1: \documentclass[makeidx]{amsart}
2: %\usepackage[active]{srcltx}
3: %\usepackage{makeidx}
4: %\renewcommand{\baselinestretch}{2}
5: \usepackage{amsmath}
6: \usepackage{amssymb}
7: \usepackage{epsfig}
8: \usepackage{hyperref}
9: \newtheorem{Proposition}{Proposition}[section]
10: \newtheorem{Remark}[Proposition]{Remark}
11: \newtheorem{Corollary}[Proposition]{Corollary}
12: \newtheorem{Lemma}[Proposition]{Lemma}
13: \newtheorem{Theo}[Proposition]{Theorem}
14: \newtheorem{Theorem}{Theorem}[section]
15: \newtheorem{Consequence}[Proposition]{Consequence}
16: \newtheorem{Definition}[Proposition]{Definition}
17: \newtheorem{Note}[Proposition]{Note}
18: \newtheorem{Convention}[Proposition]{Convention}
19: \newcommand {\z}{{\noindent}}
20: \def\blackslug{\hbox{\hskip 1pt \vrule width 4pt height 8pt depth 1.5pt
21: \hskip 1pt}}
22: \def\qed{\quad\blackslug\lower 8.5pt\null\par}
23: \def\HH{\mathbb{H}}
24: \def\CC{\mathbb{C}}
25: \def\RR{\mathbb{R}}
26: \def\NN{\mathbb{N}}
27: \def\ZZ{\mathbb{Z}}
28: \def\QQ{\mathbb{Q}}
29: \def\Re{\mathrm{Re}}
30: \def\Im{\mathrm{Im}}
31: \def\bchi{\mbox{\raisebox{.4ex}{\begin{large}$\chi$\end{large}}}}
32: \renewcommand{\theequation}{\thesection.\arabic{equation}}
33: \renewcommand{\thefootnote}{(\arabic{footnote})}
34: \makeindex
35: \date{ }
36: \begin{document}
37:
38:
39: \author{O. Costin$^1$ }\title{ Global
40: reconstruction of analytic functions from local expansions and a new general
41: method of converting sums into integrals}
42: \gdef\shortauthors{O. Costin} \gdef\shorttitle{Global analytic
43: reconstruction} \thanks{$1$. Department of Mathematics, Ohio State
44: University.}
45: \maketitle
46:
47:
48:
49:
50: \bigskip
51:
52:
53: \begin{abstract}
54: A new summation method is introduced to convert a relatively
55: wide family of local Taylor series and infinite sums into integrals.
56:
57: Global behavior such as analytic continuation, position of singularities,
58: asymptotics for large values of the variable and asymptotic location of
59: zeros thereby follow, through the integral representations, from the Taylor
60: coefficients at a point, say zero.
61:
62: The method can be viewed in some sense as the inverse of Cauchy's formula.
63:
64: It can work in one or several complex variables.
65:
66: There is a duality between the global analytic structure of the
67: reconstructed function and the properties of the coefficient as a function
68: of their index.
69:
70: Borel summability of a class of divergent series follow as a biproduct.
71:
72: \end{abstract}
73: \tableofcontents
74:
75: \setcounter{equation}{0}\section{Introduction}
76:
77: Finding the global behavior of an analytic function in terms of its Taylor
78: coefficients is a notoriously difficult problem.
79: In fact, there cannot exist a general solution to this problem, since
80: undecidable questions can be quite readily formulated in such terms.
81:
82: Obviously too, mere estimates on the coefficients do not provide enough
83: information for global description. But constructive, detailed knowledge of
84: the coefficients does.
85: Simple examples are
86: \begin{equation}
87: \label{eq:ln2}
88: f_1(z)=\sum_{n=1}^{\infty}\frac{z^n}{\sqrt{n}},\ \ f_2(z)=\sum_{n=1}^{\infty}\frac{z^n}{n^\pi+\ln
89: n},\ \ f_3(z)= \sum_{n=1}^\infty \frac{z^n}{n^{n+1}},\ \ f_4=\sum_{n=1}^{\infty}n^{n}z^n
90: \end{equation}
91: (Certainly, as presented, $f_4$ does not define a function; the question
92: for $f_4$ is whether the series is Borel summable.
93:
94: Coefficients with a different type of growth, say
95: $e^{\sqrt{n}}$
96: can be accomodated too.
97:
98: Using \'Ecalle-Borel (EB) summability, applicable to (trans)series defined in
99: a rather constructive fashion, briefly outlined in \S\ref{EB} and
100: \S\ref{Bsum}, integral formulas containing global information can be obtained
101: in considerably wider generality. \footnote{After developing these methods,
102: it has been brought to our attention that a duality between resurgent
103: functions and resurgent Taylor coefficients has been noted in an unpublished
104: manuscript by \'Ecalle.}.
