You are on page 1of 4

The Rogers-Ramanujan Identities: Lie Theoretic Interpretation and Proof

James Lepowsky, and Robert Lee Wilson


PNAS 1981;78;699-701
doi:10.1073/pnas.78.2.699
This information is current as of March 2007.

This article has been cited by other articles:


www.pnas.org#otherarticles
E-mail Alerts Receive free email alerts when new articles cite this article - sign up in the box at the top
right corner of the article or click here.
Rights & Permissions To reproduce this article in part (figures, tables) or in entirety, see:
www.pnas.org/misc/rightperm.shtml
Reprints To order reprints, see:
www.pnas.org/misc/reprints.shtml

Notes:
Proc. Nat. Acad. Sci. USA
Vol. 78, No. 2, pp. 699-701, February 1981
Mathematics

The Rogers-Ramanujan identities: Lie theoretic


interpretation and proof
(Euclidean Kac-Moody Lie algebras/standard modules/principally specialized characters/infinite-dimensional Heisenberg algebras)
JAMES LEPOWSKY AND ROBERT LEE WILSON
Department of Mathematics, Rutgers University New Brunswick, New Jersey 08903
Communicated by G. D. Mostow, November 17, 1980

ABSTRACT The two Rogers-Ramanujan identities, which ment of Kac-Moody Lie algebras, see refs. 12-14 and 8. Here
equate certain infinite products with infinite sums, are among the we use instead the basis and realization introduced in ref. 15.
most intriguing of the classical formal power series identities. It Let Z be the polynomial algebra C[xl, X3, X5, .. .] on the in-
has been found by Lepowsky and Milne that the product side of finitely many variables xi, X3, X5, ..., and let Z be the formal
each of them differs by a certain factor from the principally spe- power series algebra C[[xl, X3, X5, ...]]. Give Z the structure
cialized character of a certain standard module for the Euclidean
Kac-Moody Lie algebra Al(l'. On the other hand, the present au- of a graded algebra by setting deg xi = - i for i > 0, i odd. Then
thors have introduced an infinite-dimensional Heisenberg subal- Z = (jo Zj and Z' = llzo Z-,, where Ze denotes the fth ho-
gebra of Al(l' which leads to a construction of Al(') in terms of mogeneous subspace of Z. Let X denote the operator
differential operators given by the homogeneous components of
an "exponential generating function." In the present announce- -l/2 exp (I4xJi) exp (-Ed/ xi)
ment, we use 6 to formulate a natural "abstract Rogers-Ramanujan
identity" for an arbitrary standard A,(l'-module which turns out from Z to Z (where we identify a E Z with the operator which
to coincide with the classical identities in the cases of the two cor-
responding standard modules. The abstract identity equates two is multiplication by a, and where i ranges over all the positive
expressions, one a product and the other a sum, for the principally odd integers). Then X may be written as T~jEZ Xj, where Xj is
specialized character of the space tk of highest weight vectors or defined to be the jth homogeneous component (in the obvious
"vacuum states" for in the module. The construction of AlM sense) of X. For i > 0, i odd, define Bi = (i/2)a/d xi and B_
leads to a concrete realization of fi as the span of certain spaces = 2xi. Let z denote the identity operator. Then:
of symmetric polynomials occurring as the homogeneous compo-
nents of exponential generating functions. The summands in the PROPOSITION 1 (15). The operators Xj, j E Z; B. and B_j, i
Rogers-Ramanujan identities turn out to "count" the dimensions > 0, i odd; and z form a basis of a Lie subalgebra of End Z.
of these spaces. For general standard A (1)-modules, we conjec- Furthermore, [Bj, Be] = jbj _ez if j, e E 2Z + 1; [Bj, Xj] =
ture that the abstract identities agree witfh generalizations of the 2Xj +e if =j E 2Z + 1, e E Z;+[- Xe] = (-ljz if j + e = 0;
Rogers-Ramanujan identities due to Gordon, Andrews, and [Xi, Xe] 0 if j + e E 2Z, j 4?t 0; and [Xj, Xe] = 2(-1)
Bressoud. Bj+e if j + e E 2Z + 1.
It has recently been established (1-10) that many important We denote this Lie algebra by A bI(2)^ or 9.
power series identities arise as specializations of the denomi- The basis elements Bi, B ., and X. have operator degrees (on
nator and character formulas for Euclidean Kac-Moody Lie al- Z) equal to their subscripts, and z has operator degree 0. The
gebras. These formulas also lead to many interesting new identi- resulting gradation g = (03>z Ij is a Lie algebra gradation of
ties. In addition to the identities which have been proved by g called the principal gradation. A subspace of g is principally
using the theory of Kac-Moody Lie algebras, there are a num- graded if it is the direct sum of its intersections with the gj.
ber of other power series identities which are known to be re- Following ref. 15, we let B be the principally graded subal-
lated to Euclidean algebras but for which complete Lie theoretic gebra of g spanned by z and the Bi. Then e is an infinite-di-
interpretations and proofs have not been found. The best- mensional Heisenberg subalgebra of q called the principal
known examples are the Rogers-Ramanujan identities: Heisenbergsubalgebra.
Set b = go = Cz i CXO and define ho = 1/2z - XO, h, =
1 oc
q
n2
1/2z + Xo, a basis of f. Let hot h* E be the corresponding dual
V

