1: %&LaTeX
2: %------------------------------------------------------------------------------
3: % Beginning of journal.tex
4: %------------------------------------------------------------------------------
5: %
6: % AMS-LaTeX 1.2 sample file for journals, based on amsart.cls.
7: %
8: % Replace amsart by the documentclass for the target journal, e.g. tran-l.
9: %
10: \documentclass{amsart}
11: %\documentclass{tran-l}
12: %\documentclass{article}
13:
14: \usepackage{epsfig}
15: %\usepackage{psfig}
16: %\usepackage{psfrag}
17: \usepackage{graphics}
18: \usepackage{amsmath,amsfonts}
19:
20: \newtheorem{theorem}{Theorem}[section]
21: \newtheorem{lemma}[theorem]{Lemma}
22: \newtheorem{prop}[theorem]{Proposition}
23:
24: \theoremstyle{definition}
25: \newtheorem{definition}[theorem]{Definition}
26: \newtheorem{example}[theorem]{Example}
27: \newtheorem{xca}[theorem]{Exercise}
28:
29: \theoremstyle{remark}
30: \newtheorem{remark}[theorem]{Remark}
31:
32: \numberwithin{equation}{section}
33:
34: % Absolute value notation
35: \newcommand{\abs}[1]{\lvert#1\rvert}
36:
37: % Blank box placeholder for figures (to avoid requiring any
38: % particular graphics capabilities for printing this document).
39: \newcommand{\blankbox}[2]{%
40: \parbox{\columnwidth}{\centering
41: % Set fboxsep to 0 so that the actual size of the box will match the
42: % given measurements more closely.
43: \setlength{\fboxsep}{0pt}%
44: \fbox{\raisebox{0pt}[#2]{\hspace{#1}}}%
45: }%
46: }
47: \newcommand{\ba}{${\scriptstyle /}$}
48: \newcommand{\be}{${\scriptstyle \vert}$}
49: \newcommand{\Cs}{C \kern-2.2mm \raise.4ex\hbox {\ba} \kern.2mm}
50: \newcommand{\Ns}{N \kern-3.4mm \raise.35ex\hbox {\ba} \kern.4mm\hskip 1.3mm}
51: \newcommand{\bas}{${\scriptscriptstyle /}$}
52: \newcommand\Css{C \kern-1.9mm \raise.3ex\hbox {\bas} \kern.8mm}
53:
54: \begin{document}
55:
56: \title{Multi-point Taylor Expansions of Analytic Functions}
57:
58: % Information for first author
59: \author{Jos\'e L. L\'opez}
60: % Address of record for the research reported here
61: \address{Departamento de Mat\'ematica e Inform\'atica,
62: Universidad P\'ublica de Navarra, 31006-Pamplona,
63: Spain}
64: %% Current address
65: %\curraddr{Department of Mathematics and Statistics,
66: %Case Western Reserve University, Cleveland, Ohio 43403}
67: \email{jl.lopez@unavarra.es}
68: % \thanks will become a 1st page footnote.
69: \thanks{The first author wants to thanks the saving bank {\it Caja Rural
70: de Navarra} for its financial support. He also acknowledges
71: the scientific and financial support of CWI in Amsterdam.}
72:
73: % Information for second author
74: \author{Nico M. Temme}
75: \address{CWI, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands}
76: \email{nicot@cwi.nl}
77: \thanks{The authors thank the referee for the comments on the
78: first version of the paper.}
79:
80: % General info
81: \subjclass{Primary 30B10, 30E20; Secondary 40A30.}
82:
83: \date{October 1, 2002 and, in revised form, January 20, 2004.}
84:
85: %\dedicatory{This paper is dedicated to our authors.}
86:
87: \keywords{multi-point Taylor expansions, Cauchy's theorem,
88: analytic functions, multi-point Laurent expansions,
89: uniform asymptotic expansions of integrals}
90:
91: \begin{abstract}
92: Taylor expansions of analytic functions are considered with respect to several
93: points, allowing confluence of any of them.
94: Cauchy-type formulas are given for coefficients and remainders in
95: the expansions, and the regions of convergence are indicated. It is
96: explained how these expansions can be used in deriving uniform asymptotic
97: expansions of integrals. The method is also used for obtaining Laurent
98: expansions in several points as well as Taylor-Laurent expansions.
99: \end{abstract}
100:
101: \maketitle
102:
103: %\section*{This is an unnumbered first-level section head}
104: %This is an example of an unnumbered first-level heading.
105:
106: %% The correct journal style for \specialsection is all uppercase; a known bug
107: %% in amsart.cls prevents this, so input must be uppercase until it is fixed.
108: %\specialsection*{This is a Special Section Head}
109: %\specialsection*{THIS IS A SPECIAL SECTION HEAD}
110: %This is an example of a special section head%
111: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
112: %\footnote{Here is an example of a footnote. Notice that this footnote
113: %text is running on so that it can stand as an example of how a footnote
114: %with separate paragraphs should be written.
115: %\par
116: %And here is the beginning of the second paragraph.}%
117: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
118: %.
119:
120: \section{Introduction}
121: In deriving uniform asymptotic expansions of a certain class of integrals
122: one encounters the problem of expanding a function, that is analytic in some
123: domain $\Omega$ of the complex plane, in several points. The first mention
124: of the use of such expansions in asymptotics is given in \cite{Chester},
125: where Airy-type expansions are derived for integrals having two
126: nearby (or coalescing) saddle points. This reference does not give further
127: details about two-point Taylor expansions, because the coefficients in the
128: Airy-type asymptotic expansion are derived in a different way. Other mentions
129: of the use of such expansions in asymptotics is given in \cite{Raimundas}
130: and \cite{nicoi}.
131: In \cite{Raimundas}, two-point Taylor expansions are used
132: with applications to Airy-type expansions of parabolic cylinder functions.
133: In \cite{nicoi} we used two-point Taylor expansions to derive
134: convergent expansions of Charlier, Laguerre and Jacobi polynomials in terms
135: of Gamma, Hermite and Chebyshev polynomials respectively.
136:
137: To demonstrate an application in asymptotics of multi-Taylor expansions
138: we may consider contour integrals of the form
139: %
140: \begin{equation}
141: \label{intro}
142: I(\lambda;{\bf\alpha})=\int_{C} g(z)e^{-\lambda f(z,{\bf\alpha})}\,dz,
143: \end{equation}
144: %
145: where ${\bf\alpha}$ is a vector of parameters,
146: ${\bf\alpha}=(\alpha_1,\ldots,\alpha_2)$ and the phase function
147: $f(z,{\bf\alpha})$
148: has $m$ saddle points $z_1$, $z_2$,...,$z_m$. The asymptotic behaviour of these
149: integrals for large values of $\lambda$ is determined by
150: the saddle-point structure of the phase function
151: [\cite{wong}, chap. 7, sec. 6]. One method for obtaining an
152: asymptotic expansion of this integral for large values of $\lambda$
153: is based on expanding $g(z)$ at the saddle points of the phase function,
154: %
155: \begin{equation}
156: g(z)=\sum_{n=0}^\infty[a_0+a_1z+\ldots+a_{m-1}z^{m-1}]
157: (z-z_1)^n(z-z_2)^n\cdots(z-z_m)^n
158: \end{equation}
159: %
160: and substitute this expansion into (\ref{intro}). When interchanging summation
161: and integration, the result is a formal expansion in $m$ series in
162: terms of functions related with the functions
163: %
164: \begin{equation}
165: F_{n,k}(\lambda;{\bf\alpha})\equiv\int_{C}
166: z^k(z-z_1)^n(z-z_2)^n\cdots(z-z_m)^ne^{-\lambda f(z,{\bf\alpha})}\,dz,
167: \hskip 5mm k=1,2,\ldots,m-1.
168: \end{equation}
169: %
170: In \cite{Raimundas}, these functions $F_{n,k}(\lambda;{\bf\alpha})$
171: are the Airy functions,
172: whereas in \cite{nicoi} these functions are the Gamma function,
173: or the Hermite or Chebyshev polynomials.
174:
175: In a future paper we will use multi-point Taylor expansions in
176: the asymptotic analysis of integrals arising in diffraction theory, such as
177: the Bessel function integral (see \cite{Janssen} and \cite{Kirk})
178: %
179: \begin{equation}
180: \label{besel}
181: J(x,y)=\int_0^\infty t J_0(yt) e^{i(t^4+xt^2)}\,dt,
182: \end{equation}
183: %
184: which is related to the Pearcey function
185: %
186: \begin{equation}
187: \label{pearcy}
188: \int_{-\infty}^{\infty} e^{i({1\over 4}t^4+{1\over 2}xt^2+yt)}dt.
189: \end{equation}
190: %
191:
192: The Taylor-Laurent expansions
193: will be used to study integrals with two saddle points and a pole of
194: the integrand. Other applications in asymptotics include the
195: study of Hermite-Pad\'e
196: approximations to the exponential function; in \cite{Driver} integrals are
197: considered
198: with three saddle points.
199:
200: %the Pearcey-type integral
201: %%
202: %\eqn\one{
203: %F(x,y)=\int_{-\infty}^{\infty} f(t)e^{i({t^4\over 4}+x{t^2\over 2}+yt)}dt,
204: %}
205: %or the swallowtail-type integral
206: %%
207: %\eqn\two{
208: %S(x,y,z)=\int_{-\infty}^{\infty} f(t)e^{i({t^5\over 5}+x{t^3\over 3}+
209: %y{t^2\over 2}+zt)}dt,
210: %}
211: %where $x$, $y$ and $z$ are real parameters. The asymptotic behaviour of these
212: %integrals for large values of $x$ and one of both $y$ or $z$ is determined by
213: %the saddle-point structure of the phase function $[$\wong, chap. 7, sec. 6$]$.
214:
215: In a recent paper \cite{nicoii} we have introduced the theory of
216: two-point Taylor expansions, two-point Laurent expansions and
217: two-point Taylor-Laurent expansions. The purpose of the
218: present paper is to generalize that theory from 2 to $m$ points,
219: $m\ge 2$.
220: We give details on the region of convergence and on
221: representations in terms of Cauchy-type integrals of the coefficients and
222: the remainders of the expansions.
223: %Some information on this type of
224: %expansions is also given in \Whittaker, p. 149, Exercise 24.
225: %More
226: Earlier information on this type of expansions is given
227: in \cite{walsh}, Chapters 3 and 8. The theory of
228: several-point Taylor expansions was already formulated in Chapter 3 of
229: Walsh's book, although in a different setting.
230: Chapter 8 of \cite{walsh} presents also a theory of rational approximation of
231: analytic functions, but is different from the theory of multi-point Laurent
232: and Taylor-Laurent expansions presented here. Whereas the
233: multi-point polynomial approximation of Chapter 3 may be reformulated as a
234: multi-point Taylor approximation, the rational approximation of Chapter 8
235: can not be written as a multi-point Laurent or Taylor-Laurent approximation.
236: For more details, see Section 5.
