1: \documentclass[makeidx,reqno]{amsart}

2: %\usepackage[T1]{fontenc}

3: %\usepackage[latin9]{inputenc}

4: %\usepackage{geometry}

5: \usepackage[active]{srcltx}

6: \usepackage{epsfig}

7: \usepackage{hyperref}

8: \usepackage{amsmath}

9: \usepackage{amssymb}

10: %\usepackage{esint}

11: \newtheorem{Proposition}{Proposition}

12:   \newtheorem{Remark}{Remark}

13:   \newtheorem{Corollary}[Proposition]{Corollary}

14:   \newtheorem{Lemma}[Proposition]{Lemma}

15:   \newtheorem{Theo}[Proposition]{Theorem}

16:   \newtheorem{Theorem}{Theorem}

17:  \newtheorem{Consequence}[Proposition]{Consequence}

18:  \newtheorem{Definition}[Proposition]{Definition}

19:  \newtheorem{Note}[Remark]{Note}

20:  \newtheorem{Convention}[Proposition]{Convention}

21: \newtheorem{Assumptions}{Assumption}

22: \newtheorem{Normalization}{Normalization}

23: \newcommand {\z}{{\noindent}}

24:  \def\HH{\mathbb{H}}

25: \def\CC{\mathbb{C}}

26:  \def\RR{\mathbb{R}}

27:  \def\NN{\mathbb{N}}

28: \def\ZZ{\mathbb{Z}}

29: \def\QQ{\mathbb{Q}}

30: \def\Re{\mathrm{Re}}

31: \def\Im{\mathrm{Im}}

32: \def\ds{\displaystyle}

33: \renewcommand{\labelenumi}{$ $\hskip -0.4cm (\roman{enumi})}

34: \newcommand{\theitemi}{(\roman{itemi})}

35: \def\bb{b} \def\({\left(} \def\){\right)} \makeindex

36: \author{O. Costin, M. Huang} \address{Mathematics Department//The Ohio State University//Columbus, OH 43220} \title{Behavior of lacunary

37:   series at the natural boundary} \gdef\shorttitle{Natural boundaries}

38: \begin{document}

39: \begin{abstract}

40:   We develop a local theory of lacunary Dirichlet series of the form

41:   $\sum\limits_{k=1}^{\infty}c_k\exp(-zg(k)), \Re(z)>0$ as $z$

42:   approaches the boundary $i\RR$, under the assumption $g'\to\infty$

43:   and further assumptions on $c_k$. These series occur in many

44:   applications in Fourier analysis, infinite order differential

45:   operators, number theory and holomorphic dynamics among others. For

46:   relatively general series with $c_k=1$, the case we primarily focus

47:   on, we obtain blow up rates in measure along the imaginary line and

48:   asymptotic information at $z=0$.

49: 

50:   When sufficient analyticity information on $g$

51:   exists, we obtain Borel summable expansions  at points on

52:   the boundary, giving exact local

53:   description.  Borel

54:   summability of the expansions provides property-preserving

55:   extensions beyond the barrier.

56: 

57: The singular

58: behavior has remarkable universality and self-similarity features. If

59: $g(k)=k^b$, $c_k=1$, $b=n$ or $b=(n+1)/n$, $n\in\NN$, 

60: behavior near the boundary is roughly of the standard form $\Re(z)^{-b'}Q(x)$ where

61: $Q(x)=1/q$ if $x=p/q\in\QQ$ and zero otherwise.

62: 

63: The B\"otcher map at infinity of polynomial iterations of the form

64: $x_{n+1}=\lambda P(x_n)$, $|\lambda|<\lambda_0(P)$, turns out to have

65: uniformly convergent Fourier expansions in terms of simple lacunary

66: series. For the quadratic

67: map $P(x) =x-x^2$, $\lambda_0=1$, and the Julia set is the graph of this Fourier expansion in the main cardioid of the Mandelbrot set.

68: 

69: 

70: 

71: \end{abstract}

72: \maketitle

73: \tableofcontents

74: \section{Introduction}

75: 

76: Natural boundaries (NBs) occur frequently in many

77: applications of analysis, in the theory of Fourier series, in

78: holomorphic dynamics (see \cite{Douady}, \cite{Devaney}, \cite{CPAM}

79: and references there), in analytic number theory, see

80: \cite{Titchmarsh}, physics, see \cite{Kawai} and even in relatively simple ODEs such as the

81: Chazy equation \cite{Ince}, an equation arising in conformal mapping

82: theory, or the Jacobi equation. 

83: 

84: 

85: The intimate structure of NBs turns out to be particularly rich,

86: bridging analysis, number theory and complex dynamics.  

87: 

88: 

89: 

90: Nonetheless (cf. also \cite{Kawai}), the study of NBs of concrete functions is yet to

91: be completed from a pure analytic point of view. The aim of the present paper is a detailed study near the analyticity

92: boundary of a prototypical  functions exhibiting this 

93: singularity structure, classes of {\em lacunary series}.  For such functions, we

94: develop a theory of generalized local asymptotic expansions at NBs,  and explore their consequences and applications. The expansions are asymptotic in the sense

95: that they become increasingly accurate as the singular curve is approach, and in many cases exact, in that the function can be recovered from these expansions.

96: 

97: 

98: Lacunary series,  sums of the form $h(s)=\sum_{j\ge 1}{c_j}s^{g_j}$, or written

99: as Dirichlet series $f(z)=\sum_{j\ge 1}{c_j}e^{-z{g(j)}}$), where

100: $j/g_j=o(1)$ for large $j$, often occur in applications and have deep

101: connections with infinite order differential operators, as found and studied  by

102: Kawai \cite{Kawai1}, \cite{Kawai}.  Under the lacunarity assumption

103: above, if the unit disk is the maximal disk of analyticity of $h$,

104: then the unit circle is its NB (\cite{Mandelbrojt}). For

105: instance, the series

106: \begin{equation}

107:   \label{eq03}

108:  h(s)=\sum_{j=1}^{\infty}s^{2^j}\:(|s|<1)

109: \end{equation}

110: studied by Jacobi \cite{Jacobi}

111:  before the advent of modern Complex Analysis, clearly has the

112:  unit disk as a singular curve:  $h(s)\to+\infty$ as $|s|\uparrow

113:  1$ along any ray of angle $2^{-n}m\pi $ with $m,n\in\NN$.

114: 

115: 

116: We show that if $c_j=1$ and $g'\to\infty$,

117: then, in a measure theoretic sense,  $f(x+iy)$ blows up

118: as $x\to 0^+$ uniformly in $y$ at a calculable rate. We find

119: interesting {\em universality properties in the blow-up profile}. 

120: 

121: 

122: In special cases of interest, Borel summable power series, in powers

123: of the distance to the boundary,  and more generally convergent 

124: expansions

125: as series of small exponentials multiplying  Borel summed series power series\footnote{These are simple instances of Borel (or Ecalle-Borel) summed transseries. A brief summary of the definitions and properties

126: of transseries and generalized Borel summability, and references to

127: the literature are given in \S\ref{A4}. \label{f1}}

128: representations

129: can be determined on a dense set on the singularity

130: barrier. Examples are

131: \begin{itemize}

132: 

133: \item $\sum_{j\ge 1}e^{-z j^b}$ where $b>1$, or its dual $q>1$, where

134:   $ b^{-1}+q^{-1}=1$, is integer (relating to exponential sums and van

135:   der Corput dualities \cite{Montgomery}; the special self-dual case

136:   $\sum_{j\ge 1}e^{-z j^2}$, is related to the Jacobi theta function);

137: 

138: \item $\sum_{j\ge 1}e^{-z a^j}$, $1<a\in\NN$;

139: \end{itemize}

140: More generally, if $c_j=c(j)$ and $g_j=g(j)$ have suitable analyticity

141: properties in $j$, then the behavior at $z=0^+$, and possibly at other

142: points, is described in terms of Ecalle-Borel summed expansions (see

143: footnote \ref{f1}).  Then the analysis leads to a natural,

144: properties-preserving, continuation formulas across the boundary.

145: 

146: 

147: 

148: In general, the blow-up profile along the barrier is closely related to

149: {\em exponential sums,} expressions of the form

150: \begin{equation}

151:   \label{eq:es1}

152: S_N=\sum_{k=1}^N c_n\exp(2\pi i g(n)),\ \ g(n)\in\RR

153: \end{equation}

154: where for us $g'\to\infty$ as $n\to\infty$. The corresponding lacunary

155: series are in a sense the continuation of (\ref{eq:es1}) in the

156: complex domain, replacing $2\pi i $ by $-z$, $\Re(z)>0$, and letting

157: $N\to \infty$. The asymptotic behavior of lacunary series as the

158: imaginary line is approached in nearly-tangential directions is

159: described by dual, van der Corput-like, expansions. The method we use

160: extend to exponential sums, which will be the subject of a different

161: paper.

162: 

163: \section{Results}

164: \subsection{Results under general assumptions}

165: \subsubsection{Blow-up on a full measure set}

166: We consider lacunary Dirichlet series  of the form

167: \begin{equation}

168:   \label{lacn1}

169:  f(z)= \sum_{k=0}^\infty e^{-zg(k)}

170: \end{equation}

171: but, as it can be seen from the proofs, the analysis extends easily to series of the form

172: $$f(z)= \sum_{j=0}^\infty  c_j e^{-zg(j)}$$

173: under suitable smoothness and growth conditions, see

174: \S\ref{genblowup}. The results in the paper apply under the further restriction,

175: \begin{Assumptions}\label{A1}

176:   The function $g$ is  differentiable and  $g'(j)\to \infty$ as $j\to\infty$.

177: \end{Assumptions}

178: 

179: 

180: In particular, $g$ is eventually increasing. By subtracting a finite

181: sum of terms from $f$ (a finite sum is clearly entire), we arrange that $g$ is

182: {\bf increasing}.  If $g(0)=a$, we can multiply $f$ by $e^{-za}$ to

183: arrange that $g(0)=0$.

184: 

185: \begin{Normalization}

186: 

187: 

188: (i) $g$ is differentiable on $[0,\infty)$, $g'>0$ and $g'\to\infty$

189: along $\RR^+$.

190: 

191: (ii) $g(0)=0$.

192: \end{Normalization}

193: 

194: \bigskip

195: 

196: \z {\bf Notation.} We write  $\ds |H(\cdot)|\mathop{=}^\mu 1+o(1)$ if $|H(y)|dy$ converges to

197: the Lebesgue measure $d\mu(y)$.

198: 

199: \bigskip

200: 

201: Under Assumption \ref{A1}, after normalization, we have the following result,

202: giving exact blow-up rates in measure, as well as sharp pointwise  blow-up upper bounds.

203:  \begin{Theorem}\label{P1}

204: (i)  We have the uniform

205: blow-up rate in measure\footnote{It turns out that in general, $L^1$ or  a.e. convergence of $|f|$ do

206: not hold.}.

207: \begin{equation}

208:   \label{eq:asbh}

209:   |f(x+i\cdot)|^2\mathop=^\mu\int_{0}^{\infty}e^{-2g(s)x}ds\left(1+o(1)\right)

210: \end{equation}

211: It can be checked that  $ \int_{0}^{\infty}e^{-2g(s)x}ds\ \ge g^{-1}(1/x)\to\infty {\text{ as }} x\to 0^+$;  see also Note~\ref{n1}.

212: 

213: 

214: 

215: (ii) The following pointwise estimate holds:

216: \begin{equation}

217:   \label{eq:locb}

218:   \|f(x+i\cdot)\|_{\infty}\le \int_{0}^{\infty}e^{-g(s)x}ds(1+o(1))\ \ \text{as } x\to 0^+

219: \end{equation}

220: This is sharp at $z=0$, cf. Proposition \ref{Prop2},  and

221: in many cases it is only reached at $z=0$; see Proposition~\ref{Pp4}.

222: \end{Theorem}

223: 

224: 

225: \subsubsection{General behavior near  $z=0$}

226: At $z=0^+$ a more detailed asymptotic description is possible.

227: \begin{Theorem}\label{Prop2}

228:   (i) As $z\to 0^+$  we have

229:   \begin{equation}

230:     \label{eq:asympt1}

231:     f(z)=\int_0^{\infty}e^{-z g(s)}ds-\frac{1}{2}+o(1)\ \to\infty\ \ \ \text{as } z\to 0^+

232:   \end{equation}

233: In fact,

234: \begin{equation}

235:     \label{eq:asympt1}

236:    f(z)-\int_0^{\infty}e^{-z g(s)}ds=-z\int_{0}^{\infty} e^{-zu}

237:    \{g^{-1}(u)\}du

238:   \end{equation}

239: (where $\{\cdot\}$ denotes the fractional part, and we used $g(0)=0$)\footnote{As mentioned, often this maximal growth

240:     is achieved at zero but in special cases it occurs, up to a

241:     bounded function, on a dense set of measure zero.}.

242: 

243: 

244: (ii)  If $g(s)$ has a

245:   differentiable asymptotic expansion  as $s\to\infty$ in terms of (integer or noninteger)

246: powers of $s$ and $\log s$, and $g(s)\sim const. s^b $, $b>1$, then after

247: subtracting the blowing up term, $f$

248: has a Taylor series at $z=0$ (generally divergent, even when $g$ is analytic,

249: which can be calculated explicitly),

250: $$

251: f(z)=

252: \int_0^{\infty}e^{-z g(s)}ds-\frac{1}{2}+zs(z), \ \ s\in C^\infty[0,\infty)$$

253: (as an example, see (\ref{eq:srjtob}).

254: \end{Theorem}

255: 

256: \begin{Note}\label{n1}

257:  Often  $g$ has

258: an asymptotic expansion starting with a combination of powers, exponentials and

259: logs.   Let $\phi=g^{-1}$. Then $\phi(\nu x)/\phi(\nu)\to \phi_1(x)$ as $\nu\to\infty$ and $x>0$ is

260: fixed and

261: $$\int_0^{\infty}e^{-x g(s)}ds=C_gg^{-1}(1/x)(1+o(1));\ \ \ C_g=\int_0^{\infty}e^{-u}\phi_1(u)du$$

262: \end{Note}

263:  For instance, if  $g(k)=k^b,b>1$ we have, as $\rho\to 1$,

264: \begin{eqnarray}

265:   \label{nb}

266:  |f(z)|\mathop=^\mu x^{-\frac{1}{2b}} \,\,\Gamma(1+1/b)^{-1/2}2^{-\frac{1}{2b}}(1+o(1)) \ \ \ (x\to 0^+)\nonumber \\

267:  f(x)=x^{-\frac{1}{b}} \,\, \Gamma(1+1/b)(1+o(1))\ \ \ \ \ \ \ (x\to 0^+)

268: \end{eqnarray}

269: 

270: \subsubsection{Blow-up profile along barrier}

271: 

272: \begin{Theorem}\label{Cexp}

273:   (i) Assume that for some $y\in\RR$ there is a smooth increasing

274:   function $\rho(N;y)=:\rho(N)\in(0,N]$ such that the following

275:   weighted exponential sum (see \cite{Montgomery}) has a limit:

276:   \begin{equation}

277:     \label{eq:expos}

278:    S_{\rho,N}:= \rho(N)^{-1}\sum_{j=1}^N e^{i y g(j)}\to L(y)\ \ \text{as}\ \ N\to \infty

279:   \end{equation}

280:    where $\rho''$ is uniformly bounded and nonpositive for sufficiently

281:   large $k$. (Without loss of generality, we may assume $\rho''(k)\le

282:   0$ for all $k$.)  Let

283:   \begin{equation}

284:     \label{eq:H1}

285:     \Phi(x)=\int_0^{\infty}e^{-xg(u)}\rho'(u)du

286:   \end{equation}

287: Then, we have the asymptotic behavior

288:   \begin{equation}

289:     \label{eq:lim2}

290:     f(x+iy)=L(y)\Phi(x)+o(\Phi(x))\ \ \ (x\to 0^+)

291:   \end{equation}

292: (ii) As a pointwise upper bound we have:

293: \begin{equation}

294:   \label{eq:lsup}

295:   \limsup_{N\ge 0} |S_{\rho,N}|=L<\infty\ \Rightarrow  \ f(x+iy)=O(\Phi(x))\ \ \ (x\to 0^+)

296: \end{equation}

297: \end{Theorem}\label{P4}

298: 

299: \subsection{Results in specific cases}

300: 

301: 

302: 

303: 

304: \subsubsection{Settings leading to convergent expansions}

305: The cases $g(j)=j^2$ and  $g(j)=a^j$, $a>1$ are distinguished, since

306: the expansions at some points near the boundary converge.

