Harmonic Functions in the Unit Disk

Victor L. Shapiro
The American Mathematical Monthly, Vol. 109, No. 1. (Jan., 2002), pp. 37-45.
Stable URL:
http://links.jstor.org/sici?sici=0002-9890%28200201%29109%3A1%3C37%3AHFITUD%3E2.0.CO%3B2-V
The American Mathematical Monthly is currently published by Mathematical Association of America.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/maa.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact [email protected].

http://www.jstor.org
Mon Mar 10 19:24:55 2008

Harmonic Functions in the Unit Disk

Victor L. Shapiro
1. INTRODUCTION. Carl Friedrich Gauss (1777-1855) and Bernhard Riemann
(1826-1866) were two of the leading mathematicians of all time. Each made valuable
contributions to complex analysis in general and to the theory of harmonic functions
in particular. In this article, we present an interesting theorem that either of them could
easily have conjectured but neither could have proved, because the tools on which the
proof hinges had not yet been developed during their lifetimes. Our focus here is on
the use of these tools.
Let S2 denote the open unit disk in the complex plane. A function u ( x , y ) is said to
be harmonic in S2 if the following holds: u belongs to the class C2(S2)and

for all ( x , y ) in S2, where Au = a2u/ax2 a2u/ay2= uss

Gauss proved that, if u is harmonic in 52, then

+ u,,,.

whenever z ( x o ,yo; r ) is contained in S2. Here D(xo,yo; r ) is the open disk centered at
(xo,yo) with radius r . The expression on the right-hand side of ( 1 )is called the integral
mean-value of u at ( x o ,yo). The converse of the above result is also true; namely, if
u is a function from C0(S2)and ( 1 ) holds whenever B ( x o ,yo; r ) lies in 52, then u is
harmonic in S2. The two results together are referred to in the literature as the Gauss
mean-value theorem for harmonic functions.
Riemann, whose Ph.D. thesis was read and approved by Gauss, is well known for
the Cauchy-Riemann equations. These state that, if f ( z ) = u ( x , y ) i v ( x , y ) is analytic in '2, then

throughout S2. From these equations and the knowledge that the analytic function f is
in the class CCO(Q),
it follows that

which ensures that Au = 0 in '2. We conclude that u ( x , y) = Rf ( z ) ,the real part

of f , is harmonic in '2.Likewise, v ( x , y) = Zf ( z ) ,the imaginary part of f , is harmonic in S2. It is a standard fact from elementary complex analysis that, if u is
harmonic in S2, then there exists a harmonic function v such that

is analytic in S2. So u is harmonic in S2 if and only if it is the real part of an analytic

function there. We shall need this information in the sequel.
January 20021

HARMONIC FUNCTIONS IN THE UNIT DISK

37

In what follows, we shall use the expression "w(r, 8) is harmonic in

that there exists u(x,y ) harmonic in $2 such that

a"to signify

forOir<land-m<8<m.
The problem that we are going to discuss starts out with the work of an important
French contemporary of Gauss, SimCon Poisson (1781-1840), and the famous kernel
function P(z) (or in polar coordinate form, P(r, 8)) that bears Poisson's name. It is
defined for z in as follows:

To check the second equality, simply observe that

and that

where z = r cos 6 i r sin 8 . Because P(r, 8) is the real part of an analytic function, it
represents a harmonic function in the unit disk. Furthermore, using a familiar identity
for 1 - cos 8, we see that the Poisson kernel can be rewritten as

Since sin(O/2) # 0 for 8

# 0 mod 2 n , we infer that

lim P(r, 8) = 0
r+ 1

for 0 < 8 < 2n. This is clearly not the case for 8 = 0, where the corresponding limit
is m .
Next, we recall that in polar coordinates

Consequently, A(a P / M ) = a(A P ) / a 8 = 0; so we discover that the function a P / a 8 =

Pois also harmonic in the unit disk. (Dealing with the origin, where r = 0, presents a
minor technicality that is not difficult to overcome.) An easy computation reveals that
Po(r, 8) = -

r(1 - r2) sin6

[(I - r)2

+ 4r sin2(6/2)12'

from which we conclude that

lim Pe(r, 8) = 0
r-1

for every 6.

