I934-] BANACH ON LINEAR OPERATIONS 13

BANACH ON LINEAR OPERATIONS

Théorie des Opérations Linéaires. By Stefan Banach. Warsaw, Monografje
Matematyczne, Vol. I, 1932. vii+254 pp.
The present book, the first of the series published by the Monografje
Matematyzcne, is an enlarged and considerably modified translation of the
Teorja Operacyi (in Polish) by Banach. It represents a noteworthy climax of
long series of researches started by Volterra, Fredholm, Hubert, Hadamard,
Fréchet, F. Riesz, and successfully continued by Steinhaus, Banach, and their
pupils. In a short review it is impossible to give even an approximate idea of
the richness and importance of the material, entirely new to a large extent,
which is gathered in this book. Roughly speaking the book treats of the field
which was touched upon by Hildebrandt in his excellent address on linear
functional transformations in general spaces,* but naturally is considerably
more extensive and intensive. The theory of linear operations is a fascinating
field in itself but its importance is still more emphasized by numerous beautiful
applications to various problems of analysis and function theory (real and
complex). A number of these applications are contained in the book; a still
larger number of them, increasing from year to year, are being inspired by the
ideas developed in the book. An attentive reader will find a constant source of
inspiration in this monograph which undoubtedly will exercise a great influence
on the further development of the subject. A description of the contents fol-
lows.
In the Introduction (19 pp.) the author recalls some basic properties of
Lebesgue and Stieltjes integrals, and of metric spaces. Particular attention is
given to subsets of such spaces, which either are Borel measurable or possess
the property of Baire, and also to Borel measurable operations transforming
one metric space into another. In Chapter 1 (Groupes, 6 pp.) the author spe-
cializes the metric space under consideration by adding the usual postulates
for groups (with respect to the operation " + ", with inverse designated by
" — ") and also the two postulates concerning limits:
xn —* x implies ( — xn) —» ( — x) ;
xn —• x and yn -> y implies (xn + yn) -* (« + y).
It is proved that every subgroup which possesses Baire's property and is of the
second category coincides necessarily with the whole space (assumed to be
complete). The additive and linear (additive+continuous) operations are in-
troduced at this early stage. It is proved that the set where a sequence { Un(x)}
of linear operations converges, or even where it is only bounded, is either
the whole space or a set of the first category (the space is assumed to be con-
nected). Chapter 2 (Espaces vectoriels généraux, 9 pp.) deals with linear spaces
and additive operations defined on such spaces. The central point here is an
important theorem concerning a possibility of extending to the whole space

* This Bulletin, vol. 37 (1931), pp. 185-212.

14 J. D. TAMARKIN [January,

an additive functional (a real-valued operation) defined on a linear sub-space.

Theorems of this kind originated with F. Riesz (in the case of space of con-
tinuous functions) and the theorem in question was proved independently
by Hahn. As an application it is shown that every bounded sequence has a
"limit" which is distributive and possesses some other properties of the usual
limit, and is always enclosed between the limits of indétermination of the
sequence. An analogous application is made to show that an arbitrary bounded
function may be supplied with an "integral/' as well as every bounded linear
set with a "measure." In Chapter 3 (Espaces du type (F), 18 pp.) a metric com-
plete space is specialized to be linear and to satisfy three additional postu-
lates:
(x, y) = (x - y,0);
hn-*0 implies hnx —> 0, where {hn\ is a sequence of constants;
xn —» 0 implies hxn —* 0, where h is a constant.
A fundamental property of linear operations (transforming such a space E of
type (F) into a set of another space Ei, also of type (F)) is that the range of a
linear operation is either a set of the first category in Ei or Ei itself, while, in
the case when the transformation (linear) is one to one, it is also bicon-
tinuous. This chapter contains important applications to the theory of continu-
ous non-differentiable functions, the continuity of solutions of partial differen-
tial equations, and the solution of certain systems of infinitely many equations
of a general type.
The next specialization is furnished by normed spaces which are metric and
linear with the metric \\x\\ satisfying the postulates
ll* + ?llslWI+y. INI-MINI.
where c is a constant. A normed space, if it is complete, is designated as a
space (B). Such spaces have been also introduced by Wiener, independently
and almost simultaneously with Banach. Chapter 4 (Espaces normes, 25 pp.)
deals with normed spaces in general. Here we find a determination of the form
of the most general linear functional in various important special spaces,
such as spaces of all convergent sequences, (c) ; of sequences x = {£n} with

of continuous functions, (C); of measurable functions x(t) with

Jl\x(t)\m< oo, (LP).
We find also interesting applications to the theory of approximation (in vari-
ous spaces) to a given function by means of linear combinations of functions
of a given set, and to the moment problem. Chapter 5 (Espaces du type (B),
18 pp.) deals especially with normed complete spaces. Here we find the fun-
damental result that if { Un(x)} is a sequence of linear operations, bounded at
each point of the second category of a space E of type (J5), then the sequence of
norms {||£^||} is bounded. This theorem is the most general among similar
results of more or less special nature obtained by various other authors. Some
interesting applications to the comparative study of definitions of summa-
bility are also found in this chapter. We mention also some results of impor-
tance for the theory of Fredholm integral equations with kernels of "bounded"
I934-] BANACH ON LINEAR OPERATIONS 15
type, which are out of the reach of the more powerful theory developed by
von Neumann and Stone. Chapter 6 {Opérations totalement continues et as-
sociées, 10 pp.) introduces the notions of the adjoint (or conjugate) operation,
and of a completely continuous operation, whose importance is so well known
in the more special theories of integral equations and bilinear forms. Chap-
ter 7 (Suites biorthogonales, 9 pp.) contains various applications to the gen-
eral theory of biorthogonal expansions. Of a particular elegance are the results
concerning factor-sequences leaving invariant the "class" (2>) or (C) of a
Fourier series. Chapter 8 (Fonctionnelles linéaires dans les espaces du type (B),
18 pp.) is devoted to a more detailed study of various sets of linear func-
tional in spaces (B). Three important notions are introduced and discussed,
those of sets of functionals which are regularly closed, transfinitely closed,
and weakly closed. The two first are shown to be equivalent for every space
(B), while all three are equivalent in case the space (B) is also separable. Con-
ditions are given in order that a given set of functionals coincide with the space
of all the functionals (the conjugate space). Chapter 9 (Suites faiblement con-
vergentes d'éléments, 12 pp.) gives a discussion of weak convergence, an im-
portant notion introduced in special cases by F. Riesz. After all these prepara-
tions the author can easily develop his beautiful theory of linear functional
equations in Chapter 10 (Équations fonctionnelles linéaires, 20 pp.). As very
special instances of applications of results of this chapter we may mention the
theory of non-singular, and some singular, Fredholm integral equations and of
Hubert's theory of limited and completely continuous bilinear forms. The
power of the method (the basic ideas of which were previously introduced by
F. Riesz and by Hildebrandt) is due to the fact that it is free from any special
way of representation of operations in question, such as integrals or matrices,
but deals directly with operations themselves.
The last two chapters treat of entirely different questions. These chapters,
as well as the Appendix at the end of the book, were not present in the Polish
edition of the book. Chapter 11 (Isométrie, equivalence, isomorphic, 28 pp.)
discusses various types of correspondences (all of them one to one) between
two spaces E and E\ of type (Ê). These spaces are called isometric if the cor-
respondence preserves the norm, isomorphic if this correspondence is linear,
and equivalent if they are isomorphic and isometric. Some of these notions can
be of course extended to more general spaces. A rotation of E about a point
XQ is defined as an isometric transformation of E into itself, which leaves invar-
iant the point xo. Various beautiful results are found in this chapter, of which
we mention only three: Every isometric transformation of a linear normed
space into another space of the same type is linear, provided it transforms the
zero element into the zero element (Mazur and Ulam). Let Q and Qi be two
metric complete and compact sets. A necessary and sufficient condition that
they be homeomorphic is that the spaces (CQ) and (CQJ of real-valued con-
tinuous functions defined over Q and Qi respectively be isometric. Every
separable space of type (B) is equivalent to a closed linear sub-space of the space
(C) of continuous functions; every metric separable space is isometric with a
sub-space of (C), The expressions for the most general rotations in spaces (C),
(c), (2>), (P) are also found in this chapter. Chapter 12 (Dimensions linéaire, 15
pp.) introduces a method of comparing various spaces according to their lin-
16 J. D. TAMARKIN [January,

ear dimensions. We say that, for two spaces of type (F), (1) dimiE^dimiEi
if E is isomorphic with a linear closed sub-space of Ei; that dimiE^dimiEi
if also (2) dimiEi^dimiE, while d i i m £ < d i m i E i if (1) is satisfied but (2) is not.
If neither (1) nor (2) is satisfied, the linear dimensions of E and E\ are said to
be incomparable. I t should be observed that, while two isomorphic spaces are
of t h e same linear dimensions, there exist separable spaces (B) which are not
isomorphic although their linear dimensions are the same. Special attention
is given here to spaces (Lp) and (lp). I t is shown t h a t if dimz(Z>)==dimj(/«),
where/), q>l, then p = q) if Kp<2<q, then the linear dimensions of (Lp)
and (/«) are incomparable; if Kp^2, then dimj(L 2 ) < d i m i ( i > ) ; if dimi(Lp)
<dirm(Z,«), where p, q>l, then £ = g = 2; the condition p = q = 2 is necessary
and sufficient t h a t dimj(i>) =dimz(/ a ); if l<p^2t then d i m j ( I > ) > d i i m ( ^ ) .
The Appendix (18 pp.) contains an additional discussion of the weak conver-
gence of elements and of functional. T h e book closes with 21 pages of addi-
tional remarks containing historical information and statements of various
generalizations of the results in the text, as well as numerous unsolved prob-
lems. These remarks reveal t h a t the whole subject is in a state of vigorous de-
velopment. The usefulness of many of these remarks would be considerably
greater if they contained hints indicating how the results in question have been
obtained. The exposition as a rule is clear and detailed, although not easy in
several places. The number of misprints (of which only two are corrected in the
list of Errata) is not negligible.
I n conclusion the reviewer may express his conviction that Banach's mono-
graph should occupy a permanent and honorable place on the desk of every
one who is interested in the theory of linear operations, to be replaced only by
subsequent, corrected and augmented editions, which undoubtedly will follow
before long.
J. D. TAMARKIN

SAKS ON I N T E G R A T I O N

Théorie de T Intégrale. By Stanislaw Saks. Warsaw, Monograf je Matematyczne,

Vol. I I , 1933. v i i + 2 9 0 pp.
The present volume is the second one of the series Monografje Matema-
tyczne. I t is a translation, entirely revised and augmented by several important
chapters, of the author's Polish book, Zarys Teorji Calki, Warsaw, 1930. I t
fills in a serious gap in the literature of the real function theory, and of the
theory of diflferentiation and integration, which has been acutely felt during
all the recent period of vigorous growth and development of these disciplines.
While Hermite, together with a large part of his contemporaries and immediate
successors, including Poincaré, contemplated with horror the pathological
cases of functions without derivatives, the ideas and methods created pre-
cisely for handling such bad cases turned out to be extremely fruitful and
indispensable for the treatment of most classical and venerated problems of
analysis. F a r from bringing in the feared anarchy and disorder, they have al-
lowed us in many instances to reach a harmony and completeness of results
which were entirely out of the reach of older classical methods. The problems