PACIFIC JOURNAL OF MATHEMATICS

Vol. 145, No. 1, 1990

EXTENSIONS OF
HARMONIC AND ANALYTIC FUNCTIONS
SHELDON AXLER A N D ALLEN SHIELDS
Allen Shields died on 16 September 1989. This paper is dedicated to him by the first
author, with many fond memories of our joint work together.

This paper studies the extensions of harmonic and analytic func-

tions defined on the unit disk to continuous functions defined on a
certain compactification of the disk.

Introduction. This introduction will present a quick survey of our

results; the complete definitions necessary to state these results pre-
cisely are given in later sections.
Let H°° denote the algebra of bounded analytic functions on the
open unit disk in the complex plane. The maximal ideal space of H°°
is denoted by M. We can think of the open unit disk as a dense subset
of M. Carl Sundberg [11] proved that every function in BMO extends
to a continuous function from M into the Riemann sphere; he also
described several properties of these extensions. Sundberg was work-
ing in the context of functions of several real variables. In the next
section of this paper we take advantage of the tools offered by analytic
function theory to give considerably simpler proofs (in the context of
one complex variable) of Sundberg's results about extensions of BMO
functions.
In the section of this paper on nontangential limits, we prove that a
function on the disk that has a continuous extension to a small subset
of M must have a nontangential limit at almost every point of the unit
circle. We then use this result to produce a class of functions in the
little Bloch space that cannot be extended to. be continuous functions
from this small subset of M to the Riemann sphere.
The section of this paper dealing with cluster sets and essential
ranges shows how those sets can be computed from the appropriate
continuous extensions. We use these results to give a new proof of Joel
Shapiro's theorem [10] that for every function in VMO, the essential
range equals the cluster set.

The final section of the paper discusses some open questions.

2 SHELDON AXLER AND ALLEN SHIELDS

We thank Peter Lappan, Joel Shapiro, and Carl Sundberg for helpful
discussions.
Extensions of BMO functions. Recall that H°° denotes the Banach
algebra of bounded analytic functions on the unit disk D and that
M denotes the maximal ideal space of H°°. Thus M is the set of
multiplicative linear maps from H°° onto the complex numbers C.
With the usual topology (the weak-star topology that M inherits as a
subset of the dual of H°°), M becomes a compact Hausdorff space.
We will freely think of the disk D as a subset of M by identifying
each point of the disk with the multiplicative linear functional of point
evaluation. The topology that D inherits as a subset of M coincides
with the usual topology on D.
Each H°° function (thought of as a function from D to C) extends,
via the Gelfand transform, to a continuous complex-valued function
defined on M. A bounded real-valued harmonic function on D is
not necessarily the real part of an H°° function. Nevertheless, if u is
a bounded real-valued harmonic function on D, then u extends to a
continuous function from M to the reals R. To see this, let u denote
the harmonic conjugate of u, so ύ is the unique real-valued harmonic
function on D such that u + in is analytic on D and w(0) = 0.
Obviously

for all z e D. Thus log|^(^ w + / ")| is a function (of φ) on M that

agrees with u on D, and clearly this function is continuous on M.
Thus we have proved the following useful lemma, whose proof we
have taken from the proof of Lemma 4.4 of [8].

LEMMA 1. Let u be a bounded real-valued harmonic function on D.

Then the function u defined on M by

is an extension of u to a continuous function from M to R.

Throughout the paper we will follow the convention established in

the lemma above of using the same symbol to denote a function and
its extension to a larger domain. Notice that by the Corona Theorem
([5], Chapter VIII), which states that the disk D is a dense subset
of M, any continuous function on D that extends to a continuous
function on M has a unique such extension.
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 3

Because M is compact, an unbounded function u: D —• R cannot

be extended to a continuous function u: M -» R. However, if we
allow the extended function to take on the values oo and - o o , and
give [-ex), oo] the obvious topology, then an unbounded function
u: D —• R might extend to a continuous function u: M —> [—oo, oo].
Carl Sundberg ([11], Theorem 3.4) proved that every real-valued
function in BMO (bounded mean oscillation) extends to a continuous
function from M to [—00, 00]. In this paper, we define BMO to
be the set of complex-valued harmonic functions on D whose real
and imaginary parts can each be written as the sum of a bounded
real-valued harmonic function on D and the harmonic conjugate of a
bounded real-valued harmonic function on D. For a proof that this
definition is equivalent to the usual definition, see [9], page 39.
In Theorem 2 below, we give a new proof of Sundberg's theorem
by using complex analysis instead of Sundberg's techniques of trunca-
tion (Sundberg was working in several real variables rather than one
complex variable). Of course, the following theorem implies that if /
is a complex-valued function in BMO, then / can be extended to a
continuous function from M to the Riemann sphere Cu{oo} (which
has the obvious topology).

