132 v.

Strong Convergence and Weak Convergence

Since B;,. is 58-measurable, it is easy to see that each x,. (s) is strongly 58-
measurable. Therefore x(s), which is the strong limit of the sequence
{x,.(s)}, is also strongly 58-measurable.
Proof of the Lemma. Let a sequence {X,.} be strongly dense in X.
Consider a mapping I -HP,. (I) = {I (Xl), I (X2), ... , I (X,.)} of the unit
sphere 5' = {I E X'; // 1// < 1} of X' into an n-dimensional Hilbert space
l2 (n) of vectors (;1> ;2' ... , ;,.) normed by /I (;1' ;2, ... , ;,.) /I =
(,. /;jI2 )1/2 .
The space l2(n) being separable, there exists, for fixed n, a sequence
{/,.,k} (k = 1, 2, ... ) of 5' such that {q;,. (I,.,k); k = 1, 2, ... } is dense in the
image q;,.(5') of 5'.
We thus have proved that, for any 10 E 5', we can choose a subsequence
{t,.,m,.} (n = 1, 2, ... ) such that /I,.,m,. (Xi) - 10 (Xi) / < lin (i = 1, 2, ... , n).
Hence ..-colim Inm
,,. (Xi) =/O(Xi) (i= 1,2, ... ), and so, by Theorem lOin Chap-

terV, 1, we obtain that lim I,.,m,.(x) = 10 (x) for every xE X.

o. Bochner's Integral
Let X (s) be a finitely-valued function defined on a measure space
(5,58, m) with values in a B-space X; let x(s) be equal to Xi =F- 0 on
B; E 58 (i = 1, 2, ... , n) where B/s are disjoint and m (Bi) < ex:> for
i = 1, 2, ... , n, and moreover, x(s) = 0 on (5 -.1
.=1
B;). Then we can
define the m-integral J
X (s) m (ds) of x (s) over 5 by 1 Xi m (B;). By
s <=1
virtue of a limiting procedure, we can define the m-integral of more
general functions. More precisely, we have the
Definition. A function x (s) defined on a measure space (5, 58, m) with
values in a B-space X is said to be Bochner m-integrable, if there exists
a sequence of finitely-valued functions {X,. (s)} which s-converges to x (s)
m-a.e. in such a way that
lim
..-cos
J I/ x (s) - x,,(s) /I m(ds) = O. (1)

For any set BE 58, the Bochner m-integral of x (s) over B is defined by
Jx(s) m(ds) = s-lim JCB(s) x,.(s) m(ds), where CB is the
B "--?OO S (2)
defining function of the set B.
To justify the above definition, we have to verify that the s-limit on
the right of (2) exists and that the value of this s-limit is independent of
the approximating sequence of functions {X,. (s)}.
Justification of the Definition. First, x(s) is strongly 58-measurable
and consequently the condition (1) has a sense, since I/x(s) -- X,.(s) /I is
 o. Bochner's Integral 133
as shown in the proof of Petti's Theorem. From the in-
equality
111 x,,(s) m(ds) -1 x,,(s) m(ds)il = 111 (x,,(s) -x,,(s)) m(ds) I
< J II x" (s) - x,,(s) II m(ds) < J IIx.. (s) - x(s) II m(ds)
8 S

+ J IIx(s) - x,,(s) II m(ds)

s
and the completeness of the space X, we see that s-lim J x.. (s) m(ds)
t>-+OO 8
exists. It is clear also that this s-lim is independent of the approximating
sequence, since any two such sequences can be combined into a single
approximating sequence.
Theorem 1 (S. BOCHNER [1]). A strongly function x (s) is
Bochner m-integrable iff II x (s) II is m-integrable.
Proof. The "only if" part. We have IIx(s)11 < II x.. (s)1I + IIx(s) -x" (s)lI.
By the m-integrability of IIx.. (s) II and the condition (1), it is clear that
IIx(s) II is m-integrable and
J IIx(s)
B
II m(ds) < J IIx,,(s)
8
II m(ds) + J IIx(s) -
B
x,,(s) II m(ds).
Moreover, since
J IlIx,,(s) 1I-lIx,,(s) III m(ds) < J IIx,,(s) - x,,(s) II m(ds) ,
B 8

we see from (1) that lim J

t>-+OO
II x". (s) II m (ds) exists so that we have
B

8
J II x (s) II m(ds) < lim J II x" (s) II m(ds).
t>-+OO B

The "if" part. Let {x" (s)} be a sequence of finitely-valued functions

strongly convergent to x (s) m-a.e. Put
y .. (s) = x.. (s) if IIx.. (s) 1/;:;;; I/x(s) 1/(1 + 2-1),
0 if = I/x.. (s) 1/> I/x(s) 1/(1 + 2-1) .
Then the sequence of finitely-valued functions {y.. (s)} satisfies II y" (s) II <
IIx(s) II· (1 + 2-1) and lim IIx(s) - y,,(s) 11= 0 m-a.e. Thus, by the
t>-+OO

m-integrabilityof IIx(s) II, we may apply the Lebesgue-Fatou Lemma to

the functions IIx(s) - y,,(s) II < IIx(s) II (2 + 2-1) and obtain
lim Jllx(s)-y,,(s)lIm(ds) =0,
t>-+OO s
that is, x(s) is Bochner m-integrable.
Corollary 1. The above proof shows that
l"x(s) II m(ds) 111 x(s) m(ds) II,
134 V. Strong Convergence and Weak Convergence

and hence Jx(s) m(ds) is m-absolutely continuous in the sense that

B
s-lim J x(s) m(ds) = O.
m(B)--+O B

The finite additivity J

..
x(s) m(ds) = ..I"
J=1 B;
J x(s) m(ds) is clear and
I: B;
;=1

so, by virtue of the a-additivity of J II x (s) Ii m (ds) , we see that

B
J x(s) m(ds) is a-additive, i.e.,
B

J
00
B = ..I B j with m (Bj )
J=1
< <Xl implies x(s) m(ds)
00
I: B;
1=1

= "
s-lim ..I
_00)=1 B;
Jx (s) m (ds).
Corollary 2. Let T be a bounded linear operator on a B-space X into a
B-space Y. If x(s) is an X-valued Bochner m-integrable function, then
Tx(s) is a Y-valued Bochner m-integrable function, and
J Tx(s) m(ds) =
B
T
B
Jx(s) m(ds).
Proof. Let a sequence of finitely-valued functions {Yn (s)} satisfy
Ily,,(s) II < Ilx(s) II (1 + n- 1) and s-limYn(s) = x(s) m-a.e .
...-..00

Then, by the linearity and the continuity of T, we have J TYn (s) m (ds) =
B
T JYn(s) m(ds). We have, moreover, by the continuity of T,
B

IITy,,(s) II IITII·IIYn(s) II < IITII'lIx(s) II· (1 + n- 1) and

s-lim TYn(s) = Tx(s) m-a.e.

Hence T x (s) is also Bochner m-integrable and

J Tx(s) m(ds) =
B
s-lim
,,->00
J Ty,,(s) m(ds) =
B
s-lim T
_00
J Yn(s) m(ds)
B

=T
B
Jx(s) m(ds).
Theorem 2 (S. BOCHNER [lJ). Let 5 be an n-dimensional euclidean
space, )8 the family of Baire sets of 5, and m(B) the Lebesgue measure
of B. If x (s) is Bochner m-integrable, and it P (so; (X) is the parallelopiped
with centre at So E 5 and side length 2(X, then we have the differentiation
theorem
s-lim (2(X)-n
,,+0
J
x (s) m (ds) = x (so) for m-a.e. so'
P(s,;,,)
 5. Bochner's Integral 135
Proof. Put
(2ar n J
P(s,;",)
x(s) m(ds) = D (x; so' 1X).

If {xn(s)} is a sequence of finitely-valued functions such that Ilxn(s) II <

llx(s) II· (1 + n-1) and s-lim xn(s) = x(s) m-a.e., then

D (x; so, 1X) - x (so) = D (x - Xk; so, 1X) + D (Xk; so, 1X) - X (so) ,
and so
lim liD (x; so, 1X) -x (so) II < lim D(llx - Xk II; SO,1X)
"'to "'to
+ lim
"'to
IID(xk;so,1X) -Xk(SO) II + Ilxk(so)-x(so)ll·
The first term on the right is, by Lebesgue's theorem of differentiation of
numerical functions, equal to Ilx (so) - Xk(SO) II m-a.e. The second term
on the right is = 0 m-a.e., since Xk(S) is finitely-valued. Hence

lim II D (x; SO' 1X) - X (so) II < 211xk (so) - X (so) Il for m-a.e. so·
",... 0

Therefore, by letting k --+ 00, we obtain Theorem 2.

Remark. Contrary to the case of numerical functions, a B-space-
valued, a-additive, m-absolutely continuous function need not necessarily
be represented as a Bochner m-integral. This may be shown by a counter
example.
A Counter Example. Let 5 = [0, 1J and the family of Baire sets
on [0, 1J, and m (B) the Lebesgue measure of B E Consider the totality
m [1/3, 2/3J of real-valued bounded functions = (0) defined on the
closed interval [1/3,2/3J and normed by II = sup I. We define
o
an m [1/3, 2/3J-valued function x (s) = (0; s) defined on [0, 1J as follows:

j the graph in s-y plane of the real-valued function y = Yo (s),

which is the O-coordinate (0; s) of x (s), is the polygonal line
connecting the three points (0, 0), (0, 1) and (1, 0) in this order.
Then, if s oF s', we have Lipschitz' condition:
II (s - S')-1 (x(s) - x(s') II = sup I(s - S')-1 s) - s') I < 3.
o
Thus, starting with the interval function (x(s) - x(s') taking values in
m [1/3, 2/3J, we can define a a-additive, m-absolutely continuous set
function x (B) defined for Baire set B of [0, 1].
If this function x(B) is represented as a Bochner m-integral, then,
by the preceding Theorem 2, the function x (s) must be strongly differen-
tiable with respect to s m-a.e. Let the corresponding strong derivative
x' (s) be denoted by 'fJ (0; s) which takes values in m [1/3, 2/3]. Then, for
136 Appendix to Chapter V. Weak Topologies and Duality

every 0 E [1/3, 2/3J and m-a.e. s,

0= lim II h-1 (x (s
h-o
+ h) - x (s)) - x' (s) II > lim Ih-1
h-o
s+ h) (0 ;s))

-1](O;s)l·
This proves that (0; s) must be differentiable in s m-a.e. for all 0 E
[1/3,2/3]. This is contradictory to the construction of s).

References for Chapter V

S. BANACH [lJ, N. DUNFORD-J, SCHWARTZ [lJ and E. HILLE-R. S.
PHILLIPS [1].

Appendix to Chapter V. Weak Topologies and Duality

in Locally Convex Linear Topological Spaces
The present book is so designed that the reader may skip this appen-
dix in the first reading and proceed directly to the following chapters.

1. Polar Sets
Definition. Let X be a locally convex linear topological space. For
any set M £;; X, we define its (right) polar set MO by
MO = {x' EX'; sup I<x, x') I ::;; 1}. (1)
"EM
Similarly, for any set M' £;; X', we define its (lelt) polar set OM' by
OM' = {xE X; sup I<x, x') I < 1} = X (\ (M')O, (2)
"'EM'
where we consider X to be embedded in its bidual (X;)'.
A fundamental system of neighbourhoods of 0 in the weak topology
of X is given by the system of sets of the form OM' where M' ranges over
arbitrary finite sets of X'. A fundamental system of neighbourhoods of
o in the weak· topology of X' is given by the system of sets of the form
MO where M ranges over arbitrary finite sets of X. A fundamental
system of neighbourhoods of 0 in the strong topology of X' is given by
the system of sets of the form MO where M ranges over arbitrary bounded
sets of X.
Proposition. MO is a convex, balanced set closed in the weak· topology
of X'.
Proof. For any fixed x E X, the linear functional I (x') = <x, x') is
continuous in the weak· topology of X'. Thus MO = n {m}O is closed
mEM
in the weak· topology of X'. The balanced convexity of MO is clear.
 1. Polar Sets 137
An Application of Tychonov's Theorem
Theorem 1. Let X be a locally convex linear topological space, and A
a convex, balanced neighbourhood of 0 of X. Then AO is compact in the
weak* topology of X'.
Proof. Let p (x) be the Minkowski functional of A. Consider, for each
xE X, a sphere 5,. = {zE C; Izl < P(x)} and the topological product
5 = lIS,.. 5 is compact by Tychonov's theorem. Any element x' EX'
,.EX
is determined by the set of values x' (x) = <x, x'), x EX. Since x E
(P(x) + e)A for any e> 0, we see that x' E X' implies <x, x') =
«P(x) + e)a, x') with a certain aE A. Thus x E AO implies that
lx' (x) I <P(x) + e, that is, x' (x) E 5,.. Hence we may consider AO as a
subset of S. Moreover, it is easy to verify that the topology induced on
A °by the weak* topology of X' is the same as the topology induced on
AO in the Cartesian product topology of 5 = lIS".
xEX
Hence it is sufficient to prove that A °is a closed subset of S. Suppose
Y = II y (x) is an element of the weak* closure of AO in S. Consider any
"EX
e > 0 and any Xl' x 2 EX. The set of all U = II U (x) E 5 such that
"EX
Iu(xl ) -y(xl ) I < e, lu(x2 ) -y(x2 ) I < e and IU(XI + x 2 ) -Y(XI + x2 ) I < e
is a neighbourhood of y in S. This neighbourhood contains some point
x' E AO and, since x' is a continuous linear functional on X, we have
IY(XI + x 2) - y(xl ) - y(x2) I < IY(XI + x 2) - <Xl + x 2 , x') I
+ I <xl' x') - Y (Xl) I + I<x2 , x') - Y (x 2) I < 3e.
This proves that Y (Xl + x 2) = Y (Xl) + Y (x 2 ). Similarly, we provethat
Y ({3 x) = {3 Y (x), and so Y defines a linear functional on X. By the fact
that Y = II y(x) E 5, we know that IY(x) I <P(x). Since P(x) is con-
xEX
tinuous, Y (x) is a continuous linear functional, i.e., y E X'. On the other
hand, since y is a weak* accumulation point of AO, there exists, for any
e> 0 and a E A, an x' E AO such that Iy(a) - <a, x') I <e. Hence
Iy(a) I < I<a, x') 1+ e < 1 + e, and so Iy(a) I < 1, that is, yE AO.
Corollary. The unit sphere 5* = {x' E X'; II x' II < I} of the dual space
X; of a normed linear space X is compact in the weak* topology of X'.

An Application of Mazur's Theorem

Theorem 2. Let M be a convex, balanced closed set of a locally convex
linear topological space X. Then M = O(MO).
Proof. It is clear that M O(MO). If there exists an X o E O(MO) - M,
then, by Mazur's theorem 3 in Chapter IV, 6, there exists an EX'
such that <xo' > 1 and I<x, I < 1 for all x EM. The last inequality
shows that E MO and'so X o cannot belong to O(MO).
138 Appendix to Chapter V. Weak Topologies and Duality

2. Barrel Spaces
Definition. In a locally convex linear topological space X, any convex,
balanced and absorbing closed set i<; called a barrel (tonneau in Bourbaki's
terminology). X is called a barrel space if each of its barrels is a neighbour-
hood of O.
Theorem 1. A locally convex linear topological space X is a barrel
space if X is not of the first category.
Proof. Let T be a barrel in X. Since T is absorbing, X is the union of
closed sets nT = {nt; t E T}, where n runs over positive integers. Since
X is not of the first category, at least one of the (n T)' s contains an interior
point. Hence T itself contains an interior point xO. If Xo = 0, T is a neigh-
bourhood of o. If Xo # 0, then -xo E T by the fact that T is balanced.
Thus -xo is an interior point of T with xO. This proves that the convex
set T contains 0 = (xo - xo)/2 as an interior point.
Corollary 1. All locally convex F-spaces and, in particular, all B-
spaces and (Rn) are barrel spaces.
Corollary 2. The metric linear space 'tI K (Rn) is a barrel space.
Proof. Let {rpk} be a Cauchy sequence with respect to the distance

d·IS (rp, 1p ) __ 2- m1 +
= m=O PmP(ffJ (- _tp) )' h
were P.m (rp ) = sup /Dj rp ()
x /.
m ffJ tp liI;;;;m,xEK

For any differential operator Dj, the sequence {Djrpk (x)} is equi-contin-
liOUSand equi-bounded, that is,
lim sup /Dirpk (Xl) - Dirpk (x 2 ) / = 0 and sup /Djrpk (x) / < 00.
IX'_X'I j, 0
This we see from the fact that, for any coordinate x"
Dirpk(X)i < 00. Hence, by the Ascoli-Arzela theorem, there
exists a subsequence of {DJ rpk' (x)} which converges uniformly on K.
By the diagonal method, we may choose a subsequence {rpk" (x)} of {rpk (x)}
such that, for any differential operator Di, the sequence {Dirpk" (x)}
converges uniformly on K. Thus
lim Dirpk"(X) = Dirp(x) where rp(x) = lim rpk"(X),
k" --->00 k"--->oo

and these limit relations hold uniformly on K. Hence the metric space
'tIK (Rn) is complete and so it is not of the first category.
Remark. (i) The above proof shows that a bounded set of 'tI (Rn) is
relatively compact in the topology of 'tI (Rn). For, a bounded set B of
<1) (R") is contained in some 'tIK (Rn) where K is a compact set of and,
moreover, the boundedness condition of B implies the equi-boundedness
and equi-continuity of {Dirp; rp E B} for every Di. (ii) Similarly, we see
that any bounded set of (R") is a relative I y compact set of (Rn).