MATH 566 LECTURE NOTES 1: HARMONIC FUNCTIONS

TSOGTGEREL GANTUMUR
1. The mean value property
In this set of notes, we consider real-valued functions on two-dimensional domains, although
it is straightforward to generalize the results to, e.g., vector-valued functions dened on n-
dimensional domains. The symbol D
r
(z) denotes the open disk with radius r and centre z,
and we set D
r
= D
r
(0). Unless otherwise specied, in what follows will be an open subset
of R
2
.
Denition 1. Let R
2
be open. Then a continuous function u C() is said to have the
mean value property and written u MVP(), if it satises
u(z) =
1
r
2
_
Dr(z)
u(w)dw, for all D
r
(z) .
It is easy to show (by dierentiating and integrating with respect to r) that the above
condition is equivalent to
u(z) =
1
2
_
2
0
u(z + r cis )d, for all D
r
(z) ,
where we have used the shorthand notation cis = (cos , sin) R
2
. Let us also introduce
the notation Har() = u C
2
() : u = 0 for the set of harmonic functions in .
Theorem 2. Har() MVP().
Proof. Let u Har(), and let D
r
(z) and (0, r). Then obviously the following
integral must vanish:
_
D(z)
u =
_
D(z)

udl =
_
2
0

u(z + cis )d =

_
2
0
u(z + cis )d,
where commuting the derivative

with the integration over is justied because the functions

(, ) u(z + cis ) and (, )

u(z + cis ) are both continuous. Now from the

fundamental theorem of calculus we have

_
2
0
u(z + cis )d = 0
_
2
0
u(z + r cis )d 2u(z) = 0.

So viewed as a tool, the mean value property can be used to prove properties of harmonic
functions. The following converse shows that the mean value property can also be used to
prove harmonicity.
Theorem 3. MVP() Har() and MVP() C

().
Date: September 27, 2010.
1
2 TSOGTGEREL GANTUMUR
Proof. Let C

0
(D
1
) be a positive function with
_
= 1, which is also radial, i.e., (z) =
([z[) for some . The normalization of implies 2
_
1
0
r(r)dr = 1. Set

(z) =
2
(z/)
for > 0, and assume that z and < dist(z, ). Then
_

u(w)

(w z)dw =
_

u(z + w)

(w)dw =
1

2
_
|w|<
u(z + w)(
[w[

)dw
=
1

2
_
r<
_
2
0
u(z + r cis )(
r

)rdrd =
2u(z)

2
_
r<
(
r

)rdr = u(z),
which means that u =

u, and so u C

().
Now let D
r
(z) , and compute
_
Dr(z)
u(w)dw = r
r
_
2
0
u(z + r cis )d = r
r
(2u(z)) = 0.
Since D
r
(z) is arbitrary, we conclude that u = 0 in .
2. Simple consequences of the mean value property
Recall that an open set is connected if it cannot be decomposed as the disjoint union of
two open sets.
Theorem 4 (Maximum principle). Let R
2
be open, and let u C
2
() be a subharmonic
function in , i.e., u 0 in .
(a) If is connected and u(z) = sup

u at some z , then u is constant.

(b) If is bounded and u C(

), then we have sup

u max

u.
Proof. For part (a), let M = u(z) = sup

u. Then we can write the mean value property as

_
2
0
[u(z) u(z + r cis )]d = 0, for all suciently small r > 0.
The integrand is a nonnegative continuous function, so it must vanish, which implies that
u(w) = M in a small disk w D
r
(z). In other words, the set = z : u(z) = M is
open. Since u is continuous, the set z : u(z) < M = is also open. This means
that either = or = , but is not empty by hypothesis.
For part (b), there is z

with u(z) = sup

u since is bounded and u is continuous on

. If z then we are done; otherwise applying part (a) to the connected component of
that contains z concludes the proof.
Remark 5. In part (b) of the above theorem, the condition that be bounded cannot be
simply removed, as can be seen from the example u(x, y) = e
x
cos y on (0, ) (

2
,

2
).
In the following theorem we use complex notation for convenience.
Theorem 6 (Schwarz reection principle). Let C be an open set symmetric with respect
to the real axis, and let
+
= z > 0 be the part of it in the upper half plane. Moreover,
assume that u Har(), and that u(z) 0 as z
+
tends to any point on the real axis
z = 0. Then u extends to be harmonic on , and the extension satises
u( z) = u(z), z .
MATH 566 LECTURE NOTES 1: HARMONIC FUNCTIONS 3
Proof. Let

= z : z
+
, and extend u to a function on by u(z) = 0 for z z =
0 and u(z) = u( z) for z

. Then the mean value property is obviously satised at points

z
+

. But for (the remaining) points z with z = 0, since u(z) = u( z), the integral
of u over any small circle centred at z is zero, giving the mean value property for those
points.
For R
2
and a function u C(), we dene the uniform norm
|u|

= sup

[u[.
We say that a sequence of functions u
k
C() converges uniformly on to u C(), if
|u
k
u|