237:
238:
239:
240: \section{Multy-point Taylor expansions}
241: \noindent
242: We consider the Taylor expansion of an analytic function $f(z)$ in several
243: points and give information on
244: the coefficients and the remainder in the expansion. In what follows
245: empty sums and derivatives of negative order must be understood as
246: zero and empty products as one. We
247: will deal with the following set of points:
248:
249:
250: \begin{definition}
251: \label{def1}
252: We define the set
253: %
254: \begin{equation}
255: S\equiv\lbrace z_1,z_1,\ldots,z_1;
256: z_2,z_2,\ldots,z_2,\ldots;z_p,z_p,\ldots,z_p\rbrace
257: \end{equation}
258: %
259: of $m$ points consisting on $p$ different points $z_1$, $z_2$,..., $z_p$
260: ($z_i\ne z_j$ if $i\ne j$),
261: each $z_j$ repeated $m_j$ times: $m_1+m_2+\ldots+m_p=m$.
262: \end{definition}
263:
264: For clearity in the exposition, we first introduce the multi-point Taylor
265: expansion for $m$ different points $z_1$, $z_2$, $\ldots z_m$ ($m=p$, $m_j=1$)
266: in Theorem \ref{theo1}. In Theorem \ref{theo2} we assume that the points
267: $z_1$, $z_2$, $\ldots z_m$ may coalesce.
268: We will need the following elementary lemma.
269:
270: \begin{lemma}
271: \label{lema1}
272: Given $z$, $w$, $\in\Cs$, take $m$
273: different points $z_1$, $z_2$,$\ldots$, $z_m$ in $\Cs$ and define
274: %
275: \begin{equation}
276: \label{defache}
277: H_m(w,z;z_1,\ldots,z_m)\equiv{\prod_{k=1}^m(w-z_k)
278: -\prod_{k=1}^m(z-z_k)\over w-z}.
279: \end{equation}
280: %
281: Then
282: %
283: \begin{equation}
284: \label{ache}
285: H_m(w,z;z_1,\ldots,z_m)=\sum_{j=1}^m{\prod_{k=1,k\ne j}^m(w-z_k)
286: \prod_{k=1,k\ne j}^m(z-z_k)\over \prod_{k=1,k\ne j}^m(z_j-z_k)}.
287: \end{equation}
288: %
289: \end{lemma}
290:
291: \begin{proof}
292: The numerator of $H_m(w,z;z_1,\ldots,z_m)$ is a polynomial of
293: degree $m$ in the variable $w$ that vanishes at $w=z$. Therefore,
294: $H_m(w,z;z_1,\ldots,z_m)$ is a polynomial of degree $m-1$ in the variable $w$.
295: Let $P_m(w,z;z_1,\ldots,z_m)$ denote the function at the right-hand side of
296: (\ref{ache}), which is also
297: a polynomial of degree $m-1$ in the variable $w$. Moreover,
298: %
299: \begin{equation}
300: H_m(z_s,z;z_1,\ldots,z_m)=\prod_{k=1,k\ne
301: s}^m(z-z_k)=P_m(z_s,z;z_1,\ldots,z_m)
302: \end{equation}
303: %
304: for $s=1$, $2$,$\ldots m$. Hence,
305: %
306: \begin{equation}
307: H_m(w,z;z_1,\ldots,z_m)=P_m(w,z;z_1,\ldots,z_m).
308: \end{equation}
309: %
310: \end{proof}
311:
312:
313: \begin{theorem}
314: \label{theo1}
315: Let $f(z)$ be an analytic function on an open set
316: $\Omega\subset C$ and $S\subset\Omega$ with $S$ consisting
317: on $m$ different points ($m=p$).
318: Then, $f(z)$ has the multi-point Taylor expansion
319: %
320: \begin{equation}
321: \label{expani}
322: f(z)=\sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^m(z-z_k)^{n}+r_N(z),
323: \end{equation}
324: %
325: where $q_{n,m}(z)$ is the polynomial of degree $m-1$
326: %
327: \begin{equation}
328: \label{poly}
329: q_{n,m}(z)\equiv\sum_{j=1}^m a_{n,j}
330: {\prod_{k=1,k\ne j}^m(z-z_k)\over \prod_{k=1,k\ne j}^m(z_j-z_k)}
331: \end{equation}
332: %
333: and the coefficients $a_{n,j}$ of this polynomial are given by the
334: Cauchy integral
335: %
336: \begin{equation}
337: \label{coefi}
338: a_{n,j}\equiv{1\over 2\pi i}
339: \int_C{f(w)\,dw\over (w-z_j)\prod_{k=1}^m(w-z_k)^{n}}.
340: \end{equation}
341: %
342: The remainder term $r_N(z)$ is given by the Cauchy integral
343: %
344: \begin{equation}
345: \label{remi}
346: r_N(z)\equiv{1\over 2\pi i}\int_C{f(w)\,dw\over (w-z)
347: \left[\prod_{k=1}^m(w-z_k)\right]^N}
348: \left[\prod_{k=1}^m(z-z_k)\right]^N.
349: \end{equation}
350: %
351: The contour of integration $C$ is a simple closed loop which
352: encircles the points $z_1$, $z_2$,...,$z_m$ (for $a_{n,j}$)
353: and $z$, $z_1$, $z_2$,...,$z_m$ (for $r_N(z)$) in
354: the counterclockwise direction and is contained in $\Omega$ (see
355: Figure 1 (a)).
356:
357: The expansion (\ref{expani}) is convergent for $z\in O_m$, where:
358: %
359: \begin{equation}
360: \label{ere}
361: O_m\equiv\lbrace z\in\Omega, \hskip 2mm
362: \prod_{k=1}^m\vert z-z_k\vert<r\rbrace, \hskip 2cm
363: r\equiv {\rm Inf}_{w\in\Css\setminus\Omega}\left\lbrace
364: \prod_{k=1}^m\vert w-z_k\vert\right\rbrace.
365: \end{equation}
366: %
367: That is, (\ref{expani}) is convergent for $z$ inside the lemniscate
368: $\prod_{k=1}^m\vert z-z_k\vert=r$ (see
369: Figure 2; if $m=1$ this domain is a disk;
370: if $m=2$ this domain is bounded by a Cassini oval).
371: In particular, if $f(z)$ is an entire function $(\Omega=\Cs$), then the
372: expansion (\ref{expani}) converges $\forall$ $z\in\Cs$.
373: \end{theorem}
374:
375: \begin{proof}
376: By Cauchy's theorem,
377: %
378: \begin{equation}
379: \label{cauchy}
380: f(z)={1\over 2\pi i}\int_C{f(w)\,dw\over w-z},
381: \end{equation}
382: %
383: where $C$ is the contour defined above (Figure 1 (a)). We write
384: %
385: \begin{equation}
386: \label{lequal}
387: {1\over w-z}={H_m(w,z;z_1,\ldots,z_m)\over \prod_{k=1}^m(w-z_k)}
388: {1\over 1-u},
389: \end{equation}
390: %
391: where $H_m(w,z;z_1,\ldots,z_m)$ is given in (\ref{defache}) and
392: %
393: \begin{equation}
394: \label{ui}
395: u\equiv {\prod_{k=1}^m(z-z_k)\over \prod_{k=1}^m(w-z_k)}.
396: \end{equation}
397: %
398: Now we use Lemma \ref{lema1} and introduce the right hand side of
399: (\ref{ache}) and the expansion
400: %
401: \begin{equation}
402: \label{expanu}
403: {1\over 1-u}=\sum_{n=0}^{N-1}u^n+{u^N\over 1-u}
404: \end{equation}
405: %
406: in (\ref{lequal}) and this in (\ref{cauchy}). After straightforward
407: calculations we
408: obtain formulas (\ref{expani})-(\ref{remi}).
409:
410: For any $z\in O_{m}$, we can take a contour $C$ in $\Omega$ such that
411: %
412: \begin{equation}
413: \prod_{k=1}^m\vert z-z_k\vert<\prod_{k=1}^m\vert w-z_k\vert,
414: \quad \forall w\in\Cs
415: \end{equation}
416: %
417: (see Figure 1 (b)). On this contour $\vert f(w)\vert$ is bounded by
418: some constant $C$: $\vert
419: f(w)\vert\le C$. Introducing these two bounds in (\ref{remi}) we see that
420: $\lim_{N\to\infty}r_N(z)=0$ and the proof follows.
421: \end{proof}
422:
423:
424: \begin{figure}[tb]
425: %\blankbox{.6\columnwidth}{5pc}
426: \begin{center}\includegraphics[width=8cm]{fig1.eps}\end{center}
427: \centerline{(a) \hskip 6cm (b)}
428: \caption{The case $m=3$. (a) Contour $C$ in the integrals
429: (\ref{coefi}) and (\ref{remi}). (b) For
430: $z\in O_{m}$, we can take a contour $C$ in $\Omega$ which contains
431: $O_{m}$ inside and therefore,
432: $\prod_{k=1}^m\vert z-z_k\vert<\prod_{k=1}^m\vert w-z_k\vert$
433: $\forall$ $w\in\Cs$.}
434: \label{Figure 1}
435: \end{figure}
436:
437:
438: \begin{figure}[tb]
439: \begin{center}\includegraphics[width=12cm]{fig2.eps}\end{center}
440: \centerline{$r=r_a$ \hskip 4cm $r=r_b$ \hskip 4cm $r=r_c$}
441: %\blankbox{.6\columnwidth}{5pc}
442: \caption{Shape of the "lemniscate domain" $O_m$ for $m=3$. It
443: depends on the size of the parameter $r$ defined in (\ref{ere}). In
444: these pictures
445: $\vert z_2-z_3\vert<\vert z_1-z_3\vert$, $\vert z_1-z_2\vert$
446: and $r_a>r_b>r_c$.}
447: \label{Figure 2}
448: \end{figure}
449:
450:
451: We need the following lemma to consider the case of coalescing points in the
452: set~$S$.
453:
454: \begin{lemma}
455: \label{lema2}
456: Given $z,w\in\Cs$, take $m$ different points $z_1$, $z_2$,
457: ..., $z_m$ in $C$, all different from $w$ too. Let those $m$ points
458: to coalesce at $z_m$, say. Then
459: %
460: \begin{equation}
461: \lim_{z_1,z_2,..,z_{m-1}\to z_m}\sum_{j=1}^m{\prod_{k=1,k\ne j}^m(z-z_k)
462: \over (w-z_j)\prod_{k=1,k\ne j}^m(z_j-z_k)}=
463: \sum_{j=0}^{m-1}{(z-z_m)^j\over (w-z_m)^{j+1}}.
464: \end{equation}
465: %
466: \end{lemma}
467:
468: \begin{proof}
469: We first note that the identity
470: %
471: \begin{equation}
472: \sum_{j=1}^n\prod_{l=1}^{j-1}(z_n-z_l)\prod_{l=j+1}^n(z_1-z_l)=0
473: \end{equation}
474: %
475: holds for any set of points $z_1$, $z_2$,..., $z_n$, $n>1$. It may be
476: cheked in the following way: we take the two first terms of the sum,
477: which gives
478: %
479: \begin{equation}
480: (z_n-z_2)(z_1-z_3)(z_1-z_4)\cdot\cdot\cdot(z_1-z_{n-1})(z_1-z_n).
481: \end{equation}
482: %
483: Next we add
484: to this the third term of the sum, which gives
485: %
486: \begin{equation}
487: (z_n-z_2)(z_n-z_3)(z_1-z_4)\cdot\cdot\cdot(z_1-z_{n-1})(z_1-z_n).
488: \end{equation}
489: %
490: We continue
491: this process untill we add the $n-1$-th term of the sum, obtaining
492: %
493: \begin{equation}
494: (z_n-z_2)(z_n-z_3)(z_n-z_4)\cdot\cdot\cdot(z_n-z_{n-1})(z_1-z_n).
