You are on page 1of 27

The Applications of Total Positivity

to Combinatorics, and conversely


Francesco Brenti1

Introduction

Total positivity is an important and powerful concept that arises often in various
branches of mathematics, statistics, probability, mechanics, economics, and computer science ( see, e.g., [24], and the references cited there). In this paper we give
a survey of the interactions between total positivity and combinatorics.
The historical origins of these connections date from the mid 1980s when it
was observed that many total positivity techniques and results had applications to
unimodality problems arising in combinatorics. Many different fields of mathematics had been previously applied to unimodality problems, but not total positivity.
This approach proved to be extremely successful and many unimodality problems
that had resisted attack by other techniques were solved in this way. In turn, the
combinatorial problems themselves suggested the study of questions in total positivity that had never been studied before. One of these is the study of linear and
bilinear transformations that preserve the P F or T P property of a sequence or matrix, respectively. Some combinatorialists initiated a study of these questions and
then applied their results to the combinatorial problems which originally motivated
them. We survey these results in 3 and several outstanding open problems in 6
which show that the theory of linear and bilinear transformations which preserve
the P F or T P properties, though rich with non-trivial results, has yet to reach its
full maturity.
Up to this point the applications of total positivity to combinatorics had been
mainly applications of the theory of P F sequences. At the end of the 1980s it
was slowly realized that, in fact, not only P F sequences but also T P matrices are
quite ubiquitous in combinatorics. While most of the familiar combinatorial matrices (such as, for example, the binomial and Stirling matrices) could be easily proved
to be T P as an application of the general theory of total positivity, it also became
apparent that for many of them it is in fact possible to give a combinatorial interpre1

Partially supported by NSF grant DMS 9304580 and by EC grant No. CHRX-CT93-0400.

tation of their minors, thus showing, in particular, that they are nonnegative. These
were the first examples of combinatorial proofs of total positivity, though the basic
ideas used in them were already present in works of Karlin-McGregor, Lindstrom
and Gessel-Viennot. Continuing this train of thought leads to a completely combinatorial characterization of totally positive matrices, which automatically affords
a combinatorial interpretation of all their minors. This result, and several variants
of it, are extremely powerful tools for proving that an impressive variety of matrices arising in combinatorics (such as various matrices of (p, q)-Stirling numbers of
both kinds, q-Lah numbers, etc...) are indeed T P . For many of these results the
combinatorial proof is the only one known to date. The combinatorial approach
is also useful from a theoretical point of view. For example, the well known property that T P matrices are closed under products is obvious from the combinatorial
standpoint, and more difficult results (such as, e.g., that immanants of T P matrices
are nonnegative) have simpler proofs in this setting. We survey these results in 4.
Despite the power of the combinatorial approach, however, many matrices arising
from combinatorics are still only conjectured to be T P , and we survey the main
conjectures in this area in 6.
Recently there has been much interest in combinatorics in extending what is
known for the symmetric group to other finite Coxeter groups. In particular, concepts that generalize total positivity have been introduced and studied for any Coxeter group. These concepts coincide with usual total positivity in the case of the
symmetric group. These studies are still in their infancy, and we survey them in
5 and point out some promising directions for further research in 6. Given the
wide applicability that total positivity has it is very reasonable to expect that these
natural generalizations will prove equally useful.

Notation and Preliminaries

In this section we collect some definitions, notation and results that will be used
def

def

in the rest of the paper. We let P = {1, 2, 3, . . .}, N = P {0}, Q be the set of
def

rational numbers, R be the set of real numbers and R+ = {x R : x 0}; for


def

def

a N we let [a] = {1, 2, . . . , a} (where [0] = ). The cardinality of a set A will be


denoted by |A|, for r N we let
i

A 
r

def

= {T A : |T | = r}. Given a polynomial

P (x), and i Z, we denote by [x ](P (x)) the coefficient of xi in P (x).


We follow [41] for general combinatorial notation and terminology. In particular,
2

given a (finite) set T we denote by (T ) the set of all (set) partitions of T (see,
e.g., [41], p.33 for further information about partitions of a set), and by S(T ) the
def

set of all bijections from T to itself. For n P, we let Sn = S([n]). Throughout


def

this work, p, q, and x will denote independent variables. For n P we let [n]p,q =
def

def

pn1 + q pn2 + q 2 pn3 + . . . + q n1 , [n]q = [n]1,q , and [k]p,q = 0 if k 0.


Given a commutative ring F (usually Z, Q, or R) we denote by F [[x]] the ring of
formal power series in the one variable x with coefficients in F . We refer the reader
to [41], 1.1, pp. 3-8, for the fundamental properties of F [[x]] which we will use.
We follow [30], I.1, for notation and terminology related to partitions of integers.
def

Recall that given n P a partition of n is a sequence = (1 , . . . , r ) (for some


r P) of positive integers such that 1 . . . r > 0 and
the length of and write r = l().

Pr

i=1

i = n. We call r

An infinite (real) matrix M = (Mn,k )n,kN (where Mn,k is the entry in the n-th
row and k-th column of M) is said to be totally positive (or, T P , for short) if every
minor of M has nonnegative determinant. An infinite (real) sequence {ai }iN is said
to be a Polya frequency sequence (or, a P F -sequence, for short) if the infinite matrix
def

def

A = (ank )n,kN (where ai = 0 if i < 0) is totally positive. Given a matrix M we


denote by M the transpose of M. We will often work with matrices and sequences
whose elements are real polynomials over some (possibly infinite) set of independent
variables x. In this case we say that such a matrix M = (Mn,k )n,kN is x-T P if the
determinant of any minor of M is a polynomial with nonnegative coefficients and we
define xP F in the natural way. Given a sequence of independent variables {xi }iN
def

and T N we let xT =

def

iT

def

xi . We also let 0 = (0, 0, 0, . . .) and 1 = (1, 1, 1, . . .).

Total Positivity and Unimodality

As noted in the Introduction, the interactions between total positivity and combinatorics have their origin in the observation that total positivity results and techniques
can be effectively used in the solution of unimodality problems. In this section we
describe some of these problems and their solution using total positivity, as well as
total positivity results that have been motivated by unimodality problems.
A sequence of real numbers {ai }i=0,...,d is said to be unimodal if there exists an
index 0 j d such that a0 a1 . . . aj aj+1 . . . ad , and is said to be
log-concave if a2i ai1 ai+1 , for i = 1, . . . , d 1. It is easy to verify that a positive
log-concave sequence is unimodal, though a log-concave sequence need not be (take,
3

e.g., (1, 1, 0, 0, 1, 1)). We say that a polynomial

Pd

i=0

ai xi is log-concave (respec-

tively, unimodal) if the sequence {a0 , a1 , . . . , ad } has the corresponding property. It


