PACIFIC JOURNAL OF MATHEMATICS

Vol. 42, No. 3, 1972

ALGEBRAS OF ANALYTIC FUNCTIONS IN THE PLANE

WILLIAM R. ZAME

Let X be a compact subset of the complex plane and

let A be an algebra of functions analytic near X which
contains the polynomials and is complete in its natural topology.
This paper is concerned with determining the spectrum of A
and describing A in terms of its spectrum. It is shown that
the spectrum of A is, formed from the disjoint union of certain
compact subsets of C (suitably topologized) by making certain
identifications. A is closed under differentiation exactly when
no identifications need be performed, and then A admits a
simple, complete description. In particular, if X is connected,
then the completion of A is merely the restriction to X of
the algebra of all functions analytic near the union of X
with some of the bounded components of C — X.

Our principal tool in these investigations is the theory of analytic

structure in the spectrum of a function algebra developed by Bishop
in [2] and extended by Bjork in [4, 5]. We view the algebra A as
the inductive limit of function algebras and induce analytic structure
in the spectrum of A. When A is closed under differentiation, topol-
ogical considerations lead quickly to the desired results. In the general
case, we pass to the smallest algebra B containing A which is closed
under differentiation. By introducing differentiation in the spectrum
of A, we show that every continuous complex-valued homomorphism
of A may be extended to B. It follows that the spectrum of A is
obtained from the spectrum of B by making certain identifications.
When no identifications need be performed, A — B.

2. Preliminaries* If U is an open set, we let ^(U) denote the

algebra of functions analytic on U, endowed with the topology of
uniform convergence on compact sets. If V is an open subset of U,
we let ruv: ^(U) -+ ^(V) be the restriction. If X is a compact set,
έ?(X) denotes the algebra of functions on X which have analytic
extensions to a neighborhood of X. We view ^(X) as the inductive
limit (in the sense of functions) of the system {<^(U);ruv} and equip
^(X) with the inductive limit topology; i.e., the finest topology
rendering the restriction maps r^: ^(U) —* ^(X) continuous.
If A is a subalgebra of ^{X) and U is an open set containing
X, we let A{U) = {/ e <?{X): f\XeA). Similarly, if K is a compact
set containing X, we let A(K) = {/ e έ?(K): f\Xe A}. For compact
sets K, L with Kz) L, we let rKL: έ? —> έ?(L) be the restriction. Then
it is easy to see that:
811
 812 WILLIAM R. ZAME

A = inductive limit {A(U)',ruv}

= inductive limit {A(K); rKL)
and that the inductive limit topologies thus induced on A coincide
with the relative topology from g?(X). A is complete in this topology
if and only if A(U) is closed in έ?(U) for each open set U containing
X. Details of the above may be found in §2 of [11].
We also regard a subalgebra A of έ?(X) as a normed algebra
with the norm:
11/liar = sup{|/(a;) I: a? e X } .
We relate these two topologies in the following proposition.

PROPOSITION. Let A be a subalgebra of 0*{X) containing the

constants. Then the norm topology and the inductive limit topology
on A admit the same continuous complex-valued homomorphisms.

Proof. If this were not so, there would be a homomorphism φ

of Ay continuous relative to the inductive limit topology, and a func-
tion / in A such that

We could then find an open set U containing X and a function F in

A(U) such that F\X = f and \F\ < 1 on U. Then (1 - F) would
be invertible in the closure of A(U) in £?(U). Moreover, φ rυ would
be a continuous homomorphism of A{U)9 and would thus extend to
its closure. Since φ r^(l — F) = ^(1 — /) = 0, we would then have
the following contradictory chain of equalities:
1 = 0. rv[(l - F)(l - F)"1] = {φ rΌ(l - F)}{φ *v[(l - F)"1]} - 0 .
This contradiction establishes the proposition.

