Generating Functions for Domino Matchings in the $2\times k$ Game of Memory

TL;DR

This paper develops generating functions for counting domino matchings in a $2\times k$ grid with specified vertical and horizontal pairings, connecting combinatorial configurations with algebraic and diagrammatic representations.

Contribution

It introduces a formal generating function for domino matchings in the $2\times k$ grid, linking combinatorial enumeration with algebraic structures and linear chord diagrams.

Findings

Derived a generating function for aggregate matching polynomials.

Connected domino matchings to linear chord diagrams.

Provided enumeration formulas for specific pairing configurations.

Abstract

When all the elements of the multiset $\{1,1,2,2,3,3,\ldots,k,k\}$ are placed in the cells of a $2\times k$ rectangular array, in how many configurations are exactly $v$ of the pairs directly over top one another, and exactly $h$ directly beside one another --- thus forming $2\times 1$ or $1\times 2$ dominoes? We consider the sum of matching numbers over the graphs obtained by deleting $h$ horizontal and $v$ vertical vertex pairs from the $2\times k$ grid graph in all possible ways, providing a generating function for these aggregate matching polynomials. We use this result to derive a formal generating function enumerating the domino matchings, making connections with linear chord diagrams.

Figures

Equations

$$D_{0,v,h}=1,D_{1,v,h}=\left(\begin{array}{cc}0& 0\\ 1& \end{array}\right),D_{2,v,h}=\left(\begin{array}{ccc}1& 0& 1\\ 0& 0& \\ 1& & \end{array}\right),D_{3,v,h}=\left(\begin{array}{cccc}2& 4& 2& 0\\ 4& 0& 2& \\ 0& 0& & \\ 1& & & \end{array}\right),$$

$$\sum _{j=0}^n(-1{)}^j(2n-2j-1)!!{\rho }_j,$$

$$T(x,y):=\frac{1-xy}{1-y-2xy-x{y}^2+x^3{y}^3}=\sum _{k=0}^{\infty }{T}_k(x)y^k,$$

$$s(x,y):=\frac{T(x,y)}{(1-xy{)}^2},r(x,y):=(1-xy)s(x,y),R(x,y):=yr(x,y),$$

$$S(x,y):=\left(1-2xy-x{y}^2+x^3{y}^3\right)s(x,y).$$

$${\mathcal{T}}_{k,v,h}(x):=\sum _{g\in {\mathcal{G}}_{k,v,h}}g(x).$$

$$\mathcal{T}(x,y,w,z):=\sum _{k,h,v\ge 0}{\mathcal{T}}_{k,v,h}(x)y^{k-v-h}{w}^v{z}^h$$

$$=\frac{1-xy-z}{1-(1+2x)y-z-w(1-xy-z)+(xy+z)(x^2{y}^2-(1-z)z-y(1-2xz))},$$

$$R(x,y)\cdot S(x,y)\cdot r(x,y)+xyT(x,y)\cdot R(x,y)\cdot r(x,y)$$

$$+R(x,y)\cdot r(x,y)\cdot xyT(x,y)+x^2{y}^{2}T(x,y)\cdot T(x,y)\cdot T(x,y)$$

$$=y{r}^{2}S+2x{y}^{2}T{r}^2+x^2{y}^2{T}^3,$$

$$T+\frac{2yz{r}^2}{1-z(S+ys)}.$$

$$(2xyz{)}^{q-1}{\left(T+\frac{2yz{r}^2}{1-z(S+ys)}\right)}^q.$$

$$\mathcal{X}:=\sum _{k,\tilde{h}\ge 0}{y}^{k-\tilde{h}}{z}^{\tilde{h}}\left(\sum _{g\in {\tilde{\mathcal{G}}}_{k,0,\tilde{h}}}g(x)\right)=\frac{T+\frac{2yz{r}^2}{1-z(S+ys)}}{1-2xyz\left(T+\frac{2yz{r}^2}{1-z(S+ys)}\right)}.$$

$$\sum _{j=0}^{\infty }(w+z^2{)}^j{\mathcal{X}}^{j+1}=\frac{T+\frac{2yz{r}^2}{1-z(S+ys)}}{1-(2xyz+z^2+w)\left(T+\frac{2yz{r}^2}{1-z(S+ys)}\right)},$$

