ANNALES SOCIETATIS MATHEMATICAE POLONAE Series I: COMMENTATIONES MATHEMATICAE XXI (1979) ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO
Séria I: PRACE MATEMATYCZNE XXI (1979)
A. Kaminska and H. Hudzik (Poznan)
Some remarks on convergence in Orlicz space
1. Abstract. A Kozek in papers [4], [5] investigated Orlicz spaces L0 of functions defined on an abstract set T with values in a Banach space X , generated by an iV-function Ф:
X x T -+ [0, oo].
In this paper we shall prove that the d 2-condition is sufficient for equivalence of modular convergence and norm convergence in L0 . We prove also that if Ф{-, t) is continuous, ц is an atomless measure and X is separable, then the d 2-condition is necessary for equivalence of both above convergences. Moreover, we shall consider the relation between modular conver
gence and convergence in measure ц.
We are indebted to Professor J. Musielak for his kind remarks in course of preparing of this paper.
2. Introduction. Let (T, Q, f i ) be a measure space, where T is an abstract set, g is a tr-algebra of subsets of the set T, ц is a tr-finite, positive, complete measure on Q. X is a Banach space.
2.1. Definition (see [4], [5]). A function Ф: X x T ^ > [0, oo] is called an N -function if it satisfies the following conditions:
There exists a set Txe Q, /2(7^7^) = 0, such that
(a) Ф(-, •) is M x Q - measurable, where $ denotes the er-algebra of Borel subsets of X;
(b) Ф (•, t) is lower semicontinuous on X for every t e Tx ; (c) Ф (•, t) is convex for every t e Tx ;
(d) Ф(0, t) = 0 and Ф (х ,0 = Ф( — x , t ) for every x e X , teT i;
(e) there exist ^-measurable functions a: (0, oo) and k: Tx -> (0, oo) such that \\x\\ ^ 2(r) implies Ф(х,г) ^ a (t) for every teT^;
(f) there exist /i-measurable functions q: Tx- > (0, oo) and 1: 7] - > (0, oo) such that \\x\\ ^ g(t) implies Ф{х, {) ^ ± (t) for every t e l i.
If Ф fulfils conditions (a)-(d), then Ф is called an N"-function; if it satisfies conditions (a)-(e), it is called an N ’-function.
R em ark. If Ф is an ^''-function and satisfies condition (e), then
— Prace Matematyczne 21.1
lim <2>(x,0= oo for every t e T x\ if it satisfies condition (f), then Ф(-, t) IWI-=o
is continuous at zero for every te 7 i .
Let M x be a set of all /i-measurable functions defined on T with values in X. The measurability of functions is understood as in [2 ].
By the Orlicz space L0 we mean the set of all functions х(-)е. Жх for which there exists a constant к > 0 such that
1ф(кх) = J <P(kx(t), t)dg < oo.
T
The functional I0 (x) defined on L0 is modular; L0 is a modular space (see [7], [8]).
The space L0 with Luxemburg norm is a Banach space. For other properties of Orlicz spaces L0 see [4], [5].
We denote by dom I0 the set I0 (x) < oo}.
2.2. Definition. We say that a sequence of functions x„(-)gL0is modular convergent to x ( - ) e L0 if there exists a constant к > 0 such that
lim I0 (k(xn — x)) = 0.
П -*OD
Norm convergence of a sequence of functions x„(-)eL0 to x{ - ) eL0 is equivalent to
lim I0 (l(xn — x)) = 0 for all l > 0.
П -*■ 00
The function Ф satisfies the A2-condition if
Ф(2х, t) < КФ( х , t) + h(t)
for all x e X and t e Tx, where К is some positive number and h (•) is an integrable non-negative function.
In Orlicz spaces L0 generated by N -function Ф: X -* R + without para
meter t, the following d 2-conditions are applied (see [6]):
(d'2) 3 V Ф(2x) ^ КФ(х) if f i ( T ) = oo
K > 0 x e X
and for X = R l
(d2,vq) V V 3 Ф(Ъ>) s: КФ(с) if M(T) < 00.
К>0 up >0 v q
It is known that if g is atomless, then d 2-conditions are sufficient and necessary for equivalence of modular convergence and norm convergence in the case g T = oo and g T < oo, respectively. Obviously, the class of JV-func
tions with parameter t contains the class of jV-functions without parameter.
Indeed, for any N -function Ф without parameter we can put Ф(х,г)
= р(г)Ф(х), where p(t) = 1.
If an N -function Ф without parameter satisfies (d2)-condition, then it satisfies also d 2-condition with h(t) = 0.
Convergence in Orlicz space 83
If p T < oo, X = R 1, and Ф satisfies (d2,^-condition, then for every v e R 1 we have <P(2v) ^ K<P(v)+ K<P(v0). Since p T < oo, jK<P(v0)d/j. < oo
7 so Ф satisfies d 2-condition with h(t) = КФ(и0).
Now, we note that there exists iV-function Ф{х,Ь) with parameter t which satisfies d 2-condition with h(t) ф 0 and it does not satisfy /^-condi
tion with h(t) = 0. For example, let us put X = R 1 and let Ф(р,/)
= p(t) Ф1 (v), where Фх is N -function satisfying (/12 „^-condition, where vQ > 0 is the smalest number v such that Ф1{2и) ^ К Ф1(ю). Further, let p(t ) ф 0 and §p(t)dp < oo. Then for every u e [0 , oo) we have
T
Ф(2v, t) ^ КФ(с, г) + КФ(р0, t).
Since J КФ(и0, t)dp = КФ1 (и0) j p(t)dp < oo, the N-function Ф satisfies
7 T
d 2-condition with h(t) — КФ(и0, t). But Ф^т0) > 0 and p(t) ф 0, so /^-con
dition cannot be fulfilled with h(t) = 0.
There exist N -functions Ф(х,г) with parameter t such that p T = oo and |Ф (х,г)^/7 < oo for every x e X and such that for p T < oo,
T
j Ф(х, t) dp = oo for x Ф 0. It suffices to take X = R 1, T = Rn, p -Lebesgue
t " _ 2 " 1
measure, Ф(х, t) = \ \ e tj and Ф(х, 0 —
П
T T where tj arej=i j=i N
the components of t = (fl 5 tn) and Ф1 is an N-function without parameter.
The condition J Ф(х, t) dp < oo for every fixed x e X and pA < oo is often
A
used. It means that simple functions belong to Orlicz spaces L0 ([7], see also condition A in [4]).
3. Results.
Proposition 3.1. Let us assume that p T < oo, Ф is an N"-function and Ф fulfils the condition
(e') V 3 IMI > ô => Ф(х, t) > a(t),
<5 > 0 a(-)
where a(-) is a measurable function such that a(f)e(0, oo) for every t e T r . Then 1Ф (x„) -> 0 implies x„ -*■ 0 in measure p.
P roof. Let x „ ()e # x and / ф(х „ )-» 0 . Assume, for a contrary, that *„(•) does not converge in measure to zero. Then there exist e0 > 0 and rj0 > 0 such that for every N there exists nN > N such that p { t e T: \\x„N(t)\\
> £o} > I/o- Let
An = {t e T : \\x„Nm > e0}.
Hence pAN > rj0 for every N. We are finding a set B œ. T and a constant К > 0 such that p(T\B ) < and Ф(х,г) > К for ||x|| > e0 and t e B .
In fact, let Bn = {t e T : a (t) > 1/n}, where a( ) is the function from condi- 00
tion (e') with 3 = e0. We have Bn a Bn + 1, (J Bn = T for every t e T t and
n = 1
lim pBn = pT. Since p T < oo, so lim p{T \B n) = 0. Consequently, there exists
M -*■ 00 П—► 00
a set Bno satisfying the inequality p(T\B „Q) < i ^ 0-
Let В = B„0 and К = 1 /п0. If t e B , then a (t) > К and from condition (e') we obtain Ф(х, t) > ot(t) > К for ||x|| > e0. Thus, for every N, we have (As r \B ) > jriо for all N, and
1ф (*nN) = f ф {*nN (t) , t ) d n > f Ф (x„N(t), t) dn
T i4jynB
^ X/i {An n B ) ^ К ~ ~ > 0 f°r every N . The above contradiction completes the proof of the proposition.
Proposition 3.2. Let Ф be an N "-function continuous in X , finite a.e.
on T and let X be a separable Banach space. I f for every sequence {*„(•)} ®=i satisfying 1ф(х„ (•)) 0 we have *;„(•)-►() in measure p, then the function Ф satisfies the condition
(e") V V (||x||
^ 3 => Ф{х, t) >0)
for every t e T t . 5 > 0 x e XP ro o f. Let Ф does not satisfy condition (e")- We shall find a sequence (•)}*= l modularly convergent to zero, which does not converge in measure to zero.
Let L = U { t e T: Ф{х, t) = 0} and Lm — U { t e T : Ф(х, t) = 0 } ,
ХФ0 IWlèl/m
00
where n = 1 , 2 , . . . We have L = y Lm. Let be a dense subset of
m= 1 oo
the set Am = { x e X : ||x|| ^ 1/mj and let Bnm = [j { t e T : Ф(х”, t) ^ 1 /«}.
k=
1
oo
Now we will prove the following inclusion Lm a f) Bnm for every
1 ^ n= 1
m = 1 , 2 , . . .
Let then m be an arbitrary index and t e L m. Then there exists an element x 0 of the set Am such that <£(x0,t) = 0. Since {х^}^°=1 is dense in the set Am, so there exists a subsequence {х*.},*! of the sequence {хк}Г= 1 such that ||x0 — Xjt*.|j 0. Hence and from continuity of Ф we have Ф(х£, t) 0. Thus Ф(х**., t) < 1/n for some index k{. Hence t e B nm.
00
If we write Bm = f | Bnm, then the inclusion Lm <= Bm is satisfied.
n=1
The set L cannot be of measure zero, because condition (e") is not fulfilled. Hence the sets Lm (for n > N 0 and some N 0) are not of measure
Convergence in Orlicz space 85
zero either. The sets Bm are measurable, moreover, they include Lm, so цВт > 0 for n > N 0. We can find a measurable set. C such that C c: BmQ, where m0 > N 0 and 0 < fxC < oo, by the cr-finitenesg of measure ц.
Writing a = 1 /m0 we define the following sequence of the set-valued functions
This familly of multifunctions has the following properties:
1° N n{t) is a closed set in X for every n = 1 , 2 , . . . , for all t e T x because Ф is lower ’semicontinous,
2° N n(t) Ф 0 for all t e C , n = 1 , 2 ,... For, let t e C c= BmQ. From definition of the set BmQ it follows that for every index n there exists an element x™ne X such that Ф(х*” , t) ^ 1 /n and ||х™и|| ^ a. Hence x™neN„(t).
3° The graph of N„ (we denote it by Gr (N n)) is J x Q-measurable, because
Gr (Nn) = {(x, t ) e X x T: x e N„ { t )}
= {(x, t ) e X x T : ||x|| ^ а л Ф{х, t) ^ 1 /п}
= {(x, t ) s X x T: Ф{х, t) ^ 1/n) n { { x e l : ||x|| ^ a } x T ] e M x Q since Ф(-,-) is 3 x Q-measurable.
Now, in view of Theorem 5.1, 5.2 in [3], we obtain the existence of the measurable functions y„(-) such that yn{ t ) e Nn{t) for t e C and for all n = 1 , 2 ,... Let
The functions x„(-) are measurable and ||x„(t)l! ^ a for all fe C , n = 1 ,2 ,...
Moreover,
So we have a sequence of functions which modularly converges to zero, but does not converge to zero in measure. This ends the proof.
R em arks. Г Condition (e") is equivalent to the following condition:
2° If X = R 1, then conditions (e') and (e") are equivalent. (e')=>(e") is obvious. Let now (e") be satisfied and let Ô > 0 be given. If |x| > <5, then Ф(х,г) = Ф(|x |,t) ^ Ф(<5, г) for every teT j. Hence condition (e') is fulfilled for а(-) = Ф(<5,.).
3° If fiT < o o , Ф is an ЛГ-function, then norm convergence in the N„(t) = { x e l : |x|| ^ а л Ф (х Д ) ^ 1/n}.
Ф (х,г) = 0<=>x = 0 for every t s T x.
Orlicz space L0 implies convergence in measure. This fact follows from the proof of Theorem 2.4 in [4].
4° Norm convergence in Orlicz space L0 , where the measure p is infinite and Ф is an N- function, does not imply convergence in measure.
The following simple example was given by Châtelain in [1]. X = R 1, T = R+ = [0 , oo ), p-Lebesgue measure, Ф(х, t) = x 2/t. Obviously, Ф is an iV-function with parameter. Let
jyJTjn if t e [ n + \ , n + 2), I 0 otherwise.
Then x„(-)eL0 . Let К ■> 0 be any positive number. Then n + 2 jç2
1Ф(К хп) = J Ф( Кхп(^, t)dp = J --- dt = - — 0.
Г n+l n n
Hence ||x„(-)||l* -*■ 0. But p { t e T : \xn(t)\ > 1} = 1 for n = 1 ,2 ,...; therefore x„( ) does not converge in measure.
Th e o r e m 3.3. I. I f Ф is an N"-function, Ф is continuous at zero, Ф satis
fies (q') and Ф fulfils condition A2, then norm convergence and modular convergence are equivalent on L0 .
II. If Ф is an N''-function continuous at zero, X is separable, p, is an atomless measure and if norm convergence and modular convergence on L0 are equivalent, then Ф fulfils condition A2.
P roof. I. By definition, norm convergence implies modular convergence.
1° First we assume that p T < oo. Let хп(-)еЬф and / ф(х„)->>0. To establish the theorem it is enough to prove that / ф(2х„)->0.
Let Tn — [ t e T : Ф(х„(г), t) ^ h(t)}; then
1Ф{2хп) = {Ф (2xn{ t ) , t ) d p + J Ф(2xn(t), t)dp
Tn T \T n
^ ( K + 1) J Ф(х„(0, t ) d p + f ф (2xn(t), t)dp = J ln + J 2n-
T J n Tn
We have J l n~* 0 by assumption. Since / ф(х„)->0, so, by Proposition 3.1, x„(-)-> 0 in measure. Let us take any subsequence x (•) c xn(.). By the Riesz theorem, we can find a subsequence x„k (•) of the sequence x„fc(•) of functions which converge to zero almost everywhere. From the continuity of Ф at zero we have Ф(2x„k (t),t)^>0 for a.e. t e T . Moreover, if teT„, then УФ(2х„(0, t) ^ ( K + l)h(t). Hence J 2n -> 0, by the Lebesgue dominated
n ki
convergence theorem. Therefore, J 2n -+ 0.
Convergence in Orlicz space 87
2° Let fiT = оо and let e > 0 be given. Then for A <= T 1ф(2хн) = j<P(2xH(t),t)dn = S<P(2xn{ t ) , t ) dn+ J <P(2x„{t),t)dfi
T A T~,A
{Ф(2х„(0, t)dn + K
J
Ф(х„(*М<*/Н-f
h(t)d/i = J ln + J 2n + J 3.A T \A T \A
We choose a set Л in such a manner that J h(t)dpi < e and fxA < oo;
1 \A
then J 3 < e. By assumption we have J 2n-+ 0. Since /1,4 < oo we apply part 1° of this proof and we get J ln < e for n > N. Therefore, / ф(2х;„)-> 0.
II. To prove the theorem it is enough to show that dom / ф1 <= dom I02, where Ф1(х,г) = Ф{х, г) and Ф2(х,г) = Ф(2x, t ) (see Theorem 1.7 in [4];
A. Kozek established in that paper that under our assumptions the inclusion dom / ф1 c dom I02 implies condition A2).
Let us suppose that norm convergence is equivalent to modular con
vergence, but there exists х ( ) е . Ж х such that x ( ) e d o m /01 and x(-)^dom i02.
Г There exists a set A с T such that цА > 0 and Ф(2x{t), t) = go
for all t e A . Since /i is atomless, there exists a sequence of sets Ane Q such that An a A, /u(A„) > 0 and /o4„->0. Let x„( ) = x(-)z^„(-)- We have / ф (*„) 0 but l 02 (2x„) = go, a contradiction.
2° Ф (2х(0, t) < oo for a.e. t e T .
(a) ц Т < oo. Let Ak = { t e T : Ф(2x(f), t) < к}; then A 1 с A2 c ..., and 00
U Ak differs from T at most by a set of measure zero. Since /tT < oo, к = 1
we have lim /i(T \A k) = 0. Now, let *„(■) = х(-)Хги„(-); then / ф(х„)-> 0, but 1ф(2хп) =
J
Ф(2x{t), t)d/j,— j Ф (2х(0, t)dfi = ooT An
for n = 1 , 2 , . . . , because j Ф(2x(t), t ) df i = oo and j Ф(2x(t), t)dpi ^ пцА„ < oo,
a contradiction. T A
(b) fiT = oo. We can suppose that |Ф (2 x( t ) , t ) dn < oo for all A e Q ,
A
fiA < co. Let £„->0. Since 1ф(х) < со, so there exist sets An such that
M „ < GO and
j
< p (x (t), t )dji < e„. Let x„() = х { )х т\ап(У , then /ф(х„) < s„.т\лп
Therefore, / ф(х„)->0. But
/ Ф(2х„) =
J
Ф(2х (t), t) dfi— f Ф(2 x (0 , t)dfi =T \
This contradiction completes the proof.
0 0 .
References
[1] J. C h â te la in , These, 1974-1975, Université de Perpignan, Université des Sciences et Techniques du Languedoc.
[2] N. D u n fo rd and J. T. S ch w arz, Linear operators, Part I, New York 1958.
[3] C. J. H im m e lb e rg , Measurable relations, Fund. Math. 87 (1975), p. 53-72.
[4] A. K ozek, Orlicz spaces of functions with values in Banach spaces, Comm. Math. 19 (1976), p. 259-288.
[5] — Convex integral functionals on Orlicz spaces, preprint, May 1976.
[6] M. A. K r a s n o s e l’skiT, Ja. B. R u tic k iî, Convex functions and Orlicz spaces, Gosud.
Izdat. Fiz.-Mat. Literat., Moskva 1958 (in Russian).
[7] J. M u sie la k and W. O rlic z , On modular spaces, Studia Math. 18 (1959), p. 49-65.
[8] H. N a k a n o , Generalized modular spaces, ibidem 31 (1968), p. 439-449.