38

@ THE MATHEMATICAL ASSOCIATION O F AMERICA [Monthly 109

Using (2), we next see that

1
P(r. 8) = 2

+ " r" cos no,

,
Po(r, 6) = -

n=1

nrn sin nB
n=1

Let (1 w(r, .)(I, designate the uniform norm on concentric circles of radius r , i.e.,

We deduce from (4) that, as r

+1,

(For "0"and "o" notation, see Zygmund [4, p. 141.) Using the well-known inequality
sin6 2 28/n for 0 < 6 < n/2, we also observe that
lim inf (1 - r12 I PB(r, 1 - r ) ( 2 4/n.
r+l

Hence, as r + 1, / ( Po(r, .) ( I , # o((1 - r)-2). Furthermore, recognizing the prototypical role that P (r, 8) plays in the theory of harmonic functions and noting as well that
0((1 - r)-') behaviour on the part of a quantity as r + 1 implies that this quantity is
o((1 - r)-2), one might be led by the foregoing discussion to conjecture the following
theorem concerning harmonic functions in the unit disk:

Theorem. Given that w (r, 8) is harmonic in '2, suppose that

(i) lim w (r, 8) = 0 for all 8
r+l

and
(ii) (1 w(r, .) 1 ,

= o ((1 - r)-2)

as r + 1.

Then w (r, 8) is identically zero in '2.

What is especially good about a theorem of this nature (which was established in
1963 by the author [3]) is that it is a "best possible" result. For if the statement "for
all 8" in (i) were replaced by "for all 8 # 0 mod 2n," the conclusion of the theorem
would be false, as P(r, 8) shows. Also, if the "on condition in (ii) were replaced by
"0,"the conclusion of the theorem would again be false, as Po(r, 8) demonstrates.
It is clear that this is a theorem that both Gauss and Riemann could have understood
and appreciated and might even have conjectured themselves. But it is highly unlikely
that either could have proved it. The proof requires notions from point-set topology that
were first introduced by Georg Cantor (1845-1918) in the 1870s, at least four years
after the death of Riemann. It also demands a passing acquaintance with the second
category theory of sets, as published by RenC Baire (1874-1932) in 1899. Specifically,
it uses the fact that a closed subset E of the real line R cannot be written as a countable
union of closed sets that are nowhere dense in E. In the sequel, we shall refer to this
result as Baire's theorem. The proof of Baire's theorem can be found in most graduate
texts in real analysis or topology.

2. PROOF OF THE THEOREM. The question is: How does one establish a result
of the above type? The conventional way of dealing with uniqueness properties of
January 20021

HARMONIC FUNCTIONS IN THE UNIT DISK

39

harmonic functions via the maximum principle does not appear to be applicable here.
What plays to our advantage, however, is that in a the function w is the real part of an
analytic function. Hence,

where f (z) = Czocnznis the Taylor series expansion of f in

ib,, we obtain

a.Writing c,

= a, -

x
CO

~ ( r8). = no

+ ,,=I (a,, cos n8 + b, sinn8) r n ,

(5)

in which the series converges uniformly in (r, 8) for 0 5 r 5 ro < 1, where ro is any
fixed radius in (0, 1). What we also have at our disposal is the information that

r+l

a0

x
CO

(a, cos n8

+ b,, sin n8) r"

,,=I

for all 8. Now, Cantor was led to the theory of sets [I, p. 9701 when he proved that
k

(a, cos n8

+ b, sin n0)

for all but a finite number of 8 with 0 5 8 < 2n implies that the coefficients a, and
b, are zero for all n. If we can show that condition (6) implies a similar conclusion
for the coefficients in the representation of w (r, 8) in (5), we will have established our
theorem. In other words, what we have to do is to improve on Cantor's techniques
(which were, in fact, inspired by Riemann's earlier work on trigonometric series).
For simplicity, we shall assume that w evaluated at the origin is zero, i.e.,

for all 8, which is also a property enjoyed by Pg(r, 8). Using (5), we evaluate w(0, 0)
and see that

Therefore, a. = 0 and
00

(a,, cos n8

w(r, 8) =

+ b, sin no) r".

n=1
Next, we observe that for 0 < r < 1,
rl'a, = n-'

w (r, 8) cos n8 do.

@ THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 109

Talung r = 1 - n-', we can appeal to assumption (ii) in the theorem to assert that
( 1 - n-')"an = o(n2).Consequently, we learn (with a similar argument for b,) that

a,, = o (n2),

b, = o ( n 2 )

asn-oo.
We next set

(a,,cos n8

wl(r, 8 ) =
n=l

+ b,, sin n8) r

n

and

C (a,,cos n8 n2+ b,, sin no) rl',

03

W ( r .Q ) =

n=1

and see that

awl@,8 ) / a r = r-'w(r, Q ) , aW(r, Q ) / a r= F 1 w l ( r ,8 ) .

(8)

From (8) and hypothesis (i) of the theorem, we infer that both wl (r, 8 ) and W (r, 8 )
satisfy the Cauchy criterion as r -+ 1 for fixed 8. In particular, the finite limit
lim W (r, 8 ) = W ( Q )

,+ 1

exists for all 8. Plainly W ( Q )is a periodic function of period 2 n . (The function
-W(r, 8 ) is called the Riemann function associated with w(r, 8).) We shall establish
our set-theoretic proof assuming the following three facts from the Riemannian theory
of trigonometric series (see Zygmund [4, ch. 1l ] and Kline [I, p. 9681). We shall make
further comments on these facts in the appendix of this paper.

Fact 1. If W ( 8 ) is continuous in the open interval ( a , p), then W ( 8 )is a linear fz~nction in ( a , p).
Fact 2. If W ( Q )is continuous in the open interval ( a ,B), then W ( Q )is continuous
from the right at a and from the left at P.
Fact 3. I f W ( 6 ) is identically a constant, then lim,,

W (r, 8 ) = W ( 0 ) uniformly in 6.

Together, Facts 1 and 2 imply that W ( 8 ) is a linear function in the closed interval
[ a ,PI.
Proceeding with the proof of the theorem, we observe from ( 8 ) that the following
situation prevails:

Fact 4. If w(r, 6 ) is uniformly bounded for 6 in a set E and for 112 5 r < 1, then
lim,,l W ( r ,6 ) = W ( 8 ) uniformly on E.
Indeed, it is clear from ( 8 ) that both w l ( r ,8 ) and W ( r ,6 ) satisfy uniform Lipschitz
conditions in the asserted range of r and 6 . Therefore, the limit is uniform, as stated.
We see that the proof of the theorem will be complete if we can verify that

January 20021

W ( 8 ) is continuous in the open interval (0, 2 n ) .

(10)

HARMONIC FUNCTIONS IN THE UNIT DISK

41

By Fact 2, W(0) is then continuous from the right at 0 and from the left at 2n; by
periodicity, it is continuous on the whole real line. Fact 1 thus ensures that W (Q) c ,
a constant, for all 6 in R. However, by the definition of W(r, 0) and Fact 3,

Therefore W (6)

0. Consequently,
a,r'"in2 = n-'

W(r, 6) cos nQd0

as r + 1, so the coefficient a,, = 0 for all n . Similarly, b , = 0 for all n , and the
theorem is established, modulo the verification of (10).
To show that statement (10) is true, we need the point-set topology and second
category theory alluded to earlier in this article. In particular, for m = 1 , 2 , . . . we set
F, = {Q E [O. 2n] : g(0) 5 m ) ,

(11)

where

Because w(r, 0) is a (jointly) continuous function of r and 0, it is clear that g(0) is a

lower semicontinuous function. Accordingly, F,, is a closed set for each m. It follows
from hypothesis (i) of the theorem that

for every closed subset E of [O,2n].An application of Baire's theorem then leads to

Fact 5. If E is a nonempty closed subset of [O,2n], then there is an open interval J

and an index rno such that E n J is nonempty and E n J is contained in E n F,,.
Referring back to (1I), we see that w (r, 0) is uniformly bounded for 0 in E fl F,,

and 0 5 r < 1. Consequently, we infer from Fact 4 that

lim W(r, 0) = W (6) uniformly on E n J.

(12)

r-t 1

(We shall employ Fact 5 and (12) twice in what follows-once

interval and again when E is a nowhere dense perfect set.)
Next, we define

when E is a closed

Z = {6 E (O,2n) : W is not continuous at 81.

