Você está na página 1de 30

Journal of Functional Analysis 202 (2003) 247276

Multiplicator space and complemented subspaces


of rearrangement invariant space
$
S.V. Astashkin,
a
L. Maligranda,
b,
* and E.M. Semenov
c
a
Department of Mathematics, Samara State University, Akad. Pavlova 1, 443011 Samara, Russia
b
Department of Mathematics, Lulea University of Technology, 971 87 Lulea, Sweden
c
Department of Mathematics, Voronezh State University, Universitetskaya pl.1, 394693 Voronezh, Russia
Received 20 May 2002; revised 23 August 2002; accepted 7 October 2002
Communicated by G. Pisier
Abstract
We show that the multiplicator space MX of an rearrangement invariant (r.i.) space X on
0; 1 and the nice part N
0
X of X; that is, the set of all aAX for which the subspaces generated
by sequences of dilations and translations of a are uniformly complemented, coincide when the
space X is separable. In the general case, the nice part is larger than the multiplicator space.
Several examples of descriptions of MX and N
0
X for concrete X are presented.
r 2002 Elsevier Inc. All rights reserved.
MSC: 46E30; 46B42; 46B70
Keywords: Rearrangement invariant spaces; Multiplicator spaces; Lorentz spaces; Orlicz spaces;
Marcinkiewicz spaces; Subspaces; Complemented subspaces; Projections
0. Introduction
For rearrangement invariant (r.i.) function space X on I 0; 1; we will consider
the multiplicator space MX and the nice part N
0
X of the space X: The space
ARTICLE IN PRESS
$
Research supported by a grant from the Royal Swedish Academy of Sciences for cooperation between
Sweden and the former Soviet Union (Project 35147). The second author was also supported in part by the
Swedish Natural Science Research Council (NFR)-Grant M5105-20005228/2000 and the third author by
the Russian Fund of Fundamental Research (RFFI)-Grant 02-01-00146 and the Universities of Russia
Fund (UR)-Grant 04.01.051.
*Corresponding author.
E-mail addresses: astashkn@ssu.samara.ru (S.V. Astashkin), lech@sm.luth.se (L. Maligranda),
root@func.vsu.ru (E.M. Semenov).
URL: http://www.sm.luth.se/~lech/.
0022-1236/03/$ - see front matter r 2002 Elsevier Inc. All rights reserved.
doi:10.1016/S0022-1236(02)00094-0
MX is connected with the tensor product of two functions xsyt; s; tA0; 1; and
N
0
X is the space given by uniformly bounded sequence in X of projections into
Q
a;n
generated by the dilations and translations of the non-zero, decreasing function
aAX on dyadic intervals
k1
2
n
;
k
2
n
in I; k 1; 2; y; 2
n
; n 0; 1; 2; y . These
functions are given by
a
n;k
t
a2
n
t k 1 if tA
k1
2
n
;
k
2
n
;
0 elsewhere:
_
The spaces MX and N
0
X coincide when X is a separable space but in the non-
separable case the nice part can be larger than the multiplicator space. Such a
description is helpful in the proofs of properties of N
0
X and it motivates us to
investigate more the multiplicator space MX: We will describe MX for concrete
r.i. spaces X as Lorentz, Orlicz and Marcinkiewicz spaces. Suitable results on N
0
X;
especially when X is a Marcinkiewicz space M
j
; are given.
The paper is organized as follows. In Section 1 we collect some necessary
denitions and notations.
Section 2 contains results on the multiplicator space MX of a r.i. space X on
0; 1: At rst we collect its properties. After that the multiplicator space MX is
described for concrete spaces like Lorentz L
p;j
spaces, Orlicz L
F
spaces and
Marcinkiewicz M
j
spaces. The main result here is Theorem 1 which gives necessary
and sufcient condition for the tensor product operator to be bounded between
Marcinkiewicz spaces M
j
:
In Section 3, we consider a subspace N
0
X of X generated by dilations and
translations in r.i. space on 0; 1 of a decreasing function from X: The main result of
the paper is Theorem 2 showing that the multiplicator space MX is a subset of the
nice part N
0
X of X and that they are equal when a space X is separable. In the
general case, the nice part is larger than the multiplicator space (cf. Example 2). Here
we apply results on multiplicators from Section 2 to the description of N
0
X:
Special attention is taken about N
0
X when X is a Marcinkiewicz space M
j
(see
Corollary 5 and Theorem 3). Stability properties of the class N
0
with respect to the
complex and real interpolation methods are presented. There is also given, in
Theorem 7, a characterization of L
p
-spaces among the r.i. spaces on 0; 1; which is
saying that r.i. space X on 0; 1 coincides with L
p
0; 1 for some 1pppNif and only
if X and its associated space X
0
belong to the class N
0
:
Finally, in Section 4 we show that, in general, you cannot compare the results on
the interval 0; 1 with the results on 0; N and vice versa.
1. Denitions and notations
We rst recall some basic denitions. A Banach function space X on I 0; 1 is
said to be a rearrangement invariant (r.i.) space provided x
n
tpy
n
t for every
tA0; 1 and yAX imply xAX and jjxjj
X
pjjyjj
X
; where x
n
denotes the decreasing
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 248
rearrangement of jxj: Always we have imbeddings L
N
0; 1CXCL
1
0; 1: By X
0
we
will denote the closure of L
N
0; 1 in X:
An r.i. space X with a norm jj jj
X
has the Fatou property if for any increasing
positive sequence x
n
in X with sup
n
jjx
n
jj
X
oN we have that sup
n
x
n
AX and
jjsup
n
x
n
jj
X
sup
n
jjx
n
jj
X
:
We will assume that the r.i. space X is either separable or it has the Fatou
property. Then, as follows from the Caldero nMityagin theorem [BS,KPS],
the space X is an interpolation space with respect to L
1
and L
N
; i.e., if a linear
operator T is bounded in L
1
and L
N
; then T is bounded in X and
jjTjj
X-X
pC maxjjTjj
L
1
-L
1
; jjTjj
L
N
-L
N
for some CX1:
If w
A
denotes the characteristic function of a measurable set A in I; then clearly
jjw
A
jj
X
depends only on mA: The function j
X
t jjw
A
jj
X
; where mA t; tAI; is
called the fundamental function of X:
For s40; the dilation operator s
s
given by s
s
xt xt=sw
I
t=s; tAI is well
dened in every r.i. space X and jjs
s
jj
X-X
pmax1; s: The classical Boyd indices of
X are dened by (cf. [BS,KPS,LT])
a
X
lim
s-0
lnjjs
s
jj
X-X
ln s
; b
X
lim
s-N
lnjjs
s
jj
X-X
ln s
:
In general, 0pa
X
pb
X
p1: It is easy to see that % j
X
tpjjs
t
jj
X-X
for any t40; where
% j
X
t sup
0oso1;0osto1
j
X
st
j
X
s
:
The associated space X
0
to X is the space of all (classes of) measurable functions
xt such that
_
1
0
jxtytjdtoNfor every yAX endowed with the norm
jjxjj
X
0 sup
_
1
0
jxtytj dt : jjyjj
X
p1
_ _
:
For every r.i. space X the embedding XCX
00
is isometric. If an r.i. space X is
separable, then X
0
X
n
:
Let us recall some classical examples of r.i. spaces. Denote by C the set of
increasing concave functions jt on 0; 1 with j0

j0 0: Each function
jAC generates the Lorentz space L
j
endowed with the norm
jjxjj
L
j

_
1
0
x
n
t djt
and the Marcinkiewicz space M
j
endowed with the norm
jjxjj
M
j
sup
0otp1
1
jt
_
t
0
x
n
s ds:
If F is a positive convex function on 0; N with F0 0; then the Orlicz space
L
F
L
F
0; 1 (cf. [KR,M89]) consists of all measurable functions xt on 0; 1 for
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 249
which the functional jjxjj
L
F
is nite, where
jjxjj
L
F
inf l40 : I
F
x
l
_ _
p1
_ _
with I
F
x :
_
1
0
Fjxtj dt:
An Orlicz space L
F
is separable if and only if the function F satises the D
2
-condition
(i.e. F2upCFu for every uXu
0
and some constants u
0
40 and C40).
The Lorentz space L
p;q
; 1opoN; 1pqpN; is the space generated by the
functionals (quasi-norms)
jjxjj
p;q

_
1
0
t
1=p
x
n
t
q
dt
t
_ _
1=q
if qoN
and
jjxjj
p;N
sup
0oto1
t
1=p
x
n
t:
For 1ppoNand jAC the Lorentz space L
p;j
is the space generated by the norm
jjxjj
p;j

_
1
0
x
n
t
p
djt
_ _
1=p
:
We will use the CalderonLozanovski construction (see [C,M89]). Let X
0
; X
1
be a
pair of r.i. spaces on 0; 1 and rAU (rAU means that rs; t srt=s for s40 with
an increasing, concave function r on 0; N such that r0 0 and r0; t 0: By
rX
0
; X
1
we mean the space of all measurable functions xt on 0; 1 for which
jxtjplrjx
0
tj; jx
1
tj a:e: on 0; 1
for some x
i
AX
i
with jjx
i
jj
X
i
p1; i 0; 1; and with the inmum of these l as the norm
jjxjj
r
: In the case of the power function r
y
s; t s
1y
t
y
with 0pyp1; r
y
X
0
; X
1

is the Caldero n construction X


1y
0
X
y
1
(see [C,LT,M89]). The particular case
X
1=p
L
N

11=p
X
p
for 1opoN is known as the p-convexication of X dened
as X
p
fx is measurable on I : jxj
p
AXg with the norm jjxjj
X
p jjjxj
p
jj
1=p
X
(see
[LT,M89]).
For other general properties of lattices of measurable functions and r.i. spaces we
refer to books [BS,KPS,LT].
2. Multiplicator space of an r.i. space
Let X XI be an r.i. space on I 0; 1: Then the corresponding r.i. space
XI I on I I is the space of measurable functions xs; t on I I such that
x
,
tAXI with the norm jjxjj
XII
jjx
,
jj
XI
; where x
,
denotes the decreasing
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 250
rearrangement of jxj with respect to the Lebesgue measure m
2
on I I: For two
measurable functions x xs; y yt on I 0; 1 we dene the bilinear operator
of the tensor product #by
x#ys; t xsyt; s; tAI:
Denition 1. The multiplicator space MX of an r.i. space X on I 0; 1 is the set
of all measurable functions x xs such that x#yAXI I for arbitrary yAX
with the norm
jjxjj
MX
supfjjx#yjj
XII
: jjyjj
X
p1g: 1
The multiplicator space MX is an r.i. space on 0; 1 because for the product
measure we have
m
2
fs; tAI I : jxsytj4lg
_
1
0
mfsAI : jxsytj4lg dt:
Let us collect some properties of MX: First note that for any measurable set
A in I the functions w
A
#x and s
mA
x are equimeasurable, i.e., their distributions are
equal
d
w
A
#x
l m
2
fs; tAI I : w
A
sjxtj4lg
mftAI : js
mA
xtj4lg d
s
mA
x
l
for all l40: In particular, jjxjj
MX
Xjjx#w
0;1
=j
X
1jj
X
jjxjj
X
=j
X
1 gives the
imbedding
MXCX and jjxjj
X
pj
X
1jjxjj
MX
for xAMX: 2
Moreover, MX X if and only if the operator #: X X-XI I is bounded.
In particular, ML
p;q
L
p;q
for 1opoN and 1pqpp since from the ONeil
theorem (see [O, Theorem 7.4]) the tensor product #: L
p;q
L
p;q
-L
p;q
I I is
bounded.
From the equality w
0;u
#w
0;v

,
t w
0;uv
t we obtain that if XCMX; then
fundamental function j
X
is submultiplicative on 0; 1; i.e., there exists a constant
c40 such that j
X
stpcj
X
sj
X
t for all s; tA0; 1:
Some other properties of the multiplicator space MX (cf. [A97] for the
proofs):
(a)
j
MX
t jjs
t
jj
X-X
; jjs
t
jj
MX-MX
jjs
t
jj
X-X
for 0otp1 and
jjs
1=t
jj
1
X-X
pjjs
t
jj
MX-MX
pjjs
t
jj
X-X
for t41: In particular, a
MX
a
X
and b
MX
pb
X
:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 251
(b) We have imbeddings L
c
CMXCL
p
; where ct jjs
t
jj
X-X
; 0otp1;
p 1=a
X
and the constants of imbeddings are independent of X: In
particular,
b
1
MX L
N
if and only if a
X
0:
b
2
If the operator #: X X-XI I is bounded, then XCL
1=a
X
:
(c) If X is an interpolation space between L
p
and L
p;N
for some 1opoN; then
MX L
p
: In particular, ML
p;q
L
p
for 1opoNand ppqpN:
Note that the operation MX is not monotone, i.e., if X; Y are r.i. spaces on 0; 1
such that XCY then, in general, it is not true that MXCMY: Namely, consider
the r.i. space X on 0; 1 constructed by Shimogaki [S]. This space has Boyd lower
index a
X
0 with j
X
t t
1=2
and L
2
CX: On the other hand, ML
2
L
2
but
MX L
N
by b
1
:
Proposition 1. We have MMX MX with equal norms.
Proof. It is enough to show the imbedding MXCMMX: Let xAMX with
the norm jjxjj
MX
pC: Then
jjx#ujj
XII
pCjjujj
XI
8uAX:
In particular, for u y#z
,
with xed yAMX and any zAX with jjzjj
XI
p1 we
have
jjx#y#z
,
jj
XII
pCjjy#z
,
jj
XI
Cjjy#zjj
XII
:
Since
m
2
fs; tAI I : jxsy#z
,
tj4lg

_
1
0
mftAI : jxsy#z
,
tj4lg ds

_
1
0
m
2
ft; aAI I : jxsytzaj4lg ds
m
3
fs; t; aAI I I : jxsytzaj4lg

_
1
0
m
2
fs; tAI I : jxsytzaj4lg da

_
1
0
mftAI : jx#y
,
tzaj4lg da
m
2
ft; aAI I : x#y
,
tzaj4lg
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 252
for any l40 it follows that
jjx#y#z
,
jj
XII
jjx#y
,
#zjj
XII
:
Taking the supremum over all zAX with jjzjj
XI
p1 we obtain
jjx#y
,
jj
MX
supfjjx#y
,
#zjj
XII
: jjzjj
XI
p1g
supfjjx#y#z
,
jj
XII
: jjzjj
XI
p1g
pC supfjjy#zjj
XII
: jjzjj
XI
p1g Cjjyjj
MX
:
This means that xAMMX and its norm is pC: &
Note that if X MY for some r.i. space Y; then X MX: Indeed, MX
MMY MY X:
For concrete r.i. spaces, like Lorentz, Orlicz and Marcinkiewicz, we have the
following results about multiplicator space. From the above discussion we have that
if 1opoNand 1pqpN; then
ML
p;q
L
p;minp;q
: 3
Proposition 2 (cf. Astashkin [A97] for p 1). Let jAC and 1ppoN: Then
(i) L
p; % j
CML
p;j
CL
p;j
:
(ii) ML
p;j
L
p;j
if and only if j is submultiplicative on 0; 1:
(iii)
If % jt lim
s-0

jst
js
; then ML
p;j
L
p; % j
:
The proof follows from [A97] (cf. also [Mi76,Mi78]), property (b) and the fact that
MX
p
MX
p
; where X
p
is the p-convexication of X .
Proposition 3. For the Orlicz space L
F
L
F
0; 1 we have the following:
(i) If FeD
2
; then ML
F
L
N
:
(ii)
If FAD
2
; then L
%
F
CML
F
CL
F
; where
%
Fu sup
vX1
F

uv
F

v
; uX1:
(iii) If FAD
2
; then ML
F
L
F
if and only if F is a submultiplicative function for
large u; i.e., FuvpCFuFv for some positive C; u
0
and all u; vXu
0
:
Proof. (i) It follows from property b
1
and the fact that Boyd index a
L
F
0:
(ii) The imbedding L
%
F
CML
F
follows from Ando theorem [A, Theorem 6] on
boundedness of tensor product between Orlicz spaces. In fact, if xsAL
%
F
and
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 253
ytAL
F
; then I
%
F
x=l I
F
y=loNfor some lX1; and so
Fl
2
jxsjjytjp1
%
Fjxsj=lF1 Fjytj=l;
from which
I
F
l
2
x#yp1 I
%
F
x=lF1 I
F
y=loN;
that is, x#yAL
F
I I: Therefore, L
%
F
CML
F
: The second imbedding follows
from (2).
(iii) It follows directly from (ii) and it was also proved in [A,A82,Mi81,O]. &
The situation is different in the case of Marcinkiewicz spaces.
Theorem 1. Let jAC: The following statements are equivalent:
(i) MM
j
M
j
:
(ii) The tensor product #: M
j
M
j
-M
j
I I is bounded.
(iii) j
0
#j
0
AM
j
:
(iv) There exists a constant C40 such that the inequality

n
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pCj
1
n

n
i1
u
i
_ _
4
is valid for any u
i
A0; 1; i 1; 2; y; n and every nAN:
Proof. The equivalence i3ii is true for any r.i. space, in particular also for the
Marcinkiewicz space M
j
:
Implication ii ) iii follows from the fact that j
0
AM
j
:
iii ) iv: Given an integer n and a sequence u
1
; u
2
; y; u
n
A0; 1; consider the
set
A
_
n
i1
i 1
n
;
i
n
_ _
0; u
i
C0; 1 0; 1:
Then
_
A
j
0
#j
0
dm
2
pCjm
2
A;
where m
2
is the Lebesque measure on 0; 1 0; 1: Since
_
A
j
0
tj
0
s dt ds

n
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 254
and
m
2
A

n
i1
1
n
u
i
it follows that estimate (4) holds.
iv ) ii: Assume that (4) is valid. It is sufcient to prove that the inequality
jjx#yjj
M
j
pC
holds for x; yAM
j
; jjxjj
M
j
jjyjj
M
j
1 and x x
n
; y y
n
: Given x x
n
AM
j
with
jjxjj
M
j
1 and e40 there exists a strictly decreasing function z z
n
AM
j
such that
jjzjj
M
j
p1 e and zXx: Therefore, we can assume in addition that x and y are
strictly decreasing and continuous on 0; 1: We must prove the inequality
_
A
t
xtys dt dspCm
2
A
t

for any set


A
t
ft; sA0; 1 0; 1 : xtysXtg; t40:
Given t40; there exists a continuous decreasing function gt g
t
t such that
A
t
ft; s : gsXtg:
Put
P
n

_
n
i1
0; g
i
n
_ _ _ _

i 1
n
;
i
n
_ _
and
Q
n

_
n
i1
0; g
i 1
n
_ _ _ _

i 1
n
;
i
n
_ _
:
Then
P
n
CA
t
CQ
n
:
The continuity of the function g implies that
lim
n-N
mQ
n
\P
n
lim
n-N

n
i1
1
n
g
i 1
n
_ _
g
i
n
_ _ _ _
p lim
n-N
max
1pipn
g
i 1
n
_ _
g
i
n
_ _ _ _
0:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 255
The function xtys belongs to L
1
m
2
: Hence
lim
n-N
_
Q
n
\P
n
xtys dt ds 0
and
_
A
t
xtys dt ds lim
n-N
_
P
n
xtys dt ds
lim
n-N

n
i1
_
g
i
n

0
xt dt
_
i
n
i1
n
ys ds
p lim
n-N

n
i1
j g
i
n
_ _ _ _ _
i
n
0
ys ds
_
i1
n
0
ys ds
_ _
lim
n-N

n
i1
j g
i
n
_ _ _ _
j g
i 1
n
_ _ _ _ _ __
i
n
0
ys ds:
Since
j g
i
n
_ _ _ _
j g
i 1
n
_ _ _ _
40
and
_
i
n
0
ys dspj
i
n
_ _
it follows that
_
A
t
xtys dt dsp lim
n-N

n
k1
j g
i
n
_ _ _ _
j g
i 1
n
_ _ _ _ _ _
j
i
n
_ _
lim
n-N

n
i1
j g
i
n
_ _ _ _
j
i
n
_ _
j
i 1
n
_ _ _ _
:
Denoting g
i
n
_ _
u
i
and applying (4) we get
_
A
t
xtys dt dsp lim
n-N

n
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pC lim
n-N
j
1
n

n
i1
u
i
_ _
C lim
n-N
jm
2
P
n
Cm
2
A
t
;
and the proof is complete. &
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 256
Observe that we have even proved that the tensor multiplicator norm in the space
M
j
is equal
sup
0ou
i
p1;i1;2;y;n;nAN

n
i1
ju
i
j
i
n
j
i1
n

j
1
n

n
i1
u
i

:
The concavity of j implies that the supremum is attained on the set of decreasing
sequences 1Xu
1
Xu
2
X?Xu
n
40:
Remark 1. Theorem 1 can be formulated in a more general form. Let j
1
; j
2
; j
3
AC:
Then the tensor product #: M
j
1
M
j
2
-M
j
3
I I is bounded if and only if there
exists a constant C40 such that the inequality

n
i1
j
1
u
i
j
2
i
n
_ _
j
2
i 1
n
_ _ _ _
pCj
3
1
n

n
i1
u
i
_ _
5
is true for every integer n and every u
i
A0; 1; i 1; 2; y; n:
Condition (5) can be also written in the integral form
_
1
0
j
1
utj
0
2
t dtpCj
3
_
1
0
ut dt
_ _
for all functions ut on 0; 1 such that 0putp1: The last integral condition is
satised when for example
_
1
0
j
1
s
t
_ _
j
0
2
t dtpCj
3
s
for all s in 0; 1: A similar assumption appeared in papers [Mi76,Mi78].
We will nd a condition on jAC under which estimate (4) will be true.
Lemma 1. Let jAC and jtpKjt
2
for some positive number K and for every
tA0; 1: Then

n
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pK 1j1j
1
n

n
i1
u
i
_ _
for every integer n and every u
i
A0; 1; i 1; 2; y; n:
Proof. The concavity of j implies that we can suppose the monotonicity
1Xu
1
Xu
2
X?Xu
n
X0: Denote
s
1
n

n
i1
u
i
:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 257
There exists a natural number m; 1pmpn; such that u
m
p

s
p
and u
i
4

s
p
for ipm:
Since
ns

n
i1
u
i
X

m
i1
u
i
Xm

s
p
it yields that mpn

s
p
and

m
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pj1

m
i1
j
i
n
_ _
j
i 1
n
_ _ _ _
j1j
m
n
_ _
pj1j
n

s
p
n
_ _
j1j

s
p
pKj1js:
If i4m; then ju
i
pjs and so

n
im1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pjs

n
im1
j
i
n
_ _
j
i 1
n
_ _ _ _
pj1js:
Hence

n
i1
ju
i
j
i
n
_ _
j
i 1
n
_ _ _ _
pK 1j1js K 1j1j
1
n

n
i1
u
i
_ _
: &
Immediately from Theorem 1 and Lemma 1 we have the following assertion.
Corollary 1. Let jAC and jtpKjt
2
for some positive number K and for every
tA0; 1: Then
MM
j
M
j
:
Let us note that the power function jt t
a
with 0oao1 does not satisfy
inequality (4) but there are functions jAC which satisfy the estimate jtpKjt
2

for some positive number K and for every tA0; 1: This estimate gives, of course, the
supermultiplicativity of j on 0; 1:
Example 1. For each l40 there exists a alA0; 1 such that the function
j
l
t
0 if t 0;
ln
l 1
t
if 0otpal;
linear if tAal; 1;
_

_
belongs to C: Clearly, j
l
tp2
l
j
l
t
2
for every tA0; al: Consequently, j
l
satises the conditions of Lemma 1.
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 258
Remark 2. There exists a function jAC such that the tensor product acts from
M
j
M
j
into M
j
and j does not satisfy the condition jtpKjt
2
for some
positive number K and for every tA0; 1: It is enough to take jt t
a
ln
b a
t
for
0oao1; b41 and a4e
2b=1a
:
We nish this part with the imbeddings of Caldero nLozanovski construction on
multiplicator spaces.
Proposition 4. Let X
0
; X
1
be r.i. spaces on 0; 1: Then
(i)
MX
0

1y
MX
1

y
CMX
1y
0
X
y
1
:
(ii) If rAU is a supermultiplicative function on 0; N; i.e., there exists a constant
c40 such that rstXcrsrt for all s; tA0; N; then
rMX
0
; MX
1
CMrX
0
; X
1
:
Proof. (i) Observe rst that YCMX if and only if the operator #: Y
X-XI I is bounded.
Since #: MX
i
X
i
-X
i
I I; i 0; 1; is bounded with the norm p1 and the
Caldero n construction is an interpolation method for positive bilinear operators
(cf. [C]) it follows that
#: MX
0

1y
MX
1

y
X
1y
0
X
y
1
-X
0
I I
1y
X
1
I I
y
X
1y
0
X
y
1
I I
is bounded with the norm p1: Therefore, MX
0

1y
MX
1

y
CMX
1y
0
X
y
1
.
(ii) When r is a supermultiplicative function the Caldero nLozanovski construc-
tion is an interpolation method for positive bilinear operators (see [As97,M]
Theorem 2]) and the proof of the imbedding is similar as in (i). &
Note that the inclusions in Proposition 4 can be strict. For the spaces X
0
L
p;q
;
X
1
L
p;N
with 1pqopoN we have
MX
0

1y
MX
1

y
ML
p;q

1y
ML
p;N

y
L
1y
p;q
L
y
p
L
p;r
;
where 1=r 1 y=q y=p and
MX
1y
0
X
y
1
ML
1y
p;q
L
y
p;N
ML
p;s
L
p;minp;s
;
where 1=s 1 y=q: The strict imbedding L
p;r
D! L
p;minp;s
gives then the
corresponding example.
3. Subspaces generated by dilations and translations in r.i. spaces
Given an r.i. space X on I 0; 1 let us denote by
V
0
X faAX : aa0; a a
n
g:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 259
For a xed function aAV
0
X and dyadic intervals D
n;k

k1
2
n
;
k
2
n
; k 1; 2; y; 2
n
;
n 0; 1; 2; y; let us consider the dilations and translations of a function a
a
n;k
t
a2
n
t k 1 if tAD
n;k
;
0 elsewhere:
_
Then supp a
n;k
CD
n;k
and
mftAD
n;k
: ja
n;k
tj4lg 2
n
mftAI : jatj4lg for all l40:
For aAV
0
X and n 0; 1; 2; y we denote by Q
a;n
the linear span fa
n;k
g
2
n
k1

generated by functions a
n;k
in X:
Denition 2. For an r.i. function space X on 0; 1 the nice part N
0
X of X is dened
by
N
0
X aAV
0
X : there exists a sequence of projections fP
n
g
N
n0
on X such that
_
ImP
n
Q
a;n
and sup
n0;1;y
jjP
n
jj
X-X
oN
_
:
We say that X is a nice space (or shortly XAN
0
) if a
n
belongs to N
0
X for every a
from X:
We are using here similar notions as in the paper [HS99]. They were consid-
ering r.i. space X X0; N on 0; N; the corresponding set VX
faAX : aa0; supp aC0; 1; a a
n
g and the set NX of all aAVX such that Q
a
is a complemented subspace of X X0; N; where Q
a
is the linear closed span
generated by the sequence a
k

N
k1
with
a
k
t at k 1 for tAk 1; k and a
k
t 0 elsewhere:
If NX X; then they write that XAN (or say that X is a nice space).
We are putting sub-zero notions, that is, V
0
X and N
0
X; so that we have
difference between of our case of r.i. spaces on 0; 1 and their case 0; N:
Theorem 2. Let X be an r.i. space on 0; 1 and let X
0
denote the closure of L
N
0; 1 in
X: Then we have embeddings
(i) MXCN
0
X;
(ii)
N
0
X
0
CMX:
Before the proof of this theorem we will need some auxiliary results.
Let aAV
0
X and f AV
0
X
0
be such that
_
1
0
f tat dt 1: 6
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 260
Dene the sequence of natural projections (averaging operators)
P
n
xt P
n;a;f
xt

2
n
k1
2
n
_
D
n;k
f
n;k
sxsdsa
n;k
t; n 0; 1; 2; y : 7
Lemma 2. The sequence of norms fjjP
n;a;f
jj
X-X
g
N
n0
is a non-decreasing sequence.
Proof. For x xt with supp xCD
n;k
we dene
R
n;k
xt x 2t
k 1
2
n
_ _
; S
n;k
xt x 2t
k
2
n
_ _
:
Then
supp R
n;k
xCD
n1;2k1
; supp S
n;k
xCD
n1;2k
and
mftAD
n1;2k1
: jR
n;k
xtj4lg mftAD
n1;2k
: jS
n;k
xtj4lg

1
2
mftAI : jxtj4lg
for all l40: Therefore,
_
D
n1;2k1
R
n;k
xt dt
_
D
n1;2k
S
n;k
xt dt
1
2
_
D
n;k
xt dt:
Moreover,
R
n;k
f
n;k
xw
D
n;k
t f
n1;2k1
R
n;k
xw
D
n;k
t;
S
n;k
f
n;k
xw
D
n;k
t f
n1;2k
S
n;k
xw
D
n;k
t
and
mftAD
n1;j
: a
n1;j
t4lg
1
2
mftAD
n;i
: a
n;i
4lg
for all l40 and any i 1; 2; y; 2
n
; j 1; 2; y; 2
n1
:
Denote P
n
P
n;a;f
: The last equality and the equality of integrals give that the
function P
n
xt is equimeasurable with the function
P
n1
yt

2
n
k1
2
n1
_
D
n1;2k1
f
n1;2k1
sR
n;k
xw
D
n;k
s ds
_ _
a
n1;2k1
t

2
n
k1
2
n1
_
D
n1;2k
f
n1;2k
sS
n;k
xw
D
n;k
s ds
_ _
a
n1;2k
t;
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 261
where
yt

2
n
k1
R
n;k
xw
D
n;k
t S
n;k
xw
D
n;k
t:
From the above we can see that y is equimeasurable with x and so
jjP
n
xjj
X
jjP
n1
yjj
X
pjjP
n1
jj jjyjj
X
jjP
n1
jj jjxjj
X
;
that is, jjP
n
jjpjjP
n1
jj: &
Lemma 3. Let X be a separable r.i. space. If aAN
0
X; then there exists a function
f AN
0
X
0
such that (6) is fullled and for the sequence of projections fP
n;a;f
g dened
by (7) we have
sup
n0;1;2;y
jjP
n;a;f
jj
X-X
oN:
Proof. Since X is a separable space and aAN
0
X it follows that there are functions
g
n;k
AX
0
k 1; 2; y; 2
n
; n 0; 1; 2; y such that
_
1
0
g
n;k
sa
n;k
s ds 1 and
_
1
0
g
n;k
sa
n;j
s ds 0; jak;
and for the projections
T
n
xt

2
n
k1
_
1
0
g
n;k
sxs ds
_ _
a
n;k
t
we have sup
n0;1;y
jjT
n
jj
X-X
oN: Let fr
i
g
n
i1
be the rst n Rademacher functions on
the segment 0; 1: Since X is an r.i. space it follows that for every uA0; 1 the norms
of the operators
T
n;u
xt

2
n
k1
r
k
u

2
n
i1
r
i
u
_
D
n;i
g
n;k
sxs ds
_ _
a
n;k
t
are the same as the norms of T
n
: Let us consider the operators
S
n
xt
_
1
0
T
n;u
xt du

2
n
k1
_
D
n;k
g
n;k
sxs ds
_ _
a
n;k
t:
Then
jjS
n
jjp sup
uA0;1
jjT
n;u
jj jjT
n
jjpC:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 262
Therefore, we can assume that
supp g
n;k
CD
n;k
and g
n;k
is decreasing on D
n;k
:
Moreover, we shift supports of the functions g
n;k
k 1; 2; y; 2
n
to the segment
0; 2
n
and consider the averages
G
n
t 2
n

2
n
k1
t
k12
n g
n;k
t;
where t
s
gt gt s:
Then the shifts h
n;k
t 2
n
t
k12
n G
n
t generate operators
U
n
xt

2
n
k1
2
n
_
D
n;k
h
n;k
sxs ds
_ _
a
n;k
t
and we can show that jjU
n
jj
X-X
pC:
Since h
n;k
t F
n

n;k
t; where F
n
t 2
n
G
n
2
n
t for 0ptp1; it follows that
_
1
0
F
n
tat dt
_
2
n
0
G
n
ta
n;1
t dt
2
n

2
n
k1
_
2
n
0
t
k12
n g
n;k
ta
n;1
t dt
2
n

2
n
k1
_
D
n;k
g
n;k
ta
n;k
t dt 1:
Let us show that there exists a subsequence fF
n
k
tg of F
n
t which converges at
every tA0; 1:
Lemma 2 gives that the norm of the one-dimensional operator
L
n
xt
_
1
0
F
n
sxs ds
_ _
at
does not exceed jjU
n
jj
X-X
; and consequently also not C: Therefore,
jjF
n
jj
X
0 p
C
jjajj
X
: 8
By the denition of F
n
we have F
n
n
t F
n
t and
1
_
1
0
F
n
sas dsXF
n
t
_
t
0
as ds
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 263
or
F
n
tp
_
t
0
as ds
_ _
1
for all tA0; 1:
Applying Helly selection theorem (see [N]) we can choose subsequences
fF
n
g*fF
n;1
g*fF
n;2
g*y*fF
n;m
g*y
that converge on the intervals
1
2
; 1
_ _
;
1
3
; 1
_ _
; y;
1
m 1
; 1
_ _
; y;
respectively. Then the diagonal sequence f
n
t F
n;n
t converges at every tA0; 1 to
a function f t f
n
t: Using estimate (8) we obtain
_
s
0
f
n
tat dtpjjf
n
jj
X
0 jjaw
0;s
jj
X
p
C
jjajj
X
jjaw
0;s
jj
X
:
Since X is a separable r.i. space it follows that jjaw
0;s
jj
X
-0 as s-0

: Therefore, the
equalities f
n
n
f
n
and a
n
a imply that ff
n
ag is an equi-integrable sequence of
functions on 0; 1: Hence (see [E, Theorem 1.21], or [HM, Theorem 6, Chapter V])
_
1
0
f tat dt lim
n-N
_
1
0
f
n
tat dt 1:
Let mAN be xed. By the estimate jjU
n
jj
X-X
pC; the denition of f
n
; and Lemma 2
we have

2
m
k1
2
m
_
D
m;k
f
n

m;k
txt dt
_ _
a
m;k

X
pCjjxjj
X
9
for all nXm and all xAX:
Suppose that xt is a non-negative and non-increasing function on every interval
D
m;k
; k 1; 2; y; 2
m
: As above, from (8) it follows that ff
n

m;k
xw
D
m;k
g
N
m1
is an equi-
integrable sequence on D
m;k
: Hence
lim
n-N
_
D
m;k
f
n

m;k
txt dt
_
D
m;k
f
m;k
txt dt
and for all such functions xt estimate (9) implies

2
m
k1
2
m
_
D
m;k
f
m;k
txt dt
_ _
a
m;k

X
pCjjxjj
X
:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 264
Since X is an r.i. space we can prove that the above estimate holds for all xAX: The
proof is complete. &
Proof of Theorem 2. (i) At rst by the result in [A97, Theorem 1.14], we have that
aAMX if and only if there exists a constant C40 such that

2
n
k1
c
n;k
a
n;k

X
pC

2
n
k1
c
n;k
w
D
n;k

X
10
for all c
n;k
AR and k 1; 2; y; 2
n
; n 0; 1; 2; y.
Suppose that aAV
0
X-MX; that is, estimate (10) holds. If et 1; then the
operators
P
n;e
xt

2
n
k1
2
n
_
D
n;k
xs ds
_ _
w
D
n;k
t n 1; 2; y 11
are bounded projections in every r.i. space X and jjP
n;e
jj
X-X
p1 (see [KPS, Theorem
4.3]).
Dene operators R
n;a
: Q
e;n
Im P
n;e
-Q
a;n
as follows:
R
n;a

2
n
k1
c
n;k
w
D
n;k
_ _

2
n
k1
c
n;k
a
n;k
:
By the assumption aAMX or equivalently by estimate (10) we have
jjR
n;a
jj
Q
e;n
-X
pC: Therefore, for the operators
P
n;a

1
jjajj
L
1
R
n;a
P
n;e
:
We have
jjP
n;a
jj
X-X
pCjjajj
1
L
1
; n 1; 2; y :
It is easy to check that P
n;a
are projections and Im P
n;a
Q
a;n
: Therefore, aAN
0
X:
(ii) If X L
N
; then MX N
0
X
0
L
N
:
If XaL
N
; then X
0
is a separable r.i. space. In this case, by Lemma 3, for any
aAN
0
X
0
there exists a function f AV
0
X
0

0
such that (6) is fullled and for the
projections P
n;a;f
dened as in (7) we have
C sup
n0;1;2;y
jjP
n;a;f
jj
X
0
-X
0 oN: &
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 265
If x is a function of the form xs

2
n
k1
c
n;k
w
D
n;k
s; then
P
n;a;f
x jjf jj
L
1

2
n
k1
c
n;k
a
n;k
:
Therefore,

2
n
k1
c
n;k
a
n;k

X
0
pCjjf jj
1
L
1

2
n
k1
c
n;k
w
D
n;k

X
0
;
and we obtain (10) for X
0
:
If X
0
X; that is, X is a separable r.i. space, then the theorem is proved. If X is a
non-separable r.i. space, then X
0
is an isometric subspace of X X
00
: By using the
Fatou property, we can extend the above inequality to the whole space X and obtain
(10), which gives that aAMX: The proof of Theorem 2 is complete. &
Immediately from Theorem 2 and the properties of the multiplicator space we
obtain the following corollaries.
Corollary 2. If X is a separable r.i. space, then MX N
0
X:
Corollary 3. If 1opoN; 1pqpN; then N
0
L
p;q
L
p;q
for 1pqpp and
N
0
L
p;q
N
0
L
0
p;N
L
p
for poqoN:
Corollary 4. Let X
0
and X
1
be separable r.i. spaces. If X
0
; X
1
AN
0
; then
X
1y
0
X
y
1
AN
0
:
Corollaries 3 and 4 show that the class of nice spaces N
0
is stable with respect to
the complex method of interpolation but it is not stable with respect to the real
interpolation method.
Corollary 5. If jAC and jtpKjt
2
for some positive number K and for every
tA0; 1; then N
0
M
j
M
j
:
By jAC
0
we mean jAC such that lim
t-0

t
jt
0:
Theorem 3. Let jAC
0
:
(i) If
lim sup
t-0

j2t
jt
2 12
then
L
N
CN
0
M
j
CL
N
,M
j
\M
0
j
: 13
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 266
(ii) If
lim
t-0

j2t
jt
2
then
N
0
M
j
L
N
,M
j
\M
0
j
:
Proof. (i) By Theorem 2 the left part of (13) is valid for any r.i. space. Assumption
(12) implies
lim sup
t-0

jt
jts

1
s
; 0oso1
and so
jjs
s
jj
M
j
-M
j
Xs lim sup
t-0

jt
jst
1
for every 0osp1: This means that a
M
j
0:
Let xAN
0
M
j
-M
0
j
: By using Corollary 2 and property b
1
we get
xAN
0
M
0
j
MM
0
j
L
N
:
This proves the right part of (13).
(ii) We must only prove the inclusion
M
j
\M
0
j
CN
0
M
j
:
Let aAM
j
\M
0
j
; jjajj
M
j
1 and ct
_
t
0
a
n
s ds: It is well known that
dista; M
0
j
lim sup
t-0

1
jt
_
t
0
a
n
s ds:
Therefore,
g lim sup
t-0

ct
jt
40
and there exists a sequence ft
m
g tending to 0 such that
lim
m-N
ct
m

jt
m

g:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 267
Since
lim
m-N
jt
m

2
n
j
t
m
2
n

1
for every natural n; it follows that
jja
n;k
jj
M
j
X lim sup
m-N
ct
m

2
n
j
t
m
2
n

lim sup
m-N
ct
m

jt
m

lim
m-N
jt
m

2
n
j
t
m
2
n

g
for every 1pkp2
n
; n 1; 2; y.
Consider the subspaces H
n;k
=closed span fa
n;i
g
iak
: These subspaces are closed
subspaces of M
j
and a
n;k
eH
n;k
: Thus, by the HahnBanach theorem, there are
b
n;k
AM
j

n
such that b
n;k
j
H
n;k
0; b
n;k
a
n;k
1 and jjb
n;k
jj
1
jja
n;k
jj
p
1
g
: Then the
operators
P
n
x

2
n
k1
b
n;k
xa
n;k
are projections from M
j
onto Q
a;n
: Moreover, P
n
are uniformly bounded since
jjP
n
xjj
M
j
p
1
g

2
n
k1
jjxjj
M
j
a
n;k

M
j

1
g
jjxjj
M
j
:
Therefore, aAN
0
M
j
: The proof is complete. &
Example 2. There exists a non-separable r.i. space X such that MXaN
0
X:
Take X M
j
with jt t ln
e
t
on 0; 1: Since a
M
j
0 it follows that MM
j

L
N
: The function at ln
e
t
for tA0; 1 as unbounded is not in MX but it is in
M
j
\M
0
j
and by Theorem 3(ii) it shows that aAN
0
X: Therefore, N
0
XaMX:
Corollary 6. If jAC
0
and lim sup
t-0

j2t
jt
2; then j
0
AN
0
M
j
and consequently
N
0
M
j
is neither a linear space nor a lattice.
Problem 1. For 1opoNdescribe N
0
L
p;N
:
Note that ML
p;N
L
p
and N
0
L
0
p;N
ML
0
p;N
L
p
:
Theorem 4. Let X be an r.i. space X on 0; 1: The following conditions are equivalent:
(i) #: X X-XI I is bounded.
(ii) MX X:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 268
(iii) XAN
0
; i.e., N
0
X X:
(iv) There exists a constant C40 such that

2
n
k1
c
n;k
a
n;k

X
pCjjajj
X

2
n
k1
c
n;k
w
D
n;k

X
14
for all aAX and all c
n;k
AR; k 1; 2; y; 2
n
; n 0; 1; 2; y.
Proof. Implication i ) ii follows by denition ii ) iii from Theorem 2 and
iv ) i by the result in [A97, Theorem 1.14]. Therefore, it only remains to prove
that (iii) implies (iv).
First, assume additionally that X is separable. If XAN
0
; then, similarly as in the
proof of Theorem 2, for any aAX there exist C
1
40 and f AV
0
X
0
such that
_
1
0
f tat dt 1 and

2
n
k1
c
n;k
a
n;k

X
pC
1
jjf jj
1
L
1

2
n
k1
c
n;k
w
D
n;k

X
:
Let us introduce a new norm on X dened by
jjajj
1
sup

2
n
k1
c
n;k
a
n;k

2
n
k1
c
n;k
w
D
n;k

X
: c
n;k
AR; k 1; 2; y; 2
n
; n 0; 1; 2; y
_

_
_

_
:
Then jjajj
X
pjjajj
1
and jjajj
1
oNfor all aAX: By the closed graph theorem we obtain
that jjajj
1
pCjjajj
X
and (14) is proved.
Now, let X be a non-separable r.i. space. In the case X L
N
both conditions (iii)
and (iv) are fullled. Therefore, consider the case XaL
N
: Then X
0
is a separable r.i.
space. The canonical isometric imbedding X
0
CX X
00
gives that X
0
AN
0
: Let
aAV
0
X: The separability of X
0
implies

2
n
k1
c
n;k
a
m

n;k

X
0
pCjja
m
jj
X
0

2
n
k1
c
n;k
w
D
n;k

X
0
Cjja
m
jj
X
0

2
n
k1
c
n;k
w
D
n;k

X
;
where a
m
t minat; m; m 1; 2; y.
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 269
Since X X
00
has the Fatou property and a
m

n;k
a
n;k

m
it follows that

2
n
k1
c
n;k
a
n;k

X
lim
m-N

2
n
k1
c
n;k
a
n;k
_ _
m

X
pCjjajj
X

2
n
k1
c
n;k
w
D
n;k

X
and Theorem 4 is proved. &
Theorem 5. Let X be an r.i. space X on 0; 1: Then XAN
0
if and only if X
00
AN
0
:
Proof. The proof is similar to that of Theorem 4. The essential part is the proof of
the estimate (14). We leave the details to the reader.
Theorem 6. Let X be a separable r.i. space X on 0; 1: Then the following conditions
are equivalent:
(i) aAN
0
X:
(ii) The operators R
n;a
: Q
e;n
-Q
a;n
given by
R
n;a

2
n
k1
c
n;k
w
D
n;k
_ _

2
n
k1
c
n;k
a
n;k
are uniformly bounded.
(iii) The operators R
n;a
and their inverses are uniformly bounded.
Proof. i ) iii: Let aAN
0
X: Then jjR
n;a
jjpC for all n 0; 1; 2; y; by Theorem
2. Next, since aa0 there exists n
0
AN and e en
0
40 such that atXut
ew
0;2
n
0

t: Therefore, for all c


n;k
AR;

2
n
k1
c
n;k
a
n;k

X
X

2
n
k1
c
n;k
u
n;k

X
e

2
n
k1
c
n;k
w
k12
n
;k12
n
0
2
n

X
e s
2
n
0

2
n
k1
c
n;k
w
D
n;k
_ _

X
Xejjs
2
n
0 jj
1
X-X

2
n
k1
c
n;k
w
D
n;k

X
;
which shows that the inverses R
n;a

1
are uniformly bounded.
ii ) i: If the operators R
n;a
are uniformly bounded, then we have estimate (14)
or equivalently aAMX and Theorem 2(i) gives that aAN
0
X: &
Now, we present a characterization of L
p
spaces among all r.i. spaces on
0; 1:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 270
Theorem 7. Let X be an r.i. space X on 0; 1: The following conditions are equivalent:
(i) XAN
0
and X
0
AN
0
:
(ii) There exists a constant C40 such that
C
1

2
n
k1
c
n;k
w
D
n;k

X
p

2
n
k1
c
n;k
a
n;k

X
pC

2
n
k1
c
n;k
w
D
n;k

X
15
for all aAV
0
X with jjajj
X
1 and all c
n;k
AR with k 1; 2; y; 2
n
; n
0; 1; 2; y.
(iii) For any pair of functions a; f such that aAV
0
X; f AV
0
X
0
satisfying (6) the
operators P
n;a;f
dened in (7) are uniformly bounded in X:
(iv) The operator of the tensor product #is bounded from X X into XI I and
from X
0
X
0
into X
0
I I:
(v) There exists a pA1; N such that X L
p
:
Proof. i ) iv: This follows from Theorem 4.
iv ) ii: Let aAV
0
X; jjajj
X
1: Assumption (iv) implies, by Theorem 4, that

2
n
k1
c
n;k
a
n;k

X
pC
1

2
n
k1
c
n;k
w
D
n;k

X
for some constant C
1
40:
Therefore it only remains to prove left estimate in (15). For arbitrary bAV
0
X
0

such that
_
1
0
atbt dt 1 and

2
n
k1
d
n;k
b
n;k

X
0
p1d
n;k
AR
we obtain
_
D
n;k
a
n;k
tb
n;k
t dt 2
n
and

2
n
k1
c
n;k
a
n;k

X
X
_
1
0

2
n
k1
c
n;k
a
n;k
t
_ _

2
n
k1
d
n;k
b
n;k
t
_ _
dt 2
n

2
n
k1
c
n;k
d
n;k
:
Since #is bounded from X
0
X
0
into X
0
I I it follows, again by Theorem 4 used
to X
0
; that

2
n
k1
d
n;k
b
n;k

X
0
pC
2

2
n
k1
d
n;k
w
D
n;k

X
0
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 271
for some constant C
2
40; from which we conclude that

2
n
k1
c
n;k
a
n;k

X
XC
1
2
sup
d
_
1
0

2
n
k1
c
n;k
a
n;k
t
_ _

2
n
k1
d
n;k
b
n;k
t
_ _
dt
2
n
C
1
2
sup
d

2
n
k1
c
n;k
d
n;k
;
where the supremum is taken over all d d
n;k

2
n
k1
such that

2
n
k1
d
n;k
w
D
n;k

X
0
p1:
The operator P
n;e
dened as in (11) by
P
n;e
xt

2
n
k1
2
n
_
D
n;k
xs ds
_ _
w
D
n;k
t n 1; 2; y
satises jjP
n;e
jj
X
0
-X
0 p1: Therefore,

2
n
k1
c
n;k
w
D
n;k

X
sup
jjyjj
X
0 p1
_
1
0

2
n
k1
c
n;k
w
D
n;k
t
_ _
yt dt
sup
jjyjj
X
0 p1
_
1
0

2
n
k1
c
n;k
w
D
n;k
t
_ _
P
n;e
yt dt
p sup
jjP
n;e
yjj
X
0 p1
_
1
0

2
n
k1
c
n;k
w
D
n;k
t
_ _
P
n;e
yt dt
p2
n
sup
d

2
n
k1
c
n;k
d
n;k
:
Hence

2
n
k1
c
n;k
w
D
n;k

X
pC
2

2
n
k1
c
n;k
a
n;k

X
:
ii ) v: By Krivines theorem [K,LT, p. 141] for every r.i. space X there exists
pA1; N with the following property:
for any n 0; 1; 2; yand e40 there exist disjoint and equimeasurable functions
v
k
AX; k 1; 2; y; 2
n
; such that
1 ejjcjj
p
p

2
n
k1
c
n;k
v
k

X
p1 ejjcjj
p
16
for any c c
n;k

2
n
k1
; where jjcjj
p


2
n
k1
jc
n;k
j
p
_ _
1=p
: Hence, in particular, it follows
(with the notion
n
N
0 that
1 e2
n=p
p

2
n
k1
v
k

X
p1 e2
n=p
:
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 272
Let
at r
1

2
n
k1
v
k
_ _
n
t; where r

2
n
k1
v
k

X
:
Then jjajj
X
1 and a
n;k
are equimeasurable functions with r
1
v
k
for every
k 1; 2; y; 2
n
: Therefore,
1 e
1 e
2
n=p
jjcjj
p
p

2
n
k1
c
n;k
a
n;k

X
p
1 e
1 e
2
n=p
jjcjj
p
;
that is,
1 e
1 e

2
n
k1
c
n;k
w
D
n;k

p
p

2
n
k1
c
n;k
a
n;k

X
p
1 e
1 e

2
n
k1
c
n;k
w
D
n;k

p
:
Hence, by assumption (15), we have
C
1
e

2
n
k1
c
n;k
w
D
n;k

p
p

2
n
k1
c
n;k
w
D
n;k

X
pC
e

2
n
k1
c
n;k
w
D
n;k

p
; 17
where C
e
C1 e=1 e:
Let 1ppoN: If X is a separable r.i. space, then (17) implies that X L
p
: In the
case when X X
00
it is sufcient to consider r.i. spaces XaL
N
: Then X
0
satises
15 and so X
0
L
p
: Hence, X
0
X
0

0
L
p
0 and X X
00
L
p
0
0
L
p
:
Let p N: Suppose that there is a function xAX\L
N
: Then from (17) we obtain
jjxjj
X
X

2
n
k1
x
n
k2
n
w
D
n;k

X
XC
1
e

2
n
k1
x
n
k2
n
w
D
n;k

N
C
1
e
x
n
2
n
:
Since xeL
N
it follows that lim
n-N
x
n
2
n
N: This contradiction shows that
XCL
N
: The reverse imbedding is always true.
v ) iii: This follows from the estimate of the norm of natural projections in L
p
space
jjP
n;a;f
jj
L
p
-L
p
pjjajj
p
jjf jj
p
0 :
iii ) i: By denition of the operators P
n;a;f
we have that XAN
0
: We want to
show that also X
0
AN
0
: For all xAX and yAX
0
_
1
0
P
n;a;f
xtyt dt

2
n
k1
2
n
_
D
n;k
f
n;k
sxs ds
_
D
n;k
a
n;k
tyt dt
_
1
0
P
n;f ;a
ysxs ds:
Therefore, the conjugate operator P
n;a;f

n
to P
n;a;f
is P
n;f ;a
on the space X
0
: Since X
0
is isometrically imbedded in X
n
the last equality implies that the operators P
n;f ;a
are
uniformly bounded, and so X
0
AN
0
: The proof is complete. &
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 273
4. Additional remarks and results
First we describe the difference between the cases on 0; 1 and 0; N: Let X0; N
denote an r.i. space on 0; N and X fxAX0; N : xt 0 for t41g the
corresponding r.i. space on 0; 1: We use here also the notion X0; NANfrom the
paper [HS99, p. 56] (cf. also our explanation after Denition 2). Let us present
examples showing that no one of the following statements:
(i) X0; NAN;
(ii) XAN
0
implies the other one, in general.
Example 3. The Orlicz space L
F
p
0; N; where F
p
u e
u
p
1; 1opoN; belongs
to the class N: On the other hand, the lower Boyd index a
L
Fp
of L
F
p
on 0; 1 equals 0
and so ML
F
p
L
N
: Therefore, by Theorem 4, L
F
p
eN
0
.
Example 4. Consider the function
jt
t
a
if 0ptp1;
t
a
ln
b
t e 1 if 1ptoN;
_
where 0obpao1: Then j is a quasi-concave function on 0; N; i.e., jt is
increasing on 0; N and jt=t is decreasing on 0; N: Let * j be the smallest
concave majorant of j: Then
sup
0otp1;nAN
* jnt
* jn * jt
N:
In fact, for every n 1; 2; y; we can choose tA0; 1 such that nto1: Then
* jnt
* jn * jt
X
1
4
jnt
jnjt
ln
b
n e 1-N as n-N:
This implies that the Lorentz space L
* j
0; NeN(see [HS99, Theorem 4.1]). At the
same time for L
* j
L
t
a L
p;1
with p 1=a on 0; 1 we have, by (3), that ML
* j

ML
p;1
L
p;1
L
* j
; and, by Theorem 4, L
* j
AN
0
:
The reason of non-equivalences (i) and (ii) is coming from the fact that the dilation
operator s
t
in r.i. spaces on 0; N does not satisfy an equation of the form
jjs
t
xjj
X0;N
f tjjxjj
X0;N
; for xAX0; N and for all t40:
If this equation is satised, then the function f t is a power function f t t
a
for
some aA0; 1 and then the above statements (i) and (ii) are equivalent. This
observation allows us to improve, for example, Theorem 4.2 from [HS99]: if
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 274
1opoNand 1pqpN; then
NL
p;q
0; N L
p;q
31pqpp:
We can characterize NL
p;q
0; N for 1oppqoN:
Theorem 8. If 1oppqoN; then NL
p;q
0; N L
p
:
Proof. Let aANL
p;q
0; N: The spaces L
p;q
0; N are separable for qoN:
Therefore, similarly as in the proof of Theorem 2, we can show that

n
k1
c
k
a
k

L
p;q
0;N
pC

n
k1
c
k
w
k1;k

L
p;q
0;N
18
for all c
k
AR; k 1; 2; y; n; n 1; 2; y. Since
jjs
t
xjj
L
p;q
0;N
t
1=p
jjxjj
L
p;q
0;N
for xAL
p;q
0; N and all t40; 19
it follows that

n
k1
c
k
a
n
k

L
p;q
pC

n
k1
c
k
w
k1
n
;
k
n
_ _

L
p;q
;
and, by Theorem 1.14 in [A97] together with property (c), we obtain aAL
p
:
Conversely, if aAL
p
then, by using property (c), Theorem 1.14 in [A97] and
equality (19), we get (18) for all n of the form 2
m
; m 1; 2; y. The space L
p;q
has the
Fatou property, thus passing to the limit, we obtain

N
k1
c
k
a
k

L
p;q
0;N
p

N
k1
c
k
w
k1;k

L
p;q
0;N
:
Next, arguing as in the proof of Theorem 2 (see also [HS99, Theorem 2.3]) we obtain
aANL
p;q
0; N:
References
[A] T. Ando, On products of Orlicz spaces, Math. Ann. 140 (1960) 174186.
[A82] S.V. Astashkin, On a bilinear multiplicative operator, Issled. Teor. Funkts. Mnogikh Veshchestv.
Perem. Yaroslavl (1982) 315 (in Russian).
[A97] S.V. Astashkin, Tensor product in symmetric function spaces, Collect. Math. 48 (1997) 375391.
[As97] S.V. Astashkin, Interpolation of positive polylinear operators in Caldero nLozanovski spaces,
Sibirsk. Mat. Zh. 38 (1997) 12111218 (English translation in Siberian Math. J. 38 (1997)
10471053).
[BS] C. Bennett, R. Sharpley, Interpolation of Operators, Academic Press, Boston, 1988.
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 275
[C] A.P. Caldero n, Intermediate spaces and interpolation, Studia Math. 24 (1964) 113190.
[E] R.J. Elliott, Stochastic Calculus and Applications, Springer, New York, 1982.
[HM] S. Hartman, J. Mikusin ski, The Theory of Lebesgue Measure and Integration, Pergamon Press,
New York, Oxford, London, Paris, 1961.
[HS99] F.L. Hernandez, E.M. Semenov, Subspaces generated by translations in rearrangement invariant
spaces, J. Funct. Anal. 169 (1999) 5280.
[KR] M.A. Krasnoselskii, Ya.B. Rutickii, Convex Functions and Orlicz Spaces, Noordhoff Ltd.,
Groningen, 1961.
[KPS] S.G. Krein, Yu.I. Petunin, E.M. Semenov, Interpolation of Linear Operators, Nauka, Moscow,
1978 (in Russian) (English translation in Amer. Math. Soc., Providence, 1982).
[K] J.L. Krivine, Sous-espaces de dimension nie des espaces de Banach re ticule s, Ann. of Math. 104
(1976) 129.
[LT] J. Lindenstrauss, L. Tzafriri, Classical Banach Spaces, II. Function Spaces, Springer, Berlin, New
York, 1979.
[M89] L. Maligranda, Orlicz Spaces and Interpolation, Seminars in Mathematics, Vol. 5, University of
Campinas, Campinas SP, Brazil, 1989.
[M] L. Maligranda, Positive bilinear operators in Caldero nLozanovski spaces, Lulea University of
Technology, Department of Mathematics, Research Report, No. 11, 2001, pp. 117.
[Mi76] M. Milman, Tensor products of function spaces, Bull. Amer. Math. Soc. 82 (1976) 626628.
[Mi78] M. Milman, Embeddings of LorentzMarcinkiewicz spaces with mixed norms, Anal. Math. 4
(1978) 215223.
[Mi81] M. Milman, A note on Lp; q spaces and Orlicz spaces with mixed norms, Proc. Amer. Math.
Soc. 83 (1981) 743746.
[N] I.P. Natanson, Theory of Functions of a Real Variable, Vol. II, Frederick Ungar Publishing Co.,
New York, 1961.
[O] R. ONeil, Integral transforms and tensor products on Orlicz spaces and Lp; q spaces,
J. Analyse Math. 21 (1968) 1276.
[S] T. Shimogaki, A note on norms of compression operators on function spaces, Proc. Japan Acad.
46 (1970) 239242.
ARTICLE IN PRESS
S.V. Astashkin et al. / Journal of Functional Analysis 202 (2003) 247276 276