Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

42 I.

INTRODUCTION

9. Harmonic Functions and the Poisson Integral

For a fixed point y in R",n 2 2, we verify by differentiation that the


function
lY12 - 1x1'
Ix - Yl"
is a solution to the Laplace equation for all x # y . The computation is easy
but tedious; it is convenient to simplify by the change of variable z = x - y,
obtaining the function

---
-1 2(G Y ) .
1z1"-2 14" '
the first term is known to be harmonic, and only the second need be differenti-
ated.
Letf(y) be a continuous function defined on the surface of the unit sphere
Iyl = 1;for 1x1 < 1 we consider the Poisson Integral

which surely exists, the integrand being continuous on Iyl = 1. It is also clear
that we may differentiate the integral as many times as we like under the
integral sign; accordingly u(x) is a C"-function in 1x1 c 1 and satisfies the
Laplace equation Au = 0. If we consider the special case when f(y) = 1, then
u(x) is a function of radius, harmonic in the sphere for which u(0) = 1. Since
the harmonic functions which depend only on radius are all of the form
+
u(x) = A Elx12-" (or u(x) = A -i- E log 1x1 when n = 2), we see that u(x) = 1
inside the sphere. We use this fact to show that the Poisson integral takes on
the boundary valuesf(y). Choose z so that lzl = 1. Now,

We write the integral as a sum of two terms, the first being the integral taken
over ly - zI > 6, the second for Iy - zl S 6. If M = supIf(y)l, Iyl = 1, then
the first term in absolute value is bounded by
9. HARMONIC FUNCTIONS AND THE POISSON INTEGRAL 43

On the other hand, given a small E , there exists 6 so that for ly - zl < 6 we
have I f ( y ) -f(z)l < E , the second term is therefore bounded by

Thus, finally, lu(rz) - f ( z ) l 6 E + 2 M ( 1 - r2)(2/S)”,which is < 2 for


~ r suffi-
ciently close to 1 uniformly on JyI = 1. Therefore, if we define u(y) = f ( y ) on
the boundary of the sphere, we obtain a function harmonic inside the sphere,
and continuous on the closed sphere.
By a simple change of variables, we obtain the Poisson integral when we
are concerned with a sphere of radius R:

(in this formula, of course, (yl = 1). We have u(Ry) =f(y).


We have already established a mean value property for functions which
are solutions to the differential equation Au = 0 and are therefore led to the
following definition.
A function u(x) defined in an open subset G of R”is harmonic there if and
only if
(i) u(x) is locally integrable, that is, u(x) is integrable over any compact
subset of G, and
(ii) for every x in G, and any sphere S(x, r ) with center at x and radius r
which is wholly contained in G,

Note that the first hypothesis is needed so that the second makes sense.
If u(x) is a function defined in G and C 2 there for which Au = 0, we have
already seen that it satisfies a mean value theorem; that is, for any xo in G and
sufficiently small r
u(xo) = j u ( x o + rz) do(z).
We multiply this equation by r”-l and integrate over 0 < r < R to obtain

uOR”= /loR
u(xo + rz)r“-l dr dw(z)
n
44 I. INTRODUCTION

Thus u(x) is harmonic in the sense of our definition, which uses volume integrals
and not surface integrals.
We remark first that if u(x) is harmonic in G, it is continuous there, for if
we suppose that xk is a sequence in G converging to xo in G, the distances from
xk to the boundary of G are uniformly bounded from below, and we may
suppose such distances 2 2 r for some small positive r. We consider, then, the
sphere of radius 2r about xo and take k so large that S ( x k , r ) is a subset of
S ( x o , 2r). Then, writing x k ( X ) as the characteristic function of S ( x , , r ) we have

and

and the Lebesgue convergence theorem may be invoked to show that u(xk)
approaches u(xo); the integrands xk(y)u(y) converge pointwise almost every-
where to xo(y)u(y) and are uniformly bounded in absolute value by the inte-
grable function IH(y)u(y)I where H ( y ) is the characteristic function of
S(x0 2 4
9 4

We select next a point xo in G and a sphere S ( x o ,R) about it which is


wholly contained in G. We may change coordinates so that xo appears as the
origin. We form the Poisson integral for the sphere of radius R determined by
the values of u(x) on that surface