495: \end{equation}
496: %
497: But this is just the last term of the sum with opposite sign.
498:
499: Using the above identity we have
500: %
501: \begin{equation}
502: \prod_{l=1}^{k-1}(z_s-z_l)\left\lbrack
503: \sum_{j=k}^s\prod_{l=k}^{j-1}(z_s-z_l)\prod_{l=j+1}^s(z_k-z_l)
504: \right\rbrack \prod_{l=s+1}^{m}(z_k-z_l)=0
505: \end{equation}
506: %
507: for any $s=1,2,3,...,m$ and any $k=1,2,3,...,m$ with $k\ne s$. Then
508: %
509: \begin{equation}
510: \label{cero}
511: \sum_{j=k}^s\prod_{l=1}^{j-1}(z_s-z_l)\prod_{l=j+1}^m(z_k-z_l)=
512: \sum_{j=1}^m\prod_{l=1}^{j-1}(z_s-z_l)\prod_{l=j+1}^m(z_k-z_l)=0
513: \end{equation}
514: %
515: for any $s,k=1,2,3,...,m$ with $k\ne s$.
516:
517: Now, for every $s=1,2,3,...,m$, we define the
518: following polynomials of degree $m-1$ in the variable $z$:
519: %
520: \begin{equation}
521: R_s(z)\equiv\prod_{l=1,l\ne s}^m(z-z_l),
522: \hskip 2cm
523: S_s(z)\equiv\sum_{j=1}^m\prod_{l=1}^{j-1}(z-z_l)\prod_{l=j+1}^m(z_s-z_l).
524: \end{equation}
525: %
526: The zeros of $R_s(z)$ are $z_k$ for $k=1,2,3,...,m$, $k\ne s$ and from
527: (\ref{cero}),
528: $S_s(z_k)=0$ for $k=1,2,3,...,m$, $k\ne s$. Moreover, the leading
529: coeficient of $R_s(z)$
530: and of $S_s(z)$ coincide. Therefore, $R_s(z)=S_s(z)$ for $s=1,2,3,...,m$.
531:
532: Finally, define the following polynomials of degree $m-1$ in the variable $w$:
533: %
534: \begin{equation}
535: P(w,z)\equiv\sum_{j=1}^m{\prod_{k=1,k\ne j}^m(z-z_k)\prod_{k=1,k\ne
536: j}^m(w-z_k)\over
537: \prod_{k=1,k\ne j}^m(z_j-z_k)},
538: \end{equation}
539: %
540: %
541: \begin{equation}
542: Q(w,z)\equiv\sum_{j=1}^m\prod_{k=1}^{j-1}(z-z_k)\prod_{k=j+1}^m(w-z_k).
543: \end{equation}
544: %
545: For every $s=1,2,3,...,m$ we have $P(z_s,z)=R_s(z)$ and $Q(z_s,z)=S_s(z)$. But
546: $R_s(z)=S_s(z)$ and therefore $P(w,z)=Q(w,z)$. Then,
547: %
548: \begin{equation}
549: \begin{split}
550: \sum_{j=1}^m{\prod_{k=1,k\ne j}^m(z-z_k)\over(w-z_j)\prod_{k=1,k\ne
551: j}^m(z_j-z_k)}= &
552: {P(w,z)\over\prod_{k=1}^m(w-z_k)}= \\ & {Q(w,z)\over\prod_{k=1}^m(w-z_k)}=
553: \sum_{j=1}^m{\prod_{k=1}^{j-1}(z-z_k)\over\prod_{k=1}^j(w-z_k)}.
554: \end{split}
555: \end{equation}
556: %
557: Taking the limits $z_1$, $z_2$,.., $z_{m-1}$ $\to$ $z_m$ in the left- and
558: right-hand sides of these equalities, we obtain the desired result.
559: \end{proof}
560:
561:
562:
563: \begin{theorem}
564: \label{theo2}
565: Let $f(z)$ be an analytic function on an open set
566: $\Omega\subset\Cs$ and $S\subset\Omega$.
567: Then, $f(z)$ has the multi-point Taylor expansion
568: %
569: \begin{equation}
570: \label{expanib}
571: f(z)=\sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^p(z-z_k)^{nm_k}+r_N(z),
572: \end{equation}
573: %
574: where $q_{n,m}(z)$ is the polynomial of degree $m-1$
575: %
576: \begin{equation}
577: \label{polyb}
578: q_{n,m}(z)\equiv\sum_{j=1}^p
579: {\prod_{k=1,k\ne j}^p(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^p(z_j-z_k)^{m_k}}
580: \sum_{l=0}^{m_j-1}a_{n,j,l}(z-z_j)^l
581: \end{equation}
582: %
583: and the coefficients $a_{n,j,l}$ of this polynomial are given by the
584: Cauchy integral
585: %
586: \begin{equation}
587: \label{coefib}
588: a_{n,j,l}\equiv{1\over 2\pi i}
589: \int_C{f(w)\,dw\over (w-z_j)^{l+1}\prod_{k=1}^p(w-z_k)^{nm_k}}.
590: \end{equation}
591: %
592: The remainder term $r_N(z)$ is given by the Cauchy integral
593: %
594: \begin{equation}
595: \label{remib}
596: r_N(z)\equiv{1\over 2\pi i}\int_C{f(w)\,dw\over (w-z)
597: \left[\prod_{k=1}^p(w-z_k)^{m_k}\right]^N}
598: \left[\prod_{k=1}^p(z-z_k)^{m_k}\right]^N.
599: \end{equation}
600: %
601: The contour of integration $C$ is a simple closed loop which
602: encircles the points $z_1$, $z_2$,...,$z_p$ (for $a_{n,j,l}$)
603: and $z$, $z_1$, $z_2$,...,$z_p$ (for $r_N(z)$) in
604: the counterclockwise direction and is contained in $\Omega$ (see
605: Figure 1 (a)).
606:
607: The expansion (\ref{expanib}) is convergent for $z\in O_p$:
608: %
609: \begin{equation}
610: O_p\equiv\lbrace z\in\Omega, \hskip 2mm
611: \prod_{k=1}^p\vert z-z_k\vert^{m_k}<r\rbrace, \hskip 2cm
612: r\equiv {\rm Inf}_{w\in\Css\setminus\Omega}\left\lbrace
613: \prod_{k=1}^p\vert w-z_k\vert^{m_k}\right\rbrace,
614: \end{equation}
615: %
616: that is, inside the lemniscate $\prod_{k=1}^p\vert z-z_k\vert^{m_k}=r$.
617: In particular, if $f(z)$ is an entire function $(\Omega=\Cs$), then the
618: expansion (\ref{expanib}) converges $\forall$ $z\in\Cs$.
619: \end{theorem}
620:
621:
622: \begin{proof}
623: If all the points in $S$ are different,
624: we have from (\ref{poly}) and (\ref{coefi})
625: %
626: \begin{equation}
627: \label{cuco}
628: q_{n,m}(z)={1\over 2\pi i}
629: \int_C{f(w)\,dw\over \prod_{k=1}^m(w-z_k)^n}
630: \sum_{j=1}^m{\prod_{k=1, k\ne j}^m(z-z_k)
631: \over (w-z_j)\prod_{k=1, k\ne j}^m(z_j-z_k)}.
632: \end{equation}
633: %
634: This last sum may be also decomposed in the form
635: %
636: \begin{equation}
637: \begin{split} &
638: \sum_{j=1}^{m_1}{\prod_{k=1, k\ne j}^m(z-z_k)
639: \over (w-z_j)\prod_{k=1, k\ne j}^m(z_j-z_k)}+
640: \sum_{j=m_1+1}^{m_2}{\prod_{k=1, k\ne j}^m(z-z_k)
641: \over (w-z_j)\prod_{k=1, k\ne j}^m(z_j-z_k)}+\ldots \\ &
642: \sum_{j=m_{p-1}+1}^m{\prod_{k=1, k\ne j}^m(z-z_k)
643: \over (w-z_j)\prod_{k=1, k\ne j}^m(z_j-z_k)}.
644: \end{split}
645: \end{equation}
646: %
647: Now let the first $m_1$ points to coalesce to $z_1$, the second $m_2$
648: points to coalesce to $z_2$, and so on, and apply Lemma \ref{lema2} to every
649: one of the $p$
650: sums above to obtain (\ref{expanib}), (\ref{polyb}) and
651: (\ref{coefib}). Equation
652: (\ref{remib}) follows from (\ref{remi}). The proof of the convergence
653: of (\ref{expanib}) in the region $O_{p}$ is a straightforward generalization
654: of the correponding proof in Theorem \ref{theo1}.
655: \end{proof}
656:
657:
658:
659: \subsection{Explicit forms of the coefficients}
660:
661:
662: \noindent
663: Formula (\ref{coefib}) is not appropriate for numerical computations.
664: A more practical formula to compute the coefficients of the above
665: multi-point Taylor expansion is given in the following proposition.
666: First we have a definition:
667:
668: \begin{definition}
669: \label{def2}
670: Let $f(w)$ be analytic at $w$; then for $n=0,1,2,\ldots$
671: the differentential operator $D_w^n f(w)$ is defined by
672: %
673: \begin{equation}
674: D_w^n f(w)={1\over n!} {d^n\over dw^n} f(w).
675: \end{equation}
676: %
677: \end{definition}
678:
679: \begin{prop}
680: \label{prop1}
681: The coefficients $a_{n,j,l}$, for $n=1,2,3,\ldots$,
682: $j=1,2,\ldots,p$, $l=0,1,\ldots,m_j-1$
683: in the expansion (\ref{expanib}) are also given by the formula:
684: %
685: \begin{equation}
686: \label{coefibis}
687: \begin{split}
688: a_{n,j,l}=&
689: D_w^{nm_j+l}\left.\left[
690: {f(w)\over\prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}
691: \right]\right\vert_{w=z_j}+ \\ &
692: \sum_{k=1,k\ne j}^pD_w^{nm_k-1}
693: \left.\left[
694: {f(w)\over(w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
695: \right]\right\vert_{w=z_k}.
696: \end{split}
697: \end{equation}
698: %
699: \end{prop}
700:
701: \begin{proof}
702: We deform the contour of integration $C$ in equation
703: (\ref{coefib}) to any contour of the form
704: $C_1\cup C_2\cup\cdots\cup C_p$, also contained in $\Omega$,
705: where $C_k$, $k=1,2,\ldots,p$, is a simple closed loop which
706: encircles the
707: point $z_k$ in the counterclockwise direction and does not contain any
708: other point $z_j$, $j=1,2,\ldots,p$, $j\ne k$ inside (see Figure 3 (a)). Then,
709: %
710: \begin{equation}
711: \begin{split}
712: a_{n,j,l}= & {1\over 2\pi i}\sum_{k=1,k\ne j}^p
713: \int_{C_k}{f(w)\over
714: (w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
715: {dw\over (w-z_k)^{nm_k}}+ \\ &
716: {1\over 2\pi i}\int_{C_j}{f(w)\over
717: \prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}{dw\over (w-z_j)^{nm_j+l+1}},
718: \end{split}
719: \end{equation}
720: %
721: from which equation (\ref{coefibis}) follows.
