INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx

COMBINATORIAL INTERPRETATIONS OF BINOMIAL COEFFICIENT

ANALOGUES RELATED TO LUCAS SEQUENCES
Bruce E. Sagan
1
Department of Mathematics, Michigan State University, East Lansing, MI 48824-1027, USA
[email protected]
(optional)
Carla D. Savage
2
Department of Computer Science, North Carolina State University, Raleigh, NC 27695-8206, USA
[email protected]
(optional)
Received: , Revised: , Accepted: , Published:
Abstract
Let s and t be variables. Dene polynomials {n} in s, t by {0} = 0, {1} = 1, and {n} =
s {n 1}+t {n 2} for n 2. If s, t are integers then the corresponding sequence of integers
is called a Lucas sequence. Dene an analogue of the binomial coecients by
_
n
k
_
=
{n}!
{k}! {n k}!
where {n}! = {1} {2} {n}. It is easy to see that
_
n
k
_
is a polynomial in s and t. The
purpose of this note is to give two combinatorial interpretations for this polynomial in terms
of statistics on integer partitions inside a k (n k) rectangle. When s = t = 1 we obtain
combinatorial interpretations of the bonomial coecients which are simpler than any that
have previously appeared in the literature.
1
Work partially done while a Program Ocer at NSF. The views expressed are not necessarily those of
the NSF.
2
Partially supported by NSA grant H98230-08-1-0072
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 2
1. Introduction
Given variables s, t we dene the corresponding sequence of Lucas polynomials, {n}, by
{0} = 0, {1} = 1, and for n 2:
{n} = s {n 1} + t {n 2} . (1)
When s, t are integers, the corresponding integer sequence is called a Lucas sequence [10, 11,
12]. These sequences have many interesting number-theoretic and combinatorial properties.
Dene the lucanomials for 0 k n by
_
n
k
_
=
{n}!
{k}! {n k}!
(2)
where {n}! = {1} {2} {n}. It is not hard to show that
_
n
k
_
is a polynomial in s and t.
(This follows from Proposition 2.2 below). The purpose of this note is to give two simple
combinatorial interpretations of the lucanomials. They are based on statistics associated
with integer partitions inside a k (n k) rectangle. More specically, we will show
that
_
n
k
_
is the generating function for certain tilings of such and their complements with
dominos and monominos.
Various specializations of the parameters s and t are of interest. When s = t = 1, {n}
becomes the nth Fibonacci number, and
_
n
k
_
is known as a bonomial coecient. Gessel
and Viennot [8] gave an interpretation of the bonomials in terms of nonintersecting lattice
paths and asked for a simpler one. Benjamin and Plott [2, 3] gave another interpretation
in terms of tilings, but it is not as straight-forward as ours. When s = and t = 1 we
get two new interpretations of the -nomial coecients of Loehr and Savage [9]. This case
is of interest because of its connection with the Lecture Hall Partition Theorem introduced
by Bousquet-Melou and (Kimmo) Eriksson [5, 6, 7]. Finally, letting s = q + 1 and t = q
one gets new interpretations for the classical q-binomial coecients. For more information
about these important polynomials, see the text of Andrews [1].
We should mention that it is possible to derive our results from q-binomial coecient
identities using algebraic manipulations, analogous to what is done for -nomial coecients
in Section 3.3 of [9]. However, we wish to demonstrate how these results follow from simple,
combinatorial arguments.
2. Recursions
In this section we will present the recurrence relations we will need for our combinatorial
interpretations. To obtain these results, we will use a tiling model that is often useful when
dealing with Lucas sequences. See the book of Benjamin and Quinn [4] for more details.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 3
L
3
:
r r r r r r r r r
{4} = s
3
+ st + st
Figure 1: The tilings in L
3
and corresponding weights
Suppose we have n squares arranged in a 1 n rectangle. We number the squares 1, . . . , n
from left to right and also number the vertical edges of the squares 0, . . . , n left to right. A
linear tiling, T, is a covering of the rectangle with disjoint dominos (covering two squares)
and monominos (covering one square). Let L
n
be the set of all such T. The tilings in L
3
are
drawn in Figure 1 where a dot in a square represents a monomino while two dots connected
by a horizontal line represent a domino.
Let the weight of a tiling be w(T) = s
m
t
d
where m and d are the number of monominos
and dominos in T, respectively. We will use the same weight for all other types of tilings
considered below. Since the last tile in any tiling must be a monomino or domino, the initial
conditions and recursion (1) immediately give
{n + 1} =

TLn
w(T).
See Figure 1 for an illustration of the case n = 3.
For our second combinatorial interpretation, we will need another sequence of polynomials
closely related to the {n}. Dene n using recursion (1) but with the initial conditions
0 = 2 and 1 = s. If s = t = 1 then n is the nth Lucas number. A combinatorial
interpretation for these polynomials is obtained via another type of tiling. In a circular
tiling of the 1n rectangle, the edges labeled 0 and and n are identied so that it is possible
to have a domino crossing this edge and covering the rst and last squares. Such a domino,
if it exists, will be called the circular domino of the tiling. Let C
n
be the set of circular
tilings of a 1 n rectangle. So L
n
C
n
is the subset of all circular tilings with no circular
domino. For example, C
3
consists of the tilings in L
3
displayed previously together with
r r r
.
Now for n 1 we have
n =

TCn
w(T). (3)
Indeed, to show that the sum satises (1) rst note that we already have a weight-preserving
bijection for the linear tilings involved. And if T C
n
has a circular edge, then removal of
the tile covering square n1 will take care of the remainder. In order to make (3) also hold
for C
0
, we give the empty tiling of the 1 0 box weight w() = 2. Bear in mind that
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 4
considered as an element of L
0
still has w() = 1. Context will always make it clear which
weight we are using.
We start with two recursions for the Lucas polynomials. These are well known for Lucas
sequences; see [4, p. 38, Identity 73] for (4) and [4, p. 46, Identity 94] or [11, p. 201, Equation
49] for (5). Also, the proofs in the integer case generalize to variable s and t without diculty.
But we will provide a demonstration for completeness and to emphasize the simplicity of the
combinatorics involved.
Lemma 2.1. For m 1 and n 0 we have
{m + n} = {n + 1} {m} + t {m1} {n} . (4)
For m, n 0 we have
{m + n} =
n
2
{m} +
m
2
{n} . (5)
Proof.For the rst identity, the left-hand side is the generating function for T L
m+n1
.
The second and rst terms on the right correspond to those tilings which do or do not have a
domino crossing the edge labeled n, respectively. To illustrate, if m = n = 2 then the tilings
in Figure 1 are counted by {m + n}. The rst two tilings do not have a domino crossing the
edge labeled 2 and so are counted by {n + 1} {m}. The edge of the third tiling does cross
that edge and is counted by t {m1} {n}.
Multiply the second equation by 2 and consider two copies of L
m+n1
. In each tiling
in the rst copy distinguish the edge labeled m 1, and do the same for the edge labeled
m in the second copy. The set of tilings in both copies where a domino does not cross the
distinguished edge accounts for the terms corresponding to linear pairs in n {m}+m {n}.
If a domino crosses the distinguished edge m1 in a tiling T, then consider the restriction
T

of T to the rst m squares as a circular tiling by shifting it so that the domino between
squares m1 and m becomes the circular edge. Also let T

be the restriction of T to the last

n1 squares considered as a linear tiling. The pairs (T

, T

) account for the remaining terms

in m {n}. A similar bijection using the tilings with a domino crossing the distinguished
edge m accounts for the rest of the terms in n {m}, completing the proof. 2
We can use the recurrence relations in the previous lemma to produce recursions for the
lucanomials.
Proposition 2.2. For m, n 1 we have
_
m + n
m
_
= {n + 1}
_
m + n 1
m1
_
+ t {m1}
_
m + n 1
n 1
_
=
n
2
_
m + n 1
m1
_
+
m
2
_
m + n 1
n 1
_
.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 5
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
r
Figure 2: A partition contained in 5 4 and a tiling
Proof.Given m, n and any polynomials p and q such that {m + n} = p {m} + q {n},
straightforward algebraic manipulation of the denition of lucanomials yields
_
m + n
m
_
= p
_
m + n 1
m1
_
+ q
_
m + n 1
n 1
_
.
Combining this observation with Lemma 2.1, we are done. 2
3. The combinatorial interpretations
Our combinatorial interpretations of
_
m+n
m
_
will involve integer partitions. A partition is a
weakly decreasing sequence = (
1
,
2
, . . . ,
m
) of nonnegative integers. The
i
, 1 i m,
are called parts and note that we are allowing zero as a part. The Ferrers diagram of ,
also denoted , is an array of m left-justied rows of boxes with
i
boxes in row i. We
say that is contained in an m n rectangle, written m n, if it has m parts and
each part is at most n. In this case, determines another partition

whose parts are the

lengths of the columns of the complement of in mn. The rst diagram in Figure 2 shows
= (3, 2, 2, 0, 0) contained in a 5 4 rectangle with complement

= (5, 4, 2, 2).
A linear tiling of is a covering of its Ferrers diagram with disjoint dominos and monomi-
nos obtained by linearly tiling each
i
. The set of such tilings is denoted L

. Note that if
mn then T L

gives a tiling of each of its rows, while T L

gives a tiling of each

column of its complement. We will also need L

n
which is the set of all tilings in L
n
which
do not begin with a monomino. This is equivalent to beginning with a domino if n 2,
and for n < 2 yields L

0
= {} and L

1
= . We dene L

similarly. The second diagram

in Figure 2 shows a tiling in L

. In a circular tiling of we use circular tilings on

each
i
. So if
i
= 0 then it will get the empty tiling which has weight 2. The notation
C

is self-explanatory. If one views the tiling T in Figure 2 as an element of L

then
it has weight w(T) = s
6
t
7
. But as an element of C

it has w(T) = 4s
6
t
7
. As usual,
context will clarify which weight to use. We are now in a position to state and prove our
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 6
two combinatorial interpretations for the lucanomials. The rst has the nice property that
it is multiplicity free. The second is pleasing because it displays the natural symmetry of
_
m+n
m
_
.
Theorem 3.1. For m, n 0 we have
_
m + n
m
_
=

mn

TL

w(T), (6)
and
2
m+n
_
m + n
m
_
=

mn

TC

w(T). (7)
Proof.We will show that the right-hand side of (6) satises the rst recursion in Proposi-
tion 2.2 as the initial conditions are easy to verify. Given mn there are two cases. If

1
= n then the generating function for tilings of the rst row of is {n + 1}, and
_
m+n1
m1
_
counts the ways to ll the rest of the rectangle. If
1
< n then

1
= m. The generating
function for L

m
is t {m1} and
_
m+n1
n1
_
takes care of the rest.
Proving that both sides of (7) is similar using the fact that if we let f(m, n) = 2
m+n
_
m+n
m
_
then, by the second recursion in Proposition 2.2, we have f(m, n) = nf(m 1, n) +
mf(m, n 1). This completes the proof. 2
We end by noting that it would be interesting to nd analogues for
_
n
k
_
of various
known identities for ordinary binomial coecients. It might then be possible to use the
previous theorem to provide combinatorial proofs. One approach would be to apply algebraic
manipulations to the corresponding results for q-binomial coecients
_
n
k

q
. For example, this
method was used in [9] to give an analog of the Chu-Vandermonde summation for s = ,
t = 1. For general s and t it is easy to see that
_
n
k
_
= y
k(nk)
_
n
k
_
x/y
where
x =
s +

s
2
+ 4t
2
and y =
s

s
2
+ 4t
2
.
Unfortunately, this approach tends to introduce algebraic functions of s and t for which a
combinatorial interpretation is unclear.
References
[1] Andrews, G. E. The theory of partitions. Cambridge Mathematical Library. Cam-
bridge University Press, Cambridge, 1998. Reprint of the 1976 original.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY x (200x), #Axx 7
[2] Benjamin, A. T., and Plott, S. S. A combinatorial approach to Fibonomial
coecients. Fibonacci Quart. 46/47, 1 (2008/09), 79.
[3] Benjamin, A. T., and Plott, S. S. A combinatorial approach to Fibonomial
coecients: errata. Fibonacci Quart. 48, 3 (2010), 276.
[4] Benjamin, A. T., and Quinn, J. J. Proofs that really count, vol. 27 of The Dolciani
Mathematical Expositions. Mathematical Association of America, Washington, DC,
2003. The art of combinatorial proof.
[5] Bousquet-M elou, M., and Eriksson, K. Lecture hall partitions. Ramanujan J.
1, 1 (1997), 101111.
[6] Bousquet-M elou, M., and Eriksson, K. Lecture hall partitions II. Ramanujan
J. 1, 2 (1997), 165185.
[7] Bousquet-M elou, M., and Eriksson, K. A renement of the lecture hall theorem.
J. Combin. Theory. Ser. A 86, 1 (1999), 6384.
[8] Gessel, I., and Viennot, G. Binomial determinants, paths, and hook length for-
mulae. Adv. in Math. 58, 3 (1985), 300321.
[9] Loehr, N., and Savage, C. Generalizing the combinatorics of binomial coecients
via -nomials. Integers, to appear.
[10] Lucas, E. Theorie des Fonctions Numeriques Simplement Periodiques. Amer. J. Math.
1, 2 (1878), 184196.
[11] Lucas, E. Theorie des Fonctions Numeriques Simplement Periodiques. [Continued].
Amer. J. Math. 1, 3 (1878), 197240.
[12] Lucas, E. Theorie des Fonctions Numeriques Simplement Periodiques. Amer. J. Math.
1, 4 (1878), 289321.