Solution of Dirichlet Problem — Perron’s Method

S. Kumaresan
School of Math. and Stat.
University of Hyderabad
Hyderabad 500046
[email protected]

1 Harmonic Functions and Their Properties

Let Ω be a nonempty open subset of R2 or C.

Definition 1. A real valued function u : Ω → R is said to be harmonic in Ω if u is in C 2 and
uxx + uyy = 0.
Theorem 2. If f : Ω → C is holomorphic and f = u + iv then u and v are harmonic.

Proof. By Cauchy Riemann equations ux = vy and uy = −vx , so that uxx = vxy = −uyy , or,
uxx + uyy = 0. Apply the same argument to −if to conclude that v is harmonic.

Theorem 3. If u is harmonic in Ω, then

(1) ux is the real part of an holomorphic function in Ω.
(2) If Ω is simply connected them u is the real part of an holomorphic function in Ω.

Proof. Let f = ux − iuy . Since u ∈ C 2 , f is C 1 .Further

fy = uxy − iuyy = uyx + iuxx = ifx .

Thus f satisfies the Cauchy-Riemann equations and hence f is holomorphic in Ω. (1) is thus
proved.
If Ω is simply connected, there is a primitive F for f = ux − iuy . Let F = U + iV .

F 0 (z) = Ux + iVx = Ux − iUy .

But
F 0 (z) = f (z) = ux − iuy .
Comparing the two expressions of F 0 we get Ux = ux and Uy = uy . Hence U (x, y) =
u(x, y) + C. Thus u is the real part of the holomorphic function F − C in Ω. This proves
(2).

Corollary 4. 1.3 If u is harmonic in Ω and h : Ω0 → Ω is holomorphic, then u ◦ h : Ω0 → R

is harmonic.

1
Proof. Fix h(z) = z 0 ∈ Ω. Let u = <φ in a neighbourhood of z 0 , where φ is holomorphic. Then
φ ◦ h is holomorphic in a neighbourhood of z and u ◦ h = <(φ ◦ h) in the same neighbourhood.
Hence u ◦ h is harmonic.

Theorem 5 (Mean Value Property for Harmonic Functions). Let u be harmonic in the disk
B(a, R) of radius R with centre a. Then
Z 2π
1
u(a) = f (a + reiθ )dθ
2π 0

for any 0 < r < R.

Proof. Let u = <f , f holomorphic in B(a, R). then

Z 2π
1
f (a) = f (a + reiθ )dθ,
2π 0

by the mean value theorem for holomorphic functions. Taking real parts we get the result.

Theorem 6 (Maximum Principle for Harmonic functions). If u is a non constant harmonic

function in a connected open Ω, then u has no maximum or minimum in Ω.

Proof. Assume that f is non constant, harmonic in Ω and that it attains its maximum M at
a point a in Ω. By the mean value property
Z 2π
1
M = u(a) = u(a + reiθ )dθ,
2π 0

for any r such that the closed disk of radius r with centre a lies entirely in Ω. Since u is
continuous and M ≥ u(a + reiθ ) we have M = u(a + reiθ ), 0 ≤ θ ≤ 2π. Thus u is the constant
M in the disk with centre a and lying in Ω. Thus the set of points in Ω where u equals M
is non-empty, and open as well as closed. Since Ω is connected we see that u is constant
throughout Ω. This is a contradiction since f is non constant.
Applying the above argument to −u proves the assertion concerning the minimum.

Ex. 7. Prove Thm. 6 using the open mapping theorem for holomorphic functions and the
fact that any harmonic function is locally the real part of a holomorphic function.

Remark 8. Compare and contrast Thm. 6 with the maximum (modulus) principle for a
holomorphic function. In the latter case the obvious version of the minimum modulus principle
has to be modified.

Corollary 9. Let w : Ω → R be continuous where Ω is connected and open and Ω is compact.

Assume that w is harmonic in Ω. Then w attains its maximum M and minimum m on the
boundary ∂Ω of Ω. Further if u and v are two continuous real valued functions on Ω, both
harmonic in Ω, and if u = v on the boundary on Ω, then u = v on Ω.

2
Proof. If w is constant then clearly w attains its maximum and minimum on the boundary
of Ω. Assume therefore that w is non constant. Since w is continuous on compact Ω, w
attains its maximum M and minimum m in Ω. Since w is non constant and harmonic in
Ω, the maximum and minimum of w are not attained in Ω. Hence they are attained in ∂Ω.
This proves the first part of the Corollary. To prove the second part put w = u − v, which is
continuous in Ω, harmonic in Ω, and vanishes on ∂Ω. Clearly by the first part of the corollary
the maximum and the minimum of w are both zero in Ω, hence u = v in Ω.

