BOUNDED ANALYTIC FUNCTIONS ON DOMAINS OF

INFINITE CONNECTIVITY
BY
LAWRENCE ZALCMANC)

1. Introduction. This paper is a contribution to the study of bounded analytic

functions on plane domains of infinite connectivity. For a fixed domain D, the set
B{D) of all such functions is a Banach algebra under the natural (pointwise)
operations and the uniform norm. Thus, our results generalize theorems about
H™, the much-studied Banach algebra of all functions bounded and analytic on
the open unit disc.
The plan of the paper is this. §2 is devoted to a description of the kind of domains
we shall consider and to various preliminaries concerning the representation and
approximation of functions on these domains. In §3, we prove that the algebra
A(D) of all functions continuous on D and analytic on D is weak * dense in B(D).
We conclude, in passing, that for each / in B(D) there exists a sequence /„ of
functions in A{D) such that ||/„||x ^ ||/!U and/„(z)->-/(z) uniformly on compact
subsets of D; in fact,/„->/almost everywhere on dD. §4 brings together some
facts which relate the boundary behavior of functions in B(D) to the geometry of
the maximal ideal space M of B(D). §5 focuses attention on a special homomorphism
in M which has several interesting properties. In §6, we are concerned with ex-
tending a recent theorem of Sarason from the disc to our more general domains.
It was this problem that originally captured our interest and provided much of the
motivation for the research in this paper. The purpose of including these somewhat
more specialized results here is to point up some of the problems that arise in
studying infinitely connected domains and to indicate how these difficulties may be
circumvented. §7 consists of comments on the generality of our arguments and
explains how various results can be extended to a larger class of domains. Finally,
in §8, we offer a brief survey of the literature, together with some comments on the
points of tangency between our work and that of other mathematicians.
Throughout, we shall assume that the reader is familiar with the basic H™
terminology and theory as presented in [17].
This paper is the author's doctoral thesis, submitted to the Massachusetts
Institute of Technology. I should like to thank Professor Kenneth Hoffman, who
directed this work, for his patience and encouragement. I also wish to thank the
referee for his many helpful suggestions.

Received by the editors May 27, 1968 and, in revised form, August 8, 1968.
P) Part of this research was carried out while the author held a National Science
Foundation Graduate Fellowship. Research at Stanford supported by contract AF 49(638)1345.
241

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 242 LAWRENCE ZALCMAN [October

2. Preliminaries. We shall be concerned with B(D) for certain sets D whose

topological structure is especially simple. To be precise, D will always denote a
domain obtained by deleting from the punctured disc {0<|z|<l} a disjoint
sequence of closed discs {|z—xn| :£/•„} centered on the positive real axis and
accumulating only at 0. By analogy with the terminology for compact sets [29,
p. 57], we call D a domain of type (L). Such a domain is obviously the interior of
its closure (£>= (D)°), and moreover, its boundary 8D has finite length. As usual,
we denote arc length measure on 8D by ds.
We shall find the following notation convenient: A = {|z|<l}, y0 = {|z| = l}
yn = {\z-xn\=rv){n = \,2,.. .\Tn = \Jl=0yk,Y = dD. For purposes of integration,
we orient y0 in the counterclockwise direction and yn («^ 1) in the clockwise sense
and endow Tn (w=0, 1, 2,...) with the corresponding orientation. Unless other-
wise specified, R is an arbitrary plane domain. Interior points of R (or D) will be
denoted by italic letters (most often z), while the generic boundary point will be
called I or £.
At this point it seems worth remarking that it is principally for reasons of
convenience that we have stated our theorems in terms of domains of type (L) :
virtually every result will have more-or-less obvious extensions to more general
sorts of domains. Frequently we shall note these extensions explicitly ; less obvious
generalizations will be treated in §7.
Now suppose D is given. Then we have

Proposition 2.1. Let feB(D). Then f has nontangential boundary values at

almost every point of 8D.

Proof. This is an immediate consequence of the corresponding fact for H™

functions. Indeed, if £ e 8D\{0} then £ has a neighborhood N in D whose interior
is analytically equivalent to A; moreover, £ lies on an analytic subarc of 8N.
If <f>maps A conformally onto N, then/o <f>eH"; and <f> extends analytically to a
full neighborhood of <^-1(£) by the reflection principle. Hence, a nontangential
path at <f>~\0 is mapped onto a nontangential path at £. It follows that/has
nontangential boundary values almost everywhere (d8) on each yn. Since 8D\{J y„
= {0} has arclength 0 and ds«dd on each yn, these boundary values exist almost
everywhere on 8D(ds).
We denote the boundary function of/(z) e _S(D) by/(Q and observe that/(£)
is a bounded measurable function on V = 8D. The interior function /(z) and the
boundary function/(Q are related by

Proposition 2.2. Let /(z) e B{D). Then

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 243

Proof. Fix z and let Cn={|z|=en} be a counterclockwise oriented circle in D

which separates yn from yn+1. By Cauchy's theorem, we have

'»-¿t.fS'-Bt.S*-**♦♦«*
for large enough n. The integral I2(n) is bounded by ||/|| ven[d(z, C„)]-1, where
d(z, Cn) is the distance from z to C„. Thus as n -> co, /2(«) -»■0. If we take limits on
both sides, we obtain the desired conclusion.
Proposition 2.2 remains valid for domains whose (rectifiable) boundary consists
of a countable number of disjoint smooth simple closed curves r„ which accumulate
on a (totally disconnected) set S of Hausdorff linear measure 0. (S may, but need
not, intersect (J t„.) Indeed, in this case, each bounded analytic function has
appropriately defined boundary values (cf. §7), and the proof of 2.2 can be applied
with only slight modification. It seems likely that the Hausdorff measure condition
can be weakened to the requirement that 5" have zero analytic capacity [29].
Generally speaking, however, the representation of 2.2 fails utterly. Of course,
if 8R is not rectifiable, the boundary function may not exist in the proper sense;
and, if it does exist, the integral will still fail to converge absolutely. Moreover, if
dR is rectifiable or even analytic 2.2 may break down if the set at which the boundary
curves accumulate is too large. For instance, we have
Example 2.3. Let S be the slit [—iß, i¡2], and let R be the region obtained by
deleting from the slit disc A\5 a disjoint sequence An (n= 1, 2,...) of closed discs
with radii rn such that 2rn<°° ana" the A„ accumulate at every point of S (on
both sides) and at no other points. Let i-n= 3An (oriented clockwise), r0 = y0. If
f e B(R), then/has nontangential boundary values at almost every (ds) point of
(J t„. However, in general, / will not have well-defined boundary values at points
of S. We claim that there exists fe B(R) such that

(1) m *i~.[ y®-dl

Indeed, let/be a Riemann map of the complement of S in the complex sphere onto
A; then clearly fe B(R). Also,

(2) ( M)dt = 0, n=\,2.

since/is analytic in a disc containing each i-n. Now suppose

(3) **-hb.M*-
and let C<=R be a smooth curve which surrounds S once. We have

(4) jcf(z)dz = 2nif'(oo)^0,

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 244 LAWRENCE ZALCMAN [October

where the integral is taken in the counterclockwise direction and f'(oo) is the
coefficient of z_1 in the expansion of/at oo; this coefficient is nonzero since/is
schlicht in the complement of S. However,

Since
SÀLn«h-LM-Lêè«-
,« =—: -= = 1 if £ lies inside C,
(6) 27rjJcz-£
= 0 otherwise,
it follows from (2) and (6) that the right-hand side of (5) above is 0, contradicting
the conjunction of (3) and (4). Hence (1) holds, as was claimed.
The informed reader will have noted that what is of relevance to the above
construction is that 5" has positive analytic capacity; S could have been replaced
by any closed set of positive analytic capacity. For instance, one could have chosen
any compact subset of the segment (—/', i) of positive linear measure.
Remark. In 2.3 above it is not hard to see that if g e A(R) one still has
M if Ä«.
Itti Jut» £-z
However, in general, even this sort of representation will fail; one need only choose
the set at which the "nice" boundary accumulates to be sufficiently thick. For
instance, let R be a region whose boundary consists of a union of circles t„ of
finite total length plus a set of positive AC capacity [29, p. 77] (e.g., an arc of
positive two-dimensional measure) at which the circles accumulate. Then it is not
true that each function in A(R) can be represented as the Cauchy integral of its
(continuous!) boundary function taken over (J rn (cf. [29, pp. 69-76]). On the
other hand, in such cases 8R is not rectifiable. Indeed, it is easy to show that if a
set has finite Hausdorff linear measure it has zero AC capacity.
We shall also need the following result from the theory of rational approximation.
Theorem 2.4. Let D be a domain of type (L). Each function in A(D) can be
approximated uniformly on D by rational functions with poles off D. These poles may
be chosen to lie in the set {oo, xlt x2,...}.
Proof. Let gn be a C °° function defined on the plane such that
(l) gn=lon{\z\él¡n},
(2)gn=0on{|z|^2/«},
(3) pgJdzU^lOn.
Suppose feA(D), and let o0 be given. Extend /continuously to a (bounded)
function on the complex sphere and set

