Permutations

Richard P. Stanley1
Department of Mathematics
Massachusetts Institute of Technology
Cambridge, MA 02139, USA
[email protected]

version of 21 May 2015

Permutations of finite sets play a central role in algebraic and enu-

merative combinatorics. In addition to having many interesting enu-
merative properties per se, permutations also arise in almost every
area of mathematics and indeed in all the sciences. Here we will dis-
cuss three different topics involving permutations, focusing on com-
binatorics but also giving some hints about connections with other
areas.

1 Increasing and decreasing subsequences

1.1 The Erdős-Szekeres theorem
Let Sn denote the symmetric group of all permutations of [n] :=
{1, 2, . . . , n}. We write permutations w ∈ Sn as words, i.e., w =
a1 a2 · · · an , where w(i) = ai . An increasing subsequence of w of
length k is a subsequence ai1 , . . . , aik (so i1 < · · · < ik ) satisfying
ai1 < · · · < aik , and similarly for decreasing subsequence. For in-
stance, if w = 5642713, then 567 is an increasing subsequence and
643 is a decreasing subsequence. Let is(w) (respectively, ds(w)) de-
note the length of the longest increasing (respectively, decreasing)
subsequence of w. If w = 5642713 as above, then is(w) = 3 (corre-
1
This material is based upon work supported by the National Science Foun-
dation under Grant No. 0604423. Any opinions, findings and conclusions or
recommendations expressed in this material are those of the author and do not
necessarily reflect those of the National Science Foundation.
sponding to 567) and ds(w) = 4 (corresponding to 5421 or 6421).
The subject of increasing and decreasing subsequences began in
1935, and there has been much recent activity. There have been ma-
jor breakthroughs in understanding the distribution of is(w), ds(w),
and related statistics on permutations, and many unexpected and
deep connections have been obtained with such areas as representa-
tion theory and random matrix theory. A more extensive survey of
this topic appears in [51].
The first result on increasing and decreasing subsequences is a
famous theorem of Erdős and Szekeres [16].
Theorem 1.1. Let p, q ≥ 1. If w ∈ Spq+1 , then either is(w) > p or
ds(w) > q.
Seidenberg [42] gave an exceptionally elegant proof of Theorem 1.1
based on the pigeonhole principle which has been reproduced many
times. Theorem 1.1 is best possible in that there exists w ∈ Spq
with is(w) = p and ds(w) = q. Schensted [41] found a quantitative
strengthening of this result based on his rediscovery of an algorithm
(now called the RSK algorithm) of Robinson [38] which has subse-
quently become a central topic in algebraic combinatorics.
In order to explain Schensted’s work, define a partition λ of an
integer n ≥ 0 to be anP integer sequence λ = (λ1 , λ2 , . . . ) satisfying
λ1 ≥ λ2 ≥ · · · ≥ 0 and λi = n. We then write λ ⊢ n. The number
of λi > 0 is the length of λ, denoted ℓ(λ). The conjugate partition
λ′ = (λ′1 , λ′2 , . . . ) to λ has λi −λi+1 parts equal to i. Note that λ′′ = λ,
and that λ′1 = ℓ(λ), λ1 = ℓ(λ′ ).
A standard Young tableau (SYT) of shape λ ⊢ n is a left-justified
array of integers, with λi entries in the ith row, such that every integer
1, 2, . . . , n appears once, and every row and column is increasing. An
example of an SYT of shape 4421 (short for (4, 4, 2, 1, 0, . . . )) is
1 3 5 6
2 7 9 11
.
4 10
8
rsk
Schensted defines a bijection w → (P, Q) between permutations

2
w ∈ Sn and pairs (P, Q) of SYT of the same shape λ = (λ1 , λ2 , . . . ) ⊢
n. His main theorem states that

is(w) = λ1 , ds(w) = λ′1 = ℓ(λ). (1)

This result was greatly extended by Greene [25], as follows. Suppose

that the parts of λ are given by λ1 ≥ λ2 ≥ · · · ≥ λℓ > 0, and of the
conjugate partition λ′ by λ′1 ≥ λ′2 ≥ · · · ≥ λ′k > 0.
Theorem 1.2. Let w ∈ Sn , 1 ≤ i ≤ ℓ, and 1 ≤ j ≤ k. Then the size
(number of terms) of the longest union of i increasing subsequences
of w is λ1 +· · ·+λi , while the size of the longest union of j decreasing
subsequences of w is λ′1 + · · · + λ′j .
For instance, let w = 247951368. The longest increasing subse-
quence is 24568, so λ1 = 5. The largest union of two increasing sub-
sequences is 24791368 (the union of 2479 and 1368), so λ1 + λ2 = 8.
(Note that it is impossible to find a union of length 8 of two in-
creasing subsequences that contains an increasing subsequence of
length λ1 = 5.) Finally w itself is the union of the three increas-
ing subsequences 2479, 1368, and 5, so λ1 + λ2 + λ3 = 9. Hence
(λ1 , λ2 , λ3 ) = (5, 3, 1) (and λi = 0 for i > 3).
Schensted’s theorem (1) leads to a formula for the number

fp,q (n) = #{w ∈ Sn : is(w) = p, ds(w) = q}.

Namely, if f λ is the number of SYT of shape λ (given explicitly

by the famous hook length formula of Frame-Robinson-Thrall [50,
Cor. 7.21.6]), then
X
fp,q (n) = (f λ )2 .
λ⊢n
λ1 =p, λ′1 =q

This formula has a number of applications. For instance, if n = pq

then there is exactly one λ ⊢ n satisfying λ1 = p and λ′1 = q, namely,
λ = hpq i, the partition with q parts equal to p. Hence
qi
2
fp,q (pq) = f hp .

3
Applying the hook length formula yields the surprising formula
 2
(pq)!
fp,q (pq) = .
11 22 · · · pp (p + 1)p · · · q p (q + 1)p−1 · · · (p + q − 1)1

1.2 Statistical properties of is(w)

Let us now turn to statistical properties of is(w). The most basic
property is the expectation E(n):
1 X
E(n) = is(w).
n! w∈S
n

√ is not hard to deduce from the Erdős-Szekeres theorem that E(n) ≥

It
n. Hammersley made a number of contributions regarding
√ E(n),
in particular showing the asymptotic formula E(n) ∼ c n for some
π
2
≤ c ≤ e and giving a heuristic argument that c = 2. It follows
from Schensted’s theorem that
1 X
2
E(n) = λ1 f λ . (2)
n! λ⊢n

Using this formula both Vershik and Kerov [58] and Logan and Shepp
[31] showed that c ≥ 2, and Vershik and Kerov added a clever ar-
gument based on the RSK algorithm that c ≤ 2. Hence c = 2 as
suggested by Hammersley. The work of Vershik-Kerov and Logan-
Shepp also determined the limiting shape Ψ (scaled so that the Young
diagram has area 1) of the partition λ ⊢ n that maximizes f λ . More-
over, “most” permutations w ∈ Sn have limiting shape approaching
(in a suitable sense) Ψ as n → ∞. If we rotate the the Young dia-
gram of λ 90◦ counterclockwise and normalize it to have area 1, then
the limiting shape is bounded by the x and y-axes, together with the
curve

x = y + 2 cos θ
2
y = (sin θ − θ cos θ),
π
for 0 ≤ θ ≤ π. See Figure 1.

4
 2

1.5

0.5

0 0.5 1 1.5 2
Figure 1: The curve y = Ψ(x)

A major breakthrough in understanding the behavior of is(w) was

achieved in 1999 by Baik, Deift, and Johansson [4]. They determined
the entire limiting distribution of is(w) as n → ∞. It turns out to
be given by the suitably scaled Tracy-Widom distribution, which
had first appeared in connection with the distribution of the largest
eigenvalue of a random hermitian matrix.
To describe these results, write isn (w) for is(w) in order to indicate
that w ∈ Sn . Let Ai(x) denote the Airy function, viz., the unique
solution to the second-order differential equation
Ai′′ (x) = x Ai(x),
subject to the condition
2 3/2
e− 3 x
Ai(x) ∼ √ 1/4 as x → ∞.
2 πx
Let u(x) denote the unique solution to the nonlinear third order
equation
u′′ (x) = 2u(x)3 + xu(x), (3)

5
subject to the condition

u(x) ∼ −Ai(x), as x → ∞.