(13)

We propose to show that Z is the empty set, which will establish (10) and complete
the proof of the theorem.
Suppose, to the contrary, that Z is not the empty set. We first exploit Fact 2 to assert
that Z has no isolated points. From Fact 5 and (12) applied to E = [a, B ] for arbitrary
a and /3 with 0 < a < /3 < 2n, it then follows that Z is nowhere dense in [O. 2x1.
Consequently, Z is a nonempty perfect set that contains no interval. Applying Fact 5
42

@ THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 109

and (12) once again, this time with E = 7,we see that there is a closed interval [a, P ]
and a 80 in Z with 0 < a < 80 < p < 2 n such that
the restriction of W to [a, ,5]

n 2 is continuous at do.

(14)

We shall demonstrate that

W (8) is continuous at 80 as a function on (a, B).

(15)

By (13), this gives a contradiction to the fact that Q0 belongs to Z. Hence Z is empty,
and the theorem is proved.
The final missing piece of the proof is to show that, given statement (14) concerning
the nowhere dense, perfect subset Z of [O,2n], (15) follows. We prove that
W(8) is continuous from the right at Qo.

(16)

(Similar reasoning works for continuity from the left.) If 80 is the left-hand endpoint
of an open interval in (a, p ) \ z , then Fact 2 implies that W(8) is continuous from the
right at Qo. So what remains is to argue that (16) continues to hold when 80 is merely a
right-hand accumulation point of 7.We now do this.
Let E > 0. By (14), we can pick 6 > 0 so that

whenever 8 lies in 7 and Qo < Q < Q0 6. Next, choose 81 in 2 with 80 < 81 < 80 6.
Set 81 - Q0 = i . Now suppose Q* is a given point of (a, p ) \ T such that 80 < 8* <
80 i.Then Q* lies in some component interval of (a, p)\T, say (02, Q3),where 80 <
O2 < Q3 ( 61 and both 82 and Q3 are in 7 . Furthermore, by Facts 1 and 2, W is a linear
function in the closed interval [d2,03]. Since the inequality (17) holds for 8 = Q2 and
8 = 63, linearity dictates that it hold for 8 = 8* as well. What we have shown is that

whenever Qo < Q < Q0 i.

Hence, W(8) is continuous from the right at 80. Assertion (15) is thereby established, and the proof of the theorem is at last complete.
It is clear that the category-type argument on which the proof of the theorem rests
was not in the purview of either Gauss or Riemann. W. F. Osgood appears to be one of
the first analysts to make use of such a technique (as the referee was kind enough to
point out). For an example of this, see his 1901-02 paper [2, p. 311.

3. APPENDIX. We proved the theorem under the assumption that w(r, 8) given
by (5) satisfies w(0,Q) = 0 for all 8 . To prove it without this assumption, consider
the function
h (r, Q) = w (r, 8) - w (r2, 28) ,

which is harmonic in the unit disk, satisfies hypotheses (i) and (ii) of the theorem, and
has h (0,Q) = 0 for all 8 . To see that this is true, we observe from (5) that

January 20021

HARMONIC FUNCTIONS IN THE UNIT DISK

for all 8. By what we have shown, h (r, 8 ) 0. Using ( 5 ) again, an easy computation
based on the vanishing of h demonstrates that

b,, =

0 for n odd and n >_ 1,

bnj2 for n even and n 2 2,

where b, are the coefficients occurring in (5) for w(r, 8 ) . Because (k 1 ) / 2 5 k for
k >_ 1, we conclude from a simple induction on n that b, = 0 for all n >_ 1. A similar
fact holds for a,,. We thus deduce from (5) that w(r, 8 ) 5 ao. Assumption (i) of the
theorem then implies that w (r, 6 ) r 0 and completes the proof.
Fact 1 follows from Rajchman's lemma [4,p. 3531, which shows that

for all 8 in ( a , p ) , where 02and @ signify the upper and lower symmetric second
derivatives, respectively. The first condition implies that - W is convex in ( a , p ) , the
second that - W is concave in ( a , p). Hence, W is a linear function in ( a , p).
Fact 2 follows from a notion of smoothness found in Zygmund [4, p. 43 and p. 2631,

applied to the continuous function

G(8)=

C (a, sin n8 n b, cos n8)

-

n=l

The smoothness assertion in question is that, if G has a derivative from the right at
some point, then it has a derivative from the left at that point and the two are equal.
To establish Fact 2, we also need Fatou's theorem found in Zygmund [4, pp. 991001, which tells us, in particular, that

for all 8 in R, where D' and 0' signify the upper and lower symmetric first derivatives,
respectively.
If W ( 6 ) = c16 c2 in the finite open interval ( a , p ) , then it follows from (18) that
G ( 6 ) = c 1 Q 2 / 2 c26 c3 in the closed interval [ a ,p]. Hence G ( 6 ) has a derivative
from the right at a equal to cla c2. But then by smoothness, it has a derivative
from the left at a equal to this same value. Consequently, G ( 8 ) has a derivative at a
equal to cla c2, and we infer from (18) that W ( a ) = cla c2. Therefore, W ( 8 )
is continuous from the right at a . A similar argument works from the left at ,8 and
completes the proof of Fact 2.
Fact 3 follows from the observation that both a,,/n2 and b,/n2 are o(1) as n + oo.
The Riemann function associated with these coeficients is

C (a,,cos n8 n4+ b, sin n6)

00

H(6) = -

n=l

If W ( 8 ) -c/2, it follows from Rajchman's lemma and (7) that c Q 2 / 4 H ( 8 ) is a

linear function. But then kemann's localization theorem [4,p. 3301 establishes Fact 3.
In the proof of Fact 1, we also need the localization theory of trigonometric series.
For the details, we refer the reader to [3].In closing, we point out that the theorem
is, in addition, best possible from a third point of view; namely, the Lm-norm (= the
uniform norm) on concentric circles of radius r cannot be replaced by the LP-norm
with p c 00.

44

@ T H E MATHEMATICAL ASSOCIATION O F AMERICA [Monthly 109

ACKNOWLEDGEMENT. The author would like to thank the referee for his helpful suggestions

REFERENCES
1. M. Kline, Mathernatical Tlzouglzt froin Ancient to Moderiz Times, Oxford University Press, New York,
1972.
2. W. F. Osgood, Note on the functions defined by infinite series whose terms are analytic functions of a
complex variable; with corresponding theorems for definite integrals, Anizals of Math. (2)3 (1901-02)
25-34.
3. V. L. Shapiro, The uniqueness of functions harmonic in the interior of the unit disk, Proc. London Math.
Soc. 13 (1963) 639-652.
4. A. Zygmund, Trigonoinetric Series, vol. 1, Cambridge University Press, Cambridge, 1959.

VICTOR SHAPIRO served, during World War 11, as a frontline combat inedic with the 132nd Infantry Regiment in the South Pacific (Bougainville, Leyte, Cebu). Currently a nonlinear analyst, he wrote his Ph.D. thesis
at the University of Chicago on multiple trigonometric series with A. Zygmund. Shapiro's research work is
surveyed in Contemporaiy Mathematics, volume 208, a book in his honor. He walks 5 miles (almost) every
day and enjoys ballroom dancing with his wife Flo.
University of California, Riverside, Riverside CA 92521
shapiro @rlmth.ucr.erlu

OH CALCULUS, OH CALCULUS
(To: "Oh, Christmas Tree")
Oh, Calculus; Oh. Calculus,
How different seem thy branches.
Oh, Calculus; Oh, Calculus,
How different seem thy branches.
Derivatives tell us the rate,
For areas we integrate.
Oh, Calculus; Oh, Calculus,
How different seem thy branches.

Oh, Integral: Oh, Integral,

Partitions getting finer.

Oh, Integral; Oh, Integral,

Partitions getting finer.

Add more and more of less and less,

The errors disappear, we guess.

Oh, Integral; Oh, Integral,

Partitions getting finer.

Derivative, Derivative,
The limit your foundation.
Derivative, Derivative,
The limit your foundation.
A quotient, both parts growing nil,
Behold you reach a value still.
Derivative, Derivative,
The limit your foundation.

Oh, Calculus; Oh, Calculus,

United are thy branches.

Oh, Calculus; Oh, Calculus,

United are thy branches.

Because of that eternal gem,

The Fundamental Theorem.

Oh, Calculus; Oh, Calculus,

United are thy branches.

by Leon Hall and Ilene Morgan

University of Missouri-Rolla

January 20021

HARMONIC FUNCTIONS IN THE UNIT DISK