You are on page 1of 25

Joshua Bao

1
On The Number of Domino Tilings of the
Rectangular Grid, The Holey Square and Related
Problems
Joshua Bao
1
Research Science Institute
November 20
th
, 1997
1
State College Area High School, State College, PA 16801
Abstract
The study of domino tiling is novel, with many intriguing
questions. The majority of which are still unanswered.
Recently, in separate investigations, Noam Elkies and
others proved that the total number of tilings of both
the Square Grid and the Aztec diamond contain a nice
power of two. In this paper, we will generalize these
results to a wide variety of grids by combining powerful
and new ideas from graph theory, linear algebra, and
geometry.
1 Introduction
In 1992, Noam Elkies and others showed that the total number of domino tilings
of the n
th
order aztec diamond is
#A(n) = 2
n(n+1)/2
. (1)
It is curious that the total number of tilings is always a power of two. Remark-
ably, the number of ways of tiling a square grid also contains a nice exponent
of two, namely the largest exponent of two which divides into the number of
tilings of the 2n2n square grid #N(2n, 2n) is n. More specicly, #N(2n, 2n)
is of the form:
#N(2n, 2n) = 2
n
(2k + 1)
2
. where k is a positive integer. (2)
This conjecture has being proven independently by a number of authors from
the Kasteleyn formula[5] and also combinatorially[3]. In [3], Lior Pachter raised
the question of whether similar patterns may exist in other types of grids, in
particular the general n m rectangular grid. The number of domino tilings of
the rectangular grid was rst studied by Kasteleyn[7]. Applying a considerable
amount of linear algebra, he showed that the total number of tilings, N(n, m),
where mn is even, is precisely:
#N(n, m) = 2
mn/2
n

j=1
m

k=1
4

_
cos
2
j
n + 1
+ cos
2
k
m+ 1
_
. (3)
Unfortunately, this result, a double product of trig functions, reveals little about
the properties of N(n, m). In this paper, we will show that a broad power of
two theorem exist for all of N(n, m) in addition to many other types of grids.
In fact, We believe that properties similar to (1) and (2) are universal to all
tilable grids. What this power of two property represent is a subtle kind of
symmetry that exist within the geometry of the domino tilings. By exploiting
such symmetries, we provide direct combinatorial proofs, which lend themselves
directly to new generalizations and fresh insights.
2 Denitions and Notation
First, we provide a short list of denitions that the reader may not be familiar
with.
Denition 1. We dene the aztec diamond of order n, A(n) as the union of
those lattice squares [a, a + 1] [b, b + 1] R
2
(a, b Z) and lie completely in
the tilted square (x, y) : [x[ +[y[ n + 1.
Denition 2. We dene the holey square, H(n, m) as the 2n 2n square
grid with a hole of size 2m2m removed from the center.
1
Denition 3. Given a function f dened on V (G), an f-factor is a spanning
subgraph G

such that the d


G
(x) = f(x) for each point x.
Denition 4. The adjacency matrix of a graph G with vertices v1,v2, . . . ,v
n
is the matrix A
G
= (a
ij
)
n
i,j=1
, where a
ij
is the number of (v
i
, v
j
) edges.
The problem of Domino tiling is isomorphic to that of nding perfect match-
ings of graphs. For the majority of this paper we shall use the dual graph
and think of edges in the perfect matching as dominoes covering two adjacent
squares. We will, on occasion, use the two descriptions interchangeably for con-
venience. For an arbitrary graph G, we will use the notation #G for the number
of perfect matchings of G. We will also use the notation P#G for the parity of
the number of perfect matchings of some graph G. A particular domino tiling
will be represented using the symbols <, >, , . A horizontal domino is repre-
sented as < [ > and a vertical one as

. An example is <>


where we have
two vertical dominoes and one horizontal.
We will also use an X in a square to represent a square that has been
removed. Thus, the following two grids are identical. XX = Finally,
we use directional arrows to represent the or expression (i.e., two ways of
tiling a single square). For example, # >

