Generalized Rie Shues

and Quasisymmetric functions

Richard P. Stanley1
Department of Mathematics 2-375
Massachusetts Institute of Technology
Cambridge, MA 02139
version of 31 May 2001

1 Introduction.
Let xi be a probability distribution on a totally
P
ordered set I , i.e., the prob-
ability of i 2 I is xi . (Hence xi  0 and xi = 1.) Fix n 2 P = f1; 2; : : :g,
and de ne a random permutation w 2 Sn as follows. For each 1  j  n,
choose independently an integer ij (from the distribution xi ). Then standard-
ize the sequence i = i1


in in the sense of [34, p. 322], i.e., let 1 <


< k
be the elements of I actually appearing in i, and let ai be the number of i's
in i. Replace the 1's in i by 1; 2; : : : ; a1 from left-to-right, then the 2 's in
i by a1 + 1; a1 + 2; : : : ; a1 + a2 from left-to-right, etc. For instance, if I = P
and i = 311431, then w = 412653. This de nes a probability distribution
on the symmetric group Sn, which we call the QS-distribution (because of
the close connection with quasisymmetric functions explained below). If we
need to be explicit about the parameters x = (xi)i2I , then we will refer to
the QS(x)-distribution.
The QS-distribution is not new, at least when I is nite. It appears for
instance in [10, pp. 153{154][14][20][24]. Although these authors anticipate
some of our results, they don't make systematic use of quasisymmetric and
symmetric functions. Some additional work using the viewpoint of our paper
is due to Fulman [15].
1.1 Example. As an example of a general result to be proved later
(Theorem 2.1), let us compute Prob(213), the probability that a random
1 Partially supported by NSF grant DMS-9988459

1
permutation w 2 S3 (chosen from the QS-distribution) is equal to the per-
mutation 213. A sequence i = i1 i2i3 will have the standardization 213 if and
only if i2 < i1  i3. Hence
X
Prob(213) = xa xb xc:
a<bc
There is an alternative way to describe the QS-distribution. Suppose we
have a deck of n cards. Cut
P the deck into random packets of respective sizes
ai > 0; i 2 I , such that ai = n and the probability of (ai)i2I is
Y ai
Prob ((ai)i2I ) = n! xai ! :
i2I i
Then rie shue these packets Pi together into a pile P in the manner
described by Bayer and Diaconis [2] (see also [11][14]); namely, after placing
k cards on P , the probability that the next card comes from the top of
the current packet Pj is proportional to the current number of cards in Pj .
This card is then placed at the bottom of P . The ordinary dovetail or rie
shue [2, p. 294] corresponds to the case I = f1; 2g and x1 = x2 = 1=2.
In this case the original deck is cut according to the binomial distribution.
More generally, if for xed q 2 P we cut the deck into q packets (some
possibly empty) according to the q-multinomial distribution, then we obtain
the q-shues of Bayer and Diaconis [2, p. 299] (where they use a for our
q). The q-shue is identical to the QS-distribution for I = f1; 2 : : : ; qg and
x1 = x2 =


= xq = 1=q. We will denote this distribution by Uq . If
we q-shue a deck and then r-shue it, the distribution is the same as a
single qr-shue [2, Lemma 1]. In other words, if  denotes convolution of
probability distributions then Uq  Ur = Uqr . We extend this result to the
QS-distribution in Theorem 2.4.
In the next section we will establish the connection between the QS-
distribution and the theory of quasisymmetric functions. In Section 3 we use
known results from the theory of quasisymmetric and symmetric functions to
obtain results about the QS-distribution, many of them direct generalizations
of the work of Bayer, Diaconis, and Fulman mentioned above. For instance,
we show that the probability that a random permutation w 2 Sn (always
assumed chosen from the QS-distribution) has k inversions is equal to the
probability that w has major index k. This is an analogue of MacMahon's
2
corresponding result for the uniform distribution on Sn (see e.g. [33, p. 23]).
A further result is that if T 0 is a standard Young tableau of shape , then
the probability that T 0 is the recording tableau of w under the Robinson-
Schensted-Knuth algorithm is s(x), where x = (xi)i2I and s denotes a
Schur function.