If B is a topological algebra, we denote the spectrum of B (the

space of nonzero continuous complex-valued homomorphisms, with the
weak* topology) by MB. We may regard an element 6 of B as a
function of MB via the Gelfand transform b(φ) = φ{b), for each φ in
MB. If B is a normed algebra with identity, then MB is compact.
Then if A is a subalgebra of έ?{X) containing the constants, MA is
compact and a standard argument may be used to show that (see [11]):
MA = protective limit {MMU);r*v}
= projective limit {MMK);r*KL}
where r^v and r%L are the adjoints of the restrictions.
 ALGEBRAS OF ANALYTIC FUNCTIONS IN THE PLANE 813

We refer to [6] for standard material concerning function algebras.

If B is a function algebra with spectrum MB we denote its Silov
boundary by SB. We make use of the techniques developed by Bishop
and Bjork in [2, 4, 5] and assume some familiarity with these papers.
In particular, if / is an element of B we say that a component W of
C — f(SB) is /-regular of multiplicity n if for each w in W there are
at most n homomorphisms ζ in MB for which /(ζ) = w; and that for
some w there are exactly n such homomorphisms. In that case, there
is a discrete subset E of W such that for each ζ in MB such that
f(Qe(W— E), there is a neighborhood Q of ζ in MB mapped home-
1
omorphically by / onto a disk, and such that g^iflQ)' is analytic
for each g in B. The neighborhood Q is called an analytic disk about
ζ, relative to the function /.
We conclude this section with a topological lemma.

LEMMA. Let M be a compact connected real 2-manifold with

boundary and let p be a continuous map of M into the 2-sphere S2.
If p is locally one-to-one and is one-to-one on the boundary of M, then
p is one-to-one.

Proof. We will reduce to the case of a 2-manifold without

boundary. To this end, suppose that M has k boundary components
Ji, •••, e/fc Each Ji is a 1-sphere, so that p(Ji) is a 1-sphere in S2
for each i. Hence S2 — p(Ji) consists of two disjoint connected open
sets. A compactness argument, using the fact that p is locally-one-
to-one, may be used to show that there is a connected neighborhood
of Ji in M on which p is one-to-one. It follows that we may choose
a neighborhood Wi of J^ such that p is one-to-one on Wi and p(Wi)
does not intersect one of the components of S2 — p(Ji). It is easy to
see that we may attach a disk to M along Jt and extend p to this
disk; since p(Wi) lies in only one component of >S2 — p(Ji) this may
be effected in such a way that the extension remains locally one-to-
one. If we perform this surgery for each boundary component J{ we
arrive at a compact connected real 2-manifold N without boundary
and a continuous map q of N into S2 which is locally one-to-one.
If q is one-to-one then p must certainly be.
For each x in S2, the fiber q~ι{%) is compact and discrete (since
q is locally one-to-one) and hence finite. Then, using the invariance
of domain, we may choose an open set U about x such that q~ι{U)
consists of open, connected components, each mapped homeomorphically
onto U by q; thus q is a covering map. Since S2 is its own universal
covering space, it follows that g, and hence p, must be one-to-one, as
desired.
814 WILLIAM R. ZAME

3* Main results* If A is a subalgebra of &(X) that contains

the constants and the coordinate function Z, we say that A is stable
if it is complete in the inductive limit topology and each of the
algebras A(U) is closed under differentiation.
In order to see how stable algebras may arise, consider the fol-
lowing construction. Let X be a compact subset of C and let {Xa}
be a partitioning of X into disjoint closed sets. For each a let Ya
be the union of Xa with some of the bounded components of C — Xa.
Then let
A = {fe^(X):f\Xae^(Ya)\Xa for each a}.
It is easy to see that A is a stable algebra and that the spectrum
of A is the disjoint union of the Ya9 suitably topologized. The fol-
lowing theorem shows that this is the only way in which stable
algebras may arise.

THEOREM 1. Let A be a stable subalgebra of ^(X) and let Y'a be

a component of MA. Then Z\Y'a is a homeomorphism. The set Ya —
Z{ Yά) is the union of Xa = X Π Ya with some of the bounded com-
ponents of C — Xa. Finally, the collection {Xa: Yfa is a component of
MA) is a partitioning of X into disjoint closed sets and A = {f e έ?(X):
f\Xae^(Ya)\Xa for each component Yra of MA}.

Proof. Let if be a compact set whose interior contains X and

whose boundary is the disjoint union of a finite number of smooth,
simple closed curves. Let A(K)* denote the completion of the algebra
A{K) in the norm |J (I*. We proceed by examining the algebra A{K)*
and its spectrum and then passing to the projective limit.
We identify K with a subset of MMK)*. Clearly, SA(K)* is contained
in the boundary (relative to C) of K. Let Λ denote the set of points
in MMK)* having a neighborhood which is an analytic disk (relative
to the function Z). We show that MA{K)* — SAiK)* — A is at most
countable. First, a standard argument shows that the unbounded
component of C — Z(SA{K)*) is ^-regular of multiplicity 0. If T is the
boundary of this component, then it follows from [5] that there are
no points ζ of MMK)* — SAίK)* for which Z(ζ) belongs to T. We con-
clude from [5] that each component of C — Z(SA{K)*) that adjoins the
unbounded component is ^-regular of multiplicity at most 1. Similarly,
if T' denotes the boundary of one of these components, then there
is at most one point ζ in MA{K)* — SA{K)* for which Z(ξ) e T'. Then
each component of C — Z(SA(K)*) that adjoins one of these components
is Z-regular of multiplicity at most 2. Proceeding inward in this
way, we see that each component of C — Z(SA{K)*) is ^-regular of some
multiplicity. Again from [5], it follows that there is a discrete subset
 ALGEBRAS OF ANALYTIC FUNCTIONS IN THE PLANE

E of C - Z(SMK)*) such that each ξ in MA(K)* - SA(K)* for which

does not lie in E\J Z{SMK)*), is a point of Λ. Moreover, if x is in E,
then there are only finitely many homomorphisms ψ for which Z(ψ) = α?.-
Now let us turn to the points ξ of Λf^*,* — SAiK)* for which Z(f) e
Z(SA(K)*). Since the boundary of K is the finite union of smooth curves,
it follows that each boundary point is a triangle point in the sense
of Bishop [2]. Hence for each ξ in MMK)* — SA(K)* for which Z(ξ) e
Z(SMK)*)9 there is a deleted neighborhood Wξ lying in Λ. From the
compactness of the boundary of K and the fact that for each point
y of the boundary there are only finitely many homomorphisms ζ for
which Z(ζ) = y, it follows that all but finitely many of the points ξ
of MMK). - SMK)* for which Z(ζ) e Z(SMK)*) actually lie in A. Then
MMK)* — SAίK)* — A. is a countable set, as was asserted.
Now let L be a connected component of MA{K)*. We assert that
Z\ L is a homeomorphism. If this were not so, we could find homomor-
phisms φ and λ in L such that Z(0) = Z{\). Since ild ( i α * = MMK)>
we could then find an open set U containing K and a function / in
A(U) such that ^(/|ίΓ) ^ X(f\K). Since A(ί7) is complete, closed
under differentiation and contains the polynomials, it follows from a
theorem of Bishop [3] that MMU) is a 1-dimensional complex analytic
manifold and that Z on MMU) is a local analytic isomorphism. Let
p: A(U) —• A{K) be the restriction and p*: MMK) —> MMU) be its adjoint.
Then p*(L) is a compact connected subset of MAiU). By the in variance
of domain theorem, ρ*(Λ) lies in the interior of ^*(L). Hence Z is
one-to-one on the boundary of p*(L). Since Z is a local homeomorphism
on ikf^(Z7), we may find a compact connected set U containing ρ*(L)
in its interior such that Z is one-to-one on the boundary of U and
U is a 2-manifold with boundary. Regarding C as a subset of S 2 ,
we may then apply the lemma to conclude that Z is one-to-one on U
and hence on p*(L). But φ and λ restrict to different homomorphisms
of A(U) so that p*(φ) Φ pHx), while Z[p*(φ)] = Z[p*(X)]9 which is a
contradiction. It must be therefore, that Z [ L is a homeomorphism.
From the Silov idempotent theorem, it follows that each component
of MMK)* contains a component of the boundary of K. It follows that
for each component L of MA(K)*, the boundary of Z(L) coincides with
Z(SΛ(K)* Π L), so that Z(L) is formed from K f] Z{L) by the addition
of certain components of C — K Γ\ Z(L).
Now let us return to MΛ. For a component Y'a of MA9 and a
compact set K with smooth boundary, containing X in its interior,
let rκ:A(K)*—+A be the restriction and let r%: MA—*> MMK)* be its
adjoint. Let K'a be the component of MA{K)* that contains rκ(Ya)
It is clear that

