Differential Forms Orthogonal to Holomorphic Functions or Forms, and their Properties
TRANSLATIONS OF MATHEMATICAL MONOGRAPHS
VOLUME 56
Differential Forms Orthogonal to Holomorphic Functions or Forms, and their Properties by
L. A. Aizenberg Sh. A. Dautov
American Mathematical Society
Providence
Rhode Island
) WIxDEPEHL1HAJIbHbIE DOPMbI,
OPTOFOHAJIbHbIE I'OJIOMOP bHbIM 4DYHKLIH5IM HJIH fiOPMAM,
H HX CBOFICTBA
A. A. Af43EHEEPF H III. A. 2I,AYTOB H3AATEJIbCTBO «HAYKA» CI'L&HPCKOE OT L EJIEHHE HOBOCHEHPCK 1975
Translated from the Russian by R. R. Simha Translation edited by Lev J. Leifman 1980 Mathematics Subject Classification. Primary 32A25, 32FI5; Secondary 32A35, 32A45.
ABSTRACT. This book is devoted to the description of exterior differential forms orthogonal to holomorphic forms of degree n - p, 0 c p c n (in particular, to holomorphic functions if p = n) with respect to integration over the boundary of a bounded domain D in C". The MartinelliBochner-Koppelman formula, which is an integral representation of exterior differential forms, is given, and the characteristic properties of the trace of a holomorphic function on the boundary are studied. The question of representation and multiplication of distributions lying in 61 '(R2"- 1) is discussed with the aid of a-closed forms of type (n, n - 1) with harmonic coefficients.
Library of Congress Cataloging in Publication Data
Aizenberg, Lev Abramovich, 1937Differential form orthogonal to holomorphic functions or forms, and their properties. (Translations of mathematical monographs; v. 56) Translation of. Differenisial'nye formy, ortogonal'nye golomorfnym funktsiiam Hi formam, i ikh svoistva. Bibliography: p. Includes index.
1. Holomorphic functions. 2. Exterior forms. 3. Differential forms. I. Dautov, Sh. A. (Shamil' Abdullovich) II. Title. III. Series. QA331.A46313 1983 515.9
ISBN 0-8218-4508-X ISSN 0065-9282
Copyright © 1983 by the American Mathematical Society
TABLE OF CONTENTS Preface to the American Edition
vii
Preface
ix
Introduction
1
CHAPTER I. Integral representation of exterior differential forms and its immediate consequences 1. The Martinelli-Bochner-Koppelman formula 2. Theorems on the saltus of forms
5 5
14
3. Characterization of the trace of a holomorphic form on the boundary of a domain 4. Some cases of the solvability of the ca-problem
28
CHAPTER H. Forms orthogonal to holomorphic forms 5. Polynomials orthogonal to holomorphic functions
37
31
37
6. Forms orthogonal to holomorphic forms: the case of strictly pseudo-convex domains 7. The general case 8. Converse theorems
45 47
49
CHAPTER III. Properties of c?-closed forms of type (p, n -1) 9. The theorems of Runge and Morera 10.. The first Cousin problem, separation of singularities, and domains of existence 11. Theorems of approximation on compact sets
v
53 53 57
60
CONTENTS
vi
CHAPTER N. Some applications 12. Generalization of the theorems of Hartogs and F. and M. Riesz 13. On the general form of integral representations of holomorphic functions 14. Representation of distributions in D'(R2n-1) by a-closed exterior differential forms of type (n, n - 1) 15. Multiplication of distributions in D'(R2n-1)
67
Brief historical survey and open problems for Chapters I-N
85
CHAPTER V. Integral properties characterizing a-closed differential forms and holomorphic functions 16. A characteristic property of a-closed forms and forms of class B 17. Holomorphy of continuous functions representable by the Martinelli-Bochner integral; criteria for the holomorphy of integrals of the Martinelli-Bochner type 18. The traces of holomorphic functions on the Shilov boundary of a circular domain 19. Computation of an integral of Martinelli-Bochner type for the case of the ball 20. Differential boundary conditions for the holomorphy of functions
CHAPTER VI. Forms orthogonal to holomorphic forms. Weighted formula for solving the a-equation, and applications 21. Forms orthogonal to holomorphic forms 22. Generalization of Theorem 8.1 23. Weighted formula for solving the a-problem in strictly convex domains and zeros of functions of the Nevanlinna-Dzhrbashyan class
CHAPTER VII. Representation and multiplication of distributions in higher dimensions 24. Harmonic representation of distributions 25. The product of distributions and its properties 26. Examples of products of distributions
67 71
76 82
89
89
94 107 111
116
125 125
130
132
137
137 141 144
Supplement to the Brief Historical Survey
151
Bibliography
153
Subject index
163
Index of symbols
165
PREFACE TO THE AMERICAN EDITION In the six years that have elapsed since this book appeared in the USSR, many new results have been obtained in this field of multidimensional complex analysis. These results are presented in a supplement (Chapter V-VII), written
by the authors especially for the American edition. The results of A. M. Kytmanov have made the greatest impact on the contents of the supplement.
He has also written Chapter VII of the supplement at the request of the authors. We take this opportunity to thank him for this work; we also thank him and S. G. Myslivech for help in preparing the manuscript of the supplement.
Vii
PREFACE In this book we consider the problem of characterizing the exterior differen-
tial forms which are orthogonal to holomorphic functions (or forms) in a domain D C C" with respect to integration over the boundary, and some related questions. We give a detailed account of the derivation of the BochnerMartinelli-Koppelman integral representation of exterior differential forms,
which was obtained recently (1967) but has already found many important applications. A complete proof of this representation has not previously been available in our* literature. We study the properties of 8-closed forms of type (p, n - 1), 0 < p n - 1, which turn out to be the duals (with respect to the orthogonality mentioned above) to holomorphic functions (or forms) in several complex variables, and resemble holomorphic functions of one complex variable in their properties. At the end of the book, we give some applications, in particular to the problem of multiplying distributions, and also a brief historical survey and a discussion of open problems. We hope that this little 'book will be useful to mathematicians and theoretical physicists interested in several complex variables. The greater part of the results expounded below were obtained by us during the years 1970-73. They were reported in seminars at Krasnoyarsk, Moscow
State University, Urals State University (Sverdlovsk), and the Institute of Mathematics in the Siberian Division of the Academy of Sciences of the USSR. We thank the participants of these seminars for useful discussions. Above all we are grateful to G. M. Khenkin and V. P. Palamodov for valuable remarks.
' Editor's note. The authors mean "Russian". ix
INTRODUCTION A large number of results in the theory of holomorphic functions of one complex variable depend on the following, already classical, assertion: Let D be a bounded domain in the plane with smooth boundary aD, and rp(z) a continuous function on )D. Then
faDf(z)q,(z) dz = 0
(0.1)
for all functions f(z) continuous in the closed domain D and holomorphic in D if
and only if q)(z) can be extended holomorphically into D (see, for example, Mushelishvili [1], §29). Briefly, holomorphic functions, and only these, are orthogonal to holomorphic functions.
By attaching the dz in (0.1) to f(z) or to 4p(z), the same result can be reformulated in the following way, which is more convenient for our exposi-
a continuous (on aD) exterior differential form() a of type (p, 0) is orthogonal under integration on aD to all forms of type (1 - p, 0) which are tion:
continuous in D and holomorphic in D, if and only if there exists a form y of type
(p, 0) holomorphic in D and continuous in D, such that a = y JaD. Here p = 0 or 1.
The analogous problem is of great interest in the theory of functions of several complex variables:
What forms a, defined on the boundary of a domain D in C" with smooth boundary, are orthogonal to holomorphic forms .t (in particular to holomorphic functions) in the sense that
f naA t=0 a
for all forms holomorphic on D (or holomorphic in D and continuous on D)? (') In the sequel, we shall simply write "form" instead of "exterior differential form". 1
2
INTRODUCTION
Kohn and Rossi [1] have considered the more general problem of describing forms which are orthogonal to all a-closed forms of type (n - p, q). They have
shown that, if q > 0, and if at each point of aD the Levi form has q + 1 positive eigenvalues or n - q negative ones, then a C°° form a of type (p, n - q - 1) on 8D is orthogonal (in the sense of (0.2)) to all 8-closed C°° forms of type (n - p, q) on D if and only if there exists a C°° a-closed form y
of type (p, n - q - 1) on D such that y 6i = a However this solution does not include the case of interest to us, since holomorphic forms are a-closed forms of type (p, 0).
It will be proved (Chapter II) that a-closed forms of type (p, n - 1), 0 < p < n, are the duals,' with respect, to integration over the boundary of the domain (in the sense of (0.2)), of holomorphic forms of type (n - p, 0); in particular, for p = n, they are the duals of holomorphic functions. The proof makes essential use of the Martinelli-Bochner-Koppelman integral representation of exterior differential forms, as also the "barrier function" of Khenkin for strictly pseudoconvex domains. To state the result in a concrete way: forms which are orthogonal to all holomorphic forms on the closure of a strictly pseudoconvex domain must be extendable to the interior as a-closed forms, or (what amounts to the same in the case of strictly pseudoconvex domains) as a-exact forms of type (p, n - 1). On the other hand, it will be established in Chapter III that the a-closed forms of type (p, n - 1), 0 < p < n, are the analogs for C" of the holomorphic functions of one complex variable, i.e. they retain several of the important
properties of these functions which e.g. the holomorphic functions of n variables, n > 1, do not possess. Chapter I is devoted to tools that will be used and results related to the main problem. In particular, we present an elementary derivation of the MartinelliBochner-Koppelman integral representation of forms, generalizing the Martinelli-Bochner integral representation for holomorphic functions on the
one hand, and the Cauchy-Green formula for smooth functions in the case n = I on the other. Also we give a characterization of the trace of a holomorphic function on the boundary of a domain in C". In Chapter IV, some applications are indicated: we generalize the classical theorems of Hartogs and the Rieszes, describe the general form of integral representations of holomorphic functions, and construct the Martinelli-Bochner representation of a distribution lying on 6 '(RZ"- I ), by means of which we introduce a new definition for the product of distributions. In the text itself, we shall refer only to such original works as are necessary for understanding the book. All references are listed at the end of the book. After Chapters IV and VII the reader will also find a brief historical survey; in
INTRODUCTION
3
the former some unresolved questions (as of 1975) are indicated (a few of which are subsequently solved in Chapters V-VII, written in 1981 for the American edition).
We shall employ the following notation. C" is the space of n complex variables, whose points are denoted by z, , z °, °, etc. If z = (z1,. .. , z") then For z, E C" we define (z, )= and i=
IZI= (Z,Z),
ta,Z)
--Z
Jt
-,
fy"z"
Z
B(z°,r)= (z EC":Iz-z°I
IIIII=i,+ +i",
Z1-zip
az;' ... 84-aZ,' ... 7,' denotes summation over J = (j,,... ,j,,) with I ' j, < ... < jP < n. Also, Adze,, dz, =.dzj, A dz = dz, A . Adz", dz[J] = dz;, A ... Adzi,_p,
jk:k= 1,...,p,
i, <
< i"-P,
l= 1,...,n-p.
The symbol [ ] will often be employed in the above sense; e.g. z[k] _ (Z,,...,Zk-11 Zk+I+...,Z"). Let M and N be sets in C". Then aM is the boundary of M, and we define
p(z,M)=;EMIR-zl,
M'= (z EM: p(z,aM)>r),
diamM= sup r,zEM
CM denotes the complement of M in C". M is the closure of M, int M is the interior of M, and M C N means that M is compact and contained in int N. Let F be a closed set and Sl an open set in C". For m = 0, 1, ... , oo,
0 < A < 1, 0 < r < 2n, and 0
For A = 0 we shall omit the index, and regard C(p,q)(F) = q' )(F) as a space with the topology of uniform convergence of all coefficients together with ail derivatives of order < m on F.
INTRODUCTION
4
C(p,q)(SZ) denotes the projective limit of the classes C(P:q)(F), F C 2, with the projective limit topology. If
a = J'a,Jdz, A dz1 E I,J then
8a = ' !,J k= 1
al dzk A dz, A dz j, k
Z(P,4)(SZ) _ (a e C(P.q)(S2): 8a = 0)
(for m = 0, differentiation is to be understood as that in the space of distributions). Here again, if m = 0 we shall omit the index, and write e.g. ZAP q)(7) Z(p.q)(S2). Similarly we shall drop the indicies p and q if p = q = 0. For all a E C(P.q)(S2) we set supp a = {z E 9: a(z) 0},(2) and
AP(1) = Z(P,0)(.l) = Z(-P'0)(2)1 ACP(SZ) = Z(P,0)(S) = AP(SZ) n
C(P,o)(2).
AP(K) denotes the inductive limit of the spaces AP(S2), SZ 3 K, where K is a compact set in C". If a E C(P,q)(S2) and p E C°°(C") with supp p C St, then we shall suppose that yea is extended by zero to CS2, so that yea E C(P.q)(C"). The index z (or ) under a domain of integration signifies that z (respectively is the variable of integration. We shall say that D is a domain with boundary of class Cm," if D = {z E C": p(z) < 0}, where p E Cm,(C") and grad p :Y60 on D. If m = 1 and A = 0, D is said to have smooth boundary.
(2) The bar denotes closure.
CHAPTER I
INTEGRAL REPRESENTATION OF EXTERIOR DIFFERENTIAL FORMS AND ITS IMMEDIATE CONSEQUENCES
§1. The Martinelli-Bochner-Koppelman formula
1°. The definition of the determinant of a numerical matrix uses only the operations of addition and multiplication; hence we can form the determinant of a matrix whose elements lie in an arbitrary ring. A determinant of order n is the algebraic sum ... all
a12
a21
a22
I an,
an2
.
.
...
aln
azn
l)`a,,1....
.a.
,
ann I
where er is the parity of the permutation I = (i1,. .. , in ). Thus the determinant is expanded according to the usual rule, the place where a factor occurs in the summand being determined by the column to which the factor belongs.
In connection with the expansion of a determinant, it is convenient to introduce the following notation. Let 0',. . . , 0 m be n-dimensional vectors. Then (
1
is the determinant of order n = v +
m)
+v,, whose first v1 columns are 6',
the next v2 columns are 02, and so on.
We state some properties(') of determinants which will be needed subsequently. (1) These properties are trivial from the point of view of algebra; however for some of them we have not been able to rind appropriate references, and have taken the liberty of listing them for the convenience of the reader. 5
I. INTEGRAL REPRESENTATION OF FORMS
6
PROPERTY 1. If each element of the kth row (or column) of a determinant has been expressed as the sum of two terms, then the determinant is equal to the sum of two determinants having the same rows (columns) as the given determinant except the kth, which consists of the first and second summands respectively. PROPERTY 2. If two rows of a determinant are interchanged, then the determinant changes sign.
PROPERTY 3. If all the elements of a certain column commute with all the elements of an adjacent column, then the interchange of these two columns changes the sign of the determinant; if they anticommute,(2) then the determinant is unaffected by the interchange.
PROPERTY 4. If all the elements of a column (or row) are multiplied by an element which commutes with all the elements of the determinant, then the determinant also gets multiplied by this element. PROPERTY 5. If all the elements of the first column are multiplied on the left by a certain element, then the determinant also gets multiplied on the left by the same element.
For the remaining properties which we state, we assume that the entries of the determinant lie in the ring of double forms (see de Rham [1], Chapter II, §7). Observe that Property 4 yields PROPERTY 4'. If all the elements of a column (or row) are multiplied by the same function, then the determinant gets multiplied by this function. PROPERTY 6. If the rows of the determinant are linearly dependent,(3) then the determinant vanishes.
The proofs are the same as for the usual determinants. For forms, and hence also for double forms, we have the following assertion: If the ai, i = 1,...,m, are forms of class C', of degree qi in z, then
a:(at A ... /dam) M
=
(_1)Q l+...+ a.a,
n... Aa;-tA BZaiA ai+tA ...Aa,,,.
i=1
(22) By anticommutation we mean the relation ab = -ba. (3) The rows are said to be linearly dependent if some linear combination of them with functions as coefficients vanishes identically, and the coefficients do not have a common zero.
7
§1. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
From this we get
PROPERTY 7. If the elements of the ith columns 0' of a determinant are C' double forms of degree q; in z, i = 1,. - ., n, then
n
Here and in the sequel, k 8i = (aZBl,..., zoni).
z) = (w1,...,w,,) be a vector-valued function defined on U C 2°. Let C" X Cz, with the property that (w, - z> = 1. If wj E C'(U) for all j, we let
(_1)q+P(n-q-1)(n - 1 ) WP,q(W,
z) =
(27ri)nP! (n
- p)!
D1,q,n-q-1(w, aZw, atw)
/SDP n-P(az, af). Here the subscripts z and of a indicate the group of variables with respect to
which a is acting, and az and a denote the vector forms (dz1,...,dzn) and (d1,.. . , d fin) respectively. We also set WP _1 = WP,n = 0.
The kernels of the integral representation for exterior differential forms will be the forms Wp,q(W, z), for a special choice of the vector w, namely a)/l - z 12. We therefore introduce the notation w= z) UP.q(J, Z) = W,,q(t(J, z), J, z)
Observe that W0,0(w, , z) is the Cauchy-Fantappi8 kernel; for this form we shall also write
w(w, , z) = 0,0(w, , Z). We can now state the Martinelli-Bochner formula for exterior differential forms:
Let D be a bounded domain with piecewise smooth boundary, and let -y E C(p q)(D). Then
I,',q(D,Y) - Ip,q(D,ay) - alp,q-1(D,Y) _
y(z),
z ED, z
D,
(1.1)
I. INTEGRAL REPRESENTATION OF FORMS
8
where
IP.9(D, y)(z) = f
A Up,9(J, z),
(aD)1
fy(J) A Up,9-11J, Z) Dt
IP,9-I(D, Y)(z)
The proof of (1.1) will be given in 5°. This formula is valid in greater generality. For example, we could assume that D is a finite union of domains with piecewise smooth boundary, whose closures are pairwise disjoint. Also, for q = n, (1.1) holds for any open set D and any form y e C(p.n)(D) whose coefficients are absolutely integrable in D. Indeed, let B be a ball with B C D. Since (1.1) holds for B, we have for z E B
Y(Z) = a f Y(J) A Up..-11J, z) Bt
= 5f y(J) A Up',,_,(
,
D;
z) - 5f
(D\B)1
Y(J) A Up,n-I( ' Z). (1.2)
For z e B, we may differentiate the second integral under the integral sign. Now aUp.n-1 = 0 by Lemma 1.2; hence (1.2) reduces to (1.1) for q = n, under the assumptions made on D and Y. 3°. Observe that
lZ1
za=Z.
Hence for q> I D1,9,n-q-1(t' azt' art)
, z12a=I 1
Iy
z12
AD2,9-
1
I
- Z I q D1,1.9-I.n-q-1(f
- Z, azz, azt' art)
The first determinant on the right has two identical columns; hence it vanishes, by Property 3 of determinants. Repeating this procedure q - 1 times more, we get
(2 _1)9
D1,9,n-q-1(t' azt, art) =
I
ZI2q+
D1,9,n-q-1(J - Z, a=z,
aft).
9
§l. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
Proceeding analogously with the last n - q - 1 columns, we get Di,q,n-q-1(t' 5z1, ait ) (_1)q
-Z
_
((_
I2n Dl,q.n-q- I\f _'F) azZ,
(-1)g9! (n - 9 - z lzn
- 1)! I, t+ a(I, k)(k \J - ik)dijdf [I, k], k==1
(1.3)
k(iI
where I= (i1,...,iq) is a multi-index, the prime over the summation sign indicates that the summation is over increasing multi-indices, k a I means that
k is different from i1,...,iq' and the constant a(I, k) is determined by the equality
dik A dil A di[I, k] = a(I, k)di. Since
Dp n-v(az,4) =p!(n -
J
we deduce from (1.3) that U
-
Z)
(_1)p(n-q-1) (n
- 1)!
(2lTi)n X
n
E/ G a(I, k)a(J) I,J k=I
kJ
k
(1.4)
I - zlzzn df
Here a(J) is determined by
dzjAdi[J] =a(J)dz. For the proof of (1.1), we need one more form of expression for U0,q. Let
Vq = (-1)'(n - 1)! 2' (27T1)n
I
1
a(I,
k)(_1)k-Id(,
[I, k] A dr[k]dir;
k=I
then Uo.q -
1 - n at
Z
fn-2 A Vq.
(1.5)
4°. Throughout this subsection, the letter w (possibly with superscripts) z) = z),...,wnQ, z)), where z)
denotes a vector-valued function
EUC(C XCZ)\((,z):=z),and(w,-z)=1.
I. INTEGRAL REPRESENTATION OF FORMS
10
LEMMA 1.1. The determinant D I__ I W
I, a-zw 2'...' a -Zwr, a - wi+1'.
.
.,a-t.wn
remains unaltered if the vector wl is replaced by any other w defined in the same domain.
PROOF. Subtract from the given determinant the new determinant Dl
l
w, azw 2,...,
W,
I+1, atwn
By Property 1, r D1....,1(w' - w, 5Zw2,...15Zw/, atw/+I,...,arwn).
(azw',-z)_
Since (w'
we see by Prop-
erty 6 that this last determinant vanishes. LEMMA 1.2. If w is of class C2 and 0 < q < n, then
z) = (-1)p+g52Wp,q-I(w, , z). In particular, atWp 0 =
(1.6)
0.
PROOF. For 1 < q < n, we have by Property 7 azDl.q-I.n-q(n', a1w, atw) = Dq,n-q(a=w' arw)
+ (-1)q-'(n - q)DI,q-1.I,n-q-I(w, azw, aZat.w, atw). (1.7) For 0 < q < n - 1, the same property yields a1.D1,q,n-q-1(w' aZw' a3'w) _ (-1)gDq,n-q(azw, ai'w)
+gD1,q-1,I.n-q-1(w' azw, atazw, atw).
(1.8)
By Property 6, the first determinant in each of (1.7) and (1.8) vanishes. For q = n in (1.7), and for q = 0 in (1.8), the second term vanishes; hence from the definition of Wp,n-I and Wp.n we see that (1.6) holds for q = 0 and n.
For 1 < q < n - 1, the second determinants in (1.7) and (1.8) are equal to each other, since aZkw = ka2w. Here azatw is regarded as a double form. Therefore
5j(n -
q)Dl,q,n-q-1(w' aZw,
arw)]
_ z[(_l)'
'gDl,q-l,n-q(w' atw, a5x')].
Multiplying this equality by
(n - 1)! (2iri)nq! (n we obtain (1.6).
(-1)pcn-q-1)
- q)!p!(n -p)!
Dp
n p az'
11
§1. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
LEMMA 1.3. If wl and w2 are vector-valued functions of class C2 in ,
WP,((w2, , z)
-
W,o(wl,
, z) = arµo n
then
1
P! (n
P)!
XD,,,,-P(az, ar), where µo
(n - 0! (2
l 0 D1,l,j,n-i-2(wI , w2,
atw2).
In particular, w(w2, J, z)
- w(wl, J, z) = arll'o A
The proof is a verification, using Properties 3 and 7 and Lemma 1.1. Similar considerations are used in the proof of the next lemma. LEMMA 1.4. If U C CF X Cz is such that there exists a vector-valued function
z) E C2(U) holomorphic in z for fixed and satisfying (u, - z)= 1, then for all w E C2(U) WP,q(w, , z) _ (azµq-I + arµq) A
p! (np)!
DP n-P(az, a0' (1.9)
where Aq and µq are forms which are linear combinations of forms of the type
DI,l,q,r,n-q-r-2(w, u, azw, aru, arw).
(1.10)
PROOF. We first show that
Dl,q,n-q-1(w' azw' aru) = azan-q-1
(1.11)
and that for 0 < r _ n - q - 2 there exist constants Cr such that Di,q,r,n-q-r-I(w, azw, aru, 8rw) - CrDI.q,r+l,n-q-r-2(w, azw, aru, arw) = azaq- I + arp rq,
(1.12)
where aq and P are forms differing from (1.10) by a constant factor. Since u, and hence aru, is holomorphic in z, by Property 7 we have azDi,l,q-I,r,n-q-r-1(u, w, azw, aru, arw)
= Dl,q,r,n-q-r-1(u' azw' aru' arw)
+ (-1)q-l (n - q - r - 1)DI,I,q-I,r,l,n-q-r-2 (1.13) x (u, w, azw, aru, 5z5tw, arw).
1. INTEGRAL REPRESENTATION OF FORMS
12
For r = n - q - 1, the second term on the right side of (1.3) disappears, and the first term is equal to the left side of (1.11) by Lemma 1.1; hence (1.11) holds.
For 0
atw)
(-1)q- 1Dl,q,r+l,n-q-r-2(w,
=
azw' a;u, kw)
+ (_l)gDq q,r,n-q-r-1(u' azw, atu, a3'w)
+gDl,l,q-1,1,r,n-q-r-2(u, w, azw, zw'
azw).
Hence by (1.13) and Lemma 1.1 we get
azgDl.l,q-l,r,n-q-r-I(u, w, azw, atu, arw)
+5 t.(-1)g(n - q - r - 1)DI,I,q,r,n-q-r-2(u' w, azw, at.u, atw)
_ (n - r - 1)Dl,q.r,n-q-r-1(w, azw, at.u, kW)
- (n - q - r -
1)Dl,q,r+I,n-q-r-2(w' 5, W, at. u' at.w).
Dividing both sides by n - r - 1, we obtain (1.12). To prove (1.9), we now need only multiply (1.12) by co over r, and then add (1.11) multiplied by cn_q_2.
. .
c,_ 1 and sum
co.
LEMMA 1.5. UP,q` ,
Z) _ -Un-p,n-q-I(Z, 0.
PROOF. We have
(_1)q+p(n-q-1)( n - 1 ) Up,q(S' Z) =
q
(2iri)"p! (n/p-p)! Z)' 1)n
Un-p,n-q-1(Z' J) =
q,-I+(n-P)q(
(y
Z)'
(p z)) A Dp n_P(az, ar),
n 9-
(2iri)np!(n -p)! p
p
XDI,n-q-l,q(t(z, J ), att(Z, J ), azt(z, J )) A
az).
The required result now follows by Property 3, if we take into account the relation
z) _ -t(z, ).
§1. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
13
LEMMA 1.6.
w(w, J, Z) = ((2Tli))!
I
k
(-1)k-lwkdw[k]
A
l
_ (n - 1)! (-1)k-l dw[k] A d (27ri)n k - Zk where the second equality holds for Jk
Zk.
PROOF. To prove the first equality, we need only expand the determinant DI n-1(w, atw) by elements of the first column and observe that To prove the second equality, we use Lemma 1.1 to replace the first column of DI n_1(w) arw) by the column whose kth component is Gk - Zk)-1 and all other components are zero, and expand the resulting determinant according to elements of this column. 50. PROOF OF (1.1). We need the following lemma. LEMMA 1.7. If f E CI in a neighborhood of 0 E Cn, then A
lim f
-0 ItI=E
I
J
(2' i)n
d0k]
I2n-m
oft (0);
m = 0,
k
(1.14)
0; m > V.
In particular,
g(M;dL A
lira f EGO
n - (-1) n-k-I ( 27ri) n! g(0)'ak
I2n
I
Here Sjk is the Kronecker symbol (Sjk = 0 for j # k, and Sj = 1). PROOF. We have
f
ItI=E
_
E2n-m
y I2n-m I
(-1)n-k-1.
m-2n
f
(-1)n-k-1
Em-2n[
A
IS;I=e
of df A d
"I<E
_
f
4k
k (0)f,I<Edf A
f31<EI
a kk
(0
- a k (0)) df A db (1.15)
The formula for the volume of a ball yields
f
dAd_ <E
(277' )n E2n'
n
.
14
I. INTEGRAL REPRESENTATION OF FORMS
so that the first term on the right side of (1.15) gives the right side of (1.14). But the limit of the second term as z -> 0 is zero, since the integrand tends to zero as a -> 0. So the first part of the lemma is proved. To prove the second part, we need only observe that g(0)S;k.
4k We now proceed to the proof of (1.1). In view of the additivity of the operators on the left side of (1.1), we need only prove the formula for forms of the type
y=f(z)dz,Adz,. Here f E C1(D), I = (i1,...,iq) andJ = (j1,...,jo). Now
(_I)q+p(n-q-1)(n - 1 A
z) _
MW, Ad
(2vi)np! (n -p)! AD 1,q,n-q- 1 (t' azt, art) A Dp n-p(az, aJ )
(-1)q(n A
=
A
I) ajt, art) A d A dz,
(27Ti)q
z) A dz,,
and analogously
ay A U,,q($, z) = a(f(t)dfr) A Uo,q A dz,, y A UP,q-1(J, z)
=f(rS
)dpJ, A UO,q-1
A dz,.
these equalities show that (1.1) needs only to be proved for forms of type (0, q), or more precisely for forms of the type y = f(z)d.1. Finally, the kernels Uo.q are invariant under the transformations tj -+ tw, Z3 I z,(j), j = 1, ... , n, T being an arbitrary permutation. Indeed, under such a
transformation, the same rows of the two determinants occurring in the expression for Uo,q change their places. Thus it suffices to prove (1.1) for the forms of the type
y =f(z)dil A ... Adiq. For z a D, (1.1) is an immediate consequence of Stokes' formula and Lemma 1.2.
§l. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
15
Let z E D. We set
g = (1 - n)-1I -
z12-z"
99 = Uo,q - atg A V.
Since y A Uo,q has type (n, n - 1) in , and y A azg A Vq-1 is of type (n - 1, n) in , we have Uo,q)-ar(yAazgA Vq-1) =aryA Uo,q+(_1)gyAa.Uo.q-aryAazgA Vq-1 (_1)qy A arazg A
As we already observed, azatg = araag; hence it follows from (1.5) and Lemma 1.2 that (_1)qy A arazg A Vq-1 = (_1)qy A az(arg A Vq-1)
= y A a=Uo,q-1 = (_1)qy A arUo,q
This together with the preceding relation yields
dr(yA9q)=aryA
Uo,q-arIAazgAVq_1.
(1.16)
Therefore the second integral in
i(z) = f D;Y A Oq -
A eq)
(1.17)
is absolutely convergent. We shall now show that i(z) is precisely the left-hand side of (1.1). For this
we consider the form y A gVq-1. This is of type (n - 1, n) in ; hence by Stokes' theorem
-f
f y A gVq aD
y A gVq-1
(13 -=I =e)r
=
f
ay A gVq_, + (-1)9 f
(D.)r
where DE _ ( E D I
I
y A arg A Vq_1,
(1.18)
(D.)r
- z I > s}. Lemma 1.7 shows that the second integral
on the right in (1.8) tends to zero as e -+ 0, so that (1.5) and (1.18) yield
f y A Uo,q-1 = f y(.) A gVq-1 - f ay(e) A gVq_1. Dr
aDr
(1.19)
Dr
In both the integrals on the right side of (1.19), we may differentiate under
the integral sign; the first is a proper integral, while the second has a singularity of the form I z - I1-Z" after differentiating the integrand and
I. INTEGRAL REPRESENTATION OF FORMS
16
hence is absolutely convergent. Hence, the left side of (1.19) is likewise differentiable, and we get
faD
A kg A Vq-1 -f
A kg A
vq-1-af
D
Dt
y A Uo,q-1 = 0. (1.20)
Adding (1.17) and (1.20), and using the definition of Uo,q and (1.16), we obtain
i(z)=f;
ayAUoq-5 fDr yAU0,a-1
.lDI
aD
We shall show that i(z) = y(z). From (1.17) and Stokes' theorem applied to D. we get
i(z) = lim f
A OO .
C-0 jr-Zl=e
Let us compute this limit. To do this, we go over from 9q to the form Oq = (-1)Qerg A Vq - aZg A Vq-j.
From the invariance of the first differential, it follows that dt.g I K = 0, where K = ( E Cn: - z I= e), or atg IK = atg IK; hence (1.5) and the definition of B q show thatO IK= °qIK I
We set
ag
gk- ayk
__
k
ag =
Zk
aZk
Then A O q = (-1)° (fl _
x {(_1)°df, A ...
1 ) !A
0 A 1' 2 o(I, ./)(-1)' 1 df [I, .il
A k=1
I
j(41
A
A 2 (-gk)dZk .
j=1
A2' 2 o(J, k)(_1)k-1 X df [J, k] A J k6J
dzi}
17
§1. THE MARTINELLI-BOCHNER-KOPPELMAN FORMULA
(tai) h
1)!fmj(-1)q 1 gkdf, A ... Adfg A dfk A d$ [1,...,q, k] k=q+I
9
A 2 a((1,..., j=1
[j],...,q, k), j)(-1)'-I
dzq A dik + a((1,...,q), k)(-1)k-I
A d11 A ... [j] A da, A . . . Adzq
q
+df, A...AdfgAdd [1,...,4] A 2 u((1,...,[k],...,q),k) k=1 n
7, gjd.j A di1 A ... [k] ... Adiq
X (-1)k-I A
j=1
Observe also that (-1)ga((1,...,4), k)dfl A ... Adfq A dgk A df [1,...,q, k]
= 0((1,...,4), k)dfk A (df, A d3'2 A ... Adfq) A df [1,...,q, k]
= df, a((1,..., [k],...,4), k)df, A ... Adfq [k] ... Adiq A
= 0((1,..., [k],...,q), k)dfk [1,...,gIdz,
Adf, Using these equalities, we get
y
A 9q = (-1(2
1)1f(.)k=1
(-1)k-lgkdf
A ... Adzq+X. Here the form X is a sum of terms of the type
const f(0gkdf A dg[ j] A dar,
j
k, I =
iq).
I. INTEGRAL REPRESENTATION OF FORMS
18
Thus finally we get by Lemma 1.7
i(z).=lim f
E-0
E-0 I-zl=e e
f("(
(-1)n+k-1
A . . . Adzq + lim
A
(n - 1)!
0
f
f
e-O VzI=e
dJ kz I2nk)
X
(2n!)nf(z)dz' A ... Adaq = y(z). k =I
§2. Theorems on the saltus of forms
1°. In the theory of functions of one complex variable, the following results are well known (see, for example, Muskelishvili [1], §22). PROPOSITION 2.1. Let D be a domain in C' with smooth boundary, and let ip E C°'x(8D). Then the functions
P kZ) =
)d 2 Iri fD
((p+ is defined in D, and T- in CD) can be extended continuously to the closure
of the open set on which they are defined, and
T+ W - 'p- W = W),
E D.
(2.1)
To formulate the second proposition, we need some notation. Let D be a domain with smooth boundary, and let p E C(8D) and z0 E 8D. On a line _V passing through z° and not tangential to 8D, choose points z E D and z' (4 D at equal distances from z0, and consider the limit lim
Z' z'-.zo
(q)
[(P+ (z) - (P(z')]
(2.2)
are defined as in Proposition 2.1).
PROPOSITION 2.2. The limit (2.2) exists and equals p(z°). The limit is approached uniformly if the nonobtuse angle between the line V and the tangent to 8D at z° is not less than a fixed /o > 0.
It must be noted that Proposition 2.1 solves the so-called additive Riemann problem:
Let D be a domain with smooth boundary, and let qp E C°,'(8D). Find functions 4V+ E AC(D) and p E AC(CD) such that q)+ -p = q) on D.
§2. THEOREMS ON THE SALTUS OF FORMS
19
We shall need analogues of the Riemann problem and Propositions 2.1 and 2.2 for forms of several complex variables. 2°. Before proceeding to consider the analogues mentioned above, we state the following result without proof. THEOREM 2.3. Suppose 8D E and f E C"'.s(D), where m% 0 and 0 < A < 1. Then the functions defined by the integral C'"+i,A
I
f($)df [k] A d
(2.3a)
I.-z12n-2
aDr
inside and outside D can be extended to the closure of the respective open sets as and the functions defined by the integral functions of class
f
Ad
D;
(2.3b)
as functions of class Crn+2.A', 0 < A' < X.
The proof of this theorem can be carried out along the lines of Gyunter's proof ([1], Chapter II, §19) for the case of real dimension three. But the computations are much more complicated in our case.
Since the coefficients of the kernels Uo,q are derivatives of the function - z I2-2", and we may differentiate under the integral sign in (2.3a) and
CI
(2.3b), we obtain COROLLARY 2.4. If aD E C'"+ 1,a and y E C.t p;q)(D), then the forms defined
by the integral (,,q)(D, y) inside and outside D extend to the closure of the respective open sets as forms of class CC,' '), and the forms defined by I(p q- I)(D, y) extend as forms of class C(', q', with 0 < A' < A.
3°. We observed that (1.1) is the analogue of the Cauchy integral formula for
forms of several complex variables. Therefore it is natural to consider an analogue of the integral of Cauchy type an integral of the form I,,q(D, y)(z). Here D is a domain with piecewise smooth boundary, and y E Ci p,q)(aD). Let y+ denote the form defined in D by the above integral and y- the one defined in CD. THEOREM 2.5. If aD E C2 and y E C(f,q)(8D), 0 < q < n - 1, then
Y=Y+IaD - YIaD (by Corollary 2.4, y+ IaD and y -18D are defined).
