See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2607961

On the T(1)-Theorem for the Cauchy Integral

Article  in  Arkiv För Matematik · February 1970

Source: CiteSeer

CITATIONS READS
57 88

1 author:

Joan Verdera
Autonomous University of Barcelona
84 PUBLICATIONS   1,539 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Joan Verdera on 14 October 2013.

The user has requested enhancement of the downloaded file.

ESI The Erwin Schrodinger International
Institute for Mathematical Physics
Boltzmanngasse 9
A-1090 Wien, Austria

On the T(1){Theorem for

the Cauchy Integral

Joan Verdera

Vienna, Preprint ESI 817 (1999) December 27, 1999

Supported by Federal Ministry of Science and Transport, Austria

Available via anonymous ftp or gopher from FTP.ESI.AC.AT
or via WWW, URL: http://www.esi.ac.at
On the T (1)-Theorem for the Cauchy Integral
Joan Verdera

17 de juny de 1998

Abstract
The main goal of this paper is to present an alternative, real vari-
able proof of the T (1)-Theorem for the Cauchy Integral. We then
prove that the estimate from below of analytic capacity in terms of
total Menger curvature is a direct consequence of the T (1)-Theorem.
An example shows that the L1 -BMO estimate for the Cauchy Integral
does not follow from L2 boundedness when the underlying measure is
not doubling.

Introduction
In this paper we present an alternative proof of the T (1)-Theorem for the
Cauchy Integral Operator with respect to an underlying measure which is
not assumed to satisfy the standard doubling condition. This result has been
proved recently in [T1] and, independently, in [NTV1] where fairly general
Calderon-Zygmund operators are considered. The proof in [T1] exploits a
tool specic to the Cauchy kernel, called Menger curvature (see section 1 for
the denition) and is based on two main ingredients: a good  inequality
and a special argument, which is designed to make the transition from an
L2 estimate to a weak (1; 1) inequality. This argument involves analytic
capacity and consequently is of a complex analytic nature. Our approach
avoids use of complex analysis. In fact, our strategy consists in nding in
any given disc a \big piece", in the sense of Guy David [D1, D2], on which the
operator is bounded on L2. We then plug in the standard good  inequality
to control the maximal Cauchy Integral by the centered maximal operator,
as in [D1, D2]. In this second step one only needs to check that the doubling
1991 Mathematics Subject Classication: Primary 30E20, 42B20. Secondary 30C85,
42A50.
Key words and phrases: Cauchy Integral, T (1) Theorem, Analytic capacity.

1
condition is not really used in the classical arguments. Thus our proof is
actually reduced to the construction of a \big piece", which turns out to be
fairly simple because of the good positivity properties of Menger curvature.
We proceed now to state precisely the main result.
Let  be a positive Radon measure in the plane. Our goal is to estimate
the Cauchy integral operator on L2(). In view of the singularity of the
Cauchy kernel z?1  we assume that  satises the growth condition
(1) (D)  Cr(D); for each disc D;
where r(D) stands for the radius of D and C is some positive constant
independent of D. Indeed, if  has no atoms then (1) is necessary for the
L2() boundedness of the Cauchy Integral [D2, p. 56]. We say that the
Cauchy integral operator is bounded on L2() whenever for some positive
constant C one has
Z Z
(2) jC"(f)j2 d  C jf j2 d; f 2 L2(); " > 0;
where
(3) C"(f)(z) =
Z
f ( ) d( ); z 2 C :
j ?zj>"  ? z
Notice that the integral in (3) is absolutely convergent for each z, as can
readily be seen applying the Schwarz inequality and then using (1).
A necessary condition for (2) is obtained by taking as f the characteristic
function D of a disc D and restricting the domain of integration in the left
hand side of (2) to D:
Z
(4) jC"(D)j2 d  C(D); for each disc D; " > 0:
D

The T (1)-Theorem for the Cauchy integral can now be stated as follows.
Theorem. Let  be a positive Radon measure satisfying (1). Then (2) fol-
lows from (4).
We remark that if  satises the doubling condition
(5) (2D)  C(D); for each disc D;
where 2D stands for the disc concentric with D of twice the radius, then (4)
is easily seen to be equivalent to requiring that C" () belongs to BMO(),
2
uniformly in ". Hence we recover the familiar condition in the standard
formulation of the T (1)-Theorem for the operator T = C [D2, p. 30]:
C (1)  C () belongs to BMO  BMO():
In the doubling context the Theorem can readily be proved using Menger
curvature and interpolation between H 1 and BMO (see section 4).
In section 1 we gather some preliminaries including notation, terminology
and background. Section 2 contains the proof of the Theorem. In section 3
we remark that the Theorem also readily gives the estimate from below for
analytic capacity in terms of Menger curvature [Me] by purely real variable
arguments. Section 4 shows that if the doubling condition (5) fails then L2
boundedness of C does not imply the L1-BMO estimate.

1 Preliminaries
Given three distinct points z1, z2, z3 2 C one has the identity [Me]
(6)
X 1 = c(z1; z2; z3)2
 (z (2) ? z )(z
(1) (3) ? z (1) )
where the sum is taken over the six permutations of f1; 2; 3g and c(z1; z2; z3)
is the Menger curvatuve of the given triple, that is, the inverse of the radius
of the circumference passing through z1, z2 and z3. For a positive Radon
measure  the quantity
ZZZ
c( )2 = c(z1; z2; z3)2 d (z1) d (z2) d (z3)
is called the total Menger curvature of  or simply the curvature of  . Note
that we have not dened c(z1; z2; z3) for triples where at least two of the
points are the same; for such triples we may set c(z1; z2; z3) = 0.
The rst application of (6) to the L2 theory of the Cauchy Integral Op-
erator was a new proof of the L2 boundedness of the Cauchy Integral on
Lipschitz graphs, (see [V2] and [MV]). There we showed that the arc length
measure on an arc of a Lipschitz graph has nite curvature.
Later on the identity (6) was used to obtain estimates from below for
analytic capacity [Me] and to describe uniform rectiability via the mapping
properties of the Cauchy integral operator [MMV]. Impressive progress has
been made, using (6), in recent work by several authors [DM, JM, L, Ma,
T1, T2, T3], culminating in David's solution of Vitushkin's conjecture [D3].

3
 In our estimates we will use two variants of the Hardy-Littlewood maximal
operator acting on a complex Radon measure  , namely,
M (z) = sup j j(D(z; r)) ; z 2 C ;
r
r>0

and
M (z) = sup j j((DD((z;z;rr)))) ; z 2 spt ;
r>0
where D(z; r) is the open disc centered at z of radius r and spt  is the closed
support of .
It follows from the Besicovitch covering Lemma that M satises the weak
type estimate [J, p. 8]
(7) fz : M  (z) > tg  Ct?1k k;
and since
M (z)  CM (z); z 2 spt ;
because of (5), (7) also holds when M is replaced by M .
Actually the weak type (1; 1) estimate for M is a consequence of the sim-
plest standard covering lemma [S, Lemma 1, p. 12] and so there is nothing
deep in it. Although we could work only with M we prefer to keep the dis-
tinction between M and M to emphasize those steps where the Besicovitch
covering lemma necessarily comes into play. Notice that (7) coupled with the
obvious L1 estimate gives by interpolation the inequality
Z Z
M (f d  C jf jp d; 1 < p < 1:
)p
The letter C will denote either the Cauchy Integral Operator or a constant
which may be dierent at each occurrence and that is independent of the
relevant variables under consideration. The precise meaning of C will always
be clear from the context.

2 The proof
Let  be a complex Radon measure. Set
(8) C"  (z) =
Z
d ( ) ; z 2 C :
j ?zj>"  ? z

4
The integral in (8) is absolutely convergent for all z provided  is a nite
measure or, more generally, provided
(9)
Z
dj j( ) < 1:
1 + j j
Lemma 1. Let j , j = 1; 2; 3 be three complex Radon measures satisfying
(9) with  replaced by j , j = 1; 2; 3. Then
Z
X
C" ((1))C" ((2)) d(3)
 ZZZ

= c2(z1; z2; z3) d1(z1) d2(z2) d3 (z3) + R;

S"

where the sum is taken over the permutations of f1; 2; 3g,

S" = f(z1; z2; z3) 2 C 3 : jz1 ? z2j > "; jz1 ? z3j > " and jz2 ? z3j > "g
and X
Z
jRj  C M(2)(z(1))M(3)(z(1)) d(1)(z(1));

C being an absolute constant.
Proof. Set
T" = f(z1; z2; z3) 2 C 3 : jz1 ? z3j > " and jz2 ? z3j > "g;
U" = f(z1; z2; z3) 2 C 3 : jz1 ? z2j  "; jz1 ? z3j > 2" and jz2 ? z3j > "g
and
V" = f(z1; z2; z3) 2 C 3 : jz1 ? z2j  "; jz1 ? z3j  2" and jz2 ? z3j > "g:
Then
Z
C"(1)C"(2) d3 =
ZZZ
d1(z1) d2(z2) d3 (3)
T" (z1 ? z3)(z2 ? z3)
ZZZ ZZZ ZZZ ZZZ

=


+


+


=


+ I" + II"

S" U" V" S"

5
where the last identity is a denition of I" and II". To estimate I" and II" we
assume, without loss of generality, that the j are positive measures. Then
jI"j  C
ZZZ
d1 (z1) d2(z2) d3(z3)
U" jz1 ? z3j2
ZZ
C "?1M3(z1) d1 (z1) d2 (z2)
jz1 ?z2 j"
Z
 C M2 (z1)M3(z1) d1 (z1):
For II" we write
ZZZ
jII" j  "?2 d1 d2 d3
V"
ZZ
 C"?1 M3(z1) d1 (z1) d2(z2)
jz1 ?z2 j"
Z

 C M2(z1)M3(z1) d1 (z1):

Operating in a similar way for any  and then summing over  we get the
conclusion of the Lemma.
We apply Lemma 1 to 1 = 2 = f with f a (real function) in L2()
and 3 = D with D a xed disc. We then have
Z Z
(10) 2 jC"(f)j2 d + 2 Re C"(f)C" (D )f d
D
ZZZ Z 

= c2(z; w;  )f (z)f (w)D ( ) d(z) d(w) d( ) + O f 2 d :

S"

In particular taking f = D one gets

Z ZZZ
6 jC"(D )j2 d = c 2 (z; w;  ) d(z ) d(w) d( ) + O((D));
D S" \D 3

and thus
ZZZ
(11) c2(z; w;  ) d(z) d(w) d( )  C(D);
D3
provided (4) holds.
6
 It is worth pointing out that (11) was inexactly attributed in [NTV1,
p. 705]. Indeed, a rst version of (11) appears in [V2] and [MV] and later on
in [MMV] in the form at hand.
We come now to the core of the argument that produces a \big piece"
inside a given disc D.
Set ZZ
2
cD (z) = c 2 (z; w;  ) d(w) d( ); z 2 C :
2 D
By Chebischev
fz 2 D : cD (z) > t or jC"(D)(z)j > tg
Z Z 

 t?2 c2 (z) d(z) +

D jC"(D )j2 d  Ct?2(D):
D D

Hence, given 0 <  < 1 ( will be chosen later), there exists a compact
E  D such that
p p
cD (z)  C= and jC"(D )(z))  C=; z 2 E;
and
(DnE )  (D):
Set, as in [T1], Z

k(z; w) = c2(z; w;  ) d( );

D
so that Z
k(z; w) d(w)  c2D (z)  C=; z 2 E:
E
Since k(z; w) = k(w; z), Schur's Lemma now gives that if f 2 L2(E )
(= L2(E; d)) then
Z Z Z 

c2(z; w;  )f (z)f (w)D ( ) d(z) d(w) d( )

 

 
S"
Z Z Z
 jf (z)j jf (w)jk(z; w) d(w) d(z)  C f 2 d;
where C = C () does not depend on ".
Therefore from (10)
Z Z  1=2 Z  1=2 Z

jC" (f)j2 d  C jC"(f)j2 f 2 d +C f 2 d;

D D

7
and consequently
Z Z
jC"(f)j2 d  C f 2 d; f 2 L2(E ):
D
By duality this implies
Z Z
(12) jC"(f)j2 d  C f 2 d; f 2 L2(D):
E
We now need an appropriate Cotlar type inequality. For a complex Radon
measure  satisfying (9) set, for z 2 C ,
C"(z) = sup jC  (z)j
"

and
C  (z) = sup C" (z):
">0
Lemma 2. Let  and  be positive Radon measures satisfying the growth
condition
(D) +  (D)  Cr(D); for each disc D;
and such that for some " > 0
Z Z
jC"(f d)j2 d  C jf j2 d; f 2 L2():
Then
C"(f)(z)  C fM (jC"(f)j2 d )1=2 + M (jf j2 d)1=2g:
For a proof for the case  =  , which can be seen to work under our
hypothesis, we refer the reader to [T2, Lemma 3 and Theorem 4].
Combining (12) with Lemma 2 applied to  = E  and  = D  we get,
for each f 2 L2(D),
Z Z 

(13) fz 2 E : C (f)(z) > tg  Ct?2

" jC"(f)j2 d + f 2 d
E
Z
 Ct?2 f 2 d:
We now want to have the above inequality at our disposal for a general
f 2 L2(). This essentially means that, for each open disc D, C" maps

8
boundedly L2(Dc ) into L2(D) with constant independent of ". This is clear
if C" is replaced by
C (f)(z) = "lim C (f)(z); z 2 D; f 2 L2(Dc ):
!0 "
The reason is that, C (f) being holomorphic on D, we only need to apply
Carleson's Theorem twice:
Z Z Z
jC (f)j2 d  C jC (f)j2jdzj  C jf j2 d:
D @D Dc
However we wish to have a real variable proof, which could be extended to
Rn and n ? 1 dimensional kernels. This can be done painlessly and in fact is
implicit in David's paper [D1].
Lemma 3. Let
be an open disc and let  and  be positive Radon measures
satisfying
(D) +  (D)  Cr(D); for each disc D;
and (
C ) =  (
) = 0.
Then Z Z
C (f ) d  C jf j2 d; f 2 L2( ):
 2

Proof. Assume, without loss of generality, that
is centered at the origin

and let r be its radius. Given z 2
let d be the distance from z to @
. We
claim that
(14) C (f )(z)  C (f )(w) + CM (f )(w); jw ? zj  2d:
Fix " > 0. Assume rst that " < d. Then for jw ? zj  2d,
jC"(f )(z)j = jCd(f )(z)j  jC4d(f )(z)j + M (f )(w):
The same inequality holds for d  " < 4d, so that we are left with the case
4d  ". Set f1 = D(z;")f , f2 = f ? f1. Thus, for jw ? zj  2d,
jC"(f )(z) ? C" (f2 )(w)j  CM (f )(w)
and
jC"(f2 )(w) ? C" (f )(w)j  CM (f )(w)
because of standard simple estimates. Therefore the claim follows.
Set
F (w) = C (f )(w) + CM (f )(w):
9
Using (14), the simplest covering lemma [S, Lemma 1, p. 12] and the growth
condition on , one proves that (see for example [S, p. 59{60])
fz 2
: C (f )(z) > tg  C jfw 2 @
: F (w) > tgj ;
where j
j denotes one dimensional Lebesgue measure. Then
Z Z Z 

C (f )2(z) d(z)  C C (f )2(w)jdwj + M (f )2 (w)jdwj

@
@

Z
 C jf j2 d
by [D1, Proposition 5, p. 164] and [D1, Proposition 3, p. 161].
Lemma 3 and (13) now give
Z
(15) fz 2 E : C (f)(z) > tg  Ct?2
" f 2 d; f 2 L2();
which shows that E is indeed a \big piece".
The proof of the Theorem is practically complete. One last step is left: we
have to check that (15) allows us to prove an appropriate good  inequality
without resorting to a doubling condition on . For the reader's convenience
we present the well known argument, which can be found in [D2, p. 61{62].
The good  inequality we need is the following.
For each  > 0 there exists = () > 0 small enough so that
(16) fz : C"(f)(z) > (1 + )t and M (f 2)1=2(z)  tg
 21 fz : C"(f)(z) > tg:
Once (16) is established we deduce that C" satises the same Lp inequalities
as M (f 2)1=2 [D2, p. 60]. Then
Z Z
C (f)p d  C
" p jf jp d; 2 < p < 1:
In particular
Z Z
jC"(f)j d  Cp jf jp d; 2 < p < 1;
p

and by duality we get the same estimate for 1 < p < 2 and so for p = 2 by
interpolation.
10
 Let's prove (16). The set
= fz : C"(f)(t) > tg is open. Given
a 2 S \
let D(a) be the disc with center a and radius 5?1 dist(a;
c). By
the Besicovitch covering lemma
\ S can be covered by a family of discs
Dj = D(aj ) which is almost disjoint, that is, such that each point in the
plane belongs to at most N discs Dj , N being an absolute constant. Notice
that then the family f4Dj g is almost disjoint too. This is one of the key facts
in order to allow us to dispense with the doubling condition.
We are going to show that, given  > 0 and 0 <  < 1, there exists
= (; ) > 0 such that, for all j ,
(17)
fz 2 S \ Dj : C"(f)(z) > (1 + )t and M (f 2)1=2(z)  tg  (4Dj ):
Then summing over j ,
fz 2 S : C"(f)(z) > (1 + )t and M (f 2)1=2(z)  tg  N(
);
where N stands now for the constant of almost disjointness of f4Dj g. Choos-
ing  so that N = 21 we get (16).
Let's turn our attention to (17). Fix j and set D = Dj , a = aj . As-
sume, without loss of generality, that there exists b 2 S \ D such that
M (f 2)1=2(b)  t. Let w be a point in
c such that jw ? aj = dist(a;
c)
and set B = D(w; 9r) where r = jw ? aj=5 is the radius of D. Hence
D 
 D(b; 3r)  4D  B . Set f1 = fB and f2 = f ? f1. Then, for
z 2 D and   ",
1 Z
jC (f1)(z)j = jC (
f )(z)j + r jf ( )j d( )
B
 C"(
f )(z) + CM (f)(b)
 C"(
f )(z) + C t;
and so
jC(f)(z)j  jC (f2)(z)j + C"(
f)(z) + C t:
To compare C (f2)(z) with C(f)(w) we use the standard arguments (see
[D1] or [D2]). We obtain
jC(f2)(z) ? C (f2)(w)j  CM (f)(b)
and
jC (f2)(w)j  C"(f)(w)  t:

11
Therefore
C"(f)(z)  C"(
f)(z) + (1 + C )t; z 2 D:
Now choose so that 2C   and let E be a \big piece" associated to the
disc D and the number . Then
fz 2 D : C"(f)(z) > (1 + )tg  (DnE )
+  z 2 E : C ( f)(z) >  t
n o
"
2
Z
 (D) + C (t)?2 f 2 d


 (D) + C (t)?2(
)M (f 2)(b)
 ( + C ( =)2)(4D)  (4D)
provided  and are chosen small snough so that  + C ( =)2  .

3 Estimating analytic capacity from below

Let K be a compact subset of C , (K ) its analytic capacity, and let  be
a positive measure supported in K satisfying (D)  r(D) for each disc D
and c() < 1. Then [Me]
(K )  c (kk k+kc2())1=2 :
3=2
(18)
The original proof of (18) is rather simple but relies on the Garabedian's L2
description of analytic capacity [G] and thus depends on complex analysis
techniques. We give here a quick derivation of an inequality slightly bet-
ter than (18) from the T (1)-Theorem described in the preceding sections.
The reader will notice that our reasoning uses purely real variable methods.
Similar arguments have been used independently by Tolsa in [T3] for other
purposes.
Given a compactly supported positive measure , set
Z Z
E () = M(z) d(z) + c(z) d(z);
where ZZ

c2 (z) =
 c2(z; w;  ) d(w) d( ); z 2 C :
The quantity E () seems to be an appropriate candidate to play the role of
\energy" associated to the kernel 1=z.
12
Theorem. For each compact subset K of the plane,
(19) (K )  C supfE ()?1 : spt   K and kk = 1g:
If  is a positive measure supported on K such (D)  r(D) for all discs
D, then

 

E kk = kk?2E ()  kk?2(kk + c()kk1=2) = kk?3=2(kk1=2 + c())

and so (18) follows from (19).
Proof of the Theorem. Take a probability measure  supported on K with
E () < 1. By Chebischev there exists a compact subset J of K such that
(J )  2?1, and M(z)  A and c (z)  A, for all z 2 J , where A = 2E ().
Set  = jJ . Then k k  2?1 ,  (D)  Ar(D) for each disc D and
(20) c (z)  A; z 2 spt :
Clearly (20) gives (11) with  replaced by  and therefore the Cauchy Integral
is bounded on L2( ) by the T (1)-Theorem discussed in the previous sections.
We wish now to have the weak L1 inequality
(21)  fz : jC"()j > tg  CA
t kk;
where  is any nite measure in the plane and C some absolute constant.
This follows by standard Calderon-Zygmund theory if  is \doubling"
and by a simple, nice argument found recently in [NTV2] in the general case.
Dualizing the weak type inequality (21), as in [T1] or [V1], we obtainR that
there exists a  -measurable function h, 0  h < 1, with  (J )  2 h d
and jC (h d )(z)j  CA, for each z 2 C nJ . Here C (h d ) is just the locally
integrable function z1  h d . Therefore, for some absolute constant C
(K )  CA?1 = CE ()?1;
as desired.

4 Failure of the L1-BMO estimate

When  is a doubling measure the proof of the T (1) Theorem for the Cauchy
Integral is very simple, as showed in [V2] and [MV]. The reasoning goes as

13
follows. Fix a disc D and take a bounded, -measurable function f supported
on D. Because of (10) and (11) we get
Z Z  1=2 Z  1=2
jC"(f)j2 d  2 jC"(f)j2 d jC"(D )j2 d
D D
ZZZ
+ kf k2 1 c2(z; w;  ) d(z) d(w) d( )+ C kf k21 (D)
D3
Z  1=2
 C(D)1=2 jC"(f)j2 d + C(D)kf k21
D
and so
Z
(22) jC"(f)j2 d  C kf k21(D);
D
with C independent of ".
The above inequality and standard arguments show that C" maps L1 ()
boundedly into BMO() and maps the atomic version of H 1() boundedly
into L1() (see [J, p. 49]). Interpolation between BMO and L1 now gives
that C" maps L2 into L2.
By BMO() we understand the space of locally integrable functions with
respect to , such that for each disc D centered at a point in spt  one has
Z
jf (z) ? fD j d(z)  C(D);
D
C being a positive constant independent of D and
1 Z
fD = (D) f d:
D
An atom is a -measurable function a, supportedR on some disc D centered
at a point in spt , such that jaj  (D)?1 and a d = 0. The atomic
version Hat1 () of H 1 is then the set of functions of the form
1
X
j aj ;
j =1
1
X
where aj is an atom for all j and jj j < 1.
j =1
When the measure  is non-doubling one can still obtain (22) from the
hypothesis of the T (1)-Theorem, but we shall see that (22) implies neither the
14
L1 -BMO nor the Hat1 -L1 estimate. The example we shall describe is rather
simple. In fact, the measure  will be the one-dimensional Lebesgue measure
restricted to a certain subset of the real line.
Set n = 4?2n , n = 0; 1; 2; : : : , and
In = [2n?1 ; 22n?1 ];
J = n?1 ; n?1 + 2
 

n
2 2 n?1 ; n = 1; 2; ; 3 : : : :

Dene  as the one-dimensional Lebesgue measure restricted to

!
1
[
(?1; 0) [ (In [ Jn ) :
n=1

Let Dn denote the disc of radius n2?1 centered at the point n2?1 . Then the
function 1
X
h = (In ? Jn )(Dn )?12?n
n=1
lies in Hat1 (). We claim that
Z 0
jC (h)(x)j dx = +1;
?1
where for f 2 L1(R) we write
f (t) dt: Z 1
C (f )(x) = P.V.
?1 t ? x
Fix a positive integer n, and abbreviate In as both I and (a; b) and Jn as J
and (;  ). Then
C (I )(x) ? C (J )(x) = log ab ?
?x
x ? log  ? x  0; for x  0:
?x
A simple computation gives
0    

log 1 + 1t dt  l log 1 + 21l

Z Z b
C (I )(x) dx =
?1 a
and 0    

log 1 + 1t dt  l log 1 + p1 ;
Z Z 
C (J )(x) dx =
?1  l
15
where l = b ? a.
Since (Dn )  2n?1 , we conclude that
Z 0 1
X
jC (h)(x)j dx  C 2?n log ?n?1 1 = +1;
?1 n=1
as claimed.
Thus C does not map Hat1 () into L1(), although it maps L2() into
L2(). To show that L1 () is not mapped boundedly into BMO() we
resort to the most elementary fact concerning the duality between Hat1 ()
and BMO(). Namely, given an atom a and a disc D as in the denition of
atom, there exists a function b in L1 (), kbk1 = 1, for which one has
Z Z
jC (a)j d = C (a)b d
Z
= ? aC (b) d
Z
= ? a(C (b) ? C (b)D) d
1 Z
 (D) jC (b) ? C (b)Dj d:
D
Then the L1-BMO estimate would imply the Hat1 -L1 estimate, which
fails.
Acknowledgement. The author was partially supported by the grants
DGES PB96-1183 and CIRIT 1996SGR00026.

References
[D1] G. David, Operateurs integraux singuliers sur certaines courbes du
 Norm. Sup. (4) 17 (1984), 157{189.
plan complexe, Ann. Sci. Ecole
[D2] G. David, Wavelets and singular integrals on curves and surfaces,
Lecture Notes in Math. 1465, Springer-Verlag, Berlin, 1991.
[D3] G. David, Unrectiable 1-sets in the plane have vanishing analytic
capacity, Prepublication, Universite Paris-Sud (1997).
[DM] G. David, P. Mattila, Removable sets for Lipschitz harmonic func-
tions in the plane, Preprint (1997).
16
[G] J. Garnett, Analytic capacity and measure, Lecture Notes in Math.
297, Springer-Verlag, 1972.
[J] J.-L. Journe, Calderon-Zygmund operators, pseudo-dierential oper-
ators and the Cauchy integral of Calderon, Lecture Notes in Math.
994, Springer-Verlag, 1983.
[JM] H. Joyce, P. Morters, A set with nite curvature and projections of
zero length, Preprint (1997).
[L] J.-C. Leger, Courbure de Menger et rectiabilite, PhD Thesis, Uni-
versite de Paris-Sud (1997).
[Ma] P. Mattila, On the analytic capacity and curvature of some Cantor
sets with non--nite length, Publ. Mat. 40 (1996), 127{136.
[Me] M. S. Melnikov, Analytic capacity: discrete approach and curvature
of a measure, Sbornik: Mathematics 186:6 (1995), 827{846.
[MMV] P. Mattila, M. S. Melnikov, J. Verdera, The Cauchy Integral, an-
alytic capacity and uniform rectiability, Ann. of Math. (2) 144
(1996), 127{136.
[MV] M. S. Melnikov, J. Verdera, A geometric proof of the L2 boundedness
of the Cauchy integral on Lipschitz graphs, Int. Math. Res. Not. 7
(1995), 325{331.
[NTV1] F. Nazarov, S. Treil, A. Volberg, Cauchy integral and Calderon-
Zygmund operators on nonhomogeneous spaces, Int. Math. Res.
Not. 15 (1997), 703{726.
[NTV2] F. Nazarov, S. Treil, A. Volberg, Weak type estimates and Cot-
lar inequalities for Calderon-Zygmund operators in nonhomogeneous
spaces, Preprint (1997)
[S] E. M. Stein, Harmonic analysis. Real-Variable methods, orthogo-
nality, and oscillatory integrals, Princeton Univ. Press., Princeton,
N.J., 1993.
[T1] X. Tolsa, L2-boundedness of the Cauchy integral operator for con-
tinuous measures, Duke Math. J., to appear.
[T2] X. Tolsa, Cotlar's inequality and existence of principal values for the
Cauchy integral without the doubling condition, J. Reine Angew.
Math., to appear.
17
 [T3] X. Tolsa, Curvature of measures, Cauchy singular integral and ana-
lytic capacity, Thesis, Universitat Autonoma de Barcelona (1998).
[V1] J. Verdera, A weak type inequality for Cauchy transforms of nite
measures, Publ. Mat. 36 (1992), 1029{1034.
[V2] J. Verdera, A new elementary proof of L2 estimates for the Cauchy
Integral on Lipschitz graphs, Proceedings of the Conference \Geo-
metrical and algebraic aspects in several complex variables", Cetraro
(Italy) 1994, to appear.

18

View publication stats