European Journal of Combinatorics 24 (2003) 499–509

www.elsevier.com/locate/ejc

Combinatorics of generalized Tchebycheff

polynomials
Dongsu Kima, Jiang Zengb
a Department of Mathematics, KAIST, Daejeon 305-701, Republic of Korea
b Institut Girard Desargues, Université Claude Bernard (Lyon I), France

Received 3 March 2003; accepted 10 March 2003

Abstract

By considering a family of orthogonal polynomials generalizing the Tchebycheff polynomials

of the second kind we refine the corresponding results of De Sainte-Catherine and Viennot on
Tchebycheff polynomials of the second kind (Lecture Notes in Mathematics, vol. 1171, 1985,
Springer-Verlag, 120). © 2003 Elsevier Science Ltd. All rights reserved.

1. Introduction

There are a number of integrals of products of classical orthogonal polynomials with

respect to their weight functions that have a regular sign pattern (see, e.g. the classical
papers by Askey [1, 2]). One way of proving the positivity of those integral expressions—
a way that has been very fruitful in the past (see, e.g. [2, 4–9, 11] and the references
cited there)—is to construct an adapted combinatorial structure and express the integral
as a generating function for that structure. Then proving the positivity of that sign pattern
boils down to deriving some specific geometric properties for those structures. Thus, a true
analytic result appears as a corollary of a combinatorial property.
The calculation of those integrals can also be viewed as an evaluation of a formal
linear functional applied to its associated orthogonal polynomials. Note that when the
weight function is unknown the positivity of the evaluation is somehow unexpected. In
this paper we want to illustrate this by making a study of “integrals” of products of
general Tchebycheff polynomials Un (x, a), which are defined by the following three-term
recurrence relation:
Un+1 (x, a) = xUn (x, a) − λn Un−1 (x, a), U0 (x, a) = 1, U1 (x, a) = x, (1)

E-mail addresses: dskim@math.kaist.ac.kr (D. Kim), zeng@euler.univ-lyon1.fr (J. Zeng).

0014-5793/03/$ - see front matter © 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0195-6698(03)00046-5
500 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509

where λ2k = a, λ2k+1 = 1. Hence Un (2x, 1) is the Tchebycheff polynomial of the

second kind Un (x) (see [3]). Vauchassade de Chaumont and Viennot [13] have shown
how these generalized Tchebycheff polynomials with one parameter a arose naturally from
some enumerative problems in molecular biology [13]. It is also known [13, 14] that the
polynomials Un (x, a) are orthogonal with respect to the linear functional ϕ : R(a)[x] →
R(a) defined by ϕ(x 2n+1 ) = 0 and
n−1    
1
n n
ϕ(x ) =
2n
ai . (2)
n i=0 i i +1

In addition, we can readily derive from (1) the following generating function:

1 + xw + aw2
Un (x, a)wn = . (3)
n≥0
(1 + w2 )(1 + aw2 ) − x 2 w2

The aim of this paper is to give a combinatorial interpretation for the following evaluation
of ϕ at a product of generalized Tchebycheff polynomials Un (x, a):
I((n 1 , . . . , n k ); a) = ϕ(Un1 (x, a) · · · Unk (x, a)). (4)
In particular, our interpretation will imply that I((n 1 , . . . , n k ); a) is a polynomial of a
with nonnegative integral coefficients and generalize a result of De Sainte-Catherine and
Viennot on Tchebycheff polynomials of the second kind [5].
We start with some definitions and notations. A lattice path of length n is a sequence
γ = ( p0, p1 , . . . , pn ) of points pi = (x i , yi ) in N × N, with steps si = ( pi−1 , pi ) of type
up (yi = yi−1 + 1, x i = x i−1 + 1) or down (yi = yi−1 − 1, x i = x i−1 + 1). The path is
called a Dyck path if p0 = (0, 0) and pn = (n, 0). We can identify the path with its step
sequence (s1 , s2 , . . . , sn ), or its step-type sequence (t (s1 ), . . . , t (sn )) with starting point
p0 and ending point pn , where t (si ) is u, if si is up; d, if it is down. We call x i (resp. yi )
the position (resp. height) of pi . The point pi is called a rise (resp. fall) if si is up (resp.
down); pi is called a peak (resp. valley) if si is up (resp. down) and si+1 is down (resp. up).
Let n = (n 1 , . . . , n k ) be a composition of the positive integer n. Let [n] denote the set
{1, 2, . . . , n} and for j = 1, . . . , k, let S j ⊂ [n] denote the segment
S j = {i : n 1 + · · · + n j −1 + 1 ≤ i ≤ n 1 + · · · + n j −1 + n j }.
So [n] = S1 ∪ S2 ∪ · · · ∪ Sk . The parity (resp. position-parity) of S j is defined to be that
of n j (resp. n 1 + · · · + n j −1 ), for 1 ≤ j ≤ k, assuming that n 0 = 0. The local parity of i
in S j is defined to be that of i − n 1 − · · · − n j −1 . Let DSV(n) be the set of Dyck paths of
length n such that all the peak-positions belong to {n 1 , n 1 + n 2 , . . . , n 1 + · · · + n k−1 }.
Given a path γ in DSV(n), a valley pi = (i, yi ) of γ is said to be special, if yi is even
and i belongs to an even segment of odd position-parity; let EHR(γ ) (resp. SVAL(γ )) be
the number of even-height rises (special valleys) of γ . The following is our main result.
Theorem 1. For any composition n = (n 1 , . . . , n k ) of the positive integer n, we have


I(n; a) = a EHR(γ )+SVAL(γ ) . (5)
γ ∈DSV(n)
 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509 501

Fig. 1. The Dyck paths of DSV(2, 3, 2, 3).

Some remarks are in order. As we will show later, the weight function in the statement
of the above theorem is different, but inspired, from that of the corresponding moments.
If a = 1, it can be shown [5] that (4) can be written as
 
2 1
I(n; 1) = Un1 (x) · · · Unk (x) 1 − x 2 dx,
π −1
so Theorem 1 reduces to a result of De Sainte-Catherine and Viennot [5].
The set DSV(n) itself may change if n is rearranged. Moreover, in n, if all the even
integers are to the left of odd ones, as in (2, 2, 3, 3), then there is no even segment of odd
position-parity and so no path in DSV(n) will have any special valleys.
Example. The set DSV(2, 3, 2, 3) is shown in Fig. 1, where even up-steps and special
valleys are each weighted by a. Note that S1 = {1, 2}, S2 = {3, 4, 5}, S3 = {6, 7}, S4 =
{8, 9, 10}; and only S3 is an even segment of odd position-parity. Theorem 1 reads
I(n; a) = 2a 2 + a 3 for any rearrangement n of (2, 3, 2, 3).
If k = 2 and n = (m, n), then to form a Dyck path in DSV(m, n) we must have m = n
and the path should start with n up-steps and end with n down-steps, which imply the
orthogonality.
Corollary 1. For any positive integers m, n we have
ϕ(Um (x, a)Un (x, a)) = a n/2 δmn . (6)
For any proposition A we define the characteristic function χ(A) to be 1, if A is true; 0, if
false. The following is the linearization formula for Un (x, a).
Corollary 2. For any positive integers m and n,


min(m,n)
Um (x, a)Un (x, a) = a k/2+χ(mn≡0 mod 2)χ(k≡1 mod 2) Um+n−2k (x, a). (7)
k=0

Proof. Since Un (x, a) is a polynomial in x of degree n, we have the expansion

m+n
Um (x, a)Un (x, a) = c(m, n, l)Ul (x, a),
l=0

where c(m, n, l) ∈ R(a). From the orthogonality (6) we derive

c(m, n, l) = a −l/2 ϕ(Um (x, a)Un (x, a)Ul (x, a)).
502 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509

By Theorem 1, ϕ(Um (x, a)Un (x, a)Ul (x, a)) = 0 if m + n + l is odd or l > m + n.
Suppose m + n + l is even and set l = m + n − 2k with k ≥ 0, then
ϕ(Um (x, a)Un (x, a)Ul (x, a)) is the generating function of Dyck paths of length 2(m +
n − k) whose peak-positions are contained in {m, m + n}. Therefore,
 (l+k)/2−l/2
a , if m or n is odd, or k is even,
c(m, n, l) =
a (l+k+1)/2−l/2 , if m and n are even, and k is odd,
 k/2
a , if k is even,
= a k/2 , if m, n, k are odd,
 (k+1)/2
a , otherwise.
Combining the above cases, we obtain (7).

There are several ingredients in our derivation of the combinatorial interpretation for
I (n; a). First, as in [5], we obtain a preliminary combinatorial interpretation of I(n; a)
with signed weight function by interpreting Un (x, a) as weighted matching polynomials of
the segment graph on [n] and the moments ϕ(x n ) as generating functions of noncrossing
perfect matchings of the complete graph on [n]. This will be done in the next section.
The difficult part is to construct a weight-preserving sign-reversing involution or wpsr
involution on the underlying set. In Section 3 we describe an involution ψ on the set of
weighted paths with fixed starting and ending points, then in Section 4, we apply this
involution ψ to prove our main theorem.

2. Matching polynomials and Dyck path complexes

A graph on a finite set S is an ordered pair G = (S, E) where E is a set of some pairs
{s, t} of elements of S. An element of S is called a vertex of G and that of E an edge of
G. A matching µ of the graph G is a subset of edges such that no two edges of µ have a
common vertex. Given a matching µ of G, a vertex of G is said to be isolated if it does not
belong to any edge of µ. A perfect matching is a matching without isolated vertices. The
complete graph on S is the graph K S = (S, E), where E is the set of all possible edges
{s, t} with s = t. The segment graph on [n] is the graph Segn = ([n], E), where E is the
set of all edges {i, i + 1} for 1 ≤ i ≤ n − 1. We put weight on the matchings of Segn : each
isolated point is weighted x and each edge {i, i + 1} is weighted −1, if i is odd; −a, if i is
even. According to Viennot’s theory [14], it is easy to derive from (1) that the polynomial
Un (x, a) is the generating function of all matchings of Segn , i.e. the weighted matching
polynomial of Segn :


Un (x, a) = (−1)|µ| a EIND(µ) x n−2|µ| , (8)
µ∈M(Segn )

where |µ| is the number of edges of µ (thus n −2|µ| is the number of isolated vertices of µ)
and EIND(µ) the number of edges {i, i +1} with i even. It is possible to give a combinatorial
proof of (3) by using the above combinatorial interpretation (8) for Un (x, a).
Consider the complete graph K n on [n]. A matching µ of K n is noncrossing iff it has
no crossing pairs, i.e. there are no edges {i, j } and {i  , j  } of µ satisfying i < i  < j < j  .
 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509 503

Let µ be a noncrossing perfect matching of K n . The height of an edge {s, t}, s < t, in µ
is the number of edges {i, j } in µ such that i ≤ s and j ≥ t. It is known [10, 12] that the
number of
noncrossing

n  perfect matchings of K 2n with i even-height edges is a Narayana
number, n1 ni i+1 . Let EHE(µ) denote the number of edges of even height in µ. It follows
from (2) that


ϕn (x n ) = a EHE(µ) , (9)
µ∈M∗n

where M∗n is the set of noncrossing perfect matchings of K n .

Definition 1. Let n = (n 1 , . . . , n k ) be a composition of n. An edge {s, t} of K n is called
homogeneous (resp. consecutive) if both s and t belong to the same segment S j for some
j (resp. |s − t| = 1). We define a matching complex of type n to be a noncrossing perfect
weighted matching µ of K n satisfying the following:

1. There are two kinds of edges: red and blue. Blue edges are consecutive and
homogeneous.
2. Each red edge is weighted 1, if it is of odd height; a, if of even height. Each blue
edge {i, i + 1} is weighted −1, if i is of odd local parity; −a, if of even local parity.

An edge is said to be an a-edge, if its weight is a or −a; 1-edge, if its weight is 1 or −1.
Given a matching complex µ, let EINDB (µ) denote the number of blue a-edges {i, i + 1}
in µ, and let EHRE(µ) denote the number of red a-edges.
Let MC(n) be the set of matching complexes of type n. Interpreting blue edges as
matchings from the polynomials, and red edges as matchings from the moments, it follows
from (4), (8) and (9) that


I(n; a) = (−1)b(µ) a EINDB(µ)+EHRE(µ) , (10)
µ∈MC (n)

where b(µ) denotes the number of blue edges.

Although it is possible to describe our proof of the main theorem within the model of
matching complexes, it seems more convenient to switch to the model of Dyck paths as it
will be clear in the next section.
It is well known that the noncrossing perfect matchings of K n are in one-to-one
correspondence with Dyck paths of length n. Indeed, given a noncrossing perfect matching
µ of K n , we can define the step-type sequence type(µ) = t1 · · · tn of a Dyck path by setting
ti = u, if i is the smaller vertex of an edge in µ; ti = d, otherwise. Obviously the mapping
type is a bijection [10]. Thus, using this bijection and suitably weighting path steps we can
switch from a matching complex to a Dyck path complex, which is defined in the following.

Definition 2. Let n = (n 1 , . . . , n k ) be a composition of n. A path complex of type n is a

pair π = (γ , w) such that γ is a path with point sequence ( p0 , p1 , . . . , pn ), pi = (x i , yi ),
and w = (w1 , . . . , wn ) is the sequence of weights of steps si = ( pi−1 , pi ) satisfying:

• If pi is not a peak, then wi = a if si is up and yi is even; 1 if si is down or yi is odd.

504 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509

Fig. 2. Paths in H(6; 2, 4) and H(6; 1, 3) with step-type sequence (u, d, u, u, d, u).

• If pi is a peak, then wi = 1 or −1 if yi is odd and i is of odd local parity; 1 or −a

if yi is odd and i is of even local parity; a or −1 if yi is even and i is of odd local
parity; a or −a if yi is even and i is of even local parity.

If γ is a Dyck path, the corresponding path complex is called a Dyck path complex. A peak
(resp. valley) pi of γ is said to be homogeneous if i, i + 1 belong to the same S j for some
j ∈ [k]. The weight of pi is the product of wi and wi+1 . Furthermore pi is called positive
(resp. negative), if its weight is positive (resp. negative).
Let DC(n) be the set of all Dyck path complexes of type n whose negative peaks are all
homogeneous. For π ∈ DC(n), let EPNP(π) (resp. NP(π)) be the number of locally-even
position negative peaks (resp. negative peaks) of π and let EHPR(π) be the number of even
height positive rises of π.
Now we can restate (10) as follows:


I (n; a) = (−1)NP(π) a EPNP(π)+EHPR(π) . (11)
π∈DC (n)

Comparing (11) with (5), we need to find a wpsr involution on DC(n) whose fixed points
are in one-to-one correspondence with paths in DSV(n).

3. Fundamental involution

If the composition n reduces to (n) a path complex of type n is simply called a weighted
path of length n. Note that the peaks of any weighted path are homogeneous. Let H(n; k, l)
be the set of weighted paths of length n whose underlying paths go from (0, k) to (n, l).
Clearly one should require that n ≥ |k − l| and n ≡ k + l mod 2 in order to have at least
one such path in H(n; k, l).
Example. All weighted paths in H(6; 2, 4) and H(6; 1, 3) whose underlying paths have
step-type sequence (u, d, u, u, d, u) are shown in Fig. 2, where weight 1 of each down-step
is omitted.
Since a lattice path is determined by its step-type sequence when the starting and
ending points are fixed, we can encode any weighted path π ∈ H(n; k, l) by a biword
 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509 505

Fig. 3. Patterns V + , V − , L + , N − , N + and L − ( is the starting point).

 
t1 · · · tn
π = , where (t1 , . . . , tn ) is the step-type sequence and (w1 , . . . , wn ) the
w1 · · · wn
weight sequence. We say that a biword τ is a pattern of π if it is a subword of π, i.e.
π = ατβ for some biwords α and β, where the composition of words is concatenation.
In order to describe our involution we shall need the following patterns:
     
+ d u + d d u + u d u
V = , L = , N = ,
1 1 1 1 + + 1 +
     
u d d u d u u d
V− = , N− = , L− = ,
−1 1 1 − 1 + − 1
where ‘+’ means 1 or a and ‘−’ means −1 or −a, moreover, we require that the negative
peak in the pattern N − is of height at least 1. The patterns are illustrated in Fig. 3.
Definition 3. Define H∗ (n; k, l) to be the set of all weighted paths π ∈ H(n; k, l)
satisfying one of the following conditions:
(i) π starts with (n − k + l)/2 positive up-steps and ends with (n + k − l)/2 down-steps.
(ii) π has s, s > 0, negative valleys of height 0, say pr1 , pr2 , . . . , prs with r1 < r2 · · · <
rs ; it starts with (r1 −k)/2 positive up-steps and is followed by (r1 +k)/2 down-steps;
for each i , 1 ≤ i ≤ s − 1, the portion of π from pri to pri+1 is one negative peak
followed by (ri+1 −ri − 2)/2 positive up-steps and (ri+1 −ri − 2)/2 down-steps; the
portion of π after prs is one negative peak followed by (n − rs − 2 + l)/2 positive
up-steps and (n − rs − 2 − l)/2 down-steps.
Example. Two weighted paths of type (ii) in H∗ (29; 3, 2) are illustrated in Fig. 4, where
π1 and π2 coincide starting from the eighth step. Note that π1 has four negative valleys
of height 0 with abscissas 9, 11, 19, 25 respectively, while π2 has one more such valley,
(3, 0).
∗ (n; k, l) = H(n;
The following lemma gives a characterization of the set H
∗
k, l)\H (n; k, l).
∗ (n; k, l) if and only if it contains one of the
Lemma 1. Assume that k is odd. Then π ∈ H
+ − + − +
patterns V , V , N , N , L and L . −

Proof. The condition is clearly sufficient. It remains to show the ‘only if’ part.
Note that if pi , i > 1, is a negative peak of height 1 then pi−1 must be a negative valley
of height 0, and if pi , i < n − 1, is a negative valley of height 0 then pi+1 must be a
negative peak of height 1.
Let π be a path in H ∗ (n; k, l), then it must contain one of the following patterns: (1) a
negative peak of height greater than 1, (2) a valley of height greater than 0, (3) a positive
506 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509

Fig. 4. Two paths π1 and π2 in H∗ (29; 3, 2).

valley of height 0 preceded by two positive steps. Now, in case (1), it has a portion of type
V − , N − or L − ; in case (2), π has a valley pi of height greater than 0, which entails the
following four cases:

(i) if π has a positive valley p1, then it has a portion of type V + ;

(ii) if pi is negative, then it has a negative peak pi+1 of height greater than 1;
(iii) if i > 1 and pi is positive and wi−1 is positive, then it has a portion of type N + or
L+;
(iv) if i > 1 and pi is positive and wi−1 is negative, then it has a negative peak pi−1 of
height greater than 1;

in case (3), it has a pattern N + or L + .

The following result is crucial for the proof of our main theorem.
Proposition 1. Suppose that n ≥ |k − l| and n ≡ k + l(mod 2). Then there is a weight-
preserving sign-reversing involution ψ on H(n; k, l) which has no fixed point, if n > k +l;
exactly one fixed point, if n ≤ k + l. In the latter case, if k is even the unique path starts
with (n + k − l)/2 down-steps and ends with (n − k + l)/2 up-steps; if k is even the unique
path starts with (n − k + l)/2 up-steps and ends with (n + k − l)/2 down-steps.
Proof. We construct a wpsr involution ψ on H(n; k, l) as follows: let π ∈ H(n; k, l).

1. If k is even, we distinguish two cases.

• π contains a peak: Let pi be the leftmost one, we define ψ(π) by changing the
i th weight of π to its opposite (this is possible because i and yi have the same
parity when k is even).
• π contains no peak: Then π must start with (n + k − l)/2 down-steps and end
with (n − k + l)/2 up-steps, clearly n ≤ k + l. Define ψ(π) = π.
2. If k is odd, we distinguish three cases.
 
∗ (n; k, l): If π = V + β or V − β then set ψ(π) = t2 t1
• π ∈ H β; else
−w1 w2
if π contains a pattern τ = L + , N + , L − or N − then find the leftmost
 such a
ti ti+1 ti+2
portion, i.e. find the smallest i such that π = α β = ατβ
wi wi+1 wi+2
 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509 507
 
ti ti+2 ti+1
and define ψ(π) = α β. Clearly ψ has no fixed points in
wi −wi+1 wi+2
this case.
• π ∈ H∗ (n; k, l) with n > k + l: Then π is one of the following forms:
   k    k    
u d d d u d
U= α β, D= α β  ,
a 1 1 1 −a 1

where α and α  are either empty or a sequence of positive up-steps; and β and β 
are either empty or start with a down-step. In either case, define ψo (π) to be the
path of the other form with α = α  and β = β  . Clearly ψ has no fixed points in
this case.
• π ∈ H∗ (n; k, l) with n ≤ k + l: Clearly, π is the unique path of type (i) in the
definition of H∗ (n; k, l) because a path of type (ii) must have a length greater
than k + l. Define ψ(π) = π.

Summarizing all the above yields the proposition.

Example. The involution ψ on H∗ (29; 3, 2) is illustrated in Fig. 4, where ψ : π1 → π2 .

By computing the number of even up-steps of the weighted path fixed by ψ, we derive
from the above proposition that the generating function of H(n; k, l) is 0 if n > k + l,
a (n−k+l)/4 if n ≤ k + l with k even and l odd, a (n−k+l)/4 otherwise.

4. Proof of the main theorem

For any sequence m = (m 0 , m 1 , . . . , m k ) of nonnegative integers with m 0 = m k = 0,

let DC(n; m) denote the set of all Dyck path complexes in DC(n) whose underlying Dyck
paths have the ordinate m i at abscissa n 1 + · · · + n i for i = 1, 2, . . . , k. Similarly we can
define the subset DSV(n; m) of DSV(n). Thus we have the following decompositions:

DC(n) = DC(n; m), DSV(n) = DSV(n; m).
m m

We now define a wpsr involution on DC(n; m). Any Dyck path complex π in
DC(n; m) will be identified with a k-tuple (π1 , . . . , πk ) of weighted paths, where πi ∈
H(n i ; m i−1 , m i ) is the restriction of π on the segment Si for 1 ≤ i ≤ k.
For π = (π1 , . . . , πk ) in DC(n; m), if ψ(πi ) = πi for all i , set Ψ (π) = π; otherwise,
set Ψ (π) = (π1 , . . . , ψ(πi ), . . . , πk ), where i is the smallest integer such that ψ(πi ) = πi .
Clearly Ψ is a wpsr involution on DC(n; m). According to Proposition 1, there is a unique
fixed point π◦ in DC(n; m) if |m i−1 − m i | ≤ n i ≤ m i−1 + m i for all i ; no fixed point,
otherwise. Moreover, if π◦ = (π1 , . . . , πk ) with πi ∈ H(n i ; m i−1 , m i ), then each πi is
the path described by Proposition 1, i.e. if m i−1 is even (resp. odd) then πi starts with
(n i + m i−1 − m i )/2 down-steps (resp. up-steps) and ends with (n i − m i−1 + m i )/2
508 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509

Fig. 5. Bijection Φ : π → π ∗ with (n; m) = ((3, 4, 3, 4, 3, 6, 3, 2); (0, 3, 5, 2, 4, 3, 3, 2, 0)).

Fig. 6. Involution Ψ on the path complexes of DC(1, 5) with weight a or −a.

up-steps (resp. down-steps). Let F(n; m) denote the fixed point set of Ψ . Clearly, we have
EHPR(π) = EHR(π) for all π ∈ F(n; m). It follows from (11) that



I (n; a) = a EHR(π) . (12)
m π∈F (n;m)

Note that a path complex in F(n; m) has a homogeneous peak in Si if m i−1 is odd
and n i > |m i−1 − m i |, while no path in DSV(n; m) has any homogeneous peak. To
complete the proof, we construct a bijection Φ : π → π ∗ from F(n; m) to DSV(n; m) by
setting π ∗ = (π1∗ , . . . , πk∗ ), where πi∗ = πi , if m i−1 is even; πi∗ is the path starting with
(n i + m i−1 − m i )/2 down-steps and ending with (n i − m i−1 + m i )/2 up-steps, in other
words, the step-type sequence of πi∗ is the reverse of that of πi , if m i−1 is odd. Clearly, the
resulting path π ∗ is an element of DSV(n; m) and the map is a bijection.
Now we compare the weights of π and π ∗ . For each i , if π ∗ has a special valley
in Si , i.e. both m i−1 and (n i − m i−1 + m i )/2 are odd and n i is even, then πi∗ has
one more even positive up-step than πi ; otherwise, πi∗ has the same number of them as
πi . So if we weight each special valley with a, then Φ becomes weight-preserving, i.e.
EHR(π) = EHR(π ∗ ) + SVAL (π ∗ ). Our main theorem follows from (12).

Example. An example of Φ : π → π ∗ from F(n; m) to DSV(n; m) is illustrated in Fig. 5

for n = (3, 4, 3, 4, 3, 6, 3, 2) and m = (0, 3, 5, 2, 4, 3, 3, 2, 0), where the two special
valleys are circled. While the involution Ψ on the path complexes of DC(1, 5) with weight
a or −a is illustrated in Fig. 6.
 D. Kim, J. Zeng / European Journal of Combinatorics 24 (2003) 499–509 509

Fig. 7. De Sainte-Catherine and Viennot’s involution with n = (1, 5).

Remark. When a = 1, De Sainte-Catherine and Viennot [5] have given a sign-reversing

involution on DC(n), which consists of changing the leftmost homogeneous negative peak
to a positive peak and vice versa. As shown in Fig. 7, for a = 1, their involution does
not preserve weights. While it is obvious from the algebraic expression (4) that I (n; a) is
symmetric by permuting n i in n, this is not clear from our combinatorial interpretation, but
Guoniu Han (private communication) has recently constructed an involution on DSV(n)
exhibiting this symmetry.

Acknowledgements
This work was partially supported by CNRS and KOSEF:20015-102-01-2.

References
[1] R. Askey, Linearization of the product of orthogonal polynomials, in: Robert C, Gunning (Ed.), Problems in
Analysis (A Symposium in Honor of Salomon Bochner), Princeton University Press, Princeton, NJ, 1970,
pp. 131–138. CBMS Regional Conference Series in Applied Mathematics, vol. 21, SIAM, Philadelphia,
1975.
[2] R. Askey, Orthogonal Polynomials and Special Functions, CBMS Regional Conference Series in Applied
Mathematics, vol. 21, SIAM, Philadelphia, 1975.
[3] T.S. Chihara, An Introduction to Orthogonal Polynomials, in: Mathematics and its Applications, vol. 13,
Gordon and Breach Science Publishers, New York–London–Paris, 1978.
[4] A. De Médicis, D. Stanton, D. White, The combinatorics of q-Charlier polynomials, J. Combin. Theory Ser.
A 69 (1) (1995) 87–114.
[5] M. De Sainte-Catherine, X.G. Viennot, Combinatorial Interpretation of Integrals of Products of Hermite,
Laguerre and Tchebycheff Polynomials, Lecture Notes in Mathematics, vol. 1171, Springer-Verlag,
1985, pp. 120–128.
[6] D. Foata, D. Zeilberger, Laguerre polynomials, weighted derangements and positivity, SIAM J. Discrete
Math. 1 (1988) 425–433.
[7] I. Gessel, Generalized rook polynomials and orthogonal polynomials, in: q-Series and Partitions,
Minneapolis, MN, 1988, IMA Vol. Math. Appl., vol. 18, Springer, New York, 1989, pp. 159–176.
[8] M. Ismail, M. Tamhankar, A combinatorial approach to some positivity problems, SIAM J. Math. Anal. 10
(1979) 478–485.
[9] D. Kim, J. Zeng, A combinatorial formula for the linearization coefficients of general Sheffer polynomials,
European J. Combin. 22 (2001) 313–332.
[10] R.P. Stanley, Enumerative Combinatorics, vol. 2, Cambridge University Press, 1999.
[11] R. Simion, D. Stanton, Octabasic Laguerre polynomials and permutation statistics, J. Comput. Appl. Math.
68 (1–2) (1996) 297–329.
[12] R.A. Sulanke, Constraint-sensitive Catalan path statistics having the Narayana distribution, Discrete Math.
204 (1999) 397–414.
[13] M. Vauchassade de Chaumont, X.G. Viennot, Polynômes orthogonaux et problèmes d’énumération en
biologie moléculaire, Sém. Lothar. Combin. B08l (1984) 8.
[14] X.G. Viennot, Une théorie combinatoire des polynômes orthogonaux généraux, Notes de conférences
données à l’UQAM, Montréal, 1983.