Sei sulla pagina 1di 29

NONLINEAR POTENTIAL ANALYSIS ON

MORREY SPACES AND THEIR CAPACITIES

D.R. Adams and J. Xiao

Dedi ated to the memory of Thomas H. Wol

Abstra t. To study the existen e and regularity, even just partial regularity (smoothness ex-
ept for a losed ex eptional set of some measure or apa ity zero) for higher order (nonlinear)
ellipti equations (systems), we intend to develop a nonlinear potential analysis on Morrey
spa es, fra tional Riesz potentials, fra tional maximal fun tions, and Morrey apa ities of two
di erent types.

1. Introdu tion
As early as 1938, C.B. Morrey had introdu ed an Lp -growth ondition on the gradient
of solutions of ertain partial di erential equations (PDE's) in two spa e dimension that
insured their ontinuity in their domain of existen e. This ondition, whi h now bares
his name, has be ome a well known and highly useful tool for studying existen e and
regularity, even just partial regularity (smoothness ex ept for a losed ex eptional set of
some measure or apa ity zero) for single higher order ellipti equations (e.g. biharmoni
equation) or se ond order nonlinear ellipti systems.
Sin e su h solutions an usually be represented as a potential { perhaps as a Riesz
or modi ed Riesz potential of their derivatives { it is lear that a systemati study of
potentials of fun tions that satisfy Morrey's ondition would be of interest in PDE resear h,
the purpose of thisp note (motivated by the su essp of the study of potentials of fun tions
that belong to L , the so alled nonlinear or L potential theory) is to begin su h an
investigation. Some of the elements are already in the literature, espe ially beginning with
the early systemati work of G. Stampa hia [S℄ and S. Campanato [C℄ in the mid 1960's;
see also [P℄. A lemma that has also been of some use (Lemma 4.3 below, taken from [A1℄)
is a version of the lassi al Sobolev Embedding Theorem for Morrey lass potentials rather
than the usual Lebesgue lass potentials.
1991 Mathemati s Subje t Classi ation: Primary 42A16, 46E30, 47B38.
Key words and phrases: Nonlinear potentials, Morrey spa es, Morrey apa ities, CSI.

Typeset by AMS-TEX
1
2 D.R. ADAMS AND J. XIAO

Here, we intend to develop some of the analogues of the Lp -potential theory in luding:
(i) a study of the pre-dual spa e of a Morrey spa e; (ii) a study of the equivalen e of
the Riesz potential of a Borel measure versus the fra tional maximal fun tion of the same
measure, now in the Morrey norm; and (iii) to begin to develop a theory of apa ities that
is naturally asso iated with the potentials of fun tions from a Morrey spa e or from its
pre-dual spa e.
There is prior work on pre-dual spa es for a xed Morrey spa e Lp; , due to C.T. Zorko
[Z℄, the Z q; spa e, and E.A. Kalita [K℄, the spa e K q;. In this note, we produ e a third
spa e, H q;, pre-dual to Lp; and then we demonstrate that all three predual spa es oin ide
(with equivalent norms); Theorem 3.3. Furthermore, it is this versatility of pre-dual spa es
that really fa ilitates all the al ulations involving the apa ities of Se tions 5, 6 and 7.
The equivalen e of the Morrey norms of the Riesz potential of a Borel measure and
the fra tional maximal fun tion of the same measure is an extension of an earlier result of
Mu kenhoupt-Wheeden [MuW℄; see Theorem 4.2 below. The 1974 M-W result orresponds
to  = n in our result. It is the equivalen e for  < n that allows us to estimate the
apa ity C (; H q;) on a ball B (x; r) from below as the radius r tends to zero; Theorem
6.4. Here we nd an analogue of the Wol potential from the standard Lp -theory; see also
[AH, p. 110℄.
The theory of apa ities for potentials of fun tions of either the Morrey spa e Lp; or its
pre-dual H q;, q = p=(p 1); p > 1, is developed in Se tions 5 and 6 below; see de nitions
for C (; Lp;) in Se tion 5 and C (; H q;) in Se tion 6. We note here that one is based
on a dual spa e, Lp;, and the other on a spa e, H q;, not ne essarily a dual spa e. If we
represent the rst by a general symbol C (; X ) and the se ond by C (; X ), we now propose
to lassify apa ities like C (; X ) as type I apa ities, and C (; X ) as type II apa ities.
Note that if the spa e X is re exive, then C (; X ) will be both of type I and type II.
Moreover,1
a standard apa ity
1
in lassi al potential theory an nbe onsidered as type I, with
X = L  M; where L =the set of all Radon measures on R with nite total variation,
and M = the set of all Radon measures (lo ally nite regular signed Borel measures) on R n .
The orresponding type II apa ity is rather uninteresting (based loosely on X  = L1 ).
Further, we show below (Theorem 5.1 and Corollary 6.2) that C (; X )  Cap(; X ), and
C (; X )  Cap(; X ),  denoting \having the same null sets". Here Cap represents the
usual \dual apa ity" { it is usually based on the potentials of measures in M.
Finally, we investigate apa ity strong-type inequalities (CSI) for our apa ities
C (; H q;) and C (; Lp;). We get a very strong version for the type I ase (Theorem
7.4) but no CSIp for the type II ase (Example 5.4). Note that the CSI for C (; Lp),
the standard L -Riesz apa ity (see [AH℄), is intermediate between these two extreme
situations. Putting everything into Choquet-Lorentz notation (Remark 5.6), the rst ase
is I H q;  L(q;1)(C (; H q;)), whenever 0 < <  < n and = < q < 1; the se ond is
I Lp; 6 L(p; ) (C (; Lp;)) for ; 2 (0; n), 1 < p < = , and any 0 < < 1.
We would like to thank S.M. Elery and R. Hurri-Syrjanen for their typing assistan e.
2. Morrey Spa es
2.1. De nition. Let 1 < p < 1, 0    n, and f 2 Lplo (R n ). Then we shall say that f
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 3

belongs to the Morrey spa e Lp; provided


 Z 1=p
kf kLp; = sup r n jf jp < 1;
r>0;x2Rn B(x;r)
where B (x; r)  R n stands for the ball entered at x and with radius r.
Unless a spe ial remark is made, the di erential element dx is omitted when the integrals
under onsideration are the Lebesgue integrals. Clearly, k  kLp; is a norm. When  = n,
then L = L , the lassi al L -spa e on R .
p; p p n
In the ase of p = 1, f is allowed to be a measure on R n and the1;Morrey spa e is linked
with Hausdor apa ity and its Choquet integral. More pre isely, L onsists of all Radon
measures (lo ally nite regular signed Borel measures)  on R satisfying
n

kkL ; = sup n r n jj(B (x; r)) < 1;


1
r>0;x2R

where jj is the total variation measure of .


2.2 Proposition [A3, Proposition 1℄. The predual spa e of L1; is given by L1 ((n1) ) whi h
onsists of (n1) -quasi ontinuous fun tions u on R n for that
Z Z 1
kukL (n1 ) = j (1)
ujdn  =  (1) (fx 2 R n : ju(x)j  tg)dt < 1:
0 n 
1 ( )
Rn
In parti ular, the duality implies
Z

n !d
 k!kL (n1 ) kkL ; :
1 ( ) 1
R

Here and hen eforth, (d1) , 0 < d  n, denotes the d-dimensional Hausdor apa ity,
that is, X
(d1) (E ) = inf rjd ;
where the in mum is taken over all ountable overings of E  R n by open balls of radius
rj .
The following is a new hara terization of the Morrey spa e:
2.3. Theorem. Let 1 < p < 1 and 0 <  < n. Then
Z 1=p
kf kLp; = sup jf j ! ;
p (2.1)
! Rn
where the supremum is taken over all nonnegative fun tions ! on R n with
k!kL (n1)  1:
1 ( ) (2.2)
4 D.R. ADAMS AND J. XIAO

Proof. From Proposition 2.2, it follows that


Z
jf jp !  k! kL (n1 ) kf kpLp;  kf kpLp;
1 ( )
R n

for k!kL (n1 )  1. Consequently, the right-hand side of (2.1) is not greater than the
1 ( )

left-hand side of (2.1).


On the other hand, let x0 2 R n and r0 > 0 and
!0 = 1B(x ;r ) r0 n : 0 0

Then Z
k!0kL (n1) = r0 n d(n1) = 1:
( )
B(x ;r )
1
0 0

Therefore !0 satis es (2.2). In the meantime,


Z 1=p
kf kLp; = sup jf jpr0 n 1B(x ;r )
r0 >0;x0 2Rn
0 0
Rn
Z 1=p
= sup jf jp!0
r0 >0;x0 2Rn R n
Z 1=p
 sup jf j !
p
! Rn
where the supremum is taken over any nonnegative ! on Rn with (2.2). The proof is
omplete.
More importantly, (2.1) leads to a onsideration of new spa e des ribing the predual
spa e of a Morrey spa e. For q = p=(p 1), p 2 (1; 1) and 0 <  < n, we say that g is in
H q; if Z 1=q
kgkH q; = inf jgj !
q 1 q < 1; (2.3)
! Rn
where the in mum is over all nonnegative fun tions ! on R n satisfying (2.2).
2.4 Theorem. Let 1 < p < 1; q = p=(p 1) and 0 <  < n. Then the pre-dual spa e of
L p;is H q; under the following pairing:
Z
hf; gi = fg:
Rn
Moreover Z

kf kLp; = sup
fg ; f 2 Lp;; (2.4)
g Rn
where the supremum is taken over all fun tions g 2 H q; with kg kH q;  1.
Proof. If f 2 Lp; and g 2 H q; , then an appli ation of Holder's inequality yields
Z Z 1=p  Z 1=q

hf; gi 
jf kgj!1=p! 1=p  jf j !
p jgjq !1 q
Rn Rn R n
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 5

sin e if g 2 H q;, then take ! su h that RRn jgjq !1 q < 1. Then it follows that jgj! 1=p is
nite almost everywhere. This implies that g must be zero a.e. ! is zero. Thus,

hf; gi  kf kLp; kgkH q; :
In parti ular, every fun tion f 2 Lp; indu es a bounded linear fun tional on H q;.
Conversely, suppose L isn a bounded linear fun tional on H q; with theq norm kLk < 1.
Fix a ball B (x0; r0)  R . If g is supported in B (x0; r0) and f 2 L (B (x0; r0)) (the
q -Lebesgue integrable spa e on B (x0 ; r0 )), then
Z !1=q
kgkH q; = inf
!
jgjq !1 q :
B(x0 ;r0 )

Now, let !0 = 1B(x ;r ) (x)r0 n . Then


0 0

Z !1=q
kgkH q;  r0( n)(1 q) jgjq :
B(x0 ;r0 )

Hen e L indu es a bounded linear fun tional on Lq (B (x0; r0)), and a ts with some fun tion
f B 2 Lp (B (x0; r0 )). By taking Bj = B (0; j ), j = 1; 2; 3; :::, we have f Bj = f BjR on Bj , so +1

we get a single fun tion f on R n that is lo ally in Lp, and su h that L(g) = Rn fg when
g 2 H q; with support in some ball of R n . If g = 1B(x ;r ) jf jp f 1 then 0 0

Z Z !1=q
jL(g)j = jf jp  kLk r0( n)(1 q) jf j(p 1)q :
B(x0 ;r0 ) B(x0 ;r0 )

This implies that f 2 Lp; with


kf kLp;  kLk:
Clearly, (2.4) is a dire t onsequen e of the previous argument.
3. Equivalent Predual Spa es
For p > 1 there are already two hara terizations of the predual spa e of a Morrey spa e
in the literature. First, in 1986, C.T. Zorko essentially proved the following theorem.
3.1 Theorem [Z, Proposition 5℄. Let p 2 (1; 1), 1=p + 1=q = 1 and  2 (0; n). Then R a
predual spa e of Lp; is Z q; in the following sense: if g 2 Lp; and f 2 Z q; , then Rn fg
is an element of (Z q; ) . Moreover, for any L 2 (Z q; ) , there exists g 2 Lp; su h that
Z
L(f ) = fg; f 2 Z q; :
Rn
6 D.R. ADAMS AND J. XIAO

The spa e Z q; is de ned by the set of all fun tions f on R n with the norm
n X o
kf kZ q; = inf kf k gkl : f = 1 k ak < 1;
k

where ak is a (q; n )-atom and kf k gk = Pk j k j < 1, and the in mum is takenn over
all possible atomi de ompositions of f . Additionally, we say that a fun tion a on R is an
(q; n )-atom provided that a is supported on a ball B  R n and satis es
kakq  1 ; 1=p + 1=q = 1:
jB j(n )=(np)
Se ond, in 1998, E.A. Kalita obtained another des ription of the predual spa e of a
Morrey spa e as follows.
3.2 Theorem [K, Theorem 1℄. Let p 2 (1; 1), 1=p + 1=q = 1 and  2 (0; n). Then R a
predual spa e of Lp; is K q; in the following sense: if g 2 Lp; and f 2 K q; , then Rn fg
is an element of (K q;) . Moreover, for any L 2 (K q; ) , there exists g 2 Lp; su h that
Z
L(f ) = fg; f 2 K q;:
Rn

The K q; onsists of all fun tions f on R n with the quasi-norm


Z 1=q
kf kK q; = inf n
jf jq !1 q ;
R
where Z
! (x) = r (n ) 1R1+ (r jx yj)d(y; r);
Rn+1
+

and where the in mum is taken over all  2 M+ (R n++1 ) (the lass of all nonnegative Radon
measures on the upper half spa e Rn++1 = f(x; r) : x 2 R n ; r > 0g) with normalization
 (R n++1 ) = 1. Here and afterwards, 1E is the hara teristi fun tion of the set E .
Note that throughout we will often use the notation A  B to denote omparability of
the quantities, i.e., there are two nite positive onstants 1 and 2 (independent of A and
B ) satisfying 1 B  A  2 B . Similarly, we say that A & B resp. A . B if only the rst
inequality resp. the se ond inequality holds.
Therefore, a very natural question arises: What is the relationship between those three
predual spa es? Below is the answer.
3.3 Theorem. Let q = p=(p 1), p 2 (1; 1) and  2 (0; n). Then Z q; = K q; = H q;
with
k  kZ q;  k  kK q;  k  kH q; : (3.1)
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 7

Proof. In what follows, we always make the following onvention: 1=p +1=q = 1, p 2 (1; 1)
and  2 (0; n).
We prove Theorem 3.3 by verifying
K q;  H q;  Z q;  K q; :
Step 1. K q;  H q;R.
Note that if !(x) = jy xj<r r (n ) d(y; r) where  is as above, then by the Corollary
to Proposition 1 in [A3℄, it follows that
Z Z
k!kL (n1) .
1 ( )
n+1
r (n ) 1R (jx
1 y j < r)d(n1) (x)d (y; r) . 1:
R+ R n +

In other words, K q; is ontained in H q; with


kf kH q; . kf kK q; ; f 2 K q;: (3.2)
Step 2. H q;  Z q;.
Let f 2 H q;. Consider Ek = fx 2 R n : !(x) > 2k g, k 2 Z. Then
X
k!kL (n1) 
1 ( ) 2k (n1) (Ek ):
k

Sin e f 2 H q;, there is an ! su h that


Z 1=q
jf jq !1 q  2kf kH q; ;
Rn
and
k!kL (n1)  1:
1 ( )

For ea h Ek there exists a ube overing of Ek , say Jj;k su h that


X
`(Jj;k )n  < 2(n1) (Ek ):
j

Here and hereafter, `(J ) denotes the edge length of a ube J of sides parallel to the axes
of R n .
Now, we x k 2 Z and onsider S the olle tion I of all dyadi ubes I in R whi h are
n
unions of thePJj;k , namely I = fJj;k : Jj;kP
 I g. Note that I need not be one of the
Jj;k . Sin e j `(Jj;k ) n  < 1, we have j `(Jj;k )n < 1; onsequently the unions of
the Jj;k annot form arbitrarily large dyadi ubes, so ea h Jj;k must be in luded in some
maximal S dyadi S ube I 2 I . Enumerate these maximal ubes as a sequen e fIm;k gm=1 .
1
Then j Jj;k = m Im;k and the interiors of any two Im;k 's are disjoint. Noti ing
8 D.R. ADAMS AND J. XIAO

X
`(Im;k )n  `(Jj;k )n
Jj;k Im;k
as well as 0 <  < n, we nd that X
`(Im;k )n   `(Jj;k )n  :
Jj;k Im;k
Be ause ea h Jj;k is ontained in a unique Im;k , we have
X X X X
`(Im;k )n   `(Jj;k )n  = `(Im;k )n ;
m m Jj;k Im;k m
and X
`(Ij;k )n  < 2(n1) (Ek ):
j
S
Sin e R n = k Ek , it follows that R
n = Sk;j Ij;k . Upon de ning
[
j;k = Ij;k n Il;k+1 ;
l
we see Rn = [j;k j;k , and with this, we obtain f = Pj;k j;k aj;k , where
Z 1=q
j;k = `(Ij;k ) (n )=p jf jq
j;k
and Z  1=q
aj;k = f 1j;k `(Ij;k ) (n )=p jf jq :
j;k
It is easy to he k that ea h aj;k is a (q; n )-atom. To prove that f 2 Z q;, it remains to
verify that f j;k g is l1 -summable. For this, noting the following two fa ts: k!kL (n1)  1 1 ( )

and !(x)  2k+1 as x 2 j;k (in fa t, x 2 j;k implies x 2= Sl Ij;k+1, then x 2= Ek+1 and
hen e !(x)  2k+1), we apply Holder's inequality to get
X Z 1=q
kf j;k gkl .
1 `(Ij;k )(n )=p 2k=p jf jq !1 q
j;k j;k
X 1=p  X Z 1=q
. 2k `(I j;k )n  jf j!1 q
j;k j;k j;k
Z
X X 1=p  X 1=q
. 2k `(Ij;k ) n  jf j!1 q
k j j;k j;k
X 1=p  X Z 1=q
. 2k (n1) (Ek ) jf j!1 q
k j;k j;k
Z 1=q
. k!k1L=p( 1 )
1 ( )
n
jf j! 1 q
n  R
. kf kH q; :
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 9

Consequently,
kf kZ q; . kf kH q; ; f 2 Z q;: (3.3)
Step 3. Z q;  K q;.
Suppose that f 2 Z q;. So f = Pi j aj with f j g 2 l1 and ea h aj is a (q; n )-atom.
Assume that xj and rj are the enter and radius of the support ball Bj of aj , respe tively.
De ne the following two fun tions:
Aj (x) = maxfrj ; jx xj jg;
and
1 X j j jrj
! (x) = kf j gkl n + ;  > 0:
j Aj (x)
1

It is lear that ! an be written as that integral form required for the de nition of K q;.
A further appli ation of Holder's inequality implies
X
( 
n )  q 1 X  1 ja jq :
jf jq  j j jrj Aj
 j j j rj  Ajn + q j
j j
Thus
Z X Z
jf jq !1 q q
 kf j gkl1 1 j j jrj (q 1) Aj(n +)(q 1) ja jq
j
Rn j Rn
X Z
= kf j gklq1 1 j j jrj (q 1) Aj(n +)(q 1) ja jq
j
j Bj
X Z
q
 kf j gkl1 1 (
j j jrj n )(q 1) jaj jq
j Rn
X
. kf j gklq1 1 j j jrj(n )(q 1)rj (n )(q 1)
j
q
. kf j gkl1 :
That is to say,
kf kK q; . kf kZ q; ; f 2 Z q;: (3.4)
Clearly, (3.1) is a dire t onsequen e of (3.2), (3.3) and (3.4).
P
3.4 Remark. Note that kf kL ( 1 ) < 1 if and only if f = j j aj where f j g 2 l1 and
1 ( )
n 
aj is a (1; n )-atom. Here a fun tion a on R n is alled (1; n )-atom if the support
of a is ontained in a ball B  R n and kak1  jB j(n )=n .
On the one hand, if f has su h a de omposition, then the sublinearity of the Choquet
integral with respe t to dyadi Hausdor apa ity ( f. [A3℄) yields
Z X X Z
kf kL (n1) 
1 ( ) j j kaj jd(n1) . j j j jaj jd(n1) . kf j gkl : 1
Rn j j R n
10 D.R. ADAMS AND J. XIAO

But, one should observe that f is not ne essarily in the spa e L1((n1) ) sin e those
(1; n )-atoms may not be (n1) -quasi- ontinuous, see also [A3, p. 123℄. This la k of
quasi- ontinuity, however, does not hinder the appli ations of the atomi de omposition. If
we want to restore the quasi- ontinuity
P of atoms, we an modify the onstru tion above,
and obtain the majorization jf j  j j j kaj j, whi h for most purposes is a satisfa tory
substitute.
Conversely, suppose kf kL (n1 ) < 1. Using the argument for Theorem 3.3, we onsider
1 ( )

j;k = `(Ij;k )n  2k+1 ;


and
aj;k = f 1j;k `(Ij;k ) (n ) 2 (k+1) :
Then f = Pj;k j;k aj;k . It is obvious that aj;k is a (1; n )-atom, sin e jf (x)j  2k+1
whenever x 2 j;k , and f j;k g belongs to l1 sin e
X
kf j;k gkl .
1 `(Ij;k )n  2k+1 . kf kL1 ((n1) ) :
j;k

4. Riesz Potentials and Maximal Fun tions with Fra tional Orders
Re all that I   and M  denote the fra tional Riesz potential and the fra tional
maximal fun tion asso iated with a nonnegative measure on R , respe tively. That is,
n
Z
I  (x) = jx y j n d(y ); 0   n;
R n
and 
M (x) = sup r n  B (x; r) ; 0   n:
r>0
It is evident that I    M  due to the estimate

r n  B (x; r)  I  (x);
for x 2 R n and r > 0. However, if d(y) = jyj dy and x = 0 then the reverse inequality is
false. In view of this, the Morrey norm equivalen e of I #  and M  is quite 1surprising.
To do so, let us use the Fe erman-Stein sharp fun tion F of a fun tion F 2 Llo (R n ):
Z
F # (x) = sup jQj 1 jF (y) FQ jdy (4.1)
Q3x Q

where the supremum is taken over all \ oordinate" ubes Q ontaining x, and FQ denotes
the integral average of F over Q.
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 11

4.1 Lemma. Let 2 [0; n℄ and let  be a nonnegative measure with ompa t support on
R n and for whi h I   2 L1lo (R n ). Then:

(i)(I  )# (x)  M (x) for x 2 R n ;


(ii) Given a ube Q  R n and numbers t;  > 0,
jfx 2 Q : I  (x) > tgj  jfx 2 Q : (I  )# (x) > 2 1 tgj
+ jfx 2 Q : I  (x) > 2 n 1 tgj:

Proof. (i) For this omparability, see also [A2, Theorem 2.2℄ or [A1, Propositions 3.3 and
3.4℄.
(ii) For t > 0, let s = 2 n 1 t and M (t) = jfx 2 Q : I  (x) > tgj. Then an appli ation
of the C-Z (Calderon-Zygmund) de omposition theorem to Q, t and I   gives Q = P t [ Qt
and P t \ Qt = ; with the following three properties:
1) Qt = S1k=1 Qtk , where Qt1 ; Qt2; ::: are ubes of whi h two di erent elements have disjoint
interiors;
2) I  (x)  t a.e. x 2 P t ;
3) t < jQtk j 1 RQtk I   < 2n t for any Qtk 2 fQtk g.
It is worth remarking that for t and s we may hoose two C-Z de ompositions su h that
every ube of the C-Z de omposition asso iated with t is ontained in a ube of the C-Z
de omposition asso iated with s.
Let F1 be the family of ubes Qsj of the C-Z de omposition asso iated with s su h that
Qsj  fx 2 Q : (I  )# (x) > 2 1 tg (4.2)
and F2 the family of the remaining ubes of the C-Z de omposition asso iated with s.
Now, if Q0 2 F2 then there is x 2 Q0 su h that (I  )# (x)  2 1 t and by the de nition
of (I  )# , one has
Z
0
jQ j 1 j0 I   (I  )Q0 j  2 1 t (4.3)
Q

and Z
(I  )Q0 = jQ0j 1 I    2n s = 2 1 t: (4.4)
Q0

For the ubes Qtk  Q0 from the C-Z de omposition asso iated with t, we use the third
12 D.R. ADAMS AND J. XIAO

property of the C-Z de omposition, (4.3) and (4.4) to imply


X X Z
t jQtk j < t
I  
Qtk Q0 Qtk Q0 Qk
X Z X
 t
jI   (I  )Q0 j + (I  )Q0 jQtk j
Qtk Q0 Qk Qtk Q0
Z X
 0
jI   (I  )Q0 j + (I  )Q0 jQtk j
Q Qtk Q0
X
 2 1 tjQ0j + 2 1 t jQtk j:
Qtk Q0

Consequently, X
jQtk j < jQ0 j:
Qtk Q0
Thus, X  X  X
jQtk j <  jQ0j  M (s) = M (2 n 1 t):
Q0 2F2 Qtk Q0 Q0 2F2

Meanwhile, it follows from (4.2) that


X  X  X
jQtk j = jQ0j  jfx 2 Q : (I  )# (x) > 2 1 tgj:
Q0 2F1 Qtk Q0 Q0 2F1

So
 X X  X 
M (t) = + jQtk j
Q0 2F1 Q0 2F2 Qtk Q0
 M (2 n 1 t) + jfx 2 Q : (I  )
# (x) > 2 1 tgj: (4.5)
This (4.5) on ludes the proof.
4.2 Theorem. Let 1 < p < 1 and 0  ;   n. If  is a nonnegative measure on R n
then
kI  kLp;  kM kLp; : (4.6)
Proof. From Lemma 4.1 (i) as well as the estimate I    M , we need only to prove
kI  kLp; . k(I  )# kLp; : (4.7)
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 13

We rst suppose that  has ompa t support. Then (4.7) is a onsequen e of Lemma
4.1 (ii). In fa t, by Lemma 4.1 (ii) we have that for any ube Q  R n ,
Z
p
Z 1
I   = jfx 2 Q : I  (x) > tgjdtp
Q 0Z
1 Z 1
 M (2 n 1 t)dtp + jfx 2 Q : (I  )# (x) > 2 1 tgjdtp
0 Z 1 0 Z 1
= 2 p( n+1) 1
M (t)dt + (2 )
p p jfx 2 Q : (I  )# (x) > tgjdtp
Z0 Z0
= 2p(n+1) I   + (2 1 )p (I  )# p :
p
Q Q

If  = 2 1 p(n+1) , then
Z Z
p
2 1 I    2p(2+p(n+1)) (I  )# p : (4.8)
Q Q

Given a ball B (x; r) entered at x 2 R n and with radiuspr, we may nd two ubes Q1
and Q2 whi h have the enter x and the side lengths 2r= n and 2r respe tively. This,
together with (4.8) and the de nition of kkLp; , dedu es that (4.7) is true for the measure
 with ompa t support on R n .
If  does not have ompa t support, we let j be the restri tion of  to the ball B (0; j )
(with radius j about the origin 0) for j = 1; 2; :::. By (4.7) for j , we have
kI  j kLp; . kM kLp; (4.9)
for all j , with the onstant in (4.9) that does not depend on j . The inequality (4.7) for the
general  now follows from (4.9) and monotone onvergen e.
Of ourse, Lemma 4.1 and the argument for Theorem 4.2 may infer
kI  kLp  kM kLp (4.10)
whi h is the theorem of B. Mu kenhoupt and R. L. Wheeden. For a proof of (4.10) using
the so- alled \good  inequality", see also [AH, pp. 72-73℄.
4.3 Lemma [A1, Theorem 3.1℄. If 0   n, 0 <   n, 1 < p < = and p~ = p=( p),
then
kI  f kLp; . kf kLp; :
~

An immediate orollary of Lemma 4.3 and Theorem 4.2 is


14 D.R. ADAMS AND J. XIAO

4.4 Corollary. Let 0   n and 0 <   n.


(i) [A2, Theorem 3.2℄ If 1 < p < =m, m 2 N , p~m = p=( mp), and f is a C m fun tion
on R n , then
kf kLp~m; . krm f kLp; ;
where rm f denotes the m-th order gradient of f ; i.e.
 m=2
r f= 
m f; m = 2; 4; 6; :::
r(m 1)=2 f; m = 1; 3; 5; :::
for whi h  is the Lapla ian.
(ii) If 1 < p < = and p~ = p=( p), then
kM f kLp; . kf kLp; :
~

Note that in ase = 0 of Corollary 4.4 (ii) is well-known; see for example [CFr, Theorem
1℄.
5. Morrey Type II Capa ities
To begin with, we de ne the Morrey type II apa ity. For a set E  R n , 2 [0; n℄,
 2 (0; n), and p 2 (1; 1), let
C (E ; Lp;) = inf kf kpLp; : f  0 & I  f  1E
be the Morrey apa ity of E . It is not hard to see that Cn (; Lp;) is a monotone, ountably
subadditive set fun tion on the lass of all subsets of R whi h vanishes on the empty set.
Furthermore, it is an outer apa ity in the sense that
C (E ; Lp;) = inf C (G; Lp;)
where the in mum is over all open sets G  E .
The dual to a Morrey Type II apa ity is denoted Cap (). Due to Theorem 3.3 we bring
K (as the predual of Lp; , q = p=(p 1), p > 1) into play to de ne the dual apa ity.
q;
For E  R n let
Cap (E ; H q;) = sup (E ) :  2 M+ (E ) & kI  kK q;  1 ;
where M+ (E ) denotes those Radon measures that are nonnegative and have their support
in E .
Meanwhile, we onsider then orresponding weighted Riesz apa ity: For E  R n and a
nonnegative fun tion ! on R , let
! (E ) = inf kf kp
R ;p : f  0 & I  f  1
;
L (!)
p E
where Z
kf kLp(!) = n jf jp!:
p
R
The following theorem reveals the relationship between these three apa ities.
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 15

5.1 Theorem. Let 2 [0; n℄,  2 (0; n), 1 < p < 1 and q = p=(p 1). Then

C (E ; Lp;) = Cap (E ; H q;) p = sup R! ;p
 (E ); E  R n ; (5.1)


where the supremum is over all  2 M+ (R n++1 ) with k k1 =  (R n++1 ) = 1.


Proof.We start with working on the rst equality in (5.1). Without loss of generality, we
may assume that E  R n is ompa t. From [K℄ (or the proof of Theorem 2.3) it follows
that Z



kf kLp; = sup n fg ; f 2 Lp;; (5.2)
g R
where the supremum is over all g 2 H q; = K q; with kgkK q;  1.
If f  0 and I  f  1 on E , an appli ation of Holder's inequality implies
Z Z Z 1=p Z 1=q
q
(E )  I  fd = fI    f p!  I   !1 q :
E Rn Rn Rn

So by (5.2) and the de nition of k  kK q; ,


(E )  kf kLp; kI  kK q; :
And hen e it follows that
Cap (E ; H q;)  C (E ; Lp;)1=p: (5.3)
The inequality (5.3) is the half of what we want; the other half is not as easy. Using the
Minimax Theorem in [AH, Theorem 2.4.1℄ we have
Z Z
sup inf inf I  (g! 1=p)d = inf inf sup I  (g! 1=p)d: (5.4)
g   Rn   g Rn
Here the supremum
+1
is over all g 2 Lpn(+1Rn ) with kgkLp  1, as well as the in mums are over
all  2 M (R+ ) with kk1 = (R+ ) = 1 and over  2 M+ (E ) with kk1 = (E ) = 1,
+ n
respe tively. Of ourse, ! must an admissible weight or else the integral involving ! 1=p
will be unde ned; i.e., we take ! > 0 on R n for otherwise we an repla e it by one that is
positive on R n and make the integral smaller as a fun tion of .
Note that the integrand in (5.4) is jointly onvex in the argument    (the eigenvalues
of the Hessian are nonnegative). If (; ) 2 M+ (Rn++1 )  M+ (E ) is equipped with the norm
maxfkk1; kk1g, then the set
f(; ) : kk1 = 1; kk1 = 1g
is vaguely ompa t.
16 D.R. ADAMS AND J. XIAO

Now it turns out from the de nition of k  kK q; that the right-hand-side of (5.4) is equal
to
inf inf kI  kLq (! q=p) = inf kI  kK q; = Cap (E ; K q;) 1: (5.5)
Also noting that if  is taken as the Dira measure on entrated at x 2 E then
Z
I  (g! 1=p)d  I  (g! 1=p)(x); x 2 E:
Rn
we apply lower semi- ontinuity to obtain an xE 2 E su h that
I  (g! 1=p)(x)  I  (g! 1=p)(xE ); x 2 E:
Thus !
g! 1=p
I  (x)  1; x 2 E:
I  (g! 1=p)(xE )
Consequently
k g! 1=p kpLp;
C (E ; L )  
p; p :
I  (g! 1=p)(xE )
This, together with (5.5) and (5.2), infers
Z Z

inf inf I  (g! 1=p)d  C (E ; Lp;)1=p  sup g! 1=p h ;
Rn h Rn

where the supremum is taken over all h with khkK q;  1. But the right-hand-side above
does not ex eed
sup kgkLp kh! 1=p kLq :
h
Sin e we are onsidering h with the ondition
 Z 1=q
khkK q; = inf jhjq !1 q  1;
R n

we an take any su h  so that kh! 1=p kLq  1 +  for any  > 0. Hen e
sup kgkLp kh! 1=p kLq  kgkLp (1 + ):
h
Thus  Z 
inf inf I  (g! 1=p)d C (E ; Lp;)1=p  kgkLp (1 + ):
Rn
The inequality   1
Cap (E ; H q;) C (E ; Lp;)1=p  1 (5.6)
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 17

follows right away upon using (5.4), taking supremum over g with kgkLp  1 and noti ing
that  > 0 is arbitrary. Also observe that the order of the two in mums over (; ) is not
an issue be ause the integrand is jointly onvex in these two arguments. Clearly, (5.6) is
just the other half of what we desire.
For the se ond equality in (5.1), we again use the Minimax Theorem (quoted before),
but this time write Z Z
sup inf n I fd = inf sup n I fd; (5.7)
f R f R
where now the supremum ranges over all f 2 Lp (! ) with
Z
kf kpLp(! ) = jf jp!  1;
Rn
while the in mum is over all  2 M+ (E ) with kk1 = 1.
Taking the in mum over both sides of (5.7) with respe t to all  2 M+ (R n++1 ) satisfying
kk1 = 1, we again see that the right-hand-side of (5.7) then be omes
  1
inf inf kI  kLq (! q=p) = inf kI  kK q; = Cap (E ; H q;) ;

where again we have taken advantage of the onvexity of kI  kLq (! q=p) with respe t to
(; ) to inter hange the two in mums. Meanwhile, the left-hand-side of (5.7) be omes
  1=p   1=p
inf R! ;p (E ) = sup

R! ;p (E ) :

This follows from the standard Meyers format; see page 274 in [Me℄. Therefore, the proof
of (5.1) is omplete.
The above theorem enables us to estimate the Morrey apa ity of a ball in R n . First we
need:
5.2. Remark. Re all that 1E stands for the hara teristi fun tion of a set E . Then
1B(x ;r ) of the ball B (x0; r0)  R n is given by
0 0

1B(x ;r ) (x) = 10;; ifif xx 22= BB ((xx0;; rr0)):



0 0
0 0
Thus there exists a onstant independent of B (x0; r0) su h that
k1B(x ;r ) kpLp; = sup n r n jB (x; r) \ B (x0; r0)j
r>0;x2R
0 0

 sup rr ; n rn ; ifif rr > 22rr0:


 

r>0 0 0
= r0

18 D.R. ADAMS AND J. XIAO

5.3 Theorem. Let p <  < n. Then


C (B (x; r); Lp;)  r p ; B (x; r)  R n : (5.8)
Consequently
C (E ; Lp;) . (1 p
) (E ); E  R n : (5.9)
Proof. It is lear that for any y 2 B (x; r),
I  1B(x;r) (y ) & r :
Thus, by Remark 5.2 we have
C (B (x; r); Lp;) . k1B(x;r) r kpLp;  r p :
In order to obtain the lower bound, note that for any y 2 B (x; r),
I  f (y )  1B(x;r) (y ):
Hen e, it follows from Remark 5.2 and Lemma 4.3 that for p~ = p=( p),
 p~=p
C (B (x; r); L ) & k1B(x;r) kpL~ p;  r :
p;
~

Consequently
C (B (x; r); Lp;) & r p :
Clearly, the ountable subadditivity of C (; Lp;) yields (5.9).
Moreover, using Theorem 5.3 we an obtain a lo al isoperimetri -type inequality atta hed
to the Morrey apa ity.
5.4 Theorem. Let 2 [0; n℄,  2 (0; n), r 2 (0; 1) and p~ = p=( p) with 1 < p < = .
Then
jE j( p)= . rp(n )=p~C (E ; Lp;); E  B (x; r)  R n : (5.10)
Proof. We x the set E  B (x; r) in the de nition of C (E ; Lp;). By Holder's inequality
we have that if I  f  1 onZE then Z  1=p~
jE j  jI  f j  jI  f jp~ jE j1 1=p~:
E E
This, together with the boundedness of I sending Lp; to Lp~; (see also [A1, Theorem
3.1℄), yields
Z 1=p~
jE j( p)=(p)  jI  f jp~
E
Z 1=p~
 jI  f jp~
B(x;r)
 Z 1=p~
= r(n )=p~ r n jI  f jp~
B(x;r)
r (n )=p~ kI  f kLp;
~

. r(n )=p~kf kLp; :


NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 19

We are done.
We lose this se tion by showing that there exists no strong type inequality for the
potentials of Morrey fun tions:
5.5 Example. Let ; 2 (0; n), 2 (0; 1) and 1 < p < = . There is a fun tion f 2 Lp;
su h that Z 1   =p dt
C fx 2 R n : I  f (x)  tg; Lp; tp = 1: (5.11)
0 t
Proof. The example is simple: set
f (y ) = jy j =p 1B(0;1) (y )
for those  and p assumed above. Then it is easy to see that kf kLp; < 1. Next noti e
that Z

I  f (x) = jx yj njyj =p dy = jxj =p 1 + O(1) ; jxj ! 0:
B(0;1)
Thus there are onstants r0, 1 and 2 su h that
1 jxj =p  I  (x)  2 jxj =p ; jxj < r0 :
But (5.11) holds be ause
C fx 2 R n : jxj < r0 & jxj =p  tg; Lp;  t p ; t ! 1:
So if we use the Choquet-Lorentz notation, then we have a fun tion f 2 Lp; for whi h
I  f 62 L(p; ) C (; Lp;) ; 8 < 1:
Note that we trivially have 
I  f 2 L(p;1) C (; Lp;) :
5.6 Remark. Here by the Choquet-Lorentz notation in relation to the apa ity C , u 2
L(p; ) (C ) means
Z 1
 =p dt
tp C fx 2 R n : ju(x)j  tg < 1; < 1;
0 t
and 
sup
t>0
tp C fx 2 R n : ju(x)j  tg < 1; = 1:
It is worth remarking the following fa ts: First, when  = n, we have
 p 1=p
Z 1  
kI  f kL p;p .
( ) C fx 2 R : I  f (x)  tg; L dt
n p . kf kL p;p
( )
0
where L(p;p) = Lp , and L(p ;p) is a Lorentz spa e on R n with p = np=(n p). Se ond, if
 2 (0; n), then
kI  f kLp; . () . kf kLp; ;
~

where again p~ = p=( p). It appears that there is no Choquet spa e norm for I  f
to insert in (), i.e., no apa itary strongtype inequality (CSI) for the Riesz potentials of
fun tions in a Morrey spa e with respe t to the Morrey type II apa ity: C (; Lp;).
20 D.R. ADAMS AND J. XIAO

6. Morrey Type I Capa ities


This se tion is devoted to an investigation of the Morrey type I apa ity asso iated with
the predual spa e of a Morrey spa e.
We start with a apa ity type fun tional that is naturally atta hed to the spa e of M.
Riesz potentials of fun tions in H q;. The fun tional is:
R ;q;() = inf fkf kqZ q; : f  0; I  f   on suppg:
And, its dual fun tional is determined by
nZ o
S ;p; () = sup d :  2 M+ (supp) & kI  kLp;  1 :
Rn
Here  is a nonnegative bounded fun tion with ompa t support (and its support is written
as supp). All su h fun tions will be denoted by O.
6.1 Theorem. Let 2 [0; n℄,  2 (0; n), p 2 (1; 1) and q = p=(p 1). Suppose  2 O
and its restri tion to its support is ontinuous there and stri tly positive. Then

R ;q;()1=q  S ;p; (): (6.1)


Proof. For our purpose, let
n Z o
M =  2 M+ (supp()) : d = 1
Rn
and 
F = f 2 H q; : kf kZ q;  1 ;
and take an a ount of the following fun tionals:
 Z  1
R ;q;;1() = sup Minf I  fd
F Rn
and  Z  1
S ;p;;1 () = inf sup n I  fd :
M F
R
Note that sin e  is bounded below away from zero on supp, M is a ompa t subset
in the vague topology of M and that both M and F are onvex. These fa ts plus the
linearity and ontinuity of RRn I  fd insure that Fan's Minimax Theorem (see again [AH,
Theorem 2.4.1℄) applies and yields
R ;q;;1()1=q = S ;p;;1 ():
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 21

We on lude the proof by showing


R ;q;()  R ;q;;1(); S ;p; ()  S ;p;;1 ():
Nevertheless, it is enough to verify the se ond omparability sin e the argument for the
rst one is similar. To do this, rst noti e that S ;p;;1()  0 sin e 0 2 F. Also, note that
S ;p;;1 () < 1, for if not then there would be a sequen e of measures fj g  M su h
that Z
sup n I  fdj ! 0; j ! 1
f 2F R
and hen e, by the H q;-Lp; duality,
kI  j kLp; ! 0; j ! 1:
Be ause M   I   for any nonnegative measure , we obtain
kM j kLp; ! 0; j ! 1;
This implies that there isR a subsequen e fjk g onverging vaguely to a measure  2 M.
This measure  satis esR Rn d = 1. But theR onvergen e in Lp; yields  (B ) = 0 for all
balls B  R n and then Rn d = 0, violating Rn d = 1.
Due to S ;p;;1() < 1 and the H q;-Lp; duality, we see that for any  > 0 there is a
measure  2 M satisfying
Z ! 1
S ;p;;1 () < sup I  fd 1 + :
+  . kI   kLp;
f 2F Rn
1  , we get
So if  = kI   kLp;
Z
S ;p;;1 () 1 =
 . kI   kLp; d . S ;p; ();
Rn
and thus
S ;p;;1 () . S ;p; ():
To establish the reversed version of the last estimate, we may assume S ;p;() < 1,
without lossR of generality.
 1
Then for any  2 M+ (supp) with kI   kLp;  1 and f 2 F
and  = Rn d  , we have
Z Z Z  1
I  fd = fI   . d :
Rn Rn Rn
by the H q;-Lp; duality on e again. So
Z
d . S ;p;;1 ():
Rn
22 D.R. ADAMS AND J. XIAO

This implies that


S ;p; () . S ;p;;1 ():
The above dis ussion produ es
S ;p; ()  S ;p;;1 ():
We are done.
For K  R n ompa t, p 2 (1; 1), q = p=(p 1), 2 [0; n℄ and  2 (0; n), de ne
C (E ; H q;) = inf kf kqZ q; : f 0 & I  f  1E :

Also, let
Cap (K ; Lp;) = sup (K ) :  2 M+ (K ) & kI  kLp;  1 :
In an obvious manner, these two apa ities are extended to the general set E  R n .
6.2 Corollary. Let ;  2 (0; n), 1 < p < 1 and q = p=(p 1). Then
 q
C (E ; H )  Cap (E ; L ) ;
q; p; E  Rn : (6.2)

Proof. It follows from taking  = 1K (for ompa t K  R n ) in (6.1).


6.3 Remark. C (E ; H q;) is never zero when p = =(n ) > 1. This follows sin e by
duality
I  f (x)  kf kH q; jx j n Lp;
and the se ond fa tor on the right is nite for all x for our hoi e of p. Also if the Riesz
kernel is modi ed at in nity (say set equal to zero for large arguments or have exponential
de ay at in nity) then the same result holds here for 1 < p  =(n ).
6.4 Theorem. Let ;  2 (0; n). Then:
(i) For all p > 1,
Cap (B (x; r); Lp;) . rn =p ; B (x; r)  R n ; (6.3)
(ii) For all p > maxf=(n ); (n )= g,

Cap B (x; r); Lp; & rn =p ; B (x; r)  R n : (6.4)
Proof. We begin by establishing the lower bound in (6.4). We do this by rst getting an
analogue of the Wol potential for the apa ity Cap (; Lp;). To a omplish this, we use
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 23

(4.6) and estimate the quantity on the right-hand-side of (4.6). Applying Fubini's theorem
and the simple inequality:
p dt 1=p
Z 1  
M (x) . t  B (x; t)
n ; x 2 Rn ;
0 t
we get !
Z 1 Z  p
 dt
kM kpLp; . sup r n tp( n)  B (x; t) dx : (6.5)
x 2Rn;r>00 0 B(x ;r) t 0

Sin e B (x; t)  B (y; 2t) whenever jx yj < t, we have the following estimate on the
inside integral of the right-hand-side of (6.4): !
Z Z Z
 p  p 1
 B (x; t) dx =  B (x; t) d(y ) dx
B(x0 ;r) B(x0 ;r) B(x;t)
ZZ  p 1
=  B (x; t) dxd(y )
Z
jx x0 j<r;jx yj<t
 p 1
  B (y; 2t) jB (x0; r) \ B (y; t)jd(y)
Rn
:= INT(x0 ; r; t):
We put this ba k into (6.5) and get Z Z
r 1  
dt
kM kpLp; . sup r  n + tp( n) INT(x0 ; r; t) :
x 2R ;r>0
n 00 r t
Looking at the rst integral,
Z r
we nd
 dt
sup r  n INT(x0; r; t) t
x0 2R ;r>0
n 0
Z r Z 
 p 1 dt
. sup r n tp( n) tn  B (y; 2t) d(y )
nx0 2R ;r>0 0 Rn t
Z 1 Z
dt  p 1 
. tn+p( n) t n
 B (y; 2t) d(y )
0 Rn t
Z Z 1 
  p 1 dt

. n tp( n)+  B (y; t) d(y ):
R 0 t
Meanwhile, looking at theZ se ond integral, we have
1  dt
sup r n INT( x0 ; r; t)
t
x0 2R ;r>0
n r
Z 1 Z  p 1 
dt
. sup r n tp( n) tn  B (y; 2t) d(y )
nx0 2R ;r>0 r Rn t
Z 1 Z  p 1 
dt
. tp( n)+  B (y; 2t) d(y )
0 Rn t
Z Z 1  p 1 dt 
. tp( n)+  B (y; 2t) d(y ):
Rn 0 t
24 D.R. ADAMS AND J. XIAO

Therefore Z
kM kpLp; .  (y )d(y );
W ;p; (6.6)
Rn
where the homogeneous Wol potential for the apa ity Cap (; Lp;) is:
Z 1  p 1 dt
 (y ) =
W ;p; tp( n)+  B (y; 2t) : (6.7)
0 t
Now to get the lower bound in (6.4), we estimate (6.7) from above (and hen e Rgenerate
an upper estimation of kI  kLp; ). To do so, we break up the integral in (6.7) as 0r + Rr1
(where we are estimating Cap (B (x0; r); Lp;)), take d(x) = 1B(x ;r) dx, and get 0

Z r Z r
dt
f  g . tp( n)+ tn(p 1)
t
 r p+ n ; p > n  ;
0 0
and Z 1 Z 1 dt
f  g . tp( n)+ rn(p 1)
t
 r p+ n ; p > n  :
r r
To get the lower bound in (6.4), we repla e the measure  above by  = r ( +=p) 
with an appropriate onstant > 0. This, together with (6.6), (6.7) and (4.6), a hieves
kI  kLp;  1 and then yields the lower estimate in (6.4).
Next we give the upper bound in (6.3). If f = 1B(x ;r) , then I  f (x)  r for a
onstant > 0 when x 2 B (x0; r). Thus we need to al ulate k1B(x ;r) ( r ) 1 kZ q; . To
0

this end we make the following de omposition:


0

1B(x ;r) ( r ) 1 = 1B(x ;r)jB (x0; r)j( pn)=(pn) jB (x0; r)j ( pn)=(pn) ( r ) 1 :
0 0

Sin e 1B(x ;r) jB (x0; r)j( pn)=(pn) is a (q; n )-atom, one has
0

k1B(x ;r) ( r ) 1 kZ q; . r ( pn)=p ( r ) 1 = rn =p :


0

And this in term gives


 1=q
C B (x0; r); H q; . rn =p :

Therefore, the proof is omplete.


7. Choquet Integrability via Morrey Type I Capa ities
In this se tion, we study the spa es indu ed by the Choquet integrals with respe t to
Morrey type I apa ities. In parti ular, we prove that the Hardy-Littlewood maximal
operator M0 is bounded on these spa es under some natural assumptions, but also that the
fun tional R ;q; may be bounded (from both below and above) by the spa e quasi-norms.
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 25

We denote by L1 (Cap (; Lp;)) the lass of all Radon measures (i.e., lo ally nite
regular signed Borel measures)  on R n obeying
kkL1p;; = supn Capj(jK(K; L) p;) < 1;
K R
where the supremum is taken over1
all ompa t sets K  R n , and againp;jj is the total vari-
ation measure of . Also, let L (Cap (; L )) onsist of all Cap (; L ) quasi- ontinuous
p;
fun tions f on R n ( f. [A4℄), for whi h
Z

kf kLp;; = n jf jdCap ; Lp;
1

ZR1
= Cap fx 2 R n : jf (x)j  tg; Lp;dt < 1: (7.1)
0
The intermediate term is alled the Choquet integral with respe t to Cap ; Lp;.
7.1 Theorem. Let 2 [0; n℄,  2 (0; n) and p 2 (1; 1). The the pairing
Z
hf; i = n fd
R
realizes the dual of L1 (Cap (; Lp;)) as equivalent to the spa e L1 (Cap (; Lp;)).
Proof. It is not hard to see that ea h  2 L1 (Cap (; Lp;)) indu es a bounded linear
fun tional on L1(Cap (; Lp;)). In fa t, for su h a , whenever f 2 L1(Cap (; Lp;)), we
have
Z 1
jhf; ij  jj(fx 2 R n : jf (x)j  tg)dt
0 Z 1

 kkL1p;; Cap fx 2 R n : jf (x)j  tg; Lp; dt
0
= kkL1p;; kf kL1p;; :
For the onverse, sin e C0(R n ) (the lass of all ontinuous fun tions with ompa t
support on R n ) is ontained in L1 (Cap (; Lp; )), every bounded linear fun tional L on
L1 (Cap (; Lp; )) with nite operator norm kLk is given by
Z
fd; f 2 C0 (R n );
n R
for some Radon measure  on R n . However for any g 2 C0 (R n ), we have
Z Z 

n gd
jj  sup

d : 2 C0 (R n ) & j j  jgj
R Rnn o
 kLk sup k kLp;; : 2 C0 1 (R n ) & j j  jgj
 kLk  kgkLp;; : 1
26 D.R. ADAMS AND J. XIAO

Now, given a ompa t set K  Rn ,


we for e g to approa h 1K , and use the previous
estimate to get
jj(K ) . kLkCap (K ; Lp;):
Therefore we omplete the proof.
For the sake of simpli ity, we employ L1 (Cap (; Lp;))+ to denote the set of all non-
negative measures  2 L1 (Cap (; Lp;)).
7.2 Corollary. Let 2 [0; n℄,  2 (0; n) and p 2 (1; 1). If f  0 is lower semi- ontinuous
on R n , then
nZ o
kf kLp;;  sup
1 fd :  2 L1 (Cap (; Lp;))+ & kkL1p;;  1 : (7.2)
Rn

Proof. Theorem 7.1 implies that the anoni al map of L1(Cap (; Lp;)) into the se ond
dual has the quasi-norm
Z n Z o

jujdCap (; L )  sup
p;
ud :  2 L1 (Cap (; Lp;))+ & kkL1p;; 1
Rn R n

for any u 2 L1 (Cap (; Lp;)). For a nonnegative lower semi- ontinuous f , we approxi-
matenfrom below by a sequen e fj g  C0 (R n )+ (the lass of all nonnegative fun tions in
C0 (R )), j % f as j ! 1. Then
Z nZ o
j dCap (; Lp;) = sup j d :  2 L1 (Cap (; Lp;))+ & kkL1p;;  1
Rn Rn
nZ o
 sup fd :  2 L1 (Cap (; Lp;))+ & kkL1p;;  1 :
Rn
It is lear that
Cap fx 2 R n : j  tg; Lp;) ! Cap fx 2 R n : f  tg; Lp;);
as j ! 1. Consequently, it follows that
kf kLp;; 1

nZ o
. sup n
fd :  2 L1 (Cap (; Lp;))+
& 1
kkL1p;;
n ZR 1 o

. sup  fx 2 R n : f (x)  tg dt :  2 L1 (Cap (; Lp; ))+ & kkL1p;; 1
0
. kf kLp;; 1

The result now follows.


Next, we would like to give a omparable fun tional of R ;q; via the Choquet integral
with respe t to Cap (; Lp;).
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 27

7.3 Theorem. Let 2 [0; n℄,  2 (0; n), p 2 (1; 1) and q = p=(p 1). Then for any
 2 C0 (R n )+ ,
R ;q;() . kkqL1p;; : (7.3)
Proof. For  2 C0 (R n )+ and  > 0, set
(x) + ; x 2 supp;

 (x) =
0; otherwise:
Then Theorem 6.1 applies to  and gives
R ;q;()
 R ;q;( )
nZ o
+
 sup n  d :  2 M (supp ) & kI  kLp;  1
R
nZ 1  o
 sup  fx 2 R n :  (x)  tg dt :  2 M+ (supp ) & kI  kLp;  1
0
Z 1
. Cap fx 2 R n :  (x)  tg; Lp;dt
Z0

  dCap ; Lp;
n
ZR
. dCap (; Lp;) + Cap (supp; Lp; ):
Rn
So, the result follows.
However, in order to establish the reversed inequality of (7.3), we rst derive an estimate
for the Hardy-Littlewood maximal operator M0(f ) of a fun tion f with (7.1).
7.4 Theorem. Let 0 < <  < n; 1 < p < =( ), and q = p=(p 1). Then for any
f 2 H q; , Z
M0 (I  f )dCap (; Lp;) . kf kH q; : (7.4)
Rn
In parti ular,
I H q;  L(q;1)(C (; H q;)): (7.5)
Proof. Without loss of generality, we an assume f  0. Then M0 (I  f ) . I  f . So
Z
M0 (I  f )d . kf kH q; kI  kLp; :
Rn
Hen e due to Corollary 7.2, it suÆ es to prove that
Z
(I  )p . rn ; B (x; r)  R n (7.6)
B(x;r)
28 D.R. ADAMS AND J. XIAO

is valid for all measures  2 L1 (Cap (; Lp;))+ . But by (6.3),


(B (x; r))  kkL1
p;; Cap (B (x; r ); L )
p;

. kkL1p;; rn =p :
The estimation (7.6) follows from [A1, Theorem 5.1℄ with d = dx = Lebesgue n-measure.
The result there is
kI  kL ;
 . kkL ; 1

under the onditions: 0 < <  = + =p and = =( ). Re all that kI  kL ;
 <1
if and only if I   belongs to the weak-Morrey spa e L : ;

sup
t>0
t B (x; r) \ fx 2 R n : I  (x)  tg . rn  :
Clearly, L  Lp;p= when p < . Thus  = + =p gives
;

kI  kLp; . kkL ; 1 =p


+ . kkL1p;;
sin e p  =  and p < be omes p < =( ). We are done.
7.5 Theorem. Let 0 < <  < n; 1 < p < =( ), and q = p=(p 1). Then for any
 2 C0 (R n )+ ,
kM0()kqLp;; . R ;q;():
1 (7.7)
Proof. By hoosing a f 2 H q; su h that I  f   on the support of  2 C0 (Rn )+, we
have M0()  M0(I  f ), and then
Z
kM0()kLp;; =
1 M0 ()dCap (; Lp; ) . kf kZ q; ; (7.8)
Rn
by (7.4) and Theorem 3.3. Thus (7.7) follows from (7.8) and the de nition of R ;q;().
Corollary 7.2, together (7.7) and (7.3), yields an interesting onsequen e as follows.
7.6 Corollary. Let 0 < <  < n; 1 < p < =( ), and q = p=(p 1). Then:
(i) For any f 2 C0 (R n ),
kM0f kLp;;  R ;q;(f )  kf kLp;; :
1 1

(ii) For any f 2 C01 (R n ) and + =p 2 N ,


kM0 f kLp;; . kr +=p f kL :
1 1

This result orresponds ni ely to Theorem A resp. Theorem B in [A3℄ on the Choquet
integrals with respe t to Hausdor apa ity.
NONLINEAR POTENTIAL ANALYSIS ON MORREY SPACES AND THEIR CAPACITIES 29

Referen es
[A1℄ D.R. Adams, A note on Riesz potentials, Duke Math. J., 42 (1975), 765-778.
[A2℄ D.R. Adams, Le tures on Lp-Potential Theory, Department of Mathemati s, University
of Umea2, 1981.
[A3℄ D.R. Adams, A note on Choquet integral with respe t to Hausdor apa ity, in \Fun -
tion Spa es and Appli ations," Lund 1986, Le ture Notes in Math. 1302, Springer-Verlag,
1988, pp. 115-124.
[A4℄ D.R. Adams, Choquet integrals in potential theory, Publi a ions Matematiques 42
(1998), 3-66.
[AH℄ D.R. Adams and L.I. Hedberg, Fun tion Spa es and Potential Theory, A Series of
Comprehensive Studies in Math. 314, Springer-Verlag, 1996.
[C℄ S. Campanato, Proprieta di in lusione per spazi di Morrey, Ri er he Mat. 12 (1963),
67-86.
[CFr℄ F. Chiarenza and M. Fras a, Morrey spa es and Hardy-Littlewood maximal fun tion,
Rend. Mat. Appl. 3/4(1988), 273-279.
[K℄ E. A. Kalita, Dual Morrey spa es, Doklady Math. 58(1998), 85-87.
[Me℄ N. G. Meyers, A theory of apa ities for potentials of fun tions in Lebesgue spa es,
Math. S and. 26(1970), 255-292.
[MuW℄ B. Mu kenhoupt and R. L. Wheeden, Weighted norm inequalities for fra tional
integrals, Trans. Amer. Math. So . 192(1974), 261-274.
[P℄ J. Peetre, On the theory of Lp; spa es, J. Fun t. Anal. 4 (1969), 71-87.
[S℄ G. Stampa hia, The spa es Lp; and H p; and interpolation, Ann. S uola Norm. Sup.
Pisa 19 (1965), 443-462.
[Z℄ C.T. Zorko, Morrey spa e, Pro . Amer. Math. So . 98(1986), 586-592.

D.R. Adams J. Xiao


Department of Mathemati s Department of Mathemati s and Statisti s
University of Kentu ky Memorial University of Newfoundland
Lexington, Kentu ky 40506 St. John's, NL, A1C 5S7
USA Canada
Email: davems.uky.edu Email: jxiaomath.mun. a

Potrebbero piacerti anche