20
I. INTEGRAL REPRESENTATION OF FORMS
PROOF. It suffices to prove that
f Yn9,= f (Y+IaD - YIaD) Ap aD
(2.4)
aD
I(aD). We extend qp to C" as an (n - q - 1)-form of class C'. Then q) can be represented in the form
for any T E
C.,-,,-
n-q-1
2 To,
1
Pi E=
I
i=O
-q-i-
(
n
The forms Y' and -y are of type (n, q); therefore y A p. = Y ` A p1 = 0 for = 1,... ,n - q - 1. Hence we need prove (2.4) only for p E C(""-q_ 1)(C"). Let p be a function defining D; thus, D = (z E C": p(z) < 0), p E C2(C"), grad p # 0 on aD. Set D, = (z E C": p(z) + I-I < 0) and D' = D_ . Then for sufficiently large I we have D,
D C D'
C",
aD1 E C2,
aD' E C2,
and also, for any X E C2"_ I(C"),
fo =rlim fDrx=
lim
00
Thus
JaD
f
f
(Y -Y) A p =
Af
Af lim
A Un,q(J, Z)
A
f DY(J) s
A Un,qG, Z)
Z))
L
r
p(z) A
z))
The reversal of the order of integration is permissible since the integrands are
jointly continuous. By Lemma 1.5, Un,qG, z) = -Uo,n_q_1(z, ). Hence by
§2. THEOREMS ON THE SALTUS OF FORMS
21
(1.1) and Stokes' theorem we get I.
f D(Y+ -Y) n 9) = lim00 = lim
1- 00
^f[a(D'\D,)l: p(z) A Uo,n-9-
(aD)r
f
)
rr
A 5TQ) + f
(aD)r
&p(z) A Uo,n-q- 1(z,
(D'\D,),
1111
+5f =
1(z, f1-
f y A ± lim f aD
1-
t
Uo,n-q-2(Z,J)(
A
Af
(D'\D,),
3Dr
0
D'\D,).pp(z)
+ (-1)n+q ay(J) A f
(D'\D,)
5p(z) A Uo.n-q-I(Z,
(p(z) A UO,n-q-2(Z, J )1.
From the form of the kernels Uo,,., we see that the last limit above is zero as a consequence of !li m
fDr
j dfD'\D,):
z )(SZ
- Zk ) dzAdz
,
(2 . 5 )
where z) E C(8D X Dr\D,). We show that the integral
f
1
DX(D'\D,) I
-zI2n-I
(2.6)
ds dv
exists. Here dv is the 2n-dimensional volume element, and ds the area element on 8D. Since the integrand is positive, (2.6) exists if and only if the following limit exists:
lim f
e-.O
1
IaDx(D'\D,)l\(I1'-zl<e)
I
- zI2n1
duds.
Now (2.7) is a proper integral; hence we may go over to the iterated integal, and the existence of the limit (2.7) is equivalent to that of
lim f
e-.0 aD
But lime_0
ds, where ge(l) = f
dv.
1
(D\D,)\(13'-zl<E)
-zI2n-1
exists, and
0< &G)
I
-
Z I2n-I
f
v I Z I2 -1
00,
(2.8)
22
I. INTEGRAL REPRESENTATION OF FORMS
where V = UtEaD [(D'\D,) - fl C C"; hence the limit (2.8) exists by the Lebesgue dominated convergence theorem, and so (2.6) exists too. Since the integrand in (2.5) is majorized by that in (2.6), we may reverse the order of integration in (2.5), and the function
ID;Miz)(fZIZ"zk)d
[jJAd
is integrable; hence the limit in r(2.5) equals !11
n
dz A dzJ D
I
r
Z)(Jk IZ"zk)
dr [.1] A d = 0,
ID,\Dr)
since the measure of D'\D, tends to zero as l -* oo. 4°. In this subsection, we shall obtain an analogue of Proposition 2.2 for forms of type (p, 0). Let D be a domain with smooth boundary, z° E 8D, and Y E C(P,0)(aD). In the interior of the right circular cone Vo with vertex z°, the normal to aD at z° as axis, and angle between the axis and generator /3 < 7r/2,
we choose two points z E D and z' (4 D such that a (z - z ° I< z' - z ° b I z - z° , where a and b are constants with 0 < a < b < oo. Consider the limit Z,
lm [y +(z) -Y (z')].
(2.9)
THEOREM 2.6. The limit (2.9) exists and is equal to y(z°). The convergence is uniform in z0 E aD if /6, a and b are fixed.
Note that for n = 1 this is a strengthening of Proposition 2.2. PROOF. Exactly as in the proof of (1.1) we can show that it is enough to prove the theorem for (0, 0)-forms, i.e. functions. Thus let Y E C(aD). Then Y+ (z) - y -(z') = 101,0(D,
Y(z°))(z) - Io,o(D,
Y(z°))(z')
+Y(z°)[Io,o(D, 1)(z) - 1010(D, 1)(z')] By (1.1),10.0(D, lXz) = 1 and Io0(D, IXz) = 0; hence we need only prove that Z
urn
f
D
(Y(z)
-
z) -
z')) = 0.
(2.10)
By a unitary transformation and a translation, we make take z° to 0, and the
tangent plane to aD at z° to the plane a = { E C": Im J. = 0). Then the left side of (2.10) is unchanged, and aD will be defined near 0 by the system of equations ' = 'w and " = u,, + ip(w). Here 'w = (w1,...,wn_1), w = ('w, u") E a, and p E C'(U), where U is a neighborhood of 0 in the plane a
23
§2. THEOREMS ON THE SALTUS OF FORMS
and p(w) = o(I w I). Let i and i' denote the projections of z and z' on the
Im axis. Then
lz'-VI
Iz - I< tan/3lil,
Iz'I
Izl
(2.11)
a 11 I/cos s
0, and choose a (2n - 1)-ball B in a such that 1) B C U;
2) I y(e(w)) - y(O) I< e, w E U;
3)lw-1I
Let B'= Let us estimate the integral over B'. Using (1.4), we get (n - 1)!
U0,0G, Z) - U0,0(t, z') =
(2T-i)
(-l)k-I
k=l
_ X (
I yk_ z I2n
I
yk
_ rLk] n zzl2n pd[k]
Now we use condition 3) and the inequality I (w) I < CI I w I < C, I w - i I to obtain fk
fk I _Z,I2n
I - ZI -I - Z'I 2-I I
I Z I +I z' I
CIC2n+22 - + '
j=0
j=0
Iw-ilj
(2.12) Iw-i'I2n-j.
We may assume that a, = a(cos $)-' < 1; then (2.11) implies I w - 2 I
Iw-a,ilandlw-f'l>lw-a,il.Thus from (2.12)weget
It
-Z-I2n
where d depends only on a, b, C, C, and P.
Iw-a,2I2n
(2.13)
24
I. INTEGRAL REPRESENTATION OF FORMS
Similarly, Zk
Zk
I-zil2n <
I-ZI2n
IZkI
+
IZkI
<
d11191 (2.14) Iw_alf12n'
-Z.I2n
where d 1 depends only on a, b, C, C1 and ft. Finally,
da < d2ds. (2.15) Here do is the surface element on aD, and ds that on a; d2 is a quantity not depending on zo E aD. Now, from condition 2) and (2.13)-(2.15) we get
JBi (y(O -
Z) -
z'))
(n-1)]d2(d+dl)ef Iw'ZIa I2,
3e,rR2_1 Iw111ds n-
Herew=(w1,...,wn_1,un)andi=(0,...,0,iyn);henceI I=lynl and Iw-ai!l2n
Thus the last integral differs from the Poisson integral only by a constant factor and is independent off. To finish the proof, it remains only to note that urnZO Z'
z1_
z) -
z')) = 0,
and that this limit is approached uniformly since the radius of the ball B does not depend on zo E aD. COROLLARY 2.7. If aD E C, x and y E C(p o)(aD), then y = Y+ Ian - Y Ian on aD. This follows immediately from Theorem 2.6 and Corollary 2.4.
We note that, if y E C(aD), then y` are in general not extendable to D as continuous functions, and it is not possible to obtain an assertion analogous to Corollary 2.7 for continuous forms. This is shown by the following example.
EXAMPLE 2.8.(4) Let D be a domain such that aD contains a (2n - 1)dimensional ball B lying in the plane {yn = 0}, with radius R < I and center 0. On B we set
(4) This example has been constructed in analogy with the one given by Gyunter ([Ii, Chapter II, §7).
25
§2. THEOREMS ON THE SALTUS OF FORMS
Then y E C(B). We extend y as a continuous function on D. The behavior of the functions y' near B is determined by the integral
z)
71(z) = f
Let us show that y, is not bounded in any neighborhood of 0. Indeed, let z = (0,...,0, iy,). Then Y1(Z) -
(Sn - iy,,)df [n] A d CJB;
C
-ZI2n
lhI
U.
The integrals of the other terms vanish since drn A din = 0 on B. Furthermore,
[n] A d IIt-zI2n
fBt
_
-0,
since the integrand is odd and the domain of integration is symmetric about 0. Transforming to spherical coordinates, we have Y1(z) = C, fscos2(Txn)^ d s f
R
o
rIrZI
r r2d+
I In
(2.16) y,2 In
the Here, S is the (2n - 2)-sphere of radius 1, ds its surface element, angle between the x, axis and the radius vector of T, and C, # 0 a constant. The first integral on the right in (2.16) is positive; hence r2n-1dr R rzn-1dr 2 In ly2(z)I % C2f C21 R I1nrIIr2+yn o
IlnrIIr2+yn
s
nIn
C2 > 0.
As yn - 0, this integral converges to fR
a
r2n-'dr _ In 8 IlnrIr2n - InR
so that, for sufficiently small yn, IY,(z)I
C2 1n1
In 8
In R
I.
(2.17)
Since the right side of (2.17) tends to infinity as 8 - 0, it follows that y, is unbounded in every neighborhood of 0.
5°. For p < n and q > 0, the assertions analogous to those of Propositions 2.1 and 2.2 and Theorems 2.5 and 2.6 are false, as the following example shows. A
EXAMPLE 2.9. Let D be the same domain as in Example 2.8, a = din-q+1 ndzn_ 1 Adz, A ... Adz, and y = dzn A a. Then y E For
26
I. INTEGRAL REPRESENTATION OF FORMS
z E B,
Y+Ian - 7_6 = Ti Ian - Yi Ian' where
Y Since
A Up,q(J, z)
A Up,q is of type (n, n - 2)y in ', we have
Y(J) A
Z) 1B = 01
i.e. y = 0; hence for z E B we have y+ Ian - Y-Ian = 0, but y Ian 0. For such forms, the jump theorem can be obtained in the following way: Let D be a domain with a piecewise smooth boundary, and y E Cep q,(8D). Let y denote any form in C(p.q)(D) such that f Ian = Y, and let
y (z ) = Ip q(D,Y)(z ) - 8Ip,q-i(D,Y)(z±), z+ E D,
z-E CD.
THEOREM 2.10. y,, extend continuously to 3D, and
Y = Y+Ian - Y-Ian
(2.18)
PROOF. By (1.1), we have for z E D
Y+ (z) = Y(z) + Ip,q(D, aj)(z)
(2.19)
7-(Z) = I,,q(D, aY)(z)
(2.20)
and for z (Z
The integral that occurs in the definition of Ip q is absolutely convergent for all z E C"; hence II,q(D, 8Y) E C(p,q)(C"). Hence the forms y can be extended continuously to 8D preserving (2.19) and (2.10). Substracting (2.20) from (2.19), we get (2.18).
6°. We now proceed to the additive problem of Riemann for forms. We formulate it as follows: Can every y E C(p.q)(8D) be represented in the form 7 = Y1Ian - Y2Ian'
(2.21)
with y, E Z(, p,q)(D) and y2 E Z(Ip q)(CD), and it not, what are the conditions on y of the validity of such a decomposition?
THEOREM 2.11. I. If 8D E C' and y E Ccp.q,(3D), then the decomposition (2.21) holds only if there exists y e C(p,q)(C") such that f Ian = Y and 8y = 0 on 8D.
§2. THEOREMS ON THE SALTUS OF FORMS
27
C"'+',a and y E C'," (aD), m % 1, then a sufficient condition for II. If aD E (2.21) to hold is the existence of a y, E Cp,4(C") such that 11 Ian = Y and aYi A dz! Ian = 0 for any J = (jP+,,.. ,j") The forms y1 and y2 can then be
chosen in Cep;qj, 0 < A' < A, with y 2 = o(I z I
1-2n)
as
I - - > > oo.
PROOF. I. We first extend y, and y2 to all of C" as forms of class C(1,P,4)(C").
We continue to denote these extensions by y, and y2, and set
Yj - Y2
Then jlan=y,and on aD
ay=ay,-a72=0, since ay, =0forz E D and ay2 = 0 for z ED. II. We shall show that the forms y± defined in subsection 5° are a-closed. Indeed,
f
aye=alP4(D, y)=azf
aDr
aDr
Since y A Up,q is of type (n, n - 1) in , it follows by Lemma 1.2 and Stokes' formula that
ay. _ (-1)q f y(')
z)
aD;
_ (_i)q f D;aj1
) A UP,q+1(t, z) = 0.
z) and U,,q_G, z) are majorized uniformly in ; E D by a function of the form C I z I'-2x. An For sufficiently large z, the coefficients of
application of Corollary 2.4 and Theorem 2.10 concludes the proof.
For q = n - 1, we can get more information on the forms y, and Y2 occurring in (2.21). Let us denote by Bp,,,- (D) the set of a-closed forms of the type
a = 1 2 (-1)k-I aG'dz[k] .l k=1 where the J in D.
A dz,,
k
are multi-indices and the G. are harmonic functions
THEOREM 2.12. If aD E C"," and y E C(pn)(aD), m ' 1, then (2.21) holds with y, E ZAP:,,_ ,>(D) and y2 E BP,,,- I(CD) fl Ccpn_,>(CD), 0 < A'<,\, y2 = O() z I ' -2") as I z I -> oo.
If p = n, then y, E Bn,n_,(D).
28
I. INTEGRAL REPRESENTATION OF FORMS
PROOF. Let y E and Y Ian = Y. Then 12,,_2(D, y) E CI p+''2)(CD) (see Corollary 2.4), i.e. there exists /3 E CCP , 2;(C") such that
/f3=Ipn-2(D,y")on CD. Wesety1 =Y++8/3andy2=I ,,,-1(D,y)(yam have been defined in 5°). We have (see (1.4))
Up."-11J, z) =
(n - 1)! (27ri) X
(-1)k-1 J k=1
aazk (n-a(J)d[J]
di [k] A dz; 1)I('-zI2"-2
hence 1 ,"_ 1(D, y) E Bp,,,- (CD). By Theorem 2.9, Y1IaD -
Y2IOD = 'Y+
18D - (I1 .n-1(D,Y) - aI ,"-2(D,Y))
=Y+IaD-y-IaD=Y The remaining properties of y1 and y2 are proved in the same way as in Theorem 2.11. For proving the second part of the theorem, we have to use Theorem 2.5. In connection with Theorem 2.12, we note
COROLLARY 2.13. Let aD E Crn+1'A and y E Zip n_1)(D), m > 1; then y = 0 < X' 71 + a$, where y1 E B(,,,,-,)(D) f1 Z ( ' P ,n _ 1)(D) and Q E C(' n 2)(D),
< x. This follows easily from (1.1) and Corollary 2.4. §3. Characterization of the trace of a holomorphic form on the boundary of a domain
1°. Proposition 2.2 allows us to give a necessary and sufficient condition for a continuous function defined on D to be extendable holomorphically into D (see Muskhelishvili [1], §35). Theorem 2.6, being an analogue of Proposition 2.2, can be similarly used for a similar purpose in the case of functions (or more generally forms) of several complex variables.
THEOREM 3.1. Let D be a domain in C" with smooth boundary, and let y E C(p.o)(aD). Then a form y E AC,(D) such that y IaD = y exists if and only if aD
for any 0 E Z(n-p,n-l)(D)
29
§3. TRACE OF A FORM ON THE BOUNDARY
PROOF. Necessity. Let D, be the sequence of domains defined in the proof of Theorem 2.5. Then
f yne= lim f !AO= lim f d(yne)=0 /-Woo aD
1--0
aD,
D,
for any 0 E
d(yne)=a(Yne)=ay A0+(-1)pyna0=o. Sufficiency. By Lemma 1.2, at.U,,O(L z) = 0 for 0 z; hence it follows from the definition of y- (§2, 2°) and (3.1) that y 0 for z D. Then Theorem 2.6 yields
lim
Y+ (z) = Y(z°),
zED
the limit being approached uniformly in z° if z approaches z° along the norm to aD. Let us show that the form Y
y+,
z E aD, z E D,
is continuous on D. Indeed, y E and so for each e > 0 there exists S, > 0 such that, for all z0, z' E aD with I z' - z° < S,,
Iy(z') - y(z°)I < e/2. (S) Choose 8, 0 < 6 < 6,/2, such that if z' EaD and z lies on the inward normal
toaDatz'andIz-z'I<6,then IY+ (z) - 1'(z')I < e/2. Now let z° E )D, z E D, and I z - z° I < 6. Choose z' EaD such that p(z, aD) = I z - z' I . Then z lies on the normal to aD at z', and
Iz' - z°I _Iz' - zI +Iz - z°I = p(z, aD) +Iz - z°I < 26 < s,. Then
(y(z)
- Y(z°)I -Iy(z) - Y(z')I +IY(z') - y(z°)I <
E.
Thus we have proved the continuity of y at the boundary points. The continuity at interior points is obvious. (5) By the absolute value of a form, we mean the maximum of the absolute values of its coefficients.
30
I. INTEGRAL REPRESENTATION OF FORMS
It remains to prove that Y E AP(D). For a fixed z E D, on account of Lemma 1.2 we have
h(z) = ay + (z) = faDr
A
_ (-1)P fDY(s.) A
z)
z).
t
z) in some neighborLet UU) E C, ,,n_2(C") be such that E 1(C"), and (3.1) yields
hood of aD. Then
aY(z)_(-1)Pf
aD1
2°. Suppose f E C'(aD) satisfies the tangential Cauchy-Riemann equations, i.e.
dfAdzlaD=O.
(3.2)
by Stokes' theorem,
Then, for any X E faD
fax=
dfAX=0; faD
hence it is natural to call y E C,,0(8D) a weak solution of the tangential Cauchy-Riemann equations if
faaD yAax=0 for all THEOREM 3.2. Let D be a domain in C", n > 2, with smooth connected boundary. Then the space of weak solutions of the tangential Cauchy-Riemann equations is identical with the space of traces on aD of forms in ACP(D).
The proof is exactly the same as that of Theorem 3.1. We need only show that, if 8D is connected and y is a weak solution of the tangential Cauchy-Riemann "equations, then y- = 0 in CD.
Let B be a ball with D C B. Since B is a domain of holomorphy, and Band z(4 B, we have XzE so that for z
Y -(z) =
f
aD1
B
A Uz) = faD;
A arx
0.
(3.3)
aD being smooth and connected, CD is connected; and y- is harmonic since the coefficients of z) are. Hence (3.3) and the uniqueness theorem imply that y-=O in CD.
§4. CASES OF SOLVABILITY OF THE 8-PROBLEM
31
§4. Some cases of solvability of the 3-problem
1°. As we know, one of the central questions in the theory of functions of several complex variables is the solution of the a-problem, i.e. of the equation
as=Y,
(4.1)
where y is a (p, q)-form, q I. Many questions in complex analysis ar either closely related to this problem, or can be reduced to it. For example, D is a domain of holomorphy if and only if the a-problem is solvable in D for any y E Z(,,g)(D), I ' p < n (Hbrmander [1], §4.2). We need the answers to the following questions.
1. Let y E Z p,q)(C") and supp y = K. Under what conditions on y and K can (4.1) be solved with the same support? 2. Let y E Z(p,q)(D). Under what conditions on D can (4.1) be solved with the same smoothness on D? 2°. To answer these questions, we need the following result of Khenkin [I],(') sharpened by Hvrelid [1]:
Let D be a bounded strictly pseudoconvex domain with boundary of class C'. Then there exist a domain of holomorphy V, D C4 V, a vector-valued z), and positive constants e, C, and C2 such that function 1.
z) E C'"-'((V\D) X V), and
2. 3.
z) is holomorphic in z E V, for each fixed E V \ D. z), - z) 0 for E V\D, z E z) =
C1(p(f) - p(z)) + C2 I - z 12 1<e. In the rest of this section, we shall denote by u the vector-valued function
4.
z) $
P.). -iP = (P1'...' 4, OD
We recall that a domain D C C' with boundary of class C2 is said to be strictly pseudoconvex if D = {z E C": p(z) < 0), where p E C2(C"), grad p 0 on 8D, and the Levi form " a2 I
is positive for all z E aD and all w az i
(z)wtw;
r
0 such that
(z).w,+...+az (Z)
(6) We state the result only in the form we need.
"
0.
1. INTEGRAL REPRESENTATION OF FORMS
32
3°. Let us consider integrals of the form
f. y(z) A DI,I,q,r,n-q-r-2(t, U, azt,
aft) ADP n-P(az, of ),
f y(J) A DI,l,q,r.n-q-r-2(t' u, azt,
a('t) A DP.n-P(az, ar),
D
Dt
where t = ( - a)/l - z 12 as usual, and y is a form of the appropriate dimension. Our interest is in the following question: is it possible, by imposing suitable conditions on y, to extend the forms defined by these integrals to forms on V, differentiable as many times as we need? Let us first explicitly write down the singularity of the integrals: LEMMA 4.1.
D1.I.q,r,n-q-r-2(U, t, azt, kU, apt)
(f - Zk)µk'r(J, Z)
n
- Z) I2(n-r-1),Dr+1 '
kl 2I
where the L%,r are forms of type (0, q) in z and (0, n - q - 2) in , with coefficients in Cni-2((V\D) X V). The proof proceeds exactly like that of (1.4). We can now answer the question asked above. LEMMA 4.2. Let m > 3 and y E Cp q 3'X(V).
l.If y=0on V\D,then
f.y(z) A Dl,1,n-q,r-2(t, u, 5't,
att)
D
A Dn_P,P(az,
2. If
E C(p,g2 2)(V\D).
0 on D, then
f(V\D)f
A DI,1,q-r,r,n-q-2(t, U, azt, ADP,,, _P(az'
art) E C(P.9? 2)(D )'
PROOF. We shall prove the first part; the proof of the second part is entirely similar. It is sufficient to prove the lemma for m < oo.
From Lemma 4.1, we see that our assertion is a consequence of the following:
If h E
h = 0 on V\D, and qq E C,-2((V\D) X V), then
Im =fDr $
z)(-
I-ZI2(n-
f) di A dz E Cm-2(V\D).
(4.2)
33
§4. CASES OF SOLVABILITY OF THE a-PROBLEM
For E V\D, (4.2) is a proper integral, so we may differentiate under the E Cm-2(V \D). integral sign m - 2 times. Hence The derivatives of the integrand are linear combinations of expressions of the form
= h(z)Dr,
k
D(.1' 1+1
where III + I' + J + J'll < m - 2 and
2(nkr-1)XIJ, Z),
z) is continuous on (V \D) X V.
And we have Dr'-I
(4q,
M1 Z) IIII+I'II+r+l '
z))r+l 1
D1.1'
t
Jk - zk
M1
I y - Z I2(n-r-1) I
-z
12(n-r-1)+IIJ+J'II-1 '
z) E (V \ D) X D.
where Ml does not depend on
We have h = 0 on 8D = {z: p(z) = 0); hence by Hadamard's lemma (Arnol'd [1], §2.6.4) there exists hl E Cm-4'" (v') such that h = ph1; in turn, we have h I = h p I = 0 on V \ D, and so, by continuity, h I = 0 on D. Repeating
this procedure m - 3 times, we get h = pm-3h2, where h2 E C°'a(V) and h2 = 0 on V \ D. For A' such that 0 <X' < A, the function h o = h 2 lies in C((V \ D) X 5); hence M21 p(z) ICI
II /+/'Il+r+ 1
' - z 1-
1m-3+A'
I& - Z
2(n-r-1)+111+J'II -1
IIIIJ+J'll- 1
4)(i, z)
I
M3
14)(J, Z) Ir+1 .
I
- Z 12n-X'-2(r+I) '
where M2 and M3 are suitable constants. z) and the fact that p(z) < 0 for z E D and Now, from property 4) of 0 for E V\D, we have
Ip(z)l
I - Z12 < c2
T
z)I
and, further, p(z) I< C3 I - z I for z E 5 and E V \ D. Hence 101 < I
ZI
2n-a'
E V\ D, the integral of any derivative of the integrand in (4.2) is convergent uniformly in , i.e. all
where M is a constant. But this means that, for
I. INTEGRAL REPRESENTATION OF FORMS
34
of order up to m - 2 extend continuously to V\ D. Now derivatives of (4.2) is a simple consequence of the following assertion. Let D be a bounded domain in R" with boundary of class C'°, m >' 0, and let f E_C"'(D). Suppose f and all its derivatives of order up to m extend continuously to D. Then f E Cm(D). For n = 2, the proof of this statement is given in Fikhtengol'ts ([I], Paragraph 260). For n > 2 the proof is entirely analogous. V. We can now proceed to the solution of the problems formulated in V. THEOREM 4.3. Let D be a bounded strictly pseudoconvex domain with boundary of class Cni+3, and let Y E Z(p ,)(C"), m ' 0. Suppose supp y C D. Then:
1. For q < n, there exists a E C(y,q_,)(C") such that supp a C D and as = y. II. For q = n, an a E C(v,"_,)(C") such that supp a C D and as = y exists if and only if
fDµAY=0,
it EA"_p(D).
III. If a E C(p,q_,)(C") and as = y,(7) then there exists such that a# = a on CD, q > 2.
(4.3) E C( p q1
(C")
PROOF. The necessity of (4.3) is a simple consequence of Stokes' formula. Indeed, for any µ E A"_p(D).
f µAy f d(tt Aa)J D
D
It Aa=0.
aD
Let us prove I and the remaining part of II. Let V be the domain of holomorphy introduced in 2°. Since supp y C D and ay = 0, for z E V we have by (1.1)
Y(z) = 8Ip,q_t(V,Y) = alp q-i(D,Y).
(4.4)
If q = 1, then 1p q_ ,(D, Y) E Ap(CD), since supp Y C D. By Hartogs' theorem, Ip,q_,(D, y) can be extended to all of C" as a form a, E Ap(C"). It now follows from (4.4) and Corollary 2.4 that a =1p q_,(D, y) - a, is a form with the desired properties.
We now consider the case q > 1. Then, by Lemmas 1.3 and 1.4, for z E V \ D we have IP,q-t(D,
fD,y(z) A a:µq-I A p, (nl p), Dp,"-p(az, I
+5j, y(z) A µy-2 A _
1
Pn( P) Dp,"_p(az, a ),
(7) If q = n, it will be supposed that y satisfies (4.3).
(4.5)
§4. CASES OF SOLVABILITY OF THE a-PROBLEM
35
for q < n, and
Ip,n-1(D, y),(') = fzy(z) A WP,o(u, , z) D
+a f y(z) A µo A pt (n - p). DP, n- P (az, D,
(4.6)
(for convenience, we have deviated from our usual notation and interchanged the places of and z).
Using Stokes' formula and the equality Y IaD = 0, we see that the first integral in (4.5) vanishes. Since z), and consequently WP,o(u, , z), are holomorphic in z on D, the first integral in (4.6) also vanishes by virtue of (4.3). By Lemmas 1.3 and 1.4 we see that the second integrals in (4.5) and (4.6) are linear combinations of the integrals considered in Lemma 4.2; hence 2
IP.4_1(D,Y) = 8a
on V\D
(4.7)
and 0 E C( p g1-2)(V ).
Corollary 2.4 and (4.4) now show that a = JP ,q- 1(D, y) - aft is a form with ' the desired properties. It remains to prove III. By (1.1), we can represent a as
a = Ip 4_t(V, a) - Ip,4-1(V, Y) - 5 'P,4-2(V, a). As in the proof of Theorem 2.10, we can show that
IP,4-1(V, a) E Z(P,4-1)(V ) Since V is a domain of holomorphy, IP, (V' P 4- 1
a)=aat
with /31 E C( p,4_2)(V ). The proof is completed by applying (4.7) and Corollary 2.4.
THEOREM 4.4. Let D be a bounded strictly pseudoconvex domain with C'"+2
boundary and y E Z(p y)(D), m
1. Then there exists a E C( 4_t)(D) such
thataa=Y. PROOF. Consider the domain of holomorphy V introduced in 2°, and a Y E C(p,q)(Cn) such that Y = y on D and supp f C V. By (1.1),
-Ip,4(V, 8y) - 8Ip,4-1(V, Y ) For q = n we have Ip n(V, 51) = 0; hence the theorem follows in this case from Corollary 2.4.
1. INTEGRAL REPRESENTATION OF FORMS
36
Now let q < n - 1. Then, since ay = 0 on D, Lemma 1.4 yields for z E D
Y(z) y
a(5-f r
A µQ-1 A
n atµqz n
p!
PI (n
1
p)I
Dp.n-p(
(n 1_ p)i Dp n_p(az a0.
Since aY = 0 on a(V\D), the second integral above vanishes by Stokes' theorem. The first integral is a linear combination of integrals considered in Lemma 4.2; hence the proof is by appealing to Lemma 4.2 and Corollary 2.4. COROLLARY 4.5. If D is a strictly pseudoconvex domain in C", n ' 2, with boundary of class C" 1, m , 2, and g E if ag can be extended to CD as a form of class C('0-"'(CD), then g can be extended to C" as a function in Cm(C").
Also, for every extension of ag to a form a E Z(o,'(C"), there exists an extension f E Cm(C") of g such that a f = a.
PROOF. Suppose ag is extended to al E Cro I) 1'(C"). Then aa, E Ccoz>2.x(C"), and supp aa, C D. Further, if n = 2, for any u E A2(D) we have
f uAaa,= f LAaa,= faD, IA Aa,= f ju A8g=0. D
D,
aD,
Here D, is a domain with smooth boundary such that D C D, and u E A2(DI). Now Theorem 4.3, III shows that there exists h E Cm(C") such that 8h = a,
on CD. The function p = h - g is holomorphic on CD; hence by Hartogs' theorem it can be extended to an entire function. Therefore g = h - p can be extended to C" as a function of class Cm(Cn).
To prove the second part of Corollary 4.5, we note that, by Theorem 4.4, a = a f, with f, E C" (C" ). Now g - f, is holomorphic in CD; let f2 be the entire function extending it to U. Then f = f, + f2 is the desired extension. This corollary is a generalization of Hartogs' theorem to smooth functions. For domains which are not domains of holomorphy, an assertion similar to Corollary 2.5 is false. An example of a function g not extendible continuously to CD, for which ag admits a C°° extension to CD, will be constructed in §8 (see (8.1)).
CHAPTER II
FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
§5. Polynomials orthogonal to holomorphic functions
1°. Let D be a bounded domain in C", with smooth boundary, and let 0 E D. Let wk(z), k = 1, ... , n, be C' functions on aD such that, for all
zEaD, (w(z),z)= 1
(5.1)
We set(')
(_l)k-lwkdw[k] A dz,
w(w, Z) _ k=1
where w = (w,,...,w,,) and [k] signifies that dwk is to be omitted. If w(w, z) is nondegenerate on aD, then every a E C2"_1(aD) can be written as a = cp(z)w,
with q) E C(aD). Instead of discussing forms orthogonal (with respect to integration over aD) to holomorphic functions, we shall discuss functions p that are orthogonal to holomorphic functions in the sense that
faDf(z)-V(z)w(w, z) = 0
(5.2)
for all f E A(D). We shall denote the subspace of such functions q2 E C(aD) by O(aD). In certain situations, it is useful to have a description of the polynomials P(z, w) which he in O(aD). Observe that, for n = 1, P(z, w) and w(w, z) are respectively P(z, l/z) and dz/z. ()Note that w(w, z) = w(w, z,0) of §1.2°. 37
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
38
THEOREM 5.1. P(z, w) E O(8D) if and only if
I,;;;,
r
TI
Tn-1
l - T1- ... -Tn-1
ZI
Zn-1
Zn
P(Zl,...,zn, -,...,
dT1 A ... AdTn-I
T, + ... +f,_I
zi, zl, j=1
I ,...,
l>...>Zn>
zl
1
(5.3)
>
Zn
where the Qj are polynomials such that Qj(0, zl,..., l/zn) = 0, j = I,...,n. We divide the proof of this theorem into several lemmas. LEMMA 5.2. If f E A(D) and P(z, w) has the form m
2
P(z, w)
aak.pkZakw#k,
p
k=1
then
where
f aak pk Dok'0 (II /3 I(+ n - ] k
f DP(Z> w)f(Z)w = (27ri)nki1
(5.5)
Y
0
This lemma follows easily from the Cauchy-Fantappi6 formula (see (13.2)), written (for 3' sufficiently close to 0) in the form
f(0 = (n - 1)! f
f(z)w(w> z) 8D (I - (J> w))n
(27ri)n
LEMMA 5.3. In order that a polynomial P(z, w) of the form (5.4) lie in 0(8D), it is necessary and sufficient that, for all n-tuples y = (YI,. . ,y,,) of nonnegative integers, m
k=1
zak+y
k
7. a a k pk '
Ds 10 pp
(IINkI + n
1)! z=°
= 0.
(5.6)
PROOF. It follows from (5.5) that (5.6) signifies orthogonality of P(z, w) to
all monomials z''; hence the necessity is obvious. The right side of (5.5) depends neither on the form of D, nor on the concrete choice of the w;(z), i = 1,...,n. Thus, whether P(z, w) lies in O(3D) or not depends only on the properties of P(z, w) itself. In particular, P(z, w) E O(8D) if and only if P(z, w) E 0(8B(0, r)), where B(0, r) = (z: I zl 12 + + I zn 12 < r2). We have seen in Examples 1 and 2 of 3° how the w,(z) can be conveniently chosen for a
ball. Each f E A(B(0, r)) can be uniformly approximated on 8B(0, r) by polynomials in z. Hence the orthogonality of P(z, w) to all monomials zT already implies that P(z, w) E 0(8B(0, r)).
§5. POLYNOMIALS ORTHOGONAL TO HOLOMORPHIC FUNCTIONS
39
Finally, from (5.6) we deduce LEMMA 5.4. A polynomial P(z, w) of the form (5.4) lies in O(8D) if and only if, for all n-tuples y of nonnegative integers, #k j ... fink! Qak k (5.7) = 0.
(II,6k II + n - 1)!
ok_ak_y
The proof of Theorem 5.1 now reduces to comparing (5.7) with the equality
f, ;;-0,...
P (Z' r _1>0
TI
, ZI
T" _
Zn-I
flk _ak = Y
aak k
1 -TI_ ...-T_1 Zn
N"
(115k11 + n - 1)I
z
7
COROLLARY 5.5. Every polynomial P(z, w) can be represented as the sum of a polynomial Q(z, w) E O(aD) and a polynomial R(w) depending only on w.
PROOF. It suffices to consider a polynomial P(z, w) of the form (5.4) for
which p l - ak =- y is a constant vector. If, at least one coordinate of y is negative, then P E O(aD). If all the coordinates of y are nonnegative and (5.7) is satisfied, then again P E O(aD). It remains to consider the case when all the components of y are nonnegative and (5.7) does not hold. In this case, we can add a monomial ao,YwY to P, choosing ao.Y so that (5.7) is satisfied for the polynomial P + ao Yw', i.e. that this "augmented" polynomial lies in O(aD). Note the inclusions A0(aD) C O(aD) C C(aD), where A0(aD) is the space of traces on aD of functions f(z) holomorphic in D and continuous on D with f(O) = 0. From Theorem 5.1 it follows that O(aD) is not a ring for n > 1. We stress that the characterization of the polynomials in O(aD) given by Theorem 5.1 and the representation in Corollary 5.5 do not depend on the concrete form of D.
2°. For large classes of domains, the functions w1(z ), ... , w"(z) can be chosen so that
polynomials P(z, w) are dense in C(aD).
(5.8)
Then the forms P(z, w)w are dense in C2n_I(aD). The following theorem may
be considered as a new step in the characterization of forms orthogonal to holomorphic functions. THEOREM 5.6. If w; E C2(aD), i = 1,...,n, then a polynomial P(z, w) belongs to O(aD) if and only if the form Pw can be extended into D as a a-exact form of
type (n, n - 1).
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
40
For the proof, we need LEMMA 5.7. P(z, w) E O(aD) if and only if P(z, w) can be represented as a linear combination of polynomials of the following two kinds.
zaw$, with ak > a a+y(
Nk for
some k,
(5.9)
);
(5.10)
here a, /3 and y are nonnegative integral vectors, and Cx, = (fil + 1)(II $ II + n)-'.
PROOF. It follows from Lemma 5.4 that polynomials of the form (5.9) or (5.10) lie in O(aD). To prove the necessity, it is enough in view of the same lemma, to show that a polynomial m
Wy 2 aakZakWak, k=1
(5.11)
lying in O(aD) is a linear combination of polynomials of the form (5.10).
Without loss of generality, we may suppose that a' = 0 and a1 k = 2,.. - m. Then (5.11) can be written as
0,
ZakWak+Y
m
aak
k=2
ZlkWlk
(Zlk v/k - Cak+Y.lk )
m
zakw ak+y
+ 2 as*Cak+Y,lk
(5.12)
+ aowY.
k=2
Zlk t'!k
The first term in (5.12) is a linear combination of the polynomials (5.10); the remaining polynomial has lower degree in z than the initial one and also lies in O(aD). Repeating this argument, we will have (after finitely many steps) a representation of (5.11) as a sum of a linear combination of the polynomials (5.10) and a polynomial bwY; by Lemma 5.4, bwY E O(aD) only if b = 0. PROOF OF THEOREM 5.6. Sufficiency follows immediately from Stokes' theorem. We shall prove the necessity for polynomials (5.9) and (5.10) (see Lemma 5.7).
Suppose the polynomial has the form (5.9). We may suppose that k = 1. Then z awo
= (z,w,),g,Z22 ...
zj i-0'(1 -
w
and so it is enough to consider the polynomial P = z" .
a,>0.