■/W-ZWafe**
w —z 8w
'•<*>-;/!•
= ñz)gn{zHX-\\M-zd-§dxdy, w = x + iy.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 245
(Here z is not necessarily restricted to D and the integration is taken over the
entire complex plane.) It is easy to see (cf. [29, pp. 88-89]) that/, is continuous
on the whole plane and analytic on D; moreover, ||/n||oo^40to/(4/«), where ^(S)
is the modulus of continuity of (the extension of)/ Since/is uniformly continuous,
¡AHa» -> 0. Further,/-/, is analytic on {|z| ^ 1/«} since the measure (dgjdw) dx dy
has no mass there. Thus hn=f—fn is continuous on D and analytic on D as well
as a full neighborhood of 0; i.e., for large enough k = k(n), hn e A(Dk), where Dk is
the finitely connected domain bounded by I\. By a classical theorem, each function
in A(Dk) can be approximated uniformly by rational functions whose poles lie in
{co, xu x2,..., xk}. Choose n so large that ojf(4¡n)<e/80 and then pick a rational
function r with poles in {oo, xu x2, ■■., xk) such that \hn —r\nk<e¡2. We have

11/-rile^ IK-rlk+l/nb < «,

as required.
The preceding proof is T. W. Gamelin's rearrangement of an argument due to
John Garnett (cf. [29, pp. 126-128]). Actually, 2.4 is a special case of a very general
theorem of A. G. Vitushkin; it is only for completeness and for the reader's con-
venience that we have included it here. Since we have treated these matters in
considerable detail elsewhere [29], we shall not comment on them further.
Finally, we require a result from the theory of cluster sets [6]. Recall that if/
is an (arbitrary) function defined on a plane domain R the cluster set of/at ae R,
denoted by Cl (/; a), is the set of limiting values of / at a. More precisely,
ß e Cl (/; a) if and only if there exists a sequence zn e R\{a} such that zn -> a and
/(z„) -> ß. If a is a nonisolated boundary point of R, one can also define the
boundary cluster set of / at a, C1B(/; a). We write ß e C1B(/; a) if and only if
there exists a sequence {<*„}<=
dR\{a} and elements ßn e Cl (/; an) such that an -> a,
ßn -*■ß. In 1914 F. Iversen proved the following beautiful and important theorem.

Theorem 2.5. Let Rbea bounded domain and let ledRbea nonisolated boundary
point. Then iff is meromorphic on R

3(C1(/;0)CC1B (/;£).
Proof. [26, pp. 332-335].
Of course, we shall be interested in the case R = D,fe B(D).

3. Weak star density. According to 2.1 and 2.2, the correspondence between
bounded analytic functions on D and their boundary-value functions is an iso-
morphism between B(D) and a subalgebra ofLco(dD; ds). There are various ways
of seeing that this isomorphism is actually an isometry. For instance, we may
observe that, by 2.2, evaluation at z e D is a continuous homomorphism of the
boundary value algebra onto the complex numbers and hence extends to the L00
closure of this algebra. Since this latter algebra is complete, the homomorphism
must actually have unit norm. Letting z vary over D, we obtain the required
conclusion. It is now obvious that the boundary algebra was actually closed to

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 246 LAWRENCE ZALCMAN [October

begin with. We shall identify B(D) with its (closed) boundary-value algebra and,
by an abuse of language, denote the latter algebra also by B(D). A similar con-
vention will hold for A(D), which we regard as a closed subalgebra of B{D).
Thus, in the sequel, context will determine whether functions of B(D) or A(D)
are to be regarded as defined on D or on 8D.
By the weak * topology on B(D) we mean the weak topology B(D) inherits as a
closed subspace of LX(8D; ds) in the pairing (¿"(cfc), L\ds)). The principal result
of this section is

Theorem 3.1. B{D) is the weak * closure of A(D) in L™(8D; ds).

The proof will proceed by a series of auxiliary propositions. First of all, we have

Lemma 3.2. LetfeL\8D; ds) satisfy

(1) £/(£)g(£) ¿£ = 0

for all g e A(D) and set

Then, for almost every £ e 8D,f(z) ->/(£) as z ->- £ nontangentially. Moreover,

/•2JJ
lim |/(ei9)-/(rei9)l dB = 0,
i—>i-Jo
(3)
V ' f2jl
Hm \f(xn + rne^)-f(xn+reie)\ dB = 0 («=1,2,...).
r-r„+ Jo

Proof. Let, as in 2.4, Dn be the (finitely-connected) domain bounded by Tn

(« = 1,2,...) and set, for £ e r„,

(4) m=m-ihiLMdè-
Suppose g e A(Dn). Then if v is a finite measure on r\rn we have the following
balayage formula :

,. f g(í)<Kí)=
Jr\r„ f (¿f
Jr\rn \2irl Jr„ «(OÄiiö
£-f/

We shall choose dv{£)=/(f) d£. Since g e A(D) we have by (1)

(6) 0= JrÍ g(i)f{i)d(=Jr„

f +fJr\r„ •

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 247

Hence, by (5) and (6),

Combining (4) and (7) we get

(8) f g(O/n©^ = 0, geA(Dn).

Thus/n(0 e H\Dn) [22]. Therefore, for z e Dn,

'■»-¿US*
V' lid Jrn C-z 2tt¡ Jr„ \2m Jr\r„ f-£ 7 i-z

-UrßMUU,xrere^
A partial fraction decomposition and the residue theorem show that the last
integral in (9) vanishes. Thus

do) '*>-¿US*
and so

no /(*)=/»(*)+¿í £&dt.
2m Jr\rn f-z
Since /n e H1{DV), /„(z) ->/„(£) a.e. on Tn as z^îeT, nontangentially [22].
Thus, for a.a. £ e Tn

lim/(z) = limUz) + lim*L f ßö #

2-c ï-t 2-î ¿w' Jr\r„ s —z

for nontangential approach.

In a similar way, the second part of the lemma follows from the representation
(11).
A similar result, with a different proof, is given in [9].
Next we have
Proposition 3.3. Suppose h e B(D) is continuous on D\{0}. Then h can be approxi-
mated pointwise boundedly on dD\{0} by functions in A{D).

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 248 LAWRENCE ZALCMAN [October

Proof. The proof uses a technique due to Arens [2, pp. 644-646]. Let H(z) = zh(z) ;
then HeA(D). Let H(z)=0, z$D, and write
, , . n2 rr H(w)-H(z) , ,
77 JJsn w —z

where 5'n= {|w| á 1/«}. Here z is allowed to vary over the whole plane. Then hn,
being (essentially) the sum of convolutions, is continuous on D; moreover, it is
not hard to see that it is analytic on D [2, p. 644]. Thus «„(z) e A(D). Now

so

|z«n(z)-//(z)|á(|z|/Xl/«)+ (l/«)|#(z)|)(^JJs i^i).

where ¿>(S)= sup {|/f(z)| : |z| <8}. It follows that

\zhn{z)-H{z)\ S (|z|è(l/«) + (l/«)|z| |«(z)|)2«

or
|«n(z)-«(z)| í 2«è(l/«) + 2|«(z)|.
Nowè(l/«) = sup{|z«(z)| : z e ■SB}á(l/»)||A||.. Thus
|«n(z)-«(z)| ^ 4||A|U
so that ||A„|| ooá 51|A|| «,. It is shown in [2] that zhn -> H on D uniformly. It follows
that «n -> « pointwise boundedly on D\{0} and that convergence is uniform off
any neighborhood of 0. That completes the proof.
A different proof of 3.3 can be based on the technique of 2.4.
Finally, we need the following general result [19, p. 454].
Proposition 3.4. Let ^ be a finite positive measure and let S be a subspace of
Lœ(dfj.). S is weak * closed if and only if {/„}cS, /„ ->/pointwise boundedly a.e.
implies that fe S.
Proof. One direction is clear. Suppose that each boundedly pointwise convergent
sequence in S converges to an element of S. We claim S is weak * closed. It is
enough, by the Krein-Shmulian theorem [7, p. 429], to show that the unit ball U
of S is (weak *) closed. If/ is in the weak * closure of U it is also in the weak L2
closure of U and hence in the (norm) L2 closure of U, since U is convex. It follows
that some sequence in U converges to / in L2(d¡x),so an appropriate subsequence
converges pointwise a.e. (d¡x). By hypothesis, fe S.
Actually, 3.4 holds for arbitrary convex sets in L00; the proof remains unchanged.
Proof of 3.1. First we shall prove that A(D) is weak * dense in B(D). It is, of
course, enough to show that any function feL\8D; ds) which is orthogonal to
A(D) is also orthogonal to B(D). Suppose, then, that

(1) £/(£)£(£) = 0, geA(D).

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 249
Choose curves y'k (k = 0, 1, 2,...) in D concentric with yk in such a way that y'k
separates yK from F\yk and let D' denote the subdomain of D bounded by the y'k,
T' = 8D'. D' is clearly a domain of type (L). Further, since by 3.2 the analytic
extension (Cauchy transform)/(z) of/converges back to /in L1 norm on each
boundary curve yk, we may assume that the curves yk have been chosen in such a
way that/(z) eL\8D'; ds).
Now let h e B(D) be fixed. We must show

(2) JAQKQ<Z
= 0-
For each k we have (by 3.2 and the Cauchy theorem)

(3) f KWS>di=f Mh(z)dz.

Hence, since all integrals are absolutely convergent,

(4) jr RQKQ ft = £ f(z)h(z)dz.

By (1) and (4),
(5) f /(z)r(z)¿z = 0
Jv
for each rational function r with poles off D. It follows, by 2.4, that

(6) f f(z)g(z)dz = 0, geA(D').

Jr'

Now let hn be as in 3.3. Then hn e A(D') and hn(z) -* h(z) pointwise boundedly on
8D'\{0}. Hence, by (6), we have

(7) 0 = lim Í f(z)hn{z)dz= f f(z)h{z)dz.

Combining (4) and (7), we obtain (2), as required.

Since A{D)^B(D), it remains only to show that B{D) is weak * closed. We shall
verify the condition of 3.4. Suppose, then, that/,^>/pointwise boundedly a.e.
on 8D, /„ e B{D). Since /„ is a normal family, there exists g e B{D) such that
/„(z) -> g(z) uniformly on compact subsets of D. Clearly,

(8)
Hence

(9)

On the other hand, by Cauchy's theorem,

(10> h\Mdl = 0> heB(D)

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 250 LAWRENCE ZALCMAN [October

if z £ D. Thus, by the dominated convergence theorem,

(11) ^iMzmd^o
2-TTtJr L,—z
for z$ D. By (9) and (11)

(12) JrCAO-*(9)¿# = o
for any z £ <5Z).It follows by the theorem of Hartogs and Rosenthal [29, p. 110] that

(i3) £(/(o-^(owo^ = o
for any k e C(8D). Evidently, /(£)=#(£) a.e. on 8D, i.e., /e B(D). The proof is
complete.
Corollary 3.5. If ¡j. is a positive measure on 8D which is absolutely continuous
with respect to ds then B{D) is the weak * closure of A(D) in £"(4*).

Proof. Obvious.
In particular, if z e D is fixed and fiz denotes harmonic measure for z, we have
dfts<Kds,so that 3.5 applies.
As an application of 3.1 we have

Theorem 3.6. Given fe B{D), there exists a sequence {/„} of functions in A(D)
such that

(1) ll/.l- ^ 11/11

»,
(2) /»(ö-/(ö a.e. on 8D.
In particular, fn{z) -+f(z)for each z e D, and the convergence is uniform on compact
subsets of D.
Proof. By [1, p. 939] it is enough to prove that

sup{ jhfds :heB(D), ||«||„ £ l\ S sup { jgfds :geA(D), \\g\\œí l}

for each feV-(ds). In other words, we must show that the functional which/
induces on B(D) has the same norm as that of the functional which / induces on
A(D). Denote the norm of the latter functional by N. By the Hahn-Banach theorem
and the Riesz representation theorem, there exists a measure ft on 8D such that
| g dfi=§ gfds for all geA(D), and ||/*|=JV. We may choose /u to satisfy the
additional condition /j.({0})= 0. Indeed, suppose ¡j.({0})= c^0. Then 0 is not a peak
point for A(D), so there exists a probability measure v on 8D such that J"g dv=g(0)
for all geA(D) and K{°})= °- (A reference for these facts is [29, pp. 4-8].) Let
fj.' = (ji-cS0) + cv. Clearly, if g e A(D), { g 4t = J g d\x and furthermore,
||/*'|| = ||0.-c8o) +o|| Ú ||^-cSo|+c = (|H|-c) + c = |/i||;

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 251
finally, ¡x'({0})=0. Since dp-fas annihilates A(D), it follows from 6.3 below that
dfji'-fds = k ds, keL\8D; ds). Moreover, j hk ds = 0 for every h e B{D), by 3.1.
Hence, J*hfds = J*h dp for each h e B(D), so that

sup|| jhfds :heB(D), \\h\\x ï l} ï |M'| = N,

as required.
Corollary 3.7. B(D) is the weak * closure in ^{SD; ds) of the set of rational
functions with poles off D. The functions /„ in 3.6 may be chosen to be rational
functions.
Proof. By 2.4, rational functions analytic on D are uniformly dense in A(D).
4. The maximal ideal space. If R is a bounded domain and z e R then evaluation
at z gives rise to a complex homomorphism <f>zof B(R) defined by </>z(f)=f(z).
Thus, R can be regarded as a subset of M= M(R), the maximal ideal space of B(R).
In fact, the map z^<f>z embeds R homeomorphically as an open subset R of M
[17, p. 160]. It is natural to ask if the closure of R (in M) is all of M. Even in the
simplest of cases, this question, the so-called corona conjecture, is exceedingly
difficult. For R = A, the open unit disc, the answer is affirmative [5]. Armed with
this fact, it is rather easy to show that the answer remains yes if R is a finite open
Riemann surface (in particular, a plane domain with finite connectivity). On the
other hand, virtually nothing seems to be known about the case of infinite con-
nectivity : at present there are no examples of planar sets of infinite connectivity
for which the corona conjecture is known either to be true or false. Although the
results below do not shed much light on the answer to this question, they do
indicate where some of the major difficulties lie. Much of our discussion will be
framed in terms of a fixed domain D of type (L).
Recall that for a e D, Ma denotes the set of all homomorphisms <f> e M such that
4>{z)= a. We call Ma the fiber over a. Since (z —a)'1 e B(D) for a^ D, it is clear
that {Ma: a e D} is a partition of M; moreover, if a e D, then Ma={<j>a}, as we
have observed above.
If £ e 8D\{0}, it is easy to see that Afç is naturally homeomorphic to the fiber
over 1 in the maximal ideal space of //". This follows, for instance, from the fact
that each fe B(D) has the representation /=/„ +/„, where /„ is analytic on a
neighborhood of yn while /„ is bounded and analytic on the (simply-connected)
component of the complement of yn in the sphere which intersects D. Explicitly,

¿»-¿/.g* «»-¿JL8*
The reasoning for the disc also shows that, for each n, LWn ^c ¡s connected and
homeomorphic to Af(A)\A. Furthermore, the Shilov boundary X of B(D) is the
maximal ideal space of L^ÇdD; ds). These facts and their unstated ramifications

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 252 LAWRENCE ZALCMAN [October

follow in a routine fashion from the H" theory, so we shall omit their proofs.
What should be obvious at this point is that for domains of type (L) the only point
where anything new can happen is the origin. Accordingly, we shall concentrate
our attention at that point.
The first observation worth making is
Proposition 4.1. D is dense in M\M0.
Proof. Let <j>eMK, £e3Z)\{0}; we claim there exists a net {z¡}cD such that
<f>Zi
—>(f>.To simplify notation we may assume that £ e y0. Let <j> denote the restric-
tion of <j>to Hm = 5(A). By the corona theorem there exists a net {z(}in A such that
«¿2,-*■ip in M(A). In particular, <f>z¡(z)
-> ifi(z)= £ so that z¡ -*■£. Thus, we may
assume {z¡}<=Z).Let/e B(D). We have

<t>(f)
= Wo+/o) = <Mfo)
+ <Kfo)= Wo)+/o(£)
and
tzjj) = /«(*«)+/ofe) ■*Hfo)+/o(£),
so that <f>Si(f)
-> <f>(f)for each/e B{D), as required.
The argument above, dependent as it is on the case of the disc, disguises the
essential difficulty involved in the corona conjecture. The problem in question is
made more explicit in
Proposition 4.2. These are equivalent:
(1) È. is dense in M{R).
(2) Givenfu /2, ...,/„ e B(R) such that
1/iWI+ l/af»!+ ■••+ [fM ^ S > 0, zeR,
there exist gu g2, ■■-,gne B(R) such that

figi +Í2g2 +■■■+fngn = 1 on R.