2 The QS-distribution and quasisymmetric func-

tions
In this section we show the connection between the QS-distribution and qua-
sisymmetric functions. In general our reference for quasisymmetric functions
and symmetric functions will be [34, Ch. 7]. Quasisymmetric functions may
be regarded as bearing the same relation to compositions (ordered partitions
of a nonnegative integer) as symmetric functions bear to partitions. Namely,
the homogeneous symmetric functions of degree n form a vector space of
dimension p(n) (the number of partitions of n) and have many natural bases
indexed by the partitions of n. Similarly, the homogeneous quasisymmetric
functions of degree n form a vector space of dimension 2n 1 (the number of
compositions of n) and have many natural bases indexed by the compositions
of n.
A quasisymmetric function F (z) is a power series of bounded degree, say
with rational coecients, in the variables z = (zi)i2I , with the following
property: let i1 < i2 <


< in and j1 < j2 <


< jn , where ik ; jk 2 I . If
a1; : : : ; ak 2 P, then the coecient of zia


ziann is equal to the coecient of
1

zja


zjann in F (z). The set of all quasisymmetric functions forms a graded Q -
1
1

algebra denoted Q . The quasisymmetric functions that are homogeneous of

1

degree n form a Q -vector space, denoted Qn. If jI j  n (including jI j = 1),

then dim Qn = 2n 1.
P
Let = ( 1; : : : ; k ) be a composition of n, i.e., i 2 P and i = n.
Let Comp(n) denote the set of all compositions of n, so #Comp(n) = 2n 1
[33, pp. 14{15]. De ne
S = f 1; 1 + 2; : : : ; 1 +


+ k 1 g  f1; : : : ; n 1g:
3
The fundamental quasisymmetric function L = L (z) is de ned by
X
L (z ) = zi


zin :
1
i1 


in
ij <ij +1 if j 2S

It is not hard to show that when jI j  n, the set fL : 2 Comp(n)g is

a Q -basis for Qn [34, Prop. 7.19.1]. More generally, if jI j = k then the set
fL : 2 Comp(n); length( )  kg is a Q -basis for cqn, and L = 0 if
length( ) > k.
If w = w1w2


wn 2 Sn then let D(w) denote the descent set of w, i.e.,
D(w) = fi : wi > wi+1g:
Write co(w) for the unique composition of n satisfying Sco(w) = D(w). In
other words, if co(w) = ( 1; : : : ; k ), then w has descents exactly at 1 ,
1 + 2 ; : : :, 1 +


+ k 1 . For any composition of n it is easyPto see that
the (possibly in nite) series L (x) (where as always xi  0 and xi = 1) is
absolutely convergent and therefore represents a nonnegative real number.
The main result on which all our other results depend is the following.
2.1 Theorem. Let w 2 Sn. The probability Prob(w) that a permutation
in Sn chosen from the QS-distribution is equal to w is given by
Prob(w) = Lco(w )(x): 1

Proof. This result is equivalent to [18, Lemma 3.2] (see also [30,
Lemma 9.39]). Because of its importance here we present the (easy) proof.
The integers i for which i + 1 appears somewhere to the left of i in w =
w1w2


wn are just the elements of D(w 1). Let i1; : : : ; in be chosen in-
dependently from the distribution xi . Let aj = iw (j), i.e., if we write1

down i1


in underneath w1


wn, then aj appears below j . In order for

the sequence i1


in to standardize to w it is necessary and sucient that
a1  a2 


 an and that aj < aj+1 whenever j 2 D(w 1). Hence
X
Prob(w) = xa xa


xan
1 2
a1 a2 


an
ai <ai+1 if i2D(w 1)