105:
106: In particular it will follow from the results
107: below that
108: \begin{equation}
109: \label{eq:sum0}
110: f_1(z)= \frac{ z}{\sqrt{\pi}}\int_0^{\infty}\frac{dt }{(1+t)\sqrt{\ln(1+t)}(t-(z-1))}
111: \end{equation}
112: On the first Riemann sheet $f_1$ has only one singularity, at $z=1$, of
113: logarithmic type, and $f_1=O(z^{-1})$ for large $z$. General Riemann surface
114: information and monodromy follow straightforwardly (cf. (\ref{eq:mono1}). A
115: similar complex analytic structure is shared by $f_2$, which has one
116: singularity at $z=1$ where it is analytic in $\ln(1-z)$ and $(1-z)$; the
117: singularity structure is that of the function
118: \begin{equation}
119: \label{intlog}
120: \phi(z)= \oint_{0}^{\infty}\frac{e^{-u\ln(z-1)}}{(-u)^{\pi}+\ln (-u)}du
121: \end{equation}
122: where the notation $\oint_0^{\infty}$ is explained after (\ref{eq:defpv}) below. An
123: explicit formula for the singular structure can be obtained in all cases, and
124: their is a duality between the properties of the coefficients and the global
125: structure, for instance monodromy, of the reconstructed function.
126:
127: The function $f_3$ is entire; questions answered
128: regard say the behavior for large negative $z$ or the asymptotic location of
129: zeros. It will follow that $f_3$ can be written as
130: \begin{equation}
131: \label{eq:invl}
132: f_3(z)=e^{-1}\int_0^{\infty}(1+u)^{-1}G(\ln(1+u))\left[\exp\left({\frac{ze^{-1}}{1+u}}\right)-1\right]du
133: \end{equation}
134: where $G(p)=s'_2(1+p)-s'_1(1+p)$ and $s_{1,2}$ are two branches of the
135: functional inverse of $s-\ln s$, cf. \S\ref{entire}. Detailed behaviour for
136: large $z$ can be obtained from (\ref{eq:invl}) by standard asymptotics
137: methods; in particular, for large negative $z$, $f_3$ behaves like a constant
138: plus $z^{-1/2}e^{-z/e}$ times a factorially divergent series (whose terms can
139: be calculated).
140:
141:
142: The example $f_1$ is the classical polylog$(1/2,z)$ but $f_2, f_3$ or $f_4$
143: satisfy no obvious relation from which analytic control can be otherwise
144: gained. Yet they are particularly simple, in that the coefficients have an
145: explicit formula. Similar representations however can be obtained in the much
146: more general case when $f^{(k)}(0)$ are {\em analyzable functions} of $k$. This is the case for
147: many series occurring in linear differential or partial differential
148: equations, difference equations, combinatorics, number theory and other
149: fields. It is often convenient to work with a series given in terms of the
150: coefficients, even when an underlying generating problem exists \cite{OCSG,
151: OT}.
152:
153: A roughly analogous reconstruction was known in the context of nonlinear ODEs
154: for which information about location of singularities of solutions can be
155: found from their {\em transseries} representations \cite{Invent}.
156:
157:
158: As it will be clear from the proofs, the method and results would apply, with
159: minor adaptations to functions of several complex variables.
160:
161: \subsection{Evaluating series.}
162: There are many other questions amenable to this method. For instance,
163: we get that
164: \begin{equation}
165: \label{eq:limit1}
166: \lim_{z\to -1^+}\sum_{n=1}^{\infty}e^{\sqrt{n}}z^n=
167: -\frac{1}{4\sqrt{\pi}}\int_{C_1}\frac{e^{1/p}dp}{p^{-3/2}(e^p+1)}
168: \end{equation}
169: where $C_1$ starts along $\RR^+$, loops clockwise once around the origin
170: and ends up at $+\infty$. The sum would be unwieldy numerically,
171: while the integral can be evaluated accurately by standard means. Likewise, we get
172: \begin{equation}
173: \label{eq:eqsum}
174: \sum_{k=0}^{\infty}\frac{e^{i\sqrt{k}}}{k^{a}}=-\frac{\gamma
175: 2^{a-1/2}}{\sqrt{\pi}}
176: \int_Cdp \frac{e^{-\frac{1}{8p}}U(2a+1/2;\frac{1}{\sqrt{2p}})}{p^{a-1}(e^p+1)}
177: \end{equation}
178: for $a>1/2$ (when it is not hard to check that the sum is convergent, {\em
179: e.g.} by approximating the sum by an integral and estimating the remainder).
180: Here $C$ is a contour starting along $\RR^-$, circling the origin clockwise
181: and ending up at $+\infty$ and $U$ is the parabolic cylinder function
182: \cite{Abramowitz}. These sums are obtained in \S\ref{Growthrates}.
183:
184:
185:
186: \subsection{Global description from local expansions}$ $\\
187: $ $\ \ \ {\em A first class of problems} is finding the location and type of
188: singularities in $\CC$ and the behaviour for large values of the variable of
189: functions given by series with finite radius of convergence, such as those in
190: (\ref{eq:ln2}).
191:
192:
193: {\em The second class of problems} amenable to the techniques presented
194: concerns the behaviour at infinity (growth, decay, asymptotic location of
195: zeros etc.) of entire functions presented as Taylor series, such as $f_3$
196: above.
197:
198:
199: {\em The third type of class of problems} is to determine Borel summability of
200: series with zero radius of convergence such as
201: \begin{equation}
202: \label{eq:entire}
203: \tilde{f}_4=\sum_{n=0}^{\infty}n^{n+1} z^n
204: \end{equation}
205: in which
206: the coefficients of the series are analyzable.
207:
208:
209:
210: \subsection{Transseries and analyzable functions}\label{EB}
211: In the early 1980's, \'Ecalle discovered and extensively studied a broad class
212: of functions, analyzable functions, closed under algebraic operations,
213: composition, function inversion, differentiation, integration and solution of
214: suitably restricted differential equations
215: \cite{Ecalle-book,Ecalle,Ecalle2,Ecalle3}. They are described as generalized
216: sums of ``transseries'', the closure of power series under the same
217: operations. The latter objects are surprisingly easy to describe; roughly,
218: they are ordinal length, asymptotic expansions involving powers, iterated
219: exponentials and logs, with at most power-of-factorially growing coefficients.
220:
221: In view of the closure of analyzable functions to a wide class of operations,
222: reconstructing functions from series with arbitrary {\em analyzable}
223: coefficients would make the reconstruction likely applicable to series
224: occurring in problems involving any combination of these many operations.
225:
226: This paper deals with analyzable coefficients having finitely many
227: singularities after a suitable EB transform. The methods however
228: are open to substantial extension. In particular, we allow for general
229: singularities, while analyzable and resurgent functions have singularities
230: of a controlled type \cite{Ecalle}.
231:
232:
233:
234:
235:
236: \subsection{Classical and generalized Borel summation} \label{Bsum} A series
237: $\tilde{f}=\sum_{k=1}^{\infty}c_k x^{-k}$ is Borel summable if its Borel
238: transform, {\em i.e.} the formal
239: inverse Laplace transform \footnote{$\displaystyle \sum_{k=1}^{\infty}c_k \frac{1}{2\pi i}\int_{1-i\infty}^{1+i\infty} e^{px}
240: x^{-k}dx=\sum_{k=0}^{\infty}\frac{c_k p^{k-1}}{(k-1)!}=F(p)$}
241: converges to a function $F$ analytic in a neighborhood of $\RR^+$, and $F$
242: grows at most exponentially at infinity. The Laplace transform of $F$ is by
243: definition the Borel sum of $\tilde{f}$. Since Borel summation is formally
244: the identity, it is an extended isomorphism between functions and series,
245: much as convergent Taylor series associate to their sums.
246:
247: However expansions occurring in applications are often not {\em classically}
248: Borel summable, sometimes for the relatively manageable reason that the
249: expansions are not simple integer power series, or often, more seriously,
250: because $F$ is singular on $\RR^+$, as is the case of $\sum n!x^{-n-1}$
251: where $F=(1-p)^{-1}$, or
252: because $F$ grows superexponentially.
253:
254: To address the latter difficulties, \'Ecalle defined averaging and cohesive
255: continuation to replace analytic continuation, and acceleration to deal with
256: superexponential growth \cite{Ecalle-book,Ecalle,Ecalle2,Ecalle3}.
257:
258: We call \'Ecalle's technique \'Ecalle-Borel (EB) summability and ``EB
259: transform'' the inverse of EB summation. While it is an open, imprecisely
260: formulated, and in fact conceptually challenging question, whether EB summable
261: series are closed under all operations needed in analysis, general results
262: have been proved for ODEs, difference equations, PDEs, KAM resonant expansions
263: and other classes of problems \cite{Balser, Duke, Braaksma,Poincare,OCSG}. EB
264: summability seems for now quite general.
265:
266:
267: A function is analyzable if it is an EB transform of a transseries. This
268: transseries is then unique \cite{Ecalle}. Then the EB transform is the mapping
269: that associates this unique transseries to the function. Simple examples of such transseries are
270: \begin{equation}
271: \label{eq:dual}
272: f_k=\frac{1}{\sqrt{k}};\ \
273: f_k=\frac{1}{k}\sum_{j=1}^{\infty} (-1)^k\frac{\ln^j k}{k^{j\pi}};\ \
274: f_k=\sum_{j=1}^{\infty}\frac{(-z)^j j!}{k^j}+2^{-k-1}\sum_{j=0}^{\infty}\frac{(-z)^j\Gamma(j-1/2)}{\sqrt{\pi}}
275: \end{equation}
276: The first two are convergent and correspond to the first two series in
277: (\ref{eq:ln2}); the last one is divergent but Borel summable.
278:
279: EB summation consists, in the simplest cases, in replacing the series in the
280: transseries by their Borel sum. In a first stage one takes the Borel transform
281: (cf. \S\ref{Bsum}) of each component series. In (\ref{eq:dual}) the Borel
282: transforms are correspondingly,
283: \begin{equation}
284: \label{eq:EBT}
285: F(p)=\frac{1}{\sqrt{p\pi}}, \ \ F_1(p),\ \ \left[\frac{1}{1+p},\sqrt{p+1}\right]
286: \end{equation}
287: and $F_1$ given in (\ref{eq:dual2}). The last pair represents the separate EB
288: transforms of the two series in the last example in (\ref{eq:dual}).
289:
290: The EB sum of these transseries are, with $(\mathcal{L}F)(p)=
291: \int_0^{\infty}e^{-pk}F(p)dp$ the usual Laplace transform,
292: \begin{multline}
293: \label{eq:EBsum}
294: f_k=\mathcal{L}\frac{1}{\sqrt{p\pi}}=\frac{1}{\sqrt{k}};\
295: f_k=\mathcal{L}F_1;\ \
296: \mathcal{L}\frac{1}{1+p}+2^{-k-1}\mathcal{L}\sqrt{1+p}\\=
297: e^k\mathrm{Ei}(-k)+2^{-k-1}\left( \frac{1}{k}+
298: \frac{\sqrt{\pi}e^k\mathrm{erfc}(\sqrt{k})}{k^{3/2}}\right)
299: \end{multline}
300: \section{(i) Series with finite radius of convergence}
301: \z Let $\mathcal{M}$ be the functions analytic at zero, algebraically bounded
302: at infinity and which have finitely many, possibly branched, singularities in
303: $\CC$. We choose to make cuts from $z_j$ to infinity which do not
304: intersect.
305:
306: \begin{figure}
307: \epsfig{file=cuts, width=7cm,height=5cm}
308: \caption{Singularities of $f$, cuts, direction of integration and Cauchy
309: contour deformation.}
310: \end{figure}
311:
312: $\mathcal{M}$ would be a proper subclass of {\em resurgent} functions
313: \cite{Ecalle-book} if the singularities are of the form described in
314: Note~\ref{Resur} (b).
315:
316: \begin{Note}[Generalizations]
317:
318: \begin{enumerate}
319: $ $
320:
321:
322: {\rm
323: \item {It will be seen from the proofs it is sufficient
324: to have at most that algebraic growth in {\em some} sector, or with slight
325: modifications, at most exponential growth.}
326:
327: \item Also, one can allow for
328: infinitely many singularities, if some estimates for their strength is
329: available.
330: \'Ecalle averaging would allow for singularities on the line of
331: summation.
332:
333: \item Exponential growth of $F_j$ (see below) can also be accomodated,
334: provided a sufficient number of initial terms of the series are left out.
335:
336: \item Several complex variables can be treated very similarly, as it will
337: become clear.
338: \item{Other types of decay/growth of coeefficients can be accomodated, cf.
339: \S\ref{Growthrates}}
340:
341:
342: Some further generalizations are transparent, but to simplify the notation and
343: proofs we will not pursue them.}
344: \end{enumerate}
345: \end{Note}
346: \begin{Note}
347: By taking sufficiently many derivatives we can assume that
348: $f\in\mathcal{M}'=\{f\in \mathcal{M}:f=O(z^{-1-\epsilon})\ \text{as}\
349: z\to\infty\}$.
350: \end{Note}
351:
352:
353:
354: Consider a series with finite radius of convergence
355: \begin{equation}
356: \label{eq:fser}
357: f=\sum_{k=0}^{\infty}f_k z^k
358: \end{equation}
359: In the following, if $g$ is analytic on $\RR^+$ and has a singularity at zero
360: we denote by
361: \begin{equation}
362: \label{eq:defpv}
363: \oint_0^{\infty}g(s)ds
364: \end{equation}
365: the integral of $g$ around $\RR^+$, traversed towards $+\infty$ on the
366: upper side.
367:
368: While providing integral formulas, in terms of functions with known
369: singularities, which are often rather explicit, the following result can also
370: be interpreted as a {\em duality of resurgence}.
371:
372: \begin{Theorem}\label{T1}
373: (i) Assume that $f_k$ have Borel sum-like representations of the form
374: \begin{equation}
375: \label{eq:trs1}
376: f_k=\sum_{j=1}^N a_j^{-k}\oint_0^{\infty}e^{-kp}F_j(p)ds
377: \end{equation}
378: (\ref{eq:trs1}) where $F_j$ are analytic and bounded on $\RR^+$ and singular
379: at zero.
380:
381: Then, $f$ is given by
382: \begin{equation}
383: \label{eq:recon3}
384: f(z)=f(0)+ z\oint_0^{\infty}\sum_{j=1}^N \frac{F_j(\ln (1+s))ds}{(1+s)(s a_j+a_j-z)}
385: \end{equation}
386:
387: \z (ii) Furthermore, $f\in\mathcal{M}$ and the singularities, located at $a_j$
388: and are of the same type as the singularities of $F_j(\ln(1+s))$.
389:
390: \z (iii) Conversely, let $f\in \mathcal{M}'$, with singularities at
391: $\{a_1,...,a_N\}$:
392: \begin{equation}
393: \label{eq:taylozero}
394: f(z)=\sum_{k=0}^{\infty}f_k z^k;\ \ \ |z|<r
395: \end{equation}
396: Then $f_k$ have Borel sum-like representations of the form
397: \begin{equation}
398: \label{eq:trs12}
399: f_k=\frac{1}{2\pi i}\sum_{j=1}^N a_j^{-k}\oint_0^{\infty}e^{-kp}\delta_{a_j}F(a_j e^s)ds
400: \end{equation}
401: where $\delta_{a_j}F(a_j e^s)$ is the difference between the values of
402: $F$
403: on the left of and right the cut from $a_j$ to $\infty$.
404: \end{Theorem}
405:
406: The behaviour at $a_j$ and
407: at $\infty$ will follow from the proof.
408:
409: \bigskip
410: \begin{Note}\label{Resur}
411: {\rm (a) The singularities presented are on the first Riemann
412: sheet. A more global information requires like information on $F_j$.
413:
414: \bigskip
415:
416:
417:
418: \z (b) {\bf Classically Borel summable series.} The integral representations
419: in (\ref{eq:recon3}) would be true Borel sums if $F(p)=(2\pi i)^{-1} \ln p\,
420: H(p)$ with $H$ analytic at zero, as it can be easily checked.
421:
422: (c) {\bf EB Borel summable series.} More generally,
423: it is an EB sum if $F$ is analytic at $0$ in $p^{\beta_j}\ln p^{\alpha_j}$,
424: $j=1,...,N$ with $\Re(\beta_j)>0$. If such is the case, the singularity of
425: $F(p)$ at zero and the singularity of $F(\ln(1+p))$ are of the same type since
426: $\ln(1+p)$ is analytic at zero. Strictly speaking, duality of resurgence only
427: applies to these cases.
428:
429: If $F$ has an exponential or worse type singularity at zero, the
430: representation is {\em not} a generalized Borel sum in any sense. \'Ecalle
431: acceleration might bring it to the simpler case above. But we allow for wilder
432: singularities since this general result goes through.
433:
434: (d) Equation (\ref{eq:sum0}) follows straightforwardly from (\ref{eq:recon3}).
435: To obtain analytic information it suffices to take first say, the
436: imaginary part of $z$ sufficiently large. We rewrite (\ref{eq:sum0})
437: in the form
438: \begin{equation}
439: \label{eq:sum01}
440: f_1(z)= \frac{ z}{\sqrt{\pi}2}\oint_0^{\infty}\frac{ dt }{(1+t)\sqrt{\ln(1+t)}(t-(z-1))}
441: \end{equation}
442: and expand the contour of integration in (\ref{eq:sum01}) up, down and to the
443: left by $2\pi-\epsilon$, to the boundary of a strip. Say the new integral is $I_2$. Then, upon bringing $z$
444: inside the contour we collect a residue
445: \begin{equation}
446: \label{eq:mono1}
447: \frac{ 1}{\sqrt{\pi}}\frac{1 }{\sqrt{\ln z}}
448: \end{equation}
449: Now $z$ can be moved around $0$ or $-1$, on curves staying inside the strip
450: and $I_2$ remains analytic. The monodromy around $0$ and $1$ is thus solely
451: contained in the term (\ref{eq:mono1}). This analysis can be extended to
452: general polylogs, and more detailed information can be obtained. This will be
453: the subject of a different paper.}
454: \end{Note}
455:
456:
457:
458:
459: \z {\large \bf Proof of Theorem \ref{T1}}
460: \z If $f\in\mathcal{M}'$ we write the Taylor coefficients in the form
461: \begin{equation}
462: \label{eq:cf1}
463: f_n=\frac{1}{2\pi i}\oint \frac{ds f(s)}{s^{n+1}}
464: \end{equation}
465: where the contour of integration is a small circle of radius $r$ around the
466: origin. We attempt to increase $r$ without bound. In the process, the contour
467: will hang around the singularities of $f$ as shown in Figure 1. The
468: integrals converge by the decay assumptions and the contribution of the arcs
469: at large $r$ vanish.
470:
471:
472: \bigskip
473:
474:
475: \z In the opposite direction, we let $z$ be sufficiently small, and such that
476: $z a_j\notin\RR$ for all $j$. We choose the contour in such close enough to
477: zero so that $|za_je^{-p}|<\alpha<1$ all along the contour. We note that by
478: dominated convergence we have
479: \begin{multline}
480: \label{eq:converse}
481: \sum_{n=1}^{\infty}f_n z^n=\sum_{j=1}^N \oint_0^{\infty}\sum_{n=1}^{\infty}z^n
482: a_j^{-n}e^{-np}F_j(p)ds\\
483: =\sum_{j=1}^N
484: \oint_0^{\infty}\frac{z a_j^{-1}e^{-p}}{1-za_j^{-1}e^{-p}}F_j(p)dp=
485: \sum_{j=1}^N
486: z\oint_0^{\infty}\frac{F_j(\ln (1+s))ds}{(1+s)(s a_j+a_j-z)}
487: \end{multline}
488: as stated. Now the integral can be analytically continued in $z$. The nature
489: of the singularities of $f$ is examined in \S\ref{CaseI}
490:
491: \subsection{Other growth rates; examples of special sums}\label{Growthrates}
492:
493: Other growth rates can be accomodated, for instance by analytic continuation.
494: We have for positive $\gamma$,
495: \begin{equation}
496: \label{eq:eqinvl}
497: e^{-\gamma\sqrt{
498: n}}=\frac{\gamma}{2\sqrt{\pi}}\int_0^{\infty}p^{-3/2}e^{-\frac{\gamma^2}{4p}}e^{-np}dp
499: \end{equation}
500: which can be analytically continued in $\gamma$. We note forst that the
501: contour cannot be, for this function, detached from zero. Instead, we keep one
502: endpoint at infinity and, near the
503: origin, simultaneously rotate $\gamma$ and $p$ to maintain $-\gamma/p$ real and negative.
504: We get
505: \begin{equation}
506: \label{eq:eqerfc}
507: e^{\sqrt{n}}=-\frac{1}{4\sqrt{\pi}}\int_{C_1} p^{-3/2} e^{\frac{1}{4p}}e^{-np}dp
508: \end{equation}
509: where $C_1$ is described in the introduction, and (\ref{eq:limit1}) follows.
510: Eq. (\ref{eq:eqsum}) is obtained in a similar way.
511:
512:
513: Obviously, if the behavior of the coefficients is of the form $A^kf_k$ where
514: $f_k$ satisfies
515: the conditions in the paper, one simply changes the independent variable to $z'=Az$.
516:
517: \subsection{Singularity formula. Duality.}\label{CaseI}
518: The problem of the type of singularities of the resummed series
519: reduces to finding the singularity type of a Hilbert-transform-like integral
520: of the form
521: \begin{equation}
522: \label{eq:sg}
523: g(t)= \oint_0^{\infty}\frac{G(s)}{s-t}ds
524: \end{equation}
525: The singularity of $g$ at $t=0$ is the same as the singularity of $G$ at $s=0$
526: as follows from a simple calculation.
527: \begin{Lemma}[Analytic structure at $t=0$.]
528: For small $t$ we have
529: \begin{equation}
530: \label{eq:Plm1}
531: \oint_0^{\infty}\frac{G(s)}{s-t}ds=
532: 2\pi i G(t)+G_2(t)
533: \end{equation}
534: where $G_2(t)$ is analytic for small $t$.
535: \end{Lemma}
536: \begin{proof}
537: We take $t$, $|t|=\epsilon$ small, outside the contour of integration. Around $s=0$ we
538: deform the contour into a circle of radius $2\epsilon$ in the process
539: collecting a residue
540: \begin{equation}
541: \label{eq:sing0}
542: 2\pi i G(t)
543: \end{equation}
544: The new integral is manifestly analytic for $|t|<\epsilon$.
545: \begin{Note}
546: {\rm In the particular case of Borel summable series, leading to
547: expressions of the form $\int_0^{\infty}(p-t)^{-1}H(p)dp$ with $p$ analytic at
548: zero, either by converting them to the form $(2\pi i)^{-1}
549: \oint_0^{\infty}(p-t)^{-1}H(p)\ln pdp$ dp or simply writing near zero
550: $H(p)=H(t)+(H(p)-H(t))$ we see that the behaviour for small $t$ is of the form $H(t)\ln
551: t+h(t)$ with $h$ analytic.}
552: \end{Note}
553: {\bf Example:} For the function $f_2$ in the introduction, the inverse Laplace
554: transform
555: is
556: \begin{equation}
557: \label{eq:dual1}
558: \frac{1}{2\pi i}\int_{1-i\infty}^{1+i\infty}\frac{e^{xp}}{x^{\pi}+\ln x}dx
559: \end{equation}
560: where the contour can be bent backwards, to hang around $\RR^-$. Then, with
561: the change of variable $x=-u$ (\ref{eq:dual1}) becomes
562: \begin{equation}
563: \label{eq:dual2}
564: F_1(p)= \oint_{0}^{\infty}\frac{e^{-up}}{(-u)^{\pi}+\ln (-u)}du
565: \end{equation}
566: for which the singularity, at one, according to (\ref{eq:sing0}) is
567: (\ref{intlog}).