= #

+ # <> The
direction of a domino from a xed square is either up, down, left or right. We
shall say that a domino is oriented in the positive (resp. negative) direction from
a given square if it points up or to the right (resp. down or to the left). We
use (j, k) to denote the j
th
square from the left and the k
th
square from the
bottom,

to represent the union of the squares and to represent the removal


of the squares. Lastly, we will use (n) to denote the largest exponent of 2
which divides into n. For example, (56) = 3.
3 The Rectangular Grid
We begin by exploring certain recurrence relations that exist for all rectangular
grids of some xed width.
Theorem 3.1. Let N(n, m) be the number of domino tilings of the nm square
grid, then
N(n, 2) = N(n 1, 2) +N(n 2, 2) (4)
N(n, 3) = 4N(n 2, 3) N(n 4, 3) (5)
N(n, 4) = N(n 1, 4) + 5N(n 2, 4) +N(n 3, 4) N(n 4, 4) (6)
N(n, 5) = 15N(n 2, 5) 32N(n 4, 5) + 15N(n 6, 5) N(n 8, 5) (7)
2
Proof. We prove these by nding decompositions of the grids. (4) is trivial from
the following observation:
# = # <> + #

which implies N(n, 2) = N(n1, 2) +N(n
2, 2). This recurrence is identical to the Fibonacci Sequence, F
n
with initial
conditions F
0
= 1 and F
1
= 1.
To prove (5), let N
n
= N(n, 3) and denote

n
= # XXX
X
(n rows)
Observe that:
# = #

+ # <>

+ # <>

= 2# XXX
X
+ # XXX
XXX
And
# XXX
X
= # XXX
X<>
+ # XXX
X

= # XXX
XXX
+ # XXX
XXX
XXX
X
Thus we have N
n
= 2
n
+ N
n2
and
n
=
n2
+ N
n2
. We rewrite and
reindex the equations as the following.
_
_
_
N
n
= 2
n
+N
n2
2
n2
= N
n2
N
n4
2
n
= 2
n2
+ 2N
n4
(8)
Adding the three equations in (8) yields the desired result of N
n
= 4N
n2

N
n4
. Using similar arguments we prove (6). Let N
n
= N(n, 4) this time and
3
denote

n
= # XX and
n
= # X X (n rows). This time we observe that
# = # ><

= # <>

(9)
+ # <> + # <>

+ #

(10)
# X X = # X<>X + # X X


(11)
# XX = # XX<> + # XX

(12)
The above is equivalent to
_
_
_
N
n
=
n
+
n1
+N
n2
+
n1

n
= N
n1
+
n1

n
= N
n1
+
n2
(Upon Simplication)
_
_
_
N
n
= N
n1
+ 5N
n2
+N
n3
N
n4

n
=
n1
+ 5
n2
+
n3

n4

n
=
n1
+ 5
n2
+
n3

n4
Thus, grids , , and N share the same recurrence with dierent seeds! By
solving the characteristic equations for these recurrences we can obtain explicit
formulae for N(n, 2),N(n, 3),N(n, 4). See Appendix 7.1. N(n, 5) can be deter-
mined using the same method.
These results, while interesting in their own right, are not completely new (see
Appendix 7.8). We now move into uncharted ground. Properties of N(n, m)
(other than the square grid) has never being studied before. We begin by employ
Theorem 3.1 to nd (N(n, m)) for rectangles of some small xed width.
4
Theorem 3.2.
(N(n, 2)) =
_

