1 December 2004

2000 Mathematics Subject Classification. Primary 32D15, 32D10.
<ph f="cmr"> </ph><ph f="cmbx">Generalization of Drużkowski's and Gonchar's ”Edge-of-the-Wedge” Theorems </ph>

Peter Pflug

Viê ̣ t-Anh Nguyên

Peter Pflug, Carl von Ossietzky Universitat Oldenburg, Fachbereich Mathematik, Postfach 2503, D–26111, Oldenburg, Germany E-mail address : pflug@mathematik.uni-oldenburg.de Vie t-Anh Nguyen, Carl von Ossietzky Universitat Oldenburg, Fachbereich Mathematik, Postfach 2503, D–26111, Oldenburg, Germany E-mail address : nguyen@mathematik.uni-oldenburg.de

1 Introduction

The ”Edge-of-the-Wedge” type theorems deal with the continuation of holomorphic functions of several complex variables. The first theorem was discovered and proved by N. N. Bogolyubov in 1956 in connection with quantum field theory and dispersion relations. Since then, there is a long list of papers dealing with this theorem and its generalizations under various assumptions (see [16, [20and the references therein).
Here we consider a one-sided version of the ”Edge-of-the-Wedge” type theorem in the spirit of the pioneer work of Malgrange–Zerner [21. Epstein's survey article [3gives a historical discussion and motivation for this version of an ”Edge-of-the-Wedge” theorem, as well as its natural connections with theorems on separate analyticity.
The first results in this direction are obtained by Komatsu [10and Drużkowski [2, but only for some special cases. Recently, Gonchar [5, 6has proved a remarkable more general result for the one-dimensional case. In a recent work [14, the authors are able to generalize Gonchar's result to the higher dimensional case.
However, in all these cases the hypotheses on the function to extend and its domain of definition are, in some sense, rather restrictive and strong. Therefore, the main goal of our work is to establish one-sided ”Edge-of-the-Wedge” type theorems in some more general one-dimensional cases with more optimal hypotheses. Perhaps, this will be the first step towards understanding the higher dimensional case and the manifold settings in its full generality.
This paper is organized as follows.
In Section 2 we gather some necessary notions and auxiliary results. This preparation will enable us, at the end of this section, to formulate the above mentioned results of Drużkowski and Gonchar and to discuss in more details the motivation for our work.
The statements of our main results as well as an outline of their proofs are given in Section 3.
The tools which are needed for the proof of the main results are developed in Sections 4, 5, and 8.
The proofs of the main results are given in Sections 6, 7, 9, and 10.
Section 11 gives examples showing the optimality of our results. Finally, we conclude the article with some remarks and open questions.
Our approach is based on our previous work [14, the Gonchar–Carleman operator developed in [5, 6, conformal mapping theory, and a thorough geometric study of harmonic measures.
Acknowledgment. The paper was written while the second author was visiting the Carl von Ossietzky Universität Oldenburg being supported by The Alexander von Humboldt Foundation. He wishes to express his gratitude to these organizations.

2 Preliminaries

In order to recall the classical one-sided versions of the ”Edge-of-the-Wedge” theorem and to discuss in more detail our motivation, we need to introduce some notation and terminology. In fact, we keep the main notation from the previous work [14.

2.1 Planar domains with partly rectifiable boundary

We collect here some classical facts from the books by Goluzin, Koosis and Pommerenke ([4, [11, [13).
A Jordan curve is the image C : = { γ ( t ) , t [ a , b ] }   of a continuous one-to-one map γ : [ a , b ] C ,   where a , b R , a < b .   The set { γ ( t ) , t ( a , b ) }   is said to be the interior of the Jordan curve. A closed Jordan curve is the image C : = { γ ( t ) , t [ a , b ] }   of a continuous map γ : [ a , b ] C ,   which is one-to-one in [ a , b )   and which satisfies γ ( a ) = γ ( b ) .   The map γ   is called a parametrization of C .   Moreover, C   is said to be rectifiable if
sup { k = 0 n 1 | γ ( t k + 1 ) γ ( t k ) | } < ,  
where the supremum is taken over all possible positive integers n   and sequences of values t 0 , , t n [ a , b ]   such that t 0 < t 1 < < t n .   It is easy to see that this supremum is independent of the choice of a parametrization. It is called the length of C .   A (rectifiable) Jordan domain is a bounded domain in C   whose boundary is a (rectifiable) closed Jordan curve.
Consider an open set D C .   Then D   is said to be locally rectifiable at a point ζ D   if there is a neighborhood U   of ζ   in C   such that U D   is the interior of a rectifiable Jordan curve. Moreover, D   is said to be locally rectifiable on a subset A   of D   if D   is locally rectifiable at all points of A .   Now let D C   be an open set which is locally rectifiable on a set A D .   We consider the induced topology on D   with respect to the Euclidean topology of C .   Therefore, there is an open set V   in D   such that A V   and D   is locally rectifiable on V .   Denote by mes   the linear measure (i.e. the one-dimensional Hausdorff measure) in C .   Notice that when A   is a Jordan curve, then the length of A   coincides with mes ( A ) .   We fix the following terminology: A set A C   is said to be linearly measurable if it is measurable with respect to the linear measure.
Let ζ D   be a point with the following property: There is a closed neighborhood V   of ζ   such that V D   is a Jordan curve whose interior contains ζ .   Then ζ   is said to be of type I if there is a neighborhood U   of ζ   such that U D   is a Jordan domain.
Otherwise, ζ   is said to be of type II. We see easily that if ζ   is of type II, then there are an open neighborhood U   of ζ   and two Jordan domains U 1 ,   U 2   such that U D = U 1 U 2 .   Let ζ D   be a point of type I or II. ζ   is said to be a point where D   admits a tangent if there are a continuous one-to-one map γ : [ a , b ] D   and a point t 0 ( a , b )   such that γ ( t 0 ) = ζ   and the following limit exists lim t t 0 γ ( t ) γ ( t 0 ) t t 0 | t t 0 | | γ ( t ) γ ( t 0 ) | = λ .   The tangent line of D   at ζ   is, by definition, the real line L : = { t λ , t R } .   In Proposition  5.6 below we shall see that all the above notions are independent of the choice of a parametrization γ .   Let * D   denote the set of all points ζ D   where D   admits a tangent at ζ .   Given a linearly measurable set A D   such that D   is locally rectifiable on A ,   then it is well-known (see [11,p.68–69) that mes ( A * D ) = mes ( A ) .   We define the concept of angular approach regions at every point of ζ * D   as follows. Let L ζ   be the tangent line of D   at ζ   and let n ζ   be any element of L ζ \ { 0 } .   Then, for any 0 < α < π 2 ,   the Stolz region or angular approach region A α ( ζ )   is given by A α ( ζ ) : = { z D : π 2 α < | arg ( z ζ n ζ ) | < π 2 + α } ,   where arg : C ( π , π ]   is as usual the argument function. Geometrically, A α ( ζ )   is the intersection of D   with two cones of aperture 2 α   and vertex ζ .   Let ζ * D   and let U   be an open neighborhood of ζ .   We say that a function f   defined on U D   admits the angular limit λ   at ζ   if lim z A α ( ζ ) , z ζ f ( z ) = λ ,   for all 0 < α < π 2 .   We conclude this subsection with a simple example which may clarify the above definitions. Let G   be the open square whose four vertex are 1 + i ,   1 + i ,   1 i ,   and 1 i .   Define the domain D : = G \ [ 1 2 , 1 2 ] .   Then D   is locally rectifiable on G ( 1 2 , 1 2 ) .   Every point of G   is of type I and every point of ( 1 2 , 1 2 )   is of type II. Moreover, * D = ( G \ { 1 + i , 1 + i , 1 i , 1 i } ) ( 1 2 , 1 2 ) .  

2.2 Harmonic measure for an open set of C  

Let D   be a proper open subset of C { }   such that the boundary D   (with respect to C { }   ) is non-polar (and A   a subset of D .   ) Let A   be a subset of D .   Consider the characteristic function 1 D \ A ( ζ ) : = { 1 , ζ D \ A , 0 , ζ A .   Then the harmonic measure of the set D \ A   (denoted by ω ( , A , D )   ) is the Perron solution of the generalized Dirichlet problem with boundary data 1 D \ A .   In other words, one has ω ( , A , D ) : = sup u U ^ u ,   where U = U ( A , D )   denotes the family of all subharmonic functions u   on D   such that limsup D z ζ u ( z ) 1 D \ A ( ζ )   for each ζ D .   It is well-known (see, for example, the book of Ransford [15) that ω ( , A , D )   is harmonic on D .   Let D C   be an open set which is locally rectifiable on a subset A   of D .   We say that a point ζ * D   is a locally regular point relative to A   if lim z ζ , z A α ( ζ ) ω ( z , A U , D U ) = 0   for any 0 < α < π 2   and any open neighborhood U   of ζ .   Obviously, ζ A ¯ .   If, moreover, ζ A ,   then ζ   is said to be a locally regular point of A .   The set of all locally regular points relative to A   is denoted by A * .   Observe that in general A * A ,   A A * .   However, if A   is open, then A * D A * .   As an immediate consequence of the Subordination Principle for the harmonic measure (see Corollary 4.3.9 in [15), one gets
lim z ζ , z A α ( ζ ) ω ( z , A , D ) = 0 , ζ A * , 0 < α < π 2 . (2.1)
We extend the function ω ( , A , D )   to D A *   by simply setting ω ( z , A , D ) : = 0 , z A * .   Geometric properties of the harmonic measure will be discussed in Section 5 below.

2.3 Cross and separate holomorphicity

Let N N , N 2 ,   and let D j   be a planar domain which is locally rectifiable on a linearly measurable subset A j   of D j ,   j = 1 , , N .   We define an N   -fold cross X ,   its regular part X * ,   its interior X o ,   its edge A   and its regular edge A *   as
X : = X ( A 1 , , A N ; D 1 , , D N )
: = N j = 1 A 1 × × A j 1 × ( D j A j ) × A j + 1 × × A N C N ,
X * : = X ( A 1 * , , A N * ; D 1 , , D N ) ,
X o = X o ( A 1 , , A N ; D 1 , , D N )
: = N j = 1 A 1 × × A j 1 × D j × A j + 1 × × A N ,
A : = A 1 × × A N , A * : = A 1 * × × A N * .
Moreover, put ω ( z ) : = j = 1 N ω ( z j , A j , D j ) , z = ( z 1 , , z N ) ( D 1 A 1 * ) × × ( D N A N * ) .   It is clear that ω | D 1 × × D N   is harmonic.
For an N   -fold cross X : = X ( A 1 , , A N ; D 1 , , D N )   define its wedge
X ^ : = X ^ ( A 1 , , A N ; D 1 , , D N ) : = { z = ( z 1 , , z N ) ( D 1 A 1 * ) × × ( D N A N * ) : ω ( z ) < 1 } .  
Then the set of all interior points of the wedge X ^   is given by
X ^ o : = X ^ o ( A 1 , , A N ; D 1 , , D N ) : = { z = ( z 1 , , z N ) D 1 × × D N : ω ( z ) < 1 } .  
In particular, if A j   is an open set of D j   and A j * D j ,   j = 1 , , N ,   one has A A *   and X X * X ^ .   We say that a function f : X C   is separately holomorphic on X o   and write f O s ( X o ) ,   if for any j { 1 , , N }   and ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N )   the function f ( a , , a ) | D j   is holomorphic on D j .   We say that a function f : X C   (resp. f : A C   ) is separately continuous on X   (resp. on A   ) and write f C s ( X )   (resp. f C s ( A )   ), if for any j { 1 , , N }   and ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N )   the function f ( a , , a ) | ( D j A j )   (resp. f ( a , , a ) | A j   ) is continuous. In the sequel, for a subset J   of { 1 , , N } ,   we write J : = { 1 , , N } \ J .   Moreover, one often identifies z C n   with ( z , z ) ,   where z : = ( z j ) j J   and z : = ( z j ) j J .   In this paragraph, suppose that A j A j * ,   j = 1 , , N .   We say that a function f : X ^ \ ( A 1 × × A N ) C   is superholomorphic on X ^   and write f O ( X ^ ) ,   if for every proper subset J   of { 1 , , N }   (including   ) and any a : = ( a j ) j J j J A j ,   the restricted function f a : X ^ o ( A j , j J ; D j , j J ) C ,   given by f a ( z ) : = f ( a , z ) ,   where J : = { 1 , , N } \ J ,   is holomorphic.
For any λ C   and r > 0 ,   let B ( λ , r )   denote the ball with center λ   with radius r .   Using identity ( 2.1 ) we make the following observation: For any a = ( a , a ) X ^   with a = ( a j ) j J j J A j *   and a = ( a j ) j J j J D j   and any 0 < α < π 2 ,   there is a neighborhood U   of a   and an ε > 0   such that U ( j J A α ( a j ) × j J B ( a j , ε ) ) X ^ o .   We say that a function f : X ^ o C   admits an angular limit λ   at a X ^   if lim z a , z j J A α ( a j ) × j J B ( a j , ε ) f ( z ) = λ , 0 < α < π 2 .   Throughout the paper, for a subset M   of an Euclidean space, C ( M )   denotes the space of all continuous functions f : M C   equipped with the sup-norm | f | M : = sup M | f | .   Moreover, a function f : M C   is said to be locally bounded on M   if, for any point z M ,   there are an open neighborhood U   of z   (with respect to the induced topology on M   ) and a positive number K = K z   such that | f | U < K .   Finally, for an open set Ω C n ,   S ( Ω )   (resp. P S ( Ω ) ,   O ( Ω )   ) denotes the set of all subharmonic (resp. plurisubharmonic, holomorphic) functions on Ω .  

2.4 Motivations for our work

We are now able to formulate what, in the sequel, we quote as the classical one-sided version of the ”Edge-of-the-Wedge” theorem.
Theorem 1. (Gonchar [5, 6) Let D j C   be a rectifiable Jordan domain and A j   an open set of the boundary D j ,   j = 1 , , N .   Then, for any function f C ( X ) O s ( X o ) ,   there is a unique function f ^ C ( X ^ ) O ( X ^ )   such that f ^ = f   on X .   Moreover, if | f | X <   then | f ^ ( z ) | | f | A 1 ω ( z ) | f | X ω ( z ) , z X ^ ,   where X   , X o   , and X ^   denote the N   -fold cross, its interior and its wedge, respectively, associated to the A j   , D j   .
Theorem 1 admits various generalizations. The following theorem is announced by Gonchar in [5.
Theorem 2. Let D j C   be a Jordan domain and let A j   be an open set of the boundary D j   such that D j   is locally rectifiable on A j ,   j = 1 , , N .   Let f   be a function defined on the N   -fold cross X   with the following properties:
  • (i) f | X o C ( X o ) O s ( X o ) ;  
  • (ii) f   is locally bounded on X ;  
  • (iii) for any j { 1 , , N }   there is a function f j   defined on A   such that for any ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N ) ,   the holomorphic function f ( a , , a ) | D j   has the angular limit f j ( a , a j , a )   at a j   for a.e. a j A j   and f 1 = = f N = f   a.e. on A .  
1) Then there is a unique function f ^ O ( X ^ o )   such that lim z X ^ o , z ζ f ^ ( z ) = f ( ζ ) , ζ X o .   2) If, moreover, | f | X < ,   then | f ^ ( z ) | | f | A 1 ω ( z ) | f | X ω ( z ) , z X ^ o .   3) If, moreover, f   is continuous at a point a A ,   then lim z X ^ o , z a f ^ ( z ) = f ( a ) .  
On the other hand, the following result due to Drużkowski [2gives a different flavor.
Theorem 3. Let D j C   be a rectifiable Jordan domain and let A j   be an open connected set of the boundary D j ,   j = 1 , , N .   Let f   be a function defined on X   with the following properties:
  • (i) f C s ( X ) O s ( X o ) ;  
  • (ii) f   is locally bounded on X ;  
  • (iii) f | A   is continuous on A .  
Then all conclusions of Theorem 1 still hold.
Observe that all these theorems require the following very strong hypothesis:
D 1 , , D N   are rectifiable Jordan domains and the edge A   is an open set of D 1 × × D N .   Moreover, the assumptions on the boundedness and continuity of f   are rather restrictive.
The question naturally arises whether Theorems 1–3 are still true if D 1 , , D N   are open sets in C   and the edge A   is not necessarily an open set of D 1 × × D N .   In addition, if one drops the hypothesis on the local boundedness and the continuity of f ,   can one obtain a holomorphic extension of f   and what are its properties? These matters seem to be of interest especially when one seeks to generalize Theorems 1–3 to higher dimensions.
The present paper is motivated by these questions. Our first purpose is to generalize Gonchar's theorems to a very general situation, where D 1 , , D N   are, in some sense, almost general open subsets of C   and where the boundary sets A 1 , , A N   are almost general subsets of D 1 , , D N .   Our second goal is to establish, in this general context, an extension theorem analogous to Drużkowski's theorem with a minimum of hypothesis on f .  

