COMMUNICATIONS ON PURE AND APPLIED MATHEMATICS,
VOL. x, 357-360 (1957)
Notes on Integration I:
The Underlying Convergence Theorem *
W. F. E B E R L E I N * *
University of Wisconsin and New York University
1. Introduction
Let C denote the (algebra and lattice of) real-valued continuous func-
tions on a compact Hausdorff space S. An integral over S is a positive linear
functional on C-that is, a mapping I of C into the real numbers satisfying
(1) I(af+bg) = a l ( f ) + b l ( g ) ,
(2) I ( / ) 2 0 whenever f 2 0.
The following proposition dates back to Arzelal, 1885, [l] and Osgood,
1897, [8]in the classical case: S = the closed finite interval (a, b ) andI(f) =
J: f ( W X .
THEOREM. Let limnf, = f fiointwise (f,, f in C ) and If,l 5 M < co for
all n. Then lim,I(f,) = I ( / ) .
This result plays a paradoxical role in integration theory? Although
the theorem concerns only continuous functions on S , the machinery of con-
ventional proofs stemming from Borel-Lebesgue (c. 1900, measure theory
[7]) or Daniel1 (1917, functional extension [4]) involves a preliminary ex-
tension of the mapping I to a larger class of (measurable) functions on S.
On the other hand it is known (cf. Banach, 1937, [3] and F. Riesz, 1952, [9])
that the theorem itself is a natural basis for extending I . This all but circu-
lar state of affairs may account for the variety of “elementary” proofs-
of varying degrees of generality-devised by such mathematicians as
Riesz (1917, [9] and 1952, [lo]) and Hausdorff (1927, 1151) (disguised ex-
tension theory), Bieberbach-Landau (1918, set of measure 0 theory [GI),
Banach (1932, set theory [2]) and others.
*This paper represents results obtained under the sponsorship of the National Science
Foundation, Contracts NSF-G2052 and NSF-G3050, and the Air Office of Scientific Research,
Contract AF lS(600)-1634.
**The author is a Temporary Member of the Institute of Mathematical Sciences,
New York University, for the academic year 1956-57.
’Arzela’s paper actually deals with sequences of Riemann integrable functions.
*For a discussion of the role of the theorem in modern functional analysis see Wymore,
1955, [12].
357
358 W. F. EBERLEIN
Here is a new and elementary proof of the general theorem-the basis
of a neogeometric approach to integration over function spaces, with ap-
plications to numerical analysis and physics. The author is indebted to
T. H. Hildebrandt and P. D. Lax for bibliographical criticism.
2. Proof of the Theorem
We show first that compactness of S makes I "countably positive":
LEMMA 1. (stones) If\ 5 2';"\/,I implies r(Ifl)5 CTI(lf,l).
Proof: Given E > 0 and z in S, there exists an integer N ( x ) such that
c:@)
[ f (z)I < If, (z)1 +E , and the inequality persists in some neighborhood U (x)
of 2 because of continuity. Since S is compact, the open covering ( U ( z ) )
(x in S) thus obtained contains a finite subcovering {U(xj)). Let N be the
largest iV(zi). Then If1 2 2: If,/+& implies
N
Wl) r 21 J ( I L l ) + E W
co
Z'(lf~l)+E'(l)J
1
whence I(lf1) 5 CyI(lf,l), since > 0 was arbitrary.
We require some elementary properties of the L p norms on C in the
distinguished cases p = 1, 2, co. Set l l f l l m = sup If(x)l (x in S) and note
that (1) and (2) imply
(A) II(f)l 5 J(lfl) 5 Ilfllm * I(1).
Thus we may first assume I(1) > 0 (non-trivial case) and then (on replacing
I by a suitable constant multiple)
(3) I ( 1 ) = 1.
Next set (1,g) = I(fg) (f, g in C) and = (1,f ) % = Finally set
IlflII = I(lfl). Then:
(B) I (f, g) I 5 I If112 * IIglI2 (Schwarz inequality) 3
(C) Ilf+slI~+fIf-gll; = 2{1lfll~+llg11;~ (parallelogram identity),
(Dl Ilflll -= I(If1)= (If19 1 ) 5 llf112 ' l l l l l 2 = Ilfll2>
(E) I If1;1 = I ( f 7 5 I lf21Im = I If1: 1 *
3This proposition is actually the basic axiom in Stone's theory of integration (1948,
[I l]), while that of the theorem is essentially the basic axiom in Banach's theory (1937, [2]).
NOTES ON INTEGRATION I 359
Consider now some preliminary reductions of the theorem itself. We
may assume f = 0 and f, 1 0 for all n (otherwise replace f, by If,-fl).
Clearly L = lim sup, I(/,) 5 M exists, and we need only show L = 0. But
we may assume limn I(f,) = L (otherwise pass to a subsequence). The pro-
position then becomes the
REDUCED THEOREM. Let: (a) f,ZieinCfor aZZn, (b) supnIlfnlImS M < c o ,
(c) limn f, = 0 $ointwise, (d) f, 2 0 for all n, (e) limn I(f,) = L. Then L = 0.
Our procedure is to establish the reduced theorem for a “smoothed”
sequence {h,}.
Let K , be the convex hull of the set [f, : m 1 n] -that is, K , consists of
all finite combinations 2 ajf,, , where each m, 2 n, a, 2 0, and 2 a, = 1.
Clearly any sequence {g,] with g, in K , satisfies (a)- (e). Set d, =
inf [ 1 Ig/l 2 : g in K,]. Since K,+, C K , , it follows that d, S d,+, 5 M for all
n (cf. (E)), whence d = limn d, exists. Now choose gn in K , such that
llgnllz 5 dn+l/n.
LEMMA 2. limm,n~~g,-gm~/2
= 0.
Proof:
IIgn-gmIIi = 2{IIgn./Ii+ IIgmIIi}-’IIi!(gn+gm)II~
by the parallelogram identity. Now n kmimplies +(g,+gm) is inK,, whence
lli!(gn+gm)llz 2 inf [IIgl12 : ginKm1 dm *
Hence
and the right hand side converges to 4d2-4d2 = 0 when m, n -+ co (n 2 m).
Finally choose (h,} as a subsequence of (g,} such that
W
c 1
lIhm-~h,+ll12 <
(For example, pick an increasing sequence {N,} of positive integers such that
Ilgi-gillz < 2-” whenever i, j 2 N , and set 12, = gN,.) Clearly (h,} satisfies
(a)+).
We conclude by showing L = lim, I(,%,) = 0. Since lim, h, = 0 point-
wise, h, = Lm (hm-hm+,), whence /h,I (= 2: ~12,-h,+,~. But then
m m
0 S II(hn)I 5 I(IhnI) 5 XI(I12m-hm+~I) 5
n
Z:n IIhm-hrn+lll2
by Lemma 1 and (D), and the last sum converges to 0 when n -+ 03.
‘Compare with the well-known device of Riesz, 1963, [lo] p. 71.
360 W. F. EBERLEIN
Bibliography
[ l ] Arzela, C., Sulla integrazione per serie, Rom. Acc. L. Rend., Vol. 1, 1885, pp. 532-537,
56G569.
[2] Banach, S., Thhorie des ophations lintaires, Monografje Matematyczne, Vol. 1, Warsaw,
1932.
[3] Banach, S., T h e Lebesgue integral ia abstract spaces, a note to S. Saks, Theory of the Integval,
Monografje Matematyczne, Vol. 2, Warsaw, 1933 (G. E. Stechert Co., New York,
1937).
[4] Daniell, P. J., A general form of integral, Ann. of Math., Vol. 19, 1917-1918, pp. 279-294.
[5] Hausdorff, F., Beweis eines Satzes won Arzela, Math. Z., Vol. 26, 1927, pp. 136-137.
[6] Landau, E., Ein Sat2 uber Riemannsche Integrale, Math. Z., Vol. 2, 1918, pp. 350-351.
[7] Lebesgue, H., Legons sur l’intdgration, Gauthier-Villars e t Cie., Paris, 1904, 2nd Ed., 1928.
[8] Osgood, W. F., Non-uniform convergence and the integration of series term by term, Amer.
J. Math., Vol. 19, 1897, pp. 155-190.
[9] Riesz, F., #ber Integrationen unendlicher Folgen. Jber. Deutsch. Math., Vol. 26, 1917,
pp. 274-278.
[lo] Riesz, F., and Sz.-Nagy, B., Lepons d’analyse fonctionnelle, Akadbmiai Kiad6, Budapest,
1952.
[ l l ] Stone, M. H., Notes o n integration, Proc. Nat. Acad. Sci. U.S.A., I. Vol. 34, 1948, pp.
336-342, 11. pp. 447-455, 111. pp. 483-490; IV. V O ~35, . 1949, pp. 50-58.
[I21 Wymore, A. W., O n W e a k Compactness in Functional Analysis, Thesis, University of
Wisconsin, Madison, 1955.
Received January, 1957.
