You are on page 1of 415

Classics in Mathematics

George P61ya • Gabor Szego Problems and Theorems


in Analysis I

yuana97@uw.edu
Springer
Berlin
Heidelberg
New York
Barcelona
Budapest
Hong Kong
London
Milan
Paris
Santa Clara
Singapore
Tokyo

yuana97@uw.edu
This picture shows
G. P61ya (r.) and
G. Szego (I.) delivering
their original manu-
script to Springer in
Berlin in 1925 (courtesy
of G. Alexanderson) .

George P61ya
Born in Budapest, December 13,1887, George P61ya initially
studied law, then languages and literature in Budapest. He came
to mathematics in order to understand philosophy, but the
subject of his doctorate in 1912 was in probability theory and
he promptly abandoned philosophy.
After a year in GOHingen and a short stay in Paris, he received
an appointment at the ETH in Zurich. His research was multi-
faceted, ranging from series, probability, number theory and
combinatorics to astronomy and voting systems. Some of his
deepest work was on entire functions. He also worked in con-
formal mappings, potential theory, boundary value problems,
and isoperimetric problems in mathematical physics, as well as
heuristics late in his career. When P61ya left Europe in 1940,
he first went to Brown University, then two years later to
Stanford, where he remained until his death on September 7,
1985.

Gabor Szego
Born in Kunhegyes, Hungary, January 20, 1895, Szego studied
in Budapest and Vienna, where he received his Ph. D. in 1918,
after serving in the Austro-Hungarian army in the First World
War. He became a privatdozent at the University of Berlin and in
1926 succeeded Knopp at the University of Konigsberg. It was
during his time in Berlin that he and P61ya collaborated on
their great joint work, the Problems and Theorems in Analysis.
Szego's own research concentrated on orthogonal polynomials
and Toeplitz matrices. With the deteriorating situation in
Germany at that time, he moved in 1934 to Washington Uni-
versity, St. Louis, where he remained until 1938, when he moved
to Stanford. As department head at Stanford, he arranged for
P61ya to join the Stanford faculty in 1942. Szego remained at
Stanford until his death on August 7, 1985.

yuana97@uw.edu
George P61ya • Gabor Szego

Problems and Theorems


in Analysis I
Series. Integral Calculus.
Theory of Functions

Reprint of the 1978 Edition

Springer

yuana97@uw.edu
George P6lya t
Gabor Szego t

Translator:
Dorothee Aeppli
1414 Chelmsfordn Street
St.Paul,MN 55108
USA

Originally published as Vol. 193 of the


Grundlehren der mathematischen Wissenschaften

Mathematics Subject Classification (1991): 05-01,28-01,30-01,40-01

CIP data applied for

Die Deutsche Bibliothek - CIP-Einheitsaufnahme


P61ya, George:
Problems and theorems in analysis I George P6lya; Gabor Szeg6'.- [Nachdr.].- Berlin; Heidelberg; New York;
Barcelona; Budapest; Hong Kong; London; Milan; Paris; Santa Clara, Singapore; Tokyo: Springer
(Classics in mathematics)
1. Series, integral calculus, theory offunctions.-Reprint [der Ausg.] Berlin, Springer, 1978.- 1998

ISBN-13: 978-3-540-63640-3 e-ISBN-13: 978-3-642-61983-0


DOl: 10.1007/978-3-642-61983-0

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or
parts thereof is permitted ouly under the provisions of the German Copyright Law of September 9, 1965,in its
current version, and permission for use must always be obtained from Springer-Verlag. Violations are liable
for prosecution under the German Copyright Law.

@ Springer-Verlag Berlin Heidelberg 1998


Softcover reprint of the hardcover 15t Edition 1998

The use of general descriptive names, registered names, trademarks etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.

SPIN 10651015 4113143-543210- Printed on acid-free paper

yuana97@uw.edu
G. P61ya G. Szego

Problems and Theorems


in Analysis I
Series Integral Calculus Theory of Functions

Translation by D. Aeppli

Corrected Printing

Springer -Verlag
Berlin Heidelberg New York 1978

yuana97@uw.edu
George P6lya Gabor Szego
Department of Mathematics, Stanford University
Stanford, CA 94305/USA

Revised and enlarged translation of "Aufgaben und Lehrsatze aus der


Analysis I", 4th ed., 1970; Heidelberger Taschenbucher, Band 73

AMS Subject Classifications (1970): 05-01, 28-01, 30-01, 40-01

ISBN-13: 978-3-540-63640-3 e-ISBN-13: 978-3-642-61983-0


DOl: 10.1007/978-3-642-61983-0

This work is subject to copyright. AU rights are reserved, whether the whole or part of the material is
concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting,
reproduction by photocopying machine or similar means, and storage in data banks. Under § 54 of the
German Copyright Law where copies are made for other than private use, a fee is payable to
"Verwertungsgesellschaft Wort", Munich.
© by Springer·Verlag Berlin Heidelberg 1972
Softeover reprint of the hardcover 15t Edition 1972
Printing and Bookbinding: BrOhlsche UniversiUitsdruckerei, Giessen
2141!314().5432

yuana97@uw.edu
Preface to the English Edition

The present English edition is not a mere translation of the German


original. Many new problems have been added and there are also
other changes, mostly minor. Yet all the alterations amount to less than
ten percent of the text. We intended to keep intact the general plan
and the original flavor of the work.
Thus we have not introduced any essentially new subject matter,
although the mathematical fashion has greatly changed since 1924.
We have restricted ourselves to supplementing the topics originally
chosen.
Some of our problems first published in this work have given rise to
extensive research. To include all such developments would have
<:hanged the character of the work, and even an incomplete account,
which would be unsatisfactory in itself, would have cost too much labor
and taken up too much space.
We have to thank many readers who, since the publication of this
work almost fifty years ago, communicated to us various remarks on it,
some of which have been incorporated into this edition. We have not
listed their names; we have forgotten the origin of some contributions,
and an incomplete list would have been even less desirable than no list.
The first volume has been translated by Mrs. Dorothee Aeppli, the
second volume by Professor Claude Billigheimer. We wish to express
our warmest thanks to both for the unselfish devotion and scrupulous
conscientiousness with which they attacked their far from easy task.
Our thanks are also due to Dr. Klaus Peters for his unflagging interest
and wise advice and to Dr. Julius G. Baron for his kind help with the
proofsheets.

Stanford, March 1972


G. P6lya . G. Szego

yuana97@uw.edu
Preface to the First German Edition

What is good education? Giving sys-


tematically opportunity to the student to
discover things by himself.
(Condensed from H. Spencer.)

In the mathematical literature there exist a number of excellent and


comprehensive collections of problems, books of exercises, review texts
etc. The present book, in our view, differs from all these, both in its
aim and in the scope and arrangement of the material covered, as well as
in the manner in which we envisage its use. Consequently each of these
points requires some explanation.

The chief aim of this book, which we trust is not unrealistic, is to


accustom advanced students of mathematics, through systematically
arranged problems in some important fields of analysis, to the ways
and means of indepeQdent thought and research. It is intended to serve
the need for individual active study on the part of both the student and
the teacher. The book may be used by the student to extend his own
reading or lecture material, or he may work quite independently through
selected portions of the book in detail. The instructor may use it as an
aid in organizing tutorials or seminars.
This bQok is no mere collection of problems. Its most important
feature is the systematic arrangement of the material which aims to
stimulate the reader to independent work and to suggest to him useful
lines of thought. We have devoted more time, care and detailed effort
to devising the most effective presentation of the material than might
be apparent to the uninitiated at first glance.
The imparting of factual knowledge is for us a secondary considera-
tion. Above all we aim to promote in the reader a correct attitude,
a certain discipline of thought, which would appear to be of even

yuana97@uw.edu
Preface to the First German Edition VII

more essential importance in mathematics than in other scientific


disciplines.

General rules which could prescribe in detail the most useful disci-
pline of thought are not known to us. Even if such rules could be for-
mulated, they would not be very useful. Rather than knowing the correct
rules of thought theoretically, one must have them assimilated into
one's flesh and blood ready for instant and instinctive use. Therefore,
for the schooling of one's powers of thought only the practice of thinking
is really useful. The independent solving of challenging problems will aid
the reader far more than the aphorisms which follow, although as a
start these can do him no harm.
One should try to understand everything: isolated facts by collating
them with related facts, the newly discovered through its connection
with the already assimilated. the unfamiliar hy analogy with the accus-
tomed, special results through generalization, general results by means
of suitable specialization, complex situations by dissecting them into
their constituent parts, and details by comprehending them within a
total picture.
There is a similarity between knowing one's way about a town and
mastering a field of knowledge; from any given point one should be able
to reach any other point 1. One is even better informed if one can imme-
diately take the most convenient and quickest path from the one point
to the other. If one is very well informed indeed, one can even execute
special feats, for example, to carry out a journey by systematically
avoiding certain forbidden paths which are customary-such things
happen in certain axiomatic investigations.
There is an analogy between the task of constructing a well-integrated
body of knowledge from acquaintance with isolated truths and the
building of a wall out of unhewn stones. One must turn each new insight
and each new stone over and over, view it from all sides, attempt to
join it on to the edifice at all possible points, until the new finds its
suitable place in the already established, in such a way that the areas
of contact will be as large as possible and the gaps as small as possible,
until the whole forms one firm structure.
A straight line is determined by two points. Similarly, many a new
result is obtained by means of a kind of linear interpolation between

1 See e.g. problem 92 and the neighboring problems in Part VI, also problem 64
in Part VIII.

yuana97@uw.edu
VIII Preface to the First German Edition

two extreme special cases 1 • A straight line is also determined by a direc-


tion and one point. New results also frequently arise from the fortunate
coincidence of the direction of one's work with a notable special case.
Also the drawing of parallels is a valuable method by means of which new
results may be derived!!.
An idea which can be used only once is a trick. If one can use it more
than once it becomes a method. In mathematical induction the result
to be obtained and the means available for its proof are proportional,
they stand in the ratio of n + 1 to n. Hence, strengthening the statement
to be proved may also be advantageous, for we strengthen at the same
time the means available for its proof. It is also found in other circum-
stances that the more general statement may be easier to prove than
the more particular; in such cases the most important achievement
consists precisely in setting up the more general statement, in extracting
the essential, in realizing the complete picture 3 •
"Qui nimium probat, nihil probat". One should examine every proof
with suspicion to see if in fact all the assumptions stated have been
used. One should attempt to obtain the same conclusion from fewer
hypotheses, or a stronger conclusion from the same hypotheses, and
should be satisfied only when one has found counter-examples which
show that the limit of the possible has been attained.
However, one must not forget that there are two kinds of generaliza-
tion, one facile and one valuable. One is generalization by dilution, the
other is generalization by concentration. Dilution means boiling the
meat in a large quantity of water into a thin soup; concentration means
. condensing a large amount of nutritive material into an essence. The
unification of concepts which in the usual view appear to lie far removed
from each other is concentration. Thus, for example, group theory has
concentrated ideas which formerly were found scattered in algebra,
number theory, geometry and analysis and which appeared to be very
different. Examples of generalization by dilution would be still easier to
quote, but this would be at the risk of offending sensibilities.

1 See e.g. problem 139 in Part I.


2 See e.g. the first section in Chapter t of Part IV. in particular problems 13
and 14.
a It happens frequently that a hint which is appended to the problem. or the
grouping of the neighboring problems. points to a strengthening or a generaliZation
of the result which may be useful for the solution of the problem. Thus compare
with each other problems 1 and 2. 3 and 4. 5 and 7 •• and 8 of Part I.

yuana97@uw.edu
Preface to the First German Edition IX
Not all the subject matter of analysis is suitable for problems. A
collection in which all the more important fields of analysis were ex-
haustively. dealt with would necessarily become too extensive and awk-
ward. One can, of course, make a selection in many different ways. We
have placed the greatest weight on the central field of modem analysis,
the theory of functions of a complex variable. However, we have kept
ourselves somewhat apart from the common highway travelled by the
usual lectures, textbooks and collections of problems, and have, all
things being eqt:.al, given preference to those fields which lie closest to
our personal interests. We have also taken problems from more difficult
fields and such as are still very much in the developmental stage, which
have scarcely or not at all been considered as yet in the textbook litera-
ture. The table of contents will illustrate this in more detail. Certain
chapters may also be made use of by the specialist. But we have nowhere
attempted to attain the completeness of a monograph since we have
subordinated the selection of the material to our chief aim, which is to
present the material, to the best of our ability, in an arrangement that
provides guidance and suggestions to the reader.
The origin of the material is highly varied. We have made selections
from the classical body of knowledge of mathematics and also from
treatises of more recent date. We collected problems which had in part
already been published in various periodicals and in part communicated
to us verbally by their authors. We have adapted the material to our
purpose, completed, reformulated and substantially expanded it. In
addition we have published here for the first time in the form of problems
a number of our own original results. We thus hope to be able to offer
something new even to the expert.

The material is arranged in two volumes. The first comprises three


parts of a more fundamental nature, the second six parts which are
devoted to more specialized questions and applications.
Each volume presents in its first half problems and in its second
half their solutions. In the part containing the problems, especially at
the beginning of the separate chapters, there are also some explanations,
which recall the general notions and theorems needed as a background.
Often there is added to a problem a method of attack or hint. The
solutions are presented in as brief and concise a form as possible. Trivial
deductions are omitted for they should become clear enough after a
serious consideration of the problem. In exceptional cases the solution

yuana97@uw.edu
x Preface to the First German Edition

is merely sketched and the reader is referred to the relevant literature.


On occasion, extensions, other applications and unsolved problems are
touched upon.
The parts are divided into chapters and these are subdivided into
sections. Whenever an explanation follows or a new train of thought
is introduced, we indicate this by a space.
The arrangement of the problems within the chapters and sections
is the point wherein the present book differs perhaps even more from
the other similar books known to us than in the selection of the material.
Practice problems in the narrow sense which aim to clarify newly learned
theorems and concepts by means of suitable special cases occupy rela-
tively little space. Isolated problems are rare. The individual problem is
mostly incorporated into a longer series of problems which on the
average cover a section, and the organic construction of these series has
been the object of our greatest concern.
One may group problems from various points of view-according to
required previous knowledge, difficulty, method, or result. We have
not committed ourselves to any of these viewpoints, but have chosen
varying arrangements, which may reflect the different situations that
one meets in independent research. One section for example may be
concerned with a method which is explained briefly at the beginning
and is afterwards applied to the solution of as many problems of various
forms as possible and which is thus developed further and further.
Another section may deal similarly with a theorem which is stated at
the beginning (or proved, if this can be done easily and quickly) and is
then applied and specialized in several ways. Still other sections are
constructed on an ascending pattern: The general theorem appears
only after preceding special cases and small, fragmentary remarks which
suggest the result or lead to its proof. Occasionally a proof which is more
difficult is attained in several steps, through a sequence of problems;
each problem yields an auxiliary lemma, an independent part of the
proof, or some perspective, and thus forms a link in a chain of ideas,
by means of which the reader finally reaches the theorem that is to be
proved. Some sections bring "miscellaneous problems" and are more
loosely knit. They recapitulate preceding material by means of more
difficult applications or present isolated results which are of interest
in themselves.
Now and then four consecutive problems form a "proportion", in
which the fourth has the same relation to the third as the second to the

yuana97@uw.edu
Preface to the First German Edition XI
first (generalization, converse, application). Some paragraphs are devoted
to a more detailed presentation and examination of analogies l . Here the
problems are taken in turn from the two subjects which are placed in
parallel. They belong together in pairs and form what may be termed a
"continued proportion". This arrangement seems to us to be particularly
instructive.

One may approach the book with a view to finding in it opportunity


for practice for oneself, or for one's students, or simply for reading. In each
case a suitable way of using it for the particular purpose can be found
quite naturally.
The initial chapters of each of the parts mostly require comparatively
little previous knowledge. The different parts are, though not entirely,
yet largely independent of each other, and also the connection between
the sections of the same part is frequently loose, so that for example one
is not required to keep scrupulously to the given sequence of topics.
The reader who wishes to solve a problem should think not only
about what is asked, but also how and where it is asked. Many.problems,
which would be intractable even for an advanced student if set in isola-
tion, are here surrounded by preparatory and explanatory problems and
presented in such a context that with'some perseverance and a little
inventiveness it should be possible to master them. There occur of course
also really difficult problems without any preparation. These are con-
tained mostly in sections of a looser structure (miscellaneous problems)
or else occur only as isolated problems.
The hints are at the disposal of the reader but they are not intended
to be forced on him.
If you are unable to solve a problem, you should not despair. The
"Socratic method of teaching" does not aim at drilling people in giving
quick answers, but to educate by means of questions. If repeated efforts
have been unsuccessful, the reader can afterwards analyze the solution·
which is to be found in the second half of the volume with more incisive
attention, bring out the actual principle which is the salient point,
assimilate it, and commit it to his memory as a permanent acquisition.
The book, already while in process of being written, was repeatedly
made use of for the organization of practice sessions and problem-solving
seminars for students in the middle and upper semesters. In such
1 See Part II, Chapter 2; Part IV, Chapter 1, § 1; Part V, Chapter 1, § 1;
Part VIII, Chapter 1, § 4.

yuana97@uw.edu
XII Preface to the First German Edition

sessions the easier problems were discussed in the classroom and the
answers were given orally by the students, while the more difficult
problems were answered in writing by an appropriately set deadline.
Important problems serving as paradigms were solved by the instructor.
Within one semester it was possible to covel' approximately the material
of one chapter. Several chapters have been tested in this manner and in
part revised on the basis of the experience which we have gained. We
believe that we may in good conscience recommend for the organization
of practice sessions and seminars the method which we have followed:
to pose not isolated problems, but carefully considered connected se-
quences of problems. Nearly all the chapters of this book can be used as
a basis for such instruction. It is obvious that some care should be exer-
cised. Especially, for homework or examinations, it is advisable to
replace some problems by related ones.
A continuous reading of the work, in which immediately after each
problem the solution is also read, can be recommended only to more
experienced readers. On the whole this is not in the spirit of the book.
Nevertheless certain chapters are suitable for such continuou!l reading
and may be used essentially as a teaching text. However, for this pur-
pose the presentation is rather condensed; it was intended to allow some
time for thinking about the problem between its formulation and its
solution, and about the proposition between its statement and its proof.

If our undertaking has not been successful in all respects we appeal


to two extenuating circumstances: Firstly, as the plan of this work is
essentially new, we had no models which we could have followed. Se-
condly, a more extensive treatment of the various chapters would have
required so much space and the improvement of the presentation in some
aspects so much time that the carrying out of the entire plan would have
been placed in jeopardy. In the interests of the project we would be
grateful to the critical reader if he would direct our attention to possible
deficiencies which could be eliminated at a later opportunity.
Numerous friends and colleagues have made available to us un-
published items and others have assisted us by reading the manuscript
or the proofsheets. We gratefully mention by name A. Aeppli (Zurich),
P. Bemays (Gottingen), A. Cohn (Berlin), R. Courant (GOttingen),
P. Csillag (Budapest), L. Fejer (Budapest), M. Fekete (Budapest),
A. Fleck (Berlin), F. Gassmann (Zurich), A. Haar (Szeged), A. Hirsch
(Zurich), E. Jacobsthal (Berlin), L. Kollros (Zurich), J. Kurschak (Buda-

yuana97@uw.edu
Preface to the First German Edition xm
pest), E. Landau (Gottingen), E. Lasker (Berlin), K. Lowner (Berlin),
A. Ostrowski (Gottingen), M. Plancherel (ZUrich), H. Priifer (Jena) ,
T. Rad6 (Szeged), M. Riesz (Stockholm), A. Stoll (ZUrich), O. Toeplitz
(Kiel), A. Walther (Gottingen). We were also permitted to incorporate
some results from unpublished papers in the estate of A. Hurwitz, and
also in that of F. and Th. Lukacs. In particular we would like to thank
sincerely T. Carleman (Lund) and I. Schur (Berlin) for their valuable
problems and also A. and R. Brauer (Berlin), H. Rademacher (Ham-
burg), and H. Weyl (ZUrich) for their truly devoted co-operation. Our
sincere thanks are also due to the publisher who accommodated us in
every respect in spite of the present difficult times.
ZUrich and Berlin, October 1924
G. P6lya . G. Szego

yuana97@uw.edu
Contents
Part One

Infinite Series and Infinite Sequences

Chapter 1
Problem Prob· Solu-
Numbers Operations with Power Series
lem tion
§1 (1-31). Additive Number Theory, Combinatorial Problems, Page Page
and Applications • . . . . . . . . . . . 173
§2 (31.1-43.1). Binomial Coefficients and Related Problems 6 183
§3 (44-49). Differentiation of Power Series 8 185
§4 (50-60). Functional Equations and Power Series 9 186
§5 (60.1-60.11). Gaussian Binomial Coefficients 11 188
§6 (61-64.2). Majorant Series . . . . . 13 189

Chapter 2
Linear Transformations of Series. A Theorem of Cesaro
§1 (65-78). Triangular Transformations of Sequences into Se-
quences 15 191
§2 (79-82). More General Transformations of Sequences into
Sequences ... 19 194
§3 (83-97). Transformations of Sequences into Functions. Theo-
rem of Cesaro , 20 195

Chapter 3
The Structure of Real Sequences and Series
§1 (98-112). The Structure of Infinite Sequences . 23 198
§2 (113-116). Convergence Exponent . . . . . . . 25 202
§3 (117-123). The Maximum Term of a Power Series 26 203
§4 (124-132). Subseries . . . . . . . . . . . . . 28 205
§5 (132.1-137). Rearrangement of the Terms . . . . 30 207
§6 (138-139). Distribution of the Signs of the Terms 32 208

Chapter 4
Miscellaneous Problems
§1 (140-155). Enveloping Series . . . . . . . . . . . . . . . 32 209
§2 (156-185.2). Various Propositions on Real Series and Sequences 36 212
§3 (186-210). Partitions of Sets, Cycles in Permutations 42 224

yuana97@uw.edu
Contents XV
Part Two
Integration
Chapter 1 Prob- Solu-
Problem lem tion
Numbers The Integral as the Limit of a Sum of Rectangles Page Page
§1 (1-7). The Lower and the Upper Sum 46 230
§2 (8-19.2). The Degree of Approximation _ 49 231
§3 (20-29). Improper Integrals Between Finite Limits 51 236
§4 (30-40). Improper Integrals Between Infinite Limits _ 53 237
§5 (41-47). Applications to Number Theory. 55 239
§6 (48-59). Mean Values and Limits of Products 57 241
§7 (60-68). Multiple Integrals 59 244

Chapter 2
Inequalities
§1 (69-94). Inequalities _ 62 247
§2 (94.1-97). Some Applications of Inequalities 72 255

Chapter 3
Some Properties of Real Functions
§1 (98-111). Prope; Integrals _ 75 258
§2 (112-118.1). Improper Integrals 77 261
§3 (119-127). Continuous, Differentiable, Convex Functions _ 79 264
§4 (128-146). Singular Integrals. Weierstrass' Approximation
Theorem 81 270

Chapter 4
Various Types of Equidistribution
§1 (147-161). Counting Function. Regular Sequences 85 275
§2 (162-165). Criteria of Equidistribution _ 87 279
§3 (166-173). Multiples of an Irrational Number _ 88 280
§4 (174-184). Distribution of the Digits in a Table of Logarithms
and Related Questions 90 281
§5 (185-194). Other Types of Equidistribution _ 92 285

Chapter 5
Functions of Large Numbers
§1 (195-209). Laplace's Method 95 287
§2 (210-217.1). Modifications of the Method 98 291
§3 (218-222). Asymptotic Evaluation of Some Maxima 100 294
§4 (223-226). Minimax and Maximin . 101 295
Part Three
Functions of One Complex Variable. General piut
Chapter 1
Complex Numbers and Number Sequences
§1 (1-15). Regions and Curves. Working with C~mplex Vari-
abies. 103 297
§2 (16-27). Location of the Roots of Algebraic Equations _ 105 300

yuana97@uw.edu
XVI Contents
Problem Prob· Solu·
Numben lem tion
§3 (28-35). Zeros of Polynomials, Continued. A Theorem of Page Page
Gauss 108 302
§4 (36-43). Sequences of Complex Numbers. • 109 305
§5 (44-50). Sequences of Complex Numbers, Continued: Trans-
formation of Sequences . 111 307
§6 (51-54). Rearrangement of Infinite Series 112 311
Chapter 2
Mappings and Vector Fields
§1 (55-59). The Cauchy-Riemann Differential Equations 113 311
§2 (60-84). Some Particular Elementary Mappings • 115 313
§3 (85-102). Vector Fields 119 317
Chapter 3
Some Geometrical Aspects of Complex Variables
§1 (103-116). Mappings of the Circle. Curvature and Support
Function . 1~5 322
§2 (117-123). Mean Values Along a Circle. 127 324
§3 (124-129). Mappings of the Disk. Area . . • 129 325
§4 (130-144). The Modular Graph. The Maximum Principle 130 326
Chapter 4
Cauchy's Theorem • The Argument Principle
§1 (145-171). Cauchy's Formula. 133 329
§2 (172-178). Poisson's and Jensen's Formulas 138 338
§3 (179-193). The Argument Principle 140 341
§4 (194-206.2). Rouche's Theorem. 142 344
Chapter 5
Sequences of Analytic Functions
§1 (207-229). Lagrange's Series. Applications 145 347
§2 (230-240). The Real Part of a Power Series . 150 354
§3 (241-247). Poles on the Circle of Convergence . 152 357
§4 (248-250). Identically Vanishing Power Series. 153 359
§5 (251-258). Propagation of Convergence 154 360
§6 (259-262). Convergence in Separated Regions . 156 363
§7 (263-265). The Order of Growth of Certain Sequences of Poly-
nomials 157 365

Chapter 6
The Maximum Principle
§1 (266-279). The Maximum Principle of Analytic Functions 157 367
§2 (280-298). Schwarz's Lemma . 160 369
§3 (299-310). Hadamard's Three Circle Theorem. 164 374
§4 (311-321). Harmonic Functions • 165 377
§5 (322-340). The Phragmen-Lindelof Method 166 379

Author Index 385


Subject Index 388

yuana97@uw.edu
Notation and Abbreviations
We tried to be as consistent as possible in regard to notation and abbreviations
and to denote quantities of the same nature by the same symbol, at least within a
section. A particular notation might be specified for one or two sections. Otherwise
the meaning of every letter is explained anew in every problem except whim a
previous problem is referred to. A problem that is closely related to the preceding
one is introduced by the remark "continued"; if it is related to some other problem
the relevant number is mentioned, e.g. "continuation of 288".
We denote Parts by roman, Chapters (if necessary) by arabic numerals. The
numbering of the problems recommences with each Part. The problem numbers
are in boldface. Within the same Part only the number of the problem is given; if,
however, we refer to another Part its number is indicated also; e.g. if we refer to
the problem (or solution) 123 of Part II in a problem (or solution) of Part II we
write "123"; in a problem (or solution) of any other Part we write "11123".
Remarks in square b~ackets [] mean hints in the problems and quotations in
the solutions (in particular at the beginning of the solutions) or references to other
problems that are used in the proof. All other remarks are in ordinary parentheses.
A problem number quoted refers to the problem as well as to the solution except
when the one or the other is stressed, e.g. [solution 38].
Almost always references to the sources are given only in the solution. Prob-
lems that appeared already in print are quoted as such. If the author but no biblio-
graphy is mentioned the problem has been communicated to us as new. Problems
whose number is preceded by the sign * (as *5 in Part II) or contains a decimal
point (as 80.10 in Part I) are new (that is, either not contained in the original
German edition, or contained there, but essentially modified in the present English
version). If the problem is the same as in the original, but the solution has some
essential new feature, the sign * is used only in the solution. The abbreviations of
the names of journals are taken from the index of Mathematical Reviews and, if
not listed there, from World List of Scientific Periodicals Published 1900-1960,
Peter Brown, British Museum, Washington Butterworths, 1963.

The journals most often quoted are:


Acta Math. =
Acta Mathematica, Stockholm
Amer. Math. Monthly =
The American Mathematical Monthly
Arch. Math. Ph'ys. = Archiv der Mathematik und Physik
Atti Accad. Naz. Lincei Rend. = Atti dell' Accademia Nazionale dei Lincei
Cl. Sci. Fis. Mat. Natur. Rendiconti. Classe di Scienze Fisiche, Mate-
matiche e Naturali, Roma
C. R. Acad. Sci. (Paris) Ser. A-B = Comptes rendus hebdomadaires des seances
de l' Academie des Sciences, Paris, Series A
et B

yuana97@uw.edu
XVIII Notation and Abbreviations

Giom. Mat. Battagl,ini = Giornale di Matematiche di Battaglini


Jber. deutsch. Math. Verein. = J ahresbericht der deutschen Mathematiker-
Vereinigung
J. reine angew. Math. = Journal fUr die reine und angewandte Mathe-
matik
Math. Ann. = Mathematische Annalen
Math. Z. = Mathematische Zeitschrift
Nachr. Akad. WiSs. Gottingen = Nachrichten der Gesellschaft der Wissen-
schaften Gottingen
Nouv. AnnIs Math. = Nouvelles Annales de mathematiques
Proc. Lond. Math. Soc. = Proceedings of the London Mathematical
Society

The following textbooks are quoted repeatedly and they are cited by the name
of the author only or by a particular abbreviation (e.g. Hurwitz-Courant; MPR.):

E. Hecke: Vorlesungen fiber die Theorie der algebraischen Zahlen. New York:
Chelsea Publishing 1948.
E. Hille: Analytic Function Theory. Vol. I: Boston-New York-Chicago-Atlanta-
Dallas-Palo Alto-Toronto-London: Ginn & Co. 1959; Vol. II: Waltham/Mass.-
Toronto-London: Blaisdell Publishing 1962.
A. Hurwitz-R. Courant: Vorlesungen iiber allgemeine Funktionentheorie und ellip-
tische Funktionen. 4th Ed. Berlin-Gottingen-Heidelberg-New York: Springer
1964.
K. Knopp: Theory and Applications of Infinite Series, 2nd Ed. London-Glasgow:
Blackie & Son 1964.
G. Kowalewski: Einfiihrung in die Determinantentheorie. 4th Ed. Berlin: Walter
de Gruyter 1954.
G. P6lya: How to Solve It. 2nd Ed. Princeton: Princeton University Press 1971.
Quoted: HSI.
G. P6lya: Mathematics and Plausible Reasoning. Vols. 1 and 2. 2nd Ed. Princeton:
Princeton University Press 1968. Quoted: MPR.
G. P6lya: Mathematical Discovery. Vols. 1 and 2. Cor. Ed. New York: John Wiley &
Sons 1968. Quoted: MD.
E. T. Whittaker and G. N. Watson: A Course of Modern Analysis. 4th Ed. London:
Cambridge University Press 1952.

The following notation and abbreviations are used throughout the book:
an - a means: an tends to a as n _ 00.
an N b.. (read: lin is asymptotically equal to b.. ) means: b.. =1= 0 for sufficiently
a..
large nand - _ 1 as n _ 00.
b..
O(IIn). with an > O. denotes a quantity that divided by a.. remains bounded;
o(a.. ) a quantity that divided by lin tends to 0 as n _ 00.
Such notation is used analogously in limit processes other than n _ 00.
x _ a + 0 means x converges from the right (x _ a - 0 from the left) to a.
exp (x) = e". e is the base of natural logarithms.
Given 1J real numbers al' a l ..... a ... max (al. a2• .... an) denotes the largest (or
one of the largest) and min (al • a l •..•• a.. ) the smallest (or one of the smallest)
among the numbers al •. a l •..•• a... max f(x) and min f(x) have an analogous meaning
for a real function defined on the interval a. b. provided f(x) assumes a maximum

yuana97@uw.edu
Notation and Abbreviations XIX
or a minimum on a, b. Otherwise we retain the same notation for the least upper
and the greatest lower bound resp. (similarly in the case of a complex variable).
sgn x stands for the signum function:
+1 for x > 0
sgnx= { Oforx=O
-1 for x < 0
[x): denotes the largest integer that is not larger than x (x - 1 < [x] ~ x).
Square brackets, however, are also used instead of ordinary parentheses if no
misunderstanding is expected. (Their use in a very special sense is restricted to
Part I, Chap. 1, § 5.)
if is the conjugate to the complex number z.
:For the determinant with general term a;.,,.., A,I4 = 1, 2, ... ,11, we use the
abbreviated notation

Ia;.,..I~ or Ia;.,..I;.,,..=1,2, ... ,n or IaU, aA2, ... , aAn I~.


A non-empty connected open set (containing only interior points) is called a
region. The closure of a region (the union of the open set and of its boundary) is
called a domain. (As this terminology is not the most frequently used, we shall
sometimes overemphasize it and speak of "open region" and "closed domain".)
A continuous curve is defined as the single-valued continuous image ofthe inter-
val 0 ~ t ~ 1, i.e. the set of points z = x +
iy, where x = /p(t), y = lp(t), /p(t) and
lp(t) both continuous on the interval 0 ~ t ~ 1. The curve is closed if /p(0) = /p(I),
11'(0) = 11'(1), without double points if /p(t 1 ) = /p(t 2 ), lp(tl) = 1p(t2 ), tl < t 2 , imply tl = 0,
t2 = 1. A curve without double points is also called a simple curve. A not-closed,
simple, continuous curve is often referred to as simple arc.
A closed continuous curve without double points (Jordan curve) in a plane
determines two regions of which it is the common boundary.
Paths of integration of line, or complex, integrals are assumed to be continuous
and rectifiable.
(a, b) denotes the open interval a < x < b, [a, b) the half-open interval
a ~ x < b, (a, b) the half-open interval a < x ~ b, [a, b) the closed interval
a ~ x ~ b. When we need not distinguish between these four cases we use the
term "interval a, b".
"Iff" is used now and then as an abbreviation for "if and only if".

yuana97@uw.edu
Problems

Part One

Infinite Series and Infinite Sequences

Chapter 1

Operations with Power Series

§ 1. Additive Number Theory, Combinatorial Problems,


and Applications
*1. In how many different ways can you change one dollar? That
is, in how many different ways can you pay 100 cents using five different
kinds of coins, cents, nickels, dimes, quarters and half-dollars (worth 1,
0, 10, 25, and 50 cents, respectively) ?
*2. Let n stand for a non-negative integer and let A.. denote the
number of solutions of the Diophantine equation

x + by + 10z + 200 + 50v = n


in non-negative integers. Then the series

represents a rational function of C. Find it.


*3. In how many ways can you put the necessary stamps in one 'ow
on an airmail letter sent inside the U.S., using 2,4, 6, 8 cents stamps? The
postage is 10 cents. (Different arrangements of the same values are
regarded as different ways.)
4. We call B .. the number of all possible sums with value n (n a
positive integer) whose terms are 1, 2, 3, or 4. (Two sums consisting of

yuana97@uw.edu
2 Operations with Power Series

the same terms but in different order are regarded as different.} The
series

represents a rational function of C. Which one?


5. Someone owns a set of eight weights of 1, 1, 2, 5, 10, 10, 20, 50 grams
respectively. In how many different ways can 78 grams be composed of
such weights? (Replacing one weight by an other one of the same value
counts as a different way.)
6. In how many different ways can one weigh 78 grams if the
weights may be placed on both pans of the scales and the same weights
are used as in problem 5 ?
7. We consider sums of the form

£1 + £2 + 2£3 + 5£4 + 10£5 + 10£6 + 20£7 + 50£8'


where £1' £2' ... , £8 assume the values 0 or 1. We call G.. the number of
different sums with value n. Write the polynomial

as a product.
8. Let £1' £2' ... , £8 assume the values -1, 0, 1. Modify problem 7
accordingly. Let D.. denote the number of different sums of value n
Find the factorization of the following expression (function of C)
99
~ D ..e".
,,--99

9. Generalize the preceding examples by replacing the particular


values of the coins, stamps and weights by aI' a2, ••. , al •
10. An assembly of p persons elects a committee consisting of n of
its members. How many different committees can they choose?
11. There are p persons sharing n dollars. In how many ways can
they distribute the money?
12. There are p persons sharing n dollars, each getting at least one
dollar. In how many ways can they do it?
13. Consider the general homogeneous polynomial of degree n in the
p variables xl> x 2 ' ••• , xp' How many terms does it have?
14. Any weight that is a positive integral multiple of a given unit
can be weighed with the weights 1, 2, 4, 8, 16, ... on one pan of the scales,
and this can be done in exactly one way. That is, any positive integer
admits a unique representation in the binary system.

yuana97@uw.edu
Pt. I. Chap. 1. No. 5-20 3

15. A set of weights 1, 3, 9, 27, 81, ... can be used to weigh any weight
that is a positive integral multiple of a given unit if both pans of the
scales are used, and this can be done in exactly one way.
16. Write
(1 + qC) (1 + qC2 ) (1 + q~) (1 + qC8 ) (1 + qC16 ) •••
= ao + alC + a2C2 + a3C3 + ....

Find the general formula for a".


17. Consider the expansion
(1 - a) (1 - b) (1 - e) (1 - d) ...

= 1- a- b + ab - e + ae + be - abe - d + ....
What is the sign of the n-th term?
18. Prove the identity

(1 + C+ C2 + C3 + ... + C9 ) (1 + CIO + C20 + .. , + C90 )


X (1 + C100 + C200 + ... + C900 ) ... = 1 ~ C'

18.1. The first and the third problem considered in the solution of 9
(concerned with A" and en respectively) are the extreme cases of a
common generalization which, properly extended, includes also 18.
Formulate such a generalization.
18.2. In a legislative assembly there are 2n + 1 seats and three
parties. In how many different ways can the seats be distributed among
the parties so that no party attains a majority against a coalition of
the other two parties?
19.
(1 + C) (1 + C2 ) (1 + C3 ) (1 +~) ... = (1 _ C) (1 _ C3) (~ _ C5) (1 _ C7) •.••

20. Each positive integer can be decomposed into a sum of different


positive integers in as many ways as it can be decomposed into a sum
of equal or different odd positive integers. E.g. the decompositions of I)
into sums with different terms are

6, 1 + 5, 2 + 4, 1 + 2 + 3,.
with odd terms

1 + 5, 3 + 3, 1 + 1 + 1 + 3, 1 + 1 + 1 + 1 + 1 + 1.

yuana97@uw.edu
4 Operations with Power Series

21. It is possible to write the positive integer n in 2,,-1 - 1 ways


as a sum of smaller positive integers. Two sums that differ in the order
of terms only are now regarded as different. E.g. only the seven following
sums add up to 4:

1 + 1 + 1 + 1, 1 + 1 + 2, 2 + 2, 1 + 3,
1+ 2 + 1, 3 + 1.
2 + 1 + 1,
22. The total number of non-negative integral solutions of the follow-
ing Diophantine equations is n 1: +
x + 2y = n, 2x + 3y = n - 1, 3x+ 4y = n - 2; ... ,

nx + (n + 1) y = 1, (n + 1) x + (n + 2) Y = O.

23. The total number N of non-negative integral solutions of the


following Diophantine equations

x + 2y = n - 1, 2x + 3y = n -+ 4y = n - 5, ...
3, 3x
is smaller than n + 2; moreover the difference n + 2 - N is equal to
the number of divisors of n + 2 (d. VIII, Chap. 1, § 5).
24. Prove that the total number of non-negative integral solutions
of the following Diophantine equations is n:

x + 4y = 3n - 1, 4x + 9y = 5n - 4, 9x + 16y = 7n - 9, ...

25. The number of non-negative solutions of the Diophantine equa-


tion
x + 2y + 3z = n

is equal to the integer closest to (n -:-2 3)2 •


26. Let a, band n be positive integers, a and b relatively pr ime to
each other. The number of non-negative integral solutions of the equa-
tion

is equal to [:bJ or [:bJ + 1. [More may be less: to prove a more general


or a more precise theorem may be less trouble.] .
27. Let ai' a2 , ••• , at be positive integers without a common factor
different from 1 and A" be the number of non-negative integral solutions

yuana97@uw.edu
Pt. I. Chap. 1. No. 21-30 5

of

Then we have
. An 1
lim - - = .
n..,.~ n l - l a l a 2 ••• al(l - 1)!

27.1 (continued). We suppose more: we assume that a j and aj are


relatively prime whenever i =F i. Then we can assert more:
An = P(n} + Qn.
where P(x} is a polynomial with rational coefficients of degree 1 - 1 and
the sequence Qn is periodical with the period al a2 ... al :

27.2 (continued). In the particular case 26 where 1 = 2, a l = a,


az = b
An < 1, when n < ab,
An > 1, when n > ab - a - b,

Aab = 2, A ab - a - b = 0 and generally


A n+ab = An + 1.
28. The points in three-dimensional space whose Cartesian coordi-
nates x, y, z are integers are called lattice points of this space. How many
lattice points of the closed positive octant (x > 0, Y > 0, z > 0) lie on
the plane x + y + z = n? How many lattice points of the open octant
(x > O. y > 0, z > 0) are contained in this plane?
29. Let n be a positive integer. How many lattice points (Xl' x 2 , ••• , x p )
of the p-dimensional space lie in the "octahedron"
1Xl 1 + x2 1
1 + x8 1
1 + ... + 1xp 1 :s;: n ?
30. Consider those lattice points in the closed cube
-n < X, y, z :s;: n
that satisfy the condition
-s :s;: x +y +z < s
where n and s are positive integers. The number of such lattice points is
equal to

-.!...
271:
J __+ 1)3 . 2s + 1
" (S
. l2n
n--t

. t
2_
sm--t
_ _2 _
. t
dt .
-" 5m 2 sm 2

yuana97@uw.edu
6 Operations with Power Series

31. Let n > 3. The number of positive integral solutions of


x+y+z=n
that satisfy the additional conditions
x<y+~ y<z+~ z<x+y
is given by
(n + 8) (n - 2) for n even n2 - 1
-8- for n odd.
8 '

§ 2. Binomial Coefficients and Related Problems


The binomial coefficients ( ~) are defined as the coefficients in the
expansion of

(1 + z)l' = (~) + ( ~) z + (~) + ... + ( ~) z' + ... , (~) = 1,


Z2

p denotes any number. If p is a positive integer then (~ ) can be inter-


preted as the number of combinations of p objects taken r at a time, see
10. In the following problems we assume that n is a non-negative integer.
31.1. Prove that

in several ways, especially with and without a combinatorial inter-


pretation, with and without the binomial theorem.
31.2. Prove that

(~)-(~)+(;)-"'+(-1)'(:)' (_1)'(n~1).
32. (~y + cr+ (; r + ... + (: Y= (2:).
33. C;Y - c:r +C;Y - ... - (2n 2: 1Y + (::Y = (-1)" (2:).
34. Put

2: ak ;/' 2: bi = 2: e"z" ,
k=O l=O ,,=0

Deduce from these identities


e" = aob" + a1b"_1 + a2b,,_2 + ... + a"bo,
"I" = 1X0fl" + C) 1X1fln-1 + (;) 1X2fln-2 + ... + IXnflo·

yuana97@uw.edu
Pt. I, Chap. 1, No. iJ1-41 7

34.1. Given ao' aI' a2, ..• Define, for n = 0, 1, 2, ... ,

Derive hence, for n = 0, 1, 2, ... ,

a" = bo - (~ ) b1 + (;) b2 - ••• + (-1)" b".


35. Defining
x"I" = x(x - h) (x - 2h) '" (x - (n - 1) h)

we have thejdentity

(x + y)"IIt = x"I" + (~) x,,-II 1t ylilt + (;) x,,-21" y211t + ... + y"I".
36 (continued). Prove the following generalization of the multinomial
theorem

37. (n) _ 2(n) + 3(n) _ .•• + (_1)"-1 n(n) ={O for n =F 1,


1 2 3 n 1 for n = 1.
38.
(~)_ !(;)+ !(;)_"'+(_1),,-1 !(:)
1 1 1
=1+"2+'3+'" +-,;-.
39. i (_1)"-" 2211 (n ~ ~~ 1) = n + 1.

r(:)x
11=0

40• •~ (: -eX 7(1 - X)"-7 = (x -eX)2 + x(1 :- x) •


41. Put
9'(x) = ao + a1x + ar(x - 1) + y(x - 1} (x - 2) + "',
'P(x) = ao + Q; X + ;: x(x - 1) + ;: x(x - l){x - 2) + "',
then
(~ ) 9'(0) + (~ ) 9'(1) + (; ) 9'(2) + ... + (: ) 9'(n) = 2"'P(n)

yuana97@uw.edu
8 Operations with Power Series

and

(~)9'(0) -(~)9'(1) + (;)9'(2) - ... + (-1)" (:) 9'(n) = (-1)" a"nl.


42.
(~) (0 - n)2 + (~) (2 - n)2 + (;) (4 - n)2 + ...

+ C) (2v - n)2 + ... = 2"n.


43.
(~) (0 - n)2 - (~) (2 - n)2 + C) (4 - n)2 - ···1
+ (-1)"(:) (2v - n)2 + ...

= {O for n =l= 2,
8 forn = 2.
43.1. Verify the identity
(_1),,-1 x"
:E--=e
00 -x 1 1
1:(1+-+-+
1) :/'
... +-~.
00

1 n! n 1 2 3 n n!

§ 3. Differentiation of Power Series


Let y be an arbitrarily often differentiable function of z. We define
the operation (z ~)" Y by the recursion formula
(z~)"y=z~(Z!)"-ly, z!y=zy'.
E.g.

+ c1x + ... + c"x" be an arbitrary polynomial. We define


r
Let I(z) = Co

I (z !) y= coY + c1(z !) y+ ... + c" (z ! y.


44. We have
I (z !)i' = I(k) i'.
45. If I(x) is a polynomial with integral coefficients (d. VIII, Chap. 2,
§ 1) then the sum of the following series is an integral multiple of e, the
base of the natural logarithms,

1( 0) +f(l) + f(2) + ... +f(k) + ...


1! 2! k! •

yuana97@uw.edu
Pt. I, Chap. 1, No. 42-50 9

46. Define I.. (z) by

(z~)"_l_=
dz 1 - z
1"z + 2"Z2 + 3"Z3 + ... =~~,
(1 _ z)"+l
n = 1,2,3, ...

Then I.. (z) is a polynomial of degree n with positive coefficients (except


for the absolute term 1.. (0) = 0). Furthermore
1.. (1) = n!.

47. Let I(x) and g(x) be two arbitrary polynomials, assume however
that g(x) does not have any non-negative integral zeroes. The series

Y = /(0) + /(It z + ~ Z2 + /(3) z3 + ...


g(O) g(l) g(2) g(3)

satisfies the differential equation

This is soluble by quadratures.


48. Suppose that I(x) and g(x) are two polynomials relatively prime,
that the degree of g(x) is not smaller than the degree of I(x) and that
g(O) = 0, g(k) =f= 0 for k = 1, 2, ... The series
Y= 1 +Mz + /(1)/(2) Z2 + /(1)/(2)/(3) z3 + ...
g(l) g(l) g(2) g(l) g(2) g(3)

satisfies the linear homogeneous differential equation

49. The series


Y= 1 + (~)2 Z + (~)2 Z2 + ... + (1 ·3··· (2n - 1))2 Z" + ...
2 2·4 2·4··· 2n

satisfies the differential equation

z(1 - z) dz 2
d2y
+ (1 - dy 1
2z) dz - ""4 y = O.

§ 4. Functional Equations and Power Series


SO. The function
F(z) = (1 - qz) (1 - q2Z) (1 - q3z) "',

can be expanded in a power series

yuana97@uw.edu
10 Operations with Power Series

Find the coefficients Ak by using the functional equation


F(z) = (1 - qz) F(qz) .
51. Let F(z) be the function defined in 50. Find the coefficients of
the power series
1
F(z) = Bo + Blz + B~2 + Bsr + ....
52. Determine the coefficients C. in the identity
(1 + qz) (1 + qz-l) (1 + tz) (1 + qSz-l) ..• (1 + t . -1z) (1 + t .. -1Z- 1 )
= Co + C1 (z + Z-I) + C2 (r + Z-2) + ... + C.. (z" + z-").

53. Deduce the following equation from the identity 52 by taking


the limit:
n (1 + t"-lZ) (1 + q2"-1z -1) (1 -
~ ~

11=1
t") = ~
8=-00
q""z", iqi < 1.

n (1- qll) =
~

n=1
~ (-1)" q-2-,
11=-00
iqi < 1.
55.
~ "(11+1)
1 - q8. 1 - 'I' . 1 - q8 ••• _
1 - q 1 - q8 1 - q5 -11=0
~q-2-
'
iqi < 1.
58. We have for iqi < 1 the relation
1-q 1_ q8 1_nII
_ . __ • __
1-'1'
't~ • _ _
1+q 1+q8 1+q8 1+'1'
••• = 1- 2q + 2q"• - 2q9 + 2q16 - .•.

57. Let iqi < 1. Define
G(z) = -q- (1 - z)
1-q
+ 1-
_q_Z_ (1 -
q8
z) (1 - qz)

+ 1 ~ q8 (1 - z) (1 - qz) (1 - q2Z) + ....


This function satisfies the functional equation
1 + G(z) - G(qz) =' (1 - qz) (1 - q2z) (1 - qaz) .•..
58. Find the coefficients of the power series of the function G(z)
defint:d in 57:
G(z) = Do + D1z + D~2 + Dsr + ....
59. Prove the identity
II II II-I II-k+l
'" (1 - a ) (1 - a ) ... (1 - a ) _
~ k - n, n = 1, 2, 3, ...
k=1 1 - a

yuana97@uw.edu
Pt. I, Chap. 1, No. 51- 60.3 11

Use it to derive the power series of -log (1 - x). [We have


G{q-") = -no n = 0.1. 2.....
where G(z) is the function defined in 57.]
80. The power series
Zl
I(z) = 1 + 3" + 5"z' + "7z8 + ...
satisfies the functional equation

1 (1 ~ Zl) = (1 + Z2) I{z) [39].

§ 5. Gaussian Binomial Coefficients


Let nand k denote integers. and q a variable. We define the Gaussian
binomial coelficient l as
[n J= _1 - q" 1 - q"-l .•• 1 - q"-k+l
k 1-q 1- ql 1-1
for 1 ::;;: k ::;;: n and as [~J = 1 when k = O. If k is not an integer or
does not satisfy the inequality 0 ::;;: k ::;;: n we set

[~J = O.
We suppose initially that q avoids the roots of the denominators (certain
roots of unity). Sometimes we shall find it necessary to emphasize the
dependence on q; then we shall use the more explicit notation [:1
for [:J.
80.1. Show that
lim
q ..... l
[n]
k -
_ (n)
k •
Cf.10.
80.2. Show that

Pay attention to the case k = O.


80.3. Prove the identity in x

n" {l + (-IX) = ~" [:J


k=l k=O
((k-l)/2x;\.

1 Cf. C. F. Gauss: Summatio quarundam serierum singularium, Opera, Vol. 2.


especiallyp.16-17.

yuana97@uw.edu
12 Operations with Power Series

60.4.
[n t 1J = [~J + [k ~ 1J qK-k+l.
60.5. Show that GJ which we have defined as a rational function
of q is, in fact, a polynomial in q, of degree k(n - k)
k(K-k)
[ ~ J= 2 CK,k,~q"·
~=O

The coefficients CK,k,~ are positive integers and "symmetric", that is


CK,k,a< =cK,k,k(K-k)-a< .

(We may regard henceforth [~J as a polynomial, defined for all values
of q.)
60.6. According to the notation explained above [~l, is the ex-
pression which we obtain from [;J by sUbstituting q2 for q. Prove the
identity in z

iI (1 +
h=l
q2h-1Z) (1 + q2/o-1 z -1) = [2nJ + i [
n q' h=l
2n ]
n -" q'
q"'(z" + Z-h).
60.7. Let m, r, and 5 denote non-negative integers. For q = -1 the

[2;J_1 I
Gaussian binomial coefficients assume simple values:
0 when r odd
(7) when r = 25,
=

[2m2s+ 1J = [2m2s ++ 11J = (m)


-1 s . -1

60.8. Show that

[~J - [:J + [;J - ... + (-It[:J


equals 0 when n is odd but

when n is even. [Call F(q, n) the proposed expression. Then you have to
prove that
F(q, n) = (1 - qK-1) F(q, n - 2). for n > 3.

You may try to start by passing from n to n - 1 in using 60.4.]

yuana97@uw.edu
Pt. I, Chap. 1, No. 60.4-60.11 13

We consider a plane with an attached system of rectangular coor-


dinates x, y. A point of which both coordinates are integers is called a
lattice point (of the plane; we consider now only the two dimensional
space, cf. 28, 29). A line parallel to one or the other of the coordinate
axes that passes through a lattice point (and therefore through infinitely
many lattice points) is called "street". These streets constitute a "net-
work of streets" in which the lattice points are street corners; the net-
work divides the plane into square "blocks" with unit sides. Let l' and s
be non-negative integers. A shortest path in the network of streets
between the origin (0,0) and the street corner (1', s) is of length
1'+s=n;
we call it a "zig-zag path". The "area under the zig-zag path" is included
by the path, the "horizontal" coordinate axis y = 0 and the "vertical"
line x = 1'.
80.9. The number of zig-zag paths between the street corners (0,0)
and (1', s) is

80.10. The number of those zig-zag paths between the street corners
(0,0) and (1', s) the area under which is tx equals c""'''' (notation 80.5).
In order to specify one of the (:) zig-zag paths considered in 80.9
we view in succession the unit segments of which it consists starting from
(0,0) and we write x or y according as the segment viewed is parallel
to the x-axis or the y-axis. Thus the zig-zag path specified by the sequence
of letters
xxyxyyx

ends at the point (4, 3) and the area under it is 4.


Take any two letters in such a sequence; they form an inversion
generated by the zig-zag path if and only if they are different and y
comes before x. Thus, in our above example there are four inversions.
80.11. The number of inversions generated by a zig-zag path equals
the area under the path. (Thus 80.10 determines the number of certain
paths, or letter sequences, having a given number tx of inversions.)

§ 6. Majorant Series
Let

yuana97@uw.edu
14 Operations with Power Series

be any complex numbers and

PO' Pl' P2' ... , P.. , '"


be non-negative real numbers. We define

ao + alz+ a2z2 + ... + a..z" + ... = A(z).


Po + Plz + P2Z2 + ... + p"z" + ... = P(z).
If the inequalities

lao I < POI lall :=;: Pl' la2 1:=;: P2' ... , la,,1 :=;: P.. , ...
hold simultaneously for all n we use the notation
A(z) ~ P(z),
in words: "P(z) is a majorant of A (z)" or "A(z) is a minorant of P(z)".
61. If A (z) ~ P(z) and A *(z) ~ P*(z) then we have also
A(z) + A*(z) ~ P(z) + P*(z)
and
A (z) A *(z) ~ P(z) P*(z).
62. If n is a positive integer we have

63. Put

From

deduce the inequalities


la.. l<n, n = 1, 2, 3, ...

64. Let a l , a2 , ••• , at be positive integers. Prove the following rela-


tion twice, a) by applying, b) without applying, the results of 9:
(1 + z".) (1 + z".) ..• (1 + z"l)
1 1
~ (1 - z"') (1 - z"') ••• (1 - z"l) ~ 1 - z'" - z'" - z'" - ••• - z"l

64.1. Let Zl' Z2' ... , z,. denote the zeros of the polynomial
z,. + alz,.-l + a2z"-2 + ... + an and define
Sk = ~ + ~ + ... + z!

yuana97@uw.edu
Pt. I, Chap. 1, No. 61-64.2· Chap. 2, No. 65 15
for k = I, 2, 3, ... Assume that
Is" I <1 for k = I, 2, ... , n.
Then

and [III 21]


Iz"I<2 for k=I,2, ... ,n.
84.2 (continued). Show by an example that the case of equality can
be attained in the relation Ia" I < 1.

Chapter 2

Linear Transformations of Series. A Theorem of Cesaro

§ 1. Triangular Transformations of Sequences into Sequences


85. Consider the infinite triangular array of numbers

Poo,
Pro, Pw
P'IIJ' P21' P22'

Suppose that the numbers P,.. are non-negative and that the sum of
each row is 1:
P,.. > 0, p,.o + P,.l + ... + P.... = 1 for'll = 0, 1, ... , n; n = 0, 1, 2, ...
We transform any given sequence of numbers so' S1' S2' ••• , s,., ... into
a new sequence to' tl' tz, ... , t,., ... in setting

t.. = p,.oso + P"ISI + ... + Pm,S,..


Assuming that the numbers so' S1' ••• , S,., '" are real, show that the
value of t,. is contained between their minimum and their maximum.

°
We have here an important particular case of a linear transformation
of a sequence into a sequence. In defining P,.. = for'll> n we extend
our triangular array into an infinite square array, the matrix of the trans-
formation (P... ) in which P,.. is the element in the n-th row and 'II-th
column.

yuana97@uw.edu
16 Linear Transformations of Series. A Theorem of Cesaro

66 (continued). We say that the transformation is regular if from


the COR vergence of a sequence to a limit we can conclude the convergence
of the transformed sequence to the same limit, that is, if necessarily
lim ttl = s
"-+00
whenever
lim s" = S.
"-+00

The transformation defined in 65 is regular if and only if


lim P". = 0 for v = 0, 1, 2, ...
"-+00

(This is a particular Case of an important theorem of Toeplitz, see 80,


and III, Chap. 1, § 5.)
67. The existence of lim s" implies
"-+00

• 50 + 51 + 52 + ." + 5.. .-
lim = lim s...
"-+00 n+1 "-+00

68. If the sequence Pi' P2' ... , P.. , ... of positive numbers converges
to the positive value P then
"+1 _ _ _ __
lim
"-+00
VPOP1P2 ... P" = p.
68.1. The numbers ao, aI' aa, ... are positive and
a
lim _~+l = p.
K-+co a,..
"
Then lim ~ exists also and has the same value p.

69. Reduce the computation of J~ 1/:; to the computation of


lim
IJ-+oo
(1 +~)".
n
70. Let the two given sequences

satisfy the conditions:


b.. > 0, n = 0, 1, 2, ... ;
bo + b1 + b2 + ... + b.. + ... diverges;
. a..
lim7J=s.
"'-+00 II

yuana97@uw.edu
Pt. I, Chap. 2, No. 66-74 17
Then
. ao + a 1 + a2 + ... + a"
lim = S.
n-+co bO + b1 + b2 + ... + bn
71. Assume IX > O. Reduce the computation of
lim 1,,-1 + 2,,-1 + 3"-1 + ... + ",,-1
"-+-00 n'"

to the computation of
lim (n + 1)" - n" •
"-700 nl%-l

(The value of this limit is well known from calculus.)


72. Let Po' Pl' ... , Pn' ... be a sequence of posW·· numbers that
satisfy the condition
·
I1m P" -- 0 •
n-+co Po + Pl + P2 + ... + Pn
The existence of lim sn = s implies
n-+co

lim soP .. + sl Pn-l + ... + snPo =s.


n-+co Po + PI + ... + Pn
73. The two sequences of positive numbers

Po' Pv P2' ... , p", ... ;


are assumed to satisfy the conditions

lim P.. - 0 ·
11m q.. = 0.
n-+co Po + PI + P2 + ... + P" - , ,,-+00 qo + ql + q2 + ... + q"
Define a new sequence
r.. = Poq.. + Plq.. - l + P2q.. -2 + ... + P..qo' n = 0,1,2, ...
This sequence satisfies again the condition

.
lim "" = O.
"-+00 "0 + "1 + "2 + ... + ""
74. Let
Po' Pv P2' ... , P.. , ... ; .qo, qv q2' ... , q", ...
be defined as in 73, and let

be an arbitrary sequence. Consider


lim soP" + sI P"_1 + s2P,,_2 + ... + s.. Po ,
"-+00 Po + PI + P2 + ... + P"

yuana97@uw.edu
18 Linear Transformations of Series. A Theorem of Cesaro

and

If both these limits exist they are equal. (The proposition is of special
interest if lim s.. does not exist. If lim s.. exists the proposition is a
fI~co 11-+00

consequence of 72.)
75. Let (J > o. If the series

is convergent. then

lim (al
....... 00
+ a2 + as + ... + a..) n -a = O•

(Series of this kind are called Dirichlet series. Cf. VIII. Chap. 1. § 5.)
76. Assume PI> O. Ps> o. Ps> 0•... and that the sequence
Pl' P 2• P a• •.•• p .. = PI + P2 + Ps + ... + P... ••. is divergent. and
limp..p,;l = o. Then
....... 00

(Generalization of 1 + ! + ! + ... + ! ,. . ., logn.)


77. Let Pl' P2' Ps •...• P... •.. and qv q2' qs' ...• q... ... be two sequences
of positive numbers for which

. PI + P2 + Ps + ... + P.. lim qi + q2 + qs + ... + q.. = {J •


lim = IX •
....... 00 np.. . . ..... 00 nq..
IX + {J > O.
Then
lim PI qi + 2P2q2 + SPsqs + ... + nP.. q.. = ~.
....... 00 n2p.. q.. (¥ + fl
78. The series a l + as + as + ... does not necessarily converge if
(al - a..) + (a 2 - a..) + ... + (a.. _l - a.. )

is bounded as n -+ 00. If. however. the additional conditions

al >a2 >aS >···. lima.. =O


....... 00

are satisfied the series a 1 + aa + a3 + ... must converge.

yuana97@uw.edu
Pt. I, Chap. 2, No. 75-82 19
§ 2. More General Transformations of Sequences into Sequences
79. Consider the infinite matrix
POO' POI' PO?' .,.
PlO' PU' P12' .,.
P2O> P21' P22' .•.

Suppose that all the numbers PM are non-negative and that the sum in
each row is convergent and equal to 1 (P ... > 0; 1: PM = 1, for n,
.=0
v = 0, 1, 2, ...). Let so' Sl' ••• , s..' '" form a bounded sequence. Defi.ne
a new sequence to' t1 , t2 , ••• , t.. , ... by setting

t.. = P..oso + P..1s1 + P..2S2 + ... + P...s. + ....


Show that t.. has a value between the upper and the lower bou'ld of the
sequence so' S1' ••• , S .. ' ••• (whose terms are here supposed to be real).
80 (continued). The convergence of the sequence so" SI' S2' ••• to a
limit S implies the convergence of the transformed sequence to' t1, t2 , •••
to the same limit s if and only if
lim PM =0
"-+00

for each fixed v. (This is the necessary and sufficient condition of the
"regularity" of the transformation with matrix (P...); cf. 66.)
81. Assume that the series
c1 + 2c2 + 3cs + 4c4 + ... + nc.. + ...
converges. Then the series
cn + 2c"+1 + 3c"+2 + 4n.. +s + ... = t..
converges too and
lim t.. = O.
"-+00

82. Let the power series


t(x) = ao + a1x + a2x2 + ... + a..x" + ...
be convergent for X = 1 and assume 0 < ex < 1. Then the power series

t(ex) +f'(<x) h +f"(<x) /7,2


1! . 2!
+ ... + j(") (<X)
n!
/7," + ...
is convergent for h = 1 - ex.

yuana97@uw.edu
20 Linear Transformations of Series. A Theorem of Cesaro

§ 3. Transformations of Sequences into Functions.


Theorem of Cesaro
83. Let the functions
11'0(1),11'1(1), 11'2(t) , ... ,11'.. (1), •.•
be non-negative in the interval 0 < I < 1 and assume that
+ 11'1(1) + 11'2(t)
11'0(1) + ... + 11'.. (1) + ... = 1
holds for all 0 < t < 1. The sequence so' 51' 52' ••• , 5.. , ••• is supposed to
be bounded. Construct the function
(/>(1) = 5011'0(1) + 5111'1 (t) + S211'2(t) + ... + S..I1'.. (t) + ....
The range of (/>(t) will fall into the interval between the upper and lower
bounds of the sequence so' 51' 52' •.• , 5.. , .••
84 (continued). Show that

,Eeo (sol1'o(t) + SII1'I(t) + 5211'2(1) + ... + s..I1'.. (t) + ...) = 5

holds for every convergent sequence so' 51' 51' ..• for which lim s.. = 5
.. -+co
if and only if for each fixed ."
lim 11'.(t) = O.
'-+1-0

85. The infinite sequences


/10, aI' az, ... , a.. , •.. ; bo' bi' b2 , ••• , b.. , ...
satisfy the following three conditions:

b.. > 0, n = 0, 1,2, ... ;


is convergent for It I < 1,
b +bl+b.j'-+··· {
o I divergent for t = 1;

Then
/10 + alt + a2t2 + ... + a..t" + ... converges for It I < 1
and

(This proposition is due to E. Cesaro. Several applications will be given


in the sequel.)

yuana97@uw.edu
Pt. I. Chap. 2. No. 83-91 21
86. If the series
ao + al + a2 + ... + a" + ... = s
is convergent then
lim (ao + alt + a2t2 + ... + a"r + ...) = s.
1-+1-0
87. Set
n = 0, 1, 2, 3, ...
If
• So + Sl + S2 + ... + s"
hm = s
"-+00 n +1
exists, then
lim (a o + alt + a2t2 + ... + ant" + ...) = s.
1-+1-0

(This proposition goes beyond 86 only if the series


ao + al + a2 + ... + an + '" diverges (67].)
88. If the following conditions are satisfied:

b" > 0, 1; b" divergent,


,,=0
then
. ao + + a212 + ... + a"t" + ...
all
lim 2 =S,
1-+1-0 bO + bIt + b2 t + ... + bnt" + ...

provided that the series in the denominator converges for It I < 1.


89. The following limit exists and is positive provided 1X is positive:

lim (1 - t)'" (l,,-lt + 2,,-lt2 + 3,,-lt3 + ... + n,,-lt" + ...).


1-+1-0

90. If 0 < k < 1 and if k converges to 1, then

f
1
dx C'V ~lo _1_ [II 202.]
V(1-x2)(1-k2x2) 2 gl-k'
o
91. Let A" and E" be the numerator and denominator, respectively,
of the n-th convergent of the infinite continued fraction
al al al A" al al al al
T1 + J3 + is + ... and so B" = T1 + J3 + is + ... + 12n _ 3' a> O.
Assume that this continued fraction converges. Find its value applying
85 and using the series
A B
-T x", = 1; -T x".
00 00

F(x) = 1; G(x)
,,=0 n. ,,=0 n.

yuana97@uw.edu
22 Linear Transformations of Series. A Theorem of Cesaro

[F(x) and G(x) satisfy a linear homogeneous differential equation of the


second order by virtue of the recursion formulas for An and Bn.J
92. Let (1 > o. If the series
all-a + a22- a + aa3-a + ... + ann-a + ...
is convergent then we have
1.l.i~0 (1 - t)a (alt + a2t2 + aa ta + ... + ant" + ...) = 0 [75J.

93. Show that


lim
1-+1-0
V1 - t i; (t'"
n=l
- 2t2n')

exists and is negative.


94. The two given sequences

satisfy the conditions

b" > 0; :E b"tn converges for all values of t;


,,=0
. a"
hmb=s.
n-+oo n

Then a o + alt + a2t2 + ... + ant" + ... converges too for all values of t
and in addition
. ao + alt + a2t 2 + ... + ai" + ...
lim 2 =s. (Cf. IV 72.)
1-+00 bo + bIt + b2t + ... + bnt" + ...

95. If lim sn = s exists then


n-+oo

. (
hm
1-+00
So + Sl -1't.
+ S2.t2 t") -I
-2' + ... + s" , + ... e =
n.
S.

96. Assume that the sum


ao + al + a2 + ... + an + ... = S
exists. Define

+ al iTt + a2 2!t + ... + an n!t + ....


2 n
g(t) = ao
Then
e-Ig(t) dt = s. f
o
97. The Bessel function of order 0 is defined as

lo(x) = 1 - i!l!
1 (X)2 1 (X)4 (_1)m(X)2m + ....
2 + 2! 2! 2 -'" + m! mT '2

yuana97@uw.edu
Pt. I, Chap. 2, No. 92-97· Chap. 3, No. 98-102 23

We have

Chapter 3

The Structure of Real Sequences and Series

§ 1. The Structure of Infinite Sequences


98. Let the terms of the sequence aI' az, aa, ... satisfy the condition
m, n = 1, 2, 3, ... ;
then the sequence

either converges to its lower bound or diverges properly to - 00.


99. Assume that the terms of the sequence aI' az, aa, ... -satisfy the
condition

Then
an
lim - = W
n-+oo n
exists; w is finite and we have
wn - 1 < an < wn + 1.
100. If the general term of a series which is neither convergent nor
properly divergent tends to 0 the partial sums are everywhere dense be-
tween their lowest and their highest limit points.
101. Let an> 0, lim an = 0 and the series a1 + a2 + ... + an + ...
be divergent. Put a1 + a2 + '" + an = Sn and denote by [sn] the largest
n-+~

integer < Sn' Find the limit points of the sequence


Sl - [Sl]' S2 - [S2]' ... , sn - [sn]' ...
102. Assume that there exists for the sequence t 1, t2, ... , tn' ... a
sequence of positive numbers 8 1 ,82 , ... , 8 n , ••• , converging to 0, for which
tn+! > tn - 8n for all n.
Then the numbers t 1, t z, ... , tn' . " are everywhere dense between their
lowest and highest limit points.

yuana97@uw.edu
24 The Structure of Real Sequences and Series

103. Let "1' "2' ... , Vn , ••• be positive integers, "1 < "2 < V3 < ....
The set of limit points of the sequence

consists of a closed interval (of length 0 if the limit exists).


104. A subsequence whose terms are the successive partial sums of
an absolutely convergent series can be picked out from every convergent
sequence.
+
105. A sequence t 1, t2 , ••• , tn' ... that diverges to 00 contains a mini-
mum (i.e. there exists a tn such that tm > tn for all m).
106. A convergent sequence has either a maximum or a minimum or
both.
The following propositions show that even the most extravagant
sequences behave occasionally like good mannered sequences, i.e. they
show some feature of monotone sequences.
107. Let ll' l2' l3' ... , lm, ... be a sequeace of positive numbers
(positive in the sense of > 0) and let lim inf lm = O. Then there are
m-+~

infinitely many subscripts n for which In is smaller than all the terms
ll' l2' l3' ... , In-1 preceding In' (In < lk' k = 1, 2, 3, ... , n - 1).
108. Let ll' l2' l3' ... , lm, ... be a sequence of positive numbers
(positive in the sense of > 0) and let lim lm = O. Then there ar.e infini-
m-+~

tely many subscripts n for which In is larger than all the terms In+!' In+2''''
following In' (In > lnH' k = 1, 2, 3, ... ). (Not only the conclusion but
also the hypothesis is different from the one in 107.)
109. Given two sequences

m = 1, 2, 3, ...
satisfying the conditions
lim 1m = 0,
m-+~
m-+~

Then there are infinitely many subscripts n such that two different kinds
of inequalities hold simultaneously:

lns" > In-1sn-1' lnsn > In-2 sn-2' ... , lnsn > l1S1' [107, 108.J
110. If the sequence ~\ ~2, ••• , ~, '" tends to + 00 and if A is
larger than its minimum [105J then there exists a subscript n (or several

yuana97@uw.edu
Pt. I, Chap. 3, No. 103-112 25
subscripts n). n > 1, so that the quotients
L .. - L,,_1 L .. - L .. _ 2 L" - L,,_a L"
---- --2- --
3 - , .•. , n
1
are ~ A and the infinitely many quotients
L,,+1 - L .. L,,+2 - L .. L,,+S - L"
--1-- --- --
3 - ' ...
2
are all > A. [The quantities in question can be interpreted as the slopes
of certain connecting lines between the points with cartesian coordinates
(0, L o), (1, L 1 ), (2, L 2 ), ••• , (m, Lm), ... ; Lo = O.
This interpretation leads to a geometric proof of the statement.]
111. Assume that the sequence l]> l2' la' ... , lm' ... satisfies the sole
condition
lim lm = +00.
m-+oo

Let A be larger than II (A > ll)' Then there exists a subscript n, n > 1,
such that all these inequalities hold simultaneously:
1"_/1+1 + ... + 1"_1 + I" < A ~ 1"+1 + 1,,+2 + ... + 1,,+. ,
fI - - v
p = 1, 2, ... , n; 11 = 1, 2, 3, ....
If A tends to infinity then so does n.
112. Let the sequence ll' l2' ... , lm' ... satisfy the two conditions
lim lm
m-+oo
= 0,
m-+oo

and assume II > A > O. Then there exists a subscript n, n > 1, such
that the inequalities
1"_/1+1 + ... + 1"_1 + I" > A> 1"+1 + 1.. + 2 + ... + l,,+>,
fA - - v
p = 1, 2, ... , n; 11 = 1, 2, 3, ...

hold simultaneously. If A tends to 0, then n tends to infinity.

§ 2. Convergence Exponent
The convergence exponent of the sequence r1 , r2 , r a, ... , rm' ... where
o < r1 < r2 < "', lim rm = 00, is defined as the number;' having the
m .... oo
following property:
rIO' + riO' + riO' + ... + r;;;O' + ...

yuana97@uw.edu
26 The Structure of Real Sequences and Series

converges for q > A. and diverges for q < A.. (For q = A. it may converge
or diverge.) For q = 0 the series is divergent, therefore A. >0. If the
series does not converge for any q then A. = 00.
113. Show that
. logm
lim sUP-I- =A..
"'-+00 og 1'",

114. Let Xl' X 2 ' X a' ••• , X"" ••• be arbitrary real numbers, x'" =1= O.
If there exists a positive distance IS such that IXI - xk I > IS, 1< k,
I, k = 1, 2, 3, ... , then the convergence exponent of Ix1 1, Ix2 1, Ix3 1, ... ,
IX", I, ... is at most l.
115. Let {J be larger than the convergence exponent of the sequence
'1' '2' '" Then there exist infinitely many subscripts n for which the
n - 1 ineqUalities
1 1
1'2
< (~)7i, ... , 1'.. _1 < (~)7i
1'.. n 1'.. n
are satisfied. [107.]
116. Assume that the convergence exponent A. of the sequence
'1' '2' 'a' ... , , "" ... is positive and 0 <IX < A. < {J. Then there are
infinitely many subscripts such that the two types of inequalities
1
1',. > (.!!....):¥" for p = n - 1, n - 2, ... , 1,
1'.. n
1

r.
1'..
> (~)7i
n
for v = n + 1, n + 2, n + 3, ...
hold simultaneously. [109.]

§ 3. The Maximum Term of a Power Series


117. Suppose 0 < '1 < '2 < 'a < .... For what values of x, X > 0,
is the m-th term of the series

larger than all the other terms, m = 0, 1, 2, ... ?


118. Assume

• I'm
lim-=oo.
m4-oo S'"

yuana97@uw.edu
Pt. I, Chap. 3, No. 113-121 27

Then arbitrarily large values of nand r can be found for which the
following inequalities hold simultaneously [111]:

k = 0, 1, 2, 3, ...

(At the root of this fact, and of 122, lies the comparison of two power
series

Suppose that Po > 0, PI > 0, ... , Pm > 0, and that P. =l= for at
least one subscript i. Let e, e> 0, possibly e = 00, be the radius of
°
convergence of the power series
Po + PIX + P2 X2 + ... + Pmx'" + ....
The sequence
Po' PIX, P2x2, ... , Pm xm , ...
° °
converges to if < X < e. Therefore there exists [105J a maximum
term whose value is denoted by ,u(x). I.e.
Pmx'" ~ ,u(x), m = 0, 1, 2, ...
The central subscript v(x) is the subscript of the maximum term, i.e.
,u(x) = P'(X)x·(X). If several of the terms Pmxm are equal to ,u(x) we call
v(x) the largest of the corresponding subscripts. More details in IV,
Chap. 1.
119. For an everywhere convergent power series in X which is not
merely a polynomial the central subscript v(x) tends to CXJ with x.
120. The subscript of the maximum term increases as x increases.
(One might consider this situation as somewhat unusual: in the course
of successive changes the position of maximum importance is held by
more and more capable individuals.)
121. The series
Po + PIX + P2X2 + ... + Pmxm + ...
with positive coefficients and finite radius of convergence e (Pm> 0,
e > 0) is such that one term after the other, all terms in turn, become
maximum term. Then ~ is the radius of convergence of the series
(!

-+-+-+
1
Po
x
Pi
x
P2
... +-+
xm
2

Pm
....

yuana97@uw.edu
28 The Structure of Real Sequences and Series

122. The dominance of the maximum term is more pronounced in an


always convergent series than in one that does not always converge (it
is strongest in a polynomial). More exactly: let the radius of convergence
of the power series
tlo + a1x + ~X2 + ... + a".x'" + ...
be infinite and that of the power series
bo + b1y + b2y2 + ... + b".y'" + .. .
be finite. Suppose a". > 0, b". > 0, m = 0, I, 2, ...
The coefficients
bo' b1 , bs' ••. be such that all the terms b".y'" become in turn maximum
term [120J. Then a value y can be determined for certain arbitralily
x
large positive such that for these corresponding values the respective
series have the same central subscript. Let the common central sub-
script be n. Then all the following inequalities hold simultaneously:
ax" by"
-~ ~ -"- ~ 1 k = 0, I, 2, ...
a,;;" - b,.Y' - ,
00 a
[Consider the maximum term of 1: b'" z"'.]
".=0 ".

123. If there are values x* to which no y corresponds in the sense of


122 then they are "rare". They have a finite logarithmic measure, i.e.
the set of points log x*, x* exceptional value, may be covered by count-
ably many intervals of finite total length.

§ 4. Subseries
Let tl' t2, ts' ... , t.. , ... be integers, 0 < tl < t2 < ts < .... The series
a, + a, + a, + ... + a, +. ..
J I I "
is called a subseries of the series
al+~+a3+"·+a.. +"·,
124. From the harmonic series
~+.!..+~+ ... +~+ ...
1 2 3 n
remove all terms that contain the digit 9 in the decimal representation
of the denominator. The resulting subseries is convergent.
125. If all the subseries of a series converge then the series is absolu-
tely convergent.
126. Let k and 1 denote positive integers. Must the convergent series
a 1 + as + a3 + ... be absolutely convergent if all its subseries of the

yuana97@uw.edu
Pt. I, Chap. 3, No. 122-131 29

form
a" + aHI + aH21 + aH31 + ...
(subscripts in arithmetic progression) converge?
127. Let k and l be integers, k :?: 1, l :?: 2. Must the convergent series
a 1 + all + as + ... be absolutely convergent if all its subseries of the
form
a" + a", + aid' + a"" + ...
(subscripts in geometric progression) are convergent?
128. Let q;(x) denote a polynomial assuming integral values for
integral x, q;(x) = cox' + C1x'-1 + ... [VIII, Chap. 2]. Assume that
the degree is l > 1 and that the coefficient Co of x' is positive (co> 0).
The values q;(0). q;(1). q;(2), ... form a generalized arithmetic progression
of order l; since Co > 0 only a finite number of terms of the progression
can be negative. Must a convergent series a1 + as + as + ... converge
absolutely if all its subseries whose subscripts form a generalized arith-
metic progression
a9'(O) + a9'(l) + a9'(2) + a9'(S) + ...
(omitting the terms with negative subscripts) converge?
129. If the series a1 + all + as +' ... converges absolutely and if
every subseries
a, + a21 + aSI + "', l = 1, 2, 3, ...
has the sum 0 then al = as = ... = o.
130. Consider the set of points determined by all the subseries of
222 2
3"9" 27 + + + .. , + 3" + ....
This set is perfect and nowhere dense (closed and dense in itself. but
nowhere dense in the set of all real numbers). (We have to consider all
the subseries. finite and infinite. including the "empty" subseries to
which we attribute the sum 0.)
131. Let the terms of the convergent series

PI + P2 + Ps + ... + P.. + ... = s


satisfy the inequalities
Pl:?:P2:?:PS:?: .. ·•
o< P.. :=:;; P"+l + P.. +2 + P"+s + ... . n = 1, 2, 3, ...

Then it is possible to represent any number (f in the half-closed interval

yuana97@uw.edu
30 The Structure of Real Sequences and Series

o< < (f s by an infinite subseries:

+ Pt. + Pt. + ... =


PI, (1.

132. Find the series PI + P2 + Pa + ... that satisfies the conditions


1
PI =2' P.. = P.. +] + P"+2 + P"+3 + ... , n = 1, 2,3, ... ,
and verify that in this case every (f mentioned in 131 can be represented
by one infinite subseries only.

§ 5. Rearrangement of the Terms


132.1. By rearranging the factors of the infinite product

(1 + ~) (1 - !) (1 + !) (1 - ~) (1 + ~) ... = Pl,l

we obtain the infinite product

(1 +~)
P p,q = 2 (1 +~)
4 ... (1 +~)
2p

X(1-~)(1-~)"'(1--,1-)(1
3 5 2q + 1 .
+_1_)
2p + 2
...
in which blocks of P factors greater than 1 alternate with blocks of q
factors smaller than 1. (Factors of the same kind remain in the "natural"
order.) Show that
pp,q=Vf· [II 202.]
132.2. By rearranging the terms of the infinite series

-21 - -31 + -41 - -51 + ... = 5 I,] = 1 - log 2


we obtain the infinite series
111 111 1 1
5p,q = 2' + '4 + 6' + ... + 2p - 3 - 5' - ... - 2q + 1 + 2p + 2 + ...
in which blocks of P positive terms alternate with blocks of q negative
terms. (Terms of the same kind remain in the "natural" order, steadily
decreasing in absolute value.) Show that
1 p
5p,q - 5].1 =2 Iog -q'

Let r1, r2 , r3 , ••• , s1> S2' S3' ••. be two sequences of steadily increasing
natural numbers without common terms. Suppose furthermore that all
positive integers appear in one or the other of the two sequences

yuana97@uw.edu
Pt. I, Chap. 3, No. 132-136 31
(r", < r"'+1' s.. < s"+1' r", Ee: s.. for m, n = 1, 2, 3, ...). The two series

a" + a" + a" + "', as, + as, + as, + ...


(the "reds" and the "blacks") are complementary subseries of the series
a l + a2 + a3 + .... Let "1' "2' "3' ... be a sequence of integers such that
each natural number 1,2,3, ... appears once and only once in it (a
permutation of the natural numbers). The series

a." + a., + a•• + ... + a... + ...


is obtained from the series
al + az + as + ... + a.. + ...
by rearrangement. We call special attention to the rearrangements where
a, (m = 1, 2, 3, ... ) is the r",-th term before as well as after the re-
'"
arrangement, i.e. which preserve the subseries a" + a" + a" + ....
If before as well as after the rearrangement a, precedes a, and as
'" '" ..
.
precedes as for all number pairs m, n. In < n, we say that the rearrange-
ment shifts the two complementary subseries relatively to each other (and
it leaves each in its original order).
133. If one of the two to each other complementary subseries of a
convergent series is convergent the other is convergent too. A rearrange-
ment which only shifts the two subseries relatively to each other does
not change the sum of such a series.
134. If one of the two to each other complementary subseries of a
conditionally convergent series diverges to + 00 then the other diverges
to - 00. Provided that all the terms of one of these two subseries are of
the same sign it is possible to obtain an arbitrary sum for the whole
series by shifting the two subseries relatively to each other.
135. It is not possible to accelerate by rearrangement the divergence
of a divergent series with positive monotone decreasing terms.
136. By rearranging the series we can slow down arbitrarily the
divergence of a divergent series with positive terms which tend to O.
More explicitely: Assume
P.. > 0, lim P,. = 0,
"-+00
lim
11-+00
(PI + pz + ... + P..) = 00.

0< QI < Q2 < ... < Q.. < ''', lim Q,. =
"-+00
00.

Then there exists a rearranged series p" + P., + P" + ... + P.,. + ...
such that P•, +' P., + ... + P',. <
= Q,. for n = 1,2,3 ...

yuana97@uw.edu
32 Miscellaneous Problems

137. Assume that


al + a2 + as + ... + a.. + ... = s is convergent,
is divergent.
Let s' < s < sIt. By a rearrangement that leaves all the negative terms
at their places the sum s' can be realized; by a rearrangement that leaves
all the positive terms at their places the sum s" can be realized [136].

§ 6. Distribution of the Signs of the Terms


138. Assume that P.. > 0, PI ~ P" ~ Ps ~ ... and that the series
PI + P2 + Ps + ... + P.. + .. .
is divergent and the series
e)PI + e2P2 + eaPa + ... + e..p.. + ... ,
where e" is -lor +1, is convergent. If under these conditions a certain
percentage of terms is positive then it is 50%. More precisely:
~1+e2+ .. ·+e e1 +£2+· .. + e
lim inf n .. =:;: 0 < lim sup n ..
"-+00 "-+00

139. Suppose P.. > 0, PI ~ P" ~ Pa ~ ... and that the series
e1P) + e2P2 + eaPa + ... + e..p.. + ... ,
where e" is -lor +1, is convergent. Then
lim (e1
...... 00
+ e2 + ea + ... + e.. ) P.. = O.

(Notice the two wellknown extreme cases


el = e" = es = ... and e1 = -e2 = es = -e, = ....)

Chapter 4

Miscellaneous Problems

§ 1. Enveloping Series
We say that the series Ilo + a1 + ~ + ... envelops the number A if
the relations
IA - (ao + a1 + a2 + ... + a..) I < Ia.. +1 I' n = 0, 1, 2, ...

yuana97@uw.edu
Pt. I, Chap. 3, No. 137 -139 • Chap. 4, No. 140-145 33

are satisfied. The enveloping series may be convergent or divergent; if


it converges its sum is A. Assume that A, flo, ai' a2 , ••• are all real. If we
have
A - (ao + a l + a 2 + ... + all) = Olla,,+!, for all n = 0, 1, 2, ...
and 0 < 0,. < 1.
the number A is enveloped by the series ao + al + a2 + ... , and in fact
it lies between two consecutive partial sums. In this situation we say
the series is enveloping A in the strict sense. G. A. Scott and G. N. Watson
[Quart. J. pure appl. Math. (London) Vol. 47, p.312 (1917)] use the
expression "arithmetically asymptotic" for a ciosely related concept.
The terms of a strictly enveloping series have necessarily alternating
signs.
140. Suppose that f(x) is a real function of the real variable x. If
the functions \I'(x) \' \I"(x) \, ... are steadily and strictly decreasing in
the interval [0, xJ, x > 0, then f(x) is enveloped in the strict sense by its
Maclaurin series.
141. The functions
e-%; log (1 + x); (1 + x) -p, p>o;
are strictly enveloped by their Maclaurin series for x > o.
142 (continued). Prove the same for the functions l
cos x, sin x.
143 (continued). Prove the same for the functions

arc tan x, 1
Jo(X) = 1 - ffIT 2"
(,,)2 + 2! 12! (,,)4
2" - ' " [141,142] .

144. Suppose that the terms of the series ao + al + as + ... are


alternately positive and negative and that there exists a number A such
that
A - (a o + al + a2 + .. , + a,,)
assumes always the same sign as the next term, all+!' Then the series
envelops A in the strict sense.
145. If the series ao + a) + as + "', all real, n = 0, 1, 2, ... ,
envelops the real number A and if in addition \all > Iaal > ItZa I > .. .

1 Obviously only the non-vanishing terms of the Maclaurin series are to be


considered. E.g. the n-th partial sum of the Maclaurin series for COS" is
,,2 ,,4 "2,,
1 - -2! + 4! - ... + (-1)" - -,
(2n)!
n = 0, 1,2, ...

yuana97@uw.edu
34 Miscellaneous Problems

then the terms al> a2 , a3 , '" have alternating signs and A is enveloped
in the strict sense.
146. Let the function I(x) assume real values for real x, x > R > O.
If I(x) is enveloped fot x > R by the leal series a o + axl + xa~ + -~;,;- + ...
then the numbers aI' a2 , a3 , ••• have alternating signs and the series is
strictly enveloping.
147. Suppose that the real valued function I(t) is infinitely often
differentiable for t > 0 and that all its derivatives P')(t) (n = 0, 1, 2, ... )
have decreasing absolute values and converge to 0 for t -+ 00. Then the
integral

o
J
I(t) cos xt dt

is, for real x, strictly enveloped by the series

- f'---xz
(0)
+ 1'"-;4
(0)
-
fV (0)
---;G + ---xs-
fVII (0)
- ....
(Example: I(t) = e- t .)
148. The number : is enveloped by the series
31313131
4 +
4 - "8 - "8 16 16 - 32 - 32 + + + ... ,
but not in the strict sense.
149 1• Plot the first seven terms of the series
i 1 i 1 i 1
e
i
=1+-----+-+----···
11 21 31 4! 51 61

successively as complex numbers and compute so the value of ei to


three decimals.
150. Let ~ denote a ray with origin z = O. Assume that along ~ all
the derivatives of the function I(z) attain the maximum of their absolute
values at the origin and only there; i.e.
I/(n)(o) I> I/(n)(z) I

whenever z is on ~ and Iz I > O. Then:


a} The function I(t) is enveloped for every z on ~ by the Maclaurin
series
[140] .

1 In 149-155 the terms of the series are complex numbers; they are regarded
as points in the Gaussian plane (complex plane) [III 1 et seqq.J.

yuana97@uw.edu
Pt. I. Chap. 4. No. 146-155 35

b) the function F(z) = j e- I I( : )dt is for every z on ~ enveloped


by the series 0

1(0) +1'(0) +f"(0) +/",(0) + ... [147J;


Z z8 z3

•~ is the mirror image of ~ with respect to the x-axis; the integral is


taken along the positive t-axis and converges under our assumptions
on I(z).
151. The Maclaurin series of e-', log (1 + z) and (1 + z)-P, p > 0,
envelop the respective functions for mz > 0, z =l= O.
152. Let z be restricted by the following conditions:

z =l= O.
I" 00 ,.

Then the function e"2 J e-"2 dt is enveloped by the series



-1z - -1 1·3
z3 +z5
---1.3·5
- + ... Z7

(strictly enveloped if z is real).


153. Suppose that a" and btl are arbitrary complex numbers that
a
have the same argument, i.e. btl is real and positive. If at a certain point
z =l= 0 the two series "

ao + a1z + a2z'l + ... + a..z" + ... , bo + blz + b2z2 + ... + b..z" + ...
envelop the values 91(z) and 1p(z) then the combined series

envelops the value 91(z) + 1p(z) for this z. (The same is true for enveloping
in the strict sense if all the coefficients and z are real.)
154. If z lies in the sectors described in 152 the function z coth z is
enveloped by its power series
zcothz=z e' + e-' = 1 +B (2z)2 _ B (2z~ + B. (2Z)B + ...
e' _ e -. 1 2! 2 4! ;s 6!

(strictly enveloped if z is real). The coefficients B]. B 2, B3 • ••• are called


the Bernoulli numbers.
155. The function
w(z) = log r(l + z) - (z + 1) log z + z -llog (2n) [II 31]

yuana97@uw.edu
36 Miscellaneous Problems

can, for mz > 0, be written as an integral


f
co arctan'!""
w(z) = 2 2'" Z de.
o e - 1

[Cf. E. T. Whittaker and G. N. Watson, pp. 251-252.] Prove that the


resulting (divergent) Stirling series
Bl B. B.
1.2 ·z - 3·4. za + 5·6· z6 - •••

envelops the function w(z) if mz > 0 and - : <arg z ~ : .

§ 2. Various Propositions on Real Series and Sequences


158. Assume that 9'(x) is defined fOJ, positive x and that for x large
enough it can be represented in the form

9'(x) = ao + 7~ + ax: + ... + a"


~ + ...
x
a" real, k = 0, 1, 2, ... The infinite series
9'(1) + 9'(2) + 9'(3) + ... + 9'(n) + ...
converges if and only if ao = at = O.
157 (continued). Suppose 9'(n) =f= o. The infinite product
9'(1) 9'(2) 9'(3) ... 9'(n) '"
converges if and only if a., = 1, at = O.
158 (continued). Under what conditions does the following infinite
seJies converge
9'(1) + 9'(1) 9'(2) + 9'(1) 9'(2) 9'(3) + ... + 9'(1) 9'(2) 9'(3) ... 9'(n) + ... ?
159. For which positive values of IX does the following series converge
co

~ (2 - e") (2 - e"'2) ... (2 - e"''') ?


,,=1
160.
J1x-"dx = --
~n-".
o ,,=1

161. Considering positive square roots only we find

yuana97@uw.edu
Pt. I. Chap. 4. No. 156-167 37
162. Let al' a2 • ••• , a", ... be positive numbers and put

tIl = Val + -Va2 + ... + Va",


where only positive square roots are considered. The convergence of the
sequence
(t)
is related to
. log log a"
lim sup - - - = IX
n
in the following way:
if IX < 2 then (t) converges,
if IX > 2 then (t) diverges.
163 (continued). The sequence (t) is certainly convergent if the series
00

I 2-"a,,(a1a2 ... a,,)-l converges.


,,=1
164. We have for 0 < q < 1
1 - q (1 - q2)1 (1 - q' )1 (1 - q8)1 ••• = (1 _ )2
1+q 1+q2 1+q' ~ q .
165. The series
" "e,
... + I" (x) + I' (x) + f(x) + f f(El ) dEl + f dEl f f(E2) dE. + ...
000
is infinite in two directions. Suppose it does converge uniformly on some
interval. Which function does the series represent?
166. Let 'P,,(x) and 'I',,(x) denote the polynomials of degree H, n =
0, 1, 2, .•. defined by the formulas
'Po(x) = 1, 'P:(x) = 'P"-l(X), 'P,,(O) = 0,
'l'o(x) = 1, +
'I',,(x 1) - 'I',,(x) = '1'''-1 (x) , '1',,(0) =·0, n = 1, 2.3, ...
Find the two sums 'P(x) and 'I'(x) ,
'P(x) = 'Po(x) + 'PI (x) + ... + 'P,,(x) + "',
'I'(x) = 'I'o(x) + 'l'J (x) + ... + 'I',,(x) + ....
167. We define
1 1
-fS--
= Y e-12" y" = n! n en, n = 1, 2, 3, ...
" , 2
N"
"""',..

Then each interval (x", y,,), k = 1, 2, 3, ... , contains the interval


(xk+J' Yk+l) as a subinterval.

yuana97@uw.edu
38 Miscellaneous Problems

188. The sequence


a,,= ( 1 +-;-1 )"+1' , n = 1, 2, 3, ...

is monotone decreasing if and only if p ~ i.


188. The sequence
a" = (1 + !)" (1 + :), n = 1, 2, 3, ...

is monotone decreasing if and only if x ~ i.


170. Let n be a positive integer. Then we have
e
2n+2<e-
(1)"
1+-;- <2n+l' e

lim (1 + ~)" is contained


. 171. As is well-known, the number e = ,,-.00 n
in the interval
(1 + !)" < e < (1 + !)"+l [188]

In which quarter of the interval is it contained?


172. The sequence
X)"+1 ,
a" = (1 +-;- n = 1, 2, 3, ...
is monotone decreasing if "and only if 0 < x ~ 2.
173. Prove that the n-times iterated sine function

sin,. x = sin (sm,.-1 x), sin1 X = sin x,

converges to 0 as n ~ 00 if sin x > 0 and that, moreover,

r 3"
lim 1 Ins sin,. x = 1.
11-+00

174. Assume that 0 < I(x) < x and

I(x) = x - a:l' + br + re(x), lime(x) = 0


s-+O

for 0 < x < xo, where 1 < k < l and a, b positive. Put

Then we have for n ~ 00


1 1
n"-1 v" ~ [(k - 1) a] - "-1 .

175. Discuss the convergence of the series

~+~+V;+''''

yuana97@uw.edu
Pt. I. Chap. 4. No. 168-180 39
where
V1 = sin x, V2 = sin sin x, ... , Vn = sin vn - 1' •••

We can obviously assume that VI > O.


176. Prove the formula
~ - 1= u1 + U 1U 2 + U 1U2U3 + ...
where
e"n - 1
U nH = log - - - , n = 1, 2, 3, '"
Un
177. Compute

s= COS3 P
1
- 3" COS3 &p + 31 COS3 32<p -
2
1
38 COS3 33 <p + ....
178. Let an' bn , bn =f= 0, n = 0,1, 2, 3, ... , be two sequences that
satisfy the conditions:
a) the power series I(x) = ~ a~xn has a positive radius of conver-
gence r.
b) the limit
. bn
lim--=q
"-+00 b.. + 1

exists and Iq I < r.


We define
Cn = aobn + a 1bn - 1 + ... + a,.bo , n = 0, 1, 2, ...
Cn
Then b converges to t(q) as n ~ 00.
1$

179. Let
n = 1, 2, 3, ...

be arbitrary functions, Ia,.k I < A for all positive integral values of n


and k and
lim t,.(x) = 0 if 0 < x < 1.
"-+00

Then we have for fixed k, k = 1, 2, 3, ...


lim a..k = O.
1$-+00

180. Suppose that the series


a no + a n1 + a n2 + ... + a nk + ... = s", n = 0, 1, 2, .. "
have a common convergent majorant
Ao + Al + A2 + ... + Ak + ... = 5,

yuana97@uw.edu
40 Miscellaneous Problems

i.e. for each k the inequalities Ia.." I : : ; A" hold simultaneously for all n.
Assume furthermore that

exists for k = 0, 1, 2, ... Then the series

ao + a l + tlsl + ... + a" + ... = s


is convergent and
lim s.. = s.
"-+00
181. Justify the limits in 53 and 59.
181.1. Assume that

converges for i = 1, 2, 3, ... , define U. as the least upper bound of

n = 1, 2, 3, ... ,
and assume that

converges. Then the series

(*) an + a12 + a2l + alB + tlsl2 + a Sl + ... + a l .. + tlsl,"-l + "',


which you obtain by arranging the numbers in the array

azl tlsl2 azs • • • a2 ... -1 • •

aSI aS2 asS .. • . . . . • ..

~'diagonally", converges and its sum is

SI + S2 + ... + s .. + ....
(The interesting point is that the absolute convergence of the double
series EEa." is not assumed.)
182. If ex is fixed, ex> 0, and n an integer increasing to + 00, then

I' v"'-I(n" -
co
v"}-2 ex>
1-.. ( )2
_n_!!.. .
• =1 3 ex

The meaningless term with subscript v = n has to be omitted, which is


indicated by the comma at the summation sign. Notice the case ex = 1.

yuana97@uw.edu
Pt. I, Chap. 4, No. 181-185.2 41
183. Let the numbers 80 , 81 , 82 , ••• , 8", ... assume one of the three
values -1, 0, 1. Then we have
1/ ,/ V
80r2+81r2+82 2+···=2sin 4~
(n 00 BOB)B2 "'B)
,,".
,,=0 2
(The left hand side must be interpreted as the limit of

80l/2 + 81 V V
2 + 82 2 + .. , + 8" V2, n = 0, 1, 2, ...
for n ~ 00. These expressions are well defined for all n. Non-negative
square roots are used throughout.)
184. Every value x of the interval [-2,2] can be written in the form

X=80~81 V2 +82 V2 + ...


where 8", R = 0, 1, 2, ... , is either -lor +1. The representation is
unique if x is not of the form 2 cos L Tt, p, q integers, 0 < p < 2'1. These
2q
are the only numbers that may be written in the finite form

The finite representation can be extended to an infinite one in two ways:


by putting either

or
8,,+1 = -1, 8"+2 = -1,
185 (continued). The number x is of the form x = 2 cos kTt, k ra-
tional, if and only if the sequence 80 , 811 82; ... is periodical after a certain
term.
185.1. Construct a sequence of real numbers a1 , a2 , ••• , a", ... so that
the series
ai+a;+"·+a~+···

diverges for 1 = 5, but converges when 1is any odd positive integer differ-
ent from 5.
185.2 (continued). Let the set of all odd positive integers be divided
(arbitrarily) into two complementary subsets C and D (having no ele-
ment in common). Show that there exists a sequence of real numbers
a], a2 , ••• , an' ... such that the series mentioned converges when l belongs
to C and diverges when l belongs to D.

yuana97@uw.edu
42 Miscellaneous Problems

§ 3. Partitions of Sets. Cycles in Permutations


A partition of a set S is formed by disjoint subsets of S the union of
which is S. "Disjoint" (or non-overlapping) means that the intersection
of any two subsets involved is the empty set. If k subsets are involved
in the partition none of which is the empty set we speak of a partition
into k classes.
We let S; stand for the number of different partitions of a set of n
elements into k classes and T" for the total number of its different parti-
tions (into any number of classes) :
T" = s;' + oS; + ... + S:.
The S~ are called the Stirting numbers 01 the second kind.
*188. Tabulate the numbers S; for n ~ 8, 1 ~ k ~ n.
*187. Obviously

Show that

*188. Show that


s: S:+1 s~ 1
z"+1 + z"+2 + ... + z"+! + ... = z(z - 1) (z - 2) ••• (z - h) •

*189. Show that, for n > 1,


S: = :! [k" - ( ~ ) (k - 1)" + ( : ) (k - 2)" - ... + (-1)" 0"] .
*190. Show that
co s:z" (6' _ l)k
~-=
"=k n! hi
.
*191. Prove the identity in x
x" = S~x + S;x(x -(1) + ... + S:x(x -1) ... (x - n + 1).
[189, III 221.]
*192. Prove the identity of 191 independently of 189 [by a com-
binatorial argument] and derive hence a new proof for 189.
*193. Define To = 1. Then
T
o
+ T1!z + T2!z2 + ... + _
_1 _1_
T"z"
n!
+ ... = ee' -1 •

*194. Show [by a combinatorial argument] that

T,,+! = (~) T" +(~) T"_1 + (;) T"_2 + ... + (:) To·

yuana97@uw.edu
Pt. I, Chap. 4, No. 186-200 43
*195. Use 194 to prove 193 [34].
*196. Show that for n > 1

T" = e1 (1" 2" 3" )


11 + 2f + 3! + ....
We let s~ stand for the number of those permutations of n elements
that are the products of k disjoint cycles 1 • Obviously
s~ + s~ + s; + ... + s: = n!.
The s~ are called the Stirling numbers 01 the first kind.
There are 1t persons seated around k round tables (where all seats are
equal) so that at least one person sits at each table. We may regard as
essential
(1) who sits next to whom, and whether A is the left-hand or right-
hand neighbor of B.
Or we regard as essential only
(2) which people sit at the same table, no matter in which order.
The number of different seatings is s~ in case (1) and S~ in case (2).
Obviously
s~ 2 S~.
*197. Tabulate the numbers s~ for n ::;; 8, 1 < k < n.
*198. Obviously
s~ = (n - 1)!,
Show that
S~+1 = SZ-1 + ns~.
*199. Show that
x(x + 1) (x + 2) .•• (x + n - 1) = s~x + S;X2 + '" + s:x"
or, which is the same, that
x(x - 1) (x - 2) ... (x - n + 1) = (_1)"-1 s~x + ... - S:_Ix"-1 + s:x".
*200. Show that
co sZ z"
~--=- log--
1 ( 1)k .
n=k n! k! 1 - z

(Compare 200 with 190, 199 with 188, and again 199 with 191. See also
VII 54.2 and VIII 247.1.)
Define SZ as the number of partitions of a set of n elements into k
classes each of which contains more than one element.
I See, e.g., Garrett Birkhoff and Saunders MacIane: A Survey of Modern
Algebra, 3rd Ed. New York: Macmillan 1965, p. 137.

yuana97@uw.edu
44 Miscellaneous Problems

·201. Tabulate the numbers s: for n ~ 8,1~ k~ n.


·202. Obviously
~= 1 when n ~ 2, s: = 0 when n < 2k.
Show that

·203. Show that


S; = 2"-1 - n - 1, S!" = 1 • 3 . 5 ... (2n - 1).
·204. Returning to S:, show that
~,,-1 = (")
2'
S",,-2 =(")3"-6
3 4'
~,,-3 =(")("-2)("-3)
4 2
and, generally, that S: __ is a polynomial in n of degree 2a divisible by
n(n -1) (n - 2) •.• (n - a); of course, a ~ 1.
·205. Define To = 1,
- - + S;- + ... + S:-
Tn = S~
for n ~ 1. Compute T" and T" for n ~ 8.
-
·208. T,,+! = (") - + (")2 T,,_2
1 T"_l - + ... + (,,_" 1) Tl- + (") -
" To·
- T1z T2"z T"z"
·207. To + 1T" + 21 + ... + --;-r- + ... = .06
e- -1-••

·208. j"=(~)T,, -(;)T"_l +(;) T"_2 _ ... + (-l)"(:)To'


(T" = L1"To, if we use the notation of the calculus of finite differences,
see introduction to III 22O.)
·209. If the function F(e) has an n-th derivative
(!)" F(e*) = S~F'(e') e' + S;F"(e') eZz + ... + S:F(")(e S) e"s.
·210. Derive the identities in the variables z and 111:

(1)

DO ~"
~ ~ C'IJ ~ _ ("-I).
~ ~ J~ "I - e , (2)
"=o~:;;;,,

DO 11-'1
~
~ ~
~ "w z
s~--=
(1 -z)-. (3)
"=o~:;;;,, "I
either on the basis of the foregoing [10,190,200] or independently of the
foregoing.

yuana97@uw.edu
Pt. I, Chap. 4, No. 201-210 45

(On the other hand, we could regard (1), (2) and (3) as defining (:),
S; and sk' respectively, and then take them as basis and starting point
for establishing the propositions, especially the combinatorial proposi-
tions, discussed in the section herewith concluded.)

yuana97@uw.edu
Part Two

Integration

Chapter 1

The Integral as the Limit of a Sum of Rectangles

§ 1. The Lower and the Upper Sum


Let f(x) be a bounded function on the finite interval [a, b]. The
points with abscissae xO' Xl' X 2 , ... , X n _ 1 ' xn ' where

a = Xo < Xl < x2 < ... < xn - 1 < Xn = b,


constitute a subdivision of this interval. Denote by m. and M. the greatest
lower and the least upper bound, respectively, ofj(x) on the v-th subinterval
[.XV-I> Xv], v = 1, 2, ... , n. We call
n
L = 2' m.(x. - x._ 1 ) the lower sum,
.=1
n
u = 2: M.(x. - x._ 1 ) the upper sum
.=1

belonging to the subdivision %0' Xl' X 2 ' ... , Xn_l' xn . Any upper'sum is
always larger (not smaller) than any lower sum, regardless of the sub-
division considered. If there exists only one number which is neither
smaller thar.. any lowel sum nor larger than any upper sum, then this
number, denoted by the symbol
b
J f(x) dx,
a

is called the definite integral of f(x) over the interval [a, b] and f(x) is
called (properly) integrable over [a, b] in the sense of Riemann.

yuana97@uw.edu
Pt. II, Chap. 1, No. 1-2 47

Example:
1
I(x} = xl' a>O.
1 1 1
2"<;--X<-2-'
:r. .-1 • :r._1

consequently

We have
.. :r. - :r._1
"" = ~:r
.. (1
"" - - - = a- - -b.
1) 1 1
~:r:r
.=1 .-1. .=1 .-1 :r•

The number -!- - ! is therefore larger than any lower sum and smaller
than any upper sum. If we can prove that no other number with this
property exists we can conclude that
b

~-~=J~dx.
a b :r2
.
Since ~ is monotone the desired proof is. easy.
:r
Cf. e.g. G. B. Thomas: Calculus and Analytic Geometry, 2nd Ed.
Reading/Mass.: Addison-Wesley Publishing 1958, pp. 140-141.
1. Suppose that a > 0, r integer, r > 2. Show in a similar way as in
the previous example that

2. Assume that a> 0 and that r is a positive integer. Show that


b,+1 _ a,+1 . b,+1 _ a,+1
f
b ,
x dx = l' 1.e. the number 1 is larger than aU
.. r+ r+
lower sums and smaller than all upper sums.

The points of division xo' xl' x2 ' ••• , x"_l' x,. form an arithmetic
progression if

yuana97@uw.edu
48 The Integral as the Limit of a Sum of Rectangles

for 'P = 1, 2, ... , n - 1. They form a geometric progression if

for 'P = 1, 2, ... , n - 1. In the second case we assume a > O.


3. Work out the lower and the upper sum for the function C on the
interval [a, b] wlth the points of division in arithmetic progression. Find
the limit as n becomes infinite.
4. Construct the lower and the upper sum for the function ~ on the
~

interval [a, b] with the points of division in geometric progression, a > O.


Find the limit as n becomes infinite.
*5. We divide the interval [1, 2] into n subintervals by the n + 1
points
n n+l n+2 n+n
-;;-, -n-' n '···'-n-'

and consider the lower sum L,. and the upper sum U,. for the function
f(x) = ~ that belong to this subdivision. Show that
~

1 1 1
U1 = 1, Ll = 1 -"2' U2 = 1 - 2" +"3 '
and generally that the sequence Ul , L l , U2 , L 2 • " •• Un. Ln. ". is identical
with the sequence of the paltial sums of the series
1 1 1 1 1
1-"2+"3-"4+"·+2n-.-1-2n + ...
6. Consider the infinite sequence whose n-th term is the n-th partial
sum of the series
sin~ +sin2~ + ... +sinn~ + ...
1 2 n

at the point x = ~1' The sequence converges to a limit different from


n+
zero. (This fact shows that the series in question cannot converge uni-
formly in the neighbourhood of x = 0.)
7. Assume that the function f(x) mentioned at the beginning of this
chapter is the derivative of the function F(x). Denote any lower sum of
f(x) by L and any upper sum by U. Then we have
L < F(b) - F(a) < U.
(But F(b) - F(a) is not necessarily the only number satisfying this
double inequality for all Land U.)

yuana97@uw.edu
Pt. II, Chap. 1, No. 3-12 49

§ 2. The Degree of Approximation


8. Assume that 0 < ; < 1, that the function I(x) is monotone in-
creasing on the interval [0,;] and monotone decreasing on the interval
[~, 1] and that 1(;) = M. Then the difference

Lt" = ! I(x) dx - ~ [I ( ~ ) + 1 ( !)+ ... + 1 (: ) ]


tends to zero like ~ as n -+ 00. We find
n
_ M - 1(0) < Lt < M - f(l)
n = ,,= n .

9. Suppose that the function I(x) is of bounded variation on the


interval [0, 1]. The difference

,1" =! I(x) dx - ~ [I ( ~) + 1( !) + ... + 1(: )]


tends to zero like ~ as n -+ 00. In fact, calling the total variation V we
n
have

1,1,,1 <:.
10. Suppose that the function I(x) has a bounded and integrable
derivative in the interval [a, b]. We write
b-a" (
,1" = f bI(x) dx - -n- ~ 1 a + v -b-a)
n- .
a .=1
Find lim n Lt".
"-+00
11. Assume that t{x) is twice differentiable and that /,,(x) is properly
integrable over [a, b]. Then the difference

Lt"
,
= fbt(x) dx - -n-
b-a" (
~t a + (2v - b-a)
1) ~
a 0=1

tends to zero like -; as n -+ 00. More precisely, lim n 2 Lt: exists; deter-
n n~~
mine its value.
12 (continued). The difference

Lt~ = ! I(x) dx - :n ~a1 [/(a) + 2 .~~ 1 (a + 2v :n~ a1) ]


converges to zero like -; as n -+ 00. More precisely, lim n 2 Lt~ exists;
n n~~

find its value. Show, in addition, that Lt~ > 0 if f'(a) > 0 and /,,(x) > 0,
a <x ~ b.

yuana97@uw.edu
50 The Integral as the Limit of a Sum of Rectangles

13. We write
1 1 1 2 2 2
U.. = n + 1+ n + 2+ ... + 2n' V" = 2n + 1+ 2n +3 + ... + 4#- 1.
Prove that
lim U" = log 2, lim V" = log 2,
,,-+00 "-+00

lim n (log 2 - U,,) = ~ ,


,,~oo ":t:
lim n 2 (log 2 - V,,)
"-+00
=.!...
32
14. The expression
_1_
. n
+_1_
. 2n
+ ... + .
1
(n - 1) n
_ 2nn (log 2n +C -log3r)
Sln - Sln - sm -'---~
n n n
is bounded for increasing n; C is Euler's constant [solution 18].
15.
.. ,,+1 1
lim e4 n --2- W22 33 ... n")" = 1.

16. Suppose that {)(. is positive and that x" is the (only) root of the
equation
111 1
2x + x- 1+ x- 2+ ... + x - n = {)(.
in the interval (n, (0). Prove that

lim (x" - n+ ~'") = 1- e


O. [12.]

x: is the (only) root in the


" .... 00

17. Assume that {)(. is positive and that


interval (n, (0) of the equation
1 2x 2x 2x
-; + Xl - 11 + Xl - 21 + ... + Xl _ nt = {)(..
Verify that
lim
"-+00
(x: - (n-t'!") 1 + e=,") =
2 1 - e '"
O. [12.]

18. Suppose that I(x) is differentiable and that /'(x) is monotone


increasing or decreasing to zero as x ~ 00. Then the following limit
exists:
lim
_00
(It{!) + 1(2) + 1(3) + ... + I{n - 1) + l/{n) - j I{x) dX) = s.
1

Assume that /' (x) increases. Then the two inequalities

l/,(n) < i/(1) + 1(2) + 1(3) + ... + I(n -1) + l/(n) - J" I(x) dy - s < 0
1

yuana97@uw.edu
Pt. II, Chap. 1, No. 13-19.2

can be established. Note the particular cases I(x) = ~,


x
I(x) = -log x.
19. Assume that I(x) is differentiable for x ~ 1 and that I'(x) is
monotone increasing to 00 as x ~ 00. Then
.
i/(l) + 1(2) + 1(3) + ... + I(n - 1) + i/(n) = f I(x) dx + O[f'(n)].
1
More precisely,
.
0< i/(l) + 1(2) +/(3) + ... +/(n-1) +i/(n) - f I(x) dx<il'(n) -il'(l). )

19.1. We may regard the relabon [18]


lim (1 + ~ + ~ + ... + ~ - log
"-+00 2 3 n
n) = C
as the definition of Euler's constant C. Derive hence that

1 - 2'
1
+ 3"1 - ... + 2n 1_ 1 -
1
2n + ... = log 2.

19.2. The definition of C given in 19.1 is convenient. Yet it would


be desirable to approximate C by rational numbers, to represent C as
the sum ·of a series whose terms are rational. Prove that the following
series fulfills this desideratum:
1
1

-~-~+
2 3
+~+~+~-
444
1 1 1 1
-5-"6-"7-"8+

+~+~+~+~+~-
99999
1 1 1 1 1 1
- 10 - TI - 12 - 13 - 14 - 15 +
+ ....
Take the terms in the order as you read a book: from left to right and
from top to bottom. This is essential since the series is not absolutely
convergent. Its terms are non increasing in absolute value.

§ 3. Improper Integrals Between Finite Limits


Let I(x) be defined on the finite interval [a, b] except at the point
x = c, a :::; C ::;; b, in the -neighbourhood of which I(x) assumes arbitrarily

yuana97@uw.edu
52 The Integral as the Limit of a Sum of Rectangles

large values. Furthermore let I(x) be properly integrable over each closed
b
subinterv!ll of [a, b] that does not contain c. Then the integral I(x) dx J
is defined as the limit II

jl(x) dx =
II
lim (7·/(x) dx
.... -++0 II
+e+.·
//(x) dX).

(If c coincides with a or b there is only one integral to consider.) If /(x)


becomes infinite at several (finitely many) points of the interval [a, b]
the integral is defined in a similar way.
Assume that I(x) is defined for x ~ a, furthermore that it is properly
integrable over any finite interval [a, co]. Then we define
00

J /(x) dx = J/(x) dx.


til

lim
o tII-+DO II

One type of improper integral may be transformed into the other type
of improper integral by an appropriate substitution.
20. Assume that the function /(x) is monotone on the interval (0, 1).
It need not be bounded at the points x = 0, x = 1, we assume however
1
that the improper integral J I(x) dx exists. Under these conditions
o

21 (continued). If 9'(x) is properly integrable over [0, 1] we have

tp (~) f(~) + tp (.!) f(.!) + ... + tp (!:.=..!) f(!:.=..!)


J9'(x) I(x) dx.
1
lim n n n n n n =
"-+DO n 0

22. Prove in a way different from 171 that for IX> 0


lim 1",-1 + 2~-1 + ... + n",-1 =..!...
"-+DO n'" ~
23. We put
00 DO DO
I k",-1:!, I 11l-1:1 = I a..z". IX, {J > O.
i-1 1=1 .. -1
Then

exists and is non-zero. (If 0 < IX < I, 0 < {J < 1. IX + {J ~ I, Z = -1,


the product series is divergent although the two factors are convergent.)

yuana97@uw.edu
Pt. II, Chap. 1, No. 20-30 53

24. The converse of the statement of problem 20 is not valid: There


"-I
are functions, monotone in the interval (0, 1), for which lim 2: ..!.-I (!!..)
"~oo ,\=1 n n
exists but not the integral.
1
25. The integral J I(x) dx exists if I(x) is monotone in the interval
o
(0, 1), finite at x = 0 or x = 1 and if the following limit is finite

lim f(~) + f(~) + ... +f(n ~ 1).


11-+-00 n

26. Let the monotone function I(x) be defined on (0,1). Then the
equation

lim - 1"(21)
2: 1 "2- = JI(x) dx 1
,,~oo n .=1 n 0

holds under the condition that the integral at right exists.


27. If IX> 0
. 1",-1 _ 2",-1 + 3"'-1 _ .•• + (_1),,-1 n",-1
lim = O.
"-+-00 n(}l,
28. If I(x) is properly integrable over [0, 1] we have obviously

lim f(~) - f(~) + f(~) - ... + (-1)" f(~) = O.


n

Show that this is true also if I(x) is improperly integrable but monotone.
29. If I(x) is monotone for x > 0, lim B" = 0, c > 0, B" > ~ we find
"-00 n

f(e,,) + f( eft +..!.-) + f( e" + ~) + ... + f( e.. + n -1)


n n n
1
lim
"~oo n
=
0
Jf(x) dx,
provided that the integral at right exists and I(x) is finite at x = 1.

§ 4. Improper Integrals Between Infinite Limits


30. Assume that the monotone function I(x) is defined for x ~ 0

J I(x) dx exists. Then we have


00

and that the improper integral


o

JI(x) dx.
00 00

lim h(t(h) + 1(2h) + 1(3h) + ...) = lim h }; I(nh) =


h-++O h~+O ,.=] 0

yuana97@uw.edu
54 The Integral as the Limit of a Sum of Rectangles

31. The F-function is defined for IX > 0 (or mIX > 0) by the integral

F(IX) = J""e- z ,%"-1 dx.


o
Using I 89 prove that
. n,,-1 n!
F(IX)=hm ,IX>O.
....... 00 lX(lX + 1) ... (lX + n - 1)
32. As it is well known [ef. E. T. Whittaker and G. N. Watson,
p. 246] Euler's constant C can be written as an integral

C=
o
J e- z (1
00

1 - e- z
- - 1) dx.
x
Show that

C= ,!~o r(1 - t) (1 ~ t + 1 ~ t + 1 ~ t + ... + 1 ~ t" + ... )


2 8

-log 1 ~ t]'
33.
t 2- + - t 8 + ... + -t")
lim (1 - t) ( -t-
H1-0 1 + t +-
1 + t2 1 + t8
- +... =
1 + t"
log 2.

34.
lim (1 - t)2 ( -t- + 2-t
- + 3 -t - + ... + n -t")
- + ... =
2 8 1f2
-.
H1-0 1- t 1- t 1- t 2 1 _ t" 3 6

35. We have
lim
~~
V1- t(l + t + t 4 + t9 + ... + t'" + ...) = V;2
and more generally for IX > 0
. ~r;--.
lim V 1 - t (1 + t1" +.-,2" + t3" + ... + t"" + ... ) = 1F(-1).
-
' ......1-0 lX lX

36. Compute
2t
.
,!~oo t
( 1
+ t"z-+iii + t +2t 22 + ... + t 2
2
t)
+ n + ... . 2 2

37. Let IX > n (1 + n"t ) . Show that


1and put g(t) = ,,=1 00

lim log g(t) _ _ 1f_


1 - .
' ..... +00 sin~
t" lX

38. Establish the equation

J log (1 -
00

2,r2 cos 2qJ + ,r4) dx = 2n sin!p,


o

yuana97@uw.edu
Pt. II, Chap. 1, No. 31-.41 55

with help of the following identity, valid for all complex t:

sin t = eil - e- il = 11 (1
00
_~) .
t t n1rel
2 i.. =1

39. Compute the integral J" log x dx as the limit of an infinite sum
o
of rectangles corresponding to the points of division

a, aq, aq2, atf, ... , O<q<1.


40. Let k be a fixed positive number and n an integer increasing to
infinity. Then

i (n)k NVk~..:. (~)-2


k-1

,=0 1/ nn
[58]. Observe the particular cases k = 1, k = 2.

§ 5. Applications to Number Theory


41. Divide the integer n by'll, 'II = 1, 2, 3, ... , n, and call the resulting
remainder n.. E.g. 173 = 2, 1020 = 10; obviously n = n, (mod'll),
o : ;:; n. < 'II. Find the probability that n, > ; .
Solution:

consequently

In the favorable case we have

- ~,
n, > 2 thus [2nJ - 2 [~] =
II 1/
l''

in the adverse case


n, < ; , thus [2; ] - 2 [: ] = 0 [VIII 3].

The probability in question is therefore

yuana97@uw.edu
56 The Integral as the Limit of a Sum of Rectangles

As n -+ 00 this sum approaches the proper integral


1

! ([!] - 2[!])dx= ft~~K ! ([!] - 2[!J)dX


1 ~

.+1

=!i!'!~K(.:: -'~1)
1 1
= 2 ( "3 - 4" + "51 - "61 + ...) = 2 log 2 - 1 = 0,38629 ...
42 (continued). Compute
lim
ft-+co
~ (nl
n 1
+ ng2 + na3 + .. , + nft).
n
43 (continued). Compute
. n1 + n2 + ns + ... + nft
lim
8-+00 n
2 •

44 (continued). Let A .. be the number of those fractions of the form


nk •
h' k = 1, 2, ... , n, that are smaller than a gIven number tX, 0 ~ tX < 1.
A
The quotient ~ tends with increasing n towards the limit
n

f
1
l-X"
1 _ x dx. [VIII 4.]
o
45. Let O'.. (n) denote the sum of the tX-th powers of all the divisors
of n (VIII, Chap. 1, § 5) and

E .. (n) = 0'.. (1) + 0'",(2) + 0',,(3) + ... + O'.. (n) = •i=1 [: J". [VIII 81] .
Then we have for tX > 0
lim .E"i~~ = C(~ + 1) ,
ft ..... co n"+l ~ +1
where C(s) denotes Riemann's C-function (C(z) = ~ n-·, cf. VIII,
ft=l
Chap. 1, § 5). For tX > 1 we even have the inequality
1.E,,(n) _ C(Ot + 1)1 < 2C(~) - 1
n = 1, 2, 3, ...
InIX+! lX + 1 i= n '

[(! - [! J) is properly integrable over the interval [0, 1] if


XIX tX > 0,
[107J; [! ] x'" is of bounded variation if > 1.J (X

yuana97@uw.edu
Pt. II, Chap. 1, No. 42-48 57
46. Let 1'(n) = <To(n) denote the number of divisors of n. Then

1'(1) + 1'(2) + 1'(3) + ... + l'(n) = :E" [: ]


~=l

= n (logn + 2C -1) + O(V;} [VIII 79],

C is Euler's constant. [Apply the idea of 9 to ! - [! ] in the interval


(:,1), m = [V;] + 1; solution 18.]
47. Denote by 0" the number of odd divisors and by En the number
of even divisors of the integer n. E.g. 020 = 2, E20 = 4. Prove that
. 01 - El + 02 - E2 + ... + On - E"
lim = log 2.
n-+oc n

§ 6. Mean Values and Limits of Products


The arithmetic, geometric and harmonic means of the numbers
aI' a2, ... , an are
al + a2 + as + ... + an ",----- n
n
Val a2 as ... an' 1111'
-+-+-+ ... +-
a1 az as an
respectively. For the last two expressions all the a:s are supposed to be
positive. (More details in Chap. 2.)
48. Suppose that the function I(x) is defined on [a, b] and properly
integrable over this interval. Define
b-a
I5n = -n- ·
Then
lim lIn +12" + fsn + ... + In" =-= _1_ jl(x) dx,
n~oo n b - aa
1 b
.. iJ=--; Jlogf( x)4.~
lim VAn/2nlsn ... In .. = e a ,
n~oo

lim n
1 11 1
"-+00
-+-+-+ ... +-
lIn 12n Isn I....
a

These three limits are called the arithmetic, geometric and harmonic mean
of thefunctionf(x). In the last two relations the greatest lower bound off(x)
should be positive. (To improper integrals apply with caution!)

yuana97@uw.edu
58 The Integral as the Limit of a Su m of Rectangles

49. Prove the existence of


v;!
.
lim -
"-+00 "

in a way different from I 89 and that the limit is equal to the geometric
mean of I(x) = x on the interval [0. 1]. i.e. = -.!...
e
50. Let a and d be positive numbers and cail A.. the arithmetic. and
+ d. a + 2d• ...• a + (n - l)d.
G.. the geometric. mean of the numbers a. a
Then we obtain
G.. 2
lim -=- .
....... 00 A.. e

51. Let A .. denote the arithmetic. and G.. the geometric. mean of
the binomial coefficients

(~). C)· (;). .... (:).


Show that
"- "-
lim
....... 00
VA .. = 2. limVG..
.. ..... 00
-
=Ve.
52. Prove
{210g 1" for 1" ~ 1.
1
l' 0
2"
2 flog (1 - 21' cos X + 1'2) dx = 0
-
for 0 < 1" ~ 1.

53. Let l' be positive and smaller than 1; let x assume any value in
the interval [0.2n] and E denote the number closest to x for which

sin (x - E) = l' sin x.


Then we find
1 2"
-2 flog (1 -
no
21' cos E + ,s) dx = log (1 - ,.2) •

[Interpret eO". eie • l' in the complex plane.]


54. Assume that I(x) is properly integrable over [a. b]. Using the
same notation as in 48 establish
b
Ji<")rl,,
lim (1
....... 00
+ 11..fJ..) (1 + 12..fJ..) ••• (1 + 1....fJ.. ) = ~

55. Compute
lim (nZ + 1) (nZ + 2) ." (nZ n) +
....... 00 (nZ - 1) (nZ - 2) .. , (nZ - n) ,

yuana97@uw.edu
Pt. II, Chap. 1, No. 49-60 59

58. Prove the identity

(1+ _ 1)(1 __
1 )(1 +
IX-I 2tx-1
_1
3tx-1
)(1 __
1 ) ...
41X-1

" .(1 + (2n - :) IX _ 1) (1 - 1


-=-2n-IX-_---:"1)

_ + 1) IX
(n • + 2) IX
(n (n + n) IX
(n + 1) IX - 1 (n + 2) IX - 1 (n + n) IX - l'

Whence there follows that the product on the left hand side extended
1 1 1 1
to infinity has the limit 2-;>, provided that (X =F 0, 1, "2' "3' "4' ...
[(2n)! = 2"n! 1· 3·5 .. · (2n -1).]
57. Let (x, {J, tJ be fixed, tJ > 0 and

b=I+ In
d
a=I+~,
n
, d= n-.
Show that
a<-fJ
lim ~ . a + d • a + 2d ••• a + (n - 1) d = (1 + tJ)-,,- .
"-+00 b b + d b + 2d b + (n - 1) d

58. Let n and v be integers, 0 < v < n. If n and v increase to infinity


in such a way that
n
v --
lim .c2 =A,
"-+00 y-n
then
lim V; (n) = 1/ 2 e-2A·.
r
"-+00 2" v :II'.
.,
59. Let t be a fixed real number. Define z = 2neV;; • Then we have

lim j2n - 1. 2n - 2. 2n - 3... 2n - nj = (~)j' .


"-+00 Z - 1 z - 2 z - 3 z - n e

§ 7. Multiple Integrals
60. Suppose that the function f(x, y) is the second mixed derivative
of a function F(x, y),
o2F(x. y) = f(x )
oxoy ,y,
in the rectangle R
c ~ y ~ d.

yuana97@uw.edu
60 The Integral as the Limit of a SUIJl of Rectangles

The points xI" Y., P = 0, 1, 2, ... , m; 11 = 0, 1, 2, ... , n;


a =xo < Xl < x2 < ... < x"'_ l < X... = b,

c = Yo < Yl < Y2 < ... < Y.. -l < Y.. = a


determine a subdivision of the given rectangle R into "subrectangles"
RI": XI'_1 ~ x <xI" Y,-1 ~ Y ~ Y" P = 1,2, ... , m; 11 = 1, 2, ... , n.
We denote by MI'" ml" the least upper and the greatest lower bound
resp. of I(x,y) in the rectangle RI'" Defining the

upper sum U =
'" ..
1: 1: MI"(xl' - XI'_1) (y, - Y,-I)'
1'=1.=1

and the lower sum L =


'" ..
1: 1: ml"(xl' - xl'_1) (y, - Y,-l)
1'=1.=1
we find that
L ~ F(b, a) - F(b, c) - F(a, a) + F(a, c) ~ U.
81.

JJ
O~s~y~,.
log Isin (x - y) I ax ay = - ~ log 2.

[ Compute the square of the absolute value of the following determinant


in two ways
1 1 .. ·1
1 ... 6"-1
2m ]
1 ... 6 2("-1)
6 - ".
-e

82. Let I(x, y) be properly integrable over the square 0 ~ x < 1,


o~ Y < 1. Show that
1 1
.. " J JI(s,y)dsdy
,,~!2!!J1 + :2/(:, :)] = eO 0

83. Let I(x, y) be properly integrable over the square 0 < x ~ 1,


o < Y ~ 1. Compute

yuana97@uw.edu
Pt. II, Chap. 1, No. 61-68 61
84. The three-dimensional domain Il) is defined by the inequalities
-1:S;; x, y, z:s;; 1, -G:S;; X + y + z:S;; G.
Using 130 show that the volume of Il) is

66. Let 1%1' 1%2' ••• , I%/> be arbitrary positive numbers, define

I.(z) = l«,,-lZ + 2".-1,:2 + ... + na. -1z" + ... , " = 1,2, ... , P
and
11 (z) Mz) ••• I/>(z) = 1: a..z" .
.. -1
Show that

= II'" I~1-1~-1
1 2
••• ;r"'/>-1-1
/>-'1
(1- x1 - x 2 - ' " - X1>-1 )",/>-1

X dX1 dX2 ... dX/>_l'

where the integral is taken over the domain described by the p inequa-
lities xJ ~ 0, x2 ~ 0, ... , x/>_l ~ 0, xJ + x2 + ... + X/>_l :s;; 1 (p - 1
dimensional simplex).
88 (continued).
II ••• I ~.-lX;.-J ... X;~11-J (1 - Xl - x2 - ••• - x/>_1)"'r 1
r(IX 1) r(IX2 ) '" r(IXp )
Xdx1 dx2 .. ·dx"_1=r(
r IX1
+ <X2 + ... + <X/>).

87. Work out the ploduct n" (1 + I"" 15,,) (same notation as in 48,
"=1
&4) as a polynomial of degree n in 15". Prove that the term containing t5!
converges to the limit

I I •.• I/(x l(x2 ) J)


":;Os.:;Os.:;O... :;os/>:;06
••• I(x/» dX1 dX2 ••• dx/> = ~
p. "
(j I(x) dX)/> ,

when p is fixed and n increases to infinity.


88. Suppose that the 2m functions
l(x 1 ), l(x 2 ), .... I",(x),
9"(x1). 9"(x2), ••• , 9"",(x)

yuana97@uw.edu
62 Inequalities

are properly integrable over the interval [a, b). Then we have

I j 11 (x)
b b

a
qJl (x) dx
a
f 11 (x) qJ2(X) dx f 11 (x) 'P,... (X) dx
"

" " "


b

a
f I... (X) qJ2(X) dx
a a

1 b ~ b IIJ (Xl) 11 (X2) •• , 11 (X... ) I qJI (Xl) qJI (X2) •• , qJl (X... )

= m! J j ... J Mx
a a a
j ) 12(x2.) .. , Mx... ) • qJ2(X1 ) qJ2(X2) '" qJ2(X... )

~~ i~'(;:)'i~(~~) ..... :i~'(~~) I~~(~~i ~~.(;:).::: ~~(~~)

[Compute in two different ways the product of the two matrices

II/!~ 11).=1.2........ • "qJ~ 111'=1.2 ........ ;


,,=1,2, ... ,'" ,,=1.2"0"'"

Chapter 2

Inequalities

§ 1. Inequalities
Let aI' a2 , " ' , an be arbitrary real numbers. Their arithmetic mean
~l(a) is defined as the expression
al + a2 + .. , + a,.
m:(a) = n
.

If all the numbers aI' a2 , .•• , an are positive we define their geometric
and harmonic means as

~(a) = lIn 1 '


-+-+
a1 a2
... +-
an

yuana97@uw.edu
Pt. II. Chap. 1 . Chap. 2 63
respectively. If m denotes the smallest and M the largest of the numbers
Q i then

m< ~(a) <M. m< &(a) <M, m< ~(a) <M.

For &(a) and ~(a) we assume m> O. The three numbers represent
mean values of all a2 , ••• , a... The inequalities become equalities only if
all the a/s are equal. The mean values have the following properties:

_1 _ &(~) _1 _ ~(~)
&(a) - aI ' ~(a) - a'

~(a + b) = ~(a) + ~(b), &(ab) = &(a) &(b), log &(a) = ~ (log a).
Let the function I(x) be defined and properly integrable on the
interval [xl> X2J. We define the arithmetic mean ~(I) of I(x) as
1
f f(x) dx.
x,
~(/) = - -
X -X 2 I
x,

If I(x) is strictly positive, i.e. if there exists a positive constant k such


that I(x) > k for x in [Xl' x2J, the geometric and harmonic means are
defined as
1 x,
- - J logf(x)dx
&(1) = /,-x, x, , ~(I) = x 2 - Xl •

I X'dx
f(x)
x,

[48.J If m denotes the greatest lower and M the least upper bound of
I(x) on [Xl' x2J then
m< ~(I) < M, m < &(1) < M, m< ~(/) < M.
It is understood that m > 0 for &(1) and ~(I).
The following Ielations are obvious:
~(1 + g) = ~(/) + 2l(g) , &(Ig) = &(1) &(g); log &(1) = 2l(log I) .

Let ai' a2 , ... , a.. be albitrary positive numbers which are not all
equal. Then

i.e.
~(a) < &(a) < ~(a).

yuana97@uw.edu
64 Inequalities

(Theorem of the means, arithmetic, geometric and harmonic.) A very


beautiful proof! was given by Cauchy in his Analyse algebrique (Note 2;
Oeuvres Completes, SeT. 2, Vol. 3; Paris: Gauthier-Villars 1897, pp. 375-
377).
1 It is obviously sufficient to prove ~(a) > QI(a). Here is the passage referred to:
"La moyenne g~omftf'ique entf'e plusieurs nombrl's A. B. C. D • ... est toujours
injmeuf'e It leUf' moyenne al·ithmetique.
Demonstration. - Soit IJ Ie nombre des lettres A. B. C. D, ... II suffira
de prouver. qu'on a generalement

"VA BCD ... < A + B ~_Cn + D ~ (1)


ou. ce qui revient au meme.
A+B+C'+D+ ••• )"
ABCD •.• < ( . (2)
n

rc
Or. en premier lieu. on aura evidemment. pour n = 2.

A B = (A ~ By _ (A ; B < ~ By •
et l'on en conclura. en prenant successivement IJ = 4. /I = 8 •...• en fin /1 = 2m
A +B)I(C+D)I (A +B+C+D)'
ABCD< ( - 2 - -2- < 4 •

ABCDEFGH
(A+ B+ C + D)' (E + F+4 G + H)'
< -.----.--
< (A + B+ C + D; E+ F+ G + H)'

ABCD···< (
A+B+C+D+ .•• m
.
)2 (3)
2'"
En second lieu. si l' n'est pas un terme de la progression geometriquc
2.4.8,16 .....
on designera par 2'" un terme de cette progression superieur a 71, et l'on f('ra
A+B+C+D+ •..
K=;
n
puis. en revenant a la formule (3). et supposant dans Ie premier membre de cettc
formule les 2'" - n derniers facteurs egaux a K. on trouvera

ABCD .•• K2"'-" <


[A + B + C +. D ;,. ••• + (2'" - 1/) I(
]"'"
~

ou. en d'autres termes.


ABCD ... Ki" .... < K ....
On aura done }lar suite

A BCD •.• < 1(" = ( A + B + C + D + "')" •


n
ce qu'il fallait demontrer."

yuana97@uw.edu
Pt. II, Chap. 2, No. 69-71

69. Let the function I(x) be defined and properly integrable on the
interval [xJ' x2] and let I(x) have a positive lower bound. Then

or, with the notation just defined,


~(f) ::::;; &(f) ::::;; ~(f).

70. Suppose that the (not necessarily differentiable) function ljJ(t)


satisfies for arbitrary values tl , t2, tl =1= t2 the inequality
IjJ ('I ~!!) < '1'(/ 1) ~ cp(t.) •

Then the more general inequality

(
'1 + '2 + ... + ,,,) < '1'('1 ) + '1'('2) + ... + '1'(',,)
IjJ n n
holds, where the to's are arbitrary but t, =1= t; for at least one pair i, j.

A function ljJ(t) defined on the interval m ::::;; t ::;;; M is called convex


if for each pair tv t2 on [m, M], tl =1= t2 the inequality

IjJ
('1 +2 II) :- ; ; '1'(1 1) + '1'('.)
2

is satisfied. (By the solution of 70 we have then generally


(
'1 + '2 + ... + ,,,) ::::;; '1'('1) + '1'(12) + ... + '1'(/,,>
IjJ n - n
for arbitrary points tl , t2 , ••• , tIl of the interval.) If instead of the inequality
with ::::;; strict inequality is supposed «) then ljJ(t) is termed strictly
convex. If -1jJ(t) is convex, ljJ(t) is termed concave. (A more intuitive but
somewhat clumsier terminology would be "convex from below" for
convex and "convex from above" for concave.) In the sequel we will
consider bounded convex functions only; these are continuous [cf.124;
110 often useful].
71. Suppose that the function I(x) is properly integrable on the inter-
val [Xl' x2 ] and m ::::;; I(x) ::::;; M and, furthermore, that ljJ(t) is defined and
convex on the interval [m, M]. Then we have the inequality

IjJ ( Xz -
1
Xl
x.
Jf(x) dx
) ::;;; 1
X z _ Xl
x.
J1jJ[j(x)J dx •
.%, .%1

yuana97@uw.edu
66 Inequalities

72. Suppose that the function qJ(t) is defined on the interval Em. M]
and that qJ"(t) exists and qJ"(t) > 0 on Em. M]. Inthis case qJ(t) is strictly
convex. If we have qJ"(t) ~ 0 only, then qJ(t) is convex. (A function can
be convex in an interval where its second derivative does not exist at
all points.)
73. The functions
r (0 < k < 1) and log t
are concave on any positive interval;
r (k < 0 or k> 1) and tlog t
are convex on any positive interval;
log (1 + e') and Ve s + t2 (e> 0)
are everywhere convex.
74. Assume that qJ(t) is a convex function defined on Em. MJ. that
Pl' PS' ...• P.. are arbitrary positive numbers and that tl , ts' ..• , t.. are
arbitrary points of the interval Em, M]. Then we have the inequality
Pl tl + Psts + ... + P..t,,) ~ P1rp(tl ) + P2rp(tS ) + ... + P..rp(t.. ) •
(
qJ PI + P2 + ... + P.. - PI + P2 + ... + P..

75. Assume that I(x} and P(x} are two functions which are properly
f P(x) dx>O.
%.

integrable over [X1,X2 ] and that m ~l(x} ~ M, P(x) > 0 and


%,

Let qJ(t) denote a convex function defined on the interval m <t< M.


Then we have
j"P(X) f(x) dX) j"P(x) rp[f(x)] dx
qJ (
%,
%.
<
=
"-'-%,-----
%.

f P(x) dx f P(x) dx
%1 %1

76. Suppose that on the interval [m, M] the first and the second
derivative of qJ(t) exist and that qJ" (t) > O. Then we find for positive
Pl' P2' ... , P..
Pl tl + P2t2 + ... + P..t..) ~ Plrp(t1) + P2rp(tS) + ... + P..rp (I.. ) •
(
qJ PI + Ps + ... + P.. - PI + P2 + ... + p..
There is equality if and only if tl = t2 = ... = tro'
77. The functions I(x) and P(x) are assumed to be continuous on the
interval [xl> x2 ], P(x) IS sttictly positive and m < I(x) < M; the function
qJ(t) is defined on the interval em,
M], qJ(t) can be differentiated twice

yuana97@uw.edu
Pt. II, Chap. 2, No. 72-79 67
and VJ"(t) > O. Then

j"p(~) f(~) d~) j"p(.'r) q>[f(~)] d~


m ( '*' ~'*...:'~----
T '*1 - '*1
I p(~) d~ I p(~) d~
-*1 ~1

There is equality if and only if I(x) is a constant.


78. Prove the following generalization of the proposition on the
arithmetic, geometric and harmonic means: Let PI> P2' ... , P.. , at, "2, ... , a..
denote arbitrary positive nuinbers, a. {: aj for at least one pair i {: j, .
i, j = 1, 2, ... , n; then the inequalities
1>,log", +l>i10g"I+'"+1> ..log....
PI + 1'2 + ... + 1' .. < e I>,+I>I+"·+P.. < PIal + p2a2 + ... + p..a..
1'1 + p. + ... + 1'.. PI + 1'2 + ... + 1'..
~ a. a..
are satisfied. Furthermore
p, 1>1 p..
-log.., + -log... '1-'·'+ -log....
41 ". lin
p, p. 1>..
-+~+ ... +- PI +1'2 + ... + 1'..
e al 4. A,.
<. I' ,
1'1 +1'8 + ... +2
a1 as a..
1>,..,log", +1>1...10g... +'" +1>......log....
PIal + 1'2"2 + ... + p ..a.. < e 1>,..,+1>.... + ... +1>.... ..
PI +1'2 + ... + 1'..
79. Let lex) and P(x) be continuous and positive on the interval
[Xl' XII]; further, I(x) is not a constant. Then we have the inequalities
'*.
11>(,*)101/('*)"
'*-,---- '*.
I p(~) f(~) dx
<'*...:'----
'*1
I p(~) d~
%,

moreover
s.
I P(%)/(,*)log/(,*)4%
'*1 -
%,
----
f Plex)(%) 4% I'*.p(~) f(~) d~
%1

I p(~) dx %1
11>(%)/(,*)4%
e %, <...:%'--- '*...:'---- <.6 %1
'*. '*.
J f(~)
p(~) d~ I p(~) dx
%,

yuana97@uw.edu
68 Inequalities

80. Let aI' a2 , ••• , a.. , bl , b2.' ••• , b.. be arbitrary real numbers. Show
that they satisfy the ine-quality

The case of equality arises if and only if the numbers a, and b. are pro-
portionalto each other, i.e. Aa, + p.b, = 0 for v = 1, 2, ... , n, ,1.2 + p.2 > O.
(Cauchy's inequality.)
81. Suppose that I(x) and g(x) denote two functions that are properly
integrable over the interval [Xl' x 2 ]. Then

cr I(x) g(x) dX)


2
~l [I(X)]2 dx l [g(X)]2 dx.

(Schwarz's inequality.)
81.1. The numbers aI' ~, ••• , a.. , bl , b2 , ••• , b.. , lX and fJ are positive,
lX + fJ = 1. Then

a~14. + a~14 + ... + a:~ ~ (al + ~ + ... + a..)'" (bl + b2 + ... + b.. )P.
(Holder's inequality; 80 follows from the particular case where lX = fJ.)
81.2. The functions I(x) and g(x~ and the numbers lX and fJ are posi-
tive,
lX+fJ=l,

r r·
and the functions are integrable in the interval [Xl' xs]. Then

j ' [I(x)]'" [g(x)]I'dx ~ Lt/(X) dx [j'g(X) dx

(For the particular case where lX = fJ see 81.)


81.3. The numbers
bl , b2 , ... , b.. ; ... ,
lX,fJ, ... , A
are positive,
lX+fJ+,,·+A=1.
Then
a~14. ... li + a~14 ... l~ + ... + a:~ ... l!
~ (al + ~ + ... + a..)'" (bl + b2 + ... + b..)P ... (ll + l2 + ... l.. )A.
81.4. The functions I(x), g(x), ... , hex) and the numbers lX, fJ, ... , A
are positive,
lX+fJ+"'+A=l

yuana97@uw.edu
pt. II. Chap. 2. No. 80-86 69

and the functions integrable in the interval [Xl' x.J. Then

l
S, [/(x)]'" [g{x) JII ••• [hex) Jt dx :s; [l'(x)
s,
dx
] '" [ s, ]11 [S' ]!
l g(x) dx . . . ! hex) dx .

82. Let a1'~' ... ' a" be arbitrary positive numbers, not all equal.
Then the function
1

",(t) = (" +
a1 a2 + ... + a"
n
')'
is monotone increasing with t. Find the values of

(Define ",(0) so that ",(t) is continuous at the point t = 0.)


83. Assume that the function I(x) , defined on [Xl' xlJ is properly
integrable and that it has a positive lower bound. The function
1

",(t) = (X2 ~ Xl j' [/(x)]' dX),


is non-decreasing for all t. Compute
"'( - (0), "'( -1), ",(0), ",(1), "'( + (0).
For ",(0) see 82. In computing 11'( - (0) and "'(00) assume that I(x) is
continuous.
84. Let a., b., 11 = 1,2, ... , n, be arbitrary positive numbers. Prove
the inequality

i.e.
@S(a + b) ~ @S(a) + @S(b).
The relation becomes an equality if and only if a., = A.b., 11 = 1, 2, ... , n.
II. The functions I(x) and g(x) are properly integrable over the
interval [Xl' x.J and strictly positive. Then
1%' IS, IS,
- - %1Jlog(l(s) +I(s))4s ......... J
- - log/(s)4% J
- - logg(s)4s
eZI -%1
.::::.e
.1'.-%1 S
I +e
.1"1-%. %
I

i.e.

88. The functions 11 (X), 12(X); ••• , I".(x) are defined on the interval
[Xl' X.], properly integrable and strictly positive (there exists a constant

yuana97@uw.edu
70 Inequalities

K such that 0 < K ::;; I.(x), 'JI = 1, 2, ... , m, Xl::;;:x::;; Xa)' Let
'A, Pa' ... , P". denote arbitrary positive numbers. Then
<M(P]ti + P2/2 + ... + p".!".) '2! PI &(11) + P2&(l2) + ... + P".&(/".).
87. Suppose that the functions I. (x), 'JI = 1, 2, ... , m, are of bounded
variation on the interval [xl> Xg] and that Pl' P2 , ••• , P". are arbitrary
positive numbers. Define
F(x) = Plf](x) + P2Mx) + ... + P".f".(x) •
PI +P2 + ... + P".
The lengths ofthe arcs of II(X), Mx), .... I".(x) are denoted by 11> 12 , .... I".,
the length of the arc of F(x) by L. (At a point of discontinuity the jump
must be included in the length of the arc.) Then we have

L::;; PI'l + P2' 2 + ... + Pm'". .


- PI + P2 + .. , + P".
88. Let I(x) be a positive continuous and periodic function with
period 2n and let P(x) be a non-negative and properly integrable function
on [0, 2n] with positive integral. Then
2"
JP(;) f(; + x) dE
F(x) = 0=----,2,-"--

o
JP(E) dE
is positive and continuous; furtheFmore
I 2" 1 2"
2 I logF(z)dz 2 I logJ(z)dz
e"o >e"o
i.e.
&(F) '2! &(1).
89. Assume that I(x) is a periodic function with period 2n and that
P(x) is non-negative and properly integrable over the interval [0, 2n]
and that its integral is positive. If I(x) is of bounded variation then this
is true also for
2"
JP(E) f(E + x) dE
F(x) = 0'---:2'-"---

o
JP(E) dE
If I, L denote the lengths of the arcs of I(x) and F(x) on [0, 2n] then I and
L satisfy the inequality
L ::;; I.

yuana97@uw.edu
Pt. II, Chap. 2, No. 87-93 71

90. The arbitrary numbers aI' a2 , ••• , a.. and bI , b2 , ••• , b.. are positive.
We define
I
9R,,(a) = (a~ + a~ + ... + a:)" .
Then
9R,,(a + b) S;; or > 9R,,(a) + 9R,,(b) ,
according as " > 1 or ,,< 1. Equality is attained only for a, = A.b.,
v = 1, 2, ... , n, or if " = 1. (What does the proposition mean in the
case" = 2?) (Minkowski's inequality.)
91. The function j(x) is defined on [Xl' x2J, properly integrable and
strictly positive. We introduce
I

9R,,(/) = (}' [/(x)]" dX)-;


Let g(x) be a function with the same properties as j(x). Then we have
9R,,(/ + g) S;; or > 9R,,(/) + 9R,,(g) ,
according as " > 1 or " < 1.
92. Let a, A, b, B .be positive numbers, a < A, b < B. If the n
numbers aI' a2 , " ' , a.. lie between a and A, and the n numbers bI' b2 , ••• , b"
between band B we can prove that

1<
(ai + a~ + ... + a;l (bi + b~ + ... + b;)
(a l bl + a2 bz + ... + a.. b.. )2 S;;
(V~: + ffi)2
2 •

The first inequality is identical with 80. The second inequality becomes
an equality if and only if
A B
a b
k=TB n , l=TB n
-+-
a b
-+-
a b
are integers and if k of the numbers a. coincide with a and the remaining
1 (= n - k) of the a:s coincide with A, while the corresponding b:s
coincide with Band b resp.
93. Let a, A, b, B be positive numbers a < A, b < B. If the two
functions j(x) and g(x) are properly integrable over the interval [Xl' x2 ]
and if a < j(x) < A, b < g(x) S;; B on [Xl' x2] then

j' [f(x)]2 dx l' [g(X)]2 dx (l/AB + Vab )2


1< ~ x, < r ab AB

= Ctf(X) g(xi dX) 2 = 2

yuana97@uw.edu
72 Inequalities

The first inequality is identical with Schwarz's inequality.


93.1. The numbers aI' a;z, ... , a.. , T, and s are positive, < s, n> 1
.. T

and E stands for J: . Then


-1 -1
(Ea:)' < (Ea:r .
94. The'function I(x) is defined on (0,1), non-decreasing, I(x) > 0,
°
but not identically zero. Let < a < b. If all the integrals occurring

r
exist we find the inequalities

(Jx"+b fIx) ax
1- (a : ~ : lY < 1 0 1 - < 1.
f ,,24 f(x) dx f x 2b fIx) ax
o 0

The inequality on the right hand side is well known. The inequality on
the left becomes an equality if and only if I(x) is a constant.

§ 2. Some Applications of Inequalities


94.1. A solid is so located in a rectangular coordinate system that
its intersection with any straight line that is parallel to one of the three
coordinate axes is either empty or consists of just one point or just one
line segment. (Such a solid need not be convex.) Let S be the surface
area of the boundary of the solid and P, Q and R the areas of its ortho-
gonal projections onto the three coordinate planes respectively. Show
that
1
2(P2 +Q2 + R2)2 < S ~ 2(P +Q + R)
and point out simple polyhedra for which the case of equality is attained
on one or the other side.
94.2. Let E denote the area of the surface of an ellipsoid with semi-
axes a, b, c and prove that
''I (be + ea + ab) <
3 =
E ~ 'n(a 2
-
+ b2 + e2 )
3 .
[Derive, or take for granted, that
E = f f (b1cz(l + clal.ql + alb2r)~dro:
the integration is extended over the surface of the unit sphere of which
(Eo '1/, C) is a point,

yuana97@uw.edu
pt. II, Chap. 2, No. 93.1-95.3 73
and dw the surface element,
II dw = 4n.]

94.3 (continued). The surface area of the ellipsoid is larger than the
surface area of the sphere with the same volume; that is
2
E> 4n(abe)S
unless a = b = e.
95. Let us call the ratio of the electrostatic capacity to the volume
of a conductor its "specific capacity". Show that the specific capacity
of an ellipsoid with three axes is always between the harmonic and the
arithmetic means of the specific capacities of the three spheres whose
radii are equal to the three semiaxes of the ellipsoid.
In analytic terms: we have to prove the inequalities

:------:- < -1f •


~ abe
3 k+~+~
< ---::----
be + ea + ab 20 V(as + u) (bl + u) (cl + u) 3'
abe

which hold for all positive number triples (a, b, e) unless a = b = e.


95.1 (continued). The capacity of the ellipsoid is larger than the
capacity of the sphere with the same volume. That is, the upper bound
in the double inequality 95 can be replaced by the sharper (abe)-l/S.
95.2. If all the roots of the equation of degree n

are real, they are all contained in the interval with the endpoints
~
na ± -",-
1 n - 1( 2
a1 - 2n
n _ 1 a2
)2 •

95.3. We consider the non-decreasing sequence of positive numbers


1'1,1'2,l'a, ...
o< 1'1 ~ 1'2 ~ I's < ....
We set 1'1 =1',
1'1" +1'2" +1'3" + ... =5"
and assume that this series is convergent for n = 1 (and so also for
n ?: 1). Prove that

yuana97@uw.edu
74 Inequalities

and that
. -5"- .
. ( -1 )1/" = I' = bm
lim
" ....... 00 s,.. ,,~co s"'+1

(If s1> S2' S3' ••• are given, we have here a "perfect" scheme for com-
putingI'. The- n-th step yields the lower bound S;;1/" and the upper bound
s,,/s"+1 for 1', the next step improves both bounds and both bounds
converge to the desired I' as n tends to 00.
Observe that

SIZ + Sr2 + S3z3 + ... = .=1 Y. - z


zG/()
~ _z_ = _ _ z_
G(z)
00

where

G(z) = jj (1 - Y.~)
.=1

and I' is the zero of G(z) nearest to the origin. Compare 197, III 342.)
95.4 (continued). Even if s,. is not given for all values of n but only
for n = 1, 2, 4, 8, ... , 2"', ... , we can devise a scheme for computing y.
Prove that
1 1 1 1
1 (1)2 <-
-<-
51 5a
(1)4 <"'<1'<"'<-
5,
(5,)4 <-
(5a )2 <- 58 5,
51
Sa

and that
lim -
11-+00
( 5" )1/" =1'.
S2ft

95.5. A wire which forms a closed plane curve C carries a unit electric
current and exerts a force F on a unit magnetic pole in the plane of, and
interior to, C. Given A, the area enclosed by C, prove that F is a mini-
mum when C is a circle and the magnetic pole is at its center.-[Assume
that C is star-shaped with respect to the magnetic pole [111109] which is
located at the origin of a system of polar coordinates r, qJ. Then
2"
F= fd;,
c
Express A.]
96. Assume that

ap .? 0, 1:" ap • = 1:" ap • = 1,
1'=1 .=1
and
p, " = 1, 2, ... , n,

yuana97@uw.edu
Pt. II, Chap. 2, No. 95.4-97 . Chap. 3, No. 98-100 75

Then we have
Y1Y2'" y" ;;::: X 1X 2 ... x".
97. Let aI' a2, ... , a" be positive numbers, M be their arithmetic, G
their geometric mean and let e denote a proper fraction. Show that the
inequality
M-G
--c-<e
implies the inequalities

1 + e < .,; < 1 + e', i = 1, 2, ... , n
where e and e' denote the only negative and the only positive root res-
pectively of the transcendental equation
(1 + x) e- x = (1 - e)" .

Chapter 3

Some Properties of Real Functions


§ 1. Proper Integrals
98. We define
g(x) = sin2 nx + sin2 nx cos2 nx + sin2 nx cos' nx + ...
+ sin2 nx COS2k nx + ...
and
G(x) = lim g(n I x).
"~oo
Is G(x) integrable?
99. Let the function I(x) (d. also 169, VIII 240) be given by

I(x) = 1
0
q1
if

if
x is irrational

x is rational. x = - •
P
q
(p, q) = 1, q > 1.

Show that I(x) is continuous for every irrational x, discontinuous for


every rational x and properly integrable over any interval.
100. The two functions I(x) and 9'(x) are properly integrable over
[a, b]. Subdivide the interval:

X,_l<Y.<X" x._ 1 < 'Y/, < x.' v=1,2..... n.

yuana97@uw.edu
76 Some Properties of Real Functions

If max (x. - x._ 1) --+- 0 (the subinterval of maximal length converges to


0) we obtain the relation
ft b
lim I/(Y.) qJ(fJ.) (x. - x._ 1 ) = f /(x) qJ(x) ax.
""'00 .-=1 •

101. Suppose that /(x) is properly integrable over [a, b] and qJ(x)
properly integrable over [a, b + a], a> o. Then
b b
lim
~~+Oa
f lex) qJ(X + 6) ax = af lex) qJ(X) ax.

102. Let /(x) denote a properly integrable function on [a, b]. There
exist to every positive number 8 two step-functions, 'I'(x) and 'P(x), such
that for the entire interval [a. b]
'I'(x) :5: /(x) :5: 'P(x)
and
b b
f 'P(x) ax - f 'I'(x) ax < 8.
a a

It is even possible to choose 'I'(x) and 'P(x) so that their points of discon-
tinuity are equidistant.
103 (continued). If lex) is of bounded variation 'I'(x) and 'P(x) may
be chosen so that the total variation of neither exceeds the total varia-
tion of /(x).
104. We define
4[x] - 2[2x] + 1 = s(x).
Then we have (1' integer) the limit relation
1
lim
ft~oo 0
f lex) s(nx) ax = 0

for any properly integrable function lex) on [0, 1]. [Sketch s(nx), VIII 3.J
106. Let /(x) be properly integrable over [a, b]. Then we can prove
that
b
lim f lex) sin nx ax = O.
" ...... 00 •

108 (continued). Yet


b 2 b
lim
,,~oo.
f lex) Isin nx Iax ="i" f

lex) ax.

Suppose that the function /(x) is bounded on the interval [a, b] and
that this interval is subdivided by the points xo' Xl' x2 ' ••• , x.. _1 ' x.. '

yuana97@uw.edu
Pt. II, Chap. 3, No. 101-113 77

whereby

The greatest lower and the least upper bounds of I(x) in the subinterval
[x._1' x.] are denoted by m. and M. respectively. Then M. - m. is
called the oscillation of I(x) on the v-th submterval. The function I(x) is
properly integrable over [a, b] if and only if to every pair e, 'Y/ of positive
numbers a subdivision of the interval can be found so that the total
length of the subintervals for which the oscillation is larger than e is
smaller than 'Y/. [Riemann's criterion, d. I.e. 105, Riemann, p. 226.]
107. The function (~ - [~]) XX is properly integrable over [0, 1]
for IX > O.

108. If I(x) is properly integrable over [a, b] the points of continuity


of I(x) are everywhere dense on this interval.
109. The function I(x) is properly integrable ove: [a, b]. The equation
b
J (/(x))2 ax = 0
II

is satisfied if and only if I(E) = 0 at every point of continuity E of I(x),


a< E< b.
110. Assume that the function y = I(x) is properly integrable over
[a, b], that m ::;; I(x) < M on this interval and that the function tp(y)
is continuous on em, M]. Then also tp[f(x)] is properly integrable over
[a, b].
111. If I(x) and tp(y) are properly integrable tp[f(x)] need not be
properly integrable. [98, 99.]

§ 2. Improper Integrals

112. If I(x) is a monotone function on the interval (0, 1] and if


1
Jx"1(x) ax exists then
o
lim x"+1/(x) = O•
......0

J
co

113. If I(x) is monotone on the interval [1,00) and if %"/(x) ax


exists then 1

lim
...... co
X H1 I(x) = o.

yuana97@uw.edu
78 Some Properties of Real Functions

114. Determine those pairs of real values Pfor which the integral
r~
(x,

Icos xl# dx
o
is convergent.
114.1. Construct a function I(x) that takes positive values and is
bounded and integrable in any finite subipterval of [0,00) and such that
00

f [I(x))'" dx
o
converges for (X = 1 but diverges for any real value of (X different from 1.
115. The functions
iI(x), Mx), ... , I,.(x) ,

are properly integrable over any finite interval and they satisfy the
following conditions:
On every finite interval we :pave lim I,.(x) = I(x) uniformly in x.
"-+00

f
00

There is a function F(x) such that 1I,. (x) 1 :::;;: F(x) and F(x) dx
exists.
Then the limit and the integral can be interchanged:
00 00

lim
n~oo - 0 0
f I,.(x) ax = f
-00
I(x) ax.

[Analogous to I 180.J
116. Prove 58 using VI 31.
117. If the Dirichlet series [VIII, Chap. 1, § 5J
":J l- S + a22- s + as3- s + ... + a,.n- s + ... = D(s)

converges for s = (f, (f > °we obtain for s > (f

f
00

D(s) res) = P(y) yS-l dy,


o
where
a1e-" + a2e- 2" + ase-S" + ... + a,.e-"" + ... = P(y).
118. Suppose that I(x) is properly integrable over every finite interval
00

and that f I/(x) 1 dx exists. Then we have

2
f f = -n_ f
+00 +00 +00

lim I(x) sin nx dx = 0, lim I(x) 1sin nx 1 dx I(x) dx.


11-)000 -"X) n~oo_oo oo

yuana97@uw.edu
Pt. II, Chap. 3, No. 114-120 79

118.1. There are rational functions R(x) such that for an arbitrary
function I(x) of the real variable x

J I(R(x» dx = J I(x) dx
00 00

provided that the integral on the right hand side exists. Show that this
property belongs to those, and only to those rational functions that are
of the form
R(x) = 6(X -IX -~-~ -
X-IX1 X-IX2
••• -~)
X-IX"

where 6 = +1 or -1, IX, lXI' 1X2' ••• ,IX" are real, and PI' P2' ••• ' P.. are po-
sitive, numbers; n ;;;::: o. (The case n = 0 must be interpreted as meaning
R(x) = 6(X - IX).)

§ 3. Continuous, Differentiable, Convex Functions


119. Are there actually functions of three variables?
More precisely: Is it possible to write every real function I(x, y, z)
of three variables with the help of two functions lP(x, y) and 1p(u, z) in
the form of
I(x, y, z) = 1p(IP(x, y), z) ?
Discuss the question:
(1) if I(x, y, z), lP(x, y), 1p(u, z) are defined for all real values of the
variables,
(2) if these functions are defined for all real values of the variables
and are continuous.
1198. Setting x + y = S(x, y), xy = P(x, y) we can write
yz + zx + xy = S{P(x, y), P[S(x, y), z]};
in this formula xy + yz + zx is composed of four functions of two
variables "boxed" in each other. Prove that it is impossible to express
I(x, y, z) using only three functions of two variables boxed in one
another if these functions are defined for all pairs of real values and if
they are arbitrarily often differentiable. [We have to show that
I(x, y, z) = yz + zx + xy cannot be represented by any of the combina-
tions 1P{1p[x(x, y), zJ, z}; 1P[1p(x, z), X(y, z)], 1P{1p[x(x, y), zJ, x}.J
120. Let the function I(x) have a continuous derivative on (a, b).
Decide whether it is possible to find for each point ~ of this interval
two points Xl' x2 , Xl < ~ < x 2 , such that
1(x2) - 1(x1) = I' (~).
X2 - Xl

yuana97@uw.edu
80 Some Properties of Real Functions

121. Assume that the function I(x) is differentiable on [a. bJ. but
not a constant and thatf(a) = f(b) =0. Then there exists at least one point €
on (a. b) for which
4
J I(x) dx.
b
II'(E) I > (b _ a)2
II

122. The function I(x) is twice differentiable. Then there exists a


point Ein (xo - '. Xo + ,) where
3 x.+'
I" (E) = r J
x,-,
[j(x) - I(xo)] dx.

122.1. When is the mean value of a function in an interval a simple


mean of those two values that the function takes at the endpoints of the
interval?
Assume that I(x) is defined. bounded and integrable in the interval
[a. b]. Introduce the abbreviations
I(u) = u. I(v) = V.

---
1
v-u .
J I(x) dx ~ W
v

and determine the most general function I(x) satisfying


U+V
W=-2-
for all u and v subject to the condition
a<u<v<b.
122.2 (continued). Analogous question for
W=VUV;
assume that I(x) > 0 for a :::;;: x < b.
122.3 (continued). Analogous question for
2UV
W=u-+-v;
assume that I(x) > 0 for a :::;;: x :::;;: b.
123. The numbers Po. Pl' P2' ...• P" • ... are non-negative. at least two
of the P. 's do not vanish. Then the logarithm of the series
Po + PIC + P2e'1x + ... + p"e"x + ...
is a function of x. convex on every interval where the series converges.
124. A bounded convex function [po 65 J is everywhere continuous
and it is even everywhere differentiable from the left and from the right.

yuana97@uw.edu
Pt. II, Chap. 3, No. 121-129 81
125. Suppose that the real function f(x) is defined on a finite or
infinite interval and that it has there a continuous derivative I'(x).
Consider the points of intersection of all the horizontal tangents of the
curve y = lex) with the y-axis, i.e. the set M of all points
y = f(x) for which f' (x) = O.
Prove that the set M cannot fill out an entire interval. (This proposition
admits a farreaching generalization.)
126. If a monotone sequence of continuous functions converges on a
closed interval to a continuous function it converges uniformly. (Theo-
rem of Dini.)
127. Prove th~ following counterpart to 126: If a sequence of
monotone (continuous or discontinuous) functions converges on a closed
interval to a continuous function it converges uniformly.

§ 4. Singular Integrals. Weierstrass' Approximation Theorem


128. If the functions
... , P,,(t) ,
are continuous on the interval [a, b] and if they satisfy the conditions
b
P,,(t) ~O, J P,,(t) dt = 1, n = 1, 2, 3, .,,'
"
then the terms of the sequence
b b b
J Pl (t) f(t) dt, J P2(t) f(t) dt, ... , J P,,(t) f(t) tit, ".
" " "
are between the minimum and the maximum of the continuous func+ion
f(t). (Cf. I 65, 179, 183.)
129. Let x be a fixed point of the interval [a, b] considered in 128.
In order that
b
lim
11"",+00 a
J P,,(t} f(t) dt = f(x)

holds for all functions f(t} continuous on [a, b] it is necessary and suffi-
cient that

lim
"-+00
(7"
a
P,,(t) dt + j P,,(t) dt) =
s+a
0

for all positive values of e for which a < x - e < x + e < b (if x = a
or x = b the first or second integral resp. under the limit sign has to be
omitted.) (Cf. 166, 180, 184.)

yuana97@uw.edu
82 Some Properties of Real Functions

130. We have
J e-
00

lim 8 al 1(1) dt = lim I(t),


...... +0 0 , ..... 00

provided that the integral on the left and the limit on the right hand side
exist.
131. If the integral

J (-1(1) dt
00

o
converges for A = IX and for A = p, IX < p, it converges for IX < A :s;; p
and it represents a continuous function of A on that interval.
132. We assume that
Pl(x,I), P2(x,I), ... , p,.(x,t), ...
are continuous functions of x and t, a :s;; ; :s;; b, and that for each n
b
p,.(x, t) > 0,
..J p,.(x, t) dl = 1.

Let 1(1) denote a continuous function. The functions


b
I,.(x) =
..J p,.(x, t) I(t) dt,
n = 1, 2, 3, •.•

lie between the minimum and the maximum of 1(1) on [a, b] for any x,
a < x < b; i.e. min I(x) :s;; I,.(x) :s;; max I(x). Furthermore
.. ~s~b .. ~s~b

lim I,.(x) = I(x)


, . ..... 00
for a < x < b,
provided
lim (7'p,.(x, t) dt
#-+00 a
+s+aj p,.(x, I) dt) = 0

uniformly for a + 8 :s;; X < b - 8, 8 fixed and positive. The convergence


is uniform on any closed subinterval of (a, b).
133. Let I(x) denote a continuous function on [0, 1]. The convergence
of
. 1 3 5
, . ..... 00 2
2n+l 1
lim - . - . - ... - - I(t) [1 - (x - 1)2]" dt = I(x)
2 4 2n 0
J
is uniform for 8 < X < 1 - 8, 0 < 8 < t, 8 fixed.
134. Let I(x) be a continuous periodic function with period 2n. Then

lim "2-
, . ..... 00
1

nn 0
J I(t)
2,.
( x-t)2
_
sinn-
dt
2-
sin x_ _
t = I(x) ,
2
the convergence is uniform for all x.

yuana97@uw.edu
Pt. II, Chap. 3, No. 130-141 83
135. Every function defined and continuous on the finite interval
[a, b] can be approximated uniformly on [a, b] by polynomials to any
degree of accuracy. (Weierstrass' approximation theorem.)
138. Every continuous function that is periodic with period 2:n: can
be uniformly approximated by trigonometric polynomials [VI, § 2] to
any assigned degree of accuracy. (Weierstrass' approximation theorem.)
137. Let I(x) denote a function that is properly integrable over [a, b]
([0, 2:n:]). Two polynomials (trigonometric polynomials), P(x) and P(x),
can then be found for any positive 8 so that for a < x < b (0 ::;; x ::;; 2:n:)
P(x) ::;; I(x) ::;; P(x)
and
b b
f f
2,. 2,. )
P(x) dx - P(x) dx < 8 (/ P(x) dx - / P(x) dx < 8 •
" "
138. The n-th moment of a function I(x} is given by
b
f I(e) f'de.
"
If all the moments of a function that is defined and continuous on the
finite interval [a, b] vanish then the function vanishes identically.
139. If all the moments
b
f I(e) f'de, n = 0, 1, 2, ... ,

of a function that is properly integrable over the interval [a, b] vanish


then the function I(x) vanishes at every point of continuity.
140. If the first n moments vanish,
b b b b
f
a
I(x) dx = J I(x) x dx = f I(x) x
A a
2 dx = ... = f I(x) x.. - 1 dx.= 0,
a

of a function /(x) defined and, continuous on the finite or infinite


interval (a, b) then the function changes sign (V, Chap. 1, § 2) at least
n times in the interval (a, b) unless it is identically O.
141. The 2n-th and (2n + 1)-th trigonometric moment (Fourier
constants, cf. VI, § 4) of a function with period 2" are defined as
2,. 2,.
f cos nx I(x) dx and f sin nx /(x) dx.
o o
If the first 2n + 1 trigonometric moments of a continuous function I(x)
with period 2" vanish then I(x) changes sign at least 2n + 2 times in
any interval of length> 2" (V, Chap. 1, § 2) unless I(x) is identically O.

yuana97@uw.edu
84 Some Properties of Real Functions

142. Let the function cp(x) be defined and continuous for x ~ o.


Suppose that the integral
co
J(k) = f e- h cp(x) dx
o
converges for k = ko and that it vanishes for a sequence of k's increasing
in arithmetic progression:

J(ko) = J(ko +!%) = J(ko + 2£x) = ... = J(ko + n!%) = ... = 0, !% > o.
Then cp(x) vanishes identically.
143. The r-function

r(s) - lim S
n nl
- ,,-+co s(s + 1) ••• (s + n)
can be written as an integral [31]. Use this fact to prove that r(s) does
not have any zeroes. [r(s + 1) = sr(s), 142].

We associate with each function that is defined on [0, 1] the poly-


nomials

K" (x) = >~ j ( : ) ( : ) x· (1 - x)" -v, n = 0, 1, 2, ...

This polynomial is bounded on [0,1] from below by the greatest lower


bound, and from above by the least upper bound, of j(x) and it coincides
with j(x) at the endpoints.
144. Work out the polynomials K,,(x), n = 0, 1, 2, ... for

j(x) = 1, j(x) = x, j(x) = x2 , j(x) = ~.

145. Let x be any point on [0, 1] and

1= 1: (:) xV(1 -
>=0
x)"-V = EI + Ell,
where EI refers to the subscripts for which I'll - nx I < n 3f4 and Ell to
those for which I'll - nx I > n 3f', n > 1. Then
1
1 --
ElI <4"n 2.

146. Let j(x) be continuous on [0, 1]. The polynomialsK,,(x) converge


uniformly to j(x) on [0, 1]. (New proof of Weierstrass' theorem, 135.)

yuana97@uw.edu
Pt. II. Chap. 3. No. 142-146· Chap. 4. No. 147-150

Chapter 4

Various Types of Equidistribution

§ 1. Counting Function. Regular Sequences


In the sequel we are considering monotone sequences of positive num-
bers. The counting lunction N(r) of such a sequence r1 , r2 , ••• , r", ... ,
0< r1 < r2 < r3 < ... < r" :::;;: ... , is defined as the number of those
r,,'s that are not larger than r, r~ 0:
N(r) = 1: 1.
',,:;;'
(If I(t) is a function of t then 1: I(r,,) denotes the sum
',,:;0'
l(r1 ) + l(r2 ) + ... + I(r".), r". < r < r".+1.) E.g. if r1 = 1, r2 = 2,
r3 = 3, ... then N(r) = [r].
N(r) is a piecewise constant, non-decreasing function whose jumps are
integers and which is everywhere continuous on the right.
147. If I(t) is differentiable and t'(t) properly integrable. t > 0, then
,
1: I(r.. ) = N(r) I(r) - J N(t) t'(t) dt.
',,:;0' 0

148. Let N(r) denote the counting function of the sequence


r1 , r:,!, r3 , ••• , r", ... , which increases to infinity. Then

Iimsup-=N(r) Iimsup-, n Ii mm - = Iimin-,


·fN(r) ·fn
'-+-00'" "-+-00"" '400 r n400""

lim sup .log N(r) = lim sup log n , IimIn . £ log N(r) = ImIn
i · flog
- -n.
'-+00 log r "-+00 log r" '-+00 log r "-+00 log r"

149. The counting function N(r) and the convergence exponent A of


the sequence r1 , r2 , rs •... , r" • ... [I, Chap. 3, § 2] are connected by the
relation
.
11m log N(r) ,
sup - - - = A.
'-+00 log r
150. A function L(r) defined and positive for r > 0 is termed slowly
increasing if it is monotone increasing and satisfies the condition
· L(2r) - 1 .
1Im---
'-+00 L(r)
Sho.w that
c>O.

yuana97@uw.edu
86 Various Types of Equidistribution

151. Suppose that L(r) is positive for r> 0, monotone increasing


and that for r sufficiently large

L(r) = (log 1')"" (log2 1')"" ••• (log" 1')"''', IXl > O.

[log" x = log"_l (log x).] Then L(r) is slowly increasing.


152. If L(r) is slowly increasing then
- 0
lim log L(r) -
r~oo log r
.

153. If N(r) denotes the counting function of the sequence


1'1' 1'2' rs ' ... , rn , ••• and if

N(r) N rtL(r) ,

where L(r) is slowly increasing, 0 < A. < 00, then A. is the convergence
exponent of the sequence 1'1' r2 , 1'3' ••• , rn , •.•

A sequence r1 , 1'2' r8 , ••• , r.. , ... of the type considered in 153 will
be called a regular sequence in the sequel, 154-159. Later on (e.g. IV
A
59-IV 65) sequences for which N(r) N :'(r) will also be termed regular.
If we take the term in this broader sense also the prime numbers 2. 3, 5,
7, 11, ... form a regular sequence and the propositions 153-159 remain
valid without alteration.
154. The counting function of a regular sequence with convergence
exponent A. satisfies the relation
r N(cr) _ A
C> O.
r-!.~ N(r) - c,

155. Let N(r) be the counting function of the regular sequence


r1' r2 • rs •...• rn •... with convergence exponent A. and t(x) be a piecewise
constant function on the interval (0, c], c > O. Then

156. The limit relation in 155 is also valid if t(x) denotes a properly
integrable function on [0, c].
157. Let N(r) denote the counting function of the regular sequence
r1 , r2 , •••• rn , ••• with convergence exponent A. and let IX > O. Then

lim N()
1
L:
(r~ )"'-A = J x",-A dx = -.
1 ).
A
'-+00 t' 'n~" r 0 ex

yuana97@uw.edu
Pt. II. Chap. 4. No. 151-161 87

158 (continued).

lim- ~ ~
1 (1' )-"'-A =J X co
-",-A
-
A
A
dx=-.
'-+00 N(r)
',. > , l' 1 ex

159. Assume that N(r) is the counting function of the regular se-
quence 1']. 1'2' ••• , 1',., ••• with convergence exponent A, that I(x) is defined
for x> 0 and properly integrable over every finite interval [a, b],
o < a < b and that furthermore
I/(x) I < x"'-A In a neighbourhood of x = 0
and
I/(x) I < X-",-A in a neighbourhood of x = 00, IX> O.
Show that

~~ N~I') ,.~/(:") = /coAX~)dX.


160. Suppose that the function I(x) is defined and monotone on the
interval (0, 1] and that it satisfies in the neighbourhood of x = 0 the
inequality
I/(x) 1< X",-A, IX> O.

The counting function and the convergence exponent of the positive


sequence 1'1,1'2' ... ,1'", ••• are called N(r) and A resp.; let 0 < A < 00.
Then

liminf
'-+00
l
N (l') ~ 1(:")< I 1(x"+) dx < lim sup N l(r) ~ 1(:").
',.:;0' \ 0 '-+00 ' .. :;0'
[1115.]

The sequence 1'.. need not be regular.


161. The function I(x) is defined for x > 0, is positive and decreasing
and satisfies the inequalities
I(x) < X",-A in the neighbourhood of x = 0
I(x) < X-",-A in the neighbourhood of x = 00. IX> O.
The sequence 1'1' 1'2' rs •... , ' .. , ... is defined as in 160. Then

[1116.]

§ 2. Criteria of Equidistribution
A sequence of the form

yuana97@uw.edu
88 Various Types of Equidistribution

is called equidistributed in the interval [0, 1] if all the Xl' X 2 ' XS ' ••• , X .. ' '"
are on [0, 1] and if for every function that is properly integrable over
[0,1] the following equation holds:
(*) lim !(xI )
"-+00
+ !(x2 ) + ... + !(x.. )
n
= JI(x) dx.
0

°
The term "equidistribution" is explained by the following criterion:
162. A sequence Xl' X 2 ' XS ' ••• , X .. ' ••• , ~ x.. ~ 1, is equidistributed
on [0,1] if and only if the "probability" of a term x,. to fall into a certain
subinterval of [0, 1] is equal to the length of that subinterval. More
precisely, if the sequence has the following property: Let [IX, (3] be an
arbitrary subinterval of [0,1] and 'II,.(IX, (3) denote the number of x/s,
'11= 1, 2, ... , n, on [IX, {3], then

• V .. (<X , {J)
lim - - = {3 -IX. [102.]
n4-oo n
163. Let [IX, (3] be an arbitrary subinterval of [0,1] and S.. (IX, (3)

°
denote the sum of the x/s, '/I = 1, 2, , ... , n, that fall into [IX, {3]. A se-
quence Xl' X 2 ' Xs ' ... , x .. ' ... , < X < 1, is equidistributed if and only if
. 5,. (<X, Il) {Ji _ <xi
lim---=--.

°
"-+00 n 2
164. A sequence Xl' X 2 ' XS ' ••• , X .. ' ••• , < x,. < 1, is equidistributed
on [0, 1] if and only if for every positive integer k

k
Xl + X 2k + Xsk + ... + x..k 1
..li-+~ n =k+l' [137.]

165. A sequence Xl' X2 ' XS ' ••• , X.. ' ••• , °< x.. < 1, is equidistributed
on [0, 1] if and only if the two equations

lim cos 2nkxl + cos 2nkx2 + ... + cos2nkx,. = 0,


n

°
" .... 00

lim sin 2nkxl + sin 2nkx2 + ... + sin 2nkx,. =


"-+00 n

hold for every positive integer k. [137.J

§ 3. Multiples of an Irrational Number


166. Let 0 be an irrational number. The numbers
X,. = On - [OnJ
are equidistributed on the interval [0, IJ.

yuana97@uw.edu
Pt. II, Chap. 4, No. 162-172 89

167. Let 0 denote an irrational number. Put E" = 1 or E" = 0 accord-


ing as the integer next to 1&0 is larger or smaller than nO. Let a and d,
a > 0, d > 0, stand for two integers. Then we find that
,,-I
1: 8,,+1111
lim _"=_0__ 1
"-+00 n = 2'
168. Let 0 be an irrational number and tx be defined as tx = qO, q
integer, q 9= o. As z converges to ;,m.. along the ray arg z = 2ntx, the
function
I(z) = :J; (nO - [nOJ) z", z arbitrary complex, Iz I < 1,
,,=1
increases to 00 in such a way that
lim (1 _ r) l(re 271i"') = _1_. [188] .
'-+1-0 23nq
169. Determine for real x the function
I(x) = lim cos2 nx + cos· 2nx + cos6 3nx + ... + cos2.. n:n:x •
"-+00 n

170. The decimal fraction


o= 0.12345678910111213 ...
(the natural numbers listed consecutively) represents an irrational
number. According to 166 the numbers
nO - [nO], n = 1, 2, 3, ...
are everywhere dense on the interval [0, 1J. Show that this is already
the case for the subset
10"0 - [10"OJ, n = 0,1, 2, 3, ...
171. The number
111 1
e= 1 +1T +21 +31 + ... +nr + ...
is irrational. [VIII 258.] Prove that the only limit point of the set
n! e - en! eJ, n = 1, 2, 3, ...
is zero.
172. Suppose that the polynomial P(x) = a1 x + a2 x 2 + ... + a,x'
has at least one irrational coefficient. Then the numbers
P(n) - [P(n)] , n = 1, 2, 3, ...
have infinitely many limit points.

yuana97@uw.edu
90 Various Types of Equidistribution

173. Let 0 be an irrational number, x,. = nO - [nOJ, n = 1, 2, 3, ...


and let ~1' ~2' ~3' ••• , ~,., •.• be a monotone decreasing sequence of
positive numbers whose sum diverges. Then we find for any properly
integrable function I(x) on [0, 1J that
. IX 1!{x1) + IX2/{x 2) + ... + IX,.!{X,.) Jl
lim
,.~~ IX1 + IX2 + ... + IX,.
= 0 I(x) dx.

§ 4. Distribution of the Digits in a Table of Logarithms


and Related Questions
174. The function g(t} has the following properties for t > 1:
(1) g(t) is continuously differentiable;

°
(2) g(t) is monotone increasing to <X> as t ~ <x>;
(3) g' (t) is monotone decreasing to as t ~ <x>;
(4) tg' (t) tends to <X> as t ~ <x>.
Then the numbers
x" = g(n) - [g(n)J, n = 1, 2, 3, ...

°
are equidistributed on the interval [0, 1J.
175. Suppose that a > 0, < C1 < 1. The sequence
x,. = ana - [an"J
is equidistributed on the interval [0, 1J.
176. Let a > 0, C1 > 1. The numbers
x,. = a (log n)" - [a (log n)"J


are equidistributed on [0, 1J.
°
177. For < a < 1, ~ the series
sin 1"~ + sin 2"~ + sin 3"~ + ... + sin n"~ + ...
~ ~ ~ ~

is absolutely convergent if and only if e > 1.


178. Suppose that the square roots of the natural numbers 1, 2, 3, ...
are written up one below the ,other in an infinite array. Examine the
digits at the j-th decimal place (to the right of the decimal point), j > 1.
Each digit 0, 1, 2, ... , 9 appears on 'the average equally often. More
precisely: let vg(n) denote the number of those integers < n whose
square roots show a g at the j-th decimal place. Then
• Pg{n) 1
lim -n- = -10 , g = 0, 1, 2, ... , 9.
°
,.~~

179. Assume a> and x,. = a log n - [a log nJ, n = 1, 2, 3, ...


and that the arbitrary function I(x) is defined and properly integrable

yuana97@uw.edu
Pt. II. Chap. 4. No. 173-182 91

over [0. 1]. Then the limit relation


lim f(x l )
n-+oo
+ f(x 2 ) + ... + f(xn)
n
=
0
j I(x) K(x, ~) dx
holds provided that n increases to infinity in such a manner that xn ~~,
e
o ~ < 1. The function K(x, ~) is given by

K(x,~) = j log q rt-HI


q- 1

log q rt-<
if 0

if ~
<x<
<x <1,
~

q= el/a , 0< ~ < 1;


q-1
logq
K(x, 0) = K(x, 1) = - 1 rt·
q-
180 (continued). The limit points of
f(x l ) + f(x 2 ) + ... + f(xn)
n = 1, 2, 3, ...
n
cover an entire interval 1 = lea, I) which depends on a and I only.
This interval degenerate into a point if and only if I(x) = c. c a
constant, at each point of continuity. What can you say about lea. f) if
a is a very large or a very small positive number?
181. Suppose that the common logarithms (to the base 10) of the
natural numbers 1, 2, 3, 4•... are listed below each other in an infinite
table of logarithms. Consider the digits at the i-th decimal place (to the
right of the decimal point), i > 1. There exists no definite probability
for the distribution of the digits O. 1. 2•...• 9 in this sequence. More
exactly: let vg(n) denote the number of those integers < n whose loga-
rithms show the digit g at their j-th decimal place. Then the quotients
"g(n)
- - d 0 no t have a 1"lmlt as n~oo: Th'
elr li' .
mIt pomts f'll .
lOut an entIre
n
interval of positive length.
182. The function get) has the following properties for t > 1:
(1) get) is continuously differentiable;
(2) get) is monotone increasing. to 00 as t ~ 00 ;
(3) g' (t) is monotone decreasing to 0 as t ~ 00;
(4) tg'(t) ~ 0 as t ~ 00.
(Cf. 174.) Then the numbers
Xn = g(n) - [g(n)]. n = 1, 2, 3, ....
are everywhere dense on the interval [0, 1] but they are not equidistri-
buted. Their distribution is characterized by the following limit theorem:
Let the function I(x} be properly integrable over the interval [0, 1]. If n

yuana97@uw.edu
92 Various Types of Equidist;ribution

increases to infinity so that x,. _~, 0 < ~ < 1, then


!(x1) + !(x2) + ... + j(X,.)
lim
n-+co n
=/(~)

holds provided that I(x) is continuous at x = ~. If I(x) has a simple


discontinuity (jump) at the point x = ~, the set of limit points of
l(x1 ) + !(x2 ) + ... + lex,,)
n
covers the interval rJ(~ - 0), I(~ + 0)]. The statement is true also for
~ = 0 or ~ = 1 if 1(1) = 1(0) and if I(x) is extended so that it becomes a
periodic function with period 1. [Then 1(1 + 0) = I( +0),/(1 - 0)= I( -0).]
183. The sequence
x,. = a (log n)a - [a (log n)a], n = 1, 2, 3, ...
is for 0 < (1 < 1 everywhere dense on the interval [0, 1] but not equi-
distributed. [178, 179.]
184. Assume that the square roots of the logarithms of the
natural numbers 1, 2, 3, 4, ... are tabulated below each other in an
infinite array. Consider the digits at the i-th decimal place (to the right
of the decimal point) i > 1. There exists no definite probability for the
distribution of the digits 0, 1, 2, ... , 9 at the i-th decimal place. More
exactly: Let vg(n) denote the number of integers k among the first n
integers for which Vlog k has the digit g at the j-th decimal place. Then
" (n)
the quotients-g- , n = 1, 2, 3, ... are everywhere dense between 0 and 1.
n

§ 5. Other Types of Equidistribution


185. Imagine in the p-dimensional space a rectilinear uniform motion
described by the equations x.(t) = a. + O.t, a., 0. constants, v = 1, 2, ... , p,
°
t time. If the numbers 01> 2 , ••• , Op are rationally independent (i.e. if
nl OI + n2 02 + ... + npOp = 0, n1 , n2 , ••• , np rational, has the only solu-
tion n i = n2 = ... = np = 0) any function l(x1 , x2 , ... , xp), that is
periodic in Xl' x2 ' ••• , xp with period 1 and properly integrable over the
unit cube 0 ::s: x. ::s: 1, v = 1, 2, ... , p, satisfies the relation
. 1 1 1 1
JI(x (t), x (t) , ... ,xp(t») dt = J J... Jl(x
1
lim T i 2 1, x2 ' " ' ' xp) dX I dx2 ' " dxp.
1-..00 0 0 0 0

188. Let be arbitrary constants, 0 <


IXl> lXI' PI' PI IXI < IXs < 1,
o < PI < P2 < 1. The conditions
IXI ::s: x - [x] < lXI' PI ::s: y - [y] < Pa

yuana97@uw.edu
Pt. II, Chap. 4, No. 183-189 93
determine an infinite number of rectangles with sides parallel to the
axes and congruent mod 1, i.e. they are congruent by translations
parallel to the axes through integral lengths. The equations x = a + (J1t,
y = b + (J.j, a, 81 , b, 8. constants, t time, define a linear uniform motion.
Let T(t) denote the sum of the time intervals up to time t the moving
point is spending in one of the above mentioned rectangles. In the case
where 81 : 82 is irrational we can establish the relation

lim Tt(t) = (oc. - ~) ({J. - (Jl) .


1-+00

187. A billiard ball is moving rectilinearly with constant speed on a


smooth square table with surface ~. The ball is reflected by the cushion
each time according to the law of reflection (angle of incidence = angle
of reflection). Suppose that the tangent of the angle between the direction
of the motion and a side of the billiard table is irrational. We denote by
T(t) the sum of the time intervals up to the time t that the moving ball
spends in a certain subregion of size f. Then
lim !(t) = 1. .
/-+00 t lj

The numbers
1 2 3 n
n' n' n' ... , -;-, n = 1, 2, 3, ...

which appear in the construction of sums of rectangles (subdivision


according to an arithmetic progression) are in a certain sense equi-
distributed. A similar type of equidistribution comes up in the next two
problems.
188. Let "1,.. "2,.. "3'" ... ,"'1'" denote the positive integers that are
smaller than n and relative prime to n; their number is rp = rp(n) [VIII
25J. Then

lim
(1'n1") +1 (1'-n2") +1 (1'-n3") + ... +1 (1''1''')
1 - -
n = f
1
f(x) dx
"-+00 q>(n) 0

holds for any properly integrable function f(x) on [0, 1J. [VIII 35.]
189. We write down in increasing order all reduced fractions ~ 1
whose numerators and denominators are among the numbers 1, 2, 3, ... , n:

(Farey series

w1 = ! ' ... , wN = !, N = N(n) = rp(1) + rp(2) + ... + rp(n)).

yuana97@uw.edu
94: Various Types of Equidistribution

Then the relation


. I(w)) + l(w2 ) + I(ws ) + ... + l(wN ) Jl
lim N = I(x) dx
.~~ 0
holds for any properly integrable function I(x) on [0,1]. [170.]

Some of the number sequences occurring in the preceding problems


were equidistributed, i.e. the probability that a number would fall into
a certain interval was proportional to its length, e.g. 166, 175, 188. This
will not be the case in the following examples: For these sequences there
exists a certain probability density according to which the density of
points may be different in different subintervals. A similar case appeared
already in 169. ,_
190. Suppose that I(x) is a properly integrable function on [0, V!]
and that there exists a positive number p such that x-PI(x) is bounded
on this interval. We set

Vn(:) =5,., " = 0, I, ... , n; n = 1,2,3, ...


2"

J (1Yn/ 2e-
Then
lim I(so.) + I(sl.. ) + l(s2.. ) + ... + I(s.... ) = dx .
Vn
00 2SI)
/

• ~DO -DO

191. Let
Xl .. , X 2.. , ••• , X .... ' -1 < x... < 1, " = 1, 2, ... , n,
be the zeros of the n-th Legendre polynomial p .. (x) [VI 97] and A be
real, A > 1. Then

Iog(l + Xl") + log (1 + X2.. ) + ... + log (1 + X.... )


A A A I
V--
A + AI - 1
lim
.~DO n = og 2A '
where the positive value of the square root is considered. [Use 203.]
192 (continued). Let k be any positive integer. Show that

Iim ~.. + ~.. + ... + X!.. = -1 In cos...v d~o.'v. 0.


[1179.]
"-+00 n no
193. Let
XIII' x2." ••• , X....' -1 < x... < 1, ,,= 1, 2, ... , n,
denote the zeros of the n-th Legendre polynomial p .. (x), and I(x) be a
properly integrable function on [-1,1]. Then
lim I(x l .. ) + l(x2.. ) + ... + I(x.... ) = ~ j
I(cos -D) d{} •
• -+00 n no

yuana97@uw.edu
Pt. II, Chap. 4, No. 190-194· Chap. 5, No. 195-197 95

194. Assume that IX ::;;: x ::;;: {J is an arbitrary subinterval of [-1, 1]


and that V.. (IX, (J) is the number of zeros in [IX, (J] of the n-th Legendre
polynomial. Then
lim " .. (IX, fJ) = arccos IX - arccos p •
"-+00 n n

The points x... are not equidistributed on the interval [-1,1] but the
values arccos x.. are equidistributed on [0, on]. We may interpret the
interval [-1, 1] as the horizontal diameter of a circle and each point x
as the normal projection of two points of the circumference onto the
diameter. We are facing here an equidistribution on the circumference
but not on the diameter.

Chapter [)

Functions of Large Numbers


§ 1. Laplace's Method
195. Let PI' P2 , ••• , PI> a1, a2 , ••• , aJ be arbitrary positive numbers.
Then

exists and it is equal to the largest among the numbers al' ~, ... , al .
196. Under the same hypotheses as in 195

197. Let /(x) be an arbitrary polynomial whose zeros are all real
and positive and for which

_f(~) = c
f(~) 0
+ ex
I
+ c2x2 + ... + ex"
..'
...

Show that

exists and that it is equal to the smallest zero of /(x).

yuana97@uw.edu
96 Functions of Large Numbers

198. The two functions <p(x) and f(x) are continuous and positive on
the interval [a, b]. Then

,,~~ -V/ f/J(x) [I(x)]" dx


exists and is equal to the maximum of f(x) on [a, b].
199. Under the same hypothesis as in 198
b
JIp(x) [f(x)]"+1 ax
lim II b = max f(x).
Boo JIp(x) [f(x)J" dx
II

200. Let k be a positive constant and a < E< b. Show that for
a, b, E, k fixed and n -+ 00
Jbe-A"(%-E)' dXNYkn""'
1m
II

201. The functions f/J(x), h(x) and f(x) = e"(%) are defined on the
finite or infinite interval [a, b] and satisfy the following conditions:
(1) f/J(x) [I(x)]" = f/J(x) e""'*") is absolutely integrable over [a, b];
n = 0,1,2, ...
(2) The function h(x) attains its maximum only at the point E in
(a, b); moreover, the least upper bound of h(x) is smaller than h(E) on
any closed interval that does not contain E; there is, furthermore, a
neighbourhood of Ewhere hIt (x) exists and is continuous; finally hIt (E) < 0,
(3) f/J(x) is continuous at x = E, f/J(E) =1= O. Then the following asymp-
totic formula holds as n -+ 00 1

j f/J(X) [I(x»)" dx N'ii(E) [I(E))"+1/2 V'- nf~~~) = f/J(E)


II
e,,"(El y' - n:~~) •
1 On the use of such integrals Laplace has this to say: ... On est souvent conduit
a des expressions qui contiennent tant de termes et de facteurs, que les substitutions
num6riques y sont impraticables. C'est ce qui a lieu dans les questions de probabilit6,
lorsque l'on considere un grand nombre d'6v6nements. Cependant il importe alors
d'avoir la valeur num6rique des formules, pour connaltre avec quelle probabilit6
les r6sultats que les 6v6nements d6veloppent en se multipliant sont indiqu6s. II
importe surtout d'avoir la loi suivant laquelle cette probabilit6 approche sans cesse
de la certitude qu'elle finirait par atteindre, si Ie nombre des 6v6nements devenait
infini. Pour y parvenir, je consid6rai que les int6grales d6finies de diff6rentielles
multipli6es par des facteurs 6lev6s a de grandes puissances, donnaient par l'int6gra-
tion, des formules compos6es d'un grand nombre de termes et de .facteurs •.•
He adds the following remark on his method of which 201 describes the first
step: ... un proc6d6 qui fait converger la s6rie avec d'autant plus de rapidit6, que
la formule qu'elle repr6sente est plus compliqu6e; en sorte qu'il est d'autant plus
exact, qu'il devient plus n6cessaire ... (Essai philosophique sur les probabilit6s'
Oeuvres, Vol. 7. Paris: Gauthier-Villars 1886, p. XXXVIII.)

yuana97@uw.edu
Pt. II. Chap. 5. No. 198-207 97
[We consider only a neighbourhood of E and expand h (x) in powers of
(x - E) up to terms of the second order.]
202. Let n be an integer, n ~ + 00. Using the fact that
,. ,.
2"
. 2.. X dX= f
f S10
2"
2..
cos x X=
d 1 • 3 ••• (2n - 1) n
2.4 ••• 2n 2'
o 0
prove that
1 • 3 ••• (2n - 1) 1
2·4···2n
N
V-'
nn

203. We assume thatA is real, A. > 1; p .. (x) denotes the n-th Legendre
polynomial. As 'n ~ 00
(A+~HI/2
P .. (A) N -1
V2nn , .
VA2 - 1

The positive value of the roots must be used. [VI 88.]


204. The Bessel function 1.(t) can be defined by Hansen's expansion

+ 2 ~ i·1. (t) cosvx •


00

ei/cOSJ& = ]0 (t)
• =1

Derive the following asymptotic formula:


I
1.(it) N i· V-~-
2m
, t~ + 00, '/I = 0, 1, 2, ...

205. Show that for positive n, n ~ + 00,


r(n + 1) = i e-"%" dx N (: r 2~n
V

and, more accurately,

(:r r(n + 1) = V~n + o(~). [18, I 167.J

208. Let k and 1 be real numbers, k > 1. Prove that for n ~ + 00


- 1)" (_k_). ,
(nk n+ ') (k¥2nn
1
N H I+ 2
k - 1

207. Assume that IX is real and that t is positive and increases to


infinity. Then

yuana97@uw.edu
98 Functions of Large Numbers

208. Let 0 < !X < 1. The following approximation is justified as


1'-+ + 0:

1 exp ( : -TX) dx N V1 ~/X T-2(1~"') -1 exp (1: /Xl' -1:",).

209. Let !X > o. As t -+ 00 we obtain

§ 2. Modifications of the Method


210. Let !X and f3 be two real constants. Then the relation

-n!1 "+,,,yK"
0
J +P B ( 1 )
e-xx" dx = A + -= + 0 -=
Vn Vn
holds where

A =./-
1
J"'-"2
edt,
t'
-
B- ./-
_1_(f3 _ /X2 +
3
2) e- ~ .
r 21l -00 r 21l
211. We denote by A a positive proper fraction and by x" the only
positive root of the transcendental equation
x, x"
1 +-+-+
X
1! 2!
... +-=AC
n!
[V 42].

As n -+ 00 the root is given by

x" = n + V; + f3 +-0(1),
!X

where !X and f3 satisfy the equations:


1 t'
/X 2 + 2
,/_ J e
00 __

2 dt = A, f3=-3-·
r 21l '"
212 (Continuation of 201). Let !X denote a real constant. Then, for
n-+oo,

213. The functions 9'(x) , h(x) and I(x) = eh(x) are defined on the finite
or infinite interval [a, bJ and satisfy the following conditions:

yuana97@uw.edu
Pt. II, Chap. 6, No. 208-216 99

(1) tp(x) [I(x})" = tp(x) e"II(%) is absolutely integrable over [a, b],
n = 0, 1,2, ...
(2) The value of the function h(x} at a point Eof (a, b) is larger than
its least upper bound in any closed interval to the left of E which does
not contain E. Moreover there is a neighbourhood of E in which h"(x)
exists and is bounded. Finally h' (E) > O.
(3) tp(x) is continuous at x = E, tp(E) =1= o.
Prove, for n -+ 00, the asymptotic formula

H cdO~+l.
j "tp(x) [I(x)]" dx ~ :~~ ePlI'(e) • n""'(~)-l • e""(~),
.
where IX and {J stand for real constants.
214. Let E denote the only real root of the transcendental equation
e1HE = 1. Then we have for n -+ 00
1 ~"+"'log"+11
I
n.
J
0
ex" dx ~ nA B ,

where IX and {J are real constants and

A=IXI+~_~
~ 2 '

215. Suppose that n is odd and let-x" denote the only real root
of the equation
x x"
1 +-+-+ ... +-=0
X 2

I! 2! n!
[V 74].

As n -+ 00 x" is asymptotically given by

x" = En + IX log n + (J + 0(1),


where E is the only real root of the equation e1HE = 1 and IX and {J are
given by
(J = 1 +~ ~log (V~ 1 +~)
2n-~-.
218. Assume that the function g(x) is monotone increasing for positive
x and that
lim g(x} =
%-++00
+ 00,
We define

yuana97@uw.edu
100 Functions of Large Numbers

If there is a positive number" such that


g(!U')
lim - -
s-++oo g(x)

exists and is a continuous function of (¥ for 1 - " :s;; (¥ < 1 +" then
• log a..
lim - - = 1.
"-+00 g(n)

The method of problem 201 to evaluate functions of large numbers can


be generalized in the following way: We have to estimate an integral of
the form

J rp(x) Mx) Mx) •• ·/.. (x) dx = J rp(x) .f


b b
1 (s)+II,(s)+ ... +II,,("l dx
a a

where the functions h1 (x), h2(X), ... , h.. (x) 3,(e positive on (a, b) and attain
their maximum at the same interior point E. Then we approximah.
h.(x) = h.(E) + i h;'(E) (x - E)2 + ... by h.(E) + 1h;'(E) (x - E)2
and the integral by

We have supposed that rp(E) :::j= 0, moreover h;(E) = 0, h:' (E) < 0 as
condition for the maximum at the point E, and
-h~(E) - h;(E) - ... - h~(E) = s. The method can be justified in
many instances and it is capable of adaptation and refinement.
217.
. 1"I ."
11m
"-+00
."
n! 2211COS"
." ."
(2ne' -.1) (2ne' - 2) (2ne' - 3) ••• (2ne' - n) I
df} = 2",.

i- and recall 59, 115.]


-11

[Put f} =

217.1. Analogy to 201 suggests sufficient conditions under which


for n~ +00
JJrp(x, y) e""(s.,,l dx dy N rp(E, 'I) e"II(E.'1l 2" 2
1ft n Y". ,/,,,,, - "sy
where the partial derivatives of second order hsz, h"", and hsy are taken
at the point (E, 'I). Give a full statement and a proof.

§ 3. Asymptotic Evaluation of Some Maxima


218. The function
Y;(x - 1) (x - 2) •.. (x - n) a-",

yuana97@uw.edu
Pt. II, Chap. 5, No. 217-223 101
where a > 1, has the maximum M .. on the interval en, + 00). It can be
approximated by
M.. 1 1
-
nl
N -
J!2; ---."",-,
(a _ 1).. +1/2'
[16.]
219. The function
x{x2 - 12) (x 2 - ~) ••• (x2 _ n2) a-x,
where a > I, has the maximum M .. on the interval (n, + 00). It can be
approximated by
M .. N~(2Va)2"+1
n!1 2n a-1' [17.]
220. We define Vi = Qo(x),
V; (1 - ;) (1 - ;) ... (1 - =) = Q..(x), n= 1, 2, 3, ...
The sequence of functions
... ,
is uniformly bounded for x > 0 if a > 2; it is not uniformly bounded if
0< a< 2.
221. We define x = Po (x),

n = I, 2, 3, ...

The sequence of functions


P 1 {x) a-x, P2 (x) a-x, ... ,
is uniformly bounded for x > 0 if a ~ 3 + V8; it is not uniformly
bounded if 0 < a < 3 + V8.
222. Assume that a > 0, 0 < # < 1 and that M .. is the maximum
of e-(x+a%l') x" in the interval (0, + 00). We find

lim -T
(M )"-1' = e- a •
""00 n.

§ 4. Minimax and Maximin


*223. The function I(x, y) is continuous in the rectangle
a <x ~ a',
Then its maximum for a given x and b < Y < b' (along a segment
parallel to the y-axis)
max I(x, y) = tp(x)
y

is a continuous function of x.

yuana97@uw.edu
102 Functions of Large Numbers

We interpret the surface


z = f(x, y)
in a rectangular coordinate system x, y, z with vertical z-axis as a topo-
graphical surface in a mountainous region. Then the curve
z = !p(x)
(in the x, z-plane) is the skyline of the range (as it appears when seen
from a faraway point of the y-axis) and the minimax
min !p(x) = min max f(x, y)
" "Y
(the minimum of the maxima) refers to the lowest point in the skyline.
*224 (continued). Show that
max min f(x, y) < min max f(x, y) .
Y" x Y
*225 (continued). Which one of the two signs, < and =, is valid in
the example
f(x, y) = 1 - (x - y + 1)2,
a = 0, a' = 2, b = 0, b' = 4?
*226. Add to the assumptions of 223 that f(x, y) > O. Then
]

lim
"400
[r (I
a b
[f(x, y)]" dy)-l dX]
"= min max f(x, y) .
" Y

yuana97@uw.edu
Part Three

Functions of One Complex Variable

General Part

Chapter 1

Complex Numbers and Number Sequences

§ 1. Regions and Curves. Working with Complex Variables


The complex variable z is written in the form
z= x + iy = "e'f (x, y, ", {} real, " ~ 0, {} taken mod 2..,;).
We call
x = 8lz the real part of z, y = 3z the imaginary part of z,
l' = Izl the absolute value of z (also modulus),
{} = arg z the argument or amplitude of z.
z
The number = x - iy = "e- i6 is the conjugate of z.
1. The number z + z is real, z - z is purely imaginary, zz is real
and not negative.
2. What sets of points in the z-plane are characterized by the con-
ditions:
8lz> 0; 8lz~ 0; a < 3z < b; IX~ arg Z::;, {J; 8lz = 0;
1 1
IZ-Zol=R; Iz-zol<R; Iz-zol::;'R; R::;,lzl::;'R'; 8l z = R

(a, b, IX, {J, R, R' real, Zo complex, a < b, IX < (J < IX+ 2..,;, 0 < R < R') 1
3. What sets of points in the z-plane are characterized by the condi-
tions
Iz - al + Iz - bl = k; Iz - al + Iz - bl::;, k, k > 01

yuana97@uw.edu
104 Complex Numbers and Number Sequences

4. What open set of the z-plane is characterized by the condition


IZ2 + az + bl < R2?
For what values of R is this set connected, for what values of R is it not
connected?
5. Assume Ia I < 1. For any point of the complex z-plane

I1z -- iiza I
is either < 1, or = 1, or > 1, and so the whole plane is divided into
three subsets. Describe them.
6. Suppose t)la > O. For any point of the z-plane

I; ~ ;1
is either < 1, or = 1, or > 1, and so the whole plane is divided into three
subsets. Describe them.
7. Let (x, {J be real; a complex; (x, {J and a are fixed. Suppose that the
complex variables Zl and z, satisfy the relation
IXZ;Zl + aZ1Z2 + aZ1Z2 + {JZ2Z2 = O.
If (X{J - aii < 0 the points 2. lie on a circle, possibly a line segment. (The
Z2
left hand side of the equation is called a Hermitian form of the variables
Zl and Z2')
8. Let a and b be positive constants and the real variable t signify
time. Describe the curves given by the three equations

Zl = . + a,t
~a Z2 = -~ 'b e-it ,Z = .
~a + at - 1'b-
e it .

9. Describe the motion of the point


... +6,
z = (a + b)eil - be b •

where a, b are positive constants and t denotes time.


10. Let the radius vector r and the argument f) be functions of the
time. The complex function Z = re'/) of the real variable t is represented
by the motion of a point in a plane. Compute the components of velocity
and of acceleration parallel and perpendicular to the radius vector.
[Differentiate Z twice with respect to t.]
z"
11. For what values of Z is the absolute value of the n-th term I"
n.
of
z z2 z"
1+-+-+"'+-+'"
1! 21 nl

yuana97@uw.edu
Pt. III, Chap. 1, No. 4-15 106
(exponential series in the complex plane) larger than the absolute value
of any other term? n = 0, 1, 2, ...
12. For what values of z is the absolute value of the n-th term of
the series
1 + ~ + z(z - 1) + z(z - 1) (z - 2) + ... +
1 1·2 1·2·3

+ 1) ~ (z)
+ z(z - 1) .. , (z - n
1.2 .. ·n +"'=kJ
,,=0 n
(binomial series for (1 + t)' with t = 1 and complex z) larger than the
absolute value of any other term of this series? n = 0, 1, 2, ...
13. We put

Po(z) = z, P,,(z) = z(1- ~:) (1 - ;:) (1 - ~) ... (1 - ::),


1J = 1, 2, 3, ...
For what values of z is IP,,(z) I larger than IPo (z) I, IPI (z) I, ... , IP"-1 (z) I,
IP"+1 (z) I, ... ? (P,,(z) is the n-th partial product in the product expansion
of sin nz.)
n
14. We assume that the real functions I(t) and qJ(t) are defined on the
interval a s::: t < b, that I(t) is positive and continuous and qJ(t) properly
integrable. Then

Ij I(t) ei<p(I) dt I s::: j I(t) dt.


Equality holds if and only if the function qJ(t) assumes the same value
mod 2:17: at all its points of continuity.
15. Let qJ(t) be defined for t > 0 and be properly integrable over any
finite interval. If

f f
00 00

e-(I+i<p(l»)dt = P, e- 2(1+i<p(I») dt = Q,
o o
then
14P2 - 2QI < 3.
Equality holds if and only if qJ(t) assumes the same value mod 2:17: at
all its points of continuity.

§ 2. Location of the Roots of Algebraic Equations


We consider polynomials of degree n

P(z) = aoz" + a1Z"-1 + a2Z.. -2 + ... + a" -z1+,a,

yuana97@uw.edu
106 Complex Numbers and Number Sequences

with arbitrary coefficients; ao =f= 0 is often assumed. The complex number


Zois called a zero of this polynomial if
aoz..o + atzo.. -1 + a2z0.. -2 + ••• + a.. _ 1 Zo + a" = 0.
(zo is a root of the algebraic equation P(z) = 0.) If ZI' Z2' ... , z.. are the n
zeros of the polynomial P(z) we can write
P(z) = ao{z - ZI) (z - Z2) •.. (z - Z,.)
as is proved in algebra.
16. A polynomial of the form
z.. - PIZ,.-t - P2Z,.-2 - ... - P.. _IZ - P,.,
where PI >- 0, P2 > 0, ... , P,. > 0, PI + P2 + ... + Pn > 0, has just one
positive zero.
17. If Zo is a zero of the polynomial
z.. + alz"-l + a2z.. - 2 + ... + a..
then IZo I is not larger than the only positive zero Cof the polynomial
z.. - lall Z"-l - la2 1 Z"-2 - ... - I a.. ,.
18. Assume a" =f= O. The absolute value of none of the zeros of the
polynomial
P(Z) = z.. + a1z"-t.. + a2z.. - 2 + ... + a"
is smaller than the only positive zero Cof the polynomial
z" + lall Z,,-l + la2 1z"-2 + ... + la.. -ll Z -la"l.
19. All the zeros of the polynomial z" + c are on the circle centred at
Z = 0 with radius I c 11 /".
*20. Let c1 ' c2 ' ••• , cn be positive numbers and
c1 + C2 + ... + c" < 1.
The absolute values of the zeros of the polynomial
z" + a1z.. - 1 + a2z"-2 + '" + a"
are not larger than
-
M -max (JaIl ,
C1
VJa21
C2
, ... , VJa~T) .
c"
*21. The absolute values of the roots of the equation
z" + a1z"-1 + a2z"-2 + .. , + a.. = 0
are not larger than the largest among the numbers

Vn Ia. I;
3
Vn la 1, 2 ~, ••• J

yuana97@uw.edu
Pt. III, Chap. 1, No. 16-25 107
also they are not larger than the largest of the numbers

k = 1. 2. 3••..• n;

and they are certainly smaller than the largest of the numbers

k = 1.2•...• n.
22. Assume
Po > PI > P2 .•. > P,. > O.
The polynomial
Po + P1z + P2 Z2 + '" + p,.z"
cannot have a zero in the unit disc 1z 1 ::;; 1.
23. Suppose that all the coefficients Pl' P2' ...• P,. of the polynomial
Poz" + P1z"-1 + ... + P,._lZ + P,.
are positive. Then the zeros of this polynomial lie in the annulus
IX < 1Z 1 < {3. where IX is the smallest. {3 the largest among the values
P,.
... ,
P,.-l·
24. Let ao• at. a2 • •••• a,. be digits (ill. the ordinary decimal notation.
that is integers between 0 and 9 inclusively) n > 1. a,.~ 1. Then the
zeros of the polynomial
ao + a1z + a2z2 + ... + a,.z"
are either in the open left half-plane or in the open disk

Izl<1+;:'.
The best upper bound that may replace the last number is between 3
and 4.
25. We assume that all the zeros of the polynomial
P(z) = + a1z,.-1 + ... + a,._lz + a..
aoz"

are in the upper half-plane 3z > O. Let a. = IX. + i{3.. IX.. (3. real.
11= O. 1. 2•...• n. and
U(z) = IXoz" + IX1z"-1 + ... + IX.. _1Z + IX,..
V(z) = {3oz" + {31z"-1 + .. , + {3.. -IZ + (3,..
Then the polynomials U(z) and V(z) have only real zeros.

yuana97@uw.edu
108 Complex Numbers and Number Sequences

28. Let P(z) = 0 stand for an algebraic equation of degree n all


the zeros of which are in the unit circle Iz \ < 1. Replacing each coefficient
of P{z) by its conjugate we obtain the polynomial P(z). We define
P*(z) = z" P(Z-l). The roots of the equation P(z) + P*(z) = 0 are all
on the unit circle \z\ = 1.
27. Suppose that the polynomial P(z) of degree n, n ;;;::: 2, assumes
the values tx and P for z = a and z = b, respectively, where a =t= b and
tx =t= p. Let <t denote the closed domain bounded by two arcs of circle
the boundary whereof is the set of those points at which the line
segment a, b subtends the angle !!... n
Show that to each point" on
the line connecting tx and p there exists a point z in <t such that" = P(z).

§ 3. Zeros of Polynomials, Continued. A Theorem of Gauss


28. If all the complex numbers zl' Z2' ••• , z" (considered as points in
the complex plane) are on the same side of a straight line passing through
the origin, then
ZI + Z2 + ... + z" =t= 0, ~+~
81 82
+ ... + ~=t=0.
8"

29. Suppose z]' Z2' ••• , z" are arbitrary complex numbers that add
up to zero. Any straight line t through the origin separates the nUJ1lbers
ZI' Z21 ••• , z" so that there are some z:s on each side of t unless all the z:s
lie on t itself.
30. Let zl' Z2' ••• , z" be arbitrary points of the complex plane,
m 1 > 0, m 2 > 0, ... , m" > 0, m 1 + m 2 + ....+ m" = 1 and
Z = m]z1 + m2z2 + ... + m"z".
Then there are points z. on both sides of any straight line through z
except when all the z;s lie on that straight line.
We can interpret the numbers m 1, m 2 , ••• , m" as masses fixed at the
points Z1' Z2' ••• , z". Then the point z defined in 30 is the center of gravity
of this mass distribution. If we consider all such mass distributions at
the points Z1' Z2' ••• , z" the corresponding centers of gravity cover the
interior of a convex polygon, the smallest one containing the points
ZI' Z2' ••• , z". The only exception arises when all the points are on a
straight line. Then the centers of gravity fill out the interior of the smal-
lest line segment that contains all the points Z1' Z2' ... , z".
31. The derivative P'(z) of P(z) cannot have any zeros outside the
smallest convex polygon that contains all the zeros of P(z) (considered
as points in the complex plane). Those zeros of P'(z) that are not zeros

yuana97@uw.edu
Pt. III; Chap. 1. No. 26-36 109
of P(z) lie in the interior of the smallest convex polygon (the smallest
line segment) that contains the zeros of P(z).
32. Let z1' Z2' " ' , z,. be arbitrary complex numbers, z,. 9= z. for all
/-' 9= v, /-', v = 1. 2•...• n. We consider all the polynomials P(z) that
vanish only at the points z1' Z2' ••• ,z,. (having there zeros of arbitrary
order). The set of the zeros of the derivatives P'(z) of all these polynomials
is everywhere dense in the smallest convex polygon that contains
z1' Z2' ... , z,..
33. Let P(z) denote a polynomial. The zeros of cP'(z) - P(z), c 9= 0
lie in the smallest (infinite) convex polygon that contains the rays
parallel to the vector c starting from the zeros of P(z).
A zero of cP' (z) - P(z) appears on the boundary of this region only
in one of the following two cases: (a) the zero in question is also a zero
of P(z); (b) the region in question degenerates into a ray.
34. Let el' e2' ... , ep be positive numbers, a1' a2 , ... , a p arbitrary
complex numbers. and let the polynomials A (z) and B(z) of degree p
and p - 1 resp. be related by
B(z)
A (z)
=~ +~
z - al z - a2
+ ... +..-!L...
z- a p
Suppose that the polynomial P(z) is a divisor of
A(z) P"(z) + 2B(z) P'{z), i.e.
A(z) P"(z) + 2B(z) P'(z) = C(z) P(z),
where C(z) denotes a polynomial. Then the zeros of P(z) lie in the smallest
convex polygon that contains the numbers aI' a2 • ••• , ap '
36. If a polynomial f(z) whose coefficients are all real has only-real
zeros then this is true also for its derivative f'(z). If f(z) has complex
zeros then they appear in pairs. the two zeros forming a pair are mir-
ror images to each other with respect to the real axis; they are complex
conjugates. We draw all those disks the "vertical" diameters of which
are the line segments connecting the conjugate zeros of such pairs. If
I' (z) has any complex zeros they lie in these disks. [Examine the
imaginary part of j(~; .]
§ 4. Sequences of Complex Numbers
38. Assume that the numbers z1' Z2' ••• , z,., ... are all in the sector
-(X ~ arg z ~ < ~ . Then the series
(X. (X

Zl -+ Z2 + ... + z,. + ... and 1zll + 1z21 + ... + 1z.. 1+ ...


are either both convergent or both divergent.

yuana97@uw.edu
110 Complex Numbers and Number Sequences

37. Suppose that the numbers Zl' Z2' .•• , z.. ' ... are all in the half-
plane 9lz > 0 and that the two series
+ Z2 + ... + z.. + ... and zi + ~ + ... + z! + ...
ZI

converge. Then IZI 12 + IZ21 2 + ... + IZ" 12 + ... converges too.


38. There exist complex sequences ZI' Z2' ••• , z.. ' ... for which all the
series
~+~+"'+z!+"" k = 1, 2, 3, ...
converge and all the series
IZI Ik Iz21 k+ + ... +
Iz.. 11I + "',
k = 1, 2, 3, ...
diverge.
39. Let ZI' Z2' ••• , z.. ' ... be arbitrary complex numbers. If there
exists a positive distance lJ such that Iz/ - zk I ; : : lJ for I < k,
t, k = 1, 2, 3, ... the convergence exponent of the sequence
IZ11, IZ21, IZ31, .. · is at most 2. [1114.]
40. The limit points of the complex numbers
1'''' + 2'''' + 3'''' + .. , + n'''' , lX real, lX
~
<::::: 0, n = 1, 2, 3, ...
n
fill out the entire circle with radius (1 + lX2)-1/2 and center at the origin.
[The expression in question is closely related to a sum of rectangles.]
41. Find the locus of the limit points of the complex sequence
ZI' Z2' ••• , z,,' ... , where

z" = (1 + :) (1 + ;) (1 + ~) ... (1 + ~).


42. Put
(1 + V~) (1 + V~ ) ... (1 + :; ) = z"
and connect the points Z,,_1 and z" by a straight line. The distance
between these two points is always 1. The polygonal line connecting the
successive points approaches with increasing n an Archimedean spiral;
if ~" = r"e"''', r" > 0, 0 < 'P" - 'P.. - l < ~, then
r - r,,-1 1
lim " =-, lim~=~.
"-+00 Vi" - Vi"-1 2 "-+00 Vi" 2
it

43. Let t be a fixed real number and put z = 2neY;;- . Then

"-+00
·
I1m y-;;
-
n z(z - 1)(z -
2'n!
2) ... (z - n)
=e -/' .

[II 59; II 10 slightly modified.]

yuana97@uw.edu
Pt. III, Chap. 1, No. 37-46 111
§ 5. Sequences of Complex Numbers, Continued:
Transformation of Sequences
By means of the infinite triangular array

we transform an arbitrary infinite sequence zO' Zt •...• Z" • ••• into a new
sequence WOo wI' w2 • •••• w... ...
n = 0.1.2 •...
The triangular array is called convergence preserving if it transforms every
convergent sequence zO' Zt. Z2' •••• z" • ••• into a convergent sequence
WOo wI' w2 • •••• w" • ... (Cf. I. Chap. 2.) The array is convergence preserving
if it fulfills the following condition. consisting of two parts:
(1) lim a". = a. exists for all fixed v;
11-+ co

(2) with the notation


"
Ila".. I=C" •
• =0
the sequence at. 0'2' •••• a". ... is convergent and the sequence
0'0'
Co. C1' C2 • •••• C" •... is bounded. [0. Toeplitz: Prace mat.-fiz. Vol. 22.
pp. 113-119 (1911); H. Steinhaus: Prace mat.-fiz. Vol. 22. pp. 121-134
(1911); T. Kojima: T6hoku Math. J. Vol. 12. pp.291-326 (1917);
I. Schur: J. reine angew. Math. Vol. 101. pp. 79-111 (1921).J
44. Prove the easier part of the above mentioned proposition: If
the conditions (1) and (2) are satisfied the array preserves convergence.
[I88.I80.J
45. What conditions must the series "0 +"t + "2 + ... + + ... "II
satisfy in order that its Cauchy product [I 34. II 23. VIII. Chap. 1.
§ 5J with any convergent series Vo + vt + v2 + ... + v" + ... results
in a convergent series
+ (UOvt + utvo) + (uOV2 + UtVt + u 2vO) + .. .
"OVo

+ (UOVII + "Jv.. _ t + ... + U,,_IVl + uIIVO) + ... ?


48. What condition must the series u 1 + u 2 + ... + UII + ... satisfy
in order that its Dirichlet product [VIII. Chap. 1. § 5J with any conver-

yuana97@uw.edu
112 Complex Numbers and Number Sequences

gent series vt + V2 + ... + v.. + ... results in a convergent series


UtV1 + (U 1V2 + U 2Vt ) + (UtVS + USVt ) + ... + ~ U,V .. + ... ?
,/.. --,
47. The sequence of factors
"0' "1' "2' ... , " .. '
turns any convergent series ao + at + a2 + ... + a.. + ... into a con-
vergent series
"oao + a1 "t + "2a2 + ... + ""a" + ...
if and only if the series

converges.
48. The existence of
lim (u 1
"-+00
+ u2 + ... + U,,_t + CUll) =I¥
implies the existence of
lim (ut
"-+00
+ u 2 + ... + u.. _t + u..) = I¥

in two cases only: if c = 0 or if me > !, but not if me ~ !, c =1= o.


49. Let u o' u 1 ' u 2 ' ••• , U.. ' ••• be arbitrary complex numbers. For
what values of c does the existence of
. ( Uo + u t + ... + U,,)
lim u"
"-+00.
+c 11 +1
imply the existence of lim U .. ?
"-+00
SO. If the Dirichlet series
+ a22- s + as3- s + ... + a..n- s + ...
a1 1-s
converges for s = + iT, T real, > 0, then
(I (I, (I

lim {l - W (a t + a t2 + a t S + ... + a"t" + ...) = o. [192.]


'-+1-0 t 2 S

§ 6. Rearrangement of Infinite Series


51. If every subseries of a series with complex terms converges the
series converges absolutely.
52. Assume that the series 1zl 1+ 1z21 + ... + 1z.. 1+ ... diverges.
Then there exists a direction 01 accumulation, that is a real number I¥

yuana97@uw.edu
Pt. III, Chap. 1, No. 47-54 113

such that those t~rms of the series Zl. + Z2 + ... + z,. + ... that are
contained in the sector €X - e < arg z < €X + e constitute an absolutely
divergent sub series for any e > o.
53. If lim z,. = 0 and if the positive real axis is a direction of accu-
+ +
"~oo
mulation of the conditionally convergent series Zl Z2 Z3 ., .•••

then there exists a subseries Z" + z'. + z'. + ... the real part of which
diverges to + 00 and the imaginary part converges to a finite number.
54. If the series Zl + Z2 + Z3 + ... is convergent, but not absolutely
convergent, any value represented by a point of a certain straight line
can be obtained as the sum of the series rearranged in a suitable order.
[Consider two complementary subseries shifted relatively to each other;
52, 53, I 133, I 134.]

Chapter 2

Mappings and Vector Fields

If we associate each point Z of some domain ~ of the z-plane with a


certain complex value w according to a given law then w is called a
function of z. Two geometrical interpretations of the functional relation
are particularly useful. One uses one plane, the other two planes. The
value w belonging to the point z (or, if more expedient, w) can be thought
of as a vector acting on the point z; in this way a vector field is defined in
the domain ~. In the other interpretation, the value w associated with
the point z in the z-plane is conceived as a point in another complex
plane (w-plane). In this way the domain ~ is mapped onto a certain point
set of the w-plane.

§ 1. The Cauchy-Riemann Differential Equations


Let u(x, y) and v(x, y) be two real functions of the two real variables
x and y. Then w = u + iv is a function of the variable z = x + iy. The
function w = u + iv of z = x + iy is called analytic in a certain open
region if u and v are continuous as well as their first partial derivatives
and satisfy the Cauchy-Riemann differential equations

yuana97@uw.edu
114 Mappings and Vector Fields

Observe the combination

axa (U +.) 1 a( .)
W =-;- 8y U + w =
dftl
dz •

55. Are the functions


z, Z2., Izl, i
analytic ?
55.1. Assume that fez) is analytic, use the notation
w = u + iv =f(z) = f(x + iy)

as above and use subscripts to denote partial derivatives in the usual


way. Verify that
2
Uz + Vz2 = 2
U" + v"2 = Uz2 + u"2 = Vz2 + v"2 = UzV" - U"Vz = Idi
dw 12
.
55.2 (continued). Prove that

55.3 (continued). Let tp(x, y) and 'I'(x, y} denote functions of the two
real variables x and y having continuous derivatives of first order;. they
can also be considered as functions of U and v where I'(z) =1= O. Verify that

tpz'l'z + tp"'I',, = (tp..'I'.. + tpv'l'v) I


dW l2
dz I.
55A (continued). If tp(x, y) has continuous derivatives of the second
order, also
tpn + tp"" = (tp.... + tpvv) I!; 12 .
55.5 (continued). Assume that a, b, c and d are real constants,
ad-bc=l,
and consider

Then
2 2
lP..
2
+ lPv
2=
lP.... + lPw
=
".
•.
lPz + lP" lPn + lP"" IJ

56. Find the analytic function of z that vanishes for z = 0 and has
the real part
U = 1 + 2X2 _ 2". + (x~ + ,,1)2 •
57. We denote by a and b, a < b, two fixed real numbers, by z a
variable point in the half-plane 3z > 0 and by w the variable angle

yuana97@uw.edu
Pt. III, Chap. 2, No. 55-61 115

under which the interval [a, b] is seen from the point z. If possible find
an analytic function the real part of which is w.
58. Show that for any analytic function f(z) = f(x + iy)

(a~B + ~:) If(x + iy) 12 = + iy) 12 •


411'(x
59. Show that for any analytic function of z = x + iy
2 I . 12 _ 4\/,(x+ iy) \2
( a2
ax. + IJyB
a )
log (1 + f(x + ~y) ) - (1 + \/(x + iy) \2)2 •
§ 2. Some Particular Elementary Mappings
The Cauchy-Riemann differential equations express the fact that an
analytic function brings about a conformal mapping of the z-plane into
the w-plane. (Preservation of the angles including sense.)
The import of the Cauchy-Riemann differential equations for vector-
fields will be discussed later. Cf. § 3.
80. We consider an orthogonal coordinate system ~, 'YJ, C in three
dimensional space. An arbitrary point (~, 'YJ, C) of the sphere
~2 + 'YJ2 + C2 = 1 (unit sphere) is projected from the point (0,0, 1)
(north-pole of the sphere) into the plane C= 0 (equatorial plane). Let the
projection be (x, y, 0). Express x + iy in terms of ~, 'YJ, C and ~, 'YJ, C in
terms of x and y. (Stereographic projection.)
81 (continued). Let the point P on the plane C= 0 be the stereo-
graphic projection of the point P' on the unit sphere. A rotation through
the angle :n; of the unit sphere around the ~-axis moves the point P' to the
point P". This point P" is then projected stereographically into the
point P'" of the ~, 'YJ-plane. Let the point P have the coordinates x, y, 0
and the point P'" the coordinates u, v, O. Express u + iv in terms of
x +iy.
We introduce on the unit sphere the geographic coordinates 0 and cp
(longitude and latitude) whereby

-:n; < 0< :n;,

The sphere is described by


~ = cos cp cos 0, 'YJ = cos cp sin 0, C = sin cpo
We consider now the circular cylinder tangent to the unit sphere
~2 + 'YJ2 + C2 =
1 along the equator (great circle in the plane C = 0).
Imagine a system of coordinates (u, v) on the cylinder that becomes a
cartesian system when the cylinder is unrolled. Let the point u = 0,

yuana97@uw.edu
116 Mappings and Vector Fields

v = 0 coincide with (I, 0, 0), the positive u-axis be a generatrix pointing


upwards and let the v-axis on the cylinder coincide with the equator.
The values of v varie on the cylinder in the same sense as 0 from -n to n.
The points obtained in this way fill out an infinite strip of width 2n with
the u-axis as center-line when the cylinder is unrolled.
Mercator's projection establishes a conformal one to one correspon-
dence between the unit sphere and the u, v-cylinder (unrolled into a strip).
The point on the cylinder corresponding to the point qJ, 0 on the sphere
has the coordinates
1l = log tan (~ + :), v = O.

62. Into which lines does the Mercator projection transform the
meridians and parallel circles? What are their images under the stereo-
graphic projection?
63 (continued). The point P of the unit sphere is stereographically
projected onto the point (x, y, 0) of the plane z = 0 and the image of P
under Mercator's projection is u, v on the cylinder. Express x + iy in
terms of u + iv.
64. Suppose z = eW • To what curves in the z-plane do the two fami-
lies of lines 9iw = const. and 3w = const. in the w-plane (which are
orthogonal to each other) correspond?
65. Along which curves of the z-plane is the real part of Z2 constant?
Along which curves is the imaginary part constant? The two families of
curves form an orthogonal system; why?
66. Which curves in the z-plane are transformed by w = Vi into
the lines 9iw = const. in the w-plane ? Same question for 3w = const.
67. The mapping ~ = cos z transforms the line 9iz = const.. of the
z~plane into hyperbolas, the lines 3z= const. into ellipses of the w-plane.
68. Consider the function z = w+ew , z = x + iy, w = u + iv. Find
the equations of the curves in the x, y-plane that are mapped onto the
lines u = const. and v = const. respectively. What corresponds to the
lines v = 0, v = n?
69. Given the function w = e' compute the area of the image of the
square a - e < x < a + e, -e < y < e, 0 < e < n, Z = x + iy. Find
the ratio of the two areas and its limit as e converges to O.
We define the linear enlargement (enlargement ratio, stretching, change
of scale) of the mapping w = /(z) at a point z where /(z) is regular to be
the ratio of the length of the line element at the point w = I(z) (in the
w-plane) to the length of the line element at the point z (in the z-plane).

yuana97@uw.edu
Pt. III, Chap. 2, No. 62-73 117

This ratio is equal to 1t' (z) I. We define the area enlargement to be the anal-
ogous ratio of the area elements; it is equal to It'(Z) 12. A curveL in the
z-plane is therefore transformed into a curve in the w-plane of length

L
J It'(z) Ildzl·
An area A in the z = x + iy-plane is mapped onto an area in the w-plane
of size
JJ 1t' (z) 12 dx dy.
A

The change in direction of the line element under the mapping w = I(z)
is .equal to arg I.' (z). It is called the rotation at the point z and it is deter-
mined up to a multiple of 2n at any point where t' (z) ::!= O. The branch
for which -n < arg f'(z) ::;: n is usually adopted.
70. The function w = cos z, z = x + iy, yields a one to one con-
formal mapping of the rectangle
O< Xl
<
= X
<
= X2 < "2
n
'
onto a domain bounded by parts of confocal ellipses and hyperbolas
[67J. Compute the area of this domain.
71. Consider the function w = Z2. What is the locus of those points
at which the linear enlargement equals some given constant? Analogous
question for the rotation.
72. Let a be an increasing positive parameter. Determine the region
onto which the function w = e' , z = x + iy, maps the variable square
-a < x < a, -a < y < a. Up tc what value of a is this region covered
only once? For what values of a is the image covered exactly n times?
73. We examine the image of the closed disk 1 z 1 < r under the func-
tion w = eZ • Suppose r is continuously increasing. There is on the ray
arg w = IX a point that is covered by the image of the disk growing with
r at least as often as any other point of that ray for all values of r. Where
is this point?
The regular function w = I(z) is called schlicht (or univalent) in the
region ffi if it does not assume in ffi any value more than once. E.g. the
function I(z) = Z2 is schlicht in the upper half-plane 3z > 0; the function
V:Z is schlicht in the z-plane cut open along the positive real axis; the
function eZ is schlicht in the horizontal strip -n < 3z ::;: n but not in
any wider horizontal strip [72] etc. A schlicht function w = I(z) estab-
lishes a conformal one to one correspondence between the region ffi and
a region 6 of the w-plane. Very often it is useful to consider the point

yuana97@uw.edu
118 Mappings and Vector Fields

at infinity as an ordinary point. In fact, stereographic projection onto


the sphere maps the point at infinity onto a point that plays no special
role on the sphere. If I(z) is schlicht in the region 91 the derivative I' (z)
does not vanish in 91. The converse is not true [72].
74. The function w = Z2 + 2z + 3 is schlicht in the open disk
Izl < 1.
75. The function w = Z2 is schlicht in the upper half-plane ,3z > 0
and maps it onto the w-plane cut along the non-negative real axis.
76. Let IX be real and 1a 1 < 1. The function
W =e'IX z - -a
. ___
1 - az
maps the unit disk Iz 1 ~ 1 univalently onto itself. [5.] What is the locus
of those points at which the linear enlargement equals some given con-
stant?
77. Assume that C is a circle inside the unit disk. Then there exists
a transformation of the unit disk onto itself of the type

that maps the circle C onto a circle centred at the origin.


78. Find a function that transforms the upper half-plane ,3z > 0
onto the disk 1wi < 1 so that z = i is transformed into w = O.
79. The function w = ! (z + !) is schlicht in the open unit disk
and maps it onto the w-plane cut open along the real line segment
-1 < w < 1. What curves correspond to the rays starting from the
origin? What is the image of the unit circle?
SO. Find a function that maps the annulus 0 < '1 < Izl <'2 onto
the area bounded by the two confocal ellipses
Iw - 21 + 1w + 21 = 4a1 • 1w - 21 + 1w + 21 = 4a2• 1 < a 1 < a2 •
[The result of 79 can be used if the given constants satisfy the relation
a1 - Va~ - 1 a2 - Va~ - 1

(the roots are positive).]


81. Transform the upper half of the unit disk 1z 1 < 1. ,3z > 0 onto
the upper half-plane [79]. At which points of the z-plane is the linear
enlargement equal to ~ ? At which points is the rotation ±; ?
82. Map the upper half of the unit disk Izl < 1. ,3z> 0 onto the
w-plane cut along the non-negative real axis in such a way that

yuana97@uw.edu
Pt. III, Chap. 2, No. 74-84 119

z = 0 corresponds to w = o. z = 1 to w = 1. z = i to w + 00. Which


point corresponds to z = -1 ?
83. Let 0 ::s;; eX < {J < 2n. The function
2",
w = (e- i «z)I1-«
maps the sector eX < arg z < {J onto the w-plane cut open along the
non-negative real axis.
84. Let 0 ::s;; eX < {J < 2n. Map the circular sector
eX<argz<{J. Izl<l
onto the unit disk 1w 1 < 1.

§ 3. Vector Fields
We use a special notation (slightly different from the one used in
other parts of the chapter) to discuss vector fields that are defined by
analytic functions of a complex variable. The independent variable is
denoted by

X. Y. r. {} real. r > O. Let


f = f(z) = rp + itp = rp(x. y) + itp(x. y)
be an analytic function of z; rp. 1jJ real. We write
df .
dz = w= u - ~V.

u. v real; w = u - iv is again an analytic function. Its Cauchy-Riemann


equations, obtained by separating the real and imaginary parts of
8
8x (u -
.)
~v = T1 Oyc. (u .
- w),
are
8u 8v CII 8v
8y=8x' 8:r -8y'
We assign the vector tv = u + iv to the point z = x + iy. In this way
we obtain a 'vector field in the horizontal z-plane. This, in turn, defines
a three-dimensional vector field. It is obtained by assigning to a given
point in space with vertical projection z = x + iy the same vector tv
that is assigned to the point z. In this way all points along the same
perpendicular to the z-plane play the same role.
This vector field is irrotational: the first Cauchy-Riemann differential
equation

yuana97@uw.edu
120 Mappings and Vector Fields

expresses the fact that the rotation vanishes. (It is obvious that the other
two components of the rotation vanish.) Furthermore the vector field
is a solenoidal vector field: the second Cauchy-Riemann differential
equation
OU + Ov· = 0
ox &y
shows that the divergence vanishes. (The third term usually appearing
in the expression for the divergence is obviously 0.)
85. Show that
orp
v=-.
ey
(<p(X, y) is the potential of the vector field; the curves <p(x, y) = const.
are the level lines.) Furthermore
u=Ctp
8y'
v=-~. Uh

(V'(x, y) is the conjugate potential or stream function or stream potential;


the lines V'(x, y) = const. are the stream lines or the lines of force, depend-
ing on the physical interpretation of the vector 70.)
86. The lines <p(x, y) = const. and the lines V'(x, y) = const. are
orthogonal to each other.
87. We have
(Laplace's equation).

88. Connect two points zl = Xl + iYI and Z2 = x2 + iY2 in the vector


field by a curve L whose line element forms the angle l' with the positive
x-axis. Prove that
J (u cos + v sin 1') ds =
L
l' <P(X2' Y2) - <p(xl> YI)'

i.e. the line integral of the tangentIal component of 70 is equal to the


potential difference (work).
89. Using the same notation as in 88 establish the relation
J(u sin
L
l' - v cos 1') ds = V'(x 2, Y2) - V'(x1 , YI)'

i.e. the line integral of the normal component of 70 is equal to the change
of the stream function (flux of force). (The normal to the curve L is
pointing to the right as one moves from Zl to Z2:
- i (cos l' + i sin 1') = sin l' - i cos 1'.)
The vector field generated by an analytic function can be interpreted
as an electrostatic, magnetostatic or gravitational field. The vector

yuana97@uw.edu
Pt. III, Chap. 2, No. 85-92 121
field may also be interpreted as the field of a steady flow of heat or
electricity. In this case the vector w is the gradient and as such it is
proportional to the intensity of the flow. Finally the vector field can be
also thought of as the field of the irrotational steady flow of an in-
compressible fluid.
90. The steady flow of an incompressible fluid of constant density (!
and variable pressure p, subject to no forces, and moving parallel to the
complex plane is described by the equations I
uou+vau+~op=O u~+v(lv+~op=O
ox ox ' ex oy
+:
Oy g f! Oy ,

~: = O.
If w = u - iv is an analytic function the components of the vector
w= u +iv and
P = Po - : (uS + VI) (Bernoulli's equation)
(Po constant) satisfy these equations.
91. The function
1
w=-
z
determines a vector field. Find tbe direction and absolute value of w
at the point z = re,6, the potential, the conjugate potential, the level
lines and the stream lines. (That part of the vector field that lies in the
annulus 0 < rl < Iz I < rg can be considered as the electrostatic field
between two condenser plates of a Leyden jar, or as the field of the
heat flow in the chimney of a factory.)
92. Let qJ denote the potential and "P the conjugate potential of the
vector field described in 91. Suppose that Zl and Zg are arbitrary points
on the circles Iz I = r l and Iz 1= r g, respecth'ely, and that qJl and qJ2 are
the values of the potential at zl and Z2' Find
qJ2 - qJI

(difference of the potentials between the two condenser plates of the


Leyden jar).
The conjugate potential"P in 91 turns out to be infinitely multivalued.
Study the change in value along an arbitrary closed curve L without
double points that contains the circle IZ I = rl and lies in the annulus
rl < Izi < r2• For a given point z of L let "P be the value of the stream

1 Cf. e.g. A. Sommerfeld: Mechanics of Deformable Bodies. New York: Acade-


mic Press 1950, p. 86,

yuana97@uw.edu
122 Mappings and Vector Fields

potential in z before. and "P' be the value after. describing L exactly once
in the positive sense. Evaluate
"P' -"P.

(Flow of force passing from one condenser plate of the Leyden jar to the
other.) Furthermore find

(~apacity of the cylindrical condenser plate per unit of the generatrix.)


93. Discuss the same questions as in 91 for the vector field determined
by
i
W=--.
z

(Stationary magnetic field of force generated by an infinite straight


conductor perpendicular to the z-plane.) Are the potential ({J and the
conjugate potential"P uniquely determined in this field?
94. Two infinite straight conductors are perpendicular to the z-plane
piercing it at the points z = -1 and z = 1. They carry cuuents of the
same intensity but in opposite directions. Determine the stream lines and
the level lines in the magnetic field so generated.
95. There are n infinite straight conductors perpendicular to the
z-plane piercing it at the points Zl' Z2' •••• z,.. They carry currents of the
same direction. There exist at most n - 1 points in the z-plane where
the generated magnetic force vanishes (points of equilibrium); these
points lie in the smallest convex polygon that contains the points
zl' Z2' •••• z,.. [When all the vectors of the field are rotated through 90°
the last statement is an evident consequence of the mechanical inter-
pretation of the field.]
96. Two confocal ellipses are given with foci at z = - 2. z = 2 and
semi-axes 2al' 2b1 and 2az• 2bz

a~ - b~ = a~ - b~ = 1.

Find an irrotational and solenoidal vector field in the region bounded


by the two ellipses such that these ellipses are level lines. (Electrostatic
field in a condenser whose plates are confocal elliptic cylinders.) What
do the stream lines and level lines look like? Find the capacity [92].
[Map the region between the ellipses onto the region bounded by two
concentric circles. Cf. 80 and 91.]

yuana97@uw.edu
Pt. III, Chap. 2, No. 93-99 123
97. Suppose 0 < a < b, (X < {J < (X + 231;. In the domain defined
by the inequalities
a~lzl~b, (X<argz~{J

determine an irrotational solenoidal vector field for which the bounding


circular arcs are stream lines and the bounding straight segments are
level lines (electric current in a plate of constant thickness).
We call1p1 and 1p2 the values of the conjugate potential on Iz 1 = a
and Iz 1 = b resp., fIJI and flJa the values of the potential on arg z = (X and
arg z = {J resp. Compute

(Resistance, except for a factor depending on the thickness and the


specific resistance of the plate.)

When the problem is more difficult it is advisable to consider at the


same time three (possibly four) planes: the z-plane (stream plane); the
w-plane (velocity plane) and the I-plane (potential plane); and possibly
the w-plane. The notation suggests the flow of a fluid. Since I = fIJ + i1p
and w = u - iv = iz are analytic functions of the complex variable
z = x + iy the Z-, w- and I-planes are conformally mapped onto each
other. These three planes are mapped onto the w-plane with preservation
of the angles but with reversed orientation. The w-plane in particular
is the mirror image of the w-plane with respect to the real axis. The
stream lines and the level lines in the stream plane (z-plane) correspond
to lines parallel to the axes in the potential plane (I-plane). The two
undetermined real constants contained in fIJ and 1p correspond to a trans-
lation of the I-plane.
98. Find an irrotational solenoidal vector field outside the unit circle
(Iz 1 > 1); wought to be = 1 for z = 00 and tangent to the unit circle.
(A fluid passing a circular pillar; at a considerable distance off the pillar
the flow is uniform.) [By reasons of symmetry the two segments of the
real axis inside the vector field have to be stream lines. We may expect
that the horizontal component of weverywher.e points to the right. Find
the mapping of the stream plane onto the potential plane.]
99. At which points of the vector field determined in 98 does the
velocity vector vanish? (Stagnation points.) In how many points of the
w
field does assume the same value? Where is the pressure p minimal,
where is it maximal? [90.] What is the resultant pressure exerted on the

yuana97@uw.edu
124 Mappings and Vector Fields

pillar? Rotate all the vectors of the field through 90°: Give a physical
interpretation of the vector field so obtained.
100. The figure represents the contours of a vector field determined
by the following conditions: The irrotational and solenoidal field covers
the entire upper, and part of the lower, half-plane; it is symmetric with
respect to the imaginary axis. For z = 00 the velocity Ui becomes -i,
for z = 0 we have Ui = O. The following stream lines are known: the
positive imaginary axis, the two pieces of real axis from z = 0 to z = 1
and from z = 0 to z = -I (C to A, A to B and A to D in the figure);

Stream plane (z-plane)

the direction of Ui is indicated by arrows. The two remaining curvilinear


parts of the boundary, which lead from z = I and z = -I (B and D in
the figure) to - i 00 should be determined so that they are at the same
time stream lines and lines of constant velocity, i.e. w is tangent to,
and Ito I constant on, both curves. Draw the contours of the images of
the field in the iii-plane and in the I-plane. (Stagnant water, "dea~ water",
or infinite wake of constant pressure, behind a planklike barrier perpen-
dicular to the direction of floW.) The segment from z = -I to z = I
represents the barrier, the stagnation point z = 0 is at its center. The
undisturbed flow is assumed to be homogeneous with constant velocity
-i. Throughout the wake the pressure is constant; this implies, according
to Bernoulli's equation [90], that lUi I = const. along the stream line that
separates the stagnant from the flowing water.
101 (continued). With the help of a conformal mapping express w
in terms of I and then z in terms of I [82]. Determine the width of the
wake at a great distance.
102 (continued). Assuming that the density is (! = 1, compute the
total pressure against the barrier.

yuana97@uw.edu
Pt. III, Chap. 2, No. 100-102· Chap. 3, No. 103-109 125

Chapter 3

Some Geometrical Aspects of Complex Variables

§ 1. Mappings of the Circle. Curvature and Support Function


103. Suppose that the point z is moving with uniform angular velocity
1 on the circle Iz I = r. Find thr vector (its magnitude and direction)
that represents the velocity of the image w (in the w-plane) of the moving
point z under the mapping w = j(z).
104. We consider the image of the circle Iz I = r in the w-plane under
the mapping w = t(z). What is the shortest distance of the tangent at
the point w = t(z) from the origin of the w-plane?
105. The point z moves with uniform angular velocity 1 on the circle
Iz I = r. What is the angular velocity ofthe vector drawn from the origin
to the point w = t(z} in the w-plane?
106. The image of the circle Iz I = r under the mapping w = t(z) has
at the point w = t(z) the curvature
1 + rozj"(z}
1 III f(z)

I! Izf(z) I

[We are dealing with the angular velocity of the rotation of the tan-
gent.]
107 (continued). The sign of the curvature depends on the situation
of a fixed point (e.g. the origin) not on the trajectory of the moving point
w = t(z): The fixed point may be to the right or to. the left of this tra-
jectory and on its concave or convex side. Explain this dependency.
[Note the example w = z.. + a, n real, a complex.]
108. We assume that in the motion defined in 103 the point w = t(z)
describes in the positive direction a closed curve without double points.
This curve is everywhere convex (seen from outside) if and only if for
Izl =r
j"(z}
gtz f'(z) > -1.

109. A closed curve without double points is called starshaped with


respect to a point in its interior if any ray from this point intersects the
curve at exactly one point (all the points of the curve can be "seen"
from this point). Let the image of the circle Izl = r under the mapping
w .-: t(z) be a closed curve without double points described in the positive
sense [103]. The curve is star-shaped with respect to the origin w = 0 if

yuana97@uw.edu
126 Some Geometrical Aspects of Complex Variables

and only if on Izl =r


f'(z)
ffiz f(z) > o.
110. The image of the circle 1z 1 = r under w = fez) is convex if and
only if the function w = zl' (z) maps 1z 1 = r onto a curve that is star-
shaped with respect to the origin.
111. The set of points with respect to which a closed curve is star-
shaped forms a convex set. Prove this purely geometric proposition for
analytic curves with the help of 109.
Let Sf denote a finite convex domain in the w = u + iv-plane. For
a fixed angle q; the expression
u cos q; + v sin q; = ffiweiq;
assumes a certain maximum h{q;) in Sf. The function h{q;) is periodic
with period 2n and is called the support function of Sf. The straight line

u cos q; + v sin q; - h(q;) = 0

is a line of support-; its normal points away from Sf and forms the angle q;
with the positive u-axis. If Sf extends to infinity these definitions are
modified insofar as a finite maximum exists only in a sector with an
opening ::;; n. The two cases: the infinite strip and the half-plane are
exceptions; then there exist only two lines of support, or only one,
respectively.
112. Find the support function of the convex domain that consists
of the point a = 1ale"'" .
113. The function w = fez) establishes a one to one correspondence
between the disk Iz 1::;; r and the convex domain Sf. Suppose that I{z)
is regular and I'{z) =F 0 at a certain point z on the boundary, Iz 1 = r. In
this case a definite line of support (tangent) passes through the boundary
point w = fez) of Sf. Express the corresponding quantities q; any h{q;)
in terms of fez).
114. The function w = log (1 + z) maps the disk 1z 1 = 1 onto an
infinite domain contained in the strip - ~ <,3w < ;. Its support
function (defined only for - ~ < q; ::;; ;) is
h(q;) = cos q; • log (2 cos q;) + q; sin q;.
115. The function w = ~I arcsin iz maps the disk 1z 1 < 1 onto a
finite convex domain which has two corners and lies in the strip

yuana97@uw.edu
Pt. III, Chap. 3, No. 110-118 127

I
-1'& ::;; 3w ::;; 1'&. Its suppOrt function is
cos rp log (V cos 2qJ + V2 cos rp)2 + 2 sin rp arcsin (Vi" sin rp) :If

h(rp) = for 0 ::;; rp ~ '4;

1'& sin rp for :::;; rp ::;; ; ;

h(rp + 1'&) = h( -rp) = h(rp) .


116. The equation we-Ul+1 = z defines w = fez). It maps the disk
1z 1 ::;; 1 onto a finite convex domain which has one corner and lies in
the half-plane mw < l. Its support function is

h(rp) = cos rp for - : ~ rp ~ : ;

.
tn t he sector '4
n < rp < 4'
7n 1't'IS gIven nc f orm
. ,tn paramet'

h(rp) eiq> = 1 : w m(l - w)

whereby w describes the boundary, 1we-VJ+l1 = l.

§ 2. Mean Values Along a Circle


117. With z = e,1} we find for k, l = 0, 1, 2, 3, ...

for k =F l,
for k=l.

The functions 1, z, Z2, zS, ... form an orthogonal system on the unit
circle.
118. Let fez) denote a regular function on the disk Izi ~,. The arith-
metic mean [II 48] of fez) on the circle 1z1 = , is defined by

1
"2
f2""1}
f(,e') d{} =

lim
fIr) + f(r(u,,) + f(rcu!) + ... + f(rcu:- 1 )
,
no "-+00 n
2",
where co" = e" . Show that

f- f f(,e'I}) d{}
no
2,.
= f(O) .

This means: If an analytic function is regular at every point of a closed


disk its value at the center of the disk is equal to the arithmetic mean of
its values on the bounding circle.

yuana97@uw.edu
128 Some Geometrical Aspects of Complex Variables

119. Let the function I(z) be regular and different from zero for
Izi < 1'. Prove that the geometric mean of I/(z) Ion the circle Izl = l'
1 2,.
2" I log/f("'/)j4/) .. - - - - - - - , , - - -
e"o = lim VI/(r) I(rw.. ) I(rw;) "'/(rw:- 1= 1/(0) I.
...... 00
1)

[log I(z) is regular for 1z I ::;; 1'.]


120. The function I(z) is regular in the disk Iz I < l' and does not
vanish for z = 0; the zeros of I(z) in the disk are Z1' Z2' ••• ' z..' where each
multiple zero is represented with its mUltiplicity. Then the geometric
mean of I/(z) I on the circle Iz 1 = l' is
1 2,.
2" I log/f(,.'/)l/d/)
.
..

e "0 = 1/(0) liz z


1 2
:'~I
..

(f(z) = (z - Z1) (z - Z2) ••• (z - z.. ) I·(z), I·(z) regular and different
from 0 for Izl < 1'.]
121. Under the hypothesis of 120 the geometric mean of I/(z) I on
the disk 1z I ::;; 1',
1 ,2",
,.,1 f f log/f(~,'/)/gdgd/)
g(r) = e 00

is always smaller than the geometric mean of I/(z) I on the circle Izl = 1',
1 2,.
2" f log/J("'/)/d/)
<M(r) = e "'0

We have in fact
.. ( /.,/1+/.,/1+ ... +/ ... /1)
9(1')
- - - e- -
2
1-
..,
1

(}I(,.) - •
122. Let I(z) = ao al z + a2 z2 + a..z" be a regular + '" + + ...
function for Iz I ::;; 1'. The arithmetic mean of I/(z) 12 on the circle
jzl=1'is
1
J I/(re'/)) 12 dD =
2,.
2n: lao 12 + la1 12 1'2 + la2 12 1'4 + ... + Ia.. 12 1'2" + ....
o
123. Assume that I(z) = ao + arz + a2 z2 + .,. + a..z" + ... is regu-
lar for Iz I ::;; 1. The partial sums
s.. (z) = ao + a1z + a2 z2 + ... + a..z", n = 0, 1, 2, ...
of the power series of I(z) have the following minimum property: If P(z)
denotes any polynomial of degree n the integral
2",
~ J I/(e
2n: 0
i /)) - P(e'/)) 12 dD

yuana97@uw.edu
Pt. III, Chap. 3, No. 119-127 129
is a minimum if and only if P,,(z) = s,,(z). The minimum equals
la"+112+ la"+212+ la"+312+ ....
§ 3. Mappings of the Disk. Area
124. We assume that the function
/(z) = tlo + a1z + ~Z2 + ... + a"z" + ...
is regular for Izi ~.,. The mapping w = /(z) transforms the disk Izl ~.,
into a domain of the w-plane. The w-values to which several z-values,
Iz 1 <." correspond have to be counted with the proper multiplicity.
The area of the image is
n(lal 12r + 21~12.,4. + 31as12 -y6 + ... + n la,,1 2r" + ...).
(The area is the additive combination of the areas of the images of the
disk Izi =., under the functions w = a"z", n = 0, 1, 2, ... )
125. Let
co
W = /(z) = ~
~ a" z" = ... + a + a-"+1Z-,,+1 + ... + a -1Z-1
-II
z-"
"==-00
+ ao + a1z + ... + a"z" + ...
be regular in the annulus ., ~ Iz 1 ::s;; R. The area of the image (count
covering with the proper multiplicity) is equal to
co
n I n la,,12(R2" - r").
11=-00

128. Assume that

91(Z) = cz + Co + c: + ;c + ... + Izc"i + "', C'* 0,


is regular and schlicht for I~ 1 > .,. The range of w = 91(z) is a proper
subset of the w-plan~. The area of its complement is
n{lcI2.,2 _Iclll _.2I c.11 _ 31 cal l _ ••• ) .
,.2 ,-' ,..

127. Suppose that the function


co
W = /(z) = ~
~ anzn = ... + a-nz-n + a-n+lz-n+l + ... + a -1 Z-1
11=--00

+ ao + a1z + ... + anz" + ...


is regular and establishes a one to one relationship between the points
of the circle 1z 1 = ., and its image L. The area of the domain bounded by

yuana97@uw.edu
130 Some Geometrical Aspects of Complex Variables

Lis

"=-00
The area is considered positive or negative depending on whether the
moving image w of the point z describing the circle in positive sense
leaves the area enclosed by L on the left or on the right.
128. Let I(z) be regular on the disk Izi < r and let ](e) denote the
area of the image of the disk Iz I :::;;: e, 0 < e < r, under the mapping
w = I(z). Then

f J~\ .r J/(ri6) 12
, 2,.

4 de = df} - 2n 1/(0) 12.


o 0
129. Assume that the function
c
W = m(z)
T
= cz + c0 + --.!Z
+ --.!
z2
C
+ ... + -Z,. + ...
CIl

is regular outside the circle Iz I = r and that it maps the domain Iz J > r
univalently into the closed exterior of a curve L in the w-plane. We
assume a homogeneous mass distribution on the circle Izi = r and on
the curve L in the w-plane, a distribution such that arcs which correspond
to each other under the mapping w = qJ(z) carry the same mass. The
mass distribution defined in this way on L has a certain center of gravity
e (conformal center of gravity of L). We find
E=co'

§ 4. The Modular Graph. The Maximum Principle


Let the function I(z) = u + iv be regular in a domain 'Il of the
z = x + iy-plane which we conceive as horizontal. We assign to each
point z in 'Il the point over the z-plane with cartesian coordinates x, y, C,
where /
c= I/(z)12=u 2 +v2 •
The surface obtained in this way appropriately represents the variation
of the modulus of the function I(z). We will call it the modular graph!.
Jensen [Acta Math. Vol. 36, p. 195 (1912)] calls it an "an3.1ytic land-
scape".
130. We take the cylinder over the disk Iz I <r and intersect it with
the modular graph of the function
I(z) =ao +a!z +a2z2 + ... +a,.z" + ....
1 There is no connection with the modular group or with elliptic functions.

yuana97@uw.edu
Pt. III, Chap. 3, No. 128-137 131
The volume of that part of the cylinder that is contained between the
z-plane and the modular graph

= :nr2 (Iaol! + la1llr! + aI 2 11 1" + ... + I a.. 12r2" + ...)


1 2 3 n+1 .
131. Let y(z) denote the angle between the x, y-plane and the tangen-
tial plane of the modular graph at z, I/(z) 12. Then

tan y(z) = 2 I/(z) II I'(z) I.

132. A point of the modular graph with a horizontal tangent plane


belongs to one of two types, it is a "pit" or a "saddle point" : If the tan-
gent plane is the x, y-plane, then there are only isolated points of the
graph on it, pits. If the tangent plane is not the x, y-plane it intersects
the modular graph along a curve (level line) 2n branches of which meet
under equal angles, ::' at the point of contact which is a saddle point.
(For a mountain pass, a saddle point in an actual landscape, we expect
n = 2.) The 2n regions of the modular graph determined locally in this
way Jie alternately above and below the tangent plane. (All the pits are in
the x, y-plane, all the saddle points above, at diverse heights.)
133. The intersection of the modular graph of a polynomial with real
zeros only and of a plane perpendicular to the x-axis is a convex curve
the lowest point of which Jies in the x, C-plane.
134. Let I(z) be a regular and single-valued function in the disk
Iz - zo I <1' andMbe the maximum of I/(z) I for z on the circle Iz - Zo I = 1'.
Then
It(zo) I ~ M.
Equality holds if and only if I(z) is a constant.
135. The function I(z) is assumed to be regular and single-valued in
the domain 1). The maximum of I/(z) Ion the boundary of 1) is denoted
by M. Then
I/(z) 1< M
in the interior of 1) unless I(z) is a constant. (The Maximum Modulus
Principle, or briefly Maximum Principle; d. Chap. 6.)
138. What does the maximum principle say about the modular
graph?
137. The n points PI' P 2, ••• , p .. are given in a plane, P is a variable
point in tliis plane. In any domain 1) the function of the point P
PPI ' PP2 '" pP..

yuana97@uw.edu
132 Some Geometrical Aspects of Complex Variables

(PP. is the distance between the points P and P.) assumes its maximum
on the boundary.
138. Let t(z) be regular, single-valued and non-vanishing in the
domain ~. If t(z) is not a constant, It(z) I can assume its minimum only
at boundary points of ~.
139. We assume that the given points PI' P2 ' ••• , P,. are all inside a
circle of radius Rand P is moving along this circle. Then
,. ==--==---::
Vpp1
.pp ... pp
2,.

(the geometric mean of the n distances P P~) attains a maximum > R


and a minimum < R unless all the P:s coincide with the center of the
circle.
140 (continued). The same statement as in 139 is true for the maxi-
mum of the arithmetic mean
PP; + PP; + ... + PP,.
n

of the n distances P p., but it is not true for the minimum.


141 (continued). The same statement as in 139 holds for the mini-
mum of the harmonic mean
n
111
=+=+
PP PP
... +=
PP,.
I 2
of the n distances but not for the maximum.
142. Consider the domain bounded by a closed level line (It(z) I is
constant along this c'urve) without selfintersections and lying inside the
region where t(z) is regular. It contains at least one zero of t(z) unless
t(z) is a constant.
143. Given n points PI' P2 "'" P,. in a plane in which P varies. The
locus of the points P for which the product of the distances

PPI ' PP2 ••• PP,. = const.

is a "lemniscate with n foci". (The standard lemniscate represents the


special case n = 2, d. 4.) Show that a lemniscate with n foci can never
consist of more than n separate closed branches.
144. Let the function t(z) be regular on the disk Iz I :::;; 'Y and zo,
Izo I = be a point at which It(z) I assumes its maximum. Then
'Y, zo:;(:oi
is real and positive. [103, 132.]

yuana97@uw.edu
Pt. III, Chap. S, No. 1S8-144 • Chap. 4, No. 145-150 133

Chapter 4

Cauchy's Theorem. The Argument Principle


§ 1. Cauchy's Formula
145. Put
2". 8._ 1 + 8.
ro = e", z_ = aro·, /;. = - - 2 - - '

'P=1,2, ... ,n; zo=z", a fixed, a=l=O.


Compute the sum
Zl -

·0 + 8
z- 81 + Z3 - Z2 + ... +" 8 - Z
,,-I
Cl CI Ca C" '
which converges to the integral ¢ d: along Iz I = Ia I as n -+ 00.
146. Let k denote an integer different from -1 and L a closed curve
without double poillts and of finite length; if k < -2, L does not pass
through z = 0; furf:hermore the points z1' Z2' ... , z" are consecutive points
on L. Show that the integral ~ Z"dz vanishes; try to approximate it
by a sum of the form L

~(ZI - zo) + ~(Z2 - ZI) + ... + ((z" - Z,,_I)' Zo = z". [II 1, II 2.]
147. Evaluate
¢1~z'
along the ellipse z = x + iy, x2 - xy + y2 + X + y = O.
148. Show that
"
J
2"
xdfJ :n:
xz+ sinlfJ = 2V1 + Xl
when x> O.
o
149. Prove the formula

J (1 +
2,.
2 cost'/o)"cosnfJ d-D =
1-1'-2I'cosfJ
2:n:
V1-21'-SI'I
(1 - I' - V1 -
21'2
21' - sr)" ,
o
1
-1<'<3' n=O,1,2" ..
150. Evaluate the curvilinear integral
tf. (1 - Xl - y2) Y dx + (1 + x 2 yl) X dy +
'j.J 1 + 2X2 - 2y2 + (X2 + )'2)2

along an ellipse with the foci (0, -1) and (0, 1),

yuana97@uw.edu
134 Cauchy's Theorem. The Argument Principle

151. We have for 0 < Sh < 1


ins
f'"'X--1e- i " dx = r(s) e- 2
o
152. The equation

J
o
'"'
sin (?)
;r"
dx = _1_r(~)sin(~
n-l n n
:t)
2

holds for n > 1.


153. We find, assuming p > 0, 0 < IX < ;, n = 0, I, 2, ... , that

f'"' e -"Pcosc< •
SlD
(../.I' ) .. dX = -
~. SlDlX X
1r(n-+-1) . (n + SlD----.
1) .ox
o P P P
Note the special case IX = pn.
154. Assume p > 0, x> 0, p fixed, x variable. Show that
+'"'
lim x'"'+l
%~'"'
f
0
e- IP cos xt dt = r(p + 1) sin ' ; .
155. Let a > O. The integral

along the straight line s = a + it, - 00 < t < 00, parallel to the imag-
inary axis converges absolutely for all real values of IX. The integral
turns out to be
if IX :::;;; 0,
if IX >0.
156. We call p(t) the largest term of the series
"I t"
1+-+-+'"+-+'''
11 21 n!
Assume A > 0 and let z be the only positive root of the equation
A - z - e-' = O.
Then

{I
[ -:t J
1

-,",
+ '"' sin ; ei("+I-lj..
u
du= 0
if
if

yuana97@uw.edu
Pt. III, Chap. 4, No. 151-161 135
157. The Legendre polynomials can be defined as the coefficients of
the expansion into a power series [VI 91]

-:;===1==== = Po(x) + !J(x) + P (X) + ... + P,,(X) + ....


2
V1 - 28% + r
z r z8 z"+l

Deduce from this Laplace's formula (VI 88)


1 1 1;-;;--; d
P,,(x} = - J(x +/Xyx2-1)"V 01_
n -1 1- 012

and the Dirichlet-Mehler formula


l' ,.
P,,(cosD) =.!f V2 cos
n
(n t
(cost - cos D)
+ I) dt =.!f V2 sin
n
(n + I)
t
(cosD - cost)
dt,
o l'

O<D< :n;.
(The square roots are positive.)
158. Let ~ denote the half-strip
ffiz > 0, -:n; < 3z < :n;
and L the boundary of ~ consisting of three straight pieces. We orient
L so that ~ is on the right hand side of L. The integral

-2'
.t f ~dC=E(z}
,e
'''' L .. - z

defines a function E(z) for points z on the left hand side of L. Show that
E(z) is an entire function which assumes real values for real z.
159 (continued). We find
1
2-: J ~e dC = 1.
nJ L

180 (continued). The function

is bounded outside ~. Inside ~ the function

is bounded.
181. We define
2'
z(z-1)(r-2) ... (z-n) =I,,(z}.

yuana97@uw.edu
136 Cauchy's Theorem. The Argument Principle

Show that
~/,,(I) dz
lim !_!=2" .
11-+00 ~1/,,(I)lldll=J,
1_1=2"
where the integrals are computed along the positively oriented circles.
[II 217.]
182. Assume that /(z) is regular in the disk 1z 1 ~ r and different from
oon the cire1e 1z 1 = r. The largest value oORz ~~:~ on 1z 1 = r is atleast
equal to the number of zeros of /(z) in Izl < r.
183. Let Zl' Z2' ••• , z" be arbitrary but distinct complex numbers
and L be a closed continuous curve without double points, enclosing all
the z:s. The function /(z) is supposed to be regular inside of. and on. L.
Then
P(z) = ~ ~ /(1;.) co(C) - co(z) de
2m 'j' co(C) C- Z '
L

where w(z) = (z - Zl) (z - Z2) ••• (z - z,,). is the uniquely determined


polynomial of degree n - 1 that coincides with /(z) at the points
Zl' Z2' .... z".
184. The function /(z) is analytic on the segment a ~ z ~ b of the
real axis and it assumes there real values. The closed curve L is conti-
nuous without double points and encloses the segment a ~ %~ b; /(z)
is regular inside L. Let zl' %2' " .• z" denote arbitrary points of the real
segment [a, b]. Then there exists a point %0 on [a, b] such that

~ /(1) d% = ~ /(1) dz.


'j' (I - 11 ) (I - 12 ) ... (I - I,,) 'j' (I - 10)"
L L

185. Assume that the entire function F(%) satisfies the inequality

F(x
1 + iy) 1 < CIf!"!
in the entire %-plane. z = x + iy; C and (! are positive constants. Then

d(F(I) ) +00 e(-l)" F(n;)


dz sin et = - ,,=~oo (et - nn)1 •

Example: F(z) = cos (!Z.


168. The entire function G(z) is supposed to satisfy the inequality
IG(x + iy) I < cee!y!

yuana97@uw.edu
Pt. III. Chap. 4. No. 162-170 137

on the entire z-plane, z = x + iy; C, (! positive constants; in addition


assume that G(z) is an odd function, G( -z) = -G(z). Then
(_l)"G(n +!)n)
I
co
G(z) (!
2(!z cos (!Z = .. -0 «n + !) n)a - ~za •

Example: G(z) = sin (!z.


187. We suppose that the function /(z) is regular in the disk Izl :s;; 1.
Then
1 1 f
f /(x) dx = 2---;~ /(z) log z dz = 2~ ~ I(z) (log z - in:) dz,
o m 1_1-1 m 1_1=1
1 1
f ,t/(x) dx = 2m" -- ~ z"/(z) dz, k> -1; k 9= 0, 1, 2, ...
o e -11.1=1
We integrate with respect to x along a straight line from 0 to 1, with
respect to z along the positively oriented unit circle. We start at the
point z = 1 with the branch of log z that is real, and the branch of z"
that is positive, for positive z.
168. The function /(z) is regular in the unit disk 1z 1 :s;; 1 and satisfies
the condition

r 1/(e'")
2,.
1 d{) = 1.
IJ

j!
Let k> -1. Then

1 when k is an integer,
f
I
1 ,t/(x) dx :s;; 1
o ---- when k is not an integer.
21Rin knl'
169. Let IX > -2. The quadratic form of the infinitely many real
variables Xl' x 2, X 3, •••
i i X}.X,..

A-I ,..=1 1 + I' + IX


is bounded, i.e. there exists a constant M independent of n such that

i; i; xAx,.. I<M
I}.=1 ,..=1 1 + I' + IX

whenever the variables xl' x2 ' ••• , x.. satisfy the condition
~ + x~ + ... + x! = 1; n = 1, 2, 3, ... We may choose M = n: if IX is
an integer and M = ISlfiIXn
. n I if IX is not an integer.
170. We assume that the functions 11 (z), 12(Z)' ... , I.. (z), ... are regular
in the open region 9t and that they converge uniformly to the function

yuana97@uw.edu
138 Cauchy's Theorem. The Argument Principle

I(z) in any closed domain inside m. Then the limit function I(z) is regular
in m.
171. The complex function I(z) = u(x, y) + iv(x, y) of the real
variables x and y is defined and continuous in a region mof the z = x + iy
plane. Moreover we assume that the integral
9> I(z) dz
along any circle inside mvanishes. Then I(z) is an analytic function of
the complex variable z, regular in the entire region m. [Compute the
variation of the area integral
F,(z) = ff I(z + e + if}) de df}
"+'1';::;;,'
when the leal or the imaginary part of z varies.]

§ 2. Poisson's and Jensen's Formulas


172. The function I(z) is assumed to be analytic in the open disk
Iz I < 1, bounded on the closed disk Iz I s 1 and continuous, possibly
with the exception of a finite number of points. Then
2"
;-nof l(e i6) d{} = 1(0) .
(More general than 118.)
173. The function I(z) is regular in the disk Iz I < R; let 0 < r < R.
Then we have Poisson's formula
i6 1 f2:71 i9 RI - 1'1
/(re ) = 2n I(Re) R2 _ 2Rrcos (9 _ 0) + rade.
o
174. Suppose that the function I(z) is regular and bounded in the
half-plane mz ~ O. Then, for x> 0,
1 +00 . _ .
/(x + iy) = -
n
f
-00
I(if}) d arctan !L2.
x

175. We assume that the. function I(z) is meromorphic in the disk


Iz I < 1, regular and non-zero on the boundary and at the origin. The
zeros of I(z) in Iz I s 1 are aI' a 2 , ... , a",
and the poles bl , b2 , ''', b,. (mul-
tiple zeros and poles are listed with correct multiplicity). Then we have
jensen's formula
. 1 1 1
log 1/(0) I + log laJ + logjaJ + ... + log la",1
- log fb;r
1
-
1
logfbJ - ... - log fbJ = 2n
1 1
! log I/(e' ) I
2,. '6
d{}.

yuana97@uw.edu
Pt. III, Chap. 4, No. 171-177 139
[Draw circles of radius e around the zeros and poles in the open disk
\z\ < 1 so that these circles do not have points in common with each
other nor with the circle Iz \ = 1. Connect the e-circles with the unit
circle by paths that do not intersect each other (e.g. by radii of the unit
circle if all the poles and zeros have different argument, d. diagram).

By excluding the e-disks and the connecting paths we reduce the disk
\z \ < 1 to a simply connected region 91,. Compute the integral f log!(Z) dz
along the positively oriented boundary of 91,.]
178. The function I(z) is meromorphic in the disk \z\ < R, regular
and different from 0 on the circle \z \ = R and it has inside the circle
the zeros aI' a2 , ••• , a". and the poles bl' b2 , ••• , b,., counted with correct
multiplicities. If the point z = reilJ , r < R, is neither a zero nor a pole
we have
".
log I/(z) 1 + ,.~ log I
I RI -
(a,. _ z) RI- P~,. log I
a,.,.z I RI -
(b. -
b.z
-:ifli
I

1 2,. i9 RI-rt
= 2:r!; flog I/CRe ) 1 RI _
o
2Rf'COS (8 _D) + rt d8 .
177. The function I(z) is meromorphic in the half-plane mz > 0,
regular and non-zero on its boundary and it has inside the half-plane
the zeros al' a2 , ••• , a". and the poles b1 , b2 , ••• , b,., counted with correct
multiplicity. If I(z) is regular at infinity (but also under weaker condi-
tions on its behaviour at infinity) and if I(z) is regular and non-zero at
z = x + iy, x > 0, we have

IZ+ii i ,. Iz+b I
log I/(z) \ + 2
".
,.=1
log
I
z _a!!..l- .=1
,.
2 log i z _ b
I •
P

1 +00 _
= n- flog I/Ci1J) Id arctan !L.1..
x
-00

yuana97@uw.edu
140 Cauchy's Theorem. The Argument Principle

178. The function f(z) is regular in the domain

r~ Izl ~ R,
different from zero on the boundary of Si) and it has in Si) the zeros
al' a2 , ••• , am' a,. = r,.et6II, ft = 1, 2, ... , m. U smg
. the e
d f'lnltIon
..
log If(ee
iIJ ) 1 = U(e,
-D) we have the formula
:tr

1 ,.) 1 +z-
~ ( ;- - ; cos -D,. = nR f U(R, -D) cos -D d-D
~~<R ,. _~
2
R

+ 2~ f(l!~ - ;2) [U(e,.;) + U(e,- ;)Jde +X(R),


,
where X(R) is bounded as R -+ ex>; r, f(z) are fixed.
b," log f(z) (1Z2 +'R2 1)dZ
T has to be computed along a path analogous
to the one in 175.]

§ 3. The Argument Principle


179. Prove 25 by examining the change of arctan ~~:~ as x increases
from - ex> to + ex> along the real axis.
We consider a closed, continuous, oriented curve in the z-plane that
avoids the origin. If, starting from an arbitrary point, z describes the
entire curve in the given direction (returning to its starting point) the
argument of z changes continuously and its total variation is a multiple,
2nn, of 2n. The integer n is called the winding number of the curve.
180. Every ray from the origin intersects the curve in question at
leas tin 1 times.
In the sequel (181-194) L denotes a closed continuous curve without
double points and Si) the closed interior of L. The function f(z) is assumed
to be regular in Si), except possibly at finitely many poles, finite and non-
zero on L. As z moves along L in the positive sense the point w = f(z)
describes a certain closed continuous curve the winding number of
which is equal to the number of zeros inside L minus the number of poles
inside L. [The Argument Principle. Cf. Hille, Vol. I, p. 253.] The pro-
position remains true also when f(z) is only continuous and non-zero
on L.

yuana97@uw.edu
Pt. III, Chap. 4, No. 178-189 141
181. The functions p(z) and 'P(z) are regular in i), except possibly
at finitely many poles, finite and different from 0 on the boundary L
of i). We define I(z) = p(z) 'P(z). The winding number of the image under
w = I(z) of L is equal to the sum of the winding numbers of the images
of Lunder w = p(z) and w = 'P(z) respectively.
182. Prove the argument principle for a polynomial.
183. The argument principle implies: If p(z) is regular and non-zero
in the domain i) the winding number of the image of L (the boundary
curve of i)) generated by w = p(z) vanishes; i.e. the argument of p(z)
is a single-valued function on L. Deduce the general statement of the
principle from this particular case.
184. The real trigonometric polynomial

am cos ml} + b",sinml} + am+! cos (m + 1)1) + b",+! sin (m + 1) I} + ... +


+ a" cos nl} + b" sin nl}
has at least 2m and at most 2n zeros in the interval 0 < I} < 2n. [Ex-
amine
P(z) = (am - ibm) z'" + (am+! - ibm+) z"'+) + ... + (a" - ib~) z".]
185. Let 0 < a l < az < ... < a". Then the trigonometric polynomial
flo +a l cos I} + a2 cos 21} + ... + a" cos nl}
has 2n distinct zeros in the interval 0 ~ {J < 217' 2n. [22.] [Consequently it has
real zeros only, VI 14.]
186. The function I(z) is meromorphic in the interior of the curve L
and regular on the curve L. If \a \ is larger than the maximum of \/(z) \
on L then I(z) assumes the value a inside L just as often as it has poles
there.
187. The function
w = e'" - e-'"
assumes any value w with positive real part once and only once in the
half-strip ffiz > 0, -I < 3z < I·
188. Suppose that I(z) is regular in the closed disk \z \ <r and
univalent on the circle \z \ = r, that is it assumes there no value more
than once. Then the image of \z \ = r under w = I(z) has the same orien-
tation as \z \ = r and the function I(z) is univalent (schlicht) also in the
disk \z\ < r.
, Xl

189. The zeros of the function fe 2 dx lie with the exception of


o

yuana97@uw.edu
142 Cauchy's Theorem. The Argument Principle

Z = 0 inside the region ffi Z2 < O. [Cornu spiral, ct. e.g. A. Sommerfeld:
Optics. New York: Academic Press 1954, pp. 243-244].
190. The function /(z) is single-valued, regular and does not assume
a certain value a in the annulus., < Iz I < R. All the closed continuous
curves without double points that enclose the circle Iz I = ., and that lie
inside the annulus are mapped by the function w = /(z) - a onto curves
in the w-plane with the same winding number.
191. The function /(z) is regular in the domain il and its absolute
value on the boundary curve L of il is constant. As z moves on the curve
L the argument of /(z) changes monotonically. (Whence a new proof of
142.)
192. The function /(z) has one zero more than /' (z) inside Lunder
the hypothesis of 191. (More informative than 142.) This means geo-
metrically: Inside a closed level line without double points the modular
graph has more pits than saddle points, namely precisely one more.
193. If /(z) is regular in the domain il and /'(z) does not vanish in
il the mapping w = /(z) of il is not necessarily schlicht [72]. If however
I/(z) I is constant on the boundary of il the mapping in question has to
be schlicht.
§ 4. Rouche's Theorem
194. We suppose that /(z) and qJ(z) are two functions that are regular
in the interior of il, continuous in the closed domain il and that further-
more I/(z) I > IqJ(z) I on the boundary L of il. Then the function
/(z) + qJ(z) has exactly the same number of zeros inside il as /(z).
195. Let A. be rea], A. > 1. The equation
zi- z = 1
has exactly one root in the disk Iz I = 1. This root is real and positive.
196. Let A. be real, A. > 1. The equation
A. - z - e-' = 0
has only one root in the half-plane ffiz > 0 which, consequently, is real.
197. A function (not necessarily schlicht) that maps the closed unit
disk onto a domain contained in the open unit disk has exactly one fixed
point. I.e. if /(z) is regular in the disk Iz I < 1 and if I/(z) I < 1 in Iz I ~ 1
then the equation /(z) - z = 0 has exactly one root in Iz I ~ 1.
198. The entire function r;~i assumes each value infinitely often in
the half-strip
-d < 3z < d, ffiz < 0 (d arbitrary) .

yuana97@uw.edu
Pt. III, Chap. 4, No. 190-204 143
199. Let I(t) be a real-valued twice continuously differentiable
function on the interval 0 ~ t ~ 1. If 1/(1) I > 1/(0) I the entire function
1
F(z) = f I(t) sin zt dt
o
has infinitely many real zeros and only a finite number of complex zeros;
if 0 < 1/(1) I < 1/(0) I it has only a finite number of real, and infinitely
many complex, zeros. [The zeros of F(z) behave with respect to reality
like the zeros of 1(0) -/(1) cos z.]
_ . Let a be a constant, Ia I > 2.5. The power series

F
1 + -za + -a'z2 + -a'z2 + ... + -a'" + ... =
Z"
(z)

defines an entire function which does not vanish on the boundary of the
annulus

and has exactly one zero inside the annulus, n = 1, 2, ... [Examine
the maximum term on the circle Iz I = Ia 12.. , I 117.]
201 (continuation of 170). Let @) denote the set of all zeros of all
the functions I.. (z), n = 1, 2, 3, '" in m. If the limit function I(z) does
not vanish identically its zeros in mare identical with the limit points
of @) in m. (The term "limit point" is used here to mean a point an
arbitrary fixed neighborhood of which contains at least one zero of I.. (z)
for all sufficiently large n.)
202. The functions
... , I.. (z) ,
are schlicht in the unit disk Iz I < 1 and converge in any smaller disk
Iz I < l' < 1 uniformly to a not everywhere constant limit function I(z).
Then the function I(z) is schlicht in the unit disk Iz I < 1.
203. Let gl (z), g2(Z), ... , g.. (z) , '" be entire functions which have real
zeros only. If
lim g.. (z) = g(z)
"-+00

uniformly in any finite domain, the entire function g(z) can have only
real zeros.
204. Assume

a~ 0, d>O.

yuana97@uw.edu
144 Cauchy's Theorem. The Argument Principle

The entire function


"
~ a. cos (a + 'Pd) Z
has only real zeros.
205. Suppose that I(t) is a positive valued, never decreasing function,
1
defined on the interval 0 < t < 1 and that its integral JI(t) dt is finite.
o
The entire function
1
JI(t) cos zt dt
o
has real zeros only. [185.]
206. The domain 'l) contains the segment a < z ~ b of the real axis.
The functions 11 (z), 12(z), ... , I,,(z), ... are regular in 'l), they assume real
values for real z and they have no zeros on [a, b]. If these functions con-
verge in 'l) uniformly to a not identically vanishing limit function I(z)
then I(z) has no zero on the segment a ~ z < b. - This statement is
lalse.
206.1. The analytic functions 11 (z), 12 (z) • ... , I,,(z) are regular and
single-valued in the connected closed domain 'l); let c1• c2 • ••• , cn denote
constants. If the function
c1/1(z) + c2/2(z) + ... + cJn(z)
does not vanish identically the number of its zeros in 'l) cannot exceed a
certain upper bound which depends on 11 (z)./2{z), ... • In(z) and 'l) but
does not depend on c1' c2' ...• cn' (206.2 is less general but more precise.)
206.2. Let AI' A2• .•• , Al denote real n umbers
Al < A2 < '" < Al
and mI , m2 • ••• , ml positive integers,

Let 11 (z), 12 (z), ... , In (z) stand for the functions

taken in this order, and N for the number of those zeros of the function

yuana97@uw.edu
Pt. III. Chap. 4. No. 205-206.2 145
that are contained in the horizontal strip
cx<3z <{3.
Assuming that
lell + le2 1+ ... + le",,1 > 0, le"-mj+ll + ... + le.. -II + Ie.. I > 0
show that
(Aj ~_~1.~_(1J ~.~ _ n +1< N ::::;; ().j - "I) (/J. -.-x) +n _ 1.
2n - - 2n

(206.1 is more general but less precise.)

Chapter 5

Sequences of Analytic Functions

§ 1. Lagrange's Series. Applications


The power series
a1z + a2z2 + ... + a..z" + ... = W
which converges not only for z = 0 and for which al =1= 0 establishes a
conformal one to one mapping of a certain neighbourhood of z = 0 onto
a certain neighbourhood of w = O. Consequently the relationship between
z and w can also be represented by the expansion
blw + b2w2 + ... + b..w" + ... =.t,
albl = 1. To compute the second series from the first we set
1
q>(z).
al + a2 z + asz- + ... + a..z"-1 + ... =
9

The equation
z
W = 9'(z)'
where q>(z) is regular in a neighbourhood of z = 0, q>(O) =1= 0, implies

_ co
z - 2: 1iI
W" [d. -I [9'(X)]"]
,,-I .
,,=1· dx ",=0

More generally, if I(z) is regular in a neighbourhood of z = 0, then

(L) I(z) = 1(0) + i :; [d"-I j'(.~_rtx)t] .


.. =1 dx ",=0

yuana97@uw.edu
146 Sequences of Analytic Functions

(Biirmann-Lagrange series, d. Hurwitz-Courant, p. 137; Whittaker and


Watson, p. 129.)
207. We use the same notation and hypothesis as before and expand

j(z) , = i; w" [d" j(x) [q:(X)J"] .


1 - wrp (z) ,,=0 n! dx" %=0

Derive this formula from Lagrange's formula or Lagrange's formula


from this one by correctly using the generality of both formulas. [The
existence of one formula for a certain I(z) implies immediately the other
formula for another I(z).]
208. Prove the formula in 207 directly by expressing the coefficient
of w" as a Cauchy integral.
209. Expand in ascending powers of w the solution of the transcen-
dental equation

that vanishes for w = O.


210 (continued). Expand e'" in powers of w = ze-' where IX is an
arbitrary constant.
211. Expand in ascending powers of w the solution x of the trinomial
equation
1- x +
wxfJ = 0
that becomes 1 for w = O.
212 (continued). Expand XX in powers of w, where IX denotes an
arbitrary constant. (XX = Y is the solution of the trinomial equation

1 - 'lftX + wyPftX = 0.)

213 (continued). Note the cases p = 0, 1,2, -1,1 and derive 209,
210 by taking the limit in 211, 212.
214. Evaluate the sum of the power series
1 + i; (n + ~)" w"
,,=1 n.
What is its radius of convergence?
215. Prove 156 with the help of the results of 214.
216. Let IX and p denote rational numbers. Then the series

1 e e
+ (iX iii) w + ~ 2#) w2 + ... + ~ n~ W" + ...
represents an algebraic function of w.

yuana97@uw.edu
Pt. III, Chap. 5, No. 207-219 147
217. Arrange the successive powers of the trinomial 1 + w + w 2 in a
regular triangular array
1
1 + w +w2
1+2w+3w2 +2w3 +w'
1 + 3w + 6w2 + 7w3 + 6w' + 3w5 + w6
The sum of the middle terms (in boldface) is
1 + w + 3w2 + 7w3 + ... = 1 .
Vi - 2w - 3wl

218. Arrange the successive powers of the binomial 1 + w in a regular


triangular array (Pascal's triangle)
1
l+w
1 +2w +w2
1+3w+3w2 +w3
1 + 4w + 6w 2 + 4w3 + w'
Find the sum. of the middle terms (in boldface) and. more generally.
the sum of any column.
219. Find the generating functions of the polynomials P,,(x). p~,IlI(x).
L<:l (x). defined by the formulas

(1) P (x)·= _1_ d" (x 2 _ I)" (Legendre's polynomials);


.. 2"nl dx"

(2) (1 - x)'" (1 + x)1l p("',IlI(x) = (-1)" !:.... (1 - x)"+'" (1 + x)"+/I


" 2"n! dx"
tx> -1. P> -1 (Jacobi's polynomials);

(3) e-%x"'L("I(x) =...!... d" e-"x"+C< tx> -1


" n! dx" •

(generalized Laguerre's polynomials).


(Cf. VI 84, VI 98, VI 99. The generating function of the Legendre poly-
nomials is the series
Po(x) + Pl(x) w + P2 (x) w2 + ... + P,,(x) w" + ...
= 1 +xw +---w 3x
2
1
+ .... 2 -
2

yuana97@uw.edu
148 Sequences of Analytic Functions

the sum of which has to be found as a function of x and w; similarly in


the other two cases.)
We define as usual
LfF(z) = F(z + 1) - F(z),
Lf2F(z) = .1 [LfF(z)] = F(z + 2) - 2F(z + 1) + F(z) ,

.1 "F(z) . F(z+ n) - C) F(z + n - 1) + (;) F(z + n - 2)


- ... + (-1)" F(z)

220. Let s be a constant of sufficiently small modulus. Then the


following formulas are valid for F(z) = esz :

(1) F(z) = F(O) + :! LfF(O) + z(z 2~ 1) Lf2F(0) + ... +


+ z(z - 1) ···n(~ - n + 1) Lf"F(O) + ... ;

(2) F' (z) = LfF(z) - ! .1 F(z) + ~ .1 F(z) _ ... +


2 3

+(_1),,-1 ~Lf"F(z) + ... ;

(3) F(z) = F(O) + ;! F'(l) + z(z 2~ 21 F"(2) + ... +


+ z(z - n)"-l F(")(n) + ... [210J;
n!