Equation (3) is known as the Painlevé II equation, after Paul Painlevé

(1863–1933).
Now define the Tracy-Widom distribution to be the probability
distribution on R given by
 Z ∞ 
2
F (t) = exp − (x − t)u(x) dx . (4)
t

It is easily seen that F (t) is indeed a probability distribution, i.e.,

F ′ (t) ≥ 0, limt→∞ F (t) = 1, and limt→−∞ F (t) = 0. Let χ be a ran-
dom variable with distribution F , and let χn be the random variable
on Sn defined by √
isn (w) − 2 n
χn (w) = .
n1/6
We can now state the remarkable results of Baik, Deift, and Johans-
son.

Theorem 1.3. As n → ∞, we have

χn → χ in distribution,

i.e., for all t ∈ R,

lim Prob(χn ≤ t) = F (t). (5)

n→∞

As an example of the use of Theorem

√ 1.3, we state the following
improvement to the formula E(n) ∼ 2 n.

Corollary 1.4.
√
E(isn ) − 2 n
Z
lim = t dF (t) (6)
n→∞ n1/6
= −1.7711 · · · .

6
 The starting point for the proof of Theorem 1.3 is a formula of
Gessel [23] that determines the numbers

uk (n) = #{w ∈ Sn : isn (w) ≤ k}.

1 2n


For instance, it is known that u2 (n) = Cn := n+1 n
(a Catalan
number), but no simple formula exists for general uk (n). Define
X x2n
Uk (x) = uk (n) 2 , k ≥ 1
n≥0
n!
X x2n+i
Ii (2x) = , i ≥ 0.
n≥0
n! (n + i)!

The function Ii is the hyperbolic Bessel function of the first kind

of order i.

Theorem 1.5. We have

k
Uk (x) = det I|i−j| (2x) i,j=1
. (7)

Gessel’s theorem (Theorem 1.5) reduces the theorem of Baik, Deift,

and Johansson to “just” analysis, viz., the Riemann-Hilbert problem
in the theory of integrable systems, followed by the method of steep-
est descent to analyze the asymptotic behavior of integrable systems.
For further information see the survey [13] of Deift.

1.3 Matchings
There are many extensions and generalizations of the theory of in-
creasing and decreasing subsequences. See for instance equation (15)
and the last paragraph of Section 2.3 below. We will conclude this
section with a different generalization: crossings and nestings of
matchings. A (complete) matching on the set [2n] may be defined as
a partition M = {B1 , . . . , Bn } of [2n] into n two-element blocks Bi .
Thus B1 ∪ B2 ∪ · · · ∪ Bn = [2n], Bi ∩ Bj = ∅ if i 6= j, and #Bi = 2.
(These conditions are not all independent.) It is not immediately
apparent what should be the analogue of increasing and decreasing

7
 1 2 3 4 5 6 7 8 9 10

Figure 2: A matching on [10]

subsequences; in particular, what plays the role of the symmetry

a1 · · · an 7→ an · · · a1 that interchanges increasing and decreasing sub-
sequences of permutations?
Write Mn for the set of matchings on [2n]. We represent a matching
M ∈ Mn by a diagram of 2n vertices 1, 2, . . . , 2n on a horizontal line
in the plane, with an arc between vertices i and j and lying above
the vertices if {i, j} is a block of M. Sometimes we identify the block
{i, j} with the arc connecting i and j. Figure 2 shows the diagram
corresponding to the matching
M = {{1, 5}, {2, 9}, {3, 10}, {4, 8}, {6, 7}}.
Let M ∈ Mn . A crossing of M consists of two arcs {i, j} and
{k, l} with i < k < j < l. Similarly a nesting of M consists of two
arcs {i, j} and {k, l} with i < k < l < j. The maximum number
of mutually crossing arcs of M is called the crossing number of M,
denoted cr(M). Similarly the nesting number ne(M) is the maximum
number of mutually nesting arcs. For the matching M of Figure 2,
we have cr(M) = 3 (corresponding to the arcs {1, 5}, {2, 9}, and
{3, 10}), while also ne(M) = 3 (corresponding to {2, 9}, {4, 8}, and
{6, 7}).
Define
fn (i, j) = #{M ∈ Mn : cr(M) = i, ne(M) = j}.
It is well-known that
X X
fn (0, j) = fn (i, 0) = Cn . (8)
j i

In other words, the number of matchings M ∈ Mn with no crossings

(or with no nestings) is the Catalan number Cn . Equation (8) was
given the following generalization by Chen et al. [10].

8
Theorem 1.6. For all i, j, n we have fn (i, j) = fn (j, i).

The key to proving Theorem 1.6 is to find a “matching analogue”

rsk
of the RSK algorithm w → (P, Q) which Schensted connected with
increasing and decreasing subsequences (equation (1)). This algo-
rithm associates a matching M on [2n] with a pair Φ(M) = (R, S)
of oscillating tableaux of length n and the same shape. A standard
Young tableau of size n may be regarded as a sequence of partitions
of length n, starting with the empty partition and adding one square
at a time to the diagram of the partition. An oscillating tableau is
defined similarly, except that we can either add or delete a square at
each step. An example of an oscillating tableau of length 6 and shape
31 (the partition (3, 1)) is (∅, 1, 11, 21, 2, 21, 31). The pair (R, S) can
also be regarded as a single oscillating tableau of length 2n and shape
∅, by merging R with the reverse of S (identifying the last partition in
R with the last in S). Oscillating tableaux were first defined (though
not with that name) by Berele [7] in connection with the represen-
tation theory of the symplectic group. The matching analogue of
equation (1) is the following.

Theorem 1.7. Let M be a matching on [2n] and let Φ(M) = (∅ =

λ0 , λ1 , . . . , λ2n−1 , λ2n = ∅) be the corresponding oscillating tableau of
length 2n and shape ∅. Then ne(M) is equal to the the largest part
of any of the λi ’s, while cr(M) is equal to the most number of parts
of any of the λi ’s.

Theorem 1.6 follows readily from Theorem 1.7, via the bijections
M 7→ Φ(M) 7→ Φ(M)′ 7→ Φ−1 (Φ(M)′ ), where Φ(M)′ is obtained
from Φ(M) by conjugating all the partitions that appear in it.
√We can now ask for a matching analogue of the formula E(n) ∼
2 n for the expectation of is(w), or more generally, of the limiting
distribution given by Theorem 1.3. This question can be reduced to
finding the limiting distribution of ds(w) where w is a fixed-point free
involution in S2n . This problem had earlier been solved by Baik and
Rains [5][6], yielding the following result.

9
Theorem 1.8. We have for random (uniform) M ∈ Mn and all
t ∈ R that
√ !  Z ∞
nen (M) − 2n

t 1/2 1
lim Prob ≤ = F (t) exp u(s)ds ,
n→∞ (2n)1/6 2 2 t

where F (t) is the Tracy-Widom distribution and u(s) the Painlevé II

function. The same result holds with nen replaced with crn . In par-
ticular, the expectation M(n) of nen (M) or crn (M) satisfies M(n) ∼
√
2n.

2 Alternating permutations
2.1 The basic generating function
A permutation w = a1 a2 · · · an ∈ Sn is called alternating if a1 > a2 <
a3 > a4 < · · · . In other words, ai < ai+1 for i even, and ai > ai+1 for
i odd. Similarly w is reverse alternating if a1 < a2 > a3 < a4 > · · · .
Let En denote the number of alternating permutations in Sn . (Set
E0 = 1.) For instance, E4 = 5, corresponding to the permutations
2143, 3142, 3241, 4132, and 4231. The number En is called an Euler
number because Euler considered the numbers E2n+1 , though not
with the combinatorial definition just given. For a more extensive
survey of alternating permutations and Euler numbers, see [53].
The involution

a1 a2 · · · an 7→ n + 1 − a1 , n + 1 − a2 , · · · , n + 1 − an (9)

on Sn shows that En is also the number of reverse alternating per-

mutations in Sn .
The fundamental enumerative property of alternating permuta-

10
tions is due to Desiré André [1] in 1879. We have
X xn
En = sec x + tan x (10)
n≥0
n!
x2 x3 x4 x5 x6
= 1+x+ + 2 + 5 + 16 + 61
2! 3! 4! 5! 6!
7 8 9
x x x x10
+272 + 1385 + 7935 + 50521 +··· .
7! 8! 9! 10!
Note that sec x is an even function (i.e, sec(−x) = sec x), while
tan x is odd (tan(−x) = − tan x). It follows from equation (10) that
X x2n
E2n = sec x (11)
n≥0
(2n)!
X x2n+1
E2n+1 = tan x. (12)
n≥0
(2n + 1)!