$${\rho }_j(k,v,h):=[x^j]{\mathcal{T}}_{k,v,h}(x).$$

$${\rho }_j(k,v,h)=\sum _{g\in {\mathcal{G}}_{k,v,h}}{g}_j.$$

$$D_{k,v,h}=\sum _{j=0}^n(-1{)}^j(2n-2j-1)!!{\rho }_j(k,v,h),$$

$$D(y,w,z):=\sum _{k,h,v\ge 0}{D}_{k,v,h}{y}^k{w}^v{z}^h.$$

$$D(y,w,z)={\int }_0^{\infty }dt{e}^{-t}\frac{1}{2\pi i}{\oint }_{C_{ϵ}}\frac{dx}{x\sqrt{1+2x}}\mathcal{T}\left(\frac{x}{t},\frac{-yt}{x},yw,yz\right),$$

$$[y^n]\mathcal{T}(x,y,w,z)=\sum _{j=0}^n{\Omega }_{j,n}(w,z)x^j=\sum _{j=0}^n\left(\sum _{h,v\ge 0}{\rho }_j(n+h+v,v,h)w^v{z}^h\right)x^j.$$

$$[y^n]{\int }_0^{\infty }dt{e}^{-t}\mathcal{T}(x{t}^{-1},yt,w,z)=\sum _{j=0}^n(n-j)!{\Omega }_{j,n}(w,z)x^j.$$

$$[x^n]\left(\sum _{q=0}^{\infty }(-1{)}^q\frac{(2q-1)!!}{q!}{x}^q\right)\left(\sum _{j=0}^n(n-j)!{\Omega }_{j,n}(w,z)x^j\right)$$

$$=(-1{)}^n\sum _{j=0}^n(-1{)}^j(2n-2j-1)!!{\Omega }_{j,n}(w,z).$$

$$D(y,$$

$${\int }_0^{\infty }dt\frac{e^{-t}}{(1+(1-w)y-(1-z{)}^2{y}^2)\sqrt{1-\frac{2ty(1-(1-z)y)}{(1+(1-z)y)(1+(1-w)y-(1-z{)}^2{y}^2)}}}$$

$$=\sum _{j=0}^{\infty }(2j-1)!!\frac{y^j(1-(1-z)y{)}^j}{(1+(1-z)y{)}^j(1+(1-w)y-(1-z{)}^2{y}^2{)}^{j+1}}.$$

$$E(y,w,z):={\mathcal{L}}^{-1}\left\{y^{-1}D(y^{-1},w,z)\right\}.$$

$$D(y,w,1)=\sum _{j=0}^{\infty }\frac{(2j-1)!!y^j}{(1+(1-w)y{)}^{j+1}}\to E(y,w,1)=\frac{e^{y(w-1)}}{\sqrt{1-2y}},$$

$$D(y,1,z)=\frac{1}{\left(1-(1-z)y\right)}\sum _{j=0}^{\infty }\frac{(2j-1)!!y^j}{{\left(1+(1-z)y\right)}^{2j+1}},$$

$$D(y,z,z)=\sum _{j=0}^{\infty }\frac{(2j-1)!!y^j(1-(1-z)y{)}^j}{(1+(1-z)y{)}^j{\left(1+(1-z)y-(1-z{)}^2{y}^2\right)}^{j+1}}.$$

$${\cal F}_{\ell}(y)\coloneqq\sum_{k\geq 0}\Biggl{(}\sum_{\begin{subarray}{c}h,v\geq 0\\ h+v=k-\ell\end{subarray}}D_{k,v,h}\Biggr{)}\,y^{k}=[z^{-\ell}]\lim_{z\to\infty}D\left(\frac{y}{z},z,z\right).$$

$${\mathcal{F}}_0=\frac{1}{1-y-y^2},{\mathcal{F}}_1=\frac{2y^3}{(1-y)(1-y-y^2{)}^2},{\mathcal{F}}_2=\frac{y^2(1+3y+6y^2+y^3+3y^4)}{(1-y{)}^2(1-y-y^2{)}^3}.$$

$$D_{k,h}=D_{k-1,h}+L_{k,h}-D_{k-1,h-1}.$$

Full text

1 Introduction

The game of memory consists of the placement of a set of distinct pairs of cards in a rectangular array. The present author [1] considered the enumeration of the configurations in which exactly $p$ of the pairs are placed directly beside, or over top of one another, thus forming $1\times 2$ or $2\times 1$ dominoes. In this paper we consider the case of $2\times k$ arrays in more detail. In Figure 1 we show a configuration of the case $k=4$ with $h=1$ horizontal dominoes, and $v=1$ vertical dominoes.

The enumeration of these configurations always carries a factor of $k!$, which counts the orderings of the $k$ distinguishable pairs. It is therefore easier to drop this factor, and thus treat the pairs as indistinguishable. We can then think of the domino enumeration problem in different ways: as a Brauer diagram111Terada [2] and Marsh and Martin [3] have considered Brauer diagrams in the context of combinatorics., as a chord diagram (cf. Krasko and Omelchenko [4]), or unfolded as a linear chord diagram (cf. Cameron and Killpatrick [8]). In the present paper we provide a generating function for the numbers $D_{k,v,h}$ which count, considering the pairs to be indistinguishable, the number of configurations with $h$ horizontal, and $v$ vertical dominoes. It is clear that $D_{k,v,h}=0$ for $h+v>k$; considering the $D_{k,v,h}$ as matrices with $v\geq 0$ indexing rows, and $h\geq 0$ indexing columns, the entries below the anti-diagonal are therefore all zero. The first few values are as follows:

[TABLE]

where we have omitted the aforementioned zero entries. The sum of the numbers on the anti-diagonal are the Fibonacci numbers, which count the domino tilings of the $2\times k$ array222Graham, Knuth, and Patashnik [6, p. 320] give an account.. The sum of all numbers in the matrix333We recall that the double factorial is given by $n!!\coloneqq\prod_{k=0}^{\lceil{n\over 2}\rceil-1}(n-2k)$, and we define $0!!=(-1)!!=1$. is $(2k-1)!!$, which is the number of ways of placing the cards, modulo re-labelling of the pairs.

The strategy we will employ is to consider the matching numbers of the $2\times k$ grid graph, whose vertices represent the $2\times k$ array of cards, and whose edges define the possible domino matchings. In Figure 2 we show this grid graph for the case $k=4$. The present author [1, Section 3, Theorem 4] provided a method for computing the number of [math]-domino configurations (i.e., configurations without any matched pairs) on any analogous graph $G$. Let $\rho_{j}$ be the number of $j$-edge matchings444A $j$-edge matching is defined as a set of $j$ pairwise non-adjacent edges, none of which are loops. on $G$. Then the number of [math]-domino configurations is given by

[TABLE]

where $n$ is the number of pairs. We may therefore compute the $D_{k,v,h}$ by computing the matching numbers for the graphs which arise from removing $v$ vertical, and $h$ horizontal vertex pairs (and their incident edges), from the $2\times k$ grid graph in all possible ways.

2 Preliminaries

The board associated with the grid graph is defined as follows (cf. Riordan [10, p. 163]). In Figure 2, we show the board for the case $k=4$. We color the vertices of the grid graph black and white, in a checkerboard pattern. The columns (rows) of the board represent the black (white) vertices. The cells of the board represent the edges of the graph. The vertical edges correspond to the cells on the central diagonal, while the horizontal edges correspond to the upper and lower diagonals. The rook or matching polynomial $g(x)=\sum g_{j}x^{j}$ encodes the number $g_{j}$ of $j$-edge matchings on the graph and enjoys two important properties. The first comes from partioning the set of all matchings into two sets: those which contain a specific edge, and those which do not. One can then develop the associated board using the property that the rook polynomial of a board $B$ is equal to that of $B$ with a given cell removed (corresponding to the absence of the specific edge in the matching), plus $x$ times the rook polynomial of $B$ with the entire row and column containing that cell removed (corresponding to the presence of the specific edge in the matching). An example is shown in Figure 3. The second property stems from graphs consisting of disconnected components; the disconnectedness implies that the number of matchings of each component are independent of one another. We then have that if a board can be separated into regions whose cells share no common row or column with another region (as in the right hand side of Figure 3), the rook polynomial factorizes into a product of the polynomials for the regions.

