betaFeedback

WorksMathematics

On sign patterns of Fourier coefficients of theta quotients

Abstract

Motivated by conjectures of Schlosser and Zhou on sign patterns of Fourier coefficients of infinite products, we study two families of theta quotients and their positive integral powers. We determine asymptotic formulas for their Fourier coefficients and use them to obtain explicit eventual periodic sign patterns. Our results provide new instances of the sign phenomena observed by Schlosser and Zhou and illustrate how dominant cusps determine the resulting sign patterns. These results were formalized and verified in Lean by AxiomProver.

Introduction and statement of results

In this paper, for m∈N≥3m \in\mathbb{N}_{\geq3} and ℓ∈Z∖{0}\ell\in\mathbb{Z} \setminus\{0\}, we study the sign pattern of the Fourier coefficients for the quotient

(q,qm−1;qm)∞ℓ(q2,qm−2;qm)∞ℓ.(1)\frac{(q,q^{m-1};q^m)_\infty^\ell}{(q^2,q^{m-2};q^m)_\infty^\ell}. \tag*{(1)}

Here and throughout, ∣q∣<1|q|<1 and for a∈Ca\in\mathbb{C} and n∈N0∪{∞}n\in\mathbb{N}_0\cup\{\infty\}, we set (a)n=(a;q)n:=∏j=0n−1(1−aqj)(a)_n=(a;q)_n:=\prod_{j=0}^{n-1}(1-aq^j). For r∈Nr\in\mathbb{N} and a1,…,ar∈Ca_1,\ldots,a_r\in\mathbb{C}, we let (a1,…,ar)n:=(a1,…,ar;q)n:=(a1)n⋯(ar)n(a_1,\ldots,a_r)_n:=(a_1,\ldots,a_r;q)_n:=(a_1)_n\cdots(a_r)_n. The primary motivation for considering this problem goes back to the sign pattern of Fourier coefficients of the Rogers–Ramanujan continued fraction

R(q):=q151+q1+q21+q31+⋯=q15(q,q4;q5)∞(q2,q3;q5)∞.R(q):=\frac{q^{\frac{1}{5}}}{1+\frac{q}{1+\frac{q^2}{1+\frac{q^3}{1+\cdots}}}}=q^{\frac{1}{5}}\frac{(q,q^4;q^5)_\infty}{(q^2,q^3;q^5)_\infty}.

In this case, using the Circle method, the sign pattern of the nn-th Fourier coefficient for nn sufficiently large was proved by Richmond and Szekeres [16 (3.9)]. Andrews [2 Corollary 1] then gave a proof for the sign pattern of all coefficients via qq-series dissections. Later, using dissections of Ramanujan’s theta functions, Ramanathan [15 Theorem 5 and Theorem 6] and Chan–Yesilyurt [7 Theorem 1.2 and Theorem 1.3] studied periodicity of the sign of Fourier coefficients for a family of infinite products resembling (1). Recently, using similar techniques, Baruah and Sarma [3 Theorem 2 and Theorem 3] investigated the sign pattern of the coefficients of (1) with m=5m=5 and ℓ=±5\ell=\pm5. Using the Circle method, Schlosser and Zhou [17] studied the sign pattern of Fourier coefficients of theta quotients similar to (1) and made several conjectures. Three of the authors [4 Theorem 1.2] settled one of the conjectures of Schlosser and Zhou [17 Conjecture 22]. Other results in this direction can be found in [8 Section 2] and [11].

Returning to (1), for m∈N≥3m \in\mathbb{N}_{\ge3} and ℓ∈N\ell\in\mathbb{N} we write

Fmℓ(q):=(q,qm−1;qm)∞ℓ(q2,qm−2;qm)∞ℓ=:∑n≥0cm[ℓ](n)qn.(2)F_m^{\ell}(q) := \frac{\left(q,q^{m-1};q^m\right)_{\infty}^{\ell}}{\left(q^2,q^{m-2};q^m\right)_{\infty}^{\ell}} =: \sum_{n\ge0} c_m^{[\ell]}(n)q^n. \tag*{(2)}

and study the asymptotic growth and sign pattern of cm[ℓ](n)c_m^{[\ell]}(n) as n→∞n \to\infty.

We first consider the case mm odd. For m=3m=3, we trivially have Fmℓ(q)=1F_m^{\ell}(q)=1 and hence cm[ℓ](n)=0c_m^{[\ell]}(n)=0 for all n∈Nn \in\mathbb{N}. For m∈N>3m \in\mathbb{N}_{>3} odd, define

ϕm:=12(1−2m)(1−3m).(3)\phi_m := \frac{1}{2}\left(1-\frac{2}{m}\right)\left(1-\frac{3}{m}\right). \tag*{(3)}

Our first result is the following:

Theorem 1.1. Let m∈N>3m \in\mathbb{N}_{>3} be odd and ℓ∈N\ell\in\mathbb{N}. Then, as n→∞n \to\infty,

cm[ℓ](n)∼((1−1m)(1−3m)ℓ2m)14(−1)ncos⁡(π(ϕmℓ−nm))eπ2(1−1m)(1−3m)ℓmnn34.c_m^{[\ell]}(n) \sim\left(\frac{\left(1-\frac{1}{m}\right)\left(1-\frac{3}{m}\right)\ell}{2m}\right)^{\frac{1}{4}} \frac{(-1)^n \cos\left(\pi\left(\phi_m\ell-\frac{n}{m}\right)\right)e^{\pi\sqrt{\frac{2\left(1-\frac{1}{m}\right)\left(1-\frac{3}{m}\right)\ell}{m}n}}}{n^{\frac{3}{4}}}.

In particular, for n≫1n \gg1,

sgn⁡(cm[ℓ](n))=(−1)nsgn⁡(cos⁡(π(ϕmℓ−nm))).\operatorname{sgn}\left(c_m^{[\ell]}(n)\right)=(-1)^n\operatorname{sgn}\left(\cos\left(\pi\left(\phi_m\ell-\frac{n}{m}\right)\right)\right).

Next, we explicitly determine the sign in Theorem 1.1. For a property PP, define

χP:={1if P holds,0otherwise.\chi_P := \begin{cases} 1 & \text{if } P \text{ holds}, \\ 0 & \text{otherwise}. \end{cases}

Moreover let

Sm,1[ℓ]:={[m+12−5ℓ2,3m−12−5ℓ2]if ℓ is even,[m−5ℓ−12,2m−5ℓ+12]if ℓ is odd.S_{m,1}^{[\ell]} := \begin{cases} \left[\frac{m+1}{2}-\frac{5\ell}{2},\frac{3m-1}{2}-\frac{5\ell}{2}\right] & \text{if } \ell\text{ is even}, \\ \left[m-\frac{5\ell-1}{2},2m-\frac{5\ell+1}{2}\right] & \text{if } \ell\text{ is odd}. \end{cases}

Corollary 1.2. Let ℓ∈N\ell\in\mathbb{N}, m∈N>3m \in\mathbb{N}_{>3} odd with m>6ℓm>6\ell, and define 0≤r≤2m−10 \le r \le2m-1 via n≡r(mod2m)n \equiv r \pmod{2m}. Then, for n≫1n \gg1,

sgn⁡(cm[ℓ](n))=(−1)r+⌊ℓ/2⌋+χr∈Sm,1[ℓ].\operatorname{sgn}\left(c_m^{[\ell]}(n)\right)=(-1)^{r+\lfloor\ell/2\rfloor+\chi_{r\in S_{m,1}^{[\ell]}}}.

Next, we consider the growth of cm[ℓ](n)c_m^{[\ell]}(n) for m∈N≥4m \in\mathbb{N}_{\ge4} even and ℓ∈N\ell\in\mathbb{N}.

Theorem 1.3. We have, for m∈N≥4m \in\mathbb{N}_{\ge4} even and ℓ∈N\ell\in\mathbb{N}, as n→∞n \to\infty,

cm[ℓ](n)∼ℓ142ℓ+54m14sin⁡ℓ(2πm)(−1)neπ2ℓmnn34.c_m^{[\ell]}(n) \sim\frac{\ell^{\frac{1}{4}}}{2^{\ell+\frac{5}{4}}m^{\frac{1}{4}}\sin^{\ell}\left(\frac{2\pi}{m}\right)}\frac{(-1)^n e^{\pi\sqrt{\frac{2\ell}{m}n}}}{n^{\frac{3}{4}}}.

In this case, we have the following sign-pattern of cm[ℓ](n)c_m^{[\ell]}(n).

Corollary 1.4. We have, for m∈N≥4m \in\mathbb{N}_{\ge4} even, ℓ∈N\ell\in\mathbb{N}, and n∈N0n \in\mathbb{N}_0,

sgn⁡(cm[ℓ](n))=(−1)n.\operatorname{sgn}\left(c_m^{[\ell]}(n)\right)=(-1)^n.

We next turn to ℓ<0\ell<0. Writing ℓ=−λ\ell=-\lambda with λ∈N\lambda\in\mathbb{N}, for m∈N>3m \in\mathbb{N}_{>3} with gcd⁡(m,6)=1\gcd(m,6)=1, set

Fmλ(q):=(q2,qm−2;qm)∞λ(q,qm−1;qm)∞λ=:∑n≥0γm[λ](n)qn.\mathcal{F}_m^{\lambda}(q) := \frac{\left(q^2,q^{m-2};q^m\right)_{\infty}^{\lambda}}{\left(q,q^{m-1};q^m\right)_{\infty}^{\lambda}} =: \sum_{n\ge0}\gamma_m^{[\lambda]}(n)q^n.

If m≡±1(mod6)m \equiv\pm1 \pmod{6} and λ∈N\lambda\in\mathbb{N}, define

Tm[λ](n):=cos⁡(π(λ6−n3±(2n−λ6m+λm2))).T_{m}^{[\lambda]}(n) := \cos\left(\pi\left(\frac{\lambda}{6}-\frac{n}{3}\pm\left(\frac{2n-\lambda}{6m}+\frac{\lambda}{m^2}\right)\right)\right).

We show the following result:

Theorem 1.5. Let m∈N>3m \in\mathbb{N}_{>3} with gcd⁡(m,6)=1\gcd(m,6)=1 and λ∈N\lambda\in\mathbb{N}. Then, as n→∞n \to\infty,

γm[λ](n)∼((1−1m)(1+1m)λ6m)14Tm[λ](n)eπ2(1−1m)(1+1m)λ3mnn34.\gamma_m^{[\lambda]}(n) \sim\left(\frac{\left(1-\frac{1}{m}\right)\left(1+\frac{1}{m}\right)\lambda}{6m}\right)^{\frac{1}{4}} \frac{T_m^{[\lambda]}(n)e^{\pi\sqrt{\frac{2\left(1-\frac{1}{m}\right)\left(1+\frac{1}{m}\right)\lambda}{3m}n}}}{n^{\frac{3}{4}}}.