For this reason E2n is sometimes called a secant number and E2n+1
a tangent number. Note that equations (11) and (12) can be used to
define sec x and tan x, and hence also cos x = 1/ sec x and sin x =
tan x/ sec x. The subject of “combinatorial trigonometry” seeks to
develop as much of trigonometry as possible based on these combi-
natorial definitions. As a first step, the reader is invited to show that
sec2 x = 1 + tan2 x [50, Exer. 5.7].

2.2 Other occurrences of Euler numbers

There are numerous situations not directly related to alternating per-
mutations in which the Euler numbers appear. A selection of these
occurrences of Euler numbers is given here.

• Complete increasing binary trees. A (plane) binary tree on the

vertex set [n] is defined recursively by having a root vertex v
and a left and right subtree of v which are themselves binary
trees or are empty. A binary tree is complete if every vertex
either has two children or is an endpoint. A binary tree on

11
 Figure 3: The two complete binary trees with five vertices

the vertex set [n] is increasing if every path from the root is
increasing. Figure 3 shows the two complete binary trees with
five vertices. Each one has eight increasing labelings, so there
are 16 complete increasing binary trees on [5].

Theorem 2.1. The number of complete increasing binary trees

on [2m + 1] is the Euler number E2m+1 . (There is a similar but
more complicated statement for the vertex set [2m] which we do
not give here.)

• Flip equivalence. The Euler numbers are related to increasing

binary trees in another way. The flip of a binary tree at a vertex
v is the binary tree obtained by interchanging the left and right
subtrees of v. Define two increasing binary trees T and T ′ on
the vertex set [n] to be equivalent if T ′ can be obtained from T
by a sequence of flips. Clearly this definition of equivalence is an
equivalence relation; the equivalence classes are in an obvious
bijection with increasing (1,2)-trees on the vertex set [n], that
is, increasing (rooted) trees so that every non-endpoint vertex
has one or two children. (These are not plane trees, i.e., the
order in which we write the children of a vertex is irrelevant.)
Figure 4 shows the five increasing (1,2)-trees on four vertices,
so f (4) = 5.

Theorem 2.2. We have f (n) = En (an Euler number).

• Simsun permutations.
Define a simsun permutation to be a permutation w = a1 · · · an ∈
Sn such that for all 1 ≤ k ≤ n, the subword of w consisting of

12
 1 1 1 1 1

2 2 2 4 2 3 3 2
3 3 4 3 4 4

Figure 4: The five increasing (1,2)-trees with four vertices

1, 2, . . . , k (in the order they appear in w) does not have three

consecutive decreasing elements. For instance, w = 425163
is not simsun since if we remove 5 and 6 from w we obtain
4213, which contains the three consecutive decreasing elements
421. Simsun permutations were named after Rodica Simion and
Sheila Sundaram and were first described in print by Sundaram
[56, §3]. They are a variant of a class of permutations due to
Foata and Schützenberger [17] known as André permutations.
For further information on simsun permutations see Chow [12].
We have chosen here to deal only with simsun permutations
because their definition is a little simpler than that of André
permutations. Simion and Sundaram prove in their paper the
following basic result on simsum permutations.
Theorem 2.3. The number of simsun permutations in Sn is
the Euler number En+1 .

Another proof of Theorem 2.3 was given by Maria Monks (pri-

vate communication, 2008). She gives a bijection between sim-
sun permutations in Sn and the increasing (1, 2)-trees on the
vertex set [n] discussed above. Simsun permutations have an
interesting connection with the cd-index of Sn , discussed below.

• Orbits of chains of partitions. A partition π of the set [n] is a

collection {B1 , . . . , Bk }Sof nonempty subsets of [n] (called the
blocks of π) such that Bi = [n] and Bi ∩ Bj = ∅ if i 6= j. Let
Πn denote the set of all partitions of [n]. If π, σ ∈ Πn , then
we say that π is a refinement of σ, denoted π ≤ σ, if every
block of π is contained in a block of σ. The relation ≤ is a
partial order, so Πn becomes a partially ordered set (poset).

13
Note that the symmetric group Sn acts on Πn in an obvious
way, viz., if B = {a1 , . . . , aj } is a block of π and w ∈ Sn , then
w · B := {w(a1 ), . . . , w(aj )} is a block of w · π.
Let M(Πn ) denote the set of all maximal chains of Πn , i.e., all
chains
π0 < π1 < · · · < πn−1 ,
so that for all 0 ≤ i ≤ n − 2, πi+1 is obtained from πi by
merging two blocks of πi . Thus πi has exactly n − i blocks.
In particular, π0 is the partition into n singleton blocks, and
πn−1 is the partition into one block [n]. The action of Sn on Πn
induces an action on M(Πn ). For instance, when n = 5 a class
of orbit representatives is given by the five chains below. We
write e.g. 12−34−5 for the partition {{1, 2}, {3, 4}, {5}}, and
we omit the first and last element of each chain.
12−3−4−5 < 123−4−5 < 1234−5
12−3−4−5 < 123−4−5 < 123−45
12−3−4−5 < 12−34−5 < 125−34
12−3−4−5 < 12−34−5 < 12−345
12−3−4−5 < 12−34−5 < 1234−5

Theorem 2.4. The number of orbits of the action of Sn on

M(Πn ) is the Euler number En−1 .

Theorem 2.4 was first proved by Stanley [44, Thm. 7.7] by

showing that the number of orbits satisfies the same recurrence
as En−1 . By elementary representation theory, the number of
orbits of Sn acting on M(Πn ) is the multiplicity of the trivial
representation in this action. This observation suggests the
problem of decomposing Sn -actions on various sets of chains
in Πn into irreducible representations. The first results in this
direction appear in [44, §7]. Many further results were obtained
by Sundaram [56]. Theorem 2.4 can also be proved by giving a
simple bijection between the orbits and increasing (1, 2)-trees
on [n − 1], as defined earlier in this section

14
• Polytope volumes. Euler numbers occur as (normalized) vol-
umes of certain convex polytopes. The first polyope, which
we call the zigzag order polytope Pn , consists of all points x =
(x1 , . . . , xn ) ∈ Rn satisfying

xi ≥ 0, 1 ≤ i ≤ n

x1 ≥ x2 ≤ x3 ≥ · · · xn .
To compute its volume, for each alternating permutation w =
a1 a2 · · · an ∈ Sn , let w −1 = b1 b2 · · · bn . Let

Pw = {(x1 , . . . , xn ) ∈ Pn : xb1 ≤ xb2 ≤ · · · ≤ xbn }.

It is easy to see that each Pw is a simplex with volume 1/n!.

One can check using the theory of P -partitions [48, §4.5] that
the Pw ’s have disjoint interiors and union Pn . Hence vol(Pn ) =
En /n!.
The second polytope is called the zigzag chain polytope Cn . It
consists of all points x = (x1 , . . . , xn ) ∈ Rn satisfying

xi ≥ 0, 1 ≤ i ≤ n

xi + xi+1 ≤ 1, 1 ≤ i ≤ n − 1.
The polytope Cn first arose in [14] and [43]. A “naive” method
for computing the volume is the following. For 0 ≤ t ≤ 1 let
Z t Z 1−x1 Z 1−x2 Z 1−xn−1
fn (t) = ··· dx1 dx2 · · · dxn . (13)
x1 =0 x2 =0 x3 =0 xn =0

Clearly f (1) = vol(Cn ). Differentiating equation (13) yields

fn′ (t) = fn−1 (1 − t). There are various ways to solve this re-
currence for fn (t) (with the initial conditions f0 (t) = 1 and
fn (0) = 0 for n > 0), yielding
X
fn (t)xn = (sec x)(cos(t − 1)x + sin tx).
n≥0

15
Putting t = 1 gives
X
fn (1)xn = sec x + tan x,
n≥0

so we conclude that vol(Cn ) = En /n!.

A more sophisticated proof uses the earlier mentioned fact that
vol(Pn ) = En /n!. Given (x1 , . . . , xn ) ∈ Rn , define ϕ(x1 , . . . , xn ) =
(y1 , . . . , yn ) ∈ Rn by

1 − xi , if i is odd
yi =
xi , if i is even.
It is easily checked that ϕ is an affine transformation taking Pn
onto Cn . Since the homogeneous part of ϕ has determinant ±1,
it follows that ϕ is a volume-preserving bijection from Pn onto
Cn , so vol(Cn ) = vol(Pn ) = En /n!. This argument appeared in
Stanley [46, Thm. 2.3 and Exam. 4.3].
The polytope Cn has an interesting connection to tridiagonal
matrices. An n × n matrix M = (mij ) is tridiagonal if mij = 0
whenever |i − j| ≥ 2. Let Tn be the set of all n × n tridiagonal
doubly stochastic matrices M, i.e., n × n (real) tridiagonal ma-
trices with nonnegative entries and with row and column sums
equal to 1. Thus Tn is a convex polytope in a real vector space
of dimension n2 (or of dimension 3n − 2 if we discard coordi-
nates that are always 0). It is easy to see that if we choose
the n − 1 entries m12 , m23 , . . . , mn−1,n arbitrarily, then they
determine a unique tridiagonal matrix M with row and column
sums 1. Moreover, in order for M to be doubly stochastic it is
necessary and sufficient that mi,i+1 ≥ 0 and

m12 + m23 ≤ 1, m23 + m34 ≤ 1, . . . , mn−2,n−1 + mn−1,n ≤ 1.

It follows that Tn is linearly equivalent to Cn−1 (in fact, Tn

projects bijectively to Cn−1 ). Moreover, the relative volume
(volume normalized so that a fundmental parallelopiped in the
intersection of the integer lattice with the affine span of Tn has
volume 1) of Tn is En−1 /(n − 1)!.

16
 • Singularities. V. I. Arnold [2] (see also [3] for a followup) has
shown that the Euler number En+1 is equal to the number
of components of the space of real polynomials f (x) = xn +
a1 xn−1 +· · ·+an−1 x whose critical points (zeros of f ′ (x)) are all
real and whose n−1 critical values (the numbers f (c) where c is
a critical point) are all different. For instance, when n = 3 the
polynomials x3 + ax2 + bx form a real plane. The critical points
are real if and only if b ≤ a2 /3. Two critical values coincide in
the region b < a2 /3 if and only if b = a2 /4 or b = 0. These
two curves cut the region b < a2 /3 into E4 = 5 components.
Arnold interprets this result in terms of morsifications of the
function xn+1 ; see his paper for further details. Arnold goes on
to deduce a number of interesting properties of Euler numbers.
He also extends the theory to morsifications of the functions
xn + y 2 and xy + y n , thereby producing Bn and Cn analogues
of Euler numbers (which correspond to the root system An ).

2.3 Longest alternating subsequences

In Section 1.2 we discussed properties of the length is(w) of the
longest increasing subsequence of a permutation a1 · · · an ∈ Sn . We
can ask whether similar results hold for alternating subsequences of
w ∈ Sn . In particular, for w ∈ Sn define as(w) (or asn (w) to make
it explicit that w ∈ Sn ) to be the length of the longest alternating
subsequence of w. For instance, if w = 56218347 then as(w) = 5, one
alternating subsequence of longest length being 52834. Our source
for most of the material in this section is the paper [52].
It turns out that the behavior of as(w) is much simpler than that
of is(w). The primary reason for this is the following lemma, whose
straightforward proof we omit.

Lemma 2.5. Let w ∈ Sn . Then there is an alternating subsequence

of w of maximum length that contains n.

Lemma 2.5 allows us to obtain explicit formulas by induction.

17
More specifically, define

ak (n) = #{w ∈ Sn : as(w) = k}

bk (n) = a1 (n) + a2 (n) + · · · + ak (n)

= #{w ∈ Sn : as(w) ≤ k}.

For instance, b1 (n) = 1, corresponding to the permutation 1, 2, . . . , n,

while b2 (n) = 2n−1, corresponding to the permutations u1 , u2 . . . , ui ,
n, v1 , v2 , . . . , vn−i−1 , where u1 < u2 < · · · < ui and v1 > v2 > · · · >
vn−i−1 . Using Lemma 2.5, we can obtain the following recurrence for
the numbers ak (n), together with the initial condition a0 (0) = 1:
n  
X n X
ak (n + 1) = (a2r (j) + a2r+1 (j))as (n − j). (14)
j=0
j 2r+s=k−1
r,s≥0

This recurrence can be used to obtain the following generating

function for the numbers ak (n) and bk (n). No analogous formula is
known for increasing subsequences.
Theorem 2.6. Let
X xn
A(x, t) = ak (n)tk
n!
k,n≥0
X xn
B(x, t) = bk (n)tk .
k,n≥0
n!
√
Set ρ = 1 − t2 . Then
2/ρ 1
B(x, t) = −
1 − ρ ρx ρ
1− e
t

A(x, t) = (1 − t)B(x, t).

Many consequences can be derived from Theorem 2.6. In particu-

lar, there are explicit formulas for ak (n) and bk (n). For instance, for

18
k ≤ 6 we have

b2 (n) = 2n−1
1
b3 (n) = (3n − 2n + 3)
4
1 n
b4 (n) = (4 − 2(n − 2)2n )
8
1
b5 (n) = (5n − (2n − 5)3n + 2(n2 − 5n + 5))
16
1
b6 (n) = (6n − 2(n − 3)4n + (2n2 − 12n + 15)2n ).
32
We can also obtain explicit formulas for the moments of as(w). For
instance, to obtain the mean (expectation)
1 X
D(n) = as(w),
n! w∈S
n

we compute
X ∂
D(n)xn = A(x, 1)
n≥1
∂t
6x − 3x2 + x3
=
6(1 − x)2
X 4n + 1
= x+ xn .
n≥2
6

Thus
4n + 1
D(n) = , n ≥ 2,
6
a remarkably simple formula. Note that (not surprisingly) D(n)√ is
much larger than the expectation E(n) of is(w), viz., E(n) ∼ 2 n.
Similarly the variance
1 X
V (n) = (as(w) − D(n))2
n! w∈S
n

19
is given by
8 13
V (n) = n− , n ≥ 4.
45 180
Now that we have computed the mean and variance of as(w), we
can ask whether there is an “alternating analogue” of the Baik-Deift-
Johansson formula (5). In other words, can we determine the scaled
limiting distribution

asn (w) − 2n/3

K(t) = lim Prob √ ≤t ,
n→∞ n

for t ∈ R? It turns out that the limiting distribution is Gaussian. It

is a consequence of results of Pemantle and Wilson [35] and Wilf [60],
and was proved directly by Widom [59]. More precisely, we have
√
t 45/4
1
Z
2
K(t) = √ e−s ds. (15)
π −∞

2.4 Umbral enumeration of subsets of alternat-

ing permutations
We now consider the enumeration of alternating permutations having
additional properties, such as having alternating inverses or having
no fixed points. The main tool is a certain character χτn of the sym-
metric group Sn , first considered by H. O. Foulkes [20][21], whose
dimension is En . We will not define this character here but will
simply state some results that follow from known connections be-
tween symmetric functions and permutation enumeration. We will
use umbral notation [24][39] for Euler numbers. In other words, any
polynomial in E is to be expanded in terms of powers of E, and then
E k is replaced by Ek . The replacement of E k by Ek is always the
last step in the evaluation of an umbral expression. For instance,

(E 2 − 1)2 = E 4 − 2E 2 + 1 = E4 − 2E2 + 1 = 5 − 2 · 1 + 1 = 4.