307: 

308: \begin{Proposition}\label{b=2}

309: \z If $b=2$, then we have the identity \[

310: f(z)=\frac{1}{2}\sqrt{\frac{\pi}{z}}-\dfrac{1}{2}+\sqrt{\frac{\pi}{z}}\sum_{k=1}^{\infty}e^{-\frac{k^{2}\pi^{2}}{z}}\]

311: 

312: \end{Proposition}

313: Clearly, this is most useful when $z\to 0$. It also shows the identity

314: associated to the Jacobi theta function

315: \begin{equation}

316:   \label{eq:ide2}

317:   f(z)=\frac{1}{2}\sqrt{\frac{\pi}{z}}-\dfrac{1}{2}

318: +\sqrt{\frac{\pi}{z}}f\left(\frac{k^2\pi^2}{z}\right)

319: \end{equation}

320: \begin{Proposition}\label{Prop4}

321:   If $g(j)=a^j$, $a>1$,  then, as $z\to 0^+$, $f(z)$

322:   is convergently given by

323:  \begin{multline}\label{convt}

324:   f(z)=-\dfrac{\log \zeta}{\log a}+\sum_{n=1}^{\infty}\frac{(-\zeta )^{n}}{n!(1-a^{n})}+c_{0}+\frac{1}{\log a}\sum_{k\neq0}\Gamma\left(-\frac{2k\pi i}{\log a}\right)\zeta ^{\frac{2k\pi i}{\log a}}\\=-\dfrac{\log \zeta }{\log

325: a}+\sum_{n=1}^{\infty}\frac{(-\zeta )^{n}}{n!(1-a^{n})}

326: +c_{0}-

327: \frac{1}{2\pi i}\int_{0}^{\infty}\log_{\RR}[-(s/\zeta )^{\frac{2\pi i}{\log a}}]e^{-s}ds

328:  \end{multline}

329: where $\zeta=1-e^{-z}$.

330: (Here $\log_{\RR}$ is the usual branch of the log with a cut along

331:  $\RR^-$ and {\bf not} the log on the universal covering of

332:  $\CC\setminus \{0\}$.)

333: 

334: It is clear that for $a\in\NN$ the transseries

335: can be easily calculated for any $z=\rho\exp(2\pi i m/a^j)$, $(m,j)\in\NN, 0\le \rho<1$ since

336: $$f(\rho \exp(2\pi i m/a^j))=\sum_{n=1}^{j} \rho^{a^j}\exp(2\pi i m a^{n-j})+f(\rho)$$

337: where the sum is a polynomial, thus analytic.

338: \end{Proposition}

339: 

340: 

341: 

342: \subsubsection{Borel summable transseries representations; resurgence (cf. \S\ref{A4})}

343: 

344: 

345: 

346: 

347: When $c_j\equiv 1$ it is clear that the growth rate as $z\to 0^+$

348: majorizes the rate at any point on $i\RR$. There may be {\bf no} other

349: point with this growth, as is the case when $g(j)=j^b$, $b\in (1,2)$

350: as seen in

351: see Proposition \ref{Pp4} below, or densely many if, for instance,

352: $g(j)=j^b$, $b\in\NN$, or when $g(j)=a^j$, $a\in\NN$. The behavior

353: near points of maximal growth merits special attention.

354: 

355: 

356: 

357: For $g=j^b$, $1<b\ne 2$, $f$ has asymptotic expansions which do not,

358: in general converge.  They are however generalized Borel summable.

359: 

360: 

361: Define $d$ by \begin{equation}

362:   \label{eq:eqdu}

363:   \frac{1}{b}+\frac{1}{d}=1

364: \end{equation}

365: 

366: 

367: 

368:  \begin{Theorem}\label{PP3}{ Let $g=j^b;\,b>1$.  Then,

369: 

370: 

371: (i)   The asymptotic series

372: of $f(z)$ for small $z$,

373: \begin{equation}

374:   \label{eq:tr1}

375:   \tilde{f}_0=\Gamma\left(1+\dfrac{1}{{b}}\right)z^{-\frac{1}{{b}}}-\dfrac{1}{2}+\frac{i}{2\pi}\sum_{j=1}^{\infty}(1-(-1)^{j{b}})\zeta(j{b}+1)b_{j}z^{j}

376: \end{equation}

377: is Borel summable  in $X=z^{-1/(b-1)}$, 

378:    along any ray $\arg(X)=c$ if $c\ne-\arg k^qs_{\pm}, k\in\NN $,   where

379: $s_{\pm}=t_{\pm}\mp 2\pi i t_{\pm}^{1/b}$ and   $t_{\pm}=( \pm  2\pi i/b)^{b/(b-1)}$. More precisely, (a)

380:  \begin{equation}

381:     \label{eq:eqfz}

382:    f(z)=\Gamma(1+1/b)z^{-1/b}-\frac{1}{2}+ z^{-b/(b-1)}\int_0^{\infty}e^{-z^{-1/(b-1)s}}H(s)ds

383:   \end{equation}

384:   where $H(s)=H_{b}(s^{b-1})$, where $H_{b}$ is analytic at zero and $H_{b}(0)=0$; (b) $H$ is

385: analytic on the Riemann surface of the log, with square root branch points

386: at all points of the form $k^qs_{\pm}, k\in\NN$ and  (c) making appropriate cuts (or working on Riemann surfaces), $u^{-(b-1)^2/b}H$ is bounded at infinity.

387: 

388: If $\arg(X)\in(\theta_-,\theta_+)$, then $\mathcal{LB}\tilde{f}_0=f$ \footnote{Note that 

389: the variable of Borel summation, or {\em critical time}, is not

390: $1/z$ but $z^{-1/(b-1)}$.}  In a general complex direction, $f$ has a nontrivial transseries, see (iii).

391: 

392: 

393: (iii) For a given direction $\varphi$,

394: $\sigma$ be $\pm 1$ if $\pm \varphi>0$ and $0$ otherwise. If $\sigma \arg z\in (\theta_{\sigma }, \pi/2)$, then the transseries

395: of $f$ is

396: 

397: \begin{equation}

398:  \tilde{f_0}(z)+\sigma \frac{i}{2\pi}\sum\limits _{k=1}^{\infty}

399: e^{-{s_{{\sigma}}k^{{q}}}{x^{-q+1}}}\sum\limits _{j=0}^{\infty}c_{j}(\sigma k)^{\frac{-2j{b}+2-{b}}{2({b}-1)}}z^{\frac{2j-1}{2({b}-1)}}

400: \end{equation}

401: and it is Borel summable as well.}

402:  \end{Theorem}

403: 

404: 

405: 

406: 

407: 

408:  \begin{Note}

409:    {\rm The duality $k^b\leftrightarrow k^d$ is the same as in van der Corput

410: formulas; see \cite{Montgomery}.}

411:  \end{Note}

412: \subsubsection{Examples} (i)

413:  For $b=3$, $f$ the transseries is given by

414: \begin{equation}

415: \tilde{f}_0(z)+\sum_{k\in\ZZ}\sigma e^{-\kappa_3k^{3/2}z^{-1/2}}\left[\left(\dfrac{\pi

416: i k}{6}\right)^{\frac{1}{4}}{z}^{-\frac{1}{4}}+\dfrac{5}{32}\dfrac{(ik)^{\frac{7}{4}}}{6^{\frac{3}{4}}\pi^{\frac{5}{4}}}z^{\frac{1}{4}}+...\right]

417: \end{equation}

418: where  with $\kappa_{3}=\pi^{3/2}\sqrt{\frac{32}{27}}(-1)^{\frac{1}{4}}$ and

419: $$ \tilde{f}_0(z)=\frac{\Gamma(4/3)}{z^{1/3}}-\dfrac{1}{2}-\dfrac{z}{120}+\frac{z^{3}}{792}+...$$

420: 

421: 

422: \z (ii) For $b=3/2$, with $\kappa=32 i\pi^3/27$, $f$ has the Borel summable transseries

423: \begin{equation}

424:   \label{eq:3/2f}

425:  \tilde{f}_0(z)+ \sigma \sum\limits _{k=1}^{\infty}e^{\kappa \sigma k^{3}z^{-2}}

426: \left[\dfrac{4\sqrt{2}(\sigma i)^{-\frac{1}{2}}\pi}{3}\dfrac{k^{\frac{1}{2}}}{z}+

427: \dfrac{(\sigma i)^{\frac{1}{2}}}{16\sqrt{2}\pi^{\frac{5}{4}}}\dfrac{z^2}{k^{\frac{5}{2}}}+...

428: \right]

429: \end{equation}

430: where $\theta_{\pm}=\pi/4$  and

431: $$\tilde{f}_0(z)=\Gamma\left(\text{\small $\frac{5}{3}$}\right)z^{-\frac{2}{3}}-\dfrac{1}{2}-\dfrac{3\zeta(\frac{5}{2})}{16\pi^{2}}z+\frac{1}{240}z^{2}+\frac{315\zeta(\frac{11}{2})}{2048\pi^{5}}z^{3}+...$$

432: 

433: 

434: 

435:  \subsubsection{Properties-preserving extensions beyond the barrier}

436: It is natural to require for an extension beyond the barrier that it

437: has the following properties:

438: \begin{enumerate}

439: \item It reduces to usual analytic continuation when the latter exists.

440: 

441: \item It commutes with all properties  with which analytic continuation

442: is compatible (principle of preservation of relations, a  vaguely

443: stated concept; this requirement is rather open-ended).

444: 

445: 

446: \end{enumerate}

447:  Borel summable series (more generally transseries)

448:      or suitable convergent representation representations allow for

449:      extension beyond the barrier, as follows. In the case $g(j)=j^b$

450:      (and, in fact in others in which $g$ has a convergent or summable

451:      expansion at infinity), $f(z)$ can be written, after Borel

452:      summation in the form (see

453:      \S\ref{AC1}) \begin{equation} \label{eq:eqac} z^{-{d}}\int_0^{

454:          \infty}e^{-z^{-1/(b-1)}

455:          s}H_1(s)ds=z^{\frac{1}{b-1}-d}\int_0^{ \infty}e^{-t}H_1(t

456:        z^{1/(b-1)})dt=:\int_0^{\infty}e^{-s}F(s,z)ds \end{equation} where $H_1$ is analytic near the

457:      origin and in $\CC$ except for arrays of isolated singularities

458:      along finitely many rays. Furthermore, $H_1$ is polynomially

459:      bounded at infinity. This means that the formal series is

460:      summable in all but finitely many directions in

461:      $z$.

462:      \begin{Definition}

463:        We define the Borel sum continuation of $f$ through point $z$ on the

464:        natural boundary, in the direction $d$, to be

465:        the Borel sum of the formal series of $f$ at $z$ in the direction $d$,

466:        if the Borel sum exists.

467: 

468: This extension simply amounts to analytically

469:        continuing $F(s,z)$ in $z$, in (\ref{eq:eqac}), for

470:        small $s$ and then analytically continuing

471:        $F(s,z)$ in $s$ for fixed $z$.

472:        along $\RR^+$.

473:      \end{Definition}

474: 

475:  \subsubsection{Notes} (cf. Appendix \S\ref{Fi}.) \begin{enumerate} \item The Borel sum provides a natural, properties-preserving,

476:        extension \cite{OCtransseries}. Borel summation commutes with all common operations

477:        such as addition, multiplication, differentiation.  Thus,

478:        the function and its extensions will  have the same

479:        properties.

480: 

481: 

482: \item

483: Also, when a series converges, the Borel sum coincides with the usual

484: sum.  Thus, when analytic continuation exists, it coincides with the

485: extension.

486: 

487: \item

488: 

489:  With or without a boundary, the Borel sum of a

490: divergent series changes as the direction of summation crosses the

491: {\em Stokes directions} in $\CC$.  Yet, the properties of the family of

492: functions thus obtained are preserved.

493: There may then exist extensions {\em along} the barrier as well. Of course,

494: all this cannot mean that there is analytic continuation across/along

495: the boundary.

496: \end{enumerate}

497: 

498: 

499: 

500: See also Eq. (\ref{convt}), a convergent expansion, where $z$ can also

501: be replaced by $-z$.  In this case, due to strong lacunarity, the extension

502: {\em changes} along {\em every} direction, as though there existed densely

503: many Stokes lines.

504: 

505: \subsection{Universal behavior near boundary in specific cases}

506: \begin{figure}

507:   \centering

508:   \includegraphics[scale=0.3]{Riemann1}

509:   \caption{The standard function $Q(x)$. The points above $Q=0.05$ are

510: the only ones present in the actual graph.}

511:   \label{fig:1}

512: 

513:   \includegraphics[scale=0.3]{barrn40}

514:   \caption{Point-plot graph of $Q_{4,4}$, normalized to

515:     one.}

516:   \label{fig:1}

517:  \end{figure}

518: \begin{figure}

519:   \includegraphics[scale=0.4]{fig030}

520:   \caption{Point-plot graph of $Q_{3,3}$; $Q_{3,2}$ follows from it

521: through the transformation (\ref{eq:dual}).

522: }

523:   \label{fig:1}

524: 

525: 

526: \end{figure}

527: 

528: 

529: In many cases, $\phi(g^{-1}(u/x))$ has an asymptotic expansion, and then

530: $\Phi$ in turn has an asymptotic expansion in $x$.

531: Detailed behavior along the boundary can be obtained in special cases such as  $g(j)=j^b;\,b\in\NN$,

532: or $g(j)=j^{(b+1)/b};\,b\in\NN$. Properly scaled sums converge to everywhere discontinuous

533: functions.

534: \begin{Theorem}\label{Pp4}

535:   (i) If $g(j)=j^b$, $b\in\NN$, we have $\displaystyle \limsup

536:   x^{1/b} |f(x +iy)|<M<\infty$. We let

537:   $Q_{b,d}(s)=\displaystyle \lim_{x\to

538:     0^+}x^{1/d}f(x+2\pi i s)$ (whenever it exists); then,

539: 

540: \begin{multline}

541:   \label{eq:bd2}

542: \frac{Q_{b,b}(s)}{\Gamma(1+b^{-1})}=\begin{cases} \displaystyle  \,\,\frac{1}{n}\sum_{l=1}^{n}e^{-2\pi

543: i\dfrac{m}{n}l^{b}}, \ s=\frac{m}{n}, \ m,n\in \NN\ \text{relatively prime} \\ \\

544: 0\ \ \ a.e. \end{cases}

545: \end{multline}

546: 

547: (ii) For $b=3/2$ and $z\to 0^+$, $f$ grows like $z^{-2/3}$. For any other point

548: on $i\RR$ the growth is slower, at most $(\Re\, z)^{-1/2}$. Furthermore,

549: 

550: 

551: \begin{equation}\label{case3/2}

552:   \frac{Q_{3/2,1/2}(\sqrt{s})}{\sqrt{6\pi i}s^{1/4}}=\begin{cases}\displaystyle\frac{1}{n}\sum_{l=1}^{n}e^{-2\pi

553: i\dfrac{m}{n}l^{3}},\ \ s =\dfrac{16}{27}\frac{n}{m}, \ n,m\text{ as in (i)}\\ \\

554:  0\ \ a.e.\end{cases}

555: \end{equation}

556: In particular, we have the profile duality relation

557: \begin{equation}

558:   \label{eq:dual}

559:   Q_{3/2,1/2}(s)=\frac{\sqrt{6\pi i}s^{1/2}}{\Gamma(4/3)}Q_{3,3}\(2^83^{-6}s^{-2}\)

560: \end{equation}

561: for any $s\in\RR$ for which $Q_{3/2}(s)$ and/or

562: $Q_{3,3}(s)$ is well defined, for instance, in the cases given in

563: (\ref{eq:bd2}) and (\ref{case3/2}).