Riordan [9, 10, p. 230] (McQuistan and Lichtman [11] give connections to dimer models in Physics) provided the generating function for the rook polynomials of the $2\times k$ grids

[TABLE]

where $T_{k}(x)$ are the rook polynomials. Riordan also provided similar generating functions for several related grids, whose boards are shown in Figure 5,

[TABLE]

For example, the rook polynomial corresponding to the board on the left hand side of Figure 3 is $R_{3}(x)\,r_{3}(x)+x\,T_{3}(x)\,T_{2}(x)$.

3 Generating functions for matching numbers

We now turn our attention to the set ${\cal G}_{k,v,h}$ of graphs which arise when $h$ horizontal, and $v$ vertical vertex pairs are removed, in all possible ways, from the $2\times k$ grid graph. The four graphs belonging to ${\cal G}_{3,0,1}$ are displayed in Figure 4. Each of the graphs $g\in{\cal G}_{k,v,h}$ will have an associated matching polynomial, which with a slight abuse of notation, we will denote $g(x)$. The combinatorial object of interest will be the sum of these polynomials

[TABLE]

Theorem 1**.**

The generating function for the ${\cal T}_{k,v,h}(x)$ is given by555The power of $y$ is taken to be $k-v-h$; this is a useful parameterization for computing the generating function for the $D_{k,v,h}$.

[TABLE]

where the power of the variable $w$ (respectively $z$) corresponds to the number of removed vertical (respectively horizontal) vertex pairs, while the power of the variable $y$ corresponds to the number of remaining vertex pairs after these removals have taken place. The power $j$ of the variable $x$ corresponds to $j$-edge matchings in the resulting graphs.

Proof.

The removal of a horizontal vertex pair from the grid graph corresponds to the deletion of a cell on the lower or upper diagonal of the board, together with its entire row and column. In Figure 6, boards resulting from the deletion of two horizontal vertex pairs are shown. When the boards are developed using the black cells in Figure 6, various shapes arise; these are shown in Figure 5. For example, the configuration shown on the left in Figure 6 corresponds to

[TABLE]

where the $\cdot$ denotes ordinary multiplication, and has been included to aid in the following explanation. The first term corresponds to the removal of both the black cells, the second (third) term to the additional removal of the row and column containing the first (second) black cell, and the last term to the removal of the rows and columns containing both black cells. The multiplication also accounts666The reader is referred to Flajolet and Sedgewick [5, p. 16] for an account of the basic symbolic method. for the ordered sum over all possible positions of the removed horizontal vertex pairs777The exception is when the two removed horizontal vertex pairs are directly over top of one another in the grid graph; these coincident configurations will be accounted for below.. When the row and column containing a black cell is removed, the overall board is shortened, and hence earns a factor of $y$; this is why the expression $xy$ appears for each such removal. It is straightforward to see that the corresponding expression for the board shown on the right in Figure 6 differs from the above only in the first term, which becomes $R(x,y)\cdot s(x,y)\cdot R(x,y)=y^{2}r^{2}s$.

We now focus on generalising to the removal of an arbitrary number $\tilde{h}$ of non-coincident (see Footnote 7) horizontal vertex pairs from the grid graph. In keeping with previous notation, we call this set of graphs $\widetilde{\cal G}_{k,0,\tilde{h}}$. We begin by accounting for the terms which arise from the first step in the development of the associated boards, i.e. when the black cells are removed from the boards. For the time being, we exclude the terms which arise when any rows and columns containing those black cells are also removed. We introduce the variable $z$, the power of which corresponds to the number of horizontal vertex pairs removed from the grid graph. When $\tilde{h}=0$, we have simply $T$. When $\tilde{h}=1$, we have $2zyr^{2}$, where the factor of 2 arises as the removed horizontal vertex pair can correspond to a cell on the upper, or the lower diagonal of the board. As we have seen in the previous paragraph, for $\tilde{h}=2$, we have $2z^{2}(yr^{2}S+y^{2}r^{2}s)=2z^{2}yr^{2}(S+ys)$, where again the factor of 2 accounts for swapping the diagonals which the two cells (corresponding to the removed vertex pairs) are taken from. It follows that the general term for $\tilde{h}>0$ is $2yzr^{2}\left(z(S+ys)\right)^{\tilde{h}-1}$. Performing a sum over $\tilde{h}\geq 0$ we obtain the expression

[TABLE]

