Abstract
A d-dimensional polycube is a face-connected set of cells on \(\mathbb {Z}^d\). Let \(A_d(n)\) denote the number of d-dimensional polycubes (distinct up to translations) with n cubes, and \(\lambda _d\) denote their growth constant \(\displaystyle \lim _{n \rightarrow \infty } {\scriptstyle \frac{A_d(n+1)}{A_d(n)}}\). We revisit and extend the method for the best known upper bound on \(A_2(n)\). Our contributions include the following.
-
1.
We prove that \(\lambda _2 \le 4.5252\);
-
2.
We prove that \(\lambda _d \le (2d-2)e+o(1)\) for \(d \ge 2\) (already improving significantly the known upper bound on \(\lambda _3\) to 9.8073); and
-
3.
We implement an iterative process in three dimensions, improving further the upper bound on \(\lambda _3\) to 9.3835.
Similar content being viewed by others
Notes
The notation \(\textsf {L} ^{(3)}\) should not be confused with the notation \(\textsf {L} ^k\) used in the previous section.
The source code can be found in https://github.com/mshalah/polyominoes_polycubes_upperbounds .
The weight functions \(W_1(x,y),\dots ,W_{20}(x,y)\) can be found in
https://github.com/mshalah/polyominoes_polycubes_upperbounds .
It is hard to identify the larger root with a small resolution since the function is extremely steep at the vicinity of this root. A Maple program provided that this root is approximately 0.46377209419657440900.
References
The On-line Encyclopedia of Integer Sequences, http://oeis.org
Aleksandrowicz, G., Barequet, G.: Counting \(d\)-dimensional polycubes and nonrectangular planar polyominoes. Int. J. Comput. Geom. Appl. 19, 215–229 (2009)
Aleksandrowicz, G., Barequet, G.: Counting polycubes without the dimensionality curse. Discret. Math. 309, 576–583 (2009)
Barequet, G., Barequet, R.: An improved upper bound on the growth constant of polyominoes. Electronic Notes in Discrete Mathematics 49(Supplement C), 167–172 (2015). The 8th European Conference on Combinatorics, Graph Theory, and Applications
Barequet, G., Moffie, M., Ribó, A., Rote, G.: Counting polyominoes on twisted cylinders. INTEGERS: Electron. J. Combin. Number Theory 6, 1–37 (2006)
Barequet, G., Rote, G., Shalah, M.: \(\lambda > 4\): an improved lower bound on the growth constant of polyominoes. Commun. ACM 59, 88–95 (2016)
Barequet, G., Shalah, M.: Improved upper bounds on the growth constants of polyominoes and polycubes. In: Latin American Symposium on Theoretical Informatics, pp. 532–545. Springer (2021, postponed from 2020 due to COVID-19)
Barequet, R., Barequet, G., Rote, G.: Formulae and growth rates of high-dimensional polycubes. Combinatorica 30, 257–275 (2010)
Conway, A.: Enumerating 2d percolation series by the finite-lattice method: Theory. J. Phys. A Math. Gen. 28, 335–349 (1995)
Eden, M.: A two-dimensional growth process. In: Proceedings of 4th Berkeley Symposium on Mathematical Statistics and Probability, pp. 223–239. Berkeley, CA (1961)
Flajolet, F., Sedgwick, R.: Analytic Combinatorics. Cambridge University Press, Cambridge (2009)
Gaunt, D., Peard, P.: 1/d-expansions for the free energy of weakly embedded site animal models of branched polymers. J. Phys. A Math. Gen. 33(42), 7515 (2000). http://stacks.iop.org/0305-4470/33/i=42/a=304
Gaunt, D., Sykes, M., Ruskin, H.: Percolation processes in \(d\)-dimensions. J. Phys. A Math. Gen. 9, 1899–1911 (1976)
Georgakopoulos, A., Panagiotis, C.: On the exponential growth rates of lattice animals and interfaces, and new bounds on \(p_c\). arXiv preprint arXiv:1908.03426 (v1) (2019)
Guttmann, A. (ed.): Polygons, Polyominoes, and Polycubes, vol. 775. Springer, Netherlands (2009)
Heydenreich, M., Matzke, K.: Expansion for the critical point of site percolation: the first three terms. Combinatorics, Probability and Computing pp. 1–25 (2021). https://doi.org/10.1017/S0963548321000365
Jensen, I.: Counting polyominoes: A parallel implementation for cluster computing. In: Proceedings of the International Conference on Computational Science, part III, vol. 2659, pp. 203–212. Springer (2003)
Klarner, D.: Cell growth problems. Can. J. Math. 19, 851–863 (1967)
Klarner, D., Rivest, R.: A procedure for improving the upper bound for the number of \(n\)-ominoes. Can. J. Math. 25, 585–602 (1973)
Lubensky, T., Isaacson, J.: Statistics of lattice animals and dilute branched polymers. Phys. Rev. A 20, 2130–2146 (1979)
Luther, S., Mertens, S.: Counting lattice animals in high dimensions. J. Stat. Mech. Theory Exp. 9, 546–565 (2011)
Madras, N.: A pattern theorem for lattice clusters. Ann. Comb. 3, 357–384 (1999)
Madras, N., Soteros, C., Whittington, S., Martin, J., Sykes, M., Flesia, S., Gaunt, D.: The free energy of a collapsing branched polymer. J. Phys. A Math. Gen. 23, 5327–5350 (1990)
Mertens, S., Lautenbacher, M.: Counting lattice animals: a parallel attack. J. Stat. Phys. 66, 669–678 (1992)
Rands, B., Welsh, D.: Animals, trees and renewal sequences. IMA J. Appl. Math. 27, 1–17 (1981)
Read, R.: Contributions to the cell growth problem. Can. J. Math. 14, 1–20 (1962)
Redelmeier, D.: Counting polyominoes: Yet another attack. Discret. Math. 36, 191–203 (1981)
Sykes, M., Glen, M.: Percolation processes in two dimensions: I. low-density series expansions. J. Phys. A Math. Gen. 9, 87–95 (1976)
Acknowledgements
The authors would like to thank Andrei Asinowski for many helpful discussions, and especially for contributing the singularity analysis given in Appendix A. M. Shalah acknowledges the support by The Eric and Wendy Schmidt Postdoctoral Grant for Women in Mathematical and Computing Sciences. We would also like to thank the three reviewers for their very helpful comments.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Work on this paper by the first author has been supported in part by ISF Grant 575/15 and by BSF Grant 2017684. A preliminary version of this paper [7] appeared in LATIN’20 (held in January 2021). Part of this research was performed while the second author was affiliated with Stanford University, CA.
Appendices
Appendix A: Singularity Analyses of the Diagonals of the Generating Functions in \(d=2,3\)
In this appendix, we use a standard method for analyzing the singularities of the diagonals of our bivariate functions.
1.1 A.1 \(d=2\)
Our goal is to analyze the diagonal of the bivariate generating function
By using the standard transformation \(G(x,y) := g^{(2)}(x/y,y)/y\), the diagonal coefficients of \(g^{(2)}(\cdot ,\cdot )\) form the coefficient of \(y^{-1}(\cdot )\) in \(G(\cdot ,\cdot )\). Now, by applying residue analysis, and, in particular, the theorem about the residues of a rational function, we find that the diagonal is the function
Hence, we need to determine the asymptotic behavior of \([x^n]q(x)\). To this aim, we use the singularity-analysis method of Flajolet and Sedgewick [11, P. 393, Thm. VI.4]. Our final goal is to find the dominant singularity r of q(x), and to find an asymptotically equivalent function of the form \((1-z)^\alpha \), for which asymptotic expressions are known.
In this case, the dominant singularity is one of the roots of \(1{-}4x{-}4x^2\), namely, \(r=(\sqrt{2}-1)/2 \approx 0.21\). (The other root is \(s=-(\sqrt{2}+1)/2 \approx -1.21\).) We rewrite q(x) as \(-\frac{2x-1}{4x}(1-4x-4x^2)^{-1/2} \ -1/(4x)\). The term \(-1/(4x)\) is negligible for the asymptotics. For \(x=r\), the prefactor \(-\frac{2x-1}{4x}\) yields \(1/\sqrt{2}\).
In order to separate the singular part of \((1-4x-4x^2)^{-1/2}\), and then to use the standard result concerning the asymptotics of \((1-z)^\alpha \), we use the substitution \(z=x/r\) and rewrite
For the part \(\big (4(rz-s)r\big )^{-1/2}\), we can just substitute \(x=r\) (equivalently, \(z=1\)) because this part is not singular at this point. We have that \(4(rz-s)r = 2 \sqrt{2} (\sqrt{2}-1)\), and, thus, we obtain another prefactor \(1/\sqrt{2(2-\sqrt{2})}\).
Finally, for the part \((1-z)^{-1/2}\) which is singular at \(x=r\), we use the corresponding “standard scale” result, \([z^n](1-z)^{-1/2} \sim \frac{1}{\sqrt{\pi n}}\) [ibid, P. 388]. Therefore, \([x^n](1-x/r)^{-1/2} \sim \frac{1}{\sqrt{\pi n}} (1/r)^n\), and we finally have that
concluding that the growth constant of this function is \(2(\sqrt{2}+1) \approx 4.82843\), as we already found in Sect. 4.1.
1.2 A.2 \(d=3\)
The analysis of the 3-dimensional case in much more involved than that of the 2-dimensional case. We describe here only the main ingredients of the analysis without getting into the many technical difficulties. In this case, it is not possible to write the generating function q(x) for the diagonal explicitly, however, we can still determine its asymptotic behavior using symbolic computation. Using Newton polygon in order to determine the number of small roots and resultants, and then finding a polynomial equation satisfied by the sum of (implicitly given) residues, we obtain the equation
In order to determine the dominant singularity, we consider the exceptional set, where the discriminant of this equation vanishes (see [11, Sec. VII.7.1]). This discriminant has two positive roots, approximately 0.1020 and 0.4638.Footnote 6 The growth constant of our sequence is obviously larger than 3. (This can be justified in many ways, for example, by the fact that the growth constant in two dimensions is already greater than 4.) Therefore, we conclude that the dominant singularity is about \(r \approx 0.101965\), and, accordingly, the growth constant is \(1/r \approx 9.8073\), as we already found in Sect. 4.2.
Appendix B: Klarner and Rivest’s Second Part of Condition (\(\varvec{*}\)) is Incorrect
In this appendix, we explain why we believe that Klarner and Rivest did not use in their program the second part of condition (\(\varvec{*}\)) (marked with a red frame), as they state it in their paper [19, p. 600]:
1.1 B.1 Argument 1
Consider the set \(\textsf {C} _4\), that contains all twigs that have exactly four black cells, or fewer black cells and no white cells. Let us enumerate the twigs that have one, two, or three black cells and no open cells. We can easily observe the following.
-
1.
There is only one twig (\(L_1 \in \textsf {L} \), shown in Fig. 4) with one dead cell and no open cells.
-
2.
There are only two twigs (see Fig. 10) with two dead cells and no open cells.
-
3.
There are only six twigs with three dead cells and no open cells. The sequences corresponding to these six twigs are \(L_2 *L_2 *L_1\), \(L_2 *L_4 *L_1\), \(L_3 *L_1 *L_1\), \(L_4 *L_2 *L_1\), \(L_4 *L_4 *L_1\), and \(L_5 *L_1 *L_1\).
We now show that there are 400 twigs with four dead cells and no open cells, which, together with twigs with less dead cells, sum up to a total of 409 twigs in the set \(\textsf {C} _4\). Each twig T with 4 dead cells corresponds to a sequence \((\alpha , \beta , \gamma , \delta )\) of four elements of \(\textsf {L} \) (see Fig. 4), such that \(T= \alpha *\beta *\gamma *\delta \). (Note that T contains its forming twigs in reverse order.) Trivially, in total, there are \(|\textsf {L} |^4=5^4=625\) sequences of four elements of \(\textsf {L} \). However, some of these sequences are invalid, namely, they do not represent valid twigs. For example, the only valid sequence that starts with \(L_1\) is of length 1 since \(L_1\) has no open cells. It is easy to verify that the following sequences are exactly those that are invalid since their prefixes correspond to configurations with less than four dead cells and no open cells: \(S_1=(L_1,\beta ,\gamma ,\delta )\), \(S_2=(L_2,L_1,\gamma ,\delta )\), \(S_3=(L_4,L_1,\gamma ,\delta )\), \(S_4=(L_2,L_4,L_1,\delta )\), \(S_5=(L_4,L_2,L_1,\delta )\), \(S_6=(L_2,L_2,L_1,\delta )\), \(S_7=(L_4,L_4,L_1,\delta )\), \(S_8=(L_3,L_1,L_1,\delta )\), and \(S_9=(L_5,L_1,L_1,\delta )\).