= Lco(w ) (x): 2
1

4
 For some algebraic results closely related to Theorem 2.1, see [12].
Two special cases of Theorem 2.1 are of particular interest. The rst is
the q-shue distribution Uq . Let des(u) denote the number of descents of
the permutation u 2 Sn, i.e.,
des(u) = #fi : w(i) > w(i + 1)g:
It follows easily from Theorem 2.1 or by a direct argument (see [34, p. 364])
that the probability ProbUq (w) that a random permutation in Sn chosen
from the distribution Uq is equal to w is given by
 1 ) + n 1
ProbUq (w) = q des( w q n:
n
This is just the probability of obtaining w with a q-shue, as de ned in [2,
x3], con rming that Uq is just the q-shue distribution.
For xed n, the uniform distribution U on Sn is given by
U = qlim U:
!1 q
(1)
Because of this formula many of our results below may be considered as gen-
eralizations of the corresponding results for the uniform distribution. These
results for the uniform distribution are well-known and classical theorems in
the theory of permutation enumeration.
The second interesting special case of Theorem 2.1 is de ned by
xi = (1 t)ti 1 ;
where I = P and 0  t  1. Given u = u1u2


un 2 Sn, de ne u =
v1v2


vn , where vi = n + 1 un+1 i. (Equivalently, u = w0uw0, where
w0(i) = n + 1 i.) Set e(u) = maj(u), where maj denotes major index (or
greater index ) of u [33, p. 216], i.e.,
X
maj(u) = i:
i2D(u)
It follows from [34, Lemma 7.19.10] that
te(w )(1 t)n
Prob(w) = (1 t)(1
1

t2 )


(1 tn) : (2)

5
 The QS-distribution de nes a Markov chain (or random walk) Mn on Sn
by de ning the probability Prob(u; uw) of the transition from u to uw to be
Prob(w) = Lco(w ) (x). The next result determines the eigenvalues of this
1

Markov chain. Equivalently, de ne

X
n (x) = Lco(w )(x)w 2 R Sn ;
1

w2Sn
where R Sn denotes the group algebra of Sn over R . Then the eigenvalues of
Mn are just the eigenvalues of n(x) acting on R Sn by right multiplication.
2.2 Theorem. The eigenvalues of Mn are the power sum symmet-
ric functions p (x) for  ` n. The eigenvalue p(x) occurs with multiplicity
n!=z , the number of elements in Sn of cycle type .
The above theorem appears implicitly in [16, Thm. 4.4] and more explic-
itly in [23, Note 5.20]. See also [13, x7] and [34, Exer. 7.96]. It is also a
special case of [3, Thm. 1.2], as we now explain.
Bidigare, Hanlon, and Rockmore [3] de ne a random walk on the set R
of regions of a ( nite) hyperplane arrangement A in R n as follows. (Precise
de nitions of the terms used below related to arrangements may be found in
[3]. For further information see [9].) Let F be the set of faces of A, i.e., the
nonempty faces of the closure of the regions of A. Let wt be a probability
distribution on F . Given R 2 R and F 2 F , de ne F
R to be that region
R0 of A closest to R that has F as a face. (There is always a unique such
region R0.) Now given a region R, choose F 2 F with probability wt(F ) and
move to the region F
R.
Consider the special case when A is the braid arrangement Bn, i.e., the
set of hyperplanes ui = uj for 1  i < j  n, where u1; : : : ; un are the
coordinates in R n . The set of regions of Bn can be identi ed in a natural way
with Sn, viz., identify the region ua < ua <


< uan with the permutation
w given by w(ai) = i. The faces of Bn correspond to ordered set partitions
1 2

 of 1; 2; : : : ; n, i.e.,  = (B1 ; : : : ; Bk ) where Bi 6= ;, Bi \ Bj = ; for i 6= j ,

and [Bi = f1; 2; : : : ; ng; viz.,  corresponds to the face de ned by ur = us if
r; s 2 Bi for some i, and ur < us if r 2 Bi and s 2 Bj with i < j . De ne the

6
type of  to be the composition
type() = (#B1; : : : ; #Bk ) 2 Comp(n):
Given = ( 1 ; : : : ; k ) 2 Comp(n), de ne the monomial quasisymmetric
function M (z ) by X
M (z ) = zi


zikk :
1
1

i1 <


<ik
(See [34, (7.87)].) Now if F is a face of Bn corresponding to the ordered
partition , then set
wt(F ) = Mtype() (x):
It is easy to see that wt is a probability distribution on the set of faces of
Bn; in fact, for indeterminates (zi)i2I we have
X  X n
Mtype()(z) = zi :