Proof. This is an easy consequence of the definition of the topology of M(R).
For details, see [17, p. 163].
Even very weak versions of condition (2) of 4.2 yield interesting information
about B(R). For instance, we have

Proposition 4.3. Let R be a bounded plane domain, le8R. Then the following
conditions are equivalent.
(1) IffeB{R) satisfies |z-£| + |/(z)| aS>0 on R then there exist g,heB(R)
such that (z-£)g+/« = l on R.
(2) Iff s B(R)thenf(M^C\ (/; £).
Proof. Suppose (2) is false. Then there exists fe B(R) such that 0 e/(Afc) but
0 £ Cl (/; £). Clearly, / satisfies the condition of (1) for an appropriately small 8,
so (z-C)g+fh= 1 for some g, « e B(R). Choosing <f>eMc such that 4>(J')=/(</>)
=0,
we obtain a contradiction. Conversely, if (1) is false for some/e B(R) it follows

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 253
from the Gelfand-Mazur theorem that there exists <j>
e M such that <f>(f)=<f>(z—£,)
=0. Clearly 0 e Mt. It is obvious that 0 e/(Mc)\Cl (/; £).
It would be extremely interesting to have an example of a domain for which
conditions (1) and (2) above do not hold. For such a domain, of course, the
corona conjecture would fail in the worst possible way. We should point out that
the inclusion Cl (/; 0c/(^c) always holds. Indeed, suppose z„ -> £ and/(zn) ->-/}.
Let / be the ideal in B(R) of all functions h such that h(zn) -> 0. / is a proper
ideal and is thus contained in a maximal ideal, i.e., there exists $ e M(R) such that
<l>(h)
= 0 for all h el. Since z-£,/-£ e I, we have <¿(z-£)=0, <j>(f-ß)=0. The first
equality shows that <j>eMKand the second gives <j>{f)=ß.
When R = A, it is known that (1) of 4.3 is true. Of course, this follows from
the corona theorem, but it can also be obtained as a consequence of the Riesz-
Nevanlinna factorization theory for functions in Hx (see [17, p. 162] for details).
Hence, if D is a domain of type (L) and £ g 8D\{0} it follows (by decomposing
integrals, for instance) that/(MÇ) = C1 (/; £). The proof breaks down for £=0;
entirely different techniques are required to treat this point.

Proposition 4.4. Letfe B(D) and suppose \z\ + |/(z)| ä S > Ofor z e D. Then there
exist g, he B(D) such thatf(z)g(z) + zh(z)=l, ze D.
Proof. Let y = {|z|=e} be a circle in D such that |/(z)|>8/2 on and inside y.
Let Ty be that part of 8D lying inside of y and set

m aM _Lf_L^_J_f J_dl_
K) g{) 2ni)yf(0 £ 2ni]ry M)i-Z
where the integrals are taken in the counterclockwise direction. Then

as z -> 0. It follows from (1) that g is analytic off the compact set bounded by r„
and it is clear from (2) and the maximum modulus principle that g is bounded;
hence geB(D). Since / is bounded away from 0 near 0 we have g(z)/(z)—1
= 0(|z|), z -> 0. Set h(z) = (\ -g(z)f(z))lz. Done.
The elegant proof of 4.4 was communicated to me by Lennart Carleson.
Combining 4.3 and 4.4 we obtain
Proposition 4.5. Letfe B(D). Thenf(M0) = Cl (/; 0).
Corollary 4.6. M0 is connected.
Proof. (Cf. [18, p. 188].) First of all, we note that M0 is a hull; indeed, it is the
zero set of the function z. Hence [20, p. 116], M0 is the maximal ideal space of the
Banach algebra B(D)jI, where / is the ideal of all functions whose Gelfand trans-
forms vanish identically on M0. Suppose M0=K0 u Ku where the K, are disjoint
compact sets. By the Shilov idempotent theorem [20, p. 168], there exists/e B(D)/I

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 254 LAWRENCE ZALCMAN [October

such that/^=y on K¡. Let n: B(D)-^ B(D)¡I be the quotient map, and choose
feB(D) such that Tr(f)=f. Clearly,/(A/o) = {0, 1}. But /(M0) = Cl (/; 0) by 6.5,
and Cl (/; 0) is connected since D is locally connected at 0 [5, p. 3]. This contra-
diction shows that M0 must be connected.
There is an alternative argument, independent of Proposition 4.4, which is also
worth presenting. We argue as above to conclude that/(Mo) = {0, 1}. Since Cl (/; 0)
is connected and contained in f(M0) it must be a singleton, say Cl(/;0)={0}.
We will show that K1 does not meet the closure cl (D) of D in M. First of all, note
that since K1= MQn{/=l}, we must have/(<£)=l for each <j>eK^. Suppose that
<j>
e Kx C\ cl (D). Then there exists a net {z¡} in D such that <j>Zt -> <f>in M ; i.e.,
«(z¡) -* h(4>)for each « e B(D). Taking « = z, we have zt -> 0 in D. Thus,/(z¡) -> 0
since 0 is the unique cluster value of/at 0. This contradicts/(z¡) ->/(0) = 1. Thus
A^ncl(Z>)=0. We now obtain a contradiction by observing that the Shilov
boundary of B(D) is contained in cl (£>), yet meets each component of M.
The content of the preceding paragraph can be summarized in
Proposition 4.7. Let R be a bounded plane domain, fe B(R). If t,e8R and R is
locally connected at £ then MKis connected.
It is easy to see that if R fails to be locally connected at £, Mc can be disconnected.
In fact, the example of 2.3 shows that this may happen even if R is the interior of
its closure.
In concluding this section, let us point out that when the conditions of 4.3 are
satisfied the Gelfand theory allows us to reduce (2) of 4.2 to a local condition. More
precisely, we have
Proposition 4.8. Let R be a bounded plane domain and suppose /(MC) = C1 (/; £)
for every fe B(R) and every £ e 8R. Then these are equivalent:
(1) Ê is dense in M(R).
(2) Givenfx,f2, ...,/„ e B(R) such that
\fi(z)\ + |/2(z)| + • • • + |/n(z)| a; 8 > 0, zeR,
there exist for each £ e 8R functions gx(z; £), g2(z; £),..., g„(z; £) e B(R) such that
hdz) = A(z)gl(z; £)+••• +fn(z)gn(z; £)
is bounded away from 0 in a neighborhood (in R) oft.
Proof. Clearly, the / do not all vanish at any point of R. If <f>
e Afc(£ £ 8R),
then <f>(fj)=0,l&jZn, implies that ¿(nt) = 0; but <¿(«c)=¿0,since 0£C1(«C;£).
Hence, if (2) holds the/ belong to no common maximal ideal. It follows that there
exist gug2,-..,gne B(R) such that/ gx +/2g2 + • • • +/ng„ s 1. By 4.2, we are done.
In particular, it follows from 4.8 and 4.4 that for domains of type (L) it is enough
to exhibit functions gu ..., gn e B(D) such that
\fi(z)gi(z)+ ■■■+fn(z)griz)\ ^ r¡ > 0 near 0

for some 77>0 (the/'s having been given).

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 255
5. A distinguished homomorphism. In this section we shall be concerned with
the problem of how the metric structure of a domain affects its function theory.
Throughout, D will be a domain of type (L) with boundary curves y0 = {|z| = l},
yn={\z-xn\=rn}, n=l,2,....
Let Y be a compact set in the plane, x e Y. We say x is a peak point for A(Y)
if there exists a function /e A(Y) such that (l)/(x) = l and (2) |/(z)|<l for
z e Y\{x}. If R is a domain bounded by finitely many simple closed curves it is easy
to see that each point of 8R is a peak point for A(R). For domains of infinite
connectivity, however, not every point of the boundary need be a peak point. In
fact, we have
Theorem 5.1. 0 is a peak point for A(D) if and only if
CO

n=l An

Proof. The proof is immediate from [29, p. 58] and 2.4.

Theorem 5.1 should be compared with Beck's theorem [3] on "peak points" for
bounded analytic functions. We should also point out that nothing so simple as
5.1 is true even for domains whose structure is very close to that of domains of
type (L) ; for details see [29, pp. 60-62].
Now suppose that D is given and 0 is not a peak point for A(D). Then 2 rjxn < co,
so í/ju(£)= (2it/)"1£"1 di is a finite measure on 8D. By Cauchy's theorem, r(0) =
jr r(£) dp(t) for each rational function analytic on D; hence, by Theorem 2.4 we
have

/(0) = £/(£)^(£), feA(D).

It follows that the measure p is multiplicative on A(D). This property persists in

the (weak *) limit, so that, by 3.1,

J" ghdp = (£ g dp}(£ h dp}, g,he B(D).

Thus, <f>0(h)
= j h dp is a complex homomorphism of B(D); and, since <j>0(z) =0
<f>0
e M0. If Kds is another measure which represents evaluation at 0 on A(D)

/(0) = £/(£)*(£) ds, feA(D),

the same argument shows that t/tK,given by

Mh)= ! h(0K(0ds, heB(D),

Jr
is an element of M0. In fact, <f>0
= <l'Kfor any such K. Indeed, writing
(2tti)-!£-x ¿£-tf ¿fe= Hds, HeL\ds),

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 256 LAWRENCE ZALCMAN [October

we have

j/WH(Qds = 0, feA(D).

Since A(D) is weak * dense in B(D)

f «(£)//(£)ds= 0, heB(D),

which proves our assertion.

