Further results on the exponent of convergence of zeros of solutions of certain higher order linear differential equations (Q1932161)

Estimates of the growth of solutions for the linear differential equations \[ f^{(k)}+ a_{k-1}(z)f^{(k-1)}+\dotsb +a_{0}(z)f=0, \eqno(*) \] \(a_j\), \(j\in \{0, \dotsc, k-1\}\), being entire functions, are well-known. [\textit{I. Laine}, Nevanlinna theory and complex differential equations. Berlin: W. de Gruyter (1992; Zbl 0784.30002)]. For an entire function \(f\) with the zero sequence \((z_n)\), let \(\lambda(f)=\inf \{ \alpha>0: \sum_n |z_n|^{-\alpha}<\infty\}\) be the convergence exponent of its zeros. It is well-known that \(\lambda(f)\leq \sigma(f)\), where \(\sigma(f)\) denotes the order of \(f\). \textit{J. Heittokangas} and \textit{J. Rättyä} [Math. Nachr. 284, No. 4, 412--420 (2011; Zbl 1220.34109)] recently proved the following result. {Theorem A.} Let \(\lambda \geq 1\), and suppose that the coefficients \(a_j\), \(j\in \{0, \dotsc, k-2\}\) are entire, and \(a_{k-1}\equiv 0\). Then \(a_j\), \(j\in \{0, \dotsc, k-2\}\) are polynomials satisfying \(\deg a_j \leq (k-j)(\rho-1) \) for all \(j\in \{0, \dotsc, k-2\}\) if and only if all solutions of (2) satisfy \(\lambda(f)\leq \lambda\). Analogous result is proved for the case when the coefficients \(a_j\) are analytic in the unit disc \(D\). But lower estimates of \(\lambda(f)\), where \(f\) is a solution of (*), is more delicate in general. \textit{S. B. Bank} and \textit{J. K. Langley} [Proc. Edinb. Math. Soc., II. Ser. 30, 455--469 (1987; Zbl 0596.30049)] established conditions under which all non-trivial solutions of the equation \[ y^{(k)}+(\Pi (z)e^{P(z)}+Q(z))y=0, \] where \(P\), \(Q\) are polynomials, \(\Pi\) is an entire function, satisfy \(\lambda(f)=\infty\). \textit{K. Ishizaki} [Hokkaido Math. J. 26, No. 2, 421--434 (1997; Zbl 0885.34003)], \textit{K. Ishizaki} and \textit{K. Tohge} [J. Math. Anal. Appl. 206, No. 2, 503--517 (1997; Zbl 0877.34009)] studied the differential equation of the form \[ f''+ (e^{P_1(z)}+ e^{P_2(z)}+Q_0(z))f=0,\eqno(1) \] where \(P_1(z)=\zeta_1 z^n+ \dots\), \(P_2(z)=\zeta_2 z^m+ \dots\), \(\zeta_1 \zeta_2 \neq 0\) are non-constant polynomials, \(Q_0\) is an entire function of order less than \(\max\{n,m\}\), and \(e^{P_1(z)}\) and \(e^{P_2(z)}\) are linearly independent. They proved the following. {Theorem B.} Suppose that \(m=n\), and that \(\zeta_1\neq \zeta_2\) in (1). If \(\zeta_1/\zeta_2\) is non-real, then, for any non-trivial solution of (1), we have \(\lambda(f)=\infty\). { Theorem C.} Suppose that \(m=n\), and that \(\zeta_1/\zeta_2=\rho>0\) in (1). If \(0<\rho<\frac 12\) or \(Q_0(z)\equiv 0\), \(\frac 34<\rho<1\), then, for any non-trivial solution of (1), we have \(\lambda(f)\geq n\). In the present paper, the authors consider exponents of convergence for solutions of (*) when \(a_0\) is of a special form. { Theorem 1.1. } Let \(a_0(z)=\sum_{j=1}^l Q_j(z)e^{P_j(z)}\), \(l\geq 3\), where \(P_j(z)=\zeta_j z^n+\dots\), \(1\leq j\leq l\), are polynomials of degree \(n\), \( a_m\), \(Q_m\), \(1\leq m\leq k-1\), are entire functions of order less than \(n\), \(k\geq 2\). (i) If \(\zeta_1/\zeta_2\) is non-real, \(0<\lambda_j=\frac{\zeta_j}{\zeta_2}<\frac 1k\), \(3\leq j\leq l\), then, for any non-trivial solution \(f\) of (*), we have \(\lambda(f)=\infty\). (ii) If \(0<\zeta_1/\zeta_2< \frac 1{2k} \), \(0<\lambda_j\) and \(\sum_{j=3}^l \lambda_j <1\), then, for any non-trivial solution \(f\) of (*), we have \(\lambda(f)\geq n\). This generalizes some previous results of the second author and others, (cf. [\textit{J. Tu} and \textit{X.-D. Yang}, Kodai Math. J. 33, No. 2, 251--266 (2010; Zbl 1203.30033)]).