is well known that if

Pd

i=0

ai xi is a polynomial with nonnegative coefficients and

with only real zeros, then the sequence {a0 , a1 , . . . , ad } is log-concave and unimodal
(see, e.g., [15], Thm. B, p.270). Log-concave and unimodal sequence arise often
in combinatorics, algebra, geometry and computer science, as well as in probability
and statistics where these concepts were first defined and studied. Even though
log-concavity and unimodality have one-line definitions, to prove the unimodality
or log-concavity of a sequence can be a very difficult task requiring the use of intricate combinatorial constructions or of refined mathematical tools. The number
and variety of these tools has been constantly increasing and is quite bewildering
and surprising. They include, for example, classical analysis, linear algebra, the
representation theory of Lie algebras and superalgebras, the theory of symmetric
functions, and algebraic geometry. We refer the interested reader to [42] (see also
[12]) for a survey of many of these techniques, problems, and results.
The theory of total positivity was first used to attack unimodality problems in
1989 ([5]). In retrospect, this is not too surprising since P F sequences (i.e., Toeplitz
T P matrices) are necessarily log-concave and unimodal. Nonetheless, the theory
of total positivity was at that time not very well known outside of the fields of
probability, statistics, and matrix theory, and thus very few combinatorialists were
aware of it. One classical result on P F sequences that proved to be especially useful
is the characterization theorem (see, e.g., [24], Theorem 5.3, p. 412).
Theorem 3.1 Let {ai }iN be a sequence of real numbers with a0 = 1. Then {ai }iN
is a P F sequence if and only if there exists a 0 and sequences {i }iN , {i }iN
R+ such that

i0

i +

i0

i < + and

az

ai z = e

i0

(1 + i z)

i0

(1 i z)

(1)

i0

for all z in some open disc around the origin in the complex plane.
This fundamental result, on which much of the theory of P F sequences is based (see,
e.g., [24]), was conjectured by Schoenberg ([39], p.367) and first proved by Edrei
([16], see also [2]). It is interesting to observe that Theorem 3.1 has turned out to
be of fundamental importance also in the theory of representations of the infinite
4

symmetric group where it was independently discovered and proved by Thoma ([49],
Satz 2).
For the purposes of combinatorics one needs a formal version of Theorem 3.1,
i.e., one which holds in the ring of formal power series R[[z]] without any question
of convergence (note that the RHS of (1) is not a well defined element of R[[z]] since
neither

i0 (1

+ i z) nor

i0 (1

i z) converge formally in R[[z]] (see, e.g., [41],

1.1, Proposition 1.1.9, p. 6)).


The following consequence of Theorem 3.1 is a restatement of Theorems 4.6.2
and 2.2.4 of [5], and is the result that is actually used in combinatorics.
Theorem 3.2 Let {ai }iN be a sequence of real numbers such that

ai xi =

i0

d
X

bi xi

i=0

(1 x)d+1

in R[[x]], for some d P and b0 , . . . , bd R such that


following are equivalent:

(2)

,
Pd

i=0 bi

6= 0. Then the

i) {a0 , a1 , a2 , . . .} is a P F sequence;
ii) {b0 , . . . , bd } is a P F sequence;
iii) the polynomial

Pd

i
i=0 bi x

has nonnegative coefficients and only real zeros.

There are two main reasons why this particular consequence of Theorem 3.1 is
so useful and interesting for combinatorics. The first one is that formal power
series of the form (2) are extremely common in enumerative combinatorics. In
fact, it is a well known and classical result in the theory of rational generating
functions that a formal power series

i0

ai xi is of this form if and only if there

exists A(x) R[x] such that A(i) = ai for all i N (see, e.g., [41], 4.3, Corollary
4.3.1). Furthermore, it is often the case that if the ai s have some combinatorial
interpretation then the bi s also do, and conversely, so the equivalence of i) and ii)
is often useful. The second reason lies in the fact that there are many polynomials
arising in combinatorics that are known (or conjectured) to have only real zeros
(see, e.g., [42], 3, p. 504, and [12], 3). This fact has always puzzled and intrigued
combinatorialists as they saw no philosophical reason why this peculiar analytic
phenomenon should occurr so often in combinatorics. However, the equivalence of
ii) and iii) in Theorem 3.2 (which, incidentally, was already known before Theorem
5

3.2, see [17]) shows that the reality of the zeros of a polynomial having nonnegative
coefficients is linked to the nonnegativity of certain determinants formed with its
coefficients, and this strongly suggests that, in such cases, these determinants should
have a combinatorial significance. Thus, combinatorialists were led to the conclusion
that, given a combinatorial sequence, it should often be the case that the minors
of its associated Toeplitz matrix also count something. This was already known
in some cases (see, e.g., [30], I.5, eq. (5.12)), and has stimulated research in this
direction, which has produced several beautiful combinatorial results (see, e.g., [19],
[9], [36], [43], Corollary 2.9 and the comments following it), conjectures, and open
problems.
We now illustrate a particular combinatorial application of Theorem 3.2 and then
give references to many others. Let (P, ) be a finite partially ordered set (or, poset,
for short). A map : P [i] (i P) is said to be order preserving if x  y implies
(x) (y) for all x, y P , and is a linear extension if it is order preserving and a
def

bijection (so i = |P |). Given two linear extensions , : P [p] (where p = |P |)


we let
def

d(, ) = |{i [p 1] : ( 1 (i)) > ( 1 (i + 1))}|.

(3)

One can then show (see, e.g., [40], I.2, p. 6, Definition 3.2, p. 8, and Proposition
8.3, p.24) that the polynomial
def

W (P ; x) =

xd(,) ,

(4)

where the sum is over all linear extensions of P , does not depend on , but
only on P (hence the notation W (P ; x) for it). Linear extensions of posets are an
important topic both in combinatorics as well as in theoretical computer science,
where they are usually called topological sortings (see, e.g., [25], [26], [56]), and the
polynomial W (P ; x) has been widely studied (see, e.g., [40], [5], [54]). In particular,
the following conjecture was made about it in [31].
Conjecture 3.3 Let P be a (finite) poset. Then W (P ; x) has only real zeros. In
particular, W (P ; x) is log-concave and unimodal.
The above conjecture (which was later generalized by Stanley, see [5], 1.2) is open
even for the unimodality statement, and has been verified for |P | 8 (there are
19,449 such posets) ([48], p.5). However, Conjecture 3.3 can be proved in several
important cases using Theorem 3.2, thanks to the following result.
6