Thus wt de nes a Bidigare-Hanlon-Rockmore random walk on the regions
of Bn . Let P (R; R0) denote the probability of moving from a region R to a
region R0 .
2.3 Theorem. Let R correspond to u 2 Sn and R0 to uw. Then
P (R; R0) = Lco(w )(x): 1

Hence the Bidigare-Hanlon-Rockmore random walk just de ned on the regions

of Bn is isomorphic to the random walk on Sn de ned by the QS-distribution.
Proof. By symmetry we may suppose that u = 12


n, the identity
permutation. The faces F of R0 for which F
R = R0 correspond to those
ordered partitions  obtained from w 1 = w1


wn by drawing bars between
certain pairs wi and wi+1, where there must be a bar when i 2 D(w 1). The
elements contained between the consecutive bars (including a bar before w1
and after wn) form the blocks of , read from left-to-right. For instance, if
w 1 = 582679134, then one way of drawing bars is j58j267j9j1j34j, yielding
the ordered partition  = (f5; 8g; f2; 6; 7g; f9g; f1g; f3; 4g). It follows that
X
P (R; R0) = M (x);

7
where runs over all compositions re ning co(w 1). From the de nition of
L (x) we have X
M (x) = Lco(w )(x); 1

and the proof follows. 2

The next result determines the convolution of the QS(x)-distribution with
the QS(y)-distribution. We write xy for the variables xi yj in the order xi yj <
xr ys if either i < r or i = r and j < s.
2.4 Theorem. Suppose that a permutation u 2 Sn is chosen from
the QS(x)-distribution, and a permutation v 2 Sn is chosen from the QS(y)-
distribution. Let w 2 Sn. Then the probability that uv = w is equal to
Lco(w ) (xy). In other words, if  denotes convolution then
1

QS(x)  QS(y) = QS(xy):

Equivalently, in the ring R Sn we have
n (x) n (y ) = n (xy ):
Theorem 2.4 is equivalent to [23, (62)][30, Thm. 9.37]. For the Uq -
distribution it was also proved in [2, Lemma 1], and this proof can be easily
extended to prove Theorem 2.4 in its full generality.

3 Enumerative properties of the QS-distribution

Suppose that X is any subset of Sn. Let w 2 Sn be chosen from the QS-
distribution. It follows from Theorem 2.1 that the probability that w 2 X is
given by X
Prob(w 2 X ) = Lco(u ) (x): 1 (3)
u2X
There are many known instances of the set X  Sn for which the right-hand
side of (3) can be explicitly evaluated. Most of this section will be devoted
to some examples of this type.

8
 Our rst result involves the symmetric function ln(z) de ned by
ln(z) = n1 (d)pn=d
X
d (z );
djn
where  denotes the usual number-theoretic Mobius function. More infor-
mation on ln(z) may be found in [30, Ch. 8] or [34, Exer. 7.88{7.89].
3.1 Theorem. Let w be a random permutation in Sn, chosen from
the QS-distribution. The probability Prob((w) = ) that w has cycle type
 = h1m 2m


i ` n (i.e., mi cycles of length i) is given by
1 2

Y
Prob((w) = ) = hmi [li](x);
i1
where brackets denote plethysm [26, xI.8][34, p. 447].
The proof of Theorem 3.1 follows from [30, Thm. 8.23 and Thm. 9.41(a)].
In the special case of the Uq -distribution, the result appears in [11, Thm. A].
As above let maj(w) denote the major index of w 2 Sn, and let inv(w)
denote the number of inversions of w [33, pp. 20{21], i.e.,
inv(w) = #f(i; j ) : 1  i < j  n; w(i) > w(j )g:
Let In(j ) (respectively, Mn(j )) denote the probability that w 2 Sn satis es
inv(w) = j (respectively, maj(w) = j ) under the QS-distribution.
3.2 Theorem. We have
Mn(j ) = In(j )
XX Mn(j )tj zn =
YY
1 ti 1x z
1 : (4)
j
n0 j 0 (1 t)(1 t2)


(1 tn) i1 j 2I
Proof. The standardization w of a sequence i = i1i2


in satis es
maj(w) = maj(i) and inv(w) = inv(i). Hence by de nition of the QS-
distribution we have
X X
Mn (j )tj = tmaj(i)xi


xin 1
j i =i1


in
X X
In(j )tj = tinv(i) xi


xin :
1
j i =i1


in

9
Thus if
X
F  ( t) = tmaj(v)
v
X
G(t) = tinv(v) ;
v
where v ranges over all permutations of the multiset f1 ; 2 ; : : :g, then
1 2

X X
Mn(j )tj = F (t)m (x)
j `n
X X
In(j )tj = G (t)m(x);
j `n
where m (x) denotes a monomial symmetric function. But it is known [32,
(45), p. 97][33, Prop. 1.3.17] that
 
F(t) = G (t) =  ; n ; : : : ;
1 2 t
a q-multinomial coecient in the variable t. Hence Mn(j ) = In(j ).
Now let h (x) denote a complete homogeneous symmetric function [34,
x7.5]. It is easy to see [34, Prop. 7.8.3] that
h (1; t; t2; : : :) = (1 t)(1 t21)


(1 ti ) :
Y

i
Hence
X X
Mn (j )tj = (1 t)


(1 tn) h (1; t; t2; : : :)m(x):
j `n
Equation (4) thenQfollows immediately from the expansion [34, (7.10)] of the
Cauchy product i;j (1 xi yj ) 1. Equation (4) is also equivalent to [17,
(78)]. 2

P Other expansions of the Cauchy product lead to further formulas for

j Mn (j )t . In particular, from [34, (7.20), Thm. 7.12.1, and Cor. 7.21.3]
j

10
there follows (using notation from [34])
X
M n ( j ) tj =
X
tb() (1Q t) (1


(1 tn ) s (x)
j `n u2 th(u) ) 
X (1 t)


(1 tn )
1
=
(1 t  )


(1 tl ) z p (x):
1
(5)
`n
A famous result of MacMahon asserts that
#fw 2 Sn : maj(w) = j g = #fw 2 Sn : inv(w) = j g:
Since the uniform distribution U on Sn satis es U = limq!1 Uq , it follows
that Theorem 3.2 is a generalization of MacMahon's result. In fact, the proof
we have given of Theorem 3.2 shows that it is equivalent to the equidistribu-
tion of maj and inv on the set of permutations of any nite multiset (where
the underlying set is linear ordered). This result was also known, at least
implicitly, to MacMahon. For further references and some generalizations,
see [5].
In principle one can use Theorem 3.2 to derive the moments of maj(w)
(or equivalently inv(w)). Let E denote expectation with respect to the QS-
distribution and EU expectation with respect to the uniform distribution.
The next result obtains the rst two moments of maj(w), stated for simplicity
as the expectations of maj(w) and maj(w)(maj(w) 1) (instead of maj(w)2).
3.3 Corollary. Choose w 2 Sn from the QS-distribution. Then
 
E (maj(w)) = EU (maj(w)) 12 n2 p2 (x) (6)
E (maj(w)(maj(w)  1)) 
= EU (maj(w)(maj(

w) 1)) (7)
3 n +4 1 p2 (x) + 34 n3 p3(x) + 32 n4 p2 (x)2: (8)
Note. It is easy to see (e.g. [22, p. 16]) that
 
EU (maj(w)) = 21 n2
EU (maj(w)(maj(w) 1)) = n(n 1)(n1442)(9n + 13) :
11
Proof of Corollary 3.3. Let
X
n(t) = Mn (j )tj :
j
Then
E (maj(w)) = 0n(1):
When we di erentiate (5) term-by-term and set t = 1, the only surviving
terms will be from  = h1ni and  = h21n 2i. Hence

E (maj(w)) = dt (1 t(1)


t(1)n t ) n1! p1(x)n
d n


(1 t )


(1 tn) 1 n
+ (1 t2 )(1 t)n 2 2
(n 2)! p1(x) p2(x) : 2 (9)
t=1
P
Now p1(x) = xi = 1 and
 
d (1 t)


(1 tn) 1
dt (1 t)n n! t=1 = EU (maj(w)):
It is routine to compute the second term on the right-hand side of (9), ob-
taining equation (6).
A similar computation using E (maj(w)(maj(w) 1)) = 00n(1) yields (8);
we omit the details. 2
It is clear from the above proof that for general k  1, E (maj(w)k ) is
a linear combination, whose coecients are polynomials in n, of power sum
symmetric functions p(x) where  ` n and `()  n k. Here `() denotes
the length (number of parts) of .
We next consider the relationship between the QS-distribution and the
Robinson-Schensted-Knuth (RSK) algorithm [34, x7.11]. Recall that this
algorithm associates a pair (T; T 0) of standard Young tableaux (SYT) of the
same shape  ` n with a permutation w 2 Sn. We call T 0 the recording
tableau of w, denoted rec(w). The shape  of T 0 is also called the shape of
w, denoted sh(w) = .

12
 3.4 Theorem. Choose w 2 Sn from the QS-distribution. Let T be an
SYT of shape  ` n. Then the probability that rec(w) = T is given by
Prob(rec(w) = T ) = s(x);
where s denotes a Schur function.
Proof. Let
XT = fw 2 Sn : rec(w) = T g:
It follows from [34, Thm. 7.19.2 and Lemma 7.23.1] that
X
Lco(w ) (x) = s(x):
1

w2XT
The proof now follows from equation (3). 2
3.5 Corollary. Choose w 2 Sn from the QS-distribution, and let  ` n.
Then
Prob(sh(w) = ) = f s(x);
where f  denotes the number of SYT of shape  (given explicitly by the
Frame-Robinson-Thrall hook-length formula [34, Cor. 7.21.6]).
Proof. There are f  SYT's T of shape , and by Theorem 3.4 they all
have probability s (x) of being the recording tableau of w. 2
Note. If the RSK algorithm associates (T; T 0) with w 2 Sn, then call
T the insertion tableau of w, denoted ins(w). Since ins(w) = rec(w 1) [34,
Thm. 7.13.1], it follows that
Prob(ins(w) = T ) = f Lco(w) (x):

Note. The probability distribution Prob() = f  s (x) on the set of all

partitions  of all nonnegative integers is a specialization of the z-measure
of Borodin and Olshanski, as surveyed in [7] (see also [29]).
It is easy to give an expression for the probability that a permutation w 2
Sn chosen from the QS-distribution has a xed descent set S  f1; 2; : : : ; n
13
1g. In general, if  and  are partitions with    (i.e., i  i for all i), then
= denotes the skew shape obtained by removing the diagram of  from that
of  [34, p. 309]. One can then de ne (see [34, Def. 7.10.1]) the skew Schur
function s= (x). A border strip (or rim hook or ribbon ) is a connected skew
shape with no 2  2 square [34, p. 245]. Let = ( 1 ; : : : ; k ) 2 Comp(n).
Then there is a unique border strip B (up to translation) with i squares
in row i. For instance, the border strip B(3;2;1;2;4) looks like

Some special properties of border strip Schur functions sB are discussed in

[34, x7.23].
3.6 Theorem. Let w be a random permutation in Sn, chosen from
the QS-distribution. Let = ( 1 ; : : : ; k ) 2 Comp(n), and recall that S =
f 1; 1 + 2 ; : : : ; 1 +


+ k 1g  f1; 2 : : : ; n 1g. Then the probability
that w has descent set S is given by
Prob(D(w) = S ) = sB (x):
Proof. Immediate from Theorem 2.1 and the fact [34, Cor. 7.23.4] that
X
sB = Lco(w ) :
1 2
w2Sn
=co(w)

4 Longest increasing subsequences.

It is a fundamental result of Schensted [31] (see also [34, Cor. 7.23.11]) that
if sh(w) =  = (1; 2; : : :), then 1 is the length is(w) of the longest in-
creasing subsequence of w. (There is a generalization due to Greene [19][34,
Thm. 7.23.13 or A1.1.1] that gives a similar interpretation of any i.) Thus
14
the expected value EU (n) of is(w) for w 2 Sn under the uniform distribution
is given by
EU (n) = EU (is(w)) = n1! 1 f  2 :
X

`n
(See [34, Exer.
p 7.109(b)].) It was shown by Vershik and Kerov [35] that
EU (n)  2 n. Later Baik et al. [1] obtained a much stronger result, viz.,
the exact distribution (suitably normalized) of 1 (or equivalently is(w)) in
the limit n ! 1. (This result was extended to any i in [8][28][21].) We can
ask whether similar results hold for the QS-distribution. Many results along
these lines appear in [4][6, Ex. 2][8][27][29]. Here we make some elementary
comments pointing in a di erent direction from these papers.
It is clear from Corollary 3.5 that
X
E (is(w)) = 1 f s(x); (10)
`n
where as usual E denotes expectation with respect to the QS-distribution.
This formula can be made more explicit in the special case of the distribution
Uq . Identify  ` n with its Young diagram, and let c(u) denote the content
of the square u 2  as de ned in [34, p. 373].
4.1 Theorem. We have
EUq (is(w)) = n1! 1 f  2 1 + q 1c(u) :
X
Y

(11)
`n u2
Proof. Let s (1q ) denote s (x) where x1 =


= xq = 1 and xi = 0 for
i > q. It is well-known (e.g. [34, Cor. 7.21.4 and 7.21.6]) that
Y
s(1q ) = fn! (q + c(u)):
u2
Since s (x) is homogeneous of degree n, the proof follows from equation (10).
2
Note that EUq (is(w)) is a polynomial in q 1, which can be written
!
EUq (is(w)) = EU (is(w)) + n1! 1 f  2 c(u) 1q +


:
X
X
(12)
`n u2

15
Let us mention that
X X X i  X 0 
c(u) = i(0
i i ) = 2
i
2 ;
u2
where 0 = (01 ; 02; : : :) denotes the conjugate partition to . If we regard n
as xed, then equation (12) shows the rate at which EUq (is(w)) converges to
EU (is(w)) as q ! 1. It is therefore of interest to investigate the asymptotic
behavior of the coecient
!
1 X
F1(n) = n! 1 f 

2 X
c(u)
`n u2
as n ! 1 (as well as the coecient of q i for i > 1). Numerical evidence
suggests that F1 (n)=n is a slowly increasing function of n. The largest value
of n for which we have done the computation is n = 47, giving F1 (47)=47 
0:6991. Eric Rains suggests that there is some reason to suspect that F (n)=n
grows like n1=6 .
Acknowledgment. I am grateful to Jinho Baik, Francois Bergeron,
Ken Brown, Persi Diaconis, Alain Lascoux, Andrei Okounkov, Craig Tracy,
and an anonymous referee for providing helpful suggestions and additional
references.

References
[1] J. Baik, P. Deift, and K. Johansson, On the distribution of the length of
the longest increasing subsequence of random permutations, J. Amer.
Math. Soc., to appear.
[2] D. Bayer and P. Diaconis, Trailing the dovetail shue to its lair, Ann.
Applied Probability 2 (1992), 294{313.
[3] P. Bidigare, P. Hanlon, and D. Rockmore, A combinatorial description of
the spectrum of the Tsetlin library and its generalization to hyperplane
arrangements, Duke Math. J. 99 (1999), 135{174.
[4] Ph. Biane, seminar talk at U.C. Berkeley, March 2000.
16
 [5] A. Bjorner and M. Wachs, q-hook length formulas for forests, J. Com-
binatorial Theory (A) 52 (1989), 165{187.
[6] A. Borodin and A. Okounkov, A Fredholm determinant formula for
Toeplitz determinants, Integral Equations and Operator Theory 37
(2000), 386{396, math.CA/9907165.
[7] A. Borodin and G. Olshanski, Z -measures on partitions, Robinson-
Schensted-Knuth correspondence, and = 2 random matrix ensembles,
preprint, math.CO/9905189.
[8] A. Borodin, A. Okounkov, and G. Olshanski, On asymptotics of
Plancherel measures for symmetric groups, J. Amer. Math. Soc. 13
(2000), 481{515, math.CO/9905032.
[9] K. Brown and P. Diaconis, Random walks and hyperplane arrangements,
Ann. Probability 26 (1998), 1813{1854.
[10] P. Diaconis, J. Fill, and J. Pitman, Analysis of top to random shues,
Combinatorics, Probability, and Computing 1 (1992), 135{155.
[11] P. Diaconis, M. McGrath, and J. Pitman, Rie shues, cycles, and
descents, Combinatorica 15 (1995), 11{29.
[12] G. Duchamp, F. Hivert, and J.-Y. Thibon, Une generalisation des fonc-
tions quasi-symetriques et des fonctions symetriques non commutatives,
C. R. Acad. Sci. Paris Ser. I Math. 328 (1999), 1113{1116.
[13] G. Duchamp, A. Klyachko, D. Krob, and J.-Y. Thibon, Noncommuta-
tive symmetric functions III: Deformations of Cauchy and convolution
algebras, Discrete Math. Theor. Comput. Sci. 1 (1997), 159{216.
[14] J. Fulman, The combinatorics of biased rie shues, Combinatorica 18
(1998), 173{184.
[15] J. Fulman, Applications of symmetric functions to cycle and subsequence
structure after shues, preprint, math.CO/0102176.
[16] A. Garsia and C. Reutenauer, A decomposition of Solomon's descent
algebra, Advances in Math. 77 (1989), 189{262.

