1: \magnification=\magstephalf
2: \input lopez0
3: %\overfullrule=0pt
4:
5: \input references
6:
7: %\bigskip
8: %\bigskip\bigskip
9:
10: \tfont
11: \centerline {Two-point Taylor Expansions of Analytic Functions}
12: \bigskip\bigskip %\bigskip
13: \afont
14: \cl{Jos\'e L. L\'opez$^1$ and Nico M. Temme$^2$}
15: \medskip
16: \cfont
17: \cl{$^1$
18: Departamento de Mat\'ematica e Inform\'atica,}
19: \cl{Universidad P\'ublica de Navarra,
20: 31006-Pamplona,
21: Spain
22: }
23: \cl{$^2$
24: CWI,
25: P.O. Box 94079,
26: 1090 GB Amsterdam,
27: The Netherlands}
28: \cfont
29: \cl{e-mail: \tt jl.lopez@unavarra.es, nicot@cwi.nl}
30: \medskip
31: %\centerline {}
32:
33:
34: \bfont
35: \parindent=25pt
36: {\pfont ABSTRACT}\par\noindent
37: {\narrower\noindent
38: Taylor expansions of analytic functions are considered with respect to two
39: points. Cauchy-type formulas are given for coefficients and remainders in
40: the expansions, and the regions of convergence are indicated. It is
41: explained how these expansions can be used in deriving uniform asymptotic
42: expansions of integrals. The method is also used for obtaining Laurent
43: expansions in two points.
44: %
45: %
46:
47:
48:
49:
50: \vskip 0.3cm \noindent
51: \cfont
52: 2000 Mathematics Subject Classification:
53: %\par\noindent
54: \bfont
55: 30B10, 30E20, 40A30.
56: \par\noindent
57: \cfont
58: Keywords \& Phrases:
59: %\par\noindent
60: \bfont
61: two-point Taylor expansions, Cauchy's theorem,
62: analytic functions,
63: two-point Laurent expansions,
64: uniform asymptotic expansions of integrals.
65:
66: \par\noindent
67: {\parindent 35pt
68: \cfont
69: \item{Note:\quad}
70: \bfont Work carried out under project MAS1.2
71: Analysis, Asymptotics and Computing.
72: This report has been accepted for publication in
73: {\sl Studies in Applied Mathematics}.
74: \par
75: %\bigskip
76: }
77: }
78: \rm
79: \parindent=15pt
80: %
81: %
82: \sect{Introduction}
83: %
84:
85: \noindent
86: In deriving uniform asymptotic expansions of a certain class of integrals
87: one encounters the problem of expanding a function, that is analytic in some
88: domain $\Omega$ of the complex plane, in two points. The first mention
89: of the use of such expansions in asymptotics is given in \ref\Chester,
90: where Airy-type expansions are derived for integrals having two
91: nearby (or coalescing) saddle points. This reference does not give further
92: details about two-point Taylor expansions, because the coefficients in the
93: Airy-type asymptotic expansion are derived in a different way.
94:
95: To demonstrate the application in asymptotics we consider the integral
96: %
97: \eqn\CFU{
98: F_b(\omega)={1\over 2\pi i}\int _{\C} e^{\omega({1\over3} z^3-b^2z)} f(z)\,dz,
99: }
100: where $\omega$ is a large positive parameter and $b$ is a parameter that may
101: assume small
102: values. The contour starts at $\infty e^{-i\pi/3}$ and terminates at
103: $\infty e^{i\pi/3}$, and lies in a domain where the function $f$ is
104: analytic.
105: In particular, $f$ is analytic in a domain that contains the saddle
106: points $\pm b$ of the exponent in the integrand. One method for obtaining an
107: asymptotic expansion of $F_b(\omega)$ that holds uniformly for small values of
108: $b$ is based on expanding $f$ at the two saddle points:
109: %
110: \eqn\ftexp{
111: f(z)=\sum_{n=0}^\infty A_n(z^2-b^2)^n + z \sum_{n=0}^\infty
112: B_n(z^2-b^2)^n,
113: }
114: and substitute this expansion into \CFU. When interchanging summation
115: and integration, the result is a formal expansion in two series in
116: terms of functions related with Airy functions. A Maple algorithm for
117: obtaining the coefficients $A_n$ and $B_n$, with applications to
118: Airy-type expansions of parabolic cylinder functions, is given in
119: \ref\Raimundas.
120:
121: In a future paper we use expansions like \ftexp\ in order to derive
122: convergent expansions for orthogonal polynomials and hypergeometric
123: functions that also have an asymptotic nature. The purpose of the
124: present paper is to give details on the two-point Taylor expansion
125: \ftexp, in particular on the region of convergence and on
126: representations in terms of Cauchy-type integrals of coefficients and
127: remainders of these expansions. Some information on this type of
128: expansions is also given in \ref\Whittaker, p. 149,
129: Exercise 24.
130:
131: Without referring to applications in asymptotic analysis we include
132: analogous properties of two-point Laurent expansions and of another
133: related type, the two-point Taylor-Laurent expansion.
134:
135: %
136: \sect{Two-point Taylor expansions}
137: %
138:
139: \noindent
140: We consider the expansion \ftexp\ in a more symmetric form and give
141: information on
142: the coefficients and the remainder in the expansion.
143:
144: \noindent
145: {\bf Theorem 1.} {\it Let $f(z)$ be an analytic function on an open set
146: $\Omega\subset\Cs$
147: and $z_1$, $z_2\in\Omega$ with $z_1\ne z_2$. Then, $f(z)$ admits the two-point
148: Taylor expansion
149: %
150: \eqn\expani{
151: f(z)=\sum_{n=0}^{N-1}\left\lbrack
152: a_n(z_1,z_2)(z-z_1)+a_n(z_2,z_1)(z-z_2)\right\rbrack
153: (z-z_1)^n(z-z_2)^n+r_N(z_1,z_2;z),
154: }
155: %
156: where the coefficients $a_n(z_1,z_2)$ and $a_n(z_2,z_1)$ of the expansion
157: are given by the
158: Cauchy integral
159: %
160: \eqn\coefi{
161: a_n(z_1,z_2)\equiv{1\over 2\pi i(z_2-z_1)}\int_{\cal C}{f(w)\,dw
162: \over (w-z_1)^n(w-z_2)^{n+1}}.
163: }
164: %
165: The remainder term $r_N(z_1,z_2;z)$ is given by the Cauchy integral
166: %
167: \eqn\remi{
168: r_N(z_1,z_2;z)\equiv{1\over 2\pi i}\int_{\cal C}{f(w)\,dw\over
169: (w-z_1)^N(w-z_2)^N
170: (w-z)}(z-z_1)^N(z-z_2)^N.
171: }
172: %
173: The contour of integration ${\cal C}$
174: is a simple closed loop which encircles the points $z_1$ and $z_2$ (for
175: $a_n$) and $z$, $z_1$ and $z_2$ (for $r_N$) in
176: the counterclockwise direction and is contained in $\Omega$ (see
177: Figure 1 (a)).
178:
179: The expansion \expani\ is convergent for $z$ inside the Cassini oval (see
180: Figure 2)
181: $$
182: O_{z_1,z_2}\equiv\lbrace z\in\Omega, \hskip 2mm
183: \vert(z-z_1)(z-z_2)\vert<r\rbrace
184: $$
185: where
186: $$
187: r\equiv {\rm Inf}_{w\in\Css\setminus\Omega}\left\lbrace
188: \vert(w-z_1)(w-z_2)\vert\right\rbrace.
189: $$
190: In particular, if $f(z)$ is an
191: entire function
192: $(\Omega=\Cs$), then the
193: expansion \expani\ converges $\forall$ $z\in\Cs$.
194: }
195:
196: \noindent
197: {\bf Proof.} By Cauchy's theorem,
198: %
199: \eqn\cauchy{
200: f(z)={1\over 2\pi i}\int_{\cal C}{f(w)\,dw\over w-z},
201: }
202: %
203: where ${\cal C}$ is the contour defined above (Figure 1 (a)). Now we write
204: %
205: \eqn\lequal{
206: {1\over w-z}={z+w-z_1-z_2\over (w-z_1)(w-z_2)}{1\over 1-u},
207: }
208: %
209: where
210: %
211: \eqn\ui{
212: u\equiv {(z-z_1)(z-z_2)\over (w-z_1)(w-z_2)}.
213: }
214: %
215: Now we introduce the expansion
216: %
217: \eqn\expanu{
218: {1\over 1-u}=\sum_{n=0}^{N-1}u^n+{u^N\over 1-u}
219: }
220: %
221: in \lequal\ and this in \cauchy. After straightforward calculations we
222: obtain \expani-\remi.
223:
224: For any $z\in O_{z_1,z_2}$, we can take a contour $\cal C$ in $\Omega$ such
225: that
226: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$ $w\in{\cal
227: C}$ (see Figure 1 (b)).
228: In this contour $\vert f(w)\vert$ is bounded by some constant $C$: $\vert
229: f(w)\vert\le C$.
230: Introducing these two bounds in \remi\ we see that
231: $\lim_{N\to\infty}r_N(z_1,z_2;z)=0$ and
232: the proof follows.
233: \hfill $\boxe$
234:
235: %
236: \bigskip
237: %
238: \centerline{\epsfxsize=8cm \epsfbox{fig1.adb.eps}}
239: %
240: \centerline{(a) \hskip 6cm (b)}
241: \parindent=10pt
242:
243: \noindent
244: {\bf Figure 1}. {\cfont (a) Contour $\cal C$ in the integrals
245: \expani-\remi. (b) For
246: $z\in O_{z_1,z_2}$, we can take a contour $\cal C$ in $\Omega$ which contains
247: $O_{z_1,z_2}$ inside and therefore,
248: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$ $w\in{\cal C}$.}
249: \vskip 2mm
250:
251: %
252: \rm
253: \parindent=15pt
254:
255: %
256: \bigskip
257: %
258: \centerline{\epsfxsize=10cm \epsfbox{fig2.adb.eps}}
259: %
260: \centerline{(a) $4r>\vert z_1-z_2\vert^2$ \hskip 1.5cm (b) $4r=\vert
261: z_1-z_2\vert^2$
262: \hskip 1.5cm (c) $4r<\vert z_1-z_2\vert^2$}
263: \parindent=10pt
264:
265: \noindent
266: {\bf Figure 2}. {\cfont Shape of the Cassini oval depending on the relative
267: size of the parameter $r$ and the focal distance $\vert z_1-z_2\vert$.}
268: \vskip 2mm
269:
270: %
271: \subsect{An alternative form of the expansion}%
272: %
273:
274:
275: \smallskip
276: \noindent
277: The present expansion of $f(z)$ in the form \expani\ stresses the
278: symmetry of the expansion with respect to $z_1$ and $z_2$. In this
279: representation it is not possible, however, to let $z_1$ and $z_2$
280: coincide, which causes a little inconvenience (the coefficients
281: $a_n(z_1,z_2)$ become infinitely large as $z_1\to z_2$;
282: the remainder $r_N(z_1,z_2;z)$ remains well-defined).
283: An alternative way is the representation (cf. \ftexp),
284: $$
285: f(z)=\sum_{n=0}^{\infty}\left\lbrack
286: A_n(z_1,z_2)+B_n(z_1,z_2)\,z\right\rbrack
287: (z-z_1)^n(z-z_2)^n,
288: $$
289: and we have the relations
290: %
291: $$
292: \eqalign{
293: A_n(z_1,z_2)&=-z_1a_n(z_1,z_2)-z_2a_n(z_2,z_1),\cr
294: B_n(z_1,z_2)&= a_n(z_1,z_2)+a_n(z_2,z_1),\cr
295: }
296: $$
297: %
298: which are regular when $z_1\to z_2$. In fact we have
299: %
300: $$
301: \eqalign{
302: A_n(z_1,z_2)&={1\over 2\pi i}\int_{\cal C}{w-z_1-z_2\over
303: [(w-z_1)(w-z_2)]^{n+1}} f(w)\,dw,\cr
304: B_n(z_1,z_2)&={1\over 2\pi i}\int_{\cal C}{f(w)\,dw\over
305: [(w-z_1)(w-z_2)]^{n+1}}.\cr
306: }
307: $$
308: %
309: Letting $z_1\to0$ and $z_2\to0$, we obtain the standard Maclaurin
310: series of $f(z)$ with even part (the $A_n$ series) and odd part
311: (the $B_n$ series).
312:
313: \rm
314: \parindent=15pt
315:
316: %
317: \subsect{Explicit forms of the coefficients}%
318: %
319:
320:
321: \smallskip
322: \noindent
323: Definition \coefi\ is not appropriate for numerical computations.
324: A more practical formula to compute the coefficients of the above
325: two-point Taylor expansion is given in the following proposition.
326:
327: \noindent
328: {\bf Proposition 1.} {\it Coefficients $a_n(z_1,z_2)$ in the expansion
329: \expani\ are
330: also given by the formulas:
331: %
332: \eqn\ao{
333: a_0(z_1,z_2)={f(z_2)\over z_2-z_1}
334: }
335: %
336: and, for $n=1,2,3,...$,
337: %
338: \eqn\coefibis{
339: a_n(z_1,z_2)=\sum_{k=0}^n{(n+k-1)!\over
340: k!(n-k)!}{(-1)^{n+1}nf^{(n-k)}(z_2)+(-1)^kkf^{(n-k)}(z_1)\over
341: n!(z_1-z_2)^{n+k+1}}.
342: }
343: %
344: %}
345:
346: \noindent
347: {\bf Proof.} We deform the contour of integration ${\cal C}$ in equation
348: \coefi\ to any
349: contour of the form ${\cal C}_1\cup{\cal C}_2$ also contained in $\Omega$,
350: where
351: ${\cal C}_1$ (${\cal C}_2$) is a simple closed loop which encircles the
352: point $z_1$ ($z_2$) in the
353: counterclockwise direction and does not contain the point $z_2$ ($z_1$)
354: inside (see Figure 3 (a)). Then,
355: $$
356: \eqalign{
357: a_n(z_1,z_2)= & {1\over 2\pi i(z_2-z_1)}\left\lbrace\int_{{\cal C}_1}{f(w)\over
358: (w-z_2)^{n+1}}{dw\over (w-z_1)^n}+\int_{{\cal C}_2}{f(w)\over
359: (w-z_1)^{n}}{dw\over (w-z_2)^{n+1}}\right\rbrace= \cr &
360: {1\over (z_2-z_1)}\left\lbrace{1\over (n-1)!}{d^{n-1}\over dw^{n-1}}
361: \left.{f(w)\over(w-z_2)^{n+1}}\right\vert_{w=z_1}+
362: {1\over n!}{d^{n}\over dw^{n}}
363: \left.{f(w)\over(w-z_1)^{n}}\right\vert_{w=z_2}\right\rbrace. \cr}
364: $$
365: From here, equations \ao-\coefibis\ follows after straightforward computations.
366: \hfill$\boxe$
367:
368:
369: %
370: \bigskip
371: %
372: \centerline{\epsfxsize=12cm \epsfbox{fig3.adb.eps}}
373: \centerline{(a) \hskip 4cm (b)\hskip 4cm (c)}
374: %
375: \parindent=10pt
376:
377: \noindent
378: {\bf Figure 3}. {\cfont (a) The function $(w-z_2)^{-n-1}f(w)$ is analytic
379: inside
380: ${\cal C}_1$, whereas $(w-z_1)^{-n}f(w)$ is analytic inside ${\cal C}_2$.
381: (b) The function $(w-z_2)^{-n-1}g_1(w)$ is analytic inside
382: ${\cal C}_1$, whereas $(w-z_1)^{-n}g_2(w)$ is analytic inside ${\cal C}_2$.
383: (c) The function $(w-z_2)^{-n-1}g(w)$ is analytic inside
384: ${\cal C}_1$, whereas $(w-z_1)^{-n}f(w)$ is analytic inside ${\cal C}_2$.}
385: \vskip 2mm
386: %
387: \rm
388: \parindent=15pt
389:
390: %
391: \subsect{Two-point Taylor polynomials}%
392: %
393:
394: \smallskip
395: \noindent
396: Next we can define the two-point Taylor polynomial of the function $f(z)$ at
397: in the following way:
398:
399: \noindent
400: {\bf Definition 1.} {\it Let $z$ be a real or complex variable and $z_1$
401: and $z_2$ ($z_1\ne z_2$) two real or complex numbers. If $f(z)$ is
402: $n-1-$times differentiable at those two points, we define the two-point
403: Taylor polynomial of $f(z)$ at $z_1$ and $z_2$ and degree $2n-1$ as
404: $$
405: P_n(z_1,z_2;z)\equiv
406: \sum_{k=0}^{n-1}\left\lbrack
407: a_k(z_1,z_2)(z-z_1)+a_k(z_2,z_1)(z-z_2)\right\rbrack
408: (z-z_1)^k(z-z_2)^k,
409: $$
410: where the coefficients $a_k(z_1,z_2)$ are given in \ao-\coefibis.}
411:
412: \noindent
413: {\bf Proposition 2.} {\it In the conditions of the above definition, define
414: the remainder of the
415: approximation of $f(z)$ by $P_n(z_1,z_2;z)$ at $z_1$ and $z_2$ as
416: $$
417: r_n(z_1,z_2;z)\equiv f(z)-P_n(z_1,z_2;z).
418: $$
419: Then, (i) $r_n(z_1,z_2;z)=o(z-z_1)^{n-1}$ as $z\to z_1$ and
420: $r_n(z_1,z_2;z)=o(z-z_2)^{n-1}$ as $z\to z_2$.
421: (ii) If $f(z)$ is $n-$times differentiable at $z_1$ and/or $z_2$, then
422: $r_n(z_1,z_2;z)={\cal O}(z-z_1)^n$ as $z\to z_1$ and/or
423: $r_n(z_1,z_2;z)={\cal O}(z-z_2)^n$ as $z\to z_2$.}
424:
425: \noindent
426: {\bf Proof.} The proof is trivial if $f(z)$ is analytic at $z_1$ and $z_2$
427: by using \remi.
428: In any case, for real or complex variable, the proof follows after
429: straightforward
430: computations by using l'H\^opital's rule and \ao-\coefibis.
431: \hfill $\boxe$
432:
433:
434:
435: \noindent
436: {\bf Remark 1.} {%\it
437: Observe that the Taylor polynomial of $f(z)$
438: at $z_1$ and $z_2$
439: and degree $2n-1$ is the same as the Hermite's interpolation polynomial of
440: $f(z)$
441: at $z_1$ and $z_2$ with data $f(z_i)$, $f'(z_i)$,...,$f^{(n-1)}(z_i)$,
442: $i=1,2$.}
443:
444:
445: %
446: \sect{Two-point Laurent expansions}
447: %
448:
449: \noindent
450: In the standard theory for Taylor and Laurent expansions much analogy
451: exists between the two types of expansions. For two-point expansions, we
452: have a similar agreement in the representations of coefficients and remainders.
453:
454: \noindent
455: {\bf Theorem 2.} {\it Let $\Omega_0$ and $\Omega$ be closed and open sets,
456: respectively, of the complex plane, and $\Omega_0\subset\Omega\subset\Cs$. Let
457: $f(z)$ be an analytic function on $\Omega\setminus\Omega_0$
458: and $z_1$, $z_2\in\Omega_0$ with $z_1\ne z_2$.
459: Then, for any $z\in\Omega\setminus\Omega_0$,
460: $f(z)$ admits the two-point Laurent expansion
461: %
462: \eqn\expanii{
463: \eqalign{
464: f(z)= & \sum_{n=0}^{N-1}\left\lbrack
465: b_n(z_1,z_2)(z-z_1)+b_n(z_2,z_1)(z-z_2)\right\rbrack
466: (z-z_1)^n(z-z_2)^n+ \cr &
467: \sum_{n=0}^{N-1}\left\lbrack
468: c_n(z_1,z_2)(z-z_1)+c_n(z_2,z_1)(z-z_2)\right\rbrack
469: (z-z_1)^{-n-1}(z-z_2)^{-n-1}+ \cr & r_N(z_1,z_2;z), \cr}
470: }
471: %
472: where the coefficients $b_n(z_1,z_2)$, $b_n(z_2,z_1)$, $c_n(z_1,z_2)$ and
473: $c_n(z_2,z_1)$ of the expansion are given, respectively, by the Cauchy
474: integrals
475: %
476: \eqn\coefii{
477: b_n(z_1,z_2)\equiv{1\over 2\pi i(z_2-z_1)}\int_{{\Gamma}_1}
478: {f(w)\,dw\over (w-z_1)^n(w-z_2)^{n+1}}
479: }
480: %
481: and
482: %
483: \eqn\coefiii{
484: c_n(z_1,z_2)\equiv{1\over 2\pi i(z_2-z_1)}\int_{{\Gamma}_2}
485: (w-z_1)^{n+1}(w-z_2)^{n}f(w)\,dw.
486: }
487: %
488: The remainder term $r_N(z_1,z_2;z)$ is given by the Cauchy integrals
489: %
490: \eqn\remii{
491: \eqalign{
492: r_N(z_1,z_2;z)\equiv & {1\over 2\pi i}\int_{{\Gamma}_1}{f(w)dw\over
493: (w-z_1)^N(w-z_2)^N
494: (w-z)}(z-z_1)^N(z-z_2)^N- \cr &
495: {1\over 2\pi i}\int_{{\Gamma}_2}{(w-z_1)^N(w-z_2)^Nf(w)dw\over
496: w-z}{1\over(z-z_1)^N(z-z_2)^N}.\cr}
497: }
498: %
499: In these integrals, the contours of integration ${\Gamma}_1$ and
500: ${\Gamma}_2$ are
501: simple closed loops
502: contained in $\Omega\setminus\Omega_0$ which encircle the
503: points $z_1$ and $z_2$ in the counterclockwise direction.
504: Moreover, ${\Gamma}_2$ does not contain
505: the point $z$ inside, whereas ${\Gamma}_1$ encircles ${\Gamma}_2$ and the
506: point $z$
507: (see Figure 4 (a)).
508:
509: The expansion \expanii\ is convergent for $z$ inside the Cassini annulus
510: (see Figure 5)
511: %
512: \eqn\domainii{
513: A_{z_1,z_2}\equiv\lbrace z\in\Omega\setminus\Omega_0, \hskip 2mm
514: r_2<\vert(z-z_1)(z-z_2)\vert<r_1\rbrace }
515: %
516: where
517: $$
518: r_1\equiv {\rm Inf}_{w\in\Css\setminus\Omega}\left\lbrace
519: \vert(w-z_1)(w-z_2)\vert\right\rbrace,\quad
520: r_2\equiv{\rm Sup}_{w\in\Omega_0}\left\lbrace
521: \vert(w-z_1)(w-z_2)\vert\right\rbrace.
522: $$
523: }
524:
525: \noindent{\bf Proof.} By Cauchy's theorem,
526: %
527: \eqn\cauchybis{
528: f(z)={1\over 2\pi i}\int_{{\Gamma}_1}{f(w)dw\over w-z}-
529: {1\over 2\pi i}\int_{{\Gamma}_2}{f(w)\,dw\over w-z},
530: }
531: %
532: where ${\Gamma}_1$ and ${\Gamma}_2$ are the contours defined above. We
533: substitute
534: \lequal-\ui\ into the first integral above and
535: $$
536: {1\over w-z}={z_1+z_2-z-w\over (z-z_1)(z-z_2)}{1\over 1-u}, \hskip 2cm
537: %}
538: %%
539: %with
540: %%
541: %\eqn\uii{
542: u\equiv {(w-z_1)(w-z_2)\over (z-z_1)(z-z_2)},
543: $$
544: into the second one. Now we introduce the expansion \expanu\ of the factor
545: $(1-u)^{-1}$ in both integrals in \cauchybis. After straightforward
546: calculations we obtain
547: \expanii-\remii.
548:
549: For any $z$ verifying \domainii, we can take simple closed loops
550: ${\Gamma}_1$ and
551: ${\Gamma}_2$ in $\Omega\setminus\Omega_0$ such that
552: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$
553: $w\in{\Gamma}_1$ and
554: $\vert(z-z_1)(z-z_2)\vert>\vert(w-z_1)(w-z_2)\vert$ $\forall$ $w\in{\Gamma}_2$
555: (see Figure 4 (b)).
556: On these contours $\vert f(w)\vert$ is bounded by some constant $C$: $\vert
557: f(w)\vert\le C$.
558: Introducing these bounds in \remii\ we see that
559: $\lim_{N\to\infty}r_N(z_1,z_2;z)=0$ and
560: the proof follows.