H1 (1 - q5n 1I)(1 q~n-4)


- I: (1 q) (I q')
[1] basis. Let ko, k, E Z, ko, k, 2 0 and set A = koh* + klh*. The
standard a-module V with highest weight A (2) is the unique
...

n=l
ndI
and irreducible module generated by an element vo (called the high-
est weight vector, unique up to a nonzero scalar factor) satisfying
h-vo = A(h)vo for h E b and gj v0 = 0 for allj > 0.
TI (1- q5n-2)(1 - q5n-3) = + E - q)... ( - qn) [2] Following ref. 7, we define the level of V to be the non-neg-
n-Inl ative integer ko + kj. Since z is central in q and z = ho + hl,
z acts on V as multiplication by the level of V. Let V(e) (4 = 0,
(See ref. 11 for the history of these identities.) 1) be the standard module with highest weight he. The two
In order to explain the known connection between these standard modules V(o), V(l) of level 1 are called the fundamental
identities and Euclidean Lie algebras, we describe the Euclid- or basic modules and may be described as follows:
ean Lie algebra All, also called !t(2). For the definition of PROPOSITION 2 (15). V(e) (4 = 0, 1) is isomorphic to Z, with
Al(1) in terms of generators and relations and a general treat- 6 acting (irreducibly) as above, and with X acting as the jth
homogeneous component of
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore be hereby marked "advertise- l/2 (-1)e 1 exp (I 4x1/i) exp (- a/a xi)
ment" in accordance with 18 U. S. C. §1734 solely to indicate this fact.
699
700 Mathematics: Lepowsky and Wilson Proc. Vad Acad. Sci. USA 78 (1981)
i ranging through the positive odd integers.
A XI ... X; * Vov j 2 0, where each xe E and at most n of the xe
.

The principal gradation of induces in the natural way the lie outside B. Each V[.j is principally graded, since we may
gradation U(g) = EDj=1 U(j)j
of the universal enveloping algebra
of . Let V be a standard g-module, with highest weight vector
choose the xe in the definition to be elements of the homoge-
neous basis {Bk, Xi, z} of g. For n 2 -1, set Q[n] = a n v.
vo. The principal gradation of V is the decomposition V = Then each [nj is principally graded, and
@fEZ V= Ej20 Vx where V. = U(g)j'vo. For both fundamental
modules V, V Z for allj 0-i.e., the principal gradation 0= -ll C_ Rol C RI C.. C f
agrees with the gradation of Z.
Let q be an indeterminate, and consider the formal power is a filtration of Q such that fl = U fl[n. Hence:
series ring Z[[q]]. Let V be a standard module, and let W be n
PROPOSmION 5. We have
a principally graded subspace of V-i.e., W = ESj, W-, where
W,= wfn Vj. Following ref. 6, we define the principally
specialized character SEW) of W to be the power series X(Q) =
X(Z nj/1[n-11)
n-O

x(W) = E (dim Wj) qj EE Z[[q]]. We call this formula the abstract Rogers-Ramanujan identity
for V, because of the following main result, which gives our Lie
p20
theoretic interpretation and proof of [1] and [2]:
If U is a graded subspace of W, W/U has an obvious gradation, THEOREM 1. For V of level 3, the abstract Rogers-Ramanujan
and we extend the notion of principally specialized character identity coincides with the classical identity [1] or [2]. Specif-
to this situation. ically, let X = koh; + k1h*. If ko = 1, k, = 2, then X(Q) is the
In refs. 5 and 6, the principally specialized character A(V) is product side of [1] (see [4]) and for all n 2 0,
computed and is found to have an infinite product expansion:
PROPOSITION 3 (5, 6). Let V have highest weight X = kKh X(Q[n1/f[n-1]) = qn2/(1 - q) ... (1 - qn)
+ klh;, with ko,ki 1EZ, ko, k 2: 0. Let F =H, 1 (1 -q21 )
Then and if ko = 0, k1 = 3, then X(fQ) is the product side of [2] (see
[5]) and for all n > 0,
X(V) = F' H (1 qf)-l H (1 qfn)
n-1
-
n-1
-

n-O x(QIn]/Q[n-fl) = q"2n/(l - q) ... (1 - qfn).