To state the sign pattern of γm[λ](n)\gamma_m^{[\lambda]}(n) for m≡±1(mod6)m \equiv\pm1 \pmod{6}, we define

Sm,2[λ],±:={[m+12+λ6,3m−16+λ6]if λ≡0(mod6),[(3∓1)m+46+λ−16,(9∓1)m−26+λ−16]if λ≡1(mod6),[(3∓2)m+56+λ−26,(9∓2)m−16+λ−26]if λ≡2(mod6),[λ+36,m+λ−36]if λ≡3(mod6),[(3±2)m+76+λ−46,(9±2)m+16+λ−46]if λ≡4(mod6),[(3±1)m+86+λ−56,(9±1)m+26+λ−56]if λ≡5(mod6).\mathcal{S}_{m,2}^{[\lambda],\pm}:= \begin{cases} \left[\frac{m+1}{2}+\frac{\lambda}{6},\frac{3m-1}{6}+\frac{\lambda}{6}\right] & \text{if }\lambda\equiv0\pmod{6},\\ \left[\frac{(3\mp1)m+4}{6}+\frac{\lambda-1}{6},\frac{(9\mp1)m-2}{6}+\frac{\lambda-1}{6}\right] & \text{if }\lambda\equiv1\pmod{6},\\ \left[\frac{(3\mp2)m+5}{6}+\frac{\lambda-2}{6},\frac{(9\mp2)m-1}{6}+\frac{\lambda-2}{6}\right] & \text{if }\lambda\equiv2\pmod{6},\\ \left[\frac{\lambda+3}{6},m+\frac{\lambda-3}{6}\right] & \text{if }\lambda\equiv3\pmod{6},\\ \left[\frac{(3\pm2)m+7}{6}+\frac{\lambda-4}{6},\frac{(9\pm2)m+1}{6}+\frac{\lambda-4}{6}\right] & \text{if }\lambda\equiv4\pmod{6},\\ \left[\frac{(3\pm1)m+8}{6}+\frac{\lambda-5}{6},\frac{(9\pm1)m+2}{6}+\frac{\lambda-5}{6}\right] & \text{if }\lambda\equiv5\pmod{6}. \end{cases}

Corollary 1.6. Let λ∈N\lambda\in\mathbb{N}, m∈N>3m \in\mathbb{N}_{>3} with m≡±1(mod6)m \equiv\pm1 \pmod{6} and λ<m2\lambda< \frac{m}{2}, and assume that n≡3r+2mj(mod6m)n \equiv3r+2mj \pmod{6m} with 0≤r≤2m−10\le r\le2m-1 and 0≤j≤20\le j\le2. Then, for n≫1n\gg1,

sgn⁡(γm[λ](n))=(−1)r+⌊λ6⌋+χr∈Sm,2[λ],±{1if λ≡0,1,2(mod6),−1if λ≡4,5(mod6),(m3)if λ≡3(mod6).\operatorname{sgn}\left(\gamma_m^{[\lambda]}(n)\right)=(-1)^{r+\left\lfloor\frac{\lambda}{6}\right\rfloor+\chi_{r\in\mathcal{S}_{m,2}^{[\lambda],\pm}}} \begin{cases} 1 & \text{if }\lambda\equiv0,1,2\pmod{6},\\ -1 & \text{if }\lambda\equiv4,5\pmod{6},\\ \left(\frac{m}{3}\right) & \text{if }\lambda\equiv3\pmod{6}. \end{cases}

where (⋅⋅)\left(\frac{\cdot}{\cdot}\right) denotes the Jacobi-symbol.

The paper is organized as follows. In § 3, we state an exact formula of Zuckerman, preliminary results on the Jacobi theta function, and Ingham’s Tauberian theorem. Proofs of Theorem 1.1 and Theorem 1.2 are given in § 5. We show Theorem 1.3 and Theorem 1.4 in § 6. § 7 is devoted to the proof of Theorem 1.5 and Theorem 1.6. Finally, we conclude this paper with § 8 by proposing questions for future research.

Acknowledgements

The first and the second author have received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 101001179). The research of the fourth author was supported by grants from the Research Grants Council of the Hong Kong SAR, China (project numbers HKU 17314122 and HKU 17305923). The authors are grateful to Ken Ono for using AxiomProver to formalize the results of this paper and to verify the proofs in Lean. These formal verifications were carried out by AxiomProver. The relevant files are available in the following repository: https://github.com/bkanemath/Sign-Patterns-of-Fourier-Coefficients-of-Theta-Quotients.

Preliminaries

Exact formulas of Rademacher and Zuckerman

Exact formulas of Rademacher and Zuckerman. We require an exact formula for Fourier coefficients of weakly holomorphic modular functions. Rademacher [13] perfected the Circle Method of Hardy and Ramanujan [10] and obtained an exact formula for the partition function. This was later extended to weakly holomorphic forms of nonpositive weight in results of Rademacher–Zuckerman [14] and Zuckerman [18] and by Ono and the second author [6]. To state the result, let IαI_\alpha denote the usual II-Bessel function. We allow weakly holomorphic modular forms with a multiplier system. Throughout, for a∈Na \in\mathbb{N}, b∈Zb \in\mathbb{Z}, with gcd⁡(a,b)=1\gcd(a,b)=1, we denote by [b]a[b]_a the inverse of bb (mod⁡a)(\operatorname{mod} a).

Theorem 2.1. Let κ≤0\kappa\le0. Suppose that ff is a weakly holomorphic modular form of weight κ\kappa on a subgroup Γ\Gamma of finite index in SL2(Z)\mathrm{SL}_2(\mathbb{Z}). Assume that, for some δ≥0\delta\ge0, f(τ)=∑n≫−∞a(n)qn+δf(\tau)=\sum_{n\gg-\infty}a(n)q^{n+\delta}. For h,k∈N0h,k \in\mathbb{N}_0 with 0≤h<k0 \le h < k and gcd⁡(h,k)=1\gcd(h,k)=1, define fh,kf_{h,k} by

f(1k(h+iz))=(iz)κfh,k(1k([−h]k+iz))(Re⁡(z)>0).f\left(\frac{1}{k}(h+iz)\right)=\left(\frac{i}{z}\right)^\kappa f_{h,k}\left(\frac{1}{k}\left([-h]_k+\frac{i}{z}\right)\right)\qquad(\operatorname{Re}(z)>0).

Write fh,k(τ)=:∑r≫−∞ah,k(r)qr+δh,kckf_{h,k}(\tau)=:\sum_{r\gg-\infty}a_{h,k}(r)q^{\frac{r+\delta_{h,k}}{c_k}}. Here ck∈Nc_k\in\mathbb{N} and δh,k∈[0,1)\delta_{h,k}\in[0,1). Then, we have, for n+δ>0n+\delta>0,

a(n)=2πiκ(n+δ)κ−12∑k≥11k∑0≤h<kgcd⁡(h,k)=1e2πi(n+δ)hk∑r+δh,k<0ah,k(r)e2πick(r+δh,k)[−h]k(∣r+δh,k∣ck)1−κ2I−κ+1(4πk∣r+δh,k∣(n+δ)ck).a(n)=2\pi i^\kappa(n+\delta)^{\frac{\kappa-1}{2}}\sum_{k\ge1}\frac{1}{k}\sum_{\substack{0\le h<k\\ \gcd(h,k)=1}}e^{\frac{2\pi i(n+\delta)h}{k}} \sum_{r+\delta_{h,k}<0}a_{h,k}(r)e^{\frac{2\pi i}{c_k}(r+\delta_{h,k})[-h]_k} \left(\frac{|r+\delta_{h,k}|}{c_k}\right)^{\frac{1-\kappa}{2}} I_{-\kappa+1}\left(\frac{4\pi}{k}\sqrt{\frac{|r+\delta_{h,k}|(n+\delta)}{c_k}}\right).

The Jacobi theta function

The Jacobi theta function. Recall the Jacobi theta function

ϑ(w;τ):=∑n∈Z+12qn22e2πin(w+12)(w∈C, q:=e2πiτ).\vartheta(w;\tau):=\sum_{n\in\mathbb{Z}+\frac{1}{2}}q^{\frac{n^2}{2}}e^{2\pi i n(w+\frac{1}{2})}\qquad(w\in\mathbb{C},\ q:=e^{2\pi i\tau}).

Let z∈Cz\in\mathbb{C} with Re⁡(z)>0\operatorname{Re}(z)>0, k∈Nk\in\mathbb{N}, h∈Zh\in\mathbb{Z} with gcd⁡(h,k)=1\gcd(h,k)=1. Define ωh,k\omega_{h,k} by1

η(1k(h+iz))=eπi12k(h−[−h]k)ωh,k−1z−12η(1k([−h]k+iz)).\eta\left(\frac{1}{k}(h+iz)\right)=e^{\frac{\pi i}{12k}(h-[-h]_k)}\omega_{h,k}^{-1}z^{-\frac{1}{2}}\eta\left(\frac{1}{k}\left([-h]_k+\frac{i}{z}\right)\right).

Moreover, we require the following properties of ϑ(w;τ)\vartheta(w;\tau) from [4 Lemma 2.1 (1), (2), (3)], [19 Proposition 1.3 (7), (8)], and [9].

Lemma 2.2. Suppose that w∈Cw\in\mathbb{C} and τ∈H\tau\in\mathbb{H}.

(1) Suppose that z∈Cz\in\mathbb{C} with Re⁡(z)>0\operatorname{Re}(z)>0, k∈Nk\in\mathbb{N}, and h∈Zh\in\mathbb{Z} with gcd⁡(h,k)=1\gcd(h,k)=1. Then we have

ϑ(w;1k(h+iz))=eπi4k(h−[−h]k)e−3πi4ωh,k−3ize−πkw2zϑ(iwz;1k([−h]k+iz)).\vartheta\left(w;\frac{1}{k}(h+iz)\right)=e^{\frac{\pi i}{4k}(h-[-h]_k)}e^{-\frac{3\pi i}{4}}\omega_{h,k}^{-3}\sqrt{\frac{i}{z}}e^{-\frac{\pi k w^2}{z}}\vartheta\left(\frac{iw}{z};\frac{1}{k}\left([-h]_k+\frac{i}{z}\right)\right).

(2) Suppose that λ,μ∈Z\lambda,\mu\in\mathbb{Z}. Then, with ζ:=e2πiw\zeta:=e^{2\pi i w}, we have

ϑ(w+λτ+μ;τ)=(−1)λ+μq−λ22ζ−λϑ(w;τ).\vartheta(w+\lambda\tau+\mu;\tau)=(-1)^{\lambda+\mu}q^{-\frac{\lambda^2}{2}}\zeta^{-\lambda}\vartheta(w;\tau).

(3) We have

ϑ(w;τ)=−iq18e−πiw∏n≥1(1−qn)(1−e2πiwqn−1)(1−e−2πiwqn).\vartheta(w;\tau)=-iq^{\frac{1}{8}}e^{-\pi i w}\prod_{n\ge1}(1-q^n)(1-e^{2\pi i w}q^{n-1})(1-e^{-2\pi i w}q^n).

(4) We have

ϑ(wτ;−1τ)=−i−iτeπiw2τϑ(w;τ).(4)\vartheta\left(\frac{w}{\tau};-\frac{1}{\tau}\right)=-i\sqrt{-i\tau}e^{\frac{\pi i w^{2}}{\tau}}\vartheta(w;\tau). \tag*{(4)}

(5) For λ∈Q\lambda\in\mathbb{Q}, qλ22ϑ(λτ;τ)q^{\frac{\lambda^{2}}{2}}\vartheta(\lambda\tau;\tau) is a modular form of weight 12\frac{1}{2} (on some congruence subgroup).

Moreover, using [4 Lemma 2.1 (3)], Theorem 2.2 (3), (4), we have the following lemma.

Lemma 2.3. Suppose that τ∈H\tau\in\mathbb{H} and a,b∈Ra,b \in\mathbb{R}.

(1) If 0<a<10<a<1, then we have, as τ→i∞\tau\to i\infty,

ϑ(aτ+b;τ)=−ie−πibq18−a22(1+O(qmin⁡{a,1−a})).\vartheta(a\tau+b;\tau)=-ie^{-\pi ib}q^{\frac{1}{8}-\frac{a^{2}}{2}}\left(1+O\left(q^{\min\{a,1-a\}}\right)\right).

(2) For b∉Zb\notin\mathbb{Z}, we have, as τ→i∞\tau\to i\infty,

ϑ(b;τ)∼−2sin⁡(πb)eπiτ4.\vartheta(b;\tau)\sim-2\sin(\pi b)e^{\frac{\pi i\tau}{4}}.

(3) If 0<b<10<b<1, then we have, as z→0z\to0,

ϑ(aiz+b;iz)∼−e−2πiab+πiae(−b2−14+b)πzz.\vartheta(aiz+b;iz)\sim-\frac{e^{-2\pi iab+\pi ia}e^{\left(-b^{2}-\frac{1}{4}+b\right)\frac{\pi}{z}}}{\sqrt{z}}.

(4) For a∉Za\notin\mathbb{Z}, we have, as z→0z\to0,

ϑ(aiz;iz)∼−2isin⁡(πa)ze−π4z.\vartheta(aiz;iz)\sim-\frac{2i\sin(\pi a)}{\sqrt{z}}e^{-\frac{\pi}{4z}}.

We also require the following asymptotics of the II-Bessel function.

Lemma 2.4. We have, as x→∞x\to\infty,

I1(x)∼ex2πx.I_{1}(x)\sim\frac{e^{x}}{\sqrt{2\pi x}}.

A Tauberian theorem

We require a modified version of Ingham’s Tauberian Theorem [12], stated in [5 Theorem 1.1]. For Δ≥0\Delta\geq0, define

RΔ:={x+iy:x,y∈R, x>0, ∣y∣≤Δx}.R_{\Delta}:=\{x+iy:x,y\in\mathbb{R},\ x>0,\ |y|\leq\Delta x\}.

Proposition 2.5. Let B(q)=∑n≥0b(n)qnB(q)=\sum_{n\geq0}b(n)q^{n} be a power series whose radius of convergence is at least one and assume that the b(n)b(n) are non-negative and weakly increasing. Also suppose that ν,β,γ∈R\nu,\beta,\gamma\in\mathbb{R} with γ>0\gamma>0 exist such that

B(e−t)∼νtβeγtas t→0+,B(e−z)≪∣z∣βeγ∣z∣as z→0,(5)B(e^{-t})\sim\nu t^{\beta}e^{\frac{\gamma}{t}}\quad\text{as }t\to0^{+},\qquad B(e^{-z})\ll|z|^{\beta}e^{\frac{\gamma}{|z|}}\quad\text{as }z\to0, \tag*{(5)}

with the latter condition holding in each region RΔR_{\Delta} for Δ≥0\Delta\geq0. Then we have

b(n)∼νγβ2+142πnβ2+34e2γnas n→∞.b(n)\sim\frac{\nu\gamma^{\frac{\beta}{2}+\frac{1}{4}}}{2\sqrt{\pi}n^{\frac{\beta}{2}+\frac{3}{4}}}e^{2\sqrt{\gamma n}}\quad\text{as }n\to\infty.

Moreover, we require the asymptotic behavior of (q)∞(q)_{\infty} as q→1q\to1.

Lemma 2.6. We have, with q=e−zq=e^{-z},

(q)∞∼2πze−π26zas z→0.(q)_{\infty}\sim\sqrt{\frac{2\pi}{z}}e^{-\frac{\pi^{2}}{6z}}\quad\text{as }z\to0.

Proof of Theorem 1.1 and Theorem 1.2

In this section, we consider the case that mm is odd. For m∈N>3m \in\mathbb{N}_{>3} and ℓ∈N\ell\in\mathbb{N}, we study the behavior of Gmℓ(τ)G_m^\ell(\tau), where

Gm(τ):=ϑ(τ;mτ)ϑ(2τ;mτ)=q12Fm(q),G_m(\tau) := \frac{\vartheta(\tau;m\tau)}{\vartheta(2\tau;m\tau)} = q^{\frac{1}{2}}F_m(q),

as τ→hk∈Q\tau\to\frac{h}{k} \in\mathbb{Q}. For a∈Ra \in\mathbb{R}, we write a=⌊a⌋+{a}a = \lfloor a \rfloor+ \{a\}. We have the following lemma (for general mm). For m∈N>3m \in\mathbb{N}_{>3}, ℓ,k∈N\ell,k \in\mathbb{N}, d:=gcd⁡(m,k)d := \gcd(m,k) with μ:=md\mu:= \frac{m}{d}, and κ:=kd\kappa:= \frac{k}{d}, define

Am,ℓ(h,k):=πℓμκ({μhm}−{μhm}2−{2μhm}+{2μhm}2).A_{m,\ell}(h,k) := \frac{\pi\ell}{\mu\kappa}\left(\left\{\frac{\mu h}{m}\right\}-\left\{\frac{\mu h}{m}\right\}^{2}-\left\{\frac{2\mu h}{m}\right\}+\left\{\frac{2\mu h}{m}\right\}^{2}\right).

Lemma 3.1. Let m∈N>3m \in\mathbb{N}_{>3} and ℓ,k∈N\ell,k \in\mathbb{N}. For 0≤h≤k0 \le h \le k with gcd⁡(h,k)=1\gcd(h,k)=1, we have, as z→0z \to0,

Gmℓ(hk+izk)=eAm,ℓ(h,k)z+Oh,k(1).G_m^\ell\left(\frac{h}{k}+\frac{iz}{k}\right)=e^{\frac{A_{m,\ell}(h,k)}{z}+O_{h,k}(1)}.

Proof. It is enough to prove the claim for ℓ=1\ell=1. Indeed, the general case follows by taking the ℓ\ell-th power. Define

gm(τ):=Gm(τm)=ϑ(τm;τ)ϑ(2τm;τ).(6)g_m(\tau):=G_m\left(\frac{\tau}{m}\right)=\frac{\vartheta\left(\frac{\tau}{m};\tau\right)}{\vartheta\left(\frac{2\tau}{m};\tau\right)}. \tag*{(6)}

Then

Log⁡(gm(1k(h+iz)))≡∑j∈{0,1}(−1)jLog⁡(ϑ(j+1mk(h+iz);1k(h+iz))),(7)\operatorname{Log}\left(g_m\left(\frac{1}{k}(h+iz)\right)\right)\equiv\sum_{j\in\{0,1\}}(-1)^j\operatorname{Log}\left(\vartheta\left(\frac{j+1}{mk}(h+iz);\frac{1}{k}(h+iz)\right)\right), \tag*{(7)}

where, here and throughout, we use ≡\equiv to denote that the identity holds modulo 2πiZ2\pi i\mathbb{Z}. Now by Theorem 2.2 (1) with w=j+1mk(h+iz)w=\frac{j+1}{mk}(h+iz) (j∈{0,1}j\in\{0,1\}), we have, as z→0z\to0,

Log⁡(ϑ(j+1mk(h+iz);1k(h+iz)))≡−Log⁡(z)2−π(j+1)2h2m2kz+Log⁡(ϑ(i(j+1)hmkz−j+1mk;1k([−h]k+iz)))+Oh,k(1).\begin{aligned} \operatorname{Log}\left(\vartheta\left(\frac{j+1}{mk}(h+iz);\frac{1}{k}(h+iz)\right)\right) \\ &\equiv-\frac{\operatorname{Log}(z)}{2}-\frac{\pi(j+1)^2h^2}{m^2kz}+\operatorname{Log}\left(\vartheta\left(\frac{i(j+1)h}{mkz}-\frac{j+1}{mk};\frac{1}{k}\left([-h]_k+\frac{i}{z}\right)\right)\right)+O_{h,k}(1). \end{aligned}

Plugging this into (3.2), we obtain, as z→0z\to0,

Log⁡(gm(1k(h+iz)))=3πh2m2kz+∑j∈{0,1}(−1)jLog⁡(ϑ(ij(j+1)hmkz−j+1mk;1k([−h]k+iz)))+Oh,k(1)=3πh2m2kz+∑j∈{0,1}(−1)jLog⁡(ϑ((j+1)(hmτ1+α1);τ1))+Oh,k(1),(8)\begin{aligned} \operatorname{Log}\left(g_m\left(\frac{1}{k}(h+iz)\right)\right) \\ &=\frac{3\pi h^2}{m^2kz}+\sum_{j\in\{0,1\}}(-1)^j\operatorname{Log}\left(\vartheta\left(\frac{ij(j+1)h}{mkz}-\frac{j+1}{mk};\frac{1}{k}\left([-h]_k+\frac{i}{z}\right)\right)\right)+O_{h,k}(1) \\ &=\frac{3\pi h^2}{m^2kz}+\sum_{j\in\{0,1\}}(-1)^j\operatorname{Log}\left(\vartheta\left((j+1)\left(\frac{h}{m}\tau_1+\alpha_1\right);\tau_1\right)\right)+O_{h,k}(1), \tag*{(8)} \end{aligned}

where

α1:=−h[−h]k+1mkandτ1:=1k([−h]k+iz).\alpha_1:=-\frac{h[-h]_k+1}{mk}\qquad\text{and}\qquad\tau_1:=\frac{1}{k}\left([-h]_k+\frac{i}{z}\right).

By Theorem 2.2 (2) with λ=⌊a⌋\lambda=\lfloor a\rfloor, μ=0\mu=0, w={a}τ+bw=\{a\}\tau+b, we have

ϑ(aτ+b;τ)=(−1)⌊a⌋e−2πi⌊a⌋bq−⌊a⌋22−⌊a⌋{a}ϑ({a}τ+b;τ).\vartheta(a\tau+b;\tau)=(-1)^{\lfloor a\rfloor}e^{-2\pi i\lfloor a\rfloor b}q^{-\frac{\lfloor a\rfloor^2}{2}-\lfloor a\rfloor\{a\}}\vartheta(\{a\}\tau+b;\tau).

Applying this with a=hma=\frac{h}{m}, b=α1b=\alpha_1, and τ=τ1\tau=\tau_1, we obtain,

Log⁡(ϑ((⌊hm⌋+{hm})τ1+α1;τ1))≡−πi⌊hm⌋2τ1−2πi⌊hm⌋{hm}τ1+Log⁡(ϑ({hm}τ1+α1;τ1))+Oh,k(1).(9)\begin{aligned} \operatorname{Log}\left(\vartheta\left(\left(\left\lfloor\frac{h}{m}\right\rfloor+\left\{\frac{h}{m}\right\}\right)\tau_1+\alpha_1;\tau_1\right)\right) \\ &\equiv-\pi i\left\lfloor\frac{h}{m}\right\rfloor^2\tau_1-2\pi i\left\lfloor\frac{h}{m}\right\rfloor\left\{\frac{h}{m}\right\}\tau_1+\operatorname{Log}\left(\vartheta\left(\left\{\frac{h}{m}\right\}\tau_1+\alpha_1;\tau_1\right)\right)+O_{h,k}(1). \tag*{(9)} \end{aligned}

Now, by Theorem 2.3 (1), (2) with a={hm}a=\left\{\frac{h}{m}\right\}, b=α1b=\alpha_1, and τ=τ1\tau=\tau_1, we have,

Log⁡(ϑ({hm}τ1+α1;τ1))≡πi4τ1−πi{hm}τ1+Oh,k(1)(as z→0).\operatorname{Log}\left(\vartheta\left(\left\{\frac{h}{m}\right\}\tau_1+\alpha_1;\tau_1\right)\right)\equiv\frac{\pi i}{4}\tau_1-\pi i\left\{\frac{h}{m}\right\}\tau_1+O_{h,k}(1)\quad\text{(as }z\to0\text{)}.

Plugging this into (9), we obtain, as z→0z\to0,

Log⁡(ϑ((⌊hm⌋+{hm})τ1+α1;τ1))≡πi4τ1+π(h2m2+{hm}−{hm}2)1kz+Oh,k(1).(10)\begin{aligned} \operatorname{Log}\left(\vartheta\left(\left(\left\lfloor\frac{h}{m}\right\rfloor+\left\{\frac{h}{m}\right\}\right)\tau_1+\alpha_1;\tau_1\right)\right) \\ &\equiv\frac{\pi i}{4}\tau_1+\pi\left(\frac{h^2}{m^2}+\left\{\frac{h}{m}\right\}-\left\{\frac{h}{m}\right\}^2\right)\frac{1}{kz}+O_{h,k}(1). \tag*{(10)} \end{aligned}

In the same way, we have, as z→0z\to0,

Log⁡(ϑ((⌊2hm⌋+{2hm})τ1+2α1;τ1))≡πi4τ1+π(4h2m2+{2hm}−{2hm}2)1kz+Oh,k(1).\begin{aligned} \operatorname{Log}\left(\vartheta\left(\left(\left\lfloor\frac{2h}{m}\right\rfloor+\left\{\frac{2h}{m}\right\}\right)\tau_1+2\alpha_1;\tau_1\right)\right) \\ &\equiv\frac{\pi i}{4}\tau_1+\pi\left(\frac{4h^2}{m^2}+\left\{\frac{2h}{m}\right\}-\left\{\frac{2h}{m}\right\}^2\right)\frac{1}{kz}+O_{h,k}(1). \end{aligned}

Plugging this and (10) into (8), we obtain, as z→0z\to0,

Log⁡(gm(1k(h+iz)))≡πkz({hm}−{hm}2−{2hm}+{2hm}2)+Oh,k(1).\operatorname{Log}\left(g_m\left(\frac{1}{k}(h+iz)\right)\right)\equiv\frac{\pi}{kz}\left(\left\{\frac{h}{m}\right\}-\left\{\frac{h}{m}\right\}^2-\left\{\frac{2h}{m}\right\}+\left\{\frac{2h}{m}\right\}^2\right)+O_{h,k}(1).

Exponentiating and using (6) with τ↦mτ\tau\mapsto m\tau, m=dμm=d\mu, and k=dκk=d\kappa, we conclude the proof. □

Next, we maximize ∣Gmℓ(1κ(h+iz))∣\left|G_m^\ell\left(\frac{1}{\kappa}(h+iz)\right)\right|.

Lemma 3.2. For m∈N>3m\in\mathbb{N}_{>3} odd and ℓ∈N\ell\in\mathbb{N}, ∣Gmℓ(1κ(h+iz))∣\left|G_m^\ell\left(\frac{1}{\kappa}(h+iz)\right)\right| is maximal, as z→0z\to0, for hκ∈{m−12m,m+12m}\frac{h}{\kappa}\in\left\{\frac{m-1}{2m},\frac{m+1}{2m}\right\}.

Proof. Again we may assume ℓ=1\ell=1. By Theorem 3.1, as z→0z\to0, ∣Gm(1κ(h+iz))∣\left|G_m\left(\frac{1}{\kappa}(h+iz)\right)\right| is maximal for the maximal value of

{1μκ(3{μhm}2−{μhm})if 0≤{μhm}<12,1μκ(3{μhm}2−5{μhm}+2)if 12≤{μhm}<1.(11)\begin{cases} \frac{1}{\mu\kappa}\left(3\left\{\frac{\mu h}{m}\right\}^{2}-\left\{\frac{\mu h}{m}\right\}\right)&\text{if }0\le\left\{\frac{\mu h}{m}\right\}<\frac{1}{2},\\ \frac{1}{\mu\kappa}\left(3\left\{\frac{\mu h}{m}\right\}^{2}-5\left\{\frac{\mu h}{m}\right\}+2\right)&\text{if }\frac{1}{2}\le\left\{\frac{\mu h}{m}\right\}<1. \tag*{(11)} \end{cases}

Let d∣md\mid m. If d=1d=1, then μ=m\mu=m, κ=k\kappa=k, and {μhm}={h}=0\left\{\frac{\mu h}{m}\right\}=\{h\}=0. So, by Theorem 3.1, Gm(1k(h+iz))=Oh,k(1)G_m\left(\frac{1}{k}(h+iz)\right)=O_{h,k}(1), as z→0z\to0. We next assume d>1d>1. Plugging in μ=md\mu=\frac{m}{d} and κ=kd\kappa=\frac{k}{d}, by (11), we need to maximize (note that dd is odd since mm is odd)

{d2mk(3{hd}2−{hd})if 0≤{hd}<12,d2mk(3{hd}2−5{hd}+2)if 12≤{hd}<1.\begin{cases} \frac{d^2}{mk}\left(3\left\{\frac{h}{d}\right\}^{2}-\left\{\frac{h}{d}\right\}\right)&\text{if }0\le\left\{\frac{h}{d}\right\}<\frac{1}{2},\\ \frac{d^2}{mk}\left(3\left\{\frac{h}{d}\right\}^{2}-5\left\{\frac{h}{d}\right\}+2\right)&\text{if }\frac{1}{2}\le\left\{\frac{h}{d}\right\}<1. \end{cases}
{d2mk(3j2d2−jd)if h∈dN0+j, 0≤j≤⌊d−12⌋,d2mk(3j2d2−5jd+2)if h∈dN0+j, ⌈d+12⌉≤j≤d−1.\begin{cases} \frac{d^2}{mk}\left(\frac{3j^2}{d^2}-\frac{j}{d}\right) & \text{if } h \in d\mathbb{N}_0+j,\ 0 \le j \le\left\lfloor\frac{d-1}{2}\right\rfloor,\\ \frac{d^2}{mk}\left(\frac{3j^2}{d^2}-\frac{5j}{d}+2\right) & \text{if } h \in d\mathbb{N}_0+j,\ \left\lceil\frac{d+1}{2}\right\rceil\le j \le d-1. \end{cases}

Since mm is odd and d∣md \mid m, making the change of variables j↦d−jj \mapsto d-j in the second case, we thus need to maximize

{d2mk(3j2d2−jd)if h∈dN0+j, 0≤j≤d−12,d2mk(3j2d2−jd)if h∈dN−j, 1≤j≤d−12.(12)\begin{cases} \frac{d^2}{mk}\left(\frac{3j^2}{d^2}-\frac{j}{d}\right) & \text{if } h \in d\mathbb{N}_0+j,\ 0 \le j \le\frac{d-1}{2},\\ \frac{d^2}{mk}\left(\frac{3j^2}{d^2}-\frac{j}{d}\right) & \text{if } h \in d\mathbb{N}-j,\ 1 \le j \le\frac{d-1}{2}. \tag*{(12)} \end{cases}

Observe that for d=3d=3 these expressions vanish. So for the remainder of the proof, we assume d>3d>3. For mm and dd fixed the expressions in (3.7) are maximized for j=d−12j=\frac{d-1}{2}. The corresponding contribution is (d−1)(d−3)4mk\frac{(d-1)(d-3)}{4mk}. Moreover, one sees that this is maximal for kk minimal with gcd⁡(m,k)=d\gcd(m,k)=d, which occurs for k=dk=d. The expression we need to maximize in this case is (d−1)(d−3)4md\frac{(d-1)(d-3)}{4md}. As d∣md\mid m, this is maximized for d=md=m. The corresponding values of hh are m−12\frac{m-1}{2} and m+12\frac{m+1}{2} which are clearly coprime to k=mk=m. □

Lemma 3.3. For m∈N>3m\in\mathbb{N}_{>3} odd, ℓ∈N\ell\in\mathbb{N}, and h∈{m−12,m+12}h\in\left\{\frac{m-1}{2},\frac{m+1}{2}\right\}, we have, as z→0z\to0,

Fmℓ(e2πim(h+iz))∼Cm,hℓe(14−1m+34m2)πℓzF_m^\ell\left(e^{\frac{2\pi i}{m}(h+iz)}\right)\sim C_{m,h}^{\ell}e^{\left(\frac{1}{4}-\frac{1}{m}+\frac{3}{4m^2}\right)\frac{\pi\ell}{z}}

with

Cm,h:={−ie5πi2m−3πim2if h=m−12,ie−5πi2m+3πim2if h=m+12.C_{m,h}:= \begin{cases} -ie^{\frac{5\pi i}{2m}-\frac{3\pi i}{m^2}} & \text{if } h=\frac{m-1}{2},\\ ie^{-\frac{5\pi i}{2m}+\frac{3\pi i}{m^2}} & \text{if } h=\frac{m+1}{2}. \end{cases}

Proof. Note that it is enough to prove the case ℓ=1\ell=1. By Theorem 2.2 (3), for m∈N>3m\in\mathbb{N}_{>3} odd, we have, as z→0z\to0,

Fm(e2πim(m−12+iz))∼−ieπimϑ(12−12m+izm;iz)ϑ(1−1m+2izm;iz).F_m\left(e^{\frac{2\pi i}{m}\left(\frac{m-1}{2}+iz\right)}\right)\sim-ie^{\frac{\pi i}{m}}\frac{\vartheta\left(\frac{1}{2}-\frac{1}{2m}+\frac{iz}{m};iz\right)}{\vartheta\left(1-\frac{1}{m}+\frac{2iz}{m};iz\right)}.

Moreover, by definition, c_m^{[1]}(n)\in\mathbb{R}. So, Fm(q‾)=Fm(q)‾F_m(\overline{q})=\overline{F_m(q)}, and thus we only need to prove the claim for m−12\frac{m-1}{2}. Theorem 2.3 (3) finishes the proof. □

Proof of Theorem 1.1. By Theorem 2.2 (5), q−3ℓ2m+ℓ2Fmℓ(q)=q−3ℓ2mGmℓ(τ)q^{-\frac{3\ell}{2m}+\frac{\ell}{2}}F_m^\ell(q)=q^{-\frac{3\ell}{2m}}G_m^\ell(\tau) is a modular function. By Theorem 3.2 its maximal growth occurs at the cusps m−12m\frac{m-1}{2m} and m+12m\frac{m+1}{2m}. Using Theorem 2.4, Theorem 3.3, and (1.3), Theorem 2.1 now gives, as n→∞n\to\infty,

cmℓ(n)∼((1−1m)(1−3m)ℓ2m)14(−1)ncos⁡(π(ϕmℓ−nm))eπ2(1−1m)(1−3m)ℓnmn34.c_m^\ell(n)\sim\left(\frac{\left(1-\frac{1}{m}\right)\left(1-\frac{3}{m}\right)\ell}{2m}\right)^{\frac14}\frac{(-1)^n\cos\left(\pi\left(\phi_m\ell-\frac{n}{m}\right)\right)e^{\pi\sqrt{\frac{2\left(1-\frac{1}{m}\right)\left(1-\frac{3}{m}\right)\ell n}{m}}}}{n^{\frac34}}.

Here we require that the right-hand side does not vanish, which we now prove. Precisely, we need to show that, for n∈Nn\in\mathbb{N} and ℓ∈N\ell\in\mathbb{N},

cos⁡(π(ϕmℓ−nm))≠0.\cos\left(\pi\left(\phi_m\ell-\frac{n}{m}\right)\right)\ne0.

For this, it suffices to prove that ϕmℓ−nm∉Z+12\phi_m\ell-\frac{n}{m}\notin\mathbb{Z}+\frac12. Suppose for contradiction that, for some j∈Zj\in\mathbb{Z},

ϕmℓ−nm=j+12⇒(m−2)(m−3)ℓ−2mn=2m2j+m2,\phi_m\ell-\frac{n}{m}=j+\frac12\Rightarrow(m-2)(m-3)\ell-2mn=2m^2j+m^2,

which cannot hold because the left-hand side is even whereas the right-hand side is odd, as mm is odd. This concludes the proof. □

Proof of Corollary 1.2. By Theorem 1.1, for n≫1n \gg1 with n≡r(mod2m)n \equiv r \pmod{2m},

sgn⁡(cm[ℓ](n))=(−1)rsgn⁡(cos⁡(π(ϕmℓ−rm))).(13)\operatorname{sgn}\left(c_{m}^{[\ell]}(n)\right)=(-1)^{r}\operatorname{sgn}\left(\cos\left(\pi\left(\phi_{m\ell}-\frac{r}{m}\right)\right)\right). \tag*{(13)}

Thus, it suffices to determine sgn⁡(cos⁡(π(ϕmℓ−rm)))\operatorname{sgn}\left(\cos\left(\pi\left(\phi_{m\ell}-\frac{r}{m}\right)\right)\right). We only consider the case ℓ\ell even, the case ℓ\ell odd is shown analogously. Using (3), we write

π(ϕmℓ−rm)=π(ℓ2+3ℓm2−Rm).(14)\pi\left(\phi_{m\ell}-\frac{r}{m}\right)=\pi\left(\frac{\ell}{2}+\frac{3\ell}{m^{2}}-\frac{R}{m}\right). \tag*{(14)}

where

R:=Rr,ℓ:=r+5ℓ2.(15)R:=R_{r,\ell}:=r+\frac{5\ell}{2}. \tag*{(15)}

Using (14), we obtain

cos⁡(π(ϕmℓ−rm))=(−1)ℓ2cos⁡(π(Rm−3ℓm2)).(16)\cos\left(\pi\left(\phi_{m\ell}-\frac{r}{m}\right)\right)=(-1)^{\frac{\ell}{2}}\cos\left(\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)\right). \tag*{(16)}

Now note that by (15), we have

5≤R<5m2.5\le R<\frac{5m}{2}.

Next, we split the range of RR. For 5≤R≤m−125\le R\le\frac{m-1}{2}, we have

0<π(Rm−3ℓm2)<π2.0<\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)<\frac{\pi}{2}.

Thus

sgn⁡(cos⁡(π(Rm−3ℓm2)))=1for5≤R≤m−12.(17)\operatorname{sgn}\left(\cos\left(\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)\right)\right)=1\quad\text{for}\quad5\le R\le\frac{m-1}{2}. \tag*{(17)}

Next, for m+12≤R≤3m−12\frac{m+1}{2}\le R\le\frac{3m-1}{2}, we have

π2<π(Rm−3ℓm2)<3π2.\frac{\pi}{2}<\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)<\frac{3\pi}{2}.

Thus, we obtain

sgn⁡(cos⁡(π(Rm−3ℓm2)))=−1form+12≤R≤3m−12.(18)\operatorname{sgn}\left(\cos\left(\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)\right)\right)=-1\quad\text{for}\quad\frac{m+1}{2}\le R\le\frac{3m-1}{2}. \tag*{(18)}

Finally, for the remaining range of RR, i.e., 3m+12≤R<5m2\frac{3m+1}{2}\le R<\frac{5m}{2}, we have

3π2<π(Rm−3ℓm2)<5π2.\frac{3\pi}{2}<\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)<\frac{5\pi}{2}.

Thus, we obtain

sgn⁡(cos⁡(π(Rm−3ℓm2)))=1for3m+12≤R<5m2.\operatorname{sgn}\left(\cos\left(\pi\left(\frac{R}{m}-\frac{3\ell}{m^{2}}\right)\right)\right)=1\quad\text{for}\quad\frac{3m+1}{2}\le R<\frac{5m}{2}.

Combining this, (17), (18), (16), (15), and (13), we obtain the claim.

Proof of Theorem 1.3 and Theorem 1.4

In this section, we consider the case that mm is even. We first determine the asymptotics of Fmℓ(−e−z)F_m^{\ell}(-e^{-z}) as z→0z \to0. Note that, by the fact mm even and (1.2), we have

Fmℓ(−q)=(−q,−qm−1;qm)∞ℓ(q2,qm−2;qm)∞ℓ=∑n≥0(−1)ncm[ℓ](n)qn.(19)F_m^{\ell}(-q)=\frac{(-q,-q^{m-1};q^m)_{\infty}^{\ell}}{(q^2,q^{m-2};q^m)_{\infty}^{\ell}}=\sum_{n\ge0}(-1)^n c_m^{[\ell]}(n)q^n. \tag*{(19)}

We set

dm[ℓ](n):=(−1)ncm[ℓ](n).(20)d_m^{[\ell]}(n):=(-1)^n c_m^{[\ell]}(n). \tag*{(20)}

Lemma 4.1. For n∈N0n \in\mathbb{N}_0, dm[ℓ](n)≥0d_m^{[\ell]}(n) \ge0 and {dm[ℓ](n)}n≥0\{d_m^{[\ell]}(n)\}_{n\ge0} is weakly increasing.

Proof. From (4.1), it directly follows that dm[ℓ](n)≥0d_m^{[\ell]}(n) \ge0 for n∈N0n \in\mathbb{N}_0. To prove that {dm[ℓ](n)}n≥0\{d_m^{[\ell]}(n)\}_{n\ge0} is weakly increasing, it suffices to show that (1−q)Fmℓ(−q)(1-q)F_m^{\ell}(-q) has non-negative coefficients. By (4.1), we have

(1−q)Fmℓ(−q)=(−qm+1,−qm−1;qm)∞ℓ(1−q)ℓ−1(qm+2,qm−2;qm)∞ℓ(1-q)F_m^{\ell}(-q)=\frac{(-q^{m+1},-q^{m-1};q^m)_{\infty}^{\ell}}{(1-q)^{\ell-1}(q^{m+2},q^{m-2};q^m)_{\infty}^{\ell}}

which has non-negative coefficients. This concludes the proof.

Next, we determine the asymptotics of Fmℓ(−e−z)F_m^{\ell}(-e^{-z}), as z→0z \to0.

Lemma 4.2. We have, as z→0z \to0,

Fmℓ(−e−z)∼eπ2ℓ2mz2ℓsin⁡ℓ(2πm).F_m^{\ell}(-e^{-z}) \sim\frac{e^{\frac{\pi^2\ell}{2mz}}}{2^{\ell}\sin^{\ell}\left(\frac{2\pi}{m}\right)}.

Proof. Note that it is enough to prove the case ℓ=1\ell=1. By (4.1) and Theorem 2.2 (3), we have

Fm(−q)=−ie−πiτ(qm;qm)∞2ϑ(2τ;2mτ)(q2m;q2m)∞ϑ(τ;mτ)ϑ(2τ;mτ).F_m(-q)=-\frac{ie^{-\pi i\tau}(q^m;q^m)_{\infty}^{2}\vartheta(2\tau;2m\tau)}{(q^{2m};q^{2m})_{\infty}\vartheta(\tau;m\tau)\vartheta(2\tau;m\tau)}.

Using this with τ=iz2π\tau=\frac{iz}{2\pi}, Theorem 2.6 and Theorem 2.3 (4) imply the claim.

Proof of Theorem 1.3. By Theorem 4.1, dm[ℓ](n)d_m^{[\ell]}(n) satisfies the monotonicity and non-negativity required to apply Theorem 2.5. Moreover, Theorem 4.2 shows that Fmℓ(−e−z)F_m^{\ell}(-e^{-z}) satisfies (2.1) with ν=12ℓsin⁡ℓ(2πm)\nu=\frac{1}{2^{\ell}\sin^{\ell}\left(\frac{2\pi}{m}\right)}, β=0\beta=0, and γ=π2ℓ2m\gamma=\frac{\pi^2\ell}{2m}. By Theorem 2.5, we have, as n→∞n \to\infty,

dm[ℓ](n)∼ℓ142ℓ+54m14sin⁡ℓ(2πm)eπ2ℓnmn34.d_m^{[\ell]}(n)\sim\frac{\ell^{\frac14}}{2^{\ell+\frac54}m^{\frac14}\sin^{\ell}\left(\frac{2\pi}{m}\right)}\frac{e^{\pi\sqrt{\frac{2\ell n}{m}}}}{n^{\frac34}}.

By (4.2), we conclude the proof.

Proof of Theorem 1.4. The claim follows from Theorem 4.1 and (4.2).

Proof of Theorem 1.5 and Theorem 1.6

For m∈N>3m \in\mathbb{N}_{>3} and λ∈N\lambda\in\mathbb{N}, we study the main growth of Gmλ(τ)G_m^{\lambda}(\tau), where

Gm(τ):=ϑ(2τ;mτ)ϑ(τ;mτ)=q−12Fm(q),G_m(\tau):=\frac{\vartheta(2\tau;m\tau)}{\vartheta(\tau;m\tau)}=q^{-\frac12}F_m(q),

as τ→hk∈Q\tau\to\frac{h}{k} \in\mathbb{Q}. Noting that Gm=Gm−1G_m=G_m^{-1}, Theorem 3.1 implies the following:

Lemma 5.1. Let m∈N>3m \in\mathbb{N}_{>3} and λ,k∈N\lambda,k \in\mathbb{N}. For 0≤h<k0 \le h < k with gcd⁡(h,k)=1\operatorname{gcd}(h,k)=1, d:=gcd⁡(m,k)d := \operatorname{gcd}(m,k) with μ:=md\mu:= \frac{m}{d} and κ:=kd\kappa:= \frac{k}{d}, we have,

Gmλ(hk+izk)=e−Am,λ(h,k)z+Oh,k(1)(as z→0).\mathcal{G}^{\lambda}_{m}\left(\frac{h}{k}+\frac{iz}{k}\right)=e^{-\frac{A_{m,\lambda}(h,k)}{z}+O_{h,k}(1)}\qquad( \text{as } z\to0).

Next, we maximize ∣Gmλ(1k(h+iz))∣\left|\mathcal{G}^{\lambda}_{m}\left(\frac{1}{k}(h+iz)\right)\right|.

Lemma 5.2. For m∈N>3m \in\mathbb{N}_{>3} with m≡±1(mod6)m \equiv\pm1 \pmod{6} and λ∈N\lambda\in\mathbb{N}, ∣Gmλ(1k(h+iz))∣\left|\mathcal{G}^{\lambda}_{m}\left(\frac{1}{k}(h+iz)\right)\right| is maximal, as z→0z \to0, for hk∈{m∓16m,5m±16m}\frac{h}{k} \in\left\{\frac{m\mp1}{6m},\frac{5m\pm1}{6m}\right\}.

Proof. By taking the negative of (3.6) we see that, as z→0z \to0, ∣Gmλ(1k(h+iz))∣\left|\mathcal{G}^{\lambda}_{m}\left(\frac{1}{k}(h+iz)\right)\right| is maximal for the maximal value of

{1μκ(−3{μhm}2+{μhm})if 0≤{μhm}<12,1μκ(−3{μhm}2+5{μhm}−2)if 12≤{μhm}<1.(21)\begin{cases} \frac{1}{\mu\kappa}\left(-3\left\{\frac{\mu h}{m}\right\}^{2}+\left\{\frac{\mu h}{m}\right\}\right) & \text{if } 0\le\left\{\frac{\mu h}{m}\right\}<\frac{1}{2},\\ \frac{1}{\mu\kappa}\left(-3\left\{\frac{\mu h}{m}\right\}^{2}+5\left\{\frac{\mu h}{m}\right\}-2\right) & \text{if } \frac{1}{2}\le\left\{\frac{\mu h}{m}\right\}<1. \tag*{(21)} \end{cases}

Let d∣md\mid m. If d=1d=1, then μ=m\mu=m, κ=k\kappa=k, and {μhm}={h}=0\left\{\frac{\mu h}{m}\right\}=\{h\}=0. So, by Theorem 5.1, Gmλ(1k(h+iz))=Oh,k(1)\mathcal{G}^{\lambda}_{m}\left(\frac{1}{k}(h+iz)\right)=O_{h,k}(1) as z→0z\to0. Next, we assume d>1d>1. Plugging in μ=md\mu=\frac{m}{d} and κ=kd\kappa=\frac{k}{d}, by (5.1), making the change of variables j↦d−jj\mapsto d-j in the second case, thus we need to maximize

{d2mk(−3j2d2+jd)if h∈dN0+j, 0≤j≤d−12,d2mk(−3j2d2+jd)if h∈dN−j, 1≤j≤d−12.(22)\begin{cases} \frac{d^{2}}{mk}\left(-\frac{3j^{2}}{d^{2}}+\frac{j}{d}\right) & \text{if } h\in d\mathbb{N}_{0}+j,\ 0\le j\le\frac{d-1}{2},\\ \frac{d^{2}}{mk}\left(-\frac{3j^{2}}{d^{2}}+\frac{j}{d}\right) & \text{if } h\in d\mathbb{N}-j,\ 1\le j\le\frac{d-1}{2}. \tag*{(22)} \end{cases}

We note that for mm and dd fixed these expressions are maximized if ∣6j−d∣\lvert6j-d\rvert is minimal. Now ∣6j−d∣\lvert6j-d\rvert is minimal for j=d∓16j=\frac{d\mp1}{6}. In this case, (5.2) is maximal for k=dk=d. The expression we need to maximize in this case is d−112m\frac{d-1}{12m}. This is maximized for d=md=m. The maximum value is 112(1−1m2)\frac{1}{12}\left(1-\frac{1}{m^{2}}\right), which is positive for m∈N>3m\in\mathbb{N}_{>3}. Since j=m∓16j=\frac{m\mp1}{6}, we have m=6j±1m=6j\pm1 and hence

gcd⁡(m,j)=gcd⁡(6j±1,j)=1.\operatorname{gcd}(m,j)=\operatorname{gcd}(6j\pm1,j)=1.

Thus, for h∈mN0+jh\in m\mathbb{N}_{0}+j (for the plus sign) and k=mk=m (resp. h∈mN−jh\in m\mathbb{N}-j and k=mk=m for the minus sign), we have gcd⁡(h,k)=1\operatorname{gcd}(h,k)=1. Taking the representative 0≤h<m0\le h<m gives h=j=m∓16h=j=\frac{m\mp1}{6} and h=m−j=5m±16h=m-j=\frac{5m\pm1}{6}. □

Next, we determine the growth of Fmλ(e2πim(h+iz))\mathcal{F}^{\lambda}_{m}\left(e^{\frac{2\pi i}{m}}(h+iz)\right) as z→0z\to0.

Lemma 5.3. For m∈N>3m\in\mathbb{N}_{>3} with m≡±1(mod6)m\equiv\pm1\pmod{6}, λ∈N\lambda\in\mathbb{N}, and h∈{m∓16,5m±16}h\in\left\{\frac{m\mp1}{6},\frac{5m\pm1}{6}\right\}, we have,

Fmλ(e2πim(h+iz))∼Cm,hλe(112−112m2)πλz(as z→0),\mathcal{F}^{\lambda}_{m}\left(e^{\frac{2\pi i}{m}}(h+iz)\right)\sim C^{\lambda}_{m,h}e^{\left(\frac{1}{12}-\frac{1}{12m^{2}}\right)\frac{\pi\lambda}{z}}\qquad( \text{as } z\to0),

where

Cm,h:={eπi6+πi6m+πim2if h=m∓16,e−πi6+πi6m+πim2if h=5m±16.C_{m,h}:= \begin{cases} e^{\frac{\pi i}{6}+\frac{\pi i}{6m}+\frac{\pi i}{m^{2}}} & \text{if } h=\frac{m\mp1}{6},\\ e^{-\frac{\pi i}{6}+\frac{\pi i}{6m}+\frac{\pi i}{m^{2}}} & \text{if } h=\frac{5m\pm1}{6}. \end{cases}

Proof. Again it is enough to consider the case λ=1\lambda=1. Moreover, by definition, γm[1](n)∈R\gamma^{[1]}_{m}(n)\in\mathbb{R}. So, Fm(q‾)=Fm(q)‾\mathcal{F}_{m}(\overline{q})=\overline{\mathcal{F}_{m}(q)} and thus we only need to prove the claim for h=m∓16h=\frac{m\mp1}{6}. Using Theorem 2.2 (3), for h=m∓16h=\frac{m\mp1}{6}, we have, as z→0z\to0,

Fm(e2πim(m∓16+iz))∼eπi6e∓πi6mϑ(13+13m+2izm;iz)ϑ(16+16m+2izm;iz).\mathcal{F}_{m}\left(e^{\frac{2\pi i}{m}\left(\frac{m\mp1}{6}+iz\right)}\right)\sim e^{\frac{\pi i}{6}}e^{\mp\frac{\pi i}{6m}}\frac{\vartheta\left(\frac{1}{3}+\frac{1}{3m}+\frac{2iz}{m};iz\right)}{\vartheta\left(\frac{1}{6}+\frac{1}{6m}+\frac{2iz}{m};iz\right)}.

Now, Theorem 2.3 (3) finishes the proof. □

Proof of Theorem 1.5. By Theorem 2.2 (5), q3λ2m−λ2Fmλ(q)q^{\frac{3\lambda}{2m}}-\frac{\lambda}{2}F_m^{\lambda}(q) is a modular function. By Theorem 5.2 its maximal growth occurs at the cusps m∓16m\frac{m\mp1}{6m} and 5m±16m\frac{5m\pm1}{6m}. Using Theorem 5.3 and Theorem 2.4, Theorem 2.1 now gives, as n→∞n\to\infty,

γm[λ](n)∼((1−1m)(1+1m)λ6m)14Tm[λ](n)eπ2(1−1m)(1+1m)λn3mn34.\gamma_m^{[\lambda]}(n)\sim\left(\frac{\left(1-\frac{1}{m}\right)\left(1+\frac{1}{m}\right)\lambda}{6m}\right)^{\frac{1}{4}}\frac{T_m^{[\lambda]}(n)e^{\pi\sqrt{\frac{2\left(1-\frac{1}{m}\right)\left(1+\frac{1}{m}\right)\lambda n}{3m}}}}{n^{\frac{3}{4}}}.

Here we require that the right-hand side does not vanish, which we now prove. Precisely, for n,λ∈Nn,\lambda\in\mathbb{N} and m∈N>3m\in\mathbb{N}_{>3} with gcd⁡(m,6)=1\gcd(m,6)=1, we need to show

Tm[λ](n)≠0.T_m^{[\lambda]}(n)\ne0.

The proof is analogous to the argument given in Theorem 1.1. □

Proof of Theorem 1.6. By Theorem 1.5, for n≫1n\gg1,

sgn⁡(γm[λ](n))=sgn⁡(Tm[λ](n)).\operatorname{sgn}\left(\gamma_m^{[\lambda]}(n)\right)=\operatorname{sgn}\left(T_m^{[\lambda]}(n)\right).

Since n≡3r+2mj(mod6m)n\equiv3r+2mj\pmod{6m}, we obtain, using that m≡±1(mod6)m\equiv\pm1\pmod{6},

Tm[λ](n)=(−1)rcos⁡(π(λ6±1m(r−λ6+λm))).(23)T_m^{[\lambda]}(n)=(-1)^r\cos\left(\pi\left(\frac{\lambda}{6}\pm\frac{1}{m}\left(r-\frac{\lambda}{6}+\frac{\lambda}{m}\right)\right)\right). \tag*{(23)}

We write λ=6s+a\lambda=6s+a with 0≤a≤50\le a\le5. Since m≡±1(mod6)m\equiv\pm1\pmod{6}, we have

R=Ra:=a±m−16+r−s∈Z.R=R_a:=a\frac{\pm m-1}{6}+r-s\in\mathbb{Z}.

Noting that

λ6±1m(r−λ6+λm)=s±1m(R+λm),\frac{\lambda}{6}\pm\frac{1}{m}\left(r-\frac{\lambda}{6}+\frac{\lambda}{m}\right)=s\pm\frac{1}{m}\left(R+\frac{\lambda}{m}\right),

and using (23), we have,

Tm[λ](n)=(−1)r+scos⁡(π(Rm+λm2)).(24)T_m^{[\lambda]}(n)=(-1)^{r+s}\cos\left(\pi\left(\frac{R}{m}+\frac{\lambda}{m^2}\right)\right). \tag*{(24)}

Since 0<λ<m20<\lambda<\frac{m}{2} by assumption, we have

0<λm2<12m.0<\frac{\lambda}{m^2}<\frac{1}{2m}.

Thus, it is not hard to see that

sgn⁡(cos⁡(π(Rm+λm2)))=(−1)χR (mod⁡2m)∈[m+12,3m−12].(25)\operatorname{sgn}\left(\cos\left(\pi\left(\frac{R}{m}+\frac{\lambda}{m^2}\right)\right)\right)=(-1)^{\chi_{R\,(\operatorname{mod}2m)\in\left[\frac{m+1}{2},\frac{3m-1}{2}\right]}}. \tag*{(25)}

We now apply (24) and (25), distinguishing residue classes of λ\lambda modulo 6.

Case 1: λ≡0(mod6)\lambda\equiv0\pmod{6}. Here a=0a=0 and s=λ6s=\frac{\lambda}{6}. Thus R=r−sR=r-s. We have

R(mod2m)∈[m+12,3m−12]⟺r(mod2m)∈[m+12+λ6,3m−12+λ6].R\pmod{2m}\in\left[\frac{m+1}{2},\frac{3m-1}{2}\right]\Longleftrightarrow r\pmod{2m}\in\left[\frac{m+1}{2}+\frac{\lambda}{6},\frac{3m-1}{2}+\frac{\lambda}{6}\right].

It is not hard to show that this interval lies in [0,2m−1][0,2m-1]. Thus we can skip the condition r(mod2m)r\pmod{2m}. By (24), we conclude the claim in this case.

Case 2: λ≡1(mod6)\lambda\equiv1\pmod{6}. As the proof of this case is analogous to the case λ≡0(mod6)\lambda\equiv0\pmod{6}, we skip it here.

Case 3: λ≡2(mod6)\lambda\equiv2\pmod{6}. Again the proof follows that of the case λ≡0(mod6)\lambda\equiv0\pmod{6} and we therefore skip it here.

Case 4: λ≡3(mod6)\lambda\equiv3 \pmod{6}. Here a=3a=3 and s=λ−36s=\frac{\lambda-3}{6}. Then R=±m−12+r−sR=\frac{\pm m-1}{2}+r-s. We have

R(mod2m)∈[m+12,3m−12]⟺r(mod2m)∈[m+12−±m−12+s,3m−12−±m−12+s].R \pmod{2m} \in\left[\frac{m+1}{2},\frac{3m-1}{2}\right] \Longleftrightarrow r \pmod{2m} \in\left[\frac{m+1}{2}-\frac{\pm m-1}{2}+s,\frac{3m-1}{2}-\frac{\pm m-1}{2}+s\right].

We first assume that m≡1(mod6)m\equiv1\pmod{6}. Then our condition becomes

r(mod2m)∈[λ−36+1,m+λ−36].r \pmod{2m} \in\left[\frac{\lambda-3}{6}+1,m+\frac{\lambda-3}{6}\right].

It is not hard to see that this again lies in [0,2m−1][0,2m-1]. The claim follows in this case as (m3)=1\left(\frac{m}{3}\right)=1.

Next, assume that m≡−1(mod6)m\equiv-1\pmod{6}. Then our condition becomes

r(mod2m)∈[m+s+1,2m+s]=[m+s+1,2m−1]∪[2m,2m+s]≡[m+s+1,2m−1]∪[0,s](mod2m),\begin{aligned} r \pmod{2m} \in[m+s+1,2m+s]=[m+s+1,2m-1]\cup[2m,2m+s] \\ &\equiv[m+s+1,2m-1]\cup[0,s]\pmod{2m}, \end{aligned}

which is the complement of [s+1,m+s][s+1,m+s] in [0,2m−1][0,2m-1]. Thus

(−1)χr∈[m+s+1,2m−1]∪[0,s]=(−1)1+χr∈[s+1,m+s].(-1)^{\chi_{r\in[m+s+1,2m-1]\cup[0,s]}}=(-1)^{1+\chi_{r\in[s+1,m+s]}}.

Noting that (m3)=−1\left(\frac{m}{3}\right)=-1, the claim follows in this case.

Case 5: λ≡4(mod6)\lambda\equiv4\pmod{6}. Here a=4a=4 and s=λ−46s=\frac{\lambda-4}{6}. Then R=2(±m−1)3+r−sR=\frac{2(\pm m-1)}{3}+r-s. Thus we have

R(mod2m)∈[m+12,3m−12]R \pmod{2m} \in\left[\frac{m+1}{2},\frac{3m-1}{2}\right]
⟺r(mod2m)∈[m+12−2(±m−1)3+s,3m−12−2(±m−1)3+s].\Longleftrightarrow r \pmod{2m} \in\left[\frac{m+1}{2}-\frac{2(\pm m-1)}{3}+s,\frac{3m-1}{2}-\frac{2(\pm m-1)}{3}+s\right].

If m≡1(mod6)m\equiv1\pmod{6}, the above condition becomes

r(mod2m)∈[−m6+λ−46+76,5m6+λ−46+16].r \pmod{2m} \in\left[-\frac{m}{6}+\frac{\lambda-4}{6}+\frac{7}{6},\frac{5m}{6}+\frac{\lambda-4}{6}+\frac{1}{6}\right].

Noting that the left endpoint is negative, we split the interval as

[−m6+λ−46+76,−1]∪[0,5m+16+λ−46]≡[0,5m+16+λ−46]∪[11m+76+λ−46,2m−1](mod2m).\begin{aligned} \left[-\frac{m}{6}+\frac{\lambda-4}{6}+\frac{7}{6},-1\right]\cup\left[0,\frac{5m+1}{6}+\frac{\lambda-4}{6}\right] \\ &\equiv\left[0,\frac{5m+1}{6}+\frac{\lambda-4}{6}\right]\cup\left[\frac{11m+7}{6}+\frac{\lambda-4}{6},2m-1\right]\pmod{2m}. \end{aligned}

This is the complement of [5m+76+λ−46,11m+16+λ−46]\left[\frac{5m+7}{6}+\frac{\lambda-4}{6},\frac{11m+1}{6}+\frac{\lambda-4}{6}\right] in [0,2m−1][0,2m-1]. The claim follows in this case.

We next consider the case m≡−1(mod6)m\equiv-1\pmod{6}. Then our condition becomes

r(mod2m)∈[7m+76+λ−46,13m+16+λ−46].(26)r \pmod{2m} \in\left[\frac{7m+7}{6}+\frac{\lambda-4}{6},\frac{13m+1}{6}+\frac{\lambda-4}{6}\right]. \tag*{(26)}

Note that 7m+76+λ−46<2m\frac{7m+7}{6}+\frac{\lambda-4}{6}<2m and 13m+16+λ−46>2m\frac{13m+1}{6}+\frac{\lambda-4}{6}>2m. Thus we write the interval in (26) as

[7m+76+λ−46,2m−1]∪[2m,13m+16+λ−46]≡[0,m+16+λ−46]∪[7m+76+λ−46,2m−1](mod2m).\begin{aligned} \left[\frac{7m+7}{6}+\frac{\lambda-4}{6},2m-1\right]\cup\left[2m,\frac{13m+1}{6}+\frac{\lambda-4}{6}\right] \\ &\equiv\left[0,\frac{m+1}{6}+\frac{\lambda-4}{6}\right]\cup\left[\frac{7m+7}{6}+\frac{\lambda-4}{6},2m-1\right]\pmod{2m}. \end{aligned}

This is the complement of [m+76+λ−46,7m+16+λ−46]\left[\frac{m+7}{6}+\frac{\lambda-4}{6},\frac{7m+1}{6}+\frac{\lambda-4}{6}\right] in [0,2m−1][0,2m-1], yielding the claim in this case.

Case 6: λ≡5(mod6)\lambda\equiv5\pmod{6}. Here a=5a=5 and s=λ−56s=\frac{\lambda-5}{6}. Then R=5(±m−1)6+r−sR=\frac{5(\pm m-1)}{6}+r-s. We have

R(mod2m)∈[m+12,3m−12]R \pmod{2m} \in\left[\frac{m+1}{2},\frac{3m-1}{2}\right]
⟺r(mod2m)∈[m+12−5(±m−1)6+s,3m−12−5(±m−1)6+s].\Longleftrightarrow r \pmod{2m} \in\left[\frac{m+1}{2}-\frac{5(\pm m-1)}{6}+s,\frac{3m-1}{2}-\frac{5(\pm m-1)}{6}+s\right].

We first assume that m≡1(mod6)m \equiv1 \pmod{6}. Then our condition becomes

r(mod2m)∈[4−m3+λ−56,2m+13+λ−56].r \pmod{2m} \in\left[\frac{4-m}{3}+\frac{\lambda-5}{6},\frac{2m+1}{3}+\frac{\lambda-5}{6}\right].

Noting that 4−m3+λ−56<0\frac{4-m}{3}+\frac{\lambda-5}{6}<0, we rewrite the above interval as

[4−m3+λ−56,−1]∪[0,2m+13+λ−56]≡[0,2m+13+λ−56]∪[5m+43+λ−56,2m−1](mod2m).\begin{aligned} \left[\frac{4-m}{3}+\frac{\lambda-5}{6},-1\right]\cup\left[0,\frac{2m+1}{3}+\frac{\lambda-5}{6}\right] \\ &\equiv\left[0,\frac{2m+1}{3}+\frac{\lambda-5}{6}\right]\cup\left[\frac{5m+4}{3}+\frac{\lambda-5}{6},2m-1\right]\pmod{2m}. \end{aligned}

This is the complement of [2m+43+λ−56,5m+13+λ−56]\left[\frac{2m+4}{3}+\frac{\lambda-5}{6},\frac{5m+1}{3}+\frac{\lambda-5}{6}\right] in [0,2m−1][0,2m-1] and this claim follows in this case.

We next assume m≡−1(mod6)m \equiv-1 \pmod{6}. In this case, our condition becomes

r(mod2m)∈[4m+43+λ−56,7m+13+λ−56].r \pmod{2m} \in\left[\frac{4m+4}{3}+\frac{\lambda-5}{6},\frac{7m+1}{3}+\frac{\lambda-5}{6}\right].

Noting that 4m+43+λ−56<2m\frac{4m+4}{3}+\frac{\lambda-5}{6}<2m and 7m+13+λ−56>2m\frac{7m+1}{3}+\frac{\lambda-5}{6}>2m, we rewrite the above interval as

[4m+43+λ−56,2m−1]∪[2m,7m+13+λ−56]≡[0,m+13+λ−56]∪[4m+43+λ−56,2m−1](mod2m).\begin{aligned} \left[\frac{4m+4}{3}+\frac{\lambda-5}{6},2m-1\right]\cup\left[2m,\frac{7m+1}{3}+\frac{\lambda-5}{6}\right] \\ &\equiv\left[0,\frac{m+1}{3}+\frac{\lambda-5}{6}\right]\cup\left[\frac{4m+4}{3}+\frac{\lambda-5}{6},2m-1\right]\pmod{2m}. \end{aligned}

This is the complement of [m+43+λ−56,4m+13+λ−56]\left[\frac{m+4}{3}+\frac{\lambda-5}{6},\frac{4m+1}{3}+\frac{\lambda-5}{6}\right] in [0,2m−1][0,2m-1] and the claim follows in this case. □

Open questions

The results of this paper suggest several directions for future research.

(1) It would be interesting to extend Theorems 1.1, 1.3, and 1.5, by considering theta quotients of the form (for a,b,m,ℓ∈Na,b,m,\ell\in\mathbb{N} with a≠ba \ne b)

ϑ(aτ;mτ)ℓϑ(bτ;mτ)ℓ.\frac{\vartheta(a\tau;m\tau)^\ell}{\vartheta(b\tau;m\tau)^\ell}.

For which choices of aa, bb, mm, and ℓ\ell do the Fourier coefficients have an eventual periodic sign pattern? Can this sign pattern be determined explicitly? More generally, can one characterize the dominant cusps in terms of aa, bb, and mm?

(2) In Theorem 1.5, we assume gcd⁡(m,6)=1\gcd(m,6)=1. Can this assumption be removed? More precisely, how do the dominant cusps and the resulting sign pattern change if 2∣m2\mid m or 3∣m3\mid m?

(3) Our results raise a more general question: under what conditions do the Fourier coefficients of a modular theta quotient have an eventually periodic sign pattern?

Declarations

Data Availability. Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

Conflicts of interest. The authors declare that there are no conflicts of interest.

Funding. Funding declarations can be found in the Acknowledgements Section.

Numerics

In the following tables, we list the sign patterns of cm[ℓ](n)c_{m}^{[\ell]}(n) and γm[ℓ](n)\gamma_{m}^{[\ell]}(n) for m∈{5,7}m \in\{5,7\}, 1≤ℓ≤61 \le\ell\le6, and 70≤n≤m+6970 \le n \le m+69.

(m,ℓ)(m,\ell)(5,1)(5,1)(5,2)(5,2)(5,3)(5,3)(5,4)(5,4)(5,5)(5,5)
sgn⁡(cm[ℓ](n))\operatorname{sgn}\left(c_{m}^{[\ell]}(n)\right)+−+−−+ - + - -+−+−−+ - + - -+−+−++ - + - ++−+−++ - + - +−−+−+- - + - +
sgn⁡(γm[ℓ](n))\operatorname{sgn}\left(\gamma_{m}^{[\ell]}(n)\right)++−−−+ + - - -+++−−+ + + - -+++−−+ + + - -−++−−- + + - -−+++−- + + + -

Table 1.

(m,ℓ)(m,\ell)(5,6)(5,6)(7,1)(7,1)(7,2)(7,2)(7,3)(7,3)
sgn⁡(cm[ℓ](n))\operatorname{sgn}\left(c_{m}^{[\ell]}(n)\right)−++−+- + + - ++−+−+++ - + - + ++−+−+−++ - + - + - +−−+−++−- - + - + + -
sgn⁡(γm[ℓ](n))\operatorname{sgn}\left(\gamma_{m}^{[\ell]}(n)\right)−−++−- - + + -+++−−−++ + + - - - ++++−−−−+ + + - - - -++++−−−+ + + + - - -

Table 1.

(m,ℓ)(m,\ell)(7,4)(7,4)(7,5)(7,5)(7,6)(7,6)
sgn⁡(cm[ℓ](n))\operatorname{sgn}\left(c_{m}^{[\ell]}(n)\right)−+−−+−+- + - - + - +−+−++−+- + - + + - +−+−+−++- + - + - + +
sgn⁡(γm[ℓ](n))\operatorname{sgn}\left(\gamma_{m}^{[\ell]}(n)\right)−++++−−- + + + + - -−−+++−−- - + + + - -−−++++−- - + + + + -

Table 1. Sign patterns for cm[ℓ](n)c_{m}^{[\ell]}(n) and γm[ℓ](n)\gamma_{m}^{[\ell]}(n).

University of Cologne, Department of Mathematics and Computer Science, Weyertal 86-90, 50931 Cologne, Germany

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

The University of Hong Kong, Department of Mathematics, Pokfulam, Hong Kong

Email address: [email protected]

References

  1. [1]G. Andrews, The theory of partitions, Cambridge University Press, 1980.
  2. [2]G. Andrews, Ramanujan’s “Lost” Notebook III. The Rogers–Ramanujan continued fraction, Adv. Math. 41 (1981), 186–208.DOI
  3. [3]N. Baruah and A. Sarma, Sign patterns and congruences of certain infinite products involving the Rogers–Ramanujan continued fraction, Ramanujan J. 67 (2025), 1–25.arxiv.org/abs/2503.08517
  4. [4]K. Bringmann, B. Heim, and B. Kane, On a sign-change conjecture of Schlosser and Zhou, J. Math. Anal. Appl. 548 (2025), p. 129339.arxiv.org/abs/2406.03453
  5. [5]K. Bringmann, C. Jennings-Shaffer, and K. Mahlburg, On a Tauberian theorem of Ingham and Euler–Maclaurin summation, Ramanujan J. 61 (2021), 55–86.
  6. [6]K. Bringmann and K. Ono, Coefficients of harmonic Maass forms, Proceedings of the 2008 University of Florida Conference on partitions, q-series, and modular forms, 2008.
  7. [7]S. Chan and H. Yesilyurt, The periodicity of the signs of the coefficients of certain infinite product, Pacific J. Math. 225 (2006), 13–32.DOI
  8. [8]S. Chern, Asymptotics for the Taylor coefficients of certain infinite products, Ramanujan J. 55 (2021), 987–1014.arxiv.org/abs/1902.10839
  9. [9]M. Eichler and D. Zagier, The theory of Jacobi forms, Progress in Math. 55, Birkhäuser, 1985.DOI
  10. [10]G. Hardy and S. Ramanujan, Asymptotic formulae in combinatory analysis, Proc. London Math. Soc. (2) 17 (1918), 75–115.
  11. [11]B. He and L. Li, Some conjectures of Schlosser and Zhou on sign patterns of the coefficients of infinite products, arXiv:2509.10023.
  12. [12]A. Ingham, A Tauberian theorem for partitions, Ann. Math. 42 (1941), 1075–1090.
  13. [13]H. Rademacher, On the partition function p(n), Proc. Lond. Math. Soc. (3) 43 (1937), 241–254.DOI
  14. [14]H. Rademacher and H. Zuckerman, On the Fourier coefficients of certain modular forms of positive dimension Ann. Math. 39 (1938), 433–462.DOI
  15. [15]K. Ramanathan, Generalisations of some theorems of Ramanujan, J. Number Theory 29 (1988), 118–137.DOI
  16. [16]B. Richmond and G. Szekeres, The Taylor coefficients of certain infinite products, Acta Sci. Math. (Szeged) 40 (1978), 347–369.
  17. [17]M. Schlosser and N. Zhou, On the infinite Borwein product raised to a positive real power, Ramanujan J. 61 (2023), 515–543.
  18. [18]H. Zuckerman, On the coefficients of certain modular forms belonging to subgroups of the modular group, Trans. Amer. Math. Soc. 45 (1939), 298–321.DOI
  19. [19]S. Zwegers, Mock theta functions, Ph.D. Thesis, Universiteit Utrecht (2002).

Paper details

vlnc:W_8uumc5e4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

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