Clearly, we have that \(|S_1|=5^3=125\), \(|S_2|=|S_3|=5^2=25\), and \(|S_4|=|S_5|=|S_6|=|S_7|=|S_8|=|S_9|=5\). There remain exactly twenty invalid sequences, as we proceed to show. The twigs \(L_4\) and \(L_5\) (see Fig. 4) cannot be concatenated to any of the configurations (a–j) (Figs. 11 and 12) since this would violate the first part of condition (\(\varvec{*}\)), that is, a cell of \(L_4\) or \(L_5\) would overlap with an occupied cell of the configuration or a cell marked as forbidden. This results in \(2{\cdot }10=20\) more invalid sequences. Thus, we find that the number of twigs with four dead cells is
Hence, adding items (1–3) above (twigs with less than four dead cells), we obtain that \(|\textsf {C} _4| = 400 + 1 + 2 + 6 = 409\), which is precisely the number provided for \(|\textsf {C} _4|\) by Klarner and Rivest as well (see Table 1). On the other hand, restricting the construction of \(\textsf {C} _4\) further with the second part of condition (\(\varvec{*}\)) would imply that, for example, concatenating \(L_1\) to configuration (d) is invalid, which would decrease the size of \(\textsf {C} _4\). Therefore, we conclude that Klarner and Rivest did not use the second part of condition (\(\varvec{*}\)).
1.2 B.2 Argument 2
Consider P, the pentomino shown in Fig. 13. The spanning tree of P corresponds to the sequence \(L_3 {*} L_2 {*} L_4 {*} L_1 {*} L_1\) (which is equivalent to concatenating \(L_1\) twice to configuration (d) in Fig. 11). In terms of elements of \(\textsf {C} _4\), this is a sequence of length two: its first element T is the twig corresponding to adding \(L_1\) to configuration (d), and its second element is \(L_1\). By definition, the two twigs T and \(L_1\) belong to \(\textsf {C} _4\). However, if we were to use the second part of condition (\(\varvec{*}\)), then T would be discarded as an element of \(\textsf {C} _4\) because a forbidden cell of \(L_1\) would overlap with a cell of T. In such a situation, \(\textsf {C} _4\) would not be a complete set of building blocks for polyominoes, and P would have no corresponding sequence of weight \(x^5 y^5\) of elements of \(\textsf {C} _4\). Therefore, the method would have failed to provide an upper bound on \(\lambda _2\) if the second part of condition (\(\varvec{*}\)) had been used, as some polyominoes (such as P) would be overlooked.
Appendix C: Weight Function in Two Dimensions
The following is the weight function \(W_{21}(x,y)\), from which we computed the upper bound \(\lambda _2 \le 4.5252\).
Appendix D: Weight Function in Three Dimensions
The following is the weight function for \(i=9\), from which we computed the upper bound \(\lambda _3 \le 9.3835\).
Appendix E: Computing the Radius of Convergence
Let f(x,y) be a rational bivariate generating function. Klarner and Rivest [19, §3] showed how to compute the radius of convergence of the diagonal of f(x,y). This requires a change of variable in order to apply the residue theorem. The diagonal function \(f_D(z) = \sum _n l(n,n)z^n\) could then be written as a sum of residues. The following is our Maple implementation of this method. (The function f(x, y) corresponds to the generating function from Sect. 3.1.)
Rights and permissions
About this article
Cite this article
Barequet, G., Shalah, M. Improved Upper Bounds on the Growth Constants of Polyominoes and Polycubes. Algorithmica 84, 3559–3586 (2022). https://doi.org/10.1007/s00453-022-00948-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00453-022-00948-6