0. Recall that continuity and the Riemann integral are preserved under uniform
limits.
Theorem 7. If u
k
Har() converges uniformly on to u, then u Har().
Proof. We have u C(). Let D
r
(z) . Then we have
2u
k
(z) =
_
2
0
u
k
(z + r cis )d, for all k.
Since the left hand side converges to 2u(z), and the integrand converges uniformly to
u(z + r cis ), the mean value property over D
r
(z) is satised for u, and since D
r
(z) is
arbitrary we have u Har().
3. Derivative bounds
Noting that partial derivatives of harmonic functions are also harmonic, and by using the
mean value property for the partial derivatives, we can bound the derivatives of harmonic
functions by the size of the function itself. Recall that for = (
1
,
2
) with [[ = 1, the
directional derivative along is dened by

u =
1

x
u +
2

y
u.
Theorem 8. Let u Har() and

D
r
(z) . Then for any R
2
with [[ = 1, we have
[

u(z)[
2
r
max
Dr(z)
[u[.
If in addition u 0 in

D
r
(z), we have
[

u(z)[
2
r
u(z).
Proof. Since u = 0 and u is smooth, we have

u = 0 in , i.e.,

u Har(). Using the

mean value property and the divergence theorem, we get

u(z) =
1
r
2
_
Dr(z)

u(w)dw =
1
r
2
_
2
0
u(z + r cis )( cis )rd.
Now if u 0, since the inner product [ cis [ 1, we infer

u(z)
1
r
_
2
0
u(z + r cis )d =
1
r
2u(z).
On the other hand, for general u, we have
[
x
u(z)[
1
r
max
Dr(z)
[u[ 2 =
2
r
max
Dr(z)
[u[.

The following can be called the Liouville theorem for harmonic functions.
Corollary 9. If u Har(R
2
) is bounded above or bounded below, then u is constant.
4 TSOGTGEREL GANTUMUR
Proof. If u a for some constant a R, then set v = u a, or if u b for some constant
b R, then set v = b u. In either case, we have v 0 and v Har(R
2
). Now let z R
2
and apply the above theorem to v, inferring
[

v(z)[
2
r
v(z), for any r > 0,
which gives

v 0 in any direction .
Now we estimate higher derivatives.
Corollary 10. Let u Har() and

D
r
(z) . Then for j, k 0 and n = j + k, we have
[
j
x

k
y
u(z)[
2
n
n
n
r
n
max

Dr(z)
[u[.
Proof. We prove it only for k = 0, since the general case is completely analogous. All the
derivatives
m
x
u are harmonic, so with = r/n, we have
[
n
x
u(z)[
2

max

D(z)
[
n1
x
u[ . . .
_
2

_
n
max

Dn(z)
[u[,
which is what we need.
4. Analyticity
Let us recall the Taylor theorem for two variables, with a somewhat simplied hypothesis
on the smoothness of the function involved.
Theorem 11. Let f C
n
(

D
r
(z)), and h = (h
1
, h
2
) D
r
(0). Then we have
f(z + h) =

j+k<n

j
x

k
y
f(z)
j!k!
h
j
1
h
k
2
+

j+k=n
R
j,k
(z)h
j
1
h
k
2
,
with the following estimate on the error term
[R
j,k
(z)[ sup
w

Dr(z)

j
x

k
y
f(w)
j!k!

.
Denition 12. A function f C

() is called (real) analytic at z if the following

Taylor series converges in a neighbourhood of z.
f(z + h) =

j=0

k=0

j
x

k
y
f(z)
j!k!
h
j
1
h
k
2
.
Moreover, f is said to be (real) analytic in if it is analytic at each z . The set of analytic
functions in is denoted by C

().
Theorem 13. Har() C

().
Proof. Let D
r
(z) and (0, r). Then we have
[R
j,k
(z)[ sup

D(z)

j
x

k
y
u
j!k!

2
n
n
n
j!k!(r )
n
max

Dr(z)
[u[
Using the crude bounds
_
n
j
_
2
n
and n
n
e
n
n!, we get
2
n
n
n
j!k!
=
_
n
j
_
2
n
n
n
n!
4
n
e
n
,
MATH 566 LECTURE NOTES 1: HARMONIC FUNCTIONS 5
and with this in mind, for [h[ the Taylor remainder term can be estimated as
[R
j,k
(z)h
j
1
h
k
2
[
_
4e
r
_
n
max

Dr(z)
[u[.
We see that the Taylor series converges in D

(z) whenever <

r
4e+1
.
As a consequence of analyticity, we have the following rigidity that is typical in analytic
setting. What is remarkable is that there is no condition on the size of D relative to ; we
only need D nonempty and open.
Theorem 14 (Identity theorem). Let u Har() with a connected open set, and let u 0
in an open subset D . Then u 0 in .
Proof. Each
j,k
= z :
j
x

k
y
u(z) = 0 is closed, so the intersection =

j,k

j,k
is also
closed. But is also open, because z implies that u 0 in a small disk centred at z by
a Taylor series argument. Since is nonempty by hypothesis, we conclude that = .
The following corollary records the fact that a harmonic function is completely determined
by its restriction to any open subset of its domain of denition. In other words, if it is at
all possible to extend a harmonic function (dened on an open set) to a bigger domain, then
there is only one way to do the extension.
Corollary 15. Let u, v Har() with a connected open set, and let u v in an open
subset D . Then u v in .