Real Analysis Exchange

Vol. 28(2), 2002/2003, pp. 381–394

B. Bongioanni, Departamento de Matemática (FIQ-UNL), and

IMAL-CONICET, Santa Fe, Argentina. e-mail: bbongio@[Link]
L. Forzani, Departamento de Matemática (FIQ-UNL), and
IMAL-CONICET, Santa Fe, Argentina. e-mail: lforzani@[Link]
E. Harboure, Departamento de Matemática (FIQ-UNL), and
IMAL-CONICET, Santa Fe, Argentina. e-mail: harbour@[Link]

WEAK TYPE AND RESTRICTED WEAK

TYPE (p, p) OPERATORS IN ORLICZ
SPACES

Abstract
Rt
Let (Ω, µ) be a finite measure space, Φ(t) = 0 a(s) ds and Ψ(t) =
Rt
0
b(s) ds, where a and b are positive continuous functions defined on
[0, ∞). Consider the associated Orlicz spaces LΦ (Ω) and LΨ (Ω). In
this paper we find a relationship between a and b to assure that T , a
sublinear and positive homogeneous operator of restricted weak type
(p, p) and of type (∞, ∞), maps LΨ (Ω) into LΦ (Ω). If the two Orlicz
spaces are normable, our results imply the continuity of T . This relation
between a and b is sharp since it is shown to be necessary for operators
like the one side maximal operators related to the Cesàro averages.

1 Introduction
Let (Ω, µ) be a finite measure space and M(Ω) be the space of measurable
functions from Ω into R. Let Ψ be a nondecreasing continuous function such
that Ψ(0) = 0 and limt→∞ Ψ(t) = ∞. The family of functions
Z
LΨ (Ω) = {f ∈ M(Ω) : Ψ(|f |) dµ < ∞ for some  > 0 }
Ω

is called an Orlicz Space. For more details see Rao and Ren [6].
If f is a measurable function, we define µf : (0, ∞) → [0, ∞], the distribu-
tion function of f , as µf (s) = µ({x ∈ Ω : |f (x)| > s}) for all s > 0.
Key Words: Orlicz Space, Interpolation, Maximal Function
Mathematical Reviews subject classification: 42B25, 46E30
Received by the editors August 29, 2002

381
382 B. Bongioanni, L. Forzani and E. Harboure

Let T be a sublinear and positive homogeneous operator defined on a

subspace D ⊂ M(Ω) and taking values on M(Ω). We assume that D contains
all the characteristic functions of sets of finite measure and has the property
that whenever f ∈ D and g is a truncation of f , then g ∈ D. Such an operator
T is of weak type (p, p) if there exists

p a constant A such that for any measurable
function f ∈ D, µT f (s) ≤ As kf kp for all s > 0 .
T is of restricted weak type (p, p) if there
 exists a constant
p A such that for
A ∞ 1/p
R
any measurable function f ∈ D, µT f (s) ≤ s 0 µf for all s > 0. Finally,
T is of type (∞, ∞) if there exists a constant B such that for any measurable
function f ∈ D, kT f k∞ ≤ Bkf k∞ .

Remark 1. In terms of Lorentz spaces Lp,q , an operator T is of restricted

weak type (p, p) if there exists a constant C such that kT f kp,∞ ≤ Ckf kp,1 for
all f ∈ D. If a sublinear and positive homogeneous operator satisfies the weak
type (p, p) inequality for all characteristic functions of sets of finite measure,
then the operator is of restricted weak type (p, p) (see [7]).

In the sequel we will work with functions Φ and Ψ given by Φ(t) =

Rt Rt
0
a(s) ds and Ψ(t) = 0 b(s) ds for all t ≥ 0, where a and b be positive
continuous functions defined on [0, ∞).

2 Statement of the Theorems

In [4] the Hardy-Littlewood maximal function is studied in the torus T and,
under some assumptions on a and b, it is found that
Z Z
0 0
Φ(|M f |) ≤ C + C Ψ(C 0 |f |) for all f ∈ M(T) (1)
T T

if and only if
Z t
a(s)
ds ≤ Cb(Ct) for all t > 1 . (2)
1 s
In [2] the authors consider the maximal function in the context of spaces of
homogeneous type solving the problem under somehow more general assump-
tions on Φ and Ψ. Also, in [3] a similar problem is solved for the fractional
maximal function of order 0 < α ≤ 1, being the Hardy-Littlewood maximal
function a particular case.
The properties of M used to prove (2) implies (1), are only the weak type
(1, 1) and type (∞, ∞). In consequence it is easy to extend these results to
operators of weak type (p, p) with p > 1 and type (∞, ∞) as follows.
Restricted Weak Type (p, p) Operators in Orlicz Spaces 383

Theorem 2.1. Let T be of weak type (p, p) with p ≥ 1, and of type (∞, ∞).
If for some constant C, a and b satisfy
Z t
a(s)
tp−1 ds ≤ Cb(Ct) for all t > 1 (3)
1 sp

then, there exists a constant C 0 such that Ω Φ(|T f |) dµ ≤ C 0 +C 0 Ω Ψ(C 0 f ) dµ

R R

for all f ∈ D .
Remark 2. We notice that our assumptions on a and b allow us to obtain
Kolmogorov type inequalities.
A model operator which plays the role of M in this case is the maximal
Mp , acting on Lebesgue measurable functions on [0, 1] given by
 Z 1/p
1 p
Mp f (x) = sup f (4)
I∈I,x∈I |I| I

with I the family of all intervals contained on [0, 1].

For this operator we have the following theorem analogous to the results
on [4]. In particular, it says that condition (3) is sharp.
Theorem 2.2. Let p ≥ 1 and b monotone on [1, ∞). There exists a constant
C 0 such that
Z Z
Φ(Mp f ) ≤ C 0 + C 0 Ψ(C 0 |f |) for all f ∈ M([0, 1]) (5)
[0,1] [0,1]

if and only if (3) holds.

For p > 1 there exist operators which are of restricted weak type (p, p)
but not of weak type (p, p). Examples of these are the maximal operators
associated to Cesàro averages of order α with 0 < α < 1, defined for f ∈
M([0, 1]) by
Z c
+ 1
Mα f (x) = sup α
|f (s)|(c − s)α−1 ds for x ∈ [0, 1] (6)
x<c<1 (c − x) x

and
Z x
1
Mα− f (x) = sup α
|f (s)|(s − c)α−1 ds for x ∈ [0, 1]. (7)
0<c<x (x − c) c

which are known to be of restricted weak type (1/α, 1/α) but not of weak type
(1/α, 1/α).
384 B. Bongioanni, L. Forzani and E. Harboure

It is well known that if we start with an operator of weak type (p, p) or

restricted weak type (p, p), the Marcinkiewicz Interpolation Theorem gives the
same boundedness results for the intermediate spaces Lq with p < q < ∞.
One of the mainstay of this paper is to see what occurs when we are dealing
with the more general class of Orlicz spaces.
The results of the following theorems show that the properties of bound-
edness differ if we start with an operator of weak type (p, p) or with one of
restricted weak type (p, p), p > 1.
Theorem 2.3. Let T be of restricted weak type (p, p) with p > 1, and of type
(∞, ∞). If for some constant C, a and b satisfy
Z t 1/p Z ∞ 1/p0
a(s) −p0 /p
sup ds b(Cs) ds < ∞, (8)
t>1 1 sp t

then there exists a constant C 0 such that Ω Φ(|T f |) dµ ≤ C 0 +C 0 Ω Ψ(C 0 |f |) dµ

R R

for all f ∈ D .
We now introduce the linear
R x operator Hp with p > 1 defined for f ∈
1 1/p−1
M([0, 1]) by Hp f (x) = x1/p 0
f (s)s ds for x ∈ [0, 1]. It is easy to see
that the Hp operator is of restricted weak type (p, p) and of type (∞, ∞).
Remark 3. Note that if f is decreasing then Hp f is also decreasing; in fact,
if x and y are in [0, 1] and x < y, using that f is decreasing, we have
Z x Z y  1/p−1
1 1 x x x
Hp f (x) = 1/p f (s)s1/p−1 ds = 1/p f ( t) t dt
x 0 x 0 y y y
Z y Z y
1 x 1
= 1/p f ( t)t1/p−1 dt ≥ 1/p f (t)t1/p−1 dt = Hp f (y).
y 0 y y 0

Also, as it is easy to realize from its form, H1/α is related to Mα+ and Mα− .
The next theorem tells us that, as in the case of weak type, condition (8) for
a and b is sharp.
Theorem 2.4. Let p > 1 and b monotone on [1, ∞). There exists a constant
C 0 shuch that
Z Z
Φ(Hp f ) ≤ C 0 + C 0 Ψ(C 0 |f |) for all f ∈ M([0, 1]) (9)
[0,1] [0,1]

if and only if condition (8) holds.

We should mention that similar results to Theorem 2.4 were obtained in
[1] in terms of norm inequalities and under more restricted assumptions on Φ
and Ψ. From Theorem 2.4 we can derive the following consequence.
Restricted Weak Type (p, p) Operators in Orlicz Spaces 385

Corolary 2.5. Let 0 < α < 1 and b monotone on [1, ∞). There exists a
constant C 0 such that
Z Z
− 0 0
Φ(Mα f ) ≤ C + C Ψ(C 0 |f |) for all f ∈ M([0, 1]) (10)
[0,1] [0,1]

if and only if condition (8) holds with p = 1/α.

Proof. The operator Mα− is of restricted weak type (1/α, 1/α) and of type
(∞, ∞). Thus the sufficiency of condition (8) follows from Theorem 2.3. The
necessity follows from the fact that for f ∈ M([0, 1]) we have Mα− f (x) ≥
Hp f (x) for almost all x ∈ [0, 1], and the result is a consequence of Theorem
2.4.

The same is true for Mα+ since Mα+ f (x) = Mα− g(−x) with g(x) = f (1/2−x)
for all x ∈ [0, 1], and these two functions have the same distribution function.

Remark 4. It is not hard to find the largest spaces that are mapped into Lp .
In fact, if we thake for a(t) = tp−1 the best possible function b satisfying either
(3) or (8) we get Mp maps the space Lp log L into Lp , whereas Mα− and Mα+
map the space Lp (log L)p into Lp .

Remark 5. In particular, Remark 4 implies that condition (8) is strictly

stronger than (3). In fact, the pair a(t) = tp−1 and b(t) = tp−1 log(t + 1)
satisfies (8) but not (3).

3 Proofs of the Theorems

The proof of Theorem 2.1 requires the following lemma which tells us how to
control the size of the distribution of T f in terms of the distribution of f .

Lemma 3.1. Let T be an operator of weak type (p, p), p > 1, and of type
(∞, ∞) with constants A and B respectively. Then, for every function f in
the domain of T ,

(4Ap)p ∞ p−1
Z
µT f (t) ≤ s µf (s) ds for all t > 0
tp t/4B

Proof. Let f ∈ D be given. For t > 0 let us define


f (x) if |f (x)| > t/2B
f t (x) =
0 otherwise
386 B. Bongioanni, L. Forzani and E. Harboure

and ft = f − f t . Since T is sublinear

µ({|T f | > t}) ≤ µ({|T f t | > t/2}) + µ({|T ft | > t/2}). (11)

From the boundedness of T in L∞ we have |T ft (x)| ≤ t/2, which implies

µ({|T ft | > t/2}) = 0. On the other hand, using the weak type (p, p) of T , we
get  p
2A t
µ({|T f t | > t/2}) ≤ kf kp . (12)
t
Since µf t (s) = µf (t/2B) for s ∈ (0, t/2B), µf t ≤ µf and the fact that µf is
decreasing, we have
!
Z ∞ Z Z t/4B ∞
kf t kpp = p sp−1 µf t (s) ds = p + sp−1 µf t (s) ds
0 0 t/4B
Z t/4B Z ∞
p−1
= pµf (t/2B) s ds + p sp−1 µf t (s) ds
0 t/4B
Z t/2B Z ∞
≤ pµf (t/2B) p−1
s ds + p sp−1 µf (s) ds (13)
t/4B t/4B
Z t/2B Z ∞
≤p µf (s)sp−1 ds + p sp−1 µf (s) ds
t/4B t/4B
Z ∞
≤ 2p sp−1 µf (s) ds
t/4B

and this completes the proof.

Proof of Theorem 2.1. Let f be a function in D
Z Z ∞ Z 1 Z ∞ 
Φ(|T f |) dµ = a(t)µT f (t) dt = + a(t)µT f (t) dt
Ω 0 0 1
Z ∞
≤ µ(Ω)Φ(1) + a(t)µT f (t) dt.
1

From Lemma 3.1, Fubini’s Theorem and inequality (3) we get

Z ∞ Z ∞ !
(4Ap)p ∞ p−1
Z
a(t)µT f (t) dt ≤ a(t) s µf (s) ds dt
1 1 tp t/4B
Z ∞ Z 4Bt !
p p−1 a(t)
= (4Ap) µf (s) s dt ds
1/4B 1 tp
Restricted Weak Type (p, p) Operators in Orlicz Spaces 387

∞
C(4Ap)p
Z
≤ b(4BCs)µf (s) ds
(4B)p−1 1/4B
Z ∞
Ap p
≤( ) 4BC b(4BCs)µf (s) ds
B 0
Z
Ap
= ( )p Ψ(4BCf )dµ
B Ω

and the proof is complete.

Proof of Theorem 2.2. Since the operator Mp is simultaneously of weak

type (p, p) and of type (∞, ∞) the “only if” part follows from Theorem 2.1.
Suppose that (5) holds. Let t > 1 fixed and ft = tχ[0,1/tp ) being in
M([0, 1]).
 Z x 1/p  p Z x 1/p
1 p t
Mp ft (x) ≥ [ft (y)] dy = χ[0,1/tp ) dy
x x 0
( 0
t if x ∈ [0, 1/tp )
≥ 1
.
x1/p
if x ∈ (1/tp , 1)

Then for 1 < s < t,

λMp ft (s) = |{x : Mp ft (x) > s}|

1
≥ |{x ∈ (0, 1/tp ] : t > s} ∪ {x ∈ (1/tp , 1] : > s}|
x1/p
1
= |(0, 1/tp ] ∪ {x ∈ (1/tp , 1] : x < }|
sp
1
= |(0, 1/sp ]| = .
sp
Therefore it follows that
Z 1 Z ∞
Φ(|Mp ft |) = a(s)λMp ft (s) ds
0 0
Z t Z t
(14)
a(s)
≥ a(s)λMp ft (s) ds ≥ ds.
1 1 sp

b(s)
On the other hand, let L = lim inf . If this limit is zero it is easy to
s→∞ sp−1
Ψ
see that there is a function f in L that Mp f = ∞ everywhere in [0, 1]. In
Ψ(s)
fact, since b is monotone we have also lim inf p = 0 and then we can choose
s→∞ s
388 B. Bongioanni, L. Forzani and E. Harboure

Ψ(tn )
an increasing sequence of numbers tn , n ≥ 1, such that tn > 2n and <
tpn
∞
1 X hP
n−1 1 Pn 1

. Then the function f = t n χI n
, with In = k=1 p ,
tn k=1 p
tn
has
2n n=1
b(s)
this property. So we only need to consider b such that lim inf p−1 > 0
s→∞ s
( since b is monotone it must be nondecresing on [1, ∞). Since λft (s) =
and
1/tp if 0 < s < t
we have
0 if s ≥ t,
1 ∞
C0 t
Z Z Z
Ψ(C 0 |ft |) = C 0 b(C 0 s)λft (s) ds = p b(C 0 s) ds
0 0 t 0
(15)
C0 t C 0 b(C 0 t)
Z
0
≤ C Ψ(1) + p b(C s) ds ≤ C 0 Ψ(1) +
0
.
t 1 tp−1
b(s) b(s)
Now since lim inf = L > 0, there exists s0 such that M ≤ p−1 for
sp−1
s→∞ s
L
s > s0 , with M = 1 if L = ∞ and M = when L is finite. Then from (14)
2
and (15) we have
Z t Z 1 Z 1
a(s) 0 0
ds ≤ Φ(|Mp ft |) ≤ C + C Ψ(C 0 |ft |)
1 sp 0 0
C 02 b(C 0 t)
≤ C 0 + C 02 Ψ(1) +
tp−1
C + C Ψ(1) b(s0 t) C 02 b(C 0 t)
0 02
≤ + ≤ Ct1−p b(Ct).
M s0p−1 tp−1 tp−1
( )
0 02 C 0 + C 02 Ψ(1)
with C = max C , s0 , C + . Since C is independent of t, (3)
M s0p−1
follows.
To prove Theorem 2.3 we need the analogous to Lemma 3.1.
Lemma 3.2. Let T be an operator of restricted weak type (p, p) and of type
(∞, ∞) with constants A and B respectively. Then, for every function f in
the domain of T ,
" #p
4A ∞
Z
1/p
µT f (t) ≤ µf (s) ds for all t > 0 (16)
t t/4B
Restricted Weak Type (p, p) Operators in Orlicz Spaces 389

Proof. Let f ∈ D and t > 0. We define f t and ft as in the proof of

Lemma 3.1. Then we have µ({|T ft | > t/2}) = 0. Since T is of restricted weak
type, µf t (s) = µf (t/2B) for s ∈ (0, t/2B), µtf ≤ µf and the fact that µf is
decreasing, we have
p
2A ∞
 Z
t 1/p
µ({|T f | > t/2}) ≤ µf t (s) ds
t 0
 p " Z t/4B Z ∞ ! #p
2A 1/p
= + µf t (s) ds
t 0 t/4B
 p " Z ∞ #p
2A t 1/p 1/p
= µf (t/2B) + µf t (s) ds
t 4B t/4B
 p "Z t/2B Z ∞ #p
2A 1/p 1/p
≤ µf (s) ds + µf (s) ds
t t/4B t/4B
" #p
4A ∞
Z
≤ µf (s)1/p ds
t t/4B

Proof of Theorem 2.3. This proof resembles that of Theorem 1 in [5].

Suppose (8) holds, that is, there exists a constant D such that for all t > 1,
Z t 1/p Z ∞ 1/p0
a(s) −p0 /p
ds b(Cs) ds ≤ D. (17)
1 sp t

Let f be a function in the domain of T . From Lemma 3.2,

Z Z ∞ Z 1 Z ∞ 
Φ(|T f |) dµ = a(s)µT f (s) ds ≤ + a(s)µT f (s) ds
Ω 0 0 1
Z ∞ " Z ∞ #p
4
≤ Φ(1)µ(Ω) + Ap a(s) µf (t)1/p dt ds
1 s s/4B
 p Z ∞  Z ∞ p
A 1
= Φ(1)µ(Ω) + a(s) µf (t/4B)1/p dt ds.
B 1 s s
Z ∞ 1/pp0
−p0 /p
Now, if we call h(t) = b(Cr) dr and g(t) = µf (t/4B)1/p we
t
have by Hölder’s inequality and Fubini’s Theorem
Z ∞  Z ∞ p
1
a(s) g(t) dt ds
1 s s
390 B. Bongioanni, L. Forzani and E. Harboure

Z ∞  Z ∞ p
1 1
= a(s) g(t)h(t)b(Ct)1/p dt ds
1 s s h(t)b(Ct)1/p
Z ∞ Z ∞  Z ∞ p/p0
a(s) p −p0 /p −p0
≤ (g(t)h(t)) b(Ct) dt b(Cr) h(r) dr ds
1 sp s s
Z ∞ (Z
t Z ∞ p/p0 )
p a(s) −p0 /p −p0
= [g(t)h(t)] b(Ct) b(Cr) h(r) dr ds dt.
1 1 sp s

Since integration by parts yields

Z ∞ Z ∞ 1/p0
−p0 /p −p0 0 −p0 /p
b(Cr) h(r) dr = p b(Cr) dr
s s

and
Z t Z s −1/p0 Z t 1/p
a(s) a(r) a(r)
dr ds = p dr ,
1 sp 1 rp 1 rp
using inequality (17) twice we have
Z ∞
(Z
t Z ∞ p/p0 )
p a(s) −p0 /p −p0
[g(t)h(t)] b(Ct) b(Cr) h(r) dr ds dt
1 1 sp s
Z ∞ Z ∞ (Z
tp/(p0 )2 )
0 p/p0 a(s) p −p0 /p
= (p ) [g(t)h(t)] b(Ct) b(Cr) dr ds dt
1 1 sp s
Z ∞ (Z
t Z s −1/p0 )
0 p/p0 p/p0 p a(s) a(r)
≤ (p ) D [g(t)h(t)] b(Ct) dr ds dt
1 1 sp 1 rp
Z ∞ Z t 1/p
0 0 a(r)
≤ p(p0 )p/p Dp/p [g(t)h(t)]p b(Ct) dr dt
1 1 rp
Z ∞ Z ∞ −1/p0
0 p/p0 p p −p0 /p
≤ p(p ) D [g(t)h(t)] b(Ct) b(Cr) dr dt
1 t
∞ 0
p(p0 )p/p Dp
Z Z
0
≤ p(p0 )p/p Dp g(t)p b(Ct) dt ≤ Ψ(4BCf ) dµ
1 C Ω

and this completes the proof.

Proof of Theorem 2.4. Since the operator Hp is simultaneously of restricted

weak type (p, p) and of type (∞, ∞), from Theorem 2.3 the “only if” part is
done.
Restricted Weak Type (p, p) Operators in Orlicz Spaces 391

To prove the “if” part, suppose that (9) holds. We first assume that b has
the property Z ∞
0
b(s)−p /p ds < ∞. (18)
1
Let t > 1 fixed. For s > 0 let
0
ht (s) = At b(Cs)−p
h 0 R∞ 0
i−p
with At = tb(Ct)−p /p + t b(Cs)−p /p ds and C > max{(C 0 )2 , 1} such
R∞ 0 0
that 1 b(Cs)−p /p < (C 0 )−p /p . Observe thatin this case the monotonicity
of b and condition (18) imply that b is increasing and lim b(s) = ∞. Then,
s→∞
ht is decreasing, lims→∞ ht (s) = 0 and h−1t (s) is well defined for s > 0. Now
consider ft ∈ M([0, 1]) defined by ft = h−1 t χ(0,yt ) , with yt = min{ht (t), 1}.
The distribution function of ft is for s > 0,
λft (s) = |{x ∈ (0, 1] : ft (x) > s}|
= |{x ∈ (0, 1] : h−1
t (x) > s and x < yt }|
= |{x ∈ (0, 1] : x < ht (s) and x < yt }|
= min{ht (s), ht (t), 1}.
From this and the fact that b is increasing we get
Z Z ∞
0 0 02
C Ψ(C |ft |)dλ = C b(C 0 s)λft (s) ds
[0,1] 0
 Z t Z ∞ 
≤ C ht (t) b(Cs) ds + b(Cs)ht (s) ds
0 t
 Z ∞ 
≤ C tb(Ct)ht (t) + b(Cs)ht (s) ds
t
 Z ∞ 
0 0
≤ CAt tb(Ct)−p /p + b(Cs)−p /p ds
t
 Z ∞ −p0 /p
0
−p0 /p
≤ C tb(Ct)−p /p + b(Cs) ds
t
Z ∞ −p/p0
−p0 /p
≤C b(Cr) dr .
t

Thus, by the choice of C we have

Z Z ∞ −p/p0
0 0 0 −p0 /p
C +C Ψ(C |ft |) ≤ 2C b(Cr) dr . (19)
[0,1] t
392 B. Bongioanni, L. Forzani and E. Harboure

On the other hand we will see that

1
λHp ft (s) ≥ for all s ∈ (1, t), (20)
sp
and therefore
Z Z ∞ Z t
a(s)
Φ(|Hp ft |)dλ = a(s)λHp ft (s) ds ≥ ds. (21)
[0,1] 0 1 sp

Thus from (21) and (19) we have

Z t Z Z
a(s) 0 0
ds ≤ Φ(|Hp ft |) ≤ C + C Ψ(C 0 |ft |)
1 sp [0,1] [0,1]
Z ∞ −p/p0
−p0 /p
≤ 2C b(Cr) dr .
t

Since C does not depends on t, we get (8).

It remains to prove (20). Since lims→∞ ht (s) = 0, we have limx→0 ft (x) =
∞ and limx→0 Hp ft (x) = ∞ (for all g decreasing, Hp g ≥ g). Also Hp ft
is continuous and decreasing (see Remark 3) on (0, 1]. Then the image of
Hp ft is the interval [Hp ft (1), ∞). For Hp ft (1) < s < t, we have λHp ft (s) =
|{x : Hp ft (x) > s}| = xs with xs ∈ Z (0, 1] such that s = Hp ft (xs ) =
Z xs xs p
1 1/p−1 1 1/p−1
f t (x)x dx. Then x s = f t (x)x dx .
xs
1/p
0 sp 0
We have Hp ft (1) < t = h−1 t (ht (t)) = ft (ht (t)) ≤ Hp ft (ht (t)). Since
Hp ft (1) < Hp ft (ht (t)) and Hp ft is decreasing, ht (t) ≤ 1. Thus, yt = ht (t)
and xs = λHp ft (s) ≥ λHp ft (t) ≥ λft (t) = yt . Therefore
Z xs Z yt Z ht (t)
ft (x)x1/p−1 dx ≥ ft (x)x1/p−1 dx = h−1
t (x)x
1/p−1
dx
0 0 0
Z (ht (t))1/p Z ∞
= h−1 p
t (y )dy = t(ht (t)) 1/p
+ (ht (r))1/p dr
0 t
 Z ∞ 
1/p −p0 /p −p0 /p
= At tb(Ct) + b(Cr) dr = 1
t

1
Then xs ≥ p if Hp ft (1) < s < t. If 1 < s < Hp ft (1), obviously λHp ft (s) =
s
1
1 > p and we get (20). To finish the proof of Theorem 2.4 it remains to
s
consider the case when Z ∞
0
b(s)−p /p ds = ∞. (22)
1
Restricted Weak Type (p, p) Operators in Orlicz Spaces 393

We will show that, in this situation, Hp does not map LΨ ([0, 1]) into LΦ ([0, 1]).
We may suppose that b is increasing. If b were not increasing, we could
take b ≤ b̃ increasing satisfying (22) and it is enough to do the following
construction with b̃ instead of b since LΨ̃ ⊃ LΨ .
Consider the function f = h−1 χ[0,1] in M([0, 1]), where

0
Kb(x)−p
h(x) = R p
x
1/2
b−p0 /p ds

for x ≥ 1 and K such that h(1) = 1. Note that h is Rdecreasing and so f is well
∞ 0 0
defined. First we see that f is in LΨ ([0, 1]). Since 1 b−p /p = ∞ and b−p /p
∞
R xn −p0 /p on [1, ∞), there exists a sequence {xn }n=1 with
is continuous and decreasing
xn > 1, such that 1 b = n and limn→∞ xn = ∞. Then, from the fact
that λf (s) = h(s) for s > 1, we have

Z Z ∞ Z 1 Z ∞
Ψ(|f |) = b(s)λf (s) ds ≤ b(s) ds + b(s)h(s) ds
[0,1] 0 0 1

and

∞ ∞ 0
b(s)−p /p
Z Z
1
b(s)h(s) ds = R p ds
K 1 1
s
b−p0 /p
1/2
x1 0 ∞ Z xn+1 0
b(s)−p /p b(s)−p /p
Z X
≤ p ds + R xn
p ds
b−p0 /p
R
1
1
1/2
b−p0 /p n=1 xn 1

∞ R xn+1 −p0 /p Rx 0
1 X
1
b − 1 n b−p /p
= R p + R xn p
b−p0 /p


1
b−p0 /p
1/2 n=1 1
∞
1 X 1
= R p + p
< ∞.
1 0
b−p /p n=1
n
1/2

Now we will see that Hp f is not in LΦ ([0, 1]) by showing that Hp f (x) = ∞
for all x ∈ [0, 1]. Since Hp f is decreasing on [0, 1], it is enough to show
394 B. Bongioanni, L. Forzani and E. Harboure

Hp f (1) = ∞. In fact,
Z 1 Z ∞
1 1 −1 1/p−1 1
Hp f (1) = 1/p h (r)r dr ≥ 1/p h(r)1/p dr
K 1/p K 0 K 1
Z ∞ 0 ∞ Z xn+1 0
b(s)−p /p X b(s)−p /p
= Rs ds = s ds
b−p0 /p b−p0 /p
R
1 1/2 n=0 xn 1/2
∞ Z xn+1 0 ∞ R xn+1 −p0 /p Rx 0
X b(s)−p /p X
1
b − 1 n b−p /p
≥ R xn+1 ds = R1 Rx
n=0 xn 1/2
b−p0 /p n=0 b−p0 /p + n+1 b−p0 /p
1/2 1
∞
X 1
= R1 = ∞.
b −p0 /p +1+n
n=0 1/2

References
[1] A. Cianchi, Hardy inequalities in Orlicz Spaces, Trans. AMS, 351, No.
6 (1999), 2456–2478.

[2] I. Genebashvili, A. Gogatishvili, V. Kokilashvili and M. Krbec, Weight

theory for integral transforms on spaces of homogeneous type, Pitman
Monographs and Survey in Pure and Applied Mathematics 92, Longman,
Harlow, 1998.
[3] E. Harboure, O. Salinas and B. Viviani, Orlicz Boundedness for Certain
Classical Operators, Colloquium Mathematicum 2002, 91 (No. 2), 263–
282.
[4] Hiro-O Kita, On Maximal Functions in Orlicz Spaces, Proc. AMS 124,
(1996), 3019–3025.
[5] B. Muckenhoupt, Hardy Inequality with weights, Studia Mathematica, T.
XLIV. (1972), 31–38.
[6] M. M. Rao and Z. D. Ren, Orlicz Spaces, Marcel Dekker Inc., New York,
Basel, Hong Kong, 1991.
[7] E. M. Stein and Guido Weiss, Fourier Analysis on Euclidean Spaces,
Princenton University Press, 1971.