HI (1 - q ) H (1 - qn), [3] Let V be arbitrary of positive level k and highest weight
n-1 n-1
A = Who + k1h . Assume that ko ' k1. We have:
n-ko+1 n-ki+1
CONJECTURE. Let j = 0 if k is even, j = 1 if k is odd. Let t
where all the congruences are mod (ko + k, + 2). This expression = [k/2] + 1. Then for all n . 0,
is symmetric in ko and kl. In particular, the level 3 modules have
the following principally specialized characters: If ko = 1, k, W(Q E Cnf (ko, k1) qX
Q[n- 1]) =
= 2, then too
X(V) = F.(IlnH l (1 - q5n-1)( -q5n-4))-l [4] where Cn]e (ko, k1) denotes the number of partitions d, + ...
+ dn of e such that 0 < di . di+,; di + 2 ' di+t-l; if diet+2
and if ko = 0, k, = 3, then 'd + 1 then di + ... +d+2 ko (mod 2 - j); and at most
X(V) = F'(HIn:.l (1 - q5n-2)( -q5n-3))-l [5] ko of the di = 1.
This conjecture generalizes Theorem I and, for ko / k1, it
Thus it is known that the product sides of the Rogers- asserts that the abstract Rogers-Ramanujan identity for V agrees
Ramanujan identities are obtained (by dividing by F) from for- with Bressoud's unified formulation (19, 20) of the generalized
mulas for the principally specialized characters of appropriate Rogers-Ramanujan identity (due to Gordon, Andrews, or Bres-
Al"l)-modules. A complete Lie theoretic interpretation of these soud) whose product side is the expression multiplying F in [3].
identities would require explanations (in terms of the structure (The conjectured identities for the cases ko = k, > 1 appear to
of the module V) of the factor F and of the sum sides of the be new; the identities for small values of ko and k, are trivial.)
identities. We announce such explanations in this note. The The computation of x(n01W/AIn-k1) in Theorem 1 is difficult
details will appear in Advances in Mathematics. and has several steps, which we now outline. Let V be as in the
Remark: For general ko and k, with ko 0 kj, [3] is F times conjecture. Define the completions of a, of U(g), and of V as
the product side of a generalized Rogers-Ramanujan identity follows: C(g) = 1jEZ&' C(U(g)) = Ijnz U(g)j, and C(V) =
due to Gordon (16), Andrews (11, 17, 18), or Bressoud (19, 20). HIjEZ V.. For x E C(U(g)) and v E C(V), we say that x v is de-
Denote by o+ the abelian subalgebra of spanned by the B3 fined if, for all t E Z, YjEz xj v, j is an absolutely convergent
(i > 0, i odd), and by p the subalgebra Cz ED !+. Let V have series in Vt, with respect to any norm on the finite-dimensional
positive level, and set (1 = {v EE VI¢'+v = 0}, the space of space V,. (The subscripts denote homogeneous components.)
"vacuum states" for in V. View fl as a p-submodule of V, and In this case, we take x-v to be the element of C(V) with 1xyvt
form the induced s-module U(Q) ®u(p,) Q. The map f: U(s) as its component of degree t. For ; E C:*, define X(C) =
OuM fl -* V sending u 09 w to u-w (u E U(!), w E fl) is a well-
defined B-module map. We have:
jeZ {iXJ E C(g) and
PROPOSITION 4. The map f is an u-module isomorphism, fl
is a principally graded subspace of V, and x(V) = F X(Q) E-(;) =
exp (I-2;i/kj)B_ E C(U(g)).
We define the !-filtration of V j odd
We have:
THEOREM 2. Fix n 2 0. Let O s m s n, (l..., C. E C* and
as follows: For all n 2 0, V[n] is the span of all the expressions ..ll>... > 1I.mI Then
Lepowsky and Wilson
Mathematics: Proc. Natl. Acad. Sci. USA 78 (1981) 701
E (E-(tj
( X.t@.
- )X(m-j)-... -VO of g isomorphic to g itself, and restricting V(1) to this subalgebra.
Theorem 1 is now proved by computing the dimensions of
is defined(vo being a highest weight vector of V), and fl[nl is the relevant concrete spaces of polynomials. We prove (equal)
the span of the homogeneous components of these elements of upper and lower bounds on these dimensions. The lower
C(V). bounds are obtained by reducing mod 2 and using the viewpoint
Recall the fundamental modules V(e), e = 0, 1 (Proposition of divided power algebras (cf. ref. 21).
2). By embedding V as the g-submodule of V(O)
... 01V(O) 0 V(1) ...09 V(1) (where there are ke factors We thank V. G. Kac for stimulating conversations. During this work,
V(e)) generated by 1 ...0 1, we can realize fln] concretely J.L. was partially supported by a Sloan Foundation Fellowship and
as a space of symmetric polynomials: Define sg(j) = 1 if 1 c
National Science Foundation Grants MCS76-10435 and MCS78-02439,
and R.L.W. was partially supported by National Science Foundation
j s ko, sg(j) = -1 if ko + 1 ' j ' k. Consider the polynomial Grant MCS77-03608 and a Faculty Academic Study Program Grant
algebra Cx[x. 1 ' j ' k; i > 0, i odd], with the grading given from Rutgers University.
by setting deg X, =-i. For in the set KN of functions from
{1, ..., n} into {1, k}, define sg(O) = HIe=1 sg[(fe)]. For 1. Macdonald, I. G. (1972) Invent. Math. 15, 91-143.
a fixed E KN, u, v E= {1, ..., n}, we say that u v if O(u) 2. Kac, V. G. (1974) Funkt. AnaL Ego Prilozheniya 8, 77-78. (English
= 4(v). For 1 j ' k, i > 0, i odd, define yj = 4xJ./i and zi translation: Functional Analysis and its Applications 8, 68-70).
3. Moody, R. V. (1975) Proc. Am. Math. Soc. 48, 43-52.
Y, - (1/k) (Y1, + + Yk,). Then {zj'|1 c j k - 1; i >
<
4. Lepowsky, J. (1978) Adv. Math. 27, 230-234.
0, i odd} is an algebraically independent set, but zjX =1 i = 0 5. Lepowsky, J. & Milne, S. (1978) Proc. Natl Acad. Sci. USA 75,
for all i > 0, i odd. Now define 578-579.
6. Lepowsky, J. & Milne, S. (1978) Adv. Math. 29, 15-59.
fn: ¢n C [[zj ill -- j c< k -1; i > 0, i odd]] 7. Feingold, A. & Lepowsky, J. (1978) Adv. Math. 29, 271-309.
8. Kac, V. G. (1978) Adv. Math. 30, 85-136.
by 9. Lepowsky, J. (1979) Ann. Sci. Ecole Norm. Sup. 12, 169-234.
In 10. Lepowsky, J. (1980) Adv. Math. 35, 179-194.
fn (vI,.v) sg(O) exp ( V4 Zoe) i) 11. Andrews, G. E. (1976) The Theory of Partitions, Encyclopedia
OEkkx e=1 i>O of Mathematics and its Applications, ed. Rota, G.-C. (Addison-
i odd Wesley, Reading, MA), Vol. 2.
12. Kac, V. G. (1968) Izv. Akad. Nauk SSSR Ser. Mat. 32, 1323-1367
(English translation: Mathematics of the USSR-Izvestija 2,
(Vu VV)2
- (V. + vV)2. 1271-1311).
U>t U>C
U-V U t 13. Moody, R. V. (1968) J. Algebra 10, 211-230.
14. Garland, H. & Lepowsky, J. (1976) Invent. Math. 34, 37-76.
Then: 15. Lepowsky, J. & Wilson, R. L. (1978) Comm. Math. Phys. 62,
THEOREM 3. Under the identification of V as a g-submodule 43-53.
of V(O) ... V(O) V(1) ... V(M), fl[n is the span of the 16. Gordon, B. (1961) Am. J. Math. 83, 393-399.
homogeneous components Of fm(V, ... ,vm)for 0 c m c n and 17. Andrews, G. E. (1967) J. Combin. Theory 2, 422-430.
18. Andrews, G. E. (1967)J. Combin. Theory 2, 431-436.
V1, ..., Vm E C.
19. Bressoud, D. (1979) J. Combin. Theory 27, 64-68.
For V of odd level, there is a second realization of Vas aspace 20. Bressoud, D. (1980) Mem. Am. Math. Soc. 24, No. 227.
of polynomials, obtained by viewing Eez §ke as a subalgebra 21. Block, R. & Wilson, R. L. (1975) Comm. Algebra 7, 571-589.

You might also like