20
Similarly,
   
E E 2 E 3
(1 + t) = 1 + Et + t + t +···
2 3
1 1
= 1 + Et + (E 2 − E)t2 + (E 3 − 3E 2 + 2E)t3 + · · ·
2 6
1 1
= 1 + Et + (E2 − E1 )t2 + (E3 − 3E2 + 2E1 )t3 + · · ·
2 6
1 1
= 1 + 1 · t + (1 − 1)t2 + (2 − 3 · 1 + 2 · 1)t3 + · · ·
2 6
1 3
= 1+t+ t +··· .
6
A typical result is the following. Let g(n) (respectively, g ∗ (n)) de-
note the number of fixed-point-free alternating involutions (respec-
tively, fixed-point-free reverse alternating involutions) in S2n . Set
X
G(t) = g(n)xn
n≥0

= 1 + t + t2 + 2t3 + 5t4 + 17t5 + 72t6 + 367t7 + · · ·

X
G∗ (t) = g ∗ (n)xn
n≥0

= 1 + t2 + t3 + 4t4 + 13t5 + 59t6 + 308t7 + · · · .

Theorem 2.7. We have the umbral generating functions

 (E 2 +1)/4
1+t
G(t) =
1−t
G(t)
G∗ (t) = .
1+t
Another class of alternating permutations with a nice enumeration
are cycles of length n. For this particular case the resulting formu-
las can be “deumbralized” to give more explicit formulas. Let b(n)
(respectively, b∗ (n)) denote the number of alternating (respectively,
reverse alternating) n-cycles in Sn .

21
Theorem 2.8. (a) If n is odd then
1X
b(n) = b∗ (n) = µ(d)(−1)(d−1)/2 En/d .
n
d|n

(b) If n = 2k m where k ≥ 1, m is odd, and m ≥ 3, then

1X
b(n) = b∗ (n) = µ(d)En/d .
n
d|m

(c) If n = 2k and k ≥ 2 then

1
b(n) = b∗ (n) = (En − 1). (16)
n
(d) Finally, b(2) = 1, b∗ (2) = 0.
Note the curious fact that b(n) = b∗ (n) except for n = 2. Can
Theorem 2.8 be proved combinatorially, especially in the cases when
n in a prime power (when the sums have only two terms)?
There are umbral formulas for enumerating alternating and reverse
alternating permutations according to the length of their longest
increasing subsequence. These results are alternating analogues of
i
equation (7). For any integer i we use the notation
P D F (x)n for the
nth formal derivative of the power series F (x) = n≥0 an x , where
in particular
X xn+1
D −1 F (x) = an
n≥0
n+1

and D −i−1 = D −1 D −i for all i ≥ 1.

Theorem 2.9. Let αk (n) (respectively, αk′ (n)) denote the number of
alternating (respectively, reverse alternating) permutations w ∈ Sn
whose longest increasing subsequence has length at most k. Let
X
exp E tan−1 (x) = cn (E)xn


n≥0
1 1
= 1 + Ex + E 2 x2 + (E 3 − 2E)x3 + · · · .
2 6
22
Define
X xn
A1 (x) = cn (E)
n≥0
n!
√
A2 (x) = 1 + x2 A1 (x)
A1 (x)
A3 (x) = √ ,
1 + x2
and for 1 ≤ r ≤ 3 and k ≥ 1 define

Br,k (x) = det(D j−i Ar (x))ki,j=1 .

We then have:

• If n is odd, then the coefficient of xn /n! in the umbral evaluation

of B1,k (x) is αk (n) = αk′ (n).

• If n is even, then the coefficient of xn /n! in the umbral evalua-

tion of B2,k (x) is αk (n).

• If n is even, then the coefficient of xn /n! in the umbral evalua-

tion of B3,k (x) is αk′ (n).

Note that it follows from Theorem 2.9 that αk (2m + 1) = αk′ (2m +
1). This result is also an immediate consequence of the involution (9).
It is known (see [54]) that α2 (n) =
C⌈(n+1)/2⌉ for n ≥ 3 and α2′ (n) =
1 2i
C⌈n/2⌉ for n ≥ 1, where Ci = i+1 i
(a Catalan number). Can these
formulas be deduced directly from the case k = 2 of Theorem 2.9?
Similarly, J. Lewis [29] has shown that
(
′
f (m,m,m) , n = 2m
α3 (n) =
f (m−1,m,m+1) , n = 2m + 1,

where (as in Subsection 1.1) f λ denotes the number of SYT of shape

λ. No such results are known for α4′ (n) or α3 (n).
Is there an asymptotic formula for the expected length L(n) of the
longest increasing subsequence of an alternating permutation w ∈ Sn

23
as n → ∞, analogous to the result (2) of Logan-Shepp and Vershik- √
Kerov for arbitrary permutations? It is easy to see that L(n) ≥ n.
Is it true that limn→∞ loglogL(n)
n
= 12 ? Is there in fact a limiting distri-
bution for the (suitably scaled) length of the longest increasing sub-
sequence of an alternating permutation w ∈ Sn as n → ∞, analogous
to the Baik-Deift-Johansson theorem (5) for arbitrary permutations?
Two closely related problems are the following. Let aλ denote the
number of SYT of shape λ ⊢ n and descent set {1, 3, 5, . . . } ∩ [n − 1],
as defined in [50, p. 361]. (These are just the SYT Q that arise from
alternating permutations w ∈ Sn by applying the RSK algorithm
w 7→ (P, Q).) What is the (suitably scaled) limiting shape of λ as
n → ∞ that maximizes aλ and similarly that maximizes aλ f λ ? (For
the shape that maximizes f λ , see Figure 1. Could this shape also
maximize aλ ?)

2.5 The cd-index of the symmetric group

Let w = a1 · · · an ∈ Sn . The descent set D(w) of w is defined by
D(w) = {i : ai > ai+1 } ⊆ [n − 1].
A permutation w is thus alternating if D(w) = {1, 3, 5, . . . } ∩ [n − 1]
and reverse alternating if D(w) = {2, 4, 6, . . . }∩[n−1]. For S ⊆ [n−1]
let
βn (S) = #{w ∈ Sn : D(w) = S}.
The numbers βn (S) are fundamental invariants of w that appear in a
variety of combinatorial, algebraic, and geometric contexts. Here we
explain how alternating permutations are related to the more general
subject of permutations with a fixed descent set.
We first define for fixed n a noncommutative generating function
for the numbers βn (S). Given a set S ⊆ [n − 1], define its character-
istic monomial (or variation) to be the noncommutative monomial
uS = e1 e2 · · · en−1 , (17)
where 
a, if i 6∈ S
ei =
b, if i ∈ S.

24
For instance, D(37485216) = {2, 4, 5, 6}, so uD(37485216) = ababbba.
Define
X
Ψn = Ψn (a, b) = uD(w)
w∈Sn
X
= βn (S)uS . (18)
S⊆[n−1]

Thus Ψn is a noncommutative generating function for the numbers

βn (S). For instance,

Ψ1 = 1
Ψ2 = a+b
Ψ3 = a2 + 2ab + 2ba + b2
Ψ4 = a3 + 3a2 b + 5aba + 3ba2 + 3ab2 + 5bab + 3b2 a + b3 .

The polynomial Ψn is called the ab-index of the symmetric group Sn .

The main result of this section is the following.
Theorem 2.10. There exists a polynomial Φn (c, d) in the noncom-
muting variables c and d such that

Ψn (a, b) = Φn (a + b, ab + ba).

The polynomial Φn (c, d) is called the cd-index of Sn . For instance,

we have

Ψ3 (a, b) = a2 + 2ab + 2ba + b2 = (a + b)2 + (ab + ba),

so Φ3 (c, d) = c2 +d. Some values of Φn (c, d) for small n are as follows:

Φ1 = 1
Φ2 = c
Φ3 = c2 + d
Φ4 = c3 + 2cd + 2dc
Φ5 = c4 + 3c2 d + 5cdc + 3dc2 + 4d2
Φ6 = c5 + 4c3 d + 9c2 dc + 9cdc2 + 4dc3 + 12cd2 + 10dcd + 12d2 c.