Theorem 3.4 Let P be a poset. Then


X

(P ; i)xi =

i0

W (P ; x)
(1 x)|P |+1

(5)

in Z[[x]], where
(P ; i) = |{f : P [i] : f is order preserving}|,

(6)

for i N.
Note that (5) and the remarks following Theorem 3.2 imply that, for a given poset
P , (P ; i) is a polynomial function of i. The above result was first proved by Stanley
in [40], Proposition 8.3, p. 24. We therefore obtain immediately, from Theorems 3.2
and 3.4, the following equivalent conditions for the validity of Conjecture 3.3, which
first appeared in [5], 5.7, Theorem 5.7.2.
Theorem 3.5 Let P be a poset. Then the following conditions are equivalent:
i) W (P ; x) has only real zeros;
def

ii) {w0 (P ), . . . , w|P |(P )} is a P F sequence, where wi (P ) = [xi ](W (P ; x)) for i =
0, . . . , |P |;
iii) {(P ; i)}iN is a P F sequence.
Note that since the hard part of Theorem 3.2 is the implication i) iii) it is
reasonable to expect that proving iii) should be easier than proving i), in Theorem
3.5. This is also confirmed by the fact that the combinatorial meaning of (P ; i) (see
(6)) is simpler than that of wi (P ) (see (3), (4), and part ii) of Theorem 3.5). In fact,
for many classes of posets one knows formulas for (P ; i), but not for wi (P ) (see,
e.g., [5], 5.2, 5.4, 5.5, and 5.6). However, even these known formulas do not make it
clear whether the sequence {(P ; i)}iN is P F or not. This led combinatorialists to
study sufficient conditions (usually suggested by the known formulas for (P ; i)) for
a sequence to be P F , and eventually led to the following result. Given a polynomial
A(x) R[x], having only real zeros, we denote by (A) (respectively (A)) its
smallest (respectively, largest) zero.
Theorem 3.6 Let A(x) R[x] be a polynomial having only real zeros. Suppose
that A(x) = 0 for all x {((A), 1] [0, (A))} Z. Then the sequence {A(i)}iN
is P F .
7

The above result (which actually holds under slightly milder hypotheses, see [5],
Theorem 4.3.4) was first proved in [5], see Theorem 4.4.1. It is worth noting that,
even though it was motivated by purely combinatorial reasons, Theorem 3.6 is a
result on total positivity, with no combinatorics in it. This theme of combinatorics
stimulating results in total positivity, and conversely, will come up again and again
in this survey, and is in fact its main message.
Using Theorem 3.6 it is indeed possible to prove Conjecture 3.3 in many interesting cases, for example for Gaussian posets, disjoint unions of chains, and certain
Ferrers posets (we refer the reader to [5], Chapter 5, for the definitions of these
classes of posets and further details). In each case, the proof follows immediately
from known formulas for (P ; i) using Theorems 3.5 and 3.6.
If P1 and P2 are posets then one can form their disjoint union P1 P2 (see, e.g.,
[41], 3.2, for a discussion of this and other operations on posets). It is then well
known, and easy to see from (6), that
(P1 P2 ; i) = (P1 ; i) (P2 ; i)
for all i N. Thus Theorem 3.5 and Conjecture 3.3 naturally lead to the question
of whether the Hadamard (i.e., componentwise) product of two P F sequences is
again P F . It was already known to Karlin (see also [5], 6.5, p. 77, for an explicit
example) that this is in general false, though it is true for eventually vanishing
sequences (see, e.g., [5], Theorem 4.7.8). However, as noted after Theorem 3.4,
the sequence {(P ; i)}iN has the special property that (P ; i) is a polynomial
function of i. Combinatorialists tried to find out if this particular question had been
considered before (see, e.g., [4], and [5], 4.7, p. 55), then studied it themselves (see,
[5], 4.7), and were finally led to the following result (see, [52], and [53], Theorem
0.3).
Theorem 3.7 Let A(x), B(x) R[x] be such that {A(i)}iN and {B(i)}iN are
P F sequences. Then {A(i)B(i)}iN is a P F sequence.
Once again, this is a total positivity result that was motivated purely from combinatorics. Just as for Theorem 3.6, Theorem 3.7 has also many consequences in
combinatorics, and allows the proof of Conjecture 3.3 in other interesting cases such
as for trees and series-parallel posets. All these applications are described in detail
in [54]. Though Theorems 3.6 and 3.7 were motivated by Conjecture 3.3, they have
8

found many other applications in combinatorics. For a description of some of these


we refer the reader to [5], Chapter 6, and [53].
But there are more developments in total positivity that have arisen from Theorems 3.6 and 3.7. A closer look at (5) reveals what is really going on. Let A(x) R[x]
and write
A(x) =

d
X

ai

i=0

def

x+di
d

where d = deg(A). Then an easy application of the binomial theorem (see, e.g.,
[41], p. 16) implies that
X

A(i)xi =

i0

d
X

ai xi

i=0

(1 x)d+1

in R[[x]]. Therefore, by Theorem 3.2, the statement {A(i)}iN is a P F sequence is


equivalent to {a0 , . . . , ad } is a P F sequence, and the relationship between (P ; i)
and W (P ; x) is really that of expanding (P ; x) (considered as a polynomial in x)
in terms of the basis

x+di 
d

def

(where d = deg((P ; x))) and then changing

to xi , for i = 0, . . . , d.

x+di 
d

These considerations lead naturally to the following definitions and general problems. Let {vi (x)}i=0,...,d be an ordered basis of Vd . Define, for r P {},
def

P Fr [vi ] =

( d
X

ai vi (x) R[x] : {a0 , . . . , ad } is a P Fr sequence

i=0

def

and let P F [vi ] = P F [vi ], for brevity. Then Theorem 3.6 can be interpreted as
giving a sufficient condition for a polynomial in P F [xi ] to be also in P F

hx+di i
d

Thus, one may ask if similar results exist for other pairs of bases of Vd . Indeed, often
one does not even need any extra conditions, as the following result shows (see [5],
Chapter 2, Theorems 2.6.1, 2.6.2, and 2.6.3).
Theorem 3.8 Let d P. Then the following strict inclusions hold:
PF

hx i

P F [hxii ] P F [xi ] P F [(x)i ],

PF



x+di

P F2
P F2



hx i
i



x+di
d

P F [(x)i ] P F2

PF

P F [(x)i ],

P F2 [hxii ] P F2



P F2

hx i
i

hx i

hx i
i

hx i
i

P F2 [(x)i ],

, P F [hxii ] P F2
9

hx i
i

Sometimes, simple conditions suffice, (see [5], Theorem 3.2.1).


Theorem 3.9 Let A(x) R[x] be a polynomial of degree d. Then the following are
equivalent:
i) A(x) P F

hx i
i

P F1

hx+di i

ii) {A(i)}iN is a P F sequence.


The preceding result can equivalently be stated as saying that P F
PF

hx+di i
d

hx i
i

P F1

hx+di i
d

for all d P. Theorems 3.8 and 3.9, though not motivated directly by

combinatorics, have also found numerous applications in enumerative combinatorics


(see [5], Chapter 6, [6], [7]).
Note that Theorem 3.7 can be restated, by Theorem 3.2 and the remarks following Theorem 3.7, as the assertion that P F
def

hx+di i
d

PF

hx+ei i
e

PF

hx+d+ei i
d+e

(where, for A, B Vd , AB = {ab : a A, b B}). Since it follows from Theorem

3.2 that P F [xi ]P F [xi ] P F [xi ], it is natural to wonder whether similar product
theorems hold for some of the other bases considered in Theorem 3.8. Indeed,
the following holds, (see Proposition 4.2 and 4.4 in [10], and Theorem 4.5 and the
formula at the top of page 153 in [51]).
Theorem 3.10 P F [(x)i ]P F [(x)i ] P F [(x)i ].
On the other hand P F [hxii ]P F [hxii ] 6 P F1 [hxii ] since, for example, hxi(1 + 2hxi) =
2hxi2 hxi, and therefore P F

hx i
i

PF

hx i
i

6 P F1

hx i
i

also.

