JOURNAL OF COMBINATORIAL THEORY, Series A 40, 136-149 (1985)

Harmonics on Posets
DENNIS STANTON*

School of Mathemalics, University of Minnesota,

Minneapolis. Minnesota 55455

Communicated by the Managing Editors

Received September 15, 1983

Given a finite ranked poset P, for each rank of P a space of complex valued
functions on P called harmonics is defined. If the automorphism group G of P is
sufftciently rich, these harmonic spaces yield irreducible representations of G. A
decomposition theorem, which is analogous to the decomposition theorem for
spherical harmonics, is stated. It is also shown that P can always be decomposed
into posets whose principal harmonics are orthogonal polynomials. Classical exam-
ples are given. 0 1985 Academic Press, Inc.

1. INTR~OUCTI~N

It is well known that metric association schemes [3] and distance tran-
sitive graphs [ 1] lead to discrete orthogonal polynomials. In the classical
examples, the orthogonality of distinct irreducible representations of their
automorphism groups implies the orthogonality relation for these
polynomials [6,9]. For each classical case there are also posets associated
with the polynomials [4, 5, 81. These posets allow another technique to
decompose the representations-harmonics. Although this has been carried
out in each individual case, no general theorem has been given. The pur-
pose of this paper is to state such a theorem. These results are given in
Theorems 2.17 and 2.24.
Given a poset P, a vector space of complex valued functions on P,
Harm(n), is defined. The space Harm(n) is an analogue of spherical har-
monics. Recall that for spherical harmonics, the functions on the sphere
Sp ~ ’ decompose

(1.1)

* Partially supported by NSF Grant MCS 834872.

136
0097-3165/85 $3.00
Copyright 0 1985 by Academic Press. Inc.
All rights of reproduction in any form reserved.
 HARMONICS ON POSETS 137

where Harm(n)/,,-! is an irreducible representation of the rotation group

SO(p). Theorem 2.17 and 2.24 are analogues of (1.1) for finite posets.
Delsarte [S] used certain posets (regular semilattices) to define an
association scheme. He was abie to compute the eigenvalues of the
association scheme in each example by Mobius inversion. Part of our
hypotheses (A.1 t(A.4) include a nice group action on the poset which
assures regularity (see Proposition 2.8). In Theorem 2.17 we give an explicit
formula for these eigenvalues. They are spherical functions for G-functions
on a double coset space H\G/H lying in the irreducible representation
Harm(41x,v. Thus they correspond to the ultraspherical polynomials in
(1.1). In Section 4 we show that P can be decomposed into posets whose
spherical functions are orthogonal polynomials. Some examples of posets P
and the corresponding spherical functions are given in Section 5.

2. THE DECOMPOSITION INTO HARMONICS

In this section we prove our main theorems, Theorems 2.17 and 2.24.
Throughout this paper, a ranked poset P is a finite graded poset [a],
P = U,“=, X,,, where rank (X,,) = n, X0 = (0) is the unique minimal element
of P, and X, is the set of maximal elements of P. We do not assume that
XN = (1 }, in fact the hypotheses below will insure that X, has at least
N+ 1 members.
Let L2(P) be the set of all complex valued functions on P. Then sections
of the zeta function of P provide a basis {f” : CLE P} for L2( P)

ffbj3
(2.1)
otherwise.

Put

Hom(n)=span{f”: UEX~>. (2.2)

so that Hom(n)cL’(P), and define d: L’(P) + L’(P) on the basis

(f’: REP) by

df”=Cfa, a covers 8. (2.3)

B
Put [4]
Harm(n) = Horn(n) n Ker d. (2.4)

Let G be the group of automorphisms of P. Clearly G acts on L2(P) and

Horn(n). It is easy to verify that the action of G commutes with d, so that
138 DENNISSTANTON

Harm(n) is also a G-module. In this paper we put conditions on G and P

so that the restriction of the functions in Harm(n) to X, (Harm(n)\,,) is
an irreducible G-module, and the representation of G on L2(X,) is mul-
tiplicity free with the decomposition