We now account for those terms which arise in the development of the boards whenever the row and column containing a black cell are further removed. As was explained above, the calculus of the rook polynomial implies that we gain a factor of $xy$ for each such a removal. In fact, we gain a factor of $2xyz$; the 2 since the black cell could be on the upper, or lower diagonal, and the $z$ to account for the corresponding removal of the vertex pair from the grid graph. Once one row and column have been removed, the board is split into two independent boards. If we remove the columns and rows of $q-1$ different black cells, we will have $q$ independent boards; the corresponding terms are then given by

[TABLE]

We can also interpret this expression from the point of view of the grid graph itself. The factor $(2xyz)^{q-1}$ corresponds to the removal of $q-1$ horizontal vertex pairs, where the coincident edge below or above is chosen in the matching. The factor 2 comes from choosing either the upper or the lower vertex pair to remove. This leaves us to deal with $q$ separate and independent grids. Each of these $q$ grids is enumerated by a factor of Equation (3), where $T$ corresponds to the case where no further horizontal vertex pairs are removed. The fraction, on the other hand, corresponds to the case where at least one further horizontal vertex pair is removed (but the coincident edge is not chosen in a matching).

Taking the sum over $q>0$, we obtain

[TABLE]

It remains to account for the removal of vertical vertex pairs, and coincident horizontal vertex pairs, from the grid graph. The removal of a single pair of coincident horizontal vertex pairs is equivalent to the removal of two neighboring vertical vertex pairs. The effect of either of these removals on the board is again to break it into two independent boards. Thus we find that the removal of coincident horizontal and vertical vertex pairs is accounted for as follows:

[TABLE]

where the coefficient of $w^{v}$ corresponds to the removal of $v$ vertical vertex pairs. Simplifying this expression, we obtain Equation (1). ∎

4 From matching numbers to domino-counting generating functions

We now use the result of Theorem 1 to compute the number of configurations with exactly $h$ horizontal, and $v$ vertical dominoes. Let the $\rho_{j}(k,v,h)$ be defined888We use the notation $[y^{n}]\,f(y)$ to represent the coefficient of $y^{n}$ in the Taylor expansion of $f(y)$. as follows:

[TABLE]

We remind the reader the combinatorial significance of this quantity. The set ${\cal G}_{k,v,h}$ of graphs is obtained from removing $v$ vertical, and $h$ horizontal vertex pairs from the $2\times k$ grid graph in all possible ways. Each graph $g\in{\cal G}_{k,v,h}$ has a number $g_{j}$ of $j$-edge matchings. We then have that

[TABLE]

As mentioned in the Introduction, the number $D_{k,h,v}$ of configurations with exactly $h$ horizontal, and $v$ vertical dominoes is given by

[TABLE]

where $n=k-h-v$. We define the corresponding generating function in the usual way,

[TABLE]

We now translate Equation (5) into an operation on ${\cal T}(x,y,w,z)$.

Theorem 2**.**

The generating function $D(y,w,z)$ may be obtained using the following integral representation:

[TABLE]

where the contour integral with respect to $x$ is taken around a small circle containing the origin.

Proof.

We consider the coefficient of $y^{n}$ in the Taylor expansion of ${\cal T}(x,y,w,z)$, and define $\Omega_{j,n}(w,z)\coloneqq[x^{j}y^{n}]\,{\cal T}(x,y,w,z)$. We have

[TABLE]

Under the integration in $t$, the replacements $x\to xt^{-1}$ and $y\to yt$ dress this result by a factor of $\int_{0}^{\infty}dt\,t^{n-j}e^{-t}=(n-j)!$

[TABLE]

We now consider the coefficient of $x^{n}$ in the expression obtained by multiplying Equation (6) with $(1+2x)^{-1/2}$

[TABLE]

We may, therefore, compute this quantity by further scaling $y\to y/x$ and taking the residue at the origin after an overall multiplication by $x^{-1}$. The factor of $(-1)^{n}$ is absorbed by a final replacement $y\to-y$. The variables $w$ and $z$ are also scaled by $y$, so that the unnatural parameterization discussed in Footnote 5 is rectified. ∎

Corollary 3**.**

The generating function $D(y,w,z)$ is given by

[TABLE]

Proof.

We note from Equation (1) that ${\cal T}\left(\frac{x}{t},\frac{-yt}{x},yw,yz\right)=Ax/(Bx+Ct)$, where $A,B$, and $C$ are functions of $y,w$, and $z$. The contour integration replaces $x\to-Ct/B$ in the factor $(1+2x)^{-1/2}$. The integration over $t$ is interpreted as acting on the Taylor expansion of the resulting expression. ∎

This is not a convergent series999It may be interpreted as the real part (taking $y,w$, and $z\in\mathbb{R}$) of the expansion of the following expression, asymptotic in $y^{-1}$: $\sqrt{\frac{2(1+(1-z)y)}{y(1-(1-z)y)(1+(1-w)y-(1-z)^{2}y^{2})}}F\left(\sqrt{\frac{(1+(1-z)y)(1+(1-w)y-(1-z)^{2}y^{2})}{2y(1-(1-z)y)}}\right)$, where $F$ is Dawson’s integral; Nijimbere [7] gives a modern account of the asymptotic expansion of this function and its relatives., and hence we cannot benefit from an analytic generating function, with which questions about asymptotic behavior could easily be answered. In order to convert this formal generating function into a convergent series, we could take an inverse Laplace transform in $y$ to form an exponential generating function

[TABLE]

Performing this transform is not straightforward, however in the simple case of counting only vertical dominoes, it is feasible, and yields a well-known result A055140,

[TABLE]

which counts the number of matchings of $2k$ people with partners (of either sex) such that exactly $v$ couples are left together. Unfortunately, we have been unable to perform the transform for the case of counting only horizontal dominoes A325754

[TABLE]

but in the next section we derive the exponential generating function by appealing to known results for the $1\times 2k$ problem.

It is also interesting to consider the case where vertical and horizontal dominoes are not distinguished, i.e., $D(y,z,z)$. The present author [1, Section 4] considered this sequence previously. We can now provide a generating function for these numbers A325753

[TABLE]

The present author [1, Section 4.2] also made several conjectures101010A. Howroyd [12] has proven several of these. for generating functions for the so-called $(k-\ell)$-domino configurations, when the number of dominoes is $\ell$ less than the maximum value $k$. These can now be readily calculated as follows:

[TABLE]

The first few such generating functions are

[TABLE]

The function ${\cal F}_{0}$ is the generating function for the Fibonacci numbers, giving the number of domino tilings. The function ${\cal F}_{1}$ is that for the path length of the Fibonacci tree of order $k$, A178523. The sequences corresponding to the cases $\ell=2,\dots,5$ appear as A318267 to A318270, respectively; before the results presented here were available these were obtained by using data produced by a computer program to fix the coefficients in the numerators of the generating functions, based on a guess for the pattern of the denominators. Hence only a small range of values of $\ell$ were achieved.

5 Connections to linear chord diagrams

Kreweras and Poupard [13] solved the problem of counting the $h$-domino configurations on the $1\times 2k$ grid graph (i.e., a path of length $2k$). Cameron and Killpatrick [8] recently revisited this case in the context of linear chord diagrams, and provided a derivation of the corresponding exponential generating function. Let $L_{k,h}$ be the number of $h$-domino configurations on the $1\times 2k$ grid graph. We seek to establish a correspondence with the number $D_{k,h}$ of $h$ horizontal domino configurations on the $2\times k$ grid graph, where we allow any number of vertical dominoes.

Theorem 4**.**

The numbers $L_{k,h}$ and $D_{k,h}$ are related by the following recursion relation:

[TABLE]

Proof.

We begin by unfolding the the vertices of the $2\times k$ grid graph, as shown in Figure 7, to give the vertices of the $1\times 2k$ grid graph. We then note that the central pair of vertices does not correspond to a horizontal domino in the $2\times k$ graph, but rather to a vertical one. The configurations counted by $L_{k,h}$ may be divided into two sets: those with a domino on the central pair and those without. Those configurations with a domino on the central pair are counted by $D_{k-1,h-1}$, as the central pair is effectively deleted, leaving the $2\times(k-1)$ grid graph with $h-1$ horizontal dominoes. In Figure 8, we provide a pictorial interpretation of this relation. The configurations counted by $D_{k,h}$ can be similarly divided, but since the central pair this time represents a vertical domino, those configurations with this vertical domino are equal in number to $D_{k-1,h}$. ∎

Jovovic [14], and Cameron and Killpatrick [8], provide the exponential generating function

[TABLE]

Translating the recursion relation from Theorem 4 into a differential equation for the exponential generating function, we obtain

[TABLE]

This elementary, non-homogeneous, first-order ODE may be solved using an integrating factor.

Corollary 5**.**

The exponential generating function for $D_{k,h}$ is as follows:

[TABLE]

where we have expressed the result in terms of the imaginary error function $\operatorname{Erfi}$.

Proof.

The method of an integrating factor may be used to solve Equation (8). ∎

6 Asymptotic growth and distributions

The asymptotics of the exponential generating functions $E(y,w,1)$ and $E(y,1,z)$, given in Equations (7) and (9), respectively, can be analyzed using the usual machinery of analytic combinatorics. Let ${\cal V}_{k}$ be the random variable defined as the number of vertical dominoes in a random configuration on the $2\times k$ grid. Similarly, let ${\cal H}_{k}$ be the analogous random variable counting horizontal dominoes. The probability distribution functions $V_{k,v}\coloneqq P({\cal V}_{k}=v)$ and $H_{k,h}\coloneqq P({\cal H}_{k}=h)$ are computed as follows:

[TABLE]

Taking derivatives of $E(y,w,1)$ by $w$, we can compute the factorial moments of $V_{k,v}$. We note that

[TABLE]

where $\sim$ indicates asymptotic growth in $k$, and so the $m^{\text{th}}$ factorial moment of $V_{k,v}$ is asymptotically $(1/2)^{m}$, consistent with a Poisson distribution of mean $1/2$. A similar argument can be made for $H_{k,h}$, where the corresponding mean is found to be $1$. Indeed, Kreweras and Poupard [13] (see also Cameron and Killpatrick [8]) proved that the asymptotic factorial moments for the distribution of dominoes on the $1\times 2k$ grid graph are all equal to one; the case of horizontal dominoes on the $2\times k$ grid graph must have the same asymptotic behavior, since the matchings differ only at a single site: the vertices shown in red in Figure 7.

We expect, in the $k\to\infty$ limit, the occurrences of vertical and horizontal dominoes to be independent, and so the distribution $P_{k,p}\coloneqq P({\cal H}_{k}+{\cal V}_{k}=p)$ of dominoes (vertical or horizontal) should also be Poisson with mean $1/2+1=3/2$. We calculate $P_{k,p}$ as follows:

[TABLE]

The present author [1, Section 2, Theorem 1] proved that the mean of this distribution, exact in $k$, is given by

[TABLE]

which gives the expected result of $3/2$ in the $k\to\infty$ limit. We do not have an expression for the exponential generating function for the $D_{k,v,h}$, and so cannot follow the same line of reasoning used above to establish the equality of the remaining factorial moments. A different strategy111111The author thanks Stephan Wagner for providing this proof. may be used, however, to prove that, indeed,

Theorem 6**.**

[TABLE]

Proof.

We begin by noting that $P_{k,p}$ is (up to the denominator $(2k-1)!!$) the coefficient of $y^{k}z^{p}$ in $D(y,z,z)$, i.e.

[TABLE]

We now define the coefficients $a_{j,n}$ as follows:

[TABLE]

We then have that

[TABLE]

Let $P_{k}(z)$ be defined as follows:

[TABLE]

We now proceed to prove

[TABLE]

which is equivalent to the statement of the theorem. We accomplish this by placing bounds on the $a_{k-\ell,\ell}$. Note that

[TABLE]

The following inequalities hold coefficient-by-coefficient:

[TABLE]

It then follows that

[TABLE]

or, equivalently

[TABLE]

In the $k\to\infty$ limit these bounds become equal. In order to establish the form of $P_{k}(z)$ in this limit, we consider

[TABLE]

and note that the bounds in this inequality are equal to $(3/2)^{\ell}/\ell!$ in the limit, thus

[TABLE]

It remains to show that (the limit of) the sum in the definition of $P_{k}(z)$ is convergent. For this we note that

[TABLE]

and, further,

[TABLE]

is a convergent series; by dominated convergence it follows that

[TABLE]

∎

7 Acknowledgements

The author would like to thank the anonymous referee for a very careful reading of the manuscript and for suggesting several improvements. He also thanks Stephan Wagner for providing the proof to Theorem 6.