The area generating function
for simple2column polyominoes
with hexagonal cells
Abstract
Columnconvex polygons were first counted by area several decades ago, and the result was found to be a simple, rational, generating function. In this chapter we generalize that result. Let a column polyomino be a polyomino whose columns can have connected components. Then columnconvex polygons are equivalent to 1convex polyominoes. The area generating function of even the simplest generalization, namely to 2column polyominoes, is unlikely to be solvable. We therefore define a class of polyominoes which interpolates between columnconvex polygons and 2column polyominoes. We derive the area generating function of that class, using an extension of an existing algorithm. The growth constant of the new class is greater than the growth constant of columnconvex polyominoes. A rather tight lower bound on the growth constant complements a compelling numerical analysis.
1 Introduction
The enumeration of polyominoes is a topic of great interest to chemists, physicists and combinatorialists alike [27]. In chemical terms, any polyomino (with hexagonal cells and with no internal holes) is a possible benzenoid hydrocarbon. In combinatorics, polyominoes are of interest in their own right because several polyomino models have mathematically appealing exact solutions. Furthermore, they are also relevant to various problems of tilings [25]. Polyominoes are extensively studied, in one form or another, in a wide variety of problems of interest to physicists. In particular, we note their investigation under the name lattice animals, in the study of percolation [29, 33], in the graphical representation of the Ising model, and its extension to the Potts model, and in the study of the properties of branched polymers [28, 43, 48].
They are also a representative of a class of problems that appear to be unsolvable—notably the enumeration, by area or perimeter, of selfavoiding polygons, polyominoes and other classes of graphs that are relevant to the Ising model and to percolation. The principal line of attack on such problems is to simplify them until they are solvable, in the hope that the essential physics is not destroyed in the process. That is the approach taken in this chapter, where the model proposed, while still solvable, is closer to the ultimate problem of full polyomino enumeration than has previously been attained. Further, by use of BousquetMélou’s [4] and Svrtan’s [21] upgraded version of the Temperley methodology [44], we give the solution of one problem previously out of reach due to its complexity. This development may spur further advances in this class of problem.
One frequently cited polyomino model is that of columnconvex polygons ^{3}^{3}3We distinguish between polygons and polyominoes in that the former cannot have internal holes. As a consequence, the perimeter generating function for polygons has a nonzero radius of convergence, whereas for polyominoes the radius of convergence is zero.. There exist two main versions of columnconvex polygons: the first composed of square cells and the second of hexagonal cells. Both versions have a rational area generating function. For the version with square cells, the area generating function was found independently by Pólya [42] in 1938 or 1969 and by Temperley [44] in 1956. That was perhaps the earliest major result in polyomino enumeration. For the version with hexagonal cells, the area generating function was found by Klarner in 1967 [34]. The growth constant of squarecelled columnconvex polygons is , while the growth constant of hexagonalcelled columnconvex polygons is . (By the growth constant we mean the limit , where denotes the number of celled elements in a given set of polyominoes.) In both cases the area generating function is a simple pole, so that
There exist some models which are supersets of columnconvex polygons and are still solvable. Those models are called convex polygons [32], prudent polygons [23], cheesy polyominoes [16], polyominoes with cheesy blocks [17], columnsubconvex polyominoes [19], and simple2column polyominoes [18]. The former two models can be enumerated by perimeter and area, whereas the latter four models have been enumerated only by area.
This chapter is a kind of companion to [18]. Namely, the enumeration of simple2column polyominoes by area was first done in [18]. In the present chapter, we again enumerate simple2column polyominoes by area, but this time the cells are hexagons, whereas in [18] the cells were squares. As mentioned above, we make use of BousquetMélou’s [4] and Svrtan’s [21] upgraded version of Temperley’s methodology [44]. The computations are rather long and intricate. Also, these computations are very similar to those which are done detailedly in [18]. Therefore in this chapter we only give an outline of the proof, though with enough detail that the method may be applied by others to new problems. Incidentally, in [18], simple2column polyominoes are called by their original name (coined by S. Feretić): simplexduplex polyominoes. The name simple2column polyominoes was suggested subsequently (during the preparation of [20]) by Tony Guttmann.