Y* = protective limit {K'a\ r%L} .

S16 WILLIAM R. ZAME

From the description of K'a derived above, it follows that Z\ Y?a is a

homeomorphism and that Z{Yra) is the union of I f ] Z(Y«) = Xa with
some of the bounded components of C — Xa.
If / belongs to A, then it is in A(K)* for some compact K with
smooth boundary containing X in its interior. Since Z is a homeomor-
phism on each component of MMK)*, it follows that A, the set of points
in MMK)* having neighborhoods which are analytic disks, is all of
MMK)* - SMK)*. Now we may see that f\(KΓ)Z(L')) belongs to
έ?{Z{L')) I (K n Z{L')) for each component U of MMK)*. It follows that
/1 (X Π Z( Y£) belongs to <?{Z{ Y'a)) \ (X Π Z( Y'a)) for each component Y'a
of MA.
Finally, suppose that U is an open set containing X and that /
is a function in έ?{U) such that f\(Xp[Z(Y'a)) belongs to

for each component Y'a of MA. For each such Y«, choose a compact set
Ka with smooth boundary containing X in its interior and such that
Z(L'a) c (U U Z(Y«)) where L'a is the component of MMK)* that contains
r%(Yfa). If Yβ is sufficiently close to Yc[, we may choose Kβ to be j8Γα.
Then the compactness of MA enables us to choose a single compact
set Kf with smooth boundary, containing X in its interior, and such
that Z{L") c (J7U Z(Y'a)) for each a, where L" is the component of
MA{κn* that contains r%,(Y'a). Without loss, we may assume that every
component of Kr contains a point of X. Then for each component
L' of MMKt). we see that /1 (Kf n Z(L')) belongs to έ?(Z(L')) \ (K' Π Z(L'))
The Silov idempotent theorem and the Arens-Calderon theorem then
imply that f\Kf belongs to A(K')*. Since A is complete and Kr
contains X in its interior, it follows that / 1 X belongs to A, which
completes the proof.
The above theorem gives a complete description of stable algebras.
In what follows, we use stable algebras to describe the structure of
more general subalgebras of έ?{X). We let Abe a complete subalgebra
of έ?{X) containing the polynomials and let Ao be the smallest stable
algebra containing A; Ao is the completion of the algebra generated
by the functions in A together with all their derivatives. We let
i: A —> Ao be the inclusion and i*: MAQ —> MA be its adjoint (the restric-
tion map).