722: \end{proof}
723:
724:
725: \begin{figure}[tb]
726: %\blankbox{.6\columnwidth}{5pc}
727: \begin{center}\includegraphics[width=12cm]{fig3.eps}\end{center}
728: \centerline{(a) \hskip 4cm (b)\hskip 4cm (c)}
729: \caption{Integration contours $C_k$ for $p=3$
730: and $q=1$. (a) The function
731: $\prod_{s=1,s\ne k}^p(w-z_s)^{-nm_s}f(w)$ is analytic
732: inside $C_k$ for $k=1,2,\ldots,p$.
733: (b) The functions $\prod_{s=1,s\ne k}^p(w-z_s)^{-nm_s}g_k(w)$ and
734: $\prod_{s=1,s\ne k}^p(w-z_s)^{(n+1)m_s}g_k(w)$ are analytic inside
735: $C_k$ for $k=1,2,\ldots,p$.
736: (c) The functions $\prod_{s=1,s\ne k}^p(w-z_s)^{-nm_s}g_k(w)$,
737: $\prod_{s=1,s\ne k}^q(w-z_s)^{-nm_s}
738: \prod_{s=q+1,s\ne k}^p(w-z_s)^{nm_s}g_k(w)$ and
739: $\prod_{s=1,s\ne k}^q(w-z_s)^{-(n+1)m_s}\ \times\ $
740: $\prod_{s=q+1,s\ne k}^p(w-z_s)^{(n+1)m_s}g_k(w)$ are analytic inside
741: $C_k$ for $k=1,2,\ldots,p$.}
742: \label{Figure 3}
743: \end{figure}
744:
745:
746: \subsection{Multi-point Taylor polynomials}
747:
748:
749: \noindent
750: In Theorem \ref{theo1} we have assumed that the function $f(z)$ is analytic in
751: $\Omega$. If $f(z)$ is not analytic in
752: $\Omega$ but has a finite number of derivatives at $z_1$,
753: $z_2$,...,$z_p$, we can still define the multi-point Taylor polynomial of the
754: function $f(z)$ at $z_1$, $z_2$,...,$z_p$ in the following way:
755:
756: \begin{definition}
757: \label{def3}
758: Let $z$ be a real or complex variable. If $f(z)$ is
759: $Nm_k-1-$times differentiable at $z_1$, $z_2$,...,$z_p$,
760: we define the multi-point
761: Taylor polynomial of $f(z)$ at the points of $S$ and degree $mN-1$ as
762: %
763: \begin{equation}
764: P_N(z)\equiv\sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^p(z-z_k)^{nm_k},
765: \end{equation}
766: %
767: where $q_{n,m}(z)$ is the polynomial of degree $m-1$
768: %
769: \begin{equation}
770: q_{n,m}(z)\equiv\sum_{j=1}^p
771: {\prod_{k=1,k\ne j}^p(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^p(z_j-z_k)^{m_k}}
772: \sum_{l=0}^{m_j-1}a_{n,j,l}(z-z_j)^l
773: \end{equation}
774: %
775: and the coefficients $a_{n,j,l}$ are given in (\ref{coefibis}).
776: \end{definition}
777:
778: \begin{prop}
779: \label{prop2}
780: In the conditions of the above definition, define
781: the remainder of the
782: approximation of $f(z)$ by $P_N(z)$ at the points of $S$ as
783: %
784: \begin{equation}
785: r_N(z)\equiv f(z)-P_N(z).
786: \end{equation}
787: %
788: Then, (i) $r_N(z)=o(z-z_k)^{Nm_k-1}$ as $z\to z_k$, $k=1,2,\ldots,p$.
789: (ii) If $f(z)$ is $Nm_k-$times differentiable at $z_k$ for some $k$, then
790: $r_N(z)=O(z-z_k)^{Nm_k}$ as $z\to z_k$.
791: \end{prop}
792:
793: \begin{proof}
794: The proof is trivial if $f(z)$ is analytic at every
795: $z_1$, $z_2$,...,$z_p$ by using (\ref{remib}).
796: In any case, for real or complex variable, the proof follows
797: by using l'H\^opital's rule and (\ref{coefibis}).
798: \end{proof}
799:
800:
801: \begin{remark}
802: \label{rem1}
803: Observe that the Taylor polynomial of $f(z)$ at the points of $S$
804: and degree $mN-1$ is the Hermite's interpolation polynomial of
805: $f(z)$ at $z_1$, $z_2$,...,$z_p$ with data $f(z_k)$,
806: $f'(z_k)$,...,$f^{(Nm_k-1)}(z_k)$, $k=1,2,\ldots,p$.
807: \end{remark}
808:
809:
810: \section{Multi-point Laurent expansions}
811:
812: \noindent
813: In the standard theory for Taylor and Laurent expansions much analogy
814: exists between the two types of expansions. For multi-point expansions, we
815: have a similar resemblance in the representations of coefficients and remainders.
816:
817:
818: \begin{theorem}
819: \label{theo3}
820: Let $\Omega_0$ and $\Omega$ be closed and open sets,
821: respectively, of the complex plane, and $\Omega_0\subset\Omega\subset\Cs$. Let
822: $f(z)$ be an analytic function on $\Omega\setminus\Omega_0$
823: and $z_1$, $z_2$,...,$z_p\in\Omega_0$ (That is, $S\in\Omega_0$).
824: Then, for any $z\in\Omega\setminus\Omega_0$,
825: $f(z)$ has the multi-point Laurent expansion
826: %
827: \begin{equation}
828: \label{expanii}
829: f(z)= \sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^p(z-z_k)^{nm_k}+
830: \sum_{n=0}^{N-1}t_{n,m}(z)\prod_{k=1}^p(z-z_k)^{-(n+1)m_k}+ r_N(z),
831: \end{equation}
832: %
833: where $q_{n,m}(z)$ is the polynomial of degree $m-1$
834: %
835: \begin{equation}
836: \label{polybis}
837: q_{n,m}(z)\equiv\sum_{j=1}^p
838: {\prod_{k=1,k\ne j}^p(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^p(z_j-z_k)^{m_k}}
839: \sum_{l=0}^{m_j-1}a_{n,j,l}(z-z_j)^l
840: \end{equation}
841: %
842: and the coefficients $a_{n,j,l}$ of this polynomial are given by the
843: Cauchy integral
844: %
845: \begin{equation}
846: \label{coefii}
847: a_{n,j,l}\equiv{1\over 2\pi i}
848: \int_{\Gamma_1}{f(w)\,dw\over (w-z_j)^{l+1}\prod_{k=1}^p(w-z_k)^{nm_k}}.
849: \end{equation}
850: %
851: Also, $t_{n,m}(z)$ is the polynomial of degree $m-1$
852: %
853: \begin{equation}
854: \label{polyi}
855: t_{n,m}(z)\equiv\sum_{j=1}^p
856: {\prod_{k=1,k\ne j}^p(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^p(z_j-z_k)^{m_k}}
857: \sum_{l=0}^{m_j-1}b_{n,j,l}(z-z_j)^l,
858: \end{equation}
859: %
860: where the coefficients $b_{n,j,l}$ of this polynomial are given by the
861: Cauchy integral
862: %
863: \begin{equation}
864: \label{coefiii}
865: b_{n,j,l}\equiv{1\over 2\pi i}
866: \int_{\Gamma_2}\prod_{k=1}^p(w-z_k)^{m_k(n+1)}{f(w)\,dw\over (w-z_j)^{l+1}}.
867: \end{equation}
868: %
869: The remainder term $r_N(z)$ is given by the Cauchy integrals
870: %
871: \begin{equation}
872: \label{remii}
873: \begin{split}
874: r_N(z)\equiv & {1\over 2\pi i}\int_{\Gamma_1}{f(w)\,dw\over (w-z)
875: \prod_{k=1}^p(w-z_k)^{Nm_k}}\prod_{k=1}^p(z-z_k)^{Nm_k}- \\ &
876: {1\over 2\pi i}\int_{\Gamma_2}\prod_{k=1}^p(w-z_k)^{Nm_k}{f(w)\,dw\over w-z
877: }{1\over\prod_{k=1}^p(z-z_k)^{Nm_k}}.
878: \end{split}
879: \end{equation}
880: %
881: In these integrals, the contours of integration ${\Gamma}_1$ and ${\Gamma}_2$
882: are simple closed loops contained in $\Omega\setminus\Omega_0$ which
883: encircle the
884: points $z_1$, $z_2$,...,$z_p$ in the counterclockwise direction.
885: Moreover, $z$ is not inside ${\Gamma}_2$,
886: whereas ${\Gamma}_1$ encircles ${\Gamma}_2$ and the
887: point $z$ (see Figure 4 (a)).
888:
889: \noindent
890: The expansion (\ref{expanii}) is convergent for $z$ inside the
891: "lemniscate annulus" (see Figure 5)
892: %
893: \begin{equation}
894: \label{domainii}
895: A_{p}\equiv\lbrace z\in\Omega\setminus\Omega_0, \hskip 2mm
896: r_2<\prod_{k=1}^p\vert z-z_k\vert^{m_k}<r_1\rbrace,
897: \end{equation}
898: %
899: where
900: %
901: \begin{equation}
902: \label{erei}
903: r_1\equiv {\rm Inf}_{w\in\Css\setminus\Omega}\left\lbrace
904: \prod_{k=1}^p\vert w-z_k\vert^{m_k}\right\rbrace,\quad
905: r_2\equiv{\rm Sup}_{w\in\Omega_0}\left\lbrace
906: \prod_{k=1}^p\vert w-z_k\vert^{m_k}\right\rbrace.
907: \end{equation}
908: %
909: \end{theorem}
910:
911: \begin{proof}
912: By Cauchy's theorem,
913: %
914: \begin{equation}
915: \label{cauchybis}
916: f(z)={1\over 2\pi i}\int_{{\Gamma}_1}{f(w)dw\over w-z}-
917: {1\over 2\pi i}\int_{{\Gamma}_2}{f(w)\,dw\over w-z},
918: \end{equation}
919: %
920: where ${\Gamma}_1$ and ${\Gamma}_2$ are the contours defined above.
921: First we assume that the $m$ points of the set
922: $S$ are all distinct and later we will let the first $m_1$
923: points to coalesce to $z_1$, the second $m_2$ points to coalesce to
924: $z_2$ and so on. We
925: substitute (\ref{lequal})-(\ref{ui}) into the first integral above and
926: %
927: \begin{equation}
928: {1\over w-z}=-{H_m(w,z;z_1,\ldots,z_m)\over \prod_{k=1}^m(z-z_k)}{1\over 1-u},
929: \hskip 1cm
930: u\equiv {\prod_{k=1}^m(w-z_k)\over \prod_{k=1}^m(z-z_k)},
931: \end{equation}
932: %
933: where $H_m(w,z;z_1,\ldots,z_m)$ is defined in (\ref{defache}),
934: into the second one. Now we introduce the expansion (\ref{expanu}) of
935: the factor
936: $(1-u)^{-1}$ in both integrals in (\ref{cauchybis}).