L2(X,) = 6 Harm(n)I (2.5)

n=O
Moreover, the spherical function for Harm(n)lxN can be explicitly com-
puted. Of course (2.5) is analogous to the spherical harmonics decom-
position of L2(Sn--I). A survey is given in [9].
We assume that
(A.l) P is a meet semilattice,
(A.2) P is a lower semimodular,
(A.3) G is transitive on X,;

and, if x(O) E X, is fixed and H = { g E G: x’“)g = x(O)},

(A.4) the non-empty H-orbits on X, are .Q,,, = (NE X,: c1 A x(O) E

Xnej}, O<i<n.
Recall [2] that (A.l) implies that any pair x, YE P have a unique greatest
lower bound, the meet x A y. Lower semimodular [2] means that if c E P
covers a and 6, a #b, then there is some dE P covered by both a and b.
Note that (A.2) does not apply for a, b E X,. For (A.4), note that
(x A y) g = xg A yg, so that the H orbits on X, are always contained in
some Sz,;.

PROPOSITION 2.6. If P and G satisfy (A.1 ), (A.3), and (A.4) then G is

transitive on each X,,.

Proof. Let c(E X, so that (A.4) implies that CIESZ,~ for some i. Let
Ct<XEXN, and choose (by (A.3)) gE G such that xg= x(O). Then
ccg< xg = x(O), so clg E Q2,,. Since H is transitive on QnO, clearly G is tran-
sitive on X,. 1
The hypothesis that 1X0 ) = 1 can be assumed without loss of generality.
For if (A.l), (A.3), and (A.4) hold then x(O) 2 a for all LXE X0. Clearly (A.3)
and (A.4) imply that x 3 czfor all XEX~, c1E X0. Since Q,, # 0, choose
x E X, such that x(O) A x E X0. For the meet x(O) A x E X0 to exist, clearly
1X,(= 1.

PROPOSITION 2.7. Let g E G act on X, x X, by (~1,x) g = (ag, xg), CC

E X,,
XE X,. If P and G satisfy (A.l), (A.3) and (A.4) then the non-empty G-
 HARMONICSONPOSETS 139

orbits on X,xX, are A(r,N,s)={(a,x):aEX,, xEXN, a A xEX,>,

O<s<r.
Proof Clearly Q,,-. x x(O) c A(r, N, s) so A(r, N, s) # 0. Given (a, x),
(a’, x’) E A(r, N, s) choose g, g’ E G such that xg = x(O) and x’g’ = x(O).
Clearly ag, a’g’ E Q,,_ s, so by (A.4) choose h E H such that ag = cr’gh, so
(a, x) g = (a’, x’) g’h. 1
Next we see that the hypotheses yield a regular semilattice as defined by
Delsarte [ 51.

PROPOSITION 2.8. If P and G satisfy (A.1 ), (A.3), and (A.4), then P is a

regular semilattice.
Proof: We verify (i), (ii), and (iii) in the definition of a regular semilat-
tice [5, p.2311. For (i), let yEXN, ZEX,, z<y, and B(z,y)= (u~X,:z<
u 6 y }. Clearly B(z, y) g = B(zg, yg) for g E G. Because g permutes P,
Proposition 2.7 implies that 1B(z, y)l is independent of z and y. For (iii), let
(a,l?)EA(r, N,r-j) and put C(~‘,~)={(~,,-)EX,~XX,:~~~, b<y,
ad:}. Again C( y, a)g= C(yg, ug) so that ) C( y, a)1 is independent of
(a, y) E A( r, N, r -j) by Proposition 2.7. For (ii), any u E X, satisfies
u = x A J for some x, y E X,. This is because QN,N-,, # 0 and G is trans-
itive on X,. Then (iii) with 2 = x and h E X, gives (ii). [
Note that Proposition 2.7 also implies that if tl E X,, x E X,, CI,< x, then
the number of maximal chains c~,~ from c( to x is independent of GLand x.
The hypothesis (A.2) has not been used yet. It is not necessary to obtain
the main results. However, in view of Proposition 2.10, it will be used in
Section 4.

LEMMA 2.9. If P satisfies (A.1 ) and (A.2), and w covers us’, then for any
XE P, rank(x A w’)>rank(x A w)- 1.
Proof Put k = rank(w) - rank(x A w) and w = w. > w1 > .. > wk =
x A w be a maximal chain. If w, = w’, then w’ > x A w and x A w’ =
x A w. Otherwise w, # w’, and w covers w, and w’. By (A.2), w1 and w’
cover some u2. If u2 = w2 again we are done. Continuing in this way we
build a maximal chain w’ > v2 > . . . > uk+ I, where wj covers vi+, . Thus
X A u”>Vk+,, and rank(x A w’) > rank(x A w)- 1. u

PROPOSITION 2.10. If P satisfies (A.l) and (A.2), then d(x, y)=

N - rank(x A y) is a metric on X, x X,.
Proof. Clearly the triangle inequality suffices, d(x, y) + d( y, z) 3 d(x, z).
Let d(x,y)=k, d(y,z)=j, and y>y,> ... >yj=y A z be a maximal
140 DENNIS STANTON

chain. From Lemma 2.9, rank( yi A x) > N - k - i. Both x and z contain

vj A x, thus rank(x A z)>N-k-j. 1
It is clear that G preserves the metric d on X,, thus G acts as
automorphisms on X,.
The representation L2(X,) is the permutation representation of G
induced from the subgroup H. In order to show that Harm(n) is an
irreducible submodule of L2(X,), it is sufficient to show that Harm(n)
has a unique right H-invariant function f, such that f,(x”‘) = 1. The
function f,, is the spherical function [6]. We investigate the right H-
invariant functions in Harm(n) by finding the right H-invariant functions
in Horn(n).

LEMMA 2.11. A basis for the right H-invariant functions in Horn(n) is

{F,,:O<i<n}, where F,i=C,Enn,f”.
Proof: This is clear, because g .f a = f p, /3 = ag - ‘. Each Fnj is non-zero
because Q,i is non-empty. i

PROPOSITION 2.12. If P and G satisfy (A.1 )(A.4), then

dFni=a,-1.,F,-,.i+b,,-,.;-*E;t~l,i-1,
where

and
b,-,,i=I(yE~,,,+,:y>aJI f0
for some given a E Q, _ ,,f_
Proof: Because d commutes with the action of G, dF,, is also right H-
invriant. From Lemma 2.11, dF,,j is a linear combination of F,, ~ lJ. Clearly,

dFti= If ‘, a E SZni, a covers /I.

d

If a E QHi, then rank(a A x(O)) = n - i. By Lemma 2.9, if a covers a, then

rank@ A x”‘)>n - i- 1, so /?EQ,_ l,i or PEQ,_,,~+,. The constant
b n ~ I,i # 0, because there is some /? covered by a on a maximal chain from
a A X(O) to a. fl
Now suppose f, = C;=. c,F,, E Harm(n) is right H-invariant. Then
Proposition 2.12 implies

a,- l,ici + b”- l,iCi+ 1 = 0, O<i<n-1. (2.13)

 HARMONICS ON POSETS 141

Because 6, _ l,i # 0, the solution to (2.13) is

i-l a
(-l)iq= fl fALco. (2.14)
j=O n--1J
Thus Harm(n) contains a unique (up to scalar multiple) function f, which
is right H-invariant. Since f, 1XN is constant on the orbits Q,, it remains to
evaluate F,; at aEQ,.

PROPOSITION 2.15. rf a E Q,, then F,Ja) = 1(/J E Q,;: b d a} I.

Proof This is clear from (2.1) and Lemma 2.11. 1
Clearly Proposition 2.15 implies that F,Jx’~‘) = 6,d,, where
ti,,= l{aeX,,:a<x’“‘}l #O. Thus, if co= 1,

f,,(x’“‘) = 4, # 0, (2.16)

and we have obtained the following theorem.

THEOREM 2.17. If P and G satisfy (A. 1 )-(A.4), then for each 0 6 n < N,
Harm(n)1 xN occurs in ,5*(X,) with multiplicity one. The normalized spherical
function for Harm(n)1 x,Vis

fn(%ty) = t c,F,,,(Q,)/d,,,
i=O

where ci is given by (2.14) and F,,,(Q,) is given by Proposition 2.15.

In order to establish (2.5), it is sufficient to show that the submodules
Harm(n) of L’(X,) are distinct. We will verify (2.5) directly.
Any fE Horn(n) can be uniquely written f =C, c, f”, aeX,,, for some
constants c,. We identify f with the 1X,1-tuple (c,), aEX,. We consider
(c,) as a column vector. Let A,, be the I X, _, I x 1X,, I incidence matrix of
x n-1 x X,. It is easy to see that f E Harm(n) if, and only if, A,c = 0. Since
A,,: Horn(n) + Hom(n - l),

Horn(n) = Harm(n) + A;f(Hom(n - 1)). (2.18)

Iterating (2.18) we have
Horn(N) = Harm(N) + A; Harm(N- 1) + ... + AZ... Af Harm(O).
(2.19)
Given f E Harm(n) which corresponds to c, f = C, c, f ‘, a E X,, the value
off(y), YEX, is
fb)=&3 aEX,,a<y. (2.20)
a
142 DENNISSTANTON

Clearly Az...A,T+, is the 1X,( x 1X, 1 matrix whose (y, c() entry is the
number of maximal chains from CIto y. Under (A.1 ), (A.3), and (A.4) this is
a constant c~,~ for c(6 y. Thus

cN.nf(Y)= LG~~X+,4W (2.21)

Thus, as functions on X,,

Harm(n)1 XN= A’, . . . AT,, Harm(n), (2.22)

and (2.19) implies

Hom(N)I xN=Harm(N)I,,+ ... +Harm(O)l,,. (2.23)

Clearly, Hom( N) I xN = L*(X,), and L2(XN) has N+ 1 independent right H-

invariant functions. Thus the right-hand side of (2.23) also has these
functions. Each Harm(n)/,, contains one such function, thus they are dis-
tinct. Since the representation is unitary, the decomposition (2.5) is
orthogonal.

THEOREM 2.24. If P and G satisfy (A.l) - (A.4), then

L*(X,) = 6 Harm(n)lx,L
tZ=O

is the decomposition of L*(X,) into distinct irreducible representations of G.

The representation L*( X,) is multiplicity free, and the spherical functions are
given in Theorem 2.17.

3. REMARKS ON THE MAIN THEOREMS

The hypothesis (A.2) is not necessary to prove Theorems 2.17 and

Theorem 2.24. Without (A.2), Proposition 2.12 could have n - i + 1 terms
rather than two terms. Nevertheless, the recurrence that results (similar to
(2.13)) has a unique solution because the leading term is non-zero. So
Theorem 2.17 holds without (A.2). Our proof of Theorem 2.24 from
Theorem 2.17 did not use (A.2). We concentrate on (A.2) in the next sec-
tion.
Clearly (2.18) and Theorem 2.24 imply

Horn(n)1 xN= 6 Harm(i)] xN. (3.1)

i=O

One can ask if Horn(n) L*(X,) has a similar decomposition into

irreducibles. In general, Harm(i)1 x, is not irreducible (see [7,
 HARMONlCSON POSETS 143

Theorem 6.231). The assumption (A.4) is crucial and applies only to H

defined from the top level XN.
If we put H, = ( g E G: yg = y> for a fixed y E X,, the intertwining
functions for Harm(n)lxm, those functions constant on the double cosets
H,,,\G/H, are also given in Theorem 2.17. They are fn(Qmi), where m
replaces N in Proposition 2.15.

4. LOWER SEMIMODULARITY AND ORTHOGONAL POLYNOMIALS

Delsarte [S] showed that X,xX, is an association scheme if P is a

regular semilattice. With the assumption (A.2) X, is a metric space, and
one can ask if the association scheme X, xX, is metric. Then the eigen-
values of the association scheme, fn(QNj), are orthogonal polynomials. In
this section we first show if the Hasse diagram of X, u X,- , is connected,
then X,x X, is a metric association scheme. The bulk of this section,
however, considers P when X, u X, _, is not connected. Then we show (in
Theorem 4.9) that P can be decomposed in a nice way relative to the
assumptions (A. 1 )-(A.4).
Clearly, X, x X, is a metric association scheme [3] if, and only if, for
any x E Q,,, 0 d k < N, there is a y E Qly.k + i such that ?c A y E X,- r . This
implies there is some path x,>cq,<x, >a1 < ... >c(,-i <.xN, cc;~X,-,,
xi E 52,; in the bipartite graph X, u X,- , . By (A.4), there is a path from
x(O) to x in X,uX,- , for any xc X,. Thus X, u X, _ , is connected. If
X, u X,- I is connected, the same argument produces y EL~~,~+ 2 for
XE!z,,.

PROPOSITION 4.1. If P and G satisfy (A. 1 )-( A.4), then X, x X,,, is a

metric association scheme under the relations Ri = {(x, v) E X, x X, :
.Y A ,v E X, ~, > I!J and on1.vif; X, u X, ~ , is connected as a bipartite graph.
It is clear that in Proposition 4.1 the metric don X, x X, coincides with
the graph distance on X,v x X,. So X, also is a distance transitive graph. If
fEHarm(n)l,,,
Tf = A,f, (4.2)
where

(Tf )(a) = cf (Ph d(B, a) = 1, % PEXN. (4.3)

LJ
In (4.3) if we put f =f, and c(= x(O), then we see that

&=fn(Q,,), O<n<N, (4.4)

144 DENNISSTANTON

are the distinct eigenvalues of T. Alsof,(a,) is a polynomial of degree j in

1” CL 91.
Next we consider the case when X, u X, _ r is not connected as a bipar-
tite graph. We show that P can be decomposed into ranked posets
inheriting (A. 1 k(A.4). We give a definition to describe the decomposition.

DEFINITION 4.5. Suppose P and Q are finite ranked posets. Then P. Q

is the finite ranked poset obtained by attaching to each maximal element
X~E P a copy Qj of Q. The minimal element of Qi is identilied with xi.
Elements from distinct copies of Q are incomparable in P. Q.
Clearly, if rank(P) = N and rank(Q) = ikf, then rank(P. Q) = N + M. Let
(X,1 =p+ 1 and {Qi}pzO be the copies of Q in P.Q.

PROPOSITION 4.6. The ranked pose? P. Q satisfies (A.1 )-(A.4) if, and
only is, both P and Q satisfy (A.l)-(A.4).
Proof. The hypothesis (A.l) is clear: for ME P, ye Qi, a A y = c1 A xi,
and for z E Q,, j # i, y A z = xi A xj. Clearly (A.2) follows for P. Q,
because a E XN+ , can cover exactly one b E X,. The other covering
relations are entirely in Q or P. For (A.3), let the automorphism groups of
P and Q be G and K, respectively. It is clear that the automorphism group
of P*Q is G=GxKx ... x K, where K occurs p + 1 times. If x0 E X, and
x(O) E Q, n X,, ,,,, are fixed, the subgroup R of G fixing x(O) is i? = H x Ix
Kx ..* x K, where I= {k E K: x”‘k = x’“‘}. It is easy to check that fi is
transitive on X,,,. If P. Q satisfies (A.3), we verify that
C=GXKX ‘.. xK, ~~=HxIxKx “. x K, and that P and Q satisfy
(A.3). The asumption (A.4) is easy to check, with

szN+n,i(P'Q) = Qni(Qo), OQi<n<M,

(4.7)
zjIJJ CQjnX,v+.), O<n<i<M,

where A = (1 <j<p: x0 A xjEX,+.-i); and

Qni(P. Q) = fini( O<ibnQN. 1 (4.8 1

We prove the following theorem in a sequence of lemmas.

THEOREM 4.9. SupposeP and G satisfy (A. 1 )-(A.4) but XN u X, _, is

not connected as a bipartite graph. Then P = R. Q for some ranked posets R
and Q which satisfy (A.1 )-(A.4). Moreover, the Hasse diagram of Q is con-
nected.
 HARMONICS ON POSETS 145

LEMMA 4.10. SupposeP and G satisfy (A. 1 )-( A.4). In the bipartite graph
A’,,,u X,- 1, ij”G, is the component containing x(O), then Gon X, = Uf= o Q,,
for some k3 1.
Proof. Let the connected components of X, v X,- , be Go,..., G,. Any
g E G is also an automorphism of X, u X,,-, . Because G is transitive on
X,, Gi is isomorphic to Go, 1 < i <p. The subgroup H fixes x(O) and so
must map Go to itself. This implies that the H-orbits on X, and X,+, are
either contained in Go or disjoint from Go. Because Sz,,, # @,
GonXNxQN,,. So choose k > 1 such that Go n XNx lJf==, Q,, and
Q N,k + I n(GonXN)=12(. Suppose QNm, m>k+2 is also in GonXN, and
choose m minimal. Then there is a path x(O) > a0 < x, > a1 < . . . > a,+ 1<
XsEBNm such that X~E XN, a,E X,-l. Clearly x, A x,-,=a,-,~XN-l
so that Proposition 2.10 implies x,- 1E QN,,-, , QN,m, or QN,m+ r.
By minimality of m, x,- 1ESZN,mp, is ruled out. Continuing, we obtain
d(xCo’, x1) 3 m which contradicts d(x”‘, x,) = 1. i

LEMMA 4.11. Zf VEX, and VEX, are in different components of

xNuxN-,, then x A y EX, ~i, wherej > k.
Proof: Suppose x A y E X, ~,, 0 <j 6 k. From (A.3), choose g E G such
that xg = x(O). Then xg A yg = x”’ A yge X,-,, so ygEQN,. Thus
ygeG,nX,, xg=x’“‘EGonX,, and xg and yg are connected in
XNUXN-~,. 4

LEMMA 4.12.
Zf aEX,-,, O<j<k, and a<x for some xEGonX,,
then no y EQNs, s > k, contains a.
ProoJ: Ifnot,a<x A y,sox A y~X~~,,O<l<k.Sinces>k,yisin
a different component of X, u X,- , than x. This contradicts Lem-
ma4.11. 1
Lemma 4.12 implies that by considering the covering relations, we can
extend Go to Go which is a connected component of the Hasse diagram of
x,u “’ ux,p,. We clearly have G,u “. uc,=X,u ... uX&k with
G, = Go. The poset Q in Theorem 4.9 will be Go.

LEMMA 4.13. For O<j<k, t?onXN_j=Uf:~QN-j,i.

Proof Clearly Go n X,,-j is a union of H-orbits QN-,,;. Let x E fiNi and

x A x(O) < y, < ... < yi = x be a maximal chain from x A x(O) to x.
Clearly ~,EX,-;+~ and yp A x(‘)EX~-~. Thus Y~EQ~-~+~,~ is in the
same component as x. Since Go n X, = IJt= o QNi, this implies that
GonxN-j ISZ,,-,,~-~, k>i>j>O, and that QN-j,i, i>k-j is in some
other component. m
146 DENNIS STANTON

LEMMA 4.14. Ifk is given by Lemma 4.11, 152,-k,01 = IeOnX,_,\ = 1.

ProojI After Lemma 4.13 it remains to show ( R,-,,, 1= 1. Choose
XCQNk c G,, n X, and g E G such xco)g = x. Because g maps x(O) to x, g
maps Go to itself. Thus g permutes 52,_k,o. Since x(O) contains all elements
of 52N-k,03 x(“)g = x does also. Since x(O) A x E x,-k is uniquely defined,
1aN-k,OI = 1. This completes the proof of Theorem 4.9 with Go = Q and
uy=‘2 Xi = R. B
The spherical functions fn for P can be given by the spherical functions
for Q and R in Theorem 4.9. For c(E R, tl A x(O)= a A x0, so for
Proposition 2.12, the constants a,- ,,i and b,- l,i are computed directly
from R if 0 < < N. The solution (2.14) is the same as R’s If 0 <j< M- 1,
note that Proposition 2.12 here becomes
dFN,,~j.,-j=bFN.,-j~l,M~j-l (4.15)

for some constant b. So the solution to (2.13) has cj = 0 for i > M-j. For
0 < i < M-j the constants a, _ l,i and b,- I,i are computed directly from
Q, z Q. So the solution (2.14) is the same as Q’s. It only remains to
restrict f, to 52, + Mj using (4.7) with n=M.

THEOREM 4.16. Suppose P satisfies Theorem 4.9 and the spherical

functions for Q and R are g,, 0 9 m < M, and f,, 0 <n < N, respectively.
Then the value of the spherical function 7, for Harm(l)\ XN+M is

(9 702 N+Mj)=oT M<j, N+l,<l<N+M,

=g/- N(QMj(Q)), O<j<M, N+ 1 <l<N+M;

(ii) J;(Q N+ ~j) =f~WLj- M(R)), M<j, 0<16N,

= 1, O<j<M,O<l<N.

5. EXAMPLES

EXAMPLE 1. For 2N < v let P= U,“=, X,, where X, is the set of all k-
dimensional subspaces of a v-dimensional vector space I/ over GF(q).
Clearly (A. 1)-(A.4) are satisfied. In Proposition 2.12 we see that
a,-l.i=(qN+‘-n+‘-l)/(q-l) (5.1)
b,~I,;=(q”-n+l-qN+‘-“+l)/(q-l). (5.2)
From Theorem 2.17 the spherical function is
 HARMONICSON POSETS 147

where (A)i = nj:A (1 - Aq’) and [;I = (q)n (q), ’ (q);-lk is the q-binomial
coefficient. A transformation [6, 41 shows that

f,(Q,)= Qn(q -‘; q”‘-“-‘, cN-‘, N; q) (5.4)

as q-Hahn polynomials,

EXAMPLE 2. Let P = lJ,“=, X, be the poset of isotropic subspaces [7]

of a vector space V over GF(q) equipped with a non-singular bilinear form.
For c(E X,, dim CI= k, and X, is the set of maximal isotropic subspaces.
Again (A.l)-(A.4) are satisfied (see [7]). We take P which corresponds to
a skew-symmetric form, so that the group which acts on P is the symplectic
group Sp(2N, q). Then

a,,_,,i=(qNp”+‘- l)/(q- l), (5.5)

b,,~,,;=(q*(N~‘i+l)-qN~‘,+‘)/(q-l). (5.6)

From Theorem 2.17 the spherical function is

It can be shown [7] that

(5.8)

a q-Krawtchouk polynomial.

EXAMPLE 3. Let [N] = (1, 2,..., N} and [k] = { 1, 2 ,..., k}, k> 2. Let P
be the poset of functionsf: A + [k], where A is any subset of [IN]. Clearly
g contains f if g is an extension of J: We can identify f with an N-tuple
(a,,..., aN), where a,=f (i), in A; ai = 0, i# A. Clearly the wreath product
Sk wr S, acts on P so that (A.1 t(A.4) are satisfied. Also

a n-,,i=N-n+ 1, (5.9)
bn-,.i=(N-n+ l)(k- 1). (5.10)

From Theorem 2.17 the spherical function is

(5.11)

This is a Krawtchouk polynomial [S],

f,r(QNj) = Kn(j; (k - 1Ilk; N). (5.12)

148 DENNISSTANTON

EXAMPLE 4. Let V and W be N and M dimensional vector spaces over

GF(q). Let P be the set of ordered pairs (a, T) [S], where c1is a subspace of
V and T is a linear map T: tl -+ W. We say (~1,T) < (fi, S) if a c /I and
S(,=T. It can be verified that (A.1 )-(A.4) hold with
G = GL( V) x GL( W) * Hom( V, W). The action is (g,, g,, X)(ct, S) = (ccgi,
g;1%2+Xl,,). Then
a,-,,i=q’(qN-“+‘-l)/(q-l) (5.13)
b,_,,i=(qM-qi)(qN--+f - l)/(q- 1). (5.14)

From Theorem 2.17 the spherical function is

/.(RNj)=[f]-’ 2 (-l)$)
I=0

[:I:][;]q ilNpj--n+i--M)/(qpM),.
(5.15)

This can be rewritten as a normalized afline q-Krawtchouk polynomial

I2317
fn(Q,)=Rn(q-‘, KM, N; 4). (5.16)

In Example 3 X, is the Hamming association scheme. Example 2 gives

the q-analogue for k = 2. The top level X,,, thus is a q-analogue of an N-
dimensional cube. Example 1 is the q-Johnson scheme. If q = 1 in Exam-
ple 1 the Johnson scheme and the Boolean algebra are obtained. Each of
Examples 14 is a metric association scheme. The next example is not.

EXAMPLE 5. Let P be a rooted tree of height N and degree d+ 1. Thus

each a E P is covered by exactly d nodes of P, and a covers a unique /I E P.
Clearly P = Uf= o X,, where 1X, 1= dk, and X, u X,- i is not connected.
An iteration of Lemma 4.6 implies that G = (S,)P, p = dN - ’ +
dNe2+ ... + 1, acts on P and satisfies (A.lp(A.4) with 152,1 =dj-dip’,
ja 1. Similarly, Theorem 4.16 implies that
fntQNj) = l3 j<N-n,
= -(d-l)-‘, j=N-n+l, (5.17)
= 0, j>N-n+l.

More generally, if P is a rooted tree of height N so that each node of level

i, 1 <i<N, covers ai nodes, it is clear that 1x1 =~~..‘a,,

(5.18)
 HARMONICS ON POSETS 149

and
LCs2Nj) = l, j6N-n,
= -(C&-l)-‘, j=N-n+l, (5.19)
= 0, j>N-n+l.
An easy computation of the L2-norm gives
n-1
dim Harm(n)(X,=n ai . (a, - 1).
i= 1 [ 1
Briefly we mention three other examples which, although they do not
(5.20)

satisfy (A.1 )-(A.4), do satisfy Theorem 2.17 and have X, u X,,, _ , connec-
ted as a bipartite graph. The first two are the analogues of Example 4 for
skew-symmetric and hermitian matrices over a finite field (see [S]). The
poset P does not have a unique meet x A y so (A.l) is not satisfied.
Nevertheless (A.2) is satisfied and X, is a metric space. The groups G is
transitive on X,,, ((A.3)), and the H-orbits on X, can be parametrized so
that analogues to Proposition 2.12, and (2.13) and (2.14) hold. For the
third example take P of Example 3 for k = 2. Let p be those elements of P
contained in 8,; those elements in X, with an even number of 1s. Then P
is a ranked poset which satisfies (A.lk(A.3). Half of the orbits Qniyj are
empty so (A.4) does not hold. Nevertheless, an analogue of Theorem 2.17
does hold.
The hypotheses (A.1 )-(A.4) might be of some use in the classification of
P-polynomial association schemes. Are all such infinite families of schemes
that are supported by a poset in (A.l)-(A.4) given by Examples lG4?

REFERENCES

1. N. BIGGS, “Algebraic Graph Theory,” Cambridge Tracts in Mathematics Vol. 67, Cam-
bridge Univ. Press, London/New York, 1974.
2. G. BIRKHOFF, “Lattice Theory,” American Mathematical Society Colloquium Publications
Vol. 25, Amer. Math. Sot., Providence, R.I., 1973.
3. P. DELSARTE, An algebraic approach to the association schemes of coding theory, Philips
Res. Rep. Suppl. 10 (1973).
4. P. DELSARTE, Hahn polynomials, discrete harmonics, and t-designs, SIAM .I. Appl. Math.
34 (1978), 157-166.
5. P. DELSARTE, Association schemes and t-designs in regular semilattices, J. Combin. Theory
Ser. A 20 (1976), 230-243.
6. C. DUNKL, Orthogonal functions on some permutation groups, Relations between com-
binatorics and other parts of mathematics, Proc. Sympos. Pure Math. 34 (1979), 129-147.
7. D. STANTON, Some q-Krawtchouk polynomials on Chevalley groups, Amer. J. Math. 102
(1980). 625-662.
8. D. STANTON, A partially ordered set and q-Krawtchouk polynomials, J. Combin. Theor)
Ser. A 30 (1981), 276284.
9. D. STANTON, “Special Functions: Group Theoretic Aspects and Applications,” pp. 87-128.
Reidel, Boston, 1984.