THEOREM 2. Let u: D -» R be in BMO. Then u can be extended

to a continuous function from M to [—00,00].

Proof. Every real-valued function in BMO is the sum of a bounded

real-valued harmonic function and the harmonic conjugate of a bound-
ed real-valued harmonic function. By Lemma 1, every bounded real-
valued harmonic function extends to a continuous real-valued func-
tion on M, so we can assume that u = ϋ, where v is a bounded
real-valued harmonic function on D.

We can assume without loss of generality that

where |M|oo denotes the supremum of \υ\ on D. Thus eϋ~ιv is an

analytic function mapping D into the open right half-plane, which we
denote by RHP. Let F: RHP —• D be the analytic function defined
by
F{z)
= JTΪ
 SHELDON AXLER AND ALLEN SHIELDS

ϋ ιv ϋ iv
Then F oe ~ is a bounded analytic function, and hence F o e ~
extends to a continuous function h on M, taking values in the closed
disk D.
ι
Let F~ : Z) —• RHP U {oc} be the continuous function defined by

ι
where RHP u {00} is given the obvious topology. Note that F o F
ι ϋ iv
is the identity on RHP, and thus (F~ o h)\D = e ~ . The abso-
lute value function maps RHP U {00} continuously to [0, oc], with
ι ϋ
\(F~ oh)\r>\ = e . Finally, the logarithm function maps [0, oc] con-
ι
tinuously to [-00, oc], with log \{F~ o h)\n\ = v see Figure 3.

_ , 1 + z

F(
CRHP ^ z + 1
D F
-^ ~z R
RHP
1
H P U {00} U [0, 00] *% [-00, 00]

n h y

— •

extension of £

FIGURE 3
ι
Thus the function log \(F~ oh)\ is a continuous function from M
into [-oc, oo] that agrees with v on D. D
Because each function in H°° has a unique extension to a continu-
ous complex-valued function on M, we can think of H°° as a closed
subalgebra of C(M) and apply results from the theory of function
algebras. In particular, there is a smallest closed subset X oΐ M,
called the Shilov boundary of H°° , such that

sup{|/(p)| :φeM} = sup{|/(p)|: Ψ e * } for every / e H°°.

Because the map that sends / to f\χ is an isometry, we see that
H°° is isometric to a closed subalgebra of C(X). Thus by the Hahn-
Banach Theorem and the Riesz Representation Theorem, for each
φ e M, there is a positive regular Borel measure on X such that for
every / e H°° , f{φ) can be computed by integrating / against this
measure. Special properties of H°° (see [7], p. 182) show that this
measure is unique; we denote it by dφ. We will denote the closed
support of this measure by supp φ, so supp φ is a closed subset of
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 5

X, which is a closed subset of M. To summarize: For each φ G M,

there is a unique positive regular Borel measure dφ on X such that

f(φ) = f fdφ for every / G H°°.

Jx
We know that if / G BMO, then / extends to a continuous func-
tion from M to the Riemann sphere C U {00}, and so it is natural
to ask whether this extension of / satisfies the above equation. Be-
cause the extension of / may take on the value 00, it is not clear that
the right-hand side of the above equation makes sense. However, as
Sundberg proved ([11], Theorem 3.7), everything works out as nicely
as possible. We reprove Sundberg's results (Propositions 4 and 6 be-
low) using complex analysis rather than Sundberg's more complicated
technique of truncation.
The following proposition clearly implies that if u is a real-valued
function in BMO, φ G M, and u{φ) = —00, then u = —00 on supp φ .

PROPOSITION 4. Let u be a real-valued function in BMO, and sup-

pose that φ G M is such that u{φ) = 00. Then u = 00 on supp φ.