We consider next the difference, u(x) = w(x) - u(x) in the closed sphere
1x1 5 R.Since each of the terms is continuous in the closed sphere, this function
is also continuous and vanishes on the boundary. Suppose M = sup v(x) for
1x1 S Rand that M > 0. There is then a point zo where u(zo) = Mand (zo(< R.
Choose r = R - (zoIand note that
1
0 -
ISrI SS(.a.,) M - V(Y) d y = M - ~(20)= 0,

whence u(y) = M almost everywhere in S(zo, r ) ; from the continuity of v(y),


the equation even holds everywhere in that sphere. Clearly, there is a sequence
of points in that sphere approaching the boundary and the corresponding
values of the continuous u(y) must approach zero, contradicting M > 0. A
similar argument shows that the minimum of u(x) in the sphere is 0, whence
u ( x ) = w(x). But w(x), a Poisson integral, is known to be C" and a solution of
Aw = 0 inside the sphere; hence u(x) has the same properties. Since the sphere
was about an arbitrary point xo in G, it follows that the harmonic function
u(x) is C" in G and satisfies the differential equation Au = 0.
We prove two well-known results about harmonic functions in R".
9. HARMONIC FUNCTIONS AND THE POISSON INTEGRAL 45

Liouville's Theorem: Let u(x) be harmonic in all of R" and bounded


there by M : then u(x) = constant.

PROOF: We choose a point xo and estimate lu(xo) - u(0)I.We take the


spheres S(0, R ) and S(xo, R), where R > 41xol, and write their characteristic
functions xo, x , to obtain

the integrand vanishes inside the sphere about the origin of radius R - lxol
+
and outside the sphere of radius R IxoI, and so is supported by a set of
measure at most

by the mean value theorem, this is bounded by o,(R + Ixol)"-'21xol. It follows


that

a quantity which converges to 0 with increasing R . Since R may be taken


arbitrarily large, it follows that u(x) = constant = u(0).

Schwarz Reflection Principle: Let u ( x ) be defined and continuous in


the closed hemisphere 1x1 S 1, x, 2 0. Suppose u(x) = 0 for x, = 0 and that
u(x) is harmonic in the open hemisphere. Then u(x) may be extended to the
lower hemisphere by the equation

~ ( ~ 1 x 3Z , * * * , x n - l ? x n ) =
-u(xI,x2,..*,~,-1,
-xn)

to obtain a function continuous in 1x1 5 1 and harmonic in 1x1 c 1.

PROOF: Given x in R", we write x' for the reflection, that is, the point
with the same coordinates except x; = -x, . Our extension formula then is
conveniently written u ( x ) = -u(x'). Let f ( y ) be the restriction of u(x) to
1x1 = 1, x, 2 0; we extend f(y) by the equation f ( y ) = - f ( y ' ) to obtain a
function defined and continuous on the sphere lyl = I . We form the Poisson
integral of the extended function
46 I. INTRODUCTION

and note that this function satisfies the equation w(x') = - w(x) since 1 - lx'I2
= I - 1x1' and Ix' - yl = Ix - y'l; thus, w(x') is the Poisson integral off(y')
which is the negative off(y). From this equation it also follows that w(x)
vanishes when x, = 0 (since then x = x'). Thus, in the upper hemisphere, the
harmonic function w(x) has the same boundary values as the harmonic func-
tion u(x) ;their difference, which is also harmonic, vanishes on the boundary,
is continuous in the closed hemisphere, and cannot have a maximum or
minimum inside by an argument which we have already used. Thus that dif-
ference vanishes identically and u(x) and w(x) coincide in the upper hemisphere.
Hence u(x) is extended by w(x) as asserted in the Reflection theorem.
In Section 1 we made the important observation that a family of functions,
analytic in some region G of the complex plane and uniformly bounded there
was an equicontinuous family. The same result holds in a more general context:
we consider an infinite sequence uk(x) of functions harmonic in a region G of
R" and uniformly bounded there by the constant M.Let K be a compact subset
of G, d the diameter of K, and R the distance from K to the boundary of G.
[f x 1 and x2 are any two points of K,the estimate used in the proof of Liouville's
theorem shows that if Ixl - x2 I is sufficiently small, say smaller than R/4, then

and the coefficient of Ixr - x2 I above is then a uniform Lipschitz constant for
functions in the sequence relative to the compact K. Since G is a countable
union of such compacts, there exists a subsequence uk,(x) converging uni-
formly on all compact subsets of G to a limit u(x) which is necessarily con-
tinuous. The limit is even harmonic, since the mean value property for u(x) is
an immediate consequence of the mean value property for the functions of
the sequence via the Lebesgue convergence theorem.
When a function u(x) is harmonic in a region G and is also nonnegative,
more can be said.
Suppose u(x) is a nonnegative, harmonic function in the ball 1x1 < R + E ;
from the Poisson integral representation for u, we have for all x in 1x1 < R ,

since u(0) = j u ( R y ) dw(y) and R - 1x1 is a lower bound for the distance from x
to the sphere lyl = R. This is called Harnack's inequality and it enables us to
establish the following remarkable theorem.
9. HARMONIC FUNCTIONS AND THE POISSON INTEGRAL 47

Theorem: Let uk(x) be a sequence of positive harmonic functions in a


region G and xoa point of G such that the values U k ( x 0 ) are bounded: then there
exists a subsequence uk,(x) converging uniformly on compact subsets of G to
a positive, harmonic limit in G.

PROOF: It is enough to show that the sequence is uniformly bounded


on any compact subset of G which is connected and which contains xo. If K
is such a set, let r be smaller than one quarter of the distance from K to the
boundary of G; we can cover K by a finite number of overlapping balls of
radius r. From the Harnack inequality, the boundedness of the sequence of
functions at some point of such a ball implies its boundedness at all other
points of the ball, indeed, the sequence is uniformly bounded on the ball.
Since there are only finitely many balls in the covering, the sequence is uni-
formly bounded on K.

Theorem: The most general function u(x), positive and harmonic in the
ball 1x1 < R is of the form

where v is a positive Radon measure on the sphere Iyl = 1.

PROOF: Let Rk be a sequence of positive numbers converging increas-


ingly to R; for 1x1 < Rk we have

The measure dvk(y) = u(Rky)d o ( y ) is positive and has total mass u(0); from
Helly's theorem it follows that there is a weakly convergent subsequence of
these measures, that is, there exists a positive Radon measure vo on Jyl = 1
such that the integrals j f ( y ) dvk,(y)converge to / f ( y ) dvo(y) for all functions
f ( y ) continuous on the sphere. Since the Poisson kernel is continuous on the
sphere, the formula of the theorem is then valid.
The Poisson integral is particularly useful in the study of harmonic and
analytic functions in the complex plane, and it is then convenient to write the
integral in terms of polar coordinates. If z = reieis a point in the disk lzl < 1
and y is a point on the boundary, that is, y = e'W,the kernel
1 - lzl2 1 - r2
becomes
Iz - YI2 1 + r2 - 2r cos(e - o)
48 I. INTRODUCTION

and the Poisson integral is written


u(z) = u(reie)
= -1J 2n 1 - r2
211 1 + r 2 - 2r cos(e - w ) u(e'O) d o .
From the previous theorem we have the following corollary.

Corollary: The most general function u(z) positive and harmonic in the
disk IzI < 1 is of the form

where v is a positive Radon measure on the circle IzI = 1.

Another easy and useful result is the following.

Theorem: The most general functionf(z), analytic in the disk IzI < 1
and having a positive real part there, is of the form

where Cis the constant Im[f(O)] and v a positive Radon measure on the circle
IZI = 1.

+
PROOF: The functionf(z) = u(z) io(z) has positive real part and so
the harmonic u(z) is the Poisson integral of a positive measure v . Since the
+
real part of (e'" z)/(ei" - z) is the Poisson kernel, the formula determines
+
an analytic function U(z) iV(z) for which U(z) = u(z). It follows that the
function so obtained differs from f(z) by a constant and the theorem readily
follows.

At the end of Section 2, we obtained a canonical factorization for functions


analytic and bounded in the disk. This is now easily extended to the following
representation.

Theorem: The most general functionf(z), analytic and bounded in the


disk IzI < 1, is of the form
10. SMOOTH FUNCTIONS 49

where Cis a constant, 1 a nonnegative integer, B ( z ) a Blaschke product, and v


a positive Radon measure on the circle.
A proof is scarcely necessary; the function g(z) occurring in the canonical
factorization off(z) has a negative real part.

10. Smooth Functions

In most of this section we consider functions .f(x) defined on the real axis
and C" there, that is, having continuous derivatives of all orders. Such func-
tions need not be analytic, as the example which we shall now construct shows.
Let
f(x) = 0 for x 5 0
= exp( - l/x) for x > 0;

it is clear thatf(x) is C" everywhere on R', except perhaps at the origin. The
function is clearly continuous at the origin, since limx,,f(x) = 0. We next
invoke the following eleme'ntary lemma.

Lemma: Letf(x) and g(x) be continuous function on the real axis such
that g(x) is the derivative off(x) at all points x different from 0; thenf'(0)
exists and equals g(0).

PROOF: By the mean value theorem, the functionf(x) is Lipschitzian


in the interval [O, 11 sincef(x) - f ( O ) =f'(O.u)x (the mean value theorem only
requires that the derivative exist in the open interval (0, x)) and therefore the
difference is bounded in absolute value by llgllmlxl.A similar argument shows
that the function is Lipschitzian i n [ - 1,0] and hence, it is absolutely con-
tinuous in [ - 1, 13 and therefore the indefinite integral of its derivative. We
infer that

f(x) = f(- 1) + do d t
-1

and since g(t) is continuous, f ( x ) is differentiable at the origin with derivative


do) *
Now let r(x) be any rational function of x, f ( r ) , the function introduced
above. We have
lim r ( x ) f ( x ) = 0 ,
x-0

You might also like