The Scottish Book

R.
Daniel Mauldin
The Scottish Book

Mathematics from The Scottish Café,
with Selected Problems from The New Scottish Book
Second Edition
The Scottish Café, birthplace of the Scottish Book problems, as it appeared in a postcard from
the early 1970s. The Scottish Café is on the right, with the Café Roma on the left. According to
Stanisław Ulam, this scene has changed little from the period preceding World War II
R. Daniel Mauldin
The Scottish Book

Mathematics from the Scottish Café
with Selected Problems from The New
Scottish Book
Second Edition
R. Daniel Mauldin
Department of Mathematics
University of North Texas
San Diego, CA, USA
ISBN 978-3-319-22896-9 ISBN 978-3-319-22897-6 (eBook)

DOI 10.1007/978-3-319-22897-6
Library of Congress Control Number: 2015947275
Mathematics Subject Classification (2010): 01A60.
Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 1981, 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, 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.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.

springer.com)
Preface to the Second Edition
The Scottish Book, wrapped in the mists of the past, is a legend in the mathematics
world. Its fascinating story and the legendary figures who formulated the problems
in the book continue to hold our attention. It represents the best of café mathematics,
an informal, free-wheeling style of mathematical conversation and interaction that
seems almost lost today. For the solution to some of the problems, prizes were
offered, ranging from the famous live goose, a bottle of whiskey of measure > 0,
one kilo of bacon to one small beer.
One can imagine the atmosphere at the Scottish Café while the problems were
being formulated. We should take to heart the playfulness and enjoyment that is
on display, a good model for encouraging and stimulating mathematics at any age.
One can also see that a few of the problems must have been stated after spending
some time drinking tea or perhaps a brandy or two. In his book Adventures of a
Mathematician, Ulam describes what a session (one lasting at least seventeen hours)
at the café might be like and sketches some of the central figures. During the years I
knew and worked with Ulam, he loved the café mode of discussion. Gian-Carlo Rota
gives a detailed description of this in his article ‘The Lost Café’, in From Cardinals
to Chaos, edited by N. G. Copper (Cambridge University Press, 1989). The entire
school of mathematics in Lwów is wonderfully presented by Roman Duda in Pearls
From a Lost City (American Mathematical Society, 2014), a translation of his 2008
Polish version. The tragedy that befell so many of the figures around the book has
been traced many times, including the article by Joanna Diane Caytas, ‘Survival of
the Scottish Book: A Phoenix from the Holocaust of Polish Mathematics’, available
on the internet. There is even a collection of poems about the book by Susana H.
Case, The Scottish Café (Slapering Hot press, 2002).
The problems and the ideas behind continue to effect mathematics today. In
the 35 years since the first edition of the book, many more problems have been
solved or partially solved. But even today, quite a few remain unsolved. In view
of this, I decided to gather new commentaries and update some of the old ones.
The appendices of this edition include a list of the unsolved and partially solved
problems together with those that have no commentary, a list of unsolved prize
problems, a list of problems posed by each author, and a list of problems by subject
v
vi Preface to the Second Edition
area. Besides correcting many errors in the first edition, for clarification a few
changes in Ulam’s translation of the original Scottish Book have been made.
In addition to some of the lectures given at the 1979 Scottish Book conference,
this edition also includes a brief history of Wrocław’s New Scottish Book and some
selected problems from it.
This edition would not be possible without the generous contributions and
suggestions of the commentators. I offer them my heartfelt gratitude. I want to thank
Kirby Baker, Larry Lindsay, Bill Bernard, and Sue DeMeritt for their assistance
in getting this project underway. I thank Al Hales, Joe Buhler, and Jan Mycielski
for their counsel. I thank Allen Mann, Christopher Tominich, and Benjamin Levitt
at Birkhäuser for their support during the preparation of this new edition. Finally,
I thank Diana, my wife, for supporting me throughout this long project.
It is my sincere hope that this collection will bring to the reader, as it has to me,
many hours of enjoyment and an image of what must have been a most wonderful
place, The Scottish Café.
San Diego, CA, USA R. Daniel Mauldin

Preface to the First Edition
Once while working on a problem, someone was kind enough to point out that
my problem was in the “Scottish Book.” I had not heard of the Scottish Book
and certainly did not realize that this book had no connection with Scotland. But,
since that introduction, I’ve become more and more aware of the magic of the
mathematicians and the mathematics involved in its birth.
The Scottish Book offers a unique opportunity to communicate with the men (no
women were on the scene, as I understand it) and ideas of a time and place, Lwów,
Poland, which have had an enormous influence on the development of mathematics.
The history of the Scottish Book as detailed in the following lectures by Ulam,
Kac, and Zygmund provides amazing insights into the mathematical environment
of Lwów before World War II.
There are many collections of problems, but this set has become world-renowned.
Perhaps, a primary reason for this renown is that the problems are clearly and simply
formulated, accessible to the general mathematical community, and yet strike at the
heart of the concepts involved.
It is my pleasure and honor to edit this version of the Scottish Book, which
includes a collection of some of the talks given at the “‘Scottish Book Conference”’
held at North Texas State University in May of 1979. The purpose of the conference
was to examine the history, development, and influence of the Scottish Book. As
John Oxtoby toasted at the conference, there was a “‘condensation of Poles”’ at
this conference. Among them were some of the original contributors to the Scottish
Book, Professors Ulam, Kac, and Zygmund. Their edited talks appear here, together
with the talk given by one close to them in spirit and collaboration, Professor
Paul Erdós. Also presented here is a talk by a member of a younger generation,
Professor Andrzej Granas, in which one problem of Schauder is discussed with
its many-faceted implications and connections. It should come as no surprise that
the conference was held in Texas; the mathematical similarities between the Texas
school and the Polish school have long been noted, beginning with the fact that the
first American to publish in Fundamenta Mathematicae is R. L. Moore.
From a glance at the problems, one sees that they cover a wide range of
mathematics. I think this simply reflects the wide interests of the unusual group,
vii
viii Preface to the First Edition
which assembled the collection. The problems are concentrated in the areas of
summability theory, functional and real analysis, group theory, point set topology,
measure theory, set theory, and probability. It is likewise easy to confirm that some
of the contributors to the Book were, as R. H. Bing toasted, the “‘leading lights”’ in
these fields.
I have attempted to obtain an appropriate commentary for each of the problems,
although quite a few of the problems remain without comment. Some of these, as
well as a number of problems with comments, remain unsolved to this day. For
others I simply failed to get an appropriate expert comment (I would be grateful for
contributions from readers of this edition).
Following a problem there may appear the word Addendum. This indicates
a comment that was entered into the Scottish Book during the time when the
problems were being collected in Lwów. Later commentaries, remarks, and solution
to problems which are presented here for the first time follow the original addenda.
The problems and original addenda appear here essentially as they have in the
two earlier English-language editions of the Scottish Book, both edited (and one
produced) by Stanisław Ulam. The first, in 1957, was a mimeographed version of
Ulam’s own translation from the original languages in which the problems were
inscribed in the Book (mostly Polish), which he distributed on personal request
from his professional base at Los Alamos National Laboratory. By 1977, the volume
of requests addressed to both Professor Ulam and the Los Alamos Laboratory’s
library made it only reasonable to prepare a somewhat more formal edition.
This edition again presented only the translated problems and their contemporary
addenda, and has been distributed by the Los Alamos laboratory since then. The
recent reconcentration and expansion of interest in the Book, including the 1979
conference, had made a place for a new edition, including a collection of at least
some of the work which has been stimulated by the Scottish Book problems in the
years since they were first collected.
This project enjoyed the aid of several individuals and institutions, beginning
with the encouragement of Stan Ulam and Gian-Carlo Rota. I sincerely thank
all of the commentators for their generosity in providing the commentaries and
suggestions. It is obvious that a major contributor to this edition is Jan Mycielski.
His encouragement and constant flow of comments and references kept life in the
project. Bill Beyer provided many significant comments on the formulation of the
problems.
The Scottish Book conference, which was held in Denton in May 1979, focused
our efforts. It was my hope that some of the spirit of that time and place would
be recaptured at the conference. Perhaps, it was through the contributions of the
speakers including R. D. Anderson and D. A. Martin, both of whom traced some of
the most outstanding work in their fields back to the Scottish Book.
Preface to the First Edition ix
The National Science Foundation, through grant MCS-79-0971 and North Texas
State University, provided funding for the conference. A number of commentaries
were written under the auspices of a Faculty Research grant from North Texas State
University. I sincerely thank Lynn Holick for the superb typing, and the people at
Birkhäuser Boston for their help in bringing the project to fruition.
San Diego, CA, USA R. Daniel Mauldin

1981
Preface to the Limited Los Alamos
Edition of 1957
The enclosed collection of mathematical problems has its origin in a notebook,

which was started in Lwów, in Poland in 1935. If I remember correctly, it was
S. Banach who suggested keeping track of some of the problems occupying the
group of mathematicians there. The mathematical life was very intense in Lwów.
Some of us met practically every day, informally in small groups, at all times of the
day to discuss problems of common interest, communicating to each other the latest
work and results. Apart from the more official meetings of the local sections of the
Mathematical Society (which took place Saturday evenings, almost every week!),
there were frequent informal discussions mostly held in one of the coffee houses
located near the University building - one of them a coffee house named “Roma,”
and the other “The Scottish Coffee House.” This explains the name of the collection.
A large notebook was purchased by Banach and deposited with the headwaiter of
the Scottish Coffee House, who, upon demand, would bring it out of some secure
hiding place, leave it at the table, and after the guests departed, return it to its secret
location.
Many of the problems date from years before 1935. They were discussed a great
deal among the persons whose names are included in the text, and then gradually
inscribed into the “‘book”’ in ink. Most of the questions proposed were supposed
to have had considerable attention devoted to them before an “‘official”’ inclusion
into the “‘book”’ was considered. As the reader will see, this general rule could not
guarantee against an occasional question to which the answer was quite simple or
even trivial.
In several instances, the problems were solved, right on the spot or within a short
time, and the answers were inscribed, perhaps some time after the first formulation
of the problem under question.
As most readers will realize, the city of Lwów, and with it the “Scottish Book,”
was fated to have a very stormy history within a few years of the book’s inception.
A few weeks after the outbreak of World War II, the city was occupied by the
Russians. From items at the end of this collection, it is seen that some Russian
mathematicians must have visited the town; they left several problems (and prizes
for their solutions). The last date figuring in the book is May 31, 1941. Item Number
xi
xii Preface to the Limited Los Alamos Edition of 1957
193 contains a rather cryptic set of numerical results, signed by Steinhaus, dealing
with the distribution of the number of matches in a box! After the start of war
between Germany and Russia, the city was occupied by German troops that same
summer and the inscriptions ceased.
The fate of the Scottish Book during the remaining years of war is not known to
me. According to Steinhaus, this document was brought back to the city of Wrocław
by Banach’s son, now a physician in Poland. (Many of the surviving mathematicians
from Lwów continued their work in Wrocłow. The tradition of the Scottish Book
continues. Since 1945, new problems have been formulated and inscribed and a
new volume is in progress.)
A general word of explanation may be in order here. I left Poland late in 1935 but,
before the war, visited Lwów every summer in 1936, ’37,’38, and ’39. The last visit
was during the summer preceding the outbreak of World War II, and I remember just
a few days before I left Poland, around August 15, the conversation with Mazur on
the likelihood of war. It seems that in general people were expecting another crisis
like that of Munich in the preceding year, but were not prepared for the imminent
world war. Mazur, in a discussion concerning such possibilities, suddenly said to
me “ A world war may break out. What shall we do with the Scottish Book and
our joint unpublished papers? You are leaving for the United States shortly, and
presumably will be safe. In case of a bombardment of the city, I shall put all the
manuscripts and the Scottish Book into a case which I shall bury in the ground.” We
even decided upon a location of this secret hiding place; it was to be near the goal
post of a football field outside the city. It is not known to me whether anything of
the sort really happened. Apparently, the manuscript of the Scottish Book survived
in good enough shape to have a typewritten copy made, which Professor Steinhaus
sent to me last year (1956).
The existence of such a collection of problems was mentioned on several
occasions, during the last 20 years, to mathematical friends in this country. I have
received, since, many requests for copies of this document. It was in answer to such
oral and written requests that the present translation was made. This spring in an
article, “Can We Grow Geniuses in Science?”, which appears in Harper’s June 1957
issue, L. L. Whyte alluded to the existence of the Scottish Book. Apparently, the
diffusion of this small mystery became somewhat widespread, and this provided
another incentive for this translation.
Before deciding to make such an informal distribution, I consulted my teacher
and friend (and senior member of the group of authors of the problems), Professor
Steinhaus, about the propriety of circulating this collection. With his agreement, I
have translated the original text (the original is mostly in Polish) in order to make it
available through this private communication.
Even as an author or co-author of some of the problems, I have felt that the only
practical and proper thing to do was to translate them verbatim. No explanations or
reformulations of the problems have been made.
Many of the problems have since found their solution, some in the form of
published papers (I know of some of my own problems, solutions to which were
Preface to the Limited Los Alamos Edition of 1957 xiii
published in periodicals, among them Problem 17.1, Z. Zahorski, Fund. Math., Vol.
34, pp. 183–245 and Problem 77(a), R. H. Fox, Fund. Math., Vol. 34, pp. 278–287).
The work of following the literature in the several fields with which the problems
deal would have been prohibitive for me. The time necessary for supplying the
definitions or explanations of terms, all very well understood among mathematicians
in Lwów, but perhaps not in current use now, would also be considerable. Some
of the authors of the problems are no longer living, and since one could not treat
uniformly all the material, I have decided to make no changes whatsoever.
Perhaps, some of the problems will still present an actual interest to mathe-
maticians. At least the collection gives some picture of the interests of a compact
mathematical group, an illustration of the mode of their work and thought; and
reflects informal features of life in a very vital mathematical center. I should be
grateful if the recipients of this collection were willing to point out errors, supply
information about solutions to problems, or indicate developments contained in
recent literature in topics connected with the subjects discussed in the problems.
It is with great pleasure that I express thanks to Miss Marie Odell for her help
in editing the manuscript and to Dr. Milton Wing for looking over the translated
manuscript.
Los Alamos, NM, USA S. Ulam

May 1977
Preface to the Limited Los Alamos
Edition of 1977 Monograph
Numerous requests for copies of this document, addressed to Los Alamos Scientific
Laboratory library or to me, appear to make it worthwhile (after a lapse of some
20 yr) to reprint, with some corrections, this collection of problems.
This project was made possible through the interest and active help of Robert
Krohn of the laboratory.
It is a pleasure to give special thanks to Dr. Bill Beyer for his perspicacious
review of the changes and the revised version of some formulations. Thanks are due
to Martha Lee Delanoy for editorial work.
Los Alamos, NM, USA S. Ulam

May 1977
xv
Contents
Part I The Scottish Book Conference Lectures

1 An Anecdotal History of the Scottish Book, S. Ulam . . . . . . . . . . . . . . . . . . . . . 3
2 A Personal History of the Scottish Book, Mark Kac . . . . . . . . . . . . . . . . . . . . . . 13
3 Steinhaus and the Development of Polish Mathematics, A. Zygmund . . 21
4 My Scottish Book ‘Problems’, Paul Erdös . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5 KKM-Maps, Andrzej Granas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.1 Definition and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Ky Fan fixed point theorem and the minimax inequality. . . . . . . . . . . . . . 39
5.3 KKM-maps and variational inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.4 KKM-maps and the theory of games. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.5 Bibliographical and historical comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Part II The Scottish Book Problems

6 Problems with Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7 Appendices to the Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7.1 Problem Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
7.2 Authors of the problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
7.3 Problems of Note. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Part III A Brief History of Wrocław’s New Scottish Book

8 Lwów of the West, P. Biler, P. Krupski, G. Plebanek
and W. A. Woyczyński . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Part IV The New Scottish Book

9 Selected Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
xvii
Part I
The Scottish Book Conference Lectures
Chapter 1
An Anecdotal History of the Scottish Book,
S. Ulam
For those readers who may not know, I should start by saying that the so-called
Scottish Book is an informal collection of problems in mathematics. It was begun
in Lwów, Poland—my home town—in 1935; how and why will be explained in
due course. Most of the problems are due to a few local mathematicians, myself
included. Actually, many of the earlier problems originated well before 1935—
perhaps 6 or 7 years before—during the period when I was still a student. As a
budding mathematician, I regularly attended all the seminars and lectures in my field
of interest, and made friends with several of the older, established mathematicians.
I was then able to take part in the informal discussions—generally among two or
three of us at a time—which were a standard feature of mathematical life in pre-
World War II Lwów. For several years I was invariably the youngest person in
any such group; ultimately, Mark Kac made his appearance, and I lost my special
position to him, my junior by some five years.
The story of the Scottish Book could also be called the “Tale of Two Coffee
Houses,” the Café Roma and, right next to it, the Café Szkocka, or Scottish Café.
These two establishments are situated on a little square 100 or 200 yards from the
University of Lwów. A few years ago my friend Mazur—one of the more prolific
authors represented in the Scottish Book—sent me a post card which shows these
two coffee houses as they were in the early 70s (and presumably still are). The
postcard has been reproduced as the frontispiece of this edition of the Scottish Book.
So far as I can tell, nothing has changed since the days before World War II.
For our story, the Café Roma was, in the beginning, the more important of the two
coffee houses. It was there that the mathematicians first gathered after the weekly
meetings of our local chapter of the Polish Mathematical Society. The meetings
were usually held on Saturday in a seminar room at the University—hence close
to the Cafés. The time could be either afternoon or evening. The usual program
consisted of four or five ten-minute talks; half-hour talks were not very common,
and hour-long talks were mercifully rare. There was of course some discussion at
the seminar, but the really fruitful discussions took place at the Café Roma after the
meeting was officially over.
© Springer International Publishing Switzerland 2015 3

R.D. Mauldin, The Scottish Book, DOI 10.1007/978-3-319-22897-6_1
4 1 An Anecdotal History of the Scottish Book, S. Ulam
Among the senior mathematicians who frequented the Café Roma, the most
prominent was undoubtedly Banach. The other full professors or associate profes-
sors were Stożek, Ruziewicz, and Łomnicki. There were also younger lecturers and
docents as well as a few students like myself. Kuratowski, who was a professor
at the Polytechnic Institute, and Steinhaus, who was at the University, preferred
a more elegant and genteel pastry shop. But Banach, Mazur and various visitors,
including Sierpiński, patronized the Roma. There we sat discussing mathematics,
lingering over a single cup of coffee or a glass of tea for three or four hours at a
time—something one can still do in some Paris cafés.
Besides mathematics, there was chess. Auerbach was a very strong player.
Frequently he would play a game or two with Stożek or Nikliborc while Banach
watched and, of course, kibitzed. That is something I too love to do, although I
know it can be extremely annoying to serious players. Many years later I read a
story about an Englishman who habitually kibitzed in his club. When the players
objected violently, he wrote a letter to the Times saying, “As a free Englishman, I
believe I have the right to express my opinions freely, and evaluate the position for
both players.”
But above all, we mathematicians continued the discussions which had been aired
earlier at the meetings of the Mathematical Society. The whole atmosphere, in Lwów
especially, was one of enthusiastic collaborations; people were really interested
in each other’s problems. This was true in Warsaw too, where there was much
collaboration among the topologists, set theoreticians and logicians. In Lwów, the
interest was not only in set theory, but, owing to Steinhaus’ and Banach’s influence,
also in functional analysis and several other fields.
It was Steinhaus who discovered Banach; in fact, he used to say that this was
his greatest discovery. Steinhaus was a young professor in Kraków, a city about two
hundred miles west of Lwów. One evening while walking in a park, he overheard
two young men sitting on a bench discussing the Lebesgue integral. Lebesgue’s
integral was a rather new theory at the time (this was 1917). Steinhaus was intrigued
and started talking to the two young men, one of whom was Banach. Steinhaus was
greatly impressed, and he encouraged Banach to continue his studies. Banach, by
the way, was a very eccentric person in his habits and personal life. He would not
take any examinations at all, disliking them intensely. But he wrote so many original
papers and proposed so many new ideas that he was granted a doctor’s degree several
years later without passing any of the regular exams. All this happened at the end of
World War I around 1919.
Collaboration was of course not unknown in other mathematical centers. For
example, the book by Felix Klein on the history of mathematics in the 19th century
mentions that groups of mathematicians (small groups—pairs, triplets at most)
discussed mathematical problems in Göttingen. This was not so prevalent in Paris.
At one time I thought it would be interesting to try to write a history of
the development of mathematical collaboration. I have the impression that the
profusion of joint papers is a rather recent trend. Not being a historian, I don’t
know the detailed course of the development of mathematics in Italy in the days
of the Renaissance, but it is certainly true that even in antiquity mathematicians
were writing letters to each other. There are letters of Archimedes containing
1 An Anecdotal History of the Scottish Book, S. Ulam 5
mathematical problems and theorems. As for joint papers, I don’t know their
ancient history. There are of course some very famous 19th Century papers like
Russell-Whitehead. Atiyah-Singer is an example of a more recent and celebrated
mathematical paper, as are the Kac-Feynman formulae, and so on.
Today collaboration and the number of joint papers seem to be increasing. In
fact, joint papers appear to constitute a sizable proportion of all current original
mathematical work. There is almost as much as in physics, where, especially in
experimental physics, there may be ten or twenty authors for a single paper. In
theoretical physics too, one sees quite a few joint papers.
This seems to me a curious phenomenon both epistemologically and psycho-
logically. Somehow, in some cases, collaboration is more fruitful than the efforts
of a single individual. Certain single individuals still produce the main ideas, but
it is interesting to compare this parallel work to work on computers. Without any
doubt, the human brain, even in a single person, operates on many parallel channels
simultaneously. This is not so on present day computers, which can only perform
one operation at a time.
The germ of cooperation between elements exists in the brain of even very
primitive animals, and is well-known in mammals. Certainly, creative activity in
mathematics requires the putting together of very many elements. Suppose we had
two or three brains working together in parallel on a subject; it is fascinating to
speculate on what this might lead to. No one doubts that we shall witness the
development of computers able to work in parallel; in fact, this development has
already begun. Of course, it is dangerous to be too certain of what the future will
bring. I cannot refrain from quoting a statement by Niels Bohr: “It is very difficult
to predict; especially the future.”
But to go back to the Polish school of Mathematics, to the cafés and to the
Scottish Book—I should point out that the subjects studied partook of a certain
novelty. Set theory itself was still rather new, and set theoretical topology was newer
yet. The theory of functions of real variables and the idea of function spaces were
to some extent fostered and developed in Poland, and in Lwów specifically.
Another point that should be made is that the definition of Banach spaces gave
a very general framework and yet embraced many examples, each having, so to
speak, a different flavor; these were sufficiently different to excite great interest.
Generalizations about objects that are too similar to each other are less interesting.
But where one can identify common properties of objects which appear quite
different from each other, it is comparable to the living world where there exist so
many species that are close but not alike. The richness depends on the combination
of diversity and similarity.
Those who have followed the subject know that there are many different types of
Banach spaces: the space of continuous functions, Hilbert space, not to mention
the finite-dimensional Banach spaces with different Minkowski metrics, spaces
of measurable functions, analytic functions, and so on, all having made their
appearance implicitly in problems of mathematical analysis. And that class of spaces
and transformations is of special interest mainly because the spaces deal with non-
compact phenomena.
The usual approximation methods, the “epsilon” approaches, are almost by

definition suitable for treating compactness in the classes of objects under dis-
cussion. Many problems of analysis however, are, so to say, very noncompact
and yet somehow homogeneous and amenable to methods of a general analysis
with limiting processes, which are encompassed by Banach’s original definition.
A similar definition was given independently by Norbert Wiener, but as he wrote
in his autobiography, he somehow lost interest in the subject without developing a
theory of such spaces. Some of the problems in the little collection gathered in the
Scottish Book deal with function spaces.
I have not yet explained how this collection came about. Let us therefore go
back to the Café Roma and Banach. He used to spend hours, even days there,
especially towards the end of the month before the university salary was paid.
One day he became irritated with the credit situation at the Roma and decided
to move to the Szkocka next door, a mere twenty yards away. Stożek and some
chemists and physicists continued to frequent the Roma, but the Scottish Café now
became the meeting place of a smaller group of mathematicians, including Banach,
Mazur, myself, and occasionally some others. It is owing to this that so many of the
problems in this collection are entered in our names. There were of course visitors,
my friend Schreier among others, but the regular habitués were just the three of us.
How did the book come about? One day Banach decided that because we talked
about so very many things, we should write the ideas down whenever possible in
order not to forget them. He bought a large and well-bounded notebook in which we
started to enter problems. The first one bears the date July 17, 1935. This was while I
was still living in Poland, before I received an invitation from von Neumann to visit
him in Princeton. (It was during this visit to Princeton that the late G. D. Birkhoff, at
a tea at von Neumann’s, asked me whether I would come to Harvard to join the Soci-
ety of Fellows there. I accepted of course, and consequently was able to remain in
the United States.) During the summers I used to return to Poland to visit my family
and my mathematical friends. These were the summers of 1936, ’37, ’38, and ’39.
The notebook was kept at the Scottish Café by a waiter who knew the ritual—
when Banach or Mazur came in it was sufficient to say, “The book please,” and he
would bring it with the cups of coffee.
As years passed, there were more and more entries by other Polish mathemati-
cians, Borsuk for instance—a topologist friend of mine from Warsaw—and many
others. The “Book” grew to become a collection of some 190 problems, of which
by now, nearly fifty years later, about three-quarters have been solved. Some of the
problems were entered without too much previous work or thought; a few were
solved on the spot. All of this is noted in the book.
The document stayed in Poland. On my last prewar trip in the summer of 1939,
Mazur, more realistic about the world situation than I (I thought we would only
see more crises like that of Munich or Czechoslovakia), said he believed a great
war was imminent. He said that our results, about countable groups among other
subjects, some of which are unpublished to this day, should not be lost, so he
proposed that when war came he would put the book in a little box and bury it
where it could be found later, near the goal post of a certain soccer field. I don’t
know whether this is the way the Scottish Book was preserved or not, for when I
saw Mazur a few years ago in Warsaw I forgot to ask him about it. At any rate, the
Scottish Book survived the war and was in Banach’s possession. When Banach died
in 1945, his son Stephan Banach, Jr. (now a neurosurgeon in Warsaw) found it, and
showed it to Steinhaus immediately after the war. Steinhaus then copied it verbatim
by hand, and in 1956 sent this copy to me at Los Alamos. I translated it, and had
some three hundred mimeographed copies of the translation made. I had to pay for
this myself—Los Alamos is a government laboratory, and one cannot use taxpayers’
money for such frivolous purposes. I mailed these copies to various universities both
here and abroad, and also to a few friends. Since then, as the book became known
in mathematical circles, people kept writing to Los Alamos for copies. There were
so many requests over the years that the laboratory decided in 1977 to print another
edition, under the supervision of W. A. Beyer. Photocopies of the Polish original
have been preserved. If someone were interested in graphology or handwriting—
the handwriting of Banach or Mazur, for instance—he could look at it. (Some is
reproduced on pages 14 and 15.)
So much for the origin of the “Book.” Many problems are still unsolved, and
according to experts, have some value. I think it is fair to say that these problems
did exert an influence on the development of some subjects in the areas of functional
analysis, in the theory of infinite series, in real variable theory, in topology, in the
theory of probability (including measure theory), in group theory, and so on. It was
later in the game in Lwów that algebraic problems became of interest. Schreier and I,
along with Mazur, began to discuss problems concerning groups, as well as various
questions in the theory of Lie algebras. I remember that when I first learned about
the latter at the age of twenty-two or twenty-three, they seemed too formal to me.
Only later did I begin to appreciate their importance and applications. There were
also some problems in geometry.
It was the variety of examples and a certain concreteness in these abstract ideas
that made this whole subject, for me and perhaps for many other mathematicians,
so vivid and alive. There are examples of spaces, examples of transformations,
examples of functions, of sets. Recently, by the way, and quite by chance, I came
upon the following phrase of Shakespeare’s, in Henry VIII: “Things done without
example, in their issue are to be feared.” Is this an anti-“new-math” statement? I can
certainly agree with the sentiment, even if, as I suspect, the word “example” was
meant in quite a different way.
Central to the theme I am trying to develop is this class of examples which have
something, but not too much, in common. Here we see almost a biological or genetic
development, an evolutionary development of the objects which mathematics
creates and which take on a life of their own. In the beginning, in the foundations
of mathematics, you might say there are only sets, and next come spaces. In the
next stage, where the sets are “animated,” we have topology. Further development
results in greater specificity, ergo, metric spaces. One could go on to mention certain
algebras, and so forth. These, we might say, correspond to nouns. When we start
operating on them, that is, when we consider transformations and functions, it is like
introducing verbs into the language. It occurred to me long ago that many words
in the common language can, in the mind of some young person of imagination,
become germs of a mathematical theory. What is topology if not the study of an
elaboration of the word “continuous” or “continuity”? There are many other words
which could stimulate people to build theories, or at least “mini-theories.”
I can give some examples of how my own interest was originally stimulated
by problems of the Scottish Book type. For example, one important thread going
through some of the problems is the idea of “approximate,” or more properly
“epsilon-approximability” by finite or generally simpler structures. Many problems
of the book deal with properties of approximation, of reduction from the infinite to
the finite. Of course it is the finite that interests physicists, but the idea of infinity, as
all of us know, is useful because it puts in a more succinct way some properties of
very large or very small numbers, just as the infinitesimal calculus is more concise
and efficient than the calculus of finite differences. So the study of infinity per se and
its relation to finite approximation is of great interest. I am speaking vaguely here,
in general terms, but one can find many concrete examples in the Scottish Book.
Speaking of concreteness, I should like to say that there may be a more tangible
aspect to the ultra-set-theoretical investigation of very high cardinals, the incredibly
large infinities which may be measurable. These big sets, cardinal numbers in our
speculations, do throw as it were a shadow on the lower infinities. And indeed, there
are more concrete or semi-concrete formulations or expressions of mathematical
objects suggested by speculation on the existence of these superinfinities.
Speaking of “epsilons,” I want to mention a number of little amusements I have
indulged in over the years concerning what I call “epsilon stability,” not just of
equations and their solutions, but more generally of mathematical properties.
As an example of this “epsilon stability,” consider the simple functional equation:
f (x + y) = f (x) + f (y), i.e., the equation defining the automorphism of the group
of real numbers under addition. The “epsilonic” analogue of this equation is
|g(x + y) − g(x) − g(y)| < ε . The question is then: Is the solution g necessarily near
some solution f̄ of the strictly linear equation? As D. Hyers and I showed, the answer
is yes. In fact, |g − f | < ε , with the same epsilon as above. This is not a very deep
theorem. What about the more general case? Suppose I have a group for which I
replace the group operation by one that is “close” to it in some appropriate sense.
This of course requires a notion of distance in a group. The result of the replacement
is an “almost endomorphism.” Then we may ask: Is it of necessity “near” a strict
endomorphism? The answer is not known in general, even for compact groups.
Recently, D. Cenzer obtained an approximation result for some easy groups, e.g.,
the group of rotations on a circle.
In the same spirit, we may take the idea of a transformation which is an isometry,
a transformation which preserves distances. What about transformations which do
not exactly preserve distances but change them by very little, i.e., at most by a given
ε > 0? Suppose I have a transformation of a Banach space or some space which
transforms into itself, and where every distance is changed by less than a fixed ε . Is
such a transformation “near” one which is a true isometry? Hyers and I proved, in
a series of short papers, that this is true for Euclidean space, for Hilbert space, for
the C space, and so forth. If you have such a transformation, it must then be within
a fixed multiple of a strictly true isometry.
Recently I became more ambitious and looked at some other mathematical
statements from this point of view. One could try to “epsilonize” in this sense
theorems on projective geometry, on conics, and so on. More generally, take as
an example some famous theorem like the theorem on functions with an algebraic
addition. It is a well-known statement that the only functions which satisfy an
algebraic addition theorem are, in addition to sine, cosine and elementary functions,
the elliptic functions. One could ask (perhaps this question is not yet properly
formulated): Is it true that a function which “almost” satisfies an algebraic addition
theorem must be “almost” an elliptic function?
And in a similar vein: If we have a function which is differentiable, let us say five
times, and its derivative vanishes and changes sign at a point, then any sufficiently
differentiable function which is sufficiently close, in the sense of absolute value
alone, must also have a vanishing fifth derivative at a nearby point. This is almost
trivial to prove, though at first it seems false. Why is this true? Because the fifth
derivative can be obtained by finite differences. This is all very nice and easy for
functions of one variable. For functions of several variables the analog becomes
interesting and not too well-known or established. The same is true, mutatis
mutandis, for spaces of infinitely many dimensions, and is of possible interest to
physicists as a general “stability” property.
Finally, I want to mention another class of problems which appears here and there
in the Scottish Book. These problems are attempts to characterize certain spaces or
certain transformations. For example, suppose one wants to characterize the Hilbert
space among other Banach spaces by some properties of homogeneity or by the
wealth of isometric transformations into itself which it allows. There is already a
result in a finite number of dimensions due to Auerbach, Mazur, and myself on one
way to characterize an ellipsoid.
We wrote a paper where we proved that a convex body, all of whose sections
through a certain point are affine to each other, must be an ellipsoid. We did not
prove it for all dimensions, only in three dimensions. This paper appeared before
World War II in Monatshefte für Mathematik. It is merely another example of what I
mean by a characterization. Recently this topic has been developed by many people,
notably by Anderson in this country and Pelczyński in Poland.
There are other common threads going through the problems of the Scottish Book
but it would not be true to say that the problems are all cast in a similar mold. Some
are just momentary curiosities, spur-of-the-moment thoughts of the habitués of the
Scottish Café, or of casual visitors such as von Neumann. After 1939, one notes
a curious change: suddenly the contributors include Russian names, the names of
well-known mathematicians like Sobolev. This was after the city was occupied by
Russia, in September 1939.
Today more problem books are appearing. There are problem sections in the
“Notices” and in the American Mathematical Monthly. Another little Scottish
Book is being kept in Boulder, where I was a professor for some ten or twelve
years; J. Mycielski is keeping track of it. There is another one currently kept in
Wrocław, Poland—I don’t know about Hungary. Erdős has written monumental
papers containing selections or collections of problems in set theory and in number
theory. One of them, written jointly with R. Graham, is not yet published, but I have
seen the manuscript—it is a very interesting and exciting book.
Finally let me mention a few ideas which are not in the Scottish Book but which
I remember from conversations and discussions with Mazur. As an example, Mazur
and I discussed the possibility of establishing, at first only purely mathematical, but
later physical objects, which could replicate or almost replicate themselves. This
was a very sketchy and premature idea. Years later, as is well known, von Neumann
discussed this question in some detail.
We also considered the purely theoretical (at the time) possibility of comprehen-
sive computing machines. Neither of us had sufficient knowledge of electronics to
even approximate the present schemata, but we discussed the concept on a purely
abstract level.
We had some other very curious conversations. I specifically remember discus-
sions among ourselves and with visitors about what is now known as nonlinear
mathematics—truly a strange expression, for it is like saying “I will discuss nonele-
phant animals”—it was more specific than that. In fact Mazur and Orlicz had started
a study of polynomial operations; their paper appeared in Studia Mathematica. Then
we discussed iterations in one or more variables of transformations showing the sort
of phenomena which very recently I and many other mathematicians have studied
both experimentally and theoretically, and which seem to now present some interest
even for physicists. But to go into this would take me too far afield.
From the original Scottish Book; the handwriting in the Mazur-Orlicz

problem (bottom) is Mazur’s.
From the original Scottish Book; the handwriting in the Banach-Ulam

problem is Banach’s.
Chapter 2
A Personal History of the Scottish Book,
Mark Kac
It is a special pleasure to be introduced by my old friend Erdős. The use of the

adjective “old” is slightly depressing, and I would like to forget about it, but
somehow Erdős will not let me do it.
I should like to begin my remarks by pointing out the remarkable thing that we
celebrated the Scottish Book in Denton, Texas. It is remarkable not only because
of the energy, dedication and interest of one man, namely Dan Mauldin, but it is
also, for me at least, typically American. It represents the kind of combination
of generosity and sentiment which runs through the whole history of this young
civilization. I cannot think of any other country on the surface of the earth which
would be interested in celebrating a somewhat obscure event which occurred in
another country in what now seems like the dim past. And so on my own behalf,
and I am sure also on behalf of all my former and present compatriots, I would like
to express our thanks not only to Dan Mauldin but also to the spirit of America in
Denton, Texas.
Before I come to Mathematics and to my connection (tenuous as it was) with the
Scottish Book let me engage in a little of what Stan Ulam, quoting Disraeli, referred
to as “anecdotage.”
As you can see by perusing the Scottish Book, a significant number of problems
were inscribed by distinguished foreign mathematicians who passed through Lwów.
One of the most famous of these visitors and probably the most famous one, was
Henri Lebesgue.
Lebesgue came to Lwów in May 1938 to receive an honorary doctorate from
the University. At that time, since Stan Ulam, who was the Secretary of the Lwów
Section of the Polish Mathematical Society, was away in the USA, I was substituting
for him and was given the extremely pleasant job of showing Lebesgue around
the city. I reminisced about this event in 1974 in Geneva, when the centennial of
Lebesgue’s birth was celebrated. My remarks were published in L’Enseignment
Mathématique in French, and were later translated into Polish (not by me since
my knowledge of my mother tongue is no longer sufficiently reliable). Today I give

14 2 A Personal History of the Scottish Book, Mark Kac
you an abbreviated English version of these remarks. In fact I will tell you only two
stories, one of which is directly connected with the Scottish Café, the birthplace and
home of the Scottish Book.
At the time of his visit Lebesgue was no longer interested in anything but
elementary mathematics; he refused to discuss measure, integrals, projection of
Borel sets, or anything of that sort. He gave two lectures, both extremely beautiful,
but entirely elementary: one on construction by ruler and compass, and the other on
iterated radicals.1
As a footnote to the political atmosphere of those days it may be of interest to
record the following. The Polish press, which was inept above and beyond the call of
duty, confused Lebesgue with Hadamard. Hadamard was a known leftist. Lebesgue,
on the other hand, was a man of rather conservative views, though by no means a
reactionary. He was greeted upon arrival by a violent editorial against the leftist,
communist French professor being honored by the Poles. The confusion was soon
cleared up, but nobody bothered with a retraction. So you can see the press is the
same the world over, and not much has changed in this respect over the years.
As I showed Lebesgue around the city he was extremely disappointed with
me—he was very much interested in the churches, and wanted to know all about
their history, and I was unable to provide him with much information on that subject.
Lwów by the way was an extremely interesting city from the religious point of view,
because it was, with the possible exception of Jerusalem, the See of all three lines of
Catholicism. There were in fact three archbishops in Lwów, representing the Roman,
Greek, and Armenian branches.
The Armenian Cathedral, one of the most beautiful churches in Europe, espe-
cially interested Lebesgue. To his chagrin I could not tell him anything about it,
and I was equally disappointed by Lebesgue’s refusal to discuss measure, integrals,
and other mathematical topics. Still, we became reasonably friendly, and he merely
pitied me as one doomed to some terrible fate for lack of interest in history.
That afternoon we had a 5 o’clock reception for Lebesgue in the Scottish Café.
Fewer than 15 people attended, which goes to show how small the number of
mathematicians was in those days. The waiter gave all of us menus, and not realizing
that Lebesgue was not a Pole he gave him one too. Lebesgue looked at the menu for
about 30 seconds with utmost seriousness and said, “Merci, je ne mange que des
choses bien definies” (Thank you, I eat only well-defined things). At this moment I
had an inspiration, and by changing a little a well-known phrase of Poincaré directed
against Cantorism I said, “Ne mangez jamais que des objets susceptibles d’être
définis par un nombre fini de mots” (Never eat things which cannot be defined in
1 Professor Granas brought with him to the Texas conference a copy of Summaries of the
Proceedings of the meetings of the Mathematical Society in Lwów. From these documents we
can ascertain that one of the lectures took place on May 25, 1938, and was on iterated square roots.
It also turned out that my recollection as reported in my Geneva talk was not entirely correct, but
the errors are minor.
2 A Personal History of the Scottish Book, Mark Kac 15
a finite number of words). “Ah,” said Lebesgue, “you are familiar a little bit with
Poincaré’s philosophy,” and I think that he forgave me at that moment my ignorance
of the history of the Armenian Cathedral.
My second remark of an anecdotal nature has to do with the beautiful talk
by Professor Martin, my former colleague at Rockefeller University, which was
presented at the conference in Texas. It should be of interest, because it characterizes
the way Steinhaus felt about mathematics, and especially about the axiom of
determinacy. I am sure of this because I attended lectures by Steinhaus at both
Rockefeller and the Courant Institute in the early sixties. I also had many occasions
to speak to him about it.
I will now give Steinhaus’ “proof” of the determinacy of the Ulam game.
We of course all remember the Ulam game, where Player One picks a zero or one
and Player Two picks a zero or one, and one then constructs what Tony Martin called
a decimal binary (which is an excellent name for what ordinary mortals call simply
a binary). [Editor Note: The term binary decimal goes back to Turing’s classic 1937
paper on computable numbers and Hardy and Wright’s introduction to the theory of
numbers.] If it falls into a set E Player One wins, and if it is not in E Player Two
wins. The question is: Is there a winning strategy for either one of the players? Here
is a “proof” that there is one:
Let me denote by x1 , x2 , . . . the moves of Player One, and by y1 , y2 , . . . the
moves of Player Two. I will give in logical symbols, which I use very infrequently,
the statement that Player One has a winning strategy:
x1 y1 x2 y2
∃ x1 (y1 ) ∃ x2 (y2 ) · · · + + + + · · · ∈ E.
2 22 23 24
It says, “There is a first move of Player One such that for every first move of Player
Two there is a second move of Player One, such that for every second move of Player
x1 y1
Two, etc., the fraction + 2 + · · · belongs to E.” This is merely a transcription in
2 2
logical symbols of the statement that there is a strategy for Player One.
Now suppose there is no such strategy; then you put the symbol ∼ in front of the
string of quantifiers in the formula above and use the DeMorgan rule, obtaining
x1 y1 x2 y2
(x1 ) ∃ y1 (x2 ) ∃ y2 · · · + + + +··· ∈
/ E.
2 22 23 24
Now if you translate this into human language, it means that Player Two has a
winning strategy. So if Player One doesn’t have a strategy, Player Two has a strategy,
and, consequently, the axiom of determinacy in this case merely allows one to use
DeMorgan’s law for an infinite number of quantifiers. Now of course you can see
where the difficulty comes in. It is that difficulty which plagues the whole beastly
subject, and it is, namely, where you ask, “How does one know whether something
does or does not belong to set E?” It is here, of course, that we get into all the
difficulties, and Steinhaus merely felt—and I have enormous sympathy for it—that
his axiom had a chance to distinguish those sets E that are worthy to be called sets
from those that are not.
Axioms like the axiom of choice allow us—give us a legal license—to create
certain objects and then call them sets. Steinhaus thought that his axiom would be
of the kind that would distinguish between constructible and nonconstructible sets.
This little argument reminds me—and now I am only almost serious; up to this
point I was dead serious—of an imperfect analogy with what happens in quantum
mechanics where certain statements, although they sound perfectly all right, are
not allowable. For instance, when you say, “The amount of energy in a radiation
field in a subvolume,” then it sounds like a perfectly well-defined thing. But if you
really follow the dicta of quantum mechanics, you have to express it in terms of a
Hermitian operator—every physical quantity has to be represented by a Hermitian
operator—and it turns out that it is not unique. In fact, how to interpret this may
very well depend on the method of measurement. You have something of the sort
here—nothing is really defined until you come to grips with saying, “How do you
know whether a number constructed by an infinite number of operations does or
does not belong to a set?”
Now, one final observation in connection with other people’s involvement in the
Scottish Book Conference, namely with Professor Zygmund’s, who referred in his
talk to one of the greatest Polish discoveries, the category method. As a matter
of fact, this discovery is so well known that one does not even recognize what a
remarkable discovery it was. It was remarkable because it showed that sometimes it
is easier to prove that most objects have a certain property than to exhibit a particular
example.
Professor Zygmund asked about the rearrangement of the Fourier series in
connection with the question of convergence, and bemoaned the fact, which many
of us bemoan, that there is no decent, sensible measure in the set of all permutations.
However, if one goes back to the Polish invention of the method of category, then
of course the set of all permutations can be easily metrized by the Frechet trick.
Consequently, the concept of sets of first and second category is perfectly well
defined. There is in fact a book by Professor Oxtoby, who attended the conference
(and I even ascertained from him that it was published in 1971 by Springer-Verlag),
called Measure and Category. The message of the book is that whenever both
can be defined and whenever the measure is reasonable, then second category
and measure one, other than in very exceptional situations, are the same. One can
rephrase Professor Zygmund’s question to ask whether the set of all permutations of
Fourier series which lead to divergence is of second category. A very simple case—
similar abut much simpler—was considered by my colleague at the time, and still
a good friend, Professor Ralph Palmer Agnew of Cornell, in response to a question
posed during a conversation we had many years ago. If you take a conditionally
convergent series of real numbers, then of course we know that it can be rearranged
so as to make it converge to any prescribed number, and it can also be rearranged
into a divergent series. Now it is easy to prove, and in fact Agnew proved it (it was
published around 1940 in the Bulletin of the American Mathematical Society) that
the set of permutations which lead to divergent rearrangements is indeed of second
category. You might say that everything bad which one might expect to happen is
going to happen in a plentiful sort of way.
Now to some of the more personal things. I am not really, in a certain sense,
a product, or at least not a typical product, of the Polish school. When I came to
Lwów as a student in October 1931, I did not know any of the great masters; my
first contact was with the late Marceli Stark, a remarkable man and a tremendously
well-educated mathematician who died recently and to whose memory I would like
to pay tribute. I was very concretely minded, and I still am—in fact even more so.
Yet I felt a little bit that I also ought to do these abstract things, and Steinhaus, whom
I met a little later, said, “You shouldn’t; you must earn the right to generalize.” I have
not yet earned that right.
I became interested in probability theory in a way that I am not even going to tell
you in detail, because I can’t give you a full autobiography. Some day I am going
to get even with Stan Ulam and write my own adventures, which, however, are not
nearly as exciting as his.
It was through Steinhaus that I became interested in probability theory, and,
with the exception of one problem (I put altogether four problems into the Scottish
Book—numbers 126, 161, 177, and 178) and I really do not know why I put it in; it is
not even properly stated—these problems deal directly or indirectly with probability
theory. The first one is a minor technicality, which Hinčin proved in response to a
letter.
The problem that I cannot for the life of me remember how and why I thought
of it, is the problem of characterizing continuous functions, φ (x, y), such that if A
and B are real symmetric matrices, then φ is positive definite (Problem no. 177).
Now, because of noncommutativity, φ (A, B) is not properly defined. But that is
easily remedied if φ is a polynomial in two variables—one simply replaces φ by
a symmetrized polynomial, in which case it makes perfect sense, and the question
can still be asked. Whether it is of any interest I have no idea. I do not have
any recollection as to why it interested me at the time, and I probably should
have appealed to Dan Mauldin to put this problem in a footnote because there
is no particular reason to bother the next generation with this one—unless in the
meantime I remember what it was I really wanted.
The first, as I have already told you, was a minor technical problem, but the fourth
(Problem no. 178) has a certain degree of interest, and I may as well say what it is.
It is unsolved not because it is necessarily difficult, but because nobody has tried.
I am not going to give any prizes for it. It might, however, be of some interest to
those of you who are analytically-minded.
There is a well-known theorem of Cramér that if a product of two characteristic
functions φ1 (ξ ), φ2 (ξ ) is exp(−ξ 2 /2) then both φ1 and φ2 must themselves be
Gaussian, i.e.,
φ1 = exp(−α1 ξ 2 + β1 ξ )
φ2 = exp(−α2 ξ 2 + β2 ξ )
with α1 + α2 = 1/2 and β1 + β2 = 0. (In Problem 178 the theorem is slightly

misstated.)
In probabilistic terms, if a sum of two independent random variable is Gaussian,

then the random variables themselves must be Gaussian. Similar theorems hold
for stable distributions. My Problem 178 raised the question of whether other
distributions can be similarly characterized. One must, of course, get away from
the product since the product is intimately tied to addition of random variables and
therefore to stable distributions, and I hit upon
−1
1 1
+ −1
x y
as a candidate for the characterization of the class of characteristic functions

1
, α > 0.
1 + αξ 2
The problem is closely related to the following problem which is perhaps of greater
general interest:
What are the functions F(x, y) of two variables such that F(φ1 (ξ ), φ2 (ξ )) is a
characteristic function of a probability distribution whenever φ1 and φ2 are?
I strongly suspect that F must be a function of the product xy, i.e., F(x, y) ≡
G(x, y) with G satisfying some additional conditions, but I have no idea how to go
about proving it.
The only one of my four problems which was destined to have a future was
Problem 161. There is not much point in going into details since an interested
reader can consult my 1949 address, “Probability Methods in some problems of
analysis and number theory” (Bull. Am. Math. Soc. 55, (1949), 390–408). Echoes
of this problem are still reverberating, as witness a recent paper by I. Berkes,
“A Central Limit Theorem for Trigonometric Series with Small Gaps,” (Z. für
Wahrsch., 47 (1979), 157–161), but the original source will only become known
with the publication of the Scottish Book. As it is, not even my 1949 address is
cited, which is some kind of a price one must pay for pioneering.
Problem 161 bears the date June 10, 1937, which was five days after I repeated
the ancient oath, “Spondeo ac polliceor . . .” and was awarded the degree of Doctor
of Philosophy of the John Casimir University in Lwów. Actually, not knowing Latin,
I got into my head that in spondeo the accent is on the second syllable and not,
as is correct, on the first. Steinhaus, who was my sponsor (promotor) and who
was a stickler for proper usage of all languages, used to make me practice the
correct pronunciation before the actual ceremony. When the moment arrived for
me to reply to a Latin oath read with pomp (though not with pomposity) by the
Rector Magnificus I forgot all the practice and put an emphatic accent on the wrong
(second) syllable. Steinhaus cringed and so did my father, who knew Latin and who
journeyed to Lwów to witness the occasion.
Returning to the Scottish Book, I would like to point out that although the
problems in it range over most of the principal branches of Mathematics, one branch
is conspicuously absent, and that is Number Theory. The reason is simple, and it
is that Number Theory was not in vogue in Poland at the time. Sierpiński in his
younger years (and also toward the end of his life) did important and interesting
things in Number Theory, and the Warsaw school did produce two “mutants”:
A. Walfisz (who left Poland for the Soviet Union and was Professor at Tiblisi) and
S. Lubelski. There was even a serious journal, Acta Arithmetica (which continues to
this day), devoted to Number Theory, but this beautiful and important area was far
from the forefront of mathematical preoccupation in Poland before World War II.
I cannot remember at all how I came to think about number theoretic problems
in connection with Probability Theory, but I do remember making what appeared to
me then to be a great discovery (it wasn’t).
If φ (n) is the familiar Euler function, one has

φ (n) 1
= ∏ 1−
n p|n
p
which can be written in the form

φ (n) ρp (n)
= ∏ 1−
n p p
where

1, p | n,
ρp (n) =
0, p n.
This of course was a well-known elementary fact, but the method also yielded at
once

φ (n) ρp (n)
M = ∏ M 1−
n p p

1 1 1
= ∏ 1− + 1− ,
p p p p
for all such that the infinite product converges, and hence one had a handle on the
distribution of φ (n)/n.
When late in November 1938 I left for the United States, the boat (M.S. Pilsudski,
sunk in the early days of World War II) stopped for about six hours in Copenhagen,
which gave me a chance to meet Professor Børge Jessen. I communicated my
number theoretic discovery to him only to learn that the same result had been
obtained and already published by I.J. Schoenberg. The probabilistic nature of
the result was, however, somewhat hidden in Schoenberg’s proof, and I had
the advantage (because of my deep involvement with the normal distribution in
unexpected contexts, as illustrated by Problem 161) of being—so to speak—on

the ground floor. It was therefore a small step to suspect that the number of prime
divisors ν (n) of n given by the formula
ν (n) = ∑ ρp (n)
should behave like a sum of independent random variables and hence be normally
distributed after subtracting an appropriate
√ mean (log log n) and scaling down by
an appropriate standard deviation ( log log n). But here my ignorance of Number
Theory proved an impediment. The number of terms in the sum ∑ ρp (n) depends on
n, preventing a straightforward application of the Central Limit Theorem. I struggled
unsuccessfully with the problem until I stated my difficulties during a lecture in
March 1939 in Princeton. Fortunately Erdős was in the audience and he perked up
at the mention of Number Theory. He made me repeat my problem, and before the
lecture was over he had a proof. Thus did the Normal Distribution enter Number
Theory and thus was born its probabilistic branch. While stretching a bit the
historical truth I hereby assign the role of godmother of this branch to the Scottish
Book.
Chapter 3
Steinhaus and the Development of Polish
Mathematics, A. Zygmund
The origin and history of the Scottish Book is described by Professor Ulam in his
own lecture and I could not add much here.
The book is a product of one of the mathematical schools in Poland, that of
Lwów, while I myself, born and educated in Warsaw, belonged to what was then
known, both in Poland and abroad, as the Warsaw mathematical school. There was
a close collaboration between individuals of both schools, and though my personal
contact with Lwów was rather loose, I was very much interested in the work going
on there, and it had considerable influence on my own work.
In what follows I shall give a few facts about the development of Polish
Mathematics, limiting myself to those which have some pertinence to the Scottish
Book.
The Polish mathematical school of the period 1919–1939 was an interesting
phenomenon, first because of its achievements, and secondly because of the place
and circumstances in which it arose. One might say that before 1919 there had
been Polish mathematicians but there was no Polish mathematical school. The
rapid growth of Polish Mathematics after 1919 was partly spontaneous, helped by
the recent freeing of the country from foreign occupation, and partly a result of
thoughtful planning.
The development of Polish Mathematics was in the first place due to Janiszewski,
Mazurkiewicz, and Sierpiński in Warsaw and to Banach and Steinhaus in Lwów.
The role of Janiszewski here was particularly significant and unique. Born in 1888,
he died in 1920 and so did not live to see the fruition of his ideas, but he was the
chief planner of the Polish school. A talented mathematician (topologist) himself,
he realized the difficulties of organizing good mathematical research in a country
without a strong and continuous mathematical tradition. His idea was that the surest
and quickest way to success would be first through concentration on a particular
mathematical discipline which would be the main source of interest and of problems
for a larger group of mathematicians, and secondly, through starting a mathematical

22 3 Steinhaus and the Development of Polish Mathematics, A. Zygmund
publication specializing in this selected branch of mathematics. Once a strong point

was established, gradual extension of interest to other fields of mathematics was
expected.
At that time the theory of Sets, Topology, and Real Variables were attracting a
number of Polish mathematicians. It was natural to make a starting point here and
in 1920 the first volume of the publication Fundamenta Mathematicae appeared
in Warsaw. It was a success from the start. It gave an outlet to Polish mathematical
production and attracted foreign papers. Before September 1939, thirty-two volumes
of Fundamenta had been published.
Let me pass to another Polish mathematical school, that of Lwów. (After 1945
the city of Lwów was no longer within the boundaries of Poland.) When we think
of Lwów mathematics, two names usually come to our minds. One of them is
Stefan Banach; the other is Hugo Steinhaus, Banach’s teacher and later collaborator.
While the importance of Banach’s mathematical work is widely recognized, and
the name is essentially an adjective in Mathematics, few people outside Poland
appreciate the importance of Steinhaus’ influence on Polish Mathematics. Without
Steinhaus, Banach as we know him probably would not have existed, and Polish
Mathematics would have had a different character. It is for this reason that I would
like to devote most of the time at my disposal to the role played by Steinhaus in the
development of Polish Mathematics. It is my personal feeling that despite generally
high respect for his work, Steinhaus’ role is not sufficiently appreciated here. In
what follows, I would like to indicate some of the achievements of Steinhaus and
his collaborators.
Born in 1888, Steinhaus studied in Germany and Paris before the outbreak of
the first World War. When he returned to Poland, just before the war, he was
appointed first a docent and then a professor at the University of Lwów. That was the
beginning of his impact on Polish mathematics, for he brought from abroad, to what
was a rather provincial mathematical milieu, not only new ideas but also personal
contacts with outstanding foreign mathematicians, which were very beneficial to
Polish mathematics and contributed very much to its development. Let me illustrate
this by one story.
While in Germany, Steinhaus had become a personal friend of Otto Toeplitz,
a German mathematician who was also partly of Polish origin. Under Steinhaus’
influence, Toeplitz published a short paper in a relatively little-known Polish
mathematical periodical Prace Matematyczno-Fizyczne, which mostly published
Polish papers. The title of the paper was “Über lineare Mittelbildungen”; it appeared
in Volume 22 (1911) of the Prace and is essentially a paper about the method of
condensation of singularities. I was curious that a German mathematician should
want to publish such a paper in a very little-known Polish journal. But looking
back, one may say that this was an important step in the development of modern
functional analysis. Let me be more specific.
The main result of Toeplitz was as follows. Given a matrix {amn } (m, n = 0, 1, . . .)
of real or complex numbers, we may associate with every numerical sequence {sn },
n = 0, 1, . . . a transformed sequence
tm = ∑{amn sn } (m = 0, 1, . . .).
n
3 Steinhaus and the Development of Polish Mathematics, A. Zygmund 23
The problem was to find necessary and sufficient conditions for the matrix {amn }
to have the property that every convergent sequence {sn } is transformed into a
convergent sequence {tn }. The problem is, in today’s perspective, very elementary
and Toeplitz in his paper (that was in 1911) gave such necessary and sufficient
conditions. One of those conditions, the basic one, is very familiar by now. It is
∑ |amn | ≤ constant, for all m.

n
Toeplitz proved both the necessity and the sufficiency of his conditions. Sufficiency
alone was proved independently and at about the same time by the American
mathematician Silverman, and the theorem itself is occasionally quoted as the
Toeplitz-Silverman theorem.
Obviously, each tm is a linear operation defined in the space of sequences {sn },
and for mathematicians in Lwów interested in functional analysis Toeplitz’ result
raised a question of abstract generalizations. In 1928, Banach and Steinhaus sent
a paper to Fundamenta giving one such generalization. The main result of the
paper was as follows: Let {um (x)} be a sequence of bounded linear operations
defined in a normed linear space E, and let Mum be the norm of the operation um .
If supm um (x) is finite for every point x belonging to a set F of the second category
in E (in particular, if it is finite for every x ∈ E), then the sequence Mum is bounded.
In other words, there is a constant M such that
{|um (x)| ≤ M|x| for x ∈ E and m = 1, 2, . . .}.
Of course, the result of Toeplitz is a consequence of this general theorem. Sierpiński,

the editor of Fundamenta, gave a paper to Saks, his former pupil, for refereeing,
and I remember that Saks showed me the manuscript and pointed out that the
argument could be much simplified by replacing the rather cumbersome method
of condensation of singularities by the application of the notion of sets of second
category. For example, if the functionals are not uniformly bounded, then by merely
considering sets of first and second category one can prove, without computation,
the existence of a point at which all the functionals are unbounded. The paper
appeared in a revised form in Fundamenta in 1928, and marks an important point
in the development of functional analysis through the application of sets of first and
second category. It is perhaps regrettable that the paper, rewritten by Saks, nowhere
mentions the fact that it was he who introduced the new method, and the authorship
of the method remains unknown except to very few people.
Let me mention another result (due to Steinhaus himself) which had considerable
influence upon my own work. The story begins with a theorem of Hurwitz which
gives the following. Suppose we have a power series ∑ cn zn of radius of convergence
equal to 1. The function then must have at least one singularity on the circle of
convergence. Hurwitz proved that if we select a suitable sequence {±1}, the series
∑(±cn )zn is nowhere continuable across the circle |z| = 1. In this connection,
Steinhaus proved the following: if instead of ±1 we introduce a Gaussian random
24 3 Steinhaus and the Development of Polish Mathematics, A. Zygmund
variable, then what happened in the case of Hurwitz for a particular sequence of
signs becomes true, in the new situation, with probability 1. In other words, if we
introduce Gaussian random variable into the coefficients of a power series of finite
radius, then with probability 1 this series becomes nowhere continuable, and what
was initially an individual situation, which occurred due to a special selection of
the values of the random variable, tends to be a general phenomenon. This result
of Steinhaus was the beginning of a certain development, randomization of series,
which plays a distinctive role in the theory of functions, both of real and complex
variable. Steinhaus used a special method and for this reason he had to use Gaussian
random variables, but it turns out that this is merely a special case of a much more
general theorem. For example, under the assumptions of the theorem of Hurwitz
almost all series ∑(±cn )zn are nowhere continuable.
It was Steinhaus’ idea to introduce methods of probability into construction of
functions with required properties. Here the property was noncontinuability, but
there are many similar situations. Let me describe one which is elementary but very
useful.
Let φ0 (t) be a function of period 1 which is equal to 1 for 0 ≤ t < 1/2 and to −1
for 1/2 ≤ t < 1. Let φn (t) = φ0 (2n t), n = 0, 1, . . .. The φn (t)—called Rademacher
functions—form an orthonormal system on 0 ≤ t ≤ 1 and are known to possess
the following properties: For any sequence {cn }, n = 0, 1, 2, . . . of real or complex
numbers, if ∑ |cn |2 < ∞, then the series ∑ cn φn converges almost everywhere and
its sum is in Lp on the interval 0 ≤ t ≤ 1, no matter how large p is (it is even
exponentially integrable). If, on the contrary, ∑ |cn |2 = ∞, then ∑ cn φn not only
diverges almost everywhere but is almost everywhere nonsummable by any linear
method of summability. Consider now any trigonometric series
∞
(1/2)a0 + ∑ (an cos nx + bn sin nx)
n=1
with, say, real coefficients. Then using very elementary methods, one can show that
if ∑(a2n + b2n ) < ∞, then almost all series St = ∑ φn (t)(an cos nx + bn sin nx) converge
almost everywhere and are in the class Lp for every finite p, and if ∑(a2n + b2n ) = ∞
then almost all St are nonsummable by any linear method of summability, and in
particular, are not Fourier series. The situation is rather typical and the method when
applied to various series leads to examples illustrating various points of the theory
of functional developments. It is an important method in the theory of Fourier series.
Let me mention in this connection one problem to which I do not know the
answer but which intrigues me. There is a celebrated theorem of Carleson which
says that the Fourier series of a function in L2 converges almost everywhere. On the
other hand, Kolmogorov and Zahorski showed that there is an L2 function whose
Fourier series suitably rearranged diverges almost everywhere. Thus if we do not fix
the order of terms in a Fourier series, we may have convergence almost everywhere
as well as divergence almost everywhere. The question naturally arises which
situation occurs “more frequently.” The question may be a little foolish and may
3 Steinhaus and the Development of Polish Mathematics, A. Zygmund 25
have no obvious answer since we have no measure in the space of rearrangements

of natural numbers. Still, it is of a certain interest since in analysis we have situations
when a sequence of functions has no “natural” ordering (take, for example, a general
orthogonal system).
The school of Lwów is technically no longer in existence and its organ Studia
Mathematica, begun in 1932, is now being published in Warsaw. But the influence
of the work of its founders and their pupils continues and grows in various
Polish mathematical centers. The names of Banach, Steinhaus, Schauder, Kaczmarz,
Auerbach, Ulam, Mazur, Orlicz, Nikliborc, Schreier, Ruziewicz, Kac, and others
symbolize the achievements of this school.
Chapter 4
My Scottish Book ‘Problems’, Paul Erdös
I shall discuss several problems which are connected with the Scottish Book.
Let me start with a problem which F. Bagemihl and I solved. Everybody knows
Riemann’s theorem: A nonabsolutely convergent series of real numbers has the
property that for any preassigned number a the series can be reordered to converge
to a. Bagemihl and I proved that for the Cesàro sum of the series there are three
possibilities: On reordering the domain of convergence is just one point or it is
the whole line, or it is an arithmetic progression (this last possibility does not
occur in Riemann’s theorem). Our paper appeared in Acta Mathematica in 1954;
later we found that it is Problem 28 of the Scottish Book, due to Mazur. Some
interesting questions remain. First of all one could investigate what happens with
other summability methods, for example with Ck , the kth Cesàro mean, and more
complicated summability methods. Lorentz and Zeller proved that for an arbitrary
analytic set there is a matrix summability method such that by reordering one can
get that analytic set. But this still leaves the problem of what happens if one uses a
decent summability method like Ck or Abel or some other fixed scheme. Bagemihl
and I did not investigate what happens for complex series under reordering and
Cesàro summability. There are interesting possibilities here. For example, there is
a very pretty theorem by Steinitz: for a reordered complex series, there are three
possibilities for the convergent sums; they constitute (a) a single point, (b) a flat, or
(c) the whole complex plane. The analogue for n-dimensional vectors also holds.
I believe it was Banach who raised the question as to whether this theorem can
be generalized to function spaces, including, of course, Hilbert space. This was
answered in the negative by Marcinkiewicz or Mazur.
The Scottish Book’s Problem 8, due to Mazur, is a very nice question. There is
a classical theorem which states that the Cauchy product of two convergent series
U and V need not be convergent, but the product series is always C1 -summable to
the sum UV. Mazur asks the converse: Is every series summable by the first mean
representable as the Cauchy product of two convergent series? I tried to do this but
I couldn’t, and it should be looked at by somebody who knows more about it than I.

28 4 My Scottish Book ‘Problems’, Paul Erdös
Problems 22 (Ulam-Schreier) and 99 (Ulam) are as follows:

(Problem 22) Is every set z of real numbers a Borel set with respect to sets G which
are additive groups of real numbers?
(Problem 99) Can every set in the plane be gotten by Borel operations on squares?
Ulam and I settled many of these questions long ago, but we never got around to
publishing our results. These results were rediscovered and published by the Indian
mathematician B.V. Rao; when Rao sent me a preprint I urged him to publish.
Naturally, I did not tell him that Ulam and I had already done the work. Eventually
he found out, however, and asked me why I hadn’t said anything about it. I replied
that this was the one respect in which I did not want to imitate Gauss, who had the
nasty habit of “putting down” younger mathematicians by telling them he had long
ago obtained their supposedly new results.
I have some remarks on Problem 88. This problem is a curious question about
infinite series, due to Mazur:
Consider a sequence of numbers {an }∞ n=1 with the following property: if x1 , x2 , . . .
is a bounded sequence, then

∞ ∞ ∞

∑ ai xi + ∑ ai+1 xi + ∑ ai+2 xi + · · ·
i=1 i=1 i=1
converges. Is it true that ∑n=1 n|an | converges?
I have no idea why it should be true and I haven’t been able to settle it. [Editor’s
Note: The problem has been solved; see Commentary.]
Now let me talk about some of the problems which don’t seem to be very difficult
but still may be of some interest even now, after many years. One of these is a
very pretty conjecture by Borsuk which says that if one has a set of diameter 1 in
n-dimensional space it can be decomposed into n + 1 sets of diameter < 1. This
is trivial on the line, easy in the plane, difficult in 3-space, and unsolved higher
(I suspect that it is false for sufficiently high dimension). There is another difference
between two and three dimensions. For two dimensions one knows the extremal
solution to be an equilateral triangle. The decomposition is as follows: Construct
the circumscribed circle about the equilateral triangle, and draw three radii so that
they divide the circle into three equal areas.
4 My Scottish Book ‘Problems’, Paul Erdös 29
II III
This induces the desired decomposition of the triangle—the three sets are
congruent and have diameter about .88, and that is the extreme situation.
Now in three dimensions nobody knows what the extreme result is. The three-
dimensional case was done by Eggleston and then independently and much more
simply by Grünbaum and Heppes.
There is a very simple theorem of Steinhaus which says that the difference set of a
set of positive measure (say on the line) contains an interval. This is an almost trivial
theorem by our present standards. It follows instantly from the Lebesgue density
theorem, and therefore by this method one obtains the following theorem: Any set
of positive measure on the line, or, more generally, in k-dimensional space, contains
all finite sets in the space to within a similarity transformation. The proof is almost
immediate because by the Lebesgue density theorem there is an interval or a sphere
in which the density is as close to 1 as one wishes, and therefore it follows that set
will contain a set which is similar to any finite set. A related question is Problem 146
(Ulam), to which the answer is negative. For a set of positive measure (on the line,
say) one can find an interval in which the density is > 1 − ε . Problem 146 asks: Can
one determine how fast the density will tend to 1 as a function of the length of the
interval? It is easy to see that the answer is negative—no general statement can be
made as to how fast the density will tend to 1.
I have the following problem; perhaps it is easy, but it has remained unsolved
for so long that I should offer $100. Consider a sequence of positive reals {xn }∞ n=1 ,
with xn → 0. Does there exist a set of positive measure which does not contain a
set similar to this sequence? If the answer is yes, it would show that this simple
extension of Steinhaus’ theorem does not hold for infinite sets. In this case one
could ask for the minimum of the measure of a set which has this property. I don’t
think the problem is difficult, but perhaps it is not quite trivial.
An amusing consequence of Steinhaus’ theorem is the following: A set of infinite
measure in the plane contains the vertices of some triangle with a preassigned area.
If one wants to find a triangle of area 1 whose vertices are required to lie in the set,
it is easy to do it. The same is true of a set E in the plane which has a line which
E intersects in a set of positive measure, and which has points arbitrarily far from
the line, i.e., it contains triangles of any area. This is an immediate consequence of
Steinhaus’ theorem; I want to pose a slightly different problem.
Is it true that there is an absolute constant c so that a set with planar measure
> c contains three points which form a triangle with area 1? I don’t know what
the answer is. The extremal case might take the form: Choose a circle so that the
inscribed equilateral triangle has area 1, and take the interior of that circle. Then
this set will not contain a triangle with area 1 and the corresponding area may be the
minimum value of c. This surely should be disprovable if it is false.
Incidentally, there is an interesting question of Ulam and myself, the proof of
which is lost. We had the following result: Take an ideal in the integers and take the
Boolean algebra modulo that ideal—for example, one can take the Boolean algebra
of subsets of integers modulo the finite sets. In other words, two subsets are distinct
if they differ by an infinite set. First of all we wanted to prove that there are 2c
nonisomorphic ideals; we didn’t completely do this, but it has been done in the
meantime.
The thing which is lost is the following: Consider the ideal of the finite sets, the
ideal of the sets of density zero, and the ideal of the sets of logarithmic density zero.
Now it is clear that if a sequence has density 0, then it has logarithmic density 0, and
it is clear that the converse is false. (The integers between n! and 2(n!) clearly don’t
have density 0 and clearly do have logarithmic density 0. By clearly, I mean a good
freshman should be able to do it, although it’s not completely trivial.) We proved
easily that the algebra modulo finite sets is not isomorphic to the other two because
the Boolean algebra modulo finite sets has no upper bound and the other two have.
Now we allegedly proved that the Boolean algebra modulo the sequences of density
0 and logarithmic density 0 are not isomorphic. When I first visited Ulam in 1943 or
1944 in Madison we had the proof, then six months later we had forgotten the proof,
and had to reconstruct it, so it seems that the proof should have been correct. Now
the proof is gone and nobody can prove it. This problem should be settled; perhaps
I should offer a hundred dollars for a proof (or a disproof) that these two Boolean
algebras are not isomorphic. If it is trivial I well deserve to have to pay the hundred
dollars.
I also want to mention a very nice problem of Tarski which should be settled:
squaring the circle. Can a square and a circle of the same area be decomposed
into a finite number of congruent parts? This is a very beautiful problem, and
rather well known. If it were my problem I would offer $1000 for it—a very nice
question, possibly very difficult. Really one has no obvious method of attack. In
higher dimensions this is no longer true. As everybody knows, this is the famous
Banach-Tarski paradox, the basic idea of which really goes back to Hausdorff in his
1914 book. [Ed. Note: This problem was solved by M. Laczkovich in his beautiful
paper, Equidecomposability and discrepancy: a solution to Tarksi’s circle squaring
problem, J. Reine Angew. Math. 404 (1990), 77–117. His solution has led to many
research directions today (2015).]
Now let me say a few things about the Cauchy equation f (x + y) = f (x) + f (y).
Assume f (x + y) − f (x) is a continuous function of x for every h. I conjectured
that f (x) = g(x) + h(x), where g(x) is continuous and h(x) is a Hamel function.
I didn’t know how to prove it, so I did the next best thing—I guessed who would
be able to prove it. I wrote to de Bruijn, and he proved my conjecture. The paper
appeared in the Niewu Archief vorr Wiskunde about 28 years ago. I also conjectured
that if f (x + h) − f (x) is a measurable function of x, for every h, then f (x) can
be decomposed into three parts: g(x) measurable, h(x) Hamel, and r(x), where
r(x + h) − r(x) is zero almost everywhere. This conjecture has recently been proved
by Laczkovich, a young Hungarian mathematician. [Ed. Note. See M. Laczkovich,
Functions with measurable differences, Acta Math. Acad. Sci. Hungar. 35 (1980),
217–235. Laczkovich wrote a survey about all of this: The difference property. In:
Paul Erdős and his Mathematics (editors: G. Halász, L. Lovász, M. Simonovits
and V. T. Sós), Springer, 2002, Vol. I, 363–410.] The nicest problem here is due to
Kemperman; if it were my problem I would offer $500 for it. This is the problem
of Kemperman: If for every x and positive h, 2f (x) ≤ f (x + h) + f (x + 2h), then f (x)
is monotonic. Now at first sight this seems harmless—it looks as though anyone
could prove it or find a counterexample, but nobody has succeeded. It is rather
easy to show that if f (x) is measurable and satisfies the above condition property
it must be monotonic. That is a simple exercise, but one can define such a function
which is not monotonic on the rational numbers, so one has to use more than just a
countable subset, and this is all I know about it. Nobody else has made any progress
on this problem—the question remains open. I think it is very surprising that this
should be so difficult. [Ed. Note. This problem was solved by M. Laczkovich, On
Kemperman’s inequality 2f (x) ≤ f (x + h) + f (x + 2h). Colloq. Math. 49, (1984),
109–115, also see his paper: A generalization of Kemperman’s functional inequality,
2f (x) ≤ f (x + h) + f (x + 2h). General Inequalities 3 Oberwolfach, 1981, Proceed-
ings, edited by E. F. Beckenbach; Birkháuser, (1983), 281–293. As Laczkovich
points out these papers deal with the following general unsolved problem. Suppose
f is a function defined on the real line satisfying the inequality
a1 f (x + b1 h) + · · · + an f (x + bn h) ≥ 0
for every real x and nonnegative h. For which values ai , bi does this condition imply
that f is monotonic?]
Let me talk about a different kind of problem which still bears some relation
to the problems discussed above. Kakutani and I proved in 1942 the following the
theorem: c = ℵ1 ⇐⇒ the real line (continuum) can be decomposed as the union of
ℵ0 Hamel bases. In other words, if c = ℵ1 , then one can decompose the real line into
countably many rationally independent sets, and conversely. This is not very hard
to prove; it appeared in in Bulletin of the AMS in 1943. Now I ask the following
question: Can an n-dimensional Euclidean space be decomposed into countably
many sets so that each set has the property that the distances are all distinct? The
relation of this to the previous theorem is that if one decomposes the real line into
ℵ0 Hamel bases and if one takes a Hamel basis, then all distances will be distinct
between two points of the Hamel basis, because of the rational independence. In a
Hamel basis, with elements a, the numbers a1 − a2 are all distinct. So for the real
line the theorem of Kakutani and myself settles it, but for the plane there are already
great difficulties. Nevertheless R.O. Davies, the English mathematician, succeeded

in carrying out the proof. Recently I had a letter from K. Kunen in which he says
that he has proved my conjecture for n-dimensional space, and that the proof is very
complicated. [See K. Kunen, Partitioning Euclidean space. Math. Proc. Cambridge
Philos. Soc. 102 (1987), 379–383.] So it seems this problem is settled now. For every
n, n-dimensional space can be decomposed into the union of countably many sets
Sk where each Sk has the property that any four points give six distinct distances; in
other words, all the distances are defined. The continuum hypothesis must be used;
without the continuum hypothesis, it cannot possibly be true. It is already wrong
for the line. If c > ℵ1 and one decomposes (theorem of Hajnal and myself) the
real line into ℵ0 sets, then there are always four points which determine only four
distances, i.e., at least one of the sets contains four points which determine only
four distances. It will contain, namely, the following configuration: • • • •. This
is not hard to prove, using a partition theorem due to Hajnal and myself. If each
point of a set of size ℵ1 is connected to every point of a set of size ℵ2 and the edges
of the resulting graph are colored with ℵ0 colors, then there is a monochromatic
circuit of length four in the graph. From this one can obtain the above configuration.
Now in Hilbert space the situation is completely different. Hajnal and I have an easy
example in Hilbert space of a set of power c where all triangles are isosceles, so in
this case one can’t even find three points where all distances are distinct. I asked, and
Posa settled, the following question: Is there a set of power c in Hilbert space so that
every subset of power c contains an n-dimensional regular simplex? Every subset of
power c contains an equilateral triangle or regular simplex and this is true even for
infinite dimensional simplices if the continuum hypothesis is assumed, so Hilbert
space behaves completely differently even in this simple case. Now it frequently
happens in problems of this sort that the infinite dimensional case is easier to settle
than the finite dimensional analogues. This moved Ulam and me to paraphrase a
well-known maxim of the American armed forces in World War II: “The difficult
we do immediately, the impossible takes a little longer,” viz: “The infinite we do
immediately, the finite takes a little longer.”
There is a beautiful theorem of Sierpiński. I remember how surprised I was when
I first saw it. If c = ℵ1 the plane can be decomposed into two sets S1 and S2 so that
every vertical line meets S1 in a countable set and every horizontal line meets S2
in a countable set. It is a very simple theorem by present standards but it was very
startling then. I made the following generalization, which is also very simple. Split
the lines into two classes. Then if c = ℵ1 one can decompose the plane into two sets
so that S1 meets every line of the first class in a countable set and S2 meets every line
of the second class in a countable set. Thus one is not restricted to the vertical and
horizontal lines. The proof is almost trivial but rather standard, and there are various
generalizations. R.O. Davies has investigated and settled almost all the problems
here.
Now there is a very pretty three-dimensional generalization of Sierpiński which
goes as follows: If c = ℵ1 , three-dimensional space can be decomposed into three
sets, E1 , E2 , E3 , so that every line parallel to the x-axis meets E1 in a finite set,
every line parallel to the y-axis meets E2 in a finite set, and every line parallel to
the z-axis meets E3 in a finite set. It is a very pretty theorem. This is necessary
and sufficient for c = ℵ1 . After the war Sierpiński returned to elementary number
theory, which was his first love. He did his first work in number theory and his
last work in number theory: in between he did set theory and real functions. This,
however, was one of the few really new things which he did after the war. Actually
when I first lectured in Poland in 1956 on the partition calculus of Rado and myself,
certain things were called Sierpińskizations. But Sierpiński was not very interested
in that—he really was very much more interested by that time in number theory.
At any rate, this is a very pretty theorem. Kuratowski has various generalizations,
and Hajnal and I raised the following problem which was one of the few problems
in partition calculus which was unsolved until very recently. This is the problem
(I offered $50 for it): Choose three sets of power ℵ1 : A, B, and C. Take the triples
(x, y, z), x ∈ A, y ∈ B, z ∈ C, and decompose them in an arbitrary way into two
classes. Then is it true that there is a set A1 of size ℵ0 in A, a set B1 of size ℵ0 in B,
and a set C1 of size ℵ0 in C so that all triples from A1 , B1 , and C1 are in the same
class? That was one question which remained unsolved. During the last meeting on
logic and set theory in Cambridge, Prikry and Mills thought of this negatively, and
they disproved it. I think the paper will appear soon.
I will close with some comments on number theory. In 1947 there was a meeting
in Oslo, and MacLane handed me a paper to referee, by a young man called Mills,
the son of the older Mills. He proved the following theorem: There is a real number
c > 1 so that [c3n ] is a prime, for every n. I was very excited just because he
had written down an arbitrarily large prime which seems far better than humanity
deserves in this case. Well, I was a little disappointed when I looked at the paper,
which was very nice and the first of its kind, but in a way it was cheating—it has
nothing to do with primes. All one needs to know about the primes is that there is a
prime between two consecutive cubes and then one constructs the c by a descending
sequence of integers which are products of primes. So actually one doesn’t get a
single new prime. It is a nice remark but it is useless for the theory of primes. The
existence of a polynomial of many variables which whenever positive is also a prime
is of no use to number theory. It tells nothing about primes and a similar result holds
for any recursively enumerable set.
Chapter 5
KKM-Maps, Andrzej Granas
5.1 Definition and examples
Let E be a real vector space and X ⊂ E be an arbitrary subset. A set-valued function

G : X → 2E is called a Knaster-Kuratowski-Mazurkiewicz map or simply a KKM-
map1 if
s
conv{x1 , . . . , xs } ⊂ G(xi )
i=1
for each finite subset {x1 , . . . , xs } ⊂ X.

We now give some examples of KKM-maps.
(i) Variational problems. Let C be a convex subset of E and φ : C → R be a convex
functional2 combination ∑ λi xi in C; if φ : C → R is convex, then the sets
{ y ∈ C | φ (y) < λ } and { y ∈ C | φ (y) ≤ λ } are convex for each λ ∈ R. For
each x ∈ C, let
G(x) = { y ∈ C | φ (y) ≤ φ (x) }.
We show that G : C → 2C is a KKM-map. For a contradiction let y0 = ∑ λi xi

be convex combination in C such that y0 ∈ / ni=1 G(xi ). Then φ (xi ) < φ (y0 ) for
i = 1, 2, . . . , n and this means that each xi lies in { x | φ (x) < φ (y0 ) }; since this
set is convex we have a contradiction φ (y0 ) = φ (∑ λi xi ) < φ (y0 ).
(ii) Best approximation. (a) Let E = (E, · ) be a normed linear space, C ⊂ E be
a convex set and f : C → E be a map. For each x ∈ C let
G(x) = { y ∈ C | fy − y ≤ fy − x }.
1 2E stands for the set of subsets of E and conv(A) for the convex hull of A ⊂ E.
2 Recall that a real-valued functional φ on C is convex if φ (∑ λi xi ) for any convex

36 5 KKM-Maps, Andrzej Granas
We show that G : C → 2C is a KKM-map. Indeed, let y0 = ∑ λi xi be a convex

combination in C. If y0 ∈ / ni=1 G(xi ), we would have fy0 − y0 > fy0 − xi
for each i = 1, 2, . . . , n, i.e., that xi lies in the open ball { x ∈ E | fy0 − x <
fy0 − y0 }. Since this ball is convex it contains y0 ∈ conv{x1 , . . . , xn } and we
have a contradiction: fy0 − y0 < fy0 − y0 .
(b) Let E be a vector space, C ⊂ E be convex, p be a seminorm on E and let
f : C → E be any map. For each x ∈ C let
G(x) = { y ∈ C | p(fy − y) ≤ p(fy − x) }.
The same argument as in the previous example shows that G : C → 2C is a

KKM-map.
(iii) Variational inequalities. Let E = (H, ( , )) be a pre-Hilbert space, C a convex
subset of H and f : C → H any map. For each x ∈ C, let
G(x) = { y ∈ C | (fy, y − x) ≤ 0 }.
We show that G : C → 2E is a KKM-map. Indeed, let y0 ∈ conv{x1 , . . . , xn }. If

y0 ∈/ ni=1 G(xi ), we would have (fy0 , y0 − xi ) > 0 for each i = 1, 2, . . . , n, i.e.,
that each xi lies in the set { x ∈ E | (fy0 , y0 ) > (fy0 , x) }. Since this set is convex
it also contains y0 = ∑ λi xi and we have a contradiction: (fy0 , y0 ) < (fy0 , y0 ).
The principle of KKM-maps

The following fundamental result represents a version of the well-known Knaster-
Kuratowski-Mazurkiewicz theorem [19], which was used in their simple proof of
Brouwer’s fixed point theorem:
Theorem 5.1. Let E be a vector space, X an arbitrary subset of E, and G : X → 2E

a KKM-map such that each G(x) is finitely closed3 . Then the family { G(x) | x ∈ X }
of sets has the finite intersection property.
Proof. We argue by contradiction, so assume that n1 G(xi ) = 0. / Working in the

finite-dimensional subspace L spanned by {x1 , . . . , xn }, let d be the Euclidean metric
in L and C = conv{x1 , . . . , xn } ⊂ L. Note that because each L ∩ G(xi ) is closed in
L, and since n1 L ∩ G(xi ) = 0/ by hypothesis, the function φ : C → R given by
x → ∑n1 d(x, L ∩ G(xi )) does not vanish. We now define a continuous map f : C → C
by setting
1 n
f (x) = ∑ d(x, L ∩ G(xi )) · xi .
φ (x) i=1
3 A subset A ⊂ E is finitely closed if its intersection with each finite dimensional linear subspace
L ⊂ E is closed in the Euclidean topology of L.

5.1 Definition and examples 37
By Brouwer’s fixed point theorem, f would have a fixed point x0 ∈ C. Letting

I = { i | d(x0 , L ∩ G(xi )) = 0 }, the fixed point x0 cannot belong to { G(xi ) | i ∈ I };
however,
x0 = f (x0 ) ∈ conv{ xi | i ∈ I } ⊂ { G(xi ) | i ∈ I }
and, with this contradiction, the proof is complete.

As an immediate corollary we obtain:
Theorem 5.2. (Ky Fan [7]). Let E be a topological vector space4 , X ⊂ E an

arbitrary subset, and G : X → 2E a KKM-map. If all the sets G(x) are closed in
E, and if one is compact, then { G(x) | x ∈ X } = 0.
/
We now observe that the conclusion G(x) = 0/ can be reached in another way
which avoids placing any compactness restriction on the sets G(x); it involves using
an auxiliary family of sets and a suitable topology on E:
Theorem 5.3. Let E be a vector space, X an arbitrary subset of E, and G : X → 2E

a KKM-map. Assume there is a set-valued map Γ : X → 2E such that G(x) ⊂ Γ (x)
for each x ∈ X, and for which x∈X G(x) = 0/ =⇒ x∈X Γ (x) = 0./
Because of (1.1) the proof is obvious.
Simple Applications
We give now some simple applications of KKM-maps.
Theorem 5.4. (Mazur-Schauder [22]) Let E be a reflexive Banach space and C a

closed convex set in E. Let φ be a lower-semicontinuous5 convex and coercive (i.e.,
|φ (x)| → ∞ as x → ∞) functional on C. If φ is bounded from below, then at some
x0 ∈ C the functional φ attains its minimum.
Proof. Let d = inf{ φ (x) | x ∈ C }; because φ is coercive, we can find a number

ρ > 0 such that K = B(0, ρ ) ∩ C = 0/ and φ (x) > d + 1 for all x ∈ C\K. It is enough
now to show that there is a point x0 ∈ K such that φ (x0 ) ≤ φ (x) for all x ∈ K. For
each x ∈ K, let G(x) = { y ∈ K | φ (y) ≤ φ (x) }; since d = inf φ (x), the theorem will
be proved by showing G(x) = 0. / Since G : K → 2E is a KKM-map (cf. example
(i)), the conclusion is obtained by observing that in the weak topology of E each
G(x) (being closed and convex) is compact.
4 All topological spaces are assumed to be Hausdorff.

5A map f : X → R on a topological space is lower semicontinuous (l.s.c.) if { x ∈ X | f (x) > λ }
is open for each λ ∈ R; it is upper semicontinuous (u.s.c.) if { x ∈ X | f (x) < λ } is open for each
λ ∈ R.
The next result generalizes one of the forms of the Schauder fixed point theorem;
it follows that the principle of KKM-maps is in fact equivalent to the Brouwer fixed
point theorem.
Theorem 5.5. (Ky Fan [10]) Let C be a compact convex set in a normed space E
and let f : C → E be continuous. Assume further that, for each x ∈ C with x = f (x)
the line segment [x, f (x)] contains at least two points of C. Then f has a fixed point.
Proof. Define G : C → 2E by
G(x) = { y ∈ C | y − f (y) ≤ x − f (y) }.
We know (cf. example (ii)) that G is a KKM-map. Because f is continuous, the sets
G(x) are closed in C, and therefore compact. Consequently, we find a point y0 such
that y0 ∈ x∈C G(x) and hence y0 − f (y0 ) ≤ x − f (y0 ) for all x ∈ C. We show
that y0 is a fixed point: the segment [y0 , f (y0 )] must contain a point of C other than
y0 , say x = ty0 + (1 − t)f (y0 ) for some 0 ≤ t < 1; then y0 − f (y0 )leqty0 − f (y0 )
and, since t < 1, we must have y0 − f (y0 ) = 0.
Theorem of Tychonoff and two of its generalizations
Theorem 5.6. (Tychonoff [31]) Let C be a compact convex set in a locally convex
topological space E. Then every continuous f : C → C has a fixed point.
Proof. Let {pi }i∈I be the family of all continuous seminorms in E. For each
i ∈ I set
Ai = { y ∈ C | pi (y − f (y)) = 0 }.
A point y0 ∈ C is a fixed point for f if and only if y0 ∈ i∈I Ai . By compactness of C

we need show only that j∈J Aj = 0/ for each finite subset J ⊂ I. Define G : C → 2E by
G(x) = { y ∈ C | ∑ pj (y − f (y)) ≤ ∑ pj (x − f (y)) }.

j∈J j∈J
It is easy to verify that G is a KKM-map; consequently, there is a point y0 ∈ C

such that
∑ pj (y0 − f (y0 )) ≤ ∑ pj (x − f (y0 ))

j∈J j∈J
for all x ∈ C. This clearly implies that pj (y0 − f (y0 )) = 0 for j ∈ J and thus
y0 ∈ j∈J Aj .
The Tychonoff fixed point theorem (1.6) is a special case of the following result
of Ky Fan [10] which extends Theorem (1.5) to locally convex spaces:
5.2 Ky Fan fixed point theorem and the minimax inequality 39
Theorem 5.7. Let C be a compact convex set in a locally convex topological vector
space E and let f : C → E be continuous. Assume that, for each x ∈ C with x = f (x),
the line segment [x, f (x)] contains at least two points of C. Then f has a fixed point.
Proof. Assume f (x) = x for all x ∈ C. Then for some continuous seminorm p we
would have infy∈C p[f (y) − y] > 0. Define G : C → 2C by G(x) = { y ∈ C | p(fy − y) ≤
p(fy − x) }. Since G is a compact valued KKM-map (cf. example (ii)b), we get a
point y0 ∈ C such that
0 < p(fy0 − y0 ) ≤ p(fy0 − x) for all x ∈ C.
Now the same simple argument as in (1.5) gives a contradiction p(fy0 − y0 ) <
p(fy0 − y0 ). The proof is completed.
As an immediate application of (1.7) we derive a fixed point theorem for inward
and outward maps in the sense of B. Halpern. Let C be a convex subset of a vector
space E; for each x ∈ C, let
IC (x) = { y ∈ C | there exists y0 ∈ C and λ > 0 such that y = x + λ (y0 − x) }
and
OC (x) = { y ∈ C | there exists y0 ∈ C and λ > 0 such that y = x − λ (y0 − x) }.
A map f : C → E is said to be inward (resp. outward) if f (x) ∈ IC (x) (resp. f (x) ∈

OC (x)) for each x ∈ C.
Theorem 5.8. Let C be a convex compact subset of a locally convex topological
vector space E. Then every continuous inward (resp. every continuous outward)
map f : C → E has a fixed point.
Proof. The case of an inward map follows directly from (1.7); if f is outward, then
g : C → E given by x → 2x − f (x) is inward with the same set of fixed points as f
and the conclusion follows.
5.2 Ky Fan fixed point theorem and the minimax inequality
The following result is an important application of the KKM-map principle:

Theorem 5.9. (Ky Fan [10]) Let C be a nonempty compact convex set in a linear
topological space E and let A : C → 2C be a set-valued map such that
(i) A−1 y6 is open for each y ∈ C;
(ii) Ax is convex nonempty for each x ∈ C. Then there is a w ∈ C such that w ∈ Aw.
6 If A : X → 2Y , then A−1 : Y → 2X is defined by the condition x ∈ A−1 ⇐⇒ y ∈ Ax.

Proof. Define G : C → 2C by y → C\A−1 y; each G(y) is a nonempty set closed in

C, therefore compact. We observe that C = { A−1 y | y ∈ C }:
Given any x0 ∈ C choose a y0 in the nonempty set Ax0 ; then x0 ∈ A−1 y0 . Thus
{ G(y) | y ∈ C } = 0/ and G cannot be a KKM-map. Therefore for some convex
/ si=1 G(yi ) and hence w ∈ C\ si=1 G(yi ) = si=1 A−1 yi .
combination w = ∑si=1 λi yi ∈
−1
Thus w ∈ A y for each i = 1, 2, . . . , s and therefore yi ∈ Aw for each i = 1, 2, . . . , s.
Since Aw is convex we get w = ∑ λi yi ∈ Aw and the proof is completed.
Using Theorem (2.1) we shall now derive two other generalizations of the
Tychonoff Theorem. We first establish the following.
Theorem 5.10. Let C be a nonempty compact convex subset of a linear topological
space E, V an open convex nbd of 0 and let f : C → E be a continuous map such
that f (C) ⊂ C + V. Then there is an x0 ∈ C satisfying f (x0 ) ∈ x0 + V.
Proof. Define A : C → 2C by
Ax = { y ∈ C | fx − y ∈ V }.
Each Ax is convex and each A−1 y is open. Supposing Ax0 = 0/ for some x0 , we get
f (x0 ) ∈
/ C + V contrary to f (C) ⊂ C + V. Thus by the Ky Fan fixed point theorem
we get x0 ∈ Ax0 for some x0 ∈ C, i.e., f (x0 ) ∈ x0 + V and the proof is complete.
Theorem 5.11. (Schauder-Tychonoff) Let C be a convex subset of a locally convex
linear topological space E and let f : C → C be a continuous compact map (i.e.,
f (C) is relatively compact in C). Then f has a fixed point.
Proof. Let V be a convex symmetric nbd of 0. Because E is locally convex it is
enough to show that f has a V-fixed point, i.e., a point x0 such that f (x0 ) ∈ x0 +V. Let
¯ and let K = conv{x1 , . . . , xk }.
{xi +V}ki=1 be a finite covering of the compact set f (C)
Since f (K) ⊂ K + V there is by Lemma (2.2) a point x0 ∈ K ∩ C such that f (x0 ) ∈
x0 + V and the proof is completed.
Theorem 5.12. (Ky Fan-Iokhvidov) Let C and K be two convex compact subsets of
a locally convex space E and let f : C → E be a continuous map such that f (C) ⊂
C + K. Then there is a point x0 ∈ C such that f (x0 ) ∈ x0 + K.
Minimax inequality
The following result due to Ky Fan [11] represents an analytic formulation of the
Ky Fan fixed point theorem and at the same time generalizes the Mazur-Schauder
theorem (2.4):
Theorem 5.13. (Minimax inequality) Let C be a compact convex set in a topologi-

cal vector space. Let f : C × C → R be a real-valued function such that:
(i) y → f (x, y) is l.s.c. on C for each x ∈ C;
5.3 KKM-maps and variational inequalities 41
(ii) x → f (x, y) is quasi- concave7 on C for each y ∈ C. Then miny∈C supx∈C f (x, y) ≤
supx∈C f (x, x).
Proof. Note that y → sup f (x, y) is l.s.c. and hence its minimum miny∈C supx∈C f (x, y)
on the compact set C exists. Let μ = supx∈C f (x, x); clearly, we may assume that
μ < ∞. Define G : C → 2E by
G(x) = { y ∈ C | f (x, y) ≤ μ }.
As in Example (i), it can be easily verified that G is a KKM-map; furthermore,

each G(x) is compact because y → f (x, y) is l.s.c. By Theorem (1.2) we infer that
/ and hence there is a y0 ∈ C such that f (x, y0 ) ≤ μ for all x ∈ C; this
x∈C G(x) = 0
clearly implies the assertion of the theorem and thus the proof is completed.
Among numerous applications of the Ky Fan minimax inequality we mention the
following fundamental existence theorem in potential theory:
Theorem 5.14. Let X be a compact space and G : X × X → R+ a continuous

function such that G(x, x) > 0 for all x ∈ X. Then there exists a positive Radon
measure μ on X such that
G(x, y)d μ (y) ≥ 1
for all x ∈ X, and
G(x, y)d μ (y) = 1
for x in the support of μ .
For a proof, we refer to Ky Fan [11].
5.3 KKM-maps and variational inequalities
KKM-maps can be used to get some of the basic facts in the theory of variational
inequalities.
Let (H, ( , )) be a Hilbert space and C be any subset of H. We recall that a map
f : C → H is called monotone8 on C if (f (x) − f (y), x − y) ≥ 0 for all x, y ∈ C. We say
that f : C → H is hemi-continuous if f |L∩C is continuous for each one-dimensional
flat L ⊂ H.
7 Recall that a real-valued function φ defined on a convex set C is quasi-concave if the set { x ∈ C |
f (x) > r } is convex for each r ∈ R; φ is quasiconvex if −φ is quasi-concave.
φ : H → R is Gateaux differentiable, then φ : H → H is monotone if and only if φ is convex;
8 If
thus, the notion of a monotone operator arises naturally in the classical context of the calculus of
variations.
Theorem 5.15. (Hartman-Stampacchia [15]) Let H be a Hilbert space, C a closed

bounded convex subset of H, and f : C → H monotone and hemi-continuous. Then
there exists a y0 ∈ C such that (f (y0 ), y0 − x) ≤ 0 for all x ∈ C.
Proof. For each x ∈ C, let
G(x) = { y ∈ C | (f (y), y − x) ≤ 0 };
the theorem will be proved by showing { G(x) | x ∈ C } = 0.

/
We know (cf. example (iii)) that G : C → 2H is a KKM-map. Consider now the
map Γ : C → 2H given by
Γ (x) = { y ∈ C | (f (x), y − x) ≤ 0 };
we show that Γ satisfies the requirements of (2.4):

(i) G(x) ⊂ Γ (x) for each x ∈ C. For, let y ∈ G(x), so that 0 ≥ (f (y), y − x). By
monotonicity of f : C → H we have (f (y) − f (x), y − x) ≥ 0 so 0 ≥ (f (x), y − x)
and y ∈ Γ (x).
(ii) Because of (i), it is enough to show { Γ (x) | x ∈ C } ⊂ { G(x) | x ∈ C }.
Assume y0 ∈ Γ (x). Choose any x ∈ C and let zt = tx + (1 − t)y0 = y0 − t(y0 −
x); because C is convex, we have zt ∈ C for each 0 ≤ t ≤ 1. Since y0 ∈ Γ (zt )
for each t ∈ [0, 1], we find that (f (zt ), y0 − zt ) ≤ 0 for all t ∈ [0, 1]. This says that
t(f (zt ), y0 − x) ≤ 0 for all t ∈ [0, 1] and, in particular, that (f (zt ), y0 − x) ≤ 0 for
0 < t ≤ 1. Now let t → 0; the continuity of f on the ray joining y0 and x gives
f (zt ) → f (y0 ) and therefore that (f (y0 ), y0 − x) ≤ 0. Thus, y0 ∈ G(x) for each
x ∈ C and Γ (x) = G(x).
(iii) We now equip H with the weak topology. Then each Γ (x), being the
intersection of the closed half-space { y ∈ H | (f (x), y) ≤ (f (x), x) } with C,
is closed convex and bounded and therefore weakly compact.
Thus, all the requirements in (1.4) are satisfied; therefore, { G(x) | x ∈ C } = 0/
and, as we have observed, the proof is complete.
Corollary 5.1. (Browder-Goedhe-Kirk) Let C be a closed bounded convex subset
of H and F : C → C a nonexpansive map, i.e., Fx − Fy ≤ x − y for all x, y ∈ C.
Then F has a fixed point.
Proof. Putting f (x) = x − F(x) for x ∈ C, we verify by simple calculation that
f : C → H is a continuous monotone map; so by theorem (3.1) there is a y0 ∈ C such
that (y0 − Fy0 , y0 ) = (fy0 , y0 − x) ≤ 0 for all x ∈ C. By taking in the above inequality
x = F(y0 ) we get y0 = Fy0 , and the proof is complete.
Corollary 5.2. (Nikodym [26]) Let C ⊂ H be a closed bounded convex set. Then
for each x0 ∈ H there is a unique y0 ∈ C with x0 − y0 = inf{ x0 − x | x ∈ C }.
5.4 KKM-maps and the theory of games 43
Proof. Uniqueness being evident, let f : C → H be given by y → y − x0 ; clearly,

f is continuous and monotone. By (3.1) there is y0 ∈ C with (y0 − x0 , y0 − x) ≤ 0
for all x ∈ C; this being equivalent to x0 − y0 = infC x0 − x, the assertion of the
theorem follows.
5.4 KKM-maps and the theory of games
The notion of a KKM-map can be used to establish general geometric results which
have many applications in the theory of games.
The Coincidence Theorem and the Minimax Principle
Theorem 5.16. (Ky Fan) Let X ⊂ E and Y ⊂ F be nonempty compact convex sets
in the linear topological spaces E and F. Let A, B : X → 2Y be two set-valued maps
such that
(i) Ax is open and Bx is a nonempty convex set for each x ∈ X;
(ii) B−1 y is open and A−1 y is a nonempty convex set for each y ∈ Y. Then there is
an x0 ∈ X such that Ax0 ∩ Bx0 = 0.
/
Proof. Let Z = X × Y and define G : X × Y → 2E×F by (x, y) → Z − (B−1 y × Ax);
each G(x, y) is a nonempty set closed in X × Y, therefore compact. As in the proof of
Theorem (2.1) one verifies easily that G cannot be a KKM-map. Therefore there are
elements z1 , . . . , zn in Z such that conv(z1 , . . . , zn ) is not contained in n1 G(zi ), so that
some convex combination w = ∑n1 λi zi ∈ / n1 G(zi ). Because Z is convex, the point w
belongs to Z, so w ∈ Z − 1 G(zi ) = n1 B−1 (yi ) × Axi . Writing w = (∑ λi xi , ∑ λi yi )
n
we have ∑n1 λi xi ∈ B−1 (yi ) for each i = 1, . . . , n and ∑n1 λi yi ∈ Axi for each i = 1, . . . , n.
The first inclusion shows each yi ∈ B(∑n1 λi xi ) and therefore that ∑ λi yi ∈ B(∑ λi xi ).
The second inclusion shows each xi ∈ A−1 (∑ λi yi ), therefore ∑ λi xi ∈ A−1 (∑ λi yi ),
and consequently ∑ λi yi ∈ A(∑ λi xi ). Thus, A(∑ λi xi ) ∩ B(∑ λi xi ) = 0, / and the proof
is complete.
We give an immediate application to game theory by establishing a general
version of the von Newmann minimax principle due to M. Sion [29].
Theorem 5.17. (Minimax principle) Let X and Y be two nonempty compact convex
sets in the linear topological spaces E and F. Let f : X × Y → R satisfy
(i) y → f (x, y) is lsc and quasi-convex for each fixed x ∈ X;
(ii) x → f (x, y) is usc and quasi-concave for each fixed y ∈ Y. Then maxx miny
f (x, y) = miny maxx f (x, y).
Proof. Because of upper semicontinuity, maxx f (x, y) exists for each y and is a
lower semicontinuous function of y, so miny maxx f (x, y) exists; similarly,
maxx miny f (x, y) exists. Since f (x, y) ≤ maxx f (x, y) we have miny f (x, y) ≤
miny maxx f (x, y); therefore,
max min f (x, y) ≤ min max f (x, y).

x y y x
We shall show that inequality cannot hold. Assume it did; then there would
be some r with maxx miny f (x, y) < r < miny maxx f (x, y). Define A, B : X → 2Y
by Ax = { y | f (x, y) > r } and Bx = { y | f (x, y) < r }. We verify that these set-
valued maps satisfy the conditions of the coincidence theorem: Each Ax is open
by lower semicontinuity of y → f (x, y), each Bx is convex by the quasi-convexity
of y → f (x, y), and is nonempty because maxx miny f (x, y) < r. Since A−1 y = { x |
f (x, y) > r } and B−1 y = { x | f (x, y) < r }, we find in the same way that each A−1 y
is nonempty and convex and each B−1 y is open. Then by the coincidence theorem,
there would be some (x0 , y0 ) with y0 ∈ A(x0 ) ∩ B(x0 ), which gives the contradiction
r < f (x0 , y0 ) < r. Thus, the inequality cannot hold, and the proof is complete.
The Intersection Theorem and the Nash Equilibrium Theorem
Given a cartesian product X = ∏ni=1 Xi of topological spaces, let X j = ∏i=j Xi and

let pi : X → Xi , pi : X → X i denote their projections; write pi (c) = xi and pi (x) = xi .
Given x, y ∈ X we let
(yi , xi ) = (x1 , . . . , xi−1 , yi , xi+1 , . . . , xn ).
The following geometrical result of Ky Fan [8] generalizes the coincidence

theorem:
Theorem 5.18. Let X1 , X2 , . . . , Xn be nonempty compact convex sets in linear
topological spaces and let A1 , A2 , . . . , An be n subsets of X such that
(i) for each x ∈ X and each i = 1, 2, . . . , n,
Ai (x) = { y ∈ X | (yi , xi ) ∈ Ai }
is convex and nonempty;

(ii) for each y ∈ X and each i = 1, 2, . . . , n,
Ai (y) = { x ∈ X | (yi , xi ) ∈ Ai }
is open. Then n
i=1 Ai = 0.
/
Proof. As in (4.1) define G : X → 2X by y → X\ ni=1 Ai (y); one verifies that G is not
a KKM-map and if a convex combination w = ∑ λi xi ∈ / G(xi ), then w ∈ ni=1 Ai .
5.5 Bibliographical and historical comments 45
As an immediate corollary:
Theorem 5.19. (Nash equilibrium theorem [24])

Let X1 , X2 , . . . , Xn be nonempty compact convex sets each in a topological vector
space. Let f1 , f2 , . . . , fn be n real-valued continuous functions defined on X = ∏ni=1 Xi
such that for each y ∈ X and each i = 1, 2, . . . , n the function xi → fi (xi , yi ) is quasi-
concave on Xi . Then there is a point y0 ∈ X such that fi (y0 ) = maxxi fi (xi yi0 ).
Proof. We briefly indicate the proof. Given ε > 0, define for each i = 1, 2, . . . , n,
Aεi = { y ∈ X | fi (y) > max fi (xi , yi ) − ε }.

xi ∈Xi
ε
i=1 Ai = 0.
n
One verifies easily that the conditions of (4.3) are satisfied and hence /
Then by a compactness argument one gets a point y0 ∈ X such that y0 ∈ n ε
i=1 Ai for
each ε > 0, and this point y0 satisfies the assertion of the theorem.
The coincidence theorem of Ky Fan also has applications in areas other than
the theory of games. Among such applications we will mention the following result
which extends the Tychonoff fixed point theorem to an important class of nonlocally
convex spaces:
Theorem 5.20. (Ky Fan [8]) Let E be a linear topological space with sufficiently
many continuous linear functionals9 , and let C be a convex and compact subset of E.
Then every continuous map f : C → C has a fixed point.
5.5 Bibliographical and historical comments
1. In the special case when X is the set of vertices of a simplex in Rn , Theorem (1.1)
was discovered by Knaster-Kuratowski-Mazurkiewicz [19]; their method of
proof was based on Sperner’s Lemma. The abbreviation KKM stands for
Knaster-Kuratowski-Mazurkiewicz. The principle of KKM-maps (Theorem
(1.1)), established in a somewhat different form by Ky Fan [7], represents an
infinite-dimensional analog of the Knaster-Kuratowski-Mazurkiewicz theorem;
its formulation and the proof are taken from Dugandji-Granas [5]. Ky Fan
demonstrated the importance of the principle of KKM-maps by giving numerous
applications to various fields.
Theorem (1.4) of Mazur-Schauder [22] (and an earlier Theorem (3.3) of
Nikodym [26]) initiated the abstract approach to problems in calculus of
variations. Mazur and Schauder gave applications of Theorem (1.4) to a number
of concrete problems in calculus of variations; these results, however, were never
published (cf. Scottish Book Problem 105).
9 For example, the Hardy spaces H p (0 < p < 1) have this property but are nonlocally convex.
Theorem (1.6) of Tychonoff [31] gives a positive answer to the second part of
Problem 54. The proof of (1.6) given here is due to Ky Fan [7]. Theorems (1.7)
and (1.8) were established by Ky Fan [10] and Browder [3] respectively.
2. Theorem (2.1) was established (in a different form) by Ky Fan [7]; the formula-
tion of (2.1) given here is due to Browder [3], who obtained it from the Brouwer
theorem.
Theorem (2.3), due (in a slightly less general form) to Hukuhara [16] (cf. also
an earlier result by Mazur [21]), gives a positive answer to the third part of
Problem 54; the proof of (2.3) given here is due to Lassonde [20]. Theorem (2.4),
established by Ky Fan, generalizes an earlier result of Iokhvidov [17].
The fact that the minimax inequality of Ky Fan is equivalent to Theorem (2.1)
is proved in Ky Fan [11]. For other applications of the Ky Fan fixed point
theorem (2.1) (or of the minimax inequality) the reader is referred to Ky Fan
[11], Lassonde [20], and Browder [4].
3. Variational inequalities (the systematic study of which began around 1965)
have in recent years assumed increasing importance in many applied prob-
lems (cf. the survey by Stampacchia [30] for an introductory account and
further references). The proof of Theorem (3.1) is from Dugundji-Granas [5].
The same type of proof works for semimonotone operators in the sense of
F. Browder (cf. Lassonde [20]). For more general results, see also Brezis-
Nirenberg-Stampacchia 479 [2], Lassonde [20], and Mosco [23].
4. The Coincidence Theorem (4.1) is a special case of the intersection theorem (4.3)
proved by Ky Fan in [9]. Theorem (4.2), established by Sion [29] evolved from
several earlier results; in the special case when X ⊂ Rn , Y ⊂ Rk are simplexes and
f is bilinear, Theorem (4.2) was discovered by von Neumann [25], who deduced
it from the Brouwer theorem. The direct proof of (4.2) is a modification of an
earlier proof by Ky Fan [8] and is taken from Dugundji-Granas [6]. The proof of
the Nash equilibrium theorem is due to Ky Fan [9]. For more general results and
further references, see Browder [3], Lassonde [20], and Ky Fan [12].
In connection with Theorem (4.5) we remark that the first part of Problem 54
remains unanswered; it is not known whether a compact convex subset of an
F-space has the fixed point property. Theorem (4.5) represents the best-known
partial answer to this question. For other fixed point results in nonlocally convex
spaces, see Klee [18], Granas [14], and also Riedrich [27], Granas [13], where
further references will be found.
Problem 54 was an inspiration for numerous later investigations both in fixed
point theory and in nonlinear functional analysis. The literature is too extensive
to be summarized here and we refer to Dugundji-Granas [6] and Granas [13] for
bibliographies on the topics in fixed point theory related to this problem.
References 47
References
1. S. Banach, Die Theorie der Operationen und ihre Bedeutung für die Analysis, Proc. Int.
Congress in Oslo (1936), 261–268.
2. H. Brezis, L. Nirenberg, G. Stampacchia, A remark on Ky Fan’s minimax principle, Boll.
Unione Mat. Ital. (4) 6 (1972), 293–300.
3. F.E. Browder, The fixed point theory of multi-valued mappings in topological vector spaces,
Math. Ann. 177 (1968), 283–301.
4. F.E. Browder, On a sharpened form of the Schauder fixed-point theorem, Proc. Natl. Acad. Sci.
USA, 74 (1977) 4749–4751.
5. J. Dugundji, A. Granas, KKM-maps and variational inequalities, Ann. Scu. Norm. di Pisa, V
(1978), 69–682.
6. J. Dugundji, A. Granas, Fixed point theory, Vol. I, Monografie Matematyczne 61, Warszawa,
1982. Springer Monographs in Mathematics, Springer-Verlag, New York, 1982.
7. K. Fan, A generalization of Tychonoff’s fixed point theorem, Math. Ann. 142 (1961), 305–310.
8. K. Fan, Sur un theoreme minimax, C.R. Acad. Sci. Paris, Groupe 1, 259 (1962), 3925–3928.
9. K. Fan, Applications of a theorem concerning sets with convex sections, Math. Ann. 163
(1966), 189–203.
10. K. Fan, Extensions of two fixed point theorems of F.E. Browder, Math. Z. 112 (1969), 234–240.
11. K. Fan, A minimax inequality and applications, in O. Shisha, ed., Inequalities III, Academic
Press, New York and London, 1972, 103–113.
12. K. Fan, Fixed point and related theorems for non-compact convex sets. Game theory and related
topics (Proc. Sem, Bonn and Hagen, 1978), pp. 151–156, North Holland, New York, 1979.
13. A. Granas, Points fixes pour les applications compactes: Espaces de Lefschetz et la théorie
de l’indice. Cours professe pendant la session du Seminaire de Mathematiques Superieures a
l’Université de Montreal (1973), Presses de l’Université de Montreal, 1980.
14. A. Granas, Sur la methode de continuité de Poincare, C.R. Acad. Sci. Paris 200 (1976),
983–985.
15. P. Hartman, G. Stampacchia, On some non-linear elliptic differential equations, Acta Math.
115 (1966), 271–310.
16. M. Hukuhara, Sur l’existence des points invariants d’une transformation dans l’espace
fonctionnel, Jap. J. Math. 20 (1950), 1–4.
17. I.S. Iokhvidov, On a lemma of Ky Fan generalizing the fixed-point principle of A.N. Tikhonov
(Russian), Dokl. Akad. Nauk SSSR 159 (1964), 501–504. English transl.: Soviet Math. Dokl. 5
(1964), 1523–1526.
18. V. Klee, Leray-Schauder theory without local convexity, Math. Ann. 141 (1960), 286–296.
19. B. Knaster, C. Kuratowski, S. Mazurkiewicz, Ein Beweis des Fixpunktsatzes für
n-dimensionale Simplexe, Fund. Math. 14 (1929), 132–137.
20. M. Lassonde, Multiapplications KKM en analyse non lineaire, These Ph.D. Math., Univ. de
Montreal (1978).
21. S. Mazur, Un theoreme sur les points invariants, Ann. Soc. Polon. Math. XVII (1938), 110.
22. S. Mazur, J. Schauder, Über ein Prinzip in der Variationsrechnung, Proc. Int. Congress Math.,
Oslo (1936), 65.
23. U. Mosco, Implicit variational problems and quasi variational inequalities, in Lecture Notes
in Mathematics, 543: Non-linear operators and the calculus of variations, Bruxelles (1975),
Springer-Verlag, Berlin-Heidelberg-New York (1976), 83–156.
24. J. Nash, Non-cooperative games, Ann. of Math. 54 (1951), 286–295.
25. J.V. Neumann, Ueber ein okonomisches Gleichungssystem und eine Verallgemeinerung des
Brouwerschen Fixpunktsatzes, Ergebnisse eines Mathematischen Kolloquiums 8 (1935),
73–83.
26. O. Nikodym, Sur le principe de minimum dans le probleme de Dirichlet, Ann. Soc. Polon.
Math. X (1931), 120–121.
27. T. Riedrich, Vorlesungen über nichtlineare Operator-gleichungen, Teubner-Texte, Teubner-

Verlag, Leipzig (1975).
28. J. Schauder, Einige Anwendungen der Topologie der Funktionalraume, Recueil mathematique
I (1936), 747–753.
29. M. Sion, On general minimax theorems, Pacific J. Math. 8 (1958), 171–176.
30. G. Stampacchia, Variational inequalities, Summer School on theory and applications of
monotone operators, Venice, (1968).
31. A. Tychonoff, Ein Fixpunktsatz, Math. Ann. 111 (1935), 767–776.
Front Row l. to r.: Stan Ulam, Mark Kac, Paul Erdos, A. Zygmund. Back Row l.
to r.: Dan Mauldin, Tony Martin, A. Granas, R. D. Anderson.
Part II
The Scottish Book Problems
Chapter 6
Problems with Commentary
Stefan Banach’s Last Picture, at the End of the War in Poland. Sent to Stan Ulam by
his son.
PROBLEM 1: BANACH
July 17,1935
(a) When can a metric space [possibly of type (B)] be so metrized that it will
become complete and compact, and so that all the sequences converging
originally should also converge in the new metric?
(b) Can, for example, the space c0 be so metrized?
Commentary
A space of type (B) is the terminology from Banach’s monograph, Théorie des
Opérations Linéaires, Warszawa, 1932, for a Banach space.

52 6 Problems with Commentary
There is probably no satisfactory answer to part (a). This can be seen as follows.
First, notice that if X is a metric space, then X admits a new metric under which X
is compact and such that all sequences which converge in the original metric should
converge in the new metric if and only if there is a continuous one-to-one map f of
X (with the original metric) onto a compact metric space. Next, notice that if g is
a function from [0, 1] into [0, 1] and X is the graph of g in the unit square provided
with the usual Euclidean metric, then the projection of X into the first axis is a
continuous one-to-one map of X onto [0, 1]. Since there are 2c maps of [0, 1] into
[0, 1], a majority of the spaces X so obtained are very strange.
However, there are restricted cases of this general problem which seem to be
unresolved. For example:
Let X be a complete separable metric space. Are there some simple conditions
such that there is a continuous one-to-one map of X onto a compact metric space?
For example, if X is a locally compact separable metric space, then there is such
a map. The space N N or equivalently the space of all irrational numbers has this
property [4]. Also if X = ∏∞ n=1 Xn , where each Xn has this property, then X also has
this mapping property. In particular, Rω has this property. Since it is now known that
every infinite-dimensional Banach space is homeomorphic to Rω [1, 2] the answer
to part (b) is yes. This was first proved in a different way by Klee [3].
1. R. D. Anderson and R. H. Bing, A complete elementary proof that Hilbert space is homeomor-
phic to the countable infinite product of lines, Bull. Amer. Math. Soc. 74 (1968).
2. M. I. Kadeč, On homeomorphisms of certain Banach spaces, Dokl. Akad. Nauk SSSR 92 (1955),
465–468.
3. V. Klee, On a problem of Banach, Coll. Math. 5 (1957), 280–285.
4. W. Sierpiński, Sur les images continues et biunivoques de l’ensemble de tous les nombres
irrationels, Mathematica 1 (1929), 18–21.
R. Daniel Mauldin
PROBLEM 2: BANACH, ULAM

(a) Can one define, in every compact metric space E, a measure (finitely additive)
so that Borel sets which are congruent should have equal measure?
(b) Suppose E = E1 + E2 + . . . + En , and E1 ∼ = ··· ∼
= E2 ∼ = En and {Ei } are disjoint;
1 1 1
then we write Ei = E. Can it occur that E = E, n = m, if we assume that
n n m
1
E are Borel sets and E is compact?
n
Commentary
Two Borel sets A and B are congruent means there is an isometry of A onto B.
By a result of Tarski ([3], Theorem 16.12) problems (a) and (b) are equivalent to
each other.
The answer is yes if the space E is also supposed to be countable [4].
6 Problems with Commentary 53
It is known that for every compact metric space there exists a Borel measure such
that congruent open sets have equal measures (see [1, 2]). It follows that if two Borel
sets are congruent by an isometry which extends to some open sets then those Borel
sets have equal measures.
1. J. Mycielski, Remarks on invariant measures in metric spaces, Coll. Math. 32 (1974), 105–112.
2. , A conjecture of Ulam on the invariance of measure in Hilbert’s cube, Studia
Math. 60 (1977), 1–10.
3. A. Tarski, Cardinal Algebras, Oxford University Press, 1949.
4. R. O. Davies and A. J. Ostaszewski, Denumerable compact metric spaces admit isometry-
invariant finitely additive measures, Mathematika 26 (1979), 184–186.
Jan Mycielski
Second Edition Commentary
Problem 2 remains open for uncountable compact metric spaces. This problem
has given rise to several interesting related results and questions concerning the
existence of countably additive, metrically invariant σ -finite measures on locally
compact metric spaces.
We say a metric space (X, d) is locally homogeneous if for any two points x, y ∈ X
there is some ε > 0 and an isometry φ of the open ball B(x, ε ) onto B(y, ε ) which
sends x to y. We say that a measure (finitely additive or countably additive) on the
Borel subsets of X is metrically invariant provided whenever A and B are congruent
Borel sets, then μ (A) = μ (B). We say μ is open-invariant provided μ (A) = μ (B),
whenever A and B are congruent open sets. Bandt and Babaki in [1] proved the
following.
Theorem 1. Let (X, d) be locally compact and locally homogeneous and let A be
a compact subset of X with non-empty interior. Then there is a unique metrically
invariant Borel measure μ on X with μ (A) = 1.
This class includes all locally compact groups with a left-invariant metric.
We note the following example of Bandt and Baraki demonstrating why one
cannot expect countable additivity without local homogeneity. Let X = ∪∞ n=1 Cn ,
where Cn is the n-dimensional cube [0, 1/n]n and let d be the Euclidean metric on
each Cn and let d(x, y) = max{1/n, 1/m}, for x ∈ Cn , y ∈ Cm , n = m. For each n,
Lebesgue measure λn on Cn induces an open-invariant measure λn on X and positive
linear combinations of these are open-invariant. However, there is no σ -finite
metrically invariant measure, μ . For otherwise, for some n, μ (Cn ) > 0. Then a
subcube of Cn with edge length 1/(n + 1) will have positive measure. Thus, Cn+1
would impossibly contain uncountably many disjoint isometric copies with the same
measure.
Let me mention that Bandt and Baraki also incorporated this example into
hyperspaces answering a question of McMullen:
Theorem 2. Let n > 1 and let K (Rn ) be the space of all compact convex subsets of
Rn with dH , the Hausdorff metric. Then there is no nontrivial σ -finite Borel measure
on K (Rn ) which is invariant even under all isometries from the whole space into
itself. (On the other hand, there is an open-invariant measure.)
Ulam posed the following specific problem concerning metrically invariant
measures.
Open Problem [Ulam]: Let μ be Lebesgue product measure on [0, 1]∞ . Is μ
metrically invariant with respect to metrics of the type
d(x, y) = (∑ a2n · (xn − yn )2 )1/2 , where a = (an ) ∈ 2 ?
Mycielski in his referenced 1977 paper showed that invariant open sets A and
B have the same measure under these metrics but whether arbitrary invariant Borel
sets have the same measure remains open. There has been some progress. J. Fickett
in [2] proved the following.
Theorem 3. Lebesgue measure is metrically invariant on [0, 1]∞ , the Hilbert cube,
provided the sequence an is very rapidly decreasing:
1/2n
an
lim = 0.
n an−1
1. Christoph Bandt and Gebreselassie Baraki, Metrically invariant measures on locally homoge-
neous spaces and hyperspaces. Pacific J. Math. 121 (1986), no. 1, 13-28.
2. James W. Fickett, Approximate isometries on bounded sets with an application to measure
theory. Studia Math. 72 (1982), no. 1, 37-46.
R. Daniel Mauldin

Theorem. It is proved very simply that a compact set cannot be congruent to a
proper subset of itself.
Commentary
A stronger theorem has been proved by A. Lindenbaum [1], namely a set in a

compact metric space which is both an Fσ and a Gδ cannot be congruent to a
proper subset of itself. For sets which are only Fσ or only Gδ this is false as the
set { ein : n = 1, 2, . . . } in the unit circle and its complement show.
1. A. Lindenbaum, Contributions à l’étude de l’espace métrique I, Fund. Math. 8 (1926), 209–222.
Jan Mycielski
PROBLEM 4: SCHREIER
Theorem. If {ξn } is a bounded sequence, summable by the first mean to ξ , then
almost every subsequence of it is also summable by the first mean to ξ .
Commentary
Problem 4 probably arose in the context of the “normal numbers” discoveries of

Borel and others, dealing with sequences {tn } of 0’s and 1’s; if t = ∑∞ −n
1 tn 2 , then
Lebesgue measure on [0, 1] enables one to speak of a random sequence {tn }, and to
prove theorems about the density of 1’s in such a sequence. [Rend. Circ. Matem.
Palermo 27 (1909), 247–271] Input also came from the foundations of probability
theory; if the probability of 1 is p, and {tn } is the sequence of observations, and this
is (C, 1) summable to p, then one expects that a random choice of a subsequence
from {tn } would have the same property. The result cited to J. Schreier in Problem 4
is a generalization of this: if {xn } is bounded and (C, 1) summable to L, then so
is almost every subsequence. (See Birnbaum and Schreier, Studia Math. 4 (1933),
85–89; Birnbaum & Zuckerman, Amer. J. Math. 62 (1940), 787–791.)
In 1943, independently motivated, Buck and Pollard proved the following
assertions: (1) if {xn } is divergent, so are almost all its subsequences; (2) if every
subsequence of {xn } is (C, 1) summable, {xn } is convergent; (3) if {xn } is not (C, 1)
summable, neither are almost all its subsequences; (4) there is a sequence {xn }
that is (C, 1) summable but such that almost all its subsequences fail to be (C, 1)
summable; (5) if {xn } is summable to L and ∑∞ 1 (xn /n) < ∞, then so are almost all
2
its subsequences. (See Bull. Amer. Math. Soc. 49 (1943), 924–931.) “Almost all”
was interpreted in terms of the standard mapping between selection sequences of
0’s and 1’s, and [0, 1]. Of these five assertions, (2) was a special case of an earlier
result of Buck [MR 5, 117] showing that a sequence {xn } must be convergent if
every subsequence of it is summable by any fixed regular matrix method. (See also
Agnew [MR6, 46] and Buck [MR 18, 478].) In subsequent papers, assertion (1)
was extended to generalized sequences and cluster sets by Buck [MR 5, 235] and
Day [MR 5, 236] and a best possible strengthening of assertion (5) was obtained by
Tsuchikura [MR 12, 820] who proved that the condition ∑∞ 1 (xn /n) < ∞ could be
2
∞ 2
replaced by ∑1 xR = o(n / log log n). Techniques used involved properties of the
2
Rademacher functions Rn (t), and the strong law of large numbers.

R.C. Buck
Madison, Wisconsin
February, 1979
PROBLEM 5: MAZUR
Definition. A sequence {ξn } is asymptotically convergent to ξ if there exists a
subsequence of density 1 convergent to ξ .
Theorem. (Mazur) In the domain of all sequences this notion is not equivalent to
any Toeplitz method.
How is it in the domain of bounded sequences?

Addendum. We have the following theorems:

(1) If a method (aik ) sums all the asymptotically convergent sequences, the aik = 0
for k > k0 , i = 1, 2, . . . and there exist finite lim aik for k = 1, . . . , k0 , such that
the method sums all the sequences.
(2) If a method (aik ) sums all the convergent sequences and every bounded
sequence summable by the sequence is asymptotically convergent, then there
exists a sequence of increasing integers {kn } with density 1, such that for
every bounded sequence {ξn } summable by this method, the sequence {ξkn }
is convergent.
From (1) it follows that there does not exist a permanent method summing all
the asymptotically convergent sequences; from (2) it follows that a permanent
method summing all bounded asymptotically convergent sequences must also
sum other bounded sequences.
Mazur
July 22, 1935
Commentary
In 1980, matrix summability has lost the interest it had in the ’30s. However, this
problem, which remains unsolved, dealt with a concept that perhaps ought to receive
more attention. A sequence (of numbers, functions, operators, etc.) may diverge and
yet have a subsequence of density d > 0 that converges; if d = 1, Mazur called the
sequence “asymptotically convergent.” Samples: (a) A measurable transformation
T is mixing if lim m(T −n A ∩ B) = m(A)m(B) for each pair of measurable sets A and
B; T is called weakly mixing if the sequence is asymptotically convergent [Halmos,
Lectures on Ergodic Theory, Chelsea Publ. 1956]. (b) Let f (x) be an entire function
of exponential type c < π with
∞ log |f (x)f (−x)|
dx < ∞.
1 x2
Then the sequence sn = n−1 log |f (n)| has a subsequence of positive density that
converges to zero [Levinson, Gap and Density Theorems, Amer. Math. Soc. Colloq.
1940].
It has been noted that asymptotic convergence is closely related to Cesàro
summability. Thus, a bounded sequence {xn } that is (C, 1) summable to one of
its outer limit points must be asymptotically convergent to it; this is also related
to what is called strong (C, 1) summability. There are enough results to suggest
that a more structured theory lies in the background. A key result is the additivity
theorem for generalized asymptotic density, regarded as a finitely additive measure
(See [1, 2, 4, 5]).
1. R. C. Buck, The measure theoretic approach to density, Am. J. Math. 68 (1946), 560–580.
2. , Generalized asymptotic density, Am. J. Math. 75 (1953), 335–346.
3. , Some Remarks on Tauberian Conditions, Quart. J. Math. Oxford Ser. (2) 6
(1955), 128–131.
4. , Convergence theorems for finitely additive integrals, J. Indiana Math. Soc.
(N.S.) 23 (1959), 1–9.
5. A. R. Freedman, On the additivity theorem for n-dimensional asymptotic density, Pacific J.M̃ath.
49 (1973), 357–363.
6. C. T. Rajagopal, Some theorems on convergence in density, Publ. Math. Debrecen 5 (1957),
77–92.
7. Hans Rohrbach, Bodo Volkmann, Zur Theorie der asymptotische Dichte, J. Reine Angew. Math.
192 (1953), 102–112.
8. , Verallgemeinerte asymptotische Dichten, J. Reine Angew. Math. 194 (1955),
195–209.
R. C. Buck
PROBLEM 6: MAZUR, ORLICZ

Prize: Bottle of wine, S. Mazur
Is a matrix, finite in each row and invertible (in a one-to-one way), equivalent to
a normal matrix?
The terminology used in the formulation of Problem 6 can be found in S. Mazur’s

paper [1]. The paper gives a functional analysis approach to summability theory.
The subject was quite popular at the beginning of the 20th century and then mostly
abandoned. In particular, nowadays the meaning of “normal matrix” is very different
from that in Problem 6. Probably for this reason the problem escaped an attention
of readers of the Scottish Book. An answer to Problem 6 is in negative. To present
it we recall some definitions from [1] and introduce some new ones. So:
c is the Banach space of all convergent sequences of real numbers with the norm
||x||∞ = supn∈N |αn | for x = (αn )n∈N ∈ c.
s is the space of all sequences of real numbers equipped with the topology of
coordinate convergence. It is an F space or Fréchet space. For x = (αn )n∈N ∈ s
we write x(n) for αn .
T = (θn,m )n,m∈N is said to be a Toeplitz matrix if there are only finite nonzero entries
in each row. (Note that the present meaning of Toeplitz matrix is different).
For a Toeplitz matrix T = (θn,m )n,m∈N and x = (αn )n∈N the sequence T(x) is defined
by T(x)(n) = ∑∞ m=1 θn,m αm . This formula gives a 1 − 1 correspondence between
Toeplitz matrices and continuous linear operators in s. The matrix product of
Teoplitz matrices is well defined, it is a Toeplitz matrix, and it corresponds to the

composition of the corresponding operators in s. If a Toeplitz matrix T corresponds
to a 1 − 1 and “onto” operator in s, then there is unique inverse matrix T −1 ( it has
to be a Toeplitz matrix then).
Toeplitz matrices T, R are said to be equivalent if {x ∈ s : T(x) ∈ c} = {x ∈ s :
R(x) ∈ c}
A matrix D = (δn,m )n,m∈N is said to be normal if δn,m = 0 and δn,n = 0 for all n < m.
A normal matrix D is a Toeplitz matrix, it has an inverse matrix which is normal

and in which the entries on the diagonal are the inverses of corresponding diagonal
entries in D. Problem 6 can be reformulated as follows. Is it true that for each
invertible Toeplitz matrix T there exits a normal matrix D such that the T −1 (c) =
D−1 (c). The last equality means that the matrix TD−1 which is a continuous and
1 − 1 linear map of s onto itself transforms c onto itself. By the Closed Graph
Theorem in F spaces it follows that the matrix TD−1 corresponds to a bounded
linear invertible operator L in c.
A matrix T = (θn,m )n,m∈N is called semi-normal if there exists a sequence of

nonsingular square matrices Ak = (αn,m k )
1≤n,m≤lk such that for the sequence s0 =
0, sk = l1 + . . . + lk we have θn,m = αn−s
k
k−1 ,m−sk−1
for each sk−1 < n, m ≤ sk and
θn,m = 0 if n ≤ sk < m for some k. Then (lk ) is called a rank sequence and (Ak ) a
diagonal sequence of matrices of T.
It is clear that each semi-normal matrix is a Toeplitz matrix. An easy inspection
shows that if T, R are semi-normal matrices with the same rank sequence (lk ) and
the diagonal sequences of matrices (Ak ) of T and (Bk ) of R then the matrix TR
is semi-normal with rank sequence (lk ) and diagonal sequence of matrices (Ak Bk ).
Each semi-normal matrix T with diagonal sequence (Ak ) is invertible and the inverse
matrix T −1 is semi-normal which has the same rank sequence as T and (A−1 k ) is the
diagonal sequence of matrices of T −1 . Each normal matrix T is semi-normal. For
such matrix each sequence (lk ) of positive integers is a rank sequence of T and
the corresponding diagonal matrix sequence Ak = (αm,n k )
1≤n,m≤lk is defined as in
the definition of semi-normal matrix, i.e. αm,n = θn+sk−1 ,m+sk−1 . We will construct a
k
semi-normal matrix T which is not equivalent to a normal matrix. For each k ∈ N let
Ck be a matrix with 2k rows and k columns with entries equal to ±1 and such that
k
no two rows are identical. Then the columns of Ck are orthogonal vectors in R2 . Let
Ak be any square nonsingular matrix of rank 2k in which the last k columns form
the matrix Ck . Such a matrix exists in view of linear independence of the columns
from Ck . Let T = (θn,m )n,m∈N be a semi-normal matrix such that (Ak ) is its diagonal
sequence. Then (lk ) = (2k ) is the rank sequence of T and sk = 2k+1 − 2 for k ∈ N.
Assume that T is equivalent to a normal matrix D. Then as explained earlier the
normal matrix B = D−1 = (βn,m )n,m∈N is such that TB is a matrix which defines a
bounded invertible linear operator L in c. Let us fix an integer r > 3(||L||˙||L−1 ||)2 .
Finally let em be the sequence such that em (n) = 0 for n = m and em (n) = 1.
For m ∈ (sr − r, sr ] and the sequence L(em ) let fm denote the sequence such that
L(em )(n) = fm (n) for m ≤ sr and fm (n) = 0 for n > sr . The first sr−1 coordinates
of the both sequences are equal to 0 as well. For these reasons and since L−1
corresponds to a semi-normal matrix with the rank sequence (2k ) we obtain that the
first sr coordinates of L−1 (fm ) and of L−1 (L(em )) are the same. Thus ||L−1 ||||fm ||∞ ≥
|L−1 (fm )(m)| = |L−1 L(em )(m)| = 1. But ||fm ||∞ = supsr−1 <n≤sr | ∑m≤j≤sr θn,j βj,m | =
∑m≤j≤sr |βj,m |. The last equality follows because there exists n ∈ (sr−1 , sr ] such that
θn,j = sign(βj,m ) and θn,j = ±1 for all sr − r < j < sr . Hence
∑ ∑ |βj,m | ≥ r/||L−1 ||.

sr −r<m≤sr m≤j≤sr
Because TB corresponds to a bounded operator L in c the sum of absolute values of

entries in each row does not exceed ||L||. Therefore for each n ∈ (sr−1 , sr ] we obtain
the inequality
||L|| ≥ ∑sr −r<m≤sr | ∑m≤j≤sr θn,j βj,m |. Now averaging over n ∈ (sr−1 , sr ] and taking
into account the Khinchin inequality:
√ 1
2−r ∑sr−1 <n≤sr | ∑sr −r≤j≤sr θn,j γj | ≥ (1/ 3)(∑sr −r≤j≤sr γj2 ) 2 ≥ √13r ∑sr −r≤j≤sr |γj |
which holds for each sequence (γj )sr −r<j≤sr we arrive at
||L|| ≥ 2−r ∑ ∑ | ∑ θn,j βj,m | =

sr−1 <n≤sr sr −r<m≤sr m≤j≤sr
1
∑ 2−r ∑ | ∑ θn,j βj,m | ≥ √ ∑ ∑ |βj,m |.
3r sr −r<m≤sr m≤j≤sr
sr −r<m≤sr sr−1 <n≤sr m≤j≤sr
This inequality together with the preceding one gives r ≤ 3(||L||||L−1 ||)2 which
yields a contradiction with the initial choice of r.
1. S. Mazur, Eine Anwendung der Theorie der Operationen bei der Untersuchung der
Toeplitzschen Limitierungsverfahren, Studia Math. 2 (1930).40–50.
Stanisław Kwapień
PROBLEM 7: MAZUR, BANACH

Are two convex infinite-dimensional subsets of a Banach space [of type (B)]
always homeomorphic?
Commentary
Keller [7] proved that all infinite-dimensional compact convex subsets of Hilbert
space are homeomorphic with the cube [0, 1]ℵ0 , and this is easily extended to
Fréchet spaces. Klee [9] showed that if K is a locally compact closed convex subset
of a Banach space then there are cardinal numbers m and n with 0 ≤ m ≤ ℵ0
and 0 ≤ n < ℵ0 such that K is homeomorphic with either [0, 1]m × (−∞, ∞)n or
[0, 1]m × [0, ∞). The various possibilities indicated are topologically distinct.
Klee [8] showed that Hilbert space is homeomorphic with all of its closed convex
bodies. Extending this and results of Stoker [10] and Corson and Klee [5], Bessaga
and Klee [2] showed that if K is a closed convex body in an arbitrary topological
linear space E, then K is homeomorphic with a closed halfspace in E or with the
product (for some finite m ≥ 0) of [0, 1]m by a closed linear subspace of codimension
m in E. From this, another result of Bessaga and Klee [3], and theorems of Kadeč [6]
and Anderson [1] on topological equivalence of Fréchet spaces, it follows that every
infinite-dimensional Fréchet space is homeomorphic with all of its closed bodies.
Most of the above material appears in the book of Bessaga and Pełczyński [4].
1. R. D. Anderson, Hilbert space is homeomorphic to the countable infinite product of lines, Bull.
Amer. Math. Soc. 72 (1966), 515–519.
2. C. Bessaga and V. Klee, Two topological properties of topological linear spaces, Israel J. Math.
2 (1064), 211–220.
3. , Every non-normable Fréchet space is homeomorphic with all of its closed
convex bodies, Math. Ann. 163 (1966), 161–166.
4. C. Bessaga and A. Pełczyński, Selected Topics in Infinite-Dimensional Topology, Monografie
Matematyczne, Vol. 58, Polish Scientific Publishers, Warsaw, 1975.
5. H. Corson and V. Klee, Topological classification of convex sets, Convexity (V. Klee, ed.),
Amer. Math. Soc. Proc. Symp. Pure Math. 7 (1963), 37–51.
6. M.I. Kadeč, A proof of the topological equivalence of all separable infinite-dimensional
Banach spaces (Russian), Funktsional. Anal. i Prilozhen. 1 (1967), 61–70.
7. O.H. Keller, Die Homeomorphie der kompakten konvexen Mengen in Hilbertschen Raum,
Math. Ann. 105 (1931), 748–758.
8. V. Klee, Convex bodies and periodic homeomorphisms in Hilbert space, Trans. Amer. Math.
Soc. 74 (1953), 10–43.
9. , Some topological properties of convex sets, Trans. Amer. Math. Soc. 78 (1955),
30–45.
10. J.J. Stoker, Unbounded convex point sets, Amer. J. Math. 62 (1940), 165–179.
V. Klee
PROBLEM 8: MAZUR
Prize: Five small beers, S. Mazur
(a) Is every series summable by the first mean representable as Cauchy product of
two converging series? Or else, equivalently,
(b) Can one find for each convergent sequence {zn } two convergent sequences {xn },
{yn } such that
x1 yn + x2 yn−1 + · · · + xn y1
zn = .
n
The problem reflects Mazur’s interests in the summability theory. It was solved
negatively in 1984, three years after Mazur had died. The solution was obtained
independently by P.P. B. Eggermont, Y.J. Leung, [1] and by S. Kwapień, A.
Pełczyński [2]. The problem is closely related to Problem 88. Both of them can
be easily reduced to problems on Hankel matrices. In 1982 V.V. Peller, [3], obtained
deep results on these matrices. The same author in [4] showed how to derive
solutions to Problems 8 and 88 from a theorem in [3] in an easy way. Beside
the paper [4] contains results on quantitative aspects of Problems 8 and 88 which
strengthen theorems and answer questions from [5].
Another solution to Problem 8 and its extension to some other classes of sequences
were obtained by C. Lennard and D. Redelet, [6].
1. P.P.B. Eggermont and Y.J. Leung, On a factorization problem for convergent sequences and on
Hankel forms in bounded sequences, Proc. Amer. Math. Soc. 96 (1986),269–274.
2. S. Kwapień and A. Pełczyński, On two problems of S.Mazur from The Scottish Book, Lecture
at the Colloquim dedicated to the memory of Stanisław Mazur, Warsaw Univeristy, (1985)
(unpublished).
3. V.V. Peller, Estimates of functions of power bounded operators on Hilbert space, J. Operator
Theory 7 (1982),341–371.
4. V.V. Peller, On S.Mazur’s Problems 8 and 88 from the Scottish Book, Studia Math. 180
(2007),191–200.
5. A. Pełczyński and F. Sukochev, Some remarks on Toeplitz mutltipliers and Hankel matrices,
Studia Math. 175 (2006),175–204
6. C. Lennard and D. Redelet, The Mazur product map on Hardy type sequence spaces, J. Math.
Anal. Appl. 350 (2009),384–392.

Theorem (Ulam). If E is a class of sets, each finite, each of which contains at most
n elements, and such that every n + 1 of these sets have a common element, then
there exists an element common to all sets of E.
Remark
The following is a slightly stronger statement of this theorem: If E is a class of sets

which contain a set with less than m elements and every m sets of E have an element
in common, then there exists an element common to all sets of E.
The proof is immediate.
J. Mycielski
PROBLEM 10: BANACH, MAZUR

Let H be an arbitrary abstract set and E the set of all real-valued functions defined
on H. The sequence xn (t) → x(t) (such that t ∈ H, xn , x ∈ E) if lim xn (t) = x(t) for
each t ∈ H.
Theorem. Each linear functional f (x) defined in E is of the form
n
f (x) = ∑ αi x(ti ),
i=1
where αi and ti do not depend on x.
Commentary
Apparently, Banach and Mazur never published a proof of this theorem. An

argument is given below. Note that the statement (in the terminology of that time)
that f is a linear functional includes the condition that f is continuous; i.e., if xn
converges pointwise to x, then f (xn ) converges to f (x).
Let B denote the space of bounded real-valued functions on H with the uniform
norm. Notice that f is a continuous linear functional on B, since if xn → 0, then
f (xn ) → 0. So there is a finitely additive set function μ defined on all subsets of H
so that if x ∈ B, then
f (x) = xd μ .
H
Let {An }∞n=1 be a sequence of pairwise disjoint subsets of H. If f ( χAn ) = cn = 0

for infinitely many n, then f (xn ) = 1 for infinitely many n, where xn → 0 pointwise.
This contradiction establishes that μ is countably additive and that there do not exist
infinitely many pairwise disjoint sets with nonzero measure. From this it follows that
there are finitely many points t1 , . . . , tn in H and numbers α1 , . . . , αn so that
n
f (x) = ∑ αi x(ti )
i=1
for all x ∈ B.
Finally, if x is an unbounded real-valued function defined on H, then x is the
pointwise limit of the functions xn , where xn (t) = n if |x(t)| ≥ n and xn (t) = x(t)
otherwise. Thus,
n
f (x) = ∑ αi x(ti )
i=1
for all x ∈ E. Clearly, every linear functional (in today’s terminology) defined on E
is of the given form if and only if H is finite.
R. Daniel Mauldin
PROBLEM 10.1: MAZUR, AUERBACH, ULAM, BANACH

Theorem. If {Kn }∞ n=1 is a sequence of convex bodies, each of diameter ≤ a and
the sum of their volumes is ≤ b, then there exists a cube with the diameter c = f (a, b)
such that one can put all the given bodies in it disjointly.
Corollary. One kilogram of potatoes can be put into a finite sack.
Determine the function c = f (a, b).
Commentary
Known as the “sack of potatoes” theorem, the first published proof is due to
Kosiński [1]. There it is established that in k-dimensional Euclidean space the bodies
can be put in a rectangular parallelepiped with edges 3a, 3a, . . . , 3a, (a + k!b/ak−1 ).
An
√ exact computationk−1 of the function f (a, b) is not given, but clearly f (a, b) ≤
k max{3a, a + k!b/a }. For k ≥ 3 Moon and Moser [3] give an improvement
of Kosinski’s main lemma; it follows that the bodies can be put in a rectangular
parallelepiped with edges 2a, 2a, . . . , 2a, 2(a + k!b/ak−1 ), and a similar estimate for
f (a, b) can be derived. Several related questions are also investigated in [3] and [2].
1. A. Kosiński, A proof of an Auerbach-Banach-Mazur-Ulam theorem on convex bodies, Coll.
Math. 4 (1957), 216–218.
2. A. Meir and L. Moser, On packing of squares and cubes, J. Combin. Theory 5 (1968), 126–134.
3. J.W. Moon and L. Moser, Some packing and covering theorems, Coll. Math. 17 (1967),
103–110.
Branko Grünbaum

Assume that there is a measure defined in the space of all integers. This measure
is finitely additive and any single point has measure zero. Let us extend this measure
to product spaces over the set of integers (finite of infinite products) in such a way
that the measure of a subproduct equals the numerical product of the measures of
its projections.
(a) Is the set of all sequences convergent to infinity measurable?
(b) Is the set of all pairs (x, y) where x, y are relatively prime measurable?
(c) Theorem (Schreier). The set of all pairs (x, y) where x < y is nonmeasurable.
Remark. A set is not measurable if a measure can be defined in it at least two
different ways and still satisfy the conditions above.
Solution
We show that the answer to (a) is no and that the answer to (b) depends on the
measure in question.
Let X be a set, B an algebra of subsets of X, μ a real-valued, finitely additive,
nonnegative function defined on B. The proof of the following theorem may be
found in [1].
Theorem. If E ∈ P(X) − B, then there is a finitely additive extension, μ̂ , of μ to
the algebra A , generated by the sets in B and E. Moreover, any two extensions of
μ take the same value on E if and only if
sup{ μ (B) : B ∈ B and B ⊆ E }

= inf{ μ (B) : B ∈ B and B ⊇ E }.
Let us remark that if the preceding equality holds, then there is a unique extension
of μ to A . In particular, if inf{ μ (B) : B ∈ B and B ⊇ E } = 0, then there is a unique
extension of μ to A .
Let us show that the answer to (a) is no. Let C = { xn ∈ N N : limn→∞ xn = +∞ }.
Let μ be a finitely additive probability measure defined on all subsets of N which
gives measure zero to singletons. Let M be the algebra of subsets of N N generated
by D, all sets of the form A1 × A2 × · · · × An × · · · where for each i, Ai ⊆ N and for
all but finitely many i, Ai = N. Let m be the unique finitely additive measure defined
on M such that
∞
m(A1 × A2 × · · · ) = ∏ μ (Ai ).
i=1
Suppose B ∈ M , B ⊆ C and B = 0. / There is some k such that B is of the form

T × N × N × · · · where T is a subset of N k and T = 0.
/ Let (x1 , . . . , xk ) ∈ T. Clearly,
this finite sequence can be extended to an infinite sequence xn which does not
converge to infinity. This contradicts the fact that T ⊆ C. Thus,
sup{ m(B) : B ∈ M and B ⊆ C } = 0.
Similarly, if T ∈ M and T ⊇ C, then T = N N . It follows from the theorem quoted

above that C is not measurable.
Let us show that the answer to (b) depends on what probability measure μ is
under consideration. Let W be the algebra of subsets of N × N generated by all sets
of the form A × B, where A, B ⊆ n and let R = { (x, y) : x and y are relatively prime }.
Let us note that a set K is in W if and only if K can be expressed as
n
K= Ai × Bi , (+)
i=1
where the sets Ai are pairwise disjoint.

Let μ be a 0-1 valued measure defined on all subsets of N such that μ gives
measure zero to singletons and μ (Q2 ) = 1, where
Q2 = { 2n : n ∈ N }.
Notice that R ⊆ Q2 × (N − Q2 ) ∪ (N − Q2 ) × N. From this it follows that
0= sup{ μ × μ (K) : K ∈ W and K ⊆ R }

= inf{ μ × μ (K) : K ∈ W and K ⊇ R }.
Thus, R is measurable with respect to the product measure μ × μ .

Let ν be a 0-1 valued measure defined on all subsets of N such that ν gives
measure zero to singletons and ν (P) = 1, where P is the set of primes.
Suppose A × B ⊆ R and ν × ν (A × B) = ν (A)ν (B) > 0. Since ν is 0-1 valued,
there is some z > 1 with z ∈ A ∩ B. But then (z, z) ∈ B. This contradiction shows that
sup{ ν × ν (K) : K ∈ W and K ⊆ R } = 0.
Suppose K ∈ W and K ⊇ R. Consider an expression for K of the form (+). Since
Aq ∪ · · · ∪ An ⊇ N − {1}, there is some i so that ν (Ai ) = 1. Let g be a prime, g ∈ Ai .
Since {g} × P − {g} ⊆ K, P − {g} ⊆ Bi . Thus, ν × ν (K) ≥ 1. It follows from the
quoted theorem that R is not measurable with respect to ν × ν .
The theorem stated by Schreier can be proven by similar methods.
This problem naturally leads to the following problem. For each n, n = 2, 3, 4, . . .,
let Un be the algebra of universally measurable subsets of N n . In other words,
for each finitely additive probability measure μ defined on all subsets of N which
vanishes on singletons, let Mn (μ ) be the algebra of all subsets on N n which are
measurable with respect to μ n , and let Un be the intersection of all such families.
Stan Williams has proved the following theorem.
Theorem. A subset E of N n , for n = 2, 3, . . . is universally measurable if and only
if there is a set B in the algebra generated by product sets and a finite subset F of N
such that B ⊂ E and
E−B ⊂ { πi−1 (F) : 1 ≤ i ≤ n },
where πi is the projection of N n into the ith coordinate.

In particular, if n = 2, E is universally measurable if and only if E is in the algebra
generated by product sets.
Added 2014. Williams also characterized the universally measurable sets for
infinite product spaces.
1. G. Birkhoff, Lattice Theory, Amer. Math. Soc. Coll. Publ. 25, revised edition, New York, 1967.
2. S. C. Williams, Universally measurable sets of finitely additive product measures, Illinois J.
Math. 33 (1989), 451–463.
R. Daniel Mauldin
PROBLEM 12: BANACH

A surface S is homeomorphic to the surface of a sphere and it has:
(a) a tangent plane everywhere
(b) a continuously varying tangent plane.
Is S equivalent to the surface of a geometric sphere? (That is to say, does there exist
a homeomorphism of the whole space which transforms the given surface S into the
surface of the sphere?)
Solution
We use the definition of a tangent plane that is stated in Problem 156: A plane T(q)
in Euclidean three-dimensional space E3 is tangent to a topological 2-sphere S at
the point q ∈ S if for every ε > 0, there exists a round ball B with center q such
that any straight line joining q to a point of S ∩ B − {q} makes an angle of size less
than ε with the plane T(q). We say that a topological sphere S has a continuous
family of tangent planes if S has a unique tangent plane at each of its points and,
for any sequence {qi } of points of S converging to a point q, the sequence {T(qi )}
converges to T(q).
Using current terminology, we can state Problem 12 as follows:
Is a 2-sphere S tame in E3 if S has a continuous family of tangent planes? We use
3
E to denote Euclidean 3-dimensional space. A topological 2-sphere S is defined to
be tame in E3 if there is a homeomorphism of E3 onto itself that carries S onto the
graph of x2 + y2 + z2 = 1; otherwise, S is wild in E3 . The existence of wild spheres
in E3 became known in the early 1920s with Alexander’s description of a “horned
sphere” [1] and Antoine’s construction of a wild Cantor set in E3 [2, 3]. References
for numerous other examples of wild spheres can be found in surveys of work on
embeddings of surfaces in E3 [5, 6]. It is our purpose in this note to describe a wild
sphere that has a continuous family of tangent planes.
We notice, with Example 1 below, that the definition stated above for a tangent
plane does not imply that there is a unique tangent plane at each point. However,
we do not permit this situation in our definition of a continuous family of tangent
planes.
Example 1. Let S denote the topological 2-sphere that is obtained by revolving the
graph of |x|1/2 +|z|1/2 = 1 about the z-axis. Any vertical plane that contains the point
(0, 0, 1) would, under the definition given above for a tangent plane, be tangent to S
at the point (0, 0, 1). In this example, any normal line to S at (0, 0, 1) fails to pierce
S at (0, 0, 1). (We say that a line L pierces S at a point q ∈ L ∩ S if there exist two
points p and r of L such that q is between p and r on L and the open intervals (p, q)
and (q, r) of L are in different components of the complement of S.)
We describe, with Example 2, a 2-sphere S which has a continuous family of
tangent planes such that there are points of S where the normal line does not
Figure 12.1 B
Figure 12.2
pierce S. While this sphere is tame in E3 , we use the example in our description,
with Example 3, of a wild sphere that has a continuous family of tangent planes.
Then we state a theorem about spheres that are pierced by their normal lines.
Example 2. Let D1 denote a disk in the xy-plane as indicated in Figure 12.1. (The
boundary of D1 is the union of three arcs AB, AC, and BC, where AB and AC are
closed intervals of two intersecting lines and BC is an arc of a circle that is tangent
to both lines.)
We let S = D1 ∪ D2 , where D2 is another disk, indicated as follows, which has

the same boundary as D1 . Let P be any vertical plane that intersects the interior
of D1 . The intersection of P with D2 is required to be a curve of the type indicated
in Figure 12.2. (Symmetry and similarity are not required.) This can be done so that
S has a continuous family of tangent planes.
Example 3. We describe a sphere S that is wildly embedded in E3 and has a
continuous family of tangent planes. To do this, we follow a procedure described by
Fox and Artin [9] to “entangle” S in the vicinity of the point A so that a wild sphere
S is obtained. Let D1 be a disk that is embedded in E3 as indicated in Figure 12.3.
(This is similar to the wild embedding of an arc described in Example 1.2 of the
paper by Fox and Artin.)
The wild disk D1 is in the disk D1 of Example 2 except near the overcrossings
(or undercrossings) where D1 is raised slightly above the plane of D1 to avoid self-
intersections. Now we place another disk D2 over D1 in a manner analogous to the
way D2 was placed over D1 in Example 2. We do this so that D1 and D2 have the
same boundary and D1 ∪D2 is a 2-sphere S in the closure of the bounded component
of the complement of the sphere S of Example 2. This can be done so that the
wild sphere S has a continuous family of tangent planes. For any vertical plane P
that intersects the interior of S , each component of P ∩ Int S would be an open
disk with a boundary as indicated in Figure 12.2, except that the adjustments near
the overcrossings might not permit the lower edge to be straight. If we picture the
disk D1 as a roadway approaching the point A, the heights of the overcrossings (or
D⬘1
D1
Figure 12.3
Figure 12.4
undercrossings) should approach zero and the slope of the roadway should approach
zero. Also, the ratio of the height of D2 above D1 to the width of the road should
approach zero as the road approaches A. We rely on the work of Fox and Artin in
the paper cited above to know that the topological sphere S is wild in E3 .
Theorem. If the topological 2-sphere S in E3 has a tangent plane at each point and,
for each p ∈ S, the line normal to S at p pierces S at p, then S is tame in E3 .
Proof. It follows from the definition of a tangent plane for a sphere S that each point
p ∈ S is the vertex of a solid double cone K that does not intersect S except at p. Since
the normal line to S at p pierces S at p, it follows that the two components of K − p
are in different components of E3 − S. (See Figure 12.4.) Cannon [7] and Bothe [4]
have independently shown that the existence of such a double cone at each point of
S implies that S is tame in E3 .
Remarks. We can construct a wild sphere like Alexander’s horned sphere [1] so that
it has a continuous family of tangent planes. We would first construct a tame sphere
S such that the disk x2 + y2 ≤ 1 in the xy-plane is in S and each vertical section of
the interior of S would be like Figure 12.2. Then a wild sphere S similar to the one
described by Alexander [1] would be constructed in S together with the bounded
component of its complement. The Cantor set of points where S is locally wild
would be a subset of the circle x2 + y2 = 1 in the xy-plane. As in Example 3, we
would need to exercise some control on cross-sections, slopes, heights, and ratios as
we approach a point of x2 + y2 = 1 along the horns.
A 2-sphere S would be tame in E3 if for each point q of S there is plane that
contains q and does not intersect S − q. We can readily show that the interior of
such a sphere is convex, and this implies that the sphere is tame. A more general,
and much more difficult theorem has very recently been proved by Daverman and
Loveland [8]. They have shown that a 2-sphere S is tame in E3 if there is a δ > 0
such that, for each p ∈ S, there is a round ball B of diameter δ such that p ∈ B and B
does not intersect the interior of S.
1. J. W. Alexander, An example of a simply connected surface bounding a region which is not
simply connected, Proc. Nat. Acad. Sci. U.S.A. 10 (1924), 8–10.
2. , Remarks on a point set constructed by Antoine, Proc. Nat. Acad. Sci. U.S.A. 10
(1924), 10–12.
3. L. Antoine, Sur L’homeomorphisme de Deux Figures et de leurs Voisinages, J. Math. Pures
Appl. 4 (1921), 221–325.
4. H. G. Bothe, Differenzeirbare Flächen sind zahm, Math. Nachr. 43 (1970), 161–180.
5. C. E. Burgess, Embeddings of surfaces in Euclidean three-space, Bull. Amer. Math. Soc. 81
(1975), 797–818.
6. C. E. Burgess and J. W. Cannon, Embeddings of surfaces in E3 , Rocky Mountain J. Math. 1
(1971), 259–344.
7. J. W. Cannon, -Taming sets for crumpled cubes II: Horizontal sections in closed sets, Trans.
Amer. Math. Soc. 161 (1971), 441–446.
8. R.J. Daverman and L.D. Loveland, Any 2-sphere in E3 with uniform interior tangent balls is
flat, Canadian J. Math. 33 (1981), 150–167.
9. R.H. Fox and E. Artin, Some wild cells and spheres in three-dimensional space, Ann. of Math.
(2) 49 (1948), 979–990.
C.E. Burgess
University of Utah
PROBLEM 13: ULAM

Let E be the class of all subsets of the set of integers. Two subsets K1 , K2 ∈ E are
called equivalent or K1 ≡ K2 if K1 − K2 and K2 − K1 are at most finite sets. There is
given a function F(K) defined for all K ∈ E; its range is contained in E and
F(K1 + K2 ) ≡ F(K1 ) + F(K2 )

F(compl. K) ≡ compl. F(K).
Question: Does there exist a function f (x) (x and f (x) natural integers) such that
f (K) ≡ F(K)?
Commentary
The answer is no. As is well known today there are 2c ultrafilters over the
integers E. For each ultrafilter U, let FU (X) be defined by FU (X) = E if X ∈ U,
FU (X) = 0,
/ otherwise. Each ultrafilter defines a distinct homomorphism F. Also the
corresponding functions f would have to be different, but there are only 2ℵ0 such
maps f .
Richard Laver
PROBLEM 14: SCHAUDER, MAZUR
Let f (x1 , . . . , xn ) be a function defined in the cube Kn . Let us suppose that f
possesses almost everywhere all the partial derivatives up to the rth order and the
derivatives up to the order r − 1 are absolutely continuous on almost every straight
line parallel to any axis. All the partial derivatives (up to the order r) ∈ Lp , p > 1.
Does there exist a sequence of polynomials {wi } which converges in the mean in
the pth power to f and in all partial derivatives up to the order r?
For r = 1 this was settled positively by the authors. An analogous problem exists
for domains other than Kn .
PROBLEM 15: SCHAUDER

Let f (x1 , . . . , xn ) be a function defined in Kn , i.e., in the n-dimensional cube. Does
there exist for every n some pn ≥ 2 such that if f ∈ Lpn then there exists a function
u(x1 , . . . , xn ) continuous on Kn :
(a) vanishing on the boundary of Kn ,
(b) possessing first derivatives on almost every line parallel to the axes and
absolutely continuous,
(c) possessing almost everywhere second partial derivatives (∈ Lpn ) and satisfying
the equation : Δ u = f .
The author proved that for n = 2, 3; pn = 2. Mazur observed that for n = 4, pn
cannot be equal to 2. For which n does there exist a pn > 2?
Solution
To present our solution to this problem, let Q = (− 12 , + 12 )n , Q̄ = [− 12 , + 12 ]n be the

open and closed unit cubes in Rn , respectively. Let ∂ Q be the boundary of Q. Fix
p > n.
Theorem: Given f ∈ Lp (Q) there exists a continuous function u on Q̄ with the
following properties.
• u = 0 on ∂ Q.
• For each x = (x1 , · · · , xn ) ∈ Q and j = 1, · · · , n, the partial derivative uj (x) :=
limh→0 [u(x1 , · · · , xj−1 , xj + h, xj+1 , · · · , xn ) − u(x1 , · · · , xn )]/h exists.
• For each j, i = 1, · · · , n, and for almost every (x1 , · · · , xi−1 , xi+1 , · · · , xn ) ∈

(− 12 , 12 )n−1 , the function F : xi → uj (x1 , · · · , xn ) is absolutely continuous.
• For each j, i, and for almost every x ∈ Q, the limit uij (x) := limh→0 [uj (x1 , · · · , xi−1 ,
xi + h, xi+1 , · · · , xn ) − uj (x1 , · · · , xn )]/h exists.
• For almost every x ∈ Q, we have ∑ni=1 uii (x) = f (x).
Remarks: This answers Schauder’s problem. The methods used to prove it were out
of reach when the problem was proposed, but are now standard.
Notation and Definitions
• We let ej = (0, 0, · · · , 0, 1, 0, · · · , 0) denote the jth unit vector in Rn .

• We write c, C, C , etc. to denote constants depending only on n, p. These symbols
may denote different constants in different occurrences.
• For 0 < α < 1, we write ||F||Lip(α ) = supx=y |F(x)−F(y)| |x−y|α . Note that ||F||lip(α ) = 0
if F is constant.
• A rectangular box in Rn is a Cartesian product I = (a1 , b1 ) × · · · × (an , bn ) of
open intervals (aj , bj ) with aj < bj .
The volume of I = (a1 , b1 ) × · · · × (an , bn ), written as |I|, is defined as the product
(b1 − a1 ) · (b2 − a2 ) · · · · · (bn − an ).
p
• We write Lcomp (Rn ) to denote the space of all functions F ∈ Lp (Rn ) that vanish
a.e. outside a compact set.
• We write C0∞ (Rn ) to denote the space of C∞ functions of compact support on Rn .
• The adjective“smooth" means C∞ unless we say otherwise.
• We write B(x, r) to denote the open ball about x with radius r in Rn .
• If Ω ⊂ Rn , then Ω̄ denotes the closure of Ω .
• The expression f ∗ g denotes the convolution, f ∗ g(x) = Rn f (y)g(x − y)dy.
Background
We start with basic properties of the Newtonian potential. References to the

literature and/or sketch of proofs may be found at the end of this article.
ϕ (x−y)
Let T ϕ (x) = −cn Rn |y|n−2 dy
(or T ϕ (x) = −const R2 (ln |y|)ϕ (x − y)dy if n = 2) for ϕ ∈ C0∞ (Rn ).

Then T ϕ is smooth. Define
∂ ∂2
Tj ϕ := ∂ xj (T ϕ ) and Tij ϕ = ∂ xi ∂ xj (T ϕ ) for ϕ ∈ C0∞ .
(A) For some α ∈ (0, 1) depending only on n, p, we have the following result. (the
Sobolev inequality).
p
T and Tj extend to operators from Lcomp (Rn ) to Lip(α ); we continue to
denote these extensions by T, Tj , respectively.
We have the inequalities:
• ||Tf ||Lip(α ) ≤ C(R)||f ||Lp (Rn ) if supp f ⊂ B(0, R); here C(R) depends only on
n, p, R.
• ||Tj f ||Lip(α ) ≤ C||f ||Lp (Rn ) .
(B) Moreover, Tij extends to a bounded operator from Lp (Rn ) to Lp (Rn ). We con-
tinue to denote this extension by Tij . We have the inequality (Calderón-
Zygmund): ||Tij f ||Lp (Rn ) ≤ C||f ||Lp (Rn )
(C) If f ∈ Lp (Rn ) and f = 0 a.e. in a neighborhood of x0 , then Tf , Ti f , Tij f are
smooth in a neighborhood of x0 .
(D) (Dirichlet Problem)
• Let f be a continuous function on ∂ Q. Then there exists a continuous
function u on Q̄, such that u = f on ∂ Q and u is harmonic in Q.
• Let Ω ⊂ Rn be a bounded open set with a smooth boundary ∂ Ω , and let f be
a continuous function on ∂ Ω . Then there exists one and only one continuous
function u on Ω̄ such that u = f on ∂ Ω and u is harmonic in Ω . Moreover,
if f is smooth in a neighborhood of some given x0 ∈ ∂ Ω , then u|Ω̄ ∩B(x0 ,δ )
extends to a smooth function on B(x0 , δ ) for some small δ > 0.
(E) (Strong Maximal Function)
For f ∈ Lp (Rn ) and x ∈ Rn , define
Mf (x) = sup{ |I|
1
I |f (y)|dy: I is a rectangular box containing x}
= sup{ |I|
1
I |f (x − y)|dy: I is a rectangular box containing 0}.
Mf is called the “strong maximal function” of f .
Then we have the inequality:
||Mf ||Lp (Rn ) ≤ C||f ||Lp (Rn ) for all f ∈ Lp (Rn ).
A computation with the strong max function

∞
We fix a smooth function θ (t) supported in [− 12 , 12 ], with −∞ θ (t)dt = 1. For δ > 0
−n x1
and x = (x1 , · · · , xn ) ∈ R , define θδ (x1 , · · · , xn ) = δ θ ( δ ) · · · θ ( xδn ).
n
Then Rn θδ (x)dx = 1, and

|θδ (x1 , · · · , xn )| ≤ δCn ∏nj=1 I[− 1 δ , 1 δ ] (xj ), where I denotes the indicator function.
2 2
1 h
For h > 0, δ > 0, F ∈ Lp (Rn ), we will study the quantity h 0 ( θδ ∗ F)(x + tei )dt =
θδ ,h ∗ F(x), where θδi ,h (x) = 1h 0h θδ (x + tei )dt.
i
For x = (x1 , · · · , xn ) ∈ Rn , we have

h1 C
|θδi ,h (x)| ≤ 0 h · [ δ n ∏j=i I[− δ ,+ δ ] (xj ) · I[− δ , δ ] (xi + tei )]dt ≤
C
|I| II (x), where I =
2 2 2 2
∏nj=1 Ij and
δ δ
Ij =(− , + ) j = i
2 2
δ δ
Ij =(−( + h), +( + h)) j=i
2 2
Consequently,
1 h C
| (θδ ∗ F)(x + tei )dt| ≤ |θδi ,h (y)||F(x − y)|dy ≤ |F(x − y)|dy ≤ C MF(x).
h 0 Rn |I| y∈I
Thus, we have shown that
1 h
| (θδ ∗ F)(x + tei )dt| ≤ CMF(x)
h 0
for all δ , h > 0, i = 1, · · · , n, F ∈ Lp (Rn ).

More generally, the same argument shows that
1 h
| (θδ ∗ F)(x + tei )dt| ≤ CMF(x + sei )
h 0
for any s ∈ (0, h) and therefore that
h h
| (θδ ∗ F)(x + tei )dt| ≤ C MF(x + sei )ds
0 0
for any δ , h > 0, i = 1, · · · , n and F ∈ Lp (Rn ).
The Analogue of Problem 15 on Rn
The operators T, Tj , Tij commute with convolution with θδ when acting on C0∞ , hence
p
also when acting on Lcomp (Rn ).
p
Let f ∈ Lcomp (R ). For x ∈ Rn , h > 0 and j = 1, · · · , n we argue as follows.
n
h−1 [Tf (x + hej ) − Tf (x)] =
lim h−1 [(θδ ∗ Tf )(x + hej ) − (θδ ∗ Tf )(x)]

δ →0
(since Tf is continuous)
= lim h−1 (T(θδ ∗ f )(x + hej ) − T(θδ ∗ f )(x))

δ →0
(since T commutes with convolution with θδ )
h
= lim {h−1 ∂j (T(θδ ∗ f ))(x + tej )dt}
δ →0 0
(by the fundamental theorem of calculus; note that T(θδ ∗ f ) is smooth since θδ ∗ f ∈
C0∞ (Rn ))
h
= lim h−1 Tj (θδ ∗ f )(x + tej )dt
δ →0 0
(by definition of Tj ).
Therefore,
(*)
h−1 [Tf (x + hej ) − Tf (x)] − Tj f (x) =
h
lim h−1 Tj (θδ ∗ f )(x + tej )dt − lim θδ ∗ Tj f (x)
δ →0 0 δ →0
(since Tj f is continuous)
h
= lim h−1 Tj (θδ ∗ f )(x + tej )dt − lim Tj (θδ ∗ f )(x)
δ →0 0 δ →0
(since Tj commutes with convolution with θδ )
h
= lim h−1 [Tj (θδ ∗ f )(x + tej ) − Tj (θδ ∗ f )(x)]dt.
δ →0 0
For each δ > 0, the Sobolev inequality (see (2A)) gives
|Tj (ϕδ ∗ f )(x + tej ) − Tj (ϕδ ∗ f )(x)| ≤ C|t|α ||θδ ∗ f ||Lp (Rn ) ≤ C|t|α ||f ||Lp (Rn ) .
Hence,
h
|h−1 [Tj (θδ ∗ f )(x + tej ) − Tj (θδ ∗ f )(x)]dt| ≤ Chα ||f ||Lp (Rn ) ,
0
and therefore (*) tells us that
|h−1 [Tf (x + hej ) − Tf (x)] − Tj f (x)| ≤ Chα ||f ||Lp (Rn ) .
Consequently,
Tj f (x) = lim h−1 [Tf (x + hej ) − Tf (x)].

h→0+
Similarly,
Tj f (x) = lim h−1 [Tf (x + hej ) − Tf (x)].

h→0−
Thus,
Tj f (x) = lim h−1 [Tf (x + hej ) − Tf (x)]

h→0
for all x ∈ Rn , j = 1, · · · , n, f ∈ Lp (Rn ).

p
We have shown that Tj f = ∂j Tf for all f ∈ Lcomp , not just for f ∈ C0∞ .
p
Next, for f ∈ Lcomp , x ∈ R , h > 0, and i, j = 1, · · · , n, we study the quantity
n
h−1 [Tj f (x + hei ) − Tj f (x)] =
lim h−1 [(θδ ∗ Tj f )(x + hei ) − (θδ ∗ Tj f )(x)]

δ →0
(since Tj f is continuous)
= lim h−1 [Tj (θδ ∗ f )(x + hei ) − Tj (θδ ∗ f )(x)]

δ →0
(since Tj commutes with convolution with θδ )
h
= lim h−1 ∂i Tj (θδ ∗ f )(x + tei )dt
δ →0 0
(by the fundamental theorem of calculus; note that Tj (θδ ∗ f ) is smooth, since θδ ∗
f ∈ C0∞ (Rn ))
h
= lim h−1 Tij (θδ ∗ f )(x + tei )dt
δ →0 0
(by definition of Tij )
h
= lim h−1 θδ ∗ Tij f (x + tei )dt
δ →0 0
(since Tij commutes with convolution with θδ ).

Therefore, by the results of section 3, we have
1
| [Tj f (x + hei ) − Tj f (x)]| ≤ CM(Tij f )(x)
h
and
h
(!) |Tj f (x + hei ) − Tj f (x)| ≤ C M(Tij f )(x + sei )ds.
0
The strong maximal inequality and the Calderón-Zygmund inequality now show
that the function
x → sup |h−1 [Tj f (x + hei ) − Tj f (x)]|

h>0
is pointwise less than or equal to CM(Tij f )(x) and thus has Lp -norm at most
C||M(Tij f )||Lp (Rn ) ≤ C ||Tij f ||Lp (Rn ) ≤ C ||f ||Lp (Rn ) .
Now define
εij f (x) = lim sup |h−1 [Tj f (x + hei ) − Tj (x)] − Tij f (x)|.
h→0
Then our previous inequality and the boundedness of Tij on Lp yields the inequality
(∗∗) ||εij f ||Lp (Rn ) ≤ C||f ||Lp (Rn )
p
for all f ∈ Lcomp (Rn ).
On the other hand, the definition of εij shows that
0 ≤ εij (f + g)(x) ≤ εij (f ) + εij (g).
Moreover, for functions θ ∈ C0∞ (Rn ) we have εij (θ ) = 0, since Tj θ is smooth and
Tij θ = ∂i Tj θ .
Given ε > 0 and f ∈ Lcomp p
(Rn ), we find a function θ ∈ C0∞ (Rn ) with
||f − θ ||Lp (Rn ) < ε , and write f = θ + (f − θ ).
Therefore,
0 ≤ εij (f ) ≤ εij (θ ) + εij (f − θ ) = εij (f − θ ).
and consequently, by (∗∗), we have
||εij (f )||Lp (Rn ) ≤ ||εij (f − θ )||Lp (Rn ) ≤ C||f − θ ||Lp (Rn ) ≤ Cε .

Since ε > 0 is arbitrarily small, we see that εij (f ) = 0 almost everywhere. That
is, for almost every x ∈ Rn , we have
Tij f (x) = lim h−1 [Tj f (x + hei ) − Tj f (x)].

h→0+
Similarly, for almost every x ∈ Rn , we have
Tij f (x) = lim h−1 [Tj f (x + hei ) − Tj f (x)].

h→0−
Thus, for almost every x ∈ Rn , we have Tij f (x) = lim h−1 [Tj f (x + hei ) − Tj f (x)].
h→0
Thus, at almost every x ∈ Rn , we have Tij f (x) = ∂i (Tj f )(x). We have shown this
p
for all f ∈ Lcomp , not just for f ∈ C0∞ .
∞
For θ ∈ C0 (Rn ), we have
n n
∑ Tii θ = ∑ ∂i2 T θ = Δ T θ = θ ,
i=1 i=1
since T inverts the Laplacian when acting on θ ∈ C0∞ . Since Tij is bounded on Lp ,
we conclude that
n
∑ Tii f = f (almost everywhere) for all f ∈ Lp (Rn ).
i=1
Next, we study the absolute continuity of
xi → Tj f (x1 , · · · , xn ) for fixed x1 , · · · , xi−1 , xi+1 , · · · , xn .

p
Fix i and j ∈ {1, · · · , n}, and f ∈ Lcomp (Rn ).

Given x := (x1 , · · · , xi−1 , xi+1 , · · · , xn ) ∈ Rn−1 , we define

F x (xi ) := Tj f (x1 , · · · , xi−1 , xi , xi+1 , · · · , xn )
for xi ∈ R.
Let Iν = [aν , bν ](ν = 1, · · · , N) be pairwise disjoint intervals.
Applying our previous inequality (!), with x = (x1 , · · · , xi−1 , aν , xi+1 , · · · , xn ) and
h = bν − aν , we learn that

|F x (bν ) − F x (aν )| ≤ C M(Tij f )(x1 , · · · , xi−1 , y, xi+1 , · · · , xn )dy.
y∈Iν
Summing over ν , we find that
N

∑ |Fx (bν ) − Fx (aν )| ≤ C y∈∪ν Iν
M(Tij f )(x1 , · · · , xi−1 , y, xi+1 , · · · , xn )dy
ν =1
∞ 1 1
≤ C( [M(Tij f )(x1 , · · · , xi−1 , y, xi+1 , · · · , xn )]p dy) p · (∑ |Iν |) p
−∞ ν
by Hölder’s inequality, where p is the exponent dual to p.

It follows that F x (xi ) is an absolutely continuous function of xi , provided
∞
(#) [M(Tij f )(x1 , · · · , xi−1 , y, xi+1 , · · · , xn )]p dy < ∞.
−∞
However, we have already seen that M(Tij f ) ∈ Lp (Rn ), and therefore (#) holds
for almost every (x1 , · · · , xi−1 , xi+1 , · · · , xn ) ∈ Rn−1 .
Therefore, the function xi → Tj f (x1 , · · · , xn ) is absolutely continuous on R, for

p
almost every (x1 , · · · , xi−1 , xi+1 , · · · , xn ); this holds for any f ∈ Lcomp (Rn ).
This concludes our study of the analogue of Problem 15 in Rn .
Passing from Rn to the Unit Cube
Let f ∈ Lp (Q), Q = (− 12 , 12 )n .
p
We introduce a function f̃ ∈ Lcomp (Rn ), defined by
f̃ (x1 , · · · , xn ) =
n
∑ (−1)i1 +···+in f ((−1)i1 (x1 −i1 ), · · · , (−1)in (xn −in ))· ∏ I(− 1 , 1 ) (xν −iν ).
2 2
(i1 ,i2 ,··· ,in )∈Zn ν =1
|i1 |,··· ,|in |≤100
In particular, f̃ = f on Q.
Let Rj,σ : Rn → Rn be the reflection of Rn with respect to the hyperplane Hj,σ =
{(x1 , · · · , xn ) : xj = σ }(σ = ± 12 ).
Thus, Hj,σ (x1 , · · · , xn ) = (x1 , · · · , xj−1 , 2σ − xj , xj+1 , · · · , xn ).
Then f̃ + f̃ ◦ Rj,σ = 0 on the cube (−10, +10)n , as one checks by elementary
computation.
The operator T commutes with Rj,σ when acting on C0∞ , hence also when acting
p
on Lcomp . It therefore follows from the result given in section (2C) that
T f̃ + T f̃ ◦ Rj,σ
is smooth on the cube (−1, 1)n .

In particular, T f̃ is smooth on the face Q̄ ∩ Hj,σ of the unit cube since Rj,σ =
identity on that face.
Applying the results of section 4 to f̃ , we now learn the following.
Let U = T f̃ |Q̄ . Then
• U is continuous on Q̄
• For each x ∈ Q and j = 1, · · · , n, the limit Uj (x) := limh→0 h−1 [U(x + hej ) − U(x)]
exists.
• For almost every x ∈ Q, and for each i, j = 1, · · · , n, the limit Uij (x) :=
limh→0 h−1 [Uj (x + hei ) − Uj (x)] exists.
• For each i, and for almost every (x1 , · · · , xi−1 , xi+1 , · · · , xn ) ∈ (− 12 , 12 )n−1 , the
function
xi → Uj (x1 , · · · , xi−1 , xi , xi+1 , · · · , xn )
is absolutely continuous on [− 12 , + 12 ].
• The restriction of U to {(x1 , · · · , xn ) ∈ Q̄ : xj = σ } is smooth, for each σ = ± 12
and each j = 1, · · · , n.
• ∑ni=1 Uii = f almost everywhere in Q.
Let V be the solution of the Dirichlet problem
⎡ ⎤
V continuous on Q̄,
⎢ ⎥
⎣ V = U on ∂ Q, ⎦ (see section (2D))
V harmonic in Q
Then
• V is continuous on Q̄
• For each x ∈ Q and j = 1, · · · , n, the limit Vj (x) = limh→0 h−1 [V(x + hej ) − V(x)]
exists.
• For each x ∈ Q and i, j = 1, · · · , n, the limit Vij (x) = limh→0 h−1 [Vj (x + hei ) −
Vj (x)] exists.
• The restriction of V to {(x1 , · · · , xn ) ∈ Q̄ : xj = σ } is smooth for each σ = ± 12 , j =
1, · · · , n.
• ∑ni=1 Vii (x) = 0 for all x ∈ Q.
We will prove that
xi → Vj (x1 , · · · , xi−1 , xi , xi+1 , · · · , xn ) is a smooth function on [− 12 , + 12 ], for
(†)
each (x1 , · · · , xi−1 , xi+1 , · · · , xn ) ∈ (− 12 , 12 )n−1 .
To see this, we may assume without loss of generality, that i = n.

Fix x◦ = (x1◦ , · · · , xn−1
◦ ) ∈ (− 1 , 1 )n−1 , and pick ε > 0 so small that |x◦ | + 10ε <
2 2 j
1
2
for each j.
Pick a smooth curve Γ as in Figure 15.1 containing vertical line segments starting
at (± 12 , 0).
the curve Γ
∋ 2
1 1
2 2
Fig. 15.1
Then define Ω ⊂ Rn to consist of all (x1 , · · · , xn−1 , xn ) ∈ Rn s.t. the point

(∑n−1 ∈ R2 lies in the shaded region enclosed by Γ and the x-axis.
1 (xi − xi ) , xn )
0 2
(We include the interval (− 12 , 12 ) × {0} on the x-axis, but we do not include the
curve Γ in the shaded region.)
Then
• Ω is a bounded domain with a smooth boundary in Rn .
• ∂ Ω contains a neighborhood of (x10 , · · · , xn−1
0 , 1 ) in Rn−1 × { 1 } and a neighbor-
2 2
hood of (x1 , · · · , xn−1 , − 2 ) in R
0 0 1 n−1 × {− 2 }.
1
• Ω ⊂ Q.
Now V is a harmonic function on Ω , V is continuous on Ω̄ , and V|∂ Ω is smooth
in a neighborhood of
1 1
(·) (x10 , · · · , xn−1
0
, ) and (x10 , · · · , xn−1 , − ).
2 2
According to the result given in section (2D), V extends to a smooth function on the
union of Ω with small balls centered at the two points (·). In particular, the function
1 1 1 1
(··) xn → V(x10 , · · · , xn0 , xn ) is smooth on [− , − + 2δ ]and on [ − 2δ , ]for small enoughδ .
2 2 2 2
In [− 12 + δ , + 12 − δ ], that function is smooth, simply because V is harmonic.
Therefore, the function (··) is smooth on [− 12 , + 12 ], completing the proof of (†).
We now define u = U − V. The known properties of U and V immediately imply

that u has all the properties asserted in the statement of the theorem on the 1st page.
This work was partially supported by NSF grant #: DMS-1265524. I am grateful

to Will Crow for TeXing this note.
References
In this section we provide references and/or sketch of proof for the background
results in Section 2.
(A) The proof of Theorem 2 part (iii) in Stein’s SINGULAR INTEGRALS AND
DIFFERENTIABILITY PROPERTIES OF FUNCTIONS, Princeton U. Press
1970 contains the essential ideas of the proof.
(B) See A.P. Calderon and A. Zygmund, On the existence of certain singular
integrals, Acta Math 88,1952, pp 339–393
(C) follows from the integral formulas in our references for (A) and (B) together
with the Lebesgue dominated convergence theorem.
(D) First Bullet point. If the restriction of f to each face is a trigonometric polyno-
mial, then a solution is easily produced by explicit formulas. One can then pass
to the general case by approximating every continuous f uniformly by "simple"
f ’s as described above. The harmonic extensions of the approximating f ’s
converge uniformly, thanks to the maximum principle for harmonic functions.
Second bullet point. Existence of solutions of the Dirichlet problem is proven
e.g. in Partial Differential Equations:Second Edition by L. C. Evans, Graduate
Studies in Math vol 19, AMS.
To obtain the smoothness of u near x0 as stated at the end of (D) one extends u
to be zero outside Ω . Let U be the resulting extension, and let F be the Laplacian
of U (as a distribution). Then F is supported on the boundary of Ω , and involves
terms containing u and its normal derivative (let’s call it v) on the boundary of
Ω . Since U must agree with f as we approach the boundary from the inside,
we obtain a singular integral (or pseudodifferential) equation for v in terms of f .
That equation is elliptic, hence v is smooth wherever f is smooth. In particular,
f and v are smooth in a neighborhood of x0 in the context of (D). Since u inside
Ω is given by solving Laplacian U = F on Rn , it follows that u is smooth up to
the boundary of Ω and may therefore be extended to a neighborhood of x0 as
asserted in (D).
(E) See B. Jessen, J. Marcinkiewicz and A. Zygmund, Note on the differentiability
of multiple integrals, Fundamenta Math. 25 (1935), pp 217–234.
CHARLIE FEFFERMAN
PROBLEM 15.1: MAZUR, ORLICZ
Prize: Two small beers, S. Mazur
Is a space E, of type F, for which there exists a sphere K which is bounded,
necessarily of type (B)? (A sphere is bounded if and only if χn ∈ K, and if the
numbers tn → 0, then tn χn → 0.)
Addendum. Negative answer: It suffices to take for E the space of numerical

sequences
∞
x= ∑ ξn
n=1
such that
∞
∑ |ξn |p < ∞; 0 < p < 1,
n=1
with ordinary operations, and

1/p
∞
x = ∑ | ξn | p
.
n=1
Instead of the space p one can also take Lp which consists of real-valued functions
x = x(t) in [0, 1], measurable, and such that
1
|x(t)|p dt < ∞,
0
with ordinary algebraic operations and

1
1/p
x = |x(t)| dt
p
.
0
MAZUR
May 1, 1937
Remark A space of type (F) is the terminology from Banach’s monograph,

Théorie des Opérations Linéaires, for a Fréchet space, a completely metrizable
topological vector space.
PROBLEM 16: ULAM

Find a Lebesgue measure in the space of all measurable functions satisfying the
following conditions:
If {Hn } are measurable sets contained on the line {x = xn }, then the set of all
measurable functions ff (x), satisfying the condition f (xn ) ∈ Hn has a measure equal
to |H1 | · |H2 | . . . where |Hn | denotes the measure of the set Hn .
Addendum. Such a measure exists.
S. Banach
May 15, 1941

Solution
If one takes M to be the set of all measurable functions from R into R, then there is
no such measure. This is easily seen by considering the set A = { f ∈ M : 0 ≤ f (1)
≤ 1 } and the set B = { f ∈ M : 0 ≤ f (1) ≤ 1 and 0 ≤ f (2) ≤ 2 }. If there were such
a measure, then the measure of A would be 1, while the measure of the subset B of
A would be 2.
On the other hand, if one takes M to be the space of all measurable functions
from R into I, the unit interval, then the answer is yes, even if one interprets the
words “Lebesgue measure” to mean a regular Borel measure in M relative to the
topology induced by the product topology of I R . (Obviously, it must have been this
case which Banach considered.) To prove this we first construct the standard product
measure μ in I R . Thus μ is defined over the σ -algebra Σ of subsets of I R generated
by all subsets of the form { f : f (t) ∈ A }, where t ∈ R and A is a measurable subset
of I. Now we use the following properties of Σ and μ .
(a) If t1 , . . . , tn are different real numbers and A1 , . . . , An are measurable subsets of
I, then

μ { f : f (t1 ) ∈ A1 , . . . , f (tn ) ∈ An } = λ (A1 ) · . . . · λ (An )
where λ is the Lebesgue measure.

This follows of course from the definition of μ .
(b) The completion of Σ relative to μ contains a Borel subsets of I R , in other words
for every Borel set B ⊆ I R there exist sets B0 ⊆ B ⊆ B1 , with B0 , B1 ∈ Σ and
μ (B0 ) = μ (B1 ).
This follows from a theorem on Haar measures, see [1, §64, Theorem H,
or 2]. For convenience of the reader we shall prove it directly, and our proof
generalizes to products of second countable spaces.
Proof. Since the completion Σ̄ of Σ is a σ -field, to prove (b) it is enough to show
that all open sets V ⊆ I R are in Σ̄ . Let I1 , I2 , . . . be a countable basis of open sets
for I. Let B be a basis of open sets for I R which consists of all cylinders over finite
nonempty products of the In ’s. Then V is a union of some sets Cs ∈ B (s ∈ S). Let
S0 ⊆ S be a countable set such that μ ( s∈S0 Cs ) is maximal. We put B0 = s∈S0 Cs .
Let T ⊆ R be a countable set such that B0 is a cylinder over an open subset of I T .
For any A ∈ B we put A∗ = (the cylinder in I R over the projection of A into I T ). Thus
A ⊆ A∗ , A∗ is open, and the range of the function ∗ is countable. Let us prove that
if A ∈ B and A ⊆ V then μ (A∗ − B0 ) = 0. First it is clear that A∗ − B0 ! ∈ Σ . Then
A = A∗ ∩ U, where U ∈ B and U is a cylinder over an open set in I S , where S ∩ T = 0. /
By the definition of B0 we have μ (A∪B0 ) = μ (B0 ). Hence 0 = μ (A−B0 ) = μ ((A∗ −
B0 ) ∩ U) = μ (A∗ − B0 ) · μ (U). Since μ (U) > 0 it follows that μ (A∗ − B0 ) = 0. Now
we put
B1 = { A∗ : A ∈ B, A ⊆ V }.
Since the range of ∗ is countable, we have μ (B1 ) = μ (B0 ). Also it is clear that
B0 , B1 ∈ Σ and B0 ⊆ V ⊆ B1 .
Remarks. Since the above proof uses the separability of I in an essential way it may
be interesting to recall the following facts related to (b) which hold for arbitrary
compact spaces and Baire measures:
(1) If μ is a product of Baire measures μt in compact spaces Ct , then μ is a Baire
measure in ∏ Ct .
Proof. By the Tychonoff theorem and the Stone-Weierstrass theorem every
continuous real-valued function φ (x) over ∏ Ct can be uniformly approximated
by functions of the form p(f1 (x(t1 )), . . . , fn (x(tn ))), where p is a polynomial and
fi is a continuous function over Ci . Hence φ is measurable relative to μ and hence
μ is a Baire measure.
(2) Every Baire measure (in Halmos’ sense [1]) is a locally compact space and has
a unique extension to a regular Borel measure (again in Halmos’ sense). This is
proved in [1, § 54, Theorem D]. However, this is not true for the more widely
used definitions. This has been demonstrated by Fremlin in a preprint.
If μ is a regular Borel measure in a compact space C and μ0 is the Baire
restriction of μ , then, by (1), μ02 is a Baire measure in C2 . However, Fremlin [4]
has given an example which shows that it is not necessarily true that μ 2 be
consistent with the unique Borel extension of μ02 . This solves a problem of
Bledsoe and Morse [3]. (If μ is a Borel measure in C which is not regular,
then μ 2 need not be a Borel measure in C2 (e.g., let C = { α : α ≤ ω1 } with
the interval topology. Thus C is compact. For any Borel set B ⊆ C, we put
μ (B) = 1, if B has a closed uncountable subset, and μ (B) = 0 otherwise. Thus
μ is a nonregular Borel measure, and, as is easily seen, the diagonal of C2 is not
μ 2 -measurable.)
(c) The inner measure of M in I R is 0 and the outer measure of M is 1.
Proof. For any f ∈ I R let
Af = { g : |{ t : f (t) = g(t) }| ≤ ℵ0 },
then Af intersects every nonempty set in Σ . Of course if f is measurable then Af ⊆ M

and if f is not measurable then Af ∩ M = 0.
/ And (c) follows.
In connection with this proof of (c) we have the following more special question.
Let μ0 be a probability Baire measure in a compact space C. Let μ¯0 S be the regular
Borel extension of the product measure μoS in CS . Let M ⊂ CS be any set invariant
under countable changes, i.e.,
|{ t : f (t) = g(t) }| ≤ ℵ0 =⇒ (f ∈ M ⇐⇒ g ∈ M),
and 0/ = M = CS (in particular |S| > ℵ0 follows). Is it true that the inner μ0S measure
of M is 0 and the outer μ0S measure of M is 1?}
By (a), (b), and (c) the outer measure μ ∗ restricted to the class of sets of the
form M ∩ X, where X ∈ Σ (= the completion of Σ ), has all the properties required in
Problem 16, restricted to functions in I R .
We are indebted to A. Hajnal, A. Mate, A. Ramsay, and R.J. Gardner for their
help in writing this commentary.
1. P.R. Halmos, Measure theory, Van Nostrand Publ. Co., New York, 1950.
2. E. Hewitt and K.A. Ross, Abstract Harmonic Analysis, Vol. I, Springer-Verlag, New York, 1963.
3. W.W. Bledsoe and A.P. Morse, Product measures, Trans. Amer. Math. Soc., 79 (1955), 173–215.
4. D.H. Fremlin, Products of Radon Measures: A counterexample, Canad. Math. Bull. 19, (1976),
no. 3, 285–289.
R. Daniel Mauldin and J. Mycielski
PROBLEM 17: ULAM

Prove a converse of Poisson’s theorem: that is, given a sequence of urns
containing white balls (1) and black ones (0), with unknown composition {pn }
and given also the result xn of drawing from each urn in turn, prove that with
probability 1,
1 n
lim
n→∞ n
∑ xi = p
i=1
implies that
1 n
lim
n→∞ n
∑ pi = p.
i=1
(See the Commentary to Problem 94.)

PROBLEM 17.1 ULAM
Let f be a continuous function defined for all 0 ≤ x ≤ 1. Does there exist a perfect
set of points C and an analytic function φ so that for all points of the set C we have
f ≡ φ?
Remark
Z. Zahorski (Sur l’ensemble des points singuliers d’une fonction d’une variable
réelle admettant les derivées de tous les ordres, Fund. Math. 34 (1947), 183–245)
showed that the answer is no and also raised a number of related problems.
A natural question is: What happens if one relaxes the requirement on the interpolant
g, asking it to be not analytic but just smooth? In particular, in the paper mentioned
above Z.Zahorski asked: Is it true that there is a function f ∈ C(I), I = [0, 1], such
that for every g in C∞ (I) (or in Cp (I)) the set
E(f , g) = {t : f (t) = g(t)}
is at most countable?
Here are some results in the subject, obtained in the 1980s - 1990s.
1. S. Agronski, A. Bruckner, M. Laczkovich, and D. Preiss proved that for every
f ∈ C(I) there exists g ∈ C∞ (I) such that E(f , g) is infinite [ABLP-85]. These
authors proved also that for every f ∈ C(I) there exists g ∈ C1 (I) such that E(f , g)
is a perfect set. The following result of Z.Buczolich [B-88] provides a jump to
“almost” C2 −functions: For every f ∈ C(I) there is a convex function g such that
E(f , g) is perfect.
2. However , it is not possible, in general, to get the interpolation by C2 − function
[O-94]: There is a Lipshitz function f on I, such that the set E(f , g) is at most
countable whenever g ∈ C2 (I). This gives the positive answer to Zahorski’s
question above. In fact, one cannot “jump” over any integer in the following
sense: Given n ∈ N there exists f of smoothness (n − ε ) so that E(f , g) is at most
countable whenever g is of smoothness (n + ε ).
3. On the other hand, one can “jump” from 1 to 2: For every f ∈ C1 (I) one can find
g ∈ C2 (I) such that E(f , g) is a perfect set ([O-94]). As a contrast , let us mention
the following result from that paper: There exists f ∈ C2 (I) such that no function
g ∈ C3 (I) interpolates f on a perfect set. It can be stated in a more general form:
Given n ≥ 2 there exists f ∈ Cn (I) such that every function g ∈ Cn+1 (I) may have
only finitely many points of tangency of maximal order with f ,
References
[ABLP-85] S.Agronsky, A.M.Bruckner, M.Laczkovich and D.Preiss, Convexity conditions and

intersections with smooth functions, Trans. Amer. Math. Soc. 298 (1985),659–677.
[B-88] Z.Buczolich , Sets of convexity of continuous functions, Acta Math.Hungar 52 (1988),
291–303.
[O-94] A.Olevskii, Ulam-Zahorski problem on free interpolation by smooth functions, Trans.
Amer. Math. Soc. 342 (1994) , 713–727.
A. Olevskii
PROBLEM 18: ULAM

Let a steady current flow through a curve in space which is closed and knotted.
Does there exist a line of force which is also knotted (knotted = nonequivalent
through any homeomorphism of the whole space R3 with the circumference of a
circle)?
We refer to the curve through which the current flows as the wire, and to lines of
force as (magnetic) field lines. Wires are understood to be simple closed curves.
This problem asks whether there always, for every knotted wire, exists a knotted
field line. That question remains unanswered. One much weaker interpretation of the
question, though, is whether there can exist a knotted field line, i.e., whether there
is some knotted wire which has some field line which is a simple closed knotted
curve.
That question has been answered in the affirmative using computer-assisted proof
techniques.
(a) (b)
(c)
Fig. 18.1 A knotted field line, shown (a) with the wire and (b) without the wire. The wire is drawn
in bold. A planar knot diagram is shown in (c).
(a) (b)
Fig. 18.2 An unknotted field line, shown (a) with the wire and (b) without the wire. The wire is
drawn in bold.
Theorem 4 (Minton 2015 [9]). Figures 18.1 and 18.2 depict two closed field lines
for the same wire. The wire is piecewise-linear and forms a trefoil knot. The field
line in Figure 18.1 is a trefoil knot, and the field line in Figure 18.2 is an unknot.
The piecewise-linear trefoil knot was chosen for simplicity and computational
efficiency; certainly, other choices could have been made. In fact, the computer
proof yields the following strengthening essentially for free.
Corollary 1. There is a wire with any desired knot type admitting a simple, closed
field line equivalent to the trefoil knot (respectively, to the unknot).
Magnetic field lines are generically closed for planar wires, but for nonplanar
wires they are not. The topology of field lines can be quite nontrivial [11], even
for simple examples [8]. Because they are not all closed, field lines can exhibit
interesting topological behavior like interlocking [15]. Studying the linking of field
lines has applications in physics ranging from solar dynamics [18] to tokamaks and
fusion energy [12]. There are explicit examples known of electromagnetic fields
in which every two field lines are linked [13, 6]. There are even explicit examples
known of electromagnetic fields in which every torus knot appears as a field line [7].
These are solutions of Maxwell’s equations, but they do not correspond to magnetic
fields arising from current through a single wire. A modern perspective on magnetic
field lines is that they are governed by a Hamiltonian system with one and a half
degrees of freedom [4, 16]. This allows one to invoke the theory of Hamiltonian
systems, implying properties like chaos and existence of certain periodic solutions
(i.e., closed field lines) [12]. This argument does not a priori give us control over
the topological structure of the closed field lines, though.
Computer assistance has been profitably used in rigorous mathematical proof
for some time now, with perhaps the most notable examples being the Four-Color
Theorem [1] and Hales’ proof of the Kepler Conjecture [5]. It has also proven
useful for showing existence of certain periodic solutions to differential equations
[17, 2, 3]. This is essentially our problem.
Because magnetic field lines obey an action principle [4], they can be viewed in
an existing computer-assisted proof framework [10]. Theorem 4 was proven with
the following basic procedure: find an approximate solution using (nonrigorous)
numerical calculations — this part uses standard scientific computing techniques —
and then prove existence of a true solution near to the approximate solution using
the following fixed-point theorem.
Existence Theorem. Let U and V be Banach spaces, let f : U → V be a differen-
tiable function, and let T : V → U be a linear operator. Suppose x0 ∈ U and ε > 0
are such that
||T||
|f (x0 )| · <ε
1 − ||idV − Df (x) ◦ T||
for all x ∈ B(x0 , ε ). Then there exists x∗ ∈ B(x0 , ε ) such that f (x∗ ) = 0.
This theorem converts the problem of finding a continuous object that exactly
satisfies a set of conditions into a problem of computing bounds on continuous
objects. Computers are relatively ill-suited to the former, but (e.g., using interval
arithmetic [14]) they are well-suited to the latter. We apply the Existence Theorem
to our problem by looking for a zero of the function f : x(t) → x (t)−B(x(t)), where
B is the magnetic field in question. The main complications arise from the fact that
the natural domain and codomain of this function f are infinite-dimensional; one
must reduce to a finite-dimensional problem by controlling for the effects of higher
Fourier modes [10, §III.4]. The details of the proof can be found in the report [9].
In the context of Theorem 4, what the computer actually proves is this: there is
a closed field line which is within 10−10 relative units of the provided approximate
curve. But in particular this means that, up to the precision of the diagram, the true
solution is identical to the figure shown.
Note that the conditions of the Existence Theorem are not necessarily satisfied
near every zero x∗ of every function f : the derivative Df (x∗ ) must be right-invertible.
So this technique is widely but not universally applicable. Generally speaking, for
a problem such as this, invertibility can be arranged by suitable choice of f if the
solution is isolated. So the fact that field lines are not generically closed — and thus
a closed field line is not likely to be in a continuous family of closed field lines —
is actually critical for the proof to work.
Finally, note also that this proof technique is designed to handle individual cases;
it does not give us a way to prove that knotted field lines always exist. If true, proving
that will require new techniques.
1. Kenneth Appel and Wolfgang Haken. Every planar map is four colorable, volume 98 of
Contemporary Mathematics. American Mathematical Society, Providence, RI, 1989. With
the collaboration of J. Koch.
2. Gianni Arioli and Hans Koch. Integration of dissipative partial differential equations: A case
study. SIAM Journal on Applied Dynamical Systems, 9(3):1119–1133, 2010.
3. CAPD - computer assisted proofs in dynamics, a package for rigorous numerics. http://capd.
ii.uj.edu.pl/. Accessed: 2015-04-14.
4. John R Cary and Robert G Littlejohn. Noncanonical Hamiltonian mechanics and its application
to magnetic field line flow. Annals of Physics, 151(1):1–34, 1983.
5. Thomas C. Hales. A proof of the Kepler conjecture. Ann. of Math. (2), 162(3):1065–1185,
2005.
6. William T. M. Irvine and Dirk Bouwmeester. Linked and knotted beams of light. Nature
Physics, 4:716–720, Aug 2008.
7. Hridesh Kedia, Iwo Bialynicki-Birula, Daniel Peralta-Salas, and William T. M. Irvine. Tying
knots in light fields. Phys. Rev. Lett., 111:150404, Oct 2013.
8. Keith Leon McDonald. Topology of steady current magnetic fields. American Journal of
Physics, 22(9):586–596, 1954.
9. Gregory T. Minton. Knotted field lines and computer-assisted proof. Preprint.
10. Gregory T. Minton. Computer-assisted proofs in geometry and physics. PhD thesis,
Massachusetts Institute of Technology, Sep 2013.
11. Tomoshige Miyaguchi, Makoto Hosoda, Katsuyuki Imagawa, and Katsuhiro Nakamura.
Topology of magnetic field lines: Chaos and bifurcations emerging from two-action systems.
Phys. Rev. E, 83:016205, Jan 2011.
12. P. J. Morrison. Magnetic field lines, Hamiltonian dynamics, and nontwist systems. Physics of
Plasmas, 7(6):2279–2289, 2000.
13. Antonio F. Rañada. A topological theory of the electromagnetic field. Letters in Mathematical
Physics, 18(2):97–106, 1989.
14. Nathalie Revol and Fabrice Rouillier. Motivations for an arbitrary precision interval arithmetic
and the MPFI library. Reliable Computing, 11(4):275–290, 2005.
15. Joseph Slepian. Lines of force in electric and magnetic fields. American Journal of Physics,
19(2):87–90, 1951.
16. G. F. Torres del Castillo. Debye potentials and Hamiltonians for magnetic field line flow.
Journal of Mathematical Physics, 36(7):3413–3418, 1995.
17. Yoshitaka Watanabe. A computer-assisted proof for the kolmogorov flows of incompressible
viscous fluid. Journal of Computational and Applied Mathematics, 223(2):953–966, 2009.
18. A R Yeates and G Hornig. Dynamical constraints from field line topology in magnetic flux
tubes. Journal of Physics A: Mathematical and Theoretical, 44(26):265501, 2011.
Gregory T. Minton
PROBLEM 19: ULAM

Is a solid of uniform density which will float in water in every position a sphere?
Commentary
Only a few special cases have been solved. In two dimensions there are counterex-
amples for density 1/2. For the limiting case of density 0, in two or three dimensions
the answer is a qualified yes. If the body is central symmetric of density 1/2 the
answer is yes, in any dimension.
The two-dimensional version of the problem concerns a cylinder of uniform
density which floats in every position, having the axis parallel to the water surface,
Figure 19.1 Two possible solutions. The line segment rotates within the curve, and in each
position cuts off half the area and half the perimeter.
and compatible with Archimedes’ law. H. Auerbach [1] showed that in the case of
density 1/2, the cylinder need not be circular, or even convex, and gave a class of
examples. We reproduce his illustration of two of them (Fig. 19.1).
For dimension 2 or 3 in the limiting case of density 0 the body must rest on a
plane in every position. L. Montejano [2] showed that the shell of the body must be
a sphere, and noted the example of a ball from which a smaller concentric ball had
been removed. The proof given would seem to generalize to arbitrary dimension.
For arbitrary dimension d and density 1/2, if S is star-shaped, symmetric,
bounded and measurable, then it differs from a ball by a set of measure 0. This
follows from Theorem 1.4 in R. Schneider [3]:
Let Ωd be the unit sphere |u| = 1, and , the inner product. “If φ is an even
measure on Ωd satisfying Ωd |u, v|dφ (u) = 0 for each v ∈ Ωd , then φ = 0.” See
also Problem 6331, Am. Math. Monthly 88 (1981), 150.
The simplest unsolved case seems to be that of dimension 2, central symmetry,
and density other than 0 or 1/2.
1. H. Auerbach, Sur un problème de M. Ulam concernant l’équilibre des corps flottants, Studia
Math. 7 (1938) 121–142.
2. L. Montejano, On a Problem of Ulam concerning a characterization of the sphere, Studies in
Applied Math. 53, (1974), 243–248.
3. R. Schneider, Functional equations connected with rotations and their geometric applications,
L’Enseignement Math. (2) 16 (1970), 297–305.
D. Hensley
The problem is still open in dimension three and only a few special cases have been
solved.
In three dimensions, there are no solutions, other than the sphere, in the limit
ρ → 0 or 1, [5]. Also, there are no nontrivial solutions among star-shaped objects
with central symmetry for density ρ = 12 , [4] and [6]. F. Wenger has recently
proposed a perturbation expansion scheme starting from the sphere for objects with
central symmetry and ρ = 12 [11], as well as for bodies with arbitrarily shape and
ρ = 12 [12]. His results point out towards the existence of many nontrivial solutions
in these wider classes of shapes, even though the proof is incomplete in that the
convergence of the perturbation series has not been examined. Furthermore, no
attempt to construct actual solutions of the problem, in dimension three, has been
reported. In this dimension, Várkonyi [8], following the spirit of Auerbach [1], took
a different approach to construct neutrally floating objects of density ρ = 12 with
cylindrical symmetry.
The two-dimensional version of the problem concerns a cylinder of uniform

density ρ which floats in water in equilibrium in every position, having the axis
parallel to the water surface. This problem is connected to several dynamical
systems, as for example, the tire track problem [7], the problem of the existence
of closed carrousels [3] and the problem of determining the trajectory of a charge
moving in a perpendicular parabolic magnetic field.
H. Auerbach [1] showed that in the case of density ρ = 12 , a cylinder of uniform
density ρ which floats in water in equilibrium in every position, having the axis
parallel to the water surface, need not to be circular, or even convex and gave a class
of examples. All these examples coincide with the Zindler curves [13] which, with
a suitable geometric construction, are in one to one correspondence with the family
of the figures of constant width. Auerbach’s illustration of two of them are given in
Figure 19.1.
If a figure D of density ρ floats in equilibrium in every position then the water
surface divides the boundary of D in constant ratio, say, σ : 1 − σ . We call σ the
perimetral density of D. In [9], F. Wenger was able to obtain noncircular solutions
for ρ = 12 = σ , by a perturbative expansion around the circular solution. These
figures have a p-fold rotational symmetry and have (p − 2) different perimetral
densities. On the other hand, Bracho, Montejano, and Oliveros [2] proved that if the
perimetral density σ is 13 or 14 , then the solution is circular. Later, in [10], F. Wegner
was able to give nontrivial explicit solutions to this two-dimensional version.
1. H. Auerbach, Sur un probleme concernant l’equilibre des corps flottant, Studia Math. 7 (1938),
121–142.
2. J. Bracho, L. Montejano and D. Oliveros, A clasification theorem for Zindler carrousels,
Journal of Dynamical and Control Systems. 7 (2001), 367–384.
3. J. Bracho, L. Montejano and D. Oliveros, Carrousels, Zindler curves and the floating body
problem, Period. Math. Hungar. 49 (2004), 1–17.
4. K. J. Falconer, Applications on a result on spherical integration to the theory of convex sets,
Amer. Math. Monthly 90 (1983), 690–693.
5. L. Montejano, On a problem of Ulam concerning the characterization of the sphere, Studies
Appl. Math. 53 (1974), 243–248.
6. R. Schneider, Functional equations connected with rotations and their geometrical applications,
L’Enseignement Math. 16 (1970), 297–305.
7. S. Tabachnokov, Tire track geometry; variations on a theme, Israel Journal of Math. 151
(2006), 1–28.
8. P. L. Varkonyi, Neutrally floating objects of density 12 , in three dimensions, Studies Appl. Math.
130 (2013), 295.
9. F. Wegner, Floating bodies of equilibrium Studies Appl. Math. 111 (2003), 167–183.
10. F. Wegner, Floating bodies of equilibrium in 2D, the tire track problem and electrons in a
parabolic magnetic field, arXiv:physics/0701241v3 (2007).
11. F. Wegner, Floating bodies of equilibrium in three dimensions. The central symmetric case.
arXiv:0803.1043 (2008).
12. F. Wegner, Floating bodies of equilibrium at density 12 in arbitrary dimensions.
arXiv:0902.3538 (2009).
13. K. Zindler, Über konvexe Gebilde II, Monatsh. Math. Phys. 31 (1921), 25–57.
Luis Montejano
PROBLEM 20: ULAM
Consider one-to-one and continuous transformations of the plane of the form
x = x, y = f (x, y) and y = y, x = g(x, y) and also transformations which result
from composing the above a finite number of times. Can every homeomorphic
transformation be approximated by such?
(Analogous problem for the n-dimensional space)
Remark
H.G. Eggleston (A property of plane homeomorphisms, Fund. Math. 42 (1955),

61–74) has proved the answer is no. He also shows that if the plane R×R is replaced
by the compact square [0, 1] × [0, 1] then the answer is yes. For related material see
Problem 47 and the accompanying commentary.
Jan Mycielski
PROBLEM 20.1: MAZUR, ORLICZ
For every positive integer n determine the smallest positive integer kn with the
following property: If f (x1 . . . . , xn ) is an irreducible polynomial, there exist points
(x11 , . . . , x1n ), . . . , (xkn 1 , . . . , xkn n ),
such that
f (λ1 x11 + . . . + λkn xkn 1 , . . . , λ1 x1n + . . . + λkn xkn n ),
considered as a polynomial of the variables λ1 , . . . , λkn , is irreducible. Is the

sequence kn bounded? (x11 , . . . , xkn n and λ1 , . . . , λkn are real or complex variables.)
Commentary
According to Professor Orlicz, Problems 20.1, 27, and 56 emerged in connection

with some problems which he and Mazur were considering [1, 2]. The exact
meaning of the problems they were considering seems to have become obscured.
Problems 20.1 and 56 still seem to be unsolved.
1. S. Mazur and W. Orlicz, Sur la divisibilité des polynomes abstraits, C. R. Acad. Sci. Paris 202
(1936), 621–623.
2. S. Mazur and W. Orlicz, Sur les fonctionnelles rationnelles, C. R. Acad. Sci. Paris 202 (1936),
904–905.
PROBLEM 21: ULAM

Can one make from the disc x2 + y2 ≤ 1 the surface of a torus using transforma-
tions with arbitrary small counter-images? (That is to say, for every ε > 0 there
should exist a transformation called f (p) of the disc into the torus, such that if
|p1 − p2 | ≥ ε then f (p1 ) = f (p2 ).)
Remark
This problem was solved negatively by T. Ganea in his paper “On ε -maps onto
manifolds,” Fund. Math. 47 (1959), 35–44. Also, compare with the commentary to
Problem 97.
Jan Mycielski
PROBLEM 22: ULAM, SCHREIER
Is every set Z of real numbers a Borel set with respect to set G which are additive
groups of real numbers? (Can any set Z be obtained through the operations Σ
performed countably many times and through operations of forming differences of
sets, starting with sets G such that if x, y belong to the set G, then x − y also belongs
to G?)
Remark
The answer to the question is no as it stands. This is because every set Z of real
numbers which is in the Borel field generated from the additive groups has the
property that if z ∈ Z, then −z ∈ Z. However, with this modification, the problem
seems to be unsolved. Also, see Erdős’ comments about this problem in his lecture.
R. Daniel Mauldin

DEFINITION A. A function defined in a certain n-dimensional region is called
monotonic in this region if, in every subregion, it assumes its maximum and
minimum on the boundary. A function is called a saddle function if, after subtracting
an arbitrary linear function, it is always monotonic.
DEFINITION B. Let C be a plane region; C = a Jordan curve which is its
boundary; K = a space curve over C with one-to-one projection. (That is to say,
two different points of K have different projections on C.) I shall say that the curve
K satisfies the triangle condition with a constant Δ , if the steepness of the plane
defined by any three different points of K √ is always ≤ Δ . By the steepness of a
plane z = ax + by + c, we mean the number a2 + b2 .
Rado (and later J. von Neumann) proved this theorem: The surface (function)
defined in a convex region C, which is continuous z = f (x, y), and is a saddle
function, and whose boundary curve satisfies the triangle condition with the constant
Δ satisfies a Lipschitz condition with the same constant Δ . That is to say, for any
two points (x1 , y1 ) and (x2 , y2 ) ∈ C we have:

|f (x1 , y1 ) − f (x2 , y2 )| = Δ (x1 − x2 )2 + (y1 − y2 )2 .
Problem A. What can one say if the boundary curve is assumed to be merely
continuous? For example, is a Lipschitz condition satisfied in every closed domain
contained entirely in the interior of C? Problem B. Can one prove anything
analogous to Rado’s theorem for functions of a greater number of variables (n ≥ 2)?
PROBLEM 24: MAZUR

Prize: Two Small Beers, S. Mazur
In a space E of type (B), there is given an additive functional F(x) with the
following property: If x(t) is a continuous function in 0 ≤ t ≤ 1 with values in E,
then F(x(t)) is a measurable function. Is F(x) continuous?
Remark
The answer is yes (See I. Labuda and R.D. Mauldin, Problem 24 of the Scottish
Book, Coll. Math. 48 (1984), 89–91).

Recently the theory of integral equations was generalized for singular integral
equations; that is to say, in which the integral expression K(s, t)g(t)dt is con-
sidered as an improper integral in the sense of Cauchy. Under certain additional
assumptions, the three well-known theorems of Fredholm (for equations with fixed
limits) are also valid. In the sense of the theory of operations, equations of this type
are probably not totally continuous in the corresponding spaces of type (B).
Problem. Find a new class of linear operations F(x), which contains as special
cases the integral equations of the above type (improper) and for which Fredholm
theorems do not hold anymore. The equations are of type: y = x + F(x).

Prize: One small beer, S. Mazur
Let E be a space of type (F0 ) and {Fn (x)} a sequence of linear functionals in
E converging to zero uniformly in every bounded set R ⊂ E. Is the sequence then
convergent to zero uniformly in a certain neighborhood of zero? [E is a type (F0 )
means that E is a space of type (F) with the following condition: If xn ∈ E, xn → 0
and the number series ∑∞ ∞
n=1 |tn | is convergent, then the series ∑n=1 tn xn is convergent.
R ⊂ E is a bounded region if xn ∈ R, and if the numbers tn → 0, then tn xn → 0.]
Addendum The answer is negative.

M. Eidelheit
June 4, 1938
Commentary
The answer is really negative, even in locally convex spaces. It follows from
Słowikowski’s example of a Montel space which is not a Schwartz space [1]. The
example is also presented in [2, p. 149]. The space is the following.
Let k, m, n1 , n2 be positive integers. Let n = (n1 , n2 ) and let
ak,m,n1 ,n2 = nk max(1, 2m−n1 ).
Let X be the space of all double sequences
x = {xn1 ,n2 }
such that
xk,m = sup ak,m,n1 ,n2 |xn1 ,n2 |

n1 ,n2
with the topology determined by the pseudonorms xk,m . We put
Fn (x) = a1,1,n1 ,n2 xn1 ,n2 .

The functionals Fn do not tend uniformly to 0 on any neighborhood of zero, because

a1,1,n1 ,n2
lim sup > 0.
n→∞ ak,m,n1 ,n2
On the other hand, for an arbitrary bounded set A we have
lim ak,m,n1 ,n2 sup |xn1 ,n2 | = 0

x∈A
for every k and m. In particular for m = k = 1, this implies that the functionals Fn
tend uniformly to 0 on each bounded set A.
1. W. Słowikowski, On (S)- and (DS)-spaces, Bull. Acad. Pol. Sci. Cl. III 5 (1957), 599–600.
2. S. Rolewicz, Metric Linear Spaces, Monografie Matematyczne Vol. 56, Polish Scientific
Publishers, Warszawa, Poland 1972.
Stefan Rolewicz
Prize: Five small beers, S. Mazur
Let E be a complex space of type (B); F(x), G(x) complex polynomials defined
in E. Let us assume that there exist elements xn ∈ E such that |xn | ≤ 1 and F(xn ) → 0,
G(xn ) → 0. Does there exist then an element x0 such that F(x0 ) = 0, G(x0 ) = 0?
Addendum The answer is positive. If there is no x0 ∈ E, such that F(x0 ) = 0,
G(x0 ) = 0, then there exist complex polynomials φ (x), ψ (x) in E with the property
that
F(x)φ (x) + G(x)ψ (x) ≡ 1.
Mazur, Orlicz
April 4, 1939
Commentary

1. S. Mazur and W. Orlicz, Sur la divisibilité des polynomes abstraits, C. R. Acad. Sci. Paris 202
(1936), 621–623.
904–905.
PROBLEM 28: MAZUR

Prize: Bottle of wine, S. Mazur
Let
∞
∑ an ,
n=1
be a series of real terms and let us denote by R the set of all numbers a for which
there exists a series differing only by the order of terms from
∞
∑ an ,
n=1
summable by the method of the first mean to a. Is it true that if the set R contains
more than one number but not all the numbers, then it must consist of all numbers
of a certain progression α x + β (x = 0, ±1, ±2, . . .)?
The same question for other methods of summation. [It is known that
(1) There exists a series
∞
∑ an
n=1
such that R consists of all terms of a sequence given in advance α x + β (x =

0, ±1, ±2, . . .);
(2) If the sequence {an } is bounded, then R consists of either one number or
contains all the numbers—the first case occurs only when the series ∑ an is
absolutely convergent.]
Commentary
Let A be a linear method of summation (for example, a matrix method) which

for some real series ∑∞ 0 an produces a sum s = A − ∑ an , while other series are
perhaps not A-summable. For a given method A and a series ∑ an we consider
all rearrangements ∑ ank of the series, single out those among them that are
A-summable, and consider the A-sums s. The set of all s may be called the “Riemann
set” of the method A and the series ∑ an . For example, Riemann’s theorem is that
for ordinary convergence, the Riemann set of a series is either a point or the whole
real line R. Steinitz’ theorem asserts that the Riemann set of ordinary convergence
for a series with complex terms is either one point, or a line, or the whole complex
plane. Bagemihl and Erdős [Acta Math. 92 (1954), 35–53] answered the first part
of Problem 28: for (C, 1) summability, the Riemann sets are precisely as described
there.
Lorentz and Zeller [Acta Math. 100 (1963), 149–169] answered the second part
of Problem 28 negatively. They proved that Riemann sets of matrix methods A may
be almost absolutely arbitrary. More exactly: a subset S of R is a Riemann set of
some regular matrix method and some series ∑ an is and only if S is an analytic set.
(Thus, all Borel sets S are Riemann sets.)
In particular, S may be any countable set. [But it is not known that this can
be combined with the additional property for series of bounded terms mentioned
in the addendum to Problem 28.] Obviously, there are many open problems here
of different types. The exact determination of all Riemann sets of some “natural”
summation method (such as the Abel or Euler method) is probably quite difficult.
G.G. Lorentz
The University of Texas at Austin
PROBLEM 29: ULAM
Is the group Hn of all homeomorphisms of the surface of an n-dimensional
sphere simple? (In the following sense: the component of identity does not contain
a nontrivial normal subgroup.) It is known (Schreier-Ulam) that the theorem holds
for n = 2 and the component of identity of Hn does not contain any closed, normal
proper subgroups for any n.
Commentary
The problem for the orientation-preserving homeomorphisms of S1 was solved

by Schreier and Ulam [4], and in 1947, Ulam and von Neumann showed the
comparable result for S2 [5].
The more general problem for Sm , m > 1 was partially solved in 1958 [1], with
conditions on a space X and group G of autohomeomorphisms of X guaranteeing
that every element of G is the product of six elements of G, each of the six being
a conjugate of an arbitrary non-identity element of G or its inverse. Examples of
spaces and groups satisfying the conditions include all stable autohomeomorphisms
of Sm (m > 1) and all autohomeomorphisms of the Cantor set, the rationals, the
irrationals, the Hilbert cube, the universal curve, etc. (An autohomeomorphism of a
manifold is called stable if it is the product of finitely many autohomeomorphisms
each supported on a cell.) It follows from the well-known 1968 results of Kriby and
Siebenmann that the annulus conjecture is true and thus all orientation-preserving
autohomeomorphisms of Sm are stable (m > 5) (later shown for m = 5). Thus
with the similar earlier known results for m = 1, 2, 3, Problem 29 is settled in the
affirmative for m = 4 (and the case m = 4 would follow from the annulus conjecture
for S4 ).
Later the author, in unpublished work, and then Nunnally [3] showed that for
a slightly different class of spaces and groups including all the examples cited
above, every element of G is the product of at most three conjugates of any
non-identity element of G. Note that the inverse need not be used. It is not hard
to see that in general two conjugates are not sufficient. But Nunnally showed that
for any “dilation” g (as, for example, a motion from one pole toward the other on
Sm ), two conjugates of g do suffice.
Other related papers dealing with inner automorphisms of G are by Fine and
Schweigert [2] and several papers by Whittaker in the early 1960s [6].
1. R.D. Anderson, The algebraic simplicity of certain groups of homeomorphisms, Am. J. Math.
80 (1958), 955–963.
2. N.J. Fine and G.E. Schweigert, On the group of homeomorphisms of an arc, Ann. of Math. 62
(1955), 237–253.
3. Ellard Nunnally, Dilations on Invertible Spaces, Trans AMS 123 (1966), 437–448.
4. J. Schreier and S. Ulam, Eine Bemerkung über die Gruppe der topolischen Abbildungen der
Kreislinie auf sich selbst, Studia Math. 5 (1934), 155–159.
5. S.M. Ulam and J. von Neumann, On the group of homeomorphisms of the surface of a sphere,
(abstract), Bull. AMS vol. 53 (1947), 506.
6. J.V. Whittaker, On isomorphic groups and homeomorphic spaces, Ann. of Math. 78 (1963),
74–91, MR27#737.
R.D. Anderson
PROBLEM 30: ULAM

Two elements a and b of a group H are equivalent if there exists h ∈ H such
that there is a relation a = hbh−1 . Two pairs of elements: a , a and b , b are called
simultaneously equivalent if there exists h ∈ H such that we have a = hb h−1 and
a = hb h−1 .
Question: For which groups does it suffice for simultaneous equivalence of two
pairs of elements a , a and b , b that every combination of the elements a and a
be equivalent to the corresponding combination of the elements b and b ? (The
necessity of this condition is obvious.)
PROBLEM 31: ULAM

June 18, 1936
In a metric group which is complete and compact, is the set of elements
equivalent to a given element always of first category? Does this theorem hold under
the additional assumptions that the group is connected or simple?
Addendum. Banach, Mazur counterexample:
Sa = eix → ei(x+a) , Tb = eix → e−i(x+b) .
Commentary
The solution of Banach and Mazur is a misunderstanding, since by the set of

conjugates of an element a in a group G, Ulam means { xax−1 : x ∈ G }. Of course in
every matrix group trace(xax−1 ) = trace(a). Hence in most matrix groups the set of
conjugates of any element is of dimension smaller than the dimension of the group.
This observation need not a priori extend to all groups mentioned in the problem
and so it remains unsolved.
(For the transfer of some results from matrix groups to all compact groups see
J. Mycielski, Some properties of connected compact groups, Coll. Math. 5 (1958),
162–166.)
J. Mycielski
PROBLEM 32: ULAM
Let G be a compact metric group (the group operation we shall denote by ×).
Does there exist for every ε > 0 a finite number of elements of the group:
a1 , a2 , . . . , aN for which we can define a group operation (denoted by the symbol o)
so that with respect to this operation the given finite system forms a group and:
(1) (ai × aj , ai oaj ) < ε ; i, j = 1, 2, . . . , N [(a, b) denotes the distance between the
elements a, b].
(2) The inverses of the elements ai (i = 1, 2, . . . , N) with respect to the two
operations are distant from each other by less than ε ?
Remark
A.M. Turing [Annals of Mathematics, 39 (1938), 105–111] showed that the only
finitely approximate Lie groups are the compact abelian groups. Thus SU(3) would
be a counterexample.
PROBLEM 33: ULAM

Two sequences of sets of real numbers An and Bn are called equivalent if there
exists an arbitrary function f mapping the set of all numbers into itself in a one-to-
one way and such that f (An ) = Bn . Questions:
(α ) Is every sequence An of projective sets equivalent to a certain sequence of Borel

sets?
(β ) Is every sequence of measurable sets—in the sense of Lebesgue—equivalent
to a sequence of Borel sets? Can one prove that there exists a sequence not
equivalent to any sequence of sets which are Lebesgue measurable?
Addendum. There exist sequences of projective sets and sequences of measurable

sets nonequivalent to sequences of Borel sets. (Communicated by Mr. Szpilrajn,
who obtained additional results concerning this notion of equivalence of sequences
of sets.) (Fund. Math. 26)
Ulam
August 1, 1935
Commentary
The answer to both α and β is no. In fact, Szpilrajn (Sur l’équivalence des suites
d’ensembles et l’équivalence des fonctions, Fund. Math. 26 (1936), 302–326)
showed that there is a sequence of (PCA) ∩ (CPA) = Δ21 sets which is not equivalent
to any sequence of Borel sets.
S. Ulam
PROBLEM 34: ULAM
A class K of sets of real numbers has the following properties:
(1) The class K contains all sets measurable in the sense of Lebesgue.
(2) If A ∈ K and B ∈ K, then A − B ∈ K.
(3) If An ∈ K, then ∑ An ∈ K.
(4) If the whole space is decomposed into a noncountable number of sets Aγ all
disjoint, each noncountable and each belonging to K, then there exists in the
class K a set which contains exactly one element from each of the sets Aγ .
Question: Is the class K the class of all subsets of our space?
Commentary
The answer is negative (see [1]). Under natural additional assumptions however,
the answer is positive. Those additional assumptions are that K be invariant under
translation and that for some integer n > 1 for every partition of R into sets of
cardinality n there exists a selector which belongs to K. This was proved in [2].
It was also proved in [2] that under the supposition of CH (more precisely the
supposition that every union of less than 2ℵ0 sets of measure zero is of measure
zero), there exists a σ -field F of subsets of R such that (1) it includes the field of
Lebesgue measurable sets, (2) it does not contain all subsets of R, (3) for every
partition of R into sets of cardinality 2ℵ0 there exists a selector which belongs to
F, (4) F is closed under images by all rational functions with real coefficients. The
proof of [2] depends on the algebraic structure of R and it is not known if one could
achieve invariance of F, e.g., under all homeomorphisms of R onto itself.
1. E. Grzegorek and B. Weglorz, Extensions of filters and fields of sets (I), Journal Austral. Math.
Soc. 25, Series A, (1978), 275–290.
2. B. Weglorz, Large invariant ideals on algebras, Alg. Univ. 13 (1981), 41–55.
Jan Mycielski
PROBLEM 35: ULAM
Is projective Hilbert space (that is to say, the set of all diameters of the unit sphere
in Hilbert space metrized by the Hausdorff formula) homeomorphic to the Hilbert
space itself?
Commentary
The answer is no because Hilbert space is simply connected and projective Hilbert
space is not simply connected (because it has a double covering).
W. Holsztynski
PROBLEM 36: ULAM

Can one transform continuously the full sphere of a Hilbert space into its
boundary in such a way that the transformation should be identity on the boundary?
Addendum. There exists a transformation with the required property given by
Tychonoff.
Commentary
This was answered affirmatively by a construction of Kakutani [2]. It was later

shown by Klee [3] that there is a homeomorphism h of the unit ball { x : x ≤ 1 }
onto the punctured unit ball { x : 0 < x ≤ 1 } such that h is the identity on
the boundary { x : x = 1 }. By results of Bessaga [1], h can even be made a
diffeomorphism. With f (x) = h(x)/h(x), f is a very nice transformation of the
desired sort.
1. C. Bessaga, Every infinite-dimensional Hilbert space is diffeomorphic with its unit sphere, Bull.
Acad. Polon. Sci., Ser. Sci. Math. Astr. et Phys. 14 (1966), 27–31.
2. S. Kakutani, Topological properties of the unit sphere of a Hilbert space, Proc. Imp. Acad. Tokyo
19 (1943), 269–271.
Soc. 75 (1953), 10–43.
V. Klee
PROBLEM 37: ULAM

A class of sets K is called a ring if: whenever A ∈ K, B ∈ K, then both (A + B) and
(A−B) ∈ K. Two rings of sets K and L are isomorphic if one can make correspond to
every set of the ring K, in a one-to-one fashion, a set of the ring L so that the sum of
sets goes over into the sum, the difference into the difference, and the counterimage
contains all the sets of the ring K. Questions:
(α ) How many nonisomorphic rings of sets of real numbers exist?

(β ) How many nonisomorphic rings of sets of integers exist?
(γ ) Is the ring of projective sets isomorphic to the ring of Borel sets?
Analogous questions for rings in the sense of countable addition, i.e., countable
summation of sets which belong to K also belongs to K.
Commentary
With no loss of generality assume that the problem is formulated for Boolean
algebras of sets, BAs. Concerning (α ), it is known that there are at least 2c such
c
BAs [1]. It is still open whether there can be 22 such (the maximum possible).
Similarly, concerning (β ), there are at least c such (by the folklore result that there
are c non-isomorphic denumerable BAs); the question is open whether there are 2c
such (the maximum possible).
The answer to (γ ) is clearly no: since both BAs are atomic, any isomorphism
between them must preserve infinite unions which exist in one or the other. But the
BA of projective sets is not closed under countable unions [2, p. 12].
The same questions were asked for σ -fields. For part (α ), the same remark
and open question holds. For part (β ), there are exactly ℵ0 such, by an obvious
argument. Part (γ ) is still open.
1. J.D. Monk and R.M. Solovay, On the number of complete Boolean algebras, Alg. Univ., 2 (1972),
365–368.
2. W. Sierpiński, Les ensembles projectifs et analytiques, Mémorial des Sciences Mathématiques,
Fascicule CXII, Gauthier-Villars, Paris, 1950.
J.D. Monk
PROBLEM 38: ULAM

Let there be given N elements (persons). To each element we attach k others
among the given N at random (these are friends of a given person). What is the
probability PkN that from every element one can get to every other element through
a chain of mutual friends? (The relation of friendship is not necessarily symmetric!)
Find limN→∞ PkN (0 or 1?).
Solution
First, if k ≥ 2, the resulting graph is connected with probability tending to 1. Here a

is joined to b if a knows b or b knows a. This may be seen as follows.
Suppose that the graph, G, which has N vertices, is not connected. Then it must
be possible to split G into two parts, G1 and G2 , so that |G1 | = r, |G2 | = N − r where
3 ≤ r ≤ N − 3 and there is no edge connecting an element of G1 to an element of
G2 . The probability that we do not join any of the r points of G1 to any of the N − r
points of G2 does not exceed

r−1 N −1 r N −r−1 N − 1 N−r
.
k k k k
Since k ≥ 2, this last estimate is less than

2r 2(N−r)
r N −r
.
N −1 N −1
Thus, the probability that there is a split is less than

2r
N r N − r 2(N−r)
∑ N −1 N −1
. (1)
3≤r≤N−3 r
To see that this sum goes to zero as N → ∞ notice that if 3 ≤ r < (N/8) and N >
8e2 /(8 − e2 ), then (er/N)r > (e(r + 1)/N)r+1 . Thus,
2r
N r N − r 2(N−r)
∑ r N −1 N −1
3≤r<N/8
r r
N r 2r
≤c ∑ ≤ c ∑ er
3≤r<N/8
r N N
≤ k/N 2 ,
where k and c are constants. Also, for all N,

2r
N r N − r 2(N−r) N e N/3
∑ r N −1 N −1
≤
3 3
N ≤r≤ N
8 3
and
2r
N r N − r 2(N−r) N e N/2
∑ r N −1 N −1
≤
2 4
.
N ≤r≤ N
3 2
These inequalities imply that the sum in (1) converges to zero.

Second, if k = 1, the resulting graph is connected with probability tending to
zero. If k = 1, we may consider this as a problem on random mapping functions
[1, p. 66]. Let f map {1, . . . , n} into itself by setting f (i) = j, provided i “knows” j.
Katz and Rényi proved the following theorem: If C(n) denotes the number of
connected mapping functions, then
C(n)
lim = (π /2)1/2 .
n→∞ nn−1/2
Thus,
lim P1N = 0.
N→∞
1. J.W. Moon, Counting Labelled Trees, Canadian Mathematical Monographs, No. 1, William
Clowes and Sons, Limited, London, 1970.
P. Erdős
Stan Ulam’s seminal question concerning the connectivity of random graphs with
given (out-)degree foreshadowed the emergence of the subject of the evolution of
random graphs, which was pioneered by Erdős and Renyi more than two decades
later. This has subsequently become a major stream in the flow of probabilistic
combinatorics. It is interesting that Erdős’ comments on the problem only treated
the (easier) problem for undirected graphs, whereas Ulam’s question dealt with the
(more difficult) case of directed graphs.
Ron Graham
More Second Edition Commentary and Solution
One interpretation of Ulam’s question is that it asks not about the limiting behavior
of the probability that certain random directed graphs are strongly connected, but
about whether associated undirected graphs are connected. We now formalize the
problem.
Consider a directed graph Dm,n and an undirected graph Um,n . Dm,n and Um,n
use the same set of n vertices, which we label v1 , . . . , vn . We construct Dm,n as
follows: independently for each vertex vi , we randomly select m distinct vertices
other than vi and draw directed edge from vi to each of these m vertices. Using Dm,n
we then construct Um,n as follows: we connect vi and vj by an undirected edge if
and only if Dm,n contains both − v→ −→
i vj and vj vi . Let Pm,n denote the probability that
Um,n is connected. For large n, we√wish to know how Pm,n depends on m. We show
√ occurs near m = n log n: for any ε > 0 and n sufficiently
that a transition √ large, if
m < (1 − ε ) n log n, then Pm,n is close to 0, whereas if m > (1 + ε ) n log n, then
Pm,n is close to 1.
We remark that a related question is when Dm,n is strongly connected. This
question has a different transition that probably occurs around m = log n, since Dm,n
is likely to have vertices with no in-edges when m is much smaller than log n, but
not when m is much bigger than log n.
Define Si = {vj : Dm,n contains −v→
i vj }.
Lemma 1 Fix any vertices w1 , . . . , wk and sets of vertices A1 , . . . , Ak such that
wi ∈/ Aj for any i, j. Then the probability that there is no directed edge in Dm,n from
Ai to wi for any i is at most
(|A1 |+···+|Ak |)
m
1− .
n−1
Proof. Since the edges emanating from any one node are chosen independently of
the edges emanating from any other node, the probability in question is:
n
∏ Pr(for all j such that vi ∈ Aj , there is no directed edge from vi to wj ) . (1.1)
i=1
Let ai denote the number of sets Aj that contain vi . The probability that there is no
edge from a particular vi to the wj ’s for which vi ∈ Aj is
n−m−1 ai
a (n − m − 1)(n − m − 2) · · · (n − m − ai ) n−m−1
n−1
i
= ≤ .
ai
(n − 1)(n − 2) · · · (n − ai ) n−1
Thus, the probability in (1.1) is at most

ai (a1 +···+an ) (|A1 |+···+|Ak |)
n
n−m−1 n−m−1 n−m−1
∏ n−1
=
n−1
=
n−1
.
i=1
√
Theorem 1 Fix ε , δ > 0. For n sufficiently large, if m = (1 − ε ) n log n, then
Pm,n < δ .
Proof. We say that a vertex is isolated if it has no incident edge in Um,n . Obviously,
Um,n cannot be connected if it has any isolated vertex. We will show that, as n gets
large, if m has the specified value, then Um,n is likely to have isolated vertices. To do
so, we compute the expected number of isolated vertices and the standard deviation;
we then apply Chebyshev’s Inequality.
Let p denote the probability that a particular vertex (say v1 ) is isolated in Um,n ,
and let Xm,n denote the number of isolated vertices in Um,n . The vertex v1 is isolated
if and only if, for each vj ∈ S1 , the vertex v1 is not in Sj . These events are independent
and all have probability 1 − n−1 m
, so
m
m
p = 1− .
n−1
We deduce that
m
m
EXm,n = np = n 1 − .
n−1
Next we consider the variance of Xm,n . Define q to be the probability that a

particular pair of vertices (say v1 and v2 ) are both isolated. In order to compute
q, we consider cases depending on whether or not v2 ∈ S1 and v1 ∈ S2 .
m 2

Case (1): v1 ∈/ S2 and v2 ∈ / S1 . This occurs with probability 1 − n−1 . Next
observe that v1 and v2 are both isolated if and only if the following two conditions
hold: for every vj ∈ S1 , we have v1 ∈/ Sj , and for every vj ∈ S2 , we have v2 ∈
/ Sj . By
Lemma 1, this occurs with probability at most
|S1 |+|S2 | 2m
m m
1− = 1− .
n−1 n−1
Thus case (1) contributes at most

2m+2
m
1−
n−1
to q.
Case (2): v1 ∈
/ S2 but v2 ∈ S1 . This occurs with probability

m m
1− .
n−1 n−1
Now v1 and v2 are both isolated if and only if the following two conditions hold:
for every vj ∈ S1 − {v2 }, we have v1 ∈
/ Sj , and for every vj ∈ S2 , we have v2 ∈
/ Sj . By
Lemma 1, this occurs with probability at most
|S1 |−1+|S2 | 2m−1
m m
1− = 1− .
n−1 n−1
Thus case (2) contributes at most

2m
m m
1−
n−1 n−1
to q.
Case (3): v1 ∈ S2 but v2 ∈
/ S1 . By symmetry, case (3) makes the same contribution
to q as case (2).
We need not worry about the case v1 ∈ S2 and v2 ∈ S1 , because then v1 and v2
could not be isolated. Summing cases (1), (2), and (3),
2m 2
m m
q ≤ 1− 1+ .
n−1 n−1
We next deduce the variance of Xm,n .

m 2m 2
m m m
2
EXm,n = np + n(n − 1)q ≤ n 1 − +n 2
1− 1+ .
n−1 n−1 n−1
Thus
m 2m 2
m m m
2
EXm,n − (EXm,n )2 ≤ n 1 − + n2 1 − .
n−1 n−1 n−1
Let σ denote the standard deviation of Xm,n . The value 0 is EXm,n /σ standard
deviations away from the mean, so by Chebyshev’s inequality, the probability
that Xm,n = 0 is at most σ 2 /(EXm,n )2 = (EXm,n 2 − (EX
m,n ) )/(EXm,n ) . Thus the
2 2
probability that there are no isolated vertices is at most

m m
m 2
m 2m
2 − (EX
EXm,n m,n )
2 n 1 − n−1 + n2 1 − n−1 n−1
≤ 2m
(EXm,n )2 n2 1 − m n−1
−m 2
1 m m
≤ 1− + . (1.2)
n n−1 n−1
√
Recall that m = (1 − ε ) n log n. Thus for large n, the first term in (1.2) is
approximately em /(n−1) /n ≈ e(1−ε ) log n /n ≈ n(1−ε ) −1 . Thus for sufficiently large
2 2 2
n, both terms in (1.2) are less than δ /2, and so the probability that there are no
isolated vertices (and hence Pm,n ) is less than δ .
Lemma 2 For integers m > i ≥ 0, a > j ≥ 0, and b > k ≥ 0, suppose m = a + b
and i = j + k. Then
(a − 1)(a − 2) · · · (a − j) (b − 1)(b − 2) · · · (b − k) aj bk
≤ i . (1.3)
(m − 1)(m − 2) · · · (m − i) m
Proof. We use induction on the pair j, k. For the base cases, we note that since
m > a, b, we have
(a − 1)(a − 2) · · · (a − j) aj
≤ j
(m − 1)(m − 2) · · · (m − j) m
and
(b − 1)(b − 2) · · · (b − k) bk
≤ k.
(m − 1)(m − 2) · · · (m − k) m
Now suppose that j and k are both nonzero. Suppose aj and bk are both less than i
m.
Then

j k i i im
i = j+k = a+ b< a+ b= = i,
a b m m m
which is a contradiction. Thus either aj or bk must be at least as big as mi . We can

assume without loss of generality that aj ≥ mi . Then a−j
a ≤ m . Therefore, if we
m−i
replace (a − j) and (m − i) by a and m, respectively, in the left-hand side of (1.3),

then the value does not decrease. The result now follows by induction.
√ 2 Fix 1/4 > ε > 0 and δ > 0. For n sufficiently large, if m =

Theorem
(1 + ε ) n log n , then Pm,n > 1 − δ .
Proof. Fix a set S of i vertices, where i ≤ n/2, and let Sc denote the complement
of S. We will say that S is “isolated” if there are no edges in Um,n from S to Sc .
We consider the probability that S is isolated. Let w1 , . . . , wi denote the vertices of
S, and for 0 ≤ k ≤ i, let Ak be the set of vertices x ∈ Sc such that there is an edge
−
w→k x. For some k, if exactly j of the m out-edges from wk end in S, then Ak has size
m − j. This occurs with probability
i−1 n−i
j m−j
n−1
m
m! (i − 1)(i − 2) · · · (i − j)(n − i)(n − i − 1) · · · (n − i − m + j + 1)
=
j!(m − j)! (n − 1)(n − 2) · · · (n − m)

m (i − 1)(i − 2) · · · (i − j)(n − i)(n − i − 1) · · · (n − i − m + j + 1)
= .
j (n − 1)(n − 2) · · · (n − m)

n−i m (i − 1)(i − 2)· · ·(i − j)(n − i − 1)(n − i − 2)· · ·(n − i − m + j)
= ,
n−i−m+j j (n − 1)(n − 2)· · ·(n − m)
which, for large n, is less than

m (i − 1)(i − 2) · · · (i − j)(n − i − 1)(n − i − 2) · · · (n − i − m + j)
2 .
j (n − 1)(n − 2) · · · (n − m)
By Lemma 2, this is at most

j
m i (n − i)m−j
2 .
j nm
S is isolated if and only if there is no directed edge from Ak to wk for any k. By

Lemma 1, the probability that this occurs is at most
j m−j i
m
m i (n − i)m−j m
∑2 j (n)m
1−
n−1
j=0
mi j i
(n − i)(n − m − 1) m
m i(n − 1)
=2 i
n(n − 1) ∑ (n − i)(n − m − 1)
j=0 j
mi mi
(n − i)(n − m − 1) i(n − 1)
=2i 1+
n(n − 1) (n − i)(n − m − 1)
mi
i (n − i)(n − m − 1) + i(n − 1)
=2
n(n − 1)
mi
i (n − i)(n − 1) + i(n − 1) − (n − i)m
=2
n(n − 1)
mi
(n − i)m 2 2 i
=2i 1 − < 2e−(n−i)m /n . (1.4)
n(n − 1)
√
Now using the fact that m = (1 + ε ) n log n , the right-hand side of (1.4) is less
than
i i
2e−(1+ε ) = 2n−(1+ε ) (n−i)/n .
2 (log n)(n−i)/n 2
n
We now multiply by i to bound the expected number of isolated sets of size i:
i
n
2n−(1+ε ) (n−i)/n
2
i
ne i i
2e 1−(1+2ε )(n−i)/n i 2e −2ε +(1+2ε )i/n i
2n−(1+ε ) (n−i)/n ≤
2
≤ n = n .
i i i
(1.5)
We now consider two cases. First suppose i ≤ ε n. Then the quantity in the right-hand
side of (1.5) is less than

2e −2ε +(1+2ε )ε i −ε /2 i
n < 2en
i
(since ε < 1/4). For any ε > 0, we can choose n sufficiently large that 2en−ε /2 is
arbitrarily small, and so we can make
εn i
∑ 2en−ε /2
i=1
arbitrarily small.
Finally consider the case ε n < i ≤ n/2. Let x = i/n. Then the quantity in the
right-hand side of (1.5) is less than

2e x i 2e x−1 i 2e i
n ≤ n ≤ √ .
xn ε ε n
Again, for any ε > 0, we can choose n sufficiently large that 2e

√
ε n
is arbitrarily
small, and so we can make
n/2 i
2e
∑ √
ε n
i=ε n
arbitrarily small. √
Thus for m = (1 + ε ) n log n , as n goes to infinity, the expected number of
isolated sets in Um,n of size ≤ n/2 goes to 0. But if Um,n is disconnected, then there
must be an isolated set of size ≤ n/2, so the probability that Um,n is connected goes
to 1.
√
Corollary 1 Fix ε , δ > 0. For n sufficiently large, if m ≥ (1 + ε ) n log n, then
√
Pm,n > 1 − δ , and if m ≤ (1 − ε ) n log n, then Pm,n < δ .
Proof. A straightforward coupling argument shows that (for fixed n), as m increases,
the probability that Um,n is connected cannot decrease. Therefore, this corollary
follows from Theorems 1 and 2.
Douglas S. Jungreis
PROBLEM 39: AUERBACH

The absolute value of a number x satisfies the following conditions:
(1) φ (x) ≥ 0, φ (x) ≡ 0, 1
(2) φ (x + y) ≤ φ (x) + φ (y)
(3) φ (xy) = φ (x)φ (y).
The only continuous functions satisfying these conditions are: φ (x) = |x|α , where
α is constant and 0 < α ≤ 1. Do there exist discontinuous functions with the above
properties?
Addendum This follows from Lebesgue’s theorem [See for example, E. Kamke,
Zur Definition der affinen Abbildung, Jahresb. d.D.M.V., 36 (1927): There exists
a complex function of a complex variable w = f (z) discontinuous and such that:
f (z1 + z2 ) = f (z1 ) + f (z2 ), f (z1 z2 ) = f (z1 )f (z2 ); φ (x) = |f (x)| satisfies (1), (2), (3),
and is discontinuous.]
S. Mazur
April 10, 1937
Commentary
That the only continuous functions satisfying the conditions are: φ (x) = |x|α , 0 <
α ≤ 1, follows of course from considering the functional equation φ (xy) = φ (x)φ (y)
and Cauchy’s equation into which it may be transformed: f (z1 + z2 ) = f (z1 ) + f (z2 ).
For a discussion of Cauchy’s equation, see [1].
The page references to the Kamke paper is pp. 145–156. Slightly more precisely
than is stated in the addendum, Kamke considered the following problem. Suppose
f (z) has the following properties:
(1) f (z) is defined for each z.
(2) f (z) takes on each value exactly once.
(3) f (z1 + z2 ) = f (z1 ) + f (z2 ).
(4) f (z1 z2 ) = f (z1 )f (z2 ).
In the real case, Darboux showed that f (z) = z. In the complex case there are also
solutions f (z) = z and f (z) = z̄. There are discontinuous solutions in the complex
case, as was proved by Steinitz (1910), Ostrowski (1913) and Noether (1916).
Kamke gives a construction for the discontinuous solutions using the well-ordering
principle.
It is now well known that there are 2c ring automorphisms of the complex
numbers [2, p. 157]. Since each such automorphism f must be the identity on the
positive rational numbers, it follows that if f is continuous, then f must be the
identity on the real numbers.
Thus, if |f | is continuous, then f must be either the identity or the conjugation
map. So, if f is any of the 2c discontinuous ring automorphisms of the complex
numbers, then φ (x) = |f (x)| is discontinuous and satisfies (1), (2), and (3).
1. J. Aczel, Lectures on functional equations and their applications, Academic Press, New York-
London, 1966.
2. N. Jacobson, Lectures in Abstract Algebra, III, Van Nostrand, New York, 1964.
W.A. Beyer and R. Daniel Mauldin


Can one define a completely additive measure function for all the projective
sets on the interval (0, 1) which, for Borel sets, coincides with Lebesgue measure?
In particular, can one define such a measure on the ring of sets of the sets P(A)
(projective)? All this with the additional requirement that congruent sets should
have equal measure.
Commentary
It is now known that this problem is connected with the axioms of set theory.
For example, if ZF+ “there is an inaccessible cardinal” is consistent, then it is
consistent that all sets are Lebesgue measurable [5]. If the axiom of projective
determinacy is consistent, then all projective sets are Lebesgue measurable [3]. If
there is a projective well-ordering of the real numbers into type ω1 (which is the
case under Gödel’s axiom of constructibility), then there is no countably additive
measure defined on all projective sets which coincides with Lebesgue measure [4].
Kakutani and Oxtoby (2) and Hulanicki [1] obtained some absolute results
concerning extensions of Lebesgue measure.
1. A. Hulanicki, Compact abelian groups and extensions of Haar measures, Rozprawy Mat. 38
(1964), 58 pp. (MR31 #270.)
2. S. Kakutani and J.C. Oxtoby, Construction of a non-separable invariant extension of the
Lebesgue measure space, Ann. of Math., (2) 52 (1950), 580–590. (MR12-246.)
3. D.A. Martin, Descriptive Set Theory: Projective Sets, in Handbook of Mathematical Logic,
Edited by John Barwise, Studies in Logic, volume 90, North-Holland Publ. Co., New York,
1977.
4. R.D. Mauldin, Projective well orderings and extensions of Lebesgue measure, Coll. Math., 46
(1982), 185-188.
5. R.M. Solovay, A model of set theory in which every set of reals is Lebesgue measurable, Ann.
of Math. 92 (1970), 1–56.
6. S.M. Ulam, Problems in Modern Mathematics, John Wiley, New York, 1960.
R. Daniel Mauldin
PROBLEM 41: MAZUR

Does there exist a 3-dimensional space of type (B) with the property that every
2-dimensional space of type (B) is isometric to a subspace of it? This is equivalent to
the question: Does there exist in the 3-dimensional Euclidean space a convex surface
W which has a center 0 with the property that every convex curve with a center is
affine to a plane section of W through 0? More generally, given an integer k ≥ 2,
does there exist an integer i and an i-dimensional space of type (B) such that every
k-dimensional space of type (B) is isometric to a subspace of it; given k, determine
the smallest i.
Commentary
By very simple reasoning, Grunbaum [2] showed that no 3-dimensional Banach

space is isometrically universal for all 2-dimensional Banach spaces. Bessaga [1],
with more complicated reasoning, obtained the result with “3” replaced by
“finite.” Further refinements were contributed by Melzak [5], Klee [3], and
Lindenstrauss [4].
1. C. Bessaga, A note on universal Banach spaces of finite dimension, Bull. Acad. Polon. Sci. Ser.
Sci. Math. Astr. Phys. 6 (1958), 249–250.
2. B. Grünbaum, On a problem of S. Mazur, Bull. Res. Council Israel (Sect F), 7 (1958), 133–135.
3. V. Klee, Polyhedral sections of convex bodies, Acta Math. 103 (1960), 243–267.
4. J. Lindenstrauss, Notes on Klee’s paper: “Polyhedral sections of convex bodies,” Israel J. Math.
4 (1966), 235–242.
5. Z. Melzak, Limit sections and universal points of convex surfaces, Proc. Amer. Math. Soc., 9
(1958), 729–734.
V. Klee
PROBLEM 42: ULAM

To every closed, convex set X, contained in a sphere K in Euclidean space, there
is assigned another convex, closed set f (X), contained in K, in a continuous manner
(in the sense of the Hausdorff metric); does there exist a fixed point, that is to say, a
closed convex X0 such that f (X0 ) = X0 ?
Theorem (Mazur). Let E be a class of convex closed sets contained in a sphere
K with the properties:
(1) If A ∈ E, B ∈ E, then also λ A + (1 − λ )B ∈ E, for 0 ≤ λ ≤ 1 [λ A + (1 − λ )B
denotes the set of points λ x + (1 − λ )y for x ∈ A and y ∈ B];
(2) If An ∈ E and the sequence {An } converges to A, then A ∈ E.
Suppose that f (x) is a continuous function in E whose values belong to E; then,
there exists a fixed point; that is, an X0 ∈ E such that f (X0 ) = X0 . Examples of such
a class E are, for instance, the class of all closed, convex sets contained in K with
diameter not greater than a given number φ > 0.
There is a canonical embedding of the set of all nonempty convex compact subsets
of a linear space into another linear space. The latter is normed if so is the former.
This embedding preserves the convex combinations. In problem 42, the image of
this embedding would be a compact convex subset in a normed space and the result
(existence of a fixed point) follows from Schauder’s theorem.
Michael Zarichnyi
PROBLEM 43: MAZUR

Prize: One bottle of wine, S. Mazur
Definition of a certain game. Given is a set E of real numbers. A game between
two players A and B is defined as follows: A selects an arbitrary interval d1 ;
B then selects an arbitrary segment (interval) d2 contained in d1 ; then A in his
turn selects an arbitrary segment d3 contained in d2 and so on. A wins if the
intersection d1 , d2 , . . . , dn , . . . contains a point of the set E; otherwise, he loses. If
E is a complement of a set of first category, there exists a method through which
A can win; if E is a set of first category, there exists a method through which B
will win. Problem. It is true that there exists a method of winning for the player
A only for those sets E whose complement is, in a certain interval, of first category;
similarly, does a method of win exist for B if E is a set of first category?
Addendum. Mazur’s conjecture is true.
S. Banach
August 4, 1935
Modifications of Mazur’s Game
(1) There is given a set of real numbers E. Players A and B give in turn the digits 0
or 1. A wins if the number formed by these digits in a given order (in the binary
system) belongs to E. For which E does there exist a method of win for player
A (player B)?
Ulam
(2) There is given a set of real numbers E. The two players A and B in turn give
real numbers which are positive and such that a player always gives a number
smaller than the last one given. Player A wins if the sum of the given series
of numbers is an element of the set E. The same question as for (1). [Ed. See
Problem 67 for another modification.]
Banach
Commentary
The first published paper on general finite games with perfect information is
Zermelo’s [14]. Here, in Problem 43, we have the first interesting definition of
an infinite one. A proof of the solution of Banach which is announced here was
published by Oxtoby [5]. This theorem constitutes a very useful characterization of
meager or comeager sets. E.g., this characterization immediately yields that the set
of real numbers, such that in their binary representation 1 has a frequency, is meager
(yet, by the strong law of large numbers, the set of those numbers in which 1 has
frequency 1/2 has a complement of measure zero). For other applications of this
criterion, see [4].
The theorem of Banach also has the following corollary (to be compared with the
theorems which follow): If C is a class of subsets of the real line R which is closed
under preimages by continuous functions f : R → R, then every set E ∈ C has the
property of Baire iff for every set E ∈ C the game of Mazur is determined (i.e., one
of the players has a winning strategy).
A similar game was introduced by Morton Davis [1]. Here a set E ⊆ {0, 1}ω is
given and Player I chooses a finite sequence of 0’s and 1’s, Player II chooses 0 or
1 and again I chooses a finite sequence of 0’s and 1’s and II chooses 0 or 1, etc.,
ω times. If the juxtaposition of the consecutive choices belongs to E then I wins,
otherwise II wins. Davis has proven that I has a winning strategy iff E has a perfect
subset, and II has a winning strategy iff E is countable or finite.
Ulam’s modification (1) of Mazur’s game is particularly important. Let C be
a class of subsets of the unit interval [0, 1] which is closed under some natural
operations, e.g., under finite unions and preimages by Borel measurable functions.
Then, if for every set E ∈ C the Ulam game is determined, every set E ∈ C has the
property of Baire, is Lebesgue measurable, and is either countable or has a perfect
subset and the same is true for the complement of E (see [1, 11, 12, 13]).
The conjecture that all projective sets are determined is called the axiom of
projective determinacy and is often used in the theory of projective sets since it
has very natural or fitting consequences which cannot be proved otherwise (and are
inconsistent with V = L), see [10].
Ulam’s game is more general than Mazur’s or Davis’ in the sense that for every
game of the latter kinds an equivalent game of the first kind can be defined (see [11]).
In spite of this generality all known proofs of the existence of nondetermined games
of Ulam require the axiom of choice for uncountable families of sets of reals.
Therefore, and because of its excellent consequences, it is a plausible conjecture
that if we remove from set theory the full axiom of choice and put in the axiom
of determinacy (which tells that for every E ⊆ [0, 1] the Ulam game is determined)
the resulting theory is consistent. But this conjecture is beyond the reach of present
methods because the axiom of determinacy yields the consistency of set theory with
very strong axioms of infinity [Solovay, Martin, Harrington], whence (by Gödel)
one cannot prove the opposite.
The best one can do, therefore, is to prove for larger and larger classes C that
for all E ∈ C the Ulam game is determined. The strongest results in this direction
were obtained by D.A. Martin. In [7] he proves this for the class of all Borel sets.
This theorem is outstanding because, although it pertains to sets of low set-theoretic
rank, it still uses the full power of Zermelo-Skolem set theory (often incorrectly
called Zermelo-Fraenkel set theory). Friedman had proven earlier [3] that this could
not be demonstrated in Zermelo’s set theory. In [6] Martin proves that if there exists
a cardinal κ having the Erdős-Rado partition property κ → (ω1 )< ω then for all
2
analytic sets E the Ulam game is determined. Finally, in [8] he proves, under the
assumption of the existence of some extremely large cardinals (called iterable), that
for all E ∈ Σ12 (continuous images of complements of analytic sets) the Ulam game
is determined. Those results of Martin constitute brilliant examples of the impact of
strong axioms of infinity upon the theory of projective sets. The problem of proving
the full axiom of projective determinacy from some strong axiom of infinity is still
open, and is perhaps the most outstanding problem of set theory.
Why don’t we abandon the axiom of choice and accept in its place the axiom of
determinacy, in spite of the fact that this removes such artificial phenomena as para-
doxical decompositions of the sphere and yields so many “positive” consequences?
The answer is that it seems more natural to restrict those consequences (and the
axiom) to a suitable class C as above, e.g., the class of projective sets, without
violating our basic intuitions about the class of all sets (of which the axiom of choice
is a part) and without wrecking the unicity of our fundamental theory.
Banach’s modification (2) is of a more special character than that of Ulam. Some
work has been done on such modifications (see [2, 9]). It appears that the specific
question of Ulam and Banach (the question of who is the winner) does not lead to
nontrivial answers. One may observe that a countable set is avoidable by any of the
players and that the existence of a winning strategy means that the corresponding set
has a perfect subset of a particular shape. For studies of some other infinite games
with perfect information, see [11] and the commentary to Problem 67. Other such
games were recently studied by F. Galvin, R. McKenzie, R. Laver, J. Mycielski,
K. Prikry, and many others. A paper of F. Galvin, J. Mycielski, R. Solovay and
some others is in preparation.
1. Morton Davis, Infinite games of perfect information, Advances in game theory, Ann. Math.
Studies No. 52, Princeton University Press, Princeton, N.J. 1964, 85–101.
2. A. Ehrenfeucht, G. Moran and S. Shelah, Size direction games over the real line, Israel J. Math.
14 (1973), Part I, 163–168, Part II, 418–441, and Part III, 442–449.
3. H. Friedman, Higher set theory and mathematical practice, Ann. of Math. Logic 2 (1971),
326–357.
4. J. F. Lynch, Almost sure theories, Ann. of Math. Logic 18 (1980), 91–135.
5. J.C. Oxtoby, The Banach-Mazur game and Banach category theorem, Contributions to the
theory of games, vol. III Ann. of Math Studies no. 39, Princeton 1957, 159–163.
6. D. A. Martin, Measurable cardinals and analytic games, Fund. Math. 66 (1970), 287–291.
7. D. A. Martin, Borel determinacy, Ann. of Math. 102 (1975), 363–371.
8. D. A. Martin, Infinite games, Proc. of the Int. Congress of Mathematicians in Helsinki, 1978,
269–273, Acad. Sci. Fennica, Helsinki, 1980.
9. G. Moran, Existence of nondetermined sets for some two person games over the reals, Israel
J. Math. 9 (1971), 316–329.
10. Y. Moschovakis, Descriptive Set Theory: A Foundational Approach, North-Holland, Amster-
dam, 1980.
11. J. Mycielski, On the axiom of determinateness, Part I, Fund. Math. 53 (1964), 205–224, and
Part II, Fund. Math. 59 (1966), 203–212.
12. J. Mycielski and H. Steinhaus, A mathematical axiom contradicting the axiom of choice, Bull.
Acad. Polon. Sci. Sér. Math., Astronom. Phys. 10 (1962), 1–3.
13. J. Mycielski and S. Świerczkowski, On the Lebesgue meaurability and the axiom of determi-
nateness, Fund. Math. 54 (1964), 67–71.
14. E. Zermelo, Über eine Anwendungen der Mengen lehre auf die Theorie des Schachspiels, Proc.
Fifth Congress of Math., Cambridge 1912, vol. II, 501.
Jan Mycielski
The investigation of the determinacy of infinite games is the most distinctive and
intriguing development of modern set theory . . . .
Akihiro Kanamori, Introduction to the Handbook of Set Theory [7, page 44]
Infinite games of various types – not only the three types of games in Problem
43 of the Scottish Book – now appear in many branches of set theory. In some cases,
the game is introduced for the reason indicated in the questions of Mazur, Ulam,
and Banach: the existence of a winning strategy for Player I (or II) is a useful way
of characterizing some interesting property. In other cases, the interest lies in the
determinacy of the game.
This commentary will be confined to the three types of games in Problem 43,
which will be discussed here in reverse order. Since Jan Mycielski’s 1981 commen-
tary there have been new results on each of the three. All of the results involve the
determinacy of the game, not the original question.
Before discussing these games, it is necessary to digress briefly for the benefit
of those readers who are unfamiliar with set theory (at the level of Jech [5] or
Kanamori [6]). To such a reader, there is much about this commentary which might
seem strange, for three reasons, which in increasing order of strangeness are as
follows. First, we work in ZF (not ZFC), and consider a number of propositions
which contradict the axiom of choice. Second, we will be concerned solely with
various propositions which are not decidable in ZF (and in some cases not in ZFC),
and the relationship between two such propositions. One type of relationship is
implication. The other type involves consistency. It may be that (ZF + P) and (ZF +
Q) are equiconsistent. And it may be that (ZF + P) implies the consistency of (ZF +
Q); if so, then the consistency of (ZF + P) implies the consistency of (ZF + Q) but,
by Gödel’s Theorem, the consistency of (ZF + Q) does not imply the consistency of
(ZF + P). In this case, we say that P has greater consistency strength than Q. And
third, large cardinal axioms are involved.
A large cardinal axiom asserts that there exists a cardinal number satisfying a
particular property. That property is often technical and its motivation will not be
evident to the non-specialist. It will also not be evident that cardinals satisfying
that property are large. Large cardinal axioms are linearly ordered by consistency
strength, and it is standard practice in set theory to gauge the consistency strength of
a proposition by comparison to large cardinal axioms. Some – though certainly not
all – set theorists hold the following three beliefs: first, that it is meaningful to say
that a mathematical statement is true, even when there is no possibility of proving
it; second, large cardinal axioms are probably true; third, the collection of axioms
that constitute “the usual axioms of set theory" should be expanded to include large
cardinal axioms. For a more serious discussion of these matters, see Kanamori [6]
and Maddy [11].
For any set X with at least two members and any set E ⊂ X ω , the game GX (E) is
the infinite game in which Players I and II alternately play elements x0 , x1 , x2 , . . . of
X, and Player I wins iff the sequence (x0 , x1 , x2 , . . .) is in E. When X = 2 (= {0, 1}),
this is the game of Ulam’s question. For any such X, let ADX be the proposition that
for every E ⊂ X ω , GX (E) is determined. AD2 and ADω are equivalent; this is the
axiom of determinacy, denoted simply as AD. AD and ADR are the only axioms of
this form consistent with ZF. For a history of determinacy axioms, see Larson [10].
Clearly ADR implies AD. It is also clear, since a strategy for a game on ω can be
identified with a real number, that AD implies that AD holds in L(R), the smallest
model of ZF containing all reals and all ordinals. But ADR does not hold in L(R),
so AD does not imply ADR . In fact, ADR has greater consistency strength than
AD. This follows from the fact that AD holds in L(R), together with the theorem of
Solovay [17] that ADR implies that “the sharp" of L(R) exists.
For any E ⊂ R, the Banach game, B(E), is the game of Banach’s question. The
axiom of determinacy for Banach games, ADB, is the proposition that every Banach
game is determined.
One of the consequences of the 1981 publication of the Scottish Book is that Tony
Martin learned about Banach games. He showed that one could simulate a game on
2 by a Banach game [4, 1.2]. That is, he defined a function E → E∗ from 2ω into
R and then proved that if Player I has a winning strategy for the game B(E∗ ) then
he also has a winning strategy for G2 (E), and similarly for Player II. This clearly
shows that ADB implies AD, and it also provides evidence that Banach’s original
question is unanswerable.
Since Banach games are games on reals, trivially ADR implies ADB. Thus
ADR ⇒ ADB ⇒ AD, and since ADR is stronger than AD it is not possible that both
arrows can be reversed. This leads Martin to ask how strong ADB is – as strong as
AD, as strong as ADR , or somewhere in between? There are two questions here, one
of implication, the other of consistency strength.
Note that a strategy for a game on reals is not a real, but rather is – modulo
some coding – a set of reals. Freiling [4, 1.3 and 1.4] provided a partial answer to
Martin’s question by proving that if there is any winning strategy for the game B(E)
then there is a winning strategy (set of reals) which is fairly simple to define – so
simple that it is in the model L(R). Therefore, the relationship between ADR and
ADB is the same as the relationship between ADR and AD: ADR implies that ADB
holds in L(R) and the sharp of L(R) exists, hence ZF + ADB is consistent. This still
left open the question of how ADB is related to AD.
Using the work of Freiling and some unpublished work of Martin, Becker [2]
then proved that AD implies ADB. Hence AD and ADB are equivalent, relative
to ZF.
This equivalence has local versions, which we state using the notation of
descriptive set theory (see [8] or [13]). Let Γ be a collection of pointsets in R
and in 2ω . Call Γ nice if every Δ13 set is in Γ , Γ is closed under finite unions
and intersections and Γ is closed under preimages by Δ12 functions. For example,
the projective pointclasses Πn1 , Σn1 and Δ1n for n ≥ 3 are nice, as is the class of
projective sets and the class of all sets. For any nice Γ the following are equivalent:
for any E ⊂ R, if E ∈ Γ then B(E) is determined; for any E ⊂ 2ω , if E ∈ Γ then
G2 (E) is determined. The latter fact is called Γ -determinacy. The equivalence of
AD and ADB is the special case in which Γ is all sets. For the other Γ ’s mentioned
above, the equivalent propositions are compatible with the axiom of choice, and
many set theorists would say that they are (probably) true.
Why should we believe that for projective E ⊂ 2ω , the determinacy of Ulam’s
game G2 (E) is true (or even consistent)? One of the reasons for this belief is that
large cardinal axioms imply that G2 (E) is determined. Proving the determinacy
of definable games from large cardinal axioms is arguably the most important
development in set theory since Cohen’s invention of forcing in 1963. In 1981,
when Mycielski’s commentary was published, this subject was still in its infancy.
The strongest result mentioned there has now been improved in two ways: the large
cardinal axiom has been weakened and the collection of definable games has been
enlarged beyond Σ21 . (There is a typo in that commentary: Σ12 should be Σ21 .)
The relevant large cardinal axiom was discovered by Hugh Woodin and now
bears his name. A cardinal κ is a Woodin cardinal if for each function f : κ → κ there
exists an elementary embedding j : V → M with critical point λ < κ closed under
f such that V(j(f ))(λ ) ⊂ M. Woodin was led to formulate this definition by previous
work of Foreman, Magidor, Shelah and himself, work which was not directly related
to proving determinacy.
Shortly after Woodin formulated the above definition, Martin-Steel [12] proved
that if there exist n Woodin cardinals with a measurable cardinal above them then
1 -determinacy is true. Later Woodin proved that the existence of ω Woodin
Πn+1
cardinals with a measurable above them implies that for every E in L(R), G2 (E) is
determined [14, 8.24]; hence L(R) is a model of AD. And he proved from a stronger
large cardinal axiom that there is a model of ADR [18, 8.3].
All proofs of definable determinacy from large cardinal axioms have the follow-
ing form. The game G2 (E) is simulated by a game of the form Gκ (E∗ ), where κ is a
large cardinal and E∗ is a closed subset of κ ω . Closed games are determined. What
makes the proof work is that E∗ is a very special type of closed set, namely the set
of branches of a homogeneous tree. For information on homogeneous trees, see [6]
and [10].
There is a known upper bound on the amount of definable determinacy that can
be proved from large cardinal axioms. Work of Abraham, Shelah, and Woodin (see
[1]) shows that one can always produce a forcing extension of the universe in which
there is a nondetermined Δ21 game. From this it follows that no large cardinal axiom
can imply Δ21 -determinacy.
It is not possible for determinacy axioms, or any other statement about small
sets, to imply the actual existence of large cardinals, since one can pass to the
initial segment of the universe below the least inaccessible cardinal and thereby
get a model with no large cardinals but no change in the small sets. Consistency, as
opposed to implication, is another matter. Woodin proved that the theories (ZFC +
there exist infinitely many Woodin cardinals) and (ZF + AD) are equiconsistent.
(A proof of the forward direction is given in [18, 6.1]; a proof of the reverse
direction is given in [9, 6.2].) He also proved some (rather technical) level-by-level
consistency results; but there are still some open problems in this area.
Finally, we consider Mazur’s game. Proving that it is true (as opposed to merely
consistent) that for all projective E ⊂ R, Mazur’s game is determined, that is, every
projective set has the property of Baire, requires axioms just as strong as those
needed to prove that for every projective E ⊂ 2ω , G2 (E) is determined. But in
contrast to the games of Ulam and Banach, the determinacy of Mazur’s game has
very low consistency strength. Solovay [16] proved that assuming the consistency
of an inaccessible cardinal – which is the weakest large cardinal axiom – the theory
(ZF + every set has the property of Baire) is consistent. At the time Mycielski’s
commentary was written it was open whether the above theory was equiconsistent
with ZFC. Shelah [15] later proved that it is.
1. U. Abraham and S. Shelah, Coding with ladders a well ordering of the reals, Journal of
Symbolic Logic 67 (2002), 579–597.
2. H. Becker, Determinacy of Banach games, Journal of Symbolic Logic 50 (1985), 110–122.
3. M. Foreman and A. Kanamori (Eds.), Handbook of Set Theory, Springer, 2010.
4. C. Freiling, Banach games, Journal of Symbolic Logic 49 (1984), 343–375.
5. T. Jech, Set Theory, third edition, Springer, 2003.
6. A. Kanamori, The Higher Infinite, second edition, Springer, 2003.
7. A. Kanamori, Introduction, in [3], 1–92.
8. A. S. Kechris, Classical Descriptive Set Theory, Springer, 1995.
9. P. Koellner and W. H. Woodin, Large cardinals from determinacy, in [3], 1951–2119.
10. P. B. Larson, A brief history of determinacy, in: A. S. Kechris, B. Löwe and J. R. Steel (Eds.),
Large Cardinals, Determinacy and Other Topics: the Cabal Seminar, volume IV, to appear.
11. P. Maddy, Believing the axioms, I and II, Journal of Symbolic Logic 53 (1988), 481-511 and
736–764.
12. D. A. Martin and J. R. Steel, A proof of projective determinacy, Journal of the American
Mathematical Society 2 (1989), 71–125.
13. Y. N. Moschovakis, Descriptive Set Theory, second edition, American Mathematical Society,
2009.
14. I. Neeman, Determinacy in L(R), in [3], 1877–1950.
15. S. Shelah, Can you take Solovay’s inaccessible away?, Israel Journal of Mathematics 48
(1984), 1–47.
16. R. M. Solovay, A model of set theory in which every set of reals is Lebesgue measurable,
Annals of Mathematics 92 (1970), 1–56.
17. R. M. Solovay, The independence of DC from AD, in: A. S. Kechris and Y. N. Moschovakis
(Eds.), Cabal Seminar 76-77, Springer, 1978.
18. J. R. Steel, The derived model theorem, http://www.math.berkeley.edu/$\thicksim$steel/
papers/dm.ps
Howard Becker
From l to r: Stan Ulam, John Oxtoby, and Tony Martin
PROBLEM 44: H. STEINHAUS

A continuous function z = f (x, y) represents a surface such that through every
point of it there exist two straight lines contained completely in the surface. Prove
that the surface is then a hyperbolic paraboloid. Do the same without assuming
continuity of f .
Addendum. This problem was solved affirmatively by Banach—also without
assuming continuity. The proof is based upon the remark: Any two straight lines
on this surface either intersect or else their projection on the plane xy are parallel.
July 30, 1935

PROBLEM 45: BANACH
Let G be a metric group which is complete and non-Abelian; U1 (x), U2 (x), . . . ,
Un (x) multiplicative operations defined in G and with values belonging to G. Prove
that if the operation U(x) = U1 (x)U2 (x) · · · Un (x) is of a Baire class, then it must be
continuous. This statement is true for n = 2.
Commentary
It is well known that if U is a homomorphism of G into G which has the Baire

property, then U is continuous, so the statement is true for n = 1. I have been unable
to locate a proof for n = 2.
R. Daniel Mauldin
PROBLEM 46: BANACH

Is the sphere in a space of type (B) unicoherent? (That is to say, in every
decomposition of it into continua A, B, is the intersection AB connected?)
Addendum. An affirmative answer to Prof. Banach’s problem follows from the
following theorem of Borsuk: In every space which is connected, locally connected,
complete and unicoherent, there exists a simple closed curve which is a retract.
In general linear spaces, in which the multiplication is continuous, an affirmative
answer to Prof. Banach’s problem follows from my theorem in Fund. Math., Vol. 26,
p. 61.
S. Eilenberg
PROBLEM 47: BANACH

Can every permutation of a matrix {Aik } i, k = 1, 2, . . . , ∞ be obtained by
composing a finite number of permutations in such a way that the rows go over
into rows and columns into columns? (Vide Problem 20: Ulam)
Commentary
This problem was solved by M. Nosarzewska in 1951—the answer is yes. For

further results and references, see E. Grzegorek, On axial maps of direct products
II, Coll. Math. 34 (1976), 145–164.
J. Mycielski
Let E be a set of real numbers which is countable, closed, and bounded. W is
the set of all continuous real-valued functions defined on E. Is the space W [if we
define the norm of a function f ∈ W as follows: f = maxx∈E f (x)] isomorphic to
the space c of all convergent sequences?
Addendum. The answer is affirmative. (The solution is unpublished.)
Mazur
February 15, 1939
Commentary
A negative response can be gleaned from a beautiful paper of Jozef Schreier (Studia
Math. 2 (1930), 58–67). Though Schreier carries out his construction in [0, 1] he
actually shows that if the ω th derived set of the compact metric space K is nonvoid,
then there exists a uniformly bounded sequence (fn ) of continuous real-valued
functions on K which is weakly convergent to zero yet admits no subsequence
whose arithmetic means are norm convergent to zero. In particular, if one considers
the ordinal ray [0, ω ω + 1) in its order topology, the resulting compact countable
metric space imbeds homeomorphically in [0, 1] yet (because the ω th derived set
of [0, ω ω + 1) is nonvoid) C([0, ω ω + 1)) is not isomorphic to c (a Banach space is
easily seen to have the delightful property that each weakly null sequence admits a
subsequence whose arithmetic means are norm null).
Much more can be said here and perhaps the best way of indicating this was
pointed out by C. Bessaga and A. Pełczyński (Studia Math. 19 (1960), 53-62), who
classified the isomorphic types of spaces C([0, α + 1)) for ordinals α < ω1 . Their
result: if α < β < ω1 , then C([0, α + 1)) is isomorphic to C([0, β + 1)) if and only if
β < α ω . It follows that the isomorphic types of spaces W as described in Problem 48
are uncountable in number.
Joseph Diestel
Kent, Ohio
Schreier showed that inside of [0, 1], you can find a countable set K such that in
C(K), there is a sequence which while uniformly bounded and pointwise convergent
to 0 has the property that the arithmetic means of any subsequence are at least 1/2
in the sup norm. In fact, a close analysis of Schreier’s construction uncovers the fact
that if K is any compact metric space whose ω th derived set is non-void, then there is
a sequence in C(K) which is uniformly bounded, converges pointwise to 0 but none
of whose subsequences have norm convergent arithmetic means. Since any weakly
null sequence in C(K) has a subsequence whose arithmetic means are norm null,
the answer to problem 48 is ’no’. A beautiful exposition on extension of Schreier’s
result can be found in the paper of N. Farnum (Canad. J.Math. 26(1974), 91–97).
Joseph Diestel
Kent, Ohio

Does there exist a space E of type (B) with the property (W) which is universal
for all spaces of type (B) with the property (W)? One should investigate this question
for the following properties (W):
(1) the space is separable and weakly compact (that is, from every bounded
sequence one can select a subsequence weakly convergent to an element).
(2) The space contains a base (countable).
(3) The adjoint space is separable.
The space E is universal isometrically (or isomorphically) for spaces of a given
class K if every space of this class is isometric (or isomorphic) to a linear subspace
of the space E.
Commentary
This problem was solved negatively for property (1) in 1967 by W. Szlenk of
Warsaw (Studia Math. 30 (1968), 53–61), who showed that if X is any separable
Banach space that contains isomorphs of all separable reflexive Banach spaces, then
X ∗ is nonseparable. Szlenk’s solution makes heavy use of a mode of derivation
the idea for which comes from the work of Z. Zalcwasser (Studia Math. 2
(1930), 63–67) and of D.C. Gillespie and W.A. Hurewicz (Trans. AMS, 32 (1930),
527–543).
More recently, using other work of W.A. Hurewicz (Fundamenta Math. 15
(1930), 4–17), Jean Bourgain (Proc. Amer. Math. Soc. 79 (1980), 241–246) proved
the theorem that if a separable Banach space B is universal for all separable
reflexive Banach spaces, then B is universal for all separable Banach spaces! We
also refer the reader to Haskell Rosenthal’s presentations of the result (entitled
“On applications of the boundedness principle to Banach space theory according
to J. Bourgain,” in Expose 5, Publ. Math. Univ. Pierre et Marie Curie, 29, Univ.
Paris VI, 1979.
(2) was solved by S. Banach and S. Mazur (cf. Banach’s book Theorie des
operations lineaires), who showed C[0, 1] is universal for all separable Banach
spaces. C[0, 1] has a (in fact, the original) Schauder basis.
The problem was solved negatively for (3) by P. Wojtaszczyk (Studia Math.
37 (1970), 197–202). Modifying the method of Szlenk, he showed that if X is
a separable Banach space for which every separable reflexive Banach space is
isomorphic to a subspace of X, then X is not isomorphic to a dual space. Jean
Bourgain noticed (Bull Soc. Math. Belg. Ser. B 31 (1979), 87–117) that if X is a
Banach space that contains an isomorph of C(K) for every countable compact subset
of [0, 1], then X contains an isomorph of every separable Banach space. Since C(K)∗
is isomorphic to 1 whenever K is a compact countable subset of [0, 1], this gives a
substantial improvement to Wojtaszczyk’s results.
Joseph Diestel
Kent, Ohio
With regard to (2), A. Pełczyński (Studia Math. 32 (1969), 247-268) showed there
exists a Banach space U with a (Schauder) basis such that every Banach space with
such a basis is isomorphic to a complemented subspace of U; in the same vein
M. I. Kadets (Studia Math. 40 (1971), 85-89) established the existence of a
separable Banach space V with the bounded approximation property such that every
separable Banach space with the bounded approximation property is isomorphic
to a complemented subspace of V. Using the so-called Pełczyński decomposition
technique, it can be shown that U and V are isomorphic. Very recently, calling on
categorical idea developed by R. Fraisse (Publ. Sci. Univ. Alger. 1 (1954) 35-182), J.
Garbulinska-Wegrazyn (Banach J. Math. Anal. 8 (2014), 211–220) has shown how
to construct an isometric version of the Kadets- Pełczyński space.
Those interested in (3) and the mentioned work of Bourgain would be well
advised to read P. Dodos (J. Functional Anal. 260 (2011), 1285–1303).
Joseph Diestel
Kent, Ohio
PROBLEM 50: BANACH
Prove that the integral of Denjoy is not [Editor note: The word not was left out
in the Los Alamos edition] a Baire functional in the space S (that is to say, in the
space of measurable functions).

The formulation of this problem is rather vague. However it was speculated that the
results in the paper [1] could be construed as a solution to this problem; see, e.g., the
review of H. Becker, MR0934228 (89g:03067) of [2] (which contains a summary of
the results in [1]).
We present below a plausible precise interpretation of Problem 50 and explain
how the results in the above paper provide a positive solution.
Denote by S the set of (Lebesgue) measurable functions f : [0, 1] → R (we use

here the interval [0, 1] but of course everything below works for any interval [a, b]).
We let for f , g ∈ S , f ∼ g ⇐⇒ f = g, a.e.. Let S = S / ∼ be the space of measurable
functions (modulo equality a.e.). This is a topological vector space with the topology
induced by the invariant (under translation), complete, separable metric
1 |f (x) − g(x)|
d([f ]∼ , [g]∼ ) = dx.
0 1 + |f (x) − g(x)|
We refer to Chapter 7 of the book [3], for a detailed introduction to the Denjoy
integral. We will only need the following facts.
The Denjoy integral is defined on a subset DI of S ∫ (the set of Denjoy
integrable functions) such that f ∈ DI , g ∼ f =⇒ g ∈ DI . Let DI = {[f ]∼ : f ∈
DI }. For f ∈ DI the Denjoy integral of f is a continuous function F on [0, 1],
uniquely determined up to a constant, and we denote by I (f ) = F(1) − F(0) the
corresponding definite integral. Moreover f ∼ g =⇒ I (f ) = I (g), so I descends
to a unique function I : DI → R, given by I([f ]∼ ) = I (f ). The crucial property of
the Denjoy integral is now the following: If F : [0, 1] → R is a differentiable function
with F = f , then f ∈ DI and I (f ) = F(1) − F(0), i.e, the Denjoy integral recovers
the primitive of any derivative.
We can now formulate a precise version of Banach’s problem: Prove that the
function I : DI → R is not in the Baire class of functions (from the separable
metrizable space DI into R), i.e., the smallest class of functions containing
the continuous functions and closed under limits of pointwise convergent sequences
of functions. Equivalently this means that I is not a Borel function (i.e., the preimage
of some open set is not Borel in DI).
Under this interpretation, it is a corollary of the results in [1] that this is true. This
can be seen as follows.
Let C = C([0, 1]) be the Banach space of continuous functions on [0, 1] and
consider the infinite product space CN , a Polish space. Let CN be the subset of
CN consisting of all pointwise convergent sequences of continuous functions and
for f̄ = (fn ) ∈ CN, let lim f̄ be its pointwise limit, which is clearly in S . It is easy to
check that the function L(f̄ ) = [lim f̄ ]∼ from CN into S is Borel. Let now D be the
subset of CN consisting of all sequences f̄ such that lim f̄ is a derivative (of some
differentiable function on [0, 1]). Then LD = L|D is a Borel function from D into DI.
Thus if I was a Borel function, so would be the composition I ◦ LD from D to R. In
particular, the set of all f̄ ∈ D such that I (lim f̄ ) > 0 would be a Borel subset of D,
contradicting Theorem 4 of [1].
On the other hand, it can be shown that the set DI is coanalytic in S and that the
Denjoy integral I : DI → R is Δ11 -measurable on DI, i.e., the preimage of any open
set is both analytic and coanalytic in DI. This is due to Ajtai (unpublished). A proof
can be also given using the techniques in [1].
References
[1] R. Dougherty and A.S. Kechris, The complexity of antidifferentiation, Advances in Mathemat-
ics, (88)(2) (1991), 145–169.
[2] A.S. Kechris, The complexity of antidifferentiation, Denjoy totalization, and hyperarithmetic
reals, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif.,
1986), 307–313, Amer. Math. Soc., Providence, RI, 1987.
[3] R.A. Gordon, The Integrals of Lebesgue, Denjoy, Perron, and Henstock, Graduate Studies in
Mathematics, Volume 4, Amer. Math. Soc., 1994.
A. S. Kechris
PROBLEM 51: MAZUR

Is a set of functions, measurable in [0, 1] with the property that every two
functions of the set are orthogonal, at most countable? (I do not assume that the
functions are square-integrable!)
(b) An analogous question for sequences: Is the set of sequences with the
property that any two sequences {εn }, {ηn } of this set are orthogonal, that is
∞
∑ εn ηn = 0,
n=1
at most countable?
Addendum. Solved by Mazurkiewicz.
PROBLEM 52: BANACH

Show that the class of functions which are continuous and defined in the interval
[0, 1] and which have everywhere a derivative, does not form a Borel set in the space
C of all continuous functions in (0, 1). One can show that it is not a set Fσ and also
it is the complement of an analytic set.
Addendum. Solved by Mazurkiewicz.
Commentary
Mazurkiewicz proved that this set forms a coanalytic subset of C which is not a
Borel set [3]. It is also true that this set is of the first category (meager) [6, p. 45].
Mazurkiewicz also showed that the set of all continuous functions f on the unit
square for which there is some y so that ∂ f /∂ x exists at (x, y) for all x in [0, 1] forms
a PCA set which is not a CPCA set [4].
It has been shown that the set of all continuous nowhere differentiable functions
forms a coanalytic subset of C which is not a Borel set [2]. It can be shown that
the set of functions of Besicovitch also forms a coanalytic subset of C [7] and is
nonempty [5]. Of course, almost every path in Brownian motion is a continuous
nowhere differentiable function [1].
1. A. Dvoretzky, P. Erdős, and S. Kakutani, Nonincrease everywhere of the Brownian motion

process, Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and
Probability, University of California Press (1961), 103–116.
2. R.D. Mauldin, The set of continuous nowhere differentiable functions, Pac. J. Math. 80 (1979),
199–205. Correction, Pac.J. Math. 121 (1986), 110-111.
3. S. Mazurkiewicz, Über die Menge die differenzierbaren Funktionen, Fund. Math. 27 (1936),
244–249.
4. S. Mazurkiewicz, Eine projektive Menge der Klasse PCA in Funktionalraum, Fund. Math. 28
(1937), 7–10.
5. A.P. Morse, A continuous function with no unilateral derivatives, Trans. Amer. Math. Soc., 44
(1938), 496–507.
6. J.C. Oxtoby, Measure and Category, Springer-Verlag, New York, 1970.
7. S. Saks, On the functions of Besicovitch in the space of continuous functions, Fund. Math. 12
(1928), 244–253.
R. Daniel Mauldin
PROBLEM 53: BANACH

A surface element C (i.e., a one-to-one continuous image of a disc) has the
following property: For every ε > 0 one can find η > 0 such that any two points
of C with a distance less than η can be connected by an arc contained in C with a
length less than ε . Show that C has a finite area and almost everywhere a tangent
plane.
Addendum. There exists a surface element C of the form z = f (x, y), 0 ≤ x, y ≤ 1,
satisfying the above conditions but without possessing a finite area.
Mazur
August 1, 1935

A convex, closed, compact set H is transformed by a continuous mapping U(x)
on a part of itself. H is contained in a space of type (F). Does there exist a fixed
point of the transformation?
(b) Solve the same problem for arbitrary linear topological spaces or such
spaces in which there exist arbitrarily small convex neighborhoods.
[A solution exists for spaces of type (F0 ); in the more general theorem, H need
not be compact; only U(H) is assumed compact.]
Remark
This problem has led to an incredible number of fixed point theorems. This topic is
discussed in Andrzej Granas’ lecture published in this edition of the Scottish Book
(pp. 45–61)—Problem 54 is discussed in the last section of his talk.
The second and third parts of the problem have a positive solution; the first part
of Problem 54 is still unsolved.
A. Tychonoff [Ein Fixpunktsatz, Math. Ann. 111 (1935), 767–776] provided a

proof of the fixed point theorem for locally convex spaces. Versions of Schauder’s
fixed point theorem were proved by Glicksberg, Fan, Krasnoselskii, Schaefer, and
many other authors. In 2001, Robert Cauty [Solution du problème de point fixe
de Schauder, Fund. Math. 170 (2001) 231–246] considered Schauder’s problem
in the general case; however, later it was discovered that his proof (as well as its
elaboration in [T.Dobrowolski, Revisiting Cauty’s proof of the Schauder conjecture,
Abstr. Appl. Anal. 2003, no. 7 , 407–433]) contained a gap.
In 2005, R. Cauty [Rétractes absolus de voisinage algébriques, Serdica Math.
J. 31 (2005), no. 4, 309–354] introduced the notion of algebraic ANR. He applied
the theory of algebraic ANRs to the fixed point theorem for compact upper semi-
continuous multivalued maps with acyclic compact point images. In particular, this
gave an affirmative solution of Schauder’s problem.
M. Zarichnyi
PROBLEM 55: MAZUR
There is given, in an n-dimensional space E or, more generally, in a space of type
(B), a polynomial W(x) bounded in an ε -neighborhood of a certain nonbounded set
R ⊂ E (an ε -neighborhood of a set R is the set of all points which are distant by less
than ε from R). Does there exist a polynomial V(x) and a polynomial of first degree
φ (x) such that
(1) W(x) = V(φ (x));
(2) The set φ (R), that is to say the image of the set R under the mapping φ (x), is
bounded?
Addendum. In the case of Euclidean spaces, a solution for a finite system of
polynomials:
There exists a linear substitution with determinant = 0 under which all the
polynomials of the given system go over into polynomials depending on a smaller
number of the given variables. (Studia Math. 5.) [See also Problem 75.]
Auerbach
August 3, 1935

In a space E of type (B) there is given a functional F(x) of degree m and
discontinuous. “F is of degree m” means that for x0 , h0 ∈ E there exist numbers
a0 , . . . , am such that F(x0 + th0 ) = a0 + ta1 + . . . + tm am for rational t. Do there then
exist points xn ∈ E such that xn → 0 and |F(xn + x)| → +∞ or even only
lim |F(xn + x)| = +∞

n→∞
for all x ∈ E? Not solved even for finite-dimensional spaces E.
Commentary

1. S. Mazur and W. Orlicz, Sur la divisibilité des polynomes abstraits, C. R. Acad. Sci. Paris, 202
(1936), 621–623.
904–905.
PROBLEM 57: RUZIEWICZ

Given are two functions w(h) and φ (h), decreasing with |h| to 0, and satisfying
the conditions
w(h)
lim = ∞,
h→0 |h|
and
w(h)
lim = ∞.
h→0 φ (h)
Does there exist a function satisfying the conditions:
|f (x + h) − f (x)| < w(h) (1)
and

f (x + h) − f (x)
lim sup = ∞?
(2)
h→0 φ (h)
We shall prove that the answer to Stanisław Ruziewicz’s question is affirmative. We

denote by x the distance from x to the nearest integer.
Lemma 1. For every x ∈ R we have
max(|x + (1/3) − x|, |x + (2/3) − x|) ≥ 1/6 (3)
and
max(|x − (1/3) − x|, |x − (2/3) − x|) ≥ 1/6. (4)
Proof. We may assume x ∈ [0, 1]. Suppose |x + (1/3) − x| < 1/6 and |x +
(2/3) − x| < 1/6. If x ∈ [0, 1/2], then these inequalities imply x + (1/3), x +
(2/3) ∈ [1/2, 1]. From these we infer |(2/3) − 2x| < 1/6 and |(1/3) − 2x| < 1/6,
which is impossible. If x ∈ [1/2, 1], then necessarily x + (1/3), x + (2/3) ∈ [1, 3/2].
Thus we have |2x − (5/3)| < 1/6 and |2x − (4/3)| < 1/6, which is also impossible.
This proves (3).
Applying (3) with −x in place of x and using the fact that the function x is even,
we obtain (4).
Lemma 2. Suppose that v : [0, ∞) → R is continuous, increasing, concave, and
satisfies v(0) = 0 and limx→+0 v(x)/x = ∞. Then, for every sequence (cn ) of positive
numbers converging to zero there exist a function g : R → R and a subsequence
(cnk ) of (cn ) such that
(i) |g(x + h) − g(x)| ≤ 2v(|h|) for every x, h ∈ R, and
(ii)
1
max(|g(x + (cnk /3)) − g(x)|, |g(x + (2cnk /3)) − g(x)|) ≥ v(cnk )
12
and
1
max(|g(x − (cnk /3)) − g(x)|, |g(x − (2cnk /3)) − g(x)|) ≥ v(cnk )
12
for every x ∈ R and k = 1, 2, . . ..
Proof. We define the sequence (ni ) of positive integers and the sequences (λi ), (ai )
of positive numbers by induction. We put n1 = 1, λ1 = v(c1 ) and a1 = 1/c1 . Suppose
that ni , λi = v(cni ) and ai = 1/cni have been defined. Since limx→+0 v(x)/x = ∞ and
v(cn ) → 0, we can choose an index ni+1 > ni such that v(cni+1 )/cni+1 > 16v(cni )/cni
and v(cni+1 ) < v(cni )/16. We put λi+1 = v(cni+1 ) and ai+1 = 1/cni+1 . In this way we
defined the numbers ni , λi and ai for every i = 1, 2, . . ..
It is clear from the construction that λi+1 < λi /16 and ai+1 λi+1 > 16ai λi hold for
every i. Therefore, we have
∞
1 1 16
∑ i k+1
λ < λ · 1 +
16
+
16 2
+ . . . = λk+1
15
i=k+1
and

1 1
∑ λi ai < λk ak · + 2 + . . . = λk ak /15.
16 16
i<k
for every k.
We define g(x) = ∑∞
i=1 λi ai x, and prove that g satisfies the requirements. We
have
∞
g(x + h) − g(x) = ∑ λi (ai x + ai h − ai x) =
i=1 (5)
= Ak + λk (ak x + ak h − ak x) + Bk
for every x, h ∈ R and k = 1, 2, . . ., where

∞
Ak = ∑ λi (ai x + ai h − ai x) and Bk = ∑ λi (ai x + ai h − ai x).
i<k i=k+1
Since |x + h − x| ≤ |h| and |x + h − x| ≤ 1/2 for every x and h, it follows that
|Ak | ≤ ∑ λi ai |h| < λk ak |h|/15,

i<k
1 ∞ 8 1
|Bk | ≤ ∑ λi < 15 λk+1 < 30 λk
2 i=k+1
and
|g(x + h) − g(x)| ≤ |Ak | + λk ak |h| + |Bk |. (6)
If |h| > c1 , then

∞
|g(x + h) − g(x)| ≤ ∑ λi < (16/15)λ1 = (16/15)v(c1 ) ≤ (16/15)v(|h|),
i=1
and thus (i) is true in this case.

If 0 < |h| ≤ c1 , then there exists an index k such that cnk+1 ≤ |h| ≤ cnk . By (6) we
have
16 8
|g(x + h) − g(x)| ≤ λk ak |h| + λk+1 . (7)
15 15
We have λk+1 = v(cnk+1 ) ≤ v(|h|). Since v is concave and v(0) = 0, it follows that
the function v(x)/x is decreasing on (0, ∞). Therefore, by |h| ≤ cnk we have
v(cnk ) v(|h|)
≤ ;
cnk |h|
that is, λk ak |h| ≤ v(|h|). Thus, by (7), we obtain |g(x + h) − g(x)| ≤ (24/15)v(|h|),
which proves (i).
Now we prove (ii). Let x ∈ R and k ≥ 1 be fixed. By Lemma 1, there are h1 , h2 ∈
{cnk /3, 2cnk /3} such that
|ak x + ak h1 − ak x| ≥ 1/6 and |ak x − ak h2 − ak x| ≥ 1/6.
(Note that ak = 1/cnk .) Thus, by (5), we obtain
1 1 1
|g(x + (−1)j−1 hj ) − g(x)| ≥ − λk ak hj + λk − λk ≥
15 6 30
2 1 1 1 1
≥ − λk + λk − λk > λk = v(cnk )
45 6 30 12 12
for j = 1, 2, which completes the proof.
Lemma 3. Let u : [0, ∞) → R be an increasing function such that
limx→+0 u(x) = u(0) = 0 and limx→+0 u(x)/x = ∞. Then there exists a function
v : [0, ∞) → R such that limx→+0 v(x)/x = ∞, v is increasing, continuous, concave,
v ≤ 2u on [0, ∞), and v = u at the points of a sequence converging to zero.
Proof. Let x0 > 0 be an arbitrary point, and put s0 = u(x0 )/x0 . Since u is increasing
and limx→+0 u(x)/x = ∞, it follows that for every s ≥ s0 there is a smallest positive
number d(s) such that u(d(s))/d(s) = s. Indeed, it is easy to check that d(s) =
inf{x > 0 : u(x) ≤ sx} has this property. Note that u(x)/x > s for every 0 < x < d(s).
We put d0 = d(s0 ). Since limx→+0 u(x) = 0, there is a point 0 < x1 < min(d0 , 1)
such that u(x1 ) < u(d0 ). We put s1 = u(x1 )/x1 and d1 = d(s1 ). Let 1 be the
linear function whose graph connects the points (d1 , u(d1 )) and (d0 , u(d0 )).
Then 1 (0) > 0, and thus we can select a point 0 < x2 < min(d1 , 1/2) such that
u(x2 ) < 1 (0). We put s2 = u(x2 )/x2 and d2 = d(s2 ). Let 2 be the linear function
whose graph connects the points (d2 , u(d2 )) and (d1 , u(d1 )). Then 2 (0) > 0, and
thus we can select a point 0 < x3 < min(d2 , 1/3) such that u(x3 ) < 2 (0). We put
s3 = u(x3 )/x3 and d3 = d(s3 ). Continuing this process we define the points dn for
every n ≥ 0 and the linear functions n for every n ≥ 1.
It is clear that (dn ) is a strictly decreasing sequence converging to zero, and that
u(x)/x > u(dn )/dn = sn for every x ∈ [0, dn ) and n = 0, 1, . . .. Thus the sequence (sn )
is strictly increasing.
We define v(0) = 0, v(x) = u(d0 ) for every x ≥ d0 , and v(x) = n (x) for every
x ∈ [dn , dn−1 ] and n = 1, 2, . . .. It is easy to check that v is continuous, increasing,
and concave on [0, ∞), satisfies limx→+0 v(x)/x = ∞, and equals u at the points dn .
We prove that v(x) ≤ 2u(x) for every x ≥ 0. This is clear for x ≥ d0 (since u
is increasing). Let n ≥ 1 be fixed; we prove that v(x) = n (x) ≤ 2u(x) for every
x ∈ [dn , dn−1 ]. Since u is increasing and u(x) > sn−1 x for every x < dn−1 , it follows
that u(x) ≥ max(u(dn ), sn−1 x) for every x ∈ [dn , dn−1 ]. Thus it is enough to show
that
n (x) ≤ 2 · max(u(dn ), sn−1 x) (8)
if x ∈ [dn , dn−1 ]. Let a = u(dn )/sn−1 . Then a > u(dn )/sn = dn . Since the slope of n
is smaller than sn−1 , it follows that
n (a) < u(dn ) + sn−1 (a − dn ) < u(dn ) + sn−1 a = 2u(dn ).
Therefore n (x) < 2u(dn ) in the interval [dn , a]. Since n (a) < 2u(dn ) = 2sn−1 a,
n (dn−1 ) = u(dn−1 ) = sn−1 dn−1 < 2sn−1 dn−1 and n is linear, it follows that n (x) <
2sn−1 x for every a ≤ x ≤ dn−1 . This proves (8) and the lemma.
Now we turn to the solution of the problem. Let w and φ be given as in the
problem. We put w1 (h) = min(w(−h), w(h)) for every h ≥ 0. Then w1 is increasing
on [0, ∞), and satisfies w1 (0) = 0 and limx→+0 w1 (x)/x = ∞. By Lemma 3, there
exists a function v such that v is increasing, continuous, concave, v ≤ 2w1 on [0, ∞),
limx→+0 v(x)/x = ∞, and v = w1 at the points of a sequence (dn ) converging to zero.
For every n we have v(dn ) = w1 (dn ) = min(w(−dn ), w(dn )), and thus either
v(dn ) = w(dn ) or v(dn ) = w(−dn ). Therefore, taking a suitable subsequence of (dn )
we can find a sequence (cn ) of positive numbers tending to zero such that either
v(cn ) = w1 (cn ) = w(cn ) for every n, or v(cn ) = w1 (cn ) = w(−cn ) for every n.
Applying Lemma 2 we obtain a function g satisfying (i) and (ii). We prove that
f = g/5 satisfies the requirements of the problem. For every x and h = 0 we have
|f (x + h) − f (x)| ≤ 2v(|h|)/5 ≤ 4w1 (|h|)/5 ≤ 4w(h)/5 < w(h),
and thus (1) is satisfied.

If v(cn ) = w(cn ) for every n, then we use the first inequality in (ii) of Lemma 2,
and choose a sequence (hk ) such that hk ∈ {cnk /3, 2cnk /3} and
|g(x + hk ) − g(x)| ≥ v(cnk )/12 = w(cnk )/12 ≥ w(hk )/12
for every k. Then
|f (x + hk ) − f (x)| 1 w(hk )
lim ≥ · lim = ∞. (9)
k→∞ φ (hk ) 60 k→∞ φ (hk )
Therefore, in this case we have

f (x + h) − f (x)

lim sup =∞ (10)
φ (h)
h→+0
for every x.
In the case when v(cn ) = w(−cn ) for every n, we use the second inequality in (ii)
of Lemma 1, and choose a sequence (hk ) such that hk ∈ {−cnk /3, −2cnk /3} and
|g(x + hk ) − g(x)| ≥ v(cnk )/12 = w(−cnk )/12 ≥ w(hk )/12
for every k. Then we have (9) again, proving that

f (x + h) − f (x)

lim sup =∞ (11)
φ (h)
h→−0
holds for every x. This shows that f satisfies the requirements.

Remark 1. We proved slightly more than what was asked in the problem. We
showed that there exists a function f satisfying (1), and either (10) for every x or
(11) for every x. It is clear from the proof that if the function φ is even, then f
satisfies

f (x + h) − f (x)
lim sup = lim sup f (x + h) − f (x) = ∞ (12)
φ (h) φ (h)
h→+0 h→−0
for every x.
In the general case, however, we cannot expect (12) to be true. Moreover, it is
possible that whenever a function f satisfies (1), then

f (x + h) − f (x)
lim = 0.
(13)
h→+0 φ (h)
Consider the following simple example. Let 0 < α < β < γ < δ < 1 be given
numbers, and define w and φ as follows: w(h) = hα , φ (h) = hβ if h ≥ 0, and
w(h) = (−h)γ , φ (h) = (−h)δ if h < 0. It is clear that w and φ satisfy the conditions
of the problem.
Suppose (1). If h > 0, then
|f (x + h) − f (x)| = |f ((x + h) − h) − f (x + h)| ≤ w(−h) = hγ ,
and thus

f (x + h) − f (x) γ
lim sup ≤ lim sup h = 0.
φ (h) β
h→+0 h→+0 h
M. Laczkovich

A set E1 of real numbers precedes the set E2 , which we denote by E1 pE2 , if:
(1) E1 is of a lower homoie class than E2 (E1 < E2 ),
(2) There does not exist a set E3 so that E1 < E3 < E2 .
(a) Do there exist sets A, B, C, and {An }, n = 1, 2, . . . , N, (N > 1), such that
ApBpC and ApA1 pA2 p . . . pAn pC?
(Remark: For n = 2 such sets exist; cf. Fund. Math., 15, p. 95.)
(b) Do there exist sets A, B, C, and {An }, n = 1, 2, . . . ad inf. such that ApBpC
and ApA1 pA2 pA3 p . . . ad inf., and An < C for n = 1, 2, 3, . . .?
Commentary
This problem concerns dimensional types as defined by Fréchet [1, p. 30]. If X

and Y are topological spaces, then the type of X is ≤ the type of Y (symbolized
by dX ≤ dY) provided there is a homeomorphism of X into Y. The spaces X and
Y are of the same type (dX = dY) provided dX ≤ dY and dY ≤ dX. If X and Y are
homeomorphic, then dX = dY. It is easily seen that the converse is not true. The type
of X is less than the type of Y provided dX ≤ dY but there is no homeomorphism
of Y into X. This is what is meant by the expressions “E1 < E2 ” or “E1 is of lower
homoie class than E2 ” in this problem. There is a discussion of this concept and
some early results in [2].
Apparently, the problems posed here are still open.
1. M. Fréchet, Les espaces abstraits, Gauthier-Villars, Paris, 1928.
2. W. Sierpiński, General Topology, University of Toronto Press, Toronto, 1956.
R. Daniel Mauldin
Can one decompose a square into a finite number of squares all different?
Commentary
A square dissected into a finite number of squares no two the same is now referred
to as a perfect squared square or simply a perfect square. The earliest published
mention of the problem which has been found is in a paper of 1925 by Zbigniew
Moroń [11]. Moroń was concerned, however, with the dissection of a rectangle into
squares. He stated that Ruziewicz had asked if a rectangle could be made up of
different squares and presented two such rectangles in answer. A letter from Prof.
Władysław Orlicz in 1977 gives some details [9]. He and Moroń were schoolmates
studying mathematics at the University of Lwów and in about 1923–1924 both were
junior assistants in the Institute of Mathematics. Stanisław Ruziewicz, professor of
mathematics, proposed the problem (presumably in his mathematical seminar) of

dissecting a rectangle into unequal squares, which he said he had heard of from
mathematicians of the University of Cracow. The students worked diligently on the
problem without success until they were all surprised by Moroń’s solutions.
Moroń observed that if there were two different dissections of the same rectangle
such that each square of one dissection is different from each square of the other,
they can be put together with two added squares to form a square dissected into
squares which are all different provided that neither of the two dissected rectangles
contained a square equal to one of the two added squares (two examples are given
in Figures 59.1 and 59.2).
It may be that Ruziewicz also mentioned dissecting a square at that time; in any
event, he did so later. The problem was believed by some to be impossible. Kraitchik
in 1930 stated that the Russian mathematician N.N. Lusin had communicated to him
the proposition (believed to be true though not demonstrated) that it was not possible
to dissect a square into a finite number of different squares [10].
Figure 59.1
Figure 59.2
Moroń did not state how he obtained his two dissected rectangles; obviously,
they must have been found empirically. Some others were found over a dozen years
later by Sprague. He succeeded in making up two different dissected rectangles
the same size and forming a perfect square according to Moroń’s observation,
published in 1939 [12]. The square has 55 component squares and is 4205 units
in size; its structure is shown in Figure 59.1. It contains five disjoint subsets of
squares arranged into five rectangles, two pairs of which are Moroń’s two rectangles
magnified different amounts.
A second perfect square was published, by listing the sides of the component
squares, only a few months later [13]. This is shown in Figure 59.2; it has 28
elements (the component squares) and is 1015 units in size. This square was the
result of the work of four students at Trinity College, Cambridge, R.L. Brooks,
C.A.B. Smith, A.H. Stone, and W.T. Tutte, who had been working on the problem
of squaring rectangles and squares during the years 1936–1938. Their paper [4] is a
classic; see also [14] for a very interesting expository account of their work.
A distinction is made between a squared rectangle (square) which contains a
subset of elements, more than one and less than all, which are themselves arranged
into a rectangle, and one which does not; the former are designated compound and
the latter simple. The Cambridge group founded and developed the theory of simple
squared rectangles. They proved that every simple squared rectangle with n elements
can be produced from the complete set of 3-connected planar graphs of order n + 1.
This is done by considering the graph to be an electrical network with an emf placed
in one branch and the other branches having unit resistance. The relative values of
the currents in these branches are calculated from Kirchhoff’s laws. These values
represent the relative values of the sides of the component squares of a dissection
and the connectivities of the branches of the network represent the manner in
which the component squares are arranged. They also produced a number of perfect
squares, including several departing from Moroń’s suggested type, and developed
some theory concerning them. The treatment utilizes electrical network theory and
graph theory, and makes contributions in each of these fields.
Bouwkamp described the method of [4] in more physical terms in 1946–1947 [1]
and listed the 3-connected planar graphs with up to 14 edges (by drawings) and the
simple squared rectangles with up to 13 elements (by giving their elements in a
certain order). This reached the practical limit of what could be done systematically
and completely by hand. Later, in 1960, complete sets of 3-connected planar graphs
with up to 19 edges were produced by computer [3], and from these complete sets of
simple squared rectangles, also by computer. A catalogue of the rectangles with up
to 15 elements was published [2]. The methods used are described in Duijvestijn’s
1962 thesis [5].
One relevancy of simple squared rectangles to Problem 59 is that compound
perfect squares can be produced from them. A summary of various methods of
combining squared rectangles was given in [8], which also introduced another
method of producing compound perfect squares of the type shown in Figure 59.3,
by means of which many compound perfect squares of order higher than 24 were
produced. In 1948 Willcocks had constructed the compound perfect square with
only 24 elements [15, 16] which is shown in Figure 59.3. It remained the lowest
order perfect square until 1978 and is the lowest order compound perfect square.
This was shown by a computer search by Duijvestijn and Leeuw based on a method
proposed by Federico [17]. Over two thousand perfect squares of order higher than
24 have been produced.
On the other hand, the production of simple perfect squares is still not possible by
any general direct method. The Cambridge group in [4] and later papers developed
a theoretical method by means of which some simple perfect squares of a special
type but of a high order were produced. Simple perfect squares of another special
type, of orders 25 and 26, were produced by Wilson in 1967 [17].
The other relevancy of the systematic production of simple perfect rectangles is
that if carried out to a sufficiently high order one with equal sides might be found,
a simple perfect square. This was done up to order 19 by Duijvestijn in his 1962
Figure 59.3
thesis [5]. He found that there were no simple perfect squares below order 20. The
work of testing for order 19 squares, for example, required the construction of the
complete set of nonisomorphic 3-connected planar graphs with 20 edges. However,
further work on some of them was eliminated a priori by virtue of certain theorems
in the basic paper [4]. Even so, the computer time was high. The construction
of the graphs necessarily results in a considerable amount of duplication and to
eliminate the duplicates and easily calculable numerical identifying characteristic
of graphs, invariant under isomorphism, was developed. An incidental result of
the construction of complete sets of 3-connected planar graphs was an advance in
Euler’s problem of the enumeration of convex polyhedra, since these graphs are
isomorphic with the graphs of convex polyhedra.
Duijvestijn returned to the problem in recent years. With improvements in details
of the methods and the availability of a faster computer he was able to complete the
search through order 21. In 1978 he announced that there was one, and only one,
Figure 59.4
simple perfect square of order 21 and none of lower order [6]. The square is shown
in Figure 59.4. Scientific American called this a pluperfect square, which term is
deserved not only because it is the lowest order perfect square possible but also on
account of the elegance of its construction.
An extended historical review of this subject is available in [9].
Acknowledgements
I am indebted to Michael Goldberg for making the drawings for Figures 59.1
and 59.4 and to P. Leeuw for drawing Figures 59.2 and 59.3 by computer.
1. C.J. Bouwkamp, On the dissection of rectangles into squares. I., Nederl. Akad. Wetensch.
Proc., AA49 (1946), 1176–1188; II, III, A50 (1947), 58-71, 72–78.
2. C.J. Bouwkamp, A.J.W. Duijvestijn, and P. Medema, Tables Relating to Simple Squared
Rectangles of Orders Nine through Fifteen, Technische Hogeschool, Eindhoven, Netherlands,
Aug. 1960, 360 pages.
3. C.J. Bouwkamp, A.J.W. Duijvestijn, and P. Medema, Tables of c-nets of Orders 8–19
inclusive, Phillips Research Laboratories, Eindhoven, Netherlands, 2 vols. 1960; unpublished,
available in UMT file of Mathematics Computation; see Math. Comp., 24 (1970), 995-997 for
description. (The basic paper [4] and others refer to 3-connected planar graphs as c-nets.)
4. R.L. Brooks, C.A.B. Smith, A.H. Stone, and W.T. Tutte, The dissection of rectangles into
squares, Duke Math. J. 7 (1940), 312–340.
5. A.J.W. Duijvestijn, Electronic Computation of Squared Rectangles, Dissertation, Technische
Hogeschool, Eindhoven, Netherlands, 1962, 96 pages; also in Phillips Res. Rep. 17 (1962),
523–612.
6. A.J.W. Duijvestijn, Simple perfect squared square of lowest order, J. Comb. Theory, B25
(1978), 240-243. See also Sci. Amer., 238 (June 1978), 86–88
7. A.J.W. Duijvestijn, P.J. Federico, and P. Leeuw, Compound Perfect Squares, Amer. Math.
Monthly 89 (1982), 15-32.
8. P.J. Federico, Note on some low-order perfect squared squares, Canad. J. Math. 15 (1963),
350–362.
9. P.J. Federico, Squaring Rectangles and Squares; A Historical Review with Annotated Bibli-
ography, in Graph Theory and Related Topics, J.A. Bondy and U.S.R. Murty, Eds., Academic
Press, 1979, 173–196.
10. M. Kraitchik, La Mathématique des Jeux ou Récréations Mathématiques, Stevens Frères,
Brussels, 1930, 272.
11. Z. Moroń, O rozkladach prostokatow na kwadraty (On the dissection of a rectangle into
squares), Przeglad Mat. Fiz., 3 (1925), 152-153.
12. R. Sprague, Beispiel einer Zerlegung des Quadrats in lauter verschiedene Quadrate, Math.
Zeit., 45 (1939), 607–608.
13. A.H. Stone, Problem E401, Amer. Math. Monthly, 47 (Jan. 1940), 48.
14. W.T. Tutte, Squaring the square, Sci. Am., 199 (Nov. 1958), 136–142, 166. Reprinted with
addendum and enlarged bibliography in Martin Gardner, The 2nd Scientific American Book of
Mathematical Puzzles and Diversions, pp. 186–209, 250, Simon & Schuster, New York, 1961;
also paperback edition.
15. T.H. Willcocks, Problem 7795 and solution, Fairy Chess Rev. 7 (1948), 97 (Aug.), 106 (Oct.).
16. T.H. Willcocks, A note on some perfect squares, Canad. J. Math., 3 (1951), 304–308.
17. John C. Wilson, A method for finding simple perfect squared squarings, Ph.D. Thesis,
University of Waterloo, Waterloo, Ontario, Canada, 1967, 80 pages plus 72 pages computer
output.
P. J. Federico

Can one, for every ε > 0, represent the surface of a sphere as a sum of a
finite number of regions which are smaller in diameter than ε , closed, connected,
congruent, and have no interior point in common? We assume that the boundaries
of these sets are: (a) polygons, (b) curves of finite length, (c) sets of measure zero.
PROBLEM 61: STEINHAUS

(a) Determine the surfaces z = f (x, y) such that in each of their points there intersect
two plane curves congruent to each other.
(b) Determine the surface z = f (x, y) such that in each point there intersect two
plane curves congruent to one of them (for every point the same curve).
(cf. Problem 44)
Addendum. All surfaces of revolution have this property; whether these are the only
ones is not known.
Ruziewicz
July 31, 1935
PROBLEM 62: MAZUR, ULAM

In a group G there are given groups Gn , n = 1, 2, . . . ad inf. with the following
properties: G = G1 + G2 + . . . + Gn + . . ., Gn ⊂ Gn+1 , Gn is isomorphic to G1 . Is G
isomorphic to G1 ?
Addendum. As R. Baer remarked, the answer is trivial: Gn is the group of
numbers with the denominator n, G = ∑ Gn = the group of rational numbers.

The set E of elements of a group G we call a base if E spans a group which is
identical with G, but no proper subset of the set E has this property. If there is a base
in a group G, does there exist a base for every subgroup H of it?
Commentary
The answer is no. Several examples of groups with a minimal set of generators
having subgroups without minimal sets of generators were given by V. Dlab. (On a
problem of Mazur and Ulam about irreducible generating systems in groups, Coll.
Math. 7 (1959), 171–176).
Jan Mycielski
PROBLEM 64: MAZUR

In a space E of type (B) there are given two convex bodies A and B and their
distance from each other is positive. (A convex body is a convex set which is closed,
bounded, and possesses interior points.) Does there exist a hyperplane H which
separates the two bodies, A, B? That is to say, a hyperplane which has the property
that one of the bodies lies on one, the other on the other side of this hyperplane.
[Hyperplane means a set of all points x satisfying the equation F(x) − c = 0, where
F(x) is a linear functional = 0, and c a constant.]
Addendum The theorem is true even when the two bodies are not disjoint, but do
not have common interior points.
Eidelheit
January 11, 1936
Commentary
The result of Eidelheit is, of course, now well known. The theorem stated here
was published in M. Eidelheit, “Zur Theorie der konvexen Mengen in linearen
normierten Räumen,” Studia Math. 6 (1936), 104–111. See also S. Kakutani, “Ein
Beweis des Satzes von Eidelheit über konvexe Mengen,” Proc. Imp. Acad. Japan,
13 (1937), 93–94. I do not know if there now is a stronger theorem for normed
linear spaces. For the strongest form of the separation theorem in finite-dimensional
Euclidean space, see F.A. Valentine, Convex Sets, McGraw-Hill, New York (1964),
p. 66.
W.A. Beyer
PROBLEM 65: MAZUR

In a space E of type (B) there is given a convex set W, containing 0 and nowhere
dense. Is the smallest convex set containing W, symmetric with respect to 0 (that
is to say, the set generated by elements x − y, where x ∈ W, y ∈ W), also nowhere
dense?
Addendum. False—the set W composed of functions which are nondecreasing, in
the space (C) of all continuous functions is convex and nowhere dense. It contains
0. The convex set containing W and symmetric with respect to zero contains all
functions of bounded variation and is not nowhere dense.
Mazur
PROBLEM 66: MAZUR

The real-valued function z = f (x, y) of real variables x, y possesses the first partial
derivatives ∂ f /∂ x, ∂ f /∂ y and the second partial derivatives ∂ 2 f /∂ x2 , ∂ 2 f /∂ y2 . Do
there exist then almost everywhere the mixed second partial derivatives ∂ 2 f /∂ x∂ y,
∂ 2 f /∂ y∂ x. According to a remark by Prof. Schauder, this theorem is true with
the following additional assumptions: The derivatives ∂ f /∂ x, ∂ f /∂ y are absolutely
continuous in the sense of Tonelli, and the derivatives ∂ 2 f /∂ x2 , ∂ 2 f /∂ y2 are square
integrable. An analogous question for n variables.
Remark
This problem was just settled in the negative by V. Mykhaylyuk and A. Plichko
in their paper On a Mazur problem from the "‘Scottish Book" concerning second
partial derivatives. The paper will appear in Colloquium Mathematicum. April
2015.
PROBLEM 67: BANACH

August 1, 1935
A modification of Mazur’s game [See Problem 43].)
We call a half of the set E [in symbols, (1/2)E] an arbitrary subset H ⊂ E such
that the sets E, H, E − H are of equal power.
(1) Two players A and B give in turn set Ei , i = 1, 2, . . . ad inf. so that Ei = (1/2)Ei−1 ,
i = 1, 2, . . ., where E0 is a given abstract set. Player A wins if the product
E1 E2 · · · Ei Ei+1 · · · is vacuous.
(2) The game, similar to the one above, with the assumption that Ei = 1/2[E0 −
E1 − . . . − Ei−1 ], i = 2, 3, . . . ad inf., and E1 = (1/2)E0 . Player A wins if E1 +
E2 + . . . = E0 .
Is there a method of win for player A? If E0 is of power cofinal with ℵ0 , then

player A has a method of win. Is it only in this case? In particular, solve the problem
if E0 is the set of real numbers.
Addendum. There exists a method of play which will guarantee that the product of
the sets is vacuous. The solution was given by J. Schreier.
August 24, 1935
Commentary
The solution of J. Schreier was published in [2].

The strategies for both games are easy to describe: One well orders E with a
relation <. Then for game (1), given any X ⊂ E with |X| = |E| one always chooses
the subset of all those elements of X which have an immediate predecessor in X
relative to <. Then, after ω steps the intersections is empty. For game (2) one
chooses the subset of all elements of X which have no immediate predecessors in X.
After ω steps E is covered.
Many related games were studied in [1], e.g., the following game: A set S is
given. Player I cuts S into two pieces, Player II chooses one of them, then again I
cuts the chosen piece and II chooses one of them, etc., for ω steps. Player II wins if
the intersection of the chosen parts is empty and Player I wins otherwise. Then it is
proved that I has a winning strategy iff |S| ≥ 2ℵ0 and II has a winning strategy iff
|S| ≤ ℵ0 . For more material on games, see the commentary to Problem 43.
1. F. Galvin, J. Mycielski, R.M. Solovay, and others, a long paper in preparation.
2. J. Schreier, Eine Eigenschaft abstrakter Mengen, Studia Math. 7 (1938), 155–156.
J. Mycielski
PROBLEM 68: ULAM

There is given an n-dimensional manifold R with the property that every section
of its boundary by a hyperplane of n − 1 dimensions gives an n − 2dimensional
closed surface (a set homeomorphic to a surface of the sphere of this dimension).
Prove that R is a convex set. This question was settled affirmatively for n = 3 by
Schreier. (That is to say, a manifold contained in E3 , such that every section by a
plane gives a single simple closed curve, must be convex.)
Commentary
This problem remains unsolved. Of course, one could restate the problem to take
into account the possible lower dimensional intersections of the manifold with an
(n − 1)-dimensional hyperplane.
Schreier [3] showed that each 2-dimensional surface in R3 , each of whose
nondegenerate planar sections is a Jordan curve, is convex. Aumann [1] proved that
a continuum K in R3 is convex provided that for each plane P, P ∩ K and P − K is
connected. Aumann also showed that a continuum K in Rn is convex if and only
if Rn − K is connected and the intersection of K with each (n − 1)-dimensional
hyperplane is convex.
Aumann [2] later claimed the following theorem. A closed bounded subset K
of Rn is convex if and only if K ∩ P is simply connected for all 2-dimensional
hyperplanes P.
It is rather interesting that a problem formulated and partially solved by Schreier
and Ulam should have been included in the first volume of Deutsche Mathematik,
now available in an expurgated edition.
1. G.G. Aumann, Eine einfache Characterisierung der convexen Kontinuen im R3 , Deutsche
Mathematik 1 (1936), 108.
2. G. Aumann, Über Schnitteigenschaften convexer Punktmengen im R3 , Deutsche Mathematik 1
(1936), 162–165; Swets and Zeitlinger N.V. Amsterdam, 1966.
3. J. Schreier, Über Schnitte convexer Flächen, Bull. Int. Ac. Pol. Series A (1933), 155–157.
R. Daniel Mauldin
We may interpret this problem as follows: “if R is a compact, (n + 1)-dimensional

manifold with boundary in Rn+1 for which every n-dimensional hyperplane H that
meets R in more than a point has H ∩ ∂ R an (n − 1)-sphere, is R convex?”
Schreier [3] showed first that a two-dimensional surface in R3 , each of whose

nondegenerate planar sections is a Jordan curve, is the boundary of a convex body.
Montejano [4] generalized Schreier’s theorem as follows: “Let N be a closed,
connected n-manifold topologically embedded in Rn+1 . Suppose that for every

n-dimensional hyperplane H that meets N in more than a point, H − N has exactly
two components. Then N is the boundary of a convex (n + 1)-body.”
As a corollary of this theorem, Problem 68 was settled affirmatively.
The following interpretation of Ulam’s problem was also proved by Montejano
in [4]: Let 1 ≤ k ≤ n. If R is a compact (n + 1)-manifold with boundary in Rn+1
for which every k-dimensional plane H that meets the interior of R has H ∩ ∂ R a
(k − 1)-sphere, then R is a convex (n+1)-body.
In [4], the following homological characterization of convexity which is closely
related with Problem 68 is considered. Let 1 ≤ k ≤ n and let K be a compact
subset of Rn+1 , suppose that for every k-dimensional plane H, H ∩ K is acyclic,
then K is convex. This theorem was also obtained by Kosinski [7] and generalizes
the following results of Aumann [1, 2], mentioned in the first edition of this book:
a continuum K in R3 is convex provided that for each plane P, P ∩ K and P − K
is connected; a continuum K in Rn is convex if and only if Rn − K is connected
and the intersection of K with each (n − 1)-dimensional hyperplane is convex; and
a closed subset K of Rn is convex if and only if K ∩ P is simply connected for
all 2-dimensional hyperplanes P. Later Montejano and Shchepin [5] and [6] gave
homological characterizations of convex sets in terms of acyclic support sets.
1. G. Aumann, Eine einfache Characterisierung der convexen Kontinuen in R3 , Deutsche Mathe-
matik 1 (1936), 108.
2. G. Aumann, Über Schnitteigenschaften convexer Punktmengen im R3 , Deutsche Mathematik 1
(1936), 162-165. Swets and Zeitlinger N.V. Amsterdam, 1966.
3. J. Schreier, Über Schnitte convexen Flächen, Bull, Int. Ac. Pol. Series A (1933), 155-157.
4. L. Montejano, About a problem of Ulam concerning flat sections of manifolds, Comment. Math.
Helvetici 65 (1990), 462-473.
5. L. Montejano and E. Shchepin, A characterization of convex sets in terms of acyclic support
sets, Bull. London Math. Soc., 28 (1996), 501-504.
6. L. Montejano and E. Shchepin, Topological tomography in convexity, Bull. London Math. Soc.
34 (2002), 353-358.
7. A. Kosinski, A theorem on families of acyclic sets and its applications, Pacific J. Math., 12
(1962), 317-325
Luis Montejano
The problem of characterizing the spaces of type (B) among the metric spaces.
There is given a complete metric space E with the following properties:
(1) If p, q ∈ E, there exists exactly one x ∈ E, such that x is a metric center of the
couple (p, q);
(2) If p, q ∈ E, there exists exactly one x ∈ E, such that q is a metric center of the
couple (p, x).
Is the space E isometric to a certain space of type (B)? [Every space of type (B)
has the properties of (1) and (2).]
Definition of a metric center of a couple of points (p, q): We take the set of all
points x ∈ E such that px + xq = pq; we denote it by R. By R1 we denote the set of
all points r ∈ R such that rx ≤ d(R)/2 for every x ∈ R, where d(R) is the diameter of
the set R; we denote by Rn+1 the set of all points r ∈ Rn such that rx ≤ d(Rn )/2 for
all x ∈ Rn . One can show that the intersection R1 R2 · · · Rn · · · contains at most one
point; if such a point exists, we call it the metric center of the pair (p, q).
Addendum. The answer is negative.
S. Mazur
December 21, 1936
Commentary
Lobaczewski’s geometry is a counterexample. A more elementary example is a

hyperbolic paraboloid with distances measured on geodesics.
Stefan Rolewicz
PROBLEM 70: ULAM

Prove the following lemma: Let f1 (x1 , x2 , . . . , xn ; t1 , t2 , . . . , tr ); 0 ≤ xi ≤ 1;
0 ≤ tj ≤ 1; i = 1, . . . , n; j = 1, . . . , r be a polynomial with variables xi and tj real-
valued and vanishing identically at the point X = (0, 0, . . . , 0; t1 , t2 , . . . , tr ); ε a
positive number. There exists then a polynomial f2 in the same variables and
constants K and ρ both positive and independent from ε (both K and ρ = 1?)
such that the following conditions are satisfied.
(1) |f1 (x1 , . . . , xn ; t1 , . . . , tr ) − f2 (x1 , . . . , xn ; t1 , . . . , tr )| < ε .
(2) The derivatives with respect to the variables x at the point xi = 1; i = 1, . . . , n imi-
tate the behavior of the polynomial; that is to say, if T and T denote two sets of
variables t1 , . . . , tr and t1 , . . . , tr so that |f2 (x1 , . . . , xn ; T ) − f2 (x1 , . . . , xn ; T )| <
ε , then we have for every i:

∂ f (x , . . . , x ; t , . . . , t ) ∂ f2 (x1 , . . . , xn ; t1 , . . . , tr )
2 1 n 1
r
− < Kε .
∂ xi ∂ xi
x1 =x2 =...=xn =0 x1 =x2 =...=xn =0
(3) The derivatives with respect to at least one of the variables x at the point x1 =
x2 = . . . = 0 are essentially different from zero. That is to say, there exist points
T ∗ and T ∗∗ such that

∂ f (x , . . . , x ; T ∗ ) ∂ (x , . . . , ∗∗ )
2 1 n f 2 1 xn ; T
− > ρ.
∂ xi ∂ xi
x1 =x2 =...=xn =0 x1 =x2 =...=xn =0
From an affirmative solution, i.e., from this lemma, there would follow an
affirmative answer to Hilbert’s problem concerning introduction of analytic
parameters in n-parameter groups. (The problem was solved for compact groups
by von Neumann.)
PROBLEM 71: ULAM

Find all the permutations f (n) of the sequence of natural integers which have the
property that if {nk } is an arbitrary sequence of integers with a density α , then the
sequence f (nk ) has also a density α , in the set of all integers.
PROBLEM 72: MAZUR

Let E be a space of type (F) with the following property: If Z ⊆ E is a compact
set, then the smallest closed convex set containing Z is also compact. Is E then a
space of type (F0 )? [See Problem 26 for a definition of (F0 ).]
Commentary
Mazur’s theorem states that the closed convex hull of a compact subset of a Banach
space is compact [1]. Problem 72 then asks for a partial converse to this result. It was
answered by Mazur and Orlicz [2] who showed that an F-space X is locally convex
if and only if whenever xn → 0 in X, tn ≥) and ∑ tn ≤ 1 then the series ∑ tn xn is
bounded (i.e., has bounded partial sums). Thus if the convex hull of every compact
set is bounded, then X is locally convex. In particular, spaces of type (F0 ) are locally
convex and the answer to Problem 72 is affirmative.
1. S. Mazur, Über die kleinste konvexe Menge, die eine gegebene kompakte Menge enthält, Studia
Math. 2 (1930), 7–9.
2. S. Mazur and W. Orlicz, Sur les espaces métriques linéares (1), Studia Math. 10 (1948),
184–208.
N.J. Kalton

Let cn be the smallest number with the property that if F(x1 , . . . , xn ) is an arbitrary
symmetric n-linear operator [in a space of type (B) and with values in such a space],
then
sup F(x1 , . . . , xn ) ≤ cn sup F(x, . . . , x).

xi ≤1, x≤1
i=1,2,...,n
It is known (Mr. Banach) that cn exists. One can show that the number cn satisfies
the inequalities

nn 1 n n
n!
≤ cn ≤ ∑ k · kn .
n! k=1
Is cn = nn /n!?
Commentary
The answer to this problem is yes for any real normed linear spaces and it is now
a standard fact in the field of infinite dimensional holomorphy. R.S. Martin proved
that cn ≤ nn /n! in his 1932 thesis [11] with the aid of an n-dimensional polarization
formula. His argument was published a few years later in [15] by A.E. Taylor.
Although this polarization formula was known to Mazur and Orlicz [12, p. 52],
it appears that they used the case x = 0 rather than the case x = − ∑n1 hk /2, which
gives the best estimate. Extremal examples in 1 and L[0, 1] showing that cn = nn /n!
are given in [5, 10], and [16]. For expositions, see [4, p. 48] and [13, p. 7]. Note
that by the Hahn-Banach theorem, there is no loss of generality in this problem if
all multilinear mappings are taken to be complex valued.
S. Banach [1] showed in 1938 that cn = 1 when only real Hilbert spaces are
considered. (He also assumed separability, though this assumption is not needed.)
His result can be deduced quite easily from [3, Satz 9] or [9, Th. IV]. For modern
expositions, see [2, p. 62], [6], or [8]. It is shown in [6] that Banach’s result and an
improvement by Szego of Bernstein’s inequality for trigonometric polynomials are
easily deduced from each other. For complex Lp -spaces, 1 ≤ p < ∞, it is conjectured
in [6] that
|p−2|
nn p
cn ≤ ,
n!
and this is proved when n is a power of 2. It also follows from [6] that
nn/2 (n + 1)(n+1)/2
cn ≤
2n n!
holds for J ∗ -algebras [7]. (In particular, the space C(S) of all continuous complex-
valued functions on a compact Hausdorff space S and, more generally, any
B∗ -algebra is a J ∗ -algebra.) Since cn = 1 for the space C(S) with S a two point set
[6, p. 154], it is natural to ask whether this holds for any compact Hausdorff space S.
If so, then it is easy to deduce that the Bernstein inequality holds for polynomials
on C(S). (See [6, p. 149].)
A natural generalization of Problem 73 is the following: Let k1 , . . . , dn be
nonnegative integers whose sum is n and let c(k1 , . . . , kn ) be the smallest number
with the property that if F is any symmetric n-linear mapping of one real normed
linear space into another, then
sup F(x1k1 , . . . , xnkn ) ≤ c(k1 , . . . , kn ) sup F(x, . . . , x).

xi ≤1, x≤1
i=1,2,...,n
where the exponents denote the number of coordinates in which the base variable
appears. It is shown in [6] that if only complex normed linear spaces and complex
scalars are considered, then
k1 ! · · · kn ! nn
c(k1 , . . . , kn ) = (1)
k1k1 · · · knkn n!
(where 00 = 1) but there are many cases where (1) does not hold when real normed
linear spaces are considered.
1. S. Banach, Über homogene Polynome in (L2 ), Studia Math. 7 (1938), 36–44.
2. J. Bochnak and J. Siciak, Polynomials and multilinear mappings in topological vector spaces,
Studia Math. 39 (1971), 59–76.
3. J.G. van der Corput and G. Schaake, Ungleichungen für Polynome und Trigonometrische
Polynome, Composito Math. 2 (1935), 321–361; Berichtigung ibid., 3 (1963), 128.
4. H. Federer, Geometric Measure Theory, Springer-Verlag, Berlin-Heidelberg-New York 1969.
5. B. Grünbaum, Two examples in the theory of polynomial functionals, Riveon Lematematika
11 (1957), 56–60.
6. L.A. Harris, Bounds on the derivatives of holomorphic functions of vectors, Analyse Fonction-
nelle et Applications, Leopoldo Nachbin editor, Hermann, Paris 1975.
7. , Bounded symmetric homogeneous domains in infinite dimensional spaces,
Lecture Notes in Math. 364, Springer-Verlag, Berlin, 1974, 13–40.
8. L. Hörmander, On a theorem of Grace, Math. Scand. 2 (1954), 55–64.
9. O.D. Kellogg, On bounded polynomials in several variables, Math. Zeit. 27 (1928), 55-64.
10. J. Kopeć and J. Musielak, On the estimation of the norm of the n-linear symmetric operation,
Studia Math. 15 (1955), 29–30.
11. R.S. Martin, Thesis, Cal. Inst. of Tech., 1932.
12. S. Mazur and W. Orlicz, Grundlegende Eigenschaften der polynomischen Operationen. I–II,
Studia Math. 5 (1934), 50–68, 179-189.
13. L. Nachbin, Topology on Spaces of Holomorphic Mappings, Springer-Verlag, New York, 1969.
14. T.J. Rivlin, The Chebyshev Polynomials, Wiley, New York, 1974.
15. A.E. Taylor, Additions to the theory of polynomials in normed linear spaces, Tôhoku Math. J.
44 (1938), 302–318.
16. , (Review of a paper of Kopéc and Musielak), Math. Reviews 17 (1956), 512.
17. D.R. Wilhelmsen, A Markov inequality in several dimensions, J. Approx. Theory 11 (1974),
216–220.
Lawrence A. Harris
Mathematics Department
University of Kentucky
Lexington, Kentucky 40506
Given is a polynomial
W(t1 , . . . , tn ) = ∑ ak1 ,...,kn t1k1 · · · tnkn

k1 +...+kn =n
in real variables t1 , . . . , tn , homogeneous and of order n; let us assume that

|W(t1 , . . . , tn )| ≤ 1 for all t1 , . . . , tn such that |t1 | + . . . + |tn | ≤ 1. Do we then have
nn
|ak1 ,...,kn | ≤ ?
k1 ! · · · kn !
Commentary
The answer to this problem is yes and the solution is an easy consequence of the
solution to Problem 73. Indeed, let F be the symmetric n-linear map on 1n such that
W(x) = F(x, . . . , x) for all x ∈ 1n . Then applying the multinomial theorem [12, p. 52]
with x = t1 e1 + . . . + tn en and the uniqueness of the representation for W(t1 , . . . , tn ),
we obtain
n!
ak1 ,...,kn = F(ek11 · · · eknn ), (2)
k1 ! · · · kn !
where e1 , . . . , en is the standard basis for 1n . The desired estimate follows. Note
that the problem of determining the best estimate α (k1 , . . . , kn ) in Problem 74 is
equivalent to the problem of determining c(k1 , . . . , kn ) in Problem 73; for, if F is a
symmetric n-linear map satisfying F(x, . . . , x) ≤ 1 for all x ≤ 1 and if x1 ≤
1, . . . , xn ≤ 1, then the polynomial W(t1 , . . . , tn ) = F(x, . . . , x), where x = t1 x1 +
. . . + tn xn , satisfies the hypotheses of Problem 74 and
n!
ak1 ,...,kn = F(x1k1 , · · · , xnkn ).
k1 ! · · · kn !
Thus
n!
α (k1 , . . . , kn ) = c(k1 , · · · , kn ). (3)
k1 ! · · · kn !
The general problem of obtaining estimates on the coefficients of polynomials
in m variables which satisfy a given growth condition on Rm can be solved with
the aid of the following generalized polarization formula: Let W(t1 , . . . , tm ) be any
homogeneous polynomial of degree n and let ak1 ,...,km be the coefficient of t1k1 · · · tm
km
in its expansion. For each i = 1, . . . , m, choose distinct real numbers xi0 , . . . , xiki
and put
Γij = ∏(xij − xi ), 0 ≤ j ≤ ki ,

=j
with Γij = 1, if ki = 0. Then
W(x1j1 , . . . , xmjm )
ak1 ,...,km = ∑ , (4)
Γ1j1 · · · Γmjm
where the sum is taken over all 0 ≤ j1 ≤ k1 , . . . , 0 ≤ jm ≤ km . Note that one can
convert any polynomial p of degree ≤ n in m − 1 variables to a homogeneous
polynomial W of degree n in m variables by defining

t1 tm−1
W(t1 , . . . , tm ) = tm
n
p ,..., .
tm tm
One can obtain estimates on the left-hand side of (4) by estimating the right-hand
side of (4) and minimizing. (A reasonable first choice is xij = ki /2 − j.) To prove
(4), observe that if p(t) is a polynomial of degree ≤ ki , then the coefficient of tki
in the Lagrange interpolation formula for p is ∑kj=0
i
p(xij )/Γij and apply this to each
variable of W.
For example, we show that the improved estimate
nn
|ak1 ,...,kn | ≤ r (5)
k1 ! · · · kn !
holds in Problem 74, where
n
1 + e−2 ki
r= , =∑ .
2 i=1 2
Indeed, choose xi0 = 2, xi1 = 0, xi2 = −2 for i = 1, . . . , , xi0 = 1, xi1 = −1 for

i = + 1, . . . , n − , and xi0 = 0 for i = n − + 1, . . . , n. Then by (4),

1
|a2...21...10...0 | ≤ ∑ (n − 2j)n ≤ 2− nn r , (6)
4 j=0 j
where the last inequality follows from (1 − 2j/n)n ≤ e−2j . Clearly
c(k1 , . . . , kn ) ≤ c(2, . . . , 2, 1, . . . , 1, 0, . . . , 0),
and this together with (3) and (6) implies (5).

A related problem of interest is to find a Banach space analogue of Markov’s
theorems; that is, to find the smallest number Mn,k with the property that if P is any
polynomial of degree ≤ n mapping one real normed linear space into another, then
sup D̂k P(x) ≤ Mn,k sup P(x),

x≤1 x≤1
where D̂k p(x)y = dk /dtk p(x + ty) |t=0 . It is not difficult to show that
(k) (k)
Tn (1) ≤ Mn,k ≤ 22k−1 Tn (1), (7)
where Tn is the Chebyshev polynomial of degree n. (See [14, p. 119].) Indeed,

let P be any real-valued polynomial of degree ≤ n on a real normed linear space
and suppose |P(x)| ≤ 1 for all x ≤ 1. Let x ≤ 1, y ≤ 1 and −1 ≤ s ≤ 1.
Define q(t) = P(φ (t)), where φ (t) = [x − sy + t(x + sy)]/2, and note that φ (t) ≤
(1 + t)/2 + (1 − t)/2 = 1 when −1 ≤ t ≤ 1. Then q is a polynomial of degree ≤ n
satisfying |q(t)| ≤ 1 for −1 ≤ t ≤ 1, so |q(k) (1)| ≤ |T (k) (1)| by [14, 1.5.11] and
(k)
clearly q(k) (1) = 2−k D̂k P(x)(x + sy). Hence the map s → D̂k Tn (1) on [0, 1] and
k k
D̂ P(x)y is the coefficient of s in this polynomial. Therefore,
(k)
|D̂k P(x)y| ≤ 2k−1 [2k Tn (1)]
by [14, p. 57]. Thus (7) follows by the Hahn-Banach theorem.

Note that the value of Mn,k is unchanged when only real-valued polynomials on
12 are considered. It is shown in [6] and [9] that Mn,1 = n2 when only real Hilbert
(k)
spaces are considered and it would be interesting to know whether Mn,k = Tn (1)
for all 1 ≤ k ≤ n in this case. See also [17].
1. S. Banach, Über homogene Polynome in (L2 ), Studia Math. 7 (1938), 36–44.
2. J. Bochnak and J. Siciak, Polynomials and multilinear mappings in topological vector spaces,
Studia Math. 39 (1971), 59–76.
3. J.G. van der Corput and G. Schaake, Ungleichungen für Polynome und Trigonometrische
Polynome, Composito Math. 2 (1935), 321–361; Berichtigung ibid., 3 (1963), 128.
4. H. Federer, Geometric Measure Theory, Springer-Verlag, Berlin-Heidelberg-New York 1969.
5. B. Grünbaum, Two examples in the theory of polynomial functionals, Riveon Lematematika
11 (1957), 56–60.
6. L.A. Harris, Bounds on the derivatives of holomorphic functions of vectors, Analyse Fonction-
nelle et Applications, Leopoldo Nachbin editor, Hermann, Paris 1975.
7. , Bounded symmetric homogeneous domains in infinite dimensional spaces,
Lecture Notes in Math. 364, Springer-Verlag, Berlin, 1974, 13–40.
8. L. Hörmander, On a theorem of Grace, Math. Scand. 2 (1954), 55–64.
9. O.D. Kellogg, On bounded polynomials in several variables, Math. Zeit. 27 (1928), 55-64.
10. J. Kopéc and J. Musielak, On the estimation of the norm of the n-linear symmetric operation,
Studia Math. 15 (1955), 29–30.
11. R.S. Martin, Thesis, Cal. Inst. of Tech., 1932.
12. S. Mazur and W. Orlicz, Grundlegende Eigenschaften der polynomischen Operatoren I–II,
Studia Math. 5 (1935), 50–68, 179-189.
13. L. Nachbin, Topology on Spaces of Holomorphic Mappings, Springer-Verlag, New York, 1969.
14. T.J. Rivlin, The Chebyshev Polynomials, Wiley, New York, 1974.
15. A.E. Taylor, Additions to the theory of polynomials in normed linear spaces, Tôhoku Math. J.
44 (1938), 302–318.
16. , (Review of a paper of Kopéc and Musielak), Math. Reviews 17 (1956), 512.
17. D.R. Wilhelmsen, A Markov inequality in several dimensions, J. Approx. Theory 11 (1974),
216–220.
Research supported in part by the National Science Foundation.
Lawrence A. Harris
University of Kentucky
PROBLEM 75: MAZUR

In Euclidean n-dimensional space E, or, more generally, in a space of type (B)
there is given a polynomial W(x). α is a number = 0. If the polynomial W(x)
is bounded in an ε -neighborhood of a certain set R ⊂ E, is it then bounded in a
δ -neighborhood of the set α R (which is the set composed of elements α x for x ∈ R)?
(See Problem 55.)
Addendum. From the solution of Problem 55, it follows that the theorem is true
in the case of a Euclidean space.
Mazur
PROBLEM 76: MAZUR

Given in 3-dimensional Euclidean space is a convex surface W and point 0 in its
interior. Consider the set V of all points P defined by the property that the length
of the interval P0 is equal to the area of the plane section of W through 0 and
perpendicular to this interval. Is the set V convex?
PROBLEM 77: ULAM

Prize for (a), one bottle of wine, S. Eilenberg
(a) Let A and B e two topological spaces such that the spaces A2 and B2 are
homeomorphic. Is then the space A homeomorphic to the space B?
(b) Let A and B be two metric spaces such that A2 is isometric to B2 . Is A isometric
with B?
(c) Let A and B be two abstract groups such that A2 and B2 form isomorphic groups.
Is A isomorphic with B?
[We understand by A2 (resp. B2 ) the set of ordered pairs of elements of the set A (or
B).] A topology [or, in Problem (c), the group operation] in such sets is defined, for
example, in the “Euclidean” manner: by the square root of the sum of squares of the
distances between projections.
Commentary Parts (a) and (b)
A number of papers have been devoted to part (a). In 1947, R. H. Fox [7] gave
an example of two nonhomeomorphic compact 4-dimensional manifolds whose
cartesian squares are homeomorphic. In 1960, J. Glimm [10] noticed that the
cartesian square of the contractable open manifold, which is not homeomorphic to
E3 , described by J.H.C. Whitehead in [17], is homeomorphic to E6 . This result was
generalized to a class of contractable open 3-manifolds by D.R. McMillan, Jr. [13].
In 1964, K.N. Kwun proved [11] that if α is a simple arc in En and β is a simple
arc in Em , then En /α × Em /β is homeomorphic to En+m . Since there exist wild arcs
in En for n ≥ 3 such that their complement is not simply connected, the result of
Kwun gives us a class of “cartesian elements” or roots of E2n , for n ≥ 3, which
are not open topological manifolds. Another class of cartesian elements of E2n was
constructed by A.J. Boals [3] in 1970. This class includes the famous dog bone
space of Bing [1, 2]. K.W. Kwun and F. Raymond constructed nontrivial “cartesian
elements” of the cube [0, 1]2n , for n ≥ 2 [12]. In 1978, an analogous result for the
Hilbert cube was published by Cerin [6]. H. Torunczyk [16] proved a number of
very general results which imply, for example, that if A and B are cell-like finite
dimensional continua in the Hilbert cube Q, the Q/A × Q/B is homeomorphic to Q.
It is not known whether there exist nonhomeomorphic, 3-dimensional, compact
manifolds A and B such that A2 and B2 are homeomorphic. It is not known whether
there exist 3-dimensional nonhomeomorphic polyhedra A and B so that A2 is
homeomorphic to B2 .
There are some cases for which part (a) has a positive answer. R.H. Fox [7]
showed that the answer is yes if A and B are 2-dimensional compact manifolds
with or without boundary. Recently, W. Rosicki, in a dissertation [15] upon which
this commentary is based, gave an affirmative answer in case A and B are com-
pact 2-dimensional polyhedra. Related problems were considered by Borsuk [4],
H. Patkowska [14], Furdzik [8], and Cauty [5].
Part (b) was solved in the negative by G. Fournier [9]. However, it is open
whether there is an affirmative solution to (b) in case A and B are complete metric
spaces. In fact, part (b) is open in the case where A and B are assumed to be compact.
1. R.H. Bing, A decomposition of E3 into points and tame arcs such that the decomposition space
is topologically different from E3 , Ann. of Math. 65 (1957), 484–500.
2. , The Cartesian product of a certain nonmanifold and a line is E4 , Ann. of Math.
70 (1959), 399–412.
3. A.J. Boals, Non-manifold factors of Euclidean space, Fund. Math. 68 (1970), 159–177.
4. K. Borsuk, Sur la décomposition des polyèdres en produits cartesiens, Fund. Math. 31 (1938),
137–148.
5. R. Cauty, Sur les homéomorphismes de certain produits de courbes, Bull. Acad. Polon. Sci.
Sér. Sci. Math., 27 (1979), 413-416.
6. Z. Čerin, Hilbert cube modulo an arc, Fund. Math. 101 (1978), 111–119.
7. R.H. Fox, On a problem of S. Ulam concerning Cartesian products, Fund. Math. 34 (1947),
27S–287.
8. Z. Furdzik, On the uniqueness of decomposition into Cartesian product of two curves, Bull.
Acad. Polon. Sci., 14 (1966), 57–61.
9. G. Fournier, On a problem of S. Ulam, Proc. Amer. Math. Soc., 29 (1971), 622.
10. J. Glimm, Two Cartesian products which are Euclidean spaces, Bull. Soc. Math. France, 88
(1960),131–135.
11. K.W. Kwun, Product of Euclidean spaces modulo an arc, Ann. of Math., 79 (1964), 104–108.
12. K.W. Kwun and F. Raymond, Factors of cubes, Amer. J. Math., 84 (1962), 433–440.
13. D.R. McMillan, Jr., Cartesian products of contractable open manifolds, Bull. Amer. Math. Soc.,
67 (1961), 510–514.
14. H. Patkowska, On the uniqueness of the decomposition of finite dimensional ANR’s into
Cartesian products of at most 1-dimensional spaces, Fund. Math., 58 (1966), 89–110.
15. W. Rosicki, O Problemie Ulama Dotyczacym Kartezjanskich Kwadratou Wieloschianow 2-
wymiarowych, Rozprowo Doktorska, Gdansk, 1979; On a problem of S. Ulam concerning
Cartesian squares of 2-dimensional polyhedra., Fund. Math. 127 (1987), 101-125.
16. H. Toruńczyk, On CE-images of the Hilbert cube and characterization of Q-manifolds, Fund.
Math., 106 (1980), 31-40.
17. J.H.C. Whitehead, A certain open manifold whose group is unity, Quart. J. Math., 6 (1935),
268–279.
R. Daniel Mauldin
Commentary—Part (c)
For abelian (i.e., commutative) groups this was one of the three “test problems” in
the first edition of [4]. I did not know of its appearance in the Scottish Book. My
idea was to show how Ulam’s theorem for countable torsion groups could really be
used to answer explicit questions.
Jónsson [3] gave a negative answer with A and B torsion-free of rank two.
Crawley [2] gave an example for torsion groups (of course uncountable). Corner [1]
exhibited a countable torsion-free abelian group G which is isomorphic to G⊕G⊕G
but not to G ⊕ G—this is even more spectacular.
1. A.L.S. Corner, On a conjecture of Pierce concerning direct decompositions of torsion-free
abelian groups, Proc. of Coll. on Abelian groups, Budapest, 1964, 43–48.
2. P. Crawley, Solution of Kaplansky’s test problems for primary abelian groups, J. Alg. 2 (1965),
413–431.
3. B. J’onsson, On direct decompositions of torsion-free abelian groups, Math. Scand., 5 (1957),
230–235.
4. I. Kaplansky, Infinite Abelian groups, U. of Michigan Press, 1954, 1969.
I. Kaplansky

August 2, 1935
Find all the surfaces with following property: Through every point of the surface
there lie two curves congruent, respectively, to two given curves A and B. Compare
Problem 61. (Such a surface is, for example, a cylinder: the curves A and B are here a
circle and a straight line.)

A polynomial y = U(x) maps in a one-to-one fashion, a space X of type (B) onto
a space Y of type (B); the inverse of this mapping x = U −1 (y) is also a polynomial.
Is the polynomial y = U(x) of first degree? Not decided even in the case when X
and Y are a Euclidean plane; in this case, the question is given for a one-to-one
mapping t = φ (t, s), s = ψ (t, s) of a plane onto itself where φ (t, s), ψ (t, s) are
polynomials; the inverse mapping has also the form t = Φ (t , s ), s = Ψ (t , s ) where
Φ (t , s ), ψ (t , s ) are polynomials. Is the mapping affine; that is to say, of the form
t = a1 t + b1 s + c1 , s = a2 t + b2 s + c2 where a1 b2 − a2 b1 = 0?
Addendum. Trivial. In the Euclidean space:
y1 = x1 + k
y2 = x2 + φ2 (x1 )
y3 = x3 + φ3 (x1 , x2 )
..
.
yn = xn + φn (x1 , . . . , xn−1 )
where k is an arbitrary constant and φ2 , . . . , φn are arbitrary polynomials in their

variables. The inverse mapping is obvious at once.
PROBLEM 80: MAZUR

Let E be a complete metric space; we denote by E∞ a complete metric space
formed by the set of all sequences {en } of elements of E. By a distance between
two such sequences {en }, {en } we understand the number
∞
(e , e )
∑ 2−n 1 + (e
n n
, e )
n n
n=1
[For e , e ∈ E we denote by (e , e ) the distance between the elements e , e ]. If R
is a given set contained in E, then we denote by Rδ the set of all sequences {rn } of
elements of R, and by Rσ the set of all sequences {rn } of elements of R such that
rn = r0 almost always; r0 is a fixed element of R. Is it true that: If the set R is an Fσ
set but not closed, then Rδ is an Fσ δ set but not an Fσ ; if the set R is an Fσ δ but not
an Fσ , then Rσ is and Fσ δ σ but not an Fσ δ ; more generally, if R is an F2ξ +1 but not
an F2ξ , then Rδ is an F2ξ +2 , but not an F2ξ +1 , and if R is an F2ξ +2 but not an F2ξ +1 ,
then Rσ is an F2ξ +3 but not an F2ξ +2 (F0 = F, F1 = Fσ , F2 = Fσ δ , F3 = Fσ δ σ , . . .)?
Investigate in particular the case when the space E is compact or of type (B) or of
type (F).
Commentary
If X is a metric space, let F0 (X) be the family of all closed subsets of X. For each
ordinal α > 0, let
Fα (X) = { K : K = Fn , where each Fn ∈ { Fβ (X) : β < α } },
if α is odd, and let

Fα (X) = { K : K = Fn , where each Fn ∈ { Fβ (X) : β < α } },
if α is even. Limit ordinals are considered even.

Thus, in the terminology of this problem it is true that if R ∈ F2α +1 (E), then
Rδ ∈ F2α +2 (E∞ ) and if R ∈ F2α (E), then Rσ ∈ F2α +1 (E∞ ), for 0 ≤ α < ω1 . The
problem is whether these estimates are sharp. Specifically,
(a) if R ∈ F1 (E) − F0 (E), does Rδ ∈ F2 (E∞ ) − F1 (E∞ )?
(b) if R ∈ F2 (E) − F1 (E), does Rσ ∈ F3 (E∞ ) − F2 (E∞ )?
(c) if R ∈ F2α +1 (E) − F2α (E), does Rδ ∈ F2α +2 (E∞ ) − F2α +1 (E∞ )?
(d) if R ∈ F2α +2 (E) − F2α +1 (E), does Rσ ∈ F2α +3 (E∞ ) − F2α +2 (E∞ )?
If all of these questions have affirmative answers, then one would have an elegant
method of generating Borel sets of arbitrarily high class. One could simply take a
subset of E whose structure is known and iterate the described procedure through
E∞ , (E∞ )∞ , . . .. However, the solutions to these general problems are unknown.

There are some positive solutions in specific cases.
One can show that if E is compact, then the answer to (a) is yes as follows:
Let R = Kn , where each Kn is compact and suppose Rδ = Tn , where each
Tn is closed in E∞ . Suppose R is not compact. Then for each n, R = πn (Tn ), since
Tn is compact and πn (Tn ) is compact where πn is the projection of E∞ onto the
nth coordinate. For each n, let rn ∈ R − πn (Tn ). Then {rn }∞ n=1 ∈ Rδ − Tn . This
contradiction establishes (a) in case E is compact.
Using a variation of the procedure described by Mazur and a wonderful appli-
cation of Brouwer’s fixed point theorem, Sikorski [2] and Engelking, Holsztynski,
and Sikorski [1] gave a positive solution to the iterative process in a special case.
Sikorski [3] also used the Brouwer fixed point theorem to give a specific example
of an analytic set which is not a Borel set. In [1], Sikorski raised a problem which is
closely associated with Problem 80.
1. R. Engelking, W. Holsztynski, and R. Sikorski, Some examples of Borel sets, Coll. Math., 15
(1966), 271–274.
2. R. Sikorski, Some examples of Borel sets, Coll. Math., 5 (1958), 170–171.
3. , On an Analytic Set, Bull. Acad. Sci. Pol., 14 (1966), 15–16.
R. Daniel Mauldin

August 6, 1935
Plane of
Symmetry
Axis of
Surface
A hyperbolic paraboloid and a plane are composed, in two ways, of curves which
are imbedded in the surface (AA; BB), straight lines and parabolas. Do there exist
other surfaces of this kind? Are they composed of (AB), (CD)? Is it true that such
surfaces, namely, all surfaces having at each point two intersecting curves congruent
to A and B, respectively (exceptis excipiendis), are necessarily of the form z = f (x)+
g(y)? (The plane, sphere, and circular cylinder are considered trivial.) (Compare
Problems 44 and 61)
Commentary
This problem is not clearly stated. We will make an attempt to interpret it. Let us
disregard the plane since the plane as an example of a composed surface is usually
trivial. With this exception, the first sentence says that a hyperbolic paraboloid
is composed in two ways of straight lines and parabolas. (That the hyperbolic
paraboloid and the hyperboloid of one sheet are the only doubly ruled surfaces
is well known. See, for example, M. Spivak, A Comprehensive Introduction to
Differential Geometry, Vol. 3, 2nd edition, page 345, problem 14, Publish or Perish,
Inc., 1979.)
The notation (AA; BB) might have the following interpretation. Suppose one can
find on a surface four families of curves (A1 ), (A2 ), (B1 ), and (B2 ) such that
a. All curves of the family (A1 ) are congruent to one another; all curves of the
family (A2 ) are congruent to one another and to the curves (A1 );
b. Likewise for (B1 ) and (B2 );
c. Through each point of the surface there pass four curves, one from each family.
One then needs to show that the hyperbolic paraboloid has this property where
(A1 ) and (A2 ) consist of straight lines and (B1 ) and (B2 ) consist of parabolas. Of
course, with this interpretation, any surface of revolution with a plane (Fig. 81.1) of
symmetry perpendicular to the axis of revolution has these properties. Such a surface
is fibered by circles parallel to the plane of symmetry and if we take any curve A
at all on the circle which meets each circle in one point, we see that the surface
is generated by congruent copies of A. And, provided A is not itself symmetric in
the plane of symmetry of the surface, there will be a second system of curves on
the surface congruent to A (viz. the reflections of the first system in the plane of
symmetry). In fact, we see that the surface contains a noncountable infinity of pairs
of curve-systems (A1 , A2 ), (B1 , B2 , . . . having the properties (a) and (b) above with
one curve of every system passing through any chosen point of the surface. It may
be a reasonable conjecture that only the hyperbolic paraboloid, other than surfaces
of revolution, has properties (a), (b), and (c).
In the third sentence, Steinhaus asks if there are surfaces composed of (AB),
(CD). He might mean that (AB) ≡ (AA; BB) and then is asking if the structure with
the structure (AA; BB) can have also a different structure (CC; DD).
A Reviewer [Ed. who requested to remain anonymous.]


August 6, 1935
f (t) is independent (in the sense of correlation) from y1 (t), y2 (t), . . . , yn (t) (0 ≤
t ≤ 1), if, for every function of n-variables F(y1 , y2 , . . . , yn ) and for every 4-tuple of
numbers (α1 , β1 , α2 , β2 ) the sets which are defined as follows: A = Et (α1 ≤ f (t) ≤
B1 ), B = Et (α2 ≤ F(y1 (t), . . . , yn (t)) ≤ β2 ) have the property that |AB| = |A| · |B|.]
Problems
(1) Is a set of functions mutually independent (that is to say, each independent of
all the other n) at most countable?
(2) Does a system like that have to be complete and orthogonal, or only complete?
Remark: The notion of independence introduced above is what natural scientists

call “complete lack of correlation.” (Their definitions are, however, not too precise.)
Addendum. Under the assumption that the functions which are independent are
integrable, together with their th power, we have the following relation:
1 1
yk1 (t)yk2 (t) · · · yk (t)dt = ∏ yki (t)dt.
0 i=1 0
It follows immediately that the system {φi (t)} where φi (t) = yi (t) − 01 yi (t) is
orthogonal. If we assume that yi (t) ∈ L, then the system is “lacunary” (and therefore
cannot be complete). Lacunarity follows in this case from the relation

1 n n
1
∑ φi (t) dt ≥ M ∑ φi2 (t)dt,
0 i=1 i=1 0
1 2
where M does not depend on n nor does it depend on the sequence 0 φi (t)dt.
October 12, 1935
Commentary
The solution to part (1) is as follows. Let I be a set of mutually independent func-
tions; then all but countably many members of I are constant almost everywhere.
To see this, let I = { arctan φ : φ ∈ I }. Then I ⊂ L2 ([0, 1]). Let I = { φ ∈ I :
variance(φ ) > 0 }. If arctan φ ∈ I − I , then φ is constant almost everywhere.
Since B = { X − E(X)/σ (X) : X ∈ I }, is an orthonormal subset of L2 ([0, 1]), B,
and therefore I , is countable.
One answer to (2) is given by Steinhaus at the end of the problem. The following
may be a more elementary argument for the fact that a sequence of independent
random variables Xi ∈ L2 ([0, 1]) cannot span L2 .
We can assume without loss of generality that Xi is the constant 1 and that
σ (Xi ) > 0, for i > 1. For each i > 1, set Yi = Xi − E(Xi )/σ (Xi ). Then 1, Y1 , Y2 , . . .
is an independent, orthonormal sequence. Also, Y2 · Y3 is orthogonal to each of
these functions and, by independence, E(Y22 Y32 ) = E(Y22 )E(Y32 ) = 1. But, if the Xi ’s
spanned L2 , then we would have Y2 · Y3 = 0, which is inconsistent.
D. Stroock

We assume that a continuous function f (x) satisfies at every point the condition

f (x + h) − f (x)
lim <M
h→0 ha
(M a constant, 0 < a < 1, a constant). Does the function f (x) satisfy a Hölder
condition? (It is easy to prove that in every interval of a certain dense set of intervals,
Hölder condition holds with the exponent a and with the same constant.)
Addendum. The answer is negative. We define the function f (x) as a triangular
one in intervals 1/n − xn , 1/n + xn with the height 1/n.
Marcinkiewicz

One assumes that for a convex surface in the 3-dimensional space all its plane
sections by planes going through a fixed point 0 inside the surface are projectively
equivalent. Is this surface an ellipsoid?
PROBLEM 85: BANACH

Does there exist a sequence of measurable functions {φn (t)} (0 ≤ t ≤ 1),
belonging to L2 , orthogonal, normed, complete, and such that the development of
every polynomial is divergent almost everywhere?
(b) The same question if, instead of polynomials, we consider analytic functions
for 0 ≤ t ≤ 1.
One can prove that the answer to Question (a) is an affirmative one if we admit
only polynomials of degree less than n (n arbitrary given ahead of time).
Constructions of divergent Fourier series go back to A.N.Kolmogorov and

D.E.Menshov. In particular, Menshov (1923) discovered that there is an orthonormal
system of functions {ϕn } such that some series
∑ c(n)ϕn , {c(n)} ∈ l2
diverges almost everywhere (a.e.)
The subject was developed substantially in the beginning of 1960s. This progress
was inspired by Kolmogorov-Zahorski’s theorem which says that a trigonometric
L2 − series may diverge a.e. after a suitable rearrangement of its terms.
See [O-75] for a presentation of results obtained in the area.
In particular, one can find there (p.115) the following result of A. Krantsberg
[K-74]: one can construct a complete orthonormal system in L2 (I) such that Fourier
expansion of every (nontrivial) continuous function with respect to this system
diverges a.e.
This result answers Banach’s Problem 85 and also its stronger version -
Problem 186.
References
[K-74] A.S.Krantsberg, On divergent orthogonal Fourier series, Mat. Sb. 93 (1974), 540–553 (in
Russian).
[O-75] A.M.Olevskii, Fourier series with Respect to General Orthogonal Systems. Springer, 1975.
A. Olevskii
PROBLEM 86: BANACH

Given a sequence of functions {φn (t)} orthogonal, normed, measurable, and
uniformly bounded; can one always complete it, using functions with the same
bound, to a sequence which is orthogonal, normed, and complete? Consider the
case when infinitely many functions are necessary for completion.
K. S. Kazarian, in his article On a problem of S. Banach from the Scottish

Book, Proc. Amer. Math. Society 110 (1990), 881-887 showed that the answer
is no. Working with the Haar system of functions, it is shown that there is
an ONS {Φn } on the interval [0, 1] such that the dimension of the orthogonal
complement is infinite and there is no uniformly bounded orthonormal basis for
the orthogonal complement. In fact, stronger theorem are proven in the paper.
Kazarian mentions that Banach probably knew how to construct counterexamples
in case the complement is finite dimensional. Kazarian mentions a letter from B.
S. Kašin. in which it is related that he and A. M. Olevskii had obtained similar
results. Finally, Kazarian mentions a letter from Z. Cielsielski to Kašin where it is
stated that S. Kaczmarz solved the problem in 1936 with the help of the Banach
method. Kazarian was unable to find any published result of Kaczmarz along these
lines. I also made queries about Kaczmarz’s result, but no one was able to provide a
reference. Perhaps it is lost.
R. Daniel Mauldin
PROBLEM 87: BANACH
Let y = U(x) be an operation which is continuous and satisfies a Lipschitz
condition. The operation is defined in Lβ (β ≥ 1) and its image is also contained
in Lβ . We assume that for a certain α > β there exists a constant Mα such that if
x ∈ Lα then U(x) ∈ Lα and U(x)α ≤ Mα xα . Show that for every γ such that
β < γ < α there exists an Mγ with the property: If x ∈ Lγ , then U(x) ∈ Lγ and
U(x)γ ≤ Mγ xγ . This theorem is true under the additional assumption that U is
a linear operation (follows from a theorem of M. Riesz). Banach showed that the
theorem is true if α = ∞.
[ x(t)γ = ( 1 γ 1/γ .]
0 |x(t)| dt)
Commentary
The fundamental objects of linear and nonlinear functional analysis are operators
which map one Banach space (or more generally, topological vector space) into
another. One of the methods of studying operators is an interpolation theory, called
the theory of interpolation spaces. This theory has been applied to other branches
of analysis (e.g., Fourier series, Schauder basis, partial differential equations,
numerical analysis, approximation theory), but it is also of considerable interest
in itself. In order to explain the essence of these questions let us define linear
interpolation spaces.
Let A, B, Ai , Bi , i = 0, 1, be complex Banach spaces (or quasi-Banach spaces,
i.e., topological vector spaces which are complete and locally bounded) and let A,
Ai , i = 0, 1 be continuously embedded into some fixed Hausdorff topological vector
space A , likewise B, Bi , i = 0, 1 into B; A is intermediate between A0 and A1 , and
B is intermediate between B0 and B1 in the sense that
A0 ∩ A1 ⊂ A ⊂ A0 + A1 , B0 ∩ B1 ⊂ B ⊂ B0 + B1 ,
with the continuous inclusions ([1], p. 24–25).

The pair of spaces (A, B) is called a linear interpolation pair with a constant
C > 0 between the pairs of spaces (A0 , B0 ) and (A1 , B1 ) if for any linear continuous
operator L from Ai into Bi , i = 0, 1, L (or its appropriate unique extension L̂) is a
linear continuous operator from A into B and LA→B (or L̂A→B ) is majorized by
C max(LA0 →B0 , LA1 →B1 ). If Ai = Bi , i = 0, 1 and A = B, we shall say shortly that
the space A is a linear interpolation space with constant C > 0 between the spaces
A0 and A1 .
The first linear interpolation theorem was the M. Riesz theorem [22] of 1926,
formulated as inequalities for bilinear forms. This theorem was improved and
presented in an operator form by G.O. Thorin [26], who used the classical method of
analytic functions (the three-line theorem). Thorin’s theorem states that (Lp , Lq ) is a
linear interpolation pair with constant 1 between the pairs (Lp0 , Lq0 ) and (Lp1 , Lq1 ),
where 1/p = (1 − θ )/p0 + θ /p1 , 1/q = (1 − θ )/q0 + θ /q1 , 0 < θ < 1, and is now
generally known as the Riesz-Thorin theorem. A further important generalization
was the theorem of J. Marcinkiewicz. Its proof, obtained by the real method, was
published by A. Zygmund [28]. Next, E.M. Stein and G. Weiss [24] have proved
an important generalization of the Riesz-Thorin and Marcinkiewicz theorems. All
those theorems, however, concern Lp spaces or Lorentz Lpq spaces. Another theorem
which should be mentioned ! here, although not known in the literature, is the
Orlicz interpolation theorem [18] of 1954 exceeding Lp spaces and concerning the
interpolation of continuous function spaces and Lipschitz function spaces.
The theory of linear interpolation was created and has been developed by
many authors in the last twenty years, among them J.L. Lions, E. Gagliardo,
A.P. Calderon, S.G. Krein, and J. Peetre. Almost all information on this problem,
its applications and bibliography can be found in the following monographs:
P.L. Butzer and H. Berens [3], J.L. Lions and F. Magenes [7], J. Bergh and
J. Löfström [1], H. Triebel [27], S.G. Krein, Ju. I. Petunin, and E.M. Semenov [5].
Problem 87 is the first problem concerning nonlinear interpolation. Before giving
a positive solution of the problem we shall introduce some definitions which will
simplify the formulation of the problem and the theorems connected with this
problem.
The pair of spaces (A, B) is called a semi-Lipschitz (Lipschitz) interpolation pair
with constant C > 0 between the pairs (A0 , B0 ) and (A1 , B1 ) if for any operator U
(possibly nonlinear) which maps Ai into Bi , i = 0, 1 and satisfies the conditions
Ua − UbB0 ≤ M0 a − bA0 a, b ∈ A0

UaB1 ≤ M1 aA1 a ∈ A1
[ Ua − UbB1 ≤ Mi a − bAi a, b ∈ Ai , i = 0, 1 ]
then U (or its appropriate unique extension, denoted as U instead of Û) maps A into
B and
UaB ≤ C max(M0 , M1 )aA a∈A

[ Ua − UbB ≤ C max(M0 , M1 )a − bA a, b ∈ A ].
If Ai = Bi , i = 0, 1 and A = B, we can say shortly that the space A is a semi-Lipschitz

(respectively Lipschitz) interpolation space with constant C > 0 between A0 and A1 .
Using these formulations, Banach’s problem can be presented as follows:
Is Lγ [0, 1] a semi-Lipschitz interpolation space

with constant C > 0 between Lβ [0, 1] and Lα [0, 1] for (B)

any γ , 1 ≤ β < γ < α ≤ ∞?
Banach notes that he proved this theorem for α = ∞; it is not known whether this
proof was published.
In [11] the problem of the Riesz-Thorin theorem for Lipschitz operators is
considered. The problem in this particular case has the form:
Is Lγ [0, 1] a Lipschitz interpolation space with

constant C > 0 between Lβ [0, 1] and Lα [0, 1] (M)
for any γ , 1 ≤ β < γ < α ≤ ∞?
W. Orlicz [19] proved that if 0 < z < ∞ then any Orlicz space LΦ (0, z) with the
Luxemburg norm
!
z |f (t)|
f LΦ = inf r > 0 : Φ dt ≤ 1
0 r
is a semi-Lipschitz and Lipschitz interpolation space with constant C > 0 between

L1 (0, z) and L∞ (0, z). This contains an answer for problems B and M if β = 1 and
α = ∞. Next, G.G. Lorentz and T. Shimogaki in [10] have given a generalization of
Orlicz’ theorem, replacing β = 1 by β ≥ 1. Following G.G. Lorentz, T. Shimogaki
and W. Orlicz, we shall prove the following theorem.
Theorem 1. Let Φ (u) = 0u (u − t)β dξ (t), u > 0, where 1 ≤ β < ∞ and ξ is a
positive nondecreasing left continuous function with ξ (0) = 0. Then
(i) LΦ (0, z) is a semi-Lipschitz interpolation space with constant 1 between
Lβ (0, z) and L∞ (0, z);
(ii) LΦ (0, z) is a Lipschitz interpolation space with constant 1 between Lβ (0, z)
and L∞ (0, z) if 0 < z < ∞;
(iii) LΦ (0, ∞) is a semi-Lipschitz interpolation space with constant 1 between
Lβ (0, ∞) and L∞ (0, ∞) if Φ satisfies Δ2 -condition;
Proof. (i) (Lorentz-Shimogaki [10]). We may assume that Mβ = M∞ = 1. Then
by Fubini’s theorem,
z |Uf (t)|
!
β
Φ (|Uf (t)|)dt = (|Uf (t)| − s) dt dξ (s)
0 0 Es
∞
!
β
= (|Uf (t)| − s) dt dξ (s),
0 Es
where Es = { t ∈ (0, z) : |Uf (t)| > s } for s > 0. We consider an s-truncation f (s) of
the function f :

(s) f (t), if |f (t)| ≤ s,
f =
s |ff (t)
(t)| , if |f (t)| > s.
Since |Uf | − |Uf |(s) = |Uf − (Uf )(s) |, then
z ∞
!
(s) β
Φ (|Uf (t)|)dt = |Uf (t) − (Uf ) (t)| dt dξ (s)
0 0 Es
∞ z
!
= |Uf (t) − (Uf )(s) (t)|β dt dξ (s)
0 0
∞
β
= Uf − (Uf )(s) Lβ dξ (s).
0
From the assumption we have U(f (s) )L∞ ≤ f (s) L∞ ≤ s, hence the inequality
|Uf − (Uf )(s) | ≤ |Uf − U(f (s) )| a.e. (see [23, 14]) follows. Then
z ∞
β
Φ (|Uf (t)|)dt ≤ Uf − U(f (s) )Lβ dξ (s)
0 0
∞ z
β
≤ f − f (s) Lβ dξ (s) = Φ (|f (t)|)dt
0 0
or
Uf LΦ ≤ f LΦ , f ∈ LΦ .
Taking U0 f = (Uf )/ max(Mβ , M∞ ), we obtain the theorem for the general case.
Proof. (ii) and (iii) ([19], see also [12] and [15]). We take any fixed f0 ∈ Lβ (0, z)∩
L∞ (0, z) and define an operator T by
U(f + f0 ) − Uf0
Tf = , f ∈ Lβ (0, z) ∪ L∞ (0, z).
max(Mβ , M∞ )
Now T satisfies assumptions (i) for f ∈ Lβ (0, z). Hence
Tf LΦ ≤ f LΦ , f ∈ LΦ (0, z) ∩ Lβ (0, z).
This means that
Uf − Uf0 LΦ = max(Mβ , M∞ )T(f − f0 )LΦ

≤ max(Mβ , M∞ )f − f0 LΦ
for f ∈ LΦ (0, z) ∩ Lβ (0, z). Let z < ∞.

For arbitrary f , g ∈ LΦ (0, z), we consider the truncations f (n) , g(n) . Then U(f (n) )
and U(g(n) ) converge to Uf and Ug, respectively, in the Lβ (0, z)-norm. Therefore,
for a properly chosen sequence ni , U(f (ni ) ) and U(g(ni ) ) converge almost every-
where to Uf and Ug. Since f (n) ∈ L∞ (0, z) and |f (n) − g(n) | ≤ |f − g|, by the Fatou
property of the Luxemburg norm we obtain
Uf − UgLΦ ≤ lim U(f (ni ) ) − U(g(ni ) )LΦ

i→∞
≤ max(Mβ , M∞ )f − gLΦ .
If z = ∞ and Φ satisfies Δ2 -condition, then Lβ (0, ∞) ∩ L∞ (0, ∞) is dense in LΦ (0, ∞)

and U can be uniquely extended to LΦ (0, ∞).
Since Φ (u) = uγ for β ≤ γ < ∞ has a representation as Φ in theorem 1, we obtain

a positive answer to problem B and M for α = ∞.
In [9, 12, and [13]], there are generalizations of the Orlicz and Lorentz-
Shimogaki theorems. Those generalizations replace Orlicz spaces with rearrange-
ment invariant spaces. Further considerations on the interpolation of Lipschitz
operators of weak type can be found in [23] and [14].
Now we shall deal with any Banach space. For 0 < ω < 1 and 1 ≤ p ≤ ∞, let
(A0 , A1 )θ ,p;K = { a ∈ A0 + A1 :
∞
aθ ,p;K ≡ (t−θ K(t, a))p dt/t)1/p < ∞ },
0
where
K(t, a) ≡ K(t, a; A0 , A1 )
= inf{ a0 A0 + ta1 A1 : a0 ∈ A0 , a1 ∈ A1 , a = a0 + a1 }.
The real method presented above is called the K-method. The properties of this
interpolation space are characterized in [1, 3, 27]. There also exist other real
methods such as the trace method and the mean method, which are equivalent to the
K-method. The method which, in general, is not equivalent to the K-method is the
complex method introduced independently by J.L Lions and A.P. Calderon (see [1]).
We shall prove that the K-method can be applied to nonlinear interpolation.
Theorem 2 (Lions [6], Peetre [20], see also [15]). Let A1 ⊂ A0 and B1 ⊂
B0 . For all θ , 0 < θ < 1 and all p, 1 ≤ p ≤ ∞ the pair of Banach spaces
((A0 , A1 )θ ,p;K , (B0 , B1 )θ ,p;K ) is a semi-Lipschitz interpolation pair with constant 1
between the pairs (A0 , B0 ) and (A1 , B1 ).
The proof is immediate and is obtained by showing that K(t, Ua; B0 , B1 ) ≤
max(M0 , M1 )K(t, a; A0 , A1 ), for t > 0. J.L. Lions [6] proved this theorem using the
trace method with the additional assumptions: B0 is reflexive and U : A1 → B1 is a
continuous operator. Theorem 2 (stated as above) was proved by J. Peetre [20] and
myself [15]. Further generalizations of Theorem 2, together with its applications,
can be found in [20, 25, and [15]].
Using s-truncation, it can be shown that (Lβ , Lα )θ ,γ ;K = Lγ with equivalent
norms, where 0 < β < α < ∞, 1/γ = (1 − θ )/β + θ /α and 0 < θ < 1 (see [1],
th. 5.2.1). This fact and theorem 2 gives us a positive answer to problem B.
J.L. Lions [6, 21] put forward the following problem:
Does Theorem 2 hold for the complex

interpolation method? (L)
A negative solution with one operator and one Banach space, where Ai = Bi , i = 0, 1,
was obtained by M. Cwikel [4]; a negative solution for a family of operators and a
family of Banach spaces can be found in [15]. Arriving at this solution was difficult,
since for many important and well-studied Banach pairs, the complex interpolation
spaces coincide with suitably chosen real interpolation spaces. In [15] it is shown
that with an additional assumption that the operator U : A0 → B0 is differentiable
in the sense of Fréchet, Lions’ problem has a positive solution.
The theory of nonlinear interpolation has not been studied thoroughly. Hence,
some problems, such as the two given below, emerge.
Problem 1. Does the assumption on the continuity of the operator U : A1 → B1
in Theorem 2 imply continuity of the operator U : (A0 , A1 )θ ,p;K → (B0 , B1 )θ ,p;K ?
Problem 2. Does Lions’ problem have a positive solution with the assumption
of:
(a) differentiability of the operator U : A0 → B0 in the sense of Gateaux;
(b) Lipschitz condition on the operator U : A1 → B1 ?
If the spaces Ai , Bi , i = 0, 1 satisfy some additional conditions, Problem 1 has a

positive solution (see [15, th. 7.5]).
Theorem 3 ([15, th. 5.3]). Let A1 ⊂ A0 and B1 ⊂ B0 . For any θ , 0 < θ < 1
and any p, 1 ≤ p < ∞, the pair of Banach spaces ((A0 , A1 )θ ,p;K , (B0 , B1 )θ ,p;K ) is a
Lipschitz interpolation pair with constant 1 between the pairs (A0 , B0 ) and (A1 , B1 ).
Proof. (The method is based on the paper of W. Orlicz [19]). For any fixed a1 ∈ A1
we set for a ∈ A, Ta = U(a + a1 ) − Ua1 . Then
Ta − TbB0 = U(a + a1 ) − U(b + a1 )B0

≤ M0 a − bA0
for a, b ∈ A0 , and
TaB1 = U(a + a1) − Ua1 B1

≤ M1 aA1
for a ∈ A1 . From Theorem 2 we get
Taθ ,p;K ≤ max(M0 , M1 )aθ ,p;K
for a ∈ (A0 , A1 )θ ,p;K . Hence,
Ua − Ua1 θ ,p;K = T(a − a1 )θ ,p;K

≤ max(M0 , M1 )a − a1 θ ,p;K .
Because p < ∞, A1 is dense in (A0 , A1 )θ ,p;K (see [1, th. 3.4.2(b)]). Taking any a, b ∈
(A0 , A1 )θ ,p;K there are sequences (an ), (bn ) in A1 convergent to a and b, respectively,
in the · θ ,p;K norm. Hence,
Ua − Ubθ ,p;K ≤ Ua − Uan θ ,p;K + Uan − Ubn θ ,p;K

+ Ubn − Ubθ ,p;K
≤ max(M0 , M1 )( a − an θ ,p;K + an − bn θ ,p;K
+ bn − bθ ,p;K )
≤ max(M0 , M1 )(2a − an θ ,p;K + a − bθ ,p;K
+ 2bn − bθ ,p;K ).
Taking n → ∞ we get Ua − Ubθ ,p;K ≤ max(M0 , M1 )a − bθ ,p;K , a, b ∈

(A0 , A1 )θ ,p;K . From Theorem 3 we obtain a positive solution of problem M.
The papers in [8, 2, and [15]] contain some general considerations on the subject
of Lipschitz operator interpolation in rearrangement invariant spaces or arbitrary
Banach spaces.
1. J. Bergh, J. Löfström, Interpolation spaces. An introduction, Springer, Berlin, 1976.
2. F. Browder, Remarks on nonlinear interpolation in Banach spaces, J. Functional Anal., 4
(1969), 390–403.
3. P.L. Butzer, H. Berens, Semi-groups of operators and approximation, Grundlehren Math. Wiss.
145, Berlin, 1967.
4. M. Cwikel, A counterexample in nonlinear interpolation, Proc. Amer. Math. Soc., 62 (1977),
62–66.
5. S.G. Krein, Jǔ.I. Petunin, E.M. Semenov, Interpolation of linear operators (in Russian), Nauka,
Moscow, 1978.
6. J.L. Lions, Some remarks on variational inequalities, Proc. Internat. Conf. Functional Analysis
and Related Topics (Tokyo 1969), Univ. of Tokyo Press, Tokyo, 1970, 269–282.
7. J.L. Lions, F. Magenes, Problèmes aux limites non homogènes et applications (in Russian),
Mir, Moscow, 1971.
8. G.G. Lorentz, T. Shimogaki, Interpolation theorems for operators in function spaces, J.
Functional Anal., 2 1968), 31–51.
9. , Majorants for interpolation theorems, Publ. Ramanujan Inst., 1 (1969),
115–122.
10. , Interpolation theorems for the pairs of spaces (Lp , L∞ ) and (L1 , Lq ), Trans.
Amer. Math. Soc., 159 (1971), 207–221.
11. L. Maligranda, Riesz Thorin theorem for Lipschitz operators, V Session of Functional Analysis,
Poznan 1978 (abstract will appear in Functiones et Approximatio X).
12. , Interpolation of Lipschitz operators for the pairs of spaces (L1 , L∞ ) and (l1 , c0 ),
Comment. Math. Prace Mat., 21 (1979), 327-339.
13. , Interpolation of Lipschitz operators for the pairs of spaces (Lp , L∞ ) and (lp , c0 ),
0 < p < ∞, Functiones et Approximatio 9 (1980), 107-115.
14. , A generalization of the Shimogaki theorem, Studia Math., 71 (1981), 69-83.
15. , Interpolation of nonlinear operators in Banach spaces, Thesis, University of A.
Mickiewicz Poznan 1979 (in Polish).
16. J. Marcinkiewicz, Sur l’interpolation d’opérations, C. R. Acad. Sci. Paris, 208 (1939),
1272–1273.
17. W. Orlicz, Ein Satz über die Erweiterung von linearen Operationen, Studia Math., 5 (1934),
127–140.
18. , On a class of operations over the space of continuous vector-valued functions,
Studia Math., 14 (1954), 285–297.
19. , On a class of operations over the space of integrable functions, Studia Math.,
14 (1954), 302–309.
20. J. Peetre, Interpolation of Lipschitz operators and metric spaces, Mathematica (Cluj), 12
(1970), 325–334.
21. Problems in interpolation of operators and applications (Problem list of the Special Session on
Interpolation of Operators and Applications), Notices of the AMS, 22 (1975), 124–126.
22. M. Riesz, Sur les maxima des formes bilinéares et sur les fonctionelles linéares, Acta Math.,
49 (1926), 465–497.
23. T. Shimogaki, An interpolation theorem on Banach function spaces, Studia Math., 31 (1968),
233–240.
24. E.M. Stein, G. Weiss, An extension of a theorem of Marcinkiewicz and some of its applications,
J. Math. Mech., 8 (1959), 263–284.
25. L. Tartar, Interpolation non linéare et régularité, J. Functional Anal., 9 (1972), 469–489.
26. G.O. Thorin, An extension of a convexity theorem due to M. Riesz, Kungl. Fysiogr. Sällsk. i
Lund Förh., 8 (1938), 166–170.
27. H. Triebel, Interpolation theory, function spaces, differential operators, North-Holland, Ams-
terdam, 1978.
28. A. Zygmund, On a theorem of Marcinkiewicz concerning interpolation of operations, J. Math.
Pures Appl., 35 (1956), 223–248.
Lech Maligranda
The Mathematical Institute of the Polish Academy of
Sciences, Poznań, Poland
PROBLEM 88: MAZUR

Given is a sequence of real numbers (an ) with the property that for every
bounded sequence (xn ) the series |a1 x1 + a2 x2 + . . . + an xn + . . . | + |a2 x1 + a3 x2 +
. . . + an+1 xn + . . . | + . . . + |am x1 + am+1 x2 + . . . + am+n−1 xn + . . . | + . . . converges. Is
the series
∞
∑ n|an |
n=1
convergent?
Remark: If sequences of numbers (a1n ), (a2n ), . . . , (amn ) are given with the
property that for every bounded sequence of numbers (xn ) the series |a11 x1 +a12 x2 +
. . . | + |a21 x1 + a22 x2 + . . . + a2n xn + . . . | + . . . + |am1 x1 + am2 x2 + . . . + amn xn + . . . | +
. . . converges, then, according to a remark by Mr. Banach, the series
∞
∑ (|am1 | + |am2 | + . . . + |amn | + . . .)
m=1
can diverge.
This problem was solved in negative by S. Kwapień and A. Pełczyński, The main
triangle projection in matrix spaces and its applications, Studia Math. 34 (1970)
43–68. It is closely related to Problem 8. Via duality arguments its negative solution
provides a negative solution to Problem 8. For more details we refer to the papers
from References in Commentary to Problem 8.
PROBLEM 89: MAZUR

Let W be a convex body located in the space (L2 ), and such that its boundary Wb
does not contain any interval; let xn ∈ W, (n = 1, 2, . . .), x0 ∈ Wb , and in addition
let the sequence (xn ) converge weakly to x0 . Does then the sequence (xn ) converge
strongly to x0 ? It is known that this statement is true in the case where W is a sphere.
Examine this problem for the case of other spaces.
The first part of this problem can be answered in negative by the following example :
let W = {(αn ) ∈ 2 : ∑∞ ∞
n=1 αn /n ≤ ∑n=1 αn /n} and let xn = en , n ∈ N be the standard
2
base sequence in 2 , x0 = 0. The answer to the second question in Problem 89 is:

whenever in Banach space there is weakly convergent sequence which is not norm
convergent we can produce a similar example without the property.
PROBLEM 90: ULAM, AUERBACH

It is known that every semisimple Lie group (e.g., the projective group in n
variables) contains four elements generating a dense subgroup. Can one lower the
number 4?
Commentary
Every connected semisimple Lie group G has a free subgroup with two free
generators which is dense in G (see [1]). Related results and references are given
in [2].
1. M. Kuranishi, On everywhere dense imbeddings of free groups in Lie topological groups,
Nagoya Math. J. 2 (1951), 63–71.
2. J. Mycielski, Almost every function is independent, Fund. Math., 81 (1973), 43–48.
J. Mycielski
PROBLEM 91: MAZUR

A convex body W with a center is given, in the n-dimensional Euclidean space. it
is affine to its conjugate body. Is W then an ellipsoid? The answer is negative in the
case when n is an even number; for odd n the problem is not solved. It is equivalent
to this: If a space of type (B) of n dimensions is isometric to its conjugate space, is
it then isometric to the Euclidean space?
Commentary
The answer is negative for odd dimensions as well. If the dimension is n = 3,

a simple example due to K. Leichtweiss (Zur expliziten Bestimmung der
Norm der selbstadjungieren Minkowksi-Räume, Resultate der Mathematik,1
(1978), 61–87) is obtained by taking as W the convex polyhedron given
in a cartesian system of coordinates as the convex hull of the eight points
±(1, 1, 1), ±(1, 1, −1), ±(1, 0, 0), ±(0, 1, 0). The conjugate (dual, polar) convex
polyhedron W ∗ has as vertices the points ±(−1, 1, −1), ±(1, −1, −1), ±(0, 1, 0),
±(1, 0, 0). Thus W is isometric to W ∗ , although it clearly is not an ellipsoid.
Analogous examples can be constructed in all odd dimensions ≥ 3.
Branko Grünbaum
PROBLEM 92: MAZUR

Given is a bounded sequence of numbers (sn ). There exist sequences of numbers
(n ) with the property that:
(1) n > 0 (n = 1, 2, . . .);
(2) 1 + 2 + . . . = ∞;
(3) The sequence (1 s1 + . . . + n sn )/(1 + . . . + n ) converges.
Do there exist sequences (n ) which, in addition to properties (1), (2), and (3),
satisfy the condition:
(4a) The sequence (n ) is fully monotonic; that is, all the differences Δn1 = n −
n+1 , Δn2 = Δn1 − Δn+1
1 , . . . are nonnegative;
or only the condition:

(4b) The sequence (n ) is nonincreasing.
If two sequences are given (n ), (n ) which satisfy the conditions (1), (2), (3),
(4a), or merely (1), (2), (3), (4b), then can the limits
1 + . . . + n sn
lim
n→∞ + . . . + n
1
and
1 + . . . + n sn
lim
n→∞ + . . . + n
1
be different?
Addendum. There exist sequences (n ), (n ) satisfying conditions (1), (2), (3),
(4b) such that for a certain bounded sequence sn composed of 0s and 1s, the two
limits
1 + . . . + n sn
lim
n→∞ + . . . + n
1
and
1 + . . . + n sn
lim
n→∞ + . . . + n
1
exist but are different.
Mazur
August 10, 1935
PROBLEM 93: MAZUR

Let R be a planar set. The system of functions x = f (t), y = g(t) (0 ≤ t ≤ 1) is
called a parametric description of the set R, if the set of points (f (t), g(t)) is identical
with R. Assume that for a given set R there exists a parametric description x = f1 (t),
y = g1 (t) for which the functions f1 (t), g1 (t) are continuous and there also exists
another parametric description x = f2 (t), y = g2 (t) where the functions f2 (t), g2 (t)
are of bounded variation; does there exist a parametric description of R: x = f (t), y =
g(t) so that the functions f (t), g(t) are simultaneously continuous and of bounded
variation? Assume that for a given set R there exist parametric descriptions x = f (t),
y = g(t) for which the functions are of bounded variation and continuous—for every
such description, we determine the length d(f (t), g(t)) of the set R and we take the
lower bound of the numbers d(f (t), g(t)) denoted by d; does there exist a parametric
description of R: x = f0 (t), y = g0 (t) also with functions of bounded variation and
continuous and such that d(f0 (t), g0 (t)) = d? The same problem in the case of the
n-dimensional Euclidean space.
Addendum.∗ The theorem is true; we can represent R by functions x = f (t), ξ
y = gξ (t), continuous and of bounded variation, in such a way that the length of the
curve (by Jordan’s definitions) is at most twice the Carathéodory measure of R
A.J. Ward
March 23, 1937
∗ Original manuscript in English.
Commentary
For references to the early work of Gołab and Ważewski, where the solution
announced here is proved, as well as related work, see V. Faber, J. Mycielski and
P. Pedersen, On the shortest curve which meets all the lines which meet a circle,
Ann. Polon. Math. 44 (1984), 249–266.
Jan Mycielski
PROBLEM 94: Z. LOMNICKI, ULAM

Let limn→∞ kn /n = f < 1, where always kn < n. Prove that
P 0, P < f
lim x1 · · · xkn (1 − xkn +1 ) · · · (1 − xn )dx1 · · · dxn dp =
n→∞ 0 K 1, P ≥ f
where
K = { (x1 , . . . , xn ) : x1 + . . . + xn = nP, 0 ≤ xi ≤ 1, i = 1, . . . , n }.
Compare Problem 17.

Addendum. This conjecture was proved by S. Bochner in April, 1936—he even

gave the order of convergence. A paper on this topic will appear in Annals of Math.
S. Ulam 1936
Commentary
The problem is better formulated as follows:

Let (ξi , Xi ), i = 1, 2, . . . be independent random vectors, where each ξi is (a priori)
uniformly distributed in (0, 1) and each Xi takes the value 1 with probability ξi and
the value 0 with probability 1 − ξi . Thus the two components of each vector are
dependent. It is easy to compute the conditional expectations below:
1
0 ξ ξ dξ 2
E{ ξ | Xi = 1 } = 1
= ;
0 ξ dξ
3
(1)
1
0 ξ (1 − ξ )d ξ 1
E{ ξ | Xi = 0 } = 1
= .
0 ξ dξ
3
In fact, owing to the stated independence, if Fi denotes the σ -field of all Xj for j ≥ 1
except Xi , the conditional expectations above are not affected if we adjoin Fi to the
two conditions shown. Therefore,

1 N 2 m 1 N −m
∑ E{ ξ | X1 + . . . + XN = m } = 3 N + 3 N .
N i=1
This is the curious discovery of Bochner (see [1], Formula (10)).

Given all Xi , i ≥ 1, the random variables { ξ , i ≥ 1 } are independent with a
posteriori expectations determined by the Xi ’s as shown in (1). Hence, if (m/N) → t,
then

1 N 1 N
1
P ∑ ξi → 3 (1 + t) N ∑ Xi → t = 1
N i=1
(2)
i=1
by the classical law of large numbers (Borel’s form is sufficient). Bochner gave the
following explicit analytic formula (cf. his (3) which is really the same):

1 N N

P ∑ ξi ≤ p ∑ Xi = m
N i=1 i=1
= 2N ∑
ξ1 · · · ξm (1 − ξm+1 ) · · · (1 − ξN )dξ1 · · · dξN .
1≤i≤N ξi ≤pN
0≤ξi ≤1
Therefore, if (m/N) → t, the latter integral converges to 0 if p ≤ (1/3)(1 + t) and

to 1 if p > (1/3)(1 + t). (Note: The equality case is included because the integral
has the same value if the constraint ∑Ni=1 ξi ≤ pN is changed to ∑Ni=1 ξi < pN.)
This result proved by Bochner by the method of Fourier transforms. He pointed
out that, contrary to what might be facilely conjectured, the critical value of p is
(1/3)(1 + t) and not t. (Only when t = 1/2 do these two values coincide, and the
original problem concerns only this case.) The latter is indeed the critical value if
the ξi ’s take the values 1 or 0 with probability 1/2 each. This is seen by reevaluating
the conditional expectations in (1) under the new a priori distributions for the ξi s,
as Bochner did. (When the ξi ’s are constants the result reduces to Problem 17 which
can be done by bounded convergence.) His improvement of (2) in obtaining the
speed of convergence and a central limit theorem can also be derived, presumably,
by applying (by now) well-known limit theorems to the sequence of independent
ξi ’s, but using their a posteriori moments. Surely it was a remarkable achievement
in 1936.
K.L. Chung
1. S. Bochner, A converse of Poisson’s theorem in the theory of probability Annals of Math., 37

(1936), 816–822.
PROBLEM 95: SCHREIER, ULAM

Is the group R of real numbers (under addition) isomorphically contained in the
group S∞ of all permutations of the sequence of natural integers?
Addendum. The answer is affirmative.
Schreier, Ulam November, 1935
Commentary
The rationals Q can be embedded into S∞ by letting Q act on itself by left

translations. The direct product of countably many copies of S∞ can be embedded
into S∞ since the integers may be decomposed into countably many disjoint infinite
subsets. Hence, a countable direct product of Q’s may be embedded into S∞ . This
last direct product, however, is isomorphic to the reals under addition since both
are vector spaces of the same dimension over Q. Schreier and Ulam noted this
argument; Schreier and Ulam also asked if every Lie group can be embedded (as an
abstract subgroup) into S∞ . In particular, can SO(3) be embedded into S∞ ? It is easy
to check that this question has an affirmative answer if and only if SO(3) has a
subgroup of countable index. It is unknown at this time whether SO(3) has such a
subgroup. It is simple to check that the analogous question for second countable
connected locally compact groups reduces to the connected Lie group case, for
any second countable connected locally compact group is a projective limit of a
sequence of Lie groups. It is also simple to check that any totally disconnected
second countable locally compact group can be embedded as an abstract group
into S∞ , for any such group has a neighborhood basis of the identity consisting of
compact open subgroups. The embedding may be constructed by letting the group
act on a suitably chosen sequence of (countable discrete) quotient spaces.
Robert R. Kallman
The obvious generalization of Problem 95 was stated by Ulam ([5], p. 58) as follows:
“. . . is every Lie group isomorphic (as an abstract group) to a subgroup of the group
S∞ ?” In case one regards any discrete group to be a zero-dimensional Lie group,
one should perhaps refine Ulam’s question to ask if every Polish (i.e., complete
metrizable separable) Lie group can be abstractly embedded into S∞ ? As noted in
the original commentary, Schreier and Ulam proved that the additive group of the
reals can be embedded into S∞ .
A certain amount of progress has been made on this problem. In fact if F is any
field whose cardinality is less than or equal to 2ℵ0 , then GL(n, F) is isomorphic to
a subgroup of S∞ ([3]). Therefore any connected matrix Lie group can be abstractly
embedded into S∞ . In particular, any semisimple complex analytic group can be
abstractly embedded into S∞ since any such group has a faithful finite dimensional
representation ([1], p. 200, Theorem 3.2). The same is true for any simply connected
solvable analytic group ([1], p. 219, Theorem 3.1) or, more generally, any semidirect
product G = B η H, where H is a reductive analytic group and B is a simply
connected solvable analytic group, normal in G ([1], p. 223, Theorem 4.3).
However, not every Lie group is isomorphic to a matrix Lie group. For example,
the simply connected cover SL(2, R) of SL(2, R) has center isomorphic to (Z, +)
but any semisimple Lie group with faithful continuous matrix representation has
finite center ([1], Proposition 4.1, p. 221). More generally, any nontrivial covering
group of SL(2, R) has no continuous faithful finite dimensional representation ([1],
Exercise 1, p. 210) and the same is true for a group covered by the Heisenberg group
([1], Exercise 1, p. 225). Therefore the most general case of Ulam’s extrapolation
of Problem 95 does not follow from the matrix result and the general Lie group
question remains unanswered.
Consider the following two questions:
Question #1: Let G be a group with a finite central cyclic subgroup Z such that
G/Z can be injected into S∞ . Can G be injected into S∞ ? In particular, is this true if
Z = Z2 ?
Question #2: Let G ⊂ S∞ be a subgroup and let Z be a central cyclic subgroup of G.
Can G/Z be injected into S∞ ?
If one is optimistic and believes that these two questions have affirmative answers,
then Ulam’s general Lie group question has an affirmative answer. To see this,
recall that the fundamental group of every connected Lie group is a finitely
generated abelian group. The fundamental theorem of finitely generated abelian
groups implies that such a group is a finite direct sum of cyclic groups. Therefore
every connected Lie group is of the form G/Z, where G is a simply connected
Lie group and Z is a finitely generated central subgroup of G. On the other hand,
Ado’s theorem ([2]) implies that every finite dimensional Lie algebra has a faithful
finite dimensional representation. Therefore basic facts about Lie groups imply that
every simply connected Lie group covers a matrix Lie group, which, of course, can
be injected into S∞ . In this same circle of ideas, recall that if G is a matrix group
and Z ⊂ G is a finite central subgroup, then G/Z has a faithful finite dimensional
representation ([1], Lemma 3.1, p. 199) and therefore can be injected into S∞ . On
the other hand, and more generally, it is an easy exercise to prove that if G ⊂ S∞
is a subgroup and N ⊂ G is a finite normal subgroup, then G/N can be injected
into S∞ .
Again, if one is optimistic and believes that Ulam’s general Lie group question
has an affirmative answer, one might well ask if still more might be true. For
example, can every locally compact Polish group be algebraically injected into
S∞ ? This has a ring of plausibility since every locally compact group has an
open subgroup which is a projective limit of Lie groups ([4], Yamabe’s Main
Approximation Theorem, p. 175). The left regular representation of a locally
compact group G is a homeomorphism onto its (closed) range in the unitary group
of the Hilbert space L2 (G). Therefore one might further ask if the unitary group of a
separable Hilbert space can be algebraically injected into S∞ ? Unfortunately this is
asking too much, for it is known that any algebraic injection of the unitary group of
an infinite dimensional separable Hilbert space into a separable topological group
is continuous. It is an easy consequence of the spectral theorem that the unitary
group is connected in the strong operator topology. But there is no continuous
injection of the connected unitary group into the totally disconnected group S∞ .
The homeomorphism group of [0, 1] cannot be injected into S∞ for similar reasons.
A fortiori the same is true for the universal Polish groups the homeomorphism group
of the Hilbert cube and the isometry group of the universal Urysohn space.
1. Gerhard Hochschild, The Structure of Lie Groups, Holden-Day, San Francisco, 1965.
2. Nathan Jacobson, Lie Algbras, Dover, New York, 1979, ISBN 0-486-63832-4.
3. Robert R. Kallman, Every reasonably sized matrix group is a subgroup of S∞ , Fundamenta
Mathematicae, volume 164, Issues 1–2, 2000, pp. 35 – 40.
4. Dean Montgomery and Leo Zippin, Topological Transformation Groups, Robert E. Krieger,
New York, 1974.
5. S.M. Ulam, Problems in Modern Mathematics, Wiley, New York, 1964 (first published under
the title A Collection of Mathematical Problems, Wiley, New York, 1960).
Robert R. Kallman
PROBLEM 96: ULAM

Can the group S∞ of all permutations of integers be so metrized that the group
operation (composition of permutations) is a continuous function and the set S∞
becomes, under this metric, a compact space? (locally compact?)
Addendum. One cannot metrize in a compact way.
Schreier, Ulam November, 1935
Commentary
Two solutions of generalizations of this problem have appeared in the literature.

Gaughan [1] showed that there is no nontrivial, locally bounded, Hausdorff topo-
logical group structure on S∞ . Kallman [2] showed that S∞ has a unique topology
(the usual one) under which it is a complete separable metric group.
1. E.D. Gaughan, Topological Group Structures of Infinite Symmetric Groups, Proceedings of the
National Academy of Sciences U.S.A., 58 (1967), 907–910.
2. R.R. Kallman, A Uniqueness Result for the Infinite Symmetric Group, Studies in Analysis,
Advances in Mathematics Supplemental Studies, 4 (1979), 321–322.
Robert R. Kallman
PROBLEM 97: KURATOWSKI, ULAM
Two sets (spaces) A AND B are called quasihomeomorphic if, for every ε ,
there exists a continuous mapping fε of the space A onto the space B such that
the counterimages are smaller than ε (that is to say, from |x − x | > ε it follows
that f (x ) = f (x )) and, conversely, a continuous mapping gε with counterimages
smaller than ε of the space B onto the space A. Problem: Are two manifolds
(topological spaces such that every point has a neighborhood homeomorphic to
the n-dimensional Euclidean sphere) which are quasihomeomorphic, of necessity
homeomorphic?
Commentary
For dimensions strictly greater than 4, the problem has an affirmative solution. In
fact, the following much stronger statement is true:
Theorem [5] If Mn , n ≥ 5, is a compact manifold without boundary, then there
is an ε > 0 such that if f : Mn → Nn is a map with diamf −1 (x) < ε for each x ∈ Nn ,
then f is homotopic to a homeomorphism.
The proof uses techniques of Siebenmann [11] and Chapman and Ferry [3] which
arise out of the general Kirby-Siebenmann [8] approach to the study of topological
manifolds. Further developments in this general area are due to Chapman and to
Quinn [10].
In dimensions ≤ 2 the problem is easily seen to be true. In dimension 3, work
of Waldhausen [12], Armentrout [1], and Siebenmann [11] is certainly relevant; see
also Hamilton’s paper [7]. While the details have not been worked out, this work,
together with the approach of [3] and [5] may well add up to a proof of Problem 97
modulo the Poincaré conjecture in dimension 3. Nothing seems to be known in
dimension 4.
More interestingly, no counterexample seems to be known to the following: If X

and Y are compact ANRs and for each ε > 0 there are surjective ε -maps f : X → Y
and g : X → Y, then X and Y are homeomorphic. Perhaps a reasonable first step
would be to show that X and Y are homotopy equivalent. [2, 4, 6], and [9] are
relevant. Compare the commentary to Problem 21.
1. S. Armentrout, Cellular decompositions of 3-manifolds that yield 3-manifolds, Bull. Am. Math.
Soc. 75 (1969), 453–456.
2. I. Bernstein and T. Ganea, Remark on spaces dominated by manifolds, Fund. Math., 47 (1959),
45–56.
3. T.A. Chapman and S. Ferry, Approximating homotopy equivalences by homeomorphisms,
American J. of Math. 101 (1979), 583–607.
4. S. Eilenberg, Sur les transformations á petites tranches, Fund. Math., 30 (1938), 92–95.
5. S. Ferry, Homotoping ε -maps to homeomorphisms, Amer. J. of Math., 101 (1979), 567–582.
6. T. Ganea, On ε -maps onto manifolds, Fund. Math., 47 (1959), 35–44.
7. A.J.S. Hamilton, The triangulation of 3-manifolds, Q. J. Math. Oxford, 27 (1976), 63–70.
8. R.C. Kirby and L.C. Siebenmann, On the triangulation of manifolds and the Hauptvermutung,
Bull. Amer. Math. Soc. 75 (1969), 742–749.
9. S. Mardešic and J. Segal, ε -mapping onto polyhedra, Trans. Amer. Math. Soc. 109 (1963),
146–164.
10. F. Quinn, Ends of maps. I, Ann. Math. 110 (1979), 275–331.
11. L.C. Siebenmann, Approximating cellular maps by homeomorphisms, Topology 11 (1972),
271–294.
12. F. Waldhausen, On irreducible 3-manifolds which are sufficiently large, Ann. Math., 87 (1968),
56–88.
Steve Ferry, The Institute for Advanced Study

and The University of Kentucky

Do there exist a finite number of analytic transformations of the n-dimensional
sphere into itself, f1 , . . . , fn , such that by composing these transformations a finite
number of times, one can approximate arbitrarily any continuous transformation
of the sphere into itself? How is it for one-to-one transformations? (Analytic here
means differentiable any number of times.)
Remark
The first problem is still open. The second problem about homeomorphisms has
a positive answer—see J. Schreier and S. Ulam, Über topologische Abbildungen
der euklidischen Sphären, Fund. Math., 23 (1934), 102–118. Further comments
are in Stanislaw Ulam, Sets, Numbers, and Universes; Selected Works, edited by
W.A. Beyer, J. Mycielski, and Gian-Carlo Rota, in the series Mathematicians of
Our Time, MIT Press, Cambridge, Mass., 1974.
PROBLEM 99: ULAM

By a product set in the unit square, we understand the set of all pairs (x, y) where
x belongs to a given set A, y to a given set B. Do there exist sets which cannot be
obtained through the operations of forming countable sums and differences of sets
starting from product sets? Do there exist nonprojective sets with respect to product
subsets?
Commentary
If the continuum hypothesis or Martin’s axiom holds, then the answer to the first
question if no [2, 6]. In fact, under either of these assumptions, we have Rσ δ =
P(I 2 ), where R = { A × B : A, B ⊆ [0, 1] = I }.
If every subset of I 2 is generated from R by the operations of forming countable
unions and differences, then all sets are generated by some countable stage [1].
A. Miller [5] has shown that the stage at which all sets are generated can be any
ordinal α , 2 ≤ α < ω1 . This problem has some interesting connections to other
problems of set theory, for example, the existence of Q-sets [4] and whether the
continuum is real-valued measurable [3]. In connection with this it is unknown
whether a universal analytic set is in B(R), the Borel field generated by R, if
the continuum is real-valued measurable.
The situation regarding the second question does not seem to be clear. It also
seems to be unknown whether every subset of I 2 can be analytic with respect to R
and yet B(R) = P(I 2 ).
1. R.H. Bing, W.W. Blesdoe, R.D. Mauldin, Sets generated by rectangles, Pac. J. Math., 51 (1974),
27–36.
2. K. Kunen, Inaccessibility properties of cardinals, Ph.D. Thesis, Department of Mathematics,
Stanford University, August 1968.
3. R.D. Mauldin, Countably generated families, Proc. Amer. Math. Soc. 54 (1976), 291–297.
4. , On rectangles and countably generated families, Fund. Math. 95 (1977),
129–139.
5. A. Miller, On the length of Borel heirarchies, Ann. Math. Log. 16 (1979), 233–267.
6. B.V. Rao, On discrete Borel spaces and projective sets, Bull. Amer. Math. Soc., 75 (1969),
614–617.
R. Daniel Mauldin
Second edition Commentary
Let S (R) be the family of sets obtained by applying the Souslin operation to sets
in B(R)). Miller [3] Theorem 6: It is relatively consistent with ZFC that S (R) =
P(I 2 ) and B(R) = P(I 2 ). This answers the last question of the commentary
above.
In regard to the second question: “Do there exist nonprojective sets with respect
to product subsets?” We might define the projective sets with respect to product
subsets to be the smallest family of sets containing
{A1 × A2 × · · · × An : n < ω and Ai ⊆ R}
and closed under complementation, countable unions, and projection. In his thesis
[2] Kunen showed that in the random real model or Cohen real model that
B(R) = P(I 2 ) by showing that no well-ordering of the reals is in B(R). For
a different argument for why B(R) = P(I 2 ) in the Cohen real model, see Miller
[3] Remark 4 page 180. It is also true that by combining results of Rothberger
[4] and Bing, Bledsoe, Mauldin [1], that certain cardinal arithmetic, for example,
2ω = ℵ2 + 2ω1 = ℵω1 , implies that B(R) = P(I 2 ), see Miller [3] Remark 5 page
180. I am not sure if any of these arguments can be generalized to show that it is
consistent that not every subset of the plane is a projective set with respect to the
product subsets.
1. Bing, R. H.; Bledsoe, W. W.; Mauldin, R. D.; Sets generated by rectangles. Pacific J. Math. 51
(1974), 27–36.
2. Kunen, Kenneth; Inaccessibility Properties of Cardinals. Thesis (Ph.D.) Stanford University.
1968. 124 pp, ProQuest LLC
3. Miller, Arnold W.; Generic Souslin sets. Pacific J. Math. 97 (1981), no. 1, 171–181.
4. Rothberger, Fritz; A remark on the existence of a denumerable base for a family of functions.
Canadian J. Math. 4, (1952). 117–119.
Arnold W. Miller, January 2015
PROBLEM 100: ULAM, BANACH

Let Z be a closed set contained in the surface of the n-dimensional sphere. Does
there exist a sequence of homeomorphic mappings of the surface of the sphere onto
itself, converging to a mapping of the surface onto Z?
Addendum. For n = 2, affirmative answer by Borsuk.
Solution
First, we notice that the answer is yes if Z is a singleton, e.g., for the case of the circle
S1 , the iterates of the map e2π it → e2π it have the required property for Z = {1}. For
Sn we take an appropriate fibration into circles and do a similar thing.
If Z has more than one point, then by choosing an appropriate coordinatization
of Sn we can assume that the north pole and the south pole are in Z. Then let h move
each point p down toward the south pole along a great circle by the angular distance
(1/2)dist(p, Z). Then it is easy to check that h is continuous, that h(p) = p for each
p ∈ Z, and that hm (p) → Z as m → ∞ for each p ∈ Sn . To check that h is one-to-one,
we note that the great circles of our movement meet only at the poles. Also, if p
and q are on the same great circle, and p is above q, then clearly h(p) = h(q), if the
arc from p to q meets Z. If the arc from p to p does not meet Z, then dist(p, Z) <
2dist(p, q) + dist(q, Z). Thus h is one-to-one.
Jan Mycielski
PROBLEM 101: ULAM
A group U of permutations of the sequence of integers is called infinitely
transitive if it has the following property: If A and B are two sets of integers, both
infinite and such that their complements to the set of all integers are also infinite,
then there exists in the group U an element (permutation) such that f (A) = B. Is the
group U identical with the group S∞ of all permutations?
Solution
The answer is no. Let G be the group of all permutations of N = {1, 2, 3, . . .} such
that for each π ∈ G there is a finite partition P of N such that π is order preserving
on each set in P. It is clear that G has the desired property, since there exists a π ∈ G
which maps A onto B preserving order and N − A onto N − B preserving order. On
the other hand, G is a proper subgroup of S∞ since a permutation which reverses the
order in all the blocks n2 , n2 + 1, . . . , (n + 1)2 − 1 does not belong to G.
A. Ehrenfeucht
Note
S. Ulam informs us that C. Chevalley had found the solution and a number of
interesting related results shortly after the end of the second world war.
PROBLEM 102: ULAM

(a) Let ε be a positive number; p and q two points of the unit square. In the first case
let the point p be fixed and q wander at random. In the other case, assume that
both points move at random. Is the probability of approach of the two points p
and q within a distance ≤ ε of each other, after n steps, greater in the first case
than in the second?
(b) Let a, b denote two rotations of a circle of radius 1 through angles a, b. Let ε be a
positive number. We define a set of pairs Eε1 (a, b) as follows: Two rotations a, b
belong to it if (na − b) mod 2π is smaller than ε earlier than (na − nb) mod 2π ;
that is to say, for smaller n than is the case for (na − nb) mod 2π . We denote by
Eε2 (a, b) the complement of the set of pairs Eε1 (a, b) with respect to the set E of
all pairs. Which of the two sets Eε 1 (a, b), Eε2 (a, b) has greater measure? (Show
that asymptotically these sets have equal measures.)

Does there exist a separable group S, universal for all locally compact groups?
(That is, a group such that every locally compact group should be continuously
isomorphic with a subgroup of it?) The authors deduced from J. von Neumann’s
representation of compact groups the existence of a compact group, universal for all
compact groups.
Commentary
This problem, if taken literally, is false by cardinality considerations. A better

formulation of the question might be the following: Does there exist a Polish
group S such that every second countable locally compact group G is continuously
isomorphic to a subgroup of S? The answer to this question is yes. The existence of
Haar measure on such a G shows that G is continuously isomorphic to a subgroup
of U(H), the unitary group on the complex separable infinite dimensional Hilbert
space H, by considering the left regular representation of G on L2 (G). It will follow
from results given later that S cannot be locally compact and second countable.
This problem has a number of variants and perturbations, most of which are
not quite so easy to settle; some of them are considered below. Unless otherwise
noted, G will denote a typical locally compact group with a countable basis
for its topology. The prerequisites for the following discussion may be found in
Montgomery and Zippin’s book on topological groups, Hochschild’s book on Lie
groups, and Kaplansky’s book on infinite abelian groups.
Is there a locally compact group S such that every compact group is isomorphic
to a subgroup of S? This is false by cardinality considerations. However, there is
a compact metric group S such that every compact metric group is continuously
isomorphic to a subgroup of S. Take S to be the product of countably many copies
of U(n), for each positive integer n, where U(n) is the unitary group on a complex
n-dimensional Hilbert space. This is the Peter-Weyl theorem.
Is there a locally compact group S such that S/S0 is compact and such that
every connected G is isomorphic as an abstract group to a subgroup of S? The
answer is no. In fact there is no such S for all G which are noncompact centerless
simple Lie groups. Let K be the maximal compact normal subgroup of S. S/K is
a Lie group whose connected component of the identity is of finite index. For any
G, either G intersects K only in the identity or G is contained in K. In the latter
case, let π be any finite dimensional unitary representation of K. Then π (G) is the
identity or is isomorphic with G. Choose a π such that the latter holds. Now any
solvable subgroup of a compact connected Lie group has an abelian subgroup of
finite index. But any noncompact simple Lie group G has a solvable subgroup with
no abelian subgroup of finite index. Hence, G intersects K only in the identity for
all noncompact simple Lie groups. Hence,! if S exists, we may assume that S is a
Lie group such that S/S0 is finite. Since each G is simple and S/S0 is finite, each
G is contained in S0 . Hence, we may assume that S is a connected Lie group. Each
G intersects the radical of S in the identity, so we may further assume that S is a

centerless semisimple Lie group. But this is impossible, for there is a finite upper
bound on the length of a maximal solvable series of subgroups of S, but the length
of a maximal solvable series of subgroups of an arbitrary noncompact simple G has
no upper bound.
Note that there is a countable abelian group S such that every countable abelian
G is isomorphic to a subgroup of S. Take S to be the direct sum of countably many
copies of Q and the Zp,∞ ’s. However, there is no countable group S such that every
countable group G is isomorphic to a subgroup of S, for the number of two-element
subsets of such an S is countable, but the number of nonisomorphic countable groups
with two generators is uncountable.
Finally, is there a Polish group S such that every Polish group G may be injected
continuously into S? Is there a Polish group S such that every Polish group G
is abstractly isomorphic to a subgroup of S? Is there a second countable locally
compact group S so that every countable G is isomorphic to a subgroup of S? For this
last question such an S cannot be connected since the group of all finite permutations
Sf has a simple subgroup of index 2, either Sf is contained in a compact group or
Sf ∩ K = (e). In the latter case, there is an injection of Sf to the Lie group S/K.
Now in both cases there is therefore a faithful matrix representation of Sf . But this
is a contradiction, for any nontrivial representation of Sn occurs on a vector space of
dimension of least n, and Sf = n≥1 Sn . A slight modification of this argument also
shows that S cannot have the property that S/S0 is compact.
Robert R. Kallman
It is now known ([1]) that every Polish group is topologically isomorphic to a closed
subgroup of the homeomorphism group of the Hilbert cube. It is also known that
every Polish group is topologically isomorphic to an isometry group of some Polish
space and that every isometry group of a Polish space is topologically isomorphic
to a closed subgroup of the isometry group of the universal Urysohn space ([2]).
1. Vladimir Vladimirovich Uspenskij, A universal topological group with a countable base,
Functional Analysis and Its Applications, volume 20, 1986, pp. 160–161.
2. Vladimir Vladimirovich Uspenskij, On the group of isometries of the Urysohn universal metric
space, Commentationes Mathematicae Universitatis Carolinae, volume 31, number 1, 1990,
pp. 181–182.
Robert R. Kallman

Let f (x, y, z, p, q) denote a function of five variables possessing a sufficient
number of derivatives and satisfying the inequalities: f > M(|p|z+a + |q|z+a ); M
constant, a > 0.
One has to find a minimum of the integral where z = z(x, y), p = zx , and q = zy :
f (x, y, z, p, q)dx dy (1)

Ω
(the region Ω should be sufficiently regular), among all z which possess all the
first, possibly also the second continuous derivatives, and which assume the same
values on the boundary. One may assume that the given boundary value has a
given number of derivatives with respect to the arc length of the boundary curve.
Expression (1) is assumed regular. A similar condition for free boundary conditions.
Prove the existence of a function, minimizing in a given class. (Regular problem:
fpp fqq − 4fpq > 0.)

The question is to find a system of functions x(u, v), y(u, v), z(u, v) minimizing
the parametric variational problem

xu yu

· · · f (x, y, z, X, Y, Z)du dv, X = , etc. (2)
xv yv
corresponding to Problem 104. It is allowed to change the class of admissible

functions; these could be, for example, functions which are absolutely continuous
in the sense of Tonelli. If not, then the problem is not solved. Mazur and Schauder
solved Eq. (2) in the case when f does not contain x, y, z explicitly (even without any
conditions analogous to those in Problem 104) but only within the class of functions
absolutely continuous in the sense of Tonelli. Even this case (x, y, x does not appear)
was not solved for functions x(u, v), . . . , z(u, v) sufficiently regular.
PROBLEM 106: BANACH

Prize: One bottle of wine, S. Banach
Let
∞
∑ xi
i=1
be a series [xi are elements of a space of type (B)] with the property that under a
certain ordering of its terms the sum = y0 , under some other ordering, equals y1 .
Prove that for every real number there exists an ordering of the given series such
that the sum of it will be: y0 + (1 − )y1 . In particular, consider the case where xi
are continuous functions defined on the interval (0, 1). The convergence according
to norm means uniform convergence.
Addendum. This does not hold in the space L2 and also not in C. We define, for
every n, 2n functions fn,i (x) as follows:
i−1 i
fn,i (x) = 1, <x< if i = 2n
2n 2n
i−1 i
fn,i (x) = −1, <x< if i = 2n
2n 2n
fn,i (x) = 0 otherwise.
Consider the orderings:
0 ≡ f1,1 + f1,3 + f1,2 + f1,4 + f2,1 + f2,22 +1 + f2,2

+ f2,22 +2 + f2,3 + f2,22 +3 + f2,4 + f2,22 +4 + . . .
1 ≡ f1,1 + f1,2 + f1,3 + f2,22 +1 + f2,22 +2 + f1,4
+ f2,22 +3 + f2,22 +4 + . . .
Since the fi,k assume integer values, one cannot order the series in such a way that it
converges in L2 to 0 < < 1.
Marcinkiewicz
Commentary
The addendum does not completely make sense as it stands. In particular, the defi-
nition of fn,i for i > 2n seems not to apply for (0, 1), if this is the interval the writer
of the addendum is considering. Problem 106 inquires about the generalization to
Banach spaces of the theorem of Steinitz [5] which asserts that if a series of vectors
in Rm , ∑ vi , is convergent, then its sums (allowing all orderings of the terms) form
a flat in Rm . For early discussions of the generalization of Steinitz’s theorem to
abstract spaces, see Wald [6], Hadwiger [2, 3], and Pracher [4]. As is stated by
Damsteeg and Halperin [1], Steinitz’s theorem follows from the theorem that for
ci ≥ 0, i = 1, 2, and positive integer m there is a finite constant Km (c1 , c2 ) with
the property: whenever, for any n, the m-dimensional vectors u1 , v1 , v2 . . . , vn , u2
satisfy |u1 | ≤ c1 , |u2 | ≤ c2 , |vi | ≤ 1 for all i and u1 + ∑ni=1 vi + u2 = 0, then by
reordering the vi it is possible to satisfy |u1 + ∑hi=1 vi | ≤ Km (c1 , c2 ) for h = 1, 2, . . . , n.
Assume now that Km (c1 , c2 ) denotes the least possible such constant. Obviously,
Km (c̄1 , c̄2 ) ≥ Km (c1 , c2 ) if c̄1√≥ c1 and c̄2 ≥ c2 . Damsteeg and Halperin [1] have
proved that Km (0, 0) ≥ (1/2) m + 3 and thus that this method of proof of Steinitz’s
theorem cannot be used to generalize Steinitz’s theorem to Hilbert space.
One can adjust the definition given in the addendum to provide a counterexample.
However, perhaps the following geometric description, due to Israel Halperin is
more transparent.
For every half-open interval J = [a, b), let JL = [a, (a + b)/2) and JR = [(a +
b)/2, b). Also, let χ (J) denote the characteristic function of J. Now, let I = [0, 1).
Then IL , IR , ILL , ILR , . . . are determined by the above definitions.
Consider two sequences (different orderings of the same elements):
(s1 ) χ (I), −χ (I), χ (IL ), −χ (IL ), χ (IR ), −χ (IR ), χ (ILL ),

−χ (ILL ), χ (ILR ), −χ (ILR ), χ (IRL ), −χ (IRL ), . . .
and
(s2 ) χ (I), −χ (I), χ (IL ), χ (IR ), −χ (IL ), χ (ILL ), χ (ILR ),

−χ (IR ), χ (IRL ), χ (IRR ), −χ (ILL ), χ (ILLL ), χ (ILLR ), . . .
The first sequence sums to 0 in every Lp , 0 < p < ∞. This is easily seen by grouping
each odd term with its successor. The second sequence sums to 1 in every Lp .
This can be seen be grouping the terms in s2 after the first term into sets of three
consecutive terms:
−χ (I), χ (IL ), χ (IR ), −χ (IL ), χ (ILL ), χ (ILR )
and noticing that the sum of the three terms in each set is zero.
The final statement in the addendum states a reason why no rearrangement of
this series converges to any constant function , 0 < < 1. Finally, since L2 can be
embedded in C, this gives an example in C.
1. I. Damsteeg and I. Halperin, The Steinitz-Gross theorem on sums of vectors, Trans. Royal Soc.
Canada Sect. III. 44, series 3 (1950), 31–35.
2. H. Hadwiger, Über das Umordnungsproblem im Hilbertschen Raum, Math. Zeit., 46 (1940),
70–79.
3. , Über das Umordnungsstärke und eine Erweiterung des Steinitzschen Satzes,
Math. Ann., 118 (1943), 702–717.
4. K. Prachar, Über bedingt konvergente Vektorreihen im Banach’schen Raum, Monatsh. Math.,
54 (1950), 284–307.
5. E. Steinitz, Bedingt konvergente Reihen und konvexe Systeme, Journal Reine u. Angew. Math.,
143 (1913), 128–175.
6. A. Wald, Reihen in topologischen Gruppen, Ergebnisse eines Math. Koll. Wien (1933), 59.
R. Daniel Mauldin
W.A. Beyer
The issue of Marcinkiewicz’s addendum to Problem 106 was resolved by the Soviet
probabilist Kornilov who was able to reconstruct what Marcinkiewicz was talking
about. The example is similar to that constructed by Israel Halperin and included
in the Commentary of Mauldin and Beyer. This example was used by V. M.
Kadec (Functional Anal. Appl 20 (1986)) to solve Banach’s problem completely
when he was able to surgically implant a suitable sequence of finite dimensional
versions of this example in any infinite dimensional Banach space. Kadec’s main
tool ( other than his personal cleverness) was the celebrated Dvoretzky Spherical
Sections theorem which says that any infinite dimensional Banach space contains
subspaces of arbitrarily large dimension that are nearly isometric to Euclidean
spaces of the same dimension. Kadec’s theorem was generalized by W. Banaszczyk
(J. Angew. Math. 403 (1990), 187–200 and Studia Math. 107 (1993), 213–222); the
upshot of Banaszcyk’s work is that the Levy-Steinitz theorem holds in a (locally
convex) Fréchet space if and only if that space is nuclear. Remarkably, the key
to Banaszczyk’s proof is a sharpening of Dvoretzky’s result due to V. Milman
(Proc. AMS 94 (1985), 445–449). Alas, a companion to Banaszczyk’s Fréchet space
theorem was uncovered by J. Bonet and A. Defant (Israel J. math. 117 (2000), 131-
156), who investigated what the situation is in spaces dual to nuclear Fréchet spaces.
A beautiful exposition of the original Levy-Steinitz theorem is due to Peter
Rosenthal ( Amer. Math. Monthly 94 (1987), 342–351) while the little book of M.
I. Kadets and V. M. Kadets (Series in Banach Spaces, Birkhauser Verlag, 1987) is
chock full of interesting examples and the full proof of V. M. Kadets’ result as well
as a proof of Dvoretzky’s result. Regarding the Dvoretzky result no better source can
be found than the Springer Lecture Notes in Mathematics volume of V. D. Milman
and G. Schechtman Asymptotic Theory of Finite Dimensional Normed Spaces.
Joe Diestel
Kent, Ohio
PROBLEM 107: STERNBACH

Does there exist a fixed point for every continuous mapping of a bounded plane
continuum E, which does not cut the plane, into part of itself? The same for
homeomorphic mappings of E into all of itself.
Commentary
This problem was well known when Sternbach recorded it in the Scottish Book.
In a recent conversation Professor Kuratowski mentioned that he, Mazurkiewicz,
and Knaster first considered this problem in the late 1920s. Ayres [1] proved in
1930 that every locally connected nonseparating plane continuum has the fixed-
point property for homeomorphisms. In 1932 Borsuk [1] introduced the concept
of a retract to prove that every locally connected nonseparating plane continuum
has the fixed-point property (for all continuous functions). In 1938 Hamilton [17]
showed that this problem is related to the notion of an indecomposable continuum.
Let G be a bounded simply connected plane domain whose closure does not separate
the plane and whose boundary is hereditarily decomposable. Hamilton proved that
the closure of G has the fixed-point property for homeomorphisms. Bell [2] in
1967 and Sieklucki [26] in 1968 proved that every nonseparating plane continuum
that has a hereditarily decomposable boundary has the fixed-point property. In
1972 Hagopian [20] extended his theorem to every nonseparating plane continuum
with the property that every pair of its points can be joined by a hereditarily
decomposable subcontinuum.
In 1951 Cartwright and Littlewood [16] proved that every homeomorphism of a
nonseparating plane continuum onto itself that can be extended to an orientation-
preserving homeomorphism of the plane has a fixed point. Recently Bell [3] proved
that every homeomorphism of a nonseparating plane continuum onto itself that can
be extended to the plane has a fixed point.
The question whether a nonseparating plane continuum has the fixed point
property for mappings is one of the most famous unsolved problems in plane
topology. There was a rekindling of interest in this problem when Bellamy [5]
constructed an example of a locally planar tree-like continuum without the property.
However, in spite of concerted efforts, we have been unable to convert Bellamy’s
example to a nonseparating plane continuum without the fixed point property for
mappings.
R. H. Bing
The fixed point problem for a planar continuum, nonseparating the plane, is one of
the most persistent unsolved problems in topology [7]. K. Borsuk proved that there
is no generalization to higher dimensions: there exists a cellular continuum in R3
without the fixed point property [12].
The planar problem seems to have been considered already by L.E.J. Brouwer
in [14]; E.R. Reifenberg [25] reconstructs Brouwer’s proof of the Cartwright-
Littlewood theorem in [25]. In 1912, Brouwer [15] proved his famous “Transla-
tionssatz” from which it follows that an orientation preserving homeomorphism of
the plane such that the orbit of some point is bounded must have a fixed point.
In 1977, Morton Brown [13] used the Brouwer Translation Theorem to give a
very concise proof of the Cartwright-Littlewood theorem, but not Bell’s theorem
for arbitrary homeomorphisms of the plane, without the assumption of preserving
orientation. The Brouwer Translation Theorem does not generalize to orientation
reversing homeomorphisms of the plane. In 1981, S.M. Boyles [9] constructed such
homeomorphism without a fixed point and the orbit of every point bounded. It is still
not known whether a nonseparating plane continuum has the fixed point property
even with respect to homeomorphisms of the continuum.
In 1982, Bell announced a version of the Cartwright-Littlewood theorem for
analytic maps. In 2013, A. Blokh, A., R. Fokkink, J. Mayer, L. Oversteegen, and
E. Tymchatyn [8] extended the theorem to compositions of open and monotone
maps.
Brouwer also considered a locally connected “circular continuum” invariant
under an orientation reversing homeomorphism of the plane showing that the
homeomorphism possesses two fixed points in the continuum. This theorem has
been extended to any continuum in the plane with two invariant complementary
domains by K. Kuperberg [21], and further by J.P. Boroński [10], who proved that
if h is a planar orientation reversing homeomorphism of the plane, the continuum X
is invariant under h, and there are at least n + 1 complementary domains, then the
set of fixed point of h in X has at least n + 1 components.
A very significant result obtained independently by Bell [3], K. Sieklucki [26],
and S.D. Iliadis [22] is that a nonseparating plane continuum without the fixed
point property must contain an indecomposable continuum in the boundary. In 1971,
C. Hagopian [19] proved that an arcwise connected, nonseparating plane continuum
possesses the fixed point property. His 1972 paper generalizes this result and also
the Bell-Iliadis-Sieklucki theorem.
O.J. Hamilton [18] proved in 1951 that chainable (arc-like) continua have the
fixed point property with respect to continuous maps. In 1990, P. Minc extended
the theorem to weakly chainable planar continua [23]. Having interest in showing
that there is just one pseudo-arc (a homogeneous chainable continuum) Bing [6]
showed that all chainable continua are planar. The tree-like continuum without the
fixed point property constructed by Bellamy is not embeddable in the plane. In
1999, Minc gave an example of a non-planar weakly chainable tree-like continuum
without the fixed point property [24]. The fixed point problem for planar, tree-like
continua is still open.
1. W.L. Ayres, Some generalisations of Scherrer fixed-point theorem, Fund. Math., 16 (1930),
332–336.
2. H. Bell, On fixed point properties of plane continua, Trans. Amer. Math. Soc., 128 (1967),
539–548.
3. , A fixed point theorem for plane homeomorphism, Bull. Amer. Math. Soc., 82
(1976), 778–780.
4. H. Bell, A fixed point theorem for plane homeomorphisms, Fund. Math., 100 (1978), 119–128.
5. D.P. Bellamy, A tree-like continuum without the fixed point property, Houston J. of Math., 6
(1980), 1–13.
6. R. H. Bing, Snake-like continua, Duke Math. J., 18 (1951), 653–663.
7. R. H. Bing, The elusive fixed point property, Amer. Math. Monthly, 76 (1969), 119–132.
8. A. Blokh, A., R. Fokkink, J. Mayer, L. Oversteegen, and E. Tymchatyn, Fixed point theorems
for plane continua with applications, Mem. Amer. Math. Soc., 224 (2013), no. 1053.
9. S.M. Boyles, A counterexample to the bounded orbit conjecture, Trans. Amer. Math. Soc., 266
(1981), 415–422.
10. J. P. Boroński, Fixed points and periodic points of orientation-reversing planar homeomor-
phisms, Proc. Amer. Math. Soc., 138 (2010), 3717–3722.
11. K. Borsuk, Einige Satze über stetige Streckenbilder, Fund. Math., 18 (1932), 198–213.
12. K. Borsuk, Sur un continu acyclique qui se laisse transformer topologiquement en lui même
sans points invariants, Fund. Math., 24 (1935), 51–58.
13. M. Brown, A short proof of the Cartwright-Littlewood theorem, Proc. Amer. Math. Soc., 65
(1977), 372.
14. Brouwer, L. E. J., Über eineindeutige, stetige Transformationen von Flächen in sich, Math.
Ann., 69 (1910), 176–180.
15. Brouwer, L. E. J., Beweis des ebenen Translationssatzes, Math. Ann., 72 (1912), 37–54.
16. M.L. Cartwright and J.E. Littlewood, Some fixed point theorems, Ann. of Math., 54 (1951),
1–37.
17. O.H. Hamilton, Fixed points under transformations of continua which are not connected im
kleinen, Trans. Amer. Math. Soc., 44 (1938), 18–24.
18. Hamilton, O. H., A fixed point theorem for pseudo-arcs and certain other metric continua, Proc.
Amer. Math. Soc., 2 (1951), 173–174.
19. C.L. Hagopian, A fixed point theorem for plane continua, Bull. Amer. Math. Soc., 77 (1971),
351–354.
20. , Another fixed point theorem for plane continua, Proc. Amer. Math. Soc., 31
(1972), 627–628.
21. K. Kuperberg, Fixed points of orientation reversing homeomorphisms of the plane, Proc. Amer.
Math. Soc., 112 (1991), 223-229.
22. S.D. Iliadis, Positions of continua on the plane, and fixed points, Vestnik Moskov. Univ. Ser. I
Mat. Meh., 25 (1970), 66–70.
23. P. Minc, A fixed point theorem for weakly chainable plane continua, Trans. Amer. Math. Soc.,
317 (1990), 303–312.
24. P. Minc, A weakly chainable tree-like continuum without the fixed point property, Trans. Amer.
Math. Soc., 351 (1999), 1109–1121.
25. E.R. Reifenberg, On the Cartwright-Littlewood fixed point theorem, Ann. of Math., 61 (1955),
137–139.
26. K. Sieklucki, On a class of plane acyclic continua with the fixed point property, Fund. Math.,
63 (1968), 257–278.
K. Kuperberg
Lex Oversteegen
PROBLEM 108: BANACH, MAZUR, ULAM
Let E be a space of type (B) which has a basis and H a set everywhere dense
in E.
(1) Does there exist a basis whose terms belong to H?
(2) The same question under the additional assumption that the set H is linear.
Addendum. Affirmative answer.
Krein

October 16, 1935
Given are n functions of a real variable: f1 , . . . , fn . Denote by R(f1 , . . . , fn ) the
set of all functions obtained from the given functions through rational operations
(expressions of the form

k
∑ ak1 ···kn f1 1 · · · fnkn
k k
.
∑ bk1 ···kn f1 1 · · · fn n
Must there always exist, in the set R, a function f such that its indefinite integral
does not belong to the set R?
An analogous question in the case where we include in the set R all the functions
obtained by composing functions belonging to R.
Addendum. An affirmative answer for the first question was found by Docents,
Dr. S. Kaczmarz and Dr. A. Turowicz.
March, 1938
Commentary
The solution to the first question appears as S. Kaczmarz and A. Turowicz, “Sur
l’irrationalité des intégrales indefinies,” Studia Mathematica 8 (1939), 129–134.
Their solution is more general and is the following: Let fi (x), 1 ≤ i < ∞, be an
infinite sequence of functions of a real variable that are assumed to be finite and
summable in an interval (a, b) (−∞ ≤ a < b ≤ ∞). Denote by Z the functions
g(x) = R(f1 (x), f2 (x), . . . , fn (x)) where R(y1 , . . . , yn ) is an arbitrary rational function
with real coefficients and n is arbitrary. A function g(x) in Z is defined for all x in
(a, b) for which the denominator is not zero. Then for each interval (α , β ) such that
a < α < β < b there exists a function g0 (x) in Z such that (1) g0 (x) is finite and
x
summable in (α , β ), (2) the function F(x) = −∞ g0 (t)dt is not in Z. If a and b are
both infinite, one can take α = a and β = b.
The title of the paper is well chosen. The theorem states that if one starts with
a countable collection F of functions and denotes by Z the closure of F under the
operation of forming rational functions of functions in F, then there are functions in
Z whose integrals are not in Z. That is, integrals of functions in Z may be “irrational.”
The proof depends on two lemmas.
Lemma 1. If rj (xi ; 1 ≤ i ≤ k), 1 ≤ j ≤ k + 1, are rational functions of k variables,
then there exists a polynomial G(yi ; 1 ≤ i ≤ k + 1) of k + 1 variables which is not
identically zero such that G(ri (xj ; 1 ≤ j ≤ k); 1 ≤ i ≤ k + 1) ≡ 0.
Lemma 2. Suppose Q(log |x − c1 |, log |x − c2 |, . . . , log |x − cn |) ≡ 0 for all x in
(α , β ), Q being a polynomial in n variables and ci being outside (α , β ). Then
Q(y1 , . . . , yn ) ≡ 0.
The result of Kaczmarz and Turowicz seems to be unrelated to the theory of
Liouville-Ritt-Risch of integration in finite terms. Also, the theorem appears to be
nonconstructive in the sense that a function g0 (x) is not exhibited. The reviewer is
not aware of any work on the second question.
W.A. Beyer
The phrasing of Problem 109 is somewhat vague, and is subject to interpretation.

Also, it is possible that the original question formulated by Mazur and Ulam was
different. Indeed, in [1] Kaczmarz and Turowicz quote the first part of Problem 109
as follows.
Let f1 (x), . . . , fn (x) be a set of n real functions continuous in an interval (a, b),
and let Z be the set of the functions of the form R(f1 (x), . . . , fn (x)), where R denotes
an arbitrary rational function of n variables. Does there exist in Z a function which
is summable in (a, b) and whose indefinite integral does not belong to Z?
That is, according to Kaczmarz and Turowicz, in the original problem the func-
tions f1 , . . . , fn were continuous. In order to formulate the possible interpretations of
the problem, let us start with fixing the ideas.
Let R denote the set of all real-valued functions defined on subsets of the real
line. We shall denote by Fc the family of all functions f : I → R such that I is a
subinterval of R and f is continuous on I. The domain of f ∈ R is denoted by D(f ).
We define addition, multiplication, division, and composition on R as usual with
the obvious conventions D(f + g) = D(f · g) = D(f ) ∩ D(g), D(f /g) = D(f ) ∩ {x ∈
D(g) : g(x) = 0} and D(f ◦ g) = {x ∈ D(g) : g(x) ∈ D(f )}.
For every family F ⊂ R we denote by E(F ) the set of all functions obtained
from F and from the constant functions through addition, multiplication, division,
and composition.
The second question of Problem 109 asks if for every finite system F ⊂ R (or
F ⊂ Fc ) there exists an f ∈ E(F ) such that the indefinite integral of f does not
belong to E(F ). It is not specified what we should mean by the indefinite integral
of f . In the Kaczmarz-Turowicz variant the function f is summable on (a, b), so
it is very likely that by the indefinite integral of f Mazur and Ulam meant the
integral function x → xx0 f dx. But we could also interpret the indefinite integral as
the primitive (antiderivative) of f . In this case we have to assume that f is defined
on an interval. Summing up, we interpret the problem as follows.
Is it true that for every finite system F ∈ R (or F ∈ Fc ) there exists an f ∈ E(F )
such that f is defined a.e. on an interval I, f is summable on I, and its integral
function on I does not belong to E(F ) (or, f is defined on an interval I, f has an
antiderivative F on I and F does not belong to E(F ))?
In the sequel we answer three of these questions and give a partial answer to the
fourth one.
I. First we show that if there are no restrictions on the functions fi , then the answer
is negative, for both interpretations of the indefinite integral. We prove that there
are functions f , g ∈ R such that Fc ⊂ E(f , g). Since the integral functions and the
antiderivatives belong to Fc , this will solve the problem in the negative.
Let A and B be subsets of R such |A| = |B| = 2ω and the map (x, y) → x + y
(x ∈ A, y ∈ B) is injective. (Let, e.g., A be the set of all numbers in [0, 1] whose
decimal expansion only contains the digits 0 and 1, and let B = 2A.) Let g be an
injective map from R into A.
The cardinality of the family Fc is of the continuum, and thus there is a bijection
φ from B onto Fc . We put X = {(x, y) ⊂ R2 : y ∈ B, x ∈ D(φ (y))}, and define
f (g(x) + y) = φ (y)(x) ((x, y) ∈ X).
Since the map (x, y) → g(x) + y is injective on X, it follows that f is well defined on
the set {g(x) + y : (x, y) ∈ X}.
Let h ∈ Fc be arbitrary. If h = φ (y), then we have h(x) = f (g(x) + y) for every
x ∈ D(h). This proves that h ∈ E(f , g).
II. We show that if the functions f1 , . . . , fn are supposed to be continuous, and if

we interpret the indefinite integral as the antiderivative of a function defined on
an interval, then the answer to the question is again negative. We shall prove the
following.
Theorem 1. There are continuous functions f1 , . . . , f6 ∈ C[0, 1] and
f7 , f8 : R → R such that whenever f ∈ E(f1 , . . . , f8 ) is defined on an interval, then the
indefinite integral of f belongs to E(f1 , . . . , f8 ).
Note that if f ∈ E(f1 , . . . , f8 ) is defined on an interval I, then f is continuous on I,
and thus the integral function of f is the same as the antiderivative of f .
The proof of Theorem 1 is based on the following observation.
Theorem 2. There are continuous functions g, φ1 , . . . , φ5 ∈ C[0, 1] with the follow-
ing property. For every Lipschitz function f : [0, 1] → R there are constants α , β and
γ1 , . . . , γ5 such that
5
f (x) = α · ∑ g (β φi (x) + γi )
i=1
for every x ∈ [0, 1].

Proof. If f ∈ R, A ⊂ D(f ) and f is Lipschitz on A, then we shall denote by Lip(f ; A)
the infimum of the numbers K such that |f (x) − f (y)| ≤ K · |x − y| for every x, y ∈ A.
The set
K = {f ∈ C[0, 1] : |f (0)| ≤ 1 and Lip(f ; [0, 1]) ≤ 1}
is a compact, convex subset of the Banach space C[0, 1]. Then the intersection of K
with any open ball is also convex, hence connected. This shows that K is a compact
and locally connected subset of C[0, 1]. Therefore, by the Hahn-Mazurkiewicz-
Sierpiński theorem [3, Theorem 2, §50, II, p. 256], K is the continuous image of
[0, 1]. Let ψ be a continuous map from [0, 1] onto K. We define
Ψ (x, y) = ψ (y)(x) (x, y ∈ [0, 1]);
then Ψ is a continuous function defined on the unit square.

Now we apply a variant of Kolmogorov’s theorem representing continuous

functions of several variables using functions of one variable and addition [2,
Statement (5), p. 231]. We obtain that there are functions g, φ1 , . . . , φ5 ∈ C[0, 1] and
positive constants λ1 , λ2 such that
5
Ψ (x, y) = ∑ g (λ1 φi (x) + λ2 φi (y)) (1)
i=1
for every x, y ∈ [0, 1].

Let f : [0, 1] → R be Lipschitz. Then there is a constant ε > 0 such that ε · f ∈ K.
By the choice of the function Ψ , there is a y ∈ [0, 1] such that ε · f (x) = Ψ (x, y) for
every x ∈ [0, 1]. Writing ci for λ2 φi (y) (i = 1, . . . , 5), (1) gives
5
ε · f (x) = ∑ g (λ1 φi (x) + ci ) (x ∈ [0, 1]),
i=1
which completes the proof.
Proof of Theorem 1. Let g, φ1 , . . . , φ5 as in Theorem 2. We put f1 = g and fi = φi−1

for every i = 2, . . . , 6. In addition, we define f7 (x) = x and f8 (x) = arctan x for every
x ∈ R. We prove that the functions f1 , . . . , f8 satisfy the requirements. We put E0 =
E(f1 , . . . , f8 ).
Suppose that f ∈ E0 is defined on an interval I. Then f is continuous on I.
It is clear that the integral function of f is Lipschitz on every closed and bounded
subinterval of I. Therefore, it is enough to prove the following: if F is defined on an
interval I and if F is Lipschitz on every closed and bounded subinterval of I, then
F ∈ E0 .
First suppose that I = [a, b] is a closed, bounded interval. Then F is Lipschitz on
[a, b], and thus the function x → F((b − a)x + a) (x ∈ [0, 1]) is Lipschitz on [0, 1]. By
Theorem 1 we have
5
F((b − a)x + a) = α · ∑ f1 (β fi+1 (x) + γi )
i=1
for every x ∈ [0, 1]. Thus

5
1
F(x) = α · ∑ f1 β fi+1 · (x − a) + γi
i=1 b−a
for every x ∈ [a, b]. Therefore, we have

5
F = α · ∑ f1 (β fi+1 (δ f7 + λ ) + γi )
i=1
on [a, b], where δ = 1/(b − a) and λ = −a/(b − a). This proves F ∈ E0 .

Next suppose that I = (a, b) is a bounded open interval. Then F is Lipschitz on

the interval [a + 2−n , b − 2−n ] for every n ≥ n0 , where n0 is a positive integer with
2−n0 < (b − a)/2. Let
Mn = max{1 + |F(x)| : x ∈ [a + 2−n , b − 2−n ]} and
Kn = 1 + Lip(F; [a + 2−n , b − 2−n ])
for every n ≥ n0 . It is easy to construct a Lipschitz function h ∈ C[a, b] such that

h(a) = h(b) = 0, h(x) > 0 for every a < x < b,
|h(x)| ≤ 1/Kn if x ∈ [a + 2−n , a + 2−n+1 ] ∪ [b − 2−n+1 , b − 2−n ] and
Lip(h; [a + 2−n , a + 2−n+1 ]) = Lip(h; [b − 2−n+1 , b − 2−n ]) < 1/Mn
for every n ≥ n0 . Let

F(x) · h(x) if x ∈ (a, b),
k(x) =
0 if x = a or x = b.
Then h and k are both Lipschitz functions on [a, b], and thus, as we proved above,
they belong to E0 . Since F = k/h, we obtain F ∈ E0 .
If I = (a, b] or I = [a, b), then an obvious one sided modification of the previous
argument shows that F ∈ E0 .
Next suppose that I = R. Then the function G(x) = F(tan x) is Lipschitz on every
closed subinterval of (−π /2, π /2). Therefore, we have G ∈ E0 . Thus we have F =
G ◦ f8 ∈ E0 .
A similar argument applies if I = [0, ∞) (or I = (0, ∞)). Put J = [0, π /2) (or J =
(0, π /2)) and H(x) = F(tan x) for every x ∈ J. Then H is Lipschitz on every closed
subinterval of J. Therefore, we have H ∈ E0 . Thus we have F = H ◦ f8 ∈ E0 .
Using suitable linear substitutions, one can check the statement in the cases when
I is one of [a, ∞), (−∞, b], (a, ∞) or (−∞, b).
III. Now we turn to the case when the functions fi are supposed to be continuous, and
we interpret the indefinite integral as the integral function of a summable function
defined a.e. on an interval. This case seems to be the most difficult among the
variants, and we only can give a partial solution. We show that the answer to the
corresponding question is affirmative, even if we start with a countable system of
continuous functions, assuming that some elementary functions are among them.
The precise statement is the following.
Theorem 3. For every countable set of continuous functions F ⊂ Fc there is a

function f ∈ E(F ∪ {ex , log x, sin x, arcsin x}) such that f is defined on R except one
point, f is summable on R, and the integral function of f does not belong to E(F ∪
{ex , log x, sin x, arcsin x}).
Proof. Let F ⊂ Fc be countable, and put F1 = F ∪ {ex , log x, sin x, arcsin x} and
E1 = E(F1 ). One can prove that there is a sequence of functions ωi : [0, ∞) → R (i =
1, 2, . . .) such that limδ →0+ ωi (δ ) = 0 for every i, and whenever f ∈ E1 is defined on
[0, 1], then the modulus of continuity of f |[0,1] is dominated by one of the functions
ωi . That is,
max{|f (x) − f (y)| : x, y ∈ [0, 1], |x − y| ≤ δ } ≤ ωi (δ )
for every δ > 0. (See the proof of [4, Theorem 6.4].) Let xi ∈ [0, 1] be a strictly
decreasing sequence such that xi → 0 and 0 < xi < ωi (2−i ) for every i. If F is defined
on [0, 1], F(0) = 0 and F(xi ) > 2−i for every i, then the modulus of continuity of F
cannot be dominated by any one of the functions ωi , and thus F ∈ / E1 .
We shall construct a function f0 ∈ E1 such that f0 is defined on R,
f (x) = x−2 ·f0 (1/x) is summable on R, and the integral function F(x) = ox f dx
satisfies F(xi ) > 2−i for every i. By the choice of the numbers xi this will imply
F∈ / E1 , and will complete the proof.
We shall construct f0 such that |f0 (x)| ≤ 1/x2 holds for x ≤ −1. This will ensure
that f is summable on the intervals (−∞, −1], [−1, 0] and [1, ∞). We have
xi 1/xi+1
f dx = f0 dx
xi+1 1/xi
for every i. We shall construct f0 such that the value of this integral will be between
2−i and 21−i for every i = 2, 3, . . .. This will ensure that f is summable on [0, 1], and
F(xi ) > 2−i for every i.
In order to construct f0 , we apply Theorem 6.4 of [4]. We find that for every pair
of continuous functions g : R → R and ε : R → (0, ∞) there is a function f0 ∈ E1
such that |f0 (x) − g(x)| < ε (x) for every x ∈ R. It is clear that with a suitable pair of
functions g and ε , f0 will have the required properties.
1. S. Kaczmarz and A. Turowicz, Sur l’irrationalité des intégrales indéfinies, Studia Math. 8 (1939),
129-134.
2. Jean-Pierre Kahane, Sur le théorème de superposition de Kolmogorov, J. Approximation Theory
13 (1975), 229-234.
3. K. Kuratowski, Topology, vol 2, Academic Press, 1968.
4. M. Laczkovich and I. Z. Ruzsa, Elementary and integral-elementary functions, Illinois J. Math.
(44) (2000), 161-182.
M. Laczkovich
PROBLEM 110: ULAM

October 1, 1935; Prize: One bottle of wine, S. Ulam
Let M be a given manifold. Does there exist a numerical constant K such that
every continuous mapping f of the manifold M into part of itself which satisfies
the condition |f n x − x| < K for n = 1, 2, . . . [where f n denotes the nth iteration of
the image f (x)] possesses a fixed point: f (x0 ) = x0 ? (By a manifold, we mean a set
such that the neighborhood of every point is homeomorphic to the n-dimensional
Euclidean sphere.) The same under more general assumptions about M (general
continuum?).
Addendum. An affirmative answer in the case where M is a locally contractible
2-dimensional continuum.
March, 1936
J. von Neumann observed that from the n-dimensional theorem an affirmative

answer would follow for Hilbert’s problem concerning the introduction of analytic
parameters in n-parameter groups.
March, 1936
Updated Commentary
The second part of the problem has been answered in the negative by W. Kuperberg,
who gave an example of a 1-dimensional metric continuum, which for every K > 0
admits a fixed-point free K-involution. Subsequently, W. Kuperberg and P. Minc
proved that the Cartesian product of the Hilbert cube Q and the circle S1 has the
property: For every K > 0 there exists a dynamical system Φ on Q × S1 such that for
each x ∈ Q × S1 the trajectory Φ (t, x) is of diameter smaller than K, and Φ (n, x) = x
for each nonzero integer n. Thus, by defining f (x) = Φ (1, x), the authors have found
a fixed-point-free homeomorphism f satisfying the above property |f n (x) − x| < K,
for n = 1, 2, . . ., and defined on an absolute neighborhood retract.
The problem for a manifold has been answered in the negative by K. Kuperberg
and C. Reed [9] who gave an example of a C∞ dynamical system Φ on the 3-
dimensional Euclidean space R3 with all trajectories bounded by the given constant
K and no rest points (in fact, Φ (n, x) = x for any x ∈ R3 ). Again, by taking
f (x) = Φ (1, x), the authors obtain an example of a fixed-point free homeomorphism
of R3 onto R3 such that |f n (x) − x| < K for any iteration f n of f . An example of a
dynamical system with the same properties can be constructed on a closed manifold
S1 × S1 × S1 .
T C
T4δ
T2δ
W. Kuperberg
C. Reed
The examples by W. Kuperberg and P. Minc, which were given in 1979, are
described in the survey [8].
The dynamical system on R3 with uniformly bounded orbits in [9] is based on
a construction of a plug, first constructed by F.W. Wilson in [12]. In fact, without
explicitly specifying, Wilson’s paper yields a C∞ dynamical system on any smooth,
connected, metrizable, boundaryless n-manifold, n ≥ 3, non-compact or of Euler
characteristic zero, with each orbit contained in an element of a given open cover,
thus answering Ulam’s question. The Kuperberg-Reed plug possesses invariant
annuli, a property which led to the radius inequality method used by K. Kuperberg
in [7] for the construction of a smooth aperiodic dynamical system on S3 (Seifert
conjecture [11]), and by G. Kuperberg and K. Kuperberg [5] for the construction
of real analytic counterexamples to the Modified Seifert conjecture. The C1 Seifert
conjecture was solved in the negative earlier by P.A. Schweitzer [10]. A modification
of Schweitzer’s plug by G. Kuperberg [6] gave a volume preserving, aperiodic,
C1 dynamical system Φ on S3 ; the total space of the line bundle tangent to Φ
has a symplectic structure, and Φ is Hamiltonian with respect to this structure.
The classical Hamiltonian Seifert conjecture for S3 embedded in R4 was solved
by V.L. Ginzburg and B.Z. Gürel [3]. It is not known whether there exists a volume
preserving, C∞ counterexample to the Seifert conjecture.
The following question posed by G. Kuperberg remains open: Does there exist a
non-singular, volume preserving dynamical system on R3 with uniformly bounded
orbits?
There are examples of non-singular dynamical systems on R3 with bounded
(though not uniformly) orbits:
1. B. Brechner and R.D. Mauldin [2], which is based on K. Borsuk’s example in [1]
of an acyclic compact subset of R3 without the fixed point property and cannot
be modified to be volume preserving.
2. G.S. Jones and J.A. Yorke [4], which can be modified to be volume preserving.
1. K. Borsuk, Sur un continu acyclique qui se laisse transformer topologiquement en lui même
sans points invariants, Fund. Math. 24 (1935), 51–58.
2. B.L. Brechner and R.D. Mauldin, Homeomorphisms of the plane, Pacific J. Math. 59 (1975),
375–381.
3. V.L. Ginzburg and B.Z. Gürel, A C2 -smooth counterexample to the Hamiltonian Seifert
conjecture in R4 , Ann. of Math. 158 (2003), 953–976.
4. G.S. Jones and J.A. Yorke, The existence and nonexistence of critical points in bounded flows,
J. Differential Equations 6 (1969), 238–246.
5. G. Kuperberg and K. Kuperberg, Generalized counterexamples to the Seifert conjecture, Ann.
of Math. 144 (1996), 239–268.
6. G. Kuperberg, A volume-preserving counterexample to the Seifert conjecture, Comment. Math.
Helv. 71 (1996), 70–97.
7. K. Kuperberg, A smooth counterexample to the Seifert conjecture, Ann. of Math. 140 (1994),
723–732.
8. K. Kuperberg, W. Kuperberg, P. Minc and C.S. Reed, Examples related to Ulam’s fixed point
problem, Topol. Methods in Nonlinear Analysis 1 (1993), 173–181.
9. K. Kuperberg and C. Reed, A rest point free dynamical system on R3 with uniformly bounded
orbits, Fund. Math. 114 (1981), 229–234.
10. P.A. Schweitzer, Counterexamples to the Seifert conjecture and opening closed leaves of
foliations, Ann. of Math. 100 (1974), 386–400.
11. H. Seifert, Closed integral curves in 3-space and isotopic two-dimensional deformations, Proc.
Amer. Math. Soc. 1 (1950), 287–302.
12. F.W. Wilson, On the minimal sets of non–singular vector fields, Ann. of Math. 84 (1966),
529–536.
K. Kuperberg
PROBLEM 111: SCHREIER
Does there exist an uncountable group with the property that every countable
sequence of elements of this group is contained in a subgroup which has a
finite number of generators? In particular, do the groups S∞ and the group of all
homeomorphisms of the interval have this property?
Commentary
The answer to the first problem is yes. It can be obtained by taking the union of
an appropriate chain of groups obtained by amalgamation using the fact that every
countable group is a subgroup of a group with two generators.
The second question remains open.
J. Mycielski
Fred Galvin gave an affirmative answer to the second question in his paper:
Generating countable sets of permutations, J. London Math. Soc. 51 (1995), 230-
242. Whether the homeomorphism group of the interval has this property remains
open.
R. Daniel Mauldin
PROBLEM 112: SCHREIER

Is an automorphism of a group G which transforms every element into an
equivalent one of necessity an inner automorphism?
Commentary
The answer is no. Burnside gave an example of a finite group with outer automor-
phisms mapping every conjugacy class onto itself. For further work and references,
see G.E. Wall, Finite groups with class-preserving outer automorphisms, J. London
Math. Soc.22 (1947), 315–320.
Jan Mycielski
PROBLEM 113: SHREIER

Let C denote the space of continuous functions of a real variable (under uniform
convergence in every bounded interval); let F(f ) denote an operation which is
continuous, which has an inverse which maps C onto itself, and such that it maps
the composition of two functions f (g) into the composition of F(f ) and F(g).
Is F(f ) of the form F(f (t)) = hfh−1 (t), where h is a continuous function strictly
monotonic in this interval (−∞, +∞) and
lim h(t) = −∞, lim h(t) = +∞?

t→−∞ t→+∞
Solution
Theorem. Let T be a continuous automorphism of the semigroup C onto C. Then

there is a homeomorphism h of R onto R so that T(f ) = h−1 fh.
Notice that an element f of C is constant if and only if fg = f , for all g in C. From
this it follows that T takes constant functions to constant functions. For each x ∈ R,
let w(x) be the number such that the constant function x̄ is taken to w(x)¯ by T. Notice
that w is a one-to-one map of R onto R. Also, since T is continuous (under uniform
convergence on compact sets), w is continuous. Therefore, w is a homeomorphism.
Let h = w−1 . We plan to show that T(f ) = h−1 fh, for all f in C.
Let H be the subset of C consisting of all homeomorphisms of R. It can
be checked that T(H) = H. Thus, T|H is a continuous automorphism of the
group of homeomorphisms of R. It follows that T|H is inner [1]. So, there is a
homeomorphism k such that T(f ) = k−1 fk for all homeomorphisms f .
In order to see that k = h, fix x0 and for each n, set hn (x) = (1/n)(x − x0 ) + x0 .
Then {hn }∞ n=1 is a sequence of homeomorphisms converging to x¯0 . Thus,
T(hn ) = k hn k converges to T(x¯0 ) = h−1¯(x0 ). In particular, (Thn )(k−1 (x0 )) =
−1
k−1 (hn (x0 )) = k−1 (x0 ) converges to h−1 (x0 ). Thus, k = h.

Now fix f ∈ C and x0 ∈ R. We will show that (T(f ))(h−1 (f (x0 )) = h−1 (f (x0 )).
For each n, let hn be a homeomorphism of R such that hn (x0 + n) = x0 + 1/n.
Then the sequence fhn converges to the constant function f (x¯ 0 ) uniformly on
every interval. Thus, T(fhn ) converges to T(f (x¯ 0 )) = h−1 (f¯(x0 )). Since T(fhn ) =
Tf (h−1 hn h) and T(fhn )(h−1 (x0 )) → h−1 (f (x0 )), it follows that (Tf )(h−1 (x0 )) =
h−1 (f (x0 )). The theorem follows from this.
1. N.J. Fine and G.E. Schweigert, On the group of homeomorphisms of an arc, Ann. Math. 62
(1955), 237–253.
R. Daniel Mauldin
PROBLEM 114: AUERBACH, ULAM
The circumference of a circle can be approximated by a one-to-one continuous
image of a half line p in an essential manner; that is to say, the Abbildungsgrad of
the transformation obtained by central projecting of the line into the circumference
is equal to +∞ and the approximated circle is the set of points of condensation.
Is it possible to approximate analogously the surface of a sphere in the
3-dimensional space by a one-to-one continuous image of a plane?
P K
Remark
The answer is no. See A. Calder, For n > 1 any map Rn → Sn is uniformly homotopic
to a constant, Indag. Math. 34 (1972), 32–36, and A. Calder, Uniformly trivial maps
into spheres, Bull. Amer. Math. Soc. 81 (1975), 189–191.
PROBLEM 115: ULAM

Does there exist a homeomorphism h of the Euclidean space Rn with the
following property? There exists a point p for which the sequence of points hn (p) is
everywhere dense in the whole space. Can one even demand that all points except
one should have this property? For a plane such a homeomorphism (with the desired
property only for certain points) was found by Besicovitch.
Commentary
The answer to the first question is yes for n ≥ 2 [8, 10] (and of course no for n = 1).
In this part of the problem it does not matter whether the sequence hn (p) is taken
to refer to the full orbit of p or just to the positive semiorbit (because if some point
has a dense orbit, then a dense Gδ set of points have dense positive and negative
semiorbits [9, p. 70]), but it makes a difference in the second part. If the sequence
is taken to refer to the positive semiorbit of p, the answer to the second question
is always no. By a theorem of Dowker [6, Th. K] (applied to Rn ∪ {∞}), if some
point has a dense orbit, then there are points whose positive semiorbit is nowhere
dense and lies outside any given sphere. More generally, Homma and Kinoshita [7,
Th. 5] have shown that under any continuous mapping of Rn into itself there is a
dense set of points whose positive semiorbit closure is a proper subset of Rn . See
also Birkhoff [4, p. ! 202]. If the sequence hn (p) is taken to refer to the full orbit of
p, the second question remains open for every n ≥ 2. Besicovitch [2, 3] constructed
a class of homeomorphisms of R2 that are aperiodic except for a fixed point at 0.
Each has a point whose positive semiorbit is dense in R2 , but also a point other than
0 whose orbit is bounded. In the case n = 2 the words “except one” are important—
it is known that not every orbit can be dense in R2 [5]. Indeed, by a theorem of
Brouwer (see [1, Prop. 1.2]), if h is a homeomorphism of R2 onto itself, then either
h2 (p) = p for some p or else hn (p) → ±∞ for every p. For n ≥ 3 the second question
appears to remain open even when the words “except one” are deleted.
1. S.A. Andrea, On homeomorphisms of the plane which have no fixed points, Abh. Math. Sem.
Univ. Hamburg 30 (1967), 67–74.
2. A.S. Besicovitch, A problem on topological transformation of the plane, Fund. Math. 28
(1937), 61–65.
3. , A problem on topological transformation of the plane II, Proc. Cambridge
Philos. Soc. 47 (1951), 38–45.
4. G.D. Birkhoff, Dynamical systems, Amer. Math. Soc. Colloq. Pub. Vol. 9, New York, 1927.
5. B.L. Brechner and R.D. Mauldin, Homeomorphisms of the plane, Pacific J. Math. 59 (1975),
375–381.
6. Y.N. Dowker, The mean and transitive points of homeomorphisms, Ann. Math. (2) 58 (1953),
123–133.
7. T. Homma and S. Kinoshita, On the regularity of homeomorphisms of En , J. Math. Soc. Japan
5 (1953), 365–371.
8. J.C. Oxtoby, Note on transitive transformations, Proc. Nat. Acad. Sci. U.S.A. 23 (1937),
443–446.
9. , Measure and category, Springer-Verlag, New York, 1971.
10. V.S. Prasad, Ergodic measure preserving homeomorphisms of Rn , Indiana Univ. Math. J. 28
(1979), 859–867.
John C. Oxtoby
The most interesting still unresolved questions from Problem 115 in the Scottish
Book ask if there exists a homeomorphism of Rn or of Rn \ {0} with the full orbit of
every point dense. Such a homeomorphism is called minimal because the smallest
non-empty closed invariant subset is the entire space. In full generality the problem
is still open, but in addition to the partial results cited in the commentary above there
have been some more recent developments. The case n = 2 has been completely
resolved. As mentioned in the commentary the Brouwer plane translation theorem
(see, e.g., [1] for a modern proof) implies there can be no minimal homeomorphism
of R2 . The proof that no minimal homeomorphism exists for the punctured plane
was provided by P. Le Calvez and J.-C. Yoccoz [4]. Le Calvez and Yoccoz prove,
in fact, that the plane with any finite number of punctures does not admit a minimal
homeomorphism. A shorter proof using the Conley index was provided by J. Franks
[2]. The question has also been resolved in the orientation reversing case for R3 . L.
Hernández-Corbato, P. Le Calvez, and F. Ruiz del Portal prove in [5] that there is
no minimal orientation reversing homeomorphism of R3 . The other cases for Rn ,
n ≥ 3, with or without a deleted point, remain open.
The non-existence of forward minimal homeomorphisms for Rn is referenced in
the commentary above. However, it is worth mentioning a more general (and earlier)
result of W.H. Gottschalk (Theorem B of [3]) which asserts there is no forward
minimal homeomorphism of any non-compact locally compact space.
1. J. Franks, A New Proof of the Brouwer Plane Translation Theorem, Ergodic Theory and
Dynamical Systems, 12 (1992), 217–226.
2. J. Franks, The Conley index and non-existence of minimal homeomorphisms, Illinois Journal
of Math., 43 (1999) 457–64.
3. W. H. Gottschalk, Orbit-closure decompositions and almost periodic properties. Bull. Amer.
Math. Soc. 50, (1944). 915–919.
4. P. Le Calvez and J.-C. Yoccoz, Un théorème d’indcice pour les homéomorphismes du plan au
voisinage d’un point fixe, Annals of Math. 146 (1997), 241–293.
5. L. Hernández-Corbato, P. Le Calvez, and F. Ruiz del Portal, About the homological discrete
Conley index of isolated invariant acyclic continua, Geom. Topol., 17, (2013), 2977–3026.
John Franks

Let G be a compact group. It is known that almost every (in the sense of
Haar measure) couple of elements φ , ψ ∈ G generates in G an everywhere dense
subgroup. Let there be given a sequence {cn } of zeros and ones. Let us put fn = φ
if cn = 0, fn = ψ if cn = 1. Prove that for almost every pair φ , ψ and almost every
sequence {cn } the sequence f1 , f1 f2 , f1 f2 f3 , . . . is everywhere dense in G. Investigate
whether this sequence is uniformly dense; that is, for every region V ⊂ G we
should have lim qn /n = measure of V, if qn denotes the number of the elements
of f1 , f1 f2 , . . . , f1 f2 · · · fn which fell into V. Investigate also whether an analogous
theorem holds for similar sequences of images of a point p obtained with the aid
of two transformations Φ (p) and Ψ (p), which are strongly transitive mappings of
the space S into itself preserving measure.
Commentary
The problem in its general form still seems to be open. A deep result of
Veech can, however, be applied in order to construct sequences of the form
f1 , f1 f2 , . . . , f1 f2 · · · fn , . . . that are “uniformly dense” (or, in modern terminology,
“uniformly distributed”) in the compact group G. Suppose φ , ψ ∈ G generate a
dense subgroup of G, and let y1 , y2 , . . . be nonconstant sequence with yn = φ or ψ
for each n. Then, according to Veech, there exists a sequence r1 , r2 , . . . of positive
integers such that, putting fn = yrn for each n, the sequence f1 , f1 f2 , . . . , f1 f2 · · · fn , . . .
is uniformly distributed in G. The result of Veech, which refers to more general
sequences y1 , y2 , . . ., can be found in “Some questions of uniform distribution,”
Ann. Math. (2) 94 (1971), 125–138.
H. Niederreiter
We begin with a discussion about the first sentence of the problem. It is clear that
some additional assumptions on the group are needed in order that almost every
pairs of elements generate a dense subgroup. In particular, the group needs to
be connected, since otherwise both generators may be contained in the connected
component with positive probability.
In addition, the topology must have a base of small cardinality. It has been
shown by Hofmann and Morris (Weight and c, J. Pure Appl. Algebra 68 (1990), no.
1–2, 181–194.) that a pair of elements generating a dense subgroup in a compact
connected Hausdorff topological group exists if and only if the topology has a
base of cardinality at most c. Even when there is a pair of elements generating a
dense subgroup, the set of such pairs can still be non-measurable, e.g. in the case
(R/Z)c . Indeed, in this group any Borel set may depend on at most countably many
coordinates.
Auerbach (Sur les groupes linéaires bornés (III), Stud. Math. 5 (1934), no. 1,
43–49.) proved that almost every pair generates a dense subgroup in a compact
connected linear group. This was generalized by Schreier and Ulam (Sur le nombre
des générateurs d’un groupe topologique compact et connexe, Fund. Math. 24
(1935), no. 1, 302–304) to compact connected metrizable groups with almost every
in the sense of measure theory replaced by generic in the sense of Baire category.
They used von Neumann’s partial solution of Hilbert’s 5’th problem. It seems that
a very similar argument would work to show the claim for almost every pairs. We
believe that this was the intended setting of the problem.
Probability measures and random walks on semigroups have been studied
extensively in the past century. A comprehensive exposition of the subject can be
found in the book of Högnäs and Mukherjea (Probability measures on semigroups,
Springer, 2011). In particular, Theorem 3.12 (Chapter 3, p 212) gives a positive
answer to the problem:
Theorem 1. Let G be a compact metrizable group, and let μ be a Borel probability
measure on it. Denote by Zn the random walk generated by μ , that is Zn is the
product of the first n elements of a sequence of independent random elements of
G with probability law μ . If the support of μ is not contained in a proper closed
subgroup of G, then Zn equidistributes in the sense that
1 N
∑ f (Zn ) →
N n=1
fdmHaar
almost surely for all continuous function f ∈ C(G).

By the result of Schreier and Ulam, this holds in particular when μ = 1/2
(δφ + δψ ) for almost every pair φ , ψ ∈ G.
(The condition that suppμ is not contained in a proper closed subgroup is a
standing assumption stated in the beginning of Section 3.3 (p. 198) of the book.)
For the reader’s convenience we outline briefly an argument proving the above
theorem. Breiman (The strong law of large numbers for a class of Markov chains,
Ann. Math. Statist. 31 (1960), no.3, 801–803) proved the following “Law of large
numbers” that is applicable to our situation. Let {Xn }n∈N be a stationary Markov
chain on a statespace G that is a compact Hausdorff topological space. Suppose
that for each continuous function f ∈ C(G), the conditional expectation E(f (Xn )|
Xn−1 = x) is a continuous function of x ∈ G. Suppose further that the common law
of Xn is the only stationary measure with respect to the transition probabilities. Then
1 N

P lim
N→∞ N
∑ f (Xn ) = fdmHaar X1 = x = 1
n=1
for every point x ∈ G.

We apply this theorem as follows. We let X0 be a random element with law mHaar
that is independent of the random walk Zn . Then we set Xn = X0 Zn for all n and apply
the theorem with x = 1, the unit element.
It remains to show that the Haar measure is indeed the unique stationary measure.
We recall that the random walk is generated by the probability measure μ , and a
measure ν is stationary if and only if μ ∗ ν = ν . Here and everywhere below μ ∗ ν
denotes the convolution of μ and ν . We show that 1/N ∑Nn=1 μ ∗n → mHaar in the
weak-* topology. If ν is stationary, then
1 N ∗n
ν= ∑ μ ∗ ν → mHaar ∗ ν = mHaar
N n=1
and hence ν = mHaar indeed.

One way to verify the claimed convergence is via spectral theory. For g ∈ G, we
denote by Ug the unitary operator acting on L2 (G) defined by Ug f (h) = f (hg). We
set U = Ug d μ (g). With this notation, our claim is equivalent to
1 ∞ n
∑U f →
N n=1
fdmHaar
for all f ∈ C(X). Since Ug f has the same modulus of continuity for all g ∈ G, it is
enough to show the above convergence in L2 .
If Uf = f for some f ∈ L2 (G), then Ug f , f = 1 for μ -almost all g. Since Ug f , f
is continuous, Ug f , f = 1 follows for all g ∈ suppμ . Since Ug is unitary, this
implies Ug f = f , and this holds for all g ∈ G, since suppμ is assumed to generate
a dense subgroup. Thus f must be constant. Using Riesz’s generalization of the
von Neumann mean ergodic theorem for U, we conclude that (1/N) ∑∞ n=1 U f →
n
fdmHaar in L for all f ∈ L (G). We also note that L (G) can be decomposed as an
2 2 2
orthogonal sum of finite dimensional U-invariant subspaces due to the Peter Weyl
theorem, hence the ergodic theorem can be reduced to the finitely dimensional case
in this situation.
J. Bourgain and P.P. Varjú
PROBLEM 117:FRÉCHET.
Original manuscript in French
Consider a Jordan curve which has a tangent (oriented) at every point. Does
there exist at least one parametric representation of this curve where the coordinates
are differentiable functions of the parameter and where the derivatives of the three
coordinates do not vanish simultaneously?
Addendum.∗ In general, no; but we can represent the curve with functions of a
parameter t in such a way that dx/dt, dy/dt, dz/dt exist (and are not all zero), except
for a set N of values of t, such that m(N) = 0 and also the set of points of the curve,
corresponding to N, has Caratheodory measure zero (Fund. Math. 28).
A.J. Ward
March 23, 1937
∗ Original manuscript in English
PROBLEM 118: FRÉCHET.

Original manuscript in French
Let Δ (n) be the greatest of the absolute values of determinants of order n whose
terms are equal to ±1. Does there exist a simple analytic expression of Δ (n)
as a function of n; or, more simply, determine an analytic asymptotic expression
for Δ (n).
Commentary
This problem is misattributed. Hadamard published his famous paper [5] with a
partial solution of the problem in 1893, when Fréchet was barely 15 years old.
Already in 1867 Sylvester [7] studied “Hadamard” matrices, although it seems
that he was not aware of the connection between these matrices and the maximal
determinant problem. Hadamard [5] proved that any complex n-square matrix A,
with entries not greater in absolute value than 1, satisfies
| det(A)| ≤ nn/2 . (1)
If all the entries in A are real, then equality can hold in (1) if and only if
AAT = nIn , (2)
which implies that all the entries of A are ±1. A (1. − 1) n-square matrix with
determinant ±nn/2 is called a Hadamard matrix. It easily follows from (2) that
an n × n Hadamard matrix can exist only if n = 1, 2, or n ≡ 0 mod 4. It has
been conjectured that Hadamard matrices exist for all such n. The conjecture
is unresolved although Hadamard matrices have been constructed for an infinite
number of values of n. The smallest n for which the conjecture is undecided is 268.
For a comprehensive listing of the orders for which Hadamard matrices are known,
see [6].
Inequality (1) implies that
Δ (n) ≤ nn/2 (3)

for all n, and if n ≡ 0 mod 4 then
Δ (n) = nn/2 ,
provided that an n × n Hadamard matrix exists.

If n > 2 and n ≡ 0 mod 4, then Δ (n) < nn/2 , and the bound in (1) can be
improved. Barba [1] showed that if n is odd, then
Δ (n) ≤ (2n − 1)1/2 (n − 1)(n−1)/2 (4)

∼ (2/e)1/2 nn/2
= 0.85776nn/2 .
Ehrlich [4] sharpened Barba’s bound for n ≡ 3 mod 4, and n ≥ 63:
Δ (n) ≤ 2 · 113 · 7−7/2 (n − 3)(n−7)/2 n7/2 (5)

∼ 2 · 113 · 7−7/2 e−3/2 nn/2
= 0.65452nn/2 .
For the case n ≡ 2 mod 4, Wojtas [8] proved that
Δ (n) ≤ 2(n − 1)(n − 2)(n−2)/2 (6)

∼ (2/e)nn/2
= 0.73576nn/2 .
The same result was obtained independently by Ehrlich [3]. It is also known [1, 9]
that equality holds in (6) for all n ≡ 2 mod 4, n ≤ 62, such that no prime factor of
the squarefree part of n − 1 is congruent to 3 mod 4. The bound in (6) therefore is,
in a sense, the best possible.
Comparing the bounds in (3), (4), (5), (6), and taking into consideration the
known cases of equality, it appears that there is no simple analytic expression for
Δ (n), nor does there exist an analytic asymptotic expression for Δ (n). Nevertheless,
Clements and Lindström [2] have shown that for any n,
n(n/2)(1−C(n)) < Δ (n) ≤ nn/2 ,
where C(n) = log2 (4/3)/ log2 n. It follows that
log Δ (n) ∼ log nn/2 .
1. G. Barba, Intorno al teoremo di Hadamard sui determinanti a valore massimo, Giorn. Mat.
Battaglini 71 (1933), 70–86.
2. G.F. Clements and B. Lindström, A sequence of (±1)-determinants with large values, Proc.
Amer. Math. Soc. 16 (1965), 548–550.
3. H. Ehrlich, Determinantenabschätzungen für binäre Matrizen, Math. Z. 83 (1964), 123–132.

4. , Determinantenabschätzung für binäre Matrizen mit n ≡ 3 mod 4, Math. Z. 84
(1964), 438–447.
5. J. Hadamard, Résolution d’une question relative aux déterminants, Bull. Sci. Math. 10 (1963),
240–246.
6. J. Seberry, A computer listing of Hadamard matrices, Proc. International Conf. on Combinato-
rial Theory, Canberra, 1978.
7. J.J. Sylvester, Thoughts on inverse orthogonal matrices, simultaneous sign-successions, and
tessellated pavements in two or more colours, with applications to Newtons’s rule, ornamental
tile-work, and the theory of numbers, Phil. Mag. (4) 34 (1867), 461–475.
8. M. Wojtas, On Hadamard’s inequality for the determinants of order non-divisible by 4, Colloq.
Math. 12 (1964), 73–83.
9. C.H. Yang, On designs of maximal (+1, −1)-matrices of order n ≡ 2 (mod 4), Math. Comp. 22
(1968), 174–180, and 23 (1969), 201–205.
Henryk Minc
University of California
Santa Barbara, CA 93106
PROBLEM 119: ORLICZ
Does there exist an orthogonal system composed of functions uniformly bounded
and having the property possessed by the Haar system, that is to say, such that the
development of every continuous function in this system is uniformly convergent?
Commentary
A.M. Olevskii [1] has shown that neither C[0, 1] nor L1 [0, 1] have a Schauder basis
that is orthonormal and uniformly bounded. The results of Olevskii have been
sharpened in some directions by S.T. Szarek [2] who has shown in particular that
every normalized Schauder basis of L1 [0, 1] contains a subsequence whose span is
≈ . These results were further sharpened by Szarek in [3] and by Kwapien and
Szarek in [4].
1. A.M. Olevskii, Fourier series with respect to general orthogonal systems, Springer-Verlag, 1975.
2. S.T. Szarek, Bases and biorthogonal systems in the spaces C and L1 , Arkiv Math. 17 (1979),
255–271.
Joseph Diestel
Kent, Ohio
The results of Olevskii were also sharpened by S. Kwapien and S. Szarek (Studia
Math. 66 (1970), 189–200) and S. Szarek In turn the results of Kwapein, Olevskii,
and Szarek were established by J. Bourgain(Trans. Amer. math. Soc. 285 (1984),
133–139) for the dish algebra and L1 /H01 , using considerably more complicated
arguments.
Joseph Diestel
Kent, Ohio
PROBLEM 120: ORLICZ

Let xni be a sequence of powers with integer exponents on the interval (a, b) and
∞
1
∑ ni = +∞.
i=1
Give the order of approximation of a function satisfying a Hölder condition by

polynomials:
N
∑ ai xni .
i=1
This problem is now rather well-understood. Major results were obtained by M.v.
Golitschek, D.J. Newman, T. Ganelius, D. Leviatan, J. Bak, J.Tzimbalario, L. Marki,
G. Somorjai and J. Szabados and others.
The seminal result is as follows. Let Λ = {0 = λ0 < λ1 < · · · < λN } and EΛ (f ) =
infaj f − ∑j aj xλj p for f ∈ Lp [0; 1], 1 ≤ p < ∞ or f ∈ C[0; 1] for p = ∞. Let B(z) =

z−λj − p1 B(1+iy)
∏j=1 z+λj + p 1 , ε = supy≥0 1+iy and ωp (f , δ ) = sup|h|≤δ f (· + h) − f (·)p – be the
p-modulus of continuity. The main result is
EΛ (f ) ≤ Aωp (f , ε )
for some absolute constant A; the estimate is sharp, i.e. it can not hold for A < 1/600.
The history, references, and proofs can be found in G. G. Lorentz, M. v.
Golitschek, Y. Makovoz "Constructive Approximation: Advanced Problems"
Springer, 1996, Ch. 11.
K. S. Rjutin
PROBLEM 121: ORLICZ

Give an example of a trigonometric series
∞
∑ (an cos nx + bn sin nx)
n=1
everywhere divergent and such that

∞
∑ (a2+
n
ε
+ b2+ ε
n ) < +∞
n=1
for every ε > 0.
The required example can be constructed by the following results of S.Sh. Galstyan.
Theorem 1. [2]. Let {αn : n ≥ 0} be a nonincreasing sequence of positive
numbers tending to zero and ∑∞ ∞
n=0 αn = ∞. Then there is a series ∑n=0 cn e
2 inx of class
H , the real and imaginary parts of which diverge everywhere, and cn = O(αn ).
1
Theorem 2. [3]. Let {αn : n ≥ 0} be a nonincreasing sequence of positive

numbers tending to zero and ∑∞ n=0 αn = ∞. Then there exists a subsequence {αnk :
2
∞
k ≥ 0} such that the series ∑k=0 αnk cos(nk t)diverges everywhere.
Theorems 1 and 2 are sharp. We recall that, by a famous theorem of
L. Carleson [4], if an and bn are real numbers and ∑∞ n=1 (an + bn ) < ∞, then the
2 2
∞
series ∑n=1 (an cosnx + bn sin nx) converges almost everywhere.
Using a result of Körner, an explicit example of such a series has been given by
Akita, Gotô and Kano [1].
Everywhere divergent complex trigonometric series with rapidly decreasing
coefficients are studied in [5, chapter 8, §9].
1. Akita, M.; Gotô, K; and Kano, T., A problem of Orlicz in the Scottish Book, Proc. Japan Acad.
Ser. A Math. Sci. 62 (1986), 267–269.
2. Galstyan, S.Sh., Everywhere divergent trigonometric Fourier series with rapidly decreasing
coefficients (Russian), Mat. Sb. (N.S.) 122 (164) (1983), N. 2, 157–167; English translation:
Mathematics of the USSR - Sbornik, 50 (1985), N. 1, 151–161.
3. Galstyan, S.Sh., Everywhere divergent trigonometric series (Russian), Mat. Zametki 37 (1985),
N. 2, 186-191; English translation: Math. Notes 37 (1985), N. 1-2, 105–108.
4. Carleson, L., On convergence and growth of partial sums of Fourier series, Acta, Math. 116
(1965), N. 1–2, 135–157.
5. Kahane, J.-P., Some random series of functions, D. C. Heath and Company, Massachusetts,
1968.
Sergei Konyagin

Does there exist in every space of type (B) of infinitely many dimensions, a series
which is unconditionally convergent but not absolutely? (A series
∞
∑ xn
n=1
is called unconditionally convergent if it converges under every ordering of its terms

and absolutely convergent if the series
∞
∑ xn
n=1
converges.
Commentary
In 1950, A. Dvoretsky and C.A. Rogers (Proc. Nat. Acad. Sci. USA 36 (1950), 192–
197) showed that if every unconditionally convergent series in a Banach space X is
absolutely convergent then X must be finite dimensional; this gives an affirmative
response to Problem 122. Close on the heels of the Dvoretsky-Rogers solution came
a new and stunning approach to a whole circle of related problems, developed by
A.Grothendieck. Central to the Grothendieck program is the idea of a p-absolutely
summing operator: A bounded linear operator T between the Banach spaces X and Y
is p-absolutely summing if there exists a constant K > 0 such that given x1 , . . . , xn ∈
X the inequality

n n
∑ Txk p ≤ K p sup ∑ |x∗ xk |p : x∗ ≤ 1
k=1 k=1
holds. A quick check in case p = 1 shows that the operator T : X → Y is 1-absolutely

summing if and only if T takes unconditionally convergent series into absolutely
convergent series; for general p ≥ 1, T is p-absolutely summing if and only if
whenever ∑n |x∗ xn |p is finite for each x∗ ∈ X ∗ then ∑n Txn p is finite. Through the
work of Grothendieck and A. Pietsch (Studia Math. 28 (1967), 333–353), one can
conclude that if a normed linear space X has the property that for some p ≥ 1 the
series ∑n xn p converges whenever ∑n |x∗ xn |p does for each x∗ ∈ X ∗ then X must
be finite dimensional.
Though the results of Grothendieck and Pietsch might appear to be but a marginal
improvement of that of Dvoretsky and Rogers such is far from the truth. On the
one hand, the theory of p-absolutely summing operators (and related classes of
operators) has played a central role in the revival of Banach space theory especially
as it relates to other areas of mathematical endeavor (particularly harmonic theory).
On the other hand, the theory of p-absolutely summing operators is instrumental in
providing a more complete answer to Problem 122, particularly for Fréchet spaces
(F0 spaces in the Polish terminology). In fact, Grothendieck (Memoir American
Mathematical Society, volume 16 (1955)) was able to classify those Fréchet spaces
in which unconditionally convergent series are absolutely convergent (such spaces
are called nuclear) and showed that many of the important non-normed spaces of
analysis are indeed nuclear.
The above synopsis only touches the tip of a mathematical iceberg. Improve-
ments of the Grothendieck-Pietsch results have been obtained by B. Maurey and
A. Pe}lczyński (Studia Math. 54 (1976), 291–300) and H. König (preprint from
Bonn University). The original Dvoretsky-Rogers proof was to lead Dvoretsky to
his famous “ε -spherical sections” theorem, recently given a definitive treatment by
T. Figiel, J. Lindenstrauss and V. Milman (Acta Math. 139 (1977), 53–94). The
theory of nuclear spaces has been extensively developed, principally by the Soviet
school (a good report on which can be found in the articles of B. Mityagin appearing
in the 1978–79 Seminaire Functional Analyse, Ecole Polytechnique).
In addition to these developments, the finite dimensional structure and the theory
of p-summing operators have led to the Maurey-Rosenthal dichotomy (c.f. H.P.
Rosenthal, Studia Mathematica 58 (1976), 21–43).
Joseph Diestel
Kent, Ohio
Given are three sets A1 , A2 , A3 located in the 3-dimensional Euclidean space and
with finite Lebesgue measure. Does there exist a plane cutting each of the three sets
A1 , A2 , A3 into two parts of equal measure? The same for n sets in n-dimensional
space.
Addendum Solution in “Z Topologii,” Mathesis Polska 1936.
Commentary
I have not seen the solution referred to (in Mathesis Polska 1936); perhaps it was
the one circulating orally in Princeton in 1941. The theorem for 3 sets, A1 , A2 , A3 in
R3 , was aptly named the “Ham Sandwich Theorem.” The proof began by bisecting
the “ham” A3 by a continuously varying plane, and observing that the measure-
differences of the parts in which it cut the two “slices of bread” A1 , A2 provided
an antipodal map from S2 to R2 . The case n = 2 of what has become known as the
Borsuk-Ulam theorem [2] then guarantees that some point of S2 is mapped to the
origin, giving a plane that bisects all 3 sets.
The case n = 2 of the Borsuk-Ulam theorem can be proved without much formal
topological apparatus; hence, the device of bisecting A3 first allows the proof to be
completely elementary, as is indicated briefly in [5] and in full in [3, pp. 120–123].
However, the use of the Borsuk-Ulam theorem in full strength (an antipodal map
from Sn to Rn maps onto the origin, n = 1, 2, . . .) gives an easier proof that one can
bisect each of n given sets (of finite measure) in Rn by some hyperplane, and in fact
that bisection behaves like a linear condition on algebraic varieties; one can bisect 5
given sets in the plane by a conic, and so on (see [5]).
It has been pointed out that there is no need to use the same measure on all the
sets. For instance, in the original case n = 3, one might with some realism bisect the
volumes of the two slices of bread, and the surface area of the ham.
The situation for ratios other than bisection, in R1 and R2 , was also investigated
in [5]. In R1 , a necessary and sufficient condition on positive numbers α1 , α2
that, given sets A1 , A2 of finite Lebesgue measure, there always exists an interval
(possibly infinite) in R1 whose intersection with Ai has measure αi times that of
Ai (i = 1, 2), is that α1 = α2 = reciprocal of an integer greater than 1. In R3 , the
same condition on α1 , α2 is necessary in order that there always exist a circle (or
straight line) cutting off αi times the measure of Ai (i = 1, 2). The Ham Sandwich
Theorem shows that when α1 = α2 = 1/2 the condition is sufficient; but the case
α1 = α2 = 1/3 remains (like the others) unsettled. To prove sufficiency here, it
would suffice (by an approximation argument) to prove that given two finite subsets
B1 , B2 of the plane, Bi ! having 3ni points, there exists, a circle having exactly
ni points of Bi in its interior (i = 1, 2). A recent postscript to [5] is supplied by
[1], which points out that the sketch of the elementary argument in [5] is too
sketchy; in general, one cannot bisect A3 by a continuously varying plane taking one
position in each direction. In [5] it was intended, but not stated, that one should first
approximate the given sets in measure by sets having positive measure everywhere.
Having bisected the new sets, one applies an obvious limiting process to bisect the
given ones. (In [3] the difficulty is avoided by assuming the sets Ai are open.)
The Borsuk-Ulam theorem and some analogous theorems about continuous
functions on spheres have led to very extensive developments. See [4], under the
headings “Theorems about Sn of Borsuk-Ulam Type” and “Theorems of Dyson,
Kakutani, Yamabe-Yujobo,” pp. 1117–1125.
1. Richard Arens, On Sandwich Slicing, Topology, Vol. II (Proc. Fourth Colloq., Budapest, 1978),
57–60, Colloq. Math. Soc. János Bolyai, 23, North-Holland, Amsterdam, 1980.
2. K. Borsuk, Drei Sätze über die n-dimensionale euklidische Sphäre, Fund. Math. 20 (1933),
177–190.
3. W.G. Chinn and N.E. Steenrod, First concepts of topology, Random House, New York 1966.
4. N.E. Steenrod (ed.), Reviews of Papers in Algebraic and Differential Topology, Topological
Groups and Homological Algebra, Part II, Amer. Math. Soc., 1968.
5. A.H. Stone and J.W. Tukey, Generalized “Sandwich” Theorems, Duke Math. J. 9 (1942),
356–359.
A.H. Stone
University of Rochester
Rochester, NY 14627
PROBLEM 124: MARCINKIEWICZ

What can one say about the uniqueness for the integral equation
1
y(t)f (x − t)dt = 0, 0 ≤ x ≤ 1? (1)
0
I know that if the sequence of integrals fk (x) = 0x fk−1 (t)dt, f0 = f ; k = 1, 2, 3, . . .

is complete in L2 then the only solution of Eq. 1 is y ≡ 0. This is the case also if f
is of bounded variation and f (0) = 0. Finally, if Eq. 1 possesses even one nonzero
solution, y, then every (iterated) integral of y also satisfies this equation.
I conjecture that if f (0) = 0 and f is continuous, then Eq. 1 has only the solution
y ≡ 0.
The answer is affirmative. This follows from the following more general theorem of
E. C. Titchmarch (Published in 1926!) (The zeros of certain integral functions, Proc.
London Math. Soc. 25 (1926); Theorem VII.) If Φ and Ψ are integrable functions
such that
x
Φ (t)Ψ (x − t)dt = 0
0
almost everywhere in the interval 0 < x < K, then Φ (t) = 0 almost everywhere
in (0, λ ) and Ψ (t) = 0 almost everywhere in (0, μ ), where λ + μ ≥ K. In his
book Operational Calculus (Elsevier, 1960) Jan Mikusinski proves Titchmarsh’s
theorem in the second chapter. He proves it following a particularly simple argument
found by Ryll-Nardzewski in 1952. As Mikusinski writes, Ryll-Nardzewski never
published this proof.
Z. Buczolich
M. Laczkovich
PROBLEM 125: INFELD

Originating in physics
We shall say a decent function of two variables f (x, y) satisfies the condition A if
there exists a function y = φ (x) such that
!
xfx + yfy = 0 (1)
4fx fy =1 (2)
for y = φ (x). [We see that φ (x) exists for f (x/y) and for f = ax + by, if ab = 1/4.]
Do we have the criterion: For every function f (x, y) satisfying A there exists F(x/y)
such that F(x/φ (x)) = f (x, φ (x)) (with the exception of the case of f = ax + by)?
PROBLEM 126: M. KAC

If:
1 !
0 f (x)dx = 0, (1)
1 2
0 f (x)dx = ∞, (2)
show that
n
1 f (x)
lim exp i √ dx = 0
n→∞ 0 n
(It is known that if 1 2

0 f (x)dx = A then the above limit = e−1/2 ).
Addendum. Solved affirmatively by A. Khintchin; it will appear in the fourth
communiqué on independent functions in Studia Math., Vol. 7.
PROBLEM 127: KURATOWSKI

Is it true that in every 0-dimensional metric space (in the sense of Menger-
Urysohn) that every closed set is an intersection of a sequence of sets which are
simultaneously closed and open? (The answer is affirmative for metric separable
spaces.)
Commentary
The problem is equivalent to the question whether for every metric space X, the
condition ind X = 0 implies that Ind X = 0 (see [1, p. 9] for the definitions). Indeed,
if every closed subset of X is an intersection of a sequence of open and closed sets,
then for every pair A, B of disjoint closed subsets of X there exists an open-and-
closed set U ⊂ X such that A ⊂ U ⊂ X\B: the set U can be defined by the formula
U= ∞ i=1 (Ui \Wi ), where U1 , U2 , . . . and W1 , W2 , . . . are decreasing sequences of
open-and-closed sets satisfying A = ∞ ∞
i=1 Ui and B = i=1 Wi (cf. [1, p. 155]). It
seems that the last implication, implicit in [1], was first explicitly stated in [4]. When
solving, in the negative, the famous problem whether the dimensions ind and Ind
coincide in metric spaces, P. Roy defined in [2] a metric space X such that ind X! = 0
and Ind X = 1 (a detailed discussion of this example is contained in [3]). Roy’s space
belongs among the most difficult examples in general topology.
1. J. Nagata, Modern dimension theory, Groningen, 1965.
2. P. Roy, Failure of equivalence of dimension concepts for metric spaces, Bull. Amer. Math. Soc.
68 (1962), 609–613.
3. , Nonequality of dimensions for metric spaces, Trans. Amer. Math. Soc. 134
(1968), 117–132.
4. J. Terasawa, On the zero-dimensionality of some non-normal product spaces, Sci. Rep. Tokyo
Kyoiku Daigaku Sec. A 11 (1972), 167–174.
R. Engelking
PROBLEM 128: NIKLIBORC

There is given, in a 3-dimensional space, a solid T which is unicoherent and
homogeneous. Let V(P) = T dtM /rPM . Assume that V(P) is a polynomial in P in
all of T + S. S is the surface of T. Show that T is an ellipsoid. It is known that if this
polynomial is of second degree then the theorem is true.
T rPM
M
S
dt
M
There are given two closed spaces S and S1 , each homeomorphic to the surface
of the sphere and constituting the boundary of a solid T. Suppose that V(P) =
T dtM /rPM is a constant in T1 (the solid boundary by S1 ). Prove that S and S1 are
homothetic ellipsoids. It is known that if S and S1 are homothetic, then they are
ellipsoids.
T
S
S1
T1
PROBLEM 130: KACZMARZ

Let {fn (t)} be a system of uniformly bounded, orthogonal, lacunary functions.
Does there exist a constant γ > 0, such that for every finite system of numbers
c1 , c2 , . . . , cn we have:
n
max |c1 f1 (t) + · · · + cn fn (t)| ≥ γ ∑ |ck |?
k=1
Remark: The system is lacunary if, for every p > 2 there exists a constant Mp , such
that
1/2
1 n
∑ c2k
p
|c1 f1 + · · · + cn fn |p dt ≤ Mp .
0 1
Let us formulate Kaczmarz’s problem in more modern notation. Let (Ω , μ ) denote

a probability space and let {φ1 , φ2 , . . .} denote an orthonormal system (OS) of
complex-valued functions on Ω . Such a system is said to be Sidon with constant
γ if for all coefficients {aj } one has
sup | ∑ aj φj (x)| ≥ γ ∑ |aj |. (1)

x∈Ω j∈N j∈N
An OS is said to be Λ (p) with constant Mp < ∞ if for all coefficients {aj }

1/2
|| ∑ aj φj ||Lp (Ω ) ≤ Mp ∑ |aj | 2
.
j∈N j∈N
An OS which is Λ (p) for all p ≥ 2 is said to be Λ (∞). With this notation, Kaczmarz’s
problem may be reformulated as: Is a uniformly bounded Λ (∞) orthonormal system
Sidon?
The topic of “Sidonicity" has been extensively studied in the context of characters
on groups. However, many problems regarding Sidon sets/systems can be formu-
lated in the context of general bounded orthonormal systems. We will report on
several recent results in this more general setting here. Detailed proofs will appear
in [3].
Let us briefly recall the development of the theory of Sidon sets/systems in the
character setting. In 1960 Rudin introduced Λ (p) sets and constructed a subset of
the integers which is Λ (∞) but which is not Sidon. See Section 3.2 and Theorem
4.11 of [9]. This provides a negative answer to Kaczmarz’s problem, although there
is no evidence there that Rudin was aware of the problem’s provenance. Let us
briefly discuss Rudin’s construction. He first proved that a Sidon set must be Λ (∞)
and, more restrictively, the set’s Λ (p) constants must satisfy Mp p1/2 . From this he
deduced that the size of the intersection of a Sidon set with an arithmetic progression
of size n must be log n. Rudin was then able to give a combinatorial construction
of a set which (1) had too large of an intersection with a sequence of arithmetic
progressions to be Sidon, yet 2) was Λ (p) for all p. He established the second
property by combinatorial considerations after expanding out Lp norms in the case
of even integer exponents.
On the other hand, much in the spirit of Kaczmarz’s problem, Rudin asked if
the stronger condition Mp p1/2 characterizes Sidon sets. In 1975 Rider [8] proved
that the Sidon condition (1) is equivalent to the following (superficially) weaker
Rademacher-Sidon condition. We say that {φj } has the Rademacher-Sidon property
with constant γ̃ if the following inequality is satisfied for all coefficients {aj }
sup | ∑ rj (ω )aj φj (x)|dω ≥ γ̃ ∑ |aj | (2)

x∈Ω j∈N j∈N
where rn denote independent Rademacher functions. In 1978 Pisier [6] proved

that Rudin’s condition Mp p1/2 implies the Rademacher-Sidonicity property.
Collectively these results show that Rudin’s condition characterizes Sidonicity in the
character setting. We note that both Rider’s and Pisier’s arguments make essential
use of properties of characters.
√
It is well known that Rudin’s condition Mp ≤ C p is equivalent to the condition
that
1/2
|| ∑ aj φj ||ψ2 ≤ C ∑ |aj |2
j j
where || · ||ψ2 is the Orlicz norm associated to the function ψ2 (x) := e|x| − 1. We
2
will denote this condition as ψ2 (C ).

Given the developments in the character setting, a natural relaxation of Kacz-
marz’s problem would be to ask if the ψ2 condition implies Sidonicity in the case of
general uniformly bounded orthonormal systems. Our first result is a construction
of an OS that gives a negative answer to this question.
Theorem 1. For all large n, there exists a real-valued OS {φ0 , φ1 , . . . , φn } with
n + 1 elements satisfying ||φj ||L∞ ≤ 7 and satisfying the ψ2 (C) condition with some
universal constant C, and such that
n n
1
|| ∑ aj φj ||L∞ √ ∑ |aj |
i=0 log n j=0
for some choice {ak }.

This construction makes essential use of Rudin-Shapiro polynomials.
On the other hand, the following result does provide a generalize of Pisier’s
theorem to the setting of general uniformly bounded orthonormal systems.
Theorem 2. Let {φj } be a ψ2 uniformly bounded OS. Then the OS obtained as a
5-fold tensor, {φj ⊗ φj ⊗ φj ⊗ φj ⊗ φj }, is Sidon. Moreover, if the 5-fold tensor of an
OS is Sidon, then the system itself is Rademacher-Sidon.
When applied to a system of characters it follows from the homomorphism property
that the system itself must be Sidon. It is an interesting problem to determine if the
result holds for the two-fold tensor. A corollary of this result is that the ψ2 condition
implies the Rademacher-Sidon property. We will discuss several proofs of this fact.
In fact, orthogonality beyond the ψ2 condition is not required.
Theorem 3. Let φ1 , φ2 , . . . denote a set of functions on a probability space (Ω , μ )
such that ||φj ||L2 = 1 and satisfying the ψ2 (C) condition. Then
sup | ∑ rj (ω )aj φj (x)|dω ≥ γ̃ ∑ |aj |. (3)

x∈Ω j∈N j∈N
with γ̃ := γ̃ (C).
This theorem will be a corollary of the following more general result.

Proposition 1. Let φ1 , φ2 , . . . , φn be a system of functions satisfying the ψ2 (C)
condition and ||φj ||L2 = 1. Let x1 , x2 , . . . , xn denote (real or complex) vectors in
a normed vector space satisfying ||xj || ≤ 1 and λ1 , λ2 , . . . , λn scalars. Then the
estimate

n n

∑ |λj |φj (ω )xj dω ≥ β ∑ |λj |
j=1 j=1
implies

n n

∑ λj rj (ω )xj dω ≥ γ ∑ |λj |
j=1 j=1
for γ := γ (β , C).
Let us explain how Proposition 1 implies Theorem 3. By truncation it suffices to
prove (3) for a finite system, as long as the bounds do not depend on the size of the
system. We then have that

n n n

∑ |λj | = ∑ |λj ||φj (x)| dx ≤ 2
∑ |λj |φj (x)φj (y)
j=1
dx. (4)
j=1 j=1 Ly∞
Using the ψ2 (C) hypothesis, we may apply Proposition 1 to replace the functions
{φj (x)} with Rademacher functions and remove the absolute values. This gives us

n n

γ ∑ | λj | ≤ ∑ λj rj (ω )φj (y) dω
j=1
j=1
Ly∞
which is Theorem 3.
Another variant of Kaczmarz’s problem would be to ask if the ψ2 (C) condition
implies that a system contains a large Sidon subsystem. In this direction recall the
Elton-Pajor Theorem (see [4] and [5]):
Theorem 4. (Elton-Pajor) Let x1 , x2 , . . . , xn denote elements in a real or com-
plex Banach space, such that ||xi || ≤ 1. Furthermore, for Rademacher functions
r1 , r2 , . . . , rn assume that γ n ≤ || ∑ni=1 ri (ω )xi ||dω . Then there exists real constants
c := c(γ ) > 0 and β := β (γ ) > 0 and a subset S ⊆ [n] with |S| ≥ cn such that
β ∑ |aj | ≤ || ∑ aj xj ||
j∈S j∈S
for all complex coefficients {ai }i∈S .

It immediately follows from Theorem 3 and the Elton-Pajor Theorem that a

finite uniformly bounded OS satisfying the ψ2 (C) condition must contain a Sidon
subsystem of proportional size. More precisely:
Theorem 5. Let φ1 , φ2 , . . . , φn be a system of functions satisfying ||φj ||L2 = 1,
||φj ||L∞ ≤ M and the ψ2 (C) condition. Then there exists a subset S ⊆ [n] of
proportional size |S| ≥ α (C, M)n such that
sup | ∑ aj φj (x)| ≥ γ ∑ |aj |.

x∈Ω j∈S j∈S
where γ = γ (C, M).

One interesting consequence of Proposition 1 is that one may replace the
Rademacher functions in the hypothesis of the Elton-Pajor Theorem with any
complex-valued functions satisfying the ψ2 (C) condition.
Our approach to Theorem 2 and Proposition 1 is rather elementary. The proofs
proceed by showing that one may efficiently approximate a bounded system
satisfying the ψ2 (C) condition by a martingale difference sequence. Once one is
able to reduce to a martingale difference sequence, one may apply Riesz product-
type arguments. This approach provides a new and elementary proof of Pisier’s
characterization of Sidon sets. It is worth noting that the first author obtained a
different elementary proof of Pisier’s theorem in 1983 [2]. The approach there,
however, like Pisier’s, relies on the homomorphism property of characters.
We have also found an alternate approach to Proposition 1 based on more
sophisticated tools from the theory of stochastic processes such as Preston’s theorem
[7], Talagrand’s majorizing measure theorem [11] and Bednorz and Latała’s recent
characterization of bounded Bernoulli processes. This approach yields a superior
bound for the size of γ (β , C) and allows for the following extension to more general
norms.
Theorem 6. Let φ1 , φ2 , . . . , φn be a ψ2 (C) system, uniformly bounded by M and let
x1 , x2 , . . . , xn be vectors in a normed space X. Then
n n
|| ∑ φj (ω )xj ||dω CM || ∑ rj (ω )xj ||dω . (5)
j=1 j=1
−1
In particular, one may take γ (C, β ) β C min M, log β1 in Proposition 1
for ψ2 (C) systems uniformly bounded by M.
Approximating ψ2 systems
The key ingredient in the proofs of Theorem 2 and Proposition 1 is the following:
Lemma 1. Let φ1 , φ2 , . . . , φn be real-valued functions on a probability space (Ω , μ )

such that
||φj ||L2 = 1 and ||φj ||L∞ ≤ C (6)

1/2
n n

∑ aj φj ≤C ∑ |aj | 2
(7)
j=1 j=1
ψ2
for all coefficients {aj }. For ε > 0, there exists a subset S ⊆ [n] such that |S| ≥
δ (ε , C)n and a martingale difference sequence {θj }j∈S satisfying ||θj ||L∞ ≤ C such
that:
||φj − θj ||L1 ≤ ε , (8)
and such that there exists an ordering of S, say j1 , j2 , . . . , jn , with

" #
E θjs |θjs , s < s = 0. (9)
−1
Moreover, one may take δ (ε , M) C−2 ε 2 log Cε .
For technical reasons we will require the following slightly stronger coefficient
version of this result.
Lemma 2. Let φ1 , φ2 , . . . , φn be functions uniformly bounded by C satisfying the
hypotheses of Proposition 1, and λ1 , λ2 , . . . , λn be complex coefficients such that
∑nj=1 |λj | = 1. Then there exists a set S ⊆ [n] and a martingale differences sequence
θj1 , θj2 , . . . indexed by elements of S satisfying (8) and (9), such that ∑j∈S |λj |
δ (C, ε ) > 0.
Proofs of these results may be found in [3].
Tensor systems: Theorem 2
We now describe how to obtain the first part of Theorem 2 from Lemma 2. For the
sake of exposition, we prove the result for real-valued systems. The complex case
can be handled in a similar manner. See [3].
Theorem 7. Let {φj } be an OS uniformly bounded by C and satisfying the ψ2 (C)
condition. Then the OS obtained as a 5-fold tensor, {φj ⊗ φj ⊗ φj ⊗ φj ⊗ φj }, is Sidon.
Proof. Let φ1 , φ2 , . . . , φ5 denote independent copies of the system {φi } on prob-
ability spaces Ω1 , Ω2 , . . . , Ω5 , respectively. Furthermore let Ω̃ := ⊗5s=1 Ωs and
(1) (2) (3) (4) (5)
let ri , ri , ri , ri , ri denote independent Rademacher functions on a distinct
probability space T. For a fixed set of coefficients {ai } and ε > 0, applying Lemma
1 gives a martingale difference sequence, {θj }, with the following properties:

C −1 n
∑ |ai | C −2 2
ε log
ε ∑ |ai |, (10)
i∈A i=1
for all i ∈ [n]
||θi || ≤ C, (11)
||φi − θi || ≤ ε . (12)
For 0 < δ < 1 and αi ∈ [−1, 1], define

(1) (2) (1) (3) (2)
μ(α ,δ ) := ∏
T i∈A
1 + δ αi ri θi (x1 ) ∏ 1 + δ αi ri ri θi (x2 ) ∏ 1 + δ αi ri ri θi (x3 ) ×
i∈A i∈A

(3) (4) (4)
∏ 1 + δ αi ri ri θi (x4 ) ∏ 1 + δ αi ri θi (x5 ) dω .
i∈A i∈A
Expanding out the product, and defining νS (x) := ∏i∈S θi (x), we see that
$
5
μ( α , δ ) = ∑ δ |S| ∏ αi ∏ θi (x1 ) . . . θi (x5 ) = ∑ δ |S| ∏ αi νS (xj ). (13)
S⊆A i∈S i∈S S⊆A i∈S j=1
Assuming δ < C we clearly have that
||μ(α ,δ ) ||L1 (Ω̃ ) = 1. (14)
To each subset S ⊆ A we may associated a Walsh function on, say, the probability
space T in the usual manner. In particular, let r1 , r2 , . . . , rm denote a system
of Rademacher functions on T and form the associated Walsh system element
associated to S by WS (y) := ∏i∈S ri (y). Given f such that ||f ||Lx∞ ≤ C, observe that

∑ C −2|S|
WS (y) νS , f ≤ ∏ 1 + C ri (y)θi (x)
−2
=1
S⊆A i∈A
Lx1
where we have used |C−2 θi (x)f (x)| ≤ 1. Since the function of y defined by the
expression on the left above is uniformly bounded by 1 and thus has L2 (T) norm at
most 1, Bessel’s inequality gives us that
∑ C−4|S| | νS , f |2 ≤ 1. (15)

S⊆A
Using (13), we have that

% &
$
5 ' ( ' (
μ( α , δ ) , φs = δ ∑ αj | θj , φj |5 + ∑ δ |S| ∏ αj | νS , φj |5 .
s=1 j∈A S⊆A j∈S
|S|≥2
We will estimate each of these terms separately. We start by estimating the second
using (15). Provided C8 δ 2 < 1, this gives
' (
∑ δ |S| ∏ αj | νS , φj |5 ≤ C8 δ 2 .
S⊆A j∈S
|S|≥2
We now consider the first term. By orthogonality and (12) we have that
' ( ' ( ' ( 3
| θj , φi | 5 ≤ | θj , φi | 2 φj , φi + ε .
From this and (15) we have, for j ∈

/ A, that

' ( 5 ' (

∑ |αj | θj , φi | ≤ ∑ | θj , φi |5 ≤ C4 ε 3 .
j∈A j∈A
For i ∈ A, using again (15), we have that

' (5
5
∑ αj θj , φi − αi |θi , φi | ≤ C4 ε 3 . (16)
j∈A
Finally we have
| φi , θi | ≥ φi , φi − | φi , φi − θi | ≥ 1 − Cε . (17)
Setting αj = sign(aj ) for j ∈ A, the preceding estimates imply

% &
n $
5
∑ ai φi , μ ( α , δ )
i=1 s=1

n
≥ δ ∑ |ai | θi , φi − δ 5
∑ |ai | ε − δ 3
∑ |ai | ε 3 − δ 2 ∑ |ai |.
i∈A i∈A i∈A
/ i=1
Using (17), provided ε C−1 , we have that

% &
n $
5
1 n n
∑ ai φi , μ ( α , δ ) ≥δ ∑ |ai | − C ε ∑ |ai | − C δ ∑ |ai | .
2 i∈A
4 3 8
i=1 s=1 i=2 i=1
Recalling (10), we have that the quantity above is

1 2 2 C −1 n
≥δ C ε log − C4 ε 3 − C8 δ ∑ |ai |.
2 ε i=1
The result follows by an appropriate choice of δ and ε .

We now prove the second part of Theorem 2, namely:
Proposition 2. Let {φi } denote a real-valued OS uniformly bounded by M such that
the k-fold tensored system {⊗ks=1 φs } is Sidon. Then {φi } has the Rademacher-Sidon
property.
Let k ≥ 2. If {⊗ki=1 φi } is Sidon, we have that

n k n

∑ ai gi (ω ) ∏ φi (xi ) dω ≥ c ∑ |ai |.
i=1
i=1 L∞ (Ω̃ ) i=1
We then claim that

n n k

∑ ai gi (ω )φi (x) dω ∑ ai gi (ω ) ∏ φi (xi ) dω .
i=1 i=1
L ∞ (Ω ) i=1 L∞ (Ω̃ )
Recognizing that each side can be interpreted as the expectation of the supremum of
a Gaussian process, this inequality follows from Slepian’s comparison lemma once
one has established the following lemma.
Lemma 3. In the notation above we have
1/2 1/2
n k k √ k n
2 2 2
∑ |ai | ∏ φi (xs ) − ∏ φi (xs ) ≤ k ∑ ∑ |ai | φi (xs ) − φi (xs ) .
i=1 s=1 s=1 s=1 i=1
Proof. Using the elementary inequality |∏ni=1 ai − ∏ni=1 bi | ≤ ∑ni=1 |ai − bi | for
sequence of real numbers of modulus less than or equal to 1, we have that
1/2 1/2
n k k k n
2 2 2
∑ |ai | ∏ φi (xs ) − ∏ φi (xs ) ≤M k−1
∑ ∑ |ai | φi (xs ) − φi (xs ) .
i=1 s=1 s=1 s=1 i=1
1/2
√ k n
2

∑ ∑ |ai | φi (xs ) − φi (xs )
2
≤ k .
s=1 i=1
To summarize, we have shown that

n n

∑ ai gi (ω )φi dω M c ∑ |ai |.
i=1
L ∞ (Ω ) i=1
One can replace the Gaussians random variables with Rademacher functions using
a truncation argument and the contraction principle. Alternatively, one may apply
Proposition 1. See [3] for details. This completes the proof.
A Counterexample: Theorem 1
The purpose of this section is to prove Theorem 1. We start with the following
elementary fact:
Lemma 4. Let 10 < n, p be positive real numbers. Then
) √
log nn−1/p ≤ p.
√ √
Proof. The claim is equivalent to pn1/p ≥ log n, or pn2/p ≥ log n. Taking
logarithms, this inequality is equivalent to log p + 2p log n ≥ log log n. For a fixed
n, the minimum of the left-hand side occurs when 1p − p22 log n = 0, or p = 2 log n.
Thus we have
2
log p + log n ≥ log log n + log 2 + 1 ≥ log log n,
p
which establishes the claim.

Next we estimate the Λ (p) constant of the first n elements of the Walsh system.
Lemma 5. Let Wi denote the i-th Walsh function defined on the probability space
Ω1 . Then
√ 1/2
log n n √ n
√ || ∑ ai Wi ||p p ∑ |ai | 2
.
n i=1 i=1
Proof. By the Hausdorff-Young inequality we have that

√ √
log n n log n
n
√ || ∑ ai Wi ||p ≤ √ n1/p −1/2 ( ∑ |ai |2 )1/2
n i=1 n i=1
1/2
) n
≤ log nn −1/p
∑ |an | 2
.
i=1
Applying Lemma 4 completes the proof.

For a fixed large n, let σi ∈ {−1, +1} be chosen such that
n √
|| ∑ σi Wi ||L∞ (Ω1 ) ≤ 6 n. (18)
i=1
In other words, ∑ni=1 σi Wi is a Walsh Rudin-Shapiro polynomial. The existence of

the coefficients σi is guaranteed, for instance, by Spencer’s “six standard deviations
suffices" theorem [10]. Next let ri denote independent Rademacher functions on Ω2 .
Furthermore define
⎛ 2 ⎞
n
log n
Ψ := ⎝1 + 2 ∑ ri ⎠ ,
n i=1

where Ω Ψ d μ = 1 + logn n . We now define an OS φ0 , φ1 , . . . , φn on the measure
space (Ω , Ψ d μ ) where Ω = Ω1 × Ω2 . For 1 ≤ i ≤ n define
−1/2 √
1 log n log n
φi := √ 1+ ri − √ σi Wi
Ψ n n
√
log n
where ||φi ||L∞ ≤ 1 × 1 × (1 + √
n
)≤ 2. Next define
−1/2 √
1 log n log n n 1 n
φ0 := √
Ψ
1+
n ∑ ri + √n ∑ σi Wi .
n i=1 i=1
−1/2 √
log n
Using that 1 + logn n ≤ 1, √1
Ψ n ∑ni=1 ri ≤ 1 and √1n ∑ni=1 σi Wi ≤ 1 by (18)
we then have that
√
log n
||φi ||L∞ ≤ 1×1× 1+ √ ≤ 7.
n
We now verify that this system satisfies orthonormality relations. For 1 ≤ i ≤ n

−1 √ 2
1 log n log n
|φi |2Ψ d μ = 1+ ri − √ σi Wi Ψ d μ
Ω ΩΨ n n

log n −1 log n
= 1+ 1+ = 1.
n n
For 1 ≤ i, j ≤ n we have
−1 √ √
1 log n log n log n
φi φjΨ d μ = 1+ ri − √ σi Wi × rj − √ σj Wj Ψ d μ
Ω Ω Ψ n n n
√ √
log n −1 log n log n
= 1+ ri − √ σi Wi × rj − √ σj Wj d μ = 0.
n Ω n n
Next we consider φ0 . We have
−1 √ 2
log n log n n 1 n
Ω
| φ0 | Ψ d μ = 1 +
2
n Ω
∑ ri + √n ∑ σi Wi d μ
n i=1 i=1

log n −1 log n
= 1+ 1+ = 1.
n n
For 1 ≤ i ≤ n, we have
√ √
log n −1 log n n 1 n log n
Ω
φ0 φiΨ d μ = 1 +
n Ω
∑
n i=1
ri + √ ∑ σi Wi × ri − √ σi Wi d μ
n i=1 n
√ √
log n −1 log n log n
= 1+ √ − √ = 0.
n n n
This completes the verification that the construction gives a uniformly bounded OS.
Next we verify the ψ2 (C) condition.
Lemma 6. The OS φ0 , φ1 , . . . , φn satisfies the ψ2 (C) condition for some fixed C
independent of n.
Proof. Let p ≥ 2, and ∑ni=1 |ai |2 = 1. We have
1/p 1/p
n
| ∑ni=1 ai φi |p n
|| ∑ ai φi ||Lp = Ψ dμ ≤ | ∑ ai φi | d μ
p
i=1 Ω |Ψ |p/2 Ω i=1
n √
log n n
≤ ||a0 ψ0 ||Lp (Ω ) + || ∑ ai ri ||Lp (Ω1 ) + √ || ∑ ai σi Wi ||Lp (Ω2 ) .
i=1 n i=1
Estimating the first term trivially, the second term using Khintchine’s inequality, and
the third using Lemma 5 gives us that
n
√
|| ∑ ai φi ||Lp (Ω ) p.
i=1
This completes the proof.

Finally, we show that these systems are not uniformly Sidon in n.
Lemma 7. There exists coefficients {a0 , a1 , . . . , an } with unit 1 norm, such that
n
1
|| ∑ ai φi ||L∞ (Ω ) √ .
i=0 log n
Proof. Set a0 = √1
log n
and ai = 1n , for 1 ≤ i ≤ n. Then

1 n log n −1/2
1 1
− √ ψ0 + ∑ ai φi = √ 1+ ×
log n n i=1 Ψ n

1 n 1 n n
log n n
1
− ∑ ri + ∑ ri − √ ∑ σn Wn + 3/2 ∑ σn Wn
n i=1 n i=1 n log n i=1 n i=1

n

1 log n 1 1
≤ √ + √ ∑ σn Wn √
log n n n i=1 log n
where we have used (18).

This completes the proof of Theorem 1.
1. W. Bednorz and R. Latała, On the boundedness of Bernoulli proceses, Ann. Math. 180 (2014),
1167–1203.
2. J. Bourgain, Sidon sets and Riesz products. Ann. Inst. Fourier (Grenoble) 35 (1985), no. 1,
137–148.
3. J. Bourgain and M. Lewko, Sidoncity and variants of Kaczmarz’s problem, Preprint.
4. J. Elton, Sign-embeddings of ln1 . Trans. Amer. Math. Soc. 279 (1983), no. 1, 113–124.
5. A. Pajor, Plongement de lk1 complexe dans les espaces de Banach. (French) [Embedding of
complex lk1 in Banach spaces] Seminar on the geometry of Banach spaces, Vol. I, II (Paris,
1983), 139–148, Publ. Math. Univ. Paris VII, 18, Univ. Paris VII, Paris, 1984.
6. G. Pisier, Ensembles de Sidon et processus gaussiens. (French) C. R. Acad. Sci. Paris Sér. A-B
286 (1978), no. 15, A671–A674.
7. C. Preston, Banach spaces arising from some integral inequalities. Indiana Univ. Math. J. 20
1970/1971 997–1015.
8. D. Rider, Randomly continuous functions and Sidon sets. Duke Math. J. 42 (1975), no. 4,
759–764.
9. W. Rudin, Trigonometric series with gaps. J. Math. Mech. 9 1960 203–227.
10. J. Spencer, Six standard deviations suffice. Trans. Amer. Math. Soc. 289 (1985), no. 2,
679–706.
11. M. Talagrand. Regularity of Gaussian processes. Acta Math., 159(1–2):99-149, 1987.
J. Bourgain
M. Lewko
PROBLEM 131: A. ZYGMUND

Given is a function f (x), continuous (for simplicity), and such that
1
f (x + t) − f (x)
lim dt < ∞, for x ∈ E, |E| > 0.
h→0 h t
Is it true that the integral
1 f (x + t) − f (x)
dt
0 t
may not exist almost everywhere in E? Similarly, for other Dini integrals?
Remark
This problem was raised by Professor Zygmund in a lecture given in Lwów in the
early thirties. The problem was solved positively by Marcinkiewicz. The solution
was published in his paper “Quelques théorèms sur les series et les fonctions,” Bull.
Math. Seminar at the University of Wilno, 1938. Although the original journal is
practically inaccessible now, the paper is reproduced in Marcinkiewicz’s collected
papers, published by the Polish Academy of Science, Warsaw, 1964.
PROBLEM 132: W. SIERPIŃSKI

February 25, 1936
Does there exist a Baire function F(x, y) (of two real variables) such that for every
function f (x, y) there exists a function φ (x) of one real variable (depending on the
function f ) for which f (x, y) = F(φ (x), φ (y)) for all real x and y.
Commentary
This problem remains unsolved. Sierpiński (Sur une function universelle de deux
variables réelles, Bull. Acad. Sci. Cracovie A (1936), 8–12) showed that assuming
the continuum hypothesis there is a function F(x, y) which possesses the mapping
properties described in the problem.
R. Daniel Mauldin
A function F : R × R → R is universal iff for every other function f (x, y) there

exists a function φ (x) with
f (x, y) = F(φ (x), φ (y))
for all x, y.
In Larson, Miller, Steprans, Weiss [1] we show that it is relatively consistent
with ZFC that there is no universal function (Borel or not), and we show that it
is relatively consistent that there is a universal function but no Borel universal
function. We also prove some results concerning higher arity universal functions.
For example, the existence of an F such that for every G there are unary h, k, j such
that for all x, y, z
G(x, y, z) = F(h(x), k(y), j(z))
is equivalent to the existence of a 2-ary universal F. However the existence of an F

such that for every G there are h, k, j such that for all x, y, z
G(x, y, z) = F(h(x, y), k(x, z), j(y, z))
follows from a 2-ary universal F but is strictly weaker.

1. Larson, Paul B.; Miller, Arnold W.; Steprans, Juris; Weiss, William A. R.; Universal functions.
Fund. Math. 227 (2014), no. 3, 197–246.
Arnold Miller, January 2015

PROBLEM 133: EILENBERG
There is given in a metric space E, a family of sets which are open-and-closed,
covering the space E. Find a family of sets which are simultaneously open and
closed and disjoint, covering the space E and smaller than the preceding family.
Remarks:
(1) A family of sets K is smaller than the family K1 if every set of the family K is
contained in a certain set of the family K1 .
(2) This problem includes Problem 127, of Prof. Kuratowski.
(3) For separable spaces, the solution is trivial.
Commentary
In general no such family can be found. Roy’s space X (cf. the Commentary to
Problem 127) satisfies the equality ind X = 0, so that (∗ ) X has a base consisting of
open-and-closed sets, and yet, since dim X > 0, X has an open—and by (∗ ) an open-
and-closed—cover which does not admit a refinement by pairwise disjoint open-
and-closed sets (see R. Engelking, Dimension Theory, Warszawa 1978, Proposition
3.2.2).
R. Engelking

Is the Cartesian product K1 × K2 of two indecomposable continua, K1 and K2 , of
necessity an indecomposable continuum?
Commentary
Let U be a nonvoid open subset of K1 with Ū = K1 . Let x be a point of K2 . Then

(Ū × K2 ) ∪ (K1 × {x}) is a proper subcontinuum of K1 × K2 with nonvoid interior
U × K2 . Every proper subcontinuum of an indecomposable continuum is nowhere
dense, so K1 × K2 must be decomposable. This is essentially the argument used by
F.B. Jones [Amer. J. Math. 70 (1948), 403–413] to show that the product of any two
nondegenerate continua is aposyndetic. Jones’ paper is one of the earliest published
results concerning the decomposability of products of continua, though it seems
unlikely that earlier workers could have been totally unaware of results along this
line.
While much interesting work continues to be done on indecomposable continua,
very little of it involves products since indecomposability is always lost.
Wayne Lewis
Tulane University

Is non-unicoherence of a locally connected continuum an invariant of locally
homeomorphic mappings?
The following example due to F.Marty, Bull. Sciences Math., II série, 61 (1937),
169-172, provides a negative answer. Let S and E be a sphere and an ellipsoid in R3 ,
both centered at zero, with E touching S exactly at two poles, X = S ∪ E and let f :
X → Y glue the antipodal points of S and E, respectively. Then X is not unicoherent,
f is a local homeomorphism, but f (S) and f (E) are projective planes with exactly
one point in common, hence their union Y is unicoherent.
More details and results on this topic can be found in a paper by A.Lelek, Fund.
Math. 45 (1957), 51–63.
Roman Pol

Can an interior mapping (that is to say, one such that open sets go over into open
sets) increase the dimensions?
Remark: This question occupied R. Baer, who obtained some partial results.
Addendum. A. Kolmogoroff, Annals Math. 38 (1937), 36–38, gave an example
of a continuous, interior mapping which increased the dimension from 1 to 2.
B. Knaster

Given is a continuous mapping f of a compact space X, such that dim X >
dim f (X) > 0. Does there exist a closed set Y ⊂ X such that dim Y < dim f (Y)?
In particular, does there exist, for every continuous mapping of a square into the
interval, a closed 0-dimensional set whose image consists of a certain interval? We
assume about the set X that it has the same dimension in every one of its points.
Commentary
The particular question about mappings from a square onto an interval was settled by
J.L. Kelley [2] in a paper containing results of his dissertation under G.T. Whyburn.
Kelley states in the paper that he owes the theorem to S. Eilenberg and L. Zippin.
Using a neat category argument be proves the following:
Theorem If f is a mapping from the unit square I 2 onto the unit interval I, then
there is a closed, totally disconnected subset K of I 2 such that f (K) = I.
Indeed, Kelley shows that if A is the space of all subsets of I 2 (with the Hausdorff
metric) which map onto I under f and ε > 0, then the subset of A consisting of those
members of A all of whose components have diameter less than ε is a dense open
set in A. Thus the set of all members of A which are closed and totally disconnected
is a dense Gδ in A.
In the same paper, using a similar category argument, Kelley also proves that if
f is a monotone open mapping of a compact metric space onto a finite dimensional
metric space Y, then there is a closed and totally disconnected subset K of X such
that f (K) = Y if and only if the set of points on which f is one-to-one is a totally
disconnected subset of Y.
We found only three other papers with results related to Problem 137 and we have
not obtained copies of these papers. The results given below are from the reviews of
these papers.
In 1955, A Kosinski [4] proved that if f is a monotone open mapping from a

compact metric space X onto a finite dimensional space Y and f −1 (y) is nondegen-
erate for each y ∈ Y and ε > 0, then there is a finite collection O1 , O2 , . . . , On of open
sets with disjoint closures,each of diameter less than ε and such that f maps ni=1 Oi
onto Y, a result which follows from Kelley’s earlier theorem. In 1956, B. Knaster
[3] showed that Kosinski’s result cannot be proved without the assumption that f
be open. In 1956, R.D. Anderson [1] showed that if X is a compact metric space
which is either 1-dimensional or a subset of a 2-manifold and f is a monotone open
map from X onto a compact metric space Y and f −1 (y) is nondegenerate for each
point y ∈ Y, then there is a closed and totally disconnected subset K of X such that
f (K) = Y. In addition Anderson gave examples to show that his result could not be
established under the assumption that X be n-dimensional for n ≥ 2 or even that X
be a manifold for n ≥ 4.
We have a slight improvement on Kelley’s theorem about mappings from I 2
onto I. We can show that if X is a locally compact, locally connected metric space
with the property that each connected open subset of X is cyclically connected and f
is a mapping from X onto I, then there is a Cantor set K in X such f (K) = I. We note
that Kelley’s argument applies for mappings from I n onto I and perhaps it could
be altered to obtain our result. Our proof is elementary and somewhat similar to
Kelley’s but is not a category argument. In view of our result and Knaster’s example
it seems to us that there are interesting questions which are special cases of the
original problem, and which to our knowledge remain unanswered. And it seems
appropriate to us to consider special cases with restrictions on the range space. We
suggest two problems. If X is a continuum which is 2-dimensional at each of its
points, and f maps X onto I, must there be a closed, 0-dimensional subset of X which
maps onto I under f ? If f maps the unit cube I 3 onto I 2 and the set of points at which
f is one-to-one is totally disconnected, must there be a closed 0-dimensional subset
of I which maps onto I 2 under f ?
1. R.D. Anderson, Some remarks on totally disconnected sections of monotone open mappings,
Bull. Acad. Polon. Sci. Cl. III 4 (1956), 329–330.
2. J.L. Kelley, Hyperspace of a continuum, Trans. Am. Math. Soc. 52 (1942), 22–36.
3. B. Knaster, Sur la fixation des decompositions, Bull. Acad. Polon. Sci. Cl. III 4 (1956), 193–196.
4. A. Kosinski, A theorem on monotone mappings, Bull. Acad. Polon. Sci. Cl. III 3 (1956), 69–72.
William S. Mahavier
North Texas State University
May 17, 1936
(a) Any compact convex set located in a linear space of type (B0 ) is an absolute
retract.
(b) A set compact and convex, in the sense of Wilson, is an absolute retract. [A set
Y ⊂ X is a retract with respect to X is there exists a continuous function f ∈ Y X
such that f (y) = y for y ∈ Y. We call a compact space an absolute retract if it
is a retract in every space which is metric, separable, containing it.] Absolute
retracts have the fixed point property: (vide K. Borsuk, Fund. Math. 17.) A set
X is convex in the sense of Wilson if, for every x, y ∈ X and 0 ≤ t ≤ 1 there exists
one and only one point z ∈ X such that ζ (x, z) = tζ (z, y) = (1 − t)ζ (x, y).
Commentary
These results have been extended, and placed in a context that does not involve any
hypothesis of separability, compactness, or completeness of the metric space, nor
any restriction on the convexity.
Call an (arbitrary) metric space Z an AR (Absolute Retract) if it is a retract of
each metric space containing it as a closed subset; and call an (arbitrary) metric
space Z an ES(M) (Extensor Space for Metric spaces) if for every (arbitrary) metric
space X and each closed A ⊂ X, every continuous f : A → Z has an extension F :
X → Z. It was shown by Dugundji (An extension of Tietze’s theorem, Pac. J. Math.
1 (1951), 353–367) that (a) Every convex subset (not necessarily closed) of any
locally convex linear space is an ES(M), and (b) A metric space is an ES(M) if and
only if it is an AR.
It is also proved in the same paper that the surface S = { x : x = 1 } of the
unit ball B in a normed linear space is an ES(M) if and only if S is compact. This
result answers a generalization of Problem 36; it shows that Brouwer’s fixed-point
theorem for the unit ball B of any normed linear space L (this theorem is equivalent
to the non-retractability of B onto S) is valid if and only if L is finite-dimensional.
J. Dugundji
PROBLEM 139: ULAM

Is every one-to-one continuous mapping of the Euclidean space into itself
equivalent to a mapping which brings sets of measure 0 into sets of measure 0?
Addendum. Theorems: von Neumann
(a) A compact group of transformations of the Euclidean space is equivalent to a
group of transformations carrying sets of measure 0 into sets of measure 0.
(b) Let fn be a sequence of one-to-one mappings of Euclidean space. There exists a
homeomorphism h such that the mappings hfn h−1 carry sets of measure 0 into
sets of measure 0.
PROBLEM 140: ULAM

Two mappings (not necessarily one-to-one) f and g of a set E into a part of itself
are called equivalent if there exists a one-to-one mapping h of E into itself such that
f = hgh−1 . What are necessary and sufficient conditions for the existence of such
an h?
Commentary
It is necessary and sufficient that the algebras E, f and E, g be isomorphic. In
other words, the directed graphs over E with the arrows x → f (x) and x → g(x),
respectively, should be isomorphic. The problem has been extensively studied for
the case when E is the unit interval and the functions are required to be continuous
(see, e.g., Jan Mycielski, On the conjugates of the function 2|x| − 1 in [−1, 1], Bull.
London Math. Soc. 12 (1980), 4–8, where other references are given).
Jan Mycielski
PROBLEM 141: ULAM

In the group M of one-to-one measurable transformations of the circumference
of a circle into itself, two transformations which are rotations through different
irrational angle are not equivalent. An analogous theorem holds for the group of
transformations of the surface of the n-dimensional sphere into itself.
PROBLEM 142: ULAM; Theorem: GARRETT, BIRKHOFF

For every abstract group G there exists a set Z and a subset X contained in the
square of the set Z : X ⊂ Z 2 , such that the group G is isomorphic to a group of
all one-to-one transformations f of the set Z into itself, under which the mapping
(x, y) → (f (x), f (y)) carries the set X into itself.
Commentary
A group G is isomorphic to Aut(P) for some poset P. Hence G is also isomorphic to

Aut(L) for the distributive lattice 2P . This is proved in my paper: Sobre los grupos
de automorfismos, Revista de la Unión Mat. Argentina 11 (1946), 247–256.
Garrett Birkhoff
PROBLEM 143: MAZUR

Let K denote the class of functions of two integer-valued variables x, y such that:
(1) The functions x, y, O, x + 1, xy belong to K;
(2) If the functions a(x, y), b(x, y), c(x, y) belong to K, then the function f (x, y) =
c(a(x, y), b(x, y)) also belongs to K;
(3) If the function a(x, y) belongs to K, then the function f (x, y) for which f (0, y) =
1, f (x + 1, y) = a(x, f (x, y)) belongs to K.
Does the class K contain the function
1 forx = y,
d(x, y) = ?
0 forx = y
Commentary
Evidently, the variables should be restricted to integers ≥ 0. So far as I know,

no solution to this problem has been published. An affirmative solution will be
presented here. Indeed, it will be shown that all primitive recursive functions of
two variables are definable. Of course, no other functions are definable.
The particular function requested by Mazur was sgn |x − y|. In the classical
terminology, which we shall follow, this is the characteristic function of x = y. It
would be equivalent to ask for the characteristic function of x = y. The use of this
function is central to the further development. Was Mazur aware of this?
In [2], I made a study of restricted schemes for obtaining all primitive recursive
functions. That paper starts with the standard definition of primitive recursive
functions as those obtained from certain initial functions (identity, zero, and
successor) by repeated substitutions and primitive recursions. It then considers
various restrictions of the recursion scheme, and asks what functions should be
adjoined to the initial functions in order that all primitive recursive functions can
be obtained using the restricted scheme being studied. Some improvements of my
results may be found in Gladstone [1].
Notice that the function defined by Mazur’s recursion scheme depends on only
one of the variables. That is, the scheme is essentially a definition by recursion
of a function of one variable. It will not change which functions of two variables
are definable if we allow functions of any number of variables, include all of the
usual initial functions together with the function F(x, y) = xy, and allow unrestricted
substitution, but allow recursion only for defining functions of one variable.
The recursion scheme for defining functions of one variable has the form
F0 = a, FSx = B(x, Fx).
Here the function F is defined in terms of a function B of two variables. To agree

with Mazur’s scheme, we must use only a = 1.
For any number c, the function cx may be defined by c0 = 1, csx = cx · c. We put
x
sgn x = 00 . Next, we define the function Fx = |x − 1| by F0 = 1, FSx = x. We can
use this to define the predecessor function Px = |x − 1| · sgn x.
We can now define addition in an important special case. If, for every x, either
Gx = 0 or Hx = 0, then
Gx + Hx = P(SGx · SHx).
This enables us to define functions by piecing. For example, if
Ax when Cx = 0,
Fx =
Bx when Cx > 0,
then
Fx = Ax · 0Cx + Bx · sgn Cx,
and this sum can be defined as above. This is very useful.

Let Rx be the distance from x to the smallest number of the form 6n ≥ x. We see
that
PRx if Rx > 0,
R0 = 1, RSx =
P(5x) if Rx = 0.
This is a legitimate definition, since piecing is allowed. But x = y if and only if

2x · 3y is a power of 6, and this is expressed by R(2x · 3y ) = 0. Thus the characteristic
function of equality if 0R(2 ·3 ) , and the function sgn |x − y| requested by Mazur is
x y
sgn R(2x · 3y ). This completes the affirmative solution of Mazur’s problem.

The above construction would not be possible if we used a = 0 instead of a = 1.
Indeed, as is easily seen, we could then obtain only functions which are monotone
in each variable. However, if we adjoined 0x as well as xy to the initial functions,
then all of the other functions defined above could be obtained. Indeed, if c > 0,
then cx − 1 can be defined by a recursive definition with a = 0, and cx = S(cx − 1).
We have P0 = 0, PSx = x, and can put |x − 1| = Px + 0x . This addition is definable,
as explained above. Finally, we may define Fx = RSx by a recursion with a = 0, and
then we have Rx = FPx + 0x .
We want to show that all primitive recursive functions can be obtained using
a = 1, or using a = 0 and adjoining 0x as well as xy to the initial functions. Now
in [2], I showed that all primitive recursive functions can be obtained if we adjoin
x + y and Q to the usual initial functions, where Q is the characteristic functions of
squares. I used only the recursion scheme with a = 0. However, the use of a = 0
may be replaced by the use of a = 1. Indeed, if Fx is defined by
F0 = 0, FSx = B(x, Fx),
then Gx = SFx is defined by
G0 = 1, GSx = SB(x, PGx),
and we can then obtain Fx = PGx. Thus it will be sufficient to define x + y and Q,
using a = 0.
We first define the function Fx = [x1/2 ] by
SFx if Sx = (SFx)2 ,
F0 = 0, FSx =
Fx otherwise.
This is a legitimate definition, since we allow piecing and know the characteristic
function of equality. But x is a square if and only if x = [x1/2 ]2 , so we obtain a
definition of Q.
We may define the function Fx = [log2 x] for x > 0 in a quite similar way. Let
SFx if Sx = 2SFx ,
F0 = 0, FSx =
Fx otherwise
We can now define addition by x + y = [log2 (2x · 2y )].

It follows that all primitive recursive functions can be obtained by adjoining xy to
the initial functions and allowing recursion only to define functions of one variable,
with the restriction that a = 1. The same result holds with a = 0, if we adjoin xy and
0x to the initial functions. Also, we see that Mazur’s class consists of all primitive
recursive functions of two variables.
Gladstone [1, §4] showed that instead of adjoining x + y and Q to the initial
functions, as in [2], it would be sufficient to adjoin only x + y, provided that we did
not restrict the value of a. An examination of his proof shows that only a = 0, 1, 2
are used, and it is easily seen that a = 1 alone would be sufficient. We could allow
just a = 0 if we also adjoined 0x to the initial functions. Thus the same results are
obtained with x + y as with xy.
My proof was suggested in part by the form of Gladstone’s proof. He first defined
the characteristic function of powers of 2, and then noted that x = y if and only if
2x + 2y is a power of 2. I first defined the characteristic function of powers of 6, and
then noted that x = y if and only if 2x · 3y is a power of 6.
1. M.D. Gladstone, Simplifications of the recursion scheme, J. Symbolic Logic, 36 (1971),
653–665.
2. R.M. Robinson, Primitive recursive functions, Bull. Amer. Math. Soc. 53 (1947), 925–942.
Raphael M. Robinson
University of California Berkeley
Let K denote a sphere in a separable space of type (B). Does there exist a one-to-
one mapping of K into the interval 0 ≤ x ≤ 1 under which the image of every open
set in K is a set of positive measure?
Solution
The answer to the problem is yes. This may be seen as follows. Let {xn }∞ n=1 be a
countable dense subset of K. For each pair of positive integers n, m, let T(n, m) =
{ x ∈ K : x − xn = 1/m }. Set T = { T(n, m) : n ≥ 1, m ≥ 1 } and set R = K − T.
Then R is a dense Gδ subset of K which is 0-dimensional and dense-in-itself. Let M
be a copy of the Cantor set lying in J, the set of all irrational numbers between 0 and
1, and such that the Lebesgue measure of M is zero. Of course, J − M is a dense-
in-itself 0-dimensional complete separable metric space. According to a theorem of
Mazurkiewicz [1, p. 144], there is a homeomorphism g of R onto J − M. Let h be a
one-to-one Borel measurable map of T onto I − (J − M). Set f (x) = g(x), if x ∈ R
and f (x) = h(x), if x ∈ K − R. Then f is a one-to-one Borel measurable map of K
onto I. Clearly, if U is an open subset of K, then f (U) has a positive measure.
1. K. Kuratowski, Topology, Volume I, Academic Press, New York, 1966.
R. Daniel Mauldin
PROBLEM 145: ULAM
Given is a countable sequence of sets An . Find necessary and sufficient conditions
for the possibility of introduction of a countably additive measure m(An ) such that
m(∑ An ) = 1, m(p) = 0; (p) denotes a set composed of a single point. [Possibly a
stronger condition: m(Ap ) = 0 for a certain given subsequence pk .] We demand that
the measure should be defined for each of the sets of a Borel ring of sets over the
sequence An .
Commentary
A solution is given by S. Banach in Sur les suites d’ensembles excluant l’existence

d’une mesure, Coll. Math. 1 (1948), 103–108.
Jan Mycielski
PROBLEM 146: ULAM
It is known that in sets of positive measure there exist points of density 1 [that
is to say, points with the property that the ratio of the length of intervals to the
measure of the part of the set contained in these intervals tends to 1 (if the length
of the interval converges to 0)]. Can one determine the speed of convergence of this
ratio for almost all points of the set?
Commentary
It will be more convenient to write about the natural product measure μ in the
Cantor space C = {0, 1}ω , i.e., the measure μ defined by the formula μ (A) =
(the Lebesgue measure of the set { ∑∞ i=0 xi /2

i+1 : (x , x , . . .) ∈ A }). Let A ⊂ C
0 1
be μ -measurable. For any ε > 0, we put S(n, ε , A) = { (x0 , . . . , xn−1 ) ∈ {0, 1}n :
2n μ (A ∩ V(x0 , . . . , xn−1 )) > 1 − ε }, where V(x0 , . . . , xn−1 ) = { (y0 , y1 , . . .) ∈ C :
(y0 , . . . , yn−1 ) = (x0 , . . . , xn−1 ) }. By the Lebesgue density theorem (see [2]) and an
easy compactness argument we have the following.
Theorem. For every ε > 0
lim |S(n, ε , A)|/2n = μ (A).

n→∞
On the other hand, it is not hard to prove that for every sequence ε0 , ε1 , . . . such
that 1 ≥ ε0 ≥ ε )1 ≥ · · · and εn → 0 there exists a measurable set A ⊆ C such that
μ (A) = 1 − ε0 and
S(n, εn , A) = φ
for n = 0, 1, . . ..
For related material see [1].
1. A. Ehrenfeucht and J. Mycielski, An infinite solitaire game with a random strategy, Colloq.
Math. 42 (1979), 115-118.
2. J.C. Oxtoby, Measure and Category, Springer-Verlag, New York, 1970.
Jan Mycielski
PROBLEM 147: AUERBACH, MAZUR
September 4, 1936
Suppose that a billiard ball issues at the angle 45◦ from a corner of the rectangular
table with a rational ratio of the sides. After a finite number of reflections from the
cushion will it come to one of the remaining three corners?

Let P(x1 , . . . , xn ) denote a polynomial with real coefficients. Consider the set
of points defined by the equation P(x1 , . . . , xn ) = 0. A necessary and sufficient
condition for this set not to cut the Euclidean (real) space is: All the irreducible
factors of the polynomial P in the real domain should be always nonnegative or
always nonpositive.
We will interpret “not to cut” to mean that the real zero set of P has (non-
empty) path-connected complement. Problem 148 admits an elementary solution
via reduction to. . .
Lemma 1. If Q ∈ R[x1 , x2 ] and each irreducible factor of Q is always nonnegative

on R2 or always nonpositive on R2 , then the complement of the real zero set of Q is
path-connected.
Auerbach’s Problem naturally leads to important recent advances in parts of algo-
rithmic
algebraic geometry: polynomial factorization and nonnegativity. Algorithms that are
practical and efficient (as of early 2015) for multivariate polynomial factorization
over C are detailed in [Gao03, CL07]. Algorithms that take numerical instability
into account (in the coefficients and/or the final answer) include [G+04, Zen09].
For an arbitrary f ∈ R[x1 ] with degree D and exactly t monomial terms, all general
algorithms for factorization over R have complexity super-linear in D. However,
such an f has at most 2t − 1 real roots (thanks to Descartes’ Rule). So counting just
the degree 1 factors may admit a faster algorithm when t is fixed. Such algorithms,
with complexity polynomial in log D (and the total bit size of all the coefficients
when f ∈ Z[x1 ]), appear in [BRS09] (counting bit operations, for t = 3) and [Sag14]
(counting field operations, for any constant t). The multivariate case is touched upon
in [Ave09, Gre15].
The relationship between nonnegativity and sums of squares was advanced by
Hilbert and Artin, and has since evolved into a beautiful intersection of optimization,
real algebraic geometry, and convexity. See, e.g., [BPT12] and the references
therein.
Our key reduction to Lemma 1 hinges on the following fact:
Lemma 2. (See [Sch00, Thm. 17, Pg. 75, Sec. 1.9]) Suppose P∈R[x1 , . . . , xn ]\R[x1 ]
is irreducible in R[x1 , . . . , xn ]. Then there is a polynomial
Φ ∈ R[w1 , . . . , wn−1 , y1 , . . . , yn−1 ] \ {0}
with the following property: If α2 , β2 , . . . , αn , βn ∈ R and Φ (α2 , . . . , αn ,

β2 , . . . , βn ) = 0, then P(x1 , α2 x2 + β2 , . . . , αn x2 + βn ) is irreducible in R[x1 , x2 ]. !
"
Solution to Problem 148: The case n=1 follows immediately upon observing that,
up to real affine transformations, the only irreducible non-constant polynomials in
R[x1 ] are x1 and x12 + 1. So assume n ≥ 2 and let Z be the zero set of P in Rn .
Sufficiency: Let u = (u1 , . . . , un ) and v = (v1 , . . . , vn ) lie in Rn \ Z. Since Z is
closed, u (resp. v) in fact lies in an open neighborhood U (resp. V) of points
contained in the same connected component of Rn \Z as u (resp. v). Since the
complement of any real algebraic hypersurface is open, Lemma 2 implies we
can find α2 , β2 , . . . , αn , βn such that the specialization Q(x1 , x2 ) := P(x1 , α2 x2 +
β2 , . . . , αn x2 + βn ) satisfies: (a) each irreducible factor Pi of P specializes to an
irreducible factor Qi of Q and (b) Pi is nonnegative on all of Rn (resp. nonpositive
on all of Rn ) if and only if Qi is nonnegative on all of R2 (resp. nonpositive on
all of Rr). In particular, the condition Φ = 0 from Lemma 2 enables us to pick
(β2 , . . . , βn ) arbitrarily close to (u2 , . . . , un ), and (α2 , . . . , αn ) arbitrarily close to
(v2 − u2 , . . . , vn − un ), so that both Q(u1 , 0) and Q(v1 , 1) are nonzero.
Let W denote the zero set of Q in R2 and note that the real 2-plane
H := {(x1 , α2 x2 + β2 , . . . , αn x2 + βn )}(x1 ,x2 )∈R2 ⊂ Rn
yields a pair (H, H ∩Z) affinely equivalent to (R2 , W). By Lemma 1 (and Conditions
(a) and (b)), R2 \W is path-connected, and thus H\Z is path-connected. Moreover,
by our choice of the αi and βi , both U and V intersect H\Z. So U and V, and thus u
and v, are connected by a path in Rn n\Z. "
!
(Necessity): Suppose now that Rn \Z is path-connected, but P has an irreducible
factor Pi attaining both positive and negative values on Rn . Then Pi must be positive
(resp. negative) at some point u+ (resp. u− ) in Rn \Z since Rn \Z is open. By
assumption, there is a path in γ : [0, 1] −→ Rn \Z connecting u+ and u− . In particular,
Pi (γ (0))Pi (γ (1)) < 0, so by the Intermediate Value Theorem, Pi (γ (s)) = 0 for some
s ∈ (0, 1). In other words, γ ([0, 1]) intersects Z, which is a contradiction. "
!
To prove Lemma 1 we will apply the following two facts:
Proposition 3. If X ⊂Rr is finite, then any two points of R2 \X can be connected by
a smooth quadric curve Γ ⊂ R2 \X. "
!
Lemma 4. Suppose f , g∈C[x1 , x2 ] have respective degrees d and e, and no common
factor of positive degree. Then f = g = 0 has no more than de solutions in C2 . "
!
Proposition 3 follows easily by using an invertible affine map to reduce to the special
case of connecting (0, 0) and (1, 0) via the graph of cx1 (1 − x1 ) for suitable c: The
finiteness of X guarantees that all but finitely many c will work. Lemma 4 is a special
case of Bézout’s Theorem, but can also be easily derived from the basic properties
of the univariate resultant (see, e.g., [Sch00, App. B]).
Proof of Lemma 1: Let W denote the real zero set of Q in R2 . By Proposition 3
it suffices to prove that, under the hypotheses of Lemma 1, W is finite. It clearly
suffices to restrict to the special case where Q is non-constant and irreducible in
R[x1 , x2 ]. Note also that the irreducibility of Q and the assumption on the sign of Q
are invariant under composition with any invertible real affine
map.
Consider now any root ζ = (ζ1 , ζ2 ) ∈ R2 of Q. If δ := ∂∂xQ (ζ ), ∂∂xQ (ζ ) = 0 then,
1 2
by composing with a suitable invertible real affine map, we may assume δ = (1, 0).
In particular, by Taylor expansion, we see that Q changes sign in a non-empty
horizontal line segment containing ζ . Therefore, every root of ζ of Q must satisfy
∂Q ∂Q
∂ x1 (ζ ) = ∂ x2 (ζ ) = 0.
Let Q1 · · · Qr be the factorization of Q over C[x1 , x2 ] into factors of positive
degree, irreducible in C[x1 , x2 ]. The Galois group G := Gal(C/R) has order 2, is
generated by complex conjugation (·), and acts naturally on the Qi . In particular,
G acts trivially on Qi if and only if Qi ∈ R[x1 , x2 ]. So r must be even when r ≥ 2,
since Q is irreducible over R[x1 , x2 ]. Furthermore, r ≤2 since Qi Q̄i is invariant under
complex conjugation. So we either have r = 1 (with Q irreducible over C[x1 , x2 ]) or
r =2 (with Q̄1 = Q1 = Q̄2 = Q2 ). A simple calculation then shows that, in either case,
∂Q
∂ x has no common factors with Q. So W is finite by Lemma 4. "
!
1
I am grateful to Dan Mauldin and Joe Buhler for bringing Auerbach’s problem to
my attention. I also thank Guillaume Chèze, Michel Coste, Erich Kaltofen, Marie-
Francoise Roy, and Zbigniew Szafraniec for valuable comments.
Research partially supported by NSF grant CCF-1409020.
[Ave09] Avendaño, Martín, “The number of roots of a lacunary bivariate polynomial on a line,” J.
Symb. Comp. 44 (2009), pp. 1280–1284.
[BRS09] Bihan, Frederic; Rojas, J. Maurice; Stella, Case E., “Faster Real Feasibility via Circuit
Discriminants,” proceedings of International Symposium on Symbolic and Algebraic Compu-
tation (ISSAC 2009, July 28–31, Seoul, Korea), pp. 39–46, ACM Press, 2009.
[BPT12] Blekherman, Grigoriy; Parrilo, Pablo A.; and Thomas, Rekha R., “Semidefinite Opti-
mization and Convex Algebraic Geometry,” MPS-SIAM series on optimization, book 13, SIAM,
2012.
[CL07] Chèze, Guillaume and Lecerf, Grégoire, “Lifting and recombination techniques for
absolute factorization,” J. Complexity, 23, no. 3, pp. 380–420, 2007.
[Gao03] Gao, Shuhong, “Factoring multivariate polynomials via partial differential equations,”
Math. Comp., 72, no. 242, pp. 801–822, 2003.
[G+04] Gao, Shuhong; Kaltofen, Erich; May, John P.; Yang, Zhengfeng; and Zhi, Lihong,
“Approximate factorization of multivariate polynomials via differential equations,” in proceed-
ings of ISSAC 2004, pp. 167–174, ACM, New York, 2004.
[Gre15] Grenet, Bruno, “Bounded-degree factors of lacunary multivariate polynomials,” Math
ArXiV preprint http://arxiv.org/abs/1412.3570. Submitted for publication.
[Sag14] Sagraloff, Michael, “A near-optimal algorithm for computing real roots of sparse
polynomials,” in proceedings of ISSAC 2014, pp. 359–366, ACM, 2014.
[Sch00] Schinzel, Andrzej, Polynomials with Special Regard to Reducibility, Encyclopedia of
Mathematics and its Applications, Book 77, Cambridge University Press, 2000.
[Zen09] Zeng, Zhonggang, “The approximate irreducible factorization of a univariate polynomial.
Revisited,” in proceedings of ISSAC 2009, pp. 367–374, ACM, New York, 2009.
J. Maurice Rojas
Texas A&M University TAMU, College Station
dσ
M
S2
.P
S1
Let S1 and S2 denote closed and convex surfaces tangent at a point Q. Let S2 be
contained in the domain whose boundary is S1 .
Let
dσM
Vk (P) = k = 1, 2.
Sk rPM
Theorem: V1 (Q) > V2 (Q).

Let S denote a closed surface and f (M) a continuous function defined on S. Let
V(P) = S f (M)(1/rPM )dσM . Let us assume that the plane π has the property: If
P1 and P2 denote two arbitrary points of space, located outside a sufficiently large
sphere and symmetrically with respect to the plane π , then V(P1 ) = V(P2 ). Prove
that:
(1) The plane π is a plane of symmetry for the surface S.
(2) In points of symmetry M1 and M2 belonging to S we have f (M1 ) = f (M2 ).
PROBLEM 151: WAVRE

November 6, 1936; Prize: A “fondue” in Geneva; Original manuscript in French
Does there exist a harmonic function defined in a region which contains a cube
in its interior, which vanishes on all the edges of the cube? One does not consider
f ≡ 0.
Addendum. Does there exist an algebraic function f (z) homomorphic in every
point of a curve traced on a surface of Riemann and such that one has
f (x)
dz = 0, f (z) ≡ 0,
−x
z
the point x being contained in a certain domain? The curve will be open. One
should find f (z) and .
Prize: “Fondant” in Lwów

November 6, 1936; Prizes: For the computation of the frequency: 100 grammes of
caviar. For a proof of the existence of frequency, a small beer. For counter example:
A demitasse
A disk of radius 1 covers at least two points with integer coordinates (x, y) and
at most 5. If we translate this disc through vectors nw (n = 1, 2, 3, . . .), where w
has both coordinates irrational and their ratio is irrational, then the numbers 2, 3, 4
repeat infinitely many times. What is the frequency of these events for n → ∞? Does
it exist?
P2
P3 P3
P2 P4 P2
P3 P3
P2
Figure 152.1
Commentary
By the 2-dimensional equipartition theorem (see J.F. Koksma, Diophantische

Approximationen, repr. Chelsea 1936, Springer-Verlag,
√ Berlin-New York, 1974),
the
√ frequencies exist, 2 has the frequency 4 − √ 3 − (2/3) π , 3 has the frequency
2 3 − 4 + (π /3), and 4 has the frequency 1 − 3 + (π /3). These frequencies are
the areas of the three parts P2 , P3 , and P4 of the square [0, 1] × [0, 1] such that
if the center of a circle of radius 1 is in Pi then the circle covers i lattice points
(Figure 152.1).
Jan Mycielski
PROBLEM 153: MAZUR

Prize: A live goose, November 6, 1936
Given is a continuous function f (x, y) defined for 0 ≤ x, y ≤ 1 and the number
ε > 0; do there exist numbers a1 , . . . , an ; b1 , . . . , bn ; c1 , . . . , cn with the property that

n

f (x, y) − ∑ ck f (ak , y)f (x, bk ) ≤ ε
k=1

in the interval 0 ≤ x, y ≤ 1?
Remark: The theorem is true under the additional assumption that the function
f (x, y) possesses a continuous first derivative with respect to x or y.
Commentary
Grothendieck proved, in his thesis [3], that Problem 153 is equivalent to the
approximation problem—a problem which was considered to be one of the central
open problems of functional analysis. The statement of the approximation problem
is the following: Is every compact linear operator T from a Banach space X into a
Banach space Y a limit in norm of operators of finite rank? A Banach space Y is
said to have the approximation property if the answer to the question is positive for
every choice of X and T. Every space Y with a Schauder basis has the approximation
property and thus Problem 153 is closely related to the basis problem. (Does every
separable Banach space have a Schauder basis?) Since a Hilbert space has a basis
(and thus the approximation property) it follows from Grothendieck’s reformulation
of Problem 153 that this problem (in its original formulation) has a positive answer
if f satisfies a Lipschitz condition of order 1/2.
The answer to the approximation problem (and thus also to Problem 153 as well
as the basis problem) is negative. This was proved in 1972 by P. Enflo [2]. The
Goose promised to the solver of 153 was given to Enflo a year or so later following
his lecture on his solution in Warsaw.
By modifying Enflo’s construction, Davie [1] showed that for every α < 1/2
there is a function f (x, y) satisfying a Lipschitz condition of order α but for which
there is no approximation of the form required in the statement of Problem 153.
Further major progress related to the approximation problem was made by
A. Szankowsky. He showed [6] that there is a natural example of Banach space
which fails to have the approximation property, namely, the space of all operators
from 2 into itself with the usual operator norm. He proved also [5] that unless Y is
“very close” to a Hilbert space it has a subspace failing the approximation property.
A detailed treatment of questions related to the approximation property is
contained in [4].
1. A.M. Davie, The Banach approximation problem, J. Approx. Theory 13 (1975), 392–394.
2. P. Enflo, A counterexample to the approximation problem in Banach spaces, Acta Math. 130
(1973), 309–317.
3. A. Grothendieck, Produits tensoriels topologiques et espaces nucléaires, Memo. Amer. Math.
Soc., 16 (1955).
4. J. Lindenstrauss and L. Tzafriri, Classical Banach spaces, Vol. I, Sequence Spaces (1977), Vol.
II, Function Spaces (1979), Springer-Verlag, Berlin-New York.
5. A. Szankowski, Subspaces without the approximation property, Israel J. Math. 30 (1978),
123–129.
6. A. Szankowski, B(H ) does not have the approximation property, Acta Math. 147 (1981),
89–108.
Joram Lindenstrauss
PROBLEM 154: MAZUR

September 15, 1936
Let {φn (t)} be an orthogonal system of continuous functions and closed in C.
(a) If f (t) ∼ a1 φ1 (t) + a2 φ2 (t) + · · · is the development of a given continuous
function f (t) and n1 , n2 , . . . denote the successive indices for which an1 = 0, . . .,
can one approximate f (t) uniformly by the linear combinations of the functions
φn1 (t), φn2 (t), . . .?
(b) Does there exist a linear summation method M such that the development of
every continuous function f (t) into the system {φn (t)} is uniformly summable
by the method M to f (t)?
Roughly speaking, the question is: Let {ϕn } be an orthonormal system on a

segment I. Suppose it satisfies the Weierstrass approximation property, that is, linear
combinations are dense in C(I). Is a Fejer type theorem true ?
The question can be decomposed into two sub-questions:
First:
Is it true that every f ∈ C(I) admits approximation by linear combinations which
use only those {ϕn }, which are presented in the Fourier expansion of f with nonzero
amplitudes ?
(In this case one may say that spectrum-preserving approximation is possible).
And second:
Suppose that the system does satisfy the last property. Is it true that corresponding
linear combinations could be obtained from the Fourier partial sums by a certain
linear procedure?
Unfortunately, a counter-example can be constructed to each of the questions,
see [1].
References
[1] A.M.Olevskii, On spectrum-preserving approximation, Soviet Math. Dokl. 41 (1990), no. 2,

215–218.
A. Olevskii
PROBLEM 155: MAZUR, STERNBACH
November 18, 1936
Given are two spaces X, Y of type (B), y = U(x) is a one-to-one mapping of the
space X onto the whole space Y with the following property: For every x0 ∈ X there
exists an ε > 0 such that the mapping y = U(x), considered for x belonging to the
sphere with the center x0 and radius ε , is an isometric mapping. Is the mapping
y = U(x) an isometric transformation?
This theorem is true if U −1 is continuous. This is the case, in particular, when
Y has a finite number of dimensions or else the following property: If y1 + y2 =
y1 + y2 , y1 = 0, then y2 = λ y1 , λ ≥ 0.
PROBLEM 156: WARD

March 23, 1937; Original manuscript in English
A surface x = f (u, v), y = g(u, v), z = h(u, v), f , g, h being continuous functions,
has at each point a tangent plane in the geometric sense; also, to each point of
the surface there corresponds only one pair of values of u, v. Does there exist a
representation of the surface by functions x = f1 (u, v), y = g1 (u, v), z = h1 (u, v), in
such a manner that the partial derivatives exist and the Jacobians ∂ (f2 , f3 )/∂ (u, v),
∂ (f3 , f1 )/∂ (u, v), ∂ (f1 , f2 )/∂ (u, v) are not all zero, except for a set N of values of
u, v such that the corresponding set of points of the surface has surface measure
(in Caratheodory’s sense) zero? [Let (x, y, z) be a point of a surface S, and P a plane
through (x, y, z). Then if, for every ε > 0, there exists a sphere K(ε ) of center (x, y, z),
such that the line joining (x, y, z) to any other point of S ·K(ε ) always makes an angle
of less than ε with P, we say that P is the tangent plane to S at (x, y, z).]?
PROBLEM 157: WARD

March 23, 1937; Prize: Lunch at the “Dorothy"
f (x) is a real function of a real variable, which is approximately continuous. At
each point x, the upper right-hand approximate derivative of f (x) (that is,
f (x + h) − f (x)
lim ,
h→0+ h
neglecting any set of values of h which has zero density at h = 0) is positive. Is f (x)
monotone increasing?
Second Edition Remark
The answer is affirmative. This was proved by R. J. O’Malley in his paper:

A density property and applications, Trans. Amer. Math. Soc. 199 (1974), 75–87,
MR 50 13402. In fact, O’Malley’s monotonicity theorem is much stronger than the
statement in Problem 157.
PROBLEM 158: STOÏLOW

May 1, 1937; Original manuscript in French
Construct an analytic function f (x) continuous in a domain D admitting there
a perfect discontinuum set P of singularities such that f (P) is a discontinuous set.
Such a function would permit one to form a “quasi-linear” function; that is to say,
one which has the following properties:
(1) The function is continuous and univalent in the whole plane z.
(2) The function tends toward ∞ for |z| → ∞.
(3) The function has a perfect set of singularities.
See: S. Stoïlow, Remarques sur les fonctions analytiques continues dans un

domaine où elles admettant un ensemble parfait discontinu de singularités, Bull.
Math. Soc. Roum. Sci. 38 (1936), 117–120.

May 22, 1937
Let Φ denote the set of all continuous functions defined in (0, 1), f (0) = 0, 0 ≤
f (x) ≤ 1 for 0 ≤ x ≤ 1. Let
∞
P(x) = ∑ an xn
n=0
be a power series and let Pk (x) denote the kth partial sum.
Does there exist a power series P(x) with the following property; for every ε >
0 there exists N(ε ) such that for every function f ∈ Φ there exist n ≤ N, so that
|f (x) − Pn (x)| < ε ?
Addenda. In this formulation the answer is negative. Since
max | sin2 2n π x − sin2 2m π x| = 1 (m = n),

0≤x≤1
there does not exist a function which approximates both sin2 2n π x and sin2 2m π x
with a precision ≤ 1/3 in the interval 0, 1. If the requested universal N(1/2)
existed, then by taking
sin2 2π x, sin2 22 π x, . . . , sin2 {2N(1/2)+1 π x}, sin2 {2N(1/2)+2 π x},
we would conclude on the basis of Problem 159 that for a certain k ≤ N(1/3), the
polynomial Pk (x) would approximate simultaneously sin2 2n π x and sin2 2m π x for
n = m with precision ≤ 1/3.
Sternbach
Let Φ denote an arbitrary set of continuous functions defined in (0, 1), f (0) = 0,
|f (x)| ≤ N; in order that the set φ should have the requested property, it is necessary
and sufficient that the functions of the set Φ be equicontinuous.
Mazur
PROBLEM 160: MAZUR

June 10, 1937
Let G denote a metric group.
(1) Let the group G be complete and have the property that for every ε > 0, every
element a ∈ G has a representation a = a1 a2 · · · an , where (ak , e) < ε . Is the
group G connected in the sense of Hausdorff? (That is to say, it cannot be
represented as a sum of two disjoint, closed sets = 0.)
/
(2) If a group G is connected in the sense of Hausdorff, is it then arcwise connected?
Commentary
Part (1) is still open. It is not known whether every topological group generated
by every neighborhood of the identity is connected. However, for locally compact
groups, the answer is yes [1]. It is known (see [2]) that a compact group is connected
if and only if each of its elements has an nth root for each integer n.
A. Gleason remarks that part (1) is open even for the completion of the infinite
cyclic group Z with special metrizations which he defines as follows:
Let p1 , p2 , . . . be a sequence of positive integers and a1 , a2 , . . . be a decreasing
sequence of real numbers such that
= 1 > a2 >a3 > · · · and an → 0;
(1) a1
pn+1
(2) an − 1 ≥ 1;
pn
(3) for every m there exist integers k and n1 , . . . , nk , all ni larger than m, such that
the greatest common divisor of pn1 , . . . , pnk is 1.
E.g., the sequences pn = (n + 1)! − 1 and an = (1/n) satisfy (1), (2), and (3) with
k = 2.
For any x ∈ Z we put
N(x) = inf { ∑|ki |ai : ki ∈ Z, ∑ki pi = x } .
Theorem 1. N is a norm in Z, i.e., N(x + y) ≤ N(x) + N(y), N(−x) = N(x),

N(0) = 0 and N(x) > 0 for all x = 0. Moreover the metric N(x − y) turns Z into a
dense topological group which is generated by every neighborhood of 0.
Proof. The first two conditions for a norm are obvious from the definition of N. To
prove the third we show first
(4) if |x| ≤ pn and x = 0 then N(x) ≥ an . By (1), if N(x) ≥ 1 then (4) is true.
Suppose that N(x) < 1. Consider any representation x = ∑ ki pi such that the largest i
for which ki = 0, call it i0 , satisfies i0 > n. Since | ∑ ki pi | = |x| ≤ pn ≤ pi0 −1 it follows
that
pi0
∑|ki | ≥ − 1 + |ki0 |.
pi0 −1
Hence, by (1) and (2), ∑ |ki |ai ≥ 1. Therefore there exists a representation x = ∑ ki pi
in which i ≤ n for all i with ki = 0. Hence, since x = 0, N(x) ≥ an .
By (1) and (4) N(pn ) = an → 0. Thus Z is dense at 0 and hence everywhere. Also
by (3), Z is generated by every neighborhood of 0. Q.E.D.
Theorem 2 N(Z) is dense in the interval [0, sup N(Z)].

Proof. Since Z is generated by every neighborhood of 0 for every ε > 0 and x, y ∈ Z
there exists a sequence z1 , . . . , zn such that z1 = x, zn = y and N(zi − zi+1 ) < ε for
i = 1, . . . , n − 1. Hence N(Z) is dense in the interval [N(x), N(y)].
Problem 1. Is sup N(Z) < ∞ possible? (Notice that if

an pn+1
→ +∞,
pn
. p /
n
then sup N(Z) = ∞, in fact N → ∞.)
2
Problem 2. Are there sequences a1 , a2 , . . . and p1 , p2 , . . . satisfying (1), (2), and
(3) and such that Z can be partitioned into two non-empty sets X and Y such that for
any sequences x1 , x2 , . . . and y1 , y2 , . . . where xi ∈ X and yi ∈ Y, the three series
∞ ∞ ∞
∑ N(xi − xi+1 ), ∑ N(yi − yi+1 ), ∑ N(xi − yi )
1 1 1
cannot all converge? (A positive solution of this problem would yield a negative
answer to Mazur’s problem, since the Hausdorff completion of Z would then be a
disconnected complete metric abelian group generated by every neighborhood of
the identity, its partition into two non-empty closed sets being given by the closures
of X and Y.)
Part (2) of Problem 160 has a negative answer. The simplest example of a
compact metric connected and not locally connected (in fact indecomposable) group
is the subgroup of K ω , where K is the circle group which consists of the sequences
(x0 , x1 , . . .) ∈ K ω such that xi = xi+1
2 for i = 0, 1, . . . (such groups are called Van
Danzig solenoids).
For other open problems on connected groups, see [3].
1. D. Montgomery and L. Zippen, Topological transformation groups, Interscience 1955.
2. J. Mycielski, Some properties of connected compact groups, Coll. Math. 5 (1958), 162–166.
3. J. Mycielski, On the extension of equalities in connected topological groups, Fund. Math. 44
(1957), 300–302.
Jan Mycielski
Remark
A negative solution to part (1) was given by T. Christine Stevens in her paper,
Connectedness of complete metric groups, Colloq. math. 50 (1986), 233–240.
PROBLEM 161: M. KAC

Let rn be a sequence of integers such that

n−1
lim rn − ∑ rk = ∞.
n→∞
k=1
One has then

!
sin 2π r1 x + · · · + sin 2π rn x
lim → E a < √ < b
n→∞ 0≤x≤1 n
1 b
e−y dy.
2
=√
π a
2
(One can put, for example, rn = 2n .)
Problem: Is the theorem true for rn = 2n ?
Remark
This problem is discussed in Mark Kac’s conference lecture, on pages 17–26.

Also, see Kac’s discussion in his address: Probability methods in some problems
of analysis and number theory, Bull. Amer. Math. Soc. 55 (1949), 390-408.

July 3, 1937; Prize: Dinner at “George’s"
We assume that f (x) IS measurable (L), periodic, f (x + 1) = f (x) and f (x) = +1
or −1. Do we have, almost everywhere,
lim sup f (nx) = +1, lim inf f (nx) = −1?

n→∞ n→∞
More generally: If fn (x) are measurable, uniformly bounded, and fn (x+1/n) ≡ fn (x),
do we have then
lim sup fn (x) = constant almost everywhere?

n→∞
lim inf fn (x) = constant almost everywhere?

n→∞
Addendum. A more general theorem, formulated by Professor Banach, is true: If f (x)

is an arbitrary measurable function with period 1, then one has almost everywhere
the relations:
lim f (nx) = → essential upper bound f (x),

n→∞ 0≤x≤1
lim f (nx) = → essential lower bound f (x),

n→∞ 0≤x≤1
M. Eidelheit
October 16, 1937
Second Edition Remark
The theorem quoted by Eidelheit was published in a more general form in S.

Mazur and W. Orlicz, Sur quelques propriétés de fonctions périodiques et presque-
périodiques, Studia Math. 9 (1940), 1–16; MR 3-107 (Théorème 1, p. 5). While
Eidelheit attributes this theorem to Banach, Mazur and Orlicz do not mention
Banach’s name.
As for the second half of problem 162, Mazur and Orlicz give an equivalent
formulation and a partial answer on pp. 13–14 of their paper.
PROBLEM 163: J. von NEUMANN

July 4, 1937; Prize: A bottle of whiskey of measure > 0
Given is a completely additive and multiplicative Boolean algebra B. That is to
say:
(1) B is a partially ordered set with the relation a ⊂ b.
(2) Every set S ⊂ B has the least upper (greatest lower) bound Σ (S)(Π (S)). [We
write: Σ (a, b) = a + b, Π (a, b) = ab, Σ (B) = 1, Π (B) = 0.]
(3) We have a general “distributive law” (a + b)c = ac + bc.
(4) Every element a ∈ B has an (according to (3), unique) “inverse” in B: a+(−a) =
1, a(−a) = 0.
A measure in B is a numerical function:
= 0, for a = 0,
1. μ (a)
> 0, for a = 0.
2 . ai ∈ B (i = 1, 2, . . .), ai aj = 0 for i = j
imply μ (Σi (ai )) = Σi μ (ai ).
Obviously, one has to determine:

(5) If S ⊂ B, (a, b ∈ S, a = b) =⇒ ab = 0, then S is at most countable.
Question: When does there exist a “measure” in B? Remarks: As one verifies

without difficulty, the following “generalized distributivity” law is necessary!
(6) Let ai1 ≤ ai2 ≤ · · · for i = 1, 2, 3, . . ., then we have
∏ ∑(aij ) = ∑ ∏(aij(i) )
i j j(i) i
without the assumption that ai1 ≤ ai2 ≤ · · · this characterizes, according to Tarski,
the “atomic” Boolean algebras.
(1) to (5) do not imply (6). Counterexample: The Boolean algebra of Borel sets,
modulo sets of first category. Example for (1) to (6): Measurable sets (or Borel
sets) modulo sets of measure 0 when one employs Lebesgue measure. Is (5), (6)
sufficient?
Commentary
This is really two problems. The more specific one—given a complete Boolean
algebra B, do conditions (5) (the “countable chain condition”) and (6) (a distributive
law) imply the existence of a (finite, strictly positive, countably additive) measure
on B—is still not completely solved. It is known that the answer no is relatively
consistent with the usual axioms (ZFC) of set theory. In fact, I have given a
counterexample [8, Th. 5, pp. 164–166] assuming the falsity of Souslin’s hypothesis;
and it is known that there are models of set theory in which Souslin’s hypothesis is
false [6, 12]. There are also models in which Souslin’s hypothesis is true [11]; it
is not known whether the answer to Problem 163 would then be affirmative. (But I
conjecture that the answer, in ZFC, is always no.)
The more general problem raised here by von Neumann is that of finding
(assuming conditions (1)–(6)) necessary and sufficient conditions for the existence
of a measure. Such conditions have been given (a) by Maharam [8], simplified by
Hodges and Horn [4], (b) by Kelley [7]. A simpler condition and a variant on the
conditions of Kelley were given by Ryll-Nardzewski (quoted in [7, p. 1176]). Under
stronger hypothesis, simpler conditions have been given by Horn and Tarski [5].
A survey of this question is in Sikorski’s book [10, pp. 201–204]. None of these
answers is entirely satisfactory, in that they require the existence of a sequence of
subsets of B with certain properties, and this is not easy to verify in specific cases.
Perhaps this is inevitable from the nature of the question. A different (but also not
easily applicable) answer is implicit from the structure theory of [9]; it is necessary
and sufficient that B be isomorphic to a countable direct sum B1 +!B2 + · · · , where
Bn is either an atom or the measure algebra of some product of unit intervals with
Lebesgue product measure.
The method of attack in [8] led to a further interesting question. If B does have a
measure μ , it is then an abelian group (under symmetric difference) with invariant
metric d(x, y) = μ (xΔ y). Thus a necessary condition is that B be metrizable. It will
then have an invariant metric ρ ; and on defining λ (x) = ρ (x, o) (where o is the zero
element) we obtain a “continuous outer measure” λ on B; that is, a finite nonnegative
function, vanishing only at o, satisfying (i) λ (x ∨ y) ≤ λ (x) + λ (y), and (ii) if x1 ≥
x2 ≥ · · · and limn xn = o then limn λ (xn ) = 0. In [8] a further condition was imposed
to pass from a continuous outer measure to a measure. But it was asked whether the
existence of a continuous outer measure on B implies by itself the existence of a
measure. This, the “control measure” question is still open. For some partial results
and applications, see [1]! and [2].
The counterexample in [8] (assuming Souslin’s hypothesis false) has the follow-
ing remarkable property: each countably generated complete subalgebra B1 of B
does have a measure, though B does not. This raises two further open questions
concerning complete Boolean algebras with the countable chain condition:
(a) Is there such an algebra, with the above remarkable property, even if Souslin’s
hypothesis is true?
(b) If each countably generated complete subalgebra of B (a complete Boolean
algebra satisfying (5)) has a measure, what further conditions ensure that B has
a measure? (Of course, the measures on the subalgebras need not be consistent
with one another.) Would the existence of a continuous outer measure be
sufficient?
Von Neumann’s Problem 163 leads naturally to an even more fundamental one:
Given a (finitely additive) Boolean algebra A, under what conditions will A admit
a finitely additive (strictly positive, finite) measure? The solution of this problem
was the first step in Kelley’s treatment of the countably additive case; he showed
that A admits a finitely additive measure if and only if the nonzero elements of
A can be partitioned into countably many subsets each having positive “intersection
number” [7, pp. 1166, 1167]. That this condition does not hold automatically, even if
A satisfies the countable chain condition (5), is shown by an example of Gaifman [3].
As mentioned above, the complete Boolean algebras with countably additive
measures have an easily described structure [9]. It would be very interesting to have
a structure theory for finitely additive measures, the structure of which can be much
more complicated.
1. J.P.R. Christenson, Some results with relation to the control measure problem, Vector space
measures and applications Proc. Conf. Univ. Dublin, Dublin 1977), II, Lecture Notes in Math.
vol. 645, Springer-Verlag, Berlin-New York 1978, 27–34.
2. J.P.R. Christenson and W. Herer, On the existence of pathological submeasures and the
construction of exotic topological groups, Math. Ann. 213 (1975), 203–210.
3. H. Gaifman, Concerning measures on Boolean algebras, Pacific J. Math. 14 (1964), 61–73.
4. J.L. Hodges and A. Horn, On Maharam’s conditions for a measure, Trans. Amer. Math. Soc. 64
(1948), 594–595.
5. A. Horn and A. Tarski, Measures in Boolean algebras, Trans. Amer. Math. Soc. 64 (1948),
467–497.
6. T. Jech, Non-provability of Souslin’s hypothesis, Comment. Math. Univ. Carolinae 8 (1967),

291–305.
7. J.L. Kelley, Measures on Boolean algebras, Pacific J. Math. 9 (1959), 1165–1177.
8. Dorothy Maharam, An algebraic characterization of measure algebras, Ann. of Math. (2) 48
(1947), 154–167.
9. , On homogeneous measure algebras, Proc. Nat. Acad. Sci. USA 28 (1942),
108–111.
10. R. Sikorski, Boolean Algebras, 2nd. ed., Ergebnisse der Math. u. ihrer Grenzgebiete, N.S.
vol. 25, Springer-Verlag, Berlin 1964.
11. R.M. Solovay and S. Tennenbaum, Iterated Cohen extensions and Souslin’s problem, Ann. of
Math. (2) 94 (1971), 201–245.
12. S. Tennenbaum, Souslin’s problem, Proc. Nat. Acad. Sci. USA 59 (1968), 60–63.
Dorothy Maharam
PROBLEM 164: ULAM
Let a finite number of points including 0 and 1, be given on the interval [0, 1], a
number ε > 0, and a transformation of this finite set into itself T, with the following
property: For every point p, |p, T(p)| > ε . Let us call a “permissable step” passing
from the point p to T(p) or to one of the two neighbors (points nearest from the left
or from the right side) of the point T(p).
Question: Does there exist a universal constant k such that there exists a point
p0 from which, in a number of permissable steps [k/ε ] one can reach a point q which
is distant from p0 by at least 1/3?
PROBLEM 165: ULAM

Prize: Two bottles of wine
Let pn be a sequence of rational points in the n-dimensional unit sphere. The first
N points p1 , . . . , pN are transformed on N points (also located in the same sphere)
q1 , . . . , qN , all different. We define a transformation on the points pn , n > N, by
induction as follows: Assume that the transformation is defined for all points pν , ν <
n, and their images are all different. This mapping has a certain Lipschitz constant
Ln−1 . The Lipschitz constant of the inverse mapping we denote by Ln−1 . We define
the mapping at the point pn so that the sum of the constants Ln + Ln should be
minimum. (In the case where we have several points satisfying this postulate we
select one of them arbitrarily.)
Question: Is the sequence {Ln + Ln } bounded?
PROBLEM 166: ULAM

Let M be a topological manifold, f a real-valued continuous function defined on
M. We denote by GM f the group of all homeomorphic mappings T of M onto itself
such that f (T(p)) = f (p) for all p ∈ M.
Question: If N is a manifold not homeomorphic to M, does there exist f0 such
that GM
f0 is not isomorphic to any Gf ?
N
PROBLEM 167: ULAM

Let S denote the surface of the unit sphere in Hilbert space. Let f1 , . . . , fn
be finite system of real-valued, continuous functions defined on S. Let T be a
continuous transformation of S into part of itself. Does there exist a point p0 such
that fν (T(p0 )) = fν (p0 ), ν = 1, . . . , n?
Commentary
Klee [2] constructed a homeomorphism h of S onto the entire Hilbert space E, and
Bessaga [1] showed h could even be made a diffeomorphism. Now choose z ∈ E ∼
{0}, let f be a continuous linear functional on E such that f (z) = 0, and for each
x ∈ E let t(x) = x + z. Finally, let f1 = fh and T = h−1 th. Then f1 is a differentiable
real-valued function on S and T is a diffeomorphism of S onto S. If f1 (T(p0 )) =
f1 (p0 ), then with y = h(p0 ) it follows that f (t(y)) = f (y). That is impossible, for
f (t(y)) = f (y) + f (z).
1. C. Bessaga, Every infinite-dimensional Hilbert space is diffeomorphic with its unit sphere, Bull.
Acad. Polon. Sci. Ser. Sci. Math. Astronom Phys. 14 (1966), 27–31.
Soc. 74 (1953), 10–43.
V. Klee
PROBLEM 168: ULAM

Prize: Two bottles of beer
Does there exist a sequence of sets An such that the smallest class of sets
containing these, and closed with respect to the operation of complementation and
countable sums, contains all the analytic sets (on the interval)?
Commentary
If the continuum hypothesis or Martin’s axiom holds, then the answer is yes [1].
B.V. Rao [3] and R. Mansfield [2] showed that there is no sequence of sets An having
the required properties which are Lebesgue measurable.
Finally, Rao [4] showed that if one assumes the axiom of determinacy, then the
answer is no.
1. R.H. Bing, W.W. Bledsoe, R.D. Mauldin, Sets generated by rectangles, Pac. J. Math. 51 (1974),
27–36.
2. R. Mansfield, The solution to one of Ulam’s problems concerning analytic sets II, Proc. Amer.
Math. Soc. 26 (1970), 539–540.
3. B.V. Rao, Remarks on analytic sets, Fund. Math. 66 (1969/70), 237–239.
4. , Remarks on generalized analytic sets and the axiom of determinateness, Fund.
Math. 69 (1970), 125–129.
R. Daniel Mauldin
In Miller [1] Theorem 3 it is shown to be relatively consistent with ZFC that there
is no countable family H of sets of the reals such that every analytic set is in the
σ -algebra generated by H .
1. Miller, Arnold W.; Generic Souslin sets. Pacific J. Math. 97 (1981), no. 1, 171–181.
Arnold W. Miller
PROBLEM 169: E. SZPILRAJN
Does there exist an additive function μ (E), equal for congruent sets, defined for
all plane sets, and which is an extension of the linear measure of Caratheodory?
(0 ≤ μ (E) ≤ +∞)?
Commentary
In this problem, “additive” means “finitely additive” (for “countably additive” the
answer would, of course, be negative). The answer is yes, see, e.g., J. Mycielski,
Finitely additive invariant measures (I), Coll. Math. 42 (1979), 309–318.
J. Mycielski

Is every plane set all of whose homeomorphic plane images are Lebesgue
measurable (L), measurable absolutely? [That is to say, measurable with respect
to every Caratheodory function (“Massfunction”).]
This is true for linear sets; for plane sets an analogous theorem is true if
one replaces homeomorphisms by generalized homeomorphisms in the sense of
Mr. Kuratowski.
Addendum. Affirmative answer follows from an unpublished result of von
Neumann.
Commentary
The solution announced at the end of the problem holds for all Rn and rests on the
following theorem of von Neumann, later proved by Oxtoby and Ulam [1, 2]: If μ
is a Borel probability measure on I n = [0, 1]n , then μ is homeomorphic to the usual
product Lebesgue measure if and only if it is positive for nonempty open sets, zero
for points, and μ (∂ I n ) = 0. To deduce the solution of the problem from this theorem
suppose that A ⊆ Rn is not measurable relative to some Caratheodory measure μ .
Without loss of generality, we can assume A ⊆ I n and μ is homeomorphic

to Lebesgue measure by a homeomorphism h, but then h(A) is not Lebesgue
measurable.
1. J.C. Oxtoby and S.M. Ulam, Measure-preserving homeomorphisms and metrical transitivity,
Ann. of Math. (2) 42 (1941), 874–920.
2. J.C. Oxtoby and V.S. Prasad, Homeomorphic measures in the Hilbert Cube, Pac. J. Math. 77
(1978), 483–497.
Jan Mycielski
PROBLEM 163: J. SCHEIER, S. ULAM
Let T(A) denote the set of all mappings of a set A into itself. An operation is
defined for pairs of elements of the set T: U(f , g) = h for all h ∈ T(A). (U(f , g) ≡ 1.)
Assumptions:
(1) U(f , g) is associative; that is, U(f , U(g, h)) = U(U(f , g), h).
(2) U(f , g) is invariant with respect to permutations of the underlying set; i.e., if p
is a permutation of the set A, then U(p−1 fp, p−1 gp) = p−1 U(f , g)p.
Theorem. U(f , g) = f (g) (composition).
Trivially, if U is any solution to the problem, then so also is U (f , g) := U(g, f ).

In particular, there is the trivial solution U(f , g) = g ◦ f (which should probably be
excluded).
Here is an example of a U of a different form. Let A be a finite set. Define the
function s : AA → AA by the rule s(f ) = Id if f ∈ Sym(A) and s(f ) = f otherwise.
Set U(f , g) = s(f ) ◦ s(g). Note that s(f ◦ g) = s(f ) ◦ s(g) if f and g are in the image
of s. The proof that U satisfies the desired properties is immediate from the further
relations: s2 (f ) = s(f ) and s(f σ ) = (sf )σ for σ in Sym(A) and f ∈ AA , where f σ =
σ −1 f σ .
For an arbitrary set A one can set t(f ) = Id if f : A → A is an onto function and
t(f ) = f otherwise. Then W(f , g) = t(f ) ◦ t(g) works with the same proof as above.
However, there is another possible interpretation of Problem 171. The third line
“U(f , g) = h for all h ∈ T(A)” may mean that one insists that the product be onto. To
this end, define V(f , g) = f ◦ g if both f and g are in Sym(A) and otherwise V(f , g) =
s(f ) ◦ s(g). This V satisfies properties (1) and (2), and moreover is surjective in the
sense that any h is of the form h = V(f , g) for some f , g ∈ AA .
Daniel Goldstein
PROBLEM 172: M. EIDELHEIT

June 4, 1938
A space E of type (B) has the property (a) if the weak closure of an arbitrary
set of linear functionals is weakly closed. [A sequence of linear functionals fn (x)
converges weakly to f (x) if fn (x) → f (x) for every x.]
The space E of type (B) has the property (b) if every sequence of linear
functionals weakly convergent converges weakly as a sequence of elements in the
conjugate space Ē.
Question: Does every separable space of type (B) which has property (a) also
possess property (b)?

July 23, 1938
Let A denote the set of all linear operations mapping a given space of type (B)
into itself. Is the set of operations in A which have continuous inverses dense in A
(under the usual norm)?
Commentary
It is now well known that the set of invertible linear operators in an infinite
dimensional Hilbert space is connected and open, but is not dense. See Problem 109
in P. Halmos: A Hilbert space problem book, Van Nostrand, Princeton, 1967. For
a characterization of the closure of the invertible operators, see R. Bouldin, Proc.
Amer. Math. Soc. 108 (1990), 721-726.
R. Daniel Mauldin

July 23, 1938
Let U(x) be a linear operation defined in a space of type (B0 ) mapping this space
into itself and such that the operation x− λ U(x) has inverses for sufficiently small λ .
Can we then have
(x − λ U)−1 = x + λ U(x) + λ 2 U(U(x)) + · · ·?
PROBLEM 175: BORSUK

August 10, 1938
(a) Is the product (Cartesian) of the Hilbert cube Q with the curve which is shaped
like the letter T, homeomorphic with Q?
(b) Is the product space of an infinite sequence of letters T homeomorphic to Q?
Commentary
The answer to this striking problem has been given by R.D. Anderson [1], who
showed that (a) the products T × ∏i [−1, 1]i and ∏i [−1, 1]i are strongly homeomor-
phic, and (b) if Yn × ∏i Xi is for each n strongly homeomorphic to ∏i Xi and the Xi s
are compact then ∏i Yi is homeomorphic (∼ =) to ∏i Xi . The details of Anderson’s
proof have never been published, but the definition of “strongly homeomorphic”
and another proof of (b) appeared in [15]. Anderson’s proof of S(a) depended on a
construction of lattice-isomorphic bases of open sets in T × Q and Q, respectively.
By the same method he showed that (c) countable products of nondegenerate dendra
yield Hilbert cubes.
The above results and Anderson’s further study of the topological properties of
the Hilbert cube initiated intense investigations on infinite products of compact
ARs. A. Szankowski [12] has given an alternate proof of (c) based on a Hilbert-
cube analogue of the lakes of Wada construction. A crucial step has been made
by J.E. West, who in a series of papers [15, 16, 17] has developed a technique of
establishing homeomorphisms of product spaces which enabled him to show that
(d) the class X = { X : X × Q ∼ = Q } is closed with respect to the mapping cylinder
construction, and (e) if X1 , X2 and X1 ∩ X2 are in X then so is X1 ∪ X2 . He also proved
that if X × Q ∼ = Q then X × ∏i [−1, 1]i is strongly homeomorphic to ∏i [−1, 1]i ,
thereby establishing that (f) ∏i Xi ∼ = Q whenever all the Xi s are in X and contain
more than one point. Both Szankowski’s and West’s proofs heavily depended on the
properties of Z-sets discovered by Anderson in [2].
West’s theorems have been re-proved by several authors who, in particular,
replaced West’s “interior-approximation” technique by the use of a theorem of
M. Brown giving a sufficient condition for the inverse limit of a sequence of
compacta to be homeomorphic to each of them. See West’s expository article [18]
for a closer discussion. (Subsequent to this article were [20, 5, 10, 11, 13, 6, 9].)
In 1975, R.D. Edwards gave the definitive characterization of Q-factors by
proving that X × Q ∼ = Q whenever X is a compact AR (see [16]). Combined with the
Anderson-West result (f) mentioned above, this shows that infinite cartesian mul-
tiplication of nondegenerate compact ARs always yields Hilbert cubes. A slightly
stronger result is that X ∼
= Q whenever X is a compact AR and for each n there are
nondegenerate spaces X1 , . . . , Xn with X ∼
= X1 × · · · × Xn (see [14]).
In fact, West and Edwards established their results more generally for factors of
manifolds modeled on Q; i.e., West showed that Y = { Y : Y × Q is a Q-manifold }
is closed with respect to the mapping cylinder construction, and Edwards showed
that locally compact ANRs are in Y . The implications are crucial to (1) identifying
Q-manifolds and specifically copies of the Hilbert cube, and (2) theory of ANRs. Not
going into details we mention here that West’s results were the basis for both Curtis-
Schori’s solution of Wojdysawski’s problem if hyperspaces of Peano continua were
Hilbert cubes [7, 8] and West’s [19] solution of Borsuk’s problem on finiteness of
homotopy type of compact ARs. Similarly, Edwards’ theorem was the basis for a
general characterization of Q-manifolds (see [14]).
Results analogous to that of Edwards remain valid also for non-locally compact
ANRs once Q is replaced by a suitable normed linear space (see [18] and [14] for
references).
K. Borsuk’s Problem 175 appeared to be of great significance, especially if
viewed as an obvious specification of the more general problem of determining
which infinite products are homeomorphic to the Hilbert cube and how the singu-
larities of ARs behave under infinite multiplication (the latter is in the spirit of other
questions of Borsuk; see [4]). It was raised 18 years before any nontrivial factor of
a finite- or infinite-dimensional cube was constructed [3] and, with Wojdysawski’s
question on hyperspaces of Peano continua (posed in Fundamenta Math. in 1939),
has stimulated the very basic research on the topology of the Hilbert cube.
1. R.D. Anderson, The Hilbert cube as a product of dendrons, Amer. Math. Soc. Notices 11 (1964),
572.
2. , On topological infinite deficiency, Michigan Math. J. 14 (1967), 365–383.
3. R.H. Bing, The cartesian product of a certain nonmanifold and a line is E4 , Ann. of Math. 70
(1959), 399–412.
4. K. Borsuk, Theory of Retracts, Polish Scientific Publishers, Warsaw, 1967.
5. M. Brown and M. Cohen, A proof that simple-homotopy equivalent polyhedra are stably
homeomorphic, Michigan Math. J. 21 (1974), 181–191.
6. T.A. Chapman, Lectures on Hilbert Cube Manifolds, CBMS Lecture Notes #28, 1976.
7. D.W. Curtis and R.M. Schori, 2X and C(X) are homeomorphic to the Hilbert cube, Bull. Amer.
Math. Soc. 80 (1974), 927–931.
8. , Hyperspaces of Peano continua are Hilbert cubes, Fund. Math. 101 (1978),
19–38.
9. A. Fathi, A. Marin Y.M. Visetti, New proofs of Chapman’s CE mapping theorem and West’s
mapping cylinder theorem, Proc. Amer. Math. Soc. 67 (1977), 327–334.
10. M. Handel, The Bing staircase construction for Hilbert cube manifolds, Gen. Topology and
Appl. 9 (1978), 19-28.
11. , On certain sums of Hilbert cubes, preprint.
12. A. Szankowski, On factors of the Hilbert cube, Bull. Polon. Acad. Sci 17 (1969), 703–709.
13. H. Torunczyk, On factors of the Hilbert cube and metric properties of retractions, Bull. Polon.
Acad. Sci. 24 (1976), 757–765.
14. , Characterization of infinite-dimensional manifolds, Proceedings of the Inter-
national Conference on Geometric Topology (Warsaw, 1978) , 431–437, PWN, Warsaw, 1980..
15. J.E. West, Infinite products which are Hilbert cubes, Trans. Amer. Math. Soc. 150 (1970), 1–25.
16. , Mapping cylinders of Hilbert cube factors, General Topol. and Appl. 1 (1971),
111–125; Part II, 5 (1975), 35–44.
17. , Sums of Hilbert cube factors, Pacific J. Math. 54 (1974), 293–303.
18. , Cartesian factors of infinite-dimensional spaces, Topology Conference, Vir-
ginia Polytechnic Inst. and State University, Springer Lecture Notes in Math. #375, 1974,
249-268.
19. , Mapping Hilbert cube manifolds to ANRs, A solution of a conjecture of
Borsuk, Annals of Math. 106 (1977), 1–18.
20. R.Y.T. Wong and Nelly Kroonenberg, Unions of Hilbert cubes, Trans. Amer. Math. Soc. 211
(1975), 289–297.
H. Torunczyk

September 12, 1938
In a ring of type (B) (normed, complete linear ring with the norm satisfying
the condition: |xy| ≤ |x| |y|) containing a unit element there is given an element a
possessing an inverse a−1 .
(n) (n) (n)
Question: Does there exist a sequence of polynomials c0 I + c1 a + · · · + cm an
converging to a−1 ? (I = the unit element, c are numbers.)
Addendum. Answer is negative. Example: The ring of linear operations U(x) of
the space (C) into itself; U(x) = x(t2 ), 0 ≤ t ≤ 1.
M. Eidelheit
November 11, 1938
PROBLEM 177: M. KAC

September 11, 1938
What are the conditions which a function Φ (x, y) must satisfy in order that for
every pair of Hermitian matrices A and B the matrix Φ (A, B) is “positive definite”?
Remark
This problem is discussed in Mark Kac’s conference lecture, on pages 17–26.
PROBLEM 178: M. KAC

Let
1
φ (x, y) = .
1
x + 1y − 1
Prove that if
+∞ +∞

1
φ eiξ x dσ1 (x), eiξ x dσ2 (x) = ,
−∞ −∞ 1+ξ2
then σ1 (x) = α1 e−β1 |x| and σ2 (x) = α2 e−β2 |x| . (This is analogous to Cramer’s
theorem that if
+∞ +∞
eiξ x dσ1 (x) × eiξ x dσ2 (x) = e−ξ
2 /4
,
−∞ −∞
then σ1 (x) and σ2 (x) are of the form e−β1 ξ and e−β2 ξ .)
2 2
Remark
This problem is discussed in Mark Kac’ conference lecture, on pages 17–26.
PROBLEM 179: OFFORD

January 10, 1939; Original manuscript in English
If a0 , . . . , an are any real or complex numbers and if εν = ±1, ν = 1, 2, . . . , n, then
the following theorem is true:
|a0 + ε1 a1 + ε2 a2 + · · · + εn an | ≥ min |aν |

0≤ν ≤n
except for a proportion of at most A/n1/4 of the 2n sums.

Problems:
(i) To find a short proof of this result. √
(ii) When the as are all equal to 1 the size of the exceptional set is (A/ n)2n . Is this
the right upper bound whatever the numbers aν ?
Commentary
Littlewood and Offord (On the number of real roots of a random algebraic equation
(III), Mat. Sbornik, 12 (1943), n
√ 277–285) showed that the proportion of the 2 sums
which fail is at most c log n/ n and Erdős (On a lemma of Littlewood
√ and Offord,
Bull. Amer. Math. Soc., 51 (1945), 898–902) improved this to c/ n. Kleitman (On a
lemma of Littlewood and Offord on the distribution of certain sums, Math. Zeitschr.
90 (1965), 251–259) improved the bound to

n
"n#
2
and showed this is the best possible. Thus the answer to the second question is
no. Kleitman (On a lemma of Littlewood and Offord on the distribution of linear
combinations of vectors, Adv. in Math. 5 (1970), 155–157) generalized the problem
and result to vectors in Hilbert space.
Littlewood and Offord (op. cit.) used their result to consider the number of real
roots of an equation of the form
n
∑ εν aν xν = 0, εν = ±1.
0
Let
n
M = ∑ |aν |.
0
They showed that except for a set of these equations of the proportion

log log n
O ,
log n
the remainder of the equations have not more than

M
10 log n log ) + 2(log n) 5
|a0 an |
real roots for each equation.
W.A. Beyer
PROBLEM 180: KAMPÉ DE FÉRIET

May 16, 1939
Let ν (t, E) be a stationary random function (in the sense of E. Slutsky, A. Khin-
chine): (E a random event)
⎫
ν (t, E) = 0 ν (t, E)2 = constant⎬ for all t
ν (t, E)ν (t + h, E) = function of h alone ⎭ −∞ < t < +∞.
Does there exist a random variable A which, with uniform probability, assumes
every value α between 0 and 1,
Prob [A < α ] = α , 0 ≤ α ≤ 1,
such that
(1) E = φ (α ),
(2) ν [t1 , φ (α )] and ν [t2 , φ (α )] are two independent functions (in the sense of
H. Steinhaus) for every couple t1 , t2 (t1 = t2 )?

Find a continuous function (or perhaps an analytic one) f (x), positive and such
that one has
∞
∑ f (x + n) ≡ 1
n=−∞
√
(identically in x in the interval −∞ < x < +∞); examine whether (1/ π )e−x is
2
such a function; or else prove the impossibility; or else prove uniqueness.

√
Addendum. The function (1/ π )e−x does not have the property—this follows
2
from the sign of the second derivative for x = 0 of the expression
+∞
1
∑ √π e−(x+n) .
2
−∞
H. Steinhaus
We take a function g(x) positive, continuous, and such that
+∞
∑ g(x + n) = f (x) < +∞
n=−∞
in the interval (−∞, +∞) for example, g(x) = e−x and the function g(x)/f (x)
satisfies the conditions.
S. Mazur
December 1, 1939
PROBLEM 182: B. KNASTER

December 31, 1939; Prize: Small light beer
The disk cannot be decomposed into chords (not reducing to a single point), but
a sphere can be so decomposed (noneffectively). Give an effective decomposition
of a sphere into chords. The same for the n-dimensional sphere for “chords” of
dimension k ≤ n − 2.
Commentary
This problem is still open. Concerning the partition of a spherical surface and the
plane into arcs, see J.H. Conway and H.T. Croft: Covering a sphere with congruent
great circle arcs, Proc. Cambridge Phil. Soc. 60 (1964), 787–800.
Jan Mycielski
PROBLEM 183: BOGOLUBOW

February 8, 1940; Prize: A flask of brandy; Original manuscript in French
Given is a compact, connected, and locally connected group of transformations
of the n-dimensional Euclidean space. Prove (or give a counterexample) that one
can introduce in this space such coordinates that the transformations of the group
will be linear.
Remark
The answer is no. One of the earliest counterexamples (given by Conner and Floyd)
is presented by G.E. Bredon, Introduction to Compact Transformation Groups,
Vol. 46, Pure and Applied Mathematics, Academic Press, New York, 1972, 58–61.
PROBLEM 184: S. SAKS

February 8, 1940; Prize: One kilo of bacon
A subharmonic function φ has everywhere partial derivatives ∂ 2 φ /∂ x2 ,
∂ φ /∂ y2 . Is it true that everywhere Δ φ ≥ 0?
2
Remark: It is obvious immediately that Δ φ ≥ 0 at all points of continuity of

∂ 2 φ /∂ x2 , ∂ 2 φ /∂ y2 , therefore on an everywhere dense set.
PROBLEM 185: S. SAKS

Is it true that for every continuous surface z = f (x, y) (0 ≤ x ≤ 1, 0 ≤ y ≤ 1) the
surface area is equal to
2 2
1 1 f (x + h, y) − f (x, y) f (x, y + h) − f (x, y)
lim + + 1 dxdy.
h→0 0 0 h h
Remark: The theorem is true for curves [for surfaces it was given by
L.C. Young, but the proof (cf. S. Saks, Theory of the Integral, 1937) contains
an essential error].
Commentary
S. Saks had written, in Theory of the Integral (Warsaw 1937, p. 182): “. . . as proved
by L.C. Young . . .
2
F(x + α , y) − F(x, y)
S(F; I0 ) = lim
α ,β →0 I0 α
2 1/2
F(x, y + β ) − F(x, y)
+ +1 dxdy.” (1)
β
Here I0 is any rectangle with sides parallel to the axes, and S(F; I0 ) is the area
of the continuous surface z = F(x, y) over I0 . Young (An expression connected with
the area of a surface z = F(x, y), Duke Math. J. 11 (1944), 43–57) states that the
proof is based on a false inequality (which appears near the bottom of 183); the
existence of an error had been pointed out by V. Jarník and also by T. Rado and
P.V. Reichelderfer. Young writes, “The error was not mine, but I am partly to blame
for suggesting that the theorem could easily be proved in this sort of way, and I
failed to detect the error during proof reading.” (According to Saks’ preface, Young
“greatly exceeded his role of translator in his collaboration with the author.”) Young
then analyzes the situation in depth.
Denote the integral in (8.4) by S∗ (α , β ). Young observes that the correct part
of Saks’ proof establishes only that S(F; I0 ) ≤ limα ,β →0 S∗ (α , beta). He shows
√ √
that there is strict inequality for I0 = [0, 2] × [0, 2] and F(x, y) = g(x + y),
where g is the well-known singular continuous monotone function constant on the
complementary intervals of Cantor’s set, extended similarly to the whole line. By
taking the average F and two functions corresponding to rotations of the surface
through angles ±π /3, he obtains a surface z = F0 (x, y) such that after an arbitrary
rotation of the axes there is still strict inequality.
Young also establishes the following definitive criterion. In order that, for a
continuous surface z = F(x, y) of finite area, this area be the limit of S∗ (α , β ), it
is necessary and sufficient that there exists a decomposition of I0 into two Borel sets
E1 , E2 such that on E1 the function F is absolutely continuous in x on the sections
determined by almost all constant values of y, while in E2 it is absolutely continuous
in y on the sections determined by almost all constant values of x.
Young’s condition is presumably not necessary (as well as sufficient) for the
equality S(F; I0 ) = limh→0 S∗ (h, h) with which Problem 185 is concerned, but it
might be regarded as neither sensible nor interesting to seek a criterion for the
possession of this rather artificial property, tied as it is to a particular choice of
axes. Modern research has moved in the direction of results independent of such a
choice.
Roy O. Davies
PROBLEM 186: S. BANACH

March 21, 1940
Does there exist a sequence {φi (t)} of functions, orthogonal, normed, and
complete in the interval (0 ≤ t ≤ 1) with the property that for every continuous
function f (t), 0 ≤ t ≤ 1 (not identically zero) the development
∞ 1
∑ φi (t) 0
f (t)φi (t) dt
i=1
is at almost every point unbounded?
Remark
The answer is yes. See the commentary to problem 86

PROBLEM 187: P. ALEXANDROFF

April 19, 1940; Original manuscript in French
(1) Let P be a mutilated polyhedron (that is to say, in its simplest decomposition one
has deleted a certain number of simplices of arbitrary dimension) contained in
Rn . Rn −P is then also a mutilated polyhedron. We understand by the Betti group
of this polyhedron the usual Betti group in the sense of Vietoris. The duality
law of Alexander is then true for mutilated polyhedra. Prove that if P ⊂ Rn
is a mutilated topological polyhedron (that is to say, a topological image of a
mutilated polyhedron) the duality theorem of Alexander still holds.
(2) Prove (or refute) the theorem: For every Hausdorff space which is bicompact,
the inductive definition of dimension is equivalent to the definition given with
the aid of coverings (Uberdeckungen).
(3) Prove (or refute) the impossibility of an interior continuous transformation of a
cube with p dimensions into a cube of p dimensions for p < q.
Commentary
Concerning part 2, examples of compact spaces whose covering and inductive

dimensions are distinct were given by Lunc [5] and Lokuchevsky [3]. A description
of Lokuchevsky’s example can be found in [2, 178–180].
Concerning part 3, an open (and monotone) mapping of I m onto I n with m <
n was defined by L. Keldyš [3]. Similar examples were announced earlier by
R.D. Anderson [1].
1. R.D. Anderson, Some upper semi-continuous collections of continuous curves filling up R3 ,
Bull. Amer. Math. Soc. 59 (1953), 559.
2. R. Engelking, Dimension Theory, North-Holland, New York, 1978.
3. L. Keldyš, Transformation of a monotone irreducible mapping into a monotone-open one, and
monotone-open mappings of the cube which raise the dimension. Mat. Sb. 43 (1957), 187–226.
4. O.V. Lokuchevsky, On the dimension of bicompacta, Doklady 67 (1949), 217–219.
5. A.L. Lunc, A bicompactum whose inductive dimension is greater than its dimension defined by
means of coverings, Doklady 66 (1949), 801–803, MR 11–46.
Ryszard Engelking
PROBLEM 188: S. SOBOLEW
April 20, 1940; Prize: For a solution of the problem, a bottle of wine; Original in
both Polish and Russian
One has proved the existence of a Cauchy problem

u = φ0 (x1 , . . . , xn−1 )
xn =0

∂ u
= φ1 (x1 , . . . , xn−1 ),
∂ xn
xn =0
for the quasilinear partial differential equation of the form

n n
∂ 2u
∑ ∑ Aij ∂ xi ∂ xj = F
j=1 i=1
of the hyperbolic type (where Aij and F depend on x1 , . . . , xn , u, ∂ u/∂ x1 , . . . ,

∂ u/∂ xn ), if the function φ0 possesses square integrable derivatives up to the order
[n/2] + 3 and function φ1 partial derivatives up to the order [n/2] + 2 (also square
integrable). We assume in addition that the derivatives of functions Aij and F with
respect to ∂ u/∂ xi and to u are continuous.
For the nonlinear equation of the general form

∂u ∂ u ∂ 2u ∂u
φ x1 , . . . , xn , u, ,..., ,..., 2 =0
∂ x1 ∂ xn ∂ x1
2 ∂ xn
one can easily show the existence of a solution if only φ0 has derivatives up to the
order [n/2] + 4, and φ1 up to the order [n/2] + 3, square integrable. One should
construct an example of such an equation and such boundary conditions having
derivatives of the order less by 1, square integrable, such that the solution would
not exist, or else lower the number of derivatives necessary for the existence of a
solution to the number necessary in the case of quasilinear equations. (This latter
number cannot be lowered any more as shown by known examples.)
Problem 188.1 M. EIDELHEIT

November 27, 1940
Let z(x, y) be a function absolutely continuous on every straight line parallel to
the axes of the coordinate system In the square 0 ≤ x, y ≤ 1. Let f (t) and g(t) be
two absolutely continuous functions in 0 ≤ t ≤ 1 with values also in (0, 1). Is the
function h(t) = z(f (t), g(t)) also absolutely continuous? If not, then perhaps under
the additional assumptions that
1

1 ∂ z p 1

1 ∂ z p
dxdy < ∞, dxdy < ∞,
0 0
∂x 0 0
∂y
where p > 1.
As this edition went to press, a negative solution to Eidelheit’s problem has been
given by L. Maligranda, V. Mykhaylyuk, and A. Plichko.
PROBLEM 189: A. F. FERMANT

Original manuscript in Russian
Let w = f (z) be a regular function in the circle |z| < 1, f (0) = 0, f (0) = 1.
We shall call the “principal star” of this function the following one-leafed star-like
domain: On the leaf of the Riemann surface corresponding to the function w = f (z)
to which the point w = f (z) = 0 belongs, we take the biggest one-leafed region
belonging to the surface.
Prove the theorem: The principal star of the function w = f (z) contains a circle
of a radius not less than an absolute constant B (generalization of a theorem of
A. Bloch).
Commentary
The answer does not appear to be known. The idea of the principal star goes back to
W. Gross who used it for entire functions. E. Landau showed that the biggest disk
on the Riemann surface centered at 0 can be arbitrarily small. Otherwise, nothing
seems to be known about the star as defined in the problem.
1. W. Gross, Über die Singularitäten analytischer Funktionen, Monatshefte f. Math. u. Phys. 29
(1918), 3-47.
2. E. Landau, Der Picard-Schottkysche Satz und die Blochsche Konstante, Sitz. Der. d. Preuss. Ak.
d. Wiss. phys.-math. Kl. 1926.
Lars V. Ahlfors
PROBLEM 190: L. LUSTERNIK

February 4, 1941; Prize: For the solution, a bottle of champagne for the solver;
Original manuscript in Russian
Let there be given in the Hilbert space L2 an additive functional f (x) defined on
a part of L2 , and a self-adjoint operator A. If f is linear, then it is an element of L2
and Af = f (Ax). Let us extend the operation A over all additive functionals f by the
formula: Af = f (Ax). If there is a point λ of the continuous spectrum of A, then we
can find an infinite set of additive functionals f , not identically equal to zero, for
which (A − λ E)(f ) ≡ 0.
These f (x) can be considered as, so to say, ideal associated elements for the
points λ of the continuous spectrum since the properties of the continuous spectrum
are reflected on the structure of the sets of the ideal associated elements.
Commentary
This problem does not seem to be precisely formulated since there is no yes or no
answer requested or conjecture stated. By the term “linear” in the second sentence
is meant additive and continuous, as was common in those days. See Banach,
Theorie des operations Linearires, Monografje Matematyczne, 1932, page VI. In
the third sentence, f presumably should have domain containing the range of A. The
proposer’s method of associating a functional f with A and λ ,
f (Ax) = λ f (x) ∀x ∈ L2 ,
anticipates, in a rather vague way, the construction of generalized eigenfunctions

as invented by Gelfand and Kostyuchenko in 1955 (see Gelfand and Shilov,
Generalized Functions, Academic Press, Volume 3, 1967, Chapter 4, and Volume 4,
1964,Chapter 1, Section 4). In the Gelfand and Kostyuchenko theory, let A be an
operator in a linear topological space Φ . A linear functional f ∈ Φ such that
f (Ax) = λ f (x)
for every x ∈ Φ is called a generalized eigenvector of A corresponding to the

eigenvalue λ . It is then shown that if we have a rigged Hilbert space Φ ⊂ H ⊂ Φ
and if the operator A can be extended to a unitary or self-adjoint operator in H, then
the system of generalized eigenfunctions of A is complete.
W.A. Beyer
R. Daniel Mauldin

April , 1941
Auxiliary definitions: I call a measure every nonnegative, completely additive set
function defined on a certain completely additive class of sets K, subsets of a fixed
set χ and such that μ (χ ) = 1. The measure μ is convex (according to M. Fréchet,
“sans singularités”), if for every set A such that μ (A) > 0 there exists a set B ⊂ A,
such that μ (A) > μ (B) > 0. The measure μ is separable if there exists a countable
class D ⊂ K, such that for every η > 0 and every M ∈ K there exists L ∈ D,
such that μ [(M − L) + (L − M)] < η . The class K is a class of sets stochastically
independent with respect to μ if μ (A1 A2 · · · An ) = μ (A1 )μ (A2 ) · · · μ (An ) for every
disjoint sequence {Ak } of sets belonging to K.
Definitions of a base. The class B ⊂ K is called a base of a measure μ if
(1) B is a class of sets stochastically independent with respect to μ ; and
(2) All sets of the class K can be approximated, up to sets of measure 0, by sets of
the smallest countably additive class of sets containing B.
Remarks: Let Bn denote the set of numbers from the interval 0, 1, whose nth
binary digit = 1. The sequence {Bn } is a base for the Lebesgue measure in the
interval 0, 1. It follows easily that every convex, separable measure has a base. In
the known examples of nonseparable measures, there also exists a base.
Problem: Does every convex measure possess a base?
Solution
The answer is no. One counterexample would be provided by the direct sum of the
Lebesgue measure spaces 2ℵ0 and 2c (scaled to make the total measure 1). That this
space is a counterexample follows from the following theorem.
Theorem A non-atomic (= convex, in Szpilrajn’s terminology) measure has a

base if and only if the corresponding measure algebra is homogeneous.
Recall that a measure μ defined on a σ -algebra K of subsets of a set X is

homogeneous provided that h(A) = h(X), for every measurable subset A of positive
measure, where h(A) = least cardinal of a family F of measurable subsets of A such
that every measurable subset of A differs by at most a null set (with respect to μ
restricted to A) from a member of the Borel field of subsets of A generated by F.
Sketch of proof: If the measure algebra E of (X, μ ) is homogeneous, then
according to the results in [1], E is isomorphic to the measure algebra of 2m (with
Lebesgue product measure) for some infinite cardinal m, and this clearly provides
a “base” for μ , in the sense of Problem 191. Conversely, if μ has a base B in this
sense, then E is isomorphic to the measure algebra of 2m where m = cardinality of
B: and this is well known (and easily seen) to be homogeneous.
The space 2ℵ0 ⊕ 2c is not homogeneous, since h(2m ) = m, for every infinite
cardinal m.
1. D. Maharam, On homogeneous measure algebras, Proc. Nat. Acad. Sci. USA 28 (1942),
108–111.
Dorothy Maharam
PROBLEM 192: B. KNASTER, E. SZPILRAJN

May, 1941
Definition. A topological T has the property (S) (of Suslin) if every family of
disjoint sets, open in T, is at most countable.
Definition. A space T has property (K) (of Knaster) if every noncountable
family of sets, open in T, contains a noncountable subfamily of sets which have
elements common to each other. Remarks:
(1) One sees at once that the condition (K) implies (S) and, in the domain of metric
spaces, each is equivalent to separability.
(2) B. Knaster proved in April 1941 that, in the domain of continuous, ordered sets,
the property (K) is equivalent to separability. The problem of Suslin is therefore
equivalent to the question whether, for ordered continuous sets, the property (S)
implies the property (K).
Problem (B. Knaster and E. Szpilrajn). Does there exist a topological space (in
the sense of Hausdorff, or, in a weaker sense, e.g., spaces of Kolmogoroff) with the
property (S) and not satisfying the property (K)?
Remark:
(3) According to Remark (2), a negative answer would give a solution of the
problem of Suslin.
Problem (E. Szpilrajn). Is the property (S) an invariant of the operation of

Cartesian product of two factors?
Remarks:
(4) One can show that if this is so, then this property is also invariant of the
Cartesian product of any number of (even noncountably many) factors.
(5) E. Szpilrajn proved in May 1941 that the property (K) is an invariant of the
Cartesian product for any number of factors and B. Lance and M. Wiszik
verified that if one space possesses property (S), and another space has property
(K), then their Cartesian product also has property (S).
Commentary
B. Knaster proved in April 1941 that the existence of a Souslin line was equivalent
to the existence of a connected, linearly ordered space without property (K). The
square of a Souslin line does not have property (S). But E. Szpilrajn proved in May
1941 that a product of any family of spaces with property (K) has property (K).
B. Lance and M. Wiszik verified that the product of a space with property (S) and
a space with property (K) has property (S). Thus these two problems came to be
posed.
We now know that the answer to both questions is independent of Zermelo
Frankel set theory.
Definitions: We say that a topological space has the associated property if every
uncountable family of open sets has:
(c) (for caliber ℵ1 ) an uncountable subfamily with nonempty intersection.
(p) (for precaliber ℵ1 ) an uncountable subfamily with the finite intersection
property.
(K) (for Knaster) an uncountable subfamily each two of whose members have
nonempty intersection.
(S) (for Souslin; now called ccc for countable chain condition) at least two members
with nonempty intersection.
Clearly (c) → (p) → (K) → (S).

Property (c) is actually a topological property. There is a simple subset of 2ω1
with property (p) but not (c); for compact T2 spaces (p) and (c) are clearly equivalent.
However (p), (K), and (S) depend only on the associated Boolean algebra. If one
assumes Martin’s axiom together with the negation of the continuum hypothesis
(MA + ¬CH), then (p), (K), and (S) are all equivalent (independently proved
by K. Kunen, F. Rowbottom, R. Solovay; see I. Juhasz, Martin’s axiom solves
Ponomarev’s problem, Bull. Acad. Polon. Sci. Ser. Sci. Math. Ast. Phys. 18 (1970),
71–74. Thus it is consistent with ZFC that the answers to both questions be yes.
The existence of a Souslin line, long known to be independent of ZFC (with or
without CH: see Souslin’s conjecture, Amer. Math. Monthly 76 (1969), 1113–1119)
clearly implies that the answer to both questions is no.
More recently (p), (K), and (S) have all been shown to be different if the
continuum hypothesis holds. See K. Kunen and F. Tall, Between Martin’s axiom
and Souslin’s hypothesis, Fund. Math. 102 (1979), 173–181) for a proof that (K)
does not imply (p) under CH (proved independently by P. Erdős, R. Laver and
independently and more simply, F. Galvin, Chain conditions and products, Fund.
Math. 108 (1980), 33–48) have shown that if CH holds there are spaces with (S)
whose square does not have (K). Thus, (S) does not imply (K) and the product
question also has a negative answer if one assumes CH. For other proofs that ccc
being productive does not imply (K) when CH holds, see (M.L. Wage, Almost
disjoint sets and Martin’s axiom, J. Symbolic Logic, 44 (1979), 313–318) and (E. van
Douwen and K. Kunen, L-spaces and S-spaces in ℘(ω ), Topology Appl. 14 (1982),
143–149).
M.E. Rudin

May 31, 1941
The “expected” number of matches: 7
Probability that x ≤ 9 −→ 0.68
“The probable” number of matches: 6
The probability that x ≤ 6 is 0.5
(Two boxes with fifty matches)
(The exact solution requires lengthy computations.)
Commentary
These calculations seem to be the result of a problem which Steinhaus called the
Banach match box problem. Since nowadays most mathematicians do not smoke,
it should be explained that a certain mathematician always carries two boxes of

matches, one in his right pocket and one in his left pocket. He picks a box at random
to light his pipe. Initially, the boxes each have N matches. When he first finds a
box empty, what is the distribution of the number of matches in the other box?
The distribution is given in Feller, An Introduction to Probability Theory and its
Applications, Vol. 1, 2nd edition, John Wiley & Sons, 1957, 157.
We observe that these calculations make some sense if “median” is replaced by
“mean” and the value “.45” in the fourth line of the Los Alamos edition is replaced
by “.95”. It may be that there were transcription errors. Perhaps there is another
interpretation whereby these calculations make sense. [Editor note. To see several
papers devoted to the Banach match box problem, check MathSciNet.]
W.A. Beyer
Chapter 7
Appendices to the Problems
7.1 Problem Subject Index
absolutely continuous functions: 188.1

analytic functions: 17.1, 158, 181, 189
analytic sets: 28, 52, 168
approximation property: 153
approximation theory: 14, 120, 159
associative mappings: 171
automatic continuity: 45
automorphism: 112
Baire function: 132
Baire property: 45, 50
basis: 108
Banach match box problem: 193
Banach Spaces (spaces of type (B)): 1, 7, 24, 27, 41, 46, 47, 64, 69, 87, 106, 108,
122, 144, 153, 155, 172, 173, 174, 176, 190
billiards: 147
Boolean algebras: 37, 163
Borsuk-Ulam theorem: 123
Cauchy problem: 188
compact groups: 32
constrained minima: 104, 105
convergence of rearrangements: 106
control measure problem: 163
Convexity: 7, 10.1, 19, 23, 36, 41, 42, 54, 64, 65, 68, 72, 76, 84, 89, 91, 138,
149, 191
curves: 93, 117
Denjoy integral: 50
descriptive set theory: 22, 33, 40, 43, 50, 52, 80, 168

284 7 Appendices to the Problems
dimension raising maps: 136, 137

diophantine approximation: 152
distribution of sums: 179
dynamical systems: 102, 110, 115, 141, 147, 164
Eidelheit separation theorem: 64
ellipsoid: 84, 91, 128, 129
essential map: 114
everywhere divergent trigonometric series: 121
fixed points: 42, 54, 107, 115, 167
floating bodies: 19
Fréchet spaces (spaces of type (F): A space of type (F) is the terminology from
Banach’s monograph, Théorie des Opérations Linéaires, for a Fréchet space, a
completely metrizable topological vector space.) 15.1
Functional analysis: 10, 24, 26, 72
functional equations: 39
functions of several variables: 23, 66, 70
general topology 58, 77, 127, 133, 134 135, 136, 192
geometric topology 12, 21, 29, 35, 36, 46, 68, 107, 110, 114, 115, 123, 167,
175, 187
geometry: 44, 59, 61, 76, 78, 81, 182
graph theory: 38, 140, 164
group theory: 29, 30, 62, 63, 77, 142, 160, 183
Hadamard’s determinant problem: 118
Ham sandwich theorem: 123
harmonic functions: 151
Hilbert cube: 7, 12, 77, 110, 170, 175
Hilbert’s fifth problem: 70
homeomorphisms: 20, 97, 98, 100, 170
homeomorphism group: 29
homogeneous polynomials: 74
homothetic: 129
indecomposable continua: 134 Integral Equations: 25, 124
interpolation: 87
lacunary: 130
Lebesgue density: 146
Lie groups: 90
Lipschitz functions: 23, 87, 165
local isometry: 155
magnetic fields: 18, 19
manifolds 21, 68, 166
matrix theory: 6, 47, 177
measure theory: 2, 11, 16, 40, 139, 144, 145, 146, 163, 169, 170, 191
multilinear mappings: 73
orthogonal functions: 51, 85, 86, 119, 130, 154, 186
partial differential equations: 15, 125, 188
7.2 Authors of the problems 285
permutations: 46, 71, 95, 96, 101

physics: 18, 19, 125
Poisson distribution: 17
Polish groups: 31, 45
polyhedron: 187 polynomials: 20.1, 70, 75, 79, 128, 148
polynomials in Banach spaces: 55, 56
probability theory: 17, 82, 94, 126, 161, 178, 180, 193
quasi-homeomorphisms: 97
Real analysis: 1, 2, 3, 4, 5, 6, 8, 50, 57, 83, 109, 113, 131, 157, 162, 165, 181
real roots of random algebraic equations: 179
recursion theory: 143
set theory: 9, 13, 22, 34, 37, 40, 43, 99, 168
Schauder basis: 119, 153
Schauder fixed point theorem: 54
S∞ : 71, 95, 96, 101, 111
spaces of type (F0 ) (Defined in Problem 26): 26, 72
spaces of type (B0 ): 138 strategy: 43, 67
subharmonic functions: 184
summability: 4,5, 8, 28, 88, 92, 154
surface area: 185
surface integrals: 149, 150, 185
Suslin line: 192
theory of games: 43
theory of surfaces: 44, 23, 53, 61, 78, 156
topological dimension: 127
transformation groups: 183
unconditional convergence: 122
uniform distribution in groups: 116
universal Banach spaces: 49
universal groups: 103 unicoherent: 46, 128, 135
7.2 Authors of the problems
Alexandroff: 187
Auerbach: 39, 83, 84, 148
Auerbach, Banach, Mazur, Ulam: 10.1
Auerbach, Mazur: 147
Auerbach, Ulam: 90, 114
Banach: 1, 12, 43, 45, 46, 47, 50, 52, 53, 67, 85, 86, 87, 106, 186
Banach, Mazur: 7, 10, 48, 49
Banach, Mazur, Ulam: 108
Banach, Ulam 2, 3, 11, 40, 100
Bogolubow: 183 Borsuk: 175
286 7 Appendices to the Problems
Eidelheit: 172, 173, 174, 176, 188.1

Eilenberg: 133, 134, 135, 136, 137, 138
Fermant: 189
Fréchet: 117, 118
Infeld: 125
Kac: 126, 161, 177, 178
Kaczmarz: 130
Kampé de Fériet: 180
Knaster: 182
Knaster, Szpilrajn: 192
Kuratowski: 127
Kuratowski, Ulam: 97
Lomnicki, Ulam: 94
Lusternik: 190
Marcinkiewicz: 124
Mazur: 5, 8, 24, 28, 41, 43, 51, 55, 64, 65 66, 72, 75, 76, 80, 88, 89, 91, 92, 93, 143,
153, 154
Mazur, Orlicz: 6, 9, 15.1, 20.1, 26, 27, 56, 73, 74, 79, 122
Mazur, Schauder: 14
Mazur, Sternbach: 155
Mazur, Ulam: 62, 63, 69, 109, 144
Nikliborc: 128, 129, 149, 150
Offord: 179
Orlicz: 119, 120 121
Ruziewicz: 57, 58, 59, 60, 159
Saks: 184, 185
Schauder: 15, 23, 25, 54, 104, 105
Scheier: 4, 111, 112 113
Scheier, Ulam: 22, 95, 98, 103, 116, 171
Sierpinski: 132
Sobolew: 188
Steinhaus: 44, 61, 78, 81, 82, 123, 152, 162, 181, 193
Sternbach: 107
Stoilow: 158
Szpilrajn: 169, 170, 191
Ulam: 13, 16, 17, 17.1, 18, 19, 20, 21, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 42, 43,
68, 70, 71, 77, 96, 99, 101, 102, 110, 139, 140, 141, 142, 145, 146, 164, 165, 166,
167 168
von Neumann: 163
Ward: 156, 157
Wavre: 151
Zygmund: 131
7.3 Problems of Note 287
7.3 Problems of Note
We end the part of this book dedicated to the original Scottish Book with three
lists of particularly notable problems. They are highlighted not necessarily for their
mathematical content (which is, of course, fascinating and rich), but for their sundry
extrinsic values.
The numbers below refer to the problem numbers in Chapter 6.
• Prize problems: 6, 15.1, 24, 27, 28, 43, 77, 106, 110, 115, 151, 152, 153, 157,
162, 163, 165, 168, 182, 183, 184, 188, 190
• Partially solved problems: 1, 2, 18, 37, 55, 75, 80, 92, 97, 98, 99, 111, 115,
168, 187
• Unsolved Problems, Problems with No Commentary, and Problems of Unknown
Status: 14, 20.1, 22, 23, 25, 30, 31, 45, 56, 58, 60, 61, 70, 71, 76, 78, 81, 84, 102,
104, 105, 107, 118, 125, 128, 129, 147, 150, 151, 155, 156, 158, 164, 165, 166,
172, 174, 180, 182, 184, 188, 189, 190
Part III
A Brief History of Wrocław’s New
Scottish Book
Chapter 8
Lwów of the West, P. Biler, P. Krupski,
G. Plebanek and W. A. Woyczyński
At the end of World War 2, the Yalta/Potsdam Agreements made Lwów part of the
Soviet Union’s Ukraine Republic, and the Silesian capital, Wrocław, part of Poland.
Most of the faculty of the Lwów’s Jan Casimir University moved 320 miles west to
the newly Polish city of Wrocław. That included several major figures of the Lwów
(and Warsaw) School of Mathematics, settled at the new University and Polytechnic
of Wrocław led by the former Rector (President) of the Lwów University, Stanisław
Kulczyński, a biologist. Figure 1 shows the dramatic shift West of the political
boundaries between Germany, Poland and the Soviet Union executed in 1945, and
the related move of the academic community from Lwów University to Wrocław.
At the very end of WW 2 Wrocław was surrounded by the Soviet army, and
surrendered on May 6, 1945, a couple of days before Germany formally capitulated.
But the result of the four-month siege of Festung Breslau was devastating, and some
90 percent of the city, including the university buildings was in ruin. Today the city
is completely rebuilt and returned to its past glory. The main university building
(constructed during the period 1728-1740 after the university was established in
1702 as a continuation of a Jesuit College) is again dominating the view of the
Oder River downtown, and the new Mathematical Institute building where the
New Scottish Book is currently stored is crowning the Grunwaldzki Square axis
(see, Fig. 2). And, for 2016, Wrocław was designated by the European Union as
the Europe’s Culture Capital of the Year, the fact which is advertised across the
continent with the moniker WrocLOVE.
The New Scottish Book (NSB) was conceived in 1946 as a continuation of the
original Lwów Scottish Book (SB). The transition may seem seamless as the last
entry in SB was contributed by Hugo Steinhaus on May 31, 1941, and he also
contributed the first entry in the New Scottish Book on July 2, 1946.
But in-between those dates the four mathematicians (see, Fig. 3), who are
recognized as the founding fathers of the Wrocław School, had very different
traumatic experiences during the German occupation. Hugo Steinhaus, Edward
Marczewski, and Bronisław Knaster were of Jewish background, and had to hide

292 8 Lwów of the West, P. Biler, P. Krupski, G. Plebanek and W. A. Woyczyński
Figure 1 Poland in 1939,

and in 1945. The Lwów
community moves to
Wrocław.
to avoid concentration camps. Steinhaus, aged 58 in 1945, who spent the war years
as a forester’s assistant in the Carpathian woods, came with an established reputation
as a world class mathematician, a student of David Hilbert in Göttingen, the founder
of the Lwów school, and the “discoverer” of Stefan Banach. Marczewski (the name
he adopted during the war to avoid persecution, his original name was Szpilrajn), a
measure theorist, only 38 in 1945, and the youngest of the group, was arrested by the
Germans towards the end of the war and, in an improbable coincidence, was sent
to Festung Breslau as a forced laborer to help clear the rubble from Grunwadzki
Square at the exact location where the new Institute of Mathematics was to be
built under his leadership in the 1960s (see, Fig. 2). Knaster, a Warsaw topologist
8 Lwów of the West, P. Biler, P. Krupski, G. Plebanek and W. A. Woyczyński 293
Figure 2 Wrocław from 1945 to 2015. From left to right: Destruction of the 1945 siege;
the restored main building of the University; Institute of Mathematics of the University.
Figure 3 The Founding Fathers of the Wrocław School of Mathematics: From left to right: Hugo
Steinhaus, Władysław Slebodziński, Bronisław Knaster, and Edward Marczewski.
(see, Fig. 4) who spend the war years hiding in Lwów, earning subsistence by
serving (with Stefan Banach and other Lwów scientists) as a human “guinea pig”
in the German-run Rudolf Weigl Institute developing typhus vaccines for the war
effort.
Władysław Ślebodziński, a distinguished differential geometer, although not
Jewish, did not avoid Auschwitz and spent the war years in the concentration camp
lucky to survive, but with his health damaged.
The significance of Steinhaus’ entries to the SB and NSB was more than
symbolic. His series of paper “Sur les fonctions indépendentes,” based on the
ideas from his pioneering 1923 Fundamenta Mathematica paper, “Les probabilités
dénombrables et leur rapport à la théorie de la mesure” that was the precursor of
Kolmogorov’s Grundbegriffe, was the inspiration for the probability group. The first
six papers in the series, written jointly with his student Mark Kac (later at Cornell)
were published in Lwów in Studia Mathematica between 1936 and 1940, resumed
with the paper number 7 published in the same journal in 1948, with the last, tenth,
appearing in 1953. Immediately, more problems were contributed by Marczewski,
and Knaster, as well as many distinguished Polish mathematicians who survived
the war: A. Alexiewicz, S. Goła̧b, A. Mostowski, W. Orlicz, W. Sierpiński, R.
Sikorski, J. Szarski, T. Ważewski, Z. Zahorski. At the initiative of Marczewski,
most of the problems entered in NSB have been reprinted in the journal Colloquium
Mathematicum established in Wrocław by the Founding Fathers.
The combination of the mathematical interests of the Founding Fathers was a
strike of luck for Wrocław as they provided diverse background for the future
Figure 4 New Scottish Book notebooks in their new elegant covers.
developments. Steinhaus was a leader in functional analysis, probability theory,

and orthogonal expansions, Marczewski’s work in measure theory (and later in
universal algebras) was already significant before the war, Knaster had been the
leader of the topology seminar at Warsaw University in the 1930s (together with
Mazurkiewicz) where the attendees included Samuel Eilenberg (later at Columbia),
Szpilrajn-Marczewski, Nachman Aronszajn (later at University of Kansas), and
Kazimierz Kuratowski, and which attracted up to thirty participants on a regular
basis. And Ślebodziński was already acknowledged in the differential geometry
community as the inventor of the concept of Lie derivative. So, from the very
beginning, there was a solid and broad basis for creation of the mathematical
research center in Wrocław.
The New Scottish Book (in Polish, Nowa Ksiȩga Szkocka) was active from
1946 through 19791 after which date there were no entries. with many entries by
some of the most prominent researchers of the second half of the 20th century.
Mathematicians from all over the world started visiting Wrocław and contributing
to NSB. Poland, although governed by the communist party, was not like the
Soviet Union. The communications with the West, while controlled, remained open,
and the Catholic Church never stopped playing an important role in the society
(unlike the Orthodox Church in Russia which was completely dismantled by the
communists). So already in July of 1946 we find NSB contributions by Gustave
Choquet from Paris, and Vaclav Jarnik from Prague.
The international reputation of the Wrocław mathematics community made
it also very influential in running the local academic establishment. By 1953
Marczewski was elected Rector of Wrocław University, and his position facilitated
1 Just seven problems were entered between 1979 and 1987, the time of a historical Solidarity
movement in which many Wrocław mathematicians played major roles.

8 Lwów of the West, P. Biler, P. Krupski, G. Plebanek and W. A. Woyczyński 295
more visits from well-respected foreign mathematicians. Alfred Renyi visits from
Budapest, and Szolem Mandelbrojt from Paris, both contributing problems to NSB.
The Mandelbrojt problem led directly to Louis de Brange’s proof of the Bieberbach
conjecture. We also see the rise of Czesław Ryll-Nardzewski, a young Wrocław
mathematician who was to make the name for himself by seminal contributions in
ergodic theory, harmonic and complex analysis, probability theory and foundations
of mathematics (joined later in the latter field by Jan Mycielski), and had a
paramount influence on the life of the mathematical community in Wrocław (and,
more broadly, in Poland).
By the end of the 1950s, we also see the emergence of another young superstar,
Kazimierz Urbanik (later, also Rector of the University of Wrocław) who started
out with a series of papers offering an alternative development of the Gelfand-Itô
theory of generalized stochastic processes. In the 1960s Jean-Pierre Kahane visits
from Paris, contributes several problems to NSB, and develops a close and lasting
interaction with the Wrocław harmonic analysis group led by Ryll-Nardzewski
and Stanisław Hartman (and later, by later Andrzej Hulanicki), and we see a
steady stream of distinguished visitors (and contributors to NSB) such as Pavel
Sergeevich Aleksandrov from Moscow, R.H. Bing from Madison, Wisconsin, and
Aryeh Dvoretzky from Jerusalem.
One of the last problems (No. 961) was entered in October 1979 by Fulvio
Ricci, from Turin, the intellectual exchange with the international mathematical
community was very much alive. Just in October of 1978, Paul Erdös visited from
Budapest, and contributed a problem, and Hubert Delange from Orsay, suggested
a problem of finding extremal points of the set of entire functions satisfying an
exponential inequality.
Initially, the NSB, a collection of soft cover notebooks, was stored in the
joint library of the Institute of Mathematics of the University of Wrocław and
the Wrocław Polytechnic, with access guarded by the inimitable Ms Marietta
Wilanowska, the chief librarian, an approach quite different from the flamboyant
Scottish Café environment of the original SB. After some twenty years of its
existence the NSB was moved to the library located on the top floor of the new
Mathematical Institute building with a spectacular view on the Oder River and
the Cathedral Island with its several romanesque and gothic churches dating back
to the 12th Century. In 2014, on the initiative of Dean Biler, the booklets were
sent to the bookbinder and given an elegant hard leather-bound covers pictured
in Fig 5. But they are still stored by the chief librarian under lock-and-key with
an on-demand access. In 1959, young Wrocław mathematicians, Henryk Fast and
Stanisław Świerczkowski, prodded by Edward Marczewski, prepared a privately
typed little volume of 70 pages, where they collected NSB problems for the years
1946 -1958.
P.B., P.K., and G.P,

Institute of Mathematics,
University of Wrocław,
Wrocław, Poland
W.A.W.,
Department of Mathematics, Applied Mathematics and Statistics, and
Center for Stochastic and Chaotic Processes
in Science and Technology,
Case Western Reserve University,
Cleveland, OH, USA
Part IV
The New Scottish Book
At least 280 problems can be classified as topological ones. The vast majority of
them emerged from research activities of the famous Topology Seminar conducted
by Bronisław Knaster in Wrocław in the years 1946–1977. ‘CM P. . . ’ refers to
problems published in Problem Section of Colloquium Mathematicum.
Instytut Matematyczny, Uniwersytet Wrocławski, pl. Grunwaldzki 2/4, 50-384 Wrocław,

Poland Piotr.Biler@math.uni.wroc.pl; Pawel.Krupski@math.uni.wroc.pl;
Grzegorz.Plebanek@math.uni.wroc.pl
Chapter 9
Selected Problems
Instytut Matematyczny, Uniwersytet Wrocławski, pl. Grunwaldzki 2/4, 50-384 Wrocław, Poland
Piotr.Biler@math.uni.wroc.pl; Pawel.Krupski@math.uni.wroc.pl;
Grzegorz.Plebanek@math.uni.wroc.pl
1. July 2, 1946; Hugo Steinhaus (Wrocław) (in Polish)

Let 0.a1 a2 a3 . . . and 0.b1 b2 b3 . . . be decimal representations of transcendental
numbers. Is 0.a1 b1 a2 b2 a3 b3 . . . also a transcendental number?
4. July 6, 1946; Gustave Choquet (Grenoble) (in French)
Let E be a compact metric space covered by a finite number of open sets Ci of
diameter less than ε > 0. Let δi denote the diameter of the boundary of Ci , and
A(ε ) = inf ∑ δi2 over all such coverings. Finally, define A = limε →0 A(ε ). Show that
A = 0 for every space E of dimension 1, in the sense of Menger.
Prize: a bottle of champagne (to be drunk in Paris) for a positive answer; a bottle
of bordeaux for a negative answer.
5. July 7, 1946; Edward Marczewski (Wrocław) (in Polish)

Let μ be a probability measure on a σ -algebra A of subset of a space X. Two
subfamilies P1 and P2 of A are stochastically independent if μ (P1 ∩P2 ) = μ (P1 )·
μ (P2 ) for every Pi ∈ Pi , i = 1, 2.
Two real-valued measurable functions f1 , f2 : X → R are stochastically indepen-
dent in the sense of Kolmogorov if P1 and P2 are stochastically independent, where
Pi = {fi−1 (Z) ∈ A : Z ⊆ R}. Functions f1 and f2 are stochastically independent in
the sense of Steinhaus if the corresponding condition is satisfied by the families
Pi = {fi−1 (a, b) : a < b}.
Functions stochastically independent in the sense of Kolmogorov are stochasti-
cally independent in the sense of Steinhaus. Does the converse implication hold?
CM 1, P3
Answer: It is noted that the answer is ‘yes’ if the measure in question is the
Lebesgue measure, which was proved by S. Hartman. In general, the answer is ‘no,’
which follows from an unpublished note by S. Banach.

300 9 Selected Problems
Remark: First counterexample was published a few years later. Apparently, the
answer is positive whenever the measure μ is perfect.
7. November 29, 1946; Bronisław Knaster (Wrocław) (in Polish)

The Borsuk-Ulam Antipodal Theorem (K. Borsuk, Drei Sätze über die n-dimen-
sionale euklidische Sphäre, Fund. Math. 20 (1933), 178, Satz II, known as the
“Antipodensatz”) was generalized by Hopf (H. Hopf, Eine Verallgemeinerung
bekannter Abbildungs- und Überdeckungssätze, Portugaliae Math. 4 (1944), 131) as
follows: For each continuous map f (p) of the n-dimensional sphere Sn of diameter
d onto the n-dimensional Euclidean space En and for δ ≤ d, there exists at least one
pair of points p1 , p2 in Sn such that the distance between them is δ and f (p1 ) = f (p2 ).
I noticed that from a certain “tripod” problem (which provokes me to be called
“affaire Dreifuss” (“Dreifuß problem”), [with an allusion to “l’affaire Dreyfus,” of
course; note of the editor]) of Steinhaus one can derive, among others, the following
topological content:
Given any continuous f (p) from S2 onto E1 , does there exist at least one triple
p1 , p2 , p3 in S2 which is isometric (i.e., congruent) to a given triple of points in S2
and such that f (p1 ) = f (p2 ) = f (p3 )?
The problem is open. Is the following more general problem true: Given any
continuous f (p) from Sn onto Ek , k = 1, 2, . . . n, does there exist at least one set
P = {p1 , p2 , . . . , pn−k+2 } which is isometric to a given (n − k + 2)-point subset of Sn
and such that f (p1 ) = f (p2 ) = · · · = f (pn−k+2 )?
How many and when do such sets P exist?
Prize: A dinner with vodka in the restaurant “Monopol” (in Wrocław) for a proof
of the theorem; a coffee with a cake in the University café for disproving.
CM 1, P4
Answer: Yes. for S2 , see E.E. Floyd, Real-valued mappings of spheres, Proc. Amer.
Math. Soc. 6 (1955), 957-959, CM 1, P4, R1; the positive answer in full generality
was given by R. P. Jerrard, On Knaster’s conjecture, Trans. Amer. math. Soc. 170
(1972), 385–402, see CM 30, P4, R3.
14. December 13, 1946; Stanisław Goła̧b (Kraków) (in Polish)

Prove that in the two-dimensional Minkowski geometry the length of the indicatrice
(“Eichkurve”), measured in the induced metric, is at least 6. J. Kawaler – a mathe-
matician from Wadowice – was supposed to prove this. However, I could not contact
him from the beginning of the war.
16. December 13, 1946; Zygmunt Zahorski (Kraków) (in Polish)
Let us call a function f smooth iff
f (x − h) + f (x + h) − 2f (x)
lim = 0.
h→0 h
9 Selected Problems 301
[NB continuous functions with this property form the Zygmund class in modern
terminology.]
Question 1: A smooth function must it be measurable?
(If it were measurable this is of Baire’s first class.)
Question 2: Determine the category and the measure of the set of points where a
continuous smooth function has the derivative.
It is easy to prove that this set is of power continuum, and each function ϕ with
ϕ (x) ∈ [f (x), f (x)] has the Darboux property.
Answer: Q1; No.
CM 1, P8, P9, p. 32; R1, R2, p. 148. CM 5, P220, P221, p. 234.
41. March 9, 1947; Stanisław Hartman (Wrocław) (in Polish)

It is easy to see that lim inf | cos n|n = lim inf | sin n|n = 0. A bit harder is to prove
lim sup(cos n)n = 1 and lim sup | sin n|n = 1. Find lim inf(cos n)n , lim inf(sin n)n ,
lim inf(sin n)n .
Prize: A coffee and cakes in café “Bagatela,” Wrocław.
51. May 11, 1947; Mieczysław Biernacki (Lublin) (in Polish)

Is it true that the Cauchy product of two convergent series with monotone decreasing
terms has eventually monotone terms?
Answer: No. An example (in French) is given by Gustave Choquet, May 13, 1947.
52. May 11, 1947; Jan Mikusiński (in Polish)

The following theorem is true: If a function f satisfies the differential equation
f (n) (x) + f (x) = 0 on an interval [a, b], f (a) = f (b) = 0, then b − a ≤ N(n) for some
quantity which depends on n only. Determine the least N(n). The solution is known
for n = 2, . . . , 6.
70. June 20, 1948; Karol Borsuk (Warsaw) (in Polish)
Let A be a compact 0-dimensional subset of the Euclidean space En , n ≥ 4.
Can the fundamental group of En \ A be non-trivial?
CM 1, P45
71. June 20, 1948; Karol Borsuk (Warsaw) (in Polish)
A set A is said to be a topological divisor of a space E if there is a set B such that E
is homeomorphic to the Cartesian product A × B.
Must any divisor of a polyhedron be a polyhedron?
CM 1, P46
76. September 29, 1948; Eduard Čech (Prague) (in French)

If dim E ≤ n is defined by the existence of arbitrarily small finite open covers of
order ≤ n + 1 (i.e., each point of E belongs to at most n + 1 elements of the cover)
and if E is completely normal, is then the dimension of a subspace less than or equal
to the dimension of entire space E? The answer is affirmative if each closed subset
of E is Gδ , in particular if E is metrizable.
CM 1, P53
102. May 03, 1950; Bronisław Knaster (Wrocław) (in Polish)
A point p of a set Z is of order n if there exist arbitrarily small neighborhoods of
p whose boundaries intersect Z in exactly n points; in the case when n grows ad
infinity if diameters of the neighborhoods approach 0, p is said to be of order ω .
A curve is regular if it consists of points of order ≤ ω . A cycle is a continuous
image of the interval such that each pair of points lies on a simple closed curve. The
cardinality of the preimage of a point is called the multiplicity of the map at this
point.
Question: Does there exist, for each cycle which is a regular curve, a continuous
map from a circle onto the cycle of multiplicity at each point equal to the order of
the point?
Prize: A coffee with cakes for “yes”, a cod (a meat-ball for Ms. Nosarzewska [who
solved many problems in topology but didn’t like fishes; note of the editor]) for
“no.”
107. May 26, 1950; M. Katětov (Prague) (in Czech)

Let P be a normal topological space. Is it true that for every decreasing sequence
(Fn )n of closed subsets of P, if ∞
n=1 Fn = 0,
/ then there is a sequence (Gn )n of open
subsets of P such that Fn ⊆ Gn for every n and ∞ n=1 Gn = 0?/
CM, P96
Answer: No. In the modern terminology this amounts to asking if every normal
space is countably paracompact. The class of countably paracompact spaces was
introduced by Dowker and Katetov in 1951; see section 5.2 in Engelking’s book.
The answer is ‘no’ and every counterexample is now called a Dowker space. The
first Dowker space (of large size) was constructed by M.E. Rudin in 1971. The
question whether there are small Dowker spaces became quite popular in set-
theoretic topology (see M.E. Rudin’s survey in Handbook of Set Theoretic topology
(North Holland, 1984)); cf. Borel measures by R.J. Gardner and W.F. Pfeffer,
ibidem.
108. May 31, 1950; M. Katětov (Prague) (in Czech)

Is every perfectly normal space fully normal?
CM, P97
Answer: No. A space is fully normal if every open cover has a open star refinement.
By Theorem 5.1.12 in Engelking’s book a space is fully normal if (and only if) it
is paracompact. See historical remarks in Engelking, p. 389 in the 1977 edition; by
5.5.3 the answer is negative.
110. 1950; M. Katětov (Prague) (in Czech)

Which Banach spaces are paracompact in the weak topology?
CM, P98
Remark: H.H. Corson showed in 1961 that for the weak topology of a Banach
space paracompactness is equivalent to the Lindelöf property (Trans. Amer. Math.
Soc. 101 (1961), 1-15). There is a well-studied natural class of Banach space which
are weakly Lindelöf — weakly compactly generated spaces (WCG). R. Pol showed
that a Banach space may be weakly Lindelöf but not WCG.
120. October 28, 1950; Edward Marczewski (Wrocław) (in Polish)

Is every indirect product of two countably additive set functions μ and ν defined on
algebras A ⊆ P(X) and B ⊆ P(Y) is countably additive on the product algebra in
X × Y? What if A and B are σ -algebras?
Here an additive set function λ defined on A ⊗ B is called an indirect product
of μ and ν if λ (A × Y) = μ (A) for every A ∈ A and λ (X × B) = ν (B) for every
B ∈ B.
Answer: No, see E. Marczewski; Cz. Ryll-Nardzewski, Remarks on the compact-
ness and non direct products of measures, Fund. Math. 40 (1953), 165–170. Indirect
product of measures became essential in investigating compact measures in the
sense of Marczewski and perfect measures, see, e.g., J. Pachl, Disintegration and
compact measures, Math. Scan. 43 (1978), 157–168 and W. Adamski, Factorization
of measures and perfection, Proc. Amer. Math. Soc. 97 (1986), 30–32.
121. November 13, 1950; Roman Sikorski (Warsaw) (in Polish)

Is every Hausdorff space with a countable open basis a continuous interior image
of a separable metric space? (A mapping is interior if it sends open sets onto open
sets).
CM 2, P78
Answer: Yes. A.S. Schwarz, On a problem of Sikorski, Uspekhi Mat. Nauk,
12:4(76) (1957), 215.
129. December 1, 1950; Stanisław Hartman (Wrocław) (in Polish)

Is it true that the limit of a uniformly convergent sequence of functions on the
interval [0, 1] satisfying the Darboux property has also this property?
Answer: Henryk Fast gave a negative answer on January 4, 1951 (not recorded in
NSB).
131. January 4, 1957; Edward Marczewski (Wrocław) (in Polish)

Suppose that μ1 , μ2 are measures such that some (indirect) product of them is
compact. Do μi have to be compact themselves?
Answer: A measure μ is compact in the sense of Marczewski if there is a family
of μ measurable sets C such that μ (A) = sup{μ (C) : C ⊆ A, C ∈ C } and whenever
(Cn )n is a sequence in C with the finite intersection property then n Cn = 0.
/ The
answer is ‘yes’ by a highly nontrivial result due to Pachl that a restriction of a
compact measure to any sub-σ -algebra is compact.
Remark: The Lebesgue measure on [0, 1] is trivially compact in the sense of
Marczewski. However the problem, if any probability measure on some σ -algebra
A ⊆ Bor[0, 1] is compact has not been fully answered. The positive solution
under the continuum hypothesis was given by D.H. Fremlin (Weakly α -favourable
measure spaces, Fund. Math. 165 (2000), 67–94).
132. February 24, 1951; Stanisław Hartman (Wrocław) (in Polish)

It is known that for almost every real x the sequence {n!x}mod1 has the equipartition
property. Find such an x explicitly.
√
Answer: Hugo Steinhaus, February 26, 1951. E.g., the number x = ∑∞ k=1 k! [ξ k k]
1
has this property for every real ξ .
150. May 12, 1951; Kazimierz Kuratowski (Warsaw) (in Polish)

Can a generalized (i.e., non-compact) Janiszewski space be of dimension > 2?
Czesław Ryll-Nardzewski has proved that the Cantor set C, so every perfect
0-dimensional separable space, is homogeneous in the strongest (known) sense: if
two compact subsets F1 and F2 of C are homeomorphic, then each homeomorphism
between their boundaries can be extended to a homeomorphism of C onto itself
that maps F1 onto F2 (and conversely—the condition is necessary). This has been
presented at today’s meeting of the Polish Mathematical Society, Wrocław Branch.
Question: Are perfect 0-dimensional spaces the only ones with this type of
homogeneity?
Prize: Two isometric (leather) gloves.
See B. Knaster and M. Reichbach, Notion d’homogénéité et prolongements des
homéomorphies, Fund. Math. 40, (1953), 180193.
189. May 14, 1952; Karol Borsuk (Warsaw) (in Polish)

Is this true that a curve whose Cartesian square is homeomorphic to a subset of the
Euclidean 3-space is either an interval or a circle?
194. July 2, 1952; Czesław Ryll-Nardzewski (Wrocław) (in Polish)
If a complex function f (x) = ∑∞ n=−∞ an e
int satisfies ∞
∑n=−∞ |an | < ∞, is it true that
|f (x)| has absolutely convergent Fourier series?
Answer: No; Jean-Pierre Kahane (Montpellier, Paris), Sur les fonctions sommes
de séries trigonométriques absolument convergentes. C. R. Acad. Sci., Paris 240
(1955), 36–37.
197. April 12, 1951; Adam Rybarski, Abraham Götz (Wrocław) (in Polish)
Consider a convex surface and the shortest paths between two given distinct points
on that surface. The shortest path could be nonunique, as is in the case of antipodal
points on a sphere. Is it true that the sphere is the only surface such that if the shortest
paths are nonunique, there are infinitely many such paths?
199. September 25, 1952; Alfrèd Rényi (Budapest) (in English)
What are necessary and sufficient conditions regarding the system {fn (x)}, n =
1, 2, . . . , of independent measurable functions, defined in the interval [a, b], which
ensure the completeness of the system {f1m1 (x)f2m2 (x) . . . fnmn (x)}, where m1 , m2 , . . .
run independently over all nonnegative integers and n = 1, 2, 3, . . . ?
200. September 25, 1952; Alfrèd Rényi (Budapest) (in English)
Let us suppose that we know the projections of a probability measure (or mass
distribution) of the plane on an enumerable infinite set of straight lines. Is the
distribution uniquely determined or not?
Remark. I have proved that the distribution is uniquely determined in case the
distribution is such that its characteristic function is an analytic function of both
variables; this is true, e.g., in the case when the distribution is bounded.
Answer: No; A. Heppes, On the determination of probability distributions of more
dimensions by their projections. Acta Math. Acad. Sci. Hung. 7 (1956), 403–410.
215. December 16, 1952; J. Novák (Prague) (in Czech)

Is it true that in the power set P(X) of the set X of cardinality ℵ1 every sequence
has a converging subsequence? (A sequence (An )n of subsets of X converges to A if
A = n k≥n Ak = n k≥n Ak .)
CM, P135
Answer: Equivalently, this is the question if the Cantor cube 2ω1 is sequentially
compact in its usual product topology. The question was related to the fact that
the answer is clearly negative under the continuum hypothesis (CH). However,
the statement is independent of the usual axioms of set theory — it holds true
under Martin’s axiom and the negation of CH, see D.H. Fremlin, Consequences
of Martin’s axiom (CUP, 1984).
220. March 25, 1953; Bronisław Knaster (Wrocław) (in Polish)

How many topological types are there of 0-dimensional dense-in-itself Gδ ’s?
In particular, how many topological types are there among them having a homeo-
morphic image in the Cantor set C that differs from C on a non-compact subset of
left ends of contiguous intervals?
CM 3, P123
240. October 23, 1953; Bronisław Knaster (Wrocław) (in Polish)
Is it true that for each continuous mapping of the projective plane onto a (necessar-
ily) bounded continuum in the Euclidean plane there exist three points of the first
plane which are mapped onto a single point of the second one?
252. May 31, 1954; Stanisław Mrówka (in Polish)
Does the Axiom of Choice follow from the theorem that each T1 compact topology
can be enriched to a T2 compact one?
Remark: The theorem above together with the fact that the Tikhonov product of
two-point spaces is compact imply the Axiom of Choice.
272. September 7, 1955; Jan Mycielski (Wrocław) (in Polish)
Does there exist a set M of 2ℵ0 continuous functions on the interval [0, 1] such that
for every f , g ∈ M, f = g, the limit
df f (x + h) − f (x)
(x) = lim
dg h→0 g(x + h) − g(x)
does not exist for any x ∈ [0, 1]?

Remark. Using Baire’s category method one can show the existence of such a set
of power ℵ1 , so that the problem is of interest without assuming the Continuum
Hypothesis.
Answer: Yes; J. de Groot, A system of continuous, mutually non-differentiable
functions. Math. Z. 64 (1956), 192–194.
322. September 27, 1956; Paul Erdös (Budapest) (in English)

Prove that for every integer n > 0
4 1 1 1
= + +
n x y z
is solvable in positive integers x, y, z.

328. December 23, 1956; Hugo Steinhaus (Wrocław) (in Polish)

Let p ≥ 3 denote a prime number and consider regular p-polygon. Prove (or
disprove) that any three diagonals of this polygon do not intersect at a point.
Prize: a picture of the regular 23-gon.
Answer: Yes; proof in: H. T. Croft, M. Fowler, On a problem of Steinhaus about
polygons. Proc. Cambridge Phil. Soc. 57 (1961), 686–688.
334. January 25, 1957; Hugo Steinhaus (Wrocław) (in Polish)

Suppose that {nk } is an increasing sequence of integers. Conjecture: either the circle
|z| = 1 is a natural boundary (singular, “coupure”) for the function f (z) = ∑∞ nk
k=1 z ,
or f is a rational function.
338. February 10, 1957; Edward Marczewski (Wrocław) (in Polish)
Two metric spaces are called quasi-homeomorphic if they admit continuous map-
pings from the first one onto the second one and from the second one onto the first
one, both with arbitrarily small fibers.
What is an analogue of the Menger-Nöbeling theorem for quasi-homeomorphisms?
In other words, for each natural n, find the least number m = φ (n) such that every
separable (or compact) space of dimension ≤ n is quasi-homeomorphic to a subset
of the Euclidean m-space.
Clearly, n ≤ φ (n) ≤ 2n + 1. [It is known that for compact spaces (1) the dimension
is an invariant of quasi-homeomorphisms (K. Kuratowski and S. Ulam, Fund. Math.
20 (1933), 252) while (2) topological embeddability, e.g., in the plane—is not (a
Borsuk’s note submitted to CM)].
Answer: T. Ganea proved in Bull. Math. Polon. Sci. 7 (1959), 27–32, that φ (n) =
2n + 1.
344. May 14, 1957; Szolem Mandelbrojt (Paris) (in French)

Let f (z) = ∑∞ ∞
n=1 an z and ϕ (z) = ∑n=1 bn z be univalent functions for |z| < 1. Prove
n n
that ∑ n z is also a univalent function for |z| < 1. If it is true, then this will settle
an bn n
the Bieberbach conjecture (i.e., if a1 = 1, then |an | ≤ n).

Answer: No. Counterexamples are in: B. Epstein, I.J. Schoenberg, On a conjecture
concerning schlicht functions. Bull. Amer. Math. Soc. 65 (1959), 273–275.
C. Loewner, E. Netanyahu, On some compositions of Hadamard type in classes of
analytic functions. Bull. Amer. Math. Soc. 65 (1959), 284–286.
A much more general result was proved by L. de Branges in A proof of the
Bieberbach conjecture, Acta Math. 154 (1985), 137–152.
349. May 14, 1957; Marshall H. Stone (Chicago) (in English)

Let ∑nk=0 pk f (n−k) = 0 be a differential equation with sufficiently smooth coefficients
pk on the interval a < x < b; finite or infinite, with p0 > 0 there. Let nab be the
number of solutions in L2 (a, b), na — the number of solutions in L2 (a, c) but not in
L2 (a, b) (where a < c < b), and nb — the number of solutions in L2 (c, b) but not in
L2 (a, b). Find the relation, if any, which must hold among the integers nab , na , nb ;
and discover properties of the coefficients p0 , . . . , pn which condition these integers.
In the case n = 2, the works of A. Wintner and Ph. Hartman should be considered.
350. May 15, 1957; Bronisław Knaster (Wrocław) (in Polish)
There exists a continuous complex-valued function on the circle that admits each of
its values at two points (e.g., antipodal ones) but there is no such, i.e. of constant
multiplicity 2, real-valued function either on a circle or on an interval.
More generally, prove that none of the continuous functions on an interval with
values in an arbitrary topological space has multiplicity 2.
Characterize topologically the class of continua such that no continuous function
defined on them has constant multiplicity n ≥ 2 (n finite).
CM 5, P223
Comments: (Knaster, April 4, 1958): O. Hanner proved it for an interval but the
paper in Fund. Math. 45 (1958) cannot appear. It turned out that this and many other
theorems on 2-to-1 functions were done by Americans: Harold in Duke Math. J. 5
(1939) and Amer. J. Math. 62 (1940), Roberts in Duke Math. J. 6 (1940), Martin and
Roberts in Trans. Amer. Math. Soc. 49 (1941), Gilbert in Duke Math. J. 9 (1942)
and Civin in Duke Math. J. 10 (1943). It has been possible to discover these results
because the war-time Zentralblatts, etc. have arrived only now!
376. March 14, 1958; Boris Gnedenko (Kiev) (in Russian)

Let ξ1 , ξ2 , . . . , ξn , . . . be mutually independent random variables such that for
some positive constants b1 , b2 , . . . , bn , . . . , the law of the random variable bn ξn is
independent of n. Find the class of limit laws of the sums
ξ1 + · · · + ξn
sn = − An ,
Bn
with some constants An and Bn → ∞ as n → ∞.

383. April 22, 1958; Tudor Ganea (Bucarest) (in French)
Does there exist a finite-dimensional compact absolute retract X which is homoge-
neous, i.e., for a, b ∈ X there exists a homeomorphism h : X → X such that h(a) = b?
CM 6, P275
393. May 28, 1958; Johannes de Groot (Amsterdam)
Does there exist a (plane) continuum which does not admit any continuous mapping
to itself, except trivial ones?
Comments: (Knaster, January 6, 1959): I posed the same question at the Topologi-
cal Seminar in Warsaw University on October 22, 1930. It is recorded in the seminar
book. I have found it out due to Professor K. Borsuk’s letter in which he reminded
me of the fact.
Answer: Similar questions were also asked by J. de Groot [Fund. Math. 42

(1955), 203–206] (about the existence of a connected set with the property) and
by R. D. Anderson in 1955 (about continua with this kind of rigidity). H. Cook
in [Fund. Math. 60 (1967), 241–249] presented two famous examples. The first
one is an indecomposable, 1-dimensional continuum admitting only the identity
and constant self-maps. His second example is a hereditarily indecomposable 1-
dimensional continuum such that any continuous map between its nondegenerate
subcontinua must be the identity map. Since both examples are non-planar, the
Knaster - de Groot question was open until 1985 when T. Maćkowiak [Houston
J. Math. 11 (1985), no. 4, 535–558] described a hereditarily decomposable arc-
like (hence planar) continuum for which the identity is the only mapping onto a
nondegenerate subcontinuum. A little bit later, in [Dissertationes Math. (Rozprawy
Mat.) 257 (1986)], he constructed an arc-like hereditarily decomposable continuum
with the same rigidity as the second Cook’s example, i.e. with no nonconstant,
nonidentity maps between subcontinua.
Cook’s and Maćkowiak’s examples have found many interesting applications in
continuum theory, fixed-point theory, topological groups, functional analysis (see
citations in MR).
394. May 28, 1958; Johannes de Groot (problem raised by H. de Vries)
(Amsterdam)
Does there exist a rigid topological group T? (rigid means: the automorphism group
of T equals identity).
Comments: The problem reduces to groups whose all elements are of order 2.
CM 7, P286
Answer: The problem should be meant as to whether there is a topological group
without automorphisms different from translations. A solution was given by J. van
Mill in 1983 [Trans. Amer. Math. Soc. 280 (1983), 491-498] who constructed a
separable metric, connected and locally connected topological group H, the only
autohomeomorphisms of which are group translations. Moreover, for each x ∈ H,
each autohomeomorphism of H \ {x} is the identity map.
Under the Continuum Hypothesis, he obtained a connected and locally connected
topological group H with a stronger rigidity property: each continuous self-map of
H is either constant or a translation.
399. May 30, 1958; Paul Turán (Budapest) (in English)

Let F(n, A, c) be the set of those α ’s in 0 < α < 1 for which the inequality α − μq <
A
q2
is soluble with an n ≤ q ≤ cn. Denoting by |F(n, A, c)| the measure of this set, it
is to be proved that limn→∞ |F(n, A, c)| exists (and as it is easy to see, positive and
less than 1).
460. March 12, 1959; Bronisław Knaster (Wrocław) (written in Polish by

A. Lelek)
It follows from the Hahn-Mazurkiewicz theorem that for each continuous function
that maps an interval onto a square the set of points which belong to at least 3-point
preimages is dense.
Does every continuous function mapping an n-dimensional cube onto an (n + k)-
dimensional cube have a dense set of at least (k + 2)-point preimages?
462. September 9, 1959; Mark Kac (Ithaca, NY) (in English)
[Problem from Bell Telephone Laboratories, Murray Hill, NJ, US] If n points are
placed on the surface of a three-dimensional sphere and if they repel each other
according to the inverse square law, can one characterize the equilibrium position(s)
of the points?
466. October 10, 1959; Zbigniew Semadeni (Warsaw) (in Polish)
A Banach space X has property Ps (is said to be s-injective) if for every Banach
superspace Y ⊇ X there is a projection from Y onto X of norm ≤ s. Is it true that a
space with property Ps is isomorphic to a P1 -space?
CM, P308
Answer: This is still open (as far as I know, some form of the question appears on a
recent list of problems composed by H.G. Dales). Though 1-injective Banach spaces
were characterized in early fifties, there has been a little progress in understanding
s-injective spaces for s > 1.
467. October 10, 1959; Aleksander Pełczyński (Warsaw) (in Polish)

Is there a projection from ∞ [0, 1] onto its subspace of elements with countable
support.
CM, P309?
Answer: No, see A. Pełczyński, V. Sudakov, Remark on non-complemented
subspaces of the space m(S), Colloq. Math. 9 (1962) 85–88.
Remark: The question was related to the classical Phillips theorem, saying that c0
is not complemented in ∞ .

I call a dendroid any arcwise connected continuum such that between each pair of
its points it contains only one irreducible continuum.
Characterize topologically planable dendroids (i.e., having a homeomorphic image
in the plane).
CM 8, P323
485. March 7, 1960; Kazimierz Urbanik (Wrocław) (in Polish)
Let E be a perfect subset of the positive half-line containing the origin. Show that
if for each ε > 0 the Hausdorff dimension of E ∩ [0, ε ] is positive, then the additive
semigroup generated by E contains the entire half-line.
CM 8, P322, p. 139.
I call a dendroid any arcwise connected and hereditarily unicoherent continuum
(this definition is equivalent to that given in Problem 471). If, in the definition, one
replaces the arcwise connectedness with the λ -connectedness (i.e., every two points
can be joined by an irreducible subcontinuum of type λ in the sense of Kuratowski,
Topologie II, p. 137, reference), then I call such continuum a λ -dendroid.
Question: Does the Borsuk’s fixed-point theorem for dendroids (Bull. Polon.
Math.Sci. 2 (1954), 17–20) extend to λ -dendroids?
Answer: This was a famous problem in fixed-point theory of continua.
J. J. Charatonik commented (NSB, October 7, 1969) he has a positive answer
but his argument was wrong. The problem was solved by Roman Mańka, a Ph.D.
student of Knaster, in 1974 (Fund. Math. 91 (1976), no. 2, 105–121; see also an
extensive review MR0413062 (54 #1183)). He proved that λ -dendroids have the
fixed-point property even for upper semi-continuous continuum-valued mappings.
533. April 6, 1961; Jean-Pierre Kahane (Montpellier, Paris) (in French)

Determine a condition on the sequence of integers {λn } such that continuous
functions with the Fourier series ∑∞
n=−∞ an exp(iλn x) are either everywhere smooth
(C∞ ) or nowhere (on any interval) smooth.
545. April 14, 1961; Aleksander Pełczyński (Warsaw) (in Polish)
Is the space S of all measurable functions on [0, 1] (with the asymptotic convergence)
homeomorphic to the Hilbert space 2 ?
CM 9, P365
Let R be a metric compact absolute retract.
Is it true that for each two metric compact infinite spaces Q1 and Q2 the spaces
CR (Q1 ) and CR (Q2 ) are homeomorphic?
(Symbol CR (Q) denotes the space of continuous functions from Q to R with the
uniform convergence).
Remark: If R is convex, the answer is positive.
CM 9, P366

Let R be a set homeomorphic to letter T.
Is the space CR (Q) homeomorphic to the Hilbert space for an arbitrary metric
compact infinite space Q?
CM 9, P367
554. April 24, 1961; Bronisław Knaster (Wrocław) (in Polish)

Let F be a family of dendroids.
Does there exist a continuum C such that all D ∈ F are continuous images of C?
For curves which are not dendroids the answer is negative (Z. Waraszkiewicz, Fund.
Math. 22 (1934), 180–205).
555. June 2, 1961; Karl Menger (Chicago) (in English)
Let k be a positive integer. What is the dimension of the set of all points of the
Hilbert space 2 that have exactly k irrational coordinates?
559. June 26, 1961; Alexander Doniphan Wallace (New Orleans)
Does the closed 2-cell admit a continuous associative multiplication such that
{x : x2 = x} coincides with the boundary?
CM 9, P381
562. September 16, 1961; Clifford Hugh Dowker (London)
It can be shown by direct computation that the set R of points with all coordinates
rational in a non-separable real Hilbert space has the same dimension in terms of
coverings as in terms of neighbourhoods of points; indeed, dim R = R = 1.
Problem: To find a sufficient condition on a space X for equality of dimensions,
dim X = X; moreover, a condition satisfied by the above space R.
CM 9, P383
563. November 07, 1961; Jerzy Mioduszewski (Wrocław) (in Polish)
Let f , g : Y → Y be continuous functions onto a triod (letter Y) which commute, i.e.,
fg = gf . Does there exist a point x such that f (x) = g(x)?
CM 9, P383
565. November 12, 1961; E. Sparre Andersen (Århus) (in English)
Let X1 , X2 , . . . be independent random variables, all with distribution function F(x),
and let an = Prob (X1 + · · · + Xn > 0). Is it possible to choose F(x) in such a way that
the sequence a1 , a2 , . . . is divergent and Cesàro summable of order 1? It is known
that there exists F(x) such that a1 , a2 , . . . is not Cesàro summable.
569. February 2, 1962; Yuriı̆ Mikhaı̆lovich Smirnov (Moscow) (in Russian)
Is every metric space R countable-dimensional (i.e., R = ∞
i=1 Ni , dim Ni = 0) if it
has transfinite dimension R?
CM 10, P412
570. February 2, 1962; Yuriı̆ Mikhaı̆lovich Smirnov (Moscow) (in Russian)
Is any weakly infinite-dimensional space (compact) with a countable base
countable-dimensional?
A space R is weakly infinite-dimensional if for every sequence of pairs of closed
subsets An , Bn , An ∩ Bn = 0,
/ there are closed sets Cn separating R between An and
Bn with ∞ n=1 nC = 0.
/
CM 10, P413
582. May 26, 1962; Pavel Sergeevich Aleksandrov (Moscow) (in Russian)
Let n be a natural number and τ an uncountable cardinal number.
Does there exist a compactum Anτ of dimension dim = n and weight τ which is
universal for all at most n-dimensional compacta of weight ≤ τ (i.e., containing
topologically every such compactum)?
CM 10, P422
What are compacta which are images of an n-dimensional cube under 0-dimensional
open continuous mappings?
In particular, can a p-dimensional cube, p ≥ 3, be mapped by such mapping onto a
q-dimensional cube for q > p?
CM 10, P423
Is it true for each compact space X that X = X?
I proved the inequality dim X ≤ X for compact spaces (1940), Yu. M. Smirnov
proved it for strongly paracompact spaces.
Does there exist a paracompact space for which X < dim X?
592. May 26, 1962; Aleksander Pełczyński (Warsaw) (in Polish)
Can every infinite extremally disconnected compact Hausdorff space Q be continu-
ously mapped onto the generalized Cantor set Dτ , where τ is the weight of Q?
CM 10, P425
601. August 30, 1962; R. H. Bing (Madison, WI)
Suppose P is a pseudoarc, U is an open subset of P and h is a homeomorphism of P
onto itself that leaves each point of U fixed. Must h be the identity homeomorphism?
CM 10, P431
603. August 30, 1962; R. H. Bing (Madison, WI)
Is a 2-sphere S tame if it is homogeneous under a space homeomorphism? That is—
for each pair of points p, q of S there is a homeomorphism h : E3 → E3 such that
h(S) = S, h(p) = q.
(This problem was proposed by a student named Becker in one of my seminars).
604. August 30, 1962; R. H. Bing (Madison, Wisc.)
Is a 2-sphere S in E3 tame if it can be approximated from either side by a singular
2-sphere? That is—for each component U of E3 \ S and each ε > 0 there is a map
of S into U such that f moves no point more than ε .
CM 10, P429
608. 1962; Arieh Dvoretzky (Jerusalem) (in English)
Let C be a convex symmetric body in R4 . If all orthogonal three-dimensional
projections of C can be mapped on one another in C, is C an ellipsoid?
The answer is yes if C ⊂ R3 and two-dimensional projections are considered.

H. Auerbach, S. Mazur, S. Ulam, Sur une propriété caractéristique de l’ellipsoïde.
Monatshefte Math. 42 (1935), 45–48.
647. May 24, 1963; Richard D. Anderson (Baton Rouge, LA)
Does there exist any homeomorphism α of the Cantor set C onto itself such that for
any homeomorphism β of C onto itself there exists a mapping η of C onto C such
that ηα = β η ?
CM 10, P460
Answer: No. E. Nunnally, There is no universal-projecting homeomorphism of the
Cantor set. Colloq. Math. 17 (1967), 51–52.
CM 17, P460, R2
648. May 24, 1963; Richard D. Anderson (Baton Rouge, LA)

Let X be a compact metric continuum which is locally the product of the Cantor
set and the open interval. Suppose one arc component (or, alternatively, all arc
components) is dense in X.
Does X admit a continuous flow (one-parameter group of homeomorphisms) fully
transitive on each arc component?
661. June 21, 1963; Jun-iti Nagata (Osaka)
Is every n-dimensional metric space topologically imbedded in a topological
(Cartesian) product of n 1-dimensional metric spaces? (Covering dimension).
Every n-dimensional metric space can be imbedded in a product of (n + 1)
1-dimensional metric spaces. The prediction for the problem of Prof. Borsuk is
negative.
CM 12, P463
674. November 21, 1963; Jan Jaworowski (Warsaw) (in Polish)
A space X is called strongly contractible to a point x ∈ X if there is a homotopy con-
tracting X to x which keeps x fixed. A compact metric space, strongly contractible
to each of its points, of finite dimension is an absolute retract, since it is locally
contractible.
Do there exist compact, strongly contractible to each of its points spaces which are
not absolute retracts?
CM 12, P476
677. January 11, 1964; Sibe Mardešić (Zagreb)
Is there a simple closed curve C in the plane such that every straight line intersects
C exactly in 0,1 or ℵ0 points?
Let X be a metric disc-like continuum, i.e., such that it can be ε -mapped onto a
2-disc I × I for each ε > 0.
Does X possess the fixed point property?

Is dim X = X = X for all locally connected compact spaces?
684. January 11, 1964; Sibe Mardešić and P. Papić (Zagreb)
Let X be a locally connected continuum which is at the same time the image of a
totally ordered compact space K under some continuous mapping.
Is X also obtainable as the continuous image of some totally ordered continuum C?
728. May 1, 1965; Nachman Aronszajn (Lawrence, KS) (in Polish)
Let f (t) = ∑Nk=1 ak rk (t) be a linear combination of Rademacher functions rk , E ⊂
(0, 1) a set of measure |E| ∈ (0, 1]. Consider the inequality
N
c ∑ |ak |2 ≤ |f (t)|2 dt
k=1 E
with a constant c which depends on |E| only. Such an inequality is false for |E| ≤ 12 ,
#
with f = r1 − r2 . This is true for |E| ∈ 23 , 1 , and I do not know what happens if
#
|E| ∈ 12 , 23 .
774. February 12, 1967; Bronisław Knaster (Wrocław) (in Polish)
Do there exist n-dimensional continua, 1 < n < ∞, which contain no nontrivial (i.e.,
proper, non-singleton) continuous self-images?
777. February 21, 1967; Roman Duda (Wrocław) (in Polish)
Is it true that if a polyhedron can be decomposed as the Cartesian product of two
spaces which are different from it, then it can also be decomposed as the product of
two different from it polyhedra?
CM 19, P634
822. October 28, 1968; Władysław Narkiewicz (Wrocław) (in Polish)
Show that if f (n) ≥ 0 is an integer valued multiplicative function such that for all j
and N ≥ 2
1 1
lim |{n ≤ x : f (n) ≡ j (mod N)}| = ,
x→∞ x N
then f (n) = n.
H. Delange, Sur les fonctions multiplicatives à valeurs entières. C. R. Acad. Sci.,
Paris, Sér. A 284 (1977), 1325–1327.
823. March 3, 1969; Wojbor A. Woyczyński (Wrocław, Cleveland) (in Polish)
Let Ω , F , m be a measure space, B – a Banach space, and B – the algebra of
Borel sets on [0, 1], and
M : Ω × B % (ω , A) → Mω (A) ∈ B
such that for each fixed A ∈ B, M is F -measurable, and for each fixed ω ∈ Ω ,
M is a B-valued measure on B. It is known that for each fixed ω ∈ Ω there
exists a finite positive measure mω on B such that mω (A) ≤ Mω (A), A ∈ B,
and limmω (A)→0 Mω (A) = 0, where M(A) = sup |∑ni=1 ai M(Ai )| with the upper
bound taken over all |ai | ≤ 1 and finite partitions of A into Ai , cf. N. Dunford, J.
Schwartz, Linear Operators I, (IV.10.5). Can we choose mω in a measurable way,
i.e. so that for each A ∈ B the map Ω % ω → mω (A) ∈ R+ is F -measurable?
Answer. Zbigniew Lipecki (Wrocław), January 14, 1974.
Yes, B∗ has a denumerable set of functionals separating points in B (with a sketch
of proof).
827. March 11, 1969; Roman Duda (Wrocław) (in Polish)
Does every metric separable connected space have dimension ≤ 1 if it is hereditarily
locally connected (i.e., any connected subset is locally connected)?
If yes, can such space be compactified with a hereditarily locally connected
continuum?
CM 21, P682
832. June 14, 1969; Zbigniew Zieleźny (Wrocław, Buffalo) (in Polish)
Let Kx,y be a (Schwartz) distribution on Rn × Rn which is C∞ with respect to x,
and for each x0 Kx0 ,y is in Ey (a compactly supported distribution). Define the
convolution (symbolically) Tu = Kx,x−y uy dy, where u ∈ D (Rn ). Question: When
Tu ∈ E implies u ∈ E , for every u ∈ D , i.e. when the convolution operator T is
hypoelliptic?
850. November 3, 1970; Jean-François Méla (Paris) (in French)
Does there exist a set E = (nk ) of positive integers, nk < nk+1 , such that both the
properties below a), b) are satisfied?
a) There is a constant C > 0 such that for each p ≥ 1 there exists a bounded measure
3 (nk ) = 1 for 1 ≤ k ≤ p, μ
μ on the torus whose Fourier transform satisfies μ 3 (nk ) =
0 for k > p, with μ ≤ C;
b) There exists a function bounded on E which cannot be continued onto Z as the
Fourier transform of a measure on the torus.
854. November 6, 1970; Dame Mary Lucy Cartwright (Cambridge) (in English)
By a theorem of Nemytskii and Stepanov the necessary and sufficient condition for
a set to be a minimal set of a uniformly almost periodic flow is that it is a connected
separable Abelian group. In 1940 Kodaira and Abe showed that if such a group
is of (n − 1)-dimensions and embedded in Rn , then it is an (n − 1)-dimensional
torus Tn−1 . Does this remain true without the hypothesis of almost periodicity?
By Denjoy’s irregular case on the torus it seems necessary to assume that the
minimal set contains an (n − 1)-dimensional ball. In particular, does it remain true
for solutions of autonomous systems of differential equations in n dimensions if
the minimal set of a solution contains an (n − 1)-dimensional cylinder formed by
the product of a small (n − 2)-dimensional ball in the plane normal to the solution
through the center of the ball and an interval of length 1 of all solutions through
points of the ball? This is true for n = 2 when the ball reduces to a point. Is it true
for n > 2, if it is true for R2 is it true for other n-dimensional manifolds?
867. December 8, 1971; Kazimierz Urbanik (Wrocław) (in Polish)
Let D denote the space (càdlàg) of right continuous functions on [0, 1] with limits on
the left, with the Skorokhod topology. Moreover, let μ be a Borel measure induced
on D by a homogeneous process with independent increments. Show that μ does
not have structure of Hilbertian measure, i.e. there is no linear set H ⊂ D which
is μ -measurable and has the structure of a Hilbert space with a μ -measurable inner
product with the topology stronger than the Skorochod topology, and μ (H ) = 1.
For the Wiener process the answer is yes (Stanisław Kwapień (Warszawa) and
Małgorzata Guerquin (Wrocław)).
868. March 01, 1972; Bronisław Knaster (Wrocław) (in Polish)
Is every closed subset of an absolute retract X the fixed-point set of a continuous
map f : X → X?
CM 27, P846
871. April 25, 1972; Karol Borsuk (Warsaw) (in Polish)
Can any locally connected continuum in the Euclidean 3-space E3 , disconnecting
E3 , be continuously mapped into itself without a fixed point?
CM 27, P847
872. April 25, 1972; Karol Borsuk (Warsaw) (in Polish)
Does every dendroid have a trivial shape?
Does any finite union of dendroids have the planar shape (i.e., the shape of some
plane compactum)?
CM 27, P848
873. April 28, 1972; Wojbor A. Woyczyński (Wrocław, Cleveland) (in English)
Let X be a Banach space. The gradient g : SX → SX of the norm . in X is defined
by the equality g(x)ξ = limt→0 1t (x + tξ − x). Does the gradient in Lp -spaces,
1 < p < 2, satisfy the Hölder condition with exponent p − 1?
Answer: Yes; W. Woyczynski, Random series and laws of large numbers in some
Banach spaces. Teor. Veroyatn. Primen. 18 (1973), 361–367. reprinted: Theory
Probab. Appl. 18 (1973), 350–355.
893. May 30, 1974; Richard M. Schori (Baton Rouge, LA)

If X is a compact metric space, by 2X we mean the collection of all nonempty closed
subsets of X metrized with the Hausdorff metric. If ε > 0, the ε -hyperspace of X,
2Xε = {A ∈ 2X : diameter A ≤ ε }.
If X is a compact ANR, does there exist a metric on X (perhaps any convex metric
will suffice) such that for sufficiently small ε > 0, 2Xε is a) locally contractible, b) an
ANR, c) of the same homotopy type as X, d) a Hilbert cube manifold, e) a Hilbert
cube manifold homeomorphic to X × Q, where Q is the Hilbert cube?
Furthermore, if X ∈ AR, is 2Xε homeomorphic to Q for each ε > 0?

If e) and this last question are both true, then X ∈ AR implies that X × Q ∼
= Q. If this
is the case, then, by Chapman, each compact ANR has finite homotopy type.
CM 35, P981
898. December 13, 1974; Sam B. Nadler, Jr.
Let Hn denote the Hausdorff metric for the space Cn of all nonempty subcon-
tinua of Euclidean space Rn . Let An = {A ⊂ Rn : A is an arc}, Σn = {A ⊂ Rn :
A is a simple closed curve}, Tn = {A ⊂ Rn : A is a simple triod}, Pn = {A ⊂ Rn :
A is a pseudoarc}.
Question 1: Is A3 homogeneous?
By the Schönflies theorem, A2 is homogeneous; note: homeomorphisms
h : R2 → R2 lift up to A2 . But in R3 there are wild arcs. Are these arcs really
wild in A3 ?
Prize: 1 bottle of wine.
Question 2: Same as Question 1 for the spaces Σn , Tn and An≥3 .
Bing has shown Pn is a residual Gδ in the space Cn .
Question 3: Is Pn homogeneous and, if so, is it a topological group?
If Pn is homogeneous, then I would conjecture that Pn is homeomorphic to 2 !
This would be interesting because then every functional analysist would know what
a pseudoarc is — namely, a point in the Hilbert space!
909–910. January 2, 1975; Bronisław Knaster (Wrocław) (in Polish)

Let GFPP denote the existence of a point x ∈ F(x) for any upper semi-continuous
function defined on points x of a continuum X whose values are subcontinua
F(x) ⊂ X. R. Mańka proved (Fund. Math. 1976) property GFPP for all hereditarily
decomposable and hereditarily unicoherent X (so called λ -dendroids).
Question 1: For what continua X does property GFPP of X induce property GFPP
of the hyperspace C(X) of all subcontinua of X?
Question 2: For what X and F does property GFPP of X induce property GFPP
of F(X)?
911. January 2, 1975; Bronisław Knaster (Wrocław) (in Polish)
Is (and when) a deformation retract of a contractible continuum (in particular, of a
contractible fan) contractible?
912. January 2, 1975; Bronisław Knaster (Wrocław) (in Polish)
For what continua X is the uniform limit of a sequence of confluent maps X → Y (in
the sense of J. J. Charatonik, Fund. Math. 56 (1964)) a confluent map?
Prize: For solving any of Problems 908–913: an opportunity to tell two jokes of
moderate length.
915. April 8, 1975; Nigel J. Kalton (Swansea) (in English)

Let (E, τ ) be an F-space with a separating dual. Prove or disprove that E contains
an infinite dimensional weakly closed subspace G and a closed infinite dimensional
subspace H such that G ∩ H = {0} and G + H is closed.
918. June 8, 1975; Aline Bonami (Orsay) (in French)
Consider the set Λ of all the integers of the form 2k 3 , k, ∈ N. Using the Paley-
Littlewood theory one can prove that there exist constants Ap , p > 2, such that for
each trigonometric polynomial P with its spectrum in Λ , P(x) = ∑λ ∈Λ aλ exp(iλ x),
one has Pp ≤ Ap P2 . Does there exist a direct proof?
Moreover, can one prove more precisely that ∑λ ∈Λ |aλ |2 < ∞ implies
exp (μ | ∑ aλ exp(iλ x)|) ∈ L1 for each μ > 0?
921. July 30, 1975; F. Burton Jones (Riverside, CA)
After 40 years, does there exist a normal Moore space (i.e., a space satisfying
R. L. Moore’s axiom 0 and parts 1,2,3 of axiom 1) which is not metric?
One would hope for a “honest” example of such a space without some weird set
theory assumption)
922. October 10, 1975; Fleming Topsøe (Copenhagen) (in English)
Is it possible to find in Q – the space of rationals, a sequence (Pn )n≥1 of probability
measures such that Pn converges in topology of weak convergence to some measure
P, and such that the only compact subset K ⊂ Q for which Pn K → PK holds is the
empty set? With Q replaced by a Polish space such an example does not exist.
Prize: A week in Copenhagen for an example.
920. July 30, 1975; F. Burton Jones (Riverside, CA)

After 50 years, Professor Knaster’s question: “Is the circle the only homogeneous
plane continuum?” has been reduced to:
Is every hereditarily indecomposable homogeneous plane continuum a pseudo-arc?
i.e., is it chainable or circularly chainable?
923. December 11, 1975; David P. Bellamy (Warsaw)
Can every finite-dimensional metric hereditarily indecomposable continuum be
embedded into a finite product of pseudo-arcs?
As a trivial application of my results in: Mapping hereditarily indecomposable
continua onto a pseudo-arc, Lecture Notes in Math. 375 (1974), 6–14, it can be
established that (1) every Hausdorff hereditarily indecomposable continuum can be
embedded into some product of pseudo-arcs; (2) every metric hereditarily indecom-
posable continuum can be embedded into a countable product of pseudo-arcs.
CM 37, P1006
Remark: The question was repeated by the author in [Proceedings of the
International Conference on Geometric Topology, PWN Warszawa, 1980, p. 459]
and [Open Problems in Topology II, Edited by E. Pearl, Elsevier, Amsterdam,
2007, p. 259]. R. Pol proved in [Topology Proc. 16 (1991), 133–135] that each
n-dimensional hereditarily indecomposable metric continuum can be embedded into
the product of n one-dimensional hereditarily indecomposable metric continua.

If M is an open n-manifold for n ≥ 2 and β M \ M is connected, is β M \ M an
aposyndetic continuum?
This is true if M = Rn .
CM 37, P1007
What are necessary and sufficient conditions to ensure that a metric continuum X
cannot be mapped onto the cone over itself?
If X is an indecomposable Hausdorff continuum with infinitely many composants,
is the cardinality of the set of composants of X always equal to 2m for some infinite
cardinal number m?
929. March 23, 1977; Tadeusz Maćkowiak (Wrocław)
Let a continuum X be such that dim X ≥ 2. Does it follow that X contains an
hereditarily indecomposable continuum?
Remark: By a well-known Mazurkiewicz theorem, any metric continuum of
dimension at least 2 contains a nondegenerate indecomposable subcontinuum.
932. August 3, 1977; Ryszard Frankiewicz (Wrocław)

Can the spaces β D(ω1 ) \ D(ω1 ) and β N \ N be homeomorphic?
D(ω1 ) = the discrete space of cardinality ω1 .
941. April 05, 1978; James E. West (Ithaca, NY)
p
Let E →
− B be a locally trivial fiber bundle with fiber F and suppose E, B, and F to
be compact metric.
2p
Under what conditions (if any) is 2E −→ 2B a locally trivial fiber bundle with fiber
2F ? In particular, if E, B, F are ANR’s? What about Hurewicz fibrations? Serre
fibrations?
CM 43, P1187
p
Let E →− B be a Hurewicz fibration with each fiber homeomorphic to a given Hilbert
cube manifold F.
If E and B are also Hilbert cube manifolds, under what conditions is p a locally
trivial bundle?
CM 43, P1188

p
Let E →− B be a Hurewicz fibration with B = the Hilbert cube, E a compact ANR,
and each fiber of p a nondegenerate absolute retract.
Must there exist two cross-sections, σ and τ of p, with disjoint images? (Problem
of West and Toruńczyk).
If not, give a characterization of those cross-sections σ for which there exists a
cross-section τ with image disjoint from that of σ .
CM 43, P1189
945. September 15, 1978; Charles L. Hagopian (Sacramento, CA)
Does every arcwise connected disc-like continuum have the fixed-point property?
CM 44, P1201
946. October 21, 1978; Paul Erdös and Endre Szemerédi (Budapest) (in English)
Let 1 ≤ a1 < · · · < an be n integers. Let f (n) be the smallest integer so that the set
ai + aj , ai aj , 1 ≤ i ≤ j ≤ n, contains at least f (n) distinct integers. Szemerédi and I
proved ( > 0 is an absolute constant)
(∗) n1+ < f (n) < n2 exp( log n/ log log n).
The upper bound may give the right order of magnitude; f (n) > n2−ε seems certain.
Denote by F(n) the smallest integer so that there are at least F(n) distinct integers
of the form ∑ εi ai , ∏ aεi i , εi = 0 or 1. We conjecture F(n) > nk for every k if
n > n0 (k). We proved F(n) < n log n/ log log n which perhaps gives the right order
of magnitude for F(n).
953. May 15, 1979; Hubert Delange (Orsay) (in English)
Let H (C) be the vector space of entire functions of one complex variable, with the
topology of convergence on compact sets. In H (C) consider the set S consisting of
those entire functions F which satisfy |f (x + iy)| ≤ exp(|y|) for every real x and y
(which is equivalent to |f (n) (x)| ≤ 1 for every n ≥ 0 and every real x). S is a compact
convex set. Find its extremal points.
The functions z → C exp(iaz), where |C| = 1 and a ∈ [−1, 1], are extremal points.
If there were no other extremal points, every function of S could be put into the form
1
f (z) = −1 exp(itz) d μ (t), μ – a complex measure on [−1, 1]. But this is not the
case. Example: f (z) = 0z sinu u du = p.v. −1 1
exp(itz) dtt .
961. October 27, 1979; Fulvio Ricci (Torino, Pisa) (in English)
Given two pseudomeasures T and U on the torus T such that T̂(0) = Û(0) = 0,
consider the distributional derivative (PT PU ) of the pointwise product PT PU , where
PT and PU are primitives of T and U, respectively. In general, (PT PU ) is not a
pseudomeasure. This product has been introduced by J. Benedetto.
Are there closed subsets E of the torus T which are not sets of strong spectral
resolution, and such that if T and U above have support in E, then (PT PU ) is a
pseudomeasure?
4 5
Let E = {0} ∪ 1
2m + 21n : m, n ≥ 0 . What is the answer to the first question in this
case?
Remark. If U ∈ M (E), then (PT PU ) ∈ A (E), see F. Ricci, A multiplicative
structure on some spaces of pseudomeasures on the circle and related properties,
Bull. Sci. Math., II. Ser. 103 (1979), 423–434.

The Scottish Book - Mathematics From The Scottish Café

Uploaded by

Copyright:

Available Formats

The Scottish Book - Mathematics From The Scottish Café

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Scottish Book - Mathematics From The Scottish Café

Uploaded by

Copyright:

Available Formats

R.

The Scottish Book

ISBN 978-3-319-22896-9 ISBN 978-3-319-22897-6 (eBook)

Library of Congress Control Number: 2015947275

Mathematics Subject Classification (2010): 01A60.

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media (www.

San Diego, CA, USA R. Daniel Mauldin

San Diego, CA, USA R. Daniel Mauldin

The enclosed collection of mathematical problems has its origin in a notebook,

Los Alamos, NM, USA S. Ulam

Los Alamos, NM, USA S. Ulam

Part I The Scottish Book Conference Lectures

Part II The Scottish Book Problems

Part III A Brief History of Wrocław’s New Scottish Book

Part IV The New Scottish Book

© Springer International Publishing Switzerland 2015 3

The usual approximation methods, the “epsilon” approaches, are almost by

From the original Scottish Book; the handwriting in the Mazur-Orlicz

From the original Scottish Book; the handwriting in the Banach-Ulam

It is a special pleasure to be introduced by my old friend Erdős. The use of the

© Springer International Publishing Switzerland 2015 13

with α1 + α2 = 1/2 and β1 + β2 = 0. (In Problem 178 the theorem is slightly

In probabilistic terms, if a sum of two independent random variable is Gaussian,

as a candidate for the characterization of the class of characteristic functions

which can be written in the form

unexpected contexts, as illustrated by Problem 161) of being—so to speak—on

© Springer International Publishing Switzerland 2015 21

publication specializing in this selected branch of mathematics. Once a strong point

∑ |amn | ≤ constant, for all m.

{|um (x)| ≤ M|x| for x ∈ E and m = 1, 2, . . .}.

Of course, the result of Toeplitz is a consequence of this general theorem. Sierpiński,

have no obvious answer since we have no measure in the space of rearrangements

© Springer International Publishing Switzerland 2015 27

Problems 22 (Ulam-Schreier) and 99 (Ulam) are as follows:

converges. Is it true that ∑n=1 n|an | converges?

great difficulties. Nevertheless R.O. Davies, the English mathematician, succeeded

5.1 Definition and examples

Let E be a real vector space and X ⊂ E be an arbitrary subset. A set-valued function

for each finite subset {x1 , . . . , xs } ⊂ X.

G(x) = { y ∈ C | φ (y) ≤ φ (x) }.

We show that G : C → 2C is a KKM-map. For a contradiction let y0 = ∑ λi xi

G(x) = { y ∈ C | fy − y ≤ fy − x }.

© Springer International Publishing Switzerland 2015 35

We show that G : C → 2C is a KKM-map. Indeed, let y0 = ∑ λi xi be a convex

G(x) = { y ∈ C | p(fy − y) ≤ p(fy − x) }.

The same argument as in the previous example shows that G : C → 2C is a

We show that G : C → 2E is a KKM-map. Indeed, let y0 ∈ conv{x1 , . . . , xn }. If

The principle of KKM-maps

Theorem 5.1. Let E be a vector space, X an arbitrary subset of E, and G : X → 2E

Proof. We argue by contradiction, so assume that n1 G(xi ) = 0. / Working in the

L ⊂ E is closed in the Euclidean topology of L.

By Brouwer’s fixed point theorem, f would have a fixed point x0 ∈ C. Letting

x0 = f (x0 ) ∈ conv{ xi | i ∈ I } ⊂ { G(xi ) | i ∈ I }

and, with this contradiction, the proof is complete.

Theorem 5.2. (Ky Fan [7]). Let E be a topological vector space4 , X ⊂ E an

Theorem 5.3. Let E be a vector space, X an arbitrary subset of E, and G : X → 2E

G(x) = { y ∈ C | fy − y ≤ fy − x }.

G(x) = { y ∈ C | y − f (y) ≤ x − f (y) }.

Proof. Define G : C → 2C by y → C\A−1 y; each G(y) is a nonempty set closed in

the theorem will be proved by showing { G(x) | x ∈ C } = 0.

Proof. Uniqueness being evident, let f : C → H be given by y → y − x0 ; clearly,

Given a cartesian product X = ∏ni=1 Xi of topological spaces, let X j = ∏i=j Xi and

Let {An }∞n=1 be a sequence of pairwise disjoint subsets of H. If f ( χAn ) = cn = 0