In Section 2, we discuss the chemical relevance of polyominoes with hexagonal cells. In Section 3, we define the model. In Section 4 we give the formula for , a generating function for simple2column polyominoes, in which the variable is conjugate to the area and is conjugate to the number of twocomponent columns of the polyomino. A truncated version of the proof is given in Section 5. In Section 6, we discuss the asymptotic behaviour of , and give a tight lower bound on the growth constant, as well as a very accurate estimate. In Section 7 we conclude, outlining further work prompted by our results.
Note that our solution essentially gives detailed information only about the area generating function. The additional parameter counts columns of a certain type. Thus we cannot give perimeterarea phase diagrams which are relevant to the description of vesicle collapse. Indeed, as is shown in [20], the perimeter generating function has zero radius of convergence (as is the usual case for polyominoes), which precludes such a phase diagram in its usual form.
2 Chemical relevance of polyominoes
with hexagonal cells
Polyominoes with hexagonal cells can be used to model benzenoid hydrocarbons. Benzenoid hydrocarbons are a class of versatile conjugated organic molecules that have constantly been studied by a variety of researchers. This is so because benzenoid hydrocarbons found uses in experimental and theoretical research in chemistry [7, 11, 26], environmental chemistry [22, 40], chemical technology [2], cancer research [9, 24, 31], ecology [6, 37], combinatorial chemistry [35], graph theory [30, 45], theory of aromaticity [39], computational chemistry [3, 36, 46, 47], polymer science [43], etc.
Additionally, polyominoes and benzenoid structures under various disguises have been discussed regularly at the Dubrovnik MATH/CHEM/COMP Symposia. For example, this class of structures has been discussed there as fractal benzenoids [41], as columnconvex animals [10, 14, 15], in algebraic studies of benzenoids [8], in leapfrog transformation [1], maximum matchings and eigenvalues of benzenoid graphs [13], in phototoxicity [12], ringcurrents in benzenoid hydrocarbons [38], etc.
Apparently research on various aspects of this class of structures continues to date and the present chapter is in this direction. However, before we proceed, we wish to point to perhaps the first scientific study on hexagonal structures. This study is due to Croatian polymath Rugjer Bošković (1711–1787). He studied the structure of honeycomb. Honeycomb is made of hexagonal cells. Bošković hypothesized that the hexagonal lattice is not chosen by chance. By using hexagonal lattice, bees economize on building material. Bošković published this study, entitled De apium cellulis, in 1760.
3 Definition of the model
There are three regular tilings of the Euclidean plane, namely the triangular tiling, the square tiling, and the hexagonal tiling. We adopt the convention that every square or hexagonal tile has two horizontal edges. In a regular tiling, a tile is often referred to as a cell. A plane figure is a polyomino if is a union of finitely many cells and the interior of is connected. Observe that, if a union of hexagonal cells is connected, then it possesses a connected interior as well, as a connected union of hexagonal tiles must be connected through shared edges. Topologically, a connected union of square cells may be connected only at a shared vertex. Such unions are forbidden by the definition of polyominoes however.
Let and be two polyominoes. We consider and to be equal if and only if there exists a translation such that .
From now on, we concentrate on the hexagonal tiling. When we write “a polyomino”, we actually mean “a hexagonalcelled polyomino”.
Given a polyomino , it is useful to partition the cells of according to their horizontal projection. Each block of that partition is a column of . Note that a column of a polyomino is not necessarily a connected set. On the other hand, it may happen that every column of a polyomino is a connected set. In this case, the polyomino is a columnconvex polygon. See Figure 1.
By a 2column polyomino, we mean a polyomino in which columns with three or more connected components are not allowed. Thus, each column of a 2column polyomino has either one or two connected components. A simple2column polyomino is such a 2column polyomino in which consecutive twocomponent columns are not allowed. If is a column of a simple2column polyomino, and is a (left or right) neighbour of a twocomponent column, then must be a onecomponent column. See Figure 2.
4 The area generating function for
simple2column polyominoes
with hexagonal cells
If a polyomino is made up of cells, we say that the area of is . Let denote the set of all simple2column polyominoes with hexagonal cells.
In Theorem 2 below, we state a formula for the generating function
Theorem 1
The generating function is given by
(1) 
where
In the above formulae, it will be noticed that (a) some of the numbers have an overline, and (b) no result is given for , , , , , and .
This is both to save space, and to highlight the close similarity between certain quantities. For all the quantities defined above, the overlines may be ignored. To obtain the formula for from the formula for , replace by . To obtain the formula for from the formula for , replace the by and change to . To obtain the formula for from the formula for , and also to obtain the formula for from the formula for , replace each of the ’s by . To obtain the formula for from the formula for , and also to obtain the formula for from the formula for , replace each of the ’s by and change into .
5 Proof of Theorem 1
Let be a simple2column polyomino and let have at least two onecomponent columns. Then we define the lower pivot cell of to be the cell which is the lower right neighbour of the bottom cell of the secondlast (i.e., secondrightmost) among the onecomponent columns of . We also define the upper pivot cell of to be the upper right neighbour of the top cell of the secondlast among the onecomponent columns of .
Let be a simple2column polyomino and let have at least one twocomponent column. Then we define the lower inner pivot cell of to be the upper right neighbour of the top cell of the lower component of the last among the twocomponent columns of . We also define the upper inner pivot cell of to be the lower right neighbour of the bottom cell of the upper component of the last among the twocomponent columns of .
Observe that the lower pivot cell of a simple2column polyomino is not necessarily contained in . The same holds for the upper pivot cell, the lower inner pivot cell and the upper inner pivot cell of .
Let denote the set of those simple2column polyominoes whose last (i.e., rightmost) column is a onecomponent column. It is convenient to first compute a generating function for the set , and thence a generating function for all simple2column polyominoes. So, let
Next, we define three generating functions in two variables, and : Let , and .
The generating functions and are related by
(2) 
Henceforth the notation will be abbreviated as .
In order to obtain a functional equation for the generating function , we are going to suitably partition the set . The blocks of the partition will be denoted , and the parts of coming from these blocks will be denoted , respectively.
First, we define to be the set of those which have no other onecomponent column than the last column. We have .
Let be the set of those which have the following two properties: the secondlast column is a onecomponent column, and the last column contains the lower pivot cell of . We have .
Let be the set of those which have the following two properties: the secondlast column is a onecomponent column, and the last column does not contain the lower pivot cell of . We have .
Thus, is the set of those whose secondlast column is a onecomponent column.
Let be the set of those which have the following three properties: the secondlast column is a twocomponent column and the thirdlast column is (necessarily) a onecomponent column, the lower component of the secondlast column and the thirdlast column have no edge in common, the lower pivot cell of is contained in the upper component of the secondlast column. We have .
Let be the set of having the following three properties: the secondlast column is a twocomponent column and the thirdlast column is a onecomponent column, the lower component of the secondlast column and the thirdlast column have no edge in common, and the lower pivot cell of is contained in the hole of the secondlast column. We have .
The definition of is obtained from the definition of by writing the word “upper” where the definition of says “lower”, and by writing the word “lower” where the definition of says “upper”. We have .
The definition of is obtained when the changes just described are made to the definition of (instead of to the definition of ). We have .
Thus, is the set of those which, in addition to having a twocomponent secondlast column, also have the property that is not a polyomino.
Let be the set of which have the following three properties: the secondlast column is a twocomponent column and the thirdlast column is a onecomponent column, each of the two components of the secondlast column has at least one edge in common with the thirdlast column, and the last column also has at least one edge in common with each of the two components of the secondlast column. We have .
Let be the set of which have the following four properties:

the secondlast column is a twocomponent column and the thirdlast column is a onecomponent column,

each of the two components of the secondlast column has at least one edge in common with the thirdlast column,

the last column has at least one edge in common with the lower component of the secondlast column, but does not have any edges in common with the upper component of the secondlast column,

the last column does not contain the lower inner pivot cell of .
We have .
Let be the set of having the following four properties:

the secondlast column is a twocomponent column and the thirdlast column is a onecomponent column,

each of the two components of the secondlast column has at least one edge in common with the thirdlast column,

the last column has at least one edge in common with the lower component of the secondlast column, but does not have any edges in common with the upper component of the secondlast column,

the last column contains the lower inner pivot cell of .
We have
The definition of is obtained from the definition of by writing the word “upper” where the definition of says “lower”, and by writing the word “lower” where the definition of says “upper”. We have .
The definition of is obtained when the changes just described are made to the definition of (instead of to the definition of ). We have .
Thus, is the set of those which, in addition to having a twocomponent secondlast column, also have the property that is a polyomino. This means that is the set of all whose secondlast column is a twocomponent column.
The sets form a partition of the set . We have , and the expressions for are given above. Putting these things together, we get a functional equation for . It is convenient to write that functional equation as
(3)  
We solved equation (3) by iteration, as is usually done in the upgraded Temperley method. The iteration ended in
(4)  
From equation (4), we got a system of three linear equations in three unknowns: , and . The first equation is just the case of (4). The second equation is obtained by differentiating (4) with respect to and then setting . The third equation is obtained by differentiating (4) twice with respect to , and then setting .
Once the linear system is solved, relation (2) tells us how to obtain the soughtafter generating function . (To solve the linear system, we made use of the computer algebra package Maple.)
5.1 A corollary to Theorem 1
By setting , from Theorem 1 we obtain the well known result, discovered by Klarner [34]:
Corollary 1
The area generating function for columnconvex polygons with hexagonal cells is given by
6 A bit of asymptotic analysis
We write to denote the coefficient of in a power series .
Formula (1) is very complicated, but still allows us to quickly compute Taylor polynomials of any reasonable degree. We actually computed the Taylor polynomial of degree for the function . It turned out that
Then, for , we divided the coefficient by the coefficient . This quotient gradually stabilizes, so that for we have
Simple concatenation arguments, first used by Klarner [34], enable one to prove that . (The letter denotes the growth constant.) In this way, making use of , we find that , which is less than 1% below the best numerical estimate. Unfortunately, finding upper bounds is much more difficult.
Next, for , we divided by . Once again, the quotient stabilizes: for we have . So, there is plenty of evidence that the coefficient (i.e., the number of celled simple2column polyominoes) has the asymptotic behaviour . The dominant singularity of is a simple pole, located at .
For comparison, the area generating function for columnconvex polyominoes is
The number of celled columnconvex polyominoes has the asymptotic behaviour. The dominant singularity of is a simple pole, located at .
As stated in [49], the area generating function for all polyominoes is
and the growth constant of all polyominoes is . Admittedly, the growth constant of all polyominoes is considerably greater than , the growth constant of simple2column polyominoes.
7 Conclusion
We have defined a class of polyominoes that interpolates between columnconvex polyominoes and all polyominoes. The former have been solved, while the latter remain unsolved. The area generating function of columnconvex polyominoes is just . On the contrary, the area generating function of our interpolating model (called simple2column polyominoes) is a cumbersome rational function of series. The area generating functions of columnconvex polyominoes and of simple2column polyominoes both have a simple pole singularity, located at and , respectively. For all polyominoes, the corresponding singularity is at , and the singularity is of the form , rather than a simple pole [49]. We have also given the rigorous lower bound , where is the growth constant of simple2column polyominoes.
In terms of possible extensions of this work, it is probably possible to compute the area generating function of simple2column polyominoes. Here, by a simple2column polyomino we mean a 2column polyomino in which runs of two consecutive twocomponent columns are allowed, but it is forbidden for three consecutive columns to each have two connected components. One reason for doing this is that the growth factor is expected to be greater than that for simple2column polyominoes, and may set the benchmark in this regard. At present the situation is that for columnconvex polygons the growth constant is , for simple2column polyominoes the growth constant is , and for all polyominoes the growth constant is . The polyomino model with a growth constant closest to the actual value for polyominoes is a directed model called multidirected polyominoes [5] with a growth constant of . It would be interesting to compute the area of simple2column polyominoes to see if they had a growth constant closer still to that for polyominoes.
8 Acknowledgement
Svjetlan Feretić wishes to thank Professor Anthony J. Guttmann for providing useful information, with appropriate references included, concerning the role that polyominoes play in theoretical physics.
References
 [1] Babić D., Trinajstić N. and Klein D. J., A note on a variant of the leapfrog transformation of chemical graphs, Croat. Chem. Acta, 1994, 67, 37–44.
 [2] Balaban A. T. and Nenitzescu C. D., Dehydrogenating condensations of aromatics (Scholl and related reactions), pp. 979–1047 in: G. A. Olah (Ed.), FriedelCrafts and Related Reactions, Vol. 2, WileyInterscience, New York, 1964.
 [3] Balasubramanian K., Kaufman J. J., Koski W. S. and Balaban A. T., Graph theoretical characterization and computer generation of certain carcinogenic benzenoid hydrocarbons and identification of bay regions, J. Comput. Chem., 1980, 1, 149–157.
 [4] BousquetMélou M., A method for the enumeration of various classes of columnconvex polygons, Discrete Math., 1996, 154, 1–25.
 [5] BousquetMélou M. and Rechnitzer A., Lattice animals and heaps of dimers, Discrete Math., 2002, 258, 235–274.
 [6] Butler J. D., Air pollution, smoking and lung cancer, Chem. Brit., 1975, 11, 358–363.
 [7] Clar E., Polycyclic Hydrocarbons, Academic Press, London, 1964.
 [8] Cyvin S. J., Cyvin B. N. and Brunvoll J., Allbenzenoid systems: an algebra of invariants, Croat. Chem. Acta, 1993, 66, 101–112.
 [9] Daudel P. and Daudel R., Chemical Carcinogenesis and Molecular Biology, Wiley, New York, 1966.
 [10] Delest M. P. and Dulucq S., Enumeration of directed columnconvex animals with given perimeter and area, Croat. Chem. Acta, 1993, 66, 59–80.
 [11] Dias J. R., Handbook of Polycyclic Hydrocarbons. Part A. Benzenoid Hydrocarbons, Elsevier, Amsterdam, 1987.
 [12] Estrada E. and Patlewicz G., On the usefulness of graphtheoretic descriptors in predicting theoretical parameters. Phototoxicity of polycyclic aromatic hydrocarbons (PAHs), Croat. Chem. Acta, 2004, 77, 203–211.
 [13] Fajtlowicz S., John P. E. and Sachs H., On maximum matchings and eigenvalues of benzenoid graphs, Croat. Chem. Acta, 2005, 78, 195–201.
 [14] Feretić S., A new coding for columnconvex directed animals, Croat. Chem. Acta, 1993, 66, 81–90.
 [15] Feretić S., The columnconvex polyominoes perimeter generating function for everybody, Croat. Chem. Acta, 1996, 69, 741–756.
 [16] Feretić S., Polyominoes with nearly convex columns: A semidirected model, Math. Commun., submitted. arXiv:0910.4573
 [17] Feretić S., Polyominoes with nearly convex columns: A model with semidirected blocks, Math. Commun., to appear. arXiv:0910.4780
 [18] Feretić S., The area generating function for simplexduplex polyominoes, Electron. J. Combin., submitted. arXiv:0910.4794
 [19] Feretić S. and Guttmann A. J., Polyominoes with nearly convex columns: An undirected model, Glas. Mat. Ser. III, to appear. arXiv:0910.4791
 [20] Feretić S. and Guttmann A. J., Two generalizations of columnconvex polygons, J. Phys. A: Math. Theor., in press.
 [21] Feretić S. and Svrtan D., On the number of columnconvex polyominoes with given perimeter and number of columns, in: A. Barlotti, M. Delest, R. Pinzani (Eds.), Proc. Fifth FPSAC Conference, Firenze, 1993, pp. 201–214.
 [22] Futoma D. J., Smith S. R., Smith T. E. and Tanaka J., Polycyclic Aromatic Hydrocarbons in Water Systems, CRC Press, Boca Raton, FL, 1981.
 [23] Garoni T. M., Guttmann A. J., Jensen I. and Dethridge J. C., Prudent walks and polygons, J. Phys. A: Math. Theor., 2009, 42, 095205 (16 pp).
 [24] Gelboin H. V. and Ts’o P. O. P. (Eds.), Polycyclic Hydrocarbons and Cancer, Vols. 1–3, Academic Press, New York, 1978–1981.
 [25] Grünbaum B. and Shephard G. C., Tilings and Patterns, W. H. Freeman and Company, New York, 1987.
 [26] Gutman I. and Cyvin S. J., Introduction to the Theory of Benzenoid Hydrocarbons, SpringerVerlag, Berlin, 1989.
 [27] Guttmann A. J. (Ed.), Polygons, Polyominoes and Polycubes, Lecture Notes in Physics, Vol. 775, SpringerVerlag, Berlin, 2009.
 [28] Hakim V. and Nadal J. P., Exact results for D directed animals on a strip of finite width, J. Phys. A: Math. Gen., 1983, 16, L213–L218.
 [29] Hammond A., Critical exponents in percolation via lattice animals, Electron. Comm. Probab., 2005, 10, 45–59.
 [30] Harary F., Graph Theory, 2nd ed., AddisonWesley, Reading, MA, 1971.
 [31] Harvey R. G., Activated metabolites of carcinogenic hydrocarbons, Acc. Chem. Res., 1981, 14, 218–226.
 [32] James W. R. G., Jensen I. and Guttmann A. J., Exact generating function for 2convex polygons, J. Phys. A: Math. Theor., 2008, 41, 055001 (26 pp).
 [33] Jensen I., Enumerations of lattice animals and trees, J. Statist. Phys., 2001, 102, 865–881.
 [34] Klarner D. A., Cell growth problems, Canad. J. Math., 1967, 19, 851–863.
 [35] Knop J. V., Müller W. R., Szymanski K. and Trinajstić N., Computer Generation of Certain Classes of Molecules, SKTH/Kemija u industriji, Zagreb, 1985.
 [36] Knop J. V., Szymanski K., Jeričević Ž. and Trinajstić N., Computer generation and identification of carcinogenic bay regions in benzenoid hydrocarbons, Int. J. Quantum Chem., 1983, 23, 713–722.
 [37] Lunde G. and Bjørseth A., Polycyclic aromatic hydrocarbons in longrange transported aerosols, Nature, 1977, 268, 518–519.
 [38] Mallion R. B., Topological ringcurrents in condensed benzenoid hydrocarbons, Croat. Chem. Acta, 2008, 81, 227–246.
 [39] Minkin V. I., Glukhovtsev M. N. and Simkin B. Y., Aromaticity and Antiaromaticity: Electronic and Structural Aspects, Wiley, New York, 1994.
 [40] Neff J. M., Polycyclic Aromatic Hydrocarbons in the Aquatic Environment: Sources, Fates, and Biological Effects, Applied Science, London, 1979.
 [41] Plavšić D., Trinajstić N. and Klein D. J., Clar structures in fractal benzenoids, Croat. Chem. Acta, 1992, 65, 279–284; this report has been stimulated by the initial paper on the subject, Klein D. J., Cravey M. J. and Hite G. E., Fractal benzenoids, Polycyclic Aromat. Compd., 1991, 2, 163–182.
 [42] Pólya G., On the number of certain lattice polygons, J. Combin. Theory, 1969, 6, 102–105.
 [43] Privman V. and Švrakić N. M., Directed Models of Polymers, Interfaces, and Clusters: Scaling and FiniteSize Properties, Lecture Notes in Physics, Vol. 338, SpringerVerlag, Berlin, 1989.
 [44] Temperley H. N. V., Combinatorial problems suggested by the statistical mechanics of domains and of rubberlike molecules, Phys. Rev., 1956, 103, 1–16.
 [45] Trinajstić N., Chemical Graph Theory, 2nd revised ed., CRC Press, Boca Raton, FL, 1992.
 [46] Trinajstić N., Jeričević Ž., Knop J. V., Müller W. R. and Szymanski K., Computer generation of isomeric structures, Pure Appl. Chem., 1983, 55, 379–390.
 [47] Trinajstić N., Nikolić S., Knop J. V., Müller W. R. and Szymanski K., Computational Chemical Graph Theory: Characterization, Enumeration, and Generation of Chemical Structures by Computer Methods, Ellis Horwood Ltd., Simon & Schuster, Chichester, 1991.
 [48] Viennot G., Problèmes combinatoires posés par la physique statistique, Astérisque, 1985, 121–122, 225–246.
 [49] Vöge M. and Guttmann A. J., On the number of hexagonal polyominoes, Theoret. Comput. Sci., 2003, 307, 433–453.