Abstract
We obtain an exact formula for the cubic partition function and prove a conjecture by Banerjee, Paule, Radu and Zeng.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and statement of results
Let n be a non-negative integer. A non-increasing finite sequence of positive integers that sums to n is called a partition of n. We denote by p(n) the number of partitions of n, which can be defined by the coefficients of the following q-series [1]
The function p(n) is one of the most well-studied functions throughout number theory and satisfies numerous remarkable identities. One example is the following identity discovered by Ramanujan [1, 10]:
An immediate consequence is one of Ramanujan’s celebrated congruences which asserts that for any non-negative integer n,
Similar to (1.1), we define the cubic partition function a(n) by
Chan [6] proved that the function a(n) satisfies an identity similar to (1.1) above
see [6]. His result is closely related to Ramanujan’s cubic continued fraction, see [7]. For an introduction to Ramanujan’s cubic continued fraction, see [2, 4]. Furthermore, from the generating function (1.2), it is obvious that a(n) is the number of partition pairs \((\lambda , \mu )\) such that \(|\lambda | + |\mu | = n\) and \(\mu \) consists of only even numbers. These connections gave rise to the term “cubic partitions”.
The first 30 values of a(n) are as follows: 1, 1, 3, 4, 9, 12, 23, 31, 54, 73, 118, 159, 246, 329, 489, 651, 940, 1242, 1751, 2298, 3177, 4142, 5630, 7293, 9776, 12584, 16659, 21320, 27922, 35532, 46092.
There is another combinatorial interpretation of the cubic partition function. Suppose we have a real polynomial equation of degree n. Then, a(n) counts the number of possibilities for the roots to be real and unequal, real and equal (in various combinations), or simple or multiple complex conjugates. For example, for a real cubic polynomial, one can have 3 real distinct roots, 3 distinct real roots, 3 real roots and two of them are equal, 3 equal real roots, and3 real roots and two are equal, 3 real roots, and 1 real root with a complex conjugate pair of 2 roots. Hence, \(a(3) = 4\).
Another natural question about p(n) is how fast it grows. In 1918, Hardy and Ramanujan [8] showed the following with their celebrated Circle Method
Hardy and Ramanujan made extensive use of the fact that the generating function (1.1) is almost a modular form. More precisely,
where \(\eta (z)\) is the Dedekind eta function defined on the upper half-plane \(\mathbb {H} = \{\tau \in \mathbb {C}\mid \text {Im}\tau > 0\}\) by
The function \(\eta (\tau )\) is a modular form of weight \(\frac{1}{2}\) and satisfies the transformation law
For the last definition, we used that one can always assume \(c>0\) or (\(c=0\) and \(d=1).\) Here, s(h, k) is the Dedekind sum
A proof can be found in [3]. Later, Rademacher [9] perfected the circle method and obtained the exact formula
where \(A_k(n)\) is the Kloosterman sum given by
Note that the series (1.6) converges very fast and that the first term recovers the result by Hardy and Ramanujan (1.3). Here, we want to find an exact formula of Rademacher type for the cubic partition function a(n), by using a result of Zuckerman [12] extending Rademacher’s work and it can be seen as the apotheosis of the classical Circle Method. Using the Circle Method, Zuckerman computed exact formulas for Fourier coefficients of weakly holomorphic modular forms of arbitrary non-positive weight on subgroups of \({\text {SL}}_2(\mathbb {Z})\) with finite index in terms of the cusps of the underlying subgroup and the negative coefficients of the form at each cusp. We state the relevant results of Zuckerman in Sect. 2 and prove the exact formula for a(n) in Sect. 3. Similar to the case above, we will obtain as an immediate consequence the asymptotic main term for
Finally, we are going to prove a conjecture by Banerjee, Paule, Radu and Zeng [5] which predicts the following asymptotic formula for \(\log a(n)\)
2 Zuckerman’s result
The following two sections follow Zuckerman’s work [12] closely. Most of the results are completely taken over and translated into more modern language. His method, due to its generality, involves numerous technical parameters which appear in the Circle Method and do not have obvious meaning without knowing the context they arise in. The interested reader should thus consult [12] for a more detailed account.
Let \(\Gamma \) be a subgroup of \({\text {SL}}_2(\mathbb {Z})\) of finite index and let F be a weakly holomorphic modular form of weight \(k=-r, r>0.\) Thus, \(F(\tau )\) satisfies a transformation equation of the form
where \(\varepsilon = \varepsilon (a,b,c,d)\) lies on the unit circle and depends only on the transformation. If \(c\ne 0,\) then c is taken to be positive and we choose the branch of the argument such that
Since \(\Gamma \) is of finite index in the modular group, we can choose a complete finite system of representatives for the inequivalent cusps of \(\Gamma ,\) which we will denote by \(P_1, \ldots P_s,\) where
In order to treat all our cusps symmetrically, we will have to assume that the point at infinity does not belong to our chosen set of representatives. This can always be achieved by considering a \(\Gamma \)-equivalent rational point instead. We will obtain a set of Fourier expansions \(f_1, \ldots , f_s\) corresponding to our set of cusps in the variable \((\tau - P_g)^{-1}.\) Consider now any transformation \(\gamma \in {\text {SL}}_2(\mathbb {Z}),\) which must not necessarily belong to our subgroup \(\Gamma .\) We then write
where \(F^{*}(\tau )\) is now a usual Fourier expansion in \(\tau ,\) obtained from one of the expansions
where \(c_g>0\) will be specified below. Note that by choosing the identity matrix in (2.1) we get a usual Fourier expansion \(F(\tau ),\) which we want later. Corresponding to each of the cusps \(P_g\), we can find a transformation in \(\Gamma \)
which can be written in the form
We can recover the transformation in the usual form as
We then define \(\alpha _g\) by the equation \(\varepsilon e^{\frac{\pi i}{2}} = e^{-2\pi i \alpha _g},\) where \(0\le \alpha _g < 1\) and \(\varepsilon \) is taken with respect to \(\tau '.\)
We now describe how to find \(F^{*}(\tau )\) in terms of the Fourier expansions \(f_g.\) We therefore choose any transformation \(\gamma \in {\text {SL}}_2(\mathbb {Z}).\) Then, \(\frac{a}{c}\) is a rational point if \(c\ne 0\) or the point at infinity if \(c=0.\) In any case, \(\frac{a}{c}\) is \(\Gamma \)-equivalent to exactly one of our cusps \(P_g.\) We can therefore find a transformation
which takes \(P_g\) into \(\frac{a}{c}.\) We can then express \(F^{*}(\tau )\) in terms of the Fourier expansion \(f_g\) as follows
where \(\kappa = \pm 1\) is taken as the solution to the following equations
Furthermore, one obtains
This gives the Fourier expansion in spirit of (2.1) at any rational point \(\frac{a}{c},\) including the point at infinity. When executing the circle method with these transformation equations there are numerous parameters that appear. First of all consider as usual the Farey fractions \(\frac{h}{k}, \gcd (h,k) = 1, k>0, h\ge 0.\) Then, consider the point \(P=P_g - \frac{k}{hc_g}\) which is either a rational point or the point at infinity. In any case, it is \(\Gamma \)-equivalent to one of our cusps \(P_\beta ,\) where \(\beta = \beta (h,k,g).\) There exists a transformation in \(\Gamma \) which takes this point into \(P_\beta \)
Associated to this transformation is the parameter \(\sigma _{h,k}^{(g)}\) implicitly defined as the solution of the equations
It satisfies \(-c_g \le \sigma _{h,k}^{(g)} \le c_g, \sigma _{h,k}^{(g)} \ne 0.\) The remaining technical parameters necessary to state Zuckerman’s result are
Finally, Zuckerman’s result reads now as follows.
Theorem 2.1
(Zuckerman’s exact formula) Assume the notation and hypotheses above and denote by \(I_r\) the Bessel function of order r. If \(n + \alpha _g > 0,\) then we have
3 Exact formula for the cubic partition function
We recall that the cubic partition function was defined by the identity
Using the relation
we can view \(\eta (2\tau )\) in light of (1.5) as a modular form of weight \(\frac{1}{2}\) on the subgroup \(\Gamma _0(2)\) of \({\text {SL}}_2(\mathbb {Z})\) with multiplier
Hence, we find that
is a weakly holomorphic modular form of weight \(-1\) with multiplier and thus we can apply Zuckerman’s result to find the Fourier coefficients of \(F(\tau )\) and therefore a formula of a(n). The group \(\Gamma _0(2)\) has two inequivalent cusps 0 and \(\frac{1}{2}.\) We consider the transformations
which we write as
Thus, we find that \(c_1 = 2\) and \(c_2 = 4.\) Moreover, we can now determine \(\alpha _1\) and \(\alpha _2.\) Since \(F(\tau )\) has the multiplier
a short calculation shows that \(\alpha _1 = \alpha _2 = \frac{7}{8}.\) Since the point at infinity is \(\Gamma _0(2)\)-equivalent to \(\frac{1}{2},\) we obtain that \(F^{*}(\tau )\) is given in terms of the Fourier expansion \(f_2.\) To find the exact expression for \(F^{*}(\tau )\), we follow Zuckerman’s method by choosing a transformation that maps \(\frac{1}{2}\) to the point at infinity. Obviously,
Hence, \(\kappa = -1\) and
Furthermore, we consider the point \(\frac{a}{c} = \frac{0}{1},\) which corresponds to the transformation \(\tau \mapsto -\frac{1}{\tau }.\) In this case, the obvious equivalent point is 0 itself and the identity is the corresponding transformation. Thus, in this case
We now determine \(a_{-1}^{(g)}\) and \(\mu _{-2}^{(g)}.\) In the first case, we have on the one hand
with \(\varepsilon ^{*} = e^{\frac{7\pi i}{8}}.\) On the other hand, we have by definition that
Comparing coefficients we see that \(\mu _{-1}^{(2)} = 1 \) and \(a_{-1}^{(2)} = -2e^{\frac{\pi i}{8}}.\) Furthermore, (3.1) implies that the Fourier coefficients are exactly a(n) shifted down by one. Thus, we can recover a(n) through the relation \(n=m+1.\) A similar calculation shows that \(\mu _{-1}^{(1)} = 1\) and \(a_{-1}^{(2)} = \frac{e^{-\frac{\pi i}{2}}}{\sqrt{2}}.\)
We now continue to determine the remaining parameters only for \(g=2\) as the Fourier expansion at \(\frac{1}{2}\) gives us the exact formula. For \(\beta (h,k,2)\), it is required to find transformations that send \(P_g-\frac{k}{c_gh}\) into \(P_\beta .\) Let \(h', h'', h'''\) be any solutions of
Then, the following table gives the remaining parameters needed to apply Theorem 2.1:
k | a | b | c | d | \(\beta \) | \(\sigma _{h,k}^{(2)}\) | \(\delta _{h,k}^{(2)}\) | \(G_{h,k}^{(2)}\) | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
\( \equiv 1 \,\,(\textrm{mod}\,\,{2})\) | \(h'''\) | \(\frac{(k-2h)h'''-1}{4h}\) | \(2h''' + 4h\) | \(\frac{(k-2h)h'''-1}{2h}- 2h + k\) | 2 | \(-1\) | 1 | \(\left( \frac{h'''k-1}{2kh} + 1\right) \pi \) | |||||
\( \equiv 0 \,\,(\textrm{mod}\,\,{4})\) | \(h'\) | \(\frac{\left( \frac{k}{2}-h\right) h'- 1}{2h}\) | \(2h' + 2h\) | \(\frac{\left( \frac{k}{2}-h\right) h' - 1}{h} + \frac{k}{2} - h \) | 2 | \(-2\) | 1 | \(\left( \frac{h'k-2}{kh} + 1\right) \pi \) | |||||
\( \equiv 2 \,\,(\textrm{mod}\,\,{4})\) | h | \(\frac{k-2h}{4}\) | \(4h''\) | \(\frac{kh'' + 1}{h} - 2h''\) | 1 | 4 | \(-1\) | \(-\left( \frac{4h''k+4}{kh}\right) \pi \) |
Since the series in Theorem 2.1 converges absolutely, the terms can be rearranged and putting everything together, keeping in mind the shift \(n = m+1,\) it follows that
where
Changing the order of summation one last time, we arrive at the following
Theorem 3.1
For \(n \ge 1\), we have the exact formula
where \(l' = l\) if \(l \equiv 1 \,\,(\textrm{mod}\,\,{2}),\) \(l' = l\) if \(l \equiv 0 \,\,(\textrm{mod}\,\,{4}),\) \(l' = \frac{l}{2}\) if \(l \equiv 2 \,\,(\textrm{mod}\,\,{4}).\)
It is easy to see that the series converges rapidly and that the term \(l=1\) dominates the rest of the sum. Thus, we may conclude that
For the Bessel function of order \(\nu \in \mathbb {N}\), the following asymptotics for \(z \in \mathbb {R}\) are known,
see [11], where \(a_k(\nu )\) is defined as
Thus, we find that the first-order approximation is
We now obtain the following asymptotics.
Corollary 3.2
We have as \(n \rightarrow \infty \)
As a final consequence, we prove a conjecture by Banerjee, Paule, Radu and Zeng [5].
Corollary 3.3
We have
Proof
We need to shift our approximation from \(n-\frac{1}{8}\) to n. Thus, write
Together with the second-order approximation for the Bessel function
we obtain
Taking logarithms, we find as conjectured
\(\square \)
Remark
We can refine our proof and obtain stronger asymptotics of the form
for some real \(c_i\) by taking higher order expansions of the involved functions.
References
Andrews, G.: The Theory of Partitions, Encycl. Math. and Its Appl., vol. 2, G.-C. Rota ed., Addison-Wesley, Reading (1976) (Reissued: Cambridge University Press, Cambridge (1998))
Andrews, G., Berndt, B.: Ramanujan’s Lost Notebook. Part I. SpringerVerlag, New York (2005)
Apostol, T.: Modular Functions and Dirichlet Series in Number Theory. Springer-Verlag, Berlin (1976)
Berndt, B.: Ramanujan’s Notebooks. Part III. Springer-Verlag, New York (1991)
Banerjee, K., Paule, P., Radu, C., Zeng, W.: New inequalities for \(p(n)\) and \(\log p(n)\). Ramanujan J. 61, 1295–1338 (2022)
Chan, H.: Ramanujan’s cubic continued fraction and an analog of his “most beautiful identity". Int. J. Number Theory 6(03), 673–680 (2010)
Chen, W., Lin, B.: Congruences for the Number of Cubic Partitions Derived from Modular Forms. arxiv:0910.1263
Hardy, G., Ramanujan, S.: Asymptotic formulae in combinatory analysis. Proc. Lond. Math. Soc. Ser. 2(17), 75–115 (1918)
Rademacher, H.: A convergent series for the partition function \(p(n)\). PNAS 23(2), 78–84 (1937)
Ramanujan, S.: Collected Papers, Cambridge University Press, Cambridge, 1927; reprinted by Chelsea, New York, 1962; reprinted by American Mathematical Society, Providence (2000)
Temme, N.: Asymptotic Methods for Integrals. World Scientific, Singapore (2014)
Zuckerman, H.: On the coefficients of certain modular forms belonging to subgroups of the modular group. Trans. Am. Math. Soc. 45, 298–321 (1939)
Acknowledgements
The author wishes to thank Kathrin Bringmann and Walter Bridges for suggesting this problem, William Craig and Andreas Mono for sharing their knowledge on modular forms and helpful suggestions, and Johann Franke for helping me verify the results numerically. Finally, the author wishes to thank the referee for the careful review and helpful suggestions for improvements in the paper.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The author received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 101001179).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Mauth, L. Exact formula for cubic partitions. Ramanujan J (2024). https://doi.org/10.1007/s11139-024-00871-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11139-024-00871-0