25
If we define deg(c) = 1 and deg(d) = 2, then the number of cd-
monomials of degree n − 1 is the Fibonacci number Fn . It is not hard
to see that all these monomials actually appear in Φn (c, d). Thus
Φn (c, d) has Fn terms, compared with 2n−1 terms for Ψn (a, b).
There are several known proofs of Theorem 2.10. Perhaps the most
natural approach is to define an equivalence Prelation on Sn such that
for each equivalence class C, we have that w∈C uD(w) is a monomial
in c = a+ b and d = ab+ ba. Such a proof was given by G. Hetyei and
E. Reiner [26]. See [49, §1.6] for an exposition of this proof. The proof
of Hetyei and Reiner leads to a somewhat complicated combinatorial
interpretation of the coefficients of Φn (c, d) which makes it apparent
that they are nonnegative. It is reasonable to ask whether there is
a more “direct” description of the coefficients. Such a description
was first given by D. Foata and M.-P. Schützenberger [17] in terms of
the André permutations mentioned in Section 2.2. We state here the
analogous result for simsun permutations (as defined in Section 2.2),
due to R. Simion and S. Sundaram.

Theorem 2.11. Let µ be a monomial of degree n − 1 in the non-

commuting variables c, d, where deg(c) = 1 and deg(d) = 2. Replace
each c in µ with 0, each d with 10, and remove the final 0. We get
the characteristic vector of a set Sµ ⊆ [n − 2]. Then the coefficient of
µ in Φn (c, d) is equal to the number of simsun permutations in Sn−1
with descent set Sµ .

For example, if µ = cd2 c2 d then we get the characteristic vector

01010001 of the set Sµ = {2, 4, 8}. Hence the coefficient of cd2 c2 d in
Φ10 (c, d) is equal to the number of simsun permutations in S9 with
descent set {2, 4, 8}.
Note that every cd-monomial, when expanded in terms of ab-
monomials, is a sum of distinct monomials including bababa · · · and
ababab · · · . These monomials correspond to descent sets of alter-
nating and reverse alternating permutations, respectively. Hence
Φn (1, 1) = En . This fact also follows immediately from Theorems 2.3
and 2.11. Since the coefficients of Φn (c, d) are nonnegative and the
expansion of every cd-monomial contains bababa · · · and ababab · · · ,

26
it follows that

βn (S) ≤ βn (1, 3, 5, . . . ) = βn (2, 4, 6, · · · ) = En . (19)

Moreover, with just a little more work it is easy to see that equal-
ity holds in equation (19) if and only if S = {1, 3, 5, . . . } or S =
{2, 4, 6, . . . }. This result is originally due to Niven [34] and de Bruijn
[9]; the proof based on the cd-index appears in [47, Thm. 2.3(b)].
All three references actually prove a more general result about when
βn (S) ≤ βn (T ).

3 Reduced decompositions
Let w ∈ Sn , and let

p = ℓ(w) = #{(i, j) : i < j, w(i) > w(j)},

the number of inversions of w, or the length of w when we regard

Sn as a Coxeter group. Write si = (i, i + 1) ∈ Sn , 1 ≤ i ≤ n − 1.
It is easy to see that w cannot be written as a product of fewer
than p adjacent transpositions si . A reduced decomposition of w
is a sequence (r1 , r2 , . . . , rp ) of integers 1 ≤ ri ≤ n − 1 such that
w = sr1 sr2 · · · srp . Write R(w) for the set of reduced decompositions
of w and r(w) = #R(w), the number of reduced decompositions
of w. A theorem of Tits [57][8, Thm. 3.3.1(ii)] asserts that any two
reduced decompositions can be obtained from each other by applying
the relations si sj = sj si if |i − j| ≥ 2, and si si+1 si = si+1 si si+1 . For
instance, one of the reduced decompositions of w = 24351 is (using
abbreviated notation where we omit commas and parentheses) 12324.
It follows that R(w) = {12324, 12342, 13234, 31234} and r(w) = 4.
Given a permutation w ∈ Sn , define a power series Gw in the
infinitely many variables x1 , x2 , . . . by
X X
Gw = xi1 · · · xip . (20)
(r1 ,...,rp )∈R(w) 1≤i1 ≤···≤ip
ij <ij+1 if rj <rj+1

27
Thus Gw is homogeneous of degree p = ℓ(w). For instance, let w =
321, so R(w) = {121, 212}. Then
X X
G321 = xi xj xk + xi xj xk .
1≤i<j≤k 1≤i≤j<k

Note that G321 is a symmetric function of the xi ’s. The homogeneous

symmetric functions of degree p have a well-known basis consisting
of the Schur functions sλ for λ ⊢ p. In particular, the expansion of
G321 in terms of Schur functions is simply G321 = s21 (where 21 is
abbreviated notation for the partition (2, 1, 0, 0, . . . )). The following
key lemma was proved by Stanley [45]. Several other proofs have
subsequently been given. Perhaps the most straightforward is a proof
of Fomin and Stanley [19] based on the nilCoxeter algebra of Sn .

Lemma 3.1. For all permutations w ∈ Sn , the power series Gw is a

symmetric function.

By Lemma 3.1 we can define coefficients αwλ for w ∈ Sn and λ ⊢

p = ℓ(w) by X
Gw = αwλ sλ . (21)
λ⊢p

Because Gw is an integer linear combination of monomials it follows

that αwλ ∈ Z, but at this point it is unclear whether αwλ ≥ 0. Note
that it follows immediately from the definition of Gw that r(w) is the
coefficient of x1 x2 · · · xp in Gw . On the other hand, the coefficient
of x1 x2 · · · xp in sλ (where λ ⊢ p) is just f λ , the number of SYT of
shape λ. Hence we obtain the following “formula” for r(w).

Corollary 3.2. Let w ∈ Sn , ℓ(w) = p. Then

X
r(w) = αwλ f λ .
λ⊢p

Of course the usefulness of Corollary 3.2 depends on what we can

say about the numbers αwλ. (Recall from Section 1 that f λ is given
explicitly by the “hook length formula,” so we may regard f λ as
known.) The simplest situation is when Gw = sλ for some λ ⊢

28
n. We can say precisely when this happens. Define a permutation
a1 a2 · · · an ∈ Sn to be 2143-avoiding or vexillary if there do not exist
1 ≤ h < i < j < k ≤ n for which ai < ah < ak < aj . (Numerous
equivalent definitions exist.)
Suppose that w = a1 · · · an ∈ Sn . For 1 ≤ i ≤ n − 1 define

ci = #{j : i < j ≤ n, aj < ai },

and define λ(w) to be the partition whose parts are the ci ’s (sorted
into decreasing order).

Theorem 3.3. We have Gw = sλ for some λ if and only if w is

vexillary. In this case λ = λ(w), so r(w) = f λ(w) .

For instance, let w = 5361472 ∈ S7 . Then w is vexillary, and

(c1 , . . . , c6 ) = (4, 2, 3, 0, 1, 1, 0). Hence λ(w) = 43211, Gw = s43211 ,
and r(w) = f 43211 = 2310.
An especially interesting special case of Theorem 3.3 occurs for the
permutation w0 = n, n − 1, . . . , 1 ∈ Sn . Note
that w0 is the unique
n
longest permutation in Sn , and ℓ(w0 ) = 2 . Clearly w0 is vexillary,
and λ(w0 ) = (n − 1, n − 2, . . . , 1). We obtain from the hook length
formula the following corollary.

Corollary 3.4. For w0 ∈ Sn we have

n


!
r(w0 ) = f (n−1,n−2,...,1) = n−1 n−2
2
n−3
.
1 3 5 · · · (2n − 1)

Corollary 3.4 suggests to an enumerative combinatorialist the prob-

lem of finding a “simple” bijective proof that r(w0 ) = f (n−1,n−2,...,1) .
A remarkable such proof was given by Edelman and Greene [15]. It is
also a consequence of the theory developed by Edelman and Greene
that the coefficients αwλ of equation (21) are nonnegative. An ele-
gant combinatorial interpretation of αwλ was later given by Fomin
and Greene [18], which we now describe.
A semistandard Young tableau (SSYT) T of shape λ is a Young
diagram of shape λ with a positive integer inserted into each square so
that the rows are weakly increasing and columns strictly increasing.

29
The row reading word of T is the sequence obtained by reading the
entries of T beginning with the first row from right-to-left, then the
second row right-to-left, etc.
Theorem 3.5. Let w ∈ Sn , ℓ(w) = p, and λ ⊢ p. The coefficient
αwλ is equal to the number of SSYT of shape λ whose row reading
word is a reduced decomposition of w.
As an example of Theorem 3.5, let w = 4152736. The SSYT whose
row reading words are a reduced decomposition of w are

123 123 1236

34 346 34
56 5 5,

with row reading words 3214365, 3216435, and 6321435. Hence

r(w) = f 322 + f 331 + f 421 = 21 + 21 + 35 = 77.

The symmetric function Gw has a connection with representation

theory. Let ϕ : Sp → GL(m, C) be an ordinary (complex) represen-
tation of Sp of dimension m. The irreducible representations ϕλ of
Sp are indexed by partitions λ ⊢ p. Let mϕ (λ) be the multiplicity
of ϕλ in ϕ, and define the (Frobenius) characteristic of ϕ to be the
symmetric function X
ch ϕ = mϕ (λ)sλ .
λ⊢p

Now let µ be the (Young) diagram of any partition µ ⊢ p. There

is a standard way, involving column antisymmetrizers of tabloids of
shape µ, of constructing a module Mµ , called the Specht module, that
affords the irreducible representation ϕµ . See e.g. Sagan [40, §2.3].
The Specht module construction can be carried out for any finite
diagram D, i.e., any finite subset of an infinite grid of squares, giving
a module MD for a representation ϕD of Sp , where p = #D.
For any permutation w ∈ Sn we can define its diagram Dw as
follows. Consider the squares of an infinite quadrant of squares as
being coordinatized by P × P, where P = {1, 2, . . . }. If 1 ≤ i ≤ n
and w(i) = j, then remove all squares (h, k) with h = i and k > j,

30
 Figure 5: The diagram of the permutation w = 351624

or with k = j and i > h. What remains will be Dw , which is easily

seen to have p = ℓ(w) squares. For instance, Figure 5 shows the
diagram of w = 351624, with a dot in the center of the squares (i, j)
for which w(i) = j (using matrix coordinates, i.e., rows increase from
top-to-bottom and columns from left-to-right). The following result
is due to Kraśkiewicz and Pragacz [28] (first written up in 1986 but
not published until 2004); see also Kraśkiewicz [27].
Theorem 3.6. Let w ∈ Sn , with diagram Dw . Then ch ϕDw = Gw ,
so in particular dim ϕDw = r(w).
It is still not understood for arbitrary diagrams D how to decom-
pose MD into irreducible modules or even to give a “nice” description
of dim MD , though there is a generalization of Theorem 3.6 due to
Reiner and Shimozono [37]. Some further cases where dim MD and
ch ϕD can be computed are considered by Liu [30].
The symmetric functions Gw are also connected with flag varieties.
The basic result is the following. Let Fln denote the set of all com-
plete flags 0 = V0 ⊂ V1 ⊂ · · · ⊂ Vn = Cn of subspaces in Cn (so
dim Vi = i). The set Fln has the natural structure of a smooth pro-
jective variety; in fact, it is canonically isomorphic to the coset space
GL(n, C)/B, where B is the Borel subgroup of all invertible upper
triangular matrices. For every w ∈ Sn there is an affine subvariety
Ωw of Fln called a Schubert variety (see e.g. [22, Chap. 10]). The clo-
n
sure Ωw represents a class [Ωw ] in the cohomology H 2(( 2 )−p) (Fln ; C),
and these classes form a basis for H ∗ (Fln ; C).

31
 We want a more explicit description of the cohomology ring H ∗ (Fln ; C)
with its distinguished basis [Ωw ], w ∈ Sn . To this end define the
Schubert polynomial Sw = Sw (x1 , . . . , xn−1 ) by
X X
Sw = xi1 · · · xip .
(r1 ,...,rp )∈R(w) 1≤i1 ≤···≤ip
ij <ij+1 if rj <rj+1
ij ≤j

(There are several equivalent definitions; we choose the one most

convenient here.) Note the similarity to the definition (20) of Gw . In
fact, if we define for w ∈ Sn and N ≥ 1 the permutation 1N × w ∈
Sn+N by
(1N × w)(i) = w(i) + N if 1 ≤ i ≤ n,
(1N × w)(n + 1) < (1N × w)(n + 2) < · · · < (1N × w)(n + N),
then it is clear that
Gw = lim S1N ×w .
N →∞

For this reason Gw is sometimes called a stable Schubert polynomial.

Now let Rn = C[x1 , x2 , . . . , xn ]/In , where the ideal In is generated
by the elementary symmetric functions e1 , e2 , . . . , en in the xi ’s. Then
there is an algebra isomorphism

ϕ : Rn → H ∗ (Fln ; C),

such that for w ∈ Sn we have ϕ(Sw0w ) = [Ωw ], where w0 = n, n −

1, . . . , 1. For a proof, see e.g. [32, (A.5)][33, §3.6.4]. This isomorphism
shows the primary geometric significance of Schubert polynomials.
There is an interesting identity involving reduced decompositions
that is related to Schubert polynomials.

Theorem 3.7. Let w ∈ Sn and ℓ(w) = p. Then

X
r1 r2 · · · rp = p! Sw (1, 1, . . . , 1).
(r1 ,r2 ,...,rp )∈R(w)

Theorem 3.7 was first proved by Macdonald [32, (6.11)]. Subse-

quently a simpler proof, based on the nilCoxeter algebra of Sn , was

32
given by Fomin and Stanley [19]. This proof was extended to give a
q-analogue of Theorem 3.7 originally conjectured by Macdonald [32,
(6.11q ?)].
It is known [32, (4.7)][33, Prop. 2.6.7] that Sw is a single monomial
if and only if w = a1 · · · an is 132-avoiding (also called dominant), i.e.,
there does not exist i < j < k such that ai < ak < aj . The number
of 132-avoiding permutations in Sn is the Catalan number Cn . Note
in particular that w0 is 132-avoiding. It follows from Theorem 3.7
that if w is 132-avoiding then
X
r1 r2 · · · rp = p!. (22)
(r1 ,r2 ,...,rp )∈R(w)

There is a curious analogue of equation (22), going back to Cheval-

ley [11] and made more explicit by Stembridge [55] in the case w =
w0 , connected with degrees of Schubert varieties, Bruhat order, etc.
We simply state the result here; for further details see [36]. Let (i, j)
denote the transposition interchanging i and j. Given w ∈ Sn of
length p, let
T (w) = {((i1 , j1 ), (i2 , j2 ), . . . , (ip , jp )) : w = (i1 , j1 )(i2 , j2 ) · · · (ip , jp )
and ℓ((i1 , j1 ) · · · (ik , jk )) = k for all 1 ≤ k ≤ p}.
Then for w = w0 (so p = n2 ) we have

X
(j1 − i1 )(j2 − i2 ) · · · (jp − ip ) = p!. (23)
((i1 ,j1 ),(i2 ,j2 ),...,(ip ,jp ))∈T (w0 )

For instance, if w = 321 then (abbreviating the transposition (i, j)

as ij)
T (321) = {(12, 23, 12), (23, 12, 23), (12, 13, 23), (23, 13, 12)}.
Hence we get
1 · 1 · 1 + 1 · 1 · 1 + 1 · 2 · 1 + 1 · 2 · 1 = 3!.
Open problem. Is the similarity between equation (22) in the case
w = w0 and equation (23) just a coincidence? Can either equation
be given a combinatorial proof?

33
References
[1] D. André, Développement de sec x and tg x, C. R. Math. Acad.
Sci. Paris 88 (1879), 965–979.

[2] V. I. Arnold, Bernoulli-Euler updown numbers associated with

function singularities, their combinatorics and arithmetics, Duke
Math. J. 63 (1991), 537–555.

[3] V. I. Arnold, Springer numbers and morsification spaces, J. Al-

gebraic Geometry 1 (1992), 197–214.

[4] J. Baik, P. Deift, and K. Johansson, On the distribution of the

length of the longest increasing subsequence of random permu-
tations, J. Amer. Math. Soc. 12 (1999), 1119–1178.

[5] J. Baik and E. Rains, Symmetrized random permutations, in

Random Matrix Models and Their Applications, Math. Sci. Res.
Inst. Publ. 40, Cambridge University Press, Cambridge, 2001,
1–19.

[6] J. Baik and E. Rains, The asymptotics of monotone subse-

quences of involutions, Duke Math. J. 109 (2001), 205–281.

[7] A. Berele, A Schensted-type correspondence for the symplectic

group, J. Combin. Theory Ser. A 43 (1986), 320–328.