17
[17] I. M. Gelfand, D. Krob, A. Lascoux, B. Leclerc, V. S. Retakh, and J.-Y.
Thibon, Noncommutative symmetric functions, Advances in Math. 112
(1995), 218{348.
[18] I. Gessel and C. Reutenauer, Counting permutations with given cycle
structure and descent set, J. Combinatorial Theory (A) 64 (1993), 189{
215.
[19] C. Greene, An extension of Schensted's theorem, Advances in Math. 14
(1974), 254{265.
[20] A. R. Its, C. A. Tracy, and H. Widom, Random words, Toeplitz deter-
minants and integrable systems. I, preprint, math.CO/9909169.
[21] K. Johansson, Discrete orthogonal polynomial ensembles and the Plan-
cherel measure, preprint, math.CO/9906120.
[22] D. E. Knuth, The Art of Computer Programming, vol. 3, Sorting and
Searching, second ed., Addison-Wesley, Reading, Massachusetts, 1998.
[23] D. Krob, B. Leclerc, and J.-Y. Thibon, Noncommutative symmetric
functions II: Tranformations of alphabets, Internat. J. Algebra Comput.
7 (1997), 181{264.
[24] G. Kuperberg, Random words, quantum statistics, central limits, ran-
dom matrices, preprint, math.PR/9909104.
[25] B. F. Logan and L. A. Shepp, A variational problem for random Young
tableaux, Advances in Math. 26 (1977), 206{222.
[26] I. G. Macdonald, Symmetric Functions and Hall Polynomials, second
ed., Oxford University Press, Oxford, 1995.
[27] A. Okounkov, In nite wedge and random partitions, preprint,
math.RT/9907127.
[28] A. Okounkov, Random matrices and random permutations, Internat.
Math. Res. Notices 2000, 1043{1095, math.CO/9903176.
[29] A. Okounkov, SL(2) and z-measures, preprint, math.RT/0002135.

18
[30] C. Reutenauer, Free Lie Algebras, Oxford University Press, Oxford,
1993.
[31] C. E. Schensted, Longest increasing and decreasing subsequences,
Canad. Math. J. 13 (1961), 179{191.
[32] R. Stanley, Ordered structures and partitions, Mem. Amer. Math. Soc.,
no. 119 (1972).
[33] R. Stanley, Enumerative Combinatorics, vol. 1, second printing, Cam-
bridge University Press, New York/Cambridge, 1996.
[34] R. Stanley, Enumerative Combinatorics, vol. 2, Cambridge University
Press, New York/Cambridge, 1999.
[35] A. M. Vershik and S. V. Kerov, Asymptotic behavior of the Plancherel
measure of the symmetric group and the limit form of Young tableaux,
Soviet Math. Dokl. 18 (1977), 527{531; translated from Dokl. Akad.
Nauk SSSR 233 (1977), 1024{1027.

19