CYCLIC SIEVING, PROMOTION, AND REPRESENTATION

THEORY
BRENDON RHOADES
Abstract. We prove a collection of conjectures of D. White [37], as well as some
related conjectures of Abuzzahab-Korson-Li-Meyer [1] and of Reiner and White [21],
[37], regarding the cyclic sieving phenomenon of Reiner, Stanton, and White [22]
as it applies to jeu-de-taquin promotion on rectangular tableaux. To do this, we
use Kazhdan-Lusztig theory and a characterization of the dual canonical basis of
C[x11 , . . . , xnn ] due to Skandera [27]. Afterwards, we extend our results to analyzing
the fixed points of a dihedral action on rectangular tableaux generated by promotion and evacuation, suggesting a possible sieving phenomenon for dihedral groups.
Finally, we give applications of this theory to cyclic sieving phenomena involving
reduced words for the long elements of hyperoctohedral groups and noncrossing
partitions.

1. Introduction
Suppose that we are given a finite set X equipped with the action of a finite
cyclic group C generated by c. In studying the combinatorial structure of the action of C on X, it is natural to ask for the sizes of the fixed point sets X 1 =
2
(|C|1)
X, X c , X c , . . . , X c
. Indeed, the cardinalities of the above sets determine the
cycle structure of the image of c under the canonical homomorphism C SX , so
from a purely enumerative standpoint these fixed point set sizes determine the action
of C on X.
Reiner, Stanton, and White studied such actions and associated polynomials to
them which encode the sizes of all of the above fixed point sets at the same time [22].
Following their 2004 paper, we make the following definition.
Definition 1.1. Let C be a finite cyclic group acting on a finite set X and let c be a
generator of C. Let C be a root of unity having the same multiplicative order as
c and let X(q) Q[q] be a polynomial. We say that the triple (X, C, X(q)) exhibits
the cyclic sieving phenomenon (CSP) if for any integer d 0 we have that the fixed
d
point set cardinality |X c | is equal to the polynomial evaluation X( d).
Date: March 17, 2009.
1

BRENDON RHOADES

This definition generalizes Stembridges notion of the q = 1 phenomenon [34], [35]

which restricts the above definition to the case where the cyclic group C has order 2
or, equivalently, when we are given a set X equipped with an involution X X.
A few remarks are in order. First, notice that since the identity element of C fixes
every element of X, we have that X(1) = |X| whenever (X, C, X(q)) exhibits the
CSP. Also, it is easy to show that given a finite cyclic group C acting on a finite
set X, a polynomial X(q) such that (X, C, X(q)) exhibits the CSP is unique modulo
the ideal in Q[q] generated by the cyclotomic polynomial |C| (q). Finally, given any
finite cyclic group C acting on a finite set X, it is possible to show that the triple
P
i
(X, C, X(q)) exhibits the CSP, where X(q) = |C|1
i=0 ai q and ai is the number of
C-orbits in X with stabilizer order dividing i. Typically the interest in a CSP is that
X(q) may be taken to be a natural polynomial deformation of a formula enumerating
|X|.
These polynomial deformations often
come from the theory of q-numbers. For any
n 1
n N, define the q-analogue [n]q := qq1
= 1 + q + q 2 + + q n1 . Following the case
 
q
.
of natural numbers, we define further [n]!q := [n]q [n1]q . . . [1]q and nk q := [k]!q[n]!
[nk]!
q
n
It is well known that k q is the generating function for partitions which fit inside an
(nk) by k rectangle and is, therefore, a polynomial [29]. We first give the canonical
example of the CSP.
Theorem 1.1. (Reiner-Stanton-White 2004 [22]) Fix two positive integers k n.
Let X be the set of all subsets of [n] having size k and let C = Z/nZ act on X via
the long cycle (1, 2, . . . , n) Sn .
 
Then, the triple (X, C, X(q)) exhibits the CSP, where X(q) = nk q .
Reiner et. al. also proved a version of the above result for the case of multisets.

Theorem 1.2. (Reiner-Stanton-White 2004 [22]) Fix two positive integers k and n.
Let X be the set of all k-element multisets of [n]. Let C = Z/nZ act on X by the
long cycle (1, 2, . . . , n) Sn .


.
Then, the triple (X, C, X(q)) exhibits the CSP, where X(q) = n+k1
k
q

In each of these results and in all of the rest of the CSPs appearing in this paper, the
set X is a set of combinatorial objects and C is generated by a natural combinatorial
operator on X. We shall see that Theorems 1.1 and 1.2 are both implied by CSPs
regarding certain sets of tableaux and an action given by the sliding algorithm of
jeu-de-taquin promotion. Postponing definitions until Section 2, we state our three
main results. The following was conjetured by D. White [37].
Theorem 1.3. Let n be a rectangular partition and let X = SY T (). Let
C = Z/nZ act on X by jeu-de-taquin promotion.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

Then, the triple (X, C, X(q)) exhibits the cyclic sieving phenomenon, where X(q)
is the qanalogue of the hook length formula
[n]!q
X(q) = f (q) :=
.
(i,j) [hij ]q
There are several interpretations of the polynomial X(q) in the above result.
In addition to being the natural q-analogue of the hook length formula, up to a
power of q we may also interpret X(q) to be the fake degree polynomial corresponding to the -isotypic component of the action of Sn on the coinvariant algebra
C[x1 , . . . , xn ]/C[x1 , . . . , xn ]S+n (see Proposition 4.1 Stanley [32] together with Corollary 7.21.5 of [30]). Moreover, we have that X(q) is equal to the Kostka-Foulkes
polynomial K,1n (q) corresponding to and the composition (1n ) of length n consisting entirely of 1s. Finally, up to a power of q, X(q) is equal to the qanalogue
of the Weyl dimension formula for the (1, 1, . . . , 1)-weight space of the irreducible
representation of GLn (C) having highest weight . There is a column strict version
of the above result, also conjectured by White [37].
Theorem 1.4. Let k 0 and let n be a rectangular partition. Let X = CST (, k)
and let C = Z/kZ act on X via jeu-de-taquin promotion.
Then, the triple (X, C, X(q)) exhibits the cyclic sieving phenomenon, where X(q)
is a qshift of the principal specialization of the Schur function
X(q) := q () s (1, q, q 2, . . . , q k1),
where is the statistic on partitions = (1 , 2 , . . . ) given by
() = 01 + 12 + 23 + . . . .
There are several interpretations of the polynomial X(q) in the above theorem,
as well. Just as standard tableaux of a fixed shape are enumerated by the hook
length formula, the hook content formula enumerates column strict tableaux of a fixed
shape (with uniformly bounded entries). Up to a power of q, the polynomial X(q)
is the qanalogue of the hook content formula. Also, X(q) is equal to MacMahons
generating function for plane partitions which fit inside a box having dimensions 1
by 1 by k 1 weighted by number of boxes [18]. This latter interpretation can
be easily seen via the obvious bijection between column strict tableaux and plane
partitions. Finally, up to a power of q, X(q) is equal to the qanalogue of the Weyl
dimension formula corresponding to the irreducible representation of GLk (C) having
highest weight .
The previous two results have concerned standard tableaux of fixed rectangular
shape and column strict tableaux of fixed rectangular shape and arbitrary content,
respectively. We can also formulate a result suggested by Reiner and White [21]
[37] concerning column strict tableaux of fixed shape and fixed content. Specifically,

BRENDON RHOADES

suppose that n is rectangular and |= n is a composition with length () = k.

Also suppose that has some cyclic symmetry, i.e., there exists some d|k such that
the dth cyclic shift operator preserves . It can be shown that the action of promotion
on column strict tableaux acts cyclically on content compositions. Therefore, the dth
power of promotion acts on the set of column strict tableaux of shape and content
.
Theorem 1.5. Let n and |= n be as above. Let C = Z/( kd Z) = hci act on the
set of column strict tableaux of shape and content by the dth power of promotion.
th
Let be a primitive kd root of unity.
Then, for any m 0, the number of fixed points under the action of the mth power
of promotion is equal to the modulus |K, ( m)|, where K, (q) is the Kostka-Foulkes
polynomial.
This result is almost, but not quite, a CSP. Specifically, since the evaluation of a
Kostka-Foulkes polynomial at a root of unity may be strictly negative, we do not have
that the action of the dth power of promotion on the relevant set of tableaux together
with K, (q) exhibits the CSP. However, we do have cyclic sieving up to modulus.
Moreover, since the q-hook length formula is (up to a power of q) a special case of
a Kostka-Foulkes polynomial, it follows that (again up to modulus) our result on
standard tableaux is a special case of this latter result. It will turn out that Theorem
1.5 is a weight space refinement of Theorem 1.4.
The bulk of the remainder of this paper is devoted to the proofs of the above
Theorems 1.3, 1.4, and 1.5. We now give our overarching philosophy which was also
used by Stembridge [35] in relation to the action of evacuation on tableaux and the
q = 1 phenomenon. Suppose that we are given some finite set X equipped with
the action a finite cyclic group C = hci and a polynomial X(q) and we wish to show
that (X, C, X(q)) exhibits the CSP. On its face, this is a purely enumerative problem
- if we could find some formula for the evaluation of X(q) at appropriate roots of
unity and equate this with some formula enumerating the fixed point set of X under
the action of appropriate elements of C, we would be done. This direct approach,
when possible, has its merits. CSPs are certainly of enumerative interest and in
many cases show that previously studied natural q analogues of counting formulas
for fundamental combinatorial sets encode information about fixed point sets under
fundamental combinatorial operators.
However, this nave approach can often be very difficult. For example, while direct
combinatorial proofs of the CSPs for subsets and multisets in [22] exist, there is no
known enumerative proof of the CSPs for rectangular tableaux given above. We
instead take the viewpoint that, instead of being purely enumerative, CSPs often
conceal deeper algebraic structure which can be exploited in their proofs.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

Specifically, suppose that we have some complex vector space V with basis {fx | x
X} indexed by the elements of X. Suppose further that V is equipped with the
action of a group G and denote the corresponding representation by : G GL(V ).
Finally suppose that we can find some element g G whose action on V is given by
the formula
(1.1)

(g)(fx ) = fcx ,

for all x X. That is, the matrix for the action of g with respect to the given basis of
V is the permutation matrix corresponding to the action of c on X. It is immediate
d
that for any d 0, the number of fixed points |X c | of the action of cd on X is equal
to the trace of the linear operator (g d ). In symbols, if : G C is the character of
our representation,
(1.2)

|X c | = (g d),

for all d 0. So, we have reduced our problem of enumerating fixed point sets to
the evaluation of a certain character at a certain group element. If we can interpret
this character evaluation as an appropriate root of unity evaluation of X(q), then our
CSP will be proved.
At first glance, this approach may seem to complicate matters. We must first
model our action of C on X in a representation theoretic context. Moreover, the
evaluation of the character (g d) may be no easier than the enumeration of the fixed
d
point set X c , particularly if the action of C on X is easy to understand. However,
character theory is a well studied subject and provides us with much artillery with
which to attack the former problem. An elementary and fundamentally important
example of this is that if h G is a group element which is conjugate to g, we
have that (g d ) = (hd ), where the latter character may be easier to compute. So,
under nice conditions, the problem of proving a CSP may be reduced to a problem
of studying conjugacy in a group. And, as a bonus, this method of proving a CSP
reveals representation theoretic structure that may have previously gone unnoticed.
In our situation, we will gain some understanding of why the hypothesis that our
partitions be rectangular is necessary in our results. It is this approach that we will
use in proving our CSPs.
These representation theoretic methods, however, come with some fine print. It is
often too much to ask that an equation of the form (1.1) hold. Sometimes we must
content ourselves with finding a group element g G such that for all x X we have
(1.3)

(g)(fx ) = (x)fcx ,

where : X C is some function, hopefully as simple as possible. Moreover, the

identification of a polynomial evaluation at a root of unity with a character evaluation
may require more involved techniques than elementary group conjugacy. In particular,

BRENDON RHOADES

this part of the proof of our result on standard tableaux will use results from Springers
theory of regular elements.
The remainder of this paper is organized as follows. In Section 2 we review some
tableaux theoretic definitions, the algorithm of jeu-de-taquin promotion, and some
representation theoretic tools we will be using in the proofs of our CSPs (specifically
Kazhdan-Lusztig theory). In Section 3 we prove our CSP for standard tableaux. In
Section 4 we derive a slightly new perspective on the irreducible representations of the
general linear group which will lead to a proof of our CSP for column strict tableaux
in Section 5. In Section 6 we extend the general philosophy of the proofs in Section
5 to get our result concerning column strict tableaux of fixed content. In Section 7
we extend our results on cyclic actions and prove some results which enumerate fixed
points under combinatorial dihedral actions. In Section 8 we derive corollaries of our
fixed point results on tableaux for other combinatorial actions involving handshake
patterns and the reflection group Bn . We close in Section 9 with some open questions.

2. Tableaux and Representation Theory Background

We begin this section by gonig over the definitions of standard and column strict
tableaux, as well as the definition of the action of jeu-de-taquin promotion. For a
more leisurely introduction to this material, see [25] or [30].
Given a positive integer n, a partition of n is a weakly decreasing sequence of
positive integers = (1 2 k ) such that 1 + 2 + + k = n. The
number k is the length () of . The number n is the size || of . We write n
to indicate that is a partition of n. We will sometimes use exponential notation to
write repeated parts of partitions, so that 42 33 1 is the partition (4, 4, 3, 3, 3, 1) 18.
We identify partitions n with their Ferrers diagrams, i.e., is identified with
the subset of the lower right quadrant of Z Z given by {(i, j) | i [j ]}. For
example, the Ferrers diagram of the partition (4, 4, 3, 1) 12 is given by

.
Given a coordinate (i, j) in the Ferrers diagram of , the hook length hij at (i, j)
is the number of dots directly south or directly east of (i, j), the dot (i, j) included.
In the above partition h12 = 6. A partition is said to be rectangular if its Ferrers
diagram is a rectangle. The conjugate of a partition is the partition whose Ferrers
diagram is obtained by reflecting the Ferrers diagram for across the line y = x.
So, (4, 4, 3, 1) = (4, 3, 3, 2).

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

Given partitions and such that we have a set theoretic containment of Ferrers
diagrams , we define the skew partition / to be the set theoretic difference
r of Ferrers diagrams. The size of / is the difference || || and we write
/ n to denote that the skew partition / has size n.
Given a positive integer n, a composition of n is a finite sequence of nonnegative
integers = (1 , 2 , . . . , k ) which satisfies 1 + 2 + + k = n. In particular,
some of the i may be zero. We write |= n to denote that is a composition of
n. The number n is called the size || of and the number k is called the length
() of . If |= n and () = k, the composition defines a function [n] [k]
given by sending every number in the interval (1 + + i1 , 1 + + i ] to the
number i for i = 1, 2, . . . , k. We denote this function by , as well. For example, if
= (0, 2, 1, 0, 1), then : [4] [5] is given by (1) = 2, (2) = 2, (3) = 3, (4) = 5.
Given a partition = (1 , 2 , . . . , k ) of length k, define ck to be the cyclically
rotated composition (2 , 3 , . . . , k , 1 ). This defines an action of the order k cyclic
subgroup of Sk generated by (1, 2, . . . , k) on the set of all compositions of n having
length k.
Let n and let |= n be a composition with () = k. A column strict tableau
T of shape with content is a filling of the Ferrers diagram of with 1 1 s, 2 2 s,
. . . , k k s such that the numbers increase strictly down every column and weakly
across every row. An example of a column strict tableau of shape (4, 4, 3, 1) is
1 1 3 4
3 3 4 6
4 5 5
6
.
Given a partition n and a composition |= n, the set of all column strict
tableaux of shape and content is denoted CST (, k, ), where k := () reminds us of the maximum possible entry in our tableau. The set of all column
strict tableaux
S of shape with entries at most k is denoted CST (, k), so that
CST (, k) = |=n,()=k CST (, k, ). The tableaux obtained by requiring that entries increase weakly down columns and strictly across rows are called row strict and
we have the analogous definitions of RST (, k, ) and RST (, k). Finally, a tableau
is called standard if its content composition consists entirely of 1 s. We denote by
SY T () the set of all standard tableaux of shape and make the canonical identification of SY T () with CST (, n, 1n ) for n. Given any partition n, we
have the column superstandard tableaux CSS() obtained by filling in the boxes of
within each column, going from left to right. For example,
1 4 7
CSS((3, 3, 2)) = 2 5 8
3 6
.

BRENDON RHOADES

For a partition n, f := |SY T ()| is famously enumerated by the FrameRobinson-Thrall hook length formula [8]:
n!
(2.1)
f =
.
(i,j) hij
Also given a positive integer k, the Schur function associated to the partition in k
variables is the element of the polynomial ring C[x1 , . . . , xk ] given by
X
X
(2.2)
s (x1 , . . . , xk ) =
x1 1 . . . xk k .
|=n,()=k T CST (,k,)

That is, the Schur function associated to is the generating function for column strict
tableaux of shape weighted by their content compositions.
The sets X involved in the most important CSPs proved in this paper will be
standard tableaux with fixed rectangular shape, column strict tableaux with fixed
rectangular shape and uniformly bounded entries, and column strict tableaux with
fixed rectangular shape and specified content. The cyclic actions on each of these
sets will be based on jeu-de-taquin promotion, a combinatorial algorithm which we
presently outline. The action of promotion has received recent attention from Stanley
[31], who considers a related action on linear extensions of finite posets, and from
Bandlow, Schilling, and Thiery [2], who derive crystal theoretic uniqueness results
related to the action of promotion on column strict tableaux.
Roughly speaking, promotion acts on tableaux by deleting all of the highest possible
entries, sliding the remaining entries out while preserving the column strict condition,
and then increasing all entries by one and filling holes with one so that the resulting
object is a column strict tableaux. More formally, suppose that we are given a
partition n and a positive integer k. We define the jeu-de-taquin promotion
operator j : CST (, k) CST (, k) as follows. Given a tableau T in CST (, k),
first replace every k appearing in T with a dot. Suppose that there is some dot in the
figure so obtained that is not contained in a continuous strip of dots in the northwest
corner. Then, choose the westernmost dot contained in a connected component of
dots not in the northwest corner of the figure. Say this dot has coordinates (a, b),
where a > 0 and b < 0. By our choice of dot, at least one of the positions (a 1, b) or
(a, b + 1) must be filled with a number - i.e., the positions immediately west or north
of our dot. If only one of these positions has a number, interchange that number and
the dot. If both of these positions have a number, interchange the greater of these
two numbers and the dot (if the two numbers are equal, interchange the number
at the northern position (a, b + 1) and the dot). This interchange moves the dot
one unit north or one unit west. Continue interchanging the dot with numbers in
this fashion until the dot lies in a connected component of dots in the northwest
corner of the resulting figure. If all of the dots in the resulting figure are not in a
connected component in the northwest corner, choose the westernmost dot contained

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

in a connected component of dots not in the northwest corner of the figure and slide
this new dot to the northwest corner in the same way. Iterate this process until all
dots are contained in a connected component in the northwest corner of the figure.
Now increase all the numbers in the figure by 1 and replace the dots with 1 s. The
resulting figure is j(T ).
Example 2.1. Let T be the following element of CST ((4, 4, 3, 1), 6):
1 1 3 4
3 3 4 6
T = 4 5 5
6
.
We compute the image j(T ) of T under jeu-de-taquin promotion.
1 1 3 4
3 3 4 6
T = 4 5 5
7

1 1 3 4
3 3 4
4 5 5
7

1 3 4
1 3 4
3 5 5
4

1 3
1 1 2 4
1 3 4 4
2 4 5 5
7 3 5 5
7 4 6 6
= j(T )
4
5
Notice that the content of T is (2, 0, 3, 3, 2, 2) and the content of j(T ) is (2, 2, 0, 3, 3, 2)
- that is, j acts by cyclic rotation on content compositions. Also notice that the result
of applying j to T depended on our considering T as an element of CST ((4, 4, 3, 1), 6).
If we had considered T as an element of CST ((4, 4, 3, 1), k) for any k > 6, then we
would have
2 2 4 5
4 4 5 7
j(T ) = 5 6 6
7
.
For the proof of the following lemma about promotion, see [25] or [30].
Lemma 2.1. Let T be in CST (, k).
1. The tableau j(T ) is well-defined, i.e., independent of the order in which we chose
dots to slide northwest in the algorithm.
2. The tableau j(T ) is column strict with entries k.
3. If T has content , then j(T ) has content ck .
So, j is indeed a well defined function CST (, k) CST (, k). By running the
defining algorithm for j in reverse, we see that j is a bijection and we call its inverse
j 1 jeu-de-taquin demotion. By Part 3 of the above lemma, j restricts to an operator

10

BRENDON RHOADES

SY T () SY T () where we consider the upper bound k to be equal to ||. We

call this restriction promotion, as well, and retain the notation j for it.
We now review the construction of the Kazhdan-Lusztig basis for the Hecke algebra
Hn (q) and the symmetric group algebra C[Sn ], the construction of the KazhdanLusztig cellular representations of Sn , and the interaction of these representations
with the combinatorial insertion algorithm of RSK. See [4], [9], [25], and [30] for a
more leisurely introduction to this material.
A fundamental result in the representation theory of the symmetric group Sn is
that we have an isomorphism of C[Sn ]-modules
M
f W ,
C[Sn ]
=
n

where W is the irreducible representation of Sn indexed by the partition . Viewing

the left regular representation as a well understood object, it is natural to study bases
of the left hand side of the above isomorphism which allow the visualization of the
decomposition on the right hand side and, in particular, facilitate the study of the
irreducibles W . The natural basis {w | w Sn } fails miserably in this regard - every
element of the symmetric group acts as an n! by n! permutation matrix with respect
to this basis, rendering the above isomorphism invisible. It turns out a basis for C[Sn ]
in which the above isomorphism is evident arises in a natural way when one studies
algebras which generalize the symmetric group algebra.
In a fundamental 1978 paper [14], Kazhdan and Lusztig studied the representation
theory of the Hecke algebra Hn (q), which is a quantum deformation of the symmetric
group algebra C[Sn ] and reduces to C[Sn ] in the specialization q = 1. Hn (q) admits
an involution which restricts to the identity on C[Sn ] and it is natural to ask whether
there are bases of Hn (q) which are invariant under this involution. The answer is yes,
and up to certain normalization conditions these bases are unique. Specialization of
this basis of Hn (q) at q = 1 yields a basis of C[Sn ] which, as we will see, can be viewed
as upper triangular with respect to the above isomorphism. Also amazingly, the
avatars of the irreducibles W so obtained will interact very nicely with combinatorial
notions such as descent sets of tableaux and RSK insertion.
The symmetric group Sn has a Coxeter presentation with generators s1 , s2 , . . . , sn1
and relations

(2.3)

si sj = sj si

for |i j| > 1,

(2.4)

si sj si = sj si sj

for |i j| = 1,

(2.5)

s2i = 1

for all i.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

11

We interpret si to be the adjacent transposition (i, i + 1). The length (w) of a

permutation w Sn is the minimum value of r so that w = si1 . . . sir for some adjacent
transpositions sij . Call such a minimal length word reduced. We define the left descent
set of w to be the subset DL (w) of [n 1] given by DL (w) := {i | (siw) < (w)}. The
right descent set of w, DR (w), is defined analogously. In this paper, we will denote
permutations in Sn by their cycle decomposition, their expressions as words in the
Coxeter generators si , and by their one line notation. In this latter system, writing
w = w1 w2 . . . wn means that w sends 1 to w1 , 2 to w2 , and so on.
Given a partition = (1 , 2 , . . . , k ) n, we define the Young subgroup of Sn
indexed by to be the subgroup S of Sn which stabilizes the sets
{1, 2, . . . , 1 }, {1 + 1, . . . , 1 + 2 }, . . . , {1 + + k1 + 1, . . . , 1 + + k = n}.
For example, if = (3, 2, 2) 7, then the associated Young subgroup of S7 is given
by S = S{1,2,3} S{4,5} S{6,7} . In general, we have a natural direct product decomposition S
= S1 Sk which implies the corresponding order formula
|S | = 1 ! k !.
Sn comes equipped with a (strong) Bruhat order, the partial order given by the
transitive closure of w
S v if and only1if there exists some (not necessarily simple)
reflection t T := wSn ,i[n1] wsi w such that v = tw and (w) (v). Unless otherwise indicated, writing w v for permutations w and v will always mean
comparability in Bruhat order. The identity permutation 1 is the unique minimal
element of Sn under Bruhat order. The long element wo whose one-line notation is
n(n 1) . . . 1 is the unique maximal element in Sn under Bruhat order.
Let q be a formal indeterminate. The Hecke algebra Hn (q) is the C(q 1/2 )-algebra
with generators Ts1 , Ts2 , . . . , Tsn1 subject to the relations:
(2.6)

Tsi Tsj = Tsj Tsi

for |i j| > 1,

(2.7)

Tsi Tsj Tsi = Tsj Tsi Tsj

for |i j| = 1,

(2.8)

Ts2i = (1 q)Tsi + q

for all i.

It turns out that if w is a permutation in Sn and si1 . . . sir is a reduced expression

for w, then the Hecke algebra element Tw := Tsi1 . . . Tsir is independent of the choice
of reduced word for w. Moreover, the set {Tw | w Sn } forms a basis for Hn (q) over
the field C(q 1/2 ). Finally, it is obvious that the specialization of the defining relations
of Hn (q) to q = 1 yields the classical group algebra C[Sn ].
It follows from the definition of Hn (q) that the generator Tsi is invertible for all i,
with Ts1
= 1q (Tsi (1 q)). Therefore, for any permutation w in Sn we have that the
i
algebra element Tw is invertible, being a product of invertible elements. With this in
mind, define an involution D of Hn (q) by D(q 1/2 ) = q 1/2 and D(Tw ) = (Tw1 )1 and

12

BRENDON RHOADES

extending linearly over C. In the latter formula, the inverse in the subscript is taken
in the symmetric group Sn and the inverse in the exponent is taken in the Hecke
algebra Hn (q). Notice that in the specialization at q = 1, the involution D is just the
identity map on C[Sn ].
Theorem 2.2. (Kazhdan-Lusztig
[14]) There is a unique basis
P

1/2 (w)
{Cw (q) = (q
)
vSn Pv,w (q)Tv | w Sn } of Hn (q) such that

1. (Invariance) D(Cw (q)) = Cw (q) for all permutations w Sn ,

2. (Polynomality) Pw,v (q) Z[q] always,
3. (Normalization) Pw,w (q) = 1 for any w Sn ,
4. (Bruhat compatibility) Pv,w = 0 unless v w in Bruhat order, and
5. (Degree bound) The degree of Pv,w (q) is at most (1/2)((w) (v) 1).
The basis in the above theorem is called the Kazhdan-Lusztig (KL) basis of Hn (q).
Its specialization to q = 1 is a basis of C[Sn ], but to make some results in this paper
look cleaner we will throw
signs and call {Cw (1) | w Sn } the KL basis
P in some (w)(v)

of C[Sn ], where Cw (1) = vSn (1)

Pv,w (1)v. These signs will not seriously
affect the representation theory. The polynomials Pv,w (q) are the KL polynomials
and are notoriously difficult to compute for general v and w. The KL basis of either
Hn (q) or C[Sn ] leads to the definition of the KL representation, which is just the left
regular representation of either algebra viewed with respect to this basis.
In light of the degree bound in the above theorem, we recall a statistic (v, w) on
ordered pairs of permutations v, w Sn by letting (v, w) = [q ((w)(v)1)/2 ]Pv,w (q).
So, (v, w) is the coefficient of the maximum possible power of q in Pv,w (q). By
Bruhat compatibility and polynomality, we have that (v, w) = 0 unless v w and
also (v, w) := (w) (v) is odd. Moreover, we introduce a symmetrized version of
given by [u, v] := max{(u, v), (v, u)}.
The KL function can be used to get a recursive formula for the Pu,v (q) which will
be of technical importance to us in what follows.
Lemma 2.3. ([14], Equation 2.2c) If u w and i DL (w), then
X (v,w)
(2.9)
Pu,w (q) = q 1c Psi u,si w (q) + q c Pu,siw (q)
q 2 (v, siw)Pu,v (q),
si v<v

where c = 1 if i DL (u) and c = 0 otherwise.

The KL function is well behaved with respect to left and right multiplication by
the long element wo Sn , as well as taking the inverses of the permutations involved.
For a proof of the following lemma, see for example [4].
Lemma 2.4. Let u, v Sn . We have that (u, v) = (wo v, wou) = (vwo , uwo) =
(wo uwo, wo vwo ). Also, we have that (u, v) = (u1 , v 1).

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

13

Starting from the KL representation, we get a natural preorder on permutations in

Sn . Specifically, for u and v in Sn , say that u L v if and only if there exists some i in
[n1] such that Cv (1) appears with nonzero coefficient in the expansion of the product
si Cu (1) in the KL basis of C[Sn ]. The transitive closure of L defines a preorder L
on Sn (i.e., a reflexive, transitive relation which need not be antisymmetric). The
preorder L is called the left KL preorder. The right KL preorder R is defined in
exactly the same way, but by instead considering the right regular representation of
Hn (q) or C[Sn ]. The two sided KL preorder LR is the transitive closure of the union
of L and R .
Given any set X equipped with a preorder , we can define an equivalence relation
on X by x y if and only if x y and y x. The preorder induces a partial order
on the equivalence classes X/ also denoted defined by [x] [y] if and only if for
some elements x [x] and y [y], x y. The equivalence classes of permutations in
Sn so defined via the preorders L , R , and LR are called the left, right, and two
sided KL cells, respectively. Remarkably, these cells can be identified via an explicit
combinatorial algorithm.
from counting the dimensions on both sides of the isomorphism C[Sn ]
=
LIt follows

n f W that the sets of permutations w in Sn and ordered pairs (P, Q) of standard

tableaux of the same shape with n boxes are in bijection. The Robinson-SchenstedKnuth (RSK) algorithm gives an explicit bijection between these sets. For a detailed
definition of its algorithm see [25] or [30]. In this paper, w 7 (P (w), Q(w)) will
always mean that w row inserts to (P (w), Q(w)). For example, in S6 we have that

1 3 4 5
1 3 4 6
.
, 2
623415 7 2
6
5
Define the shape sh(w) of a permutation w to be the shape of either P (w) or Q(w).
So, sh(623415) = (4,1,1).

To further explore the interaction between the RSK algorithm and the algebraic
properties of the Coxeter group Sn , we introduce the notion of a descent set of a
tableau. Given n and T SY T (), the descent set D(T ) of T is the subset of
[n 1] defined by i D(T ) if and only if i + 1 occurs strictly south and weakRoughly
speaking, promotion acts on tableaux by deleting all of the highest possible entries,
sliding the remaining entries out while preserving the column strict condition, and
then altering entries and filling holes so that the resulting object is a column strict
tableaux.ly west of i in T . For example, if
1 3 5
T = 2 4 7
6
,

14

BRENDON RHOADES

then D(T ) = {1, 3, 5}. In the case of rectangular tableaux, we will later generalize
the descent set to another combinatorial object called an extended descent set which
will help us greatly in proving our representation theoretic results.
The RSK algorithm behaves in a predictable way with respect to taking inverses,
finding descent sets, and left and right multiplication by the long permutation wo . To
show this, we recall Sch
utzenbergers combinatorial algorithm of evacuation. Given
an arbitrary partition n and a tableau T in CST (, k) for some k 0, define
the image e(T ) of T under evacuation as follows. First, embed T in the northwest
corner of a very large rectangle. Then rotate the rectangle by 180 degrees, moving T
to the southeast corner. Then, for i = 1, 2, . . . , k, replace each i occurring in T with
k i + 1. Finally, use the jeu-de-taquin sliding algorithm to move the boxes of T from
the southeast corner to the northwest corner of the rectangle. The resulting tableau
is e(T ).
Lemma 2.5. Let n and let T CST (, k, ) for some k 0 and some composition = (1 , 2 , . . . , k ).
1. e(T ) is a well-defined element of CST (, k), that is, independent of the choices
involved in embedding or sliding.
2. The content of e(T ) is (k , k1 , . . . , 1 ).
3. The operator e is an involution, that is, e(e(T )) = T always.
With Part 2 of the above lemma as motivation, given a composition = (1 , . . . , k ),
we define wok to be the reverse composition (k , k1 , . . . , 1 ). It follows from the
above lemma that e restricts to an involution on SY T () for arbitrary partitions
. Evacuation interacts nicely with RSK. Proofs of these results can be found, for
example, in [25] and [4].
Lemma 2.6. Let w Sn and suppose w 7 (P, Q).
1. w 1 7 (Q, P ).
2. wo w 7 (P , e(Q) ).
3. wwo 7 (e(P ) , Q ).
4. wo wwo 7 (e(P ), e(Q)).
5. i DL (w) if and only if i D(P ).
6. i DR (w) if and only if i D(Q),
where T denotes the conjugate of a standard tableau T .
RSK leads to a natural triple of equivalence relations on Sn . We say that two permutations v, w Sn are left Knuth equivalent if we have P (w) = P (v). Analogously,
v and w are right Knuth equivalent if we have the equality Q(w) = Q(v). Finally, we
have an equivalence relation given by w v if and only if w and v have the same
shape. Amazingly, these algorithmic, combinatorial equivalence classes agree with
the KL cells.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

15

Theorem 2.7. Let w and v be permutations in Sn .

1. w and v are left Knuth equivalent if and only if w and v lie in the same left KL
cell.
2. w and v are right Knuth equivalent if and only if w and v lie in the same right KL
cell.
3. w and v have the same shape if and only if w and v lie in the same two-sided KL
cell.
So, we may interpret the partial orders induced by the left and right KL preorders
as partial orders on the set of standard tableaux with n boxes. It can be shown
that these two partial orders are identical. Similarly, the induced partial order on
two-sided KL cells can be identified with a partial order on all partitions of n. It
can be shown that this latter partial order is just the dominance order dom defined
by dom if and only if for all i 0 we have the comparability of partial sums
1 + 2 + + i 1 + 2 + + i . We also have the following change-of-label
result, which shows that the symmetrized KL -function behaves well with respect
to a translation of Knuth class.
Theorem 2.8. (Change of label) Identify permutations with their images under RSK.
Let U1 , U2 , T1 , and T2 be standard tableaux with n boxes, all having the same shape.
We have that
1. [(U1 , T1 ), (U1 , T2 )] = [(U2 , T1 ), (U2 , T2 )] and
2. [(U1 , T1 ), (U2 , T1 )] = [(U1 , T2 ), (U2 , T2 )].
Thanks to change of label, we can define [P, Q] for two standard tableaux P and
Q of the same shape to be the common value of [(P, T ), (Q, T )] = [(T, P ), (T, Q)],
where T is any standard tableau having the same shape as P and Q.
With these results in hand, we are ready to define a powerful avatar of the Specht
modules. Let n be a partition and choose an arbitrary tableau T SY T ().
By
2.7 and its following remarks we have a left action of Sn on S0,T :=
L Theorem

w C{Cw (1)}, where the w in the direct sum ranges over all permutations such that
either sh(T ) <dom sh(w) or T = P (w). Also by the paragraph following Theorem 2.7
L
the module S0,T contains an Sn -invariant submodule S1,T := w C{Cw (1)}, where
now w ranges over all permutations such that sh(T ) <dom sh(w). It therefore makes
sense to define S ,T to be the quotient module
S ,T := S0,T /S1,T .
The vector space S ,T carries an action of the group algebra C[Sn ]. We can identify
the basis vectors of S ,T with elements of SY T () via associating to a tableau U of

shape the image of C(T,U

) (1) in the above quotient.
It is natural to ask to what degree the module S ,T depends on the choice of
standard tableau T . In this direction, it turns out that the action of the Coxeter

16

BRENDON RHOADES

generators si on the space S ,T is completely determined by the symmetrized KL function and the purely combinatorial descent set of a tableau. Identifying the basis
elements of S ,T with tableaux in SY T (), we have that the action of si on S ,T is
given by the explicit formula
(
P
if i D(P )
P
(2.10)
si P =
P + iD(Q) [P, Q]Q if i
/ D(P ).

This formula can be proven via some reasonably explicit manipulations of the KL
basis and KL polynomials [9]. In particular, the action of si does not depend on the
tableau T that we chose in the construction of S ,T . It follows that the matrices giving
the left action of si with respect to the given bases of S ,T and S ,U are literally equal
(up to reordering basis elements) for any two standard tableaux T, U SY T ().
With this strong isomorphism in hand, we define the C[Sn ]-module S to be the
C[Sn ]-module S ,T for any choice of T SY T (). Again remarkably, the modules S
are precisely the irreducible representations of Sn . For an exposition and extended
version of the following result, see [9].
Theorem 2.9. The module S is isomorphic as a C[Sn ]-module to the irreducible
representation of Sn indexed by the partition .
S is called the (left) KL cellular representation indexed by . The basis elements
of S are homomorphic images of a subset of the KL basis of C[Sn ] and are in
natural bijection with the elements of SY T (). This avatar of the Specht modules will
turn out to be very useful in the representation theoretic modeling of combinatorial
operators.
An earlier use of the modules S in this way is due to Berenstein, Zelevinsky [3]
and Stembridge [35] and concerns the action of the evacuation operator e. Since e has
order 2, one would hope to find an order 2 element in the symmetric group Sn which
maps (up to sign) to the permutation matrix corresponding the evacuation under the
KL cellular representation. It turns out that the long element does the trick.
Theorem 2.10. (Berenstein-Zelevinsky [3], Stembridge [35]) Identify the basis vectors
of the Kazhdan-Lusztig (left) cellular representation corresponding to with elements
P SY T (). Denote this representation by : Sn GL(S ). Let wo Sn be the
long element.
Then, up to a plus or minus sign, we have that (wo ) is the linear operator which
sends P to e(P ), where e is evacuation.
Informally, this result states that given any partition n, the image of the long
element wo Sn models the action of evacuation on SY T (). It will turn out that for
rectangular partitions , the image of the long cycle (1, 2, . . . , n) in Sn will model the

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

17

action of promotion on SY T (). A straightforward application of the MurnaghanNakayama rule determines whether the sign appearing in the above theorem is a plus
or a minus. The issue of resolving this sign will be slightly more involved for us and
will be taken care of by direct manipulation of the KL basis. In the next section we
show precisely how this is done.
3. Promotion on Standard Tableaux
Given a rectangular shape = ba with ab = n, we want to determine how the
operation of jeu-de-taquin promotion on SY T () interacts with the left KL cellular
representation S of shape . Our first goal is to show that the promotion operator
j interacts nicely with the function.
Define a deletion operator d : SY T (ba ) SY T (ba1 (b 1)) by letting d(U) be the
(standard) tableau obtained by deleting the n in the lower right hand corner of U. It
is easy to see that d is a bijection. Our first lemma shows that d is well behaved with
respect to the function.
Lemma 3.1. For any U, T SY T (ba ), we have that [U, T ] = [d(U), d(T )].
Proof. The line of reasoning here is similar to one used by Taskin in Lemma 3.12 of
[36].
Recall that for any partition n we have the associated column superstandard
tableau CSS() defined by inserting the numbers 1, 2, . . . , n into the diagram of first
from top to bottom within each column, then from left to right across columns. Define
permutations u = u1 . . . un , v = v1 . . . vn Sn and t = t1 . . . tn1 , w = w1 . . . wn1
Sn1 by their images under RSK:
u 7 (U, CSS(ba )),
v 7 (T, CSS(ba )),
w 7 (d(U), CSS(ba1 (b 1)),
t 7 (d(T ), CSS(ba1 (b 1)).
Using the definition of the RSK algorithm its easy to check that, using one-line
notation for permutations in Sn ,
u1 . . . un = w1 w2 . . . wna (n)wna+1 . . . wn1 ,
v1 . . . vn = t1 t2 . . . tna (n)tna+1 . . . tn1 .
By a result of Brenti ([5], Theorem 4.4), we have an equality of Kazhdan-Lusztig
polynomials Pw,t (q) = Pw(n),t(n) (q). For 1 k n, define permutations w (k) and
t(k) in Sn by w (k) := sk sk+1 sn1 (w(n)), t(k) := sk sk+1 sn1 (t(n)). Here w(n)
(respectively t(n)) is the permutation in Sn whose one-line notation is w1 . . . wn1 n

18

BRENDON RHOADES

(respectively t1 . . . tn1 n). By the above remarks it follows that w (na+1) = u and
t(na+1) = v.
We claim that for all k, Pw(k) ,t(k) (q) = Pw(k+1) ,t(k+1) (q). It will then follow by induction and Brentis result that Pu,v (q) = Pw,t(q). To see this, notice that w (k) = sk w (k+1) ,
t(k) = sk t(k+1) , and the transposition sk satisfies sk DL (w (k+1) ) DL (t(k+1) ). These
conditions together with Lemma 2.3 in the case c = 1 imply that we have the following
polynomial relation:
(3.1) Pw(k) ,t(k) (q) = Psk w(k+1) ,sk t(k+1) (q)
= Pw(k+1) ,t(k+1) (q) + qPsk w(k+1) ,t(k+1) (q)

(r,t(k) )
2

(r, t(k+1) )Psk w(k+1) ,r (q).

sk r<r
(i1)

(i1)

However, we also have that both w

and t
map i to n for all i. Therefore,
(k)
(k+1)
(k) (k+1)
w t
and the Bruhat interval [w , t
] is empty. This implies that the only
surviving term in the above sum is Pw(k+1) ,t(k+1) (q) and we get that Pu,v (q) = Pw,t(q), as
desired. To complete the proof, we make the easy observations that (u, v) = (w, t)
and that the pairs (u, v) and (w, t) both lie in the same right cell. Therefore, by
taking the left tableaux of the pairs (u, v) and (w, t), we get the desired equality of
coefficients.

Observe that the proof of Lemma 3.1 shows a stronger statement regarding an
equality of KL polynomials rather than just an equality of their top coefficients.
It is easy to show using Lemma 2.4 and Part 4 of Lemma 2.6 that [e(P ), e(Q)] =
[P, Q] for any standard tableaux P and Q having the same arbitrary shape. We use
this fact and the above result to get the desired fact about the action of j.
Proposition 3.2. Let P, Q be standard tableaux which are either both in SY T (ba ) or
SY T (ba1 (b 1)). We have that [P, Q] = [j(P ), j(Q)].
This proposition says that, in the special case that our shape is a rectangle or a
rectangle missing its outer corner, the action of promotion preserves the function.
This does not hold in general for other shapes. Indeed, a counterexample may be
found in the smallest shapes which do not meet the criteria of Proposition 3.2: =
(3, 1) and = (2, 1, 1). In the case of = (3, 1), promotion acts on the set SY T ((3, 1))
via a single 3-cycle:


1 2 3
1 3 4
1 2 4
.
4
, 2
, 3

However, it is easy to check that in S4 we have [4123, 2134] = 0, [2134, 3124] = 1,

and [3124, 4123] = 1. Since for any standard tableau T the column reading word of
T (i.e., the element of S|T | obtained by reading off the letters in T from bottom to top,
then from left to right) has insertion tableau T , we obtain the claimed counterexample.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

19

This lemma and the above example give some representation theoretic justification
for why we need our hypotheses that be rectangular in Theorems 1.5 through
1.7, but perhaps leave mysterious why these results fail in general for rectangles
minus outer corners. We will soon give combinatorial intuition for why the strictly
rectangular hypotheses are needed.

Proof. (of Proposition 3.2) We prove this first for the case of rectangular shapes.
The case of shapes which are rectangular minus an outer corner will follow easily. We introduce a family of operators on tableaux. Define a creation operator
c : SY T (ba1 (b 1)) SY T (ba ) by letting c(T ) be the tableau obtained by adding
a box labeled n to the lower right hand corner of T . Given a tableau T of shape /,
let r(T ) be the tableau obtained by rotating T by 180 degrees and let (T ) be the
tableau obtained by replacing every entry i in T with |T | + 1 i. Given a standard
tableau T which is not skew, let J(T ) be the tableau obtained by embedding T in
the northwest corner of a very large rectangle, and playing jeu-de-taquin to move the
boxes of T to the southeast corner. Observe that we have the following relations of
operators:
I. r = r
II. e = rJ
III. j = rcrJd
Let P and Q be standard tableaux of shape ba . With the above definitions, we have
the following chain of equalities:
[P, Q] = [d(P ), d(Q)]

(Lemma 3.1)

= [ed(P ), ed(Q)]
= [rJd(P ), rJd(Q)]

(II)

= [crJd(P ), crJd(Q)]

(Lemma 3.1)

= [ecrJd(P ), ecrJd(Q)]
= [rJcrJd(P ), rJcrJd(Q)]

(II)

= [rcrJd(P ), rcrJd(Q)]

(crJd(P ), crJd(Q) are rectangles)

= [rcrJd(P ), rcrJd(Q)]

(I)

= [j(P ), j(Q)].

(III)

This completes the proof for the case of rectangular shapes. For the case of shapes
which are rectangular minus an outer corner, notice that we have the following equality of operators on SY T (ba1 (b 1)):
IV. j = djc.

20

BRENDON RHOADES

Let P and Q be in SY T (ba1 (b 1)). We have the following chain of equalities:

[P, Q] = [c(P ), c(Q)]
= [jc(P ), jc(Q)]

(Lemma 3.1)
(c(P ), c(Q) are rectangles)

= [djc(P ), djc(Q)]
= [j(P ), j(Q)].

(Lemma 3.1)
(IV)

This completes the proof of the lemma.


To better understand the action of promotion, we introduce a new combinatorial
set related to rectangular tableaux which will be the ordinary tableau descent set
with the possible addition of n. Given a standard tableau P of rectangular shape
with n boxes, define the extended descent set of P denoted De (P ) [n], as follows.
For i = 1, 2, . . . , n 1, say that i De (P ) if and only if i is contained in the ordinary
descent set D(P ). To determine if n is contained in De (P ), consider the tableau U
with entries {2, 3, . . . , n} obtained by deleting the 1 in P and playing jeu-de-taquin to
move the resulting hole to the southeastern corner. The entry n is either immediately
north or immediately west of this hole in U. Say that n is contained in De (P ) if and
only if n appears north of this hole in U. The fundamental combinatorial fact about
the extended descent set is that promotion acts on it by cyclic rotation.
Lemma 3.3. Let P SY T (ba ). For any i, we have that i De (P ) (mod n) if and
only if i + 1 De (j(P )) (mod n).
Example 3.1. Let b = 4 and a = 3 and let P be the following element of SY T (4, 4, 4):
1 2 4 9
P = 3 5 8 11
6 7 10 12 .
Applying the promotion operator to P yields
1 2 3 5
j(P ) = 4 6 9 10
7 8 11 12 .
It can be shown that De (P ) = {2, 4, 5, 9, 11} and De (j(P )) = {3, 5, 6, 10, 12}.
For partitions n other than rectangles, it is not in general possible to define an
extended descent set which agrees with the ordinary descent set on the letters [n 1]
and on which j acts with order n. Indeed, it can be shown that promotion acts with
order 6 on SY T ((3, 2)) and (3, 2) is a partition of 5. This gives some combinatorial
intuition for why we needed the strictly rectangular hypotheses in our results.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

21

Proof. We prove the equivalent statement involving jeu-de-taquin demotion. Let Q :=

j 1 (P ) denote the image of P under the demotion operator and suppose that i
De (P ) for i [n]. We want to show that i 1 De (Q) (mod n). This is shown in
several cases depending on the value of i. Let U be as in the above paragraph and
let be the jeu-de-taquin path in the rectangle ba involved in obtaining U from P .
Case I. i {2, 3, . . . , n 1}.
For this case, observe that a single jeu-de-taquin slide does not affect the (ordinary)
descent set of a tableau. Here we extend the definition of the descent set of a tableau
to a tableau with a hole in it in the obvious way.
Case II. i = n.
Since P is rectangular, must contain n. Since n is in De (P ), the entry immediately
above n is also contained in . Thus, n moves up when U is constructed and we get
that n 1 is immediately above n in Q. Therefore, n 1 is a descent of Q, as desired.
Case III. i = 1.
We must show that n De (Q). To do this, we consider the action of demotion on
Q. Let be the jeu-de-taquin path in Q involved in sliding out a hole at the upper
left, so that ends in the lower right hand corner of Q at the entry n. We want to
show that the final edge of is a downward edge into n.
To do this, define a subset S of the path by (x, y) S if and only if (x+ 1, y) .
Here we consider diagrams of partitions to be subsets of the fourth quadrant of the
plane intersected with the integer lattice Z2 which have their upper leftmost coordinate equal to (1, 1). Since starts at (1, 1) and ends at (b, a), the projection
of S onto the x-axis is the full interval [1, b 1]. If the final edge of were not
a downward edge into n, it would follow that intersected the set S nontrivially.
Choose (x, y) with x minimal so that (x, y) S.
We claim that (x, y + 1) is not contained in . Indeed, if (x, y + 1) were contained
in , then since is a jeu-de-taquin path the (x, y)-entry of Q is less than the
(x + 1, y + 1)-entry of Q, contradicting the fact that P is standard. So, either x = 1
or (x1, y) . If x = 1, then since (1, y) but (1, y + 1)
/ , we must have that
y = 1. But (1, 1) S contradicts 1 being a descent of P . If (x 1, y) , then
the fact that the projection of S onto the x-axis is [1, b1] contradicts the minimality
of x. We conclude that the final edge of is a downward edge into n, as desired. 
We prove a technical lemma about the image of the long cycle under the KL
representation. This lemma will be used to pin down a constant in a Schurs Lemma
argument in the proof of Proposition 3.5.
Lemma 3.4. Let cn = (1, 2, . . . , n) Sn be the long cycle. Let = ba be a rectangle.
Identify permutations with their images under RSK. The coefficient of the KL basis

22

BRENDON RHOADES

element C(j(CSS()),CSS())
(1) in the expansion of C(CSS(),CSS())
(1)cn in the KL basis
a1
of C[Sn ] is (1) .

Proof. Define permutations u, v Sn by

u 7 (CSS(), CSS()),
v 7 (j(CSS()), CSS()).
It follows that the one-line notation for u and v is
u =a(a 1) . . . 1(2a)(2a 1) . . . (a + 1) . . . n . . . (n a + 1),
v =(a + 1)a . . . 31(2a + 1)(2a) . . . (a + 3)2(3a + 1)(3a) . . . (2a + 3)(a + 2) . . .
n(n 1) . . . (n a + 2)(n + 2 2a).
By inspection, both of these are 3412 and 4231 avoiding, so therefore by smoothness
considerations (see Theorem A2 of [14]) all of the KL polynomials Px,u (q) for x Sab
are equal to 1 and we have the corresponding formulas:
Cu (1) =

(1)(x,u) x,

xSab

Cu (1)cn =

(1)(x,u) xcn .

xSab

The unique Bruhat maximal element xo cn for which xo Sab has one-line notation
xo cn = (a+1)a . . . 32(2a+1)(2a) . . . (a+2)(3a+1) . . . (2a+2) . . . n(n1) . . . (na+2)1.
Right multiplying by c1
n yields that
xo = a(a1) . . . 21(2a) . . . (na)(na1) . . . (n2a+1)(n1)(n2) . . . (na+1)(n).
Therefore, xo and u differ by a cycle of length a and (1)(xo ,u) = (1)a1 . It follows
that there exist numbers y C,
X
Cu (1)cn = (1)a1 Cx o cn (1) +
y Cy (1).
y<xo cn

The coefficient we are interested in is v . First, notice that the Bruhat interval
[v, xo cn ] consists of all permutations in Sn whose one line notation has the form
(a + 1)a . . . 31 (2a + 1)(2a) . . . (a + 3)2 . . . n(n 1) . . . (n a + 2)b ,

where 1 {1, 2}, 2 {2, a + 2}, . . . b1 {n 3a + 2, n 2a + 2}, b {n

2a + 2, 1}. It follows from the definition of the Bruhat order that the subposet of
such permutations is isomorphic to the Boolean lattice Bb1 of rank b 1. Moreover,
all permutations of the above form are 3412 and 4231 avoiding, meaning that for

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

every w [v, xo cn ] we have that Cw (1) =

[v, xo cn ] Sab cn = {xo cn }.

zw (1)

(z,w)

23

z. Finally, we observe that

All of the conditions in the above paragraph imply that v = (1)a1 , as desired.

We now have all of the ingredients necessary to analyze the relationship between
j and the KL representation. Specifically, we show that up to a predictable scalar, j
acts like the long cycle (1, 2, . . . , n) Sn . This gives our desired analogue of Equation
1.5.
Proposition 3.5. Let = ba be a rectangular shape with ab = n and let :
Sn GL(S ) be the associated KL cellular representation, with basis identified with
SY T (). Let cn denote the long cycle (1, 2, . . . n) Sn . Define a C-linear map
J : S S by extending J(P ) = j(P ). We have that
(3.2)

(cn ) = (1)a1 J.

Equivalently, for any standard tableau P SY T (), we have that

(3.3)

(cn )(P ) = (1)a1 j(P ).

Proof. We want to show that the operator J 1 (cn ) commutes with the action of Sn .
If we can do this, by Schurs Lemma it will follow that for some constant C
we have the equality of operators J = (cn ). In [35], Stembridge used this sort of
argument to prove Theorem 2.10.
Since Sn is generated by the simple reflections si for i = 1, 2, . . . , n1, it is enough to
show that J 1 (cn ) commutes with (si ) for each i = 1, 2, . . . , n1. From conjugation
within the symmetric group we know that (cn )(si ) = (si+1 )(cn ) always, where
we interpret sn to be the affine transposition (1, n). So, it is enough to show that
J 1 (si+1 ) = (si )J 1 always. For i = 1, 2, . . . , n 2 this will follow from the action
of Coxeter generators on S , but for i = n 1 this poses a problem since there is no
known nice formula in general for the action of sn on these modules. In fact, we show
that J 1 (cn ) commutes with (si ) for only i = 1, 2, . . . , n 2 and then apply the
branching rule for restriction of irreducible symmetric group modules to show that
these commutation relations are in fact sufficient for our purposes.
We first show that for i = 1, 2, . . . , n 2 we have
J 1 (cn )(si ) = (si )J 1 (cn ).
Given a standard tableau P , let j 1 (P ) be the unique tableau which maps to P under
promotion. From the corresponding conjugation relation in the symmetric group it
follows that
(cn )(si ) = (si+1 )(cn ).

24

BRENDON RHOADES

On the other hand, because 1 i n 2 and J acts as j on basis elements indexed

by standard tableaux, we have that for any P SY T (),
(
j 1 (P )
if i + 1 De (P )
P
J 1 (si+1 )(P ) =
1
1
j (P ) + i+1De (Q) [P, Q]j (Q) if i + 1
/ De (P )
(
j 1 (P )
if i De (j 1 (P ))
P
=
j 1 (P ) + iDe (j 1 (Q)) [P, Q]j 1 (Q) if i
/ De (j 1 (P ))
(
j 1 (P )
if i De (j 1 (P ))
P
=
j 1 (P ) + iDe (j 1 (Q)) [j 1 (P ), j 1(Q)]j 1 (Q) if i
/ De (j 1 (P ))
= (si )J 1 (P ).

The first equality is the action of the simple transpositions on the cellular KL basis,
the second is Lemma 3.3, the third is Proposition 3.2, and the fourth is again the
definition of the action of simple transpositions.
The above discussion implies that the operator J 1 (cn ) commutes with the action
of the parabolic subgroup Sn1 of Sn on the irreducible Sn -module S . Since is
a rectangle, has a unique outer corner and by the branching rule for symmetric
groups, the restriction S SSnn1 remains an irreducible Sn1 -module. Therefore, by
Schurs Lemma, we conclude that there is a number C so that
J = (cn ).
We want to show that = (1)

a1

. This follows from Lemma 3.4.

The above result states that for Specht modules of rectangular shape the image of
the long cycle cn under the KL representation is plus or minus the permutation matrix
which encodes jeu-de-taquin promotion. It is not true that any conjugate of cn in Sn
enjoys this property - indeed, many are not even permutation matrices. Observe the
analogy to Theorem 2.10 which states that the image of the long element wo under
the KL representation for an arbitrary shape is plus or minus a permutation matrix
which encodes evacuation.
Also notice that if is a partition of any nonrectangular shape, then has more
than one outer corner. This implies that the restricted module S SSnn1 is not irreducible and the above proof breaks down.
As a corollary to this we get a classical result on the order of promotion [11].
Corollary 3.6. For n rectangular, every element of SY T () is fixed by j n .
Remark 3.2. For arbitrary partitions it is not true that the order of promotion on
SY T () divides ||. In fact, this order for arbitrary is unknown. So, the hypothesis
that be rectangular in the statement of the the corollary is necessary. Moreover, it

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

25

is not in general true that for a rectangular partition n the order of promotion
on SY T () is equal to n. For an extreme counterexample, if is either a single row
or a single column, the set SY T () consists of a single element and j has order 1.
Other counterexamples exist, minimal among them = (2, 2). The author does not
know of a formula for the order of j on SY T () for arbitrary shapes . As a possible
indication of the subtlety here, given any nonrectangular partition n, there exists
a tableau P SY T () which is not fixed by the nth power of promotion [19].
For the special case of staircase shapes = (k, k 1, . . . , 1) n, the operator j 2n
fixes every element of SY T (). This leads one to hope that the KL cellular basis may
be exploited in studying the action of promotion on SY T () for staircase shapes .
Unfortunately, even for the staircase (3, 2, 1) we do not have the result analogous to
Proposition 3.2 that j preserves the function. It is, however, possible to show that
for any staircase , the action of j || on SY T () is given by transposition, which is
known to preserve the function. Taken together, the operators of transposition and
evacuation generate a group isomorphic to the Klein four group Z/2ZZ/2Z, leaving
some hope that the study of the action of this smaller group might be analyzed by our
methods. Unfortunately, the case of the staircase (3, 2, 1) also shows that no power j d
for 1 d 5 preserves the function, so it would seem difficult to use our methods
to analyze the action of any of the larger groups on the set SY T () generated by
lower positive powers of j together with evacuation.
Even in the case of rectangular shapes , the previous corollary does not give the
order of j on a specific tableau T SY T (). For the case where has 3 rows,
this order can be interpreted as the order of the cyclic symmetry of a combinatorial
object associated to T called an A2 web [20].
Let sn denote the affine transposition (1, n) in Sn . As another corollary, we get
a nice formula for the image of sn under the KL representation.
Corollary 3.7. Let = ba be a rectangular shape and identify the basis of the corresponding left KL cellular representation with SY T ().
We have that

(
P
P
sn P =
P + nDe (Q) [P, Q]Q

if n De (P )
if n
/ De (P ).

Proof. Using the notation of the proof of the above proposition, we have that sn1
commutes with the action of J 1 (c). It follows that J 1 sn (P ) = sn1 J 1 (P ). The
desired formula follows.

Finally, the above analysis leads to many equalities of coefficients.

26

BRENDON RHOADES

Corollary 3.8. Let n be a rectangle and let Y and Y be two orbits of SY T ()

under the action of j which have relatively prime orders. Given P, Q Y and P , Q
Y we have that [P, P ] = [Q, Q ].
Example 3.3. For n = 6 and = (2, 2, 2), we have that |SY T ()| = 5 and the cyclic
action of j on SY T () breaks SY T () into one cycle of size 2 and one cycle of size
3. Explicitly, the operator j acts on SY T ((2, 2, 2)) via:

1 4
1 2
1 3
1 3
1 2
2 5
3 5
2 4 2 5
3 4
3 6 , 4 6 , 5 6
4 6 , 5 6

Mapping the elements of these orbits to their column reading words in S6 , the above
Corollary implies that when v is any element of the set {321654, 521643, 431652} S6
and w is any element of the set {421653, 531642} we have that [v, w] is a constant.
Since 421653 covers 321654 in Bruhat order, we see that this common value of is
equal to 1.
An application of the above proposition yields Theorem 1.3, our desired result on
cyclic sieving in the action of jeu-de-taquin on standard tableaux. Implicit in the
statement of Theorem 1.3 is the fact that the nth power of promotion fixes every
element of SY T (). This fails if is nonrectangular. Even if we were to replace
C with the cyclic group having order equal to the order of promotion on SY T ()
for nonrectangular , the action of promotion together with the appropriate q-hook
length formula would still not in general yield a CSP. For example, consider the
shape = (3, 3, 1) 7. There are 21 standard tableaux of shape and the action of
promotion breaks SY T () up into three orbits having sizes 3, 5, and 13. Thus, the
order of promotion on SY T () is 3 5 13 = 195. It can be shown that the q-hook
length formula for the shape lambda is X(q) = (1+q+q 2 +q 3 +q 4 +q 5 +q 6 )(1+q 2 +q 4 ).
The evaluation of X(q) at a primitive 195th root of unity is not even real, let alone
equal to zero, the number of fixed points of one iteration of promotion on SY T ().
Example 3.4. In the above example of the action of promotion on SY T ((2, 2, 2)),
we have that
[6]!q
f (2,2,2) (q) =
= (1 q + q 2 )(1 + q + q 2 + q 3 + q 4 ).
[4]q [3]q [2]q [3]q [2]q [1]q
i

Letting = e 3 , we compute directly that

f (2,2,2) (1) = 5 f (2,2,2) () = 0 f (2,2,2) ( 2) = 2
f (2,2,2) ( 3) = 3 f (2,2,2) ( 4 ) = 2 f (2,2,2) ( 5 ) = 0.
This is in agreement with the fixed point set sizes:
2

|SY T ((2, 2, 2))1| = 5 |SY T ((2, 2, 2))j | = 0 |SY T ((2, 2, 2))j | = 2

3
4
5
|SY T ((2, 2, 2))j | = 3 |SY T ((2, 2, 2))j | = 2 |SY T ((2, 2, 2))j | = 0,

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

27

as predicted by Theorem 1.3.

Proof. (of Theorem 1.3) By Corollary 3.6, C does indeed act on X by promotion. Let
2i
= e n . Viewing cn = (1, 2, . . . , n) as a permutation matrix in Sn GLn (C), we
get that c is conjugate to diag(1, , 2, . . . , n1). This means that for any d 0 cdn is
conjugate to diag(1, d, 2d, . . . , d(n1) ).
Let denote the character of the irreducible representation of Sn corresponding
to . It is well-known that cn is a regular element of Sn , that is, there is an eigenvector of the image of cn under the reflection representation which avoids all reflecting
hyperplanes. A result of Springer (Proposition 4.5 of [28]) on regular elements implies
that, for any d 0, we have the character evaluation (cdn ) = d() f ( d), where
,
() = 01 + 12 + 23 + . Since = ba , we can compute that () = ba(a1)
2
d()
d(a1)
which implies that
= (1)
.
On the other hand by Proposition 3.5, (cdn ) is equal to the trace of (1)d(a1) J d ,
where J is the permutation matrix which records promotion. This trace is (1)d(a1)
d
times the number of fixed points |X j | of the action of j d on X. The desired CSP
follows.

4. A Construction of the Finite Dimensional Irreducible GLk (C)
Representations
In the last section we proved a CSP regarding rectangular standard tableaux by
modeling the action of promotion by the image of the long cycle in Sn under the
Kazhdan-Lusztig cellular representation of appropriate shape. We want to prove
an analogue of this CSP for column strict tableaux of rectangular shape having n
boxes and entries uniformly bounded by k. In analogy with the last section, we will
construct a representation having dimension CST (, k) for rectangular n under
which some group element will act as the promotion operator with respect to some
basis.
It is well known that the irreducible finite dimensional representations of the general
linear group GLk (C) are paramaterized by partitions having at most k rows and that
the dimension of the irreducible representation labeled by is equal to |CST (, k)|.
The symmetric group Sk embeds naturally into GLk (C) as the subgroup of permutation matrices and it will turn out that the long cycle (1, 2, . . . , k) can be taken to
model the action of promotion on column strict tableaux provided our shape is
rectangular. It remains, however, to find a basis for our representation under which
(1, 2, . . . , k) acts as a monomial matrix corresponding to promotion. This will involve
a construction of finite dimensional GLk (C) irreducible representations of arbitrary
shape which we now present. This construction is essentially due to Du [7] together
with results of Skandera [27] but we rederive it here for self-containment. Our basis

28

BRENDON RHOADES

will essentially be a homomorphic image of a subset of the dual canonical basis for
the polynomial ring C[x11 , . . . , xkk ] and, as such, will bear spiritual similarities to
Lusztigs canonical basis [17]. As a starting point of our construction, we introduce a
family of polynomials called immanants. The homomorphic images of an appropriate
subset of these polynomials will be the zero weight space of our representations.
For a positive integer n N, let x = (xij )1i,jn be an n n matrix of commuting
variables and let C[x11 , . . . , xnn ] be the complex polynomial ring in these variables.
We will sometimes abbreviate the latter ring as C[xij ]1i,jn . Call a polynomial
in C[xij ]1i,jn an immanant if it belongs to the C-linear span of the permutation
monomials {x1,w(1) xn,w(n) | w Sn }. Thus, immanants form an n!-dimensional
complex vector space. Given any polynomial f (x11 , . . . , xnn ) C[xij ]1i,jn and an
n n matrix A = (aij ) with entries in any commutative Calgebra R, define f (A)
to be the element f (a11 , . . . , ann ) of R obtained by applying f to A.
Following [24], define for any w Sn the w-Kazhdan-Lusztig immanant Immw (x)
by the equation
X
(4.1)
Immw (x) =
(1)(w,v) Pwo v,wo w (1)x1,v(1) xn,v(n) .
vw

Specializing to the identity permutation, we have that Imm1 (x) = det(x). So, at least
superficially, KL immanants are deformations of the determinant. It can be shown
that the KL immanants share the properties of Schur nonnegativity and total nonnegativity with the determinant [24], and that for the case of permutations w which
are 321-avoiding, the application of the KL immanant Immw (x) to the path matrix
of a planar network has a combinatorial interpretation which naturally generalizes
Lindstroms Lemma [23]. By the Bruhat triangularity of the KL polynomials, the set
{Immw (x) | w Sn } forms a basis for the vector space of immanants.

We will find it necessary to work with KL immanants defined on variable sets with
repeated entries. More precisely, given any pair of compositions , |= n, we define
the matrix x, to be (x(i),(j) )1i,jn . Note that either or both of () or () may
exceed n. We also construct the corresponding polynomial ring C[x(i),(j) ]1i,jn .
For a permutation w Sn , denote by Immw (x, ) the element of C[x(i),(j) ]1i,jn
which results in applying the w-KL immanant to the matrix x, . So, for example,
we have that Immw (x) = Immw (x1n ,1n ). In this paper we will mostly be interested in
the case where = 1n .
Example 4.1. Take n = 3 and let = (1, 2) and
given by

x11 x12
x(1,2),(1,1,1) = x11 x12
x21 x22

= (1, 1, 1). Our matrix x, is

x13
x13
x23 .

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

29

Let w = 213 S3 . Recalling that every KL polynomial for pairs of permutations in

S3 is either 0 or 1 depending on whether the pairs are Bruhat comparable, we see
that
Imm213 (x(1,2),(1,1,1) ) = x12 x11 x23 x12 x13 x21 x13 x11 x22 + x13 x12 x21 .
On the other hand, computing the KL immanant corresponding to the permutation
231 S3 yields the result
Imm231 (x(1,2),(1,1,1) ) = x12 x13 x21 x13 x12 x21 = 0.
As the above example shows, it can happen that a polynomial of the form Immw (x, )
is equal to zero. Indeed, since Imm1 (x) = det(x), such a polynomial is nonzero for
w = 1 if and only if all of the parts of and are at most 1. It is possible to
derive a criterion based on the RSK correspondence for determining precisely when
the polynomials Immw (x, ) vanish [27]. A remarkable fact, due to Skandera, is that
the nonvanishing polynomials of the form Immw (x, ) form a basis for the polynomial
ring C[x11 , . . . , xkk ] [27]. Moreover, Skandera showed that the basis so constructed is
essentially equivalent to Lusztigs dual canonical basis. We will not need the full force
of this result, and now state the theorem which will be relevant for our purposes.
Theorem 4.1. (Skandera [27]) Let k 0. The nonzero elements of the set {Immw (x, )},
where w ranges over Sn and and range over all possible compositions of n having length k, are linearly independent and a subset of the dual canonical basis of the
polynomial ring C[x11 , . . . , xkk ] in k 2 variables.
We now relate polynomial rings to representation theory. Let Y = Ck and Z = Cn
be two complex vector spaces of dimensions k and n, respectively. Let Y and Z
denote their dual spaces with standard bases {y1 , . . . , yk } and {z1 , . . . , zn }, respectively. Now the tensor product Y Z has basis xij := yi zj , for 1 i k and
1 j n. In this way, we identify the symmetric algebra Sym(Y Z ) with the
polynomial ring C[xij ]1ik,1jn . This space carries an action of the general linear
group GL(Y ) = GLk (C), where matrices act on the first component of simple tensors
by g (f h) := (f g 1) h. Taking n = k and i [n 1], viewing the adjacent
transposition si as an element of Sn GLk (C), we quote a result from [24] about
this action.
Lemma 4.2. Let w Sn . We have that
(
Immw (x)
P
(4.2) si Immw (x) =
Immw (x) + Immsi w (x) + si z>z (w, z)Immz (x)

si w > w,
si w < w.

Using the formula in the above lemma, one can show that the KL immanants
form a cellular basis for the vector space of immanants. More precisely, identifying
permutations with their images under RSK, for any n and T SY T (), we have

30

BRENDON RHOADES

that the space

VT,n,1
n := C{Imm(T,P ) (x) | P SY T ()}

C{Imm(U,S) (x) | U, S SY T ()}

>dom

is closed under the left action of Sn . This closure is essentially a consequence of the
fact that the formula in Lemma 4.2 is almost the same as the formula for the action
of si on the KL basis of C[Sn ].

Since VT,n,1
n is closed under the action of C[Sn ] for any T , it follows immediately
that the quotient space
M

C{Imm(U,S) (x) | U, S SY T ()})

VT,n,1n := VT,n,1
n /(
>dom

carries the irreducible Sn -representation corresponding to the shape . A basis for

VT,n,1n is given by the image of the set {Imm(T,P ) (x) | P SY T ()} under the canonical projection map. Letting I1n (P ) denote the image of Imm(T,P ) (x) under this
projection, we can write this basis as {I1n (P ) | P SY T ()}. The 1n appearing
in this notation will be fully justified when we broaden our scope to nonstandard
tableaux. By a change of label argument, the representation of Sn on the quotient
space VT,n,1n does not depend on the choice of the tableau T , and is given by
(
if i D(P )
I1n (P )
P
(4.3)
si I1n (P ) =
I1n (P ) + iD(Q) [P, Q]I1n (Q) if i
/ D(P ).

So far we have used KL immanants to construct representations of the group algebra

C[Sn ]. The goal of this section is to construct modules over GLk (C). To do this,
we first recall combinatorial notions of standardization and semistandardization on
tableaux. These operations give us a means to transform row strict tableaux into
standard tableaux and vice versa, when possible.
It is always possible to transform a row strict tableau into a standard tableau.
Given a partition n and a row strict tableau P RST (, k, ) for some |= n,
define the standardization std(P ) of P to be the element of SY T () given by replacing
the 1 1s in P with the numbers [1, 1 ] increasing down columns, replacing the 2 2s
in P with the numbers [1 + 1, 1 + 2 ] increasing down columns, and so on.
Example 4.2. Suppose that P is the tableau given by
1 7 8
P = 1 9
RST ((3, 2, 2), 9, (2, 1, 0, 0, 0, 0, 1, 1, 2)),
2 9
then we have that

1 4 5
std(P ) = 2 6
SY T ((3, 2, 2)).
3 7

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

31

The reverse case is more complicated. Given n and a standard tableau T

SY T () along with a composition |= n, say that T is -semistandardizable if the
sequences of numbers [1, 1 ], [1 +1, 1 +2 ], . . . , all occur in vertical strips in T which
increase down columns. Equivalently, the standard tableau T is -semistandardizable
if and only if D(T ) contains the union of the intervals [1, 1 ), [1 + 1, 2 ), . . . . Define
the -semistandardization rst (T ) of T -semistandardizable to be the element of
RST (, k, ) formed by replacing the numbers in [1, 1 ] in T by 1s, the numbers in
[1 + 1, 1 + 2 ] in T by 2s, and so on.
Example 4.3. The tableau T SY T ((3, 2, 2)) shown below is not (1, 2, 3, 1)-semistandardizable
because the 3 occurs in a higher row than the 2 and therefore rst(1,2,3,1) (T ) is undefined. Notice, however, that the tableau obtained by replacing the 3 in T with a 2,
the 4, 5, and 6 with 3s, and the 7 with a 4 is in fact row strict.
1 3 4
T = 2 5
6 7
On the other hand, the tableau U SY T ((3, 2, 2)) shown is (1, 2, 3, 1)-semistandardizable
and its (1, 2, 3, 1)-semistandardization is shown.
1 2 3
1 2 4
rst(1,2,3,1) (U) = 2 3
U= 3 5
3 4
6 7
Our terminology here may be slightly misleading since semistandard tableaux are
usually defined in the literature to be column strict rather than row strict. This
difference is cosmetic, though, and this notion of semistandardization will allow for
cleaner statements of our results.
The following is immediate.
Lemma 4.3. Let T SY T () be -semistandardizable. We have that std(rst (T )) =
T . Moreover, if U RST (, k, ), then std(U) is -standardizable.
Therefore, for any composition |= n, standardization injects RST (, k, ) into
SY T () and -semistandardization gives a bijection between the -semistandardizable
elements of SY T () and RST (, k, ). The combinatorial notion of semistandardizable tableaux can be related to the vanishing of KL immanants as follows. This will
be a key point in our construction of modules over the general linear group.
Lemma 4.4. Let U, T SY T () and let k N. We have that Imm(T,U ) (x,1n ) = 0
if and only if U is not -semistandardizable. Moreover, the set {Imm(T ,U ) (x ,1n )}
ranging over all possible |= n with () k, U , T SY T (), and U that are
-semistandardizable, is linearly independent.

32

BRENDON RHOADES

Proof. Let S denote the parabolic subgroup of Sn which setwise fixes the intervals
[1, 1 ], [1 + 1, 1 + 2 ], and so on. By Skanderas results on the dual canonical basis
[27], we have that Imm(T,U ) (x,1n ) 6= 0 if and only if the inverse image w of (T, U)
under RSK is a Bruhat maximal element of some left coset of S in Sn . It is easy to
show using the properties of jeu-de-taquin and RSK that this happens if and only if U
is -semistandardizable. The claim about linear independence follows from Theorem
4.1.

Example 4.4. As we have seen in Example 4.1, we have that Imm213 (x(2,1),(1,1,1) ) is
nonzero and Imm231 (x(2,1),(1,1,1) ) = 0. It is easy to see that the permutation 213 and
231 row insert as follows:


1 3
1 3
213 7
2
, 2


1 3
1 2
.
231 7
2
, 3
Therefore, the recording tableau for 213 is (2, 1)-semistandardizable whereas the
recording tableau for 231 is not. The nonvanishing and vanishing of the associated
polynomials is therefore predicted by Lemma 4.4.
We are finally ready to define our general linear group modules. For any T

SY T (), define VT,k

to be the space
M MM
MM

VT,k
:=
C{Imm(T,U ) (x,1n )
C{Imm(P,Q)(x ,1n )},

<dom

P,Q

where the first set ranges over all compositions |= n such that () = k and
all U SY T () which are -semistandardizable and the second set ranges over all
compositions |= n with ( ) = k and all pairs of tableaux P, Q SY T (). We

first show that VT,k

is closed under the action of GLk (C).

Lemma 4.5. VT,k

is a left GLk (C)-module.

Proof. For 1 i < k and z C, let Ei,z (resp. Fi,z ) denote the elementary matrices
in GLk (C) which are obtained by replacing the 0 in position (i, i + 1) (resp. (i + 1, i))

with a z. We show that VT,k

is closed under the action of the permutation matrices
Sk GLk (C), the Cartan subgroup H of diagonal matrices in GLk (C), and Ei,z and
Fi,z for 1 i < k and z C. Since GLk (C) is generated by these matrices, the result
will follow.

To show that VT,k

is closed under the action of Sk , let 1 i < k. We show that VT,k

. If
is closed under the action of si Sk . Let Imm(P,Q) (x,1n ) be a basis element of VT,k
we let si be the composition given by si := (1 , . . . , i1 , i+1 , i , i+2 , . . . , k ) |=
n, it follows that the polynomial si Imm(P,Q) (x,1n ) is equal to the image of w

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

33

Imm(P,Q) (x1n ,1n ) under the C-algebra homomorphism xi,j 7 x (i),j . Here w is the
permutation in Sn which is obtained by replacing the square diagonal submatrix of
size |i + i+1 | in the n n identity matrix corresponding to rows and columns in
(1 + + i1 , 1 + + i+1 ] with an antidiagonal matrix of 1s. It follows from
Theorem 2.7 and the analogous actions of Equations 2.10 and 4.4 that this latter

polynomial remains in VT,k

, as desired.

Given any diagonal matrix A = diag(a1 , . . . , ak ) in H, it is easily seen that the

immanant Imm(P,Q) (x,1n ) is an eigenvector for the operator A with eigenvalue a :=
(1)
(n)

a1 ak . Thus, VT,k
is closed under the action H.

Finally, we show that VT,k

is closed under the action of the Ei,z and Fi,z . To
simplify notation, we show this only for the case of nonzero z and Fi,z 1 . The other
cases are similar. Let = (1 , . . . , k ) be a composition of n of length k and let
w be a permutation in Sn with w 7 (P, Q). Let be the shape of P and Q. Let
I := { [n] | () = i}. The image of Imm(P,Q) (x,1n ) under the action of Fi,z 1 is
the polynomial
Y
Y
X
(1)(v,w) Pwo v,wo w (1) (xiv() + zx(i+1)v() )
x(m)v(m) .
vSn

m[n]I

Expanding out the terms in parenthesis and regrouping, we see that the above expression is equal to
Y
Y
Y
X
X
(1)(v,w) Pwo v,wo w (1)
x(i+1)v(j)
xiv()
x(m),v(m) .
z |J|
JI

jJ

vSn

IJ

m[n]I

Fix a subset J I. Let y be any permutation in Sn which fixes every letter not in I
and rearranges the letters of I so that the letters in J are mapped into a contiguous
suffix. Let |= n be the composition of n defined by
:= (1 , . . . , i1 , i |J|, i+1 + |J|, i+1 , . . . , k ).
By the discussion following Lemma 4.2, the image of
Y
Y
Y
X
(1)(v,w) Pwo v,wo w (1)
x(i+1)v(j)
xiv()
x(m),v(m)
vSn

jJ

IJ

m[n]I

under the action of y 1 is a complex linear combination of terms of the form

Imm(P ,Q ) (x,1n ), where either P and Q are both standard tableaux of shape
and P = P or P and Q are both standard tableaux of shape strictly dominating .

Since the subset J was arbitrary and since we already know that VT,k
is closed under

the action of Sk , it follows that VT,k is stable under the action of Fi,z 1 , as desired. 

34

BRENDON RHOADES

The above result can be found in [7] together with [27]. At any rate, we define VT,k
to be the quotient GLk (C)-module given by
!
M MM

VT,k := VT,k
/
C{Imm(P,Q) (x ,1n )} ,
>dom

P,Q

where the second direct sum ranges over compositions |= n with ( ) = k and
the third direct sum ranges over all standard tableaux P and Q of shape . Basis
elements of VT,k are given by the images of the polynomials Imm(T,U ) (x,1n ) for compositions = (1 , . . . , k ) |= n and -semistandardizable U. The image of the above
polynomial in VT,k shall be abbreviated I (U ), where U is the unique element of
RST (, k, ) such that std(U ) = U.
Theorem 4.6. The GLk (C)-module VT,k is isomorphic to the dual of the irreducible
finite dimensional GLk (C)-module with highest weight , where is the shape of T .
Moreover, for any |= n with () = k, the weight space of VT,k corresponding to
is equal to the C-linear span of {I (U)}, where U ranges over RST (, k, ).
Proof. We compute the Weyl character of VT,k . Let h := diag(a1 , . . . , ak ) be an element of the Cartan subgroup of GLk (C) for some nonzero complex numbers a1 , . . . , ak .
It is easy to see that the action of h on some basis element I (U) for U row strict
with content is given by
1
k
I (U).
h I (U) = a
a
1
k

(Here we recall that our action of GLk (C) came from the contragredient action.) Since
the set of all I (U) where U ranges over all row strict tableaux of shape and entries
k forms a basis for VT,k , it follows that h acts on VT,k with trace
X
X
1
k
a
a
.
1
k
|=n,()=k U RST (,k,)

1
This latter sum is the combinatorial definition of the Schur function s (a1
1 , . . . , ak ),
which is the Weyl character of the dual of the irreducible finite dimensional GLk (C)module with highest weight .

The claim about the weight space decomposition of VT,k is obvious.

Example 4.5. We illustrate our construction for the case k = 3, n = 4, and =

(2, 2). Let T be the standard tableau of shape (2, 2) given by
T =

1 2
3 4.

The space VT,3

is equal to the complex span of

|RST ((2, 2), 3)|+|RST ((2, 1, 1), 3)| |SY T (2, 1, 1)|+
|RST ((1, 1, 1, 1), 3)| |SY T (1, 1, 1, 1)| = 6 + 15 3 + 35 1 = 86

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

35

linearly independent polynomials which are obtained as above by applying KL immanants corresponding to permutations in S4 which row insert to any of the shapes
(2, 2), (2, 1, 1), or (1, 1, 1, 1) to 4 4 matrices in the 9 variables x11 , . . . , x33 with
possibly repeated rows. Also as explained above, we only consider KL immanants
corresponding to the permutations with shape (2, 2) whose insertion tableaux are

equal to T . By Lemma 4.5, the 86 dimensional space VT,3

is closed under the action
of GL3 (C).

The space VT,3 is obtained from VT,3

by modding out by the complex span of all polynomials obtained by the application of KL immanants whose associated permutations
do not have shape (2, 2) under RSK. The space VT,3 has dimension |RST ((2, 2), 3)| =
6. A complex basis of VT,3 is the set {I (U)} as defined above, where U is a row strict
tableau of shape (2, 2) and content composition satisfying () = 3. We write down
an element of this basis explicitly.

First note that the permutations in S4 with insertion tableau T are precisely 3412
and 3142. It can be shown that the KL immanants corresponding to these permutations are
Imm3412 (x) = x13 x24 x31 x42 x14 x23 x31 x42 x13 x24 x32 x41 + x14 x23 x32 x41
and
Imm3142 (x) = x13 x21 x34 x42 x13 x22 x34 x41 x13 x24 x31 x42 x14 x21 x33 x42
+ x14 x22 x33 x41 + x13 x24 x32 x41 + x14 x23 x31 x42 x14 x23 x32 x41 .
Let To be the tableau given by
To =

1 3
2 4

The set RST ((2, 2), 3) is equal to

{rst(1,2,1) (T ), rst(2,2,0) (To ), rst(2,0,2) (To ), rst(0,2,2) (To ), rst(1,1,2) (To ), rst(2,1,1) (To )}.
To find the basis element I(1,2,1) (rst(1,2,1) (T )), we apply the 3412-KL immanant to the
matrix x(1,2,1),(1,1,1,1) . This results in
Imm3412 (x(1,2,1),(1,1,1,1) ) = x13 x24 x21 x32 x14 x23 x21 x32 x13 x24 x22 x31 + x14 x23 x22 x31

and I(1,2,1) (rst(1,2,1) (T )) is the homomorphic image of this in the quotient VT,3
.

5. Promotion on Column Strict Tableaux

The goal in this section is to prove a CSP for column strict tableau using the description of irreducible GLk (C)-modules given in the last section. This will essentially
involve showing that the long cycle (1, 2, . . . , k) Sk GLk (C) acts as a monomial
matrix corresponding to promotion in the rectangular irreducible representations constructed in Section 4.

36

BRENDON RHOADES

We start by defining a collection of epimorphisms which will allow us to work with

our representations more easily. Given |= n, we define an epimorphism C-algebras
: C[xij ]1i,jn C[x(i)j ]1i,jn by the formula (xij ) = x(i)j . The relevant
vanishing properties of KL immanants may be stated cleanly in terms of the maps
.
Lemma 5.1. Let U, T SY T (). We have that (Imm(U,T ) (x)) is nonzero if and
only if U is -semistandardizable, in which case (Imm(U,T ) (x)) = Imm(U,T ) (x,1n ).
Proof. This is straightforward from the definition of the KL immanants and together with Lemma 4.4.

We record the action of the long cycle in Sn on immanants (which correspond to
standard tableaux). This will be related to the action of the long cycle in Sk on
semistandard tableaux via the maps .
Lemma 5.2. Let = ba be a rectangle and let P SY T (). Let cn = (1, 2, . . . n)
Sn be the long cycle. We have that cn I1n (P ) = (1)b1 I1n (j(P )).
Proof. This is straightforward using the corresponding result for the KL basis of C[Sn ]
(Proposition 3.5) and the action of Sn on immanants given in Lemma 4.2.

We also record how the combinatorial operations of promotion and standardization
commute.
Lemma 5.3. Let n be a rectangle and = (1 , . . . , k ) |= n. We have the
following equality of operators on SY T ():
(5.1)

j rst = rstck j k ,

where ck is the composition of n of length k given by ck := (k , 1 , , k1 )

and the right hand side is defined if and only if the left hand side is defined.
Proof. Observe that for any tableau P SY T (), i + 1 occurs strictly south and
weakly west of i in P if and only if i D(P ). For i < n this is equivalent to the
condition i De (P ). Therefore, this follows from the definition of jeu-de-taquin as
well as the cyclic action of jeu-de-taquin on the extended descent set of rectangular
tableaux from Lemma 3.3.

The following lemma shows how the allow the transfer of information from the
standard to the semistandard case.
Lemma 5.4. Let cn = (1, 2, . . . , n) be the long cycle in Sn and let ck = (1, 2, . . . , k)
be the long cycle in Sk . For = (1 , . . . , k ) |= n we have a left action of cn on

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

37

C[xij ]1i,jn and d maps C[x(i)j ]1i,jn into C[xck (i)j ]1i,jn . We have the following
commutative square.
cnk

(5.2)

C[xij ]1i,jn
C[xij ]1i,jn

ck
C[x(i)j ]1i,jn C[xck (i)j ]1i,jn
ck

Proof. The commutativity of this diagram is easily checked on the generators xij of
the algebra in the upper left.

Recall that in the last section we constructed representations VT,k of GLk (C) for a
fixed standard tableau T with n boxes. Let VT,k, be the subspace of VT,k generated by
the elements {I (U)}, where U ranges over RST (, k, ). By Theorem 4.6, the spaces
VT,k, give the weight space decomposition of the irreducible representation VT,k . In
terms of this weight space decomposition, the commutative square (5.2) implies the
following commutative square:
cnk

(5.3)

VT,n,1n
VT,n,1n

ck .
VT,k, VT,k,ck
ck

Using this square, we can relate the action of ck on VT,k to promotion.

Proposition 5.5. Given U RST (, k, ) for = ba rectangular and = (1 , . . . , k ) |=
n, we have that
(5.4)

ck I (U) = (1)k (b1) Ick (j(U)).

Proof. We have the following chain of equalities:

ck I (U) = ck cnk I1n (std(U))
= ck ((1)k (b1) Ick (rstck j k std(U))
= (1)k (b1) Ick (j(U)).
The first equality comes from the commutative square (5.3) and Lemma 5.2. The
second comes from Lemmas 5.2 and 5.3. The third is again Lemma 5.3 and the
definition of ck .

Corollary 5.6. For rectangular, the order of j on RST (, k) (or CST (, k)) is
equal to k unless consists of a single column (respectively, single row) and k = ||.

38

BRENDON RHOADES

Proof. By Proposition 5.5, every element of RST (, k) is fixed by the operator j k . If

is any rectangular shape other than a column (provided RST (, k) is nonempty),
its easy to produce a row strict tableau of content composition having no nontrivial
cyclic symmetry.

Example 5.1. If = (2, 2) and k = 3, the set CST ((2, 2), 3) contains 6 elements
and promotion acts as the following permutation:


1 1
2 2
1 1
2 2 , 3 3 , 3 3
which does indeed have order k = 3.



1 2
1 2
1 1
2 3 , 3 3 , 2 3

Finally we are able to prove Theorem 1.4. As in the standard tableau case of
Theorem 1.3, the hypothesis that is rectangular is necessary. For arbitrary shapes
, the order of the operator j on CST (, k) is unknown. Also as with the standard
case, even if we removed the rectangular condition and replaced C by the cyclic group
of size the same as the order of j on CST (, k) (still taking X(q) to be the principal
specialization of s ), this result would not be true.
Example 5.2. Keeping with the earlier example of = (2, 2) and k = 3, we compute
that ((2, 2)) = 2 and s(2,2) (x1 , x2 , x3 ) = x21 x22 +x22 x23 +x21 x23 +x1 x22 x3 +x1 x2 x23 +x21 x2 x3 .
Maintaining the notation of Theorem 5.7, we therefore have that X(q) = 1 + q + 2q 2 +
2i
q 3 + q 4 . Letting = e 3 , we see that
X(1) = 6 X() = 0 X( 2 ) = 0.
These numbers agree with the fixed point set sizes:
2

|X 1 | = 6 |X j | = 0 |X j | = 0,
as predicted by Theorem 1.4.

Proof. (of Theorem 1.4) We prove the equivalent assertion which is obtained by re
placing X by RST (, k) and X(q) by q ( ) s (1, q, q 2, . . . , q k1). We fix a standard
tableau T of shape and consider the action of (1, 2, . . . , k) on VT,k .
2i

Let = e k C. Suppose U RST (, k, ), = (1 , . . . , k ) satisfies cm

k I (U) =
I (U). Since ck maps elements of the form I(1 ,...,k ) (P ) for P RST (, k, (1 , . . . , k ))
to elements of the form I(k ,1 ,...,k1 ) (Q) for Q RST (, k, (k , 1 , . . . , k1)), it follows that i j (mod m) implies that i = j . Using the facts that ( ) = ab(b1)
2

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

39

and 1 + + k = n = ab we get that

( m )( ) = (e

2i
k

mab(b1)
2

= (e k )mn(b1)
= (ei )(1 ++m )(b1)
= (1)(1 ++m )(b1) .
However, we already know that
(1 ++m )(b1)
cm
I (j m (U)).
k I (U) = (1)

It follows that for any r 0, the coefficient of I (U) in crk I (U) is r( ) if j r (U) = U
and 0 otherwise. Therefore, the trace of the operator crk on the space VT,k is equal to

r
r( ) |X j |.
On the other hand, the operator crk is conjugate to diag(1, r , (2r) , . . . , (k1)r )
in GLk (C). It is easy to see that if P is row strict with content , then I (P )
is an eigenvector for the latter operator with eigenvalue (1r )1 ( r )2 ( 2r )3 .
It follows that the trace of crk on V is the specialization of the Schur function

s (1, q, q 2, . . . , q k1) at q = r . The desired CSP follows.
As a pair of corollaries to Theorem 1.4, we can prove Reiner-Stanton-Whites results
from the introduction quite easily.
Proof. (of Theorem 1.1) Interchange n and k in Theorem 1.4 and take to be a single
column.

Proof. (of Theorem 1.2) Interchange n and k in Theorem 1.4 and take to be a single
row.

6. Promotion on Column Strict Tableaux with Fixed Content
In this section we fix a rectangular partition = ba with ab = n, a positive integer
k and a composition |= n with () = k such that has some cyclic symmetry.
We prove a near-CSP involving the action of certain powers of promotion on the set
CST (, k, ) and the Kostka-Foulkes polynomials. In representation theoretic terms,
this corresponds to a weight space refinement of our results in the last section.
Assume that for some integer d|k, we have that the composition has cyclic symmetry of order d. That is, i = j whenever i j (mod d). Since j maps the set
CST (, k, (1, . . . , k )) into the set CST (, k, (k , 1 , 2 , . . . , k1)), we have that
the dth power j d of j maps the set CST (, k, ) into itself. Note that for the special

40

BRENDON RHOADES

case d = 1, k = n, and = 1n this is the statement that j acts on the set SY T ()

of standard tableaux of shape . Since j acts with order k on the set CST (, k), we
have that j d generates an action of the cyclic group Z/( kd Z) on CST (, k, ).
For a partition n and a composition |= n, let K, (q) N[q] be the associated
Kostka-Foulkes polynomial. The Kostka-Foulkes polynomials are q-analogues of the
Kostka numbers K, which enumerate the number of column strict tableaux with
shape and content . In particular K, (1) = K, always. Moreover, we have
the polynomial equality K, (q) = K, (q) for any rearrangement of the composition . The Kostka-Foulkes polynomials are the generating function for the charge
statistic on tableaux and are also the coefficients of the change of basis matrix from
Schur functions to Hall-Littlewood symmetric functions. For more details on these
polynomials, see [16]. Up to a power of q, the Kostka-Foulkes polynomials will play
the role of X(q) in our CSP.
It should be noted that the Kostka-Foulkes polynomials have an interesting representation theoretic interpretation. Let g denote the simple Lie algebra slk (C) and let
g
= n h n+ denote the Cartan decomposition of g. For any partition n, let

V denote the irreducible representation of g indexed by . The module V has a

weight space decomposition V
= V , where the are elements of the dual algebra
h . Let X be a generic element of the nilpotent subalgebra n+ of g generated by the
positive roots. For j = 0, 1, 2, . . . , let Vj denote the subspace of the weight space V
which is killed by X j , so that V0 V1 V2 . . . . Since X is an element of the
positive subalgebra, some sufficiently high power of X must carry every element of V
outside of the weight support for the representation V , so this filtration terminates
in V for large j. This filtration of the weight space V is called the Brylinski-Kostant
(BK) filtration. It turns out that the Kostka-Foulkes polynomial K, (q) is equal to
the associated jump polynomial ([6], [13]), that is,
X
(6.1)
K, (q) =
dim(Vj+1/Vj )q j .
j0

To get our fixed point result, we will need some information about the evaluation
of Kostka-Foulkes polynomials at roots of unity. Lascoux, Leclerc, and Thibon [16]
have interpreted these evaluations in terms of ribbon tableaux. We define the relevant
combinatorial objects.
For a positive integer m, an m-ribbon is a connected skew shape with m boxes which
contains no 2 by 2 squares. The southwesternmost box of an m-ribbon is called the
head of the m-ribbon and the northeasternmost box of an m-ribbon is called the tail
of the m-ribbon.
For any skew partition /, an m-ribbon tableau of shape / is a tiling of the
diagram of / by m-ribbons with a number attached to every m-ribbon in the tiling.
Observe that the set of m-ribbon tableaux of shape is empty unless m divides |/|.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

41

The content of an m-ribbon tableau T of shape |/| is the composition of |/|/m

given by (number of 1s in T , number of 2s in T , . . . ). If an m-ribbon tableau T of
shape / exists, the sign m (/) of / is the number (1)h (T ), where h(T ) is the
sum of the heights of the ribbons in T . It can be shown that the sign of a skew shape
/ is independent of the tableau T chosen.
An m-ribbon tableau T of shape / is said to be column strict if for any ribbon R
in T with label, j the head of R does not lie to the right of a ribbon with label i > j
and the tail of R does not lie below a ribbon with label i j. Observe that for m = 1
this definition reduces to the ordinary definition of column strict tableaux. For any
m
composition |/|
, let K/,
denote the number of column strict ribbon tableaux of
m
shape / and content .
Say that a skew shape / is a horizontal m-ribbon strip if there exists a column
strict m-ribbon tableau of shape / in which every ribbon has the same label. If /
is a horizontal m-ribbon strip, it is easy to see that there exists a unique such column
strict ribbon tableau (for fixed choice of label). The following result is mentioned on
page 12 of [16].
Theorem 6.1. (Lascoux-Leclerc-Thibon [16]) Let n be a partition and |= n be
a composition. For d|n, let C be a root of unity of order d. If the multiplicity of
any part of is not divisible by d, then K, () = 0. If the multiplicity of every part
of is divisible by d, then the modulus |K, ()| is equal to the number of column
strict d-ribbon tableaux with content
. Here
is any composition of nd whose part
multiplicities are all d1 times the part multiplicities of .
We will also need a minor lemma which relates Schur function specialization to
ribbon tableau enumeration. For this lemma there is no need for our rectangular
shape hypothesis.
Lemma 6.2. Let n be a partition of arbitrary shape. Let k > 0 be a positive
integer and let d|k. Let = e2i/k be a primitive k th root of unity and let a1 , . . . , ad
be arbitrary complex numbers.
We have that
(6.2)

( d)() s (a1 , da1 , . . . , kda1 , . . . , ad , . . . , kdad ) =

K,d a1d

add d ,

where the sum on the right hand side is over all compositions |=

nd
.
k

The domino case d = k2 of the above lemma was known, for example, to Stembridge [35]. We also remark that the right hand side of the above identity can
be interpreted as a product of specializations of Schur functions. In particular let
k
((1) , (2) , . . . , ( d ) ) be the kd -quotient of the partition and let be the kd -core of .

42

BRENDON RHOADES

By the Stanton-White correspondence [33] there exists a content preserving bijection

from column strict kd -ribbon tableaux of skew shape / and kd -tuples of ordinary
k
column strict tableaux of shapes ((1) , (2) , . . . , ( d ) ). From this bijection it is fairly
easy to derive an alternative formulation of Lemma 6.2. Namely, the expression
s (a1 , da1 , . . . , kda1 , . . . , ad , . . . , kdad ) is equal to 0 unless has empty kd -core
and, if does have empty kd -core, we have the identity
k

(6.3)

d
s(i) (a1 , a2 , | . . . , ad )|.
|s (a1 , da1 , . . . , kda1 , . . . , ad , . . . , | kdad )| = |i=1

Proof. We show that both sides of this equation satisfy the same recursion.
For the left hand side of the above equation, observe that in any column strict
tableaux in CST (, k), the last kd letters {k d + 1, k d + 2, k} of [k] must
appear in a skew shape flush with the exterior of the shape . This gives a partition
of CST (, k) according to this skew shape /.
We consider the skew Schur function specialization s/ (1, d, . . . , kd) for the various skew shapes /. By a result of Lascoux, Leclerc, and Thibon [16], we have
that the above is equal to 0 if / is not a horizontal kd -ribbon strip and is equal to
k (/) if / is a horizontal kd -ribbon strip.
d

For the right hand side, given any column strict kd -ribbon tableau T of content
= (1 , . . . , d ), we have that the restriction of T to the number d is a horizontal kd ribbon strip of skew shape / for some normal shape . On the other hand, given
any normal shape such that / is a horizontal kd -ribbon strip of with k ribbons
and any column strict kd -ribbon tableau To of shape and content (1 , 2 , . . . , g
k1 ),
k
we have that To extends uniquely to a column strict d -ribbon tableau T of shape and
It follows that both sides of the above equation satisfy the same recursion
content .
relation and, thus, the above equation holds for arbitrary partitions . Also, we have
that both sides of the above equation are equal to zero when is not a kd -ribbon
tableau.

Proof. (of Theorem 1.5)
d
For any composition |= n of length k, let N,
denote the number of column
strict tableaux of shape and content which are fixed by the operator j d . Observe
d
that N,
is nonzero only if i = j whenever i j (mod d).

Let ck = (1, 2, . . . , k) denote the long cycle in Sk GLk (C). Let T be an arbitrary
standard tableau of shape and let VT,k be the irreducible GLk (C-representation
constructed in Section 4. Identify the basis elements of VT,K with symbols I (P ) for
compositions of n of length k and column strict tableaux P of shape and content

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

43

in the obvious way. By Theorem 4.6 have a weight space decomposition

M
VT,k, ,
VT,k
=

where VT,k, is the C-linear subspace of VT,k spanned by the vectors I (P ) for P
CST (, k, ). Recall that each of the subspaces VT,k, is stabilized by the Cartan
subgroup H of diagonal matrices in GLk (C) and the action of ck maps the space
VT,k, into the space VT,k,ck .
2i

Let = e k and let a1 , . . . , ad C be any nonzero complex numbers. From the

discussion in the above paragraph and Theorem 4.6, the action of the composition of
cdk with an element of the Cartan subgroup yields the following character evaluation.
(6.4)

k ()()s (a1 , da1 , . . . , kda1 , . . . , ad , . . . , kdad ) =

(6.5)

k ()()[s (x1 , . . . , xk )]xi = di ai/d =

d
X
k
k
1

d
N,
a1d add d ,

(6.6)

where the sum on the lattermost expression ranges over all compositions of n of
length k such that i = j whenever i j (mod d). By Lemma 6.2 we can interpret
the first expression in the above sequence of equalities as follows.
(6.7)

d
N,
a1d

add

K,d a1d

add d ,

where the sum on the left hand side ranges over all compositions of n having
length k such that i = j whenever i j (mod d) and the sum on the right hand
side ranges over all compositions of nd
having length d. Choosing the complex
k
numbers a1 , . . . , ad to be algebraically independent over Q, we can view this latter
equation as a polynomial identity. It follows that if is any composition of n of
length k satisfying i = j whenever i j (mod d) and if we set := (1 , 2 , . . . , d ),
we have the equality
k

d
N,
= K,d .

It remains to relate the right hand side of the above equation to the evaluation of a
Kostka-Foulkes polynomial at an appropriate root of unity. This is Theorem 6.1. 
Remark 6.1. It would be desirable to strengthen Theorem 1.5 to a genuine CSP.
One obvious way of doing this would be to find an expression f (, ) potentially

44

BRENDON RHOADES

depending on both the rectangular partition and the composition such that the
triple (CST (, k, ), hj di, q f (,) K, (q)) exhibits the CSP. That is, our polynomial
X(q) would just be a qshift of a Kostka-Foulkes polynomial. By Lascoux-LeclercThibon we know that the evaluation of the Kostka-Foulkes polynomial K, (q) itself
at the relevant roots of unity is plus or minus the correct positive number, but we do
not know of an expression f (, ) which takes this sign into account.
7. Dihedral Actions
The cyclic sieving phenomenon is concerned with the enumeration of the fixed point
set sizes when we are given an action on a finite set X by a finite cyclic group C.
However, there are many interesting combinatorial actions on finite sets where the
group in question is not cyclic. In general, given a finite group G acting on a finite set
X, we can ask for the fixed point set sizes |X g | for all g G. For the case where G is
a product of two cyclic groups, Barcelo, Reiner, and Stanton have proven an instance
of a bicyclic sieving phenomenon involving complex reflection groups where the fixed
point set sizes involved are obtained by evaluating a polynomial X(q, t) Q[q, t] in
two variables at two appropriate roots of unity. In our case, we are interested in
actions of dihedral groups D generated by promotion and evacuation on either of the
sets SY T () or CST (, k), where is a fixed rectangular shape and k 0. While it
is not yet known how to interpret the associated fixed point set sizes as evaluations
of a naturally associated polynomial, we hope that this work will give motivation to
the study of a dihedral sieving phenomenon. Some work regarding dihedral actions
is in the REU report [1].
Let = ba be a rectangular partition with ab = n and let k 0. It is possible to
show that we have eje = j 1 as operators on CST (, k). Moreover, we know that
the order of the operator j is equal to k while the order of the operator e is equal to
2. This implies that the group he, ji generated by e and j, considered as a subgroup
of the symmetric group SCST (,k), is dihedral of order 2k. Similar remarks hold for
the action of e and j on RST (, k) and SY T ().
We have already determined that the action of j on the above sets of tableaux is
modeled by the action of the long cycle of an appropriate symmetric group on an
appropriate module. By Stembridges result, the action of e can be modeled by the
action of the long element wo . So, corresponding to the dihedral group acting on
the sets CST (, k), RST (, k), and SY T (), we get an isomorphic dihedral group
generated by the long cycle and wo sitting inside Sk , Sk , and Sn , respectively. Our
first observation gives the cycle types of the operators wo and wo cn in the symmetric
group Sn .
Lemma 7.1. Let wo Sn be the long element and let cn Sn be the long cycle
n1
n
(1, 2, . . . , n). The permutation wo has cycle type 2 2 when n is even and 2 2 1 when

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

45

Figure 7.1. Conjugacy classes of reflections in dihedral groups

n

n is odd. The permutation wo cn has cycle type 2( 2 1) 12 when n is even and 2

when n is odd.

n1
2

Proof. This is straightforward from the definition of permutation multiplication. Looking at the dihedral group generated by cn and wo , this essentially asserts that if n
is odd, any element of the unique conjugacy class of reflections within the symmetry
group of an n-gon fixes one vertex and if n is even there are two conjugacy classes of
reflections fixing zero or two vertices. See Figure 7.1.

This simple lemma about permutation multiplication allows us to pin down, including sign, the action of the operators wo and wo cn on the cellular representation.
As in Section 3, given a rectangular partition = ba , identify the basis of the KL
cellular representation of Sn corresponding to with SY T ().
Proposition 7.2. Let = ba be a rectangle and let be the KL cellular representation
corresponding to . We have that
(
e(P )
if b is even
(wo )P =
(7.1)
a2
(1) e(P ) if b is odd

if b is even
e(j(P ))
( a2 1)
(wo cn )P = (1)
(7.2)
e(j(P )) if b is odd and a is even

(1) a2 e(j(P ))
if b, a are odd.
Proof. Recall that the KL cellular representation corresponding to a shape is an
irreducible Sn module of shape . By Theorem 2.10 and Proposition 3.5, we need
only verify that the signs in these equations are correct. To do this we will use the

46

BRENDON RHOADES

Murnaghan-Nakayama rule (see, for example, [25]). For any of the 3 cycle types
which appear in Lemma 7.1, there exists a corresponding removal of rim hooks from
the corresponding rectangle with the appropriate sign.
In particular, if b is even, we can fill the diagram of with n2 rim hooks of shape
2 1 in the case of wo and ( n2 1) rim hooks of shape 2 1 together with two rim
hooks of shape 1 1 in the case of wo c. In either case, the product (1)h() over all
the rim hooks in either of these fillings, where h() is the height of , is equal to 1.
If b is odd and a is even for the case of wo , we can fill the diagram of with na
2
rim hooks of shape 2 1 in the left (b 1) a box and a2 rim hooks of shape 1 2
a
in the first column. The sign of this filling is equal to (1) 2 . In the case of wo c with
the same parity conditions on a and b, we can replace the upper left 1 2 rectangle
a
in the filling with two 1 1 rim hooks, and the resulting filling has sign (1)( 2 1) .
If b and a are both odd, in the case of wo and wo cn we can fill the diagram of
with na
rim hooks of shape 2 1 in the left (b 1) a box, a2 rim hooks of shape
2
1 2 in the lower portion of the first column of , and one 1 1 rim hook in the
a
upper left hand corner of . The sign of this filling is (1) 2 .
It is possible to show that the sign of all rim hook fillings of corresponding to the
conjugacy class of either wo or wo cn is the same, either 1 or 1. The desired result
now follows from the Murnaghan-Nakayama rule.

Observe that, in this case, we were able to use the Murnaghan-Nakayama rule to
deduce the sign of our permutation matrices relatively easily. This is in contrast to
the proof of Lemma 3.4, where we needed to use special facts about the KL cellular
representation. In fact, it is possible to use the Murnaghan-Nakayama rule in a similar
way to derive Lemma 3.4 in the cases where both a and b are not odd, but for this
particular case more specific facts are needed.
We already know that the number of points fixed by the action of a given power of
j on standard tableaux is given by the evaluation of the q-hook length formula at an
appropriate root of unity. The previous lemma allows us to give the number of fixed
points of the operators e and ej on standard tableaux as a character evaluation.
Proposition 7.3. Let = ba be a rectangular partition of n, let c be the long cycle
in Sn , and let : Sn C be the irreducible character of Sn corresponding to . Let
he, ji be the dihedral subgroup of SSY T () generated by e and j and let : hwo, cn i
he, ji be the group epimorphism defined by (wo ) = e and (c) = j. For any g he, ji
and any w such that (w) = g one has
|SY T ()g | = (w).

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

More precisely, we have that

(
(wo )
if b
|SY T ()e | =
(7.3)
a2
(1) (wo ) if b

(wo cn )
a
(7.4)
|SY T ()ej | = (1)( 2 1) (wo cn )

(1) a2 (w c )
o n

47

is even
is odd
if b is even
if b is odd and a is even
if b and a are odd.

Proof. This follows easily from the above lemma.

Our next task is to extend these results to facts about the dual canonical basis. Let
be a rectangular partition and let k 0. By the results of Section 4, there exists
a basis {I (P )} for the GLk (C)-module VT,k for T SY T () fixed, where ranges
over all compositions of n of length k and P ranges over all elements of RST (, k, ).
For any choice of and P and any diagonal matrix g = diag(x1 , . . . , xk ) GL(Ck ),
1 2
the vector I (P ) is an eigenvector for g with eigenvalue x
x2 . We first show
1
how multiplication by the long elements interacts with the projection maps .
Lemma 7.4. Let be a rectangular partition and fix T SY T (). Let won be the long
permutation in Sn and let wok be the long permutation in Sk . We have the following
commutative diagram, where = (1 , . . . , k ) |= n and wok is the composition of
n of length k given by (k , k1, . . . , 1 ).
won

(7.5)

VT,n,1n
VT,n,1n

wok
VT,k, VT,k,wok
wok

Proof. The commutativity of the corresponding diagram of polynomial rings is easily

verified by computing the image under the two possible compositions of the generators
xij , as before. The commutativity of the above diagram is induced from this other
commutative diagram.

Using this lemma, we obtain the action of the long element in Sk and its product
with the long cycle Sk on the basis elements I (P ), P RST (, k, ).
Lemma 7.5. Let wok be the long element in Sk , let ck be the long cycle (1, 2, . . . , k)
in Sk , and let = ba be a rectangular partition, and let P RST (, k, ). We have

48

BRENDON RHOADES

that

(
I(k ,...,1 ) (e(P ))
if a is even
wok I (P ) =
2b
(1) I(k ,...,1 ) (e(P )) if a is odd

(b1)

I(k1 ,k2 ,...,1 ,k ) (ej(P ))

if a is even
(1) k
b
(
1+
(b1))
k
wok ck I (P ) = (1) 2
I(k1 ,k2 ,...,1 ,k ) (ej(P )) if a is odd and b is even

(1)( 2b +k (b1)) I
if b, a are odd.
(k1 ,k2 ,...,1 ,k ) (ej(P ))

Proof. This combines previously proven results about the actions of the long element
and the long cycle.

Our results about fixed points have two flavors. When k is odd, j and ej are
conjugate in the dihedral group he, ji and hence have the same number of fixed points
as operators on CST (, k). However, when k is odd, e and ej are not conjugate, so
different numbers of fixed points are possible. Parts of the next theorem come from
conjectures and results of Abuzzahab, Korson, Li, and Meyer [1] and a theorem of
Stembridge [35].
Theorem 7.6. Let = ba be a rectangular partition of n and let k 0. Assume that
at k is odd. We have that
(7.6)
(7.7)

|CST (, k)e | = (1)() s (1, 1, 1, . . . , |(1)k1)

= |CST (, k)ej |.

On the other hand, if k is even, we have that

(7.8)
|CST (, k)e | = (1)() s (1, 1, 1, . . . , |(1)k1)
(7.9)

(1)() s (1, 1, . . . , (1)k3 , (1)k2, (1)k2 )

(1)() s (1, 1, . . . , (1)k1 )

|CST (, k)ej | =
b
1+()
2

(1)
s (1, 1, . . . , (1)k1)

(1) 2 +() s (1, 1, . . . , (1)k3 , (1)k2, (1)k2 )

a
a
a
a

and b even
even, b odd
odd, b even
and b odd.

Proof. The claims about the operator e are Theorem 3.1 of [35]. The claim about the
operator ej in the case where k is odd follows from the claim about the operator j
by conjugacy, as mentioned above. We are reduced to the claim about the operator
ej in the case where k is even. As before, we prove the equivalent statement for row
strict tableaux. This can be obtained by copying the above claim verbatim, replacing
every RST with a CST , every with a , and swapping every b for every a.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

49

Since k is even, we know that the operator wo ck as an element of GL(Ck ) has

eigenvalues 1 with multiplicity k2 + 1 and 1 with multiplicity k2 1. The claim now
follows by examining the signs in Lemma 5.4 with the appropriate parity conditions on
a and b in mind. In particular, multiplication by terms of the form (1)k corresponds
to swapping a sign of one of the arguments in the appropriate Schur function.

Observe that in the above column strict result one does not have that for any
g in the group of operators generated by evacuation and promotion the fixed point
set |CST (, k)g | is equal to plus or minus the character of a general linear group
element corresponding to g. Roughly speaking, the reason for this is that Equation
(5.4) implies that the long cycle acts as promotion on column strict tableaux up to a
sign which is not in general independent of weight space.
8. Applications to other Combinatorial Actions
For a positive integer n, let Bn denote the group of signed permutations of [n] having
order 2n n!. The group Bn is a Coxeter group with generators s0 , s1 , s2 , . . . , sn1 ,
where si for i > 0 is the adjacent transposition switching positions i and i + 1 but
preserving signs and s0 switches the sign of the element in the first position. Under
this identification, the long element of Bn is the signed permutation which sends 1
to 1, 2 to 2, . . . , and n to n. Since the long element woBn in Type Bn is the
scalar transformation 1, we have in particular that woBn si = si woBn for all i, so that
reduced expressions for the long element are mapped to other reduced expressions
for the long element under cyclic rotation. Our cyclic sieving results specialize in the
case of standard tableaux of the square shape nn to analyze this cyclic action.
Theorem 8.1. Let X be the set of reduced words for the long element in Bn . Let C =
n
Z/n2 Z act on X by cyclically rotating words. Let X(q) = f n (q) be the qhook length
formula. We have that the triple (X, C, X(q)) exhibits the cyclic sieving phenomenon.
Proof. By Theorem 1.3, we have a CSP involving the cyclic action of promotion on
the set SY T (nn ) and the same polynomial X(q). Let Y be the set of standard shifted
tableaux with n2 boxes and staircase shape (see, for example, [10], [26], or [38]).
Shifted jeu-de-taquin promotion gives a cyclic action on Y .
By a result of Haiman ([10], Proposition 8.11) we have a bijection between the
sets SY T (nn ) and Y under which ordinary promotion gets sent to shifted promotion.
Moreover, the type B Edelman-Greene correspondence gives a bijection between Y
and the set X under which shifted jeu-de-taquin promotion maps to cyclic rotation
of reduced words ([11], Theorem 5.12).

We can also apply our enumeration results for standard tableaux to prove cyclic
sieving phenomena for handshake patterns and noncrossing partitions. This gives

50

BRENDON RHOADES

a new proof of a result of White [37], as well as a new proof of results of Heitsch
[12] which have biological applications related to RNA secondary structure. The
case of handshake patterns gives a very explicit realization of the dihedral actions
investigated in Section 7.
Given n N, a handshake pattern of size 2n consists of a circle around which the
points 1, 2, . . . , 2n are written clockwise and a perfect matching on the set [2n] such
that, when drawn on the circle, none of the arcs in this matching intersect. This can
be thought of as a way in which the people labelled 1, 2, . . . , 2n can all shake hands so
that no one crosses arms. Let Hn denote the set of all handshake patterns of size 2n.
Handshake patterns can be identified with the basis elements for the Temperley-Lieb
algebra Tn () (see, for example, [23]).
For n N, a noncrossing partition of [n] is a set partition (P1 |P2 | . . . |Pk ) of the set
[n] so that whenever there are integers a, b, c, d, i, and j with 1 a < b < c < d n
and a, c Pi and b, d Pj , we must necessarily have that i = j. Noncrossing
partitions have a pictorial interpretation. Drawing the numbers 1, 2, . . . , n clockwise
around a circle, a partition of [n] is noncrossing if and only if, when the blocks of
are drawn on the circle, none of the regions intersect.
Recall that a poset L is a lattice if and only if given any pair of elements x, y L,
there exist unique least upper bounds and greatest lower bounds for x and y (denoted
x y and x y, respectively). If L is a finite lattice and x L, then an element
y L is said to be a complement to x if we have x y = 1 and also x y = 0.
Here 1 and 0 denote the unique maximal and minimal elements of L. A finite lattice
L is said to be complemented if every element has at least one complement. The
set NC(n) of all noncrossing partitions of [n] is a lattice with respect to the partial
order given by refinement (however, this lattice is not a sublattice of the lattice of all
partitions since the formulas for the least upper bounds and greatest lower bounds do
not agree). The lattice NC(n) is complemented, and given a noncrossing partition
of [n], Kreweras complementation [15] gives a way to produce a complement of .
Write the numbers 1, 1 , 2, 2, . . . , n, n clockwise around a circle. Draw the blocks of
on the numbers 1, 2, . . . , n. Now draw the unique maximal noncrossing partition
of the numbers 1 , 2 , . . . , n which does not intersect the blocks of . Call this new
partition . Kreweras showed that is a complement to in NC(n). Complements
in NC(n) are not in general unique and it is an open problem to determine how many
complements there are in NC(n) for a fixed noncrossing partition of [n].
Both of the

sets Hn and NC(n) have cardinality given by the Catalan number
2n
1
Cn = n+1
. Moreover, these sets both carry an action of the cyclic group - in the
n
case of Hn given by the action of rotating the table clockwise by one position and in
the case of NC(n) given by Kreweras complementation. As a corollary of our earlier
work, we get results of White [37] and Heitsch [12].

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

51

Theorem 8.2. Let n N and let C = Z/(2n)Z. Let X(q) be the q-Catalan number
 
1
2n
.
X(q) = Cn (q) =
[n + 1]q n q
Let X be either Hn equipped with the C-action of rotation or NC(n) equipped with
the C-action of Kreweras complementation.
Then, the triple (X, C, X(q)) exhibits the cyclic sieving phenomenon.
Proof. It is well known (see, e.g., Heitsch [12]) that there exists a bijection NC(n)
Hn under which Kreweras complementation corresponds to rotation, so we need only
prove this theorem in the case where X = Hn . To do this, we use a bijection Hn
SY T ((n, n)) which maps rotation to jeu-de-taquin promotion. It is well known that
the q-Catalan numbers are equal to the q-hook length formula in the special case of
a 2 by n partition, so then the result will follow from Theorem 1.3.
To construct this bijection, given a handshake pattern h Hn , fill the top row
of a 2 by n tableau T with the smaller members of each of the n handshake pairs
in h in increasing order left to right. Fill the bottom row of T with the smaller
members of each handshake pair in increasing order from left to right. Since h is a
handshake pattern, it follows that T is a standard tableau. It is easy to see that the
correspondence h 7 T defines a bijection Hn SY T ((n, n)).
White proved that, under the above bijection, rotation maps to promotion [37].
For completeness, we prove this again here.
Given a tableau T SY T (n, n), define the ascent set A(T ) of T to be the subset
of [2n 1] defined by i A(T ) if and only if i + 1 occurs strictly north and weakly
east of i in T . It is easy to check that for i = 1, 2, . . . , 2n 2, i is contained in A(T )
if and only if i is contained in A(j(T )). Moreover, by Lemma 3.3, j acts cyclically on
the extended descent set. It is easy to check that the ascent set and extended descent
set of a two-row rectangular standard tableau T together determine the associated
handshake pattern. Moreover, it is easy to check that table rotation acts cyclically
on the ascent set and extended descent set of the associated tableau. Therefore,
the bijection between Hn and SY T ((n, n)) maps table rotation to promotion, as
desired.

White proved the portion of Theorem 8.2 involving handshake patterns directly in
the investigation of the rectangular standard tableaux CSP conjecture, thus proving
the conjecture in the case of tableaux with 2 rows. Having proven this more general
conjecture in Theorem 8.2 we get Whites result on handshake patterns as a corollary.
The set Hn also carries an action of the dihedral group D4n of order 4n. Namely,
we think of the points 1, 2, . . . 2n as the vertices of a regular 2n-gon and let D4n act

52

BRENDON RHOADES

naturally. Let r denote the reflection in D4n across the line bisecting the pairs of
vertices (1, 2n) and (n, n + 1). Let s denote rotation in D4n by one clockwise unit.
Proposition 8.3. Let X be either the set Hn or the set NC(n). X carries an action
of D4n , where r and s have the action described above in the case of Hn and act by
reflection about the line through 1 bisecting the circle and Kreweras complementation
in the case of NC(n). Let wo denote the long element of S2n and let c denote the long
cycle (1, 2, . . . , 2n) in S2n .
Then, the number of fixed points of the operators r and rs on X are given by the
formulas:
(
(n,n) (wo )
if n is even
|X r | =
(8.1)
(n,n)
(1)
(wo ) if n is odd
(8.2)

|X rs | = (n,n) (wo c)

Proof. It is easy to see, given the simplified algorithm for evacuation in the case of
rectangular tableaux, that under Whites bijection Hn SY T ((n, n)) the action of
r corresponds to evacuation. Moreover, it is simple to see that under the bijection of
Heitsch [12] Hn NC(n), the action of r corresponds to reflection across the line
bisecting the circle and going through the vertex labelled 1. Therefore, the result
follows from the specialization of Proposition 7.2 to the case a = 2 and b = n.

9. Open Problems
Most of the results in this paper have enumerated some fixed point set via some
representation theoretic interpretation of the associated action. These enumerative
results imply a collection of bijections involving standard tableaux.
Corollary 9.1. Let n be a rectangular partition and suppose d|n. Then, there is
a bijective correspondence between the set of tableaux in SY T () fixed by j d and the
n
set of column strict nd -ribbon tableaux of shape with content 1 d .
Corollary 9.2. Let n be a rectangular partition and suppose d|k are positive
integers. Then, there is a bijective correspondence between the set of tableaux in
CST (, k) fixed by j d and the set of column strict kd -ribbon tableaux of shape with
labels drawn from [ kd ].
Corollary 9.3. Let n be a rectangular partition and let |= n be a composition
with () = k. Suppose that d|k and cdk = . Then, there is a bijective correspondence between the set of tableaux in CST (, k, ) which are fixed by j d and the set of
column strict kd -ribbon tableaux of shape and content (1 , 2 , . . . , d ).
These combinatorial results naturally lead to the following problem.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

53

Problem 9.4. Write down an explicit bijection between any of the above three pairs
of sets.
As we mentioned, one of the reasons that we required to be rectangular in Theorems 1.3, 1.4, and 1.5 was that j n does not in general fix every element of SY T ()
for arbitrary partitions n. This leads to the following question.
Question 9.5. Let n be an arbitrary partition. What is the order of the promotion
operator j on SY T ()? What about on CST (, k) for some fixed k 0? What about
on CST (, k, ) for some fixed k 0 and composition |= n with () = k?
Related to determining the order of promotion on nonrectangular tableaux is the
problem of determining the cycle structure of its action.
Question 9.6. Let n be an arbitrary partition and let the cyclic group C act on
SY T () where the action is generated by jeu-de-taquin promotion. Find an explicit
polynomial X(q) such that (SY T (), C, X(q)) exhibits the CSP. Do the same for
column strict tableaux, with or without fixed content.
10. Acknowledgments
The author would like to thank Serge Fomin, Vic Reiner, Mark Skandera, Dennis
Stanton, John Stembridge, Muge Taskin, and Dennis White for many helpful conversations. Also, the author would like to thank the anonymous referees for many
helpful comments regarding correcting errors, improving exposition, and in particular
pointing out a (as presented corrected) flaw in the proof of Lemma 3.3.
References
[1] O. Abuzzahab, M. Korson, M. Li, and S. Meyer. Cyclic and dihedral sieving for plane
partitions (2005). U. Minnesota REU Report.
[2] J. Bandlow, A. Schilling, and N. Thiery. On the uniqueness of promotion operators on
tensor products of type A crystals (2008). Available on arXiv: 0806.3131.
[3] A. Berenstein and A. Zelevinsky. Canonical bases for the quantum groups of type Ar and
piecewise linear combinatorics. Duke Math. J., 82 (1996) pp. 473502.
rner and F. Brenti. Combinatorics of Coxeter groups. Springer (1995).
[4] A. Bjo
[5] F. Brenti. Combinatorial properties of the Kazhdan-Lusztig R-polynomials for Sn . Adv. Math.,
126, 1 (1997) pp. 2151.
[6] R. K. Brylinski. Limits of weight spaces. J. Amer. Math. Soc., 2 (1989) pp. 517533.
[7] J. Du. Canonical bases for irreducible representations of quantum GLn . Bull. London Math.
Soc., 24, 4 (1992) pp. 325334.
[8] J. S. Frame, G. B. Robinson, and R. M. Thrall. The hook graphs of the symmetric
group. Canad. J. Math., 6 (1954) pp. 316325.
[9] A. M. Garsia and T. J. McLarnan. Relations between Youngs natural and the KazhdanLusztig representations of Sn . Adv. Math., 69 (1988) pp. 3292.

54

BRENDON RHOADES

[10] M. Haiman. On mixed insertion, symmetry, and shifted Young tableaux. J. Combin. Thy. Ser.
A, 50, 2 (1989) pp. 196225.
[11] M. Haiman. Dual equivalence with applications, including a conjecture of Proctor. Discrete
Math., 99 (1992) pp. 79113.
[12] C. E. Heitsch. Kreweras complementation and orbits in Catalan lattices. Submitted, 2007.
[13] A. Joseph, G. Letzter, and S. Zelikson. On the Brylinski-Kostant filtration. J. Amer.
Math. Soc., 13 (2000) pp. 945970.
[14] D. Kazhdan and G. Lusztig. Representations of Coxeter groups and Hecke algebras. Inv.
Math., 53 (1979) pp. 165184.
[15] G. Kreweras. Sur les partitions non croisees dun cycle. Discrete Math., 1, 4 (1972) pp.
333350.
[16] A. Lascoux, B. Leclerc, and J.-Y. Thibon. Green polynomials and Hall-Littlewood functions at roots of unity. Euorp. J. Combinatorics, 15 (1994) pp. 173180.
[17] G. Lusztig. Canonical bases arising from quantized enveloping algebras. J. Amer. Math. Soc.,
3 (1990) pp. 447498.
[18] P. MacMahon. Combinatory Analysis, vol. 1 and 2. Cambridge Univ. Press, Cambridge
(1915).
[19] T. K. Petersen and B. Rhoades. Notes on the order of promotion (2008). In progress.
[20] P. Pylyavskyy. A2 web immanants (2008). In progress.
[21] V. Reiner (2007). Personal communication.
[22] V. Reiner, D. Stanton, and D. White. The cyclic sieving phenomenon. J. Combin. Theory
Ser. A, 108 (2004).
[23] B. Rhoades and M. Skandera. Temperley-Lieb immanants. Ann. Comb., 9, 4 (2005) pp.
451494.
[24] B. Rhoades and M. Skandera. Kazhdan-Lusztig immanants and products of matrix minors.
J. Algebra, 304, 2 (2006) pp. 793811.
[25] B. Sagan. The Symmetric Group. Springer, New York (2001).
[26] B. E. Sagan. Shifted tableaux, Schur Q-functions, and a conjecture of R. Stanley. J. Combin.
Theory Ser. A, 45, 1 (1987) pp. 62103.
[27] M. Skandera. On the dual canonical and Kazhdan-Lusztig bases and 3412, 4231-avoiding
permutations (2006). Submitted.
[28] T. A. Springer. Regular elements of finite reflection groups. Invent. Math., 29 (1974) pp.
159198.
[29] R. Stanley. Enumerative Combinatorics, vol. 1. Cambridge University Press, Cambridge
(1997).
[30] R. Stanley. Enumerative Combinatorics, vol. 2. Cambridge University Press, Cambridge
(1999).
[31] R. Stanley. Promotion and evacuation. Electronic J. Combinatorics, 15, 2 (2008-2009).
[32] R. P. Stanley. Invariants of finite groups and their applications to combinatorics. Bull. Amer.
Math. Soc. (N.S.), 1, 3 (1979) pp. 475511.
[33] D. Stanton and D. White. A Schensted algorithm for rim-hook tableaux. J. Combin. Theory
A, 40 (1985) pp. 211247.
[34] J. Stembridge. On minuscule representations, plane partitions, and involutions in complex
Lie groups. Duke Math J., 73 (1994) pp. 469490.
[35] J. Stembridge. Canonical bases and self-evacuating tableaux. Duke Math. J., 82, 3 (1996)
pp. 585606.
[36] M. Taskin. Properties of four partial orders on standard Young tableaux. J. Combin. Theory
Ser. A. In press, 2006.

CYCLIC SIEVING, PROMOTION, AND REPRESENTATION THEORY

[37] D. White (2007). Personal communication.

[38] D. R. Worley. A theory of shifted Young tableaux . Ph.D. thesis, MIT (1984).

55