[8] A. Björner and F. Brenti, Combinatorics of Coxeter Groups,

Graduate Texts in Mathematics 231, Springer-Verlag, New
York, 2005.

[9] N. G. de Bruijn, Permutations with given ups and downs, Nieuw

Arch. Wisk. 18 (1970), 61–65.

[10] W. Y. C. Chen, E. Y. P. Deng, R. R. X. Du, R. Stanley, and

C. H. Yan, Crossings and nestings of matchings and partitions,
Trans. Amer. Math. Soc. 359 (2007), 1555–1575.

34
[11] C. Chevalley, Sur les décompositions cellulaires des espaces G/B,
in Proceedings of Symposia in Pure Mathematics, 56, Part 1,
Algebraic groups and their generalizations: Classical methods
(University Park, PA, 1991), 1–23, American Mathematical So-
ciety, Providence, RI, 1994.
[12] Chak-On Chow, Counting simsun permutations by descents,
Ann. Combinatorics, to appear.
[13] P. Deift, Integrable systems and combinatorial theory, Notices
Amer. Math. Soc. 47 (2000), 631–640.
[14] E. E. Doberkat, Problem 84-20, SIAM Review 26 (1984), 580.
[15] P. Edelman and C. Greene, Balanced tableaux, Adv. in Math.
63 (1987), 42–99.
[16] P. Erdős and G. Szekeres, A combinatorial problem in geometry,
Composito Math. 2 (1935), 463–470.
[17] D. Foata and M.-P. Schützenberger, Nombres d’Euler et permu-
tations alternantes, in A Survey of Combinatorial Theory (Proc.
Internat. Sympos., Colorado State Univ., Fort Collins, Colo.,
1971), North-Holland, Amsterdam, 1973, pp. 173–187.
[18] S. Fomin and C. Greene, Noncommutative Schur functions and
their applications, Discrete Math. 193 (1998), 179–200.
[19] S. Fomin and R. Stanley, Schubert polynomials and the nilCox-
eter algebra, Advances in Math. 103 (1994), 196–207.
[20] H. O. Foulkes, Enumeration of permutations with prescribed up-
down and inversion sequences, Discrete Math. 15 (1976), 235–
252.
[21] H. O. Foulkes, Tangent and secant numbers and representations
of symmetric groups, Discrete Math. 15 (1976), 311–324.
[22] W. Fulton, Young Tableaux, Student Texts 35, London Math-
ematical Society, Cambridge University Press, Cambridge/New
York, 1997.

35
[23] I. M. Gessel, Symmetric functions and P-recursiveness, J. Com-
bin. Theory Ser. A 53 (1990), 257–285.
[24] I. M. Gessel, Applications of the classical umbral calculus, Al-
gebra Universalis 49 (2003), 397–434.
[25] C. Greene, An extension of Schensted’s theorem, Advances in
Math. 14 (1974), 254-265.
[26] G. Hetyei and E. Reiner, Permutation trees and variation statis-
tics, European J. Combin. 19 (1998), 847–866.
[27] W. Kraśkiewicz, Reduced decompositions in Weyl groups, Eu-
rop. J. Comb. 16 (1995), 293–313.
[28] W. Kraśkiewicz and P. Pragacz, Schubert polynomials and Schu-
bert functors, Europ. J. Comb. 25 (2004), 1327–1344.
[29] J. B. Lewis, Pattern avoidance and RSK-like algorithms for al-
ternating permutations and Young tableaux, arXiv:0909.4966.
[30] R. I. Liu, Matching polytopes and Specht modules, preprint,
arXiv:0910.0523.
[31] B. F. Logan and L. A. Shepp, A variational problem for random
Young tableaux, Advances in Math. 26 (1977), 206–222.
[32] I. G. Macdonald, Notes on Schubert Polynomials, Laboratoire de
combinatoire et d’informatique mathématique (LACIM), Uni-
versité du Québec à Montréal, Montreal, 1991.
[33] L. Manivel, Symmetric Functions, Schubert Polynomials and
Degeneracy Loci, SMF/AMS Texts and Monographs 6, Ameri-
can Mathematical Society, Providence, RI, 2001.
[34] I. Niven, A combinatorial problem of finite sequences, Nieuw
Arch. Wisk. 16 (1968), 116–123.
[35] R. Pemantle and M. C. Wilson, Asymptotics of multivariate
sequences. I. Smooth points of the singular variety, J. Combin.
Theory Ser. A 97 (2002), 129–161.

36
[36] A. Postnikov and R. Stanley, Chains in the Bruhat order, J.
Algebraic Combinatorics 29 (2009), 133–174.

[37] V. Reiner and M. Shimozono, Percentage-avoiding, northwest

shapes and peelable tableaux, J. Combinatorial Theory (A) 82
(1998), 1–73.

[38] G. de B. Robinson, On the representations of Sn , Amer. J. Math.

60 (1938), 745–760.

[39] G.-C. Rota and B. Taylor, The classical umbral calculus, SIAM
J. Math. Anal. 25 (1994), 694–711.

[40] B. E. Sagan, The Symmetric Group, second ed., Springer, New

York, 2001.

[41] C. E. Schensted, Longest increasing and decreasing subse-

quences, Canad. J. Math. 13 (1961), 179–191.

[42] A. Seidenberg, A simple proof of a theorem of Erdős and Szek-

eres, J. London Math. Soc. 34 (1959), 352.

[43] R. Stanley, Elementary problem E 2701, American Math.

Monthly 85 (1978), 197; solution by I. G. Macdonald and R.
B. Nelsen (independently), 86 (1979), 396.

[44] R. Stanley, Some aspects of groups acting on finite posets, J.

Combin. Theory Ser. A 32 (1982), 132–161.

[45] R. Stanley, On the number of reduced decompositions of ele-

ments of Coxeter groups, European J. Combinatorics 5 (1984),
359–372.

[46] R. Stanley, Two poset polytopes, Discrete Comput. Geom. 1

(1986), 9–23.

[47] R. Stanley, Flag f -vectors and the cd-index, Math. Z. 216

(1994), 483–499.

37
[48] R. Stanley, Enumerative Combinatorics, vol. 1, Wadsworth and
Brooks/Cole, Pacific Grove, CA, 1986; second printing, Cam-
bridge University Press, New York/Cambridge, 1996.

[49] R. Stanley, Enumerative Combinatorics, vol. 1, second edition,

in preparation; Chapter 1 available at

math.mit.edu/∼rstan/ec/ch1.pdf.

[50] R. Stanley, Enumerative Combinatorics, vol. 2, Cambridge Uni-

versity Press, New York/Cambridge, 1999.

[51] R. Stanley, Increasing and decreasing subsequences and their

variants, Proc. Internat. Cong. Math. (Madrid, 2006), vol. 1,
American Mathematical Society, Providence, RI, 2007, pp. 545–
579.

[52] R. Stanley, Longest alternating subsequences of permutations,

Michigan Math. J. 57 (2008), 675–687.

[53] R. Stanley, Alternating permutations, preprint.

[54] R. Stanley, Catalan addendum,

math.mit.edu/∼rstan/ec/catadd.pdf.

[55] J. R. Stembridge, A weighted enumeration of maximal chains in

the Bruhat order, J. Algebraic Combinatorics 15 (2002), 291–
301.

[56] S. Sundaram, The homology representations of the symmetric

group on Cohen-Macaulay subposets of the partition lattice,
Adv. Math. 104 (1994), 225–296.

[57] J. Tits, Le problème des mots dans les groupes de Coxeter, Sym-
posia Mathematics (INDAM, Rome, 1967/68), Academic Press,
London, 1969, vol. 1, pp. 175–185.

[58] A. M. Vershik and S. V. Kerov, Asymptotic behavior of the

Plancherel measure of the symmetric group and the limit form of

38
 Young tableaux (Russian). Dokl. Akad. Nauk SSSR 223 (1977),
1024–1027. English translation: Soviet Math. Dokl. 233 (1977),
527–531.

[59] H. Widom, On the limiting distribution for the longest alternat-

ing subsequence in a random permutation, Electron. J. Combin.
13(1) (2006), Article R25.

[60] H. S. Wilf, Real zeroes of polynomials that count runs and de-
scending runs, preprint, 1998.

39