Thus, if 2 rjxn < co, there exists a unique complex homomorphism <f>0 in M0
which arises as the extension to B(D) of those complex measures which represent
evaluation at 0 on A(D) and are absolutely continuous with respect to arc length.
The remainder of this section will be devoted to the study of this distinguished
homomorphism. We have

Theorem 5.2. The point 0 e 8D fails to be a peak point for A(D) if and only if
limx^0_ h(x) exists for every heB(D). In this case, the limit actually exists for
approach off any wedge \9\ < 8, and one has

«M«)= urn h(x).

jc->0-

Proof. First suppose 0 is a peak point for A(D) and let / be a function that
peaks there. Let the sequence {zn}<=Z),zn-^-0 be given. Clearly, we can choose a
subsequence {znJ such that f(zni<) -> 1 as rapidly as we like ; in particular, the znic
may be chosen in such a way that {f(znJ} is an interpolating sequence for H"
[17, pp. 194-206].Let heü™ be such that «(/(znJ) = (-l)fc. Then h°feB(D)
and « o/does not tend to a limit on {zn}.Thus, for any curve in D with an endpoint
at 0 there exists a function in B(D) which fails to tend to a limit along that curve.
Suppose next that 0 is not a peak point for A(D). Then 2 trn\xn< co, and so
2 rn\(xn - rn) < co (since rn < xn). Let

Then by Cauchy's formula/(z) = 2n=o/n(z) f°r ze D. Also,

For —i^x^O, we have

|/n(x)| g (27r)-1||/||0O27rrn(xn-rn)-1, « = 1, 2,...,

independently of jc. Thus

I \ux)\ú\\f\u2 7^-°

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 257

as N-> co, uniformly on [—\, 0]. Therefore,

limfix) = lim 2 /„(*)= J /„(O)- <f>0(f),

as required. It now follows by a classical theorem of Lindelöf (cf. [26, p. 306])
that the limit actually exists off any wedge containing the positive real axis.
Alternatively, one can show directly that if 8 > 0 is given there exists e > 0 such that
Zn°=o/n(z) converges uniformly on the set {z : \z\^e and |argz|3:8}. The result
then follows as above.
Henceforth, we shall assume that 0 is not a peak point for A(D), so that the
complex homomorphism <j>0 exists.
At this point, it should be clear that, in some sense, 4>0is related to the homo-
morphisms <f>z, ze D. Part of our problem is to make this statement precise. Of
course, Theorem 5.2 allows us to conclude that <f>0 lies in the sequential closure of
D; however, much more is true. It turns out that the relevant concept is the familiar
notion of (Gleason) part. Recall that two complex homomorphisms, </> and ifi, of a
function algebra 21lie in the same part (<f>~>fi)
if and only if ||(£—</>
| < 2. Equivalently,
<f>~>p
if and only if
sup{|ft/)|:/G«, ¡/I Sl,tff>-<9<1.
~ is an equivalence relation, and the parts of 21 are the equivalence classes into
which it partitions the maximal ideal space of 21. If 21= B(D), it is easy to see that
all of D (i.e., Û) lies in a single part; this follows, for instance, from Schwarz's
lemma applied to a chain of discs. Our claim is

Theorem 5.3. The homomorphism <j>0

lies in the same Gleason part as D.
For the proof, we shall need the following result, itself of some interest.

Proposition 5.4. Let {/„}<=B{D) and suppose /n(z)^/(z), zeD, pointwise

boundedly.Then <f,0(Jn)
-* <f>0(f).
Proof. Obviously, we may assume ¡/„¡„^l. By Vitali's theorem, fn(z)^-f(z)
uniformly on compact subsets of D. Choose disjoint curves y'n= {\z-xn\ =r'n}
(n = 1,2,...) in D so that 2 r'Jxn < °° and \Jk¿ „ y'klies in the unbounded component
of the complement of y'n; let y'o^D be a circle centered at 0 which surrounds
Unœ=iy'n- We write r'N = {J%=0y'n, T'={0} u U"=o y'n- Let D' be the subdomain of
D which has T' as its boundary. Then for ze D' and geB(D) we have

do Im« Jr,r ifiâ*.

w —z
Let e > 0 be given and choose N so large that

-L f É.
2"Jr'\r'N \w\

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 258 LAWRENCE ZALCMAN [October

This is possible since 2 rñlxn<co- Now if/e B(D) it is clear that

Mdw.
w
Hence,

\uf)-uf¿\s\±-A
I•«« Jrj, ^*+fc
w

Since T^ is a compact subset of D,fk(w) ->/(m0 uniformly on r'N as k -*>oo; thus

limsup IMfl-M/ül ^ 2e.

But £ was arbitrary, so <¿0(/)= lim </>Q(fk),

as required.
Actually, even more is true: <¡>0
remains continuous when B(D) is endowed with
the strict topology (/J-topology) [21]. This observation has also been made (in-
dependently) by Rudin [24].
Proof of 5.3. Fix xe D. We must show

sup{|/(x)| : feB(D), \\f\\x g 1, <f>0(f)

= 0} < 1.
Suppose, to the contrary, that there exists a sequence of functions {/„} in B(D)
such that <£o(/n)=0, ||/„|| ^ 1, and/n(x) -> 1 as « -> co. {/„} is a normal family, so by
extracting a subsequence and renumbering we have fn(z) ->/(z) uniformly on
compact subsets of D. Of course, / e B(D) and ||/||«>ál; since/(x) = l, it follows
that/(z)=l.By5.4
0 = lim^o(/n) = M/) = <^o(l)= I,
a contradiction.

Corollary 5.5. Evaluation at 0 lies in the same part of A(D) as D.

The identification of the parts of B(D) remains an open question. Since each
point of 8D\{0} is a peak point for A(D), any part which lies in (J {Afc: £ g 8D\{0}}
must actually lie in a single fiber. In this case, the problem reduces to the identifica-
tion of the parts of H™, for which see Hoffman [18]. Thus, only those parts which
contain points of M0 need be considered.
To conclude this section, we prove

Theorem 5.6. <j>0

does not lie in the Shilov boundary of B(D).

Proof. By [10], there exists/e B(D) such that/is nonconstant and continuous
on D\{0} and |/| = 1 on 8D\{0}. Since the Shilov boundary X of B(D) is just the
maximal ideal space of L°°(SZ); ds) (§4), |/0/>)| = l for any tfieX. On the other
hand, since </>0
lies in the same part as D, \<f>0(J)\
< 1. Hence <^0i X.
A similar proof, not depending on the results of [10], can be based on an abstract
extremal technique due to Bishop [4].

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 259
It is worth noting that for/as above Cl (/; 0) = {|z| ^ 1}. Indeed, it is clear that
ClB(/;0)c{|z| = l}; and by 2.5 d(Cl (/; 0))<=C\B(f;0). Since / is bounded and
Uf) e Cl (/; °)\C1bif; 0) it followsthat Cl (f; 0)={|z| ^ 1}.
6. A new algebra. Several years ago, Allen Devinatz asked for a characteriza-
tion of the (uniform) closure in L°° of the set of bounded measurable functions on
the circle only finitely many of whose negative Fourier coefficients fail to vanish
(Bull. Amer. Math. Soc. 71 (1965),p. 855). Recently,Donald Sarason [25], [16]
answered this question in the following way.
Theorem 6.1. [H'°, z] = H™+ C.
Here, [//"", z] denotes the uniformly closed subalgebra of L'a{dd) generated by z
and the elements of //°°, and C=C(y0) is the algebra of all continuous functions
on the unit circle. The assertion of the theorem is that each element of [H'°, z]
can be written as the sum of an H00 function and a continuous function. This
representation is, of course, far from unique since H°° n C=A is large. Not the
least remarkable feature of 6.1 is the (implicit) fact that H°° + C is an algebra, a
result which is by no means obvious.
We are interested in extending 6.1 and its corollary 6.2 to our more general
domains. Since part of the purpose of doing this is to indicate how the infinitely
connected case differs from that of the disc, it will be instructive to include a proof
of 6.1. Our proof is somewhat different from Sarason's and may be of independent
interest.
Proof of 6.1. By the Weierstrass theorem, H™ + C<=[Hx, z\. On the other hand,
[Hœ,z] obviously coincides with the uniformly closed algebra generated by H°°
and C; hence, it is enough for our purposes to show that H™ + C is a closed sub-
algebra of L". Let P be the subset of C consisting of trigonometric polynomials.
It is easy to see that Hw +P is an algebra, and that that algebra is dense in H °°+ C.
Thus, to conclude that H°° + C is a closed algebra we need only show that H" +C
is uniformly closed.
We have
/ 7T IT'1
C->Lm->L"'IH"'->L'°,
where / is the canonical injection and n is the quotient map. Clearly, w-1(7r(i'(C)))
= #°° + C; thus H™+ C is closed if and only if tt(í(C)) is. Let P: C -> C\A be the
natural projection. It is clear that C\A and tt(í(C)) are isomorphic as linear spaces
under the correspondence <t> : p(f) -*■-n{i{f)). We shall prove that O is a topological
isomorphism; from this, it is immediate that tt{í{C)) is closed, so that H^ + C is
closed, as required.
Let feC and suppose /(£) = e(£) - /z(£), t,ey, where eeZ,00 and heH™. For
keV° denote by k the harmonic extension of k to A:

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 260 LAWRENCE ZALCMAN [October

and set kr{Q = k(rt). Since/is continuous, /^/uniformly on y0 asr^-1, i.e.,

(e —«)r = (er —nr) ->(£ —«) uniformly as r -> 1. Thus

||«r-n|U ^ ||er-e||oo + 8(r) ^ 3\\e\\«»

where S(r) -*■0 as /• -*■1 and r is chosen so large that 8(r) á ||e|| „. Since «r e /I, we
have
(1) inf ||/+g|| ^ ||(s-A)+Är||0Dú ||«IU+ ||A-A,||. Ú 4|HL.
se.4

Since the left-hand side of (1) is the norm of the coset p(f) in C/A, it follows that
IkOC/n))! ^0 implies |p(/n)|| —>0. Since the reverse implication is trivial, <1>is a
topological isomorphism, and we are done.
It is easy to see that //OT+ C has as its maximal ideal space A/(A)\A and that its
Shilov boundary is the maximal ideal space W oíLx(dd) (cf. [17, p. 193]). Thus, we
can regard the algebra of Gelfand transforms (H°° + C)~ as a closed subalgebra of
C(W). Let Wa (|a| = 1) be the fiber of W over a, i.e., the set of all complex homo-
morphisms <j>eW such that <f>(z) = a. Then we have

Proposition 6.2. The algebra (H^ + C)" consists of all functions in C{W) such
that for each a e y0 there exists gae H™ satisfying ga=f on Wa.

Proof. This follows from Theorem 1.1 of [14] and the observation that the
fibers Wa are unions of maximal sets of antisymmetry for (H°° + C)~.
It is worth noting that 6.2 gives an "abstract" answer to the question posed by
Devinatz which is independent of 6.1. For the proof, we need only replace
(Hw + Cy (wherever it appears) by [Z/00,¿T\
Now suppose R is a finitely connected region whose boundary 8R consists of
finitely many disjoint simple closed curves. It is easy to extend 6.1 (and 6.2) to
this case. For simplicity, we may formulate our assertion as [B(R), C(8R)] =
B(R) + C(8R). Here all functions are understood to be defined on 8R, and
[B(R), C(8R)] is the smallest closed subalgebra of Lœ(8R; ds) which contains both
B(R) and C(8R). However, we shall not insist on the details of the proof since
another proof of this result will appear (implicitly) below.
When R is a domain of infinite connectivity the techniques that make possible
the reduction of the finitely connected case to the situation on the disc are no
longer available, and one must proceed differently. Below, we shall prove the
analogue of 6.1 for domains D of type (L). In the process, we shall obtain a proof
for the finitely connected case that differs somewhat from that already given. We
shall continue, with the obvious modifications, the notation established above.
First of all, we need the following generalization of the classical F. and M. Riesz
theorem [17, p. 46].
Proposition 6.3. Let ¡x be a finite complex Baire measure on 3D and suppose
J"r/(£) d/x(£)= 0 for eachfeA(D). Then d¡j.= k ds + c80, where keL\ds), c is a
complex constant, and 80 is the point mass at 0.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 261
Proof. Fix n, and let pn be the restriction of p to Tn, pn=p—pn. By the balayage
formula (5) of 3.2 we have

for each g e A(Dn). Let

J_ CdfiM
e*W 2m) t-i
Then
o = £ g(t) dp(i) = jg dpn+jg dpn

- •>r„
f g(X)Wn{Q+Fn(jÇ)di]=
f «(0^,(0.
Jr„

Here dvn=dpn + Fn d{. Now v„ is obviously a measure on Yn which annihilates

A(Dn); hence, by Rudin's extension of the Riesz theorem [22] to finitely connected
domains, vn is absolutely continuous with respect to arc length on Tn. Since
Fn d£«ds by definition, it follows that dpn«ds. Let n -> go to conclude that dp«ds
on each boundary curve yn. Done.
Next, we record the following elementary fact, suggested by the referee.

Lemma 6.4. Supposefe B(D). Then

sup{|z|:zGCl(/;£),£eyn}g ¡f\Ja.

Proof. This is essentially a consequence of the Poisson integral formula. Let J

be a proper open subarc of yn. Pick a simply connected region U in D whose
boundary is a simple closed curve C such that /<= c and C\J<^ D. Let Abe a Riemann
map from A onto U; h extends to a homeomorphism from A onto Ü. Clearly
/oAg5(A)=í/°°, soif zeU

(1) /(z) =f° Kh-\z)) = f /o A(£)/>(£,

A-H*))rfA(£),
where P is the Poisson kernel, A is normalized Lebesgue measure, and/o/i(£)
denotes nontangential boundary values.
Because of the reflection principle, the nontangential limits of/at £0 e /are the
same as those of/° h at A-1(£o)- Also, a subset of/ has measure 0 if and only if
its preimage under h has measure zero, so

l/-*L-»»|.-l/li|..
Now let £o g /, {zfc}c D, zk -> £0. We may assume {zjc £/. Then {A_1(zfc)}cA and
A"1(zfc)-^A"1(£o). But examination of the Poisson kernel shows that

(2) lim sup I f / o/*(£)/>(£,

A- \zk)) d\(Q ^ ll/°%-t/)ll-= ll/|7||-
k-»<*> |J|Î| =1

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 262 LAWRENCE ZALCMAN [October

The conjunction of (1) and (2) yields

limsup|/(zfc)|
fc-»oo
â I/ML á I/ÜL,
as required.
For convenience, we shall write B = B(D), A = A(D), C= C(8D), L°°=L'°(dD; ds),
L1=L1(8D; ds). All algebras will be viewed on the boundary.
Theorem 6.5. [B, C] = B+C.
Proof. It is enough to show that B+C is a closed subalgebra of L°°. Fix z0 e D
and denote by Q the set of all rational functions holomorphic on D\{z0}. By the
Hartogs-Rosenthal theorem, the restriction of Q to T is uniformly dense in C.
Thus, B+ Q is dense in B+C. It is also easy to see that B+ Q is an algebra. Indeed,
let q e Q, « e B. We have q(z) = (z-z0)'nqi(z), where q1e B n Q, and h(z) = h1(z)
+ (z—z0)"«2(z), where h1 is a polynomial of degree at most «—1 and h2eB.
Hence, qh=q1h2 + (z~z0)'nq1h1 e B+ Q, as required. It remains only to show that
B+C is uniformly closed.
Let tt: C^ CIA be the quotient map and ¿: C¡A ->(C//4)** the natural (iso-
metric) injection. Since (C/A)* = A1, we have (CIA)** = (AL)*. Let I: (A1)* ->
(A1 n L1)* be the restriction map. Then I: (C/A)** -> LM/(^X n L1)1=La>//i. Here
we have used the fact that B=(AL n L1)1, i.e. 5 is the weak * closure of A in Va
(3.1). Consider now the sequence

C-> ■n C¡A-> i (C¡A)**->I Q'1

LXIB^—> Lm,

where Q: L™-^ L^/B is the quotient map. Obviously, the image of C under this
sequence of maps is (canonically isomorphic to) B+C. It is enough, therefore, to
show that I o i o tt(C) is closed. In other words, we must show that the norms on
tt(C) induced by (C/A)** and Lx/B are equivalent.
It will be convenient to reformulate this condition in function theoretic terms.
By 6.3, each element n of A1 has the form ¡x= k ds + c80, \\n\\=j \k\ ds+\c\. Set,
for/e C,
ll/lli = sup{!//</,*| : ?eA\ H £ 1}= inf
A
\\f+g\\„,
||/||2 = sup{|J/^|:/,£^nLM|/i|| ^ 1} . inf |/+*|..

We claim that ||/||i<A||/||2 for some universal constant A>0. Actually, it is enough
for our purposes to show that, for any sequence {/n}cC, ||/„||2^0 implies
ll/nlli ~* 0- Of course, if A1 = A1n L1, i.e., if each measure in A1 is absolutely
continuous, there is nothing to prove. This situation will occur when and only
when 0 is a peak point for A. Hence, in the sequel, we may assume that 0 is not a
peak point for A.
Now suppose {/„}<=C and ||/n||2 -*■0. Without loss of generality we may assume
/n(0)=0 for all «. Since ||/B||2 = infB |/„ + «||«, we have/„ = £„-«„, where eneLx,

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 263
hn e B, ||e„|„o -> 0. Let p=k ds+cS0 be in the unit ball of A1 and suppose c^O.
Then
| JVndp| = | jfjc ds+ c/n(0)
I = jfnk ds
\ r I Ir !
¿ \ \ enk ds \ + \ \ hnk ds\.

The first term on the right-hand side is bounded by ||eB||«>J" \k\ ds< ||e„|oo -*■0.
To estimate the second term, write

f hnk ds = íhn(k + cK) ds-cihnKds,

where Kds = (2m)~1C~1 dl. Now k + cKe A1 n L1, so that the first integral above
vanishes by 3.1. Hence

[hnkds ^ \c\ \hnKds è \Mhn)\.

Since/n(0)=0, it is clear that hn is small (on V) near 0. It follows from Iversen's

theorem (2.5) and 6.4 that Cl (hn; 0) can be made to lie in any disc about the
origin, however small, if only n is chosen large enough. Since <t>0(hn) e Cl {hn; 0)
(4.5), we are done. The proof is complete.
It should be pointed out that, in case 0 is a peak point for A, our proof shows
that
(*) mf|/+*|L=inf|/+A|.
A B

for any/e C. Actually, as Professor T. W. Gamelin pointed out, (*) remains true
even if 0 fails to be a peak point. For in this case, set dm = ds + o0 and let H be the
weak * closure of A in Lœ(dm). One can show without difficulty that

inf\\f+g\\D = inf\\f+h\\L«<dm), feC.

Thus, it remains to prove that infH ||/+A| =infB ||/+A||, and this can be done by
using Iversen's theorem. Details are left to the reader. It is also worth noting that
(*) holds true for A=A(R), B=B(R), where R is a finitely connected domain with
smooth boundary.
Finally, let us note that the obvious analogue of 6.2 trivially holds true for our
algebra [B, Cf.

7. Comments; generalizations. This section is devoted to a discussion of various

extensions and generalizations of the preceding results. For convenience, the
comments are grouped together by section.
§2. A quick glance at the proof shows that 2.1 is valid for any domain R whose
(rectifiable) boundary consists of a union of disjoint simple closed analytic curves

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 264 LAWRENCE ZALCMAN [October

plus a set of (arc length) measure 0. Actually, the requirement of analyticity can be
weakened considerably: it is enough, for instance, that the boundary be con-
tinuously differentiable. For then, as before, / will have a (unique) asymptotic
value/a(£) at almost every point £ of dR; then by Lindelöf's theorem/(z)->/a(£)
as z -> £ nontangentially from within R. Of course, if the boundary is merely
Jordan,/will still have a unique asymptotic value at a set of points on the curve
that corresponds to a set of full measure on the unit circle. Quite plainly, the
existence of boundary values is a local phenomenon; the global structure of R
comes into play in the requirement that the set of accumulation points of the
"nice" boundary curves have zero arc length.
Generalizations of 2.2, 2.3, and 2.4 have already been discussed in §2; accord-
ingly, we shall offer no further comment on these results.
Iversen's theorem (2.5) is one of the most beautiful and certainly one of the most
important theorems of cluster set theory; it deserves to be better known. General-
izations have proceeded in various directions, most notably by replacing C1B(/; a)
by even smaller sets. For instance, the theorem remains true if a set of capacity 0
is deleted from the boundary. A survey of what is known (stated for the case of the
disc) is in [28, p. 5]. Classical cluster set theory has limited its attention almost
entirely to (meromorphic) functions defined on the disc; it seems high time that
more general (infinitely connected) domains were studied. Part of the reason for
including the results of §5 is to show that interesting new things can happen in the
case of infinite connectivity.
§3. Theorems 3.1 and 3.6 are valid for a much larger class of domains than those
of type (L). For instance, the linear disposition of the holes was completely
irrelevant: nothing is altered if D is replaced by a domain R whose boundary
consists of countably many simple closed analytic curves of finite total length
which accumulate at a single degenerate boundary component (a point). More
generally, the boundary curves of R may be allowed to accumulate at a finite
number of degenerate boundary components. Indeed, if A g B(R), one can use the
Cauchy integral formula to write A= 2" =i hu, where each hk is analytic in a domain
Rk=>R whose boundary accumulates at a single point. We can approximate each
hk pointwise boundedly by functions in A(Rk), and the sum of these functions
approximates h. It seems likely that 3.1 and 3.6 remain true if the set of accumula-
tion points of the boundary has Hausdorff linear measure zero or zero analytic
capacity; however, we have not pursued this line of thought to any extent. In any
case, most of the argument of 3.6 remains valid if the boundary is allowed to
accumulate at countably many degenerate boundary components.
We should point out that there are very definite limitations to the generality of
3.1 and 3.6. For instance, the example of 2.3 shows that neither theorem can hold
in this case. Indeed, if k=x^n we have \rkgdl=0, g e A(R), but §rkfdC^0.
Similarly, it is easy to see (by perturbing the curves t„ to the interior of R) that/
cannot be approximated pointwise boundedly by functions in A(R).

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 265
Since this work was done, T. W. Gamelin has developed a new approach to
these problems and has succeeded in extending 3.1 and 3.6 to a class of domains
much larger than that considered above. His idea is to apply the constructive
techniques of Vitushkin to study the possibility of pointwise bounded approxima-
tion (on R) of functions in B(R) by rational functions. These methods do not
require the hypothesis of rectifiability on dR; however, the nature of the boundary
remains central in determining the possibility of approximation.
§4. Questions involving fibers are intimately connected with the boundary
behavior of analytic functions. Thus, it is not true that the conformai equivalence
of two domains implies a topological equivalence between the fibers lying over the
boundaries. A simple example may clarify the situation.
Suppose R is a simply connected domain and <j>:/?->■ A maps R conformally
onto A. Using classical examples from the theory of prime ends, it is easy to
construct an R satisfying
(1) There exists an arc y^8R such that Cl (</>, £) = <*for any £ ey, where a is
some fixed point of the unit circle.
(2) There exists a point £0 e 8R such that Cl (<f>,£0)={aj}, where {aj is a finite,
countable, or even uncountable set of points on the circle.
If/e B(R), then 2/=/° j>~x e H°°, and this correspondence is an isomorphism
between these two algebras. Then for any / e B(R) we have (with the obvious
notation)

(1') /( U Mi) = (TfT(Ma)

(2') /(MCo)= Ui (TfT(Mai).

We restrict the rest of our comments to 4.4. This result is obviously local : it
depends only on the nature of the boundary near the point in question. To be
precise, suppose £ e 8R is a boundary component and one can find a sequence of
(rectifiable) curves Ck in R surrounding £ and tending to it. Assume further that
the Cauchy integral formula (2.2) is valid in the subregion of R cut off by Cfc;
this necessitates, in particular, that the portion of 8R lying inside Ck be rectifiable.
Then clearly the argument of 4.4 applies. Actually, it is not difficult to see that the
rectifiability requirement can be avoided. Indeed, suppose that R is a domain such
that 0 e 8R is a component of 8R and there exists an exhaustion of R by subdomains
Rn, each of which has rectifiable boundary, such that 0 is a component of 8Rn.
For a fixed/e B(R) the argument of 4.4 gives functions gn, hn e B(Rn) such that
/gn+z«„=l on Rn, and these functions are uniformly bounded. It follows that
there exist g, heB(R) such that fg+zh = 1 on R, as required. (We have assumed
that the representation of 2.2 holds on a suitable subregion of each Rn.)
§5. For further information on peak points for algebras of analytic functions
see [29] and the references cited there.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 266 LAWRENCE ZALCMAN [October

A few comments on Theorem 5.2 are in order. For certain domains, the asymp-
totic value <f>0(f)may exist for approach even more general than that allowed by the
theorem. For instance, Rudin [23] has given an example of a domain D of type (L)
such that each/G B(D) has an asymptotic value along any curve which terminates
at 0 and misses a suitable annular neighborhood of each boundary curve yk. It is
easy to see in this case that each function in B(D) has a unique asymptotic value
at 0. Of course, generally speaking, Cl (/; 0) can be large; indeed, as we have
noticed in §5, there exist functions /g B(D) such that ||/||«> = 1 and Cl (/; 0)
= {|z|^l}. More generally, Rudin [23] has shown that if £ g 8R is not an AB-
removable point for R then there exists an/g B(R) as above; cf. [3].
The metric condition which insures the existence of limÄ_0- A(x), h e B(D), can
be strengthened to guarantee the existence of lim^o- h(lc)(x), k = 0, 1, 2,..., n.
In fact, one can give conditions for every derivative to extend continuously to
{z : |z| < 1, Re z^O}. However, since the derivative of a bounded analytic function
is, in general, badly unbounded (on the disc, an Hm function may have a derivative
of unbounded characteristic), we shall not concern ourselves with this extension.
It should be noted, nonetheless, that Rudin has shown [24] that for any compact
set K on {Re z = 0} of (linear) measure 0 there exists a domain R, whose boundary
consists of K and of circles in the open right half plane which accumulate to K,
such that for each/G B(R) and each &= 0, 1,2,.. .,fm(z) can be extended to be
continuous on the closed left half plane.
See also the comments on §6.
§6. As we remarked earlier, Sarason's proof of 6.1 was somewhat different from
the one we have presented. In particular, his argument relies on the F. and M. Riesz
theorem and is closely related to the technique of 6.5. It seems rather remarkable
that 6.1 can be obtained so simply as a consequence of the geometry of the disc.
The result of 6.3 is really quite general. If R is an arbitrary plane region and if
the rectifiable simple closed curve y is a component of 8R which lies at a positive
distance from 8R\y, then the argument of 7.1 shows that each p e A(R)1 is absolutely
continuous with respect to arc length on y. It follows that if the (closed) boundary
curves of R accumulate at a countable set of degenerate boundary components
{£i, lz, ■■•} and p _|_A(R) then p = Fds+J, cn8n, where 2 k„| <°° and S„ is the
point mass at £„.
Similarly, it is a general fact for completely arbitrary R that B+Q is a sub-
algebra. It is also true that B+Q is dense in B+ C. For each point of 8R is in the
closure of R, so that each point of 8R is a peak point for the uniform closure on 8R
of rational functions analytic on 8R [29, p. 52]. By Bishop's theorem [29, p. 123],
gis dense in C=C(8R).
The proof of 6.5 depends strongly on the fact that B is the weak * closure of A,
and this fact is available to us only if 8R clusters at finitely many point components
of the boundary. However, if we assume the weak * density of A in B, the final part
of the proof remains valid even if the curves of 8R are allowed to cluster at a

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 267
countable number of point components of 8R. The crucial observation here is that
if £o is a boundary component which is not a peak point then there exists a positive
function K e L\ds) such that J Kds= 1 and J Kg ds=g((,0) for each g e A. Indeed,
by a theorem of Bishop [29, p. 8] there exists on 8R a probability measure v,
whose mass is carried entirely on the peak points for A, such that ¡ g dv=g(t,0),
g e A. Clearly, v contains no point masses. Now (£-£0) dv J_ A, so by 6.3 we have
(£ —£0) dv«ds. Thus dv«ds, i.e., dv = Kds, as required. Since A is (assumed to be)
weak * dense in B, integration against K extends to a complex homomorphism (in
Mlo) of B. The remainder of the proof of 6.5 can be carried through along the lines
of the proof in §6. We omit the details.

8. Bibliography. Relatively little has been written about bounded analytic

functions on general (infinitely connected) domains. There is, of course, the basic
paper of Havinson [15], which is concerned, in part, with extremal problems for
bounded analytic functions. Quite recently, Fisher [10] simplified some of Havin-
son's results. While the approaches of both these authors are rather general, the
problems they consider do not, for the most part, bear directly on the questions we
have studied above. One notable exception is the theorem of Fisher which says,
in particular, that each function in the unit ball of B(D) can be approximated
pointwise boundedly on D by inner functions in B(D)\ this extends a classical
result of Carathéodory. In a somewhat different direction, Rubel and Shields [21]
have studied spaces of bounded analytic functions under topologies different from
that induced by the uniform norm.
We have mentioned Rudin's papers [23], [24] and Beck's note [3] above; to these
we should add the paper of Gamelin and Rossi [12] (or, better, [13]), the primary
concern of which is the study of harmonic functions on domains of infinite con-
nectivity. In fact, certain examples in [13] actually provided the inspiration for the
results of §5; only later did we become acquainted with [23] and [24]. It is Fisher's
work [8], [9], however, that seems most closely related to our results. In particular,
[8] contains a proof that A(D) is pointwise boundedly dense in B(D) in case D
satisfies certain (rather restrictive) metric criteria. Part of the motivation for [8]
came from Voichick's paper [27], which characterized the extreme points of B(R)
for a special class of infinitely connected domains (Blaschke domains).
Finally, let us note that Gamelin's approach to the results of §3 (cf. §7), utilizing
the techniques of Vitushkin, will appear in extenso in [11].

9. Additional remarks. Since this paper was written, new information on

bounded analytic functions on infinitely connected domains has become available.
For results on pointwise bounded approximation of bounded analytic functions
see [30].
The obvious analogue of Proposition 4.4 is true for an arbitrary plane domain.
This observation is due to T. W. Gamelin. The simple proof is an amalgam of the
argument of 4.4 with the Vitushkin technique exhibited in 2.4.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 268 LAWRENCE ZALCMAN [October

In a recent U.C.L.A. thesis, Michael Behrens has settled special cases of several
of the problems left open here. In particular, he has proved the corona conjecture
for domains of type (L) which satisfy "Zrnlxn< oo and xn+1/xn^c< 1 and has
succeeded in describing the intersection of M0 with the part containing D in this
and other cases. His techniques, which are related to the nonstandard analysis of
Robinson [31], consist of an ingenious reduction of problems on D to known
results on the disc.

Added in proof. Behrens' work will appear in [32]. Gamelin [33] has recently shown
that the corona problem on an arbitrary domain is local; this enables him to exhibit
a large class of infinitely connected domains for which the corona conjecture is true.
The results of §5 have just been generalized to arbitrary domains in a beautiful paper
by Gamelin and Garnett [34].

References
1. P. R. Ahern and Donald Sarason, On some hypo-Dirichlet algebras of analytic functions,
Amer. J. Math. 89 (1967), 932-941.
2. Richard Arens, The maximal ideals of certain function algebras, Pacific J. Math. 8 (1958),
641-648.
3. Anatole Beck, A theorem on maximum modulus, Proc. Amer. Math. Soc. 15 (1964),
345-349.
4. E. A. Bishop, Abstract dual extremal problems, Notices Amer. Math. Soc. 12 (1965), 123.
5. Lennart Carleson, Interpolations by bounded analytic functions and the corona problem,
Ann. of Math. (2) 76 (1962), 547-559.
6. E. F. Collingwood and A. J. Lohwater, The theory of cluster sets, Cambridge Univ.
Press, Cambridge, 1966.
7. Nelson Dunford and Jacob T. Schwartz, Linear operators. Part I, Interscience, New York,
1966.
8. Stephen D. Fisher, Bounded approximation by rational functions, Pacific J. Math. 28 (1969),
319-326.
9. -, Rational functions, H", and H" on infinitely connected domains, Illinois J. Math.
12(1968), 513-523.
10. -, On Schwarz's lemma and inner functions, Trans. Amer. Math. Soc. 138 (1969),
229-240.
11. T. W. Gamelin, Uniform algebras, Prentice-Hall, Englewood Cliffs, N. J., 1969.
12. T. W. Gamen'n and Hugo Rossi, "Jensen measures and algebras of analytic functions",
in Function algebras, edited by F. Birtel, Scott, Foresman and Co., Chicago, 111., 1966, pp.
15-35.
13. -, Jensen measures and algebras of analytic functions, preprint version of [12].
14. Irving Glicksberg, Measures orthogonal to algebras and sets of antisymmetry, Trans. Amer.
Math. Soc. 105 (1962), 415-435.
15. S. Ja. Havinson, Analytic capacity of sets, joint nontriviality of various classes of analytic
functions, and the Schwarz lemma in arbitrary domains, Amer. Math. Soc. Transi. (2) 43 (1964),
215-266.
16. Henry Helson and Donald Sarason, Past and future, Math. Scand. 21 (1967), 5-16.
17. Kenneth Hoffman, Banach spaces of analytic functions, Prentice-Hall, Englewood Cliffs,
N. J., 1962.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 1969] BOUNDED ANALYTIC FUNCTIONS 269
18. Kenneth Hoffman, Bounded analytic functions and Gleason parts, Ann. of Math. (2) 86
(1967), 74-111.
19. Kenneth Hoffman and Hugo Rossi, Extensions of positive weak *'-continuous functional,
Duke Math. J. 34 (1967), 453-466.
20. Charles E. Rickart, General theory of Banach algebras, Van Nostrand, Princeton, N. J.,
1960.
21. L. A. Rubel and A. L. Shields, The space of bounded analytic functions on a region,
Ann. Inst. Fourier (Grenoble) 16 (1966), 235-277.
22. Walter Rudin, Analytic functions of class Hp, Trans. Amer. Math. Soc. 78 (1955),
46-66.
23. -, Some theorems on bounded analytic functions, Trans. Amer. Math. Soc. 78
(1955), 333-342.
24. -, Essential boundary points, Bull. Amer. Math. Soc. 70 (1964), 321-324.
25. Donald Sarason, Generalized interpolation in H", Trans. Amer. Math. Soc. 127 (1967),
179-203.
26. M. Tsuji, Potential theory in modern function theory, Maruzen, Tokyo, 1959.
27. Michael Voichick, Extreme points of bounded analytic functions on infinitely connected
regions, Proc. Amer. Math. Soc. 17 (1966), 1366-1369.
28. Max L. Weiss, Cluster sets of bounded analytic functions from a Banach algebraic viewpoint,
Ann. Acad. Sei. Fenn. Ser. A I 367 (1965).
29. Lawrence Zalcman, Analytic capacity and rational approximation, Springer-Verlag,
Berlin, 1968.
30. T. W. Gamelin and John Garnett, Constructive techniques in rational approximation, Trans.
Amer. Math. Soc. (to appear).
31. Abraham Robinson, Non-standard analysis, North-Holland, Amsterdam, 1966.
32. Michael Behrens, The corona conjecture for a class of infinitely connected domains, Bull. Amer.
Math. Soc. (to appear).
33. T. W. Gamelin, Localization of the corona problem, Pacific J. Math, (to appear).
34. T. W. Gamelin and John Garnett, Distinguished homomorphisms and fiber algebras, Amer.
J. Math, (to appear).

Massachusetts Institute of Technology,

Cambridge, Massachusetts
Stanford University,
Stanford, California

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use