2.

Vitali's convergence theorem

Objekttyp: Chapter

Zeitschrift: L'Enseignement Mathmatique

Band (Jahr): 47 (2001)

Heft 3-4: L'ENSEIGNEMENT MATHMATIQUE

PDF erstellt am: 03.05.2017

Nutzungsbedingungen
Die ETH-Bibliothek ist Anbieterin der digitalisierten Zeitschriften. Sie besitzt keine Urheberrechte an
den Inhalten der Zeitschriften. Die Rechte liegen in der Regel bei den Herausgebern.
Die auf der Plattform e-periodica verffentlichten Dokumente stehen fr nicht-kommerzielle Zwecke in
Lehre und Forschung sowie fr die private Nutzung frei zur Verfgung. Einzelne Dateien oder
Ausdrucke aus diesem Angebot knnen zusammen mit diesen Nutzungsbedingungen und den
korrekten Herkunftsbezeichnungen weitergegeben werden.
Das Verffentlichen von Bildern in Print- und Online-Publikationen ist nur mit vorheriger Genehmigung
der Rechteinhaber erlaubt. Die systematische Speicherung von Teilen des elektronischen Angebots
auf anderen Servern bedarf ebenfalls des schriftlichen Einverstndnisses der Rechteinhaber.

Haftungsausschluss
Alle Angaben erfolgen ohne Gewhr fr Vollstndigkeit oder Richtigkeit. Es wird keine Haftung
bernommen fr Schden durch die Verwendung von Informationen aus diesem Online-Angebot oder
durch das Fehlen von Informationen. Dies gilt auch fr Inhalte Dritter, die ber dieses Angebot
zugnglich sind.

Ein Dienst der ETH-Bibliothek

ETH Zrich, Rmistrasse 101, 8092 Zrich, Schweiz, www.library.ethz.ch

http://www.e-periodica.ch
 Fatou's proof of his lemma is very similar. It should be noted that Fatou's
long paper is one of the most important of the century. For the first time the
new theory of intgration is applied to complex function theory; there are
also fundamental applications to trigonomtrie sries.
It is not until 1908, that DCT first appears in Lebesgue (1908) p. 9-10
[16] with a sketch of the proof; the same thing happens in Lebesgue (1909)
[17] at the top of p. 50. In thse papers Lebesgue seeks to apply his new
results and finds BCT insufficient. In Lebesgue (1910) [18], in 15 on page
375, the proof of DCT is given in more dtail, still on a set of finite measure.

Sketch of his proof. Let e > 0; since g is integrable on E, there

exists a number M>o, such that
Jf g<s,
where F={g>M}; then

/ |/n -/| < 2e, and on E\F 9

the resuit follows by BCT.
Jf
Note that ail the theorems so far hve been stated and proved for sets E
of finite measure. Theredoes not seem at that time to hve been much interest
on anyone's part in extending the results and proofs for the case m(E) = +00 .
However, (excluding of course BCT) this is easily done.

2. Vitali' s convergence theorem

In 1907, before Lebesgue announced DCT, there appeared a remarkable

paper by G. Vitali [29], which, I feel, has not received its due, even from
Hawkins. In it Vitali proves the following resuit:

Let E be a set of finite measure (finiteness is essential hre). Let {/?}

be a squence of integrable functions such that n ?> ci:e. with finite f f f
a.e. Then fis integrable and Jf f n ?>
Jf
f, for every measurable subset F

of E, if and only if the intgrais

f
lfn are uniformly absolutely continuous
Ja
(uniformly in n): given e > 0, there exists > 0, such that if m(A) < 5,

then
\Ja f
n <e for qU n.

This implies that / f \f n

\
< 2e. Vitali calls this equi-absolutely continuous.
Ja
Note that this resuit generalizes at once to any finite measure space.
 Vitali first proves that uniform absolute continuity is sufficient for
/fn?>
Jf //
Jf
for ail measurable subsets F of E.

Sketch of his proof. N, let Gh = {\fn > h for some n}.

For h G \ 2h2

If F^ = E\Gh, then the squence {/?} is uniformly bounded on Th for

every /gN. Thus on ail measurable subsets of Th , convergence of intgrais
follows by BCT. On the other hand, Gh is a decreasing squence of sets
and m(f] h G h ) =o,so m(Gh) |O.So for ail sufficiently large h, the uniform
absolute continuity condition implies that the contribution of the intgrais over
Gh is small.

Vitali next proves necessity of the uniform absolute continuity condition

when the functions /? are ail non-negative.

Sketch of his proof If /fn are not uniformly absolutely continuous,

then for some e > 0, there exists for each > 0, a measurable set F
with m(F) < and n G N with /fn > s . Let t > 0 with Zt < 00 .

Jf
For each t , there exists a measurable set G t E and n t G N such that
r
m(Gi) <Si and /
Jd f n .
>e. Let T r = \J? r Gi. Then Tr decrease with r and
oo
m(T r ) K^t^O as r-> 00. For ail f>r, / /?. >e. Since /?. -^/ a.e.

as /?> 00, /
Jrr
/?. ? >
//by hypothesis, and so for each
Jv r
r,
JTr f>s>o.
Put T= D~i rr . Then m(T) =0, but [f>e>o.
Jy
Contradiction.

Finally Vitali proves necessity in the gnerai case. If /? - / a.e. on E

and /? are completely integrable on , i.e. //? ->
Jf // for
Jf
every measurable

subset FofE, then / /? are uniformly absolutely continuous, i.e. given

e>o there exists >o, such that if m(A) <S, then ffn <e for ail
n G N. (This is the deepest and hardest part of Vitali's paper, and is in a
sens 'new' even after 93 years !)

Proof (Vitali). Ail the sets that occur in this proof will be measurable,
even when this is not explicitly stated.
 Step I. If
resuit is true.
fn > 0 for ail n G N, we hve already seen above that the

Step IL Suppose now that />0 a.e. on E; we can assume that />0
on ail of E. Note first that if /? ? >
/ boundedly then BCT implies that
/ \fn ~f\ ~^ 0 as n -^ 00, and so given e > 0, there exists iVGN such

that for ail FE and n>N,we hve

//?-// <e.
Let Gn ={0 <J5 < 2", V/ > w}. Gn Gn+lG

n+1 for ail n and = (J^ Gn .

So n|o and m(E\Gn ) |0.

Now given a > 0, there exists min, a) such that if Fn Gn then

a
fj~f < fr
strictly increasing squence of positive integers n\
aU 7 min, a). Let ? >o,en |O.We
< n2n
2 < ??? <
can find

m < ???

such that for every subset Tn of Gm , .

(t)
For every positive integer n > ni, there exists a unique i G N such that
ft/+i < < z+2- For such n, put

Let F be a subset of and r w. = Gni HF. For fixed

If n i+ i <n< rii+2, then /

JT ni
gg n = gn , and so, using (f)
JT

Thus n are completely integrable on E and since g n >0 for ail n, it

follows by Step I that / gn are uniformly absolutely continuous on E. Put
cj)
n =f n ? gn . The (/>
n are completely integrable on E . To complte the proof

of Step II we must show that / <f)

n are uniformly absolutely continuous.
Observe that if nt+\ <n< n+2 , then (j) n {x) =0 for ail xG Gn/ . So for ail

xeE, lim</> n (X) =0, and for any measurable subset 1 CE, / (j) n ?> 0.

Suppose that / </>

n are n<9^ uniformly absolutely continuous. Then there
exists a>o, such that for ail >0 and TV GN, there exist F with

Irff
m(F) < jla and n>N such that
Jt
cj)
n >a. Let 771,772, ... be >0, and

such that JLrji < ? Let Fi be . a subset of E for which there exists t\ GN

with / 0A > cr . Since lim / (j) h = / ofl,0

fl , we can find z'i GN such
1 / 1 1 / Gni ni 1
-/ii
that / (j>
h >a. Now there exists /ii >0 such that if ra(F) < /ii ,
?/Gn^ rm
then f
/ </)
fl < 771 . By our assumption, there exists F2F
2 such that m(F2 ) < \i\
h
and 2 > n Zl+ i such that / cj>
t2 >a.By the same reasoning, there exists
Jv2
z'
2 (necessarily >i\) such that / <p
t2 >a.
JGnh nr 2
Now there exists fi 2 >0 such that if mT) < /i2 , then

There exist a subset F3F

3 with m(F 3 ) < /i 2 and t3t
3 > n i2 +i, such that

\
<t>t
3 > cr- Again we can find i3i
3 GN such that / </>
h >a.
Now there exists fi 3 >0 such that if ra(F) < fx 3 , then

Continue in this way to obtain an increasing squence z),^ > 0, and

h> riij^+i and sets F; such that m(Tj) < x and / 4>
tk < f]j_ u

k= 1,2, ... J- 1, but / <^

r >C 7. Let
We claim the sets Qy are disjoint. Since tj >
-

rc^+i we hve (j)tj ?oon Gm : , .

and so (f> tj+l ?oon Gni .So Q/+i HGW Since Gn increase with n, i.e. .
=0.
Gn : GGn +i for ail n, it follows that Q + i
; 1, 2, . . . J, and so
-

DQk?oif&=
Q,j are ail disjoint. Put Q= (J)^ Qy-. Then since <^. =oon Qi, . . , Q -_i , .
;

But / <j)
tj >g and / </>
tj < 7#+ A _i, so
r/ <j>
tj <7:? Hence

/&:>7T for ail j. But since <fi n are completely integrable and <fi
n (x) ? 0

for ail xGE it follows that / f <fi

tJ
. ? 0 as j?> 00 . This is a contradiction,
Ji
completing the proof of Step IL

Step 111. In the gnerai case, put g n =/??/+ 1 . Then #? ? 1 a.e.

on E, and gn is completely integrable. Hence as seen in Step 11, the resuit
is true for the gn , i.e., /gn are uniformly absolutely continuous. Hence the

same holds for / /? , completing the proof.

COMMENTS

The concept of complte integrability of a squence is weaker than weak

sequential convergence in Ll.L . Vitali was of course, in 1907, unaware of L
1 1

convergence (strong or weak) and its significance. Using Vitali 's proof that
complte integrability implies uniform absolute continuity, we see that it also
1
implies convergence in . But a direct proof that complte integrability
Ll.L

implies L convergence (without using Vitali 's resuit) seems hafd.

1

A feature of this remarkable paper is that ail. the results are stated and
proved in terms of sries of fonctions rather than squences ; thus one has to
realize that a "sries ail of whose partial sums are non-negative" corresponds
to a squence of non-negative terms, and is not to be confused with "a sries of
non-negative terms", which corresponds, of course, to an increasing squence
of non-negative terms Vitali's theorem is obviously a generalization of BCT.
!

He remarks that Beppo Levi's MCT follows from it. Of course, so also does
DCT, but Vitali did not know about DCT at the time.
The history and use of this resuit between 1907 and 1939 is something
I would like to know more about ! Hawkins [12] mentions the paper, but
quotes only what I consider to be a much less important resuit at the end of
the paper. In a footnote on p.s(This 1909 paper [17], Lebesgue says that
DCT and MCT are spcial cases of Vitali's convergence theorem. He also
DCT can be extended to sets of infinit measure. On p> 365 of the
states that
1910 paper [18] he again refers to Vitali's Theorem, saying that it gives a
necessary and sufficient condition for term by term intgration. In Leons II
[19] (p. 131) Lebesgue merely refers to the paper: M. Vitali a crit sur ce
sujet un trs important Mmoire, que je ne puis ici que signaler; this is just
before he gives DCT. In 1913, Camp, in a rather messy paper [2], gives a
generalization of Vitali's theorem to several variables.
In 1915, de la Valle Poussin, wrote a long paper [27] entitled Sur
l'intgrale de Lebesgue; this article is complementary to his book Intgrales
de Lebesgue, fonctions d'ensembles, classes de Baire [28] written at about
the same time. In the paper, in the section on convergence theorems, de la
Valle Poussin discusses Vitali's work, and in the proof of Theorem 4 on
p. 448-450 he simplifies considerably the hard part of Vitali's proof; we give
a sketch of his argument.

It is clearly sufficient to prove that if /? ? oonE,

> and / /? are not
uniformly absolutely continuous on E, then there exists F E such that
fn - 0. (We know this is true if /? >0 on E.) Let A m be a squence (to
Jf
be chosen later) such that o<Am < m+1 for ail mGN, and A m - +00,
Am+lA

and let Em ={xGE: \f

n (x)\ >Am for some nG N}. Note that the measure
of Em tends to zro.
Let e= limsup / \fn \\ e>o, else \fn \
are uniformly absolutely
continuous. Let uj >0 with u< s/6. It is fairly easy to choose A m so that
for each m G N, there exists n G N such that the following three inequalities
are satisfied:

(I)
This is done inductively: for each m, we can find n so that the first two
inequalities are satisfied and then choose A m+i (> A m ) depending on nso that
the third is satisfied. Further we can choose n increasing and
? oo with m. >

(t) implies that / \f \> e-2cu, and so there exists

n Fm Em \Em+l
JEm \Em +i

f 1
so that
m
f
n > -(e - 2u). Put F=(JFm
ra=l
(disjoint union). Note that
Fi E\Em for i=
1, . . , m?l and Fh EEm +\ for
.
/z
=m+l,m+2,
Hence for each m, and the corresponding /,

which is positive since cj < s/6. Hence ffn -*> 0, completing the proof.
Jf
The argument is very similar to Step 1 in Hahn's proof of the Vitali-
Hahn-Saks Theorem [10] given in 3, and it is at least conceivable that Hahn
got the initial impetus for his proof from de la Valle Poussin's paper. In
[27], Theorem 5 on p. 450, de la Valle Poussin shows that uniform absolute
continuity of /fn on a space of finite non-atomic measure is quivalent to :

Given s>o, there exists K>o, such that for ail n EN, / \fn \<e.
H\fn\>K}

This was rediscovered by Doob [3] 24 years later, the new criterion was
called uniform integrability, and used extensively by Doob in his study of
martingales. In 1918, H. Hahn [9, p. 1774] showed, using Vitali's resuit, that
complte integrability implies strong L convergence: this shows that Hahn
1

was aware of Vitali's paper. However de la Valle Poussin's paper seems to be

virtually unknown ! I learnt about it from the excellent set of bibliographical
rfrences on p. 223 of Hahn and Rosenthal [11]. Nagumo [21] discusses the
theorem with rfrence to Vitali, uses it, and gives a necessary and sufficient
condition for uniform absolute continuity. Vitali himself does not seem to hve
worked further on this subject. See the biographical article by A. Tonolo [26].
Among well-known books on real analysis written before World War II
only Hobson [13] refers to Vitali's paper on p. 296-299. Hobson also has
what is probably the first attempt to generalize Vitali's resuit to sets of infinit
measure ; thse are ail, in my opinion, somewhat artificial. Since the fifties some
books on analysis and/or probability hve included the concept of uniform
absolute continuity or uniform integrability, but often without any mention
of Vitali. Also, where there is a rfrence to Vitali, the resuit attributed to
him is often the quivalence of uniform absolute continuity and strong L
1
convergence, which follows from the easy part of Vitali's work, whereas
complte integrability is not mentioned. Rudin, Real and Complex Analysis
[24], is an exception -
in ail three ditions; however, in the first dition the
Vitali convergence theorem is given, by the third dition this has changed
to the Vitali-Hahn-Saks theorem. Dunford-Schwartz [6] has a comprehensive
account in Chapters 111 and IV. Unfortunately there is a slip in the statement
of Vitali's convergence theorem on p. 234.

3. The Vitali-Hahn-Saks Theorem

Vitali's convergence theorem is regarded as the origin of this theorem.

It was first stated and proved by H. Hahn [10] in 1922. Hahn's statement
and proof follow. (Both this resuit and Corollary 2 are referred to as "The
Vitali-Hahn-Saks Theorem". The resuit is obviously stronger than Vitali's
convergence theorem.)

Theorem (H. Hahn [10] Thm. XXI, pages 45-50). // m(E) < 00,
/? integrable on E, and for each measurable FE, lim /fn exists and is
n-*n

-* Jf
finite, then Ifn are uniformly absolutely continuons.

Proof. Again, ail the sets that occur in this proof will be measurable.
Suppose the intgrais are not uniformly absolutely continuous. Then there
exists e > 0 with the property that for each N e N and a > 0 there is

a measurable set Z with m(Z) <a and n 0 >N with / \f

no
\

>e.By
Jz
considering the sets where fm >0 and f <0, we obtain
nQ for each NGN,
a set M with m(M) <a and n0 >N with Jm f >-.
2 no

Step 1. We show that there exists a squence of pairwise

disjoint sets
Mv and an increasing squence of positive integers n such that
v

We start by choosing a proper subset Z x of E and m e N such that

fm > ?We observe that there exists a>o sufficiently small so that