568:
569: \end{proof}
570: \section{(ii) Entire
571: functions}\label{entire} We
572: restrict the analysis to entire functions of exponential order one, with
573: complete information on the Taylor coefficients. Such functions include of
574: course the exponential itself, or expressions such as $f_3$. It is useful to
575: start with $f_3$ as an example. The analysis is brought to the case in
576: \S\ref{CaseI} by first taking a Laplace transform. Note that
577: \begin{equation}
578: \label{eq:eq41}
579: \int_0^{\infty}e^{-xz}f(z)dz=\frac{1}{x}\sum_{n=0}^\infty\frac{n!}{n^{n+1}x^{n}}
580: \end{equation}
581: The study of entire functions of exponential order one likely involves the
582: factorial, and then a Borel summed representation of the Stirling formula is
583: needed.
584: \subsection{The Gamma function and Borel summed Stirling formula}\label{S115}
585: We have
586: \begin{multline}
587: n!=\int_0^\infty t^ne^{-t}dt=n^{n+1}\int_0^\infty e^{-n(s-\ln s)}ds\\=
588: n^{n+1}\int_0^1 e^{-n(s-\ln s)}ds+ n^{n+1}\int_1^\infty e^{-n(s-\ln s)}ds
589: \end{multline}
590:
591:
592: \z On $(0,1)$ and $(1,\infty)$ separately, the function $s-\ln(s)$ is
593: monotonic and we may write, after inverting $s-\ln(s)=t$ on the two
594: intervals to get $s_{1,2}=s_{1,2}(t)$,
595: \begin{equation}\label{e25}
596: n!= n^{n+1}\int_1^{\infty}e^{-nt}(s'_2(t)-s'_1(t))dt=n^{n+1}e^{-n}
597: \int_0^{\infty}e^{-np}G(p)dp
598: \end{equation}
599: \z where $G(p)=s_2'(1+p)-s_1'(1+p)$. From the definition it follows that $G$
600: is bounded at infinity and $p^{-1/2}G$ is analytic in $p^{1/2}$ at $p=0$.
601: Using now (\ref{e25}) and Theorem~\ref{T1} in (\ref{eq:eq41}) we get
602: \begin{equation}
603: \label{eq:lapl41}
604: \int_0^{\infty}e^{-xz}f(z)dz=\frac{1}{x^2}\int_0^{\infty}\frac{G(\ln(1+t))}{(te+(e-x^{-1}))(t+1)}dt
605: \end{equation}
606: Upon taking the inverse Laplace transform we obtain (\ref{eq:invl}).
607:
608:
609: \bigskip
610:
611:
612: More generally we obtain from Theorem~\ref{T1}, in the same way as above, the
613: following.
614: \begin{Theorem}\label{T2}
615: Assume that the entire function $f$ is given by
616: \begin{equation}
617: \label{eq:enti22}
618: f(z)= \sum_{k=1}^\infty \frac{f_k z^k }{k!}
619: \end{equation}
620: with $f_k$ as in Theorem~\ref{T1}. Then,
621: \begin{equation}
622: \label{eq:recon31}
623: f(z)=
624: \oint_0^{\infty}\sum_{j=1}^N \left[\left(e^{\frac{z}{s+a_j}}-1\right)
625: \frac{F_j(\ln (1+s))}{(1+s)}\right]ds
626: \end{equation}
627: \end{Theorem}
628: As in the simple example, the behavior at infinity follows from the integral
629: representation by classical means.
630: \section{(iii) Borel summation}
631: \begin{Theorem}\label{T3}
632: Consider the formal power series
633: \begin{equation}
634: \label{eq:enti23}
635: \tilde{f}(z)= \sum_{k=1}^\infty {f_k k! z^{k+1}}
636: \end{equation}
637: with coefficients $f_k$ as in Theorem~\ref{T1}.
638: Then the series is (generalized) Borel summable to
639: \begin{multline}
640: \label{eq:bs1}
641: \int_0^{\infty}dp e^{-p/z}p\sum_{j=1}^N\oint_0^{\infty}\frac{F_j(\ln(1+s))}{(1+s)(a_js+a_j-p)}ds\\=
642: -\sum_{j=1}^N\oint_0^{\infty}\frac{F_j(\ln (1+s))}{1+s}\left(z-a_j(s+1)e^{-\frac{a_j(s+1)}{z}}\mathrm{Ei}\left(\frac{a_j(s+1)}{z}\right)\right)ds
643: \end{multline}
644: \end{Theorem}
645: The proof proceeds as in the previous sections, taking now a Borel
646: followed by Laplace transform.
647: \subsection{Appendix: Borel summed version of $1/n!$}
648: We can use the following representation \cite{Abramowitz}
649: \begin{equation}
650: \label{eq:eqg}
651: \frac{1}{\Gamma(n)}=-\frac{ie^{-\pi i
652: z}}{2\pi}\oint_0^{\infty}s^{-z}e^{-s}ds=-\frac{ie^{-\pi i
653: z}z^{-z}}{2\pi}\oint_0^{\infty}s^{-z}e^{-zs}ds
654: \end{equation}
655: with our convention of contour integration. From here, one can proceed
656: as in \S\ref{S115}.
657: \subsection{Acknowledgments} The author is grateful to M. Kruskal, who
658: introduced him to questions of the type addressed in the paper, S.
659: Garoufalidis who presented concrete problems requiring these techniques and
660: R. D. Costin, S. Tanveer and J. McNeal for very useful discussions. Work supported in part
661: by NSF grants DMS-0406193 and DMS-0600369.
662:
663: \begin{thebibliography}{99}
664: \bibitem{Abramowitz} M Abramowitz and I A Stegun,
665: {\em Handbook of mathematical
666: functions with formulas, graphs, and mathematical tables}
667: New York :
668: Wiley-Interscience (1970).
669: \bibitem{Ecalle-book} J. \'Ecalle {\em Fonctions
670: Resurgentes, Publications Mathematiques D'Orsay, 1981}
671: \bibitem{Ecalle}
672: J. \'Ecalle {\em in Bifurcations and periodic orbits of
673: vector fields NATO ASI Series, Vol. 408, 1993}
674: \bibitem{Ecalle2} J. \'Ecalle {\em Finitude des cycles limites.., Preprint
675: 90-36 of Universite de Paris-Sud, 1990}
676: \bibitem{Ecalle3} J. \'Ecalle, F. Menous {Well behaved averages
677: and the non-accumulation theorem..} Preprint
678: \bibitem{BRBS} W. Balser, B.L.J. Braaksma, J-P. Ramis, Y. Sibuya
679: {\em Asymptotic Anal. {\bf 5}, no. 1 (1991), 27-45}
680: \bibitem{Braaksma} B. L. J. Braaksma {\em Ann. Inst. Fourier,
681: Grenoble,{\bf 42}, 3 (1992), 517-540}
682: \bibitem{Balser} Balser, W. {\em From divergent power series to
683: analytic functions, Springer-Verlag, (1994).}
684: \bibitem{Borel} Borel, E. {\em Lecons sur les series divergentes,
685: Gauthier-Villars, 1901}
686: \bibitem{Hardy} Hardy, C. G. {\em Divergent series}
687: \bibitem{Costin} O. Costin {\em IMRN 8, 377-417 (1995)}
688: \bibitem{Duke} O Costin Duke Math. J. Vol. 93, No.2 pp. 289--344 (1998).
689: \bibitem{Invent} O. Costin, R.D. Costin Inventiones Mathematicae 145, 3, pp
690: 425-485 (2001).
691: \bibitem{OT} O. Costin, S. Tanveer, in preparation.
692: \bibitem{Stokes} G. G. Stokes {\em Trans. Camb. Phil. Soc {\bf 10}
693: 106-128}. Reprinted in {\em Mathematical and Physical papers by late
694: sir George Gabriel Stokes. Cambridge University Press 1904, vol. IV,
695: 77-109}
696: \bibitem{Wasow} W. Wasow {\em Asymptotic expansions
697: for ordinary differential equations, Interscience Publishers 1968 }
698: \bibitem{Sibuya} Y. Sibuya {\em Global theory of a second order linear
699: ordinary differential equation with a polynomial coefficient , North-Holland 1975}
700:
701: \bibitem{Braaksma-K} B L J Braaksma Transseries for a class of nonlinear dierence equations J. Dier.
702: Equations Appl. 7, no. 5, 717--750 (2001).
703: \bibitem{Poincare} O. Costin, S. Tanveer, Nonlinear evolution PDEs in
704: $\RR^+\times \CC^d$: existence and uniqueness of solutions, asymptotic and
705: Borel summability (to appear in Annales de l'Institut Henri Poincaré)
706: \bibitem{OCSG} O. Costin, S. Garoufalidis, Resurgence of the
707: Kontsevich-Zagier power series (submitted).
708: \bibitem{Dingle} R.B. Dingle {\em Asymptotic Expansions: Their Derivation and
709: Interpretation, Academic Press, 1973}
710: \bibitem{Zemanian} A. H. Zemanian, Distribution Theory and Transform Analysis,
711: pp.236, Dover publications, (1987).
712: \end{thebibliography}
713:
714:
715: \end{document}
716: