betaFeedback

WorksMathematics

The Minimal excludant of ℓ\ell-regular partition

Abstract

Andrews and Newman \cite{andrewsnewmanI} introduced and studied ``mex,'' function for an integer partition λλ of a positive integer nn. The related function σmex(n)σ^{\mathrm{mex}}(n) was subsequently defined, and its generating function and other properties have been studied by several authors. In this article, for a positive integer ℓ≥2\ell \geq 2, we define the minimal excludant of an ℓ\ell-regular partition λλ of a positive integer nn. We then define σℓmex(n)σ_\ell^{\mathrm{mex}}(n), the sum of the minimal excludants over all ℓ\ell-regular partitions of nn, and obtain explicit generating functions for σℓmex(n)σ_\ell^{\mathrm{mex}}(n) in the cases ℓ=2\ell = 2 and ℓ=3\ell = 3, expressed as eta-quotients. We prove that the sequence σℓmex(n)σ_\ell^{\mathrm{mex}}(n) is weakly increasing for every ℓ\ell, and we obtain Hardy–Ramanujan type asymptotic formulas for σ2mex(n)σ_2^{\mathrm{mex}}(n) and σ3mex(n)σ_3^{\mathrm{mex}}(n) as n→∞n \to \infty. Additionally, we completely determine the parity of σ2mex(n)σ_2^{\mathrm{mex}}(n), showing that it is odd precisely at generalized pentagonal numbers, and we use this characterization to compute the density of nn for which σ2mex(n)σ_2^{\mathrm{mex}}(n) is odd versus even. Finally, for σ3mex(n)σ_3^{\mathrm{mex}}(n), we prove that it is odd infinitely often, and we establish a quantitative lower bound, showing that the number of such nn up to XX grows at least like log⁡log⁡X\log \log X.

Introduction and Statement of Results

A partition π=(π1,π2,…,πk)\pi= (\pi_1, \pi_2, \ldots, \pi_k) of a non-negative integer nn is a non-increasing sequence of positive integers summing to nn. Let p(n)p(n) denote the number of partitions of nn, and let P(n)P(n) denote the set of all partitions of nn. Grabner and Knopfmacher [10] introduced and studied an interesting partition statistic which they called the smallest gap of a partition, namely the least positive integer missing from its parts.

In 2019, Andrews and Newman [3, 4] revisited this idea, thereby rediscovering the Grabner–Knopfmacher statistic, and went on to generalize it very naturally to arithmetic progressions. They defined

σmex(n):=∑π∈P(n)mex⁡(π),\sigma^{\mathrm{mex}}(n) := \sum_{\pi\in P(n)} \operatorname{mex}(\pi),

and proved the striking identity

σmex(n)=D2(n).(1)\sigma^{\mathrm{mex}}(n) = D_2(n). \tag*{(1)}

where D2(n)D_2(n) denotes the number of two-colored partitions of nn into distinct parts; the corresponding generating function had, in fact, already appeared implicitly in the work of Grabner and Knopfmacher. They also obtained a Hardy–Ramanujan–Rademacher type exact formula for σmex(n)\sigma^{\mathrm{mex}}(n) in terms of Bessel functions and Dedekind sums, together with the asymptotic estimate σmex(n)∼146n3/4exp⁡(π2n3)\sigma^{\mathrm{mex}}(n) \sim\frac{1}{4\sqrt{6}n^{3/4}}\exp\left(\pi\sqrt{\frac{2n}{3}}\right) as n→∞n \to\infty. In the same paper, they introduced a second

2010 Mathematics Subject Classification: Primary 11P83, Secondary 11N37

Keywords: ℓ\ell-regular partition; Minimal excludant, Asymptotic formula, parity result arithmetic function

a(n):=∑π∈P(n)mex⁡(π) odd1,a(n) := \sum_{\substack{\pi\in\mathcal{P}(n)\\ \operatorname{mex}(\pi)\ \text{odd}}} 1,

and showed that σmex⁡(n)≡a(n)(mod2)\sigma_{\operatorname{mex}}(n) \equiv a(n) \pmod{2}, with a(n)a(n) is almost always even and is odd exactly when nn is of the form r(3r±1)r(3r \pm1), for some positive integer rr. In 2021, Ballantine and Merca [5] gave a combinatorial proof of the Andrews–Newman identity (1.1). Barman and Singh [6, 7], and later da Silva and Sellers [16], established further congruences and parity results for mex-related partition functions. In 2021, Chern [9] introduced the notion of the maximal excludant maex⁡(π)\operatorname{maex}(\pi), the largest integer below the biggest part of π\pi that fails to occur in π\pi, and obtained both a generating function identity for σmaex⁡(n):=∑π∈P(n)maex⁡(π)\sigma_{\operatorname{maex}}(n) := \sum_{\pi\in\mathcal{P}(n)} \operatorname{maex}(\pi) and the asymptotic relation σmaex⁡(n)∼σL(n)\sigma_{\operatorname{maex}}(n) \sim\sigma_L(n), where σL(n)\sigma_L(n) is the sum of the largest parts over all partitions of nn, with σL(n)\sigma_L(n) itself satisfying a Hardy–Ramanujan type formula due to Kessler and Livingston [14]. In 2022, Kaur, Bhoria, Eyyunni and Maji [13] have extended these investigations to mex statistics on partitions restricted in various ways, such as partitions into distinct parts and they studied the sum of minimal excludants over partitions into distinct parts σdmex⁡(n)\sigma_{\operatorname{dmex}}(n) and found a connection of Ramanujan’s sum σ(q)\sigma(q) with the sum function σdmex⁡(n)\sigma_{\operatorname{dmex}}(n). In 2026, Baruah and Talukdar [8] studied the partition functions involving least rr gaps and obtained the generating function of the related sum function in terms of eta quotients, studied their asymptotic formula. This growing literature makes clear that the minimal excludant, though a simple combinatorial rule, interacts richly with qq-series, modular forms, and analytic number theory. Motivated by the above results, we study the minimal excludant of ℓ\ell-regular partitions.

Let ℓ≥2\ell\ge2 be a positive integer. Let Pℓ(n)\mathcal{P}_{\ell}(n) denote the set of all ℓ\ell-regular partitions of nn and pℓ(n):=∣Pℓ(n)∣p_{\ell}(n) := \lvert\mathcal{P}_{\ell}(n)\rvert. For a partition λ∈Pℓ(n)\lambda\in\mathcal{P}_{\ell}(n), the ℓ\ell-regular minimal excludant is defined by mex⁡ℓ(λ):=min⁡{r≥1:ℓ∤r, r∉λ}\operatorname{mex}_{\ell}(\lambda) := \min\{r \ge1 : \ell\nmid r,\ r \notin\lambda\}. In other words, mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\lambda) is the smallest positive integer that is not divisible by ℓ\ell and does not occur as a part of λ\lambda. Let mm be positive integer such that ℓ∤m\ell\nmid m. We define

pℓmex⁡(m,n):=#{λ∈Pℓ(n):mex⁡ℓ(λ)=m}.p_{\ell}^{\operatorname{mex}}(m,n) := \#\{\lambda\in\mathcal{P}_{\ell}(n) : \operatorname{mex}_{\ell}(\lambda) = m\}.

If ℓ∣m\ell\mid m, then pℓmex⁡(m,n)=0p_{\ell}^{\operatorname{mex}}(m,n) = 0, by convention. We consider some examples to understand the above definitions.

Example 1. Let ℓ=3\ell= 3 and n=12n = 12. Then, the admissible parts are 1,2,4,5,7,8,10,11,…1, 2, 4, 5, 7, 8, 10, 11, \ldots.

  1. Consider λ=(8,2,2)\lambda= (8, 2, 2). The missing admissible parts are 1,4,5,7,…1, 4, 5, 7, \ldots. Hence mex⁡3(λ)=1\operatorname{mex}_{3}(\lambda) = 1.

  2. Let λ=(7,2,2,1)\lambda= (7, 2, 2, 1). The missing admissible parts are 4,5,8,…4, 5, 8, \ldots. Therefore mex⁡3(λ)=4\operatorname{mex}_{3}(\lambda) = 4.

  3. Let λ=(5,4,2,1)\lambda= (5, 4, 2, 1). The missing admissible integers are 7,8,10,…7, 8, 10, \ldots. Hence mex⁡3(λ)=7\operatorname{mex}_{3}(\lambda) = 7.

Notice that mex⁡ℓ\operatorname{mex}_{\ell} is unbounded.

Writing Pℓ(n)\mathcal{P}_{\ell}(n) for the set of ℓ\ell-regular partitions of nn, we let pℓmex⁡(m,n)p_{\ell}^{\operatorname{mex}}(m,n) denote the number of λ∈Pℓ(n)\lambda\in\mathcal{P}_{\ell}(n) with mex⁡ℓ(λ)=m\operatorname{mex}_{\ell}(\lambda) = m, ℓ∤m\ell\nmid m. In this article, we derive a general two-variable generating function for this statistic in Proposition 1. Our main object of study is the aggregate function

σℓmex⁡(n):=∑λ∈Pℓ(n)mex⁡ℓ(λ),\sigma_{\ell}^{\operatorname{mex}}(n) := \sum_{\lambda\in\mathcal{P}_{\ell}(n)} \operatorname{mex}_{\ell}(\lambda),

the natural analogue of σmex⁡(n)\sigma_{\operatorname{mex}}(n) for ℓ\ell-regular partitions. We obtain explicit generating functions for σℓmex⁡(n)\sigma_{\ell}^{\operatorname{mex}}(n) expressed as eta-quotients for ℓ=2\ell= 2 and ℓ=3\ell= 3.

Let λ∈P2(n)\lambda\in\mathcal{P}_2(n). Suppose that mex⁡2(λ)=2r−1\operatorname{mex}_2(\lambda)=2r-1. Then 1,3,5,…,2r−31,3,5,\ldots,2r-3 all occur as parts of λ\lambda, whereas 2r−12r-1 does not occur. We define p2mex⁡(r,n):=#{λ∈P2(n):mex⁡2(λ)=2r−1}p_2^{\operatorname{mex}}(r,n):=\#\{\lambda\in\mathcal{P}_2(n):\operatorname{mex}_2(\lambda)=2r-1\}.

Theorem 1.1. We have

∑n≥0σ2mex⁡(n)qn=(q2;q2)∞6(q;q)∞3(q4;q4)∞2=f26f13f22.(2)\sum_{n\geq0}\sigma_2^{\operatorname{mex}}(n)q^n=\frac{(q^2;q^2)_\infty^6}{(q;q)_\infty^3(q^4;q^4)_\infty^2}=\frac{f_2^6}{f_1^3f_2^2}. \tag*{(2)}

where, for brevity, we set fr=(qr;qr)∞f_r=(q^r;q^r)_\infty, r≥1r\geq1 is a positive integer.

Theorem 1.2. We have

∑n≥0σ3mex⁡(n)qn=f65f1f3f122+qf122f3f1f6.(3)\sum_{n\geq0}\sigma_3^{\operatorname{mex}}(n)q^n=\frac{f_6^5}{f_1f_3f_{12}^2}+q\frac{f_{12}^2f_3}{f_1f_6}. \tag*{(3)}

Further, we show that {σℓmex⁡(n)}n≥0\{\sigma_\ell^{\operatorname{mex}}(n)\}_{n\geq0} is weakly increasing for every ℓ≥2\ell\geq2. In particular, we find the following result.

Theorem 1.3. The sequence {σℓmex⁡(n)}n≥0\{\sigma_\ell^{\operatorname{mex}}(n)\}_{n\geq0} is weakly increasing. Moreover, it satisfies the following inequality

σℓmex⁡(n+1)−σℓmex⁡(n)≥pℓ(n)−pℓ(n−1).(4)\sigma_\ell^{\operatorname{mex}}(n+1)-\sigma_\ell^{\operatorname{mex}}(n)\geq p_\ell(n)-p_\ell(n-1). \tag*{(4)}

Using Tauberian theorem of Ingham, we establish Hardy–Ramanujan type asymptotic formulas for σ2mex⁡(n)\sigma_2^{\operatorname{mex}}(n) and σ3mex⁡(n)\sigma_3^{\operatorname{mex}}(n). In particular, we have the following results.

Theorem 1.4. We have

σ2mex⁡(n)∼122neπn/3,asn→∞.\sigma_2^{\operatorname{mex}}(n)\sim\frac{1}{2\sqrt{2n}}e^{\pi\sqrt{n/3}},\qquad\text{as}\qquad n\to\infty.

Theorem 1.5. We have

σ3mex⁡(n)∼14ne2π3n,asn→∞.\sigma_3^{\operatorname{mex}}(n)\sim\frac{1}{4\sqrt{n}}e^{\frac{2\pi}{3}\sqrt{n}},\qquad\text{as}\qquad n\to\infty.

We further determine the parity of σ2mex⁡(n)\sigma_2^{\operatorname{mex}}(n) completely, showing that it is odd precisely at generalized pentagonal numbers, exactly as in the Andrews–Newman setting, and deduce the corresponding density of nn for which σ2mex⁡(n)\sigma_2^{\operatorname{mex}}(n) is odd versus even. In particular, we get the following results.

Theorem 1.6. For every real number X≥1X\geq1, we have

#{1≤n≤X:σ2mex⁡(n) is an odd integer}=26X3+O(1),\#\{1\leq n\leq X:\sigma_2^{\operatorname{mex}}(n)\text{ is an odd integer}\}=\frac{2\sqrt{6X}}{3}+O(1),

while

#{1≤n≤X:σ2mex⁡(n) is an even integer}=X−26X3+O(1).\#\{1\leq n\leq X:\sigma_2^{\operatorname{mex}}(n)\text{ is an even integer}\}=X-\frac{2\sqrt{6X}}{3}+O(1).

Finally, we prove that σ3mex⁡(n)\sigma_3^{\operatorname{mex}}(n) is odd infinitely often, with a quantitative lower bound of order log⁡log⁡X\log\log X for the number of such nn up to XX. In particular, we obtain the following result.

Theorem 1.7. For sufficiently large XX, we have

#{0≤n≤X:σ3mex⁡(n)≡1(mod2)}≫log⁡log⁡X.\#\{0\leq n\leq X:\sigma_3^{\operatorname{mex}}(n)\equiv1\pmod{2}\}\gg\log\log X.

Preliminaries

Lemma 2.1. We have

∑t∈Zqt2=(q2;q2)∞5(q;q)∞2(q4;q4)∞2=f25f12f42.\sum_{t \in\mathbb{Z}} q^{t^2} = \frac{(q^2;q^2)_\infty^5}{(q;q)_\infty^2(q^4;q^4)_\infty^2} = \frac{f_2^5}{f_1^2f_4^2}.

Proof. See ([11]Eq. (1.5.1)). □

Lemma 2.2. For z≠0z \ne0, ∣Q∣<1\lvert Q\rvert< 1, we have

∑r=−∞∞Qr2zr=(−Qz;Q2)∞(−Qz;Q2)∞(Q2;Q2)∞.(5)\sum_{r=-\infty}^{\infty} Q^{r^2}z^r=(-Qz;Q^2)_\infty\left(-\frac{Q}{z};Q^2\right)_\infty(Q^2;Q^2)_\infty. \tag*{(5)}

Proof. See ([2] Theorem 2.8). □

Lemma 2.3 ([12]). Let B(q):=∑n=0∞b(n)qnB(q):=\sum_{n=0}^{\infty}b(n)q^n be a power series with radius of convergence 1. Assume that {b(n)}\{b(n)\} is a weakly increasing sequence of non-negative real numbers. If there are constants β,γ∈R\beta,\gamma\in\mathbb{R} and μ>0\mu>0 such that

B(e−y)∼βyγeμ/y,as y→0+,B(e^{-y})\sim\beta y^\gamma e^{\mu/y},\qquad\text{as }y\to0^+,

then we have

b(n)∼β2πμ2γ+14n2γ+34e2μnas n→∞.b(n)\sim\frac{\beta}{2\sqrt{\pi}}\frac{\mu^{\frac{2\gamma+1}{4}}}{n^{\frac{2\gamma+3}{4}}}e^{2\sqrt{\mu n}}\qquad\text{as }n\to\infty.

Lemma 2.4. For any positive integer uu and jj, and a prime number pp, we obtain

(qu;qu)∞pj≡(qpu;qpu)∞pj−1(modpj).(6)(q^u;q^u)_\infty^{p^j}\equiv(q^{pu};q^{pu})_\infty^{p^{j-1}}\pmod{p^j}. \tag*{(6)}

Lemma 2.5. Euler’s pentagonal theorem gives the expansion

f1=(q;q)∞=∑n=−∞∞(−1)nqn(3n−1)2=1+∑n=1∞(−1)n(qn(3n−1)2+qn(3n+1)2).(7)f_1=(q;q)_\infty=\sum_{n=-\infty}^{\infty}(-1)^nq^{\frac{n(3n-1)}{2}}=1+\sum_{n=1}^{\infty}(-1)^n\left(q^{\frac{n(3n-1)}{2}}+q^{\frac{n(3n+1)}{2}}\right). \tag*{(7)}

Proof. See ([2] Corollary 1.7). □

Lemma 2.6. We have

f13=(q;q)∞3=∑n=0∞(−1)n(2n+1)qn(n+1)2.(8)f_1^3=(q;q)_\infty^3=\sum_{n=0}^{\infty}(-1)^n(2n+1)q^{\frac{n(n+1)}{2}}. \tag*{(8)}

Proof. see ([11], Eq. (1.7.1) ) □

Proofs

Proposition 1. For a positive integer ℓ≥2\ell\ge2, define

Aℓ(m):=∑1≤j<mℓ∤jj.A_\ell(m):=\sum_{\substack{1\le j<m\\ \ell\nmid j}}j.

Then we have

∑m≥1ℓ∣m∑n=0∞pℓmex⁡(m,n)zmqn=(qℓ;qℓ)∞(q;q)∞∑m≥1ℓ∣mzmqAℓ(m)(1−qm).(9)\sum_{\substack{m\ge1\\ \ell\mid m}}\sum_{n=0}^{\infty}p_{\ell}^{\operatorname{mex}}(m,n)z^m q^n=\frac{(q^\ell;q^\ell)_\infty}{(q;q)_\infty}\sum_{\substack{m\ge1\\ \ell\mid m}}z^m q^{A_\ell(m)}(1-q^m). \tag*{(9)}

Proof. Suppose mex⁡ℓ(λ)=m\operatorname{mex}_{\ell}(\lambda)=m, ℓ∤m\ell\nmid m. Then

  1. every admissible integer 1≤r<m1\le r<m, ℓ∤r\ell\nmid r, must occur at least once;

  2. the part mm is absent;

  3. every admissible integer greater than mm may occur with arbitrary multiplicity.

The compulsory contribution of the parts below mm is

qAℓ(m)∏r<mℓ∤r11−qr,q^{A_\ell(m)}\prod_{\substack{r<m\\ \ell\nmid r}}\frac{1}{1-q^r},

and the unrestricted larger admissible parts contribute

∏r>mℓ∤r11−qr.\prod_{\substack{r>m\\ \ell\nmid r}}\frac{1}{1-q^r}.

Thus, we obtain

∑n=0∞pℓmex⁡(m,n)qn=qAℓ(m)∏r≥1ℓ∤r1−qm1−qr,ℓ∤m.(10)\sum_{n=0}^{\infty}p_{\ell}^{\operatorname{mex}}(m,n)q^n =q^{A_\ell(m)}\prod_{\substack{r\ge1\\ \ell\nmid r}}\frac{1-q^m}{1-q^r},\qquad\ell\nmid m. \tag*{(10)}

Since

∏r≥1ℓ∤r11−qr=(qℓ;qℓ)∞(q;q)∞,\prod_{\substack{r\ge1\\ \ell\nmid r}}\frac{1}{1-q^r} =\frac{(q^\ell;q^\ell)_\infty}{(q;q)_\infty},

we get

∑n=0∞pℓmex⁡(m,n)qn=qAℓ(m)(qℓ;qℓ)∞(q;q)∞(1−qm),ℓ∤m.(11)\sum_{n=0}^{\infty}p_{\ell}^{\operatorname{mex}}(m,n)q^n =q^{A_\ell(m)}\frac{(q^\ell;q^\ell)_\infty}{(q;q)_\infty}(1-q^m),\qquad\ell\nmid m. \tag*{(11)}

Consequently, we have

∑m≥1ℓ∤m∑n=0∞pℓmex⁡(m,n)zmqn=(qℓ;qℓ)∞(q;q)∞∑m≥1ℓ∤mzmqAℓ(m)(1−qm).(12)\sum_{\substack{m\ge1\\ \ell\nmid m}}\sum_{n=0}^{\infty}p_{\ell}^{\operatorname{mex}}(m,n)z^m q^n =\frac{(q^\ell;q^\ell)_\infty}{(q;q)_\infty}\sum_{\substack{m\ge1\\ \ell\nmid m}}z^m q^{A_\ell(m)}(1-q^m). \tag*{(12)}

Proof of Theorem 1.1 and 1.2

Proof of Theorem 1.1. Let p2mex⁡~(r,n)p_2^{\widetilde{\operatorname{mex}}}(r,n) be the number of 2-regular partitions λ\lambda of nn such that mex⁡(λ)=2r−1\operatorname{mex}(\lambda)=2r-1. We have

G2(z,q):=∑n≥0∑r≥1p2mex⁡~(r,n)z2r−1qn=∑r=1∞z2r−1∑n=0∞p2mex⁡~(r,n)qn.(13)G_2(z,q):=\sum_{n\ge0}\sum_{r\ge1}p_2^{\widetilde{\operatorname{mex}}}(r,n)z^{2r-1}q^n =\sum_{r=1}^{\infty}z^{2r-1}\sum_{n=0}^{\infty}p_2^{\widetilde{\operatorname{mex}}}(r,n)q^n. \tag*{(13)}

Since mex⁡2(λ)=2r−1\operatorname{mex}_2(\lambda)=2r-1, the compulsory odd parts are 1,3,5,…,2r−31,3,5,\ldots,2r-3 and their sum is

1+3+⋯+(2r−3)=(r−1)2.1+3+\cdots+(2r-3)=(r-1)^2.

Hence, these compulsory parts contribute q(r−1)2∏j=1r−211−q2j+1q^{(r-1)^2}\prod_{j=1}^{r-2}\frac{1}{1-q^{2j+1}}. The part 2r−12r-1 is excluded, while every larger odd part may occur with arbitrary multiplicity. Thus,

∑n=0∞p2mex⁡~(r,n)qn=q(r−1)2∏j=1r−211−q2j+1∏j=r∞11−q2j+1=q(r−1)2(1−q2r−1)(q;q2)∞.(14)\sum_{n=0}^{\infty}p_2^{\widetilde{\operatorname{mex}}}(r,n)q^n =q^{(r-1)^2}\prod_{j=1}^{r-2}\frac{1}{1-q^{2j+1}}\prod_{j=r}^{\infty}\frac{1}{1-q^{2j+1}} =\frac{q^{(r-1)^2}(1-q^{2r-1})}{(q;q^2)_\infty}. \tag*{(14)}

Combining (3.5) and (3.6) together, we get

G2(z,q)=∑r=1∞z2r−1∑n=0∞p2mex(r,n)qn=∑r=1∞z2r−1q(r−1)2(1−q2r−1)(q;q2)∞.(15)G_2(z,q)=\sum_{r=1}^{\infty}z^{2r-1}\sum_{n=0}^{\infty}p_2^{\mathrm{mex}}(r,n)q^n=\sum_{r=1}^{\infty}z^{2r-1}\frac{q^{(r-1)^2}(1-q^{2r-1})}{(q;q^2)_\infty}. \tag*{(15)}

Differentiating G2(z,q)G_2(z,q) with respect to zz and putting z=1z=1, we get

∑n≥0σ2mex(n)qn=∂∂zG2(z,q)∣z=1=∑r=1∞(2r−1)q(r−1)2(1−q2r−1)(q;q2)∞=1(q;q2)∞∑r=1∞(2r−1)q(r−1)2(1−q2r−1)=1(q;q2)∞[∑r=1∞(2r−1)q(r−1)2−∑r=1∞(2r−1)q(r−1)2q2r−1]=1(q;q2)∞[∑t=0∞(2t+1)qt2−∑t=0∞(2t+1)q(t+1)2]=1(q;q2)∞[∑t=0∞(2t+1)qt2−∑t=1∞(2t−1)qt2](16)\begin{aligned} \sum_{n\geq0}\sigma_2^{\mathrm{mex}}(n)q^n=\left.\frac{\partial}{\partial z}G_2(z,q)\right|_{z=1} \\ &=\sum_{r=1}^{\infty}(2r-1)\frac{q^{(r-1)^2}(1-q^{2r-1})}{(q;q^2)_\infty} =\frac{1}{(q;q^2)_\infty}\sum_{r=1}^{\infty}(2r-1)q^{(r-1)^2}(1-q^{2r-1}) \\ &=\frac{1}{(q;q^2)_\infty}\left[\sum_{r=1}^{\infty}(2r-1)q^{(r-1)^2}-\sum_{r=1}^{\infty}(2r-1)q^{(r-1)^2}q^{2r-1}\right] \\ &=\frac{1}{(q;q^2)_\infty}\left[\sum_{t=0}^{\infty}(2t+1)q^{t^2}-\sum_{t=0}^{\infty}(2t+1)q^{(t+1)^2}\right] \\ &=\frac{1}{(q;q^2)_\infty}\left[\sum_{t=0}^{\infty}(2t+1)q^{t^2}-\sum_{t=1}^{\infty}(2t-1)q^{t^2}\right] \tag*{(16)} \end{aligned}
=1(q;q2)∞[1+2∑t=1∞qt2]=∑t∈Zqt2(q;q2)∞=(q2;q2)∞6(q;q)∞3(q4;q4)∞2.(17)=\frac{1}{(q;q^2)_\infty}\left[1+2\sum_{t=1}^{\infty}q^{t^2}\right] =\frac{\sum_{t\in\mathbb{Z}}q^{t^2}}{(q;q^2)_\infty} =\frac{(q^2;q^2)_\infty^6}{(q;q)_\infty^3(q^4;q^4)_\infty^2}. \tag*{(17)}

Proof of Theorem 1.2. In order to prove Theorem 1.2, we need the following proposition.

Proposition 2. For any positive integer mm, with 3∤m3\nmid m, let

A3(m)=∑1≤j<m3∤jj,(18)A_3(m)=\sum_{\substack{1\leq j<m\\3\nmid j}}j, \tag*{(18)}

Then

A3r+1=3r2,A3r+2=3r2+3r+1for some non-negative integer r.(19)A_{3r+1}=3r^2,\qquad A_{3r+2}=3r^2+3r+1\quad\text{for some non-negative integer }r. \tag*{(19)}

Proof. Now we compute the quantity A3(m)A_3(m).

Case I. Suppose that m=3r+1m=3r+1. Then

A3(3r+1)=∑k=0r−1(3k+1)+∑k=0r−1(3k+2).A_3(3r+1)=\sum_{k=0}^{r-1}(3k+1)+\sum_{k=0}^{r-1}(3k+2).

Since

∑k=0r−1(3k+1)=r(3r−1)2and∑k=0r−1(3k+2)=r(3r+1)2,\sum_{k=0}^{r-1}(3k+1)=\frac{r(3r-1)}{2} \qquad\text{and}\qquad \sum_{k=0}^{r-1}(3k+2)=\frac{r(3r+1)}{2},

we obtain A3r+1=3r2A_{3r+1}=3r^2. Case II. Suppose that m=3r+2m=3r+2. Then

A3r+2=A3r+1+(3r+1)=3r2+(3r+1).A_{3r+2}=A_{3r+1}+(3r+1)=3r^{2}+(3r+1).

□\square

Proof of Theorem 1.2. From Proposition 1, we have

G3(z,q)=∑n≥0m≥13∤mp3mex(m,n)zmqn=∑m≥13∤mzmqA3(m)(1−qm)(q;q3)∞(q2;q3)∞.(20)G_3(z,q)=\sum_{\substack{n\geq0\\m\geq1\\3\nmid m}}p_3^{\mathrm{mex}}(m,n)z^mq^n=\sum_{\substack{m\geq1\\3\nmid m}}\frac{z^mq^{A_3(m)}(1-q^m)}{(q;q^3)_\infty(q^2;q^3)_\infty}. \tag*{(20)}

Using Proposition 2 in (20), we get

G3(z,q)=1(q;q3)∞(q2;q3)∞[∑r≥0z3r+1q3r2(1−q3r+1)+∑r≥0z3r+2q3r2+3r+1(1−q3r+2)].(21)G_3(z,q)=\frac{1}{(q;q^3)_\infty(q^2;q^3)_\infty}\left[\sum_{r\geq0}z^{3r+1}q^{3r^2}(1-q^{3r+1})+\sum_{r\geq0}z^{3r+2}q^{3r^2+3r+1}(1-q^{3r+2})\right]. \tag*{(21)}

Differentiating (21) with respect to zz and putting z=1z=1, we obtain

∂∂z(G3(z,q))∣z=1=∑n≥0σ3mex(n)qn:=∑m≥13∤mmp3mex(m,n)=1(q;q3)∞(q2;q3)∞[∑r≥0(3r+1)q3r2(1−q3r+1)+∑r≥0(3r+2)q3r2+3r+1(1−q3r+2)]=N3(q)(q;q3)∞(q2;q3)∞.(22)\begin{aligned} \left.\frac{\partial}{\partial z}(G_3(z,q))\right|_{z=1}=\sum_{n\geq0}\sigma_3^{\mathrm{mex}}(n)q^n:=\sum_{\substack{m\geq1\\3\nmid m}}mp_3^{\mathrm{mex}}(m,n) \\ &=\frac{1}{(q;q^3)_\infty(q^2;q^3)_\infty}\left[\sum_{r\geq0}(3r+1)q^{3r^2}(1-q^{3r+1})+\sum_{r\geq0}(3r+2)q^{3r^2+3r+1}(1-q^{3r+2})\right]=\frac{N_3(q)}{(q;q^3)_\infty(q^2;q^3)_\infty}. \tag*{(22)} \end{aligned}

where

N3(q)=∑r≥0(3r+1)q3r2(1−q3r+1)+∑r≥0(3r+2)q3r2+3r+1(1−q3r+2)=∑r≥0(3r+1)q3r2+∑r≥0q3r2+3r+1−∑r≥0(3r+2)q3(r+1)2=∑r≥0(3r+1)q3r2−∑r≥0(3r+2)q3(r+1)2+∑r≥0q3r2+3r+1=∑r≥0(3r+1)q3r2−∑r≥1(3r−1)q3r2+∑r≥0q3r2+3r+1(23)\begin{aligned} N_3(q)=\sum_{r\geq0}(3r+1)q^{3r^2}(1-q^{3r+1})+\sum_{r\geq0}(3r+2)q^{3r^2+3r+1}(1-q^{3r+2}) \\ &=\sum_{r\geq0}(3r+1)q^{3r^2}+\sum_{r\geq0}q^{3r^2+3r+1}-\sum_{r\geq0}(3r+2)q^{3(r+1)^2} \\ &=\sum_{r\geq0}(3r+1)q^{3r^2}-\sum_{r\geq0}(3r+2)q^{3(r+1)^2}+\sum_{r\geq0}q^{3r^2+3r+1} \\ &=\sum_{r\geq0}(3r+1)q^{3r^2}-\sum_{r\geq1}(3r-1)q^{3r^2}+\sum_{r\geq0}q^{3r^2+3r+1} \tag*{(23)} \end{aligned}
=1+2∑r≥1q3r2+∑r≥0q3r2+3r+1=∑r∈Zq3r2+∑r≥0q3r2+3r+1.(24)=1+2\sum_{r\geq1}q^{3r^2}+\sum_{r\geq0}q^{3r^2+3r+1}=\sum_{r\in\mathbb{Z}}q^{3r^2}+\sum_{r\geq0}q^{3r^2+3r+1}. \tag*{(24)}

Note that, for any real number rr,

(−r−1)2+(−r−1)=r2+r.(25)(-r-1)^2+(-r-1)=r^2+r. \tag*{(25)}

Using (25), we get

∑r≥0q3r2+3r+1=12∑r∈Zq3r2+3r+1=q2∑r=−∞∞(q3)r2(q3)r.(26)\sum_{r\geq0}q^{3r^2+3r+1}=\frac{1}{2}\sum_{r\in\mathbb{Z}}q^{3r^2+3r+1}=\frac{q}{2}\sum_{r=-\infty}^{\infty}(q^3)^{r^2}(q^3)^r. \tag*{(26)}

Applying Q=q3Q=q^{3} and z=q3z=q^{3} in Lemma 2.2, we have

∑r=−∞∞q3r2+3r=(−q6;q6)∞(−1;q6)∞(q6;q6)∞=2(−q6;q6)∞2(q6;q6)∞,(27)\sum_{r=-\infty}^{\infty}q^{3r^{2}+3r}=\left(-q^{6};q^{6}\right)_{\infty}\left(-1;q^{6}\right)_{\infty}\left(q^{6};q^{6}\right)_{\infty}=2\left(-q^{6};q^{6}\right)_{\infty}^{2}\left(q^{6};q^{6}\right)_{\infty}, \tag*{(27)}

since (−1;q6)∞=2(−q6;q6)∞\left(-1;q^{6}\right)_{\infty}=2\left(-q^{6};q^{6}\right)_{\infty}. Further observe that

(−q6;q6)∞=(q12;q12)∞(q6;q6)∞.(28)\left(-q^{6};q^{6}\right)_{\infty}=\frac{\left(q^{12};q^{12}\right)_{\infty}}{\left(q^{6};q^{6}\right)_{\infty}}. \tag*{(28)}

Using (27) and (28) in (26), we get

∑r≥0q3r2+3r+1=q(q12;q12)∞2(q6;q6)∞2.(29)\sum_{r\geq0}q^{3r^{2}+3r+1}=q\frac{\left(q^{12};q^{12}\right)_{\infty}^{2}}{\left(q^{6};q^{6}\right)_{\infty}^{2}}. \tag*{(29)}

Also, Using Lemma 2.1, we have

∑r∈Zq3r2=(q6;q6)∞5(q3;q3)∞2(q12;q12)∞2.(30)\sum_{r\in\mathbb{Z}}q^{3r^{2}}=\frac{\left(q^{6};q^{6}\right)_{\infty}^{5}}{\left(q^{3};q^{3}\right)_{\infty}^{2}\left(q^{12};q^{12}\right)_{\infty}^{2}}. \tag*{(30)}

Thus

∑n≥0σℓmex⁡(n)qn=1(q;q3)∞(q2;q3)∞[(q6;q6)∞5(q3;q3)∞2(q12;q12)∞2+q(q12;q12)∞2(q6;q6)∞2]=(q3;q3)∞(q;q)∞[(q6;q6)∞5(q3;q3)∞2(q12;q12)∞2+q(q12;q12)∞2(q6;q6)∞2]=(q6;q6)∞5(q;q)∞(q3;q3)∞(q12;q12)∞2+q(q12;q12)∞2(q3;q3)∞(q;q)∞(q6;q6)∞2.(31)\begin{aligned} \sum_{n\geq0}\sigma_{\ell}^{\operatorname{mex}}(n)q^{n} &=\frac{1}{\left(q;q^{3}\right)_{\infty}\left(q^{2};q^{3}\right)_{\infty}}\left[\frac{\left(q^{6};q^{6}\right)_{\infty}^{5}}{\left(q^{3};q^{3}\right)_{\infty}^{2}\left(q^{12};q^{12}\right)_{\infty}^{2}}+q\frac{\left(q^{12};q^{12}\right)_{\infty}^{2}}{\left(q^{6};q^{6}\right)_{\infty}^{2}}\right]\\ &=\frac{\left(q^{3};q^{3}\right)_{\infty}}{\left(q;q\right)_{\infty}}\left[\frac{\left(q^{6};q^{6}\right)_{\infty}^{5}}{\left(q^{3};q^{3}\right)_{\infty}^{2}\left(q^{12};q^{12}\right)_{\infty}^{2}}+q\frac{\left(q^{12};q^{12}\right)_{\infty}^{2}}{\left(q^{6};q^{6}\right)_{\infty}^{2}}\right]\\ &=\frac{\left(q^{6};q^{6}\right)_{\infty}^{5}}{\left(q;q\right)_{\infty}\left(q^{3};q^{3}\right)_{\infty}\left(q^{12};q^{12}\right)_{\infty}^{2}}+q\frac{\left(q^{12};q^{12}\right)_{\infty}^{2}\left(q^{3};q^{3}\right)_{\infty}}{\left(q;q\right)_{\infty}\left(q^{6};q^{6}\right)_{\infty}^{2}}. \tag*{(31)} \end{aligned}

Proof of Theorem 1.3. Consider the map ϕ:Pℓ(n)⟶Pℓ(n+1)\phi:\mathcal{P}_{\ell}(n)\longrightarrow\mathcal{P}_{\ell}(n+1) by ϕ(λ)=λ∪{1}\phi(\lambda)=\lambda\cup\{1\}. Since 11 is never divisible by ℓ\ell, ϕ(λ)\phi(\lambda) is still an ℓ\ell-regular partition. Moreover, ∣ϕ(λ)∣=∣λ∣+1=n+1|\phi(\lambda)|=|\lambda|+1=n+1. The map is injective, because if ϕ(λ1)=ϕ(λ2)\phi(\lambda_{1})=\phi(\lambda_{2}), then removing one copy of 11 from both partitions gives λ1=λ2\lambda_{1}=\lambda_{2}. Now, we consider two cases. Firstly, suppose that 1∈λ1\in\lambda, then mex⁡ℓ(λ)>1\operatorname{mex}_{\ell}(\lambda)>1. Adding another 11 does not change whether any other part occurs. Hence mex⁡ℓ(ϕ(λ))=mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\phi(\lambda))=\operatorname{mex}_{\ell}(\lambda). Secondly, suppose that 1∉λ1\notin\lambda, then mex⁡ℓ(λ)=1\operatorname{mex}_{\ell}(\lambda)=1. After adding 11, 1∈ϕ(λ)1\in\phi(\lambda). Therefore, the mex of ϕ(λ)\phi(\lambda) is the smallest positive integer other than 11 which is not divisible by ℓ\ell and does not occur in λ\lambda. Thus, mex⁡ℓ(ϕ(λ))>1=mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\phi(\lambda))>1=\operatorname{mex}_{\ell}(\lambda). Therefore, in both cases, mex⁡ℓ(ϕ(λ))≥mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\phi(\lambda))\geq\operatorname{mex}_{\ell}(\lambda). Since ϕ\phi is injective, we have

σℓmex⁡(n+1)=∑μ∈Pℓ(n+1)mex⁡ℓ(μ)≥∑λ∈Pℓ(n)mex⁡ℓ(ϕ(λ))≥∑λ∈Pℓ(n)mex⁡ℓ(λ)=σℓmex⁡(n).\sigma_{\ell}^{\operatorname{mex}}(n+1)=\sum_{\mu\in\mathcal{P}_{\ell}(n+1)}\operatorname{mex}_{\ell}(\mu)\geq\sum_{\lambda\in\mathcal{P}_{\ell}(n)}\operatorname{mex}_{\ell}(\phi(\lambda))\geq\sum_{\lambda\in\mathcal{P}_{\ell}(n)}\operatorname{mex}_{\ell}(\lambda)=\sigma_{\ell}^{\operatorname{mex}}(n).

Hence σℓmex⁡(n+1)≥σℓmex⁡(n)\sigma_{\ell}^{\operatorname{mex}}(n+1)\geq\sigma_{\ell}^{\operatorname{mex}}(n) (n≥0n\geq0). Therefore, the sequence {σℓmex⁡(n)}n≥0\{\sigma_{\ell}^{\operatorname{mex}}(n)\}_{n\geq0} is weakly increasing for every ℓ≥2\ell\geq2.

As defined earlier, we have pℓ(n)=∣Pℓ(n)∣p_{\ell}(n)=|\mathcal{P}_{\ell}(n)|. Every partition λ∈Pℓ(n)\lambda\in\mathcal{P}_{\ell}(n) with 1∉λ1\notin\lambda satisfies mex⁡ℓ(λ)=1\operatorname{mex}_{\ell}(\lambda)=1. After adding a part 11, its mex becomes strictly larger, i.e. mex⁡ℓ(ϕ(λ))>1=\operatorname{mex}_{\ell}(\phi(\lambda))>1= mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\lambda). If 1∈λ1 \in\lambda, then mex⁡ℓ(ϕ(λ))=mex⁡ℓ(λ)\operatorname{mex}_{\ell}(\phi(\lambda)) = \operatorname{mex}_{\ell}(\lambda). Hence

σℓmex⁡(n+1)−σℓmex⁡(n)≥#{λ∈Pℓ(n):1∉λ}.(32)\sigma_{\ell}^{\operatorname{mex}}(n+1)-\sigma_{\ell}^{\operatorname{mex}}(n)\geq\#\{\lambda\in\mathcal{P}_{\ell}(n):1\notin\lambda\}. \tag*{(32)}

Now consider the partitions in Pℓ(n)\mathcal{P}_{\ell}(n) which contain at least one part 11. Removing one occurrence of 11 gives a bijection with Pℓ(n−1)\mathcal{P}_{\ell}(n-1). Therefore,

#{λ∈Pℓ(n):1∉λ}=pℓ(n)−pℓ(n−1).(33)\#\{\lambda\in\mathcal{P}_{\ell}(n):1\notin\lambda\}=p_{\ell}(n)-p_{\ell}(n-1). \tag*{(33)}

Comparing (3.24) and (3.25), we get

σℓmex⁡(n+1)−σℓmex⁡(n)≥pℓ(n)−pℓ(n−1)\sigma_{\ell}^{\operatorname{mex}}(n+1)-\sigma_{\ell}^{\operatorname{mex}}(n)\geq p_{\ell}(n)-p_{\ell}(n-1)

□\square

Proof of Theorem 1.4. Let

C(q)=∑n≥0σ2mex⁡(n)qn=(q2;q2)∞6(q;q)∞3(q4;q4)∞2.C(q)=\sum_{n\geq0}\sigma_{2}^{\operatorname{mex}}(n)q^{n}=\frac{(q^{2};q^{2})_{\infty}^{6}}{(q;q)_{\infty}^{3}(q^{4};q^{4})_{\infty}^{2}}.

Applying the transformation formula for the Dedekind’s eta function [15], we have

(e−y;e−y)∞∼2πyexp⁡(−π26y),y→0+.(34)(e^{-y};e^{-y})_{\infty}\sim\sqrt{\frac{2\pi}{y}}\exp\left(-\frac{\pi^{2}}{6y}\right),\qquad y\to0^{+}. \tag*{(34)}

Thus

(e−2y;e−2y)∞6∼(πy)6exp⁡(−π22y)=π3y3e−π22y,(e^{-2y};e^{-2y})_{\infty}^{6}\sim\left(\sqrt{\frac{\pi}{y}}\right)^{6}\exp\left(-\frac{\pi^{2}}{2y}\right)=\frac{\pi^{3}}{y^{3}}e^{-\frac{\pi^{2}}{2y}},

while

(e−y;e−y)∞3∼(2πy)3e−π22y,(e^{-y};e^{-y})_{\infty}^{3}\sim\left(\sqrt{\frac{2\pi}{y}}\right)^{3}e^{-\frac{\pi^{2}}{2y}},

and

(e−4y;e−4y)∞2∼(π2y)2e−π212y=π2ye−π212y.(e^{-4y};e^{-4y})_{\infty}^{2}\sim\left(\sqrt{\frac{\pi}{2y}}\right)^{2}e^{-\frac{\pi^{2}}{12y}}=\frac{\pi}{2y}e^{-\frac{\pi^{2}}{12y}}.

Therefore,

C(e−y)∼π2 y−1/2eπ212y.C(e^{-y})\sim\sqrt{\frac{\pi}{2}}\,y^{-1/2}e^{\frac{\pi^{2}}{12y}}.

From Theorem 1.3, we know that the sequence {σ2mex⁡(n)}\{\sigma_{2}^{\operatorname{mex}}(n)\} is weakly increasing. Thus, by applying Lemma 2.3 with parameter

β=π2,γ=−12,μ=π212,\beta=\sqrt{\frac{\pi}{2}},\qquad\gamma=-\frac{1}{2},\qquad\mu=\frac{\pi^{2}}{12},

we obtain

σ2mex⁡(n)∼122neπn/3,as n→∞.\sigma_{2}^{\operatorname{mex}}(n)\sim\frac{1}{2\sqrt{2n}}e^{\pi\sqrt{n/3}},\qquad\text{as }n\to\infty.

□\square

Proof of Theorem 1.5. From (3.23), we have

∑n≥0σ3mex⁡(n)qn=(q6;q6)∞5(q;q)∞(q3;q3)∞(q12;q12)∞2+q(q12;q12)∞2(q3;q3)∞(q;q)∞(q6;q6)∞.\sum_{n\geq0}\sigma_{3}^{\operatorname{mex}}(n)q^{n} =\frac{(q^{6};q^{6})_{\infty}^{5}}{(q;q)_{\infty}(q^{3};q^{3})_{\infty}(q^{12};q^{12})_{\infty}^{2}} +q\frac{(q^{12};q^{12})_{\infty}^{2}(q^{3};q^{3})_{\infty}}{(q;q)_{\infty}(q^{6};q^{6})_{\infty}}.

Let

T1(q):=(q6;q6)∞5(q;q)∞(q3;q3)∞(q12;q12)∞2.T_1(q):=\frac{(q^6;q^6)_\infty^5}{(q;q)_\infty(q^3;q^3)_\infty(q^{12};q^{12})_\infty^2}.

Using (34), we get

(e−6y;e−6y)∞5∼(π3y)5e−5π236y.(e^{-6y};e^{-6y})_\infty^5\sim\left(\sqrt{\frac{\pi}{3y}}\right)^5e^{-\frac{5\pi^2}{36y}}.

and

(e−y;e−y)∞(e−3y;e−3y)∞(e−12y;e−12y)∞2∼2πy2π3y(π6y)2e−π26y−π218y−π236y.(e^{-y};e^{-y})_\infty(e^{-3y};e^{-3y})_\infty(e^{-12y};e^{-12y})_\infty^2\sim\sqrt{\frac{2\pi}{y}}\sqrt{\frac{2\pi}{3y}}\left(\sqrt{\frac{\pi}{6y}}\right)^2e^{-\frac{\pi^2}{6y}-\frac{\pi^2}{18y}-\frac{\pi^2}{36y}}.

Therefore,

T1(e−y)∼π3y−1/2eπ29y.T_1(e^{-y})\sim\frac{\sqrt{\pi}}{3}y^{-1/2}e^{\frac{\pi^2}{9y}}.

Similarly, let

T2(q):=q(q12;q12)∞2(q3;q3)∞(q;q)∞(q6;q6)∞.T_2(q):=q\frac{(q^{12};q^{12})_\infty^2(q^3;q^3)_\infty}{(q;q)_\infty(q^6;q^6)_\infty}.

As y→0+y\to0^+, we get q=e−y→1q=e^{-y}\to1 and hence q∼1q\sim1. Thus

T2(e−y)∼π6y−1/2eπ29y.T_2(e^{-y})\sim\frac{\sqrt{\pi}}{6}y^{-1/2}e^{\frac{\pi^2}{9y}}.

Note that both terms have the same exponential growth. Consequently,

T1(e−y)+T2(e−y)=∑n≥0σ3mex(n)e−ny∼(π3+π6)y−1/2eπ29y=π2y−1/2eπ29y.T_1(e^{-y})+T_2(e^{-y})=\sum_{n\geq0}\sigma_3^{\mathrm{mex}}(n)e^{-ny}\sim\left(\frac{\sqrt{\pi}}{3}+\frac{\sqrt{\pi}}{6}\right)y^{-1/2}e^{\frac{\pi^2}{9y}}=\frac{\sqrt{\pi}}{2}y^{-1/2}e^{\frac{\pi^2}{9y}}.

Applying the Lemma 2.3 with parameter β=π2\beta=\frac{\sqrt{\pi}}{2}, γ=−12\gamma=-\frac{1}{2}, μ=π29\mu=\frac{\pi^2}{9}, we get

σ3mex(n)∼14ne2π3n,as n→∞.\sigma_3^{\mathrm{mex}}(n)\sim\frac{1}{4\sqrt{n}}e^{\frac{2\pi}{3}\sqrt{n}},\qquad\text{as }n\to\infty.

Proof of Theorem 1.6

In order to prove Theorem 1.6, we need the following parity result for σ2mex(n)\sigma_2^{\mathrm{mex}}(n).

Proposition 3. We have

σ2mex(n)≡1(mod2)if and only ifn=0orn=k(3k−1)2orn=k(3k+1)2\sigma_2^{\mathrm{mex}}(n)\equiv1\pmod2\quad\text{if and only if}\quad n=0\quad\text{or}\quad n=\frac{k(3k-1)}{2}\quad\text{or}\quad n=\frac{k(3k+1)}{2}

for some positive integer $k.

Proof. From equation (17), we have

∑n=0∞σ2mex(n)qn=(q2;q2)∞6(q;q)∞3(q4;q4)∞2.(35)\sum_{n=0}^{\infty}\sigma_2^{\mathrm{mex}}(n)q^n=\frac{(q^2;q^2)_\infty^6}{(q;q)_\infty^3(q^4;q^4)_\infty^2}. \tag*{(35)}

By using Lemma 2.4, we get

∑n=0∞σ2mex(n)qn≡(q;q)∞(mod2).(36)\sum_{n=0}^{\infty}\sigma_2^{\mathrm{mex}}(n)q^n \equiv(q;q)_\infty\pmod{2}. \tag*{(36)}

From Lemma 2.5, taking modulo 2, we obtain

(q;q)∞=1+∑k≥1(qk(3k−1)2+qk(3k+1)2)(mod2).(37)(q;q)_\infty=1+\sum_{k\geq1}\left(q^{\frac{k(3k-1)}{2}}+q^{\frac{k(3k+1)}{2}}\right)\pmod{2}. \tag*{(37)}

From (37) and (36), we obtain

σ2mex(n)≡1(mod2)⟺n=k(3k−1)2orn=k(3k+1)2\sigma_2^{\mathrm{mex}}(n)\equiv1\pmod{2}\quad\Longleftrightarrow\quad n=\frac{k(3k-1)}{2}\quad\text{or}\quad n=\frac{k(3k+1)}{2}

for some integer k≥1k\geq1. □

Proof of Theorem 1.6. For a real number XX, we define

A(X):=#{n≤X:σ2mex(n) is odd},A(X):=\#\{n\leq X:\sigma_2^{\mathrm{mex}}(n)\text{ is odd}\},

and

B(X):=#{n≤X:σ2mex(n) is even}.B(X):=\#\{n\leq X:\sigma_2^{\mathrm{mex}}(n)\text{ is even}\}.

Note that for any real number kk,

k(3k−1)2≤X⟺k≤1+1+24X6.\frac{k(3k-1)}{2}\leq X\quad\Longleftrightarrow\quad k\leq\frac{1+\sqrt{1+24X}}{6}.

Similarly,

k(3k+1)2≤X⟺k≤−1+1+24X6.\frac{k(3k+1)}{2}\leq X\quad\Longleftrightarrow\quad k\leq\frac{-1+\sqrt{1+24X}}{6}.

Using Proposition 3, we get

A(X)=⌊1+1+24X6⌋+⌊−1+1+24X6⌋=1+1+24X6+−1+1+24X6+O(1)=1+24X3+O(1).(38)\begin{aligned} A(X)=\left\lfloor\frac{1+\sqrt{1+24X}}{6}\right\rfloor+\left\lfloor\frac{-1+\sqrt{1+24X}}{6}\right\rfloor\\ &=\frac{1+\sqrt{1+24X}}{6}+\frac{-1+\sqrt{1+24X}}{6}+O(1)=\frac{\sqrt{1+24X}}{3}+O(1). \tag*{(38)} \end{aligned}

and

B(X)=⌊X⌋−A(X)=X−A(x)+O(1).B(X)=\lfloor X\rfloor-A(X)=X-A(x)+O(1).

For any real number uu, we have

(1+u)1/2=1+12u−18u2+116u3−5128u4+⋯ ,(39)(1+u)^{1/2}=1+\frac{1}{2}u-\frac{1}{8}u^2+\frac{1}{16}u^3-\frac{5}{128}u^4+\cdots, \tag*{(39)}

For large xx, using (39), we get

x+a=x(1+ax)1/2=x+a2x−a28x3/2+a316x5/2−5a4128x7/2+⋯ .(40)\sqrt{x+a}=\sqrt{x}\left(1+\frac{a}{x}\right)^{1/2}=\sqrt{x}+\frac{a}{2\sqrt{x}}-\frac{a^2}{8x^{3/2}}+\frac{a^3}{16x^{5/2}}-\frac{5a^4}{128x^{7/2}}+\cdots. \tag*{(40)}

For large value of xx,

x+a=x+a2x+O(x−3/2).\sqrt{x+a}=\sqrt{x}+\frac{a}{2\sqrt{x}}+O\left(x^{-3/2}\right).

In particular

24X+13=24X3+1624X+O(X−3/2).\frac{\sqrt{24X+1}}{3}=\frac{\sqrt{24X}}{3}+\frac{1}{6\sqrt{24X}}+O\left(X^{-3/2}\right).

Thus

A(X)=263X+O(1).A(X)=\frac{2\sqrt{6}}{3}\sqrt{X}+O(1).

Therefore,

B(X)=X−263X+O(1).B(X)=X-\frac{2\sqrt{6}}{3}\sqrt{X}+O(1).

Proof of Theorem 1.7

In order to prove Theorem 1.7, we need the following lemma.

Lemma 3.1. Let nn be a positive integer such that n≡2(mod3)n\equiv2\pmod{3}. Then we have

∑s=0∞σ3mex(n−s(3s−1)2)+∑s=1∞σ3mex(n−s(3s+1)2)≡0(mod2).\sum_{s=0}^{\infty}\sigma_{3}^{\mathrm{mex}}\left(n-\frac{s(3s-1)}{2}\right)+\sum_{s=1}^{\infty}\sigma_{3}^{\mathrm{mex}}\left(n-\frac{s(3s+1)}{2}\right)\equiv0\pmod{2}.

Proof. Let

F(q):=∑n≥0σ3mex(n)qn=(q6;q6)∞5(q;q)∞(q3;q3)∞(q12;q12)∞2+q(q12;q12)∞2(q3;q3)∞(q;q)∞(q6;q6)∞.(41)F(q):=\sum_{n\geq0}\sigma_{3}^{\mathrm{mex}}(n)q^n =\frac{(q^6;q^6)_\infty^5}{(q;q)_\infty(q^3;q^3)_\infty(q^{12};q^{12})_\infty^2} +q\frac{(q^{12};q^{12})_\infty^2(q^3;q^3)_\infty}{(q;q)_\infty(q^6;q^6)_\infty}. \tag*{(41)}

Taking modulo 2 of (41), we get

F(q)≡f3f1(1+qf36)(mod2)⟺f1F(q)=f3+qf37(mod2).F(q)\equiv\frac{f_3}{f_1}\left(1+qf_3^6\right)\pmod{2}\Longleftrightarrow f_1F(q)=f_3+qf_3^7\pmod{2}.

Thus,

(q;q)∞∑n≥0σ3mex(n)qn≡f3+qf37(mod2).(42)(q;q)_\infty\sum_{n\geq0}\sigma_{3}^{\mathrm{mex}}(n)q^n\equiv f_3+qf_3^7\pmod{2}. \tag*{(42)}

From Lemma 2.5 and taking modulo 2 on it, we get

(q;q)∞≡1+∑s=1∞qs(3s−1)/2+∑s=1∞qs(3s+1)/2(mod2).(43)(q;q)_\infty\equiv1+\sum_{s=1}^{\infty}q^{s(3s-1)/2}+\sum_{s=1}^{\infty}q^{s(3s+1)/2}\pmod{2}. \tag*{(43)}

Using (43) in (42), we obtain

(∑s=0∞qs(3s−1)/2+∑s=1∞qs(3s+1)/2)(∑n≥0σ3mex(n)qn)≡f3+qf37(mod2).(44)\left(\sum_{s=0}^{\infty}q^{s(3s-1)/2}+\sum_{s=1}^{\infty}q^{s(3s+1)/2}\right)\left(\sum_{n\geq0}\sigma_{3}^{\mathrm{mex}}(n)q^n\right)\equiv f_3+qf_3^7\pmod{2}. \tag*{(44)}

Since qf37qf_3^7 contains only powers q3j+1q^{3j+1} and comparing the coefficients of qnq^n of (44), we obtain

∑s=0∞σ3mex(n−s(3s−1)2)+∑s=1∞σ3mex(n−s(3s+1)2)≡0(mod2),n≡2(mod3).(45)\sum_{s=0}^{\infty}\sigma_{3}^{\mathrm{mex}}\left(n-\frac{s(3s-1)}{2}\right)+\sum_{s=1}^{\infty}\sigma_{3}^{\mathrm{mex}}\left(n-\frac{s(3s+1)}{2}\right)\equiv0\pmod{2},\qquad n\equiv2\pmod{3}. \tag*{(45)}

Lemma 3.2. For every integer r≥2r\geq2 satisfying r≡2(mod3)r\equiv2\pmod{3}, there exists an integer n∈[2r−1,r(3r−1)2]n\in\left[2r-1,\frac{r(3r-1)}{2}\right] such that σ3mex(n)≡1(mod2)\sigma_{3}^{\mathrm{mex}}(n)\equiv1\pmod{2}. In particular, there are infinitely many integers nn for which σ3mex(n)\sigma_{3}^{\mathrm{mex}}(n) is an odd integer. ::

Proof. On contrary suppose that σ3mex(n)\sigma_{3}^{\mathrm{mex}}(n) is even for every 2r−1≤n≤r(3r−1)22r-1 \le n \le\frac{r(3r-1)}{2}. We define

S(s):=r(3r−1)2−s(3s−1)2andR(s):=r(3r−1)2−s(3s+1)2.(46)S(s):=\frac{r(3r-1)}{2}-\frac{s(3s-1)}{2} \quad\text{and}\quad R(s):=\frac{r(3r-1)}{2}-\frac{s(3s+1)}{2}. \tag*{(46)}

Observe that if r≡2(mod3)r\equiv2\pmod{3}, then r(3r−1)2≡r(mod3)\frac{r(3r-1)}{2}\equiv r\pmod{3}. Therefore, (3.37) can be rewritten as

∑s=0∞σ3mex(S(s))+∑s=1∞σ3mex(R(s))≡0(mod2),r≡2(mod3).(47)\sum_{s=0}^{\infty}\sigma_{3}^{\mathrm{mex}}(S(s))+\sum_{s=1}^{\infty}\sigma_{3}^{\mathrm{mex}}(R(s))\equiv0\pmod{2},\qquad r\equiv2\pmod{3}. \tag*{(47)}

Notice that for s≥r+1s\ge r+1, S(s)<0S(s)<0, and for s≥rs\ge r, R(s)<0R(s)<0. We use the standard convention σ3mex(n)=0\sigma_{3}^{\mathrm{mex}}(n)=0 (n<0)(n<0). Thus (3.39) reduces to

1+∑s=0r−1σ3mex(S(s))+∑s=1r−1σ3mex(R(s))≡0(mod2),(48)1+\sum_{s=0}^{r-1}\sigma_{3}^{\mathrm{mex}}(S(s))+\sum_{s=1}^{r-1}\sigma_{3}^{\mathrm{mex}}(R(s))\equiv0\pmod{2}, \tag*{(48)}

since σ3mex(S(r))=σ3mex(0)=1\sigma_{3}^{\mathrm{mex}}(S(r))=\sigma_{3}^{\mathrm{mex}}(0)=1. Now we show that S(s),R(s)S(s),R(s) lie in the interval [2r−1,r(3r−1)2]\left[2r-1,\frac{r(3r-1)}{2}\right] for 1≤s≤r−11\le s\le r-1. Notice that S(0)=r(3r−1)2S(0)=\frac{r(3r-1)}{2}, S(r−1)=3r−2S(r-1)=3r-2. Note that S(s)S(s) decreases in ss and the smallest value for 0≤s≤r−10\le s\le r-1 is S(r−1)S(r-1). Since 3r−2≥2r−13r-2\ge2r-1 (r≥1)(r\ge1), we have 2r−1≤S(s)≤r(3r−1)22r-1\le S(s)\le\frac{r(3r-1)}{2}, for 0≤s≤r−10\le s\le r-1. Similarly, for 1≤s≤r−11\le s\le r-1, again R(s)R(s) decreases with ss, and R(r−1)=2r−1R(r-1)=2r-1. Thus 2r−1≤R(s)≤r(3r−1)22r-1\le R(s)\le\frac{r(3r-1)}{2}. Therefore, S(s),R(s)S(s),R(s) lie in the interval [2r−1,r(3r−1)2]\left[2r-1,\frac{r(3r-1)}{2}\right]. By our assumption, the left-hand side of (3.40) is 1+0+0≡1(mod2)1+0+0\equiv1\pmod{2}, contradicting (3.40), which says that it is 0(mod2)0\pmod{2}. Therefore, at least one integer nn from n∈[2r−1,r(3r−1)2]n\in\left[2r-1,\frac{r(3r-1)}{2}\right] such that σ3mex(n)≡1(mod2)\sigma_{3}^{\mathrm{mex}}(n)\equiv1\pmod{2}.

Proof of Theorem 1.7

Proof of Theorem 1.7. Suppose ak≡2(mod3)a_k\equiv2\pmod{3}, for some positive integer kk. Then ak+1=ak(3ak−1)2≡2(mod3)a_{k+1}=\frac{a_k(3a_k-1)}{2}\equiv2\pmod{3}. Thus, ak≡2(mod3)a_k\equiv2\pmod{3} for every k≥0k\ge0. In particular, we choose a0=2a_0=2, a1=5a_1=5, a2=35a_2=35, and so on. By applying the Lemma 3.2 with r=akr=a_k, there exists some integer nkn_k such that

ak≤2ak−1≤nk≤ak(3ak−1)2=ak+1andσ3mex(nk)=1(mod2).(49)a_k\le2a_k-1\le n_k\le\frac{a_k(3a_k-1)}{2}=a_{k+1} \quad\text{and}\quad \sigma_{3}^{\mathrm{mex}}(n_k)=1\pmod{2}. \tag*{(49)}

Let ν\nu be the largest integer such that aν≤Xa_\nu\le X. Then aν≤X<aν+1a_\nu\le X<a_{\nu+1}. From (3.41), we have nk≥2ak−1n_k\ge2a_k-1. Since ak+1=ak(3ak−1)2a_{k+1}=\frac{a_k(3a_k-1)}{2}, we have, for ak≥2a_k\ge2, ak+1>2aka_{k+1}>2a_k. Thus the interval [2ak−1,ak+1][2a_k-1,a_{k+1}] is disjoint with the analogous interval corresponding to k+2k+2. Now the number of nkn_k for which σ3mex(nk)\sigma_{3}^{\mathrm{mex}}(n_k) is odd is at least ν\nu. Now for all k≥0k\ge0, we obtain

ak+1=ak(3ak−1)2<32ak2<2ak2.a_{k+1}=\frac{a_k(3a_k-1)}{2}<\frac{3}{2}a_k^2<2a_k^2.

On iteration, we get ak<(2a0)2k<52k⟹log⁡log⁡ak<klog⁡2+log⁡log⁡5a_k<(2a_0)^{2^k}<5^{2^k}\Longrightarrow\log\log a_k<k\log2+\log\log5. Since aν≤X<aν+1a_\nu\le X<a_{\nu+1}, we have log⁡log⁡X<log⁡log⁡aν+1<νlog⁡2+C\log\log X<\log\log a_{\nu+1}<\nu\log2+C where C=log⁡log⁡5+log⁡2C=\log\log5+\log2 is a fixed constant. Therefore, for sufficiently large X,ν≫log⁡log⁡XX,\nu\gg\log\log X.

Conclusion

In this paper, we have introduced and studied the minimal excludant statistic for ℓ\ell-regular partitions, extending the classical work of Andrews and Newman to this more general setting. Further, we obtained explicit eta-quotient representations for its generating function when ℓ=2\ell= 2 and ℓ=3\ell= 3, revealing an elegant connection between this combinatorial statistic and the theory of modular forms. We also obtained asymptotic formula for σ2max⁡(n)\sigma_2^{\max}(n) and σ3max⁡(n)\sigma_3^{\max}(n). Further we obtained that the parity result for σ2max⁡(n)\sigma_2^{\max}(n) is governed entirely by the generalized pentagonal numbers which helps us to find a density result for σ2max⁡(n)\sigma_2^{\max}(n) while σ3max⁡(n)\sigma_3^{\max}(n) is shown to be odd infinitely often, with a quantitative log⁡log⁡X\log\log X lower bound obtained through a delicate dissection argument involving Euler’s pentagonal number theorem modulo 2. These results suggest that the minimal excludant statistic for ℓ\ell-regular partitions carries rich arithmetic structure worthy of further exploration. Natural directions for future work include extending the parity and density results to σℓmax⁡(n)\sigma_\ell^{\max}(n) for ℓ≥4\ell\ge4, investigating congruences modulo primes other than 2, and studying the mex statistic for other classes of restricted partitions.

Data availability statement: There is no data associated to our manuscript.

References

Nabin Kumar Meher, Department of Mathematics, National Institute of Technology Raipur, G. E. Road, Raipur, Chhattisgarh, India 492010.

Email address: [email protected], [email protected]

References

  1. [2]G. E. Andrews, The Theory of Partition, Reprint of the 1976 original. Cambridge University Press, Cambridge (2010).
  2. [3]G. E. Andrews, D. Newman, Partitions and the minimal excludant, Ann. Comb. 23 (2019), no. 2, 249–254.DOI
  3. [4]G. E. Andrews, D. Newman, The minimal excludant in integer partitions, J. Integer Seq. 23 (2020), Article 20.2.3.
  4. [5]C. Ballantine, M. Merca, Combinatorial proof of the minimal excludant theorem, Int. J. Number Theory, 17, No. 8, 1765–1779 (2021).arxiv.org/abs/1908.06789
  5. [6]R. Barman, A. Singh, On Mex-related partition functions of Andrews and Newman, Res. Number Theory, 7, 53 (2021).arxiv.org/abs/2009.11602
  6. [7]R. Barman, A. Singh, Mex-related partition functions of Andrews and Newman, J. Integer Seq., 24 (2021), Article 21.6.3.arxiv.org/abs/2009.11602
  7. [8]N. D. Baruah, P. Talukdar, Arithmetic and asymptotic properties for some functions related to the least r-gaps of partitions, Ann. Comb., 30 (2026), 526–551.DOI
  8. [9]S. Chern, Partitions and the maximal excludant, Electron. J. Combin. 28 (2021), no. 3, Paper No. 3.13.arxiv.org/abs/1905.06304
  9. [10]P. J. Grabner, A. Knopfmacher, Analysis of some new partition statistics, Ramanujan J. 12(3) (2006), 439–454.DOI
  10. [11]M. D. Hirschhorn, The power of q. In: Developments in Mathematics,Vol. 49. Springer (2017).
  11. [12]A. E. Ingham, A Tauberian theorem for partitions, Ann. of Math. (2) 42 (1941), 1075–1090.DOI
  12. [13]P. S. Kaur, S. C. Bhoria, P. Eyyunni, B. Maji Minimal excludant over partitions into distinct parts, Int. J. Number Theory, 18 (2022), 2015–2028.
  13. [14]I. Kessler, M. Livingston The expected number of parts in a partition of n, Monatsh. Math. , 81 (1976), 203–212.DOI
  14. [15]N. Koblitz Introduction to Elliptic Curves and Modular Forms, Springer Verlag, New York (1991).
  15. [16]R. da Silva, J. A. Sellers, Parity considerations for mex-related partition functions of Andrews and Newman, J. Integer Seq. 23, no. 5 (2020), Article 20.5.7arxiv.org/abs/2004.02292

Paper details

vlnc:W_5wrh5vzn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents