'^ : with respect to A and putting A = 
afterwards. If, on the other hand, only the discrete and mutually local set of operators 
with 2g\ G Z is available, this idea does not work. In this case, we can define the Liouville 
field by its normal ordered expansion, i.e. 

:^:^#+^^^(:r:r. (3.4.1) 

Of course, we must keep in mind that this definition is afflicted with the same problems 
that we encountered in defining the exponential operator, especially since the expansion 
( p.4.1| ) contains infinitely many terms (cf. the discussion at the end of section 3.2); 



furthermore, with this definition the relation between : ip : and the exponential operator 
defined before is somewhat obscure. With this caveat, let us proceed nonetheless and 
analyze the quantum Liouville equation. To do so, we need some relations involving 
derivatives of : F :. From ( |2.2.14| ) and ( p.2.15| ), we easily derive 



d+Y{T,a) :=: [c{a) jj da" E^{a - a")e^^^^+''^e'^'^^-''"^^ :, (3.4.2) 



and 



: 5_r(r,a) :=: [c{a) jj da' E^{a - a')e'^^^^+"'^e^'^^-"^^ : . (3.4.3) 
(We put /i^ = 8 for simplicity.) For the double derivative, we obtain 

: 9+9_F(r, a) : =: e'^^^^^^^e'''^'^^") :=: e'^'^^"'") : . (3.4.4) 
A separate calculation shows that 

[:5±r(r,a):,:F(r,p):] = 0. (3.4.5) 



We can therefore differentiate ( p.4.1| ) term by term and move the derivatives to the right; 



the calculation is then almost identical with the classical case. Using the above relations. 



29 



we thus get 

d+d-ip = 2^(-ir (: F :)"-^ e'?'^ : + 

n>l 

+ 2 - 1) (: r : d+Y :: : . (3.4.6) 

n>l 

Now, it is a little exercise in re-normal ordering to show that 

: d+Y{T, a) : : d^Y{T, p) : = : (^c{a) J^^ da" {a - a")e^^("+'^)e'^"(^-'^")' 



: c a, 



J^^ da' E^{p - a')e'^^(-+-')e'^"("-^)) : (3.4.7) 



^//^l^-0+(T+(7')gV"(T-fT") 



= (-1)29 : Uaf r da' I ^ da" E^{p - a')E.{a - a"Y 
^ Jo Jo " 

. g»?'/'+('7)g'?V'"{o-') . _ (3.4.8) 

Putting a = p, we get 

: d+Y{T, a) : : d.Y{T, a) := (-1)^' : Y{r, a) :: e''^^"'") : . (3.4.9) 



Inserting this into ( |3.4.6| ), we see that for g G Z, the right hand side of ( |3.4.6| ) can be 
simplified to 

2 ^(-l)"n (: Y :)""^ : e^'^ := - : e^ : , (3.4.10) 

n>l 

which is the desired result. We conclude that under the assumptions made above the 
quantum Liouville equation is satisfied only for integer values of g, and, in particular, 
for d = 1. For the d < 1 models, g is not an integer, as is obvious from ( p.2.8 ). In the 



foregoing section, we demonstrated that locality holds to all orders if 2gX is an integer. 
For non-integer ^f, A is not integer either; but then, : e''' : cannot be a local operator. This 
seems to indicate that the quantum Liouville equation is not consistent with locality for 
d < 1\ However, we repeat that this conclusion is subject to the caveats mentioned above, 
and therefore does not necessarily imply any inconsistency of the d < 1 models. 

4 Ground Ring 

As an application of the formalism developed in the preceding sections, we shall make an 
attempt to construct the generators of the so called "ground ring" which was found 



to play an important role in characterizing the symmetry structure of the d = 1 string 
theory. Our prime concern here is to see how its structure changes when the dependence 
on the cosmological constant is fully taken into account in the operator formalism. 



30 



4.1 Euclidean Case with fi^ = 

To begin with, let us briefly recall the Euclidean case with /i^ set to zero, the case first 
discussed by Witten [|^. When the Liouville field is regarded as a free field, one can 
consider the ring of BRST invariant operators with vanishing ghost number for the holo- 
morphic and the anti-holomorphic sectors separately. Each of them is hence termed "a 
chiral ground ring" . In the holomorphic sector, the chiral ring was shown to be generated 
by the following two operators, called x and y : 

X = (^cb + -^{dX - td(f))^ ■ e^(^+^^)/v^ , (4.1.1) 

y = (^cb- -^{dX + id(f))^ ■ e-*(^-^<^)/v^ , (4.1.2) 

where b{z) and c{z) are the ghost fields, while (f){z) and X{z) are respectively the Liouville 
and the matter field with the free-field operator products 

(f){z)(j){w) ~ X{z)X{w) ^ -\n{z -w) . (4.1.3) 

These operators correspond to special discrete physical states and are invariant with 
respect to the BRST operator Qe given by 

Qe = C~nL^ - ^ ~ ^) ■ C-mC-nbm+n -inv , (4-1-4) 

where the subscript inv signifies sl{2) invariant normal ordering. The Virasoro generator 
is of the form 

= L^ + Lt, (4.1.5) 

Ln = ^E:«n-m«^:, (4.1.6) 

Lt = ^Y.--^t-m(yt--+iq{n + l)at, (4.1.7) 

where the background charge q takes the value ^/2. The anti-holomorphic chiral ground 
ring is generated by similar operators called x and y. 

In the following, we shall concentrate on the generator x and for ease of notation 
express it as 

X = jX^o + dX^o- Xd^o, (4.1.8) 

J = cb, (4.1.9) 

X = e'75^, (4.1.10) 

$0 = e"^*. (4.1.11) 

The subscript "0" emphasizes that the Liouville field is treated as a free field and the 
usual normal ordering for composite operators is understood. 



31 



4.2 Minkowski Case with //^ = 

Before tackling the fully interacting case, wc need to clarify in some detail how the free 
field case should be treated in Minkowski formulation. Many of the calculations to be 
perfomed in this subsection will be utilized in the interacting case. 

To facilitate the comparison with the Euclidean case, we shall use the canonically 
normalized Liouville field (f>{x~^) which, in terms of the field ip{x~^), is given by 

= ^^(^+)- (4.2.1) 

( To be precise, (j){x~^) is defined to be that part of (f){x'^, x") which is independent of x". 
That is, it includes the full Q-zeio mode of 0(a;"'",x~). ) Then the energy-momentum 
tensor for the Liouville sector is given by 

Ttix^) = l{d^<Pr - qdl<l> , (4.2.2) 

and the corresponding Virasoro generator is of the form (still with q = \/2) 

C'"^ = lj:--'^tmc^i--+^1^^t- (4.2.3) 

Notice that the term containing the background charge takes a slightly different form 
compared with the Euclidean case. It is straightforward to show that the exponential 
operator $o is primary with respect to this Virasoro generator provided that it is defined 
with symmetric normal ordering (see the appendix.) Specifically, 

Lf^$o(a:+)] = + (-^) n) $o(a;+) . (4.2.4) 

It is easy to check that L^-'^'s satisfy the Virasoro algebra of the form 



tM,4, tM,<I> 



(m - n)l!^^^ + (5^+n,o ( - m) + g^m j . (4.2.5) 



In order to construct the correct nilpotent BRST operator, we need Virasoro generators 
which satisfy the above algebra with the central term proportional to — m. As is 
well-known, such a standard form can be obtained by a shift 

IM,, ^ ^ + ig^5.,o = C'^ + 5n,0 . (4.2.6) 

If we denote the total Virasoro operator (including the matter part L^''^) by L^, the 
BRST operator takes the form 

Qm ^ Yl {^n + ^n,o) - ^ - n) : c_^c_„&^+„ -.inv , (4.2.7) 



32 



We are now ready to display the generator xm , corresponding to the operator x, for 
the Minkowski case. It takes the form 



3 --] X^o + -d+X% - 



(4.2.8) 



where the ghost current j is normal-ordered with respect to the sl{2) invariant vacuum as 
in the Euclidean case, and the operators $o and X are defined with the symmetric normal 
ordering. The shift of j by the amount 3/2 compared with (|4.1.8| ) can be understood as 
follows: The BRST operator Qm given in ( [4.2.7|) actually takes a better form if we make 
a change of normal-ordering from the s/(2) invariant one to the "physical" one defined by 
the rule 



-(co&o - boco) 



annihiliation for n > 1 , 
for zero mode . 



Qm then becomes 

Qm 



1 



(4.2.9) 
(4.2.10) 

(4.2.11) 



^—m^—n'-^m+n -phys • 

It is easily checked that under this normal ordering j becomes anti-hermitian and is related 
to the previous definition by 

friv _^ = f^y\ (4.2.12) 

BRST invariance of xm given above can be demonstrated as follows. First we write 
the commutator [Qm^xm] as 



[Qm, Xm] 



where 



A2 

A, 
A^ 



lim 



Ai + ^2 + A3 - -Ai 



QM,3{.x-^)X^Q{y' 



Qm, -X-d+% 



(4.2.13) 

(4.2.14) 
(4.2.15) 

(4.2.16) 
(4.2.17) 



y42, ^3 and ^4 are easy to evaluate using the fact that X and $0 are primary with respect 
to the Virasoro operators inside Qm- The result is 

5 



A2 

^3 

-'-A. 
2 



cd+ {xd+%) - 

-cd+ {d+X^^) 



-dlcx% 
1 



-dlcX^,, 

-h-d+{X%) + l-d^cX%. 
2 I 2 I 



(4.2.18) 
(4.2.19) 
(4.2.20) 



33 



Therefore, 



^2 + ^3 - 



c (dlX<l>o - Xdl<l>o) - -dlcX<l>o 



(4.2.21) 



To compute Ai, we will need the following formulae which can be obtained by straight- 
forward calculations: 



-c (Tm(x+) + l) - -cd^cb + ^dlc - |-a+c. 



i i 



(4.2.22) 
(4.2.23) 



2 \V2 



z — w 



w 



1\ 1 



z — w 2ji 



d+X, (4.2.24) 
(4.2.25) 



zw 



A{z-w)' 



z — w 2ji 



16 



$0+ : T^/(a;+)$o : 



j(x+)c(y+) = c(x+) + 2] + : 3{x+)c{y+) : 

j(x+)-9+c(y+) = c(x+)f(^^^)" 

2 \\Z — W J 



z — w 



(4.2.26) 
(4.2.27) 
(4.2.28) 



Using these formulae, it is easy to verify that all singular terms cancel, and we are left 
with 



3 1 



3 1 



(4.2.29) 



From ( [4.2.211) and ( [4.2.29|) it is evident that the sum Ai + A2 + A^ — \A^ indeed vanishes. 



4.3 Minkowski Case with /i V 

We now describe an attempt to construct the generators of the ground ring for the in- 
teracting case. As we switch on the cosmological term, left and right moving modes are 
coupled and we must necessarily consider the full ground ring. In the free field case its 



34 



generators are obtained as products of chiral generators, such as xmXm, xmVm, etc.. Since 
the interacting generators must reduce to these forms in the hmit /i^ 0, it is reasonable 
to take our candidates to be obtained from these products by appropriate replacements 
of free fields by corresponding interacting fields. More specifically, we shall consider in 
the following the operator ai which is obtained from xmXm by the replacements 

X{x-^)X{x-) — > X{x-^,x-), (4.3.1) 
d+X{x+)X{x-) — > d^X{x+,x-), (4.3.2) 
d+^o{x-^)d^^o{x-) — > a+9_$(x+, X-) , etc. (4.3.3) 

Since the commutation relations with the right and left BRST charges can be considered 
separately, we shall concentrate on the action of the left charge. Then we can effectively 
work with only the xm factor of ai ( with interacting fields ) and at the end of the 
calculation we can multiply by xm and make the replacements above. 

Wc note the remarkable fact that the weight in the exponent of $ is precisely such 
that the expansion in powers of breaks off after the first non-trivial order; namely, 
we have exactly $ = $o(l + ^)- This allows us to perform the calculations without any 
approximation. 

Let us now calculate the commutator [Qmi^m]- All the calculations which make use 
of the conformal properties of the fields X and $ proceed exactly as in the free case. The 
only place we must not use the free- field expressions is the computation of T'^{x'^)^{y^) 
and 9^$. The result using the conformal poperties is 

This is precisely the combination which vanished in the free field calculation. 
To simplify this expression it is useful to factorize $ as 

$ = $o(l+>') , (4.3.5) 

where $o is as defined for free theory and symmetric normal ordering for $o and Y is 
implicit. Now we shall prove the simple yet non-trivial result 

= d+^Q{l + Y). (4.3.6) 

To prove this, we must show that ^od+Y vanishes. First from the definition of Y we 
easily get 

d^Y{x) = ^ /da''e'5/2C(P)e-^^('^-'^'%lii"^elT^"^"''e<3/2 
8 J 



35 



^^y/2<t>+{z)^y/2^t{z)QV2cl>-{z")^V2<t>-{z") _ (4.3.7) 

A better representation is obtained if we move the second e*^/^ factor through the parts 
involving a" . Then we can express dj^Y{x) in the form 

d+Y{x) = : e^^''^"^) : F^") , (4.3.8) 

(4.3.9) 



where 



y(-) = /f! /rf^"C'(P-Z7r)e-(^-^'^)^('^-'^")e(^-^)Ti'^^"" 



^gV^<^-(^-")eV^^.-(^-"). (4.3.10) 

With this expression, we can compute the operator product between $o and at 
different arguments. We get 

$o(x+)9+y(y) = : e-^'^''^"''^-^*"^""^ 



. : g-;^<A+(x+)+V20+(s/+)-^*-(x-) . Y^-\y) . (4.3.11) 



This is seen to vanish in the hmit — * and we have the announced resuh ( 4.3.6 ) 



With the use of various formulae developed for the free field case, we then get, after 
some calculations 

= :T'f'%: {1 + Y)+ : ^dl(P%: {l + Y) + d+%d+Y . (4.3.12) 

v2 

Now look at Tf.j{x~^)^{y) in [Qm,xm]- We first re-normal order Tfj and $0 and leave the 
factor 1 + Y untouched. The result is 

\z — w 2/1 
+ : -L9^0(y)$o(i/) : (1 + Y)+ : T'^{x^)%{y) : (1 + F) . 



Substituting ( [4.3. 12[ ) and ([4.3. 13| ) into ( [4.3.4[ ) we get a non-vanishing result of order /i 



,2. 



[Qm,xm] = cXd+%d+Y. (4.3.14) 



36 



Since this result is not as expected, let us make a few checks before contemplating 
upon its implications. One check of the correctness of the calculation above is to show 
that the right hand side of ( |4.3.14| ) 

A = cXd+^^d+Y (4.3.15) 
is Qm closed. We need the following formulae: 

[Qm,c{x+)} = c^9+c(x+) (4.3.16) 

[gM,$o] = c-a+$o - ^-5+c$o (4.3.17) 

[gM,9+$o] = d+C-d+% + C-dl% 

I I 

-^^-dlc%-]-d+cd+% (4.3.18) 

[Qm.Y] = c-^d+Y (4.3.19) 

[QM.d+Y] = d+c-d+Y + c-dlY (4.3.20) 

[Qm,X] = c-d+X + \-d+cX (4.3.21) 
I A I 

Using these expressions and (^od+Y = 0, we get 

{Qm,A} = c-d+cXd+%d+Y 

-\\c-d+cXd+^Qd+Y 
14 I 

--c-d+cXd+^od+Y 
A i 

+c-d+cXd+'i>od+Y 
= 0. (4.3.22) 

Thus indeed ( ^XT5| ) is BRST closed. 

We now provide a further argument of more general nature which supports the result 
above. This argument will lead to the conclusion that as long as we do not change the 
ghost structure of the operator xm it is not possible to construct a BRST invariant. 

It is well-known that the total zero- level Virasoro generator Lq°*, including the ghost 
part, can be written as Lq°* = {Bq, Q^j}. On the other hand we have, for any operator O 
with a definite global dimension, [Lq"*, O] = \d^O. Combining these relations, we get 



4°*, O 



-d+O 



[{&o,Q},0] 

{6o,[Q,O]} + {Q,[6o,0]} . (4.3.23) 



37 



Thus if O is QM-closed, then 



= {g,[&o,C]}- (4.3.24) 



This equation states that dj^O is necessarily BRST exact and is determined solely by the 
part of O which contains the ghost c. 

Let us apply this logic to see if we can construct a BRST invariant operator of the 
form 

O — chX^ + terms not containing c . (4.3.25) 
For this class of operators, we have 

[60, O] = [bo, cbX^ = bX<^ . (4.3.26) 

Therefore 

-d+O = {Q,bX<^} 

= {Q, b}X^-b [g, X]^-bX [g, . (4.3.27) 

Using the various formulae developed previously for the calculation of xm, we can compute 
the right hand side. The result is 

d+O = 9+ f jX^ + -d+X^ - -Xd+^ - -X^ 
\ i i 2 

+-Xd+^od+Y . (4.3.28) 

( One can check explicitly that [Q, d+O] — 0.) As expected, the expression in the paren- 
thesis is precisely our candidate xm, but the additional term cannot be written as a total 
derivative. The closest we can get is 

-Xd+^od+Y = d+ ( - r dy+Xd+^od+Y{y)] . (4.3.29) 

I \l Ja J 

But the operator in parenthesis is not BRST invariant due to the lower limit of the 
integration. 

Thus we have shown that as long as we keep intact the structure of the term involving 
the c-ghost, we cannot construct a BRST invariant in terms of the the fully interacting 
field in analogy to the free field case. The situation does not improve even when we take 
into account the right-moving sector. 

This result would mean that the structure of the ground ring remains identical to 
that of the theory without interactions. Of course, one might question our use of the free 



38 



field form of the BRST operator in reacliing tliis conclusion. However, the only part of 
this operator that could be plausibly affected and altered by the interactions is the term 
with the Liouville energy momentum tensor. We are hesitant about this possibility as it 
is difficult to see how the construction can be modified without upsetting all our results 
up to this point. After all, it is the free field form of the Liouville energy momentum 
tensor that we have been using in our construction of the exponential operator in section 
3.2 (since we did not even know how to define the energy momentum tensor otherwise!). 
Despite the fact that we have been using the free field form of the BRST operator, our 
conclusion is not as trivial as it may seem. To underline this point, we note that, as far 
as the conformal properties are concerned, we do have operators such as e"^ which exhibit 
the same conformal properties as the corresponding free-field operator. It should also be 
noted that, if we express the generators of the ring in terms of the fully interacting fields, 
their forms certainly change into complicated expressions involving /x^. Since the result of 
the matrix model, which incorporates the full interaction, indicates the existence of the 
ground ring of the free-field structure, this interpretation seems to be the correct one. The 
point is that the free-field is not the Liouville field, but it is a complicated combination 
of the Liouville field. 



5 Discussion 



Clearly, difficult problems remain. The rigorous construction of the exponential Liouville 
operator still has not been accomplished. Apart from the technical difficulties discussed 
at the end of section 3.2, it is far from clear whether a perturbative expansion in the 
cosmological constant /x^ makes sense at all. As long as these problems are not solved, 
there is little point in addressing other issues, such as the question of whether the Liouville 
energy momentum tensor can be consistently defined in terms of the interacting Liouville 
field and shown to coincide with the free field energy momentum tensor, or whether the 
transformation between the Liouville field and the free field ip is canonical also at the 
quantum level. The hidden quantum group structure revealed in the construction of 
|3^ is certainly intriguing, but the existence of two mutually exclusive solutions to the 



locality condition remains an unsatisfactory feature. Perhaps the fact that there exist two 
solutions for the constant g (except for d = 1 and d = 25) plays a role in resolving this 
issue. In any case, we find it remarkable that the weights which appear in the ground 
ring operators are precisely in agreement with our discretization condition (3.3.25) and 
such that the expansion (3.2.15) has only finitely many terms. 

A possible way out may be to shelve these questions for the time being and to pursue 



39 



the study of the operator formahsm for Liouville theory along the hnes advocated by 
Gervais and collaborators [^,[^|. Rather than insisting on a rigorous construction of 



the Liouville exponential operator, these authors emphasize the importance of the hidden 
quantum group structure. This point of view is supported by the fact that many results 
can be deduced by requiring covariance with respect to the hidden SL{2)g, i.e. without 
invoking the explicit form of the Liouville operator, and are found to agree with the results 
obtained by other methods. A detailed comparison between this approach and the results 
obtained in this paper certainly merits further investigation. 



Acknowledgment 

We would like to acknowledge stimulating and helpful discussions with J.L. Gervais and 
G. Weigt. H.N. would also like to thank J. Schnittger for explaining his as yet unpublished 
results. 

Y.K. is grateful to the generous support and the hospitality of the members of the theory 
group of the Universitat Hamburg and DESY, where a part of this work was performed. 
The research of Y.K. is supported in part by the grant in aid from the Japanese Ministry 
of Education, No.04245208 and No.04640283. 



Appendix Euclidean-Minkowski Conversion 

In this appendix, we summarize how Euclidean and Minkowski formulations of conformal 
field theories are related. When there is a background charge, transcription is somewhat 
non-trivial. We shall mostly deal with the chiral sector. 

We begin with the Euclidean case. With z the "plane" coordinate, the energy- 
momentum tensor is expanded as 

n 

where L„'s are assumed to satisfy the "standard form" of the Virasoro algebra 

[Lm,Ln] = {m-n)Lm+n + —{m^ -m)5m+n,o- (A. 2) 

A primary field (j){z) of dimension A is characterized by 

<P{z) = ^0,;^-"-^, (A.3) 

n 

[LnAiz)] = z''(z^ + {n + l)A]<l){z). (A.4) 



40 



The "cylinder" coordinates 2;, which will be directly related the Minkowski light-cone 
coordinates, are defined by the conformal tranformation 



z 
z 



z = T + ia , 
z = T — ia . 



Then the energy-momentum tensor is transformed into 

2 



Tiz) 



dz 
dz . 



Tiz) 



12 



{z,z} 



where {z, z} is the Schwarzian derivative 



{z' = dz/dz, 



(A.5) 
(A.6) 



(A.7) 



(A.8) 



etc. 



For the transformation (|A.5| ) above, the Schwarzian derivative is —1/2. Thus we get 



n~z) 

The generators satisfy the algebra 



z'T{z) 



c 

24 



(A.9) 
(A. 10) 



TZ TZ 



c 

12 



that is, without a central term linear in m. 
The primary field (f){z) is transformed into 

A 



dz 
dz 



A simple calculation leads to 



, dz J 



(A.ll) 



(A.12) 



Note that the factor in front of A is changed from n + 1 to n. 

Minkowski formulation is obtained from Euclidean cylinder formulation by the replace- 
ment r — > ir, which converts the cylinder coordinates into the light-cone coordinates: 



z = Te + tcr 
z = te — icr 

izd 

izd 



i{rM - cr) 

04 



IX 

ix~ 



9_ = 



d 



dx~ 



(A.13) 
(A.14) 

(A.15) 
(A.16) 



41 



( Here and hereafter, d (d) means d/dz (d/dz).) Using these formulae, it is easy to 
obtain operator product expansions, such as (|3.1.19| ), from the corresponding ones in the 
Euchdean plane coordinate formulation. 

Before discussing the important case of a free boson with a background charge, let 
us briefly describe the notion of marginal deformation of Virasoro generators. Let L^'s 
satisfy the standard form of the Virasoro algebra ( |A.2| ). We also suppose that there exists 
a f/(l) type current with the mode oscillators am, which satisfy 



m5. 



+ [Aim + A2m'^)6m+n,o ■ 



(A.17) 
(A.18) 



Note that we allow for a term proportional to 5m+n,o in the second equation. This will 
occur later when we deal with a system with a background charge. 

With this setting one can deform the Virasoro generator in the following manner: 



Lrn = Lrn + (Bo + Bim)am + C5mfl 

Then a straightforward calculation shows 



(A.19) 



Thus if we take 



(m - n)Lm+n + ^2*-"^^ ~ m)5m+n,o 

-m'Bii2A2 + Bi)Sm+n,o + m{Bl + ^A^B^ - 2C)5, 



C = ^ + AiBo, 



if Si ^ 
if 5i = 



(A.20) 

(A.21) 
(A.22) 



Lm satisfy the standard form of the Virasoro algebra with the same central charge. 

Now let us consider a system of a free boson with a background charge q. In Euclidean 
plane coordinate formulation, (f){z, z) is expanded as 



.z,z) 



bo - iao \n{zz) + - 



(A.23) 



where the mode operators satisfy the usual commutation relations. The holomorphic 
energy-momentum tensor is given by 



T{z) 



Lf + iq{n + l)an , 



E -n~2 
n ^ 1 



(A.24) 
(A.25) 
(A.26) 



42 



satisfy the standard form of the Virasoro algebra with c = 1 + 12g^. 
It is important to note that (p transforms under conformal transformation with an 
additional inhomogeneous term proportional to the background charge q: 



{zd4>{z) + q{n + 1)) . 



(A.27) 



Now we make a conversion to the Minkowski formulation. Due to the the non-trivial 
tranformation property just mentioned, (f) undergoes the replacement 



[x'^,x ) — iq{x^ + X )—mq. 



(A.28) 



In terms of modes, this is equivalent to the shifts 



an 



b^'-tqTT, 



Uq — iq . 
an . 



(A.29) 



To convert the energy-momentum tensor, one must take into account this replacement in 
addition to the transformation ( [A.TI ). After a simple calculation we obtain 



Tix^ 



-M 



iqnar. 



- iqan 



(A.30) 

(A.31) 
(A.32) 



where the inessential overall additive constant —1/24 has been dropped. L^^ satisfies the 
Virasoro algebra of the form 



tM tM 
1 ^n 



(m - n)L^^^ + 6n,+nfl [—{m - m) + q m] , 



12 



(A.33) 



which is not of the standard form. But since the difference is linear in m, we can shift 
to cancel this term. It is easy to see that 

.2 



- iqan + -77^^,0 



2 



(A.34) 
(A.35) 



satisfies the standard form. Notice that 



—nam+n + iqm{m + 1)5„ 



(A.36) 



43 



This is of the form treated in the discussion of marginal deformation of the Virasoro 
algebra. Indeed we can identify 

Ai = iq, A2 = iq, 

Bo = -iq, 5i = 0, (A.37) 

^ = \' 

and easily check that these coefficients precisely satisfy the conditions for the marginal 
deformation. 

It is extremely useful to regard the shift ( |A.29| ) as a similarity tranformation lA defined 

by 

U = e-^'^Oe'?™°. (A.38) 

Then the conversion of (f) can be succinctly expressed as 

— > U(pU-K (A.39) 

Moreover, it is not difficult to check explicitly that this operation correctly converts the 
Virasoro generators. Namely, 

UL^U-' = , (A.40) 



where and are as defined in ( |A.25| ) and (|A.32|) . This makes obvious the previously 



mentioned fact that both and L^^ satisfy the standard form of the Virasoro algebra. 

As an application of the similarity transformation, let us give the conversion formula 
for the exponential operator e^*^ carrying the conformal weight A = — (A^/2) + Xq. If we 
denote by : :e and : :s the Euclidean and the symmetric normal orderings respectively, 
a simple calculation gives 

U : e^^(^'^") -.E ( 4^4^Yu~' = : e^<^ 15 e'^'^^'/^ _ ^^.41) 

Apart from a coordinate independent phase factor, the usual Euclidean normal ordering 
is precisely converted to the symmetric normal ordering This clearly shows the necessity 
of symmetric normal ordering adopted in the text. 

References 

[1] D.J. Gross and A.A. Migdal, Phys. Rev. Lett. 64, 717 (1990) 
M. Douglas and S. Shenker, Nucl. Phys. 335, 635 (1990) 
E. Brezin and V. Kazakov, Phys. Lett. 236B, 144 (1990) 



44 



[2] V. Kazakov, "Bosonic Strings and String Field Theories in One-Dimensional Target 
Space LPTENS 90/30 

I.R. Klebanov, "String Theory in Two Dimensions" Princeton Preprint PUPT-1271 
(1991) 

[3] A. Gupta, S.P. Trivedi and M.B. Wise Nucl. Phys. B340 (1990) 475 

[4] N. Seiberg, Prog. Theor. Phys. (Supplement) 102 (1990) 319 

[5] M. Bershadsky and I.R. Klebanov Phys. Rev. Lett. 65 (1990) 3088; Nucl. Phys. 
B360 (1991) 559 

[6] M. Gouhan and M. Li, Phys. Rev. Lett. 66 (1991) 2051 

[7] VI. L Dotsenko, Mod. Phys. Lett. A6 (1991) 3601 

[8] Y. Kitazawa, Phys. Lett. B265 (1991) 262 

[9] K. Aoki and E. D'Hoker, Mod. Phys. Lett. A7 (1992) 235 
[10] P. Di Francesco and D. Kutasov Nucl. Phys. B342 ((1990) 589 
[11] D.J. Gross, I.R. Klebanov and M.J. Newman, Nucl. Phys. B350 (1991) 621 
[12] D.J. Gross and I.R. Klebanov, Nucl. Phys. B352 (1991) 671 

[13] K. Demeterfi, A. Jevicki and J.P. Rodrigues, Nucl. Phys. 362 (1991) 173; Nucl. 
Phys. 365 (1991) 499 

[14] U.K. Danielsson and D.J. Gross, Nucl. Phys. B366 (1991) 3 

[15] A.M. Polyakov, Mod. Phys. Lett. A6 (1991) 635 

[16] B.H. Lian and G.J. Zuckerman, Phys. Lett. 266B, 21 (1991), Phys. Lett. 254B, 

417 (1991), Comm. Math. Phys. 145, 561 (1992) 
[17] P. Bouwknegt, J. McCarthy, and K. Pilch, Comm. Math. Phys. 145, 541 (1992) 
[18] J. Avan and A. Jevicki, Phys. Lett. B266 (1991) 35; Phys. Lett. B272 (1991) 17 
[19] M. Awada and S.J. Sin, Int. J. Mod. Phys. A7 (1992) 4791 

[20] A. Gerasimov, A. Marshakov, A. Mirnov, A. Morozov and A. Orlov, Nucl. Phys. 
357 (1991) 565 

[21] D. Minic, J. Polchinski and Z. Yang, Nucl. Phys. B369 (1992) 324 
[22] G. Moore and N. Seiberg, Int. J. Mod. Phys. A7 (1992) 2601 

[23] S. Das, A. Dhar, G. Mandal and S. Wadia, Mod. Phys. Lett. A7 (1992) 71; tbtd 

A7 (1992) 937 (Erratum A7 (1992) 2245; Int. J. Mod. Phys. A7 (1992) 5165 
[24] E. Witten, Nucl. Phys. B373 (1992) 187 

[25] I.R. Klebanov and A.M. Polyakov, Mod. Phys. Lett A6 (1991) 3273 
[26] E. Witten and B. Zwiebach, Nucl. Phys. B377 (1992) 55 



45 



[27] J.L. Gervais and A. Neveu, Nucl. Phys. B199 (1982) 59 ; B209 (1982) 125 ; B224 
(1983) 329 ; B238 (1984) 396 

[28] E. Braaten, T. Curtright and C. Thorn, Phys. Rev. Lett. 51 (1983) 201; Ann. 
Phys. (NY) 147 (1983) 365 

[29] E. Braaten, T. Curtright, G.Ghandour and C. Thorn, Phys. Rev. Lett. 51 (1983) 
19; Ann. Phys. (NY) 153 (1984) 147 

[30] L. Johansson and R. Marnehus, NucL Phys. B254 (1985) 201 

[31] E. D'Hoker and R. Jackiw Phys. Rev. D26 (1982) 3517 

[32] H.J. Otto and G. Weigt, Phys. Lett. 159B (1985) 341; Z. Phys. C31 (1986) 219 
[33] G. Weigt, "Critical Exponents of Conformal Fields Coupled to Two- Dimensional 

Quantum Gravity in the Conformal Gauge", Zeuthen Preprint PHE 90-15; 

"Operator Quantium Liouville Theory and Correlation Functions", Talk at Vth 

Regional Conference on Mathematical Physics, Dec. 1991 Edirne, Turkey; 

Phys. Lett. B277 (1992) 79 
[34] O. Babclon, Phys. Lett. B215 (1988) 529 

[35] J.-L. Gervais, Comm. Math. Phys. 130 (1990) 257; Comm. Math. Phys. 138 
(1991) 301; "Gravity-Matter Couphngs From Liouville Theory", Preprint LPTENS- 

91/22 (1992); 

E. Crcmmcr, J.L. Gervais and J.F. Rousscl, "The Quantum Group Structure Of 
2D Gravity And Minimal Models II: The Genus-Zero Chiral Bootstrap", Preprint 
LPTENS 93/02 (1993) 

[36] J.L. Gervais, Nucl. Phys. B (Proc. Suppl.) 25A (1992)122 

[37] E. D'Hoker, "Lecture Notes on 2-D Quantum Gravity and Liouville Theory", UCLA 
preprint UCLA/91/TEP/35 

[38] S. Kachru, Mod. Phys. Lett. A7 (1992) 1419, Nucl. Phys. B390 (1993) 173; 
M. Li, Nucl. Phys. B382 (1992) 242; 
J.F.L. Barbon, Int. J. Mod. Phys. A7 (1992) 7579 

[39] V. Dotsenko, Mod. Phys. Lett. A7 (1992) 2505 

[40] A.M. Polyakov, Phys. Lett. 103B (1981) 207 

[41] A.N. Leznov and M.V. Saveliev, Commun. Math. Phys. 74 (1980) 111; Phys. Lett. 
B79 (1978) 294; B83 (1979) 314; Lett. Math. Phys. 3 (1979) 207 ; 
D. Olive and N. Turok, Nucl. Phys. B265 (1986) 469; B220 (1983) 491 ; 
P. Mansfield, Nucl. Phys. B208 (1982) 277; 222 (1983) 419 

[42] J. Liouville, J. Math. Pures. Appl. 18 (1853) 71 



46 



[43] V. Knizhnik, A. Polyakov and A.A. Zamolodchikov, Mod. Phys. Lett. A3 (1988) 
819 



47