937: Using (\ref{ache}) and after straightforward calculations we obtain
938: %
939: \begin{equation}
940: f(z)= \sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^m(z-z_k)^{n}+
941: \sum_{n=0}^{N-1}t_{n,m}(z)\prod_{k=1}^m(z-z_k)^{-n-1}+ r_N(z),
942: \end{equation}
943: %
944: where $q_{n,m}(z)$ is given by formulas (\ref{poly}) and (\ref{coefi})
945: replacing the contour $C$ by $\Gamma_1$. Also,
946: %
947: \begin{equation}
948: t_{n,m}(z)=\sum_{j=1}^m b_{n,j}{\prod_{k=1,k\ne j}^m(z-z_k)
949: \over \prod_{k=1,k\ne j}^m(z_j-z_k)},
950: \hskip 1cm
951: b_{n,j}\equiv{1\over 2\pi i}
952: \int_{\Gamma_2}\prod_{k=1}^m(w-z_k)^{n+1}{f(w)\,dw\over w-z_j},
953: \end{equation}
954: %
955: and
956: \begin{equation}
957: \begin{split}
958: r_N(z)= & {1\over 2\pi i}\int_{\Gamma_1}{f(w)\,dw\over (w-z)
959: \prod_{k=1}^m(w-z_k)^N}\prod_{k=1}^m(z-z_k)^N- \\ &
960: {1\over 2\pi i}\int_{\Gamma_2}\prod_{k=1}^m(w-z_k)^N{f(w)\,dw\over w-z
961: }{1\over\prod_{k=1}^m(z-z_k)^N}.
962: \end{split}
963: \end{equation}
964: Now we write
965: %
966: \begin{equation}
967: t_{n,m}(z)={1\over 2\pi i}
968: \int_{\Gamma_2}\prod_{k=1}^m(w-z_k)^{n+1}f(w)\,dw\sum_{j=1}^m
969: {\prod_{k=1,k\ne j}^m(z-z_k)\over (w-z_j)\prod_{k=1,k\ne j}^m(z_j-z_k)}
970: \end{equation}
971: %
972: and repeat the steps following (\ref{cuco}) in the proof of Theorem \ref{theo2}
973: for $q_{n,m}(z)$ and $t_{n,m}(z)$.
974:
975: For any $z\in A_p$ we can take simple closed loops ${\Gamma}_1$ and
976: ${\Gamma}_2$ in $\Omega\setminus\Omega_0$ such that (see Figure 4 (b))
977: %
978: \begin{equation}
979: \prod_{k=1}^p\vert z-z_k\vert^{m_k}<
980: \prod_{k=1}^p\vert w-z_k\vert^{m_k} \hskip 1cm \forall \hskip 2mm
981: w\in{\Gamma}_1
982: \end{equation}
983: %
984: and
985: %
986: \begin{equation}
987: \prod_{k=1}^p\vert z-z_k\vert^{m_k}>
988: \prod_{k=1}^p\vert w-z_k\vert^{m_k} \hskip 1cm
989: \forall \hskip 2mm w\in{\Gamma}_2.
990: \end{equation}
991: %
992: On these contours $\vert f(w)\vert$ is bounded by some constant $C$: $\vert
993: f(w)\vert\le C$. Introducing these bounds in (\ref{remii}) we see that
994: $\lim_{N\to\infty}r_N(z)=0$ and the proof follows.
995: \end{proof}
996:
997:
998: \begin{figure}[tb]
999: %\blankbox{.6\columnwidth}{5pc}
1000: \begin{center}\includegraphics[width=10cm]{fig4.eps}\end{center}
1001: \centerline{(a) \hskip 6cm (b)}
1002: \caption{The case $p=3$.
1003: (a) Contours $\Gamma_1$ and $\Gamma_2$ in the integrals
1004: (\ref{coefii}), (\ref{coefiii}) and (\ref{remii}).
1005: (b) For $z\in A_{p}$, we can take a contour $\Gamma_2$ in
1006: $\Omega$ located between $\Omega_0$ and $A_{p}$ and a contour
1007: $\Gamma_1$ in $\Omega$ such that $A_{p}$ is inside this contour. Therefore,
1008: $\prod_{k=1}^p\vert z-z_k\vert^{m_k}<\prod_{k=1}^p\vert w-z_k\vert^{m_k}$
1009: $\forall$ $w\in\Gamma_1$ and
1010: $\prod_{k=1}^p\vert w-z_k\vert^{m_k}<\prod_{k=1}^p\vert z-z_k\vert^{m_k}$
1011: $\forall$ $w\in\Gamma_2$.}
1012: \label{Figure 4}
1013: \end{figure}
1014:
1015:
1016: \begin{figure}[tb]
1017: %\blankbox{.6\columnwidth}{5pc}
1018: \begin{center}\includegraphics[width=10cm]{fig51.eps}\end{center}
1019: \centerline{$(r_1^a,r_2^a)$ \hskip 3cm
1020: $(r_1^a,r_2^b)$ \hskip 3cm $(r_1^a,r_2^c)$}
1021: %\blankbox{.6\columnwidth}{5pc}
1022: \bigskip
1023: \begin{center}\includegraphics[width=10cm]{fig52.eps}\end{center}
1024: \centerline{$(r_1^b,r_2^b)$ \hskip 3cm
1025: $(r_1^b,r_2^c)$ \hskip 3cm $(r_1^c,r_2^c)$}
1026: \caption{Shape of the "lemniscate annulus" $A_p$ for $p=3$.
1027: It depends on the relative size of the parameters $r_1$ and $r_2$
1028: defined in (\ref{erei}). The different forms are labeled by $(r_1,r_2)$
1029: with $r_1>r_2$. In these pictures
1030: $\vert z_2-z_3\vert<\vert z_1-z_3\vert$, $\vert z_1-z_2\vert$
1031: and $r_2^a>r_2^b>r_2^c$.}
1032: \label{Figure 5}
1033: \end{figure}
1034:
1035:
1036: If the only singularities of $f(z)$ inside $\Omega_0$ are just poles at
1037: $z_1$, $z_2$,...,$z_p$, then
1038: alternative formulas of (\ref{coefii}) and (\ref{coefiii}) for computing the
1039: coefficients of the above multi-point
1040: Laurent expansion is given in the following proposition.
1041:
1042: \begin{prop}
1043: \label{prop3}
1044: Suppose that $g_k(z)\equiv (z-z_k)^{\rho_k}f(z)$,
1045: $k=1,2,\ldots,p$ are analytic functions in $\Omega$ for certain
1046: $\rho_1$, $\rho_2$,...,$\rho_k\in\Ns$. Then, for $n=0,1,2,\ldots$,
1047: coefficients $a_{n,j,l}$ and $b_{n,j,l}$ in expansion (\ref{expanii}) are
1048: also given by the formulas:
1049: %
1050: \begin{equation}
1051: \label{coefiibis}
1052: \begin{split}
1053: a_{n,j,l}= &
1054: \left.\sum_{k=1,k\ne j}^p
1055: D_w^{nm_k+\rho_k-1}\left[
1056: {g_k(w)\over(w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}\right]
1057: \right\vert_{w=z_k} + \\ &
1058: \quad D_w^{nm_j+\rho_j+l}
1059: \left. \left[
1060: {g_j(w)\over\prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}
1061: \right]\right\vert_{w=z_j}
1062: \end{split}
1063: \end{equation}
1064: %
1065: and
1066: %
1067: \begin{equation}
1068: \label{coefiibisbis}
1069: \begin{split}
1070: b_{n,j,l}= &
1071: \left.\sum_{k=1,k\ne j}^p
1072: D_w^{\rho_k-(n+1)m_k-1}
1073: \left[
1074: {g_k(w)\over(w-z_j)^{l+1}}\prod_{s=1,s\ne k}^p(w-z_s)^{(n+1)m_s}
1075: \right]\right\vert_{w=z_k}+ \\ &
1076: \quad
1077: D_w^{\rho_j-(n+1)m_j+l}\left. \left[
1078: {g_j(w)\prod_{s=1,s\ne j}^p(w-z_s)^{(n+1)m_s}}
1079: \right]\right\vert_{w=z_j}.
1080: \end{split}
1081: \end{equation}
1082: %
1083: \end{prop}
1084:
1085: \begin{proof}
1086: We deform both contours ${\Gamma}_1$ and ${\Gamma}_2$
1087: of equations (\ref{coefii}) and (\ref{coefiii}), respectively, to any
1088: contour of the form $C_1\cup C_2\cup\cdots\cup C_p$
1089: contained in $\Omega$, where
1090: $C_k$, $k=1,2,\ldots,p$ is a simple closed loop which
1091: encircles the point $z_k$ in the counterclockwise
1092: direction and does not contain the point $z_j$ $j=1,2,\ldots,p$, $j\ne k$
1093: inside (see Figure 3 (b)). Then,
1094: %
1095: \begin{equation}
1096: \begin{split}
1097: a_{n,j,l}= & {1\over 2\pi i}
1098: \sum_{k=1,k\ne j}^p\int_{C_k}{g_k(w)\over
1099: (w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
1100: {dw\over (w-z_k)^{nm_k+\rho_k}}+ \\ &
1101: {1\over 2\pi i}\int_{C_j}{g_j(w)\over
1102: \prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}
1103: {dw\over (w-z_j)^{nm_j+\rho_j+l+1}}
1104: \end{split}
1105: \end{equation}
1106: %
1107: and
1108: \begin{equation}
1109: \begin{split}
1110: b_{n,j,l}= & {1\over 2\pi i}
1111: \sum_{k=1,k\ne j}^p\int_{C_k}
1112: {\prod_{s=1,s\ne k}^p(w-z_s)^{(n+1)m_s}\over
1113: (w-z_j)^{l+1}}{g_k(w)dw\over (w-z_k)^{\rho_k-(n+1)m_k}}+ \\ &
1114: {1\over 2\pi i}\int_{C_j}\prod_{s=1,s\ne j}^p(w-z_s)^{(n+1)m_s}
1115: {g_j(w)dw\over (w-z_j)^{\rho_j-(n+1)m_j+l+1}}.
1116: \end{split}
1117: \end{equation}
1118:
1119: From here, equations (\ref{coefiibis}) and (\ref{coefiibisbis}) follow.
1120: \end{proof}
1121:
1122: \begin{remark}
1123: \label{rem2}
1124: Let $z$ be a real or complex variable.
1125: Suppose that $g_k(z)\equiv (z-z_k)^{\rho_k}f(z)$
1126: is $\rho_k-1-$times differentiable at every $z_k$ in $S$
1127: for some $\rho_k\in\Ns$, $k=1,2,\ldots,p$. Define
1128: %
1129: \begin{equation}
1130: g(z)\equiv f(z)-
1131: \sum_{n=0}^{M}t_{n,m}(z)\prod_{k=1}^p(z-z_k)^{-(n+1)m_k},
1132: \end{equation}
1133: %
1134: where $M\equiv\lfloor$Max$\lbrace (\rho_1-1)/m_1,(\rho_2-1)/m_2,\ldots,
1135: (\rho_p-1)/m_p\rbrace\rfloor$ and $t_{n,m}(z)$
1136: is the polynomial defined in (\ref{polyi}) and (\ref{coefiibisbis}).
1137: Then, the thesis of Proposition \ref{prop2}
1138: holds for $f(z)$ replaced by $g(z)$. Moreover, if
1139: $\prod_{k=1}^p(z-z_k)^{\rho_k}f(z)$ is an
1140: analytic function in $\Omega$, then the thesis of Theorem \ref{theo2}
1141: applies to $g(z)$.
1142: \end{remark}
1143:
1144: \section{Multi-point Taylor-Laurent expansions}
1145:
1146: \noindent
1147: For multi-point expansions we have the possibility (that we do not
1148: have in the standard theory) of expanding in Taylor series in some points and
1149: in Laurent series in other points.
1150:
1151:
1152: \begin{theorem}
1153: \label{theo4}
1154: Let $\Omega_0$ and $\Omega$ be closed and open
1155: sets, respectively, of the complex
1156: plane, and $\Omega_0\subset\Omega\subset\Cs$. Let $f(z)$
1157: be an analytic function on $\Omega\setminus\Omega_0$,
1158: $z_1,z_2,\ldots,z_q\in\Omega\setminus\Omega_0$ and
1159: $z_{q+1},z_{q+2},\ldots,z_p\in\Omega_0$ ($q$ points are in
1160: $\Omega\setminus\Omega_0$ and $p-q$ points are in $\Omega_0$). Write
1161: $s\equiv m_1+m_2+\cdots+m_q$. Then, for
1162: $z\in\Omega\setminus\Omega_0$, $f(z)$ has the Taylor-Laurent expansion
1163: %
1164: \begin{equation}
1165: \label{expaniii}
1166: \begin{split}
1167: f(z)= & \sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^p(z-z_k)^{nm_k}+
1168: \sum_{n=0}^{N-1}t_{n,m}^{(1)}(z){\prod_{k=1}^q(z-z_k)^{nm_k}\over
1169: \prod_{k=q+1}^p(z-z_k)^{nm_k}}+ \\ &
1170: \sum_{n=0}^{N-1}t_{n,m}^{(2)}(z){\prod_{k=1}^q(z-z_k)^{(n+1)m_k}\over
1171: \prod_{k=q+1}^p(z-z_k)^{(n+1)m_k}}+ r_N(z),
1172: \end{split}
1173: \end{equation}
1174: %
1175: where $q_{n,m}(z)$ is the polynomial of degree $m-1$
1176: %
1177: \begin{equation}
1178: \label{polybisbis}
1179: q_{n,m}(z)\equiv\sum_{j=1}^p
1180: {\prod_{k=1,k\ne j}^p(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^p(z_j-z_k)^{m_k}}
1181: \sum_{l=0}^{m_j-1}a_{n,j,l}(z-z_j)^l
1182: \end{equation}
1183: %
1184: and the coefficients $a_{n,j,l}$ of this polynomial are given by the
1185: Cauchy integral
1186: %
1187: \begin{equation}
1188: \label{coefiv}
1189: a_{n,j,l}\equiv{1\over 2\pi i}
1190: \int_{\Gamma_1}{f(w)\,dw\over (w-z_j)^{l+1}\prod_{k=1}^p(w-z_k)^{nm_k}}.
1191: \end{equation}
1192: %
1193: Also, $t_{n,m}^{(1)}(z)$ and $t_{n,m}^{(2)}(z)$
1194: are the following polynomials of degrees $s-1$ and $m-s-1$ respectivelly,
1195: %
1196: \begin{equation}
1197: \label{polyii}
1198: t_{n,m}^{(1)}(z)\equiv -\sum_{j=1}^q
1199: {\prod_{k=1,k\ne j}^q(z-z_k)^{m_k}\over \prod_{k=1,k\ne j}^q(z_j-z_k)^{m_k}}
1200: \sum_{l=0}^{m_j-1}b_{n,j,l}(z-z_j)^l,
1201: \end{equation}
1202: %
1203: where the coefficients $b_{n,j,l}$ of this polynomial are given by the
1204: Cauchy integral
1205: %
1206: \begin{equation}
1207: \label{coefv}
1208: b_{n,j,l}\equiv{1\over 2\pi i}
1209: \int_{\Gamma_2}{\prod_{k=q+1}^p(w-z_k)^{nm_k}\over
1210: \prod_{k=1}^q(w-z_k)^{nm_k}}
1211: {f(w)\,dw\over (w-z_j)^{l+1}}.
1212: \end{equation}
1213: %
1214: %
1215: \begin{equation}
1216: \label{polyiii}
1217: t_{n,m}^{(2)}(z)\equiv \sum_{j=q+1}^p
1218: {\prod_{k=q+1,k\ne j}^p(z-z_k)^{m_k}\over
1219: \prod_{k=q+1,k\ne j}^p(z_j-z_k)^{m_k}}
1220: \sum_{l=0}^{m_j-1}c_{n,j,l}(z-z_j)^l,
1221: \end{equation}
1222: %
1223: where the coefficients $c_{n,j,l}$ of this polynomial are given by the
1224: Cauchy integral
1225: %
1226: \begin{equation}
1227: \label{coefvi}
1228: c_{n,j,l}\equiv{1\over 2\pi i}
1229: \int_{\Gamma_2}{\prod_{k=q+1}^p(w-z_k)^{(n+1)m_k}\over
1230: \prod_{k=1}^q(w-z_k)^{(n+1)m_k}}
1231: {f(w)\,dw\over (w-z_j)^{l+1}}.
1232: \end{equation}
1233: %
1234: The remainder term $r_N(z)$ is given by the Cauchy integrals
1235: %
1236: \begin{equation}
1237: \label{remiii}
1238: \begin{split}
1239: r_N(z)\equiv & {1\over 2\pi i}\int_{\Gamma_1}{f(w)\,dw\over (w-z)
1240: \prod_{k=1}^p(w-z_k)^{Nm_k}}\prod_{k=1}^p(z-z_k)^{Nm_k}- \\ &
1241: {1\over 2\pi i}\int_{\Gamma_2}{\prod_{k=q+1}^p(w-z_k)^{Nm_k}\over
1242: \prod_{k=1}^q(w-z_k)^{Nm_k}}{f(w)dw\over w-z
1243: }{\prod_{k=1}^q(z-z_k)^{Nm_k}\over\prod_{k=q+1}^p(z-z_k)^{Nm_k}}.
1244: \end{split}
1245: \end{equation}
1246: %
1247: In these integrals, the contours of integration ${\Gamma}_1$ and ${\Gamma}_2$
1248: are simple closed loops
1249: contained in $\Omega\setminus\Omega_0$ which encircle $\Omega_0$ in the
1250: counterclockwise direction. Moreover,
1251: the points $z$ and $z_1$, $z_2$,...,$z_q$ are not inside ${\Gamma}_2$,
1252: whereas ${\Gamma}_1$ encircles ${\Gamma}_2$ and the points
1253: $z$ and $z_1$, $z_2$,...,$z_q$ (see Figure 6 (a)).
1254:
1255: \noindent
1256: The expansion (\ref{expaniii}) is convergent in the region (Figure 7)
1257: %
1258: \begin{equation}
1259: \label{domainiii}
1260: \begin{split}
1261: D_{q,p}\equiv\biggl\lbrace z\in\Omega\setminus\Omega_0, & \hskip 1mm
1262: \prod_{k=1}^p\vert(z-z_k)\vert^{m_k}<r_1, \\ & \left.
1263: \prod_{k=1}^q\vert(z-z_k)\vert^{m_k}<
1264: r_2\prod_{k=q+1}^p\vert(z-z_k)\vert^{m_k}\right\rbrace
1265: \end{split}
1266: \end{equation}
1267: %
1268: where $r_1\equiv$ Inf$_{w\in\Css\setminus\Omega}\left\lbrace
1269: \prod_{k=1}^p\vert(w-z_k)\vert^{m_k}\right\rbrace$
1270: and
1271:
1272: \noindent
1273: $r_2\equiv$ Inf$_{w\in\Omega_0}\left\lbrace
1274: \prod_{k=1}^q\vert(w-z_k)\vert^{m_k}
1275: \prod_{k=q+1}^p\vert(w-z_k)^{-1}\vert^{m_k}\right\rbrace$.
1276: \end{theorem}
1277:
1278: \begin{proof}
1279: By Cauchy's theorem,
1280: %
1281: \begin{equation}
1282: \label{cauchybisbis}
1283: f(z)={1\over 2\pi i}\int_{{\Gamma}_1}{f(w)\,dw\over w-z}-
1284: {1\over 2\pi i}\int_{{\Gamma}_2}{f(w)\,dw\over w-z},
1285: \end{equation}
1286: %
1287: where ${\Gamma}_1$ and ${\Gamma}_2$ are the contours defined above.
1288:
1289: First we assume that the $m$ points of the set
1290: $S$ are all distinct. Later we will let the first $m_1$
1291: points coalesce to $z_1$, the second $m_2$ points to
1292: $z_2$, and so on. We
1293: substitute (\ref{lequal})-(\ref{ui}) into the first integral above and
1294: %
1295: \begin{equation}
1296: {1\over w-z}={F_m(w,z;z_1,\ldots,z_m)\over
1297: \prod_{k=1}^s(w-z_k)\prod_{k=s+1}^m(z-z_k)}{1\over 1-u},
1298: \end{equation}
1299: %
1300: where
1301: %
1302: \begin{equation}
1303: u\equiv {\prod_{k=1}^s(z-z_k)\prod_{k=s+1}^m(w-z_k)\over
1304: \prod_{k=1}^s(w-z_k)\prod_{k=s+1}^m(z-z_k)}
1305: \end{equation}
1306: %
1307: and
1308: %
1309: \begin{equation}
1310: \begin{split}
1311: F_m(w,z;z_1,\ldots,z_m)\equiv {1\over w-z}&\left[
1312: \prod_{k=1}^s(w-z_k)\prod_{k=s+1}^m(z-z_k)- \right. \\ & \left.
1313: \prod_{k=1}^s(z-z_k)\prod_{k=s+1}^m(w-z_k)\right]
1314: \end{split}
1315: \end{equation}
1316: %
1317: into the second one. Next we introduce the expansion (\ref{expanu})
1318: of the factor
1319: $(1-u)^{-1}$ in both integrals in (\ref{cauchybisbis}). We observe that
1320: $F_m(w,z;z_1,\ldots,z_m)$ may be written as
1321: %
1322: \begin{equation}
1323: \begin{split}
1324: F_m(w,z;z_1,\ldots,z_m)=
1325: &H_s(w,z;z_1,\ldots,z_p)\prod_{k=s+1}^m(z-z_k) -\\
1326: &H_{m-s}(w,z;z_{p+1},\ldots,z_m)\prod_{k=1}^s(z-z_k),
1327: \end{split}
1328: \end{equation}
1329: %
1330: where $H_m(w,z;z_1,\ldots,z_m)$ is defined in (\ref{defache}).
1331: Using this decomposition,
1332: equation (\ref{ache}), and after straightforward calculations we obtain
1333: %
1334: \begin{equation}
1335: \begin{split}
1336: f(z)= & \sum_{n=0}^{N-1}q_{n,m}(z)\prod_{k=1}^m(z-z_k)^{n}+
1337: \sum_{n=0}^{N-1}t_{n,m}^{(1)}(z){\prod_{k=1}^s(z-z_k)^{n}
1338: \over\prod_{k=s+1}^m(z-z_k)^{n}}+ \\ &
1339: \sum_{n=0}^{N-1}t_{n,m}^{(2)}(z){\prod_{k=1}^s(z-z_k)^{n+1}
1340: \over\prod_{k=s+1}^m(z-z_k)^{n+1}}+ r_N(z),
1341: \end{split}
1342: \end{equation}
1343: %
1344: where $q_{n,m}(z)$ is given by formulas (\ref{poly}) and (\ref{coefi})
1345: replacing the contour $C$ by $\Gamma_1$. Also,
1346: %
1347: \begin{equation}
1348: t_{n,m}^{(1)}(z)=-\sum_{j=1}^s b_{n,j}{\prod_{k=1,k\ne j}^s(z-z_k)
1349: \over \prod_{k=1,k\ne j}^s(z_j-z_k)}
1350: \end{equation}
1351: %
1352: with
1353: %
1354: \begin{equation}
1355: b_{n,j}\equiv{1\over 2\pi i}
1356: \int_{\Gamma_2}{\prod_{k=s+1}^m(w-z_k)^n\over
1357: \prod_{k=1}^s(w-z_k)^n}{f(w)\,dw\over w-z_j},
1358: \end{equation}
1359: %
1360: %
1361: \begin{equation}
1362: t_{n,m}^{(2)}(z)=\sum_{j=s+1}^m c_{n,j}{\prod_{k=s+1,k\ne j}^m(z-z_k)
1363: \over \prod_{k=s+1,k\ne j}^m(z_j-z_k)}
1364: \end{equation}
1365: %
1366: with
1367: %
1368: \begin{equation}
1369: c_{n,j}\equiv{1\over 2\pi i}
1370: \int_{\Gamma_2}{\prod_{k=s+1}^m(w-z_k)^{n+1}\over
1371: \prod_{k=1}^s(w-z_k)^{n+1}}{f(w)\,dw\over w-z_j}
1372: \end{equation}
1373: %
1374: and
1375: %
1376: \begin{equation}
1377: \begin{split}
1378: r_N(z)= & {1\over 2\pi i}\int_{\Gamma_1}{f(w)\,dw\over (w-z)
1379: \prod_{k=1}^m(w-z_k)^{N}}\prod_{k=1}^m(z-z_k)^{N}- \\ &
1380: {1\over 2\pi i}\int_{\Gamma_2}{\prod_{k=s+1}^m(w-z_k)^{N}
1381: \over\prod_{k=1}^s(w-z_k)^{N}}{f(w)\,dw\over w-z
1382: }{\prod_{k=1}^s(z-z_k)^{N}\over\prod_{k=s+1}^m(z-z_k)^{N}}.
1383: \end{split}
1384: \end{equation}
1385: %
1386: Now we write
1387: %
1388: \begin{equation}
1389: \begin{split}
1390: t_{n,m}^{(1)}(z)
1391: &=-{1\over 2\pi i}
1392: \int_{\Gamma_2}{\prod_{k=s+1}^m(w-z_k)^n
1393: \over\prod_{k=1}^s(w-z_k)^n}
1394: \sum_{j=1}^s
1395: {f(w)\prod_{k=1,k\ne j}^s(z-z_k)\over (w-z_j)\prod_{k=1,k\ne j}^s(z_j-z_k)}
1396: \,dw,\\
1397: t_{n,m}^{(2)}(z)
1398: &={1\over 2\pi i}
1399: \int_{\Gamma_2}{\prod_{k=s+1}^m(w-z_k)^{n+1}
1400: \over\prod_{k=1}^s(w-z_k)^{n+1}}
1401: \sum_{j=s+1}^m
1402: {f(w)\prod_{k=s+1,k\ne j}^m(z-z_k)\over (w-z_j)\prod_{k=s+1,k\ne j}^m(z_j-z_k)}
1403: \,dw,
1404: \end{split}
1405: \end{equation}
1406: %
1407: and repeat the steps following (\ref{cuco}) in Theorem \ref{theo2}
1408: for $q_{n,m}(z)$,
1409: $t_{n,m}^{(1)}(z)$ and $t_{n,m}^{(2)}(z)$.
1410:
1411: For any $z$ verifying (\ref{domainiii}), we can take simple closed loops
1412: ${\Gamma}_1$ and ${\Gamma}_2$ in $\Omega\setminus\Omega_0$ such that
1413: (see Figure 6 (b))
1414: %
1415: \begin{equation}
1416: \prod_{k=1}^p\vert z-z_k\vert^{m_k}<
1417: \prod_{k=1}^p\vert w-z_k\vert^{m_k} \hskip 1cm \forall \hskip 2mm
1418: w\in{\Gamma}_1
1419: \end{equation}
1420: %
1421: and
1422: %
1423: \begin{equation}
1424: \prod_{k=1}^q\vert w-z_k\vert^{m_k}
1425: \prod_{k=q+1}^p\vert z-z_k\vert^{m_k}>
1426: \prod_{k=1}^q\vert z-z_k\vert^{m_k}
1427: \prod_{k=q+1}^p\vert w-z_k\vert^{m_k}
1428: \end{equation}
1429: %
1430: $\forall$ $w\in{\Gamma}_2$.
1431: On these contours $\vert f(w)\vert$ is bounded by some constant $C$: $\vert
1432: f(w)\vert\le C$. Introducing these bounds in (\ref{remiii}) we see that
1433: $\lim_{N\to\infty}r_N(z)=0$ and the proof follows.
1434: \end{proof}
1435:
1436: \begin{figure}[tb]
1437: %\blankbox{.6\columnwidth}{5pc}
1438: \begin{center}\includegraphics[width=10cm]{fig6.eps}\end{center}
1439: \centerline{(a) \hskip 6cm (b)}
1440: \caption{The case $q=2$, $p=3$.
1441: (a) Contours $\Gamma_1$ and $\Gamma_2$ in the integrals
1442: (\ref{coefiv}), (\ref{coefv}), (\ref{coefvi}) and (\ref{remiii}).
1443: (b) For $z\in D_{q,p}$, we can take a contour $\Gamma_2$
1444: located between $\Omega_0$ and $D_{q,p}$ and a contour $\Gamma_1$
1445: in $\Omega$ with $D_{q,p}$ inside this contour.
1446: Therefore,
1447: $\prod_{k=1}^p\vert z-z_k\vert^{m_k}<\prod_{k=1}^p\vert w-z_k\vert^{m_k}$
1448: $\forall$ $w\in\Gamma_1$ and
1449: $\prod_{k=1}^q\vert z-z_k\vert^{m_k}\prod_{k=q+1}^p\vert w-z_k\vert^{m_k}<
1450: \prod_{k=1}^q\vert w-z_k\vert^{m_k}\prod_{k=q+1}^p\vert z-z_k\vert^{m_k}$
1451: $\forall$ $w\in\Gamma_2$.}
1452: \label{Figure 6}
1453: \end{figure}
1454:
1455:
1456: \begin{figure}[tb]
1457: %\blankbox{.6\columnwidth}{5pc}
1458: \begin{center}\includegraphics[width=11cm]{fig71.eps}\end{center}
1459: \centerline{$(r_1^a,r_2^a)$ \hskip 1.8cm
1460: $(r_1^a,r_2^b)$ \hskip 1.8cm $(r_1^a,r_2^c)$ \hskip 1.8cm
1461: $(r_1^a,r_2^d)$}
1462: \bigskip
1463: \begin{center}\includegraphics[width=11cm]{fig72.eps}\end{center}
1464: \centerline{$(r_1^b,r_2^a)$ \hskip 1.8cm
1465: $(r_1^b,r_2^b)$ \hskip 1.8cm $(r_1^b,r_2^c)$ \hskip 1.8cm
1466: $(r_1^b,r_2^d)$}
1467: \bigskip
1468: \begin{center}\includegraphics[width=11cm]{fig73.eps}\end{center}
1469: \centerline{$(r_1^c,r_2^a)$ \hskip 1.8cm
1470: $(r_1^c,r_2^b)$ \hskip 1.8cm $(r_1^c,r_2^c)$ \hskip 1.8cm
1471: $(r_1^c,r_2^d)$}
1472: \caption{The region $D_{q,p}$ defined in Theorem \ref{theo4} is given by
1473: $D_{q,p}=O_p\bigcap B_{q,p}$, where $O_p$
1474: is the "lemniscate domain" of foci $z_1$, $\ldots$, $z_p$ and parameter
1475: $r_1$. Also,
1476: $B_{q,p}\equiv \lbrace z\in\Cs$, $\prod_{k=1}^q\vert(z-z_k)\vert^{m_k}<
1477: r_2\prod_{k=q+1}^p\vert(z-z_k)\vert^{m_k}\rbrace$. This pictures show the
1478: topologically different forms of $D_{q,p}$ depending on the
1479: relative value of $r_1$ and $r_2$ when $q=2$ and $p=3$. The pictures are
1480: labeled with $(r_1,r_2)$. In these pictures $z_1$
1481: $\vert z_1-z_2\vert<\vert z_1-z_3\vert$, $\vert z_2-z_3\vert$
1482: and $r_2^a>r_2^b>r_2^c>r_2^d$.}
1483: \label{Figure 7}
1484: \end{figure}
1485:
1486:
1487: If the only singularities of $f(z)$ inside $\Omega_0$ are just poles at
1488: $z_{q+1}$, $z_{q+2}$,...,$z_p$, then alternative formulas of
1489: (\ref{coefiv}), (\ref{coefv}) and (\ref{coefvi}) for computing the coefficients
1490: of the above two-point
1491: Taylor-Laurent expansion is given in the following proposition.
1492:
1493: \begin{prop}
1494: \label{prop4}
1495: Suppose that $g_k(z)\equiv
1496: (z-z_k)^{\rho_k}f(z)$ is an
1497: analytic function in $\Omega$ for certain $\rho_k\in\Ns$ and
1498: $k=q+1,q+2,\ldots,p$.
1499: Define $g_k(w)=f(w)$ for $k=1,2,3,\ldots q$.
1500: Then the coefficients $a_{n,j,l}$, $b_{n,j,l}$ and $c_{n,j,l}$ in the expansion
1501: (\ref{expaniii}) are also given by the formulas:
1502: %
1503: \begin{equation}
1504: \label{coefvx}
1505: \begin{split}
1506: a_{n,j,l}= &
1507: \left.\sum_{k=1,k\ne j}^q
1508: D_w^{nm_k-1}\left[
1509: {f(w)\over(w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
1510: \right]\right\vert_{w=z_k}+ \\ &
1511: \quad
1512: \left.\sum_{k=q+1,k\ne j}^p
1513: D_w^{nm_k+\rho_k-1}\left[
1514: {g_k(w)\over(w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
1515: \right]\right\vert_{w=z_k}+ \\ &
1516: \quad
1517: D_w^{nm_j+\rho_j+l}\left.
1518: \left[
1519: {g_j(w)\over\prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}\right]\right\vert_{w=z_j},
1520: \end{split}
1521: \end{equation}
1522: %
1523: %
1524: \begin{equation}
1525: \label{coefiiibis}
1526: b_{n,j,l}=
1527: \left.\sum_{k=q+1}^p
1528: D_w^{\rho_k-nm_k-1}\left[
1529: {g_k(w)\prod_{s=q+1,s\ne k}^p(w-z_s)^{nm_s}\over(w-z_j)^{l+1}
1530: \prod_{s=1}^q(w-z_s)^{nm_s}}\right]\right\vert_{w=z_k}.
1531: \end{equation}
1532: %
1533: %
1534: \begin{equation}
1535: \label{coefiiibisbis}
1536: \begin{split}
1537: c_{n,j,l}= &
1538: \left.\sum_{k=q+1,k\ne j}^p
1539: D_w^{\rho_k-(n+1)m_k-1}\left[
1540: {g_k(w)\prod_{s=q+1,s\ne k}^p(w-z_s)^{(n+1)m_s}\over(w-z_j)^{l+1}
1541: \prod_{s=1}^q(w-z_s)^{(n+1)m_s}}\right]
1542: \right\vert_{w=z_k}+ \\ &
1543: \quad
1544: D_w^{\rho_j-(n+1)m_j+l}\left. \left[
1545: {g_j(w)\prod_{s=q+1,s\ne j}^p(w-z_s)^{(n+1)m_s}
1546: \over\prod_{s=1}^q(w-z_s)^{(n+1)m_s}}\right]
1547: \right\vert_{w=z_j}.
1548: \end{split}
1549: \end{equation}
1550: %
1551: \end{prop}
1552:
1553: \begin{proof}
1554: We deform both contours ${\Gamma}_1$ in equation (\ref{coefiv}) and
1555: the contour ${\Gamma}_2$ in equations (\ref{coefvx}) and (\ref{coefvi})
1556: into any contour of the form
1557: ${C_1}\cup{C_2}\cup\cdots\cup{C_p}$ contained in
1558: $\Omega$, where ${C_k}$, $k=1,2,...,p$ is a simple closed loop
1559: which encircles the point $z_k$ in the counterclockwise
1560: direction with $z_j$ not inside ${C_k}$,
1561: $j=1,2,\ldots,p$, $j\ne k$ (see Figure 3 (c)). Then,
1562: %
1563: \begin{equation}
1564: \begin{split}
1565: a_{n,j,l}= & {1\over 2\pi i}
1566: \sum_{k=1,k\ne j}^p\int_{C_k}{g_k(w)\over
1567: (w-z_j)^{l+1}\prod_{s=1,s\ne k}^p(w-z_s)^{nm_s}}
1568: {dw\over (w-z_k)^{nm_k+\rho_k}}+ \\ &
1569: \int_{C_j}{g_j(w)\over\prod_{s=1,s\ne j}^p(w-z_s)^{nm_s}}
1570: {dw\over (w-z_j)^{nm_j+\rho_j+l+1}},
1571: \end{split}
1572: \end{equation}
1573: %
1574: %
1575: \begin{equation}
1576: b_{n,j,l}= {1\over 2\pi i}
1577: \sum_{k=q+1}^p\int_{C_k}
1578: {\prod_{s=q+1,s\ne k}^p(w-z_s)^{nm_s}\over
1579: (w-z_j)^{l+1}\prod_{s=1}^q(w-z_s)^{nm_s}}{g_k(w)dw\over (w-z_k)^{\rho_k-nm_k}},
1580: \end{equation}
1581: %
1582: %
1583: \begin{equation}
1584: \begin{split}
1585: c_{n,j,l}= &
1586: \sum_{k=q+1,k\ne j}^p{1\over 2\pi i}\int_{C_k}
1587: {\prod_{s=q+1,s\ne k}^p(w-z_s)^{(n+1)m_s}\over
1588: (w-z_j)^{l+1}\prod_{s=1}^q(w-z_s)^{(n+1)m_s}}
1589: {g_k(w)dw\over (w-z_k)^{\rho_k-(n+1)m_k}}+ \\ & {1\over 2\pi i}
1590: \int_{C_j}{\prod_{s=q+1,s\ne j}^p(w-z_s)^{(n+1)m_s}\over
1591: \prod_{s=1}^q(w-z_s)^{(n+1)m_s}}{g_j(w)
1592: dw\over (w-z_j)^{\rho_j-(n+1)m_j+l+1}}.
1593: \end{split}
1594: \end{equation}
1595: %
1596: From here, equations (\ref{coefvx}), (\ref{coefiiibis}) and
1597: (\ref{coefiiibisbis})
1598: follow.
1599: \end{proof}
1600:
1601: \begin{remark}
1602: \label{rem3}
1603: Let $z$ be a real or complex variable and
1604: suppose that $(z-z_k)^{\rho_k}f(z)$
1605: is $\rho_k-1-$times differentiable at $z_k$ for certain $\rho_k\in\Ns$.
1606: Define
1607: %
1608: \begin{equation}
1609: g(z)\equiv f(z)-\sum_{n=0}^{M}t_{n,m}^{(1)}(z)
1610: {\prod_{k=1}^q(z-z_k)^{nm_k}\over
1611: \prod_{k=q+1}^p(z-z_k)^{nm_k}}-
1612: \sum_{n=0}^{M}t_{n,m}^{(2)}(z){\prod_{k=1}^q(z-z_k)^{(n+1)m_k}\over
1613: \prod_{k=q+1}^p(z-z_k)^{(n+1)m_k}},
1614: \end{equation}
1615: %
1616: where $M\equiv\lfloor$Max$\lbrace
1617: (\rho_{q+1}-1)/m_{q+1},(\rho_{q+2}-1)/m_{q+2},\ldots,
1618: (\rho_p-1)/m_p\rbrace\rfloor$ and
1619: $t_{n,m}^{(1)}(z)$ and $t_{n,m}^{(2)}(z)$
1620: are the polynomials defined in (\ref{polyii}), (\ref{polyiii}),
1621: (\ref{coefiiibis}) and (\ref{coefiiibisbis}).
1622: Then, the thesis of Proposition \ref{prop2}
1623: holds for $f(z)$ replaced by $g(z)$.
1624:
1625: Moreover, if
1626: $\prod_{k=q+1}^p(z-z_k)^{\rho_k}f(z)$ is an
1627: analytic function in $\Omega$, then the thesis of Theorem \ref{theo2}
1628: applies to $g(z)$.
1629: \end{remark}
1630:
1631:
1632:
1633: \section{Discussion and concluding remarks}
1634:
1635:
1636:
1637: In an earlier paper \cite{nicoii} we have discussed the theory
1638: of two-point Taylor expansions, two-point Laurent
1639: expansions and two-point Taylor-Laurent expansions. In
1640: the present paper we have generalized these two-point
1641: cases to multi-point cases. We have given details on the
1642: regions of convergence and on representations of the
1643: coefficients and the remainders of the expansions in terms
1644: of Cauchy-type integrals.
1645:
1646: Multi-point Taylor expansions are related with topics from
1647: interpolation theory, in particular with the Newton
1648: interpolation theory with applications in numerical
1649: analysis. For example, applications can be found in
1650: initial and boundary value problems in connection with
1651: ordinary differential equations and in numerical
1652: quadrature of integrals.
1653:
1654: From the point of view of interpolation theory detailed
1655: information on multi-point expansions can be found in
1656: \cite{walsh}, Chapters 3 and 8. The theory of several-point
1657: Taylor expansions is discussed in Chapter 3 of \cite{walsh},
1658: although in a setting that is different from our approach.
1659: Our approach gives explicit Cauchy-type integrals of
1660: coefficients and remainders which cannot be found in
1661: Walsh's approach. In particular, we cannot find explicit
1662: formulas for the polynomials $q_{n,m}(z)$ of formula (15)
1663: as we have in (16)-(17). Knowledge of these explicit
1664: formulas is necessary to construct asymptotic expansions
1665: of integrals with several saddle points.
1666:
1667: In addition to this, our Laurent and Taylor-Laurent
1668: expansions are new. They have a formal similarity with
1669: the rational approximations of Chapter 8 of \cite{walsh}: they
1670: involve negative powers of $z$. But they are completely
1671: different. The rational approximations, in particular the
1672: Pad\'e-type approximations $P_n(z)/Q_m(z)$ are of
1673: interpolatory type. These are generalizations of the
1674: Taylor polynomial at several points: a quotient of
1675: polynomials instead of a polynomial. However, our
1676: expansions (21) or (32) have a different form and a
1677: different approximation property: they approach not only
1678: at regular points like Pad\'e-type approximations but also
1679: at singular points of $f(z)$. And of course, the regions
1680: and convergence properties in \cite{walsh} are different from
1681: ours.
1682:
1683: Apart from applying the present results in problems from
1684: interpolation theory, in particular in problems from
1685: numerical analysis, we expect to find applications in
1686: asymptotic analysis of integrals, which application area
1687: is our main motivation; see [5]. In that paper certain
1688: orthogonal polynomials have been considered and we have
1689: given new convergent expansions that also have an
1690: asymptotic property for large values of a parameter (the
1691: degree $n$ of the polynomials). Orthogonal polynomials
1692: and special functions can be studied when the variable
1693: and several parameters are large. In that case more
1694: than one or two so-called critical points occur that may
1695: give the main contributions to the integral, and
1696: expansions of analytic functions at these points gives
1697: again the possibility of constructing new convergent
1698: expansions with an asymptotic property. This method
1699: avoids the complicated conformal mapping of the phase
1700: function of the integral into a standard form (say a cubic
1701: or higher polynomial). In addition, when the critical
1702: points are multiple poles, Laurent-type expansions may be
1703: considered. A few application areas are mentioned in the
1704: Introduction, see the integral in (\ref{besel}), which we expect
1705: to approximate in terms of Airy functions and the Pearcey
1706: integral (\ref{pearcy}) and its derivative with respect to
1707: $x$ and $y$.
1708:
1709:
1710:
1711:
1712:
1713: \bibliographystyle{amsplain}
1714: \begin{thebibliography}{10}
1715:
1716:
1717: \bibitem {Chester} C. Chester, B. Friedman, and F. Ursell,
1718: \textit{An extension of the method of steepest descent.}
1719: Proc. Cambridge Philos. Soc.
1720: \textbf{53} (1957), 599--611.
1721:
1722: \bibitem {Driver} K.A. Driver and N.M. Temme,
1723: \textit{On polynomials related with Hermite-Pad\'e
1724: approximations to the exponential function.}
1725: J. Approx. Theory.
1726: \textbf{65} (1998), 101--122.
1727:
1728: \bibitem {Janssen} A.J.E.M. Janssen,
1729: \textit{On the asymptotics of some Pearcey-type integrals.}
1730: J. Phys. A: Math. Gen.
1731: \textbf{25} (1992), L823--L831.
1732:
1733: \bibitem {Kirk} N.P. Kirk, J.N.L. Connor, P.R. Curtis and C.A. Hobbs,
1734: \textit{Theory of axially symmetric cusped focusing: numerical evaluation of a
1735: Bessoid integral by an adaptive contour algorithm.}
1736: J. Phys. A: Math. Gen.
1737: \textbf{33} (2000), 4797--4808.
1738:
1739: \bibitem {nicoi} Jos\'e L. Lopez and Nico M. Temme,
1740: \textit{Asymptotic expansions of Charlier, Laguerre and Jacobi polynomials.}
1741: Accepted for publication in
1742: The Proceedings of the Royal Society of Edinburgh A (Mathematics).
1743:
1744: \bibitem {nicoii} Jos\'e L. Lopez and Nico M. Temme,
1745: \textit{Two-point Taylor expansions of analytic functions.}
1746: Stud. Appl. Math.
1747: \textbf{109} (2002), 297-311.
1748:
1749: \bibitem {Raimundas} Raimundas Vidunas and Nico M. Temme,
1750: \textit{Symbolic evaluation of coefficients in Airy-type asymptotic
1751: expansions.}
1752: J. Math. Anal. Appl.
1753: \textbf{269} (2002), 317-331.
1754:
1755:
1756: \bibitem{walsh} J. L. Walsh,
1757: \textit{Interpolation and Approximation by rational functions in
1758: the complex domain},
1759: Amer. Math. Soc., Providence, RI, 1969.
1760:
1761:
1762: \bibitem{wong} R. Wong,
1763: \textit{Asymptotic Approximations of Integrals},
1764: Academic Press, New York, 1989.
1765:
1766:
1767: \end{thebibliography}
1768:
1769: \end{document}
1770:
1771: %------------------------------------------------------------------------------
1772: % End of journal.tex
1773: %------------------------------------------------------------------------------
1774: