Boll. Un. Mat. Ital. B (7) 10 (1996), no. 4, 843-883.

Commutators of Singular Integrals

on Homogeneous Spaces
Marco BRAMANTI and M. Cristina CERUTTI
Sunto. Sia ÐX,d,.Ñ uno spazio omogeneo, K un operatore di integrale singolare, continuo da _: ÐXÑ a _: ÐXÑ
per ogni p − Ð"ß_Ñ e tale che il suo nucleo soddisfi opportune disuguaglianze puntuali. Sia a − BMOÐXÑ;
si dimostra che il commutatore
CÒK,aÓÐfÑ œ aKf
KÐafÑ
è continuo su _: per ogni p − Ð"ß_Ñ e soddisfa la seguente stima:
² C ÒK,aÓÐfÑ ² : Ÿ cÐp,K,XÑ ² a ² ‡ ² f ² :
dove ² † ² ‡ è la seminorma BMO. Si mostra anche come applicare questo risultato ad alcuni esempi
concreti di operatori di integrale singolare su spazi omogenei di misura finita o infinita, fornendo
condizioni sul nucleo di O , sufficienti affinche´ l'operatore di integrale singolare O sia ben definito e
continuo su _: .

INTRODUCTION
A well-known result of 1976 by Coifman-Rochberg-Weiss (see [CRW]) is that, if K is a Calderon-
´
Zygmund operator on _ Б Ñ and a − BMO, then the commutator
: 8

CÒK,aÓÐfÑ œ aKf
KÐafÑ
is a well defined linear continuous operator from _: into _: , for every p − Ð1,_Ñ and
² C ÒK,aÓÐfÑ ² : Ÿ cÐp,n,KÑ ² a ² ‡ ² f ² :. Ð0.1Ñ
Over the years several extensions of this result have been proved. For instance, Chanillo [Cha] in '82
proved an analogous result for fractional integral operators; Bloom [Bl] in '85 proved a weighted estimate
for the commutator of the conjugate operator; Segovia and Torrea [ST1] in '89 generalized the above result
to vector valued functions and more recently (see [ST2]) they considered higher order commutators. In this
´
paper we consider singular integral operators of Calderon-Zygmund type, defined on a homogeneous space,
and prove the Coifman-Rochberg-Weiss' theorem in this context. In another one [BC2] we deal with
commutators of fractional integrals on homogeneous spaces.
The motivation for this paper arose when, proving _: estimates for parabolic equations with VMO
coefficients, we needed to extend [CRW]'s result to parabolic Calderon-Zygmund operators (see [BC1]).
That of homogeneous spaces seems a natural setting for this problem since the basic real analysis tools, as
maximal function, sharp function and BMO, can be defined in a natural way in this context; also, the
related results, analogous to the Euclidean case, have been proved (see [CW1], [Bu]). Moreover, the present
extension of (0.1) to homogeneous spaces turns out to be the suitable tool to prove _: -estimates for some
ultraparabolic operators with Z QS coefficients. This problem is considered by Bramanti-Cerutti-
Manfredini in [BCM]. See also Remark 4.13 for other notes about applications of our results to problems
arising from P.D.E.'s.
´
In § 1-2 we discuss definitions of homogeneous space and of Calderon-Zygmund type operator, as
well as recall known results. In § 2-3 we prove our main results, namely the _: estimate (0.1) for the
commutator, respectively in the case of an infinite measure space and in a finite measure one. It turns out
that, while the infinite measure case is fairly straightforward, the finite measure one is significantly
different. We want to stress that our definition of Calderon-Zygmund operator requires _: continuity: in
other words, our results say that, under suitable assumptions on the kernel and the space, the _: estimate on
the operator K implies Ð0.1) for its commutator. So, if we are given a concrete singular kernel on a
homogeneous space, to apply this theory we need to know that an _: continuous operator is actually
defined. This problem is discussed in § 4. Namely we first discuss some general sufficient conditions on the
kernel under which _: -continuity of the singular integral can be assured, and then give specific examples
arising in the applications in which our theorem (both in the infinite and finite measure case) can be
applied.
Finally, we wish to point out that our results are obtained under quite general assumptions on the
space X, namely X does not need to be endowed with a group structure neither the measure of a ball in X
needs to behave like a fixed power of the radius.

1
Acknowledgements. We wish to thank Prof. E. Fabes for some helpful discussions.

1. SOME BASIC FACTS ABOUT HOMOGENEOUS SPACES

Several different definitions of homogeneous spaces can be found in the literature. Here we refer
particularly to [CW1,2], [MS], [Bu].
Definition 1.1. Let X be a set. A function d:X ‚ X Ä Ò0,_Ñ is called quasidistance if:
iÑ dÐx,yÑ œ 0 Í x œ y;
iiÑ dÐx,yÑ œ dÐy,xÑ;
iiiÑ there exists a constant c.   1 such that for every x, y, z − X
dÐx,yÑ Ÿ c. ÒdÐx,zÑ  dÐz,yÑÓ. (1.1)
At the end of section 4.a we will show that, for the results contained in this paper, condition 33) can be
weakened to require only "quasisymmetry", instead of symmetry. (See Remark 4.6)
If ÐX,dÑ is a set endowed with a quasidistance, the balls B< ÐxÑ ´ BÐx,rÑ œ šy − X: dÐx,yÑ  r› (for
x − X and r  0) form a base for a complete system of neighborhoods of X, so that X is a Hausdorff space.
Note that the balls are not in general open sets; if they are, then they form a base for the topology of X .
Definition 1.2. We say that ÐX,d,.Ñ is a homogeneous space if:
iÑ X is a set endowed with a quasidistance d, such that the balls are open sets in the topology induced by
d;
iiÑ . is a positive Borel measure on X, satisfying the doubling condition:
 .ÐB#< ÐxÑÑ Ÿ c. † .ÐB< ÐxÑÑ  _
0 (1.2)
for every x − X, r  0, some constant c.  1.
The numbers c. , c. in (1.1)-(1.2) will be called constants of X and we will write cÐXÑ for a constant
depending on the constants of X.
We note that every homogeneous space is separable. This follows, for instance, from Theorem (1.2) of
[CW].
A fundamental result about quasidistances is the following one:
Theorem 1.3. (Macias-Segovia, [MS]).
Let d be a quasidistance on a set X. Then there exists a quasidistance $ on X such that:
ÐiÑ d and $ are equivalent, that is there exist constants c", c# such that for every x, y − X:
c" $ Ðx,yÑ
Ÿ dÐx,yÑ Ÿ c#$Ðx,yÑ;
ÐiiÑ $ is "locally Holder
¨ continuous"; more precisely there exist ! − Ð0,1Ó and c  0 such that for every x,
y, z − X:
± $Ðx,zÑ
$Ðy,zÑ ± Ÿ c $Ðx,yÑ! Ò$Ðx,zÑ  $Ðy,zÑÓ"!.
If d is a distance, d is Lipschitz, and the previous result is trivial.
We say that f − V(! if f has bounded support and

²f²( ´ sup ± Ð ÐÑ ÑРѱ

f x f y
 _.
xÁy
d x,y
(

If ÐX,d,.Ñ is a homogeneous space with . a regular measure, then the space V(! is dense in _: for every
( − Ð0,1], p − Ò1,_Ñ. (This follows from Thm. 1.3).
We will use also the following space of test funcitons
Wœ š f − __ÐXÑ ± the support of f is bounded›.
Note that W is dense in _: for every p − Ò1,_Ñ.
The definition of standard real analysis tools, such as the maximal function Mf, the sharp function f #
,
the BMO seminorm ² f ² ‡ and the BMO space, naturally carries over to this context, namely:
MfÐxÑ œ sup
' ± f Ð y Ñ ± d . Ðy Ñ ;
x−B
B

2
where the sup is taken over all balls containing x and
' B
... œ 1
. ÐB Ñ
' ... ;
B

f ÐxÑ#
œ sup

' ± fÐyÑ
fF ± d.ÐyÑ
x−B
B

where fF œ
' fÐxÑ d.ÐxÑ ;
B

²f²‡ œ sup f ÐxÑ #

œ sup

' ± fÐyÑ
fF ± d.ÐyÑ;
x B
B

BMO ´ šf − _"69- ÐXÑ : ² f ² ‡  _ ›.

We recall three results related to these concepts, which have been proved in the context of
homogeneous spaces in [CW1], [Bu].
Theorem 1.4. (Maximal Inequality).
For every p − Ð1,_Ó there exists a constant cÐX,pÑ such that for every f − _: :
² Mf ² : Ÿ c ² f ² :.
Theorem 1.5. ÐJohn-Nirenberg LemmaÑ.
For every p − Ò1,_Ñ there exists a constant cÐX,pÑ such that for every f − BMO, every ball B:
"Î:
Š
' ± f Ðx Ñ
fF ± : d . Ð x Ñ ‹ Ÿ c ² f ² ‡ .
B

Theorem 1.6. ÐSharp InequalityÑ.

For every p − Ò1,_Ñ there exists a constant cÐX,pÑ such that for every f − _: :
if .ÐXÑ œ _: ²f²:Ÿc²f ²: #

if .ÐXÑ  _: ² f

' f ² : Ÿ c ² f ² :.
X
#

Moreover, the following two lemmas about BMO functions hold:

Lemma 1.7. Let a − BMO and M  1. Then for every ball B< ÐxÑ and every positive integer j :
± aF Q 4 <
aF < ± Ÿ c † j ² a ² ‡
where c depends on M and on the dubling constant c. .
Proof. ± aFQ<
aF< ± œ ±
' B < ŠaÐxÑ
aFQ< ‹ d.ÐxÑ ± Ÿ
Ÿ c
. ÐBQ<Ñ
' B Q< ± aÐxÑ
aFQ< ± d.ÐxÑ Ÿ c ² a ² * .
Hence:
± aFQ4<
aF< ± Ÿ ± aF<
aFQ< ±  ± aFQ<
aFQ#< ±  á  ± aFQ4 <
aFQ4< ± Ÿ c † j ² a ² -1 * .

Lemma 1.8. Let f − BMO, and let: Ú fÐxÑ if ± fÐxÑ ± Ÿ n
f8 ÐxÑ œ Û n if fÐxÑ   n
Ü
n if fÐxÑ Ÿ
n
Then f8 − BMO and:
² f8 ² ‡ Ÿ c ² f ² ‡
with c an absolute constant.
Finally, we recall the following result about homogeneous spaces of finite measure. For sake of
completeness we include the proof, having been unable to find a written reference.
Lemma 1.9. Let ÐX,d,.Ñ be a homogeneous space. Then .ÐXÑ  _ if and only if X is bounded.
Proof. We have to prove that if .ÐXÑ  _ then X is bounded (the reverse follows from the definition of
doubling condition).

3
By contradiction. Let x! be a fixed point of X and x5 a sequence of points such that r5 ´ dÐx! ,x5Ñ Å _. By
the triangle inequality we can prove that:
B Ð x! , r

2c
5
. Ñ  BÐx5, 5. Ñ œ g
r

2c
and B Ðx ! , 2c
r 5
. Ñ © BÐx5, Hr5 Ñ Ð"Þ3Ñ
with H œ c. Š"  %-". ‹ .

Then .ŠBÐx! , 5. Ñ‹
r

2c
Å .ÐXÑ so that, by (1.3), BÐx5, 5. Ñ Ä 0. But then, by doubling and (1.3),
r

2c

.ŠBÐx! , 5. Ñ‹ Ÿ .ŠBÐx5 , Hr5 Ñ‹ Ÿ cÐXÑ † .ŠBÐx5 , 5. Ñ ‹ Ä 0

2c
r r

2c

so that .ÐXÑ œ 0, contradiction.

2. COMMUTATORS OF CALDERON-ZYGMUND OPERATORS ON HOMOGENEOUS SPACES
Definition 2.1. We say that K is a Calderon-Zygmund
´ operator (CZ operator) on the homogeneous space
ÐX,d,.Ñ if::
iÑ K:_ ÐXÑ Ä _: ÐXÑ is linear continuous for every p − Ð1,_Ñ;
iiÑ there exists a measurable function k:X ‚ X Ä ‘ such that for every f − W , for a.e. x  sprt f :
KfÐxÑ œ ' kÐx,yÑfÐyÑ d.ÐyÑ ; X

iiiÑ the kernel k satisfies the following pointwise Hörmander condition:

there exist constants cO  0, "  0, M  1 such that for every x! − X, r  0, x − B< Ðx! Ñ, y  BQ< Ðx! Ñ,

± kÐx!,yÑ
kÐx,yÑ ± Ÿ c O
. ÐBÐx!;yÑÑ † Ð ! Ñ"
d x ,x

Ð ! Ñ"
d x ,y
(H1)

where, BÐx;yÑ is the ball with center x and radius 2dÐx,yÑ. We will write cÐp,KÑ for a constant depending on
the number p − Ð1,_Ñ, the _: norm of K and the constants in (H1). The following holds:
Proposition 2.2. Condition (H1) implies the following integral Hörmander condition: there exist constants
A  1, B  0 such that for every x! , x − X:
'Ð! d x ,y A d x ,x
± Ð Ñ
kÐx,yÑ ± d.ÐyÑ Ÿ
Ñ  Ð ! Ñ k x! ,y B. (H0)

(Condition (H0) is assumed in [CG], [CW1] to prove the fundamental result about _: continuity of an
integral operator which is continuous on _# . This result will be recalled and applied in § 4).
Proof. Fix x! , x − X and apply (H1) with r œ 2dÐx! ,xÑ:
' Ð ! Ñ  Ð ! Ñ ± kÐx! ,yÑ
kÐx,yÑ ± d.ÐyÑ
d x ,y 2M d x ,x
Ÿ
Ÿ cO † dÐx! ,xÑ" ' Ð ! Ñ d x ,y Ð! Ñ
2M d x ,x
d. ÐyÑ

. ÐBÐx!;yÑцdÐx!,yÑ" œ
_
œ cO † dÐx! ,xÑ" !' 4 Ð ! ÑŸ Ð ! ÑŸ 4" Ð ! Ñ .Ð Ð ! .ÑÑ†Ð Ñ Ð ! Ñ" d y
Ÿ
4œ"
2 M d x ,x d x ,y 2 M d x ,x B x ;y d x ,y

_
Ÿ cO † dÐx! ,xÑ" ! Ò 4 Ð ! ÑÓ"
4œ"
1
2 M d x ,x
œ cO † cÐM,"Ñ.

Examples of CZ operators on homogeneous spaces will be discussed in detail in § 4. For a CZ operator
K, consider the commutator
CÒK,aÓÐfÑ œ aKf
KÐafÑ
where a − BMO. (When K is implicitly understood, we shall write more briefly: T+ f œ CÒK,aÓÐfÑ). Our goal
is to prove the following estimate:
² T+f ² : Ÿ cÐp,K,XÑ ² a ² ‡ ² f ² : (2.1)
for every p − Ð1,_Ñ. The above definition of CÒK,aÓÐfÑ is, up to now, purely formal: if f − _ and a − BMO,
:
KÐafÑ may not be well defined. However, if a − BMO, by the John-Nirenberg lemma (Theorem 1.5),
4
a − _:69- for every p − Ò1,_Ñ. Therefore if f − W and a − BMO, T+ f is well defined and belongs to _:69- for
every p − Ò1,_Ñ; if, moreover, a − __ , then T+ f − _: for every p − Ð1,_Ñ. The following proposition is a
consequence of the above remark, the density of W in _: and Lemma 1.8.
Proposition 2.3. If (2.1) holds for every a − __ and every f − W, then for every a − BMO the operator T+
can be continuously extended to _: for every p − Ð1,_Ñ, and (2.1) holds for every a − BMO, f − _: .
Let us now state the first main result.
Theorem 2.4. Let K be a CZ operator; then for every p − Ð1,_Ñ there exists a constant c œ cÐp,K,XÑ such
that for every a − __ , every f − W:
² ÐT+ fÑ ² : Ÿ c ² a ² ‡ ² f ² : .
#

Proof. We shall prove the following inequality:

for every r − Ð1,_Ñ there exists a constant cÐr,K,XÑ such that
"Î< "Î<
± ÐT+ fÑ Ð–xÑ ± Ÿ c † ² a ² ‡
ŠMÐ ± Kf ± < ÑЖxÑ‹  ŠMÐ ± f ± < ÑЖxÑ‹
#
Ÿ for every –x − X. (2.2)
From (2.2) the result follows by the continuity of K and the maximal inequality (Theorem 1.4). To prove
(2.2), let us fix a ball B$ œ B$ Ðx! Ñ containing –x, and let x be any point of B$ . Let us write:
T + f Ðx Ñ
œ KÐafÑÐxÑ
aÐxÑKÐfÑÐxÑ œ
œ KŠÐa
a Ñf ; Q ‹ÐxÑ  KŠÐa
a Ñf ; Q ‹ÐxÑ
ÐaÐxÑ
a Ñ † KÐfÑÐxÑ ´
B$ B $ B$ X\B $ B$

´ AÐxÑ  BÐxÑ  CÐxÑ. (2.3)

Following [To], p. 418, we get, by Theorem 1.5, Lemma 1.7 and the _: estimate on K:
"Î<

' ± CÐxÑ
CB ± d.ÐxÑ Ÿ cÐX,rÑ ² a ² † ŠMŠ ± KÐfÑ ± < ‹Ð–xÑ‹ ;
B$ $ * (2.4)

"Î<

' ± AÐxÑ
A ± d.ÐxÑ Ÿ cÐr,K,XÑ ² a ²
B$ B$ * ŠMŠ ± f ± < ‹Ð–xÑ‹ . (2.5)

To bound the middle term, observe that

' ± BÐxÑ
BF ± d.ÐxÑ
B$ $
Ÿ 2
' ± BÐxÑ
BÐx!Ñ ± d.ÐxÑ
B$
(2.6)
and that, by ÐH1Ñ
± B Ð x Ñ
B Ð x! Ñ ± Ÿ ' X \B Q $ Ð !Ñ
x
± kÐx,yÑ
kÐx!,yÑ ± ± aÐyÑ
aF ± ± fÐyÑ ± d.ÐyÑ Ÿ $

± Ð Ñ F$ ±± Рѱ
cO † dÐx! ,xÑ" ' d.ÐyÑ Ÿ
a y a f y
Ÿ X \B Q $ Ð !Ñ .Ð Ð ! Ñц Ð ! Ñ"
x B x ;y d x ,y

"Î<w
± Ð Ñ F$ ±<w
cO † $ " ' Ð ! Ñц Ð ! Ñ" d.ÐyÑ
a y a
Ÿ X \B Ð
Q$ !xÑ . Ð B x ;y d x ,y
†

"Î<
± Рѱ<
† ' Q $ Ð !Ñ .Ð Ð ! Ñц Ð ! Ñ" d.ÐyÑ . (2.7)
f y

X \B x B x ;y d x ,y

Moreover:
' ± Рѱ<
X \B Q $ Ð !Ñ .Ð Ð ! Ñц Ð ! Ñ"
x B x ;y
f y

d x ,y
d.ÐyÑ œ
_ ± Рѱ<
œ !' 4 $Ÿ Ð ! Ñ 4" $ .Ð Ð ! Ñц Ð ! Ñ" d.ÐyÑ Ÿ
f y

4œ"
M d x ,y M B x ;y d x ,y

5
 _
Ÿ ! Ò 4 $Ó" † .Ð Ð ! † 4 $ÑÑ
1 1
' <
Ð ! 4" $Ñ ± fÐyÑ ± d.ÐyÑ Ÿ
4œ"
M B x ,2 M B x ,M

(by the doubling condition) Ÿ cÐc. , M," Ñ † $ † MÐ ± f ± < ÑЖxÑ. 1

" (2.8)
Similarly:
' ± Ð Ñ F$ ±<w
Q $ Ð !Ñ .Ð Ð ! Ñц Ð ! Ñ" d.ÐyÑ Ÿ
a y a

X \B x B x ;y d x ,y

_
Ÿ cÐc. Ñ † ! ' <w
Ò 4 $Ó" †
Ð ! 4" $Ñ ± aÐyÑ
aF$ ± d.ÐyÑ Ÿ
1

4œ"
M B x ,M

(by Theorem 1.5 and Lemma 1.7)

_ <w
Ÿ $"1
! cÐr,K,XÑ ² a ² <‡w Š 1 j
" 4‹ œ
1
"
w
cÐr,K,XÑ † $ † ² a ² ‡< . (2.9)
4œ"
M

Finally (2.2) follows from (2.3)

(2.9), and the theorem is proved.

As an immediate consequence of Theorems 2.4, 2.3 and 1.6, we obtain the _: -estimate (2.1) in the
case .ÐXÑ œ _:
Theorem 2.5. Let X be a homogeneous space with .ÐXÑ œ _, K a CZ operator, a − BMO. Then for every
p − Ð1,_Ñ the operator T+ can be continuously extended to _: , and
² T+ f ² : Ÿ cÐp,K,XÑ ² a ² ‡ ² f ² : .
Let us note that in the case .ÐXÑ  _, applying Theorem 1.4 to Theorem 2.4 we get:
' T+ f ² : Ÿ cÐp,K,XÑ ² a ² ‡ ² f ² :
² T+ f

X
(2.10)
for every a − __ , f − W, p − Ð1,_Ñ. In the next section we will show how to get the desired estimate from
this result.

3. ESTIMATE ON THE COMMUTATOR OF CALDERON-ZYGMUND OPERATORS

ON FINITE MEASURE HOMOGENEOUS SPACES
Throughout this section X will be a finite measure homogeneous space, unless otherwise stated. Let us
state the main result we are going to prove:
Theorem 3.1. Let ÐX,d,.Ñ be a homogeneous space and K a CZ operator on X such that:
3Ñ .Ð\Ñ  _ and . is regular;
iiÑ the kernels k and k ‡ satisfy condition (H1) (where k ‡ Ðx,yÑ œ kÐy,xÑÑ;
iiiÑ k satisfies the growth condition:
± kÐx,yÑ ± Ÿ .Ð Ð ÑÑ for every x, y − X, some constant c;
c
B x;y
(3.1)
Then for every a − BMO, p − Ð1,_Ñ, the operator T+ can be continuously extended to _: and
² T+ f ² : Ÿ
c²a²‡²f²:
with c depending on p, X, the _: norm of K and the constants of k and k ‡ appearing in conditions (H1),
(3.1).
Let us sketch the line of the proof of Theorem 3.1. By (2.10), to get the result we need to estimate
' ÐT+ fÑÐxÑ d.ÐxÑ.

X

Since the function T+ f cannot be written explicitely using our "abstract" definition of CZ operator, we shall
consider suitable "approximating" integral operators K% , and estimate the mean value of the commutators C
~
ÒK% ,aÓÐfÑ. Finally, we shall prove that K% converges, in an appropriate weak sense, to some operator K which
is not necessarily K, but has the same commutator.
The first step is the following:
Theorem 3.2. Suppose that k and k ‡ satisfy condition (H1), ' ± kÐx,yÑ ± d.ÐyÑ is a bounded function and
the same is true for k ‡ . Set, for f − W :
X

6
 KfÐxÑ œ ' kÐx,yÑfÐyÑ d.ÐyÑ
X

K ‡ fÐxÑ œ ' k ‡ Ðx,yÑfÐyÑ d.ÐyÑ.

X

Let a − __ , T+ and T+‡ be the commutators of K and K ‡ , respectively. Then, for every p − Ð1,_Ñ there
exists a positive constant c, depending on p, X, the _: norms of K and K ‡ , and the constants appearing in
condition (H1) for k and k ‡ , such that
² T+ f ² : Ÿ cÐp,K,XÑ ² a ² ‡ ² f ² : .
Proof. By (2.10) we have that for every a − __ , f − W:
² T+ f ² : Ÿ c ² a ² ‡ ² f ² :  .ÐXÑ"Î:" ± ' ÐT+ fÑÐxÑ d.ÐxÑ ± .
X
(3.2)
Therefore, let us consider:
' ÐT+ fÑÐxÑ d.ÐxÑ
X
œ ' d.ÐxÑ' kÐx,yÑ ÒaÐxÑ
aÐyÑÓ fÐyÑ d.ÐyÑ.
X X

Since a, f are bounded functions and .ÐXÑ  _, the integral on the right hand side of the last equation
converges absolutely and equals

' fÐyÑ d.ÐyÑ' kÐx,yÑ ÒaÐyÑ
aÐxÑÓ d.ÐxÑ œ
' fÐyÑ T+‡ 1 ÐyÑ d.ÐyÑ.
X X X

Hence:
± ' ÐT+ fÑÐxÑ d.ÐxÑ ± Ÿ ² f ² : ² T+‡ 1 ² :w .
X
(3.3)

Since also k ‡ satisfies (H1), (2.10) holds for T+‡ , too, and
w
² T+‡ 1 ² :w Ÿ c ² a ² ‡ .ÐXÑ"Î:  .ÐXÑ"Î: ± ' ÐT+‡ 1ÑÐxÑ d.ÐxÑ ± .
X
(3.4)
Again by Fubini's theorem, we get:
' ÐT+‡ 1ÑÐxÑ d.ÐxÑ œ ' ÐKaÑÐyÑ d.ÐyÑ
' ÐK ‡ aÑÐxÑ d.ÐxÑ.
X X X
(3.5)
¨
By Holder and the _: estimate on K:
w w
± ' ÐKaÑÐyÑ d.ÐyÑ ± Ÿ ² Ka ² : † .ÐXÑ"Î: Ÿ c ² a ² : † .ÐXÑ"Î: .
X
(3.6)
Without loss of generality, we can assume
' a Ð xÑ d . Ð x Ñ œ 0
X

since the commutator we are estimating is not affected by adding a constant to a. We claim that
² a ² : Ÿ c: ² a ² ‡ .ÐXÑ"Î: .
To see this, recall that, by Lemma 1.9, X is bounded, so it coincides with some ball B. Then, by Theorem
1.5,
' ± aÐxÑ ± : d.ÐxÑ œ .ÐXÑ †

X
' ± aÐxÑ
aF ± : d.ÐxÑ Ÿ c .ÐXÑ ² a ² :‡ .
B

Therefore from (3.6) we get:

± ' ÐKaÑÐyÑ d.ÐyÑ ± Ÿ
X
c ² a ² ‡ . ÐX Ñ (3.7)
An analogous estimate holds for the second term in (3.5) and from (3.2)
(3.7) the result follows.

The second step is the construction of suitable "truncated operators" to which we will apply the
previous theorem. Since we will need this construction also in § 4, while considering general homogeneous
spaces (not necessarily of finite measure) we drop for a moment the assumption .ÐXÑ  _. By Theorem
1.3, we can endow X with a quasidistance $ equivalent to d and locally Holder
¨ of exponent !. Consider the
functions <3 : Ò0,_Ñ Ä Ò0,1Ó (i œ 1, 2) defined by:
Ú! for t Ÿ "
<" ÐtÑ œ Û t
" for "  t  2
Ü" for t   2.
Ú" for t Ÿ 2
<# ÐtÑ œ Û 3
t for 2  t  3
Ü! for t   3.

7
For any % − Ð0,1Ñ set
< Ð Ñ for 0 Ÿ t Ÿ 2
< % Ð tÑ œ œ " %
t

<# Ð%tÑ for t   2

k% Ðx,yÑ œ kÐx,yÑ † <% Ð$ Ðx,yÑÑ. (3.8)
Lemma 3.3. Suppose ! − Ð0,1Ó is a number for which Theorem 1.3 (ii) holds for X, and K satisfies
condition (H1) with exponent " ; moreover, suppose k satisfies the growth condition (3.1). Then k% defined
above satisfies condition (H1), with constants independent of % . More precisely, the exponent appearing in
(H1) is # œ minÐ!," Ñ.
Remark 3.4. Observe that if k satisfies (H1) and (3.1) with respect to d and ., then it satisfies similar
conditions with respect to $ and . (for any quasidistance $ equivalent to d); in particular, (H1) holds with
the same exponent " . Therefore to prove Lemma 3.3, we don't need to distinguish d and $ : henceforth d
will be the quasidistance of X that satisfies a Holder
¨ condition of exponent ! (recall Theorem 1.3).
Proof of Lemma 3.3. Let us start by noting that the triangle inequality implies that if x − B< Ðx! Ñ and
y  BQ< Ðx! Ñ, there exist two positive constants c" , c# depending on c. , M such that
c" dÐx,yÑ Ÿ dÐx! ,yÑ Ÿ c# dÐx,yÑ Ð3.9Ñ
(assuming M  c. ).
Now, let x! − X, r  0, x − B< Ðx! Ñ, y  BQ< Ðx! Ñ.
± k% Ðx! ,yÑ
k% Ðx,yÑ ± Ÿ
Ÿ ± <% ÐdÐx! ,yÑÑÒkÐx! ,yÑ
kÐx,yÑÓ ±  ± kÐx,yÑÒ<% ÐdÐx! ,yÑÑ
<% ÐdÐx,yÑÑÓ ± œ I  II.
By definition of <% and (H1)
I Ÿ .Ð Ð O! ÑÑ † ÐÐ !! ÑÑ .
"
c d x ,x

B x ;y d x ,y
"

Also, observe that the quantity

± <% ÐdÐx! ,yÑÑ
<% ÐdÐx,yÑÑ ± ´ A
can be nonzero only if one of the following cases occurs:
1=> case: one of the numbers dÐx! ,yÑ and dÐx,yÑ is in Ð0, 2%Ñ.
28. case: one of the numbers dÐx! ,yÑ and dÐx,yÑ is in Ð 2%, % Ñ, the other is  2%. 3

1=> case.
A Ÿ ± Ð ! Ñ% Рѱ Ÿ (by Theorem 1.3)
d x ,y d x,y

"

c † dÐx! ,xÑ! † Ò Ð ! Ñ % Ð ÑÓ Ÿ
!
Ÿ
d x ,y d x,y
(by (3.9))

c † ÐÐ !! ÑÑ .
!
Ÿ
d x ,x

d x ,y
!

28. case.
A Ÿ c † % dÐx! ,xÑ! ÒdÐx! ,yÑ"!  dÐx,x! Ñ"! Ó Ÿ (by (3.9))

c † ÐÐ !! ÑÑ .
!
Ÿ
d x ,x

d x ,y
!

Therefore
II Ÿ .Ð Ð ! ÑÑ † ÐÐ !! ÑÑ ,
!
c d x ,x

B x ;y d x ,y
!

and the theorem is proved.

Let's now consider the operators:
K% fÐxÑ œ ' k% Ðx,yÑ fÐyÑ d.ÐyÑ.
X
(3.10)
Note that condition Ð3.1Ñ and the doubling condition imply

' '%  Ð .Ð Ñ
± k% Ðx,yÑ ± d.ÐyÑ Ÿ Ñ 3% .Ð Ð ÑÑ Ÿ cÐ%Ñ uniformly in x
d y

X d x,y B x;y

8
and the same is true for k ‡ . In particular, for f − W, K% f and K%‡ f are well defined and bounded.
The following _: continuity estimate on K% with constants independent of % , may be obtained
adapting to our context the proof of a known result by M. Cotlar, regarding Calderon-Zygmund ´ operators
on ‘8 (see for instance [Me], p. 241).
Theorem 3.5. Let K be a CZ operator; let the kernel k satisfy (3.1); let K% be defined as in (3.8), (3.10).
Then for every q − Ð1,_Ñ there exists a constant c independent of % such that:
"Î;
± K % f Ðx Ñ ± Ÿ c
MÐKfÑÐxÑ  ŠMÐ ± f ± ; ÑÐxÑ‹  MfÐxÑŸ for every x − X, f − _; .

Corollary 3.6. Under the same assumptions of Theorem 3.5, let

K æ fÐxÑ œ sup ± K% fÐxÑ ± .
%!
æ : :
Then K is continuous from _ to _ , for every p − Ð1,_Ñ.
The following proposition is now a consequence of Theorem 3.2, Lemma 3.3 and Corollary 3.6.
Proposition 3.7. Let K be a CZ operator; let the kernels k and k ‡ satisfy condition (H1); moreover, suppose
k satisfies (3.1). Then the following estimate holds
² CÒK% , aÓÐfÑ ² : Ÿ c ² a ² ‡ ² f ² :
_
for every p − Ð1,_Ñ, a − _ , f − W, with c depending on p and the constants involved in the assumptions,
but not on %.
Next, we need to specify in what sense the operators K% approximate K.
Theorem 3.8. Let K and K% be as in Proposition 3.7, and assume that the measure . is regular. Then there
exist a sequence %8 Ä 0 and a function b − __ such that for every p − Ð1,_Ñ, every f, g − W we have:
' f K% 8 g
X
Ä ' f Kg
X
 ' bfg for n Ä _.
X

´
A similar result has been proved for Calderon-Zygmund operators in ‘8 (see for instance [M], p.
227). The proof of this result, thought, uses tools charachteristic of ‘8 (Fourier transform, dyadic
decompositions etc.) and therefore it cannot be repeated in our context. Our proof, on the other hand, relies
entirely on topological and measure theorethical properties.
Proof. We know that:
² K% f ² : Ÿ c: ² f ² : (3.11)
for every p − Ð1,_Ñ, f − W, and so for every f − _: , with c: independent of % .
Observe that the space _: is separable. To see this, note that since X is separable (as we noted after
Definition 1.2) and the balls form a base for the topology, X has a countable base. Also, the measure is
regular, hence Borel sets can be approximated, in measure, with open sets, so that the Borel 5 -algebra is
separable, and _: is separable for every p − Ò"ß _Ñ.
From (3.11), by compactness, we can construct a sequence %8 Ä 0 such that K% 8 f converges weakly in
_: , for n Ä _, for every f belonging to some dense subset of _: . But then, again by (3.11),
' g K% 8 f Ä
~
' g Kf (3.12)
X X

w ~
for every f − _: , g − _: , in particular for every f, g − W, with K:_: Ä _: linear continuous.
~
In order to study the relationship between K and K, let f, g − W such that distÐsprt f, sprt gÑ   %!  0.
Then:
' Ðf K% 8 gÑÐxÑ d.ÐxÑ
X
œ ' sprt f
fÐxÑ d.ÐxÑ' sprt g
k% 8 Ðx,yÑgÐyÑ d.ÐyÑ œ

(for every %8  %#! Ñ

œ ' Ðf KgÑÐxÑ d.ÐxÑ.

X

Taking the limit for n Ä _, we have

~
' f Kg œ ' f Kg. (3.13)
X X

9
 Now fix g − W, let C be its support. Set C8 œ š x − X: distÐx,CÑ   1
›; then -C8 œ X\C. If
n
8
E © C8 , E a Borel set, taking f œ ;I in (3.13) we get
~ œ ' Kg,
' Kg E E

~ œ Kg a.e. in X\C. Therefore the operator

i.e. Kg
K ´ K~
K, "
is local, that is
for every g − W, K" gÐxÑ œ 0 for a.e. x  sprt g.
Let us now show that K" is the multiplication by a bounded function. Let E be a Borel set, C8 an
increasing sequence of closed sets and A8 a decreasing sequence of open sets such that
.ÐA8 \C8 Ñ Ä 0, C8 © E © A8 . Set:
bÐxÑ ´ ÐK" 1ÑÐxÑ œ ÐK" ;G8 ÑÐxÑ  ÐK" ;E8 G8 ÑÐxÑ  ÐK" ;E-8 ÑÐxÑ
\ (3.14)
and take limits in (3.14) for n Ä _. Since .ÐA8 \C8 Ñ Ä 0, K" ;E8 G8 Ä 0 in _: , and a subsequence
\

converges a.e. Since ;G8 Ä ;I in _: , K" ;G8 Ä K" ;I in _: and a subsequence converges a.e. Moreover
ÐK" ;E-8 ÑÐxÑ œ 0 for a.e. x − E, since K" is local. Then for a.e. x − E we have:
ÐK" ;I ÑÐxÑ œ bÐxÑ.
On the other hand ÐK" ;G8 ÑÐxÑ œ 0 for a.e. x  E, and therefore ÐK" ;I ÑÐxÑ œ 0 for a.e. x  E. This implies
that for every Borel set E
ÐK" ;I ÑÐxÑ œ bÐxÑ † ;I ÐxÑ
and, by linearity, an anologous identity holds for every simple function and so, by density, for every function
in W:
~
ÐKf
KfÑÐxÑ œ ÐK" fÑÐxÑ œ bÐxÑ † fÐxÑ for every f − W. (3.15)
Moreover b − __ since for every f, g − W
± ' ÐbfgÑÐxÑ d.ÐxÑ ± œ ± ' Ðg K" fÑÐxÑ d.ÐxÑ ± Ÿ
X X
c ² f ² : ² g ² :w ,
w
and, taking h − W, h   0, and letting f œ h"Î: , g œ h"Î: we have:
± ' ÐbhÑÐxÑ d.ÐxÑ ± Ÿ
X
c²h²" (3.16)
From (3.12), (3.15), (3.16) the theorem follows.

The previous theorem can also be read as a "representation theorem for CZ operators": if K is a CZ
operator, there exist a sequence %8 Ä 0 and a function b − __ such that for every f, g − W we have:

' Ðf KgÑÐxÑ d.ÐxÑ œ lim ' fÐxÑ d.ÐxÑ' k%8 Ðx,yÑgÐyÑ d.ÐyÑ  ' ÐbfgÑÐxÑ d.ÐxÑ.
X
8 X X X

In other words, the "abstract" operator K, allows the above "concrete" representation; note, however, that if
we only know that K satisfies definition 2.1 with kernel k, the function b remains unknown and K is not
really determined; on the other hand, the commutator of K is uniquely determined by the kernel.
Proof of Theorem 3.1. Let a − __ , f, g − W. By Theorem 3.8 we have:
' g C ÒK% 8 , aÓÐfÑ
X
œ

œ ' Šg a K% 8 f
gK% 8 ÐafÑ‹
X
Ä ' šg a ŠKf  bf‹
gŠKÐafÑ  baf‹›
X
œ ' g T+ f.
X

By Proposition 3.7 we have:

±' g T+ f ± œ lim ± ' g C ÒK% 8 , aÓÐfÑ ± Ÿ c ² g ² :w ² a ² ‡ ² f ² : .
X
8 X

Hence, by the density of W in _: :

² T+ f ² : Ÿ c ² a ² ‡ ² f ² :
_
for every a − _ , f − W. By Proposition 2.3, the theorem follows.

4. _: CONTINUITY OF SINGULAR INTEGRAL OPERATORS.
EXAMPLES AND APPLICATIONS
10
 In this section we want to show how our results about the commutator of a CZ operator we have
proved so far (i.e. Theorem 2.5 in the case .ÐXÑ œ _, Theorem 3.1 in the case .ÐXÑ  _) can be applied
to given concrete singular integral operators. The problem here is to assure that the kernel k actually defines
a singular integral operator, continuous on _: . Our goal is to prove this fact under some natural
assumptions on the kernel. This is possible basically through two known theorems, which will be recalled
later. Our main results are contained in Theorems 4.1, 4.2, 4.3, 4.5.
Theorem 4.1. Let ÐX,d,.Ñ be a homogeneous space, . a regular measure. Let k: X ‚ X\Öx œ y× Ä ‘ a
kernel satisfying the following conditions:
the growth condition (3.1); ÐGÑ
¨
the Hormander condition with respect to both variables, i. e. both k and k ‡ satisfy (H1); ÐHÑ
the cancellation property with respect to both variables: ÐCÑ
there exists C  0 such that for every r, R, 0  r  R  _, a.e. x
± '  Ð Ñ
r d x,y R
kÐx,yÑ d.ÐyÑ ± Ÿ C,
and the same is true for k ‡ .
Let k% be the truncated kernel defined as in § 3, and for any f − V(! , set
K% fÐxÑ œ ' k% Ðx,yÑ fÐyÑ d.ÐyÑ.
X

Then K% can be extended to a linear continuous operator from _: into _: for every p − Ð1,_Ñ, and
² K% f ² : Ÿ c ² f ² : for every f − _: ,
where the constant c depends on X, p and all the constants involved in the assumptions, but not on % .
Theorem 4.2. Suppose all the assumptions of Theorem 4.1 hold. Then there exist a sequence %8 Ä 0 and a
CZ operator K such that for every f, g − W
' f † K% 8 g Ä ' f † K g for %8 Ä 0.
X X

Moreover, if K" , K# are two operators obtained in this way by two different sequences %8 , there exists a
function b − __ such that
K " f
K# f œ b † f
for every f − _: , and therefore K" and K# have the same commutator (for which the results of sections 2-3
hold).
In the particular case of antisymmetric kernels, the previous result can be improved:
Theorem 4.3. Suppose all the assumptions of Thm. 4.1 hold. Moreover, let
kÐx,yÑ œ
kÐy,xÑ for every x, y − X, x Á y.
Then for any f, g − V(! ,
ØK% f,gÙ Ä ØKf,gÙ ´ "# ' X
' kÐx,yÑÒfÐyÑgÐxÑ
fÐxÑgÐyÑÓ d.ÐxÑ d.ÐyÑ
X
Ð4.1Ñ
for % Ä 0. In particular, the CZ limit operator is unique.
For a kernel which is not necessarily antisymmetric, we are interested in getting a better result about
the convergence of K% , that is uniqueness of the limit and strong convergence in _: . These facts can be
proved under stronger assumptions on k% . Before stating the result, we need one more
Definition 4.4. Let ÐX,d,.Ñ be an unbounded homogeneous space. For every x − X, define
rB œ sup šr  0: B< ÐxÑ œ Öx×›
(with the convention supg œ 0). We say that ÐX,d,.Ñ satisfies a reverse doubling condition if there exist
cw  1, M  1 such that for every x − X, r  rB :
.
.ÐBQ< ÐxÑÑ   cw. † .ÐB< ÐxÑÑ. Ð4.2Ñ
It can be proved that, under some general additional geometric assumptions on the space ÐX,dÑ, (4.2) is
actually a consequence of the doubling condition on .. (See [BC2]).

11
Theorem 4.5. Let ÐX,d,.Ñ be a homogeneous space such that . is regular and, if X is unbounded, the
reverse doubling condition (4.2) holds. Let k be a kernel satisfying all the assumptions of Theorem 4.1.
Moreover assume that
for a.e. x − X there exists
lim ' k% Ðx,yÑ d.ÐyÑ ´ bÐxÑ. ÐL Ñ
% Ä 0 Ð Ñ d x,y 1

Then for every f − _: , p − Ð1,_Ñ, K% f converges (strongly) in _: for % Ä 0.

Remark 4.6. (Homogeneous spaces with a quasisymmetric quasidistance).
Although, in the definition of homogeneous space, we have required the symmetry of . , the above results
still hold if . is only quasisymmetric, that is satisfies:
.ÐCß BÑ Ÿ - † .ÐBß CÑ Ð4.3Ñ
for some positive constant - , every Bß C − \ . (See Examples B and C in section 4.b for applications in
which the natural quasidistance is quasisymmetric).
Note that, if . satisfies Ð3Ñß Ð333Ñ of Definition 1.1 and (4.3), setting
. w (Bß C) œ . (Bß C)  .ÐCß BÑ
we get a symmetric quasidistance . w , equivalent to . . Therefore if 5ÐBß C) is a kernel satisfying condition
(G) or (H) with respect to (\ß .ß .Ñ, it satisfies the same condition (possibly with other constants) with
respect to (\ß . w ß .Ñ, which is a homogeneous space. Moreover, if it satisfies (C) and (G) with respect to
(\ß .ß .Ñ, let, for !  <  V  _:
G" œ šCÀ <  . w (Bß C)  V ›; G# œ šCÀ <  . (Bß C)  V /("  -Ñ›;

G$ œ šCÀ </("  - ) Ÿ . (Bß C) Ÿ <›; G% œ šCÀ V /("  - ) Ÿ . (Bß C) Ÿ ("  -ÑV ›,

with - as in (4.3). Then by (C), (G) and the doubling condition,

¸'G" 5ÐBß CÑ . .ÐCѸ œ ¸'G# 5ÐBß CÑ . .ÐCÑ  'G" G# 5ÐBß CÑ . .ÐCѸ Ÿ

\

Ÿ -  'G$ ± 5ÐBß CÑ ± . .ÐCÑ  'G% ± 5ÐBß CÑ ± . .ÐCÑ Ÿ const.

So 5 satisfies (C) with respect to (\ß . w ß .Ñ.

Analogously, we can see that if 5 satisfies (L) and (G) with respect to (\ß .ß .Ñ, then it satisfies (L)
with respect to (\ß . w ß .Ñ.
As a consequence, Theorems 4.1, 4.2, 4.3, 4.5 still hold if in the definition of homogeneous space we
weaken the assumptions, asking . to be quasisymmetric, instead of symmetric; moreover, the same is true
for the commutator estimate (Theorems 2.5-3.1).
All the theorems we have stated so far depend on the following two known results.
Theorem 4.7, by Coifman-de Guzmàn. (See [CG]. A detailed proof is exposed in [CW1]).
Let ÐX,d,.Ñ be a homogeneous space, k: X ‚ X Ä ‘ a measurable function satisfying an integral
¨
Hormander condition (H0) (see Proposition 2.2). Assume that the operator
KfÐxÑ œ ' kÐx,yÑ fÐyÑ d.ÐyÑ
X

is continuous from _# to _# . Then for every p − Ò1,2Ó there exists a constant c: depending on p, the _#
norm of K and the constants appearing in (H0Ñ, such that for every f − _#  _: :
² Kf ² : Ÿ c: ² f ² : if 1  p Ÿ 2;
² ²"
.šx − X: ± KfÐxÑ ±  t› Ÿ c" for every t  0, if p œ 1.
f

If also k ‡ satisfies condition (H0), then, by duality, we get that K is also _: continuous for p − Ð2,_Ñ.
Note that the theorem cannot be directly applied to a singular kernel, that is with the growth allowed
by condition (G), because the integral of such a kernel with an _# function does not converge.

12
 The second result we need is an extension to homogeneous spaces of the so called "TÐ1) theorem" by
David-Journe,´ [DJ]. The extension we use is due to Christ, [Chr]. (A less general extension was given by
´
David-Journe-Semmes in [DJS]). To state the theorem we need some more definitions.
Let ( − Ð0,1Ó such that V(! ÐXÑ is dense in _: ÐXÑ for every p − Ò1,_Ñ, and let ÐV!( Ñw be the dual space of
V(! .
Definition 4.8. We say that a linear continuous operator T: V(! Ä ÐV(! Ñw , continuous in the natural
topologies, satisfies the weak boundedness property (WBP) if there exists a constant c  0 such that for any
x! − X, r  0, f, g − V!( with support contained in B< Ðx! Ñ, ² f ² _ , ² g ² _ Ÿ 1, ² f ² ( , ² g ² ( Ÿ r( :
± ØTf,gÙ ± Ÿ c † .ÐB< Ðx! ÑÑ. ÐWBPÑ
Theorem 4.9, by Christ (see [Chr]).
Let ÐX,d,.Ñ be a homogeneous space, T: V(! Ä ÐV!( Ñw a linear continuous operator satisfying (WBP).
Suppose there exists a kernel k: X ‚ X\Öx œ y× Ä ‘, satisfying (G) and (H) (see Thm. 4.1) such that for
any f, g − V(! with disjoint supports
ØTf,gÙ œ ' ' kÐx,yÑ fÐyÑ gÐxÑ d.ÐyÑ d.ÐxÑ.
X X

Moreover suppose that TÐ1Ñ and T ‡ Ð1Ñ belong to BMO. Then T can be extended to a continuous operator
from _# into _# .
Remark 4.10. The definition of TÐ1Ñ and T ‡ Ð1Ñ are not in general trivial. However we will apply this
theorem to the truncated operators K% , so that the definition of K% 1 and K%‡ 1 will be clear.

4.a. PROOF OF THEOREMS 4.1, 4.2, 4.3, 4.5

Proof of Thm. 4.1. Let us consider the truncated kernels k% . We are going to check that all assumptions of
Thm. 4.9 are fulfilled, with constants independent of % . Conditions ÐC) and (G) imply that K% 1 and K%‡ 1 are
bounded functions, uniformly in %, and therefore have BMO norms bounded uniformly in % . By (G) and the
doubling condition,
'
± k% Ðx,yÑ ± d.ÐyÑ Ÿ cÐ%Ñ uniformly in x,
X

so that K% fÐxÑ is well defined for any f − V(! ; if also g − V(! we can write
ØK% f,gÙ œ ' ' k% Ðx,yÑ fÐyÑ gÐxÑ d.ÐyÑ d.ÐxÑ
X X

where the integral is absolutely convergent. To estimate ØK% f,gÙ, let f, g − V(! , sprt f, sprt g § B< Ðx! Ñ for
some x! − X, r  0. Then
K% fÐxÑ œ ' Ð Ñ k% Ðx,yÑ ÒfÐyÑ
fÐxÑÓ d.ÐyÑ  fÐxÑ † ' Ð Ñ k% Ðx,yÑ d.ÐyÑ 
d x,y r d x,y r

 'Ð d x,y Ñ  k% Ðx,yÑ fÐyÑ d.ÐyÑ œ I  II  III.

r

c ² f ² ( '  Ð Ñ .Ð Ð Ð Ñ ÑÑ d.ÐyÑ Ÿ
(
± I ± Ÿ (by ÐGÑÑ
d x,y

0 d x,y r B x;y

_
(expanding the integral as ! '
8"
r
 Ð Ñ 2r8 á and using the doubling property)
d x,y
2
n=1

Ÿ c ² f ² ( † r( .
± II ± Ÿ ± fÐxÑ ± ± ' Ð Ñ d x,y r
k% Ðx,yÑ d.ÐyÑ ± Ÿ c ± fÐxÑ ± since:
'Ðd x,y Ñ k% Ðx,yÑ d.ÐyÑ œ
r

œ ' % Ð
2 d x,y Ñ kÐx,yÑ d.ÐyÑ  '% Ð ÑŸ % k% Ðx,yÑ d.ÐyÑ ´ A  B.
r d x,y 2

c by ÐC). ±A± Ÿ
Finally, ± B ± Ÿ '% Ð ÑŸ % ± kÐx,yÑ ± d.ÐyÑ Ÿ c by (G) and the doubling condition.
d x,y 2

± III ± Ÿ c ' Ð Ñ  .б ÐРѱ ÑÑ d.ÐyÑ.

d x,y r
f y

B x;y

13
Therefore:
±' X
gÐxÑ K% fÐxÑ d.ÐxÑ ± Ÿ
± Рѱ
Ÿ c' <Ð !Ñ ± gÐxÑ ± Š ² f ² ( † r(  ± fÐxÑ ±  ' Ð Ñ  .Ð Ð ÑÑ d.ÐyÑ ‹ d.ÐxÑ Ÿ
f y

B x d x,y r B x;y

Ÿ c ² g ² _ † .ÐB< Ðx! ÑÑ Š ² f ² ( † r(  ² f ² _  ² f ² _ † sup' Ÿ Ð ÑŸ † .Ð .ÐÐ Ñ ÑÑ ‹ . d y

x r d x,y c r B x;y

where we used the assumption on the supports of f and g. Since the last integral in brackets is bounded
uniformly in r, by the doubling, we get:
± ØK% f,gÙ ± Ÿ c † .ÐB< Ðx! ÑÑ ² g ² _ Š ² f ² ( † r(  ² f ² _ ‹
with c an absolute constant. This proves that K% : V!( Ä ÐV(! Ñw is continuous, and satisfies (WBC), with
constants independent of %. Then by Thm. 4.9, K% can be extended to a continuous operator from _# into
_# , with norm uniformly bounded in % . From this fact and condition (H), by Thm. 4.7, we also get
² K% f ² : Ÿ c: ² f ² : Ð4.4Ñ
for every p − Ð1,2Ó. Since the same reasoning can be applied to K%‡ , by duality we get that (4.4) actually
holds for every p − Ð1,_Ñ. This proves Thm. 4.1.

For seak of brevity, we skip the proof of Thm. 4.2, since it is similar to that of Thm. 3.8, with minor
changes in the case .ÐXÑ œ _.
To prove Thms. 4.3 and 4.5 we need the following technical lemma, which replaces in homogeneous
spaces the use of Young inequality to estimate some "convolution" integrals.
Lemma 4.11. Let f − _" , f boundedly supported, and set, for every r  0
g< ÐxÑ œ ' Ð Ñ- .б ÐРѱ ÑÑ d.ÐyÑ.
d x,y r
f y

B x;y

Then, for every p − Ð1,_Ñ and r  0, g< − _: ; moreover g< Ä 0 in _: for r Ä  _.

Proof. Suppose first that X is unbounded. Note that, by the doubling condition, .ÐBÐx;yÑÑ is comparable
with .ÐBÐy;xÑÑ:
.ÐBÐx;yÑÑ Ÿ cÐc. , c. Ñ † .ÐBÐy;xÑÑ. (4.5)
w
Let h − _: .
± Рѱ
± ' Ðh g< ÑÐxÑ d.ÐxÑ ± Ÿ ' ± fÐyÑ ± d.ÐyÑ' Ð Ñ- .Ð Ð ÑÑ d.ÐxÑ Ÿ
h x

X X d x,y r B x;y

¨
(by (4.5) and Holder)
"Î:
.Ð Ñ
Ÿ ² h ² :w † ' ± fÐ y Ñ ± ' Ð d.ÐyÑ.
d x

X d x,y Ñ- .Ð Ð ÑÑ:
r B y;x

Now
.Ð Ñ _ .Ð Ñ
'Ð Ñ- .Ð Ð ÑÑ: œ ! ' 8  Ð ÑŸ 8" .Ð Ð ÑÑ: Ÿ
d x d x

8œ!
d x,y r B y;x 2 r d x,y 2 r B y;x

_
Ÿ ! 8" ÑÑ:
" Ÿ
1

n œ ! .Ð Ð
n
B y,2 r

(by (4.2), with n! œ max Ön: 28 r  rC × if rC  r, n! œ 0 otherwise)

_
Ÿ Š! w Ð:
"ÑÐ8"Ñ ‹ † .Ð Ð ÑÑ:
" .
1 1

8œ! Ð .Ñ c B y,r

14
 Hence:
± ' Ðh g< ÑÐxÑ d.ÐxÑ ± Ÿ
X
cÐp,XÑ ² h ² :w ' .Рб РѱÑÑ"Î:w d.ÐyÑ .
X
f y

B y;r
(4.6)

Recall that f is boundedly supported, let sprt f © BV Ðx! Ñ. Let us show that for any fixed r  0
inf .ÐBÐy,rÑÑ œ c  0. (4.7)
y − B V Ðx ! Ñ
To see this, observe that, for every y − BV Ðx! Ñ, BV Ðx! Ñ © B#-. V ÐyÑ, so that, by the doubling condition,
0  .ÐBV Ðx! ÑÑ Ÿ cÐX, r, RÑ † .ÐB< ÐyÑÑ.
From (4.6)-(4.7) we have
± ' Ðh g< ÑÐxÑ d.ÐxÑ ± Ÿ
X
cÐp,XÑ ² h ² :w † cÐX, r, RÑ ² f ² " œ cÐp,X,r,fÑ ² h ² :w
hence g< − _: . To show that g< Ä 0 in _: for r Ä _ observe that for every r   1,
± g< ÐxÑ ± Ÿ ± g" ÐxÑ ± − _: . Moreover g< ÐxÑ Ä 0 a.e. for r Ä _, as we can see applying Lebesgue's
theorem to
g< ÐxÑ œ ' ; <Ð Ñ ÐyÑ .б ÐРѱ ÑÑ d.ÐyÑ.
X X \B x
f y

B x;y

Therefore, again by Lebesgue's theorem, g< Ä 0 in _: for r Ä _.

If X is bounded, by (4.7) g< is bounded for every r, and for r sufficiently large g< ´ 0. So the lemma is
proved.

Proof of Theorem 4.3. Since for the K% 's Theorem 4.1 holds, it is enough to check the weak convergence
(4.1). By antisymmetry and with a change of variables, for any fixed f, g − V(! with support contained in
some ball B:
ØK% f,gÙ œ "# ' ' k% Ðx,yÑÒfÐyÑgÐxÑ
fÐxÑgÐyÑÓ d.ÐxÑ d.ÐyÑ.
X X

Now:
.Ð Ñ
± ØK% f,gÙ
ØKf,gÙ ± Ÿ c † %( ' d.ÐxÑ'% Ð Ñ % .Ð Ð ÑÑ 
d y

B d x,y 2 B x;y

± Ð Ñ Ð Ñ± ± Ð Ñ Ð Ñ±
 ' d.ÐxÑ'  Ð Ñ d.ÐyÑ œ I  II.
f y g x f x g y

B
% %
2
d x,y
3
.Ð Ð ÑÑ
B x;y

Now
±I± Ÿ c † %( .ÐBÑ Ä 0
while ± II ± Ä 0 by Lemma 4.11.

Proof of Theorem 4.5. For f − V(! , let us write
K% fÐxÑ œ '% Ð Ñ k% Ðx,yÑ ÒfÐyÑ
fÐxÑÓ d.ÐyÑ  fÐxÑ † '% Ð Ñ k% Ðx,yÑ d.ÐyÑ 
d x,y 1 d x,y 1

 'ŸÐ 1 d x,y Ñ % k% Ðx,yÑ fÐyÑ d.ÐyÑ œ I%  II%  III% .

3

I% œ '% Ð Ñ % k% Ðx,yÑ ÒfÐyÑ

fÐxÑÓ d.ÐyÑ  ' % Ð Ñ kÐx,yÑ ÒfÐyÑ
fÐxÑÓ d.ÐyÑ œ
d x,y 2 2 d x,y 1

œ A%  B% .
With an estimate similar to that in the proof of Thm. 4.1, ± A% ± Ÿ c ² f ² ( † %( Ä 0 for % Ä 0. The
function B% ÐxÑ is boundedly supported and converges uniformly, as % Ä 0, to
'Ð
0 d x,y Ñ kÐx,yÑ ÒfÐyÑ
fÐxÑÓ d.ÐyÑ
1
Ð4.8Ñ

which is bounded by c ² f ² ( , as a similar estimate shows. So B% converges to (4.8) in _: .

Since f is compactly supported, our assumption (L) implies that II% Ä bÐxÑ fÐxÑ uniformly, and
therefore in _: . Here we used the fact that
'% Ð d x,y Ñ k% Ðx,yÑ d.ÐyÑ
1

15
has limit for % Ä 0. This follows from (L), as can be seen for instance approximating the cutoff function <%
, which enters the definition of k% , by piecewise constant functions.
To prove that III% converges in _: , let us show that ² III%
III( ² : Ä 0 for % , ( Ä 0. Suppose for
instance 0  %  (.
III%
III( œ '  Ð ÑŸ kÐx,yÑ fÐyÑ d.ÐyÑ  '  Ð Ñ k% Ðx,yÑ fÐyÑ d.ÐyÑ 
2
( d x,y
2
%
2
% d x,y
3
%

' 2
 Ð Ñ ( k( Ðx,yÑ fÐyÑ d.ÐyÑ.
d x,y
3 Ð4.9Ñ
(

Each term in (4.9) is majorized by

± Рѱ
c' Ð Ñ  d x,y
2
( .Ð Ð ÑÑ d.ÐyÑ
f y

B x;y

which converges to zero in _: , as ( Ä 0, by Lemma 4.11. This completes the proof of Thm. 4.5.

4. b. EXAMPLES OF SINGULAR INTEGRAL OPERATORS
ON HOMOGENEOUS SPACES
We start with the following
Remark 4.12. Whenever, in a homogeneous space, the following holds
c" rR Ÿ .ÐB< ÐxÑÑ Ÿ c# rR (4.10)
for every r  0, x − X, some positive constants c" , c# , N, condition (H1) can be rewritten as:
there exist cO  0, "  0, M  1 such that for every x! − X, r  0, x − B< Ðx! Ñ, y  BQ< Ðx! Ñ :

cO † Ð Ð! ! Ñ ÑR .
"
± kÐx! ,yÑ
kÐx,yÑ ± Ÿ (H2)
d x ,x

d x ,y
"

Example A. Singular integrals with mixed homogeneities, studied by Fabes and Rivière in [FR] and
applications to parabolic equations.
Let X œ ‘R , . the Lebesgue measure, d the distance defined as follows. Let !" ,á ,!R be N real
numbers, with 1 Ÿ !" Ÿ !# Ÿ á Ÿ !R ; the equation
R #
! 4
#!4 œ
x
1
4œ" 3

allows for every x − ‘R \{0} a unique positive solution 3ÐxÑ. Set 3Ð0Ñ œ 0 and define dÐx,yÑ œ 3Ðx
yÑ. It
can be proved (see [FR]) that d is actually a distance. Introducing the polar type change of variables
Ú œ 3! "
Ý x"
Ý cos :" á cos :R# cos :R"
x# œ 3! # cos :" á sin :R"
Ûá
Ý
Ý
Ü xR œ 3! R sin :"
R
we find dx œ 3!" d3 d5 , with ! œ ! !4 and d5 the surface meaure on !R" œ Ö ± x ± œ 1×. Therefore
4œ"
we can compute:
.ÐB< ÐxÑÑ œ cÐNÑ † r! for every r  0. Ð4.11Ñ
In particular (4.10) holds, and (‘R ,d,dx) is a homogeneous space. The unit ball with respect to d is the
euclidean unit ball. Let
kÐx,yÑ œ 3HÐÐ  ÑÑ! , x

x
y

y
Ð4.12Ñ

where the function H has the following mixed homogeneity of degree zero
HÐt!" x" , t!# x# ,á , t!R xR Ñ œ HÐx" ,á ,xR Ñ, for every t  0, x − ‘R Ð4.13Ñ
¨
and satisfies the Holder condition:
there exist c  0, " − Ð0,1Ó such that for every x, y − !R" :
16
 ± H Ð x Ñ
H Ð yÑ ± Ÿ c ± x
y ± " . (4.14)
It is easy to check that the following condition (H2) is fulfilled:
if !, " are as above, there exists a constant c such that for every x! − ‘R , x − B< Ðx! Ñ, y  B#< Ðx! Ñ:

c † Ð Ð! ! Ñ Ñ .
"
± kÐx! ,yÑ
kÐx,yÑ ± Ÿ (4.15)
d x ,x

d x ,y
" !

The growth condition (G) follows from (4.11) and the definition of k. Moreover, if we assume that H
satisfies the vanishing property
'! HÐxÑ d5 ÐxÑ œ 0, (4.16)
R"
then conditions (C) and (L) are trivial. Therefore, by Thms. 4.1, 4.5, for every p − Ð1,_Ñ it is well defined
KfÐxÑ œ lim ' k% Ðx,yÑ fÐyÑ dy,
%Ä0
where the limit exists in _ sense, and K is continuous. (This result was proved with a direct approach in
:
[FR]). For this operator the commutator estimate of Thm. 2.5 holds.
Kernels of the above kind arise for instance considering parabolic operators with constant coefficients:
8
P œ ! +34 `0#304
`> ,
3ß4œ"

where +34 is a symmetric and positive matrix. Here B ´ Ð0 ß >Ñ − ‘8 " ´ ‘R (we set R œ 8  "). Let >! be
the fundamental solution for P, with pole at the origin, and let
5ÐBß CÑ œ Ð`0#304 >! Ñ(B
CÑ, for any 3ß 4 œ "ß á ß 8.
Then 5 is a kernel as in (4.12)-(4.13), with !" œ á œ !8 œ 1, !8 " œ 2. For any test function ? we can
write:
?ÐBÑ œ '‘8" >! ÐB
CÑ P?ÐCÑ .C
and, differentiating twice the above formula with respect to 0 we find
`0#304 ?ÐBÑ œ T ÞZ Þ'‘8" `0#304 >! ÐB
CÑ P?ÐCÑ .C  -Ð8Ñ † P?ÐBÑ.
Therefore, the _: -continuity of the singular integral implies _: -estimates for the (spacial) second
derivatives of ?, in terms of P? and ?.

Example B. Kolmogorov-type operators, studied by Lanconelli-Polidoro in [LP].
Let us consider the following linear partial differential operator:
; R
P ´ ! +34 `B3B4  ! ,34 B3 `B4
`> ´
3ß4œ" 3ß4œ"

;
´ ! +34 `B3B4  ØBß FHÙ
`> , H œ Ð`B" ,`B# ,á ,`BR Ñ
3ß4œ"

where ÐBß >Ñ − ‘R " , Ö+34 × is a constant, symmetric, positive ; ‚ ; matrix (3Þ/Þ the principal part is
uniformly elliptic on ‘; , ;  R ), and F is an R ‚ R constant matrix with the following structure:

Ô 0 F" ! á ! ×
ÖÖ 0 0 F# á ! Ù
Ù
F´ÖÖ 0ã 0ã 0ã ä ã Ù
Ù
á F<
Õ0 0 0 á 0 Ø

where for every 5 œ "ß á ß <, F5 is a matrix :5" ‚ :5 with rank :5 and ; œ :!   :"   á   :< ,
:!  :"  á  :< œ R . Under these assumptions, the operator P is hypoelliptic, and the matrix F induces
in the space ‘R " a (noncommutative) group of translations
ÐBß >Ñ ‰ ÐCß 7 ) œ (C  IÐ7 ÑBß >  7 ), where IÐ7 Ñ œ exp(
7 F X ),
17
such that P is invariant for left-translations. Moreover, there exists a group of dilations on ‘R " , which we
denote by šHÐ-Ñ› , where every HÐ-Ñ is a group homomorphism, acting as follows:
--!
HÐ-ÑÐBß >Ñ œ (-!" B" ,á ß -!R BR , -# >) Ð4.17Ñ
where !3 are suitable integers   1. The operator P is homogeneous of degree 2 with respect to the group of
dilations:
P Š?ÐHÐ-ÑDÑ‹ œ -# † HÐ-Ñ Š(P?)ÐDÑ‹ ,

0 "
with D œ ÐBß >Ñ. For R œ # and F œ ”
0 0•
we have the simplest example:

P œ `B#"  B" `B#

`> .
In this case the groups of translations and dilations are, respectively:
ÐB" ß B# ß >Ñ ‰ ÐC" ß C# ß 7 Ñ œ ÐB"  C" , B#  C#
7 B" ß >  7 Ñ;
HÐ-Ñ ÐB" ß B# ß >Ñ œ Ð-B" ß -$ B# ß -# >Ñ.
We can define a homogeneous norm, setting, for any D − ‘R " , D Á !,
²D² œ3 Í ± HÐ 3 ÑD ± œ 1,
1

with ± † ± œ euclidean norm; also, let ² ! ² œ 0. Then:

3Ñ ² HÐ-ÑD ² œ - ² D ² for every D − ‘R " , -  0;
33Ñ The set ÖD − ‘R " : ² D ² =1× is the euclidean sphere !R " ;
333Ñ For every D , ' − ‘R "
² D ‰ ' ² Ÿ -Ð ² D ²  ² ' ² Ñ for some constant - œ -ÐFÑ   1.
Defining:
.ÐDß ' Ñ œ ² ' " ‰ D ² ,
we have that . is a quasisymmetric quasidistance (see Remark 4.6). Since the Lebesgue's measure is left and
right-invariant under the group law, and
./> HÐ-Ñ œ -U # ,
with U œ !"  á  !R , !3 as in (4.17), we have that ± UÐDß <Ñ ± œ ± UÐ!ß <Ñ ± œ ± UÐ!ß 1Ñ ± <U # , for
every D − ‘R " and <  0. Therefore .D is a doubling measure with respect to the quasidistance . , and
(‘R " ,.Dß .Ñ is a homogeneous space.
Let >! be the fundamental solution for P, with pole at the origin, which is homogeneous of degree

U with respect to HÐ-Ñ, and let

5ÐDß ' Ñ œ Ð`B#3B4 >! Ñ(' " ‰ DÑ, for any 3ß 4 œ "ß á ß ; .
Then 5 is a G^ kernel on the above homogeneous space, for which the _: -estimate of Thms. 4.1, 4.5 and
the commutator estimate of Thm. 2.5 hold.

Remark 4.13. Applications of the commutator theorem to get a priori estimates for partial
differential operators. In the above examples, _: -continuity of singular integral operators implies _: -
estimates for (suitable) second derivatives of ?, in terms of P? and ?. The differential operator P which is
considered has constant coefficients or, more generally, is translation invariant, with respect to a suitable
group. The commutator estimate can also be used to get _: -estimates, for suitable classes of P.D.O.'s with
variable coefficients. This technique was used first by Chiarenza-Frasca-Longo [CFL1,2], who applied
Coifman-Rochberg-Weiss' Theorem to get _: estimates for uniformly elliptic operators in nondivergence
form, with Z Q S coefficients. The class Z Q S ("vanishing mean oscillation", see [Sarason]) is contained
in FQ S, and properly contains the space of uniformly continuous functions, so the _: theory for elliptic
operators with Z Q S coefficients is an interesting extension of the classical one. This kind of result has
been extended to the (uniformly) parabolic case by Bramanti-Cerutti [BC1], who proved a parabolic version
of CRW Theorem. In another paper, Bramanti-Cerutti-Manfredini [BCM], using the commutator theorem
in homogeneous spaces, prove _: estimates for a special class of ultraparabolic equations, of Kolmogorov-
Fokker-Planck type (see Example B for the translation invariant case). It should be noted that, although
18
CRW-type Theorem provides a very powerful tool, the _: -estimates for second derivatives for operators
with variable coefficients are not a direct consequence of it. In fact, in these cases, the representation
formula for the derivatives of ? is obtained by "unfreezing" the coefficients of P after having written the
representation formula for the operator with constant coefficients. Therefore the kernels one obtains are of
the kind 5ÐBß B
CÑ and are not themselves G^ kernels; instead, one has to expand 5 as a series in terms of
kernels that turn out to be G^ kernels. This technique requires a deep analysis of the properties of the
singular kernel 5ÐBß B
CÑ.
Example C. The Kohn-Laplacian on the Heisenberg group. (See [St3] for the proofs of the facts listed
below).
Let us consider ‚8 ‚ ‘ endowed with the group law:
Ð0 ,>Ñ ‰ Ð(,=Ñ œ Ð0  ( , >  =  # ImÐ0 † (–ÑÑ;
Ð0 ,>Ñ" œ Ð
0 ß
>Ñ.
Ђ8 ‚ ‘ß ‰ Ñ, or, equivalently, Б#8 " ß ‰ Ñ, is a Lie group, for which the Lebesgue measure in ‘#8 " is the
Haar measure. Moreover the dilations
HÐ-ÑÐ0 ,>Ñ œ Ð-0 ,-# >Ñ (-  0)
are group automorphisms. We can define in ‘ #8 " a homogeneous norm
"
± Ð0 ,>Ñ ± œ Ð ± 0 ± %  ># Ñ %
and a quasidistance
. ŠÐ0 ,>Ñß Ð(ß =Ñ‹ œ ± Ð(ß =Ñ" ‰ Ð0 ,>Ñ ± .
Then ‘#8 " , with this quasidistance and the Lebesgue's measure, is a homogeneous space, such that the
measure of a ball of radius < equals constant † <U , with U œ #8  #. (Therefore U is called the
homogeneous dimension of ‘#8 " ).
Consider now the Lie algebra canonically associated to the Lie group. To describe it, let us write, from
now on, Ð0 ,>Ñ œ ÐBß Cß >Ñ, with Bß C − ‘8 ; let \4 , ]4 , X , respectively, be the left-invariant vector fields on
‘#8 " that equals `Î`B4 , `Î`C4 , `Î`> at the origin. Then one can compute:
`
\4 œ `B4  #C4 `>` ;
`
]4 œ `C4
#B4 `>` ;
`
Xœ `> .
Moreover,
[\4 ß ]4 ] œ
%X for 4 œ "ß á ß 8
and all the other commutators vanish. The operator
8
_ œ
! Š\4#  ]4# ‹
4œ"

is the Kohn-Laplacian on the Heisenberg group; it is a hypoelliptic operator, homogeneous of degree two
with respect to the dilations HÐ-Ñ, and it has a fundamental solution >! , with pole at the origin, HÐ-Ñ-
homogeneous of degree
ÐU
#Ñ; moreover, >! is explicitely known:
>! Ð0 ,>Ñ œ -Ð8ÑÎÐ ± 0 ± %  ># Ñ8Î# .
For any test function ? we can write:
?ÐDÑ œ '‘#8" >! Ð' " ‰ DÑ _?Ð' Ñ . '
and, differentiating twice the above formula with respect to the vector fields \4 or ]4 we find, for example,
\3 \4 ?ÐDÑ œ T ÞZ Þ'‘#8" \3 \4 >! Ð' " ‰ DÑ _?Ð' Ñ . '  -Ð8Ñ † _?ÐDÑ.

The kernel 5ÐDß ' Ñ œ \3 \4 >! Ð' " ‰ D ) turns out to be a G^ kernel on the homogeneous space described
above; for the corresponding operator the _: estimates and the commutator estimate hold.

19
 We expect that this fact could be a starting point to get _: estimates for the "second derivatives" \3 \4
? for a class of operators with variable coefficients, modeled on the Kohn-Laplacian. (See Remark 4.13).

The next two examples are taken from [CW], pp.76-77.
Example D. Singular kernels on eucledean spaces with weighted measures. (This example is related to
some weighted singular integrals in ‘8 studied by Stein in [St1]).
Let X œ ‘8 , d. œ ± x ± # dx with 0 Ÿ #  n, d the eucledean distance. It is well known (see [GR],
p. 407) that the function ± x ± ! is an A: -weight for
n  !  nÐp
1Ñ, hence ± x ± # with 0 Ÿ #  n
is an A: -weight for every p − Ò1,_Ñ. In particular d. is doubling, so that X is a homogeneous space; one
can prove that d. satisfies also a reverse doubling condition. Note that d. is not translation invariant. Now,
let
kÐx,yÑ œ H±xÐxy±y8Ñ Š ± x ± #
± y ± # ‹ (4.18)

where H is a homogeneous function of degree zero (in the usual sense) satisfying (4.14), (4.15). In this case
(4.10) does not hold, but the following inequalities can be easily verified:
c±xy±8 ±x±# if ±xy±±x±/2
.ÐBÐx;yÑÑ Ÿ
(4.19)
c±xy±8# if ±xy± ±x±/2.

By (4.18) and (4.19), we can see that k satisfies condition (G). Moreover, one can prove that k satisfies (H)
with exponent " w œ minÐ" ,# Ñ. (" is the same number appearing in (4.15)). Finally, also (C) and (L) can be
proved, the last one following from the analogous result for classical CZ integrals on ‘8 . So also in this
case, by Thms. 4.1, 4.5, the kernel k defines a unique CZ operator, for which the commutator estimate
holds.
Example E. Singular integrals on the sphere. (The following operators, discussed in [CW1], are related to
the Riesz transform on the sphere, studied in [KV2]).
Let X œ D8" the surface of the unit ball in ‘8 , d the euclidean distance (in ‘8 ) and . the surface
measure. We know that
c" r8" Ÿ .ÐB< ÐxÑÑ Ÿ c# r8"
for every x − X, !  <  #, with 0  c"  c# .
Let us consider the kernels:
x3 y4  x4 y 3
k34 Ðx,yÑ œ ± x  y ±8

where i, j Ði Á jÑ are two fixed indexes between 1 and n. (In the following we will suppose these indexes
fixed once and for all, and we will drop them from k). One can check that kÐx,yÑ Ÿ c † ± x
y ± 8" , so that
(G) holds for k. Moreover, a condition (H2) is satisfied:
there exists c  0 such that for every x! − !8" , r  0, x − B< Ðx! Ñ, y  B#< Ðx! Ñ:
± kÐx! ,yÑ
kÐx,yÑ ± Ÿ c † ±±xx!!yx±±8 .
From this condition (H) follows. Note that, for every r, R  0,
'r±xy±R kÐx,yÑ d5 œ 0
so that ÐC) and (L) hold. Again, a unique CZ operator is well defined, for which the commutator estimate
holds, by Thm. 3.1.
Example F. This example exhibits a kernel which defines infinitely many CZ operators, all of them having
the same commutator. In other words, Thm. 4.1 and 4.2 apply, but Thm. 4.5 does not.
In ‘8 , with the usual distance and measure, define the convolution kernel:
kÐx,yÑ œ kÐx
yÑ with kÐxÑ œ 1
± x ±8 cos log ± x ± .
Then:
± k Ðx Ñ ± Ÿ 1
± x ±8 that is (G) holds;

20
 ± fkÐxÑ ± Ÿ 1
±x±8" , which implies (H);

'r±x±R kÐxÑ dx œ c8 † Ò sin log R

sin log rÓ,
so that (C) holds, but
lim '%±x±1 kÐxÑ dx does not exists.
%Ä0
By Thm. 4.1 we know that
² K% f ² : Ÿ c: ² f ² : for every p − Ð1,_Ñ
with c: independent of %. By Thm. 4.2 we also know that the commutator CÒK% ,aÓ converges weakly to
CÒK,aÓ , where K is any CZ operator of an equivalence class. For this commutator the estimate of Thm. 2.5
holds. Nevertheless, K% f does not converge in any sense to a CZ operator. We can actually compute, for any
sequence %5 Ä 0,
lim K%5 fÐxÑ œ '±xy±1 kÐx
yÑ ÒfÐyÑ
fÐxÑÓ dy  '±xy±-1 kÐx
yÑ fÐyÑ dy 
%5 Ä 0

c8 fÐxÑ † Ò lim sin log %5 Ó.
%5 Ä 0
Hence in general K%5 f does not converge, and there are sequences for which K%5 f Ä Kf  b † f, where K is
a fixed operator and b is any number in the interval Ò
c8 , c8 Ó.

References
[BC1] M. Bramanti-M. C. Cerutti: W:"ß# -solvability for the Cauchy-Dirichlet problem for parabolic
equations with VMO coefficients. Comm. in Part. Diff. Eq., 18 (9&10) (1993), 1735-1763.
[BC2] M. Bramanti-M. C. Cerutti: Commutators of fractional integrals on homogeneous spaces.
Preprint.
[BCM] M. Bramanti-M. C. Cerutti-M. Manfredini: _: -estimates for some ultraparabolic operators with
discontinuous coefficients. To appear on Journal of Math. Anal. and Appl.
[Bl] S. Bloom: A commutator theorem and weighted BMO. Trans. of A. M. S., 292, n. 1 (1985), 103-
122.
[Bu] ´ sur un espace de nature homogène.
N. Burger: Espace des fonctions à variation moyenne bornee
C. R. Acad. Sc. Paris, t. 286 (1978), 139-142.
[CFL1] F. Chiarenza-M. Frasca-P. Longo: Interior [ #ß: -estimates for nondivergence elliptic equations
with discontinuous coefficients. Ricerche di Mat. XL (1991), 149-168.
[CFL2] F. Chiarenza-M. Frasca-P. Longo: [ #ß: -solvability of the Dirichlet problem for non divergence
elliptic equations with VMO coefficients. Trans. of Am. Math. Soc., 336 (1993), n. 1, 841-853.
[Cha] S. Chanillo: A note on Commutators. Indiana Univ. Math. Jour., vol. 31, n.1 (1982), 7-16.
[Chr] M. Christ: Lectures on Singular Integral Operator. CBMS Regional Conference Series
inMathematics, 77, A. M. S. Providence, RI, 1990.
[CG] ´ Singular Integrals and Multipliers on Homogeneous Spaces. Revista
R. Coifman-M. de Guzman:
´ Matematica
de la Union ´ Argentina, 25 (1970), 137-143.
[CRW] R. Coifman-R. Rochberg-G. Weiss: Factorization theorems for Hardy spaces in several
variables. Annals of Math., 103 (1976) 611-635.
[CW1] R. Coifman-G. Weiss: Analyse Harmonique Non-Commutative sur Certains Espaces Homogènes.
Lecture Notes in Mathematics, n. 242. Springer-Verlag, Berlin-Heidelberg-New York (1971).
[CW2] R. Coifman-G. Weiss: Extension of Hardy Spaces and their use in Analysis. Bull. A. M. S., 83, n.
4 (1977), 567-645.
[DJ] G. David-J. L. Journe: ´
´ A boundedness criterion for generalized Calderon-Zygmund operators.
Annals of Mathematics, 120 (1984), 371-397.

21
[DJS] G. David-J. L. Journe-S ´
´ Semmes: Operateurs ´
de Calderon-Zygmund, ´
fonctions para-accretives et
interpolation. Rev. Mat. Iberoamericana, 1, n. 4 (1985), 1-56.
[FR] E. Fabes-N. Rivière: Singular integrals with mixed homogeneity. Studia Mathematica, 27 (1966),
19-38.
[GR] J. Garcia-Cuerva - J. L. Rubio de Francia: Weighted norm inequalities and related topics. North-
Holland, Mathematics Studies, 116. Amsterdam-New York-Oxford, 1985.
[JN] F. John-L. Nirenberg: On functions of bounded mean oscillation. Comm. Pure Appl. Math., 14
(1961), 175-188.
´
[KV1] A. Koranyi-S. ´
Vagi: ´
Integrales singulières sur certains espaces homogènes. C. R. Acad. Sc.
Paris, t. 268 (1969), serie A, 765-768.
[KV2] ´
A. Koranyi-S. ´ Singular Integrals on Homogeneous Spaces and some Problems of classical
Vagi:
analysis. Ann. Sc. Norm. Sup. di Pisa, serie III, vol. XXV (1971), pp. 575-648.
[Me] ´
Y. Meyer: Ondelettes et Operateurs ´
II: Operateur ´
de Calderon-Zygmund. Hermann, Paris, 1990.
[MS] R. A. Macias-C. Segovia: Lipschitz Functions on Spaces of Homogeneous Type. Adv. in Math.,
(1979), 257-270.
[R] N. M. Rivière: On singular integrals. Bull. of A. M. S., vol. 75 (1969), 843-847.
[St1] E. M. Stein: Note on Singular Integrals. Proc. of Am. Math. Soc., vol. 8, n. 2 (1957), 250-254.
[St2] E. M. Stein: Singular Integrals and Differentiability Properties of Functions. Princeton
University Press. Princeton, New Jersey, 1970.
[St3] E. M. Stein: Harmonic Analysis: Real-Variable Methods, Orthogonality and Oscillatory
Integrals. Princeton University Press, Princeton, NJ, 1993.
[ST1] C. Segovia-J. L. Torrea: Vector-Valued Commutators and Applications. Indiana Univ. Math. J.,
38, n.4 (1989), 959-971.
[ST2] ´
C. Segovia-J. L. Torrea: Higher Order Commutators for Vector Valued Calderon-Zygmund
Operators. To appear on: Trans of Amer. Mat. Soc.
[To] A. Torchinsky: Real variable methods in harmonic analysis. Academic Press, 1986.

Dipartimento di Matematica. Politecnico di Milano

Via Bonardi 9. 20133 Milano. Italy.

22