Combinatorica 24 (3) (2004) 427–440

C OM BIN A TORIC A
Bolyai Society – Springer-Verlag

MATCHING POLYNOMIALS AND DUALITY

BODO LASS

Received February 20, 2001

Let G be a simple graph on n vertices. An r-matching in G is a set of r independent edges.

The number of r-matchings in G will be denoted by p(G, r). We set p(G, 0) = 1 and define


n/2
the matching polynomial of G by µ(G, x) := r=0 (−1)r · p(G, r) · xn−2r and the signless


n/2
matching polynomial of G by µ(G, x) := r=0 p(G, r) · xn−2r .
It is classical that the matching polynomials of a graph G determine the matching
polynomials of its complement G. We make this statement more explicit by proving new
duality theorems by the generating function method for set functions. In particular, we
2 2
show that the matching functions e−x /2 µ(G, x) and e−x /2 µ(G, x) are, up to a sign, real
Fourier transforms of each other.
Moreover, we generalize Foata’s combinatorial proof of the Mehler formula for Hermite
polynomials to matching polynomials. This provides a new short proof of the classical
fact that all zeros of µ(G, x) are real. The same statement is also proved for a common
generalization of the matching polynomial and the rook polynomial.

1. Introduction

Let V be a finite set of vertices, n := |V |; and let G = (V, E) be a simple

 
graph, i.e. E, the set of edges, is a subset of V2 , the family of all 2-element
 
subsets of V . The complement of G is the graph G = (V, E) with E = V2 \E.
An r-matching in G is a set of r edges of G, no two of which have a
vertex in common. Clearly, r ≤ n/2. If r = n/2, then an r-matching of
G is called perfect if n is even and quasi-perfect if n is odd. Let p(G, r) be

Mathematics Subject Classification (2000): 05C70; 05A15, 05A20, 05E35, 12D10

0209–9683/104/$6.00
2004
c János Bolyai Mathematical Society
428 BODO LASS

the number of r-matchings in G, with the convention that p(G, 0) := 1. The

matching polynomial of G is (see [3], chapter 1)

n/2

µ(G, x) := (−1)r · p(G, r) · xn−2r ,
r=0

while the signless matching polynomial reads

n/2

µ(G, x) := p(G, r) · xn−2r .
r=0

In particular, µ(G, 0) counts the number of perfect matchings of G and

µ(G, 1) counts the number of arbitrary matchings of G.
These polynomials were introduced by Heilmann and Lieb [5], who, moti-
vated by statistical physics, mainly studied their zeros. They obtained many
estimations on the locations of those zeros and provided several different
proofs for their main theorem that all zeros of µ(G, x) are real. Another
proof of this theorem was obtained by Godsil, which he reproduced in his
recent book [3] together with all the classical proofs. However, all those
proofs rely on a recursive approach (via the deletion of a special vertex) to-
wards the matching polynomial. One of the purposes of this paper is to give
a short proof that avoids this traditional deletion technique.
We generalize the combinatorial proof of the Mehler formula for Hermite
polynomials imagined by Foata (see [2]). To state and derive our extension
of the Mehler formula we need an adequate algebraic tool, the algebra of
generating functions for set functions. This algebra is developed in the next
section. In section 4 we show that our generalization of the Mehler formula
immediately implies that |µ(G, x)|2 ≥ [( m x)2 ]n + 2|E| · [( m x)2 ]n−1 for
every x ∈ C, i.e. that all the zeros of µ(G, x) are real. Moreover, this fact
holds for a common generalization of the matching polynomial and the rook
polynomial.
Section 3 is entirely devoted to questions of duality. It is evident that
µ(G, x) and µ(G, x) contain the same information, a convenient relation
between those polynomials being
√
µ(G, x) = (−i)n · µ(G, xi), |V | = n, i = −1.
The observation that µ(G, x) and µ(G, x) determine µ(G, x) and µ(G, x),
however, seems to have first been made by Lovász, [8] 5.18. His proof, based
on the inclusion-exclusion principle, does not seem to be too difficult al-
though it is marked with an asterisk in his book. We prefer to give an ex-
plicit calculation involving several new duality theorems. In particular, we
 MATCHING POLYNOMIALS AND DUALITY 429

show that the matching functions e−x /2 µ(G, x) and e−x /2 µ(G, x) are, up
2 2

to a sign, real Fourier transforms of each other.