3 Statement of the main results and outline of the proofs

We are now ready to state the first main result.
Theorem A. Let D j C   be an open set and A j   a linearly measurable subset of D j   such that D j   is locally rectifiable on A j ,   mes ( A j ) > 0 ,   j = 1 , , N .   Let f : X C   be such that:
  • (i) f   is locally bounded on X   and f O s ( X o ) ;  
  • (ii) f | A   is measurable with respect to the N   -dimensional Hausdorff measure on A ;  
  • (iii) for any j { 1 , , N } ,   there is a function f j : A C   such that, for any ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N ) ,   the holomorphic function f ( a , , a ) | D j   (see (i)) has the angular limit f j ( a , a j , a )   at a j   for a.e. a j A j   and f 1 = = f N = f   a.e. on A .  
Then there exists a unique function f ^ O ( X ^ o )   with the following property:
1) There are subsets A ~ 1 A 1 A 1 * , , A ~ N A N A N *   such that
  • 1a) mes ( A j \ A ~ j ) = 0 ,   j = 1 , , N ;   1
  • 1b) f ^   can be extended to a function (still denoted by) f ^ O ( X ^ ( A ~ 1 , , A ~ N ; D 1 , , D N ) )   which admits the angular limit f ^ ( a )   at every point a X ^ ( A ~ 1 , , A ~ N ; D 1 , , D N ) \ A ;  
  • 1c) f ^ = f   on X ( A ~ 1 , , A ~ N ; D 1 , , D N ) \ A .  
In addition, f ^   enjoys the following properties:
2) If | f | X < ,   then | f ^ ( z ) | | f | A 1 ω ( z ) | f | X ω ( z ) , z X ^ o .   3) For any j { 1 , , N }   and a 0 = ( a , z j 0 , a ) A 1 * × × A j 1 * × D j × A j + 1 * × × A N * ,   if lim a a 0 , a X f ( a )   exists, then f ^   admits the angular limit lim a a 0 , a X f ( a )   at a 0 .   4) For any a 0 A * ,   if lim a a 0 , a A f ( a )   exists, then f ^   admits the angular limit lim a a 0 , a A f ( a )   at a 0 .   5) If f | A   can be extended to a continuous function defined on A * ,   then f   can be extended to a unique continuous function (still denoted by) f   defined on X * : = X ( A 1 * , . . . , A N * ; D 1 , . . . , D N )   and f ^   admits the angular limit f ( a )   at every a X *   and f 1 = = f N = f   on A A * .  
Theorem A has an immediate consequence.
Corollary A'. We keep the hypotheses and the notation of Theorem A. Suppose in addition that f C ( X o ) .   Then there exists a unique function f ^ O ( X ^ o )   with the following property:
  • 1a') f ^   can be extended to a function (still denoted by) f ^ O ( X ^ ( A 1 A 1 * , , A N A N * ; D 1 , , D N ) ) ,   which admits the angular limit f ^ ( a )   at every point a X ^ ( A 1 A 1 * , , A N A N * ; D 1 , , D N ) \ A ;  
  • 1b') f ^ = f   on ( X X * ) \ A .  
It is worthy to note that Theorem A and Corollary A' generalize, in some sense, Theorems 1–3.
Now we drop the hypothesis on local boundedness and continuity of f .   Then the examples of Drużkowski in [2(see Section 11 below) show that, without these conditions, the extended function f ^   (if it does exist) is, in general, not continuous on X ^ .   However, our second main result gives a partially positive answer to this question.
Theorem B. Let D j C   be an open set and A j   a linearly measurable subset of D j   such that D j   is locally rectifiable on A j ,   mes ( A j ) > 0 ,   and A j * D j ,   j = 1 , , N .   Let f : X C   satisfy the following properties:
  • (i) f | A C s ( A )   and f O s ( X o ) ;  
  • (ii) for any j { 1 , , N }   and for any ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N ) ,   the function f ( a , , a )   is locally bounded on D j A j   and the (holomorphic) restriction function f ( a , , a ) | D j   has the angular limit f ( a , a j , a )   at a j   for every a j A j .  
Then there are subsets A ~ 1 A 1 A 1 * , , A ~ N A N A N * ,   and a unique function f ^ O ( X ^ o )   with the following properties:
1) mes ( A j \ A ~ j ) = 0 ,   j = 1 , , N ;   2) f ^   can be extended to a function (still denoted by) f ^ O ( X ^ ( A ~ 1 , , A ~ N ; D 1 , , D N ) )   which admits the angular limit f ^ ( a )   at every point a X ^ ( A ~ 1 , , A ~ N ; D 1 , , D N ) ;   3) f ^ = f   on X ( A ~ 1 , , A ~ N ; D 1 , , D N ) .  
Observe that if f C s ( X ) O s ( X o ) ,   then conditions (i)–(ii) above are fulfilled.
Below we give some ideas how to prove Theorems A and B. Our method consists of two steps. In the first step we suppose that each D j   is a Jordan domain, j = 1 , , N .   In the second one we treat the general case. The key technique here is to use level sets of the harmonic measure. More precisely,we exhaust each D j   by the level sets of the harmonic measure ω ( , A j , D j ) ,   i.e. by D j , δ : = { z j D j : ω ( z j , A j , D j ) < 1 δ }   ( 0 < δ < 1 ) .   In order to carry out the first step, we improve Gonchar's method [5, 6and make intensive use of Carleman's formula and of geometric properties of the level sets of harmonic measures.
In the second step we apply some mixed cross type theorems (see [14) in order to prove Theorems A and B with D j   replaced by D j , δ .   Then we construct the solution for the original domains D j   by means of a gluing procedure.
Although our results have been stated for the general case N 2 ,   the proofs will be presented only in the case N = 2 .   At the end of Section 9 we will give the ideas how to get the general results.

4 Gonchar–Carleman operator

In this section we reformulate a result due to Gonchar [5, 6to our context. This result will play an important role for the proof of Theorems A and B. Let D j   be a rectifiable Jordan domain and let A j   be a linearly measurable subset of D j   such that mes ( A j ) > 0 ,   j = 1 , , N .   Let f   be a function defined on X : = X ( A 1 , , A N ; D 1 , , D N )   with the following properties:
  • (i) f | A   is measurable;
  • (ii) f O s ( X o ) ;  
  • (iii) there exist a constant C > 0   and N   functions f j : A C ,   j = 1 , , N ,   such that for any ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N ) ,   | f ( a , , a ) | D j < C ,   and f ( a , , a )   has the angular limit f j ( a , a j , a )   at a j   for a.e. a j A j ,   and f 1 = = f N = f   a.e. on A .  
For j = 1 , , N ,   let ω ~ j   be the conjugate function of ω ( , A j , D j )   such that ω ~ j ( z j 0 ) = 0   for a certain fixed point z j 0 D j   . Therefore, we can define the holomorphic functions g j ( z j ) : = ω ( z j , A j , D j ) + i ω ~ j ( z j ) ,   j = 1 , , N ,   and g ( z ) : = j = 1 N g j ( z j ) , z = ( z 1 , , z N ) D 1 × × D N .   Each function e g j   is bounded on D j ,   j = 1 , , N .   Therefore, in virtue of [4,p. 439, we may define e g j ( a j )   for a.e. a j A j   to be the angular boundary limit of e g j   at a j .   In virtue of (i), for each positive integer M ,   we define the Gonchar–Carleman operator as follows
K M ( z ) = K M [ f ] ( z ) : = 1 ( 2 π i ) N A e M ( g ( a ) g ( z ) ) f ( a ) d a a z , z D 1 × × D N , (4.1)
where d a : = d a 1 d a N ,   a z : = ( a 1 z 1 ) ( a N z N ) .   Moreover, we can extend K M   to ( D 1 A 1 ) × × ( D N A N )   in the following way. Let z = ( z 1 , , z N ) ( D 1 A 1 ) × × ( D N A N )   and J   the set of { j : 1 j N , z j A j }   . Set J : = { 1 , , N } \ J .   In the sequel one often identifies z   with ( z , z ) ,   where z : = ( z j ) j J   and z : = ( z j ) j J .   Then we define
(4.2) K M 1 , , M N ( z ) = K M 1 , , M N [ f ] ( z ) : = { f ( z ) , J = , 1 ( 2 π i ) N A e j J M j ( g j ( a j ) g ( z j ) ) f ( z , a ) d a a z , J ,
where M 1 , , M N N ,   A : = j J A j   , d a : = j J d a j   , and a z : = j J ( a j z j ) .   Moreover, we will write simply K M ( z )   in place of K M , , M ( z )   for any M N .   This notation is in accordance with ( 4.1 ).
The following Carleman Theorem due to Goluzin and Krylov (see, for example, [1,p.2) will be very useful.
Theorem 4.1. Let D   be a rectifiable Jordan domain and A   a linearly measurable subset of D   such that mes ( A ) > 0 .   Then for any function f H ( D ) ,   any relatively compact subset V D   , and any ε > 0 ,   there exists a positive integer M 0   (depending only on the sets D , A , V   and the number f H ( D )   ) such that | f ( z ) K M [ f | A ] ( z ) | < ε , z V , M M 0 ,   where K M   is given by (4.1) for N = 1   and f | A   is the angular boundary limit of f   on A .   (Notice that by [4,p.439, f | A   is linearly measurable).
  • Proof. For every z D ,   applying the Cauchy formula to the function w e M ( g ( w ) g ( z ) ) f ( w ) H ( D ) ,   M N ,   we obtain f ( z ) = 1 2 π i A e M ( g ( a ) g ( z ) ) f ( a ) d a a z + 1 2 π i D \ A e M ( g ( a ) g ( z ) ) f ( a ) d a a z .   Since | e ( g ( a ) g ( z ) ) | < 1   for a.e. a D \ A ,   the theorem follows.
Theorem 4.2. Let D j   be a rectifiable Jordan domain and A j   a linearly measurable subset of D j   such that mes ( A j ) > 0 ,   j = 1 , , N .   Let f   be a function defined on X   which satisfies (i)–(iii) above. Then 1) the following limit K ( z ) = K [ f ] ( z ) : = lim M K M ( z )   exists for all z X ^ ( D 1 A 1 ) × × ( D N A N ) ,   and its limit is uniform on compact subsets of X ^ o ;   2) there is a finite constant C 0   such that | K ( z ) | C 0 C j J dist ( z j , D j ) ( 1 e ( 1 ω ( z ) ) ) , z X ^ ( D 1 A 1 ) × × ( D N A N ) ,   where dist ( z j , D j ) : = inf ζ j D j | z j ζ j |   and C   is the constant given in (iii) above.
  • Proof. We briefly recall the argument of Gonchar in [6. Write the difference K M + 1 K M   as follows:
    K M + 1 K M = j = 1 N K M j , (4.3)
    where
    K M j : = K M + 1 , , M + 1 , M , , M j 1 K M + 1 , , M + 1 , M , , M j , j = 1 , , N .  
    Reasoning as in formulas (6)–(9) in [6and taking (iii) into account, we see that there is a constant C 0   such that
    | K M + 1 ( z ) K M ( z ) | j = 1 N | K M j ( z ) | C C 0 j J dist ( z j , D j ) e M ( 1 ω ( z ) ) (4.4)
    for z X ^ ( D 1 A 1 ) × × ( D N A N ) .   Since | e ( 1 ω ( z ) ) | < 1 ,   the desired conclusion of Part 1) follows immediately from this estimate. Finally, using ( 4.4 ), Part 2) follows.
The following version of Privalov's Uniqueness Theorem will be also needed.
Theorem 4.3. Let D C   be a domain which is locally rectifiable on a linearly measurable subset A   of D   with mes ( A ) > 0 .   Let f   be a holomorphic function on D   such that the angular limits of f   on the set A * D   are equal to 0 .   Then f 0 .  
  • Proof. Using the hypothesis, one may find a rectifiable Jordan domain U D   such that mes ( A U ) > 0   and f | U   admits the angular limit 0   at almost every point of A U .   By Privalov's Uniqueness Theorem (see [11), f | U 0 .   Hence, f 0 .  
Finally, the following result will play a key role in the proof of Theorems A and B.
Lemma 4.4. Let D   be a rectifiable Jordan domain, ( f n ) n = 0   a sequence of holomorphic functions on D   , and Δ   a linearly measurable subset of D   with mes ( Δ ) > 0 .   Assume that f n   admits the angular limit (denoted by) f n ( t )   at every point t Δ ,   n N   , 2 that lim n f n ( t ) = f 0 ( t ) ,   t Δ ,   and that sup n 0 | f n | D < .   Then ( f n ) n = 1   converges uniformly on every compact subset of D   to f 0 .  
  • Proof. Let g ( t , Δ , D ) : = ω ( t , Δ , D ) + i ω ~ ( t , Δ , D ) , t D ,   where ω ~ ( , Δ , D )   is the conjugate harmonic function of ω ( , Δ , D ) .   Since g O ( D )   is bounded, it follows from [4,p.439that for a.e. t D ,   g   admits an angular limit (denoted by g ( t )   ) at t .   Fix an arbitrary compact subset H   of D .   Recall that | f n | D < M < ,   n N .   Consequently, applying Theorem  4.1 we see that, for any ε > 0 ,   there is an M ε   such that, for any M M ε ,  
    | f n ( τ ) 1 2 π i Δ e M ( g ( t , Δ , D ) g ( τ , Δ , D ) ) f n ( t ) d t t τ | < ε , τ H , n N . (4.5)
    On the other hand, using that lim n f n ( t ) = f 0 ( t ) ,   t Δ ,   and applying Lebesgue's Dominated Convergence Theorem, we see that lim n 1 2 π i Δ e M ( g ( t , Δ , D ) g ( τ , Δ , D ) ) f n ( t ) d t t τ = 1 2 π i Δ e M ( g ( t , Δ , D ) g ( τ , Δ , D ) ) f 0 ( t ) d t t τ   for all τ H   and M > M ε .   This, combined with ( 4.5 ), implies that lim n f n ( τ ) = f 0 ( τ ) ,   τ H .   Hence, by the Montel Theorem the sequence ( f n ) n = 1   converges uniformly on compact subsets of D   to f 0 .   This completes the proof.

5 Properties of the harmonic measure and some elements of Conformal mapping theory

In this section we develop the tools needed for the proofs of Theorems A and B. In the sequel, D C   is an open set and A   is a linearly measurable nonempty subset of D   such that D   is locally rectifiable on A .   Observe that under the above assumption D   is non-polar. Let * D   be the set of all points ζ D   where D   admits a tangent. Recall from Subsection 2.1 that mes ( A * D ) = mes ( A ) .   Let P D   be the generalized Poisson integral of D .   If, in addition, A   is a Borel set, then, by Theorem 4.3.3 of [15, the harmonic measure of D \ A   is given by
ω ( , A , D ) = P D [ 1 D \ A ] . (5.1)
Next, let E   denote the unit disc of C .   Then, for an f L 1 ( E ) ,   a point ζ E   is said to be a Lebesgue point of f   if lim r 0 1 mes ( E B ( ζ , r ) ) E B ( ζ , r ) | f ( θ ) f ( ζ ) | d θ = 0 ,   where d θ   is the linear measure defined on E   . If A   is a linearly measurable subset of E   and f : = 1 A ,   then every point of A   that is a Lebesgue point of f   is called a density point of A .  
Proposition 5.1. 1) Let f L 1 ( E ) .   Then a.e. points of E   are Lebesgue points of f   and lim z ζ , z A α ( ζ ) P E [ f ] = f ( ζ ) , 0 < α < π 2 ,   for every Lebesgue point ζ   of f .   2) Let A   be a subset of positive linear measure of E   and A   the set of all density points of A .   Then mes ( A \ A ) = 0   and lim z ζ , z A α ( ζ ) P E [ 1 E \ A ] = 0 , 0 < α < π 2 ,   for every ζ A .   Moreover, if N   is a subset of E   with mes ( N ) = 0 ,   then lim z ζ , z E P E [ 1 E \ A ] = 0 , 0 < α < π 2 ,   for all interior points ζ   of A N .  
  • Proof. Part 1) is classical (see, for example, Theorem 5.4.8 in [17). The first assertion of Part 2) is also classical. Applying Part 1) to the function f : = 1 E \ A ,   the second assertion of Part 2) follows. The last assertion is almost trivial.
We recall the following well-known result due to Carathéodory and F. and M. Riesz (see [4,p.44,p.420).
Theorem 5.2. Let D C   be a Jordan domain and let Φ   be a conformal mapping from D   onto the unit disc E .   1) Then Φ   extends to a (unique) homeomorphic map (still denoted by) Φ   from D ¯   onto E ¯ .   If, in addition, D   is a rectifiable Jordan domain, then, for any linearly measurable subset A   of D ,   mes ( A ) > 0   if and only if mes ( Φ ( A ) ) > 0 .   2) Suppose that γ : [ 0 , 1 ] C   is a one-to-one real analytic map and U   is an open set such that γ ( [ 0 , 1 ] ) U D ,   and U D   is a Jordan domain. Then Φ   extends to a conformal mapping from D γ ( [ 0 , 1 ] )   onto E Φ ( γ ( [ 0 , 1 ] ) )   (i.e. Φ ( z )   exists and is nonzero for z γ ( [ 0 , 1 ] )   ).
Now we are ready to formulate the following
Definition 5.3. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D .   A point ζ A   is said to be a density point of A   3 if one of the following cases happens:
Case 1: ζ   is of type I. There are an open neighborhood U   of ζ   such that U D   is a Jordan domain and a conformal mapping Φ   from U D   onto the unit disc E   which extends homeomorphically from U D ¯   onto E ¯   such that Φ ( ζ )   is a density point of the set Φ ( ( U D ) A ) .   In other words, lim r 0 mes ( Φ ( ( U D ) A ) B ( Φ ( ζ ) , r ) ) mes ( E B ( Φ ( ζ ) , r ) ) = 1 .   Case 2: ζ   is of type II. There are an open neighborhood U   of ζ   such that U D = U 1 U 2   with Jordan domains U 1 , U 2 ,   and conformal mappings Φ j   ( j = 1 , 2 )   from U j   onto E   which extends homeomorphically from U j ¯   onto E ¯   such that Φ j ( ζ )   is a density point of the set Φ j ( U j A ) .   In other words, lim r 0 mes ( Φ j ( U j A ) B ( Φ j ( ζ ) , r ) ) mes ( E B ( Φ j ( ζ ) , r ) ) = 1 , j = 1 , 2 .  
Proposition 5.4. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D .   1) Then Definition  5.3 is independent of the choice of U   and Φ   in Case 1 (resp. U ,   Φ 1 ,   Φ 2   in Case 2).
2) Let A   denote the set of all density points of A .   Then mes ( A \ A ) = 0 .  
  • Proof. To prove Part 1), let ζ 0   be a density point of A .   We consider two cases.
    Case 1: ζ 0   is of type I. In virtue of Definition  5.3 and the assumption that ζ 0   is of type I, for each j { 1 , 2 }   let U j   be an open neighborhood of ζ 0   such that U j D   is a Jordan domain and let Φ j   be a conformal mapping from U j D j   onto E   which extends homeomorphically from U j D ¯   onto E ¯ .   Suppose that Φ 1 ( ζ 0 )   is a density point of Φ 1 ( ( U 1 D ) A ) E .   We would like to show that Φ 2 ( ζ 0 )   is a density point of Φ 2 ( ( U 2 D ) A ) E .   Let V   be an open neighborhood of ζ 0   such that V U 1 U 2 ,   V D   is a rectifiable Jordan domain and V D   is connected. We deduce that Φ j ( V D )   is an open arc of E   (which is obviously real analytic) and Φ j ( V D )   is a Jordan subdomain of E   for j { 1 , 2 } .   Let Ψ 1   be a conformal mapping from the Jordan domain Φ 1 ( V D )   onto E   which extends homeomorphically from Φ 1 ( V D ) ¯   onto E ¯ .   By Part 2) of Theorem  5.2 and the fact that Φ 1 ( V D )   is an open real analytic arc, we see that Ψ 1   extends conformally to the arc Φ 1 ( V D ) E ,   and ( Ψ 1 Φ 1 ) ( V D )   is an arc of E .   Next, consider the conformal mapping Ψ 2 : Φ 2 ( V D ) E   given by Ψ 2 ( z ) : = ( Ψ 1 Φ 1 Φ 2 1 ) ( z ) , z Φ 2 ( V D ) .   Since by Part 1) of Theorem  5.2  Φ 2 ( V D )   is a Jordan domain, Ψ 2   extends homeomorphically from Φ 2 ( V D ) ¯   onto E ¯   and satisfies the equation Φ 2 Φ 1 1 = Ψ 2 1 Ψ 1 .   Since Φ 2 ( V D ) E   is an open real analytic arc, applying again Part 2) of Theorem  5.2 , we deduce that Ψ 2   extends conformally to the arc Φ 2 ( V D ) E ,   and ( Ψ 2 Φ 2 ) ( V D )   is an arc of E .   We summarize what has been done so far: We have shown that Ψ 2 1 Ψ 1   is conformal from the arc Φ 1 ( V D )   onto the arc Φ 2 ( V D )   and Φ 2 Φ 1 1 = Ψ 2 1 Ψ 1 .   Hence, | ( Φ 2 Φ 1 1 ) ( η ) | > 0 ,   η Φ 1 ( V D ) .   Choosing η 0 : = Φ 1 ( ζ 0 )   and noting that ( Φ 2 Φ 1 1 ) ( η 0 ) = Φ 2 ( ζ 0 ) ,   we deduce that
    lim r 0 mes ( Φ 2 ( ( U 2 D ) A ) B ( Φ 2 ( ζ 0 ) , r ) ) mes ( E B ( Φ 2 ( ζ 0 ) , r ) )
    = lim r 0 mes ( Φ 2 ( ( V D ) A ) B ( Φ 2 ( ζ 0 ) , r ) ) mes ( E B ( Φ 2 ( ζ 0 ) , r ) )
    = lim r 0 mes ( Φ 2 ( ( V D ) A ) B ( Φ 2 ( ζ 0 ) , | ( Φ 2 Φ 1 1 ) ( η 0 ) | r ) ) mes ( E B ( Φ 2 ( ζ 0 ) , | ( Φ 2 Φ 1 1 ) ( η 0 ) | r ) )
    = lim r 0 mes ( Φ 1 ( ( V D ) A ) B ( Φ 1 ( ζ 0 ) , r ) ) mes ( E B ( Φ 1 ( ζ 0 ) , r ) )
    = 1 ,
    where the last identity follows from the assumption that Φ 1 ( ζ 0 )   is a density point of Φ 1 ( ( V D ) A ) E .   Consequently, Φ 2 ( ζ 0 )   is a density point of Φ 2 ( ( V D ) A ) E .   Hence the proof of Part 1) is complete in this first case.
    Case 2: ζ 0   is of type II. Then there is an open neighborhood U   of ζ 0   such that U D = U U ,   where U ,   U   are some Jordan domains. We apply the result of Case 1 to each Jordan domain U ,   U .   Hence the proof of Part 1) is finished.
    It remains to prove Part 2). We may find a sequence ( U k ) k = 1   of open sets of C   such that U k D   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains and A k = 1 ( U k D ) .   Using Part 1) and Proposition  5.1 , we see that almost every point in ( A ( D U k ) )   is a density point of this set, i.e. belongs to ( A ( D U k ) ) .   On the other hand, clearly ( A ( D U k ) ) A .   Consequently, mes ( A \ A ) k = 1 mes ( ( A ( D U k ) ) \ ( A ( D U k ) ) ) = k = 1 0 = 0 .   Hence, A   is linearly measurable and mes ( A \ A ) = 0 .   This completes the proof of Part 2).
The following classical result will be needed to prove Proposition  5.6 below (see [11,p.60or [13,p.51).
Theorem 5.5. (A theorem of Lindelöf ) Let D   be a Jordan domain and let Φ   be a conformal mapping from E   onto D .   By Part 1) of Theorem  5.2 we still denote by Φ   its homeomorphic extension from E ¯   onto D ¯ .   Let ζ = e i θ E   and let β : = lim t θ + arg ( Φ ( e i t ) Φ ( e i θ ) ) .   Then D   has a corner 4 of opening π α   ( 0 α 2 )   at Φ ( ζ )   if and only if arg Φ ( z ) Φ ( ζ ) z ζ β α ( θ + π 2 ) as z ζ , z E .  
The next proposition justifies the coherence of the definitions of the tangent line and the angular approach regions given in Subsection 2.1.
Proposition 5.6. Let D C   be an open set and let ζ * D .   1) Suppose that ζ   is either of type I or of type II. For j { 1 , 2 } ,   let γ j : [ a j , b j ] D   be a parametrization of one Jordan curve contained in D   such that γ j ( c j ) = ζ   , c j ( a j , b j )   . Suppose that the following limit exists lim t c 1 γ 1 ( t ) ζ t c 1 | t c 1 | | γ 1 ( t ) ζ | = λ 1 .   Then the following limit also exists lim t c 2 γ 2 ( t ) ζ t c 2 | t c 2 | | γ 2 ( t ) ζ | = λ 2 ,   and λ 2 = ± λ 1 .   In other words, the tangent line at ζ   is independent of the choice of parametrizations.
2) Suppose that ζ   is of type I. Suppose that U   is a neighborhoods of ζ   such that D U   is a Jordan domain. Let Φ   be a conformal mapping from D U   onto E   such that Φ ( ζ ) = 1 ,   where Φ ( ζ )   is defined as in Part 1) of Theorem  5.2 . Then for any 0 < α < β < γ < π 2 ,   and any neighborhood V   of ζ ,   there are neighborhoods R V   of ζ   and S   of 1   such that Φ ( A α ( ζ ) R ) A β ( 1 ) S Φ ( A γ ( ζ ) R ) .   3) Suppose that ζ   is of type II. Suppose that U   is a neighborhood of ζ   such that D U = U 1 U 2   with some disjoint Jordan domains U 1 , U 2   satisfying ζ * U 1 * U 2 .   For j { 1 , 2 } ,   let Φ j   be a conformal mapping from D U j   onto E   such that Φ j ( ζ ) = 1 .   Then for any 0 < α < β < γ < π 2 ,   j { 1 , 2 } ,   and any neighborhood V   of ζ ,   there are neighborhoods R V   of ζ   and S   of 1   such that Φ j ( A α ( ζ ) U j R ) A β ( 1 ) S Φ j ( A γ ( ζ ) U j R ) .  
  • Proof. First one proves Part 1). Without loss of generality suppose that γ 1 ( [ a 1 , b 1 ] ) = γ 2 ( [ a 2 , b 2 ] ) D .   Then the function γ : = γ 1 1 γ 2 : [ a 2 , b 2 ] [ a 1 , b 1 ]   is a bijective continuous map and γ ( c 2 ) = c 1 .   Therefore, it is either monotone increasing or monotone decreasing. Observe that
    lim t c 2 ( γ 2 ( t ) ζ t c 2 | t c 2 | | γ 2 ( t ) ζ | ) = lim t c 2 ( ( γ 1 γ ) ( t ) ζ t c 2 | t c 2 | | ( γ 1 γ ) ( t ) ζ | )
    = lim t c 2 ( ( γ 1 γ ) ( t ) ζ γ ( t ) c 1 | γ ( t ) c 1 | | ( γ 1 γ ) ( t ) ζ | ) lim t c 2 ( γ ( t ) c 1 t c 2 | t c 2 | | γ ( t ) c 1 | )
    = λ 1 ε ( γ ) ,
    where ε ( γ ) : = { 1 , if γ is monotone increasing 1 , if γ is monotone decreasing .   Hence, λ 2 = ± λ 1   and the proof of Part 1) is complete.
    To prove Part 2), let R   be an open neighborhood of ζ   such that R U V   and R D   is a Jordan domain. Then Φ ( R D )   is a Jordan subdomain of E .   Let Ψ   be a conformal mapping from E   onto the Jordan domain R D   such that Ψ ( 1 ) = ζ .   Then Φ Ψ   is a conformal mapping from E   onto the Jordan domain Φ ( R D )   and ( Φ Ψ ) ( 1 ) = Φ ( ζ ) .   By Part 1) of Theorem  5.2 , we have that
    lim z 1 , z E Ψ ( z ) = Ψ ( 1 ) and lim z 1 , z E ( Φ Ψ ) ( z ) = ( Φ Ψ ) ( 1 ) = Φ ( ζ ) . (5.2)
    On the other hand, observe that ζ * ( R D )   as ζ * D .   Moreover, Φ ( ζ ) * ( Φ ( R D ) )   because ( Φ ( R D ) )   contains a neighborhood of Φ ( ζ )   in E .   Consequently, applying Theorem  5.5 to the two conformal mappings Ψ   and Φ Ψ ,   we obtain the existence of the following two limits
    lim z 1 , z E arg ( Ψ ( z ) Ψ ( 1 ) z 1 ) and lim z 1 , z E arg ( ( Φ Ψ ) ( z ) ( Φ Ψ ) ( 1 ) z 1 ) , (5.3)
    where arg z   denotes the argument of a complex number z .   Choose an α   such that α < α < β .   Then using ( 5.2 ) and ( 5.3 ), it follows that there is an open neighborhood T   of 1   such that
    A α ( ζ ) Ψ ( T E ) Ψ ( A α ( 1 ) T ) , ( Φ Ψ ) ( A α ( 1 ) T ) A β ( 1 ) ( Φ Ψ ) ( T E ) . (5.4)
    Let R   be a neighborhood of ζ   such that R D = Ψ ( T E )   and S   a neighborhood of 1   such that S E = ( Φ Ψ ) ( T E ) .   Then we deduce immediately from ( 5.4 ) that Φ j ( A α ( ζ ) U j R ) A β ( 1 ) S .   The remaining inclusion of Part 2) can be proved in exactly the same manner.
    For Part 3) one argues as in the proof of Part 2). Hence the proof of the proposition is complete.
The following two lemmas will be very useful.
Lemma 5.7. Let D C   be a bounded domain with C 2   smooth boundary and let A   be a linearly measurable subset of D .   Then ω ( z , A , D ) = P D [ 1 D \ A ] ( z ) , z D .  
  • Proof. Since D   is bounded with C 2   boundary, using ( 5.1 ) it is a classical fact (see, for example, [19) that
    ω ( z , B , D ) = P D [ 1 D \ B ] ( z ) : = D P ( z , ζ ) 1 D \ B ( ζ ) d σ ( ζ ) , z D , (5.5)
    for any Borel subset B   of D .   Here P ( , )   is the Poisson kernel of D   and d σ   is the Lebesgue boundary measure of D .   Since A   is linearly measurable, there are two Borel sets A 1 ,   A 2   such that A 1 A A 2   and mes ( A 2 \ A 1 ) = 0 .   It follows from definition that
    ω ( z , A 1 , D ) ω ( z , A , D ) ω ( z , A 2 , D ) , z D . (5.6)
    On the other hand, ( 5.5 ) and the assumption on A 1   and A 2   imply that ω ( z , A 1 , D ) = ω ( z , A 2 , D ) = P D [ 1 D \ A ] ( z ) , z D .   This, combined with ( 5.6 ), gives the desired conclusion.
Lemma 5.8. Let E   be the unit disc, A   a linearly measurable subset of E ,   and u   a subharmonic function defined on E   with u 1   . Let α ( 0 , π 2 )   be such that limsup z ζ , z A α ( ζ ) u ( z ) 0 for a.e. ζ A .   Then u ω ( , A , E )   on E .  
  • Proof. Fix a point z 0 E .   We wish to show that
    u ( z 0 ) ω ( z 0 , A , E ) . (5.7)
    As in the proof of Lemma  5.7 , we may assume (without loss of generality) that A   is a Borel set. Moreover, we may suppose that u   is continuous at z 0 .   Otherwise, one considers a Poisson Modification v   of u   given by v ( w ) : = { u ( w ) , w E \ B ( z 0 , r ) P B ( z 0 , r ) [ u | B ( z 0 , r ) ] , w B ( z 0 , r ) .   where r > 0   is chosen so that B ( z 0 , r ) E .   Then by the hypothesis on u   and by Lemma 4.1.3 in [15, we deduce that v   is subharmonic on E ,   u v 1   on E ,   and v | B ( z 0 , r )   is harmonic.
    For 0 < r < 1   put u r ( z ) : = u ( r z ) ,   z E ¯ .   We obtain
    u ( z 0 ) = lim r 1 u ( r z 0 ) limsup r 1 P E [ u r ] ( z 0 ) P E [ limsup r 1 u r ] ( z 0 )
    P E [ 1 E \ A ] ( z 0 ) = ω ( z 0 , A , E ) ,
    where the second estimate holds because of Fatou's Lemma, the third one follows from the hypothesis on u ,   and the last equality is a consequence of ( 5.1 ). This proves ( 5.7 ). Hence, the proof is complete.
The following estimate will be crucial for the future development.
Proposition 5.9. Let D C   be an open set which is locally rectifiable on linearly measurable subsets A ,   N   of D   with mes ( N ) = 0 .   Then 1) all density points of A   which are elements of * D   are locally regular points of A   and mes ( A \ A * ) = 0 ;   2) for every ζ A *   and every open neighborhood U   of ζ ,   it holds sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) ω ( z , A , D ) = 0 ,   and mes ( A U ) > 0 ;   3) for any interior point ζ   of A N ,   we have lim z ζ , z D ω ( z , A , D ) = 0 .  
  • Proof. To prove Part 1), let ζ 0   be a density point of A .   There are two cases to consider according to the type of ζ 0 .   We only treat the case when ζ 0   is of type I. The second case where ζ 0   is of type I I   is almost analogous, and therefore, left to the interested reader.
    Since ζ 0   is of type I and D   is rectifiable at ζ 0 ,   there is an open neighborhood U   of ζ 0   such that U D   is a rectifiable Jordan domain. Fix a conformal mapping Φ   from U D   onto E   which extends to a homeomorphism from U D ¯   onto E ¯ .   By Definition  5.3 and Proposition  5.4 , η 0 : = Φ ( ζ 0 )   is a density point of the linearly measurable subset Φ ( ( U D ) A )   of E .   Consequently, applying identity ( 5.1 ) and Part 1) of Proposition  5.1 yields that
    sup 0 < α < π 2 limsup w A α ( η 0 ) ω ( w , Φ ( ( U D ) A ) , E ) = 0 . (5.8)
    On the other hand, one has
    ω ( z , ( U D ) A ) , U D ) = ω ( Φ ( z ) , Φ ( ( U D ) A ) , E ) , z U D . (5.9)
    Applying Theorem  5.5 to Φ 1   at η 0   and combining ( 5.8 ) and ( 5.9 ), it follows that sup 0 < α < π 2 limsup z A α ( ζ 0 ) ω ( z , ( U D ) A , U D ) = 0 .   Hence, ζ 0 A * .   This proves the first assertion of Part 1). Let A   denote the set of all density points of A .   By Part 2) of Proposition  5.4 , mes ( A \ A ) = 0 .   On the other hand, we have already shown that A * D A * .   Consequently, mes ( A \ A * ) mes ( A \ ( A * D ) ) = mes ( A \ A ) + mes ( A \ * D ) = 0 ,   which completes the proof of the last assertion of Part 1).
    Part 2) follows from the definition of regular points and the Subordination Principle (see Corollary 4.3.9 in [15).
    In virtue of Lemma  5.8 and the second assertion of Part 2) of Proposition  5.1 , the proof of Part 3) follows along the same lines as that of Part 2).
In the sequel we formulate some important stability property of the angular harmonic measure. Recall that D   is a proper open subset of C { }   such that the boundary D   (with respect to C { }   ) is non-polar. Let A   be a linearly measurable subset of * D .   Let φ : D R   be a bounded function. The associated Perron function H D , A : D R   is defined by
H D , A φ : = sup u U ^ u , (5.10)
where U ^ = U ^ ( φ , A , D )   denotes the family of all subharmonic functions u   on D   such that
limsup z ζ u ( z ) φ ( ζ ) , ζ D \ A ,
limsup z ζ , z A α ( ζ ) u ( z ) φ ( ζ ) , ζ A , 0 < α < π 2 .
In the sequel, U ^ ( A , D )   will stand for U ^ ( 1 D \ A , A , D ) .  
Proposition 5.10. (Maximum Principle) Let u S ( D )   be such that u   is bounded from the above and
limsup z ζ u ( z ) 0 , ζ D \ A ,
limsup z ζ , z A α ( ζ ) u ( z ) 0 , ζ A , 0 < α < π 2 .
Then u 0   on D .  
  • Proof. Suppose that u < M   for some M .   Let ζ 0   be an arbitrary point of A .   Fix a rectifiable Jordan domain U   such that U D   and U D   is a neighborhood of ζ 0   in D .   Applying Lemma  5.8 and Part 3) of Proposition  5.9 to u | U   yields that limsup z ζ , z U . u ( z ) 0 .   Since ζ 0   is an arbitrary point of A   , the desired conclusion follows from the classical Maximum Principle (see Theorem 2.3.2 in [15).
Using the above proposition, the corresponding results in Sections 4.1 and 4.2 of [15with respect to H D , A   (instead of H D   ) are still valid making the obviously necessary changes. In particular, we have the following (see Corollary 4.2.6 in [15):
Proposition 5.11. Let D   be a proper open subset of C { }   such that the boundary D   (with respect to C { }   ) is non-polar. Let A   be a linearly measurable subset of * D   and φ : D R   a bounded function which is continuous n.e. on D .   Then there exists a unique bounded harmonic function h   on D   such that lim z ζ h ( z ) = φ ( ζ )   for n.e. ζ D .   Moreover, h = H D φ = H D , A φ .  
In virtue of this result, Theorem 4.3.3 in [15is still valid in the context of H D , A .   More precisely,
Proposition 5.12. Let D   be a proper open subset of C { }   such that D   (with respect to C { }   ) is non-polar. Let A   be a linearly measurable subset of * D   and φ : D R   a bounded Borel function. Then H D φ = H D , A φ = P D [ φ ] .  
We also need the following
Proposition 5.13. Let D   be a proper open subset of C { }   such that D   (with respect to C { }   ) is non-polar. Let A   be a Borel subset of D   such that A * D   and mes ( A ) = 0 .   Then P D [ 1 A ] 0   on D .  
  • Proof. Suppose without loss of generality that D   is locally rectifiable on the interval [ 0 , 1 ] D   and that A   is a Borel subset of [ 0 , 1 ]   with mes ( A ) = 0 .   Since D C \ [ 0 , 1 ] ,   it follows from the Subordination Principle that P D [ 1 A ] P C \ [ 0 , 1 ] [ 1 A ] on D .   Therefore, it suffices to show that P C \ [ 0 , 1 ] [ 1 A ] 0   on C \ [ 0 , 1 ] .   To this end consider the conformal mapping Φ ( z ) : = 1 z 1   which maps C { } \ [ 0 , 1 ]   onto H : = { z C : Im z > 0 } .   It is not difficult to show that P C \ [ 0 , 1 ] [ 1 A ] = P H [ 1 Φ ( A ) ] Φ 1 0 .   This concludes the proof.
Now we arrive at one of the main results of the section
Theorem 5.14. Let D C   be a proper open subset of C { }   such that D   (with respect to C { }   ) is non-polar. Suppose that D   is locally rectifiable on a linearly measurable subset A   of D .   Let B   be a linearly measurable subset of D   with B A A *   and mes ( A \ B ) = 0 .   Then ω ( z , A , D ) = H D , B 1 D \ B   for z D .   In particular, for every subset N   of D   with mes ( N ) = 0 ,   we have ω ( z , A \ N , D ) = ω ( z , A , D )   for z D .  
  • Proof. Replacing A   by two Borel sets A 1 , A 2   such that A 1 B   and A A 2 D   and mes ( B \ A 1 ) = mes ( A 2 \ A ) = 0 ,   one gets that mes ( A 2 \ A 1 ) = 0 .   Then we conclude by the Subordination Principle and Proposition  5.13 that ω ( z , A 2 , D ) ω ( z , A , D ) ω ( z , B , D ) ω ( z , A 1 , D ) = ω ( z , A 2 , D ) , z D .   In virtue of Proposition  5.12 , we have that ω ( z , A 1 , D ) = H D , A 1 1 D \ A 1 .   On the other hand, it follows from the definition that H D , A 1 1 D \ A 1 H D , B 1 D \ B ω ( , A , D ) on D .   Combining the above three estimates, the proof of the first assertion of the theorem follows.
    The second assertion is a direct consequence of the first one.
Proposition 5.15. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D .   Let ( D k ) k = 1   be a sequence of open subsets D k   of D   and ( A k ) k = 1   a sequence of measurable subsets of A   such that
  • (i) D k D k + 1   and k = 1 D k = D ;  
  • (ii) A k A k + 1   and A k D D k   and D k   is locally rectifiable on A k   and k = 1 A k = A ;  
  • (iii) for any point ζ A   there is an open neighborhood V = V ζ   of ζ   in C   such that V D   = V D k   for some k .  
Then ω ( z , A , D ) = lim k ω ( z , A k , D k ) , z D .  
Remark 5.16. We may always choose sequences ( D k ) k = 1   and ( A k ) k = 1   such that (i)–(iii) are satisfied and the open sets D k   are bounded. For example, take D k : = D B ( 0 , k )   and A k : = A B ( 0 , k ) ,   k 1 ,   k N .  
  • Proof. Using the Subordination Principle it is easy to see that the sequence ( ω ( , A k , D k ) ) k = 1   is decreasing and the following limit u : = lim k ω ( , A k , D k )   exists and defines a subharmonic function in D .   By the Subordination Principle again, we have u ω ( , A , D ) .   Therefore, it remains to establish the converse inequality. In virtue of (iii) and of Part 2) of Proposition  5.9 , we conclude that
    sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u = 0 , ζ B , (5.11)
    where B : = k = 1 A k * .   On the other hand, in virtue of Part 1) of Proposition  5.9 , we have that mes ( ( A A * ) \ B ) k = 1 mes ( A k \ A k * ) = 0 .   Consequently, applying Theorem  5.14 , we deduce from ( 5.11 ) that u ( z ) ω ( z , A , D ) ,   z D .   This completes the proof.
Next, we introduce a notion which will be relevant for our further study.
Definition 5.17. Let D , G C   be two open sets such that G D .   A point ζ * D   is said to be an end-point of G   in D   if, for every 0 < α < π 2 ,   there is an open neighborhood U = U α   of ζ   such that U A α ( ζ ) G .   The set of all end-points of G   in D   is denoted by G D .  
The remaining part of this section is devoted to the study of level sets of the harmonic measure. We begin with the following important properties of these sets.
Theorem 5.18. Let D C   be an open set and A   a linearly measurable subset of D   such that D   is locally rectifiable on A   and mes ( A ) > 0 .   Then, for any 0 < ε < 1 ,   the ” ε   -level set” D ε : = { z D : ω ( z , A , D ) < 1 ε }   enjoys the following properties:
  • (i) Let G 1 ,   G 2   be arbitrary distinct connected components of D ε ,   then G 1 D G 2 D = .  
  • (ii) For any point ζ A * ,   there is exactly one connected component G   of D ε   such that ζ G D .  
  • (iii) G D A   is linearly measurable and mes ( G D A ) > 0   for every connected component G   of D ε .  
  • Proof. To prove (i), suppose, in order to reach a contradiction, that G 1 D G 2 D .   Fix a point ζ 0 G 1 D G 2 D .   Then, for any 0 < α < π 2 ,   there is an open neighborhood U α   of ζ 0   such that A α ( ζ ) U α G 1 G 2 .   This implies that G 1 G 2 .   Hence, G 1 = G 2 ,   which contradicts the hypothesis that G 1 G 2 .   The proof of (i) is complete.
    Next, we turn to the proof of (ii). Fix a ζ 0 A * .   In virtue of assertion (i), it suffices to show the existence of a connected component G   of D ε   such that ζ 0 G D .   Applying Part 2) of Proposition  5.9 , we see that, for every 0 < α < π 2 ,   there is an open neighborhood U α   of ζ 0   such that
    A α ( ζ 0 ) U α D ε . (5.12)
    Fix an arbitrary 0 < α 0 < π 2 ,   and let G   be the connected component of D ε   containing A α 0 ( ζ 0 ) U α 0 .   Since ( A α 0 ( ζ 0 ) U α 0 ) ( A α ( ζ 0 ) U α ) , 0 < α < π 2 ,   we deduce from ( 5.12 ) that G   also contains A α ( ζ 0 ) U α   for every 0 < α < π 2 .   Hence ζ 0 G D .   The proof of (ii) is finished.
    Finally, we prove (iii). First, we may find a sequence ( U k ) k = 1   of open sets of C   such that U k D   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains and A k = 1 ( U k D ) .   Since A   is linearly measurable, we see that in order to prove the measurability of G D A ,   it is sufficient to check that G D ( D U k )   is linearly measurable for every k 1 .   To prove the latter assertion, fix an k 0 1   and let U : = U k 0 .   Let Φ   be a conformal mapping from D U   onto E   which extends to a homeomorphic mapping (still denoted by) Φ   from D U ¯   onto E ¯ .   Applying Part 2) of Proposition  5.6 and using Definition  5.17 , we see that, for any ζ ( D U ) ,   ζ G D   if and only if Φ ( ζ ) [ Φ ( G U ) ] E .   We shall prove, in the sequel, that [ Φ ( G U ) ] E   is a Borel subset of E .   Taking this for granted, then G D ( D U )   is also a Borel set. Consequently, G D A   is linearly measurable.
    To check that [ Φ ( G U ) ] E   is a Borel set, put
    A n , m ( η ) : = { w E A ( 1 1 n ) π 2 ( η ) : | w η | < 1 m } , n , m 1 , η E . (5.13)
    For any n , m , p 1 ,   let
    (5.14) T n m p : = { η E : A n , m ( η ) Φ ( G U ) and ω ( Φ 1 ( w ) , A , D ) 1 ε 1 p , w A n , m ( η ) } .
    We observe the following:
    Geometric fact. Let η 0 E   and ( η q ) q = 1 E   such that lim q η q = η 0 .   Then A n , m ( η 0 ) q = 1 A n , m ( η q ) .   The proof of this fact follows immediately from the geometric shape of the cone A n , m ( η )   given in ( 5.13 ).
    Let ( η q ) q = 1 T n m p   such that lim q η q = η 0 E .   Using the above geometric fact, we see that A n , m ( η 0 ) Φ ( G U ) .   This, combined with ( 5.14 ) and the continuityof ω ( Φ 1 ( ) , A , D ) | E ,   implies that η 0 T n m p .   Hence, the set T n m p   is closed. Clearly, we have [ Φ ( G U ) ] E = n = 1 m = 1 p = 1 T n m p   It follows immediately from this identity that [ Φ ( G U ) ] E   is a Borel set.
    Consequently, as was already discussed before, G D A   is linearly measurable.
    To finish assertion (iii), it remains to prove that mes ( G E A ) > 0 .   Suppose, in order to reach a contradiction, that mes ( G E A ) = 0 .   Consider the following function u ( z ) : = { ω ( z , A , D ) , z D \ G 1 ε , z G .   Then clearly u S ( D )   and u 1 .   In virtue of assertions (i) and (ii) and Part 2) of Proposition  5.9 , we have that sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u ( z ) = sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) ω ( z , A , E ) = 0 , ζ ( A A * ) \ ( G D A ) .   Consequently, using the notation in ( 5.10 ), we conclude that u U ^ ( ( A A * ) \ N , D ) ,   where N : = G D A .   Since, by our above assumption, mes ( N ) = 0 ,   it follows from Theorem  5.14 that u ω ( , A , D ) .   But on the other hand, one has ω ( z , A , D ) < 1 ε = u ( z )   for z G .   This leads to the desired contradiction. Hence, the proof of (iii) is finished.
Theorem 5.19. Let D C   be an open set and A   a linearly measurable subset of D   such that D   is locally rectifiable on A   and mes ( A ) > 0 .   For any 0 ε < 1 ,   let D ε : = { z D : ω ( z , A , D ) < 1 ε } .   1) For any subset N   of D   such that mes ( N ) = 0 ,   let
U ε ( A , N , D ) : = { u S ( D ε ) : u 1 and sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u ( z ) 0 , ζ ( A A * ) \ N } .  
Then U ε ( A , N , D ) = U ε ( A , , D ) .   2) Define the ”harmonic measure of the ε   -level set” ω ε ( , A , D )   as ω ε ( z , A , D ) : = { sup u U ε ( A , , D ) u ( z ) , z D ε 0 , z A * .   Then ω ε ( z , A , D ) = ω ( z , A , D ) 1 ε , z D ε A * .  
  • Proof. Clearly, by definition, U ε ( A , , D ) U ε ( A , N , D ) .   To prove the converse inclusion, fix an arbitrary u U ε ( A , N , D ) .   Consider the following function u ^ ( z ) : = { max { ( 1 ε ) u ( z ) , ω ( z , A , D ) } , z D ε ω ( z , A , D ) , z D \ D ε .   Then u ^ S ( D )   and u ^ 1 .   Moreover, in virtue of Part 2) of Proposition  5.9 , we have that A * ( D ε ) D .   Consequently, for every ζ ( A A * ) \ N ,  
    (5.15) sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u ^ ( z ) max { sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u ( z ) , sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) ω ( z , A , D ) } .
    Observe that the first term in the latter line of ( 5.15 ) is equal to 0   because u U ε ( A , N , D ) .   In addition, in virtue of Part 2) of Proposition  5.9 , the second term in the latter line of ( 5.15 ) is also equal to 0 .   Hence, u ^ U ^ ( ( A A * ) \ N , D ) .   Consequently, by Theorem  5.14 , u ^ ω ( , A , D ) .   In particular, one has
    u ( z ) ω ( z , A , D ) 1 ε , z D , u U ε ( A , N , D ) . (5.16)
    On the other hand, as an immediate consequence of Part 2) of Proposition  5.9 , we get that ω ( , A , D ) 1 ε U ε ( A , , D ) U ε ( A , N , D ) .   This, combined with ( 5.16 ), implies the desired conclusions of Part 1) and Part 2).
An immediate consequence of Theorem  5.19 is the following Two-Constant Theorem for level sets.
Corollary 5.20. Let D C   be an open set and let A   and N   be two linearly measurable subsets of D   such that D   is locally rectifiable on A ,   mes ( A ) > 0   and mes ( N ) = 0 .   Let 0 ε < 1   and put D ε : = { z D : ω ( z , A , D ) < 1 ε } .   If u S ( D ε )   satisfies u M   on D ε   and sup 0 < α < π 2 limsup z ζ , z A α ( ζ ) u ( z ) m ,   ζ ( A A * ) \ N ,   then u ( z ) m ( 1 ω ε ( z , A , D ) ) + M ω ε ( z , A , D ) .  

6 Boundary behaviour of the Gonchar–Carleman operator

Before investigating the boundary behavior of the Gonchar–Carleman operator, we first introduce the following notion and study its properties.

6.1 Angular Jordan domains

Let E   be the unit disc. We begin with the
Definition 6.1. For every closed subset F   of E   and any real number h   such that mes ( F ) > 0   and sup x , y F | x y | < h < 1 2 2 ,   the open set Ω = Ω ( F , h ) : = ζ F { z A π 4 ( ζ ) : | z | > 1 h }   is called the angular Jordan domain with base F   and height h .  
We give below a list of properties of such angular Jordan domains.
Proposition 6.2. Let Ω = Ω ( F , h )   be an angular Jordan domain.
1) Then there exist exactly two points ζ 1 , ζ 2 F   such that | ζ 1 ζ 2 | = sup x , y F | x y |   and F [ ζ 1 , ζ 2 ] ,   where [ ζ 1 , ζ 2 ]   is the (small) closed arc of E   which is oriented in the positive sense and which starts from ζ 1   and ends at ζ 2 .   2) Write the open set [ ζ 1 , ζ 2 ] \ F   as the union of disjoint open arcs [ ζ 1 , ζ 2 ] \ F = j J ( a j , b j ) ,   where ( a j , b j )   is the (small) open arc of E   which goes from a j   to b j   and which is oriented in the positive sense, and the index set J   is finite or countable.
For j J ,   we construct the isosceles triangle with the three vertices a j ,   b j   and c j   such that the base of the isosceles triangle is the segment connecting a j   to b j ,   and c j   satisfies arg ( c j a j a j ) = 3 π 4 and arg ( c j b j b j ) = 3 π 4 .   Let [ a j , c j ]   (resp. [ c j , b j ]   ) denote the segment connecting a j   to c j   (resp.
the segment connecting c j   to b j   ). Put F 0 : = F j J ( [ a j c j ] [ c j b j ] ) .   Then F 0   is a rectifiable Jordan curve starting from ζ 1   and ending at ζ 2 .   3) Let η 1   (resp. η 2   ) be the unique point in the circle B ( 0 , 1 h )   such that arg ( η 1 ζ 1 ζ 1 ) = 3 π 4 ( resp. arg ( η 2 ζ 2 ζ 2 ) = 3 π 4 )   and that | η 1 ζ 1 |   (resp. | η 2 ζ 2 |   ) is minimal. Let F 1   (resp. F 2   ) denote the segment connecting η 1   to ζ 1   (resp. the segment connecting ζ 2   to η 2   ).
Let F 3   be the (small) closed arc of the circle B ( 0 , 1 h )   which starts from η 2   and ends at η 1   and which is oriented in the negative sense.
Then Ω   is a rectifiable Jordan domain and its boundary Γ   consists of the rectifiable Jordan curve F 0 ,   two segments F 1 ,   F 2   and the closed arc F 3 .   4) For every ε ( 0 , h 4 )   define the dilatation τ ε : E E   as follows τ ε ( z ) : = ( 1 ε ) z , z E .   Put Ω ε : = τ ε ( Ω ) \ B ( 0 , ( 1 + ε ) ( 1 h ) ) ¯ .   Then Ω ε   is a rectifiable Jordan domain and its boundary Γ ε   consists of the rectifiable Jordan curve F 0 ε : = τ ε ( F 0 ) ,   a sub-segment F 1 ε   of τ ε ( F 1 ) ,   a sub-segment F 2 ε   of τ ε ( F 2 ) ,   and a closed arc F 3 ε   of B ( 0 , ( 1 + ε ) ( 1 h ) ) .   5) Consider the projection τ : E \ { 0 } E   given by τ ( z ) : = z | z | ,   z E \ { 0 } .   For every ε ( 0 , h 4 )   notice that F 0 ε F 1 ε F 2 ε = Γ ε \ B ( 0 , ( 1 + ε ) ( 1 h ) ) .   Then the two maps
F 0 ε F 1 ε F 2 ε ζ τ ( ζ ) E ,
F 3 ε ζ τ ( ζ ) E ,
are one-to-one. In addition, for any linearly measurable subset A   of Γ ε ,   mes ( A ) 10 mes ( τ ( A ) ) .   6) Ω ε Ω   as ε 0 .   7) For any closed Jordan curve C   contained in Ω   there is an ε > 0   such that C Ω ε .   8) mes ( F \ Ω E ) = 0 .  
  • Proof. All assertions are quite simple using an elementary geometric argument.
    Therefore, we leave the details of their proofs to the reader. However, we will give the proof of the fact that Ω   is a domain. This proof will clarify Definition  6.1 .
    In virtue of the condition on F   and h   given in Definition  6.1 , we see that { z A π 4 ( ζ ) : | z | > 1 h } ,   ζ E ,   is connected, and that
    { z A π 4 ( ζ ) : | z | > 1 h } { z A π 4 ( η ) : | z | > 1 h } , ζ , η E : | ζ η | < h < 1 2 2 .  
    Hence, Ω   is a domain.
Theorem 6.3. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D   with mes ( A ) > 0 .   Then, for any 0 ε < 1   and any connected component G   of D ε : = { z D : ω ( z , A , D ) < 1 ε } ,   there are an open set U C ,   a conformal mapping Φ   , and an angular Jordan domain Ω = Ω ( F , h )   such that
  • (i) U D   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains;
  • (ii) Φ   maps E   conformally onto one connected component of U D   (notice that, in virtue of (i), U D   has at most two connected components);
  • (iii) Φ ( F ) A A * G D   and Φ ( Ω ) G .  
  • Proof. We have already shown in the proof of (iii) of Theorem  5.18 that there is a sequence ( U k ) k = 1   of open sets of C   such that U k D   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains, and A k = 1 ( U k D ) ,   and mes ( A A * G D ) > 0 .   Consequently, there is an index k 0   such that
    mes ( A A * G D ( D U ) ) > 0 , (6.1)
    where U : = U k 0 .   Suppose without loss of generality that U D   is a rectifiable Jordan domain. The remaining case where U D   is the disjoint union of two rectifiable Jordan domains may be proved in the same way. Let Φ   be a conformal mapping from E   onto D U .   By Theorem  5.2 , Φ   extends to a homeomorphic map (still denoted by) Φ   from E ¯   onto D U ¯ .   Hence, (i) and (ii) are satisfied.
    In virtue of Theorem  5.2 , it follows from ( 6.1 ) that
    mes ( Φ 1 ( A A * G D ( D U ) ) ) > 0 . (6.2)
    For any m 1 ,   let
    A m : = { η E : A 2 , m ( η ) Φ 1 ( G ) } , (6.3)
    where A 2 , m ( η )   is given by formula ( 5.13 ).
    Using the Geometric fact just after ( 5.14 ), we see that A m   is closed. On the other hand, applying Proposition  5.6 leads to Φ 1 ( A A * G D ( D U ) ) m = 1 A m .   Therefore, in virtue of ( 6.2 ), there is an index m 0   such that mes ( A m 0 Φ 1 ( A A * G D ( D U ) ) ) > 0 .   Put h : = 1 2 m 0 .   By the latter estimate one may find a closed set F   contained in A m 0 Φ 1 ( A A * G D ( D U ) )   such that mes ( F ) > 0   and sup x , y F | x y | < h .   Since h = 1 2 m 0 ,   a geometric argument shows that { z A π 4 ( ζ ) : | z | > 1 h } A 2 , m 0 ( ζ ) , ζ E .   This together with ( 6.3 ) implies that Ω = Ω ( F , h ) Φ 1 ( G ) .   Hence, (iii) is verified.
    This completes the proof.
In the sequel, the following uniqueness theorem will play a vital role.
Theorem 6.4. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D   with mes ( A ) > 0 .   Let N D   with mes ( N ) = 0 .   Let 0 ε < 1   and G   a connected component of D ε : = { z D : ω ( z , A , D ) < 1 ε } .   If f O ( G )   admits the angular limit 0   at every point of ( A A * G D ) \ N ,   then f 0 .  
  • Proof. Applying Theorem  6.3 we obtain an open set U   in C ,   a conformal mapping Φ   from E   onto D U   which extends homeomorphically to E ¯ ,   and an angular Jordan domain Ω : = Ω ( F , h )   satisfying assertions (i)–(iii) listed in that theorem.
    Consider the function f Φ : Ω C .   By the hypothesis and by Proposition  5.6 , f Φ O ( Ω )   admits the angular limit 0   at a.e point in F .   Since mes ( F ) > 0 ,   Theorem  4.3 gives that f Φ 0   on Ω .   Hence, f 0   on the subdomain Φ ( Ω )   of G .   This proves f 0 .  

6.2 Main result of the section

The boundary behavior of Gonchar–Carleman operator is described below.
Theorem 6.5. Let D j   be the unit disc E   and A j   a linearly measurable subset of D j   such that mes ( A j ) > 0 ,   j = 1 , , N .   Let f   be a function defined on X : = X ( A 1 , , A N ; D 1 , , D N )   with the following properties:
  • (i) f | A   is measurable;
  • (ii) f O s ( X o ) ;  
  • (iii) there exist a constant C > 0   and N   functions f j : A C ,   j = 1 , , N ,   such that for any ( a , a ) ( A 1 × × A j 1 ) × ( A j + 1 × × A N ) ,   | f ( a , , a ) | D j < C   and f ( a , , a )   has the angular limit f j ( a , a j , a )   at a j   for a.e. a j A j ,   and f 1 = = f N = f   a.e. on A .  
Let 0 < δ 1 < 1 ,   z = ( z 2 , , z N ) ( D 2 ( A 2 A 2 * ) ) × × ( D N ( A N A N * ) )   such that j = 2 N ω ( z j , A j , D j ) < δ 1 ,   and let G   be any connected component of D 1 , δ 1 : = { z 1 D 1 : ω ( z 1 , A 1 , D 1 ) < 1 δ 1 } .   Then there is an angular Jordan domain Ω = Ω ( F , h )   such that Ω G ,   F A 1 A 1 * G D 1 ,   and the Gonchar–Carleman operator K [ f ]   (see formulas ( 4.1 )–( 4.2 )) satisfies K [ f ] ( a 1 , z ) = lim z 1 a 1 , z 1 A α ( a 1 ) K [ f ] ( z 1 , z ) , 0 < α < π 2 ,   for a.e. a 1 F .  

6.3 Preparatory results

For the proof of Theorem  6.5 we need the following results.
In the sequel, for every function f L 1 ( E , | d ζ | ) ,   let C [ f ]   denote the Cauchy integral C [ f ] ( z ) : = 1 2 π i E f ( ζ ) d ζ z ζ , z E .   For a function F : E C ,   the radial maximal function M rad F : E [ 0 , ]   is defined by ( M rad F ) ( ζ ) : = sup 0 r < 1 | F ( r ζ ) | , ζ E .   Now we are able to state the following classical result (see Theorem 6.3.1 in Rudin's book [16)
Theorem 6.6. (Korányi-Vági type theorem) There is a constant C > 0   such that (changes) E | M rad C [ f ] ( ζ ) | 2 | d ζ | C E | f ( ζ ) | 2 | d ζ |   for every f L 2 ( E , | d ζ | ) .  
We recall the definition of the Smirnov class E p ,   p > 0 ,   on rectifiable Jordan domains.
Definition 6.7. Let p > 0   and Ω   a rectifiable Jordan domain. A function f O ( Ω )   is said to belong to the Smirnov class E p ( Ω )   if there exists a sequence of rectifiable closed Jordan curves ( C n ) n = 1   in Ω ,   tending to the boundary in the sense that C n   eventually surrounds each compact subdomain of Ω ,   such that C n | f ( z ) | p | d z | M < , n 1 .  
Next, we rephrase some facts concerning the Smirnov class E p ,   p > 0   on rectifiable Jordan domains in the context of angular Jordan domains Ω ( F , h ) .  
Theorem 6.8. 1) Let Ω   be a rectifiable Jordan domain. Then, for every f E p ( Ω ) ,   p > 0 ,   f   admits the angular limit f *   a.e. on Ω .   2) Let Ω : = Ω ( F , h )   be an angular Jordan domain and let Γ : = Ω .   For any 0 < ε < h 4 ,   let Γ ε   be the rectifiable closed Jordan curve defined in Part 4) of Proposition  6.2 . Then f E p ( Ω )   if sup 0 < ε < h 4 Γ ε | f ( z ) | p | d z | < .   In addition, for an f E p ( Ω )   , p > 0 ,   it holds that Γ | f * ( z ) | p | d z | sup 0 < ε < h 4 Γ ε | f ( z ) | p | d z | .   3) Every f E 1 ( E )   has a Cauchy representation f : = C [ f * ] .   Conversely, if g L 1 ( E , | d z | )   and E z n g ( z ) d z = 0 , n = 0 , 1 , 2 , ,   then f : = C [ g ] E 1 ( E )   and g   coincides with f *   a.e. on E .  
  • Proof. For the proof of Parts 1) and 3), see [4,p.438–441. Taking into account Parts 6) and 7) of Proposition  6.2 , Part 2) also follows from the results in [4,p.438–441.
    Hence, the proof is complete.

6.4 Proof of Theorem  6.5 .

We only give the proof of this theorem in the case N = 2 .   The proof for the general case is quite similar, therefore, it is left to the interested reader.
We fix an arbitrary z 2 0 D 2 ( A 2 A 2 * )   , 0 < δ 2 < δ 1   such that ω ( z 2 0 , A 2 , D 2 ) < δ 2 ,   and an arbitrary connected component G   of D 1 , δ 1 : = { z 1 D 1 : ω ( z 1 , A 1 , D 1 ) < 1 δ 1 } .   Applying Theorem  6.3 , we may find an angular Jordan domain Ω : = Ω ( F , h ) G   such that F A A * G D 1 .   Without loss of generality we may assume that z 2 0 D 2 ,   since the case where z 2 0 A 2 A 2 *   is trivial because of (iii). In the course of the proof, the letter C   will denote a positive constant that is not necessarily the same at each step.
Applying Theorem  4.1 , we have
f ( z 1 , a 2 ) = lim M K M [ f | A 1 × { a 2 } ] ( z 1 ) , z 1 D 1 , a 2 A 2 ,
f ( a 1 , a 2 ) = lim r 1 f ( r a 1 , a 2 ) , a 1 D 1 , a 2 A 2 .
Consequently, f | D 1 × A 2   is measurable. In addition, by (iii) this function is bounded.
Therefore, for every M N   we are able to define the function K , M ( , z 2 0 ) : D 1 C ,  
K , M ( a 1 , z 2 0 ) : = 1 2 π i A 2 e M ( g 2 ( z 2 0 ) g 2 ( a 2 ) ) f ( a 1 , a 2 ) d a 2 a 2 z 2 0 , a 1 D 1 . (6.4)
Since, in virtue of (ii)–(iii), f ( a 1 , ) O ( D 2 )   and | f ( a 1 , ) | D 2 < C   for a 1 A 1 ,   it follows from Theorem  4.1 that
lim M K , M ( a 1 , z 2 0 ) = f ( a 1 , z 2 0 ) , a 1 A 1 , (6.5)
and the above convergence is uniform with respect to a 1 A 1 .   On the other hand, by ( 6.4 ) we see that K , M ( , z 2 0 )   is measurable and bounded.
In addition, for any n = 0 , 1 , 2 , ,   taking (ii) into account, we have that D 1 K , M ( a 1 , z 2 0 ) a 1 n d a 1 = 1 2 π i A 2 ( D 1 f ( a 1 , a 2 ) a 1 n d a 1 ) e M ( g 2 ( z 2 0 ) g 2 ( a 2 ) ) d a 2 a 2 z 2 0 = 0 ,   where the first equality follows from an application of Fubini's Theorem and the second one from an application of Part 3) of Theorem  6.8 to f ( , a 2 ) ,   a 2 A 2 .   Consequently, in virtue of Part 3) of Theorem  6.8 , we can extend K , M ( , z 2 0 )   to D ¯ 1   by setting
K , M ( z 1 , z 2 0 ) : = C [ K , M ( , z 2 0 ) ] ( z 1 ) = 1 2 π i D 1 K , M ( a 1 , z 2 0 ) d a 1 a 1 z 1 , z 1 D 1 . (6.6)
Then the following identity holds
lim z 1 a 1 , z 1 A α ( a 1 ) K , M ( z 1 , z 2 0 ) = K , M ( a 1 , z 2 0 ) , 0 < α < π 2 , (6.7)
for a.e. a 1 D 1 .   Now we come back to the angular Jordan domain Ω   . We keep the notation introduced in Proposition  6.2 . Put K : = K [ f ]   and K M , M : = K M , M [ f | A ]   (see formulas ( 4.1 )–( 4.2 )). For any 0 < ε < h 4   and any z 1 Γ ε ,   applying the Cauchy integral formula, we obtain
(6.8) K , M ( z 1 , z 2 0 ) K M , M ( z 1 , z 2 0 ) = 1 ( 2 π i ) 2 D 1 \ A 1 A 2 e M ( g 1 ( z 1 ) g 1 ( t 1 ) ) + M ( g 2 ( z 2 0 ) g 2 ( t 2 ) ) f ( t 1 , t 2 ) d t 1 d t 2 ( t 1 z 1 ) ( t 2 z 2 0 ) = e M ( g 1 ( z 1 ) ( 1 δ 1 ) ) D 1 p N ( t 1 ) d t 1 t 1 z 1 .
Using the choice of G   and the hypothesis on δ 1   and δ 2 ,   it can be checked that
| e M ( g 1 ( ) ( 1 δ 1 ) ) | G 1 , | p N | D 1 C e M ( δ 1 δ 2 ) . (6.9)
Therefore, recalling the projection τ : E \ { 0 } E   is defined in Part 5) of Proposition  6.2 , we estimate
(6.10) Γ ε | K , M ( z 1 , z 2 0 ) K M , M ( z 1 , z 2 0 ) | 2 | d z 1 | C Γ ε | M rad C [ p N ] ( τ ( z 1 ) ) | 2 | d z 1 | 10 C τ ( F 0 ε F 1 ε F 2 ε ) | M rad C [ p N ] ( a 1 ) | 2 | d a 1 | + 10 C τ ( F 3 ε ) | M rad C [ p N ] ( a 1 ) | 2 | d a 1 | 20 C E | M rad C [ p N ] ( a 1 ) | 2 | d a 1 | C E | p N ( a 1 ) | 2 | d a 1 | C e M ( δ 1 δ 2 ) ,
where the first estimate follows from ( 6.8 )–( 6.9 ) and the definition of the radial maximal function, the second and the third one from Part 5) of Proposition  6.2 , the fourth estimate holds by an application of Theorem  6.6 , and the last one follows from ( 6.9 ).
On the other hand, for any 0 < ε < h 4 ,  
(6.11) Γ ε | K M + 1 , M + 1 ( z 1 , z 2 0 ) K M , M ( z 1 , z 2 0 ) | 2 | d z 1 | 2 Γ ε | K M 1 ( z 1 , z 2 0 ) | 2 | d z 1 | + 2 Γ ε | K M 2 ( z 1 , z 2 0 ) | 2 | d z 1 | C e M δ 0 ,
where the latter estimate follows from the same argument as in the proof of ( 6.8 )–( 6.10 ). We recall from Part 1) of Theorem  4.2 that lim M K M , M ( z 1 , z 2 0 ) = K ( z 1 , z 2 0 ) , z 1 Γ ε ,   and that the convergence is uniform with respect to z 1 Γ ε .   This, combined with ( 6.10 )–( 6.11 ), implies that
Γ ε | K , M ( z 1 , z 2 0 ) K ( z 1 , z 2 0 ) | 2 | d z 1 | C e M ( δ 1 δ 2 ) , 0 < ε < h 4 . (6.12)
Since we have already shown that | K , M ( , z 2 0 ) | D 1 < ,   in virtue of Part 2) of Theorem  6.8 , we deduce from ( 6.12 ) that K ( , z 2 0 ) | Ω E 2 ( Ω ) .   For every a 1 D 1 ,   let K ( a 1 , z 2 0 )   denote the angular limit of K ( , z 2 0 ) | Ω   at a 1   (if the limit exists). It follows from ( 6.12 ) and Part 2) of Theorem  6.8 that
lim M Γ | K , M ( a 1 , z 2 0 ) K ( a 1 , z 2 0 ) | 2 | d a 1 | sup 0 < ε < h 4 Γ ε | K , M ( z 1 , z 2 0 ) K ( z 1 , z 2 0 ) | 2 | d z 1 | lim M C e M ( δ 1 δ 2 ) = 0 .  
This, combined with ( 6.5 ) and Part 8) of Proposition  6.2 , implies finally that K ( a 1 , z 2 0 ) = f ( a 1 , z 2 0 ) , for a.e. a 1 F .   Hence, Theorem  6.5 has been proved.  

7 Proof of Theorem A in a special case

In this section, we prove Theorem A under the following hypotheses:
D 1 , , D N   are rectifiable Jordan domains and | f | X < .   ( * )   We first prove that Theorem A under the above hypotheses may be reduced to the case when D 1 = = D N = E .   Indeed, let Φ j   be a conformal mapping from D j   onto E   which extends to a homeomorphic map (still denoted by) Φ   from D j ¯   onto E ¯   j = 1 , , N .   Consider the function f ~   given by
f ~ ( w ) : = f ( Φ 1 1 ( w 1 ) , , Φ N 1 ( w N ) ) , w = ( w 1 , , w N ) X ( Φ 1 ( A 1 ) , , Φ N ( A N ) ; E , , E ) .  
Then we deduce easily from the hypothesis on f   that f ~   satisfies the hypotheses (i)–(iii) on the cross X ( Φ 1 ( A 1 ) , , Φ N ( A N ) ; E , , E ) .   Let f ~ ^ O ( X ^ o ( Φ 1 ( A 1 ) , , Φ N ( A N ) ; E , , E ) )   denote the extension function of f ~   provided by Theorem A in the case where D 1 = = D N = E .   Using Proposition  5.6 and Part 1) of Theorem  5.2 , we see that the function f ^   defined by f ^ ( z ) : = f ~ ^ ( Φ 1 ( z 1 ) , , Φ N ( z N ) ) , z = ( z 1 , , z N ) X ^ o ,   satisfies the conclusions of Parts 1)–4). This proves the above reduction.
We give here only the proof of Theorem A for the case N = 2 .   Since the general case is quite similar and does not require any new ideas, it is therefore left to the interested reader. In summary, from now on we assume that N = 2 ,   D 1 = D 2 = E ,   and | f | X < .   ( * * )   Using hypotheses (i)–(iii) and ( * * ) ,   we are able to apply Theorem  6.5 and obtain a function K [ f ] O ( X ^ )   (see the notation in Subsection 2.3). We define the desired extension function f ^   as follows
f ^ : = K [ f ] on X ^ \ ( A 1 × A 2 ) . (7.1)
The remaining part of the proof is divided into several steps.
Step 1: Proof of the estimate | f ^ | X ^ o | f | X .   Proof of Step 1. Let z 0 = ( z 1 0 , z 2 0 )   be an arbitrary point of X ^ o .   Then we may find an δ 1 ( 0 , 1 )   such that 0 < ω ( z 2 0 , A 2 , D 2 ) < δ 1 < 1 ω ( z 1 0 , A 1 , D 1 ) .   Let G   be the connected component of D 1 , δ 1 : = { z 1 D 1 : ω ( z 1 , A 1 , D 1 ) < 1 δ 1 }   that contains z 1 0   . By Theorem  6.3 we may find an angular Jordan domain Ω : = Ω ( F , h )   contained in G   such that F A 1 A 1 * G D 1 .   In addition, for every M N ,   applying Theorem  6.5 to the function f M ,   we obtain the function K [ f M ] O ( X ^ o )   with the following property
lim z 1 a 1 , z 1 A α ( a 1 ) K [ f M ] ( z 1 , z 2 0 ) = f ( a 1 , z 2 0 ) M = lim z 1 a 1 , z 1 A α ( a 1 ) ( K [ f ] ( z 1 , z 2 0 ) ) M , 0 < α < π 2 ,  
for a.e. a 1 F .   Consequently, arguing as in the proof of Theorem  6.4 , we get K [ f M ] ( z 0 ) = ( K [ f ] ( z 0 ) ) M , M N ,   Since z 0 X ^ o   is arbitrarily chosen, it follows from the latter identity that
K [ f M ] ( z ) = ( K [ f ] ( z ) ) M , M N , z X ^ o . (7.2)
Now we are able to conclude the proof in the same way as in [6,p.23. More precisely, taking into account ( 7.1 )–( 7.2 ) and Part 2) of Theorem  4.2 , one gets that | f ^ M ( z ) | | K [ f M ] ( z ) | C 0 | f | X M D 1 ) dist ( z 2 , D 2 ) ( 1 e ( 1 ω ( z ) ) ) , z = ( z 1 , z 2 ) X ^ o .   Extracting the M   th roots of both sides and letting M   tend to ,   the desired estimate of Step 1 follows.   Step 2: We shall prove that f ^   is the unique function O ( X ^ o )   which verifies Property 1).
Proof of Step 2. First we show that the function f ^   given by ( 7.1 ) satisfies Property 1). Without loss of generality, it suffices to prove that there is a subset A ~ 2   of A 2 A 2 *   such that mes ( A ~ 2 ) = mes ( A 2 )   and f ^   admits the angular limit f   at every point of D 1 × A ~ 2 .   For any a 1 A 1   put A a 1 : = { a 2 A 2 : f ( a 1 , ) has an angular limit at a 2 } .   By hypothesis (iii), we have mes ( A a 1 ) = mes ( A 2 ) ,   a 1 A 1 .   Consequently, applying Fubini's Theorem, we obtain that A 1 mes ( A a 1 ) | d a 1 | = mes ( A 1 ) mes ( A 2 ) = A 2 mes ( { a 1 A 1 : a 2 A a 1 } ) | d a 2 | .   Hence,
mes ( { a 1 A 1 : a 2 A a 1 } ) = mes ( A 1 ) for a.e. a 2 A 2 . (7.3)
The same reasoning also gives that
mes ( { a 1 A 1 : f ( a 1 , a 2 ) = f 1 ( a 1 , a 2 ) } ) = mes ( A 1 ) for a.e. a 2 A 2 . (7.4)
Set
(7.5) A ~ 2 : = { a 2 A 2 A 2 * : mes ( { a 1 A 1 : a 2 A a 1 } ) = mes ( A 1 ) and mes ( { a 1 A 1 : f ( a 1 , a 2 ) = f 1 ( a 1 , a 2 ) } ) = mes ( A 1 ) } .
We deduce from ( 7.3 )–( 7.5 ) that
mes ( A ~ 2 ) = mes ( A 2 ) . (7.6)
Fix an arbitrary point a 2 0 A ~ 2   and let ( z 2 n ) n = 1   be an arbitrary sequence of D 2   such that lim n z 2 n = a 2 0   and z 2 n A α ( a 2 0 )   for some fixed number 0 < α < π 2 .   Fix an arbitrary point z 1 0   of D 1   and let ( z 1 n ) n = 1   be an arbitrary sequence of D 1   such that lim n z 1 n = z 1 0 .   Clearly, we may find 0 < δ 1 < 1   such that
sup n N ω ( z 1 n , A 1 , D 1 ) < 1 δ 1 . (7.7)
Fix an δ 2   such that 0 < δ 2 < δ 1 .   Since a 2 0   is locally regular relative to A 2   and lim n z 2 n = a 2 0   and z 2 n A α ( a 2 0 ) ,   there is a sufficiently large number N 0   with
ω ( z 2 n , A 2 , D 2 ) < δ 2 , n > N 0 . (7.8)
Let G   be that connected component of the following open set D 1 , δ 1 : = { z 1 D 1 : ω ( z 1 , A 1 , D 1 ) < 1 δ 1 }   which contains z 1 0   (see ( 7.7 )). Applying Theorem  6.3 , we may find an angular Jordan domain Ω : = Ω ( F , h )   contained in G   such that F A 1 A 1 * G D 1 .   Let V   be a rectifiable Jordan domain with Ω V G ,   z 2 0 V   , and U   some neighborhood of the base F   of Ω   with V U = Ω U .   In virtue of ( 7.8 ) and of the fact that V G D 1 , δ 1 ,   we obtain that
V × { z 2 n } X ^ o , n > N 0 . (7.9)
Consequently, Theorem  6.5 yields that for any n > N 0 ,  
f ( a 1 , z 2 n ) = lim z 1 a 1 , z 1 A α ( a 1 ) f ^ ( z 1 , z 2 n ) , 0 < α < π 2 , (7.10)
for a.e. a 1 F .   Next, for any n > N 0   let
F n : = { a 1 F : a 2 0 A a 1 and f ( a 1 , z 2 n ) = lim z 1 a 1 , z 1 A α ( a 1 ) f ^ ( z 1 , z 2 n ) } , F 0 : = n = N 0 + 1 F n .
It follows from ( 7.5 ), ( 7.10 ) and the fact that a 2 0 A ~ 2   that mes ( F n ) = mes ( F ) ,   n > N 0 .   Hence
mes ( F 0 ) = mes ( F ) > 0 . (7.11)
In virtue of ( 7.9 ), consider the following holomorphic functions on V  
h n ( t ) : = f ^ ( t , z 2 n ) and h 0 ( t ) : = f ( t , a 2 0 ) , t V , n > N 0 . (7.12)
Since we have already shown in Step I that | h n | V | f | X < ,   n > N 0   or n = 0 ,   applying Part 1) of Theorem  6.8 , we may find a subset Δ   of F 0   with mes ( Δ ) = mes ( F 0 ) > 0   such that h n ,   n > N 0   (resp. h 0   ) admits the angular limit f 1 ( t , z 2 n )   (resp. f 1 ( t , a 2 0 ) )   at t Δ .   Observe that by ( 7.5 ) and the fact that a 2 0 A ~ 2   we have that lim n f 1 ( t , z 2 n ) = f 1 ( t , a 2 0 ) = f ( t , a 2 0 ) for a.e. t Δ .   Using this and ( 7.12 ), we are able to apply Lemma  4.4 to the sequence ( h n ) n = 0 .   Consequently, one gets lim n f ^ ( z 1 n , z 2 n ) = f ( z 1 0 , a 2 0 ) .   This shows that f ^   admits the angular limit f   at every point of D 1 × A ~ 2 .   Hence, f ^   satisfies Property 1).
In order to complete Step 2 we need to show the uniqueness of f ^   . To do this, let f ^ ^ O ( X ^ o )   be a function with the following property: There is a subset A ~ ~ j   of A j A j *   ( j = 1 , 2 )   such that mes ( A j \ A ~ ~ j ) = 0   and f ^ ^   admits the angular limit f   at every point of ( A ~ ~ 1 × D 2 ) ( D 1 × A ~ ~ 2 ) .   Fix an arbitrary point z 0 = ( z 1 0 , z 2 0 ) X ^ o .   Let G   be the connected component containing z 1 0   of the following open set { z 1 D 1 : ω ( z 1 , A 1 , D 1 ) < 1 ω ( z 2 0 , A 2 , D 2 ) } .   We deduce from the property of f ^   and f ^ ^   that both holomorphic functions f ^ ( , z 2 0 ) | G   and f ^ ^ ( , z 2 0 ) | G   admit the angular limit f ( , z 2 0 )   at every point of A ~ 1 A ~ ~ 1 G D 1 .   Consequently, applying Theorem  6.4 yields that f ^ ( , z 2 0 ) = f ^ ^ ( , z 2 0 )   on G .   Hence, f ^ ( z 0 ) = f ^ ^ ( z 0 ) .   Since z 0 X ^ o   is arbitrary, the uniqueness of f ^   is established. This completes Step 2.   Step 3: Proof of Part 2).
Proof of Step 3. Fix ( z 1 0 , z 2 0 ) X ^ o .   For every a 2 A 2   we have | f ( a 1 , a 2 ) | | f | A 1 × A 2 , a 1 A 1 , and | f ( z 1 , a 2 ) | | f | X , z 1 D 1 .   Therefore, the Two-Constant Theorem (see Theorem 2.2 in [14) implies that
| f ( z 1 , a 2 ) | | f | A 1 × A 2 1 ω ( z 1 , A 1 , D 1 ) | f | X ω ( z 1 , A 1 , D 1 ) , z 1 D 1 , a 2 A 2 . (7.13)
Let δ : = ω ( z 1 0 , A 1 , D 1 )   and consider the δ   -level set D 2 , δ : = { z 2 D 2 : ω ( z 2 , A 2 , D 2 ) < 1 δ } .   Clearly, z 2 0 D 2 , δ .   Recall from Step 2 that A ~ 2 A 2 A 2 * ,   mes ( ( A 2 A 2 * ) \ A ~ 2 ) = 0 ,   and
f ( z 1 0 , a 2 ) = lim z 2 a 2 , z 2 A α ( a 2 ) f ^ ( z 1 0 , z 2 ) , 0 < α < π 2 , a 2 A ~ 2 . (7.14)
Consider the following function h : D 2 , δ A ~ 2 C   defined by
h ( t ) : = { f ^ ( z 1 0 , t ) , t D 2 , δ f ( z 1 0 , t ) , t A ~ 2 . (7.15)
Clearly, h | D 2 , δ O ( D 2 , δ ) .   On the other hand, in virtue of ( 7.15 ) and the result of Step 1, we have
| h | D 2 , δ | f ^ | X ^ 0 | f | X < . (7.16)
In addition, applying Corollary  5.20 and taking ( 7.14 )–( 7.15 ) into account yields | h ( t ) | | h | A ~ 2 1 ω δ ( t , A 1 , D 1 ) | h | D 2 , δ ω δ ( t , A 1 , D 1 ) , t D 2 , δ ,   where, by Theorem  5.19 , ω δ ( t , A 2 , D 2 ) = ω ( t , A 2 , D 2 ) 1 ω ( z 1 0 , A 1 , D 1 ) .   This, combined with ( 7.13 )–( 7.16 ), implies that
| f ^ ( z 1 0 , z 2 0 ) | = | h ( z 2 0 ) | | f | A 1 × A 2 1 ω ( z 1 0 , A 1 , D 1 ) ω ( z 2 0 , A 2 , D 2 ) | f | X ω ( z 1 0 , A 1 , D 1 ) + ω ( z 2 0 , D 2 , A 2 ) .
Hence Part 2) for the point ( z 1 0 , z 2 0 )   is proved.   Step 4: Proof of Part 3).
Proof of Step 4. Let ( a 1 0 , z 2 0 ) A 1 * × D 2   be such that the following limit exists λ : = lim ( a 1 , z 2 ) ( a 1 0 , z 2 0 ) , ( a 1 , z 2 ) A 1 × D 2 f ( a 1 , z 2 ) .   We like to show that f ^   admits the angular limit λ   at ( a 1 0 , z 2 0 ) .   In virtue of assumption ( * * ) ,   we may suppose (without loss of generality) that | f | X < 1 .   For any 0 < ε < 1 2 ,   we may find an open neighborhood A a 1 0   of a 1 0   in A 1   and a positive number r > 0   such that B ( z 2 0 , r ) D 2   and
| f ( a 1 , z 2 ) λ | < ε 2 , a 1 A a 1 0 , | z 2 z 2 0 | r . (7.17)
Put
δ : = sup z 2 B ( z 2 0 , r ) ¯ ω ( z 2 , A 2 , D 2 ) . (7.18)
Since a 1 0 A 1 * ,   it follows from Part 2) of Proposition  5.9 that mes ( A a 1 0 ) > 0 .   Next, consider the level set D 1 , δ : = { z 1 D 1 : ω ( z 1 , A a 1 0 , D 1 ) < 1 δ } .   In virtue of ( 7.18 ), we can define
h ( t , z 2 ) : = f ^ ( t , z 2 ) λ , t D 1 , δ , z 2 B ( z 2 0 , r ) ¯ . (7.19)
Clearly,
| h | D 1 , δ 2 | f ^ | X ^ 0 = 2 | f | X = 2 . (7.20)
By ( 7.19 ) and using the result of Step 2, we know that for every z 2 B ( z 2 0 , r )   the holomorphic function h ( , z 2 ) | D 1 , δ   admits the angular limit f ( a 1 , z 2 ) λ   at a 1   for a 1 A ~ 1 A a 1 0 ,   where A ~ 1   is given in Step 2. Consequently, applying Corollary  5.20 and taking ( 7.17 ) and ( 7.20 ) into account, we see that | h ( t , z 2 ) | < ε 2 ( 1 ω δ ( t , A a 1 0 , D 1 ) ) 2 ω δ ( t , A a 1 0 , D 1 ) , t D 1 , δ .   Let 0 < α < π 2 .   In virtue of Theorem  5.19 and the hypothesis that a 1 0 A 1 * ,   we deduce that lim t a 1 0 , t A α ( a 1 0 ) ω δ ( t , A a 1 0 , D 1 ) = 0 .   Consequently, there is an r α > 0   such that | f ( z 1 , z 2 ) λ | = | h ( z 1 , z 2 ) | < ε , z 1 A α ( a 1 0 ) { | z 1 a 1 0 | < r α } , z 2 B ( z 2 0 , r ) .   This completes the above assertion.
Similarly, we can prove that f ^   admits the angular limit lim ( z 1 , a 2 ) ( z 1 0 , a 2 0 ) , ( z 1 , a 2 ) D 1 × A 2 f ( z 1 , a 2 )   at any point ( z 1 0 , a 2 0 ) ,   if the latter limit exists. Hence the proof of Step 4 (i.e. Part 3)) is finished.   Step 5: Proof of Part 4).
Proof of Step 5. Let ( a 1 0 , a 2 0 ) A 1 * × A 2 *   be such that the following limit exists λ : = lim ( a 1 , a 2 ) ( a 1 0 , a 2 0 ) , ( a 1 , a 2 ) A 1 × A 2 f ( a 1 , a 2 ) .   We like to show that f ^   admits the angular limit λ   at ( a 1 0 , a 2 0 ) .   Suppose without loss of generality that | f | X < 1 ,   and fix an arbitrary 0 < ε < 1 2 .   Since ( a 1 0 , a 2 0 ) A 1 * × A 2 * ,   we may find an open neighborhood A a 1 0   of a 1 0   in A 1   (resp.
an open neighborhood A a 2 0   of a 2 0   in A 2 )   such that
| f ( a 1 , a 2 ) λ | < ε 2 , a 1 A a 1 0 , a 2 A a 2 0 . (7.21)
By Part 2) of Proposition  5.9 , one gets mes ( A a 1 0 ) > 0   and mes ( A a 2 0 ) > 0 .   Consider the function
h ( z 1 , z 2 ) : = f ( z 1 , z 2 ) λ , ( z 1 , z 2 ) X ( A a 1 0 , A a 2 0 ; D 1 , D 2 ) . (7.22)
Clearly,
| h ( z 1 , z 2 ) | 2 , ( z 1 , z 2 ) X ( A a 1 0 , A a 2 0 ; D 1 , D 2 ) . (7.23)
Applying the results of Steps 1–3 to h ,   we obtain the function
h ^ : = K [ h ] on X ^ 0 ( A a 1 0 , A a 2 0 ; D 1 , D 2 ) . (7.24)
so that h ^   admits the angular limit h   on ( A ~ a 1 0 × D 2 ) ( D 1 × A ~ a 1 0 ) ,   where A ~ a 1 0 ,   A ~ a 1 0   are given by Step 2. Clearly, X ^ ( A a 1 0 , A a 2 0 ; D 1 , D 2 ) X ^ ( A 1 , A 2 ; D 1 , D 2 ) .   Consequently, arguing as in Step 1 and taking into account the above mentioned angular limit of h ^ ,   we conclude that h ^ = f ^ λ on X ^ ( A a 1 0 , A a 2 0 ; D 1 , D 2 ) .   Consequently, applying Step 3 and taking into account ( 7.21 )–( 7.24 ) and the inequality | f | X < 1 ,   we see that
| f ^ ( z 1 , z 2 ) λ | = | h ^ ( z 1 , z 2 ) | | h | A a 1 0 × A a 2 0 1 ω ( z 1 , A a 1 0 , D 1 ) ω ( z 2 , A a 2 0 , D 2 ) ( 2 | f | X ) ω ( z 1 , A a 1 0 , D 1 ) + ω ( z 2 , A a 2 0 , D 2 ) < ε 2 ( 1 ω ( z 1 , A a 1 0 , D 1 ) ω ( z 2 , A a 2 0 , D 2 ) ) 2 ω ( z 1 , A a 1 0 , D 1 ) + ω ( z 2 , A a 2 0 , D 2 ) .
Therefore, for all z = ( z 1 , z 2 ) X ^ ( A a 1 0 , A a 2 0 ; D 1 , D 2 )   satisfying
ω ( z 1 , A a 1 0 , D 1 ) + ω ( z 2 , A a 2 0 , D 2 ) < 1 3 , (7.25)
we deduce from the latter estimate that
| f ^ ( z ) λ | < ε . (7.26)
Since a 1 0   (resp. a 2 0   ) is locally regular relative to A a 1 0   (resp. A a 2 0   ), there is an r α > 0   such that ( 7.25 ) is fulfilled for z = ( z 1 , z 2 ) ( A α ( a 1 0 ) { | z 1 a 1 0 | < r α } ) × ( A α ( a 2 0 ) { | z 2 a 2 0 | < r α } ) .   This, combined with ( 7.26 ), completes the proof. Hence Step 5 (i.e. Part 4)) is finished.   Step 6: Proof of Part 5).
Proof of Step 6. In virtue of Step 5, we only need to show that f ^   admits the angular limit f   on ( A 1 * × D 2 ) ( D 1 × A 2 * ) .   To do this let ( a 1 0 , z 2 0 ) A 1 * × D 2   and choose an arbitrary 0 < ε < 1 .   Fix a compact subset K   of A 2 A 2 *   such that mes ( K ) > 0   and a sufficiently large N   such that
ε N ( 1 ω ( z 2 0 , A 2 , D 2 ) ) ( 2 | f | X ) ω ( z 2 0 , A 2 , D 2 ) < ε 2 . (7.27)
Using the hypothesis that f   can be extended to a continuous function on A 1 * × A 2 * ,   we may find an open neighborhood A a 1 0   of a 1 0   in A 1 *   such that
| f ( a 1 , a 2 ) f ( a 1 0 , a 2 ) | ε N , a 1 A a 1 0 A a 1 0 * , a 2 K . (7.28)
On the other hand,
| f ( a 1 , z 2 ) f ( a 1 0 , z 2 0 ) | 2 | f | X < , a 1 A a 1 0 A a 1 0 * , z 2 D 2 . (7.29)
For a 1 A a 1 0 A a 1 0 * ,   applying the Two-Constant Theorem to the function f ( a 1 , ) f ( a 1 0 , ) O ( D 2 )   and taking ( 7.27 )–( 7.29 ) into account, we deduce that
| f ( a 1 , z 2 0 ) f ( a 1 0 , z 2 0 ) | ε N ( 1 ω ( z 2 0 , K , D 2 ) ) ( 2 | f | X ) ω ( z 2 0 , K , D 2 ) < ε 2 . (7.30)
Since f ( a 1 , ) | D 2   is a bounded holomorphic function for a 1 A 1 ,   there is an open neighborhood V   of z 2 0   such that | f ( a 1 , z 2 ) f ( a 1 , z 2 0 ) | < ε 2 , a 1 A 1 , z 2 V .   This, combined with ( 7.30 ), implies that
| f ( a 1 , z 2 ) f ( a 1 0 , z 2 0 ) | | f ( a 1 , z 2 0 ) f ( a 1 0 , z 2 0 ) | + | f ( a 1 , z 2 ) f ( a 1 , z 2 0 ) |
< ε 2 + ε 2 = ε , a 1 A a 1 0 , z 2 V .
Therefore, f   is continuous at ( a 1 0 , z 2 0 ) .   Consequently, we conclude, by Step 5, that f ^   admits the angular limit f ( a 1 0 , z 2 0 )   at ( a 1 0 , z 2 0 ) .   Similarly, we may also show that f ^   admits the angular limit f ( z 1 0 , a 2 0 )   at every point ( z 1 0 , a 2 0 ) D 1 × A 2 * .   This completes the proof of the last step.  

8 Preparatory results

We first develop some auxiliary results. This preparation will enable us to generalize the results of section 7 to the general case considered in Theorem A.
Definition 8.1. Let Ω C n   be an open set and let A Ω .   Define
h A , Ω : = sup { u : u P S ( Ω ) , u 1 on Ω , u 0 on A } ,
h A , Ω * ( z ) : = limsup w z h A , Ω ( w ) , z Ω .
The function h A , Ω *   is called the plurisubharmonic measure of A   relative to Ω .  
Proposition 8.2. Let D C   be an open set which is locally rectifiable on a linearly measurable subset A   of D   with mes ( A ) > 0 .   Let { a j } j J   be a finite or countable subset of A   with the following properties:
  • (i) For any j J ,   there is an open neighborhood U j   of a j   such that D U j   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains;
  • (ii) A j J U j .  
For any 0 < δ < 1 2 ,   define
U j , δ : = { z D U j : ω ( z , A U j , D U j ) < δ } , j J , A δ : = j J U j , δ , D δ : = { z D : ω ( z , A , D ) < 1 δ } .
Then: 1) A A * A δ D   and A δ D 1 δ D δ ;   2) ω ( z , A , D ) δ h A δ , D * ( z ) ω ( z , A , D ) ,   z D .  
  • Proof. To prove Part 1), let a A A *   and fix an j J   such that a U j .   Then lim z a , z A α ( a ) ω ( z , A U j , D U j ) = 0 , 0 < α < π 2 .   Consequently, for every 0 < α < π 2 ,   there is an open neighborhood V α U j   of a   such that ω ( z , A U j , D U j ) < δ , z A α ( a ) V α .   This proves A A * A δ D .   To prove the second assertion of Part 1), one invokes the Subordination Principle and obtains for z U j , δ ,  
    ω ( z , A , D ) ω ( z , A U j , D U j ) < δ < 1 δ . (8.1)
    Hence, z D 1 δ .   This implies that A δ D 1 δ .   In addition, since 0 < δ < 1 2 ,   it follows that D 1 δ D δ .   Hence, Part 1) is proved.
    We turn to Part 2). Since A δ   is an open set and, by Part 1), A A * A δ D ,   it follows from Definitions  5.17 ,  8.1 that h A δ , D * ( z ) ω ( z , A A * , D ) , z D .   Hence, in virtue of Proposition  5.9 and Theorem  5.14 , it follows that h A δ , D * ( z ) ω ( z , A , D ) , z D ,   which proves the second estimate of Part 2).
    To complete Part 2), let z A δ .   Choose j J   such that z U j , δ .   We deduce from ( 8.1 ) that ω ( z , A , D ) δ 0 .   Hence, ω ( z , A , D ) δ 0 , z A δ .   On the other hand, ω ( z , A , D ) δ < 1 ,   z D .   Consequently, the first estimate of Part 2) follows. The proof of the lemma is finished.
The next result gives a nice geometric property of the interior of a wedge.
Proposition 8.3. Let N N ,   N 2 ,   D j C   a domain and A j D j   such that D j   is locally rectifiable on A j ,   mes ( A j ) > 0 ,   j = 1 , , N .   Put X ^ o : = X ^ o ( A 1 , , A N ; D 1 , , D N ) .   Then X ^ o   is a domain.
  • Proof. We only give the proof for the case N = 2 .   The general case when N > 2   can be proved in exactly the same way. Fix arbitrary points z 0 = ( z 1 0 , z 2 0 )   and w 0 = ( w 1 0 , w 2 0 )   of X ^ o .   Let
    ε : = 1 2 ( min { 1 ω ( z 1 0 , A 1 , D 1 ) ω ( z 2 0 , A 2 , D 2 ) , 1 ω ( w 1 0 , A 1 , D 1 ) ω ( w 2 0 , A 2 , D 2 ) } ) . (8.2)
    For j , k { 1 , 2 }   , j k ,   let D j , z j 0   (resp. D j , w j 0   ) be the connected component containing z j 0   (resp. w j 0   ) of the following open set
    { z j D j : ω ( z j , A j , D j ) < 1 ω ( z k 0 , A k , D k ) ε } , ( resp { z j D j : ω ( z j , A j , D j ) < 1 ω ( w k 0 , A k , D k ) ε } ) , (8.3)
    where ε   is given in ( 8.2 ).
    By Theorem  5.18 (iii), for j { 1 , 2 } ,   we may find a point a j A j A j * D j , z j 0 D j   (resp. b j A j A j * D j , w j 0 D j   ).
    For j { 1 , 2 } ,   fix a connected open neighborhood V z j 0 D j   of z j 0   and an open neighborhood V a j D j   of a j   such that
    V a j A π 4 ( a j ) is a domain , ω ( z j , A j , D j ) < ω ( z j 0 , A k , D j ) + ε , z j V z j 0 , ω ( z j , A j , D j ) < ε , z j V a j A π 4 ( a j ) . (8.4)
    Combining ( 8.2 )–( 8.4 ), we see that V z 1 0 × V z 2 0   and V z 1 0 × ( V a 2 A π 4 ( a 2 ) )   are in the same connected component of X ^ o .   The same argument also shows that V z 1 0 × ( V a 2 A π 4 ( a 2 ) )   and ( V a 1 A π 4 ( a 1 ) ) × ( V a 2 A π 4 ( a 2 ) )   are in the same connected component of X ^ o .   Hence, V z 1 0 × V z 2 0   and ( V a 1 A π 4 ( a 1 ) ) × ( V a 2 A π 4 ( a 2 ) )   are in the same connected component of X ^ o .   We apply the same argument as above to w 0 .   Consequently, one may find, for j { 1 , 2 } ,   a connected open neighborhood V w j 0 D j   of w j 0   and an open neighborhood V b j D j   of b j   such that
    V b j A π 4 ( b j ) is a domain , ω ( z j , A j , D j ) < ω ( w j 0 , A k , D j ) + ε , z j V w j 0 , ω ( z j , A j , D j ) < ε , z j V b j A π 4 ( b j ) , (8.5)
    and V w 1 0 × V w 2 0   and ( V b 1 A π 4 ( b 1 ) ) × ( V b 2 A π 4 ( b 2 ) )   are in the same connected component of X ^ o .   Recall that D 1   and D 2   are domains. Then in virtue of ( 8.4 ) and ( 8.5 ), one may find, for j { 1 , 2 } ,   a Jordan curve γ j : [ 0 , 1 ] D j   such that
    γ j ( 0 ) V a j A π 4 ( a j ) and γ j ( 1 ) V b j A π 4 ( b j ) . (8.6)
    Let
    δ : = 1 2 ( min { 1 sup t [ 0 , 1 ] ω ( γ 1 ( t ) , A 1 , D 1 ) , 1 sup t [ 0 , 1 ] ω ( γ 2 ( t ) , A 2 , D 2 ) } ) . (8.7)
    Next, we shrink V b 2   such that ω ( z 2 , A 2 , D 2 ) < δ , z 2 V b 2 .   This, combined with ( 8.4 )–( 8.7 ), implies that ( V a 1 A π 4 ( a 1 ) ) × ( V b 2 A π 4 ( b 2 ) )   and ( V b 1 A π 4 ( b 1 ) ) × ( V b 2 A π 4 ( b 2 ) )   are in the same connected component of X ^ o .   Similarly, by shrinking V a 1   if necessary, we see that ( V a 1 A π 4 ( a 1 ) ) × ( V a 2 A π 4 ( a 2 ) )   and ( V a 1 A π 4 ( a 1 ) ) × ( V b 2 A π 4 ( b 2 ) )   are in the same connected component of X ^ o .   Hence, ( V a 1 A π 4 ( a 1 ) ) × ( V a 2 A π 4 ( a 2 ) )   and ( V b 1 A π 4 ( b 1 ) ) × ( V b 2 A π 4 ( b 2 ) )   are in the same connected component of X ^ o .   In summary, combining the above fact, ( 8.5 ) and the similar conclusion for z 0 ,   we have shown that V z 1 0 × V z 2 0   and V w 1 0 × V w 2 0   are in the same connected component of X ^ o .   Consequently, z 0   and w 0   are in the same connected component of X ^ o ,   which completes the proof.
The main ingredient in the proof of Theorem A is the following mixed cross theorem.
Theorem 8.4. Let D C n   be a domain of holomorphy, Ω C   an open set, A D ,   and B   a linearly measurable subset of Ω .   Assume that A = k = 1 A k   with A k   locally pluriregular5 compact subsets of D ,   A k A k + 1 ,   k 1 .   In addition, Ω   is locally rectifiable on B   with mes ( B ) > 0 ,   and B B * .   For 0 δ < 1   put G : = { w Ω : ω ( w , B , Ω ) < 1 δ } .   Let X : = X ( A , B ; D , G )   , X o : = X o ( A , B ; D , G ) ,   and (using the notation of Theorem  5.19 ) X ^ o = X ^ o ( A , B ; D , G ) : = { ( z , w ) D × G : h A , D * ( z ) + ω δ ( w , B , Ω ) < 1 } .   Let f : X C   be such that
  • (i) f O s ( X o ) ;  
  • (ii) f   is measurable and bounded on X ;  
  • (iii) for any z A ,   lim w η , w A α ( η ) f ( z , w ) = f ( z , η ) , η B , 0 < α < π 2 .  
Then there is a unique function f ^ O ( X ^ o )   such that f ^ = f   on A × G   and lim z z 0 , w η 0 , w A α ( η 0 ) f ^ ( z , w ) = f ( z 0 , η 0 ) , 0 < α < π 2 ,   for every z 0 D   and η 0 B .   Moreover, | f ^ | X ^ o | f | X .  
  • Proof. First one proves the existence and uniqueness of f ^ .   To do this we argue as in the proof of Theorem 3.5.1 in [7. For the sake of completeness, we give here a sketchy proof. Fix an f : X C   which satisfies (i)–(iii) above.
    Step I: Reduction to the case where D   is strongly pseudoconvex and A   is a locally pluriregular compact subset of D .   One proceeds as in the first and second step in that proof. More precisely, since D   is a domain of holomorphy, we may find an exhaustion sequence ( D k ) k = 1   of relatively compact, strongly pseudoconvex subdomains D k   of D   with A k D k D .   By reduction assumption, for each k   there exists an f ^ k O ( X ^ o ( A k , B ; D k , G ) )   such that f ^ k   admit the angular limit f | X ( A k , B ; D k , G )   on X ( A k , B ; D k , G ) .   We would like to show that f ^ k + 1 = f ^ k   on X ^ o ( A k , B ; D k , G ) .   Indeed, fix an arbitrary k 0 1   and an arbitrary point ( z 0 , w 0 ) X ^ o ( A k 0 , B ; D k 0 , G ) .   Let k N   such that k k 0 .   Let D   be the connected component containing z 0   of the following open set { z D : h A k 0 , D k 0 * ( z ) < 1 ω δ ( w 0 , B , Ω ) } .   Observe that both functions f ^ k 0 ( , w 0 ) | D   and f ^ k ( , w 0 ) | D   are holomorphic and f ^ k ( z , w 0 ) = f k ( z , w 0 ) = f ^ k 0 ( z , w 0 ) , z A k D .   Since A k D   is non-pluripolar, we deduce that f ^ k 0 ( , w 0 ) | D = f ^ k ( , w 0 ) | D .   Hence, f ^ k 0 ( z 0 , w 0 ) = f ^ k ( z 0 , w 0 ) ,   which proves the above assertion.
    On the other hand, by Proposition  5.15 one gets X ^ o ( A k , B ; D k , G ) X ^ o   as k .   Therefore, we may glue f ^ k   together to obtain a function f ^ O ( X ^ o )   such that f ^   admits the angular limit f   on X   and f ^ = f   on A × G .   The uniqueness of such an extension f ^   can be proved using the argument given in the previous paragraph.
    This completes Step I. Step II: The case where D   is strongly pseudoconvex and A   is a locally pluriregular compact subset of D .   Suppose without loss of generality that | f | X < 1 .   The key observation is that we are still able to apply the classical method of doubly orthogonal bases of Bergman type.
    Next one observes that Lemma 3.5.10 in [7is still valid in the present context.
    Look at Step 3 in that proof. In the sequel, we will use the notation from [7. Let μ : = μ A , D ,   H 0 : = L h 2 ( D ) ,   H 1 : =   the closure of H 0 | A   in L 2 ( A , μ )   and let ( b k ) k = 1 H 0   be the basis from Lemma 3.5.10 in [7, ν k : = b k H 0 , k N ,   with the following property:
    k = 1 ν k ε < , ε > 0 . (8.8)
    For any w B ,   we have f ( , w ) H 0   and f ( , w ) | A H 1 .   Hence
    f ( , w ) = k = 1 c k ( w ) b k , (8.9)
    where
    c k ( w ) = 1 ν k 2 D f ( z , w ) b k ( z ) ¯ d Λ 2 n ( z ) = A f ( z , w ) b k ( z ) ¯ d μ ( z ) , k N . (8.10)
    Taking the hypotheses (i)–(iii) into account and applying Lebesgue's Dominated Convergence Theorem, we see that the formula
    c k ^ ( w ) : = A f ( z , w ) b k ( z ) ¯ d μ ( z ) , w G B , k N ; (8.11)
    defines a bounded function which is holomorphic in G .   Moreover, by (iii) and ( 8.10 )–( 8.11 ) it follows that
    lim w η , w A α ( η ) c k ^ ( w ) = c k ^ ( η ) = c k ( η ) , η B , 0 < α < π 2 . (8.12)
    Observe, as in [7and using ( 8.10 )–( 8.12 ), that we obtain the following estimates
    log | c k ^ ( w ) | log ν k log μ ( A ) log ν k , w G , k N ,
    limsup w η , w A α ( η ) log | c k ^ ( η ) | log ν k log Λ 2 n ( D ) log ν k 1 , η B , 0 < α < π 2 , k N ,
    where Λ 2 n ( D )   is the volume of D   with respect to the Lebesgue measure of C n .   This shows that for any ε > 0 ,   there is a sufficiently large N   such that for all k N ,  
    log | c k ^ | log ν k ω δ ( , B , Ω ) + ε 1 on G . (8.13)
    Take a compact set K D   and let 1 > α > max K h A , D * .   Choose an ε > 0   so small that α + 2 ε < 1 .   Consider the open set G K : = { w G : ω δ ( , B , Ω ) < 1 α 2 ε } .   By ( 8.13 ) there is a constant C ( K )   such that
    | c k ^ | G k C ( K ) ν k ω δ ( , B , Ω ) + ε 1 C ( K ) ν k α ε , k 1 . (8.14)
    Now we wish to show that
    k = 1 c k ^ ( w ) b k ( z ) (8.15)
    converges locally uniformly in X ^ o .   Indeed, by ( 8.8 ), ( 8.14 ), and Lemma 3.5.10 in [7,
    k = 1 | c k ^ | G K | b k | K k = 1 C ( K ) ν k α ε C ( K , α ) ν k α C ( K ) C ( K , α ) k = 1 ν k ε < , (8.16)
    which gives the normal convergence on K × G K .   Since the compact set K   and ε > 0   are arbitrary, the series in ( 8.15 ) converges uniformly on compact subsets of X ^ o .   Let f ^   denote this limit function in ( 8.15 ).
    Fix z 0 D   and η 0 B .   We choose a compact K   of D   as above. Moreover, we may suppose without loss of generality that K   contains a neighborhood of z 0 .   Let ε 0 > 0 .   In virtue of ( 8.16 ), there is an N 0   such that
    k = N 0 + 1 | c k ^ | G K | b k | K < ε 0 2 . (8.17)
    On the other hand, in virtue of ( 8.9 )–( 8.12 ), we may find, for any 0 < α < π 2 ,   an open neighborhood V α   of η 0   such that | k = 1 N 0 c k ^ ( w ) b k ( z ) k = 1 N 0 c k ( η 0 ) b k ( z ) | < ε 0 2 , z K , w A α ( η 0 ) V α .   This, combined with ( 8.15 ) and ( 8.17 ), implies that limsup z z 0 , w η 0 , w A α ( η 0 ) | f ^ ( z , w ) f ( z 0 , η 0 ) | < ε 0 , 0 < α < π 2 .   Since ε 0 > 0   and ( z 0 , η 0 ) D × B   can be arbitrarily chosen, we conclude that lim z z 0 , w η 0 , w A α ( η 0 ) f ^ ( z , w ) = f ( z 0 , η 0 ) , ( z 0 , η 0 ) D × B , 0 < α < π 2 .   To complete Step II, it remains to show that f ^ = f   on A × G .   To do this, fix an arbitrary ( z 0 , w 0 ) A × G .   Let G   be the connected component of G   containing w 0   .
    Recall that G = { w Ω : ω ( w , B , Ω ) < 1 δ } .   Then observe that both functions f ^ ( z 0 , ) | G   and f ( z 0 , ) | G   admit the same angular limit f   on B G Ω .   Consequently, applying Theorem  6.4 yields that f ^ ( z 0 , ) | G = f ( z 0 , ) | G .   Hence, f ^ ( z 0 , w 0 ) = f ( z 0 , w 0 ) ,   which proves the above assertion.
    This completes the proof of Step II. It remains to prove the estimate | f ^ | X ^ | f | X .   In order to reach a contradiction assume that there is a point z 0 X ^ o   such that | f ^ ( z 0 ) | > | f | X .   Put α : = f ^ ( z 0 )   and consider the function
    g ( z ) : = 1 f ( z ) α , z X . (8.18)
    Using the above assumption, it can be checked that g   satisfies hypotheses (i)–(iii) of Theorem  8.4 . Hence applying the first assertion of the theorem, there is exactly one function g ^ O ( X ^ o )   with g ^ = g   on A × G .   Therefore, by ( 8.18 ) we have on A × G :   g ( f α ) 1 .   Thus g ^ ( f ^ α ) 1   on X ^ o .   In particular, 0 = g ^ ( z 0 ) ( f ^ ( z 0 ) α ) = 1 ,   a contradiction. Hence the inequality | f ^ | X ^ | f | X   is proved.
Finally, we conclude this section with two uniqueness results.
Proposition 8.5. Let D j C   be an open set and A j   a linearly measurable subset of D j   such that D j   is locally rectifiable on A j ,   A j A j * ,   mes ( A j ) > 0 ,   j = 1 , 2 .   Let D ~ 1 C   be a domain, D 1 D ~ 1 ,   and let A ~ 1   be a linearly measurable subset of D ~ 1   such that D ~ 1   is locally rectifiable on A ~ 1 ,   and mes ( A ~ 1 ) > 0 .   Put
X ^ o : = X ^ o ( A 1 , A 2 ; D 1 , D 2 ) ,
X ~ ^ o : = X ^ o ( A ~ 1 , A 2 ; D ~ 1 , D 2 , ) .
Let f ^ O ( X ^ o ) ,   f ~ ^ O ( X ~ ^ o ) ,   and z 1 0 D 1 D ~ 1   be such that both f ^   and f ~ ^   admit the same angular limit at ( z 1 0 , a 2 )   for a.e. a 2 A 2 .   Then f ^ ( z ) = f ~ ^ ( z )   for every z = ( z 1 0 , z 2 ) X ^ o X ~ ^ o .  
  • Proof. Fix an arbitrary z 2 0 D 2   such that z 0 : = ( z 1 0 , z 2 0 ) X ^ o X ~ ^ o .   Let ε 0 : = 1 4 ( 1 max { ω ( z 1 0 , A 1 , D 1 ) , ω ( z 1 0 , A ~ 1 , D ~ 1 ) } ) ,   and put D 2 , ε 0 : = { z 2 D 2 : ω ( z 2 , A 2 , D 2 ) < 1 ε 0 } .   Applying Theorem  6.4 to f ^ | D 2 , ε 0   and f ~ ^ | D 2 , ε 0 ,   it follows that f ~ ^ ( z 1 0 , z 2 0 ) = f ^ ( z 1 0 , z 2 0 ) .   Hence, the proof is finished.
Now we are able to prove the uniqueness stated in Theorem A.
Corollary 8.6. We keep the hypotheses and the notation of Theorem A. Then there is at most one function f ^ O ( X ^ o )   which satisfies Property 1) of Theorem A.
  • Proof. It follows immediately from Proposition  8.5 .

9 Proof of Theorem A

We mainly consider the case N = 2 .   In this case we proceed by four steps. Recall that by Corollary  8.6 , the function f ^   satisfying Part 1) is uniquely determined (if exists).
In the first two steps we mainly prove the following assertion:
There are a function f ^ O ( X ^ o )   and a subset A ~ j   of A j A j *   ( j = 1 , 2 )   such that mes ( A j \ A ~ j ) = 0   and f ^   admits the angular limit f   at every point of ( A ~ 1 × D 2 ) ( D 1 × A ~ 2 ) .   ( * * * )   In fact, in the first two steps we always assume that | f | X < .   Using this and the previous observation and taking ( * * * )   for granted, we conclude the proof of Steps 1 and 2 below in exactly the same way as we did in Section 7 starting from Step 2 of that section.
Step 1: Proof of Theorem A for the case where N = 2 ,   | f | X < ,   and D 2   is a rectifiable Jordan domain.
Proof of Step 1. In virtue of Proposition  8.2 , let { a 1 j } j J 1   be a finite or countable subset of A 1   with the following properties:
  •   For any j J 1 ,   there is an open neighborhood U 1 j   of a 1 j   such that D 1 U 1 j   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains (according to the type of a 1 j   );
  •   A 1 j J 1 U 1 j .  
For any 0 < δ < 1 2 ,   define
U 1 j , δ : = { z 1 D 1 U 1 j : ω ( z 1 , A 1 U 1 j , D 1 U 1 j ) < δ } , j J 1 , A 1 , δ : = j J 1 U 1 j , δ , D 2 , δ : = { z 2 D 2 : ω ( z 2 , A 2 , D 2 ) < 1 δ } .
Moreover, for every j J 1   let
X j : = X ( ( D 1 U 1 j ) A 1 , A 2 ; D 1 U 1 j , D 2 ) ) , X j ^ o : = X ^ o ( ( D 1 U 1 j ) A 1 , A 2 ; D 1 U 1 j , D 2 ) ) , f ~ j : = f | X j . (9.1)
Using the hypotheses on f ,   we conclude that f ~ j ,   j J 1 ,   satisfies (i)–(iii) of Theorem A. Moreover, | f ~ j | X j | f | X < .   Since D 2   is a rectifiable Jordan domain and D 1 U 1 j ,   j J 1 ,   is either a rectifiable Jordan domain or the disjoint union of two rectifiable Jordan domains, we are able to apply the result of Section 7 to f ~ j .   Consequently, we obtain, for j J 1 ,   a unique function f ^ 1 j O ( X j ^ o ) ,   a subset A 1 j   of A 1 ,   a subset A 2 j   of A 2   such that
mes ( A 1 \ A 1 j ) = mes ( A 2 \ A 2 j ) = 0 , f ^ 1 j admits the angular limit f on ( ( ( D 1 U 1 j ) A 1 j ) × D 2 ) ( D 1 × A 2 j ) . (9.2)
Put
A ~ 1 : = j J 1 A 1 j and A ~ 2 : = j J 2 A 2 j , X δ : = X ( A 1 , δ , A ~ 2 ; D 1 , D 2 , δ ) , X δ ^ o : = X ^ o ( A 1 , δ , A ~ 2 ; D 1 , D 2 , δ ) . (9.3)
In virtue of Proposition  8.5 , we are able to collect the family ( f ^ 1 j | U 1 j , δ × D 2 , δ ) j J 1   in order to obtain a function f ~ ~ δ O ( A 1 , δ × D 2 , δ ) .   Next, consider the function f ~ δ : X δ C   given by
f ~ δ : = { f ~ ~ δ , on A 1 , δ × D 2 , δ f , on D 1 × A ~ 2 . (9.4)
In virtue of ( 9.1 )–( 9.4 ), we deduce that
mes ( A 1 \ A ~ 1 ) = mes ( A 2 \ A ~ 2 ) = 0 , (9.5)
and
lim z 1 z 1 0 , z 2 a 2 0 , z 2 A α ( a 2 0 ) f ~ δ ( z ) = f ( z 1 0 , a 2 0 ) , 0 < α < π 2 , z 1 0 D 1 , a 2 0 A ~ 2 , lim z 1 a 1 0 , z 1 A α ( a 1 0 ) , z 2 a 2 0 f ~ δ ( z ) = f ( a 1 0 , z 2 0 ) , 0 < α < π 2 , a 1 0 A ~ 1 , z 2 0 D 2 , δ . (9.6)
In virtue of ( 9.4 )–( 9.6 ), f ~ δ   satisfies the hypotheses (i)–(iii) of Theorem  8.4 . Applying this theorem to f ~ δ ,   we obtain, for every 0 < δ < 1 2 ,   a function f ^ δ O ( X δ ^ o ) .   In virtue of ( 9.6 ), we see that
f ^ δ = f ~ δ on A 1 , δ × D 2 , δ , lim z 1 z 1 0 , z 2 a 2 0 , z 2 A α ( a 2 0 ) f ^ δ ( z ) = f ( z 1 0 , a 2 0 ) , 0 < α < π 2 , z 1 0 D 1 , a 2 0 A ~ 2 , lim z 1 a 1 0 , z 1 A α ( a 1 0 ) , z 2 a 2 0 f ^ δ ( z ) = f ( a 1 0 , z 2 0 ) , 0 < α < π 2 , a 1 0 A ~ 1 , z 2 0 D 2 , δ . (9.7)
We are now in a position to define the desired extension function f ^ .   Indeed, one glues ( f ^ δ ) 0 < δ < 1 2   together to obtain f ^   in the following way
f ^ : = lim δ 0 f ^ δ on X ^ o = X ^ o ( A 1 , A 2 ; D 1 , D 2 ) . (9.8)
Now one has to check that the limit ( 9.8 ) exists and possesses all the required properties. This will be an immediate consequence of the following
Lemma 9.1. For any point z X ^ o   put
δ z : = 1 ω ( z 1 , A 1 , D 1 ) ω ( z 2 , A 2 , D 2 ) 2 . (9.9)
Then f ^ ( z ) = f ^ δ ( z )   for all 0 < δ δ z .  
Proof of Lemma  9.1 . Fix an arbitrary point z 0 = ( z 1 0 , z 2 0 ) X ^ o ( A 1 , A 2 ; D 1 , D 2 )   and let δ 0 : = δ z 0 .   Let 0 < δ δ 0 .   Then, by Proposition  8.2 , we see that ω ( z 2 0 , A ~ 2 , D 2 ) < 1 δ 0   and
h A 1 , δ , D 1 * ( z 1 0 ) + ω δ 0 ( z 2 0 , A ~ 2 , D 2 ) ω ( z 1 0 , A 1 , D 1 ) + ω ( z 2 0 , A ~ 2 , D 2 ) 1 δ 0
ω ( z 1 0 , A 1 , D 1 ) + ω ( z 2 0 , A 2 , D 2 ) 1 δ 0 < 1 ,
where the latter estimate follows from formula ( 9.9 ). Consequently,
z 0 X ^ o ( A 1 , δ , A ~ 2 ; D 1 , D 2 , δ 0 ) . (9.10)
On the other hand, using Part 1) of Proposition  8.2 , it is clear that
X ^ o ( A 1 , δ , A ~ 2 ; D 1 , D 2 , δ 0 ) X ^ o ( A 1 , δ , A ~ 2 ; D 1 , D 2 , δ ) X ^ o ( A 1 , δ 0 , A ~ 2 ; D 1 , D 2 , δ 0 ) . (9.11)
Moreover, in virtue of ( 9.4 ) and ( 9.7 ), we have
f ^ δ = f ~ ~ δ = f ^ δ 0 on A 1 , δ × D 2 , δ 0 . (9.12)
Next, let G   be the connected component containing z 1 0   of the following open set { z 1 D 1 : h A 1 , δ , D 1 * ( z 1 ) < 1 ω δ 0 ( z 2 0 , A 2 , D 2 ) }   Observe that, in virtue of ( 9.10 )–( 9.11 ), both functions f ^ δ | G   and f ^ δ 0 | G   are holomorphic and G A 1 , δ   is a nonempty open set. Therefore, we deduce from ( 9.12 ) that f ^ δ = f ^ δ 0   on G .   Hence, f ^ δ ( z 0 ) = f ^ δ 0 ( z 0 ) ,   which completes the proof of the lemma.   We complete the proof as follows. An immediate consequence of Lemma  9.1 is that f ^ O ( X ^ o ) .   Next, we apply Lemma  9.1 and make use of ( 9.4 )–( 9.9 ) and of the fact that X δ ^ o X ^ o   as δ 0 .   Consequently, we conclude that f ^   satisfies ( * * * ) .   Taking into account the remark made at the beginning of this section, the proof of the theorem in Step 1 is finished.   Step 2: Proof of Theorem A for the case N = 2   and | f | X < .   Proof of Step 2. We proceed using Step 1 in exactly the same way as we proved Step 1 using the result of Section 7. Hence, Step 2 is finished.   Step 3: Proof of Theorem A for the case when N = 2   and A 1 ,   A 2   are compacts.
Proof of Step 3. Recall from the hypothesis (i) of Theorem A that f   is locally bounded. Since A 1 ,   A 2   are compact, a compactness argument shows that there are a positive number M 0   and an open neighborhood U 1   (resp. U 2   ) of A 1   (resp. A 2   ) such that
| f | < M 0 on ( [ U 1 ( D 1 A 1 ) ] × A 2 ) ( A 1 × [ U 2 ( D 2 A 2 ) ] ) . (9.13)
For j { 1 , 2 } ,   fix a sequence ( D ~ j , k ) k = 1   of subdomains of D j   such that
D ~ j , k D j and D ~ j , k D j as k . (9.14)
Consider the sequence ( D j , k ) k = 1   of open sets of D j ,   the sequence of crosses