_
1 if n + 1 = 3(2d + 1),
k + 2 where n + 1 = 3 2
k
(2d + 1)
0 Otherwise.
for positive integers d and k, where k > 2.
We use the following lemma to prove the theorem. Since N(n, 2) = F
n
, we
use the two interchangeably.
Lemma 3.1.
F
n+m
= F
n1
F
m1
+F
n
F
m
Proof. We prove the lemma by the following observation,
# = # + # <>
<>
(13)
= # # (14)
+ # # (15)
Since the above grids can be of any length, we immediately see that
F
n+m
= F
n1
F
m1
+F
n
F
m
, which is our lemma.
Proof of Theorem. Observe that
F
n
1, 1, 2, 3, 1, 0, 1, 1, 2, 3, 1, 0 . . . (mod 4) (16)
for n = 0, 1, 2, 3, 4 . . . (17)
Consequently, (F
n
) = 1 i n + 1 = 3(2c + 1). We now prove the second part
by strong induction.
Base Case: FromF
n
1, 1, 2, 3, 5, 8, 13, 5, 2, 7, 9, 0, 9, 9, 2, 11, 13, 8, 5, 13, 2, 15, 1, 0 . . .
(mod 2
4
) we see that (F
n
) = 3 i n + 1 = 6(2c + 1).
Inductive Step: Assume that the theorem holds for k = m(m > 0), namely
(F
j
) = m + 2 where j = 3 2
m
(2d + 1) 1. We induct on m. Since
2j + 1 = 3 2
m+1
t 1, it suces to show that (F
2j+1
) = m + 3. By
lemma 3.1 we have, F
2j+1
= F
j
F
j+1
+ F
j1
F
j
= F
j
(F
j1
+ F
j+1
). From (17)
we have F
j1
+ F
j+1
2 (mod 4) which implies (F
j1
+ F
j+1
) = 1. Thus,
(F
2j+1
) = (F
j
(F
j1
+F
j+1
))= (F
j
) +(F
j1
+F
j+1
) = m+2 +1 = m+3
and we are done.
Theorem 3.3.
N(n, 3) =
_
2c + 1 if n is even,
0 otherwise.
for some positive integer c.
5
Proof. The theorem follows immediately from N(1, 3) = 1 and
N(n, 3) = 4N(n 1, 3) N(n 2, 3). We see that N(2d, 3) is odd.
Theorem 3.4.
(N(n, 4)) =
_

_
2 if n + 1 = 5(2d + 1),
2k + 4 where n + 1 = 5 2
k
(2d + 1)
0 Otherwise.
In the second case, N(n, 4) = 2
2k+4
(4e +3). for some positive integers d, e and
k, where k > 2.
We use the following two lemmas to prove theorem 3.4.
Lemma 3.2. For the 4 n grid,
N
n+m
= N
n
N
m
+N
n1
N
m1
+ 2
n

m
+
m

n
+
m1

n1
where,
n
=
n1

i=0
N
i

2n
=
n1

i=0
N
2i+1

2n+1
=
n

i=0
N
2i
.
Proof of Lemma. We decompose the (n +m) 4 grid. Note that only an even
number of dominoes may tile across the breaking line and the number of white
squares and black squares in a two coloring must be equal on both sides upon
the division to ensure that the the number of tilings is not 0. Thus we have,
# = # + # <>
<>
<>
<>
+ # <>
<>

+ #

<>
<>
+ # <>

<>
+ # <><>
<>
<>
<><>
and the lemma follows immediately.
Lemma 3.3. For k > 1, if (N
n
) = k, then
(
n
) = (
n+1
) = 0 (mod k/2 + 2).
(
n1
) = (
n
) = (
n+1
) = 0 (mod k/2).
6
See Appendix 7.2 for a proof.
Proof of Theorem 4. We will use the following modular cycles:
_

_
N
n
1, 5, 3, 4, 7, 1, 5, 5, 0, 5, 5, 1, 7, 4, 3, 5, 1, 1, 0, 1 . . .

n
1, 2, 7, 2, 6, 5, 6, 3, 0, 0, 5, 2, 3, 2, 6, 1, 6, 7, 0, 0 . . . (mod 2
3
)

n
1, 1, 6, 4, 2, 3, 3, 0, 0, 0, 5, 5, 6, 4, 2, 7, 7, 0, 0, 0 . . .
where n = 1, 2, 3, 4 . . .
(18)
From (18), we see that (N(n, 4)) = 2 i n + 1 = 5(2d + 1). Again, we prove
the second portion by inductive reasoning.
Base Case: From the modular cycle of N
n
mod 2
7
(See appendix 7.3), we
can infer that N
n
= 6 i n + 1 = 10(2d + 1).
Inductive Step: Assume that the theorem holds for k = m(m 1), namely
N
j
= 2
2m+4
(4r + 3) where j = 5 2
m
(2d + 1) 1. We induct on m. Since
2j +1 = 5 2
m+1
t 1, it suces to show that (N
2j+1
) = 2m+6. By lemma
3.2 we have,
N
2j+1
= N
j
(N
j1
+N
j+1
) + 2
j

j+1
+
j
(
j+1
+
j1
)
Rewriting this expression in terms of yield,
N
2j+1
= (
j+1

j1
)(
j+2

j2
) +2(
j
+
j1
)(
j
+
j+1
) +
j
(
j+1
+
j1
)
Applying lemma 3.3, the middle term is 2
2m+6
e for some even e. Thus it suces
to show,
((
j+1
j 1)(
j+2

j2
) +
j
(
j+1
+
j1
)) = 2m+ 6
Let f, g, h be positive integers. From (18),
j+2

j2
= 2(8h +1) for some h.
Since (
j+1
) = (
j+1
+
j
) = m+4 and
j+1

j1
= 2
m+4
(4r+3), if r is odd
then
j
(
j+1
+
j1
) 6 (mod 2
m+6
), else
j
(
j+1
+
j1
) 2 (mod 2
m+6
)
. Let g = [r/2] + f + h, we have (N
2j+1
) = (2
2m+5
(8g + 6)) = 2m + 6 or
N
2j+1
= 2
2m+6
(4g + 3) as desired.
The acute reader may have already noticed an evolving pattern. We will now
present a broad theorem
for all rectangular grids with even dimensions.
Theorem 3.5.
(#N(2n, 2m)) = k where 2k + 1 = gcd(2n + 1, 2m+ 1).
Proof Outline. Due to space limitations, we give an outline of the proof. First,
we can show that it suces to prove the following.
(#N(p 1, kp 1)) =
p 1
2
.
7
where p is an odd prime and k is an odd integer.
. . . O O O O
O O O O
O O O O
O O O O
. . . (19)
Divide the grid as shown. Since only an even number of tiles may cross each of
the columns marked by O, we can reduce the grid to the following. Let c and c
be even subsets of S = 1, 2, 3, . . . , p 1 and let U = 1, 2, 3, . . . , k then
#N(p 1, kp 1) =

cS
cS, cc=
#(N
_
jc,

j c
iU
((ip 1, j) +(ip + 1,

j))) (20)
#N(p 1, kp 1) =

(p 1, k) ranging over all possible reductions. (21)


where, given a tiling of the squares marked by O, we call the number of ways
of tiling each subgrid as (p 1, k). By considering all such subgrids we can
show that ((p 1, k)) > (p 1)/2 for all k, thus completing the proof.
Finally, based on numerical evidence, we make the following conjecture for the
odd case.
Conjecture 3.1.
(#N(2n + 1, 2m)) = k(3 +j),
where 2k + 1 = gcd(n + 1, 2m+ 1) and j = (n + 1).
From Theorems 3.2 and 3.4, this conjecture is true for at least m=1 and m=2.
With a computer program, we have being able to verify that the conjecture holds
for all rectangles N(a, b) with (a, b)[a 30, b 30.
4 Attacking the Holey Square
In this section, we will generalize (2) in an entirely dierent way. By removing
a m m square grid from the center of an n n gird, we obtain a new class
of grids called the Holey Squares (dened in Section 1). Remarkably, the Holey
Square also has a power of two property, namely the number of its tilings is of
the form,
Theorem 4.1.
#H(n, m) = 2
nm
k
2
.
where k is odd for at least m = 1, 2, . . . , 6. Numerical evidence suggest that k
is odd for all m.
8
First, we utilize a very special case of a theorem in [4]. We provide only a
simplied version of the proof in Appendix 7.4 which is also similar to the one
shown in [2] and [3].
Lemma 4.1. Label the diagonal squares on the 2n 2n holey square grid from
the bottom left to the top right with the labels a
1
, b
1
, a
2
, b
2
, . . . , a
n
, b
n
(see Figure
1). The number of domino tilings of the holey square with dominoes placed at
a
1
, a
2
, . . . , a
n
is dependent only on the orientation of the dominoes and not their
direction.
In [3], Lior Pachter introduced a class of grids as the following:
H
1
= H
2
= H
3
= H
4
=
where H
n
is H
n1
with two new columns of length 2n1 added to the left.
We now introduce a new class of grids, H
n,m
, dened as the following:
H
n,m
= H
n

_
2(nm)<k,j<2nm
(j, k)
For example, if we let

H = H
n,1
, then,

H
1
= null

H
2
=

H
3
=

H
4
=
Lemma 4.2. The number of domino tilings of the holey square is given by
#H(n, m) = 2
nm
(#H
n,m
)
2
.
Proof. Consider a xed orientation for the dominoes covering the a
i
s. We can
assume (using Lemma 4.1) that the directions of the dominoes are all either
down or to the right (call such a conguration reduced). Now notice that the
square grid decomposes naturally into two halves. Figure A illustrates an ex-
9
ample of a reduced conguration.
U U U U U U U U
U U U U U <>
U U U U U
U U UXX
U U XX
U U
U U
<>
Figure A
Notice that the region lled with U is equivalent to H
n
, as is its complement.
Also, notice that in any reduced conguration, every domino covers either two
Us or none at all. We have from Lemma 4.1 that N(n) = 2
nm

C
#C where
C ranges over all reduced congurations. From the remarks above it follows
that

C
#C = (#H
n,m
)
2
. The lemma follows.
Lemma 4.3. #H
n,1
= #

H
n
is odd.
Proof. We prove this by induction, using non-trivial examples. The case of

H
2
is trivial with only one tiling possible. First observe that for even n (in this case
10
n = 6):
# (22)
= # <>
<>
+ #

(23)
+# <>


(24)
The two terms in (23) are equal, so we have
11
P# = P# XX
XX
XX
(25)
Now we can repeat the above argument, working our way down the diagonal
until we reach the n
th
column.
P# XX
XX
XX
= P# XX
XX
XXXX
X
XX
X
(26)
Note that (26) can be decomposed as shown here:
# (27)
= #
XX
+ #
>

(28)
12
Now we show that the second term in (28) is in fact even. #(28) =
4
XXX
XXX
XX
+ #
>

>

(29)
In the general case, continue this reduction until we reach the lower right corner,
spitting out an even number of tilings each time. Finally, we see that the end
result is also even:
#
>

>

= 4 #
>

XXX
XXX
XXXX
(30)
Now it is apparent that, P#(25) =
P#
XX
= P#
XXXX
X
XXX
X
X
(31)
Our last shape is

H
n1
, ipped and rotated by 90

!!
The above argument also work for odd n in the same fashion. See Appendix
7.5.
Therefore, we have,
P#

H
n
= P#

H
n1
. (32)
The lemma follows immediately.
Lemma 4.4. #G is even if and only if there is a non-empty set S V (G)
such that every point is adjacent to an even number of points of S.
13
See Appendix 7.6 for a proof.
Lemma 4.5. #H
n,2
, #H
n,3
, #H
n,4
, #H
n,5
, #H
n,6
are all odd.
See Appendix 7.7 for a proof.
Proof of Theorem. The theorem follows immediately by applying Lemmas 4.1
through 4.5.
5 Other Grids
The nice power of two properties are not limited to the aztec diamond, the
rectangular grid, and the holey square. For instance, the eect of removing a
4 4 grid from the corner of N(n) is also curious in that the exponent of 2
increases with such a removal.
Theorem 5.1. Let K be the grid obtained after the removal of a 44 grid from
one corner of the 2n 2n square. The number of tilings of K (n > 2) satises:
#K(n) = 2
k
(2r + 1)
where,
k =
_

_
n + 2 i 2[n,
n + 3 i 4[(n 3),
n + 4 i 8[(n 9),
n + 5 i 16[(n 5),
n + 6 i 16[(n 13).
What if we reomoved the step diagonal of a holey square?
Conjecture 5.1. Let P be the grid obtained after the removal of any k consec-
utive edges from the step-diagonal of the holey square H(n,m). Then the number
of domino tilings of P is given by
#P = 2
nkm
(2c + 1)
2
. (33)
Recall that the number of domino tilings of the n
th
order aztec diamond is
precisely 2
n(n+1)/2
. Suppose that we were to remove [n/2] squares from each
corner of a 2n 2n square grid along the main diagonals, then we obtain the
aztec diamond of order n after removing all the forced tilings.
X X
X X
X X
X X
= X<> <>X
X X


X X
X<> <>X
14
What if we removed less than [n/2] squares from each corner?
Conjecture 5.2. Let Q be the grid obtained after the removal of t(t [n/2])
consecutive squares from each corner of a 2n 2n square grid along the main
diagonals, then the number of domino tilings of Q satises,
(#Q) = n +t(2t 1). (34)
Note that the power of two ranges from n to n(n+1)/2, which marks a transition
between the square grid
and the aztec diamond! This remarkable conjecture generalizes both (1) and
(2).
6 Final Remarks
The property that some nice pattern describes the largest exponent of two which
divides the total number of domino tilings is likely to be universal. In fact, there
is a rich source of problems to be found in the enumeration of domino tilings.
It seems that the tools needed to resolve many of the problems are just being
developed. We sincerely hope that this paper will spark more interest on this
intriguing topic and attract more mathematicians to this branch of study.
We leave the reader with the following inquiry.
Inquiry. What is the largest power of two which divides the total number of
domino tilings (or the number of perfect matchings) of an arbitrary planar grid,
G, consisting of connected columns and rows?
7 Appendices
7.1 Explicit Formulae for N(n,2),N(n,3),N(n,4)
N(n, 2) = F
n
=
1

5
_
(
1 +

5
2
)
n+1
(
1

5
2
)
n+1
_
N(n, 3) =
_
0 for odd n,

3
6
((

3 + 1)(2 +

3)
n/2
+ (

3 1)(2

3)
n/2
) for even n.
N(n, 4) = ar
n
+bs
n
+cu
n
+dv
n
,
where
15
r =
1
4

29
4

1
2

7
2

29
2
s =
1
4

29
4
+
1
2

7
2

29
2
u =
1
4
+

29
4

1
2

7
2
+

29
2
v =
1
4
+

29
4
+
1
2

7
2
+

29
2
and,
a =
1
116
_
29 11

29
_
3886 638

29
_
b =
1
116
_
29 11

29 +
_
3886 638

29
_
c =
1
116
_
29 + 11

29
_
3886 + 638

29
_
d =
1
116
_
29 + 11

29 +
_
3886 + 638

29
_
.
7.2 Proof of Lemma 3.3
Proof of Lemma 3.3. Let (N
n
) = k. Since
n+1

n
= N
n
, we can write

n+1

n
c (mod 2
k
) where c is a positive integer. Similarly, let
n+1

n1
d (mod 2
k
) and
n+1
e (mod 2
k
) From the recurrence relation de-
rived earlier and (N
n
) = k, we can write
(4
n
+
n+1
+ 2
n1

n1
) = k and (2
n
+
n+1

n1
+
n
) = k
Substitutions for c, d, and e and reindexing yields the relations, 2(d) =
2(e) = k and 2(c) 4 = k as desired.
7.3 The Modular Cycle
The full cycle of N(n, 4) mod 2
7
:
1,5,11,36,95,25,13,69,64,13,5,65,39,100,99,61,105,25,0,41,89,13,115,100,87,
49,85,61,64,21,61,41,47,36,123,53,17,113,0,17,113,85,27,36,15,9,93,117,64,
93,53,81,119,100,83,109,121,9,0,57,73,93,3,100,7,33,37,109,64,101,109,57,
127,36,107,101,33,97,0,33,97,37,43,36,63,121,45,37,64,45,101,97,71,100,67,
29,9,121,0,73,57,45,19,100,55,17,117,29,64,53,29,73,79,36,91,21,49,81,0,49,
81,117,59,36,111,105,125,85,64,125,21,113,23,100,51,77,25,105,0,89,41,125,
35,100,103,1,69,77,64,5,77,89,31,36,75,69,65,65,0,65,65,69,75,36,31,89,77,5,
64,77,69,1,103,100,35,125,41,89,0,105,25,77,51,100,23,113,21,125,64,85,125,
105,111,36,59,117,81,49,0,81,49,21,91,36,79,73,29,53,64,29,117,17,55,100,19,
45,57,73,0,121,9,29,67,100,71,97,101,45,64,37,45,121,63,36,43,37,97,33,0,97,
33,101,107,36,127,57,109,101,64,109,37,33,7,100,3,93,73,57,0,9,121,109,83,
100,119,81,53,93,64,117,93,9,15,36,27,85,113,17,0,113,17,53,123,36,47,41,61,
21,64,61,85,49,87,100,115,13,89,41,0,25,105,61,99,100,39,65,5,13,64,69,13,
25,95,36,11,5,1,1,0,1
16
7.4 Proof of Lemma 4.1
Proof of Lemma 4.1. Let M be any domino tiling of the holey square grid. Let
M

be the tiling obtained by reecting M across the diagonal and dene D =


M M

(D is allowed to consist of multiple dominoes). Notice that D is a


2-factor of the holey square grid and is therefore a disjoint union of even-length
cycles. Furthermore, since D is symmetric across across the diagonal, any cycle
maps to another cycle under the reection.
Now dene C

i
to be the cycle containing a
i
. C

i
can have at most one other
vertex on the diagonal because every vertex in C

i
has degree 2. Furthermore,
such a vertex must be of the type b
j
, for otherwise the number of vertices
enclosed by C is odd (contradicting the fact that D is a disjoint union of even
length cycles). It follows that all the cycles C

i
are distinct.
Finally, let C
i
= C

i
M be the alternating cycles in M obtained from C

i
.
By the above arguments, the alternating cycles C
i
are disjoint. Thus, there
is a bijection between any two sets of tilings with xed dominoes of the same
orientation on the a
i
s. We simply select all the dominoes on the a
i
s that have
switched direction and rotate the appropriate alternating cycles.
b
3
a
3
b
2
XX
XX
a
2
b
1
a
1
Figure 1
7.5 Proof of odd case of Lemma 4.3
Proof. We begin by using the same reduction as in the even case, namely:
P# = P# XX
XX
XXXX
X
XXX
X
X
(35)
17
It follows that (35) is equal to
# (36)
= #
XX
+ #
>

(37)
The later part of #(37) =
2 #
XX
XX
+
>
>

(38)
We can now proceed exactly like the even case:
The later part of #38 =
4 #
>
XXX
XXX
XX
+
>
>

>

(39)
18
It follows that
>
>

>

= 4 #
>
>

XXX
XXX
XXXX
(40)
Thus, P#35 =
P#
XX
= P#
XXX
XXX
X
XXX
X
X
(41)
(42)
The last shape is again

H
n1
!
7.6 proof of Lemma 4.4
Proof. Let A be the adjacency matrix of G. Let us consider [A[. Every expan-
sion term which is not zero and symmetric with respect to the main diagonal
corresponds to a domino tiling and vice versa. The other expansion terms cor-
respond to each other in pairs under reection by the main diagonal. Thus the
number of domino tilings has the same parity as [A[. So it is even if and only
if [A[ disapears over GF(2). THis occurs if and only if rows of A are linearly
dependent over GF(2), or there are elements g
1
,g
2
, . . . ,g
n
of GF(2) that not
all of them are 0 and multiplying the rows of A in order by g
i
, the sum of rows
will be 0. Let S be the set of grids for which the corresponding g
i
is 1, then S
has the property that each grid is adjacent to an even number of elements of it.
Similarly, the converse is also true.
19
7.7 Proof of Lemma 4.5
Using the same technique as in the proof of Lemma 4.3, we can show that
P#
XX
XX
XXXX
= P# XX
XX
XXXX
X
XXX
XX
XXXX
<>
XX
X
X
(43)
# XX
XX
XXXX
X
XXX
XX
XXXX
<>
XX
X
X
= # XX
XX
XXXX
X
XXX
<>XX
<>XXXX
<>XXX
XXX
X
X
(44)
+# XX
XX
XXXX
X
XXX
XX
XXXX
XXX
XXX
X
X
+ # XX
XX
XXXX
X
XXX
<>XX
<>XXXX
XXX
XXX
X
X
(45)
+# XX
XX
XXXX
X
XXX
XX
XXXX
<>XXX
XXX
X
X
(46)
20
Notice the last shape is equivalent to H
n1,2
. We can show that the three re-
maining grids from the sum are
all even by Lemma 4.4. Thus, we have P#H
n,2
= P#H
n1,2
. Similar ar-
guments also work for m=3,4,5,6.
7.8 Special Notes
By a derivation from Kasteleyns formula, Richard Stanley has proven that a
recursive formula exists for all rectangular grids. Moreover, Stanley has shown
that 2
(n+1)/2
is the least possible degree of a linear recurrence satised by
N(n, m) for some xed m. In fact, Hock and McQuistan have computed these
reccurences up to n = 10 by employing the transfer matrix method on a com-
puter. However, the short combinatorial proofs shown in Theorem 3.1 are new
except for (4) which is trivial and well known.
Also note that many proofs shown in this paper are abridged and occasion-
ally omitted to save space. The complete proofs will be made available on the
internet at web.mit.edu/jbao/www.
7.9 Acknowledgements
I would like to give my sincere thanks to Lior Pachter, my mentor and friend,
for his guidance; Professor Hartley Rogers for organizing the mentorship; Dr.
Rickert, my tutor, for his immense help with LaTeX; And RSI for making all of
this possible.
References
[1] G.H. Hardy, A Course of Pure Mathematics, Cambridge University Press
(1952).
[2] L. Pachter and P. Kim, Forcing Matchings on Square Grids, preprint (1996).
[3] L. Pachter, On the Number of Domino Tilings of the Square Grid, preprint
(1997).
[4] M. Ciucu, Enumeration of perfect matchings in graphs with reective sym-
metry, J. Combin. Theory Ser. A 77 (1997), no. 1, p 67-97.
[5] P. John, H. Sachs and H. Zernitz, Problem 5. Domino Covers in
Square Chessboards, Zastosowania Matematyki (Applicationes Mathemati-
cae) XIX, 3-4 (1987), p 635-641.
[6] W. Jockusch, Perfect matchings and perfect squares, J. Comb. Theory Ser.
A 67 (1994), p 100-115.
21
[7] P.W. Kasteleyn, The statistics of dimers on a lattice, I: The number of dimer
arrangements on a quadratic lattice, Physica 27 (1961), p 1209-1225.
[8] L. Lovasz, Combinatorial Problems and Exercises, 4(29), North-Holland
Publishing Company (1979).
[9] J. Propp, Twenty Open Problems in Enumeration of Matchings, preprint,
(1996).
[10] Richard P. Stanley, On Dimer Coverings of Rectangles of Fixed Width,
Discrete Math. (1984), no. 3, p 81-87.
[11] J.L. Hock, R.B. McQuistan, A Note On the Occupational Degeneracy For
Dimers on a Saturated Two-Dimensional Lattice Space, Discrete Apllied
Math. bf 8 (1984), 101-104, North-Hollend.
[12] N. Elkies, G. Kuperburg, M. Larsen, J. Propp: Alternating sign matrices
and domino tilings. J. Algebraic Comb. 1,111-132, 219-234 (1992).
22

You might also like