Matching polynomials generalize many classical orthogonal polynomi-
als, namely Hermite polynomials (the matching polynomials of complete
graphs), Tchebycheff polynomials of both kinds (paths and cycles) and La-
guerre polynomials (complete bipartite graphs), see [9] chapter 6. The com-
plete graph Kn on n vertices can be defined as the complement of the edgeless
graph Kn on those vertices. Clearly, µ(Kn , x) = µ(Kn , x) = xn . The Hermite
polynomials will be (by definition) the matching polynomials of the com-
plete graphs, i.e. Hen (x) := µ(Kn , x). The reader will recognize the classical
definition of the Hermite polynomials as a special case of the second equality
in our duality theorem (d/dx), replacing G by Kn .
The first step towards the duality theory developed here was the com-
binatorial interpretation of integrals over products of Hermite polynomials
(see [9], VI-34, remark 21, or the recent book [1], chapter 6.9):
 ∞
1
e−x
2 /2
µ(Kn1 ··· Knk , 0) = √ · µ(Kn1 , x) · · · µ(Knk , x) · dx,
2π −∞

where Kn1 · · · Knk denotes the vertex disjoint union of the complete
graphs Kn1 , . . . , Knk . Even this formula has been neglected in physics (for
instance, in [5] there is no consideration of duality at all) as can be seen
from the fact that Itzykson and Zuber [6] need a whole page to solve their
integral (3.5), the result being evident from the previous formula.
The second step was the realization, due to Godsil [4], that, more gener-
ally,
 ∞
1
e−x /2 · µ(G, x) · dx,
2
µ(G, 0) = √
2π −∞
the equation µ(Kn1 , x) · · · µ(Knk , x) = µ(Kn1 · · · Knk , x) being evident.
Finally, the last step of duality theory is the derivation of direct formulas
for the matching polynomials of the complementary graph. In this context
Godsil mainly proposed

n/2

µ(G, x) = p(G, r) · µ(Kn−2r , x),
r=0

but also the more explicit formula

 ∞
1
e−x
2 /2
µ(G, y) = √ · µ(G, x + y) · dx
2π −∞
430 BODO LASS

is a specialization of theorem 2.4 in [4], that Godsil did not reproduce in his
book [3].

We propose two new duality theorems by means of differential operators,

a slight modification of Godsil’s last theorem together with a scalar product
formula and, finally, the Fourier transform interpretation.
Our proofs of section 3 make use of generating functions for set functions.
We must say that the set function machinery developed in the next section
has been the adequate tool for deriving those duality theorems.

2. Algebraic tools

Let V be a finite set and

f : 2V → A
V  ⊆ V → f (V  ) ∈ A

be a set function, where A is a commutative ring with 1. Consider the

generating function
 
Ff (ν) := f (V  ) · ν V , ν ∅ := 1,
V  ⊆V

subject to the following multiplication rules (V  , V  ⊆ V ):

   +V 
ν V · ν V := ν V , where


  V  ∪ V  , if V  ∩ V  = ∅,
V + V :=
†, if V  ∩ V  = ∅, where

† + V  := †, † + † := †, and ν † := 0.

The algebra A[V ] of those generating functions is not unknown. In fact, we

have the isomorphism

A[V ]  A[v1 , . . . , vn ]/v12 , . . . , vn2 ,

if V contains n elements.
 MATCHING POLYNOMIALS AND DUALITY 431

Example. For V  ⊆ V put


(f g)(V  ) := f (V  ) · g(V  )
V  =V  V 

(f, g, f g : 2V → A). Then

Ff g (ν) = Ff (ν) · Fg (ν).

For |V | = ∞ let F (V ) be the partially ordered set of finite subsets of V . We

have the canonical projections pV  ,V  : A[V  ] → A[V  ] (V  , V  ∈ F (V ), V  ⊇
V  ) and define
A[V ] := lim A[V  ], V  ∈ F (V )
←−

in order to work with generating functions of the form

 
Ff (ν) = f (V  ) · ν V .
V  ∈F (V )

Let 
V := ν {v}
v∈V

be the generating function for the indicator function of the subsets of V of

cardinality 1 (the double use of V for the set itself on the one hand and
for an element of A[V ] on the other hand will never cause confusion). In
the product V n each subset of cardinality n occurs n! times, so that V n /n!
represents the indicator function of the subsets of the set V of cardinality
n. The identity
∞
  
f (n) · V n /n! = f (|V  |) · ν V , f : N → A,
n=0 V  ∈F (V )

provides an imbedding of the ring A![[V ]] of generating functions of expo-

nential type (usually the variable is called x instead of V ) into our ring A[V ].
It is at the origin of (almost?) all applications of A![[V ]] into combinatorics,
but requires the existence of an infinite combinatorial model depending just
on cardinalities. Consequently, A[V ] gives more flexibility and closeness to
combinatorics. In addition, A[V ] is ideally suited for computer calculations
(for more details and lots of different applications see [7]).
Remark. The ring Z![[V ]] is not noetherian, but it contains the important
functions exp(V ) and log(1 + V ).
432 BODO LASS

