APPLIED MAThEMATiC5

AND

COMFUTATIN
ELSEVIER Applied Mathematics and Computation 97 (1998) 17-22

On orbit sum values of elements of finite order

Abstract

The irreducible characters form a basis of the ring of class functions on a compact, simply connected, simple Lie group. Another basis are the orbit sums 4~ = ~ucw;. eU which are related to the irreducible characters by the inner multiplicities. There is a W-invariant sesquilinear form on the class functions - defined over a finite set of elements of finite order, under which the orbit sums are almost orthogonal. We give an exact criterion for the orthogonality given the set of elements of order dividing a fixed n and follow some remarkable properties of the values of orbit sums on elements of finite order (EFO). 1998 Elsevier Science Inc. All rights reserved.
A M S classification: 22E46; 22E30; 22E40; 22-04

Many problems of computing with representations of semisimple Lie groups [2] can be discretized in spite of the inherent continuous nature of the structures. A suitably chosen set of isolated elements of the group is often sufficient for a problem at hand. For example, some of the lowest tensor products of irreducible representations can be decomposed into the direct sum of irreducible ones using only the dimension - i.e. character values of the identity element of the group. A method generalizing this example, namely decomposition of class functions into finite sums of irreducible characters using the character values of a set of elements of finite order (EFO) was invented in [5]. However, there was no description how to find a suitable and efficient choice of EFO for a problem at hand. The goal of this note is to make that method more versatile by providing an explicit criterion for the choice of the EFOs which can split/decompose the characters appearing in the decompositions of a given class function.

1 Supported by the Schweineriacher Nationalfonds. E-mail: [email protected]. 0096-3003/98/$ - see front matter 1998 Elsevier Science Inc. All rights reserved. PII: S 0 0 9 6 - 3 0 0 3 ( 9 7 ) 10153-9

18

S. Grimm / Appl. Math. Comput. 97 (1998) 17-22

In the following let G be a simple, compact, simply connected Lie group of rank l. Fix a maximal torus T C G. Denote by ge the complexification of the Lie algebra g of G, and let it c gc be the Lie algebra of T. Denote by Q the root lattice of G spanned by the simple roots ~1,... , ~ , by P the weight lattice spanned by the fundamental weights ~ 1 , . . . , ~ and by P--+ the dominant integral weights. Then we have the exact sequence
exp 2rfi(.)
- -

(1)

where 0 c t is the coroot lattice. Note that the coroot lattice is spanned by the simple coroots ~1,... ,~t, which can be seen as the simple roots of the dual group G. We denote by d h , . . . ,d~l the fundamental coweights and by /3 the coweight lattice which they span. With the adjoint representation of G, Ad: G--~ Aut(G) we have /5 = ker(exp o Ad). (2)

Furthermore, by denoting the natural pairing t* t ~ C by (., .) we have (0~i, d)/) = 6ij and (COl,~i) = (~ij" By the exact sequence (1) the elements Tn := {x C T [ x~ = 1} of(full) order dividing n are identified with T. -~ (l/n){) {) , (3)

and by Eq. (2) the elements of adjoint order dividing n with A, := {x E T I Ad(x)" = id} "~ (l/n)/5^ .

(4)

Note that both sets are finite subgroups of T and they are stable under the Weyl group W. We consider the class functions CI(G) on G, i.e. f : G ~ C constant on conjugacy classes: for all x, g E G. As every conjugacy class intersects T (G being compact and connected), it is sufficient to look at the restriction of f to T. Furthermore, conjugation in G restricts to the action of the Weyl group W := on T. Thus to evaluate class functions on EFO in T, we only need to consider the conjugacy classes (respectively

f(gxg l) = f(x)

NG(T)/T

to~w).
Using (3) respectively (4) we can write

A,/W

An/W~--(!/5)/Wand Tn/W~_(!Q)/W,
where I~" is the affine Weyl group defined by W := Q w, and 0 is operating by translation. The fundamental domain of W in t is given by

