ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO Séria I: PRACE MATEMATYCZNE XXIV (1984)
J. Szelmeczka (Lublin)
On weak sequential compactness of generalized Orlicz spaces
In [3] there are given necessary and sufficient conditions for relative а(Ьф, Lr )-compactness of a subset of an Orlicz space Ьф. Applying the method of the proof of this theorem and making use of a criterion of weak compactness of a set in (see [2]), we shall prove a theorem of this type for generalized Orlicz spaces L0 generated by convex functions depending on a parameter.
Let ju be a finite atomless measure in an abstract set Q. A function Ф: Q x R + -+ R + — [0, oo] is called a convex (p-function, depending on a parameter f eO if Ф(г, и) is measurable with respect t in Q for every и ^ 0, and is a convex ^-function of и ^ 0 for every t e Q and Ф(г, u)/u ->0 as и -► 0 +, Ф(£, u)/u -> oo as и -> oo for a.e. te Q .
Let Ф be for every t e Q the function complementary in the sense of Young to the function Ф, with respect to u. Then, considering in the Young inequality u-v < Ф(г, u)+ u) the case of equality and writing tpx = \tp, we have
u 'tp (t,u ) = Ф(£, u)+ 4*(t, (p{t, u)) = 2u(px (t , m), where tp and ф are such that
U V
Ф(Г, m) = j (p(t, s)ds, W(t,v) = ^ ( t , s)ds
о о
and
№ (t, и) = Ф ^ м ) = $q>1(t,s)ds.
0
The function *Р15 complementary to Ф1 in the sense of Young satisfies the equality (f, v) = ^ ( t , 2v).
We have
2utp1 (t, u) = 0 (t,u)+ W (t, (p{t,u)), and so
utp(t,u) = Ф (t , u) + W(t, (p (t, u)) = 2u(p1 (t , и)
^ 2Ф(г,и) + 2Ф(Г, (p1{t,u)).
Hence
W(t, ( p ( t,u ))-2 ¥ (t, ( P i ( t , u ) ) ^ Ф(г, и), and we get
(*)
for и ^ 0, t e Q . Now, we show that the condition ] w ( t , 4 v0{t))dp < 00
Si
implies f Ф(г, ф (t, 2v0(t)))dp < со.
h
This follows from inequality (2) from [1], which may be applied to functions depending on a parameter t for every t, separately:
Ф(1,ф(и)) < ¥(t,2u).
Taking in this inequality и = 2v0(t) and integrating over Q, we obtain the required condition.
The purpose of this paper is to prove the following
Th e o r e m. Let Ф and ¥ be convex cp-functions depending on the parameter t e Q mutually complementary in the sense of Young. Let p be a finite, atomless measure on Q. Moreover, let us assume the following condition: there exists a non
negative, integrable over Q function h(t) such that
(**) V V Э V ¥ (t, 2v) + h(t) > R ¥ { t , v), R> 0 teS2 uqU) >0p >Vÿ(t)
where v0(t) is a measurable function such that J ¥ ( t, 4v0{t))dp < 00.
S2
Then we have the following equivalence :
A set A 6 L0 is bounded in the sense of the norm of L0 and (1) sup \( f - g ) d p -> 0 as p ( E ) -+ 0 for every g e L p
feA E if and only if
(2) there exists a number A0 > 0 such that the family offunctions {#(t, A0 1/|) : f e A ] is a weakly sequentially compact subset of L 1.
P ro o f. First, we shall prove the sufficiency, i.e. that (2)=>(1).
Let us write В — (Ф(г, A0 1/|): f e A ) . The set В <=: L x is a weakly compact subset of Lj if lim \gdp = 0 uniformly with respect to g e B (see [2],
H ( E ) ^ 0e
Theorem 8.11 (corollary), p. 294) hence condition (2) says that lim |Ф (г, Я0 \f\)dg = 0 uniformly with respect to f e A . ц(Е) -*0 e
(i) We first prove that A is bounded in Ьф. By the assumption, we have (3) V 3 V V ( ^ ( £ ) < <5(e) =*> |Ф ( ( , Я0 |/|) 4 /1 < e).
г > О «5(e) > O f e A E £
Let us take e= 1 and the respective <5j = <5(1). Since the measure g. is atomless, there exist measurable and pairwise disjoint sets E u E 2, ..., EnQ such that Q = E x и E2u . . .u E„0 and д(£;) < <5j for i = 1, 2 ,..., n0. Then
f &(t, Ao\f\)dfi < 1 for every f e A , i = 1, 2 ,..., n0.
Ei Hence
Ф(1, X0\ f i ) d a < ^ U ( t , l „ | / | ) ^ < — •«„ = l
J « 0 J « 0
a q
for every f e A . Thus ||/ ||ф < и0/Я0 for every f e A , and this means the boundedness of A in Ьф.
(ii) Now, we are going to prove that
su p \\f-g \ dg, -* 0, as fi(E)-+0 for every g e L F.
feA E
Applying the Young inequality, we obtain
A J |/ *0№ < J W , X0\f\)dg + jV ^ t, y l g l 'j d g
E E E
for every Я > 0 and f e A . Let us choose Я > 0 so small that )dg < oo;
Aq
Q
such а Я exists, because g e L y . By absolute continuity of the integral as a set function, for every e > 0 there exists а <5'(г) > 0 such that if g(E) < <5'(e), then
Thus
4> L 7 -I0I
Aq dg < jXe.
\ f - g \ &(t, A0 | / l № + lfi
E E
for every f e A .
10 — Prace Matematyczne 24.2
But applying (3) and taking g(E) < 0 (^Х-е), we have J<P(r,A0| / № < i àe for every f e A .
E
Hence, writing 6 = min (b'(г), <5 (4Яе)) (<S does not depend on / ) , we obtain 1 Xe
( \ f - g \ d g < ~ - ---for g(E) < Ô and every f e A .
к A 2
Thus
s u p j j f- g jd g < £ for ju(E) < 3.
f e A E
Hence
sup Jl/'gfl dg -*■ 0 as g(E)-> 0 for every g e L y .
f e A E
This concludes the proof of the implication (2)=>(1).
Now, we proceed to the proof of necessity, i.e. (1) =>(2). We may suppose without loss of generality that дф( / ) < 1 for f e A . We indirectly prove that В is a weakly sequentially compact subset of Lb i.e. that lim J Ф(Г, X0\f\)dg = О
ц(Е ) - * Ое
uniformly with respect to f e A : this means that (3) holds for some 0 < A() ^ 1.
So, let us suppose that the family [4>(t, X0\f\): f e A } is for all 0 < A0 < 1 not uniformly absolutely continuous. Thus, writing g(t) = <P(t, X0 \ f (t)\) for f e A , the relation sup J g (t) dg -> 0 as g ( E ) -> 0 does not hold. In other words,
д е в E
(4) 3 V 3 3 {/i(E) < 0 (e) and \gdg ^ e}.
e > O^(fi) > 0 geB E £
We shall show that
(5) lim sup I (g(t) = u„(t))+ dg > e0 > 0
n — 'M geB Q
for every sequence of integrable functions 0 un(t) t oo.
Let us suppose that if (5) is not true, then there is a sequence of measurable functions such that lim sup f \g(t) — un(t)\ + dg — 0, and so there
n--> oo geB n
exists a sequence 0 ^ u „ ( f ) t o o for which sup §(g(t) — un(t))+ dg -* 0, i.e.
веВ Q
V \(g(t) — u,,(t))+dg < rj for n > N^, where Nn does not depend on geB . Let v>o‘n
us write h„(t) = (g(t) — u„(t))+ : then p i ^ d g < g, n > N n uniformly with b
respect to g e B . But
h i \ = if >
nU (0 if g (t) < un(t), whence
r} > j h n(t)dg ^ j h n(t)d g = j g ( t ) d g - J u„(t)dg.
Q E E n { t : g(t)>u„(t)} £ n { f: g ( t ) > u n(t)}
However, we have for t e E n {t : g(t) > u„ (r) j = G„, g (t) > un (t), and the right- hand side of the last inequality is non-negative. Hence
I j' g(t)dg — I' un(t)dg\ < ц for n >
Gn Gn
and for an arbitrary measurable set E a Q. Let us denote by Fn(E) the expression under the sign of absolute value on the left-hand side of the last inequality. Then
.1 'gdg = f gdg+ j gdg ^ |' u„(t)dg +F„(E) + |' u„{t)dg
E E —G n G n E — Gn G n
= iu„(t)dg + Ffl(E).
E
Let us take any rj > 0 and a respective N n such that \Fn(E)\ < ц for n > Nq. Let us denote by nn the least index n such that n > N„. Then \Fn(E)\
< ц for every E. Fix nn\ then
\gdg < j u„(t)d/i + r].
E E
From the integrability of un(t) it follows that is an absolutely
E
continuous set function. Hence there exists a ô(rj) > 0 such that if g(E) < ô (rj), then \u„ (t)d(.i < //. Thus, if //(£) < <5(/у), then igdg < 2g for every g
e n E
(uniformly with respect to g). Consequently, we obtain a contradiction to (4). So we have proved condition (5).
Let a + — a for a ^ 0, a + = 0 for a * 0. Since u„(t) Î we get (g(t) — -u„{t))+ > (g{t)-u„+1{t))+, and hence
Thus
implies
sup $(g(t )-u„(t))+ dg > sup $ ( g ( t ) - u n{ t ) y dg.
geB q geB n
lim sup §(g(t) — un(t))+ dg > e0 > 0
n - » » geB Q
s u p $ ( g ( t) - u n(t))+ dg > e0 > 0 for every n.
geB Q
Hence for each n there exists an f ne A such that
(7) ]’(ф(1,/1()|/,( г ) |) - и >1(г))+ ^ > £() > 0 for n = 1, 2 ,...
О
Taking R = 2n+1 in assumption (**), we may find measurable functions vn(t) > 0 such that F(t, 2v) + h(t) > 2n+1 F (t,v) for v ^ v„(t). Let us write
E„ = {t: Ф(г,Яо |/«(01) > un(t)},
g n ( t)
=
(pi[
lя0|/„(
г)|x£n(r)] =
( P i ( t , ù o \ f n ( t ) \ ) x E„ ( t)•
Moreover, let us choose, by definition, integrable functions u„(t) ^ 0 such that u„(t)îoo, un(t) ^ 2", un(t) ^ <p(t, ij/(2vn(t)))- We have
Ё?ф(Л<0 l/J) < M / J ) ^ 1*
Hence
1 ^ $<P(t,À0\fn\)dn ^ j u„(t)dfi > 2"(£„),
n E n
and so /i( £ J < ?n; thus fi(En)-+ 0 as n-+oo.
Taking и = À0\fn{t)\ and t e E n, we have, by inequality (*), that
2j « 4>M„|/„WI) Ф(М/,(01).
On the other hand, we apply the inequalities appearing in the definition of u„(t). Then, taking t e E n, we obtain
[M ol/„(O I] > w„(0 ^ Ф ^ М ^ О ) ] ; hence
'U/.WI г ^ М м о ) .
Since (p is a non-decreasing function, we get
<P {t , Яо |/„(01} ^ <P {Ф(*> 2u„(0)} = 2u„(0>
0
ЛО =
<Pi(f, 2о|/я(01) = (f, Ао1Л(01) ^
Hence
lf/^ ' ^ 2”+1 > 2”+ 2,
for t e E n. Consequently,
2 " Ч > и ,д М < P (t,\m \) + h«) for t e E n.
We integrate both sides of this inequality over Q:
2"
Qy(dn)= 2” j Ф(г,0„(г))й^ ^ j Ф(г, |/„(0I)^+j/z(t)^
E„ E n Q
< ^ ф Ш + f h(t)du ^ 1 + J h{t)dn = C < 00,
h «
since the function #„(f) is equal to zero outside the set En. Hence Qy(gn) ^ 2~ " C for n = 1 ,2 ,...
Let us write g(t) = sup g n(t); then
Hence
g(t) = lim sup g^t).
n~+ao i < n
Y (t, g (0) < Urn sup W (t, gt (r)), t e Q,
n -* ao i < n
and so, by the Fatou lemma, we obtain
Qvi9) = I 4 ' ( t , g ( t ) ) d f i < lim js u p W(t, g ^ d g
П n->ao Q i ^ n
= lim Qy(sup(gu g2, . . . , g n))
n~*-ao
^ Qr (91) + Qv (9 2) ^ ■ •• ^ X 2 n ‘ C — С < 00 . n= 1
Thus 9 e bp. Applying the case of equality in the Young inequality, we get M-0\fH{t)xEH(t)\gH(t) = Ф(г,Яо|/„(01Х£„(0)+^(^520[„(0)-
Hence
2A0 f \fn(t)-9n(t)\dg
= j < P ( t a 0 \ f n ( t ) \ ) d g +
J
¥ ( t , 2 g n( t) ) d g > f <P(t,A0 \fn(t)\)dgEn E n E n
= J [<P(t, À0 \ fn( t ) \ ) - u „ ( t ) ] d g + J u n( t )d g
En En
> j [& (t,*o\fn(t)\)-u„(t)]dg
E n
= Ao|/„(t)|)-M„(f)]+ dg > e0 > 0 (see (7)).
x
But g (t ) = sup g n(t), and so
r, (8)
which yields
2Л-0 j If n'9\dg > 2X0 J \fn-gn\dg > e0 > 0,
E n E„
S\fn '9 \d g > -Hjy- > ° for и = 1 ,2 ,...
E n
However, we have g e b y . Moreover, g(En)-> 0. Since we assume (1), i.e.
sup J | / -g\dg -*0, as g(E)-> 0 for every g e L y ,
f e A E
sup j \f-g[dp->0,
f e A E „
we obtain
and since / пе Л , we get
J \fn'9\dn sup j \f-g\dfi^>0, as и ^ о о ,
E n f e A E„
a contradiction to (8). This proves the theorem completely.
Let us remark that assumption (**) in the theorem is needed only in the proof of necessity, i.e. (1)=>(2).
References
[1] R. B o ja n ic , J. M u s ie la k , An inequality fo r functions with derivatives in an Orlicz space, Proc.
Amer. Math. Soc. 15 (1964), 902-906.
[2] M. D u n fo r d , J. T. S c h w a r tz , Linear operators, Part I, New York, London 1958.
[3] W. A. L u x e m b u r g , A remark on a theorem o f W. Orlicz on weak convergence, Comment.
Math., Tomus specialis in honorem Ladislai Orlicz I (1978), 211-220.