564: \end{Theorem}

565: For large $n$, the sum over $l$ in Eq. (\ref{eq:bd2}) is,

566: statistically, expected to be of order one.  After $x^{\frac{1}{d}}$

567: rescaling, the template behavior ``in the bulk'' is given by the

568: familiar function $Q(x)=n^{-1}$ if $x=m/n, (m,n)\in\NN^2$ and zero

569: otherwise, shown in Fig.  \ref{fig:1}.

570: \subsection{Fourier series of the B\"otcher map and structure of Julia sets}\label{holom}

571: 

572: We show how lacunary series are building blocks for fractal structures

573: appearing in holomorphic dynamics (of the vast literature on  holomorphic

574: dynamics we refer here in particular to \cite{Beardon}, \cite{Devaney} and \cite{Milnor}).

575: In \S\ref{S5} we mention a few known facts about polynomial

576: maps. Consider for simplicity the

577: quadratic map

578: \begin{equation}

579:   \label{eq:logist1}

580:   x_{n+1}=\lambda x_n(1-x_n)

581: \end{equation}

582: It will be apparent from the proof that the results and method extend easily

583: to  polynomial iterations of the form

584: \begin{equation}

585:   \label{eq:eqpol1}

586:   x_{n+1}=\lambda P_k(x_n)

587: \end{equation}

588: with $\lambda$ relatively small.

589: The substitution $x=-(\lambda y)^{-1}$ transforms (\ref{eq:logist1}) into

590: \begin{equation}

591:   \label{eq:infty1}

592:   y_{n+1}=\frac{y_n^2}{\lambda(1+ y_n)}=f(y_n)

593: \end{equation}

594: 

595: By B\"otcher's theorem (we give a self contained proof in

596: \S\ref{S5}, for (\ref{eq:logist1}), which extends in fact to the

597: general case), there exists a map $\phi$,\footnote{The notations

598: $\phi,\psi, H$, designate different objects than those in previous

599: section.} analytic near zero, with $\phi(0)=0$,

600: $\phi'(0)=\lambda^{-1}$ so that $(\phi f \phi^{-1})(z)=z^2$. Its

601: inverse, $\psi$, conjugates (\ref{eq:infty1}) to the canonical map

602: $z_{n+1}=z_n^2$, and it can be checked that

603: \begin{equation}

604:   \label{eq:eqG}

605:   \psi(z)^2=\lambda \psi(z^2)(1+\psi(z));\ \ \psi(0)=0,\ \psi'(0)=\lambda

606: \end{equation}

607: Let $\mathcal A(\mathbb{D})$ denote the Banach space of functions

608: analytic in the unit disk $\mathbb{D}$ and continuous in  $\overline{\mathbb{D}}$, with the sup norm. We define  the linear operator $\mathfrak T=\mathfrak T_2$, on $\mathcal A(\mathbb{D})$ by

609: \begin{equation}

610:   \label{eq:defT}

611:   (\mathfrak T f)(z)=\frac{1}{2}\sum_{k=0}^{\infty}2^{-k}f(z^{2^k})

612: \end{equation}

613: This is the inverse of the operator $f\mapsto 2f-f^{\vee 2}$,

614: where $f^{\vee p}(z)=f(z^p)$.

615: Clearly, $\mathfrak T f$ is an isometry on   $\mathcal A(\mathbb{D})$

616: and it maps simple functions, such as generic polynomials, to

617: functions having $\partial \mathbb{D}$ as a natural boundary; it reproduces $f$

618: across vanishingly small scales.

619: 

620: 

621: \bigskip

622: 

623: \begin{Theorem}\label{T1}

624: 

625: (i) $\psi$ and $H=1/\psi$ are analytic in ($\lambda,z$) in $\mathbb{D}\times\mathbb{D}$ ($\lambda\in \mathbb{D}$ corresponds to the main cardioid

626: in the Mandelbrot set, see \S\ref{S5}), and continuous in $\mathbb{D}\times\overline{\mathbb{D}}$. The series

627: \begin{equation}

628:   \label{eq:formG}

629: \psi(\lambda,z)=\lambda z+\sum_{k=1}^{\infty}z\lambda^{k+1} \psi_k(z)

630: \end{equation}

631: converges in $\mathbb{D}\times\overline{\mathbb{D}}$ (and so does the series

632: of $H$), but not in  $\overline{\mathbb{D}}\times\overline{\mathbb{D}}$. Here

633: \begin{equation}

634:   \label{eq:formPhi1}

635:  \psi_1(z)=\mathfrak T z=\sum_{k=1}^{\infty}2^{-k+1}z^{2^k}

636: \end{equation}

637: (note that $\psi(z)=\frac{1}{2}z+\frac{1}{2}\int_0^zs^{-1}h(s)ds$ with $h$ given in (\ref{eq03})),  and in general

638: \begin{equation}

639:   \label{eq:frakt}

640:  \psi_k =\mathfrak{T}\left(z\sum_{j=0}^{k-1}\psi_j^{\vee 2}\psi_{k-1-j}-\sum_{j=1}^{k-1}\psi_j\psi_{k-j}\right)

641: \end{equation}

642: 

643: (ii) All $\psi_k, k\ge 1$ have binarily lacunary series: In  $\psi_k$,

644: the coefficient of $z^p$ is nonzero only if $p$ has at most $k$ binary-digits equal to $1$, i.e.,

645: \begin{equation}

646:   \label{eq:pk}

647:   p=2^{j_1}+\cdots+2^{j_k},\ \ j_i=0,1,2,... \ (\text{``$k$''is the same as in $\psi_k$})

648: \end{equation}

649: (iii) For $|\lambda|<1$, the Julia curve of (\ref{eq:logist1}) is given

650: by a uniformly convergent Fourier series, by (i),

651: \begin{equation}

652:   \label{eq:eqJ}

653:   J=\left\{-\left(\Re\, H(e^{it}),\Im\, H(e^{it})\right): t\in [0,2\pi)\right\}

654: \end{equation}

655: \end{Theorem}

656: \begin{Remark}{\rm

657: The effective lacunarity of the Fourier series makes calculations of the

658: Julia set numerically effective if $\lambda$ is not too large.}

659: \end{Remark}

660: 

661: \begin{figure}

662:   \centering

663:  \includegraphics[scale=0.3]{lambda3}\includegraphics[scale=0.3]{lambda3i}

664:   \caption{The Julia set of $x_{n+1}=\lambda x_n(1-x_n)$, for $\lambda=0.3$ and

665:  $\lambda=0.3i$,

666: calculated from the Fourier series  (\ref{eq:formG}) discarding all $o(\lambda^2)$  terms. They coincide, within

667: plot precision, with numerically calculated ones using standard iteration of maps

668: algorithms.

669: }

670:   \label{fig:3}

671: 

672: \end{figure}

673: 

674: \begin{Note} \label{N4} $ $ \hskip -1cm {\rm

675:     \begin{enumerate}

676:     \item Lacunarity of $\psi_k$ is a strong

677:     indication that $\psi$ has a natural boundary (in fact, it does have one),

678:     but not a proof.  Transseriation at {\em singular} points and

679:     summation of convergent series (here in the parameter $\lambda$)

680:     {\bf do not commute}.  The transseries of $H$ on the barrier can

681:     be calculated by transasymptotic matching, see \cite{Inventiones},

682:     but in this case is more simply found directly from the functional

683:     relation; see Note \ref{Explc}.

684: \item In assessing the fine structure

685:  of the fractal using the Fourier expansion truncated to $o(\lambda^n)$, the scale of analysis

686: cannot evidently  go below $O(\lambda^n)$.

687:     \end{enumerate}

688: }

689: \end{Note}

690: 

691: 

692: \begin{Remark}{\rm

693: 

694:     \begin{enumerate}

695: 

696: 

697:  \item For $|\lambda|$ sufficiently small,

698: Theorem~\ref{T1} provides a convenient way to determine

699: the Julia set as well as the discrete evolution on the boundary.

700: 

701: 

702: 

703:  \item  For small $\lambda$, the self similar structure

704: is seen in

705: $$\psi_1(\rho \exp(2\pi i m/2^n))=

706: \sum_{k=1}^{n-1}\frac{ \rho^{2^j}

707: \exp(2\pi i m 2^{k-n})}{2^j}+\frac{ \rho^{2^n-2}}{2^{n-1}}\psi_1(\rho)$$

708: where the sum is a polynomial, thus analytic. Up to a scale factor

709: of $2^{-n+1}$,  if $|\lambda|\ll 2^{-n+1}$,  the nontrivial structure of $\psi$

710: at $\exp(2\pi i m/2^n)$ and at $1$ are the same, see Note \ref{N4}; that is

711: $$\psi(\exp(2\pi i m/2^n))={2^{-n+1}}\psi(1)+\text{regular}+o(\lambda)$$

712: Exact transseries can be obtained for $\psi$; see  also Note~\ref {Explc}.

713: 

714: \item For iterations  of the form

715: $x_{n+1}=\lambda^{k-1}P_{k}(x)$ where $P_k$ is a polynomial of degree $k>2$,  the calculations and the results, for small

716: $\lambda$, are essentially the same. The lacunary series

717: would involve the powers $z^{k^j}$. For instance if the

718: recurrence is $x_{n+1}=\lambda^2x_n(1-3x_n+x_n^2)$, then $\psi$

719: is to be replaced by the solution of

720: $$\psi=\lambda^2\psi^{\vee 3}(1+3\psi+\psi^2)$$

721: and the small $\lambda$ series will now have

722: $\psi_1=\sum_{k=1}^{\infty}3^{-k-1}{{z^{3^k}}}$ and so on.

723:     \end{enumerate}

724:   }

725: \end{Remark}

726: 

727: 

728: 

729: \section{Proofs}

730: \begin{Lemma}\label{linv}

731:  Under the assumptions of Theorem~\ref{P1}, (iii), $h(y):=g^{-1}(y)$ also has a

732:   differentiable asymptotic power series as $y\to\infty$.

733:   \end{Lemma}

734: \begin{proof}

735:     Straightforward inversion of power series asymptotics, cf \S\ref{appendix}.

736:   \end{proof}

737: 

738: \subsection{Proof of Theorem \ref{P1}}

739: The proof of (\ref{eq:asbh}) (i)  essentially amounts to showing that

740: $|f|^2$ is diagonally dominant, in that terms containing $g(j)$ and $g(k)$

741: with $j\ne k$ are comparatively small, as shown in \S\ref{genblowup}.

742: 

743: (ii) Since $g$ is increasing on

744: $(0,\infty)$, the result follows from  the usual

745: integral upper and lower bounds for a sum. Equation (\ref{eq:asympt1})

746: follows from simple calculations, cf.  \S\ref{appendix}.

747: 

748: \subsection{Proof of  Theorem \ref{Prop2}}

749: We prove part (ii); part (i) is similar, and simpler. By standard Fourier analysis we get

750: \begin{equation}

751:   \label{eq:eq21}

752: \{u\}=\frac{1}{2}-\sum_{k=1}^{\infty}\frac{\sin2k\pi u}{k\pi}, \ \ \ \ u\notin\mathbb{Z}\end{equation}

753: where

754: \begin{equation}

755:   \label{eq:eqM}

756:   \left\|\{u\}-\frac{1}{2}+\sum_{k=1}^{M}\frac{\sin2k\pi u}{k\pi}\right\|_{\infty}\le C_M \to {\text{\rm  Si}}(\pi) \text{ as } M\to \infty

757: \end{equation}

758: where Si is the sine integral, and Si($\pi$) is the constant in the Gibbs phenomenon.

759:       Let $g_m$ be an analytic function such that

760:     $g(s)-g_m(s)=o(s^{-m})$ for large $s$.

761:     \begin{Lemma}

762:       The analysis reduces to the case where $h$ is a finite sum of powers. Indeed,

763: \begin{equation}

764:   \label{eq:eqf}

765:   f=e^{-xg(0)}+\int_0^{\infty} e^{-xu}g^{-1}(u)du+\int_{g(0)}^{\infty}e^{-xu} \{h_m(u)\}du+R_{m-1}(x)

766: \end{equation}

767: where $R_{m-1}$ is $C^{m-1}$ and $h_m$ is a truncation of the asymptotic

768: expansion of $h$, such that $h(u)-h_m(u)=o(u^{-m})$.

769:     \end{Lemma}

770:     \begin{proof}

771: We have

772: 

773: \begin{multline}

774:   \label{eq:sum2}

775:   \int_0^{\infty} e^{-xu} \{h(u)\}  du=

776: \sum_{N=0}^{\infty}\int_{g(N)}^{g(N+1)} e^{-xu}(h(u)-N) du\\=

777: \sum_{N=0}^{\infty}\int_{g_m(N)}^{g_m(N+1)} e^{-xu}(h_m(u)-N)du

778: +R_{m-1}(x)

779: \end{multline}

780: where

781: \begin{multline}

782:   \label{eq:eqR}

783: R_{m-1}(x)=\sum_{N=0}^{\infty}\left(\int_{g(N)}^{g_m(N)}+\int_{g_m(N+1)}^{g(N+1)}\right)

784:  e^{-xu} \{h(u)\}  du\\+\sum_{N=0}^{\infty}\int_{g_m(N)}^{g_m(N+1)}e^{-xu}(h(u)-h_m(u))du

785: \end{multline}

786: Using (\ref{eq:eq21}) and (\ref{eq:eqM}) the proof follows and the

787: fact that $g(N)-g_m(N)=o(N^{-m})$ and $h(u)-h_m(u)=o(u^{-m})$, the sum is rapidly convergent, and the result follows. \end{proof}

788: \begin{Lemma}

789:   If $h$ is a finite sum of powers, then $f-e^{-xg(0)}-\int_0^{\infty} e^{-xu}g^{-1}(u)du\in C^\infty$.

790: \end{Lemma}

791: \begin{proof}

792: 

793: Using (\ref{eq:eq21}) and (\ref{eq:eqM}) we have

794: \begin{multline}

795:   \label{eq:eqhm}

796:   f_1(x)=  x\int_0^{\infty}e^{-xu} \{h_m(u)\}du=\frac{1}{2}-

797: \sum_{k=1}^{\infty}\int_0^{\infty}e^{-xu}\frac{\sin 2k\pi h_m(u)}{k\pi}du\\

798: =\frac{1}{2}-

799: \sum_{k=1}^{\infty}\frac{1}{2k\pi i}\sum_{\sigma=\pm 1}\sigma  \int_0^{\infty}e^{-xu+\sigma 2k\pi i h_m(u)}du

800: \end{multline}

801:   We deform the contours of integration along the directions

802: $\sigma\alpha$ respectively, say for $\alpha=\pi/2$.  The integral

803: \begin{equation}

804:   \label{eq:vert}

805:  f_1= \int_0^{i\infty}u^n e^{2k\pi i h_m(u)}du

806: \end{equation}

807: exists for any $n$ and it is estimated by

808: \begin{equation}

809:   \label{eq:estm3}

810:  \left|\int_0^{i\infty}u^n e^{2k\pi i h_m(u)}du\right|<const.  k^{-b(n+1)}\Gamma(n(b+1))

811: \end{equation}

812: The termwise nth derivatives at  $0^+$ of the series of $f_1(x)$ converge rapidly,

813: and the result follows.