Example. If char A = 2, then we have

∞

(1 + V )−1 = (−1)n n! · V n /n!
n=0
≡1+V and
∞

log(1 + V ) = (−1)n−1 (n − 1)! · V n /n!
n=1
≡ V + V 2 /2
in the ring A![[V ]]. These identities are at the origin of lots of results of
parity in combinatorics.
  
For all t ∈ A we put (tν)V := t|V | · ν V , V  ⊆ V , and therefore
  
Ff (tν) = f (V  ) t|V | · ν V .
∅⊆V  ⊆V

It is evident that this definition is compatible with the addition and the
multiplication. Most important are the special cases t = −1 and t = 0: Ff (0) =
Ff (0·ν) = f (∅). If Ff (0) = 0, then Ff (ν)n /n! is defined for any ring A, because
a partition into n nonempty subsets can be ordered in n! different ways.
Thus we have an operation of A![[V ]] on A[V ] via the substitution G(Ff (ν))
defined for any G ∈ A![[V ]].
Finally, define for any f, g : 2V → A the function f ∗ g : 2V → A by
(f ∗ g)(V  ) := f (V  ) · g(V  )
for each V  ⊆ V and define the Hadamard product to be
Ff (ν) ∗ Fg (ν) := Ff ∗g (ν).

3. Duality theorems

Let G = (V, E) be a finite simple graph and let G = (V, E) be its complement.
We have identified V with the generating function of the indicator function
of the one-element subsets of V , and we have realized that V 2 /2 corresponds
to the indicator function of the two-element subsets of V . Similarly, we
identify E and E with the generating functions of the indicator functions of
E, E ⊆ 2V . Since every two-element subset of V either belongs to E or to E,
we have the following fundamental identity, valid in the ring A[V ]:
E + E = V2 /2.
 MATCHING POLYNOMIALS AND DUALITY 433

Let p(G) be the number of perfect matchings (if |V | ≡ 0 mod 2) or quasi-

perfect matchings (if |V | ≡ 1 mod 2), and let c(G) be the number of arbitrary
matchings of G. We denote by G[V  ] the subgraph of G induced by V  ⊆ V ,
i.e. its vertices are the elements of V  and its edges are the edges of G having
both endpoints in V  . Then the perfect matchings are counted by exp[E], the
quasi-perfect matchings by V ·exp[E], and altogether we have the identities
 
1+ p(G[V  ]) · ν V = (1 + V ) · exp[E],
∅⊂V  ⊆V
 
1+ c(G[V  ]) · ν V = exp[V + E].
∅⊂V  ⊆V

The following proposition was proved in [8], 5.18, for the case of perfect
matchings.
Proposition. c(G) ≡ p(G) mod 2.
Proof. Using the previous three identities we have

exp[V + E] = exp[V + V 2 /2− E] ≡ exp[log(1+ V )+ E] = (1+ V )·exp[E].

From the very definitions of the matching polynomials we have the fol-
lowing generating functions:
 
1+ µ(G[V  ], x) · ν V = exp[xV − E],
∅⊂V  ⊆V
 
1+ µ(G[V  ], x) · ν V = exp[xV + E].
∅⊂V  ⊆V

Considering the first equality for the complete graph K∞ on an infinite

set of vertices yields the classical generating function of exponential type for
Hermite polynomials:
∞

1+ µ(Kn , x) · V n /n! = exp[xV − V 2 /2].
n=1

We are now in a position to provide a very short proof of Godsil’s duality

theorem.
Duality theorem (Godsil).

n/2

µ(G, x) = p(G, r) · µ(Kn−2r , x).
r=0
434 BODO LASS

Proof. Using the set function algebra developed in section 2 we have:

exp[xV − E] = exp[xV − V 2 /2 + E] = exp[E] · exp[xV − V 2 /2].

We can further establish the following new identities.

d 2
Duality theorem for the matching polynomials ( dx 2 ).

2
2 /2] · µ(G, x),
d
µ(G, x) = exp[ dx
2
2 /2] · µ(G, x).
d
µ(G, x) = exp[− dx
Proof. As in the preceding proof,
exp[xV + E] = exp[V 2 /2] · exp[xV − E]
2
2 /2] · exp[xV − E],
d
= exp[ dx
2
because dxd
exp[xV −E] = V ·exp[xV −E]. The differential operator exp[− dx
d
2 /2]
d2
is the inverse of exp[ dx2 /2].
d
Duality theorem for the matching polynomials ( dx ).

µ(G, x) = e−x
2 /2 2 /2
· µ(G, dx
d
) · ex ,
x2 /2 −x2 /2
µ(G, x) = e · µ(G, − dx
d
)·e .
Proof. By the Taylor formula we know that
d
f (x + a) = exp[ dx a] · f (x)
for variables x, a and a formal power series f . It follows that
exp[−x2 /2] · exp[ dx
d
V − E] · exp[x2 /2]
= exp[−x2 /2] · exp[−E] · exp[ dx
d
V ] · exp[x2 /2]
= exp[−x2 /2] · exp[−E] · exp[(x + V )2 /2]
= exp[xV + E].
The second equality is proved in the same way.
Specializing the second equality of the preceding theorem to Hermite
polynomials, i.e. replacing G by Kn , provides the classical definition of Her-
mite polynomials. However, we could not find the specialization to Hermite
polynomials of the first equality, i.e. the differential operator Hen (d/dx), in
the literature.
Finally, we can prove several integral formulae.
 MATCHING POLYNOMIALS AND DUALITY 435


Duality theorem for the matching polynomials ( ).
 ∞
1
e−(x−y)
2 /2
µ(G, y) = √ · µ(G, x) · dx.
2π −∞

Proof. Using the invariance of the integral with respect to translations we

get:
 ∞ 
1
√ exp −(x − y)2 /2 · exp[xV − E] · dx
2π −∞
 ∞
1 
=√ exp −s2 /2 · exp[(s + y)V − E] · ds
2π −∞

1 ∞ 
= exp[yV + E] · √ exp −(s − V )2 /2 · ds
2π −∞
1 ∞ 
= exp[yV + E] · √ exp −t2 /2 · dt
2π −∞
= exp[yV + E].

For graphs G = (V  , E  ) and G = (V  , E  ) we have the following result.

Scalar product formula.
∞
1
µ(G , x) · µ(G , x) · e−x /2 · dx
2
µ(G G , 0) =√
2π −∞
= µ(G , dx
d
) · µ(G , x) .
x=0

Proof. The first equality being evident from the previous theorem, we just
have to prove the second one:
d
exp[ dx V  + E  ] · exp[xV  + E  ]
x=0
 
= exp[E  ] · exp[(x + V )V + E  ]
x=0
= exp[V  V  + E  + E  ].

Remark. If G = Kn and G = Km , then the scalar product formula counts

the number of perfect matchings of the complete bipartite graph Kn Km ,
which is equal to zero, if n = m, and equal to n!, if n = m. This is the
orthogonality of the Hermite polynomials.
n 
(−i) ∞ −(x−yi)2 /2
The previous duality theorem implies µ(G, y) = √
2π −∞
e ·
µ(G, x) · dx. This proves the following theorem.
436 BODO LASS

Duality theorem for the matching polynomial (C).

 ∞
−y 2 /2 (−i)n
exyi · e−x
2 /2
e µ(G, y) = √ µ(G, x) · dx.
2π −∞

If we call e−x /2 µ(G, x) matching function of G, then this matching func-

2

tion is even for n even and odd for n odd.

Duality theorem for the matching polynomial (R).
 ∞
2
e−y cos(xy) · e−x
2 /2 2 /2
µ(G, y) · (−1)n/2 = · µ(G, x) · dx, n even,
π 0
 ∞
2
e−y sin(xy) · e−x
2 /2 2 /2
µ(G, y) · (−1)(n−1)/2 = · µ(G, x) · dx, n odd.
π 0

Thus the matching functions of G and G are, up to an eventual multiplication

by −1, real Fourier transforms of one another.

4. Zeros

From now on every edge {u, v} ∈ E of our graph G = (V, E) will get a positive
real weight w{u,v} (we can assume that the two-element subsets of V which
are not edges get the weight zero). This weighted graph will be denoted
by Gw = (V, Ew ). In particular, Ew will be identified with the generating
function of the set function which attributes the value 0 to all subsets of
V with the only exception of the edges of G, which get their own weights.
The (weighted) matching polynomial can be defined with the help of its
generating function:
 
1+ µ(Gw [V  ], x) · ν V = exp[xV − Ew ].
∅⊂V  ⊆V

We see that every matching is counted with respect to its weight: the product
of the weights of its edges.
A Hamiltonian cycle of Gw is a cyclic order of V and its weight is the
product of the weights of its n = |V | edges corresponding to two consecu-
tive vertices in the cyclic order. In particular, if the edge corresponding to
two consecutive vertices in the cyclic order does not belong to the graph
(equivalently, has weight zero), then the weight of that “Hamiltonian cycle”
is equal to zero. Let cyc(Gw ) be the sum of the weights of all Hamiltonian
cycles of Gw , with the convention that cyc(Gw ) = 1 if n = 1. We assume
 MATCHING POLYNOMIALS AND DUALITY 437

that the weight of each edge in the complete graph Kn is equal to 1, so that
cyc(Kn ) = (n − 1)!.
A Hamiltonian path of Gw is a linear order of V and its weight is the
product of the weights of its n − 1 edges corresponding to two consecutive
vertices in the linear order. Let lin(Gw ) be the sum of the weights of all
Hamiltonian paths of Gw , with the convention that lin(Gw ) = 1 if n = 1.
Clearly, lin(Kn ) = n!.
Let us put
   
cycGw (ν) := cyc(Gw [V  ])·ν V , linGw (ν) := lin(Gw [V  ])·ν V .
∅⊂V  ⊆V ∅⊂V  ⊆V

Considering the infinite graph K∞ yields

∞
 ∞
 V
cyc(Kn ) · V n /n! = − log(1 − V ), lin(Kn ) · V n /n! = .
n=1 n=1
1−V

Usually (in undirected graphs) one does not distinguish between the two
different directions of Hamiltonian cycles or paths. In this sense cycGw (ν)
and linGw (ν) count them “twice”. Now we can prove our generalization of
the Mehler formula.
Theorem. Using the Hadamard product ∗ we have:
exp[xV − Ew ] ∗ exp[yV − Ew ]
1 1
= exp · cycGw (ν) + · cycGw (−ν) ·
2 2
 x−y 2 x+y 2
exp − · linGw (ν) − · linGw (−ν) .
2 2
Proof. Two matchings of Gw to be considered in the left hand side of the
theorem provide a partition of V into even Hamiltonian cycles (to be counted
“twice”, because the matchings can be interchanged), even (according to the
number of vertices) Hamiltonian paths (to be counted with the factor −x2 or
−y 2 , because the number of edges of the paths is odd) and odd Hamiltonian
paths (to be counted with the factor 2xy). Thus the left hand side is equal
to
 
cycGw (ν) + cycGw (−ν)
exp ·2 ·
4
 
linGw (ν) + linGw (−ν)
exp · (−x2 − y 2 ) ·
4
 
linGw (ν) − linGw (−ν)
exp · 2xy .
4
438 BODO LASS

But this is precisely the right hand side of the theorem.

Specializing to K∞ yields the Mehler formula.

Corollary (Mehler).

∞

1+ µ(Kn , x)µ(Kn , y)·V n /n! = √ 1
1−V 2
·exp[( x+y
2 ) · 1+V
2 V −( x−y
2 ) · 1−V ].
2 V

n=1

Replacing y in the previous theorem by the complex conjugate number

x yields the following corollary.

Corollary.

exp[xV − Ew ] ∗ exp[xV − Ew ]
1 1
= exp · cycGw (ν) + · cycGw (−ν) ·
2 2
exp[( m x)2 · linGw (ν) − (e x)2 · linGw (−ν)]
= exp[( m x)2 · linGw (ν)] · exp[(e x)V − Ew ] ∗ exp[(e x)V − Ew ] .

Therefore |µ(Gw , x)|2 ≥ [( m x)2 ]n +2W ·[( m x)2 ]n−1 for every x ∈ C, where
W is the sum of the weights of all edges of Gw . In particular, all zeros of
µ(Gw , x) are real.

We finish this article by considering a common generalization of the

matching polynomial and the classical rook polynomial. Thus we do not
just have our weighted graph Gw = (V  , Ew  ), but also a bipartite graph

Gw = (V  , V  ; Ew  ). In other words, we have a graph G = (V, E ) with ver-

w w
tex set V := V V  , edge set Ew := Ew
  Ew and, in particular, no edges

between vertices of V  . The weights of the edges of Gw are still assumed to
be positive, whereas the weights of the edges of Gw are supposed to be such
that for each v  ∈ V  the weights of the edges incident with v  all have the
same sign, i.e. they are all positive or all negative.
Let exp[xV  + V  − Ew  − E  ] be the generating function of our general-
w
ized matching polynomials, and let cycGw (ν) count the Hamiltonian cycles
“twice”. Moreover, let linGw (ν), linGw (ν), linGw (ν) count the Hamiltonian
paths “twice” which have both endpoints in V  , one endpoint in V  and one
endpoint in V  , both endpoints in V  , respectively. (Note that linGw (ν) is
nonnegative by our restrictions on the weights.)
 MATCHING POLYNOMIALS AND DUALITY 439

Theorem.
exp[xV  + V  − Ew 
− Ew 
] ∗ exp[yV  + V  − Ew
− Ew 
]
 
1 1
= exp · cycGw (ν) + · cycGw (−ν) ·

2 2 
x−y 2 x+y 2
exp − · linGw (ν) − · linGw (−ν) ·

2 
2
x+y
exp − · linGw (−ν) · exp[− linGw (−ν)].
2
Proof. Clearly, both sides of the equality are equal to
 
cycGw (ν)+ cycGw (−ν)
exp ·2 ·
4
   
linGw (ν)+ linGw (−ν) linGw (ν)− linGw (−ν)
exp · (−x2 −y 2 ) · exp · 2xy ·
4 4
       
linGw (ν)+ linGw (−ν) linGw (ν)− linGw (−ν)
exp (−x−y) · exp (x+y) ·
4 4
    
linGw (ν)+ linGw (−ν) linGw (ν)− linGw (−ν)
exp · (−2) · exp ·2 .
4 4

Corollary.
exp[xV  + V  − Ew

− Ew
] ∗ exp[xV  + V  − Ew
 
− Ew ]
1 1
= exp · cycGw (ν) + · cycGw (−ν) ·
2 2
exp[( m x)2 · linGw (ν) − (e x)2 · linGw (−ν)] ·
exp[−(e x) · linGw (−ν)] · exp[− linGw (−ν)]
= exp[( m x)2 · linGw (ν)] ·
exp[(e x)V  + V  − Ew
 
− Ew ] ∗ exp[(e x)V  + V  − Ew
 
− Ew ] .

Since linGw (ν) is nonnegative, all zeros of our generalized matching polyno-
mial are real.

Acknoledgements. First of all, I should like to thank cordially D. Foata to

have made the redaction of this article possible by inviting me to the IRMA
in Strasbourg and by explaining to me his combinatorial proof of the Mehler
formula. His help and support are undescribable. I also want to express my
gratitude to A. Frank for his recommendation of the beautiful book [3]
(otherwise [7], chapter 3.1, would not exist), and especially to E. Triesch
(without him [7] would not exist).
440 BODO LASS: MATCHING POLYNOMIALS AND DUALITY

References

[1] G. E. Andrews, R. Askey and R. Roy: Special Functions, Encyclopedia of mathe-

matics and its applications, Vol. 71, Cambridge University Press, 1999.
[2] D. Foata: A combinatorial proof of the Mehler formula, J. Combinatorial Theory
Ser. A 24 (1978), 367–376.
[3] C. D. Godsil: Algebraic Combinatorics, Chapman & Hall, 1993.
[4] C. D. Godsil: Hermite polynomials and a duality relation for matchings polynomials,
Combinatorica 1 (1981), 257–262.
[5] O. J. Heilmann and E. H. Lieb: Theory of monomer-dimer systems, Comm. Math.
Physics 25 (1972), 190–232.
[6] C. Itzykson and J.-B. Zuber: Matrix integration and combinatorics of modular
groups, Comm. Math. Physics 134 (1990), 197–207.
[7] B. Lass: Funktionen zählen, Diplomarbeit, 1997.
[8] L. Lovász: Combinatorial Problems and Exercises, Akadémiai Kiadó, 1993.
[9] G. Viennot: Une théorie combinatoire des polynômes orthogonaux généraux, Notes
de conférences données à l’Université du Québec à Montréal, 1983.

Bodo Lass
Lehrstuhl II für Mathematik
RWTH Aachen
Templergraben 55
D-52062 Aachen
Germany
lass@math2.rwth-aachen.de
and
Institut Girard Desargues (UMR 5028 du CNRS)
Université Claude Bernard – Lyon 1
Bâtiment Doyen Jean Braconnier (101)
43, Boulevard du 11 Novembre 1918
F-69622 Villeurbanne Cedex
France
lass@igd.univ-lyon1.fr