THEOREM 2. The map i*: MAo-+MA is onto. If Y and Yr are

components of MAQ then ί* \ Y and i* \ Yf are one-to-one and there are
at most finitely many pairs (μ, v) in Y x Yr such that i*(μ) = ί*(v).
If / * is a homeomorphism, then A — Ao.
 ALGEBRAS OF ANALYTIC FUNCTIONS IN THE PLANE 817

Proof. We show first that i* is onto. Choose a compact set K

with smooth boundary, whose interior contains Xand is dense in K,
and each component of which meets X Let A, be the (non-complete)
subalgebra of έ?(X) generated by the functions in A and all their
derivatives. Let ίκ: A(K) —• AX{K) be the inclusion and i%: MAl{K)*—>
MMK)* be its adjoint. If we show that i% is onto for each K belong-
ing to a fundamental system of neighborhoods of X, then by passage
to the protective limit, it will follow that i* is onto. So suppose
that for a particular choice of K, i% is not onto.
Let A be the set of points of MAiK)* which have a neighborhood
which is an analytic disk relative to Z. As in the proof of Theorem
1, we see that MA(K}* — SA{K)* — A = E is at most countable. In view
of the Silov idempotent theorem, no point of MMK)* is isolated, so that
there is an open subset of A disjoint from ίi(MAl(K)*). Let W be a
component of A containing such an open set. We distinguish two cases.
Regard K as a subset of MMK)* and consider first the case in
which W contains a point of K. Then W is a Riemann surface with
the local coordinate Z. If / is a function in A(K), then / is analytic
on W. Denote the derivative of / with respect to the coordinate Z
by Df and the derivative of / with respect to Z by f\ If / and / '
both belong to A{K), then the connectedness of W, together with
the fact that W contains a point of K and hence an open subset of
K, implies that Df = / ' on W.
Let h belong to A{K) and let g = h'. Define a function g on W
by g(ζ) = Dh(ζ). The analysis of the previous paragraph implies that
g = g if g = hf belongs to A(K). Thus the functions in A(K) together
with their first derivatives, extend to be analytic on W. By iteration
of this process, we may extend each of the functions g in AX{K) to
an analytic function g on W; since W is connected, this extension is
unique.
Thus if δ is a homomorphism in W, δ extends to a homomorphism
of AX(K) by defining δ(g) — g(δ). If we show that δ is a continuous
homomorphism of A^K), and thus extends to A^K)*, then we will
have that i%(5) = δ and this contradiction will complete the analysis
of this case.
To this end, let us consider the boundary of W in MMK)*. Since
W is a connected component of A, no point of A is a boundary point
of W. Thus the boundary points of W belong either to K or to E.
If p is a boundary point of W that belongs to K, the fact that the
interior of K is dense in K and that the boundary of K is smooth
implies that there is a half-disk about p belonging to K. By enlarg-
ing K slightly we may effect a modification of K so that some half-
disk around p belongs to K Π W. An argument using the compactness
of the part of the boundary of W that lies in K shows that all the
818 WILLIAM R. ZAME

boundary points of W that belong to K may be assumed to have

half-disks in KC\ W about them (modifying K as necessary).
Now consider the boundary points of W that belong to E. If q
is one of these points, then the results of [2] imply that there is a
neighborhood Q of q with the property that Q — q lies in A and
consists of finitely many components, each of which is mapped by Z
homeomorphically onto a disk minus its center. Thus we may cover
W U {q} with a Riemann surface WQ in such a way that the functions
in A(K) extend to be analytic on Wq. We may certainly do this for
each of the boundary points of W lying in E. Thus, passing to a
covering Riemann surface when necessary, and modifying W as neces-
r
sary (by enlargement of K), we arrive at a Riemann surface W
which has a subset of the interior of K as a neighborhood of its
boundary, and to which the functions in A(K) extend naturally. As
before, we see that the functions in A^K) extend to W. Hence no
function in AX(K) assumes a larger value on W than on K. It follows
that δ is indeed a continuous homomorphism of A^K).
We have shown that each compact set K whose interior contains
X and is dense in K, and all of whose components meet X, can be
modified slightly to produce another such compact set K' with the
property that ί£,: MAl(κn* —• MAiKn* is onto. Since the collection of such
sets Kr forms a fundamental system of neighborhoods of X, it follows
from a passage to the protective limit that i*: MAQ —> MΛ is onto.
Now let Y and Y' be distinct components of MA[). Considered as
a map on MAQ, Z\ Y is one-to-one and Z — Z<>i* so that ί* is certainly
one-to-one. If there are infinitely many pairs (μ, v) in Y x Yf such
that ί*(μ) = i*(v) then some point (λ, ξ) is a limit point of such pairs.
We may choose a compact set K with smooth boundary, whose interior
contains Xand is dense in K, and such that i%{Y) and i%{Yr) belong
to different components of MMKV\ say T and Ύf respectively. If / is
in A(K), then/ is analytic on T - (ΓfΊ SMK).) and Tf - {T Π SMK)*)>
and the derivative of / may be obtained, as in the first part of the
proof, by differentiating with respect to the local coordinate Z. Then
the functions fo (Z\ T)"1 and fo (Z\ Tr)~ι are analytic in a neighborhood
of Z(ξ) = Z(X) and agree to infinite order there. Now it follows that
§(i%(ζ)) — 9(iκQΨ) for each g in A^K), since A^K) is generated by
functions in A(K) and their derivatives. It follows that i%(S) = i%{ξ)
which is a contradiction. It follows that only finitely many pairs in
Y x Yr are not separated by i*, as desired.
Finally, suppose that i* is a homeomorphism, and let / be in
AQ(U) for some open set U containing X. As in the proof of Theorem
1, we may choose a compact set containing X in its interior such that
/ ((K Π Z{L)) belongs to έ?(Z(L)) \ (K Π Z{L)) for each component L of
MAίK)*. As before, we may then conclude that / belongs to A(K)*
 ALGEBRAS OF ANALYTIC FUNCTIONS IN THE PLANE 819

a n d hence t h a t f\X belongs to A. Since U is a r b i t r a r y , this com-

pletes t h e proof.

COROLLARY 1. Let X be compact and connected. If A is a com-

plete subalgebra of &(X) containing the polynomials, then there is a
compact connected set Xf containing X and such that X' — X is open
and A - <?{X').

Proof. Let Ao be the stable algebra generated by A. The Silov

idempotent theorem implies that MA is connected and the corollary
now follows quickly from Theorems 1 and 2.

The above results have easy applications to questions of approxi-

mation on open sets as well. We mention one result that seems
particularly striking.

COROLLARY 2. Let U be a connected open set and let g be an

analytic function on U that admits no analytic extension to the union
of U with any of the bounded components of C — U. Then every
analytic function on U is the limit, uniformly on compact subsets of
U, of polynomials in g and Z.

REFERENCES

1. E. Bishop, Subalgebras of functions on a Riemann surface, Pacific J. Math., 8

(1958), 29-50.
2. , Analyticity in certain Banach algebras, Trans. Amer. Math. Soc, 102
(1962), 507-544.
3. , Holomorphic completions, alalytic continuations, and the interpolation of
semi-norms, Ann. Math., 7 8 (1963), 468-500.
4. J.-E. Bjork, Analytic structures, Papers from the Summer Gatherning on Function
Algebras at Aarhus, July (1969), 19-28.
5. , On analytic structures in the maximal ideal space of a function algebra,
Papers from the Summer Gathering on Function Algebras at Aarhus, July (1969), 29-35.
6. T. W. Gamelin, Uniform Algebras, Prentice-Hall, 1969.
7. , Polynomial approximation on thin sets, Papers from the Symposium on
Several Complex Variables, Park City, Utah, (1970), 50-78.
8. H. Royden, Algebras of bounded analytic functions on Riemann surfaces, Acta
Math., 144 (1965), 113-142.
9. J. Wermer, Function rings and Riemann surfaces, Ann. Math., 6 7 (1958), 45-71.
10. , Rings of analytic functions, Ann. Math., 6 7 (1958), 497-516.
11. W. R. Zame, Algebras of analytic germs, to appear.
Received May 21, 1971.
RICE UNIVERSITY