814: \end{proof}

815: 

816: \subsection{Proof of Theorem \ref{Cexp}}

817: \begin{proof}

818:   Let $\epsilon>0$ be arbitrary, let

819: $$S_N=\sum_{j=1}^N e^{i y g(j)};\ \ \ S_0:=0$$

820: and let $N_1$ be large enough so that $|S_N/\rho(N)-L|\le \epsilon$ for $N>N_1$. Then, by looking at $e^{i\phi} f$ if necessary, we can assume that $L\ge 0$. We have

821: \begin{multline}

822:   \label{eq:eqf3}

823:   f(x+iy)=\sum_{k=1}^{\infty}e^{-xg(k)}(S_{k}-S_{k-1})=\sum_{k=1}^{\infty}\frac{S_k}{\rho(k)}(e^{-xg(k)}-e^{-xg(k+1)})\rho(k)\\=L(y)\sum_{k=1}^{\infty}(e^{-xg(k-1)}-e^{-xg(k)}){\rho(k)}+\sum_{k=1}^{\infty}(e^{-xg(k)}-e^{-xg(k+1)})d_k{\rho(k)}

824: \end{multline}

825: where $d_k=L(y)-S_k/\rho(k) \to0$ as $k\to\infty$. Now, 

826: \begin{equation}

827:   \label{eq:67}

828:  \sum_{k=1}^{\infty}e^{-xg(k)}(\rho(k+1)-\rho(k)) \sim \int_0^{\infty}e^{-xg(k)}\rho'(k)dk =:\Phi_{\rho}(x).

829: \end{equation}

830: and under the given assumptions $\Phi_{\rho}\to \infty$ as $x\to 0$. 

831: Note that $\Phi_{\rho}\to \infty$ as $x\to 0$ and the facts that $e^{-xg(k)}-e^{-xg(k+1)}>0$, $\rho(k)>0$ and $d_k\to 0$, readily imply that the last sum in (\ref{eq:eqf3}) is $o(\Phi_{\rho}(x))$; (\ref{eq:lsup}) is follows in a similar way.

832: \end{proof}

833: 

834: \subsection{Proof of Theorem\ref{Pp4}}

835: We rely on Theorem \ref{Cexp}, and analyze the case $b=3/2$; the

836: case $b\in\NN$ is simpler. We have $\beta=t/(2\pi)$. Let

837: $\check{S}_j=\sum \limits_{k=1}^{j}e^{-2\pi i \dfrac{m}{n} k^3}\!\!.$ It

838: is clear that for $m,n\in\NN$ we have $j^{-1}\check{S}_j\to \sum

839: \limits_{k=1}^{n}e^{-2\pi i \dfrac{m}{n} k^3}=L_{mn}$. On the other

840: hand, by summation by parts we get

841: 

842: \begin{multline}

843:   \label{eq:eq01}

844: S_N= \frac{2\sqrt{2}e^{\pi i/4}}{3\beta}\sum_{k=1}^{k_N-1} k^{1/2}

845:  \exp\left(-\frac{8\pi i}{27\beta^2}k^3\right)+O(N^{1/4})=O(N^{1/4})\\+

846: \frac{2\sqrt{2}e^{\pi i/4}}{3\beta}(k_N-1)^{3/2}\frac{\check{S}_{k_N-1}}{k_N-1}

847: +\frac{2\sqrt{2}e^{\pi i/4}}{3\beta}\sum_{k=1}^{k_N-1}k^{-1}\check{S}_k k(k^{1/2}-(k-1)^{1/2})

848: \end{multline}

849: and it is easy to check that, as $N\to\infty$ we have

850: \begin{equation}

851:   \label{eq:sum}

852: S_N\sim  \left(\frac{2\sqrt{2}e^{\pi i/4}}{3\beta}(1+1/3)\right) k_N^{3/2} L_{mn}=\frac{8\sqrt{2}e^{\pi i/4}}{9\beta}\sqrt{\frac{3\beta}{2}}N^{3/4}

853: \end{equation}

854: where we used the definition of $k_N$ following

855: eq. (\ref{eq:eq01}), which implies $k_N\sim \frac{3}{2}N^{1/2}\beta$. The result follows by changes of variables, using

856: (\ref{eq:lim2}) and noting that

857: \begin{equation}

858:   \label{eq:int4}

859:  \frac{3}{4} \int_0^{\infty}e^{-x k^{3/2}}k^{-1/4}dk=\frac{\sqrt{\pi}}{2\sqrt{x}}

860: \end{equation}

861: 

862: 

863: It is also clear that $|j^{-1}\check{S}_j|\le 1$ and a similar calculation

864: provides an overall upper bound.

865: 

866: 

867: 

868: Section \ref{Direct,3} provides an independent way to calculate the behavior

869: along the boundary.

870: \subsection{Proof of Proposition \ref{b=2}}

871: For $b=2$ using  (\ref{eq:eqhm}) and

872: $\int_{0}^{\infty}e^{-px}\sin2k\pi

873: p^{\frac{1}{2}}dp=\pi^{3/2}ke^{-\frac{k^{2}\pi^{2}}{x}}x^{-3/2}$

874: we immediately obtain

875: 

876: \[

877: f(x)=\frac{\sqrt{\pi}}{2\sqrt{x}}-\dfrac{1}{2}+\frac{\sqrt{\pi}}{\sqrt{x}}\sum_{k=1}^{\infty}e^{-\frac{k^{2}\pi^{2}}{x}}\]

878: 

879: \subsection{The case $g(j)=a^j$}

880: 

881: 

882: 

883: 

884: It is easy to see that, for this choice of $g$, we have the functional

885: relation $$f(z)-f(az)=e^{-z}$$

886: 

887: Since $e^{-z}\to 1$ as $z\rightarrow0+$, the leading behavior formally

888: satisfies $f(z)-f(az)\sim1$ i.e. $f(z)\sim-\dfrac{\log z}{\log a}$

889: 

890: In view of  this we let $f(z)=-\dfrac{\log z}{\log a}+G(z)$ which

891: gives  $G(z)-G(az)=e^{-z}-1$ {\em i.e.}

892: 

893: \begin{equation}

894:   \label{eq:eqg}

895:  G(z)=G(z/a)+(1-e^{-z/a})

896: \end{equation}

897: We first obtain a solution a solution $\check{G}$ of the homogeneous equation and then

898: write $G=\check{G}+h(a^z)$, where $h$ now satisfies

899: \begin{equation}

900:   \label{eq:eqhy}

901:   {h}(y)={h}(y+1)

902: \end{equation}

903:  Iterating (\ref{eq:eqg}) we obtain

904: \[

905: \check{G}(z)=\sum_{n=0}^{\infty}(1-e^{-z/a^n})=-\sum_{n=0}^{\infty}\sum_{k=1}^{\infty}\frac{(-z)^{k}}{k!a^{nk}}=\sum_{n=1}^{\infty}\frac{(-z)^{n}}{n!(1-a^{n})}\]

906: which is indeed an entire function and satisfies the functional

907: relation.  Now we return to (\ref{eq:eqhy}). Since by its connection to

908: $f(z)$, ${h}$ is obviously smooth in $y$, it can be expressed in

909: terms of its Fourier series

910: 

911: \[

912: {h}(y)=\sum_{k=-\infty}^{\infty}c_{k}e^{2k\pi iy}\] where the

913: coefficients $c_{k}=\bar{c}_{-k}$ can be found by using the original

914: function $f$. Recall that we have $$f(z)=-\dfrac{\log z}{\log

915:   a}+\check{G}(z)+{h}\(\dfrac{\log z}{\log a}\)$$ which

916: implies

917: 

918: \[

919: {h}(y)=f(a^{y})+y-\check{G}(a^{y})\]

920: \begin{Lemma}\label{Lgeom}

921:   The Fourier coefficients of the periodic function ${h}$ are given

922: by

923: \begin{equation}

924:   \label{eq:ck}

925:   c_{k}=\frac{1}{\log a}\Gamma\(-\frac{2k\pi i}{\log a}\)\:\(k\neq0\)

926: \end{equation}

927: and

928: $$c_{0}=\int_{0}^{1}f(a^{y})dy+\int_{0}^{1}ydy-\int_{0}^{1}\check{G}(a^{y})dy$$

929: Since $c_{k}\sim\sqrt{\dfrac{1}{\pi\log a}}e^{\frac{-|k|\pi^{2}}{\log a}}$,

930: (implying that the Fourier  expansion  for $h$ is valid exactly

931: for $\Re\(z\)>0$) we have

932: \[

933: f\(z\)=-\dfrac{\log z}{\log a}+\sum_{n=1}^{\infty}\frac{\(-z\)^{n}}{n!\(1-a^{n}\)}+c_{0}+\frac{1}{\log a}\sum_{k\neq0}\Gamma\(-\frac{2k\pi i}{\log a}\)z^{\frac{2k\pi i}{\log a}}\]

934: The series further resums to

935: \begin{equation}

936:   \label{eq:resum}

937:   f\(z\)=-\dfrac{\log z}{\log

938: a}+\sum_{n=1}^{\infty}\frac{\(-z\)^{n}}{n!\(1-a^{n}\)}+c_{0}-\frac{z}{2\pi

939: i}\int_{0}^{\infty}\log_{\RR}\(-s^{\frac{2\pi i}{\log a}}\)e^{-sz}ds

940: \end{equation}

941: valid for $z>0$.

942: \end{Lemma}

943: The proof is given in \S\ref{pfLgeom}.

944: 

945: 

946: Similar results hold for other rational angles if $\bb\in\NN$ (by

947: grouping terms with the same phase). For example,

948: \begin{multline}

949:   \sum_{n=0}^{\infty}e^{-2^{n}(z+\frac{2}{5}\pi i)}\\=e^{-\frac{2}{5}\pi

950: i}\sum_{n=0}^{\infty}e^{-2^{4n}z}+e^{-\frac{4}{5}\pi

951: i}\sum_{n=0}^{\infty}e^{-2^{4n+1}z}+e^{-\frac{8}{5}\pi

952: i}\sum_{n=0}^{\infty}e^{-2^{4n+2}z}+e^{-\frac{6}{5}\pi

953: i}\sum_{n=0}^{\infty}e^{-2^{4n+3}z}\\

954: =e^{-\frac{2}{5}\pi

955: i}\sum_{n=0}^{\infty}e^{-16^{n}z}+e^{-\frac{4}{5}\pi

956: i}\sum_{n=0}^{\infty}e^{-16^{n}(2z)}+e^{-\frac{8}{5}\pi

957: i}\sum_{n=0}^{\infty}e^{-16^{n}(4z)}+e^{-\frac{6}{5}\pi

958: i}\sum_{n=0}^{\infty}e^{-16^{n}(8z)}

959: \end{multline}

960: \subsection{Proof of Theorem~\ref{PP3}}\label{AC1}

961: (i) and (ii):

962:  By an argument similar to the one leading to (\ref{eq:eqhm}) we have

963: \begin{equation}

964:   \label{eq:dec3}

965:   f(z)=\int_0^{\infty} e^{-zg(s)}  ds-\frac{1}{2}+

966: \sum_{k=1}^{\infty}\frac{1}{2k\pi i}\sum_{\sigma= \pm 1}\int_0^{\sigma i\infty}e^{-zu+\sigma 2k\pi i u^{1/b}}du

967: \end{equation}

968: The exponential term ensures convergence in $k$. 

969: Taking $s^b=t$ we see that

970: \begin{equation}

971:   \label{eq:eqt1}

972:   \int_0^{\infty} e^{-z s^b}dp=\Gamma(1+1/b)z^{-1/b}

973: \end{equation}

974: We now analyze the case $\sigma=-1$, the other case being similar. A term in the sum is

975: \begin{multline}

976:   \label{eq:trm1}

977:   \frac{1}{2k\pi i}\int_0^{ i\infty}e^{-zu-2k\pi i u^{1/b}}du=

978:   \left(\frac{k}{z}\right)^{{d}}\int_0^{

979:     \infty}e^{-\nu_k(t+2\pi i t^{1/b})}dt\\=\left(\frac{k}{z}\right)^{{d}}

980: \int_0^{

981:     \infty}e^{-\nu_k s}{\check{H}_{-}}(s)ds=z^{-{d}}\int_0^{

982:     \infty}e^{-z^{-1/(b-1)} s}{\check{H}_{-}}(s/k^{{d}})ds

983: \end{multline}

984: where $\nu_k={k^{{d}}}{z^{-1/(b-1)}}$, ${\check{H}_{-}}(s)=\phi'(s)$ and 

985: \begin{equation}

986:   \label{eq:eqphi}

987:  \Phi(\phi(u)):= \phi(u)+2\pi i \phi(u)^{1/b}=u

988: \end{equation}

989: (a) Near the origin, we write $\phi=u^b H_{b}$, and we get 

990: $$H_{b}=\frac{1}{2\pi i}(1-u^{b-1}H_{b})^b$$

991: and analyticity of $H_{b}$ in $u^{b-1}$ follows, for instance, from the contractive

992: mapping principle. 

993: 

994: (b) The only singularities of $\phi$, thus of ${\check{H}_{-}}$, are the points implicitly given by $\Phi'=0$.

995: 

996: 

997: (b) It is also easy to check that for some $C>0$ we have

998: \begin{equation}

999:   \label{eq:eqH1}

1000:   |{\check{H}_{-}}(u)|\le C\frac{|u^{b-1}|}{1+|u^{b-1}|}

1001: \end{equation}

1002: uniformly in $\CC$, with a

1003: cut at the singularity, or on a corresponding Riemann surface.

1004: From (\ref{eq:eqH1}) we see that the sum

1005: \begin{equation}

1006:   \label{eq:eqsum}

1007:   H_{-}=\sum_{k=1}^{\infty} {\check{H}_{-}}(s/k^{{q}})

1008: \end{equation}

1009: converges, on compact sets in $s$, at least as fast as $const\sum k^{-b}$,

1010: thus it is an analytic function wherever {\em all} ${\check{H}_{-}}(s/k^{{d}})$ are

1011: analytic, that is, in $\CC$ except for the 

1012: points $k^d s_0$. Using an integral estimate, we get the

1013: global bound

1014: \begin{equation}

1015:   \label{eq:eqinfty}

1016:   |H_{-}|\le const \sum_{k=1}^{\infty} \frac{|u|^{b-1}}{k^q+|u|^{b-1}}\le const |u|^{(b-1)^{2}/b}

1017: \end{equation}

1018: as $|u|\to \infty$.

1019: 

1020: The function $H$ in the lemma is simply $H_{-}+H_{-}$ where 

1021: $H_+$ is obtained from $H_{-}$ by replacing $i$ by $-i$. The calculation

1022: of the explicit power series is straightforward, from (\ref{eq:eqphi}),

1023: (\ref{eq:eqsum}) and the similar formulae for $H_+$, using dominated 

1024: convergence based on (\ref{eq:eqinfty}). We provide the details for convenience.

1025: 

1026: We  write the last term in (\ref{eq:trm1}) in the form 

1027: \begin{equation}

1028:   \label{eq:deffk}

1029:   f_{k}(z)=\(\frac{k}{z}\)^{\frac{b}{b-1}}\intop_{C}e^{-s(\frac{k^{b}}{z})^{\frac{1}{b-1}}}\check{H}_{-}ds

1030: \end{equation}

1031: where  the contour $C$ is a curve from the origin to $\infty$ in the

1032: first quadrant.

1033: 

1034: 

1035: 

1036: Watson's lemma implies \[

1037: f_{k}(z)=\sum_{j=1}^{N}b_{j}k^{-jb}z^{j-1}+R_{N}(k,z)\] where

1038: $b_{j}$can be calculated explicitly from Lagrange-B\"urmann inversion

1039: formula used for the inverse function $\phi^{-1}$ and $R_{N}(k,z)\leqq

1040: Ck^{-(N+1)b}z^{N}$, for arbitrary $N\in\mathbb{N}$. It follows that

1041: \begin{equation}

1042:   \label{eq:srjtob}

1043:   f(z)\sim\Gamma(1+\dfrac{1}{b})z^{-\frac{1}{b}}-\dfrac{1}{2}+\frac{i}{2\pi}\sum_{j=1}^{\infty}(1-(-1)^{jb})\zeta(jb+1)b_{j}z^{j}

1044: \end{equation}

1045: which holds  for $z\rightarrow0$ in the right half plane in any

1046: direction not tangential to the imaginary axis. 

1047: 

1048: (iii) We  obtain the transseries (which gives us  information

1049: near the imaginary axis) by using the global properties of

1050: $\check{H}_{-}$, and standard deformation of the Laplace contour.

1051: 

1052: As $z$ goes around the complex plane, as usual in Laplace-like

1053: integrals, we rotate $s$ in (\ref{eq:deffk}) simultaneously,

1054: to keep the exponent real and positive. In the process,

1055: as we cross singularities, we collect a 

1056: contribution to the integral from the point $s_{-}$ above; the contribution

1057: is an integral around a cut originating at $s_{-}$. The singularity is integrable, and

1058: collapsing the contour to the cut itself, we get a 

1059: contribution again in the form of a Laplace transform. 

1060: This is the Borel sum (in the same variable, $z^{-1/(b-1)}$. 

1061: 

1062: Generically $s_{-}$ is a

1063: square root branch point and we have

1064: \[

1065: \frac{ds_{-}}{ds}=\sum_{j=0}^{\infty}\tilde{c}_{j}\left[s-\left(\dfrac{b

1066: i}{2\pi}\right)^{\frac{b}{1-b}}\right]^{\frac{j-1}{2}}\]

1067: The asymptotic expansion of the cut contribution, Borel summable

1068: as we mentioned, is

1069: \[

1070: \exp \left({-(2\pi)^{\frac{b}{b-1}}(b^{\frac{b}{1-b}}-b^{\frac{1}{1-b}})i^{\frac{b}{1-b}}\(\frac{k^{b}}{z}\)^{\frac{1}{b-1}}}\right)\sum_{j=0}^{\infty}c_{j}\(\frac{z}{k^{b}}\)^{\frac{2j+1}{2(b-1)}}\]

1071: The exponential term ensures convergence in $k$ of the Borel summed transseries. A similar result

1072: can be obtained for $-k.$

1073: 

1074: Thus, the transseries of $f$ is of the form

1075: 

1076: \[

1077: \tilde{f}(z)=\Gamma(1+\dfrac{1}{b})z^{-\frac{1}{b}}-\dfrac{1}{2}+\frac{i}{2\pi}\sum_{j=1}^{\infty}(1-(-1)^{jb})\zeta(jb+1)b_{j}z^{j}\]

1078: 

1079: 

1080: \[

1081: +\begin{cases}

1082: \frac{i}{2\pi}\sum\limits _{k=1}^{\infty}e^{(b-1)(\frac{2\pi}{b i})^{\frac{b}{b-1}}(\frac{k^{b}}{z})^{\frac{1}{b-1}}}\sum\limits _{j=0}^{\infty}c_{j}k^{\frac{-2jb+2-b}{2(b-1)}}z^{\frac{2j-1}{2(b-1)}} & \,-\frac{\pi}{2}\leqslant\arg z\leqslant\theta_{1}\\

1083: -\frac{i}{2\pi}\sum\limits _{k=1}^{\infty}e^{(b-1)(\frac{2\pi}{b

1084: i})^{\frac{b}{b-1}}(\frac{(-k)^{b}}{z})^{\frac{1}{b-1}}}\sum\limits

1085: _{j=0}^{\infty}c_{j}(-k)^{\frac{-2jb+2-b}{2(b-1)}}z^{\frac{2j-1}{2(b-1)}}

1086: & \,\theta_{2}\leqslant\arg z\leqslant\frac{\pi}{2}\end{cases}\]

1087: 

1088: \begin{Remark}

1089:  The analysis can be extended to the case

1090: 

1091: \[

1092: f(z)=\sum_{k=1}^{\infty}F(k)e^{-k^{\bb}z}\:(\lambda\neq0)\]

1093: by noticing that

1094: 

1095: \[

1096: f(z)=\int_{0}^{\infty}(zF(p)-F'(p))e^{-pz}[p^{\frac{1}{\bb}}]dp\] If

1097: the expression of $F(k)$ is simple, for example $F(k)$ is a finite

1098: combination of terms of the form

1099: $\mu^{k^{\frac{1}{\bb}}}k^{\lambda}(\log k)^{m}\:(m=0,1,2,3...)$,

1100: the method in this section applies with little change to calculate the

1101: transseries of $f(z)$.

1102: \end{Remark}

1103: For special values of $\bb$, asymptotic information as $z$ approaches

1104: the imaginary line can be  obtained in the following way. Let

1105: $z=\delta+2\pi i\beta$ and

1106: \[

1107: f(z)=\sum_{k=1}^{\infty}e^{-k^{\bb}(\delta+2\pi i\beta)}\:(\Re(\delta)>0)\]

1108: If $1<\bb\in\NN$, we

1109: may obtain the asymptotic behavior for all rational $\beta=\dfrac{m}{n}$,

1110: by noting that $e^{-k^{\bb}(2\pi i\beta)}=e^{-(k+n)^{\bb}(2\pi

1111:   i\beta)}$ and splitting the sum into

1112: \[

1113: f(z)=\sum_{j=0}^{\infty}\sum_{l=1}^{n}e^{-(nj+l)^{\bb}\(\delta+2\pi

1114: i\dfrac{m}{n}\)}=\sum_{l=1}^{n}e^{-2\pi

1115: i\dfrac{m}{n}l^{\bb}}\sum_{j=0}^{\infty}e^{-(nj+l)^{\bb}\delta}\]

1116: It follows that

1117: \[

1118: \sum_{j=0}^{\infty}e^{-(nj+l)^{\bb}\delta}=n^{\bb}\delta\int_{l^b/n^b}^{\infty}e^{-n^{\bb}\delta s}\left(s^{\frac{1}{\bb}}-l/n\right)ds\]

1119: \[

1120: =n^{\bb}\delta \int_{l^\bb/n^\bb}^{\infty}e^{-n^{\bb}\delta s}

1121: \(s^{\frac{1}{\bb}}-l/n\)ds-n^{\bb}\delta

1122: \int_{l^\bb/n^\bb}^{\infty}e^{-n^{\bb}\delta s}\left\{s^{\frac{1}{\bb}}-l/n\right\}ds\]

1123: by the argument  above. Since the absolute value of the

1124: fractional part  does not exceed one, we have the

1125: estimate

1126: \[

1127: \sum_{j=0}^{\infty}e^{-(nj+l)^{\bb}\delta }=n^{\bb}\delta \int_{(\frac{l}{n})^{\bb}}^{\infty}e^{-n^{\bb}\delta s}s^{\frac{1}{\bb}}ds+O(1)=\frac{1}{n}{\Gamma\(1+\dfrac{1}{\bb}\)}\delta ^{-\frac{1}{\bb}}+O(1)\:(z\rightarrow0)\]

1128: This implies

1129: \begin{equation}

1130:   \label{eq:deffz}

1131:   f(z)=\frac{1}{n}\sum_{l=1}^{n}e^{-2\pi

1132: i\dfrac{m}{n}l^{\bb}}\Gamma\(1+\dfrac{1}{\bb}\)z^{-\frac{1}{\bb}}+O(1)

1133: \end{equation}

1134: Therefore $f(z)$ either blows up like $z^{-\frac{1}{\bb}}$(when

1135: $\sum_{l=1}^{n}e^{-2\pi i\dfrac{m}{n}l^{\bb}}\neq0$) or it is bounded

1136: (when $\sum_{l=1}^{n}e^{-2\pi i\dfrac{m}{n}l^{\bb}}=0$).

1137: 

1138: The Fourier expansion of the fractional part can be used to calculate

1139: the transseries as we did for $\beta=0$, but we shall omit the

1140: calculation here.

1141: 

1142: For special values of $b$, asymptotic information is relatively easy

1143: to obtain on a dense set along the barrier.  This is the case when

1144: $\bb=\dfrac{r+1}{r}$ where $r\in\NN$; then, the transseries contains

1145: exponential sums   in terms of integer powers, $k^{r}$, a

1146: consequence of the duality relation $\dfrac{1}{b}+\dfrac{1}{d}=1$,

1147: which at the transseries level is of the form $\sum

1148: e^{-k^{b}z}\rightarrow g_{1}+\sum e^{ck^{d}z^{-d/b}}g_{2}$ where

1149: $g_{1},g_{2}$ are power series.  We illustrate this for

1150: $\bb=\dfrac{3}{2}$.

1151: 

1152: Without loss of generality, we assume $\beta<0$. The transseries of

1153: $f$ is given in (\ref{Tr3/2}).  To estimate the asymptotic behavior of

1154: $f(z)$ as $z$ approaches the imaginary line, we rewrite (\ref{Tr3/2}) as

1155: \begin{multline}

1156:   f(z)=\Gamma\(\frac{5}{3}\)z^{-\frac{2}{3}}-\dfrac{1}{2}\\+\frac{i}{2\pi}\sum_{k=1}^{\infty}\(\frac{k}{z}\)^{2}\int_{0}^{\infty}e^{-s\frac{k^{3}}{z^{2}}}\(\dfrac{3}{4\sqrt{2}(\pi

1157: i)^{\frac{3}{2}}}s^{\frac{1}{2}}-\dfrac{3i}{8\pi^{3}}s-\dfrac{105i^{\frac{3}{2}}}{256\sqrt{2}\pi^{\frac{9}{2}}}s^{\frac{3}{2}}+\cdots\)ds\\

1158: +\sum_{k=1}^{\infty}e^{\dfrac{32i\pi^{3}k^{3}}{27z^{2}}}\dfrac{4\sqrt{2}i^{-\frac{1}{2}}\pi}{3}k^{\frac{1}{2}}z^{-1}\\\frac{i}{\pi}\sum_{k=1}^{\infty}\(\frac{k}{z}\)^{2}e^{-\dfrac{32i\pi^{3}k^{3}}{27z^{2}}}\int_{0}^{\infty}e^{-s\frac{k^{3}}{z^{2}}}\(\dfrac{i^{-\frac{1}{2}}}{8\sqrt{2}\pi^{\frac{3}{4}}}(s-s_{0})^{\frac{1}{2}}+\cdots\)ds

1159: \end{multline}

1160:  Watson's Lemma implies that

1161:  \begin{equation}

1162:    \label{eq:eqtest0}

1163:    f(z)=\dfrac{4\sqrt{2}i^{-\frac{1}{2}}\pi}{3z}\sum_{k=1}^{\infty}k^{\frac{1}{2}}e^{\dfrac{32i\pi^{3}k^{3}}{27z^{2}}}+O(1)

1164:  \end{equation}

1165:  The is sum in (\ref{eq:eqtest0}) is  similar to the  sum with

1166:  $\bb=3$, and can be estimated in a similar way:

1167: 

1168: \[

1169: \sum_{k=1}^{\infty}k^{\frac{1}{2}}e^{-(y+2\pi i\frac{m}{n})}=\(\sum_{l=1}^{n}e^{-2\pi i\dfrac{m}{n}l^{3}}\)\frac{\sqrt{\pi}}{3n\sqrt{y}}+o\(\frac{1}{\sqrt{y}}\)\]

1170: 

1171: 

1172: Setting $\dfrac{32i\pi^{3}k^{3}}{27z^{2}}=y+2\pi i\dfrac{m}{n}$ , we

1173: have $z=-\dfrac{4\pi

1174: i}{3\sqrt{3}}\sqrt{\dfrac{n}{m}}$+$\(\dfrac{n}{m}\)^{\frac{3}{2}}y+o(y)$.

1175: The asymptotic behavior can be obtained for $\beta=-\dfrac{4\pi

1176: i}{3\sqrt{3}}\sqrt{\dfrac{n}{m}}$ (this includes all rationals) by

1177: substituting $y=\dfrac{32i\pi^{3}k^{3}}{27z^{2}}-2\pi i\dfrac{m}{n}$

1178: in the above estimates. Setting $z =-\dfrac{4\pi

1179: i}{3\sqrt{3}}\sqrt{\dfrac{n}{m}}$+$\delta$ ($\delta>0$), a direct

1180: calculation shows that

1181: \begin{equation}

1182:   \label{eq:eqtest}

1183:   \sqrt{\Re(z )}f(z )=\frac{\sqrt{6\pi

1184: i}}{3^{\frac{7}{4}}}\(\frac{n}{m}\)^{\frac{1}{4}}\frac{1}{n}\sum_{l=1}^{n}e^{-2\pi

1185: i\dfrac{m}{n}l^{3}}+o(1)

1186: \end{equation}

1187: \subsection{Details of the proof of

1188:  Theorem~\ref{P1}}\label{genblowup}

1189: Consider, more generally,

1190: \[

1191: f(\delta,\beta)=\sum_{k=0}^{\infty}a(k)e^{-g(k)(\delta+2\pi i\beta)}\:, \ \ \delta>0\]

1192: where $g(k)>0$ is a real function

1193: and$\int_{0}^{\infty}|a(t)|^{2}dt=\infty$.

1194: 

1195: We find  the behavior of

1196: \begin{multline}

1197: \int_{\beta_{0}}^{\beta_{1}}|f(\delta,\beta)|^{2}d\beta\\

1198: =\int_{\beta_{0}}^{\beta_{1}}\sum_{k=0}^{\infty}|a(k)|^{2}e^{-2g(k)\delta}d\beta+\int_{\beta_{0}}^{\beta_{1}}\sum_{k\neq

1199: j}a(k)\bar{a}(j)e^{-(g(k)+g(j))\delta+(g(j)-g(k))2\pi i\beta}d\beta

1200: \\

1201:   =(\beta_{1}-\beta_{0})\sum_{k=0}^{\infty}|a(k)|^{2}e^{-2g(k)\delta}\\

1202: +\frac{1}{2\pi

1203: i}\sum_{k\neq

1204: j}\frac{a(k)\bar{a}(j)}{g(j)-g(k)}e^{-(g(k)+g(j))\delta}e^{(g(j)-g(k))2\pi

1205: i\beta_{0}}\left(e^{(g(j)-g(k))2\pi i(\beta_{1}-\beta_{0})}-1\right)

1206: \end{multline}

1207: where

1208: $\beta_{0,1}\in\RR$ are arbitrary, or after $m$ integrations,

1209: \begin{multline}

1210:   F(\delta)=\int_{\beta_{m-1}}^{\beta_{m-1}+c_{m-1}}\cdots\int_{\beta_{1}}^{\beta_{1}+c_1}\int_{\beta_{0}}^{\beta_{0}+c_0}|f(\delta,\beta)|^{2}d\beta d\beta_{0}\cdots

1211:   d\beta_{m-2}\\

1212:   =c_0c_1\cdots c_m\sum_{k=0}^{\infty}|a(k)|^{2}e^{-2g(k)\delta}+O\left(\sum_{k\neq j}\left|\frac{a(k)\bar{a}(j)}{(g(j)-g(k))^{m}}\right|e^{-(g(k)+g(j))\delta}\right)

1213: \end{multline}

1214: 

1215: Note that

1216: \begin{multline}

1217:  \sum_{k\neq

1218: j}\left|\frac{a(k)\bar{a}(j)}{(g(j)-g(k))^{m}}\right|e^{-(g(k)+g(j))\delta}\\=2\sum_{k=0}^{\infty}\sum_{n=1}^{\infty}\left|\frac{a(k)\bar{a}(k+n)}{(g(k+n)-g(k))^{m}}\right|e^{-(g(k)+g(k+n))\delta}\\

1219: =2\sum_{k=0}^{\infty}|a(k)|e^{-2g(k)\delta}\sum_{n=1}^{\infty}\left|\frac{a(k+n)}{(g(k+n)-g(k))^{m}}\right|e^{-(g(k+n)-g(k))\delta}\\

1220: =O\left(\sum_{k=0}^{\infty}|a(k)|e^{-2g(k)\delta}\sum_{n=1}^{\infty}\left|\frac{a(k+n)}{(g(k+n)-g(k))^{m}}\right|\right)

1221: \end{multline}

1222: under our assumption.

1223: 

1224: If furthermore we have

1225: \[

1226: \sum_{n=1}^{\infty}\left|\frac{a(k+n)}{(g(k+n)-g(k))^{m}}\right|=o(a(k))\]

1227: then we obtain

1228: \begin{equation}

1229:   F(\delta)=c_0c_1\cdots c_m\sum_{k=0}^{\infty}|a(k)|^{2}e^{-2g(k)\delta}+O\left(\sum_{k=0}^{\infty}\frac{|a(k)|^{2}}{G(k)}e^{-2g(k)\delta}\right)

1230: \end{equation}

1231: where $G(k)>1,\, G(k)\rightarrow\infty$ as $k\rightarrow\infty$,

1232: or

1233: \begin{equation}

1234:   \label{eq:eqB}

1235:  F(\delta)\left(\sum_{k=0}^{\infty}|a(k)|^{2}e^{-2g(k)\delta}\right)^{-1} =c_0\cdots c_m+o(1)\ \ as\ \delta\to 0^+

1236: \end{equation}

1237: The result now follows from the following lemma.

1238: \begin{Proposition}

1239:   Assume

1240: 

1241: (i) $h_n:\RR\to [0,\infty)$ are locally $L^1$, and

1242: 

1243: (ii) $\displaystyle \lim_{n\to\infty}\int_{B} h_n(x_1+\cdots +x_N)dx_1\cdots dx_N= meas(B)$ for any box $B\displaystyle=\prod_{i=1}^N[a_i,b_i]$.

1244: 

1245: Then

1246:  $h_n\to 1$ in the dual of $C[\alpha,\beta]$ for any $[\alpha,\beta]$.

1247: 

1248: \end{Proposition}

1249: 

1250: \begin{proof}

1251:   We first take $N=2$, the general case will follow by induction on $N$.

1252: 

1253:   Consider the rectangle $B_{c}=(a,b-c)\times (0,c)$, $0<c<b-a$. By changing coordinates

1254: to $x+y=s,y=y'$, we get that

1255: \begin{equation}

1256:   \label{eq:Tc}

1257:  c^{-1}\int_{B} h_ndydx= \int_a^bh_n(s)T_{a,b;c}(s)ds\to meas(T_{a,b;c});\ \text{as } n\to\infty

1258: \end{equation}

1259: where $T_{a,b;c}(\cdot)$ is the function having $T_{a,b;c}$ as a

1260: graph, $T_{a,b;c}$ being an isosceles trapezoid with lower base the

1261: interval $(a,b)$ and upper base of length $b-a-2c$ at height 1.

1262: 

1263: 

1264: We also note that the indicator function of $[a,b]$, $\mathbf{1}_{ab}$,

1265: satisfies the inequalities $T_{a-c,b+c;c}\ge \mathbf{1}_{ab}\ge

1266: T_{a,b;c}$. Thus, since $c$ is arbitrary and $h_n\ge 0$, and both

1267: $meas (T_{a-c,b+c;c})$, and $meas (T_{a,b;c})$ tend to $(b-a)$ as

1268: $c\to 0$, we have

1269: \begin{equation}

1270:   \label{eq:(b-a)}

1271:   \lim_{n\to\infty}\int_a^bh_n(s)ds =(b-a) =meas([a,b])

1272: \end{equation}

1273: In particular, given $\alpha<\beta$, $\|h_n\|_{L^1[\alpha,\beta]}$ are uniformly

1274: bounded, that is, for some $C\ge 1$ we have

1275: \begin{equation}

1276:   \label{eq:L1}

1277:  \sup_{n\ge 1}\|h_n\|_{L^1[\alpha,\beta]}\le C(\beta-\alpha)

1278: \end{equation}

1279: Since a continuous function on $[\alpha,\beta]$ is approximated

1280: arbitrarily well in sup norm by finite linear combinations of

1281: indicator functions of intervals, it follows from (\ref{eq:(b-a)}),

1282: (\ref{eq:L1}) \footnote{Alternatively, and somewhat more compactly,

1283: one can prove the result without the intermediate steps (\ref{eq:(b-a)})

1284: and (\ref{eq:L1}) by upper and lower bounding continuous functions

1285: by sums of trapezoids.}

1286:  and the triangle inequality that

1287: \begin{equation}

1288:   \label{eq:concl}

1289:   \int_\alpha^\beta h_n(s)f(s)ds\to\int_\alpha^\beta f(s)ds, \ \forall\,f\in C[\alpha,\beta]

1290: \end{equation}

1291: For general $N$ we use $h^-_n(s)=\int_{B'}h_n(s+x_1+...+x_{N-1})dx_1\cdots dx_{N-1}$  and

1292: (\ref{eq:(b-a)}) to reduce the problem to $N-1$.

1293: \end{proof}

1294: 

1295: 

1296: The condition

1297: 

1298: \[

1299: \sum_{n=1}^{\infty}\left|\frac{a(k+n)}{(g(k+n)-g(k))^{m}}\right|=o(a(k))\]

1300: is satisfied, for instance, if

1301: 

1302: (1) $\exists c>0$ so that $c<|a(k)|<c^{-1}$, or $|a(k)|$

1303: decreases to 0. (Note that $g(k+n)-g(k)=g'(k+tn)n$, where

1304: $0\leqslant t\leqslant1$, and $g'(k)\rightarrow\infty$ as

1305: $k\rightarrow\infty$.)

1306: 

1307: (2) $\exists c,r$ so that $c<a(k)<k^{r}$. $g'(k)\geqslant

1308: k^{\varepsilon}$ for some small $\varepsilon>0$.

1309: \section{Proof of Theorem~\ref{T1}}\label{S5}

1310: In Appendix \S\ref{Fi} we list some known facts about iterations of  maps.

1311: 

1312: \begin{proof}[Proof of B\"otcher's theorem, for (\ref{eq:eqG})]

1313:   ({\bf Note:} this line of proof extends to general analytic maps.)

1314: 

1315:  We

1316: write $\psi=\lambda z+\lambda^2zg(z)$ and obtain

1317: \begin{equation}

1318:   \label{eq:H}

1319:   g(z)-\frac{1}{2}g(z^2)=\frac{1}{2}z+\frac{1}{2}\lambda\left[g(z)(z-g(z))+g(z^2)\right]+\frac{\lambda^2 z}{2}g(z)g(z^2)=N(g)

1320: \end{equation}

1321: Let $\mathcal{A_{\lambda}}$ denote the functions analytic

1322: in the polydisk $\mathbb{P}_{1,\epsilon}=\mathbb{D}\times

1323: \{\lambda:|\lambda|<\epsilon\}$. We write (\ref{eq:H}) in the form (see \ref{eq:defT}))

1324: \begin{equation}

1325:   \label{eq:eqH4}

1326:   g=2\mathfrak TN(g)

1327: \end{equation}

1328: This equation is manifestly contractive in the sup norm, in a ball of

1329: radius slightly larger than $1/2$ in $\mathcal{A_{\lambda}}$, if

1330: $\epsilon$ is small enough.  For $\lambda\ne 0$, evidently

1331: $\psi=\phi^{-1}$ is also analytic at zero. \end{proof}

1332: \begin{Lemma}

1333:   $\psi$ is analytic in

1334: $\mathbb{D}_1$ for all $\lambda$ with $|\lambda|<1$.

1335: \end{Lemma}

1336: \begin{proof} We have

1337: \begin{equation}

1338:   \label{eq:eqGM}

1339:   \psi(z)=\frac{\lambda}{2}\left( X+\sqrt{X^2+4X/\lambda}\right)=:F(X);\ X=\psi(z^2)

1340: \end{equation}

1341: For small $z\ne 0$, $\psi(z)=O(z)$ and thus $F(\psi(z^2))$ is well defined and analytic. Note that (\ref{eq:eqGM}) provides analytic

1342: continuation

1343: of $\psi$ from $\mathbb{D}_{\rho^2}$ to  $\mathbb{D}_{{\rho}}$, provided nowhere

1344: in $\mathbb{D}_{\rho^2}$ do we have $\psi=-4/\lambda$  (certainly the case if $\rho$ is small). We assume, to get a contradiction, that there is a $z_0$,

1345: $|z_0|=\lambda_0<1$  so that $\psi(z_0)=-4/\lambda$, and we choose the least

1346: $\lambda_0$ with this property. By the previous discussion, $\psi$

1347: is analytic in the open disk $\mathbb{D}_{\sqrt{\lambda_0}}$.

1348:  Then we use the ``backward'' iteration  $\psi(z^2)=

1349: \lambda^{-1}\psi(z)^2/(1+\psi(z))$ to calculate $\psi(z^2)$

1350: from $\psi(z)$,  starting with $z=z_0$. This is in fact

1351: equivalent to  (\ref{eq:infty1}); after the substitution $x=(-\lambda y)^{-1}$

1352: we return to (\ref{eq:logist1}), with $x_0=\lambda/4$. Using (vi) and

1353: (vii) of \S\ref{Fi}, it follows that $1/x_n\not\to 0$, that is,  $\psi(z_0^{2^n})\not \to 0$. This impossible, since $\psi$ is analytic and $\psi(0)=0$.

1354: \end{proof}

1355: 

1356: \begin{proof}[Proof of Theorem~\ref{T1}, (i)]

1357: We return to

1358: (\ref{eq:eqG}). Taking $a\in (0,1)$ $m_n=\sup \{|\psi(z)|:

1359: |z|<a^{1/2^n}$ we note that

1360: \begin{equation}

1361:   \label{eq:eqm}

1362:   m_{n+1}\le \frac{1}{2} |\lambda|(m_n+\sqrt{m_n^2+4 m_n/|\lambda|})

1363: \end{equation}

1364: The sequence of $m_n$ is bounded by the sequence of  $M_n$, defined  by replacing

1365: ``$\le$'' with ``$=$'' in (\ref{eq:eqm}).  Since $\frac{1}{2}

1366: |\lambda|(x+\sqrt{x^2+4 x/|\lambda|})<x$ if

1367: $x>A:=|\lambda|/(1-|\lambda|)$, we have $\limsup_n M_n\le A$. By the

1368: maximum principle, $|\psi(\lambda,z)|< A$ in

1369: $\mathbb{D}\times\mathbb{D}$. Thus, by Cauchy's formula in $\lambda$

1370: we have $|\psi_n(z)|\le A$ for all $n$ and $z\in \mathbb{D}$. The

1371: radius of convergence of (\ref{eq:formG}) in $\lambda$ is at least

1372: one.  By \S\ref{Fi},  (vii), the radius of convergence is exactly

1373: one. 

1374: 

1375: Indeed, note first that  (a)  if $\psi$ is analytic in

1376: $\mathbb{D}$ then $\psi'\ne 0 $ in $\mathbb{D}$, otherwise 

1377: $\psi'(z_1)=0$ would imply  $\psi'(z_1^{2^n})=0$ in contradiction

1378: with $\psi'(0)=\lambda$. This means that if there is a $z_0$, $\psi(z_0)=-4/\lambda$, then $\sqrt{z_0}$ is a singular point of $\psi$. 

1379: 

1380: Secondly, any $\lambda$ of the form $1+i\epsilon$ with small

1381: $\epsilon$ correspond to $c=1/4+1/4 \epsilon^2,$ outside the

1382: Mandelbrot set.  Thus, in the iteration (\ref{eq:infty1}), the initial

1383: condition $y_0=-4/\lambda$ implies $y_n\to 0$. We can now use the

1384: implicit function theorem to suitably match $y_n$, once it is small

1385: enough, to some value of $\psi$ near zero.  Indeed, the equation

1386: $y_n=\lambda z_{0}^{2^n}+O(z_0^{2^{n+1}})$ has $2^n$ solutions. This

1387: means that for such a $z_0$, using (\ref{eq:eqGM}) to iterate

1388: backwards and to determine $\psi(z_0)$ (noting the parallel to

1389: (\ref{eq:infty1})), we have $\psi(z_0)=-4/\lambda$, and by (a) above,

1390: $\psi$ cannot be analytic in $z$ in $\mathbb{D}$.

1391: 

1392: Formula (\ref{eq:frakt}) follows by straightforward expansion of (\ref{eq:eqG})

1393: and identification of powers of $\lambda$.\end{proof}

1394: \begin{proof}[Proof of Theorem~\ref{T1}, (ii)]

1395: The stated type of lacunarity  of $\psi_k$ follows from (\ref{eq:frakt}) by induction,

1396: noting the discrete convolution structure in $k$.\end{proof}

1397: \begin{proof}[Proof of Theorem~\ref{T1}, (iii)]

1398: Continuity of $\psi_k$ in $\overline{\mathbb{D}}$ also follows by induction from (\ref{eq:frakt}) and the

1399: properties of $\mathfrak{T}$. By  dominated convergence (applied to the

1400: discrete measure $|\lambda|^n$),  for $\lambda<1$, $\psi$ is

1401: continuous in $\overline{\mathbb{D}}$ and the Fourier series converges

1402: pointwise in $\partial{\mathbb{D}}$.

1403: 

1404: To show convergence of the Fourier series of $H$ we only need to show

1405: $\inf_{\mathbb{D}}|\psi|>0$. Now, $\psi$ clearly cannot

1406: vanish for any $z_0=\mathbb{D}$, otherwise $\psi(z_0^{2^n})=0$, would imply

1407: by analyticity $\psi\equiv 0$. If

1408: $\min_{\mathbb D_{\rho}}|\psi(\rho)|=\epsilon$ would be small enough, then

1409: $\min_{\mathbb D_{\rho^2}}|\psi(\rho)|\le O(\epsilon^2)\ll

1410: \epsilon$, contradicting the maximum principle for $z/\psi(z)$.

1411: 

1412: 

1413: The rest of the proof is straightforward

1414: calculation, using the analyticity of $\psi$. \end{proof} The extension of the small

1415: $\lambda$ analysis  to higher

1416: order polynomials is also straightforward.

1417: \begin{Note}\label{Explc} The transseries of the B\"otcher map at binary rational

1418: numbers can be calculated rather explicitly. This is beyond

1419: the scope here, and will be the subject of a different paper. A less explicit

1420: expression has been obtained in \cite{CPAM}. We note that the constant

1421:     $\log_2(2\pi)$ in (1.17) of \cite{CPAM} should be $(2\pi)/\log

1422:     2$.

1423:   \end{Note}

1424: \section{Appendix}\label{appendix}

1425: \subsection{Proof of Lemma~\ref{linv}}

1426:  For every $m$, we write $g(s)=g_m+o(s^{-m})$ where $g_m$

1427: is a finite sum, an initial sum in the asymptotic series

1428: of $g$. It is straightforward to show that $g_m^{-1}(y)$ has

1429: an asymptotic power series as $y\to\infty$. Then, in the equation

1430: $g(s)=y$ we write $s=s_m+\epsilon$ where $g_m(s_m)=y$. Then,

1431: $y=g(s)=g(s_m+\epsilon)=g_m(s_m)+g'(\xi)\epsilon+(g(s_m)-g_m(s_m))$

1432: implies $\epsilon=(g_m(s_m)-g(s_m))/g'(\xi)=o(s^{-m-\theta})$ where  $g'(\xi)\sim a x^{\theta}$. Now $\theta$ is fixed and $m$ is arbitrary, and then the result

1433: follows.

1434: \subsection{Proof of Lemma~\ref{eq:asympt1}}

1435:   We have

1436:   \begin{multline}

1437:   \label{eq:34}

1438: f-e^{-xg(0)}=-\sum_{k=1}^{\infty}k(e^{-xg(k+1)}-e^{-xg(k)})=x\int_0^{\infty} e^{-xg(s)}g'(s)\lfloor s \rfloor ds\\=x\int_{g(0)}^{\infty} e^{-xu}\lfloor g^{-1}(u)\rfloor du=x\int_{g(0)}^{\infty} e^{-xu}  g^{-1}(u)du-

1439: x\int_{g(0)}^{\infty} e^{-xu} \{g^{-1}(u)\}  du

1440: \end{multline}

1441: and we also have

1442: \begin{equation}

1443:   \label{eq:ginv}

1444:   x\int_{g(0)}^{\infty} e^{-xu}  g^{-1}(u)du=-\int_{0}^{\infty} (e^{-xg(u)})'u du=

1445: \int_{0}^{\infty} e^{-xg(u)}du

1446: \end{equation}

1447: whereas

1448: $$0<\int_0^{\infty} e^{-xu} xg'(u)\{u\}du\le \int_0^{\infty} e^{-xu} xg'(u)du=e^{xg(0)}$$

1449: \subsection{Proof of Lemma \ref{Lgeom}}\label{pfLgeom}

1450: By the Fourier coefficients formula we have

1451: \begin{multline}

1452: c_{0}=\int_{0}^{1}f(a^{y})dy+\int_{0}^{1}ydy-\int_{0}^{1}\check{G}(a^{y})dy\\

1453: c_{k}=\int_{0}^{1}f(a^{y})e^{-2k\pi iy}dy+\int_{0}^{1}ye^{-2k\pi iy}dy-\int_{0}^{1}\check{G}(a^{y})e^{-2k\pi iy}dy\\

1454: =\frac{i}{2k\pi}+\sum_{n=0}^{\infty}\int_{0}^{1}e^{-a^{n+y}-2k\pi iy}dy+\sum_{n=1}^{\infty}\int_{0}^{1}\frac{(-1)^{n}a^{ny}}{n!(a^{n}-1)}e^{-2k\pi iy}dy\\

1455:  =\frac{i}{2k\pi}+\frac{1}{\log a}\sum_{n=0}^{\infty}a^{\frac{2kn\pi i}{\log a}}\(\Gamma\(-\frac{2k\pi i}{\log a},a^{n}\)-\Gamma\(-\frac{2k\pi i}{\log a},a^{1+n}\)\)\\+\sum_{n=1}^{\infty}\frac{(-1)^{n+1}}{n!(2k\pi i-n\log a)} =\sum_{n=0}^{\infty}\frac{(-1)^{n+1}}{n!(2k\pi i-n\log a)}+\frac{1}{\log a}\Gamma\(-\frac{2k\pi i}{\log a},1\)\:(k\neq0)

1456: \end{multline}

1457: 

1458: 

1459: Note that since $\mathcal{L}^{-1}(\frac{1}{2k\pi i-n\log a})=\dfrac{1}{\log a}e^{\frac{2k\pi ip}{\log a}}\:(n\rightarrow p)$

1460: we have

1461: \begin{multline*}

1462: \sum_{n=0}^{\infty}\frac{(-1)^{n+1}}{n!(2k\pi i-n\log

1463: a)}=\int_{0}^{\infty}\sum_{n=0}^{\infty}\frac{(-1)^{n+1}e^{-np}}{n!}\dfrac{1}{\log

1464: a}e^{\frac{2k\pi ip}{\log a}}dp\\=\dfrac{1}{\log

1465: a}\int_{0}^{\infty}e^{-e^{-p}}e^{\frac{2k\pi ip}{\log

1466: a}}dp\\=\dfrac{1}{\log a}\int_{0}^{1}e^{-t}t^{\frac{-2k\pi i}{\log

1467: a}-1}dt=\frac{1}{\log a}\(\Gamma(-\frac{2k\pi i}{\log

1468: a})-\Gamma(-\frac{2k\pi i}{\log a},1)\)

1469: \end{multline*}

1470: 

1471: 

1472: The above procedure is justified for $k$ in the upper half plane.

1473: By analytic continuation the expression holds for $k$ real as well.

1474: Eq. (\ref{eq:ck}) follows.

1475: 

1476: 

1477: 

1478: 

1479: We can further resum the series in the above expression by noting

1480: that

1481: \begin{multline}

1482:    \sum_{k\neq0}\Gamma\(-\frac{2k\pi i}{\log a}\)

1483: z^{\frac{2k\pi i}{\log a}}=\sum_{k=1}^{\infty}

1484: \frac{-\log a}{2k\pi i}\int_{0}^{\infty}

1485: t^{\frac{-2k\pi i}{\log a}}e^{-t}z^{\frac{2k\pi i}{\log a}}dt

1486: \\+\sum_{k=-\infty}^{-1}\frac{-\log a}{2k\pi i}=\int_{0}^{\infty}\sum_{k=1}^{\infty}\frac{-\log a}{2k\pi i}

1487: \(\frac{z}{t}\)^{\frac{2k\pi i}{\log a}}e^{-t}dt

1488: \int_{0}^{\infty}t^{\frac{-2k\pi i}{\log a}}e^{-t}z^{\frac{2k\pi i}{\log a}}dt

1489: \\+

1490: \int_{0}^{\infty}\sum_{k=-\infty}^{-1}\frac{-\log a}{2k\pi i}

1491: \(\frac{z}{t}\)^{\frac{2k\pi i}{\log a}}e^{-t}dt=\frac{\log a}{2\pi i}\int_{0}^{\infty}

1492: \log_{\RR}\(1-\(\frac{z}{t}\)^{\frac{2\pi i}{\log a}}\)e^{-t}dt

1493: \\-

1494: \frac{\log a}{2\pi i}\int_{0}^{\infty}

1495: \log_{\RR}\(1-\(\frac{z}{t}\)^{-\frac{2\pi i}{\log a}}\)e^{-t}dt=\frac{\log a}{2\pi i}\int_{0}^{\infty}

1496: \log_{\RR}\(-\(\frac{z}{t}\)^{\frac{2\pi i}{\log a}}\)e^{-t}dt

1497: \\

1498: =-\frac{\log a z}{2\pi i}\int_{0}^{\infty}\log_{\RR}\(-s^{\frac{2\pi

1499: i}{\log a}}\)e^{-sz}ds

1500: \end{multline}

1501: which can be justified by analytic continuation, for the last

1502: expression and the sum are both analytic, and equal to each other on

1503: the real line. The logarithm $\log_{\RR}$ is defined with a branch cut along $\RR^-$.

1504: 

1505: We finally obtain the integral representation (\ref{eq:resum}),

1506: valid for $z$ in the right half plane.

1507: 

1508: \begin{Remark}

1509: Since $\log_{\RR}\(-s^{\frac{2\pi i}{\log a}}\)=i\arg\(\dfrac{2\pi \log

1510: s}{\log a}-\pi\)=2\pi i \(\left\{\dfrac{\log s}{\log a}\right\}-\dfrac{1}{2}\)$,

1511: we actually recover the last term of (\ref{eq:asympt1}).

1512: \end{Remark}

1513: 

1514: 

1515: \subsection{Direct calculations for $b\in\NN$ integer; the cases $b=3,b=3/2$}\label{Direct,3}

1516: \begin{Proposition}

1517: If $\bb$ is an integer, the behavior of \[

1518: f(\delta+2\pi i \beta)=\sum_{k=1}^{\infty}e^{-k^{\bb}(\delta+2\pi i\beta)}\:(\Re(\delta)>0)\]

1519: where $\beta=2\pi i m/n$, m and n being integers, as $\delta$ approaches

1520: 0 is

1521: \[

1522: f(\delta+2\pi i \beta)=\left[\frac{1}{n}\sum_{l=1}^{n}e^{-2\pi

1523: i\dfrac{m}{n}l^{\bb}}\Gamma\(1+\dfrac{1}{\bb}\)\right]\delta^{-\frac{1}{\bb}}+O(1)\]

1524: Therefore $f(\delta)$ either blows up like $\delta^{-\frac{1}{\bb}}$ or is

1525: bounded.

1526: 

1527: The more general case $\bb=\dfrac{r+1}{r}$ where $r$ is an integer

1528: can be treated similarly. In particular, if

1529: $\bb=\dfrac{3}{2}$, for $\beta=-\dfrac{4\pi

1530:   i}{3\sqrt{3}}\sqrt{\dfrac{n}{m}}$(this includes all rational

1531: numbers), we have

1532:  \[

1533: \sum_{k=1}^{\infty}e^{-k^{3/2}z}=\frac{\sqrt{6\pi

1534: i}}{3^{\frac{7}{4}}m^{\frac{1}{4}}n^{\frac{3}{4}}}\sum_{l=1}^{n}e^{-2\pi

1535: i\dfrac{m}{n}l^{3}}\frac{1}{\sqrt{\delta}}+o\left(\frac{1}{\sqrt{\delta}}\right)\]

1536: with $z=-\dfrac{4\pi

1537: i}{3\sqrt{3}}\sqrt{\dfrac{n}{m}}$+$\delta$($\delta\to 0^+$)

1538: \end{Proposition}

1539: 

1540: \begin{proof}

1541: 

1542: 

1543: In general, to find the asymptotic behavior of $f(z)$ we analyze

1544: the functions

1545: \begin{equation}

1546:   \label{eq:deffk}

1547:  f_{k}(z)=\int_{0}^{\infty}e^{-pz-2k\pi

1548: ip^{\frac{1}{\bb}}}dp,\ k\in\ZZ

1549: \end{equation}

1550: Letting $p=q\(\dfrac{z}{k}\)^{\frac{\bb}{1-\bb}}$ we have

1551: 

1552: \[

1553: \int_{0}^{\infty}e^{-pz-2k\pi

1554: ip^{\frac{1}{\bb}}}dp=\(\frac{k}{z}\)^{\frac{\bb}{\bb-1}}\int_{0}^{\infty}e^{-(q+2\pi

1555: iq^{\frac{1}{\bb}})(\frac{k^{\bb}}{z})^{\frac{1}{\bb-1}}}dq\]

1556: 

1557: 

1558: Next we let $s=h(q)=q+2\pi iq^{\frac{1}{\bb}}$and

1559: $f_{k}(z)=(\frac{k}{z})^{\frac{\bb}{\bb-1}}\intop_{C_1}e^{-s(\frac{k^{\bb}}{z})^{\frac{1}{\bb-1}}}\ds\frac{1}{h'(h^{-1}(s))}ds$

1560: where the contour $C_1$ is a curve from the origin to $\infty$ in

1561: the first quadrant.

1562: 

1563: We can find the asymptotic behavior of $f_{k}(z)$ using Watson's

1564: Lemma \cite{benderorszag}. By iterating the contractive map

1565: $q\rightarrow\(\dfrac{s-q}{2\pi i}\)^{\bb}$ near 0, we can easily

1566: see that $\dfrac{h^{-1}(s)}{s^{\bb}}$ is analytic in $s^{\bb-1}$,

1567: which implies $\ds\frac{1}{h'(h^{-1}(s))}$ is analytic in

1568: $s^{\bb-1}$ near 0 with no constant term.

1569: 

1570: 

1571: 

1572: 

1573: 

1574: 

1575: Now let's consider the examples $\bb=3$ and $b=3/2$; for  $\bb=3$ we

1576: have $s=h(q)=q+2\pi iq^{\frac{1}{3}}$ and

1577: $\dfrac{ds}{dq}=$$\dfrac{3i}{8\pi^{3}}s^{2}+\dfrac{15}{64\pi^{6}}s^{4}-\dfrac{21i}{128\pi^{9}}s^{6}\cdots$. Thus, the asymptotic power series is

1578: 

1579: \[

1580: f(z)\sim\Gamma\(\frac{4}{3}\)z^{-\frac{1}{3}}-\dfrac{1}{2}-\dfrac{z}{120}+\frac{z^{3}}{792}+\cdots\]

1581: 

1582: 

1583: The branch point of $h^{-1}$  is located at

1584: $s_{0}=\pi^{\frac{3}{2}}\dfrac{4\sqrt{2}}{3\sqrt{3}}(-1)^{\frac{1}{4}}$,

1585: which is between the contour $C_1$ defined above and the $x$-axis. As we start

1586: rotating $z$ from $z>0$, we have, cf (\ref{eq:deffk}),

1587: \begin{multline}

1588:   f_{k}(z)=\(\frac{k}{z}\)^{\frac{3}{2}}\intop_{C_1}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s))}ds\\

1589: =\(\frac{k}{z}\)^{\frac{3}{2}}\intop_{0}^{\infty}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s))}ds+\(\frac{k}{z}\)^{\frac{3}{2}}e^{-s_{0}(\frac{k^{3}}{z})^{\frac{1}{2}}}\intop_{C_2}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s+s_{0}))}ds\\

1590: =\(\frac{k}{z}\)^{\frac{3}{2}}\int_{0}^{\infty}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s))}ds+2\(\frac{k}{z}\)^{\frac{3}{2}}e^{-s_{0}(\frac{k^{3}}{z})^{\frac{1}{2}}}\int_{0}^{\infty}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s+s_{0}))}ds

1591: \end{multline}

1592: where the contour $C_2$ starts at $\infty$, goes clockwise around

1593: the origin, then ends at $\infty.$ Since now

1594: 

1595: $$\dfrac{ds}{dq}=\dfrac{(-\pi

1596: i)^{\frac{3}{4}}}{6^{\frac{1}{4}}}(s-s_{0})^{-\frac{1}{2}}+\dfrac{5}{6}+\dfrac{5}{16}\(\dfrac{i}{6\pi}\)^{\frac{3}{4}}(s-s_{0})^{\frac{1}{2}}+\cdots$$

1597: 

1598: We have

1599: \begin{multline}

1600:   \(\frac{k}{z}\)^{\frac{3}{2}}e^{-s_{0}z}\intop_{C_2}\frac{e^{-s(\frac{k^{3}}{z})^{\frac{1}{2}}}}{h'(h^{-1}(s+s_{0}))}ds\\=e^{-\pi^{\frac{3}{2}}\frac{4\sqrt{2}}{3\sqrt{3}}(-1)^{\frac{1}{4}}z}\(\(\dfrac{\pi

1601: i}{6}\)^{\frac{1}{4}}k^{-\frac{1}{4}}z^{-\frac{1}{4}}+\dfrac{5}{32}\dfrac{i^{\frac{7}{4}}}{6^{\frac{3}{4}}\pi^{\frac{5}{4}}}k^{-\frac{7}{4}}z^{\frac{1}{4}}+\cdots\)

1602: \end{multline}

1603: 

1604: 

1605: Therefore the transseries is

1606: \begin{multline}

1607:   \label{eq:trans1}

1608:   \tilde{f}(z)=\Gamma(\frac{4}{3})z^{-\frac{1}{3}}-\dfrac{1}{2}-\dfrac{z}{120}+\frac{z^{3}}{792}+\cdots\\

1609: +\sum_{k=1}^{\infty}e^{-\pi^{\frac{3}{2}}\frac{4\sqrt{2}}{3\sqrt{3}}(-1)^{\frac{1}{4}}(\frac{k^{3}}{z})^{\frac{1}{2}}}\left[\(\dfrac{\pi

1610: i}{6}\)^{\frac{1}{4}}k^{-\frac{1}{4}}z^{-\frac{1}{4}}+\dfrac{5}{32}\dfrac{i^{\frac{7}{4}}}{6^{\frac{3}{4}}\pi^{\frac{5}{4}}}k^{-\frac{7}{4}}z^{\frac{1}{4}}+\cdots\right]\\

1611: -\sum_{k=1}^{\infty}e^{-\pi^{\frac{3}{2}}\frac{4\sqrt{2}}{3\sqrt{3}}(-1)^{\frac{1}{4}}(\frac{-k^{3}}{z})^{\frac{1}{2}}}\left[\(\dfrac{\pi

1612: i}{6}\)^{\frac{1}{4}}(-k)^{-\frac{1}{4}}z^{-\frac{1}{4}}+\dfrac{5}{32}\dfrac{i^{\frac{7}{4}}}{6^{\frac{3}{4}}\pi^{\frac{5}{4}}}(-k)^{\frac{7}{4}}z^{\frac{1}{4}}+\cdots\right]

1613: \end{multline}

1614: The calculation for $\bb=\dfrac{3}{2}$ is similar: in this case

1615: $\dfrac{ds}{dq}=\dfrac{3}{4\sqrt{2}(\pi

1616: i)^{\frac{3}{2}}}s^{\frac{1}{2}}-\dfrac{3i}{8\pi^{3}}s-\dfrac{105i^{\frac{3}{2}}}{256\sqrt{2}\pi^{\frac{9}{2}}}s^{\frac{3}{2}}+\cdots$

1617: and the asymptotic power series is

1618: 

1619: \[

1620: f(z)\sim\Gamma\(\frac{5}{3}\)z^{-\frac{2}{3}}-\dfrac{1}{2}-\dfrac{3\zeta(\frac{5}{2})}{16\pi^{2}}z+\frac{1}{240}z^{2}+\frac{315\zeta(\frac{11}{2})}{2048\pi^{5}}z^{3}+\cdots\]

1621: 

1622: 

1623: The exponential sum is slightly different than in the previous case, for

1624: now the branch point $s_{0}=-\dfrac{32}{27}i\pi^{3}$ lies in the

1625: lower half plane, which means the contour $C_2$ can be deformed

1626: to $[0,$$+\infty)$ without passing through any singularity.

1627: 

1628: We collect the contribution from the branch point only when $\arg z$

1629: decreases to $-\dfrac{\pi}{4}$from 0. Since

1630: $$\dfrac{ds}{dq}=\dfrac{-4\sqrt{2}i^{\frac{1}{2}}\pi^{\frac{3}{2}}}{3}(s-s_{0})^{-\frac{1}{2}}+\dfrac{4}{3}+\dfrac{i^{-\frac{1}{2}}}{8\sqrt{2}\pi^{\frac{3}{4}}}(s-s_{0})^{\frac{1}{2}}+\cdots$$

1631: we have for the exponential part of the sum

1632: $$\(\dfrac{4\sqrt{2}i^{-\frac{1}{2}}\pi}{3}k^{\frac{1}{2}}z^{-1}+\dfrac{i^{\frac{1}{2}}}{16\sqrt{2}\pi^{\frac{5}{4}}}k^{-\frac{5}{2}}z^{2}+\cdots\)\exp\({\dfrac{32i\pi^{3}k^{3}}{27z^{2}}}\)$$

1633: Therefore, for small $z$ in the right half plane, the transseries  is given by

1634: \begin{multline}

1635:   \label{Tr3/2}

1636: \tilde{f}(z)=\Gamma\(\frac{5}{3}\)z^{-\frac{2}{3}}-\dfrac{1}{2}-\dfrac{3\zeta(\frac{5}{2})}{16\pi^{2}}z+\frac{1}{240}z^{2}+\frac{315\zeta(\frac{11}{2})}{2048\pi^{5}}z^{3}+\cdots\\+\begin{cases}

1637: \sum\limits _{k=1}^{\infty}e^{\frac{32i\pi^{3}k^{3}}{27z^{2}}}\(\frac{4\sqrt{2}i^{-\frac{1}{2}}\pi}{3}k^{\frac{1}{2}}z^{-1}+\frac{i^{\frac{1}{2}}}{16\sqrt{2}\pi^{\frac{5}{4}}}k^{-\frac{5}{2}}z^{2}+\cdots\); \,\,-\frac{\pi}{2}\leqslant\arg z\leqslant-\frac{\pi}{4}\\

1638: -\sum\limits

1639: _{k=1}^{\infty}e^{-\frac{32i\pi^{3}k^{3}}{27z^{2}}}\(\frac{4\sqrt{2}i^{-\frac{1}{2}}\pi}{3}(-k)^{\frac{1}{2}}z^{-1}+\frac{i^{\frac{1}{2}}}{16\sqrt{2}\pi^{\frac{5}{4}}}(-k)^{-\frac{5}{2}}z^{2}+\cdots\);\,\frac{\pi}{4}\leqslant\arg

1640: z\leqslant\frac{\pi}{2}\end{cases}

1641: \end{multline}

1642: The effect of the exponential part of the transseries affects the leading

1643: order when

1644: $z\rightarrow0$ nearly tangentially to the imaginary line.

1645: 

1646: For example, to see the effect of the first exponential term for

1647: $\bb=\dfrac{3}{2}$, we let $z^{-\frac{2}{3}}\sim

1648: e^{\frac{32i\pi^{3}}{27z^{2}}}z^{-1}$ and $z=-ire^{i\theta}$ near

1649: the negative imaginary line.

1650: 

1651: The critical curve along which the power term and the exponential

1652: term are of equal order is

1653: $\theta\sim\dfrac{9}{64\pi^{3}}r^{2}\log\dfrac{1}{r}$.

1654: \end{proof}

1655: \begin{figure}[h!]

1656:    $ $ \hskip -3cm

1657:   \includegraphics[scale=0.38]{M}

1658:   \caption{The Madelbrot set (drawn with xaos 3.1 \cite{xaos}).} \label{M1}

1659:   \label{fig:12}

1660: \end{figure}

1661: \subsection{Notes about iterations of maps}\label{Fi}

1662: 

1663: For the following, see e.g., \cite{Beardon,Devaney,Milnor}.

1664:     \begin{enumerate}

1665:    \item   For $|\lambda|<1$, three types of behavior are possible for

1666: the solution of (\ref{eq:logist1}): if the initial condition $x_0\in \mathcal{F}$, the connected component

1667: of the origin in {\em the Fatou set}, then $x_n\to 0$ as $n\to \infty$.

1668: Clearly,  $x_0\in \mathcal{F}$ if $x_0$ is small enough. If $x_0\in \overline{\mathcal{F}}^c$, the connected

1669: component of infinity in the Fatou set, then

1670: $|x_n|\to\infty$. Clearly, $x_0\in \overline{\mathcal{F}}^c$ if it is large enough. Finally, for   $x_0\in \partial{F}=J$, the Julia set, a (connected) curve

1671: of  nontrivial Hausdorff dimension invariant under

1672: the map, $\{x_n\}_n$ are  dense in $J$ and the evolution is chaotic.

1673: 

1674: 

1675: \item $J$ is the closure of the set of repelling periodic points.

1676: 

1677: \item For polynomial maps, and more generally, for entire maps, $J$ is the boundary of the set of points which converge to infinity under iteration.

1678: 

1679: \item If the maximal disk of analyticity of $\psi$ is the unit disk $\mathbb{D}_1$, then $\psi$ maps $\mathbb{D}_1$ biholomorphically onto the immediate basin $\mathcal{A}_0$ of zero.  If on the contrary the maximal disk is $\mathbb{D}_r,\,r<1$, then there is at least one other critical point in $\mathcal{A}_0$, lying in $\psi(\partial\mathbb{D}_r)=J_y$, the Julia set of \ref{eq:infty1}.

1680: 

1681: \item If $r=1$, it follows that $\psi(\partial\mathbb{D}_1)=J_y$.

1682: 

1683: 

1684: \item By the change

1685: of variable $x_n=-(1/\lambda)t_n+1/2$, (\ref{eq:logist1}) is brought to the ``$c$ form''

1686: $t_{n+1}=t_n^2+c$, $c=\lambda/2-\lambda^2/4$. The Mandelbrot set is defined

1687: as (see e.g. \cite{Devaney})

1688: \begin{equation}

1689:   \label{eq:defM}

1690:   \mathcal{M}=\{c: t_n \text{ bounded }\text{ if } t_0=0\}

1691: \end{equation}

1692: If $c\in\mathcal{M}$, then clearly $y_n$ in  (\ref{eq:infty1}) are bounded away from zero. Note that $t_0=0$ corresponds to $x_0=1/2$ implying $x_1=-\lambda/4$.

1693: \item \label{(iii)}  $\mathcal{M}$ is a compact set; it coincides  

1694: with the set of $c$ for which $J$ is connected. The

1695: cardioid $\mathcal{H}=\{(2e^{it}-e^{2it})/4:t\in [0,2\pi)\}$ is contained in

1696: $\mathcal{M}$; see \cite{Devaney}. This means $\{\lambda:|\lambda|<1\}$

1697: corresponds to the interior of $\mathcal{M}$. We have

1698:  $|\lambda|=1\Rightarrow c\in \partial \mathcal{M}\subset\mathcal{M}$.

1699: 

1700: \end{enumerate}

1701: 

1702: 

1703: \subsection{Overview of Borel summability and transseries}\label{A4}

1704: 

1705: There is a vast literature on transseries, Borel summability, and

1706: resurgence, see, for example \cite{Sauzin}. Most of the modern theory

1707: originates in Ecalle's work \cite{Ecalle}.

1708: \begin{Definition}\label{def1}

1709:   We say that $f$ is given by a Borel summable transseries for

1710:   $x>\nu$, if there exists a $\beta \in\CC$, a sequence $c_k$, with

1711:   $\Re\,c_k\ge Ck$ for some $C>0$, and a sequence of functions $Y_k$,

1712:   analytic in a neighborhood of $(0,\infty)$, having convergent Puiseux

1713:   series at zero, and $|Y_k(p)|\le |B^ke^{\nu p}|$ (where $B$ and

1714:   $\nu$ are independent of $k$) such that

1715: \begin{equation}

1716:   \label{eq:deftr}

1717:   f(x)=\sum_{k=0}^{\infty}e^{-c_k x}x^{(k+1)\beta}\mathcal{L} Y_k

1718: \end{equation}

1719: where $\mathcal{L}$ is the usual Laplace transform:

1720: \begin{equation}

1721:   \label{eq:lapl}

1722:   (\mathcal{L}Y)(x)=\int_0^{\infty}e^{-px}Y(p)dp

1723: \end{equation}

1724: The definition for other directions $\theta$ in the $x$ complex domain is obtained

1725: by changing the variable to $x'=xe^{-i\theta}$.

1726: 

1727: The Borel-Laplace summation operator is denoted by $\mathcal{LB}$.

1728: \end{Definition}

1729: \begin{Definition}

1730:   A formal power series in powers of $1/x$ is Borel summable as

1731:   $x\to\infty$ if it is the asymptotic series of $\mathcal{L} Y$,

1732:   where $Y$ is as in Definition \ref{def1}. (We note that by Watson's

1733:   Lemma \cite{benderorszag}, $(\mathcal{L}Y)(x)$ has an asymptotic

1734:   series as $x\to\infty$, which is the termwise Laplace transform of

1735:   the Taylor series of $Y$ at zero.)

1736: \end{Definition}

1737: Transseries representations contain therefore manifest asymptotic information.

1738: \begin{Definition}

1739:   A function $Y(p)$ is resurgent in $p$ in the sense of Ecalle \cite{Ecalle}, if it

1740:   is analytic on the Riemann surface of $\CC\setminus J$, where $J$ is

1741:   a discrete set, and has uniform exponential bounds along any

1742:   direction towards infinity cf. \cite{Sauzin}.\footnote{It is also required

1743: that certain relations, called bridge equations, hold; in our

1744: case it would be easy to derive them from the

1745: explicit form of $H$, but we omit this calculation.} By abuse of language,

1746:   $f(x)$ is called resurgent if it satisfies the requirements in

1747:   Definition \ref{def1} and {\bf all} $Y_k$ are resurgent.

1748: \end{Definition}

1749: This is especially useful when global information about $f$ for

1750: $x\in\CC$ is needed: deformation of contours in $p$, and collecting

1751: residues when/if singularities are crossed, provides a straightforward

1752: way to obtain this information.

1753: 

1754: 

1755: 

1756: 

1757: \begin{thebibliography}{99}

1758: \bibitem{Abramowitz} M Abramowitz and I A Stegun, Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, New York: Dover (1965).

1759:  \bibitem{benderorszag} C Bender and S Orszag, {\em Advanced

1760: Mathematical Methods for scientists and engineers}, McGraw-Hill, 1978, Springer-Verlag 1999.

1761: 

1762:  \bibitem{Beardon} A F  Beardon, {\em Iteration of Rational Functions} Springer Verlag, New York (1991).

1763:  \bibitem{OCtransseries}O. Costin, Topological construction of transseries and introduction to generalized Borel summability (AMS CONM series vol. 373, 137--177)

1764:   \bibitem{CPAM} O. Costin and M. D. Kruskal, {\em Analytic methods

1765:     for obstruction to integrability in discrete dynamical systems}

1766:   Comm. Pure Appl. Math. 58 no. 6, 723--749) (2005).

1767: \bibitem{Inventiones} O.  {}Costin and R D Costin, Invent. Math,  145, 3, pp 425-485  (2001).

1768: 

1769: 

1770: \bibitem{Devaney} R L Devaney, An Introduction to Chaotic Dynamical Systems, 2nd Edition, Westview Press (2003).

1771: \bibitem{Douady}  A Douady and J  Hubbard {\em On the dynamics  of polynomial-like mappings},  Ann. Sci. Ec. Norm. Sup. 18 pp. 287–344 (1985).

1772: 

1773: 

1774: 

1775: 

1776: \bibitem{Ecalle}  J Ecalle, Les fonctions resurgentes, vol. I, II and III, Publ. Math. Orsay, 1985.

1777: 

1778: \bibitem{Jacobi}  D C G Iacobo. Fundamenta Nova Theoriae Functionum Ellipticarum.

1779: Konigsberg:, Sumtibus Fratrum Borntraeger, 1829.

1780: \bibitem{Ince}  E L Ince,  Ordinary Differential Equations, Dover Publications (1956).

1781: \bibitem{Kawai1} T. Kawai, {Amer. J. Math.}, 109, 57--64 (1989).

1782: \bibitem{Kawai} T. Kawai, in {\em Towards the Exact WKB Analysis of Differential

1783: Equations, Linear or Nonlinear}, C.J. Howls, T. Kawai and Y. Takei, eds., Kyoto

1784: Univerisity Press, Kyoto, pp. 231--244, (2000).

1785: 

1786: \bibitem{Mahler} K

1787: Mahler,

1788: {\em An unsolved problem on the powers of $3/2$}.

1789: J. Austral. Math. Soc. 8 1968 313--321.

1790: \bibitem{Mandelbrojt}  S. Mandelbrojt {\em S\'eries lacunaires},

1791:   Actualit\'es scientifiques et industrielles, Paris, 305 (1936).

1792: \bibitem{Milnor} J. Milnor {\em Dynamics in one complex variable}, Annals of Mathematics

1793: studies Nr. 160, Princeton (2006).

1794: \bibitem{Montgomery} H. L. Montgomery, {\em Ten lectures on the interface between

1795: Analytic Number Theory and Harmonic Analysis}, Providence, R.I. : Published for the Conference Board of the Mathematical Sciences by the American Mathematical Society, (1994).

1796: \bibitem{xaos} Project Xaos, Gnu freeware, http://sourceforge.net/projects/xaos.

1797: \bibitem{Sauzin} D. {}Sauzin {\em Resurgent functions and splitting problems}

1798: RIMS Kokyuroku 1493 (31/05/2006) 48-117, \begin{verbatim}{http://arxiv.org/abs/0706.0137v1.}\end{verbatim}

1799: \bibitem{Titchmarsh} E C  Titchmarsh, {\em The Theory of the Riemann Zeta-Function}, Oxford University Press, (1987).

1800: 

1801: 

1802: \end{thebibliography}

1803: \end{document}

1804: -(1/2)*E(Pi^(1/2)*gamma^(1/2)*(N-xN)*(-i)^(1/2))/(gamma^(1/2)*(-i)^(1/2))

1805: 

1806: 

1807: 

1808: 30B50, 30B10, 30B30,40G10,34E05, 37F10,37F50