Zn"Wf, 2 ... W!",
Z2W2 .
zn^wf2
w!^ with
§5. POLYNOMIALS ORTHOGONAL TO HOLOMORPHIC FUNCTIONS
41
0 we have
By Lemma 1.6, for Zk
1)!
w(w, z) = (n
(2-i)
Z dw[k] A dz.
(5.13)
k
Therefore, for z, 0 0
Pw=
n-1 1 (2iri
)n
'
n
zal-IzZz...za-w2z...w -dw[1] A dz t
(27ri)n(Q2 )+ 1)
-IZ2 ...
l
... w -dw[1,21 A dz}
Z1
This equality can be extended to all of aD by continuity. It remains to consider the case when P E O(aD) has the form (5.10). In this case JJ
n
P(z, w) = aWa+y Ca+y.l 2 k=1
For z1
zn
In
ZkWk - Z1W1 = zawa+y 2 akzkwk k=1
0 we get from (5.13)
Pw = (n
1)! zaWa+ymY1 (-1)m-lamwmdw[m] A dz. i1
(5.14)
The condition z, 0 may be dropped since both sides of (5.14) are z,, continuous on aD. If m > 1, then a(w,wmzawa+ydw[l, m] A dz)
_ (a1 + Y, + 1)z"w*+rwmdw[m] A dz + (-1)m(am + Ym + 1)zaw"+yw,dw[1] A dz, and so (5.14) implies
Pw = aw,z"w"+ydw[1] A dz + aµ,
where a
a constant, and it a linear combination of the forms W,WmZ"w"+ydw[l, m] A dz. Further, in view of (5.13), is
Pw=
((27ri)"
az,w,z"w"+&w+aµ=P,w+aµ.
Hence P1(z, w) E O(aD), and it follows from (5.7) that a = 0. Y. We now consider examples of classes of domains for which there exists a vector-valued function w(z), z E aD, such that (5.1) and (5.8) hold and the form w(w, z) is nondegenerate.
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
42
EXAMPLE 1. Let D = (z: p(z) < 0) be a strictly pseudoconvex bounded domain with C2 boundary and containing zero; suppose that D is also linearly convex (Aizenberg [1]), which in this case means that, for every z E aD, the z1)p=, + . +((n - zn)p = 0) does not inanalytic tangent plane tersect D. We set w; =
P=,
(5.15)
ZIP;'+...+Zxpz
For this choice of wi, (5.1) holds. Let us consider the form to formed from the w; as in 1 °. LEMMA 5.8. w(w, z) is nondegenerate on aD.
PROOF. We represent aD as the union of disjoint sets 1'1,. .. , rn such that p' + 0 on Fm. Then we see by an easy computation that, on Fm, (-1)m-Ie(p)dz
w=
[m]
p1(z1Pr,+...+Znppn)
n
A dz,
where L(p) is the generalization of the Levi determinant (see Fuks [1], Chapter II, §12.3, or Vladimirov [1], Chapter III, §18.5) to the case of n variables (see Rizza [ 1 ]): 0 pz7
PZ
PZ r
... ...
PZ
pziI
E(P) _ PZ^ I
P"'If.
...
P"'J.
It remains to show that (?(p) + 0 on aD for strictly pseudoconvex domains D. But this can be easily checked by making a nonsingular linear change of variables such that the analytic tangent plane to D at a given point of aD is parallel to a coordinate hyperplane, and using the following property of L(p): if z = is a biholomorphic map, and p(z( )), then 2
E(PI) _ 40 aaI,.." "
(5.16) I
The proof of (5.16) in the general case is the same as in the case n = 2 (see Fuks [1], Chapter II, §12.3).
Let h be the locus of (z, w) in C2n as z ranges over aD. As in the proof of Lemma 5.8 (by means of the same linear change of variables), it is easy to prove LEMMA 5.9. h is a smooth manifold without complex tangent vectors.
§5. POLYNOMIALS ORTHOGONAL TO HOLOMORPHIC FUNCTIONS
43
Let b = (w: (w, z)z1 for all z E D) be the compact set dual to D, and let D be the dual of D. D is an open set (see Aizenberg [I ] and [2]). Let us also require that D be connected.(2) Then D is linearly convex in the sense of Martineau [1]. LEMMA 5.10.
h = (D XD) fl {(z,w): (z,w)= 1),
(5.17)
where h is a polynomially convex compact set.
PROOF. D may be interpreted as the set of complex direction vectors w of
analytic hyperplanes of the form a,,, = (z: (w, z) = I) that do not meet D. Hence the right side of (5.17) is the set of pairs (z, w) in C2n such that z E aD
and aw passes through z but does not meet D. From the uniqueness of the analytic tangent plane, it follows that w has the form (5.15). Thus (5.17) holds.
D and D are polynomially convex, since D and D are connected (see, for example, Makarova, Kudaiberganov and Cherkashin [1]); hence (5.17) implies the polynomial convexity of h. In view of Lemma 5.9, continuous functions on h are uniformly approximable by holomorphic functions (Harvey and (Wells [1]). These in turn can be approximated by polynomials P(z, w), because of Lemma 5.10. Thus condition (5.8) is satisfied. EXAMPLE 2. Now, in contrast to (5.15), we set w/
=ij1j z I2,
j = 1,...,n.
(5.18)
For this choice of the wj, the vector-valued function w(z) does not depend on D, and always satisfies (5.1). Let us consider conditions under which (5.8) holds. Let D be a bounded domain with 0 E D. Let B denote the uniformly closed algebra of functions on aD generated by z,,.. . , z" and w1,. .. , w,,, and let B. be its restriction to the (complex) one-dimensional analytic plane a, 0 E a.
Clearly, B separates points on aD and contains 1. Let a = (z: z = at = (a t t, ...
, a"t ), t E C'}, a E C". Then wj = aj I a I-2 t -', so that B. is generated by t and t-1. LEMMA 5.11. For (5.8) to hold, it is necessary and sufficient that, for every a, the set aD fl a have no interior points and that its complement in a consist of two connected components.
For the proof, we need the following results of Cirka [1]. Let X be a compact set in C", a subalgebra of C(X) and F
f.).
,r a family
of real-valued functions in C(X ). Then a continuous function f on X can be (2) It is not clear whether the connectedness of D is a consequence of the hypotheses of Example 1.
44
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
approximated uniformly on X by polynomials in elements of A and F if and only if
f can be approximated on every Xa = (x E X: fa(X) = as), as E R, by elements of A. PROOF OF LEMMA 5.11. Consider the functions f k = a jzk I Z I-2, j, k
f k = fk j' f k E B. Then fjk + fkj = 2 Re f k E B. These functions do not separate points lying in the same plane. By taking the restriction of f k to two planes a, and a2 and considering the functions f k it is easy to see that Re f k and Im f k separate points of different planes. Thus the set (z E aD: Re f k = bjk, in f k = cjk) is either empty or coincides with aD fl a for some a.
Sufficiency. If aD fl a satisfies the conditions of the lemma, then, by a theorem of Mergeljan [1], polynomials in t and t-' are dense in C(aD fl a). Hence tirka's result implies that B = C(aD). Necessity. If B = C(aD), then
B. = C(aD fl a). (5.19) In the interior of aD fl a, each function in Ba, being a uniform limit of holomorphic functions, is holomorphic; hence (5.19) can hold only if aD fl a
has no interior points. If the complement of aD fl a had a component w containing neither 0 nor oo, then for to E w we would have (t - to)-' E C(aD fl a); but (t - to)-' E B. by the maximum modulus principle. It remains to consider the question of the nondegeneracy of the form w for the choice (5.18) of w(z). LEMMA 5.12. Let D be a bounded domain with smooth boundary. Then w(w, z) is nondegenerace on aD if and only if, for all z E aD
(5.20)
i.e. no analytic tangent plane passes through 0 E D. The proof reduces to simple computation. For instance, if at z E aD we have p2^ + 0, then w(w, z) = I z112n
2n ,
k
da [k] A dz
_ -2 lzlp 1 + ... IZI
A dz.
p=,,
Thus the nondegeneracy of w is equivalent to (5.20).
Thus all the required conditions of 1 ° and 2° are satisfied by the function w(z) of (5.18), provided D satisfies the conditions of Lemmas 5.11 and 5.12. EXAMPLE 3. If the class of domains D is the same as in Examples 1 or 2, but p E C"+2, then, by Theorem 5.4 of Hdrmander and Wermer [1], (5.8) is valid
§6. THE CASE OF STRICTLY PSEUDOCONVEX DOMAINS
45
for polynomials in z and v, where v = (v,, ... , v" ), each v, being a C"+' function sufficiently close to w, in the C2 topology, i = 1,...,n. Thus it remains only to choose the vi such that the analogue of (5.1) holds and w(v, z) is nondegenerate on 8D. EXAMPLE 4. Let D be a domain of the class of Example 1, and let w; E C' as in Example 3. We extend the wi to some open neighborhood V of aD so that
(5.1) is preserved. Consider homeomorphisms 4p, CC"" of aD onto the boundary aD1 C V of a domain D, depending continuously in the C2 topology
on the parameter 1, 0 < I < 1, such that lim,..o q), in the C2 topology is the identity map of D. Again applying Theorem 5.4 of Hormander and Wermer, we see that, for e > 0 sufficiently small, the polynomials P(z, w) have the property (5.8) on aD1, 0 < t < E. Thus the w; of (5.15) work not only for the domains of Example 1, but also for domains which are close to them in the above sense.
§6. Forms orthogonal to holomorphic forms: the case of strictly pseudoconvex domains
1°. We now proceed to the characterization) of forms orthogonal to holomorphic functions, i.e. we shall answer the following question: Let D be a bounded domain with smooth boundary in C". Then which forms a E C,, _,)(8D) are orthogonal to functions in A"-P(D) in the sense that
fµAa=0
ao
_
for ally E A,,-P(D)? Let us denote the space of these orthogonal forms by
As already observed, in the case n = 1, for any domain D, we have a E A i P(D) if and only if a E Z(1P,oP) (1 CP,o(D ), p = 0, 1.
It is easy to give a sufficient condition for a form a to belong to A P(D). Indeed, if a = Y lap for some y E Z(P ,,_,)(D), then a E A P(D), since
f IAAa= f d(µAY)=0 aD
D
by Stokes' theorem, for any µ E A"_P(D). By a standard application of the Hahn-Banach theorem, it can be seen from is weakly dense in the space of forms (and even Theorem 2.1 that measures) from ACI(D) in the topology of C*(aD), the dual space of C(aD).
2°. For strictly pseudoconvex domains, we can show that the condition stated in 10 is not only sufficient, but also necessary. More precisely, we have
46
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
THEOREM 6.1. If D is a strictly pseudoconvex domain in C", n > 1, with 1, and if a E m then a E aD E if and only if there exists y E Cp "-Z(D) such that ay IaD = a.
Theorem 6.1, Part II of Theorem 4.3 and Part III of Theorem 4.3 for q = n are essentially equivalent. PROOF. Sufficiency was proved in 1 °. Necessity. By Theorem 2.12,
(6.1) a = a, IaD - a2 IaD' where a, E Z_ ,)(D) and a2 E I)(CD). If we extend a2 to a form I)(C"), then y = aa2 satisfies the conditions of Theorem 4.3: a2 E
fµnaa2=f µ/a2=0, D
D
aD
since a2 IaD = a - a, IaD, and a, a1 IaD E By Part III of Theorem 4.3, a2 = a2 = afl on CD, and $ E C(P.n_2)(C"). By Theorem 4.4, a, = ay1 and yI E C(p,n_2)(D). Now (6.1) yields that y = YI - Q has the desired properties. Theorem 6.1 cannot be carried over to arbitrary domains because it is not true that all 5-closed forms (which, as shown in 1 °, are orthogonal to holomorphic forms) are a-exact for every domain. Thus the following result seems natural: THEOREM 6.2. Let D = 12 \ U2,), where fl and the fli are strictly pseudo1, such that ii cc 2 and Si, n flj = 0 for i = j; let a E C',,",- I)(aD). Then a E An P(D) if and only if there exists convex domains with Cni+2 boundary, m
aE
,)(D) such that & (aD = U.
The "exotic" form of the domain in which such a characterization is given reflects to a certain extent (and for n = 2 almost completely) the true nature of things, as Theorem 8.1 shows.
PROOF. Let a E A P(D). By Theorem 2.12, a = a, IaD - a2 IaD' where al E and a2 E ZZP','_,)(CD). a2 breaks up into forms ak a° E Z'" a0, a1 a E C(P,n-2)(li). Extend the y, to C" so that k
supp Y, Cs
Sl\ U SZj, j=1
j#i and denote the extended forms also by y,. Consider $ = a2 - (ayl + +aYk). Clearly, $ = 0 on Sli, i = 1,...,k, and f = a° on Cfl. Further,
§7. THE GENERAL CASE
47
And since $IaD E A _ (D), since 8y, IaD and a2 Ian belong to A _ (S2). By Theorem 6.1, a° IaD = A,-P(U) = 1)(Q). Let qp E C°°(C"), p = 1 on CO and qp ° 0 on aYo 61 where yo E UI S2,. Then $ I3D = a(py0) IaD. Set k
it = a1 - 2 aY; - a(opYo) i=1
Then a E
1)(D), and k
(aTIIaD+a(PYo) IaD) = a.
IaD = a1 Ian
i= I
For k = 1, Theorem 6.2 admits an alternative formulation: COROLLARY 6.3. Let D1 C4 D2 be strictly pseudoconvex domains with Cm+2
boundaries, m > 1, and let a; E
i = 1,2,.... Then the a; are
restrictions to the 8D1 of the same form a E Z
1j(D2\DI) if and only if
f D&Aa1-Ja µ/ i
a2
L`
for any µ E A,,-,(52)'
3°. Just as Parts II and III of Theorem 4.3 for q = n yield Theorem 6.1, Parts I and III for q < n yield THEOREM 6.4. If D is a strictly pseudoconvex domain in C", n > 2, with Cm+2
boundary, m > 1, and if a E ZZ p Q)(CD), 1 4 q < n - 2, then there exists y E C(p,4_ 1)(C") such that 8y = a on CD.
This theorem is an analogue of Hartogs' theorem on the extension of holomorphic functions from the exterior of a compact set. §7. The general case
For the characterization of forms orthogonal to holomorphic forms in the general case, we need the notion of the envelope of holomorphy of a closed bounded domain. If a compact set K can be written as n m Dm, Dm+1 C4 where each domain D. has a schlicht (i.e. univalent) envelope of holomorphy H(Dm), then we shall say that K has a schlicht envelope of holomorphy, and define H(K) as(1 m H(Dm ). The envelope of holomorphy H(K) so obtained does not depend on the choice of the sequence Dm, m = 1, 2,.. ., and preserves
a number of properties of the envelopes of holomorphy of domains (see Aizenberg [2]).
48
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
THEOREM 7.1. If aD E m > 1, a compact set D has a schlicht envelope of holomorphy, and y E C(p,n_I)(aD), then y E if and only if there exist y1 E Z, _ IU(D) and y2 E Bp,"_ (CD) fl C(P,n_ 1)(CD), 0 < A' < A, where y 2 = O( z 11-2n ) as Jz - oo and is a-exact in CHID), such that Cm+I.a,
(7.1)
y= )'IIaD-y2IaD
If p = n, then yI E B,,,,,- 1(D). PROOF. SUFFICIENCY. Consider a E A,,-P(D). There exists a bounded do-
main Q with smooth boundary such that H(D) C Q and a E An_p(Q). By assumption, y2 = 8Q in CH(D). Now, by Stokes' theorem,
f a A y = fa aAyi
- f aAy2
_
1)
aD
aD
aD
(_1)"-p-1 f Qd(a
=(_"-p f
D
aA8y1- f aA8$ aQ
A B) = 0.
Necessity. Using Theorem 2.12, we write y in the form (7.1). Obviously,
yI E A p(D); hence y2 E A p(D). We must show that y2 is a-exact in CH(D). Let H(D) = n Qm, where Qm+ I C Qm and the Qm are strictly pseudoconvex with C°° boundary. If a E A,,-,(Q.), then, since y2 is a-closed, we have by Stokes' theorem
f
aQ.
aAy2-f
aAy2=0,
aD
i.e. y2 E An p(Qm) for all m. As in the proof of Theorem 6.1, we get that for every m there exists S. E Cp "_2(C") such that y2 = asm in CQm. By induction we shall now construct a. E C(p,n_2)(CQm), m = 1,2,..., such that 8s,;, = y2 and Sm = Sm+ on CQm. We put S; = S1. Suppose 8,, is already defined, and om = Sm+I - 8,,,. Then aem = asm+I - asm = 0 on CQm. By Theorem 6.4, om can be extended to all of
C" as a form in Z(p,x_2 (C"); denote this extension also by om and set S,+I = Sm+I - 9m. Then Sm+I = aSm+I - aom = y2 on CQm+I, and s,+I Sm=Sm+I
0m-8m=00nCQm.
If we define s as equal to Sm on CQm, then a E C( p,n_2)(CH(D)) and
as=y2. The question arises whether the second term on the right of (7.1) is necessary, or whether in the general case, as in the strictly pseudoconvex case, the characterizations of forms in may be confined to that of forms
with a-closed extensions into D. From the results of §8, it follows that the second term in (7.1) is in general necessary.
49
§8. CONVERSE THEOREMS
§8. Converse theorems
It is natural to ask whether there exist domains for which it is impossible to characterize forms in A (D) in a manner similar to Theorem 6.2, and if such
domains do exist, then to describe the class of domains for which the characterization is possible. For domains in C2, the answer is provided by the following result. THEOREM 8.1. Let D be a domain in C2 with C2 boundary, and suppose that, such that a 13D = a. Then
for every a E A' (D) n Ct2,)(aD), there exists it E k
D = w U iii, i
where Sl and the fli are domains of holomorphy, Sli C4 S2, and S2, n Sly _ 0 for
ij.
PROOF. Since D is a bounded domain with boundary of class C2, we have
D = SZ\ U; Sli, where Sl and the Sli are domains with C2 boundary and connected complements, with Sli C Sl and Sl, n SZj = 0. 1. Suppose Sl is not a domain of holomorphy. Then there exists z° E aD at which the Levi form is negative definite; hence there is a neighborhood U of z°
such that (z E U: F(z, z°) = 0)\z° C fl, where 2
2
F(z, z°)
7 i=1
a
.
(z°)(zi - zO) + 2
.i°'
i
as
Jz°)(zi - z0)lzi
az,
-
(see, for example, Gunning and Rossi [1], Chapter IX, §B). Let (p E C°°(C2), and supp 9) C U and q) = 1 in a neighborhood of z°. Put
g(z) _ q(z)[F(z, z°)]-'
(8.1)
for z E CSl. Then 8g E Z(o,1)(02), since ag = 0 at z (4 U U Sl and ag = ap(z)[F(z, z°)]-l for z E U\9. Further, if
a= 1agAdzI., 0
on ail, on aSli, i
then a E A' (D). Indeed, a satisfies the conditions of Theorem 7.1, since
a=a,laD-a2IaD,wherea,
O, anda2agAdzon Cu and0ontheili.
By assumption there is an it E Z(2,,)(D) such that a I8D = a.
50
H. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
By (1.1) applied to B\SZ, where B is any ball such that U U SZ Cc B, we have for z E B\SZ
ag n dz = f [ac B\n));
5g(e) n d n
-5f
z)
5g(e) A d A
z).
(8.2)
(B\2)r
By the same formula applied to D, we have for z E B\SZ
z) -
0 = faD)ra(U) A
z).
A
(8.3)
The first integral of (8.2) and that of (8.3) are negatives of each other, since ag Iaa = a jag. = 0 and d Ian = ag A dz Ian. Hence by adding (8.2) and (8.3) we get ag A dz = ah A dz, where h is defined by
h(z) A dz = -f
-f a A Dr
B\D)rag((') A d n
z)
U2,0Q, z).
h(z) is continuous in C2, since the integrands in (8.4) have integrable singularities (see (1.4)). Now g - h is holomorphic in B\Sl, since a(g - h) A dz = 0, so that g - h extends holomorphically into B. Therefore g extends continuously to z0.On the other hand,
lira lg(z)I = oo. Z-zo
This contradiction shows that the assumption that SZ is not a domain of holomorphy is false. 2. Suppose now that some SZ, is not a domain of holomorphy. Then there is a ball B such that all functions of A(S21) extend holomorphically to B, and B\S1; is not empty. Let ° E B\SZ;. Without loss of generality, we may suppose that
0=0andBnS2,A(zEC2:z2=0) of =
0-
U2,1(01 Z) Iaa;
on aSt
0
on ag a ndSZj,j
i.
§8. CONVERSE THEOREMS
51
Exactly as in step I above, we see that a E Al (D). Hence there is an a` E Z(21)(D) such that a IaD = a. By (1.1) applied to yyS2,, for z E SZi we get fad,);U2.1(0, J) A U2,1(J, z)
U2,1(O1 Z) =
z).
U2,1(0, f) A
(8.5)
By the same formula applied to D, for z E fl we get
O=f
a
aDj
D
z).
A
(8.6)
Again as in step 1, we see that the first integrals of (8.5) and (8.6) are negatives of each other. Hence, if we add them together, (8.5) and (8.6) yield U2,1(0, z) = ah A dz, with h defined by
h(z) A dz = - f )rU2,1(0,') A U20&, z)
- fD;a(') A U2,0(t, z). (8.7)
By Corollary 2.4, the first integral on the right above is of class C(2,0)(0i) and
the second integral is proper for z E fli, so that it also lies in C(2 0)(SZ), i.e. h E C°°(fi). On the other hand,
U21(0 Z= )
1
a1dz2 - z2d21
Iz
(21ri )2
zldz
A dz = -
14
(27ri )2 I Z I2Z2
for z2 0; hence h - z1((27ri)2 I Z I2 z2)-' is holomorphic in D\(z: z2 = 0). Then g = z2h - z1((27ri)2 I Z is holomorphic in D, and so it extends I2)-1
holomorphically to B. But g(z1,0) = z1' and B n (z2 = 0) is connected; hence g cannot be extended to 0 E B. The theorem is proved in full. COROLLARY 8.2. Let D be a domain in C2 with C2 boundary, and suppose that,
for each a E Al (D) n C(21)(8D), there exists y E C(20)(D) such that ay E Z('2.,)(5) and ay IaD = a. Then D is a domain of holomorphy. PROOF. By Theorem 8.1,
D=SZ\ U SZi, i=1
is a domain of holomorphy and k is the number of bounded components of CD. The corollary will be proved if we show that k = 0. where SZ
Suppose that k 0, and let ° lie in a bounded component of CD, and let t be the boundary of this component. Then U2,1(s°, ) E Z(21)(D) by Lemma 1.2;
52
II. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
hence U2.1U°1 0 IaD E A1(D). By hypothesis, we therefore have a y E Ct20j(D) such that U2,1(U°, z) = ay IaD. But then by the Martinelli-Bochner formula ((1.1) for p = q = 0) and Stokes' theorem we get
1=f
f) = fray = 0.
This contradiction proves the corollary.
CHAPTER III
PROPERTIES OF a-CLOSED FORMS
OF TYPE (p, n - 1)
§9. The theorems of Runge and Morera
The classical theorem of Runge asserts that a domain in C' is polynomially
convex if and only if it is simply connected. Wermer (see Wermer [1] or Gunning and Rossi [1]) has constructed an example of a domain which is biholomorphically equivalent to the tricylinder but not polynomially convex. Thus there are no necessary and sufficient topological conditions on a pair of domains in C" which ensure that homolorphic functions in the smaller domain
can be approximated by functions which are holomorphic in the larger domain. For a-closed forms of type (p, n - 1), the situation is exactly the same as that for holomorphic functions of one complex variable. THEOREM 9.1 (Runge's theorem for forms). Let Sgt C SZ2 be open sets in C". Then the following conditions are equivalent: 1. Every form lying in Z("np."_,)(S2) can be approximated in C(,n,,,_ 1)(91) by forms lying in Z(p,"_,)(St2).
2. If S22\2, = K n F with K compact and F closed in 02, and if K fl F = 0,
then K = 0.(') PROOF. I --> 2. Suppose SZ2\SZ, = K U F, with F closed in SZ2, K compact,
K fl F = 0 but K 0 0. Pick ° E K. Then A dot A ... a=
E Zip,"-p(ll)
By condition 1 we can approximate a in Cip,"_,)(21) by the ak E Z(p,,,_,)(Sl2). Let D be an open set, consisting of finitely many connected components with (')Condition 2 means precisely that each bounded component of Cf2, meets C2253
III. a-CLOSED FORMS OF TYPE (p, n - 1)
54
smooth boundary, such that K C D C 02\F. Then aD C 521, and by Stokes' theorem
f
aAdip+1A...Aden=
lim
k-.oo
aD
f
aD
= lim f 8ak A dip+I A
. Add" = 0.
D
On the other hand, by the definition of Uo,o and the Martinelli-Bochner formula (see (1.1)), we have
faDa A
p
A
... Aden = (277i)"
(n - 1)!
f
(2iri)n UO.o(J, m _ (n - 1)! aD
This contradiction proves that I - 2. We shall prove the implication 2 - 1 first only for m = oo; the case of an arbitrary m will be handled later, using Theorem 9.3. By the Hahn-Banach theorem, it is sufficient to prove that any continuous linear function Ton C(p,rt-1)(9 I) which vanishes on Z(0,0,,,- )(f 2) vanishes on
Z(p,n-I)(9'I) Now, since T is linear and continuous on C('P,n_I)(S2i), which is the projective limit of the spaces C( pp,n_ I)(L), T can be extended to a continuous linear functional on C(P,,,_ I)(L) for some compact set L C 521 (see, for example, Gel'fand and Silov [1], Chapter I, §4.1). Here we may assume that CL has no
bounded connected components contained in 521; otherwise we need only adjoin to L all the bounded components of CL contained in SZI to obtain a new compact set which does satisfy the required condition. Put
w) = T (Up,n-1(J, z)),
r E CL.
(9.1)
For a C°° function g, Eg/Lxk tends to gzk uniformly on all compact sets; hence we may differentiate (9.1) with respect to
and use Lemma 1.2 to get
8 = T(atUp -1q, Z)) _ (-1)"Ti(azUpnz)) _
z))) = 0,
where q)t. E C°°(C") is equal to 1 in a neighborhood of L, and equal to 0 in a neighborhood of E CL. This means that is holomorphic on CL. For t E C522,
D1'0 = Tz(Dt Up,n-IG, z)) = 0, since
z), and hence all its derivatives, belong to Z(p,,,_I)(S22) (for
E C92). By the uniqueness theorem, ¢ = 0 in each component of CL
§9. THE THEOREMS OF RUNGE AND MORERA
55
meeting C02. Also, 0 = 0 in the unbounded component of CL as well. This is proved for n = 1 in Hdrmander [11 (§1.3). For n > 1, 4, can be holomorphically
extended from the unbounded component of CL to all of C" by Hartogs' theorem, and lim
0.
M--
From Liouville's theorem it follows that ¢ = 0 in the unbounded component of CL. Condition 2 of the theorem and the choice of L guarantee that CL has no other components, i.e. 4 = 0 on CL. Now let a E and let D be an open set with smooth boundary consisting of finitely many connected components L C D Cs 52,. Let T E C°°(C"), (p = I in a neighborhood U of L, and supp q) C D. Then, for z E U, (1.1) gives
a(z) = II,n_, (D, a) - 5IP,n-2(D, a) - J(aD);a( ) A Up n-i(
Z) - a(qJ2
a)).
Since 4P1, n_2(D, a)/ E C(,.n_2)(C") and 8D C CL, we deduce that
Ta = f aDr
z))
A
= f Da(y) A Thus 2
- T(a('PI2,n-2(D, a)))
0.
1 is proved (for m = oo).
As in the case of one variable (see Hormander [1), §1.3), we have, in particular, for SZ2 = C" COROLLARY 9.2. Z('P,"_ I)(C") is dense in Z('nP."_ 1)(f) if and only if SZ has connected complement in the one point compactification of C".
THEOREM 9.3. Let Sl be an open set in C" and a E C(p.n_)(S2), m > 0. Then the following conditions are equivalent: 2 and every 1. For every ball B
I = (i1,...,in_p), 1
f a A d('r = 0.
(9.2)
as
2. a can be approximated in C(p,n_ I)(SZ) by forms from Z(p,n_ 1)(SZ). 3. a E Z(p,n_ I)(SZ).
Weierstrass' theorem on the limit of a sequence of holomorphic functions shows that, for n = 1, Theorem 9.3 is precisely Morera's theorem. For n > 1, a
56
in. a-CLOSED FORMS OF TYPE (p, n - 1)
form satisfying the equivalent conditions of Theorem 9.3 can be nondifferentiable. As an example of such a form we may take q)(z,)dz[n], where p(z,) is a
continuous but not differentiable function on C'. Thus it is impossible to obtain a theorem in the same formulation as Morera's theorem for n = 1. PROOF OF THEOREM 9.3. We first introduce some notation.
If g is locally integrable in Sl and f E C°°(C") with compact support L, then for all z E Q such that z - L C SZ (2) we set
(g*f)(z)=
ff g(z -
A
Cr.
2
K exp -r2 r IzIZ
forI zj
0
forI zr
Pr(z) -
(9.3)
(K is chosen so that f pr d A d = 1). If n
akjdz[k] AdZJ E C(p,n-1)(0),
a J k=1 then
7, ar(z) =
n
I (akJ * Pr)(Z)dZ [k] Adz,. J k=1
It is known (see, for example, Bremermann [1], §3.2, Lemma 2) that ar e C(-P,,,- ,)(Or), where W = (z E 11: p(z, asa) > r), and that ar converges to a in C(p,n-)(2) as r --> oo. We first prove that I - 2. Let B C4 12, and let the multi-index I' be such that dz,, A dz, = dz. Then fBaar A dz, = f Bar A dz, n
f7 fCakl'(Z ; Br
A 4 A da [k] A dz
1
=1 (f,Ba(z z
f ^( f C
(aB-t>=
/
(2) z - L denotes the vector difference, i.e. the set ( E C": = z - w, w E L).
§10. THE FIRST COUSIN PROBLEM
57
Hence aa, A dz, = 0 in SZr. Since the multi-index I is arbitrary, it follows that aar = 0. Now it is easy to see that the open sets 2r, SZ satisfy condition 2 of Theorem 9.1. Thus we may apply Theorem 9.1 with m = oo to conclude that 2.
I
For m > 0, the implication 3 -> I is an immediate consequence of Stokes' theorem. So, let m = 0 and B C St. Then B C W for sufficiently small r, since B is compact, and so
f aAdz,=lim f a,. A r--.0
aB
f aa,Adz,
aB
B n
= lim
(-1)k-1( fC
fB:
`
k=I1
aak Pr(z
- (')d / d dz n dz,
I
where I' is as before. Using the equality [pr(z - Zk = -[Pr(z - )]- and the definition of a derivative in the space of distributions, we get n
f da A dzf f a A dz, = lim r-'0 aB
B,
=lim f df r-.0 Ct
Cf k=1
1)k-
8
4k
()df / d
5a(z)pr(z-)Adz,=0. B,
This proves 3 -+ 1. The implication 2
3 is trivial for m > 1, and follows for m = 0 from the fact that differentiation is a continuous operator in the space of distributions. CONCLUSIONS OF THE PROOF OF THEOREM 9.1. By Theorem 9.3, every
a E Z_ 1)(f21) can be approximated in C(P,,,_1)(SZ1) by forms from Zip Hence Theorem 9.1 for the case m = oo can be applied to deduce the theorem in the general case.
§10. The fist Cousin problem, separation of singularities and domains of existence
For forms of type (p, n - 1), we can pose the following problem, analogous to the Mittag-Leffler problem: Let S2 be an open set in C" and (z') a discrete sequence of distinct points in St.
Suppose forms a; E Z('P,"_ 1)(SZ\(z')) are given. Does there exist an a E Z(p "_ 1)(S2 \ (z')) such that a - a, E Z(p,"_ 1 )(U) for some neighborhood U of
z'?
III. d-CLOSED FORMS OF TYPE (p, n - 1)
58
As in the case of functions, this problem can be generalized to the following one, which it is natural to call the first Cousin problem for forms: Let 52 be an open set in C", and (ff,) an open covering of S2. Suppose forms a;j E z( -P,,,- 1)(fl, n S2J) are given, satisfying the conditions
a;j+aai=O onQ nSZJ, a;J+aJk+aki=O onS2;nSlJnSZk for all i, j and k. Can we find forms a; E Zip rc_ 1)(S21) such that a. = aJ - a, on a, n S2J? The forms a,j are called the Cousin data, and the a, (if they exist) a solution of the first Cousin problem.
LEMMA 10.1. For any Q E Coo ")(S2), the equation ay = (3 has a solution YE
C(p.n-q(2).
PROOF. Let {rk} be a sequence of positive numbers decreasing to 0, such that 0, k = 1, 2,..., and let ¢k E C°°(C") be equal to 1 on Gk = SZ'k n B(0, k)
Gk+I and supp 4k C Sl. Put q'I = 4'I, and q ='k - 'Pk- I for k > 1. Then on S2.By (1.1)we have ayk=q'k
pk=OonGkand9) where Yk
f 9k$ n Up,n-1G, Z).
By Corollary 2.4, yk E C(o,n_ 1)(S2). Since IN 0 on Gk, k > 1, we get Yk E Zp.n_ 1)(Gk). As we observed in the proof of Theorem 9.3, 9'k and S2 satisfy condition 2 of Theorem 9.1; hence so do Gk and 2. Thus we can choose Xk E Zip ,r_ 1)(10), k > 1, such that the differences between the corresponding coefficients of yk and Xk, and all their derivatives of order up to min(m, k), are less (in absolute value) than 2-k on Gk_ I since Gk_ I C4 Gk. Then the series 00
Y=YI+ I (Yk - Xk) k=2
together with all its derivatives of order up to m converges uniformly in the interior of 2 to the form y, so that y E C(p,n_ 1)(9). By termwise differentiation, we get p
00
ay = (PIN + I q)k# = k=2
N pp
THEOREM 10.2. If the Cousin data are of class C"', m > 0, then the first Cousin problem has a solution which is of class C"'.
§10. THE FIRST COUSIN PROBLEM
59
Note that for the solvability, with any data, of the Cousin problem for holomorphic functions of several complex variables, it is necessary and sufficient that the first Dolbeault cohomology group be trivial. Therefore Theorem 10.2 is not valid as it stands in the case of holomorphic functions of several variables. PROOF OF THEOREM 10.2. Let (9),) be a partition of unity subordinate to the
covering (w,), i.e. the sequences (qp,) of functions and {i,} of integers satisfy the following conditions: 1. p, E C0(C") and supp p, C S2i,. 2. On any compact subset of 9, all but finitely many of the (v, vanish. 3. Zoo q),=lon fl. Put hk = 2i Tlai k; then hk E C(p,"_ 1)(Slk). Furthermore, 00
hk - hi = 1=1
00
cI(ai,k - aril) = G glajk = ask, 1=1
and so the form /1 which is equal to 8hk on Qk is well-defined. And /3 E C(p.")(Sl), since
Q6.=8hk=
41'Aa;,k.
1=1
By Lemma 10.1, there exists y E C(p,"_1)(Sl) such that 8y
The forms
ak=hk-yarethedesired ones, since aak=8hk-8y=0 andak - aj = hk
-y - h - +y=ask. In the special case when the covering consists of two sets, we have COROLLARY 10.3. Let 12, S21 and 02 be open sets in C", with 12 = 121 n 522.
Then, for any a E Z(p,"_ I)(2), there exist a. E Z( 'P,,,- 1)(0i), i = 1, 2, such that
a=a1-a2 on a. That is, for 5-closed (p, n - 1)-forms, the theorem on separation of singularities is valid for any triple of domains, whereas for holomorphic functions of several complex variables this is not always the case. THEOREM 10.4. For any domain D C C" and any p, 0 s p < n, there exists a form a E Bp,"_ 1(D) which cannot be continued across 8D as a 8-closed form. PROOF. Let (
) be a countable dense subset of 8D. Consider the function
_O0 G(z)
k
1
1
1
k2 I'"k - z I2n-2
(10.1)
If K C D is compact, then z I> r> 0 for E 8D, z E =-K and an r independent of and z. Therefore, the series (10.1) is majorized by the
in. 3-CLOSED FORMS OF TYPE (p, n - 1)
60
convergent series 1
°°
1
2n-2 k=1 k 2 r
Hence (10.1) converges uniformly in the interior of D to a harmonic function, and lim G(z) = coo. z-z°EaD
(10.2)
Clearly, the form
a=2' 2 aGdZ[k]Adzj, azk J k= I
has the desired properties (here J = (j1,. .. ,j,)). Indeed, suppose that a can be
extended continuously to a neighborhood of a point z° E D. Then all the functions GZk, and hence also all the G=k = (Gzk ),( 3) can be extended continu-
ously to a neighborhood of z°. But then G itself can be extended smoothly to this neighborhood, which is impossible in view of (10.2). §11. Theorems of approximation on compact sets
a-closed (p, n - 1)-forms enjoy approximation properties analogous to those of holomorphic functions of one complex variable on compact sets just as well as on open sets: THEOREM 11.1. Let K be a compact set in C", and a E C(p,n_1)(K), m % 1.
Suppose that a has an extension to a neighborhood of K such that as and all derivatives of order up to m - I vanish on K. Then a can be appoximated in Cpn-1)(K) by forms which are C°° and 5-closed in a neighborhood of K. If in addition CK is connected, then the approximating forms may be chosen in Z(p,n-1)(Cn).
If formulated in the same way as above for holomorphic functions of several complex variables, Theorem 11.1 would no longer be true (Fuks [2], Chapter III, §16.3). PROOF. Let D be a bounded open set, consisting of finitely many connected components with smooth boundary, such that K C D and a E C(p n_ 1)(D). We represent a as an integral by means of (1.1). By Lemma 1.2,
Ip,n-1(D, a) E Z(-P..- 1)(D). (3) This equality holds because G is real-valued.
61
§11. APPROXIMATION ON COMPACT SETS
By Corollary 2.4,
Ip.n-2(D, a) E C(p.n-2)(D),
so we can approximate a) in the topology of Cp _2(D) by forms lying in C(p.n-2). Then aI2n_2(a) is approximated in C(p.n1 1)(D) by forms lying in Z(p n_1)(D).
To prove the theorem, it remains to approximate IP,n-1(D, aa). Now as has the form CM-1
E'hJda A dz1, J
Here the hJ vanish with all derivatives up to order m - 1 on K; hence they can
be approximated in C"(D) by functions h' E C°°(D) which vanish in a neighborhood of K (see, for example, Malgrange [1], Chapter I, Lemma 4.2). Then
a1= I
,n_
(D,
A
dt)
E
1)(D)
J
The at are a-closed in a neighborhood of K (since the hi vanish in a neighborhood of K), and in this neighborhood we may differentiate the integral defining the operator Ipn_ I under the integral sign.
We shall prove by induction on m that the sequence (a') converges in to a° - Ip n_ 1(D, aa). First let m = 1. For each J, (h') converges uniformly on D to hJ, i.e., given e > 0, we can find an N such that I h' - hJ I < e on D for 1 > N. Let akJ denote the coefficient of da[k] A dzJ in a', l = 0, 1,.... Then (see (1.14)) C(p.n-n1
akJ(Z)
fD;hJ(t)(-1)k-1
n
(nl)1 IT
hJ(t))(-1)k-1
do
and dv denotes Lebesgue measure.
A dt,
-1,
(11.1)
Jk -z
I
Ce, Ip-z12n-1
fD
j
I2n
N
'r
'ff
C=
JJ>
(n - I 'f (hJ(t) (21ri) e(n - 1)t.
where
rkz
for t = 0,1,... (h,° = h1). Hence for z E K and 1
IakJ(Z) - akJ(Z)I
I
((2zri))!
U (z + D), zEK
k df A dt
III. a-CLOSED FORMS OF TYPE (p, n - 1)
62
Suppose now that our assertion has been proved for m = s. We proceed to prove it for m = s + 1. Note that hJ = 0 in a neighborhood of K, and that the kernel in (11.1) is a function of - z. Hence, by Stokes' theorem, for z E K we have aZt
akJ(Z) = fat
a, hJ(
- f D;h'i
)(-1)k-I I
I
Yk- ZZIZ"
dAd
fk- k df A d('[i]. z
(11.2)
A similar formula is also valid for the derivatives with respect to the a,. The
sequence formed by the second integral in (11.2) converges, with all its derivatives, uniformly on K; hence we see by means of the induction hypothe-
sis that the sequence {(akj)z.} converges in CS-'(K). Therefore {akJ} converges in C5(K) to akJ, and the first assertion of the theorem is proved. Now, if CK is connected, then we may also suppose that the neighborhoods
of K in which the approximating forms are defined also have connected complements. Then the second assertion follows from Theorem 9.1. We shall call a domain D in C" admissible if, for every z E OD, there exists a neighborhood UZ of z and an a E C" such that
DnUZ+ea cD
(11.3)
for0<E<1. All domains whose boundaries have continuously varying normals are admissible; in particular, all domains with smooth boundaries are admissible.
For an admissible domain D, we define a function r(z) on aD by r(z) = sup (r > 0: there exists a E C" such that (11.3) holds for a and UZ = B(z, r)). By the definition of an admissible domain, r(z) > 0. It is not hard to show that r(z) is lower semicontinuous.
THEOREM 11.2. Let D be an admissible bounded domain, and let a E Then a can be approximated in C(p,n-I)(D) by infinitely differentiable 5-closed forms defined in a neighborhood of D. Further, if a E ZAP ,,- )(D), then the approximating sequence converges to a Z(p,n-I)(D).
k
For the proof, we need
LEMMA 11.3. Let D = D, n D2 (D, DI and D2 are domains in C") and (D,\ 5) n (D2\D) = 0. Then any a E Z(,_ I)(D) can be represented in the
forma=a, -a2,Where a; EZ(pn_,)(D,),i= 1,2. If, moreover, a E C(v,"_), then a; (=- Ck n_ I)(D,), i = 1, 2.
63
§11. APPROXIMATION ON COMPACT SETS
PROOF. The hypothesis of the lemma implies that D = D1 fl D2. Choose (p1 and (P2 in C00(C") so that q21 - 9)2 1 and supp q>; fl D;\D = 0. (4p, and 4'2 exist. Indeed, choose neighborhoods U1 and U2 of D 1\ D and D2\D such that
U1 fl U2 = 0, and functions ¢i such that i'; = 0 on U and 4; > 0 on CU;, ¢; E C°°(C"). The q)1 = 4,1(4Y1 +'P2)-' and q2 = - 2(¢1 + We define two forms
Q(z) _
agv' A a
Y(Z) = f
M'2)-' will serve.)
on D1,
a4'2Aa on D2, DiUDZ);N(J) A U,,,,
1(J, z).
is well-defined since ag21 = 42 on D1 fl DD = D, and we have /3 E CAP n) (DI U D2) and y E C(p,n_1)(DI U D2). We shall prove that 16 = ay.
(11.4)
Here, the differentiation is to be understood in the sense of distributions. Approximate # in C(p,n)(D1 U D2) by forms /3. E C('P,n)(D1 U D2), and set Y;(z) = f D'UD2)i$ (J) A Up,n 1(J, z). By (1.1) we have /3, = wy,, and since the Y; approximate Y in C(u,n_ 1)(D1 U D2),
(11.4) follows from the continuity of the differentiation operator in the space of distributions.
We set a;_p;a-y,i=1,2.Then a;EZ(p,n_1)(D;),since aa,=aq,Aa
-ay=$-$=0. Further, a1-a2=TIC' -Y-gp2a+Y=(p1-q2)a
a on D. If moreover a e C(p.n_1t(D), then /3 E Cfp n)(D1 U D2) by Corollary 2.4. Then y E CAP n_1)(D1 U D2), and so a; E Cep n_1)(D;). PROOF OF THEOREM 11.2. Let a E Z(p,n_ 1)(D). Then for every S > 0 there exist domains D;, D C D,, and forms a; E Z(p,n_ 1)(D; ), i = 1, ... ,1, such that
(11.5)
on D and diam(aD fl aD;) < S. We prove this as follows. Choose N such that (2.3-')Ndiam D < 8(2n )112. In the first step, put b = infzEaD Re zl, c= supzEaD Re zl and R = supZED I Z I + 1. Then the domains D,
D' =DU {z: IzI
D2 = D U (z: Izj < R, Rez1 > c - (c - b)/3) I
64
III. s-CLOSED FORMS OF TYPE (p, n - I)
satisfy the conditions of Lemma 11.3. Hence a = a1 + a2, where a' E Z(p
n _ 1)(D).
Observe that
-zI)l=3(c-b) <3diamD.
sup z, j'EaDnaD'
At the next step, we put
b. =
inf
Rez1,
sup
C, =
zEaDnaD'
Rez1,
zEaDnaD'
i = 1,2,
and apply Lemma 11.3 to the triples consisting of the domains D',
D" = D' U (z: IzI < R, Re z1 < b1 + (c. -
b.)/3)
and
Di2 = D' U {z: IzI < R, Rez1 > ci - (c; - b1)/3). We obtain forms a'j E Z(p,n_1)(D'J) with a = all + a12 + a21 + a22. Furthermore,
sup zjEaDnaD'"
sup
z1)l = 3
z1)I
z.3'EaDnaD'
(3 )2 diam D.
After N steps, we get 21 domains D"" * "m containing D and 2^' forms a...'N E Z(p n-1)(DhI"''N) such that
I 2
a=
a,,...,N
ii... iN= I
and
N
z1)l < (13
sup 'N
diam D <
S
2n
We repeat this process 2n - 1 more times, replacing Re zI first by Im z1, then by Re z2, and so on. We obtain domains D1,. .. , D, and forms a; E Z(p,n_I)(D,) satisfying (11.5), with 8D fl aDi contained in a cube of edge 8(2n)-1/2. Hence diam(aD fl aD) < S. We see from (11.5) that it suffices to prove the theorem for forms a E Z(p,n_ I)(G), where G is a domain such that D C G and the diameter 8 of aD fl aG is small. Since r(z) is lower semicontinuous and positive, there exists 0 > 0 such that r(z) > 0. Let S < 0, and fix a z° in aD fl aG. Then there exist
r > 8 and a E C" such that_D fl B(z°, r) + ea C D for 0 < e < 1. Since 8D fl aG C B(z°, r), we have D\B(z°, r) C4 D; hence sufficiently small translations do not carry it outside G. Hence a,(z) = a(z + ea) E Z(p,n_I)(UE(D))
§I1. APPROXIMATION ON COMPACT SETS
65
(UE(D) being a neighborhood of D) for sufficiently small e. As a - 0, the of converge to ain C(p."_I)(D). Applying Theorem 9.3 to the forms a,(z) and the domains UE(D), we thus obtain the first assertion of Theorem 11.2.
Suppose now that in addition a E C(p.n_I)(D). Then, by the second part of Lemma 11.3, we see that a; E C( "_I)(D;). Thus we need only note that aE tends to a in C(,,,,_ )(D) as e 0.P THEOREM 11.4. Suppose K is a compact set of 2n-dimensional measure zero in
C". Then the space of forms of type (p, n - 1) which are C' and 5-closed in a neighborhood of K is dense in C(p,n- 1)(K).
PROOF. By the Weierstrass approximation theorem, C(p,n- )(C") is dense in C(p,n- 1)(K). Let a E C ."_ 1)(C" ), and let B be a ball containing K. By (1.1),
a = Ip,"-1(B, a) - JP,"-1(B, aa) - 5IP."-2(B, a).
(11.6)
The first and third forms on the right above lie in Z(p "_I)(B). Choosing a sequence of domains Di such that Di C4 D,_ I C B and fl D. = K, we have
Ip,n-1(B, Ba)(z) = rlim f B\D')raa(J) A UP,"-1(J, z), since the 2n-dimensional measure of K is zero; the convergence is uniform in z E K. Now the theorem follows from the fact that
fB\Dt)raa(J) A Up,.-.(J, z) E Z(P,.-I)(Di) COROLLARY 11.5. If D is a bounded domain in C" with piecewise smooth boundary and z° an arbitrary point of D, then forms of the type a1 + a2, where a1 E Zip,"-I)(5) and a2 E Zip,"_1)(C"\z°), are dense in C(p,"_I)(aD).
PROOF. aD satisfies the conditions of Theorem 11.4. Hence forms of type (p, n - 1) which are C00 and a-closed in a neighborhood of aD are dense in Thus we need only approximate forms of that kind. Let w be an C(p,nopen set containing aD, and let a E Zip,,,_I)(w). Choose a connected compact )(W).
set KCDso that z°EKand DCKUto.By Corollary 10.4,a=al+a2 on w\K, where a1 E Z(-P,x_1)(w U D) and a2 E Z('P,n_I)(CK). Since K is connected and compact, CK and C"\z° satisfy condition 2 of Theorem 9.1. Hence Z('P,n_1)(C"\z°) is dense in Z(1P,"_I)(CK), and the corollary follows.
CHAPTER IV
SOME APPLICATIONS §12. Generalization of the theorems of Hartogs and F. and M. Riesz
1°. Let D be a bounded domain with piecewise smooth boundary, for which there exists a sequence of domains Dm with smooth boundaries satisfying the following conditions: 1) Dm C D.+,, and U Dm = D. 2) For every holomorphic function f(z) in D,the integral (12.1)
f DmIfIdvm
is nondecreasing as m increases, dvm being the (2n - 1)-dimensional Lebesgue volume element on BD.. 3) If the integrals (12.1) are uniformly bounded for f(z) (in which case we shall say f(z) E H,), then f IaDm converges weakly to a summable function on 8D (also denoted by f(z), z E 8D) in the sense that f DmIfidem <
f
aD
I J { Idv;
lim M-00
f
D
fv = faDfv
for every exterior differential form v E C21,_1(5\ DI). For n = 1, the theory of the class H, has been worked out in detail (Privalov
[1], Hoffman [1]). For n > 1, the class H, can be constructed by using boundary properties of holomorphic functions of several complex variables. We indicate another possible approach. Suppose D is such that each of its sections by a complex one-dimensional analytic plane through 0 is simply connected, and that there is a C' homeomorphism J of D onto the unit balls under which each plane section of D through 0 is mapped conformally onto the section of the ball by the same plane.(') Let D. be the inverse image by ¢ of (') These two conditions are satisfied, for instance, for linearly convex domains with smooth boundary (see Yuzhakov and Krivokoleskio [I]). 67
68
IV. SOME APPLICATIONS
the ball of radius I - (m + 1)-', m = 1,2,.... Then these Dm satisfy the conditions listed at the beginning of the section. If f E H,, then, for almost all plane sections, f E H, in the one-variable sense (see Aizenberg [3], [4]), and the function f(z) I3D can be constructed from the sections utilizing the known boundary properties of holomorphic functions in H, for n = 1. Consider a smooth vector-valued function w(z) on BD for which (5.1) holds and the form w(w, z) is nondegenerate on aD. In addition to the space O(aD)
introduced in §5, we shall need two of its subspaces. Let O,(aD) be the subspace of p E C(aD) such that there exists $ E Z(°° _,)(D) for which 0 I aD = qpw, and let 02(aD) be the space of p in C(aD) such that there exists yE with ay IaD = pw. Obviously, 02(aD) C 01(8D) C 0(aD). Let us consider the problem of describing all measures orthogonal under integration on aD to the spaces O(aD), O,(aD) or 02(aD). In the classical case of n= I and D= U= (z,: z, 1< 1), we have A0(aD) = O(aD), so that the theorems which provide solutions to the above problem may be regarded as generalizations of the classical F. and M. Riesz theorem (see, for example, Privalov [ 1 ] or Hoffman [ 1 ], p. 47):
Let It be a complex-valued Borel measure on the unit circle W. Then
fzkdµ=0, au
k=0,1,2,.. .,
(12.2)
if and only if p is absolutely continuous with respect to Lebesgue measure and there exists f E H, such that
µ(E) = fEf(z,)dz, for any Borel set E C W. Various methods have been developed which lead to different generalizations of the F. and M. Riesz theorem (see, for example, Glicksberg [1], Konig and Seever [1], Rainwater [1], Heinz [11, and Khenkin [1]), but they are not applicable in our case, as a rule, since 0(aD) is not a ring if n > 1. THEOREM 12.1. I. Let It be a complex-valued Borel measure on aD. Then
1Dq)
dµ=0
for all p E O(aD)'if and only if there exists f(z) E H, such that
µ(E)= fEf(Z)w for all Borel subsets E of W.
(12.3)
§12. THE HARTOGS AND RIESZ THEOREMS
69
II. (12.3) holds for all (p E O,(aD) if and only if µ is generated by a form fw with f E H1.
III. If n > 1 and aD is connected, then (12.3) holds for all p E 02(8D) if and only if µ is generated by a form fw with f E H1. Thus, Theorem 12.1 generalizes the classical F. and M. Riesz theorem. But we may also interpret Theorem 12.1 in another way. Consider the question: under what conditions can a function f(z), defined on aD, be extended to the interior of D holomorphically? For n > 1, if aD is connected, then Hartogs' theorem(2) (Fuks [1], Chapter IV, §21.2, and Shabat [1], p. 349) says that it is sufficient that f be holomorphic in a neighborhood of D. The more general theorem of Severi (Shabat [1], p. 344, or Martinelli [1]) gives the necessary and
sufficient condition (3.2) for functions f(z) of class C' on aD. The local differential conditions may be replaced by global integral ones (§3, 2°). It follows from Theorem 6.1 that, in the case of a strictly pseudoconvex domain D, q) E O(aD) if and only if qvw is a-exact. In the case of an arbitrary domain
D, this condition is at least sufficient to ensure that p E O(aD), as Stokes' theorem immediately shows (regarding necessary and sufficient conditions, see
§7). Thus it is natural to call a measure µ a weak solution of the tangential Cauchy-Riemann equations if (12.3) holds for all p E O(aD). Then part I of Theorem 12.1 may be reformulated as follows (a similar interpretation may also be given to parts II and III): THEOREM 12.2. A complex-valued Borel measure µ on aD is generated by the
form f(o, f E H,, if and only if it is a weak solution of the tangential CauchyRiemann equations.
Part I of Theorem 12.1 is a consequence of part II, the proof of which is new even for the classical case n = 1. In order to illustrate the method, we carry out the proof for n = 1 in subsection 2°, and then indicate the modifications to be made in the case n > 1. 2°. For the proof of the Riesz theorem, we shall use the following assertions.
1)If pEC(au)and f 4,(z,)zkda, = 0,
k = 0,1,...,
(i2.4)
au
then qp E AC(U). This proposition can be proved elementarily in various ways. For instance, it
follows immediately from the classical theorem of Fejer on the uniform convergence of the sequence of arithmetic means of the partial sums of the (2) Also referred to sometimes as the Osgood-Brown theorem.
70
IV. SOME APPLICATIONS
Fourier series of the continuous function p to that same function. We need only write the Fourier series in the complex form. Now let X be a Banach space, X* its dual, and I' a linear set in X*. Consider the following closedness properties of I':
a) If xn, E t and limm_. x*,(x) = x*(x) for all x E X, then x* E t. b) If xo (Z y, then there exists an x E X such that xo*(x) = I and x*(x) = 0 for all x * E IF. 2) For a separable Banach space X, b) follows from a).
For this assertion, we refer to a simple theorem of Banach (Banach [1],
Chapter VIII, Theorem 5) asserting that the two conditions are in fact equivalent in the case under consideration .(3) PROOF OF THE F. AND M. RIESZ THEOREM. Let C* be the dual space of C(au) and F C C* the set of measures v of the form dv = f(z1)dz1, f(z1) E H1. We must show that µ E F. We first show that I' is closed in the sense off a). Let
dvm =fm(ZI)dzl, lim
M-00
fm(zl) E H1,
fU Ii(Z1)dvm=f' u
(12.5)
(z1)dvo
for all J(z1) E C(aU). If we apply (12.5) with the Cauchy kernel in the place of 4' and use the fact that a weakly convergent sequence of functionals is bounded in norm, then we see that the sequence fm(zt) converges uniformly in the interior of the disc U. The limit function fo(z1) obviously belongs to H1. To prove that dvo = f0(zl)dz1, it suffices to show that lim f P(zl)dvm
=
f P(z1)f0(z1)dz1
for all 0 in C(DU) or for a set of functions 0 which is dense in C(aU). As such a set we take the set of all polynomials P(z1, zI'). Each such polynomial can
be represented as Pi(z1) + P2(z,1), where P,(z1) is then orthogonal to the functions of class Hi in the sense of integration along W. So we must examine whether
lim IauPZ(1) fm(z,)dz, =Iau P2(1) fo(z.)dz..
M-00
Z1
(12.6)
Z1
The integrals in (12.6) are linear combinations of the values of the respective functions and their derivatives at 0; this is evident for functions in A(D), and easily follows for H1-functions by a limiting process. Now (12.6) follows from the uniform convergence of fm to f0 inside U. (3) See also Dunford and Schwartz [I], Chapter V, §7, Exercise 16.
§13. GENERAL FORM OF INTEGRAL REPRESENTATIONS
71
Thus r possesses property a), hence also b), by 2). If µ E F, there is a p E C(aU) such that f T(zl)f(zl)dzl = 0
(12.7)
f T(z,)dµ = 1.
(12.8)
for aIlfE H1, but
But (12.7) implies (12.4); hence p(z,) E A(aU) by 1). Any function in A(3U) can be approximated by polynomials in z,, and so (12.8) contradicts (12.2). LEMMA 12.3. The formula (5.5) is valid also for f E H1.
PROOF. We first extend the w;(z) smoothly to a neighborhood V of 8D, and then shrink V, if necessary, so that (z, w) 0 in V. Also we replace w; by the normalized functions w; (z, w) -', i = 1,... , n. Thus we have extended the w; (z ) to a neighborhood of aD preserving (5.1). Now we apply (5.5) to the domains
D,,,, m > mo, and then effect a passage to the limit using properties of H,-functions. PROOF OF THEOREM 12.1. The proof proceeds exactly like that of the F. and M. Riesz theorem, using the propeties of H,-functions, replacing the Cauchy kernel by the Martinelli-Bochner kernel and assertion 1) by Theorems 3.1 and
3.2. Here, instead of the decomposition P(z,, z,') = P,(z,) + P2(z,'), we use Corollary 11.5. In the special case of a domain D for which there exists a vector-valued function w(z) such that (5.1) and (5.8) hold and the form w is nondegenerate on aD (fairly large classes of such domains were presented in §5.3°), the proof can be made more elementary by appealing to (5.8) and the decomposition of P(z, w) given by Corollary 5.2 and Lemma 5.3 rather than Corollary 11.5.
§13. On the general form of integral representations of holomorphic functions
10. The problem of describing forms which are orthogonal to holomorphic functions is closely related to that of describing the general form of integral representation of holomorphic functions. By an integral representation of holomorphic functions we mean a formula of the type
f(z)
f
(13.1)
valid for all z E D and all f E A(D). Here D is a bounded domain with smooth boundary, and µz E C2i_1(aD) for fixed z E D. A form pj( ) satisfying (13.1) will be called a reproducing kernel. A reproducing kernel is a solution
72
IV. SOME APPLICATIONS
of the system inhomogeneous linear equations (13.1), and the orthogonal forms
are the solutions of the corresponding system of homogeneous equations. Hence any reproducing kernel can be obtained from a fixed one by the addition of forms orthogonal to holomorphic functions, and conversely, every
form which is orthogonal to holomorphic functions can be obtained as the difference between a fixed reproducing kernel and another suitable one. For functions in AC(D), we have the following well-known integral repre-
sentation: if z) is a vector-valued function which is of class C1 in the variable (' E aD for fixed z E D, and satisfies (w, - z) = 1 (throughout this section, w = (w',. .. , w") will denote a vector-valued function with the above properties), then (Cauchy-Fantappie formula)
f(z) = f
z)
(13.2)
forallf EAC(D)(w(w,., z) has been defined in§1.2°)(a) If we put w = t = (f- i)/J - z 12, then (13.2) becomes the well-known Martinelli-Bochner formula (see (1.1) for p = q = 0).
Observe that, if a reproducing kernel is multiplied by a function E D, z E D, satisfying rpZ E AC(D),
z E D,
q,Z(z) = 1,
then we again get a reproducing kernel. Indeed, suppose ' Then
fo
(13.3) satisfies (13.1).
=f(z)pz(z) =f(z).
It must be noted that the class of kernels of the form
z) is
invariant under biholomorphic maps (Aizenberg [5D, though it is not known whether a similar assertion holds with regard to the Cauchy-Fantappie kernel. 2°. For n = 1, the Cauchy-Fantappie formula goes over into the Cauchy integral formula
f(z) =
If
21ri aD
d Z
THEOREM 13.1. Let D be a bounded domain with smooth boundary, ands continuous reproducing kernel. Then 2I
tt.M = Iri for some
satisfying (13.3).
(°) (I3.2) is a consequence of (1.1), Lemma 1.3 and Stokes' theorem.
a
§13. GENERAL FORM OF INTEGRAL REPRESENTATIONS
73
PROOF. Since dg is nondegenerate on aD and continuous, we have
g2(')d where gZ E C(aD) for fixed z E D. Consider the function I
h:Q) =
2iri((' - z)
For every f E A(D), we have
fp
0.
By Theorem 3.1, h2 can be extended to D, to each fixed z E D, as a function in AC(D). Set
1 + 2i7i( Then p: E AC(D) and q)Z(z) = 1 for z E D. By the definition of hZ and pZ, we have
Y. For strictly pseudoconvex domains in C2, the Cauchy-Fantappi8 formula gives the general form of integral representation, as the next theorem asserts. THEOREM 13.2. Let D be a strictly pseudoconvex domain in C2 with boundary
a reproducing kernel. Then µz is of the form of class C3, and µr E w(w, , z) for some w E C1(aD). PROOF. Every point
E aD has a neighborhood U such that, on aD n U,
all differentials can be expressed either in terms of the differentials of z1, z2 and z1 or those of z z2 and z2. The U; cover 8D; choose a finite subcovering, say U1,...,U,,,. Let {qq,) be a partition of unity subordinate to the covering +µ,,,, where µj = (pju,. By the choice of the U, we {U). Then A. = µ1 + have for z E U,.
µi = a.dz1 A dz + b.dz2 A dz,
where a., b. E
(13.4)
(U). But suppµ; CS U; hence supp a, and supp b. C U.
Therefore, extending a, and b, by zero outside U, we get a., b. E and (13.4) is valid everwhere on W i.e. µ, E C(21)(aD). Hence µ E C(2,1)(aD). . Consider the form
aZ(r) = A(r) - w(t(r, z), r, z).
74
IV. SOME APPLICATIONS
It follows from (13.1) and the Martinelli-Bochner formula that aZ E A1(D). Hence aZ = ay: IaD by Theorem 6.1, with yZ E C(2,O)(D). Let g2 be the coefficient of yZ. Then g2 E C'(5). Set
w(', z) =
z) +
G2 - Z2), 1 - z,). Then (w, - z) = 1, and w E C '(aD ). By Lemma 1.6, for , + z, we have at.w2 A d
1
(27ri)
rt2(J, z) A d
I
,
2
(2ori)
=
Z,
-F-
-
/\ do
Z), , Z) + 5-Y' =
By continuity, the equality w(w, , z) = & persists for 4°. There is a kind of converse to Theorem 13.2:
z,.
THEOREM 13.3. Let D be a domain in C2 with C2 boundary, and suppose that every µ E C(Z 1)(aD) satisfying
f(z°) _ Df E A(5),
(13.5)
for some fixed z° E D is of the form w(v(f), , z°) for some v E C2(aD) such that z°)= 1. Then D is a domain of holomorphy. PROOF. We shall show that D satisfies the conditions of Corollary 8.2. Let
a E A1(D) n C(2I)(8D). Then by assumption there exists v E C2(aD) such
that (v,z°)=land
z°), ', z°) 6 + a = w(v, $, z°),
(13.6)
since the left side of (13.6) satisfies (13.5). Therefore
a = w(v, , z°) -
z°), ', z°).
By Lemma 1.3, the right side is a-exact. An application of Corollary 8.2 now finishes the proof.
5°. For n > 2, we have not been able to obtain a description of the general form of integral representation similar to that in Theorem 13.2. We have only the following weaker result. THEOREM 13.4. Let D be a strictly pseudoconvex linearly convex domain (see Example 1, §5) in C" with C3 boundary, and let µz E C21 ,(aD) be a reproducw1,...,w"("-1)/2 E C'(aD) ing kernel. Then there exist vector-valued functions
§13. GENERAL FORM OF INTEGRAL REPRESENTATIONS
75
such that
n(n- 1)/2
(-1)iw(wi, , z) +
z), , z),
(13.7)
1=1
where An is I if n(n - 1)/2 is even, and 0 otherwise. PROOF. As in the proof of Theorem 13.2, we get
IL,G) = with -y., E
(13.8)
z), , z) IaD + aYZ I8D1
C(1n,n-2)(D)
Since D is linearly convex, the form z), , z) is nondegenerate on 8D for fixed z E D (see Lemma 5.12). Lemma 1.6 implies that for each ° E aD
there is a j, 1 <j
z),
are linearly independent at °. Hence we have
j] A d
Yz IaD = 2
i*j
in a neighborhood of 0, with bi,1 E C'(aD). As in the proof of (13.4), we deduce that (13.9)
YZ IaD = I ai,j,e(J )djt[i, j] A i<j
Observe that aye = dye, since Ye is of type (n, n - 2) in . From (13.9) and the invariance of the differential we get (13.10)
a 1 IaD = I dai,j,e(J) A dtt[i, j] A i<j
zi)ej Set n,, j = zj)e,, where e; is the vector with ith coordinate 1 and all others zero. Index the pairs (i, j) with i <j by 1, and set n
I+i,+j, (21ri ) w/ w --t(J, z) + (-1) (n - 1)1 a yp
then (w', - z) = 1 for 1= 1,...,n(n - 1)/2, since
,,
(rl;, j,
- z)=O. For
z,, we have
(w',
3' ,
z) =
(n - 1)!
(-1)"-'dwi[i,]
(21ri)n
i, - zi,
= (n - 1)! (21ri)n
+
d;t [i,] A dt ail - zi,
(-1)i+-1
(-1)t-'dai,,j,,Z
= w(t(r, z), r, z)
A d('
A
jj A dg +(-1)r-1dai,,j,,Zdst
[i,, j1] A d.
IV. SOME APPLICATIONS
76
By continuity, this equality also holds for ; = z;,. Hence
da11,, =dtt[i,, j,] A d' = (-1)'[w(t, , z)-w(w', , z)].
(13.11)
Summing (13.11) over I from I to n(n - 1)/2, and using (13.8) and (13.9), we get (13.7).
§14. Representation of distributions in 6D' (R2n')
by 1-closed exterior differential forms of type (n, n - 1)
P. As is well known (see, for example, Bremermann [1], §5.9), there exists for each distribution T on the real line R' a holomorphic function f(z1) outside the support of T such that T is the difference of the boundary values of f(z1) on the real axis, in the following sense: for each q) E 6i (5) lim
f [ f(x + ie) - f(x - ie)]p(x)dx = (T, q)). R'
Such a function f is called an analytic representation of T. This concept has turned out to be very useful in the theory of distributions. For example, the product of distributions can be defined with the help of this notion (Bremermann [ 1 ], Ivanov [ I ], Itano [ 1]).
For distributions on R", n > 1, analytic representations have also been studied, but have been less useful than in the case n = 1, since they consist of 2" or n + I holomorphic functions, whose singularities are not contained in the
supports of the distributions (see Bremermann [11, §13.8; Martineau [2], Schapira [ I ], or Vogt [ 11).
In this section, we shall define the analytic representation of distributions on
R2"-' by means of a-closed forms of class Bn,n_ t. This representation has the same properties as the analytic representations in the case n = 1 (except that holomorphy of the representation is replaced by the property of being 8-closed, or more precisely of lying in the class Thus, in this situation also, a-closed forms of type (n, n - 1) can be regarded as analogues of holomorphic functions of one variable. 2°. We shall consider the space R2"-' of the variables (X,, 1) and its embedding in the space C" of the variables (z1,. .. , z,,) that is effected in the following way: z = x + ixn+j, j = 1, ... , n = 1, Zn = Xn + 'Y1X2"_
i-
j-
(5) In §§14 and 15, we use the notation and the properties of the spaces q, 9', F9, P, 0, and 0; which can be found in Bremermann [1]. In particular, the sign (....... ) is used here in a sense different from that of §§1-13.
77
§14. REPRESENTATION OF DISTRIBUTIONS
Observe that the Mar//tinelli-Bochner kernel
Un n-1U Z)
- (27ri)(n
(_I)k-1
1)!
2
X (fk X Zk)dz[k] A dz is a form of class B"."_ in z for fixed Let T E 6'. Then the expression
aT(z) = (Tt., U",n-1G, Z))
(14.1)
will be called the Martinelli-Bochner representation of T. Here j = rlJ + irin+>> 1, and =71", 71 E R2n-'.
j
THEOREM 14.1. If T E Pand K is the support of T, then aT(z) E Bn,n_ 1(CK ), and aT = O(I Z I1-2") as I z I - oo. The form aT represents Tin the sense that
tlimo
yy-
PaT-f= TaT) =
(14.2)
(f=e
for all qq E Oo. The representation of a derivative of T is the corresponding derivative of aT.
PROOF. In view of the legitimacy of differentiation under the functional sign, noted earlier in the proof of Theorem 9.1, we have
_
aT(z) -
(n - 2)!(-1)
Tt' 8Z
(27ri)
n
74 dz[k] A dz[k]
n 2n-2
= 8z 11 (n
(2 )i!)n I)"
Tr'
I2n-2
d [k] A dz[k],
I
Z
i.e. aT(z) E Bn,n_1(C"\(y = 0)). Let z° E R2-'\K. Then there exists 8 E 6D such that 8 = 1 in a neighborhood of K and / = 0 in a neighborhood of zo; therefore (see Bremermann [1], §3.11)
(To Un,n-la, z)) _ (Tr, fla)Un.n-1a, z)).
(14.3)
The right side of (14.3) is also defined in a neighborhood of z0, and we may differentiate under the "distribution sign". Hence aT(z) E Bn,n_1(CK). The last assertion of the theorem can be verified similarly. It remains to prove (14.2). Let us write n
aT(z) = 7, ajT(z)dz[j] A dz. J=I
(14.4)
IV. SOME APPLICATIONS
78
On the plane {z: y = e}, all the summands in (14.4) except the last one vanish, and
(n - 1)! (-1)n-1 [('n - i)dz[n] A dz - (fin - z")di[n] A dz[n] A da"] I2n (277i )n I - Z (n - 1)!Iy I dx
r"(In-xI2+y2)
= K(rt, x, y),
where K(ri, x, y) is the Poisson kernel for the half-space, and x = (x1,...,x2n-1). Let P(z) = (T,, K(i1, x, y)). We shall prove that
limo f T*pv dx = (T, (p) y=E
(14.5)
for all q) E 00. The integral in (14.5) exists, since T*(z) = 0(1 Z 1-2n) as oo. Further, the integral can be approximated by its defining sums, IzI which converge in & to (T,1, 1
f .-iT(x)K(X, 71, e)dx) _ (T" q)*(fl, e)), R2
and p*(rl, e) converges in to ip(71) as a -. +0 (see Bremermann [1], §5.4; the lemma in this citation can easily be carried over to the case n : 1). This proves (14.5), and hence (14.2). Thus the proof of the above theorem proceeds exactly as in the case n = 1 (Bremermann [1], §§5.1-5.4,5.6). If T E 6D', then the Martinelli-Bochner representation of T does not exist in general; nevertheless, as in the case n = 1, we can formulate THEOREM 14.2. Let T E 6D', and let K be the support of T. Then there is an aT E B","-I (CK) representing T in the sense that (14.2) holds for all sp E G1.
The proof proceeds as in the case n = 1 (Bremermann [1], §5.9). We need
only show that, if T E P and aT is its Martinelli-Bochner representation, defined in some ball B, then aT can be approximated uniformly in the interior of B by forms from B" n_ 1(C"). For this, define zI2-an ).
T(z) = (Ti, 13' -
Then, as in the proof of Theorem 14.1,
aT(z) =
(-1)'(n - 2)1
(2711)
n
8 t(z)
dz [k] A dz[k]) k=J
§14. REPRESENTATION OF DISTRIBUTIONS
79
T(z) is harmonic outside the support of T; hence aT can be uniformly approximated inside B by forms
(-I(2
)"
2)I
a(Pj(z)k,dz[k] Adz[k]),
where the Pa(z) are harmonic polynomials.
The representation (14.1) is also defined for distributions in 0', but in this case only the following result, weaker than Theorem 14.1, can be proved. THEOREM 14.3. Let T E 0', and let aT(z) be defined as in (14.1); let K be the support of T. Then aT E B"_ I(CK), aT(z) - 0 as I z --* oo and I Im z" c> 0, and (14.2) holds for all qv E 0_2n.
PROOF. The proof that aT E Bn,n_I(CK) is similar to that of the fact that aT is holomorphic when n = I (Bremermann [1], §6.5). We have (fk-Zk)It-zI-2n ,0 inFaslzl-> o0 - z I-2n s (Im zn)2-2n. We can write down similar inequali-
and I tk - zk I ties for the derivatives; hence U,,,n_ IG, z) - 0 in Oo as I z I oo and I Im zn I> c > 0. Hence we may take limits in (14.1). The rest of the proof proceeds as in the case n = I (Bremermann [1], §§5.6, 6.5); the necessary uniform estimates at the corresponding places are valid. I
We shall say that a distribution T is of class A, if there exists an aT E BB,n_I(C"\R2"-') representing T in the sense of Theorem 14.3 and aT (see (14.4)) is bounded on the sets I Im zn I'- c for all c > 0. Clearly, 0o C A. Y. We give below some examples of distributions and their representations. 1) the Dirac distribution. Then
as(z) _ Or, U., 1G, Z)) = Un,n-i(0, z)
_ (27ri) (n " 1)!Zn 2 (-1)kakda[k] Adz. Izi k=1 2) Define the distribution 6+ by
(p)=-,limo (2i7r1)!J
(IxI +E2)"
for p E 61. For n = 1, this gives the usual 6+ (Bremermann [1], §7.2). Let us show that 8+ E 6D'. The expression
(n - 1)!e 277"
f
tp(x)dx
R2"-i (IXI2+E2)"
80
IV. SOME APPLICATIONS
differs from the Poisson integral for the half-space only by a constant factor; hence its limit as a -> 0 exists and equals igp(0). Consider the integral 2iir
1
Since p E 6D, we may suppose that the integration extends over a ball B, = (x: x E R2n-', I x I< r). Further we replace cp(x) by [qp(x) - p(0)] + q2(0), and observe that x"dx
= 0,
(IXI2+E2)"
fB,
since the integrand is odd and the domain of integration is symmetric about 0. Now 2
xjf'
9'(X) -92(0) =
0
J=1
az
p(tx)dt =
2
j
xjp,(x).
l=1
The functions XkX"pk(x) I X I-2n are integrable in Br, and IXI-2n'
IXkXfPk(X)I(IXI2
+ e2)-"
hence we may take the limit under the integral sign. Thus S+ is defined. It is obviously linear, and its continuity follows from the weak completeness of 6D' (see Shilov [1], Chapter II, §9.3). Suppose qp E 6D has support outside some ball; then
f
97 (()
fR2n-
X(I2
i P(X) I
I
Ix Izn
+
R2n- I
- le I dx.
Taking limits as a --> +0, we get
s+
I9p(x)Idx
< cfR.-I IxI2n-I
i.e. 8+ = O(I X I1-2") Hence S+ E Oa for any a < 0 (see Bremermann [1], §7.2).
Finally, we write down the representation of 8+
(n - 1)!" as+(z) _
2 (_l)kikdz [k] A dz,
(21riIZI2) k=1 0,
Im Z. < 0.
Im z" > 0,
§14. REPRESENTATION OF DISTRIBUTIONS
81
3) S_ = 8 - S+ ; hence the representation of S_ has the form IM z" > 0,
0,
aa_(z) _
(n - 1)!
(-1)kzkdz [k] A dz,
I2)n
(27fi I Z
Im z" < 0.
k=1
4) The principal part P(x" I x I-2n) is defined by 12n),
(P(Xn I x
x"q)(x)dx
q)) = lim
e-.+0 R2"-I (I X
12
+ EZ)
i.
For n = 1, this distribution is denoted by P(x-1) (Bremermann [1], §7.5). Also,
n (nier"1) P IXn2 I
(14.6)
=S_
It follows that P(xn I X I-2n) E Oa for any a < 0. As in the case n = I (Bremermann [1], §7.6), it is easy to show that P(x" I x1-2n) is equivalent to the Cauchy principal value of the integral IR2"-1XnIXI-2n p(x)dx.
The representation of this distribution is obvious, taking into account (14.6) and the representations for S+ and 45) Consider the distributions 0k, k = 1,. . . , n - 1, defined by (xn + IXn+k)(P(x)dx (Bk
I'M e-+0IR2"-1
E 6l .
(IX I2+e2)n
As in Example 2, it is not hard to show that ek E O. for any a < 0. As a representing form for °k, we may take (i/2)n-' I Z 1-2.[(_ 1)n-k-1Zndz
[k] Adz +zkdz [n] Adz], Im zn > 0,
aek(z) = 0,
Im zn < 0-
6) Finally, we define the distribution 00 by
82
IV. SOME APPLICATIONS
Clearly, 00 E OQ for any a < -1, and (i'/2I z 12)n-1 [di[n]
(-1)n-k-lZnZk
A dz +
ao.(z) = i
Imz2>0,
Xda[k] Adz], Im zn < 0,
where k can have any integral value from 1 to n - 1. It is clear that all the examples are of distributions of class A. §15. Multiplication of distributions in 6
'(R2n-1)
Let T E A. Consider the representation aT of T which occurs in the definition of the class A. Recall that aT is the coefficient of da[n] A dz in aT. We denote by BA the space of all functions of the form
aT(zl, Z2,...,Zn) - oT(Z1, z2,... '2n_1' Zn),
Im Zn > 0, T E A.
BA contains differences of all representations of the T E A satisfying the conditions in the definition of the class A.
The elements of BA are harmonic functions, bounded on all the sets (IM Z" c > 0). Consider the map ¢: f E BA --> T E A such that for every p E D
lim f pfdi[n] A dz = (T, q,). +0 y=e We shall show below that ¢ is an isomorphism of BA onto A. We denote by BA the algebra generated by BA. We introduce the following equivalence relation in BB : f, - f2, if, for any (p E' ,
f
lim (f1-f2)q,dx=0. e- +0 y=E
Let .4 be the quotient space of BA by this equivalence relation, and f the image of f E BB under the natural map BB - A. We have BA C A. If
T
EBA,
j=1,2,
then by the product T, T2 we shall mean the element (11 f2) of A. We define the derivative in A by
s ak1= (af/axk) It is defined unambiguously; indeed, if f, ^- f2, then I-e
xks, f [(h1f2I=-11m
y=e
(f1 -f 2)
=0 aXkk
83
§15. MULTIPLICATION OF DISTRIBUTIONS
for any p E 0-20 i.e.
afl/aXk ^' aft/aXk. THEOREM 15.1. 1) ii is an isomorphism of BA onto A.
2) If S, T E A and the supports of S and T are disjoint, then S T = 0. 3) If S and T are defined by bounded continuous functions S(x) and T(x), then
S T = S(x)T(x). PROOF. 1) By definition, ¢ is surjective, A and BA are vector spaces, and ' is
linear; hence we need only show that the kernel of ¢ is trivial. If f E Ker then, for any p E 6D,
f e-lim +0 R2"
p(x)f(x, e)dx = 0.
(15.1)
But since the left side above is a distribution lying in 0'2n and 6 is dense in ®2n+ (15.1) holds for any y2 E 0_2n as well (Bremermann [1], §6.7).
Put
_
,
=
(n - 1)! 1,n
t2n
(Ix't12+t2n)n
where 't = (t1,...,t2n_I). The function f(x, y + e) is harmonic and bounded for y > 0; hence by Poisson's formula
f('t, t2n + e) = fR"-l p,(x)f(x, e)dx.
Lettinge- +0, wegetf(t)0forten>0. 2) Set u = 4-'T and v =
'S. We must show that, for any p E G1,
lim fR2- 1 u(x, e)v(x, e)p(x)dx = 0.
(15.2)
Outside the support of S, we have v(x, 0) = 0; hence v(x, y) = yv1(x, y) in a neighborhood of any such point. If we now choose T with support in such a neighborhood, then tp(x)v(x, e) 0 in 6 as e - +0; hence (15.2) holds (see
Shilov [1], Chapter II, §9.3). (15.2) also holds for functions with support outside that of T. Since every p E GD can be represented as the sum of functions cj with supports disjoint from that of either S or T (see Bremermann [1], §3.6), it follows that (15.2) also holds in the general case. 3) Let w = , -'(TS). Then u, v and w are harmonic extensions of S(x), T(x) and S(x)T(x) to the half-space (x: Im zn > 0), given by the Poisson integral formula. Also, u --> S, v - T, and w -> ST uniformly on any compact set (see the proof of Theorem 14.1). Hence uv -* ST uniformly on any compact set, so that uv ^- w. Q.E.D.
IV. SOME APPLICATIONS
84
We remark that, for n = 1, several definitions of the product of distributions have been introduced by means of analytic representations (see, for example, Bremermann [1], Russian pp. 245-259)* Ivanov [1], or Itano [1]), but this kind of definitions leads to difficulties for n > 1, basically for the reason that the
product of distributions with disjoint supports need not be zero; even the product of the zero distribution with a nonzero one need not be zero (Itano [1]).(6) Theorem 15.1 shows that the definition of the product given here does not lead to these difficulties. In conclusion, we present some examples. 1)
i-'Ok = (i/2)n-l zk/I z I2n, and ¢-'B0 = (i/2 I Z It follows from the identity IzI2-2n
12)n-
1
(3/8Zk)(zk/I z I2n) = -nzk/I z I4n
that
900 (a/aZk)ok=-n0k°Bk. 2)
(n - 1)] (-1)n(zn - zn) (27TiIZI2)n
and r,
-'
(
in- '
YP(xnIX1-2n)
= (2IZI2)n(zn+Zn).
We have Boo (a/axn)S = -2nS o P(xn I X I-2n),
since 1 a IzI2n-2 axn
Zn - zn IzI2n
Zn - Zn ---n Iz12n
(15.3)
Zn - Zn IzI2n
For n = 1, (15.3) goes over into the known identity
8' = -28 o P(x-' ) (see Ivanov [1), p. 18). * Editor's note. The citation is to an appendix added to the Russian translation. (6) Other methods of defining the product of distributions have been developed by Vladimirov [2]-[4].
BRIEF HISTORICAL SURVEY AND OPEN PROBLEMS FOR CHAPTERS I-IV
1°. The formula (1.1) was first proved for (0, q)-forms by Koppelman [1], and then generalized to (p, q)-forms in Aronov and Dautov [1]. Lemmas 1.1 to 1.4 and 1.6 were formulated by Koppelman [1], [2]. The elementary proof of (1.1) given in §1 is a detailed exposition of the proof in Aronov [1]. The results of §§2, 4, 6, 7 and 13 had been noted previously under stronger
smoothness assumptions. The use of Holder conditions has enabled us to sharpen these assertions in this book. Theorem 2.5 was formulated in the case q = n - 1 by Aizenberg [6], [7] and
Serbin [1], but the latter's proof is faulty. For arbitrary q, the theorem was proved by Dautov. Theorem 2.6 is due to Dautov and Kytmanov [1]. Corollary 2.7 was observed for p = 0 by Lu Qi-keng and Zhong Tong-de [I]. Example 2.9 and Theorem 2.10 were suggested by Dautov. Theorem 2.11 is also due to Dautov; the solvability of this boundary value problem had been established by Andreotti [ 1 ] and Andreotti and Hill [ 1 ], but their solution was not explicit. Theorem 2.12 is contained in Aizenberg [6], [7]. Corollary 2.13 is taken from Aizenberg [7].
Theorem 3.1 was proved by Weinstock [1] for domains with smooth boundary, but his proof works only with a stronger smoothness assumption than 8D E C', e.g. if 8D is a Lyapunov surface. The proof presented here is due to Aronov and Dautov [2], and is based essentially on the same ideas as in the case n = I (see Muskhelishvili [1], §29.3). Apart from the tangential Cauchy-Riemann equations (3.2), there are other local differential conditions
which may be imposed to secure holomorphic extension of f(z) into D (HOrmander [1], Theorem 2.6.13). The possibility of replacing local differential conditions by global integral ones seems to have been first observed by Fichera 85
86
HISTORICAL SURVEY AND OPEN PROBLEMS
[ I ] (see also Martinelli [ 1 ] and Kohn and Rossi [ 11). Theorem 3.2 is taken from
the work of Weinstock, but the proof presented here is due to Aronov and Dautov [2].
The construction of a "barrier function" (§4.2°) was given by Khenkin [1] (see also Ramirez de Arellano [1] and Hvrelid [1]). Lemma 4.2 was implicitly present in Dautov [1]. This lemma is basic for the proofs of Theorems 4.3, 4.4, 6.1 and 6.4. Theorem 4.3 is equivalent to Theorems 6.1 and 6.4 except for
minor details (this form of the assertions of Theorems 6.1 and 6.4 was proposed to the authors by V. P. Palamodov). Theorem 4.4 is due in the C°° case to Kohn [1], whose work is based on L2-estimates for the 8-Neumann problem (see also Hormander [1], Chapter IV). We remark that a solution of
the a-Neumann problem by integral formulas, as well as estimates in the uniform metric, were first obtained by Khenkin [2] and Grauert and Lieb [1] (see also Khenkin [31, Lieb [11, Q vrelid [11, Romanov and Khenkin [1), Kerzman [1] and Polyakov [ 1], [2]).
The results of §5, excluding Theorem 5.6 and Lemma 5.11, are proved in Aizenberg [8], [9]. Theorem 5.6 was published in Dautov [1]; for n = 2, it has
been obtained jointly by Aizenberg and Dautov. Lemma 5.11 is due to Kytmanov and Preobrazhenskii [1].
The results of §6 are due to Dautov [1], [2]. An assertion equivalent to Theorem 6.4 in the C°° case had been proved earlier by Kohn and Rossi [1]. In Andreotti and Hill [2], there is a local variant of the Hartogs theorem for C°° forms, from which one can also deduce a global result (of the type of Theorem 6.4). Theorem 7.1 is taken from Aizenberg [6], [7].
The propositions of §8 are due to Dautov [4].
Theorem 9.1 is a precise analogue (both in formulation and method of proof) of Runge's theorem in one variable (see Harmander [1], Theorem 1.3.1).
The results of §9 and Theorem 10.2 occur in Dautov [3]. The possibility in principle of obtaining such assertions had been observed by Malgrange [2]. The results can also be deduced from Corollary 2.13 and analogous assertions for harmonic functions (see Palamadov [1], p. 377). Corollary 10.3 and Theorem 10.4 are contained in Aizenberg's note [7].
The results of §11 are due to Dautov, and were published (except for Theorem 11.4) in [3]. Theorem 11.1 had been proved, in the case when the. compact set is the closure of a domain with smooth boundary and the form is C°°, by Weinstock [2]. For n = 1, the similar assertion is due to Browder (see Gamelin [1], Chapter II, Corollary 1.2). Theorem 11.2 for n = 1 is a known theorem of Keldysh [1] with a supplementary condition. Theorem 11.4 is a
HISTORICAL SURVEY AND OPEN PROBLEMS
87
generalization of the theorem of Hartogs and Rosenthal (see Gamelin [11, Chapter II, Theorem 8.5).
The results of §12 were obtained by Aizenberg. In less general form, they have appeared in his papers [8] and [9].
The Cauchy-Fantappie formula (13.2) was proved for convex domains by Leray [1], and for arbitrary domains by Aizenberg [10], [5], Koppelman [2], and Kenkin [1]. In the last article, the proof is close to Leray's. The footnote to formula (13.2) in fact Koppelman's proof [2]. Theorem 13.1 was noted by Aizenberg [5]; the remaining results of §13 are due to Dautov [14].
The propositions of §§14 and 15 are taken from Aizenberg and Kytmanov [1]. The definition of the product of two distributions considered in §15 was first introduced for n = 1 by Ivanov [1], [2].
2°. To conclude, we indicate some open problems which we think are of some interest.
1) The Martinelli-Bochner-Koppelman formula is a generalization of the familiar Martinelli-Bochner representation for holomorphic functions to exterior differential forms. Is there a similar generalization of the Cauchy-Fantappi6 formula?
2) Can one replace "all 8-closed forms" in Theorem 3.1 by "all the forms z), z E CD"? In other words: Is the representability of a function in a given domain by the Martinelli-Bochner formula sufficient for it to be holomorphic?(t) This question can be generalized as follows: 3) If a E Z('p,q)(D), then a is given by (1.1) without the second summand.
Conversely, let a E C(p, )(D) and suppose it can be represented by (1.1) without a second summand. Does it follow that a is 8-closed in D? The same question can be asked under weaker smoothness conditions (Holder condition, etc.).
4) Can one assert in Theorem 2.12 that y, E Bp "_ 1(D) when 0
88
HISTORICAL SURVEY AND OPEN PROBLEMS
7) Are the orthogonal polynomials P(w, z) E O(aD) dense in O(aD)? A positive answer to this question would enable us to state the analogue of the Hartogs-Riesz theorem of §12 in a way similar to the classical theorem: one
would be considering measures orthogonal to the polynomials in O(aD) described in §5.
8) Elucidate for which domains the second summand in Theorem 7.1 is necessary. The results of §§6-8 solve the problem only in certain cases (the case of strictly pseudoconvex domains from which a finite number of strictly pseudoconvex domains have been removed; the case of domains not covered by Theorem 8.1, when n = 2).
9) Characterize the class of domains for which the Cauchy-Fantappit formula is the general form of the integral representations for holomorphic functions (analogues of the Cauchy integral). The theorems of §13 show that for n = 2 this class contains the strictly pseudoconvex domains and consists only of domains of holomorphy. This problem is related to the next two problems. 10) Is the Cauchy-Fantappie formula invariant under biholomorphic maps?
11) Find the general form of the integral representations of holomorphic functions for an arbitrary bounded domain with smooth boundary.
12) Prove approximation theorems for forms in Z(,_ I), similar to the well-known results of Vitushkin [ 1 ] and Mergelyan [ 1 ].
13) Extend the definition of the product of distributions of class A in R2 n-' to arbitrary distributions. 14) The representations of distributions in §14 are closely related to their Fourier transforms when n = I (see Bremermann [1], Chapter III). Is there a similar convenient connection when n > 1? A positive answer to this question would lend further support to our view that forms in Bn,n_ I (or Z(n,n_ I)) are good analogues of the holomorphic functions of one variable. Note. After this book went to press, solutions have emerged for some of the problems listed above. Aronov and Kytmanov [1] have obtained an affirmative answer to question 2) for functions of class C'(D). Also, Kytmanov [3] has settled question 3) positively for forms in C( p,q)(D). In another paper [5] by the same author, problem 13) is solved in arbitrary dimension.
CHAPTER V
INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED DIFFERENTIAL FORMS AND HOLOMORPHIC FUNCTIONS
§16. A characteristic property of a-closed forms and forms of class B
In this section, we answer question 3) in the "Brief historical survey and open problems" section. I THEOREM 16.1. Suppose a(D) E C2, "` and y E C(v Q)(D), X > 0. Then y is a a-closed in D if and only if, for all z E D,
y(z) = Ij.q(D, y)
- aIj.q-i(D, y).
(16.1)
If the coefficients of y are harmonic in D, then the conditions on aD and the behavior of y on aD can be relaxed. THEOREM 16.2. Let D be a bounded admissible domain with piecewise smooth boundary, and suppose y E C(,, q)(D) has coeffficients harmonic in D. Then y is a-closed in D if and only if (16.1) holds.
Note that, if q = 0, then JP2,
y) = 0, and Il,q(D, y) has coefficients
which are harmonic in D. Hence from Theorem 16.2 we deduce COROLLARY 16.3. Suppose D is as in Theorem 16.2, and y e C(p o)(D ). Then,
for z E D,
y = I,,o(D, -y), if and only if the coefficients of y are holomorphic in D. 89
(16.2)
90
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
For the proof, we need to use the double form
z) = (n -1)!
ta(I, I
k) k- z
IZ"df [I, k]dz,
(a part of UP,,), as well as the following lemma. LEMMA 16.4. Let D be a bounded admissible domain in C", n > 1, and Suppose
f E C'(D) is harmonic in D. Then f can be approximated in the topology of C'(D) by linear combinations of fractions of the form I
- zk I2-2n, Zk (4 D
PROOF. Since D is admissible and bounded, the space of functions harmonic
in a neighborhood of D is dense (in the topology of C'(D)) in the space of functions of class C'(D) harmonic in D (see Weinstock [31). Hence we need only prove the lemma for functions harmonic in a neighborhood of D.
Now let f be harmonic in the closure of a domain G D D with smooth boundary. Then, by Green's formula, for z E G
f(z) = (2n
12)a2n
bas
LW, z)
a fm
-f(0) an W, z)J da, (16.3)
where a2,, is the area of the unit sphere in C", and z) =I - z I2-2n; also, af(t)/an denotes the derivative off in the direction of the outward normal to aG at , and da the surface element of aG. By replacing the integral in (16.3) z)/an by difference quotients, by the approximating Riemann sums and we obtain the assertion of the lemma. PROOF OF THEOREMS 16.1 AND 16.2. The necessity follows immediately from
the Martinelli-Bochner-Koppelman formula (1.1). Let us prove the sufficiency. If the form
7(r) = 2'
A dfr
t J satisfies (16.1), then each of the forms
Yj _
I
1'r,rdfr
also satisfies (16.1), Indeed,
IP,q(Y) =8D f I'
A
A
z)
A dzk k
= f D 2'
2' (Yy,jU)dfr A wq A
A
A dz,)
§16. a-CLOSED FORMS AND FORMS OF CLASS B
_i'
D
Adz,
1Y'
J' Io.v(D,
91
A
Similarly,
_
51p.q-1(D, Y)
510,,_ 1(D, Y) A
Hence it is enough to consider forms of the type (0, q)
Y(z) _ E'Yl(z)dzl.
I Suppose now that the conditions of Theorem 16.2 are satisfied. If (16.1) holds for y, then Y1(z) = Ip,q(D, ay) = 0 for z E D by (1.1). But the coefficients of y, are harmonic outside D, and tend to 0 as I z - oo; also, being convolutions of compactly supported bounded measurable functions with the locally integrable functions I 1_2nfk, they are continuous in C". Hence by the uniqueness theorem for harmonic functions we have y, = 0 on C". Writing out < this identity, we see that, for every multi-index I'(i, < i2 <
fD 1,
ay' k(ZI, 4k
A dal' A 2 u(I, m)adf [I, m] A d = 0.
(16.4)
mal
By Lemma 16.4, the yj( ) can be approximated in C'(D) by linear combizs), zs E D, s = 1, 2,.... Hence (16.4) yields nations of the
fD I,
kz/, afk
dfk A
A 7, o(I, m) m(Zl
aYl
df [I, mI A dg = 0.
(16.5)
afm
Summing over I in (16.5), we get
f DIaYI2df
i.e.5y=0. Suppose next that we are in the situation of Theorem 16.1. The coefficients of the form Y2 = I).q(D, y) are harmonic in D, and y2 E C(o,q)(D) (see Corollary 2.4). Further, aY2 = ay by (16.1), and hence
I02,q(D,a72) = Iq(D,ay) =0. Thus, (16.1) holds for y., and it remains to apply Theorem 16.2. We have just considered the properties of forms representable by (16.1), i.e. the Martinelli-Bochner-Koppelman formula without the second summand. It is
92
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
interesting to discuss the class of forms to which the first summand belongs. If we consider (1.1) with p = n and q = n - 1, we see that the first summand has the form In,n-1(D,
y)
- f Y(J )Un.n-1(J. z).
It is clear that this form is of class B,,,,,- 1(D). Forms of this important class have already been encountered before in this book (see Theorems 2.12, 7.1 and 14.1-14.3). Let us present some new properties of this class of forms. They can be written as Y = Yc(z)
aG k
k=1
dz [k]1 A dz,
with G harmonic in D.
THEOREM 16.5 (UNIQUENESS). Let aD E C', and suppose G E C'(5) is harmonic in D. If the restriction of Yc to aD is zero, then YG = 0. PROOF.
_
aG
0 = f DGYc(z) = fDdG A Yc = fD k=1
aZk
2
da A dz.
Hence aG/azk = 0, k = 1, ... , n, i.e. G is antiholomorphic in D, and so Yc = 0. We remark that Theorem 16.5 remains valid even for unbounded domains, if we assume G = 0(1 z I2-2n) as I z I -> oo. THEOREM 16.6. If n > I and aD is connected, and y E C(;,,"_ 1) is such that
y(z) =
1(D,Y),
z E D,
(16.6)
then y =- 0. PROOF. Consider the functions
z E D, z(4 D. These functions are harmonic (outside 3D). It follows from the properties of the potential of a simple layer that G+ = G- on 3D. By Theorem 2.5,
YM
IaD =
Yc+I8D - YG-IaDI
(16.7)
so that (16.6) implies Yc-IaD_= 0. By Theorem 16.5, Yc-= 0 outside D, i.e. G- is
antiholomorphic outside D and vanishes at infinity; hence G-= 0. Hence G+ = 0 as well. Now the assertion of the theorem follows from (16.7).
93
§16. a-CLOSED FORMS AND FORMS OF CLASS B
This theorem shows that there are no nontrivial forms representable in D by (16.6) (i.e. by (1.1) with only the first summand). The next result describes forms representable by (16.6) outside D.
THEOREM 16.7. Let a(D) be connected and of class C', and suppose G E G'(CD) is harmonic in CD, with G = O(I Z I2-2") as I z I - oo. Then YG(z) =
YG),
z a D,
if and only if G can be continued antiholomorphically into D.
PROOF. Necessity. By considerations similar to those in the proof of Theorem 16.6, we see that G- is antiholomorphic in D and that YG = YG- Hence G - Gis antiholomorphic in CD and tends to zero as I z I - oo, so that G = G-. Sufficiency. Suppose G can be continued antiholomorphically into D: denote
the continuation by G1. Then GI E C'(aD) and G, is antiholomorphic in D; hence G, E C'(D) (Shabat [1], p. 334). Let
z E D,
{G+(z),
z a D; then
YG+IaD - YG-IaD= -YG IaD = YG, Ian - YGIaD'
so that (YG+ -YG,) laD = (YG--YG) IaD'
Let D = (z E C": p(z) < 0), where p E C'(C") and grad p IaD 1 G = (20
"" aGap Igrad pI
(16.8)
0. Set
I
azk azk k=1
N. is defined near aD. It is easy to see that NG lad(' = YG IaD
where as is the surface element of aD. Now we see from (16.8) that NG+-G, = A19--G for z E aD, and G+ -G, _
G--G on aD. It follows from this and the complex form of Green's theorem that the function z)
G+(z)-G,(z), zED, G-(z) - G(z),
z
D,
4 ) = 0; hence 4 = 0. Consequently G+ (z) is harmonic in C" with = G,(z) and G-(z) = G(z). Hence YG+ = 0 and YG-= Y.
94
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
§17. Holomorphy of continuous functions representable by the Martinelli-Bochner integral; criteria for the holomorphy of integrals of the Martinelli-Bochner type
If p = 0, the kernel in (16.2) becomes the Martinelli-Bochner kernel U 0(
,
z)
-Un.n-1(Z, J)
2k=1(-l)k-1(yk
(n - 1)! I
(21ri)"
IJJ
-
[k] A d
- Z I2n
and the formula itself reduces to the Martinelli-Bochner formula for holomorphic functions:
f(z)=f D
z),
z E D,
(17.1)
which is valid if D is a bounded domain with aD piecewise smooth, and z) is not holomorphic in z f E A C(D ). The Martinelli-Bochner kernel for n > 1. Nevertheless, the following sharpening of Corollary 16.3 holds. THEOREM 17.1. If D is a bounded domain withpiecewise smooth boundary, and
the representation (17.1) is valid for an f E C'(D), then f E A(D).
We may drop the condition f E C'(D), but we must then impose additional conditions on 3D: THEOREM 17.2. If f E C(D) is representable in a bounded domain D by the Martinelli-Bochner integral (17.1), then f is holomorphic in D in each of the following cases:
a) OD E C2; b) n = 2, and aD is connected and of class C'.
If f E AC(D), then
f DfwUO,a, z) = 0,
z (4 D.
(17.2)
As it turns out, (17.2) completely characterizes the traces on aD of functions holomorphic in D: COROLLARY 17.3. In each of the two cases considered in Theorem 17.2, the condition (17.2) holds if and only if for f E C(aD) there exists F (E AC(D) such that F I aD = f.
Let Pk 1 be spherical (i.e. homogeneous and harmonic) polynomials forming an orthonormal basis for L2(aB(0,1)), where aB(0,1) is the unit sphere in C". Here, l is the degree of the polynomial, and k its index among the polynomials
95
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
of degree I occurring in the basis, so that I = 0,1, 2..., and k = 1,.
a(/),
where
v(1) - 2(n + k - 1)(2n + k - 3)!
k!(2n-2)!
COROLLARY 17.4. Suppose D is bounded, and aD is connected and of class C2.
Then an f E C(8D) is the trace on aD of a function from ACID) if and only if
f
I (-1)'
0,
[j] A
I=>
>
1= 0,1,2,...; k = 1, 2,...,a(1). For n = 1, this corollary is the well-known classical criterion, namely orthogonality to all the monomials ',1= 0, 1, 2,.... We shall not present the proof of the rather easy Theorem 17.1 here, but refer the reader to Aronov and Kytmanov [1] (see also Corollary 16.3), or to the book by Aizenberg and Yuzhakov [1], §1. We proceed to the proof of Theorem 17.2, which is more difficult.
Let f E C(8D), 8D E C'; consider the following integral of MartinelliBochner type:
+z
f
z) = f-(z),
E z
D.
e in the same manner Consider limz,z,.. Ean[f+(z) - f -(z')], where z, z' -+ so that Theorem 17.2 as in (2.9). By Theorem 2.6, this limit is equal to and Corollary 17.3 are equivalent. Therefore it suffices to prove Corollary 17.3.
We introduce some notation. Let z° E 8D, z+ E D and z-E CD, with the last two lying on the normal to 8D at z° and I z+ -z I=I z--z° I . Define Pt 3') =
- zEaD inf I - zI, QED, zi n f
1 '-Z 1 ,
(4 D,
then D = (z E C": p(z) < 0). If 8D E C2, then (see Volkov [1], §2) the following assertions hold: a) There exists a neighborhood Uof 8D such that p e C2(U).
b)Igrad pl=1in U. c)
8zk
andz EU.
(z+
8z k
(z-) =
a/laZ
k)
(z°),
k = 1,2,...,n,
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
96
Let __
NF
ap aF 8i k
n
I
az k
k=I
LEMMA 17.5. Suppose aD E C2 and F E C(aD); then lim
Z+, Z-_.Zo
(NF (z+) - NF (z-)) = 0
uniformly in z°. If NF can be extended continuously to D, then NF can be continuously extended to CD, and conversely.
PROOF. Let F E C(aD). It is clear that the second part of the lemma follows from the first. Since NF = NF+c for any constant C, we may assume F(z0) = 0. Observe that
d [k] A d IaD =
(2i)"(-1)k-I
8 k
do,
where do is the surface element of D. Therefore (y
U0,0G, Z) IaD
(n
I
1,n l)
kI
ayk
I pk Z I2ndo,
hence
Nj (z-)
NF' (2+
=-(n IT" + n!
a
do Ip-Z-I2n I J
aZk°)
(ap(Z°)/aZk)(tk - Zk )>+m=1 (aP/asm)(m - zm
IT" aD
I t _ z+ I2n+2
jk=I (aP(z°)/aZk)Uk -Zk )I=I (ap/afm)(fm - Zm) do. - Z-1 2n+2 I
Let us denote the first integral above by II, and the second by I2. By performing a unitary transformation and a translation, we may take z° to
0, and the tangent plane to aD at z° to the plane a = (w E C": Imwn = 0}. Then aD will be defined in a neighborhood of the origin by the system of equations
T ='w, where
n=un+i(w), w=('w,un)Ea.J(w)will
be of class C2 in a neighborhood W of 0 in a, and z ± will be of the form (0'. . . , 0, iy,; ).
97
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
8D is a Lyapunov surface with Holder exponent 1. Hence
I(w)I
II
w E W,
c1w12,
1,2,...,n,
aya,
n- 1, (17.3)
where z. = x. + iy, (see Vladimirov [5], §22.4; these estimates are derived there for R3, but the proofs easily carry over to R", n > 3). Hence
ask
I
I
w E W,
aPk I
(17.4)
fork = 1,. .. , n - 1, since alp
a / 8yn'
az
ayn
c2,
w E W.
l
Here the constants are independent of the point being considered. Furthermore, c31w1,
w E W.
(17.5)
Fix an e > 0. Choose a ball B in the plane a with center 0, and an a > 0, such that
1)BCw, I< e for w E B, 3) B X [-a, a] C U, 2) 1
Forz=z
+clwl2)
_
1
1
1but
I2iFYn-021
w-z12
Consequently, 00
1 -(2sv2)/Iw-zjZ
k=0
IY'-zl2k
2 91(w,
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
98
where g,(w, z) is bounded uniformly for z E B and I yn I < a. Therefore I
I
IW-ZI2n
_
1
ZI2n+2
l (1+ 2gyn-J2 Iw-=I2 g2)+ 2,Jyn - & 2
1
1+ I W -ZI2
I W- Z I2n-2
93)
(17.6)
(17.7)
,
and g2 and g3 are uniformly bounded for w E B and I Yn I < a.
Let us estimate the integral I, over the portion of aD corresponding to the ball, i.e. the integral I'. By (17.6) and (17.3), we have
_
1
-P2I
1
I y- Z-12m
I y- z+I2n
I W - Z+ I2n+2
Igzl `ca (I W I2
Further, do < d, dS, where dS is the surface element of the plane a; hence
ecs
I Y I dS
2(R2n
n
dS
+c1
B (IWI2+Yn
2)"-I
The first integral above is a Poisson integral for a half-space (see, for example, Stein [2], Chapter III, §2.1) and is independent of y , while the second integral is continuous at 0. Hence 118 I s me, where m is independent of F and the point considered.
The integral I, over the remaining portion of aD can be obviously made ±
arbitrarily small as z -+ 0. It is easy to see that the form of 12 is unchanged by the unitary transformation and translation. Hence
GI a
m(Jm-Zm)
kY, n
= 1Un - Zn) I IP f(m_1m)
mI
n-1
=1
ePrmlrn
M= I arm
m
- Zn)+iaP afn
(un+,
2+Y,,'
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
99
The integral 12 over the part of aD corresponding to B can be split into three integrals
(ap/afn)2pyn + En,=11 (ap/affmy, do I w - Z I2n+2
2in!
n
I'2 2=± X
fB
in!
(ap/afn)(un + 2 +Yn 2 - 2pyn ) + Fm=11 Iw-z+I2n+a n
X
ds' =
ip2)93da',
1+
Zn
i=1
n-I
2
a
(ax) I
+
i-1
a
2
ate) dS, yj
and each of these can be estimated in the same way as 1', using (17.3)-(17.5) and (17.7).
LEMMA 17.6. Let D C C" be a domain with C' boundary, and let F E C(aD). Set
f Fdtg n df [k, m] n d = ao
where
z) = I
HF (z), HF (z),
z E D, z E D,
- z I2-2n. Then lim z+ z-..zo
(HF (z+) - Hi (z-)) = 0,
uniformly in z 0. If HF (z) can be extended continuously to D, then Hi (z) can be extended continuously to CD, and conversely (z' and z° are as in Lemma 17.5).
PROOF. We essentially repeat the proof of Theorem 2.6. Since HF = HF+c we may suppose F(z°) = 0. As in Lemma 17.5, we make a unitary transformation and translation so that aD may be defined in a neighborhood of the origin by the system of equations
T ='w,
fin= Un+i (w)
(we preserve the notation of Lemma 17.5). In addition, ¢(w) is of class C' in a neighborhood w of 0 in the plane a, and (w) = 0(1 w I). Fix an e > 0 and choose a ball B in a with center 0 such that
1)BCw, 2) I F(e(w)) I< e for w E B,
3)Iw-z
forwEB.
100
V. INTEGRAL PROPERTIES CHARACTERIZING 8-CLOSED FORMS
Condition 3) is guaranteed by the relations
Iw - Ow)I =ip(w)I = 0(I W I),
Iw - z I
fk
IZ+I2n J
Iy-z I2nI 1-
IIJ -Z+I -IZ-1 I 2n-1
IOkI
IJ -Z I I
I
W - z+ I2n
Similarly, Zk
Zk
I-Z-I2n
-Z+I2n
dlwlzzl I2n
(17.9)
Finally,
do < d1 dS.
(17.10)
Now condition 2) on B and (17.8)-(17.10) yield
If
z-) A df [k, m] A dg]
z+) A df [k, m] A d IY"
< ed2f
B (Iw12+y,+2)
n
ed3.
Here d3 is independent of z° and F. Thus the first part of the lemma is proved, and the second part follows from the first, as in Lemma 17.5. PROOF OF COROLLARY 17.3. Suppose f as in Corollary 17.3, i.e. f -= 0. Then,
by the jump formula for integrals of Martinelli-Bochner type (Theorem 2.6), we see that f + can be extended continuously to D and its restriction to aD coincides with f. Consider now a function (p harmonic in some neighborhood of D. There exists a domain G with smooth boundary such that D C G and 9) is harmonic on G. Then, by Green's formula (16.3),
(n -
ac
g(3,
z)
a
agars' z),do,
z E G.
101
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
Replacing the integral on the right by approximating Riemann sums, and the derivatives by difference quotients, we see that q' can be approximated uniformly in some neighborhoodU(D) of D by linear combinations of fractions of the form zk), zk (4 U(D), k = 1, 2,.... Since
(n -
z)
(-1)k-1
kl
[k] A
k
it follows from (17.2) that
f
aD
2 k=1
(-l)k-1 8 d [k] A 4 = 0 4k
(17.11)
for all (p harmonic in a neighborhood of D.
Consider now a sequence of domains Dm, m = 1, 2,..., with smooth boundaries, such that D. C Dm+ 1, U r D. = D, and lim
m-oo
fD,a =
f
(17.12) DC(
for all continuous (2n - 1)-forms a on D. By Stokes' formula
(-1)" fD f+ D.
n
(-1)k-1"
kd'
[k]
Add=
fDpµf+, D.
k=1
where µf+ _ k 1
A df. The integral on the left tends
to zero by virtue of (17.11) and (17.12), so that lim
m-=
f
(17.13)
q)Af- = 0
3D,
_
for all functions qv harmonic in a neighborhood of D.
We now want to show that (17.13) holds for all q7 continuous on D. Using Lemma 17.5 we finish the proof of case a) of Corollary 17.3. Choose Dm = (z E C": p(z) < -1/m). Then clearly (17.12) holds. It is easy to verify that
dt[k] A df IaD _
(2i)"(-1)"+k-1
ak
dom,
where dam is the surface element of 8Dm. Hence
(-2i)"Nf dam. Since Nf = 0, Nj can be extended continuously to Dby Lemma 17.5, and vanishes on D. Therefore (17.13) holds for all 'p in C(D). Choosing f+ as q,,
102
V. INTEGRAL PROPERTIES CHARACTERIZING 3-CLOSED FORMS
we then get by Stokes' theorem
1+µf. = (-2i)n f
faD,,,
af+
2
aDm k=,
2
dV-+0
8 k
as m - oo. Hence f+ is holomorphic in D. We have already shown that the restriction of f+ to aD coincides with f. We now consider case b) of Corollary 17.3. Let n = 2, and suppose 8(D) is connected and of class Cl. We have proved that if f E C(D) satisfies (17.2), then (17.11) holds for f. Consider the function
z,) +
z2))
1
-z,) -a( 2-f2) IJ -zI2 This is harmonic in
in D, = C2\(: b(3,
i,) - a( f2 - i2) = 0}. Let V be
a ball containing D. Then, for each z E V, we can choose a and b such that V C D,. As is easily verified,
dgAd -
Add. 1
Thus, by (17.11), faaD fdg A d = 0,
z E V.
But since 8D is connected, the above integral vanishes for all z E CD. We can now conclude the proof of Corollary 17.3 with the aid of Lemma 17.6. Indeed, we have shown that Hj - 0; therefore, by Lemma 17.6, Hj can be extended continuously to D and vanishes on 8D. Being harmonic in D, H7 therefore vanishes identically in D. Further, a ail
ail
1
z) =
47T2
z) _ -
42
a
dg n
8Z2
aaI dg A
Thus
a%f+(z)
4r2f
f+ is holomorphic in D. PROOF OF COROLLARY 17.4. Since f - is harmonic in CD (and CD is connected), we will have f -(z) = 0 for all z E CD if we show that f -(z) = 0
103
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
outside a ball B containing D. For z E B,
z) is harmonic in B with respect
to , and hence can be represented as a series in the basis of polynomials Pk,,(f ), converging uniformly in B. Hence the equality in Corollary 17.4 implies (17.2). Conversely, if f extends holomorphically into D as a function of class AC(D), then the above-mentioned equality follows from the fact that the integrands are a-closed forms.
To obtain another application of Corollary 17.3, we need the following notion. Suppose D is a bounded domain with aD E C', and let f e C(aD). We shall say that f is holomorphically continuable in dimension one if, for each complex line 1, corresponding to the restriction f = f 611 there exists F, E C(D fl 1), holomorphic at interior (relative to 1) points of D fl 1, such that
f=F, onaDfl/. COROLLARY 17.7. If D is bounded and aD E C2, and if f E C(aD) is holomorphically continuable in dimension one, then f is holomorphically continuable into D.
This is a consequence of Corollary 17.3 and the following lemma: LEMMA 17.8. If a function is holomorphically oontinuable in dimension one, then (17.2) holds for z E CD.
PROOF. Suppose z lies in the unbounded component of CD. By performing an orthogonal linear transformation and translation in C", we may ensure that z = 0 and that the plane {z: z1 = 0) does not meet D. This can be done for z far enough from D. Thus U0,0(', z) becomes 0). We now make the change of variables v j', j = vi v ' for j = 2, .... Then I
1
0
0
2 1
V2
0
dv=-vn+ldv.
u12
1
un
0 u12
Also, Observe that
0) is mapped biholomorphically onto C"\{v: v, = 0). n
I
k=1
(-1)k-1fkd$
n
[k] =
k=2i=1
kdU[k]
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
104
where
2 YIV
Y2V I I
...
"'V
1
In
Ak=
k = 1,
,
l1
[k]
k > 1.
0,
Jlvn
J2vn
...
rtvn
Hence Uo.oG ,O)
1)Rdv1 v
n(
diF[I] v212 +
l1 + (
vRI2)
=
dvI
R I
AX('v).
v
Here, 'v = (02,... , va ). Under our change of variables, the lines (: I = t, 2
= cat) go ov r to the lines {v: v2 = c2,...,va = ca), and the boundary of D to a cert ' cycle F. If f is holomorphically continuable in dimension one, then, after th change of variables, the new function f(v) will be continuable in dimension one along all the lines (v: v2 = c2,. .. , va = CO. Furthermore,
f
frf(v) vU' A A('v) = f
v
where it is the projection v - 'v. Note that, by Sard's theorem, the sr-'('v) fl r are smooth curves for almost all 'v E ir(T). The origin lies outside the domain bounded by r; hence the inside integral above vanishes for almost all 'v, i.e.
(17.2) holds at points z sufficiently far from D. Since an integral of the Martinelli-Bochner type is harmonic, it follows that (17.2) holds on the
_ unbounded component of CD. Suppose now that z belongs to a bounded component Q of CD, and suppose z = 0 (we can always achieve this by performing a translation). Consider a cone III with vertex 0, formed by complex lines through 0, where e denotes the area of the portion of 3Q lying inside H. By the same change of variables as before, this cone goes over into the cylinder formed by the lines (v: v2 = c2,...,va = ca). And 1e can be chosen so that: 1) as e --> 0, the area of 8(Q fl II,) fl aD tends to 0, therefore
faDf(oUO.a,O) = a-.0 lim faD\a(Qnri.) fm Uo.&,O),
105
§17. HOLOMORPHY OF CONTINUOUS FUNCTIONS
and 2) aD\a(Q n H) maps to a compact set 1'E under the change of variables. Then it can be shown, as before, that
ff(v) Uv' A X('v) = 0. r. Hence (17.2) also holds for z E Q.
Corollary 17.3 is closely related to the criterion for the holomorphy of integrals of Martinelli-Bochner type. THEOREM 17.9. Suppose n > 1, and D is bounded with aD connected and of
class C'; and let F E C'(0D). If F+ is holomorphic in D, then F+ extends continuously to D and its restriction to aD coincides with F.
THEOREM 17.10. Suppose n > 1, and D is a bounded domain with aD connected and of class C2. Let F E C(aD). If F+ is holomorphic in D and extends continuously to D, then its restriction to OD coincides with F.
The content of the above two theorems is that, in contrast to the onedimensional case (integrals of Cauchy type), an integral of Martinelli-Bochner type is holomorphic in the case n > I if and only if it is a Martinelli-Bochner integral, i.e. F+ = F on aD (or equivalently, F 0). The condition that aD be connected cannot be dropped, as the example of the spherical shell D = (z: I < I z I < 2) and the function 1, 1
F E C'(0D), but F+ = 1, i.e. F+ does not coincide with F on 3B(0,1) C 3D. For the proof of Theorems 17.9 and 17.10, we need the following assertion on the saltus of derivatives of an integral of Martinelli-Bochner type (Aronov [3l).
LEMMA 17.11. Let D be a bounded domain with aD E C': D = (z E C": p(z) < 0), where p E C'(C") and gradp IaD 0. If F E C'(0D), then
Z+mo
aF+(z+)
- aF-(z)
3F(z°)
li
aZk
aZk
aft,
- Pk
aF(z°)
"
aym
Pm,
1
uniformly in z°. Here z° E 3D, z"' lie on the normal to aD at z°, z+ E D,
z- (M I z+ -z° l_I z--z° , and Pk(Z0) =
ap a
0) Igrad
PPk = Pk
106
V. INTEGRAL PROPERTIES CHARACTERIZING 2-CLOSED FORMS
PROOF OF THEOREM 17.9. We extend F to a function f of class C' in a neighborhood of D. For F, the Martinelli-Bochner formula for smooth functions (the special case p = q = 0 of (1.1)) is valid:
F(z) = 1D0
A
z) -
z)
- f aFQ) A Uo,&, z).
(17.14)
D
The second integral in (17.14) is the convolution of a bounded compactly supported measurable function with the locally integrable functions I fk - Zk I - z I-z", and hence is continuous in C". Consequently the first integral, i.e. F+ , extends continuously to D. Now F+ is holomorphic in D, and therefore I
aF+/azk = 0; hence it follows from Lemma 17.11 that the 8F-/ask extend continuously to CD and their restriction to aD has the form OF-
aF
aZ k
aZk
aF
"
k = 1,2,...,n.
Pk E aZ Pm, m=I
Multiplying (17.15) by Pk and summing over k, we then get " n " aF nn aaF
aFPk a k= IZk
7, PkPk
Pk
k=1
(17.15)
m
k= I
Zk
M=I az,"
_ 0' Pm -
since n
2 PkPk = 1. k=I
Next, consider the form
_ µF--
I
az k dz[k] A dz.
k=1
Recall that (2i)"(-1)1?+k-IPkdh,
dz[k] A da I8D = where da is the surface element of aD. Hence
µF-I8D = (-2i)" 2
k=1
aZ
Pkdo = 0. k
By Stokes' formula, we deduce that n
0= f
8D
N'F-= (-2i)" f I CD k= I
aF aZk
z
dv;
(17.16)
§18. TRACES ON THE SHILOV BOUNDARY
107
this is permissible, since F-= 0(1 z 1'-2"), and aF-/alk = 0(1 z I 2-2") as oo. (17.16) means that F- is holomorphic outside D. Since aD is IzI connected, Hartogs' theorem now shows that F- extends holomorphically into
D. However, F - 0 as I z I-. oo; hence F = 0. By Theorem 2.6, F+ extends continuously to D and F+ Ian = F. PROOF OF THEOREM 17.10. Let zo and z' be as in Lemma 17.11. Apply Lemma 17.5. In our case NF = 0, and so NF extends continuously to CD and Ni 6 = 0. The proof proceeds from this point exactly in the same way as that of Theorem 17.9. Note that neither of the Theorems 17.9 or 17.10 contains the other. They can
be regarded as results concerning the possibility of extending F from aD holomorphically into D. §18. The traces of holomorphic functions on the Shilov boundary of a circular domain
Let D be a bounded strongly starlike circular domain in C", i.e. a domain invariant under all rotations z -+ eiez, 0 < 0 < 21T, such that AD = (Az: z E D) C D for any A, 0 < A < 1. Let S denote the Shilov boundary of D, i.e. the smallest closed set lying in D and such that sup( I f(z) I : z E D) = sup( I f(z) I : z E S)
for all f E AC(D).
If D is strictly pseudoconvex, then S = aD; in general S is a subset of W. Before proceeding to the description of the traces on S of functions from A C(D), we give a characterization of polynomials orthogonal to A C(D). Let µ
be a nonnegative measure on S, invariant under the rotations z -i e'9z, 0 < 0 < 2 a, and massive on S, i.e. S \ E = S for every E C S with µ(E) = 0. Denote by [AC(D)Jµ the space of functions p E C(S) such that fsq>fdµ = 0 for all f E AC(D). If D is the unit ball and µ = w(a, z), then [AC(D)Jµ = 0(D). Let P2(µ) be the space of polynomials in z1,...,X", with the inner product
(f, g) = fsfgd
.
If the multi-indices I and J are such that 111110 II J II, then by the invariance of µ under rotations we have (z' z') = f z'a'dµ = e`a(urn-HJ11) l'zrz1dµ S
S
=
e`e(uni-i°JU)(zI,
zJ)
108
V. INTEGRAL PROPERTIES CHARACTERIZING 3-CLOSED FORMS
Hence (z', zJ) = 0; thus, if f and g are any homogeneous polynomials of different degrees, then
(f,g)=0.
(18.1)
If follows that we can choose a complete orthonormal system of polynomials {(p;) in P2(µ), where I is the degree of the homogeneous polynomial, and k its index among those of degree 1. For multi-indices I and J with 11111 < II J II, we
put
P,,l(z)
= GaJJkcIIJII-11111(2), k
where aUk = SZ 1-J Z
k
dµ.
LEMMA 18.1. A polynomial P(z, z) lies in IAC(D)]µ if and only if it is a linear combination of polynomials of the form
z'i', IIIII > IIJII, z'z'-P1J(z), IIIII
(18.2)
(18.3)
PROOF. Sufficiency. D is strongly star-like, so functions from ACID) can be approximated uniformly on D by polynomials. Since (T') is complete, we need only prove that, for a polynomial P(z, z) of the form (18.2) or (18.3) and for all k and 1,
fSP(z, z)gvldµ = 0.
(18.4)
If P(z, a) is of the form (t8.2), then by (18.1) JP(22)Tidµ=(z'q S
,z')=0,
since z'p is a homogeneous polynomial of degree IIIII + 1 > II J II. Next, if P(z, z) is of the form (18.3) and 0 IIJII - IIIII, we have (using (18.1) again) 1!/
fSP(2' z)4'idµ = (z'p
,
(92i, PiJ(z)) = 0.
Finally, if 1= II J II - IIIII, then (18.4) holds by the choice of the a,Jk. Necessity. Let P(z, z) _ 7,1,J bJZ,zJ belong to [AC(D)]µ ; then bjj(z'?'
P(z, z) II111>IIJII
II
III
- PPJ(z) )
§18. TRACES ON THE SHILOV BOUNDARY
109
i.e. P(z, Y) = P,(z, Z) + P(z), where P1 is a linear combination of polynomials
of the form (18.2) or (18.3), and P(z) is a holomorphic polynomial. Since P(z, z) E [AC(D)]µ by assumption, and P1(z, a) E [AC(D)]µ by what we have already proved, we see that P(z) E [AC(D)Iµ. Hence
f
S
IP(z)12
dµ =
f P(z) P(z)dµ = 0. s
Since ,u is a nonnegative and massive on S, it follows that P(z) = 0 on S, and hence that P(z) = 0 on U (S is the Shilov boundary). Thus P(z, a) = P1(z, a ). Q.E.D. If D is n-circular, i.e. invariant under all n-rotations z -
0<
< 21r, and if µ is also invariant under n-rotations, then as a
complete orthonormal system we an choose (alz'), consisting of monomials with a/ = (f z /z l d1t)-1 /2 In this case Lemma 18.1 becomes LEMMA 18.2. The polynomial P(z, z) lies in [AC(D)Iµ if and only if it is a linear combination of polynomials of the form z'Z1,
with lk > jk for some k,
z/zI+J( a/2
+J -
2 al+J+e,*IZ
2 ,
(18.5)
where ek is the multi-index with I at the kth place an 0 elsewhere. Lemmas 18.1 and 18.2 are analogues of Lemma 5.7. When D is the unit ball, w = z and u = w(z, a), Lemma 18.2 and Lemma 5.7 are identical.
For the proof of the main result, we need the Poisson kernel. First we construct the Szego kernel h(f, z) of the circular domain D by the formula
h(f,z)=I
k,I
It can be shown that: _ 1) z) is holomorphic in z on D and continuous in on D;
3) for all f EAC(D),
f(Z) = f
z)dµ;.
S
The Poisson kernel can be constructed from the Szego kernel:
P(3, Z) _ I h(f, z) I2 h(z, z)
(18.6)
110
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
For z E D we have h(z, z) > 0, since (18.6) and 2) above imply z
h(i, z) = fsh(z, f)h(f, z)dµt = fs I h(f, z)I dµt > 0. Further, if f E ACID ), then
f(z) = h(Z z)f(z)h(Z, z) = h(z z)
z)dµt
z)dµt.
=
(18.7)
THEOREM 18.3. Let D be a bounded strongly star-like circular domain in Cn
such that every point of its Shilov boundary S is a peak point for the algebra AC(D), and let µ be a nonnegative massive measure on S invariant under the rotations z -> e'Bz, 0 < 0 < 27r. Then a function p E C(S) is of the form f Is for some f E AC(D) if and only if
f (pPdµ=0
(18.8)
for every polynomial P of the form (18.2) or (18.3).
If D is n-circular and It is invariant under all the n-rotations z (ei9lz,,...,e'B^zn), 0 <
< 21r, then we may replace (18.2) and (18.3) above by (18.5). PROOF. The necessity follows from Lemma 18.1.
Sufficiency. Since each point of S is a peak-point for AC(D), we have for each z0 E S and any neighborhood W of z°
urn f
z-.z0 S\w
z)dpr = 0.
(18.9)
zED
The argument for this can be found in Koran [1]. Further, since and
f
z)di = 1,
z) > 0
z E D,
S
it follows from (18.9) (see, for example, Hoffman [1], p. 18) that if p E C(S), then the function ,W(z),
f(z) =
f
z E S, z)dµt.,
z E D,
§19. COMPUTATION OF MARTINELLI-BOCHNER INTEGRAL
111
is continuous on D U S. Since S is the Shilov boundary of D, ( f(rz)) therefore
converges uniformly on D as r --> 1; hence f extends continuously to D. It remains to show that, if p satisfies (18.8), then f E A(D). From (18.7), we see
that, for any g E AC(D) and any differential operator T with constant coefficients,
z))dµt = 0. It follows in particular that
z)) L=0 E[AC(D)]
7Z
From the form of the Poisson and Szego kernels, we see that the function on
the left above is a polynomial in and f. By Lemma 18.1, it is a linear combination of polynomials of the form (18.2) and (18.3). This and the hypothesis of the theorem imply that
fsz))] I:=0dµt = 0, i.e. all the coefficients of the form
8f(z) = f
z)dµt.
(18.10)
S
and all their derivatives vanish at z = 0. Since the Poisson kernel is realanalytic in z, the form (18.10) has real-analytic coefficients; hence of ° 0 on D, i.e. f E A(D). §19. Computation of an integral of Martinelli-Bochuer type for the case of the ball We consider on the unit sphere M(0, 1) the space L2(aB(0,1)) with the usual inner product
(.f, g) = L(0, f . do, 1)
where do is the surface element of the sphere. We shall identify this space with
the space G2(B(0,1)) of harmonic functions f in B(0,1) represented by a Poisson integral from L2(aB(0,1)). Let 'J's,t be the finite-dimensional vector space consisting of homogeneous harmonic polynomials of the form Ps.t =
7.
1 a1 zIz-J .
111II=s 11J11=t
112
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
The union of all the 6J',, is dense in L2(aB(0,1)). As we know, (17.1) is valid for f E G2(B(0,1)) fl A(B(0,1)):
(Mf)(z)
faB(o. I)
z) =f(z),
z E B(0, 1).
Then the Martinelli-Bochner type integral of PS , is
LEMMA 19.1 Let PS , E
of the form
MP",
_
n+s-1 n+s+t -
1Ps,t
This follows from the fact that the harmonic extension of k Ps,, from aB(0,1) into B(0, 1) is given by
1-Ifl2
kPsa + n + s + t Ia
a
Ps,r, k
and from the easily verified formula U0,0G, Z) IaB(o,1) = 11
here
-I
Z IZ
PQ, z)da;
z) is the Poisson kernel for the ball:
M,Z)_(n
_ 1)! 1-IZ 2
IZI2n
COROLLARY 19.2. If
Qs,t = 2
:E
Z
11111=s
I the R3_p,,_p are the harmonic polynomials defined by
RS-°'`-°
_
s+p+n-1
p!(s+t+n-p- 1)! (_l)j(s+t-j-2p+n-2)!lZI2jAj+PQs XI t,
j
jao
A being the Laplace operator:
0R=
a2R
j=, azjaaj
§19. COMPUTATION OF MARTINEL1I-BOCHNER INTEGRAL
113
This corollary follows from Lemma 19.1 and the Gauss representation of an
arbitrary polynomial on the sphere as a sum of homogeneous harmonic polynomials (see, for example, Sobolev [1], Chapter 13). We introduce the differential operators b and b by setting
bzkaa, bZkaa k=1
k=1
k
k
THEOREM 19.3. Suppose f E G2(B(0,1)). Then the Martinelli-Bochner type integral off has the form n
Mf=f-b4-f-
rr,,
(19.1)
Zk'rkl k=1
where IZI'-n t'IZI
/
IJIx-2f(
)dIJl,
-IZI_nf I=I
k
a=k
f E G2(B(0,1)), in the interior of B(0,1) an expansion f(t) = 2 Ps,t(0,
(19.2)
s, t,0
converging uniformly on compact subsets of B(0,1), in particular on each sphere aB(O, r), 0 < r < 1. From the definition of the class G2 (B(O, 1)), it is not hard to show that (19.2) is also valid on aB(0,1), but in the sense of convergence in L2(aB(0,1)). By Lemma 19.1, we deduce that, in the interior of B(0, 1),
Mf
n+s-1
_ o
s
t
n+s+t -
,.0 n+s+t -
1p=f-
To To prove the first part of (19.1), we observe that =IZI1-n
4i(Z)
I s, ta0 0
,In-2+t+s ps't(I
IIzI
I
z lr+s
s,O n+s+t-
B_
1
s
o
I)
I
)d1
l
Z so n+s+t - IPS"
n
apsr
klzk aik n+s+t- 1=
-
t
s
o
n+s+t -
Ps`'
114
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
The proof of the second part of (19.1) is similar. The above theorem enables us to compute Mf for any f E Gz(B(0,1)). The functions 4, and 'Yk are harmonic in B(0, 1). To compute them, it is convenient
to change over to polar coordinates. If f is polyharmonic or real-analytic in
B(0, 1), then it may be convenient to compute Mf by using the Almani representation off (see Sobolev [1], Chapter 14): fkIZI2k,
!{ = G k>O
with the fk harmonic in B(0,1). We present some examples to illustrate the applications of Theorem 19.3. EXAMPLE 1. Let f E Gz(B(0,1)), and Tf
_ (n - 2)! r (2ir)
JaB(o,
1)
-
I
z E B(0,1) , Z
the potential of a simple layer. Then Tf=IZI' -R f l=y
In particular, if Tf = F, then f = (n - 1)F + bF + bF. Consider the integral equation
f+XTf=q2,
pEGz(B(0,1))
(19.3)
on aB(0, 1). Then its solution is of the form AIzI'-R-a
f=m-
f
a+R-z
IzI
IAI < n - 1.
ICI
(19.4)
EXAMPLE 2. Let f E Lz(aB(0,1)), and
Sf = 2fBco
z),
z E aB(O,1),
be the singular integral of the Martinelli-Bochner type. Consider the singular equation
f+ XSf= p,
T E Lz(aB(O,1)).
(19.5)
If p has an expansion qv = F+s=O Pps,gs in B(0,1), where p and q are nonnegative integers, then
f=
2Xq(p + q)(n - 1) IzI-p p+q 1P (1 - A)q + (1 + X)p [(1 - A)q + (1 + X)p]2 0 (19.6)
< IXI
(n - 1)(1 + X)(p + q) (1 - A)q+ (1 +X)p
§19. COMPUTATION OF MARTINELLI-BOCHNER INTEGRAL
115
If we know only the values of p on 8B(0,1), we must replace p in (19.4) and (19.6) by its Poisson integral. Then we obtain integral representations for the solutions of (19.3) and (19.5), i.e. resolvents of the operators T and S. To compute integrals similar to Mf in more general domains than the ball, we may use the ideas of §5. We will have to require that (5.8) holds, or that the polynomials P(z, w) are dense in L2(aD). We indicate other applications of Lemma 19.1. The following theorem is an immediate consequence of the lemma.
THEOREM 19.4. For n > 1 the operator M maps L2(aB(O,1)) into itself, is bounded, and has norm 1. All positive rational numbers in [0, 1] are eigenvalues of infinite multiplicity for M. The spectrum of M is [0, 1].
Denote by PrH the orthogonal projection of G2(B(0,1)) into the Hardy space H2(B(0,1)) consisting of elements of A(B(0,1)) satisfying the condition that JaB(o, r) I f I2 do is uniformly bounded for 0 < r < 1. This operator is called Szego's operator, and has the integral representationp 1)k-IJkdJp k] A d k=1 ((PrHf)(z) = fB(0 1)f(J) (s., z))n (
For n = 1, integrals of Martinelli-Bochner type are integrals of Cauchy type, and M = PrH. For n > 1, the kernel U00 is not holomorphic, and M and PrH are not the same, as the following example shows. EXAMPLE 3. Let f = Zk. Then r is not holomorphic in B(0,1). Indeed,
ak
(z) (n - 1)I (27ri)
/
Zm)(Zn+2
kn
JaB(0.1)
Zk)
- ZI
M=I
Iz It
I2n
dF [k] A d3
Since
z)
-(n-1)1
(f mZm)m do ,
IaB(o,l) =
1
I
- z I2n
we have a
+
aZkf Iz-0 =
(n - 1)I 271n
and the integral vanishes only if n = 1.
(n - l)°8B(0,1) f
do,
dJ [m] A d .
116
V. INTEGRAL PROPERTIES CHARACTERIZING 8-CLOSED FORMS
However, the following theorem is true: THEOREM 19.5. For every f E L2(aB(0,1)),
lim Mkf = PrH f,
(19.7)
where Mk denotes the kth iterate of M.
PROOF. For each homogeneous harmonic polynomial P,,,, (19.7) follows immediately from Lemma 19.1, since PSt, 0,
PrHP,,
ift=0, ift>0.
But such polynomials are dense in G2(B(0,1)), and the family of iterates of
M is uniformly bounded in norm; hence (see Dunford and Schwartz [1], Chapter II, §3, Theorem 6) (19.7) holds on all of L2(B(0,1)). §20. Differential boundary conditions for the holomorphy of functions
Let D = (z E C": p(z) < 0) be a bounded domain with aD E C'. We consider the following problem: suppose we are given a function f E C'(D) harmonic in D, and a continuous vector field w = (w,(z),...,w"(z)) on aD, with
w, grad p ) =
l'
W az
0,
z E aD,
(20.1)
i.e. the vector w never lies in the complex tangent plane TaD(z) at z to aD for any z E aD. Then does the condition
of k
84
wk = 0,
z E OD,
(20.2)
imply that f is holomorphic in D? This problem is an analogue of the oblique derivative problem for real-valued harmonic functions. If we do not require (20.1), then there does exist an f e A(D) satisfying (20.2), e.g. D = B(0,1) C
C2, w=(0,1)andf=Z1. When w = grad p, (20.2) becomes the Neumann normal a-condition for functions harmonic in D, and the answer to our question is affirmative: THEOREM 20.1. Let D be a bounded domain in C", and let f E C(D) be harmonic in D. Then f EA(D) if and only if the a f/aZk, k = 1,... , n, can be extended continuously to D and satisfy the Neumann normal a-condition
2 k=I
of aZk
ap = 0, aZk
z E aD.
(20.3)
117
§20. DIFFERENTIAL BOUNDARY CONDITIONS FOR HOLOMORPHY
PROOF. If f E A(D), then all of/a k = 0, and (20.3) is trivial. Conversely, if a f/azk E C(D), k = 1,. . . , n, and (20.3), holds, then
(-2i)" I
µI IaD =
k=1
of aZk
ap -
aZk
do = 0
where do is the surface element of aD. As in the proof of Theorem 17.3, we now conclude easily that f E A(D). Let us reformulate our problem differently. Decompose w into its normal and tangential components to aD at a point z E aD: w(z) = a(z)grad p + T(z);
here a never vanishes on aD, and T(z)
is
orthogonal to grad p,
i.e.
(grad p, T) = 0 on aD. Then (20.2) becomes
f aP ----71
"
"
k=I
k=1 aZk aZk
of Tk(Z) azk
a(Z)
(20.4)
Consider again the form
µf=
"
of Ada. (-1) k-1 aZkdd[k]
k=1
Using (20.4), we can write it on aD as
A f IaD = 2 ak,m(z)df A da [k, m] A dz IaD,
(20.5)
k>m
where the ak.m E C(aD). Then the new formulation of our problem is: suppose f E CI(D) is harmonic in D, and (20.5) holds; does it follow that f is holomorphic in D? (20.5) reduces to the Neumann normal a-condition when all the ak,m are zero. Condition (20.5) can also be written in integral form:
AZ) = aDf fa) with
z) + (-1)" 2 d(gak.m) A df [k, m ]A
0
k>m
111
z) = -(n - 2)!(21ri)-"
- z 12-2n. It is then clear that the problem
of the holomorphy of the functions defined by a Martinelli-Bochner integral (see Theorems 17.1 and 17.2) is a special case of the present problem (the case
ak,m=0) We now present the solution to our problem for certain classes of domains D and functions ak,m. THEOREM 20.2. If n = 2, and OD is connected and of class C', and a harmonic function f E CI (D) satisfies the condition
µl IaD = a df A d IaD,
where a is a constant, then f E A(D).
(20.6)
118
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
PROOF. If we show that, for each Ps, E
(see §19 for the notation),
s,t=0,1,2,..., 2
f
I (-1)k
ap- df
k=1
aD
[k] A
(20.7)
O,
holds, then our assertion will follow from Corollary 17.4. By (20.6) and Stokes' theorem, 2
faDf(U) k=I 2 (-1)
k_,aP
ak`df [ k ] A
d
=fDPs,rµf
= f aP5',df A d ID
fa of dP3,, A aD
If t = 0, i.e. Ps., is a holomorphic polynomial, (20.7) holds. Let t > 0. We transform dPs,, A d into the form
Add.
a
This can be achieved by setting P
(formal integration with respect to 2). Then aP
a,
f
z
a Ps,r
2
2
f
a Ps.r
aP'` are
2
Thus
so that iP = 0. Also, P E
faDfq) k=1 (-1)
=
k ,aP 4k
=a
1 d [k] A d
f
aD
2
I
(-1)k-I ak
[k] A
k=t
(20.8)
Now we apply induction on t, using the fact that (20.7) holds for t = 0, and conclude from (20.8) that (20.7) is valid for all t. THEOREM 20.3. Let D = B(0,1) and suppose that the ak,, lie in AC(D). Then, if f E C'(D) is harmonic in D and satisfies (20.5), it follows that f E A(D). PROOF. Consider first the case µ1(aB(O.I) =
where Q/,o E 91,0.
A df [k,m] A d IaBto,I)'
(20.9)
§20. DIFFERENTIAL BOUNDARY CONDITIONS FOR HOLOMORPHY
119
Writing out (20.9) for the case of the ball, we have
J
E OB(0, 1).
JkJ,
Q1,0(J)I a kSm
Jj
j2l
(20.10)
J
L
We shall look for the solution of (20.10) in the form of a series
f = 2 Pt+ s,t>0
where PS,t E 6S,,,. Then (20.10) yields t
s,t '0
6ps.t
BPS.t
2
=
s,ta0
(gym
-k
BPS
(20.11)
aSk
We now extend the function on the right above harmonically into B(0,1) using Gauss' formula (see, for example, Sobolev [1], Chapter 11): if Pk is any homogeneous polynomial of degree k, then its harmonic extension Pr from M(0, 1) into B(0,1) is given by
Pr = 2 Zk-2s sao
where
4-2.,
_
k-2s+n-1 s!(k+n-s- 1)!
7 (-) 1 j(k-j-2s+n-2)!
j!
J '0
1 zzJAJ+sp
k
(20.12) In our case
Pk = Q10
(0=- - Jk
ymt)
ay t Jk
,
JJ
and k = 1 + s + t. Hence (20.11) leads to the system of equations tPs
2
8PS-1-_"4+i Q1,0(
m-
aJk
aJm
+ J2 bjjflzjOj+' [Qi,o ( BPSakr+z
+ 2 C, p-0
(JI2JAJ+1
Qto
aPs-1,t+1+,
afk
ePa m-
+z
k) + .. .
aPs-1,t+1+i Jk aim
,
(20.13)
where the a j, by,... , c j are constants. For s = 0, we have tPo,t = 0 for any t by (20.13); hence Po.t = 0 for t 0. Next we have 1P,,, = 0,. . . , tP,,t = 0, so that the series for f consists only of holomorphic summands, i.e. f is holomorphic in B(0,1).
V. INTEGRAL PROPERTIES CHARACTERIZING 8-CLOSED FORMS
120
If the ak m are arbitrary holomorphic functions, we can decompose them into sums of homogeneous polynomials and apply the earlier considerations. Then we get a system of equations of the type of (20.13), with each equation involving only finitely many summands. In the case of B(0, r), the Neumann normal a-condition has the form n
k=1
0
Zk
on aB(O, r).
a' kk
Suppose now that this condition is satisfied everywhere in B(0, r). Then it turns out that the requirement that f be harmonic can be dropped, and f will still be holomorphic in B(0, r) as before: THEOREM 20.4. Let f E C'(B(O, r)) n C°°(0), and suppose that
z E B(0, r).
= 0, 2 Zk -Lf aZ
k=1
(20.14)
k
Then f is holomorphic in B(0, r). If f E C2(B(, r)) n C°°(0) and a2f = 0, 2 zk zm aZkaZm
z E B(0, r),
(20.15)
k m= I
then f is pluriharmonic in B(0, r). PROOF. Observe that (20.14) implies that f is holomorphic on each complex line 1 through 0, and (20.15) implies that f is harmonic on each such line. Thus we need only appeal to a result of Forelli [1], which asserts that an f which is harmonic on each complex line I through 0 and C°° at 0 is pluriharmonic in B(0, r).
Rudin [1] has shown that, if a function f in B(0, r) satisfies the Laplace equation A f = 0 and the Laplace-Beltrami equation A'f = 0, where n
k m= I
a2 ZkZm aZkaZm
then f is pluriharmonic in B(0, r). Theorem 20.4 shows that, if f E C2(B) n C°°(0), then the single equation (20.15) is sufficient. The single condition Al f = 0 is not sufficient. For example, let
f=
042 -IZ212)
k! (n + 1)!
2 k>0 (n + k + 1)!
IZ12k
(20.16)
Then f is real-analytic in B(0,1), since the radius of convergence of (20.16) is 1. It is easy to verify that 0'f = 0, but f is not pluriharmonic, since
as k.m
kZ
a k
(j + 1)! (j + l)(n + 1)! (n +j + 1)! j>0
= (1=112 -IZ212) 71 m
IZ12'.
§20. DIFFERENTIAL BOUNDARY CONDITIONS FOR HOLOMORPHY
121
If f E Ck(B(0, r)), Theorem 20.4 is not true; e.g. let IZ'I4
f(Z)=Z12'
IZI4
Since In I zj I4 < I Z I4, we see that, by choosing a multi-index I with 11111 sufficiently large, we can have f E Ck(B(0, r)) for any prescribed k. Clearly f is not holomorphic in B(0, r), although I; akaf/8zk = 0. Another theorem related to the above theorems by the method of proof is the following:
THEOREM 20.5. Let Pk be an arbitrary homogeneous polynomial of degree k > 2. Then the restriction of Pk to aB(0, r) can be extended holomorphically into B(0, r) if and only if LkPk = 0 on aB(0, r), where Lk
_ j2n zj-Z.a
(n + k - 21 - 3)! (21 - ])!!21,y+
(n+k-2)!(1+1)!
tI Pk=
2
:E
a1jZ/Z-J.
11111=m 11J11=t
We extend Pk from aB(0, r) into B(0, r) by Gauss' formula; the extension,
call it PI., can be written as P. = 2,-0 Zk-2s+ where Zk_2., is defined by (20.12). P. is holomorphic if and only if n
i=1
aPT a.-=0 az
in B(0, r) (see Theorem 20.4). But n
j= 1
aZk-2s =
- S)Zk-2s
aZj
Then
z aPr
j=I
azi
_
4'Pk I.0
(t-s)(-1)'(k-2s+n- 1)(k-j-2s+n-2)!Aj+sP k s!j!(k+n-s- I)!
. (-l)'(k-2s+n-1)(k-1-s+n-2)!(t-s) I s .(l - s). (k + n - s - 1) ,
j+s=t
(-1)'t1'Pk two
s=0
i .
(-1)s(k-2s+n- 1)(k - I- s + n - 2)!(t -s) s!(l - s)!(k + n - s - 1)!
The following lemma now concludes the proof of the theorem.
122
V. INTEGRAL PROPERTIES CHARACTERIZING a-CLOSED FORMS
LEMMA 20.6. If k % 21, then
S=o
(-1)s(k-2s+n- 1)(k-l-s+n-2)! s!(1-s)!(k+n-s- 1)!
- J0, 1,
ifl>0, ifl=0,
and
(-I)S(k-2s+n- 1)(k - I - s + n - 2)!s S=0
i s.(1-s).(k+n-s1). I
(-1)'(n+k- 21- 1)!(21-2)! _ (n+k-2)!1!(1- 1)! 0, ifl=0.
ifl>0,
PROOF. Let us compute for instance the first sum, which we denote by S. We have
S
(-1)I(k-21+n- 1)(k - 21 + n - 2)! 1! (k+n - I - 1)!
+(-1)'-'(k-21+n+1)(k-21+n-1)!+ (-1)'(k-21+n-1)! f 1+k-21+n+1 k+n-1 -1)!(k+n-1-1)!L
+
1
_
(-1)'-I(k-21+n- 0! (1- 1)(k + n - 21) + 1! (k + n - 1)!
(-1)1-1(k - 21 + n)! (1- 1)
1!(k+n-1)! + (-1)t+2(k_21+n+3)(k-21+n)!
(1-2)!2!(k+n-I+ 1)!
+
(-1)1+2(k-21+n+ 1)!(I-2) 2!1(1- 2)!(k + n - 1 + 1)!
(k-l+n-2)! (1- 1)!1(k + n - 2)!
+
+
(k+n- 1)(k-I+n-2)! = 0. 1!(k + n - 1)!
The second sum can be computed in the same way.
§20. DIFFERENTIAL BOUNDARY CONDITIONS FOR HOLOMORPHY
123
COROLLARY 20.7. Let P = Eo Pk be an arbitrary polynomial, where the Pk are homogeneous polynomials of degree k. Then P can be extended holomorphically from 8B(0, r) into B(0, r) if and only if nt
k=0
k Pk IaB(0, r) = 0'
We note that the condition LkPk = 0 on 8B(0, r) may be replaced by the condition EkPk = 0 in B, where
. J az
(n + k - 21 - 3)! (21 - 1)!!211 I21+2 !+I
a
k
J=I .:E
no
(n+k-2)!(1+1)!
z
CHAPTER VI
FORMS ORTHOGONAL TO HOLOMORPHIC FORMS. WEIGHTED FORMULA FOR SOLVING THE a-EQUATION, AND APPLICATIONS §21. Forms orthogonal to holomorphic forms
In this section, we sharpen the results of §6. First some ancillary results: LEMMA 21.1. Let D be a bounded strictly pseudoconvex domain in C". Then there exist positive constants r, 6 and c such that? 6), we can perform a smooth change of I< r in I) For any with I variables T = T(z) with the properties T2 = Im z); 1) Tl = P(z) -
2)1z11C>I T(Z)I>CIZI ;ZE 6). 3) l/c >I aT/az I> c; z E II) For any z such that I p(z) 1< r in B(z, 8), we can perform a smooth change of variables T = T(') with the properties
r2=ImIQ,z); ZI
EB(Z,6);
,
3) 1/c>IaT/a I>c; E B(z,s). The function I , z) was introduced in §4.2°. PROOF. We need one more property of Khenkin's barrier function not noted in §4.2°: there exist r > 0 and a nowhere-vanishing continuous function h on {z: I p(z) < r} such that, for z E V with p(z) (< r, dz(DQ, z) l==t = dt
z) IZ=t = hdp
(see Ovrelid [1]).
From this we conclude that the maps
Ti = p(z), p {T2 = Imo (l, Z),
y
and 125'
T1 = P(r) T2 = Im(D(J, z),
VI. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
126
and grad p(z)' 0. The lemma follows
have rank 2 at points where z = easily from this. LEMMA 21.2. Let
E)adTI
(T1 +
/ I*1
rI+E>0
... dT2n
/(I T1 I +I T2 I +E + I T I2)k I T I/'
where a, k and I are real and a > -1. Then E a) I 00) E1-k+a; k - a > 1; b) Ia,k,2n-3-2m(E) = O(1)E k+a+m+5/2;
k - a - m > 5/2; m > 0.
PROOF. In the integral defining la,k,l, we introduce spherical coordinates T1 = r cos p 1, T2 = r sin T 1 cos 9P2' .. , ,
and perform the change of variables
sj = cos Tj, j = 1, 2. Then
+
(rs1
E) a r 2n-1
-/drdsl ds2
kkd(e)
k
[r(I51 I+52(1 - 51 )1/2) + e + r2]
IS
rs, +e>O
We use the inequality I s1 I +52(1 - S1 )1/2 4A st I +s2) for I s I- 1 and 0 < s2 < 1. Setting s1 = su and s2 = s(1 - I u I), we have
f
Ia,k,I(E) = 0(1) 0
(rsu + 00'r2n-1-lsdrdsdu
(rs + e + r2)k
Ju1<1
rsu+e>O
For 1= n - 1, we go over to the coordinates (v, s, u), where v = e-1rs. Since
rs+E+r2>rs+e, we get I«.k.2n-1(e) = 0(1) dv
El-k+a f0l (v +
l (vu + 1)adu + f 4e'
k
I) k
1
1
dv
k
1
(v + I)
(vu +
,du l.
-v
Assertion a) follows. In the case b), we have
f r 2mdr f 2
la,k.2n-3-2m(E) = 0(1)
= O(1) f 2r2mdrf 2
=0(1) fo
2
(re + r
2
rds
f rs + e + r2)k (rs
t
(rsu + e)arsdu
-e(rs)-I
rds
(rs + E
2)k-a-1
dr )2(k-a-m-2)
Assertion b) now follows by direct integration.
- O(1)
2mdr
!2 0
r2)k+a-2
(e +
§21. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
127
LEMMA 21.3. Let D C R" be a bounded domain with smooth boundary and let for 0 < A < 1. Then f E f e C'(D). S u p p o s e I grad f(z) I= O(1) I p(z) I'
C0 '(D). This assertion is a slight generalization of a result of Hardy and Littlewood (see, for example, Goluzin [1], Chapter IX, §5, Theorem 3).
We now improve the estimate for the solution of the a-problem given by Theorem 4.4. THEOREM 21.4. Let D be a bounded strictly pseudoconvex domain in C" with boundary of class Cni+2, and let y E Z(,njA (D), m > 0, A > 0. Then there exists a form a satisfying as = y in D, with
0
aEC(p.Q±jjz(D), aE
C"'+l.a-i/z(D
12 < A' <
if 12 < A < 1.
(21.1)
PROOF. We proceed as in Theorem 4.4. We need only show that the form
)
'
satisfies (21.1). Lemmas 1.4 and 4.1 show that it is enough to verify that (21.1) is true for the function
H(z)=
+ /(fV\D)r
(
,z)) , I
t -z 12(n-r- 1)Add, z
(21.2)
and h = 0 on D. If m = 0, we may suppose that with h E I= 0(1) I p((') h-'; for this, we have to extend y to h E C(V\D) and i all of C" by means of Theorem 3.2 and Supplement 3.3 of Malgrange [1], Chapter 1. For arbitrary m, by working out the argument in Lemma 4.2 more carefully, we see that h(t) = 0(1) I p(.) I"'-'+X. We shall carry out the rest of the proof in the case A > i . The case A < L can be handled similarly.
Differentiating under the integral sign and using the argument in Lemma 4.2, we conclude that, for z E D,
D Hl/z )I = O( )j=12 1
(v\D); I lJ, Z) 13-j I
- z I2n-3+j
II1II+11'11=m+2. Choose r and 8 according to Lemma 21.1 and split the integrals over V \ D into integrals over (V\D)\B(z, 8) and (V\D) ( B(z, 8). The first integrals
128
VI. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
can be estimated above by a constant independent of z. Thus pk
ID'.'H(z)I = 0(1) 1 +
j=0 "[(V\D)nB(-.8)1; l b((y , Z)I3-l1J
-
ZI2n-3+j
2
=0(1)1+2 j= 0
pk-1dv t 1(V,D)nB(z,8)1f (
/
y
Z)1)3-jIJ
y
C1 (P(J) - P(Z)) + C2IJ - ZI2 +IImI (
,
- Z,2,,-3+j
Here, we have used inequality 4 from §4.2°. Now make the change of variables occurring in part II of Lemma 21.1. Using the definition of the the obvious inequalities
I71-I,I,zn-I(E) = O(1)IA-I,I=,2n-11E)
and Lemma 21.2, we getforlllll + IIJII =m+2 2
ID',JH(z)I = O(l) 1 + 2 Ik-1,3-j,2n-3+j(IP(z)I) j=0
= 0(l) 11 + =
I IX-1,3,2.-3(IP(Z)I) +
h-1,12,2.-1(Ip(z)I)l
O(1)Ip(z)h-12.
Hence we conclude by Lemma 21.3 that all derivatives of H of order m + 1 (D). Thus H satisfies (21.1). lie in We next give the sharpened version of Theorem 6.1.
THEOREM 21.5. Let D be a bounded strictly pseudoconvex domain in C",
n > 1, with aD E C-+2, m 3 0, and let a E 1)(aD), X > 0. Then a E if and only if there exists a form y satisfying (21.1) such that aylaD=a.
PROOF. The sufficiency was proved in §6. We prove necessity. By Theorem 2.12, a = a1 IaD - a2 IBDI Z(InPInI-1)(D)
(21.3)
where a1 E and a2 E Zpn"-1)(CD). By Theorem 21.4, a1 = ay1 and y1 satisfies (21.1). Extend a2 to a form a2 E C(P;n-1)(C"). By (1.1), for
§21. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
129
z E (V fl G )\ D we have
a2 = a2 = In,n-1(G, a2) - I2,"-1 (G, aa2) - aI ,n-2(G, a2) (21.4) Here G is any domain of holomorphy with boundary of class Cm+2 such that D C G. Using Lemma 1.3 and the fact that aa2 = aa2 = 0 on CD, we get for
z E G\D 1a,11-1(G, 5«2W) = I( "-1)(D,
-at f aa2 /\ µo - f 8a2 A WP,0(u, , z). (21.5) D;
Here µo is the form defined in Lemma 1.3, for the choice
w1 = t = (f - Z)/I
z 12
and w2 = u = P(J, z)/(D(J, z)
The coefficients of WP,o(u, , z) are holomorphic in z, and
a2 IaD = a2 laD = al IaD - a E AnP (D ), so that the last integral on the right side of (21.5) vanishes by Stokes' theorem. The integral
Y2(0 = f aa2 A µo Dz
and its derivatives can be estimated as in the proof of Theorem 21.4 to conclude that y2 satisfies (21.1) with G\D in the place of D. Extend y2 to all of G with the same smoothness, and denote the extension also by y2. Next, since
Ip.n_1(G, a2) E Z(P n_1)(G), we have ay3 = Ip,n_1(G, a2), where y3 E C(p n_2)(G). Then it follows from (21.3)-(21.5) that a = aY Iao, where y = YI + Y2 - 73 + IP n_2(G, a2). Corollary 2.4 now finishes the proof. If we use Theorem 21.5 in place of Theorem 6.1 in the proofs of Theorems 13.2 and 13.4, then we can weaken their hypotheses: instead of the condition µi E C2 1(8D), A > 0, we need only require µ', E C2 - 1(aD), A > Z. We state the next theorem without proof. THEOREM 21.6. Let D be a bounded strictly pseudoconvex domain in C", and
suppose a is a (p, n - 1) form on aD with coefficients in L'(aD). Then a E A 1(D) if and only if there exists /3 E Zp,n_ 1(D) such that
lim f
(p(z)=-e)
Q(z) A T(z)
f «(z) A T(z)
(21.6)
aD
for any q) E C(n-P,o)(D ). Also,
f
(p(z)=-e)
(l/3(z) A aP(z)I+IP(z)I''21$(z)I)d; =
0(1)IIaIIL,.
(21.7)
130
VI. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
Here, as before, p is a function defining D; by the absolute value of a form we mean the sum of the absolute values of its coefficients, and 11 11 L. denotes
the L' norm of the absolute value of the form. (21.6) says that, in some (generalized) sense, the restriction of /3 to aD coincides with a. (21.7) says that the tangential part of /3 has the same properties as a. For forms of type (p, n - 1), the a-problem is solvable not only for domains of holomorphy but also for domains D with aD E C2 for which the restriction of the Levi form (cf. §4.2°) to the complex tangent space to aD at a point has at least one positive eigenvalue for each E aD. In the C°° case, this was done
by Kohn and Rossi [1]. By combining the methods of proof of Fischer and Lieb [1] with those of Theorem 21.4, it is possible to show that Theorem 21.4 is
valid for q = n - 1 and for domains of the above kind. After making the corresponding changes in the proof of Theorem 6.2, we thus obtain the following improved version: THEOREM 21.7. Let D = Sl\ U Off, where 2 is a strictly pseudoconvex domain with a12 E Cin+2, m 1, and theiU. C Sl are domains with boundaries of class C'"+2 such that S2j fl SZ, =0 f o r j 1, j, l = 1, ... , k, and the restriction of the
Levi form to the complex tangent space to aSZj at has at least one positive eigenvalue at each E afl,, j = 1,...,k. Let a E C(P.fl_I)(aD). Then a E An-p(D) if and only if there exists a form a such that a IaD = a and
ae
j+112('U)'
aEZ(Pn'=1j 2(D),
0 < A' < A, if 0 < A < 1/2;
1/2
§22. Generalization of Theorem 8.1 THEOREM 22.1. Let D be a bounded domain in C' with boundary of class C2,
and suppose that for each a E A1(D) f1 Ct 1)(D) there is an a E Z,,,,,_ j)(D) such that aIaD = a. Then D = U\ U i SZj, where the SZj (E Sl are domains such that Q. fl fly = 0 for j * Q, I = 1,... , k, and the restriction of the Levi form of aSZ; to the complex tangent space to all; at has at least one nonnegative eigenvalue at each E asp;, j = 1, ... , k. Theorems 21.7 and 22.1 give similar necessary and sufficient conditions on the bounded components of the complement of D in order that all forms in A I (D) have a-closed extensions into D. However, the most interesting part of the boundary, namely that of the unbounded component of the complement, remains unexplored.
131
§22. GENERALIZATION OF THEOREM 8.1
PROOF. Suppose that, for some j, Sty does not satisfy the conclusion of the theorem, i.e. there exists to E aSZj at which all the eigenvalues of the Levi form
are negative. Then there exists a neighborhood U of to such that for any E U fl SEJ the function
F((z) _ !21
t!) + `
aZ
2
8 azm
t1)(zm - tm)
a
1
vanishes in U fl C 1J if and only if z = t E aS2j U U (see, for example, Gunning and Rossi [1], Chapter IX, §B). A similar assertion was used in the first part of the proof of Theorem 8.1. It follows that there exist t' E U fl Sty, z° E U fl CU, and r > 0 such that the functionf(z) = l/FF,(z) satisfies
If(z°)I>if(z)1,
z E CSZj fl aB(t°, r),
(22.1)
and CSZj fl B(t°, r) has piecewise smooth boundary. The form
J U°,0 ,z0),
EaSZj,
E aD\aSZj,
0,
lies in A1(D) fl
hence there exists a E Z(',,,ri_1)(D) such that a IaD = a. As in the proof of Theorem 8.1, we see that U00(t, z°) = aY in Sly, where
Y(J) = -f
)wU0,0(w, zo)U,,,n-2(w, J) + D f
(w) A Un,n-2(w, J )
By Corollary 2.4, y E C(, Je_2)(SZj). By the Martinelli-Bochner formula (i.e.
(1.1) for p = q = 0) and Stokes' theorem, for any g G A CQj fl B(t°, r)) we have
g(z°) = fa(cn r1B(j°, r)); = Here t = aStf fl
fr
z°)
AY(t) + f (cf;naB(g°, r));
r). It follows that, for some constant c,
Ig(z°)I < C
max
jg(z)j.
(22.2)
('ECS3j naB(j'0, r)
But it is clear from (22.1) that if g(z) = f m(z) with m sufficiently large, then (22.2) cannot hold. This contradiction proves the theorem.
132
VI. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
§23. Weighted formula for solving the a-problem in strictly convex domains and zeros of functions of the Nevanlinna-Dzhrbashyan class
To obtain certain precise results in strictly pseudoconvex domains, it became necessary to have the solutions of the a-equation with a right-hand side that grows near the boundary expressed by a formula. For simplicity, we present the formula here only for a strictly convex domain.
Let D be a strictly convex domain in C", i.e. a domain of the form D = (z E U, p(z) < 0), where U is a neighborhood of D, p E C2(U), and the second differential (i.e. Hessian) of p is positive definite at all E U. Hence grad pIao 0. For a strictly convex domain, the Khenkin barrier function (see §4.2°) may be introduced as follows:
PG) = (PI,...,P.) =
(, ,..., a
z) -
z) = We set of p, we have
z) = (P, - z).
_LP
By Taylor's theorem and the strict convexity z12
2Re z) + ci1 p(z) > for z, E D', D C D' C U, with a constant c, > 0 depending only on Y.
Therefore 2Re
z)
2Re '(', z) >
p(z) +c1I - z12,
(23.1)
p(z) + c11 - z12.
Hence (23.2)
I1W, z)I = 0(1)I.W, z)I,
{pq) - p(Z)I - pq) - p(z) +IImI(t, z)I+I 0(1) depend only on D' and not on z, For a (p, q)-form y we define (G;.gy)(z)
z12
=
E Y.
=-IP,q_I(D, ak(J, z)Y(0)(z)
fo;a; aM, Z) A
A Ny-, n p, (n l
p)I Dp,,,-P(az, (23.4)
§23. WEIGHTED FORMULA FOR SOLVING THE 2-PROBLEM
133
Here, µ4_I is given by Lemma 1.4, with w = I = ( - z)/l.' - z 12 and u = z), and ak is defined by ak (, yy
2n k
Z)
Z)
Z))
__
-
,
k 2n-I
Z)
j=0
(4(, Z)
Z))
)JJk
(23.5)
We have kak(J, Z)
_
Z)
z)
(I
)2n_I[ iD(t,
z) $2Q, Z)
4b(t Z)
(23.6)
From Lemma 4.1 and (23.6),.and the fact that µq-, is a linear combination of forms of the type (1.10), we see that the kernels of the second term on the hence the right in (23.4) have no singularitites. Also, ak contains a factor p operator GP q can be applied to forms y such that q > 0, and suppose y E Z(1p,q)(D) and 0 > 0 are such that
pey E L(lp q)(D) and pe-''28p A y E L(p.q+,)(D). Then for all k > 0 + 1 the forms Nk = GP qY satisfy 8(3k = y in D, and pe-iQk E Lp q_I(D).
PROOF. Assume first that y E Zp,q(D). Substitute akQ, z)yQ) in the Martinelli-Bochner-Koppelman formula (1.1). Since akG, z) = 0 for E 8D, ak is holomorphic in z on D and y8y = 0, we get
Y(z) = -Ip,9(D, arak(J, z) A Y) - elp9-i(D, ak(J,y z)-y). ((
(23.7)
We transform the first term on the right above by Lemma 1.4 to Ip,9(D' fDl
= 8 z Df ;
z) A Y)
/'D;ajak(J, Z) A Y(J) A Up.9(J, Z)
ak(J, z) A Y(J) A (azNq-I + akG, Z) A
+(-1)p+Q-1f,
A µq-I
a,ak(J, Z)
atµ9) A
P' (n1 p)!
Dp,n-p(az, aJ )
;7!
n µ9 n
P!
(nl P)!
Dp n-p(aZ, 8 )
As already observed, the integrals have no singularities, so it is legitimate to z) = 0 for E 8D and bring 8Z outside the integral sign. Now, since k > 1, we see by Stokes' theorem that the second integral above vanishes.
134
VI. FORMS ORTHOGONAL TO HOLOMORPHIC FORMS
Hence (23.7) yields y = aGp qy. In the general case, we apply this equality to the form Y,(z) = y(rz) E Zp q(D) and pass to the limit as r -- 1.
We proceed to estimate the solution thus obtained. From (23.6) and the definition of Gp q, we have 1pe-1(z)I
/DzIPe-I\z)IIIGP 9Y)(z)Idvz=O(1) fD;IY(')Idvtf
IUP,9-1(J, z)I Iak(J, z)I D(J, Z) 12n- I
+I(Vk-'(J, z)I 1
+0(1)f
IY(J)
D[
aP(f)I j IPe-I( dv
IP(J )I 15r4)(J, z)I lµq-1I duz
z)
I( P'Zz)21n+1 (' )1 IµqIdv. 2n
z )I I a k_ 1(J, z )I
1
z
Here, as before, dv denotes the volume element, and the absolute value of a form is the sum of the absolute values of its coefficients. Next, using (1.4), Lemma 4.1, (23.5), (23.2), (23.3), the relation
- z)I = O(1)I - Z1,
Ia3'(Dq, Z)I =I(aPq),
and the definition of µq-1 (see Lemma 1.4), we get fD.
I0e-1(z)I I (Gp,gY)(Z)I duz = O(1) fDI IY(J )I IPk(J )I dv
Xl
Z) Ik I
I
- z 12n-1 +
D(J, Z) Ik+21
IPB-1(Z)I
ID,
Z
12n-4IPA-1(Z)Iduz
1
I P-1(z) I dvz
+0(1)f Dz
DL
1 (J, Z)
Ik+1 I Y - Z 12n-3
Hence, by the same argument as in the proof of Theorem 21.4, fD
IPe-1(z)I I (G,,gy)(z)I dvz
= O(1)j 1 + I IY(J )I
)I) + Ie-1.k+2,2n-4(IP(J )I)J
A
An application of Lemma 21.2 finishes the proof of the theorem.
§23. WEIGHTED FORMULA FOR SOLVING THE a-PROBLEM
135
We state without proof an application of Theorem 23.1.
Let D = {z: p(z) < 0) be a bounded domain in C" with C2 boundary. We denote by NN(D) (a > 0) the class of holomorphic functions F in D such that jp(z)ja-i In+IF(z)Ida2n(z)
fD
< oo
(the Nevanlinna-Dzhrbashyan class). Here dak is the k-dimensional LebesgueHausdorff measure, and ln+ t = max(ln t, 0}, t > 0.
THEOREM 23.2. Let D be a strictly convex domain in C". Then an (n - 1)dimensional analytic subset M of D is the set of zeros of a function from Na(D), a > 0, if and only if
f
M
IP(z)la+ida2"-2(z)
_
27ifM
IP(z)Ia+ida2"-2(z)
< oo,
where the Mi are the irreducible components of M, and yj is the multiplicity of Mj in M.
CHAPTER VII
REPRESENTATION AND MULITPLICATION OF DISTRIBUTIONS IN HIGHER DIMENSIONS
§24. Harmonic representation of distributions
Recently, there has been increased interest in the problem of multiplication
of distributions, and there has been a lot of work on the problem. In the International Conference on Generalized Functions held in Moscow (Novem-
ber 1980), different approaches to solving this ,problem were discussed in several papers. Here we present a method based on representing the distributions by harmonic functions.
Let R"+' be the space of the variables (x1,...,x", y) = (x, y), and let W++' = ((x, y) E R"+', y > 0), then R" = ((x, y): y = 0). We shall say that a function f defined in W,+' has finite order of growth as y - + 0, if for every compact set Min R" there exist c > 0 and m > 0 such that I f(x, y) I <
for
x E M and 0 < y < 1. Let H denote the set of harmonic functions f in R+' having finite order of growth as y -+ + 0, and K the set of harmonic functions in W++' which can be extended continuously to R+ ' U R" and vanish on R. Then K C H. Let K(x, y) be the Poisson kernel for the half-space R+
+ 1)/2) K(x, y) = r((n ,,(n+ 1)/2
y (1 X 12
+y2)`"+1)/2
If T E E'(R"), then the harmonic function T*(x, y) = TK(x - t, y) is called the Poisson representation of T.
THEOREM 24.1. If T E E', then T* E H and represents T in the following sense:
lim f T*(x, y)4)(x)dx = (T,p), +U Y=E 137
9' E 00(R,1).
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
138
Here T* extends continuously to R+ ' U (R" \supp T), and vanishes on R"\supp T, and T*(x, e) = Oq x 1-"-') as I x I -> oo. The Poisson representation of a derivative of T is the corresponding derivative of the Poisson representation of T.
This theorem generalizes Theorem 14.1, since (see § 14) C" = R2"
aT(Z) ly=e - «T(Z)
Iy=-,
= T*(XI ,...,x2,,
1,
e).
The proof of Theorem 24.1 is the same as that of. Theorem 14.1. We need only show that T* E H. But since T E E', it is bounded with respect to some seminorm sup
119)11K =
IDrq)(x)I,
K5J supp T,
xEK,11111 m
and so
(T, K(x - t, y)) I s cjjK(x - t, Y)11K < c1y-',
x E K.
THEOREM 24.2. For each T E 6 '(R"), there exists an f E H which extends continuously by 0 to R+ ' U (R" \supp T) and which is a harmonic representation of T, i.e.
lim f -
f(x, y)cp(x)dx = (T,
E (24.1). Thus 6D' = H/K.
fE
q E 6 (R").
(24.1)
H represents some Tf E 6D' via
THEOREM 24.2 generalizes Theorem 14.2 to the case of an arbitrary R" and an arbitrary T E 6D'. Thus, for distributions defined on hyperplanes, it is more convenient to construct their representations by means of the Poisson kernel rather than the Martinelli-Bochner kernel. The latter can be used for distributions defined on smooth hypersurfaces in C". The first part of Theorem 24.2 has in fact been proved already (Theorem 14.2). We proceed to the proof of the second part. LEMMA 24.3. Let f E H. Then, for any bounded domain SZ C R", there exist an index p and an F E H which can be extended continuously to R+ ' U 9, such that
f = a2pF/ay2P. For n = 1, a similar lemma for holomorphic functions is due to Tillmann [1].
PROOF. Let 0 be the Laplacian in R"and 0 that in W. Given f harmonic in
consider
F1(x, y) = f fF(x, n)dn + qp(x).
§24. HARMONIC REPRESENTATION
139
Then OF,(x, y) = af(x, l)/ay + 10(x). Choose p E C°° such that Aq' = -af(x, l)/ay in R" (this is possible since the Poisson equation can be solved in
R). Then OF1 = 0, and aF,/ay = f. If f E H, then there exist c > 0 and m > 0 such that x E St, 0
If(x, Y)I < cy-'°,
Hence F, E H and F1(x, Y)I <
y-m+l,
xESZ,0
At some step we can find an F2 E H which is bounded for x E S2 and 0 < y < I and such that a5F2/ay3 = f. The next harmonic primitive F E H of f will then be continuously extendable to R+' U Sl, since
IF(x,Yi) -F(x,Y2)I
X ESt,0
Using this lemma, it is easy to show that the limit in (24.1) exists for any f E H and q) E G1. Indeed, a2p
f _ef(x, Y)9p(x)dx = f -! 8y2p F(x, Y)4)(x)'dx
&PF(x, y)q,(x) A _ (-1)p f y-F
_
Op f -t F(x, Y)1XP dx
o
f-1)p
dx.
Consequently, since 6 is weakly dense in 6D' (Shilov [1], Chapter II, §9.3), the above limit defines a certain distribution Tf E 6D.
Suppose now that, for some f E H, the limit in (24.1) is 0 for all p E G1. Then, by Lemma 24.3, for each bounded domain St C R" we have an F E H
and an index p such that a2PF/ay2p =f and F extends continuously to R"+ ' U Q. Since f represents 0, &PF(x, 0) = 0 in SZ in the sense of distributions. Hence it follows from the ellipticity of Op in S2 (see, for example, Schapira [1], Theorem A.22) that F(x, 0) is real analytic in R. Hence F extends in a C°° way to SZ (see Brelot [1], Chapter I, §8). Then
f(x, y) = (-l)PAPF(x, y)
(-l)pA"F(x,0) = 0,
as y - +0. LEMMA 24.4. Let f be a polyharmonic function of order m (i.e. Of = 0) defined in R+ ' . Then
f(x, y) =fo(x, y) +Yfi(x, y) + ... +ym-fm-i(x, y),
(24.2)
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
140
where the f j, j = 0, 1,... , m - 1, are harmonic functions in R+ ' ; if f has finite order of growth as y -> + 0, then the f lie in H, and if f E C°°(R"+' U Sl) for an open Sl C R", then the f are C° in R"+' U 9.
The proof is by induction on m. If m = 1, (24.2) obviously holds.
Let
ef = 0, m > 1. Then we must have f = Eo -' yjf , which would give us m-2
Of= 7. y' (j+2)(j + 1)f+2 +2(j+ j=0
1)_f+1
ay
By the induction hypothesis, since A f is polyharmonic of order m - 1, m-2
j=0 so that we obtain the equations
aj=(j+2)(j+1)fj+2+ 2(j+1)
of+
j = 0,...,m - 2,
ay
for the f ; in particular, afm-1
(Pm-2 = 2(m - 1) ay
As in Lemma 24.3, we seek the solution of this equation in the form of
the
integral
fm-1(x, Y) = 2(mI
1)
f
(pm-2(x, rl)dvl +> (x).
Since ofm-1 =
2(mI-
1) a
ay
2 (x, I) +
0'
we find ' from this equation. Next, from the equation
T.-3= (m - 1)(m-2)fm-1+2(m-2) afm-2 ay we determine fm-2 in a similar way, etc.
It is clear from the proof above that the f he in H if f is of finite order of growth, and that f E C°°(0 U IV ++) for all j if f E C°°(7 U R+'). COROLLARY 24.4. If Omf = 0 and f is of finite order of growth as y --> + 0, then the sense of (24.1)as y-> +0. f(x,
Thus, polyharmonic functions also represent distributions in 6D', and in this case Tf = Tfo
§25. MULTIPLICATION OF DISTRIBUTIONS
141
§25. Multiplication of distributions and its properties
Using the harmonic representations, we can define the product of two distributions by the method of Ivanov [1], [2]. Let H* be the algebra of functions generated by H, and K* C H* the subset consisting of functions representing 0 in the sense of (24.1). We shall call the quotient H*/K* = 6D* the space of hyperdistributions, and denote by f* the image off e H* in Gj *,
i.e.f*=f+K*. The map A: 6D' - + 6 * defined as Tf - f * (f a harmonic representation of T) is an imbedding, since K C K*. Hence we shall suppose 6D' C G *. Let T, S E
C ', and let f and g be their harmonic representations. Then by the product T o S of T and S we shall mean the hyperdistribution (fg)* = fg + K* E 6D*. In the sequel, we shall be interested in the question as to when T o S is a distribution, in the sense that the limit (24.1) exists for fg and an arbitrary q) E C1. Note that if this limit exists for fg, then it exists for any function in fg + K*, and coincides with the limit for fg. We give some properties of this product. 1. S o T is well-defined, i.e. the class fg + K* is independent of the choice of the representatives f and g.
PROOF. It suffices to show that, if f E K and g E H, then fg E K*. Let p E 6D. It is easy to show that f = yf, in a neighborhood of the support of p, where f, is a real-analytic function. Hence tp E Gi for fixed y, and fqp 0 in Gi +0, while g tends to Sin 6D' as y - +0; consequently, by a lemma in as y Shilov [1] (Chapter II, § 9.3),
lim f f(x, e)cp(x)g(x, e)dx = (S,0)= 0, e.+0 R" i.e. fg E K*.
2. supp(T o S) C supp T fl supp S. In particular, T o S = 0 if supp T n supp S = 0 . Here, the support of a hyperdistribution is defined as follows. We say that f * E 6D* vanishes in an open set 12 in R" if, for any q) E Gl with support in S2,
lim fy=E f(x, y)cp(x)dx = 0,
f E f*.
Now the support supp f * of f* is by definition the complement of the largest open set in which f* vanishes. It is easy to verify by means of a partition of unity that this notion is well-defined. PROOF OF PROPERTY 2. Let qp E 6 with supp qv C R"\(supp T fl supp S) _
0. We may suppose that, for example, supp qv fl supp T = 0 (by means of a
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
142
partition of unity it is always possible to pass to functions p with arbitrarily small support). Then f can be written in the form yf, in a neighborhood of supp p, and the rest of the proof proceeds as for Property 1.
3. If T E 6D' is defined by a function T(x) E LP(loc), and S by S(x) in Lq(loc), 1/p + l/q = 1, then To S = T(x)L(x) E L'(loc). PROOF. It can be shown that, corresponding to S(x), there exists u(x, y) E H such that u(x, y) -> S(x) almost everywhere as y - +0, and for each compact set M
I1u(x,y)-s(x)IIM as y - + 0, where
(ffdx)¼,
q
IifiiM=
1.
To do this, we must use:
(a) Theorem 1 of Stein [2], (Chapter III, §2.1), asserting that, for every f e LQ(R"), its Poisson integral defines a harmonic function f(x, y) such that f(x, y) converges to f(x) almost everywhere and II f(x, y) - f(x)ll - 0 as
y - +0; and (b) the procedure in Bremermann [1], §5.9, replacing power-series expansions by expansions in harmonic polynomials. Since S(x) E LP(loc) and T(x) E LP(loc), we will have S(x)T(x) E L'(loc). Let u(x, y), v(x, y) and w(x, y) be the respective harmonic extensions to R+ ' with the properties mentioned above. Then u, v and w are harmonic represen-
tations of T(x), S(x) and T(x)S(x). Furthermore uv - w E K*. Indeed, let q) E 6D with supp 9) C M. Then
14"
(uv-w)y,dx
but I1uv-STIIM-*0asy- +0, since Ilu- TIIMand Ilv-Sll
tendto0.
4. The derivatives of hyperdistributions are defined as follows: a
aXk
__
f*
axk
f
of
+ K*.
aXk
This definition makes sense, since a f/axk E K* for f E K*. For the derivatives of the product of two distributions, we have the Leibniz rule aXk
ToS=ax k
oS+To aXk
§25. MULTIPLICATION OF DISTRIBUTIONS
143
This follows immediately from the fact that if f represents T, then 8f/8xk represents 8T/8xk. 5. The product of distributions is compatible with multiplication by multipliers,
i.e.ifaEC°°(Rn)and TE6',then aoT=aT. PROOF. Since a E C°°, its harmonic representation g(x, y) E H extends continuously to R+ ' U Rn (solution of the Dirichlet problem for the half-space).
Hence g E C°°(Rn+' U Rn) (Brelot [I], Chapter 1, §8), so that p(x)g(x, y) -+ 4p(x)a(x) in 6D as y -> +0, for any 99 E 6D. On the other hand, f(x, y) Tin 6D' as y - + 0, so that, by the lemma already mentioned (Shilov [1], Chapter II, §9.3),
f f(x, y)4p(x)g(x, y)dxe--.-,+0(T, aip) = (aT, p). y=e
6. If f E H is a homogeneous function of degree p, then its boundary value as y - +0, i.e. Tf, is also homogeneous of degree p. Indeed,
(Tf, T(tx)) = lim f _ f(x, y)p(tx)dx = lim E+() y_E
=
1
e-'.+0 t
tp+n Elimo f y=E1
f
y=E1
f(xt , y)p(x)dx t
f(x, y)9p(x)dx = tp+n (Tf, y)) .
If T E 0',, is homogeneous of degree p, then it is not hard to see that T*(x, y) is a homogeneous function of degree p, since
T*(tx, ty) _ (T,,, K(u - tx, y))
(Tu, K(i - x, y)}
= tp (Tn, K(u - x, y))' = tPT*(x, y). By means of the general procedure for constructing a harmonic representation, it can be shown that, if T E 6D' is homogeneous of degree p, then T admits a harmonic representation which is homogeneous of degree p. It follows that, if T, S E 6D' are homogeneous of degrees p and q respectively, and T o S E 6D', then T o S is homogeneous of degree p +^q. 7.
Let x' = (Xj,...,Xm), x" _ (Xm+I,...,Xn), and T E '6 '(Rm), S E
'(R"-'n). Let f(x', y) and g(x", y) be harmonic representations of T and S in respectively. Then R++' and c
Ch+K* where h is a harmonic representation of the tensor product T X S. Thus multiplication of distributions is compatible with the formation of products.
144
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
PROOF. It suffices to show that the boundary value of f(x', y)g(x", y) in the sense of (24.1) is T X S. Consider the integral
f_ y_E
g(x",y)9p(x)dx"=¢(x',e),
99 E6 .
We have ¢(x', e) E 6 (Rm) and b(x', e) -> (S, (p(x))= J(x') in 6 (Rm) as e - +0. Since J(x', e) - ¢(x') as e - +0 (Lemma 24.3), and
D, (x', e) =
f y-g(x _E
y)DX ,V(x)dx",
we have, by the same lemma, Dx '(x, e) - DX (((x')) as a --> +0. Now we apply the lemma from Shilov [1] (Chapter II, §9.3) once again and conclude
fyEf(x',
y)g(x", y)9p(x)dx
=
f
_ f(x', y -E
y)b(x', y)dx' --p (TX-, (x')) E-+0
= (Tx,, ( Sx , , qg(x))) = (T X S, (p(x))
Thus the definition of product which we have given has several natural properties; in particular, we will not have counterexamples of the Itano kind with our definition. §26. Examples of products of distributions Consider the vector space
2 of functions, generated by functions of the
form
f(x,y)=
(IxI2
P yk(x,y)+(n+l)/z
k,0,m>1,
(26.1)
where Pk(x, y) is a homogeneous polynomial of degree k and 0Pk = 0. These functions are polyharmonic of order m; actually they are the Kelvin transforms of the Pk(x, y). We first elucidate what kind of distributions they represent.
LEMMA 26.1. Suppose f is a function of the form (26.1) lying in WI, and
k + 1 -2m>0.Then
f
Pk(x, y)x°dx
R (I X 12 + y2)k-m+(n+1)/2
-0
for all monomials xa with Ilall < k + 1 - 2m. PROOF. Let
Pk(x, y)xadx AY)=fxjal (IxI2+y2)k-m+(n+1)/2
Ilall
145
§26. EXAMPLES OF PRODUCTS
We shall show that the limit
lim I(y)=c
y-. +0
exists and is finite. Indeed, by Lemma 24.4, we can write f in the form
.f =.fo +Yfl + ...
+y,n-Ifm-I;
then
M-1
ylimoI(Y) =j=0 F y limoyf
ffi(x,
y)xadx.
According to Lemma 24.4, the f are C°° on R+ ' U (R" \(0)). We now show that the limit
lim f
f (x, y)xadx
IXIG1
exists.
By Lemma 24.3, there exists F(x, y) E H extending continuously to R+' U (x E R": I x< 2) and such that 2 / ay
p
F(x, Y) = f (x, y)
Then f (x, y) _ (-1)P PF(x, y), and IXI
a F(x, aXk
y)ip(x)dx
a-
_-f !XI
I
F(x, y) axk dx + (-1)*-' f!X! = I Fpdx
'
[1].
Thus
f 1Xi
f (x, y)xadx = (-1)p f
F(x, y)OPxadx + f
!X!<1
I
DF x'do ,
!X1=1
where D is a differential operator, and do the surface element of the sphere.
We need only observe now that the F, like the f j, are C°° on R+' U ((I x I< 2)\(O)) (see the proof of Lemma 24.3). Thus the limit
lim I(y)=c
y-. +o
exists. We now transform the variables in 1(y) by setting x = rE, rllall+n-l+k-s s s(t a Y 1(Y) I=IPk s) do +y2)k-m+(n+I)/2dr o (r2 s
-
fI
I
1. Then
146
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
where Pk(x, y) = s=o y5Pk(x). If we further set r = yt, then lllall+n-I+k-sdt I/y
I(y) =yllall-k-I+2m I
(I+
s>0
0
Since I(y) has a limit as y
t2)k-m+(n+l)/2
f
(26.2)
161=1
+0, we necessarily obtain from (26.2) that /' lllall+n-I+k-s
(
J = Ylimol(y) = I
t2)k-m+(n+l)/zd1JIF1=1Pk(S)sada = 0.
S>0 (1 + The improper integral above is convergent, since II a II < k + 1 - 2m. It is easy to see that Pk(X, y)Xadx /Rn (I X 12 + y2)k-m+(n+1)/2
= y hall-k-l-2mJ=
0.
LEMMA 26.2. Suppose an f of the form (26.1) lies in W and k + 1 - 2m % 0. Then
Pk(x,
P.V. fRn
k)
Hall = k + I - 2m,
m++1)/2 = Ca,
(I X 12 +.Y 2 )
where
7rn/2r(S - l/2)Qk+I-s-m(pks-I(x)xa)
C= a
sat
- s - m)!I'(k - m + (n + 1)/2)
22(k+I-s-m)(k + 1
(26.3)
PROOF. Consider the integral
I(Y)
_
Pk(x, y)xadX
- fIxI.I (Ix12+y2) k -m+n+I )/2 (
Transforming to polar coordinates as in Lemma 26.1, we get
I(y) - f
l2k+n-s-2mdt
/Y
0
(1 + 12)
k-m+ (n+1 n+1 )122 J 161=I
The polynmials Pk have degrees k - s; hence the Pk6a have degrees 2k + 1 - 2m - s. Thus, if s is even, the degree of the homogeneous polynomial Pk6a is odd, and so
I
P:Eado = 0.
161= I
Then,
I(Y) = sit
fI/Y 0
t2k+n+1-2s-2mdt
(1 +
t2)k-m+(n+I)/2
f
2s-1 161=1Pk
(C)
a
do.
§26. EXAMPLES OF PRODUCTS
147
Clearly 1(y) -> I as y -> +0, so that t2k+n+1-2s-2mdt
op
J
=
o
(1 +
t2)k-m+(n+1)/2
2s-I(t)ta SS dQ.
f,i,=,Pk
l5S
But t 2k+n+l-2s-2mdt fo
(I + t 2)k
(
n m+(n+l)/2-2B k-s-m+12,s-2) 1
1
and
Pks-ItadQ S - 22(k+1-s-m)(k
2TIn/2Qk+l-s
m(P`2s-1(00
+ 1 - s - m)!F(k +1551 SS- s - m + n/2)
(This follows from the Gauss representation of homogeneous polynomials; see Sobolev [I], Chapter 11). Hence I = ca. Observe that if Pk expands only in even powers of y, then ca = 0 for all a such that II a II < k + I - 2m. Let
h(x) =
Qk(x) x Ilk-2m+n+1 '
where Qk is an arbitrary homogeneous polynomial of degree k in W. We denote by h the distribution obtained by regularizing h(x), i.e. f.p.fR"
p.v.f
Qk(x)(q)(x) - Fllallck+l-2m(1/a!)(aatp/axa)(0)xa)dx x I2k-2m+n+1
R"
i
if k + 1 - 2m ' 0, where the abbreviations "f.p." and "p.v." signify "(Hadamard) finite part" and "(Cauchy) principal value".
Ifk+1-2m<0,then (h, rP) = fh(x)T(x)dx,
p E' .
In other words, h is the Hadamard finite part of the integral
f h(x)p(x)dx.
148
VII. REPRESENTATION AND MULTIPLICATION OF DISTRIBUTIONS
LEMMA 26.3. Let Q be any polynomial in Rn. Then (_ 1)1y21O1Q(x) 1>o
is a harmonic extension of Q to all of Rn+ 1
This follows from direct computation. If
h(X) =
Qk(x, Y) y2)k-m+(n+1)/2 (I X 12 +
then we denote by h(x, y) the function of the form Qk(X, Y)
h(x, Y) =
+y2)k-m+(n+l)/2
(I X 12
in U.
THEOREM 26.4. The distributions h and Tti are identical.
This follows immediately from Lemmas 26.1-26.3 and the definitions of h and Th. Let 2121 (respectively X122) denote the subset of Wi consisting of functions of the form (26.1) with Pk expanding only in even (odd) powers of y. It is easy to verify that span(W 1 + W2 2) _ X12.
Theorem 26.4 implies that the boundary values of functions f in WI in the sense of (24.1) are the distributions (f(x, 0))'. 1
THEOREM 26.5. If f E X12 21 then ca
(-1)k+1+2m
ifk+1-m;? 0,
s(a)
Tf =
Hall=k+1-2m 0,
a!
'
if k+ I -m<0,
where the ca are defined by (26.3) and 8(a) = aa6/axa, 6 being the Dirac 8 -function.
This is immediate from Lemmas 26.1 and 26.2. We have thus a complete description of the class of distributions representable by the functions in W2. EXAMPLE 1. Letf(x, y) =Y/(I x 12 +y2)(n+1)/2. Then
Tj = ir(n+1)/28/r((n + l)/2).
§26. EXAMPLES OF PRODUCTS
149
We now consider the question: when is the product of two functions in 1111 again im TO Let
f
_
Pk(x, y) y2)k-m+(n+1)/2'
(I X I2 +
g
=
Ql(x, y) (I
X I2 +y
2)1-s+(m+1)/2'
then
fg
PkQF
__
(I
X I2 +y
2)I+k+(n+1)/2-(m+s-(n+1)/2)
Hence we see that fg E TZ if and only if n is odd (n + 2p - 1), m + s > p, and
An'+s-P(PkQ/) = 0. Thus the class of such functions f and g is fairly large.
THEOREM 26.6. Suppose f, g andfg lie in M. Then: a) if f, g E T"t , then fg E ll and (f(x, 0))^ o (g(x, 0))^ = (fg(x, 0))^; 1
b)iff,g ET12'then TfoTg= 0; c)iffED'21andgE02,then fgED22and (-1)k-2p-2m £(X,0) o
Ci b(a)
IIf9II=1+k+2p-2m-2s-1 18'
IIaII=/+1-2s
J
c 8(p).
For n = 1, identities of this type have been obtained by Ivanov [3]. EXAMPLE 2. Let f = (p - 1)!y/irp(I x 12 + y2)P. Then Tf = S. Let h(x) _
Qk(x)/I
X
12k with AQk
= 0; then
(P - 1)!, (Qkxa) S(a>.
h o S = (-1)k IlaII
k
k!(k+p- 1)!
Ifh=xi/Ixl2,then
(Ix12)^oS=-2p aX S. For p = n = 1, this identity reduces to the familiar identity P(1/x) 8 = 42.
SUPPLEMENT TO THE BRIEF HISTORICAL SURVEY
§16 contains the solution to Problem 3 (see "Brief historical survey and open problems" which follows Chapter IV). All the results of this section are due to Kytmanov [2]-[4]. A special case of Problem 3 (viz., Problem 2) is discussed in §17. Theorem 17.1 was proved for the ball in C" by Aronov [2], and in the general case by Aronov and Kytmanov [1]. Theorem 17.2 and Corollaries 17.3 and 17.4 were obtained by Aizenberg and Kytmanov [1], [2]. Corollary 17.7 was obtained for the ball by Agranovskii and Val'skii [1], and for arbitrary domains by Stout [ 11. Theorems 17.9 and 17.10 were proved by Kytmanov [ 11. The results of § 18 are due to Aizenberg and Dautov [ 1 ]. All the assertions of §19 except for Theorems 19.4 and 19.5 are due to Kytmanov [8]. Theorems 19.4 and 19.5 were proved by Romanov [1]; Theorem 19.5 is in fact true for arbitrary domains with smooth boundary in C" (Romanov [2]). In §20 differential criteria for the existence of holomorphic extensions of functions are presented, which are criteria different from the tangential Cauchy-Riemann equations. Theorem 20.1 was proved for domains D in Kahler manifolds with 8D E C°° and f E C°°(D) by Folland and Kohn [1]; in the form stated here, it is contained in Aronov and Kytmanov [1]. Theorem 20.4 is a reformulation of a result of Forelli [1]. The remaining results of §20 are due to Kytmanov [7]. The strengthening of Theorem 6.1 (Theorem 21.5) was obtained by Dautov (see Aizenberg and Yuzhakov [1], Theorem 26.1). The results of §22 are due to Dautov. Weighted formulas for solving the a-problem and weighted estimates (Theorem 23.1) were proved by Dautov and Khenkin [1], [2]. Theorem 21.6 is due to Khenkin [5]. For a = 0, Theorem 23.2 (for functions of the Nevanlinna class) was proved by Khenkin 141-[61 and Skoda [ 1], [2]; for a > 0 (functions of the Nevanlinna-Dzhrbashy class) it was proved by Dautov and Khenkin [1], [2].
In these papers, there is also a generalization of Theorem 23.2 to strictly pseudoconvex domains in manifolds. 151
152
HISTORICAL SURVEY
Chapter VII is devoted to the solution of Problem 13. The results of this chapter are due to Kytmanov [5], [6]. We note that, for n = 1, Theorem 24.1 is proved in Bremermann [11.(p. 49). That part of Theorem 24.2 which proves that the boundary value of a harmonic function of class H is a distribution can be deduced from a theorem of Roitberg [1] which states that, if L is an elliptic operator of order 2m in a bounded domain 52 C R" with sufficiently smooth
boundary, then every solution f of the equation Lf = 0 with finite-order singularity on as2 defines a certain distribution on aS2. Theorem 24.2 gives a method for constructing this distribution. A different method of multiplying higher-dimensional distribution by means of analytic representations has been proposed by Ivanov and Verzhbalovich [1]. We note that, with this method as with ours, it would not be possible to have counterexamples like those of Itano [1]
BIBLIOGRAPHY (Items marked * were added for the English translation.) M. L. Agranovskii and R. E. Val'skii I. Maximality of invariant algebras of functions, Sibirsk. Mat. Zh. 12 (1971), 3-12; English trans]. in Siberian Math. J. 12 (1971). L. A. Aizenberg 1. Linear convexity in C" and the separation of the singularities of holomorphic functions, Bull. Acad. Polon. Sci. Ser. Sci. Math. Astronom. Phys. 15 (1967), 487-495. (Russian) 2. The expansion of holomorphic functions of several complex variables in partial fractions, Sibirsk. Mat. Zh. 8 (1967), 1124-1142; English transl. in Siberian Math. J. 8 (1967). 3. On Temlyakov integrals and the boundary properties of analytic functions of two complex variables, Dokl. Akad. Nauk SSSR 120 (1958), 935-938. (Russian) 4. On Temlyakov integrals and the boundary properties of analytic functions of several complex
variables, Moskov. Oblast. Ped. Inst. Uchen. Zap. '77 (Trudy Kafedr Mat. 5) (1959), 5.
13-35. (Russian) Integral representations of holomorphic functions of several complex variables, Trudy Moskov. Mat. Obshch. 21 (1970), 3-26; English transl. in Trans. Moscow Math. Soc. 21 (1970).
6.
On differential forms that are orthogonal to holomorphic functions under integration over the
boundary of a domain D C C", All-Union Conf. Theory of Functions of a Complex Variable (Entire and Meromorphic Functions and Functions of Several Variables), Abstracts of Reports, Fiz.-Tekhn. Inst. Nizkikh Temperatur Akad. Nauk Ukrain. SSR, Kharkov, 1971, pp. 10-12. (Russian) 7.
On a-closed exterior differential forms of type (p, n - 1) in the space C", Dokl. Akad. Nauk SSSR 209 (1973), 1009-1012; English transl. in Soviet Math. Dokl. 14 (1973).
8.
9.
Polynomials orthogonal to holomorphic functions of sever/ complex variables and an analogue of the Riesz theorem, Dokl. Akad. Nauk SSSR 199 (1971), 255-257; English transl. in Soviet Math. Dokl. 12 (1971). Polynomials that are orthogonal to holomorphic functions of several complex variables, and a generalization of the theorems of Hartogs and F. and M. Riesz, Holomorphic Functions of
Several Complex Variables (Proc. Sem., 1972), Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1972, pp. 5-19. (Russian) Integral representations of holomorphic functions of several complex variables, Dokl. Akad. Nauk SSSR 155 (1964), 9-12; English transi. in Soviet Math. Dokl. 5 (1964). L. A. Aizenberg and Sh. A. Dautov 1 *. Holomorphic functions of several complex variables with nonnegative real part. Traces of holomorphic and pluriharmonic functions on the Shilov boundary, Mat. Sb. 99(141) (1976), 342-355; English transl. in Math. USSR Sb. 28 (1976). 10.
153
154
BIBLIOGRAPHY
L. A. Aizenberg and A. P. Yuzhakov
I'. Integral representations and residues in multidimensional complex analysis, "Nauka", Novosibirsk, 1979. (Russian)
L. A. Aizenberg and A. M. Kytmanov 1.
Representation and multiplication of distributions from R"' with the aid of 2-closed exterior differential forms of type (n, n - 1), Manuscript No. 2690-74, deposited at
VINITI, 1974; abstract published in Sibirsk. Mat. Zh. 16 (1975), 409; English transi. in Siberian Math. J. 16 (1975). A. Andreotti 1. E. E. Levi convexity and H. Lewy problem, Proc. Internat. Congr. Math. (Nice, 1970), Vol. 2, Gauthier-Villars, Paris, 1971, pp. 607-611. Aldo Andreotti and C. Denson Hill 1. E. E. Levi convexity and the Hans Lewy problem. I, Ann. Scuola Norm. Sup. Pisa (3) 26 (1972), 325-363. 2. E. E. Levi convexity and the Hans Lewy problem. II, Ann. Scuola Norm. Sup. Pisa (3) 26 (1972), 747-806. V. I. Arnol'd Ordinary differential equations, "Nauka", Moscow, 1971; English transl., M.I.T. Press, 1. Cambridge, Mass., 1973. A. M. Aronov I. The Martinelli-Bochner integral formula for exterior differential forms, Holomorphic Functions of Several Complex Variables (Proc. Sem., 1972), Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1972, pp. 167-170. (Russian) 2. Functions that can be represented by a Martinelli-Bochner integral, Some Properties of Holomorphic Functions of Several Complex Variables, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1973, pp. 35-39. (Russian) 3'. On the boundary values of the derivatives of an integral of Bochner-Martinelli type, On Holomorphic Functions of Several Complex Variables, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1976, pp. 20-28. (Russian)
A. M. Aronov and Sh. A. Dautov I. Integral representations for complex differential forms, Sibirsk. Mat. Zh. 13 (1972), 739-747; English trans]. in Siberian Math. J. 13 (1972). 2.
On a result of B. Weinstock, Some Properties of Holomorphic Functions of Several Complex Variables, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1973,
pp. 193-195. (Russian) A. M. Aronov and A. M. Kytmanov 1. The holomorphy of functions that are representable by the Martinelli-Bochner integral, Funktsional. Anal. i Prilozhen. 9 (1975), no. 3, 83-84; English trans]. in Functional Anal. Appl. 9 (1975). Stefan Banach 1. 774orie des operations lineaires, PWN, Warsaw, 1932; reprint, Chelsea, New York, 1955. M. Brelot 1*. Elements de la theorie classique du potential, 2nd ed., Centre de Documentation, Paris, 1961.
Hans Bremermann 1. Distributions, complex variables and Fourier transforms, Addison-Wesley, 1965. E. M. Chirka 1. Approximations by holomorphic functions on smooth manifolds in C", Mat. Sb. 78(120) (1969), 101-123; English transl. in Math. USSR Sb. 7 (1969).
BIBLIOGRAPHY
155
Sh. A. Dautov I. On forms orthogonal to holomorphic functions with respect to integration over the boundary of strictly pseudoconvex domains, Dokl. Akad. Nauk SSSR 203 (1972), 16-18; English trans]. in Soviet Math. Dokl. 13 (1972). 2. On forms that are orthogonal to holomorphic functions and related problems, Funktsional.
Anal. i Prilozhen. 6 (1972), no. 4, 81-82; English transl. in Functional Anal. Appl. 6 (1972). 3.
4.
8-closed forms of type (p, n - 1) as an analogue of holomorphic functions of a single complex variable, Holomorphic Functions of Several Complex Variables (Proc. Sem., 1972), Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1972, pp. 21-36. (Russian) The general form of the integral representation of holomorphic functions in C", Holomorphic
Functions of Several Complex Variables (Proc. Sem., 1972), Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1972, pp. 37-45. (Russian) Sh. A. Dautov and G. M. Kenkin 1*. Zeros of holomorphic functions of finite order and weighted estimates for solutions of the 5-equation, Preprint No. 8M, Inst. Phys. Siberian Division Acad. Sci. USSR, Krasnoyarsk, 1977. (Russian) R Zh Mat. 1978 #7B257. 2*. Zeros of holomorphic functions of finite order and weighted estimates for solutions of the a-equation, Mat. Sb. 107(149) (1978), 163-174; English transl. in Math. USSR Sb. 35 (1979), no. 4.
Sh. A. Dautov and A. M. Kytmanov 1.
The boundary values of an integral of Bochner-Martinelli type, Some Properties of Holomorphic Functions of Several Complex Variables, Inst. Fiz. Sibirsk. Otdel. Akad.
Nauk SSSR, Krasnoyarsk, 1973, pp. 49-54. (Russian) Nelson Dunford and Jacob T. Schwartz 1. Linear operators. Vol. I, Interscience, 1958. Gaetano Fichera 1. Caratterizzazione della traccia, sulla frontiera di un campo, di una funzione analitica di piit variabili complesse, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Nat. (8) 22 (1957), 706-715. G. M. Fikhtengol'ts 1. A course in differential and integral calculus. Vol. I, 4th ed., Fizmatgiz, Moscow, 1958; German transl. of 5th ed., VEB Deutscher Verlag Wiss., Berlin, 1968. Wolfgang Fischer and Ingo Lieb 1 *. Lokale Kerne and beschrankte Losungen ftir den a-operator auf q-convexen Gebieten, Math. Ann. 208 (1974), 249-265. G. B. Folland and J. J. Kohn I*. The Neumann problem for the Cauchy-Riemann complex, Ann of Math. Studies, No. 75, Princeton Univ. Press, Princeton, N.J., and Univ. of Tokyo Press, Tokyo, 1972. Frank Forelli 1*. Pluriharmonicity in terms of harmonic slices, Math. Scand. 41(1977), 358-364. B. A. Fuks 1. Introduction to the theory of analytic functions of several complex variables, Fizmatgiz, Moscow, 1962; English transl., Transl. Math. Mono., vol. 8, Amer. Math. Soc., Providence, R.I., 1963. 2. Special chapters in the theory of analytic functions of several complex variables, Fizmatgiz, Moscow, 1963; English transl., Transl Math. Mono., vol. 14, Amer. Math. Soc., Providence, R.I., 1965. Theodore W. Gamelin 1. Uniform algebras, Prentice-Hall, Englewood Cliffs, N.J., 1969.
156
BIBLIOGRAPHY
I. M. Gel'fand and G. E. Shilov 1. Generalized functions. Vol. II: Spaces of test functions, Fizmatgiz, Moscow, 1958; English trans]., Academic Press, 1968, and Gordon and Breach, 1968. N. M. Gunther [Gyunter] 1. La theorie du potential et ses applications aux problemes fondamentaux de la physique mathematique, Gauthier-Villars, Paris, 1934; Russian transl., GITTL, Moscow, 1953; English trans]. (of the Russian), Ungar, New York, 1967. I. Glicksberg 1. The abstract F. and M. Riesz theorem, J. Functional Anal. 1 (1967), 109-122. G. M. Goluzin I*. Geometric theory of functions of a complex variable, GITTL, Moscow-Leningrad, 1952;
English transl. of the 2nd ed., Transl. Math. Mono., vol. 26, Amer. Math. Soc., Providence, R.I., 1969. Hans Grauert and Ingo Lieb 1.
Das Ramirezsche Integral and die Losung der Gleichung a f = a im Bereich der beschrdnkten
Formen, Complex Analysis (Proc. Conf., Houston, Texas, 1969), Rice Univ. Studies 56 (1970), no. 2, 29-50. Robert C. Gunning and Hugo Rossi 1. Analytic functions of several complex variables, Prentice-Hall, Englewood Cliffs, N.J., 1965.
F. Reese Harvey and R. O. Wells, Jr. 1. Holomorphic approximation and hyperfunction theory on a C' totally real submanifold of a complex manifold, Math. Ann. 197 (1972), 287-318. Maurice Heins 1. On the theorem of Szego-Solomentsev, Math. Scand. 20 (1967), 281-289.
Kenneth Hoffman 1.
Banach spaces of analytic functions, Prentice-Hall, Englewood Cliffs, N.J., 1962.
Lars Hormander 1. An introduction to complex analysis in several variables, Van Nostrand, Princeton, N.J., 1966.
L. Hbrmander and J. Wermer 1. Uniform approximation on compact sets in C", Math. Scand. 23 (1968), 5-21. Mitsuyuki Itano 1. On the multiplicative products of distributions, J. Sci. Hiroshima Univ. Ser. A-I Math. 29 (1965), 51-74. V. K. Ivanov 1. The operation of multiplication for distributions of slow growth, Izv. Vyssh. Uchebn. Zaved. Matematika 1972, no. 3(118), 10-19. (Russian) 2. Hyperdistributions and the multiplication of Schwartz distributions, Dokl. Akad. Nauk SSSR 204 (1972), 1045-1048; English transl. in Soviet Math. Dokl. 13 (1972). 3'. An associative algebra of elementary generalized functions, Sibirsk. Mat. Zh. 20 (1979), 731-740; English transi. in Siberian Math. J. 20 (1979). V. K. Ivanov and T. A. Verzhbalovich 1*. On the multiplication of Schwartz distributions, Trudy Inst. Mat. i Mekh. Ural. Nauchn. Tsentr Akad. Nauk SSSR Vyp. 13 (1974), 3-10. (Russian) M. V. Keldysh 1. Sur la representation par des series de polynomes des fonctions d'une variable complexe dans de domaines fermes, Mat. Sb. 16(58) (1945), 249-258. (Russian; French summary)
BIBLIOGRAPHY
157
Norberto Kerzman 1. Holder and L P estimates for solutions of a u = f in strongly pseudoconvex domains, Comm. Pure Appl. Math. 24 (1971), 301-379. G. M. Khenkin 1. Integral representations of functions holomorphic in strictly pseudoconvex domains and some
applications, Mat. Sb. 78(120) (1969), 611-632; English transi. in Math. USSR Sb. 7 (1969). 2.
Integral representation of functions in strictly pseudoconvex domains and applications to the
5 -problem, Mat. Sb. 82(124) (1970), 300-308; English transl. in Math. USSR Sb. 11 (1970).
3. A uniform estimate for the solution of the 8 problem in a Well region, Uspekhi Mat. Nauk 26(1971), no. 3(159), 211-212. (Russian) 4`. Solutions with estimates of the H. Lewy and Poincare-Lelong equations, Dokl. Akad. Nauk SSSR 224 (1975), 771-774; English trans]. in Soviet Math. Dokl. 16 (1975). 5'. The Lewy equation and analysis on pseudoconvex manifolds. I, Uspekhi Mat. Nauk 32 (1977), no. 3(195), 57-118; English transl. in Russian Math. Surveys 32 (1977). 6*. The Lewy equation and analysis on pseudoconvex manifolds. II, Mat. Sb. 102(144) (1977), 71-108; English transl. in Math. USSR Sb. 31(1977).
J. J. Kohn 1. Regularity at the boundary of the a-Neumann problem, Proc. Nat. Acad. Sci. U.S.A. 49 (1963), 206-213. J. J. Kohn and Hugo Rossi 1.
On the extension of holomorphic functions from the boundary of a complex manifold, Ann. of Math. (2) 81 (1965), 451-472.
Heinz Kong and G. L. Seever 1. The abstract F and M. Riesz theorem, Duke Math. J. 36 (1969), 791-797. Walter Koppelman 1. The Cauchy integral for differential forms, Bull. Amer. Math. Soc. 73 (1967), 554-556. 2. The Cauchy integral for functions of several complex variables, Bull. Amer. Math. Soc. 73 (1967), 373-377. Adam Koranyi 1*. The Poisson integral for generalized half-planes and bounded symmetric domains, Ann. of Math. (2) 82 (1965), 332-350. A. M. Kytmanov 1*. A criterion for the holomorphy of an integral of Martinelli-Bochner type, Combinatorial and Asymptotic Analysis, Krasnoyarsk. Gos. Univ., Krasnoyarsk, 1975, pp. 169-177. (Russian)
2'. On certain properties of exterior differential forms of class B, On Holomorphic Functions of Several Complex Variables, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1976, pp. 184-189. (Russian) 3. A certain characteristic property of a-closed exterior differential forms, Uspekhi Mat. Nauk 31 (1976), no. 2(188), 217-218. (Russian) On an integral characteristic property of a-dosed complex differential forms, Sibirsk. Mat. Zh. 19 (1978), 788-792; English transl. in Siberian Math. J. 19 (1978). 5. The representation and product of distributions of several variables by means of harmonic
functions, Izv. Vyssh. Uchebn. Zaved. Matematika 1978, no. 1(188), 36-42; English trans]. in Soviet Math. (Iz. VUZ) n(1978). 6*. A class of multidimensional distributions, Izv. Vyssh. Uchebn. Zaved. Matematika 1980, no. 10(221), 23-28; English trans]. in Soviet Math. (Iz. VUZ) 24 (1980).
BIBLIOGRAPHY
158
7`. Some differential criteria for holomorphy of functions in C", Some Questions of Multidimensional Complex Analysis, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1980, pp. 51-64. (Russian) 8*. On the exact calculation of an integral of Martinelli-Bochner type in a ball in C", Uspekhi Mat. Nauk 36(1981), no. 3(219), 217-218; English transl. in Russian Math. Surveys 36 (1981).
A. M. Kytmanov and L. A. Alzenberg 1*. The holomorphy of continuous functions that are representable by the Martinelli-Bochner
integral, Preprint No. 5M, Inst. Fiz. Sibirsk. Otdel. Akad. Nauk SSSR, Krasnoyarsk, 1975. (Russian) R Zh Mat. 1975 # 8E 166.
2'. The holomorphy of continuous functions that are representable by the Martinelli-Bochner integral, Izv. Akad. Nauk Armyan. SSR Ser. Mat. 13 (1978), 158-169. (Russian)
A. M. Kytmanov and A. A. Preobrazhenskii 1.
On a problem about the density of the polynomials of a definite form in the space of continuous functions on the boundary of the domain in C", Izv. Vyssh. Uchebn. Zaved. Matematika 1976, no. 7(170), 50-51; English transl. in Soviet Math. (Iz. VUZ) 20 (1976).
Jean Leray 1. Le calcul differential et integral sur une variete analytique complexe (Probleme de Cauchy, III), Bull. Soc. Math. France 87 (1959), 81-180. Ingo Lieb 1. Die Cauchy-Riemannschen Differentialgleichungen auf streng pseudokonvexen Begieten. Beschrankte Losungen, Math. Ann. 190 (1970/71), 6-44. Lu Qi-keng and Zhong Tong-de 1.
An extension of Privalof's theorem, Acta Math. Sinica 7 (1957), 144-165. (Chinese;
English summary) L. Ja. Makarova, G. Kudalberganov and V. P. Cherkashin 1. A remark on polynomial and linear convexity, Izv. Akad. Nauk UzSSR Ser. Fiz.-Mat. Nauk 17 (1973), no. 1, 36-41. (Russian) Bernard Malgrange 1. Ideals of differentiable functions, Tata Inst. Fund. Res., Bombay, and Oxford Univ. Press, London, 1967. 2. Existence et approximation des solutions des equations aux derivees partieiles et des equations de convolution, Ann. Inst. Fourier (Grenoble) 6 (1955/56),271-355. A. Martineau 1. Sw la topologie des espaces defonctions holomorphes, Math. Ann. 163 (1966), 62-88. 2. Distributions et valeurs. au bord des fonctions holomorphes, Theory of Distributions (Proc. Internat. Summer Inst., Lisbon, 1964), Inst. Gulbenkian Ci., Lisbon, 1964, pp. 193-326. E Martinelli 1. Sopra un teorema di F. Seven nella teoria delle funzioni analitiche di piu variabil complesse, Rend. Mat. a Appl. (5)20(1961), 81-96. S. N. Mergelyan 1. Uniform approximations to functions of a complex variable, Uspekhi Mat. Nauk 7 (1952), no. 2(48), 31-122; English transl. in Amer. Math. Soc. Transl. (1) 3 (1962). N. I. Muskhelishvili 1. Singular integral equations, 2nd ed., Fizmatgiz, Moscow, 1962; German transl., Akademie-Verlag, Berlin, 1965. Nils Ovrelid 1. Integral representation formulas and LP-estimates for the 8-equation, Math. Scand. 29 (1971), 137-160.
BIBLIOGRAPHY
159
V. P. Palamodov 1. Linear differential operators with constant coefficients, "Nauka", Moscow, 1967; English transl., Springer-Verlag, 1970. P. L. Polyakov 1. The Cauchy- Weilformula for differential forms, Mat. Sb. 85(127) (1971), 388-402; English transl. in Math. USSR Sb. 14 (1971). 2. Banach cohomology of piecewise strictly pseudoconvex domains, Mat. Sb. 88(130) (1972), 238-255; English transl. in Math. USSR Sb. 17 (1972). I. I. Privalov 1. Boundary properties of analytic functions, 2nd ed., GITTL, Moscow, 1950; German transl., VEB Deutscher Verlag Wiss., Berlin, 1956.
John Rainwater 1. A note on the preceding paper, Duke Math. J. 36 (1969), 799-800. Enrique Ramirez de Arellano 1. Ein Divisionsproblem and Randintegraldarstellungen in der komplexen Analysis, Math. Ann. 184 (1969/70), 172-187. Georges de Rham 1. Varietes differentiables. Formes, courants, formes harmoniques, Actualites Sci. Indust., No. 1222, Hermann; Paris, 1955.
G. B. Rizza 1. Su diverse estensioni dell'invariante di E. E. Levi nella teoria dellefunzioni di pal variabili complesse, Ann. Mat. Pura Appl. (4) 44 (1957), 73-89.
Ya. A. Roltberg On boundary values of generalized solutions of elliptic equations, Mat. Sb. 86(128) (1971), 248-267; English transl. in Math. USSR Sb. 15 (1971). A. V. Romanov 1'. Spectral analysis of the Martinelli-Bochner operator for the ball in C", and its applications,
Funktsional. Anal. i Prilozhen. 12 (1978), no. 3, 86-87; English transl. in Functional Anal. Appl. 12 (1978). 2*.
Convergence of iterates of the Bochner-Martinelli operator, and the Cauchy-Riemann equation, Dokl. Akad. Nauk SSSR 242 (1978), 780-783; English transl. in Soviet Math.
Dokl. 19 (1978). A. V. Romanov and G. M. Kenkin 1. Exact Holder estimates for the solutions of the a-equation, Izv. Akad. Nauk SSSR Ser. Mat. 35 (1971),1171-1183; English transl. in Math. USSR Izv. 5 (1971). Walter Rudin 1 . Pluriharmonic functions in balls, Proc. Amer. Math. Soc. 62 (1977), 44-46. B. V. Shabat 1. Introduction to complex analysis, "Nauka", Moscow, 1969. (Russian) Pierre Schapira 1. Theorie des hyperfonctions, Lecture Notes in Math., vol. 126, Springer-Verlag, 1970. A. I. Serbin 1. On functions representable by a Martinelli-Bochner integral, Dokl. Akad. Nauk SSSR 1% (1971), 1276-1279; English transl. in Soviet Math. Dokl. 12 (1971). G. E. Shilov 1.
Mathematical analysis: Second special course, "Nauka", Moscow, 1965; English transl., Generalized functions and partial differential equations, Gordon and Breach, New York, 1968.
160
BIBLIOGRAPHY
Henri Skoda I'. Zeros des fonctions de la classe de Nevanlinna dans les ouverts strictement pseudoconvexes, C. R. Acad. Sci. Paris Ser. A-B 280 (1975), A1677-A1680. 2*.
Valeurs ou bord pour les solutions de l'operateur d", et caracterisation des zeros des fonctions de la classe de Nevanlinna, Bull. Soc. Math. France 104 (1976), 225-299.
S. L. Sobolev 1*.
Introduction to the theory of cubature formulas, "Nauka", Moscow, 1974. (Russian)
Elias M. Stein 1. Boundary behavior of holomorphic functions of several complex variables, Math. Notes, No. 11, Princeton Univ. Press, Princeton, N.J., and Univ. of Tokyo Press, Tokyo, 1972. 2*. Singular integrals and differentiability properties of functions, Princeton Univ. Press,
Princeton, N.J., 1970.
Edgar Lee Stout 1'. The boundary values of holomorphic functions of several complex variables, Duke Math. J. 44 (1977), 105-108. Heinz Gunther Tillmann 1*. Darstellung der Schwartzachen Distributionen durch analytische Funktionen, Math. Z. 77 (1961), 106-124. A. G. Vitushkin 1.
Necessary and sufficient conditions a set should satisfy in order that any function continuous on it can be approximated uniformly by analytic or rational functions, Dokl. Akad. Nauk SSSR 128 (1959), 17-20. (Russian)
V. S. Vladimirov Methods of the theory of functions of many complex variables, "Nauka", Moscow, 1964; I. English transi., M.I.T. Press, Cambridge, Mass., 1966. 2.
The Laplace transform of generalized functions of slow growth, Itogi Nauki: Sovremennye
Problemy Mat., Vol. 1, VINITI, Moscow, 1973, pp. 61-84; English transl. in J. Soviet Math. 3 (1975), no. 2. 3.
On the construction of envelopes of holomorphy for regions of a special type, and their applications, Trudy Mat. Inst. Steklov. 60 (1961), 101-144; English transl. in Amer. Math. Soc. Transl. (2) 48 (1965).
4.
Problems of linear junction of holomorphic functions of several complex variables, Izv. Akad.
Nauk SSSR Ser. Mat. 29 (1965), 807-835; English transl. in Amer. Math. Soc. Transl. (2) 71(1968).
5'. Equations of mathematical physics, "Nauka", Moscow, 1967; English transl., Marcel Dekker, New York, 1971.
Dietmar Vogt 1. Distributionen auf dem R" als Randverteilungen holomorpher Funktionen, J. Reine Angew. Math. 261 (1973), 134-145. E. A. Volkov 1*. Boundaries of subdomains, Holder weight classes and solutions in these classes of the Poisson equation, Trudy Mat. Inst. Steklov. 117 (1972), 75-99; English transl. in Proc. Steklov Inst. Math. 117 (1972).
BIBLIOGRAPHY
161
Barnet M. Weinstock 1. Continuous boundary values of analytic functions of several complex variables, Proc. Amer. Math. Soc. 21 (1969), 463-466. 2. An approximation theorem for a-closed forms of type (n, n - 1), Proc. Amer. Math. Soc. 26 (1970), 625-628. 3*. Uniform approximation by solutions of elliptic equations, Proc. Amer. Math. Soc. 41 (1973), 513-517. John Wermer 1. An example concerning polynomial convexity, Math. Ann. 139 (1959), 147-150; addendum, ibid. 140 (1960), 322-323. A. P. Yuzhakov and V. P. Krivokolesko Certain properties of linearly convex domains with smooth boundaries in C", Sibirsk. Mat. 1. Zh. 12 (1971), 452-458; English transl. in Siberian Math. J. 12 (1971).
SUBJECT INDEX a-problem, 31, 86
Formula, of Martinelli-Bochner for exterior differential forms, ((1.1)), 5, 7,..., 87 Functions orthogonal to holomorphic functions, 37
Cousin data, 58, 59 Derivatives of distributions, 82, 84 Determinant, Levi's, 42 Distributions of class A, 79, 81, 82, 88 Domain, admissible, 62 linearly convex, 41, 68, 74, 75
Kernel, reproducing, 72-74 Cauchy, 70, 71 Cauchy-Fantappib, 7,73 - , of Martinelli-Bochner, 71, 77
_ , in the sense of Martineau, 43
Problem, first Cousin, for forms, 58
- , with smooth boundary, 1, 4, 18, 22,
- , of Mittag-Leffler for forms, 57
26, ... , 86, 88
Riemann's additive, 18
- , with boundary of class C^',a, 4, 19,
- , - , for forms, 26
20,24,..., 73, 74
- , strictly pseudoconvex, 2, 31,..., 87, 88
Product of distributions, 2, 76, 82, 88
Representation, integral, of holomorphic functions, 2, 71, 88
Envelope of holomorphy of a compact set,
- , Martinelli-Bochner, of a distribu-
47
Equations,
tion, 77, 78
tangential Cauchy-Riemann,
- , analytic, of a distribution, 76, 84
30, 85
Solution, weak, of the tangential CauchyRiemann equations, 30, 69, 70
Form, a-closed, 2, 27, 46,..., 87 a-exact, 2, 39,..., 74 double, 6, 7
- , of the first Cousin problem, 58, 59
Levi, 2, 31, 49
Theorem, of F. and M. Riesa, 68-70, 71 Runge's, 53, 86 , for forms (Theorem 9.1), 53,
- , orthogonal to holomorphic forms or functions, 1, 2,..., 71, 72 Formula, of Cauchy-Fantappii, ((13.2)), 38,
..., 86
72,73,86-88
163
INDEX OF SYMBOLS Hl, 67 H(K), 47 int M, 3
A, 82 ACp(S1), 4
Ao(aD), 39 An P(D), 45 Ap(K), 4
Izl, 3
, 3
I (p,9)(D, ry)(z), 8
I(p,9)(D,7)(z), 8
A(12), 4
11111, 3
BA, 82 BA, 82
L(p), 42
Bp,n-,, 27 B(z°, r), 3 Cn, 3
M', 3
Z(p, )(O)' 4 a,., 56
an, 79 y±, 20 7+, 26 5, 79 6+, 79
M, 3
MCN, 3
&-, 81
O(3D), 37 Ol(3D), 68
Ok, 81
C11) F),3 (p,9)( C-:)'(0), 4 (p9)
02(aD), 68 P(x-1), 81
Ilo, 11
diam M, 3 dz, 3
P(S, z), 31
C;.1L')`(F), 3
µQ, 11
P(xnlxI-2n), 81
r(z), 62 supp a, 4 t(s, z), 3
dzl, 3 dz[IJ, 3
aa, 4 D, 43 DIIJ, 3 em), 5
g* f, 56
Bo, 81
µ9, 11 pr(z), 56 p(z, M), 3
v(I), 9 a(I, k), 9
Up,q(S, z), 7
3
Vq, 9
18
Wp,9('W, S, z), 7
4 (S, z), 31
z, 3
W (W, S, z), 7
ABCDE FGHIJ -EB-89876543
165