Note that Theorems 3.2, 3.7, 3.8, 3.9, and 3.10 all have as underlying theme the
study of linear transformations that preserve the P F or P F2 property of a sequence.
This topic had been studied before in total positivity (see, e.g., [24], Theorem 7.3,
p. 142) but not in such detail. Nonetheless, several open problems and conjectures
remain in the area, and we survey the main of these in section 6.

T P matrices in combinatorics

The applications of total positivity to combinatorics described in the previous section , and the total positivity results that were obtained with (direct or indirect)
motivation from combinatorics, all deal with P F sequences (i.e., Toeplitz T P matrices). After these results were obtained it was slowly realized that T P matrices
10

themselves are in fact quite ubiquitous in combinatorics. In this section we survey


the main results in this direction.
In retrospect, the first result of this kind was obtained by B. Lindstrom ([28])
though he did not state it at all in the language of total positivity, and in fact just
used it as a lemma to prove a result in matroid theory. To describe Lindstroms
result we now require a little bit of notation and terminology. Let D = (V, A) be a
directed graph (or, digraph, for short). We will always assume that D has no loops or
multiple edges, so that we can identify the elements of A with ordered pairs (u, v),
with u, v V , u 6= v. A path in D is a sequence = u1 u2 . . . un of elements of V
such that (ui , ui+1) A for i = 1, . . . , n 1, we then say that goes from u1 to un .
We say that D is locally finite if, for every u, v V , there are only a finite number
of paths from u to v. Note that this implies that D is acyclic. We say that D is
weighted if there is a function w : A R, where R is some commutative Q-algebra.
If R = R and w((u, v)) 0 for all (u, v) A then we call D a nonnegative digraph.
Let D = (V, A, w) be a locally finite, weighted, digraph. For a path = u0u1 . . . uk
in D we let
def

w() =

k
Y

w(ui1 , ui ) ,

i=1

and, for u, v V , we let


def

PD (u, v) =

w() ,

where the sum is over all paths in D going from u to v. We adopt the convention
def

that PD (u, u) = 1 for all u V (i.e. there is only one path, the empty path, from
u to u and its weight is 1). We will usually omit the subscript D when there is no
def

def

danger of confusion. Given u = (u1 , . . . , ur ), v = (v1 , . . . , vr ) V r we let


def

N(u, v) =

w(1 , . . . , r ) ,

(1 ,...,r )
def

where w(1 , . . . , r ) =

Qr

i=1

w(i ), and where the sum is over all r-tuples of paths

(1 , . . . , r ) from u to v (i.e., i is a path from ui to vi , for i = 1, . . . , r) that are


non-intersecting (i.e., i and j have no vertices in common if i 6= j). We say that u
and v are compatible if, for every Sr \ {Id}, there are no r-tuples of paths from
(u1 , . . . , ur ) to (v(1) , . . . , v(r) ) that are non-intersecting. The following fundamental
result was first proved by Lindstrom in [28].

11

Lemma 4.1 (Lindstr


oms Lemma) Let D = (V, A, w) be a locally finite, weighted
def

def

digraph and u = (u1 , . . . , un ), v = (v1 , . . . , vn ) V n be compatible. Then


N(u, v) = det [(PD (ui , vj ))1i, jn ] .

(7)

We refer the reader to [19], Corollary 2, [44], Theorem 1.2, or [28], Lemma 1 for
the proof of Lemma 4.1. We should mention that a reasoning very similar to the
one used by Lindstrom to prove his lemma had been used previously by Karlin and
McGregor in [23], though there is no doubt that Lindstrom was unaware of their
work.
The importance of Lindstroms Lemma in the discovery and proof that many
matrices arising in combinatorics are T P can hardly be overestimated, as all of these
proofs use this result, either implicitly or explicitly. The first notable application
of Lindstroms Lemma in this direction was given by Gessel and Viennot in [18]
where they construct a directed graph that enables them to give a combinatorial
interpretation of any minor of the infinite matrix of binomial coefficients, and many
other applications in enumerative and algebraic combinatorics followed (see, e.g.,
[9], [19], [36], [44]).
Aside from its direct applications to combinatorial matrices, Lindstroms Lemma
also turned out to be extremely important from a theoretical point of view. It is
easy to see (see [13], Figure 1, for a specific example) that the matrix on the RHS of
(7) is in general not T P under the hypoteses of Lemma 4.1. However, if we define
two r-tuples of vertices (u1 , . . . , ur ) and (v1 , . . . , vr ) of D to be fully compatible if
(ui1 , . . . , uik ) is compatible to (vi1 , . . . , vik ) for all 1 i1 < i2 < . . . < ik r then
it is clear that we can apply Lindstroms Lemma to any minor of (PD (ui , vj ))1i,jn
and hence conclude that it is totally positive if D is nonnegative. What is extremely
surprising, however, is that the converse statement also holds. Namely, we have the
following result.
Theorem 4.2 Let U be an n n (real) matrix. Then U is totally positive if and
only if there exists a planar, finite, nonnegative digraph D = (V, A, w), and u=
(u1 , . . . , un ), v = (v1 , . . . , vn ) V n fully compatible, such that
U = (PD (ui , vj ))1i, jn .

(8)

The preceding result was first explicitly stated and proved in [13], Theorem 3.1.
Note that Theorem 4.2 makes some well known properties of T P matrices obvious.
12

For example, it immediately implies that the product of two T P matrices is again
T P . A less obvious consequence is the following one, which had been conjectured
by Stembridge ([47]) and first proved by him in [45], though not using Theorem
4.2. Recall that given a partition of n and an n n matrix A = (ai,j )1i,jn the
immanant of A with respect to is
def

Imm (A) =

()a1,(1) . . . an,(n)

Sn

where : Sn Q is the irreducible character of Sn corresponding to the partition


(we refer the reader to, e.g., [38], for further information on the irreducible characters
of Sn ). Note that if l() = n then () = sgn() for all Sn and hence
Imm (A) = det(A), while if l() = 1 then () = 1 for all Sn and hence
Imm (A) = per(A).
Theorem 4.3 Let A be an n n (real) totally positive matrix. Then
Imm (A) 0
for all partitions of n.
For a proof of Theorem 4.3 based on Theorem 4.2 see [13], Corollary 3.8.
Despite the fact that Theorem 4.2 describes all the totally positive matrices, it
can sometimes be a difficult task, given a T P matrix U, to construct a digraph D
and u, v fully compatible such that (8) holds. Thus, the usefulness of Theorem 4.2
for actually proving that certain matrices are T P can be limited. Nonetheless, there
are other general results that can be proved using Lindstroms Lemma that are easier
def

def

def

to apply than Theorem 4.2. Let x = {xn }nN , y = {yn }nN , and z = {zn }nN
be three sequences of independent variables. We will adopt the convention that
def

def

def

xn = yn = zn = 0 if n < 0.
Theorem 4.4 Let t N. Define a matrix M = (Mn,k )n,kN by
Mn,k = zn Mnt,k1 + yn Mn1t,k1 + xn Mn1,k
def

(9)
def

if n + k P (where Mn,k = 0 if either n < 0 or k < 0), and M0,0 = 1. Then:


i) M is (x, y, z)-totally positive;
ii) every row of M is an (x, y, z)-PF sequence.
13

In what follows we denote by Mt (x, y, z) the matrix defined in the preceding theorem. Theorem 4.4 (which is proved in [13], Theorem 4.3) immediately implies the
T P property for an impressive array of matrices arising in enumerative and algebraic
combinatorics, as we now show.
For n, k N let S[n, k]p,q and c[n, k]p,q be the (p, q)-Stirling numbers of the
second and first kind, respectively. These are defined inductively by letting
Sp,q [n, k] = pk1 Sp,q [n 1, k 1] + [k]p,q Sp,q [n 1, k] ,
and
cp,q [n, k] = pn cp,q [n 1, k 1] + q[n 1]q cp,q [n 1, k] ,
def

def

if n + k P (with the convention that Sp,q [n, k] = cp,q [n, k] = 0 if either n < 0
def

def

or k < 0), and Sp,q [0, 0] = cp,q [0, 0] = 1. (We refer the reader to [50] and [13] for
combinatorial interpretations and further information about these polynomials). It
then follows immediately from these definitions and from (9) that
(Sp,q [n + 1, k + 1])n,kN = M0 (0, {pn }nN , {[n + 1]p,q }nN ) ,

(10)

(cp,q [n + 1, k + 1])n,kN = M0 ({q[n]q }nN , {pn+1}nN , 0) .

(11)

and

For n, k N let
hn i

k q

Then it is easy to see that




n+k
k

 !
q

def

k
Y

(1 q ni+1 )
.
(1 q i )
i=1

(12)

= M0 (1, 0, {q n }nN ) .
n,kN

k]p,q , and S[n,


k]p,q be the (p, q)-Stirling numbers
For n, k N let S[n, k]p,q , S[n,
of the second kind defined in 6 of [37] (to which we refer the reader for their
definition). (Note that what we denote by S[n, k]p,q is denoted by S[n, k]p,q in 6 of
[37]). Then it follows from the results in [37] and from (9) that
(S[n + 1, k + 1]p,q )n,kN = M0 (0, 1, {1 + pq[k]p,q }kN ) ,
+ 1, k + 1]p,q )n,kN = M0 (0, {q k }kN, {[k + 1]p,q }kN ) ,
(S[n
and
+ 1, k + 1]p,q )n,kN = M0 (0, {q k }kN , {q k + p[k]qp }kN ) .
(S[n
14

For n, k N let SB (n, k; q) be the q-Stirling number of the second kind of type
Bn defined in 3 of [11]. These may also be defined by
SB (n, k; q) =

n  
X
n
i=k

q ni (1 + q)i S(i, k) ,

(13)

and have the property that SB (n, k; 0) = S(n, k), while k! SB (n, k; 1) equals the
number of k-dimensional faces of the Coxeter complex of type Bn (see, e.g., [11], 3,
for further details). Then it follows from eq. (41) on p. 430 of [11] that
(SB (n, k; q))n,kN = M0 (0, (1 + q)1, {k(1 + q) + q}kN ) .
For n, k, r P let Dr (n, k) be the set of all Sn that have exactly k cycles,
each of size r. Note that |D1 (n, k)| is just the signless Stirling number of the
first kind c(n, k), while |D2 (n, k)| is the number of derangements of Sn having k
cycles, (these numbers are sometimes called the Jordan numbers, see, e.g., [14]).
The numbers |Dr (n, k)| are usually called the signless r-associated Stirling numbers
of the first kind (see, e.g., [15], p. 257, Ex. 7, for further information about these
numbers). For n, k N let cr [n, k]q be the r-associated signless q-Stirling numbers
of the first kind defined in 5 of [13] (to which we refer the reader for their definition).
It can be shown that c1 [n, k]q = c[n, k], where c[n, k] denotes a q-Stirling number of
the first kind as defined, e.g., in [27], and that cr [n, k]1 = |Dr (n, k)|. Then it follows
from Theorem 5.7 of [13] that

(cr [n, k]q )n,kN = Mr {[n 1]q }nN , 0, q

r1
Y

r1

[n i]q

i=1

nN

For n, k N let D(n, k) be the Delannoy number. This can be defined by


def

D(n, k) =


k  
X
k
n+ki
i=0

(14)

(see, e.g., [15], p.81, for the combinatorial significance of these numbers). Then it is
easy to see that
(D(n, k))n,kN = M0 (1, 1, 1) .

(15)

For k, n N let a(n, k) be the number of regions into which n hyperplanes in


general position divide Rk , (see, e.g., [57] for further information on hyperplane
arrangements). Then it follows from [15], p. 72, Ex. 2, and from (9) that
(a(n, k))n,kN = M0 (1, 1, {1, 0, 0, . . .}) .
15

For n, k N let L[n, k]q be the signless q-Lah numbers defined in 5 of [13].
For q P the L[n, k]q were first defined and studied in [1] where they arose from a
problem in statistics and where they are called the associated Lah numbers. It can
be shown that L[n, k]q reduces to the absolute value of the ordinary Lah number
L(n, k) (as defined, e.g., in [15], p.135) when q = 1. Then it follows from results in
[1] (see also [13], Theorems 4.8, and 5.17) and from (9) that
(L[n, k]q )n,kN = M0 ({0, 0, 1, 2, . . .}, 1, 0) M0(0, q1, {nq}nN ) .
It should be noted that the proof of Theorem 4.4 (see [13], Theorems 4.3 and
4.1) actually gives a combinatorial interpretation of the minors involved in i) and ii),
which can then be specialized to obtain more explicit combinatorial interpretations
in the special cases considered above. This has been carried out in detail in [9] for
the matrices in (11) and (10) when p = 1, but has not been done yet in other cases.
Theorem 4.4 is not an isolated result and similar ones appear in 4 of [13].
Despite the large applicability of Theorem 4.4 there are still many matrices arising in enumerative combinatorics which are only conjectured to be T P , we survey
the main open problems in this area in section 6.

Recent Developments

In recent years there has been a growing trend in enumerative and algebraic combinatorics to extend classical results known for permutations (i.e., for the symmetric
group) to other Coxeter groups (see, e.g., [11], [33], [34], [35]). The concept of
total positivity has been no exception, and in this section we describe these new
developments.
We require first some notation, terminology, and definitions. Let, for brevity,
def

E = Rn , and denote by (, ) its usual inner product. Given E \ {0} we denote


def

by H the hyperplane orthogonal to (i.e., H = {v E : (v, ) = 0}) and by


the orthogonal reflection through H (i.e., the only linear transformation of E into
itself that fixes H pointwise, and sends to ). In other words,
(x) = x

2(x, )

(, )

for all x E. A (finite) subset E is said to be a root system if the following


conditions are satisfied:
16

i) spans E, and 0 6 ;
ii) if , and = c for some c R then c {1, 1};
iii) () = for all ;
iv) if , then

2(,)
(,)

Z.

The elements of are called the roots of . A root system is called irreducible if
there are no 1 , 2 , 1 , 2 6= , such that = 1 2 , 1 2 = and
(, ) = 0 for all 1 and 2 .
A subgroup W of GL(E) is called a finite reflection group (or a Coxeter group)
if there exists a root system E such that W equals the subgroup of GL(E)
generated by { : }. We say that W is irreducible if is irreducible. Note
that since () = and is invertible, permutes and is determined by
| since spans E. Hence we may identify W as a subgroup of the symmetric
group on which shows, in particular, that W is indeed finite. Irreducible finite
reflection groups have been completely classified (see, e.g., [22], Chapter 2, for an
excellent exposition of this classification). It turns out that there are five infinite
families, usually denoted by An , Bn , Cn , Dn , and I2 (n) (n P, n 2) and seven
single groups, usually denoted by E6 , E7 , E8 , F4 , G2 , H3 , and H4 .
def

Now let A = E \

H . Clearly, the connected components of A (which are

called Weyl chambers) are convex open cones. Fix, once and for all, a Weyl chamber
C, and call it the fundamental chamber. The following result is fundamental, and a
proof of it can be found, e.g., in [22], Theorem 1.12.
Theorem 5.1 W acts on A and this action is simply transitive, (i.e., given two
Weyl chambers C1 and C2 there exists a unique w W such that w(C1 ) = C2 ). In
particular, there is a unique bijection between the set of Weyl chambers and W such
that C corresponds to the identity in this bijection.
i

z}|{

Now let e1 , . . . , en be the canonical basis of E = R (i.e., ei = (0, . . . , 0, 1 , 0, . . . , 0),


for i = 1, . . . , n). Given x Rn we denote by xj its j-th coordinate with respect
to this basis (so that x = (x1 , . . . , xn )). We are now ready to define the crucial
concepts of this section which are due to K. Gross and D. St. P. Richards ([21]).
Given a function K : R2 R we let
def

DW K(s, t) =

det(w)

n
Y

j=1

wW

17

K(sj , w(t)j )

(16)

for each s, t Rn . We say that K is W -totally positive (or W T P , for brevity) if


DW K(s, t) 0

(17)

for all s, t C. We say that K is W -symmetric if


DW K(s, w(t)) = DW K(w(s), t)
for all s, t Rn and w W , and denote by H the class of all W -symmetric functions
K : R2 R.
The most basic result on totally positive functions is that the convolution of two
totally positive functions is again totally positive (see, e.g., [24]). This extends to
W -total positivity (see [21], Theorem 3.5).
Theorem 5.2 Let L, M H be W -totally positive and suppose that
def

K(x, y) =

L(x, z)M(z, y)dz

converges absolutely for any (x, y) R2 (where dz denotes Lebesgue measure). Then
K H and K is W -totally positive.
In order to better understand the concept of W -total positivity (which would
otherwise be rather abstract) it is useful to look at some important special cases.
We will analyze the concept of W -total positivity for the four infinite families An ,
Bn , Cn , and Dn .
For type An it can be shown (see [21], 6) that a function K : R2 R is An totally positive for n = 1, . . . , r if and only if it is T Pr in the classical sense and
that any function K : R2 R is W -symmetric for type An for all n 2. Thus we
will not spend any more words on this case. Also, it is not hard to show from the
definitions and known facts on finite reflection groups that the concepts of Bn -total
positivity and Cn -total positivity are equivalent for any n 2.
For type Bn it is known (see, e.g., [22], 2.10) that a finite reflection group W of
type Bn can be realized by taking
def

= {ei ej : 1 i < j n} {ei : i [n]}


def

and E = Rn , and that W then equals the set of all n n signed permutation
matrices (i.e., matrices of the form P diag(1 , . . . , n ) where P is a permutation
matrix and 1 , . . . , n {1, 1}). As a fundamental chamber we may take
C = {x Rn : 0 < x1 < x2 < . . . < xn }.
18

In this case we obtain from (16) that


X

DW K(s, t) =

{1,1}n Sn

{1,1}n

n
Y

j=1

n
Y

j=1

j sgn()

n
Y

K(sj , (j) t(j) )

j=1

j det[K(si , j tj )1i,jn ]

for any K : R2 R and s, t Rn . Hence K is W -totally positive in this case if


and only if
X

1 ,...,n {1,1}

n
Y

j=1

j det[K(si , j tj )1i,jn ] 0

for all 0 < s1 < . . . < sn and 0 < t1 < . . . < tn . For example, if W is of type B2
then K is W T P if and only if

det

K(s1 , t1 ) K(s1 , t2 )
K(s2 , t1 ) K(s2 , t2 )

det

+ det

K(s1 , t1 ) K(s1 , t2 )
K(s2 , t1 ) K(s2 , t2 )

K(s1 , t1 ) K(s1 , t2 )
K(s2 , t1 ) K(s2 , t2 )

+ det

K(s1 , t1 ) K(s1 , t2 )
K(s2 , t1 ) K(s2 , t2 )

for all 0 < s1 < s2 and 0 < t1 < t2 . No characterization is known of the functions
K : R2 R that are W -symmetric in this case. However, it is not hard to check
that if K(s, t) = K(s, t) for all s, t R then K is W -symmetric for type Bn , for
all n 2.
For type Dn it is known (see, e.g., [22], 2.10) that a finite reflection group W of
type Dn can be realized by taking
def

= {ei ej : 1 i < j n}
def

and E = Rn , and that W then equals the set of all n n signed permutation
matrices that have an even number of negative entries (i.e., matrices of the form
P diag(1 , . . . , n ) where P is a permutation matrix, 1 , . . . , n {1, 1}, and
Qn

j=1

j = 1). As a fundamental chamber we may take


C = {x Rn : |x1 | < x2 < . . . < xn }.

In this case we obtain from (16) that


DW K(s, t) =

{{1,1}n :

Qn

=1}
j=1 j

19

det[K(si , j tj )1i,jn ]

for any K : R2 R and s, t Rn . Hence, by (17), K is W -totally positive in this


case if and only if
X

det[K(si , j tj )1i,jn ] 0

(18)

{1 ,...,n {1,1}: 1 ...n =1}

for all |s1 | < s2 < . . . < sn and |t1 | < t2 < . . . < tn . It is again true that if
K(s, t) = K(s, t) for all s, t R then K is W -symmetric for W of type Dn for
all n 2.
We hope that these cases will have helped the reader get a feeling for these
new concepts of total positivity. The notions described in this section have been
developed very recently in [21] and their study is still in its infancy. We discuss
some conjectures and open problems on W -total positivity in the next section.
Before closing this section we should mention that yet another generalization of
the concept of total positivity has recently been developed by G. Lusztig in [29].
Lusztig defines the concept of total positivity for any element of a split reductive
connected (real) algebraic group, and obtains usual total positivity when the group
is GL(n; R).

Conjectures and Open Problems

Despite the spectacular applications between total positivity and combinatorics surveyed in this paper an impressive array of open problems and conjectures, both of
a total positivity as well as of a combinatorial nature, still remain. In this section
we describe some of the most outstanding ones and give references to others.
We begin with some total positivity problems that arise naturally from the results
described in section 3. Theorem 3.8 gives inclusions that hold between the classes
P F [vi ] and P F2 [vi ] as {vi }i=0,...,d ranges over certain bases of Vd . It is shown in the
Appendix on p. 103 of [5] that these are all the possible inclusions except possibly
for one, which is the following.
Problem 6.1 Is it true that P F2

hx i
i

P F2 [xi ] ?

The answer is yes if d 5.


In addition to the six bases in Theorem 3.8 there are other bases that one may
wish to consider, though their choice will depend on the applications that one has
in mind. Though not directly related to combinatorics, two bases worth looking
20

at are those consisting of the Eulerian and Krawtchouk polynomials. The Eulerian
def

polynomials Ai (x) may be defined inductively by letting A0 (x) = 1 and


def

An (x) = (1 + (n 1)x)An1 (x) + (x x2 )

d
(An1 (x))
dx

(19)

for n 1 (see, e.g., [15], for the combinatorial significance and further information
about the Eulerian polynomials).
Conjecture 6.2 P F [Ai (x)] P F [xi ].
Pd

Note that the conjecture states that if the polynomial


ros and nonnegative coefficients then the polynomial

i=0

Krawtchouk polynomials may be defined by


(d)

def

Ki (x) =

i
X

(2)j

j=0

dj
ij

i=0

Pd

 
x
j

ki xi has only real ze-

ki Ai (x) also does. The

(d)

Conjecture 6.3 P F [Ki (x)] P F [(1)i xi ].


These two conjectures first appeared in [5], 3.4. Another conjecture similar to the
last two, but involving the Lagrange interpolation polynomials appears in [5], 3.4,
p. 33.
The following problem is naturally suggested by Theorem 3.10 and the comments
preceding and following it.
Problem 6.4 Is it true that P F

hx i
i

PF

hx i
i

PF

hx i
i

Regarding combinatorial problems involving total positivity, Conjecture 3.3 is certainly one of the most outstanding, but problems of a similar nature abound (see,
e.g., [42], 3, [12], 4, [6], 6, [7], 7, [11], 5, [8], 5, [10], 6). We will just mention
two of them here to give to the reader an idea of their nature.
Let G = (V, E) be a graph (i.e., V is a finite set and E

V 
2

). A function

f : V P is called a coloring of G if (x) 6= (y) for all x, y V such that


(x, y) E. Given n P we denote by PG (n) the number of colorings : V P
such that (V ) [n]. It is then well known (see, e.g., [15], 4.1, p. 179, ex. (2)) that
there exists a polynomial (G; x) Z[x] of degree |V | such that (G; n) = PG (n)
for all n P. This polynomial is called the chromatic polynomial of G and has been
studied extensively (see, e.g., [32], and the references cited there).
Conjecture 6.5 Let G be a graph. Then {(G; n)}nN is a P F2 sequence.
21

def

Conjecture 6.6 Let G be a graph. Then (1)p (G; x) P F [(x)i ], where p =


|V |.

Conjecture 6.5 first appeared in [7], see Conjecture 7.5. Conjecture 6.6 first appeared
in [55], has been verified for all graphs with |V | 9 and is known to be true in several
cases (see [7], p. 751, and [10], 5).
We now examine some total positivity problems arising from the results presented
in section 4. There are (at least) two major problems that are naturally suggested
by Theorem 4.2.
Problem 6.7 Can Theorem 4.2 be generalized to infinite matrices U = (Ui,j )i,jN?
Problem 6.8 Can Theorem 4.2 be generalized to (finite) x T P matrices?
On a more specific level, the following conjecture (which would trivially imply Theorem 4.3) is particularly intriguing.
Conjecture 6.9 Let M be a T Pr -matrix (r P). Then Imm(M) 0 for all
partitions such that l() r.
The preceding conjecture is due to J. Stembridge ([46]), and is open even for r = 2.
We mentioned in section 4 that despite the power of Theorems 4.2 and 4.4 many
matrices arising from combinatorics are only conjectured to be T P . We give here
def

two such examples. Let A(n, k) = [xk ](An (x)) (where An (x) is the n-th Eulerian
polynomial, defined by (19)) for k, n N. The numbers A(n, k) are called Eulerian
numbers and have been widely studied in combinatorics (see, e.g., [15], 6.5).
def

Conjecture 6.10 The matrix A = (A(n, k))n,kN is T P .


Note that a simple recursion relation for the numbers A(n, k) can be obtained by
extracting the coefficient of xk on both sides of (19). Also note that it is known (see,
e.g., [15], ex. 3, p. 292) that every row of A is P F .
For r P and n, k N let Sr [n, k]q be the r-associated q-Stirling number of the
second kind defined in 5 of [13] (to which we refer the reader for its definition). It
is known (see [13], Theorem 5.10) that the Sr [n, k]q s satisfy the following recursion
relation
Sr [n, k]q = [k]qr Sr [n 1, k]q +
def

hn1 i

r1 q

Sr [n r, k 1]q ,

for all n, k P (where Sr [n, k]q = 0 if either n < 0 or k < 0, and Sr [n, 0]q =
Sr [0, n]q = 0,n for n N).
22

def

Conjecture 6.11 The matrix Sr (q) = (Sr [n, k]q )n,kN is q T P .


The developments described in section 5 are still very much in their infancy and
thus open problems abound. For example, one can ask about the W -total positivity
of any one of the totally positive functions discussed in [24], for other reflection
groups W . Of all these functions, the only one which has been investigated so far
is exy , for which the following is conjectured (see the comments following Example
6.4 in [21]).
Conjecture 6.12 The function K(x, y) = exy is W -totally positive for all finite
reflection groups W .
The above conjecture has been proved if W is the Weyl group of a compact connected
Lie group by K. I. Gross and D. St. P. Richards in [21] using deep results by HarishChandra (see [21], 4, and Theorem 6.1, for details). It does seem to be the case that
W -total positivity is a much deeper concept then usual total positivity. However,
we believe that the discovery of other W -totally positive functions will be mainly a
matter of time.

References
[1] J. C. Ahuja and E. A. Enneking, Concavity property and a recurrence relation
for associated Lah numbers, Fibonacci Quart., 17 (1979), 158-161.
[2] M. Aissen, I. J. Schoenberg, and A. M. Whitney, On the generating function of
totally positive sequences, I, J. dAnal. Math, 2 (1952), 93-103.
[3] T. Ando, Totally positive matrices, Linear Algebra Appl., 90 (1987), 165-219.
[4] F. Brenti, Query 389, Notices Amer. Math. Soc., 35 (1988), 315.
[5] F. Brenti, Unimodal, log-concave, and Polya Frequency sequences in Combinatorics, Memoirs Amer. Math. Soc., no. 413 (1989).
[6] F. Brenti, Log-concavity and combinatorial properties of Fibonacci Lattices, Europ. J. Combinatorics, 12 (1991), 459-476 .
[7] F. Brenti, Expansions of chromatic polynomials and log-concavity, Trans. Amer.
Math. Soc., 332 (1992), 729-756 .
23

[8] F. Brenti, Permutation enumeration, symmetric functions, and unimodality,


Pacific J. Math., 157 (1993), 1-28.
[9] F. Brenti, Determinants of super-Schur functions, lattice paths, and dotted
plane partitions, Adv. in Math., 98(1993), 27-64.
[10] F. Brenti, G. F. Royle, and D. G. Wagner, Location of zeros of chromatic and
related polynomials of graphs, Canadian J. Math., 46 (1994), 55-80.
[11] F. Brenti, q-Eulerian polynomials arising from Coxeter groups, Europ. J. Combinatorics, 15 (1994), 417-441.
[12] F. Brenti, Log-concave and unimodal sequences in Algebra, Combinatorics and
Geometry: an update, Contemporary Math., 178 (1994), to appear.
[13] F. Brenti, Combinatorics and total positivity, J. Combin. Theory Ser. A, 71
(1995), 175-218.
[14] L. Carlitz, Note on the numbers of Jordan and Ward, Duke Math. J. 38 (1971),
783-790.
[15] L. Comtet, Advanced Combinatorics, Reidel, Dordrecht/Boston, 1974.
[16] A. Edrei, On the generating function of totally positive sequences, II, J. dAnal.
Math, 2 (1952), 104-109.
[17] A. Edrei, Proof of a conjecture of Schoenberg on the generating function of a
totally positive sequence, Canadian J. Math., 5 (1953), 86-94.
[18] I. Gessel and G. Viennot, Binomial determinants, paths, and hook length formulae, Adv. in Math., 58 (1985), 300-321.
[19] I. Gessel and G. Viennot, Determinants, paths, and plane partitions, preprint.
[20] H. W. Gould, The q-Stirling numbers of the first and second kinds, Duke Math.
J., 28 (1961), 281-289.
[21] K. I. Gross and D. St. P. Richards, Total positivity, finite reflection groups, and
a formula of Harish-Chandra, J. Approximation Theory, to appear.
[22] J. E. Humphreys, Reflection Groups and Coxeter Groups, Cambridge Studies
in Advanced Mathematics, no.29, Cambridge Univ. Press, Cambridge, 1990.
24

[23] S. Karlin and G. McGregor, Coincidence probabilities, Pacific J. Math., 9


(1959), 1141-1164.
[24] S. Karlin, Total Positivity, vol.1, Stanford University Press, 1968.
[25] D. Knuth and J. Szwarcfiter, A structured program to generate all topological
sorting arrangements, Inform. Process. Lett., 2 (1974), 153-157.
[26] A. Kalvin and Y. Varol, On the generation of all topological sortings, J. Algorithms, 4 (1983), 150-162.
[27] P. Leroux, Reduced matrices and q-log concavity properties of q-Stirling numbers, J. Combin. Theory Ser. A, 54 (1990), 64-84.
[28] B. Lindstrom, On the vector representations of induced matroids, Bull. London
Math. Soc., 5 (1973), 85-90.
[29] G. Lusztig, Total positivity in reductive groups, in Lie Theory and Geometry
123 (J. L. Brylinski et al., editors), Birkhauser, Boston-Basel-Berlin, 1994,
531-568.
[30] I. G. Macdonald, Symmetric Functions and Hall Polynomials, Oxford University Press, New York/London, 1979.
[31] J. Neggers, Representations of finite partially ordered sets, J. Combin. Inform.
System Sci., 3 (1978), 113-133.
[32] R. C. Read and W. T. Tutte, Chromatic polynomials, in Selected Topics in
Graph Theory 3 (Beineke and Wilson, editors), Academic Press, New York,
1988.
[33] V. Reiner, Quotients of Coxeter complexes and P-partitions, Memoirs Amer.
Math. Soc., no. 460 (1992).
[34] V. Reiner, Signed permutations statistics, Europ. J. Combinatorics, 14 (1993),
553-67.
[35] V. Reiner, Signed permutation statistics and cycle type, Europ. J. Combinatorics, 14 (19B93), 569-79.
25

[36] B. E. Sagan, Log concave sequences of symmetric functions and analogs of the
Yacobi-Trudi determinants, Trans. Amer. Math. Soc., 329 (1992), 795-812.
[37] B. E. Sagan, A maj statistic for set partitions, Europ. J. Combinatorics, 12
(1991), 69-79.
[38] B. E. Sagan, The Symmetric Group , Wadsworth and Brooks/Cole, Pacific
Grove, CA, 1991.
[39] I. J. Schoenberg, Some analytical aspects of the problem of smoothing, Courant
Anniversary Volume, New York, 1948.
[40] R. P. Stanley, Ordered structures and partitions, Memoirs Amer. Math. Soc.,
no. 119 (1972).
[41] R. P. Stanley, Enumerative Combinatorics , vol.1, Wadsworth and Brooks/Cole,
Monterey, CA, 1986.
[42] R. P. Stanley, Log-concave and unimodal sequences in Algebra, Combinatorics
and Geometry, Annals of the New York Academy of Sciences, 576 (1989), 500534.
[43] R. P. Stanley, Graph colorings and related symmetric functions: ideas and applications, preprint.
[44] J. R. Stembridge, Non-intersecting paths, pfaffians and plane partitions, Adv.
in Math., 83 (1990), 96-131.
[45] J. R. Stembridge, Immanants of totally positive matrices are nonnegative, Bull.
London Math. Soc., 23 (1991), 422-428.
[46] J. R. Stembridge, personal communication, April 1991.
[47] J. R. Stembridge, Some conjectures for immanants, Canad. J. Math., 44 (1992),
1079-1099.
[48] J. R. Stembridge, A Maple package for posets, preprint.
[49] E. Thoma, Die unzerlegbaren, positiv-definiten Klassenfunktionen der abzahlbar
unendlichen symmetrischen Gruppe, Math. Zeit., 85 (1964), 40-61.
26

[50] M. Wachs and D. White, p,q-Stirling numbers and set partition statistics, J.
Combin. Theory Ser. A, 56 (1991), 27-46.
[51] D. G. Wagner, The partition polynomial of a finite set system, J. Combin.
Theory Ser. A, 56 (1991), 138-159.
[52] D. G. Wagner, Enumerative combinatorics of partially ordered sets, and total positivity of Hadamard products, Ph.D. Thesis, Massachusetts Institute of
Technology, 1989.
[53] D. G. Wagner, Total positivity of Hadamard products, J. Math. Analysis and
Appl., 163 (1992), 459-483.
[54] D. G. Wagner, Enumeration of functions from posets to chains, Europ. J. Combinatorics, 13 (1992), 313-324.
[55] D. G. Wagner, Zeros of rank-generating functions of Cohen-Macaulay complexes, Proceedings of Fourth Formal Power Series and Algebraic Combinatorics
Conference, Montreal, 1992, Discrete Math., to appear.
[56] D. West, Generating linear extensions by adjacent transpositions, J. Combin.
Theory Ser. B, 58 (1993), 58-64.
[57] T. Zaslavsky, Facing up to Arrangements: Face-Count Formulas for Partitions
of Space by Hyperplanes, Memoirs Amer. Math. Soc., no.154 (1975).

School of Mathematics
Institute for Advanced Study
Princeton, NJ, 08540
U.S.A.

27

You might also like