Proof. Replacing u by u plus a bounded real-valued harmonic

function will not affect either the hypothesis or the conclusion. Thus,
as in the proof of Theorem 2, we can assume that u = ϋ , where v is
a real-valued harmonic function on D and

IMIoo < 2'

Let h be as in Figure 3. From Figure 3, we see that because ϋ(φ) =
00, we have h{φ) = 1. On Z>, h equals an H°° function (namely
ϋ iv
Foe ~ )9 so
= h(φ)= f hdφ.
Jx
lχ
On M9 h takes values in the closed unit disk Z>, and the above
equation says that the integral of h against the probability measure
dφ is 1. Thus h must be identically 1 on supp φ, which (see Figure
3) implies that the continuous extension of ϋ to M is identically 00
on supp φ . D
If φ G M and / G H°° is invertible in H°° , then Jensen's equality
(see [6], Theorem 2.1) asserts that

= f lσg\τ(f)\dφ(τ).
Jx
6 SHELDON AXLER AND ALLEN SHIELDS

To prove Proposition 6, we will need Jensen's equality for another

class of functions. Jensen's equality does not hold for an arbitrary
outer function / e H°°, but the following proposition shows that it
does hold if the real part of / is positive on D.

PROPOSITION 5. Let f e H°° be such that R e / > 0 everywhere on

D, and let φ e M. Then

= f log\τ(f)\dφ(τ).
Jx

Proof. For each positive integer n , Re(/+^) > £ on D, so (/+£)

is an invertible function in H°° . Thus

log\φ(f)\ = lim log φ I f+ -

Λ-OO V n
r / i\ r
= nlim / log τ ( / + - dφ(τ)= / log\τ(f)\dφ(τ)9
-*°°Jx \ nj Jx
where the second equality comes from Jensen's equality for invertible
functions and the last equality comes from the Monotone Convergence
Theorem. α
Now we can prove that the continuous extension of a BMO function
can be obtained by integration against the representing measures.

PROPOSITION 6. Let u e BMO, and suppose that φ e M is such

that u(φ)eC. Then ueLι(dφ) and

u{φ) = / udφ.
Jx
Proof. We first need to prove the proposition in the case when u
is a bounded harmonic function on ΰ . To do this, we can assume
that u is a bounded real-valued harmonic function on D. Then by
Lemma 1 and Jensen's equality we have

= / \og\τ(eu+i*)\dφ(τ)= ί u(τ)dφ(τ)9
Jx Jx
as desired.
Now we no longer assume that u is bounded. The paragraph above
implies that replacing u by u plus a bounded harmonic function will
not affect the conclusion. Thus, as in the proofs of Theorem 2 and
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 7

Proposition 4, we can assume that u = ϋ, where i> is a real-valued

harmonic function on D and

Hoo<f-
00
Let /2 G i / be as in Figure 3. Then (see Figure 3)
φ{h)
v(φ) = log
- φ{h)
Obviously 1 + φ{h) and 1 - φ{h) cannot both be 0, so the last ex-
pression makes sense and is not of the form — oo — (—oo). Because
ϋ(φ) G C, neither l + φ(h) nor l-φ(h) can be 0. The above equation
becomes
ϋ(φ)=log\φ(l+h)\-log\φ(l-h)\
= f log\τ(l+h)\dφ{τ)- [ log\τ(l-h)\dφ(τ),
Jx Jx
where the last equality comes from Proposition 5 (notice that h maps
D into D, so that both 1 + h and 1 - h have positive real parts on
D). Finally, the above equation becomes

//f Λf = / vdφ,
= Jx /x log
J
where the last equality, which completes the proof, comes from Figure
3. D
Nontangential limits. The Bloch space is the set of analytic func-
tions / on the disk D such that
sup{(l-|z|2)|/(z)|:zG/)}<oo.
The little Bloch space is the set of analytic functions / on the disk D
such that
(l-\z\2)f(z)-+0 as|z|Tl.
We have seen that every BMO function can be thought of as a
continuous function from M to Cu{oo}. Can this result be extended
to other well-studied classes of functions? Sundberg ([11], Section
5) gave an example of a function in the Bloch space that cannot be
extended to a continuous function from M to C U {oo} . Sundberg's
function is not in the little Bloch space. This raises the question of
whether every function in the little Bloch space can be continuously
extended to a function from M to C U {oo} . In this section we will
construct a large class of functions in the little Bloch space that do not
8 SHELDON AXLER AND ALLEN SHIELDS

have continuous extensions (with values in C U {oo}) to M or even

to the much smaller set DuX (recall that X is the Shilov boundary
of H°°).
Our principal tool will be the following theorem about nontangential
limits. When we say that a nontangential limit exists, we are allowing
oo as a possible value. Statements about almost every point of dD
refer to the usual arc length Lebesgue measure on dD. Note that in
this theorem u is not required to be analytic or even harmonic.

THEOREM 7. Let u be a continuous function on DuX taking values

in C u {00}. Then u\r> has a nontangential limit at almost every point
of 3D.

Proof. We begin by considering the special case where u can be

extended to a continuous function from M to C. By the Stone-
Weierstrass Theorem, functions of the form

/1IΓ+••• + / * ! £ ,
where fj , gj G H°° and n is an arbitrary positive integer, are dense
in C{M). Each such function has a nontangential limit at almost
every point of dD, and hence the uniform limit of any sequence of
such functions also has this property. Thus we are done in the special
case under consideration.
Now consider the case where u is a continuous function on ΰ u l
that takes on values only in C (unlike the paragraph above, we are
no longer assuming that u can be continuously extended to M). By
the Tietz Extension Theorem, there is a function v G C(M) such that
u\x = v\χ. Notice that u - V\DUX is a continuous complex-valued
function on ΰ u l that is 0 on X. By the above paragraph, v|/> has
nontangential limits almost everywhere on dD, so that by replacing u
by u -V\DUX > we can assume without loss of generality that u\χ = 0.
Let ε be an arbitrary positive number. We claim that there is a
positive number δ and an inner function b such that
{zeD: \b(z)\ >l-δ}c{zeD: \u(z)\ < ε}.
To see how this claim will complete the proof in the case under consid-
eration (u is complex valued and u\χ = 0), let {δny%Lχ, a sequence
of positive numbers and (bn)%Lx, a sequence of inner functions, be
such that

{z e D: \bn(z)\ > 1 - δn} c jz G D: \u(z)\ < i

 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 9

For each n , the function \bn\ has nontangential limit 1 at each point
of dD\En, where En is a subset of dD of measure 0. The above
inclusion thus implies that u has nontangential limit 0 at each point
of dD\(\J™=ι En), completing the proof in this case.
We now prove the claim made in the paragraph above. The set
{φ G ΰ U l : \u(φ)\ < e} is an open subset of D UX containing X.
Thus there is an open subset & of M such that
{φeDuX: \u(φ)\ < ε} = t?Π(DUX).
By the theorem on page 179 of [7], X is the intersection, taken over
all inner functions b, of the sets {φ e M: \b(φ)\ = 1} . Thus

{φeM:\b(φ)\ = l}cd?.
b inner

Taking complements (in M), the above inclusion states that the com-
pact set M\9 is contained in the union of a certain collection of open
sets, and hence in the union of a finite subcollection. Taking com-
plements again, we see that there are finitely many Blaschke products
bι,...,bn such that {φeM: \bχ(φ)\ = 1}Π -Π{φeM: \bn(φ)\ = 1}
is contained in &. Letting b = b\ - bn, we obtain a single inner
function b such that
{φeM: \b(φ)\= \}<z<9.
Finally, \b\ has a maximum on the compact set M\<9 which must be
strictly less than 1, so there is a positive number δ such that
{φeM: \b{φ)\> l-δ}c<?,
which implies that
{zeD: \b(z)\> l-δ}c{zeD: |κ(z)|<e},
completing the proof of the claim.

We have now proved the theorem in the case where u takes values
only in C. To complete the proof in the final case, where u is a contin-
uous function on DUX taking values in Cu{oo}, let P: Cϋ{oo} —• C
be a continuous function such that for every w eC, P~ι(w) is a fi-
nite set. (One way to construct such a function is to think of C u {oc}
as the unit sphere in R 3 , and let P map a point (x,y9z) on the
unit sphere in R 3 to x + iy e C. For this choice of P, if w e D,
then P~ι(w) contains two points; if w e dD, then P~ι(w) contains
one point; and if w e C\D, then P~ι(w) contains no points.) The
10 SHELDON AXLER AND ALLEN SHIELDS

function Pou is a continuous function o n ΰ u l taking values in C.

Thus by the earlier cases, P o u\r> has a nontangential limit at almost
every point of dD.
Consider a point λedD at which Pou\o has a nontangential limit
w . On each nontangential region based at λ, u must be near the set
ι ι
P~ (w). However, P~ (w) is a finite set and each nontangential
region is connected, so u must actually be near only one point in
ι
P~ (w), which means that U\D has a nontangential limit at λ. D
A sequence {n^} of positive integers is called a Hadamard sequence
if there exists a real number q > 1 such that n^i/n^ > q for every k.
The following theorem produces a class of functions in the little Bloch
space that cannot be continuously extended (with values in C U {oo})
to DUX.

THEOREM 8. Let {nk)f=ι be a Hadamard sequence. Let {ak)™=ι be

a sequence of complex numbers such that
oo
2
a^-^0 and ^
k=ι
Let

k=\
Then f is a function in the little Bloch space that cannot be extended
to a continuous function from DuX to C U {oo}.

Proof. Because (n^)^ is a Hadamard sequence and ak —• 0, /

is in the little Bloch space ([1], Theorem 2). Because («^)^ =1 is a
Hadamard sequence and X) | % | 2 = oc, at almost every point of dD,
/ fails to have a finite radial limit ([12], Chapter 5, Theorem 6.4).
Thus at almost every point of dD, / fails to have a finite nontan-
gential limit. By a theorem of Plessner and Privalov ([4], Corollary
1 to Theorem 8.1), at almost every point of dD, / fails to have a
nontangential limit (finite or infinite). Thus by Theorem 7, / cannot
be extended to a continuous function from ΰ u l to Cu{oo}. D
The theorem above seems to suggest that nothing interesting can be
said about extending Bloch functions continuously to M. However,
Leon Brown and P. M. Gauthier ([3], Theorem 4) proved that every
normal function (in particular, every Bloch function) can be extended
to a continuous function (with values in C U {oo}) defined on a large
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 11

subset of M (namely, the union of all the nontrivial Gleason parts of

M).
The cluster set and essential range. In this section we will show
that the cluster set and essential range of a function in C{M) can be
computed from values on fibers. We will then use these results and
the techniques from earlier in the paper to give a new proof of Joel
Shapiro's theorem [10] that for every function in VMO, the essential
range equals the cluster set.
If u is a function from D to Cu{oo} and λ G 3D, then the cluster
set of u at λ, denoted c\(u, λ), is defined to be the set of w e Cu{oo}
such that there is a sequence (zn)™=ι in D with zn —• λ in C and
u{zn) —• w in C u { o o } . Notice that if u extends to a continuous
function from D U {λ} to C U {00}, then cl(w, λ) is {u(λ)} .
For λedD, the fiber of M over A, denoted Mλ, is defined by

here z denotes the identity function on D.

The following proposition shows that the cluster set at a point is
equal to the image on the fiber of M over the point. Note that the
function u is not required to be analytic or harmonic

PROPOSITION 9. Let u be a continuous function from M to Cu{oo}

and let λedD. Then cl(w, λ) = u(Mλ).

Proof. First suppose that w e cl(u, λ). Thus there is a sequence

(zn) in D such that zn -> λ and u(zn) -+ w . By the compactness of
M , some φ e Mλ is in the closure of {zn} clearly u(φ) = w . Thus
cl(w, Λ) is contained in

Now suppose that w e u(Mλ). For each positive integer n, let £„

be the open unit ball on the Riemann sphere of radius l/n centered
at w. Let

0n = I φ e M: u(φ) e En and \φ(z) - λ\ < - \.

Clearly ffn is an open subset of M. Because w e u(Mλ), ffn is

non-empty. Thus by the Corona Theorem, each @n contains a point
zn G D. Clearly z Λ —• A in C and u(zn) —• w in C U {00}, so
w G cl(w, A). Hence u(Mλ) is contained in cl(w, λ), completing the
proof. D
12 SHELDON AXLER AND ALLEN SHIELDS

Suppose that u is a continuous function from M to Cu{oo}. Let

u* denote the boundary value function defined for λ e 3D by
u\λ) = lim u(rλ)

if this limit exists in Cu{oo} . By Theorem 7, u* is defined for almost

all points of dD.
We will need to consider the usual Banach algebra L°°(dD), which
consists of the bounded, Borel measurable, complex-valued functions
on ΘD. The maximal ideal space of L°°(dD) is denoted by
M(L°°(dD)) .Forψe M(L°°(dD)), define ψ* e M by

for h G H°°. By the theorem on page 174 of [7], the map from
M(L°°(dD)) to M that takes ψ to ψ* is a homeomorphism of
M{L°°{dD)) onto X.
We will need the following lemma in our proof of Proposition 11.

LEMMA 10. Let u: M -> C be a continuous function and let ψ e

M(L°°(dD)). Then
u(ψ*) = ψ{u*).

Proof. First consider the case where U\D is a real-valued harmonic

function on D. Then

u(ψ*) = log|^ ( ^ + / δ ) | = log|vί([^ + ί f l r)U

where the first equality comes from Lemma 1 and the second equal-
ity comes from the definition of ψ*. The theory of commutative
C*-algebras shows that for an arbitrary function U € L°°(dD), we
have \ψ(U)\ = ψ(\U\) and ψ{eu) = eψ^u). Thus the above equation
becomes
u(ψ*) = logψ(\[eu+iύ]*\) = logψ(eu*) = l o g έ ? ^ = ψ(u*),
completing the proof in this case.
Now consider the set of all functions v in C(M) such that v(ψ*) =
ψ(υ*). This set is closed under addition, multiplication, and uniform
limits, and by the paragraph above contains all the bounded harmonic
functions on D. By the Stone-Weierstrass Theorem, this set must be
all of C(M), completing the proof. D
The essential range of u* at λ, denoted essran(w*, λ), is defined
to be the set of w e C u {oo} such that for every open subset E
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 13

of C U {00} that contains w and every open interval / of dD that

ι
contains λ, IΠ(u*)~ (E) has positive Lebesgue (arc length) measure.
Notice that if u* is continuous at λ, then essran(w*, λ) is {u*(λ)} .
The fiber of X over λ, denoted Xχ , is defined by

The following proposition shows that the essential range of the bound-
ary value function at a point is equal to the image on the fiber of X
over the point. Note that the function u is not required to be analytic
or harmonic.

THEOREM 11. Let u be a continuous function from M to Cu{oo}

and let λedD. Then essran(w*, λ) = u(Xλ).

Proof. First we consider the case where u takes values only in C.

By the theorem on page 171 of [7],
essran(w\ λ) = {ψ(u*): ψeM{L°°(λD)) and ψ* e Xλ}.
Lemma 10 shows that the right side of the above equation equals
u{Xχ), completing the proof in the case where u is complex valued.
Now suppose that u takes values in C U {oc} . As in the proof of
Theorem 7, let P: Cu {oo} —» C be a continuous function such that
for every w e C, P~ι(w) is a finite set. Then P o u e C(M), and so
by the paragraph above the desired results holds for Pou. Thus
P(essran(w*, λ)) — essran(P ou*, λ)
= essran((P o u)*, λ) = (P o u)(Xλ) = P(u(Xλ)).
Because the above equation holds for every function P as described
above, we have
essran(w*, λ) = u{Xλ),
completing the proof. D
In this paper, we define VMO (vanishing mean oscillation) to be
the set of complex-valued harmonic functions f on D such that the
real and imaginary parts of / can each be written as the sum of a
uniformly continuous, real-valued harmonic function on D and the
harmonic conjugate of a uniformly continuous, real-valued harmonic
function on D. For a proof that this definition is equivalent to the
usual definition, see [9], page 49.
Joel Shapiro ([10], Corollary 2.2) proved that for a function in
VMO, the cluster set equals the essential range. Using the techniques
developed above, we now give a new proof of Shapiro's theorem.
14 SHELDON AXLER AND ALLEN SHIELDS

THEOREM 12. Let u be a function in VMO and let λedD. Then

cl(w, λ) = ess ran(w*, λ).

Proof. By Proposition 9 and Theorem 11, we need only show that

u(Mλ) = u(Xλ).

Obviously u(Xλ) is contained in u(Mλ). To prove the reverse inclu-

sion, let φ G Mλ. We claim that u is constant on supp φ. If this
claim is true, then by Propositions 4 and 6 (recall that dφ is a proba-
bility measure), w must equal u(φ) on supp $P furthermore, supp φ
is contained Xλ (by the first corollary on page 188 of [7]), and so
u(φ) e u(Xχ), as desired.
To complete the proof we must show that u is constant on supp φ .
Every uniformly continuous function on D is constant on Xλ (by
Theorem 11), and so we can assume without loss of generality that
u = ϋ, where v e C(D), v is real valued and harmonic on D,
v(λ) = 0, and H^Hoo < π/2. Because v(λ) = 0, the cluster set at λ of
the H°° function

is contained in R. Thus by Proposition 9, h(Mλ) is contained in

R, where h is the function defined by Figure 3. In particular, h is
real valued on supp φ . The support set of any representing measure
is anti-symmetric, so h is constant on supp$p (see [2], page 137).
From Figure 3 we can now conclude that ϋ is constant on supp φ,
completing the proof. D
Questions. Which harmonic (or analytic) functions on the disk ex-
tend to continuous functions from M to C U {oo} ? As we have seen,
the collection of such functions contains BMO. However, as Sundberg
pointed out ([11], pages 766), the set of analytic functions that can be
continuously extended to continuous functions from M to C U {oo}
is closed under squaring and so is strictly larger than the set of analytic
functions in BMO. We note that there is no reason to suspect that the
set of functions under consideration will be closed under addition.
Our guess is that a harmonic function on the disk has an extension to
a continuous function from M to the Riemann sphere if and only if
the harmonic function has bounded mean spherical oscillation, which
means that the oscillation is measured in the spherical metric rather
than the Euclidean metric.
 EXTENSIONS OF HARMONIC AND ANALYTIC FUNCTIONS 15

Is there a converse to Theorem 7? Specifically, if u is a continuous

function on D that has a nontangential limit at almost every point of
<9Z),does u have a continuous extension to DuXΊ (Note: After this
paper was circulated in pre-print form, Christopher Bishop and Oleg
Ivanov proved that this question has an affirmative answer.)

REFERENCES

[I] J. M. Anderson, Bloch functions: The basic theory, in Operators and Function
Theory (S. C. Power, editor), D. Reidel Publishing Co., Dordrecht, 1985.
[2] Andrew Browder, Introduction to Function Algebras, W. A. Benjamin, New
York, 1969.
[3] Leon Brown and P. M. Gauthier, Behavior of normal meromorphic functions on
the maximal ideal space of H°° , Michigan Math. J., 18 (1971), 365-371.
[4] E. E. Collingwood and A. J. Lohwater, The Theory of Cluster Sets, Cambridge
University Press, Cambridge, 1966.
[5] John B. Garnett, Bounded Analytic Functions, Academic Press, New York, 1981.
[6] Kenneth Hoffman, Analytic functions and logmodular Banach algebras, Acta
Math., 108(1962), 271-317.
[7] , Banach Spaces of Analytic Functions, Prentice-Hall, Englewood Cliffs,
1962.
[8] , Bounded analytic functions and Gleason parts, Ann. of Math., 86 (1967),
74-111.
[9] Donald Sarason, Function Theory on the Unit Circle, Virginia Polytechnic and
State University, 1978.
[10] Joel H. Shapiro, Cluster set, essential range, and distance estimates in BMO,
Michigan Math. J., 34 (1987), 323-336.
[II] Carl Sundberg, Truncations ofBMO functions, Indiana Univ. Math. J., 33
(1984), 749-771.
[12] A. Zygmund, Trigonometric Series, Vol. 1, second edition, Cambridge Univer-
sity Press, Cambridge, 1968.

Received November 1, 1988. Both authors were supported in part by the National
Science Foundation.

MICHIGAN STATE UNIVERSITY

EAST LANSING, MI 48824

AND

UNIVERSITY OF MICHIGAN
ANN ARBOR, MI 48109