~ z(z2 _ 12) (z2 _ 22) ••• [z2 _ (n _ 1\2] L.l 2,,-2 [F(-n + 2) _ F(-n)]
+ .,:., --
,,=1
(2n - I)! - 2

[212, 216J.
221. The four formulas mentioned in 220 hold for any polynomial
F(z). (In this case the series are obviously finite.)
222. The formulas (1), (2) given in 220 are also valid for any rational
function F(z) if the real part of z is larger than the real part of any of the
finite poles of F(z); formula (1) requires the additional condition that
F(z) be regular for z = 0, 1, 2, 3, ... -Do the formulas (3) and (4) hold
for rational but not entire functions?

yuana97@uw.edu
Pt. III, Chap. 5, No. 220-229 149
In the sequel (223-226) we use the notation

,1"ak = ak+" - (:)ak+"-1 +(~)ak+"-2 -'" + (-1)"ak·


223.
(1 - z)" 1: ak~ = 1: ,1"akz"+k
k=-oo k=-oo

where ak , k = 0, ± 1, + 2, ... , denote arbitrary constants.


224. Define
F(z) = a o + a1z + a2z2 + ... + a".7." + .. ,
and establish the relation

1: t F(llt) = ao + ,1aot + ,12aot2 + .. , + ,1"aot" + ....


225. Define
F(z) = a o + 2aiz + 2a2z2 + ... + 2a"z" + .,.
and show that

Vl+4t
1 F(1 + 21 - V1 + 41) = a
2t 0
+ ,12a -1t + ,14a -2 t2 + ... +
+ ,12n a_"t" + ....
226. Define
F(z) = 2a 1z + 2a2z2 + 2aszs + ... + 2a"z" + ... ,
and show that
~F(1+2t-V1+4t)
t 21
_
- a1
_ a_I
+(..12
LJ a o
_..12
LJ a_ 2
)t

227.
if (1 + _.:i!-=-_.1_)
,,=1 + n (n
=
1)
sin :n:z .
:n:z(l - z)

228. sin 1CZ is a single-valued function of w = z(l - z). The expansion


of sin 1CZ in powers of w contains only positive coefficients (except the
constant term) [227].
229. Prove that
d_"..;.(:n:_-_X)'--.-_"_-_l_c_os_x] 0 n = 0, 1,2, ...
[
dol'
" .<=0
> ,

yuana97@uw.edu
150 Sequences of Analytic Functions

§ 2. The Real Part of a Power Series


230. We assume that the function
fez) = ao + a1 z + a2 z2 + ... + a"z" + ... is regular in the disk Izl < R.
Express the coefficients aI' a2 , ••• , a", ... in terms of the real and the
imaginary part resp. of f(z) on the circle Iz I = r, 0 < r < R.
231 (continued). We set mflri,';) = U(r, f}) and assume that f(O) is
real and Iz I < r. Then
+ ze-;,';
J U(r, f})
1 2,. r
f(z) = 2 _;,'; df}.
no I' - ze
232 (continued). Suppose f(z) does not vanish on Izl = r and that
its zeros in the disk Izi < rare c1 ' c2 ' ... , cm' Then we have for Izl < r
1 ze-;,';
+ 2noJ log If(re',';) I
m (z - c ) I' 2,. .
+ 1'=1
I' , :
log fez) = iy ~ log 2 I' -i,'; df},
I' - C I'z I' - ze

where y is a real constant. [Consequence of 231 as 120 is a consequence


of 119.]
233. The function fez) is regular, its real part is positive in the open
disk Iz I < R and continuous in the closed disk Iz I < R. If the real part
becomes identically zero on an arc of the circle the imaginary part of
fez) changes on this arc always in the same sense: it decreases as arg z
increases.
234. Let fez) = ao + a1 z + a2 z2 + ... + a"z" + ... be regular in
the disk Iz I < R, f(re i,';) = U(r, f}) + iV(r, f}), U(r, f}), VCr, f}) are real.
Then the equation
2,. 2,.
J [U(r, f})]2 df} = J [VCr, f})]2 df}
o 0
holds for 0 < r < R provided that it holds for r = O.
235. The function
f(z) = ! + a z + ... + a"z" + ...
1

is assumed to be regular and to have positive real part in the open disk
Izl < 1. Then la,,1 < 1, n = 1,2, ... In none of these inequalities can we
replace 1 by a smaller number.
236. Suppose that the function
fez) = ao + a1 z + a2 z2 + ... + a"z" + ... is regular and that mf(z) < A
on the disk Iz 1< R. Then the inequality
Iao I + Ia1 Ir + Ia2 1r2 + ... + Ian Ir" + ... < IaoI + R 2~ I' (A - mao)

holds for 0 < r < R. Example: fez) = z + 1, R = 1, A = O.


z - 1

yuana97@uw.edu
Pt. III, Chap. 5, No. 230-240 151

237. We assume that the Laurent series

tp(z) = ;E a,.t'
11=-00

converges on the annulus 0 < Iz I < 00 (sphere from which two points
have been removed) and that z = 0 and z = 00 are essential isolated
singular points. The maximum of the real part of tp(z) on the circle
Izl = r is denoted by A (r). A(r) increases to 00 faster than any power of
r as r ~ 00 and faster than any power of ..!... as r ~ O. Precisely
/'

lim log A(r) =


'-+00 log r
+00, r log A(r)
~ --1- =
+ 00.
, log-
/'

238. We assume that the function I(z) = ao + arz + ... + a,.t' + ...
is regular in the disk Iz I < R and we denote by A U) the largest oscillation
of the real part of I(z) in Iz I < R, i.e. A (I) is the least upper bound of
191/(zl) - 91/(z2) I for IZ11 < R, IZ21 < R. Then we have
2
la1 1R ::;;"AU).
2
The factor - cannot be replaced by a smaller one.
11:
Interpret this proposition geometrically.
239. Let the function I(z) = ao + a1z + ... + a,.z" + ... be regular
in the disk Iz I < R and denote by DU) the largest oscillation of I(z)
for Iz I < R, i.e. D(/) is the least upper bound of I/(zl) - I(Z2) I for
IZ11 < R, IZ21 < R. Then we have
la1 1R ::;; "21 D(f).
The factor 1 can not be replaced by a smaller one. Interpret this propo-
sition geometrically. '
240. Let the function I(z) satisfy the following conditions:
(1) I(z) is regular, I/(z) I ::;; M Iz - s I : :; : r;
in the disk
(2) I(z) does not vanish in the closed half-disk Iz- s I< r, 91(z- s) > 0;
(3) I(z) has in the disk Iz - sl::;; i r the zeros cI ' c2 ' ... , cl • Then

/,(s) 2 MIl
- 9 1 - < - l o g - - ;E91--.
f(s) - /' If(s) I A=l s - c"

[We may assume s = 0; 232, 12O.J

yuana97@uw.edu
152 Sequences of Analytic Functions

§ 3. Poles on the Circle of Convergence


241. Suppose that the unit circle is the circle of convergence of a
certain power series and that there are only poles of first order on this
circle (no other singularities). Then the sequence of coefficients is bounded.
242. If there is only one singular point Zo on the circle of convergence
of I a,.z" and if Zo is a pole then
,.=0

243. Let I a,.z" be the expansion into a power series of a rational


,.=0
function whose denominator (relative prime to the numerator) has degree
q. The radius of convergence is called e and A,. denotes the largest among
the q numbers Ia,. I, la,.-ll, ... , la"-HII. Then
"/-- 1
lim
"-+00
I A" =-.
e
(lim not lim sup!)
244. Let v" be the number of non-zero coefficients among the n
coefficients ao' al' "', an_I' 1£ there are only poles (and no other singu-
larities) on the circle of convergence of the power series
ao + a l z + a2 z2 + ... + a,.z" + ... the number of such poles is not
smaller than
li msup-.
H-+oo
"
Vn

Example: 1 +!' + Z2k + Z3k + ... = _ l - k '


1- z
245. We assume that the coefficients ao' aI' ... , a.. , ... of the power
series ao + aiz + ... + a,,%" + ... are real and that eei " and ee-ic< are
poles and the only singularities on the circle of convergence, 0 < ()(. < 1t.
We call V" the number of changes of sign in the sequence ao' al"." a,._l' a".
Then
" =-
v
lim - l¥
. [VIII 14.]
11-+00 n :f

246. If there is also a pole among the singularities on the circle of


convergence the power series converges at no point of the circle of
convergence.
247. Suppose that the point z = 1 is a regular point of the power
series

yuana97@uw.edu
Pt. III. Chap. 5. No. 241-249 153
that converges inside the unit disk. Then the Dirichlet series. assumed
convergent for certain values of s.
D(s) = al l- s + a2 2- s + ... + a"n- s + ...
defines an entire function. In the case where the point z = 1 is a pole
of order h of the function I(z). D(s) is a meromorphic function which can
have poles only at the points s = 1. 2•... , h (only the last one must be
a pole); the poles are simple. [D(s) r(s) = J ,t-l/(e- Z) dx, cf. II 117;
o
J ,t-l/(e-
co
Z) dx is an entire function of s if (! > 0.]
1/

§ 4. Identically Vanishing Power Series


248. The sequence ao• aI' a2 • •••• a", ... is assumed to satisfy the
condition
. log la,,1
bmsup / = -h, h>O.
"-+00 In
Then the series
~(s) = 2ao + al(eS + e- S) + a2 (l2s + e-'12S) + ... +
+ a" (l;s + e- '1;s) + .. .
converges in the infinite strip
-h<ffis<h
of the s-plane. The convergence is absolute and uniform in any closed
interior strip -h + e < ms
~ h - e, e > 0; there the series defines an
analytic function ~(s). The function ~(s) vanishes identically only if
all the coefficients ao' a l • a2 • ••• , a", ... vanish. [Compute
1 a+ioo
J
-us
F(u) = -2.
11:1.
~(s) ~ ds. 5
0 < a < h, te> 0; 155.]
4-,00

249. We assume that the power series

I(z) = ao + a1z + a2.2 2 + ... + a"z" + ...


converges inside the unit disk and that I(z) and all its derivatives tend
to 0 as z tends to 1 on the real axis, i.e. lim P"l (z) = O. n = 0, 1, 2, ...
%=%-+1
Then there are two possibilities:
(1) I(z) vanishes identically;
(2) z = 1 is a singular point for I(z).

yuana97@uw.edu
154 Sequences of Analytic Functions

If the power series converges in a disk larger than the unit disk, i.e. if
. log la,,1
lim sup < 0,
"-+00 n
we are necessarily dealing with the first case. The weaker condition
. log la,,1
bmsup~<O
11-+00 yn

admits the same conclusion. [Construct the function 4>(s) of 248.]


250. The proposition 249 is not valid if the condition on the coeffi-
. . log la,,1 .
c1ents ao, aI' a2, ... , a", ... of I(z), namely lim sup ,/_ - < 0, 15 replaced
""00 vn
by
. log la,,1
hm sup--- < 0,
11-+00 n~

[Put
J
co
I(z) = e-zlJcOSIJ7I sin (x" sin,un) e- z (l-.) dx; 153, II 222.]
o

§ 5. Propagation of Convergence
The following exampl~s show that the convergence of sequences of
analytic functions is often "contagious".
251. If the series
g(z) + g' (z) + gil (z) + ". + g(") (z) + ".
converges at one single point at which g(z) is regular then g(z) is an entire
function and the series converges at every point. The convergence is
uniform in any finite domain of the z-plane.
252. If the sequence
ig'(z) i, Vigll(Z) i, ... ,
is bounded at a single point of the z-plane then g(z) is an entire function
and the sequence stays bounded at all the points of the z-plane. It even
has the same limit superior at all the points.
253. Let

be two infinite sequences, the second one being arbitrary, the first one
such that a" =f= 0, am =f= a" when m =f= n, m, n = 0, 1, 2, ... and that
1 1 1 1
-+-+-+".
ao a a
+-+".
a" 1 2

yuana97@uw.edu
Pt. III, Chap. 5, No. 250-256 155

converges absolutely. The equations


Q,,(ao) = co' Q,,(a l ) = cl ' Q,,(a,,) = c"
define a unique polynomial Q,,(z) of degree not larger than n. If the
sequence

converges at a single point z different from ao• aI' a2 , ••• it converges at


every point z, even unifonr.ly in any finite domain of the z-plane.
254. Let
... , ••• J

denote a sequence that is infinite in two directions and Q2"(Z) be the


polynomial of degree 2n that satisfies the equations
Q2,,(-n) = C_'" Q2,,(-n + 1) = c-n+l' ... , Q2n(0) = co' ... ,
Q2,,(n - 1) = Cn-l' Q2,,(n) = C",
If the sequence of polynomials
Qo(z), Q2(Z) , Q4(Z),
converges for two different non-integral values of z it converges at every
point z, even unifoImly in any finite domain of the z-plane.
255. The sequence ce' cl ' c2' ... , cn' ... is given. A polynomial Q,,(z) of
degree not exceding n can be found [VI 76] for which
Qn (0) = co' Qn,(1) = cl"' Q,,(2) = c2, ... , Qn" ( )
(n) = Cn'
These n + 1 conditions define Q,,(z) uniquely [VI 75]. If the sequence
Qo(z), Ql(Z), Q2 (z) , Q,,(z) ,
converges at a single point z, z =f= 0, it converges at every point z, in fact
uniformly in every finite domain of the z-plane.
258. We assume that in the unit disk, Iz I < 1, the functions
fo(z), 11 (z), 12(z), ... , I,,(z) , ... are regular and different from zero and that
their absolute values are smaller than 1. If lim 1,,(0) = 0, then
n ..... ""
lim I.. (z) = 0 in the entire open disk Izl < 1; the convergence is actually
....... ""
uniform in every smallt'I disk Iz I :S r < 1.
257. The harmonic {unctions
... ,
are assumed to be regular and positive in a certain open region mof the
x, y-plane. If the infinite series
uo(x, y) + ul(x, y) + u2(x, y) + ... + u,,(x, y) + ...

yuana97@uw.edu
156 Sequences of Analytic Functions

converges at a single point of 9! it converges everywhere in 9!; in fact,


it converges uniformly in any closed subdomain of 9!.
258. We suppose that the functions of the sequence
10 (z), It (z), Mz), ... , I.. (z), ... are analytic in the open region 9! and that
the sequence of their real parts converges uniformly in every closed
subdomain of 9!. Then the sequence of their imaginary parts either
diverges at all points or it converges uniformly in any closed subdomain
of9!.

§ 6. Convergence in Separated Regions


259. The series
z zZ z'
r~ + (1 -I- z) (1 -I- zZ) + (t -I- z) (1 -I- z l (1 -I- 2 Zl)

zd
+ (1 -I- z) (1 -I- z 1 (1 -I- z') (1 -I- zI) + ...
2

converges uniformly in any domain that lies either entirely inside or


entirely outside the unit circle and its sum is z or 1 according as Izi < 1
or Izi > 1 [114].
260. Let ex denote an arbitrary constant, ex =1= o. The series
1 + 2: €X(ex -I- n) II-I :r;" e
00
I
-II.%'

.. =1 n.
converges uniformly for all positive values of x; it represents the function
en for 0 < x ~ 1 and a dilferent analytic function for 1 < x < 00.
281. The sequence of functions 11 (z), Mz), ... , I.. (z), ..•

n = 1, 2, 3, ... ,

converges uniformly in any finite domain that does not contain the
imaginary axis.
282. Let
ex > 0, P > 0, ex + P = 1,
and put
!p(z) = exz + p-z1 .
The sequence of iterated functions
!p(z), 91 [!p(z)], !p{!p[!p(z)]},
converges to +1, when 9!z> 0, to -1 when 9!z < 0 and diverges when
9!z=O.

yuana97@uw.edu
Pt. III, Chap. 5, No. 258-265 157
§ 7. The Order of Growth of Certain Sequences of Polynomials
283. Let h(<p) denote the support function of the infinite convex
domain considered in 114. The sequence 11 (z), 12(z), ... , I.. (z), ...

I.. (Z) -- V:;-(z - 1) (4 - 2) ... (z - n


nl
+ 1) (z - n) -,Is(.,,)
e, n = 1, 2, 3, ... ,

where z = ,eiq>, is bounded in the entire half-plane 9lz > 0 [12, II 220];
h(<p) is the smallest function of the angle <p that keeps the sequence
bounded.
264. Let h(<p) denote the support function of the convex domain
considered in 115. The sequence 11(z), 12(z) , ... , I.. (z), ...
_~( 1- z,)( 1 _zl)
I.. (z)-~ 22 ... (1 _ZS)
n2 e-,h(.,,) , n-l,2,3,
_
... ,
12
where z = ,eiq>, is bounded in the entire z-plane [13, II 221]; h(<p) is the
smallest function of the angle <p that keeps the sequence bounded.
285. Let h(<p) denote the support function of the convex domain
considered in 116. Then

11 + : I" e-,h(.,,) < 1, n = 1, 2, 3, ... ,

z = ,eiq>, in the entire plane.

Chapter 6

The Maximum Principle

§ 1. The Maximum Principle of Analytic Functions


The values that an analytic function assumes in the different parts
of its domain of existence are related to each other: they are connected
by analytic continuation and it is impossible to modify the values in
one part without inducing a change throughout. Therefore an analytic
function can be compared to an organism the main characteristic of
which is exactly this: Action on any part calls forth a reaction of the
entire system. E.g. the propagation of convergence [251-258] can be
compared to the spreading of an infection. Mr. Borel advanced ingenious
reflections upon similar comparisons 1. We shall examine in what manner

1 E. Borel: Methodes et problemes de tbCorie des fonctions. Paris: Gauthier-


Villars 1922. Introduction.

yuana97@uw.edu
158 The Maximum Principle

the moduli of the values are related that the function assumes in
different parts of its domain of existence.
Let the function I(z) be regular in the circle 1z 1 < R; the maximum
of its absolute value on the circle Iz 1 = 1', l' < R, is denoted by M(1').
S. The maximum of I/(z) Ion the disk Izl < l' is M(1').
267. The maximum M(1') is monotone increasing with r unlrss I(z)
is a constant.
268. We assume that the function I(z) is regular in the simply connected
region Izi > R. M(1') denotes the maximum of I/(z) \ on the circle
Izi = 1', l' > R. Then M(1') is also the maximum of I/(z) \ for Iz\ >- l' and
M(1') is monotone decreasing unless I(z) is a constant.
269. Let I(z) denote a polynomial of degree n; then

Equality is attained only if the polynomial is of the form cz".


270. Suppose that I(z) is a polynomial of degree n and that
I/(~I < M
on the real interval -1 < z < 1. Then we have
I/(z) 1 $; M(a + b)"
for any z outside this interval; a and b are the semi-axes of the ellipse
through z and with foci -1 and 1. What does the proposition imply for
z-+oo?
271. We assume that I(z) is a polynomial of degree n, that El and E z
denote two homofocal ellipses with semi-axes aI' b1 and az, ba, a1 < az,
bl < b2. The maximum of I/(z) Ion El and Ea is denoted by Ml and M2
resp.; then
Ml >_~
(a1 + bl )" = (a 2 + t 2 )'"
Derive 289 and 270 from this proposition.
272. If an analytic function is regular in a closed disk and not a
constant its absolute value at the center of the disk is smaller than the
arithmetic mean of its absolute value on the boundary of the disk.
273. If the absolute value of an analytic function I(z) is constant
in an open set of the z-plane (e.~. in a disk) I(z) must be a constant.
274. We suppose that the functions <p(z) and lp(z) are regular in the
closed disk 1z 1 < 1 and non-zero in the open disk 1z 1 < 1 and that,
besides, <p(0) and 11'(0) are real and positive. If <p(z) and lp(z) have the
same modulus on the circle 1z 1 = 1 then <p(z) = lp(z) identically.

yuana97@uw.edu
Pt. III, Chap. 6, No. 266-278 159

275. The function I(z) is regular and single-valued in the interior of


the closed domain'll and continuous in'll, boundary included; the maxi-
mum of I/(z) Ion the boundary of 'll is called M. Under these conditions
we have the inequality
I/(z) 1< M
in the interior of 'll unless fez) is a constant. [This statement is stronger
than 135.]
276. Let 'll denote a domain, Cbe an inner point of 'll and ~ be the
set of those boundary points of 'll whose distance to Cdoes not exceed
e. The circle of radius e and center Cis assumed to have an arc that does
not belong to 'll and the length of which is not smaller than 2ne , n integer.
n
We suppose that the function I(z) is regular and single-valued in the
interior of 'll and continuous on the boundary and that, in particular,
I/(z) I < a at the points of ~ and I/(z) I < A at the remaining boundary
points of'll; a < A. Then
1 1
--- 1--
I/(C) I < an A n

n
n-l
[Examine the product I[C + (z - C) £0-0] in a suitable domain,
co = e2n;/n.] >=0

277. We assume that the function I(z) is regular and bounded in


the sector 0 < arg z < IX, continuous on the real axis and that lim I(x) = 0,
x real, x > O. Then the limit relation %-+00
lim I(z) = 0
"'-+00
holds uniformly in any sector 0 < arg z < IX - e< IX.

278. We denote by M a positive constant and by ffi a connected open


region. The analytic function I(z) is assumed to have the following
properties:
(1) I(z) is regular at every point of ffi;
(2) fez) is single-valued in ffi;
(3) each boundary point of ffi has for every positive number e a
neighbourhood such that at every point z of ffi in this neighbourhood
the inequality
I/(z) 1< M + e
is satisfied.

yuana97@uw.edu
160 The Maximum Principle

These conditions imply


If(z) I <M for z in ffi
and even If(z) I < M unless fez) is a constant. [Stronger than 275.J
279. Let fez) be regular and bounded on the disk Iz I < 1 and let
lim f(re iD ) = 0
.-+1

hold uniformly in a sector 1X < -0 < p, 1X < p. Then fez) vanishes iden-
tically.

§ 2. Schwarz's Lemma
280. The function fez) is assumed to be regular and If(z) I < 1 in
the disk Iz I < 1. If f(O) = 0 either the stricter inequality If(z) I < Iz I
holds for z =F 0 or fez) = ei"'z, 1X = real.
281. We denote by z = 97(C) and w = tp(C) two schlicht maps of
the unit disk ICI < 1 into the regions ffi and e; of the z- and w-plane
resp.; the images of the origin C= 0 are the points z = Zo and w = Wo
resp. In addition let 0 < (! < 1 and r and ~ be the images of the disk
ICI < (! under the above mentioned maps. We assume that w = fez) is
a regular analytic function in ffi the range of which belongs to e; and
that f(zo) = WOo Then fez) assumes in the subdomain r of mvalues that
belong to the subdomain ~ of e;. These values are in the interior of ~
unless fez) is a schlicht function that maps the region monto the region e;.
282. Let fez) be regular and If(z) I < 1 for Iz I < 1. Then

If(z) - f(O) I ::;: Iz I1 1_ --11(0)


11(0) 12
liz 1' 0< z
I I< 1.

The relation will be an equality only for the linear function


e''''z+wo
fez) = ., 1X real.
1+ woe''''z
283. Suppose that fez) is regular for Iz I < R and that A (r) denotes
the maximum of the real part of fez) for Iz I <r, 0 < r < R. Then we
have the inequality

A (r) < ~ ~ : A (0) + /: r A (R), 0 < r < R,


where lim A(r) = A(R) [A(r) increases monotonically with r, 313J.
.-+R-O
There is equality only for the linear function

fez) = Rwo + Lwo - 2A(R)] e''''z ,


1X real.
R - e''''z

yuana97@uw.edu
Ft. III. Chap. 6. No. 297-288 161

284 (continued). The maximum M(r) of the absolute value of I(z)


in the disk Iz I ~ l' is restricted by

M(,.) ~M(O) + R 2.:.1' [A(R) - A(O}] ~ ~~:M(O} + /.:.I' A(R).


285. We assume that I(z} is regular. non-zero and bounded for
Izl < R. Then
R-, 2,
M(r} < M(o}R+' M(R)R+' ,

where lim M(r) = M(R}.


' .... R-O
286. The functions ft(z), Mz), fa(z}, ... , I,.(z}, •.. are supposed to be
regular, non-zero, and of absolute value smaller than 1, for Izl < 1. If
the series
11(0) + MO) + fa(0) + ... + 1,.(0) + ...

is absolutely convergent, the series

is absolutely convergent for Iz I ~ t.


281. We assume that the function I(z) is regular, has positive real
part for Izl < 1 and that 1(0) is real. Then we have
1-l z l 1+l zl
I(O} i+l~ ~ 91/(z) ~ 1(0) l=TzT'
I3/(z) I < 1(0) 1 21z1
_ I z l2 '

I(O)! ~ 1:\ < I/(z) I </(O)! ~ 1:\, 0< Izl < 1.


There is equality only if
1 + i"'z
I(z) = Wo ., W o, IX real, wo> O.
1 - 8""Z

288. Let I(z) be a regular function with I91/(z) I < 1 for IzI < 1. If
I(O} = 0 the stronger inequality

191/(z) I <-;: arctan Iz I, 0< Iz 1< 1,


4

holds; in addition
1 + Izi
I3/(z} I ~nlog~,
2
O<l z l<1.
Equality occurs if and only if
2
I(z) = ---, log
1 + i"z
.• IX real.
nl 1- 8""Z

yuana97@uw.edu
162 The Maximum Principle

289. Suppose that the function I(z) is regular for 1z 1 < R and that
.1 denotes the oscillation of its real part in the disk, i.e
191/(zl) - 91/(z,) 1 < .1
for 1zll < R, 1z21 < R. Then the oscillation of its real part in the smaller
disk Izl :5: r, r < R, is

The oscillation of the imaginary part is restricted by the inequality

13/(zl) - 3/(z,,) 1 :5: n2<1 log RR+,.


_ ,. , IZll < r, IZ81 :5: r.
290. We denote by ;t the infinite region of the z-plane that is symme-
tric with respect to the x-axis and whose points are characterized by
the inequalities
x>O, -k(x) <y< k(x) ,
where k(x) is a positive and continuous function defined for x ~ O. Then
there exists a positive function h(x) that depends only on ;t and has the
following properties: If F(z) is regular in ;t and bounded from below,
1F(z) 1> C, then
log IF(x) I
h(x)
is bounded from above as x increases to infinity. (The theorem is parti-
cularly interesting when k(x) converges monotonically to 0; i.e. when
the region ;t is "tapering"; whereas an analytic function can increase
arbitrarily fast along a ray (IV 180) the rate of increase is limited if the
function grows to infinity in some properly chosen neighbourhood of the
ray. The result can be formulated with reference to the points of the
modular graph somewhat vaguely so: If none should fall below a certain
minimum standard none may rise arbitrarily high.)
291. Let I(z) be a regular function and I/(z) 1< 1 in the unit disk
1z 1< 1; in addition let I(z) be regular at z = 1 and 1(0) = 0, 1(1) = 1.
Then 1'(1) must be real and 1'(1) ~ 1.
292. We assume that the function I(z) is regular and I/(z) 1 < 1 in
the unit disk 1z 1 < 1, and that, besides, I(z) is regular for z = 1 and
1(1) = 1. Then the derivative /'(1) is real and
, 1 - If(z) 12 > 1- IZ 12
I (1) I t - f(Z)i2 = It - z12' Izl < 1.
293. Suppose that I(z) is regular and 3/(z) > 0 in the upper half-
plane 3z > 0 and that, besides, I(z) is regular at the point z = a of the

yuana97@uw.edu
Pt. III, Chap. 6, No. 289-298 163
real axis and I(a) = h, h real. Then t' (a) is real and positive and we have
the inequality
3_1_ > 3 _1-._ for 3z> o.
b - /(z) = (a - z\ /,(u)

294. Let the function I(z) be regular, have the zeros Zl' Z2' "., z..
and let I(z) be bounded, I/(z) 1<M, in the unit disk Izl < 1. Then the
stronger inequality

I/(z) I ~ I
z- z z- z z - z"l
~. ----:.!". ---- M
- 1 - ZlZ 1 - zaZ 1 - z..z

holds for Izl < 1. We have equality either at every point or at no point
of the open disk Izl < 1. (Proposition 280 is a special case. n = 1,
Zl = 0.)
295. Let the function I(z) be regular, have the zeros zl' Z2' ••• , z.. and
let I(z) be bounded, I/(z) 1 ~ AI, in the open half-plane 9lz> O. Then
the stronger inequality

1I(z) 1 ~
- I +
1:1 - z .:, -+ z ".~z.. + z IIM
81 Z 8.
8 .. -
8

holds for all z with 9lz > O. We have equality either at every point or
at no point of the right half-plane 9lz > O.
298. A function that is meromorphic in a closed disk and of constant
absolute value on the boundary circle is a rational function; in fact it is,
up to a constant factor, the product of linear fractional functions that
map the disk in question either onto the interior or the exterior of the
unit circle.
297. We assume that the function I(z) is regular and bounded in
the disk Iz I < 1 and vanishes at the points Zl' Z2' Z3' ••. Then

(the sum of the distances of the zetos from the unit circle) is finite or
else I(z)= O.
298. We assume that the function I(z) is regular and bounded in
the half-plane 9lz> 0 and vanishes at the points Zl' Z2' Z3' ••• outside
the unit disk in the half-plane, i.e. Iz.. 1> 1, 9lz.. > 0, n = 1, 2, 3, ...
Then the sum of the series

z.
1 1 1
91-+91-+91-+'"
81 8a

is finite or else I(z) = O.

yuana97@uw.edu
164 The Maximum Principle

§ 3. Hadamard's Three Circle Theorem


299. The sum of the absolute values of several analytic functions
attains its maximum on the boundary. Here is a more detailed statement:
The functions 11 (z), Mz), 13 (z), ... , I.. (z) are supposed to be regular and
single-valued in the domain ~. Then the function
9'(z) = 1/1 (z) 1+ IMz) I + ... + 1/.. (z) I,
which is continuous in~, assumes its maximum on the boundary of ~.
300 (continued). The function 9'(z) assumes its maximum only on
the boundary of ~ unless all the functions Mz), Mz), ... , I.. (z) are con-
stants.
301. In three dimensional space, the n points Pl' P2 , ••• , p .. are
given and P denotes a variable point. The function
m(P)
T
-- PP1 • pp2 ... P P ..

(PP7 is the distance between P and PI,) of the point P assumes its
maximum in any domain on the boundary. (Generalization of 137.)
302. We assume that the functions Mz), Mz), ... , I.. (z) are regular
and single-valued in the domain ~. Let P1' P2 , ••• , P.. denote positive
numbers. The function
9'(z) = 1/1 (z) 11'1 + IMz) II'· + ... + 1/.. (z) II'..
is continuous in ~. It reaches its maximum only on the boundary of ~
unless all the functions it(z), Mz), ... , I.. (z) are constants.
303. The function I(z) is supposed to be regular in the multiply
connected closed domain ~ and I/(z) I single-valued in ~. [f(z) is not
necessarily single-valued.] The absolute value I/(z) I attains its maximum
at a boundary point of ~. The maximum cannot be attained at an inner
point of ~ unless I(z) is a constant.
304. Let the function I(z) be regular in the disk Izl < R. Suppose
O<r1 <r2 <r3 <R.
Then
10
g
M(r)::;: log 1'2
2 -
- log 1'! 10
log 1'a - log 1'1 g
M(r)
3
+ log 1'a -
log 1'. -
log 1'110
log 1'1 g
M(r).
1

This means that in an orthogonal system of coordinates the graph of


log M(r) as a function of log r appears as a convex curve. (Hadamard's
three circle theorem.) [Examine z"'/(z) with suitably chosen IX.]
305 (continued). The function log M(r) is a strictly convex function
of log runless I(z) is of the form az"', a, IX constants, IX real: this is the
only type of function for which the inequality in 304 becomes an equality.

yuana97@uw.edu
Pt. III, Chap. 6, No. 299-311 165
306. Suppose that the function I{z) is regular for 1z 1 < R but not
of the form c:l', c constant, n integer; let

denote the arithmetic mean of I/(z) 12 on the circle Izl = r, r < R. The
function I 2{r) is monotone increasing with r and log I 2 (r) is a strictly
convex function of log r.
307. Let I{z) be regular for Izl < R;
1 2,.
2n J logl/l,,'''>l4''
QS(r) = e 0

denotes the geometric mean of I/{z) 1 on the circle Iz 1 = r, r < R. The


function QS(r) is monotone increasing with r and a convex function of
log r (in the wide sense).
308. The function I(z) is supposed to be regular fOI 1z I < R and not
a constant. We put
2,.
1
I(r) = -2
no
J
I/(re''') 1rI{}, r<R.

The function I{r) is monotone increasing with r and log I(r) is a convex
function of log r. [299,304.]
309. Let I(z) be regular and not a constant in the disk 1z 1 < R.
The function w = /(z) maps the circle Izl = r, r < R, in !he z-plane
onto a curve in the w-plane with length l(r). The ratio 12(1') is monotone
n,.
increasing with r.
310. We suppose that I(z) is regular and not constant for Iz I < Rand
that P is a positive number. We define

r< R.

The function Ip(r) is monotone increasing with r and log Ip(r) is a convex
function oflog r. (Cf. 306, 308, 307, 267 and 304, for an analogous case see
IV 19.)

§ 4. Harmonic Functions
311. An analytic function that is regular in the closed disk ~ cannot
assume real values at all the boundary points of ~ except if if is a real
constant.

yuana97@uw.edu
166 The Maximum Principle

312. If a harmonic function is regular in a closed disk its absolute


value at the center does not exceed the arithmetic mean of its absolute
values on the boundary. Under what conditions is there equality?
313. A harmonic function is supposed to be regular and single-valued
in a domain 1). Then it attains its maximum and its minimum on the
boundary and only on the boundary unless it is a constant.
314. A harmonic function that is regular in the domain 1) and vanishes
at all the boundary points of 1) is ide,tically zero.
315. The equilibrium described in solution 31 is not stable.
316. A harmonic function is assumed to be single-valued in the
domain 1) and regular with the exception of finitely many points at
which it becomes - 00 (i.e. the function converges to - 00 as z approaches
such a point). Then it assumes its maximum on the boundary of 1).
317. Let the function I(z) be regular in the disk 1 il < R. The function
w = I(z) maps the disk onto a certain domain of the w-plane that is
assumed to be star-shaped with respect to w = o. Let 1(0) = o. Then
the images of the concentric circles I z 1 = 1', l' < R, are also star-shaped
with respect to the origin.
318. Suppose that the function I(z) is regular in the disk I z 1 < R
and that it maps the disk onto a convex domain of the w-plane. Then
the image of an arbitrary circle in the open disk 1z 1 < R is also convex.
319. Let "1 (x, y), "2(X, y), ... , ",,(x, y), z = x + iy, be regular har-
monic functions in a domain 1). The continuous function

assumes its maximum on the boundary of 1).


320. Consider a regular harmonic function in the disk Izl < R. We
denote by A (1') its maximum on the circle Izl = 1', l' < R. When
o < 1'1 < 1'2 < 1'3 < R we have
A(1' \ ~ log'. -log"l A(1') + log,.. -log,.. A(1' )
21 - log", - log"1 3 log.,.. - log"1 1 '

i.e. A(1') is a convex function of log 1'.


321. Deduce Hadamard's three circle theorem 304 from 320 and
vice versa 320 from Hadamard's three circle theorem.

§ 5. The Phragmen-Lindelof Method


322. Let IX be given, 0 < IX < ;. The function I(z) is assumed to
be regular in the sector -IX ~ {} ~ IX (z = 1'e'·). In addition I(z) has the

yuana97@uw.edu
Pt. III, Chap. 6, No. 312-322 167
following properties:
(1) there exist two positive constants A and B such that
I/(z) 1< AeBJzJ for -lX ~ 1} ~ lX, Z = reilJ ;
(2) I/(z) I ~ 1 for 1} = -lX and 1} = lX

(i.e. on the boundary of the sector considered). Then the inequality


I/(z) I < 1 holds in the entire sector.
Proof: Let A be a fixed number, 1 < A < 2:' Compare I(z) with
the function e,A where we choose the branch of zA that maps the positive
real axis onto itself. The "comparison function" e,A is regular in the
entire closed sector except at the origin z = 0 where ezA is continuous.
We have

on the two bounding rays 1} = -lX and 1} = lX because 0 < AlX < ~ .
On the circular arc Izi = r, -lX < 1} < lX, we have
IezA I = e,Acos},{} ? erAcosAc< .
(The comparison function almost satisfies the conditions of the theorem
but the conclusion is not valid. Imagine that A = 1 or that A = ~
2!X
- which values are, in fact, excluded; in the first forbidden limit case
condition (1) is satisfied, in the other condition (2), but in no case are
both conditions fulfilled and in no case is ezA bounded inside the sector
in question.)
Examine now the function I(z) e-ezA, where e> 0, at a certain inner
point Zo of the sector. Enclose the point Zo in the finite (circular) sector
bounded by the rays 1} = -lX, 1} = lX and the arc of the circle Iz I = r
between these rays; r is subject to the conditions
1

r> ( - - -)A-l 2B log A


e cos.:Lx ' r>J3'
By virtue of condition (2) we have on the rectilinear boundary of the
sector
I/(z) e- ..A I ~ 1· e-.,AcosAc< ~ 1.
By virtue of condition (1) and the inequalities eyA cos AlX > 2Br, eB , > A,
we have on the bounding arc Izl = r.
!/(z)e-ez AI < Ae B,e-.,A cOsA", < Ae- B, < 1.

yuana97@uw.edu
168 The Maximum Principle

The maximum principle together with these inequalities implies that at


the inner point Zo
A
I/(zo) e-·'0 I ::;; l.
This inequality holds for any e, however small, which concludes the
proof.
323. Condition (1) of 322 can be generalized insofar as the inequality
I/(z) I < AeB1 •1
need not be required to hold in the entire sector but only on the arcs of
the circles Izl = rp Izl = r2 , ••• , Izl = r", ... intercepted by the rays
{} = -IX, {} = IX, limr" = 00. The conclusion, namely I/(z) I < 1 in the
"-+00
entire sector, remains the same. What more general curves can replace the
circular arcs?
324. We modify condition (2) of 322 in the following way: There
r
exist in the sector -IX < {} ::;; IX two curves 1 and rs connecting the
points z = 0 and z = 00 that do not intersect and along which \f(z) I ::;; l.
This modified condition together with condition (1) as stated in 322
implies the inequality If(z) I < 1 in the domain bounded by 1 and r rs.
325. The function I(z) is assumed to be regular in the half-plane
ffiz > 0 and to satisfy the following conditions:
(1) there exist two constants A and B, A > 0, B > 0 such that in
the entire half-plane

(2) we have for r ~ 0


I/(ir) I ::;; 1, I/( -ir) I ::;; 1;

(3) li m sup log !f(y)


y
I ......
;;,.
0
,~+oo

Then I(z) is bounded by 1 in the entire half-plane:


I/(z) I ::;; 1 for 9lz ~ o.
Proof: In treating 322 we extracted the desired conclusion from
the maximum principle by introducing a variable parameter (the number
e); now we introduce two parameters. Assume 1J > o. By virtue of
condition (3) the function I/(r) e-'" I of the variable r converges to 0 as
r -+ 00; it reaches its maximum F" at a certain point ro, ro > O. If
ro = 0 the maximum is F" < 1 because of (2). We choose a fixed number
A, 1 < A< 2 (e.g. A = :) and study the analytic function
t(z) e-"ze- te - i )."/4.A

yuana97@uw.edu
Pt. III, Chap. 6, No. 323-327 169

in the sector 0 < {} < ~ , e > 0; take the branch of z" that maps the
positive real axis onto itself. We see that in the above mentioned sector

cos ().{) - "':) > cos (± "':) > O.

In the same way as in 322 we derive from (1), (2) and (3) that for
o S {} < ; the absolute value I/(z) e-'1 Z I cannot be larger than 1 or F'1
whichever is larger. The same can be said for the sector - ~ < {} < O.
We claim that Ffl S 1: If Ffl were larger than 1 we would have
I/(z) e- flz I <FfI on the entire half-plane and at the point t = ro the
maximum would be reached, I/(ro) e- flr , I = Ffl (see above). This is
impossible because the maximum cannot be attained at an inner point
z = ro' Hence Ffl < 1, consequently
ffiz> O.
This is valid for any 'YJ which concludes the proof. Notice that in condi-
tion (3) the positive real axis may be replaced ,by any ray from z = 0
that goes to infinity in the half-plane ffiz > O. Such a ray cuts the half-
plane into two sectors, both with an angle smaller than 11:: Only this
fact is essential [322, also 330].
326. Let the function I(z) be regular in the half-plane ffiz > 0 and
satisfy the following conditions:
(1) there exist two constants A and B, A > 0, B > 0, such that in
the entire half-plane
I/(z) I < AeB1z1 ;
(2) I(z) is bounded on the imaginary axis,
I/(ir) I < 1, I/( -ir) I < 1, r> 0;

(3) there exists an angle 1X, - ; < 1X < ~ such that

lim log If(ri"') I = _ 00.


'-)0-+ 00 r
Such a function must vanish identically. [Examine the function eOJZI(z).
OJ> 0.]
327. The function I(z) is supposed to be regular in the half-plane
ffiz > 0 and to satisfy the conditions:

yuana97@uw.edu
170 The Maximum Principle

(1) there exist two constants A and B, A > 0, B> 0, such that
on the entire half-plane
I/(z) I< AeB /./;
(2) there exist two constants C and y, C> 0, y> 0 such that for
1':2::0
1/(±i1') I s Ce-l".
A function satisfying these conditions must vanish identically. [Examine
the function I(z) e- IIJ10g('+1).]
328. The function sin :rr;z is the smallest function that is analytic for
ffiz :2:: 0 and that vanishes at the points z = 0, 1, 2, 3, ... More precisely,
the following proposition holds:
We assume that the function I(z) is analytic in the half-plane ffiz > 0
and that it satisfies the conditions:
(1) there exist two constants A, B, A > 0, B> 0, such that for
ffiz > 0
I/(z) I < Ae B /./;
(2) there exist two constants C and y, C > 0, y> 0 such that for
1'>0
1/(±i1') I < Ce(n-l')';
(3) I(z) has the zeros 0, 1, 2, ... , n, ...
Such a function vanishes identically.
329. Let w(x) be a positive function of the positive variable x that
increases with x and tends to + 00 as x increases to + 00. A function
I(z), regular in the half-plane ffiz ?: 0, that satisfies the inequality
I/(z) I > e"'(/'!)/'/ for ffiz ?: 0
does not exist.
330. Suppose that the function I(z) is regular at any finite point of
the sector IX <D< p, bounded by 1, I/(z) I < 1, on the two rays D = IX
and D = P and that, furthermore, there exists a positive constant 6
such that

I/(z) Iexp (-Iz III:" -,,)


is bounded for IX < D S p. Then I/(z) I < 1 at every inner point of the
sector IX < D < p. [Comparison function exp (zII:;-a).J
331. Let I(z) be regular in the sector IX < D < p. If I/(z) I S 1 on
the bounding rays D = IX and D = P and if there exists for every e > 0

yuana97@uw.edu
Pt. III, Chap. 6, No. 328-335 171
an ro such that
"
(J-a<
I/(re i ") I< e" for r > ro
in the above mentioned sector then the stronger inequality
I/(z) I < 1
holds in the entire sector. [Method of 325.]
332. The function g(z) is assumed to be an entire function, M(r) be
the maximum of Ig(z) I on the circle Iz I = r. If
lim log ~(r) = 0
,-+00 Vy
then g(z) cannot be bounded along any ray. [E.g. g(z) is not bounded
along the negative real axis.]
333. Suppose that the function I(z) is not a constant and that it is
regular in the half-strip @ defined by the inequalities

x> 0,

If there exist two constants A and a, A > 0, 0 < a < 1, such that in @

I/(x + iy) I < e Aeax ,

and if ,
I/(z) I < 1
on the boundary of @ (i.e. for x = 0, - ; < Y < ~- and for x ~ 0,

y = ± ;) then I(z) satisfies the strict inequality


I/(z) 1< 1
in the interior of @. [The comparison function is of the type eehz .]
334. Let w(x) have the same properties as in 329. Every function
I(z) that is regular in the half-strip

x> 0, z = x + iy
must satisfy the inequality
I/(x + iy) f < e",(x)e X

at least at one point z = x + iy of the half-strip.


335. The assumptions of 278 are weakened insofar as (3) is satisfied
in all but possibly finitely many boundary points Zl' zz, ... , z" of ~. An
other assumption, however, is added, namely that there exists a positive
number M' for which the inequality
I/(z) 1< M'

yuana97@uw.edu
172 The Maximum Principle

holds everywhere in m. (Only the case M' > M is interesting.) This


modification of the hypothesis does not change the conclusion of 278
that under those conditions If(z) I <M, If(z) 1< M respectively. [In
the case where the point at infinity belongs to mand all the boundary
m
points of lie in the disk Iz I < r we examine the comparison function
(2r)" n
"
v=l
(z - zv)-l.J

336. The domain '1) is supposed to lie in the half-plane ,3z > 0; the
boundary contains a finite number of segments of the real axis; Q denotes
the sum of the angles under which the segments are seen from an inner
point Cof '1). Assume that fez) is regular and single-valued in the interior
of '1) and continuous on the boundary of '1), that If(z) I < A at the real
boundary points, If(z) I < a on the remaining boundary of '1),0 < a < A.
Then
D D
- 1--
If(C)I<A"a " [57.]

337. The function fez) is supposed to be regular on that piece of the


Riemann surface of log z that covers the annulus 0 < Iz I :s;; 1. If fez)
is bounded and in particular if If(z) I :s;; 1 for Iz I = 1 then If(z) I < 1 in
the entire domain.
338. Let g(z) denote an entire function but not a constant; mbe a
connected region on the boundary of which (more exactly: at whose
boundary points different from z = 00) Ig(z) I = k and in the interior of
which Ig(z) I > k, k > O. Then the point z = 00 is necessarily a boundary
point of mand g(z) is not bounded in m.
339. Let Fl and F z be two continuous curves that have a common
starting point, extend to 00 and enclose together with z = 00 a certain
region m(e.g. two rays enclosing a sector). We assume that no point of
the negative real axis belongs to m.
The function fez) is supposed to be regular on Fv r z and in the enclosed
region; in addition lim fez) = 0 as z tends to 00 along r 1 and r z. If fez)
is bounded in mthen lim fez) = 0 as z goes to 00 along any path in m.
. log z
[Examme A-+e -1og-f(z).]
z
340. Let the curves r 1 and r z have the properties described in 339.
Let fez) be bounded and regular in the region between r 1 and z and r
assume, in addition, that lim fez) = a as z tends to 00 along r 1 and
lim fez) = b as z tends to 00 along r 2 • Then we have a = b. [Consider
a+ b)2 - (a- -2-b)2 .J
(fez) - --2-

yuana97@uw.edu
Solutions

Part One

Infinite Series and Infinite Sequences

*1. [Cf. HSI, pp. 238, 252-253, ex. 20.] AiOO = 292 [2].
*2. [For an intuitive solution see G. Polya:Amer. Math. Monthly Vol. 63,
pp. 689-697 (1956). Cf. MD, Vol. I, p. 97, ex. 3.84.]

+ C + C2 + C3 + ... + cx + ...)
00

1: AnC" = (1
(1 + t,5 + t,1O + C15 + ... + C5Y + ... )
(1 + C10 + C20 + t,30 + ... + C1O: + ... )
(1 + t,25 + C50 + C75 + ... + C25" + ... )
(1 + C50 + t,100 + t,150 + ... + t,50V + ...)
1

For the numerical computation of the coefficients An expand successively


the functions
(1 - t,)-I,

(1 - C)-1 (1 - C5)-I,
(1 - C)-1 (1 - t,5)-1 (1 - C1O)-I,
(1 - t,)-1 (1 - C5)-1 (1 - t,10)-1 (1 - t,25)-I,

(1- t,)-1 (1- t,5)-1 (1- C1O)-1 (1- C25 )-1 (1- C50 )-I.

It is convenient to dispose the coefficients needed for the computation


of A 100 in a rectangular array.

yuana97@uw.edu
174 Infinite Series and Infinite Sequen;;es

*3. B5 = 15 [4].
4. The coefficient of (n in the expansion of
(( + (2 + (3 + (4)'
is equal to the number of sums of value n with s terms of value 1, 2, 3, 4,
where the order of the terms is taken into account. Therefore we have
=
1 + 2; Bi;" = 1 + (( + (2 + C3 + (4) + (( + (2 + (3 + ~)2 + ...
"~l

1-C-C2 __ '3-C"
For the numerical computation use the relation

B .. = B"_l + B"_2 + B .. _ 3 + B"_4'


which follows from the definition of Bn or from the above equation.
5. 4 = C78 [7].
6. 20 = D78 [8].
99
7. 2; C,,(" = (1 + ()2 (1 + (2) (1 + (5) (1 + (10)2 (1 + (20) (1 + (50).
99
8. 2; D"C" = ((-1 + 1 + ()2 ((-2 + 1 + (2) ((-5 + 1 + (5)
((-10 + 1 + (10)2 ((-20 + 1 + (20) ((-50 + 1 + (50).
,,~-99

9. [Cf. Euler: Introductio in Analysin infinitorum, Chap. 16, De


Partitione Numerorum; Opera Omnia, Ser. 1, Vol. 8, pp.313-338.
Leipzig and Berlin: B. G. Teubner 1922; also e.g. W. Ahrens: Mathema-
tische Unterhaltungen und Spiele, 2nd Ed., Vol. 1, pp. 88-98, Vol. 2,
p. 329. Leipzig: B. G. Teubner 1910, 1918.] The "change problem" [2]:
1
ca,) ••• (1 -
00

(1 - C ')
a
(1 - C)
aj = 2; A,,(".
,,~O

An denotes the number of non-negative integral solutions of the Diophan-


tine equation
a1x 1 + a 2x 2 + ... + a/x/ = n.

The "postage stamp problem" [4]:

_ _ _ . __
1 -
1 _._____
Ca , - Ca , -- ••• - Cal
= 1: B r"
.. ~O "".

The "first weighing problem" [5] (all the weights on one pan):

(1 + ca') (1 + ca') ... (1 + cal) = 2; e"C.


,,~O

yuana97@uw.edu
Pt. I, Solutions 3-13 175

The "second weighing problem" [6J (weights may be placed on both


pans) :
~

(C- a, +1 + Ca,) (C- a• + 1 + ca') ... (C- al + 1 + Cal) = ~ DnC".


11=-00

10. This problem is equivalent to the following: We have to weigh


on one pan of the scales an object of n units with p different weights of
one unit. According to 9 the number C" of the different possibilities is
the coefficient of en in the expansion of

(1 + C)P = 1 + (;) C+ (~) C2 + ... + (~) C" + ... + CP,


therefore:
C -(P) -
,,- n -
P!
n \ (n - P) ! •
In abstract terms: The number of different subsets of n elements
contained in a set of p elements is (~).
11. This problem is equivalent to the following: Someone owns
quarters minted in p different years. In how many different ways can
he payout n quarters? According to 9 the number of different ways is
An' the coefficient of Cn in the expansion of

-~--
(1- C)p
= 1 +(-P)
1
(-C) + ... + (-P)(-Ct + "',
n
thus
A = P(P+1) ... (p+n-1)=(p+n-1).
n 1.2 ... n p-1
12. According to 11 this number is

(p + - P) - 1) = (np-1
(np-1 - 1).
The result follows also directly from the expression (C + C2 + .. .)P.
*13. Identical with 11. Another solution: Consider the P-fold series

1: x~'x;·"· x;P = (1 - X1)-1 (1 - X,z}-l ... (1 - Xp)-l


V1J"I.· .• '''p =0,1 ,2,3, ...

and identify the x;'s with C.


Intuitive solution: Consider n + p - 1 places in a row. At the
left hand end there are a certain number of places filled with Xl' then a
place filled with a multiplication point, then a certain number of places
filled with x 2, then a multiplication point and so on, as shown:

yuana97@uw.edu
176 Infinite Series and Infinite Sequences

We have to choose P- 1 among the n + P- 1 places for the P- 1


multiplication points which can be done in (n ; ~ ; 1) different ways,
by 10. ("Combination with repetition of P different elements taken n
at a time" is the traditional term. In some of the cases that must be
admitted the multiplication points are placed in an unorthodox way.)
14. According to the first weighing problem [9 extended to infinitely
many weights we have to consider
(1 + C) (1 + C)2 (1 + ~1"4) (1 + ~1"8) ... = 1-c- 1-C' 1-CS
- _ . - - . - _ . - - ...
l-C 1-C· 1-C'
1-Cl s
1-CB
1
=--= 1
1- C +C +C2 +CS + ....
Cf. 16, 17.
15.
(C- 1 + 1 + C) (C- 3 + 1 + CSl ... (C- 3" + 1 + C3")
3 9 3"+1 3,,+1
= C-1 ~ C-3 ~ ..• C-3" C - 1 = C- N C - 1
C- 1 C3 - 1 C3" _ 1 C- 1

3,,+1 - 1
= C- N + C- N+1 + ... + CN- 1 + CN, N = 2 .

16. a" = aF ", where F" is the number of digits 1 in the binary repre-
sentation of n (its expansion in powers of 2).
17. [E. Catalan, Problem: Nouv. Corresp. Math. Vol. 6, p. 143
(1880). Solved by E. Cesaro: Nouv. Corresp. Math. Vol. 6, p. 276 (1880).]
The series in question results from the expansion into a power series of

on setting C= 1. To determine the sign of a coefficient it is sufficient to


examine the case where a = b = ... = 1. According to 16 the sign is
given by (_1)F" where F" denotes the number of ones in the binary
expansion of n.
18.
1 - CIO 1 - ClOO 1 - C1000 1
1="1". 1 - CIO·t= CIOO ••• = 1 - C·

This problem is not contained in 9. The result, however, is well known:


Any positive integer admits a unique representation in the decimal
notation. Cf. 14.
18.1. There are I kinds of coins and we have a limited number of
each kind, PI coins of the first kind, each worth a1 cents, etc. In how
many ways can we pay n cents?

yuana97@uw.edu
Pt. I, Solutions 14-19 177

Given the positive integers

Pi' PZ' ... , PI'


find E", the number of solutions of the equation
alx l + a2x 2 + ... + alxl = n

in integers Xl' X 2 ' •.. , X" subject to the condition


o< Xl :s PI' 0< X2 < Pz, "',

By the method of solution 2


~ 1 _ xa,(P.+I) 1 _ xa,(P,+I) 1 _ xa/(PJ + 1)
S E x"= -1-- -
"=0 " x a,
- ----
1 _ X4, 1 - x 41

Particular cases:
E" = e" when PI = P2 = ... = PI = 1,
E" = A" when PI = P2 = ... = PI = 00,
E a +" = D" when PI = P2 = ... = 2 and a l + az + ... + al = a.
To encompass 18 we admit 1 = 00 (properly interpreted). We have
PI = P2 = ... = 9, al = 1, a2 = 10, a3 = 100, ... in the case 18.
18.2. Particular case of 18.1; 1= 3, a l = a2 = a3= 1, PI = P2 = P3 = n,

E 2"+1 = en: 3) _ 3 (n ~ 2) = (n ~ 1).


*19. [Euler, l.c. 9.] First solution: According to solution 14 we
have
(1 + n (1 + C2 ) (1 + C' ) (1 + C8 ) '" = 1 ~ C'
(1 + C3 ) (1 + C6 ) (1 + C12 ) (1 + C24 ) ••• = 1 ~ C 8 '

(1 + C5 ) (1 + C1O ) (1 + C20 ) (1 + C'0) ••• = 1 ~ C 6' etc.


Second solution:

n (1 + C) (1
~

K(C) = ~ C2"-1)
"=1
is invariant under the substitution of C2 for Cbecause
1 - C4"-2 = (1 + C2"-I) (1- C2"-I),
i.e. [ICI<l]
K(C) = KW) = K(~) = K(C8 ) = ... = K(O) = 1.

yuana97@uw.edu
178 Infinite Series and Infinite Sequences

Third solution:

n (1 + C) = n n
00 " 00 (1 _ C2") 00 1
(1 _ C") = (1 _ ,.2"-1) .
,,=1 ,,=1 ,,=1 ..

20. [Euler, I.e. 9.] Interpret the coefficients in the expansion of the
functions in 19. The result states that the first weighing problem with all
the integers as weights admits as many solutions as the change problem
with the odd integers as coins.
21. If we omit the restriction that the terms in the sum have to be
smaller than n, i.e. if we admit also the representation n = n, then we
are dealing with the postage stamp problem [9]. The number of different
sums is equal to the coefficient of in en
1
l-C-CI- ... -C'"
or in
1 1 1- C
1- C- ca - ca - ... = C = 1 _ 2C = (1 - C)(l + 2C + 4C2 + ...)
1---
1-C
= 1 + C+ 2C2 + 4C3 + ... + 2"-1C" + .. ".
Intuitive solution: The interval 0 ~ x <n appears as a sum
of subintervals of integral lengths if some of the n -1 points 1, 2, 3, ... , n-1
are chosen as points of division. For each of these points there are two
possibilities, to be chosen or not, independently of the other n - 2
points, and so the total number of possible choices is 2"-1. [MD, Vol. 2,
p. 189, problem 3.40.1.]
22. [E. Catalan, Problem: Mathesis (Gand) Vol. 2, p. 158 (1882).
Solved by E. Cesaro: Mathesis (Gand) Vol. 3, p. 87 (1883).] The total
number of solutions is equal to the coefficient of en
in the expansion of

23. [E. Catalan: Nouv. AnnIs Math. Ser. 3, Vol. 1, p.528 (1882).
Solved by E. Cesaro: Nouv. Annis Math. Ser. 3, Vol. 2, p. 380 (1883).]
The number of solutions is equal to the coefficient of C" -1 in the expansion

yuana97@uw.edu
Pt. I. Solutions 20-25 179
of

We have
v C
2: = r(l) C+ r(2) C+ .,. + r(v) CV + "',
00

--v
C
v=1 1 -

where r(v) denotes the number of divisors of v [VIII 74J.


24. [E. Cesaro, Problem: Mathesis (Gand) Vol. 2, p.208 (1882).J
The number of solutions is equal to the absolute term (C-free) of the
infinite sum

2:
00

"v,
]"v'-(2.+1)"
(v+1)' = 2:
00

(C-(2v+1)" +C-(2V+1)("-1) +C-(2'+1)("-2) + '"


v=1 (1 - C )(1 - C ) .=1

+ 1 + C2'+1 + C2(2'+1) + ...)(_1_V _ 1 ).


1- C' 1- C(v+1)'

It is therefore sufficient to show that the absolute term of the following


sum is equal to 1 for k 2: 1:
1 1) ]"-(2v+1)k _ ]"-(2v-1)k
+ 1.]"-k
, - C·
=-
00 ( 00
"" C-(2V+1)k _ '" ., .,
ki 1 _ ]".' - -1 ]"(0-1)' - ki 1 _ ]"v'
.=1 .,., .=1 .,
A multiple of is, however, equal to (2v + 1) k or to (2v - 1) k if and
v2
only if v2 is a factor of k. Therefore the absolute term of the sum on the
right hand side vanishes, which concludes the proof.
25. [Cf. G. H. Hardy: Some Famous Problems of the Theory of
Numbers and in Particular Waring's Ploblem. Oxford 1920, pp. 9, 10.]
"Change problem" 9; put w = e2 :ni/3:
1

It is
II - .~
72
+ ~=-~)~
8
+ ~9 cos~!I~-1
3
< 32 < ~
= 72 2 .

yuana97@uw.edu
180 Infinite Series and Infinite Sequences

*26. [Proposition by P. Paoli; d. IntermCd. math. Vol. 1, p. 247-


248 (1894). Ch. Hermite, Problem: Nouv. AnnIs Math. Ser. 1, Vol. 17,
p. 32 (1858). Solved by L. Rassicod: Nouv. AnnIs Math. Ser. 1, Vol. 17,
p. 126-130 (1858).J Cf. the more general 27.1 and the more detailed 27.2.
27. [Cf. Laguerre: Oeuvres, Vol. 1. Paris: Gauthier-Villars 1898,
pp. 218-220.J A decomposition into partial fractions similar to the one
00

in 25 furnishes for ~ A"e" = (1 - C·,)-1 (1 - C·,}-l ... (1 _ Cal}-l

the "principal term"

ala;-:-:-.~ ~~d'
Since aI' a2 , ••• , al do not have a common factor the denominator of the
other terms is of degree l - 1 at most. The statement follows from this
[solution III 242]. Cf. 28. Notice also that the I-dimensional volume
characterized by the inequalities
Xl > 0, x2 > 0, xa ~ 0, ... , Xl > 0, alx l + aZx2 + aaxa + ... + alxl < n
in I-dimensional space is
1 n n n
__ e _ " _ o •• _
and

27.1. [Cf. E. Netto: Lehrbuch der Combinatorik, 2nd Ed. Leipzig


and Berlin: B. G. Teubner 1927, pp. 319-320.J We assert that
(*)
(1 - ~') (1 -
1
~') ... (1 -
_ R (_1_)
~l) - 1- ~
+ _~(£)_
1 _ ~a,., ...al

where R(x} and Sex} are polynomials, of degree l and smaller than
a 1 a2 ... ai' respectively.
By our assumption concerning the a,'s the denominator of the left
hand side of C*} is the product of (1 - ~}I by a polynomial which has no
multiple roots and is, therefore, a divisor of 1 - ~.,a'''··l. Based on this
fact the decomposition in partial fractions yields (*). The expansions of
the two terms on the right hand side of (*) yield the two terms into which
An is decomposed, d. III 242 and VIII 158, respectively.
27.2. We cUllsider various values of n one after the other.
(1) We consider the case n < abo Let x', y', x", y" be non-negative
integers such that
ax' + by' = n, ax" by" = n. +
Then
o ~ x' < b, 0 < x" < b,
a(x' - x") = -bey' - y")

yuana97@uw.edu
Pt. I, Solutions 26-27.2 181
and so x' - x" is divisible by b. Yet
-b < x' - x" < b
and so
x' - x" = O.
That is, the number of solutions is An < 2.
(2) By a slight modification of the foregoing AAI> < 2. Yet
a .b+b . 0 = a . 0 + b . a=ab
and so AAI> = 2.
(3) By the argument of the solution 27.1
(1 - ~b) (1 - ~) = T(~)
(1 - ~a) (1 - t)
is a polynomial
T(~) = ~b-a-b+l + ... + 1,
ab
T(l) = ab = 1.
Hence
T(';) - T(1) = _~b-a-b + ...
1 - ~
is also a polynomial. Now
~ ~

~ (An - An_AI» ~ = (1 - ~) ~ An~"


n=O
T(~)
= 1-~

= T(~) - 2J.!l. + _1_


1-~ 1-~

+
~

= ... _ ~-a-b ~ ~"


and so n=O

A ab - a- b = 0,
An = A n- ab + 1 >1. when n> ab - a-b.
(4) By the last result

An = A n- [~lab + [:bJ .
Yet
n- [~J ab < n - (:b - 1) ab = ab
and so, by (1),
A [n 1 =
n- - ab
0 or 1, which proves 26.
ab

yuana97@uw.edu
182 Infinite Series and Infinite Sequences

28. Put P = 3 in 11 and 12, respectively.


29. Let k be a non-negative integer. The number of solutions of
IXl I + Ix2 1+ ... + Ixp I = k is equal to the coefficient ak of Ck in the
expansion of
(1 i; akCk.
+ 2C -I- 2C2+ 2C3 + .. .)P = (11 +- CC)P = k=O
The number of lattice points in the octahedron is therefore equal to
ao + al + a2 + ... + an'
the coefficient of Cn in the expansion of
(1 + C)p (2C + 1 - C)p
(l-C)P+I (1- C)P+I

= (2C)P(1-C)-P-I + (;) (2C1- 1 (l-C)-P + (~) (2C)P-2(1-C)-P+I + .. ',


i.e. equal to

2P G) + 2t -1 (; )(p: 1) + 2t - 2(~ )(p : 2) + ... + l.


30. [G. P6lya: Math. Ann. Vol. 74, p.204 (1913).J The number of
lattice points is equal to the sum of the coefficients of C-s, C-s+1, ... ,
1, ... , CS- 1 , CS in the expansion of
(C- n + C- n+1 + ... + C- 1 + 1 + C+ ... + C- 1 + Cn)3.
k
The following relation between the series L: aJ: and its coefficients
a. holds in general: .=-k

1J2"( L: a.C· C-' dt


2n
k )
= a"
o .=-k
and
2m+l 2m+l . 2m +1
tn C--2-_ C2 - sm--2- t
L:
'=-ffl
r= C
-t
- C
t -
. 2t
sm

31. [Cf. Ch. Hermite, Problem: Nouv. AnnIs Math. Ser. 2, Vol. 7,
p. 335 (1868). Solved by V. Schlegel: Nouv. AnnIs Math. Ser. 2, Vol. 8,
p.91 (1869).J Since z = n - x - y we have X + y < n, x> 0, y> 0;
also X < 12 - X, Y - X < n - x - y ;£ x + y.
Consequently the number of solutions in question is equal to the
number of solutions of the inequalities

~-x<y<~
2 = = 2'
y> 0, x + y < n.

yuana97@uw.edu
Pt. I. Solutions 28-38 183

31.1. (1) You are one of n persons forming an assembly which elects
a committee consisting of r of its members. There are (;) possible
committees [10J. namely (: ~ ;) to which you belong and (n ~ 1) to
which you do not belong.
(2) (1 + xt
= (1 + xt-
1 (1 + x).
Use the binomial theorem and
then consider the coefficients of x' on both sides.
(3) Straightforward verification if you use the usual expression of the
binomial coefficients.
31.2. Pass from r - 1 to r by mathematical induction; use 31.1.
32. Compare the coefficients of zn in the identity

(1 + zt (1 + zt = (1 + z}2n.
33. Compare the coefficients of z2n in the identity

34. Evident.
34.1. From

there follows

[34J. See also VII 54.1.


35. We have

therefore
~ nih x
2:~~
n=O n.
= (1 + hZ)h.
Apply 34.
36. Cf.35.
37. Differentiate the identity

1 - (1- xt = C)x - (;) x2 + ... + (-1t- 1 (: )xn


and put x = 1.
38. [Problem from Ed. Times. Cf. Mathesis (Gand) Ser. 2. Vol. 1.
p. 104 (1891). Solved by Greenstreet et al.: Mathesis (Gand) Ser. 2.

yuana97@uw.edu
184 Infinite Series and Infinite Sequences

Vol. 1, p. 236 (1891).J The sum in question is equal to

I
1
1 - (1 - x)"
x
dx = Ii -1

1 - x
x" dx = 1(1 + x + x2 + .. , + X"-I) dx
1

o 0 0
1 1 1
=1+2"+3"+"'+-;;'
39. The general term of the left hand side sum is the coefficient of
x 2n +1 in !(1 + 2xt+k+l {_x2 )n-k. Therefore we have to consider the
coefficient of x 2n +1 in
"
2:! (1 + 2X)"+k+1 (_X 2)"-k = ! (1 + 2xt+1 (1
+ 2x )"+1 - 2( -x2)"+1
__ _
k=O (1 + x)

or in the power series of 1(1 + 2x)2n+2 (1 + x) -2. Division leads to a


polynomial of degree 2n, the remainder is HI - (2n + 2) (2x + 2)].
The coefficient of x 2"+1 in
1 1 2n + 2
2(1 + X)2 - 1+x

is equal to -1(2n + 2) + 2n + 2 = n + 1.
40. Decompose the sum into three terms according to
(v - nIX)2 =n2IX2 - (2nIX - 1) v + v(v - 1).
The formula

and its first and second derivatives with respect to p furnish the three
terms if p is replaced by x and q by 1 - x. Cf. II 144.
41. It is sufficient to consider n > p and
!p(x) = xIx - 1) .. ~~x - P+ 1) =(;), 1p(x) =
1
2P
(XP )-
\
'vVe have now

.t (:) G) = p! (:I~p)!.~ (: =:) = (;) 2n - p = 2"1p(n) ,

::P~ (_1).(n)(v) __ n!_ ~'(_l).(n-p)_{


v p -p!(n-p)!.-:p v-p -
Oforn>p,
(-1tforn=p.
42. This is a special case of 40 with x = IX = 1/2. It is also a special
case of 41 with
!p{x) = (2x- n)2 = n2 - 4xn + 4x2 = n 2 - 4(n - 1) x + 4x (x - 1)
1p(x) = n 2 -2(n - 1) x + x(x - 1).
We have 1p(n) = n.

yuana97@uw.edu
Pt. I, Solutions 39- 47 185
43. Special case of 41 for l]1(x) = (2x - n)2 [42]. We obtain a" =
* 2, a" 4 for n 2.
0
for n = =
43.1. (1) First multiply both sides by ~, then use 34 and 38.
(2) Let y denote either the left hand, or the right hand, side of the
desired identity and verify that, in both cases, y = 0 for x = 0 and
dy 1 - e-~
dx=--x-

44. Write I(x) = c,,(x - Xl) (x - x2 ) '" (x - x,,). We have

(z ! - x.)1 = (k - x.) I.
45. [G. Darboux, Problem: Nouv. Annis Math. Ser. 2, Vol. 7, p. 138
(1868).] We deduce from 44

i
k=O kl
f(k) 1 = I(z~) e' = e'g(z) ,
dz

where g(z) is a polynomial with integral coefficients.


46. [E. Cesaro: Elementares Lehrbuch del' algebraischen Analysis
und der Infinitesimalrechnung. Leipzig and Berlin: B. G. Teubner· 1904,
p. 872.] The functions 1"+1 and I" are connected by the recursion formula

1"+1(z) = z[f..(z) (1- z) + (n + 1) I..(z)].


Therefore the coefficients of I,,(z) = ai")z + a~")z2 + ... + a~)z" are
linked by the relations

11 = 1, 2, ... , n + 1 ·, aoe,,) = a"+l


e,,) = 0 .

This together with 11(z) = z concludes the proof. The value of 1,,(1) is
determined by
1"+1 (1) = (n + 1) 1..(1).
47. [Cf. N. H. Abel: Oeuvres, Vol. 2, Nouvelle edition. Christiania:
Gr0ndahl & Son 1881, p. 14.] If g(x) is constant then the proposition is
a consequence of 44. In assuming that the proposition holds for poly-
nomials of a degree smaller than the degree of g(x) we write

g(z ~) y = g1(Z ~) [(z ~. - x1)yJ


= g1 (z ~) (~~~) + ~(ti) z + ::~~) Z2 + ...) = I (z ~) 1 ~ z'

yuana97@uw.edu
186 Infinite Series and Infinite Sequences

The given differential equation is soluble by successive quadratures


because this is so for the equation

(z ~ - xo),.. = zy' - xoY = tp(z).


48. According to 44 both sides are equal to
1(1) z +/(1) /(2) z2 +/(1) /(2) /(3) z3 + ... + /(1) /(2) ••• /(n - 1) /(n) z" + ...
g(1) g(1) g(2) g(1) g(2) ••• g(n - 1) •

49. Apply 48 putting I(x) = (x - .)2, g(x) = x2 •


50. Comparing the coefficients of z" on both sides of the functional
equation we get
A,,(q" - 1) = A,,_lq", n = 1 2, 3, ... ; Ao = 1,
hence
"("+1)
q-2-
A" = 2 ' n = 1, 2, 3, ...
(q - 1) (q - 1) ••• (q" - 1)
51. According to the functional equation (50]
B,,(l- q") = B,,_lq, n = 1, 2, 3, ... ; Bo = 1,
therefore
q"
B,,= (1 2 ' n = 1. 2,3, ...
- q) (1 - q ) ••• (1 - q")

52. [Cf. R. Appell and E. Lacour: Principes de la theorie des fonctions


elliptiques et applications. Paris: Gauthier-Villars 1897, p.398. For
closely similar preceding work of Gauss see I.c. solution 55.] Calltp.. (z)
the expression in question:
2 _ 1 + r"+1 z
tp,,(q z) -tp,,(z)-+2il.
qz q
From this identity follows
Cpq2~+1(1 - t,,-2p) = Cp+1(l - t,,+2p+2), 'P = 0, 1, ... , n - 1,
C,,=q, ,,'
I.e.
(1 - r"+2 +2) (1 _l"+2 +4) ••• (1 _
P P q4") p'
C = 2 4 Q 2 q, 'P = 0, 1, ... , n - 1.
• (1 - q ) (1 - q ) ••• (1 - q~"- p)

53. [Jacobi: Fundamenta nova theoriae functionum ellipticarum,


§64,; Werke, Vol. 1. Berlin: G.Reimer 1881, p.234.J Take the limit
n -+ 00 in 52. [181.]
54. [Euler: Commentationes arithmeticae, Vol. 1; Opera Omnia,
Ser. 1, Vol. 2. Leipzig and Berlin: B. G. Teubner 1915, pp.249-250.]
Special case of 53: q I q3/2, Z = _qI/2.

yuana97@uw.edu
Pt. I. Solutions 48-59 187

55. [Gauss: Summatio quarundam serierum singularium, Opera,


Vol. 2. Gottingen: Ges. d. Wiss. 1863, pp.9-45.J Special case of 53:
replace q by ql/2 and z by ql/2, and apply 19 or the procedure of the third
solution of 19.
56. [Jacobi, l.c. 53, § 66; Werke, Vol. 1, p.237.J Put z = -1 in
53 and use 19.
57. Setting _qnz = an we obtain
= n
1 +G(z) -G(qz) = 1 + 1: -L:,. (1- qz) (1 - q2 Z) ••• (1 - qn-1z) (qn - 1)
n~11 - q

= 1 + a1 + a2(1 + a1) + a3 (1 + aI) (1 + a2)


+ a4 (1 + a1 ) (1 + a2 ) (1 + a3 ) + ...
= (1 + a1 ) (1 + a2) + a3(1 + all (1 + a2)
+ a4 (1 + a1 ) (1 + a2) (1 + a3) + ...

= (1 + a1 ) (1 + 2 ) (1 + a3 )
tl

+ a4 (1 + a1 ) (1 + a2 ) (1 + a3 ) + "', etc.
58.
• q q2 q3 qn
Do = G(O) =-+--
1 - q 1 - q2 +--+
1 - q3 ... +---
1 _ qn + ... ;

applying 50 and 57 we find


=
G(z) - G(qz) = 1: Anzn , G(qz) - G(q2Z) = 2: Anqnzn,
n~l n~l

=
G(tz) - G(q3Z) = 2: A nq2H zn, ... ,
n~I

hence addition of the first m equations and taking the limit m -? ex:>
yield
= A
G(z) - G(O) = 1: ~zn.
n~11 - q

59. Obviollsly G(l) = O. The functional equation in 57 implies with


the help of complete induction that

C(q-n) 1: -q-k
= k~11
n k (1 -- 1n) (1 -- qn) '" (1 - ~
k'-l) = -n, q-1 = a.
- q q q q

yuana97@uw.edu
188 Infinite Series and Infinite Sequences

Introducing (1 - q) n = y, 1 - L
n
= q we get

Let n ~ 00 for y fixed, and so q ~ 1:


00 1
J; - (1 - eY)k = -y. [181.J
1<=1 k

60. If we regard as known the sum of the series (cf. 59) we have
1 1 +z
I(z) =-= 2z log 1-=Z
and the solution is straightforward. Without supposing such knowledge
we equate the coefficients of Z2" on both sides and so obtain
" 22k ( + k) 1
k~ (-1)"-1< 2k + 1 : - k = 2n + 1.
To verify this observe that
2n ~ (n + k) = (n + k +
2k+1 n-k 2k+1
1) + (n2k+1
+ k)
and apply 39.
80.1. If 1 and m are positive integers
1 - q' 1 + q + ... + l-1
1= q'" = 1 + q + ... + q"'-l .
60.2. Obvious in the Gaussian as in the ordinary case.
60.3. Define I(x) as the left hand side of the desired identity. Then
(1 + x) ICqx) = I(x) (1 + q"x) .
Proceed hence as in 50,51, and 52.
60.4. Multiply the identity 60.3 by 1 + q"x and compare coefficients.
Or verify directly from the initial definition.
60.5. Obvious for k = 0 and k = n. For 0 < k < n from 60.4 by
mathematical induction except the last clause about symmetry for
which use the initial definition and substitute q-l for q.
60.6. Start from 60.3, substitute first
2n for n, q2 for q,
then
zq-2 n + 1 for x

yuana97@uw.edu
Pt. I, Solutions 60-61 189

and multiply finally by


3 2,,-1
q"'Z-" =!L. 'L ... -q--
z z z
The result is, in fact, identical with 52.
60.7. According as n = 2m or n = 2m + 1, the left hand side in
60.3 reduces for q = -1 to
(1 - x2)m or (1 - x2)m (1 + x).
60.8. By virtue of 60.4

,,-1
= J; (_l}",,[n - 1J + J; (_l)k [n =: 1J q"-k
"

k=O k k=1 k 1

=
,,-1
J; (_l)k (1- q,,-k-1) [n -lJ
k=O k

[n - 2J
,,-2
= J; (_l)k (1 _ q"-l) Q.E.D.
k=O k

60.9. [See e.g. MD, Vol. 1, pp. 68 -78. J Obvious when r = 0 or s = O.


Use this remark, the relation

and mathematical induction to pass from n to n + r.


60.10. [G. P6lya: J. Combinatorial Theory, Vol. 6, p.105 (1969).J
Fairly obvious when IX = 0 or IX = r(n - r). Use these cases, 60.4, and
mathematical induction.
60.11. Cut the area under the zig-zag path into vertical strips of
width 1 by equidistant parallels to the y-axis. The top of such a strip is
an x-segment (unit segment parallel to the x-axis) of the path and the
area of the strip is the number of y-segments of the path preceding
(forming an inversion with) that top x-segment.
The "Gaussian" analogues of polynomial coefficients can be used to
determine the number of certain more general letter sequences having
a given number of inversions. (Cf. G. P6lya: Proceedings of the Second
Chapel Hill Conference on Combinatorial Mathematics and its Appli-
cations: Chapel Hill N. C. 1970, pp. 381-384.]
61. Follows from the definition; for the product use 34.

yuana97@uw.edu
190 Infinite Series and Infinite Sequences

62. If al' a2 , ••• , a,. are positive we have


1 + a.z ~ ea•• , 'JI = 1, 2, ... , n.
According to 61
(1 + a1z) (1 + a2z) ... (1 + a,.z) ~ e("I+".+···+a,.)•.
From this the particular case a1 = a2 = ... = a,. = ~n follows.
63. A (z) ~ P(z) implies

J•A(z) dz ~ J•P(z) dz and ~(.) ~ eP (,).


o 0
Therefore
/,(z) 1 2
I(zf -8" ~ 1- z
leads to
j(z)
log -z- ~ log (1
1
_ Z)2'
I(z) ~ _ 1 _ = i; ftZ,.-l •
z (1 - Z)2 ,.=1

64. a) It is obvious from the combinatorial meaning of A,., B,., C,.,


defined in 9, that
0< C,. <A,.=::;: B,..
b) The first part of the statement follows immediately from
1 +z"~ _1_=
1 - za
1 +z" +z2a + ... ,
where we put a = aI' a2, ... , a, and multiply [61J. The second part
follows from
(1 - Z"I - z'" - ••• - z"",-l) (1 - z"1II)
1 - za, _ z'" - ••• - z"",-l - z"",

for m = 2, 3, ... , l; multiplication yields


(1 - zal) (1 - z"') ••• (1 _ z"l)
-----'~---,--'-_'__--:--c- ~ 1.
1 - Z"I - z'" - ••• - z"l

Multiply both sides by [(1 - z"1) (1 - z"o) ••• (1 - z"1)r1.


64.1. We shall use the notation

{ -Zlll1 } =-+-+-+
Cm
zm
cm+1
zlll+l
c",+2
zlll+2
...
for any expansion in negative powers in which the coefficient of Z-k
vanishes for k = 0, 1, 2, ... , m - 1. We shall also extend the concept

yuana97@uw.edu
Pt. I, Solutions 62-68.1 191
of a majorant series and the use of the symbol ~ to expansions in positive
and negative powers.
z,. + ~z"-1 + a2z"-2 + ... + a,. = (z - Zl) (z - Z2) ••• (z - z,,)

= z" eXP(lOg(l- :1) +10g(1- ;) + ... + log (1- :"))

~
15"+11 + ... )
z" exp ( -1 + - 1 + '" + - 1 + ---'--..,--:-
Z 2Z2 nzn (n + 1) z"+l

=1~"1 exp(1+t"~1})=Z"+z"-1+z"-2+"'+1+{;}
z
and so lak I < 1 for k = 1, 2, ... , n; apply III 21.
64.2. For the polynomial

_z _--=-
,,+1
z- 1
1
= z" + Z"-l + .. , + 1
ak =1 and Sk = -1 for k = 1, 2, ... , n.
65. Obvious.
66. Assume s" = 0 for n =1= v, Sv = 1, therefore t.. = PIS. for n ~ v.
If lim tn = lim SIS holds for this special sequence then lim P". = 0, i.e.
n-+oo "-+00 "-+00
the condition is necessary. Suppose the condition is satisfied. Let e be
any positive number. Find N such that Is" - S I< ; for all n > N; in
addition n has to be such that Pno' PIll' ... , PnN are all smaller than
4(N : 1) M' where M denotes the maximum of Is. ,. From

tIl - s = Pno(so - s) + Pnl(sl - s) + ... + Pnn(sn - s)


we now deduce

Itn - s, < (N + 1) 2M 4(N : 1) M

+ ; (P",N+1 + P.. ,N+2 + ... + Pm,) < ~ + ; .

67. Special case of 66: Pnv = -~-1 .


n+
68. Equivalent to 67: log P.. = Sw
. atl
68.1. EqUIvalent to 68, PI! =-- -.
an _ 1
This reformulation is useful in
dealing with power series.

yuana97@uw.edu
192 Infinite Series and Infinite Sequences

69. Special case of 88: put

Po = 1, PI = (~ y, 1'2 = (:r. ... , P. =(n~ 1)",


then we have
(n + 1)·+1
1'01'11'2 ..• P. = (n + 1) I •

70. Special case of 86: put


b.
1'.... = bo + bl + b2 + ... + b.'
observe

71. Special case of 70: a. = (n + 1)"'-1, b. = (n + 1)'" - n"'. Since

Iim (n + 1)'" - n'" _


-
Ii (1
m
+ lIn)'" - 1'" _ ( d "')
- -x
_
-~,
.-+'" n",-1 .-+'" lIn dx %=1

we obtain
lim 1",-1 + 2",-1 + ... + (n + 1),"-1 _ Ii (n + 1)"'-1 _ ~
- m - .
• -+'" (n + 1)'" .-+'" (n + 1)'" - n'" IX

72. [For 72-74 cf. N. E. Norlund: Lunds Universitets Arsskrift,


N. F. Avd. 2, Vol. 16, No.3 (1919).] Special case of 86:
l' = P.-. :s;;: p.-•
•, Po + PI + P2 + ... + p. - Po + PI + P2 + ... + P.-.
73. Special case of 86. Set
Po + PI + 1'2 + ... + p. = p., qo + ql + q2 + ... + q. = Q.,
1"0 + 1"1 + 1"2 + ... + 1". = R.,
and
l' = Ji._.
Q. :s;;: P.- o •
• 0 Po Q. + PI Q.-l + ... + P. Qo - Po + PI + ... + P._.
We obtain [cf.74]
Y. Poq. + PI q.-l + ... + P.qo • qo ql q.
R. - POQ.+P 1Q.-l +"'+P.Qo =PnOQo +p.I Q1 + ... +p•• Q.'
74. We write
soq. + SI q.-l + ... + s.qo
q. = qo + ql + ... + q. '

yuana97@uw.edu
Pt. I, Solutions 69-77 193
(r" as in 73). Then

t
P"Qoqo
==~~~~~~~~--~~--
+ P"-lQ1q1 + ... + PoQ"q"
" P"Qo + P,,-lQ1 + ... + PoQ"
q"Pot.Jo + q,,-l P 1t.J1 + ... + qoP"t.J"
q"Po + q.. -1 P 1 + ... + qoP"
consequently [66, 73] lim ~,,== lim q" == lim t". To establish the above
n~oo "-+00 n-+oo
identities we use the power series whose coefficients are the sequences in
question [34]:
00 00 00 00 00 00

~ i' ~ p,i == ~ P"z", ~ i' ~q/ == ~ Q"z",


k=O I~O ,,=0 k=O 1=0 ,,=0
00 ob 00 00 00 00 00

~ i' ~r/ == ~ R"z" == ~ Pki' ~q/ == ~ Pki' ~Q/,


k=O 1=0 ,,=0 k=O 1=0 k=O 1=0
00 00 00 00 00

~ p,,~ ..z" == ~ Pki' ~ s/, ~ Q"q..z" == ~ qki' ~ s/,


,,=0 k=O 1=0 ,,=0 k=O 1=0
00 00 00 00

~ R ..t ..z" == ~ Pki' ~ q/ ~ s".z'"


,,=0 k=O 1=0 ",=0
00 co co 00

== ~ Pki' ~ Q,q/ == ~ qki' ~ P,~/,


k=O 1=0 k=O 1=0
75. Put
t" == (a 1 + a2 + ... + a.. ) n- a ,
and lim s" == s. The expression
,,~oo

..
• =1

converges to 0 [66].
76. [E. Cesaro: Nouv. AnnIs Math. Ser. 3, Vol. 9, pp.353-367
(1890).] According to 70 the limit is equal to

77. [I. SchuL] Set ~


" P. == P", ~ q.
" == Q.. and assume {J> O.
• =1 .=1
We have lim Q" == lim nq.. == 00, because otherwise we would have
"-+00 n-+oo

q.. N!L,
n
q > 0; i.e. [76] Q.. N q log n -+ 00: contradiction. Also
lim P" == lim np.. == 00. If (X> 0 the same argument as for Q.. can be
n-+oo n-+oo

yuana97@uw.edu
194 Infinite Series and Infinite Sequences

used. For lX = 0 lim np.. p;;l = 00, all the more lim np.. = 00, conse-
11-+00 11-+00

quently }; P. diverges. Therefore the series }; np"q" is divergent. If


.=1 ,,=1
lX = 0 we conclude nq.. < KQ", K independent of n,
" . "
}; 1IP.q. < K }; P.Q. < KQ .. }; P. = KP..Q",
>=1 "a:;:::l
i.e.

In the case lX > () replace the proposition by


P"Q"
lim =lX+{J.
ft.-+oo "
1: liP. q•
• =1
Apply 70:
a.. = P"Q.. - p .. - 1Q.. -1' b.. = np..q.. ,
a" p.. Q.. 1
-b.. =----
np"
+---_.
nq..
---+lX +{J =
n
s.

78. Example: a1 = all = aa = ... = 1. Now assume that a.. > a.. +!,
a" -+ 0 and
al + ~ + '" + an - nan < K for n = 1, 2, 3, ...
For a given m find n such that a.. -:s la.... From
K > al + a2 + ... + a... - man + (a",+! + ... + a,,)
- (n - m) an > m(a... - a,,) ? ima...
follows that a1 + a2 + ... + a... < K + ma", < 3K for m = 1, 2, 3, ...
\Ve are dealing here with a transformation under special conditions; 66
in itself is no help.
79. Contains 65 as a particular case; the proof is the same. If Phi = 0
for l > k the matrix is called triangular (d. 65, 66) or more specifically
lower triangular. If Phi = 0 for l < k the matrix is termed upper tri-
angular.
80. Contains 66 as a special case. Proof analogous.
81. With s" = nc" + (n + 1) c,,+! + ... we can write

Obviously lim s" = 0, which implies lim til = 0 [80].


"-+00 n-+oo

yuana97@uw.edu
Pt. I, Solutions 78-85 195
82. [G. H. Hardy and J. E. Littlewood: Rend. Circ. Mat. Palermo
Vol. 41, pp. 50-51 (1916); d. also T. Carleman: Ark. Mat. Astr. Fys.
Vol. 15, No. 11 (1920).] Put

bo + b1 (1 - £x) + ... + b,,(1 - £x)" = t".


It is known from analysis [Hurwitz-Courant, pp. 32-33; Hille, Vol. I,
p. 128.] that for Iyl < 1 -£x
bo + b1 y + ... + b"y" + ... = ao + a1 (£x + y) + ... + a" (£x + y)" + ...
holds identically. Consequently
(1 _ lX)"+1
=- ------ J; al (1X. + y)
00 ,,-/I /I 00 1 00 1
J; (1 - IX) Y J; b1y
11=0 1=0 1 - (x + y) 1=0

+ y)l •
00

= (1 - £x)"+l J; SI(£X
1=0

The coefficient of y" on the left hand side is t" and on the right hand side
(1 - £x) ,,+1 [
+ (n +1 1) £XS"+1 + (n +2 2) £x2S"+2
S"

+ ( n +3 3) £x3S"+3 + ...] = t,..


J; P". = 1 (binomial formula). The present transformation has an
.=0
"upper triangular matrix" whereas the matrix considered in 65 should
be termed "lower triangular".
83. Cf. the analogous propositions of 65 and 79.
84. Cf. the analogous propositions of 66 and 80.
85. In 84 put
all
s,,=-,;-,
,.
For given 11 and e, e> 0, choose n so that

We have
b t'
IP.(t) <
+ blt + b.1F + ... + bllt
'9 II
bo

Th~ right hand side converges to a value smaller than east -> 1. The
proposition holds also if the radius of convergence is einstead of I, Q > O.

yuana97@uw.edu
196 Infinite Series and Infinite Sequences

86. [N. H. Abel, l.c. 47, Vol. 1, p. 223.] According to 85


00

1: (a o + a l + ... + a,,) tIS


ao + alt + a2t 2 + ... + a..~'" + ... = ,,-0
- 00

1: tIS
.. =0
. ao + a l + ... + a..
-+lim 1 =s.
" ..... 00

87. [G. Frobenius: J. reine angew. Math. Vol. 89, pp. 262-264
(1880).] It follows from the hypothesis that n-1a" is bounded, therefore
00

L a.. t" converges for t I I< 1. According to 85 we have


,,=0

1: (a o + a 1 + '" + a,,) t"


00 00

1: (so + 51 + ... + SIS) t"


.. =0 .. =0
-~~-------------
00 00

1: t" 1: (n + 1) tIS
,,=0 ,,=0

. 50 + 51 + '" + 5"
-+ hm 1 = s.
....... 00 n +
88. Multiplying numerator and denominator with the geometrical
series we obtain
00

a o + all+ + ... +
a 2t2 + .. . a"t "
1: (ao + a 1 + '" + an) t"
..=0
bo + bIt + b2t2 + ... + b"t" + .. .
00

1: (bo + b1 + '" + b,,) t"


.. =0
. ao + a 1 + '" + a..
-+ hm
' ...... 00 bo + b1 + ... + b.. =s.
89. Set q>(z} = log (1 + ; } - IX log (1 + ~ ). According to 156

converges, i.e. there exists a finite limit


,"-1 I
lim n >0. n.
" + 1) ••• (a + n - 1)
..... 00 ala
I 85 t - <x-1 d b _ ala + 1) ... (a + n - 1)
n pu a" - n an ,,- n! .
90. The integral in question expanded in powers of k2 is
n
2 n=l
i (1 ·32·4·
.. · (2n.. 2n- 1))2 k2".
. (1 ·3 '" (2n - 1))2 1 1
Spec1al case of 85: an = -n2 2·4 ... 2n
,bn = -,
n
s = -,
2
k2 = t.

yuana97@uw.edu
Pt. I, Solutions 86 - 92 197

91. [Cf. 0. Perron: Die Lehre von den Kettenbriichen. Leipzig:


B. G. Teubner 1913, p.353, formula (24); R. G. Archibald: An Intro-
duction to the Theory of Numbers. Columbus/Ohio: Charles E. Merrill
Publishing 1970, p. 176.J Recursion formulas for A .. and B .. :

A"+ 2 = (2n + 1) A"+1 + aA .. , B .. + 2 = (2n + 1) B"+1 + aB.. ,


n = 0, 1, 2, ... ;

This leads to the differential equation

y" = 2xy" + y' + ay,


where y stands for F(x) or G(x). Substituting v2 for a(1 - 2x) we find
2
ddv °
y2 - y = ,y = c1 ev + c2 e
-v, c1 an d c2 cons t an t, 'l.e.

Put 2x = t. Then 85 may be applied in the following manner:

A.. 1
n-;;T ,,-1
(t) Va -'1;;-
= Va e - e
A F' -
lim ~ = lim __'_2_ = lim __2~
...... 00 B.. .. . . B..
00n _1_ 1.....1-0 G'(.!...) eVa + e-Va .
n! 2"-1 2

The power series for G' (+) is diver.gent for t = 1 because all its coeffi-
cients are non-negative and lim G' (-2t ) = CXl.
1.....1-0
92. Special case of 88.

+ b1 + ... + b.. ) t" we have


00

Since (1 - t)-a-1 = '2: (bo


,,=0

b +b + b2 + ... + b"= n
(a + n) = (a + 1) (a + 2) ... (a + n) b a
o 1 n! N n,
b> O.
[Solution 89. ] According to 75 we obtain

yuana97@uw.edu
198 Infinite Series and Infinite Sequcnccs

93. According to solution 89 we have

OO( [ ;.-]) [~/~~-]-2[V~]


lim (1 - t)3/2 1;
1->1--0 ,,=1
IVnj - 2 JI ~ r = lim - ..-----~-- < O.
2. ,,->oo_~.~.2 ... 211+1
2 '6 2n

The limit is - (V~- _- _1) V;, as ~. ~. 2 ... ~'!._-t_I ~ 2 1 / n -. [II 202].


2 2 4 6 2n V"n
94. The statement 85 is true not only for t -+ 1 but also for t -+ 00.

Cf. 84. The sum of the series


bo + bit + bi!. + ... + b"t" + ...
increases to infinity as t -+ 00 because b,. > O.
95. Application of 94: s~, b" =~.
a" = n. II.
[Borel's summation,
Knopp, p. 471.]
96. We write 5" = ao + a l + ... + a", 5_ 1 = O. Then using partial
integration for the subtrahend we get [95]

97. In 96 put an = 0 for n odd


m1 35 2m-l
a" = (-1) 2'4'6 ... ~ for n = 2m.

We have 5 = (-V--~'==-) = V~ . Similarly we obtain for -1 ~ x ~1

r
1 - z :=-1 2

e-I]o(xt) dt = ___ 1 ___ .


o VI + ~2
98. [For a special case see M. Fekete: Math. Z. Vol. 17, p. 233 (1939).]
It is sufficient to consider the case where the lower bound a is finite.
a
+
Assume e > 0 and .-.!!!. < IX e. Any number n can be written in the form
m
n = qm + r where r is an integer, 0 :'S r < m-l. We define ao = O.
Then we have
a,. = aqm +. < a,,. + am + ... + am + a. = qa + a, , m

a,. aqm +, qa", + a. = -a", ---- qm


. -.- +-
a,
-n = qm + I' <
---------- -- qm + I' m qm + I' II '

a"
IXS-< (IX
-II
+ e)-----
qm
qm+1'
+-.
a.
II

yuana97@uw.edu
Pt. I, Solutions 93 - 99 199
*99. Since 2am - 1 < a2m < 2a m + 1 we have

(*) Ia2m _ am I < _1 •


2m m 2m
The series

a1 + (a2z_~)
1
1
+ (a, _ az) + (as _ at) +
4 2 8 4
... = lim ~ = w
a
" ..... 00 2"

is convergent because
la1 1+ 2-1 + 2-2 + 2-3 + ...
is, on account of (*), a majorant series. Write the integer n in the binary
system, i.e.
n = 2'" + E12m- 1 + ... + Em'
where El , E2, ... , Em are 0 or 1; according to the hypothesis

a 2", + E1a 2",-1 + ... + E",a 1 - (E1 + E2 + ... + E",)


:s;; a" < a 2", + E1a2",-1 + ... + E",a 1 + (El + E2 + ... + Em),
2m a ._ 2m_- 1 a2m-1 _ ••• _ Em a 1 <m < logn I
Ia"n _ ~ __
n 2'"
1i1
n 2",-1 n 1 = n = n log 2 .

Applying 66 with

So = 0, Sl =i, ... ,
~ ~~-1 ~
PnO = 0, P"1 = n-' ... , P",m-l = - n - ' P"m = n" P",m+l = 0, ... ,

we conclude lim n- 1 a" = w. By virtue of (*) we finally obtain


" ..... 00

Iw -amm- \ <- \a-2m - -


2m
am \ + \ a-
m
4m - a-
4m
2", \ + " ' <1- + -1 + " ' = -1 .
2m 2m 4m m

Another proof: From


a m+" + :;, < (a", + 1) + (an + 1), 1 - a",+" < (1 - a",) + (1 - a,,)
we conclude by 98 that
. a" + 1 1- a
lim--=w, lim--"=-w,
n-+oo n fI.~oo n
where -w, being a lower bound, cannot be + 00, and so w is not - 00.

We conclude further that


an + 1 1- a
-----
n
>
= w
J
--">
n -
-w.

yuana97@uw.edu
200 Infinite Series and Infinite Sequences

100. [L. Fejer: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 142, pp. 501-
503 (1906).] The proof will show that it is sufficient to discuss the case
of bounded partial sums SI' S2' S3' ... , s,,' ... Put lim inf s" = m and
"-+00
li m sup s"
"-+00
= M , 1 IS. . . .mteger, 1 > 2 an d!lf
a posItIve - --I -m = .i
u. D'IVI'de

the number line into 1 intervals by the points


-00, m +~, m + 2~, ... , M- 2~, M -~, 00.

Choose N such that Is" - s"+11 < ~ for n > N. Let, furthermore, s""
n l > N, be in the first (infinite) interval, s".' n2 > n 1, in the last (infinite)
interval. In each of the 1 - 2 intervals of length ~ there will be at least
one point s",+k (0 < k < n 2 - n l ). A similar argument holds if the
sequence does not "slowly increase" but is "slowly decreasing".
101. [Cf. G. Szegi:i, Problem: Arch. Math. Phys. Ser. 3, Vol. 23, p. 361
(1914). Solved by P. Veress: Arch. Math. Phys. Ser. 3, Vol. 25, p.88
(1917).] The interval (0, 1). Cf. 102.
102. [G. P6lya: Rend. Circ. Mat. Palermo Vol. 34, pp.108-109
(1912).] There are subsequences with arbitrarily high subscripts
t"" t",+l' ... , t"., that descend arbitrarily slowly from the lim sup to
the lim inf of the sequence. The detailed proof follows the lines of solu-
tion 100.
103.
v" V"+l
-- --
n + v" n + 1 + v,,+1

104. Let sl' S2' s3' ... , s", •.• , lim s" = s, be the sequence in question.
"-+00

Choose s" anywhere in the interval s - ~, s + ~, and more generally


s in the interval s -~, s + ~; VI < Vz < Va < .... The terms of the
'" 2" 2"
series s" + (s•• - s.) + (s•• - s,,) + ... are not larger than the terms
of Is" I + (~ + !) + (! + !) + ... .
105. Only finitely many terms of the sequence are below a certain
fixed number. Among finitely many numbers there is a smallest one.
106. If the Weierstrass least upper and greatest lower bound of the
sequence coincide there is nothing to prove. If they are different then at
least one of them is different from the limit of the sequence. This bound
is equal to the largest or smallest term of the sequence.

yuana97@uw.edu
Pt. I, Solutions 100-110 201
107. The smallest among the numbers ll' l2' la' ... , lm (m given) is
called 'Yj, 'Yj > O. According to the hypothesis there are terms of the
sequence that are smaller than 'Yj. Let n be the smallest index for which
l,. < 'Yj. Then we have
n>m; l,. < ll,l" <l2, ... ,l,,<l"_I'
108. Apply 105 to the sequence l;;l, l;;';I' l;;~2' ...
109. [G. P6lya: Math. Ann. Vol. 88, pp. 170-171 (1923).] We call
lm an "outstanding term" of the sequence if lm is larger than all the
following terms. According to the hypothesis and to 108 there are in-
finitely many outstanding terms:
In,, In., In., ... , In, > In. > In. > ....
If ly is not oustanding it lies between two consecutive outstanding terms
(for v > n l ), i.e. n'_1 < v < n,. We find successively In, -1 < In ,,
In -2 < In , ... , l. < In , consequently
" ,
(*)
From this we conclude
lim sup l",.,
,.~oo
Sn = + 00.
Otherwise In ,Sn, and consequently, according to (*), the sequence
lIS!> l2S2' ... would be bounded, contrary to the hypothesis. Apply now
107 to the sequence
l-1
n
S-1
HI 1
1
l-1
HI
S-1
HI'
In- 1 sn- 1, ... I I

110. [Concerning 110-112 d. A. Wiman: Acta. Math. Vol. 37,


pp. 305-326 (1914); G. P6lya: Acta Math. Vol. 40, pp. 311-319 (1916);
G. Valiron: Ann. Sci. Ecole Norm. sup. (Paris) Ser. 3, Vol. 37, pp. 221-
225 (1920); W. Saxer: Math. Z. Vol. 17, pp. 206-227 (1923).]
Analytic proof: We have lim (Lm - mAl = + 00. Let L" - nA
m-+~

[105] be the minimum of the sequence


Lo-O, L 1 -A, L 2 -2A, La - 3A,
Obviously

for f..l = 1, 2, ... , n; v = 1, 2, 3, ... ; n = 0 is excluded by the assumption


on A.
G'eometric proof: From the given points draw vertical rays
upwards, construct the smallest convex figure (infinite polygon) con-

yuana97@uw.edu
202 Infinite Series and Infinite Sequences

taining these rays and determine the line of supportl of slope A. We call
(n, L,.) the corner (or one of the corners) through which this line of
support passes. The lines connecting (n, Ln) with the points to the left
(in the convex domain) have a slope smaller than A, and with the points
to the right (in the convex domain) a slope larger than A.
111. [Cf. G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24,
p. 282 (1916).] Write
m = 1, 2, 3, ... , Lo = 0 [110J.
Since LI - A < 0 the difference Lo - 0 cannot be the minimum mentioned
in the solution of 110. Then 1n+! ~ A, therefore 1n+1' and consequently
n, increases to infinity simultaneously with A.
112. Put 11 + 12 + ... -+- 1", = L"" m = 1, 2, 3, .•. , Lo = O. We find
L - rnA
lim --"'---- = - A [67].
1n4-oo m
The sequence
Lo - 0, L] - A, L2 - 2A, L", - mA, ...
tends to - 00. Let the maximum be Ln - nA. The inequalities 111
question are satisfied for this subscript n. There are in the sequence
L o' L I , ... , L"" ... infinitely many terms larger than all preceding terms
[107]. Let Ls be one of them.
lS-1 +ls
--~---~-----

2 s
are all positive. If A is smaller than their minimum the subscript n
belonging to A is > s. - The points (n, Ln) are to be enclosed in an
infinite polygon convex· from above.
113. Set lim sUPJ~g rn = S. Then we have a) S ~ A.. This is obvious
",--;'00 og r",
for S = 00. If S is finite we find for 6 > 0 and for large enough m
log m < (S + 6) log 1'm' Therefore
S+2.
1'-5-2. < m- S+-.- 2;1',:5-2. converges,
m~l

1 By line of support of a closed set ~lJl we mean a line that contains at least one
point of 9Jl but such that one of the open half-plan·!s determined by this line does
not contain any point of 9R. The intersection of the closed half-planes defined by all
possible Jines of support containing ~)l is a convex domain Sf, the smallest that
contains ~Jl (the convex hull of ~m. Every line of support of ~R is a line of support
of Sf and vice versa. These concepts can be easily adapted to the case where the
ideal point (point at infinity) belongs to the set ~Jl. as is the case here. Cf. [II,
Chap. 3, §1.

yuana97@uw.edu
Pt. I. Solutions 111-119 203

i.e. S + 2e > A, S > A. Furthermore b) S < A. For A= 00 this is evident.


If A is finite 2: 1';;;A-. converges for e > O. Therefore, by the well
... =1
known particular case of 139, where en = 1, m1';;;A-8 -+ 0, i.e.
iiog m
og1'",
< A + e for large enough m, S -
~ A + e,
-
S < A.

114. Since the assumption on the x",'s is independent of the number-


ing we may assume that Ix", I = 1'"" 0 < 1'1 < 1'2 < 1'3 :::; •••• We enclose
each number x.' 11 = 1, 2, 3, ... ,11t in an interval with center x. and
length ~. These intervals do not have inner points in common and lie
completely in the interval [ -1'... -- :,1'", + :} Therefore
. logm
m~ < 21'... +~, i.e. hm sup -----
" ' .... 00 log 1'",
< 1.
-

115. According to 113 lim m1';;;fJ = O. Apply 107 with l", = m1';;;fJ.
+ 00
" ' .... 00

116. We have lim sup m1';;;'" = because otherwise we could


....... 00

find a constant K independent of m such that m1';;;"' < K, thus


1 K 1+8 ,
< ~1+8 for < /X(l + e) < A,
...
1'1%(1+-.) IX

which is contradictory to the hypothesis that A is the convergence ex-


ponent of the sequence 1'1,1'2' .•. , 1'm' .,. Furthermore m1';;;fJ -+ O. Apply
109 with
1

117. The maximum term's index is m = 0 if 0 < x < 1'1; its


index is 11'£ if 1'", < x < 1'm+l and m ?: 1. The terms increase at first
until the m-th term is reached (maximum term) and they decrease
afterwards.
118. In 111 put l", = log 1'", - log sm' k = n -I-' and k = n + 11
n:.spectively. For given A determine n according to 111 and then draw
l' from the relation A = log l' - log s". It is obvious [117] that for
y = s" the n-th term of the second power series becomes the maximum
term.
119. Let p"x" be an arbitrary term and choose m such that m> n,
Pm> O. Then we have

if x> "'-V"-- .
. P..
Pm

yuana97@uw.edu
204 Infinite Series and Infinite Sequences

120. If for a certain x a term p"x" is larger than all preceding terms,
i.e. if for a certain value of x all the inequalities

x·(p"x"-· - P.) ~ 0, v = 0, 1, 2, .•. , n - 1,

hold, then this remains true also for any larger value of x.
121. Let m be arbitrary and choose x so that p",x'" is the maximum
term. Then
1 e'"
P'" ~ PO'
On the other hand P",«()e)'" is bounded for m -+ 00, 0< () < 1. Conse-
quently
. ~r.- 1
lim VP'" = - .
"'-'00 e
122. [l.c.110.] Suppose that for a certain positive value zthe central
co a
subscript of the series ~ b'" z'" is n [121] and that yis a value for which
",=0 '"

the same n is the central subscript of ~ b"'y"'. Define x by z '- ~ .


"'~ y
Then we have
a" i' a" K'
-b "'1"~-b - , b,,~ < b"y", k = 0, 1, 2, ...
loy - "Y"
123. [l.c. 110.] Let

co a
be the successive values of the central subscript of the series ~ b'" z"'.
",=0 ...
Assume that the term with the specific subscript k is the maximum
term in the interval (C"-l' C,,) and that y" is the value for which b"y"
becomes the maximum term of the series ~ b",y"'. The method used
",=0
for the solution 122 associates these values Yl' Y2' ... with values of x
that belong to the intervals

The exceptional values x* with which no y is associated must lie in the


intervals

yuana97@uw.edu
Pt. I, Solutions 120-128 205

Thus the values log x· have to fall into intervals of length


Y
log-i + log-!
Y
+ ... + log-
Y"
+ ... = "'-+00
.
lim Y"
log- e
= log-,
Yl Y. Y"'-1 Yl Yl

where (! denotes the radius of convergence of 1: b"y".


,,=0
124. [A. J. Kempner: Amer. Math. Monthly Vol. 21, pp. 48-50.]
All the non-negative integers between 0 = 00 .. , 000 and 10'" - 1 =
99 ... 999 that are written with the nine digits 0, 1, 2, ... , 8 are obtained
by lining up these nine digits in all possible ways m at a time. Thus we
get a total of 9'" numbers. Let,,, be the n-th non-negative integer that
is written without the figure 9. If 10",-1 - 1 < '" < 10'" - 1 then
n < 9"'. Therefore
log n ...- log 9
limsup--=--< 1 [113] .
"-+00 log 1'" - log 10

More directly: The number of terms of the subseries in question with


values between 10",-1 - 1 and 10'" - 1 is 9'" - 9",-1. Consequently
the sum of the sub series is smaller than

125. Consider the two subseries which contain all positive and all
negative terms, respectively.
126. [K. Knopp: J. reine angew. Math. Vol. 142, pp. 292-293
(1913).] No. Example: Let b1 + b2 + bs + ... be convergent and
Ib11 + Ib2 1+ Ibsl + ... be divergent. Put

Noticing that n! is divisible by l if n > l and collecting all the terms which

into the series


terms.
+ ++ ++ + ...
belong to the same bIll we transform the subseries a" + ak+1 + a"'+21 + ...
b1 b2 bs except for a finite number of

127. No [128].
128. No. - Use b" of 126. Suppose that the functions !p(x) and c1>(x)
assume only positive integral values and are strictly increasing:
0< !p(1) < !p(2} < "',0 < c1>(1} < c1>(2) < c1>(3) < ... ; !p(n), c1>(n) inte-
gers. Define a new series a1 + az + as + ... with the general term
bIll b
av = 4>(m) _ 4>(m _ 1) forc1>(m -1) < 11 <c1>(m) ; a 1 = a 2 = ... = a<l>(l) = 4>tl) .

yuana97@uw.edu
206 Infinite Series and, Infinite Sequences

The inequality qJ(t",) < «P(m) < qJ(t", + 1) determines the integer t",
completely. Collecting the terms that belong to the same b", we trans-
t - t
form the series a'l'(l) +a'l'(2) + ... into the series ! 1 1P (mT -1P7':~ 1) b",.
If we set «P(x) = ~' we obtain a series a1 + as + as + ... which furnishes
a counter example for the problems 126-128: If qJ(x) is a polynomial
t - t
of degree > 2 or if qJ(x) =kJX the sequence lP(m)"'-IP(:-~ 1) is, beyond a
certain m, monotone decreasing [Knopp, p. 314].
If qJ(x) = k + lx we transform the contracted series by adding an
absolutely convergent series into the series l-1(b 1 + b2 + ... + bn + ...)
[126].
129. [A. Haar.] Since the series SI = al + a21 + aSI + ... is of
the same type as SI it is sufficient to show al = O. We denote the first
m prime numbers by PI' P2' ... , Pm' The series
S1 - + sp. + .. , + sPm)
(SPI

+ (sp,p, + sp,p. + ...)

+ (_1)'" SpIP•...f.'"
contains only a1 and the an's whose SUbscripts n are not multiples of
one of the prime numbers PI' P2' .... Pm' Each of these an's appears
exactly once [VIII 26]. I.e.

al < 1: Ian I,
n=p",+1

The condition "absolutely convergent" is essential as can easily be


shown by the example 5; ,1,(n) [VIII, Chap. 1, § 5].
"=1 n
130. [G. Cantor, ct. E. Hewitt and K. Stromberg: Real and Abstract
Analysis. Springer: New York 1965, pp. 70-71.} We obtain the set of
points in question by removing from the closed interval [0, 1] the open
middle third of the interval, then apply the same process to the remaining
two intervals and so on indefinitely. (This set is often called the Cantor
discontinuum or the Cantor ternary set.)
131. [Cf. S. Kakeya: Proc. Tokyo math.-phys. Soc. Ser. 2, Vol. 7,
p. 250 (1914); T6hoku Sc. Rep. Vol. 3, p. 159 (1915).] Write

p" +Pn+l + ... +Pn+.=P".'


,
lim p".=P
'
n. n=1,2,3, ... ,v=0,1,2, ...
v~oo

yuana97@uw.edu
Pt. I, Solutions 129-132.2 207

Assume that P'" is the first term for which P'" < (1. Either there exists
a "1 such that p .."., < (1, PII"',+1 > (1. "1 ~ 0 or p'" :::;; (1. In the second
case we have p'" = (1 because p'" > P.., - l ~ (1 (for n 1 = 1 this means
PI = S ~ (1). i.e. (1 may be represented as an infinite subseries. In the
first case we determine the first term P... with n 2 > n 1 + "1' p .."., + P". < (1.
Either there exists a "2 with p .."., + P...... < (1, P""" + P"" •• +1 ~ (1,
"2 ~ 0 or p .."., + p ... :::;; (1. In the second case we have p .."., + p ... = (1,
because p"".' + +
p ... ~ p ....., P... -l ~ (1 (n2 > n 1 +"1 + 1, because
p"".' + P"'+',+1 = p""',+1 > (1) i.e. (1 may again be written as an
infinite subseries. If this procedure never terminates (if the first case
occurs at every step) then (1 = p ..... , + p ...,•• + p .... p• + ....
132. From the relations
P.. = P"+l + P.. +2 + P.. +s + .. .
P"+1 = P"+2 + P"+s + .. .
111
we gather P.. = 2P"+1' thus PI = "2' P2 = ,., ...• P.. = 2'" .•. The
representation by infinite binary fractions is unique.
132.1. [G. P6lya, Problem: Amer. Math. Monthly Vol. 51, pp. 533-
534 (1944). Solution Amer. Math. Monthly Vol. 53, pp. 279-282 (1946).]
Define
R -.!. ~ 2. ... (2n + 1)
"-246 2n'
Then
(1 + !)(1 + ~)."(1 +2~)=R.. ,
( 1 - ~) (1 _~) ... (1 _ _ 1 ) =....!...
3 5 2n + 1 R,,'
1 1
R" t'J n"2 2n -"2 by II 202

and so the product of the first m(p + q) factors of p p•q is


1

R".p
Rmq
t'J (1-)2
q
x2 x2
132.2. See 132.1. From log (1 + x) = x -2" +"3 - ... follows
log p p•q = Sp,q +A
where the infinite series A is absolutely convergent and so its sum is
independent of P and q (of the rearrangement of its terms). Another

yuana97@uw.edu
208 Infinite Series and Infinite Sequences

proof can be derived from the expression of Euler's constant given m


the solution of II 18.
133. Insertion of appropriate vanishing terms into the two comple-
mentary subseries reduces the proposition to the termwise addition of
two convergent series.
*134. We assume that all the terms of the divergent subseries
ar , + ar , + a" + ... are non-negative. Then the complementary sub-
series as, + as, + as, + ... will be such that to each positive e there
corresponds an integer N so that
as... +as... +1 +···+as.. <e
provided that n > m > N. After this remark the proof essentially
coincides with the well known usual proof for Riemann's theorem on
the rearrangement of the terms of conditionally convergent series.
[Knopp. pp. 318; cf. also W. Threlfall: Math. Z. Vol. 24. pp. 212-214
(1926).]
135. From PI;;;::: Ps ;;;::: Ps > ... , 0 < ml < ms < ms < ... follows
that

PI + P2 + ... + P.. ;;;::: .P..., + P..., + P..., + ... + P..... .


136. Determine the "red" subseries Pr, + Pr, + Pro + ... so that
Pr.. < min (2-", Q.. - Q.. -l)' n = 1, 2, 3•...• Qo = O.
Then Pr, + Pr, + ... + Pr.. :s;: Q... the complete "red" subseries con-
verges and Q.. - (Pr, + Pr, + ... + Pr.. ) increases beyond all bounds.
The terms of the complementary subseries are successively accommodated
..
where the relations ~ Pro < Q.. permit it. This construction only shifts
;=1 •
the two complementary subseries relatively to each other.
137. [W. Sierpinski: Bull. into Acad. pol. Sci. Lett. Cracovie 1911,
p. 149.] To obtain s' the divergence of the positive sub series is slowed
down by the method used in 136.
138. [E. Cesar<>: Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat.
Natur Ser. 4, Vol. 4. 2nd Sem. p. 133 (1888); ]. Bagnera: Darboux Bull.
Ser. 2, Vol. 12. p. 227 (1888). Cf. also G. H. Hardy: Mess. Math. Ser. 2.
Vol. 41, p.17 (1911); H. Rademacher: Math. Z. Vol. 11. pp.276-288
(1921).] Put E.. = e1 + e2 + ... + e... Eo = O. We have now
.. ..-1
e1Pl +e2P2 + ... +e..p.. = ~ (E. - E._1)P. = ~ E.(P. -P.+l) + EnP.. ·
.=1 .=1

yuana97@uw.edu
Pt. I, Solutions 133-140 209

Suppose that E,. > om for n > N, IX > O. Then we have


N ,,-1
elPI + e2P2 + '" + e,.p" > 1; E~(P~ - P~+1) + IX 1; v(P~ - P~+1) + IXnp..
7=1 ~=N+l

"

where K is independent of n; therefore the right hand side tends to 00


in contradiction to the hypothesis.
139. [E. Lasker.] Set En = el + e2 + ... + e,. as in 138. The se-
quence
(E)
has the property that between two terms with different signs there must
be a vanishing term. We distinguish two cases: (1) Infinitely many
terms of the sequence (E) vanish. (2) All but finitely many terms of the
sequence (E) have the same sign. Suppose they are positive. In the first
case choose the subscript M so that EM = 0 and that we have for
M<m<n

i e.P~1 = I.=",+1
I.=".+1 i [(E. - E".) - (E._1 - E".)] p.1

(*) = I "if (E. - E".) (P. - P'+1) + (E,. - E".) p,.1 < e.
• =".+1

Let E". denote the last vanishing term ahead of E,. in the sequence (E)
i.e. E",+I' E",+2' ... , E,. have the same sign. The inequality (*) implies
I(E,. - E",) p,.1 = IE,.p,.1 < e. In the second case choose M such that
the inequality (*) holds for M s;, m < n and that EM' E M+l> E M+2 , •••
are all positive. Let E", be their minimum. Since in this case E. - Em > 0
for v > m the estimate (*) yields (E,. - E",) P,. < e, consequently

E ..P,. < e + EmP,..


Since m is fixed and P.. converges to 0 we find for n sufficiently large
E,.p,.<e.
140. The proposition is a consequence of the well-known representa-
tion of the remainder
/(x) _ (/(0) +/'(0) x +/,,(0) x2 + .... + j(") ( 0) x") = j("+1) (8x) X"+l
1! 2! n! (n + 1) I '
0<0<1,
because r+ 1) (Ox) = 0,./("+1)(0) with 0 < 0" < 1.

yuana97@uw.edu
210 Infinite Series and Infinite Sequences

141. Consequence of 140.


142. Integration of cos x ::;; 1 (equality only for
x = 0, ± 2n, ± 4n, ± 6n, ... ) implies for positive x
• Xl . Xl
SIn X < x; 1 - cos x < 2"' I.e. cos x > 1 - 2" '
. x3.. x3
x- SIn X < 3!' I.e. SID x > X - 3i '
x'. x'
2" + cos x - 1 < 4!' I.e. cosx< 1- 21 + 4i' etc.
Xl Xl

x5 ~"+1)
-"3 +"5 - ... +
x3
143. arctan x - (x (-I)" 211 + 1

= J s 2 +1
(-x )"
i+x2
dx,,, (x) =
0
~J
:If
I
cos xl
Vi-II
dt.
o 0
144. Assume tlo > 0, hence al < 0, a2 > 0, as < 0, '" Then
A - tlo < 0, A - tlo - ~ > 0, A - tlo - ~ - all < 0, •.• , i.e.
al<A-tlo<O,
0< A - tlo - ~ < all'
as < A - tlo - ~ - all < 0, ... ,
whence the proposition follows. The proof runs similarly if tlo < O.
145. Suppose e.g. that tlo ::;; A; then al cannot be negative because
in this case We would get A - (tlo + aJ = IA - (tlo + all I > lall > I~I
in contradiction to the hypothesis. Hence a l > O. Since 0::;; A - tlo < I~I
we have furthermore. A - (tlo + a l ) = A - tlo - a l < O. By similar
arguments we find a z < 0 and tlo + a l + a z < A, as > 0 and
tlo + a l + all + as > A etc. In general, the terms of an enveloping
series need not have alternating signs [148], yet the terms of a strictly
enveloping series must have alternating signs.
146. If, in 145, we assume only Iall > Iall > ... > Ia,,1 we show
in the same way that al' a2 , ... , a.. _1 have alternating signs and that
the first partial series (with subscripts ::;; n - I, n = 2, 3, 4, ... ) are
strictly enveloping. For x sufficiently large and n fixed we have in the
present case

Thus the proposition is proved for the first n - 1 terms, consequently in


general.

yuana97@uw.edu
Pt. I, Solutions 141-150 211
147. The hypothesis implies that the derivatives (t) are alternatelyr)
positive monotone decreasing and negative monotone increasing. Let
e.g. I(t), I" (t). pv (t), ... be positive and monotone decreasing, t'(t).
r
1'" (t), (t), ... be negative and monotone increasing.
R" = [I(t) cos xt dt - ( - f'!?) + f'JO) - ~ .. + (-1)" f(2":21~(0) )
,,+1 00
= (-1)
x 2,,+1 0
J 1(2"+1)(t) sin xt dt

"

- ...J sin xt dt.


f2" +1) (0)
Obviously R" has the same sign as (-1)"+1 x 2"+2 • Apply 144.
148. The sum of the first 2n - 2 terms is

_1_<_3_
3.2,,-2 2,,+1 .

The sum of the first 2n - 1 terms is


1 1
- - <2,,+1
3.2"+1 -•
The signs of the terms do not alternate.
149. The graph consists of line segments forming a kind of spiral.
Its shape justifies to some extent the expression "enveloping". We have
389 101.
+ 120 ~ + u,
i .It
e = 720
[Graph or 151.] The result is accurate up to three decimals.
150. [H. Weyl.] a) z lies on t), Izl > D. The absolute value of the
remainder

I(z) - (/(D) +f'i~) z +f'~(~) + ... +f(:~O) z") =


Z2 j
o
(z :/)" r+ 1)(t) dt
is smaller than
1tt"+1) (D)
n!
I
1'1 " = f ("+1) (0) z"+1 .
If~dr
(n + i)!
I
o
b) Assume, more generally, that f(z) is enveloped by the series
ao + a1z + azz2 + ... for z on t). Suppose now that z lies on t), Iz I > ()

yuana97@uw.edu
212 Infinite Series and Infinite Sequences

and that t is real and positive. Then trl lies on ~. Therefore

thus the integral defining F(z) is convergent. Furthermore

I
'F(z) - l nI a,,! =
ao - -11 a - -2! a. - ... - --,
! Z Zl i
• Z"

= If e
0
oo - I [
I (-;
,)
-ao- z a.tl
al' -zs-"'-7
a"t"] I
dt :::;;

:::;; foo
- o e
_II a"+I,,,+1 I = I(n +
,,+1 Idt
1) I a"+11
z z" + 1 '

151. [For e-' d. E. Landau: Arch. Math. Phys. Ser. 3, Vol. 24,
p. 104 (1915).] Cf. 150. (For mz = 0, z =1= 0 the absolute values of the
derivatives of e-' are constant, e-' however is not constant. Therefore
the remarks of 150 are still valid.) (141.)
,,,. 00 -II 00 -II

152. e 2f e 2 dt = - f
- - 1 -
e~e-Idt.
• z c
Since mz~ ~ 0 we can apply 151 (141 resp.).
153. la" + b,,! = !a,,! + !b,,!. Cf. definition.
154. [Cauchy: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 17, pp. 370-
376 (1843); d. also E. T. Whittaker and G. N. Watson, p.136, ex. 7.]
2Z1
z coth z = 1 +,,=1
I --.--+
00

iI [151,153] .
Z n 11
155. [Cf. Cauchy, I.c. 154.] Its Maclaurin series envelops arc tan z
for mz2 ~ 0, z =1= O. This has been proved in 143 for real z; we use the
same formula for the complex z in question.
158. It is sufficient to consider tp(x) = ao + ~.
;It We get
11'(1) + 11'(2) + ... + tp(n) = aon + ~ log n + 0(1).

157. A necessary condition for convergence is that lim tp(n) = ao = 1.


Apply 158 to the function Boo
log tp(x) = log ( 1 + x~ + ;It:a +...) = x~ + a -la~
2 ;lt2 + ....
158. The series in question is certainly convergent if tp(n) = 0 for a
positive integer n. Therefore we assume that tp(n) =1= 0, n = 1, 2, 3, ...

yuana97@uw.edu
Pt. I, Solutions 151-161 213
We have [88]
.. _ - - _ . _ - -
lim VI(jl(1) (jl(2) (jl(3) "'(jl(n) I =
....... 00
laol .
Hence the series converges for I aoI < 1 and diverges for I ao I > 1.
Assume ao = 1 and (for the sake of simplicity) (jl(n) > 0, n = 1, 2, 3, ...
Then we get [157]
log (jl(1) !p(2) .•• !p(n) = a 1 10gn + b + e", i.e. !p(1) (jl(2)·· . (jl(n) = eb+ s.. n""
lim e.. = 0, b is an n-free constant. Consequently we have convergence
....... 00

for a1 < -1, divergence for a 1 ~ -1.-If ao = -1 we put


(jl(n) = -lp{n), !p(1) (jl(2) ... !p(n) = (-1)" 11'(1) 11'(2) ..• lp(n).
The remainder of the series .En-2 is O(n-l). Hence, cf. also solution II 18,
11'(1) 11'(2) .. '1p(n) = eCn- 4 , + O(n- a,-1).
c is independent of n. Therefore this series converges if and only if
1: (-1)" n- a , converges, i.e. for a1 > O. Summing up: the given series
"=1
converges if and only if at least one of the following conditions is satisfied:
a) !p(") = 0 for some positive integer n; b) laol < 1; c) ao = 1, a1 < -1;
d) ao = -1, a 1 > O.
159. Special case of 158:
'" 1
m(x)
T
= 2- C
-
= '" ",2
1 - -x - -2! -x 2 + '"
convergent for IX > 1, and IX = log 2; log 4, log 8, ... are > 1.

160. J
o
1 1
ez1ogX-
n=on· o
dx = 1: --,1 J1 x" ( log-1 ) .. dx.
00

Substitute x.. +l = e- Y •

161.SetV1+V1+",+V1=t.. , then t!=1+t.. _1' t1 =1,


t,,_1 < tn' n = 2, 3, 4, ... For positive x we have x 2 < 1 + x if and only
if x < !(1 + Vb), i.e. if x is smaller than the positive root of the equation
x2 - x - 1 = O. Hence t! = 1 + tn - 1 < 1 + tn' t"_1 < t" < i(1 + Vb),
n = 2, 3, 4, ... and lim tn = t exists, 0 < t < !(1 + V5), t2 = 1 + t,
_ "-+00
i.e. t = i(1 + V5). We proceed similarly in the case of the continued
fraction where the recursion formula is

U .. = 1 + u;_\' u1 = 1, n = 2, 3, 4, ...

yuana97@uw.edu
214 Infinite Series and Infinite Sequences

182. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p.84
(1916). Solved by G. Szego: Arch. Math. Phys. Ser. 3, Vol. 25, pp. 88-
89 (1917).] If (for the sake of simplicity for v ~ 1) log log a" < v log 2,

a" < e2", then t" < -Ve2 + Ve4 + ... + Ve2" < e 1+,.V5 [181]. If, however,
a" > ef1', P> 2, then t" > e
(f)" . - If a" < 1, then, of course, log log a"
--n--
must be interpreted as - 00.

183. We prove

by complete induction. Suppose that the corresponding relation is proved


for the n quantities a2 , as, ... , a"+1' i.e. that

where

-Va2 + Vas + ... + Va" = t,


Hence
2
t"+1 < a1 + t + s < (V--
a1 + t + V-=-
2
5)2
+1
< ( t" + 2 V-
5)2
~
41
.
184. [Jacobi, I.c. 53, § 52, Corollarium: Werke, Vol. 1, pp. 200-201.]
Write 1 - q = 110, 1 + q'" = a"" m = 1, 2, 4, 8, 16, ... , then the n + 1-th
partial product is

2-2-"
_ 40
-( )2-'"
4142444 8 '''42"

The product a1 a2 a4 a8 '" converges. Cf. also VIII 78.


185. Calling the sum in question F(x) we find

F'(X) = F(x), F(x) = const.· C.


x
188. q/(x) = tp(x), 91(0) = 1, tp(X) = C = 1 + 11 + 2!
Xl
+ ... + n!x" + .. .
L1tp(x) = tp(x) , '1'(0) = 1, tp(x) = ~= 1 + ( ~) + (;) + ... + ( =) + .. .

yuana97@uw.edu
Pt. I, Solutions 162 -169 215

for x > -1. We have


_ x" () _
fJJ.. (x ) - n!' "P x -
(X)
n -
_ x(x - 1) (x - 2) ••• (x - n
n!
+ 1)
' n
= 1 2 3
'" ...
167.

log - x..
x .. +1
= log -Y.. - - -
Y .. +1
1 - ,log -Y..
12n(n + 1) Y"+1
= (n + -21) log (1 + -n1 ) - 1;

x5
log
l+x
1_ x = 2 (X xli ).
T + '3 + "5 + ... ytelds for x = 2n
1
+ 1 '

1< (n + ~) log (1 + ~) = 1 + 3(2n ~1)2 + 5(2n 1+1)l + ...


1 1
< 1 + 3[(2n + 1)2 _ 1] = 1 + 12n(n + 1) ,
therefore x.. < x.. +1 , Y.. > Y"+1' Part of 155 resp. II 205.
On the other
hand, 167 together with II 202 implies II 205 for integral n.
168. (I. Schur.] The fact that a.. is decreasing for p > 1 is obvious
from the expansion
2(n + P) ( 1 1 )
log a.. = 2n +1 1 + 3(2n + 1)2 + 5(2n + 1)' + ...
= (1 + ~ ~ i) (1+ 3(2n 1+ 1)2 + 5(2n ~ 1)' + ...)
[solution 167]. This leads to

log a.. = 1- n
1-p 1
+1" + 12n2 + 0 (1)
n3 '
thus
log an +1 - log a.. = (n
- P+ i)3 + 0 (.;)
+\ )(n n
,

hence a.. increases for n larger than a certain subscript N if P < 1. If

r
P< 0 this is true already for n > 1 as can easily be verified by expanding
(1 + ~ with help of the binomial formula.

169. We wnte a..


. = (1
1 + -;;
)"H (11++ ~)*
-;- ; the fIrst
. factor decreases

[168]; the square of the second factor


.
IS 1 + 2x
--1 + (
n +
- 1 x
nn +
1)'
2
The
condition x > t is therefore sufficient. Now expand
1 1 1 )
log a.. = 2n ( 2n +1 + 3(2n + 1)3 + 5(2n + 1)5 + ...
+ 2[X 1/X)3 1(X)5
2n + x + '3 \2n + x +"5 2n + x + ...
]
_ 2n 2x 1 0 (1)
- 2n + 1 + 2n + ;\' + 12HZ + n3 '

yuana97@uw.edu
216 Infinite Series and Infinite Sequences

Since log an - log an+1 = ~!4~--~ + 0 Ga)


the condition x > ~ is also
necessary.
170. [Cf. problem No. 1098, Nouv. AnnIs Math. Ser. 2, Vol. 11, p. 480
(1872). Solved by C. Moreau; Nouv. AnnIs Math. Ser. 2, Vol. 13, p. 61
(1874).] The first inequality means

(1 + :r+1 <e(1 +2~)


and this is a consequence of the following inequality
1
/(x) = x + x log (1 + ;) - (1 + x) log (1 + x) > 0, O<x<-.
-n

[
f'(X) = _x_ _ log 1 + x> _x_ _ 1 + x +1= 0
/(0) = 0-]
x +2 1 ,x x +2 1 x '
+"2 +"2
The second inequality is equivalent to

[169] .
171. [I. Schur. ] The number e lies in the second quarter of the interval
because
n = 1, 2, 3, ...

The first inequality follows from 170 because

1+-<
4n 1+-
n
1 ( 1) (1 +21-)-1
n'
the second inequality is contained in 169.
172. [I. Schur.] We infer that an is decreasing for 0 < x < 2 from
the equation
x

2: -1 - (- -x )2.-1
1+-- =
2n + x
log a
n
= (n + 1) log - - - -
1 _ _ x_
= (2n + 2) .=1 ,
2" - 1 2n + x
2n +x
= x2.-1 1 00 x2.-1
=.~ 2" - 1 (2;;-+~2-;;-2 + (2 - x) • v~ 21' --[ (2n + x)2.-1 .

Furthermore
x3 1 x(2 - x) ( 1)
log an = x + 3 (2n + X)2 + Tn + x + 0 n3 '

log an - log an +1 = (2n +


2x(2 - x)
x) (2n + x + 2) +0
(1)
na '

i.e. log an - log an+ 1 < 0 for n sufficiently large, if x < 0 or x> 2.
For x = 0 we have an = 1, n = 1, 2, 3, ...

yuana97@uw.edu
Pt. I. Solutions 170-176 217

173. [Proof based on a communication of E. Jacobsthal.] Cf. 174 for


lim sin" x. We have
"-+00
x3 x3 x,
+ 120 '

X - 6 < sm x < x - 6 x>o [142] .

V: - ~ (V; r V~~·l- r (V; r


From the binomial expansion we derive that

> or In- - ~ (V;


< Vn: 1 + 1~0
for constant e and sufficiently large n, n > N(e), according as e < Va
or e > Vi Let e < Va andlX > 0, lX fixed and such that sinN x> V c .
N +IX
Then
.
SlnN +1 X
>'sm VN + >VN+IX -61(VN+IX )3> VN+IX+1'
c
1X
c C C

thus sinnx > V. ,n> N. Consequently for aIle < Va, liminfV;sin"x2:e
n+1X
C
,,~OO

i.e. > Vi If e > Vs choose"", so large that sinmx < v- C •• In a similar


N+1
way as m . t h C f'IrstcaseI d
we conc .
U e smm+l x < V-==- , sIn",+2 < V~=- ,
c. c
et c. N+2 N+3
174. The sequence v" is decreasing, v" > 0, therefore lim v" = v
"-+00
exists; v = I(v) implies v = O. Consequently it is sufficient to prove the
proposition for small x. Let b' be fixed, b' > b. For sufficiently small x
we have
x - ax!' < I(x) < x - ax!' + b'xl
and when n is sufficiently large, n> N(e),

-k-I1 _ a (l)k
en
1
en -k=! > e(n + I) -k-l
or
1 ( 1)k (1)1 1
en -k=! _ a en -"=1 + b' en -k=! < e(n + 1)-k-1
depending on whether
-1 -1
e < [(k - l)a]k-l or e> [(k - 1}a]k-1.

Cf. 173. The assumption on the sign of b is not essential.


175. [J. Ouspensky, Problem: Arch. Math. Phys. Ser. 3, Vol. 20,
p. 83 (1913).] Convergence for s > 2, divergence for s < 2 [173].
176. [Cf. E. Cesaro, Problem: Nouv. AnnIs Math. Ser. 3, Vol. 7, p. 400
(1888). Solved by Audibcrt: Nouv. AnnIs Math. Ser. 3, Vol. 11, p. 35*

yuana97@uw.edu
218 Infinite Series and Infinite Sequences

(1892).] The inequalities


C-l
x>O; x<log--<
x
0,

imply that the sequence".. is steadily decreasing in the first case, " .. > 0,
and increasing in the second case, " .. < O. We have lim".. = " = 0
...... 00
because
e" - 1
" ~ log -u- for " ~ o.
The recursion formula e"" - 1 = " ..e""+1, n = 1, 2, 3, ... yields

+"1 + "1"2 + ... + "1"2'" ".. _1 + "1"2 .. ·"..e.... +1


e'" = 1

and lim "1"2 ... ".. e.... +1 = O.


...... 00

177. [C. A. Laisant, Problem: Nouv. AnnIs Math. Ser. 2, Vol. 9,


p. 144 (1870). Solved by H. Rumpen: Nouv. AnnIs Math. Ser. 2, Vol. 11,
p. 232 (1872).] s = ! cos qJ. Notice that 4 cos3 qJ = 3 cos qJ + cos 3qJ.
178. [I. Schur, Problem: Arch. Math. Phys. Ser. 3, Vol. 27, p. 162
(1918). Cf. O. Szasz: Sber. Berlin Math. Ges. Vol. 21, pp. 25-29 (1922).]
+
If e > 0 is so small that Iq I e < T then there exists a constant A
independent of nand 11 such that

Ib;:p/ < A (Iql+ e)·, 11 = 0, 1, ... , n; n = 0, 1, 2, ...

For n > m we obtain

The sum of the last two terms is absolutely smaller than

.=",+1 .=",+1
i.e. arbitrarily small with m- 1 • Choose m so large that these two terms
are smaller than e. For fixed m choose n so that the first term becomes
absolutely smaller than e.
179. [Special case of an important proposition in function theory by
Vitali. Cf. E. Lindelof: Bull. Soc. Math. France Vol. 41, p. 171 (1913).J
We show only that lim a" l = O. (Then form x- 1 / .. (x) - a..1 , etc.) Assume
...... 00
x
e > 0 and x so small that 0 < A 1 _ x < e. Then we have

Iani I < X-I 1/.. (x) I + A 1 ~ x < X-I 1/.. {x) I + e.

yuana97@uw.edu
Pt. I, Solutions 177-181.1 219
For fixed x choose n so large that Itn(x) I < ex.
180. We have Iak I < A k , k = 0, 1, 2, ... , therefore 1: ak is conver-
gent. Moreover k=O

Isn - S I < IanO - ao I + Ia"l - all + ... + Ia"m - am I + 2 1: A k •


k=m+1
Assume e > 0. Choose m large enough to render the last term smaller
than e. Having fixed m we select n so large that Ia"k - ak I < m: 1'

k = 0, 1, ... , m. Then we get

n
00

181. a) Since the infinite product (1 - q2k) converges for Iq I < 1


k=l
all its partial products lie between two positive numbers a and b, a < b.
Therefore C. as defined in 52 is bounded:

IC.I < ba- 2 t'·


Apply 180.
b) Let y < °in 59, i.e. q > 1. Then we have
qk
----;2:--'--
1+q+q + ... +q
k-1 < q,

y)-,,+.
furthermore ( 1 - -; > eY , v = 0, 1, 2, ... , thus

y
[ 1- ( 1--; )-nJ [1- (1--;y)-n+1J ... [ 1- ( ~--;y)-n+k- 1J
< (1- eY)k.
Apply 180.
181.1. We derive from the definition of U; that
Is;1 < U.
and then from our assumption concerning 1: U; that
Sl + s2 + sa + ... + s; + ... = S
is absolutely convergent. Define, for m = 0, 1, 2, ... ,
r;,m = ai ,m+1 + ai ,m+2 + "',
(thus Y; 0 = s;) and define S! as the sum of the first M terms of the
series (*). These definitions are illustrated by the relation
m co

S - S!(m+l)l = 1: Yi,m+l-i
;=1
+;=m+l
1: Y;,o'

yuana97@uw.edu
220 Infinite Series and Infinite Sequences

Generally, for arbitrary M,

S - SM· = ~
"" r·"mj
.=1
where m. depends on M and as M -+ 00

for any fixed i. Observe that


!r.,m! < 2U j •

There follows, by virtue of 180, that


S - S!r -+ O.
(The foregoing argument, without essential modification, shows that
various "geometrically defined" rearrangements of the terms of the
series (*) leave its convergence and its sum unchanged- square, rectangle,
quarter of a circle instead of a triangle.)
182. We are dealing with the limit of the series

i' n"-I(n + k),,-l [(n + k)" -


k=l-n
n"r 2 =};'
k=l-n
k12 91 (:)

for n -+ 00; the term with subscript k' 0 must be omitted. The function
91(x) is defined by
91(x) = (1 +
X),,-l ( ___ 3 _ _
(1 + x)" - 1
)2;
91(x) is continuous for -1 < x < 00 if we set 91(0) = IX-2 ; 91(x) = 1 if
IX = 1; otherwise 91(x) ,...." x1-c. for x -+ 00, 91(x) (1 + X)"'-l for x -+ -l.
f""oo/

If IX = 1 the value of the limit follows immediately from

+ 2-2 + 3-2 + 4-2 + ... = 6' .


:n;2
1
If IX =t= 1 the general term tends, for k fixed and n -+ 00, to k-291(0).
If IX > 1 91(x) is bounded for -1 < x < 00. If the maximum of
91(x) is denoted by M the series has as a majorant the series E'Mk- 2 for
n = 1, 2, 3, '" [180.]
If 0 < IX < 1 there exists a positive number M such that
1
91(x) <M for --<x<2
2 = = ,

91(X) < M(1 + X),,-1 for -1 < x< - ~ ,


91(x) < Mx1-c. for x > 2.

yuana97@uw.edu
Pt. I. Solutions 182 -184 221
Consequently
-i"
~ -.!.. (~)
..:.., k 2 f{J n
< "
-..:..,
-i"
M
k2
(_n_)l-"
n +
< Mn1-"
k -
10-1
~ k-2 -+ 0 ,
..:..,
k=l-n k=l-n k=l"

f{J(:)<Mk 1 -'" for n=1,2,3, ... , k=2n,2n+1, ...


Hence the remaining part of the series (from -tn to =) hasL"M Ik 1- 1 -"

as a majorant. [180.]
183. We have

therefore

an = EO ~ E1 V + V + .,. -+
2 E2 2 E" V2
always makes sense. To prove
. (n "EO E1 ••• Ev)
a =2sm - ~---
" 4":'"
.=0 ,,'
~

use mathematical induction. We have


n n E E '" E )
sgn an = sgn 2 sin ( 4" 1:. --;-'
O 1
= Eo,
v=O 2
and for EO =1= 0

4sin 2
n
("41: "EOE1 '''Ev )
--v- - 2 = -2 cos ~
(n 1: ".e) " 60 t 1
--v-'
.=0 2 .=0 2

= -2 cos ( -
:n;

2
+-2 v=l
:n;E1 E2 ••• E.)
1:10 ----
2'


2sm ( :n;
- 1:10 ----
E1 1>2'" Ev)
v 1
.
4 .=1 2 -
Take the limit n -+ =.
184. [Cf. S. Pincherle: Atti Accad. Sci. Torino, Cl. Sci. Fis. Mat.
Natur. Vol. 53, pp.745-763 (1917-191.8); Atti Accad. Naz. Lincei
Rend. C1. Sci. Fis. Mat. Natur. Ser. 5, Vol. 27, 2nd Sem. pp. 177-183
(1918).] Put x = 2 cos f{J, 0:::; f{J < n. Then the binary expansion IS
unique

yuana97@uw.edu
222 Infinite Series and Infinite Sequences

except in the case of ~ = P....:II:, p, q, integers, 0 < p < 'll. In this case
2q
there are- two representations possible. The equation

2cos~ = Eo V2 +E1 V2 +Ea V2 + ...


is [183] equivalent to
4qJ)
2sin~ ( 2 - - =2sin~
(ooee
~
... e)01 . "
4, 11: 4, ,,=0 2" '

or, because both arguments are in the interval ( - ;, ;), equivalent to

therefore
n= 0, 1, 2, ... ,
provided that the above mentioned exception is excluded. These equa-
tions determine the E" from the g" uniquely (and vice versa). In the
exceptional case ~ = P....:II:,
q
p, q integers, 0 < p < 'll, q ~ 2, there are
2 -

two representations of 2'11 possible:


11:

2'11 gl 1
n = go + 2' + .. , + ~-2
gq-2
+ ~-=-T + 20q + ~+1
0
+ ...
gl 0 1 1
+ 2' + .. , + 2q - 2 + ~-1 + 2q + 2H1 + ....
gq-2
= go

In this case Eo, E1 , ••• , Eq _ 2 are, as before, uniquely determined, Eq = -1,


EH1 = EH2 = ... = 1, Eq _ 1 may be chosen -lor +1. We have there-
fore

% = 2 cos ~ = -V 2 + -V2 +
Eo E1 E2 V2 + .. , + Eq _ 2 V2.
According to 183 any number written in the above way has to be of the
type 2 cos P....:II:, p, q integers, 0 < p < 'll. If q = 1 we have to modify
2q
slightly: the numbers go' ... , gq-2' Eo, ... , Eq _ 2 do not exist, ~ = ; ,
%=0.
185. The sequence g,. is periodic beyond a certain term if and only
if this is true of the sequence En'

yuana97@uw.edu
Pt. I, Solution 185-185.2 223
185.1. [For precursory heuristic considerations see MD, Vol. 2,
pp. 49-50 and 171.] We seek a solution of the system of 2 equations
u+2v+ 3w=0
u + 8v + 27w = 0
with integral values for the three unknowns. We find
u = 5, v = -4, w = 1.
We note that
u + 32v + 243w = 120.
We define, for m = 1, 2, 3, ... ,
alOm-9 -a JOm-8 -- alOm-7 -- a10m-6 -
- alOm-5 -
- m-
1/5
- ,

= _2m-l/ 5 ,

= 3m- 1/5
and

Then
(1) _ (3) - 0
slOm - slOm - ,

si~m = 120 (1 + ! + ~ + ... + ~)


and so the series considered converges to 0 conditionally for 1 = 1 and
1 = 3, and diverges to +
(Xl for 1 = 5. It is absolutely convergent for

l> 5.
185.2. [G. P6lya, Problem: Amer. Math. Monthly Vol. 51, p.593
(1944). Solved by N. J. Fine: Amer. Math. Monthly Vol. 53, pp. 283-
284 (1946).] (1) The case where D consists of just one odd number can be
settled by an easily visible extension of solution 185.1. (Why is
u + 32v + 243w =l= O? Give a reason avoiding numerical computation.)
(2) If D consists of a finite number, say h, of different odd numbers,
construct by (1) the corresponding sequences

each of which yields a divergent series just for one required exponent

yuana97@uw.edu
224 Infinite Series and Infinite Sequences

and form the sequence

which has the desired property.


(3) If D has an infinity of members use the construction (1) infinitely
often and form the desired sequence "diagonally" as suggested by
181.1. You can fulfill the condition on EU.. by considering, if necessary,
instead of the sequence

obtained at first by the construction (1) the sequence

where the positive numbers PI' Ps. Ps, ... , decrease sufficiently rapidly.
*188. List concretely the different possibilities for small n, use 187
for all n ~ 3.

n\k 2 3 4 5 6 7 8

1
2 1
3 1 3 1
4 1 7 6 1
5 1 15 25 10
6 1 31 90 65 15
7 1 63 301 350 140 21 1
8 1 127 966 1701 1050 266 28 1

*187. You belong to a set of n + 1 persons. In a partition of this


set into k subsets you may stand alone and form a subset by yourself;
there are S:-l partitions of this kind. Or you may join one of the k subsets
already formed by the others; there are kS~ partitions of this latter kind.
*188. We set, without contradicting our original definition,
S~=O if O<n<k

and let E stand for 2 . By 187


.. =0

and this proves our assertion for k if it was assumed for k - 1. The case
k = 1 is easy.

yuana97@uw.edu
Pt. I. Solutions 186-194 225

*189. The right-hand side in 188, decomposed into partial fractions,


equals
1 1 1 1 (_1)"-1 1 (-1)"
k! z - k - 1! (k -1)f z - k + 1 + '" + (k - 1)! 11 z - 1 + lifZ'
Expand in powers of Z-1 and consider the coefficient of Z-,,-I. For a
combinatorial proof see VIII 22.1.
*190. From 189, since
(e' - 1)" = e'" - (~) e("-l). + (:) e("-2), - ... + (-1)"
= ,,~o[k" -(~) (k -1)" +(:) (k - 2)" -'" + (-1)" O"J =~.
*191. Using the notation explained in the introduction to III 220,
we can present 189 in the form
" LI"O"
S"=k\'
Apply III 220 (1) to F(z) = z", see III 221; observe that L1"z" = 0 for
k > n (for z = 0 this follows from our argument).
*192. We have to paint n distinct objects (n houses of a settlement,
n faces of a polyhedron) with x different colors, by using just one color
for each object. There are obviously x" different possibilities. If exactly
k among the x colors are used, the objects of the same color form a subset,
and the k subsets so formed constitute a partition of the total set of n
objects. Hence the number of ways of using just k different colors among
the eligible x is
S~x(x - 1) (x - 2) ... (x - k + 1).
As k can be 1, 2, 3, ... , or n, this proves the assertion for any positive
integer x, and hence for indet~rminate x. In using the same facts as in
solution 191, but in the reverse order, we end up by proving 189.
*193. Use the definition of T" at the beginning and 190 at the end:
Tz'" n 5"'z" sn z", k
1 + ~ ~ -"-= 1 + ~.~ -"-=
00 00 DO 00 00.r

V
J!..J
-"-=
til nl nl
~ L-=-!L.
kl
.. = 0 ' .. =1 " = 1 ' "=1 .. ='" "=0 .

*194. You belong to a set of n + 1 persons and, in a partition of this


set, to a subset of k + 1 persons. This can happen in (~) T .. _" different
ways. In fact, the other people in your subset may be chosen in (~ )
different ways [10J and, once they are chosen, the people remaining

yuana97@uw.edu
226 Infinite Series and Infinite Sequences

outside your subset can be partitioned in T,,_Ie ways. Now, k = 0,1,2, ... ,n
exhaust all possibilities.
*195. Set
co T z"
:E --;- =
,,=0 n.
y.

Then, by 34, the relation 194 is equivalent to


dy
dz
= e'y or dy
y
= e' dz.

Integrate this differential equation and use the initial condition


y = To = 1 for z = o.
*198. After differentiating n times

e.. - 1 = _
1(
1+-+-+-+
e' e
e
e• ...) 1!
2z
2!
3
3!

set z = 0 [193]. The result can be used to prove 45.


*191. List concretely the different possibilities for small n, use 198
for all n ~ 3.

n\k 1 2 3 4 5 6 7 8

1 1
2 1- 1
3 2 3 1
4 6 11 6 1
5 24 50 35 10 1
6 120 274 225 85 15 1
7 720 1764 1624 735 175 21 1
8 5040 13068 13132 6769 1960 322 28 1

*188. In a permutation of n + 1objects which is a


product of k
distinct cycles a given object may form a cycle by itself; there are SLI
such permutations. Or the given object may enter at some place a cycle
already formed by the n other objects; there are ns~ permutations of
this latter kind.
*199. Set s~ = s~ = 0 if 1 < n < k. By 188
""+1 ,,+1
(; + n) 1: s~x" = :E (S~_1 + ns~) x" =:E S~+lx".
10=1 1<=1 11=1

Use mathematical induction.

yuana97@uw.edu
Pt. I, Solutions 195-203 227

*200.
(1 _ t) -% = i x(x + 1) (x + 2) ,••• (x +n - 1) t"
"=0 n.

00 fI s"x","
=1+ ~
~ "=1
I-"-
n'
,,=1 .

= e-%log(l-l)

= I x"
00 1
kI (log 1 - t
)"

10=0

We have first used the binomial expansion, then 199. Consider the coef-
ficient of xI'.
*201. List concretely the different possibilities for small n, use 202
for all n > 4.

n\k 1 2 3 4

1 0
2 1 0
3 1 0
4 3 0
5 1 10 0
6 1 25 15 0
7 1 56 105 0
8 .1 119 490 105

*202. You belong to a set of n + 1 persons. In a partition of this set


into k subsets of the desired kind you may be paired with another person
to form a subset of 2; there are nS~=~ partitions of this kind. Or you may
belong to a subset of more than 2 persons; there are kS; partitions of
this kind.
*203. By mathematical induction on the basis of 202, or by combina-
torial considerations:

S-"2 = SOl2 - n = -2"2- -2 - n.

For S:" see MPR, Vol. 1, p. 118, ex. 11.

yuana97@uw.edu
228 Infinite Series and Infinite Sequences

*204. In a partition enumerated by S:_II there may be


n - a-I, n - a - 2, ... , n - 2a +1 or n - 2a
subsets containing just one element each. Therefore

S"" - I I = ( a +
n )
1
5"+1
1
+ (a +n 2) 5"+2
2
+ ... + (n2a) 5211 . II

Use numerical data from 201 for a = 1, 2, 3.


*205.

n o 1 2 3 4 5 6 7 8
T" 1 1 2 5 15 52 203 877 4140
T" 1 o 1 1 4 11 41 162 715

*206. Follow the line of solution 194; observe that To = 1, Tl = o.


- -
*207. By following the line of solution 195 you arrive at the differen-
tial equation
dy
dz = (e' - 1) y.
*208. From 193 and 207
T- ft.z" _I T "Z"
~--=e ~- [34].
n! n!

*209. Obvious for n = 1. Use 187 and mathematical induction.


*210. (1) immediate consequence of 10.
(2) from 190. Or take F(t) = e(l-l)w and compute the n-th derivative
of F(e%) at the point x = 0 by using 209.
(3) See solution ~. A proof independent of 186-209 is based on
a fact quoted in solution VII 48:

Z1IJ ,lfIJ ,aw


= e1-+-+-+
2 3
...
= e-w1og(1-%).

We can give an analogous proof for (2) by using the well known fact
that the number of partitions of a set containing n elements into kl
subsets each containing 1 element, kz subsets each contaillg 2 elements,

yuana97@uw.edu
Pt. I, Solutions 204-210 229

k3 subsets each containing 3 elements, etc. is


n!

where, of course,
1kl + 2ks + 3k3 + ... = n.
To rework 186-210 starting from 210 is left to the reader as a
research project. Just one hint: If we let I denote one of the expressions
(1). (2) and (3). then
0/ 0/
oz - wi =1. z-
oz
which involves

5"+1
I< - 5"1<-1-
- k5"1<' s~+1
.. - S~_1
.. = ns~..

respectively.
For additional material on the subject treated in the section just
concluded see V 62.1, VII 54.2, VIII 22.1, 22.2, 22.3, 58.3, 247.1, AI 191.1,
191.2.

yuana97@uw.edu
Part Two

Integration

1.
,. - 1 1 1
-,-<-'---=-1--+ 1'-2 2
Xv Xv x"_l x. x,,_l

2.

3.
1:" hea +(v-1)",
v~l

b-a
where h = Xv - Xv
- 1 = -n- .

because

li m e" -
"-+0 - h - -
1 - (dedx X
)
- 1•
X~O -

4.
" v-1
~ aq (q - 1)
"'"'
v~l aqv '

,,-
with q=_v
Xv_1
x
=}/-.
'b
a
We have [3]

lim n
n-+oo
(Vr b _
a
1) = "-+00
lim e~ - 1
log b - log a
(log b - log a) = log b - log a.
n

yuana97@uw.edu
Pt. II, Solutions 1- 8 231
1
*5. Set 1 +2 + -3-1 + '" +-;1 =Hn' Then
1 n 1 1 1 1
Ln =-n""-'
'5:'----=-+-+
v n+l n+2
... + -
n+n
>=11 +-
n

1 1 1 1 1
= 1 +-+-+-+
2 3 4
... +--+---
2n - 1 2n
2 - -2 _ ... 2
2 4 2n

= 1--21 +---
1
3
1
4
1
+ ... +----.
2n - 1
1
2n
Moreover

6.
sin ...3!._ sin 2 _11:_ sin n _11:_) "
lim _11:_ ( ~~ + __n-±.!. + ... + __n + 1 = J sin x dx > O.
n-+oo n +1 _11:_ 2 n_1I:_
~__ o·Jf
n+l n+l n+l
(VI 25.)
7. The mean value theorem implies
n
F(b) - F(a) = ~ [F(x.) - F(X._l)]
.=1
n

.=1
Yet, according to the definitions we use here,
b
F(b) - F(a) = J f{x) dx
a

need not be true. [ef. V. Volterra: Giorn. Mat. Battaglini Vol. 19,
p. 335 (1881).]
k - 1 k
8. Let -n- -< ~ <-.
n Then

yuana97@uw.edu
232 Integration

± Je, (:) -1(.)]


and

-A.';: <Ix <f (~) - min [i~'-;-').t( ~ )]


..
~=1._1

-+- k11(-;;-) - I(~) = max[/(~)./(~)J - 1(0)


~=1
n n

9. [Cf. G. P6lya: Arch. Math. Phys. Ser. 3, Vol. 26, p. 198 (1917).]
The least upper bound of the expression
I/(x1) -/(xo) 1 + I/(x 2) -/(x1) 1 + ... + I/(x,,) -/(X"_1) 1
for all possible subdivisions of the interval [a\ bJ is called the total
variation of the function I(x) on [a, b] (same notation as on p. 46).
Functions of finite total variation are also called functions of "bounded
variation" .

10.
b-a
a+.-

-.d,,=:E"
~=1
f" b-a
(a+pb:~-x)/'(~.)dx,
a+(>-1)"

where a + (p - b-a
1) -n- < ~.
b-a
< a + p- - , thus
n

1
"2
(b-a)2 .
- n - .~ 'Itt. < -.d" <"2
1 (b-a)2 "
- n - .~ M.;

M~ and m. denote the least upper and the greatest lower bound of /,(x)
in the p-th subinterval. We obtain
b - a
lim nLl"
"-+00
= - 2 - [j(a) -/(b)].
11.

because
I(x) - 1(a + (2p a) = (x _ a _ (2p _ 1) b ~ a)
- 1) b ~

X t' (a + (2p - 1) b ~ a) + ~ (x _ a_ (2p _ 1) b ~ ay I" (~.) .

yuana97@uw.edu
Pt. II, Solutions 9-14 233
The intergal of the linear term over the interval
[a + (v - 1) b : a, a + vb: a] vanishes. Cf. 10.
12. We have [11]
b-II

J [f(x) -
11+ 211 + 1

L1~ = I(a)] dx

I J
II
b-II
11+(2.+1)·211+1

+ I. [/(x) - 1(a + 2v :n ~ a1)] dx


.=1 b-II
11+(2.-1)211+1

b-II

- J
11+ 211 + 1

(x - a) f'(Eo) dx
II'
b-II
II'+(!.+1)211+1

+i
• =1
! J b-II'
(x - a- 2v:n~a1Y f"(E.) dx .
11'+(2.-1)211+1

The limit in question is


(b ~4a)2 [j'(b) + 2f'(a)].
In the case of f'(a) = 0 tJ:lis result is a consequence of 11: extend the
function I(x) considered in 11 to the left of a by reflection and substitute
2n + 1 for n.
1
13. In 10 and 11 set I(x) = -1- , a = 0, b = 1. Cf. also 5.
+~
14. [For more details d. G. N. Watson: Lond. Edin. Dubl. Phil.
Mag. Ser. 3, Vol. 31, pp. 111-118 (1916).]
II-I~---:-;;-
1

.=1 Sln-

II-l(-1- - -1- - -1)


n

=-
n 1
2(n-+-+
n ... +n)
- 2
- -1n - 1
+n~
sin 'lin 'lin (n - 'II) n n
n n n
.=1

=2n
n
[logn+c+o·(-.!..)]+n[l(-.
\ n 0
1
sm nx
_-.!...._ n(l 1- )dX+O(-.!..)]
n n~ ~)

= -
2n
n
(log n + C) 2n
- -n log -2
n
+ 0(1),
with the help of 9.

yuana97@uw.edu
234 Integration

15. [E. Cesaro, Problem: Nouv. Annis Math. Ser. 3, Vol. 17, p. 112
(1888). Solved by G. P6lya: Nouv. Annis Math. Ser. 4, Vol. 11, pp. 373-
381 (1911).] Put I(x) = x log x, a = 0, b = 1 in 18. Although the hypo-
thesis of 10 is not completely satisfied the conclusion of 10 remains
valid.
18. Write P(x) = 21
x
+i .=)
_1_,
x - 11
{J = (1 - e-"')-). The equation
P(x) = IX has degree n + 1 and possesses one root in each of the intervals

(0, 1), (1, 2), (2, 3), ... , (n - 1, n), (n, (0). The largest root is x". In 12
put I(x) ={J.~'
-x
a = 0, b = 1. We find

J(J:
1

x = IX = P(x,,) = P(n + l) (J) + A~, A~>O.


o
Since P(x) is decreasing for x > n we have x" < (n + 1) {J. The mean
value theorem implies

x" < E< (n + l) (J.


The quotient on the right hand side converges to 0 [12] because

J' til _
1
dx
n P'(n + l) (J) ~ - xli'
o

J~ dx leads to (J =
1
17. The equation IX = ~ 2/J 1+ e-'"
_ . The proof
,..-x l-e'"
o
is similar to the one given in 18.
18. [J. Franel. Related to Euler's summation formula, d. Knopp,
p. 523.] We write F(x) = J" I(~) d~. Summation of the equations
1

F(v+ I) - F(,-) = l/(v) + 11'(E.), }v< E. <v +1 < "I. < v + 1;


-F(v + I) +F(v + 1) = it(v + 1) -11'("1.) v = 1, 2, 3, ... , n - 1,

yields
l/(l) + 1(2) + 1(3) + ... + I(n - + l/(n) - F(n)
1)

= !fj'(TJl) - I'(E1) + I'(TJz) - I'(Ez) + .. , + 1'("1"-1) - I'(E"_I)]'


The series

yuana97@uw.edu
Pt. II, Solutions 15-19.2 235
is convergent because the terms have alternating signs and their absolute
values converge monotonically to O. In the case of I' (x) < 0 we find

~ I'(n) < ~I'(~.. ) < - ~ U'(17 .. ) -I'(~.. ) +1'(17"+1) -f'(~"+1) +···]<0.


This proves for I(x) =..!.. the existence of
x
lim (..!.. +..!.. +..!.. + ... +..!.. -logn)= C
.. ~CO 1 2 3 n

(Euler's constant) and furnishes the inequalities


11111 1 1
2n - 8n2 <
T + "2 + 3" + ... + n- log n - C < 2n .
For I(x) = -log x we find
log n! = (n + 1) log n - n + 1 - s + 8 ...

where s is a constant and 0 < 8 .. < 8~. Stirling's formula [205] states
that 1 - s = log V2n.
19. Cf. solution 18. The sum mentioned there.
UI'(171) -1'(~1) + 1'(172) -1'(~2) + ... + 1'(17"-1) - f'(~"-l)]'
is positive in this case. Furthermore
1'(171) -/'(~2) < o. /'(7/2) -/'(~3) < O..... /'(17 .. -2) - f'(~"-1) < o.
/'(~1) > /,(1), /'(17 .. -1) < I'(n).
19.1. (£.5. The relation
lim (H" -log n) = lim (H 2.. -10g2n)
n~~ ft~~

implies
log 2 = lim (H2 .. - H .. ) .
.. ~co

19.2. [G. P61ya; see Research Papers in Statistics. Festschrift for


J. Neyman. New York: Wiley & Sons 1966. pp. 259-261.] Use notation
5. From
lim (2H.. - 2logn) = 2C,
,,~co

lim (H... - log n 2 ) = C


follows that C is the limit of
2H,,- H"'=T
2 +.!. 2n
4 ••• +-;;z
1 1 1 1 1 1
- T - "2 - 3" -"4 - 5 - ... - n2
1 1 1 312ft - 1
=T-"2-3" +"4-5- ... +-n- 2 -'

yuana97@uw.edu
236 Integration

which coincides with the partial sum of the proposed series that has
1 + 2 + 3 + .,. + (2n - 1) terms. It still must be shown that the
limit of this subsequence of partial sums is the sum of the series. (Easy.)
20. [Cf. e.g. l.c. 9.] Let I(x) be monotone increasing. [Otherwise
consider -/(x).] Then

1 J~/(X) dx < I( ~) + 1(~): ... + I(~~) ~ JI(x) dx.


o -1
"
The condition that the function is monotone increasing is essential only
in the neighbourhood of the singular points.
21. We can assume that the function I(x) increases [20] and that
I(x) > 0 [otherwise decompose I(x) =/(x) +21/(x) 1 +f(x) -21/(x) 1 and
examine the two terms separately J. Let 6 > 0, 'YJ be chosen such that
1
o < 'YJ < 1, J I(x) dx < 6. Then we have
1-'1
1 ((1-'1)"] 1-'1
lim -
,,~OO n
~ qJ(~)
.=1 n
I(~)
n
=
0
J qJ(x) I(x) dx.
If, on the other hand, M denotes the least upper bound of 19'(x) 1we can
write

I! "i 9'( :) I (:) I~ ~ "if


1
I (:) < M /I(x) dx < M6 .
• =((1-'1)"]+1 .=((1-'1)"]+1 1-'1

22. In 20 set I(x) = X"-1.


23.
a.. = 1"-l(n - 1)/1- 1 + 2"-I(n _ 2)/1- 1 + ... + (n _ 1)"-1 1/1-1

= n,,+,8-1
.. -1
~ ~ (~)"-1 (1 _ ~)/I-l N n,,+,8-1 J 1
X,,-1 (1 _ X)/I-l dx .
• =1 n n n 0

Cf.20.
24.
1 1
I(x) =x - -1 - .
- x
25. Let I(x) be monotone decreasing and finite for x = 1. The in-
equality [20]
1 I(~) + I(~) + .. , + I(~)
J I(x) dx < n n n n

.
1
-

yuana97@uw.edu
Pt. II, Solutions 20-30 237

implies that the left hand side is bounded, i.e. that


1
lim
.~+o
J I(x) dx

is finite.
26. Let I(x) be monotone decreasing. Then
2n-1 2n-1

1 2
-n I (~)
2n
1
+~
1
J I(x) dx <- -n1 ~=1
n 2
2: 1 (~2n
1)
< -I (2n- + 1J I(x) dx,
1
- n
1) 2n

2n 2n

and a fortiori

r
2n-1 1 2.. --1

2
2n-1
j I(x) dx + JI(x) dx < ~
1
j; 1 ClJ 2~ 1) < 2
~=l 0
I(x) dx + j" l(x) dx.
1
2n 2"

Similarly

2 [i-] (2lJ _ 1)
lim -
n~oo n
2: 1
v=l
--
n
= J I(x) dx.
0
1

27. Cf. 28 for I(x) = X"-l.


28.
n-1 [i-] ,,-1
~ 2: (_1)"-1 I(~) =.! 2: 1(2lJ-=--!.) - ~ 2: 1(~).
n ~=1 n n ~=1 n n ~=l n

[20 and solution 26.]


29. Cf.26.
30. Since I(x) is monotone and lim I(x) = 0, I(x) cannot change its
X~~

sign. Assume that I(x) is positive and monotone decreasing. Then we have
(m+l)k mk
J I(x) dx < h(t(h) + 1(2h) + ... + I(mh)) < J I(x) dx,
h 0

and so for m --+ ex:>

J I(x) dx ::;: h 2: I(nh) < J I(x) dx.


~ ~ ~

k ,,=1 0

The condition that I(x) be monotone is essential only for large x and in
the neighborhood of x = o.

yuana97@uw.edu
238 Integration

31. By setting I(x) = e-zx"'-l, e- A = t in 30 we get


r(lX) = lim (log~)" (1 .. - 1t
1-+1-0 t
+ 2",-lt2 + 3.. - 1t3 + ...)
00 .. -1 ,
= lim (1 - t)'"
1-+1-0
1: n.. - 1t" =
,,=1
lim
11-+00 IX(IX
n
+ 1) ... (IX + n
n.
- 1)
[189].

32. Introduce in 30 I(x) = e- z ( 1 - ~), e- IJ = t. We have


1 - e- z x

00 e- IIA 1
h 1: --nh = log -A'
11=1 1- 8
-Z
33. In 30 put I(x) = _8_ _ _Z , e-" = t and notice that
1 + 8-
r+oo
8
-z
dx=
II 1
d
+y =log2 .
.. 1 8- z y
o 0

Or apply 32 and note


00 til 00 til 00 t211
1: --II = 1: --II - 2 1: --2-" .
11=1 1 +t 11=1 1 - t 11=1 1 - t

34. The statement follows from

o
r oo
xe _zdx=
1 -
-z

8 0
Jx ':" (
1: e- liz
11=1
00 )

dx = 1: n
11=1
00 1
2 '

We can also argue in the following manner: we have [VIII 48, VIII 65]
00 II 00

1: n .. _ t -II = 1: O'.. (t) til,


11=1 ~ - t ,,=1
1
1: O'.. (n) t" = 1: (0'.. (1) + 0'.. (2) + '" + O'..(n») t".
00 00

1_ t
11=1 11=1

Noticing 45 we now apply I 88.


35. In 30 set I(x) = e- z', e-'" = t, respectively I(x) = e- z", e-"" = t.
36. The limit is n. Introduce I(x) = -+x
1
2 2' h-1 = t in 30. Observe
the formula
1 2t 2t 2t em + 8-,11
T + t + + t + 22 + ... + t + n + ... =
2 12 l 2 l n ent _ e- nt

[Hurwitz-Courant, pp. 122-123.]

yuana97@uw.edu
Pt. II. Solutions 31-42 239

37. The proposition follows from 30: I(x) = log (1 + x-a), h = t '"
Transformation of variables and partial integration lead to

r
1
--
J
~ ~

log (1 + x-"') dx = u 1
- '" du = _:rl_ •
. +u .:rl
o 0 SIn-;x

38. Apply 30 with I(x) = log (1 - 2X-2 cos 2cp + .x-4). Write
t = -~- ei'l', use the formula for sin t and square the absolute value on
k t
both sides:

39.
a
J log x dx < 1: (at -
~ 00

1: (aqn - aqn+l) log aq"+l < aq"+l) log aq"


n=O 0 ,,=0

aq logq
= a log a + - - - ~ a log a - a
1- q

for q -)- 1. A more general proposition can be deduced in analogy to 30.


40. Taking 58 into account we obtain

!!:.- k-l ]-

= 2k"Grn-~(~r·
For more details see e.g. Jordan: Cours d' Analyse, Vol. 2, 3rd Ed. Paris:
Gauthier-Villars 1913, pp. 218-221.
41. [For problems 41-47 d. G.P6Iya: Arch. Math. Phys. Ser. 3,
Vol. 26, pp. 196-201 (1917).J
42.

-
"
= l-lim(l +-~+!- + ... +~-logn)=l- C,
"-+00 2 3 n

yuana97@uw.edu
240 Integration

where C is Euler's constant. - Define


1
if>(~) =f 1 - x"
1- x
dx = F'(rx
r(rx
+ 1)
+ 1)
+C
o
and note that the relation
1 1
J~ dif>(~) = if>(1) - J if>(~) d~ = 1- C
o 0

holds; i.e. the operations of taking the limit and computing the mean
value can be interchanged [44].
43. [Cf. Cesaro, Problem: Nouv. AnnIs Math. Ser. 3, Vol. 2, p. 239
(1883). ]
n 1

lim 1_ ~ (1 - [_n-J~) = 1- J'[~J x dx


H"""*= n
.= 1 v n • x
0

44. [G. L. Dirichlet: Werke, Vol. 2. Berlin: G. Reimer 1897, p. 97-


104; d. also G. P6lya, l.c. 41, p. 197 and Nachr. Akad. Wiss. G6ttingen
1917, pp. 149-159.] We are dealing with [VIII 4]

f' ([~J - [~ - ~J) dx


1

= lim
n-+= • X X
= lim
n-+='--
n~ (~
V
- -~-)
V + ex
1 .=1
--
n
1
1 1 1 /' 1 _
= 1 -l+rx +2 - 2--+-", + .. , =. x-'
~ dx.
o -,
If ~ = t
we obtain 41.
45. [G. P6lya, l.c. 41, pp. 1£19-200.] We assume at first that ~ > 1.
Then
1

(~ + 1)
o
r[~J
1

x
x'dx = i.; n f
n=l I
n

(~+ l)x'dx= 1 +_1_


2~+1
+ _1__ + .. '
3~+ 1

n+l

= C(~ + 1).

yuana97@uw.edu
Pt. II, Solutions 43-50 241

The total variation [d. solution 9] of [! ] xo< = I(x) is

(1(1) -/(1 + 0») + (1(1- 0) -/(1 + 0») + (/(1- 0) -/(i + 0») + ...
= 1(1-0< - 2-0<) + 2-" + 2(2-0< - 3-0<) + 3-0< + ... = 2C(IX) - 1
whence the two statements follow [9]. The limit relation holds also for
IX = 1. If 0 < IX < 1 we have to examine (! - [! J)xo< and use 22.
46. [G. P6lya, l.c. 41, pp.200-201.] We write I(x) = ! - [! J;

J I(x) dx =
]

according to 42 1 - C; furthermore, the total variation


o
of I(x) in the interval (~ , 1) is 2(m - 1). We have

: ; f I(x) dx -
...
1

~ I
,=m+1
..

I ( :) I f
...

+ .=1
..

\1 e: 1 + x) -I (: ) \dx •
.
-
0

Since m
n
>..!..
m
the first term is not larger than 2(m - -.!l [9]; the second is
n
smaller than ~.
n
47.

i (0. -
.=1
E,) = i (-It-
.=1
1 [: ]

..
([:J - :) +n I
.. .-1
= I (_1),-1 (-1~ .
• =1 .=1

The first sum on the left hand side divided by n converges to 0 as n ~ 00


[28].
48. Consequence of the definition of the definite integral.
J log x dx
1
49. Special case of 20 for I(x) = log x. With regard to
d. 0

SO. Put c = =. Then


.. c + 1 c + 2 ... c + n - 1
G.. V~
n n n n
An - c n-1
[29].
-+--
n 2n

yuana97@uw.edu
242 Integration

2"
51. A .. = n + 1" Moreover

..
because ~ (n +1 - 2'P) = O. Making appropriate use of 20 we obtain
0=1

lim
"-+00
~n log G.. = lim
"-+00
i
~n 0=1 (1 -~)
n +1
log (1 _ --"-)
n +1

1 1
= f (1 - 2x) log (1 - x) dx = 2"'
o
52. In 48 set I(x) = 1 - 21' cos X + 1'2 = 11' - e'" 12, a = 0, b = 2n.
The identity
r" - 1 = n" (1' -
0=1
e2"'P,")
implies
·1 ... I
11"'2...... = (r" - 1)2 .
If l' = 1, I.... must be omitted (it vanishes); notice 20.
53. [G. Szego, Problem: Arch. Math. Phys. Ser. 3, Vol. 25, p.196
(1917). Solved by J. Mahrenholz: Arch. Math. Phys. Ser. 3, Vol. 28,
pp. 79-80 (1920).] According to the hypothesis e-'''(eie - 1') is real, i.e.
the arguments of e'" and eie - l' are equal or differ by n. Since ~ is the
number closest to x with this property e'" and eie - l' have the same
argument, which means that eie is the point of intersection of the ray
from l' parallel to the vector e'" with the unit circle. If, therefore,
o < x < n and ~' is the argument that belongs so to x + n as ~ to x,
then eie, 1', and eW are on the same line. Thus we obtain by elementary
geometry
Ieie' - 1'12 lie - 1'12 = (1 - 21' cos f + 1'2) (1 - 21' cos ~ + r) = (1 - r)2 ,
hence
1 ,.
-2
no
f [log (1 - 21' cos ~ + 1'2) + log (1 - 21' cos~' + 1'2)] dx
= -2n1 0flog
"
(1 - 1'2)2 dx = log (1 - 1'2).

yuana97@uw.edu
Pt. II. Solutions 51- 57 243

54. We have [Maclaurin series]


Ilog (1 + x) - x I < x2 for Ix I < t·
Assume I/(x) 1< M. Whenever ()n M <t

I.~: log (1 + I.n()n) - .~ I,n()n \ < ()n .~ I:n()n'


The sum on the right hand side converges to an integral. Cf. 67. Other
subdivisions of the interval [a. b] may be considered instead of the sub-
division by points in arithmetic progression, and we can choose any
point in a given subinterval as the point where we evaluate the function.
There is an obvious analogy between the integral as a limit of sums and
the limit of products considered.
55. According to 54:
1
1 f xdx
+nn
v

hm
n...,.= n n 1
V 1 1
=--
eO
=e.
• =1 1 - - - _ f xdx
nne 0
56. The product in question is
1 ·3 ·5··· (2n - 1) IX" • 2"
._------
[(n + 1) IX - 1] [(n + 2) IX - 1] ... (2nlX - 1)
(n + 1) IX + + n) IX
- 1 • (n=t--2f;X'-=1 ... (n + n)
(n 2) IX (n
= (n +~ IX - 1
1

1
- + - - - 1- [54].

e
-~ 0f 1:"
The particular case of <X = 2, i.e.
1357911 1
2' 2' -6 . 6' 10 . 10'" = V2
follows also from the product representation of cos x at the point x =: .
[Euler: Opera Omnia, Ser. 1, Vol. 17. Leipzig and Berlin: B. G. Teubner
1915, p. 419 (distorted by a misprint).]
57.
a+vd IX-fl 1
· · · - - 1 +0- - - (J-+-t·O
--
0
b + vd - n .
1+--
n

yuana97@uw.edu
244 Integration

Make use of the remark to solution 54 for I(x) = ~-1- 1 ~ x on the inter-
val [0,6].
58. Let n be even, n = 2m, v i;;."!. -+ A V"2, moreover A ~ 0, v> m.
Then we get

C;)
e:) = m
m
+
m - 1 m - (v - m - 1)
1 m + 2 ••• m + (v - m)

1 1 v-m-1 1
1 - -- -- 1 - .Im .I-
1 _~;;; Ym ... r rm
1 1 2 1 v-m 1
1 + V;;; V;;; 1 + V;;; V;;; 1 + V;;; V;;;
AV2
- I xlix
e 0 -2A'
-+-Ay2 - = e [54].
Ix"
eO

2m) 22m
Notice that ( m ~~ Ymn [202], and that, furthermore,

( 2m.: 1 ) ( 2m )

c:::: :) r m+1
= -v- c:) .
v-1

59.

.
~n
0=) 1 +

We can justify the substitution of .~_ for sin -/-;_ by expanding


2rn 2rn
sin .~_ in powers of ./_ and taking the logarithm of the product in
2 rn rn
the same manner as in 54.
60. We write for the second difference in question

F(b, d) - F(b, c) - F(a, d) + F(a, c) = LJ 2F(x, y),


R

yuana97@uw.edu
Pt. II. Solutions 58-63 245

thus
m n
LJ2p(X. y) = 1: 1: Lf2p(X, y).
R ,,=1 v=1 R"v
The mean value theorem of the differential calculus applied to
G.(X) = P(X, Yv) - F(x. Y.-J)
leads to
LJ2P(X, y) = G,(X,,) - G.(X,,_l) = (x" - X,,_J) (F:(~". Y.) - P:(~I" y.-l))
R".
= (x" - X,,_J) (Y. - Yv-1) t(~1" 'YI.), X,,_1 <~" < x"' Y.-l < 'Y}. < Y.·
Cf. 7.
61. Multiplying the determinant in question by its complex conju-
gate roW by row we find
1 1 + i-" + e 2 (J.-,,) + ... + e(n-1)(J.-,,) I
J..,,=O) •...• n-l
= nn.

On the other hand we have (Vandermonde's determinant)


)2
n
O.1 ..... n-1

j<k
.
Ie} - ll2 = TI
.
O.J ..... n - l ( . _

j<k
k
2 sin L __ .n .
n
Hence we conclude by comparison

1:
O.1 ..... n - J 2 I
~2 log sin (J:rt - -..:::
. k) I = ~ log n - (1)
2
1 - - ~ log 2.
2

j<k n Inn 2n n 2

Argument 20!
62. Cf. 54.
63. We denote the expression in question by IIn' Then the inequality
used in solution 54 yields for sufficiently large n

Ilog IIn - n21"~1'~ t (: ': I< n'.~


n 1 n ) n [
:E t (: ' :,,)]2
n

Cauchy's inequality [80] provides an upper bound for the right hand
side:

This expression converges to 0 as n -)- 00. Therefore


1 1
J f f(x.y)dxdy
lim IIn = eO 0
n""'oo

yuana97@uw.edu
246 Integration

64. [Cf. G. P6lya: Math. Ann. Vol. 74, pp.204-208 (1913).J Sub-
divide the space by the three sequences of planes
5 3 1 3 5
x, y, z = ... , - 2n ' - 2n ' - 2n ' 2n ' 2n ' 2ft' ••.

into cubes of volume~. The expression in I 30, with s = [no"], gives the
n
number of cubes whose center is in c.i). This number multiplied by n-3
converges to the given integral.
65. We have

"'''' ... ''' V",]-lv",·-1···v"'I>-1- 1(n-v -v - " ' - v


=
kI kI kI 1 2 1>-1 1 2 I> -1 )"'1>-1 '
'] +r'+"'+"1>_1:;;;"

thus

a..
"']+""+"'+"'1>-1 =
1
1>-1 ~ ~ •.. ~ --;;-
("1 )",]-1 ("2--;;- )",.-1 ••. ("1>-1
-n-
)"'1>-1-1
n n .]+ •• + ...•1>-1:;;;.. .

. (1 _ "1 _ "2 _ ... _ "1>-1 )"'1>-1 .


n n n
Cf.23.
66. We use the same notation as in solution 65. According to solu-
tion 31 we have for t --? 1 - 0

k = 1, 2, ... , p.
Introducing
00

F(z) = ~ n""+"'·+···+"'I>-lz..
.. =1
we obtain
F(t) <Xl F(CX 1 + CX 2 + ... + cxt ) (1 - t)-(""+""+"'+"'I» ;

• II (t) Mt) ... II>(t) r(~1) r(~2) ... r(~I»


lim = .
t--..l -0 F(t) r(~1 + ~2 + ... + ~I»
On the other hand this limit is, according to I 85 and 65, equal to the
integral in question. The p-fold Dirichlet-Jordan integral can be easily
computed with the help of this relation, d. E. T. Whittaker and G. N.
Watson, p. 258.

yuana97@uw.edu
Pt. II, Solutions 64-69 247

67. The term in question is


~! ~~ ... ~ I.,nf••,,··· I.p"
1 ;;;;"<"<"'<'P;;;;"

-+ JJ... JI(x l) l(x2 ) ••• I(xp) dX l dX2 ••• dxp


a;;;;",;;;;x.;;;;".;;;;xp;;;;b
1
J J ... J I(x
b b b
= p! l) !(X2 ) ••• I(xp) dX 1 dX 2 ••• dxp.
a" a

Moreover [I 62J

n" (1 + zl.,,~n) ~ (1 + zM~,,)" ~ e zM6 ,," = ezM(b-a) •


• =1

This inequality combined with the above proved proposition implies


[I180J

,,~n;, .Ii (1 + zl.,,~,,) = 1 + ;! j I(x) dx + ;~ (j I(x) dX) 2 + ...


b

+ PI
zP ( b
! I(x) dx )P
+ ... = e
z J f(x)dx
a

This new solution of 54 illustrates well the limit operations which lead
to Fredholm's solution of integral equations. Conversely 67 can be
deduced from 54 with the help of I 179.
68. MultiplicatioR by rows leads to a determinant P of order m with
the general term
n b
L.: f~~({J~)
,,=1
C'V b: a
a
J fA (x) ({JJl(x) dx; A,P. = 1, 2, .... , m.

On the other hand we have (n > m)


j(l) j(l) •• 'j(l) mIl) mIl) ••• m(l)
'''In "tn "mn Tv1n TV.H """"mH

1 (2) 1(2) •• 'j(2) m(2) m(2) ••• m(2)


P= ''In v,n "mH ...,..V1H TJ.:!H '1'''m''
] ;;;;"<"<"'<'m;;;;n ................ .

If VI' V2 , ••• , Vm assume independently all the values 1, 2, ... , n we


obtain m! P. The sum established in this way is C'V (_n_)m_times the
b-a
m-fold integral exhibited in the problem.
69. [For a fuller account of the subject studied in the present
Chap. 2 see G. H. Hardy, J. E. Littlewood and G. P6lya: Inequalities.

yuana97@uw.edu
248 Integration

Cambridge: Cambridge University Press 1952.] The statement follows


as a limit relation from the proposition on the arithmetic, geometric and
harmonic mean [48].
When we pass to the limit ">" becomes ">-", and so our proof is
not suitable for specifying the cases in which equality is attained. For an
essential remark on such cases see 109.
70. [J. L. W. V. Jensen: Acta Math. Vol. 30, p.175 (1906).] The
proof is analogous to Cauchy's proof for the inequality between the
arithmetic, geometric and harmonic means given in the footnote on
p.64 (which deals with the case qJ(t) = log t). First the statement is
proved for n = 'i' (k integer), then it is extended to arbitrary n.
71. [J. L. W. V. Jensen, l.c. 70.] Using a similar notation as in 48 we
get for each n
/ ..
( 1
+ 12.. + ." + I,.,.) < fP(f1,.) + fP(f2,.) + ... + fP(f,.,.)
qJ n = n • /
Let n increase to infinity and notice 124, 110.
72. Let t1, tz be two arbitrary points on [m, M], t1 < tz' Then

qJ (t 1) -qJ
_ (t1+ t2)
---2- + -2-qJ
t1 - t2 (t1 + t2) + (tl - t2)2 "( )
-2-
I
8 qJ!l'

t1 + t2)
qJ (tz) =qJ ( - 2
+ -2-qJ
12 - 11 (t] + 12) + (I] - 12)2 "( ).
2-I
8 qJ Tz,

h
were t1 < T1
t]+/s
< -2- an d -2---"
/1 + t•
< T2 < t2' H ence

qJ(t1) + qJ(tz) - 2IJl (~~ 12). > 0,


provided qJ"(t} > 0 on [m, M].
73. [72.]
74. [J. L. W. V. Jensen, l.c. 70.] In the case where the Pv are integers
the proposition is a consequence of 70 where PI points coincide with t1 , pz
points with t2 , ... , P.. points with t,.. Then we extend the proposition to
rational Pv; for arbitrary Pv we need the continuity of qJ(t) [124].
75. [J. L. W. V. Jensen, l.c. 70.] Introducing

I~,. = 1(Xl + lI X2 : Xl), P." = p(XJ + lI X2 : Xl), 11 = 1, 2, ... , n,


we obtain according to 73
P1n/1,. + P2n 12" + ... + Pnnl,.,.)
(
qJ P1,. + P2n + ... + Pn,.
< P1,.fP(/1,.) + P2 nfP(/2,.) + ... + P....fP(/,.,.)
= P]n + P2n + ... + P,.n
Let n increase to infinity.

yuana97@uw.edu
Pt. II, Solutions 70-81 249

78. [0. Holder: Nachr. Akad. Wiss. Gottingen 1889, p. 38.] We put

then

thus

provided that at least one of the t,,'s is different from M.


77. [Cf. G. P6lya,Problem: Arch. Math. Phys. Ser. 3, Vol. 21, pp. 370-
371 (1913).] Analogous to 78.
78. In 78 set tp(t) = -log t, and t log t resp., M> m > 0; further-
1
more replace a. by -, v = 1, 2, ... , n.
a.
79. In 77 put: tp(t} = -log t, and t log t resp., M > m >.0; then
1
replace I(x) by f(x) •
80. First proof:

Second proof: If ). and p denote real variables the quadratic


form
().al + pb l )2 + (Aa 2 + pb2)2 + ... + ().a" + pb,,)2

= A).2 + 2B)'p + Cp2 ?:: o.


Provided that ).2 + p2 > 0 the case of equality presents itself only then
when there is a particular set of )., p for which ).a. + pb. = 0,
v = 1, 2, ... , n. Therefore AC - B2 is positive or 0 as asserted.
81. By taking the limit in 80: Writing I." = I (Xl + V XI : ~) ,
g... = g (x] + vXs -: ~!) , we obtain [80]
(Il"gl"
' /,
+ J2"g2" + ... + .."g.... ) 2 $; 121.. +. f,22" + ... + /,2"". g11l
2 2
+ g2" + ... + g"2...
n - n n

Let n increase to infinity. It is also possible to adapt both methods used


in 80 to the present problem; as to the first method cf. 68.

yuana97@uw.edu
250 Integration

81.1. See 81.3.


81.2. See 81.4.
"
81.3. Let E stand for :E , set
.=1
Ea. = A, Eb. = B, ... , Et. = L,
and use 78:

a. b.
<:E ( IXA+{JB +···+J. L
1.)
=1X+{J+···+J.=1.
81A. From 81.3 by a passage to the limit or by analogy. Analogy
may be better: It may allow us to discuss the case of equality.
82. Let t =1= 0 and introduce a! = A., '/I = 1, 2, ... , n. Because of 78
we have
f- ",'(t) = Al log Al + A2log A2 + ... + A"log A"
",(t) + A2 + ... + A"
A)
Al + A2 + ... + A
- log n
"> o.
We find
"'(- 00) = min (a), ",(-1) = .t>(a), ",(0) = @(a). ",(1) = ~(a),

"'( + 00) = max (a).


Thus we have a new proof of the proposition on the relation between the
arithmetic, geometric and harmonic means.
83. Assume t =1= 0 and set [/(x)]' = F(x). Proposition 79 implies

J F(x) log F(x) dx


(_1__
%1
%

t2 1[1'(t) =
1[I(t)
%, _ 10
g x2
J1F(x) dX) ~ o.
J F(x) dx
%1 - Xl -
%,
%,

(Or taking the limit in 82.)

P(-1) = .t>(t), P(O) = @(t), P(1) = ~(f).

Let M denote the maximum of t(x) on [Xl' x2] and c) the length of a
subinterval of [xl' x2] in which t(x) > M - e. Then we have for t > 0
1

(M - e) C2 ~ ~)' < P(t)::;; M,

yuana97@uw.edu
Pt. II. Solutions 81.1- 87 251
i.e. !P(oo) = lim !P(t) = M; !P(- 00) is found in a similar way. This
''''00
proposition contains therefore a new proof for 69.
M. [H. Minkowski; cf. e.g. Hardy. Littlewood and P6lya. l.c. 69,
p. 21.] First proof: We assume 0 S;; t S;; 1 and define
,.
n
1
<p(t) = Eta, + (1 - t) b.]". Then [80]

1II"(t) 1 (,. a. - b, )2 1" ( a. - b. )2


lII(t) = nl ~ tap + (1 - t) b, - -;;- ~ tap + (1 - t) b. < O.
unless a. = Ab,. v = 1, 2•...• n.
Second proof: With log bp - log a, = t. the inequality becomes

log (1 + e, ') + log (1 + : I ") + ... + log (1 + eI ,,) > log 1 + e--"-- . ("+, +. . +,,.)
1

log (1 + ttl. however. is convex [73].


Third proof: Particular case of 81.3:
n = 2. IX = fJ = ... = A.
Also particular case of 90, k = O.
85. [Cf. W. Blaschke: Arch. Math. Phys. Ser. 3. Vol. 24, p.281
(1916).] We define
- 1
- "I
Z'-%J
Jiog[I/("l+(l -I)g(")]d,,
<p(t) = e '" ,
Schwarz's inequality implies [81]

1II"(t)
( 1 "I f(x) - g(x) d )2
lII(t) = XI - Xl! tf(x) + (1 - t) g(x) X

"I
__1_J(
Xa -Xl
f(x) - g(x)
tf(x) +
(1 - t)g(x)
)2 dx -
S;; 0
'
'"
(Or take the limit in 84.) Particular case of 91.
86. By repeated application of 85 to the functions
P1/ 1 (x). P2/ 2 (x), .... P...I... (x).
87. By definition
,.
1k = least upper bound of 1: V(x(P) - X(·-1»)2 + [/k(X(») - MX(·-1»)]2
.=1
for all possible subdivisions of the interval [Xl' X 2 ]

(Xl = x(O) < X(l) < X(2) < ... < X(,,) = x 2).

yuana97@uw.edu
252 Integration

Since V +t
C2 2 is convex [73] we get for an arbitrary subdivision

Pl l 1 + p212 + ... + P".l". > 1: 1=1


.. i;p "
V(x C.) - x C. - 1 ))2 + rt (xc.)) - t. (x
" "
C. - 1))]2

PI + P2 + ... + P". -.=1 i P

~
.
1: V(xC') - XC
.- 1 )2 +
1=1 "

[F(xCp) - F(XC
.- 1 ))]2 •
• =1
88. Put

P~") =
.-J.
2"

P(E)~, v = 1, 2, .•• , n;
C._1)2"
..
Then the sequence F .. (x) converges to F(x),
lim F .. (x) = F(x) ,
....... 00

uniformly in x, 0 :::;;; x :::;;; 2n, because

1'#'f._!.~P(<l [m +.) - t((> - 1) ': +.)] 4< < w.. IlJP(E)~,


2"

where £0.. denotes the maximum of the oscillation of f(x) in an interval of


length 2n [p.77]. Hence QJ(F) = lim QJ(F.. ) ; because of 88 QJ(F.. ) ~QJ(f).
n ft~~

89. Using the same notation as in 88 we establish that limF.. (x) =F(x)
uniformly in x, 0 =:;; x =:;; 2n: ....... 00

It /
I .2" I
.-1 ( 1)2"
P(E) [f(E + x) -/((v - 1) 2: + x)] ~I
P- .. II
2"

=:;; max (p)i [.~ 1/((v -1) ~ + E+x) - f((v -1) 2: + ~)I] dE.
Thus
IF.. (x) - F(x) I =:;; 2: ~ax CJ!l V,
JP(E) dE
o
where V denotes the total variation of f(x) on [0, 2n]. Since the length I
of the arc of f( (v - 1) ~: + x), 0 =:;; x =:;; 2n, is the same for each v,

yuana97@uw.edu
Pt. II, 'Solutions 88-92 253

the length L,. of the arc of F,.(x) can be estimated by L,. ~ I. Besides,
the limit relation
s
~ V (xl") - x l.. - 1))2 + [F(x l")) - F(X I.. - 1))]2
.. =1
s
= lim ~V(xl"') - X I",-1))2 + [F,.(xl"')) - F,.(XI"'-1))]2 ~ l
"-+00 ",-1

holds for any arbitrary subdivision of the interval [0, 2n],


0= x(O) < XII) < ... < X(s-l) < xIs) = 2n.

A particularly interesting special case of this problem is due to


F. Lukacs: The arc lengths of Fejer's means of the Fourier series of I(x)
[134] cannot be larger than the length of the arc of y = I(x). over the
interval [0, 2n]. (The "jumps" have to be included in the length of the'"
arc, also I/( +0) -/(2n - 0) I.)
90. Assume "=1= 0 [84]. Put A. = ta. + (1 - t) b., 0 < t ~ I,
v = I, 2, ... , n and «p(t) = IDl,,(A). The second derivative is
.!..-2
«p"(t) = (" - 1) (A~ + A; + ... + A:J"
{(A~ + A; + ... + A:) [(al - b))2 A~-2 + (a2 - b2)2 A 2- 2
+ ... + (a,. - b,.)2 A:- 2 ]
- (al - b1) A~-l + (a2 - b2) A;-l + ... + (a,. - b,.) A:-?}.
The quantity in curly brackets is always positive unless a. = lb. [80].
Thus sgn «p" (t) = sgn (" - 1). Therefore we have

2«p ( ~ ) ~ or > 91(0) + 91(1)


according as " :? 1 or " < 1. If" = 2, IDl" represents the distance of
the point aI' a2 • ••• , a,. in the n-dimensional space R" from the origin.
The proposition states in this case that one side of a triangle is shorter
than the sum of the other two sides.
91. In 90 put a. = l(x1 +VX!:x1), b. =g(Xl +V 2 : X1 ),
v = I, 2, ... , n and let n become infinite.
92. [For the special case a.b. = I, v = I, 2, ... , n, d. P. Schweitzer:
Mat. phys. lap. Vol. 23, pp. 257-261 (1914).] We rearrange the numbers
a. so that a l < a2 < ... ~ a,.. To determine the maximum it is then
sufficient to consider values bl :2:: b2 > ... :2:: b". (If b. < bp ' v < /-t, we
interchange b. and bp: b! + b! = b! + b; and apbp + a.b. :2:: apb. + a.bp.)

yuana97@uw.edu
254 Integration

We may also assume that not all the a7 's are equal. nOl all the b/s. i. e.
a,.b l - alb,. = (a,. - al)b l + al(bl - b,.) > O.
If n >
2 the numbers u l • u 2 • •••• U,._l' VI' v2 • •••• V"_l are defined by the
equations
a; = u.ai + v.a! • b; = u bi
7 + v b! •
7 'JI = 2. 3•... , n - 1.
We find u. > 0, v. > 0 and a.b. > u7 al bl + v.a,.b,. [80]; u. = 0 if and
only if a. = a.+ l = ... = a,., b. = b.+ l = ... = b,. and v. = 1. In a
similar way v. = 0 implies u. = 1, etc. If u. > O. v. > 0 then
a.b. > u.al bl + v.a,.b,.. Thus the expression in question is
< (pai + qa~) (pbi + qb~)
-
- (pal bl + qa,.b,.) 2
where 1 + u 2 + Us + ... + U"_l = p, v2 + va + ... + V"_l + 1 = q.
The inequality becomes an equality if and only if the u.'s and v.'s are
o or 1. P. q. are integers, a l = a2 = ... = ap • ap + l = ap +2 = '" = a,..
bl = b2 = ... = bp • bp +1 = bp +2 = ... = b,.. The last expression is

= 1 +pq( a,.bl - alb,. )2 <1 +pq (a,.bl - alb,. )2


pal bl + qa,. b,. 2 Vpal b] qa,. b,.

~ (v~: V::~)'
it is an equality if and only if palb l = qa,.b,.. If we replace a l • a,., bI> b,.
by a, A. B. b, in the term on the right hand side it does not decrease.
93. [For the special case a = b. A = B see J. KiirscMk: Mat. phys.
lap. Vol. 23. p. 378 (1914).] In 92 define

a. = I(xl + 'JI X 2 : '2.) , b. = g (Xl + 'JI X 2 : Xl) , 'JI = 1.2, ... , n.

Then let n increase to infinity.


93.1. Define
Ea: = R.
Then

< (~~
a') S
=1.

yuana97@uw.edu
Pt. II. Solutions 93-94.1

Whereas several foregoing problems of this chapter were arranged


in pairs, each problem about sums followed by a companion problem
about integrals, the present problem has no such companion. (Why is
that so? Whereas in former cases, e.g. 81.3, multiplying ~ach 1: by-.!..
n
leaves the relation in question unchanged. this is not so in the present
case.)
94. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 28, p.174
(1920).] Assume that I(x) is not constant. We show that the quadratic
form
1
Q(x, y) = JI(t) [(2a + 1) f-a:/- + 2{a + b + 1) ts+bxy + (2b + 1) f-by2] dt
o
= Ax2 + 2Bxy + Cy2
is indefinite, i.e. AC - B2 < O. Integration by parts leads to

Jt" Jt"+1+ J t"+1+


1 1 1

I(t) dt =
(1+1)1
k + 1 I(t) 0 - k 1 dl(t) = k
/(1)
+ 1- k 1 dl(t), k>O,
o 0 0

provided that 1(1) = lim I(t) is finite. Thus


1-+1-0
. 1
Q(x, y) = 1(1) (x + y)2 - J (tax + fy)2 t dl(t).
o
We have Q(l, 1) > 0, Q(l, -1) < O. If 1(1) = 00 we can also establish
by careful manipulation [112] that Q(l, -1) < O.
94.1. Assume that the solid is a polyhedron and let P., q. and 1'. stand
for the areas of the orthogonal projections of one of its faces onto the
three coordinate planes, respectively. Extending the sums over all faces
we have
S = 1:(p; + q; + ~)i ,
S ~ 1:(P. + q. + 1'.) = 2(P + Q + R},
S > [(EP.)2 + (1:q.)2 + (1:I..)2]i = 2[P2 + Q2 + R2]l,
see 93.1 and 90. A cube attains the upper bound for S and a regular
octahedron the lower bound p!ovided that the edges of the former and
the diagonals of the latter are parallel to the coordinate axes; and these
bounds remain attained even if we apply to these solids arbitrary dilata-
tions parallel to the axes.
If the solid has a smooth (differentiable) surface, integrals should be
used instead of sums.

yuana97@uw.edu
256 Integration

94.2. By symmetry

f f ~2 dw = f f '1'/2 dw = f f C2 dw = ! f f (~ + '1'/2 + C2) dw = 4; •


(1) By 80
E = ff + e2a2'1'/2 + a2b2C2)i (~2 + '1'/2 + C2)1 dw
(b2e2~2

> f f (be~2 + ea'l'/2 + abC2) dw = 4;_ (be + ea + ab) .


(2) We may, and shall, assume that e is the shortest semiaxis.

E = ab ff [ 1- ~2 :;; cl ~2 - bl ;Z c2 '1'/2 dwT


< ab ff[ 1 - 1 el:;; l
c ~2 + bl ;z cl'l'/2)J dw

~ ~- (a 2 + b2 + ell) •

To obtain the last line work backwards and use that e < a, e < b.
(3) The more elementary 94.1 yields better estimates when e is near
O. For other approximations to the area of the ellipsoid see -G. P6lya:
Publicaciones del Instituto de Mat. Rosario Vol. 5, pp.51-61 (1943).
94.3. Lower bound in 94.2 and the theorem of the means.
95. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 26, p.65
(1917). Solved by G. Szego: Arch. Math. Phys. Ser. 3, Vol. 28, pp. 81- 82
(1920).J [For a generalization see G. P6lya. and M. Schiffer: Journal
d' Analyse math. Vol. 3, p. 323 (1953/54).J
95.1. Consider the polynomial in u

w = (a 2 + u) (b 2 + u) (e 2 + u) .
By the theorem of the means
1
w- a ~~(_1
- 3 al +
__ U
1_ _l_)_...!....dW
+ b + + c + u - 3w du
l U 2

and so

A comparison with 94.3 may suggest generalizations. Cf. G. P6lya


and G. Szego: Isoperimetric Inequalities in Mathematical Physics. Prin-
ceton: Princeton University Press 1951.

yuana97@uw.edu
Pt. II, Solutions 94.2-95.5 257
95.2. [E. Laguerre: Oeuvres, Vol. 1. Paris: Gauthier-Villars 1898,
p. 93.] Let the roots be x, Xl' X 2 ' ... , x.. _ 1 ' Then, by 80,
(a 1 - x? = (Xl + x2 + ... + X .. _ 1)2 ~ (n - 1) (xi + x~ + ... + X!_l)
= (n - 1) (ai - 2a2 - ~).
The bounds proposed are the two roots of the quadratic equa,tion that
we obtain hence by considering the case of equality.
95.3. [See G. P6lya: Numerische Math. Vol. 11, pp. 315-319
(1968).] By 93.1
J
s" = '(1:",-")" > (1:1'- ..
1 1 1

.. rv • -1) .. +1 = S..+1
.. +l .

By 80
s! = (1:y;("-1)1 2y;("+1)12 )2
< L'y.-"+11:y;"-l = S"_lS"+1'
Obviously
S"+l = ....~y.- "y.-1 < y -1 ....~y.- " = y -1 s".
The rest is more obvious.
95.4. From 95.3

and so on. Also directly from 81.1

Obviously
S2" = 1:y;"y.-" < y-"s".
The result is useful in certain applications of Graeffe's method, see I.e.
95.3.
95.5. [See Hardy, Littlewood and P6lya, I.c. 69, p. 163, theor. 218.]
2"
J y2 dqJ.

r
2A =
o

l'c)'' (")'" d~
By 81.2

2" ~ < (1',; (j",.<0>)'IS d'" (2A )"'.


The case of equality, and so the minimum of F, is attained when y is
constant.

yuana97@uw.edu
258 Integration

96. [Cf. I. Schur: Sber. Berlin. Math. Ges. Vol. 22, pp.16-17
(1923).] Suppose that xl' x2 ' ••• , x.. are positive. Since log x is concave
on any positive interval

log y,. 2: a,.l log Xl + a,.2 log x2 + ... + a,... log x.. , P = 1, 2, ... , n.

Add these inequalities.


97. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 20, p.272
(1913). Solved by G. Szego: Arch. Math. Phys. Ser. 3, Vol. 22, pp. 361-
362 (1914).]
98. [Cf. ~. Steinitz, Problem: Arch. Math. Phys. Ser. 3, Vol. 19,
p.361 (1912). Solved by G. P6lya: Arch. Math. Phys. Ser. 3, Vol. 21,
p. 290 (1913).] If x is an integer g(x) = 0; if x is not an integer g(x) = 1.
If x is rational G(x) = 0; if x is irrational G(x) = 1. Any lower sum of
G(x) is 0, any upper sum over the interval [a, b] is b - a. The function
G(x) is integrable over no interval.
99. If x is irrational/(x) = 0 and if k converges to 0, x + k is either
irrational and so I(x + k) = 0, or rational, x + k = J!...,
q
and so
I(x + k) = ~;
q q
~ converges to 0 as k -+ 0.- If x = l!...,
q
rational, I(x) = ~
q
,
and x + k irrational, I(x + k) - I(x) = - ~. Any lower sum is O. We
q
now divide the interval [0, 1] into k3 equal parts. Since there are at
most 1 + 2 + ... + (k - 1) = k(k ;- 1) positive proper fractions with

denommator < k, the upper sum

IS < k(k 2- 1) 1
k3 + k2 + II1 . 1.
3

100. If we call D. the oscillation of I(x) in the interval [x._ 1 , x.] and
if [tp(x) 1< M we get

The last sum converges to O.


101. Let n denote a positive integer, ~ < b -n a and D. the oscillation
of tp(x) on the interval [ a + (v - b-a a
1) -n-' + vb-aJ
- 1, 2, "', n.
n - ,v =
Suppose that tp(x) < M. Then
b _ a .. -1 a
J Itp(x +~) -
b b -

a
q;(x) 1 dx < -:E
n .=1
(D. + D.+ 1.) + 2 Mn- .

yuana97@uw.edu
Pt. II, Solutions 96-105 259

102. Construct the lower sum L and the upper sum U that belong to
the subdivision a = Xo < Xl < ... < X" = b (as described on p. 46) :

"
U = 2; M.(x. - x._ l ),
"
L = 2; m.(x. - x._ l ).
.=1 .=1

By a proper choice of the subdivision we can attain that U - L < e.


Now we define lJI(x) as follows: lJI(x) = M. on [X'_l' x.), 11 = I, 2, ... , n-1,
lJI(x) = M" on [X,,_l' x,,]. We define 1p(x) similarly using m•. Then
b b
J lJI(x) dx =
/I
U, J1p(x) dx =
/I
:L.

The only condition imposed on the subdividing points is that the maximal
length of the subintervals [x'_l' x.], 11 = 1, 2, ... , n, converges to 0
as n increases. Therefore these points can be chosen equidistant, forming
an arithmetic progression. The functions lJI(x) and 1p(x) constructed in
the described way are continuous on the right; they could be defined
continuous on the left instead.
103. Define lJI(x) and 1p(x) as in solution 102. Then the total variation
of lJI(x) is

and that of 1p(x) is

Both are not larger than. the total variation of I(x) because I(x) assumes
on [x._ l , x.] values which are arbitrarily close to M. and m,.
104. Let 11 be an integer, 11 = I, 2, ... , n; in the first half of the
interval [11 ~ 1, :J we have s(nx) = +1, in the second halfs(nx) =-1.
Thus
1

J J .~
1 2" "

I(x) s(nx) dx = {Ie ~ 1 + Y) _/(11 ~ 1 + Y+ 2~)}dY.


o 0

The absolute value of the expression between the curly brackets is


smaller than the oscillation of I(x) on [" ~ 1 , : ] .

105. [Riemann: Werke. Leipzig: B. G. Teubner 1876, p. 240;


E. W. Hobson: The Theory of Functions of a Real Variable & The Theory

yuana97@uw.edu
260 Integration

of Fourier's Series, 2nd Ed., Vol. II. New York: Dover Publication 1957,
pp. 514-515.] We may choose a = 0, b = 2n.
2,.
f I(x) sin nx dx
o

=! Sinny.~ {/((p -
"
1) 2: +y) _/((p _ 2: + y + :)}dY.
1)

The absolute value of the expression in the curly brackets is smaller than
the oscillation of I(x) on [(p - 1) 2: ,P 2:J .
108. [L. Fejer: J. reine angew. Math. Vol. 138, p.27 (1910).] We
may choose a = 0, b = 2n.
2"
p- 2"
,,-
2" .... ....
f I(x) Isin nx Idx = I f I(x) Isin nx Idx = I I... f Isin nx Idx,
o .=1 (._1)2"
.. .=1
.
(0_1)2"

where I." denotes a value between the least upper and the greatest
lower bound of I(x) on [(p - 1) ~ , P ~-;- ] •

107. If the points of discontinuity of the bounded function I(x) have


only finitely many accumulation points on [a, b] it is possible to find
finitely many intervals of arbitrary small length outside of which I(x)
is continuous. Such a function is therefore integrable. The function in
question has the points of discontinuity x = and ! ' ! ,"" ! ' ... ;
also x = 0 if (X =·0.
108. According to Riemann's criterion it is possible to find in an
arbitrary subinterval [ao, bo] of [a, b] a smaller subinterval [aI' bIt
bl - a1 < bo ~~, in which the oscillation of I(x) is smaller than !.
Iterating this procedure we obtain a sequence of intervals [a.. , b.. ],
n = 1, 2, ... , bPI - a" < bo - ao , on which the oscillation of I(x) is smaller
2"
than : . The point
2
(X = lim a" = lim b"
K~~ n~~
lies in [ao, bo] and I(x) is conti-
nuous at (x.
109. Assume that I(E) = 0 at each point E of continuity of I(x) ,
a < E< b.

f l(x)2 dx =
b "
lim 2 I(E.)2 (Xo - X,_I)'
a n400 1'=1

yuana97@uw.edu
Pt. II, Solutions 106-114 261
where x._ 1 < ~.< x. and the maximal length of the subintervals
[x._ 1 , x.] converges to 0 as n ~ 00. According to 108 ~. can be chosen
such that I(x) is continuous at x = ~., and so f(~.) = O. Let, on the other
hand, be I(~) =t= 0 at the point of continuity ~, a < ~ < b. Then we have
for 6, 6> 0 and sufficiently small, I(X)2 >/(;)2 whenever Ix - ~I < 6,
and therefore
b ;+a
f I(X)2 dx:;::: f I(X)2 dx :;::: 61(~)2 > O.
" ;-a
110. Assume that e, fJ are given, e, fJ > 0 and that 6 is such that
Itp(Yl) - tp(Ys) I < e whenever IYl - Y21 < 6. Since I(x) is integrable, a
subdivision of [a, b] can be found for which the total length of the sub-
intervals where the oscillation of I(x) is :;::: 6 is < fJ. On the other sub-
intervals the oscillation of tp[f(x)] is at most e.
111. [Cf. C. Caratheodory: Vorlesungen fiber reelle Funktionen. Leip-
zig and Berlin: B. G. Teubner 1918, pp. 379-380.] Let I(x) be defined
as in 99 and G(x) as in 98 and
1 for Y = 0
tp(y) = {
o for'y ~ O.
Then tp[f(x)] = G(x).
112. Assume that I(x) is non-increasing. We obtain f~r 0 < x < !
j "'/(C) dC > I(x) j C"dC = x"+1 I(x) 1 - a
(1
~1
r+ 1

" "
2" 2" ,,+1
f C"/(C) dC < I(x) f C" dC = x"+1 I(x) 2 a +~ ~ .
" "
In the case of a = -1 the last factors, both positive, must be replaced
by log 2.
113. Change the variable of integration in 112: substitute ~ for x.
x
Or prove the statement directly in a similar manner as 112.
114. The integral over [0, e] exists if and only if the integral of
X2)JJ
x" ( 1 - 2" or of x"e
-111+2
Z
. . .
converges, I.e. certamly for (:J < -2; for
> -2 if and only if £x> -1. The integral over [00,00), w > 0, is
(:J
convergent for (:J < 0 if and only if £x < -1. If (:J > 0 and n an integer

yuana97@uw.edu
262 Integration

the two following integrals can be compared as n -+ 00:


("+1).. ..
f X' Icos x IJ dx N (nn;)'" f Icos X1(%+ ....)1 dx.
.... 0

The second integral lies between


.. .
f 1cos x 1(·...)1 dx and f 1cos x 1[("+1)"1" dx,
o 0

"
which increase like n -"2 [202J. For the integral to converge we must have
IX - ~ < -1. Combining all these results we find that the integral in
question is convergent if and only if either IX < -1, {J <- 2 or if
-1 < IX < ~ - 1.
114.1. Let ~,as, as, ... be a sequence of positive numbers such that
~ ~ 1
~ a.. converges and ~ ~ diverges for {J < 1, moreover a.. < "2 .
.. =1 ft=1
We may choose e.g.
1
a.. =----,,--.,-:-
n (log n)1 , n= 3,4•...
Set for n = 1, 2, 3, ...

I(x) = j..!..
a.. for n - 1 < x < n - a!,
for n - a! ~ x < n.
II..

Observe that
. .
f I(x) dx < 2a.. , f
.. -1
U(x)]'" dx > i a: + a!-IX.
"-1

115. The limit function is properly integrable over every finite inter-
val [-w, w] and lim f'" I.. (x) dx = f'" I{x) dx.
11-+00 -m -tD
Since I/(x) 1 ~ F(x) the

f I(x) dx exists.
~

integral
We have

1_£ I(x) dx -_[I.. (x) dxl ~ 1_l"'/(X) dxl + Ijl(X)dXI


+_£"'F(X) dx + [ F(x) dx + I_II(X) dx -_£/. (x) dxl.

yuana97@uw.edu
Pt. II, Solutions 114.1-117 263
The first four terms on the right hand side are arbitrarily small for
sufficiently large wand the same is true for the last term if n is suffi-
ciently large while w is fixed.

116. Set '11= ; +A" V;;: A" -+ A. VI 31 implies


V- "Vn
; (:) = 2~ f_ (cos 2;; )"cosA"xdx.
-,,'In
In 115 we put:

V;_
I,,(x) = 1 2~o (cos 2 1~)" cos An X for Ix I < n
Vn
for Ix I > n Vn .
x'
1 --
We have lim I.. (x) = I(x) = -2 e 8 cos AX. To find a function F(x) as
n-+co n
described in 115 we proceed as follows: Since
log cos x
----

is continuous and negative on (0, ;),


lim log cos x = _~, lim log Xl
cos x = _ 00
%4-+0 Xl 2 '

there exists a (absolute) constant K, K> 0, such that


log cos x
xl'
<
-K cosx < e- Kx', 0 < <
=x=2'
n

-Kx'
1 --
Thus we can pick F(x) = 2n e 4 •

" a.e-· Y = P,,(y). Then


117. Put 2:
0=1

J P,,(y) yS-l dy = (a 1- S + a 2- s + ... + a"n- S) T(s) = D,,(s) T(s).


00

1 2
o
According to 115 it is sufficient to find F(x) such that IP,,(y) yS-ll < F(y),
J F(y) dy exists. Partial summation yields
co
whereby
o
,,-I
p .. (y) = 2: D.(a) [v"e-'Y - (v + 1)" e-('+1)Y] + n"D,,(a) e-"Y.
>=1

yuana97@uw.edu
264 Integration

The function x"e- xy has its maximum, (C1e- 1)"y-", at x =~; thus
y
IP.. (y) I < Ay-", IP.. (y)y'-ll<Ays-"-l, A independent of nand
y. Set F(y) = Ay'-"-l for 0 < y < 1; F(y) = Be- Y for y > 1, B > 0,
B independent of y.
118. We have with w > 0

\l'(X) sinnxdxl ~£i/(x) Idx + jl/(x) Idx + IJ:I(X) sinnxdxl·


The first two integrals on the right converge to 0 as w -+ 00, the third
integral converges to 0 as n -+ 00, w fixed [105]. The proof of the second
part of the problem runs similarly:

l!!(x) Isin nx Idx - n JJ(x) dx


co 2co \

~(1 + !) TI/(x)ldx+(1 + !)TI/(x)ldx


-co 0)

+ 1 l,1(x) Isin nxl dx -nl,1(x) dx


0) 20) \
[108].

118.1. [G. P6lya, Prohlem: Jber. deutsch. Math. Verein. Vol. 40,
2. Abt., p. 81 (1931). Solved by G. Szeg6: Jber. deutsch. Math. Verein.
Vol. 43, 2. Abt., pp. 17-20 (1934).]
119. (1) Reduction to two functions of two variables possible: Let
tp(x, y) make different values u = tp(x, y) correspond to different pairs
of numbers x, y (e.g. a one to one correspondence between the xy-plane
and the number line u). For a given function I(x, y, z), "P(u, z) is then
found in the following way:
If u* does not belong to the range of tp(x, y), "p(u*, z) is chosen arbi-
trarily, e.g. "p(u*, z) = 1.
If u* does belong to the range of tp(x, y), it corresponds to a unique
pair x*,y*, thus u* =tp(x*,y*) ;in this case we choose "p(u*, z) = I(x*, y*, z).
Then it is true for all x, y, z: "P(tp(x, y), z) = I(x, y, z).
(2) Representation impossible, e.g. for the function
I(x, y, z) = yz + zx + xy.
Arbitrarily many pairs Xl' Yl; x2 ' Y2; ..• ; X.. ' y.. exist so that a
given continuous function tp(x, y) assumes the same value for all of
them. (They are points of the same "level line".) If tp(x, y) = const. the
statement is evident. If tp(x, y) is not a constant and if e.g. tp(x', y') = 1,
tp(x"', y"') = 3 there exists on each circular arc connecting x', y' with
x"', y'" an intermediate point x", y" for which tp(x", y") = 2.

yuana97@uw.edu
Pt. II, Solutions 118-119a 266

If t(x, Y, z) = 'P(f/I(x, y), z) as proposed, f/I(x, y) continuous, there are


arbitrarily many pairs xl> Yl; x2 ' Y2; ... ; x"' y" for which

t(Xv Yv z) = !(x2, Y2' z) = .,. = !(x", Y.. , z)


identically in z. If

(Xl + YI) z + XIYI = (X2 + Y2) Z + X2Y2


is to hold identically in z the two equations

Xl + YI = x2 + Y2' XIYI = X2Y2


must be satisfied. Eliminating YI we find

xi - (X2 + Y2) Xl + X2Y2 = O.


If Xl =1= x2, then Xl = Y2' YI = x2· This means that to a given pair Xl' YI
there exists exactly one different pair on the "level line" and not arbi-
trarily many as the proposed form asks for.
yz + zx + xY, being a symmetric function, cannot be represented
in the form 'P(f/I(Y, z), x) or 'P(f/I(z, x), Y) either.
In the same manner one can show that the continuous function
xy + yz + z cannot be written with the help of two continuous functions
of two variables boxed in each other, however it can be done with
three such functions:

XY + yz + Z = (x + z) Y + Z = S{P[S(x, z), Y], z}.


(Notation 119a.)
It is easier to discuss similar questions for narrower classes of func-
tions [119a]. We cannot represent an analytic function of three variables
that does not satisfy some algebraic partial differential equation by
boxing into each other a finite number of analytic functions of two
variables. Cf. the 13th of the "Mathematical Problems" by D. Hilbert:
Nachr. Akad. Wiss. G6ttingen 1900, p. 280.
119 a. We write t(x, Y, z) = Y~ zx +- +
xy and denote the partial
derivatives by sUbscripts.
(1) The relation t(x, Y, z) = f/I{'P[X(x, Y), z], z} for all values X; Y, z
for which y = z + x =1= 0 implies
t
:z (~) ! (::::~:) = 0,
=

thus txzty - txfyz = 0; yz + zx + xy does not satisfy this equation.

yuana97@uw.edu
266 Integration

(2) First proof: Taking the derivative with respect to x, y, z and


eliminating f/J., and f/Jx we deduce from the relation
I(x, y, z) = f/J['I'(x, z), X(y, z)) that

I" '1'" 0 I
Iy 0 Xy 1= O.
I~ '1'. X,
We can assume that '1'" =1= 0, Xy =1= O. If '1'" =f= 0 and Xy =f= 0 we may write
'P.
-=-V,
x.
-=-U,
'P" Xy
v = v(x, z), U = u(y, z), thus vy'= Ute = O. Moreover

+ I"v + I. = O.
Iyu
This leads to a contradiction: Put F = Iyu + I"v + I., I = yz + zx + xy.
Then
8F 8F ()IF
F - 8x Iy - ay I" + 8x ayl"/y = -2z.
Second proof: Take the derivatives with respect to x, y, z and set
z = -y in the three corresponding equations. The first equation implies
then 0 = f/J.,'I'". We can assume '1'" =1= 0 so that for the special values
mentioned we have f/J., = 0, as long as '1'" =f= O. I.e.

Iy = x - y = f/JxXy' I. = x + y = f/JxX••
If x =f= y, then f/Jx =f= 0, Xy =f= 0 and

~_±_'-:' x.
x -y Xy

The right hand side depends on y and z = - y only: contradiction.


(3) Differentiation of I(x, y, z) = tp{'I'[X(x, y), z], x} implies

I" = f/J.,'I'xX" + f/J",


Iy = f/J.,'I'xXy'
I. = f/J.,'I'.,
I"Y = ... + f/J.,X"Xy'l'xx'
Iyy = ... + f/J.,X~'I'xx·
In the last two equations the terms containing '1' 1. are not written down.

yuana97@uw.edu
Pt. II, Solution5120-122.2 267

We set z = -x and obtain CJl.,1fJxX,. = O. Since we may assume X,. =1= 0


and because of the third equation we conclude for x + y =1= 0 and X,. =1= 0
that 1fJx = O. Thus
1 =CJl.,XxX,.1fJxx ' 0 =CJl.,X;1fJxx '
for the special values mentioned above. These two equations contain a
contradiction because, according to the second equation, at least one of
the three functions involved has to vanish.
120. No. Example: I(x) = xlI; ~ = 0 is a point of inflection.
121. [Cf. G. P6lya: Tohoku Mat. J. Vol. 19, p. 3 (1921).] Denote the
least upper bound of II'(x) Ion [a, b) by M. Then M > 0 and
a+b
I(x) = I'(~) (x - a) < M(x - a) for a ~ x ~ -.2-'

I(x) = 1'(",) (x - b) ~ M(b - x) for a; b ~ x < b,


a < ~ < x, x < '" < b. It is not possible that both inequalities hold
identically because such a function would cease to be differentiable at
a+b
x=-2-' Thus
a+b
b
J I(x) dx < M J (x -
a a
2
a) dx
a+b
b
+ M J (b - x) dx = M -T .
(b )2

122. [W. Blaschke; Problem: Arch. Math. Phys. Ser. 3, Vol. 25,
p. 273 (1917).] According to Taylor's theorem we can write for x ~ Xo
I(x) -/(xo) = (x - x o) I'(xo) + (x - X~2 /,,{x) , Xo - r < x< Xo + 1'.
Integration and the first law of the mean for integrals yield

J [I(x) - J J~~
%0+' x.+, %.+'
I(xo)] dx = (x _/0)2 I" (x) dx = I" (~) x o)2 dx,
%0-' x,-, x.-,
because the function I"(x), being a derivative, assumes all the values
between its greatest lower and its least upper bound.
122.1. The function I(x) is linear; method of 122.2.
122.2. Introduce the abbreviations
I(a) = A, I(x) = X, I(b) = B,
and take for (u, v) first (a, x), then (x, b), then (a, b):
(x - a) VAX + (b - x) VXB = (b - a) VAB,

yuana97@uw.edu
268 Integration

and hence
1
I(x) = X = (ex + d)1 '
where c and d are appropriate constants. Convince yourself by performing
an elementary integration that a function of this form satisfies the
imposed condition.
122.3. By the method of 122.2
1
I(x) = Vlex + dl .

123. First proof: Suppose that the series}; p,.e'''' = I(x) is con-
,.=0
vergent on the interval (a, b); then it can be differentiated arbitrarily
often in that interval:
00

f'(x) = }; np,.e"'", f"(x) = }; n 2p,.e'''',


,.=0 ,.=0
00 00

I(x) f"(x) - [j'(X)]2 =}; }; Hm - n)2Pmp,.e(m+,.)x > 0 [72].


m=O ,.=0
u\
Second proof. Proposition 80 implies for Xl < x2 , a. = Vp:'e 2 '

[I e1 XI)r~ < I(x l ) l(x2).

124. [Cf. J. L. W. V. Jensen, I.c. 70, pp.187-190.] Suppose that


the function qJ(x) is convex on the interval [a, b] and that qJ(x) < G.

(
Xl + x2 + ... + X,.) < !p(xI ) + !p(x2) + ... + !p(X,.)
qJ n - n
implies that for Xl = x 2 = ... = xm = X + n {), xm+l = ... = X" = X,
X an arbitrary point of (a, b), I{) I sufficiently small, m < n,

!p(x + n6) - !p(x) > !p(x + m«5) - !p(x)


n = m .

Substituting -{) for {) and remembering that


qJ(x + m {)) - qJ(x) ~ qJ(x) - qJ(x - m b) we obtain

(*) jrp(X + n~ - rp(x)

>
> rp(x + m~) - rp(x)

rp(x) - rp(x - !,,«5) > rp(x) - rp(x - n«5)


= m = n .

yuana97@uw.edu
Pt. II, Solutions 122.3-126 269

Hence for m = 1
G - q>(x)
--n- > €p(x + ~) - €p(x) > €p(x) - €p(x - ~) > --n-'
q>(x) - G

Let ~ converge to 0 and n increase to infinity in such a way that x ± n ~


remains in (a, b). Then the continuity of €p(x) is established. Assume
~ > 0 and replace ~ by .!
1J
in (*) :

q>(x) - q> (x - : 6) q>(x) _ q>(x _ 6)


~
- m >
= ---
6 --.

I
-6
1J

Since €p(x) is continuous:


q>(x-t 6) - q>(x) ~ q>(x + 6') - q>(x)
6 - 6' 0
> ,
q>(x) - q>(x - 6) ~ q>(x) - q>(x - 6)
<u.1\' <u.
.1\

= 6' - 6
As ~ -+ 0 the first term converges to a limit l+ and the last to a limit L,
l+ > L
125. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p. 283
(1916).] The values which y = I(x) assumes on an interval of length 1
fill a closed interval of length L:

L = max I/(xl ) -III'(X) dxl ::;: 1max II' (x) I·


l(x2) I = max

Let e > O. About each point x where I' (x) 0 construct the largest =
interval for which II'(x) I < e. We denote the length of the interval by
l". The values I(x) is assuming on such an interval cover at most an in-
terval of length lxe. The points of M (y = I(x) with I'(x) = 0, a < x <b)
are enclosed in countably many intervals of total length ::;: e(b - a). This
proof shows in addition that the set M has measure O.
126. [U. Dini; d. C. Caratheodory, l.c. 111, pp.176-177; Hille,
Vol. II, p. 78.] It is sufficient to consider the following case:
Il(X) ? 12 (x) ~ ... > In (x) ~ ... , In(x) continuous, lim In (x) = 0,
n~=

o ::;: x <
1. If the convergence were not uniform, infinitely many points
Xn would exist for which In (x n) > a > 0, 0 ~ xn < 1, a independent of n.
Let ~ denote an accumulation point of the xn's and m be such that

yuana97@uw.edu
270 Integration

Im(~) < a. Determine a neighbourhood of ~ in which Im(x) < a, thus


I,,(x) < a for n ~ m. There are infinitely many x,,'sin this neighbourhood:
contradiction.
127. [Cf. G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 28,
p.J 74 (1920).] The limit function is monotone too, say monotone in-
creasing. Subdivide the interval of convergence of the sequence I,,(x),
n = 1, 2, 3, ... , into subintervals [x._ I , x.], '11= 1, 2, ... , N, so small
that I(x.) -/(x._ I ) < e, moreover choose n so large that 1/.. (x,) -/(x.) I < e
for every v. Then we have for X'_I < x < x.

l(x._ I ) - e < 1.. (x._ 1) < I,,(x) < I.. (x.) < I(x,) + e
thus 1/.. (x) - I(x) 1< 2e; we have used the hypothesis that I.. (x) is in-
creasing.
128. Obvious.
129. Assume a < x < b. (It is obvious what has to be changed in
the proof below to accommodate the cases x = a and x = b.)
b x+. x-. b
JP,,(t) I(t) dt -/(x) = J P.. (t)
a %-8
[f(t) -/(x)] dt + J + J.
a x+s

x+.
The absolute value of the first term on the right is <~ J
P.. (t) dt < ~
x-"
whenever I/(t) - I(x) I < ~ for x - e < t < x + e. The absolute values
of the other two terms are smaller than

2 max I/(t)
a:;i;I:;ob
I (7"p. (t) dt + j P,,(t) dt).
a x+"

Thus the condition is sufficient. - We define

r0 + rt :::;:; t ~ x + e - 1J,
for x - e
I(t) = t 1 for a < t < x -e and x + e < t :::;:; b,
linear for x - e :::;:; t < x -'- e + 1J and x + e - 1J :::;:; t < x + e,
0< 1J < e; I(t) is continuous. From
b x-.+'1 x+. x-a

a
J P..(t) I(t) dt -/(x) = x-aJ P.. (t) I(t) dt +x+a-'1
J P.. (t) I(t) dt + aJ P.. (t) dt
b
+ J P.. (t) dt~ 0
x+.
follows that the condition is necessary because all the terms are positive.

yuana97@uw.edu
Pt. II, Solutions 127-133 271
130. The statement of 129 can be extended to the case b = 00,

x = 00. In 129 put n = ~ , P,,(t) = Ee- II ; then


e
...
f e-·' dt = f e- al dt = 1,
00

limE 0, E
.-+0 0 o
wher.e w is positive and independent of E.
131. The substitution of e' for t changes the interval of integration
[0,00) into (-00,00). The new integral can be split into two parts,
over (- 00, OJ and [0,00) resp. We examine the two parts separately.
If the integral

f Je"'cp(t) dt
00 00

e"'/(J) Jdt =
o 0

converges for A = {J it converges for any smaller A, A = (J - E, E > 0:


writing tI'cp(t) = 'P(t) we obtain

o
j e-II'P(t) dt = e-·'" j 'P(t) dt + j e-
0
E
0
d (1 'P(T) dT) dt,
0

hence for w -+ 00

j
e-al'P(t) dt = E je- al
0 0 0
(1 'P(T) dT) dt.
This integral is, as a function of E, continuous on the right [130]:
132. Cf. 128 and the first part of the proof of 129. The number ()
defined there can be made arbitrarily small with E, independently of x
(uniform continuity).
133. [E. Landau: Rend. Circ. Mat. Palermo Vol. 25, pp.337-345
(1908).] Apply 132: a = 0, b = 1,

P,,(x, t) = ~=---ix - t)2]" _ .


J [1 - (x - t)2]" dt
o
For 0 < E < X < 1- E

x-s 1
f P,,(x, t) dt +x+sJ P,,(x, t) dt <
-.
x+.
o J [1 - (x - 1)2]" dt

According to 201, 202 this expression converges to 0 because

_a
j(l - t 2 )" dt '" 1{;;",~.~.i. ... ~.
rn 1 3 5 2n +1

yuana97@uw.edu
272 Integration

134. [L. Fejer: Math. Ann. Vol. 58, pp. 51-69 (1904).] Apply 132:
sinn x - ~)2
a = 0, b = 2n-, P.. (x, t) =2~
nn
(
. x-t
2. According to VI 18 the
sm- 2-
P.. 's are unormed", i.e.
2"
J P..(x, t) dt =
o
1;

also for 0 < e ~ x < 2n- - e


x-a 2" 1
JP..(x, t) dt + J P..(x, t) dt < ---e-.
o x+_ nsinl -
2

Since the integrand is periodic, any interval of length < 2n- can be
considered as an interior subinterval.
135. Cf. 133.
136. Cf. 134.
137. We may assume that I(x) is non-negative and [102] that it is
piecewise constant. Such a function can be obtained by 'addition and
multiplication by positive constants of functions with the following
properties
I(x) = {Ion a subinterval [tX,P] of [a, b], a ~ tX < p< b
o outside of [tX, P] .
For simplicity's sake let a < tX, p < b.
We define for sufficiently small 'YJ, 'YJ > 0,

1 + 'YJ, if tX < x <p;


1'1(x) = 1 'YJ, if a < x < tX - 'YJ or p + 'YJ < x ~ b;
linear, if tX - 'YJ < x < tX or p < x < p + 'YJ.
Then 1'1(x) - I(x) > 'YJ, furthermore
b b

J
0< 1'1 (x) dx -
a a
JI(x) dx = 'YJ(tX - 'YJ - a)

i.e. arbitrarily small with 'YJ. The function 1'1(x) is everywhere continuous.
If the polynomial P(x) is such that

/1'1(x) - P(x) / < 'YJ, then I(x) ::;: 1'1(x) - 'YJ < P(x)

yuana97@uw.edu
Pt. II. Solutions 134-140 273
and
b b b b
f
a
P(x) dx - f I(x) dx < rJ(b -
a
a) +f
a
IT/(x) dx - f I(x) dx.
a

For IX = a or (3 = b some slight changes have to be made. A similar


proposition holds for trigonometric polynomials if a = 0, b = 2;77;.
138. [M. Lerch: Acta Math. Vol. 27, pp. 345-347 (1903); E. Phrag-
men: Acta Math. Vol. 28, pp.360-364 (1904).] Let P(x) denote an
arbitrary polynomial:
b b b
f U(X)]2 dx = f I(x) U(x) - P(x)J dx + f I(x) P(x) dx
a a a
b
= f I(x) U(x) - P(x)] dx,
a

thus
b b
fa U(X)]2 dx < max
a~x~b
I/(x) - P(x) I'
4
f I/(x) Idx. [135,109.]

A similar statement holds for trigonometric moments (Fourier constants)


f I(x) cos
2"
. nx dx, n = 0, 1, 2, ...
o sm
139. Determine P(x) as in 137 and assume I/(x} I sM. Then [d.
solution 138]

< Me.
b b
J U(X)]2 dx S M f U(x) -
a a
P(x)] dx

140. Let I(x) not vanish identically and not change sign more than
b
n - 1 times. According to the first condition, f I(x) dx = 0, there exist
a
numbers Xl' x 2 ' ••• , X k ' a < Xl < X 2 < ... < x k < b, with the following
property: I(x) does not vanish identically in any of the subintervals
(a, Xl)' (Xl' x2 ), ••• , (Xk _ l , X k ), (Xk' b); I(x) does not change sign in any of
these subintervals (V, Chap. 1, § 2) but the signs alternate for consecutive
subintervals. Therefore I(x) (x - Xl) (X - x2) ••• (X - xk) has the same
sign for the entire interval (a, b) and does not vanish identically. Accord-
ing to the hypothesis we would have in the case k < n - 1
b
f I(x) (x - Xl) (x - x2) ••• (x - xk ) dx =0
a

i.e. [109] I(x) (x - Xl) (X - x2) ••• (X - xk) = 0, I(x) = 0: contradiction.

yuana97@uw.edu
274 Integration

141. [Cf. A. Hurwitz: Math. Ann. Vol. 57, pp. 425-446 (1903).]
Suppose that 1(0) > 0 and that I(x) changes sign 2k < 2n + 2 times in
the interval (0, 2n). Let xl' x~, x 2 ' x~, ... , xk, x~,
o< Xl < Xl, < x2 < x~ < ... < X" < X", < 2n,
be the points where the changes of sign occur. In analogy to solution
140 we form
. x - Xl . X - Xl . X - X2 • X- x 2 . x - x" . x - x k
I(x) SIll - - S I l l - - SIll - - SIll - - ... SIll - - - S I l l - -
2 2 2 2 2 2·
Note that
• X-IX. x-fJ 1 IX-fj
S I l l - - S I l l - - = - C O S - - --COS X - - -
1 ( lX+fJ)
2 2 2 2 2 2
(lX,{J constant, 0 < lX < 2n, 0 < (J < 2n) changes sign in the interval
(0, 2n) only at X = lX and X = (J. Use VI 10. If 1(0) = 0 consider I(x + a)
with I(a) =l= o.
142. [L.c. 138.] Writing f" e-k,l rp(t) dt = «P(x) we obtain for
k > ko 0
co

J(k) = [«P(x) e-("-ko)"]; + (k - ko) f «P(x) e-(k-ko)" dx.


o
The relations

e-"X = y, «pC log ~) = V'(Y), '1'(0) = J(ko) = 0,

define V'(y) as a continuous function on the interval [0, 1], furthermore


I
fV'(y) y.. -l dy = 0, n= 1, 2, ...
o
Hence [138] V'(Y) = 0, «P(x) = 0, rp(x) = O.
143. [M. Lerch, communicated by M. Plancherel.] If So were a zero
of the F-function So + m, m = 1, 2, ...• would be zeros too [functional
equation]. Let m be so large that + m) is positive, put meso
s = So + m + 1 = (1 + it, (1 > 1. The equation

j
o
e-""xu - 1 cos (t log x) dx = m1e-""x'-l dx
0
= mr~) =
n
0, n = 1,2,3, ...

would imply [142] that xu - 1 cos (t log x) = 0: contradiction.


144. For I(x) = 1, x, x 2 d. 140. For I(x) = e":

K,,(x) =
,=0
i e~ (:Z) x'(1 - x)"-' = (e~ X + 1- X)" = [1 + (e~ - I)X r.
yuana97@uw.edu
Pt. II, Solutions 141-148 275
145. We have [140]

1; (v -
,=0
nx)2 (:) x'(l - x)"-' = nx(l - x),
thus
~ n
n2 L;1 1 < nx(l - x) ::::;: 4" .
146. [Cf. S. Bernstein: Communic. Soc. Math. Charkow Ser. 2,
Vol. 13, pp.1-2 (1912).] We define e,,(x) = max I/(x) -/(:)1 for
1
all v for which I~n - xl < n - '. lim e,,(x) = 0
11-+00
uniformly in x, i.e.
en(x) < en' lim e" = O. Moreover
"-+00

I(x) - K,,(x) = ,~[/(X) - I( :)] (:) x'(1 - x)"-'.


According to 145 we have the inequality

I/(x) - K,,(x) 1< e" L;1 + 2M L;II < en + ~ n- 1 ,


where I/(x) 1 < M on [0, 1].
147. [Cf. J. Franel: Math. Ann. Vol. 52, pp.529-531 (1899).]
Obvious for 0 < r < r l . For rm < r < rm+l the right hand side is equal
to
ml(r) - 1[f(r2 ) -f(rl )] - 2[f(ra) - l(r2)] _ ... - (m -1) [f(rm) - ICrm- l )]
- m[f(r) - I(rm)].
But this exactly equals the expression on the left hand side. Indeed,
the formula we have proved is the formula for "partial integration" :
, ,
f ICt) dN(t) = N(r) ICr) - f N(t) rCt) dt.
o 0

148. If r"-k-I < rn _ k = ... = r" = r"+1 = ... = r"+l < r,,+l+l
(possibly k = 0 or l = 0; ro = 0) then

lim ~= O.
n-+oo rn

If rm < r < rm+l we find N(r) = m and


m+~ _ _ 1_= ~< N(r) < _~.
rm+1 rm+1 rm+1 r - rm
Analogously in the second case.

yuana97@uw.edu
276 Integration

149. Cf. 148 and I 113.


150. [E. Landau: Bull. Acad. Belgique 1911, pp.443-472. Cf.
G. P6lya: Nachr. Akad. Wiss. Gottingen 1917, pp. 149-159.] We
assume c > 1 and choose m so large that 1 < c < 2m • Then
1< L(C1") < L( 2m1' l _ L(21') £(~~~) ... L(2m1') -+ 1
L(1') L(1') - L(1') L(21') L(2 m- 11') .

If c < 1 we replace c by ...!.. and use the slowly increasing function L(cr}.
c
151. Mathematical induction shows that for positive integral k

For k = 1 the value of the limit is obvious; if k > 1 we have for suffi-
ciently large r
log,. (21') log" 1'2 logk_l (2 log 1')
1<---<--
log" l' log" l'
=------.
log"_l (log 1')

152. It is sufficient to prove


lim log L(2m) = O.
m-+co m

This relation is a consequence of the inequality L(2 )1_


L(2m- )
m
< 1 + 15, where
15 >
0 is arbitrary and m sufficiently large, m > M(I5).
153. [149, 152.]
154 N(c1') ~ A L(c1')
• N(1') c L(r) •
155. Any function that is continuous on the left and piecewise
constant can be obtained as linear combination of the functions

I(x) = { 1 for 0 < x < '1, '1 > 0


ofor all the other values of x.
For such functions the proposition reads

[154].

Cf. 156 for other functions.


156. We bound I(x} by two functions that are piecewise constant and
continuous on the left
'P(x) < I(x) < 'P(x)

yuana97@uw.edu
Pt. II. Solutions 149-158 277

so that

Jo P (1)
eA
X
c" (1)
T dx - J "p T dx < e
0
X

e > 0, e arbitrary [102]. Then

_1 ~ "p(~~)< _1 Y'
N(r) , S;;e, 1', - N(r) , ";ie,
I(rn)I' <- ~_ ~
N(r), S;;e,
p(rnI' \./
n- n- n-

lim inf and lim sup of the middle term lie therefore [155] between

]\iJ)
o
dx and
0
f p(J)
dx, which differ from f 1(3-)
0
dx by less
than e.
157. According to 147 we have

_1_
N(r) 'n:;>'
~ (rn
I'
)"'-A = rA-'"
N(r)
(N(r)
i-'"
+ (A _ IX) j N(t) t-H"'-1dt)
0

cx> 1 + ~J. -L(r)(Xl"'::-~ 0j L(t) {,,-1 dt.


For 0 < c < 1

r-"'L(cr)
L(r) e,
j t",-1 dt < r-'"
L(I')
j L(t) t"-1 dt < r-" j t"'-1 dt =
0 0
~
(X
.
,

lim inf and lim sup of the middle term lie therefore between ~-=~ and ~
where c is arbitrarily small. (X (X

158. Proposition 147 implies that


R
~ r;;"-A = N(R) R-",-i. - N(r) r-"'-A + (IX + A) JN(t) t- ex - A- 1 dt,
r<rn:;aR r

hence [152, 153]


1
~ '"
N(r)":::"
(I'~ )-x-;' =
r
-1 + ( +N(r)J.) re<+-' J~ N(t) t-:X-A-1 dt
(X

'n>r r

cx> -1 + ((X l(I'~)r'" j L(t) t-",-1 dt.


r

Let 0 < e < 2'" - 1; for sufficiently large r the inequality


L(2r) < L(r) (1 + e) holds, hence L(2Vr) < L(r) (1 + e)' for v = 1, 2, 3, ... ;
consequently
2",
_I'_J J
ex 00 IX 00

~ < L(t) t-",-1 dt < -1'- ~ L(2Vr) t- ex - 1 dt <


(X L(r) , L(I') v=1 2 v - 1,

~_
(X
i
v=1
(1 + ej" (2-"(V-1) _ 2- exV ) = 1+e
(X
2" - 1
2'" - 1 - e

yuana97@uw.edu
278 Integration

159. [As to 159-161 d. G. P6lya: Math. Ann. Vol. 88, pp. 173-177
(1923).] Generalization of 155-158. As in 156 bound I(x) by two func-
tions 11'(%) and qF(x) that coincide with _x"-A and x"-A respectively on
[0, ()), with I(x) on [15, w) and with _X-",-A and X-",-A respectively on
[w, 00). We have [157]

and [158]

Let 15 converge to 0 and w increase to infinity.


160. Let I(x) be decreasing, f1 > A.. There exist [1115] infinitely many
values of n such that
1
r,.. < (.!!...)/i '" = I, 2, ... , n - 1.
r" n '
Choose a number r, r"_1 < r< r" such that the inequalities
1
r,.. < (.!!.-)/i '" = 1, 2, ... , n - 1,
r n'

are satisfied too. Thus, since I(x) is decreasing,

'" = 1, 2, ... , n - 1.

We have N(r) =n- 1, consequently


,..+1

N~r)/(:)~n~l/k=i]>n~l! I(x~)dx,
"
(r,..) (r,..) n
-N(r) ~ 1 r - N(r)
1 ,,-1

,..=1
- ----1
~
',..~'
1r - n
- ~--
-1 J1xfJ
.!_
1 (~)
dx .
II

yuana97@uw.edu
Pt. II, Solutions 159-163 279

The integral!
o
l3-)
dx is a continuous function of fJ [131]. Similar
arguments apply if f(x) is increasing. Replace f(x) by -f(x).
161. Let 0 < IX < A < fJ. By making use of 1116 we establish simi-
larly as in 160 the existence of arbitrarily large n for which

N(-)
1
1: f (r)
00
~
1 ( 1)
-:::;,ff x--; dx+ f f xP dx.
00 ( 1)
Yn k=l Yn 0 }

Choose IX and fJ sufficiently close to A [131].


162. [For 162-166 d. H. Weyl: Nachr. Akad. Wiss. G6ttingen
1914, pp. 235-236; Math. Ann. Vol. 77, pp. 313-315 (1916).] If f(x) = 1
on the subinterval [IX, fJ] of [0, 1] and f(x) = 0 otherwise the equation
(*) on p. 88 leads to the condition lim ~.. ~p~ = fJ - IX. Now suppose
n.-+CCI n
that the condition is satisfied. To begin we note that it does not matter
whether the subinterval IX, fJ is open, halfopen or closed. The relation (*)
holds for any function that is constant (=1= 0) on a subinterval and vanishes
outside this subinterval, therefore (*) holds also for any linear combination
c1f} (x) + c2f2(x) + ... + cd/(x) of such functions f.(x) , c. constant, i.e.
for any piecewise constant function. If f(x) is properly integrable there
exist piecewise constant functions [102, with a = 0, b = 1], tp(x) and
'l'(x) such that
+ tp(x2 ) + ... + tp(xn) /(x}) + /(x2 ) + ... + /(xn )
tp(x 1 )
-=----=------
n
<
-
---'=--~---
n

1 1
The first and the last expression converge to f tp(x) dx and f 'l'(x) dx
o 0
1
resp. and both are arbitrarily close to f f(x) d~. The weaker conditions
o
• vn(O, fl)
11m ------- =fJ, 0< fJ < 1
n-+oo n
or
• vn(fl, 1)
hm -------- = 1- fJ, O<fJ<l
n-+oo n

are also sufficient. It will even do if one of these conditions is satisfied


for a set of fJ-values everywhere dense on (0, 1).
163. The condition is obviously necessary. As in 162 the interval
may be open, half-open or closed. We replace the functions lj!(x) and

yuana97@uw.edu
280 Integration

lJI(x) in 102, if a > 0 by piecewise linear functions for which the extensions
of the single line segments pass through the origin (y = kx). As in 162
we now can establish (*) for a properly integrable function on [0, 1]
that vanishes in an interval containing the origin, e.g.
. v,.({J.l)
hm - - - = 1 - {J, 0 < (J < 1 [solution 162].
n--?cx> n

164. Cf. 162. Instead of 102 use 137.


165. Cf. 162. Instead of 102 use 137.
166. The condition of 165 is satisfied because

167. In the notation of 166 we have to examine


. f(x 1 ) + f(x + ... + f(x,.)
9)
lim - , f(x) = f(J(afJ - Od +xd) .
fI~'" n
where f(J(Y) = 1 or 0 depending on whether the integer next to y is on
the right or on the left of y. (If y = n, n + t, n integer, we choose, e.g.
q;(y) = t.) 166 implies that the limit is
1 1 1
= Jf(x) dx = 0Jf(J(y)dy= 2'
o
The proposition is also valid for arithmetic progressions of higher order
[1128] as can be proved by more elaborate methods [H. Weyl, l.c. 162,
p. 326]. The result might be interpreted as an expression of a certain
degree of "irregularity" of the sequence E 1 , E2 , E3 , ••• , En' ••• [d. R. v. Mi-
ses: Math. Z. Vol. 5, p. 57 (1919).]
168. [E. Hecke: Abh. math. Sem. Hamburg Vol. 1, pp. 57 - 58
(1922).] According to 188 we have
a1 + a2 + ... + a
1: anyn
00

lim (1 - r)' = lim --,. ,


,=1-0 ,.=] 10-+00 n
provided that the limit on the right hand side exists. For
an = (nO - [nO]) e21finlX

this limit becomes, according to 166,

= J
o
1
xe2niqx dx = -. .
2mq
1

169. [E. Steinitz, Problem: Arch. Math. Phys. Ser. 3, Vol. 19, p. 361
(1912). Solved by G. P61ya: Arch. Math. Phys. Ser. 3, Vol. 21, p. 290

yuana97@uw.edu
Pt. II, Solutions 164-174 281
(1913). J The limit is the function j(x) defined in 99. For irrational x d.
166; easier in the case of rational x.
170. We obtain 10"0 - [10"OJ by multiplying the decimal fraction
by 10" and omitting all the digits to the left of the decimal point. Let
(X = (Xl(X2 ••• (Xk represent a finite decimal fraction. Choose n so that

10"0 - [10"OJ starts with the digits (Xl' (X2' ••• , (Xk and that r zeros follow.
Then
1 10"0 - [10"OJ - (X I < 10!+' .

171. The Taylor series of e is


1 1 1;"
e = 1 + T! + 2i + .,' + n! + (n + if!' 0 < 0" < 1,
hence n! e = n! + iTn' + n'i + ... + +
2 1
eO ..
n +l ' For n > 2 we have
A e ~ 3
n + 1 < n + 1 < 1, thus n! e - [n! e] = n + 1 < ~+1'
172. [Communicated by H. Prlifer; H, Weyl proved, l.c. 162, that
the set in question is everywhere dense, even equidistributed on the inter-
val [0, 1J.J For r = 1 the statement follows from 166. Assume r > 1 and
that a, is irrational [otherwise we omit the highest rational terms which
are periodic mod 1]. If the set had only a finite number of limit points
this would be true also of the remainder mod 1 of the numbers

P(n + r -1) - (" ~ 1) P(n + r _ 2) + (' -; 1) P(n + r - 3) -'"

+ (_1),-1 P(n) = a,r! (n +1' -; 1) + a,_l(r - I)!,


This, however, contradicts 166.
173. Let k be a positive integer. Then

is bounded [166J, Partial summation shows that

is bounded too [165].


174. [L. Fejer.J Let N(x) denote the counting function of the se-
quence g(l), g(2), ... , g(n), ... and let t = y(x) be the inverse function
of x = g(t). Then N(x) = [y(x)J. The conditions (1)-(4) imposed on
g(t) imply the following properties of y(x) for x ~ g(l): y(x) is continuously
differentiable, y(x) is monotone increasing to infinity as x -+ 00, y'(x)
is monotone increasing to infinity as x -+ 00, while ~i;i -+ O.

yuana97@uw.edu
282 Integration

x
x- 2 2
Since _(X) < 2 ) = -.-(-) X
2 < Xl < X, we conclude from the
yx () -I' (X
yx 2 YX1 '

above that I' (x)


x
-+ 0 as X -+ 00; furthermore
y(x e) +
y(x)- -+ 1, e fixed or
bounded, as X increases.
Let 0 < IX < 1. It is sufficient to prove [162] that
m-l
~ (N(k + /X) - N(k)) + N(m + A,,) - N(m)
k=l N(m + x,,) , m = [g(n)], A." = min (x"' IX)
converges to IX as n -+ 00. We replace N(x) by y(x). In view of the proper-
ties of y(x) we find that this proposition is equivalent to
m
lim
m ..... oo
_1_
y(m) k=l
J; (y(k + IX) - y(k)) = IX.

According to 19 this quotient is, with g(l) = xo, equal to


y(m + /X)
2y(m)
- y(m) + _1_
y(m) x.
j (y(x + IX) _ y(x)) dx + 0 ('Y'(m))
y(m)
m
= y~) f y'(~) dx + 0(1),
x.

x < ~ == ~(x) < x + IX. Since y'(x) is montone we have


m m '"
y(m) - y(xo) = f y'(x) dx <f y'(~) dx < f y'(x + IX) dx
-'"0 %0 ·X(i

= y(m + IX) - y(xo + IX) •


175. Special case of 174: g(t) = at'.
176. Special case of 174: g(t) = a (log W.
177. Assume 0 < (! < 1. We write
s,. = Isin 1a~1 + Isin ~~! + ... + Isinna~l.
174 implies for

g(t) = ;1/' I(x) = Isin 2n-x I,

lim
" ..... 00
S"
n
=
0
j Isin 2n-x I dx = .!. .
:IT.

Thus (so = 0)
j;s. - s'2 =
• =1 v~
"il s.(~ __+1_) + s"
.=1 v~ (v 1)~ n~
-+ + 00 •

yuana97@uw.edu
Pt. II, Solutions 175-180 283

178. [J. Franel: Vjschr. Naturf. Ges. Zurich Vol. 62, p. 295 (1917).J
Write in decimal notation
V-n-- c(nl . c(nlc(nlc(nl •••
123 •

(We exlude the case where all the ct"s are equal to 9 for f larger than a
certain given index.) In 175 we put (J = t, a = 1()i-1; then

Xn = 10i - 1V; - [10i - 1 V;J = O. Ct)CJ~lCJ~2 ... ,

i.e. cJ'" = [10xnJ. We find cJ") = g if and only if ;0 < xn <g~1. This
means
g+l
w 1
=f g
dx = to'

10
179. Using the notation of solution 174 we have to prove
m-l
1: + (X) + N(m + An)
f K(x, e) dx,
(N(h - N(h)) - N(m) '"
lim k=l N(m + xn) =

e,
as n --+ <Xl, m --+ <Xl, xn --+ with N(x) = [ifJ. We replace N(x) by if in
the expression in question, which is justified because mq-m --+ 0, and
obtain

The last expression is . f'"


K(x, e) dx as can be easily checked by inte-
o
gration, or better by differentiation with respect to lX.
180. The function

<p(e) =
o
f1I(x) K(x, e) dx = I
_~3i
q
q-~ (1f I(x) if dx + (q -
0 0
f
1) ~)
1(x) if dx

is continuous in the interval [0, 1J, <p(0) = <p(1); <p(e) is constant if and
only if I(x) assumes the same value at all its points of continuity (differ-
entiate). As a --+ <Xl, q converges to 1, as a --+ 0, qtends to infinity. We find
accordingly
1 1
lim
q-+ooo
f I(x) K(x, e) dx = f I(x) dx. 0

yuana97@uw.edu
284 Integration
1
As a increases to infinity J(a; f) is reduced to a point, f I(x) dx; the
o
distribution is almost uniform for large a's [176]. If 0 < ~ < 1 we find
in addition 1
lim f I(x) K(x, ~) dx = I(~)
IJ-+OO 0

for any point of continuity ~ [132]. If ~ is a point of ordinary discon-


tinuity (jump) of I(x) the limit is I(~ - 0). The limit is 1(1 - 0) for ~ = 0
and ~ = 1. If I(x) is e.g. of bounded variation J(a; f) approaches the entire
range of the function I(x) on (0, 1) as a ~ O. The jumps of I(x) at points
of ordinary discontinuity are included. For small values of a the distri-
bution is almost like the distribution in 182.
181. [J. Franel, I.c. 178, pp. 285-295.] We denote the common
logarithm of n by Log n and write it as a decimal fraction
Log n = log n = cell). cen) ce,,) cell) •••
log 10 1 2 3 .
101'-1
In 179 and 180 set a = log 10 . Then
~fI = l()i-l Log n - [l()i-ILogn] = O. C(II)
1
d") C(II) •••
1+1 1+2 ,
i.e. ci") = [lOx,,]. Thus we are concerned with the range of the continuous
function 1+1
10
cp(~) = f K(x,~) dx
.!.-
lO
Jog 10

on the interval [0, 1]; K(x,~) is defined as in 179 and q = eioi - 1 = 10101 -;
182. Let [solution 174]
m = [g(n)] , x = g(t) , t = y(x) , N(x) = [y(x)].
Because of the conditions (1)-.(4) the function y(x) is continuously
differentiable for x ?': g(l), monotone increasing to infinity as x ~ 00,
moreover y'(x) ~ 00, 2':((X) ~ 00, whence "(X(-)e) ~ 0 as x~ 00 and also
" X) " X
N~(~ e) ~ 0 if B fixed, > 0 or if B has a lower positive bound as x in-
B

creases. Assume that 0 < ~ < 1, I(xi continuous at x =~, B > 0, B


arbitrary, <5 > 0 such that I/(x) - I(~) I < B whenever Ix - ~ I < <5. We
choose n so that IXII - ~ I < : [1 101] and find
!(X1) + !(x2 ) + ... + !(xII ) I
1--=-----=--n - I(~)
<1!(x 1) -!(~)I+I/(x2)-/(m + ... + I/(x lI ) -/(~)I 2MN(m+~- 15)
= - n < n +B,

yuana97@uw.edu
Pt. II, Solutions 181-186 285

provided \/(x) \ < M. According to the hypothesis we have, however,


N(m + E- d) < N(m + E -~) -+ 0 for m -+ 00.
n N(m + E- 2)

Now suppose that ~ is a jump point of I(x) ami. that again \x" - ~ \ < ~ .
Then
_ 2M N(m + E.~ + ,u(<<5) < /(x 1) + /(x2) + ... + /(x,,)
n n
< 2M N (m +/ - d) + M(<<5);
where ,u(<<5) and M(<<5) denote the greatest lower and the least upper bound
of I(x) in the interval \x - ~ \ < «5. These inequalities imply that the
limit points lie between I(~ - 0) and I(~ + 0). Now we prove that the
limit points cover the entire interval between I(~ - 0) and I(~ + 0) :
Since I(x) is integrable there exist points of continuity of I(x) arbitrarily
closetoE(108]. LetE' and~"betwosuchpoints, 0 < E' < ~ <~" < land
let x"' and x.." (n', n" integers) be two sequences such that x"' -+ E' and
x"" -+ ~", [g(n')] = [g(n")]. We set l(x1 ) + l(x2) + ... + I(x,,) = nF",
thus F", -+ IW). F"" -+ IW'). Moreover we have for each n

\F" - F "+1 \ = I n F" /(x"+1)


+ 1 - n+1
I< n2M+ 1·
The sequence F"" F",+l' F"'+2' ... , F"" is "slowly increasing" or "slowly
decreasing" in the sense of solution I 100. It is easy to adapt the proof
to the case ~ = 0 or ~ = 1.
183. Special case of 182: g(t) = a (log W.
184. The proposition follows from 182, 183 for
. -- 10'-1 --
g(t) = 1()1-1 VLogt = V-==- Vlog t, I(x) = 1 for [lOx] = g, otherwise
log 10
I(x) = O. Cf. 178, 181.
185. [R. Weyl, l.c. 162, pp. 319-320.] It is sufficient to prove the
P 0POS1·t·Ion f or I( xl' x 2, •••• xp ) = e2"i(".%, +".%.+' .. +"p :tpl ,where k l' k 2'.· .• kP
r
are integers at least one of which does not vanish [165]. Introduce
k 1 a1 + k2a2 + ... + kpap = a, k 1 ()1 + k2()2 + ... + kp()p = ();
then
-1I 0f e2"i(a+
I
61) dt =
2"i61 1
e21ria !..--.-- -+ O.
2mOt

186. Special case of 185: p = 2, l(x1 , x2) = 1, if £Xl < Xl < £x 2,


f31 < x2 < f32' otherwise l(x1• x2) = 0 in the unit square.

yuana97@uw.edu
286 Integration

187. [Cf. D. Konig and A. Sziics: Rend. Circ. Mat. Palermo Vol. 36,
pp. 79-83 (1913).J We may assume that the motion in question takes
place in the square 0 <x < t, 0 < Y < t. By reflection in the lines
x = t, y = t three domains besides f are obtained. The four domains
form a sub domain f* of the unit square 0 < x < 1, 0 < Y < 1. Trans-
lating f* by an integer number of units parallel to the axes we construct
an infinite number of domains as in 186. The original zigzag motion in
the square 0 < x < t, 0 < Y < t can be replaced by a rectilinear motion.
Special case of 185: p = 2, l(x1, x 2) = 1 if Xl' x 2 is in f*, otherwise
l(x1, x 2) = 0 in the unit square.
188. [G. P6lya: Nachr. Akad. Wiss. Gottingen 1918, pp.28-29.J
We put V'(y) = e2nikY, k positive integer, in VIII 35. Then we obtain,
using the same notation, g(n) = 0 for n > k. I.e.

Ic,.r 1 V' I
(;) = Itl,,~t (;) I< g(t) Ig(1) I + Ig(2) I + ... + Ig(k) I
for any value of n [165j; and <p(n) -'>- 00 d. VIII 264.
189. Using the notation of 188 we set in 170

a" = I(r~,,) + 1(r~. )+ ... + 1(r:~),


b" = <p(n) n = 1, 2,3, ...
190. Apply 137 to x-PI(x). Cf. 40 for the special functions
aoxP + a1xP+1+ ... + a1xp+ l , where ao' aI' ... , al are constants. The
result generalizes a well known theorem in probability theory. Cf. e.g.
A. A. Markoff: Wahrscheinlichkeitsrechnung. Leipzig: G. B. Teubner
1912, pp.33-34. H. Cramer: Mathematical Methods of Statistics.
Princeton: Princeton University Press 1966, p. 214.
191. We denote the highest coefficient of P,.(x) by k" = J!.n)~_
2"(nl)2
[solution VI 84J. Then .

(1 + x~,. )(1 + X~,.) .,. (1 + X~,,) = k;l(-J.)-" P,.(-J.) = k;lJ.-"p,,(J.)

[49,203J.
192. We find [52]

log 1 + V1 2
21
- 1 = ~
no
flog (1 + cos1 fJ) dl} .
In 1179 put
_ ~
x - l' ank
_
-
(-1)
k
k-1 (k + 2"k + ••• + k
Xl" X X,," _ ~ f n
cos
k{)d{))
,
A -
-
2
.
n no

yuana97@uw.edu
Pt. II, Solutions 187 -199 287

193. [Cf. G. Szego: Math. termeszettud. Ert. Vol. 36. p. 531 (1918).]
Follows from 192 in analogy to 164.
194. Special case of 193: I(x) = 1 if IX < X ::;;; (3, otherwise I(x) = 0
in the interval [-1, 1].
195. We can assume a l > ~ > ... > al • Then

~he expression in square brackets converges to 1. A different proof


follows from 196 and I 68. Cf. 82.
196. [Solution 195.]
197. Let I(x) = c(x - al) (x - a2) ••• (x - at). c =l= O. Because of
f(x) 1 1 1
f(x) = x - a1 +x- as + ... + x - al

we find
--
C, , a-"-l
1 + a-"-l
2 + ... + a-"-l
I' -'0
n - " 1 2, •••

Use 195, 196. (95.3. somewhat amplified. so that it applies also to finite
sums, is much more informative. See also III 242.)
198. Let I(x) attain its maximum at x = E, a ::;;; E::;;; b, e > 0, lJ be
positive and so small that

I(E) - e < I(x) ::;;; I(E)

whenever Ix - EI < lJ, a::;;; x::;;; b. Then


e+" b b
U(~) - e]" f 91(x) dx ::;;; f 91(x) U(x)]" dx ::;;; U(E)]" f 91(x) dx.
e-" A A

If E - lJ < a the lower limit E - lJ of the integration is replaced by a,


if E + lJ > b the upper limit E + lJ is replaced by b. Take the n-th root.
let n increase to infinity and then let e converge to O. Cf. 83.
199. First proof [Po Csillag]: Let 0 < e < M = max I(x). Then
b
f 91(x) U(x) ]"+1 dx ~ f 91(x) U(x)]"+1 dx > (M - e) f 91(x) U(x)]" dx,
A 1(%)ti:;M --. /(%);<:111-0

f 91(X) U(x)]" dx > f q'(X) U(x)]" dx > C (M - -it


1(·T);o:lIt - 6 M •
()
IT.o: -2

yuana97@uw.edu
288 Integration

where the positive constant C is independent of n. Consequently


b
J'P(x) [f(x)),,+1dx J
'P(x) [f(x)]"dx
M > a b --- > (M - e) _ __;;_---'f~(,z-"'),.;=-M=---.:....-__;;_----

f 'P(x) [f(x)]"dx J 'P(x) [f(x)]"dx


f(z)"M-.
+J
'P(x) [f(x)]"dx
f(,,)<M-.
II

As n --+ 00 the last quotient converges to 1 because


b
f 'P(x) [f(x)]"dx (M - e)" f'P(x) dx
f(z)<M-. :s;: a
f 'P(x) [f(x)]"dx -
f(")~M-.
C(M - 2e )"

The proof almost involves the concept of the Lebesgue integral.


Second proof: We write
b b _ ,,-1 _ "+1
1" = J !p(x) [f(x)]" dx = J V!p(x) [f(X) ]-2 . V!p(x) [f(X))"-2 dx,
a a

then [81]
1; < 1"_11"+1' n = 1, 2, 3, ...

The sequence
I
"/1 is therefore monotone increasing. The value of the
" 198 and 168.1.
limit follows from
200. By introducing the new variable Vkn(x - E) = t we transform
the integral into
1 y;;n(b-E)

Vkn .1_
J e
-I'dt
.
-'''''(E-a)

As n --+ 00 this integral converges to j e- I' dt = V;.


201. [Laplace: Theorie analytique des probabilites, Vol. 1, Part 2,
Chap. 1; Oeuvres, Vol. 7, p. 89. Paris: Gauthier-Villars 1886. G. Darboux:
J. Math. Pures Appl. Ser. 3, Vol. 4, pp. 5-56, 377-416 (1878).
T. J. Stieltjes, Ch. Hermite: Correspondence d'Hermite et de Stieltjes,
Vol. 2. Paris: Gauthier-Villars 1905, p.185, 315-317, 333. H. Lebesgue:
AnnIs Fac. Sci. Univ. Toulouse Ser. 3, Vol. 1, pp.119-128 (1909).
H. Burkhardt: Sber. bayer. Akad. Wiss. 1914, pp.1-11. O. Perron:
Sber. bayer. Akad. Wiss. 1917, pp. 191-219.] Let e> 0, () be positive
and so small that a < E - () < E + () < band
!ptE) - e < !p(x) < !ptE) + e, h"(E) - e < h"(X) < h"(E) + e < 0,

yuana97@uw.edu
Pt. II, Solutions 200-205 289
whenever ~ - 15 < x < ~ + 15. Then
Jcp(x) en(h(x) -heel] dx = E+a
b
J cp(x) en(h(x) -hem dx + O(lXn)
,. E-a

= cp(~') J
Ha ~(x-E)'h"W')
e2 dx + O(lXn) ;
E-a
where 0 < < 1, IX depends on 8 but not on n, ~ - 15 < r < ~ + 15,
IX

~ - 15 < r' < ~ + 15. The first term on the right hand side lies between
[cp(~) _ 8] Er ei-(X-E)'[h"(EJ-e] dx and [cp(~) + 8] Er e~(X-e)·[h"(el+.] dx.
E-a E-8
According to 200 these bounds are asymptotically equal to

[cp(~) _ 8]1/ _.
V-
2n; __
[h"(;) - e] n
and [cp(~) + 8]1/ __._ 2n;__
V [h"(;) + e] n
resp.

The theorem is also true if n increases continuously to infinity.


202. [Wallis' formula.]
1 • 3 ... (2n - 1) _ ~
2 . 4 ... 2n
2n dJ" .
- n; 0 SIn X X.

Special case of 201 :

a = 0, b = n, cp(x) = ~ , I(x) = sin2 x, ~ = ~ . (Follows also from 205.)

203.
Pn(A} = 2n;
1"
J (A + V-
A2_ 1 cosx)"dx.
-"
Special case of 201 :

a = -n, b = n, cp(x) = ;n;' I(x) = A + VA2 -1 cos x,


204.

i-VJ, (it) = f J e-tcosx cos vx dx.


2"

n; 0
Special case of 201 :
1
a = 0, b = 2n, tp(x) = 2n cos vx, I(x) = e- CO' X, ~ = n, n = t.
205. [Stirling's formula.]

+ 1) = n J (e-xx)" dx.
~

r(n H +1
o

yuana97@uw.edu
290 Integration

Special case of 201 :


a =0, b =00, tp(x) =1, !(x)= e-xx, = 1. ~

For more accurate approximates see 18, I 167, I 155.


206. According to 205 we have

( nk + l) rink + 1+ 1) ( nk)1 rink + 1)


n = r(n + 1) rink - n + 1 + 1) (Xl 11k - n r(n + 1) rink - n + 1)

207. The substitution of tx for x transforms the integral into

Let t be so large that t-1 < i. The integral

f
t' t X,,-1 (e )IX dx
-:;-

r
1- 1

can be disregarded because the function !(x) = (; increases - on the


interval [0, ~ ] ' i.e. !(x) ~ V2e < eon [0, ~ ] . Apply 201 to the integral

1X~-I(: r dx:

a =!, b =00, tp(x) =x"-l, f(x) =(:r ~ = 1, n = t.


1
208. Substitute T 1-" (1 + x) for x. There results
T
__
1
1-'" exp
(1 _'" _")1 exp
- " , - T 1-.>
OX> (
T
_~_
1-"
(1 + x)' -
----~;;---- dx.
1- IXo¥)

Special case of 201 :

a = -1, b =00, tp(x) =1, h(x) = (1 + x)'" -


IX .- - -,
1 - ",x

~= 0,
1
209. Substitute e-1t' (1 + x)for x. There results
e-1t: exp (e-1iXt ~) 1= exp {e-1iXt! [x - (1 + x) log (1 + x)] } dx.

yuana97@uw.edu
Pt. II, Solutions 206-210 291
Special case of 201 :
a = -1, b = 00, lP(x) = 1, h(x) = x - (1 + x) log (1 + x) ,
1
~ = 0, n = e-10f,t-;'-·
1
210. We put 'fJ = n- H ., 0 < e < (I' The integral III question
becomes [205]

= V;n [1 n
+O(!)r -jP"-\e- X (l +x)tdx

Y
= ;n [1 + 0 (!)r"-iY"-\e-
-'I
X (l °
+ x)t dx + (V ne- l ,,2e).
Since the function e- x (1 + x) increases for x < 0 the integral over
[-1, -'fJ] is smaller than [e'l(l - 'fJ)t. We now expand the integrand
on the remaining interval
-~+~-~(lHx)-4
e- X (l + x) = e 2 3 4 • 0< 0 = O(x) < 1.
-n~(lHx)-4
The factor e 4 is of the form 1 + O(n- 1 +4 6 ). The integral be-
comes therefore

We have
x'
/-3 = 1 + n ~ + O(n-1+6,).
",,-i+lln- 1 -n~ n~
Since J e 2 dx is of order n- l the O-term of e 3 yields a contri-
-'I
bution of O(n-1+6,). Hence

Y;n [1 +O(n- 1 )r- y,,-1 e-"~ (1 +


+4 6
1
nX;)dx + O(n-1+6,)
-'I .

1 o:+pn-t x' "+lIn-1 x'


J J
3 3
=--:- e- Z (l +~)dX + O(n-1+66) = 1 e-~(l+ x _)
V2n _,,' 3Vn V2n -00 3Vn
X dx + O(n--1+6,)

yuana97@uw.edu
292 Integration

211. [Cf. A. de Moivre: The Doctrine of Chances, 2nd Ed. London


1738, pp. 41-42.] We have

K (x)
"
= ~ IX e-Xx" dx =
n!o
1_e- x (1 + -=- + x + ... + x").
I! 2!
2
n!

Therefore x" is the only positive root of the transcendental equation


K,,(x) = 1 - A. According to 210 we find for arbitrary n-free (X and {J

K,,(n + (XV -n + (J) = A + V;--


B
+ 0 ( V;-1 ) ,
where A and B have the same meaning as in 210. Determine (X and {J so
that A = 1 - A and B = O. Then x" - (n + (X V; + (J) must converge
to zero. If, on the contrary, we had x" - (n + (X V; + (J) > c > 0 for
infinitely many n we could conclude that

1 - A = K,,(x,,) >+ (X V-n + (J + c) = A + V;-


K,,(n B'
+ 0 ( V;-1 ) ,
where B' depends on (X and {J + c as B depends on (X and {J. In particular
a l (XI

B = B + - --- ce
, 1 -"2 1 -2" .
= - - ce > O. Smce A = 1 - A the last in-
V2n V2n
equality is impossible. In a similar way we show that

x" - (n + (X V; + (J) < -c < 0


cannot hold for infinitely many n.
212. By a similar computation as in 201 : Instead of 200 consider the
formula
H~ _
'I" , 1
I I
",'12k j'

e- kn (.,-<) dx '" -V-~- e -"2 dt,


a 2kn - 00

where a is real, k> 0, a, k fixed. 212 does not completely imply 210.
213. By a similar argument as in 201 we find that the integral in
question is
<+e" <+e"
'" I q>(x) e"h(x) dx = q>(t) e"h«) I e,,[h(x) -h«)] dx,
<-6 <-6
where e" = (Xn- 1 log n +
(In- 1 , b a positive constant, b so small and n
so large that q>(x) is continuous, h(x) twice continuously differentiable
and h' (x) > 0 in the interval of integration; we have ~ - fl < f < ~ + e".
Put 'Y/" = n- 3 /4 and choose n so large that e" < 'Y/" < b. On [~ - fl, ~ - 'Y/,,]
the integrand is of the order of e-nT/n h' = e-,,1/4h' where h' denotes a posi-

yuana97@uw.edu
Pt. II, Solutions 211-216 293
tive lower bound of h'(x). On the remaining interval expand up to terms
of second order,

~ - TJ,. < ~" < ~ + e,. .


Herein h"W') is bounded and n(x - ~)2 < nTJ! = n- 1, furthermore

e+.,. no "'(e) -,."


""'(e) ( log,.+R)"'(e)
J e,."'(mx-~) dx -_ e" - e
nh'(~) N
e....
nh'(~)
e-",.
214. A change of variables leads to

[205,213].

215. According to solution 211 we have

Determine the constants IX and p in 214 so that A = 0, B = 1. Then


x" - (~n + IX log n + P) must converge to 0 [solution 211].
216. Assume ~J%)
x
< G= const. for x > 1. We split the integral

a =----
e _"n ,,+1 Jco (
el-x+x-
g(,.x) )"
"XX dx
" nl 0

into four parts corresponding to the intervals (0, e), (e, 1 - e), (1 - e, 1 + e),
(1 + e, 00) where e does not depend on n, 0 < e < !, e < y and so small
that in the first interval xe 1 - x + G• < 1. In the second and fourth interval
xe1 - x < 1; choose ~ = ~(e) so small that we have evenxe1 - x +o'x < 1 and
n so large, n > N = N(e), that g(nx)
nx
<~ and ne> 1. Then, except on
the third interval, the integrand is 0(0"), 0 < 0 < 1, 0 independent of x
and n, 0 = O(e). The mean value theorem of integrals [in addition note
205] implies that
1+. -
J
n+l
+ O(VnO"), 1 - e < ~ < 1 + e.
-II

a" = cl(ne) _e-';-- (e1 - X x)" dx


n. 1-8

Hence
log a" = g(n~) + 0(1).
g(n) g(n)

yuana97@uw.edu
294 Integration
Now lim g(lX(n)) exists and the convergence is uniform on 1 - E < IX < 1 + E.
"-+00 gn - --
The limit is arbitrarily close to 1 if E is sufficiently small.
217. The integral in question can be written in the following way:

22" .. 'In t·(COSy~ -1) " ~==_.!'. ___


n!
(2n - 1) (2n - 2) (2n - 3) ••• n V; J_
- ..'I,.
n.x
~=1 .1-
dx.

2n e'" - v

Apply 115. The limit of the integrand is e- xl [59], the convergence is


uniform on any finite interval as a supplement to the proof of 59 shows.
For an appropriate F(x) in the sense of 115 cf. G. P6lya: Nachr. Akad.
Wiss. Gottingen 1920, pp. 6-7.-For the generalized Laplace formula cf.
also R. v. Mises: Math. Z. Vol. 4, p. 9 (1919).
217.1. The real-valued functions !p(x, y) and h(x, y) are defined in the
(bounded or unbounded) region m
and satisfy there the following con-
ditions:
m
(1) !p(x, y) e""(x,,,l is absolutely integrabl~ in for n = 0, 1, 2, ...
(2) The function h(x, y) attains its maximum in mat a single point
~, "I, and in the region m' that remains when we exclude from ma closed
neighbourhood of tTJ the upper bound of h(x, y) is less than h(E, "I).
Moreover, the second partial derivatives hu, hx" and h"" exist and are
continuous in a certain neighbourhood of E, "I and at the point E, "I

hu < 0, huh"., - h!" > O.


(3) !p(x, y) is continuous at the point E, "I and !p(E, "I) =l= O.
The proof is closely analogous to the proof of 201.
218. [Cf. G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24,
p. 282 (1916).] Let x"' x" > n, be the point at which the function attains
its maximum M,,:

M" = V;';-(X,. - 1) (x" - 2) ••• (x" - n) a-x" = x" -=- n (X,,) a-X,. ,
n! n! Vx" n

1 1 1 1
--+-- +~+ ... +--=loga.
2x" x" - 1 x,. - 2 x" - n

In view of 16 we derive from this that x" = ( n + !) b +


e" where
b = (1- a-I)-I, lim e" = O. Wehaveb > 1. Thus we obtain fore positive
"-+00
and n sufficiently large

((n + l~ b - E) < (~ ) < ('n + l~ b + E) .

yuana97@uw.edu
Pt. II, Solutions 217-224 295

According to 206 these two bounds are asymptotically equal to

(b - ~ (_b _)("+ llb+! -a and (b - I)" (_b_)("+!lb+i+ a


y'2nn b- 1 y' 23tn b- 1 resp.

219. Analogous to 218, use 17.


220. We put I(n, x) = IQ,,(x) 1 a- z, then we have for m - 1 < x < m,
m positive integer, a >1,
I(m - 1, x) > I(m, x) ~ I(m + 1, x) ~ .••
[III 12] so that the least upper bound of I(n, x) for fixed positive x and
variable n is attained when n < x. Cf. 218.
221. [219, 220.]
222. The point x" where M" is attained is determined by
x"
n = x" + ap~. We have n > x"' n-+oo
lim -
n
n -x"
= I, lim - - = ap. Conse-
"-+00 n"
quently x" = n - apnP + o(nP), log x" = log n - apnP- l + o(nP- l ),
thus
M"
log -n.I = n log x" -x" - a~ - n log -e
n
+ o(nP) = -anP + o(nP).

*223. Let
tp(x1) = l(x1, Yl)' tp(x~ = I(x", y~.
By the definition of tp(x)
tp(x1) ~ l(x1 , y~,
and by the continuity of I(x, y)
l(x1, y,,) > I(x", y,,) - 8

when 1Xl - x,,1 is sufficiently small. Hence


tp(x1) ~ tp(x,,) - 8

and we can interchange Xl and x" in the foregoing argument.


*224. Let

m;x(mzin/(x, y)) = I(x l , Y1)' rnzin (m:x/(x, y)) = l(x2, Y2)·

In view of the first operation (the inner one, written on the right)

yuana97@uw.edu
296 Integration

*225.
1 - (y - 1)2 when '\I > 2
max I(x, y) = 1, min I(x, y) = { - - ,
y x 1 - (3 - y)2 when y :s;: 2.
min max I(x, y) = 1, max min I(x, y) = 0
x y y "

*226. By using 198 complete the following outline: The quantity


whose limit is desired is
,.' ] -1/..
'" [ [qJ(x) -n dx '" min qJ(x) .

yuana97@uw.edu
Part Three

Functions of One Complex Variable

1. Z + Z = 2x, Z - = 2iy, z zz = r2.


2. The open right half-plane; the closed right half-plane; the open
horizontal strip bounded by the lines y = a and y = b parallel to the
x-axis; the closed sector between the two rays which form the angles IX
and {3 resp. with the positive x-axis; the imaginary axis; the circle with
center Zo and radius R; the open disk and the closed disk resp. with center
Zo and radius R; the closed annulus between the two circles with radii
Rand R' and centred at the origin; the circle with center at Z = ~ and
. R
ra dlUS2"'
3. The ellipse, and the domain bounded by the ellipse, with foci
Z = a and Z = b and the semimajor axis k if Ia - b I :::;; k. (If Ia - b I = k
the ellipse degenerates into a segment.) If k < Ia - b I no point z satisfies
the condition.
4. Let Zl and Z2 denote the two roots of the equation Z2 + az + b = O.
The region in question is the interior of the curve IZ - zlll Z - z21 = R2
with "foci" Zl and Z2' The curve is the locus of all points for which the
product of the distances to Zl and Z2 is constant, equal to R2. It consists
of two pieces for R < [Zl ~ Z2 [ and of one piece for R > ~i.!~[ . If
R = [Zl ~ z2I the curve is called lemniscate.
5. The condition in question is equivalent to
IZ - a 12 ~ 11 - az 12 or to (1 -I a 12) (I Z 12 - 1) ~ O.
The first set is the open disk IZ I < 1; the second set is the unit circle
IZ I = 1; the third set is the exterior IZ I > 1 of the closed unit disk. The
value of the expression in question is ! a 1-1 for Z = 00, thus Z = 00
belongs to the third set.

yuana97@uw.edu
298 Functions of One Complex Variable

8. The condition in question is equivalent to


la-zIZ:la+zIZ or to -m(a+fi)z:O.
a
Since a + is real and positive the condition means mz ~ 0. The first
set is the open right half-plane; the third set is the open left half-plane;
the second set is the imaginary axis. (The value of the expression in
question is for z = 00 equal to -1; z = 00 belongs to the second set.)
7. With the notation a + i <5. Zr = X + iy. the equation reads
= y Zl

!X(xz+ yZ) + 2(yx + <5y) + {J = 0.


8. Suppose that a wheel of radius a is rolling on the real axis. Let P
be a point on the wheel at a distance b from its center. The point zl
moves on a straight line. the path of the center of the wheel; the point
Z2 describes the path of the point P if the wheel would rotate around the

origin but not slide. The point z = Z1 + Zz describes a prolate. a regular.


or a curt ate cycloid according as a >. =. or < b.
9. The point describes an epicycloid.
10.
dz _dl' i8 +. i8dO
dt - dt e ~re tit'
d2z _ d." i8
tit2 - dt2 e
+ 2' dl' dO
~ dt dt e
i8 _
re
i8 (dO)2
dt
+ ~re
. i8 d2{)
dt2

(dO)2] i8 ie i8 d (
d'"
= [ dt2-r dt e +rdt r dO)
dt •
The coefficient of ei8 is the radial component. the coefficient of iei8 is
the component perpendicular to the radius.
11. In the annulus
mIl: n< Izi <n +1. n = 0.1. 2. ".
the inequalities

1< IiTz I Iz21


<
I z"-l
2! <"'<I(n-l)! <In! >I(n+l)! > ....
I I z"l I z,,+! I
are satisfied; i.e. in mIl the absolute value of the n-th term is larger
than the absolute value of any other term. On the common border of
mIl and m,,+! the absolute values of the n-th and the n + 1-st term are
equal. all the others are smaller. In general: let
ao +az+az 2 +"'+az"+'"
1 2 ,,' •
be an everywhere convergent infinite power series. Then the z-plane can
be divided by concentric circles in such a way that in each annulus the

yuana97@uw.edu
Pt. III, Solutions 6-13 299

absolute value of a certain term is largest (maximal term). The subscripts


of these terms increase as z proceeds from one annulus to the next
larger one. [1119,1 120.J
12. The circles

Iz - nl = n + 1, n = 0, 1, 2, ... ,
are tangent to one another at z = -1 where they are perpendicular to
the x-axis. They intersect with the x-axis also at the points z = 2n + 1.
<rn+1 contains <rn' In the crescent shaped region ffin inside <r" and outside
<rn - 1 the inequalities
izl >1, IZ-
IT - 2 - 1>1, ... , Iz-n+11
1
n 1>1, Iz-nl n+2 11<1, ...
n+1 <1, Iz-n- 1

are satisfied. Hence

1< 1(:)[ < I( ~)I < ... < I(n ~ 1)1 < 1(:)1 > I(n ~ JI > ... ,
i.e. the absolute value of the n-th term is larger than any other in mn •
On the common boundary of mn
and m"+1 the n-th and the n + 1-st
terms have the same absolute value, all the other terms have smaller
absolute values. (At z = -1 all the terms have the absolute value 1.)
The regions mn together with their boundaries cover the entire half-
plane mz> -1, including z = -1. If ffiz < -1, z =1= -1, the absolute
values increase monotonically, there does not exist a largest term.
13. The lemniscates
n = 1,2,3, ...

are tangent to each other and to the straight lines mz = ±3z at the
point z = O. They intersect the real axis at the points ±n Vi 2"+1
contains 2n- The inequalities

11 - :: I> 1, 11 - ;: I> 1, ... , 11 - :: I > 1,

'll---~-I<l
(n + 1)2 ' 1 1- (n +Z2 1<1 ' •• ,
2)21

hold in the "double crescent" shaped region ffin between 2"+1 and 2,. (lRo
is the region bounded by 21)' Therefore the inequalities

m
are satisfied in n , i.e. IPn (z) I is in ffi" larger than the absolute value of
any other partial product. On the common boundary of and m,,+l mn
the absolute values of the n-th and n + 1-st partial products are equal

yuana97@uw.edu
300 Functions of One Complex Variable

and larger than any other. (At z = 0 all vanish.) The mIl's together with
their boundaries fill out the region consisting of the two angles
:rr :rr 3:rr 5:rr
-"4< argz<"4'4< argz<4
and the point z = O. Outside this region and this point P,,(z) increases
with n monotonically.
= arg J f(t) eiq>(t) dt. Then
b
14. We put f}
"
IJ I(t) eiq>(t) dtl = e- iIJ J 1(t)i'P(t) dt = J I(t) cos [«p(t) -
b ' b b b
f}] dt < J I(t) dt,
II II 4 II

except if «p(t) = f} (mod 211:) at all points of continuity of «p(t).


15. [K. Lowner: Math. Ann. Vol. 89, p.120 (1923).] It is sufficient
to prove m(4p2 - 2Q) < 3. [Replace «p(t) by «p(t) + ~ where
f} = arg (4p2 - 2Q), cf. solution 14.] Now we have [II 81]

m(4p2 - 2Q) = 4 ( / e- t cos «p(t) dty - 4 (i e- sin «p(t) dty -


t

J e- 2t cos 2q>(t) dt :s;:


00

- 2
o

< 4 (i e- t Icos «p(t) I dt)


2
- 2/ e- 2t cos 2q>(t) dt :s;:

:s;: 4 J e- t cos 2 «p(t) dt - 2 J e- 2t cos 2q>(t) dt =


00 00

o 0

J (e- + 1 :s;: 4 J (e- t - +1=


00 00

= 4 t - e- 2t ) cos2 «p(t) dt e- 2t ) dt 3.
o 0

If m( 4p2 - 2Q) = 3 then cos 2 «p(t) = 1 and cos «p(t) has the same sign
at all the points of continuity of «p(t), i.e. «p(t) = 0 or «p(t) = 11: (mod 211:).
16. The function P1z- 1 + P2Z-2 + ... + p"z-" is monotone decreas-
ing from 00 to 0 as z is positive and increasing; therefore it assumes the
value 1 at exactly one positive point C. Consequently
z" - P1Z"-1 - P2z"-2 - ••• - p" > 0 or < 0
according as z > Cor z < C.
17. We have
Izo I" = lalz~-l + a2z~-2 + ... + a,,1
< la11Izol"-1 + la21I zol"-2 + ... + lanl,
hence, according to 16, IZol < C.

yuana97@uw.edu
Pt. III, Solutions 14-24 301
18. Apply 17 to a;;lz" P(Z-l).
19. The two test polynomials considered in 17 and 18 are, in this
case, identical, namely z" - Ic j.
*20. By 16 and 17 it is enough to ascertain the sign of
M" -Iall Mn-I_la2IMn-2 - ... -Ianl
> ~ - cIMM"-l - c2M2~-2 -'" - c"M" > O.
*21. From 20. Substitute for ck

respectively and add a little remark to 20 in the last case.


22. [G. Enestrom: Ofvers. K. Vetensk. Akad. Forh. 1893, pp. 405-
415; T6hoku Math. J. Vol. 18, pp. 34-36 (1920); S. Kakeya: T6hoku
Math. J. Vol. 2, pp.140-142 (1912); A. Hurwitz: T6hoku Math. J.
Vol. 4, p. 8f! (1913).] We find for Izl < 1, z =l= 1,
1(1- z) (Po + Plz + P2 Z2 + ... + Pn z") I
= Ipo - (Po - PI) z - (PI - P2) Z2 - ... - (P"-l - P,,) z" - p"Zn+11
> Po - I(Po ~ PI) z + (P1 - P2) Z2 + ... + p"z"+11
> Po - (Po - PI + PI - P2 + ... + P,,) = 0,
because (Po - PI) Z, (PI - P2) Z2, ... , PnZ"+1 cannot have all the same
argument. (Unless z? 0, in which case the proposition is trivial.) A
weaker statement, < instead of <, follows immediately from 17.
23. We replace z by ~- and.!L respectively and choose
e z
e, e > 0 (in
the second case we first multiply by z"), so that 22 can be applied.
24. We call the polynomial in question I(z). For mz > 0, Izl > 1, we
have m-z1 ?- 0, therefore

If(z) I > la an_II _ an_ 2 _ a,,_~ _ ... _ ~


I z" I = I ,,+ z I IZ l2 Izl3 Izl"
> m(a" + a"z_~) -1:1 2 - 1:1 3 -'"

9
? 1 - Iz I2 -IZI'
The last expression is ~ 0 if Iz I ~ r, where r is the positive root of the
V
equation r2 - r = 9, r = ! ((1 + 37). 3 < r < 4. The polynomial
corresponding to the number 109, 9 + Z2, has the roots + 3i.

yuana97@uw.edu
302 Functions of One Complex Variable

25. [Ch. Hermite and Ch. Biehler; d. Laguerre: Oeuvres, Vol. 1.


Paris: Gauthier-Villars 1898, p. 109.J We write

P(z) = U(z) + iV(z) = ao(z - Zl) (z - Z2) ••• (z - z,,), ao =F 0;


let x be a root of V(x) = 0 or U(x) = O. Hence
U(x) + iV(x) = U(x) - iV(x) or U(x) + iV(x) = - [U(x) - iV(x)J,
that is
ao(x - Zl) (x - Z2) ••• (x - z,,) = +ao(x - i l ) (x - Z2) ... (x - i,,).
Such an equation, however, is possible only for real x. Assume 3x > 0;
then Ix - z.1 < Ix - i.1 for 11 = 1, 2, ... , n. By the same token xcannot
lie in the open lower halfplane.
26. [Cf. 1. Schur: J. reine angew. Math. Vol. 147, p.230 (1917).]
Put P(z) = ao(z - zl )(z - Z2) •.. (z - Z,,), a o =F O. Let x be a root of
P(z) + P*(z) = o. Then IP(x) I = IP*(x) I; since P*(z) = o(l - Zl z) a
(1 - Z2Z) ••• (1 - z"z) we have

n" Ix - z.1 = .=1


.=1
n" 11 - i.x I·
Such an equation can hold only for ~ x I = 1. Assume Ix I < 1: then [5J
x- z.x
I z.1 < 11 - I for all 11, thus the first product is smaller than the
second. By the same token it is impossible that Ixl > 1. We argue
analogously in the case of P(z) + yP*(z), II'I = 1.
27. [M. Fekete.J Define I' = AP(a) + p.P(b), 0 < A < 1, A + p. = 1.
If all the zeros of P(z) - I' = ao(z -'- Zl) (z - Z2) ···(z - z.. ) were outside
the domain in question, the inequality
a -z"
--<arg--<-
:If :If
11= 1, 2, ... ,n
n b - z. n '
would hold, thus

-1'& < arg


Pta) - "
P(b) .=--,; < 1'&, in contradiction to
Pta) - "
P(b) _ -,; -
P

28. We assume that the straight line mentioned is the imaginary
axis and 9lz. > 0 for all 11 (this can always be achieved by multiplication
with a suitable ei <»; then we have also 9l ~ > 0 and
z.
Zl + Z2 + ... +z.. =F 0,
9l -(1+ -1+ ... + -1) > 0,
ZI Z2 Z"
~+~+···+~=FO.
ZI Z2 Z ..

yuana97@uw.edu
Pt. III, Solutions 25-32 303

The conclusion holds also in the case where all the points are in a
closed half-plane determined by the straight line unless all the points
lie on the line itself.
29. Cf. 28.
30. We have mi(zi - z) + m 2(z2 - z) + ... + mn(zn - z) = O. Apply
29 to the points m.(z. - z). If m.(z. - z) lies on one side of a straight
line l' through the origin, z. lies on the corresponding side of the straight
line l through z and parallel to l'.
31. [Gauss: Werke, Vol. 3, p. 112. G6ttingen: Ges. d. Wiss. 1886;
Vol. 8, p.32, 1900; Ch. F. Lucas: C. R. Acad. Sci. (Paris) Ser. A-B,
Vol. 67, pp.163-164 (1868); Vol. 106, pp.121-122 (1888). Cf. also
L. Fejer: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 145, p.460 (1907),
and Math. Ann. Vol. 65, p. 417 (1907).J First solution: The vector
determined by the complex number 1 represents a force directed
Z - a
from a to z the magnitude of which is inversely proportional to the
distance. Let Zi' z~, ... , zn denote the zeros of P(z) and let z be a zero
of P'(z) different from Zi' Z2' '00, zn' Then

P'(z)
P(z)
= i
.=iz-z,'
_1 __ = 0 i.e. __ 1
z-zi
+
z-z2
1 __ + 00' + _1___ =
z-zn
0,

this means that z represents an equilibrium position of a material point


SUbjected to repellent forces exerted by the points Zi' Z2' ..• , Zn and
inversely proportional to the distance. If Z were outside the smallest
convex polygon that contains the z.'s the resultant of the several forces
acting on z could not vanish: there could be no equilibrium. (315.)
Second solution: Using the same notation as before we have

thus
z = miz i + m z + ... + mnzn,
2 2 mi + m + ... + mn =
2 1,

where the v-th "mass" m. is proportional to Iz ~ z.12 ' V = 1, 2, 00', n.

32. [L. Fejer, O. Toeplitz.J Let C denote an arbitrary interior point


of the convex polygon in question. Then

C= AiZ i + A2z2 + ... + Anzn, Ai > 0, A2 > 0, ... , An > 0,


Ai + A2 + ... + )'n = 1,

yuana97@uw.edu
304 Functions of One Complex Variable

thus for C=F z.

A.IC - z.12
m• = A1 IC - Zl
12 + ~ I'"., -Z2
A Ie -
12 + .... +.. z.. 12 , 11 = 1, 2, ... , n.

Approximate m. by the rational numbers P;, 11 = 1, 2, ... , n,


PI + P2 + ... + P.. = P, and remember that the roots of algebraic
equations are continuous functions of the coefficients. The derivative
of the polynomial n. (z -
.=1
z.)i'. has a zero arbitrarily close to C. Since C
lies inside or on the border of at least one of the triangles determined
by three of the points z. it is sufficient for this problem to know that
the zeros of a polynomial of degree 2 are continuous functions of the
coefficients, which is obvious. [Remark due to A. and R. Brauer.J

33. [M. Fujiwara: T6hoku Math. J. Vol. 9, pp.102-108 (1916);


T. Takagi: Proc. Phys. Math. Soc. Japan Ser. 3, Vol. 3, pp. 175-179
(1921).J Set P(z) = ao(z - Zl) (z - Z2) ••• (z - z.. ) and let z denote a
point at which
P(z) - cP'(z) = 0, P(z) =F O.
Hence
P'(Z) 1 1 1 1 1
(1) ---=-+-+···+----=0.
P(z) c Z - Zl Z - Z2 Z - z.. C

Introducing
1
m1 =-1Z--1
- Zl 2 '

we can write (1) as

(2)

The first term pn the right hand side of the equation represents the center
of gravity of a certain mass distribution at the points zl' Z2' ••• , z.. ,
that means a point inside the smallest convex polygon containing all
the points z•. The second term represents a vector parallel to the vector
c. Hence the statement follows. -Cf. V 114.

yuana97@uw.edu
Pt. III. Solutions 33 - 36 305

34. [T. J. Stieltjes: Acta Math. Vol. 6, pp. 321-326 (1885); G. P6lya:
C. R. Acad. Sci. (Paris), Vol. 155, p. 767 1-769 (1912).] Let %v,
V = 1, 2, ... , n denote the zeros of P(z) and assume A (z.) =l= O. Then

p' (z.) =l= 0 because otherwise the differential equation for P(z) would
imply that pIt (z.) = 0 and repeated differentiation would show that
P(z) is identically zero. The equation

__~ ___ +_ 1_+".+===+===+.,,+ 1


3. - 3"

implies [31] that z. lies in the interior of the smallest convex polygon
that contains the points zl' Z2' .'" Z._l' z.+1' "., Zn' aI' a2 , ••• , ap (on the
line segment that contains all these points). Consider now the smallest
convex polygon that encloses Z1' Z2' ... , Zn' a1' a2, .•. , ap ' Only t.he a:s
and no z. different from the zeros of A (z) can lie on the polygon.

35. [L. J. W. V. Jensen: Acta Math. Vol. 36, p. 190 (1913); J. v. Sz.
Nagy: Jber. deutsch. Math. Verein. Vol. 31, p. 239-240 (1922).] We
denote the zeros of /(z) by z1' Z2' ••• , zn and assume
1 + ---
----- 1 + ". + ---
1 = 0,
-
z =l= z, z =l= z.' v = 1, 2, "., n.
3 - 31 3 - 32 3 - 3,.

Because of the pairwise symmetry of the zeros we have

. (1
~3 3-3
1) =0.
+ 3-Z
v=1 " Ii

The formula

Z= X
.
+ zy, Zo = Xo
+ zYo,
. ~
,-S
(_1_ + _1_) = 2 J'~
Y I (3- - (x30)- X O)2 -
Iy2 '
3 - 30 3 - Zo (3 - 30) 2

shows that the above equation can not hold when z is outside all the
circles described.
x
36. Writing zn = xn + iy.. we find IZn 1 <---"-.
- COS IX
Therefore the con-

vergence of ~ xn implies the convergence of ~ IZn I. The converse is


obvious. n=1 n=1

yuana97@uw.edu
306 Functions of One Complex Variable

37. Set z.. = x.. + iYn' The hypothesis implies in turn the convergence
of
00 00 00.

1; x .. ' 1; mz! = 1; (x; - y!), 21; x! - 1; (x! - Y!) = 1; Iz.. 12 •


.. =1 .. =1 .. =1 .. =1 .. =1 .. =1
..
38. Example: z .. = e2ni..& (log (n + 1))-\ () irrational, 1; e2nipkIJ is
bounded as n -')- 00 [solution II 166, Knopp, p. 3Ui]. >=1
39. We assume that all the numbers z.. are different from zero and
that they are arranged according to increasing magnitude,
t) < IZ11 < IZ21 < IZal < .... We enclose each zp, 11 = 1, 2, ... , m, in a

circle with center Z. and radius ~ . These circles have no common inner
points and are completely contained in 1Z 1< Izm 1+ : . Therefore

mn 4" < n
{}2 ( Iz". I + 2"" )2.
' 1.e.
l' log m
1~':!!P log \z".\
<
=
2 [1113].

40. The expression in question is

The first factor is everywhere dense on the unit circle [I 101]; the second
is a sum of rectangles and converges to
1
Jx
1
ia dx = --.-'
o 1+~(X

41 •
lim IZ
....... 00 ..
12 = (1 +~) (1 + ~) (1 +~)
12 3 ... (1 + ~ \ ...
22 2 nIl

_ (Sin:nx) _ e" - e- n
- ~ x=i - ---~.

Writing i + n = VI + n2 e2ni6n , 0 < {}.. < I, we have tan 2:nD.. = ~; n


thus the series {}1 + {}2 + {}a + ... + {}n + ... diverges and lim {}.. = O.
....... 00

We find

According to 1101 the limit points of z.. cover the entire circle

(
Izl='--~
e:'- e-n)i.

yuana97@uw.edu
Pt. III, Solutions 37 - 44 307
42.
r2=lzI2=
n n
(1+-1)(
1
1 +21) ( + -n1) =_·_···--=n+1
-"·1 2 3 n+1
12 n '

V-- I = 1' "n~ I'n_1


Izn+1 -z" 1=1_~1
n +1 rp,,- rp,,-1
= Vn+i- V; C\Jn(1/1 +
1
arctan- V
1
n
-1)
V;
1 1
="2- 8n+····
Hence rIO C\J 1 fP" [I 70].
43. According to II 59 and II 202 the absolu te value of the expression
in question converges to e- I •• Therefore it is sufficient to prove
. t
V;
2
nsm
t "
2n sin V-
log 2 - ~ arctan - - - - t - = 0(1),
n .=1 2n cos V:--1/
where - ~ < arctan x < -i- . Let n > t2, then [I 142]
I . t I
~ sm V; ::;: 2 V; It I < 2 V; I~ = O(n-l).
2
2n cos / : -1/ - 2n (1 _ ;:) _ 1/ - n - t

Thus arctan x = x - 3"


x8
+ ... can be replaced by x. The statement
now follows from
1
log 2 - - 2:" 1
= O(n- 1 ).
n v=] 2 - [: + 2 (1 - cos V~ )]
This estimate can be justified by a slight extension of II 10: if we
replace in the sum Lin the term a+ I( a+ v~;-~) by I (a + vb : e,,)
where e" = O(n-1) we still obtain LI" = O(n-1). (We assume that the
interval of definition of I(x) contains the interval a - eN < x < b + eN'
N < n, and that I(x) is bounded on this interval.)
44. Assume that lim
n-+oo
(1" = (1, e" < K and lim z" = z,
n-+oo
Iz,,1 < M,
Iz I < M, K and M independent of n. The series

IX = ao + a1 + a2 + ... + a" + ...


and

yuana97@uw.edu
308 Functions of One Complex Variable

are absolutely convergent; let E be an arbitrary positive number and


N = N(E} be such that Iz" - zI < E whenever n > N. We define
w = (0' - IX) Z + f3 and deduce from

w" = ( !o")a. z + " a.z. + " (a". - a.) (z. - z)


0'" - .~ .~
that

i a./+/ .=,,+1
Iw-w"I<MIO'-O'"I+MI .="+1 i a.z.1
N
+ 2M ~ Ia". - a.1 + 2KE .
• =0

45. [Cf. 1. Schur: J. reine angew. Math. Vol. 151, pp.l00-l0l


(1921); F. Mertens: J. reine angew. Math. Vol. 79, pp. 182-184 (1875).]
We denote by V" and W" the partial sums of the series .J; v" and.
,,=0
.J; (uov" + U1V,,_1 + ... + u"vo)' The two sums are related by the equa-
,,=0
tion
w" = u"Vo + U" __ I VI + ... + UoV", n = 0,1,2, ...
In order that a convergent sequence V" generates a convergent sequence
W" it is necessary that the sums lu,,1 + Iun-Il + ... + juoI are bounded,
+
i.e. the series U o u 1 + u 2 + ...
must be absolutely convergent. In this
case the conditions (1) and (2) (cf. p. 111) are automatically satisfied.
Therefore the absolute convergence of the series
U o + u1 + u 2 + ... + u" + ...
is the desired necessary' and sufficient condition.
46. [Cf. 1. Schur, I.c. 45, pp. 103-104; T. J. Stieltjes: Nouv. Annis
Math. Ser. 3, Vol. 6, pp. 21-0-213 (1887).] Let V" and W .. be the partial
sums of the series ~ ~ (~Utv!!...) . Now [VIII 81]
v" and,,=1
,,=1 !I" t

W" = U1 V" + U2 V[i] + USV[i-] + ... + u"V[~]' n= 1,2,3, ...


The coefficient of Vk is equal to the sum of those u/s for which [~ ] = k.
Set 11 = [V;]. For those values of l which are less or equal to 11 the coeffi-
cient of V[.i] is precisely u/. This follows from the fact that for 2 < l :::;; 11
n n n n
I - 1- T= -/(1 - 1) > [2 > 1.

yuana97@uw.edu
Pt. III, Solutions 45-49 309
In order that the sum of the absolute values of the coefficients in the n-th
row be bounded the same must hold for Iu) I + Iu 2 1+ ... + Iu. I, that
is, u l + u2 + ... + U" + ... is absolutely convergent. Hence the validity
of the other conditions follows, and so the absolute convergence of the
series u l + u 2 + ... + u" + ... is the desired necessary and sufficient
condition.
47. [R. Dedekind, cf. Knopp, p. 315; Hadamard: Acta Math. Vol. 27,
pp. 177-183 (1903).-Cf. I. Schur, l.c. 45, pp. 104-105.] Put

A" = ao + a) + a2 + ... + a", B" = yoao + ylal + Y2a2 + ... + y"a",


then
,,-1
B" = 1: (y. -
• =0
Yv+1) A. + y"A" .
48. [Cf. G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p. 282
(1916). Solved by S. Sidon: Arch. Math. Phys. Ser. 3, Vol. 26, p.68
(1917).] Set

euo = zo, Uo + u l + ... + U"_l + eu" = Z,,' n = 1, 2, 3, ... ,


U o + u l + ... + u"_l + u" = W", n = 0, 1, 2, ... ,

1: u"C" = U(C), 1: Z"C" = Z(C), 1: W"C" = W(C).


"=0 ,,=0 ,,=0
By comparing the coefficients we find that

W(C) = 1U~C~, Z(C) = 1U~C)C + (e - 1) U(C) ,


and hence
Z(C)
W(C) = c + (1 - c) C•
Assume e ::j= O. Compare the coefficients in the last equation:
1)"
W" =
(c -
--+1-
? Zo + (c - ?1),,-1 Z1 + ... + -c 2
- 1
c c
1
- 7.,,_) + - Z", n = 0, 1, 2, ...

According to the criterion on p.111 i; (c :-+!)" must be absolutely


"=0 c
convergent. This is the case if and only if Ic : ~ I < 1, i.e. if and only if
me> t.
49. [I. Schur: Math. Ann. Vol. 74, pp. 453-456 (1913.)] The case
e = -k, k positive integer, can be excluded from the start; example:
n ) U o + u l + ... + u"
u" = ( k _ 1 ' u" + e n+ 1 = 0 for k > 2, u" = log (n + 1)

yuana97@uw.edu
310 Functions of One Complex Variable

+ U 1 + ... + U
for k = 1 [169]. The case e = 0 is obvious. We put un +e U
0
'n +1
n

= Zn' Un = wn and, as on p. 111,

Multiplying the relation

(n + 1) zn - nZn_l = (n + 1) wn - nWn _ 1 + ewn , n = 1, 2, 3, ...

byr(;(~ : ~) 1) and adding the first n equations we obtain

r(n + c + 2) ;, F(v + c + 1)
r(n + 1) w.. - T(e + 2) Wo = '~I r(v+ 1) [(v + 1) z. - VZ,_I]

,,-1
_ r(n + c + 1) _ Y' r(v + c + 1) _
- r(n + 1) (n + 1) zn e "-'
.=1
r(" + 1) z. T(e + 2) Zo,
i.e.
n+1 r(n+1) " - I r (v+c+1)
w -
" - n +c+ 1 n
z-e ""
r(n + c + 2)"';;"
-
r(v + 1)
Z
.'
.=0

For fixed v one finds an. '" - e r(t-; ~_ ~fl n- c- I [1155]. Thus !~~ an.
exists if and only if me > -1. Assume me > -1 and put
U = U = U ='" = _1_ then Z = 1 W = _1_ thus
o 1 2 1+c' " '" l+c'
1
ano + anI + an2 + ... + a"" = 1 + c.
With me = y we have
I A l'
Ir(nr(n+ +c +1) 1) 1< I F(n + 1) I < B -1'-1
n, Ir(n + c + 2) n ,

where A and B are constants, independent of n. Hence


,,-I
lanol + IanI I + ... + lan,n-II < lei ABn- 1' - I J: v1'
.=0

lei AB of1 x1'dx = I lAB


-+ _c__ [1122].
1 +y
The desired necessary and sufficient condition is therefore me > -1.
SO. Set an = cx.. + ip.. , cx,,' P.. real. The relation

yuana97@uw.edu
Pt. III, Solutions 50-55.1 311

(binomial series) and a generalisation of the proof of 175 imply


lim (a 1 + a2 + '" + a,,) n- a = 0, i.e.
"'-+00

lim (IX1
n.,.,oo
+ IX2 + ... + IX,,) n- a = lim (PI
ft.-i"OO
+ P2 + ... + P,,) n- a = o.
Now I 92 can be applied to both power series
IXlt + IX2t2 + ... + IX"t" + "', PIt + P2t2 + ... + p"t" + "',
hence
lim (1 -
1.-+1-0
W(IX1t + IX2f + ... + IX"t" + ...)
= lim (1 -
1.-+1-0
W(PIt + P2t2 + ... + p"t" + ...) = O.
51. Whenever the four subseries consisting of the terms in the four
quadrants (ffiz > 0, 3z 2: 0, etc.) converge, the series converges abso-
lutely.
52. By successive bisection: Assume that all the terms z,,' z,,' ... lie
in the sector -81 ~ arg Z < -82 and that \Z" \ + \Z" \ + ... diverges.
CoItstruct the two subseries with terms in -81 < arg Z < {}l ; {}B and in
{}l ; {}B < arg Z < -82 resp. At least one of the two is divergent.
53. Choose a finite number of terms zm = xm lj- iYm from each of the
successive sectors

(- ;. ;).(- :' :) .... , (-;'2~)''''


The different sectors should contribute different terms so that the
points z'h' Z'h+l' .... Z'h+k that correspond to (- ; . ;) are in this
sector and that
1< X" + X" + ... + X'h+k < 2.
54. [More on this topic: P. Levy: Nouv. AnnIs Math. Ser. 4. Vol. 5.
pp. 506-511 (1905); E. Steinitz: J.reine angew. Math. Vol. 143.pp. 128-
175 (1913).] Let the direction of the positive real axis be the direction
of accumulation (which can always be arranged by multiplication with
a suitable ei "') and z" + z" + ... be the subseries chosen in 53; apply
1134 to the real part. 1133 to the imaginary part.
55. Z = x + iy. Z2 = X2 - y2 + 2ixy are analytic, \.;\ = VX2 + y2
z
and = x - iy however are not analytic.
55.1. Cauchy-Riemann equations.

yuana97@uw.edu
312 Functions of One Complex Variable

SS.2. Obvious from the Cauchy-Riemann equations once the exist-


ence of the second derivatives is known.
SS.3. As SS.1. The result is useful in some problems of mathematical
physics, see e.g. G. P6lya and G. Szeg6: Isoperimetric Inequalities in
Mathematical Physics. Princeton: Princeton University Press 1951,
p.95.
SSA. [SS.2]. Useful in the same manner as SS.3.
SS.5. [SS.3, SSA.]
56. (1) Put I(x + iy) = u + iv, so u is known and

I(x + iy)= u(x, y) + i (1 v:(x, 0) dx + j v;(x, y) dY)


= u(x, y) +i (- j
o•
u;(x, 0) dx +/
0
u:(x, y) dY)
is uniquely determined.
(2) We seek the function I(z) among those functions that are conjugate
complex for conjugate complex values of z:
- (z + z) (1 + zz) z z
I(z) + I(z) = 2u(x,y) = 1 + z'A + Z2 + z2"Z2 = 1+ Z2 + 1 + ;:2 '
hence I(z) = -1z2 ' because the function is uniq uely determined according
+z
to (1) and because rational functions of z are analytic. [Hurwitz-Courant,
p. 47, p. 282.]
57. First solution: Express w in terms of x and y, verify
::~ + ::~ = 0 and determine 3/(z) by integration.
Second solu tion: n - 3 log (z - a) = n + mi log (z - a) is the
angle under which the real axis from a to 00 is seen from z. Thus
I(z) = n + i log (z - a) - [n + i log (z - b)] + ic = ic + i logi =: : '
where c stands for a real constant.
58. Put I(x + iy) = u(x, y) + iv(x, y). We denote the partial deri-
vatives of u and v by ur ' uy , uxx ' ••• Differentiation yields
2
0 2 (U 2
ox + V 2) = 2(Ur2 + v"2 + UUxx + VVxx ) ,
02
oy2 (
U
2
+ V 2) = 2(U~,2 + Vy2 + UUyy + VVyy ) ,
[SS.1, 55.2].
59. Cf. solution 58; in addition note that
(uu" + VV,,)2 + (uu y + VVy )2 = (u 2 + v2) (u; + v;)
= I/(x + iy) 1211'(x + iy) 12.

yuana97@uw.edu
Pt. III. Solutions 55.2-66 313
60. By comparison of similar triangles:

x : ~ = Y : fl = 1: 1 - C. thus x + iy- = ~1 ~ ii '


2
+ y2 =
~2+7J2
x2 (1 _ C)I = 1 _ C - 1,
b ~ ~+~-1
~= Xl + yl + l' fl = x2 + y2 + l' C= Xl + yZ + 1 .
61. We have in turn [60]
. + ~y
P. x
. -_ ~1 +_ iC'7 JP:
,
Eo fl, C, "
P :~, -fl, -"';
,.

therefore

u
+ w. _ ~ - i7J _
- 1 + C-
(x - iy) (1 - C) _ x - iy _ _1_
1 +C +
- xl yl - X iy • +
62. In the case of Mercator's projection: straight lines parallel to the
axes on the unrolled cylinder, generatrices and directrices on the cylinder
itself. This property together with the condition of conformity determines
Mercator's projection completely (d. e.g. E. Goursat: Cours d'analyse
matMmatique, Vol. 2, 3rd Ed. Paris: Gauthier-Villars 1918, p. 58). In
the case of stereographic projection: rays from the origin and concentric
circles with center at the origin.

63• X +. -- ~ +
~y 1
i7J _ coscpei8 _
-~
~ - 1 -smcp
. -
t an (!!!...
2
+~) i8
4 e ,
hence
x + iy = e"+iv.
64. With w = u + iv we have
Izl=e", argz=v.
The curves in question are concentric circles, centred at the origin, and
rays perpendicular to the circles. [62, 63.]
65. Comparison of the real and imaginary parts of
w = u + iv = Z2 = (x + iy)2 yields
U=X2 _ y2, v=2xy.
The curves u = const. and v = const. form two families of hyperbolas.
Since the mapping is cpnformal the images in the z-plane of the straight
lines u = const. and v = const. in the w-plane are orthogonal.
66. Put z = x + iy, w = u + iv; then x = u 2 - v2, Y = 2uv;
elimination of v and u resp. yields the parabolas y2 = 4u2(U 2 - x) as
images of the lines u = const. and the parabolas y2 = 4v 2 (V 2 + x) as

yuana97@uw.edu
314 Functions of One Complex Variable

images of the lines v = const. All these parabolas have the common axis
y = 0 and the focus x = Y = O. Two orthogonal parabolas pass through
every point z, z :oF 0, of the z-plane.
67. Let z = x + iy, W = u + iv. Then
. ei(x+iy) + e-i(x+iy) eY + e- Y e- Y - eY •
U +w= 2 ' thus u =--2--cOS x, v = 2 smx;
u2 v2
The lines x = const. are mapped onto the hyperbolas - 2 - - -;--2-- = 1,
cos x SIn x
.
and the hnes y = const. onto the elliPses-(eY+e
.
---=-)2 + -(--=---)2 = l.
~

Y
~

e Y-e Y
~2- -~-2-

They have the common foci W = -1, W = 1. The two families of curves
are perpendicular to each other (confocal conics).
68. Elimination of v and u respectively from x = u e" cos v and +
y = v + e" sin v leads to
x - u = e" cos (y ~Ve2" - (x - U)2) , y- v= e"'-(y-v)cotv sin v.
The line v = 0 is transformed 'into y = 0 and the line v = n into the
twice covered line segment y = n, - 00 < x ~ -l.
69. The image of the square is bounded by the two rays arg w = 8
and arg w = - 8 and the two circles 1wi = e"+t and 1wi = ea - B • The
area is therefore 8(e2a +2B - e2a - 2B ). The ratio in question is

70.
J J If'{z) J J Isin (x + iy) 12 dx dy.
~ ~ ~ ~

dx dy =
12

Because of
1sin (x + iy) 12 = sin (x + iy) sin (x - iy) = -I cos 2x + i(e2Y + e- 2Y )
the integral is
X2 ~ Xl (e2Y, _ e2y,_ e- 2y , + e-2Y1) _ Y2 ~ YI (sin 2X2 - sin 2x l ).

For Xl = 0, x2 = ; , YI = 0, Y2 = Y we obtain one quarter of the area


of an ellipse with semi-axes
eY + e- Y eY - e- Y
a =---2- , b =~-2-'

71. f' (z) = 2z. On circles with center at the origin, i.e. I z 1 = const.
and on rays from the origin, i.e. arg z = const. •

yuana97@uw.edu
Pt. Ill. Solutions 67 -78 315

°
72. When < a ;;;; n the image of the square is the simply covered
region bounded by the two circles with radii ea and e- a and the two rays
arg w = -a and arg w = a. When a > n either part or all of the region
is covered several times. If a = nn the image is covered exactly n times,
except certain points of the real axis that are covered only n - 1 times.
73. The intersection of the ray arg w = IX with the circle 1wi = 1;
observe the vertical line segment within the disk 1Z 1 < r that intersects
as many of the parallels 3z = IX +
2kn, k = 0, ± 1, ± 2, '" as possible:
it lies on ffiz = 0, i.e. on the image of 1wi = 1 (VIII 16; N(r, a, IX) assumes
its largest possible value for r, IX fixed if log a = 0.)
74. Assume Zl =1= Z2' 1zll < 1, 1z21 < 1. Then
z~ + 2Z2 + 3 - (zi + 2Z1 + 3) = (Z2 - Zl) (Z2 + Zl + 2) =1= 0.
75. If Z = reiD, r> 0, °< {} < n, then w = Rei6 = r2e2iD . R = r2,
o= °
2{}, and so R > 0, < 0 < 2n. If, on the other hand, Rand 0 are
given r, and {} are completely determined.
76. The function in question is schlicht on the closed unit disk
1Z 1 < 1, where furthermore 1wi < 1 [5]. The inverse function is

a + e-ilXw
Z = -~--.-,
1 + ae-"'w
and so the dependence of Z on e-ilXw is of the same nature as the
dependence of w on z. Hence each value w, 1wi < 1, is assumed. The
locus of the points with constant linear enlargement is given by the rela-
tion
rr=-alal l = const.
1 -
z
2
2

°
If a =1= these are certain arcs of circles centred at ~ (reflection of a
a
with respect to the unit circle).
77. [ef. A. Winternitz: Monatsh. Math. Vol. 30, p. 123 (1920).]
According to the hypothesis we have \ : := ~z I= const. along the circle
C, i.e. a and ~ (0 and 00 if a = 0) are the pair of harmonic points common
a
to C and the unit circle. Let Zo denote the center of C and r be its radius,
Zo =1= 0, r < 1 -I Zo I. Then a, 1
a 1< 1, satisfies the quadratic equation

(a-zo)C -zu)=r2 or (lal-lzol)(I~I-lzol)=r2,


arg a = arg zo; IX arbitrary.
78. w = a Z
i where a is constant, la 1 = 1.
Z +
-
t

yuana97@uw.edu
316 Functions of One Complex Variable

79. Write z = 1'eiD • Then w becomes


1 1
-+1' --1'
l' { } ' l' • {}
W =-2-COS -~-2-SIn .

The circles Iz I = 1', 0< l' < 1, are mapped onto confocal ellipses with
1 1

T T ' the common foci are w =


-+1' --1'

semiaxes and +1, W = -1. The


rays {} = const. are transformed into confocal hyperbolas with the same
foci, W = +1, W = -1. The two families of conics are orthogonal to
each other. If Izl = 1, z = i·e, then w = cos{}. Consequently,ifzdescribes
the unit circle then W describes the segment -1 < w ~ 1 twice.
SO. The function w = kz + h~ , 0 < k1'l < k1'2 < 1 maps the annulus
in question onto the region bounded by the two ellipses with foci w = - 2
and w = 2 and major semiaxes k1'l + -h1 and k1'2 + -h1 respectively.
1'1 1'2
Put

k=a1 - v4=! = al-V~.


1'1 1'a

1
z+-
81. The function w =- ~ ,z= ,e iD maps the upper half of the
unit disk into the upper half-plane [79]:
1 1
-+1' --1'

w= - l'
- 2- +'~ -l' 2 - SIn• {} ; '\Jw
cos { } ~ > 0 f or 0 < {} < n.

The linear enlargement

1:';11=t
for the points z = x + iy for which
Ix2 - y2 - 1 + 2ixYl2 = IX2 - y2 + 2ixyl2, i.e. x2 - y2 = t.
They define an equilateral hyperbola which intersects the real axis at
the points z = ± V~ . The rotation becomes
~-1
Z2 :rr.
arg --2- =±2

when 91 ~
Z
= 1, i.e. ,2 = cos 2{}. These points lie on the lemniscate

/z -V~ liz +V~/ = ~ .

yuana97@uw.edu
Pt. III, Solutions 79-89 317
1
z+-
82. The auxiliary function, = - ~ maps the region in question
onto the upper half-plane 3' > 0 [81]. The origin z = 0 corresponds to
,= 00, z = i to , = 0 and z = ± 1 to , = + 1. The function w = :2
transforms the upper half-plane 3' > 0, note 75, into the w-plane cut
open along the non-negative real axis. The point' = 00 corresponds to
w = 0, ,= 0 to w = 00, ,= ±1 to w = 1. The mapping function
having the required properties is therefore given by

W = (- - 1
2)2 = (1 + Z2)2 •
4Z2

Z +-Z

The images of z = ± 1 are both at w = 1 but on different sides of the cut.


83. arg w = R 2~ (arg z - ex), i.e. 0 < arg w < 2n.
,.,-IX
"
84. The first auxiliary function, = (e- i " z)Il-e< maps the circular
sector onto the upper half of the disk 1'1
< 1. The second auxiliary
C +~-
function s = - ___
2
C [81] maps the disk onto the upper half-plane
3s > O. Apply 78.
85. Follows from

u - iv : [rp(x, y) + i1jJ(x, y)] = ~:-


= uX t oy
[rp(x, y) + i1jJ(x, y)]
by separation of real and imaginary parts.
86. They are the images in the z-plane of the lines 'lftl = const. and
31 = const. parallel to the axes in the I-plane under the conformal mapp-
ing I = I(z).
87. Follows from 85 by virtue of the Cauchy-Riemann differential
equation ~: + ~.~ = O. Also the function 1jJ(x, y) satisfies Laplace's
differential equation.
88. With u cos T + v sin T = 'Ift(u - iv) eiT the integral in question
becomes
= '1ft fiz eiT ds = '1ft fiz dz= 'Ift[f(Z2) - I(Zl)]
L L

where dz denotes the directed line element with modulus ds and argu-
ment T.
89. u sin T - v cos T = Z'\'(u - iv) eiT ; d. 88.

yuana97@uw.edu
318 Functions of One Complex Variable

90. The third equation is identical with the second Cauchy-Riemann


differential equation. We get the first two by differentiation, keeping
the first Cauchy-Riemann differential equation in mind:
lOp
- - = - u8x
--v-
au
= - u8x- - v8y'
-
Ov au 8u
f! 8x 8x

l8p 8u Ov Ov 8v
--= -u--v-= -u--v-.
f!8y 8y 8y 8x 8y

91. The vector w= -.!.. ei8 forms


I'
the angle l} with the positive real
axis, its modulus is -.!... The functions in question are up to a constant
I'

I(z) = log z, 9'(x, y) = log r = log V + y2,


XS tp(x, y) = {} = arctan L.
x

The level lines are concentric circles around the origin, the stream lines
are rays perpendicular to these circles.
92.
9's - 9'! = logrs -logr! = log I'z,
1'1
tp' - tp = 231;,

4n
-'- '"
1 ( ,
- "')
----=--.
qiz - qil 2 log 1'2
1'1

93. The amplitude of w is equal to {} +; , the modulus is : .


Furthermore we have (up to an additive constant)

I(z) = - i log z, 9'(x, y) = {} = arctan Lx ,

tp (x, y) = - log r = -log VX2 + y2.


The level and stream lines are the stream and level lines resp. of 91.
The potential 9' is infinitely multi valued.
94. According to 93 the field of force is described (up to a real
constant factor) by

W
2i
= ZI-=t' t h us I()
z = '1 z - l
~ og z + l ' tp = log z + IZ-ll 1 ' - 9' = arg z -l
z+ 1.

The level lines are circles through the points z -" -1 and z = +1, the
stream lines are circles too, namely the ones with respect to which the
points z = -1 and z = +1 are mirror images of each other (circles of
Apollonius).

yuana97@uw.edu
Pt. III, Solutions 90-97 319
95. We are looking for the points z for which [93]
iAI i'-2 iAn
--------···---=0,
Z - zl Z - z2 Z - z,.
i.e.
Al A2 A,.
- - + -Z-- +
Z - zl z2
" ' + -Z-- = z,.
0;

the positive numbers AI' ~, ... , A.,. are proportional to the intensities of
the currents. Cf. 31, in particular the first solution given.
96. The vector field is generated by an analytic function I(z) for
which fRl = const. on the given ellipses. The function [80]

z = kZ + k~' 2kZ = z- VZ2 - 4

maps the region bounded by the two ellipses onto the annulus'l < IZ 1< '2'
The semi-axes and the radii are related by

provided that a l > a2 and the positive roots are chosen. The problem is
now reduced to 91 : w = ! defines a vector field in the Z-plane for which
the concentric circles around the origin Z = 0 are level lines, i.e.
fR J~ = const. Consequently the same is true for Z = const. I I
J dZ ~dz-
dz Z -
-J V dz
z2-4
along the given ellipses in the i-plane, i.e.

w= - V 1
zl-4
, I(z) = log (z - VZ2 - 4) •

The stream lines are confocal hyperbolas, the level lines confocal ellipses
with foci - 2 and 2. The relations between the potentials are

The capacity is
1

97. Put [93]


i
w = -""""i; "PI = -log a, "P2 = -log b, fill = (X, fIl2 = p.

yuana97@uw.edu
320 Functions of One Complex Variable

The resistance is (the sign is not important) equal to


fJ-1X
logb-loga·

98. According to 85 the unit circle Iz I = 1 is a stream line. If the


constant value of the conjugate potential along the unit circle and on
the real axis inside the vector field is assumed to be zero the function
I = I(z) transforms the unit circle Iz I = 1 into a segment of the real axis.
In view of 79 set
I(z) = k (z + !) + ko,
Since w= 1 for z = 00 we have k = 1, i.e.

w=1-~.
Z2

w
99. The stagnation points are z = ± 1; assumes the same values at
each pair of points that are symmetric with respect to the origin. There-
fore the resultant total pressure on the pillar vanishes [d. 90]. The
pressure is minimal or maximal when 11 - :21 is maximal or minimal
resp., i.e. for z = +i and z = ± 1 resp. Rotation of all the vectors through
90° generates a field of force which admits the following interpretation:
A homogeneous electrostatic field is disturbed by a circular, insulated
wire perpendicular to the direction of the field. (The most simple ex-
ample of electrostatic influence.)
100. [G. Kirchhoff: Vorlesungen tiber Mechanik, 4th Ed. 1897,
pp. 303-307; A. Sommerfeld: Mechanics of Deformable Bodies. New
York: Academic Press 1950, pp.215-217.] Supplementary continuity
condition: the boundary of the wake (the stagnant water) stretches
to infinity where Iwi = 1; since Iw I is constant on the entire boundary
w
it has to be equal to 1. The direction of is known along the boundary
segments AB, AD (barrier) and the magnitude of wis known along the-
w
boundary lines BC, DC (along the wake), is completely known at the
four points A, B, C, D. Noticing th"at either the direction or the magni-
w
tude of is constant on the respective parts of the boundary we find a
half-circle in the w-plane as image of the boundary of the field of flow.
We fix the constant contained in I [po 123] so that 1= 0 corresponds
to the stagnation point z = o. Then the left and right "banks" of the
positive real axis of the I-plane correspond to the streamlines ABC and
Abc, respectively. The I-plane cut along the positive real axis corres-
ponds to the whole field of flow; it is not possible that only a subregion

yuana97@uw.edu
Pt. III, Solutions 98-101 321
of the I-plane so cut should correspond to the field of flow because
W = :: C'V i as z -+ 00, thus I C'V iz. The same point, but associated with
the left or right bank of the cut in the I"plane, corresponds to the two
points Band D, respectively, because of symmetry. Note that a dilata-
tion of the I-plane with w
retained, causes the same dilatation of the
z-plane because dl = w dz. Now we dilate the I-plane so that I = 1
becomes the image of z = ±l. In this way we attribute a numerical
value to l. - If a one to one relationship can be established between the
corresponding parts of the z-, w-
and I-planes we can find out [d. 188J
whether the interior is to the left or to the right as one moves in the
direction given by ABCDA (see the last line given in the table). In the
following diagrams the points in the different planes corresponding to A
are also called A; B, C, D are used similarly.

z iii w I
A 0 0 0 0
B 1 1 1 1
C 00 -i i 00
D -1 -1 -1 1
region lies to the left right left left

C
Velocity plane (iV-plane)
CD
o
(w-plane)
A 8
A
o
8 e=en
------0---0---
Potential plane (f-plane)

101. According to 82 we have


4w 2 1 1 + V1 - I
1= (f+-W2)2' i.e. w VI
where Vl- I becomes 1 as I = O. Observe the continuous change of the
value of w on the two banks of the cut in the I-plane. Hence

z= J dl
o w
= 2 Vi + VI V1 - I + arcsin VI.
z= x + iy = 2 VI + ~ - i [I F ~- log (VI + VI - 1)].

The first formula for z is to be used when 0 < I < 1, the second when
I > 1 (positive roots) ; it furnishes the boundary of the wake:

yuana97@uw.edu
322 Functions of One Complex Variable

z = l = 2 + ; for I = 1; X N 2 VI, y N -I, thus the width of the wake


is 2x N 4 VfYTat a great distance from the barrier.
102. If the pressure is P = c - t 1W 12 at the point z [90] it is in
particular equal to PI = C - t on the boundary and therefore every-
where in the wake. The total pressure desired is
+1 +1 1
J
-I
(P - PI) dz =
-I
J t (1 - 1W 12) dz = J (1 -
0
w2) dz

= J (1 -
I
w2) -
dJ
= J 4 V1 -
I -- V-
I d 1= 'Te.
o W 0

103. Put z = rei/}. Since the sign of the angular velocity is positive, {}
is increasing, i.e. z describes the circle in the positive sense. We have
~~ = iz and the velocity vector in question is
dJ(z) _ dJ(z) dz - . 1'( )
dD - dz dD - ~z z.
104. Let w denote the angle through which the vector w (radius
vector) has to be rotated in the positive sense to fall into the direction
of the vector iz/, (z) (tangential vector, 103). Then the distance in question
is given by
c>< iz!'(z)
. ;;j 7(z) -- ffiz!, (z) J(z)
I/(z) 1 SIll W = I/(z) 1 I iz!,(z) I = !z!'(-;rr-'
J(z)
105. The amplitude of the vector in question is 3 log I(z). The angular
velocity is therefore, with z = rei/},
~ Q< I I() = Q< d logJ(z) dz = Q<!'(z) . = roj'(z)
dD~ og z ~ dz dD ~ J(z) U U~ J(z) •

106. The curvature is ~ = ~~ , where de denotes the change in the


direction of the velocity vector of I(z), i.e. according to 103, the change
of 3 log iz/,(z); dS denotes the line element of the curve described by
I(z). According to 103 we have ~! = 1iz/,(z) I; consequently, with z = rei/},

de -103 log iz!'(z) 3 d log z!J1 dz


1 dD dz d{}
(! dS - liz!'(z) I Iz!'(z) I
dO
The curvature is positive or negative according as the velocity vector
turns in the positive or the negative sense.

yuana97@uw.edu
Pt. III, Solutions 102 -113 323
107.

concave
I convex

to the left + -
to the right -
+

If w = zn + a, n ~ 0, then ~(! = sgnrn n . The four possibilities are already


shown by the special cases r = 1, la I> 1, n = 1, or n = -1, where
w = 0 is chosen as reference point.
108. [106, 107.J
109. The angular velocity of the vector w = t(z) is always positive.
[105.J
110. The proposition follows from 108 and 109 or from the following
consideration: the angle between dw, w = t(z), and the positive real
axis is given by the argument of iz/,(z) [103J. Convexity means that this
argument is always changing in the same sense, which property coincides
with the definition of the image of the circle under w = z/, (z) as being star-
shaped.
111. [Thekla Lukacs.J Let a and b be two points with respect to
which the image of the circle Iz I = r is star-shaped. Then

j'(z) j'(z)
ffiz j(z) -=----;; > 0, ffiz j(z) _ b > 0,
i.e.
ffiz/, (z) t(z) - a > 0, ffiz/, (z) t(z) - b > o.
Hence for A > 0, p, > 0, A + p, = 1

ffiz/, (z) t(z) - (Aa + p,b) > 0, thus ffiz j(z)


j'(z)
_ (Aa + pb) >0.
The proposition can also be easily proved by elementary geometric
reasoning.
112. h(cp) = ffiaeiq> = lal cos (cp - ex).
113. (1) The velocity vector [103J iz/,(z) forms the angle cp + ;
with the positive real axis: cp = arg z/' (z) = Slog z/' (z).
(2)
h(m) = ffizj'(z) j(z)
r Izj'(z) I -
[104.]; note the sign.

yuana97@uw.edu
324 Functions of One Complex Variable

114.
iO
z e2 {}
cp =3Iog~-
l+z
=3Iog--=-
2 {} 2'
[113J.
cos 2"

115. For : <cp < 3: the support function is identical with the one
of the point ni [112J. Outside this sector we find on the basis of 113
iO
2z 2e 2 {}
cp = 3 log V-==-
1+z2
= 3 log V
2cos{} 2"'

116. cp = arg zj'(z) = arg~-,


l~w
h(m) T
= I~-I
l~w
91(1 - w) if the line
of support is a tangent to the boundary; h(cp) = cos cp if the line of
support passes through the corner w = 1.
117. ~
211: 0
J ei(k-I)O df) =
2"
~ J ei"odf) =
2"

211: 0
0 or 1 depending on whether

the integer n = k - l is different from zero or not.


118. Follows from the expansion
l(reiO ) = ao + alreiO + a2r2e2iO + ... + a"r"ei"o + ...
by integration and application of 117.
119. Apply 118 to log I(z) and consider the real part.
120. [Jensen's Formula, d. 175.J The geometric mean of the single
factor 1z - z.1 can be computed with the help of II 52, the geometric
mean of 1/*(z) 1 with the help of 119.
121. For I(z) =F Oin Izl < rthe means are equal: g(r) =@(r) = 1/(0) 1
[119J. If I(z) is the prod.uct I(z) = 11(z) 12(Z) of two functions 11(z) and
12(Z) that are regular for 1 z 1 s r, its mean is equal to the product of
the corresponding means. It is therefore sufficient to examine the parti-
cular case I(z) = (z - zo), 1Zo 1 < r. According to II 52
1 2"
- f logl'';o -z,ldO
@(r) = e2" 0 = max (r, 1zo)1 = r,
,.211
----.-
1
f J logle" "0 -z,lededo I J
2 '
, 0
max(log{>,loglz,\)ed" log,-
1
~
2
+ I2" , I',
g(r) = e'" 0 0 =e =e
122. [Cf. M. A. Parseval: Mem. par. divers savants Vol. 1, pp. 639-
648 (1805); A. Gutzmer: Math. Ann. Vol. 32, pp. 596-600 (1888).J

yuana97@uw.edu
Pt. III, Solutions 114-127 325

123. We set P(z) = Xo + Xlz + X2 Z2 + ... + x"z" with arbitrary


complex coefficients xO' Xl' ... , X". Then

This expression becomes a minimum if and only if the first n + 1 terms


vanish.
124. [As to 124-127 d. L. Bieberbach: Rend. Circ. Mat. Palermo
Vol. 38, pp. 98-112 (1914); Sber. Berlin Math. Ges. 1916, pp. 940-955
and T. Carleman: Math. Z. Vol. 1, pp. 208-212 (1918).] Particular case
of 125; replace rand R by 0 and r respectively.
125. With w = I(x + iy) = u + iv the area in question becomes

dudv = If
,1~x'+y'~R'
I o(x,
O(U~ldxd
y) Y= JJ'IO(U~ Ie de dO..
,
R

0
2"

o(x, y)

We have (d. p. 117)

o(u,--':'l = ou OV _ OU OV = (OU)2 + (OV)2 = Io(u + iv) 12 = II'(z) 12


o(x, y) ox oy oy ox ox ox ox '

!t rC~'"
hence [d. 122]

F = II'(ee iD ) 12 e de dO. = 2n n2la" 12 e2" - ] ) de

'"
= n 1: n la,,12 (R2" - r2").
n=-oo

126. Particular case of 127. As to the orientation d. 188 or 190.-


What does the result suggest when c = O? [124.]
127. The area is given as the sum of elementary triangles provided
with a sign, bounded by the arcs of the curve L and radii. The sign is

I
positive or negative depending on whether the point w = 0 is to the left
or to the right of the bounding oriented arc. It coincides with the sign of
sin OJ introduced in solution 104. Thus the area is

= t ii
o
izl' (z) 11/(z) 1 sin OJ dO. = t }" 'iRzl' (z) I(z) d1?
0 ~
2n 00 00
z = re' .
=1'iRf 1: kakr"eik8 1: a/e- iI8 do.,
o k=-oo 1=-",

yuana97@uw.edu
326 Functions of One Complex Variable

128. According to 124 we have

[122].

129. [Cf. K. Lowner and Ph. Frank: Math. Z. Vol. 3, p. 84 (1919).]


The density function of the mass distribution in question is proportional
to 19/(Z} 1-1,
because J Itp'(z} 1-1
Idw lover an arbitrary arc of L gives the
length of the corresponding arc of the circle IZ I = r. Therefore
2,. 2,.
EJIll" (z) 1-1 Idw I = J tp(z) III" (Z) 1-11 dwl, i.e. EJ r d{} = J tp(re;') r d{}.
L L 0 0

r
130. The volume is given by

o
j 0
I/(e i ') 12 e de d{} = 231: j (i I
0 n=O
an 12 e2n +l) de,
where Z = ei'.
131. [J. L. W. V. Jensen: Acta Math. Vol. 36, p. 195 (1912).]

cos y = [1 + (!~2 + (~rJ -i, 2


tan y = (!:Y + (~y .
The Cauchy-Riemann differential equations together with the relation
C= u 2 + v2 imply

! tan y 2 = ( u: +v :r + ( ~ + ~Yu v = (u 2 + v2) [(::t + (!:YJ .


132. Let Zo be a point with horizontal tangent plane. According to
131 we have either I(zo) = 0 or I(zo) =l= 0, I'(zo) = O. As to the first case
one has to note that an analytic function has only isolated zeros. In the
second case we consider I'(zo) = I"(zo) = ... = 1(/-1) (zo) = 0, 1(/)(zo) =l= 0
[I ~ 2, saddle point of order I - 1], thus

+T
j(/)(z)
I(zo + h) = I(zo} hi + ... ,
C= I/(zo + h} 12 = I/(zo) 12 + I(zo/(I;~ZO) J( + l(zoll;~Zo) hi + ...
= I/(zo) 12 + A Ihll cos (kp-IX) +"',
where h = Ihi eUp; A, IX are real constants, A ~ 0, determined by I(zo)
and t")(zo)' The 21 values of II' that correspond to the 21 directions of the
branches joining at Zo are

k = 1, 2, ... , 21.

yuana97@uw.edu
Pt. III, Solutions 128-139 327
The sign of A cos (kp - IX) is alternately positive and negative between
these directions.
133. Suppose that the polynomial in question is given by
ao(z - IX I ) (z - IX 2) ... (z - IX,,), lXI' IX 2, ... , IX" real. Then

C= lao 12 n" [(x - IX.)2 + y2] > lao 12 n" (x- IX.)2, iJ2C
ay2> o.
• =1 .=1

134. 122 implies

It(zo) 12 + If'i~o) r + If'~~o) r + ... + Itt:~Zo) r


1'2 1" 1'2" + ...
2,.
= 2171: f I/(zo + re''') 12 dO. < M2,
o
i.e. I/(zo) I <M. In the case of equality we must have
!'(zo) = t"(zo) = ... = I(")(zo) = •.. = 0
which means I(z) = I(zo), constant.
135. The continuous function I/(z) I has to attain its maximum in the
closed domain ~. According to 134I/(z) I cannot assume its largest value
at an interior point Zo of ~.
136. The piece cut off from the modular graph by an arbitrary
cylinder perpendicular to the x, y-plane has its highest point on the
boundary unless the modular graph is parallel to the x, y-plane. There
are no "peaks" in an "analytic landscape".
137. Geometric interpretation of the proposition that a polynomial
(z - Zl) (z - Z2) ... (z - z,,) always assumes its maximum on the bound-
ary of any domain [135].
138. The function t~Z) is regular in ~ [135J.
139. Assume Iz.1 < R, 'V = 1, 2, ... , n. The function
1 " -----------,----
I(z) = R V(R2 - ZlZ) (R2 - Z2z) ••• (R2 - z"z)

has a regular branch for Iz I <R, furthermore I(z) =1= 0 for Iz I < R.
According to ~35 and 138 1/(0) 1= R must lie between the maximum
and the minimum of It(z) Ion Izl = R. We have for Izl = R [5]

I/(z) 1= V'-IZ-Z---:-ll-:-lz--Z-2-1·-··-Iz-- z"l·


The only exception occurs in the case I(z) = const., i.e. if
Zl = z2 = ... = z" = o.

yuana97@uw.edu
328 Functions of One Complex Variable

140. We find for \z\ = R


1 (R2 - zlz) + (R2 - z2z) + ... + (R 2 - Z"Z)/
R<max 1------------------------------
- R n

(-
1 IR2 - zlzl + IR2 - z2z1 + ... + IR2 - ZnZI)
<max
- R n
Iz - z]! + Iz - z2: + ... + Iz - z,,1
=max-------------------- [5].
n
The inequality becomes an equality only if ZI + Z2 + ... + zn = 0 and
all the z:s have the same argument, i.e. z. = 0, v = 1, 2, ... , n. Notice
the particular case n = 4, Zl = 1, Z2 = i, za = -1, Z4 = -i, R> 1.
The arithmetic mean of the profections of the distances in question into
the diameter through P is already equal to R.
141. The function
1 1
R2 - zlz + R2 - + ... + R2 -
Z2Z Zn Z

is regular for \z\ < R. Hence [135] for \z\ = R


. _ _n_ _ _ . n < n < R
mIll n 1 -mIn n
1: ---- 1: -2----
R = In
max 1: -2-=-
R \ = .
.=llz - z.1 v=IIR -z.zl .=1 R -z.z
.2"
'J-
.4"
1---.

Note the special case n = 3, ZI = eU}, = 1, z2 = e;{}' = e 3 ,zs = e;{}' = e 3 ,


R > 5. Then we have for z = Rei {}, v = 1, 2, 3,
1 1 00 Pk[cos (I) - I).)]
Iz-z.1 =VR2+1-2RCOS(I)-I).)=k~ RHI

where Pk(x) denotes the k-th Legendre polynomial [VI, § 11];


Po(cost'l) = 1, PI (cos t'I) = cost'l, P 2 (cost'l) ="4
1
+ "43 cos 2{}. Hence
[VI91J

00 Pk[cos (I) - I)I)J + Pk[cos (I) - 1)2)J + Pk[cos (I) - I)s)J


+~ 3Rk+!

~ ~ _ ;_1__ ~ R-5 >~


> R + 4R3 "'-' RHI -
k=S
R + 4R3(R - 1) = R •

142. If f(z) =F 0 everywhere inside the level line the absolute value
\ f(z) \ attains its maximum and its minimum on the boundary according
to 135 and 138. This implies that \f(z) \ must be constant in the interior,

yuana97@uw.edu
Pt. III, Solutions 140-146 329

and so !(z) = const. Geometric interpretation: Since there is no peak


inside a closed level line on the modular graph there must be at least
one pit unless the modular surface is a horizontal plane.
143. At least one zero must lie inside any closed line [142] along
which the absolute value of the polynomial !z - Zl) (z - Z2) ••• (z - z,,)
is constant. There are only n zeros.
144. The theorem is not valid for !(z) = const. Thus we may assume
!(zo) =F o. If there is a saddle point on the circle Iz I = r the projection
of at least one of the sectors with points above the saddle point mentioned
in 132 protrudes into the interior of Izl < r. Therefore Zo cannot be a
saddle point, i.e. !:(zo) =F O.
We put !(z) = wand consider the image of the circle Izl = r in the
w-plane. The point farthest from w = 0 is Wo = !(zo), the curve has a
definite tangent at Wo because /,(zo) =F O. The tangent, that is the vector
izol'(zo) [103J, is perpendicular to the vector Wo = !(zo) (obvious for
geometric reasons). Hence iZf~~;o) is purely imaginary. In the neigh-
bourhood of Wo the side of the image curve turned towards the o~igin
corresponds, according to the hypothesis, to the side of the circle. turned
towards the origin. Therefore iZ~~:;o) must be positive imaginary.
145. [Cf. A. Pringsheim: Sber. bayer. Akad. Wiss. 1920, p.145;
1921, p. 255.]
.. a(w' - w·- I ) " w t - w- t . n .
21; • .-1 = 2 1; t -I = 2m tan n~ 2;7n.
,=1 a(w +w J ,=1 W +w
146. [A. Pringsheim, I.c. 145.] We define

C(k) =

I
- k
k ~ 1 (.(-1 + .(=~z. + ... + .()
+1 1 ('(~f Z;l + ~~iZ.-2 + ... + Z.-!l.(+l)
for k = 0, 1, 2, .. , ,

fork = -2,-3, ... ;


then
~+l _ zk+l
Cik)(Zl - zo} + C~k)(Z2 - Zl) + ... + C~)(z" - Z"_I) = -~r- = O.

The total length of L is called t, R is the largest, r the smallest distance


of L from the origin z = 0; assume R > 1. The quantity

is the length of an inscribed polygon, thus < t. The points Zl' Z2' ... , z"
can be chosen so that for a given 15. 15 > 0, and sufficiently large n,

yuana97@uw.edu
330 Functions of One Complex Variable

Iz. - Z._11 < 15, 'JI = 1, 2, ... , n. Assume k > 0, then

/ (C~k) -z:)(z.-z._l) 1= k ! 11 (1,-1-1,) +z,(z!=t-z!-l) +" 'liz, -z._11


~ k ~ 1 [k + (k - 1) + .. , + 1 + 0] Rk - 1 /z. - z._112 ,
consequently

I.~ z!(z. - Z,_l) I < : Rk-l.~ Iz. -z._112 < : R k- 1 Z d.

The case R ~ 1 is similar. If k < -2 we have to use r instead of R for


the estimate.
147. [Cf. G. N. Watson: Complex Integration and Cauchy's Theo-
rem. Cambro Math. Tracts No. 15, p. 66 (1914).] The interior of the given
ellipse is described by the inequality
x 2 - xy + y2 + X + y < O.
i i i .
Only one, Zo = - Vi1 - V2 ,of the four poles, V"2 +
V2 ' of the mte- ±
grand lies inside the ellipse. Therefore
J. ~~ = 2:ni = :n_ (-1 + i).
'f1+z4 4z~ 2V2
148.
31

4i f2"

x2
x df}
+ sin2 f}
= f 31

ix df}_
sin 2 f} + x 2
= f -.!:!!__ = J.
31

sin f} - ix 'f Z2
2dz
+ 2zx-1 '
o -31 -31

with z = eifJ • The integral is taken along the circle Iz I = 1 which includes
only the pole at Zo = -x + V1+~. Hence
4:ni 2:ni
Zo + zOl =V1 + ;2'
149. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 24, p.84
(1916). Solved by J. Mahrenholz: Arch_Math. Phys. Ser. 3, Vol. 26, p. 66
(1917).]

o
,
r
231
(1+ 2 COS1~)n B,nfJ
1 - r - 2r cos f)
.
df) =.!. J. (1 + z + z2)n
i 'f (1 - r) z - r(l + Z2)
dz
,

the integral is taken along the circle Iz I = 1. If r ~ 0, -1 < r < 31 '


we have /: - 1/ > 2, therefore the quadratic equation

yuana97@uw.edu
Pt. III, Solutions 147 -152 331
(1 - r) z - r(1 + Z2) = 0 has its two roots separated by the unit circle.
Let e be the root inside the unit circle, i.e. IeI < 1,
(1 - r) z - r(1 + Z2) = -r(z - e) (z - ~). We obtain

2n; (fl)n.
r(~-fl) r '
f!
.
IS real, f! =
1 - r -
2r
V1 - 2r - 3r2
.
150. With the notation z = x + iy the integral can be written
~ rl Z d? - z d'i + zz(z dz - z dZ) = ~ rl (1 + r) z dz - (1 + Z2p dZ
2i'f 1+z2+z2+z2r 2i 'f (1 + Z2) (1 + Z2)

=3f 1~:2 =32ni=2n.


The foci of the ellipse are the poles of the integrand.
151. We have for w > 0
n

J x'-le- x dx + wSi f
ro ~ ~ ro
eisO-roeiO df} - e2 J x'-le- ix dx = O.
o 0 0

Let w converge to + 00. Then the first integral converges to r(s) and
the third to the integral in question. The modulus of the second integral is
" n "
"2 ~-B 2"
< A Je-rocosO df} = A J e-rocosO df} + A Je-rocosO df} « An; e- rosins +As,
o 0 "
--8
2
2
~ I~l'sl n; 1
where A = e2 ,0 < e < 2' By setting e = k ' 9ls < k < 1, we see
Q)

that the second term converges to O.


152. The integral is
1 co ~-2
= -
n 0
xn J
sin x dx.

We conclude from 151 that for real s, 0 < s < 1,

J x'-l sin x dx = r
00

(s) sin ~s .
o
This integral converges for -1 < 9ls < 1 and the right hand side is
regular for the same s values. Hence the formula holds for -1 < 9ls < 1.

yuana97@uw.edu
332 Functions of One Complex Variable

153. [Cf. Correspondance d'Hermite et de Stieltjes, Vol. 2. Paris:


Gauthier-Villars 1905, p.337; cf. G. H. Hardy: Mess. Math. Vol. 46,
pp. 175-182 (1917).]
00. 1 _i"+l)" "+!.-1
J
e-""""x" dx = - e
c
/' e-'z fA dz.
p.
J
The last integral is taken along the ray arg z = lX and it is equal to
+00
r(n +p. 1),
,,+1 1
J e-"x -;;-- dx =
o
because the integrand, with z = (!e,{J, 0 < D< lX, converges to 0 as :
tends to 0; the convergence is uniform in D. For lX = p:lt, 0 < I-' < t we
get the function
e-"Pcos"n sin (x'" sin /-l7'C) ,

all the Stieltjes moments of which vanish without the function vanishing
itself (no contradiction to II 138, II 139). According to E. Borel [LeQons
sur les series divergentes. Paris: Gauthier-Villars 1901, pp. 73-75; cf.
also G. P6lya: Astronom. Nachr. Vol. 208, p. 185 (19J.9)] a similar state-
ment can not be true for a function I (x) , I/(x) I < e- k Y-;-, k> 0, k = const.
Our formula shows that V';-
cannot be replaced in this theorem of Borel
by a lower power, x"', I-' < t, of x. H. Hamburger has proved [Math. Z.
Vol. 4, pp. 209-211 (1919)J that V; cannot even be replaced by (1~:X)2
by showing that

o
j exp (- nY;-- log
lJog;r)'
x) sin (Y;-IO~~++ n) x"dx = 0,
+ re' (log x)' n'
n = 0, 1, 2, ...

154.
+00 +00 +00
x"'+1 J e- 1Pcos xt dt = xl' J sin xt . ~-le-1I' dt =
e,,··-6, dz, 3 J
o 0 0

where the variable of integration is z = x"'tI', 1-'-1 = v and x- p = ~. Rotate


the line of integration through a small positive angle, put ~ = 0; then
rotate the line of integration in the positive sense until arg z = p.; .
155. We replace in the integral

j'
a+,T
1.
-2 e'"
-,-ds, T>a,
ret s
a-,T

the rectilinear path of integration by the semicircle over the segment


(a - iT, a + iT) to the right or to the left according as lX < 0 or lX > o.

yuana97@uw.edu
Pt. III, Solutions 153-157 333

In the first case the integral does not change and is absolutely smaller
1 e- e-
than 231 T2 nT = 2T ; in the second case it decreases by IX (residue at

the pole s = 0) and the new integral is absolutely < 2~ (T e":- a)1 nT.
Now let T increase to + 00.
156. [The case A = 1 + e-1 is due to H. Weyl.] We have

lI(t) = n!
r
--
t"
for n < t < n + 1,
thus (the interchange of summation and integration can be justified
by various arguments)

u !!
+00 +00 sin-e2

! f I :
co fI. in"
-IA -;t" ~ _I_e_ dt d .
e e ~ nl U
,,=-00 1=0 ,,=0

=~J
2m
J e<"-
+00 +00 .

u
1 el(-A.-i"+ei ,,) dtdu
-00 0

1
= 2ni
J
+00

u(,l
.
e'" -
+ iu _ i") duo
1
-00

This integral is equal to the sum of the residues in the upper half-plane
~-1 .
because the integral of . along the half-circle u = re,IJ,
u(,l + iu - e'U)
o ::;;: {} < n, converges to 0 as l' tends to infinity. The only pole in the
upper half-plane is u = iz [196], the corresponding residue is -.!...
z
(Cf.
215, IV 55.)
157. [Dirichlet: J. reine angew. Math. Vol. 17, p. 35 (1837); Mehler:
Math. Ann. Vol. 5, p.141 (1872).] Assume -1 < x < I, x = cos{},
0< {} < n. Then [cf. G. Szego: Orthogonal Pelynomials, 3rd Ed. 1967,
pp.86-90]
p .. (cos{}) =-2'
1 :P V z .. dz,
m 1-2zcos{}+zl
where the (positively oriented) path of integration encloses the two
singular points eiIJ and e-- iIJ • We may contract the path to a straight line

yuana97@uw.edu
334 Functions of One Complex Variable

segment from i 6 to e- i6 (Laplace's formula) or to the arc -D- ~ arg z < D-


of the unit circle (Dirichlet-Mehler formula). (In both cases we should
proceed carefully because the integrand becomes infinite at the endpoints,
although only of the order !.) When z goes around the singular point
the integrand changes only its sign. Thus

..Q.)
P (cos·v 2
=-.
J'
1
(cos 01} + i~sinol})"i sin oI}fh
" 2m
-1
VI - 2 (cos 01) + i~ sinol}) cosol} + (cos 01) + i~ sin 0I})1

2 J6 i'" . ie" de
= 2ni -6 VI - 2e" cos 01} + l" .
We obtain the third expression either by contracting the path of inte-
gration to the arc D- ~ arg z < 231: - D- of the unit circle or by changing
the variable and replacing D- by n - D- in the second expression.
[P,,(- cosD-) = (-1)" P,,(cosD-).]
158. If z is real and negative, any two pieces of L that are symmetric
to each other with respectto the real axis contribute conjugate complex
values. We denote by L .. the negatively oriented boundary of the half-
strip mz > lX, -31: <,3z < 31: (in particular Lo = L). If z is not in ~ and
if lX > 0 the integral along each of the L..'s has the same value. By letting
lX increase to + 00 we successively extend E(z) analytically over the
entire plane.
159. The integral along L = Lo has the same value as the integral
along L .. , lX ~ o. Since r+in = r-
in = e- r the contributions of the

horizontal parts of L .. cancel each other. The vertical segment of L ..


supplies
1
2n
J" e...+iy d"y.
-"
This value is independent of lX, hence =1, as can be seen for lX -+ - 00.

160. (1) Let z be outside ~. We have [169]


E(z) = -..!..+..!..~
g gB2m,
[(-c +~)e·CdC.
C-z
L
Evaluate the integral not along L but along L', an inner parallel curve
to L at the distance 15 (boundary of the region ffiz> 15, -31: + 15 <,3z < n -15)
whereby 0 < 15 < ; . Note that the real integral
j e e-.eCOSd de
2
o
converges.

yuana97@uw.edu
Pt. III. Solutions 158-165 335

(2) Let z be in the rectangle -1 < ffiz < O. -n < 3z < n. Accord-
ing to the residue theorem we have

for all z in the rectangle and then. by virtue of analytic continuation.


in the half-strip ~. Instead of L_l choose the outer parallel curve at the
distance 6. 0 < 6 < ; . as path of integration.
161. The numerator is
n 2" 2ni
= 2ni 17 (-1)"-- ! ( _ ) , = -, .
0=0 v n v. n.
The denominator is

f
n
22ncos19 2n
= df)e-.:>- [II 217].
-n lll2ni19 _ vi n!
0=1
Explanation: The principal parts of both integrals stem from an arc
centred at z = 2n whose length is of order V;. Along it the argument of
dz is close to ; and the argument of In(z) is nearly O. [43.]
162. The number of zeros in the disk Iz I < r is:

=
1
2ni
rf. j'(z)
':f j(z) dz
1
= 2n
f 2"
j'(z) 1
z j(z) df) = 2n
f 2"
j'(z)
ffiz j(z) df).
'19
z = re' .
1'1=' 0 0

163. The function w(C~ ~ ;(Z) is a polynomial of degree n - 1 in z.


Furthermore
v = 1. 2..... n.

164. Apply Cauchy's integral formula to both sides. The proposition


is now identical with V 97.
165. Let e > O. Consider that region of the z-plane where all the
I n; I>
inequalities z - e. n = O. ±1. ±2•... are satisfied (the riddled
plane). There exists a constant K depending on e such that in the entire
riddled plane

yuana97@uw.edu
336 Functions of One Complex Variable

It is sufficient to check this in the region - ~ ::;: x < ~, Iz I > e,


o < e < i; . The integral
~ 1 rf. F(C) dC
2:ni l' sin eC ~- Z)2

along the circle ICI = (n + ~); converges to 0 as n ~ 00 because


IF(C) (sin eC)-ll < C]( along the path of integration. Compute the sum
of the residues. [Hurwitz-Courant, pp. 118-123.J
166. We substitute in 165 G(z + ~) for F(z) and then z - ~ for z:

00 (_l)"G(n+J~)
d (G(Z) ) 1; e e
- dz cos ez = -::~-oo (ez - (n +!) n)2 .

We now combine the terms with the subscripts nand - n - 1:

.!-. (G(Z)
dz cos ez -
) _ £ (-1)" (n + i~) ( e + i)
,,~O
G
e [ez - (n :n]2 + [ez + (n(!)
+ i) :n]2
and integrate

G(z) _
COS(F--n~o(-)
£ 1 n G (n + !) :n) (1
- - e - ez-(n+!):n+ez+(n+!):n'
1)
The constant of integration has to be 0 because there must be an odd
function on both sides.
167. The functions I(z) log z and :i'1(z) are regular in the domain
described in the diagram.

168. Since Ilog z - inl < n fOl Izl = 1167 implies

j
I I(x) dx I < ~.
If k is an integer we replace I(z) by :i'1(z); if k is not an integer the second
formula in 167 is used.

yuana97@uw.edu
Pt. III, Solutions 166-171 337

169. [D. Hilbert, d. H. Weyl: Diss. Gottingen 1908, p. 83; F. Wiener:


Math. Ann. Vol. 68, p. 361 (1910); I. Schur: J. reine angew. Math. Vol.
140, p. 16 (1911); L. Fejer and F. Riesz: Math. Z. Vol. 11, pp. 305-314
(1921).J Introduce in 168

I(z) = 2~ (Xl + X2z + Xa z2 + ... + X"Z"-1)2, k = IX + 1.


Then [122J
2"
J I/(eit?) I dO- = xi + X~ + x~ + ... + X! = 1
o
and

J
o
1
Jt'1(x) dx = - 1: 1: XAX,.. J ~+"'+<x-1 dx = - 1:
1 ""
2n A=l ,..=1
1
0
1 "
2n A=l ,..=1
xx
1:" _A_,.._.
A + P. + IX
170. Let L denote a closed continuous curve without double points,
let L lie completely inside 91, and let z be a point in the interior of L.
According to Cauchy's integral theorem we have

I,,(z) 1.
=-2 :n(C)
mj'~-z
~ dC.
L

From this and from the fact that


lim f,,(C) = f(C)
,,-+= C- Z C- z
is uniformly valid on L it follows that 1(') is continuous on L. Further-
more

The last function is regular inside L.


171. [T. Carleman.J We call F,(z) the area integral of lover the disk
with center z, radius r and boundary circle Cr' The element of arc of C,
is dz = eiy Idz I. If X varies and its increment is L1x, i.e. if the area of
integration is displaced through L1x in the direction of the x-axis, the
change of the area per element of arc is Idz IL1x sin t" as is obvious by
geometric considerations; i.e.
8F,(z)
a-
x
= ~ 1sin t" Idz I·
c,
By a similar argument we obtain
BF,(z)
-By = - ~ 1 coSt" Idzl,
c,

yuana97@uw.edu
338 Functions of One Complex Variable

hence by virtue of the hypothesis

Therefore [po 113] F,(z) is analytic as well as y-2F,(z) and finally [170]

. F,(z)
hm - 2 -
,-+0 rn
= f(z) •

Instead of all the circles we could consider all the curves that are similar
to a given closed curve without double points and similarly situated.
172. To show that the difference
2,. 2,.
Jf(e i {}) d~ - 2nf(O) = J [f(e i {}) - f(re i {})] d~, 0~r < 1 [118]
o 0

vanishes we prove that this expression can be made arbitrarily small with
1 - r. We take each point of discontinuity as the center of an open disk
of radius e so that different disks have no common point. Removing these
"small" disks from the unit disk we obtain a domain in which f(z) is
uniformly continuous. We now split the integral into two parts: The first
part consists of the integral along'the arcs inside the above mentioned
disks, the second part consists of the integral along the remaining piece
of the circle. The first part can be made arbitrarily small with e [f(z)
is bounded] and then, once e is fixed, the second part can be made
arbitrarily small when 1 - r is sufficiently small.
173. Cf. 174. Cf. also 231, Hurwitz-Courant, p. 327, and E. Hille,
Vol. II, p. 36l.
174. It is convenient to consider the following general situation:
Let 'Il be a simply connected domain in the C-plane and assume that
the mapping tp(C) = Z is conformal in the interior of 'Il, sufficiently
continuous on the boundary and that it establishes a one to one rela-
tionship between 'Il and the disk !Z! < 1; let the point C= z correspond
to Z = O. We have C = tp-l(Z) where '1'-1 denotes the inverse function of
'1'. We "transplant" the function f(C), which is regular in 'Il, from the C-
plane to the Z-plane by defining

f(C) = f[tp-l(Z)] = F(Z).


We have
dZ
F(O) = ~ F(Z) 2niZ '

yuana97@uw.edu
Pt. III, Solutions 172-177 339

the integral computed along the positively oriented circle IZ! = 1 [172].
The change of variable Z = lJ'(e), for which

F[lJ'(C)] = I(e), lJ'(z) = 0,


leads to
I(z) = rK f(C) dtp(C) •
:r 2nitp(C) •
the integral is taken along the positively oriented boundary of '1). 173
and 174 are particular cases:
tp'(C) dC ,.2) de
'1): lei < R, -;Pm = R2 -
(R2 -
2Rr cos (e - If) + ,.2 ,

Z - C tp'(C) dC 2.¥ d'YJ


lJ'(e) = z + C' -;Pm = - x2 + tTl _ y)2 '
e= it], z= x + iy.
175. [J. L. W. V. Jensen: Acta Math. Vol. 22, pp. 359-364 (1899).
Cf. E. Hille, Vol. II, pp. 189-190.] The integrand is a single valued
function on 9t.; 9t. contains the point r = 0 if e is sufficiently small
(hypothesis) and then the value of the integral is 2ni log 1(0). We denote
by IX], IX2' '" , IXm' PI' ... ,p,. the endpoints of the paths that connect the
e-circles around al , a2 , ... , am' b1 , ... , b,. with the unit circle Iz I = 1
(IIX] I = ... = IIXm I = IPll = ... = IP,.I = 1). The loop starting at lXI"
turning around af' along the e-circle and ending at IXI' contributes, as
e'""* 0,
"'I'
-n~2 . jdZ -=-n~og-
Z
2' I IXI'
af'
af'
(af' is assumed to be a simple zero). The circle Izl = 1 contributes
2,.
J log I(eif}) i df}. Take the imaginary part on both sides of the equation
o
/J
J log I(re,f}) i df} -
2,..
+ 2ni 1: log b =
mIX"
2ni 1: log af' V
2ni log 1(0).
o 1'=1 I' .=1 •

The method used in 120 leads to another proof.


176. Cf. 177. Cf. also 232.
177. [F. and R. Nevanlinna: Acta Soc. Sci. Fennicae Vol. 50, No.5
(1922).] Same notation as in 174. Consider a function I(e) that is mero-
morphic in '1), and different from 0 and 00 on the boundary of '1) and at
the inner point z; inside '1) the zeros are a], a2 > . . . , am and the poles

yuana97@uw.edu
340 Functions of One Complex Variable

bJ> b2 , ... , bn • The "transplanted" function F(Z) = 1[1p-1(Z)J has the


zeros AI' = 1p(al') and the poles B. = 1p(b.). Then [175J
m 1 n 1 dZ
log IF(O) 1 +I'~log IAI'I-.~log IB.I = ~log /F(Z)/2niZ'

the integral taken along IZ 1 = 1. Hence by a change of variables [174J


m 1 n 1 d'P(C)
I'~ log 1'P(apl! - .~log 1'P(b.) I = ~ log I/(C) 12ni'P(C) - log I/(z) I,
the integral taken along the boundary of il. The signs on the left hand
side are in evidence: 11p(al') 1 < 1, 11p(b.) 1 < 1; ~~~~) is positive. Suppose
that I(C) is variable and I/(z) 1 is fixed; then the essence of the formula
can be roughly expressed as follows: zeros in the interior increase,
poles in the interior decrease the moduli of the boundary values.
176, 177 are a consequence of Jensen's general formula as 173 and
174 follow from the generalized formula given in 174. - Also the proof
of 120 can be suitably generalized [174]. -The condition that I(z) be
different from 0 on the boundary can always be weakened [120J, that
I(z) be regular on the boundary can be weakened in many cases.
178. [Cf. F. Nevanlinna: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 175,
p. 676 (1922); T. Carleman: A.rk. Mat., Astron. Fys. Vol. 17, No.9, p. 5
(1923).J Put a circle of radius e, e sufficiently small, around a!, and
connect the e-circle with the half-circle Izl = R, - ; <arg z < ;
so that the connecting paths do not intersect. Removal of the e-circles
and the connecting paths reduces il to a simply connected region ffi.;
its boundary is the appropriate path of integration. The loop around
a!, starting and ending at zl" IZI'I = R, contributes

Take the real part of the expression

"
1+j-R) (-
+-
£log I(Re
2 . 2 cos 11
'D ) -R- dO. +
(' 1
y2
1)
+ R2 log I(ty) dy
.

-2

- '2
"
+ flog I(re )
iD (-iD
e re iD)
- y - + RZ dO. -
m
2n 17
(1a - aI'
R2 -
1 -)
Z +
~I'
R2 = O.
"
+- 1'=1 I' I'
2

yuana97@uw.edu
Pt. III, Solutions 178-184 341

The formula is so simple because the differential (:2 + ~2) ~z is always


real on the half-circle Iz I = R, mz> 0 and the function ! - ~2 is
purely imaginary.
179. We introduce ao = lao I e·J"I =l= 0, z - z~ = r~ei<p., thus

U(z) + iV(z) = aa(z - Zl) (z - Z2) ••• (z - z,,)

when z = x is real and increasing from - + <X> all the arguments


<X> to
increase from -:n; to 0, hence arctan ~~) = I' + rp] + rp2 + ... + rp"
increases by n:n;. The quotient ~~)) = tan (I' + rp] + rp2 + ... + rp,,)
therefore assumes the value 0 n times and becomes infinite n times.
This method shows in addition that the zeros of U(x) and V(x) alternate.
180. Obvious, because the argument changes continuously.
181. Consequence of the argument principle: Compare the zeros and
poles of the factors with the zeros and poles of the product and take
into account the points that are special (pole or zero) for both rp(z) and
1jJ(z).-Directly: arg I(z) = arg rp(z) + arg 1jJ(z).
182. A polynomial is the product of linear factors z - zo' where Zo
denotes a zero. According to 181 it is sufficient to prove the proposition
for a linear function z - zoo The transformation w = z - Zo translates
the curve L through the vector -zoo If Zo is inside L the point w = 0 is
inside the image of L, the winding number is equal to 1. In the other
case the winding number is equal to O.
183. If I(z) is regular in i) with the possible exception of finitely
many poles and if it is different from 0 on the boundary L of i) then it
can be written as a product, I(z) = R(z) rp(z), of a rational function R(z)
and a regular function rp(z) without zeros in i), that is \I. function rp(z)
to which the restricted principle of the argument as stated in the problem
applies. [181, 182.]
184. [A. Hurwitz: Math. Ann. Vol. 57, p.444 (1903). Cf. also Ch.
Sturm: J. Math. Pures Appl. Vol. 1, p. 431 (1836).] The number of zeros
has to be < 2n by virtue of VI 14. The winding number of the curve
described by P(z), where z = ri~, r> 0 and {} varies from 0 to 2:n;, is·at
least m because P(z) has a zero of ordet m at the origin. Hence it inter-
sects the imaginary axis at least 2m times. Consider r = 1 - e, e> 0
and suitably chosen, or r = 1 depending on whether P(z) has zeros on the
circle Iz I = 1 or not. An essentially different proof follows from II 141.

yuana97@uw.edu
342 Functions of One Complex Variable

185. The polynomial P(z) = ao + a1z + a2 z2 + ... + a,.z" has n


zeros inside the unit circle Iz I = 1 [0]. Hence the winding number of
the image of the unit circle under w = P(z) is equal to n. Apply 180 to
the positive and to the negative part of the imaginary axis.
188. [A. ·Ostrowski.] Imagine in the w-plane a circle of radius
M = max I/(z) I and center w = a. This circle does not contain the point
.0nL
W = O. The image of L under the mapping w = a - I(z) lies inside the
above mentioned circle, therefore its winding num ber is o.
187. For z = iy, -1- < y < 1-, w = ~i)' - e-:n;)' = 2i sin ny. For
z = x ± Ii, x > 0, w = +i(~ + e-:nZ). Thus, when z moves along the
boundary in the positive direction w describes the imaginary axis from
+ ioo to - ioo. Let z = x + iy, x fixed, and -1- < Y < 1. The decom-
position
w = ~z • e:n;y - e-:nz • e- my = (e:nz _ e-:nZ) cos ny + i(e:nz + e-:nZ) sin ny
shows that w describes the right half of the ellipse centred at the origin
and with semi-axes ~z - e-:nz and ~z + e-ru: as z moves on the line
mz = x, -I ~ 3z < I·
Therefore, if Wo is an arbitrary point in the right half-plane 9iw > 0,
x can be chosen so large that the following situation is met: As z moves
on the boundary of the rectangular region
0< mz < x, -I < 3z < 1
the winding number of the curve described by w - Wo in the w-plane
is equal to 1. Cf. also 188.
188. Let Wo denote an arbitrary point in the interior of the image C
of the circle Iz I = 1'. The translation given by the vector -wo moves C
to C'; i.e. C' is the image of the circle Iz I = l' under the mapping
w = I(z) - woo According, to the hypothesis its winding number must
be +1 or -1. Since I(z) - Wo is regular the winding number has to be
non-negative [argument principle], thus it is +1. The function I(z) - Wo
has, therefore, exactly one zero in the disk Iz I < 1'. In an analogous
way we show that a point Wo outside f cannot belong to the range of
I(z) in Izl < l' (the corresponding winding number is 0).
189. We map the circular sector, - : <arg z < : ' Izl < 1', by

J e-'2 z dx into the w-plane. The image of the two radii


• 1.
means of w =
o
can be obtained by rotating through 45 0 , and reflecting and rotating,
respectively, the right half of the Cornu spiral, described by Sommerfeld,

yuana97@uw.edu
Pt. III, Solutions 185-192 843

l.c. The image of the bounding circular arc is near the point w = V;
[IV 189]. We choose r so small that the image of the arc z = riB,
- : <1} < : ' intersects the images of the radii only at the points that
.:rI • n
1- -1-
correspond to z = re 4 and z = re 4. We now consider the image of
the boundary of the above mentioned sector outside the circle Iz I = r.
It has the winding number O. Hence w =F 0 when - : < arg z < ; .
Different proof in V 178.
190. The integral
1 ,(. I'(z) dz
2ni 'f l(z) - a
is an integer. Therefore it cannot change if the path of integration is
deformed continuously. The curves in question can be transformed into
each other by a continuous deformation.
191. Let I(z) =1= 0, then I(z) =F 0 on L. The function
log I(z) = log R + iO is regula at every point of L. Now (clear when 0'/1
and os coincide with ox and 8y, respectively, d. p. 113]
8logR 8@
-----av = as '
where ! denotes the differentiation in the direction of the outer normal
and ! in the direction of the positive tangent at the point z of L. The
derivative on the left hand side is positive [maximum principle] hence
the derivative on the right hand side has to be positive too. The image
of L is a circle (possibly described several times in the same sense).
192. [B. Riemann: Werke. Leipzig: B. G. Teubner 1876, pp.106-
107; H. M. Macdonald: Proc. Lond. Math. Soc. Vol. 29, pp. 576-577
(1898); d. also G. N. Watson: Proc. Lond. Math. Soc. Ser. 2, Vol. 15,
pp. 227-242 (1916); Whittaker and Watson, p. 121.] We assume that
f'(z) = Riei8 ~~ =F 0 on L (same notation as in 191). Let the winding
numbers of the curves described by I(z) and I' (z) be denoted by Wand
W' resp. (W is the winding number of a circle which may be described
several times in the same sense). Thus 2n(W' - W) is equal to the change
of the argument of ~~ as z describes the curve L. Let ds = Idz I be the
line element of L, ~~ = ~~ ~; . The first factor is always real, therefore
only the change of the argument of the second factor is important. The
argument of ~; = I~;I is equal to the negative value of the argument of

yuana97@uw.edu
344 Functions of One Complex Variable

dz. The change in direction of dz, i.e. the change in direction of the tangent
vector along a closed curve without double points, is 2,,;; this can be
easily seen in the case of a polygon. Hence W' - W = -1. If I' (z)
vanishes on L we consider a level line inside but sufficiently close to L.
The geometric formulation of the proposition can also be verified directly;
d. H. M. Macdonald, l.c., or MPR, Vol. 1, pp. 163-164.
193. The function I(z) is not a constant because f'(z) =1= 0; I(z) has
exactly one zero in the interior of ~ [192]. Therefore w = I(z) describes
a (circular) curve with the winding number 1 as z describes the boundary
of ~. The winding number of the path described by I(z) - Wo is 1 or 0
depending on whether IWo I < or IWo I > the constant modulus of I(z)
on the boundary of ~.
194. [E. Rouche: J. de l'Ec. Pol. Vol. 39, p. 217 (1862).] We may
assume that I(z) and fP(z) are regular on L because I(z) and I(z) + fP(z)
are different from 0 and I/(z) I > IfP(z) I sufficiently close to L. The func-
tion 1 + ~~;~ has on L a positive real part and so, as z describes L,
the change of its argument is equal to o. Furthermore I(z) + fP(z) =
I(z) (1 + ~~:~) thus [181] the image of Lunder I(z) has the same winding
number as the one under I(z) + fP(z).
195. Special case of 194: I(z) = ze A- z , fP(z) = -1, L is the unit circle.
Since z;-· increases with z on 0 < z < 1 from 0 to ;-1 > 1 there is a
root on this segment.
196. If z = iy, Y real, we have lit - iy I >it > 1 = le- iY I. If Izl is
sufficiently large, mz > 0, we have Iit - z I > 1 ~ Ie- z I. Moreover
it - z = 0 has the root z = it in the right half-plane. Special case of 194:
I(z) = it - z, fPCz) = -e-', L a sufficiently large half-circle in mz > 0
and its closing diameter. The only root is real because the modular graph
is symmetric with respect to the vertical plane through the real axis.
197. [Cf. G. Julia: J. Math. Pures Appl. Ser. 8, Vol. 1, p. 63 (1918).]
Special case of 194: on the unit circle Izl = 1 > I/(z) I.
198. [Example of a general theorem of G. Julia: Ann. Sci. Ecole
Norm. Sup. (Paris) Ser. 3, Vol. 36, pp. 104-108 (1919).] Let R,. be the
rectangle with the four corners n ± t ± id, n integer, d fixed, d > 0
and V';Tite z = x + iy = reiD. Because of 1155 we have on the boundary
of R" as n -+ 00
log IF(z) I N (x - t) log r - y{} - x,
so that the minimum of IF(z) I on R" tends to 00 as n -+ 00. On the other
hand we find Isin ;;r;z I > c on the boundary of R,. where c is independent

yuana97@uw.edu
Pt. III. Solutions 193-200 345

of nand c > o. Hence the minimum of

_1
1r(z)j
1= Isin nz1lr(1
n
_ z) I
converges on the boundary of R_ .. to + 00 as n -+ 00. Thus we have for
arbitrary a and sufficiently large n

Ir~z)1 > lal


1
on the boundary of R_ .. whereas at the center of R_ .. r(z) = O. Apply
1
1M to fez) = r(z) , 9'(z) = -a.
189. Integrating by parts twice we obtain

zF(z) = /(0) -/(1) cos z + ! [1'(1) sin z - / t"(t) sin zt dt]


= /(0) - /(1) cos z + 9'(z) .
We draw circles of radius e around all the zeros of the periodic function
/(0) - /(1) cos z, where e > 0 and 2e smaller than the distance between
any two zeros. In the z-plane from which all the e-disks have been lifted
1(0) _'PX~) cos z
converges to 0 as z -+ 00. [To begin with prove this for
the strip -:n~ mz~ :n.] With the exception of finitely many zeros
zF(z) has, therefore, in any disk the same number of zeros as the function
/(0) -/(1) cos z [1M], that is 1. This zero is necessarily real in the case
1/(1) I> 1/(0) I where the disk is cut in half by the real axis; the non-real
zeros of the functions that assume real values for real z appear in pairs
[solution 196].
200. The term z"a-'" assumes its role as maximum term of the
series

on the circle Iz I = Ia 1
2.. - 1 and abandons it on the circle Iz \ = \a \2"+1

[I 117]. To study the dominance of the maximum term in between those


two circles notice the formula

F(z) - z"a-'" z z z z z z
~'-z"-a--""" = -a2-. . +-1 + a-2"-+1 . -a2-"+-S + -a2-"+ J • -a2-. . +-S • a-2-"+-5 + ...
+-+_._+_._._+
a 2.. - 1
z
a 2"- a 2.. - S
z z
J a 2"- 1 a2n - S a2.. - 5
z z z
....

yuana97@uw.edu
346 Functions of One Complex Variable

On the circle Iz I= Ia 12.. the corresponding terms of the two subseries on


the right hand side have the same absolute value. thus

I,!(z) - z"a-"'I < 2(-~ +~._1_ +~._1_._1_ + ...)


z"a-'" I lal lal lals lal lal s lali
2 1 21al8
< Tiii 1 - Ia 1-3 = Ia IS - 1 < 1.
because the only positive root of the equation z3 - 2Z2 - 1 = 0 is
smaller than 2.0 [18]. Hence F(z) has in the disk Izl < lal 2.. the same
number of zeros as z"a-..•• namely n. The disk Iz I < Ia 12.. - 2 contains. by
virtue of the same proof. n - 1 zeros. -Cf. V 178.
201. [A. Hurwitz: Math. Ann. Vol. 33. pp. 246-266 (1889).] The
closed disk D has a as its center. lies completely in mand contains no
other zeros of I(z) than possibly a. We have I/(z) I > 1/.. (z) -/(z) Ion the
boundary of D when n is sufficiently large. Apply 194.I,,(z) -/(z) = tp(z).
More generally: Each subdomain of m
on the boundary of which there
are no zeros of I(z) contains exactly the same number of zeros of I.. (z)
as of I(z) if n is sufficiently large. Important for the applications!
202. The limit function is regular in the unit disk Iz 1 < 1 [170].
Assume that l(zl) = l(z2) for ZI =l= Z2' 1zll < 1. 1z21 < 1; consider the
sequence I.. (z) -1.. (z1)' n = 1. 2•... which converges to I(z) -/(zl).
In a disk that has its center at Z2' lies completely inside the unit circle
and does not contain zl.I.. (z) -I.. (z) would have to vanish for sufficiently
large n [201]: contradiction.
203. [170.201.]
204. In the case where a and d are integers the proposition is proved
in the same way as proposition 185 because the zeros of the polynomial
aolf + a 1lf+ d + a 2t'+2d + ... + a..t'+nd lie in the disk Iz I ~ 1 [23].
If a and d are rational z has to be replaced by a suitable multiple of z.
If a and d are irrational approximate these two constants by rational
numbers and apply 203.
205. [G. P6lya: Math. Z. Vol. 2. p. 354 (1918).] We have [II 21]
1 ,,-1
JI(t) cos zt dt = ..lim
o ...,.00
I.=1
~ 1(~) cos ~ z
n n n
[185.203].

208. Counter-example
1
I ..(z) = Z2 +-;-. n = 1. 2. 3•... ; :1): Izl~ 2; a = -1. b= +1.
208.1. [G. P6lya. Problem: Jber. deutsch. Math. Verein. Vol. 34.
2. Abt.. p. 97 (1925). Solved by R. Jungen: Jber. deutsch. Math. Verein.

yuana97@uw.edu
Pt. III, Solutions 201-208 347

Vol. 40, 2. Abt., pp.6-7 (1931).] We may assume without loss of


generality that Mz), fs(z), ... , f,,(z) are linearly independent and
1c1 12 + 1c2 12 + ... + 1c,,12 = 1
so that the set of coefficients cl' c2 ' ••• , c" can be conceived as a point c
on the surface 6 of the unit sphere in 2n dimensions. If there were no
finite upper bound of the nature stated, there would exist an infinite
sequence of points c', c", ... such that the linear combination correspond-
ing to c(m) has not less than m zeros in ~. This sequence has at least one
limit point c(co) on 6, yet the linear combination corresponding to c(co)
has only a finite number of zeros in ~. Hence contradiction to the last
remark of solution 201.
208.2. [G. P6lya, Problem: Jber. deutsch. Math. Verein. Vol. 34,
2. Abt., p.97 (1925). Solved by Nikola Obreschkoff: Jber. deutsch.
Math. Verein. Vol. 37, 2. Abt., pp. 82-84 (1928).] Apply the argument
principle to a rectangle with corners
-a + ilX, a + ilX, a + ifJ, -a + ifJ
where a is sufficiently large. Use V 75 and solution 180 in considering the
horizontal sides.
, dw
207. From z - w91(z) = 0 follows 1 - w91 (z) = 91(z) dz' Thus La-
grange's formula (L) (p.145) for fez) implies by differentiation with
respect to w
/,(z) IP~ =
1 - WqJ'(z)
1:
,,=1 (n -
w"-l
i)!
[d"-l/,(X) IP(X) [1P(X))"-1]
dx"-1 :1:=0 '

i.e. the formula to be proved for the function t'(z) 91(z). The family of
admissible functions f(z)- is identical with the family of functions t'(z) 91(z),
where t'(z) is the deIivative of an admissible function, because 91(0) =l= O.
Thus 207 leads to Lagrange's formula (L), p. 145, by integration.
208.
~ [d"j(X) [IP(X)]"] = -.!.., ,(. j(C) [1P(Cl]" dC,
n! dx" %=0 2:/tI- ':J' C" C
1: w:n.
,,=0
[d"j(X) [:(X)
dx
J"] .. =0
= ~
2m
,(. j(C) dC
':J' C
i
,,=0
(WIP(C) )" ,
C
integrated along a circle around the center C= 0 and for w so small that
1C1 > 1w91(C) 1 along the path of integration. Then the path of integration
encloses the same number of zeros of C - w91(C) as of C [194], i.e. exactly
one. Denoting this single zero by z we further find
1 ,(. j(C) dC j(z)
2ni'Y C- wlP(C) = 1 - wlP'(z) •

yuana97@uw.edu
348 Functions of One Complex Variable

209. [L. Euler: De serie Lambertiana, Opera Omnia, Ser. 1, Vol. 6.


Leipzig and Berlin: B. G. Teubner 1921, p. 354.] In (L), p.145, set
ljI(z) = e', j(z) = Z,
2w2 32w3 w" n"-l
Z= W + 2T +31 + ... +-n-!- + ....
210. Introduce in (L), p. 145, ljI(z) = e', j(z) = eat',
at. _ 1 + "'";, IX(IX + n}"-1 w."
e -
,,=1 n!
211. We write x = 1 + z, ljI(z) = (1 + z)P, j(z) = 1 + z; (L), p. 145,
yields
P "
= 1 + z = 1 + 2: (n .: 1) wn .
00

x
,,=1
212. [Cf.l.e. 209, p. 350. ] Set x = 1 + z, ljI(z) = (1 + z)P, I(z) = (1 + z)";
(L), p. 145, implies
y= xat = (1 + z)at = 1 + i (IX +n pn- 1- 1) IXW"
n

,,=1
213. We obtain for {1 = 0, fJ = 1 the binomial series, for fJ = 2

(1- JI~)at
2w
= 1 + lX i (IX +
"=1
2n -
n - 1
1) w" .
n'
for f1 = -1 essentially the same series; for f1 = i

(1Y;-;--2
1+~ +;
)2 at
= 1
(IX + ~ -
+ lXn~' n _ 1
00 1) "
:.

Put x = 1 + ~ , w = ; , lX = af1; fix~, w, a and let f1 increase to + 00.

The equation in 211 becomes


w= ~ (1 + ~ t P c-J ~e-~ .
214. By setting ljI(z) = e', I(z) = eotl and applying 207 we obtain
00 (n + IX)" w" eat. eat'
,,~--n-!--- = 1- we' = 1- z'

where z has the same meaning as in 209. The radius of convergence is


. (n +
= hm ------
IX)"
(n + 1)! = e-1 •
n~oo n! (n + 1 + IXt+1
215. We are dealing with

~
"'"
"=0"
Jn+1
t" e-J.I dt = ~
n! "'"
,,=0
J 1
(n + IXt e-J.(n+<»
n!
,
dlX.
0

yuana97@uw.edu
Pt. III, Solutions 209-219 349

The convergence of the series

i(n + IX)" e-A"=~ 1 - z'


,,= O nl

is uniform for 0 =:;;: IX =:;;: 1, and z is given by the equation ze-' = e- A,


Iz I < 1 [214, 209]. Thus the integral in question becomes
= J J
e,,(,-A) dlX
1-z
= e,-A - 1
(1-z)(z-1)
= _1_ .
1-z
o
We verify immediately that C=A - zsatisfies theequationA-C-e-c=O;
A - z is real and positive.
216. [Regarding 216-218, 225, 226, cf. G. P6lya: Enseignement
Math. Vol. 22, pp. 38-47 (1922).] We put 9'(z) = (1 +z)lI, I(z) = (1 +z)"
and apply 207:
co (IX + Pn) " (1 + z)" x"'+1
,,~ n W = 1 _ wfl(1 + z)II-J = (1 - fl) x + fl '
where 1 + z = x and x is the root of the equation 211 which is algebraic
for rational p.
217. [Cf. L. Euler: Opuscula analytica, Vol. 1. Petropoli 1783,
pp. 48-62.] We introduce 9'(z) = 1 + z + Z2, I(z) = 1 and apply 207:
1 - w - V1 - 2w - 3w2
Z= .
2w '

co w"'[a"(l + x + X 2)"] 1 1
,,~ nT ax" %=0 = 1 - w(l + 2z) = V 1 - 2w - 3wl •

218. The sum of the terms in the k-th column to the left of the middle

e
column is [216]

1 + (k ~ 2) W + ~ 4) w2 + ... + (k : 2n) w" + ...


=_1_(1-V~)"
Vl-4w 2w .

219. [Cf. Jacobi: Werke, Vol. 6. Berlin: G. Reimer 1891, p.22.]


It is sufficient to consider the cases (2) and (3).
(2) Suppose E =1=-1, E =1= 1; choose some branch of (1- E)'" and of
(1 + E)II if IX and p are not integers. Introduce in 207
9'(z) = (~ + Z~2 - 1; I(z)...:.- (1 - E - z)'" (1 + E + z)lI,
1(0) = (1 - E)'" (1 + E)II.

yuana97@uw.edu
350 Functions of One Complex Variable

co

On the right hand side we find (1 - E)'" (1 + E)fJ 1: p!;.fJ)(E) w", whereas
.. =0
w = tp~(
z) implies
-
1- ;w - y'1 - 2~w + Wi
Z= W '

I(z) = 2"'+fJ (1 - E)'" (1 + E)fJ (1 - w + VI - 2Ew + w2 )-", X


(1 + w + VI - 2Ew + w2 ) -fJ , 1 - wrp' (z) = VI - 2Ew + w'l..
Thus

,,=0

=V1=2"'+fJ (l-w +Vl- 2Ew + w2 )-"' (1 +w +Vl- 2Ew +~)-fJ.


1-2~w+w2

The two cases E= -1, E= 1 can be discussed directly [solution VI 98].


(3) Assume E=t= 0 and put in 207
9'(z) = z + E; I(z) = e-<'H)(z + E)"', 1(0) = e-eE"'.
00

On the right hand side we have e-eE'" 1: L~"')(E) w"; the relation w = ~z)
yields ,,=0 tp

~w ~'" fIJ~t' 1 - wrp' (z) = 1 - w,


z=l_w' /(z) = (1 _ w)'" e ,
therefore

The case E = 0 can be treated directly [solution VI 99].


220.
LieU = eU(eS - 1) , Li"e" = eSl(eS - I)" .

(1) (1 + w)" = 1 +~ w + z(z 2~ 1) w2 + ...


+ z(z - 1) ... (z - n + 1) w" + ....'
n!
where w = e' - 1, valid for Ie' - 11 < 1, z arbitrary.

(2) 5 = eS - 1 - ~ (eS - 1)2 +~ (e' - 1)3 - ...

+ (_1)"-1~(e'
n
-I)" + ... , Ie' -11 < 1.

(3) e"--1
- + TI5e
z +z(z - 2) (e')2
-2!-- S
5 + ... +z(z
- --- -nt-
;j-
1
(
51:'-s)" + "',

yuana97@uw.edu
Pt. III, Solutions 220-222 351

follows from 210 with z = -5, IX = -z. AC':ording to II 205 we have


( ) ,,-1
Z Z -n~ (56")" ~ (_1)"-1 ze-'(2n)-1 n- t (5es+1 )" ,

hence the series converges whenever \56"+1\ ~ 1; at any rate the formula
holds in a neighbourhood of 5 = O.
(4) in 212 put: x = eS , {J = 1, thus w = 6"/2 - e- s/2 , IX = Z,

Replace 5 by -5 and add the two formulas:


S
e' +e
-s,
= 1 +I co
~ (Z +m- 1) ( S
e2 _ e - "2
S )2...
2 ... =12m 2m - 1

Replace 5 by -5, subtract and rearrange and then apply 216 to x = 6",
{J = 1, W = c/2 - e- s/ 2 , IX = z -1:

eS' - e-SZ = ;.
2 ""'2
..!.. (Z + m -
2m-1
1) (ei _ e-i)2. . -2 (6" _ -S)
e. .
... =1

The sum of the last two formulas produces the desired expansion of eSl •
We remember Stirling's formula and write sin nz as an infinite product
to obtain
Z(Z + m -
-- 1) (
ei -e - i)2'" Zsin 1U (. is )2... m - i .,
Sln-
2m 2m-1
~--=-
J/:n; 2

this shows convergence for sin I i: I< 1; the formula is certainly correct
in a neighbourhood of 5 = O.
221. [With respect to the formulas (2) and (3) cf. N. H. Abel:
Oeuvres, Vol. 2, Nouvelle edition. Christiania: Gr0ndahl & Son 1881,
pp. 72-73.] Expand in 220, F(z) = c', both sides in ascending powers
of 5 and compare the coefficients of ;!" .We have
s"
LI" F(z) = LI" ( 1 +,.
S

1.
z + ,.
2.
Z2 + ... +,. z" + ... = I ,. LI"z" .
S2

k.
S"

,,=ok.
) co

222. It is sufficient to prove (1) and (2) for F(z) = (z - W)-I; for
other rational functions differentiate with respect to w, split them
up into partial fractions and apply 221. The formulas (1) and (2) hold
for F(z) = eSZ if 5 is real and negative. [Solution 220.] Multiplication
with e-sfI) d5 and integration over - CX) < s < 0 yield (1) and (2) for

yuana97@uw.edu
352 Functions of One Complex Variable

F(z) = (z - W)-I. It is easier to discuss the range, in which the formulas


are valid, in the following way: Since
A"(
LJZ-W)-1 (-1)" n!
(z - w) (z - w + 1) ... (z - w + n)
=-:-----:---:---'------'..,.,---c,-----:-
,

we have to prove
(1) 1 1 z(z - 1) ... (z - n)
w- ~ w(w -
oa

z- w= - 1) ... (w - n - 1) ,

1 oa nl
(2) - (z _ W)2 = - ~ (z -
,,=0
w) (z - w + 1) ... (z - w + n + 1) •

We obtain (2) by replacing in (1) wand z by W - z - 1 and -1 respec-


tively. Formula (1) is derived by a limit process from the identity
l i z z(z - 1)
w - z= w(w - w+ 1) + w(w - 1) (w - 2) + ...
z(z - 1) ... (z - n +
1) z z - 1 z - n 1
+ w(w - 1) (w - 2) ... (w - n) +Ww - 1 ... w - n w-- z '
which is easily proved by complete induction. If wand z are different
from one another and from 0, 1, 2, 3, ... the remainder is [d. II 31]
(-z)(-z + 1) ... (-z + n)
n-'n! -w(-w + 1) ... (-w + n) w - z

r(-w) n llJ - .
~ r(-z) w - z '

The formulas (3) and (4) cannot be valid in a non-empty region for
non-entire rational functions: Otherwise the partial sums of order nand
2n resp. would converge uniformly in any finite domain according to 255
and 254, thus F(z) would be regular everywhere: contradiction.
223. The identity in question is purely formal; it represents the
combination of the infinitely many equations that define the quantities
L1"ak , k = 0, ±1, ±2, ...
224. Since the expansion of 1 : t does not contain a constant term
the coefficient f!' in the expansion of 1 ~ tF(l : t) depends on ao, aI' ... , a"
only. Therefore we can restrict ourselves to the case where F(z) is a poly-
nomial. Every polynomial however can be written as a linear combination
of the speci!;l.l polynomials (1 - z)"', m = 0, 1, 2, ... ; besides we have for
two sequences ak and bk and two constants c1 and c2 the linear relation
L1"(c1 ak + c2 b",) = c1 L1"a", + c2 L1"b". Consequently it is sufficient to prove
the statement for F(z) = (1 - z)"', m = 0, 1, 2, ... In this case L1"ao is

yuana97@uw.edu
Pt. III, Solutions 223-229 353

equal [223] to the coefficient of z" in the expansion of (1 - z)" • (1 - z)'"


= (1- z)"+"', i.e. = (_l),,(n ~ m). Thus
i;
,,=0(-1)
,,(n +n m) ('=(1 + 1t)",+1" = 1
1
+tF l+t'
(t)
225. It is sufficient to prove the statement for F(z) = (1 - z)"',
m = 0, 1,2, ... [solution 224]. Then the quantity L12"a_" is [223] equal
to the coefficient of z" in the expansion of

(1 -z)2 " k _ (1
.:;., ;a"z' - - -+2F(z-l)
-z)2"F(z) --- _ (1
-z)2"+,,,1 + (-l)"'z-'"
2
"=-00

i.e. = (-i)" en ! m). [Solution 218.]


228. We put F(z) = (1 - z)'" - 1, m = 1, 2, 3, ... [solution 224,
225]. Then L12"a_"+l - L1 2"a_"_1 is equal [223] to the coefficient of z"
in the expansion of
(1 - Z)2" (Z-l - z) i; a"i< = (1 - Z)200 (Z-1 - z) F(z) -2 F(Z-l)
1<=-00

i.e.
= (-i)" {en +: + 1) - en:;1+ I)}
= (_1)"+l~(2n
n+l
+m+
n
1).
Set p = 2, IX = m, W = -t in solution 213.
*227.

sin nx _
:n:x (1 - x) - ,,=1
iI (1 + .n:. .) (1 __n~_)
+
- iI (1 + n(n + X))
1 - ,,=1
x(1 -
1) .

228. [1. Schur.] By carrying out the multiplication in the last infinite
product in 227.
229. Apply (L), p. 145, to

<p(z) = T ~-z, fez) = sin 1I:Z, W = z(l- z).


We find
. ; A" .. A = [d"-l(l - x)-n:n: cos :n:x]
sm1l:z=':;"i w , n 1 [228].
.. =1 n. dx"- .,=0

yuana97@uw.edu
354 Functions of One Complex Variable

230. We split I(z) into the real and imaginary parts,


l{reifJ ) = U(r, D) + iV{r, D), U{r, D), V(r, D) real, a.. = b.. + ie", b.. , e..
real. The series
U(r, D) = bo + 1: r"(b.. cos nD - e.. sinnD) ,
.. =1

V{r, D) = Co + 1: r"{e.. cos nD + b.. sin nD)


.. =)

are uniformly convergent for 0 <D< 2n, hence [117]

1 2" .
a.. = b.. + ie.. = --;; f U{r, D) e- ...fJ dD
nr 0
'In

= ~ f- V{r,D) e- infJ dD, n = 1, 2, 3, ...


nr" 0

231. Solution 230 implies

I{z) = flo + ~l (b" + ie,,) z" = 2~ i" U{r, D) [1 + 2 ~l (ze~ifJn dD.


232. 1st special case: I{z) does not vanish in the disk Iz I ::;;: r. Apply
231 to the function log I(z), which is regular in Iz I < 1'.
2nd special case: I(z) = (Zz- c~.!., lei < 1'. Since log I/(reifJ ) 1= 0
" - cz
[5]
the integral on the right hand side vanishes.
Any function that is regular in the disk Iz I < l' and different from 0
on the circle Iz I = l' can be written as a product of special functions of the
types 1 and 2. The condition that I(z) be non-zero on the boundary
Iz I = l' can finally be dropped because both sides depend continuously on
1'. - Different solution with the help of 178, in the manner of solution 58.
233. Assume 0 < IX < 2n and that ReifJ , 0 < D < IX, is a point of
the arc in question. We use the same notation as in 231; U(r, D) = 0 for
o < D < IX; 1(0) is supposed to be real. Two limit operations lead to
. . 1 2" 1 + i(fJ-9)
lim l(re,fJ) = I(Re,fJ) = '2
' ..... R-O
f
n ..
U(R, e)----;fJ_9) de,
1 - e

. 1 2,.
f
(j-e
3/(Re,fJ) = 2n U(R, e) cot - 2- de,
.
d . 1 2" e-2
f
(j
d{j3/(Re'fJ )=-2n U(R,e)(sin--i-) de::;;:O.
"

yuana97@uw.edu
Pt. III, Solutions 230- 237 355

234. We retain the notation of 230:


2n
J [U(r, D)]2 dD =
~

2~ b~ + ~ ~ ,-2"(b; + c;) ,
o .. =1
2n
J [V(r, D)]2 dD = c~ + ~ ..~ r "(b; + c;).
~

2~ 2
o =1

235. [Cf. C. Caratbeodory: Rend. Circ. Mat. Palermo Vol. 32,


pp. 193-217 (1911).] In the notation of 230 we have R = 1, "0 = 1 and

a.. = . .
:/lTO
1
J2" U(r, D) e-<n8
.
dD, Ia.. I ~ .. J U(r, D) dD =
1 2n
:/lTO T
1
.. J

o < r < 1; n = 1, 2, 3, ...


Let r converge to l.-Example:

f(z) = ~ !!-; = ~ + z + Z2 + ... + z.. + ....


236. The function
~
2
A - f(Rz)
A - mao
+
i3ao = ~_ ~
2
1: a .. R" z..
2 .. =1 A - mao

satisfies the conditions of 235. Hence


2(A - mao)
Ia.. I : : ; - -..- , .
I.e. ~
~
la.. I""~ 2(A-ffiao) ~
~ (r) .
Ii .
R .. =1 .. =1

The upper bound is attained for the function given as example.


237. It is sufficient to prove the first equality (to obtain the second
-1
we replace z by~).
z
The function ~ a.. z" remains bounded for Izl >- 1,
n=-oo

I....~ ~ a.. z" I<


-1 I
M. We denote by A *(r) the maximum of the real part of
~

the entire function ~ anz" on the circle 1z I = r; then we have for r > 1
[solution 236] .. =0
A*(r) > ~ao + lla" 1r", n = 1, 2, 3, ... ;
furthermore A (r) > A *(r) - M, thus
A (r) > ~ao - M +11 a.. I r" .
If a.. is different from 0 we conclude
lim inf ~g A(r) > n.
,~~ logr-
There are arbitrarily large n's for which a.. =F O.

yuana97@uw.edu
356 Functions of One Complex Variable

238. [E. Landau: Arch. Math. Phys. Ser. 3, Vol. 11, pp.32-34,
(1907); d. F. Schottky: J. reine angew. Math. Vol. 117, pp.225-253
(1897).] It is sufficient to prove the inequality Ima1 1 R < ! L1(1); for
if IX is a real constant the largest oscillation of the real part of I(e i .. z) is
also L1 (I) and the inequality Ime'"' a1 I R :::;; ! L1 (I) for all IX implies
Iall R < ! L1 (I). - Let A denote the arithmetic mean of the upper and
lower bounds of m/(z) in the disk Iz I < R. We find Im/(z) - A I < IL1 (I)
for Iz I < R. Besides [230]
2"
n,a1 = / [m/(,e i")] e- i " d{}, O<,<R,
o
thus
AI
mma1 =/ b
[m/(,ei")-A]cos{}d{}, n,lma1 1:::;;-tJ /
b
Icos{}ld{}=2L1(1).
o 0
Let, converge to R.-For R = 1
I(z) =-.!..logl- z =iz +~i' +~z5
2i l+z 3 5
+ ... '
we have
L1(1)=;, la1 1=l.
In geometrical terms the theorem reads as follows: A disk is mapped
conformally but not necessarily univalently onto a region. The width
of this region in any direction is at least equal to : times the product
of the radius of the disk and the linear enlargement at the center of the
disk. In the above mentioned special case the image of the disk is the
vertical strip - ~ < m/(z) < ~ .
239. [E. Landau, O. Toeplitz: Arch. Math. Phys. Ser. 3, Vol. 11,
pp. 302-307 (1907).] The least upper bound of I/(z) -ie -z) I on
Iz I < R is denoted by D*(/), then D*(I) < D(I). Let 0 < , < R.
2"
4ma 1 =/ [f(,e~") - I( -,e''')] e- i " d{}
o
implies 4n, Iall < D*(I) . 2n < D(f) . 2n; let, converge to R. If I(z) is
linear, I(z) = 40 + a1z, then D(I) = 2 Ia1 r R. The proposition admits
the following geometrical interpretation: A disk is mapped conformally
but not necessarily schlicht onto a region. The maximum distance of
two boundary points (diameter) of this region is at least equal·to the
product of the diameter of the disk and the linear enlargement at the
center. In the special case mentioned the image is an open disk.

yuana97@uw.edu
Pt. III, Solutions 238- 242 357
240. [Cf. E. Landau: Math. Z. Vol. 20, pp. 99-100 (1924).] We derive
from 232 by differentiation and by setting Z = 0 [117]
/,(0)
- -- =
m'l
L:(- -
C) + -1 J (log M
~~
2"
-
. .
log It(re"';)1l e-·D d'l'}

r
flO) p~l Cp ,,2. nr 0 '

- ffi j(~: < i ffi (c


J'~l
1
p
- ~.) + :r 0
(log M - log It(reiD ) I) d'l'}

= ~
m
ffi ( -1 - e- p - -log
2
- r) +- 2 M
~
p~l CP r2 r Icp I " If(O) !
[120]. According to the condition (2) we have ffic p < 0; thus for I' >1

For I' <:: 1 we have


ep
ffi ( - - - 2 log -
y) ICpl + 2 log ICpl
- < O.
l' ICpl <~-
- r r
The inequality x + 2 log x < 5) holds for 0 < x < : because the left
hand side increases with x and e2/ S ( : Y< 1, i.e. 8e < 27.
241. The power series in question can be written in the form
c1 c2 Ck n
+1_ + ... + 1 _ + L: b"z
00.. 00

L: a"z
,,~o
= 1 _ zz
1 2 k
z z zz
,,~o
,
"-
IZll = IZ21 = ... = IZk 1= 1, lim sup Vlb" I < 1, whence
"-+00

a.. = CIZ~ + c2Z; + ... + ckzZ + b", Ib.. 1 < B.


242. The radius of convergence is assumed to be 1, thus zoZo = 1.
The series can be written as
cO.+c1z+ ... +ckzk n
+ ..L: b"z
00 " 00

,,~o
L: a"z = (z
0
_ z)k+l
~o
,
"-
Co + c1Z + ... + cklo =F 0, c
O k =F 0, lim sup Vlbn I < 1; whence for n > k
"-+00

yuana97@uw.edu
358 Functions of One Complex Variable

The expression in parentheses converges to


Co + C1Z0 + ... + cklo =1= 0 as n --+00.

Cf. also I 178 and II 95.3, II 197.


243. [G. P6lya: J. reine angew. Math. Vol. 151, pp. 24-25 (1921).]

The power series P(z) L; anz" = L; bnzn can be made to satisfy the
n=O ,,=0
conditions of 242 by an appropriate choice of the polynomial
P()
z = Co + c1 z + c2 z£ +... + cq_2z...0-2 + z·
..0-]. Ibnl
Ibl --+ e implies
"- 1 ,,+1
VIb,,1 = -
(!
[I 68]. Furthermore

Ibn 1= Icoa"
+ clan_1 + ... + Cq-2an-H2 + an-HI I
< A,,(ICol + Icil + '" + Icq- 2 + 1). 1

244. Let k be the number of poles. We set an = lXn + bn,


lim sup VIbn I = b <1 , i; lXnZ" rational with exactly k poles on the
n-+oo (! ,,=0

circle of convergence Iz I = e. We choose B so small that b + B- < _1_ - B.


According to 243 we have

r '(~ - r-k+l]
(!

max (llX" I, IlX n- 1I, ... , IlXn-k+l /) > max [( ~ - B B

for n sufficiently large; i.e. IlX; I > (; - BY' > Ib; I for at least one n,
n > n> n - k -+ 1; thus a; = lX; + b; =1= O. Consequently v" > ; - c,
C independent of n. The proposition holds also in the case where the multi-
plicity of the poles is not taken into account, but the proof has to be
approached differently.
245. [J. Konig: Math. Ann. Vol. 9, pp. 530-540 (1876).] If the
poles are of order k we have an = Ank-1e- n (sin (nlX + b) + Bn), A, lX, 15
real, limB" = 0 [solution 242]. Assume that A > 0,0 < 21J < lX <:n; - 21J
"-+00
and that IBnl < sin 1J whenever n > N. If the distance between nlX + 15
and the closest multiple of :n; is larger than 1J then an has the same sign
as sin (nlX + b). In the case where n > N and where a" does not have
the same sign as sin (nlX + b) the coefficients an- l and an+ l have certainly
the same signs as sin ((n - 1) lX + b) and sin ((n + 1) lX + b) resp.;
an- l and a,,+l have different signs because -1J < nlX + 15 - m:n; < 1J
implies -:n; + 1J < (n - 1) lX + 15 - m:n; < -1J,
1J < (n + 1) lX + 15 - m:n; < :n; - 1J. Therefore the number of changes of

yuana97@uw.edu
Pt. III, Solutions 243-248 359
sign between a,,_l a" is the same
as between sin ((n - 1) l¥ + <5) sin (nl¥ + <5) sin ((n + 1) <5 + l¥).
Now use VIII 14.
248. The radius of convergence of the series I a"z" is assumed to be
"=0
1. If the series converges at some point of the circle of convergence we
have lim a" = 0, thus [I 85]
"-+00

a
lim (1 - z) (ao + alz + ... + a"z" + ...) = lim ...!!.. = 0
.-+1 -0 "-+00 1

hence. the point z = 1 cannot be a pole.


247. [M. Fekete: C. R. Acad. Sci. (Paris) Ser. A-B, Vol. 150,
pp.l033-1036 (1910); G. H. Hardy: Proc. Lond. Math. Soc. Ser. 2,
Vol. 8, pp. 277-294 (1910).] Let
I(e- X ) = C_/IX- ll + C_h+1X-h+l + ... , h~ 0, c_ It =F 0
and (!, 0 < (! < 1, be so small that this series converges uniformly and
absolutely for Ixl < (! (disregard the terms with a negative exponent).
Then we obtain, at first for ffis > h,

Multiplication by [T(S)]-l eliminates all the poles except possibly the


poles at s = h, h - 1, ... , 1, if h > 1. In this case there is a pole at s = h
because c_ It =F O.
00
248. The series I e-"Yn, l¥ > 0, is convergent. The integral F(u)
"=1
converges because (P(a + it) is bounded for all values of t. Term by term
integration [II 115] yields

F(u) = ~ ~-~
,,=0 2:rn.
J-
a+1 OO _

es(Vn-u)
52
_

+ e-s(V,,+u) ds.
a-seo

V
By virtue of 155 we have for m - 1 <u< m V

i.e. F(u) is a piecewise linear function whose derivative is

yuana97@uw.edu
360 Functions of One Complex Variable

V
in the interval m - 1 < u < m. If (]>(s) vanishes identically then V
F(u) = Oforu> 0, thus am + am+! +am+ 2 + ... = 0 for m = 1,2,3, ...
249. We have (]>(2k+!) (0) = 0; (]>(2k)(0) vanishes also because
d k
( z dZ) I(z)
00 00 00

= 1: a,.nkz" and lim 1: a,.nkz" = 1: a,.nk; i.e. (]>(s) = O.


,,=0 .-+1 .. =0 ,.=0

250. The function I(z) described in the problem has the following
properties
(1) I(z) is regular for Iz I < 1 because the integral converges absolutely
when mz : : :;
1; we have
"
a" = f
~
e-(X+xPcosP") sin (.xl' sin ,un) : ,- dx, n = 0,1,2, ... ;
o .
(2) an cannot vanish for n = 0, 1, 2, ... , because
Ie-(x+xPcosP") sin (.xl' sin ,un) I < e- x [solution 153];
(3) for Iz I < 1, the derivatives are

f
~

/")(z) = e-xPcospnsin (.xl' sin,un) e-x(l-:)x" dx, n = 0, 1, 2, ... ;


o
this integral converges absolutely and uniformly for mz < 1; thus, if z
tends to 1 along the real axis,

J e-XPCOS/l1' sin (x P sin,un) x,. dx =


~

lim I(")(z) = 0 [153] ;


'-+1 0
(4)
Ia I <
"
.!...
n10
j e-(x+rPcospn) x" dx <.!...
n!
+ .!...max
n!x~o
(e-(x+rPcospn) x .. + 2 ) j x- 2 dx,
1

hence [II 222]


. log la,,1
hmsup--- < -cos,un < O.
n~oo nil

'f' (1(. y;-- log x) . (Y;-IOg x+ 1(.)


The use 0 f H am b urger s unctIon, exp - (log x)2 + 1(.2 SIll (log X)2 + 1(.2 '
given in solution 153, instead of e-xPcosp" sin (x P sin ,un) shows in a
la,.1 log lanl
-7:-- cannot be replaced even by (logn)2 --V:-·
log
similar way that in 249
251. [G. P6lya, Problem: Arch. Math. Phys. Ser. 3, Vol. 25, p. 337
(1917). Solved by H. Priifer, K. Scholl: Arch. Math. Phys. Ser. 3, Vol. 28,
p. 177 (1920).] We assume

yuana97@uw.edu
Pt. III, Solutions 249-254 361

and that the series in question converges for z = a, i.e. that


Co + c1 + c2 + ... + c.. + ... converges, thus
IC"'+1I + C"'+k+1 + ... + c"'+k+ .. 1 < e for m sufficiently large, k,
n = 0, 1, 2, 3, ... Then
Ig(m)(z) + g(fII+1)(Z) + .. , + g(m+ ..)(z) I
=
I2:
00
~=O
(c"'+11 + C",+l +11 + ... + c",+"+1I) -k-I-
(z - a)
"I < eeI.-al .
252. Write g(z) as a series:
g(z) = a
o
+ a111 (z - z)
0
+ 21a (z -
2 Z )2
0
+ ... + nl
a.. (z - Z )..
0
+ ....
"-
2 1, ... , Via.. I, ...
-
The sequence lall, Vla is assumed to be bounded,
"-
lim sup
...... 00
Via.. I = A. There exists a number N to any given e, e> 0,
such that Ia.. I < (A + e)" whenever n > N, consequently
g(")(z) = \a.. + a"l~.!. (z - zo) + a"2~2 (z - zO)2 + ···1
( A + e) .. +1 (A + e) .. +:l
< (A + e)" +---11- Iz -zol + 21 Iz - zol2 + ...
= (A + e)" e(A +o*-zol.
Hence lim sup
n~~
VI g(") (z) I :s;; A = lim sup
"~~
Vt g(") (zo)l. That means the
limit superior in question is at no point z larger than at any other
point ZOo
253. [J. Bendixson: Acta Math. Vol. 9, p. 1 (1887).] The proof
proceeds along similar lines as the proof of 254. Notice

Q"+1(z) - Q.. (z) = rt (


r ... 1- a:) ~ 1'.. Ii (1 - a:)'
254. [Cf. N. E. Norlund: Differenzenrechnung. Berlin: Springer
1924, p. 210.J We consider the product p .. (z) defined in 13. Then we can
write

where 1'" and <5.. are constants. Let a and b be the two points of conver-

gence. The two series 2: A" and 2: B", (r"a + <5,,) P,,(a) = A",
n=O .. =0

yuana97@uw.edu
362 Functions of One Complex Variable

(y"b + 6,,) P,,(b) = B", converge. Since the series

i \P,,(Z) _
,,=1 P,,(IX)
P"_I(z)
P,,_I(IX)
1= i ,IP"-I(Z)
,,=1 P,,_I(IX)
~-=-IX:I'
IX - n
converges uniformly (p,,(z) -;.. ~:;rtZ) for IX =a or IX = band z an arbi-
trary point in any finite domain of the z-plane so do the series
00 P,,(z) 00 P,,(z)
2: A"p" (a)
,,=0
, "~B,,p,,(b)
[Knopp, p. 348]. We have however
00 z _ b 00 P,,(z) z _ a 00 P,,(z)
2: (y"z + 6,,) P,,(z)
"=0
= ----=-b
a
2: A" -P-(
,,=0 " a
) + - _b 2: B" P (b)
a ,,=0 "

255. According to VI 76 the polynomials are


c (z - 2) c"z(z - n)"-l
Q,,(z) = Co +C1z +~- + ... + n!

Let a be a point of convergence for the sequence Q,,(z), a =f= 0,


c"a(a - n)"-l
= a", 2: a" convergent. For n > Iz I, n > Ia Ithe following
n! ,,=0
product can be expanded into a power series:

(1- :r"+1(1- :)"-1=ea-z(1+~' +A~~ + .. -),


A', A", ... depend on a and z but not on n. The series with the general
term

(1- :r"+1(1- =),,-1 - (1- n: lr" (1- n: 1)" =n~:-:A:) + ...


converges absolutely, therefore

Co E a" az ( 1 -
+ ,,=1
00 a
-;- )-"+1 (1 - -;-Z)"-l .
[Knopp, p. 348] converges too.
256. Cf. 285. The theorem can be formulated more generally: The
functions t,,(z) are regular and different from 0 and the absolute value
of each is smaller than 1 in the region m. If at one point a of mwe have
lim t,,(a) = 0 then lim t,,(z} = 0 everywhere in
n~~ n~~
the convergence is m;
in fact uniform in every closed subdomain of m. To prove this cover m
with appropriately overlapping disks.

yuana97@uw.edu
Pt. III, Solutions 255- 260 363
257. [A. Harnack: Math. Ann. Vol. 35, p.23 (1890).] We denote
by v.. (x, y) the harmonic conjugate of u.. (x, y) and put

g..(z) = e-.... (x,)')-iv.. (X. y), z=x + iy, and I.. (z) = go(z) gl(Z) ••• g..(z).

Suppose that "1: u.. (x, y) diverges at some point Zo = Xo + iyo of 9l.
.. =0
Then we would find lim I.. (zo) = 0, consequently [solution 258]
"-+00
lim I.. (z) = 0 everywhere on 9l: contradiction.
"-+00
By using III 285 to a fuller extent one proves first that the conver-
gence is uniform in a closed disk inside 9l and centred at the presupposed
point of convergence. The region 9l is then successively covered with
appropriately overlapping disks.
258. Put I.. (z) = u.. (x, y) + iv.. (x, y). The imaginary part is given by

J----ax- dy -
X,),
au.. (x, y) au.. (x, y)
v..(x, y) = v..(xo' Yo) + -ay- dx,
the path of integration is any curve in 9l that connects the arbitrarily
chosen fixed point xo, Yo with the variable point x, y. The integral conver-
ges for n -+ 00 uniformly in any subdomain of 9l because the sequences
of the partial derivatives a:' a:
au au
converge in any subdomain of 9l [cf.
230]. The sequence v.. (x, y) converges therefore if and only if the sequence
v.. (xo' Yo) converges, in which case it converges uniformly in any sub-
domain of 9l.
259. Let /Zo, aI' az, ... denote arbitrary numbers. The identity

ao(l - a l ) + aoal (l - az) + aoalaz(l - as) + ...


+ /Zoal ... a.. _l (l - a..} = ao - aoalaz ... a..
implies

provided the last limit exists. Put


1
ao = 1, a.. = 2"-1 ' n = 1, 2, 3, ... [114].
1 +z
280. The power series

1+ i
.. =1
£x(£x + ~)"-l w"
n.

yuana97@uw.edu
364 Functions of One Complex Variable

has the radius of convergence e-t [214]; therefore the series in question
converges in any connected region of the x-plane where Ixe-% I < e-t
and there it represents an analytic function. The interval 0 :s;; x < 1 as
well as the interval 1 < x < 00 can be imbedded in such a region, one
in mt , the other in ~, but the intervals cannot both be imbedded in the
same region. According to 210 the infinite sum is equal to e"'% if x is
sufficiently small, thus it is equal to e"'% for x in Suppose that mt.
1 < x < 00 and that x' is defined by xe-% = x' e-%', 0 < x' < 1. The
series in question stays the same, therefore its sum is elU' =f= e"'%. - The
series converges also for x = 1 [220, (3)], its sum is e'" according to Abel's
theorem [I 86].
261.

I.. (z) =.t ([-:J --;)v' +


n 1: v·- 1
1.

,,=1
For mz > 0 we have
f ([~J- ~)x' dx + 1
]

.. ([ n ] )'1
n) ( v
lim I.. (z) = lim '~l -;- - n + 1 = ~_ xl
-;- -;- X

. . . "" . . . "" i (~)'-1 ~


.=] n n
f x·- 1dx
0

= z(z + 1) -1 C(z + 1) [II 45].


For mz < 0, mz = x we obtain

I.~ ([:J - :)v-I < ..~ ~ < {~~::1~, :~:: : =~:=f=

where A is independent of n. Furthermore

lim
.
1: V-- 1 = 1'-1 + ~-1 + ... + n·- 1 + .. , = C(l - z) =f= 0
"-+00,,=1

[VIII 48]. Hence in this case


lim I.. (z) = l.
...... ""
262. [G. Polya, Problem: Arch. Math. Phys. Ser. 3, Vol. 25, p. 337
(1917). Solved by H. Priifer: Arch. Math. Phys. Ser. 3, Vol. 28, pp. 179-
180 (1920).] We put
tp(z) - 1
tp(=) + 1 = VI(C) ,

yuana97@uw.edu
Pt. III, Solutions 261-263 365
then
q.>[q.>(z)] - 1 q.>{q.>[q.>(z)]} - 1 { }
q.>[q.>(z)] + 1 = tp[tp(C)J, q.>{q.>[q.>(z)]} + 1 = tp tp[tp(C)J .....
C+IX-{l
tp(C) = C1 + (IX - (l) C
and the statement becomes: the sequence
tp(C). tp[tp(C)J. tp{tp[tp(Cm.···
converges to 0 if ICI < 1 and converges to 00 if ICI > 1. converges to 1
or diverges if ICI = 1.
Let ICI = r. r < 1 and denote by M(r) the maximum of

I1 C+IX-{l I
+ (IX - (l) t
for ICI < r; we find M(r) < 1 [5J and monotone increasing with r [267J.
The inequalities
Itp(C) I <rM(r). Itp[tp(C)J I < Itp(C) I M[rM(r)J < r[M(r)J2.
Itp{tp[tp(C)]} I ::;: Itp[tp(C)J I M{r[M(r)J 2 } < r[M(r)]3.
follow from the definition of tp(C). We reason analogously if ICI > 1.
Ass ume finally ICI = 1. Then
Itp(C) I = Itp[tp(C) I = Itp{tp [tp(C)J} I = ... = 1. If the sequence in question
converges to a limit point Co' ICol = 1. we have
r r Co + IX - {J.
<'0 = <'0 1 + (IX _ (J.) Co' i.e. Co = 1.
263. Put (z - !)J: - 2,)~ - n) = P,,(z). We find on the positive
n.
imaginary axis

z=iy. y> O.IViyP,,(iy)1 2 =y(1 +Y;)(1 +Y;)"'(1 +~:)<si~~n~


2n
For fixed z we have
~ (-1)" P,,(z) n' = r(ll_ z)'

Consequently the sequence is bounded. e.g. in the half-disk ffiz > o.


Iz I ::;: 1. In the crescent
Iz - n I >n. Iz - n - 11 < n + 1
the absolute value of P,,(z) is larger than the moduli of
Pn-1(z), P"_2(Z) • ...• Pn +1(z) • ... [12]. On the inside border of the

yuana97@uw.edu
366 Functions of One Complex Variable

crescent, where Iz - n I = n, z = 2n cos pei'fJ, we find, if Iz I > I, that

IV-p ( )I I Vzr(z) I I Vi z'-i e-' I 'I'(zn)


Z "z = r(z _ n) r(n + 1) = (z _ n)'-"-i e-z+"n"+l e-"I e '

= I(z ~ nY liz ~ nIn+! e'l'(z,,,) = I(2 cos pe-itpyitp I' e'l'(z,n) ,


where tp(z, n) remains bounded for all pairs of values z, n. [For a proof of
Stirling's formula to the extent as it is used here ct. e.g. Whittaker and
Watson, pp. 248-253.] On the outside border of the crescent the same
estimate holds for P"+l (z) and P,,(z) because IP"+1 (z) I = IP,,(z) I on
Iz - n - 11 = n + 1. The modulus of P,,(z) increases as z moves from
the inner to the outer border of the crescent on a ray from the origin with
slope p, 0 < p < ; , because all the factors (z - I), (z - 2), ... , (z - n)
are increasing as can be seen by a geometric consideration.
264. [Cf. N. E. N6rlund,l.c. 254, p. 214.] On the inside border of the
crescent cut out by two lemniscates
2n2 cos 2p s: r2 < 2(n + 1)2 cos 2p
but in the exterior of the unit circle we have the relation

I z( 1 - ::) (1 - ;:) ... (1- ~:) = Ir(z~znt[;(:~) 1)]11


1

= I(Zz-n
+ n)zl.lzZ;--;; nll"+i e'l'(z,,,) = l(eitpV~ + l),itpl' e'l'(z,n)
n eUl'V2cos2/p-1

= I (e-itp V2 cos 2p + e- 2itp)i'P 12 ' e'l'(z,n) ,

where tp(z, n) is bounded [13, 263J.

265. The maximum of 1(1 )~ I


+ : along the ray z = reitp, p fixed,
is independent of n; put z = nC, C= eeitp. The maximum of

iIi log 11 + CI = +"3


1 ~
cos fP - ~ cos 2p cos 3p - ...

is obtained by differentiation with respect to eand some subsequent work;


it is equal to cos p and is reached for e = 0 if - : < p s: : . If
"4
:rt
< p < 7:rt
"4 t he maXlmum
. . tt' d h
IS a ame w en

1 e2+ecos/p
- 2" log (e2 + 2e cos p + 1) + el + 2e cos-~ + 1
= ffl (log ~- + 1 - ;;:_1_) = 0,
,+1 ,+1

yuana97@uw.edu
Pt. III, Solutions 264- 271 367

Ie + 11 > 1 and has the value


1 -
Tn ffi log (e + 1) = mffi1 (1 - ,-+1
1) .
Put = w; w then satisfies the equation Iwe-w+l I = 1; Iwi < 1,
:.._1_
C+ 1
1 w d h . . ..
cp = arg -=- = arg -1- - an t e maXImum m questIon IS
C -w
= \__l-w
w _\ ffi(l - w).

The discussion of the sign of the derivative can be replaced by the


examination of the domain in 116.- The appearance of convex curves
in 263-265 is no accident; d. G. P6lya: Math. Ann. Vol. 89, pp. 179-
191 (1923). [Also Math. Z. Vol. 29, pp. 549-640 (1929).]
266. [135.J
267. [135.]
268. The function I(z) = l(e-1 ) is regular in the open disk Ie I < ~
and M(,) is the maximum of I/(e-1) Ion the circle lei =~ [266, 267J.
r
269. The function/(z) is regular in the "punctured" plane Iz I > 0,
zn
including the point z = ex) [268].
270. [So Bernstein: Communic. Soc. Math. Charkow, Ser. 2, Vol. 14
(1914); M. Riesz: Acta Math. Vol. 40, p.337 (1916).] Apply 268 to
e- I( +2~-) [79J. We find for e =',' > 1" = a + b,
n

(C+-;:-1) .f (C+T1)1
/ ~2~
--cn~ < m:~ ---,,,--1 <
-2--
M.
In the case z _ ex) the proposition becomes: the maximum modulus of a
polynomial of degree 11 on the interval -1 < z < 1 is at least equal to
the absolute value of its highest coefficient multiplied by 2- n •
271. We may assume that the axes of El and E2 coincide with the
axes of the coordinate system and that the foci are z = ± 1. Let the
two ellipses be the images of the circles Iz I = '1 and Iz I = '2' 1 < '1 < '2'
under the mapping z = ~ (e ++), then '1 = a1 + b '2 =
1, a2 +b2•

Noting 268 we proceed as in 270.


The extreme case in which El degenerates into the twice covered
real segment [-1, IJ leads to proposition 270. If the two foci coincide
we obtain two circles and the problem is the same as 269.

yuana97@uw.edu
368 Functions of One Complex Variable

m. We may assume without loss of generality that 1(0) is real and


1(0) > 0; put I(z) = l(eilJ) = U(e, {}) + iV(e, {}), then

The upper bound for 1(0) is reached if Vee, {}) = 0 for 0 < {} < 2n, hence
=
I(z) 1(0) [230].
m. [134.]
274. The function I(z) = :~:~ is regular for Iz I < 1. It is also
regular on the unit circle, where I/(z) I = 1 unless lp(z) = O. If Zo is a zero
of lp(z) it is also a zero of 91(z) and with the same multiplicity [otherwise
z = Zo were a zero or a pole of I(z) which is impossible: at other points
of the unit circle, arbitrarily close to zo, we have I/(z) I = 1]. We drop
the common factors of 91(z) and lp(z) and so obtain the regular function
I(z) which is different from zero in the closed disk Iz I ~.1 and whose
modulus is equal to 1 on the unit circle, I/(z) I = 1 for Iz I = 1, therefore
[138] I(z) =
c, Ic I = 1. Since 91(0) and 11'(0) are real and positive we have
c = 1.
275. The absolute value I/(z) I is a real continuous function in ~;
it assumes therefore its maximum in ~. This is impossible at an inner
point if I(z) is not a constant [134].
276. [Cf. E. Lindelof: Acta Soc. Sc. Fennicae, Vol. 46, No.4, p.6
(1915).] A rotation through 2nv around a point C maps the domain onto
n
the domain ~.and the set 58 onto the set 58., 'V = 0, 1, 2, ... , n - 1,
580 = 58, ~o =~. The intersection 3 (largest common subdomain) of
the domains ~o' ~l' ••• , ~.. _) contains Cas an inner point. Those inner
points of 3 that can be connected with ~ by a continuous curve in the
interior of 3 form a region 3*. The boundary of 3* consists, according
to the hypothesis and the construction, of certain points of the sets
580 , 58), ... , 58.. _). The absolute value of the function I(C + (z - C) £0-')
is < A at all the boundary points of 3* [275] and < a at those boundary
points of 3* that belong to 58•. The absolute value of the function
I[C + (z - C)] I[C + (z - C) £0-1] ... I[C + (z - C) £0-"+1]
is therefore not larger than aA"-l at all the boundary points of 3*,
consequently [275] also at the inner point z = C.

yuana97@uw.edu
Pt. III, Solutions 272- 281 369

2n. [Cf. E. Lindelof, l.c. 278. J Assume IX < n without loss of


generality because we can consider f(#), with (J suitably chosen, instead
of f(z). Draw a circle around an arbitrary point of the ray arg z = !(IX - 6)
that is tangent to the ray arg z = IX. The chord cut by this circle from the
real axis is always seen from the center under the same angle. Apply
276 where Cl) is identified with the portion of the disk in the upper half-
plane and \8 with the chord on the real axis; thus lim fez) = 0 along
arg z = !(IX - 6). By modification of the conclusion we show that the
convergence lim fez) = 0 is uniform in the sector 0 < arg z < !(IX - 6).
Repeat the argument for the rays
3 7 15
arg z = 4" (IX - e), 8 (IX - 6), 16 (IX - 6), ...

278. [Cf. E. Lindelof, l.c. 276. J Let R denote the least upper bound
of If(z) I in ffi. There exists at least one point P in ffi or on the boundary
of ffi so that in the intersection of ffi with a sufficiently small disk around
P the least upper bound of If(z) I is equal to R.
If there is no such point P in ffi then If(z) I < R in ffi, but then there
exists a boundary point P with the required property. According to
condition (3) we have R ~ M, i.e. If(z) 1< Min ffi.
In case there exists at least one point P, z = zo' of the described type,
then If(zo) I = R. Along a sufficiently small circle around Zo we have
If(z) I < R, thus, according to 134, fez) = constant.
279. [Po Fatou: Acta Math. Vol. 30, p. 395 (1905).J Put w = e2ni/".
If n is sufficiently large then

lim fez) f(wz) f(w 2 z) .•• f(w"-J z) = 0,


r->1-0

the convergence is uniform in the unit disk, 0 ~ {} < 2n. The function
fez) f(wz) f(w 2z) .. ·f(w"- l z) vanishes identically according to 278. [This
proposition is not an immediate consequence of 275.J
280. [R. A. Schwarz: Gesammelte mathema1ische Abhandlungen,
Vol. 2. Berlin: Springer 1890, pp. 110-l11.J Apply 278 to the function
f(z) which is regular in the disk Iz I < 1.
z
281. [E. Linde16f: Acta Soc. Sc. Fennicae, Vol. 35, No.7 (1908).
Concerning problems 282-289 d. P. Koebe: Math. Z. Vol. 6, p.52
(1920), where also ample bibliography is provided.J We denote by
C = tp-1(W) the inverse function of w = tp(C). The function

F(C) = tp-1{f[q>(C) J}

yuana97@uw.edu
370 Functions of One Complex Variable

satisfies the conditions of 280. Hence IF(C) I ::;: e for ICI ::;: e; there is
equality only if F(C) = e''''C, ()(. real. The inequality states that the points
F(C) lie in the disk IC I ::;: e, that means that the points tp[F(C)) = l[tp(C)]
lie in the domain 9; z = tp(C) represents an arbitrary value in t. In the
extreme case we have tp(e''''C) = l[tp(C)), i.e. I(z) = 'I' [e''''tp-l (z)], where
C = tp-l(Z) is the inverse function of z = tp(C), ()(. is real. This is the most
general function that maps 9l univalently onto @) and z = Zo onto
w = Wo [IV 88].
282. [C. Caratheodory: Math. Ann. Vol. 72, p.107 (1912).] Apply
281 to the following special case: 9l: the disk Iz I < 1; @): the disk
Iwl < 1; Zo = 0, Wo = 1(0),
tp(C) = C, (") C+ Wo
'1''' =l+iiioC·

The subdomain t is the disk Iz I < e, t is the image of ICI < e under the
function w = tp(C), thus 9 is also a disk. The points of 9 satisfy the relation
IW - -ICI (1 - !Woll ) <
Wo I -
1 - IWol 1
11 + iii~CI = e 1 - IWo I (I •
The inequality becomes an equality only if I(z) = tp[e''''tp-l(z)] = tp(e''''z) =
.'" _+ ~o , in which case 11
=~
1 + woe""::
+ woe''''zl = 1 -lwollzl, i.e.
arg z = arg Wo - ()(. + :rt.
283. Apply 281 to the following special case: 9l: disk Iz I < R;
@): half-plane 9lw < A (R); Zo = 0, Wo = 1(0), 9lwo = A (0),

tp(C) = RC, tp(C) = Wo + [iii;:= ~A(R)] C = Wo + [wo + Wo - 2A(R)] I ~ C.


t is the disk Iz I < eR = r, 9 is the image of ICI < e under the mapping
w = tp(C). The points of 9 satisfy the inequality
- C
9lw = 9lwo + [wo + Wo - 2A(R)] 9l 1 _ C< 9lwo - 2 [9lwo - A(R)] T+ (I
(I

1-(1 2(1
= 1 + (l 9lwo + 1 + (I A (R).
It is an equality only if I(z) = tp[e''''tp-l(z)] = 'I' (e"" ~).
284. The following relation [solution 283] holds

Iwl::;: IWol + [2A(R) - Wo - wo] 1~ (I = M(O) + 1 ~ (I [A(R) - A(O)],


which is a weaker statement than 236.
2BS. Apply 283 to log I(z) :
9l log I(z) = log I/(z) I ::;: log M(r).

yuana97@uw.edu
Pt. III, Solutions 282- 289 371
286. Proposition 285 implies
2!~
Iln(z) 12 < 1/,,(0) 1 Hizi
For Iz 1< ~ the exponent is 2 ~ ~ 1:1> 1, consequently Iln(z) 12 < 1In (0) I.
287. Use 281: ffi: the disk Izi < 1; 6: the half-plane ffiw> 0;
Zo = 0, Wo = 1(0) > 0,
1+C
9'(C) = C. 'P(C) = Wo 1 - C•

t: the disk 1z 1 :::;; e, s: the disk whose boundary circle intersects the real
axis orthogonally at the points Wo ~1 + Q and Wo e . The radius of this
11 -
-e +e
circle is Wo 1 ~e {!2 • The points of s satisfy the inequalities

Wo
1- e <
1 + e=
ro
<JtW
<
= Wo
1+ e
1 _ e'
I~ 1<
",w = Wo 1 -
2e
(i '

Wo
1-
1 + ee <1 I:::;; l+e
= W - Wo 1 _ e'

We have equality only in the case where I(z) = 1p[ei "'9'-l(z)] = 1p(i"'z),
IX real.

288. Special case of 281: ffi: the disk 1z 1 < 1, 6: the vertical strip
-1 < ffiw < 1, Zo = 0, Wo = 0,
2 l+C
9'(C) = C, 'P(C) = -:-log
t:rr;
-1 -
f" ;
~

t is the disk 1z I < e. The points 1C1< eare mapped by ~ ~ ~ onto the disk
that intersects the real axis orthogonally at the points 1 + e and 1 - e.
1-e l+e
This disk lies completely in the sector whose bisector is the real axis
and whose center angle is 2 arctan -2e
1 2 = 4 arctan e. Therefore s lies
-e
<~ arctan e. Besides we have in S
completely in the strip 1 ffiw 1 -:rr;

13wl= ~IOgl~~~I< ~log~~:.


Equality is attained only for I(z) = 'P[ei "'9'-l(z)] = 1p(ei "'z), IX real.
289. We may assume that R = 1, LI = 2, and 1 ffi/(z) 1 < 1 for
I z 1 < 1 as in 288; 1(0) = Wo however is arbitrary in the strip -1 < ffiw < 1.

Depending on the choice of Wo each disk Iz I < e < 1 is mapped onto a


domain that contains the range of I(z). We have to consider the maxi-
mum width of these domains in the direction of the real and the imaginary
axis respectively while Wo moves over the entire strip. It is obviously

yuana97@uw.edu
372 Functions of One Complex Variable

sufficient to examine only real wo-values, -1 < Wo < 1. In this case


in",. imf1,
2 e 2 +iC 2 1+ie--2-C
w = tp(C) = -:-log
"I'
.
''''0
= Wo + -:- log
'31'
.
'1mB.
.
1-ie 2 -·C 1-ie-2-C
The image of 1C1 = e in the w-plane is convex [318], moreover it is
symmetric with respect to the real axis because tp(C) is real for real C-
values. Thus mw assumes its maximum and its minimum for real C= ±e.
The width in the horizontal direction is therefore

2 1+ie--2-e
imllo (
2 imllo
1- ie--2 -e )
wo+-:-Iog
'31
..
1mt'o
- wo+-:-Iog-
In
.
1mB,
1 - ie-2-e 1 + ie-2- e
nwo
ecosT nwo
ecos- )
4 2-
= - ( arctan + arctan .
n . nwo . nwo
1 + esm·2 1- esm-2
As Wo varies between -1 and +1 the derivative with respect to Wo has
aIways the same sIgn . as -SID . nw
T' · ·IS equa1to -;:
oThe maxImum 8
arctan e
and it is reached for Wo = O.
The oscillation of 3w cannot be larger than twice the least upper
bound of 13w 1given in 288; proof similar.
290. [H. Bohr: Nyt. Tidsskr. Mat. (B) Vol. 27, pp. 73-78 (1916).]
We choose 'YJ so that I'YJI = 1, 'YJF(l) > O. Picking out the branch of
log 'YJF(z) that is real for z = 1 we set
w = I(z) = log 1JF(z) - log c .
log 1JF(1) - log c

We have 1(1) = 1, ml(z) > O. Apply 281 by identifying mwith st, ~


with the right half-plane, Zo with 1, Wo with 1. The functions z = 9'(C),
w = tp(C) are supposed to be normed in such a way that they transform
real C-values into real z- and w-values; in addition let 9'(1) = ex>, '1'(1) = ex>
[IV 119]. Hence
l+C
tp(C) = 1 - C'

Assume x > 1 and x = 9'(e), 0 < e < 1. According to 281 the range
of I(z) in the w-plane is contained in the image of the disk 1C1~ e under
the mapping w = tp(C). The image of the disk is a disk whose points are
at a distance of at mosttp(e) from the origin: i.e. I/(x) 1~ tp(e). Hence
we can define hex) = tp[9'-I(X)].

yuana97@uw.edu
Pt. III, Solutions 290- 294 373

291. [K. Lowner.J According to 280 we have I/(z) 1 < Izl for Izl < 1;
hence for positive z, 0 < z < 1,
.!:. -
I 1-z
J(z) 12 1.
-
By taking the limit z ~ 1 we find
11'(1) 1> 1.
Assume that arg I' (1) A sufficiently small vector with the direction
3: '
= IX.

1}, ~ < 1} < attached to z = 1, points towards the inside of the


unit circle and is mapped by the function w = I(z), /'(1) =f: 0, onto an
analytic curve segment through the point w = 1 with the direction
1} +
IX. Since I/(z) 1< 1 for 1z 1< 1 the direction is restricted by the
condition
..::2 < 1} + IX < 3n2
for any admissible 1}. Thi.s is possible for IX = 0 only. Cf. 144.
292. Let Zo be fixed, 1Zo 1<.1, I(zo) = wo, 1wo 1< 1. Choose the
constants e and 'YJ so that e 11. - :'0 = 'YJ 11 - ~o = 1, 1e I = I'YJ I = 1. Put
-zo - Wo

z - Zo
e- Z 'YJ W - Wo W.
1-_- = ,
- zoz
-1--- =
- wow
The function W = F(Z) defined with the help of w = I(z) satisfies the
hypothesis of 291. Therefore

1 <F'(I) = (dW) (dW) (dZ) = 1)(1 - Iwor) 1'(1) _~ FO)2 .


- dw w=l dz z=l dZ Z=l (1 - w o) e(1 - IZoI2)
293. [G. Julia: Acta Math. Vol. 42, p.349 (1920).J Let Zo be fixed,
3zo > 0, I(zo) = wo, 3wo > O. Choose the constants e and 'YJ so that
_ 0 = 1, el =
- _0 = 'YJ-
ea-z b-w 1 1'YJ 1 = 1. Define
a-zo b -Wo

Then [291, 292J


1 < (dW)
dZ Z=l
= (dW)
dw w=b
I'(a) (dz)
dZ Z=l
= 1)(wo - iii~
(b - wo)
I'(a) (a - ZO)2 •
e(zo - zo)
1 - zlz 1 - zzz 1 - znz
294. The function I(z) - - . . .. - - is regular in the
z- zl z - Zz Z - Zn

disk 1z 1 < 1, its absolute value is smaller than M +


e, e > 0, sufficiently
close to any boundary point [5J. Apply 278.-Different proof by careful
application of 176.

yuana97@uw.edu
374 Functions of One Complex Variable

-+ -+
295. The function t(z) ~!...--.:. ~ ... -"-- is regular in the half-
Z+Z

Zl - Z Zz - Z Z .. - Z

plane mz > 0; its absolute value is smaller than M + e, e > 0, suffi-


ciently close to any boundary point [6]. Apply 278.-A different proof is
based on 177. In fact, both methods go beyond the particular case of
the half-plane mz> 0; both can be easily adapted to a generalized
proposition which relates to 294 as 281 relates to Schwarz's lemma 280.
296. The function t(z) is assumed to be meromorphic with the zeros
aI' a2 , ••• , am and the poles b1 , b2 , ••• , b.. (counted with proper multi-
plicity) in the disk 1z 1 < 1 and It(z) 1 = c > 0 for 1z 1 = 1. The function

t(z) nm 1-az

1'=]
--I'
Z - ap
n ___
.. z-b
.=1 1 - b.z
v = <p(z)

is regular and non-zero for Iz 1 < 1, its modulus is constant, equal to c,


for 1z 1 = 1. Therefore <p(z) is a constant [142J.
297. [W. Blaschke: Sber. Naturf. Ges. Lpz. Vol. 67, p. 194 (1915).]
Let IX be real, 0 < IX < 1, t{lX) =j= O. 294 implies that the product

.ii I:.--~.I does not diverge to O. Hence the series

2 1 + L.:
1_IX
IX
OO

(1 -
Iz.I)
.=1
converges.
298. Let IX be real, IX > 1, t{lX) =j= O. Proposition 295 implies that
the product iI I
.=1
z. - IX \ does not diverge to O. Consequently the series
z. + IX

Iz
L.: 1-~ 12) =L.:-~2~>
IX z +Z • -.
~ zL.:m-
1
'=1/1 + : I 2z.z. -
00 ( 00 00

.=1 Iz. + IX (1 + IX) .=1 z.

converges.
299. [T. Carleman; cf. also P. Csillag: Mat. phys. lap. Vol. 26,
pp. 74-80 (1917).] Let Zo be an inner point of '1) and put It.{zo) 1 = e.t.(Zo)'
'11= 1, 2, ... , n. (In the case t.{zo) = 0 we choose e. = 1.) The function
F(z) = edl{Z) + e2t 2 (Z) + '" + e..I.. {z)
is regular and single-valued in '1); F(z) assumes its largest absolute value
at a boundary point ZI of '1). Hence
<P(Zl) > IF(Zl) 12 IF(zo) I = <p(zo)'

yuana97@uw.edu
Pt. III, Solutions 295-302 375
300. It is more convenient to prove the statement as follows than
to refine the proof of 299: Let Zo be an inner point of il and the disk
1z - Zo 1< r be inside il. Addition of the inequalities [272]
1 2,.
(*) 1I. (Zo) 1 ~ 2n f 1/.(zo + re',')) 1dO, " = 1, 2, ... , n
o
yields
1 2,.
fI/(zo) < 2n f
o
fI/(zo + re',')) dO.
If one of the summands (*) satisfies a strict inequality the sum satisfies
a strict inequality; i.e. if at least one of the I:s is not a constant the
maximum cannot be attained at an inner point zoo
301. [G. Szego, Problem: Jber. deutsch. Math. Verein. Vol. 32,
2. Abt., p. 16 (1923).] It is sufficient to prove that rp(P) attains its
maximum on the boundary of any plane domain il whereby the points
PI' P2 , ••• , p .. are not necessarily in the same plane. We introduce an
orthogonal coordinate system in il, x, y; z = x +
iy. Now we have to
show: A function of the form
.
n (Iz -
.=1
a.12 + b;) I,
where a. are arbitrary complex, and b. real, constants, " = 1, 2, ... , n,
attains its maximum on the boundary of any domain of the z-plane.
Work out the product [299].
302. Let Zo be an inner point of il, where some of the functions,
called I,..(z), do not vanish and other functions, called I.(z), do vanish.
(One or the other type may be absent.) Let the radius r be so small that
the disk 1z - Zo 1< r lies completely inside il and does not contain any
other zero of the functions I.(z) but ZOo The functions 1,..(z)P,.. are regular
in this disk. According to 299 there exists a point Z1' 1Z1 - Zo 1 = r such
that

,.. 1I,.. (Z1) I '" > J:


J: 1/,..(zo) IP"',
P
,..
Obviously
J: 1I. (ZI) IP' > 0= J: 1I. (zo) IP',

i.e. fI/(Z1) > fI/(zo); in fact fI/(Z1) > rp(zo) if at least one I,..(z) is not a constant
or, in the other case, if at least one I.(z) does not vanish identically. Since
il is closed and fI/(z) continuous there exists a point in il where the func-
tion fI/(z) assumes its maximum: it cannot be an inner point except in the
particular case mentioned in the problem.

yuana97@uw.edu
376 Functions of One Complex Variable

303. Since 'l) is closed the function Ij(z) I, which is single-valued and
continuous, attains its maximum in 'l). Proposition 134 shows that this
cannot happen at an inner point of 'l) except in the case where j(z) is a
constant.
304. [J. Hadamard: Bull. Soc. Math. France Vol. 24, p. 186 (1896);
O. Blumenthal: Jber. deutsch. Math. Verein. Vol. 16, p.108 (1907);
G. Faber: Math. Ann. Vol. 63, p. 549 (1907). Hurwitz-Courant, pp. 429-
430; E. Hille, Vol. II, pp. 410-411.] The function ZXj(z) is not single-
valued in the annulus r 1 < Izl s r 3 , its modulus however is. Hence the
maximum of 1ZXj(z) 1is either 1 M(r1 ) or raM(rs) [303]. Choose (X so that
(*) 1M(r1) = r;M(rs)'
Considering a specific point on the circle 1z 1= r2 we see that
~M(r2) S 1M(r1) = r;M(rs)'
We introduce the value (X from (*). (The condition that j(z) be regular
and Ij(z) 1single-valued on the punctured disk 0 < Izl < R is sufficient.)
305. The maximum of ZXj(z).is reached at a point of the circle 1z I = r2 ,
i.e. in the interior of the annulus r1 :::; 1z 1< r3, only if ZXj(z) is a constant.
306. Put j(z) = ao + a1z + a2z2 + ... + a"z" + .... The integral
12 (r) becomes
""
12(r) =Iao 12 + la112 r2 + la212 r4 + '" + la,,1 2 r2" + ... = ~ p"r",

where p" > 0 and at least two p,.'s are non-zero [II 123].
307. Assume that j(z) is not a constant and that none of the zeros
Z1' Z2' ... , z,. of j(z) in the disk 1z 1< r coincides with z = 0 (for simplicity's
sake assume also j(O) = 1). Then we have [120]
log &(r) = nlogr -loglz11-log IZ21-'" -log Iz"l.
Hence the graph of log &(r), as a function of log r, consists in a sequence
of straight pieces with monotone increasing slopes. The change of slope'
for log r = log ro is caused by the appearance of additional zeros on the
circle Iz I = roo The increase in slope is equal to the number of such zeros
counted with appropriate multiplicity.
308. [G. H. Hardy: Proc. Lond. Math. Soc. Ser. 2, Vol. 14, p. 270
(1915).] Suppose 0 < r1 < r2 < r3 < R. Define the functions e(&), F(z)
by the relations
2"
e(tt) j(r2eiD ) = Ij(r2eiD ) I, 0 < tt < 2n, F(z) = 2~ f j(ze ifJ ) e(tt) dtt.
o

yuana97@uw.edu
Pt. III. Solutions 308-312 377

The function F(z) is regular in the disk Iz I < 1'3 and its absolute value
teaches the maximum on the boundary, say, at the point 1'se"·. Hence
1(1'2) = F(1'2) :::; IF(1'se'''·) I :::; 1(1's)'
that means 1(1') is not decreasing. Determine the real number IX by the
equation
1'~1(1'1) = r31(1's) •
The absolute value of the function z"'F(z), which is regular in the annulus
1'1:::; Izi :::; 1'3' is single-valued. Hence [303]
1'21(1'2) max Iz"'F(z) I :::; 1'i1(1'1) = 1';1(1'3),
= r;F(1'2):::;
'.=01'1=0'.
hom which the convexity property of 1(1') follows [304].
2,.
309. l(1') = f
1f'(1'e''') 11' d{) [308].
o
310. Define e2 ,.iP/" = 00., ." = 1, 2, ... , n. Let 0 :::; 1'1 < 1'2 < R. There
exists [302] a point ,rae'''·, on the circle Iz I = 1'2 such that

n1 .=1
" 1 "
1: 1/(1'1 00.) II> < n 1: 1/(1'awi"·) II>·
.=1
As n -t 00 this inequality becomes
11>(1'1) :::; 11>(1'a)'
Assume 0 < 1'1 < 1'2 < 1'3 < R, IX real. The functions
'" '" '"
zl> l(w 1z), zl> l(w2z), ... , zl> I(w"z)
are regular in the annulus 1'1:::; Izl < 1'3' however, only their moduli are
necessarily single-valued. All the same [303, 302] we may conclude that
the sum of the p-th power of their absolute values assumes its maximum
on the boundary of the annulus. Applying the same arguments as in
304, 308 and taking the limit we establish the behaviour of 11>(1') with
regard to convexity. -Cf. II 83 for the limit cases p = 0 and p = 00.
311. We may assume that the center of ~ is at the origin. Apply 230.
The proposition states in other words: A harmonic function that is
regular in a closed disk and that vanishes on the bounding circle vanishes
identically.
312. We denote by u(x, y), z = x + iy, a harmonic function that is
regular in the disk (x - xo)2 + (y - YO)2 < 1'2. The value of u(x, y) at
the center is
1 2,.
u(xo, Yo) = 2
no
u(xoJ + l' cos {), Yo + l' sin {) d{} [118],

yuana97@uw.edu
378 Functions of One Complex Variable

consequently
1 2,.
IU(Xo. Yo) 1 < 21f f IU(Xo + l' cos {}. Yo + l' sin (}) 1 d{}.
o
The inequality becomes an equality if
1 2,.
-2
1fo
f [Iu(xo +1'cos{}. Yo +1'sin{}) 1± u(xo +1'cos{}.Yo +1'sin{})]d{} = O.

where the sign depends on whether u(xo• Yo) > 0 or u(xo, Yo) :$; O. The
integrand must vanish identically. i.e. u(x. y) cannot change sign on the
given circle (u(x. y) possibly becomes 0 at some points).
313. Suppose that the point xo. Yo at which the maximum is reached
is an interior point of 'il. Choose r so small that the disk with radius l'
and center xo. Yo lies in the interior of 'il. The equation
1 2,.
-2-
1fo
f [u(xo. Yo) - u(xo + l' cos {}. Yo + l' sin (})] d{} = 0 [solution 312]

implies
u(xo, Yo) - u(xo + l' cos {}, Yo+ r sin (}) = 0, 0 < {} :$; 2n,
i.e. [311] u(x, y) = const.
314. Follows from 313.
315. log Iz - zll + log Iz - z21 + ... + log Iz - z.. 1= ~ log P(z) ,
which is the potential of the system of forces in question, is a harmonic
function and as such it does not have a maximum nor a minimum at a
regular point. Stability would require a minimum of potential energy.
316. Remove the finitely many exceptional points from 'il by enclos-
ing them in circles so small that these disks have no points in common
and do not contain any point at which the function reaches its maximum
in 'il. Apply 313 to the remaining domain.
317. According to 188 the orientation of the image of the circle
1z 1 = R is preserved. Therefore 109 can be applied. The harmonic func-

tion
~/'(z)
I(z)

which is regular in the disk 1z 1 < R [f(z) being schlicht has the only and
simple zero z = 0] is positive for I z 1 = R. Hence it is positive on any
smaller concentric circle Iz/ = l' < R [313]. The images of the circles
/z 1 = r are star-shaped with respect to the origin according to 109.

yuana97@uw.edu
Pt. III, Solutions 313-322 379

318. According to 188 the orientation of the image of the circle


1.:1 = R is preserved. Thus we can apply 108. The harmonic function
f"(z)
mz j'(z) + 1,
which is regular on the disk I.: 1< R [1' (.:) =1= 0 because 1(.:) is schlichtJ
is positive for I.: I = R. Hence it is positive on any smaller concentric
circle 1.:1 = l' < R. According to 108 the images of the circles 1.:1 = l'
are convex. The statement is now proved in the case where the inner
circle is concentric to the disk 1.:1 < R.
Let the inner circle lie anywhere inside the disk I.: I < R. We build
up w = 1(.:), the given function, by combining two functions: a linear
mapping E= 1(.:) of the disk I.: I s;;: R onto itself whereby the given inner
circle is transformed into one with center at the origin E= 0 [77] and a
second one w = g(E) = l(l-l(E»); in fact, W = 1(.:) = g(l(.:»). By considering
W = g(E) we reduce the problem to the previously discussed special case.
319. Proof analogous to the proof of 299.
320. [Cf. A. Walther: Math. Z. Vol. 11, p.158 (1921).] Let u(x, y),
.: = x + iy, be the harmonic function in question. The function
u(x, y) + IX log 1', IX arbitrary real constant, is regular in the annu-
lus 1'1 s;;: I.: I <1'3. Its maximum there is either A (1'1) + IX log 1'1 or
A(1'3) + IX log 1'3. Define IX so that
[313,304].
321. [Cf. A. Walther, l.c. 320.] (1) The three circle theorem proved
in 320 holds also for harmonic functions that have in the disk 1.:1 < R
finitely or infinitely many isolated singularities provided their accumu-
lation points are not.interior points of the disk and that, in addition,
the function tends to - 00 as .: approaches a singular point [316J.
Apply this generalized three circle theorem to the function
mlog 1(.:) = log I/(.:} I·
(2) Let u(x, y), .: = x + iy, be a regular harmonic function in the
disk 1.:1 < Rand v(x, y) denote the conjugate harmonic. Now apply
304 to the function 1(.:) = e*,Yl+iv(x,y) ; 1/(.:) 1= e*'Y), hence log M(1') = A (r).
322. [Concerning the method used and the problems 322-340 ct.
in the first place E. Phragmen and E. LindelOf: Acta Math. Vol. 31,
p.366 (1908); then P. Persson: These (Uppsala 1908) and E. Lindelof:
Rend. Circ. Mat. Palermo Vol. 25, p. 228 (1908); also E. Hille, Vol. II,
pp.393-397.]

yuana97@uw.edu
380 Functions of One Complex Variable

323. The arcs may be replaced by any continuous curves that cross
the sector connecting the two rays. Their minimal distance from the
origin must increase beyond all bounds.
324. All the estimates of solution 322 remain valid in the domain
bounded by r 1 and r'l..
325. Cf. solution p. 168.
328. The function e0>61(z) satisfies the conditions (1), (2) and the modi-
fied condition (3) of 325 [final remark in the proof] however large OJ.
Consequently
I/(z) I ~ e-O>% for 9lz = x ~ O.
Let OJ increase to infinity.
327. Put arctan x : 1 = 1p. Then

z log(z + 1) = (1' cos {} + il' sin {}) [! log (1''1. + 21' cos {} + 1) + i1p],
hence for - ; ~ {} < ; , l' > 1,
m[ -z log (1 + z)] = 1'1p sin {} - il' log (1'2 + 21' cos {} + 1) . cos {}
nAn
< 1'2" - rlogl' COS·v ~ 1'2".

Let 0 < (J < 2,.n . The function


~ I(z) e- II•Jog (.+1l

satisfies the conditions (1), (2) and (3) of 328 with Ii = O.-Instead of
quoting 328, 325 we could apply the ideas developed in those solutions
to the function

+ 1) -
I(z) exp (OJz - {Jz log (z ee- iAn!4- zA) •

Then we set e = 0 and finally OJ = + 00.

328. [F. Carlson: Math. Z. Vol. 11, p.14 (1921); These, Uppsala
1914.] First solution. 327 can be applied to -!(z) . The existence of
smnz
an inequality
! !(z)
:smnz
1< A'eBlzl

(with A' > A) is best established first outside and then inside the circles
Iz - n I = i, n = 0, 1, 2, ... [solution 165].

yuana97@uw.edu
Pt. III, Solutions 323 - 332 381

Second solution: 178 and the fact that the terms on the left hand
side are positive imply

~ Ii - n2 ~ 2n1J"(1e2 - n21) [2 log C + 2(n -


"(11-') y) e] de + C,•
p-l 1

where C and C' denote constants. The left hand side is N log n, the right
hand side N n - " log n: contradiction. This method can be generalized
n
[F. and R. Nevanlinna, l.c. 177].
329. Assume e > 0; the function q:>(z) =

_ 6- is regular in the half-
plane 9h > O. We have [f(z)t

Iq:>(z) I ~1 for mz = 0 and Iq:>(z) I ~1 for Iz I = r,


if r is so large that w(r) > ! .hence Iq:>(z) I ~ 1 in the entire half-plane
»lz;;::: 0; and finally, as e -+ 0, le'l < 1: contradiction. (Borderline case
of 290: The region % becomes a halfplane.)
.IJ<+fJ
-,-
330. Assume e> 0, h = ee 2, 0 < (1 < «5, (1({J - IX) < n. Apply
the reasoning of 322 to the function

F(z) = I(z) exp (_ (hz)fJ~IJ< -a) .


There results the conclusion: IF(z) I < 1 in the entire sector. Let e
converge to O. - The proposition could also be reduced to theorem 322
with the help of a function that maps the sector IX ~ arg z ~ {J onto the
n ~(/l - IX)
sector -y < arg z < y, y ="2 - 2 ' and leaves z = 0 and z=oo
unchanged.
331. In the special case IX = - ; , (J = ; the statement is weaker
than 325. The proof involves the function
f .fJ+IJ<,.,. )
I(z) exp \ - 7Je -, fJ-IJ< 2" zfl-IJ< •

Start by proving with the help of 330 that the maximum of the modulus
of I(z) on the bisector is not larger than 1 [325].
332. Assume Ig( -r) I < C: apply 331 to the function g~) in the
sector -n < {} < n; this leads to Ig(z) I ~ C in the entire plane, thus
g(z) is a constant.

yuana97@uw.edu
382 Functions of One Complex Variable

We can also examine


8

g()
Z e
-'1V-;-e( -il) J:.

This function is analytic in the open sector 0 < {} < 11: and continuous
in, the closed sector 0 <{}< 11:. Cf. 325.-Also 325 can be applied to
C-Ig(Z2). - The function SiV~:;- illustrates how strong the theore:rp. is.
333. Let a < b < 1. The absolute value of the test function c b•
Ieeb(x+iy) I =eebxcos by ;
cosb~ 1 A
thus on the boundary of mit is > e 2 > 1. Let 1> b _ a log -----n
ecosb2"

Then we find on the boundary of the rectangle 0 < x < 1, - ; ::;: y < ;
the inequality

334. Assume the con trary: consider the function


cp(z) = c' U(z)] -6, e>O,z=x+iy,
in the rectangle

where Xl is chosen so that ew(xI) > 1. On the boundary of this rectangle


we have
Icp(iy) I < eCosy • e-ew(O)< e, Icp (x ± i ~)I < 1,
ICP(XI + iy) I ::;: er'[cosy-ew(x,il < 1;
hence in the interior Icp(z) I < e. As e -+ 0 we obtain
I eel i < e,
thus e.g. ee < e: contradiction. The role of the function ee' is explained
by 290 and 187.
335. It is sufficient to consider the case hinted at in the problem
[linear transformation]. Take e > 0; the function

q;(z) = j(z) !2 ( ~r r
Z Zv

yuana97@uw.edu
Pt. III, Solutions 333- 339 383

is regular in the intersection 91, of 91 and the disk \z \ < r and its modulus
is single-valued in 91, [303J. The maximum of I/(z) Ion the circle \z\ = r
is denoted by M(r). Taking into account the condition on I(z) on the
boundary of ffi [278J we find that in 91,
q;(z) < max [M, M(r)J.
Hence for e -+ 0
\/(z) \ < max [M, M(r)].
If, in particular, r' is also admissible [see hint] and r' < r, then we have
M(r') ::;;; max [M, M(r)J.
On the other hand [268]
M(r) < M(r').
Hence the alternative: Either M(r) > M, thus M(r) = M(r'), I(z) = oonst.
[268] or M(r) < M, \ t(z) \ < M, even \/(z) \ < M [278].
336. We assume that n boundary sections of ~ lie on the real axis.
From a variable point z in the upper half-plane they are seen under the
angles wI> w 2 , ••• , W n . Determine a regular and analytic function q;v(z)
in the half-plane 3z > 0 so that n91q;.(z) = w. [57J and put
cf>(z) = a .( ~y,(Z)+Q>'(Z)+"'+Q>n(z).

The function t(z) cf>(Z)-l is regular and bounded in the interior, < 1 on
the boundary, of ~ with the possible exception of 2n boundary points
[335].
337. The function l(eU ) is single-valued, regular and bounded in the
half-plane 91u < O. We have It(ell ) I < 1 at all the boundary points of this
half-plane except at the boundary point u = 00. [335.]
338. If z = 00 were not a boundary point we would have [135
suffices] necessarily Ig(z) I ::;;; k in ffi: contradiction. He1\ce z = 00 is a
boundary point of ffi. If g(z) were bounded in 91 we could use 335 (only
excluded point z = (0) which would lead to Ig(z) I < k in ffi.
339. There exists, according to the hypothesis, a constant M, M > 0,
such that It(z) 1< M in the region ffi bounded by 1 and 2• Choose r r
r r
R > 1 and so large that It(z) I < e on 1 and 2 outside the circle Iz I= R.
Take the branch of log z that is real for positive z. This branch is regular
and single-valued in 91, where Ilog z I < log Iz I + n provided that
Iz I > R. The inequality
1M (log R + n) + e (log z + n)1 > M (log R + n) + e (log Iz\ + n)

yuana97@uw.edu
384 Functions of One Complex Variable

holds and for Iz I ~ R both addends on the right hand side are positive.
Thus we obtain for Izl = R, z in m,

IM(log R+
log z
n) + e(log z + n) I(z) <
I log R+ n
M(log R + n)
M
=.
1

If Izl ~ R, z on r1 or on rs we find
log z I( ) 1 log Izi +n 1
1M(log R + n) + e(log z + n) Z <
+ n) e = .e(log Izl
Hence we have, by virtue of 335, at each point z in mbut outside the
circle Izl = R
I/(z) I< I~(log z + n) ~g~(IOg R + n) I.
The right hand side becomes smaller than 2e when Iz I gets sufficiently
large.
340. Assume, if possible, a =f= b and consider two disks Dl and D.
in the w-plane, with no points in common, centered at a and b respec-
tively. Outside the two disks the expression I(w - a ~ by - C' ; by I
has a positive minimum = e. Proposition 339 applied to the function
(/(z) - a ~ by _(~; by implies that the absolute value of this function
is smaller than e in the region m
bounded by l and rs whenever r
Iz I > R = R(e): we consider only such points z. Find two points, Zl
r r
on l and Z2 on 2, such that WI = I(ZI) is in DI and w 2 = I(Z2) is in D2
and join the two points by a curve in m. The image in the w-plane of

for which 1 e;by


this curve leads from DI to Dso Therefore there exists on it a point w = I(z)
(w - a ~ by - I > e: contradiction.

yuana97@uw.edu
Author Index

Numbers refer to pages. Numbers in italics refer to original contributions.

Abel, N. H. 185, 196, 351, 364 Darboux. G. 185. 288


Ahrens, W. 174 Dedekind. R. 309
Appell, P. 186 Dini. U. 81. 269
Archibald, R. G. 197 Dirichlet. P. G. L. 18. 78. 111. 112. 135,
Audibert 217 153. 240. 246. 309. 333

EnestrOm. G. 301
Bagnera, J. 208
Euler. L. 50. 51. 54. 57, 174. 177. 178.
Bendixson, J. 361
186, 234. 240. 243. 348. 349
Bernoulli, D. 121, 124
Bernoulli, Jakob 35 Faber. G. 376
Bernstein, S. 275, 367
Farey. J. 93
Bessel, F. W. 22, 97
Fatou, P. 369
Bieberbach, L. 325
Fejer. L. 200. 253. 260, 272. 281, 303.
Biehler, Ch. 302
337
Blaschke, W. 251, 267, 374
Fekete. M. 198. 302. 359
Blumenthal, O. 376
Fine. N. J. 2211
Bohr, H. 372
Fourier. J. 83. 253. 273
Borel. E. 157, 198, 332
Franel. J. 234, 275. 283. 284
Brauer. A. 304 Frank. Ph. 326
Brauer, R. 304
Fredholm. 1. 247
Briggs, H. 92. 284 Frobenius, G. 196
Biirmann. H. 146 Fujiwara, M. 304
Burkhardt, H. 288
Gauss. C. F. 11. 186, 187.303
Cantor, G. 206 Goursat. E. 313
Caratheodory, C. 261. 269, 355. 370 Greenstreet,E. 183
Carleman. T. 195. 325. 337. 340. 374 Gutzmer. A. 324
Carlson. F. 380
Catalan. E. 176. 178 Haar. A. 206
Cauchy, A. L. 64. 68. 111. 113. 119. 133. Hadamard. J. 164. 166.309. 376
146, 221, 245. 248. 312, 317. 326. 330. Hamburger. H. 332. 360
337 Hansen. P. A. 97
Cesaro. E. 15, 20, 176, 178, 179, 185, Hardy. G. H. 179, 195, 208. 247. 251.
193.208.217.234.240 257, 332. 359. 376
Cornu 142, 342 Harnack. A. 363
Cramer. H. 286 Heeke. E. 280
Csillag. P. 287. 374 Hermite. Ch. 104, 180, 182, 288,302.332

yuana97@uw.edu
386 Author Index

Hewitt, E. 206 ~ehler, F. G. 135, 333


Hilbert, D. 265, 337 ~ercator, G. 116, 313
Hille (textbook) 140, 195, 269, 338, 339, ~ertens, F. 308
376, 379 Minkowski, H. 71, 251
Hobson, E. W. 77, 259 ~ises, R. v. 280, 294
milder, O. 68, 249 Moivre, A. 292
Hurwitz, A. 274, 301, 341, 346 ~oreau, C. 216
Hurwitz-Courant (tex~book) 146, 195,
23~ 31~ 33~ 33~ 376 Nagy, J.v. Sz. 305
Jacobi, C. G. J. 147, 186, 187, 214, 349 Netto, E. 180
JacobsthaJ, E. 217 Nevanlinna, F. 339, 340, 381
Jensen, J. L. W. V. 130, 138, 248, 268, NevanJinna, R. 339, 381
305, 324, 326, 339, 340 Norlund, N. E. 192, 361, 366
Jordan, C. 239, 246
Julia, G. 344, 373 Obreschkoff, N. 347
Ostrowski, A . .'342
Kakeya, S. 206, 301 Ouspcnski, J. 211
Kempner, A. 205
Kirchhoff, G. 320 Paoli, P. 180
Knopp, K. 205 Parseval, ~. A. 324
Knopp (textbook) 198, 206, 208, 234, Pascal, B. 147
306, 309, 362 Perron, O. 197, 288
Koebe, P. 369 Persson, P. 379
Konig, D. 286 Phragmen, E. 273, 379
Konig, J. 358 Pincherle, S. 221
Kojima, T. 111 Plancherel, M. 274
IHirscMk, J. 254 Poisson, S. D. 138
P6lya, G. 173, 182, 189, 200, 201, 202,
Lacour, E. 186 207, 214, 223, 232, 234, 235, 239, 240,
Lagrange, J. L. 145, 146,347 241,246, 247, 249, 251, 255, 256, 257,
Laguerre, E. 147, 180,257, 302 258, 264, 267, 269, 270, 276, 278, 286,
Laisant, C. A. 218 294, 305, 309, 312, 330, 346, 347, 349,
Landau, E. 212, 271, 276, 356, 357 358, 360, 364, 367
Laplace, P. S. 96, 120, 135, 288, 294, HSI: 173
317, 334 MD: 178, 189, 223
Lasker, E. 209 ~PR: 227, 344
Laurent, P. A. 151 Pringsheim, A. 329
Lebesgue, H. 288 Priifer, H. 281, 360, 364
Legendre, .~. ~. 94, 95, 97, 135, 147, 328
Lerch, ~. 273, 274 Rademacher, H. 208
Levy, P. 311 J{assicod, L. 180
Lindelof, E. 218, 368, 369, 379 Riemann, B. 46, 56, 77, 113, 119, 172,
Littlewood, J. E. 195, 247, 251, 257 208, 259, 260, 311, 312, 326, :34:1
Lowner, K. 300, 326, 373 Riesz, F. 337
Lucas, Ch. F. 303 Riesz, M. 367
Lukacs, F. 253 Houche, E. 142, 344
Lukacs, Th. 323 Rumpen, H. 218

Macdonald, H. M. 343, 344 Saxer, W. 201


~IacIaurin, C. 33, 34, 212, 243 Schiffer, M. 256
~IahrenhoJz, J. 242, 330 Schlegel, V. 182
~Iarkoff, A ..\. 286 SchoJ], K. 360

yuana97@uw.edu
Author Index 387
Schottky, F. 356 Threlfall, W. 208
Schur, I. 111, 193, 215, 216, 218, 258, Toeplitz, O. 16, 111, 303, 356
302, 308, 309, 337, 353
Schwarz, H. A. 68, 72, 160, 369
Valiron, G. 201
Schweitzer, P. 253
Veress, P. 200
Scott, G. A. 33
Vitali, G. 218
Sidon, S. 309
Volterra, V. 231
Sierpinski, W. 208
Sommerfeld, A. 121, 142, 320, 342
Steinhaus, H. 111 Wallis, J. 289
Steinitz, E. 258, 280, 311 Walther, A. 379
Stieltjes, T. J. 288, 305, 308, 332 Waring, E. 179
Stirling, J. 36, 42, 43, 234, 289, 351, Watson, G. N. 33, 233, 330, 343
366 Weierstrass, K. 83, 200
Sturm, Ch. 341 Weyl, H. 211, 279, 280, 281, 285, 333,
Szasz, O. 218 337
Szego, G. 200, 214, 242, 256, 258, 264, Whittaker-Watson (textbook) 36, 54,
287, 333, 375 146, 212, 246, 343, 366
SzUcs, A. 286 Wiener, F. 337
Wiman, A. 201
Takagi, T. 304 Wintemitz, A. 315
Taylor, B. 267, 281

yuana97@uw.edu
Subject Index
Roman numerals refer to parts, Arabic numerals to problems or pages. Example:
III304 164 refer to problem 304 of Part III on page 164.

Abel's theorem I 86 21 Euler's (or Mascheroni's) constant


Argument principle 140 II 19.1, II 32 51, 54
Euler's function tp(n) II 188 93
Bernoulli numbers I 154 35
Bernoulli's equation II! 90 121 Farey series I! 189 93
Bessel functions I 97, I! 204 16, 97 Force, flux of III 89 120
Binomial coefficients, Gaussian 11 lines of III 85 120
ordinary 6 Fourier constants I! 141 83
Biirmann-Lagrange series 146
Gamma-function r(s) II 31, II 143
Cantor's discontinuum III 130 206 54,84
Capacity III 92 122 Graeffe's method II 95.4 257
Carlson's theorem III 328 170
Cauchy-Riemann differential equa- Hadamard's three circle theorem
tions 113 III304 164
Center of gravity of a mass distribu- Hamburger's function III 153, III 250
tion 108 332,360
Central subscript 27 Hermitian form III 7 104
Change problem I 9 174
Conformal center of gravity III 129 Inequality, Cauchy's II 80 68
130 Holder's II 81.1 68
Conformal mapping 115 Minkowski's II 90 71
Convergence exponent 25 Schwarz's I! 81 68
Convex (concave) function 65 Integral, definite 46
Counting function 85 improper 52
multiple 60
Differences, finite 148, 149
Dini's theorem II 128 81 Jacobi polynomials III 219 147
Direction of accumulation III 52 112 jensen's formula III 175 138
Dirichlet-Mehler formula III 157 135
Dirichlet series I 75, II 117, III 247 Lagrange (Biirmann) series 145
18, 78, 153 Laguerre polynomials III 219 147
Laplace equation III 87 120
Enlargement, area 117 Lattice points I 28 5, 13
lirtear 116 Legendre polynomials III 157, III 219
Enveloping series 32 135, 147
Equidistribution 87 zeros I! 191 94

yuana97@uw.edu
Subject Index 389
V~vellines III 85 120 Saddle point III 132 131
Line of support I 110 202, 126 Schlicht (univalent, simple) function
Linear transformation of sequences 15 117
convergence preserving 111 Schwarz's lemma III 280 160
luwer (upper) triangular I 79 194 Slowly increasing (decreasing) function
matrix I 85, 179 15, 19 II 150 85
regular I 88, 180 16, 19 Star-shaped III 109 125
triangular I 85 15 Step-function II 102 76
Stereographic projection III 80 115
Majorant (minorant) series 14 Stirling's formula (for r) II 205 97
Mapping 113 numbers, 1st kind 43
Maximum principle III 135 131 numbers, 2nd kind 42
Maximum term 27 series I 155 36
Means, arithmetic, geometric, har- Stream function III 85 120
monic of functions II 48 57, 63 lines III 85 120
of numbers 57, 62, 63 potential III 85 120
Mercator's projection 116 Sum, upper, lower 46, 60
Minimax 101 Subseries 28
Modular graph 130 complementary 31
Moments of a function II 138 83 Support function 126
line of I 110 202, 126 I
Orthogonal functions III 117 127
Oscillation II 108 77 Univalent (schlicht, simple) function
117
Partition of a set 42
Phragmen-LindelOf method III 322 166
Pit III 132 131 Variation, total, of a function II 9 232
Poisson's formula III 173 138 Vector field 113, 119
Polynomial with integral values I 128 29 conjugate potential III 85 120
Postage stamp problem I 9 174 irrotational 119
Potential III 85 120 potential III 85 120
soleueidal 120
Rearrangement of infinite series 30,
112 Wallis' formula II 202 289
Regular sequence 86 Weierstrass' approximation theorem
Riemann's C-function II 45 56 II 135, II 138 83
Rotation 117 Weighing problem I 9 174
RoucM's theorem III 194 142 Winding number 140

yuana97@uw.edu
Die Gmndlehren der mathematischen Wissenschaften
A Series of Comprehensive Studies in Mathematics
A Selection

10. Schouten: Der Ricci-Ka\kiil


23. Pasch: Vorlesungen fiber neuere Geometrie
41. Steinitz: Vorlesungen fiber die Theorie der Polyeder
45. AlexandrofflHopf: Topologie. Band 1
46. Nevanlinna: Eindeutige analytische Funktionen
57. Hamel: Theoretische Mechanik
63. Eichler: Quadratische Formen und orthogonale Gruppen
91. Prachar: Prlmzahlvertellung
102. NevanlinnalNevanlinna: Absolute Analysis
114.~c~e:lfomology
127. lfermes: Enumerability, Decidability, Computability
131.lIirzebruch: Topological Methods in Algebraic Geometry
137.lfandbook for Automatic Computation. Vol.1IPart b: Graullfill/langmaack:
Translation of ALGOL 60
138. Hahn: Stability of Motion
139. ~thematische lfilfsmittel des Ingenieurs. 1. Tell
140. ~thematische lfilfsmittel des Ingenieurs. 2. Tell
141. ~thematische lfilfsmittel des Ingenieurs. 3. Tell
142. ~thematische lfilfsmittel des Ingenieurs. 4. Tell
143. Schur/Grunsky: Vorlesungen fiber Invariantentheorie
144. Well: Basic Number Theory
145. ButzerlBerens: Semi-Groups of Operators and Approximation
146. Treves: Locally Convex Spaces and Linear Partial Differential Equations
147. Lamotke: Semisimpliziale aigebraische Topologie
148. Chandrasekharan: Introduction to Analytic Number Theory
149. Sario/Oikawa: Capacity Functions
150. IosifesculTheodorescu: Random Processes and Learning
151. Mandl: Analytical Treatment of One..<Jim.ensional Markov Processes
152. lfewittIRoss: Abstract lfarmonic Analysis. Vol. 2: Structure and Analysis for Compact
Groups. Analysis on Locally Compact Abelian Groups
153. Federer: Geometric Measure Theory
154. Singer: Bases in Banach Spaces I
155. Milller: Foundations of the Mathematical Theory of Elektromagnetic Waves
156. van der Waerden: Mathematical Statistics
157. ProhorovlRozanov: Probability Theory. Basic Concepts. Limit Theorems. Random
Processes
158. ConstantinesculCornea: Potential Theory on lfarmonic Spaces
159. Kothe: Topological Vector Spaces I
160. AgrestIMaksimov: Theory of Incomplete Cylindrical Functions and Their
Applications .
162. Nevanlinna: Analytic Functions
163. StoerlWitzgall: Convexity and Optimization in Finite Dimensions I
164. SariolNakai: Classification Theory of Riemann Surfaces
165. Mitrinovic: Analytic Inequalities
166. GrothendiecklDieudonne: Elements de Geometrie Algebrique I
167. Chandrasekharan: Arithmetical Functions
168. Palamodov: Linear Differential Operators with Constant Coefficients
170. Lions: Optimal Control of Systems Governed by Partial Differential Equations
171. Singer: Best Approximation in Normed Linear Spaces by Elements of Linear
Subspaces
172. Biihlmann: ~thematical Methods in Risk Theory
173. Ma~eda: Theory of Symmetric Lattices

yuana97@uw.edu
174. StiefeVScheifele: Linear and Regular Celestial Mechanics. Perturbed Two-body
Motion - Numerical Methods - Canonical Theory
175. Larsen: An Introduction to the Theory of Multipliers
176. GrauertlRemmert: Analytische Stellenalgebren
177. Fltigge: Practical Quantum Mechanics I
178. Fltigge: Practical Quantum Mechanics II
179. Giraud: Cohomologie non abelienne
180. Landkof: Foundations of Modem Potential Theory
181. LionslMagenes: Non-Homogeneous Boundary Value Problems and Applications I
182. LionslMagenes: Non-Homogeneous Boundary Value Problems and Applications II
183. Lions/Magenes: Non-Homogeneous Boundary Value Problems and Applications m
184. Rosenblatt: Markov Processes. Structure and Asymptotic Behavior
185. Rubinowicz: Sommerfeldsche Polynommethode
186. Handbook for Automatic Computation. VoL 2 WilkinsonlReinsch: Linear Algebra
187. SiegeVMoser: Lectures on Celestial Mechanics
188. Warner: Harmonic Analysis on Serni-Simple Lie Groups I
189. Warner: Harmonic Analysis on Serni-Simple Lie Groups II
190. Faith: Algebra I: Rings, Modules, and Categories
191. Faith: Algebra II: Ring Theory
192. Mallcev: Algebraic Systems
194.lgusa: Theta Functions
195. Berberian: Baer*-Rings
196. Athreya/Ney: Branching Processes
197. Benz: Vorlesungen tiber Geometrie der Algebren
198. Gaal: Linear Analysis and Representation Theory
199. Nitsche: Vorlesungen tiber Minirnalfliichen
200. Dold: Lectures on Agebraic Topology
201. Beck: Continuous Flows in the Plane
202. Schmetterer: Introduction to Mathematical Statistics
203. Schoeneberg: Elliptic Modular Functions
204. Popov: Hyperstability of Control Systems
205. Nikol'skii: Approximation of Functions ofSeveml Variables and Imbedding
Theorems
206. Andre: Homologie des Algebres Commutatives
207. Donoghue: Monotone Matrix Functions and Analytic Continuation
208. Lacey: The Isometric Theory of Classical Banach Spaces
209. Ringel: Map Color Theorem
210. GihmanlSkorohod: The Theory of Stochastic Processes I
211. ComforUNegrepontis: The Theory ofUltrafilters
212. Switzer: Algebraic Topology - Homotopy and Homology
214. van der Waerden: Group Theory and Quantum Mechanics
215. Schaefer: Banach Lattices and Positive Operators
216. P61ya/Szego: Problems and Theorems in Analysis II
217. Stenstrom: Rings of Quotients
218. GihmanlSkorohod: The Theory of Stochastic Processes II
219. Duvant/Lions: Inequalities in Mechanics and Physics
220. Kirillov: ElementS of the Theory of Representations
221. Mumford: Algebraic Geometry I: Complex Projective Varieties
222. Lang: Introduction to Modular Forms
223. Bergh/UifstrOm: Interpolation Spaces. An Introduction
224. Gilbarg/Trudinger: Elliptic Partial Differential Equations of Second Order
225.Schtitte:ProofTheory
226. Karoubi: K-Theory. An Introduction
227. GrauertlRemmert: Theorie der Steinschen Riiume
228. SegallKunze: Integmls and Operators

Springer-Verlag Berlin Heidelberg New York

yuana97@uw.edu
Springer
and the
environment
At S~ringer we firmly believe that an
international science publisher has a
special obligation to the environment,
and our corporate policies consistently
reflect this conviction.
We also expect our business partners -
paper mills, printers, packaging
manufacturers, etc. - to commit
themselves to using materials and
production processes that do not harm
the environment. The paper in this
book is made from low- or no -chlorine
pulp and is acid free, in conformance
with international standards for paper
permanency.

Springer

yuana97@uw.edu

You might also like