1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %%older versions at july15.tex,aug10.tex,aug16.tex

3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

4: \documentclass[12pt]{amsart}

5: %\documentclass[reqno,12pt]{amsart}

6: \usepackage{graphicx,amsmath,amssymb}

7: 

8: %\usepackage{showkeys}

9: 

10: \headheight=6.15pt

11: \textheight=9.25in

12: \textwidth=6.5in

13: \oddsidemargin=0in

14: \evensidemargin=0in

15: \topmargin=-.375in

16: 

17: \numberwithin{equation}{section}

18: 

19: \newcommand{\C}{\mathbb C}

20: \newcommand{\R}{\mathbb R}

21: \newcommand{\Z}{\mathbb Z}

22: \newcommand{\N}{\mathbb N}

23: %Derivatives:

24: %%

25: \renewcommand{\d}{\prime} 

26: \newcommand{\dd}{{\prime \prime}} 

27: \newcommand{\ddd}{{\prime \prime \prime}} 

28: \newcommand{\dddd}{{\prime \prime \prime \prime}} 

29: \renewcommand{\Re}{{\rm Re}\,}

30: \renewcommand{\Im}{{\rm Im}\,}

31: \newtheorem{theorem}{Theorem}[section]

32: \newtheorem{lemma}[theorem]{Lemma}

33: \newtheorem{proposition}[theorem]{Proposition} 

34: \newtheorem{corollary}[theorem]{Corollary}

35: \newtheorem{untheorem}{Theorem}

36: \newtheorem{unlemma}{Lemma}

37: \newtheorem{conjecture}{Conjecture}

38: \newtheorem{claim}{Claim}

39: \newtheorem{problem}{Problem}

40: \newtheorem{definition}{Definition}

41: \newtheorem*{remark}{Remark}

42: \newtheorem*{remarks}{Remarks}

43: \renewcommand{\theuntheorem}{\hspace*{-4pt}}

44: \renewcommand{\theunlemma}{\hspace*{-4pt}}

45: \renewcommand{\theconjecture}{\hspace*{-4pt}}

46: \renewcommand{\theclaim}{\hspace*{-4pt}}

47: \renewcommand{\theproblem}{\hspace*{-4pt}}

48: \renewcommand{\thedefinition}{\hspace*{-4pt}}

49: \renewcommand{\thefootnote}{}

50: 

51: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

52: 

53: \begin{document}

54: \title[]

55: {Schr\"odinger type eigenvalue problems  with polynomial potentials: Asymptotics of  eigenvalues}

56: \author[]

57: {Kwang C. Shin}

58: \address{Department of Mathematics, University of Missouri, Columbia, MO 65211, USA}

59: \date{November 4, 2004}

60: 

61: \begin{abstract}

62: For integers $m\geq 3$ and $1\leq\ell\leq m-1$, we study the eigenvalue problem $-u^\dd(z)+[(-1)^{\ell}(iz)^m-P(iz)]u(z)=\lambda u(z)$ with the boundary conditions that $u(z)$ decays to zero as $z$ tends to infinity along the rays $\arg z=-\frac{\pi}{2}\pm \frac{(\ell+1)\pi}{m+2}$ in the complex plane, where $P(z)=a_1 z^{m-1}+a_2 z^{m-2}+\cdots+a_{m-1} z$ is a polynomial. We provide asymptotic expansions of the eigenvalue counting function and the eigenvalues $\lambda_{n}$. Then we apply these to the inverse spectral problem, reconstructing some coefficients of polynomial potentials from asymptotic expansions of the eigenvalues. Also, we show for arbitrary $\mathcal{PT}$-symmetric polynomial potentials of degree $m\geq 3$ and all symmetric decaying boundary conditions that the eigenvalues are all real and positive, with only finitely many exceptions.

63: \end{abstract}

64: 

65: \maketitle

66: 

67: \begin{center}

68: {\it Preprint.}

69: %{\it To appear in: When the paper has been accepted, the Journal Title goes here}

70: \end{center}

71: 

72: \baselineskip = 18pt

73: 

74: %

75: 

76: 

77: \section{Introduction}

78: \label{introduction}

79: In this paper, we study  Schr\"odinger operators with any polynomial potential of degree $m\geq 3$ with complex coefficients, under   decaying boundary conditions along two rays to infinity in the complex plane, and provide asymptotic expansions of the eigenvalue counting functions and the eigenvalues. Then we will use these to  reconstruct some coefficients of polynomial potentials from asymptotic expansions of the eigenvalues,  and to show that all but finitely many eigenvalues of every $\mathcal{PT}$-symmetric oscillator with  a polynomial potential are real and positive.

80: 

81: For integers $m\geq 3$ fixed and $1\leq \ell\leq m-1$, we are considering the non-standard eigenvalue problems

82: \begin{equation}\label{ptsym}

83: H_{\ell,P} u(z,\lambda):=\left[-\frac{d^2}{dz^2}+(-1)^{\ell}(iz)^m-P(iz)\right]u(z,\lambda)=\lambda u(z,\lambda),\quad\text{for some $\lambda\in\C$},

84: \end{equation}

85: with the boundary condition that 

86: \begin{equation}\label{bdcond}

87: \text{$u(z,\lambda)\rightarrow 0$ exponentially, as $z\rightarrow \infty$ along the two rays}\quad \arg z=-\frac{\pi}{2}\pm \frac{(\ell+1)\pi}{m+2},

88: \end{equation}

89: where $P$ is a polynomial of degree at most $m-1$ of the form 

90: \begin{equation}\nonumber

91: P(z)=a_1z^{m-1}+a_2z^{m-2}+\cdots+a_{m-1}z,\quad a_j\in\C\,\,\text{\,for $1\leq j\leq m-1$}.

92: \end{equation}

93: 

94: The anharmonic oscillators $H_{\ell,P}$ with the various boundary conditions \eqref{bdcond} are considered in \cite{Bender2,Shin2}. The most studied case is 

95: when $m$ is even and $\ell=\frac{m}{2}$, for example, see \cite{AA,Bender-1,Bender3,HR,HMM,MAS,Simon}. In this case, $H_{\ell,P}$ is an Schr\"odinger operator in $L^2(\R)$. This is self-adjoint if all the coefficients of the polynomial $(-1)^{\ell}(iz)^m-P(iz)$ in $z$ are real, and non-self-adjoint if a coefficient of $(-1)^{\ell}(iz)^m-P(iz)$ is non-real. The case when $\ell=1$ has been studied extensively in recent years in the context of $\mathcal{PT}$-symmetric theory \cite{Bender,Dorey,Handy2,Shin,Sibuya}. 

96: 

97: 

98: Throughout this paper, we use the integer $m\geq 3$ for the degree of the polynomial potential and integer $\ell$ with  $1\leq\ell\leq m-1$. We will use $\lambda_n$, depending on the potential and the boundary condition, to denote  the eigenvalues of $H_{\ell,P}$, without explicitly indicating their dependence on the potential and the boundary condition. 

99: Also, we let $$a:=(a_1,a_2,\ldots, a_{m-1})\in\C^{m-1}$$ be the coefficient vector of $P(z)$. 

100: 

101: If a nonconstant function $u$ satisfies \eqref{ptsym} with some $\lambda\in\C$ and the boundary condition \eqref{bdcond}, then we call $\lambda$ an {\it eigenvalue} of $H_{\ell,P}$ and $u$ an {\it eigenfunction of $H_{\ell,P}$ associated with the eigenvalue $\lambda$}. Also, the {\it geometric multiplicity of an eigenvalue $\lambda$} is the number of linearly independent eigenfunctions  associated with the eigenvalue $\lambda$. For each $\ell$ fixed, we number the eigenvalues $\{\lambda_{n}\}_{n\geq n_0}$ in the order of nondecreasing magnitudes, counting their ``algebraic multiplicities'', where the integer $n_0$, depending on the potential and the boundary condition, is due to our method of proof of Theorem \ref{main_thm1}. In Theorem \ref{main_thm1} we show that for every large $n\in\N$, there exists $\lambda_n$ satisfying  \eqref{main_result} below. However, we do not know the number of eigenvalues ``near'' zero, and this is why we need to have $n_0$ in numbering the eigenvalues.

102: 

103: 

104: 

105: Before we state our main theorems, we first introduce some known facts by Sibuya \cite{Sibuya} about the eigenvalues $\lambda_n$ of $H_{\ell,P}$.

106: \begin{theorem}\label{main2}

107: The eigenvalues $\lambda_{n}$ of $H_{\ell,P}$ have the following properties.

108: \begin{enumerate}

109: \item[(I)] The set of all eigenvalues is a  discrete set in $\C$.

110: \item[(II)] The geometric multiplicity of every eigenvalue is one.

111: \item[(III)] Infinitely many eigenvalues, accumulating at infinity, exist.

112: \item[(IV)] When $\ell=1$ the eigenvalues have the following asymptotic expansion

113: \begin{equation}\label{Bender_exp}

114: \lambda_{n} =\left(\frac{\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)\sqrt{\pi}\left(n+\frac{1}{2}\right)}{\sin \left(\frac{\pi}{m}\right)\Gamma\left(1+\frac{1}{m}\right) } \right)^{\frac{2m}{m+2}}[1+o(1)]\quad\text{as $n$ tends to infinity},\quad n \in \N.

115: \end{equation}

116: \end{enumerate} 

117: \end{theorem}

118: 

119: This paper contains results on direct and inverse spectral probelms, and their applications to $\mathcal{PT}$-symmetric potential problems. Theorem \ref{main_thm1} below is the main result, regarding asymptotic expansions of  ``eigenvalue counting functions''. The other results below in the Introduction are  deduced from   Theorem \ref{main_thm1}.

120: 

121: \subsection*{Direct spectral problem}

122: Here,

123: we first introduce the following theorem, regarding asymptotic expansions of a kind of eigenvalue counting functions, where we use multi-index notations with

124: $$\alpha=(\alpha_1,\alpha_2,\dots,\alpha_{m-1})\in\left(\N\cup\{0\}\right)^{m-1},\quad \text{and}\,\,\, \beta=(1,2,\dots,{m-1}).$$  Also, we use $|\alpha|=\alpha_1+\alpha_2+\cdots+\alpha_{m-1}$, $\alpha!= \alpha_1!\alpha_2!\cdots\alpha_{m-1}!$ and $a^{\alpha}=a_1^{\alpha_1}a_2^{\alpha_2}\cdots a_{m-1}^{\alpha_{m-1}}$.

125: \begin{theorem}\label{main_thm1}

126: For  $a\in\C^{m-1}$, the eigenvalues $\lambda_n$ of $H_{\ell,P}$ satisfy

127: \begin{equation}\label{main_result}

128: \left(2n+1\right)\pi i=\sum_{j=0}^{\lfloor\frac{m+2}{2}\rfloor}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}-\frac{j-1}{m}}+o(1),\quad\text{as $n\to\infty$},

129: \end{equation}

130: where $\lfloor x\rfloor$ is the largest integer that is less than or equal to $x$, and where

131:  the error $o(1)$ term is uniform on any compact set of $a\in\C^{m-1}$, and 

132: \begin{equation}\label{dmlj_def}

133: d_{\ell,j}(a)=\left\{

134:                     \begin{array}{cl}

135:                     2i\sqrt{\pi}\sin\left(\frac{\ell\pi}{m}\right)\frac{\Gamma\left(1+\frac{1}{m}\right)}{\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)} \,\,\, &\text{if $j=0$,}\\

136: &\\

137:                    -4i\sum_{k=1}^j(-1)^{(\ell+1)k}K_{m,j,k}b_{j,k}(a)\sin\left(\frac{(j-1)\ell\pi}{m}\right)\cos\left(\frac{(j-1)\pi}{m}\right)\,\,\, &\text{if $1\leq j\leq\frac{m+1}{2}$,}\\

138: &\\

139: \eta_{m,\ell}(a)\,\,\, &\text{if $j=\frac{m+2}{2}$,}\\

140:                     \end{array}\right.

141: \end{equation}

142: where 

143: \begin{equation}\label{bjk_def}

144: b_{j,k}(a)={\frac{1}{2}\choose{k}}\sum_{\substack{|\alpha|=k\\ \alpha\cdot\beta=j}}\frac{k!}{\alpha!}\,a^{\alpha},\quad 1\leq k\leq j\leq \frac{m+2}{2},

145: \end{equation}

146: \begin{equation}\nonumber

147: \eta_{m,\ell}(a)=\left\{

148:                     \begin{array}{cl}

149:                     (-1)^{\frac{\ell-1}{2}}\frac{4\pi i}{m}\sum_{k=1}^{\frac{m+2}{2}}b_{\frac{m+2}{2},k}(a) \quad &\text{if $m$ is even and $\ell$ is odd,}\\

150: &\\

151:                    0\quad &\text{ otherwise,}

152:                     \end{array}\right.

153: \end{equation} 

154: and

155: \begin{equation}\nonumber

156: K_{m,j,k}

157: =\left\{

158:                     \begin{array}{cl}

159: -\frac{2}{m}

160: \quad &\text{if $j=k=1$},\\

161: &\\

162:                   -\frac{2k-1}{m+2-2j}B\left(k-\frac{j-1}{m},\,\frac{1}{2}+\frac{j-1}{m}\right)   \quad &\text{if $1\leq k\leq j\leq\frac{m+1}{2}$, $j\not=1$},\\

163: &\\

164:                  \frac{2}{m}\left(\ln 2-\frac{1}{1}-\frac{1}{3}-\dots-\frac{1}{2k-5}-\frac{1}{2k-3}\right)  \quad &\text{if $m$ is even, $1\leq k\leq j=\frac{m+2}{2}$,}

165:                     \end{array}\right.\nonumber

166: \end{equation}

167: where $B(\cdot,\cdot)$ is the beta function.

168: \end{theorem}

169: We obtain  \eqref{main_result} by investigating the asymptotic expansions of an entire function whose zeros are the eigenvalues.  Sibuya \cite{Sibuya} got \eqref{Bender_exp} by using the first order asymptotic expansion of the entire function. 

170: 

171: Next, we let $N_{\ell}(t)$, $t\in\R$, be the eigenvalue counting function, that is, $N_{\ell}(t)$ is the number of eigenvalues $\lambda$ of $H_{\ell,P}$ such that $|\lambda|\leq t$.  Then the following theorem on an asymptotic expansion of the eigenalue counting function is a consequence of Theorem \ref{main_thm1}.

172: \begin{theorem}\label{main_thm2}

173: Let $a\in\C^{m-1}$ be fixed. Suppose that $\Re\left(d_{\ell,j}(a)\right)=0$ for $1\leq j\leq\frac{m+2}{2}$. Then $N_{\ell}(t)$ has the asymptotic expansion 

174: \begin{equation}\label{count_ft}

175: N_{\ell}(t)=\frac{1}{2\pi i}\left(\sum_{j=0}^{\lfloor\frac{m+2}{2}\rfloor}d_{\ell,j}(a)t^{\frac{1}{2}-\frac{j-1}{m}}-\pi i\right)+O(1),\quad\text{as $t\to\infty$,}

176: \end{equation}

177:  where the error $O(1)$ is uniform for any compact set of $a\in\C^{m-1}$. 

178: \end{theorem}

179: \begin{proof}

180: In Theorem \ref{ineq_eq} below, we show that $|\lambda_n|<|\lambda_{n+1}|$ for all large $n\in\N$.

181: 

182: Suppose that $|\lambda_n|\leq t<|\lambda_{n+1}|$. Then since

183: $$\left(n+1+\frac{1}{2}\right)^{\frac{2m}{m+2}}=\left(n+\frac{1}{2}\right)^{\frac{2m}{m+2}}+O\left(n^{\frac{m-2}{m+2}}\right), \quad\text{as $n\to\infty$},$$

184: we see from Theorem \ref{eigen_asy} below that $|\lambda_{n+1}|-|\lambda_{n}|=O\left(n^{\frac{m-2}{m+2}}\right)$.

185: Thus, 

186: $$\lambda_n^{\frac{1}{2}-\frac{j-1}{m}}=t^{\frac{1}{2}-\frac{j-1}{m}}\left(1-\frac{t-\lambda_n}{t}\right)^{\frac{1}{2}-\frac{j-1}{m}}=t^{\frac{1}{2}-\frac{j-1}{m}}\left(1+O\left(\frac{t-\lambda_n}{t}\right)\right)=t^{\frac{1}{2}-\frac{j-1}{m}}+O\left(1\right).$$

187: Hence, replacing $\lambda_n^{\frac{1}{2}-\frac{j-1}{m}}$ in \eqref{main_result} by $t^{\frac{1}{2}-\frac{j-1}{m}}+O\left(1\right)$, and  solving the resulting equation for $n$ complete the proof.

188: \end{proof}

189: 

190:  Next, we improve the asymptotic expansion \eqref{Bender_exp} of the eigenvalues $\lambda_{n}$ of $H_{1,P}$. In particular, we will prove the following, which essentially invert \eqref{main_result} to get $\lambda_n$ in terms of $n$.

191: \begin{theorem}\label{eigen_asy}

192: For each $a\in\C^{m-1}$, there exist some constants $e_{j}(a)\in\C$, $2\leq j\leq \frac{m+2}{2}$, such that

193: \begin{equation}\label{main_asyeq}

194: \lambda_{n}=\lambda_{n,0}+\sum_{j=2}^{\lfloor\frac{m+2}{2}\rfloor}e_{j}(a)\lambda_{n,0}^{1-\frac{j}{m}}+o\left(\lambda_{n,0}^{1-\frac{1}{m}\lfloor\frac{m+2}{2}\rfloor}\right),\quad\text{as $n\to+\infty$},

195: \end{equation}

196: where the error term is uniform for any compact set of $a\in\C^{m-1}$, and where

197: \begin{equation}\nonumber

198: \lambda_{n,0}=\left(\frac{\sqrt{\pi}\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)}{\sin\left(\frac{\ell\pi}{m}\right)\Gamma\left(1+\frac{1}{m}\right)}\left(n+\frac{1}{2}\right)\right)^{\frac{2m}{m+2}},

199: \end{equation}

200: and $e_j(a)$, $1\leq j\leq \frac{m+2}{2}$, are defined recurrently by $e_1(a)=0$ and

201: \begin{align}

202: &e_{j}(a)=-\frac{2m}{m+2}\left(\frac{d_{\ell,j}(a)}{d_{\ell,0}(a)}+\sum_{\substack{|\alpha|=k\geq 2\\ \alpha\cdot\beta=j}}{\frac{1}{2}+\frac{1}{m}\choose k}\frac{k!}{\alpha!}\, e(a)^{\alpha} +\sum_{r=2}^{j-2}\frac{d_{\ell,r}(a)}{d_{\ell,0}(a)}\sum_{\substack{|\alpha|=k\\ \alpha\cdot\beta=j-r}}{\frac{1}{2}+\frac{1-r}{m}\choose k}\frac{k!}{\alpha!}\, e(a)^{\alpha}\right),\nonumber

203: \end{align}

204: where $e(a)=(e_{1}(a),e_{2}(a),\dots,e_{m-1}(a))$.

205: \end{theorem}

206: We note for the first summation in the definition of $e_{j}(a)$ that $\alpha\cdot\beta=j$ implies $\alpha_i=0$ whenever $i\geq j$. Also, for the second summation, we point out that $\alpha\cdot\beta=j-r\leq j-2$ implies $\alpha_i=0$ whenever $i\geq j-1$.

207: 

208: The asymptotic expansions of the eigenvalues of $H_{\ell,P}$ with $\ell=\lfloor\frac{m}{2}\rfloor$ and $\ell=1$ have been studied by a number of people. For example,

209:  Maslov \cite{MAS}  computed the first three terms of asymptotic expansions of $\lambda_n^{\frac{3}{4}}$, where $\lambda_n$ are the eigenvalues of 

210: \begin{equation}\nonumber

211: -\frac{d^{2}}{dx^{2}}u+x^{4}u=\lambda u,\quad u\in L^2(\R). 

212: \end{equation}

213: 

214: Helffer and Robert \cite{HR} considered 

215: \begin{equation}\nonumber

216: -\frac{d^{2k}}{dx^{2k}}u+x^{2m}u+p(x)u=\lambda u,\quad u\in L^2(\R),

217: \end{equation}

218: where $k,\, m$ are positive integers and where $p(\cdot)$ is a {\it real}  polynomial of degree at most $2m-1$. They obtained existence of asymptotic expansions of eigenvalues to all orders, and suggested an explicit way of computing the coefficients of the asymptotic expansion. In particular, for the case when the potential is $\varepsilon x^4+x^2$, $\varepsilon>0$, Helffer and Robert \cite{HR} computed the first nine terms of the asymptotic expansion of $\lambda_n^{\frac{3}{4}}$. 

219: 

220: Fedoryuk \cite[\S 3.3]{Fedoryuk} considered \eqref{ptsym} with complex polynomial potentials and with \eqref{bdcond} for $\ell=\lfloor\frac{m}{2}\rfloor$, and showed the existence of asymptotic expansions of the eigenvalues to all orders. 

221: Note that there appear to be typographical errors in \cite[\S 3.3]{Fedoryuk}.  For example, when $m$ is even (and $\ell=\frac{m}{2}$) the leading coefficient of the asymptotic expansion of $\lambda_n$ in \cite[\S 3.3]{Fedoryuk}  is

222: $\left(2\frac{\sqrt{\pi}\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)}{\Gamma\left(1+\frac{1}{m}\right)}\right)^{\frac{m+2}{2m}}$, which  is different from  $\left(\frac{\sqrt{\pi}\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)}{\Gamma\left(1+\frac{1}{m}\right)}\right)^{\frac{2m}{m+2}}$ found in \cite{AA} and again in Theorem \ref{eigen_asy} above. 

223: 

224: Next, we point out some differences between work of Fedoryuk \cite[\S 3.3]{Fedoryuk} and the present work.  Fedoryuk \cite[\S 3.3]{Fedoryuk} showed the existence of asymptotic expansion of the eigenvalues to all orders while we do not. On the other hand, we treat all decaying boundary conditions \eqref{bdcond} with $1\leq\ell\leq m-1$,

225: while Fedoryuk \cite[\S 3.3]{Fedoryuk} studied the case  $\ell=\lfloor\frac{m}{2}\rfloor$ only. Moreover, we computed more coefficients $e_j(a)$ explicitly, and our methods  are different from Fedoryuk's.

226:   

227: \subsection*{Inverse spectral problem}

228: Here, we introduce results on inverse spectral problems, but first the following corollary is an easy consequence of Theorems \ref{main_thm1} and \ref{eigen_asy}, regarding how the coefficients of the asymptotic expansions depend on $a\in\C^{m-1}$.

229: \begin{corollary}\label{corollary_1}

230: Let $1\leq j\leq \frac{m+2}{2}$ be  a fixed integer. Then we have the following.

231: \begin{enumerate}

232: \item[(i)] $d_{\ell,j}(a)$ and $e_{j}(a)$ are  polynomials in $a_1,a_2,\dots,a_{j-1}, a_j$.

233: \item[(ii)]  $d_{\ell,j}(a)$ and $e_{j}(a)$ do not depend on $a_{j+1},a_{j+2},\dots, a_{m-1}$.

234: \item[(iii)] If $(j-1)\ell$ is a multiple of $m$, then $d_{\ell,j}(a)\equiv 0$, and $e_{j}(a)$ does not depend on $a_j$.

235: \item[(iv)] If $(j-1)\ell$ is not a multiple of $m$, then $d_{\ell,j}(a)$ and $e_{j}(a)$ depend linearly on $a_j$.

236: \end{enumerate}

237: \end{corollary}

238: \begin{proof}

239: Statements on $d_{\ell,j}(a)$ are direct consequences of the definition of $d_{\ell,j}(a)$ in Theorem \ref{main_thm1}. One can use statements on $d_{\ell,j}(a)$ and induction on $j$ to prove  statements on $e_{j}(a)$.

240: \end{proof}

241: Next, one can reconstruct some coefficients of the polynomial potential from the asymptotic expansion of the eigenvalues.

242: \begin{theorem}

243:  Let $1\leq j\leq\frac{m+1}{2}$ be a fixed integer. Suppose that $a_k$ is known whenever  $1\leq k\leq j$ and $(k-1)\ell$ is a multiple of $m$. If $(j-1)\ell$ is a multiple of $m$, then the asymptotic expansions of the eigenvalues $\lambda_{n}$ of $H_{\ell,P}$ of type \eqref{main_asyeq} with an error term $o\left(n^{\frac{2m-2j}{m+2}}\right)$ uniquely and explicitly determine $a_j$. 

244: \end{theorem}

245: \begin{proof}

246: From the asymptotic expansion of the eigenvalues, one gets $e_2(a),e_3(a),\dots,e_j(a)$ that are explicit polynomials in $a_1, a_2,\dots,a_j$.

247: Then since we know $a_k$ if $(k-1)\ell$ is a multiple of $m$, Corollary \ref{corollary_1} says that we can find  all $a_1, a_2,\dots,a_j$.

248: \end{proof}

249: \subsection*{Applications to $\mathcal{PT}$-symmetric potentials}

250: One says that $H_{\ell,P}$ is $\mathcal{PT}$-symmetric if the potential $V$ satisfies $\overline{V(-\overline{z})}=V(z)$, $z\in\C$, that is equivalent to $a\in\R^{m-1}$. Here, we prove the partial reality of the eigenvalues $\lambda_n$ of $\mathcal{PT}$-symmetric $H_{\ell,P}$. But first, we show the following theorem, regarding monotonicity of modulus of $\lambda_n$ for all large $n\in\N$.

251: \begin{theorem}\label{ineq_eq}

252: For each $a\in\C^{m-1}$ there exists $M>0$ such that $|\lambda_{n}|<|\lambda_{n+1}|$ if $n\geq M$. 

253: \end{theorem}

254: \begin{proof}

255: See  Theorem 3 in \cite{Shin2} for the proof of the case when $\ell=1$. One can see that proof of Theorem 3 in \cite{Shin2} can be easily adapted for the cases when $2\leq\ell\leq m-1$.

256: \end{proof}

257: 

258: Now we are ready to  prove the following theorem on the partial reality of the eigenvalues $\lambda_n$ of $H_{\ell,P}$.

259: \begin{theorem}\label{main_theorem}

260: Suppose that $a\in\R^{m-1}$. Then 

261: all but finitely many eigenvalues of $H_{\ell,P}$ are real and positive. Hence $\Re\left(d_{\ell,j}(a)\right)=0$ for all $1\leq j\leq\frac{m+2}{2}$, so that the counting function formula in Theorem \ref{main_thm2} is valid.

262: \end{theorem}

263: \begin{proof}

264: When $H_{\ell,P}$ is $\mathcal{PT}$-symmetric (i.e., $a\in\R^{m-1}$),  $u(z,\lambda)$ is an eigenfunction associated with an eigenvalue $\lambda$ if and only if $\overline{u(-\overline{z},\lambda)}$ is an eigenfunction associated with the eigenvalue $\overline{\lambda}$. Thus, the eigenvalues either appear in complex conjugate pairs, or else are real. So Theorem \ref{ineq_eq} implies Theorem \ref{main_theorem}.

265: \end{proof}

266: In recent years, these $\mathcal{PT}$-symmetric operators have gathered considerable attention, because ample numerical and asymptotic studies suggest that many of such operators have real eigenvalues only even though they are not self-adjoint. In particular, the differential operators $H$ with some polynomial potential $V$ and with the boundary condition \eqref{bdcond} have been considered in \cite{Bender,Bender2,CGM,KS,Ali1,Shin1,Shin,Simon,Znojil} and references therein. The rigorous proof of reality and positivity of the eigenvalues of  $\mathcal{PT}$-symmetric operators with certain classes of polynomial potentials and with the boundary condition \eqref{bdcond} for $\ell=1$,

267: was given by Dorey, Dunning and Tateo \cite{Dorey} in 2001 and by the present author \cite{Shin} in 2002. 

268: 

269: However, there are some $\mathcal{PT}$-symmetric polynomial potentials that produce a finite number of non-real eigenvalues \cite{Bender-1,Pham,Delabaere,Handy2,Handy1} for some particular classes of polynomial potentials.  So without any further restrictions on the real coefficients $a_k$, Theorem \ref{main_theorem} is the most general result one can expect about reality of eigenvalues of $\mathcal{PT}$-symmetric operators with polynomial potentials.

270: 

271: This paper is organized as follows. In Section \ref{prop_sect}, we will introduce work of Hille \cite{Hille} and Sibuya \cite{Sibuya}, regarding properties of solutions of \eqref{ptsym}. Also, we introduce  entire functions $W_{-1,\ell}(a,\lambda)$ whose zeros are closely related with the eigenvalues of $H_{\ell,P}$, due to Sibuya \cite{Sibuya} (c.\ f., Section \ref{asymp_eigen}). In Section \ref{sec_4}, we then provide asymptotics of the entire function $W_{-1,1}(a,\lambda)$ as $\lambda\to\infty$ in the complex plane \cite{Shin2}, improving the  asymptotics of $W_{-1,1}(a,\lambda)$ in \cite{Sibuya}.  In Section \ref{sec_5}, we  provide asymptotic expansions of $W_{-1,\ell}(a,\lambda)$ as $\lambda\to\infty$ in $\C$. In Section \ref{asymp_eigen}, we investigate how the zeros of $W_{-1,\ell}(a,\cdot)$ are related with the eigenvalues of $H_{\ell,P}$. In Sections \ref{sec_7} and \ref{sec_8}, we prove Theorem \ref{main_thm1}. In Section \ref{sec_9}, we prove Theorem \ref{eigen_asy}. Finally, in the Appendix we compute $K_{m,j,k}$ in Theorem \ref{main_thm1}, that is originally given in terms of certain integrals \eqref{K_definition}.

272: 

273: \section{Properties of the solutions}

274: \label{prop_sect}

275: In this section, we introduce work of Hille \cite{Hille} and Sibuya \cite{Sibuya} about properties of the solutions of \eqref{ptsym}. 

276: 

277: 

278: First, we scale equation \eqref{ptsym} because many facts that we need later are stated for the scaled equation. Let $u$ be a solution of (\ref{ptsym}) and let $v(z,\lambda)=u(-iz,\lambda)$. 

279:  Then $v$ solves

280: \begin{equation}\label{rotated1}

281: -v^\dd(z,\lambda)+[(-1)^{\ell+1}z^m+P(z)+\lambda]v(z,\lambda)=0,

282: \end{equation}

283: where $m\geq 3$ and $P$  is a  polynomial (possibly, $P\equiv 0$) of the form

284: $$P(z)=a_1z^{m-1}+a_2z^{m-2}+\cdots +a_{m-1}z,\quad a_k\in\C.$$

285: 

286: When $\ell$ is odd, \eqref{rotated1} becomes 

287: \begin{equation}\label{rotated}

288: -v^\dd(z,\lambda)+[z^m+P(z)+\lambda]v(z,\lambda)=0.

289: \end{equation} 

290: Later we will treat the case when $\ell$ is even.

291: 

292: Since we scaled the argument of $u(\cdot,\lambda)$, we must rotate the boundary condition. We state them in a more general context by using the following definition.

293: \begin{definition}

294: {\rm {\it The Stokes sectors} $S_k$ of the equation (\ref{rotated})

295: are

296: $$ S_k=\left\{z\in \C:\left|\arg (z)-\frac{2k\pi}{m+2}\right|<\frac{\pi}{m+2}\right\}\quad\text{for}\quad k\in \Z.$$ }

297: \end{definition}

298: See Figure \ref{f:graph1}.

299: \begin{figure}[t]

300:     \begin{center}

301:     \includegraphics[width=.4\textwidth]{realgraph1.eps}

302:     \end{center}

303:  \vspace{-.5cm}

304: \caption{The Stokes sectors for $m=3$. The dashed rays represent $\arg z=\pm\frac{\pi}{5},\,\pm\frac{3\pi}{5},\, \pi.$}\label{f:graph1}

305: \end{figure}

306: 

307: It is known from Hille \cite[\S 7.4]{Hille} that every nonconstant solution of (\ref{rotated}) either decays to zero or blows up exponentially, in each Stokes sector $S_k$. 

308:  That is, one has the following result.

309: \begin{lemma}[\protect{\cite[\S 7.4]{Hille}}]\label{gen_pro}

310: ${} $

311: 

312: \begin{itemize}

313: \item [(i)] For each $k\in\Z$, every solution $v$ of (\ref{rotated})  is asymptotic to 

314: \begin{equation} \label{asymp-formula}

315: (const.)z^{-\frac{m}{4}}\exp\left[\pm \int^z \left[\xi^m+P(\xi)+\lambda\right]^{\frac{1}{2}}\,d\xi\right] 

316: \end{equation}

317: as $z \rightarrow \infty$ in every closed subsector of $S_k$.

318: 

319: \item [(ii)] If a nonconstant solution $v$ of \eqref{rotated} decays in $S_k$, it must blow up in $S_{k-1}\cup S_{k+1}$. However, when $v$ blows up in $S_k$, $v$ need not be decaying in $S_{k-1}$ or in $S_{k+1}$.

320: \end{itemize}

321: \end{lemma}

322: Lemma \ref{gen_pro} (i) implies that if $v$ decays along one ray in $S_k$, then it decays along all rays in $S_k$. Also, if $v$ blows up along one ray in $S_k$, then it blows up along all rays in $S_k$. This is essentially why we said in the Introduction that the boundary conditions \eqref{bdcond} with $1\leq\ell\leq m-1$ represent all decaying boundary conditions.

323: 

324: Still with $\ell$ odd, the two rays in \eqref{bdcond} map, by $z\mapsto -iz$,  to the rays $\arg(z)=\pm\frac{(\ell+1)\pi}{m+2}$ which are the center rays of the Stokes sectors $S_{\frac{\ell+1}{2}}$ and $S_{-\frac{\ell+1}{2}}$. Thus, the boundary conditions \eqref{bdcond} on $u$ become

325: \begin{equation}\nonumber

326: \text{$v(\cdot,\lambda)$ decays to zero in the Stokes sector $S_{\frac{\ell+1}{2}}$ and $S_{-\frac{\ell+1}{2}}$.}

327: \end{equation}

328: 

329: When $\ell$ is even, we let

330:  $y(z,\lambda)=v(\omega^{-\frac{1}{2}}z,\lambda)$ so that \eqref{rotated1} becomes

331: \begin{equation}\label{rotated2}

332: -y^\dd(z,\lambda)+[z^m+\omega^{-1} P(\omega^{-\frac{1}{2}}z)+\omega^{-1}\lambda]y(z,\lambda)=0,

333: \end{equation}

334:  where

335: $$\omega=\exp\left[\frac{2\pi i}{m+2}\right]$$

336: and hence, $\omega^{-\frac{m+2}{2}}=-1$.

337: For these cases, the boundary conditions \eqref{bdcond} become

338: \begin{equation}\nonumber

339: \text{$y(\cdot,\lambda)$ decays to zero in the Stokes sector $S_{\frac{\ell+2}{2}}$ and $S_{-\frac{\ell}{2}}$.}

340: \end{equation}

341: 

342: Next we will introduce Sibuya's results, but first we define  a sequence of complex numbers $b_j$ in terms of the $a_k$ and $\lambda$, as follows. For $\lambda \in \C$ fixed, we expand

343: \begin{eqnarray}

344: &&(1+a_1z^{-1}+a_2z^{-2}+\cdots+a_{m-1}z^{1-m}+\lambda z^{-m})^{1/2}\nonumber\\

345: &=&1+\sum_{k=1}^{\infty}{\frac{1}{2}\choose{k}}\left( a_1z^{-1}+a_2z^{-2}+\cdots+a_{m-1}z^{1-m}+\lambda z^{-m}\right)^k\nonumber\\

346: &=&1+\sum_{j=1}^{\infty}\frac{b_j(a,\lambda)}{z^j},\qquad\text{for large}\quad|z|.\label{b_def}

347: \end{eqnarray}

348: Note that $b_1,\,b_2,\,\ldots,\,b_{m-1}$ do not depend on $\lambda$, so we write $b_j(a)=b_j(a,\lambda)$ for $j=1,\,2,\dots,\,m-1$. So the above expansion without the $\lambda z^{-m}$ term still gives $b_j$ for $1\leq j\leq m-1$. We further define $r_m=-\frac{m}{4}$ if $m$ is odd, and $r_m=-\frac{m}{4}-b_{\frac{m}{2}+1}(a)$ if $m$ is even.

349: 

350: 

351:  The following theorem is a special case of Theorems 6.1, 7.2, 19.1 and 20.1 of Sibuya \cite{Sibuya} that is the main ingredient of the proofs of the main results in this paper. 

352: %The coefficient vector $a$ of $P$ is allowed to be complex, here. 

353: \begin{theorem}\label{prop}

354: Equation (\ref{rotated}), with $a\in \C^{m-1}$, admits a solution  $f(z,a,\lambda)$ with the following properties.

355: \begin{enumerate}

356: \item[(i)] $f(z,a,\lambda)$ is an entire function of $z,a $ and $\lambda$.

357: \item[(ii)] $f(z,a,\lambda)$ and $f^\d(z,a,\lambda)=\frac{\partial}{\partial z}f(z,a,\lambda)$ admit the following asymptotic expansions. Let $\varepsilon>0$. Then

358: \begin{align}

359: f(z,a,\lambda)=&\qquad z^{r_m}(1+O(z^{-1/2}))\exp\left[-F(z,a,\lambda)\right],\nonumber\\

360: f^\d(z,a,\lambda)=&-z^{r_m+\frac{m}{2}}(1+O(z^{-1/2}))\exp\left[-F(z,a,\lambda) \right],\nonumber

361: \end{align}

362: as $z$ tends to infinity in  the sector $|\arg z|\leq \frac{3\pi}{m+2}-\varepsilon$, uniformly on each compact set of $(a,\lambda)$-values . 

363: Here

364: \begin{equation}\nonumber

365: F(z,a,\lambda)=\frac{2}{m+2}z^{\frac{m}{2}+1}+\sum_{1\leq j<\frac{m}{2}+1}\frac{2}{m+2-2j}b_j(a) z^{\frac{1}{2}(m+2-2j)}.

366: \end{equation}

367: \item[(iii)] Properties \textup{(i)} and \textup{(ii)} uniquely determine the solution $f(z,a,\lambda)$ of (\ref{rotated}).

368: \item[(iv)] For each fixed $a\in\C^{m-1}$ and $\delta>0$, $f$ and $f^\d$ also admit the asymptotic expansions, 

369: \begin{align}

370: f(0,a,\lambda)=&[1+o(1)]\lambda^{-1/4}\exp\left[L(a,\lambda)\right],\label{eq1}\\

371: f^\d(0,a,\lambda)=&-[1+o(1)]\lambda^{1/4}\exp\left[L(a,\lambda)\right],\label{eq2}

372: \end{align}

373: as $\lambda\to\infty$ in the sector $|\arg(\lambda)|\leq\pi-\delta$, uniformly on each compact set of $a\in\C^{m-1}$, where 

374: \begin{align}

375: L(a,\lambda)=\left\{

376:                     \begin{array}{rl}

377:                     &\int_0^{+\infty}\left(\sqrt{t^m+P(t)+\lambda}- t^{\frac{m}{2}}-\sum_{j=1}^{\frac{m+1}{2}}b_j(a)t^{\frac{m}{2}-j}\right)\,dt \quad \text{if $m$ is odd,}\\

378:                    &\int_0^{+\infty}\left(\sqrt{t^m+P(t)+\lambda}- t^{\frac{m}{2}}-\sum_{j=1}^{\frac{m}{2}}b_j(a)t^{\frac{m}{2}-j}-\frac{b_{\frac{m}{2}+1}}{t+1}\right)\,dt  \quad \text{if $m$ is even.}

379:                     \end{array}

380:                          \right. \nonumber

381: \end{align}

382: \item[(v)] The entire functions  $\lambda\mapsto f(0,a,\lambda)$ and $\lambda\mapsto f^\d(0,a,\lambda)$ have orders $\frac{1}{2}+\frac{1}{m}$.

383: \end{enumerate}

384: \end{theorem}

385: \begin{proof}

386: In  Sibuya's book \cite{Sibuya}, see Theorem 6.1 for a proof of (i) and (ii); Theorem 7.2 for a proof of (iii); and Theorem 19.1 for a proof of (iv).  Moreover, (v) is a consequence of (iv) along with Theorem 20.1 in \cite{Sibuya}. Note that properties (i), (ii) and (iv) are summarized on pages 112--113 of Sibuya \cite{Sibuya}. 

387: \end{proof}

388: 

389: \begin{remarks}

390: {\rm (I) Uniformness of the error term in Theorem \ref{main_thm1}  is essentially due to uniformness of error terms in \eqref{eq1} and \eqref{eq2}. One can check this by carefully following our proofs. In this paper, we omit this part of the proof.

391: 

392: 

393: (II) Throughout this paper, we will deal with numbers like $\left(\omega^{\nu}\lambda\right)^{s}$ for some $s\in\R$, and $\nu\in\C$. As usual, we will use

394: $$\omega^{\nu}=\exp\left[\nu \frac{2\pi i}{m+2}\right]$$

395: and if $\arg(\lambda)$ is specified, then

396: $$\arg\left(\left(\omega^{\nu}\lambda\right)^{s}\right)=s\left[\arg(\omega^{\nu})+\arg(\lambda)\right]=s\left[\Re(\nu)\frac{2\pi}{m+2}+\arg(\lambda)\right],\quad s\in\R.$$

397: If $s\not\in\Z$ then the branch of $\lambda^s$ is chosen to be the negative real axis.

398: }

399: \end{remarks}

400: 

401: Next, we provide an asymptotic expansion of $L$ in \cite{Shin2}.   But first, we recall that 

402: for $1\leq k\leq j\leq \frac{m+2}{2}$,

403: \begin{equation}\nonumber

404: b_{j,k}(a)={\frac{1}{2}\choose{k}}\sum_{\substack{j_p\not=j_q\text{if}\, p\not=q,\, i_p\geq 1\\ i_1+\cdots+i_s=k\\ i_1j_1+\cdots+i_sj_s=j}}\frac{k!}{i_1!i_2!\cdots i_s!}a_{j_1}^{i_1}a_{j_2}^{i_2}\cdots a_{j_{s}}^{i_{s}}={\frac{1}{2}\choose{k}}\sum_{\substack{|\alpha|=k\\ \alpha\cdot\beta=j}}\frac{k!}{\alpha!}\,a^{\alpha}.

405: \end{equation}

406:  Then

407: \begin{equation}\nonumber

408: b_j(a)=\sum_{k=1}^j b_{j,k}(a).

409: \end{equation}

410: Also, we define

411: \begin{equation}\label{K_definition}

412: K_{m,j,k}=\left\{\begin{array}{rl}

413: &\int_0^{\infty}\left(\frac{t^{mk-j}}{\left(t^m+1\right)^{k-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)\,dt,\,\,1\leq k\leq j\leq \frac{m}{2}\,\,\text{or}\,\,k=j=0,\\

414: &\int_0^{\infty}\left(\frac{t^{mk-\frac{m}{2}-1}}{\left(t^m+1\right)^{k-\frac{1}{2}}}-\frac{1}{t+1}\right)\,dt,\,\,1\leq k\leq j= \frac{m+2}{2} \,\,\text{for $m$ even,}

415: \end{array}

416: \right.

417: \end{equation}

418: and define

419: \begin{equation}\label{K_deF}

420: K_{m,0}(a)=K_{m}=K_{m,0,0},\quad K_{m,j}(a)=\sum_{k=1}^jb_{j,k}(a)K_{m,j,k},\quad 1\leq j\leq \frac{m+2}{2}.

421: \end{equation}

422: See Appendix for $K_{m,j,k}$ in terms of some gamma functions as in Theorem \ref{main_thm1}.

423: \begin{lemma}\label{asy_lemma}

424: Let $m\geq 3$ and $a\in\C^{m-1}$ be fixed. Then there exist constants $K_{m,j}(a)\in\C$, $0\leq j \leq\frac{m}{2}+1$, such that 

425: \begin{equation}

426: L(a,\lambda)=\left\{\begin{array}{rl}

427: &\sum_{j=0}^{\frac{m+1}{2}}K_{m,j}(a)\lambda^{\frac{1}{2}+\frac{1-j}{m}}+O\left(|\lambda|^{-\frac{1}{2m}}\right)\,\,\text{if $m$ is odd,}\\

428: &\sum_{j=0}^{\frac{m}{2}+1}K_{m,j}(a)\lambda^{\frac{1}{2}+\frac{1-j}{m}}-\frac{b_{\frac{m}{2}+1}(a)}{m}\ln(\lambda)+O\left(|\lambda|^{-\frac{1}{m}}\right)\,\,\text{if $m$ is even,}

429: \end{array}

430: \right.\nonumber

431: \end{equation}

432: as $\lambda\to\infty$ in the sector $|\arg(\lambda)|\leq \pi-\delta$, uniformly on each compact set of $a\in\C^{m-1}$. 

433: \end{lemma}

434: \begin{proof}

435: See \cite{Shin2} for a proof.

436: \end{proof} 

437: Sibuya \cite{Sibuya} proved the following corollary, directly from Theorem \ref{prop}, that will be used later in Sections \ref{sec_4} and \ref{sec_5}.

438: \begin{corollary}\label{lemma_decay}

439: Let $a\in\C^{m-1}$ be fixed. Then

440: $L(a,\lambda)=K_{m}\lambda^{\frac{1}{2}+\frac{1}{m}}(1+o(1))$ as $\lambda$ tends to infinity in the sector $|\arg \lambda|\leq \pi-\delta$, and hence

441: \begin{equation}\label{re_part}

442: \Re \left(L(a,\lambda)\right)=K_{m}\cos\left(\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))

443: \end{equation}

444: as $\lambda\to\infty$ in the sector $|\arg (\lambda)|\leq \pi-\delta$.

445: 

446: In particular, $\Re \left(L(a,\lambda)\right)\to+\infty$ as $\lambda\to\infty$ in any closed subsector of the sector $|\arg(\lambda)|<\frac{m\pi}{m+2}$. In addition, $\Re \left(L(a,\lambda)\right)\to-\infty$ as $\lambda\to\infty$ in any closed subsector of the sectors $\frac{m\pi}{m+2}<|\arg(\lambda)|<\pi-\delta$.

447: \end{corollary}

448: 

449: Based on Corollary \ref{lemma_decay},

450: Sibuya \cite[Theorem 29.1]{Sibuya} also proved the  asymptotic expansion \eqref{Bender_exp} of the eigenvalues of $H_{\ell,P}$ for $\ell=1$.

451: 

452: Also, Sibuya \cite{Sibuya} constructed solutions of \eqref{rotated} that decays in $S_k$, $k\in\Z$. Before we introduce this, we let 

453: \begin{equation}\label{G_def}

454: G^{\ell}(a):=(\omega^{(m+1)\ell}a_1, \omega^{m\ell}a_2,\ldots,\omega^{3\ell}a_{m-1})\quad \text{for}\quad \ell\in \frac{1}{2}\Z.

455: \end{equation}

456: Then we have the following lemma, regarding properties of $G^{\ell}(\cdot)$.

457: \begin{lemma}\label{lemma_25}

458: For $a\in\C^{m-1}$ fixed, and $\ell_1,\ell_2,\ell\in\frac{1}{2}\Z$,

459: $G^{\ell_1}(G^{\ell_2}(a))=G^{\ell_1+\ell_2}(a)$, and

460: \begin{equation}\nonumber

461: b_{j,k}(G^{\ell}(a))=\omega^{((m+2)k-j)\ell}b_{j,k}(a),\quad \ell\in\frac{1}{2}\Z.

462: \end{equation}

463: 

464: If $\ell\in\Z$ then

465: \begin{equation}\nonumber

466: b_j(G^{\ell}(a))=\omega^{-j\ell}b_j(a).

467: \end{equation}

468: \end{lemma}

469: 

470: 

471: 

472: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

473: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

474: %\section{Eigenvalues are zeros of an entire function}\label{entire_sect}

475: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

476: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

477: 

478: %In this section, we will prove that the eigenvalues are 

479: %zeros of an entire function.

480:   

481: Next, recall that the function

482:  $f(z,a,\lambda)$  in Theorem \ref{prop} solves (\ref{rotated}) and decays to zero exponentially as $z\rightarrow \infty$ in $S_0$, and  blows up in $S_{-1}\cup S_1$. One can check that the function

483: $$f_k(z,a,\lambda):=f(\omega^{-k}z,G^k(a),\omega^{2k}\lambda),\quad k\in\Z,$$

484:  which is obtained by scaling $f(z,G^k(a),\omega^{2k}\lambda)$ in the $z$-variable, also solves (\ref{rotated}). It is clear that $f_0(z,a,\lambda)=f(z,a,\lambda)$, and that $f_k(z,a,\lambda)$ decays in $S_k$ and blows up in $S_{k-1}\cup S_{k+1}$ since $f(z,G^k(a),\omega^{2k}\lambda)$ decays in $S_0$. Since no nonconstant solution decays in two consecutive Stokes sectors (see Lemma \ref{gen_pro} (ii)), $f_{k}$ and $f_{k+1}$ are linearly independent and hence any solution of (\ref{rotated}) can be expressed as a linear combination of these two. Especially,  for each $k\in\Z$ there exist some coefficients $C_k(a,\lambda)$ and $\widetilde{C}_k(a,\lambda)$ such that

485: \begin{equation}\label{stokes}

486: f_{k}(z,a,\lambda)=C_k(a,\lambda)f_{0}(z,a,\lambda)+\widetilde{C}_k(a,\lambda)f_{-1}(z,a,\lambda).

487: \end{equation}

488: We then see that 

489: \begin{equation}\label{C_def}

490: C_k(a,\lambda)=-\frac{W_{k,-1}(a,\lambda)}{W_{-1,0}(a,\lambda)}\quad\text{and}\quad \widetilde{C}_k(a,\lambda)=\frac{W_{k,0}(a,\lambda)}{W_{-1,0}(a,\lambda)},

491: \end{equation}

492: where $W_{j,\ell}=f_jf_{\ell}^\d -f_j^\d f_{\ell}$ is the Wronskian of $f_j$ and $f_{\ell}$. Since both $f_j,\,f_{\ell}$ are solutions of the same linear equation (\ref{rotated}), we know that the Wronskians are constant functions of $z$. Also, $f_k$ and $f_{k+1}$ are linearly independent, and hence $W_{k,k+1}\not=0$ for all $k\in \Z$. 

493: 

494: Also, the following is an easy consequence of \eqref{stokes} and \eqref{C_def}. For each $k,\ell\in\Z$ we have 

495: \begin{align}

496: W_{\ell,k}(a,\lambda)&=C_k(a,\lambda)W_{\ell,0}(a,\lambda)+\widetilde{C}_k(a,\lambda)W_{\ell,-1}(a,\lambda)\nonumber\\

497: &=-\frac{W_{k,-1}(a,\lambda)W_{\ell,0}(a,\lambda)}{W_{-1,0}(a,\lambda)}+\frac{W_{k,0}(a,\lambda)W_{\ell,-1}(a,\lambda)}{W_{-1,0}(a,\lambda)}.\label{kplus}

498: \end{align}

499: 

500: Moreover, we have the following lemma that is useful  later on.

501: \begin{lemma}\label{shift_lemma}

502: Suppose $k,\,j\in\Z$. Then

503: \begin{equation}\label{kplus1}

504: W_{k+1,j+1}(a,\lambda)=\omega^{-1}W_{k,j}(G(a),\omega^2\lambda),

505: \end{equation}

506: and $W_{0,1}(a,\lambda)=2\omega^{\mu(a)}$, where

507: \begin{eqnarray}

508: \mu(a)=\left\{

509:               \begin{array}{rl}

510:               \frac{m}{4}  \quad &\text{if $m$ is odd,}\\

511:               \frac{m}{4}- b_{\frac{m}{2}+1}(a) \quad &\text{if $m$ is even.}

512:               \end{array}

513:                          \right. \nonumber

514: \end{eqnarray}

515: 

516: \end{lemma}

517: \begin{proof}

518: See Sibuya \cite[pages 116-118]{Sibuya}.

519: \end{proof}

520: 

521: We let $\nu(a)=\frac{m}{4}-\mu(a)$, that is, 

522: \begin{eqnarray}

523: \nu(a)=\left\{

524:                          \begin{array}{rl}

525:                          0 & \quad \text{if $m$ is odd,}\\

526:                          b_{\frac{m}{2}+1}(a)& \quad \text{if $m$ is even.}

527:                           \end{array}

528:                          \right. \label{def_nu}

529: \end{eqnarray}

530: 

531: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

532: \section{Asymptotics of $W_{-1,1}(a,\lambda)$}\label{sec_4}

533: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

534: In this section, we introduce asymptotic expansions of $W_{-1,1}(a,\lambda)$ as $\lambda\to\infty$ along the rays in the complex plane \cite{Shin2}. 

535: 

536: 

537: 

538: First, we provide an asymptotic expansion of  the Wronskian  $W_{0,j}(a,\lambda)$ of $f_0$ and $f_j$ that will be frequently used later.

539: \begin{lemma}\label{lemma7}

540: Suppose that $1\leq j\leq \frac{m}{2}+1$. Then for each $a\in\C^{m-1}$,

541: \begin{equation}\label{sec_eq1}

542: W_{0,j}(a,\lambda)=[2i\omega^{-\frac{j}{2}}+o(1)]\exp\left[L(G^{j}(a),\omega^{2j-m-2}\lambda)+L(a,\lambda)\right],

543: \end{equation}

544: as $\lambda\to\infty$ in the sector

545: \begin{equation}\label{sector0}

546: -\pi+\delta\leq \pi-\frac{4j\pi}{m+2}+\delta \leq \arg(\lambda)\leq \pi-\delta.

547: \end{equation}

548: \end{lemma}

549: %

550: 

551: Next, we provide an asymptotic expansion of $W_{-1,1}(a,\lambda)$ as $\lambda\to\infty$ in the sector near the negative real axis.

552: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

553: \begin{theorem}\label{thm_neg}

554: Let $m\geq 3$, $a\in\C^{m-1}$ and $0<\delta<\frac{\pi}{m+2}$ be fixed. Then

555: \begin{equation}\label{asy_1}

556: W_{-1,1}(a,\lambda)=[2i+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)+L(G(a),\omega^{-m}\lambda)\right],

557: \end{equation}

558: as $\lambda\to \infty$ along the rays in the sector

559: \begin{equation}\label{sector1}

560: \pi-\frac{4\pi}{m+2}+\delta\leq \arg(\lambda)\leq \pi+\frac{4\pi}{m+2}-\delta.

561: \end{equation}

562: \end{theorem}

563: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

564: \begin{proof}

565: This is an easy consequence of Lemma \ref{lemma7} with $j=2$ and \eqref{kplus1}.

566: \end{proof}

567: 

568: Also,  for integers $m\geq 4$ we provide an asymptotic expansion of $W_{-1,1}(a,\lambda)$ as $\lambda\to\infty$ in the sector $|\arg(\lambda)|\leq \pi-\delta$.

569: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

570: \begin{theorem}\label{zero_thm}

571: Let $a\in\C^{m-1}$ and $0<\delta<\frac{\pi}{2(m+2)}$ be fixed. 

572: If $m\geq 4$ then 

573: \begin{align}

574: W_{-1,1}(a,\lambda)=&[2\omega^{\frac{1}{2}+\mu(a)}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(a,\lambda)\right]\nonumber\\

575: &+[2\omega^{\frac{1}{2}+\mu(a)+2\nu(a)}+o(1)]\exp\left[L(G(a),\omega^{2}\lambda)-L(a,\lambda)\right],\label{tot_asy}

576: \end{align}

577: as $\lambda\to\infty$ in the sector 

578: \begin{equation}\label{sector41}

579: -\pi+\delta \leq \arg(\lambda)\leq \pi-\delta.

580: \end{equation} 

581: \end{theorem}

582: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

583: 

584: 

585: 

586: We provide an asymptotic expansion of $W_{-1,1}(a,\lambda)$ as $\lambda\to\infty$ along the rays in the upper half plane. 

587: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

588: \begin{corollary}\label{lemma_up}

589: Let $m\geq 4$, $a\in\C^{m-1}$ and $0<\delta<\frac{\pi}{m+2}$ be fixed. Then

590: \begin{equation}\nonumber

591: W_{-1,1}(a,\lambda)=[2\omega^{\frac{1}{2}+\mu(a)}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(a,\lambda)\right],

592: \end{equation}

593: as $\lambda\to\infty$ in the sector $\delta \leq \arg(\lambda)\leq \pi-\delta$.

594: Also,

595: \begin{equation}\nonumber

596: W_{-1,1}(a,\lambda)=[2\omega^{\frac{1}{2}+\mu(a)+2\nu(a)}+o(1)]\exp\left[L(G(a),\omega^{2}\lambda)-L(a,\lambda)\right],

597: \end{equation}

598: as $\lambda\to\infty$ in the sector $-\pi+\delta \leq \arg(\lambda)\leq -\delta$.

599: \end{corollary}

600: \begin{proof}

601: We will determine which term in \eqref{tot_asy} dominates in the upper and lower half planes.

602: 

603: Since, by \eqref{re_part},

604: \begin{equation}\nonumber

605: \Re(L(a,\lambda))=K_m\cos\left(\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1)),

606: \end{equation}

607: we have

608: \begin{align}

609: &\left[\Re(L(G^{-1}(a),\omega^{-2}\lambda))-\Re(L(a,\lambda))\right]-\left[\Re(L(G(a),\omega^{2}\lambda))-\Re(L(a,\lambda))\right]\nonumber\\

610: &=K_m\left[\cos\left(-\frac{2\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)-\cos\left(\frac{2\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right]|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\

611: &=2K_m\sin\left(\frac{2\pi}{m}\right)\sin\left(\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1)).\nonumber

612: \end{align}

613: Thus, the first term in \eqref{tot_asy} dominates as $\lambda\to\infty$ along the rays in the upper half plane, and the second term dominates in the lower half plane. This completes the proof. 

614: \end{proof}

615: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

616: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

617: \begin{proof}[Proof of Theorem ~\ref{zero_thm}]

618: In \cite{Shin2}, $C(a,\lambda)$ is used for $\frac{W_{-1,1}(a,\lambda)}{W_{0,1}(a,\lambda)}$ and asymptotics of $C(a,\lambda)$ are provided.  Notice that $W_{-1,1}(a,\lambda)=2\omega^{\mu(a)}C(a,\lambda)$.

619: 

620: Theorem 13 in \cite{Shin2} implies

621:  \eqref{tot_asy}  for the sector 

622: \begin{equation}\label{sector4}

623: \pi-\frac{4\lfloor\frac{m}{2}\rfloor\pi}{m+2}+\delta \leq \arg(\lambda)\leq \pi-\frac{4\pi}{m+2}-\delta.

624: \end{equation} 

625: Theorem 14 in \cite{Shin2} implies that

626: \begin{align}

627: W_{-1,1}(a,\lambda)=&[2\omega^{\frac{1}{2}+\mu(a)}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(a,\lambda)\right]\nonumber\\

628: &+[2\omega^{1+2\mu(a)+4\nu(a)}+o(1)]\exp\left[-L(G^2(a),\omega^{2-m}\lambda)-L(a,\lambda)\right],\nonumber

629: \end{align}

630: as $\lambda\to\infty$ in the sector $\pi-\frac{8\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi-\delta$. One can check that the first term dominates in this sector, by using an argument  similar to that in the proof of Corollary \ref{lemma_up}.

631: 

632: Also, Theorem 15 in \cite{Shin2} implies that

633: \begin{align}

634: W_{-1,1}(a,\lambda)=&[2\omega^{1+2\mu(a)}+o(1)]\exp\left[-L(a,\omega^{-m-2}\lambda)-L(^{-2}(a),\omega^{-4}\lambda)\right]\nonumber\\

635: &+[2\omega^{\frac{1}{2}+\mu(a)+2\nu(a)}+o(1)]\exp\left[L(G(a),\omega^{-m}\lambda)-L(a,\omega^{-m-2}\lambda)\right],\nonumber

636: \end{align}

637: as $\lambda\to\infty$ in the sector $\pi+\delta\leq\arg(\lambda)\leq\pi+\frac{8\pi}{m+2}-\delta$. One can check that the second term dominates in this sector. Then we replace $\lambda$ by $\omega^{m+2}\lambda$ to convert the sector here to $-\pi+\delta\leq\arg(\lambda)\leq-\pi+\frac{8\pi}{m+2}-\delta$. This completes the proof.

638: \end{proof}

639: 

640: \begin{theorem}\label{thm_sector2}

641: Let $m=3$ and let $a\in\C^{m-1}$ and $0<\delta<\frac{\pi}{m+2}$ be fixed. Then

642: \begin{align}

643: W_{-1,1}(a,\lambda)=&[-2\omega^{-\frac{5}{4}}+o(1)]\exp\left[L(G^{4}(a),\omega^{-2}\lambda)-L(a,\lambda)\right]\nonumber\\

644: &-[2i\omega^{\frac{5}{2}}+o(1)]\exp\left[-L(G^2(a),\omega^{-1}\lambda)-L(a,\lambda)\right],\nonumber

645: \end{align}

646: as $\lambda\to\infty$ in the sector 

647: $-\delta \leq \arg(\lambda)\leq \pi-\delta$.

648: Also,

649: \begin{align}

650: W_{-1,1}(a,\lambda)=&[-2i\omega^{\frac{5}{2}}+o(1)]\exp\left[-L(a,\omega^{-5}\lambda)-L(G^{-2}(a),\omega^{-4}\lambda)\right]\nonumber\\

651: &+[2\omega^{\frac{15}{4}}+o(1)]\exp\left[L(G(a),\omega^{-3}\lambda)-L(a,\omega^{-5}\lambda)\right],\nonumber

652: \end{align}

653: as $\lambda\to\infty$ in the sector $\pi+\delta \leq \arg(\lambda)\leq \delta$.

654: \end{theorem}

655: \begin{proof}

656: See Theorems 14 and 15 in \cite{Shin2} for a proof.

657: \end{proof}

658: 

659: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

660: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

661: \section{Asymptotics of $W_{-1,n}(a,\lambda)$}\label{sec_5}

662: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

663: In this section, we will provide asymptotic expansions of $W_{-1,n}(a,\cdot)$, zeros of which will be closely related with the eigenvalues of $H_{n,P}$.

664: 

665: First, we treat the cases when  $1\leq n <\lfloor\frac{m}{2}\rfloor$.

666: \begin{theorem}

667: Let $1\leq n <\lfloor\frac{m}{2}\rfloor$ be an integer. Then $W_{-1,n}(a,\cdot)$  admits the following asymptotic expansion 

668: \begin{equation}\label{up_asy}

669: W_{-1,n}(a,\lambda)=[2\omega^{\frac{2-n}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right],

670: \end{equation}

671: as $\lambda\to\infty$ in the sector 

672: \begin{equation}\label{zero3_sector}

673: -\frac{2(n-1)\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi -\frac{4n\pi}{m+2}+\delta.

674: \end{equation}

675: 

676: Also,

677: \begin{align}

678: W_{-1,n}(a,\lambda)&=[2\omega^{\frac{2-n}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right]\nonumber\\

679: &+[2\omega^{\frac{2-n}{2}+\mu(G^{-1}(a))}+o(1)]\exp\left[L(G^n(a),\omega^{2n}\lambda)-L(a,\lambda)\right],\label{zero_asy}

680: \end{align}

681: as $\lambda\to\infty$ in the sector 

682: \begin{equation}\label{zero2_sector}

683: -\frac{2(n-1)\pi}{m+2}-\delta\leq\arg(\lambda)\leq-\frac{2(n-1)\pi}{m+2}+\delta.

684: \end{equation}

685: \end{theorem}

686: \begin{proof}

687: First we will prove \eqref{up_asy} for the sector

688:  \begin{equation}\label{zero1_sector}

689: -\frac{2(n-1)\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi -\frac{4n\pi}{m+2}-\delta

690: \end{equation}

691:  and the second part of the theorem by induction on $n$.

692: 

693: The case when $n=1$ is trivially satisfied by  Theorem \ref{zero_thm} and Corollary \ref{lemma_up} since $\mu(a)+2\nu(a)=\mu(G^{-1}(a))$.

694: 

695: Suppose that \eqref{up_asy} holds in the sector \eqref{zero1_sector} for $n-1$. From this induction hypothesis we have

696: \begin{align}

697: W_{0,n}(a,\lambda)&=\omega^{-1}W_{-1,n-1}(G(a),\omega^2\lambda)\nonumber\\

698: &=[2\omega^{-\frac{n-1}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(a,\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\label{1steq}

699: \end{align}

700: as $\lambda\to\infty$ in the sector

701: \begin{equation}\nonumber

702: -\frac{2(n-2)\pi}{m+2}+\delta\leq\arg(\omega^2\lambda)\leq\pi -\frac{4(n-1)\pi}{m+2}-\delta,

703: \end{equation}

704: that is,

705: \begin{equation}\label{dom_sector}

706: -\frac{2n\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi -\frac{4n\pi}{m+2}-\delta.

707: \end{equation}

708: 

709: Also, from Lemma \ref{lemma7} if $1\leq j\leq\frac{m}{2}+1$, then we have 

710: \begin{equation}\label{2ndeq}

711: W_{0,j}(a,\lambda)=[2i\omega^{-\frac{j}{2}}+o(1)]\exp\left[L(G^{j}(a),\omega^{2j-m-2}\lambda)+L(a,\lambda)\right],

712: \end{equation}

713: as $\lambda\to\infty$ in the sector

714: \begin{equation}\nonumber

715:  \pi-\frac{4j\pi}{m+2}+\delta \leq \arg(\lambda)\leq \pi-\delta.

716: \end{equation}

717: We solve \eqref{kplus} for $W_{\ell,-1}(a,\lambda)$ and set $\ell=n$ to get

718: \begin{equation}\label{kn_asy}

719: W_{-1,n}(a,\lambda)=\frac{W_{-1,0}(a,\lambda)W_{n,k}(a,\lambda)}{W_{0,k}(a,\lambda)}+\frac{W_{0,n}(a,\lambda)W_{-1,k}(a,\lambda)}{W_{0,k}(a,\lambda)}

720: \end{equation}

721: Set $k=\lfloor\frac{m}{2}\rfloor$. Then since $1\leq k-n<k=\lfloor\frac{m}{2}\rfloor$, using \eqref{kplus1},

722: \begin{align}

723: W_{-1,n}(a,\lambda)&=\frac{2\omega^{\mu(G^{-1}(a))}W_{0,k-n}(G^n(a),\omega^{2n}\lambda)}{\omega^{n-1}W_{0,k}(a,\lambda)}+\frac{W_{0,n}(a,\lambda)W_{0,k+1}(G^{-1}(a),\omega^{-2}\lambda)}{\omega^{-1}W_{0,k}(a,\lambda)}\nonumber\\

724: &=\frac{2\omega^{\mu(G^{-1}(a))}[2i\omega^{-\frac{k-n}{2}}+o(1)]\exp\left[L(G^{n}(a),\omega^{2k-m-2}\lambda)+L(G^n(a),\omega^{2n}\lambda)\right]}{\omega^{n-1}[2i\omega^{-\frac{k}{2}}+o(1)]\exp\left[L(G^{k}(a),\omega^{2k-m-2}\lambda)+L(a,\lambda)\right]}\nonumber\\

725: &+\frac{[2\omega^{-\frac{n-1}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(a,\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right]}{\omega^{-1}[2i\omega^{-\frac{k}{2}}+o(1)]\exp\left[L(G^{k}(a),\omega^{2k-m-2}\lambda)+L(a,\lambda)\right]}\nonumber\\

726: &\times [2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{k}(a),\omega^{2k-m-2}\lambda)+L(G^{-1}(a),\omega^{-2}\lambda)\right]\nonumber\\

727: &=[2\omega^{\frac{2-n}{2}+\mu(G^{-1}(a))}+o(1)]\exp\left[L(G^n(a),\omega^{2n}\lambda)-L(a,\lambda)\right]\nonumber\\

728: &+[2\omega^{\frac{2-n}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\label{asy_sector}

729: \end{align}

730: where we used \eqref{1steq} for $W_{0,n}(a,\lambda)$ and \eqref{2ndeq} for everything else, provided that $\lambda$ lies in \eqref{dom_sector} and that

731: \begin{align}

732: \pi-\frac{4\left(\lfloor\frac{m}{2}\rfloor-n\right)\pi}{m+2}+\delta \leq &\arg(\omega^{2n}\lambda)\leq \pi-\delta\nonumber\\

733: \pi-\frac{4\lfloor\frac{m}{2}\rfloor\pi}{m+2}+\delta \leq &\arg(\lambda)\leq \pi-\delta\nonumber\\

734: \pi-\frac{4\left(\lfloor\frac{m}{2}\rfloor+1\right)\pi}{m+2}+\delta \leq &\arg(\omega^{-2}\lambda)\leq \pi-\delta,\nonumber

735: \end{align}

736: that is,

737: \begin{equation}\nonumber

738: -\frac{2n\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi -\frac{4n\pi}{m+2}-\delta.

739: \end{equation}

740: Thus, the second part of the theorem is proved by induction.

741: 

742: Next in order to  prove the first part of the theorem for the sector \eqref{zero1_sector}, we will determine which term in \eqref{asy_sector} dominates as $\lambda\to\infty$. To do that, we look at

743: \begin{align}

744: &\Re\left(L(G^{-1}(a),\omega^{-2}\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right)-\Re\left(L(G^{n}(a),\omega^{2n}\lambda)-L(a,\lambda)\right)\nonumber\\

745: &=K_m\left[\cos\left(-\frac{2\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)-\cos\left(\frac{2(n-1)\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right.\nonumber\\

746: &-\left.\left(\cos\left(-\frac{2n\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)-\cos\left(\frac{m+2}{2m}\arg(\lambda)\right)\right)\right]|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\

747: &=2K_m\sin\left(\frac{n\pi}{m}\right)\left[\sin\left(\frac{(n-2)\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right.\nonumber\\

748: &\qquad\qquad\qquad\qquad\qquad\qquad\left.+\sin\left(\frac{n\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)\right]|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1))\nonumber\\

749: &=4K_m\sin\left(\frac{n\pi}{m}\right)\cos\left(\frac{\pi}{m}\right)\sin\left(\frac{(n-1)\pi}{m}+\frac{m+2}{2m}\arg(\lambda)\right)|\lambda|^{\frac{1}{2}+\frac{1}{m}}(1+o(1)),\label{dom_set}

750: \end{align}

751: that tends to positive infinity as $\lambda\to\infty$  (and hence the second term in \eqref{asy_sector} dominates) if $-\frac{2(n-1)\pi}{m+2}+\delta\leq\arg(\lambda)\leq \pi-\frac{4n\pi}{m+2}-\delta$. 

752: 

753: 

754: We still need to prove \eqref{up_asy} for the sector 

755: \begin{equation}\label{asy_rem}

756: \pi-\frac{4n\pi}{m+2}-\delta\leq\arg(\lambda)\leq \pi-\frac{4n\pi}{m+2}+\delta,

757: \end{equation}

758: for which we use induction on $n$ again.

759: 

760: When $n=1$, \eqref{up_asy} holds by Lemma \ref{lemma_up}.

761: 

762: Suppose that \eqref{up_asy} in the sector \eqref{asy_rem} for $n-1$ with  $2\leq n<\lfloor\frac{m}{2}\rfloor$. 

763: Then \eqref{kplus1} and \eqref{kn_asy} with $k=n+1$  yield

764: \begin{equation}\nonumber

765: W_{-1,n}(a,\lambda)

766: =\frac{2\omega^{\mu(G^{-1}(a))}W_{0,1}(G^n(a),\omega^{2n}\lambda)}{\omega^{n-1}W_{0,n+1}(a,\lambda)}+\frac{W_{-1,n-1}(G(a),\omega^2\lambda)W_{0,n+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,n+1}(a,\lambda)}.

767: \end{equation}

768: If $\pi-\frac{4n\pi}{m+2}-\delta\leq\arg(\lambda)\leq \pi-\frac{4n\pi}{m+2}+\delta$, then $\pi-\frac{4(n-1)\pi}{m+2}-\delta\leq\arg(\omega^2\lambda)\leq \pi-\frac{4(n-1)\pi}{m+2}+\delta$. So

769: \begin{align}

770: W_{-1,n}(a,\lambda)

771: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^n(a))}}{\omega^{n-1}W_{0,n+1}(a,\lambda)}+\frac{W_{-1,n-1}(G(a),\omega^2\lambda)W_{0,n+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,n+1}(a,\lambda)}\label{last_eq}\\

772: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^n(a))}}{\omega^{n-1}[2i\omega^{-\frac{n+1}{2}}+o(1)]\exp\left[L(G^{n+1}(a),\omega^{2n-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

773: &+\frac{[2\omega^{\frac{3-n}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(a,\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right]}{[2i\omega^{-\frac{n+1}{2}}+o(1)]\exp\left[L(G^{n+1}(a),\omega^{2n-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

774: &\times [2i\omega^{-\frac{n+2}{2}}+o(1)]\exp\left[L(G^{n+1}(a),\omega^{2n-m}\lambda)+L(G^{-1}(a),\omega^{-2}\lambda)\right]\nonumber\\

775: &=[-2i\omega^{\frac{3-n}{2}+\mu(G^{-1}(a))+\mu(G^n(a))}+o(1)]\exp\left[-L(G^{n+1}(a),\omega^{2n-m}\lambda)-L(a,\lambda)\right]\nonumber\\

776: &+[2\omega^{\frac{2-n}{2}+\mu(G^{n-1}(a))}+o(1)]\exp\left[L(G^{-1}(a),\omega^{-2}\lambda)-L(G^{n-1}(a),\omega^{2(n-1)}\lambda)\right],\nonumber

777: \end{align}

778: where we use the induction hypothesis for $W_{-1,n-1}(G(a),\omega^2\lambda)$ and use \eqref{sec_eq1} for $W_{0,n+1}(a,\lambda)$ and $W_{0,n+2}(G^{-1}(a),\omega^{-2}\lambda)$.

779: Next, we use an argument similar \eqref{dom_set} to complete the induction step. Thus, the theorem is proved.

780: \end{proof}

781: Next we investigate $W_{0,\lfloor\frac{m}{2}\rfloor}(a,\lambda)$.

782: \begin{theorem}\label{bod_thm}

783: If $m\geq 4$ is an even integer, then

784: \begin{align}

785: W_{-1,\lfloor\frac{m}{2}\rfloor}(a,\lambda)

786: &=-[2\omega^{2+\mu(G^{-1}(a))+\mu(G^{\frac{m}{2}}(a))}+o(1)]\exp\left[-L(G^{\frac{m+2}{2}}(a),\lambda)-L(a,\lambda)\right]\nonumber\\

787: &-[2\omega^{2+\mu(a)+\mu(G^{\frac{m-2}{2}}(a))}+o(1)]\exp\left[-L(G^{\frac{m-2}{2}}(a),\omega^{m-2}\lambda)-L(G^{m}(a),\omega^{m-2}\lambda)\right].\label{thm_eq1}

788: \end{align}

789: as $\lambda\to\infty$ in the sector 

790: \begin{equation}\label{zero4_sector}

791: -\pi+\frac{4\pi}{m+2}-\delta\leq\arg(\lambda)\leq-\pi+\frac{4\pi}{m+2}+\delta.

792: \end{equation}

793: 

794: If $m\geq 4$ is an odd integer, then

795: \begin{align}

796: W_{-1,\lfloor\frac{m}{2}\rfloor}(a,\lambda)

797: &=[2\omega^{\frac{5}{4}}+o(1)]\exp\left[-L(G^{\frac{m+1}{2}}(a),\omega^{-1}\lambda)-L(a,\lambda)\right]\nonumber\\

798: &+[2\omega^{\frac{5}{4}}+o(1)]\exp\left[L(G^{m+1}(a),\omega^{-2}\lambda)-L(G^{\frac{m-3}{2}}(a),\omega^{m-3}\lambda)\right].\label{thm_eq2}

799: \end{align}

800: as $\lambda\to\infty$ in the sector 

801: \begin{equation}\nonumber

802: -\pi+\frac{4\pi}{m+2}+\delta\leq\arg(\lambda)\leq-\pi+\frac{6\pi}{m+2}+\delta.

803: \end{equation}

804: \end{theorem}

805: \begin{proof}

806: We will use \eqref{last_eq} with $n=\lfloor\frac{m}{2}\rfloor$, that is,

807: \begin{equation}\nonumber

808: W_{-1,\lfloor\frac{m}{2}\rfloor}(a,\lambda)=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{\lfloor\frac{m}{2}\rfloor}(a))}}{\omega^{\lfloor\frac{m}{2}\rfloor-1}W_{0,\lfloor\frac{m}{2}\rfloor+1}(a,\lambda)}+\frac{W_{-1,\lfloor\frac{m}{2}\rfloor-1}(G(a),\omega^2\lambda)W_{0,\lfloor\frac{m}{2}\rfloor+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,\lfloor\frac{m}{2}\rfloor+1}(a,\lambda)}.

809: \end{equation}

810: When $m$ is even, say $m=2k$, 

811: \begin{align}

812: W_{0,k+2}(G^{-1}(a),\omega^{-2}\lambda)&=W_{m+2,k+2}(G^{-1}(a),\omega^{-2}\lambda)\nonumber\\

813: &=-\omega^{-k-3}W_{-1,k-1}(G^{k+2}(a),\omega^{2k+4}\lambda)\nonumber\\

814: &=\omega^{-2}W_{-1,k-1}(G^{k+2}(a),\omega^{2}\lambda).\nonumber

815: \end{align}

816: So

817: \begin{align}

818: W_{-1,k}(a,\lambda)&=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{0,k+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,k+1}(a,\lambda)}\nonumber\\

819: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{-1,k-1}(G^{k+2}(a),\omega^{2}\lambda)}{\omega^2 W_{0,k+1}(a,\lambda)}\nonumber

820: \end{align}

821: Since $\lambda$ lies in \eqref{zero4_sector}, 

822: $$

823: -\frac{2\left(\lfloor\frac{m}{2}\rfloor-2\right)\pi}{m+2}+\delta\leq-\pi+\frac{8\pi}{m+2}-\delta\leq\arg(\omega^2\lambda)\leq\pi-\frac{4\left(\lfloor\frac{m}{2}\rfloor-1\right)\pi}{m+2}+\delta.

824: $$

825: \begin{align}

826: &W_{-1,k}(a,\lambda)\nonumber\\

827: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{0,k+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,k+1}(a,\lambda)}\nonumber\\

828: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{-1,k-1}(G^{k+2}(a),\omega^{2}\lambda)}{\omega^2 W_{0,k+1}(a,\lambda)}\nonumber\\

829: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}[2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{k+1}(a),\omega^{2k-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

830: &+\frac{[2\omega^{\frac{3-k}{2}+\mu(G^{k-1}(a))}+o(1)]\exp\left[L(a,\lambda)-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)\right]}{\omega^2 [2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{k+1}(a),\omega^{2k-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

831: &\times[2\omega^{\frac{3-k}{2}+\mu(G^{2k}(a))}+o(1)]\exp\left[L(G^{k+1}(a),\lambda)-L(G^{2k}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\

832: &=[-2i\omega^{\frac{3-k}{2}+\mu(G^{-1}(a))+\mu(G^{k}(a))}+o(1)]\exp\left[-L(G^{k+1}(a),\lambda)-L(a,\lambda)\right]\nonumber\\

833: &-[2i\omega^{\frac{3-k}{2}+\mu(G^{k-1}(a))+\mu(G^{2k}(a))}+o(1)]\exp\left[-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)-L(G^{2k}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\

834: &=-[2i\omega^{\frac{3-k}{2}+\mu(G^{-1}(a))+\mu(G^{k}(a))}+o(1)]\exp\left[-L(G^{k+1}(a),\lambda)-L(a,\lambda)\right]\nonumber\\

835: &-[2i\omega^{\frac{3-k}{2}+\mu(a)+\mu(G^{k-1}(a))}+o(1)]\exp\left[-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)-L(G^{2k}(a),\omega^{2(k-1)}\lambda)\right],\nonumber

836: \end{align}

837: where we used \eqref{up_asy} for $W_{-1,k-1}(G(a),\cdot)$ and $W_{-1,k-1}(G^{k+2}(a),\cdot)$, and \eqref{sec_eq1} for $W_{0,k+1}(a,\cdot)$. Finally, we use $\omega^{-\frac{m+2}{4}}=-i$, to get the desired asymptotic expansion of $W_{-1,\lfloor\frac{m}{2}\rfloor}(a,\lambda)$.

838: 

839: Next we investigate the case when $m$ is odd, say $m=2k+1$ (so $\lfloor\frac{m}{2}\rfloor=k$).

840: \begin{equation}\nonumber

841: W_{0,\lfloor\frac{m}{2}\rfloor+1}(a,\lambda)=\omega^{-1}W_{-1,\lfloor\frac{m}{2}\rfloor}(G(a),\omega^{2}\lambda)

842: \end{equation}

843: and

844: \begin{align}

845: W_{0,k+2}(G^{-1}(a),\omega^{-2}\lambda)&=W_{m+2,k+2}(G^{-1}(a),\omega^{-2}\lambda)\nonumber\\

846: &=-\omega^{-k-2}W_{0,k+1}(G^{k+1}(a),\omega^{2k+2}\lambda)\nonumber\\

847: &=\omega^{-\frac{1}{2}}W_{0,k+1}(G^{k+1}(a),\omega^{-1}\lambda).\nonumber

848: \end{align}

849: Similarly to the proof of the theorem for $m$ even,

850: \begin{align}

851: W_{-1,k}(a,\lambda)

852: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{0,k+2}(G^{-1}(a),\omega^{-2}\lambda)}{W_{0,k+1}(a,\lambda)}\nonumber\\

853: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}W_{0,k+1}(a,\lambda)}+\frac{W_{-1,k-1}(G(a),\omega^2\lambda)W_{0,k+1}(G^{k+1}(a),\omega^{-1}\lambda)}{\omega^{\frac{1}{2}} W_{0,k+1}(a,\lambda)}\nonumber\\

854: &=\frac{4\omega^{\mu(G^{-1}(a))+\mu(G^{k}(a))}}{\omega^{k-1}[2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{k+1}(a),\omega^{2k-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

855: &+\frac{[2\omega^{\frac{3-k}{2}+\mu(G^{k-1}(a))}+o(1)]\exp\left[L(a,\lambda)-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)\right]}{\omega^{\frac{1}{2}} [2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{k+1}(a),\omega^{2k-m}\lambda)+L(a,\lambda)\right]}\nonumber\\

856: &\times[2i\omega^{-\frac{k+1}{2}}+o(1)]\exp\left[L(G^{2k+2}(a),\omega^{2k-m-1}\lambda)+L(G^{k+1}(a),\omega^{-1}\lambda)\right]\nonumber\\

857: &=[-2i\omega^{\frac{3-k}{2}+\mu(G^{-1}(a))+\mu(G^{k}(a))}+o(1)]\exp\left[-L(G^{k+1}(a),\omega^{-1}\lambda)-L(a,\lambda)\right]\nonumber\\

858: &+[2\omega^{\frac{2-k}{2}+\mu(G^{k-1}(a))}+o(1)]\exp\left[L(G^{2k+2}(a),\omega^{-2}\lambda)-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)\right]\nonumber\\

859: &=-[2i\omega^{\frac{3-k}{2}+\frac{m}{2}}+o(1)]\exp\left[-L(G^{k+1}(a),\omega^{-1}\lambda)-L(a,\lambda)\right]\nonumber\\

860: &+[2\omega^{\frac{2-k}{2}+\frac{m}{4}}+o(1)]\exp\left[L(G^{2k+2}(a),\omega^{-2}\lambda)-L(G^{k-1}(a),\omega^{2(k-1)}\lambda)\right],\nonumber

861: \end{align}

862: where we use \eqref{up_asy} for $W_{-1,k-1}(G(a),\cdot)$, and use \eqref{sec_eq1} for $W_{0,k+1}(a,\cdot)$ and  $W_{0,k+1}(G^{k+1}(a),\cdot)$. Finally, we use $\omega^{\frac{m+2}{4}}=i$, to get the asymptotic expansion of $W_{-1,\lfloor\frac{m}{2}\rfloor}(a,\lambda)$.

863: This  completes the proof.

864: \end{proof}

865: 

866:  The {\it order of an entire function} $g$ is defined by

867: $$\limsup_{r\rightarrow \infty}\frac{\log \log M(r,g)}{\log r},$$

868: where $M(r, g)=\max \{|g(re^{i\theta})|: 0\leq \theta\leq 2\pi\}$ for $r>0$.

869: If for some positive real numbers $\sigma,\, c_1,\, c_2$, we have $\exp[c_1 r^{\sigma}]\leq M(r,g)\leq  \exp[c_2 r^{\sigma}]$ for all large $r$, then the order of $g$ is $\sigma$. 

870: \begin{corollary}\label{zero_free}

871: Let $1\leq n\leq \frac{m}{2}$. Then

872:  the entire functions $W_{-1,n}(a,\cdot)$  are of order $\frac{1}{2}+\frac{1}{m}$, and hence they have infinitely many zeros in the complex plane. Moreover,  $W_{-1,n}(a,\cdot)$ have at most finitely many zeros in the sector 

873: $$

874: -\frac{2(n-1)\pi}{m+2}+\delta\leq\arg(\lambda)\leq\pi+\frac{4\pi}{m+2}-\delta.

875: $$

876: \end{corollary}

877: 

878: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

879: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

880: \section{Relation between eigenvalues of $H_{\ell,P}$ and zeros of $W_{-1,n}(a,\cdot)$}\label{asymp_eigen}

881: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

882: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

883: In this section, we will relate the eigenvalues of $H_{\ell,P}$ with zeros of some entire function $W_{-1,n}(a,\cdot)$.

884: 

885: Suppose $\ell=2k-1$ is odd with $1\leq \ell=2k-1\leq m-1$. Then \eqref{ptsym} becomes \eqref{rotated} by the scaling $v(z,\lambda)=u(-iz,\lambda)$, and $v(\cdot,\lambda)$ decays in the Stokes sectors $S_{-k}$ and $S_{k}$. Since $f_{k-1}$ and $f_{k}$ are linearly independent, for some $D_k$ and $\widetilde{D}_k$ one can write

886: $$f_{-k}(z,a,\lambda)=D_k(a,\lambda)f_{k-1}(z,a,\lambda)+\widetilde{D}_k(a,\lambda)f_{k}(z,a,\lambda).$$

887: Then one finds

888: $$

889: D_k(a,\lambda)=\frac{W_{-k,k}(a,\lambda)}{W_{k-1,k}(a,\lambda)}\quad\text{and}\quad \widetilde{D}_k(a,\lambda)=\frac{W_{-k,k-1}(a,\lambda)}{W_{k,k-1}(a,\lambda)}.

890: $$

891: Also it is easy to see that $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $D_k(a,\lambda)=0$ if and only if $W_{-k,k}(a,\lambda)=0$.

892: Since $W_{-k,k}(a,\lambda)=\omega^{k-1}W_{-1,2k-1}(G^{-k+1}(a),\omega^{-2k+2}\lambda)$, by Corollary \ref{zero_free} $W_{-k,k}(a,\lambda)$ has at most finitely many zeros in the sector 

893: $

894: -\frac{2(2k-2)\pi}{m+2}+\delta\leq\arg(\omega^{-2k+2}\lambda)\leq\pi+\frac{4\pi}{m+2}-\delta,

895: $

896: that is,

897: $$

898: \delta\leq\arg(\lambda)\leq\pi+\frac{4k\pi}{m+2}-\delta.

899: $$

900: This is true for all $a\in\C^{m-1}$ fixed. 

901: 

902: Next, by symmetry one can show that 

903: $W_{-k,k}(a,\lambda)$ has at most finitely many zeros in the sector $\pi\leq\arg(\lambda)\leq 2\pi-\delta$. We look at $H_{\ell,P}$ with $P(z)$ replaced by $\overline{P(\overline{z})}$ whose coefficient vector is $\overline{a}:=(\overline{a}_1,\,\overline{a}_2,\dots,\overline{a}_{m-1})$. Then one sees that

904: $W_{-k,k}(a,\lambda)=0$ if and only if $W_{-k,k}(\overline{a},\overline{\lambda})=0$. The latter has at most finitely many zeros in the sector $\delta\leq\arg(\overline{\lambda})\leq\pi.$  Thus, $W_{-k,k}(a,\lambda)$ has at most finitely many zeros in the sector $\pi\leq\arg(\lambda)\leq 2\pi-\delta$, and has infinitely many zeros in the sector $|\arg(\lambda)|\leq\delta$ since it is an entire function of order $\frac{1}{2}+\frac{1}{m}\in(0,1)$.

905: 

906: Suppose that $\ell=2k$ is even with $1\leq \ell=2k\leq m-1$. Then \eqref{ptsym} becomes \eqref{rotated2} by the scaling $y(z,\lambda)=u(-i\omega^{-\frac{1}{2}}z,\lambda)$, and $y(\cdot,\lambda)$ decays in the Stokes sectors $S_{-k}$ and $S_{k+1}$. We then see that the coefficient vector $\widetilde{a}$ of the polynomial $\omega^{-1}P(\omega^{-\frac{1}{2}}z)$ becomes

907: $$\widetilde{a}=G^{-\frac{1}{2}}(a).$$

908: 

909: Now one can express $f_{-k}$ as a linear combination of 

910:  $f_{k}$ and $f_{k+1}$ as follows. 

911: $$f_{-k}(z,\widetilde{a} ,\omega^{-1}\lambda)=\frac{W_{-k,k+1}(\widetilde{a},\omega^{-1}\lambda)}{W_{k,k+1}(\widetilde{a},\omega^{-1}\lambda)}f_{k}(z,\widetilde{a},\omega^{-1}\lambda)+\frac{W_{-k,k}(\widetilde{a},\omega^{-1}\lambda)}{W_{k+1,k}(\widetilde{a},\omega^{-1}\lambda)}f_{k+1}(z,\widetilde{a},\omega^{-1}\lambda).$$

912: Thus,  $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $W_{-k,k+1}(\widetilde{a},\omega^{-1}\lambda)=0$.

913: Since

914: $$W_{-k,k+1}(\widetilde{a},\omega^{-1}\lambda)=\omega^{k-1}W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda),$$ 

915:   by Corollary \ref{zero_free},

916:  $W_{-k,k+1}(\widetilde{a},\omega^{-1}\lambda)$ has at most finitely many zeros in the sector 

917: $

918: -\frac{2(2k-1)\pi}{m+2}+\delta\leq\arg(\omega^{-2k+1}\lambda)\leq\pi+\frac{4\pi}{m+2}-\delta,

919: $

920: that is,

921: $$

922: \delta\leq\arg(\lambda)\leq\pi+\frac{4k\pi}{m+2}-\delta.

923: $$

924: This is true for each $a\in\C^{m-1}$. So one can show that 

925: $W_{-k,k+1}(\widetilde{a},\omega^{-1}\lambda)$ has at most finitely many zeros in the sector $\pi\leq\arg(\lambda)\leq 2\pi-\delta$ by symmetry, similar to the cases when $\ell$ is odd.   Thus, $W_{-k,k+1}(a,\lambda)$  has infinitely many zeros in the sector $|\arg(\lambda)|\leq\delta$.

926: 

927: %%%%%%%%%%%%%%%%%%%%

928: 

929: \section{Proof of Theorem \ref{main_thm1} when $1\leq\ell<\lfloor\frac{m}{2}\rfloor$}\label{sec_7}

930: In this section, we prove Theorem \ref{main_thm1} for $1\leq\ell<\lfloor\frac{m}{2}\rfloor$. 

931: 

932: We first treat the case when $\ell$ is odd. 

933: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $1\leq\ell<\lfloor\frac{m}{2}\rfloor$ is odd]

934: Suppose that $1\leq \ell=2k-1<\lfloor\frac{m}{2}\rfloor$.

935: Recall that when $\ell$ is odd, $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if 

936:  $W_{-k,k}(a,\lambda)=0$. Then since $W_{-k,k}(a,\lambda)$ has at most finitely many zeros outside the sector $|\arg(\lambda)|\leq\delta$, 

937: all the eigenvalues $\lambda$ of  $H_{\ell,P}$ lie in the sector $|\arg(\lambda)|\leq\delta$ if $|\lambda|$ is large enough.

938: 

939: Since 

940: \begin{equation}\nonumber

941: W_{-k,k}(a,\lambda)=\omega^{k-1}W_{-1,2k-1}(G^{-k+1}(a),\omega^{-2k+2}\lambda),

942: \end{equation}

943:  we will use \eqref{zero_asy} to investigate asymptotics of large eigenvalues. Suppose that $W_{-k,k}(a,\lambda)=0$ and $|\lambda|$ is large enough. Then from \eqref{zero_asy} with $n=2k-1$, and with $a$ and $\lambda$ replaced by $G^{-k+1}(a)$ and $\omega^{-2k+2}\lambda$, respectively, we have

944: \begin{align}

945: \left[1+o(1)\right]&\exp\left[L(G^{k}(a),\omega^{2k}\lambda)-L(G^{-k}(a),\omega^{-2k}\lambda)\right]\nonumber\\

946: &\times\exp\left[L(G^{k-1}(a),\omega^{2k-2}\lambda)-L(G^{-k+1}(a),\omega^{-2k+2}\lambda)\right]=-\omega^{2\nu(G^k(a))}.\nonumber

947: \end{align}

948: Next the term $\left[1+o(1)\right]$ can be  absorbed into the exponential function so that we get 

949: \begin{align}

950: &\exp\left[L(G^{k}(a),\omega^{2k}\lambda)-L(G^{-k}(a),\omega^{-2k}\lambda)\right]\nonumber\\

951: &\times\exp\left[L(G^{k-1}(a),\omega^{2k-2}\lambda)-L(G^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-\omega^{2\nu(G^k(a))}.\label{sim_eq1}

952: \end{align}

953: 

954: For each odd integer $1\leq\ell=2k-1<\lfloor\frac{m}{2}\rfloor$, we define

955: \begin{align}

956: h_{m,\ell}(\lambda)=& L(G^{k}(a),\omega^{2k}\lambda)-L(G^{-k}(a),\omega^{-2k}\lambda)\nonumber\\

957: &+L(G^{k-1}(a),\omega^{2k-2}\lambda)-L(G^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1),\nonumber

958: \end{align}

959: where the error term $o(1)$ is the same as that in \eqref{sim_eq1}.

960: Then by Corollary \ref{lemma_decay},

961: \begin{align}

962: h_{m,\ell}(\lambda)&=K_m\left(e^{\frac{2k\pi}{m}i}-e^{-\frac{2k\pi}{m}i}+e^{\frac{2(k-1)\pi}{m}i}-e^{-\frac{2(k-1)\pi}{m}i}\right)\lambda^{\frac{m+2}{2m}}(1+o(1))\nonumber\\

963: &=2iK_m\left(\sin\left(\frac{2k\pi}{m}\right)+\sin\left(\frac{2k\pi}{m}-\frac{2\pi}{m}\right)\right)\lambda^{\frac{m+2}{2m}}(1+o(1))\nonumber\\

964: &=4iK_m\cos\left(\frac{\pi}{m}\right)\sin\left(\frac{(2k-1)\pi}{m}\right)\lambda^{\frac{m+2}{2m}}(1+o(1))\quad\text{as}\quad\lambda\to\infty,\label{assco_eq}

965: \end{align}

966: in the sector $|\arg(\lambda)|\leq\delta$. Since $K_m>0$ and $0<\frac{(2k-1)\pi}{m}<\pi$, the function $h_{m,\ell}(\cdot)$ maps the region $|\lambda|\geq M_1$ for some large $M_1$ and $|\arg(\lambda)|\leq\delta$ into a region containing $|\lambda|\geq M_2$ for some large $M_2$ and $|\arg(\lambda)-\frac{\pi}{2}|\leq\varepsilon_1$ for some $\varepsilon_1>0$. So there exists a sequence of $\lambda_{n}$ in the sector $|\arg(\lambda)|\leq\delta$ such that $\exp\left[h_{m,\ell}(\lambda_{n})\right]= -\omega^{2\nu(G^k(a))}$ for all large $n\in\N$.

967: 

968: So, from \eqref{sim_eq1}, we have

969: \begin{equation}\nonumber

970: h_{m,\ell}(\lambda_{n})

971: =\ln\left(-\omega^{2\nu(G^k(a))}\right)

972: =\left(\frac{4\nu(G^k(a))}{m+2}+2n+1\right)\pi i,\quad\text{for all large $n\in\N$}.

973: \end{equation}

974: Thus, since 

975: \begin{align}

976: &-\frac{\nu(G^{k}(a))}{m}\ln(\omega^{2k}\lambda)+\frac{\nu(G^{-k}(a))}{m}\ln(\omega^{-2k}\lambda)\nonumber\\

977: &-\frac{\nu(G^{k-1}(a))}{m}\ln(\omega^{2k-2}\lambda)+\frac{\nu(G^{-k+1}(a))}{m}\ln(\omega^{-2k+2}\lambda)\nonumber\\

978: =& \frac{\nu(G^{k}(a))}{m}\left(-\ln(\omega^{2k}\lambda)+\ln(\omega^{-2k}\lambda)

979: +\ln(\omega^{2k-2}\lambda)-\ln(\omega^{-2k+2}\lambda)\right)\nonumber\\

980: =& -\frac{8\nu(G^{k}(a))}{m(m+2)}\pi i,\nonumber

981: \end{align}

982: from Lemma \ref{asy_lemma},

983: \begin{equation}\label{const_term}

984: \left(\frac{4\nu(G^k(a))}{m+2}+2n+1\right)\pi i=\sum_{j=0}^{\lfloor\frac{m+2}{2}\rfloor}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}-\frac{8\nu(G^k(a))}{m(m+2)}\pi i+o(1),

985: \end{equation}

986: where for $0\leq j\leq \frac{m+2}{2}$, the coefficients $d_{\ell,j}(a)$ are given by

987: \begin{align}

988: d_{\ell,j}(a)&=K_{m,j}(G^{k}(a))\omega^{2k\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k}(a))\omega^{-2k\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

989: &+K_{m,j}(G^{k-1}(a))\omega^{2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k+1}(a))\omega^{-2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}.\nonumber

990: \end{align}

991: Since $\ell=2k-1$, for $1\leq j\leq\frac{m+2}{2}$,

992: \begin{align}

993: d_{\ell,j}(a)&=K_{m,j}(G^{\frac{\ell+1}{2}}(a))\omega^{(\ell+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-\frac{\ell+1}{2}}(a))\omega^{-(\ell+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

994: &+K_{m,j}(G^{\frac{\ell-1}{2}}(a))\omega^{(\ell-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-\frac{\ell-1}{2}}(a))\omega^{-(\ell-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

995: &=4i\sum_{k=1}^j(-1)^kK_{m,j,k}b_{j,k}(a)\sin\left(\frac{(1-j)\ell\pi}{m}\right)\cos\left(\frac{(1-j)\pi}{m}\right),\nonumber

996: \end{align}

997: where we used Lemma \ref{lemma_25} along with \eqref{K_deF}.

998: Notice that \eqref{assco_eq} shows 

999: $$d_{\ell,0}(a)=4iK_{m,0}\sin\left(\frac{\ell\pi}{m}\right)\cos\left(\frac{\pi}{m}\right).$$

1000: If $m$ is even, then we redefine $d_{\ell,\frac{m+2}{2}}(a)$ as the sum of $d_{\ell,\frac{m+2}{2}}(a)=0$ above and $-\frac{4\nu(G^k(a))}{m}\pi i$ (c. f., equation \eqref{const_term}).

1001: Finally, note that $\nu(G^k(a))=(-1)^k\nu(a)=\eta_{m,\ell}(a).$ This completes the proof.

1002: \end{proof}

1003: 

1004: Next we prove Theorem ~\ref{main_thm1} for $1\leq\ell<\lfloor\frac{m}{2}\rfloor$ is even.

1005: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $1\leq\ell<\lfloor\frac{m}{2}\rfloor$ is even]

1006: 

1007: Proof is very similar to the case when $1\leq\ell<\lfloor\frac{m}{2}\rfloor$ is even. Let $\ell=2k$ for some $k\in\N$. 

1008: 

1009: Recall that  $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if  $W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)=0$. Then from \eqref{zero_asy},  we have

1010: \begin{align}

1011: &\exp\left[L(G^{k+1}(\widetilde{a}),\omega^{2k+1}\lambda)-L(G^{-k}(\widetilde{a}),\omega^{-2k-1}\lambda)\right]\nonumber\\

1012: &\times\exp\left[L(G^{k}(\widetilde{a}),\omega^{2k-1}\lambda)-L(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)+o(1)\right]=-1,\label{sim_eq2}

1013: \end{align}

1014: where we used $\left[1+o(1)\right]=\exp[o(1)]$ again.

1015: 

1016: Like in the case when $1\leq\ell<\lfloor\frac{m}{2}\rfloor$ is odd, from Lemma \ref{asy_lemma},

1017: \begin{equation}\nonumber

1018: \left(2n+1\right)\pi i=\sum_{j=0}^{\frac{m+2}{2}}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1),

1019: \end{equation}

1020: where for $0\leq j\leq \frac{m+2}{2}$, 

1021: \begin{align}

1022: d_{\ell,j}(a)&=K_{m,j}(G^{k+1}(\widetilde{a}))\omega^{(2k+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k}(\widetilde{a}))\omega^{-(2k+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1023: &+K_{m,j}(G^{k}(\widetilde{a}))\omega^{(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k+1}(\widetilde{a}))\omega^{-(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1024: &=\sum_{k=1}^j(-1)^kK_{m,j,k}b_{j,k}(a)\left(\omega^{-j\frac{\ell+1}{2}+(\ell+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{j\frac{\ell+1}{2}-(\ell+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right.\nonumber\\

1025: &\qquad\qquad\qquad\qquad\qquad\qquad\left.+\omega^{-j\frac{\ell-1}{2}+(\ell-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{j\frac{\ell-1}{2}-(\ell-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right)\nonumber\\

1026: &=4i\sum_{k=1}^j(-1)^kK_{m,j,k}b_{j,k}(a)\sin\left(\frac{(1-j)\ell\pi}{m}\right)\cos\left(\frac{(1-j)\pi}{m}\right),\nonumber

1027: \end{align}

1028: where we used $\widetilde{a}=G^{-\frac{1}{2}}(a)$ and $\ell=2k$ as well.

1029: \end{proof}

1030: 

1031: %%%%%%%%%%%%%%%%%%%%%

1032: %%%%%%%%%%%%%%%%%%%%%%

1033: \section{Proof of Theorem \ref{main_thm1} when $\ell=\lfloor\frac{m}{2}\rfloor$ and when $\frac{m}{2}<\ell\leq m-1$.}\label{sec_8}

1034: In this section, we prove Theorem \ref{main_thm1} for $\ell=\lfloor\frac{m}{2}\rfloor$. We first prove the theorem when $m$ is even, and later, we will treat the cases when $m$ is odd. Then at the end of the section, we will prove the theorem when $\frac{m}{2}<\ell\leq m-1$, by scaling.

1035: \subsection{When $m$ is even} We further divide the case into when $\ell$ is odd and when $\ell$ is even.

1036: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $m$ is even and $\ell=\lfloor\frac{m}{2}\rfloor$ is odd]

1037: Suppose that $m$ is even and $\ell=\frac{m}{2}=2k-1$ for some $k\in\N$. Since $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $W_{-k,k}(a,\lambda)=0$ if and only if  $W_{-1,2k-1}(G^{-k+1}(a),\omega^{-2k+2}\lambda)=0$.

1038: 

1039: Suppose that $W_{-1,2k-1}(G^{-k+1}(a),\omega^{-2k+2}\lambda)=0$. If $|\lambda|$ is large enough, then from  \eqref{thm_eq1},

1040: \begin{align}

1041: &\exp\left[L(G^{3k-1}(a),\omega^{2k-2}\lambda)-L(G^{k+1}(a),\omega^{-2k+2}\lambda)\right]\nonumber\\

1042: &\times\exp\left[L(G^{k-1}(a),\omega^{2k-2}\lambda)-L(G^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-\omega^{4\nu(G^{k}(a))}.\nonumber

1043: \end{align}

1044: 

1045: From Lemma \ref{asy_lemma},

1046: \begin{equation}\nonumber

1047: \left(\frac{8\nu(G^{k}(a))}{m+2}+2n+1\right)\pi i=\sum_{j=0}^{\frac{m+2}{2}}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+\frac{16(k-1)\nu(G^{k}(a))}{m(m+2)}\pi i+o(1),

1048: \end{equation}

1049: where for $0\leq j\leq \frac{m+2}{2}$, the coefficients $d_{\ell,j}(a)$ are given by

1050: \begin{align}

1051: d_{\ell,j}(a)&=K_{m,j}(G^{3k-1}(a))\omega^{2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{k+1}(a))\omega^{-2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1052: &+K_{m,j}(G^{k-1}(a))\omega^{2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k+1}(a))\omega^{-2(k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1053: &=K_{m,j}(G^{\frac{3m+2}{4}}(a))\omega^{\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{\frac{m+6}{4}}(a))\omega^{-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1054: &+K_{m,j}(G^{\frac{m-2}{4}}(a))\omega^{\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-\frac{m-2}{4}}(a))\omega^{-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1055: &=K_{m,j}(a)\left(\omega^{-j\frac{3m+2}{4}+\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{-j\frac{m+6}{4}-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right.\nonumber\\

1056: &\qquad\qquad\qquad\qquad\qquad\qquad\left.\omega^{-j\frac{m-2}{4}+\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{j\frac{m-2}{4}-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right)\nonumber\\

1057: &=4iK_{m,j}(a)\sin\left(\frac{(1-j)\pi}{2}\right)\cos\left(\frac{(1-j)\pi}{m}\right),\nonumber

1058: \end{align}

1059: where we used Lemma \ref{lemma_25} as well as some other thing as before. In addition, we used 

1060: \begin{align}

1061: &\omega^{-j\frac{3m+2}{4}+\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}=-\omega^{j\frac{m+2}{4}-\frac{m+2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)},\nonumber\\

1062: &\omega^{-j\frac{m+6}{4}-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}=-\omega^{-j\frac{m+2}{4}+\frac{m+2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}.\label{sine_eq}

1063: \end{align}

1064: \end{proof}

1065: 

1066: 

1067: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $m$ is even and $\ell=\lfloor\frac{m}{2}\rfloor$ is even]

1068: Let $\ell=\frac{m}{2}=2k$ for some $k\in\N$. Then

1069: $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)=0$.

1070: If $W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)=0$ and $|\lambda|$ large enough, then from  \eqref{thm_eq1},

1071: \begin{align}

1072: &\exp\left[L(G^{3k+1}(\widetilde{a}),\omega^{2k-1}\lambda)-L((G^{k+2}(\widetilde{a}),\omega^{-2k+1}\lambda)\right]\nonumber\\

1073: &\times\exp\left[L(G^{k}(\widetilde{a}),\omega^{2k-1}\lambda)-L(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)+o(1)\right]=-1.\nonumber

1074: \end{align}

1075: 

1076: Then like before,

1077: \begin{equation}\nonumber

1078: \left(2n+1\right)\pi i=\sum_{j=0}^{\frac{m+2}{2}}d_{\ell,j}(\widetilde{a})\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1),

1079: \end{equation}

1080: where for $0\leq j\leq \frac{m+2}{2}$, 

1081: \begin{align}

1082: d_{\ell,j}(a)&=K_{m,j}(G^{3k+1}(\widetilde{a}))\omega^{(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}((G^{k+2}(\widetilde{a}))\omega^{-(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1083: &+K_{m,j}(G^{k}(\widetilde{a}))\omega^{(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-k+1}(\widetilde{a}))\omega^{-(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1084: &=K_{m,j}(G^{\frac{3m+2}{4}}(a))\omega^{\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}((G^{\frac{m+6}{4}}(a))\omega^{-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1085: &+K_{m,j}(G^{\frac{m-2}{4}}(a))\omega^{\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-\frac{m-2}{4}}(a))\omega^{-\frac{m-2}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1086: &=4i\sum_{k=1}^j(-1)^kK_{m,j,k}b_{j,k}(a)\sin\left(\frac{(1-j)\pi}{2}\right)\cos\left(\frac{(1-j)\pi}{m}\right).\nonumber

1087: \end{align}

1088: \end{proof}

1089: 

1090: 

1091: \subsection{When $m$ is odd} We divide the case into when $\ell$ is odd and when $\ell$ is even.

1092: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $m$ and $\ell=\lfloor\frac{m}{2}\rfloor$ are odd]

1093: Let $m$ and $\ell=\frac{m-1}{2}=2k-1$ be odd. Suppose that $\lambda$ is an eigenvalue of $H_{\ell,P}$. Since $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $W_{-1,2k-1}(G^{-k+1}(a),\omega^{-2k+2}\lambda)=0$,

1094:  if  $|\lambda|$ is large enough, then from  \eqref{thm_eq2},

1095: \begin{align}

1096: &\exp\left[L(G^{k-1}(a),\omega^{2k-2}\lambda)-L(G^{3k+1}(a),\omega^{-2k}\lambda)\right]\nonumber\\

1097: &\times\exp\left[-L(G^{k+1}(a),\omega^{-2k+1}\lambda)-L(G^{-k+1}(a),\omega^{-2k+2}\lambda)+o(1)\right]=-1.\nonumber

1098: \end{align}

1099: 

1100: Then, 

1101: from Lemma \ref{asy_lemma},

1102: \begin{equation}\nonumber

1103: \left(2n+1\right)\pi i=\sum_{j=0}^{\frac{m+1}{2}}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1),

1104: \end{equation}

1105: where for $0\leq j\leq \frac{m+2}{2}$, 

1106: \begin{align}

1107: d_{\ell,j}(a)&=K_{m,j}(G^{\frac{m-3}{4}}(a))\omega^{\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{\frac{3m+7}{4}}(a))\omega^{-\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1108: &-K_{m,j}(G^{\frac{m+5}{4}}(a))\omega^{-\frac{m-1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{-\frac{m-3}{4}}(a))\omega^{-\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1109: &=K_{m,j}(a)\left(\omega^{-j\frac{m-3}{4}+\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{-j\frac{3m+7}{4}-\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right.\nonumber\\

1110: &\left.-\omega^{-j\frac{m+5}{4}-\frac{m-1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-\omega^{j\frac{m-3}{4}-\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\right)\nonumber\\

1111: &=2iK_{m,j}(a)\left(\sin\left(\frac{(m-3)(1-j)\pi}{2m}\right)+\sin\left(\frac{(m+1)(1-j)\pi}{2m}\right)\right)\nonumber\\

1112: &=2iK_{m,j}(a)\sin\left(\frac{(m-1)(1-j)\pi}{2m}\right)\cos\left(\frac{(1-j)\pi}{m}\right),\nonumber

1113: \end{align}

1114: where we used 

1115: \begin{align}

1116: &\omega^{-j\frac{3m+7}{4}-\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}=\omega^{j\frac{m+1}{4}-\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)},\nonumber\\

1117: & \omega^{-j\frac{m+5}{4}-\frac{m-1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}=-\omega^{-j\frac{m+1}{4}+\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}.\label{sine_eq2}

1118: \end{align}

1119: \end{proof}

1120: 

1121: 

1122: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $m$ is odd and $\ell=\lfloor\frac{m}{2}\rfloor$ is even]

1123: Let $\ell=\frac{m-1}{2}=2k$ for some $k\in\N$. Then

1124: $\lambda$ is an eigenvalue of $H_{\ell,P}$ if and only if $W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)=0$.

1125: If $W_{-1,2k}(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)=0$ and $|\lambda|$ large enough, then from  \eqref{thm_eq2},

1126: \begin{align}

1127: &\exp\left[L(G^{k}(\widetilde{a}),\omega^{2k-1}\lambda)-L(G^{-k+1}(\widetilde{a}),\omega^{-2k+1}\lambda)\right]\nonumber\\

1128: &\times\exp\left[-L(G^{3k+3}(\widetilde{a}),\omega^{-2k-1}\lambda)-L(G^{k+2}(\widetilde{a}),\omega^{-2k}\lambda)+o(1)\right]=-1.\nonumber

1129: \end{align}

1130: 

1131: Then, 

1132: from Lemma \ref{asy_lemma},

1133: \begin{equation}\nonumber

1134: \left(2n+1\right)\pi i=\sum_{j=0}^{\frac{m+1}{2}}d_{\ell,j}(\widetilde{a})\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1),

1135: \end{equation}

1136: where for $0\leq j\leq \frac{m+2}{2}$, 

1137: \begin{align}

1138: d_{\ell,j}(a)&=K_{m,j}(G^{k}(\widetilde{a}))\omega^{(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}((G^{-k+1}(\widetilde{a}))\omega^{-(2k-1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1139: &-K_{m,j}(G^{3k+3}(\widetilde{a}))\omega^{-(2k+1)\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{k+2}(\widetilde{a}))\omega^{-2k\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1140: &=K_{m,j}(G^{\frac{m-3}{4}}(a))\omega^{\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}((G^{-\frac{m-3}{4}}(a))\omega^{-\frac{m-3}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1141: &-K_{m,j}(G^{\frac{3m+7}{4}}(a))\omega^{-\frac{m+1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}-K_{m,j}(G^{\frac{m+5}{4}}(a))\omega^{-\frac{m-1}{2}\left(\frac{1}{2}+\frac{1-j}{m}\right)}\nonumber\\

1142: &=\sum_{k=1}^j(-1)^kK_{m,j,k}b_{j,k}(a)\sin\left(\frac{(m-1)(1-j)\pi}{2m}\right)\cos\left(\frac{(1-j)\pi}{m}\right).\nonumber

1143: \end{align}

1144: \end{proof}

1145: Theorem \ref{main_thm1} for $1\leq\ell\leq\frac{m}{2}$ has been proved. Next we prove Theorem \ref{main_thm1} for $\frac{m}{2}<\ell\leq m-1$, by scaling $z\mapsto -z$.

1146: \subsection{When $\frac{m}{2}<\ell\leq m-1$}

1147: \begin{proof}[Proof of Theorem ~\ref{main_thm1} when $\frac{m}{2}<\ell\leq m-1$]

1148: Suppose that  $\ell>\frac{m}{2}$. If $u(\cdot,\lambda)$ is an eigenfunction of $H_{\ell,P}$, then $v(z,\lambda)=u(-z,\lambda)$ solves 

1149: \begin{equation}\nonumber

1150: -v^\dd(z,\lambda)+\left[(-1)^{-\ell}(-iz)^m-P(-iz)\right]v(z,\lambda)

1151: =\lambda v(z,\lambda),

1152: \end{equation}

1153: and

1154: \begin{equation}\nonumber

1155: \text{$v(z)\rightarrow 0$ exponentially, as $z\rightarrow \infty$ along the two rays}\quad \arg z=-\frac{\pi}{2}\pm \frac{((m-\ell)+1)\pi}{m+2}.

1156: \end{equation}

1157: The coefficient vector of $P(-z)$ is

1158: $((-1)^{m-1}a_1,(-1)^{m-2}a_2,\dots,(-1)^{1}a_{m-1}).$ Certainly, 

1159: $$\sin\left(\frac{(m-\ell)\pi}{m}\right)=\sin\left(\frac{\ell\pi}{m}\right),\quad \sin\left(\frac{(1-j)(m-\ell)\pi}{m}\right)=(-1)^j\sin\left(\frac{(1-j)\ell\pi}{m}\right).$$ Also, one can find from \eqref{bjk_def} that for $1\leq k\leq j\leq\frac{m+2}{2}$,

1160: $$b_{j,k}((-1)^{m-1}a_1,(-1)^{m-2}a_2,\dots,-a_{m-1})=(-1)^{mk-j}b_{j,k}(a_1,a_2,\dots,a_{m-1}).$$

1161: Moreover,

1162: $d_{m-\ell,j}((-1)^{m-1}a_1,(-1)^{m-2}a_2,\dots,-a_{m-1})=d_{\ell,j}(a_1,a_2,\dots,a_{m-1}).$

1163: This completes proof of Theorem \ref{main_thm1}.

1164: \end{proof}

1165: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1166: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1167: \section{Proof of Theorem \ref{eigen_asy}}\label{sec_9}

1168: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1169: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1170: In this section, we prove Theorem \ref{eigen_asy}.

1171: 

1172: We begin with

1173: \begin{equation}\nonumber

1174: \left(2n+1\right)\pi i=\sum_{j=0}^{\lfloor\frac{m+2}{2}\rfloor}d_{\ell,j}(a)\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1),\quad\text{as $n\to\infty$},

1175: \end{equation}

1176: \begin{equation}\nonumber

1177: \frac{\left(2n+1\right)\pi i}{d_{\ell,0}(a)}=\lambda_{n}^{\frac{1}{2}+\frac{1}{m}}+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}\frac{d_{\ell,j}(a)}{d_{\ell,0}(a)}\lambda_{n}^{\frac{1}{2}+\frac{1-j}{m}}+o(1).

1178: \end{equation}

1179: 

1180: Let 

1181: $$c_{j}(a)=\frac{d_{\ell,j}(a)}{d_{\ell,0}(a)},\quad 1\leq j\leq\frac{m+2}{2}.$$

1182: Introduce the decomposition

1183: $

1184: \lambda_{n}=\lambda_{n,0}+\lambda_{n,1},

1185: $

1186: where 

1187: \begin{equation}\nonumber

1188:  \lambda_{n,0}=\left(\frac{\left(2n+1\right)\pi i}{d_{\ell,0}(a)}\right)^{\frac{2m}{m+2}}\,\,\text{and}\quad \frac{\lambda_{n,0}}{\lambda_{n,0}}=o\left(1\right).

1189: \end{equation}

1190: Then  we have

1191: \begin{align}

1192: \lambda_{n,0}^{\frac{1}{2}+\frac{1}{m}}

1193: &=\lambda_{n,0}^{\frac{1}{2}+\frac{1}{m}}\left(1+\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^{\frac{1}{2}+\frac{1}{m}}+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{\frac{1}{2}+\frac{1-j}{m}}\left(1+\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^{\frac{1}{2}+\frac{1-j}{m}}+o(1)\nonumber\\

1194: &=\lambda_{n,0}^{\frac{1}{2}+\frac{1}{m}}\left(1+\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1}{m}\choose k}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\right)\nonumber\\

1195: &+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{\frac{1}{2}+\frac{1-j}{m}}\left(1+\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1-j}{m}\choose k}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\right)+o(1).\nonumber

1196: \end{align}

1197: Thus,

1198: \begin{align}

1199: 0&={\frac{1}{2}+\frac{1}{m}\choose 1}\frac{\lambda_{n,1}}{\lambda_{n,0}}+\sum_{k=2}^{\infty}{\frac{1}{2}+\frac{1}{m}\choose k}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\nonumber\\

1200: &+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}\left(1+\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1-j}{m}\choose k}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\right)+o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right),\nonumber

1201: \end{align}

1202: and hence

1203: \begin{align}

1204: &\frac{\lambda_{n,1}}{\lambda_{n,0}}+\sum_{k=2}^{\infty}\frac{{\frac{1}{2}+\frac{1}{m}\choose k}}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\nonumber\\

1205: &+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}\left(\sum_{k=1}^{\infty}\frac{{\frac{1}{2}+\frac{1-j}{m}\choose k}}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\left(\frac{\lambda_{n,1}}{\lambda_{n,0}}\right)^k\right)+o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right)\nonumber\\

1206: &=-\frac{1}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}.\label{asy_eq4}

1207: \end{align}

1208: Thus, one concludes

1209: $

1210: \frac{\lambda_{n,1}}{\lambda_{n,0}}=\lambda_{n,2}+\lambda_{n,3},

1211: $

1212: where

1213: \begin{equation}\label{ex_eq1}

1214: \lambda_{n,2}=-\frac{1}{{\frac{1}{2}+\frac{1}{m}\choose 1}}c_{1}(a)\lambda_{n,0}^{-\frac{1}{m}}\,\,\text{ and }\,\,\lambda_{n,3}=o\left(\lambda_{n,0}^{-\frac{1}{m}}\right).

1215: \end{equation}

1216: 

1217: Notice that $\lambda_{n,2}=0$ since $c_{1}(a)=0$. Hence, from  \eqref{ex_eq1} along with \eqref{asy_eq4} we have

1218: \begin{align}

1219: &\lambda_{n,3}+\sum_{{k_1}=2}^{\infty}\frac{{\frac{1}{2}+\frac{1}{m}\choose {k_1}}}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\lambda_{n,3}^{k_1}+\sum_{j=2}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}\left(\sum_{{k_1}=1}^{\infty}\frac{{\frac{1}{2}+\frac{1-j}{m}\choose {k_1}}}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\lambda_{n,3}^{k_1}\right)+o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right)\nonumber\\

1220: &=-\frac{1}{{\frac{1}{2}+\frac{1}{m}\choose 1}}\sum_{j=2}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}.\label{asy_eq5}

1221: \end{align}

1222: 

1223: Suppose that 

1224: $\frac{\lambda_{n,1}}{\lambda_{n,1}}=+\lambda_{n,4}+\lambda_{n,6}+\cdots+\lambda_{n,2s}+\lambda_{n,2s+1},$

1225: where $\lambda_{n,2s+1}=o\left(\lambda_{n,0}^{-\frac{s}{m}}\right)$ and

1226: $

1227: \lambda_{n,2t}=e_{t}(a)\lambda_{n,0}^{-\frac{t}{m}},\,\, 2\leq t\leq s<\frac{m+2}{2}$ for some  $e_{t}(a)\in\C$.

1228: Then from \eqref{asy_eq4}

1229: \begin{align}

1230: &\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1}{m}\choose k}\left(\lambda_{n,4}+\cdots+\lambda_{n,2s}+\lambda_{n,2s+1}\right)^k\nonumber\\

1231: &+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1-j}{m}\choose k}\left(\lambda_{n,4}+\cdots+\lambda_{n,2s}+\lambda_{n,2s+1}\right)^k+o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right)\nonumber\\

1232: &=-\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}.\nonumber

1233: \end{align}

1234: Hence,

1235: \begin{align}

1236: &\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1}{m}\choose k}\left(\lambda_{n,4}+\cdots+\lambda_{n,2s}\right)^k\nonumber\\

1237: &+\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}\sum_{k=1}^{\infty}{\frac{1}{2}+\frac{1-j}{m}\choose k}\left(\lambda_{n,4}+\cdots+\lambda_{n,2s}\right)^k\nonumber\\

1238: &=-\sum_{j=1}^{\lfloor\frac{m}{2}+1\rfloor}c_{j}(a)\lambda_{n,0}^{-\frac{j}{m}}-{\frac{1}{2}+\frac{1}{m}\choose 1}\lambda_{n,2s+1}+o\left(\lambda_{n,0}^{-\frac{s+2}{m}}\right)+o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right).\label{asy_eq11}

1239: \end{align}

1240: Next,

1241: \begin{align}

1242: &\left(\lambda_{n,4}+\cdots+\lambda_{n,2s}+\lambda_{n,2s+1}\right)^k\nonumber\\

1243: &=\left(e_{2}(a)\lambda_{n,0}^{-\frac{2}{m}}+e_{3}(a)\lambda_{n,0}^{-\frac{3}{m}}+\cdots+e_{s}(a)\lambda_{n,0}^{-\frac{s}{m}}+o\left(\lambda_{n,0}^{-\frac{s}{m}}\right)\right)^k\nonumber\\

1244: &=\sum_{k_1=0}^{k}{k\choose k_1}\left(e_{2}(a)\lambda_{n,0}^{-\frac{2}{m}}+e_{3}(a)\lambda_{n,0}^{-\frac{3}{m}}+\cdots+e_{s}(a)\lambda_{n,0}^{-\frac{s}{m}}\right)^{k-k_1}o\left(\lambda_{n,0}^{-\frac{k_1s}{m}}\right)\nonumber\\

1245: &=\left(e_{2}(a)\lambda_{n,0}^{-\frac{2}{m}}+e_{3}(a)\lambda_{n,0}^{-\frac{3}{m}}+\cdots+e_{s}(a)\lambda_{n,0}^{-\frac{s}{m}}\right)^{k}

1246: +o\left(\lambda_{n,0}^{-\frac{s+2}{m}}\right)\nonumber\\

1247: &=\sum_{\substack{i_p\geq 0,\,j_p\not=j_q\,\,\text{if}\,\, p\not=q\\i_1+\cdots+i_t=k}}\frac{k!}{i_1!\cdots i_t!}e_{j_1}(a)^{i_1}e_{j_2}(a)^{i_2}\cdots e_{j_t}(a)^{i_t}\lambda_{n,0}^{-\frac{i_1j_1+\cdots+i_tj_t}{m}}+o\left(\lambda_{n,0}^{-\frac{s+2}{m}}\right).\nonumber

1248: \end{align}

1249: We use this in \eqref{asy_eq11} to see that the left hand side of \eqref{asy_eq11} is a power series in $\lambda_{n,0}^{-\frac{1}{m}}$. 

1250: Then comparing coefficients of $\lambda_{n,0}^{-\frac{j}{m}}$, $1\leq j\leq s$, we have

1251: \begin{equation}\label{asy_eq12}

1252: -c_{j}(a)=\sum_{\substack{|\alpha|=k\\ \alpha\cdot\beta=j}}{\frac{1}{2}+\frac{1}{m}\choose k}\frac{k!}{\alpha!}e(a)^{\alpha}+\sum_{r=1}^{j-2}c_{r}(a)\sum_{\substack{|\alpha|=k\\ \alpha\cdot\beta=j-r}}{\frac{1}{2}+\frac{1-r}{m}\choose k}\frac{k!}{\alpha!}e(a)^{\alpha}.

1253: \end{equation}

1254: Moreover, if $\frac{s+1}{m}\leq \frac{1}{2}+\frac{1}{m}$ (i.\ e., $s+1\leq \frac{m+2}{2}$) then there exists some constant $e_{s+1}(a)\in\C$ such that

1255: \begin{equation}\label{asy_eq14}

1256: \lambda_{n,2s+1}=e_{s+1}(a)\lambda_{n,0}^{-\frac{s+1}{m}}+o\left(\lambda_{n,0}^{-\frac{s+1}{m}}\right).

1257: \end{equation}

1258: Now we let $\lambda_{n,2s+1}=\lambda_{n,2s+2}+\lambda_{n,2s+3}$ where $\lambda_{n,2s+2}=e_{s+1}(a)\lambda_{n,0}^{-\frac{s+1}{m}}$ and $\lambda_{n,2s+3}=o\left(\lambda_{n,0}^{-\frac{s+1}{m}}\right)$.

1259: 

1260: If $s+1> \frac{m+2}{2}$ then $\lambda_{n,0}^{-\frac{s+1}{m}}$ could be smaller than the error term $o\left(\lambda_{n,0}^{-\frac{1}{2}-\frac{1}{m}}\right)$ in \eqref{asy_eq11}, and hence we cannot deduce existence of $e_{s+1}(a)$ like we do in \eqref{asy_eq14}. This completes proof of Theorem \ref{eigen_asy}.

1261: 

1262: \begin{remark}

1263: A first few $e_{j}(a)$ are as follows.

1264: \begin{align}

1265: e_{2}(a)&=-\frac{2m}{m+2}\frac{d_{\ell,2}(a)}{d_{\ell,0}(a)},\quad

1266: e_{3}(a)=-\frac{2m}{m+2}\frac{d_{\ell,3}(a)}{d_{\ell,0}(a)},\nonumber\\

1267: e_{4}(a)&=-\frac{2m}{m+2}\frac{d_{\ell,4}(a)}{d_{\ell,0}(a)}+\frac{3m(m-2)}{(m+2)^2}\left(\frac{d_{\ell,2}(a)}{d_{\ell,0}(a)}\right)^2,\nonumber\\

1268: e_{5}(a)&=-\frac{2m}{m+2}\frac{d_{\ell,5}(a)}{d_{\ell,0}(a)}+\frac{4m(m^2-3m-3)}{(m+2)^3}\frac{d_{\ell,2}(a)}{d_{\ell,0}(a)}\frac{d_{\ell,3}(a)}{d_{\ell,0}(a)},\nonumber\\

1269: e_{6}(a)&=-\frac{2m}{m+2}\frac{d_{\ell,6}(a)}{d_{\ell,0}(a)}+\frac{m(m-6)}{(m+2)^2}\left(\frac{d_{\ell,3}(a)}{d_{\ell,0}(a)}\right)^2\nonumber\\

1270: &+\frac{2m(m-6)}{(m+2)^2}\frac{d_{\ell,2}(a)}{d_{\ell,0}(a)}\frac{d_{\ell,4}(a)}{d_{\ell,0}(a)}+\frac{m(m-2)(9m-2)}{3(m+2)^3}\left(\frac{d_{\ell,2}(a)}{d_{\ell,0}(a)}\right)^3.\nonumber

1271: \end{align}

1272: \end{remark}

1273: 

1274: 

1275: %%%%%%%%%%%%%%%%%%%%%%%%%% Appendix A %%%%%%%%%%%%%%%%%%%%%%%%

1276: \appendix

1277: \section{Computing $K_{m,j,k}$} \label{sA}

1278: \renewcommand{\theequation}{A.\arabic{equation}}

1279: \renewcommand{\thetheorem}{A.\arabic{theorem}}

1280: \setcounter{theorem}{0}

1281: \setcounter{equation}{0}

1282: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1283: \begin{theorem}

1284: Let $m\geq 3$ be an integer. Then 

1285: \begin{equation}\nonumber

1286: K_m=K_{m,0}=\int_0^{\infty}\left(\sqrt{1+t^m}-t^{\frac{m}{2}}\right)\,dt=\frac{\sqrt{\pi}}{2\cos\left(\frac{\pi}{m}\right)}\frac{\Gamma\left(1+\frac{1}{m}\right)}{\Gamma\left(\frac{3}{2}+\frac{1}{m}\right)}.

1287: \end{equation}

1288: \end{theorem}

1289: \begin{proof}

1290: Substitute $\sqrt{u}=\sqrt{1+t^m}-t^{\frac{m}{2}}$. Then 

1291: \begin{align}

1292: \int_0^{\infty}\left(\sqrt{1+t^m}-t^{\frac{m}{2}}\right)\,dt&=\frac{1}{2^{\frac{2}{m}}m}\int_0^1\left((1-u)^{\frac{2}{m}-1}u^{\frac{1}{2}-\frac{1}{m}-1}+(1-u)^{\frac{2}{m}-1}u^{\frac{3}{2}-\frac{1}{m}-1}\right)\,du\nonumber\\

1293: &=\frac{1}{2^{\frac{2}{m}}m}\left(B\left(\frac{2}{m},\frac{1}{2}-\frac{1}{m}\right)+B\left(\frac{2}{m},\frac{3}{2}-\frac{1}{m}\right)\right),\nonumber

1294: \end{align}

1295: where $B(z,w)$ is the beta function. Then we use the following to complete the proof.

1296: \begin{align}

1297: &\Gamma(z+1)=z\Gamma(z),\quad \Gamma(z)\Gamma(1-z)=-z\Gamma(-z)\Gamma(z)=\frac{\pi}{\sin(\pi z)}\nonumber\\

1298: &B(z,w)=\int_0^1(1-u)^{z-1}u^{w-1}\,du=\frac{\Gamma(z)\Gamma(w)}{\Gamma(z+w)},\quad \Gamma(2z)=\frac{2^{2z-\frac{1}{2}}}{\sqrt{2\pi}}\Gamma(z)\Gamma\left(z+\frac{1}{2}\right).\label{gamma_eq}

1299: \end{align}

1300: \end{proof}

1301: 

1302: \begin{theorem}

1303: Let $m\geq 3$ and $1\leq k\leq j\leq \frac{m+2}{2}$. Then

1304: \begin{align}\nonumber

1305: K_{m,j,k}&=\int_0^{\infty}\left(\frac{t^{mk-j}}{(1+t^m)^{k-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)\,dt\nonumber\\

1306: &=\left\{

1307:                     \begin{array}{cl}

1308: -\frac{2}{m}

1309: \quad &\text{if $j=k=1$},\\

1310: &\\

1311:                   -\frac{2k-1}{m+2-2j}B\left(k-\frac{j-1}{m},\,\frac{1}{2}+\frac{j-1}{m}\right)   \quad &\text{if $1\leq k\leq j\leq\frac{m+1}{2}$, $j\not=1$},\\

1312: &\\

1313:                  \frac{2}{m}\left(\ln 2-\frac{1}{1}-\frac{1}{3}-\dots-\frac{1}{2k-5}-\frac{1}{2k-3}\right)  \quad &\text{if $m$ is even, $1\leq k\leq j=\frac{m+2}{2}$.}

1314:                     \end{array}\right.\nonumber

1315: \end{align}

1316: \end{theorem}

1317: \begin{proof}

1318: The case when $j=k=1$ is an easy consequence of 

1319: $$\frac{d}{dt}\left(\sqrt{1+t^m}-t^{\frac{m}{2}}\right)=\frac{m}{2}\left(\frac{t^{m-1}}{(1+t^m)^{\frac{1}{2}}}-t^{\frac{m}{2}-1}\right).$$

1320: 

1321: Suppose that $1\leq k\leq j\leq\frac{m+1}{2}$, $j\not=1$. Then since

1322: \begin{align}

1323: &\frac{d}{dt}\left(\frac{t^{mk-(j-1)}}{(1+t^m)^{k-\frac{1}{2}}}-t^{\frac{m}{2}-(j-1)}\right)\nonumber\\

1324: &=(mk-(j-1))\left(\frac{t^{mk-j}}{(1+t^m)^{k-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)-m(k-\frac{1}{2})\left(\frac{t^{m(k+1)-j}}{(1+t^m)^{(k+1)-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right),\nonumber

1325: \end{align}

1326: we have

1327: \begin{align}

1328: \int_0^{\infty}\left(\frac{t^{mk-j}}{(1+t^m)^{k-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)dt

1329: &=\frac{m(k-1)-(j-1)}{m(k-1)-\frac{m}{2}}\int_0^{\infty}\left(\frac{t^{m(k-1)-j}}{(1+t^m)^{(k-1)-\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)dt\nonumber\\

1330: &=\frac{\Gamma\left(k-\frac{j-1}{m}\right)\Gamma\left(1-\frac{1}{2}\right)}{\Gamma\left(k-\frac{1}{2}\right)\Gamma\left(1-\frac{j-1}{m}\right)}\int_0^{\infty}\left(\frac{t^{m-j}}{(1+t^m)^{\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)dt.\nonumber

1331: \end{align}

1332: Next, we use the substitution $\sqrt{u}=\sqrt{1+t^m}-t^{\frac{m}{2}}$ to show 

1333: $$\int_0^{\infty}\left(\frac{t^{m-j}}{(1+t^m)^{\frac{1}{2}}}-t^{\frac{m}{2}-j}\right)dt=-\frac{2^{\frac{2(j-1)}{m}}}{m}B\left(1-\frac{2(j-1)}{m},\frac{1}{2}+\frac{(j-1)}{m}\right).$$

1334: Finally, we use equations in \eqref{gamma_eq} to complete the proof for $1\leq k\leq j\leq\frac{m+1}{2}$, $j\not=1$.

1335: 

1336: Finally, if $m$ is even and $j=\frac{m+2}{2}$, then we use integration by parts, for  $R>0$,

1337: \begin{equation}\nonumber

1338: \int_0^{R}\frac{t^{mk-\frac{m}{2}-1}}{\left(t^m+1\right)^{k-\frac{1}{2}}}dt=\left.\frac{1}{m\left(-k+\frac{3}{2}\right)}\frac{t^{m(k-1)-\frac{m}{2}}}{\left(t^m+1\right)^{(k-1)-\frac{1}{2}}}\right|_0^R+\int_0^Rt^{m(k-1)-\frac{m}{2}-1}\frac{1}{\left(t^m+1\right)^{(k-1)-\frac{1}{2}}}dt,

1339: \end{equation}

1340: and hence

1341: $$\int_0^{\infty}\left(\frac{t^{mk-\frac{m}{2}-1}}{\left(t^m+1\right)^{k-\frac{1}{2}}}-\frac{t^{m(k-1)-\frac{m}{2}-1}}{\left(t^m+1\right)^{(k-1)-\frac{1}{2}}}\right)\,dt=-\frac{2}{m\left(2k-3\right)}.$$

1342: Also, one sees that 

1343: $$\int_0^{\infty}\left(\frac{t^{\frac{m}{2}-1}}{\left(t^m+1\right)^{\frac{1}{2}}}-\frac{1}{t+1}\right)\,dt=\frac{2\ln 2}{m}.

1344: $$

1345: 

1346: \end{proof}

1347: \subsection*{{\bf Acknowledgments}}

1348: The author thanks Mark Ashbaugh, Fritz Gesztesy, Richard Laugesen, Boris Mityagin, Grigori Rozenblioum and Alexander Turbiner for helpful discussions and references. He also thanks Richard Laugesen for reading a part of this manuscript and suggestions for improving its presentation. 

1349:  

1350: \begin{thebibliography}{10}

1351: 

1352: \bibitem{AA}

1353: A. Avila.

1354: \newblock Convergence of an exact quantization scheme.

1355: \newblock {\em Comm. Math. Phys.}, 249:305--318, 2004.

1356: 

1357: \bibitem{Bender-1}

1358: C. M. Bender, M. Berry, P. N. Meisinger, V. M. Savage and M. Simsek.

1359: \newblock Complex WKB Analysis of Energy-Level Degeneracies of Non-Hermitian Hamiltonians,

1360: \newblock{\em J. Phys. A: Math. Gen.}, 34:L31--L36, 2001.

1361: 

1362: \bibitem{Bender}

1363: C. M. Bender and S. Boettcher.

1364: \newblock Real spectra in non-Hermitian Hamiltonians having $\mathcal{PT}$-symmetry.

1365: \newblock {\em Phys. Rev. Lett.}, 80:5243--5246, 1998. 

1366: 

1367: \bibitem{Bender2}

1368: C. M. Bender and A. Turbiner.

1369: \newblock Analytic continuation of eigenvalue problems.

1370: \newblock {\em Phys. Lett. A}, 173:442--446, 1993.

1371: 

1372: \bibitem{BC}

1373: C. M. Bender and S. A. Orszag.

1374: \newblock {\em Advanced Mathematical Methods for Scientists and Engineers}.

1375: \newblock McGraw-Hill, New York, 1978.

1376: 

1377: \bibitem{Bender3}

1378: C. M. Bender and T. T. Wu.

1379: \newblock Anharmonic oscillator.

1380: \newblock {\em Phys. Rev.}, 184:1231--1260, 1969.

1381: 

1382: 

1383: \bibitem{CGM}

1384: E. Caliceti, S. Graffi and M. Maioli.

1385: \newblock Perturbation theory of odd anharmonic oscillators.

1386: \newblock {\em Comm. Math. Phys.}, 75:51--66, 1980.

1387: 

1388: 

1389: \bibitem{Pham}

1390: E. Delabaere and F. Pham.

1391: \newblock Eigenvalues of complex Hamiltonians with $\mathcal{PT}$-symmetry I, II.

1392: \newblock {\em Phys. Lett. A}, 250:25--32, 1998.

1393: 

1394: \bibitem{Delabaere}

1395: E. Delabaere and D. T. Trinh.

1396: \newblock Spectral analysis of the complex cubic oscillator.

1397: \newblock {\em J. Phys. A: Math. Gen.}, 33:8771--8796, 2000. 

1398: 

1399: 

1400: \bibitem{Dorey}

1401: P. Dorey, C. Dunning and R. Tateo.

1402: \newblock Spectral equivalences, Bethe ansatz equations, and reality properties in $\mathcal{PT}$-symmetric quantum mechanics.

1403: \newblock {\em J. Phys. A: Math. Gen}, 34:5679--5704, 2001.

1404: 

1405: 

1406: \bibitem{Dorey2}

1407: P. Dorey and R. Tateo.

1408: \newblock On the relation between Stokes multipliers and $T$-$Q$ systems of conformal field theory.

1409: \newblock {\em Nucl. Phys. B}, 563:573--602, 1999.

1410: 

1411: \bibitem{Fedoryuk}

1412: M. V. Fedoryuk.

1413: \newblock {\em Asymptotic Analysis}.

1414: \newblock Springer-Verlag, New York, 1993.

1415: 

1416: 

1417: \bibitem{Handy2}

1418: C. R. Handy. 

1419: \newblock Generating converging bounds to the (complex) discrete states of the $P^2 + iX^3 + i\alpha X$ Hamiltonian.

1420: \newblock {\em J. Phys. A: Math. Gen.}, 34:5065--5081, 2001.

1421: 

1422: \bibitem{Handy1}

1423: C. R. Handy, D. Khan, Xiao-Qian Wang and C. J. Tymczak.

1424: \newblock Multiscale reference function analysis of the  $\mathcal{PT}$ symmetry breaking solutions for the $P^2 + iX^3 + i\alpha X$ Hamiltonian.

1425: \newblock {\em J. Phys. A: Math. Gen.}, 34:5593--5602, 2001.

1426: 

1427: \bibitem{HR}

1428: B. Helffer and D. Robert.

1429: \newblock Asymptotique des niveaux d'\'energie pour des hamiltoniens a un degre de libert\'e.

1430: \newblock {\em Duke Math. J.}, 49:853--868, 1982.

1431: 

1432: 

1433: \bibitem{Hille}

1434: E. Hille.

1435: \newblock {\em Lectures on Ordinary Differential Equations}.

1436: \newblock Addison-Wesley, Reading, Massachusetts, 1969.

1437: 

1438: \bibitem{HMM}

1439: F. T. Hioe, D. MacMillen and E. W. Montroll.

1440: \newblock Quantum theory of anharmonic oscillators: energy levels of a single and a pair of coupled oscillators with quartic coupling.

1441: \newblock {\em Phys. Reports}, 43 (7):305--335, 1978.

1442: 

1443: 

1444: \bibitem{KS}

1445: S. M. Klishevich and M. S. Plyushchay.

1446: \newblock Nonlinear holomorphic supersymmetry, Dolan-Grady relations and Onsager algebra.

1447: \newblock {\em Nucl. Phys. B}, 628: 217--233,2002.

1448: 

1449: 

1450: 

1451: \bibitem{MAS}

1452: V. P. Maslov.

1453: \newblock {\em Th\'orie des perturbations et m\'ethodes asymptotiques}.

1454: \newblock Dunod, Paris, 1972.

1455: 

1456: 

1457: \bibitem{Ali1}

1458: A. Mostafazadeh.

1459: \newblock Pseudo-Hermiticity versus PT Symmetry: The necessary condition for the reality of the spectrum of a non-Hermitian Hamiltonian.

1460: \newblock {\em J. Math. Phys.}, 43:205--214, 2002.

1461: 

1462: 

1463: \bibitem{Shin1}

1464: K. C. Shin.

1465: \newblock On the eigenproblems of $\mathcal{PT}$-symmetric oscillators.

1466: \newblock {\em J. Math. Phys.}, 42:2513--2530, 2001.

1467: 

1468: \bibitem{Shin}

1469: K. C. Shin.

1470: \newblock On the reality of the eigenvalues for a class of $\mathcal{PT}$-symmetric oscillators,

1471: \newblock {\em Comm. Math. Phys.}, 229(3):543--564, 2002.

1472: 

1473: \bibitem{Shin2}

1474: K. C. Shin

1475: \newblock Eigenvalues of $\mathcal{PT}$-symmetric oscillators with polynomial potentials,

1476: \newblock {\em Preprint: {\tt math.SP/0407018}}, 23 pages, 2004.

1477: 

1478: \bibitem{Simon}

1479: B. Simon.

1480: \newblock Coupling Constant Analyticity for the Anharmonic Oscillator.

1481: \newblock {\em Ann. Phys.}, 58:76--136, 1970.

1482: 

1483: 

1484: 

1485: 

1486: \bibitem{Sibuya}

1487: Y. Sibuya.

1488: \newblock {\em Global theory of a second order linear ordinary differential equation with a polynomial coefficient}.

1489: \newblock North-Holland Publishing Company, Amsterdam-Oxford, 1975.

1490: 

1491: \bibitem{TIT}

1492: E. C. Titchmarsh.

1493: \newblock {\em Eigenfunction expansions, Part I}.

1494: \newblock Oxford at the Clarendon Press,  1958.

1495: 

1496: \bibitem{Znojil}

1497: M. Znojil.

1498: \newblock Spiked and  $\mathcal{PT}$-symmetrized decadic potentials supporting elementary $N$-plets of bound states.

1499: \newblock {\em J. Phys. A: Math. Gen.}, 33:4911--4916, 2000. 

1500: 

1501: \end{thebibliography}

1502: 

1503: %

1504: {\sc email contact:}  kcshin@math.missouri.edu

1505: %

1506: \end{document}

1507: 

1508: