Theory of Bases

Springer Tracts in Natural Philosophy
Volume 18
Edited by B. D. Coleman
Co-Editors: R.Aris . L. Collatz . J. L. Ericksen
P. Germain' M. E. Gurtin . M. M. Schiffer
E. Sternberg . C. Truesdell
J iirg T. Marti
Introduction to the
Theory of Bases
Springer-Verlag Berlin Heidelberg New York 1969

Jiirg T. Marti
Department of Mathematics
University of Illinois, Urbana
ISBN·13: 978·3·642·87142·9 e·ISBN·13: 978·3·642·87140·5

DOl: 10.1007/978·3·642·87140·5
All rights reserved. No part of this hook may be translated or reproduced in any form without written permission from
Springer·Verlag. © by Springer·Verlag Berlin' Heidelberg 1969. Library of Congress Catalog Card Number 73-83680
Title No. 6746
Softcover reprint ofthe hardcover 1st edition 1969

To Rita
Preface
Since the publication of Banach's treatise on the theory of linear

operators, the literature on the theory of bases in topological vector
spaces has grown enormously. Much of this literature has for its origin
a question raised in Banach's book, the question whether every sepa-
rable Banach space possesses a basis or not. The notion of a basis
employed here is a generalization of that of a Hamel basis for a finite
dimensional vector space. For a vector space X of infinite dimension,
the concept of a basis is closely related to the convergence of the series
which uniquely correspond to each point of X. Thus there are different
types of bases for X, according to the topology imposed on X and
the chosen type of convergence for the series.
Although almost four decades have elapsed since Banach's query,

the conjectured existence of a basis for every separable Banach space
is not yet proved. On the other hand, no counter examples have been
found to show the existence of a special Banach space having no basis.
However, as a result of the apparent overconfidence of a group of
mathematicians, who it is assumed tried to solve the problem, we have
many elegant works which show the tight connection between the theory
of bases and structure of linear spaces. In the more general setting of
a separable locally convex topological vector space or a complete linear
metric space, the basis problem is now solved; there actually are ex-
amples of such spaces which have no basis. By the nature of the prob-
lem, the methods of proof used in the theory of bases are those of
functional analysis.
A few conditions sufficient for a sequence to form a basis of a certain

type are now known. Moreover bases have been constructed for most
of the separable Banach spaces which are presented as examples in
text-books on topological vector spaces. For instance, the trigono-
metrical system is a basis for the Hilbert space L2 [0,2n]. On the other
hand, if one assumes the existence of a basis for an abstract Banach
space X, one obtains valuable indications on the structure of X or of
closed linear subspaces of X. The assertions may concern weak sequen-
VIII Preface
tial completeness, separability, reflexivity, dimension, or weak con-

ditional compactness of bounded sets.
Some generalizations of the concept of a basis for a Banach space

have greatly enriched the theory. On the one hand, it is natural to see
how results on Banach spaces may be generalized to complete linear
metric spaces or locally convex topological vector spaces. The definition
of a basis, on the other hand, may be generalized itself. If one replaces
the elements of a basis by linear subs paces of a Banach space or an
F-space, one obtains decompositions of these spaces, and these decom-
positions exhibit some properties similar to those of the bases. It is
worth noting that decompositions exist for every Banach space of in-
finite dimension. Carrying the generalization of a basis one step further,
one first discards the idea of a series expansion, requiring of a biorthog-
onal system {x",f;j only that {x,,} is total in the space X; such a
system is called a dual generalized basis. Moreover, if f,,(x)=O for
all ), implies that x=O for all x in X, the dual generalized basis is
said to be a Markushevich basis. Finally, when the requirement of
totalness, common to all definitions of bases and decompositions de-
scribed above, is dropped along with the requirement of countability,
we obtain the concept of a generalized basis for a topological vector
space X.
Although there are presently more than two hundred publications

on the theory of bases, up to now no text-book has been issued which
collects and systematizes the essential results of the subject. This tract
is an attempt to meet such a need.
The first chapter contains a short introduction to the working tools

from functional analysis. The theorems are given there without proofs.
This is justified, since there are now many well-written standard works
on this subject. In Chapter II the fundamental theorems on uncondi-
tional and absolute convergence of series in Banach spaces are derived.
Definitions and properties of the most important types of bases for
Banach spaces, together with examples of bases in some well-known
spaces of this type are given in Chapter III. The fourth chapter deals
with the connections of bases, projections, orthogonality and simple
<~ -spaces, as well as with equivalent bases for Banach spaces. The
known facts on the structure of Banach spaces with bases are explained
in Chapter IV, and the following chapter is concerned with bases for
Hilbert spaces. Decompositions are introduced in Chapter VII, and
the application of both bases and decompositions in the theory of
B-algebras, including compact operators, operators of finite rank, and
Preface IX
the theory of proper 1Hings, are discussed in Chapter VIII. Some of

the most interesting results on generalized bases for topological vector
spaces are presented in the final chapter.
I thank the Editor, Professor B. D. Coleman for his friendly co-

operation and the Springer-Verlag for the careful preparation of this
book.
Urbana, Illinois, February 1969 JURG T. MARTI

Contents
I. Linear Transformations . . . 1
1. Linear Topological Spaces 1
2. Linear Transformations. . 7
3. Conjugate Spaces and Weak Topologies. 10
4. Special Banach Spaces . . . . . . 14
II. Convergence of Series in Banach Spaces 18

1. Relations among Different Types of Convergence. 18
2. Unconditional and Absolute Convergence. 22
III. Bases for Banach Spaces. . . . . . . . . . 28

1. Bases Corresponding to Different Topologies 28
2. Biorthogonal Systems . . . . . . . . . 30
3. Shrinking and Boundedly Complete Bases. . 34
4. Unconditional Bases . . . . . . . . . . . 38
5. Absolutely Convergent Bases and Uniform Bases. 42
6. T-Bases . . . . . . . . 43
7. Bases for Special Spaces 47
IV. Orthogonality, Projections and Equivalent Bases 55

1. Bases and Projections. . . . . . . . . . . 55
2. Orthogonality, simple .IV;. -Spaces and Monotone Bases 60
3. Equivalent Bases. . . . . . 62
V. Bases and Structure of the Space 69

1. Bases, Completeness and Separability. 69
2. Bases and Reflexivity. . . . 72
3. Criteria for Finite Dimension 77
VI. Bases for Hilbert Spaces. . . . 79

1. Monotone and Orthonormal Bases . 79
2. Unconditional Bases for Hilbert Spaces . 83
XII Contents
VII. Decompositions 86
1. Decompositions of F-Spaces . 86
2. Decompositions of Banach Spaces 90
VIII. Applications to the Theory of Banach Algebras 98

1. Two-Sided Ideals of Operators of Finite Rank 99
2. n-Rings . . . . . . . . . . . . . . . . . 99
3. Proper n-Rings of Schauder Decompositions. 102
4. Minimal Schauder Decompositions. . . . 107
5. Banach Algebras and Unconditional Bases 109
IX. Some Results on Generalized Bases for Linear

Topological Spaces . . . . . . . . . . . . 113
1. Definition and Fundamental Properties of Generalized
Bases. . . . . . . . . 114
2. Dual Generalized Bases. 118
3. Examples. . . . . . . 120
4. Similar Bases . . . . . 122
5. Continuity of the Coefficient Functionals 125
Bibliography . . . . . . 130
Author and Subject Index. 146

CHAPTER I
Linear Transformations
In the four paragraphs of this chapter we present some basic
definitions and facts from functional analysis, as well as applications
in special spaces. These preliminaries will be used in the subsequent
chapters. Since many introductions to functional analysis are now
available, in order to save space, we omit proofs of all of the lemmas,
theorems and corollaries given here. Moreover, one will find here only
the working tools which are really needed for the development of the
theory of bases. We begin by defining various abstract spaces, and we
list their most important properties. Then we investigate linear trans-
formations of one space into another, continue with some facts on
conjugate spaces, and conclude with results for several spacial spaces.
It is supposed that the reader is familiar with the notions of a field,
a linear space (also called a vector space or a linear vector space), a
subspace of a linear space, an algebra, and with some elementary defi-
nitions from the theory of sets and the theory of measure and integration.
1. Linear Topological Spaces
A mapping (sometimes also called map, function, transformation or

operator) f: X -> Y from (of) a set X into a set Y is a rule which assigns
to each member x of X a unique member f(x) in Y. For any subset
A of X we write f(A) to denote the set {f(X)IXEA}. X is called the
domain of f and f(X), the image of X under f, is called the range of f.
If f(X)= Y we say that f is a function of X onto Y. f is one-to-one if
and only if f(xd=f(x 2 ),X I ,X 2 EX always implies Xl =x 2 • For any
subset A of Y let us denote JI(A)={XEXlf(x)EA}. f-I(A) is called
the inverse image of A. In particular, the inverse image of a single point
y in Y will be denoted by f - I (y). Iff is one-to-one and onto, X =f - I (y)
defines a function f - I of Y onto X, called the inverse (function) of f
If A is any subset of X, the real function XA' defined by XA(X) = 1, XE A,
and by XA (x) = 0, x ¢ A is called the characteristic function of A.
2 I. Linear Transformations
A topology for a set X is a family, of subsets of X, called open sets,

such that the void set, the set X, the union of arbitrary many open sets,
and the intersection of finitely many open sets are open (i.e. are'in ,).
The set X, endowed with the topology, is a topological space. If, and
" are two topologies for X and if ,c r', then r is said to be weaker than
,', and,' is said to be stronger than 1:. r is equal to r' if and only if ,=r',
A subset A of X is closed if its complement with respect to X is open.
The intersection A of all closed sets containing a subset A of X is called
the closure of A. A subset B of A is dense in A if A c B. X is separable
if there exists a countable dense set in X. A topological space Y is a
(topological) subspace of X if and only if Y c X and the sets which are
open in Yare precisely the intersections of Y with the open subsets of
X. A neighborhood of a set A in X is a set containing an open set which
contains A. A collection B of open sets of X is a base (for the topology
of X) at the point x in X if for any neighborhood N of x, there exists an
open set A in B such that x E A c N. A mapping I of X into another
topological space Y is continuous if and only if I-l(A) is open in X
for every open set A in Y. A one-to-one mapping I of X onto Y such
that I (A) is open in Y if and only if A is open in X, is called a homeomor-
phism. If such a homeomorphism exists, X and Yare said to be home-
omorphic.
If A is a set in which a relation ~ is defined with the following
properties: (i) if ), ~ Jl and Jl ~ v, then A~ v; (ii) A ~ A and (iii) for A, JlE A
there is a v in A such that A~ v and Jl ~ v; then A is said to be a directed
set. The notation A ~ Jl is equivalent to Jl ~ A. A net {x;.} in a topological
space X is a map of a directed set A into X. If A is the directed set of
integers i = 1,2, ... , then, evidently, {x;} is a sequence in X. A net {x;.}
in X converges to a point x in X written x Ie ~ X or x = lim x;. if to
Ie
every neighborhood N of x there corresponds a Jl in A such that x), E N
for every }, ~ Jl. If such a point x exists it is called a limit. A topological
space X is a H ausdorfJ space if and only if for distinct points x and y
in X there are disjoint neighborhoods of x and y in X.
Lemma 1. A toplological space is a Hausdorff space if and only if
every net has at most one limit.
A net {xJ in a topological space X has a cluster point x in X if for
every neighborhood N of x and JlEA there is a A~Jl for which x;..EN.
A closed subset A of X contains the cluster points of all nets in A. A
subset A of X is compact if and only if each net in A has a cluster point
in A. A is sequentially compact if and only if every sequence in A has a
cluster point; in other words if and only if each sequence in A has a
subsequence which converges to a point of X. A is conditionally compact
if and only if A is compact.
I. Linear Topological Spaces 3
Let cP always denote the field of real (or complex) numbers IR (or C
respectively) and let X be a linear space over CPo Elements of cP are
called scalars. A finite sequence Xl' ... , Xn in X is linearly independent
if and only if L O:iXi=O, with O:i in CP, implies 0: 1 =0:2 = ... O:n=O.A subset
i~n
B of X is a Hamel basis for X if and only if every X in X has the unique

representation X = I O:iXi with O:i in cP and Xi in B, where n is an arbitrary
i~n
but finite integer which may depend on X. A finite number of elements

in B is necessarily linearly independent. A Hamel basis for X always
exists and all Hamel bases have the same cardinal number. This cardinal
number defines the dimension of X. If there is a finite Hamel basis for
X, X is said to be finite dimensional and the Hamel basis is simply
called a basis for X. The set of all finite linear combinations of points
in a subset A of X is denoted by sp A, the span of A, and the closure of
this linear subspace of X is denoted by sp A. If Y is a linear subspace
of X, the factor space XjY is the set of all sets of the form
X+ Y( = {x+ ylYE Y}) with X in X. The algebraic operations in XjY
are defined by the equations
(x+ Y)+(y+ Y)=(x+y)+ Y, X,YEX,
o:(X+ Y)=o:x+ Y, O:ECP,XEX.
With the operations thus defined, the class X j Y is a linear space.

Moreover, an isomorphism of X into another linear space Y is a one-
to-one map f of X into Y such that f(0:1x+0:2x2)=0:1f(x1)+0:2f(x2),
0: 1 ,0: 2 E CP, Xl' X 2 EX. X and Yare said to be isomorphic if and only if
there exists an isomorphism of X onto Y.
A linear topological space is a linear space X with a topology such
that addition and scalar multiplication are continuous simultaneously
in both variables. Consequently, for each X in X, each open (closed) set
A in X and each nonzero 0: in CP, the sets x+A and o:A are open (closed).
A local base B in X is a base at the point 0; and it is clear that then x + B
is a base at the point x for each x in X. Two linear topological spaces
X and Y over the same field are called topologically isomorphic if there
exists a linear map of X onto Y which is a homeomorphism; if there
exists such a map (which is a continuous isomorphism of X onto Y
with continuous inverse), it is called a topological isomorphism. A set
Y is a linear (topological) subspace of X if and only if Y is both a linear
subspace of the linear space X and a topological subspace of X.
Lemma 2. If X is Hausdorff, then every .finite dimensional linear
subspace of X is closed.
If A is a subset of X, then it is known that sp A is a (closed) linear

subspace of X. A is total in X if and only if sp A =X. A is bounded if
and only if for each neighborhood N of 0 there is a positive real number
t such that ActN.
Lemma 3. A compact subset of a linear topological space X, and
hence also a convergent sequence, is bounded.
A subset A of X is said to be symmetric if - x is in A whenever x is.
A is said to be convex if x,YEA always implies tx+(1-t)YEA for
every real number t in the interval [0,1]' A is said to be circled if IXAcA
for every IX in lP with IIXI::;;; 1 and A is said to be absorbing if for any x
in X there exists an e > 0 such that ex is in A. If the collection of convex
neighborhoods of 0 is a base at the point 0 (i. e. a local base in X), then
the topology of X is said to be locally convex and X is called a locally
convex space. A locally convex (linear topological) space is called
barrelled or a barrel space if each barrel is a neighborhood of 0, where
a barrel is a closed convex circled absorbing set in X.
Lemma 4. Each neighborhood of 0 in a linear topogical space X
contains a circled absorbing neighborhood of 0 in X.
A metric on a set X is a non-negative function p, defined for each
pair of points x,y in X, subject to the conditions (i) p(x,y)=O if and
only if x=O, (ii) p(x,y)=p(y,x), and (iii) p(x,y)::;;;p(x,z)+p(Z,y),ZEX
(triangle inequality). X, endowed with the metric p, is called a metric
space. In a metric space X the set of all points {yld(x,y)<e} is called
the open ball of radius e about x. The metric topology for X is the weakest
topology whose open sets contain the open balls of X.
Lemma 5. The set of all open balls about xforms a base for the metric
topology at the point x in X.
Lemma 6. A metric space is a Hausdorff space.
If lP is the field of real or complex numbers and if p is defined by
p(IX,P) = 11X-/3I, IX,PElP, then p is a metric on lP and the topology induced
by this metric is called the usual topology (for lP).
Theorem 7. Everyone-dimensional Hausdo1:ff linear topological space
over the field lP is topologically isomorphic with lP in its usual topology. If
O#Xl EX, a topological isomorphism of lP onto X is defined by TIX=IXX 1 ,
IXElP.
In a metric space X a sequence {x n} converges to x in X if and only
if limd(xn,x)=O. {x n} is a Cauchy sequence if and only if limd(xm,xn)=O,
n m,n
i. e. if for every e > 0 there is an n. such that d(xm' xn) < e whenever
m,n~ n•. X is said to be complete if every Cauchy sequence is convergent
I. Linear Topological Spaces 5
in X. If X is complete, then it is easy to see that every net {x;J in X such

that limd(x",x/l)=O (a Cauchy net) is convergent in X.
",/l
Theorem 8. If X is a complete linear metric space under each of two
metrics, and if one of the corresponding topologies is weaker than the
other, then the two topologies are equal.
°
A subset A of a metric space is said to be totally bounded if for every
e > there is a finite set N of points in A such that for any x in A there
is a y in N for which d(x, y) < a (i. e. there is a finite a-net in A).
Lemma 9. A subset A of a metric space is compact if and only if it is
closed and sequentially compact. Moreover, A is compact if and only if
A is complete and A is totally bounded.
Let X be a linear space. A metric P on X is said to be invariant if
p(x,y)=p(x- y,O), X,YEX. Let p be an invariant metric on X with the
additional properties (i) limp(lXnx,O)=O whenever {lXn} is a sequence
in tP, x is in X and lim IXn =
n
n
° in the usual topology for tP, and (ii)
lim p(IXXmO)=O whenever IX is in tP, {x n} is a sequence in X and
n
lim p(xn,O)=O. Then the real number Ilxll, defined by Ilxll = p(x,O),
n
defines the quasinorm of an element x in X and X endowed with this
quasi-norm is a quasi-normed linear space. It follows easily that Ilxll =0
if and only if x=O, Ilx+ yll ~ Ilxll + Ilyll and II-xii = Ilxll. An F-space
is a complete quasi-normed linear space.
Theorem 10. An F-space is a linear topological space.
If the quasi-norm in a quasi-normed linear space X satisfies
IllXxll = 11X111xll, IXEtP, XEX, it is called a norm and X is called a normed
linear space. The closed set U={xEXlllxll~l} is called the unit ball
ofX.
Theorem 11. A normed linear space X is separable if and only if there
exists a total sequence in X.
Theorem 12. A bounded closed subset of a normed linear space X is
compact if and only if X is finite dimensional.
If Y is a closed linear subspace of the normed linear space X and if
we introduce the norm Ilx+ YII=inf{llyIIIYEx+ Y} for each element
x + Y of the factor space X / Y we have
Theorem 13. X / Y is a normed linear space.
A Banach space is a complete normed linear space. The metric
topology induced by the norm in a Banach space X is called the norm
2 Springer Tracts, Vol. 18 - Marti

6 1. Linear Transformations
°
or strong topology for X. Since Iltx+(l-t)yll~tllxll+(1-t)llyll, X,YEX
and tE [0, 1], the open balls about are convex so that, by Lemma 5,
each Banach space X is obviously locally convex.
Theorem 14. 1f a linear space X can be made into a Banach space by
two dUJerent choices of a norm, Ilxll and Ilxll' and if one of them defines a
weaker topology than the other, then there exist constants M 1 and M 2
such that O<M 1 ~ Ilxll'/llxll ~M 2 < CXJ for all x#O in X. (i.e. II II and 1111'
are equivalent) .
A (real or complex) Hilbert space is a linear space H over the field cP
(of real or complex numbers) together with a CP-valued function L .),
called inner product, defined for each pair of points x,y in H. This function
satisfies the following conditions:
(i) (x,x»O, xEH; (x,x)=O ifand only if x=O;
(ii) (x+ y,z)=(x,z)+(y,z), x,y,zEH;
(iii) (ax,y) = a(x, y), aE CP, X,YE H;
(iv) (x,y)=(y,x), x,YEH, and
(v) H is complete with respect to the metric defined by the norm
Ilxll = (x,x)±, XE H.
Theorem 15. H is a Banach space and for any x,y in H we have
l(x,y)1 ~ Ilxllllyll (Schwarz's inequality).
The orthocomplement A1- ofa set A in H is the set A 1- = {xEHI(x,A)= O}.
If A and A' are closed linear subspaces of H such that A n A' =
A + A' = H we write H = A E8 A' and H is called the direct sum of A
and °
and A'. Let AH denote (A1-)1-.
Lemma 16. If A is a closed linear subspace of H, then A1- is also a
closed linear subspace of H and we have H =A E8 A1-. Furthermore AH=A.
Let (5/1V be the Kronecker symbol, defined by (5/1v=O, f.1#v; (5/1/1= 1,
where f.1 and v are elements of an arbitrary index set. A sequence {xJ
in H is said to be orthonormal if and only if (x i,x)=(5ij' If {xJ is
such an orthonormal sequence, then Bessel's inequality applies, i. e.
w
l: l(x,xJI 2 ~ IIxl12 for every x in H.
i~ 1
A special example of a (finite dimensional) Hilbert space is the
Euclidean space En, denoted by [Rn if its field cP is [R and by C" if cP = C,
t?e li~ear sp~ce of all n-tuples a = {al>"" ani
of n.umbers in [R (re~pec
tIvely m C) wIth norm (or length) Iiall = (a, a)2 of a m En, where the mner
n
product is defined by (a,/3)= l: aiPi' a,/3EE n. A set {/31, ... ,/3n} of n
i~ 1
2. Linear Transformations 7
orthonormal vectors in En is called an orthonormal basis for En and it is

a well known fact that such a basis always exists.
Theorem 17. Each n-dimensional real (complex) Banach space is
topologically isomorphic to [Rn (to en respectively).
A non-empty subalgebra Y of an algebra X is called a left ideal of
X provided that x yE Y for all x in X and all y in Y. It is a right ideal if
the latter condition is replaced by y x E Y for all x in X and y in Y. If Y is
both a left and a right ideal, then it is called a two-sided ideal. If Y i= X,
Y is called proper. Y is maximal if it is proper and is not properly con-
tained in any proper ideal of the same type. An algebra X is commutative
if and only if x y = y x, x, yE X. In this case every ideal is two-sided.
The radical in a commutative algebra X is the intersection of all the maxi-
mal ideals in X. X is semi-simple if its radical consists of only the vector 0
in X. A B-algebra X is a complex Banach space which is also an algebra
over the field C, with unit e such that Ilell = 1 and Ilx yll < Ilxllllyll,
X,YEX. An example of a B-algebra is (the space) B(Y) (cf. section 2),
where Y is a Banach space.
Theorem 18. Any semi-simple commutative B-algebra has a unique
norm topology.
A B-algebra X is (algebraically) isomorphic to a B-algebra Y if and
only if there is an isomorphism T of the linear space X onto the linear
space Y such that T(x x') = T(x) T(x'), x, x' E X. If one does not request
that T is one-to-one, T is called a homomorphism of the algebra X onto
the algebra Y. X and Yare said to be topologically isomorphic B-algebras
if T is also a homeomorphism of the Banach space X onto the Banach
space Y. If T maps the identity of X onto the identity of Y then T is said
to be proper.
Theorem 19. If X is a commutative B-algebra which is isomorphic to
a semi-simple commutative B-algebra Y, then X and Yare topologically
isomorphic.
2. Linear Transformations
Let X and Y be linear topological spaces over the same scalar field tP.
A transformation T: X ---> Y is said to be additive if for all x 1 and x 2 in
X we have T(Xl +x z)= TXl + Tx z . An additive transformation T:X ---> Y
is said to be linear if T(rxx)=rxTx, rxEtP, XEX. The null-space of T,
denoted by T- l (0), is the set {xEXI T(x) = O}. The linear transformation
T is continuous at the point x in X if for every neighborhood V of Tx
there corresponds a neighborhood U of x with T(U)c V. T is continuous

if it is continuous at each point x of X.
Lemma 1. If T is continuous anywhere, then it is continuous.
Lemma 2. T is continuous if and only if lim TXA = T(limxA) for every
convergent net {x A} in X . A A
Lemma 3. If X, Y and Z are linear topological spaces and if both
T 1 : X -> Yand Tz : Y -> Z are continuous, then the product T z Tl : X -> Z
is also continuous.
A family {1)J of linear transformations of X into Y is equicontinuous
if and only if for each neighborhood V of 0 in Y there is a neighborhood
U of 0 in X such that TA (U) C V for each member 1). in {TA}.
Theorem 4. (Barrel theorem) Let {1)J be a family of continuous
linear transformations of a barrel space to a locally convex space such
that the set {TAX} is bounded for each x in X. Then the family is equicon-
tinuous.
A linear transformation T: X' -> Y is an extension of a linear trans-
formation T:X->Y, if XcX' and Tx=Tx, xEX. Conversely, Tis
said to be the restriction of T. The linear transformation I: X -> X,
defined by I x = x, X E X is called the identity of X. If X is a metric space
with metric p and if Y is a metric space with metric p', then the linear
map T:X -> Y is said to be uniformly continuous if for every e>O there
corresponds a <5 >0 such that p(x,x')<<5, x,x' EX, implies p'(Tx, Tx')<e.
Applied to this situation one has, if Y is complete,
Theorem 5. If a set D is dense in X and T: D-> Y is uniformly continuous
on D, then T has a unique extension T: X -> Y and T is uniformly contin-
uous on X.
Theorem 6. A continuous linear one-to-one transformation of one
complete linear metric space onto another has a continuous linear inverse.
Theorem 7. A linear transformation of an F-space into another is
continuous if and only !f it maps bounded sets into bounded sets.
Theorem 8. Let {TA} be a net of continuous linear transformations of
an F-space X into another F-space Y. If li~ T;.x exists for all x in a
total subset S of X, and if {TAX} is a bounded set in Y for each x in X,
then there exists a continuous linear transformation T: X -> Y such that
Tx = lim TAX, XEX.
A
Let T:D-> Y be a linear transformation into an F-space Y, with

domain D in an F-space X. Tis said to be closed if, whenever xnED,
2. Linear Transformations 9
xn-x, Txn-y, it always follows that xED and Tx= y. It is clear that
T is closed whenever D is closed and T is continuous.
Theorem 9. If T:D- Y is closed and if T- 1 exists, then T- 1 is also
closed.
Theorem 10. (Closed graph theorem) If T is closed and D=X,
then T is continuous.
Lemma 11. Let T: D- Y be a linear transformation with domain D in
a normed linear space X into another, Y. Then T is continuous if and only
if T has a finite norm II Til, defined by II Til = sup{11 Txll Illxll:':; 1, XE D}.
In the situation described in the above lemma, T is said to be bounded.
Theorem 12. If T is a bounded linear transformation on Dc X to Y,
then T has a unique bounded linear extension T' on jj and we have
IIT'II=IITII·
An isometric isomorphism of a Banach space X onto another, Y, is a
topological isomorphism of X onto Y for which II Txll = Ilxll, XE X.
Theorem 13. If T is a bounded linear transformation of a Banach
space X onto another, Y, then Y is topologically isomorphic to the factor
space XIT-l(O).
The linear space B(X, Y) (B(X)) of all bounded linear transformations
of a Banach space X into a Banach space Y (into X respectively) with the
definition of the norm which has preceded, is a Banach space. The topo-
logy in B(X, Y) defined by this norm is called the uniform operator topology.
An element of B(X) is called an endomorphism of X. Next, we have as a
consequence of Theorem 8:
Theorem 14. (BANACH-STEINHAUS) Let {Tn} be a sequence of opera-

tors in B(X, Y). If lim
n
Tnx exists for all x in a total subset S of X, and if
supll T"xll < 00 for all x in X, then there exists a T in B(X, Y) such that
n
Tx = lim T"x,
n
XE X and one has II Til:.:; supll
n
T"II < 00.
Theorem 15. A linear transformation T:D- Y with domain D in X

has a bounded inverse if and only if there is a constant M > 0 such that
IITxl1 ~Mllxll for all x in D. In this case IIT-llI:.:; 11M.
An operator in B(X, Y) is said to be compact if it maps the unit ball
of X onto a conditionally compact subset of Y.
Theorem 16.Let Tl EB(X, Y) and T2EB(Y,Z). If one of the operators

Tl or T2 is compact, then the product T2 Tl in B(X,Z) is also compact.
Theorem 17. The set of compact operators in B(X, Y) is closed in the

uniform operator topology of B(X, Y).
A linear transformation P:X -.X is called a projection in X (or of

X on P(X), or of X onto P(X)) if pZ = P. We call attention to the fact
that the definition of P does not necessarily imply that P is in B(X). In
this connection, we must know that X is the direct sum M EB N of two
linear subspaces M and N if and only if M + N = X and M n N = 0, in
other words, if and only if every element x in X can be written uniquely
as a sum X=X I +x z , where Xl is in M and X z is in N.
Theorem 18. If P is a projection in X, then X = P(X) EB (I - P) (X).
Conversely, if M and N are (closed) linear subspaces of X such that
X=MEBN, and if P:X-.X is defined by Px=x l , XEX, where Xl is
the unique component of X in M, then P is a (continuous) projection of X
onto M.
Theorem 19. If M is a linear subspace of X, there exists a projection
of X on M.
We observe that there may be more than one projection of X onto

M. Next, let X be a complex Banach space. The spectrum (J(A) of an
operator A E B(X) is the complement in C of the set of all points A of C
for which (AI_A)-l exists and is in B(X). R(A,A)=(AI-A)-l iscalled
the resolvent of A.
Theorem 20. (J(A) is a non-empty compact subset of {AIIAI::::; IIAII, )oEI[:}.
If AE(J(A) is such that AI-A is not o_ne-to-one, then A is called an
eigenvalue of A, in this case there exists an x#O such that Ax=),x.
3. Conjugate Spaces and Weak Topologies
Let X be a linear topological space over the field CPo The conjugate
space X* of X is the set of all continuous linear functions of X into tP
and the elements of X* are called continuous linear functionals. A subset
A of X* is said to be total over X if for any X in X, x*(x}=O for all x*
in A implies that X = O.
Theorem 1. If X is locally convex and HausdOl:ff, then X* is total over X.
The weak topology for a linear topological space X is the topology
for X obtained by taking as a base at the points x of X the neighborhoods
N(x,T,E)={yilx*(y-x)1 <(;,X*Ef},
3. Conjugate Spaces and Weak Topologies II
r
where 1'>0 and is a finite subset of X*. If X is a Banach space, the
weak topology for X is, of course, weaker than the initial (i. e. norm)
topology for X.
Theorem 2. The weak topology for X is locally convex. It is Hausd0l11
whenever the topology for X is locally convex and H ausdorfJ
Theorem 3. x* belongs to X* if and only if x* is continuous in the
weak topology for X.
A set X is said to be weakly closed (weakly bounded) if and only if it
is closed (bounded) in the weak topology for X.
Theorem 4. If X is locally convex, then a convex subset of X is closed
if' and only if it is weakly closed, and a subset of X is bounded if and only
if' it is weakly bounded.
Theorem 5. Let X be defined as before and let Y be a lillear subspace
of x. Then to every continuous linear functional f on Y there exists an
X*EX* such that x*(y)=f(y),YEY.
Theorem 6. Let X be locally convex, let Y be a closed linear suhspace
of X and suppose that there exists an x in X which is not in Y. Then there
exists an x* in X* such that x*(y)=O, yE Y and x*(x)= 1.
Let now X be a Banach space over the field cP and let X*
be the conjugate space of X endowed with the norm defined by
Ilx*11 =sup{lx*(x)lillxll::::::; I}, X*EX*. The set Y~= {x*EX*lx*(Y)=O}
is called the orthogonal complement of the linear subspace Y of X.
Lemma 7. X* is a Banach space.
Theorem 8. Let Y be a closed linear subspace of X. Then y* is iso-

metrically isomorphic to the factor space X*/Y~.
It is clear that X* itself possesses a conjugate (Banach) space, X**,
the second conjugate space and we denote the sequence of spaces obtained
by continued conjugation by X, X*,X**,X***, .... We have the following
important theorem on the existence of extensions of continuous linear
functionals:
Theorem 9. (HAHN-BANACH) rf' Y is a linear subspace of X, then
to every y* in y* there exists an x* in X* such that x*(y)= y*(y), yE Y
and Ilx*11 = Ily*ll·
The next theorem shows that for every x =J y in X there is at least
one x* in X* such that x*(x)=Jx*(y). In other words there are enough
elements in X* to distinguish between points of X:
Theorem 10. To each x =f. 0 in X there exists an x* in X* with Ilx*11 = 1

and x*(x)= IIxli. Therefore, Ilxil = sup{lx*(x)li IIx*II ~ 1} for every x in X.
A determining manifold for X is a closed linear subspace r of X* such
that IIxil =sup{lx*(x)li IIx*II ~ 1, X*Er}. Obviously, X* itself is a deter-
mining manifold for X.
Theorem 11. If X* is separable, so is X.
However, the conjugate space of a separable Banach space X need
not be separable. An example is X = 11 (cf. section 1.4 b) where It is iso-
metrically isomorphic to the non-separable space 100 • The linear trans-
formation J:X-+X** defined by Jx(x*)=x*(x), X*EX* is called the
natural embedding of X into X**.
Theorem 12. J is an isometric isomorphism of X into X**.
Let A be an arbitrary index set. Then the following two theorems
are consequences of the principle of uniform boundedness.
Theorem 13. Let {X.lcIAEA} be an indexed set in X. Then
sup {IIx;.II i},EA} <00, whenever sup {lx*(x.lc)liAEA} <00 for allx*in X*.
Theorem 14. Let {T.lcIAEA} be a family of bounded linear trans-
formations of X into a Banach space Y. Then sup {II TAlI iAEA} < 00 if
and only if sup {ly*(T.lcx)liAEA} < 00, XEX, y*E Y*.
A net {X.lc} in X is said to be weakly convergent if lim x*(x.lc) exists
;.
for each x* in X*, it is said weakly convergent to a point x in X if
lirnx*(x.lc)=x*(x) for every x* in X*. {X.lc} converges to x in the weak
A
topology if and only if it converges weakly to x. A subset A of X is said

to be (conditionally) weakly sequentially complete if every weakly
convergent sequence in A converges weakly (i.e. in the weak topology
for X) to an element (of X) of A.
Theorem 15. If {xn} is a weakly convergent sequence in X, then
sup IIx"II < 00. If it converges weakly to x in X, then XE sP {x,,} and
"
IIxil ~ sup IIx"II.
"
We remember that the weak topology for X is weaker than the norm
topology for X (defined by the neighborhoods N(x,<:) = {yi IIy- xII < <:}
of the points x in X). It is known that the weak topology for X is a
metric topology if and only if X* is separable.
Theorem 16. Let S be a total set in X* and let {x,,} be a sequence in X
such that supllx"lI<oo and for some x in X, limy*(xn)=y*(x),Y*ES.
n
Ihen {xn} converges to x in the weak topology for
"
x.
3. Conjugate Spaces and Weak Topologies 13
X is reflexive if the natural embedding J: X ~ X** is onto. Finite

dimensional vector spaces are examples of reflexive spaces.
Theorem 17. If X is reflexive, it is weakly sequentially complete.
Theorem 18. The following statements are equivalent:
(i) X is reflexive.
(ii) X is topologically isomorphic to a reflexive Banach space.
(iii) X* is reflexive.
(iv) The unit ball V of X is sequentially compact in the weak topology for X.
(v) Every closed linear subspace of X is reflexive.
(vi) Each separable closed linear subspace of X is reflexive.
Theorem 19. If X* is separable and the unit ball V of X is weakly
sequentially complete, then X is reflexive.
The weak* topology for the conjugate space X* can be introduced by

taking as a base at the points x* of X* the neighborhoods
N(x*,r,G)={y*il(y*-x*)(x)I<G, XEr},
where G > 0 and r is a finite subset of X. A net {xi} in X* converges

to x* in the weak* topology for X* if and only if li~xt(x)=x*(x) for
all x in X. The weak* topology for X* is weaker than the weak topology
for X*. The weak* topology for X* is a locally convex topology. It is
a metric topology on bounded sets if and only if X is separable.
Theorem 20. (BANACH) A linear functional f on X* is of the form
f(x*)=x*(x),x*EX*, with a certain x in X depending on!, if and only
if f is continuous in the weak* topology for X*.
Theorem 21. (ALAOGLU) The unit ball V* of X* is compact in the
weak* topology for X*. ff X is separable the compactness of V* is sequen-
tial.
Theorem 22. Let J be the natural embedding of X into X** and let
V, V** be the unit balls in X, X** respectively. Then J(V) is weakly*
dense in V** and J(X) is weakly* dense in X**.
The adjoint of a bounde~ linear operator T of X into a Banach
space Y is the mapping T*: Y*~X*, defined by T*y*(x)=y*(Tx),
y*E Y*, XEX. Of course, T* is linear.
Theorem 23.11 T*II = II Til and 1* is the identity in B(X*).
Theorem 24. If T is onto, then T* is a topological isomorphism of
y* onto X*nT-l(O)-L.
14 L Linear Transformations
Theorem 25. If T is a topological isomorphism of X onto Y, then T* is

a topological isomorphism of y* onto X*. In this case one has (T- 1)*
=(T*)-l.
Theorem 26. If S is in B(X, Y), and T is in B(Y,Z), where X, Y and Z

are all Banach spaces, then (S T)* = T* S*.
Theorem 27. T* is compact if and only if T is compact.
Theorem 28. If P is a continuous projection in X, then P* is likewise
a continuous projection in X* and P* (X*) = (1- P)(X)\ (1* - P*)(X*)
=P(X)-L.
Let H be a Hilbert space and T an operator in B(H). There is a unique

operator T* in B(H), called the Hilbert space adjoint of T, which is
defined by (Tx,y)=(x, T* y), x,YEH. T is called self-adjoint if T* = T.
Theorem 29. Let P of- 0 be a projection in H. Then (1 - P)(H) = P(H)-L
if and only if IIPII =
1.
A projection with the properties written in the foregoing theorem is

said to be orthogonal.
Theorem 30. A projection in H is orthogonal if and only if it is self-
adjoint.
4. Special Banach Spaces
At this place we write down some properties of a few special Banach

spaces and some theorems on these spaces which are of interest in the
theory and application of bases.
Let s be the linear space (over the field cP of real or complex numbers)
of all sequences rx = {rx n } in CP.
a) The set c in s of all convergent sequences, endowed with the norm
Ilrxll = sup Irxnl is a separable Banach space. Co is the closed linear subspace
n
of c of all sequences converging to zero. Neither c nor Co are weakly
sequentially complete and both, c and co, are not reflexive. c* and c6
are both isometrically isomorphic to the space 11 which is described
under b).
Theorem 1. (PHILLIPS) If J is the natural embedding of Co into C6*
and if {xn} is a sequence in C6** which converges to zero in the weak*
topology for c6**, then lim IIJ* xnll = o.
n
4. Special Banach Spaces 15
b) The set lp, 1 ~ p < OCJ of all elements a of s with finite norm
00 ~l/P
lIall =
(
n~l lanl) is a separable weakly sequentially complete Banach
space. 11 is not reflexive. If l<p<OCJ and l/p+l/q=l, Ip is reflexive
and I; is isometrically isomorphic with Iq, where the corresponding
00
isometric isomorphism T is given by T f3(a) = L ai f3i' aE Ip, f3 Elq.

i= 1
Taking aE Co and f3 Ell in the preceding definition, T defines an iso-
metric isomorphism of 11 onto c~. It is not misleading, therefore, to
identify sometimes c~ with 11 or l; with Iq •
For p= 1 we have the following important theorem:
Theorem 2. In 11 weak and strong convergence of sequences are the

same.
In the special case p=2, Ip is a Hilbert space (with inner product
00
(a,f3) = L an/jn)'
n=l
c) The set 100 of all bounded sequences a in IP with norm lIall = sup lanl
n
is a Banach space which is neither separable nor weakly sequentially
complete. It is easy to see that the set of all characteristic functions of
the subsets of the set of positive integers forms a total set in 100 , The
00
mapping T: 100 ~I'L defined by T f3(a) = L anf3n' aE ll' f3E 100 , is an
n= 1
isometric isomorphism of 100 onto It. It is known that weak and weak*
convergence of sequences are equivalent in I!,. Furthermore, every
bounded linear transformation of 100 into a weakly sequentially complete
Banach space sends weak Cauchy sequences into strongly convergent
sequences.
Theorem 3. Every separable Banach space is isometrically isomorphic
to a closed linear subspace of 100 ,
d) Let S be a compact metric space and let X be a Banach space.
The linear space C(S, X) over the field IP, of all continuous ( = uniformly
continuous) vector valued functions f:S~X for which the norm
IIfll = sup {lIf(s)lIisES} is finite is a separable Banach space, where
IIf(s)1I is the norm of f(s) as an element of X. If X = IP we simply
write C(S). We recall that a set E in C(S,X) is equicontinuous if and
only if to every 8> 0 and every s in S there is a neighborhood N of s
in S (which may depend on s) such that sup {lIf(s)-f(t)lIitEN} <8
for all f in E.
Theorem 4. Let {f,,} be an equicontinuous sequence in C(S,X). If fn

converges pointwise (i. e. for each point s in S) in X to a function f in
C(S, X), then fn converges uniformly (on S) to f.
Theorem 5. (STONE-WEIERSTRASS) Let A be a closed subalgebra of
C(S) which contains the characteristic function XS of S (which is the unit
in C(S)) and which contains A=U(-)lfEA}, if IP = 1[:. Moreover, if for
every pair s, t of distinct points in S there is an f in A with f(s)oIf(t),
then A is dense in C(S).
As a special instance we take S = [0, lJ and X = IP and we write
for the corresponding space C[O,l]. It is known that C[O,l] is not
weakly sequentially complete. However C* [0, l] is weakly sequentially
complete, but not separable.
e) Let S be a compact interval in IR and let p be a real number with
1 ~ p < 00. The linear space Lp(S) is the set of all equivalence classes
of functions f:S->IP which are measurable and p th-power integrable
on S. Lp(S) is a separable and weakly sequentially complete Banach
space with norm Ilfllp=[jlf(sWdsrp. Hence, iff and g are in Lp(S),
then the sum f+g is in Lp(S) and Ilf+gllp~ Ilfllp+ Ilgllp which is
M inkowski's inequality.
Theorem 6. If 1 < p < 00 and lip + llq = 1, Lp(S) is reflexive and L';(S)
is isometrically isomorphic with Lq(S); the corresponding isometric
isomorphism T:Lq(S)->L';(S) .is given by Tg(f) = S f(s)g(s)ds,fELp(S),
gELiS). S
Furthermore, if f is in Lp(S) and g is in Lq(S), then the product

fg is in LdS) and Ilfglll ~ II flip Ilgllq which is Holder's inequality.
L 2 (S) is a Hilbert space (with inner product (f,g) = ff(s)g[s)ds,
S
f,gEL 2 (S)). Very useful is the following theorem on the possibility of
interchanging the order of integration in a double integral.
Theorem 7. (FUBINI-ToNELLI) Let Sand T be compact intervals in
IR and let f (s, t) be a IP-valued measurable function on S x T such that
f (s,·) is in LI (T) almost everywhere in S and such that Sf (s, t) d t is in
LI(S), Then T
I (~f(s,t)dt)ds = Hif(S,t)ds)dt.
Theorem 8. For 1 ~ p < 00 the subset C(S) of Lp(S) is dense in Lp(S).

Theorem 9. (Lebesgue dominated convergence theorem) If g is in
LI (S) and if Un} is a sequence in Ll (S) such that Ij~(s)1 ~ Ig(s)1 almost
4. Special Banach Spaces 17
everywhere in S, which converges almost everywhere to a function f on

S. Thenfisin L i (S) and we have limJfn(s)ds=Sf(s)ds.
n s s
Theorem 10. If {fn} is a monotone increasing sequence of non-
negative real measurable functions on S, converging almost everywhere
to ajitnctionJ. Then limSfn(s)ds = Sf(s)ds.
n s s
Next, let {xn} be the trigono~etrical system, given by x o(s)=(2n)-t,

X2n- 1 (S) = n-' sin(n S), X2n(S) = n-' cos (ns), n = 1,2, ... , SE [0,2 n J.
Theorem 11. For each x in L p [0,2n], 1 <p< 00, the series
00 2"
L S xn(s)x(s) d s· Xn converges to x in the norm topology for Lp [0,2 n J.
n=O 0
Corollary 12. The sequence {xn} is total in L p [0,2n], 1 <p< 00.

t) Let D be the open unit disc {zllzl < I} in the complex plane IC.
The linear space A(D) of all complex valued functions f which are
analytic on D and continuous on [j is a Banach space if the norm in
A(D) is defined by Ilfll = sup {If(z)lizED}. Let Zo be an arbitrary point
in D. Then every function f in A(D) has the Taylor series expansion
fez) = I (f(n)(zo)/n!)(z-zot which converges absolutely and uni-

n=O
formly for z in any closed disc {zllz-zol~r} of radius r which is con-
tained in D. The Taylor series expansion is unique since we have
Theorem 13. If two power series I IXn(z-zot and I f3n(z-zot

n=O n=O
converge on a neighborhood of Zo absolutely to the same sum, then IXn = f3n
for all n.
g) Let D be defined as under t) and let 11 be the planar Lebesgue
measure in D. Let A 2 be the linear space of all complex valued functions
on D which are analytic in D and square-integrable with respect to 11.
Defining the inner product (J,g) of f and gin A2 byef, g)= ff(z)g(z)dl1(z),
A 2 becomes a Hilbert space. D
References for Chapter I: BOURBAKI [1], DAY [2], DIEUDONNE [2],

DUNFORD and SCHWARTZ [1], EDWARDS [2], HALMOS [1 and 2], HAUS-
DORFF [1], HILLE and PHILLIPS [1], KELLEY and NAMIOKA [1], KOTHE [2],
RICKART [1], TAYLOR [4], WILANSKY [2], YOSIDA [1] and ZYGMUND [1].
CHAPTER II
Convergence of Series in Banach Spaces

It is essential for understanding the concepts of bases of different
types to have some knowledge about the various definitions of conver-
gence of a series and to stress its hierarchy and its interconnections.
This is done in the first paragraph. There we also present the proof of
the important Orlicz-Pettistheorem which claims the equivalence of
weak and strong unconditional convergence of series in Banach spaces.
The second paragraph contains Riemann's theorem which asserts that
absolute and unconditional convergence of series in finite dimensional
vector spaces are the same, and the famous Dvoretzky-Rogers theorem.
The latter states the existence in every infinite dimensional Banach
space of an unconditional series which is not absolutely convergent, a
fact, which has been conjectured for about twenty years and which has
been settled down by DVORETZKY and ROGERS in 1950.
1. Relations among Different Types of Convergence
Then; are several different definitions for unconditional convergence

of a series in a Banach space X. However, it can be shown that all forms
defined in the following definition are equivalent (see HILDEBRANDT [1],
ORLlCZ [2] and DAY [1]). It is assumed that in each case the series
LXi is convergent in X and we take X= LX;.
Definition 1.
(i) LX; is unconditionally (or reordered) convergent if for every per-
mutatioin p of the integers the series LXp(i) converges in X.
i
(ii) LXi is unordered convergent
.
if lim LXi=x,
GEE.
where L is the set
l lEa
of all finite subsets of the set of integers, directed by ~.

1. Relations among Different Types of Convergence 19
(iii) I Xi is subseries convergent if for every increasing sequence {nJ

i
of integers the series Ix", converges to some element of x.
i
(iv) I Xi is bounded multiplier convergent iffor each bounded sequence

i
{IXJ in flJ, the series I lXi Xi is convergent to some element of X.
i
We note that the convergence in all the preceding definitions is the

strong convergence in X. Before showing the equivalence of (i) to (iv)
we prove the famous Orlicz-Pettis theorem which links sub series con-
vergence of series in the strong and weak topologies of X.
Theorem 2. (ORLICZ-PETTIS) If a series is subseries convergent in
the weak topology of X, then it is subseries convergent in the strong
topology of X.
Proof. Let the series of the theorem be L Xi and let ~ be the set of
;
all increasing sequences fl of positive integers. By assumption there
exists for each fl in ~ an x/1 in X such that
L x*(x;)=x*(x/1)
iEJl
for every x* in X*. Since subseries convergence in flJ is equivalent to

absolute convergence (cf. to Definition 2.1 and to Theorem 2.2 for the
more general case), it follows that L Ix*(x;)1 < 00 for all x* in X*.
i= 1
Now, T:X*-./ 1 , defined by Tx* = {X*(Xi)} is clearly linear. T is closed
since from lim X; = x*, {x;} c X*, and from the existence of lim T X;
n n
in 11 it follows lim Tx; = lim {x; (x;)} = {lim X; (x;)} = {x* (x;)} = Tx*. But
n n n
T, being closed and defined everywhere on X*, must be bounded (1.2.1 0).
Since x/1 is the weak limit of a sequence in X, we have X/1ESP{X;} (1.3.15).
Let now {x;} be a sequence in the closed unit ball U* of X*, and let
{y;} be its restriction on Y = sp {x;}. Since Y is separable, by Theorem
1.3.21, the unit ball in y* is sequentially compact in the weak* topology
of Y*. Thus there exists a subsequence {y;) of {y;}, converging to an
element Y6 in y* in the weak* topology of Y*. As a consequence of the
Hahn-Banach theorem there is a continuous extension X6 of Y6 to the whole
space X. Let X/1 be the characteristic function of fl. Evidently, X/1 is in 100
and the corresponding functional f", defined by j~(IX)= LX/1(i)IX;= L (X;
i iEJl
for every (X = {IX;} in 11' is an element of If. It then follows
20 II. Convergence of Series in Banach Spaces
iEIl
Now by 1.4.c, {XJLI.uE~} is a total set in 100 and since the mapping which
relates XJL and fJL is an isometric isomorphism of 100 onto Ii, UJLI.uE~} is
a total set in Ii. Hence, using the fact that the boundedness of T implies
that of the sequence Tx;, Tx; '''., we conclude from Theorem I.3.16,
that TX;j converges to Tx~ in the weak topology of Ii' Since weak and
strong convergence in 11 is the same (1.4.2) it is clear that T(U*) is com-
pact (1.1.9).
Next, let {1k} be a sequence of operators ef 11 to itself defined by
1k1X={O,,,.,O,lXbfik+l,,,.}~OBviously, 111k11=1 for all k so that {Td is
an equicontinuous set of functions on the compact space T( U*) to 11'
Since lim 1k1X=0 for each IX in T(U*) it follows from Theorem 1.4.4 that
k
liJ?sup{II1kIXIIIIXE T(U*)} =0, hence that
lim I Ix*(xJI =0 (1)

k i=k
uniformly on U*. Finally,
x JL - ~ Xill,,:;;sup f~ IX*(Xi)llx*EU*}
Il
,<k l'''k J
,,:;;supt~k Ix*(xi)11 X*E u*}.
Because the last term converges to zero for k---+oo, IXi is strongly
i
subseries convergent and the proof of the theorem is finished.
Theorem 3. The conditions (i), (ii), (iii) and (iv) of Definition 1 are
equivalent conditions for unconditional convergence of a series Xi in X. I
i
=
Proof. (i) (ii). Let the series be reordered convergent and suppose
that lim I Xi = X is not true. Then there exists an e> 0 such that for
crE1: .
all o"E~ there is a O"'~O" with Ilx- ,I Xiii ~e. Since ~Xi is convergent
'EfT
lEa' I
we can choose an 110 such that Ilx- i~n Xiii <e/2 for l1~no' We define
the sequence 0" 1 ~ 0" 2 ~ ... ~ 0" n ~ ••• in ~ such that the odd numbered sets
O"n contain all the integers less than or equal to max[l1o,sup{iliEO"n}]
and the even numbered sets O"n satisfy Ilx- ,I Xiii ~e. It is now apparent
lElTn
1. Relations among Different Types of Convergence 21
that which
contradicts the hypothesis that LXi is reordered convergent. Conse-

quently, lim L Xi=X,
(fer: .
L Xi = x. Then for every 3> 0 there exists a a

lEU
(ii) => (i). Let lim e such

(fer.
Ilx- tXil1 <3 for all

lEU
that a~ae' Ifp is any permutation of the (positive)

integers, there exists an index no such that ae~{p(l), ... ,p(no)}. Hence
for n?> no we have Ilx - .L xp(illl <3, X
showing that ~XP(il = for every p.
l~n I
(ii) => (iii). If X= lim L Xi exists, then for every 3> 0 there is an
aeE .
element ae in 1: for which Ilx- tXill<3 if a~ae' Let {nJ be any in-
lEU
creasing sequence of integers and let ap=aeU{n1, ... ,n p} for positive

integers p. Then
for every q?> p and p such that np_ l?> sup {iii Ea.}. Therefore tL
l~n
Xn i }
is a Cauchy sequence and the completeness of X implies the convergence

of LX ni '
L Xi
i
(iii) => (ii). Suppose (iii) is true, but that lim does not exist.
ueI: .
lEU
Then there exists an 3>0 and sequences a: and ai: of mutually

disjoint sets in 1: such that i~f ~ Xi - . L_ Xi II.
Thus by ~ Xi
lEak
I ?> 23.
lEak
II.
lEak
I
+ II.L- Xiii?> II.L+ X i - .L_ Xiii ?>23, there is a sequence {ad of mutually
Xiii ?>3. Now,
lEak lEak lEak
disjoint sets in 1: (with ak=a: or ai:) such that il~f t~k

let {Td be a subsequence of {a k} such that sup{iliETd <inf{iliETk+ d
for all k. Then it is clear that a sub series of LXi' L L Xi is not Cauchy.
i k ietk
Since X is complete this contradiction shows that lim L Xi exists and

the limit must obviously be x. <TEl: iEU
(iii) => (iv). Let LXi be subseries convergent. Then by (1) there is
i
each 3> 0 an index n such that for all k?> n,

Given a bounded sequence {IX;} in cP we thus have for every p, q?;;n,
/lit lXiXi/i =SUP{lit lXiX*(Xi)IIIIX*11 ~ 1}

P
~s~p IlXil sup{tIX*(XJIIIIX*11 ~ 1}

<2ssup
, IlXil.
(iv) then follows as an immediate consequence of the completeness
of X.
(iv) => (iii). Suppose LXi is bounded multiplier convergent. Let {xnJ
i
be an arbitrary subsequence of {x;}. To obtain (iii) we have only to take a
multiplying sequence {IXJ, given by IXj = 1 if.i is contained in the set
{n l' n 2 , •. . } and IX j = 0 else. Finally, this completes the proof of the theorem.
The result at the end of the proof (ii) => (i) suggests the following
Corollary 4. If the series LXi is unconditionally convergent then
i
LXp(i)= LXi for every permutation p of the integers.
i
2. Unconditional and Absolute Convergence

00
Definition 1. A series L Xi in a Banach space X is absolutely conver-

i= 1
gent if and only if L Ilxill is convergent.

i= 1
It is clear that every absolutely convergent series is unconditionally
convergent. However, the converse is not true in an infinite dimensional
Banach space as we see in the sequel.
Theorem 2. (RIEMANN) In a finite dimensional Banach space X,
unconditional convergence of a series is equivalent to absolute convergence
of the series.
Proof. Since an n-dimensional Banach space X over the field ~
(or C) is topologically isomorphic to the Euclidean space ~n (respectively
en) (1.1.17), it is sufficient to show that every unconditionally convergent
series L IX; in ~n is absolutely convergent in ~n. Let lXi = {lXki} and let
(L
i= 1
the norm in ~n, as usual, be given by Ilml = lf3kI2)t, f3E~n. If
k:S.n
1 ~ k ~ n we define the increasing sequences (j';: and (j'k such that i E (j'!
when ever IXki ~ 0 respectively. Since I
lXi is subseries convergent
i= 1
2. Unconditional and Absolute Convergence 23
t
~here are co~stants M > 0 for which s~p II. I + CX i I ~ M t, where (J m
ISthe set of mtegers ~ m. Then !E"m""k
vn I I Icxd
n
I Ilcxdl ~
i~m k=l i~m
~ 2 vn I max I
n
lakd
k =1 ± iEamnq~
2 vn I m}xl. I ~ cxkil
n
=
k =1 - IECim(\ak
~ 2 vn ~ m;x I . I CX I
n
0 i
k- 1 IEGmnak
2 vn I max Mt <00,
n
~
k= 1 ±
and the proof is complete.
In order to prove the Dvoretzky-Rogers theorem which states that in
every infinite dimensional Banach space there exists an unconditionally
convergent series which is not absolutely convergent, we prove a geomet-
rical lemma about symmetric convex bodies in the Euclidean space [Rn
(by a body we mean the closure of a bounded open set in [Rn). For the
rest of this chapter we assume X to be infinite dimensional.
Lemma 3. Let V be a symmetric convex body in [Rn. Then there are
points f31"'" f3n on the boundary of V such that Ii; 1 I CX i f3iE V for
i~r
any r~n and any cx#O in [Rn, where lir=llcxll[1+(r(r-l)/n}~J.

Proof. Let u 1 , ... , Un be an orthonormal basis for [Rn and let
cx = I CXiU i' CXE [Rn. We inscribe in Vthe ellipsoid E of maximum volume,
i~n
i.e. the set E={cxlcxE[Rn,(Acx,cx)~l} for which IdetAI- 1 is maximal,

where A is a positive definite symmetric linear operator of [RH. There is
an orthogonal transformation U of [Rn such that the corresponding
matrix of U A U- 1 is diagonal with diagonal elements a 1 , ••. , an> O.
Thus U(E)={CXlcxE[Rn,.I aicx?~l} is an ellipsoid of maximum volume
z:=Sn
in U(V). Next, the linear transformation T of [Rn whose corresponding
matri~ is {at (\J, maps U(E) onto the unit ball B of [Rn and U(V) onto
TU(V) which is again a symmetric convex body, and B is an ellipsoid
of maximum volume in T U(V). Since Ii; 1 I CXif3iE V if and only if
i~r
Ii,:-l I CX i TU f3iE TU(V), it is clearly sufficient to solve the problem with

i::Sr
V replaced by TU(V).
We now suppose that for some i~n there is an orthonormal basis

... , Un for !Rn and i points 131, ... , f3i of contact of T U(V) with B
U 1,
such that with 13k = L f3 kj U j, k= 1, ... , i,
j~n
(i) f3kk"):.O, f3kj=O, j>k,

(ii) L f3fj= 1-f3fk~(k-1)/n.
j<k
Since E has maximum volume in V there is at least one point of contact,
say 131 of B with TU(V). Let first U 1=f31 and let U 2 ""'U n be any
vectors completing an orthonormal basis in !R n• Thus the above con-
ditions are satisfied for i = 1. To show it for all i ~ n we assume it for
i and prove it for i + 1.
For 8>0 we consider the ellipsoid Ee of points in !Rn which satisfy
n
(1)
j$;i j=i+ 1
Since the "volume" ratio of Ee to B is [(1 +8+8 2 )/(1 +8)]i(n-i» 1, we

infer that there exists a point <I.e on the boundary of T U(V) which is
in Ee' Becalfse B is contained in T U(V), <I.e satisfies L <I.;j"):. 1. Together
with (1) this gives j<;n
j$;i j=i+ 1
Due to the compactness of T U(V), there is a sequence of 8'S tending

to zero such that <I.e converges to some point f3i + 1 of contact of the
boundaries of Band T U(V). Dividing (2) by 8 we have in the limit
n
(n-i)Lf3f+1,j-i L f3f+1,j~0. (3)

j~i j=i+ 1
The basic vectors U i + 1, ... , Un' SO far have no influence on the condi-
tions (i) and (ii). So we can choose U i + 1 orthogonal to U l' ... , U i in
sp{u 1,· .. ,U i,f3i+1} and we may complete the set {u 1, ... ,u i+d to anew
orthonormal basis of !Rn • It is clear that the coordinates of f3 j , j ~ i do not
change and that, under this coordinate transformation, the values of
the two sums in (3), as well as that of L f3f+ 1,j are invariant. Using
j~n
this, the evident equations f3i+1,j=0 for j>i+1, and the equation
L f3f+ 1,j= 1, (i) and (ii) follow immediately for i replaced by i + 1,
j~i
hence for all i ~ n (from the fact that the problem is symmetric with
respect to the origin it is clear that f3i may be chosen such that fiii is
non-negative for all i ~ n).
Having shown the existence of points 131'"'' f3H on the boundary

of T U(V), satisfying (i) and (ii), let IX be a vector of [Rr. Then, using
(i), (ii) and Schwarz's inequality, one obtains
.I IXJf3;- U;)II ~ .I IIX;I IIf3;-u;II

I l:::;;r l::::;r
.I IIX;I [~. f3't + (1- f3;;)2Jt

,:::;;r J<l
~ .I IIX;I [I.f3't+ 1 -f3f;Jt

,::::;r J<l
~ I IIX;I [2(i-l)/n]t
i~r
Hence with Ar= II IX II [1 + (r(r-l)/n)t] we get
and the lemma is proved.

Lemma 4. Let c 1, ... , Cr be any positive numbers. Then there exists
points x1, ... ,xr inXwith IIx;l12=c;for i=l, ... ,r and such that
where J1.r is any subset of the numbers 1, ... , r.

Proof. We take n = r(r - 1) and choose a set of n linearly independent
elements Y1, ... ,Yn of X. Then V= {IXIIXE [Rn, t~n IXjYjll ~ I} is a sym-
metric convex body in [Rn, because for IX+,IX-EV and t in [0,1],
I .I tIX! Yj+ .I (1- t)IX.i Yjll ~ [t+ (1- t)] mfx II.I IXJ Yjll ~ 1 (the sym-
~~n J~n }~n
metry is evident). Let 131"'" f3r be r points on the boundary of V such
that (211IXII)-1 I IXjf3j EV for any IX in [Rr, where the existence of the set
j:::;;r
{f3j} is given by the preceding lemma. Let AILr = 2 (~ Cit Defining
lEJlr
x;=Cl I f3ijYj, i~r,

j:::;;n
it is clear that IIxdl 2=Ci. Since }';,l I cl f3i is in V, we finally get
I .I xil12 = II.I cl JI::::;

lEf.Lr lEllr n
f3ijYi112 =}';,II.I [A.;,l.I cl f3i] Yil12 ~A.;,=4.I ci·
) ~n lEJlr J lEJlr
Remark. Before we prove the next theorem we remember that there
ct = 00
00 00
exist sequences {cJ of positive terms such that I Ci< 00, I

and such that i~l (:~~ CiY < 00 for some strictl;~:creasin~:;quence
1
{nbnZ,n3, ... } of integers (used in the proof of the following theorem): Let
00 00
Ci=(i)-2 and let ni =2 i - 1, i:;::,l. Then I Ci=n 2 /6, I cl=l+!+t+···

1 i= 1 i= 1
=00 and i~l (:~~1 ciY ~i~1(2i-lr2(i-1)}r=(1-1/]!2)-1<oo.

Theorem 5. Let {cJ and {nJ be one of the objects defined above.
00
Then there exists an unconditionally convergent series I Xi in X with

IIxi ll 2 =C i for i= 1,2,.... i= 1
Proof. By the preceding lemma we can choose a sequence {xJ in X such

that Ilxil1 2=Ci and II.I,XiI12~4I,Ci' k=1,2, ... , where ()~ is any sub-
lE(J'k lEak
sequence of ()k= {n k + 1, ... ,nk+ I}'

Let () be any increasing sequence of integers. Choosing v so large
that for £ > 0,
00
I (.I c )± <£/2,
k= v lEak
i
we have for any ()pq= {p,p+ 1, ... ,q} with q:;::, p>n v ,
(4 I
00
~ I ci )±
k =v ielTpqncrkna
00
Finally, since X is complete, this shows that the series I Xi is subseries,

i= !
and hence unconditionally convergent. This finishes the proof of the
theorem.
The following main theorem is now an immediate consequence of
Theorem 5.
Theorem 6. (DVORETZKY-RoGERS) The unconditionally convergent
series coincide with the absolutely convergent series in the Banach space
X !f and only if X is finite dimensional.
Proof. By Theorem 5 there is an unconditionally convergent series

D ~ 'l
I Xi such that I Ilxill = I ct = 00, hence which is not absolutely con-

i=! i=! i=!
vergent. On the other hand, if X is finite dimensional, then by Theorem 2
the unconditionally convergent series coincide with the absolutely
convergent series.
References for Chapter II: DAY [2], DVORETZKY and ROGERS [1],
HILDEBRANDT [1] and HILLE and PHILLIPS [1].
CHAPTER III
Bases for Banach Spaces
Throughout this chapter (and the next two) the basic space X will be
a Banach space. According to the three most common used topologies,
there are bases for the strong, the weak and the weak* topologies for
X, whose definitions are given in the first paragraph. It is shown that
every basis for X is a Schauder basis, a basis with continuous linear
coefficient functionals. The next paragraph shows under which conditions
a biorthogonal system is a basis for X, the equivalence of strong and
weak Schauder bases for X and relations between bases for X and bases
for the adjoint space X*. Three paragraphs are devoted to retro-,
shrinking, boundedly complete, unconditional, absolutely convergent
and uniform bases. Some applications of summability methods on the
theory of bases are given in the sixth section and in the last paragraph
bases for the special spaces Co, Ip(1 ~p < (0), C[O,1], Lp[O,1] (1 ~p < (0),
L 2 [O,2n] and A2 are considered.
1. Bases Corresponding to Different Topologies
In this section we shall be concerned with the definitions and some

basic facts pertaining to bases for Banach spaces in the different possible
topologies. Let X be a Banach space over the field cP of real (or complex)
numbers and let X* be its conjugate space. The topologies considered
are the strong topology (= norm topology of X), the weak topology
(= X* topology of X) and, when the Banach space is a conjugate space,
the weak* topology ( = X topology of X*).
Definition 1. A (weak, weak*) basis for a Banach space X over the
field cP is a sequence {xJ in X such that to every element x in X there
corresponds a unique sequence {aJ in cP for which x = lim La i Xi in the
strong (weak, weak* respectively) topology. n i<n
1. Bases Corresponding to Different Topologies 29
The elements in {cx i } naturally depend linearly on x and they are

called the coefficient functionals of the basis {xJ Moreover, that each
CX i is a unique function of x implies that every element in the sequence
{x;} is non-zero.
Definition 2. A (weak, weak*) basis which has (weakly, *-weakly)
continuous coefficient functionals is said to be a (weak, weak*) Schauder
basis.
It will soon be shown that every weak basis for X is a basis for X
and that every basis for X is a Schauder basis for X. However, that
every weak* basis for a conjugate space X* is a weak* Schauder basis
for X* is not the case as will be clear by an example appearing later in
section 7 (Theorem 4).
Theorem 3. Ina Banach space X every basis {x;} for X is a Schauder
basis for X.
Proof. Let Y be the vector space over the field cJj of all sequences
y= {cx;}, CXiEcJj for which li~ ,I CXiX i exists. Hence s~p II ,I CXiXi11 < 00
l:::;n l~n
for each yE Y. Clearly, the function Ilyll on Y, given by Ilyll = s~p L~n cxix;11
defines a norm on Yand supplied with this norm, Y becomes a normed
linear space. Now, let T: Y~X be the linear transformation defined by
x=lim I cx;x;. Since every x in X has a unique expansion of this form,
n i~n
T is one-to-one and onto. Obviously Ilxll = IITYII ~ Ilyll so that Tis
continuous. As we show later Y is complete and thus a Banach space.
Hence T is a topological isomorphism (Theorem I.2.6) and we have
lanlllxnli = 11;~n a;x;- ;"'~-1 cx;x;11 ~211yll ~21IT-lllllxll· From this and
from the uniqueness of the expansion coefficients CX; we conclude that
each an is a continuous linear functional on X.
It remains to show that Y is complete. Let {yp} be a Cauchy se-
quence in Y (each y defining a sequence {cx p ;} c cJj). Since for all i,
Icxp;- aq;lllxdl ~ 2s~p L~n (cx p;- CX q;) Xi II = 211Yp- Yqll, and since cJj is complete,
there is a sequence {a i} c cJj such that lim cx pi = a i. Given 8> 0, there is an

p
index r such that Ilyp- Yrll <8/3 forall p~r.Hence L~n (CXpi-ari)Xill <8/3,
p~r, uniformly for all n. Taking thelimit onp we obtain s~pt~n (cxi-CXr;)Xill
~8/3. Now, since YrE Y, there is an index n~, depending on r, such that
Ili~n CXriXil1 <8/3, m~n~n". Hence for each m,n~n, and m~n we have
30 III. Bases for Banach Spaces
IlitctiXill~lli~}cti-ctr;)Xill+lli~nctriXill<e. Consequently, y={ct;} is an

element of Yand by what has preceded, Y = lim Yp. Y is complete, there-
fore, and the theorem is verified. p
Corollary 4. Every weak basis for X is a weak Schauder basis for X.

Proof. Y is defined similarly as in the proof of the foregoing theorem,
with lim L ct i Xi = X in the weak topology for X. By (1.3.15), again,
n i~n
s~p" i~n ct iXi I < 00. Since it will be apparent that Y is still a Banach space,
the first part of the proof for the theorem applies also in this situation
(where T is, naturally, defined by a weak limit). Thus each ctn is again a
strongly (and hence a weakly (1.3.3)) continuous linear functional on X.
If {Yp} is a Cauchy sequence in Y we have again a sequence
{ctJ in cP such that limctpi=ct i. The continuity of T (by (1.3.15),
p
II Tyll ~ s~p L~n ctiXil1 = Ilyll with y = {ctJ) and the completeness of X imply
that Typ converges, say, to an element Xin X. Given 8> 0, there is an index
r such that Ilyp - yqll <8/3 for all p,q ~ r. Hence s~plli~ (ctpi - ctq;)Xill <8/3
for p,q~r and it follows that s~plli~}cti-ctqi)xill~e/3, q~r. Now,
there is a fixed index q~r for which Ilx- Tyqll <8/3 and for each
X*EX* with Ilx*II~1 there is an H depending on x* and on q, such
G,
that Ix*C~n ctqi xi-TYq)l<e/3, H~HG· Hence for every H~nG we have
Ix* C~n ctiXi-X)1 ~ Ili~ (cti-ctq;)Xill + Ix* C~n ctqiXi- TY q) I+ IITyq-xll < e.
Thus Y is complete, since {ctJ is in Yand is the strong limit of {ctpJ in Y,
and so we are done.
Remark. A generalization of the above theorem to complete metric
linear spaces with translation-invariant metric, given in the last chapter
(Theorem IX.5.2), deserves mention. That is, the local convexity hypo-
thesis in the theorem may be omitted.
2. Biorthogonal Systems
Biorthogonal systems playa central role in the theory of bases, since

every basis for a Banach space X, together with its associated sequence
of coefficient functionals constitutes a biorthogonal system. On the other
hand, provided some conditions are satisfied, a biorthogonal system is
a basis for some closed linear subspace of X.
2. Biorthogonal Systems 31
Definition 1. Let {xJ and {xT} be sequences in X and X* respectively.

{Xi,XT} is called a biorthogonal system for X if x1(x j)=bij. The endo-
morphisms Un of X, defined by Unx= I x1(x)x i for alln and all x in X,
are called expansion operators. ;~n
Evidently, the operators Un are projections of X with the properties

U", Un= U minim,n}'
The following theorem was first proved by Banach, but based on
stronger assumptions as stated here:
Theorem 2. Let {Xi' xT} be a biorthogonal system for X such that
sUPlx*(Unx)1 < 00, XEX, X*EX*. Then {x;} is a basis for SP{Xi} and {xT}
is a basis for sp{x1}.
Proof. By hypothesis suplx*(Unx)1 < 00, XEX, X*EX*. Thus there

n
isaconstant M<oo such that supllUnll~M (1.3.14). Letxbeinsp{xJ

°
n
Then we have for every s> an indexj and an element y)n sp {Xl" .. ,xJ
such that Ilx- y)1 <so But Ilx- Unxll ~ Ilx- yjll + IIYj- Unyjll + II UnYj- Unxll
and UnYj= Yj for each n?- j, in virtue ofthe biorthogonal relations. Hence
Ilx- Unxll ~(1 +M)s for all n?-j so that x=lim Unx=lim I X1(X)Xi'
n n i~n
The coefficients x1(x) for the expansion of x are unique, since

lim I CXiXi=O, CXiEP, i= 1,2, ... , as is easy to see by multiplication with
n i~n
xj and by use of xj (xJ = bij , implies cx j = °
for all j.
On the other hand, let x* be in sp{xn. Since II U;II = II Unll, in just
the same way as above, it follows x*=limU;x*=lim I x*(x;)x1-
n n i~n
Similarly, the x*(xJ are unique and the theorem is proved.

Remark. We observe that Theorem 2 has an obvious generalization
(cf. Corollary IX.S.6) to barrelled topological vector spaces, in that
situation the proof is then based on the Barrel theorem.
Corollary 3. Let {Xi' xT} be a biorthogonal system for X. Then {xJ

is a basis for sp{x;} if and only if 1~supIIUnll<00.
n
Proof. The sufficiency follows immediately from the preceding
theorem. To prove the necessity, let {xJ be a basis for sp{xJ Then,
since x=limUnx, we have Ilxll~supIIUnxll<oo for all x in X, by
n n
(1.3.14) implying 1 ~ supll U nil < 00.
n
Theorem 4. {xJ is a (weak) basis for X if and only if there is a se-

quence {xT} in X* such that {x;,xT} is a biorthogonal system for X and
I x1(x)x i converges strongly (respectively weakly) to x for each x in X.
i~n
Proof. Let {Xi' xi} be a biorthogonal system for X such that

I Xl"(X)X i converges weakly or strongly to X for each x in X. Then it is
i~n
clear that s~p[.I xl"(x)x*(x i)[ < 00, XEX, X*EX* and by (1.3.15),
l~n
sp{xJ=X. Therefore, the foregoing theorem applies and {xJ is a basis

for X. Conversely, suppose that {xJ is a (weak) basis for X. Then there
exists to each x in X a unique sequence {aJ in q, such that I ai Xi
i::f;n
converges (weakly) to x. The uniqueness implies the existence of linear
functionals xl" on X such that xl"(x)=a i for every x in X. From Theorem
1.3 and Corollary 1.4 we know that these functionals are continuous,
hence elements of X*. Moreover, the uniqueness of the sequence {xT(x)}
implies the biorthogonal relations xl"(x)= bij. Thus {Xi' xi} is a biortho-
gonal system for X and the proof of the theorem is complete.
Remark. From now on we conveniently write for a basis both symbols
{xJ, as well as {Xi' xi}. It is practical to use the first one if no reference
is made to the associated sequence of coefficient functionals {xi} and
by use of the second one the introduction and definition of {xi} becomes
unnecessary.
The following corollary is known as weak basis theorem given (with-
out proof) already in Banach's monograph [1,p.238]. In 1959 it has been
established by BESSAGA and PELCZYNSKI [5] for the more general case
of a Frechet space.
Corollary 5. {xJ is a basis for X if and only if it is a weak basis for x.
Proof. Let {Xi} be a weak basis for X. Then, according to the foregoing
theorem, there is a sequence {xi} in X*, biorthogonal to {x;}, and such
that I Xl"(X)X i converges weakly to x for each x in X. As we have shown
i"$;n
in the first part of the proof of the theorem, {x;} then is a basis for X.
Conversely, since strong convergence implies weak convergence: it
follows that a strong basis for X is a weak basis for X and the proof of
the corollary is finished.
Theorem 6. Let {Xi' xi"} be a biorthogonal system for X. If I x* (x;) xl"
i~n
converges to x* in the weak* topology of X* for every x* in .x; *, then {x;}

constitutes a basis for X.
Proof. By hypothesis, lim I x*(xi)X[(x)=x*(x) for each x* in X*

n i~n
and x in X. This implies S~P[.L x*(xi)xl"(x)[ < 00 for each x* in X* and
I$; n
X in X so that, by Theorem 2, {Xi'Xi"} is a basis for sp{x;}. To prove the
theorem we only must show that sp {x;} = X. If we suppose the contrary,
2. Biorthogonal Systems 33
then there is an x in X which is not in sp{xJ In this case there exists an

X* inX* such that x*(x)=l and x*(x;}=O for all i(1.3.6). Consequently,
O=lim L x*(x;)xT(x)=x*(x)= 1 which is the desired contradiction.
n i~n
Theorem 7. If {x;,xT} is a basis for X, then {xT,Jx;} is a weak*

Schauder basis for X*. Conversely, if {xt,xT*} is a weak* Schauder
basisfor X*, then {xT*} is a sequence in J(X) and {J-I xT*,xT} is a basis
forX.
Proof. We recall that J is the natural embedding of X into its second
adjoint X**. Let {x;,xT} be a basis for X. Since Jx(xn=xT(x)=bij,
{xT, J x;} is a biorthogonal system for X* and by (1.3.20) each J x;
is weak* continuous. For each x* in X* and x in X it follows that
l~ [X*-.L Jx;(x*)xT] (x)=li~x* [X-.L xT(x) x;] =0. To prove the
l~n l~n
uniqueness of the coefficients J x;(x*), we take a sequence {IX;} in cP and

assume that lim L IX;XT(X)=O for all x in X. Then with x=x j , we
n i~n
obtain IXj = 0 for all j and the first part of the theorem is proved.
On the other hand, let {xT} be a weak* Schauder basis for
X* with coefficient functionals xi*,xi*, .... Then we have x*(x)
= lim L xT*(x*)xT(x) for every x* in X* and every x in X, and the linear
n i:f:Sn
functionals xT* on X* are continuous in the weak* topology of X*.
This implies (1.3.20) the existence of elements x;, i= 1,2, ... , in X such
that xT*(x*)=x*(x;} for every x* in X*, and hence that XT*EJ(X).
Consequently, x*(x)=lim L x*(x;)xT(x) for each x* in X* and x in X.
11 i~n
In other words, L xT(x)x; converges weakly to x for all x in X. Since
i $; n
by the equation xT(x)=xj*(xn and by uniqueness of the coefficient
sequence {xj*(xT)} belonging to xT, it is clear that xT(x)= bij , it follows
at once from Theorem 4 that {Xi,XT} constitutes a weak basis for X, and
hence (by Corollary 5) a basis for X.
Corollary 8. Let X be a non-separable Banach space. Then X* has no
we{lk* Schauder basis (and no basis either).
Corollary 9. Let X be reflexive. If {xi,xT} is a basis for X, then
{xT,Jx;} is a basis for X*. Conversely, if {xT,xT*} is a basis for X*,
then {xT*} is a sequence in J(X) and {J-I xT*,xT} is a basis for X.
Proof. The first part becomes clear from the fact that for reflexive X,
the weak* and weak topologies for X* are equivalent. This implies that
{xT,Jx;} is a weak basis for X* and by Corollary 5, a basis for X*.
Conversely, if {xT,xT*} is a basis for X*, then it is a weak* Schauder
basis for X* (by (1.3.20) and since strong convergence in X* implies

weak* convergence in X*), and so {J-I xi*,xt} is a basis for X.
Theorem 10. (WILANSKY) Let {Xi,Xt} be a basis for x. Then
s~p II.I x*(xJxi I < 00 and s~p II.I x**(x{)xill < 00 for each x* in X*
l~n l~n
and each x** in X**.

Proof. Corollary 3 shows that s~pl.I x*(xJXi(x)l<oo for all x in
l::%;n
X* and all x in X. As a consequence of the principle of uniform bound-
edness (1.3.14), we then have s~p lIi~n x*(xJxtll < 00. On the other hand,
s~pl.I x**(x{)x*(xJI~s~p"x**"II.I x*(xi)xill <00 for all x** in X**

l~rJ l~n
and all x* in X*. Finally, the proof of the theorem follows by application
of the same principle once more.
Definition 11. A basis {xt} for X* is called a retro-basis if its biortho-
gonal sequence {xi*} is contained in leX).
It is apparent that each retro-basis for X* is a weak* Schauder basis
for X*.
Theorem 12. Let {xi} be a retro-basis for X* with corresponding bior-
thogonal sequence {JxJ Then {Xi' xi} is a basisfor X, and I X**(X{)X i
converges in X if and only if x** is an element of leX). i"""
Proof. By hypothesis, {Xi,Xt} is a biorthogonal system so that since
strong convergence in X* implies weak* convergence in X*, Theorem 6
applies and {xJ is a basis for X. Then it is clear that the series given in the
theorem converges if x** is in leX). On the other hand, let the series con-
verge (to an element y in X) and define Yn= L x**(x{)x i . For every x* in
X*, ly(x*)=li~lYn(x*)=li~ .L x**(X{)X~(;J=x*ili~ .I lXi(X*)Xi)

l~n \ l~n
=x**(x*) and we finally see that x** is in leX).
3. Shrinking and Boundedly Complete Bases

Definition 1. A basis {xJ for X with associated sequence of expansion
operators {UJ is shrinking if and only if
lim sup {Ix*(x- Unx)lillxll ~ I} =0 for each x* in X*.
n
We note that by Corollary 2.3, lim sup {lx*(y)1 iYESP {xn,x ll +l,
II
x n + 2 , .•. ,}, Ilyll ~ I} =0 for each x* in X* is an equivalent condition for

shrinking. Furthermore, an equivalent name for a shrinking basis {xJ
is that of a weakly uniform basis, since it will be immediately clear from
3. Shrinking and Boundedly Complete Bases 35
the definition that the expansion Vnx for x converges in the weak
topology of X, uniformly on the unit ball of X.
Definition 2. A basis {xJ for X is said to be monotone!f IIVnxl1 is
a non-decreasing function of n jar all x in X.
Theorem 3. A basis {xJ for X is monotone if and only if sup IIVnll:( 1.
n
Proof. We assume {xJ to be not monotone. Then there is an n
and an x such that II V n+1 xii < II Vnxll· But this implies II Vnll;:;: II Vn Vn+ 1 xiii
IIVn+lxll=IIVnxII/IIVn+lxll>l. This shows that a basis {xJ with
sup II Vnll:( 1 is monotone. The converse is obvious.
n
Theorem 4. Let {Xi,XT} be a basis for X. Then the following two

statements are equivalent:
(i) {xJ is a shrinking basis for X.
(ii) {xT} is a basis for X* .
Moreover, {xJ is monotone if and only if {xT} is also monotone.
Proof. (i)=(ii). If {Xi,XT} is shrinking, then for all X*EX*,
O=lim sup {Ix*(x- Vnx)llllxll:(
n
I} = lim
n
sup {I(J - Un*)x*(x)llllxll:( I}
= li~ Ilx* - Vn* x*11 = li~ I x* - i~n x*(xi}xi" I = li~ I x* - i~n J xJx*)xi" II·
Since Jxi(xj) = xj(xJ= c)ij' it follows from Theorem 2.4 that {xi",Jx;}
is a basis for X*.
(ii)=(i). Suppose that {xi", xi"*} is a basis for X*. Then
x*(xJ=lim I xj*(x*)xj(xJ=xi"*(x*)
n j~n
for every x* E X* and for all i. Hence xi"* = J Xi and so

0= li~ Ilx* - i~n J xi(x*)xi" II = li~ sup {Ix*(x)- it. X*(XJxi"{X) 1[llxll:( I}
= lim sup {Ix*(x - Vnx)lillxll:( I}
n
which shows that the basis {xJ is shrinking. Since
s~p{lli~n xi"*(x*)xi"lllllx*ll:( I}
= s~p sup {Ii~n J xi(x*)xi"(x)I[llxll:(1, Ilx*11 :( I}
= s~p sup {Ix* C~n xi"{x) Xi) 1111xll:( 1, Ilx*ll:( I}
= s~p sup {lli~n xi"(X)Xill[llxll :( I} = s~p IVnll,

the proof is complete.
Corollary 5. Let {Xi' xT} be a basis for X. If this basis is shrinking,

then {xi",Jx;} is a basis for X*. Conversely, if {xi",xi"*} is a basis
for X*, then xi"* = J Xi for all i.
Corollary 6. A basis {Xi' xT} for X is shrinking if and only if
Sp {xT}=X*.
Proof. The necessity is clear from Corollary 5. Conversely, let
Sp {xT} =X*. From Corollary 2.3 we infer that sup IX*(Unx)1 < 00,
n
XEX, X*EX* so that by Theorem 2.2, {xT} is a basis for X*. This
finally shows that the basis {x;} is shrinking.
Theorem 7. A basis {Xi' xi"} for X is shrinking if and only if a bounded
sequence {Yk} in X converges weakly to zero whenever lim Xi"(Yk) = 0
fori=1,2,.... k
Proof. Without loss of generality we can take {Yk} in X such that
sup Ilhll ~ 1. Let us first assume that the basis is shrinking. Then for
k
each x* =1= 0 in X* there is for every 8> 0 an index n such that
sup {Ix*(x- Unx)lillxll ~ 1} <8/2. Also, by hypothesis, there is an index
ko such that for k;:? k o, IXi"(Yk)1 < 8 (21Ix*11 .L Ilxjl!f 1, i = 1, ... , n. Hence
J~n
IX*(Yk)1 ~ IX*(h- Unh)1 + IX*(UnYk)1

<8/2 + /x* (~n xi"(Yk) Xi) /
~ 8/2 + Ilx*11 L Ixi"(h) Illxi"11 <8
i~n
or, lim X*(Yk)=O for every x* in X*.

k
To show the sufficiency, we suppose that {xi,xT} is not shrinking.
Then there exists an 8> 0, an x* in X*, a sequence {Yk} in X and a
sequence {md of strictly increasing integers such that x* (h) > 8 and
hESp{Xmk+l, ... ,Xmk+,}. Thus limxi"(h)=O for i=1,2, .... But h
k
converges not weakly to zero so that the basis must be shrinking.
Definition 8. A basis {x;} for X is boundedly complete if for each
sequence {IX;} in cP for which s~p II.L
IXiXi11 <00, there exists an x in
X such that x=lim L
(XiXi' ."'n
n i~n
Theorem 9. Let {Xi' xT} be a shrinking basis for X. Then {xT} is a

boundedly complete basis for X*.
Proof. Let {IXJCCP be a sequence such that sup Ily;ll<oo, where
n
Y; = L IXiXi". According to Corollary 5, {xi",J xJ is a basis for X* and
i$,n
3. Shrinking and Boundedly Complete Bases 37
it is then clear that lim y~(xJ = (Xi' Thus the Banach-Steinhaus theorem
n
applies so that there exists a y* in X* with IIY*II ~ s~p IIY~II < 00, such
that limy~(x)=y*(x) forallxinX.Consequently, y*=lim L y*(x)xj
n m j~m
=Iim L xjlimy~(x)=lim L (XiXt in the strong topology for X*.

m j~m n m i~m
This concludes the proof of the theorem since we have shown that the
basis {xr} is boundedly complete.
Theorem 10. If {Xi,Xr} is a boundedly complete basis for X, then
I x**(xt)X i converges in X for each x** in X**.
i~n
Proof. For each y** in X** there is a sequence {yJ in X for which
y**(x*)=limJYix*)=limx*(y), X*EX*, because by I.3.22, J(X) is
J J
dense in X** in the weak* topology of X**. Moreover, from the weak
convergence of Yj we obtain (I.3.15) that sup Ily)1 < 00. Then
J
sup II I y**(xt)Xi II = sup II lim I xt(y)xill ~sup II Un II' supll yjll< 00 for
n i::;;n n J i~n n J
every y** in X**, where Un is defined in 2.1 and where we have used
Corollary 2.3. This, and the bounded completeness of the basis for
X imply the convergence of I y**(xt)x i in X.
i$;n
Theorem 11. Let {Xi,Xn be a basis for X such that I x**(xt)Xi

i:5;n
converges weakly to an element of X for each x** in X**. Then X is
topologically isomorphic to a conjugate space.
Proof. We define sp {xr}1.= {x**lx**EX**, x**(x*)=O for all x* in
sp {xr}}. Let x** be in sp {xr}1. n J(X). Then for every x* in X*,
x*(rlx**)=lim Ixt(J-lx**)x*(xJ=lim Ix**(xt)x*(xJ=O so that
n i~n Il i~n
x**=O, or sp{xr}1.nJ(X)={O}. Moreover, for each x** in X**
there is an x in X such that lim I x**(xt)x*(xJ=x*(x), X*EX*. We
n i~n
have [x** -J xJ(x*) =x**(x*) - x*(x)= x**(x*)-lim I x**(xt)x*(xi)=O
n i~n
for each x* = xt, i = 1,2, ... and hence for each x* in sp {xn. Therefore,
X** =J(X)EB sp {xr}1.. Consequently, there is a projection of X** on
J(X) and (I. 2. 13) J(X) is topologically isomorphic to the factor space
X** /sp {xr} 1.. By I.3.8 it finally follows that the last quantity is
topologically isomorphic to sp {xr}*. This concludes the proof, since
then X is topologically isomorphic to sp {xr}*.
Corollary 12. If X has a boundedly complete basis, then X is topologi-
cally isomorphic to a conjugate space.

4. Unconditional Bases
Definition 1. A basis {Xi' xr} for X is said to be unconditional iffor all

00
X in X the series L Xi(X)Xi converges unconditionally.

i= 1
Now, let L be the set of all finite subsets 11 of the set of all positive
integers and let {xJ be an unconditional basis for X. Furthermore,
let Y be the vector space of all sequences y= {O(i}, O(iECP, for which
L O(iXi converges to an element X in X. Let S be the set of all sequences
i~n
00
{Yi} in cp. withly;I ~ 1. Since for each x* in X* the sum L O(iX*(Xi) is

i= 1
00
absolutely convergent (Theorem 11.2.2), L 100iX*(xi)1 is finite and an

i= 1
upper bound for sup {1&;YiO(ix*(xi)II{YJES,IlEL}. The principle of
uniform boundedness (1.3.14) then implies that sup {II ~ YiO(iXi III {yJES,
IlEL} <00. Next, we define a norm in Yby lIy,,=sUP{II~YiO(ixilll{YJES,

IlEL}, and show that Y is complete:
Lemma 2. Y is a Banach space and is topologically isomorphic with X.

Proof. Let T: Y~X be given by Ty=lim L O(iXi, yE Y. Since {xJ is
n i:<E;n
a basis for X, Tis one-to-one and onto. Because II TYII ~ Ilyll, Tis bounded.
If yp= {O(PJ is a Cauchy sequence in Y, due to the completeness of X,
Typ converges to an element X in X. From the definition of norm in Y it
follows directly that 100Pi-O(qilllxill=II(O(Pi-O(q)x;lI~IIYp-yqll. Hence
there is a sequence {O(J C cP such that limp
O(p., = O(i for all i and the rest
of the proof follows exactly the lines of the rest of the proof for Theorem
1.3.
Lemma 3. If {O(i}E Y, and {PJ is a sequence such that IP;I ~ 10(;1 for all
i, then {PJ is in Yand II {PJ II ~ II {O(J II·
Proof. Since it may happen that O(i = 0 for some i, we use the
convention that 0/0=0. Then II{PJII=suP{II~YiPiXilll{YJES'IlEL}
=sUP{II~Yi(P;/O(i)O(iXilll{~JES' IlEL}~SUP{II~YiO(iXilll{YJES, Il EL }
= II{O(JII. Finally, if lim L O(iXi=O, it
m,n i:;::::m
follows from Lemma 2 and the
4. Unconditional Bases 39
n
above estimate that lim L PiXi=O. By completeness of X, {Pi}

mIn i=m
then is
in Yand we have the lemma.
Based on these two lemmas we obtain
Theorem 4. Let X* be separable. Then each unconditional basis for X
is shrinking.
Proof. We suppose that an unconditional basis {xJ for X is not
shrinking. Then there exists an e> 0, an x* in X* of norm one, a sequence
{yJ in X and a sequence {mJ of strictly increasing indices such that
IIYdl=l,x*(Yi»e and YiESp{xmi+l"",Xm;+.} (If the basis is not shrink-
ing, then there is an e > 0 and 0"# x* E X* such that for every given
index p there is an X¢SP{X1""'Xp} in X with Ilxll = 1 and Ix*(x)1 ~2e.
Since {xJ is a basis for X, there is a q> p with Ilx- Uqxll <e/llx*ll, where Uq
is an expansion operator of {xJ Hence with Y = [lx*(Uqx)l/x*(Uqx)] Uqx
one obtains x*(y)= Ix*(Uqx)1 ~ Ix*(x)I-llx- Uqxll >e, where now
YESp{X p +1,""x q } and Ilyll > I-e). Thus there is a sequence raJ in tP
such that Yi= L ajx j . Now, let {PJ be any sequence in tP. Using
j=mi+ 1
Lemmas 2 and 3 we obtain
.L PiYi11 =II.L Pi. mf

I t::::n l~n j=mi+ 1
ajxjll~MII.L IPd. mf
l~n }=mi+ 1
ajxjll
=M Ili~n IPilYi11 ~M i~n IPdX*(Yi»eM i~n IPil,

where M is a constant, 0 < M ~ 1. On the other hand, it follows immedi-
ately that II.L PiYi I ~
l~n
.L
IPilllYil1 = .L
IPd and the linear transformation
l:::;n l~n
Ton 11 (over the field tP) to X, defined by T{yJ = lim L YiYi, {Yi}El 1,
n i~n
is bounded. Since JJT{yJJJ ~eMJJ{yJJJ, {yJE1 1, T- 1 is also bounded so
that T is a topological isomorphism of 11 onto T(ll) c X (1.2.15). Hence
Ii is topologically isomorphic with T(ll)* (1.3.25), which is again topologi-
cally isomorphic with the factor space X* /T(ll)l. (1.3.8). But this is
. impossible since X* is separable and Ii is not (we observe that if {yj} is
a countable dense set in X*, so is {yj+T(ll)l.} in X*/T(ll)l.). This
contradiction leads to the conclusion that {xJ is shrinking and the theo-
rem is proved.
Theorem 5. If {xJ is an unconditional basis for X and X is weakly
sequentially complete, then {Xi} is boundedly complete.
Proof. All spaces occurring in the proof are supposed to be over the
same field. We use the facts that c~* is isometrically isomorphic with 100
4*
(I.4.a and c), that Co is separable (1.4.a), 100 is not separable (1.4.c) and that
J(co) is weak* dense in C6* (I.3.22). Consequently, for every y** in C6*
there is a sequence {yj in Co such that y**(x*)=li~x*(y;), , X*EC6. If Co
would be weakly sequentially complete, then y** would be in J(c o),
i.e. Co would be isometrically isomorphic to 100 , which is impossible.
Thus Co is not weakly sequentially complete. Since by (I.3.15), every
weakly converging sequence {yj in the weakly sequentially complete
space X converges weakly to an element in Sf>{yj, every closed linear
subspace of X is itself weakly sequentially complete, hence can not be
topologically isomorphic with Co (cf. proof of Corollary V.3.2).
Now, if {x;} is not boundedly complete, then, as we will demonstrate,
there exists a subspace of X which is topologically isomorphic to Co,
and this contradiction shows that {Xi} must be boundedly complete. In
order to show this, we suppose the existence of a sequence {lXj in <l> such
that sup {II~ lXiXi 11\.uE~} ~ 1 and such that i~n lXiXi does not converge
to any element of X. Thus, the series is not Cauchy in X and there exists
an 8>0 and sequences of integers {nJ and {mJ such that ni~mi<ni+l
for all i and IIj~n,lXiXi\\):!8 for each i. Next, let Zo be the set of all se-
quences 13= {f3J in Co which have only a finite number of non-vanishing
m,
elements and let T:Zo-"X be defined by Tf3=lim L f3i L IXjXj' f3EZo·
n i~n j=ni
By Lemma 2 and 3 there is a constant M > 0 such that
IITf3l1 =11 ~f3i j~, IXjXj11 ~MII~llf3llj~, IXjXj11

~Mllf3llsup {lit IXjXjll\.uE~} ~Mllf3ll·
Therefore, IITII ~M. By (1.2.12), since Zo is dense in Co, T has a unique
continuous linear extension, T, on the whole space Co. From the same
lemmas we obtain also
liT 1311 = Illi~.L f3i .t IXjXjll):! M-

l::;;n )-ni
1
sup {II.~ 6ik f3i.t IXjXj111 k= 1,2, ...}
1-1 )-ni
~ M' ,up {IP.lIII OJXJII b 1,2, },,' M~ 'IIPII, pcco·

By (1.2.15) this estimate implies that T has a bounded inverse. Thus Co
is topologically isomorphic to T(co)c X and the proof of the theorem
is complete.
4. Unconditional Bases 41
Theorem 6. If X* is separable and {XioXi} is an unconditional basis

for X, then {xi", J x;} is an unconditional basis for X*.
Proof. By Theorem 4, {Xi' xi} is a shrinking basis for X. Theorem 3.9
and Corollary 3.5 imply that {xi",J x;} is a boundedly complete basis
for X*. Let then {m;} be any indefinitely increasing sequence of integers.
Since {x;} is unconditional the subseries L x~i(x)xmi converges in X for
i~n
each x in X. Thus s~pl.I JXm,(X*)X~,(x)'I<oo, XEX, X*EX*. As a con-

l~n
sequence of Theorem I.3.14 it follows that St;plIJn JXmi(X*)X~ill <00,

X*EX*. But since {xi} is a boundedly complete basis for X* we infer
that for every x* in X* the subseries I J xm,(x*) X~i converges in X*.
i $;n
Because {m;} was chosen arbitrarily, the expansion I J xi(x*)xi" for X*
i~n
in the basis {xi} converges unconditionally (IL1.3) and this finishes the
proof.
Combining the results of Theorem 4.6,3.9 and 4.4 we get immediately
Corollary 7. If {Xi,Xi} is an unconditional basis for X and if X* is
separable, then {xi",J x;} is an unconditional boundedly complete basis for
X*.
Corollary 7 applies in the special instance where X is reflexive, since
then the separability of X implies the separability of X* (1.3.11).
Theorem 8. If {Xi' xi"} is an unconditional basis for X and X* is weakly
sequentially complete, then {xi} is an unconditional basis for X*.
Proof. Let L be the set of all subsets of the set of all positive integers.
00
As a consequence of Riemann's theorem (11.2.2), I Ix*(xJxi"(x)1 is finite

i= 1
for each x in X and x* in X*. Therefore, sup{ I~ x*(xi)xi"(x)lflEL } < 00.
'Ell
(I.3.14) then implies that SUP{II~X*(Xi)Xi"lllflEL}<oo, hence that
s~p {Ix** (~ x* (Xi) xi") II fiE L } < 00 for every x** in X**. This shows that
L x** [x*(xi)xTJ is absolutely convergent for all x** in X**. Now,

i=l 00
since X* is weakly sequentially complete, the series I x*(xi)xi" is sub-

i= 1
series convergent in the weak topology of X*. But due to the Orlicz-
Pettis theorem, the series is strongly subseries convergent and hence
unconditionally convergent in X*. Moreover, the limit element must be
x*, since the series converges to x* in the weak* topology of X* (Theorem
2.7). Since x*(Xi)=JXi(x*), we have proved that {Xt,JXi} is an uncondi-

tional basis for X*.
Corollary 9. (KARLIN) There is no unconditional basis for the space
C[0,1 J.
Proof. C* [0,1] is weakly sequentially complete (L4.d). The corollary
then follows from the preceding theorem since C* [0,1], again by (l.4.d)
is non-separable.
5. Absolutely Convergent Bases and Uniform Bases
Definition 1. A basis {Xi,Xn for X is called absolutely convergent if

00
L Xt(X)Xi converges absolutely for every x in X.

i= 1
It is clear that every absolutely convergent basis is automatically an
unconditional basis.
Theorem 2. If X possesses an absolutely convergent basis, then X is
topologically isomorphic to 11 ,
Proof. Let {Xi' xn be an absolutely convergent basis for X. We define
the mapping T:X -+11 by T X = {xt(x)llxdl}, XEX. Clearly, T is linear. Let
{Yj} be a sequence in X such that li1p.Yj= YEX and such that TYj con-
J 00 .
verges to some element {exi} in 11 , Then li1p. L Ixt(y)llxdl-exd =0. But this
Ji= 1
implies that Ty = {ex i }, since xt(y)llxill = xt (lif1Yj) Ilxdl = lif1xt(y)llxill
=ex i for all i. Thus Tis closed and by (1.2.10) it follows that Tis bounded.
00
Furthermore, T is onto, because for every {ex;} in 11 , X= L (exJllxill)xi

i=l
is in X and Tx= {ex;}. On the other hand we obtain that Ilxll
00
= II Ii!? i~n xt(x)xill ~ i~llxt(x)lllxill = IITxl1 for each x in X. As a result

of this (1.2.15), X is topologically isomorphic with 11 ,
00
Definition 3.A basis {xi,xn for X is uniform if L Xt(X)Xi converges

uniformly for all x in the unit ball of x. i= 1
Lemma 4. If X is n-dimensional, then there exists a basis {x 1 , ••• ,xn }

for X with associated biorthogonal sequence of coefficient functionals
{Xi*""'Xn*}.In X* .
6. T-Bases 43
Proof. X has a Hamel basis, since each linearly independent set of n

elements {x 1, ... , x n } in X forms such a basis. The existence of a biortho-
gonal sequence of coefficient functionals {xi, ... ,x:} follows from Theo-
rem III.2.4 or directly from the following consideration: Due to (I.1.17)
there is a topological isomorphism T of X onto En. Let {yJ be an ortho-
normal basis for En. Then X=T-1[t1(TX,Y;)Yi]=it/TX'Yi)T-1Yi'

Since the set {T- 1yJ is linearly independent in X, the set {(Tx,y;)} is
unique for each x in X. Moreover x{:X--dJ, defined by X{(x)=(Tx,y;)
is linear and, by Ix{(x)1 ~ IITlllly;llllxll, bounded. This implies that
{T- 1Yi,Xi} is a basis for X and we are done.
Theorem 5. There exists a uniform basis for X if and only if X is finite
dimensional.
Proof. Let X be finite dimensional, say of dimension n. Then by the
n
lemma there exists a basis {x 1, ... ,xm xi, ... ,x:} for X. Since X= I X{(X)X i
i= 1
for all x in X, the basis is obviously uniform. Conversely, the assumption
of a uniform basis {Xi' xi} for X with expansion operators Un, n = 1,2, ... ,
implies that li~III - Unll =Ii~sup {llx- i~nX{(X)XiIIIIIXII ~ 1} =0. Since
each Un is bounded and with finite dimensional range, by (I.1.12), each
Un is compact. It thus follows (I.2.17) that I is also compact, a fact which
is possible only in' finite dimensional spaces (I.1.12). This completes the
proof of the theorem.
6. T-Bases
We shall now treat summability methods and their relevance to

Banach spaces. Use is made of infinite matrices with special properties
which are known as consistent or Toeplitz matrices. We begin with.
Definition 1. Let T= {tij} be an infinite matrix with entries in CPo
Then a sequence {x;} is T-limitable to x in X if Yi=lim I tijX} exists
for all i and Yi converges to x in X. n j"fn
Definition 2. An iriflnite matrix T is consistent if and only if, whenever

a sequence in X converges to x in X, the sequence is also T-limitable to
x in X.
Theorem 3. An iriflnite matrix T= {tiJ is consistent if and only if the
following conditions are satisfied:
00
(i) L Itijl~M<oo,
j= 1
i? 1,
(ii) li~ tij=O, j?1 and

!
00
(iii) lim L tij= 1.

j= 1
Proof. Necessity. Let {C(j} be a convergent sequence in cP with limit
c(o and let XEX and X*EX* be such that x*(x}= 1. Then, since Tis
consistent and since C(iX converges to c(o X, Yi = L tijC(jX also converges
to C(ox, Hence j~l tijC(j = x* (~1 tijC(jX) = ~=*l(Yi} converges to C(o·
This shows that T is consistent in the numerical case.

Therefore we consider the linear functionals Ti on the space Co, given
00
by TJC(} = L
tijC(j, C(= {C(;}ECo. We first fix i and show that each Ti
j= 1
is bounded on Co. Assuming the contrary would imply that Ti is not in
00
c~, hence that .L Itijl = 00 (1.4.a). In other words the sequence {.L It ij l}
)=1 J~n
would be not Cauchy in IR. Hence there would be an c > 0 and an
increasing sequence {m n } such that L Itijl>c. We then can take

j=mn + 1
C(i=ti)(l1ltijl) if mn<i~mn+l and tij=lO, else C(i=O. Clearly {C(J, so
00 mn+l 00
chosen, is in Co· But L tijC(j? L {1/n} L Iti)?c L (1/11) = 00

j=1 n=1 j=m n +l n=1
which is a contradiction. Next, since lim Ti(C() exists for every C(,
i
SUpITJC(}I<oo,C(EC o, and the uniform boundedness principle (1.3.14)
i~plies supIITill~M<oo.
!
Thus Ij=I tijC(jl~MSUPIC()'i=1'2'00.
1 J
and
choosing C(j=ti}ltijl for j~11 if tij=lO and C(j=O else, we obtain

n
L Itijl ~ M, 11 = 1,2'00" from which (i) follows immediately. Finally,
j= 1 00
let C(n)={bnJ and C(0)={1,1,00.}. Then O=li~C(\n)=li~ L tijC(jn)

! ! j= 1
00
=lim tin>n=1,2, 00., which is (ii) and 1=limC(\0)=lim L tijC(jO)

i 00 l l j== 1
= lim L t ij , which is (iii). Sufficiency. It is convenient to choose a

j= 1
!
sequence {x;} in X, converging to x in X, which is such that sup IIXil1 ~ 1.

!
6. T-Bases 45
By condition (ii) and (iii) and the convergence of {x;} there exists for
every B>O an n and an io depending on n, such that for i~io,
j~n Itijl <B/6, 1 1- j~l tij I<B/3 and Ilx-xjll <B/(3M),j~n. Hence, due to
(i), II x- Jl tijXjl1 ~ Ilx ~ - j~l tij) II + t~n ti/x-x)11 + Ilj~n ti/X-X)II

<B/3+2B/6+MB/(3M)=B,i~io. This shows that the sequence {x;}
is T-limitable to x and the proof of the theorem is finished.
Definition 4. Let T be an infinite matrix with entries in tP. Then a

sequence {x;} in X is called a T-basis for X if for every x in X there is a
unique coefficient sequence {()(;} in tP such that the sequence ()(iXi} {L
is T-limitable to X. i"in
We observe that the uniqueness implies that each Xi is non-zero and
that each basis for X is a T-basis for X with T={biJ
Theorem 5. Let T= {t mn } be a consistent invertible lower triangular
matrix and let {xJ be a T-basis for X. Then there is a unique sequence
{xi} in X* such that x;"(X)=()(i' XEX and that x;"(X)=b ij .
p,roof. Let Y be the linear space of those sequences Y = {()(J in tP for
which the partial sums L
()(iXi are T-limitable to an element x of X
Cf) i~n
and let zp= L tpn L ()(iXi= L tpn L ()(iXi' We define the norm in
n= 1 i~n n~p i~n
Yby Ilyll =sup Ilzpll, yE Y and we will show that the normed linear space
p
Y is complete.
If T - 1 = {urnn}, which is again a lower triangular matrix, we note that
00
L ()(iXi = L unpzp = L unpzp. Now, let {Yn} be a Cauchy sequence

i~n p= 1 p~n
in Y, where Yn = {()(nJ It is then clear that lim ()(ni= ()(i exists and we will
n
prove that Y = {()(J E Y and that Y = lim Yn in Y. To this end let
n
zpn = L tpj L ()(niXi' Obviously, Ilzp-zqll ~ Ilzp-zpnll + Ilzpn+zqnll

j~p i~j
+ Ilzqn+zqll and Ilzpn-zpmll ~ IIYn- Ymll· Taking the limit on m we get

for every £>0 an n such that Ilzpn-zpll ~B/3, independently of p. Since
YnE Y we can choose an r, depending on n, such that for p, q~r,
Ilzpn-zqnll <B/3. This implies Ilzp-zqll <B, p, q~r so that yE Y. That
Y is complete finally follows from Ily - Ynll = sup Ilzp - Z pnll which, by
the above, can be made arbitrarily small. p
Next, let S: Y-* X be the one-to-one linear transformation defined by

Sy=x. Evidently, IISII ~ 1 so that, since Sis onto, S-l is also bounded
(1.2.6). Defining linear functionals xr on X by xr(x)= CXi' we obtain

I x!(x) I = Icxli = II CX l Xl 11/11 xlii ~ II X111-lll u 11 zlll ~ II Xl 11":11 U11ll1yll
~llxlll-llu11IIIS-1111Ixli. In a similar way one has for i>1,lxr(x)1
= t~i CXj Xj - j~i CXjX j II/"xdl ~ 211 xdr 1 sup {t~k u kp zp III i-1 ~ k ~ i}
~21lxdrl KdIS-lllllxll, where Ki=SUP {L IUkPlli-l ~k~i} <00.
p:i;k
Thus each xr is an element of X*. Finally, the consistency of T
00
implies that liJ;l1 L tij= 1 for all n, hence that L 0niXi is T-limitable
L j=n i~m
to Xn which demonstrates that xr (xn) = bin and the theorem is proved.
Theorem 6. Let {Xi} be a total sequence in X with biorthogonal set {xn
in X* such that for some consistent matrix T={tij}, lim tni xj(x)Xj L L
exists for n=1,2, ... and s~pll.f
,=
tni~.Xj(X)Xjll<ooP f~~Pea:~ix in X.
1 l:!fl
Then {xJ is a T-basisfor X and {xr(;)} is the coefficient sequence corre-
sponding to X. 00
Proof. Let Tn:X ~X, n= 1,2, ... be defined by Tnx= L tni L xj(x)Xj'
i= 1 j~ r
The hypothesis and the principle of uniform bounded ness (1.3.14) then
imply that supllT,,1I ~M < 00. Since sp{xJ =X we have for each x in X
n
and every s>O an m and an elementy Ym in sp{xdi~m} such that Ilx- Ymll
<so On the other hand, since Tis consistent and since L Xr (Ym) Xi = Ym for
j~i
all i?>m, there is an n, such that IIYm- T"Ym"=!!Ym-.f tni

r.= 1
~.XiYm)Xjll
J~t
<s, n?>n,. But then, Ilx-T"xlI~lIx-Ymll+IIYm-T"Ymll+IIT,,(Ym-x)1I

~(2+ M)s if n?> n, so that L xj(x)x j is T-limitable to x. It now remains
j~i
to show that the coefficient sequence {xr(x)} for x is unique. This

00
follows from the assumption lim L tni L CXjXj=O for some sequence
n i= 1 j~i
{IX;} in <P, and by multiplication of this equation with x!. Then
O=X!(li~.f tni L.lXjXj)=C(mli~.I tni=lXm(li~.I tni-li~.L tni)=lXm.

1=1 J~l l=m 1=1 l<m
Hence {cx;} consists of zeros and the sequence {xr(x)} is unique.
Example. It is easy to see that the matrix T= {tij}, given by
t .. = {Iii, j~i,
1) 0, j>i,
is consistent. The corresponding T-basis for X is called a Cesaro basis
for X.
7. Bases for Special Spaces
Though it is still an open question wether there exists a basis for

every separable Banach space, bases are known for almost all of the usually
encountered examples of infinitely dimensional separable Banach
spaces. We begin with the spaces which are the easiest to treat, the spaces
of sequences, Co and Ip, 1 ~p< 00. We use the symbol bi for the sequence
{bij} in If>, where bij is, as usual, Kronecker's delta.
Theorem 1. {b;} is a monotone unconditional and non-boundedly
complete basis for Co with biorthogonal sequence {b i } in c~ = 11'
Proof. If (X={(X;} belongs to co, then it is clear that (X = lim L (Xibi'

n i~n
As an immediate consequence of the obvious equation (Xi = bi"((X), where
bi"EC~=ll is given by bi"=b i , and Theorem 2.4, {c5;} is a basis for co.
The basis is unconditional since the definition II(XII = s~pl(X;I
, of the norm
in Co implies that the expansion for (X is subseries, hence unconditionally
convergent (11.1.3). That the basis is monotone is immediately clear
from the definition of norm in co. The following counter-example shows
that the basis is not boundedly complete: The sequence (X= {1, 1, ... } in
If> is such that s~p II.L I
(X/); = 1. But· .L
(Xi bi is not convergent in Co and
l~n l~n
this contradiction concludes the proof of the theorem.

Theorem 2. Let xij = 1 for j ~ i and xij = 0 for j> i. Then with
Xi = {xij}, {x;} is a basis for Co which is not unconditional.
Proof. It is easy to verify that the sequence {xi"} in 11' given

by xi" = bi - bi+ 1, is biorthogonal to {x;}. Since sp {x;} = Co and since
s~plI Un (X II =s~p lIi~nxi"((X)x;1I =s~p lIi~n((Xi-(Xi+ l) Xi l =s~psup{l(Xj-(Xnllj < n}
~2supl(Xnl =211(X11, Corollary 2.3 implies that {Xi} is a basis for co. We
n
use an example to show that {x;} is not unconditional. Let (X be an element
of Co given by (X = {( _1)i + 1 Ii}. We then take the sub series of all odd terms
in the expansion of (X:
00 00
i=l,iodd i=l,iodd
Finally, from the divergence of the series 1 +!+t+... we conclude

that the series expansion for rx is not subseries, hence not unconditional
convergent (II.1.3) and the proof of the theorem is complete.
Theorem 3. {C5;} is a monotone unconditional retro-basisfor lp, 1 ~ p< 00.
The basis is absolutely convergentfor p= 1 but notfor p> 1.
Proof. The biorthogonal sequence to {C5J is {C5i}, C5t E I; = Iq (lip + 11q

=1) given by C5t=C5;.Sincefor rxElp,C5i"(rx)=rx;, LC5t(rx)C5; converges to
i~n
rx in the topology of each of the spaces lp, 1 ~ p < 00. Therefore, the
theorem is a consequence of Theorem 2.4 and {C5J is a retro-basis
since for all i, C5; is an element of Co as well as of lq, 1 < q < 00. To show
that {C5J is unconditional, let {nJ be any increasing sequence of
integers. Since Ip is complete and because t~1l C5:,(rx)C5 n ,11 = C~1l Irxn,lpf/P
~ (~lrxIP) l/p for each rx in Ip and each finite set of integers fl, the expan-
!EIl
sion for rx is subseries convergent. Thus {C5J is unconditional for 1 ~ p < 00
(1I.1.3). Next, let p> 1 and let pi be such that 1 < pi < p. Then rx = {i- pl/P}
00 00 00
is an element of Ip because IlrxlI P= L U-PI/P)P= L i-PI~1+ St-P'dt

;=1 ;=1 1
n
= l-(l-p')-l < 00. But L IIC5t(rx)C5dl = L Irx;l = L i-pl/p~ St-pl/Pdt
i~n i:S;n i~n 1
=(1-p'lp)-1(n1-pl/p-1) which diverges as n->oo, showing that in this
case the convergence of the series expansion for rx is not absolutely con-
vergent. However, the basis is absolutely convergent for p = 1 since
00 00
L IIC5t(rx)(5dI = L Irxd = Ilrxll· The monotony finally follows directly from

;= 1 ;= 1
the definition of the norm in Ip.
Theorem 4. Let Xl = {1,0,0, .. .},X2 = { -1, 1,0,0, .. . },X3 = {1,0, 1,0,0, ... },
.... Then {xJ is a weak* as well as an absolutely convergent basis for 11,
but {xJ is neither a retro-basis nor a weak* Schauder basis for II.
Proof. It can be verified directly that xT = {l, 1, -1,1, -1, ... } and
xt = C5;, i> 1, are biorthogonal functionals in IT = 100 • Let rx be in II. Then
Since the last term converges to zero with n, Theorem 2.4 implies that
{xJ is a basis for 11. The basis is absolutely convergent since
+L 1+ 2 L
00 00 00
L Ilxt(rx)xdl = Irx1 (rx/ Irxjl ~ 311rxll·

;=1 j=2 j=2
Finally, {Xi} cannot be a retro-basis because x! ist not in Co. {Xi} can
not be a weak* Schauder basis for 11 since then Theorem 2.7 would imply
that X!EC o which is a contradiction. However, {x;} is a weak* basis for
11 since it is a basis for 11 and since lim I lXi(Xi)j=O,j~ 1, implies 0=1X2
= 1X 3='" and thus 1X1 =0. n i';;n
Recalling that XS denotes the characteristic function of a set S we
have
Theorem 5. The following sequence in C[O,IJ (Schauder's system) is
a monotone basisfor C[O,IJ:
Xo(t) = X[O,lj(t),
Xl (t)= t X[O,l](t),
X2(t) = Xl (2 t) + X(O, 1](2 t-I) - Xl (2t -1),
Proof. Let xEC[O,IJ be arbitrary and let YnEC[O,IJ for n~ 1 be

defined by Yn(t)=x(t), t=0,I/2n,2/2n,3/2n, ... ,I, variing linearly between
these points. Since xEC[O,IJ it is uniformly continuous (l.4.d) on [O,IJ
and the polygonal functions Yn converge uniformly to x. Clearly,
YnEsp{xO, ... ,X2n}. Hence {Xi} is total in C[O,I]. Furthermore it is
evidentthat II.I lXiXill,,;;ll. I
,::E;n l~n+ 1
lXiXil1 forall n~I andarbitrarynumbers
1X1,1X2,'" in €P. Thus the theorem is a consequence of Theorem IV.l.S.
The inequality also implies the monotony of the basis {Xi}'
Unfortunately, here (and in the following proof) we make use of a
theorem which is proved later in chapter IV. But we feel that this is
justified, since we like to have some applications in classical Banach
spaces already now and it is desirable to have them all together.
Theorem 6. Haar's system, which is given by

Xl (t) = X[O,l](t),
X2n+ j(t)=2n/2[X[O,l](2n + 1 t-2j+2)- X(O,l](2n + 1 t-2j+ I)J,

j=I, ... ,2n, n=0,I,2, ... ,
is a monotone basis for Lp[O,IJ, 1";;p< 00.
Proof. Let xEC[O,IJ be arbitrary and letthe function YiESp{x)j";;2i},

i~O be defined by Yi(t)=X(t), t=j/2 i,j= I,2, ... ,2 i . Since x(t) is continu-
ous, it is uniformly continuous (l.4.d). Hence the step functions Yi converge
with i uniformly to x on [0,1]. Therefore, li~llx- Yill

, =0 and, since the
subspace C[O,lJ of Lp[O,lJ is dense in Lp[O,lJ, l::::;p<oo (1.4.8), Haar's
system {x;} is total in Lp[O,l]. For YEsp{xjli<i}, i=2,3,4, ... we have
forallrxE€f>,
It t~-C
Vy(t)+rxx;(tWdt = -'-2-' [ly(tt)+rx2 n / 2 IP + ly(tt)-rx2 n/2I PJ
where ti =sUp{tltESUpp(X i)} and t i- =inf{tltEsupp(x i)}· Thus, IIYII

::::; Ily+rxxill and according to Theorem IV.U, {x;} is a basis for L p[O,lJ
and this basis is obviously monotone. Since the functionals on L p[O,lJ,
1
J
defined by xi(x)= x(t)xi(t)dt, i= 1,2,... are bounded and satisfy
o
XnX)=bij and since the associated biorthogonal sequence of a basis
is unique, we notice that {xi} is just the biorthogonal sequence of {xJ
This completes the proof of the theorem.
Theorem 7. Let {Xi} be the sequence of functions given by x o(t)=(2n)--!-,
x 2i _ 1 (t)=n--!-sin(it), X2i(t)=n--!-cos(it), 0::::;t::::;2n, i= 1,2, .... Then {x;},
the trigonometrical system, is an unconditional basis for the Hilbert space
L 2[0,2n].
Proof. It is easy to see that {x;} is an orthonormal system for L2 [0,2 nJ
(i.e. (X;,X)=biJ Let the sequence {xi} in L'HO,2nJ be defined by
xi(x) = (x, Xi)' XEL 2 [0,2n]. Since {Xi' xi} is evidently a biorthogonal
system for L2 [0,2 n J and since, according to Theorem 1.4.11,
lim L xi(x)xi=x for each x in L 2 [0,2nJ, Theorem 2.4 implies that
n i:%.n
{x;} is a basis for L2 [0,2 n]. Furthermore, it is possible to show that
{Xi} is unconditional. Let {n;} be any increasing sequence of
integers. Then, since II i t X:'(X)xn, II = II itp (X, Xn) Xn, II = itp l(x,xn,W
00 00
::; I I(X,XiW, and since by Bessel's inequality, I I(X,Xi)l::::; Ilx11 2 ,

i=O
the completeness of L2 [0,2 nJ implies that the sub series I (x, xn) x ni
i= 1
is convergent in L [0,2 nJ for every x in L2 [0,2 nJ and from (ILL 3)
the assertion follows.
To proof the next theorem we make use of the Rademacher functions IjJno
defined for n=1,2, ... ontheinterval [O,lJ oflRby IjJn(s)=signsin(2 nns),
where signs= -1, =0 or = 1, according to s<O, =0 or >0, respec-
tively. Let E,;t={sll/ln(s)zO} and denote by lEI the Lebesgue-measure

of a measurable subset of [0,1]' It is then clear that
1
JI/In(s)l/Im(s)ds= IE: n E,;'; I+ IE; n E';; I-IE; n E,;'; I-IE: nE';; I

o
=(1 + bnm )(!+!) - (1-b nm )(!+!)= bnm ,
1. e. {I/I n} is an orthonormal system in the Hilbert space L z [0,1]'
Theorem 8. The sequence {x;} of Theorem 7 forms a basis for Lp [0,2 n],
1 < p < 00, which is not unconditional for p =1= 2.
Proof. Let 1 < p < 00 and let q = p/(P -1). First we assume L p [0,2 n]
to be real. We take in L~ [0,2 n] (which is by (1.4.6) isometrically iso-
Z1<
morphic to L q [0,2n]) a sequence {xt} defined by xt(x) = Jxi(t)x(t)dt,
o
x E Lp [0,2 n], i = 0, 1, .... Clearly, the system {Xi' xt} is biorthogonal
and since by Theorem 1.4.11, li~llx-i~nxt(x)xill=O for every x in
Lp [0,2 n], as a consequence of Theorem 2.4, {x;} is a basis for Lp [0,2 n].
First we restrict ourself to a fixed p with 1 <p<2. Now, if for every
xi: h[0,2n] we would have i~O Ixt(xW<oo, then since !lit Xt(X)Xi/1:
= L Ix{(xW and since L z [0,2n] is complete, L xt(x)x i converges
i=n i~n
to an element yin L z [0,2n] in the topology of L z [0,2n], where Ilfllz

denotes the norm offas an element of L z [0,2n]. Due to Holder's inequal-
ity, IlflI P = T If(t)IP dt ~ [T If(t)IP(ZIP)dtT . (2n)1-pIZ=(2n)1-PI2I1fll~

/Z
forallfin Lp[0,2n].Henceyisin L p[0,2n] and LXt(x)xi converges

i~n
toyin Lp [0,2 n]. But because {Xi} is a basis for L p[0,2n] we have x=y
so that x is in L z [0,2n]. However, this result is impossible because for
instance x(t)=t-t, 0~t~2n is in L p[0,2n] but not in L z[0,2n]. This
00
shows the existence of an element x in L p[0,2 n] such that L Ixt(xW = 00.

i= 1
Let us now suppose that {x;} forms an unconditional basis for L p [0,2 n].
Then there exists for each x in L p[0,2n] (cf. section 4) a constant M>O
such that IlL xt(x)xill < M for every finite set J1 of positive integers.
lEJl 1
Taking pi=n- 2 (lxi i_l(xW+lxt(xW)t, we have for i=1,2, ... , by a

suitable choice of cP i,
With this notation one gets

21<
f I~PiCOS(it+CPi)IPdt<MP.
o
Let {l/! n} be Rademacher's orthonormal system in L2 [0,1]' Next, let
sE[O,l] be fixed, let m be any positive integer and let Jl+ and Jl- be the
set of integers i, i ~ m, for which l/! ;(s) is positive, respectively negative.
Then
r I.L Picos(it+ CPi)l/!;(s)IPdt= f

21< 21<
I.~ piCOS(it+cp;)-.L Picos(it+ cp;)IPdt

J l~m ~EJl rEp.
o 0
~{~[[I;E. P;COS(itH;)I'dtrr
~(2M)P<4MP.
As a result of an inequality for lacunary series (KACZMARZ and STEINHAUS

[1], p. 132),
I.L CXi 'Pi(s)IPds,

1
( .L Icxi12)P12 ~Kp f
l~m l~m
o
where {cx;} is any set of real numbers and K P is a positive constant
depending on p only, one has
[ .2: pfCOS2(it+cp;)lPI2~Kpf
l::::Sm J
1
I.L Picos(it+cp;) 'Pi(S)IPds.
l~m
o
Integration of both sides over (0,2 n) and inverting the order of integration
becomes
2"
PI
f [ .L pf cos (it+ cp;)J 2 dt~4KpMP.
2
l~m
o
Therefore, the sum in the square brackets converges with m almost

everywhere in (0,2n) to a p/2-th power integrable function (1.4.10).
From the definition of a Lebesgue integral it follows that there exists
in (0,2n) a set E of measure lEI =n and a constant N>O such that
[~/fCOS2(it+CPi)J/2 ~N for every tin E. Integrating the sum over
the set E becomes
fcos 2(it + CPi)dt= L~l pi cos (it+ cpMt":;;' nN

00 00
i~l pi 2 2 /p
E E
where we have used Lebesgue's dominated convergence theorem to

justify the interchange oflimit and integral. But for i ~ 1,
n/(4i) n/4
Scos 2(i t + cpJdt ~ 4 i J sin 2(i t)dt = 4 J sin 2 t dt > 0.
E 0 0
Thus I pi < 00 and this contradiction shows that for 1 < p< 2 the
i= 1
basis is not unconditional.
Finally, to treat the case 2 < p < 00 we assume {xJ to be an uncon-
ditional basis for L p [0,2 n J. Then by Theorem 4.6, {xl'} is an unconditio-
nal basis for L~ [0,2 n], so that by (1.4.6), {xJ is an unconditional basis
for Lq [0,2 n], 1 < q < 2. Since this is a contradiction the assertion of the
theorem follows. Finally, the complex case results from the fact that our
biorthogonal system {Xi' xT} is an (unconditional) basis for real L p [0,2 n]
if and only if it is an (unconditional) basis for complex L p [0,2 n J.
Remark. The trigonometrical system of the preceding theorem does
not form a basis for Ll [0,2 n J. F or, supposing the contrary, the sequence
of expansion operators {V n} of {xJ would be bounded (III.2.3). Since
2n
J
(V 2nX)(S) = (2 n)- 1 (sin it)- I sin(n +i) t xes - t)dt, 0":;;' s,,:;;. 2 n, XE Ll [0,2 n]
o
(the formula is known as Dirichlet's integral (DUNFORD and SCHWARTZ
[1 J, p. 359)), this would imply that 00 > supll Vnll ~ supsup{11 V 2nxlllllxli
2n: n n
":;;'1, XE LI [0,2 n J} ~ sup(2 n)- 1 S (sin it)- II sin (n +i) t Idt (hint: take
n 0
x(t)= l/s, O,,:;;.t,,:;;.s, x(t)=O else, and choose s>o arbitrarily small}.
2n n
But S(sinit)-llsin(n+!)tldt ~2 JS-l sin(2n+ l)sds~2[1 +t+~+···
o 0
+ 1/(4 n + 1)] which contradicts the above estimate. Hence {xJ cannot
be a basis for L 1 [0,2 n J.
Similarly, the trigonometrical system {xJ cannot be a basis for C[0,2 nJ
2n
either. For (2n)- I J (sinit)-llsin(n+!)tldt":;;'sup{I(V 2nx)(0)llllxll ":;;'1,
o
XE C[0,2 n]}":;;' sup{11 V 2nxlllllxli ":;;'1, XE C[0,2 n]}. The first inequality is
obtained by choosing for X continuous functions of norm one in C[0,2 n]
which approximate the function signsin(n+!)t with respect to the norm
in L I [0,2nJ.
The two results can be obtained as well from the knowledge of an
element x in L I [0,2n] whose Fourier series diverges in L 1 [0,2n]

(DUNFORD and SCHWARTZ [1J, p. 359), and from the existence of a

function x in C [0,2 n J whose Fourier series diverges at a point of the
co
interval [0,2nJ (ZYGMUND [1J, p. 167), using the fact that I X{(X)Xi
is just the formal Fourier series of x. i =0
References for Chapter III: BANACH [1], DAY [2], GELBAUM [1], JAMES
[4], KARLIN [2], RETHERFORD [4], SINGER [3, 7 and 12], WILANSKY [1] and
ZYGMUND [1].
CHAPTER IV
Orthogonality, Projections and Equivalent Bases
If there is a basis for a closed linear subspace Y of Banach space X,

a very general condition allows to define a projection of X on Y. On the
other hand, projections are very useful tools for existence proofs of
bases. Indeed, they can be applied in the proof of the existence theorem
of NIKOL'SKII given in the first paragraph. The theorem is quite essential
in the theory and in the application ofthe theory of bases and, accordingly,
we have to make reference to it in many of the subsequent theorems and
corollaries. The second paragraph refers to the very nice fact that the
concept of orthogonality can be extended from Hilbert spaces to the
normed linear spaces. It gives the relation between total orthogonal
systems, simple ~ -spaces and monotone bases. The last section is
concerned with equivalent bases, block bases and the theorem that every
infinite dimensional Banach space contains an infinite dimensional
subspace with a basis.
1. Bases and Projections

Let {Xi'Xi"} be a basis for the Banach space X. We recall that the
expansion operators Un are defined by Unx= I xt(x)x i for all nand
i~n
x in X (III.2.1). For each m and n, Urn U n= Umin{m,n}, implying that Un is

a projection in X. In the whole chapter IV, the word projection will be
used in the sense of a continuous projection.
Theorem 1. Let Y be a closed linear subspace of X. If {yJ is a basis
for Yand if there is a sequence {xi"} in X* such that xt(y)=bij and
such that L Xt(X)Yi converges in X for all x in X, then P, defined by
Px=lim L Xt(X)Yi' XEX, defines a projection of X on Y.
i~n
n i~n
Proof. Since the series for P x converges in X for XE X we have

liP xii ~ s~plli~n xt(x) Yill < 00, XE X. By (1.3.14) there is then a constant
5*
56 IV. Orthogonality, Projections and Equivalent Bases
M such that liP xii :;S;;Mllxll so that P is an endomorphism of X. Clearly,

P(X) c Y. Since {yJ is a basis for Y there is a unique sequence {IXJ in ljJ
such that y=lim L lXiYi, yE Y. Due to xi(y)=b ij , Py=lim L lXiPYi
n i~n n i~n
= lim L lXiYi = Y so that P(X) = Y. Thus P is a projection of X on Y,
n i:E;n c
concluding the proof of the theorem.

Corollary 2. Let {Xi'X[} be a basis for X. If L x**(X7)Xi converges
i~n
in X for all x** in X**, then there exists a projection of X** on J(X).
Proof. Let J' be the natural embedding of X* into X***. We take

Y=J(X)cX**. Thenx=lim L xT(x)x i implies thatJx = lim L Jx(x[)Jx i
n i~n n i~n
=lim L J'xT(Jx)Jx i, and {Jx;} is a basis for Y. Moreover, {Tx[}

" i~n
is a sequence in X*** such that J'x[(Jx)=Jxj(x[)=xT(x)=bij and
~uch that i~' J' x[(x**)J Xi (= J i~' x**(XT~ Xi) converges for. all x**
111 X**. Thus the assumptions of the preced111g theorem are fulfIlled and
J [li~ .L x**(xJx;] defines a projection of X** on J(X).
l::s.n
Theorem 3. Let Y be a closed linear subspace of X. If {y;} is a basis

for Yand if P is a projection of X on Y, then there exists a unique sequence
{z[} inX* such that zT(yJ=bij and such that Px=lim L ZT(X)Yi,xEX.
n i~n
Proof. Let {y[} be the associated biorthogonal sequence to

{yj. For all i the Hahn-Banach theorem insures the existence of an
extension yi' of YT to the whole space X. Since .vi' E X* we can
define P* yi' = zT E X* and we obtain the relations zT(Y) = P* y['(y)
= y[,(Py) = YT(y)=b ij , Now, Px=lim LYT(Px)y;=lim' L y[,(Px)Yj
n i~n n i~n
=lim LP*y['(x)Yi=lim LZT(x)Yi for every x in X. It remains to

n i!S:n n i~n
show that the sequence {z[} is unique. If {z[} is another sequence

with the properties of {z[}, then we have lim L (zi -Z[')(X)Yi=O
n i:%n
for all x in X. Multiplying through by yj we get (zj-zj')(x)=O for
every x in X and allj. Hence the sequence {zT} is unique and the theorem
is proved.
Theorem 4. Suppose that {Xi' x[} is a basis for X and that P is a
projection of norm one of X** on J(X). Iffor every x* in X* there is a
unique x*** in X*** for which Ilx***11 = Ilx*11 and such that x*(x)
= x*** (J x) for all x in X, then L x**(x[)x i converges in X for all
i~n
x** in X**.
1. Bases and Projections 57
Proof. Since {Xi} is a basis for X we havefor each x** in X**, J- 1 Px**
= lim L
X1(J-l P x**) Xi = lim L
P x**(x1) Xi = lim L
J' x1(P x**) Xi
n i::::;'n n i~n n i$:n
=lim L P*J'x1(x**)x i, where J' is the natural embedding of X*
n i $;n
into X***. It remains to insure the relations P*J'x1=J'x1, hence
that L
x**(x1)x i converges in X for each x** in X**:
i~n
Let Fi be the restrictions of J'x1 to J(X). Then IlFill = sup {1F;(Jx)1

IXEX, Ilxll ~ 1} =sup{IJ'x1(Jx)llxEX, Ilxll ~ 1} = sup{lx1(x)llxEX, Ilxll ~ 1}
= IIx111 = IIJ' x111· Moreover, P* J' x1 is an element of X*** and
P* J'x1(Jx)=J'x1(P Jx)=J'x1(Jx)=F;(Jx), XEX. Hence P* J'Xi is
an extension of Fi to the domain X** and we have IIFil1 ~ IIP* J' x111.
On the other hand, since IIP*II = IIPII = 1. we obtain IIP* J' x111 ~ IIJ'x111·
Thus IIP*J'x111=IIJ'x111=llx111 and since x1(x)=Jx(x1)=J'x1(Jx)
= p* J' x1 (J x) for all X in X it follows from the hypothesis that
P* J' x1 = J' x1, concluding the proof of the theorem.
Theorem 5. (NIKOL'SKli) A total sequence {x;} oJnon-zero elements
in X is a basis Jar X iJ and only iJ there is a constant M ~ 1 such that
II .L (XiXill~MII.L (XiXill Jar each n,m with n~m and arbitrary coeffi-
I ~n 1::::;111
cients (Xl"'" (Xm in cPo

Proof. If {x;} is a basis for X we define the constant M ~ 1 by
M = sup II Unll (III.2.3). Let {(X;} be any sequence in cP and let nand m be
n
positive integers such that n~m. Then L (XiXi= U" L (XiXi and
Ili~n (XiXill ~M Ili~m (XiXill·
i~n i~m
Conversely, we assume the inequality of the theorem to be satisfied.

From this inequality it follows that every finite collection {xili~n}
is linearly independent. Then we define the subspaces E1 c E2 c ... c En c ...
of X by En=sp{xdi~n}. Let now Pnn be the identity mapping from En
into X. Moreover, we define the projections P"m of Em on En' m>n, by
Pnm(X+ y)=x for each x in En and y in sp{xiln< i~m}. From the obvious
equation IlPnnll = 1 and the inequality it follows that IlPnm(x + y)11 = Ilxll
~Mllx+yll, hence that IIPmnll~M for every m~n. Now, we define
the linear transformation P~:D-'>En by P~x=Pnmx, xEE m, m< 00,
where D={xlxEEm,m<oo}. Thus, IIP~II~M and since the sequence
{x;} is total in X we have jj = X. Hence P~ has a unique linear extension,
say Pn, on X and IIPnl1 ~ M (1.2.12). Since En is closed, Pn has range En
so that P" is a projection of X on En"
The existence of the sequence {Pn } enables us to show that the total
sequence {Xi} is a basis for X: For each x in X and every 8>0 we have
an index n such that inf {llx - YIIIYE En} < 8. Hence there is a y in En for
which \\x - y\\ < a and we have \IPmx-x\\ ~ \IPmx- y\\ + \\y-x\\
=\IPm(x- y)\\ + \\x- y\\ «M + l)a for all m~n. This implies that
Pnx converges strongly to x. Let XTEX*, i= 1,2,... be defined by
xT(x) Xi = (Pi - Pi-l)X, XE X, taking Po = O. Since xT(x)x i = (Pi - Pi-l)Xj
=bijXi, {xi,xT} is a biorthogonal system for X. Now Un= I (P;- P;-l)= Pm
i~n
so that sup\\Un\\~M. Finally, since sp{xJ=X, Corollary III.2.3

implies that {xJ is a basis for X and we have the theorem.
We note that the first part of Theorem II1.2.2 again may easily be
obtained as a corollary of the above theorem: By the principle of uniform
boundedness one obtains from the hypothesis that sup \\ Un \\ < 00. Now,
n
let x=.I (XiXi, (Xi E<1>, i= 1, ... ,m. Consequently, II.I (Xixill = \\Unx\\
l~m l~n
~s~ \\Un\\lli~m (Xixill for each n~m, and by the preceding theorem
{xJ is a basis for sp {xJ
Theorem 6.1f r is a determining manifold for X and if {xn} is a sequence

in X such that
(i) O<inf\\xn\\ ~sup \\xn\\ < 00,
n n
(ii) lim x*(xn)=O for any x* in

n
r,
then there exists a subsequence {xn,} of {xn} with n l = 1 which is a
basis for sp {xnJ.
Proof. By (i) we obviously may assume that \\xn\\ = 1, n= 1,2, '" . Let
n l , ... , nk be an arbitrary increasing finite set of integers, Zk= sp {xnili~k}
and let Sk={zEZkl\\z\\=l}, the unit sphere in Zk' Because Zk is finite
dimensional, it is clear that Sk is compact (1.1.12). Thus there are elements
Zl>""Zm in Sk such that inf{\\z-zi\\lzESk , i~m}«2k)-2/4 (i.e.
there exists a finite a-net with respect to Sk)' Now, there exist elements
z1, ... ,z; in rcx* such that 1-(2k)-2/4~\zT(zi)\~1 and \\zT\\=1.
From the hypothesis (ii) we infer that there exists an integer nk+ 1 > n k
for which sup{IZnXnk+)\li~m}«2k)-2/4. Next, let Z be in Sk and
(X in <1>, \(X\~2. Then \\z+(Xxnk+l\\~\(X\\\xnk+l\\-\\z\\~1>1-(2k)-2.
On the other hand, if \(X\ < 2, then there is an index i for which liz - ziI\
«2k)-2/4, and
\\Z+ (XX nk +
1 \\ ~ \zT(z+ (X xnk+J \ ~ \ZT(Zi)\-\zT(z- Zi)\-\ZT((Xx nk +)\
~ 1-(2k)-2 /4-\\z-ziI\-2\zT(x nk +1)\
~ 1-(2k)-2.
1. Bases and Projections 59
Starting with n 1 = 1 we can find in this wayan increasing sequence

{nk} such that for arbitrary ct.k in ([>,
.I ct.;Xn'II~(1-(2kr2)-111. I ct.;xn,ll'
II l~k l~k+l
for k=1,2, .... Thus for 1~p<q we have
n (l-(2k)-2t 1 II.I ct.;xn,ll·

q-l
.I ct.;Xn,11 ~
II l~P k=p l~q
Since
n (1-(2k)-2t n (1-(2k)-2)-1
q-l
k=p
1 ~
OCJ
k=l
= .
n/2
sm(n/2)
= n/2,
Theorem 5 implies that {xnJ is a basis for sp {xnJ c X and the theorem
is verified.
Theorem 7. (GRINBLYUM) A sequence {x;} which is total in X is a
basis for X if and only if there exists a positive constant ct. such that for
all n,
whereS n is the unit sphere Sn={XIXEX~,llxll=1} in X~=sp{Xl>""xn},

Xn=SP{Xn+l,Xn+2''''} and dist(Sn,Xn)=inf{llx-ylllxESn,YEX n} is
the distance from Sn to X no
Proof. Necessity. Let {x;} be a basis for X which by Theorem 111.2.4
and Corollary 111.2.3 has a corresponding sequence {Un} of expansion
operators, such that 1~supIIUnll<CX). We define ct.E(O,1] by
n
ct.=(s~p IIUnlltl. Since each Un is a continuous projection, (I - Un)(X)
is closed and since sp{xn+1,xn+z, ...} c(I - u,,)(X), one has Xnc(I - U,,)(X).
Moreover, it is clear that Sn is the unit sphere of Un (X). Hence
dist(Sn'X n) = inf {llx- ylllxE Un (X), Ilxll = 1, YEXn}
~ inf {llx- ylllxE Un(X), Ilxll = 1, YE(I - Un)(X)}
~ ct. inf{11 Un(x - y)lllxE Un (X), Ilxll = 1, YE(I - Un)(X)}
= ct..
Sufficiency. Let {ct.;} be an arbitrary sequence in ([> and let n,m be any
integers with n<m. Supposing that dist (Sn,Xn)~ct.>O, it then follows
that
.I lXiXi!! = II!!.I lXiXi!!-I.I lXiX i + !!.I lXiXi!!-1

!! l~rn l~n l~n z.::S;n
._I
l-n+1
lXiX i II·!!.I lXiXi!!
,':::::;n
~ 1X!!i~n lXiXi!!
(if In lXiXi=O, !!i~m lXiXi!!~IX!!i~n lXiXi!! is trivially satisfied). Thus
Theorem 5 applies with M = IX-I + 1 and we are done.
2. Orthogonality, simple JV1 -Spaces and Monotone Bases

It is very useful to extend the concept of orthogonality, which is
well known in Hilbert spaces, to normed linear spaces. There are several
possibilities to do this and we restrict ourself to the one which seems to
be the simplest and best suited for application to the theory of bases
in a Banach space X:
Definition 1. We say that an element x of X is orthogonal to an element

y of X if and only if Ilxll ~ Ilx + IX yll for every IX in qJ.
Clearly this definition, introduced by JAMES [3J is homogeneous in
x and y but not necessarily symmetric nor additive from the left (or
from the right) as it is in a Hilbert space. We have the following result
for Hilbert spaces H (we use the word orthogonal throughout in the
sense of Definition 1):
Theorem 2. Two elements x and y of H are orthogonal if and only
if (x,y)=O.
Proof. Let (x,y)=O. Then IIxI12 = (x + IXY-IXY,X +IXY-IXY)= Ilx +lXyW
-lllXyI12~llx+lXyI12 for every IX in qJ. Conversely, let IlxI12~llx+lXyI12.
This implies IllXyI12+2Re(x,IXY) to be non-negative. But this expression
hasminimumvalue -1(x,y)1 2/1IyI12 so that (x,y) must vanish.
Definition 3. A sequence {xJ in X is said to be an orthogonal system
if sp {Xi Ii ~ n} is orthogonal to xn + I for every n.
It follows that the elements XI"'" Xn of an orthogonal system are
linearly independent.
Definition 4. A Banach space X is an ~ -space if there is an arbitrary
index set A and a family {Na laEA} of finite dimensional subspaces,
directed by inclusion, whose union is dense in X and such that Na is the
range of a projection .r:, of norm one of X. As a special instance, an JV 1-
space is simple if A is the set of positive integers, Nl C N2 c··· c Nnc ... and
dimNn=n.
2. Orthogonality, simple .AI; -Spaces and Monotone Bases 61
Evidently, a simple ~-space is separable. It is known that Lp-

spaces (1 ~ p < (0) and the space CCS), where S is a compact metric
space, are A~ -spaces (MICHAEL and PELCZYNSKI [1]) and that
Co, lp, Lp[O, 1J(1 ~p< (0) and separable Hilbert spaces provide examples
for simple JY; -spaces.
Theorem 5. (MAZUR) In each simple A~-space X there exists a
monotone basis for X.
Proof. There exists a set {rn of projections of unit norm of X such
00
that P1(X) cP2(X)c"',dimp;(X)=i and UP;(X) is dense in X. Let

i= 1
{x;} be a sequence of non-zero elements in X such that Xl EIl(X) and
XiEP;(X) "Ker P;-1 for i> 1, where Ker P; denotes the set {xEXIp;x=O}.
Moreover, let {a;} be an arbitrary sequence in C[>. Then
From this one gets inductively for n~m,
so that by Theorem 1.5 {x;} is a basis for X. {xt} is defined to be the

biorthogonal sequence to {xJ Finally, the basis is monotone, since
for XEX,
Ili~n Xi(X)Xi II ~ li~ Ili~n xi(x)xill
= Illi~ i~n xi(x)xill = Ilxll'
Theorem 6. If {x;} is a monotone basis for X, then {x;} is a total

orthogonal system.
Proof. If {Xi' xt} is a basis for X then the sequence {X;} is obviously
total in X. The monotony implies that II.I xiCx)xill ~ II. L xiCx)xill for
l~n l~n+ 1
each x in X. We assume YEsp{xili~n} and aEc[>, both arbitrary, and
take x=y+ax n+1. Since X= I XiCX)Xi and the x;'s are linearly
i~n+ 1
independent one has Y=Lxi(x)xi and a=x;+lCX). Hence

i~n
Ilyll ~ Ily+ax n+111, showing the orthogonality property. This completes

the proof of the theorem.
Theorem 7. A Banach space with a total orthogonal system is a simple
.A/~ -space.
Proof. Let {Xi} be a total orthogonal system in the Banach space

X and define Nn=sp{xdi~n}. Obviously, N ncN n + 1 , n= 1,2, ... and
dim Nn=n. Now by Nikol'skii's theorem, {xJ is a basis for X which,
clearly, must be monotone. Moreover, it follows that Ili~n xt(x)xill ~ Ilxll
for all n and X in X, whefl~ {xt} is the associated biorthogonal sequence
to {xJ Hence Pn:X --+X, given by Pn(x)= I Xt(X)Xi' XEX, is a
i~n
projection of norm one of X onto N n which implies that X is a simple

.;V i-space.
As a result of the preceding three theorems we have
Corollary 8. Ina Banach space X the following three statements
are equivalent:
(i) X is a simple .;V i-space.
(ii) There exists a monotone basis {xJ for X.
(iii) {xJ is a total orthogonal system in X.
3. Equivalent Bases
Definition 1. Let X and Y be Banach spaces, {x;} a basis for X and
{yJ a basis for Y. Then {Xi} and {yJ are said to be equivalent if for a
sequence {a;} c IP the convergence of I a i Xi in X is equivalent to the
convergence of I aiYi in Y. i~n
i:::;n
Theorem 2. The bases {xJ for X and {yJ for Yare equivalent if and
only if there is a topological isomorphism Tof X onto Y such that TXi= Yi'
i= 1,2, .... 00
Proof. Sufficiency. Let I aix i be the unique expansion for some

i= 1 00
X in X. Since T is a topological isomorphism, Tx= I ai TXi and this

00 i=l
expansion is unique, because I ai TXi=O implies x=O and hence
i= 1
ai=O for all i. Taking y= Tx and Yi= TXi' i= 1,2, ... , every Y in Y
00
thus has the unique expansion Y= I aiYi and the sufficiency is verified.
i= 1
Necessity. From the proof of Theorem IlLl.3 we know that X is topo-
logically isomorphic to a Banach space X' and that Y is topologically
isomorphic to a Banach space Y'; X' and Y' consisting of the vector
space of all sequences {aJ in IP for which lim I aixi and lim L aiYi
n i~n n i~n
respectively, exist, and which have norm s~p Ili~n aixil and s~p Ili~n aiYil1
respectively. The hypothesis implies that the identity mapping l' of X'
3. Equivalent Bases 63
into Y' is onto. Since the vanishing of the norm of {IXJ in Y' implies
IX; = 0 for all i, hence implies the vanishing of the norm of {IXJ in X', l' is
one-to-one. The following argument leads to the conclusion that l'
is closed: Let {IXJ, {PJ, {lXmJEX', m= 1,2, ... be such that
li~s~p 11;~n (IXm;-IX;)x;II=O and that li~s~p 11;~n (IXm;-P;)Y;II=O. Since
{xJ and {yJ are bases, uniqueness shows that xn,Yn#O for each
fixed n. For every e>O we then have an m for which
s~p max {11;~n (IXm; -1X;)x;11 ' 11;~n (IXm; - PJ Y;II} < e/2 .
Therefore,
1
IlXn-Pnl ~ IlXmn-lXnl + IlXmn-Pnl ~ Ilxnllll(lXmn-lXn)xnll
1
+ IIYnllll(lXmn-Pn)Ynll
~e/llxnli +e/llYnll
and we have {IXJ = {PJ, i. e. l' is closed. But l' is defined on the whole
of X', hence l' is bounded (1.2.10). From these properties of l' one con-
cludes that X' and Y' are topologically isomorphic so that X is topologi-
cally isomorphic to Y (1.2.6). Finally, since the topological isomorphisms
Tx and Ty of X (of Y) onto X' (onto Y' respectively) are such that
T x x;={c5ij} and that Ty y;={c5ij} it is clear that the transformation
T: X --+ Y of the theorem can be defined by Tx = Ty- 11' Tx x, X E X,
hence such that Tx; = Y; for all i and the theorem is verified.
Remark. The theorem remains true if X and Yare supposed to be
locally convex complete linear metric spaces (see ARSOVE [5J, Theorem 1).
Theorem 3. Let {x;,xt} be a basis for x. If the sequence {yJ in X
00
satisfies I Ilxtllllx;-y;il<l, then {y;} is an equivalent basis for sp{yJ

;= 1
Proof. For i~m one has for arbitrary IX;'S in <P,

IIX;I = I.I xt(lXjx)1 ~ II.I IXjXjllllxtll.
J~m J~m
Thus
m
.I lX;y;11 ~ II.I lX;x;11 + .~ 11X;illx;- y;il

II l::::;m l~m 1-1
~ II.I IX;X;llll +.~ Ilxtllllx;- Y;III

l~m 1-1
~211;~m lX;x;ll·
By Theorem 1.5 there is a constant M ~ 1 such that
for any m, nand m:::; n. Hence
n
:::;2M(II.I
l::S::n
OC;y;II+.I loc;lllx;-y;lI)
1= 1
00
:::;2MII.I oc;y;II+2MII.I oc;x;II.I

l~n ,=
Ilxillllx;-ydl
l~n 1
:::;KII;~nOC;y;II'
where
Again by Theorem 1.5, this implies that {y;} is a basis for sp{y;}. Since
the above estimates show that
ltm oc;y; II:::; 211;tm oc;x;11
:::; :11;tmOC;y;lI,
it !s clear that {x;} is equivalent to {yJ
Theorem 4. Let Yand Z be closed linear subspaces of X and let {y;}
and {z;} be basesfor Y and Z respectively. Moreover, let P be a projectiOli
of X on Yand let {yi} be the associated biorthogonal sequence to {yJ If the
00
condition IIPII I Ilyilllly;-zdl < 1 is verified, then there exists a projec-

;= 1
tion of X on Z.
Proof. Let T: X -+ X be defined by

00
Tx=x-Px+ I yi(PX)Zi' XEX.

;= 1
exist. From
III - Til =SUp{lIpX- i~l Yt(PX)Zilll xEX,llxll ~ I}

=SUP{lli~l Yt(PX)(Yi-Zi)111 xEX,llxll ~ I}
00
~IIPII L IlytIIIIYi- zill<l,

i= 1
it follows that 0< I-III - Til ~ IITII ~ 1 + III - Til <2. Thus T is a topo-
logical isomorphism of X onto itself (1.2.15). From the definition of Tit
is easy to see that T(Y)c:Z. Moreover, T has the property that {TyJ
00
={z;}. Because for any ZEZ with Z= L (XiZi, L (XiYi= L (XiT-lZi

i= 1
= T- 1 L (XiZi converges with n to an element, say y, of Y and since
i~n i~n
i~n
then Ty=z, Tmaps Yonto Z. Finally, by TPT- 1 TPT-1=TPT- 1,
it follows immediately that T P T - 1 is the desired projection of X on Z.
Definition 5. Let {Xi} be a basis for X, {(Xi} a sequence in ifJ and {Pn}
a strictly increasing sequence of positive integers. A sequence {Yn} of non-
Pn+l
zero elements in X, given by Yn= L (XiXi, is called a block basis with

respect to {x;}. i=Pn+ 1
Theorem 6. A block basis {Yn} with respect to a has is {Xi} for X is

a basis for SP{Yn}.
Proof. This follows immediately from Theorem 1.5.
Theorem 7. (BESSAGA-PELCZYNSKI) If {Xi,Xt} is a basis for X and
if there is a sequence {Yn} in
X such that infllYnl1 >0 and limxt(Yn)=O
n n
for all i, then there exists a subsequence {Yp) of {y n} which is a basis for
sp{Yp) and which is equivalent to a block basis with respect to {x;}.
Proof. By hypothesis there exists an e>O with infllYnl1 ~e and
n
strictly increasing sequences of integers {Pn} and {qn} such that for n ~ 1
(starting with ql = 1),
and
8MII.
l=qn+l
I +1
X{(YpJxill ~8/2n,
where M ~ 1 is the constant which occurs in Theorem 1.5. Let
qn+l
Zn= I X{(YpJx i, n~ 1.
i=qn+ 1
Then we have for n ~ 1,
8~ IIYPn l1 = IIJI X{(YpJXi I

~II.~ X{(ypJXil +IIZnll+II._ ~ + X{(ypJXill~8/8+IIZnll.
1-1 l-qn+l 1
Therefore infllznll >8/2. Moreover,

n
28 8
I2-
00
~- n =-.
8M n=l 4M
Since {zn} is a block basis with respect to {x;}, there exists to {zn} a
biorthogonal sequence {z;} in sp{zn}*. Now, using Theorem 1.5 (cf.
also Corollary III.2.3), one has for all m,
Ilz~11 = sup{lz~(z) Illzll ~ 1,zEsp {zn}}
=2Msup{lz~(z)112Mllzll ~ 1, ZESP{Zn}}
~2M SUP{lz~(z)l!m~n Z{(Z)Zili + Ili';~-l Z{(z)zill)!O #ZESP {zJ}

~2M/llzmll <4M/8.
This shows that
00
n=l
Hence by Theorem 3, {YPJ is a basis for sp{YpJ which is equivalent to
the block basis {zn}.
Corollary 8. Let Y be an infinite dimensional closed linear subspace
of X and let {x;} be a basis for X. Then Y contains an infinite dimensional
closed linear subspace with a basis which is equivalent to a block basis
with respect to {xJ
Proof. Let {xn be the biorthogonal sequence belonging to {xJ For

m
each n there exists a y" in Y of the form Yn=lim

m
L (XiXi, (XiEP, such that
.
l=n
IIYnl1 = 1. Otherwise, for some n, we would have xt(Yn)=O for all i>n,
Thus there is a sequence {Yn} of unit norm in Y such that xt(Yn) = for
n>i, hence such that limxt(Yn)=O for all i. To complete the proof it
°
i. e. Yc sp {xAj:~ n}, and this is impossible since Y is infinite dimensional.
n
remains to apply Theorem 7.
The next two corollaries are consequences of the following interesting
lemma which is due to Banach and Mazur.
Lemma 9. If X is a separable Banach space, then X is isometrically
isomorphic to a closed subspace of the Banach space C[O,I].
Proof. Let U* be the unit ball of X*. Then, by Theorem I.3.21, U* is
sequentially compact in the weak* topology of X*. Hence there exists
a continuous transformation of Cantor's triadic point set P in [0,1], onto
U* (HAUSDORFF [1], p. 134 and 197). Since T is continuous on P we can
extend it linearly to the whole interval [0,1], the result T' being a continu-
ous transformation on [0,1] onto U*. Therefore, T'(x)EC[O,I] for
every x in X. Now, there is a point tin P such that IT'(x)1 = Ilxll (I.3.10),
where T,' is the functional in U* corresponding to the point t. But since
II T'(x)11 = sup{1 T,'(x)lltE[O,I]} ~ Ilxll we have II T'(x)11 = Ilxll. Evidently
T":X--?C[O,I], defined by T"x=T,'(x), tE[O,I], is then linear and
isometric (i.e. IIT"xll=llxll for all x in X). This shows finally (1.2.15)
that T" is an isometric isomorphism and that T"(X)cC[O,I].
Corollary 10. Each infinite dimensional Banach space X contains an
infinite dimensional subspace with a basis.
Proof. Without loss of generality we may assume that X is separable.
By the foregoing lemma, there exists an isometrical isomorphism T of
X into C[O,I] and, by Theorem III.7.5 this space has a basis. From the
preceding corollary we know that T(X) contains an infinite dimensional
linear subspace Y with a basis, say {yJ Since T is an isometric isomor-
phism, it is clear that {T- 1 yJ is a basis for the infinite dimensional
subspace T- 1 (y) of X.
°
Corollary 11. If a sequence {Yn} in a separable Banach space X conver-
ges weakly to and infllYnl1 >0, then a subsequence {YPJ of {Yn} forms
a basis for sp {y pJ. n
Proof. Let T be the transformation used in the proof of the above
corollary and let {zn be the associated biorthogonal sequence of a basis
for C[O,I]. Since Tmaps X into C[O,I], T* maps C*[O,l] into X*. Thus
lim zt(TYn) = lim T* zt(Yn)=O for all i, for Yn converges weakly to zero.
n n
Due to infll TYnl1 = infllYnl1 >0 and Theorem 7 there exists a subsequence
n n
{YPJ of {Yn} such that {Typ,,} is a basis for sp{TYpJ. This finally shows
that {YPJ is a basis for sp{YpJ.
References for Chapter IV: BESSAGA [1], BESSAGA and PELCZYNSKI [2],
DAY [3], GELBAUM [1], GRINBLYUM [1], JAMES [1-3], LINDENSTRAUSS [1],
MICHAEL and PELCZYNSKI [1], NIKoL'sKII [1] and SINGER [12].
CHAPTER V
Bases and Structure of the Space

From the hypothesis of the existence of a basis of a certain type for
a Banach space X, conclusions can be drawn on the structure of X, e. g.
properties of X like weak sequential completeness, separability, re-
flexivity, weak conditional compactness and the dimension (finite or
infinite). In the first section some results are established on the first two
properties of X or of X*. The following paragraph has to do with re-
flexivity. In both sections the hypothesis of an unconditional basis plays
an important role, while in the last paragraph criteria for finite dimension
of X are given in terms of absolutely convergent and uniform bases for X.
1. Bases, Completeness and Separability
Theorem 1. A Banach space X with a basis is separable.

Proof. Evidently, a basis for X is a total sequence in X. Therefore, by
Theorem 1.1.11, X is separable.
Corollary 2. If X* possesses a weak* Schauder basis, then X is sepa-
rable.
Proof. The corollary is an immediate consequence of the above
theorem and of Theorem IlL2. 7.
Theorem 3. If X has a boundedly complete unconditional basis, then
X is weakly sequentially complete.
Proof. Let {Xi,Xt} be the basis for X and let {Yi} be a weakly con-
vergent sequence in X (which must not necessarily converge in the
weak topology of X). Then there is a constant K such that st1;pllyj ll ~K
)
(1.3.15). Next, we define rxi=limxi"(y) for all i. From this it follows that
)
li?t~n(rxi-xi"(y))xill =0

70 v. Bases and Structure of the Space
for all n. Thus t~n(XiXill~s~plli~nxi"(Y)Xill~Ms~pIIYjll~MK, where

M = supll Unil and Un are the expansion operators of the basis {x;} for
n
X. Since {xJ is boundedly complete, I (Xi Xi converges to an element
x in X and one has (Xj=xi"(x). j~n
Now, if Yj does not converge weakly to x, there is an x* in X* of
norm one and for some £>0 there is a subsequence {yj} of {yJ such
that with Zj=x-yj, Rex*(z»e for allj. Again since {x;} is a basis
for X, and due to hypothesis, there are for every £' > 0 increasing se-
quences {nJ and {mj} such that Ilznj-Umjzn)l<e' and IIUmjznj+lll<e'.
Furthermore, let {fJJ be any sequence in ,p. Using Lemmas IIL4.2 and
3, we obtain with a constant C, 0 < C ~ 1, taking U rna = 0,
I jt/jZnj I ~ I jt/j(Umj - Umj_)znj 11- tt/j(Zn j - UmjZn) I
- L t / j Umj-l Znj I
>cll.±
J=l
'fJ)(Umj-Umj_,)Znjll-2e'.~ IfJ) J=l
n n
~C I IfJ) Rex* [(U mj - Umj_,)znjJ - 2e' L IfJ)
j=l j=l
~C I IfJjl [Rex*(zn)-llznj - U mjzn)I-11 Umj-l zn)IJ

j= 1
n
>[C£-2(C+1)e'J I IfJJ
j= 1
But
We can take e' < £ Cj2(C + 1) so that the linear transformation T :Ic-'X
(naturally, 11 is over the same field as X), defined by T{y;} = lim "YjZn,,
n L..
i~n
{y;} E 11> is bounded. Since II T {y;} II ~ [C £ - 2(C + 1)£'J II{ y;}ll, {y;} E 11 ,
T has a bounded inverse (1.2.15). Hence T is a topological isomorphism
of 11 with T(11). T(ll) is then weakly sequentially complete, because 11
is (l.4.b, cf. also proof of Corollary V.3.1). Since by the Hahn-Banach
1. Bases, Completeness and Separability 71
theorem every f in T(lI)* has a bounded linear extension to X, since

{znJcT(/ 1 ) (due to znj=T{bj;}), and since by hypothesis liJPx*(zn)
J
exists for all x* in X*, liJP f(zn) exists for all f in T(/l)*' Hence there
J
is a Z in T(lI) for which limx*(zn)=x*(z), X*EX*. But then one has
J
z=lim
n i~n
I X{(Z)X; = lim
II
I
i~n}
x;liJPx{(zn) = lim
n
I
i~n
x;[x{(x)-limx{(y~)]
J
= lim
n
I
i~n
x;[X{(x) -liJP X{(Yj)] = lim
] n
I
i~n
x;[rx; -lim X{(Y)] = O. Conse-
J
quently, contrary to our assumption, y~, and hence yj' must converge
weakly to x. Thus X is weakly sequentially complete and we are done.
Applying Corollary III.4.7 we immediately obtain
Corollary 4. If X possesses an unconditional basis and if X* is sepa-
rable, then X* is weakly sequentially complete.
A further corollary one gets, from the combination of the above with
Theorem III.4.8.
Corollary 5. If {x;, xi} is an unconditional basis for X, then {xi} is
an unconditional basis for X* if and only if X* is weakly sequentially
complete.
Theorem 6. If X has an unconditional basis, then X is weakly sequen-
tially complete if and only if no subspace of X is topologically isomorphic
with co.
Proof. The necessity is proved just in the first part of the proof for
Theorem III.4.5. The sufficiency is clear by Theorem 3 if the condition
that no subspace of X is topologically isomorphic with Co implies that
each unconditional basis for X is boundedly complete. But this impli-
cation is proved exactly in the second part of the proof for Theorem
II1.4.5 and this concludes the proof of our theorem.
Theorem 7. If X has an unconditional basis, then the basis is shrinking
if and only if no subspace of X is topologically isomorphic to 11'
Proof. Sufficiency. We need only to show that if the basis is not
shrinking, then there is a subspace of X which is topologically isomor-
phic to 11' But this is established exactly in the principal part of the
proof for Theorem II 1.4.4.
Necessity. We assume that the basis is shrinking. Then Theorem
III.3.4. warrants the existence of a basis for X* and, evidently, X* is
separable. Now, if a subspace of X is topologically isomorphic to 11'
then by (I.3.8 and 25), a factor space of X* is topologically isomorphic
to the non-separable space Ii (cf. 1.4.c). By this contradiction no subspace
of X can be topologically isomorphic to 11 and the theorem now follows.
6*
72 v. Bases and Structure of the Space
2. Bases and Reflexivity
Theorem 1. Let {Xi> xn be a basis for X such that sp {xn = X*.

Then X is reflexive if and only if the basis is boundedly complete.
Proof. By Corollary III.3.6 it is clear that {xi,xn is a shrinking
basis for X and according to Corollary IIL3.5, {xi", J x;} then is a basis
for X*. Now let X be reflexive. Since sp{x i} =X this implies that
sP {J x;} = X** so that {xi", J x;} is also shrinking. Thus, by Theorem
III.3.9, {J Xi} is a boundedly complete basis for X**. Since X is reflexive
and J is an isometric isomorphism of X onto X**, {x;} is a boundedly
complete basis for X.
Conversely, let {Xi' xn be boundedly complete and let U and U**
be the unit balls in X and X** respectively. Because J(U) is weak* dense
in U** (1.3.22), it follows that for every x** in U** there is a sequence
{Yn} in U such that x**(x*)= limJ Yn(x*) for each x* in X*. Thus
, n
1:: x**(xt)xi=lim 1:: JYn(xt)xi=lim 1:: xi"(Yn)xi=lim UjYn and by Cor-

i~j n i~j n i~j n
ollary 111.2.3, Ili~j x**(xt)xill~s~PIIUnll<CX). But the bounded com-
pleteness of the basis then implies the existence of an x in X such
that x = lim L x** (xt) Xi. Since (x** - J x) (xj) = x** (xj)
n i~n
-lim 1:: x**(xt)J x;(xj)=x**(xj)-lim 1:: x**(xt)xj(xi)=O for all xj

n i~n n i~n
and since sp{xn=X*. we have X**EJ(X) for every x** in X**. Hence
X is reflexive as we wished to prove.
Theorem 2. (JAMES) A Banach space X having a basis is reflexive if
and only if the basis is both shrinking and boundedly complete.
Proof. Let {Xi' xn be a shrinking basis for X. Then, by Corollary

IIL3.6, sp {xn = X*. If the basis is also boundedly complete, by the
last theorem, X must be reflexive. On the other hand, suppose that X is
reflexive and has a basis {Xi' xi"}. From Corollary IIL2.9 then follows
that {xi", J x;} is a basis for X*. On account of Theorem III.3.4, {x;}
is shrinking. Clearly sp {xn = X*. Hence Theorem 1 implies that
{Xi' xn is boundedly complete and the theorem is proved.
Theorem 3. (GELBAUM) If X has a basis {Xi' xi"}, X* is weakly
sequentially complete and 1::
x**(xt)x i converges in X for all x** in
X**, then X is reflexive. i";n
Proof. We take an arbitrary Y*EX* and define the sequence {y:} by
Y: = 1:: Y*(Xi)XT- By hypothesis
i~n
2. Bases and Reflexivity 73
x**(y:)= L Y*(Xi)X**(Xt) = y*( L X**(Xt)Xi)

i~n i~n
converges for each x** in X** to x**(z*), where z* is an element

in X*. But one has
n
(L xt(x)xi) = y*(x)
z*(x)=Jx(z*)=lim Jx(y:) = lim y*
n i~n
forallxinX,implying y*=z*. Hence x**(y*)=li~x** (.L Jxi(y*)x t )

l~n
for each x** in X** and since Jxixt)=xt(X)=Oii' {xt,Jx} is a

weak basis for X* which, by Corollary 111.2.5, is a basis for X*. But
{xt,Jx;} is a retro-basis for X* and we see, using Theorem 111.2.12
that X is reflexive.
Theorem 4. If X* has a retro-basis {xt} with associated biorthogonai
sequence {Jx;} and if L x**(Xt)Xi converges in X for all x** in X**,
then X is reflexive. i';;n
The theorem is a corollary to Theorem 111.2.12.
Let J' be the natural embedding of X* into X***, defined in the
same manner as the natural embedding J:X-+X**
Lemma 5. There is a projection of norm one of X*** on J'(X*).
Proof. We use the linear transformation Q:X***-+X*, given by the
condition Qx***(x)=x***(Jx), x***eX***, xeX. Evidently, IIQx***1I
=sup{lx***(Jx)llxeX, Ilxll ~ 1}~sup{lx***(x**)llx**eX**, Ilx**1I ~ 1}
= IIx***II· Since QJ' x*(x)=J' x*(Jx)=Jx(x*)=x*(x), x*eX*. xeX, Q
is onto and since for all x* in X*, IIQJ'x*II=lIx*ll, we have IIQII=l.
Now, let P=J' Q. Then P(X***)=J'(X*). Finally, because QJ' x* =x*
for every x* in X*, we obtain p 2 x***=J'(QJ')Qx***=J'Qx***
= P x*** and P is a projection. Since IIPII ~ IIJ'III1QII = 1 and since
liP x***11 = IIx***II, x*** e P(X***), P is of norm one.
Based on this lemma we have
Theorem 6. If X** has a retro-basis and iffor every x** in X** there
is a unique x**** in X**** for which Ilx****11 = Ilx**1I such that x**(x*)
=x****(J' x*) for all x* in X*, then X is reflexive.
Proof. Let {xt*, J' xt} be a retro-basis for X**. Then Theorem
III.2.12 implies that {xt, xt*} is a basis for X*. Since by the preceding
lemma for every X there is a projection of unit norm of X*** on J' (X*),
the assumptions of Theorem IV.1.4 are satisfied. Hence L x***(xt*)xt
i~n
converges in X* for all x*** in X***. As a consequence ofthe foregoing

theorem, X* is reflexive. But this is equivalent to the reflexivity of X
(1.3.18), and the theorem is proved.
74 V. Bases and Structure of the Space
Theorem 7. If X has an unconditional basis and X** is separable,

then X is reflexive.
Proof. The separability of X** implies that of X* (I.3.11). Then, on
account of Corollary 1.4, X* is weakly sequentially complete. Hence
the unit ball U* in X* is conditionally weakly sequentially complete.
Let {y:} be a sequence in U*, weakly converging to a point y* of X*.
Then there is a y** in X**, Ily**11 = 1, such that. lIy*11 = y**(y*) (I.3.1O).
But y**(y*)=limy**(y:)=::::;suplly:lI=::::;l implies that U* is weakly
n n
sequentially complete. This fact, combined with the hypothesis, yields
(I.3.19) that X* and hence X is reflexive.
Theorem 8. If X has an unconditional basis and if no subspace of X
is topologically isomorphic to 11, then every bounded set in X is sequentially
compact in the weak topology for X.
Proof. We take an arbitrary bounded sequence {Yn} in X and define
{xJ to be an unconditional basis for X with associated biorthogonal
sequence {xt}. Since sup IxnYn)1 =:: :; IlxT11 sup IIYnl1 < 00 and since every
n n
bounded closed subset of ~ is sequentially compact in the usual topology
for ~, one can choose a subsequence {zn} of the sequence {Yn} such
that there exist the limits
lXi=limxT(zn), i= 1, 2, ....
n
If {Pn} and {qn} are arbitrary increasing sequences of positive integers,

then
suPllzPn -ZqJ =::::;2supllYnll < 00
n n
and
Suppose now that {z Pn - ZqJ is not weakly convergent to O. Then there

isanx*inX*with Ilx*ll=l and a subsequence {w n } of {zpn-ZqJ such
that infllwnll~inflx*(wn)I>O. The definition of {w n } implies limxT(wn)=O
n n n
for all i. Thus, by Theorem IV.3.7, a subsequence {v n } of {w n } is a
basis for sp {v n }, equivalent to a block basis with respect to the basis
{x;}. It is clear that the basis {v n } is unconditional, since the basis {x;}
is unconditional.
For IX = {IX;} E11' lim L lXi Vi exists, since sup Ilvnll =:: :; sup Ilwnll
n i~n n n
=:: :; sup IlzPn - ZqJI < 00, and the limit element is in sp {v n }. On the other
n
hand, if I lXi Vi is convergent, since the basis {v n } is unconditional,
i~n
2. Bases and Reflexivity 75
co
we have L 100jllx*(vj)1 < 00 as a result of Riemann's theorem (II.2.2).
j=l
But because inflx*(vn)I>O this is true only if 0( is in 11 , Therefore we
n
can define the linear transformation T of 11 onto sP {v n}, by TO(
= lim L O(j Vj, 0( Ell' Evidently II Til::::;; sup II vnll < 00. The fact that {vn}
n i~n n
is a basis for sp{Vj} and the assumption TO(=O for any 0( in 11 imply
O(j = 0 for all i, hence that 0( is the zero element of 11 , Thus T is one-to-one,
and, by (1.2.6), is a topological isomorphism. That a subspace of X is
topologically isomorphic with 11 contradicts our hypothesis, thus
{zPn -ZqJ converges weakly to O.
Therefore the sequence {zn} converges in the weak topology for X,
because otherwise there would exist a y* in X* and sequences {Pn}
and {qn} of increasing integers such that infly*(zpn -ZqJl >0 which,
n
by the above results, is impossible. Thus every bounded sequence in X
contains a subsequence which converges in the weak topology for X
and the proof of the theorem is complete.
Theorem 9. In every separable non-reflexive Banach space X there
exists a subspace with a non-shrinking basis.
Proof. By (1.3.18), the unit ball U of X can not be weakly sequentially
compact. We thus have two possibilities: (i) X is not weakly sequentially
complete. In this case there exists a weakly convergent sequence {xn}
in X which has no weak limit in X. (ii) X is weakly sequentially complete
but U is not weakly sequentially compact. If this is the case, there is a
sequence {xn} in U such that no subsequence of {xn} is a weak Cauchy
sequence.
(i) Without loss of generality one may take Xl =0. Let J(J') be the
natural embedding of X(X*) into X** (into X*** respectively). Since
supllxnll<oo (1.3.15), we can define xi*EX** by xt*(x*)=limx*(x n),
n n
X*EX*. Obviously xt*¢J(X). Next, let x:*=xt*-Jxn, n=2,3, ....
It follows that supllx:*II::::;;llxt*ll+supllxnll<oo and, since xt*¢J(X)
n n
that infllx:*11 =infllxt* -J xnll >0. Since Ilx**11 = sup{lx**(x*)I\llx*ll::::;; 1,
n n
x*EX*}=sup{lx***(x**)llllx***II::::;; 1, X***EJ'(X*)}, J'(X*) is a deter-
mining manifold for X**. Moreover, lim J' x*(x:*) = xt*(x*) -limJ xix*)
n n
=xt*(x*)-limx*(xn)=O for all x* in X*. Thus Theorem IV.1.6 implies
n
theexistenceofasubsequence {X:k*} of {xt*,x!*,x~*, ... } with n 1 =1,
· h 'IS a b
W h IC ' rlor sp
aSIs - {**} - {**
Xnk =Sp Xl , Xl** - J Xnz , Xl** - J Xn3 ' ... }
=sP {xt* J x n2 ,J x n3 ,···}·
Let now Y=sP{x nz 'x n3 , ... } and Z=sP {X:k*+.} =sp{xt*-Jxn2 ,
xt*-Jxn3""}' Furthermore, let l' be the identity in B(sp{x:k*}) and
76 v. Bases and Structure ofthe Space
let U 1 be the first expansion operator of the basis {X:k*}. U 1 is a con-

tinuous projection (III.2.3) with one dimensional range spanned by xT*.
We then define the bounded linear transformation T: Y-+Z by
Ty=(1'-U 1)Jy,YEY. T is one-to-one since for YEY, O=Ty
= (1' - U 1) J Y implies that J Y is in the range of U 1 which is possible
only if y=O. T is onto, since any ZEZ has the unique representation
as a sum z=axT* +J Y, aEcfJ, yE Y (according to xf*¢J(Y) one has
axT*=Jy=O whenever axT*+Jy=O) so that z-axT*EJ(Y) and
Tr 1(z'-axT*)=(l'-U 1)(z-axT*)=z. It thus follows from (1.2.6)
that T is a topological isomorphism of Y onto Z.
x:
Moreover, by (IV.l.5) it is clear that {T- 1 k*+,} is a basis for Y,
which is non-shrinking as we will show by the following argument: If
{T- 1x:k*+,} would be shrinking, then tbis would imply that limy*(T- 1X:k*)
k
=0 for all Y*EY* , hence that limz*(x**)=limT*z*(T-
k"k k
1x**)=0
nk for
every z* EZ*. Since the restriction to Z of every x*** in X*** is
automatically in Z* and since then lim x*** (J xnJ = x*** (xT*),
k
X***EX***, we have by (1.3.15) the result that xT*EJ(X) which is
impossible. This shows that under assumption (i), {T - 1 k*+,} is ax:
non-shrinking basis for the subspace Y of X.
(ii) Let {gm} be a countable dense set of non-zero elements in X.
For every m there is a g!EX* with IIg!1I = 1 and g!(gm) = IIgmll (I.3.10).
Then r=sp{g!} in X* is a determining manifold for X, since for XEX,
sup{lx*(x)llllx*1I ~ 1,X*Er} ~ Ilxll ~inf(llx-gmll + Ig!(x)1 + Ig!(gm-x)l)
m
~suplg!(x)1 ~sup{lx*(x)llllx*11 ~ 1,X*Er}. Next, the following diago-
m
nal procedure gives a subsequence {Yn} of {x n} such that limg!(Yn)
n
exists for m= 1,2, .... Since xT(U) is sequentially compact in cfJ (1.1.12
and 9), there is a subsequence {X~1)} of {x n } such that limgHx~1)) exists
n
and we take Y1 =X\1). Similarly, there is a subsequence {X~2)} of
{X~1),X~1), ... } such that limgHx~2») exists and we take Y2=X\2). Con-
n
tinuing in this way it is easy to see that {Yn} has the requested properties.
Since {Yn} can not be a weak Cauchy sequence, there is an X*EX* of
norm one and there are increasing sequences of integers {Pk} and {qk}
such that i~flx*(Ypk-YqJI>O. The sequence {zn}={YPn-yqJ fulfils
the hypothesis of Theorem IV. 1.6, because O<inflx*(zn)1 ~infllznll
n n
~suPllznll ~2suPIIYnll ~2
n n
and limg!(zn)=
n
limg!(Y
k
Pk - yqJ=O, m= 1,2, ...
(we observe that, given g* Erand 8> 0, by the last inequalities and by
r=sp{g!}, there is an m such that Ilg*-1:11 <8, where f! is in
sp{gT, ... ,g!}, and an no depending on m such that If!(zn)I<8,
n~no. Hence Ig*(zn)I~21Ig*-f!II+lf!(zn)I<38, n~no and one has
3. Criteria for Finite Dimension 77
limg*(zn)=O for all g*Er). Thus a subsequence {znJ of {zn} forms a

n
basis for Sf> {znJ. Finally, due to inflx*(znJI >0, this basis is not shrink-
k
ing and the theorem now follows.
It is known that a Banach space X is reflexive if and only if every
separable closed linear subspace of X is reflexive. We can now prove the
following stronger proposition:
Theorem 10. (PELCZYNSKI) A Banach space X is reflexive if and only
if every subspace of X with a basis is reflexive.
Proof. By (1.3.18) and Theorem 1.1, the necessity is obvious. On the
other hand, let us assume that X is not reflexive. Then, again by (1.3.18),
there would be a separable closed linear ~ubspace Y of X which is not
reflexive, since otherwise X itself would be reflexive. Furthermore, the
preceding theorem would imply the existence of a subspace Z of Y with
a non-shrinking basis. But by hypothesis Z must be reflexive so that by
Theorem 2 the basis for Z must be shrinking. This contradiction shows
that X is reflexive and we are done.
The next corollary is immediate:
Corollary 11. In every non-reflexive Banach space there exists a sub-
space with a non-shrinking basis.
3. Criteria for Finite Dimension
We observe that if X is infinite dimensional and has an absolutely

convergent basis, then, as a result of Theorem 111.5.2, X is topologically
isomorphic with 11 • Thus, as a corollary of this theorem we obtain
Corollary 1. If X has an absolutely convergent basis and X* is weakly
sequentially complete, then X is finite dimensional.
Proof. Supposing X to be infinite dimensional, then it is topologi-
cally isomorphic with 11 • Hence by (1.3.25) X* is topologically isomorphic
with IT and thus with 100 (1.4.c). Therefore, 100 would also be weakly
sequentially complete. This is easy to see by the following argument. Let
T be a topological isomorphism of a weakly sequentially complete
Banach space Y onto a Banach space Z and let {Zk} be an arbitrary
weakly converging (not necessarily in the weak topology of Z) sequence
in Z. Then lim T* z*(T- 1 zk)=limz*(zk) exists for all z* in Z*. Since T*
k k
is a topological isomorphism of Z* onto y* (1.3.25), T- 1(Zk) converges
weakly in Y, to an element y of Y because Y is weakly sequentially
complete. The weak sequential completeness of Z then follows from
78 V. Bases and Structure of the Space
limz*(Ty-zk)=lim T* z*(y- T- 1 Zk)=O, Z*EZ*. Finally, the result that

k k
100 is weakly sequentially complete leads to a contradiction (l.4.c) so that
X is of finite dimension.
Corollary 2. Let X have an absolutely convergent basis and let X*
be separable. Then X can not be infinite dimensional.
Proof. In just the same way as in the proof of Corollary 1 it can be
shown that 100 would be separable under the hypothesis stated in this
corollary. But this is not the case (I.4.c). Therefore, X is finite dimensional.
References for Chapter V: BESSAGA and PELCZYNSKI [3], DAY [2],

GELBAUM [1], JAMES [4], KARLIN [2], PELCZYNSKI [4] and TAYLOR [1].
CHAPTER VI
Bases for Hilbert Spaces

The separable Hilbert spaces are the only class of abstract spaces for
which the existence of a Schauder basis is warranted. It is indeed a
classical fact that in every separable Hilbert space H there exists a total
orthonormal sequence, hence an orthonormal basis for H. It follows that
a basis for H is orthonormal if and only if it is normal and that such a
basis is unconditional. On the other hand a basis {Xi' xi"} for H is
ao ao
unconditional if and only if both L I(X,Xi)1 2 and L l(x,xrW exist for
every x in H. i=1 i=1
1. Monotone and Orthonormal Bases
Let H be a separable Hilbert space.

Theorem 1. There exists a monotone basis for every separable Hilbert
space.
Proof. Since H is assumed to be separable there exists a sequence
{Xi} which is a dense set in H. Defining N n =sp{xili::::;n},n=1,2, ... ,it is
clear that the sequence {N n} of finite dimensional subspaces of H is
directed by inclusion and that sp {N n} = H. Without loss of generality
one may suppose that every finite collection of elements of {Xi} is lin-
early independent. Thus, dimNn=n. Moreover, the orthogonal pro-
jections of H on N n all have norm one. Consequently, H is a simple
.Af1-space and H has a monotone basis by Theorem IV.2.5.
Remark. As a consequence of Corollary 111.2.9, {Xi' xi"} is a basis
for H if and only if .{ xr, Xi} is a basis for H.
Definition 2. A basis {Xi' xi"} for H is called orthonormal if xr = Xi
for all i, normalized if IIXil1 = 1 for all i, and normal if Ilxrll = IIXil1 = 1
for all i.
80 VI. Bases for Hilbert Spaces
In this chapter we use the well known fact that the transformation
T:H-+H* defined by Ty(x)=(x,y), x,YEH, is a one-to-one additive
isometric map of H onto H* and that T(ay)=CiTy, aEiP, YEH. We
customarily denote the elements Ty and y by the same symbol. It is
evident that an orthonormal basis is normal, since then Ilx iI1 2 =(X i,X i)
= (Xi' xt)= xi"(x i) = 1 and in the same way one obtains Ilxfll = lo
Theorem 3. An orthonormal basis for H is monotone.
i~n i~n
is a non-decreasing function of n for every x in X.

We now have the following classical theorem:
Theorem 4. There exists an orthonormal basis {Xi} for Hand Par-

seval's identity Ilx112= L I(X,Xi)1 2 holds for every x in H.
i= 1
Proof. H, as a separable space, has a total sequence {zJ of elements

in H (1.1.11). Without loss of generality we may assume each finite subset
of {z;} to be linearly independent. To this sequence we now apply an
orthogonalization process, which, for sake of completeness, will be
shortly described here:
Let us take Xl = z dllz 111 and let us define recurrently Yn+ 1= Zn+ 1
n
- L (Zn+1'X;)X i, Xn+1 =Yn+dIIYn+lll, n=1,2, .... Obviously IIx111=1

i= 1
and (Y2'X 1)=0. Because Z2 and Xl are linearly independent we observe
that Y2 #0. Hence IIx211 = 1 and (X 2,X 1)=0. Repeating this argument
one obtains Ilxnll = 1 and (xn,x1)=0, n=2,3, .... By the same procedure
we can show that (xn,xm)=O, n>m. Therefore, (xn,xm)=b nm .
Clearly, the elements X1""'Xn are linearly independent. Hence the
subspaces spanned by X1""'Xn and by Zl""'Zn are the same, showing
that {Xi} is likewise a total sequence in H. Now, if {aJ is an arbitrary
sequence in iP, I L aixiI12=.L la;l2~.L lail2 =
l~n t~n l~m
II.L aixil12 for eachn,m,
l::::;m
n~m. It then remains to invoke the theorem of NIKOL'SKII (IV.loS) to
infer that {xJ is an (orthonormal) basis for H.
Finally, the continuity of the inner product shows Parseval's iden-
tity:., i.e. it shows that IlxI12=(li,?1i~n(x,xi)x;,x)=li,?1i~}x,xi)(xi'X)
i= 1
Corollary 5. Every separable infinite dimensional Hilbert space H is

isometrically isomorphic to [2'
1. Monotone and Orthonormal Bases 81
Proof. By Theorem 4 there exists an orthonormal basis {x;} for H.

Since for every x in H, by the same theorem, Ilx112= L l(x,xiW, the
i= 1
linear transformation T:H ->12 defined by Tx= {(X,Xi)}, x EH, is an isom-
etry (i. e. II Txll = Ilxll, XE H) into 12. Hence T is one-to-one and we prove
that T is onto: Let {IX;} be an arbitrary element in 12 • Then the series
i~l lXiXi converges to some x in H, because H is complete and II itp lXiXi 112
q
= L: IlXd 2• But since {x;} is a basis for H, by uniqueness, we have
i=p
lXi=(X,XJ for all i. Therefore, {IX;} is in T(H) and the theorem now
follows.
Definition 6. The unit ball U of a Banach space X is strictly convex
provided that Iltx+(l-t)yll<l whenever X,YEU, Ilxll=IIYII=l, x#y
and O<t< 1.
Lemma 7. If the unit ball of the conjugate X* of a Banach space X
is strictly convex and if Z is an arbitrary proper closed linear subspace
of X, then for each z* in Z* there is a unique norm preserving extension
x* in X*.
Proof. We suppose to have two norm preserving extensions x* and
y* to the whole space X (at least one such extension exists by the Hahn-
Banach theorem (1.3.9)). Then z*(z)=(tz*+(l-t)z*)(z)=(tx*
+(l-t)y*)(z), ZEZ and hence Ilz*II:::;lltx*+(l-t)y*ll. But if O<t<l
and x*#y*, one has Iltx*+(l-t) y*ll/llz*ll<l which contradicts the
foregoing inequality. Hence we must conclude that x* = y* as we wished
to prove.
Corollary 8. If Z is an arbitrary proper closed linear subspace of H,
then for each z* in Z* there is a unique norm preserving linear extension
onH.
Proof. One has only to show that the unit ball U of H is strictly convex.
Let X,YEU, Ilxll=IIYII=l, x#y and O<t<1. Then Iltx+(1-t)yI12
= t 2+ (1-t)2+ 2(t-t 2)Re(x,y)= 1- 2(t-t 2)+2(t-t 2)Re(x,y) = l-(t- t 2)
[2-2Re(x,y)] = 1-(t-t2)llx- y112< 1.
Theorem 9. A basis for H is orthonormal if and only if it is normal.
Proof. Let {xi,xn be a basis for H such that Ilxdl = Ilxrll = 1 for
all i. Moreover, let j be an arbitrary index and let P j be the orthogonal
projection on the one-dimensional subspace M j of H, spanned by Xj.
On account of (1.3.23), Pj is a topological isomorphism of Mj onto
Pj 1(0)1-. Next, it is clear by Corollary III.2.9 that Mj = sp(xj'), where

xj' is the restriction of xj to M j • It follows Ilxj'l1 = Ixj'(xj)I = Ixj(xj)1 = 1
=llxjll. Moreover, let yjEPj(O)1- be defined by yj=Pjxj'. Then
Ilyjll::::; IIPjllllxj'l1 = Ilxj'll. But Ilxj'll = Ixj'(x)1 = Ixj'(Pj x) I= IPj xj'(xj)I
= Iyj (x) I::::; Ilyjll implies that Ilyjll = Ilxjll. Since Pj is orthogonal,
Pi l(O)=Mt· Thus Pi 1(0)1-= Mt1-=Mj (1.1.16) and the equations
yj(xj)=Pj xj'(x) = xj'(PjXj) = xj'(x) = xj(xj) show that yj is an exten-
sion of xj' to H. Hence both xj and yj are norm preserving extensions
of xj' to H. By the preceding corollary norm preserving extensions to
H of xj' are unique. Thus yj = xj and because yj E M j' there is an IY. in
tP such that axj=xj. Finally, the relation 1 =xj(x)=(xj,axj)=lY.llxjI12
= IY. shows that IY. = 1, hence that xj = Xj' This implies that the basis
{Xi' xn is orthonormal and that (Xi' x) = x{(x) = (lij'
Since the converse of this proof is trivial (cf. below Definition 2) the
theorem is proved.
Example 10. {(Ii} is an orthonormal basis Jor the Hilbert space 12 ,
Proof. Clear by Theorem III.7.3.

Example 11. The sequence {x n}, given by x n(z)=(n+1)/n)1- zn, ZEC,
Izl<1, n=0,1,2, ... , is an orthonormal basis Jor the Hilbert space A2
(cf (l.4.g)).
Proof. Let D be the open unit disc in the complex plane C, D = {zllzl < 1}.
That {x n } is an orthonormal sequence is proved by evaluating
(xn' x m)=n- 1(n+ 1)1-(m+ 1)1- Jz nzmdJ1(z)

D
1 21<
=n- 1(n+ 1)1-(m+ 1)1- J J ei(n-m)",pn+m pdpdcp
o 0
= n- l(n+ 1)1-(m+ 1)1-(n + m+ 2)-12n(lnm= (lnm'

Now, let J be an arbitrary element in A2. Since the Taylor senes
L IY.nZn for J(z) converges absolutely and uniformly in each closed
n=O
disc Dr={zllzl::::;r}, with r<1, one gets
00 r 27t
JJ(z)x nCz)dJ1(z) = L IY.m(n+ 1)/n)1- J J ei(m-n)",pn+m pdpdcp
m=O o 0
00
= L IY.m(n+1)jn)1-(n+m+2)-12n(lnmrn+m+2
m=O
2. Unconditional Bases for Hilbert Spaces 83
Since f(z)xn(z} is integrable over D one has

(f,xn)=lim J f(z)x n(z}djl(z)=(Xn(n/(n+1»)t.
r~l Dr
Because (Xn Zn = (f, Xn)Xn(Z), zED, the series L (f, xn)xiz) converges
n=O
absolutely and uniformly to f(z) in each compact subset of D. The
00
application of Bessel's inequality shows that L Since
r ~pl(f,
l(f,xn)12~lIfIl2.
n=O
tt}f, xJxn = xnW, the series n~o (f, xn)xn converges strongly
to an element, say g, in A2. Moreover, since a subsequence of
Lto (f, Xn)Xn(Z)} converges to g(z) almost everywhere on D, one has
g=f, and by Theorem III.2.4, {xn' xn} is an (orthonormal) basis for A2.
2. Unconditional Bases for Hilbert Spaces
Theorem 1. An orthonormal basis for H is unconditional.

Proof. Let {Xi} be such a basis for H and let {ni} be an increasing
sequence of integers. Then in view of Parseval's identity (Theorem 1.4)
00
r
or in view of Bessel's inequality, for every X in H the series L I(x, XiW
is convergent. Therefore, since II itm (x, xni)xnf = itm I(x, Xnf)l~ 1 for each
00
m~n and since H is complete, the series L (x, Xi)X i is subseries, and
i=l
hence unconditionally convergent (Theorem II.l.3). This finishes the
proof of the theorem.
Next, let :E be the set of all finite subsets of the set of all positive
integers. We then have
Lemma 2. (ORLlCZ) Let S be a compact interval in ~ and let {Ii} be a
sequence in L 2(S). If sup {II ~.fi III jlE:E} <00, then one has s~p i~}.fi112<00.
Proof. We make use again of Rademacher's system {lJ'n}, defined

below Theorem II!.7.7. Bessel's inequality applied to L (XilJ1,(Xi E «:P,
i= 1,2, ... , yields i~n
This inequality implies that almost everywhere on S
!Ii~n fi(S)lJ'i(t) 12 d t .
1
i~n Ifi(S)1 2 ~
We now define the set
fl~ = {i IlJ'i(t)~O, i= 1, ... , n}, tE[O, 1J .
Integrating over S on both sides of the last inequality and applying the
Fubini-Tonelli theorem then gives
= S II.LlJ'i(t)j;112 d t
o l~n
~! [II i~ fill+ L~ fillTdt

~ 4sup {II ~ ffl flEL}<oo,
as we wished to prove.
The following theorem is based on this lemma:
Theorem 3. Let {Xi' xr} be a normalized basis for H. Then the basis
is unconditional if and only if L
I(x, x i)1 2 and L
I(x, xnl 2 converge for
each x in H. i:E;n i:E;n
Proof. "If part": Let x and y be arbitrary elements of H. From the
hypothesis and through application of Schwarz's inequality it follows
co
immediately that L I(x, xi") (y, x;)1 < 00. For every increasing sequence
(.L (x, x!)x y)
i= 1
of integers {m;}, therefore, li~ m ;' exists. H, as a
l~n
reflexive space, is weakly sequentially complete (1.3.17). From this we

co
infer that the series L (x, xi") Xi is sub series convergent in the weak
i= 1
topology of H. By Theorems II.l.2 and 3 the series is then unconditionally
convergent in H. Hence {x;} is an unconditional basis for H.
"Only if part": Let {Xi,Xr} be a normalized unconditional basis for
H. Because H*( =H) is separable and reflexive, Theorem III.4.6 implies
that {xt, x;} is an unconditional basis for H. By Corollary VI. 1.5, every
separable Hilbert space is isometrically isomorphic to l2 and hence also
with the Hilbert space L2 [0, 1] (of course the spaces are all over the
2. Unconditional Bases for Hilbert Spaces 85
same field tP). From the section below Definition 111.4.1 we know that
{II III
sup ~ (x, xT) Xi fl E 1: } < 00 for every x in H. Thus, taking
Ii = (x, xT) T Xi' where T is an isometric isomorphism of H onto L2 [0, 1],
the assumptions of the preceding lemma are fulfilled. Hence
sup L l(x,xT)1 2=sup L lI(x,xT)XiIl 2=sup L 11/;11 2<00. In the same
n i~n n i.::s::n n i.::s;;n
way we infer that sup
n
L I(x, XJI2 = sup L I(x, Xi)
i:S;n n i::;;n
(Xi' xTW
~sup L II(x,Xi)xTI1 2<00, taking /;,=(X,Xi) TxT, using again the forego-
n i~n
ing lemma, and the fact that SUP{L II(x,xi)xTlllflE1:} < 00. This com-
pletes the proof of the theorem. ie/l
References for Chapter VI: GELBAUM [1], KARLIN [2] and ORLICZ [2].

CHAPTER VII
Decompositions
It is quite natural to generalize the concept of a basis for a space

X by taking a sequence of linear (not necessarily closed) subspaces of X
instead of a sequence of elements of X and, in the same time, to define
X to be an F-space. Such a basis of subspaces is called a decomposition
and, if the subspaces are closed, a Schauder decomposition. The closure
of the subspaces is intimately connected with the continuity of projections
onto these subspaces. If X is a Banach space, it turns out that X always
has a decomposition, but there are (non-separable) Banach spaces
which do not have a Schauder decomposition. Concerning the existence
of Schauder decompositions of Banach spaces, a theorem is verified
which corresponds to the theorem of NIKOL'SKII for the existence of a
basis. Finally, it is shown that the notions of a weak Schauder decom-
position and a Schauder decomposition in a Banach space are equivalent.
1. Decompositions of F-Spaces
Unless otherwise stated, X is always an F-space. The concept of a

decomposition has been introduced by FAGE [lJ and GRINBLYUM [5J.
It has been developped and generalized to Banach and F-spaces by
Mc-ARTHUR and RETHERFORD [1 J, RETHERFORD [1 J, RUCKLE [2J, and
SANDERS [2, 3J.
Definition 1. A sequence {M;} of (not necessarily closed) linear sub-
spaces of X is a (weak) decomposition of X if and only if for each x in
X there exists a unique sequence {x;} such that XiEMi for all i and
x = lim I Xi in X (in the weak topology for X, respectively).
n i~n
The uniqueness implies the existence of (not necessarily continuous)
associated projections Pi of X on Mi such that PiPj=bijPj.
Definition 2. If all subspaces Mi of a (weak) decomposition {MJ of
X are closed, then {M;} is called a (weak) Schauder decomposition of x.
1. Decompositions of F-Spaces 87
As will soon be seen, a Schauder decomposition of X is a decompo-

sition of X such that each of the associated projections is continuous.
For a Banach space X it is clear that every Schauder decomposition of
X is a weak Schauder decomposition of X.
Lemma 3. If {MJ is a Schauder decomposition of X, then for each i
there exists a closed linear subspace Ni of X such that M/BNi=X,
Proof. Let i be fixed and set M = M i , the summation symbol I' = I
OC>} j j*i
and N = { xEXlx= j~~ xj,xjEMj . Since {M} is a decomposition of X,
we have the unique representation X=X i + I' Xj for each x in X, where
j= 1
the sequence {Xj} (depending on x) is such that xjEMj' j= 1,2, ....
Hence MuN =X and M ( I N = {O}, the equivalent statement to
M tf) N = X. It now remains to show that N is closed. For this purpose,
let Y be the linear space defined by Y= {{xj}lli~ .I
Xj exists in X,
J~n
xjEM j for allj~. Since each sequence converging in X is bounded (1.1.3),
J
we can define a translation-invariant metric 1111 on Y by the function
where IIxll is the quasi-norm of x in X. It is easy to see that Lemma

1.1.4 impli~s that the metric 1111 on Y is a quasi-norm on Y. By the
following argument, Y is complete and hence an F-space: Let for every
8>0 be an n such that JJ{x pj -xqj}JJ<8 for any p,q~nE' Then we
E
obtain s~p II Xpn-Xqn I ~2s~pll.~ (xpj-Xq)11 =2JJ{xpj -xqj}JJ <28, and

J~n
since each Mj is complete (a closed subset of a complete metric space

obviously is complete) there exists a sequence {XOj} in X such that for
all j, limxpj=xOjEMj. Moreover {xoj}=lim{xpj} in the topology of
p p
Y, because for each n there exists a Pn ~ n such that II j~n (XOj - xPn) II < 8,
E
which implies that JJ{XOj-xpj}JJ =s~p t~n(xOj-Xp)1I ~s~p t~n (XorXpn) II

+s~pllj~n(Xpnj-xp)II<28 for all p~nE' Since {Xnei}EY there is an
mE~nE for which /ljt Xneill <8 for all n,m~me' Thus
7'
88 VII. Decompositions
for all n,m=m•. Hence lim I XOj exists in X, we have {XoJE Y and
n j~n
Y is complete.
We now define the linear transformation T: Y-7X by T({x j})
=lim I Xj' By uniqueness the last series vanishes if and only if {Xj} =0,
n j~n
hence T is one-to-one. That T is onto follows immediately from its defi-

nition and the fact that {M j } is a decomposition of X. Next we prove
that T is a topological isomorphism: //T({x j})// = IIJl Xjll ~s~p t~n Xjll
= /I {X j}/I. Therefore, by (1.2.7), T is continuous and this shows that T - 1
is also continuous (1.2.6).
Finally, let Yo be the limit of a convergent sequence Yl'Y2"" in N,
00 00
where Yo= I YOj' YOjEM j, and where h= I' hj' YkjEM j, hi=O, k= 1,
j= 1 j= 1
2, .... Using the fact that T is a topological isomorphism, we conclude
that {Yoj}=T-IYo=T-llimh=lim{hj}' But from the definition of
k k
the metric in Y it follows immediately that YOi=limYki=O, hence that
k
YoE N. Therefore N is closed and the proof of the lemma is complete.
Corollary 4. If {MJ is a weak Schauder decomposition of a Banach
space X, then for each i there exists a closed linear subspace Ni of X such
that MJBNi=X,
Proof. The corollary can be verified with almost the same formalism
used to prove the lemma. But now, Y is defined by weakly convergent
series, and by Theorem 1.3.15 one can use the same metric on Y. The
same theorem makes clear that {zp} is a Cauchy sequence in X, where
zp=weak-li,?1.I xpillzp-zqll ~s~p II.I (xpj-Xq)II)· Since X is com-
J~n \ J~n
plete it is then possible to define zo=limzp and one can show that
p
zo=weak-lim I XOj: Given X*EX* of norm one, there is a fixed

n j~n
p?:cn. such that Ilzo-zpll <c; and an m. (depending on p) such that
Ix* (zp- j~n Xpj) I<c;, n?:cm•. Hence
Ix*(zo- j~n xo j)! ~ IIzo-zpll + Ix* ~P- j~n x pj)/ +s~p t~}Xpj-XOj)11 <4c;
for all n?:c m.. Defining T by a weak limit, the rest of the proof applies
without modification.
Theorem 5. In an F-space X the following three conditions are equi-
valent:
I. Decompositions of F-Spaces 89
(i) There is a sequence {M;} of closed linear subspaces of X which is a

Schauder decomposition of X.
(ii) There is a sequence {P;} of continuous projections of X such that
PiPj=(jijPj and lim I Pix=x for each x in X.
n i~n
(iii) There is a sequence {Qn} of continuous projections of X such that
QmQn=Qmin(m,n) and I~Qnx=x for each x in X.
The subspaces Mi and the projections Pi and Qi are related by Pi(X)=M i
and Qn= I Pi' n= 1,2, ....
i~n
Proof. Let (i) be given. We define the linear transformations Pi of X

onto Mi by PiX=X i, i= 1,2, ... , where lim I Xi is the unique series
n i~n
expansion for XEX. Evidently the P;'s are (not necessarily continuous)
projections and PiP j = (jijP j . By the following reasoning, uniqueness
implies the closure of Pi for each i: By Lemma 3 there exists a closed
linear subspace Ni such that M/£J Ni=X, Let Yv be a sequence in X
converging to Y such that Piyv converges to Z in X. Then lim(Yv-PiYv)
= y-z. Since PiYvEMi and Yv-PiYvENi' it follows that zEM i and V
y-zEN i· Thus Piy=PiZ=z and, therefore, Pi is closed. By the closed

graph theorem (I.2.10), the P;'s then are continuous and (ii) is proved.
Conversely, if (ii) holds, let Mi=Pi(X) for all i. Since Pi is continu-
ous, Mi is closed. For every x in X we have x=lim L PiX. To show
n i~n
the uniqueness of the sequence {PiX} we assume a sequence {y;} such
that lim I Yj=O and YiEMi for all i. But then, by continuity of each
n j~n
Pi' O=P;lim I Yj=lim I PiYj=PiYi= Yi' Hence (i) is satisfied.
n j~n n j~n
Given (ii), taking Qn = L Pi' we have only to verify that Qm Qn
i~n
Pi = Qmin(m,n) and (iii) follows.

i::::;m j~n i=min(m,n)
Finally, from (iii) with P 1=Q1' Pi =Qi-Qi-1' i=2,3, ... , we have PiPj
=(Qi- Qi-1)(Qj- Qj-1)= Qmin(i,j)- Qmin(i,j-1)- Qmin(i-1,j)+ Qmin(i-1,j-1)
= (ji/Qj - Qj- 1) = (jijP j, which proves (ii), and where, for convenience,
we have taken Qo = O. This finishes the proof of the theorem.
As we have done in the case of bases, we sometimes denote a de-
composition {M;} of X by {Mi'P;} or, since Mi=Pi(X), by {Pi(X),PJ
Theorem 6. Let X be an F -space and {Qn} a sequence ofcontinuous
00
projections of X such that Qn Qm = Qmin(n,m) and sp U Qn(X) = X. Then

n= 1
lim Qn X= X for each x in X
n
if and only if {Qn X} is a bounded sequence
in X for each x in X.
Proof. The necessity is obvious from (1.1.3) and the convergence of

{Qn x }.
On the other hand, if x is in the set Y= {yIYEQm(X),m< <Xl}, then
for some m< <Xl we have XEQm(X) so X= Qmx. Since limQnx
n
=limQnQmx=Qmx=x for all x in Qm(X), limQnx=x for all x in Y.

n n
By hypothesis, {Qnx} is a bounded sequence for each x in X. Hence by
(1.2.8), the limit Q x = lim Qn x exists for each x in X and Q is continuous
n
and linear. But since Qx=x on Yand Y=X, Qx=x on X and the
theorem now follows.
2. Decompositions of Banach Spaces

Throughout this section X is a Banach space.
Theorem 1. Every infinite dimensional Banach space X has a decom-
position.
Proof. According to Corollary IV.3.10 there exists a closed linear
subspace Y of X with a basis. Let P be a projection of X on Y. Such a
projection exists since we do not assume that P is continuous. Let now
Ml be the nullspace P- 1 (0)={XEXIPx=0} of P.1f {y;} is a basis for
Y we define the other subspaces M i , i = 2, 3, ... to be the linear subspaces
spanned by Yi-l, i=2,3, ... respectively. Thus {M;} is a decomposition
of X. However, {M;} must not necessarily be a Schauder decomposition
of X as we see in the subsequent Corollary 3.
Lemma 2. There is no continuous projection of 100 on its subspace co.
Proof. Let {x:} CC6 be a sequence such that Ilx:11 = 1, n= 1,2, ...
and such that limx:(x)=O for each x in co. To see that such a sequence
n
exists we have only to take x: = bnEC6( = Ii), where bn = {bnJ If there
would be a continuous projection P of 100 on co, then lim x: (P y)=O for
n
each Y in 100- Hence {p* x:} would be a sequence in C6**( = I!) which
converges to 0 in the weak* topology of C6**. If J is the natural (iso-
metrically isomorphic) embedding of Co into C6*, then by Phillips' lemma
(1.4.1) it follows that limllJ* P* x:11 =0. Since J* P* X:EC6 we have
n
IIJ* p* x:11 =sup {IJ* P* x:(x)11 xEco,llxll ~ I}

= sup {lx:(P J x)11 xEco,llxll ~ I}
= sup {lx:(x)1 I xEco,llxll ~ I}
=llx:ll=l,
and this contradiction verifies the lemma.
It will turn out that the following corollary is only a special case of
Dean's theorem (2.13):
Corollary 3. There exists a decomposition of the Banach space 100
which is not a Schauder decomposition.
Proof. In the proof of the preceding theorem we take X = 100 and
P(X)= Y=c o' By Theorem 111.7.1 there is a basis {bJ for co, where
bi = {b ij }. But from the above lemma we know that P is not continuous.
Hence M I = P - 1 (0) is not closed, since otherwise P would be conti-
nuous. This shows that {MJ is not a Schauder decomposition of 100 ,
Lemma 4. If Y is any separable closed linear subspace of 100 such that
Co eYe 100 , there is a continuous projection of Yonto co.
Proof. From the separability of Y it follows the existence of a de-
numerable dense set in the unit ball of Y. Retaining only those elements
of this set which are linearly independent and which are not in Co, we
get a set {XI,X2""} in Y. We first suppose that there is only a finite
number n of elements in {X I ,X2""}' Obviously we have con sp{xili:::;n}
={O}. Let xbean element of y"=coEBsp{xili:::;n} and x=zo+ I>:iXi
i~n
(lXiE<P) its unique decomposition into components in Co and spxi,i:::;n

respecti vel y.
If Xi= {xij}, we define the (column) vectors vj = {x lj , .. . ,xnj},j = 1,2, ...
These vectors are in the unit ball Un of I':x" the Banach space of all n-tuples
{f3I"'" f3n} of scalars, with the corresponding norm given by the ex-
pression SUP{If3kll k= 1, ... ,n}. Since I':x, is finite dimensional, Un is
sequentially compact (1.1.9 and 12). Thus there is a nonvoid set l: in Un'
the set of all cluster points of the sequence {v j }. Moreover, there exists
for each v= 1,2, ... a 2/2 v-net iffv with respect to Un" By definition, there
Wl
is to each Vj a vector V ) = {wiv]"", w~1} in iffv such that Ilvj-w)V)11 <2/2v
and such that only a finite number of vectors in the sequence {W)V)} are
not infinitely repeated. We then may derive from {W)v)} another se-
quence {wlV )} in iff v in which each vector is infinitely repeated. This can
be done by taking wjl)=O and by altering a finite number of vectors
W)V) (v> 1) in {wlv )} such that for the corresponding new vectors W)V)
we have WlV)Eiff v, Ilwlv)-W)v-I)11 <6/2 v and inf{llwlV)-vlll VEl:} <2/2v.
Furthermore, since for v = 1, sup Ilw)V) - w)v-I)II < 6/2v, the set of vectors
J
W)v) = {wj'?, ... , w~1} has the property that lim W)V) exists in I':x, uniformly
with respect to j. At last we define sequen~es {yjV)} and {YjV)} by yjV)
-{
- W(v)
il ' W(v)
i2 '··· } an d Yi
-(V)_{-(v)
- Wil ' W-(V)} t' I
i2 '... respec lve y.
Let z(v) = Zo + I lXi(yjV) - .W)). Since only a finite number of entries of
L lXi(yjV) -
i~n
the vector jijv)) are different from zero it is clear that z(v) is
i~n
in Co. Since Co is complete, since y\V) converges for all i ~ n to Xi with

V-H() and since y!V) is a Cauchy sequence for all i~n, z(v) converges to
an element z in Co.
Then
II z(V) II ~ Ilzo+ i~n aiy!V) I + t~n aiW)11
~ Ilzo+ i~n aiy!V) I + I z(v) + i~n aiW)11

=21I zo + i~n aiY!V) II
~211xll +2 t~n ai(xi- y!V») I '
where we have used the fact that each entry of I aiY!') is infinitely
i~n
repeated (since each column vector tv)V) is infinitely repeated).
Next, let P n be the transformation of Yn into Co, which is given by
Pnx=z, XEYn" P n is linear, since for x'=z~+ I a;Xi' Pn(x+x')
i~n
=li~ [zo+z~+.I (ai+a;)(y!V)- Y!'»)] =Pnx+Pnx'. That P n is a (not

l~n
necessarily continuous) projection of Yn is clear from its definition.

Moreover, P n is continuous, since IlPnxll = Illi~z(V)11 =li~lIz(V)11 ~21Ixll.
00
Finally, if there is no finite n such that Yn= Y, since sp U Yn= Y, there
n=l
is a unique continuous extension P of the P n's to all of Y with norm ~ 2
(1.2.12). Again, since Co is closed, P is a continuous projection of Yon Co.
Remark. It is known that P has norm 2. However, for our purposes
it is sufficient to know the existence ofa continuous projection of Yon Co.
Lemma 5. Any separable Banach space having a subspace which is
topologically isomorphic to Co admits a continuous projection onto that
subspace.
Proof. Let X be a separable Banach space, let Y be a subspace of X
and let T be a topological isomorphism of Yonto co. If Ty= {zi(y)},
yE Y, each zi belongs to y* and there is a constant K > 0 such that
Ilzill ~K. By the Hahn-Banach theorem (I.3.8) there is an extension
Z['EX* ofziwith Ilzi'll= Ilzill· Then T':X~/oo,givenby T'x={z['(x)},
XEX, evidently is an extension of T to X. T' is bounded, since IIT'xll
=s~plz['(x)I~llxllsupllzi'll=llxlls~pllzill~Kllxll. Thus T'(X) is a
r r r
separable closed linear subspace of 100 which, by Lemma 4, admits a
continuous projection, say P', of T'(X) on co. Consequently, P= T- 1 P' T'
is likewise continuous, and is a projection of X on Y, since p 2 = T- 1

P'T'T- 1P'T'=T- 1P'T'= P.
TheQrem 6. Every separable Banach space X which has a subspace
which is topologically isomorphic to Co has a Schauder decomposition.
Proof. The preceding lemma shows the existence of a continuous

projection P of X on the subspace Y of X which is topologically iso-
morphic to co. Hence X = (I - P)(X) EB Y. Let T be the topological iso-
morphism of Co onto Yand let {b i } be a basis for Co whose existence is
proved in Theorem III. 7.1. Since then, sp {T bj } = Y and, using for in-
stance Theorem IV. 1.5, since t~n lXi Tbi II ~ II Til t~n lXibill ~ IITlllli~m lXibill
~IITIlIIT-1111Ii~m lXi Tbi II for each n,m with n~m and arbitrary
scalars lXi' {TbJ is a basis for Y. Finally, the set {Mi} is the desired
Schauder decomposition of X, where M 1=(1 -P)(X) and Mi is the
one-dimensional subspace of X spanned by Tb i - 1 , i=2,3, ....
Theorem 7. Let {MJ be a sequence of closed subspaces in X such that
00
sp UMi=X, Then {Mi} is a Schauder decomposition of X if and only

i=l
if there exists a constant K ~ 1 such that II i~n Xi I ~ K II i~m Xi I for all
n,m with n~m and for all sequences {Xi} with XiEMi'
Proof. Necessity. Let {Mi} be a Schauder decomposition of X with
corresponding sequence of projections {Pi} of X on each of the sub-
spaces Mi' Then x=lim L PiX for every XEX and the uniform bound-
n i~n
edness principle (1.3.14) implies the existence of a constant K~ 1 such
that s~p t~n Pill ~K. Since the projections in the set {Pi} are mutually
orthogonal it follows that L P i = L P L Pi
j for n~m. Therefore
i~n j~n i~m
t~n Pixil ~K t~m Pixil for every XEX. Setting Xi=PiX, this is the
necessary condition.
Sufficiency. Let {En} be a sequence of closed subspaces of X, given
by En=SP UMi' Starting from the obvious equation E1 =M 1 we
i~n
proceed recursively to show that En = M 1 $ ... EB M n (hence to show

that En n Mn+ 1 = {O}). We suppose that En=M 1 EB ... EB Mn for some
n~ 1 and take an arbitrary xEEnn M n+1. Clearly X then has the unique
decomposition X= LXi' XiEM i. The inequality of the hypothesis im-
i~n
plies that IIxll = t~n xiii ~K Ili~n xi-xll=O. Thus En n Mn+ 1= {O}, im-
plying En+l=M1EB"'EBMn+l' Let then Qnm for m~n be the pro-

jection of Em onto En. From the unique decomposition X= I Xi' XiEM i,
i~m
for any xEEm and from IIQnmx11 = Ili~n Xiii ~K IIJmXil1 =KllxlI then
follows that the norm of each of the projections Qnm is bounded by K.
Since for fixed n, Qnm is the restriction of Qn.m + 1 to Em for all m ~ nand
since sp {M;} = X it is, through extension by continuity, clear that for
all of the projections Qnm (with domain Em) there exists a unique linear
extension Qn on the whole space X (1.2.12). Moreover the norm of Qn
is also bounded by K. Qn is a projection with range En> since En is closed
and since Q;x=QnnQnx=Qnx for all XEX. Again because
00
sp U Mi=X, we have for each XEX and every 8>0 an index p such
i= 1
that inf {lIx- ylli yEEp} <8. This implies IIx-Qnxll ~ lIy-Qnxll + IIx- yll
=IIQn(x-y)II+lIx-yll~(K+1)8 for n~p, where we have chosen a
YEEp for which IIx- yll ~8. Consequently, limllx-Qnxll =0 for every
n
XEX.
Finally, since with Qo=O for each i~l onehas Mi=(Qi-Qi-l)(X),
and since for arbitrary m,n one has QmQn=Qrnin(m.n)' by Theorem 1.4,
{M;} is a Schauder decomposition of X.
Corollary 8. Let {P;} be a sequence of mutually orthogonal continuous
00
projections of X (i.e. PiPj=bijP) such that sp U Pi(X)=X. Then

i= 1
{Pi(X)} is a Schauder decomposition of X if and only if there exists a
constant K~ 1 such that s~p Ili~n Pill ~K.
Proof. To show the sufficiency, let Mi=Pi(X) and let m,n be posi-
tive integers such that m~n. Since by hypothesis, PiPj=bijP j , we have
for XEX, II.IpiXII= II.I Pj.I Pixll~KII.I Pixll· Since X was arbi-
l~n }~n l~m '~m
trary, the preceding theorem shows immediately that {M;} is a Schauder
decomposition of X. On the other hand, if {P;(X)} is a Schauder de-
composition of X, we have shown that there is a constant K~ 1 for
which s~p t~n Pill ~K, and this is the necessary condition.
Corollary 9. If X is a Banach space with a Schauder decomposition

{Pi(X)} of nontrivial subspaces of X, then each sequence {x;} of non-zero
vectors Xi in Pi(X) is itself a basis for sp {x;}.
Proof. The corollary follows at once from Theorem 7 and an appli-

cation of Theorem IV.l.S.
Theorem 10. {M;} is a weak Schauder decomposition of X if and only

if it is a Schauder decomposition of X.
Proof. Let {Mi,PJ be a weak Schauder decomposition of X. Given
YEX, let (in be the set (in = {i~nIPi(y)#O} (depending on y), and let
y=sp{Piy}. Since for all XEX, I PiX converges with n weakly to X
i:O;:;n
in X it is clear by (1.3.15) that yE Y, and that x*(x)=lim I x*(Pix),
n i~n
X*EX*. Therefore, by (VII.1.4), (1.2.18) and (1.3.14) there exists a con-
stant M>O such that s~p t~n Pill ~M. Using PiPj=bijPj, one ob-
tains for every sequence {IXJ in 1[>, and for m~n, 112;:m lXiPiyl1
= II.I Pj.I lXiPiyl1 ~MII.I lXiPiyll· Thus by Nikol'skil's theorem
J~m lEO'n lEU"
(IV. 1.5), {Piy} is a basis for Y. This implies that there are unique coeffi-
cients f3iEI[> for which y=lim I f3iPiY in the strong (and hence in the
n .
lEan
weak) topology of X. But by hypothesis, f3i= 1 for all i so that {Mi,PJ

is a Schauder decomposition of X.
Conversely, if {M;} is a Schauder decomposition of X, then I PiX
i~n
converges in the strong, hence also in the weak topology for X to x. The
uniqueness is shown by the following reasoning: Suppose that {zJ is a
sequence in X such that ZiEMi and li~x*(.I Zi) =0 for all X*EX*.
I~n
Then each Zi must vanish since x*(zi)=li~P{ x*(.I Zj) =0 for all x*
)~n
in X*. Thus {MJ is a weak Schauder decomposition of X and the proof

of the theorem is complete.
Theorem 11. Let X be reflexive. Then {Mi,P i} is a Schauder decom-
position of X if and only if there is a Schauder decomposition of X* with
associated sequence of projections {Pt}.
Proof. Let {Mi,P;} be a Schauder decomposition of X and let
Qn= I Pi' Then x*(x)=x*(limQnx) =limx*(Qnx)=limQ; x*(x) for
i~n n n n
all X in X and all x* in X*. Since X is reflexive this implies that
limx**(Q; x*)=x**(x*) for every x** in X** so that Q; x* converges
n
to x* in the weak topology of X*. Hence for each x* in X*, as a conse-
<Xl
quence of (1.3.15), sup IIQ; x*11 < 00, and x* is in

n
sp U Q;(X*).
n=1
The
<Xl
last property implies that sp U Q;(X*)=X*. Furthermore, Q;Q!

n=l
= Q!in(m,n), because Q; Q! x* (x) = x* (Qm Qn x) = x* (Qmin(m,n) x) = Q!in(m,n)
X*(X}. Now, by Theorem 1.6, Theorem 1.5 (iii) and by what has
preceded, {Pi(X*)} is a Schauder decomposition of X*, where
{Pi} = {Qi,Q!-Qi,Q~ -QI, ... }.
On the other hand, applying again this procedure, it follows that
{Pi*(X**)} is likewise a Schauder decomposition of X**. Let J be the
natural embedding of X onto X** and let Pi=J- 1 Pi* J. Then Pi is a
L L
continuous projection of X and lim Pi X = r 1 lim Pi* J x = r 1 J x
n i~n n i:::;;n
=X and, again by Theorem 1.5 (ii), it is clear that {Pi(X)} is a Schauder
decomposition of X.
Corollary 12. If {Pi(X)} is a Schauder decomposition of X and X*
has the property that the convergence of a sequence in the weak* topology
of X* implies the convergence of that sequence in the weak topology of
X*, then {Pi(X*)} is a Schauder decomposition of X*.
We observe that there exist Banach spaces which are not reflexive
but which although have the convergence property mentioned in the
above corollary. An example of such a space is the conjugate space I!
of 100 (I.4.c). Based on this property and on the fact that each bounded
linear transformation of 100 into a reflexive Banach space maps weak
Cauchy sequences into strongly convergent sequences (I.4.c) we can now
prove the following theorem, and the subsequent proposition which is
used to verify the theorem.
Theorem 13. (DEAN) The Banach space 100 does not have a Schauder
decomposition.
Proof. We suppose that 100 has a Schauder decomposition, {Pi(loo)},
and show that this assumption leads to a contradiction. Without loss
of generality one may suppose that each subspace Pi(loo} is nontrivial.
We first choose a sequence {Xi} with xiEP;(l cx,} of norm one and a
sequence UJ with fiEP;(loo}*' also of norm one, and such that fi(x i)= 1
(which is possible by (1.3.1O)). According to the Hahn-Banach theorem
each fi has a norm-preserving extension xi in I!. It follows that Pi xi(xj}
=xi(Pixj)=bijxi(xi)=bijfi(xi)=bij. Let Y=sp{Pi xi} in I!. Since
IIPi xiii =sup{IPi x{(x)llllxll:( 1,XEloo } ~Pi xi(x;)= 1, and since by
Corollary 12 {Pi (I!)} is a Schauder decomposition of I!, Corollary 9
implies that {Pi xi} is a basis for Y. Moreover, we then have x*
=li~.IPix* in I! for every x*EI!. Since Ix*(xi)I:(IIPjx*(xi)1
l~n )<1
+llx*-j~/jx*11 and j~/jx*(X;)=j~iX*(PjXi)=O, Xi converges with

100-
i weakly to 0 in
Next we consider the linear transformation T: 100 --> Y*, given
by Tx(y*) = y*(x), xEloo' y*E Y. T is continuous, since IITxl1
=sup{ly*(x)li y*E Y,lly*11 ~ 1} ~ Ilxli. As we soon will show (Proposition

14), Y is reflexive. Since then y* is also reflexive (1.3.18) and thus weakly
sequentially complete (1.3.17), this implies that T sends the weakly con-
vergent sequence {Xi} in 100 , into the sequence {Tx i } which converges
strongly in y* (l.4.c), say to F. But from the definition of T it follows
immediately that li~ Tx;(y*) = li~y*(x;)=O for all y*E Yc I~, hence
1 1
that TXi converges to 0 in the weak* topology of Y*. There-

fore, F=O so that li~IITxill =0 which contradicts the estimate
1
IITxdl =sup{ly*(xi)li y*E Y,lly*11 ~ 1} ~ IPi" xt(x i)I/11P7 x711 ~ 1/supllPnll >0
n
(by Corollary 8, supllPnll<oo). It now remains to show the following
n
Proposition 14. Y is reflexive.

Proof. Let {z;} be a sequence in the unit ball of Y. Since {Pi" xn
is a basis for Y one has the unique expansion z; = lim L rxni Pt xt
m i~m
with rxniE tP, and by (111.2.3) there is a constant M > 0 for which
sup Irxnd ~2M/llPt x711. Thus there is an increasing sequence of positive
n
integers {n 1 ) such li~rxn'j, 1 exists and is, say, rx 1 (1.1.12 and 1.1.9).
J
Likewise, there exists a subsequence {n 2j } of {n 1 ) such that li~rxn2j, 2 = rx 2 •
J
Continuing in this way (this method is called a diagonal process) one
obtains
. a subsequence {z;jj} such that z;.JJ = lim L rx nJJ... i Pt x7 and
1lmrx n.. i=rx i .
m.~
I~m
j G~~en £>0 there is for any fixed xE/ro a k such that IIj~k PjXl1 <£,
sothat I(z;pp -Z;qq)(x)1 ~ I.I 1=1
(rxnpp,i-rxnqq)P7 x7(.L Pjx) I +211.f Pjxll
J<k J=k
< I L (rx npp , i-rxnqq)Pj xj(x)1 +2£ and this is smaller than, say 3£, if
j<k
P and q are chosen large enough. Thus {z;)x)} is a Cauchy sequence
in tP for each x in lro and the Banach-Steinhaus theorem implies the
existence of a z* in l~ which is the weak* limit of {z;jJ in I~. But then
one has limz;
j
.. =z* in the weak topology of I~ (l.4.c) and by (1.3.15)
JJ
z* is in sp {z;} c Y. Now, since every FE y* has an extension x'f* in

l~* (1.3.9), li~F(z;)=limx'f*(z;)=x'f*(z*)=F(z*),
j JJ j JJ
{z;}
JJ
converges
to z* in the weak topology of Y. Finally, this shows that Y is weakly
sequentially compact and hence reflexive (1.3.18).
References for Chapter Vll: DEAN [1], FAGE [1], GRINBLYUM [5],
McARTHUR and RETHERFORD [1], RETHERFORD [1], RUCKLE [2], SANDERS
[2, 3] and SOBCZYK [1].
CHAPTER VIII
Applications to the Theory

of Banach Algebras
In the first paragraph of this chapter we assume to have to do with

Banach spaces having a basis, as is the case for separable Hilbert spaces
or for most of all known concrete separable Banach spaces. Whenever
this hypothesis is fulfilled we can draw conclusions on the approximation
problem of compact linear operators by linear operators of finite rank.
If X is a Banach space with a basis, or at least with a Schauder
decomposition, then, as a beautiful application to the theory of Banach
algebras, there are many results on the properties of the corresponding
proper n-rings. A n-ring is a commutative Banach algebra, which is a
subset of the vector space s over the field CP, of all sequences IX = {IXJ in CP,
which contains the elements {1, 1, 1, ... }, {1, 0, 0, ... }, {O, 1,0,0, ... }, ... ,
and whose elements satisfy some simple inequality. It turns out that the
set r= {IXIIXES such that I lXiPiX is (weakly) convergent for every x in X}
i~n
is a (weak) n-ring, the proper (weak) n-ring of the decomposition {Pi (X), PJ.
Moreover, the proper weak n-rings and the proper n-rings are equivalent,
semi-simple, and isometrically isomorphic with the set of operators II
in B(X), defined by A(IX)x=lim I lXiPiX, XEX, IXEr. Based on an
n i~n
existence theorem for Schauder decompositions, which is a generalization
of the corresponding existence theorem for bases established by
NIKOL'SKII, we derive an alternative existence theorem for Schauder
decompositions of X which is related to the existence of proper n-rings.
If the proper n-rings of Schauder decompositions of Banach spaces
are partially ordered in a natural way by inclusion, there exists a minimal
n-ring with a corresponding minimal Schauder decomposition. The
basis for a special space of sequences, W o, ist shown to be minimal.
Naturally, the properties of a proper n-ring of a basis depend on the
type of the basis. Some results are established in the special case of
unconditional bases.
2. n-Rings 99
1. Two-Sided Ideals of Operators of Finite Rank
Definition 1. A linear transformation on a Banach space X to another

Banach space Y is said to be of finite rank if its range is finite dimensional.
Theorem 2. If X and Yare Banach spaces and either Y has a basis or
X* has a retro-basis, then each compact linear transformation T:X -> Y
is approximable in norm by linear operators of finite rank.
Proof. Let U x be the unit ball in X and let first have Ya basis with
the corresponding sequence of expansion operators {Un}. Then T(U x )
is conditionally compact and by (1.4.4) lim Un Tx = Tx, uniformly on
n
U x' Clearly Tn = Un T is of finite rank and the preceding consideration
yields limIITn-TII=O.
n
On the other hand, let X* have a retro-basis with the associated set of
expansion operators {Vn*}. The adjoint T* is also compact (1.3.27),
therefore T*(U y.) is conditionally compact and lim V; T* y* = T* y*
n
uniformly on U y •• Hence limllTVn- Til = lim IIV; T* - T*II =0. Again,
n n
Tn= TVn is of finite rank, where we note that we have used the fact
that by Theorem III.2.12 each Vn* is the adjoint of a linear operator Vn
of finite rank. This completes the proof of the theorem.
Definition 3. Let X be a Banach space, F(X) c B(X) the set of all
linear operators of finite rank of X and ~(X)cB(X) the set of all com-
pact endomorphisms of X.
The classes F(X) and ~(X) are known to be two-sided ideals in

B(X), by (1.2.17) ~(X) is closed in the uniform operator topology of
B(X) and for F(X) we have the following corollary to Theorem 2:
Corollary 4. Let X be a Banach space with a basis. Then ~(X) is the
closure of F(X) in the uniform operator topology of B(X).
. By Theorem 2 and Theorem 111.2.7 the corollary remains true, if,
instead of X having a basis, we only assume that X* has a weak* Schau-
der basis.
2. n-Rings
We consider the complex vector space s of all sequences rx = {rx;} of

complex numbers. Defining the product of rx,{3ES by rx{3= {rxi{3;}, s
becomes a commutative algebra. Moreover, it is clear that e = {1, 1, ... }
is the identity of s. Let 6n= {6 n;}, where 6ni denotes Kronecker's symbol.
100 VIII. Applications to the Theory of Banach Algebras
Lemma 1. Each subalgebra t of s containing e and On' n = 1,2, ... , is

semi-simple.
Proof. We define the set un = {ex- onexlexE t}. It is immediately verified
that Un is an ideal of t. We show that Un is a maximal ideal: Suppose Un
to be a proper subset of a maximal ideal u. Then we can take an element
f3EU which is not in Un' implying f3n =I 0. But then we have eEU, since
f3;; 1 f3EU, (e - onHe - {3;; 1 (3)EU and {3;; 1 f3 + (e - on)(e - f3;; 1 (3) = e. Thus,
if eE U, one has U= t, hence U is not maximal. This contradiction shows
that Un is maximal. Since nUn = 0, t is semi-simple and we are done.
n
Now, if as a special case, the subalgebra t of s consists of all elements

ex for which the norm
Ilexll oo =suplexnl (1 )
n
is finite, then the underlying vector space of s is the Banach space 100 ,
Since Ilexf3lloo~llexlloollf3ll00' ex,f3El oo ' 100 is also a commutative B-algebra.
Hereafter, we use common symbols for the commutative B-algebras and
the corresponding complex Banach spaces.
Lemma 2. Let tes be a B-algebra such that onEt for all n. Then t is
a subset of 100-
Proof. Let 11'11 be the norm on t. Since lexnlilonil = Ilexonil ~ Ilexllllonll,
it follows immediately from the definition of Ilexooll that Ilexll oo ~ Ilexll,
hence that t is in 100-
Definition 3. A commutative B-algebra tes with norm 11'11 on t is

called a n-ring if and only if e, OnE t, n = 1,2, ... and there is a constant
K>O such that supllXnexl1 ~Kllexll for every ex in t, where XnEt is defined
by Xn= L 0i' n
i~n
The commutative B-algebras 100 and c based on the complex Banach

spaces 100 and care n-rings. This is clear since the norms in both 100 and
c are defined by (1). It is called attention to the fact that every n-ring is
a linear subset of 100 though its topology is not the same as that of 100
in general. Since an-ring t is a semi-simple commutative B-algebra it
follows from a known theorem (1.1.18):
Lemma 4. The norms by which a subalgebra t of s forms a n-ring are

all equivalent.
Next, let t be the closure of an-ring t in the topology of 100- Then the
identity mapping of t into the closed linear subspace t of 100 is continuous
2. n-Rings 101
by the inequality Iiall oo ~ Iiall derived in the proof of Lemma 2. There

are two possibilities: Either t#t or t=t (as sets). In the second case,
by (1.1.14) the norm 11'11 is equivalent to the norm 11'1100'
Definition 5. An-ring t is said to be of the second category in 100 if
and only if t= t. Otherwise t is said to be of the first category.
Clearly 100 and care n-rings of the second category in 100- To show
the existence of a n-ring of the first category in 100 , we need the following
lemma.
Lemma 6. If t is a n-ring of the second category in 100 , then c c t.
Proof. Let aEC and define aEC by a=liman- Then e,(jiEt,
i= 1,2, ... , imply that Pn=ae+ I (ai-a)<>iEt. Bec~use Pn converges to
i~n
a in the topology of 100 , a is in t. But t = t since t is ofthe second category

in 100 , and we obtain aE t. 00
Lemma 7.Let wcs be the set of all a insforwhich Ila i -a i +1l<oo.

Then i= 1
(i) w is properly contained in c, 00
(ii) the function 1IIIw:w->1R defined by Iiall w= s~plail + I lai-a i+11 is

a norm and ' i= 1
(iii) w is a n-ring of the first category in 1 00 ,
Proof. (i) Let aEW. From ~ap-aql =\:t>ai-ai+1)\ ~:t~ lai-ai+11,

p<q and the convergence of Ila i -a i + 11 it follows that ai is conver-
i= 1
gent, hence that a is in c. Thus we c. However, since {I, -1, t, - i, ... }
is an element of c, but not of w, it is evident that w is properly contained
in c.
(ii). Obviously Iiall w< 00 for each a in wand the function 1IIIw:w->1R
is a norm on w.
(iii). We first prove that w is complete. Let {Pp} be a Cauchy se-
quence in w. Then we have for every s>O an n such that IIPp-Pqllw<s,
p,q;;:' n. From the definition of norm we thus obtain IPpi- Pqil < s, p,q~ n.
Hence there is a Po in s such that limPPi=poi, i=,1,2, .... Moreover, for
p
any m there is a q;;:'n such that IIPqi-Poi-(P q,i+l-Po,i+l)l<s.

Since then i';m
Po is in wand lim
p
Pp = Po·
Next we show that W is a (commutative) B-algebra. As a direct result

of the definition of w we see that the identity e is contained in wand
that Ilell w= 1. Let r:x,/3EW. Since then r:x,/3Ec and
00 00
i= 1 i= 1
00 00
~supl/3l
J
L
i=l
lr:xi-r:xi+11+ suplal
J
L l/3i-/3i+11,
i=l
it follows that a/3Ew. We verify that 11r:x/3llw~ Ilall wll/3llw:

00
11r:x/3llw= SUplr:xi/3d +
l
L lr:xi/3i-r:xi+1/3i+ 11
i= 1
00 00
Now, Wis a n-ring, because (\EW, n= 1,2, ... and since

n-1
IIXnr:xllw=s~plail +
l-..;::n
L Ir:xi-ai+11 + lanl
i= 1
00
~2suplail +
l
L lr:xi-r:xi+1
i= 1
1 ~21Iallw'
Finally, W cannot be of the second category in 100 , since by Lemma 6

this would imply W= c which contradicts (i). This completes the proof
of the lemma.
3. Proper n-Rings of Schauder Decompositions
In this section we discuss endomorphisms A(a) of a Banach space

X which are related somehow to an element r:x of the commutative algebra
s. Let {MJ be a Schauder decomposition of X with associated sequence
of projections PuP z , ... of X onto the nontrivial subs paces M 1 ,M 2 , ••.
respectively. By r we define the following subset of s:
r={cxlr:xES, limLr:xiPix existsforall XEX}.
11 i::Sn
To each aEr there corresponds a linear transformation A(cx):X -+X,

given by
A(r:x)x=lim L CXiPiX. (1)
n i~n
3. Proper n-Rings of Schauder Decompositions 103
We have
Lemma 1. A(()() is bounded for every ()(Er.
Proof. Let ()(Er. Since A(Xn()() is represented by a finite sum and all
of the P;,s are bounded, A(Xn()() is an endomorphism of X. The conver-
gence of (1) implies that supIIA(Xn()()xll < 00, XEX. As a consequence of
n
the Banach-Steinhaus theorem (I.2.14) it follows IIA(()()II < 00.
Next, we define the subset II = {A(()()I()(Er} of B(X). The following
lemma shows how II is related to r.
Lemma 2. The linear transformation A: r-+ II is one-to-one.
Proof. Let A(()()=O. Due to the orthogonality property PiPj=OijP j
we then have O=A(()()PiX=()(iPiX for all XEX and all i. Since the sub-
spaces Mi are nontrivial, each Pi must be different from zero. Thus ()(i=O
for all i, i. e. ()( = 0 and this shows that A is one-to-one.
Lemma 3. r is a subset of the space 100 •
Proof. Let ()(Er. Again from A(()()PiX=()(iPiX,XEX we obtain
l()(iIIIPiXII::::; IIA(()()IIIIPixll, hence 11()(1100 = sup I()(i I::::; IIA(()()II < 00.
Theorem 4. II is a semi-simple commutative B-algebra. There exists a

norm on r such that r is a n-ring and that A is a proper isometric isomor-
phism of r onto lI.
Proof. Using the orthogonality relation PiP j = oijP j one gets for
A (()(), A({3)ElI, A(()()A({3)x=lim L ()(i PiA ({3)x=lim L Pi X = A(()({3)x, ()(i {3i
n i~n n i:$;n
XEX and thus A(()({3)ElI. Since A(e)x=lim L Pix=x, XEX, one has
n i~n
A(e)=I. Therefore A is a proper (algebraic) isomorphism of r onto n.
Moreover, II is a Banach space: Let {()(m} be a sequence in r such
that A(()(m) converges in the uniform operator topology of B(X), say to
AoEB(X). Since A(()(m)PiX=()(miPiX for each i and each XEX, we have
an ()(o~s such that lim()(mi=()(Oi. Moreover, lim A (()(m) Pi X= AOPix.
m m
From L ()(oiPix=lim L ()(miPix=limA(()(m) L Pix=A o L PiX and the

i~n m i~n m i~n i~n
boundedness of Ao we obtain that Ao = A(()(o)E lI. This shows that II is
a closed subset of B(X), and since B(X) is complete, II is also complete.
Through the mapping 11·11 :r-+~, given by II()(II = IIA(()()II, we now
introduce a norm on r and finally show that r is a n-ring. Corollary
VII.2.8 implies the existence of a constant K ~ 1 such that
IIA(Xn()()11 = IIA(()()A(Xn)11 = IIA(()() i~n Pill ::::;KIIA(()()II
8*
for all IY.Er. Hence IlxnlY.ll = IIA(XnlY.)11 ~KIIA(IY.)II =KIIIY.II, and for lY.,pEr
we have IIIY.PII = IIA(IY.P)II = IIA(IY.)A(P)II ~ IIA(IY.)IIIIA(P)II = IIIY.IIIIPII. The
completeness of r follows from the fact that A is an isometric isomor-
phism of r onto II. Because of these properties, and since e, c5 n Er,
n = 1,2, ... , r is a n-ring. Finally, the semi-simplicity of II is a consequence
of the semi-simplicity of r, where the semi-simplicity of r follows from
Lemma 2.1.
Definition 5. r is said to be the proper n-ring of the Schauder decom-
position {Mi} of X.
The next theorem shows that the notions of strong and weak proper
n-rings of Schauder decompositions of X (corresponding to strong and
weak convergence of the series expansion which defines the operator
A(IY.)) are the same.
Theorem 6. Let {Mi' Pi} be a Schauder decomposition of x. If for
00
some IY. in s the series L lY.i PiX is weakly convergent for each xEX,
i= 1
then it is convergent in the strong topology for X for all xEX.
Proof. Let r be the proper n-ring of {MJ Obviously, XnlY.Er for
every finite n. By hypothesis, x*[A(XnlY.)x] converges for each x in X
and each x* in X*. Thus sup Ix*(A(XnlY.)x)1 < 00, XEX, X*EX* which,
n
as a consequence of the uniform boundedness principle (1.3.14), implies
sup IIA(XnlY.)11 ~ M, where M is some positive constant. Since for every
n
X in X and every a>O we have an index p such that Ilx-xpll <a/2M,
where xp= L PiX, we have
i~p
IIJm lY.iPixl1 ~ Ilitm lY.iPi(X-Xp)11 + Ilitm lY.iPiXpl1

= II[A(XnlY.)-A(Xm-llY.)](X-Xp)11
~2Mllx-Xpll <a,
for every n?:m>p. Thus {.L lY.iPiX}
l~n
is a Cauchy sequence and since
X is complete, it is shown that the series converges strongly in X.
Theorem 7. A sequence of closed subspaces {MJ in X is a Schauder
decomposition of X if and only if the following conditions are satisfied:
(i) There is an isometric isomorphism A from an-ring r into B(X),
(ii) for each i, A(c5J is a projection of X onto Mi and
00
(iii) X =sp U Mi·

i= 1
3. Proper n-Rings of Schauder Decompositions 105
Proof. Necessity. (i) is a consequence of Theorem 4, (ii) follows from

the definition (1) of A which yields A(bi)=P i, and (iii) is a necessary
condition for {Mi} to be a Schauder decomposition of X.
Sufficiency. If (i) is satisfied, then since A is a homomorphism,
A(bi)A(bj)=A(bibj)=bijA(b). Defining Pi = A(b i), which, by (ii) is a
projection of X onto M i, one obtains PiP j = bijPj ' Since A is an isometric
isomorphism we have Ilocll = IIA(oc)11 on r, where lIocll is the norm of
OCEr. Taking now OC=Xn yields !!i~n PiX!! =IIA(Xn)xlI~IIXnllllxll· Sincer
is a n-ring there exists a constant K~1 such that IIxnll~Kllell=K.
This implies t~n PiX!! ~ K Ilxll and by Corollary VII.2.8, {Mi} is a
Schauder decomposition of X with associated sequence of projections
{Pi}'
Corollary 8. The n-ring r of the above theorem is the proper n-ring
of {M;}.
Proof. Since for OCEr, L OCiPiX= L ociA(bi)x= L A(ocbi)x

L PiX
i~n i~n i~n
=A(a) and since {M;} is a Schauder decomposition of X,

i~n
lim L OCiPiX exists for each XEX. Hence A(oc)x=lim L OCiPi X,

n i~n n i~n
showing that r is the proper n-ring of {Mi}'
Corollary 9. Theorem 7 holds if the condition (i) is replaced by:
(i') There is an isomorphism A from an-ring r onto a B-algebra II c B(X).
Proof. Again, by Theorem 4, (i') is a necessary condition. Since now
A is only an isomorphism, in the sufficiency proof one has to show that
A is homeomorphic: Since a n-ring is semi-simple and the semi-sim-
plicity of r implies the semi-simplicity of II = A(r), by a known theorem
(1.1.19), A is a homeomorphism. Hence there exists a constant M>O such
that sup{IIA(oc)lllllocll ~1,ocEr} ~M. We now obtain !!i~/iX!! ~KMllxll
and the rest of the proof is the same as that for Theorem 7.
In the following we investigate the problem of the extension of a
topological isomorphism from a proper n-ring r of a Schauder decom-
position of X, onto a B-algebra II c B(X).
Definition 10. A topological isomorphism from r into B(X) is n-
maximal whenever it cannot be extended to a topological isomorphism
from r' into B(X), where r' is a n-ring properly containing r.
Theorem 11. Let r be the proper n-ring of a Schauder decomposition
of X. Then the isometric isomorphism A ofr into B(X) is n-maximal.
Proof. If A is not n-maximal, then there exists a topological isomor-

phism A' of an-ring r' which contains r properly, onto a B-algebra
II'::::J II. Let then r:t. be an element of r' which is not in r. Because Xnr:t.Er,
n=1,2, ... ,
(2)
i~n
holds for every x in X. Let K' be the constant of Definition 2.3 which
corresponds to r'. Then
Ili~n r:t.iPiXII ~ IIA'(Xnr:t.}llllxll ~ IIA'IIIIXnr:t.llr' Ilxll ~K'IIA'IIIIr:t.llr' Ilxll·
Since x=lim L PiX we have for each x in X and every e>O an index
n such that ~o:"';, q~n, IIJp PjXl1 <e. Thus
.t
II l-P r:t.iPiXII = II.L r:t.iPi.t PjXl1 <K'IIA'IIIIr:t.llr' e
t!ESq J-P
and since X is complete, the right hand side of(2} is convergent. Therefore
r:t. is in r and this contradiction leads to the desired result that A is n-
maximal.
The concept of a proper n-ring can be extended in an obvious way
to a weak* Schauder basis {xi, xi*} of X*. Let r* be the subset of all r:t.
OC!
in s such that L r:t.ixi*(x*}xi exists in the weak* topology for all x*

i= 1
in X*. For r:t.Er* we then define the mapping A+(r:t.}:X*--+X* by
A+(r:t.}x*(x}= L r:t.ixi*(x*}xi(x}, XEX, X*EX*.

i= 1
Since {xJ is a basis for X with biorthogonal sequence {xi} if and only if
{xi} is a weak* Schauder basis for X* with biorthogonal sequence
{xi*} such that xi*(x*}=x*(xJ for all X*EX* (III.2.7), we have
A+(r:t.}x*(x}= L r:t.ixi(x}x*(x i}, XEX, X*EX*. (3)

i= 1
Next, let r be the proper n-ring of {Xi} and A the corresponding iso-
metric isomorphism of r into B(X}. From (3) it is apparent that A + (Xnr:t.)
is the adjoint of A(Xnr:t.} for all n and, as a consequence of the uniform
boundedness principle (1.3. 14}, supIIA(Xnr:t.}II<oo. Since limA(xnr:t.}x
n n
exists in X for each x in the total set {Xi}' it follows from the Banach-
Steinhaus theorem that A(r:t.}x exists for all x in X. Hence r*cr and
by (1) and (3),
x*(A(r:t.}x}= A+ (r:t.}x*(x), XEX, X*EX*,
4. Minimal Schauder Decompositions 107
for alllY. in r. This immediately shows that A+(IY.)=A*(IY.), the adjoint

of A (IY.), and that r* = r (as sets). Defining analogously the norm on
r* by IIIY.II = IIA*(IY.)II and using the fact that B(X) is isometrically
isomorphic to B(X*) (1.3.23), it is clear that r* is a 1Hing, the proper
n-ring of the weak* Schauder basis {xi} for X*, and one gets the
following result:
Theorem 12. A weak* Schauder basis for X* and its corresponding
basis for X have the same proper n-ring.
4. Minimal Schauder Decompositions
Let r be the set of all Schauder decompositions of Banach spaces.

Then it may happen that the proper n-ring I' y of an element y in r is
contained (as a set) in the proper n-ring I' y' for all y' in r. In this case
we write
Definition 1. A Schauder decomposition y in r is said to be minimal if
rye I' y' for every y' in r; then r y is called the proper minimal n-ring of y.
To show the existence of a minimal Schauder decomposition we use
the following lemmas and remember that every basis for a Banach space
X automatically generates, by its basis elements, a Schauder decompo-
sition of X.
Lemma 2. The subalgebra w of s (cf Lemma 2.7) is included in the
proper n-ring I' yfor each y in r.
Proof. Let X be the corresponding Banach space of an arbitrary
y= {P;(X), PJ in r with proper n-ring r y • Now, for any IY. in w we have by
Abel's formula (which is to be verified by an elementary calculation)
(1)
i!f:n i~n j~i i~n
By Corollary VII.2.8 there exists a positive constant K such that

s~p II i~n Pixll ~Kllxll for every x in X. Therefore
.I ~,Pjxll
II l~n (lY.i-lY.i+ 1) }~l ~ ,I IlY.i-lY.i+ 11 II I, PjXl1
l~n )~l
~K Ilxll I IlY.i-lY. i+11, (2)

i~n
Combining (1) and (2) yields

IIA(XnlY.)xll = n
L~n lY.iPiXl1 ~K Ilxll (I IY. + 11 + i~n IlY.i-lY.i+1 1) ~KllxlllllY.llw·
Since {Pi(X), PJ is a Schauder decomposition of X we have for each x
in X and every 8>0 an index n such that for p,q~n, II.t P i xll<8.
Therefore, ,= p
.t
II I-P aiPixll= II.I
l~q
rt.iPi.t PjXII<KIIrt.llw8
J-P
and because X is complete, I rt.iPiX converges in X. Moreover,

i~n
IIA(rt.)11 ~K 11rt.llw for every rt. in w. This shows that w, as a set, is contained
in r.
We now introduce the ideal Wo of w, defined by
wo={rt.irt.EW, limrt.n=O}.
n
In other words we have Wo = wnc o. Then we show the following

Lemma 3. The sequence {bJ is a basis for the Banach space woo
Proof. We recall that the norm in Wo is given by 11rt.llw=supllXil

00 i
+ I lrt. i -rt. i + 1 1,rt.EWo· For any rt.in Wo, we have Xnrt.EWO and
i= 1
00
11rt.-XnlXllw~2suplrt.n+;I+ IIrt.n+i-rt.n+i+ll.
i~n i= 1
Since rt.EWocCo, the right hand side of the above inequality tends to
zero with increasing n. Thus sp {bJ = Wo0 On the other hand it is clear
that for m ~ n and any f3 in s,
.I f3ibill =
I l~n w
Ilxnf3llw~llxmf3llw= II.I f3ibill
l~m w
.
Hence, by Theorem IV.1.5, {bJ is a basis for W0 0
Lemma 4. W is the proper n-ring of the basis {bJ for wo°

Proof. We know from the preceding lemma that {b i } is a basis for
Wo and we denote the corresponding proper n-ring by r. By Lemma 2,
wcr. Conversely, we can show that rcw: Let rt.Er. From the Lemmas
n
3 and 2.7 we then infer suplrt.il+ IIrt.i-rt.i+ll=IIXnrt.llw+lrt.n-rt.n+ll
1 i= 1
-Irt.nl~ t~n rt.ibiL +1rt.n+ll=IIA(rt.)xnllw+llrt.bn+lll/llbn+lll~IIA(IX)IIIIXnllw

+111X11=3111X11, where 1IIIw is the norm used in the space Wo and 1111 is
the norm used in r. From the definition of the norm in W it is now clear
that 11rt.llw~311rt.11, hence that rt.EW. As desired,this implies r=w.
Theorem 5. The basis {bJ for Wo is minimal, its proper n-ring is W.
5. Banach Algebras and Unconditional Bases 109
This theorem is an immediate consequence of the preceding three

lemmas.
Theorem 6. A sequence of closed linear subspaces {M;} in a Banach
space X is a Schauder decomposition of X if and only if the following
conditions are satisfied:
(i) There exists a continuous isomorphism Afrom the n-ring w into B(X),
(ii) for each i, A(b i ) is a projection of X on Mi and
00
(iii) X =sp U Mi·
i= 1
Proof. In view of the inequality IIA(a)II::(K Iiall w , aEW, derived in

the proof of Lemma 2, the necessity is an immediate consequence of
Theorem 3.7 and Lemma 2. The sufficiency proof is analogous to that
of Theorem 3.7, where instead of r one only has to take the n-ring w.
5. Banach Algebras and Unconditional Bases
Definition 1. A basis for a Banach space X is of the first or second

category according to its proper n-ring is of the first or second category.
Theorem 2. A basis for X is of the second category if and only if there
is a positive constant M such that supIIA(Xna)11 ::(Mllall oo for all a in loo-
n
Proof. To show the necessity, let the basis be of the second category.
Then there exists a constant M>O such that Ilall::(Mllall oo for each a
in the proper n-ring r of the basis. This is clear by (1.2.6) since Iiall oo ::( Iiall
for each a in r (VIII.2.2) and since the identity maps r in a one-to-one
manner onto the Banach space r (with the topology of 1(0). Evidently
Xna is in rfor each a in 1 Hence IIA(Xna)11 = IIXnal1 ::( M IIXnal1 00::( M Iiall 00
00 •
and the assertion follows. On the other hand, suPIIA(Xna)11 ::(Mllall oo

n
for all a in r is a sufficient condition for the basis to be of the second
category, because then IIA(a)xll ::(supIIA(Xna)xll ::(Mllall oo Ilxll for every
n
x in X. Consequently, Iiall = IIA(a)11 ::(Mllalioo' showing that the conver-

gence in 100 implies the convergence in r. Thus, using again (VIII.2.2)
r = r and this concludes the proof of the theorem.
Theorem 3. A basis {Xi' x'[} for X is of the second category if and only
if it is unconditional.
Proof. Let {x;} be unconditional. A(a)x exists in X for each a in
100 and each x in X, because unconditional convergence of a series, by
Theorem 11.1.3, implies bounded multiplier convergence of the series.
Hence the proper n-ring of {Xi} contains 100 and since r c 100 we have, as
a set, r= 100 . We now proceed to define the transformations Tn: 100 -+B(X)
by T,,(IX)=A(XnlX),IXEloo,n= 1,2, .... Since for fixed IX in 100 , sup 11T,,(IX)xll
n
=supIIA(XnlX)xll < 00, XEX, we have by the principle of uniform bound-
n
edness (1.3.14), sup II T" (IX) II < 00, IXEloo- Evidently each of the transfor-
n
mations T" is bounded. Applying now the same principle again it then
follows that sup suP{IIA(XnlX)111111X11 00 ~ 1} = sup II Tnll < 00 so that by
n n
Theorem 2, {x;} is of the second category.
Conversely, if {x;} is ofthe second category, then IIA(XnlX)11 ~MlllXlloo
for all IX in 100 and all n. Let {n;} be an arbitrary increasing sequence of
integers and let IX in 100 be such that IX j = 1 for j in {n i } and IXj=O else.
For every 8>0 we have an index p such that Ilx-xpll <8/(2M), where
xp= L
Xr(X)Xi and where x is a fixed but arbitrary element of
{.L lXiXr(X)Xi}
i~p
X. We show that is a Cauchy sequence: If n~m> p,

z:::;;n
then Ilitm lXiXr(X) Xi \I ~ \lit lXiXr<X-XP)Xi\l + \lit lXiXr<Xp) Xi \I

=lieA(xnlX)- A(Xm-1 1X )] (x -Xp)11 ~2M Ilx - xpll <8. Since L lXiXr(X)X i
= L x*(x)x i, and {n;} was supposed to be arbitrary, the basic series
i~n
ni:::;;n
00
L xr(x)x i is sub series convergent, hence unconditionally convergent
i= 1
(11.1.3) and the theorem is proved.
Corollary 4. An element IX of s is in the proper n-ring of an unconditional
basis if and only if it is in 1 00 •
Corollary 5. {x;} is an unconditional basis for X if and only if the

following conditions are satisfied:
(i) There exists a topological isomorphism A of 100 into B(X),
(ii) for each i, A(b i) is a projection of X with one-dimensional range
spanned by Xi and
(iii) sp{x;} =X.
Proof. The necessity is a consequence of Theorem 3.7, the preceding

corollary and Theorem. 1.1.14. Conversely, with IIAII < 00 and in just the
same way as in the sufficiency proof of Theorem 3.7 it follows that {x;}
is a basis for X. Since the n-ring is 100 , {Xi} is, by definition, of the second
category and thus, by Theorem 3, unconditional. This finishes the proof
of the corollary.
Theorem 6. Let {xi,xn be an unconditional basis for X and let r be
its proper n-ring. Then the spectrum of A(IX), IXEr, is the closure of the set
5. Banach Algebras and Unconditional Bases III
of eigenvalues {CX1,CX2""} of A in C and the resolvent R(A,A(cx)) of A(cx)

is given by R(A,A(cx))=A(Ae-cx)-l).
Proof. Since A(cx)xi=lim

n
L CXjXj(Xi)Xj=CXiX i, each value ai in C is an
jf5;;n
eigenvalue of A(cx) with eigenfunction Xi' On the other hand, if A is not

in the closure of the set {CX 1,CX 2 , ... } in C, then we have it:J.fIA-cxil
, >0.
Now, from (AI -A(cx))x=lim L (A-CXi)Xt(X)x i, XEX, through multipli-
n i~n
cation of both sides by (A-cxT 1xj, one obtains (A-CX)-l xj(AI - A (cx)) X
=xj(x) for allj. Since X= lim L
xt(x)x i, the assumption (AI - A(cx))x=O
n i~n
then implies X= 0 for every x in X. Thus AI - A(cx) is one-to-one and this

property shows that {AI_A(CX)tl exists. Because (Ae-cx)-l is in 100 ,
Corollary 4 implies that (Ae-cx)-l is in r. Therefore, A(Ae-cx)-l) is
bounded on X. Hence A(Ae-cx)-l)x= (AI - A(cx)t 1 (AI - A (cx))
L
A(Ae- CX)-l)X = (AI - A(cx)t llim (A -cxr lxt(X)(A- CXi)X i = (AI - A (cx)t 1X
n i~n
for all x in X, so that R (A, A (cx)) = (AI - A(cx)t 1 = A(Ae- CX)-l). Finally,
since the spectrum of A(cx) is closed in C (I.2.20), and since R(A,A(cx))
is bounded for each complex A which is not in the closure of {CX 1,CX 2, ... }
in C, the proof of the theorem is finished.
For the special value A=O it follows
Corollary 7. Under the conditions stated in the foregoing theorem,
whenever cx is in r, A(cx) has a bounded inverse if and only if it:J.flcx;l >0.
If this is the case, A(CX)-1=A(cx- 1). '
The topological isomorphism A may be used to investigate relations
between the unconditional bases for X and the orthogonal bases for X.
The last concept is explained in the following definition.
Definition 8. A basis {x;} for X is orthogonal if and only if the ine-
quality
is satisfied for all sequences cx in s and all disjoint subsets J1 and v of 1:,
where 1: is the set of all finite subsets of the set of positive integers.
Theorem 9. (SINGER) Each unconditional basis for X is an orthogonal
basis for a Banach space X', where X' is the same linear space as X,
supplied with an equivalent norm. .
Proof. Let {x;} be an unconditional basis for X. We then define a
new norm on the linear space X by
This yields
Ilxll = IIA(e)xll :::;;sup{IIA(a)xll!lIalloo:::;; 1}= Ilxll'.
°
On the other hand, as we have shown in the proof of Theorem 2, there
is a constant M > such that
Ilxll' :::;;sup{MlIall oo Ilxll!lIalloo:::;; 1} =Mllxll·
This shows that the primed and the unprimed norms on X are equivalent.
Moreover, for 13 =1= 0,
IIA(f3)xll' = sup{IIA(a)A(f3)xll!lIalloo:::;; 1}
:::;;sup{IIA(af3)xll\lIall",:::;; 1}
= 11131100 sup{ IIA(a 13) xll/llf3ll 00 Illall 00 :::;; 1}
:::;; 111311 "'_ sup{IIA(a)xll!lIall 00:::;; 1} = 11131100 Ilxll'·
Hence the inequality IIA(f3)xll':::;; 11131100 Ilxll' holds for every 13 in 100 and
every x in X. Let XuE/", be the characteristic function of the set flEL.
Choosing 13 = XJl and =. x L aixi,
IEJlUV
VEL, aEs, we get II?= aixill' lEJl
= IIA(XJl)xll':::;; IlxJlII",llxll' = Ilxll' = II L aixill' and this completes the proof

of the theorem. iEJlUV
References for Chapter VllI: KADEC [3], SINGER [8] and YAMAZAKI
[1-3].
CHAPTER IX
Some Results on Generalized Bases

for Linear Topological Spaces
Having introduced the basis concept for Hilbert and Banach spaces,
certain generalizations appear to be at least as important. First of all
one discards of almost all requirements used for the definition of a basis.
Beginning with the absolute minimum one takes a linear topological
space X, does not use the concepts of totalness and countability and
avoids all mention of series expansions. Thus, the starting point will be
a family {x,,} of elements of X and a biorthogonal family {f,,} of
continuous linear (coefficient) functionals on X. The biorthogonal system
{x"J,,} is said to be maximal if there is no biorthogonal system in which
it is properly contained. A biorthogonal system with respect to X is
called a generalized basis for X if, in addition, XEX and !,,(x)=O for
all A implies x = O. A generalized basis is always a maximal biorthogonal
system. If the set of basis elements {x ,,} of a biorthogonal system
{x"J,,} in X is total in X, then {x"J.J is called a dual generalized basis
for X. Moreover, if such a basis is also a generalized basis for X, it is
called a Markushevich basis for X if the set {x,,} is countable, and an
extended Markushevich basis for X if {x,,} is not countable. Finally,
introducing again the concept of a series expansion for elements of X,
then a Markushevich basis for X becomes a Schauder basis for X. Hence
one has the following hierarchy in terms of increasing generality:
Schauder basis
~
Markushevich basis
~
extended Markushevich basis
(= total generalized basis)
~~
generalized basis dual generalized basis
~ ~
maximal biorthogonal system --------» biorthogonal system
114 IX. Some Results on Generalized Bases for Linear Topological Spaces
Now, let {x .. ,f.. } be a generalized basis for X. It is shown that {x .. } is

unique. If X is locally convex, it is also shown that {f..} is unique if
and only if {x .. } is total in X. On the other hand, the elements {x .. }
of a dual generalized basis for a locally convex space X are unique if and
only if that basis is also a generalized basis for X. Furthermore, if X is
also separable and Hausdorff, there exists a dual generalized basis for
X, and there is a Markushevich basis for every separable Banach space.
It is worth noting that the basis problem is now solved for separable
locally convex linear topological spaces. But it is, as we know, not yet
solved for the special case of separable Banach spaces. The solution of
the first problem is given in the form of a counter-example (Theorem 1.11):
It has been shown that the (Banach) space I:' is weak* separable, but
has no basis in the (locally convex) weak* topology. It is also known
that there are separable F-spaces with no basis (Corollary 5.3).
After presenting some examples in special spaces we investigate
generalized bases for complete linear metric spaces with translation-
invariant metric, X and Y, which are similar and show that such bases
exist if and only if X and Yare topologically isomorphic. In Banach
spaces, bases are similar if and only if they are equivalent. If X = Y and
if {x;} is a Schauder basis for the complete linear metric space X with
translation-invariant metric, then, as a special case, we obtain the Paley-
Wiener theorem which gives sufficient conditions for a sequence in X
to be also a Schauder basis for X which is similar to {x;}. The final
section deals with continuity of the coefficient functionals of bases in
complete linear metric spaces and with the connection of bases for the
weak and the initial topology respectively in locally convex and in
barreled linear topological spaces. The result that every basis for a
Banach space is a Schauder basis for that space cannot be generalized
to locally convex linear topological spaces, since there exists a basis for
the locally convex linear topological space 11 (endowed with its weak*
topology) which is not a Schauder basis (cf. Example 111.7.4).
1. Definition and Fundamental Properties

of Generalized Bases
Let X be a linear topological space over the field cI>, X* its conjugate
space and let A be an index set of arbitrary cardinality.
Definition 1. The do.uble-family {x .. , f .. } is a bio.rtho.go.nal system
if X .. EXand f .. EX* fo.r all AEA and if f ..(x,.}=b ..f.I' where b..f.I is the
Kro.necker symbol. {x .. ,f.. } is maximal with respect to. X if there is no.
bio.rtho.go.nal system which co.ntains {x .. ,f.. } pro.perly.
1. Definition and Fundamental Properties of Generalized Bases 115
Definition 2. A biorthogonal system {XA,!A} is a generalized basis

for X if fA (x) =0, AEA implies x=o for all x in X. {XA,!A} is total if
the finite linear combinations of {x A} are dense in X (i.e. if {x J is total
in X). If A is (not) countable, such a basis is called an (extended) M arkus-
hevich basis for X.
We note that by the condition fA(x)=O, AEA=>x=O, {fA} is total
over X, as well as total in X* in the weak* topology for X* (BOURBAKI
[1], III-V, p. 51).
Theorem 3. A generalized basis for X is a maximal biorthogonal
system with respect to X.
Proof. If there is an x in X and an f in X* for which f(x) = 1 and
fA(x)=O on A we have indeed x=O. Thus f(x)=O and this contra-
diction verifies the theorem.
We observe that a maximal biorthogonal system may fail to yield a
generalized basis. This is shown in the last example of paragraph 3. As
in the case of a basis we often refer to the set {x J as a generalized basis
for X and {fA} will be called the family of coefficient functionals corre-
sponding to {x A}'
Let now D(A) be the linear space of all scalar valued functions on A
in which a function z is said to be zero if and only if Z(A) = 0 for every
A in A. Let the coefficient mapping F: X -+ D(A) be the linear transfor-
mation defined by F(x)= {fA(X)}. Then the following theorem is
evident from the definition of a generalized basis.
Theorem 4. A biorthogonal system {x A,!A} is a generalized basis for
X if and only if F is one-to-one.
An important consequence of this theorem is the uniqueness of the
basis elements of a generalized basis.
Theorem 5. A given family {fA I AEA} in X* can be the family of
coefficient functionals for at most one generalized basis {x J for x.
Proof. If there is another generalized basis, {x~, fJ for X, then
fA(x~) = b AI' = fA(xl')' A, f.lE A.
Hence x~ = xI' for all f.lE A, since fA(x~ - xI')
= fA(x~)-fA(xl')= 0, AEA.
Theorem 6. (ARSOVE) A generalized basis {x A} for a locally convex

linear topological space X has a unique family {fA} of coefficient func-
tionals if and only if it is total.
Proof. Given a total generalized basis {XA,!A} for X which has

another family {fn
of coefficient functionals, distinct of {fA}' then
f~(xl')=fA(xl')=bAI',}',f.lEA. Thus (f~-fA)(x)=O for every fixed A in A
and every x which is a finite linear combination of x;s. From the total-
ness of {x;.} and the continuity of f~ -fA we deduce that (f~ -fA)(X) =
for each x in X, hence that {fA} is unique. On the other hand, suppose
°
that {f.,.} is unique, but that {x A} is not total with respect to X. Then,
by Theorem 1.3.6 there is a non-zero fin X* such that f(x .. )=O on A.
Let v be a fixed element of A. (f.. +b"vf) (X/L)=b"/L shows that
{X ..,f.. +b ..vj} is a biorthogonal system with respect to X. Assuming
(f.. +b .. J)(x)=O for some x in X we infer that f .. (x) = - f(x)f.. (x v)
=jA(-f(x)x,), or in other words, F(x)=F(-f(x)x.). But since F is
one-to-one one has X= -f(x)x v • Hence f(x)=f( -f(x) xv) = -f(x)f(x v )
=0 and f .. (x)=O, ..lEA, which implies that x=o. Thus beside {x .. ,f;.}
there is another generalized basis for X, {x .. ,f.. +b"vj}, distinct from
{x .. ,f.. } and this contradiction leads to the conclusion that the family
{x .. } is total in X.
Theorem 7. (MARKUSHEVICH) There exists a M arkushevich basis for
every separable Banach space X.
Proof. By hypothesis there is a total sequence {xJ in X in which

every finite subset of elements is linearly independent. According to
the lemma of BANACH-MAZUR (Iy'3.9) there exists an isometric iso-
morphism T of X into the Banach space C [0, 1]. Evidently, sp {TxJ
= T(X). By the orthogonalization process used in the proof of Theorem
V1.1.4 to construct an orthonormal basis for a separable Hilbert space H
from a total sequence for H, one obtains from {T xJ likewise an ortho-
normal set, {Yi}' in C[0,1J if the inner product in C[O, 1J is defined
1
by (y,y')=Sy(t)Y1t)dt,y,y'EC[0,1]. Of course, in this case, the
o
sequence {yJ derived from {Tx i } is also total, but must not necess-
arily be a basis for T(X). Now, let Zi= T- 1 Yi and let the sequence
{zt} in X* be defined by zt(x)=(Tx, Y;), XEX, observing that Ilztll
~ II TIIIIYill· Because zt(z)=(T T- 1 Yj' Yi)= (Yj' Yi)= bij, {Zi' zt} is a
biorthogonal system for X and, obviously, one has sp {z;} = X. Assum-
°
ingnow zt(x)=O foralliandagivenxinX,onehas (TX'Yi)=Zt(X)=O
for all i. Since for every e > there is a set ()( 1, ... , ()(n in If> such that
IITx- i~n ()(iYill<e, one has
(Tx, Tx)=(Tx, Tx)- I I(Tx,y;)1 2

i~n
i~n i~n
=(Tx- I ()(iY;, Tx- I ()(iYi)

i~n i~n
1. Definition and Fundamental Properties of Generalized Bases 117
<8 2 .
The arbitrariness of 8 implies that Tx(t)=O almost everywhere on

[0, 1] and, because this function is continuous, that T x = 0. From the
fact that T is one-to-one one infers that x=O. Thus {Zi' zr} is a Markus-
hevich basis for X and the theorem now follows.
Definition 8. A sequence {xJ in a linear topological space X is a
basis for X if for each x in X there is a unique sequence {aJ in cP such
that x=lim L
aix i in the topology ofX.
n i~n
Evidently, each expansion coefficient ai' by h(x)=a i, defines a

linear functional h on X. However, as is shown later in section 5, the
coefficient functionals h must not necessarily be continuous.
Definition 9. A basis for a linear topological space X with continuous
coefficient functionals h, defined by h(x) = ai' XE X, i = 1, 2, ... is called
a Schauder basis for X.
Theorem 10. Every Schauder basisfor X is a Markushevich basisfor X.
Conversely, a Markushevich basis {xi,h} for X is a Schauder basis for
X if and only if x=lim L
h(X)Xi for every x in X.
n i~n
Proof. Let {Xi' h} be a Schauder basis for X. The first assertion
follows from the obvious fact that the sequence {xJ is total in X and
from the implication h(x)=O, i= 1, 2, ... =x=lim L
h(X)Xi=O. On the
n i~n
other hand, let {Xi,h} be a Markushevich basis such that x = lim L h(X) Xi
n i~n
for each XEX. In this case, for each x in X, the sequence U;(x)} of
expansion coefficients in cP is unique. This follows from the biorthogona-
lity property h(x) = [V Let {fJJ be another set of expansion coeffi-
cients for x. Then one has 0=h [li~ .L
(fj(x)- fJj}X j] =h(x) - fJi for all i.
)~n
This completes the proof of the theorem.

The following theorem gives, in the form of a counter-example, the
solution of the basis problem for locally convex spaces.
Theorem 11. If the weak* topology is assigned to I!, then I! is a
locally convex separable linear topological space which has no basis.
Proof. By (I.3.22), J(ll) is weak* dense in l'l'*( = I!). Since J(ll) is

separable (l.4.b), and hence separable for the weak* topology of I!,
it follows that I!, endowed with its (locally convex) weak* topology,
is separable.
On the other hand, if one assumes that I! has a weak* basis {xr},
then to every x* in I! there is a unique sequence {a;} in cp such that
x* = lim L ai xT in the weak* topology of I!. But by (1.4.c) it is apparent
n i~n
that the series converges weakly to x*. Thus {xr} is a weak basis for
I"! which, since I"! is a Banach space, turns out to be a basis for I! (111.2.5).
Thus I"!, and so also 100 (1.3.11) would be separable in its strong topologies.
This contradiction (1.4.c) completes the argument.
2. Dual Generalized Bases
Let X, X*, F and A have the same meaning as defined at the beginning
ofthe preceding section. There are then two types ofbiorthogonal systems
{xA,fJ:
(i) {x A,fA} is such that F is one-to-one.
(ii) {Xl,fA} is such that {x A} is total in X (i.e. sp{xA}=X).
It follows that the set of all biorthogonal systems of type (i) are the
generalized bases and that the set of all biorthogonal systems satisfying
(i) and (ii) are the (extended) Markushevich bases. As we shall see, the
biorthogonal systems oftype (ii) are in some sense dual to those oftype (i).
Definition 1. A biorthogonal system satisfying (ii) is called a dual
generalized basis for X.
Theorem 2. (KLEE) If X is separable, locally convex and HAUSDORFF
then there exists a dual generalized basis for x.
Proof. Since X is separable there is a sequence {yJ with sP {yJ = X,
and without loss of generality {Yi} may be assumed such that every
finite subset of it is linearly independent. Let N n be the linear subspace
of X spanned by {Y1, ... ,Yn}. First of all, let X1=Y1 and X2=Yz. Then
there are linear functionals f1 and f2 in N! with the properties f1 (x 1) = 1,
f1 (x 2)=O and f2(X 2)= 1, f2(X 1)=O (1.1.2 and 1.3.6). By (1.3.5) both f1
and f2 have continuous linear extensions x! and x!, respectively, onto
all of X. Thus x!, X!EX* and xT(x)=b ij , i,j= 1, 2. We now proceed
recursively. Let {x 1, ... ,xn,x!, ... ,x:} be such that sp{x 1, ... ,xn}=Nn
and that xT(xj)=b ij , i,j,,;;n. We observe that by the biorthogonality
relations, the set {Xl> ... , x n } is linearly independent. Then we define
Xn+1=Yn+1- L XT(Yn+1}X i, and it is clear that the set {X1,···,X n+1}
i~n
is linearly independent and that SP{X1, ... ,Xn+1}=Nn+1. Finally, as

above, we find X:+ 1EX* such that x:+ 1(x)=O, xEN n, X:+1(X n +1)=1.
Therefore, by a simple computation, xt(x)=b ij , i,j,,;;n+ 1. This shows
2. Dual Generalized Bases 119
that by induction one can obtain a biorthogonal system {x;,xt} for X

with sP {x;} = X, which is a dual generalized basis for X.
The following theorem appears to be dual to Theorem 1.6.
Theorem 3. A dual generalized basis {x A,fA} for a locally convex
linear topological space X has a unique family of basis elements {x A} if
and only if the corresponding coefficient mapping is one-to-one (i.e. if and
only if {x A,JA} is also a generalized basis).
Proof. Let {x~} be another family of basis elements which is distinct
of {x A}, but has the same family of coefficient functionals. Then F(x~ - x,,)
={JA(X~-X,,)}={OA,,-OA"}=O for all ilEA. The uniqueness now
follows assuming that F is one-to-one.
If, on the other hand, we make the hypothesis that F is not one-to-one
we can show that the family {x J is not unique. By this hypothesis,
there is a non-zero x in X such that Fx=O, hence that fA(x)=O on A.
Let then v be a fixed element of A. The equations fA(x"+o,,vx)=oA'"
A,IlEA, show that {XA+OAVX,fJ is a biorthogonal system with respect
to X. Moreover, since XEX and since sP {X A} =X one has, as we see
later, SP{XA+O.;.VX}=X. This shows that there is another dual gener-
alized basis for X, {XA +OAVX,JA}, which is distinct from {XA,JA}'
hence that {x A} is not unique. Thus, assuming {x A} to be unique it
follows that F is one-to-one and this is the desired result.
It remains to show that SP{XA+OAVX}=X. Let N be a given neigh-
borhood of 0 in X. Since X is locally convex and fv is continuous, N
contains a convex circled (Theorem I.1.4) neighborhood N' of 0 such
that /v (N') C {IY.IIIY.I < 1, IY.E CP}. Let now y be an arbitrary fixed element
in X. Since sP {x J = X there is a linear combination L PAX A' where
Ay is a finite subset of A containing v, such that 2(y- L PAXA) EN'.

"EA,.
On the same reasons there is a finite linear combination L IY.AX A such
that 2(1+ IPv l)(x- L IY.AX"AEAx
)
AEAx
EN'. From the assumptions it follows
that IlY.vl= Ifv(x- L IY.AXA)I«l+IPvl)-l if vEAx. We have with

lEAx
Axy=AxuAy and the convention that IY.A(PA)=O for A¢Ax (¢Ay,

respectively),
L
AEAxy
YA(XA+O AV L
JLEAxy
lY."x,,)= L
v=/=. AEAxy
(PA-YVIY.;)xA+yv(xV+ L
AEAxy
IY.AXA)
9*
where Yv=(l+a v)-lpv and h=PA-YVaA' 2¥v, turn out to be finite

scalars. But then,
y- L h(XA+bAVx)= y- L h (XA +bAV L a/lx/l)

AEAxy AeAxy IlEA xy
- Yv (x- L a/lx/l)
lLeA xy
=y- L PAXA-YV (x- L aAx A).

AeA xy AEAxy
Since IYvl:::;IPvl(l-la vl)-l<l+IPvl one gets 2yv (x- L aAx A) EN'.

AeA xy
As a consequence of the convexity of N', it finally follows that
y- L YA(xA+bAvX)EN' and the statement sp{XA+bAVX}=X is
AEAxy
verified.
3. Examples
Example 1. Let B(A) be the Banach space of all bounded functions
x:A~cP, the norm being Ilxll =sup{lx(2)1!2EA}. For every 2 in A we
define XA=XA' where XA(Il)=b A/l is the characteristic function of 2. The
functionals fA' defined by fA(x)=x(2) are obviously bounded, hence
continuous, and the set {x A,fA} is biorthogonal. {x A,f;,.} is a generalized
basis for B(A), since in addition to the properties shown above, fA (x) = 0
on A implies x(2)=O on A, hence x=O.
Example 2. In the preceding example we take A to be the set of positive
integers. Then B(A) = 100 , and though 100 is non-separable, it has a countable
generalized basis.
Example 3. Let Y be the subspace of 100 of all elements x such that
lim(l/n) L x(2) exists. It is clear that Y is closed in 100 and that {XA,j~}
n ).~n
of the foregoing examples provides a generalized basis for Y. In Y, the

functional f on Y, used in the proof of Theorem 1.6 may be given explicitly:
If f: Y ~CP is defined by f(x)=lim(l/n) L x(2), thenfis evidently non-
n
A~n
zero, but vanishes at each xA. Thus {x A' fA} is not a dual generalized basis
for Y.
Example 4. Let A(D) be the Banach space described in 1.4.f. If Zo is
any real number such that 0 < Zo < 1, every x in A(D) has the Taylor series
expansion
x(n)(z)
L __0 (z-zot
cy,
x(z)= (1)
n=O n!
3. Examples 121
in {zllz-zol<1-zo}cD. Let now {xn} bea sequence offunctions in

A{D), given by
(2)
and let Un} be a sequence of linear functionals on A{D), defined by

x(n)(zo)
fn(x) = - - ,- , n=O, 1,2, .... (3)
n.
That the functionals fn are continuous becomes clear by the Cauchy
estimates Ix(n)(zo)1 =n! sup{lx(z)lllz-zol ~ 1-zo}/{1-z o)" from which
follows Ilfnll ~(1-zo)-n. Since fn(xm)=Dnm and since fn{x)=O, n=O, ... ,
implies x(z)=O in {z Ilz-zol < 1-zo} and, by analytic continuation,
in the whole of D, the system {xn.!n} is a generalized basis for A(D).
00
However, {xn.!n} is not a Schauder basis for A{D), because L (z - zo)"

n=O
is the Taylor series of the function (1 + Zo - z) - 1 in the open disc
{z Ilz-zol < 1}and because this series is unique (Theorem 1.4.13), but
not convergent at the point z=zo-1 inDo
Example 5. Let {D k } be a sequence of closed discs in C, given by

Dk={zllzl~1-rk}, k=1,2, ... and let each C(D k) be the Banach
space of continuous complex functions on D k. In the linear space AF{D)
over the field C, of all holomorphic functions on D = {z Ilzl < 1}, the
00
function IIII :AF{D)-+~, given by Ilxll = L rkllxllk/(l + Ilxllk), where

k= 1
Ilxllk is the norm in C(D k ) , describes a quasi-norm on AF(D). We show
that AF(D) is complete in this metric and hence an F-space. For this
purpose, let {xn} be a Cauchy sequence in AF(D). Since Ilx11J(1 + Ilxllk)
~2kllxll we have Ilxllk~2k+lllxll for Ilxll~2-k-1. Since each of the
spaces C(D k) is complete, xn{z) converges pointwise and uniformly on
each Dk. Therefore, the limit is a function x(z) which is holomorphic on D,
such that limllx-xnll =0, and the assertion is verified.
n
In the same way as in the preceding example, it can be shown that the
system {xn' fn} given by (2) and (3) is a generalized basis, but not a
Schauder basis for AF(D). The only change in the proof is for the
continuity of fn. We have Ifn(x)I=lx(n)(zo)l/n!~llxIIJ{1-2-k-zo)n
~2k+lllxll/{1-rk-zo)n, if k is chosen such that zo<1-rk and if
Ilxll is taken smaller than r k - 1 • As a result of WALSH [1, p. 26J the poly-
nomials in Z are dense in each C(Dk)' Hence {xn} il total in AF{D)
which means that {xn.!n} is a dual generalized basis for AF(D), and,
by the above, a Markushevich basis for AF(D).
On the other hand, we take a sequence {xn} in AF(J)), given by

n
xn(z)= L Zi, n=O, 1,2, .... Furthermore, the equations

i=O
X(n) (0) x(n+ 1)(0)

fn(x) = ---;;! - (n+ I)!' n=O, 1, ... ,
define a sequence of linear functionals on AF(D). In quite the same

way as above it can be shown that eachfn is continuous, hence an element
of AF*(D). Since fn(xm)=O for m<n and for m?:n,
fn(xm)
[1 m
= -'-.1.... (. _ )'
'"
.,
l. .
z,-n -
1 - U~ nm L
m .,
1. zi-n-l
]
n. ,=n 1 n. (n+l)! i=n+1(i-n-l)! z=O
= 1- (1- bnm )= (\m,

the system {xn,fn} is biorthogonal. To verify that {xn,fn} is maximal,
we assume the contrary. This would imply the existence of a non-zero
element f of AF*(D) vanishing at each X n • In view of the fact that each
function in the space AF(D) may be approximated pointwise and uni-
formly on each Dk by polynomials in z (WALSH [1], Theorem I.17), the
sequence {zn}, and hence also the sequence {xn}, is total in AF(D}.
Hence for every x in AF(D) there exists a sequence {Yn} in AF(D}
such that each Yn is a finite linear combination of elements in {xn}, and
such that limYn=x. Thus f(x)=limf(x-Yn}+limf(Yn}=limf(x-Yn)
n n n n
= 0, as a consequence of the continuity of f. Since x was arbitrary we
have f=O, which is the desired contradiction.
However, {xn,fJ is not a generalized basis for AF(D), since it is
quite easy to find a non-zero x in A F(D) for which fn(x) = is true for
all n=0,1,2, ... : Let x(z)=(I-z)-l. Then fn(x)=[(I-z)-n-l
°
-(1- z)-n-2]z=0 =0, n=O, 1, ... , but Ilxll ?:tlixil d(1 + IlxI11)=t'2/(1 +2)
= 1/3> 0. But from the preceding considerations it is clear again that
{xn' fn} is a dual generalized basis.
4. Similar Bases
In this section we consider a special relationship between generalized

bases for complete linear metric spaces with translation-invariant metric,
the similarity of such bases.
Definition 1. Let X and Y be complete linear metric spaces with
translation-invariant metric, let A be an arbitrary index set and let {x)J,
{y"J be generalized bases for X and Y respectively, which have the same
4. Similar Bases 123
index set A. {x;.} and {y;.} are similar, if there exist families {f;.} and
{g;.} of coefficient functionals for {x;.} and {y;.} respectively such that
F(X)= G(Y), where F and G are the coefficient mappings determined by
each family of coefJicient functionals.
The definition naturally implies that both X and Yare over the same
field CP.
Theorem 2. (ARSOVE-EDWARDS) {xJ and {y;.} are similar if and
only if there exists a topological isomorphism T of X onto Y such that
y;.=TX;.,AEA.
Proof. To show the sufficiency let T be a topological isomorphism of
X onto Y such that y;. = T x;. on A. Given the family of coefficient func-
tionals {f;.} for {x;.} wechoose {gJ for {Y;.} such that g;.(y)=f;.(T-1y),
yE Y. Since g;.(Yp.)=f;.(T- 1 Txp.) =f;.(xp.) = b;.p. and since g;.(y)=O on A
implies T- 1y=O and thus y=O, {g;.} is a family of coefficient func-
tionals for {yJ and from f;.(x)=f;.(T- 1 Tx)=g;.(Tx) we have the
result that F (X) = G( Y).
For the necessity condition we put Z = F(X) = G(Y). Observing that F
and G are ono-to-one one has two possibilities of metrizing Z. Let Px and
py be the metrics defined in X and Y respectively. Then one can define
as a metric on Z either p~ by p~(z, 0) = px(F- 1 z, 0) or p~ by p~(z, 0)
= py( G - 1 Z, 0), Z E Z, and in both cases Z becomes a complete linear
metric space (with translation-invariant metric). We can show that p~
and p'y define equal topologies for Z.
Taking p = p~ + p~ it is evident that p, again is a translation-invariant
metric on Z. If {zn} is a Cauchy sequence in this metric, then {zn} is
also a Cauchy sequence in both metrics p~ and p~ on Z, and there exist
limits Zx and Zy of {zn} in the p~ and p~ topologies respectively.
Since each functional f;. is continuous we infer that f;.(F- 1 zx)
=f;.(F-l(p~-)limzn) =f;.((px-)limF-1z n) = limf;.(F-1z n) = limzn(A)
n n n n
= lim g;.(G- 1 Zn)= gj(py- )lim G- 1 Zn)= g;.(G-l(p~_ )limzn )= g;.(G- 1Zy)
n n n
=Zy(A)=f;.(F-l Zy) for every A in A. Because {x;.,J;'} is a generalized
basis we inferlhat F-1zX=F-1z y, and, since F is one-to-one, that
Zx=Zy. Hence {zn} converges in the metric p to the point Zx. This
shows that Z is complete in the metric p.
Because the topologies induced by p~ and p~ are both weaker than
that defined by p, it follows from (1.1.8) that p~ and p~ induce the same
topology for Z. Therefore, F and G are topological isomorphisms of X
and Y, respectively, onto the linear topological space Z (1.2.6). Defining
T by T=G- 1F, T is the required topological isomorphism of X onto
Y and, finally we have TX;.=G-1F x;.=G- 1{fp.(x;.)}=G- 1{bp.;,}=y;'
which concludes the proof of the theorem.
Since T, as a topological isomorphism, preserves additional properties

such as totalness of basic sequences or convergence of series in X, one
has the following
Corollary 3. If a generalized basis for a linear metric space with
translation-invariant metric is an (extended) M arkushevich basis or a
Schauder basis, then the similar generalized bases have the same properties.
Moreover, bases for Banach spaces are similar if and only if they are
equivalent.
The last statement in the corollary is a consequence of Theorem IV.3.2.
Theorem 4. Let {x)J (AE A) be a total generalized basis for the com-
plete metric linear space X with translation-invariant metric (p(x, y)
=p(x-y,O)=llx-yll, X,YEX). If {h}(AEA) is a family of points in X
and rx, 0 < rx < 1, is such that
for all finite sequences AI,"" An in A and all finite sequences rx 1 ,··., rxn
in ([>, then
(i) {y;J is a total generalized basis for X (which is similar to {x,,}),
(ii) and there is a topological isomorphism T of X with itself such that
y,l = Tx,l on A and (l-rx)llxll ~ IITxl1 on X.
Proof. Let D be the set of all finite linear combinations of elements in
the set {X,l}. One has D=X, since {x,,} is total inX. Let U:D--->X be
a linear operator defined by Ux= I f"(x)(x,, - h), xED, which, by the
AEA
inequality of the hypothesis, is uniformly continuous. Hence U has a
unique uniformly continuous extension V:X --->X (1.2.5). Moreover, the
hypothesis implies that Iwnxll ~rxnllxll for all n;?:O and all x in X.
Now, for S:X --->X, given by the absolutely convergent series expan-
00
sion SX= I vkx, XEX, it follows IISxll~(l-rx)-lllxll, XEX. On the

k=O
following reason S is one-to-one. Suppose that S x = O. Due to the abso-
lute convergence of the series for S x we then have for every 8> 0 an
index n such that Ikto V k XII < 8/2 for each m;?: n, and this shows that
Ilxll~[t~ Vkx\\ + \\Vkt/kxll <8/2+rx8/2<8. Ther!fore, S is! continuous

isomorphism of X into itself. Since S(x - V x) = I Vk X- I Vk+ 1 X= x,
k=O k=O
it is clear that T:X --->X, defined by Tx=x- Vx, XEX, is the (continu-
ous) inverse of S, hence a topological isomorphism of X with itself. From

the definition of V, Tx .. =y .. , AEA, follows immediately. If g .. :X-+iP is
defined by g.. (y) = f .. (T- 1 y), YEX, it is apparent that {y..,g .. } is a
generalized basis for X which is similar to {x .. ,!.. }. Finally, because
T(D) = T(D) = X, {y .. } is total in X and the theorem is proved.
As a corollary of the above theorem one obtains the famous Paley-
Wiener theorem. The theorem originally was derived in the framework
of the Hilbert space L2 in 1934 (PALEY and WIENER [lJ, p. 100). Then
it has been generalized for Banach spaces (BOAS [1 J, SCHAFKE [1 J) and
finally settled down for complete metric linear spaces by ARSOVE [6].
Let now X be the space defined in Theorem 4.
Theorem 5. (PALEY-WIENER) Let {Xi} be a Schauder basis for x.
If {yJ is a sequence in X and IX is a real number in (0,1) such that
for all finite sequences IX 1, ... , IXn in iP, then

(i) {yJ is a Schauder basis for X and
(ii) there is a topological isomorphism T on X onto itself such that Yi = TXi,
i=1,2, ... , and(l-IX)llxll~IITxll onx'
Proof. In Theorem 4 we define A to be the set of all positive integers
and we infer that {yJ is a Markushevich basis for X and that (ii) holds.
Due to the existence of T each element y in X then has the series expan-
OCJ
sion I IXiYi, where {IXJ is the coefficient sequence of the series expan-
i= 1
OCJ OCJ
sion I IXiX i for T- 1 y. The first series is unique, since I IXiYi=O im-
i= 1 i= 1
OCJ OCJ
plies I IXiXi=T- 1 I IXiYi=O, and so IXi=O, i=1,2, ... , according to

i= 1 i= 1
the assumption that {xJ is a Schauder basis for X. Hence {yJ is also
a Schauder basis for X and we are done.
5. Continuity of the Coefficient Functionals
In this section we establish some generalizations of the theorem that

every basis for a Banach space is a Schauder basis (111.1.3) and of the
fact that every weak basis in a Banach space is a basis (111.2.5). We
observe that the concept of a basis for a linear topological space X is, in
some sense, more general than that of a generalized basis for X, in that
the coefficient functionals of a basis for X may fail to be continuous. It

is easy to give an example of a basis which is not a Schauder basis (cf.
also to Theorem IIL7.4):
Example 1. Let Y be the space of all real functions expandable as
absolutely summable power series on the interval [0,1) with the topology of
uniform convergence on compact subsets of [0,1). Let {x;} be the set in
Y defined by Xi{t) = ti, i = 0,1, .... Then {x;} is a basis for Y which has
coefficient functionals which are not continuous.
Proof. From the uniqueness of the expansion coefficients of in [0,1)
absolutely summable power series (L4.13) it follows that {xJ is a basis
for Y. It is clear that for each polynomial p in t, the coefficient func-
tional fl, determined by the expansion coefficient (Xl' is given by fl(p)
=limC 1 (p(t)-p{0)). By the following procedure we can find a sequence
1-0
{yn} in Y such that limYn=O in the topology of Y, but such that
n
limfl{Yn) = 1. Since (by 1.4.5) the polynomials are dense in the Banach
n
space C [0,1] we can choose to each n a polynomial Zn in Y such that
sup { IZn(t) - (1- n t) I I tE [O,lln]} ~ lin and that sup{ IZn(t)11 t E(lin, I)}
t
J
~ lin. Obviously Yn' defined on [0,1) by Yn{t)= zn(t/)dt' is a polynomial.
o
Therefore, fl(Yn)=limC 1 Yn{t)=Zn{O)E[l-l/n, 1 + lin] for n= 1,2, ....
t_O
1
But because suP{IYn(t)11 tE[0,1)}~Jlzn{t)ldt~1/(2n)+1In=3/(2n), the
sequence {Yn} converges to °in theo topology of Yand the assertion is
verified.
Theorem 2. (NEWNS) In a complete metric linear space X over IR (or
C) which has a translation-invariant metric, every basis for X is a Schauder
basis.
Proof. Since the metric p in X is translation-invariant we use the
.I II.I I
notation Ilxll =p(x,O), XEX. Let {Xi'!J be a basis for X. Since for each
x in X, J;(X) Xi converges to x, it is clear that Ilxll /= s~p fJX)Xi < 00.
l~n l~n
Therefore p'{X,y)= Ilx- yll ' defines a new metric on X which is stronger
than p. Later on we shall show that X is also complete in the metric p'.
Theorem U.S then ensures that p and p' define the same topology on X.
Because Ilfn(x)xnll = lIit/i{X)Xi- i"'~-l fi{x)xill ~21Ixll/, and on account
of the fact that fn{x) is a continuous odd function of fn(x)xn (1.1.7), each
linear functional fn is continuous in the metric p', and by what has
preceded, also in the metric p. In the following we show that X is com-
plete in the metric p'.
Let {Yk} be a Cauchy sequence in X in the metric p'. From the con-
tinuity of each fn in the metric p' we then infer that Un(Yk)} is a Cauchy
sequence in iP = IR( = q, hence converges with k to some !Xn in iP. By
hypothesis one has for every e > 0 an integer p such that for any m and
n;;:m,
(1)
Taking the limit on q one obtains
n
and since X is complete in the metric p, L !XiXi is p-convergent in X,
i= 1
say to some point Y in X. Now, putting m= 1 in (1) yields in the limit
as q-H/J,
Therefore, limIlYp-YII'=O, X is complete in the metric p' and by this

p
argument the theorem is verfied.

Denote by S any compact interval in IR and let Lp(S), 0< p < 1, be
the set of all equivalence classes of measurable functions f:S~iP for
which Ilfll=Jlf(s)IPds is finite. It is known (DUNFORD and SCHWARTZ
s
[1], p. 171) that the function IIII:L/S)~IR defines a quasi-norm on
L/S) and that, endowed with this quasi-norm, Lp(S) is an F -space.
This space now provides an interesting counter-example for the basis
problem in F -spaces:
Corollary 3. 4(S), 0< p < 1, are examples of separable F -spaces
which have no basis.
Proof. Since there are no nontrivial continuous linear functionals on

Lp(S) (DAY [1]), L~(S)={O}, Lp(S) has no Schauder basis and thus, by
the preceding theorem, no basis.
The separability of Lp(S)(O < p < 1) is not easy available in literature
(SINGER [15], p. 454). We therefore shall scetch a way to get this result.
Let first f E Lp(S) be non-negative. Then f may be approximated by a
sequence Un} of simple measurable functions which converges from
below tofalmost everywhere on S (RUDIN [1], p. 15). Since If(s)- fn(s)IP
~ If(s)IP on S, by the Lebesgue dominated convergence theorem (1.4.9),

fn converges to f in the topology of Lp(S). Next, according to Lusin's
theorem (RUDIN [1J, p. 53) there is for any fn and every 8>0 a function
gEC(S) such that g(s)~fn(s) on S, and g(s)= fn(s) except on a set of
measure <8 in S. Thus Ilfn-gll ~8suP{lfn(s)IP ISES}, which shows,
based on the separability of C(S) (1.4.d), that Lp(S) is also separable.
The generalization to real or complex f's is familiar.
Theorem 4. Let X be a locally convex linear topological space. Then
every weak (extended) M arkushevich basis for X is an (extended)
Markushevich basis for x.
Proof. The conjugate space X* of X under the weak topology is the
same as that obtained under the initial topology (1.3.3). It is thus clear
that the coefficient functionals are also continuous in the initial topology
of X. Since the set of basis elements in X is total in X in the weak topo-
logy of X, it is total in X in the initial topology of X (1.3.4). These pro-
perties finally show that every weak (extended) Markushevich basis for
X is an (extended) Markushevich basis for X.
Theorem 5. Let X be a barrelled (topological linear) space. Then
every weak Schauder basis for X is a Schauder basis for x.
Proof. Let {Xi.!J be a weak Schauder basis for X. First of all, ac-
cording to Theorem 1.10 and the preceding theorem, {Xi.!J is a Mar-
kushevich basis for X. Thus {xJ is a total set in X which permits to
choose for each x in X a sequence {Yn} in X, converging to x, and such
that YnEsp{xili~n}. As we soon will show, the family {Tn} of continu-
ous linear transformations of X into itself, defined by Tnx= L
fi(X) Xi'
i:S;n
XEX, n= 1,2, ... , is equicontinuous in the initial topology of X. Hence
x=x+lim Tn(x- Yn)
n
=x+lim L fi(X- Yn)x i

n i~n
=limyn+lim( L fi(X)X i - Yn)
n n i~n
= lim
n
L fi(X)X i
i~n
and {Xi.!J is a Schauder basis for X (Theorem 1.10).

It now remains to show the equicontinuity property. Since lim Tn X= X
n
in the weak topology of X for each x in X, the sequence {Tn X} is bounded
in the initial topology for each x in X (1.1.3 and 1.3.4). Using the fact
that X is a barrel space we invoke the Barrel theorem (1.2.4) to infer that
the family {Tn} is equicontinuous in the initial topology and this finishes
the proof of the theorem.
I.I I
Corollary 6. Let X be a barrelled topological vector space and let
{Xi'!;} be a biorthogonal system Jor X such that s~p J(Xi)Ji(X) < 00,
l~n
XEX, JEX*. Then {Xi'!;} is a Schauder basis Jor sp{x;}.
Proof. Evidently, supIJ(Tnx)1 < 00, XEX, JEX*. Hence for all X,
n
{Tnx} is weakly bounded, and, by (I.3.4), bounded in X. The rest of the
proof is analogous to that of the theorem.
References for Chapter IX: ARSOVE [4,5], ARSOVE and EDWARDS [1],
DAVIS [1], DIEUDONNE [1], EDWARDS [2], KLEE [1], MARKUSHEVICH [1],
NEWNS [1] and SINGER [12, 15].
Bibliography
ABDELHAY, J.
[1] Caracterisation de l'espace de Banach de toutes les suites de nombres reels
tendant vers zero. C. R. Acad. Sci. Paris 229, 1111-1112 (1949).
AKUTOWICZ, E. J.
[1] Construction of a Schauder basis in some spaces of hoI om orphic functions
in the unit disc. Colloq. Math. 15,287-296 (1966).
ALAOGLU, L.
[1] Weak topologies of normed linear spaces. Ann. of Math. (2) 41, 252-267
(1940).
ALTMAN, M. S.
[1] On biorthogonal systems. Doklady Akad. Nauk SSSR (N. S.) 67, 413-416
(1949) (Russian). Math. Rev. 11, 114 (1950).
[2] On bases in Hilbert space. Doklady Akad. Nauk SSSR (N. S.) 69, 483-485
(1949) (Russian). Math. Rev. 11, 525 (1950).
ARSOVE, M. G.
[1] The Pincherle basis problem and a theorem of Boas. Math. Scand. 5, 271-275
(1957).
[2] Proper bases and automorphisms in the space of entire functions. Proc.
Amer. Math. Soc. 8, 264-271 (1957).
[3] Proper Pincherle bases in the space of entire functions. Quart. J. Math.
(Oxford) (2) 9, 4D-54 (1958).
[4] Proper bases and linear homeomorphisms in the space of analytic functions.
Math. Ann. 135,235-243 (1958).
[5] Similar bases and isomorphisms in Frechet spaces. Math. Annalen 135,
283-293 (1958).
[6] The Paley-Wiener theorem in metric linear spaces. Pacific J. Math. 10,
365-379 (1960).
ARSOVE, M. G., and R. E. EDWARDS
[1] Generalized bases in topological linear spaces. Studia Math. 19, 95-113
(1960).
BABENKO, K. I.
[1] On conjugate functions. Doklady Akad. Nauk SSSR (N. S.) 62, 157-160
(1948) (Russian). Math. Rev. 10, 149 (1949).
BANACH, S.
[1] Theorie des operations lineaires. Warsaw, 1932.
BANACH, S., and S. MAZUR
[1] 2ur Theorie der linearen Dimension. Studia Math. 4, 100-112 (1933).
[2] Sut la divergence des series orthogonales. Studia Math. 9, 139-155 (1940).
Bibliography 131
BARI, N. K.
[1] Biorthogonal systems and bases in Hilbert space. Moskov Gos. Univ. UC.
Zap. 148, Matematika 4,69-107 (1951) (Russian). Math. Rev. 14, 289 (1953).
BARIC, L. W.
[1] Some notes on sequences which are similar or related to a Schauder basis.
Duke Math. J. 35, 1-7 (1968).
BARIC, L. W., and W. RUCKLE
[1] Matrix transformations of Schauder bases. Studia Math. 28, 275-278
(1966/67).
BESSAGA, c.
[1] Bases in certain spaces of continuous functions. Bull. Acad. Pol. Sci. ClII,
5, 11-14 (1957).
[2] On topological classification of complete metric spaces. Fund. Math. 56,
25t-288 (1964/65).
[3] Topological equivalence of unseparable reflexive Banach spaces. Ordinal
resolution of identity and monotone bases. Bull. Acad. Polon. Sci. Ser. Math.
Astronom. Phys. 15, 397-399 (1967).
BESSAGA, c., and A. PELCZYNSKI
[1] An extension of the Krein-Milman-Rutman theorem concering bases to the
case of Bo-spaces. Bull. Acad. Pol. Sci. cm, 5, 379-383 (1957).
[2] On bases and unconditional convergence of series in Banach spaces. Studia
Math. 17, 151-164 (1958).
[3] A generalization of results of R. C. James concerning absolute bases in
Banach spaces. Studia Math. 17, 165-174 (1958).
[4] On subspaces of a space with an absolute basis. Bull. Acad. Pol on. Sci. Ser.
Sci. Math. Astr. Phys. 6, 313-315 (1958).
[5] Properties of bases in Bo-spaces. Prace Mat. 3, 123-142 (1959) (Polish).
Math. Rev. 23, 760 (1962).'
[6] Spaces of continuous functions (IV). Studia Math. 19, 53-62 (1960).
[7] Some remarks on homeomorphisms of Banach spaces. Bull. Acad. Polon.
Sci. Ser. Sci. Math. Astronom. Phys. 8, 757-761 (1960).
BOAS, R. P.
[I] General expansion theorems. Proc. Nat. Acad. Sci. USA 26,139-143 (1940).
BOCKAREv, S. V.
[1] Unconditional bases. Mat. Zametki 1, 391-398 (1967) (Russian). Math.
Rev. 35, 135 (1968).
BONDAREv, V. G.
[1] Weak reflexivity of spaces with Schauder basis. Vestnik Moskov. Univ. Ser.
I Mat. Meh. 22 no. 4, 46-49 (1967) (Russian). Math. Rev. 35, 865 (1968).
BOURBAKI, N.
[I] Espaces vectoriels topologiques, Ch. I-V, Elements de mathematique V.
Paris, 1953-55.
BUCK, R. C.
[1] Expansion theorems for analytic functions, Conference on functions of a
complex variable, Univ. of Michigan, 409-419 (1953).
CALKIN, J. W.
[1] Two-sided ideals and congruences in the ring of bounded operators in Hil-
bert space. Ann. of Math. (2) 42, 839-873 (1941).
132 Bibliography
CANTURIJA,Z.A.
[1] Some properties of biorthogonal systems in Banach space and their appli-
cation to spectral theory. Soobse. Akad. Nauk Gruzin. SSR 34, 271-276 (1964)
(Russian). Math. Rev. 30, 775 (1965).
[2] On the stability of bases of Banach spaces. Soobse. Akad. Nauk Gruzin.
SSR 36,269-272 (1964) (Russian). Math. Rev. 30, 87 (1965).
[3] Some properties of T-bases. SoohSe. Acad. Nauk Gruzin SSR 37, 271-274
(1965) (Russian). Math. Rev. 30, 956 (1965).
[4] On some properties of biorthogonal systems in Banach space. Thbilis.
Sahelmc. Univ. Srom. Mekh.-Makh. Mech. Ser. 110, 263-280 (1965) (Geor-
gian). Math. Rev. 33, 1356 (1967).
[5] On a problem of P. L. Ul'janov on the order of growth of the powers of a
polynomial basis. Mat. Zametki 1, 415-424 (1967) (Russian). Math. Rev. 34,
1500 (1967).
CEITLIN, JA. M.
[1] Unconditionality of a basis and partial order. Izv. Vyss. Ucebn. Zaved.
Matematika 51,98-104 (1966) (Russian). Math. Rev. 33, 1075 (1967).
[2] Reflexivity of spaces with a basis. Sibirsk. Math. Z. 8, 475-479 (1967)
(Russian). Math. Rev. 35, 624 (1968).
CIESIELSKI, Z.
[1] On Haar functions and on the Schauder basis of the space C [0,1]. Bull.
Acad. Pol. Sci. 7, 227-232 (1959).
[2] Some properties of Schauder bases of the space C [0, 1]. Bull. Acad. Pol on.
Sci. 7, 141-144 (1960).
[3] Properties of the orthonormal Franklin system. Studia Math. 23, 141-157
(1963).
CrVIN, P., and B. YOOD
[1] Quasi-reflexive spaces. Proc. Amer. Math. Soc. 8, 906-911 (1957).
CUTTLE, Y.
[1] On quasi-reflexive Banach spaces. Proc. Amer. Math. Soc. 12, 936-940
(1961).
DADIe, I.
[1] Variations of finite sets of independent vectors in Hilbert spaces. Glasnik
Mat. Ser. III 1 (21) 51-55 (1966).
DAVIS, W. J.
[1] Dual generalized bases in linear topological spaces. Proc. Amer. Math.
Soc. 17, 1057-1063 (1966).
[2] M-similarity and isomorphisms in Bo-spaces. Proc. Amer. Math. Soc. 19,
332-335 (1968).
[3] Schauder decompositions in Banach spaces. Bull. Amer. Math. Soc. 74,
1083-1085 (1968).
DAVIS, W. J., and D. W. DEAN
[1] The direct sum of Banach spaces with respect to a basis. Studia Math. 28,
209-219 (1966/67).
DAVIS, W. J., D. W. DEAN, and I. SINGER
[1] Complemented subspaces and A systems in Banach spaces. Israel 1. Math.
6,303-309 (1968).
DAY,M.M.
[1] The spaces Uwith O<p< 1, Bull. Amer. Math. Soc. 46,816-823 (1940).
Bibliography 133
[2] Normed linear spaces. Berlin-Gottingen-Heidelberg, 1962.

[3] On the basis problem in normed spaces. Proc. Amer. Math. Soc. 13,
655-658 (1962).
DEAN, D. W. (see also DAVIS, W. J.)
[1] Schauder decompositions in (m). Proc. Arner. Math. Soc. 18, 619-6i3
(1967).
DIEUDONNE, J.
[1] On biorthogonal systems. Michigan Math. J. 2, 7-20 (1953).
[2] Foundations of modern analysis. New York, 1960.
DIXMIER, J.
[1] Sur les bases orthonormales dans les espaces prehilbertiens. Acta Sci. Math.
Szeged. 15, 29-30 (1953).
DUBINSKY, E. L., and J. R. RETHERFORD
[1] Schauder bases and Kothe sequence spaces. Bull. Acad. Polon. Sci. Ser.
Math. Astronom. Phys. 14,497-501 (1966).
[2] Schauder bases in. compatible topologies. Studia Math. 28, 221-226
(1966/67).
DUFFIN, R. J., and J. J. EACHUS
[1] Some notes on an expansion theorem of Paley and Wiener. Bull. Amer.
Math. Soc. 48, 850-855 (1942).
DUNFORD, N.
[1] Uniformity in linear spaces. Trans. Amer. Math. Soc. 44,305-316 (1938).
DUNFORD, N., and J. T. SCHWARTZ
[I] Linear operators, New York, I (1958), II (1963).
DVORETZKY, A., and C. A. ROGERS
[I] Absolute and unconditional convergence in normed linear spaces. Proc.
Nat. Acad. Sci. USA 36, 192-197 (1950).
DYNIN, A. S., and B. S. MITJAGIN
[I] Criterion for nuclearity in terms of approximative dimension. Bull. Acad.
Polon. Sci. Ser. sci. math., astr. et phys. 8,535-540 (1960).
EACHUS, J. J. (see DUFFIN, R. J.)
EBERLEIN, W. F.
[I] Weak compactness in Banach spaces I. Proc. Nat. Acad. Sci. USA 33,
51-53 (1947).
EDWARDS, R. E. (see also ARSOVE, M. G.)
[I] Integral bases in inductive limit spaces. PacificJ. Math. 10, 797-812 (1960).
[2] Functional analysis, theory and applications. New York, 1965.
ELLIS, H. W., and I. HALPERIN
[I] Haar functions and the basis problem for Banach spaces. J. London Math.
Soc. 31, 28-39 (1956).
ELLIS, H. W., and D. G. KUEHNER
[I] On Schauder bases for spaces of continuous functions. Canad. Math. Bull.
3, 173-184 (1960).
FAGE, M. K.
[I] Idempotent operators and their rectification. Doklady Akad. Nauk SSSR
(N. S.) 73, 895-897 (1950) (Russian). Math. Rev. 12, 186 (1951).
[2] The rectification of bases in Hilbert space. Doklady Akad. Nauk SSSR
10 Springer Tracts, VoL 18 - Marti

134 Bibliography
FLEMING, R. J., R. D. MCWILLIAMS, and J. R. RETHERFORD

[1] On w*-sequential convergence, type p* bases and reflexivity. Studia Math.
25, 325-332 (1965). .
FOGUEL, S. R.
[1] Biorthogonal systems in Banach spaces. Pacific J. Math. 7, 1065-1072
(1957).
[2] On bases in C ([0, 1]) and L1 ([0,1]). Rev. Roumaine Math. Pures Appl. 10,
931-960 (1965).
FOIAS, C., and 1. SINGER
[1] Some remarks on strongly linearly independent sequences and bases in
Banach spaces. Rev. math. pures appl. 6, 589-594 (1961).
FRANKLIN, P.
[1] A set of continuous orthogonal functions. Math. Ann. 100, 522-529 (1928).
FRINK, O.
[1] Series expansions in linear vector spaces. Amer. J. Math. 63, 87-100 (1941).
FULLERTON, R. E.
[1] Geometric structure of absolute basis systems in a linear topological space.
Pacific J. Math. 12, 137-147 (1962).
GAPOSKIN, V. F.
[1] On unconditional bases in the spaces Lp (p > 1). Uspehi Mat. Nauk (N. S.)
13,179-184 (1958) (Russian). Math. Rev. 20,1094 (1959).
[2] On certain properties of unconditional bases in the spaces Lp (p> 1). U spehi
Mat. Na~ (N. S.) 14, 143-148 (1959) (Russian). Math. Rev. 22, 988 (1961).
[3] Unconditional bases in Orlicz spaces. Uspehi Mat. Nauk 22, no. 2 (134),
113-114 (1967) (Russian). Math. Rev. 34, 1500 (1967).
[4] The existence of unconditional bases in Orlicz spaces. Funkcional. Anal. i
Prilozen. 1,26-32 (1967) (Russian). Math. Rev. 36, 1100 (1968).
GARLING, D. J. H.
[1] Symmetric bases oflocallyconvex spaces. Studia Math. 30,163-181 (1968).
GELBAUM, B. R.
[1] Expansions in Banach spaces. Duke Math. J. 17, 187-196 (1950).
[2] A nonabsolute basis for Hilbert space. Proc. Amer. Math. Soc. 2, 720-721
(1951).
[3] Notes on Banach spaces and bases. An. Acad. Brasil. Ciencias 30, 29-36
(1958).
GELBAUM, B. R., and J. GIL DE LAMADRID
[1] Bases of tensor products of Banach spaces. Pacific J. Math. 11, 1281-1286
(1961 ).
GELFAND, 1. M.
[1] Normierte Ringe. Mat. Sbornik N. S. 9 (51), 3-24 (1941).
[2] Remark on the work of N. K. BARI "Biorthogonal systems and bases in
Hilbert space". Moskov. Gos. Univ. Ucenye Zapinski 148, 224-225 (1951)
(Russian). Math. Rev. 14,289 (1953).
GIL DE LAMADRID, J. (see also GELBAUM, B. R.)
[1] On finite dimensional approximation of mappings in Banach spaces. Proc.
Amer. Math. Soc. 13, 163-168 (1962).
GORDON, 1. A.
[1] Certain sufficient criteria for stability of complete orthonormal bases in
L [0, 1] with respect to the operation of averaging. Functional Anal. Theory of
Bibliography 135
Functions, No. I, 13-21 Izdat. Kazan. Univ., Kazan 1963 (Russian). Math.
Rev. 36, 1101 (1968).
GRINBLYUM, M. M.
[1] Certains theoremes sur la base dans un espace du type (B). Doklady Akad.
Nauk SSSR (N. S.) 31, 428-432 (1941) (Russian). Math. Rev. 3, 49 (1942).
[2] Biorthogonal systems in Banach space. Doklady Akad. Nauk SSSR (N. S.)
47, 79-82 (1945) (Russian). Math. Rev. 7, 125 (1946).
[3] Sur la theorie des systemes biorthogonaux. Doklady Akad. Nauk SSSR
(N. S.) 55, 287-290 (1947).
[4] On a property of a basis. Doklady Akad. Nauk SSSR (N. S.) 59, 9-11
(1948) (Russian). Math. Rev. 10, 307 (1949).
[5] On the representation of a space of type B in the form of a direct sum of
subspaces. Doklady Akad. Nauk SSSR (N. S.) 70, 749-752 (1950) (Russian).
Math. Rev. 11, 525 (1950).
GRUNBAUM, B.
[1] Some applications of expansion constants. Pacific 1. Math. 10, 194-201
(1960).
GURARII, V. I.
[1] On inclinations of spaces and conditional bases in Banach space. Doklady
Akad. Nauk SSSR 145, 504-506 (1962) (Russian). Math. Rev. 27, 553 (1964).
[2] Bases in spaces of continuous functions. Doklady Akad. Nauk SSSR 148,
483-495 (1963) (Russian). Math. Rev. 26, 556 (1963).
[3] Some geometric characteristics of subspaces and bases in Banach spaces.
Colloq. Math. 13, 59-63 (1964) (Russian). Math. Rev. 31, 467 (1966).
[4] The index of sequences in C and the existence of infinite dimensional
separable Banach spaces having no orthogonal basis. Rev. Roumaine Math.
Pures Appl. 10, 967-971 (1965) (Russian). Math. Rev. 34, 591 (1967).
[5] Bases for sets in Banach spaces. Rev. Roumaine Math. Pures Appl. 10,
1235-1240 (1965).
[6] Bases in spaces of continuous functions on compacta and some geometric
questions. Izv. Akad. Nauk SSSR Ser. Mat. 30, 289-306 (1966) (Russian).
Math. Rev. 34, 591 (1967).
[7] Subspaces and bases in spaces of continuous functions. Doklady Akad.
Nauk SSSR 167, 971-973 (1966) (Russian). Math. Rev. 33, 1358 (1967).
GURARII, V. I., and M. 1. KADEC
[1] Minimal systems and quasi-complements in Banach space. Doklady Akad.
Nauk SSSR 145,256-258 (1962) (Russian). Math. Rev. 26, 1276 (1963).
GUREVIC, L. A.
[1] On unconditional bases. Uspehi Mat. Nauk (N. S.) 8, no. 5 (57),153-156
(1953) (Russian). Math. Rev. 15, 631 (1954).
[2] A basis in the space of abstract functions. Doklady Akad. Nauk SSSR 136,
12-15 (1961) (Russian). Math. Rev. 24, 411 (1962).
[3] Conic tests for bases of absolute convergence. Problems of Math. Phys.
and Theory of Functions, II. p. 12-21. Naukova Dumka, Kiev, 1964 (Russian).
Math. Rev. 33, 1358 (1967).
HAAR, A.
[I] Zur Theorie der orthogonal en Funktionssysteme, I, Math. Ann. 69,
331-337 (1910), II, ibid. 71,38-53 (1911).
HALMos, P. R.
[I] Finite dimensional vector spaces. Princeton, 1958.
[2] A Hilbert space problem book. Princeton, 1967.
10*
136 Bibliography
HAUSDORFF, F.
[1] Mengenlehre. New York, 1944.
HILDEBRANDT, T. H.
[1] On unconditional convergence in normed vector spaces. Bull. Amer. Math.
Soc. 46, 959-962 (1940).
HILDING, S. H.
[1] Note on completeness theorems of Paley-Wiener type. Ann. of Math. (2)
49,953-955 (1948).
HILLE, E., and R. S. PHILLIPS
[1] Functional analysis and semi-groups. Amer. Math. Soc. Colloquium Publ.
31 (rev. ed.) (1957).
ISTRATESCU, V.
[1] Uber die Banachrliume mit zlihlbarer Basis I, Rev. Math. Pures Appl.
(Bucarest) 7, 481-482 (1962), II, Revue Roumaine Math. Pures Appl. 9,
431-433 (1964).
LYER, V. G.
[l] On the space of integral functions (III). Proc. Amer. Math. Soc. 3, 874-883
(1952).
JAMES, R. C.
[1] Orthogonality in normed linear spaces. Duke Math. J. 12,291-302 (1945).
[2] Orthogonality and linear functionals in normed linear spaces. Trans. Amer.
Math. Soc. 61, 265-292 (1947).
[3] Inner products in normed linear spaces. Bull. Amer. Math. Soc. 53, 559-566
(1947).
[4] Bases and reflexivity of Banach spaces. Ann. of Math. 52,518-527 (1950).
[5] A non-reflexive Banach space isometric with its second conjugate space.
Proc. Nat. Acad. Sci. USA 37,174-177 (1951).
[6] Projections in the space (m). Proc. Amer. Math. Soc. 6, 899-902 (1955).
[7] Separable conjugate spaces. Pacific J. Math. 10, 563-571 (1960).
[8] Characterizations of reflexivity. Studia Math. 23, 205-216 (1964).
[9] Weak compactness and reflexivity. Israel J. Math. 2, 10 1-119 (1964).
[10] Weakly compact sets. Trans. Amer. Math. Soc. 113, 129-140 (1964).
JAMES, R. C., and J. R. RETHERFORD
[I] Unconditional bases and best approximation in Banach spaces. Bull. Amer.
Math. Soc. 75, 108-112 (1969).
JONES, O. T., and J. R. RETHERFORD
[1] On similar bases in barrelled spaces. Proc. Arner. Math. Soc. 18, 677-680
(1967).
JULIA, G.
[1] Exemples des structures des systemes duaux de l'espace hilbertien. C. R.
Acad. Sci. (Paris) 216, 465-468 (1943).
KACZMARZ, S., and H. STEINHAUS
[1] Theorie der Orthogonalreihen. Warsaw, 1935.
KADEC, M. 1. (see also GURARII, V. 1.)
[I] On conditionally convergent series in the space Lp. Uspehi Mat. Nauk
[2] Linear dimension of the spaces Lp and lp. Uspehi Mat. Nauk (N. S.) 13,
95-98 (1958) (Russian). Math. Rev. 21, 59 (1960).
Bibliography 137
[3] Bases and their spaces of coefficients. Dopovidi Akad. Nauk Ukrain. RSR
1, 1139-1140 (1964).
[4] Topological equivalence of all separable Banach spaces. Soviet Math.
Doklady 7, 319-322 (1966).
[5] Nonlinear operator-bases in a Banach space. Teor. Funkcll Funkcional.
Anal. i Prilozen. Vyp. 2,128-130 (1966) (Russian). Math. Rev. 34,1188 (1967).
KADEc, M. I., and A. PELCZYNSKI
[1] Bases, lacunary sequences and complemented subspaces in the spaces Lp.
Studia Math. 21, 161-176 (1962).
[2] Basic sequences, biorthogonal systems and nonning sets in Banach and
Frechet spaces. Studia Math. 25, 297-323 (1965).
KAKUTANI, S.
[1] Some characterizations ofEuklidean spaces. Jap. J. Math. 16,93 -97 (1939).
KARLIN, S.
[1] Unconditional convergence in Banach spaces. Bull. Amer. Math. Soc. 54,
148-152 (1948).
[2] Bases in Banach spaces. Duke Math. J. 15,971-985 (1948).
KELLEY, J. L., and I. NAMIOKA
[1] Linear topological spaces. Princeton, 1963.
KLEE, V.
[I] On the borelian and projective types of linear subspaces. Math. Scand. 6,
189-199 (1958).
KOTHE, G.
[I] Probleme der linearen Algebra in topo1ogischen Vektorraumen. Proc.
internat. sympos. on linear spaces 1960. Jerusalem, Academic Press 1961.
[2] Topologische lineare Raume. Berlin-Heidelberg-New York, 1966.
KOSTYUCENKO, A., and A. SKOHOROD
[I] Ona theoremofM. K. Bari. UspehiMat. Nauk(N. S.)8, no. 5 (57),165-166
(1953) (Russian). Math. Rev. 15, 632 (1954).
KozLOv, V. YA.
[I] On bases in the space L [0, I]. Mat. Sbornik N. S. 26 (68), 85-102 (1950)
(Russian). Math. Rev. 11, 602 (1950).
[2] On a generalization of the concept of a basis. Dok1ady Akad. Nauk SSSR
(N. S.) 73,643-646 (1950) (Russian). Math. Rev. 12, 110 (1951).
KREIN, M., D. MILMAN, and M. RUTMAN
[I] A note on a basis in Banach space. Comm. lnst. Sci. Math. Mec. Univ.
Kharkoff (Zapinski lnst. Mat. Mech.) (4) 16, 106-110 (1940) (Russian).
Math. Rev. 3, 49 (1942).
[2] On a property of a basis in Banach space. Kark. Zap. Matern. Obsh. (4) 16,
182 (1940) (Russian).
KUEHNER, D. G. (see ELLIS, H. W.)
LINDENSTRAUSS, J.
[I] Extension of compact operators. Mem. Amer. Math. Soc. 48 (1964).
[2] On a subspace of the space 1. Bull. Acad. Po1on. Sci., Ser. Sci. Math.
Astronom. Phys. 12, 539-542 (1964).
LINDENSTRAUSS, J., and M. ZIPPIN
[I] Banach spaces with a unique unconditional basis. J. Funct. Anal. 3,115-125
(1969).
138 Bibliography
LruSTERNIK, L. A., and W. I. SOBOLEW

[1] Elemente der Funktionalanalysis. Berlin, 1955.
LOZANOVSKII, G. JA.
[I] On Banach lattices and bases. Functional. Anal. i Prilozen. 1, no. 3, 92
(1967) (Russian). Math. Rev. 36, 633 (1968).
MACPHAIL, M. S.
[I] Absolute and unconditional convergence. Bull. Amer. Math. Soc. 53,
121-123 (1947).
MADDAUS, I.
[I] On completely continuous linear transformations. Bull. Amer. Math. Soc.
44,279-282 (1938).
MARCINKIEWICZ, J.
[I] Quelques theoremes sur les series orthogonales. Ann. Soc. Polon. Math.
16,84-96 (1937).
MARKUS, A. S.
[I] A basis of root vectors of a dissipative operator. Soviet Math. Doklady 1,
599-602 (1960).
MARKUSHEVICH, A. I.
[I] Sur les bases (au sens large) dans les espaces lineaires. Doklady Akad.
Nauk SSSR (N. S.) 41, 227-229 (1943).
[2] Sur la meilleure approximation. Doklady Akad. Nauk SSSR (N. S.) 44,
262-264 (1944).
[3] On bases in the space of analytic functions. Mat. Sbornik 17 (59), 211-252
(1945) (Russian). Math. Rev. 7, 425 (1946).
MARTI, J. T.
[I] On integro-differential equations in Banach spaces. Pacific J. Math. 20,
99-108 (1967).
[2] Extended bases for Banach spaces. To appear.
[3] On bases, compactness and weak convergence in the Banach space Ap.
To appear.
MAZUR, S. (see BANACH, S.)
MAZUR, S., and W. ORLICZ
[I] Sur les espaces metriques lineaires I, II. Studia Math. I, 10, 184-208
(1948), II, 13,137-179 (1953).
McARTHUR, C. W.
[1] On relationships among certain spaces of sequences in an arbitrary Banach
space. Canad. J. Math. 8, 192-197 (1956).
[2] The weak basis theorem. Colloq. Math. 17,71-76 (1967).
McARTHUR, C. W., and J. R. RETHERFORD
[1] Unifonn and equicontinuous Schauder bases of subspaces. Canad. J. Math.
17, 207-212 (1965).
[2] Some remarks on bases in linear topological spaces. Math. Ann. 164, 38-41
(1966).
McKINNEY, R. L.
[1] Positive bases for linear spaces. Trans. Amer. Math. Soc. 103, 131-148
(1962).
MCWILLIAMS (see FLEMING, R. J.)
Bibliography 139
MICHAEL, E., and A. PELCZYNSKI

[1] Separable Banach spaces which admit In approximations. Israel Math. J. 4,
189-198 (1966).
MILMAN, V. D. (see also KREIN, M.)
[I] Certain properties of unconditional bases. Soviet Math. Doklady 6,
656-659 (1965).
[2] Some properties of sequences of elements of Banach spaces. First Republ.
Math. Conf. of Young Researchers, Part II, Akad. Nauk Ukrain. SSR Inst.
Math., Kiev, 480-489 (1965) (Russian). Math. Rev. 34,885 (1967).
MITJAGIN, B. S. (see also DYNIN, A. S.)
[I] Approximative dimension and bases in nuclear spaces. Uspehi Mat. Nauk
16,63-132 (1961) (Russian). Math. Rev. 27, 554 (1964).
MURRAY, F. J.
[1] On complementary manifolds and projections in spaces Lp and Ip. Trans.
Amer. Math. Soc. 41, 138-152 (1937).
[2] The analysis of linear transformations. Bull. Amer. Math. Soc. 48, 76-93
(1942).
Sz. NAGY, B.
[1] Expansion theorems of Paley-Wiener type. Duke Math. J. 14, 975-978
(1947).
NAMIOKA, I. (see KELLEY, J. L.)
NEWNS, W. F.
[1] On the representation of analytic functions by infinite series. Phil. Trans.
Royal Soc. London (A) 245,429-468 (1953).
NGUYEN THANH VAN
[1] Bases de Schauder dans certains espaces vectoriels topologoques. Ann.
Fac. Sci. Univ. Toulouse (4) 28, 139-147 (1965).
[2] Sur les bases de Schauder de l'espace des fonctions holomorphes dans une
domaine simplement connexe. C. R. Acad. Sci. Paris Ser. A-B 264, A 1053
bis 1055 (1967).
NGUENVAN KHUE
[1] Test for unconditional convergence bases. Izv. Vyss. Ucebn. Zaved. Mate-
matika 69,68-74 (1968) (Russian). Math. Rev. 36, 1099 (1968).
NIKOL'SKII, V. N.
[I] The best approximation and a basis in a Frechet space. Doklady Akad.
Nauk SSSR (N. S.) 59,639-642 (1948) (Russian). Math. Rev. 10, 128 (1949).
[2] Some questions of best approximation in a function space. Uc. Zap.
Kalininsk. Pedagog. Inst. 16, 119-160 (1954) (Russian). Math. Rev. 17, 175
(1956).
OLUBUMMO, A.
[I] Operators of finite rank in a reflexive Banach space. Pacific J. Math. 12,
1023-1027 (1962).
ORLICZ, W. (see also MAZUR, S.)
[I] Beitdige zur Theorie der Orthogonalreihenentwicklungen II. Studia Math.
1,241-255 (1929).
[2] Uber unbedingte Konvergenz in Funktionenraumen I. Studia Math. 4,
33-37 (1933).
[3] Some remarks on the absolute convergence of biorthogonal expansions in
the space C. Ann. Univ. Sci. Budapest Eotvos Sect. Math. 3-4, 217-222
(1960/61).
140 Bibliography
PALEY, R. E. A. C., and N. WIENER

[I] Fourier transforms in the complex domain. Amer. Math. Soc. Colloquium
Publ. no. 19, New York, 1934.
PALEY, R. E. A. C., and A. ZYGMUND
[I] On some series of functions (I). Proc. Cambro Phil. Soc. 26,337-357 (1930).
PECK, N. T.
[1] On non locally convex spaces II. Math. Ann. 178,209-218 (1968).
PELCZYNSKI, A. (see also BESSEGA, C., KADEC, M. I., and MICHAEL, E.)
[1] On B-spaces containing subspaces isomorphic to the space co. Bull. Acad.
Pol. Sci. Cl. III,S, 797-798 (1957).
[2] A connection between weakly unconditional convergence and weak com-
pleteness of Banach spaces. Bull. Acad. Polon. Sci. ser. Math. Astr. Phys. 6,
251-253 (1958).
[3] Projections in certain Banach spaces. Studia Math. 19, 209-228 (1960).
[4] A note to the paper of I. Singer "Basic sequences and reflexivity of Banach
spaces". Studia Math. 21, 371-374 (1962).
[5] Some problems on bases in Banach and Frechet spaces. Israel J. Math. 2,
132-138 (1964).
[6] On simultaneous extension of continuous functions. Studia Math. 24,
285-304 (1964).
[7] A proof of the Eberlein-Smulian theorem by an application of basic sequen-
ces. Bull. Acad. Polon. Sci., Ser. Sci. Math. astr. phys. 12,543-548 (1964).
PELCZYNSKI, A., and I. SINGER
[1] On non-equivalent bases and conditional bases in Banach spaces. Studia
Math. 25, 5-25 (1964).
PELCZYNSKI, A., and W. SZLENK
[1] An example of a non-shrinking basis. Rev. Roumaine Math. Pures Appl.
10,961-966 (1966).
PETTIS, B. J.
[I] On integration in vector spaces. Trans. Amer. Math. Soc. 44, 227-304
(1938).
PHILLIPS, R. S. (see also HILLE E.)
[I] On linear transformations. Trans. Amer. Math. Soc. 48, 516-541 (1940).
PIETSCH, A.
[I] Nukleare lokalkonvexe Rliume. Berlin, 1965.
[2] F-Rliume mit absoluter Basis. Studia Math. 26, 233-238 (1966).
POLLARD, H.
[I] Completeness theorems of Paley-Wiener type. Ann. of Math 45,738-739
(1944).
POLTAVSKII, L. N.
[I] Orthogonality in Lp-spaces. Vestnik Moskov Univ. Ser. I Mat. Meh. 22,
no. 1,47-50 (1967) (Russian). Math. Rev. 35, 139 (1968).
PRIGORSKrI, V. A.
[I] On some classes of bases in ,Hilbert space. Uspehi Mat. Nauk 20, no. 5 (125),
231-236 (1965) (Russian). Math. Rev. 34, 317 (1967).
[2] On quadratically stable bases of subspaces. Mat. Issled. 2, 164-168 (1967)
(Russian). Math. Rev. 36, 633 (1968).
PTAK, V.
[I] Biorthogonal systems and reflexivity of Banach spaces. Czechosl. Math. J.
9, 319-326 (1959).
Bibliography 141
RADEMACHER, H.
[I] Einige Siitze tiber Reihen von allgemeinen Orthogonalfunktionen. Math.
Ann. 87, 112-138 (1922).
REAY, J. R.
[I] Unique minimal representations with positive bases. Amer. Math. Monthly
73,253-261 (1966).
RETHERFORD, J. R. (see also DUBINSKY, E. L., JAMES, R. C., JONES, O. T., FLEMING,
R. J., and McARTHUR, C. W.)
[1] Basic sequences and the Paley-Wiener criterion. Pacific. J. Math. 14,
1019-1027 (1964).
[2] w*-bases and bw*-bases in Banach spaces. Studia Math. 25, 65-71 (1964).
[3] Bases, basic sequences and reflexivity of linear topological spaces. Math.
Ann. 164,280-285 (1966).
[4] Shrinking bases in Banach spaces. Amer. Math. Monthly 73, 841-846
(1966).
[5] Some remarks on Schauder bases of subspaces. Rev. Roumaine Math.
Pures Appl. 11, 787-792 (1966).
[6] On Cebysev subspaces and unconditional bases in Banach spaces. Bull.
Amer. Math. Soc. 73, 238-241 (1967).
[7] Some characterizations of Co and /1. Canad. Math. Bull. 10, 39-52 (1967).
RICKART, C. E.
[I] General theory of Banach algebras. Princeton, 1960.
ROGERS, C. A. (see DVORETZKY, A.)
RUCKLE, W. H. (see also BARIC, L. W.)
[I] Schauder decompositions and bases. Dissertation, Florida State Univ.,
1963.
[2] The infinite sum of closed subspaces of an F-space. Duke Math. J. 31,
543-554 (1964).
[3] Infinite matrices which preserve Schauder bases. Duke Math. J. 33,547-550
(1966).
[4] On the construction of sequence spaces that have Schauder bases. Canad.
J. Math. 18, 1281-1293 (1966).
[5] On the characterization of sequence spaces associated with Schauder bases.
Studia Math. 28, 279-288 (1966/67).
[6] Lattices of sequence spaces. Duke Math. J. 35, 491-503 (1968).
RUDIN, W.
[I] Real and complex analysis. New York, 1966.
Russo, J. P.
[I] Monotone and e-Schauder bases of subspaces. Canad. J. Math. 20, 233-241
1968).
RUTMAN, M. (see KREIN, M.)
RUTOWITZ, D.
[I] Absolute and unconditional convergence in normed linear spaces. Proc.
Cambro Phil. Soc. 58, 575-579 (1962).
SAIDUKOV, K. M.
[I] On the Lebesgue constants of bases in the space of continuous functions.
Functional Anal. Theory of Functions no. I, Izdat. Kazan Univ. Kazan,
122-133 (1963) (Russian). Math. Rev. 34, 1190 (1967).
142 Bibliography
[2] On the order of growth of the degrees of a polynomial basis. Functional

Anal. Theory of Functions no. I, Izdat. Kazan Univ. Kazan, 134-138 (1963)
(Russian). Math. Rev. 34, 1501 (1967).
[3] A criterion for a basis in the space of continuous functions. Izv. Vyss.
Ucebn. Zaved Matematica 52, 178-182 (1966) (Russian). Math. Rev. 33, 302
(1967).
SANDERS, B. L.
[I] On a generalization of the Schauder basis concept. Dissertation the Florida
State University (1962).
[2] Decompositions and reflexivity in Banach spaces, Proc. Amer. Math. Soc.
16,204-208 (1965).
[3] On the existence of (Schauder) decompositions in Banach spaces, Proc.
Amer. Math. Soc. 16, 987-990 (1965).
SCHAEFER, H. H.
[1] Halbgeordnete lokalkonvexe Vektorraume. Math. Ann. 135, 115-141
(1958).
[2] Topological vector spaces. New York, 1966.
SCHAFFER, J. J.
[I] Another characterization of Hilbert spaces. Studia Math. 25, 271-276
(1965).
SCHAFKE, F. W.
[I] Das Kriterium von Paley und Wiener 1m Banachschen Raum. Math.
Nachr. 3, 59-61 (1949).
SCHAUDER, J.
[1] Zur Theorie stetiger Abbildungen In Funktionalraumen, Math. Z. 26,
47-65 (1927).
[2] Eine Eigenschaft des Haarschen Orthogonalsystems, Math. Z. 28, 317- 320
(1928).
SCHMIDT. E.
[1] Entwicklung willkiirlicher Funktionen nach Systemen vorgeschriebener.
Math. Ann. 63, 433-476 (1907).
SCHWARTZ, J. T. (see DUNFORD, N.)
SEMADENI, Z.
[1] Product Schauder bases and approximation with nodes in spaces of con-
tinuous functions. Bull. Acad. Polon. Sci. Ser. Sci. Math. Astronom. Phys. 11,
387-391 (1963).
SINGER, 1. (see also DAVIS, W. J., FOIAS, C., and PELCZYNSKI, A.)
[1] Elementary proof of a theorem of S. R. Foguel on biorthogonal systems in
Banach spaces. Rev. Math. pures app!. 3, 305-307 (1958).
[2] Sur les espaces de Banach it base absolue, canoniquement equivalents it
un dual d'espace de Banach, C. R. Acad. Sci. Paris 251, 620-621 (1960).
[3] Weak* bases in conjugate Banach spaces. Studia Math. 21, 75-81 (1961).
[4] On Banach spaces with a symmetric basis. Rev. Math. Pures et Appl.
(Bucarest) 6, 159-166 (1961) (Russian). Math. Rev. 26, 797 (1963).
[5] Basic sequences and reflexivity of Banach spaces. Studia Math. 21,351-369
(1962).
[6] Some characterizations of symmetric bases, Bull. Acad. Pol. Sci., Serie
math., astr. et phys. 10, 185-192 (1962).
Bibliography 143
[7] On Cesaro bases in Banach spaces. Rev. math. pures app!. (Bucarest) 7,
135-142 (1962).
[8] On a theorem of I. M. Gelfand. Uspehi Mat. Nauk 17, 169-176 (1962)
(Russian). Math. Rev. 24, 656 (1962).
[9] On Banach spaces reflexive with respect to a linear subspace of their con-
jugate space II, III. Math. Ann. 145, 64-76 (1962), Rev. math. pures app!. 8,
139-150 (1963).
[10] Weak* bases in conjugate Banach spaces II. Rev. Math. Pures App!.
(Bucarest) 8, 575-584 (1963).
[11] On bases in quasi-reflexive Banach spaces. Rev. Math. pures App!.
(Bucarest) 8,309-311 (1963).
[12] Bases in Banach spaces I, II and III. Studie si cercetari matematice 14,
533-585 (1963); 15,157-208 (1964); 15, 675-725 (1964) (Rumanian).
[13] A proof of the Dvoretzky-Rogers theorem, Israel 1. Math. 2, 249-250
(1964).
[14] Bases and quasi-reflexivity of Banach spaces. Math. Ann. 153, 199-209
(1964).
[15] On the basis problem in topological linear spaces. Rev. Roumaine Math.
Pures App!. 10,453-457 (1965).
[16] Bases in Banach spaces. Grund!. d. math. Wiss. 154, Berlin-Heidelberg-
New York, in preparation.
SIRETCHI, G.
[l] On certain spaces with a base. An. Univ. Bucaresti Ser. Stiint Natur. Mat.-
Mech. 13,141-144 (1964).
SKOHOROD, A. (see KOSTYUCENKO, A.)
SOBCZYK, A.
[I] Projection of the space (m) on its subspace (co). Bull. Amer. Math. Soc. 47,
938-947 (1941).
SOBOLEW, W. I. (see LJUSTERNIK, L. A.)
SOLOMYAK, M. Z.
[I] On orthogonal basis in Banach space. Vestnik Leningrad. Univ. 12, 27-36
(1957) (Russian). Math. Rev. 19,45 (1958).
STEINHAUS, H. (see KACZMARZ, S.)
SZILENK, W. (see also PELCZYNSKI, A.)
[I] Une remarque sur I'orthogonalisation des bases de Schauder dans I'espace
C. Colloq. Math. 15, 297-301 (1966).
TAYLOR, A. E.
[I] The extension of linear functionals. Duke Math. 1. 5, 538-547 (1939).
[2] The weak topologies of Banach spaces. Proc. Nat. Acad. Sci. USA 25,
438-440 (1939).
[3] A geometric theorem and its application to biorthogonal systems. Bul!.
Amer. Math. Soc. 53, 614-616 (1947).
[4] Introduction to functional analysis. New York, 1958.
TOEPLITz, O.
[1] Uber allgemeine lineare Mittelbildungen. Prace Math. Fiz. 22, 113-119
(1911).
TROJANSKII, S.
[I] The topological equivalence of the spaces coG':) and l(~). Bull. Acad. Polon.
Sci. Ser. Sci. Math. Astronom. Phys. 15, 389-396 (1967) (Russian). Math.
Rev. 36, 1099 (1968).
144 Bibliography
TSENG, Y. Y.
[1]. On generalized biorthogonal expansions in metric and unitary spaces.
Proc. Nat. Acad. Sci. USA 28,35-43 (1942).
VAlIER, F. S.
[1] On the basis in the space of continuous functions defined on a compact set.
Doklady Akad. Nauk SSSR 101,589-592 (1955) (Russian). Math. Rev. 16,
1031 (1955).
VAlc, B. E.
[1] On some properties of unconditioal bases. Uspehi Mat. Nauk 17 no. 6
(108), 135-142 (1962) (Russian). Math. Rev. 26, 797 (1963).
[2] Some stability properties of bases. Soviet Math. Doklady 5, 1141-1144
(1964).
[3] Characteristic properties of unconditional bases and theorems of stability.
Izv. Vyss. Ucebn Zaved Matematika 47, 24-36 (1965) (Russian). Math. Rev.
34, 1191 (1967).
VANICEK, J.
[1] Biorthogonal systems in a Banach space (Engl. summary). Acta Fac. Rev.
natur. Univ. Comenian. Math. 6, 319-325 (1961).
[2] Biorthogonal systems and limit methods. Casopis Rest. Mat. 87, 17-21
(1962) (Czech.). Math. Rev. 24, 533 (1962).
[3] Approximating sequences in Banach spaces. Casopis Rest, Mat. 87, 52-62
(1962) (Czech.). Math. Rev. 24, 533 (1962).
VANICEK, J., and H. VANICKOVA
[1] On the space of holomorphic functions. Casopis Rest. Mat. 86, 433-438
(1961) (Czech.). Math. Rev. 24, 616 (1962).
VANICKOVA, H. (see VANICEK, J.)
VINIKUROV, V. G.
[1] On biorthogonal systems spanning a given subspace. Doklady Akad. Nauk
SSSR (N. S.) 85,685-687 (1952) (Russian). Math. Rev. 14, 183 (1953).
VIzrrEI, V. N.
[1] Stability of bases consisting of subspaces of a Banach space. Studies in
Algebra and Math. Anal. Izdat "Karta Malovenjaske", Kishinev, 34-44
(1965) (Russian). Math. Rev. 34, 884 (1967).
VOLENEC, V.
[1] Variations of orthogonal basic sets in Euclidean space. Glasnik Mat. Ser.
III 1 (21) 51-55 (1966).
WALSH, J. L.
[1] Interpolation and approximation by rational functions in the complex
domain. Amer. Math. Soc. Colloquium Publ. 20, 1965.
WEILL, L. J.
[1] Stability of bases in complete barrelled spaces. Proc. Amer. Math. Soc. 18,
1045-1050 (1967).
WIENER, N. (see PALEY, R. E. A. C.)
WILANSKY, A.
[1] The basis in Banach space. Duke Math. J. 18, 795-798 (1951).
[2] Functional analysis. New York, 1964.
WOJTYNSKI, W.
[1] On bases in certain countably-Hilbert spaces. Bull. Acad. Polon. Sci. Ser.
Sci. Math. Astronom. Phys. 14,681-684 (1966).
Bibliography 145
YAMAZAKI, S.
[I] Normed rings and unconditional bases in Banach spaces. Sci. Pap. ColI.
Gen. Educ. Univ. Tokyo 14, 1-10 (1964).
[2] Normed rings and bases in Banach spaces. Sci. Pap. ColI. Gen. Educ. Univ.
Tokyo 15, 1-13 (1965).
[3] Remerk to "Normed rings and bases in Banach spaces". Sci. Pap. Coil.
Gen. Educ. Univ. Tokyo 16, 25-26 (1966).
YOOD, B. (see CIVIN, P.)
YOSIDA, K.
[I] Functional analysis. Berlin-Gottingen-Heidelberg, 1965.
ZAHARJUTA, V. P.
[I] Continuable bases in spaces of analytic functions of one and several vari-
ables. Sibirsk. Mat. Z. 8, 277-292 (1967) (Russian). Math. Rev:35, 1092 (1968).
ZIPPIN, M. (see also LINDENSTRAUSS, A.)
[I] On a certain basis in Co. Israel J. Math. 4, 199-204 (1966).
[2] On perfectly homogeneous bases in Banach spaces. Israel J. Math. 4,
265-272 (1966).
ZYGMUND, A. (see also PALEY, R. E. A. C.)
[I] Trigonometrical series. Warsaw, 1935.
Author and Subject Index
Absorbing subset 4 Basis,

Adjoint of a bounded linear Schauder basis 29, 117
operator 13 second category basis 109
Alaoglu theorem 13 shrinking 34
Algebra, similar generalized 123
semi-simple 7 T-Basis 45
commutative 7 total generalized 115
Arsove, M.G. 129 unconditional 38
Arsove theorem 115 uniform 42
Arsove-Edwards theorem 123 weak 28
weak* 28
B-algebra 7 weakly uniform 34
topology isomorphic 7 Bessaga, C. 68, 78
Ball, open 4 Bessaga-Pelczynski theorem
unit 5 65
Banach, S. 54 Bessel's inequality 6
Banach space 5 Biorthogonal system 31,114
Banach theorem 13 maximal 114
Banach-Steinhaus theorem 9 Boas, R. P 125
Barrel 4 Body 23
Barrel space (barrelled space) 4 Bounded subset 4
Barrel theorem 8 Bourbaki, N. 17
Base, at a point 2
local 3 Cantor's triadic (-ternary) point set
Basis 3, 28, 117 67
absolutely convergent 42 Cauchy net 5
block basis 65 Cauchy sequence 4
boundedly complete 36 Characteristic function
Cesaro basis 46 Circled subset 3
dual generalized 118 Closed graph theorem 9
equivalent 62 Closed set 2
extended Markushevich 115 Closure 2
first category basis 109 Cluster point 2
generalized 115 Coefficient functionals of a basis
Hamel basis 3 29, 115
Markushevich basis 115 Coefficient mapping 115
monotone 35 Compact set 2
normal 79 conditionally compact set 2
normalized 79 sequentially compact set 2
orthogonal 111 Conditionally weakly sequentially
orthonormal 79 complete set 12
retro-basis 34 Conjugate space 10
Author and Subject Index 147
Consistent infinite matrix 43 Gelbaum, B. R. 54, 68, 78, 85

Continuous mapping 2 Gelbaum theorem 72
Convergence 2 Grinblyum, M. M. 68, 86, 97
weak convergence 12 Grinblyum theorem 59
Convergence of a series,
absolute 22 Haar's system 49
bounded multiplier 19 Hahn-Banach theorem 11
reordered 18 Halmos, P.R. 17
subseries 19 Hamel basis 3
unconditional 18 Hausdorff, F. 17,67
unordered 18 Hilbert space 6
Convex subset 4 Hilbert space adjoint 14
Hildebrandt, T.H. 18,27
Davis, W.J. 129 Hille, E. 17, 27
Day, M.M. 17,18,27,54,68,78 HOlder's inequality 16
Dean, D.W. 97 Homeomorphic space 2
Dean theorem 96 Homeomorphism 2
Decomposition 86 Homomorphism 7
minimal Schauder decomposition
107 Ideal, left, right, two-sided, proper,
weak, Schauder decomposition maximal 7
86 Identity 8
Dense set 2 Inequality,
Determining manifold 12 Bessel's 6
Dieudonne, J. 17, 129 Holder's 16
Dimension of a linear space 3 Minkowski's 16
Directed set 2 Schwarz's 6
Direct sum 6, 10 triangle 4
Dirichlet's integral 53 Inner product 6
Domain 1 Invariant metric ( = translation
Dunford, N. 17 invariant metric) 5
Dvoretzky, A. 27 Inverse 1
Dvoretzky-Rogers theorem 27 Inverse image
Isometric Isomorphism 9
Edwards, R.E. 17,129 Isomorphic space 3
Eigenvalue 10 topologically isomorphic space 3
Endomorphism 9 Isomorphism 3
Equivalent norms 6 isometric 9
Equicontinuous family of linear topological 3
transformations 8 n-maximal topological 105
Euclidean space 6
Expansion operator 31 James, R.C. 54,68,78
Extension of a linear transformation James theorem 72
8
Kaczmarz, S. 52
Factor space 3 Kadec, M.l. 112
Fage, M. K. 86, 97 Karlin, S. 54, 78, 85
Field 1 Karlin corollary 42
Fourier series 53 Kelley, J.L. 17
F-space 5 Klee, V. 129
Fubini-Tonelli theorem 16 Klee theorem 118
Function I Kothe, G. 17
Functional, continuous linear 10 Kronecker symbol 6
148 Author and Subject Index
Lacunary series 52 Orthogonal system 60

Lebesgue dominated convergence Orthonormal sequence 6
theorem 16
Lindenstrauss, J. 68 Paley, R.E.A.C. 125
Limit 2 Paley-Wiener theorem 125
Linearindependence 3 Pelczynski, A. 68, 78
Linear space 1 Pelczynski theorem 77
Linear transformation 7 Pettis, B. J. 18
bounded 9 Phillips, R. S. 17,27
closed 8 Phillips theorem 14
compact 9 Principle of uniform boundedness
continuous 8 12
continuous at a point 7 Projection 10
equicontinuous family of 8 orthogonal 14
of finite rank 99 proper ideal 7
uniformly continuous 8 isomorphism 7
n-ring 104
Map = mapping 1 1Hing 100
Markushevich, A.1. 129 of the first or second category
Markushevich basis 115 101
Markushevich theorem 116 proper 104
Maximal biorthogona1 system 114 proper minimal 107
Maximal ideal 7
Mazur theorem 61 Quasi-norm 5
McArthur, C. W. 86, 97 Quasi-normed linear space 5
Metric, metric space 4
Michael, E. 68
Minkowski's inequality 16 Rademacher's function system
50
Radical 7
Namioka, I. 17
Range 1
Natural embedding 12
Reflexive space 13
Neighborhood 2 Resolvent of an operator
Net 2
10
a-net 5 Restriction of a linear
Newns theorem 126 transformation 8
Newns, W.F. 129 Retherford, J.R. 54,86,97
Nikol'skii, V. N. 68
Rickart, C. E. 17
Nikol'skii theorem 57
Riemann theorem 22
Norm 5 Rogers, C. A. 27
Null space 7 Ruckle, W.H. 86,97
One-to-one
Onto 1 Sanders, B. L. 86, 97
Open ball 4 Scalar 3
Open set 2 Schlifke, F. W. 125
Operator, see linear transformation Schauder's system 49
Orlicz, W. 18, 85 Schwartz, J. T. 17
Orlicz-Pettis theorem 19 Schwarz's inequality 6
Orthocomp1ement 6 Self-adjoint operator 14
Orthogonal complement 11 Semi-simple algebra 7
elements 60 Sequence 2
projection 14 Cauchy 4
Author and Subject Index 149
Sequence, Subspace 2
orthonormal 6 linear topological 3
T-limitable 43 Symmetric subset 4
Sequential compactness 2
Separable space 2 Taylor, A.E. 17,78
Set, Topology 2
bounded 4 equal 2
closed 2 locally convex 4
compact 2 metric 4
conditionally compact 2 norm 5
conditionally weakly sequentially strong 6
complete 12 stronger 2
sequentially compact 2 uniform operator topology 9
total 4 usual 4
total over 10 weak 10
weakly bounded 11 weak* 13
weakly closed 11 weaker 2
weakly sequentially complete 12 Total subset 4
simple .#i.-spaces 60 Totally bounded set 5
Singer, I. 54, 68, 112, 129 Transformation 1
Sobczyk, A. 97 additive 7
Space, continuous at a point 7
Banach space 5 linear 7
barrelled space (Barrel space) 4 Translation invariant metric 5
complete metric 4 Triangle inequality 4
conjugate 10 Trigonometrical system 50
Euclidean 6 Two-sided ideal 7
factor space 3
F-space 5 Uniform boundedness principle
finite dimensional 3 12
Hausdorff 2 Uniformly continuous linear
Hilbert space 6 transformation 8
homeomorphic (to) 2 Uniform operator topology 9
linear topological 3 Unit 7
locally convex 4
metric 4 Weak basis theorem 32
./11; -space 60 Weak convergence 12
simple 60 weakly bounded set 11
normed linear 5 weakly closed set 11
quasi-normed linear 5 Weakly sequentially complete 12
reflexive 13 Wiener, N. 125
separable 2 Wilansky, A. 17,54
topological 2 Wilansky theorem 34
Span 3
Spectrum of an operator 10 Yamazaki, S. 112
Steinhaus, H. 52 Y osida, K. 17
Stone-Weierstrass theorem 16
Strictly convex 81 Zygmund, A. 17,54
Springer Tracts in Natural Philosophy
Vol. 1 Gundersen: Linearized Analysis of One-Dimensional Magnetohydrodynamic

Flows
With 10 figures. X, 119 pages. 1964. Cloth DM 22,-; US $ 5.50
Vol. 2 Walter: Differential- und Integral-Ungleichungen und ihre Anwendung bei
Abschatzungs- und Eindeutigkeitsproblemen
Mit 18 Abbildungen. XIV, 269 Seiten. 1964. Gebunden DM 59,-; US $ 14.75
Vol. 3 Gaier: Konstruktive Methoden der konformen Abbildung
Mit 20 Abbildungen und 28 Tabellen. XIV, 294 Seiten. 1964
Gebunden DM 68,-; US $ 17.00
Vol. 4 Meinardus: Approximation von Funktionen und ihre numerische Behandlung
Mit 21 Abbildungen. VIII, 180 Seiten. 1964. Gebunden DM 49,-; US $ 12.25
Vol. 5 Coleman, Markovitz, Noll: Viscometric Flows of Non-Newtonian Flnids
Theory and Experiment
With 37 figures. XII, 130 pages. 1966. Cloth DM 22,-; US $ 5.50
Vol. 6 Eckhaus: Studies in Non-Linear Stability Theory
With 12 figures. VIII, 117 pages. 1965. Cloth DM 22,-; US $ 5.50
Vol. 7 Leimanis: The General Problem of the Motion of Coupled Rigid Bodies about
a Fixed Point
With 66 figures. XVI, 337 pages. 1965. Cloth DM 48,-; US $ 12.00
Vol. 8 Roseau: Vibrations non lineaires et theorie de la stabilite
Avec 7 figures. XII, 254 pages. 1966. Pleine toile DM 39,-; US $ 9.75
Vol. 9 Brown: Magnetoelastic Interactions
Vol. 10 Bunge: Foundations of Physics
With 5 figures. XII, 311 pages. 1967. Cloth DM 56,-; US $ 14.00
Vol. 11 Lavrentiev: Some Improperly Posed Problems of Mathematical Physics
With 1 figure. VIII, 72 pages. 1967. Cloth DM 20,-; US $ 5.00
Vol. 12 Kronmiiller: Nachwirkung in Ferromagnetika
Mit 92 Abbildungen. XIV, 329 Seiten. 1968. Gebunden DM 62,-; US $ 15.50
Vol. 13 Meinardus: Approximation of Functions: Theory and Numerical Methods
Vol. 14 Bell: The Physics of Large Deformation of Crystalline Solids
With 166 figures. X, 253 pages. 1968. Cloth DM 48,-; US $ 12.00
Vol. 15 Buchholz: The Confluent Hypergeometric Function with Special Emphasis on
its Applications
XVIII, 238 pages. 1969. Cloth DM 64,-; US $ 16.00
Vol. 16 Slepian: Mathematical Foundations of Network Analysis
XI, 195 pages. 1968. Cloth DM 44,-; US $ 11.00
Vol. 17 Gavalas: Nonlinear Differential Eqnations of Chemically Reacting Systems
With 10 figures. IX, 107 pages. 1968. Cloth DM 34,-, US $ 8.50
Vol. 18 Marti: Introduction to the Theory of Bases
XII, 150 pages. 1969. Cloth DM 32,-; US $ 8.00

Theory of Bases

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

Theory of Bases

Uploaded by

Copyright:

Available Formats

Springer Tracts in Natural Philosophy

Springer-Verlag Berlin Heidelberg New York 1969

ISBN·13: 978·3·642·87142·9 e·ISBN·13: 978·3·642·87140·5

Title No. 6746

Softcover reprint ofthe hardcover 1st edition 1969

Since the publication of Banach's treatise on the theory of linear

Although almost four decades have elapsed since Banach's query,

A few conditions sufficient for a sequence to form a basis of a certain

tial completeness, separability, reflexivity, dimension, or weak con-

Some generalizations of the concept of a basis for a Banach space

Although there are presently more than two hundred publications

The first chapter contains a short introduction to the working tools

the theory of proper 1Hings, are discussed in Chapter VIII. Some of

I thank the Editor, Professor B. D. Coleman for his friendly co-

Urbana, Illinois, February 1969 JURG T. MARTI

II. Convergence of Series in Banach Spaces 18

III. Bases for Banach Spaces. . . . . . . . . . 28

IV. Orthogonality, Projections and Equivalent Bases 55

V. Bases and Structure of the Space 69

VI. Bases for Hilbert Spaces. . . . 79

VIII. Applications to the Theory of Banach Algebras 98

IX. Some Results on Generalized Bases for Linear

Author and Subject Index. 146

1. Linear Topological Spaces

A mapping (sometimes also called map, function, transformation or

A topology for a set X is a family, of subsets of X, called open sets,

B of X is a Hamel basis for X if and only if every X in X has the unique

but finite integer which may depend on X. A finite number of elements

(x+ Y)+(y+ Y)=(x+y)+ Y, X,YEX,

o:(X+ Y)=o:x+ Y, O:ECP,XEX.

With the operations thus defined, the class X j Y is a linear space.

If A is a subset of X, then it is known that sp A is a (closed) linear

in X. If X is complete, then it is easy to see that every net {x;J in X such

2 Springer Tracts, Vol. 18 - Marti

orthonormal vectors in En is called an orthonormal basis for En and it is

there corresponds a neighborhood U of x with T(U)c V. T is continuous

Let T:D-> Y be a linear transformation into an F-space Y, with

Theorem 14. (BANACH-STEINHAUS) Let {Tn} be a sequence of opera-

Theorem 15. A linear transformation T:D- Y with domain D in X

Theorem 16.Let Tl EB(X, Y) and T2EB(Y,Z). If one of the operators

Theorem 17. The set of compact operators in B(X, Y) is closed in the

A linear transformation P:X -.X is called a projection in X (or of

We observe that there may be more than one projection of X onto

3. Conjugate Spaces and Weak Topologies

Theorem 8. Let Y be a closed linear subspace of X. Then y* is iso-

Theorem 10. To each x =f. 0 in X there exists an x* in X* with Ilx*11 = 1

topology if and only if it converges weakly to x. A subset A of X is said

X is reflexive if the natural embedding J: X ~ X** is onto. Finite

The weak* topology for the conjugate space X* can be introduced by

where G > 0 and r is a finite subset of X. A net {xi} in X* converges

Theorem 25. If T is a topological isomorphism of X onto Y, then T* is

Theorem 26. If S is in B(X, Y), and T is in B(Y,Z), where X, Y and Z

Let H be a Hilbert space and T an operator in B(H). There is a unique

A projection with the properties written in the foregoing theorem is

4. Special Banach Spaces

At this place we write down some properties of a few special Banach

isometric isomorphism T is given by T f3(a) = L ai f3i' aE Ip, f3 Elq.

Theorem 2. In 11 weak and strong convergence of sequences are the

Theorem 4. Let {f,,} be an equicontinuous sequence in C(S,X). If fn

Furthermore, if f is in Lp(S) and g is in Lq(S), then the product

Theorem 8. For 1 ~ p < 00 the subset C(S) of Lp(S) is dense in Lp(S).

everywhere in S, which converges almost everywhere to a function f on

Next, let {xn} be the trigono~etrical system, given by x o(s)=(2n)-t,

/lit lXiXi/i =SUP{lit lXiX(Xi)IIIIX11 ~ 1}

~s~p IlXil sup{tIX(XJIIIIX11 ~ 1}

Proof. By hypothesis suplx(Unx)1 < 00, XEX, XEX*. Thus there