561: \hfill $\boxe$
562:
563:
564: %
565: \bigskip
566: %
567: \centerline{\epsfxsize=10cm \epsfbox{fig4.adb.eps}}
568: %
569: \centerline{(a) \hskip 6cm (b)}
570: \parindent=10pt
571:
572: \noindent
573: {\bf Figure 4}. {\cfont (a) Contours $\Gamma_1$ and $\Gamma_2$ in the integrals
574: \expanii-\remii. (b) For $z\in A_{z_1,z_2}$, we can take a contour
575: $\Gamma_2$ in
576: $\Omega$ situated between $\Omega_0$ and $A_{z_1,z_2}$ and a contour
577: $\Gamma_1$ in $\Omega$ which contains $A_{z_1,z_2}$ inside. Therefore,
578: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$ $w\in\Gamma_1$
579: and
580: $\vert(w-z_1)(w-z_2)\vert<\vert(z-z_1)(z-z_2)\vert$ $\forall$
581: $w\in\Gamma_2$.}
582: \vskip 2mm
583:
584: %
585: \rm
586: \parindent=15pt
587:
588: %
589: \bigskip
590: %
591: \centerline{\epsfxsize=10cm \epsfbox{fig51.adb.eps}}
592: %
593: \centerline{(a) $4r_1>4r_2>\vert z_1-z_2\vert^2$ \hskip 6mm
594: (b) $4r_1>\vert z_1-z_2\vert^2=4r_2$
595: \hskip 6mm (c) $4r_1>\vert z_1-z_2\vert^2>4r_2$}
596: %
597: \bigskip
598: \centerline{\epsfxsize=7cm \epsfbox{fig52.adb.eps}}
599: %
600: \centerline{
601: %(d) $r_1=r_2=\vert z_1-z_2\vert^2$ \hskip 1cm
602: (d) $4r_1=\vert z_1-z_2\vert^2>4r_2$
603: \hskip 2cm (e) $\vert z_1-z_2\vert^2>4r_1>4r_2$}
604: \parindent=10pt
605:
606: \noindent
607: {\bf Figure 5}. {\cfont Shape of the Cassini annulus depending on the relative
608: size of the parameters $r_1$, $r_2$ and
609: the focal distance $\vert z_1-z_2\vert$.}
610: \vskip 2mm
611:
612: %
613: \rm
614: \parindent=15pt
615:
616:
617:
618: If the only singularities of $f(z)$ inside $\Omega_0$ are just poles at
619: $z_1$ and/or
620: $z_2$, then
621: an alternative formula to \coefii\ and \coefiii\ to compute the
622: coefficients of the above two-point
623: Laurent expansion is given in the following proposition.
624:
625: \noindent
626: {\bf Proposition 3.} {\it Suppose that $g_1(z)\equiv (z-z_1)^{m_1}f(z)$
627: and $g_2(z)\equiv (z-z_2)^{m_2}f(z)$ are
628: analytic functions in $\Omega$ for certain $m_1$, $m_2\in\Ns$. Then,
629: for $n=0,1,2,...$,
630: coefficients $b_n(z_1,z_2)$ and $c_n(z_1,z_2)$ in the expansion \expanii\ are
631: also given by the formulas:
632: %
633: \eqn\coefiibis{
634: \eqalign{
635: b_n(z_1,z_2)= & \sum_{k=0}^{n+m_1-1}\left(\matrix{n+m_1-1 \cr
636: k}\right){(-1)^{k+1}(n+1)_kg_1^{(n+m_1-k-1)}(z_1)\over
637: (n+m_1-1)!(z_1-z_2)^{n+k+2}}+ \cr &
638: \sum_{k=0}^{n+m_2}\left(\matrix{n+m_2 \cr
639: k}\right){(-1)^{k}(n)_kg_2^{(n+m_2-k)}(z_2)\over
640: (n+m_2)!(z_2-z_1)^{n+k+1}}, \cr}
641: }
642: %
643: where $(n)_k$ denotes the Pochhammer symbol and
644: %
645: \eqn\coefiibisbis{
646: \eqalign{
647: c_n(z_1,z_2)= & -\sum_{k=0}^{m_1-n-2}k!\left(\matrix{m_1-n-2 \cr
648: k}\right)\left(\matrix{n \cr k}\right){(z_1-z_2)^{n-k-1}
649: g_1^{(m_1-n-k-2)}(z_1)\over (m_1-n-2)!}+ \cr &
650: \sum_{k=0}^{m_2-n-1}k!\left(\matrix{m_2-n-1 \cr
651: k}\right)\left(\matrix{n+1 \cr k}\right){(z_2-z_1)^{n-k}
652: g_2^{(m_2-n-k-1)}(z_2)\over (m_2-n-1)!}. \cr}
653: }
654: %
655: In these formulas, empty sums must be understood as zero. Coefficients
656: $b_n(z_2,z_1)$ and
657: $c_n(z_2,z_1)$ are given, respectively, by \coefiibis\ and \coefiibisbis\
658: interchanging $z_1$, $g_1$ and $m_1$ by $z_2$, $g_2$ and $m_2$ respectively.
659: }
660:
661: \noindent
662: {\bf Proof.} We deform both, the contour ${\Gamma}_1$ in equation \coefii\ and
663: ${\Gamma}_2$ in equation \coefiii, to any
664: contour of the form ${\cal C}_1\cup{\cal C}_2$ contained in
665: $\Omega$, where
666: ${\cal C}_1$ (${\cal C}_2$) is a simple closed loop which
667: encircles the point $z_1$ ($z_2$) in the
668: counterclockwise
669: direction and does not contain the point $z_2$ ($z_1$) inside (see Figure 3
670: (b)). Then,
671: $$
672: \eqalign{
673: b_n(z_1,z_2)= & {1\over 2\pi i(z_2-z_1)}\left\lbrace\int_{{\cal
674: C}_1}{g_1(w)\over
675: (w-z_2)^{n+1}}{dw\over (w-z_1)^{n+m_1}}+ \right. \cr & \left.
676: \int_{{\cal C}_2}{g_2(w)\over
677: (w-z_1)^{n}}{dw\over (w-z_2)^{n+m_2+1}}\right\rbrace = \cr &
678: {1\over z_2-z_1}\left\lbrace{1\over (n+m_1-1)!}{d^{n+m_1-1}\over dw^{n+m_1-1}}
679: \left.{g_1(w)\over(w-z_2)^{n+1}}\right\vert_{w=z_1}+
680: \right. \cr & \left.
681: {1\over (n+m_2)!}{d^{n+m_2}\over dw^{n+m_2}}
682: \left.{g_2(w)\over(w-z_1)^{n}}\right\vert_{w=z_2}\right\rbrace \cr}
683: $$
684: and
685: $$
686: \eqalign{
687: & c_n(z_1,z_2)= {1\over 2\pi i(z_2-z_1)}\left\lbrace\int_{{\cal C}_1}
688: {(w-z_2)^{n}g_1(w)\over (w-z_1)^{m_1-n-1}}dw+
689: %\right. \cr & \left.
690: \int_{{\cal C}_2}
691: {(w-z_1)^{n+1}g_2(w)\over (w-z_2)^{m_2-n}}dw\right\rbrace= \cr &
692: {1\over z_2-z_1}\left\lbrace\left.{d^{m_1-n-2}\over
693: dw^{m_1-n-2}}
694: \left\lbrack{(w-z_2)^{n}g_1(w)\over (m_1-n-2)!}
695: \right\rbrack\right\vert_{w=z_1}+
696: %\right. \cr & \left.
697: \left.{d^{m_2-n-1}\over dw^{m_2-n-1}}
698: \left\lbrack{(w-z_1)^{n+1}g_2(w)\over (m_2-n-1)!}
699: \right\rbrack\right\vert_{w=z_2}\right\rbrace
700: \cr}
701: $$
702: From here, equations \coefiibis\ and \coefiibisbis\ follow
703: after straightforward computations.
704: \hfill$\boxe$
705:
706: {\bf Remark 2.} {%\it
707: Let $z$ be a real or complex variable and $z_1, z_2$
708: ($z_1\ne z_2$) two real or
709: complex numbers. Suppose that $g_1(z)\equiv (z-z_1)^{m_1}f(z)$
710: is $n-$times differentiable at $z_1$ and $g_2(z)\equiv (z-z_2)^{m_2}f(z)$
711: is $n-$times differentiable at $z_2$. Define
712: $$
713: g(z)\equiv f(z)-\sum_{n=0}^{M-1}\left\lbrack
714: c_n(z_1,z_2)(z-z_1)+c_n(z_2,z_1)(z-z_2)\right\rbrack
715: (z-z_1)^{-n-1}(z-z_2)^{-n-1},
716: $$
717: where $M\equiv$Max$\lbrace m_1,m_2\rbrace$. Then, the thesis of Proposition 2
718: holds for $f(z)$ replaced by $g(z)$. Moreover, if
719: $(z-z_1)^{m_1}(z-z_2)^{m_2}f(z)$ is an
720: analytic function in $\Omega$, then the thesis of Theorem 1
721: applies to $g(z)$.}
722:
723:
724: %
725: \sect{Two-point Taylor-Laurent expansions}
726: %
727:
728:
729: \noindent
730: {\bf Theorem 3.} {\it Let $\Omega_0$ and $\Omega$ be closed and open
731: sets, respectively, of the complex
732: plane, and $\Omega_0\subset\Omega\subset\Cs$. Let $f(z)$
733: be an analytic function on $\Omega\setminus\Omega_0$,
734: $z_1\in\Omega_0$ and $z_2\in\Omega\setminus\Omega_0$. Then, for
735: $z\in\Omega\setminus\Omega_0$, $f(z)$ admits the Taylor-Laurent expansion
736: %
737: \eqn\expaniii{
738: \eqalign{
739: f(z)= & \sum_{n=0}^{N-1}\left\lbrack
740: d_n(z_1,z_2)(z-z_1)+d_n(z_2,z_1)(z-z_2)\right\rbrack
741: (z-z_1)^n(z-z_2)^n+ \cr &
742: \sum_{n=0}^{N-1} e_n(z_1,z_2)
743: (z-z_2)^{n}(z-z_1)^{-n-1}+r_N(z_1,z_2;z), \cr}
744: }
745: %
746: where the coefficients $d_n(z_1,z_2)$, $d_n(z_2,z_1)$ and $e_n(z_1,z_2)$
747: of the expansion are given by the Cauchy integrals
748: %
749: \eqn\coefiv{
750: d_n(z_1,z_2)\equiv{1\over 2\pi i(z_2-z_1)}\int_{{\Gamma}_1}
751: {f(w)\,dw\over (w-z_1)^n(w-z_2)^{n+1}}
752: }
753: %
754: and
755: %
756: \eqn\coefv{
757: e_n(z_1,z_2)\equiv{z_1-z_2\over 2\pi i}\int_{{\Gamma}_2}
758: {(w-z_1)^{n}\over(w-z_2)^{n+1}}f(w)\,dw.
759: }
760: %
761: The remainder term $r_N(z_1,z_2;z)$ is given by the Cauchy integrals
762: %
763: \eqn\remiii{
764: \eqalign{
765: r_N(z_1,z_2;z)\equiv & {1\over 2\pi i}\int_{{\Gamma}_1}{f(w)\,dw\over
766: (w-z_1)^N(w-z_2)^N
767: (w-z)}(z-z_1)^N(z-z_2)^N- \cr &
768: {1\over 2\pi i}\int_{{\Gamma}_2}{(w-z_1)^Nf(w)\,dw\over (w-z_2)^N
769: (w-z)}{(z-z_2)^N\over(z-z_1)^N}.\cr}
770: }
771: %
772: In these integrals, the contours of integration ${\Gamma}_1$ and
773: ${\Gamma}_2$ are
774: simple closed loops
775: contained in $\Omega\setminus\Omega_0$ which encircle $\Omega_0$ in the
776: counterclockwise
777: direction. Moreover, ${\Gamma}_2$ does not contain
778: the points $z$ and $z_2$ inside, whereas ${\Gamma}_1$ encircles
779: ${\Gamma}_2$ and the points
780: $z$ and $z_2$ (see Figure 6 (a)).
781:
782: The expansion \expaniii\ is convergent in the region (Figure 7)
783: %
784: \eqn\domainiii{
785: D_{z_1,z_2}\equiv\lbrace z\in\Omega\setminus\Omega_0, \hskip 2mm
786: \vert(z-z_1)(z-z_2)\vert<r_1
787: \hskip 2mm {\rm and}\hskip 2mm
788: \vert z-z_2\vert<r_2\vert z-z_1\vert\rbrace }
789: %
790: where $r_1\equiv$ Inf$_{w\in\Css\setminus\Omega}\left\lbrace
791: \vert(w-z_1)(w-z_2)\vert\right\rbrace$
792: and $r_2\equiv$ Inf$_{w\in\Omega_0}\left\lbrace \vert(w-z_2)(w-z_1)^{-1}
793: \vert\right\rbrace$.
794: }
795:
796: \noindent
797: {\bf Proof.} By Cauchy's theorem,
798: %
799: \eqn\cauchybisbis{
800: f(z)={1\over 2\pi i}\int_{{\Gamma}_1}{f(w)\,dw\over w-z}-
801: {1\over 2\pi i}\int_{{\Gamma}_2}{f(w)\,dw\over w-z},
802: }
803: %
804: where ${\Gamma}_1$ and ${\Gamma}_2$ are the contours defined above. We
805: substitute \lequal-\ui\ into the first integral above and
806: %
807: \eqn\equalbisbis{
808: {1\over w-z}={z_2-z_1\over (z-z_1)(w-z_2)}{1\over 1-u}, \hskip 2cm
809: %}
810: %%
811: %where
812: %%
813: %\eqn\uiii{
814: u\equiv {(w-z_1)(z-z_2)\over (z-z_1)(w-z_2)}
815: }
816: %
817: into the second
818: one. Now we introduce the expansion \expanu\ of the factor $(1-u)^{-1}$ in
819: both integrals in \cauchybisbis. After straightforward calculations we obtain
820: \expaniii-\remiii.
821:
822: For any $z$ verifying \domainiii, we can take simple closed loops
823: ${\Gamma}_1$ and ${\Gamma}_2$ in $\Omega\setminus\Omega_0$ such that
824: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$
825: $w\in{\Gamma}_1$ and
826: $\vert(z-z_1)(w-z_2)\vert>\vert(w-z_1)(z-z_2)\vert$ $\forall$ $w\in{\Gamma}_2$
827: (see Figure 6 (b)).
828: On these contours $\vert f(w)\vert$ is bounded by some constant $C$: $\vert
829: f(w)\vert\le C$.
830: Introducing these bounds in \remiii\ we see that
831: $\lim_{N\to\infty}r_N(z_1,z_2;z)=0$ and
832: the proof follows.
833: \hfill $\boxe$
834:
835: %
836: \bigskip
837: %
838: \centerline{\epsfxsize=10cm \epsfbox{fig6.adb.eps}}
839: %
840: \centerline{(a) \hskip 6cm (b)}
841: \parindent=10pt
842:
843: \noindent
844: {\bf Figure 6}. {\cfont (a) Contours $\Gamma_1$ and $\Gamma_2$ in the integrals
845: \expaniii-\remiii. (b) For $z\in D_{z_1,z_2}$, we can take a contour $\Gamma_2$
846: situated between $\Omega_0$ and $D_{z_1,z_2}$ and a contour $\Gamma_1$
847: in $\Omega$ which
848: contains $D_{z_1,z_2}$ inside. Therefore,
849: $\vert(z-z_1)(z-z_2)\vert<\vert(w-z_1)(w-z_2)\vert$ $\forall$ $w\in\Gamma_1$
850: and
851: $\vert(w-z_1)(z-z_2)\vert<\vert(z-z_1)(w-z_2)\vert$ $\forall$ $w\in\Gamma_2$.}
852: \vskip 2mm
853:
854: %
855: \rm
856: \parindent=15pt
857:
858: %
859: \bigskip
860: %
861: \centerline{\epsfxsize=10cm \epsfbox{fig71.adb.eps}}
862: %
863: \centerline{(a) $4r_1>\vert z_1-z_2\vert^2$, $r_2>1$ \hskip 3mm
864: (b) $4r_1>\vert z_1-z_2\vert^2$, $r_2=1$
865: \hskip 3mm (c) $4r_1>\vert z_1-z_2\vert^2$, $r_2<1$}
866: %
867: \bigskip
868: \centerline{\epsfxsize=10cm \epsfbox{fig72.adb.eps}}
869: %
870: \centerline{(d) $4r_1=\vert z_1-z_2\vert^2$, $r_2>1$ \hskip 3mm
871: (e) $4r_1=\vert z_1-z_2\vert^2$, $r_2=1$
872: \hskip 3mm (f) $4r_1=\vert z_1-z_2\vert^2$, $r_2<1$}
873: %
874: \bigskip
875: \centerline{\epsfxsize=11cm \epsfbox{fig73.adb.eps}}
876: %
877: \centerline{(g) $4r_1<\vert z_1-z_2\vert^2$, $r_2>1$ \hskip 3mm
878: (h) $4r_1<\vert z_1-z_2\vert^2$, $r_2=1$
879: \hskip 3mm (i) $4r_1<\vert z_1-z_2\vert^2$, $r_2<1$}
880: \parindent=10pt
881:
882: \noindent
883: {\bf Figure 7}. {\cfont The region $D_{z_1,z_2}$ defined in Theorem 3 is
884: given by
885: $D_{z_1,z_2}=D_1\bigcap D_2$, where $D_1$
886: is the Cassini oval of focus $z_1$ and $z_2$ and parameter
887: $r_1$. On the other hand,
888: for $r_2<1$ ($r_2>1$), $D_2$ is the interior (exterior)
889: of the circle of center
890: $z_1+(1-r_2^2)^{-1}(z_2-z_1)=z_2+r_2^2(r_2^2-1)^{-1}(z_1-z_2)$
891: and radius $\vert z_1-z_2\vert r_2/\vert r_2^2-1\vert$.
892: For $r_2=1$, $D_2$ is just the half plane
893: $\vert z-z_2\vert<\vert z-z_1\vert$. The
894: shape of the Cassini annulus depends on the relative
895: size of the parameters $\sqrt{r_1}$, $\sqrt{r_2}$ and
896: the focal distance $\vert z_1-z_2\vert$.}
897: \vskip 2mm
898:
899: %
900: \rm
901: \parindent=15pt
902:
903:
904: If the only singularities of $f(z)$ inside $\Omega_0$ are just poles at
905: $z_1$, then
906: an alternative formula to \coefiv-\coefv\ to compute the coefficients of
907: the above two-point
908: Taylor-Laurent expansion is given in the following proposition.
909:
910: \noindent
911: {\bf Proposition 4.} {\it Suppose that $g(z)\equiv (z-z_1)^{m}f(z)$ is an
912: analytic function in $\Omega$ for certain $m\in\Ns$. Then, coefficients
913: $d_n(z_1,z_2)$ and $d_n(z_2,z_1)$ in the expansion \expaniii\ are also
914: given by the formulas:
915: %
916: \eqn\do{
917: d_0(z_1,z_2)= {f(z_2)\over z_2-z_1}-\sum_{k=0}^{m-1}
918: {1\over (m-k-1)!}{g^{(m-k-1)}(z_1)\over(z_2-z_1)^{k+2}}, \hskip 1cm
919: }
920: %
921: $$
922: d_0(z_2,z_1)= {1\over m!}{g^{(m)}(z_1)\over z_1-z_2},
923: $$
924: and, for $n=1,2,3...$,
925: %
926: \eqn\coefiiibis{
927: \eqalign{
928: d_n(z_1,z_2)= & -{(-1)^n\over n!}\left\lbrace \sum_{k=0}^{m+n-1}
929: {(n+k)!\over k!(m+n-k-1)!}{g^{(m+n-k-1)}(z_1)\over(z_2-z_1)^{n+k+2}}+
930: \right. \cr &
931: \left. n\sum_{k=0}^{n}
932: {(n+k-1)!\over
933: k!(n-k)!}{f^{(n-k)}(z_2)\over(z_1-z_2)^{n+k+1}}\right\rbrace, \cr}
934: }
935: %
936: %
937: \eqn\coefiiirebis{
938: \eqalign{
939: d_n(z_2,z_1)= & -{(-1)^n\over n!}\left\lbrace n\sum_{k=0}^{m+n}
940: {(n+k-1)!\over k!(m+n-k)!}{g^{(m+n-k)}(z_1)\over(z_2-z_1)^{n+k+1}}+ \right.
941: \cr &
942: \left.\sum_{k=0}^{n-1}
943: {(n+k)!\over
944: k!(n-k-1)!}{f^{(n-k-1)}(z_2)\over(z_1-z_2)^{n+k+2}}\right\rbrace.\cr}
945: }
946: %
947: For $n=0,1,2,...$, coefficients $e_n(z_1,z_2)$ are given by
948: %
949: \eqn\coefiiibisbis{
950: e_n(z_1,z_2)= {(-1)^n\over n!}\sum_{k=0}^{m-n-1}
951: {(n+k)!\over k!(m-n-k-1)!}{g^{(m-n-k-1)}(z_1)\over(z_2-z_1)^{n+k}}.
952: }
953: %
954: }
955:
956: \noindent
957: {\bf Proof.} We deform both, the contour ${\Gamma}_1$ in equation \coefiv\ and
958: the contour ${\Gamma}_2$ in equation \coefv\ to any
959: contour of the form ${{\cal C}_1}\cup{{\cal C}_2}$ contained in
960: $\Omega$, where ${{\cal C}_1}$ (${{\cal C}_2}$) is a simple closed loop
961: which encircles the point $z_1$ ($z_2$) in the counterclockwise
962: direction and does not contain the point $z_2$ ($z_1$) inside (see Figure 3
963: (c)). Then,
964: $$
965: \eqalign{
966: d_n(z_1,z_2)= & {1\over 2\pi i(z_2-z_1)}\left\lbrace\int_{{\cal C}_1}{g(w)\over
967: (w-z_2)^{n+1}}{dw\over (w-z_1)^{n+m}}+
968: \right. \cr & \left. \int_{{\cal C}_2}{f(w)\over
969: (w-z_1)^{n}}{dw\over (w-z_2)^{n+1}}\right\rbrace= \cr &
970: {1\over (z_2-z_1)}\left\lbrace{1\over (n+m-1)!}{d^{n+m-1}\over dw^{n+m-1}}
971: \left.{g_1(w)\over(w-z_2)^{n+1}}\right\vert_{w=z_1}+
972: \right. \cr & \left.
973: {1\over n!}{d^{n}\over dw^{n}}
974: \left.{f(w)\over(w-z_1)^{n}}\right\vert_{w=z_2}\right\rbrace, \cr}
975: $$
976: an analog formula for $d_n(z_2,z_1)$, and
977: $$
978: \eqalign{
979: e_n(z_1,z_2)= & {z_1-z_2\over 2\pi i}\int_{{\cal C}_1}
980: {g(w)\over(w-z_2)^{n+1}}{dw\over (w-z_1)^{m-n}}= \cr &
981: (z_1-z_2){1\over (m-n-1)!}{d^{m-n-1}\over dw^{m-n-1}}
982: \left. {g(w)\over (w-z_2)^{n+1}}\right\vert_{w=z_1}. \cr}
983: $$
984: From here, equations \do-\coefiiibisbis\ follow
985: after straightforward computations.
986: \hfill$\boxe$
987:
988: {\bf Remark 3.} {%\it
989: Let $z$ be a real or complex variable and
990: $z_1$ and $z_2$ ($z_1\ne z_2$) two real or
991: complex numbers. Suppose that $(z-z_1)^{m}f(z)$
992: is $n-$times differentiable at $z_1$ for certain $m\in\Ns$ and $f(z)$
993: is $n-$times differentiable at $z_2$. Define
994: $$
995: g(z)\equiv f(z)-\sum_{n=0}^{m-1}
996: e_n(z_1,z_2)(z-z_1)^{-n-1}(z-z_2)^{n}.
997: $$
998: Then, the thesis of Proposition 2
999: holds for $g(z)$. If moreover, $(z-z_1)^{m}f(z)$ is an
1000: analytic function in $\Omega$, then the thesis of Theorem 1
1001: applies to $g(z)$.}
1002:
1003:
1004:
1005: %
1006: \sect{Acknowledgements}
1007: %
1008:
1009: \noindent
1010: J. L. L\'opez wants to thank the C.W.I. of Amsterdam for its scientific and
1011: financial support during the realization of this work. The financial
1012: support of the saving bank {\it Caja Rural de Navarra} is also acknowledged.
1013:
1014: \listrefs
1015:
1016:
1017: \end
1018:
1019:
1020: