hep-th/9311168

HD-THEP-93-43

POINCARE POLYNOMIALS AND LEVEL RANK

DUALITIES IN THE N = 2 COSET
arXiv:hep-th/9311168v1 29 Nov 1993

CONSTRUCTION 1

Christoph Schweigert 2
Institut für theoretische Physik,
Philosophenweg 16
69120 Heidelberg, Germany

Abstract
We review the coset construction of conformal field theories; the emphasis is on the
construction of the Hilbert spaces for these models, especially if fixed points occur. This
is applied to the N = 2 superconformal cosets constructed by Kazama and Suzuki. To
calculate heterotic string spectra we reformulate the Gepner construction in terms of
simple currents and introduce the so-called extended Poincaré polynomial. We finally
comment on the various equivalences arising between models of this class, which can be
expressed as level rank dualities.

1 Introduction
Conformal field theory in general and N = 2 superconformal field theories in particular have
provided a deeper understanding of many issues in both mathematics and physics.
In physics, string theory is still one of the most exiting paths towards a quantization of
gravity and also implications in the realm of statistical physics, e.g. for universality classes
of two-dimensional critical behaviour, have been established. Moreover, deep and beautiful
1
Invited talk given at the III. International Conference on Mathematical Physics, String Theory and Quan-
tum Gravity, Alushta, Ukraine, June 1993. To appear in Theor. Math. Phys.
2
supported by Studienstiftung des deutschen Volkes

1
relations to other types of quantum field theories, e.g. Toda theories, Chern-Simons theories
and integrable systems have been unraveled. Conformal field theories provide also an important
testing ground for the algebraic theory of superselection sectors.
Mathematics has largely benefitted from conformal field theory as well: it gave rise to a
new and rich interplay between various branches: algebraic and differential geometry, number
theory, infinite dimensional Lie algebras, the theory of C ∗ algebras, commutative algebra, just
to enumerate some of them.
The interest in N = 2 superconformal theories was initially motivated in string theory by
the fact that – together with charge quantization – N = 2 superconformal symmetry on the
world sheet yields a space time supersymmetric string spectrum, but nowadays an indepen-
dent motivation to study these models comes from their beautiful intrinsic structure and their
deep connection to other objects in mathematical physics, e.g. to two-dimensional topological
quantum field theories and possibly even to general conformal field theories.
Among the various ways to construct these models – non-linear sigma models with the target
space a Calabi--Yau manifold, (infra-red) fixed points of the renormalization group flow on
Landau--Ginzburg potentials etc. – exactly solvable models are distinguished by two important
properties. Not only can one calculate in these models – at least in principle – exactly, i.e. non-
perturbatively correlation functions, but, what is even more fundamental, the full field content
of these models is known. This allows for an explicit calculation of the behaviour under modular
transformations, giving complete information on quantum dimensions and fusion rules. So it
is in these models that one hopes to really identify the fundamental symmetries of quantum
physics. Our main interest in this talk will be in coset theories, where the mathematical
framework is the theory of affine Lie algebras, but, of course, there are also other completely
solvable formulations, e.g. the Coulomb gas approach.

2 The coset construction

Starting from a complex, affine (untwisted) Kac-Moody algebra at level k :

[Jna , Jm
b
] = fcab Jn+m
c
+ km δm+n,0 κab (2.1)

we obtain by the Sugawara construction

1 X a b
Ln := ∨ κab : Jm+n J−m : (2.2)
2k + 2g m,a,b

the Virasoro algebra:

c 3
[Ln , Lm ] = (n − m)Ln+m + (n − n)δn+m,0 (2.3)
12
with central charge c = kk+g
dim g
∨ . Many quantities of interest of a WZW theory can be entirely

described in terms of the simple Lie algebra g that is generated by the zero mode currents. So
in the equations above κ denotes the Killing form of g and g ∨ the dual Coxeter number; colons
indicate normal ordering. Imposing unitarity, which is a natural requirement in string theory,

2
but not necessarily in statistical mechanics, we end up with a rational WZW theory at integer
level k.
The Virasoro central charges obtained this way always obey c ≥ rank g, with equality
exactly for simply-laced Lie algebras at level k = 1; so the various minimal series cannot be
obtained this way. This was one motivation for the coset construction [1], which turned out to
be a powerful tool to construct new conformal field theories.
Consider a subalgebra h embedded in g. It is easy to show that

Lg/h
n := Lgn − Lhn (2.4)

generates a Virasoro algebra with central charge cg/h = cg − ch .

While on the level of the symmetry algebra the construction is straightforward, we have to
emphasize the fact that we are still far away from having constructed a conformal field theory.
The question we have to worry about is: what is the Hilbert space of the theory? As it turns
out, this is a highly non-trivial question and at present no completely satisfactory answer to it
is known.
A first guess is to decompose any irreducible unitary representation HgΛ of g in irreducible
representations Hhλ of the subalgebra h:
M
HgΛ = HΛλ ⊗ Hhλ , (2.5)
λ

what leads to the following relation between the characters:

X
χgΛ (τ ) = bΛλ (τ )χhλ (τ ). (2.6)
λ

The Hilbert space of the coset is then supposed to be all HΛλ , its characters are the branching
functions bΛλ .
But right here we start running into problems. To see what happens let us have a look at
the critical Ising model, i.e. the minimal model with c = 21 , which has three primary fields: the
vacuum with conformal weight ∆ = 0, the twist field σ with ∆ = 1/16 and the energy operator
ǫ with ∆ = 12 . It can be described by the coset:

(A1 )1 ⊕ (A1 )1
. (2.7)
(A1 )2
Here subscripts are used to indicate the levels of the various algebras. A field is labeled by
three quantum numbers: Φpq r , where unitarity restricts p and q to be 0 or 1, and 0 ≤ r ≤ 2.
We immediately see that group theoretical selection rules for the coupling of the two A1
representations force all fields with p + q − r 6= 0 mod 2 to vanish. A second glance at the
branching functions of the coset (2.7) reveals that, apparently, each field occurs twice in the
spectrum. The vacuum, e.g. is represented by Φ00 11 01 10 11 00
0 or Φ2 , σ by Φ1 or Φ1 , and ǫ by Φ0 or Φ2 .
A more careful analysis reveals that moreover modular invariance is spoiled.
Here a new requirement we impose on a conformal field theory enters: modular invariance.
In string theory, in a string loop expansion à la Polyakov, modular invariance is used to factor

3
out the ‘large’ diffeomorphisms in the diffeomorphism group. In fact, we require even more
than just modular invariance of the partition function; we also require that the characters form
an unitary representation of the modular group.
The situation encountered in the coset representation of the critical Ising model generalizes:
some branching functions vanish while other non-vanishing branching functions turn out to be
identical. There are three possible reasons for a branching function to vanish: group theoretical
selection rules like in the case of the Ising model, the occurrence of null states 3 and a slightly
more complicated combination argument [3], but for which no example is known. On the other
hand, in general, S-matrix elements between vanishing and non-vanishing branching functions
do not vanish, so it is impossible to simply delete the corresponding rows and lines in the
S-matrix without spoiling unitarity of the S-matrix.
Before explaining the way out of this situation, it is appropriate to recall some results from
the theory of simple currents (for a review see [3]). A simple current J is a primary field in the
theory for which the fusion product with any field of the theory yields just one other field:

J ⋆ Φ = Φ′ . (2.8)

We define the monodromy charge Q of Φ relative to J as Q = ∆(J) + ∆(Φ) − ∆(Φ′ ). In the

case of WZW theories Q can be shown to be simply the conjugacy class.
Simple currents give rise to modular invariants which can be different from the diagonal one.
Important for our purposes is the case of integer spin simple currents. In this case a non-trivial
modular invariant exists in which only fields with vanishing monodromy charge occur. It is
given by:
NX 2
X N0 a −1

Z= χJ i a (2.9)
a,Q(a)=0
Na i=0

Here Na is the length of the orbit of the simple current J, N0 is the length on the orbit which
contains the vacuum, i.e. N0 is the order of the simple current.
These invariants are suited to take care of group theoretical selection rules: non-vanishing
fields can be characterized by the fact that they have vanishing monodromy charge relative to
a subgroup of the group of simple currents of g tensored with (the complement of) h. 4
We find thus that the true fields of the theory are the orbits of the identification group; in
the literature this is referred to as ‘field identification’ [4, 5]. In fact, this seems to be a generic
feature of reduction procedures; e.g. in a system with first class constraints it is well known
that we have to factor out the action of a ‘gauge’ group, too, to obtain a consistent physical
system.
In our case problems arise if orbits of different lengths occur. Namely, we want to keep just
one representative of of every orbit to have a unique vacuum; in other words, we have to divide
Z by N02 . But this would lead to non-integer coefficients in the partition function of shorter
orbits, what is incompatible with the interpretation of Z as a partition function. The shorter
3
This is the case for conformal embeddings (which yield trivial cosets with c = 0) and also for the ‘maverick’
cosets described in [2].
4
The modular invariants of the maverick cosets are in general not simple current invariants.

4
orbits which are termed ‘fixed points’ require thus a special treatment, and here we clearly need
some additional input.
The idea suggested by the prefactors of the complete squares in (2.9) is that every fixed
point of length Nf < N0 has to be resolved in N0 /Nf distinct physical fields. This procedure
possibly introduces some arbitrariness, as S-matrix and characters for the individual physical
fields are a priori unknown.
For the S-matrix elements involving fixed points we make the following ansatz [3]:

Nf Ng
Sefi gj = Sf g + Γfijg , (2.10)
N0
where the indices i, j count resolved fields and S is the naive S-matrix. Modular invariance can
be shown to imply the following sum rules for the characters Xfi and the S-matrix elements of
fixed points:
X
Xfi = Xf (2.11)
i
X X
Γfijg = 0 = Γfijg (2.12)
i j

In most cases the Γ matrices and also the character modifications needed to fulfill (2.11) can
be described in terms of a different WZW theory, which is usually called ‘fixed point theory’.
A list of these fixed point theories can be found in [3].
We emphasize that only after having found a consistent resolution of the fixed points we
have really constructed a conformal field theory. Unfortunately, no general results concerning
existence or uniqueness of the resolution are known.

3 N = 2 coset theories
In the sequel we will focus our attention on a special subclass of cosets models, namely the
N = 2 coset models constructed by Kazama and Suzuki [6, 7]. As is well known, the N = 2
algebra is generated by the stress energy tensor T , two spin 3/2 supercurrents G± and one spin
1 u(1)-current J. With respect to J the supercurrents G± have charge ±1 . For the following
considerations it is convenient to define
1 1
G(1) := √ (G+ + G− ) G(2) := √ (G+ − G− ). (3.1)
2 i 2

Then T and G(1) or G(2) generate an N = 1 superconformal algebra.

We are now going to describe the construction proposed by Kazama and Suzuki [7]. Given
a reductive subalgebra h of a simple Lie algebra g, the embedding h ֒→ g induces also an
embedding (‘tangent space embedding’) of h ֒→ so(2d) , where 2d = dim g − dim h . We can
thus consider the diagonal embedding

h ֒→ gk ⊕ so(2d)1 , (3.2)

5
where subscripts again denote levels. The choice of this embedding can be motivated by a super-
symmetric extension of the coset construction using super Kac-Moody algebras. In particular,
the so(2d) part corresponds to bosonized free fermions, and it is clear that the corresponding
modular invariant should be chosen.
These models can be shown to have always N = 1 supersymmetry, where the supercurrent
is given by
2 i
G(1) = Gg/h = Gg − Gh = (κāb̄ : j ā Jbb̄ : − fāb̄c̄ : j ā j b̄ j c̄ :). (3.3)
k 3k
Here κ is the Killing form, f are the structure constants of g. The bar over the indices indicates
that the sum is only over elements in the orthogonal complement of h in g relative to the Killing
form. Jb are the purely bosonic currents and j ā the fermionic currents transforming in the vector
representation of so(2d).
One may now ask in which cases the symmetry algebra can be enlarged to an N = 2
algebra. To investigate this question we make the most general ansatz in terms of normal
ordered products of fields for a spin 3/2 current and plug it into the N = 2 algebra:
2 i
G(2) = (hāb̄ : j ā Jbb̄ : − Sāb̄c̄ : j ā j b̄ j c̄ :). (3.4)
k 3k

This leads to a system of algebraic equations for h and S. This system was analyzed
in [6] in the case of regular embeddings h ֒→ g, and for special embeddings in [8]. The resulting
classification can be summarized as follows: a subalgebra h of g yields an N = 2 supersymmetric
coset of the form (3.2) if and only if the Dynkin diagram of h can be obtained from the (non-
extended) Dynkin diagram of g by removing at least one node. The situation turns out to be
particularly simple if the node corresponds to a cominimal weight, i.e. a weight with Coxeter
label equal to one. One can easily show [9] that then the corresponding homogeneous space is
hermitian symmetric. In fact most of the examples considered in the literature belong to this
special subclass, in particular to the so-called projective cosets
SU(n+1)k ⊕ SO(2n)1
. (3.5)
SU(n)k+1 ⊕ U(1)

The N = 2 minimal models can be recovered from the projective cosets by setting n = 1.
At this point one may ask whether all rational N = 2 conformal field theories can be
represented in the form (3.2). We cannot answer this question. Actually one representation of
an N = 2 coset is known that is not of this form, namely at c = 1:
(A1 )2 ⊕ (A1 )2
(3.6)
(A1 )D
4

But, as this model is a minimal model, it can alternatively also be described in the form (3.2).
In [9] all tensor products of N = 2 cosets that have central charge c = 9 and are thus
suitable for the inner sector of a heterotic string compactification have been classified. To state
the result of our classification we count cosets; as different modular invariants of the same coset
yield different conformal field theories (and thus superstring vacua) the number of the latter

6
is certainly higher. Unfortunately, it is unknown, as a classification of modular invariants of
general WZW theories is still missing.
There are 168 tensor products of minimal models, 190 tensor products of projective cosets
and minimal models, 123 tensor products involving other hermitian symmetric cosets [4] and
198 tensor products involving at least one coset that is not a hermitian symmetric coset [9].
Here some trivial group theoretical identifications have already been taken into account, such as
C2 ≡ B2 . Using non-trivial relations, such as the level rank dualities discussed below (compare
section 6 and [10]) the number of distinct coset conformal field theories is reduced even further,
e.g. for those involving non-hermitian symmetric cosets from 198 to 112.

4 Explicit Calculations in N = 2 coset theories

As should be clear from the preceding discussion, an N = 2 coset has to be given a sense
as a conformal field theory, just like any other coset theory. Therefore first the u(1) part has
to be specified and the levels of the simple ideals hi of h. The latter can be shown to be
ki = Ii (k + g ∨) − hvi , where Ii is the index of the embedding hi ֒→ g.
Then group theoretical selection rules or, equivalently, the identification rules have to be
determined. In [9] a simple formula for the number of identification currents was derived which
is useful to check the completeness of a set of selection rules:
Y
|Gid | = Qi0 b ),
IC ( h (4.1)
i
i

where Qi0 is the u(1) charge of the missing simple root (whose node in the Dynkin diagram
has been deleted) and IC is the index of connection, which is equal to the number of conjugacy
classes.
Interesting quantities are the number N27 of massless generations and N27 of anti-generations
of the heterotic string compactification which has a c = 9 tensor product of coset theories in
its inner sector. For simplicity, we restrict ourselves to the diagonal modular invariant.
Several problems have to be overcome: firstly, in coset models, only the fractional part of
the conformal weight can be easily obtained. The integer part can be obtained in principle
from a character decomposition using the Weyl-Kac character formula, but this is extremely
tedious and in practice hardly feasible. A way out is to work with Ramond ground states which,
due to spectral flow, provide equivalent information. An index-like argument [11] shows that a
complete set of representatives ΦeΛxλ
for all Ramond ground states is given by the formula

e = w(Λ + ρ ) − ρ ,
λ (4.2)
g h

where w runs over all elements of the Weyl group of g such that λ e is a highest weight of
h, including an u(1)-weight. As this formula has been derived from the Weyl-Kac character
formula, it automatically takes care of null states. For arbitrary states null states are a severe
problem; this makes e.g. the determination of E6 singlets very cumbersome in practice. In
fact, their number has only been determined for tensor products of minimal models, where the

7
representation theory of the N = 2 algebra provides an independent powerful handle on null
states.
Also, in general, the superconformal charge q can be read off easily only mod 2; only for
Ramond ground states the following formula [12, 13, 9] holds:
d ξ0 Q
q= − l(w) − , (4.3)
2 k + g∨
where l(w) is the length of the Weyl group element given by (4) and ξ0 some factor of propor-
tionality. Using equation (4.3) it was shown [9] that for any known N = 2 coset theory the set
of Ramond ground states is symmetric under charge conjugation.
Once we know all Ramond ground states or, equivalently, due to chiral flow, the ring of chiral
primary fields, we can in principle implement the method of β vectors [14] . In practice, this
turns out to be rather inconvenient, but there is a different approach, which has the additional
benefit to provide also the very important insight that the spectra do not depend on the details
of the resolution procedure for the fixed points, contrary to what one might expect.
In [15] modular invariance was used to express the Euler number χ = 2(N27 − N27 ) in terms
of the Poincaré polynomial P :

1 MX −1
χ= P (e2πd(r,s)/M .) (4.4)
M r,s=0

Here M is the smallest common denominator of all u(1)-charges in the chiral ring and d(·, ·)
stands for the largest common divisor of two integers.
Fixed points of the identification group have to be carefully taken into account, of course.
However, one can show that any orbit containing at least one representative of (4) yields after
resolution exactly one Ramond ground state, no matter how long the orbit is. It is important
to realize that this result holds irrespective of the details of the resolution procedure and that,
as a consequence of (4.4), χ does not depend on these details either.

5 The extended Poincaré polynomial

The method of the extended Poincaré polynomial finally allows for a separate calculation of
N27 and N27 .
If we want to build out of a tensor product C9 of N = 2 superconformal field theories with
c = 9 a heterotic string theory, we have to perform several projections. We will sketch below
how this can be described in terms of simple currents and explain the resulting prescription
encoded in the ‘extended Poincaré polynomial’. In a second step we shall comment on the case
c = 3 + 6n . This case is of much practical interest as we have to resort to it in some cases to
remove ambiguities in the resolution of fixed points.
After splitting off the contribution of the bosonic space time coordinates and applying the
bosonic string map [16], we can describe the heterotic string in a conformal field theory language
as the tensor product
(D5 )1 ⊕ (E8 )1 ⊕ C9 . (5.1)

8
The first two factors will provide for the right movers the gauge multiplet, for the left movers,
they describe the contribution of the fermionic coordinate fields. As the only purpose of the E8
factor is to provide a phase in the S-matrix such that the fermions are correctly reproduced,
we will drop it in our discussion from now on.
First, to obtain supersymmetry on the world sheet, we have to align the boundary conditions
in the various theories such that all fields are either Ramond or Neveu--Schwarz . Therefore,
it is important to note that TF := Φ0v 0 is a spin 1/2 simple current of order 2 that is present
in any N = 2 coset theory. Its monodromy charge is 0 for fields in the Neveu--Schwarz sector
and 1/2 in the Ramond sector.
Alignment is thus equivalent to enlarging the chiral algebra by all bilinears TF (i)TF (j)
(which have conformal dimension 3). In the D5 part we set TF (0) := v, which is, just like any
other field of (D5 )1 , a simple current.
Space time supersymmetry requires the projection on even 5 values of the u(1) charges [14].
To implement this projection we note that the Ramond ground state R0 with highest u(1)
c
charge is always a simple current with conformal dimension h = 24 . One can show [9] that
0s
there is always one representative of R0 of the form Φ0Qs ; from this explicit form its monodromy
charge can be easily seen to be half of the superconformal charge. The desired projection it is
thus equivalent to including the integer spin simple current Stot := (s, R0 ) in the chiral algebra.
Here s is the spinor simple current of (D5 )1 . Stot has been termed spinor current in [4]. We will
see below that its presence in the chiral algebra assures the existence of a space time gravitino
in the corresponding heterotic string spectrum.
In a conformal field theory language a heterotic string theory thus amounts to a conformal
field theory (5.1) with the modular invariant generated by the integer spin simple current Stot
and all bilinear combinations TF (i)TF (j).
It is now easy to recover the massless spectrum of the heterotic string. To obtain the proper
interpretation we recall that in one chiral sector of the theory, e.g. for left movers, we have to
apply the bosonic string map: the D5 ⊕ E8 part is mapped on a so(2)1 theory by interchanging
vector and scalar and changing the sign of the spinor and conjugate spinor representation in
the partition function. This map preserves the modular transformation properties and allows
for a description of the fermionic coordinates of the string.
As we work in a purely bosonic description, fields are massless if ∆ = ∆ ¯ = 1. Let us first
explain how in this formulation the generic part of the string spectrum arises which provides
the supergauge- and supergravity-multiplets. Two fields that occur in any N = 2 theory in the
inner sector are the vacuum and the two Ramond ground states with highest and lowest u(1)-
charge. The massless right moving fields that are tensored with the vacuum of the inner sector
have ∆ = 1 and, due to the charge selection rule, q = 0, ±2. These conditions are fulfilled
for the currents of E8 ⊕ D5 and the transverse bosons. In the modular invariant described
above these fields are paired with the following left movers: (v, 0) what yields for the transverse
bosons the graviton (as well as an antisymmetric tensor and the dilaton as the trace) and for the
currents the gauge multiplets. Applying Stot in the left moving sector yields the superpartners
of the gauge bosons and the graviton.
5
Here we formulate the condition after applying the bosonic string map, what explains the difference to what
the reader might expect, namely projection on odd values [14].

9
 In the right moving sector, we also find in the complete square of the identity the fields Stot
and Stot † , as well as the 0 of D5 tensored with the u(1)-current of the N = 2 algebra. According
to the well known branching of the adjoint representation of E6 to the adjoint representation,
the spinor, conjugate spinor and scalar of D5 , these fields extend the gauge symmetry from
E8 ⊕ D5 to E8 ⊕ E6 . In particular cases, if more fields are present, one can even further extend
both the gauge symmetry for right movers and the supersymmetry for the left movers.
To explain how massless (anti-)generations transforming in the 27 resp. 27 representations
of E6 arise, we remark that massless states that are vectors of D5 have ∆ = 21 and q = ±1 in
C9 , i.e. they are (anti-)chiral fields. Acting twice with Stot † on the vector tensored with a chiral
primary field with q = 1 yields a spinor tensored with a Ramond ground state and in a second
step 0 tensored with an anti-chiral state with q = −2; these states combine in a 27 of E6 .
Starting with an anti-chiral field and applying Stot instead we obtain states transforming in a
27 of E6 . These states can be paired with spinors or conjugate spinors in the left moving sector;
together they give rise to the generations and anti-generations and their CP T conjugates.
To extract information on the spectra we introduce the following notation: denote by hp,q
the number of fields which are in both the left and the right moving part of C9 chiral primaries
and have superconformal charge p resp. q; p, q are integers smaller than d := c/3. These
numbers can be seen as analogues to the Hodge numbers of a Calabi-Yau threefold. In fact, we
find the usual symmetries: hp,q = hq,p , as we started from a left right symmetric invariant, and
hp,q = hd−q,d−p , due to the conjugation symmetry on the chiral ring. Note that if the vacuum is
not paired with any chiral primary field other than the unique chiral primary field with q = 3c ,
we have h0,1 = h0,2 = 0; in the corresponding heterotic string compactification neither gauge
symmetry nor space time supersymmetry is extended. As this is the most interesting case we
P
will restrict ourselves to it from now on. The Euler number is given as usual χ := (−1)p+q hp,q .
The discussion above shows that the number N27 of massless generations transforming in
the 27 representation of E6 is equal 6 to h1,1 , or equivalently to the number of fields in the
theory, which are in both sectors spinors of D5 tensored with a Ramond ground state with
superconformal charge − 21 . The massless anti-generations N27 transforming in the 27 of E6
can be correspondingly characterized by the fields which are spinors and Ramond ground state
with charge − 12 in one sector and conjugate spinors and Ramond ground state with charge + 21
in the other sector.
We are thus interested in the structure of the relevant simple current orbits. Let us first
look at the orbits of Stot : as we are only interested in the massless spectrum we start with an
arbitrary Ramond ground state (s, R(1) , . . .) . Suppose now that, on the orbit, we encounter
(Jv ǫ0 v, TFǫi Ri′ ), where ǫi is 0 or 1 . This state – which is massive unless all ǫi vanish – is paired
in the simple current invariant with the original state in the other sector of the theory. But
the chiral algebra contains also all bilinears of the form (Jv , TF (i)): we thus find within the
same complete square of the partition function the corresponding massless state, for which all
ǫi vanish, too. If the D5 part is a spinor this yields a generation; conjugate spinors correspond
to anti-generations.
6
Our notation is different from the one used for Calabi-Yau manifolds: there the superconformal charge
in both sectors is defined with a relative minus sign, so the number of generations corresponds to the Hodge
number h1,−1 = h1,2 of the manifold.

10
 The information on the orbit of Stot is very conveniently encoded in the extended Poincaré
polynomial [4]. To start with, we define it on each factor of the tensor product separately. As
any simple current has finite order, the orbit has some periodicity which we first factor out for
convenience: for any Ramond ground state R we define NR to be the smallest power of the
spinor current such that (S NR )R is equal to R or TF R. We define ǫ(R) to be +1 in the first
and −1 in the second case. The extended Poincaré polynomial is now defined as :
X tq X X
P(t, x) = N
[ xm − xn ]. (5.2)
R 1 − ǫ(R)x R m∈F+ n∈F−

The sum is over all Ramond ground states R, q is the superconformal charge of the chiral
primary field connected via spectral flow. The sets F± are defined by the prescription: m ∈ F+
iff (Stot )m R is a Ramond ground state and n ∈ F− iff (Stot )n R is TF applied to a Ramond ground
state; in particular all m, n are even. The extended Poincaré polynomial is not a polynomial
in the new variable x, but rather a series with periodic coefficients. We remark that we recover
the ordinary Poincaré polynomial as P(t, 0).
We obtain the extended Poincaré polynomial for a tensor product by the following multi-
plication: given the extended Poincaré polynomials Pi (t, xi ) of the factors, first perform the
ordinary product of polynomials and then delete all terms in which the powers of the xi do not
coincide. This procedure implements the simple observation that, in order to have a Ramond
ground state of the tensor product, we need Ramond ground states in each factor of the theory.
The statements about the corresponding string compactification can be rephrased in terms
of the extended Poincaré polynomial. First, note that our assumption that the symmetry is
not enlarged translates into the requirement that the polynomial in x multiplying t0 is equal to
P
1 + x2 . N27 and N27 can be read off from the polynomial in x multiplying t1 : let p(x) = am xm
denote one period of this series. As the action of any of the bilinears (v, TF (i)) and of (Stot )2
changes the conjugacy class in the D5 theory we find generations if am > 0 and m = 0 mod 4
or am < 0 and m = 2 mod 4; the other cases correspond to anti-generations. Put differently
P
we find N27 + N27 = |am | and N27 − N27 = p(i).
The formalism of the extended Poincaré polynomial allows for an easy calculation [4, 9] of
N27 and N27 even if fixed points are present. Namely, after writing down those parts of the
extended Poincaré polynomial which are not affected by fixed points and taking into account
some evident structure of the fixed points we are left with only a few candidates for the extended
Poincaré polynomials. As pointed out above, the ordinary Poincaré polynomial and hence the
Euler number do not depend on the resolution procedure. For all possible extended Poincaré
polynomials of a given theory we calculate χ from N27 and N27 . If we do not get the correct
Euler number (4.4), we can exclude the candidate. This works surprisingly well. But, as the
Hodge numbers turn out to be relatively robust against changes of the parameters left in the
extended Poincaré polynomial, it is important to have many tensor products in which a given
model appears in order to have enough consistency conditions to single out the true extended
Poincaré polynomial.
This is our main motivation to generalize this formalism to tensor products of cosets with
conformal charge c = 3 + 6n. Here in general, we have no string interpretation at hand, so we
can replace the D5 factor by some other Dd factor. However, we have to require that the current

11
(s, R0 ), with conformal weight d/8 + c/24 has integer spin. This fixes d to d = −2n − 1 mod 8.
(We recover the previous situation for d = 5, n = 1.) It is important to note that, as the
S matrices of Dd and Dd+4 coincide, the choice of d does not affect the fusion rules. (Note
however, that the T matrices coincide only for Dd and Dd+24 .)
We now implement analogous projections, i.e. take the simple current invariant induced by
all bilinears in the TF (i) and (s, R0 ), and obtain the extended Poincaré polynomial by exactly
the same prescription as in the c = 9 case. Again massless states that are spinors or conjugate
spinors in Dd are Ramond ground states of C3+6n . The charge selection rule implies that
states paired with spinors have superconformal charge q ≡ − 21 mod 2 and for conjugate spinors
+ 12 mod 2. The chiral primary fields connected via spectral flow have thus charge q ≡ n mod 2
for spinors resp. n + 1 for conjugate spinors. This shows that we can recover the Euler number
from the polynomials multiplying all odd powers of t in the extended Poincaré polynomial
and summing up all contributions. Comparing this result with the result of (4.4), which was
derived in [15] for all c = 3 + 6n, we obtain new consistency conditions on the coefficients in
the extended Poincaré polynomial that can arise in the resolution procedure. We remark that
P
in general we can only read off q hp,q (±1)p+q from the extended Poincaré polynomial; this is
sufficient to determine all Hodge numbers separately only for n ≤ 1.

6 Old and New Level Rank Dualities

We have already seen (3.6) that distinct Lie algebraic cosets can describe the same conformal
field theory. In fact, this phenomenon occurs rather frequently: in [10] the equivalence of four
series of N = 2 coset models was established. As these identities arise by exchanging (a simple
function of) the rank and the level of the algebras involved, they are commonly referred to as
‘level rank dualities’.
Based on well-known level rank dualities between WZW theories [17], which are however
not isomorphisms between these theories, we were able to set up an isomorphism between the
following N = 2 coset models (the notation is taken from [4] and [9]):
(B, 2n + 1, 2k + 1) ≡ (B, 2k + 1, 2n + 1) (6.1)
(B, 2n, 2k + 1) ≡ (B, 2k + 1, 2n)|D (6.2)
(BB, m + 2, 1) ≡ (CC, 2, 2m + 1) (6.3)
(CC, n, k) ≡ (CC, k + 1, n − 1). (6.4)
The subscript D in the last line indicates that one must take the D-type modular invariant for
these models rather than the diagonal one. The last two identities have been suggested in [7]
because of the corresponding symmetry of the central charges. They have also been observed
in the string spectra and in the extended Poincaré polynomials in [4] .
In [10] the identity of these models was established by the construction of one-to-one corre-
spondence T between all primary fields (including a careful treatment of the fixed points). As
an example, in the case of the last duality (6.4) which reads in the full notation
(Cn )k ⊕ (D2n−1 )1 (Ck+1 )n−1 ⊕ (D2k+1 )1
≡ , (6.5)
(Cn−1 )k+1 ⊕ u(1)2(k+n+1) (Ck )n ⊕ u(1)2(k+n+1)

12
T is constructed by relating (Cn )k and (Ck )n resp. (Cn−1 )k+1 and (Ck+1)n−1 by the level rank
duality of WZW theories [17] and by a prescription for the map on the u(1) and Dd parts of
the theory.
The mapping T can be shown to preserve both ring structures present in these models:
on the one hand, as T preserves the S-matrix – and thus the fusion ring – and respects the
conformal dimensions, it is an intertwiner for the whole modular group. On the other hand, the
chiral ring structure and the superconformal charges are respected, too. Moreover, reasoning
along the lines given in [18] it should not be too hard to show that also the branching functions
coincide.
The arguments given in [10] show that the two respective theories are exactly identical as
conformal field theories, and do not merely represent different points in the moduli space of
one theory. In fact, changing the moduli generically also changes the conformal dimensions and
even the fusion structure of the theory, as can be easily seen e.g. when looking at the situation
at c = 1. (An arbitrary marginal deformation of a rational conformal field theory does not even
lead to rational conformal field theory.)
We expect that the techniques used in the cases above also allow for an explanation of the
level rank duality in the A-series on the level of Hilbert spaces which was proven in [7] for the
symmetry algebras.

7 Conclusions
One may, of course, extend the analysis presented above and also include non diagonal modular
invariants. For many purposes a complete survey of the spectra occuring in this class of N = 2
models would be helpful. One might get a better feeling of whether the old suspicion is true that
all rational conformal field theories are related in some way to coset models, possibly including
additional orbifoldizations.
It is also important to obtain more information about the massless spectrum of these theo-
ries, e.g. the number of E6 singlets. One may also ask whether the models presented in this talk
admit a description as a Calabi-Yau manifold or via a Landau-Ginzburg potential. Finally it
would be interesting to get some insight in whether all coset models with N = 2 superconformal
symmetry admit a description in the form (3.2).

Acknowledgments: The results presented in this talk are mostly the outcome of a very
pleasant collaboration with Jürgen Fuchs. I would also like to thank M. Kreuzer, W. Lerche,
A.N. Schellekens and M.G. Schmidt for stimulating and helpful discussions. Finally I would
like to thank the organizers of the III Conference on Mathematical Physics for their efforts
and for having given me the possibility to present these results. Financial support from the
Studienstiftung des deutschen Volkes is gratefully acknowledged.

13
References
[1] P. Goddard, A. Kent, and D.I. Olive, Phys. Lett. B 152 (1985) 88
P. Goddard, A. Kent, and D.I. Olive, Commun. Math. Phys. 103 (1986) 105
[2] D.C. Dunbar and K.G. Joshi, Mod. Phys. Lett. A 8 (1993) 2803
[3] A.N. Schellekens and S. Yankielowicz, Int. J. Mod. Phys. A 5 (1990) 2903
[4] A.N. Schellekens, Nucl. Phys. B 366 (1991) 27
[5] D. Gepner, Phys. Lett. B 222 (1989) 207
[6] Y. Kazama and H. Suzuki, Phys. Lett. B 216 (1989) 112
[7] Y. Kazama and H. Suzuki, Nucl. Phys. B 321 (1989) 232
[8] C. Schweigert, Commun. Math. Phys. 149 (1992) 425
[9] J. Fuchs and C. Schweigert, preprint hep-th/9304133, Nucl. Phys. B, in press
[10] J. Fuchs and C. Schweigert, preprint hep-th/9307107, to appear in Ann. Phys.
[11] W. Lerche, C. Vafa, and N.P. Warner, Nucl. Phys. B 324 (1989) 427
[12] T. Eguchi, T. Kawai, S. Mizoguchi, and S.K. Yang, Rev. Math. Phys. 4 (1992) 329
[13] D. Gepner, preprint PUPT-1130, May 1989
[14] D. Gepner, Phys. Lett. B 199 (1987) 380
[15] E. Buturović, Nucl. Phys. B 352 (1991) 163
[16] F. Englert, H. Nicolai, and A.N. Schellekens, Nucl. Phys. B 274 (1986) 315
[17] E.J. Mlawer, S.G. Naculich, H.A. Riggs, and H.J. Schnitzer, Nucl. Phys. B 352 (1991) 863
[18] D. Altschüler, Nucl. Phys. B 313 (1989) 293

14