Development of the Valiron-Levin theorem on the least possible type of entire functions with a given upper \(\rho\)-density of roots (Q906264)

Let \(\sigma_\rho(f)=\limsup\limits _{r\to+\infty} r^{-\rho} \ln M(r,f)\), where \(M(r,f)=\max\{|f(z)|: |z|=r\} \), denote the type of an entire function \(f\) of order \(\rho\). For its zero sequence \(\Lambda=\Lambda_f=(\lambda_n)\), the counting function and the Nevanlinna counting function are defined by \(n_\Lambda(r)=\sum_{|\lambda_n|\leq r} 1\) and \(N_\Lambda(r)=\int_0^r \frac{n_\Lambda(t)}{t} dt\), \(\Lambda\not\ni 0\), respectively. The upper (lower) \(\rho\)-density of the sequence \(\Lambda\) is defined by \(\overline{\Delta}_\rho(\Lambda)=\limsup\limits_{r\to+\infty} r^{-\rho} n_\Lambda(r)\) (\(\underline{\Delta}_\rho(\Lambda)=\liminf\limits_{r\to+\infty} r^{-\rho} n_\Lambda(r)\)), the averaged upper (lower) \(\rho\)-density is defined by \[ \overline{\Delta}^*_\rho(\Lambda)=\limsup\limits_{r\to+\infty} r^{-\rho} N_\Lambda(r)\qquad \big(\underline{\Delta}^*_\rho(\Lambda)=\liminf\limits_{r\to+\infty} r^{-\rho} N_\Lambda(r)\big). \] It is well known [\textit{B. Ya. Levin}, The distribution of the zeros of entire functions (Russian). Moskau: Staatsverlag für technisch-theoretische Literatur (1956; Zbl 0111.07401)] that, for entire functions of order \(\rho\in (0,1)\), \[ \overline{\Delta}^*_\rho(\Lambda_f)\leq \sigma_\rho(f)\leq \frac{\pi \rho}{\sin\pi \rho}\overline{\Delta}^*_\rho(\Lambda_f). \eqno(1) \] The lower bound follows from Jensen's formula and is valid for any positive \(\rho\). \textit{G. Valiron} [Ann. Fac. Sci. Univ. Toulouse, III. Ser. 5, 117--257 (1914; JFM 46.1462.03)] proved the sharp estimate \(\overline{\Delta}_\rho^*(\Lambda)\geq \frac{\overline{\Delta}_\rho(\Lambda)}{e\rho}\), \(\rho>0\), hence \[ \frac{\overline{\Delta}_\rho(\Lambda_f)}{e\rho} \leq \sigma_\rho(f), \quad \rho>0. \eqno(2) \] But the lower bounds in (1) and (2) can be improved under additional restrictions on the zero distribution. \textit{G. G. Braichev} [Dokl. Math. 86, No. 1, 559--561 (2012); translation from Dokl. Akad. Nauk, Ross. Akad. Nauk 445, No. 6, 615--617 (2012; Zbl 1260.30014)] established sharp lower bounds for \(\sigma_\rho(f)\) in terms of averaged densities of \(\Lambda_f\) in the case when \(\rho\in (0,1)\) and all zeros are positive. Theorem A. Let \(f\) be an entire function of order \(\rho\in(0,1)\) with positive zeros, \(\overline{\Delta}_\rho^*(f)=\overline{\Delta}^*>0\), and \(\underline{\Delta}_\rho^*(f)\geq \underline{\Delta}^*>0\). Then the following bound holds \[ \sigma_\rho(f)\geq C^*\Bigl(\frac{\underline{\Delta}^*}{\overline{\Delta}^*}, \rho\Bigr) \overline{\Delta}^*, \eqno (3) \] where \(C^*(\theta, \rho)\) is given implicitly. For any \(\rho\in(0,1)\), there exists an entire function \(f\) with positive zeros such that in (3) equality is attained. Theorem B. Let \(f\) be an entire function of order \(\rho\in(0,1)\) with positive zeros, \(\overline{\Delta}_\rho^*(f)=\overline{\Delta}^*>0\), and \(\underline{\Delta}_\rho^*(f)\geq \underline{\Delta}^*>0\). Then the following sharp bound holds \[ \sigma_\rho(f)\geq \rho e C(\rho)\overline{\Delta}^*_\rho(f), \eqno (4) \] where \(C(\rho)=\max_{a>0} a^{-\rho} \ln (1+a)\). \smallskip Since \(C(\rho)> (\rho e)^{-1}\) [\textit{A. Yu. Popov}, Vestn. Mosk. Univ., Ser. I 2005, No. 1, 31--36 (2005); translation in Mosc. Univ. Math. Bull. 60, No. 1, 32--36 (2005; Zbl 1101.30009)], (4) implies that the lower bound in (1) cannot be attained in the class of entire functions of order \(\rho\in (0,1)\), and with all zeros on a ray. Let \({\mathbb C}_\alpha^\infty\), \(0\leq \alpha\leq 2\pi\), denote the set of all sequences that tend to \(\infty\), and such that, if \(\alpha>0\), then, for any \(\varepsilon\in (0, \alpha)\), there exists an angle of measure \(\alpha-\varepsilon\) with vertex at the origin such that all elements (possibly, except for a finite subset) of the sequence are located outside this angle. The aim of the paper is to improve the lower bound in (1) when \(\Lambda_f\in {\mathbb C}_\alpha^\infty\). Denote \[ C_{\alpha}(\rho;k)=\inf \big\{ \sigma_\rho(f): \overline{\Delta}_\rho(f)=1, \underline{\Delta}_\rho(f)\geq k, \Lambda_f\in {\mathbb C}_\alpha^\infty\big\}, \quad \alpha\in [0, 2\pi], k\in[0,1], \rho>0, \] and \( C_{\alpha}(\rho):=C_{\alpha}(\rho;0)\). It follows from [\textit{R. P. Boas jun.}, Entire functions. New York: Academic Press Inc. (1954; Zbl 0058.30201)] that \(C_0(\rho, k)\geq \frac{e^{k-1}}\rho\), \(0<k\leq 1\). Theorem 2.1 shows that this is sharp. Theorem 2.1. For any \(\rho>0\) and \(k\in [0,1]\), there exists an entire function \(f\) of order \(\rho\) such that: (1) \( \overline{\Delta}_\rho(f)=1\); (2) \(\underline{\Delta}_\rho(f)=k\); (3) \(\sigma_\rho(f)=\overline{\Delta}^*_\rho(f)=\frac{e^{k-1}}\rho\); (4) \(\underline{\Delta}_\rho^*(f)=\frac k\rho\). Set \(\varphi_\rho(t,a)= \frac{1}{2t^\rho} \log (t^2+2at+1)\), \(t>0\), \(0\leq a\leq 1\), \(0<\rho<1\). Theorem 3.1. For any \(\alpha\in [\pi, 2\pi]\) and \(\rho\in (0,1)\), the following relation holds: \[ C_\alpha(\rho)=\max_{t>0} \varphi_\rho(t, -\cos \frac\alpha 2). \] Moreover, the maximum is attained at a function with zeros on the positive ray if \(\alpha=2\pi\), and on the pair of rays \(\arg z =\pm (\pi -\frac\alpha 2)\) if \(\pi\leq \alpha<2\pi\). In particular, \(C_{2\pi}(\rho)=C(\rho)\), \(C_{ \pi}(\rho)=\frac12 C_{2\pi}(\frac \rho2)\). \smallskip It follows from the Lindelöf theorem (see, e.g., Levin's book [loc. cit.]) that \(C_\alpha(\rho)=\infty\) for \(\rho \in \mathbb N\) and \(2\pi-\frac \pi \rho <\alpha\leq 2\pi\). Nevertheless, the following upper estimates for \(C_\alpha(\rho)\) are valid. Theorem 4.1. \[ C_{2\pi} (\rho) \leq 2^{1-\rho} \bigg(1+\max_{r\geq 2} \frac{\log (r-1)}{r}\bigg)< 1.28\cdot 2^{1-\rho}, \quad 1<\rho<2, \] \[ C_{2\pi} (\rho) \leq \frac 2\rho \Bigl(\frac{\rho-1}{\rho} \Bigr)^{\rho-1}<\frac{2^{2-\rho}}\rho,\quad \frac {3}{2}<\rho<2, \] \[ C_{2\pi} (\rho)<1,\quad \rho>2. \]