S. Grimm / Appl. Math. Comput. 97 (1998) 17 22

19

F:={XEtl

(ai,X) ~>O,i= l...l,(~o,X)~< l},

where s0 = ~I=1 ni~v is the highest root of the root system of G. Thus we get

A,/W~_FNlb
n

and Tn/W~_FM1Q.
n

To any X = ( l / n ) ~ = , siGi E F A (1/n)P we set So := n - ~it=l nisi and no := 1. The n~ are called the numerical marks of G. If So,..., st have no common divisor the l + 1 tuple [So,..., st] are called the Kac coordinates of X (following [1,4]). The quantity ~ = 0 n~s~ is the adjoint order of the element X having Kac coordinates [s0,...,st]. We can therefore easily describe all elements of a given adjoint order. Note that, by abuse of notation, we call an element X E (l/n)/~ EFO, describing exp 2 n i x E T. By the Peter-Weyl theorem we know that the irreducible characters ;~, 2 E P++ of the simple finite dimensional G modules span a dense subspace of the class functions CI(G). Another set of functions spanning the same subspace of CI(G) are the orbit sums: ~b~ = E
/~EW2

e~"

Remark 1. The relation between the orbit sums and the irreducible characters is 2. given by the (inner) multiplicities m,. 7,x z E
/~EP+ + g/'/u (~#, x

where rot, is the dimension of the weight space VU := {v E V~[tv = #(t)v, Vt E T} in the irreducible representation V~ with highest weight 2. An efficient method to calculate these multiplicities is given in [3]. Usually e ~ is viewed as a formal exponential, denoting an element of Z[P]. Here we consider it as a function of T by e~(x) = exp 2~i(2,X), where X E t with x = exp2~iX. If, in addition, X has Kac coordinates [So,..., st] and we express 2 in the root basis as 2 = ~ i ai~i we get eX(x) = exp

2~i--~--"aisi ,

whereM = Enisi.

,=0

Definition 1. Let A be a finite, abelian subgroup of T. Define {., % : ZIP] Z[P] -~ C by

(fi ,f2)A := E f l (x)f2(x), xEA

20

S. Grimm I Appl. Math. Comput. 97 (1998) 17-22

The idea is to use this product to decompose class functions into linear combinations of characters respectively orbit sums. For this we need to be able to decide whether two functions take different values on a given set of EFOs. This leads us to the following definition. Definition 2. Let A be any subset of T, and S,,$2 two subsets of P. We say that A separates S1 from $2, if for each pair (21,22) E & $2 with 21 ).2 there is an x E A such that e< (x) e ~2(x); for f = ~ a a e ~" and g = ~ b ~ e ", A separates f from g, if A separates {2 [ a ~ 0 } f r o m { p [ b ~ 0 } ; in case $1 = &, A separates S,, ifA separates S from &; and for f = ~ axe~, A separates f , if A separates f from f . The following proposition shows why it is useful to use the orbit sums as basis of the class functions: Proposition 1 (Moody and Patera [5]). Let A be a finite W-stable (Abelian) subgroup of T of order g. Let 2, # E P++. 1. I f A separates W2 from W# then (~,, C,)A = a~u" gIW'ZI" 2. In any case (c~, go,)A is a nonnegative integral multiple We denote the nonnegative integral factor by

of gl W2I.

vA(L ~) .-

glW21

It is the cardinality of the set of all weights in Wp which are not separated from {2} by A. In case 2 = p we set VA(2) := VA(2, 2). The above proposition suggests the possibility to use discrete Fourier-like methods on class functions, with the additional restriction that the weights appearing need to be separated by the given set of EFO, i.e., f o r f = y~ a~b~ given by its values on the set A which separates all pairs ~b~,~b, ~ {~ [ a). 0} we find

(f, ~.)~
This leads to the question which weights are separated by a given set of EFOs. Corollary 2 (Transitivity of nonseparation). Let 2,#,v E P++ be dominant weights with 2 5k v. l f A does not separate q~z from (a~ and 49~from C~v, then A does not separate c~ from qa~, i.e.,

(~., ~,)~ o,

(~,, ~,.)~ o ~

(~, ~v)~ o.

S. Grimm I AppL Math. Comput. 97 (1998) 17~2

21

Proof. There are 2' E W2, kt', V" E W~t and v' E Wv such that e;"-'(x)=l and e ~...... ( x ) = l for a l l x E A .

Furthermore, as #' and y" are in the same orbit, there is a w E W with wp" = p'. Therefore

e "~'-w'' (x) = e ~''-t''+'('''-'') (x) = e z "(x)e "O''-v)(x)

~ (x)e ' - ' ( w

t/

,t

I x ) = 1 for all x E A .

[]

The following proposition is the main result of this paper. It describes the orthogonality of the orbit sums given the set T, or A,. Proposition 2. Let {21,...,2k} be any set of representatives of ( P / n Q ) / W

(respectively ( P / n P ) / W ) and A = A, (respectively Tn). Then A separates ~ from (a;.j, (i # j). More precisely

Let A = A. and suppose that (~b~,,~921)A O. Thus (from Proposition 1) " there exists/~ E WJ.i and v E W2j such that e 2 7 t i ( p , X } = e 2 ~ i ( v , X ) for all X E A,. I Write p - v = ~ k = l (ak/c)~k, where c is the determinant of the Cartan matrix and ak 6 2~. Taking the pairing with (1/n)(ok we get
Proofi

< !)a.
p
-

v, (ok

= -nc

2~,

Vk.

Thus nclak respectively nl(ak/c) for all k and thus p - v ~ nQ. N o w 2/and 2/are representatives of sets of the form W e + nQ, and therefore 2i = 2j. In case A = T, we proceed as above, expressing # - v in the weight basis, and pairing with (1/n)~. [] Let { x l , . . . ,xk} be a set of representatives of A / W . F r o m Proposition 2, it follows that the matrix

(x,).,
,

(x,)]
l

...,

has maximal rank (equal to k), and from Corollary 2 it follows that for any y E P++ there is a unique 2i such that

(a~(x) = cq~;.,(x) f o r a l l x E T.,

22

S. Grimm / Appl. Math. Comput. 97 (1998) 17-22

where c is a constant given by

c~,(e) _ [Stabw(2i)] 4h,(e) ]Stabw(#)] '

Corollary 3. For any weight # E P, the values of the orbit sum (#~ on An or Tn are completely determined (up to a factor) by the class Wp + nQ respectively Wp +nP. For the case A = An we can construct a set of representatives of ( P / n Q ) / W using the same idea used to describe the conjugacy classes of EFO in An. The fundamental domain in t* of nQ x W is

E t* I (4,

.< n},

where go = ~ ~ , is the highest root of the dual root system or, in other words, ff~i is the highest short root. Given any weight 2 = ~ a~o~i E P++ this translates to the condition ~ ~ai ~<n. Therefore we find a set of representatives for ( P / n Q ) / W as

Moreover, this set has the following property: For any 2 E BA, and any dominant weight/~ < 2 follows I~ E BA.. Note that Ba. is the unique set of representatives with this property. In addition, it is easy to see that A, separates W2 for all 2 E BA. with (2, ~0) < n, i.e. in this case we have VA()O = 1. For the case A = T, we find a set of representatives as follows: Note that W acts trivially on P / Q respectively nP/nQ, and we therefore get

nP W = (nQ x W) nn-~FQ.
Thus we find the representatives by taking into account the action of nP/nQ = P / Q on P / n Q / W . This action can be deduced from [4].

References
[1] V.G, Kac, Automorphisms of finite order of semisimple Lie algebras, J. Funct. Anal. Appl. 3 (1969) 252-254. [2] J.E. Humphreys, Introduction to Lie Algebras and Representation Theory, Graduate Texts in Mathematics, vol. 9, Springer, Berlin, 1972. [3] R.V, Moody, J. Patera, Fast recursion formula for weight multiplicities, Bull. Am. Math. Soc. 7 (1982) 237-242. [4] R.V, Moody, J. Patera, Characters of elements of finite order in simple Lie groups, SIAM J. Algebraic Discrete Methods 5 (1984) 359-383. [5] R.V. Moody, J. Patera, Computation of character decompositions of class functions on compact semisimple Lie groups, Math. Comp. 48 (1987) 799-827.