Ex. 10. Show that the hypothesis of compactness of ∂Ω in the above corollary can not be
dropped by exhibiting a continuous function on the closed upper half plane which is harmonic
in the open upper half plane and which does not attain its maximum or the minimum on the
closed upper half plane.

Corollary 9 motivates the following question: Given f ∈ C(∂Ω) does there exist u ∈ C(Ω)
which is harmonic in Ω and agrees with f on ∂Ω?
This is known as the Dirichlet problem.
Remark 11. If the domain Ω does not have a compact closure the solution may not be
unique. Consider, for example, Ω = {z : =z ≥ 0} and the function f (z) = 0 for real z. Then
1
u1 (z) = =e z on Ω = {z : =z > 0}, u1 (z) = 0 for real z, and, u2 (z) = 0 for all z ∈ Ω, are
two distinct functions, both harmonic in the open upper half plane, continuous on the closed
upper half plane and which agree with f on the real axis, the boundary of Ω. Indeed a simpler
function v(z) = v(x + iy) = y also serves as an illustrative example. It is also a nonconstant
(but unbounded) harmonic function on the upper half plane and vanishes on the boundary.
Remark 12. There exist Ω and a continuous bounded real valued function on ∂Ω such that
the associated Dirichlet problem has no solution. Consider Ω = {z : 0 < |z| ≤ 1}. Then
∂Ω = {|z| = 1} ∪ {0} Let f (z) = 0 if |z| = 1 and f (0) = 1. The associated Dirichlet problem
for this data has no solution. If possible let u be a solution of the Dirichlet problem for
boundary value f . Then by the removable singularity theorem (see Thm. 13) u is harmonic
in the unit disk. But by the mean value property of harmonic functions u(0) = 0 which is a
contradiction. Thus the Dirichlet problem has no solution in this case.
Theorem 13. If u is harmonic and bounded in a deleted neighbourhood of z0 , then u(z0 ) can
be so defined that the resulting new function u is harmonic in all of the given neighbourhood.

Proof. Without loss of generality assume that z0 = 0 and 0 ≤ |z| < R is the neighbourhood
of 0 in which u is bounded and harmonic. For the boundary function u |∂B(0,R) we have a
harmonic function h on B(0, R) which has a continuous extension up to the boundary and
which agrees on the boundary with u. If we can show that u ≤ h and h ≤ u on B(0, R) \ {0},
it will follow that u = h in B(0, R) \ {0} and thus u can be defined at 0 to be h(0).
Let v = (u − h) or v = (h − u). Consider, for ε > 0,

vε (z) = v(z) + ε log(|z|/R), 0 < |z| < R.

Then vε is harmonic on 0 < |z| < R and vε (z) = 0 on |z| = R. As v is bounded

lim supz→0 vε (z) ≤ 0. Hence vε (z) ≤ 0 on 0 < |z| ≤ δ for some δ. Since vε is harmonic
in the annulus δ ≤ |z| ≤ R and non-positive on the boundary of the annulus, we have vε ≤ 0

3
in the annulus. Letting δ → 0 we see that vε ≤ 0 on the punctured disk 0 < |z| < R. Letting
ε → 0 we see that v ≤ 0. Since v was any of the functions u − h or h − u, we see that h = u
and the theorem is proved.

Before going any further let us remark that the concept of harmonic function on any
Riemann surface X can be defined in an obvious way:
We say that u : X → R is harmonic if for any co-ordinate chart (U, z), the function
f ◦ z −1 : z(U ) → R is harmonic. This is equivalent to saying that with respect to local
co-ordinates z = (x, y), uxx + uyy = 0, i.e., u satisfies Laplace equation on X locally. Note
that these notions are well-defined by the holomorphic compatibility of the charts.
The maximum principle also continues to be true for harmonic functions on a Riemann
surface. For let u be real valued and harmonic in a connected open set Ω of a Riemann surface
X. Suppose u attains its maximum at a ∈ Ω. Let (U, z) be a co-ordinate chart at a with
closure lying in Ω. Then u ◦ z −1 is harmonic in z(U ) and attains its maximum at z(a). By
maximum principle for harmonic functions on the plane we see that u ◦ z −1 is constant on
z(U ), hence u is constant on U . One can now proceed as before and conclude that the set of
points on Ω where the maximum is attained is open and closed and since it is nonempty it is
all of Ω, Ω being connected. Again, as before, if Ω ⊂ X is a connected domain with compact
closure, u and v are continuous real valued functions on Ω, harmonic in Ω and if u = v on
∂Ω, then u = v on all of Ω. We thus see that the Dirichlet problem for a connected domain
Ω with compact closure has at most one solution.

2 Dirichlet Problem in the Unit Disk

In spite of the negative results in the last section we shall show in this section that the
Dirichlet problem has a solution in any disk. In fact, we establish this for the unit disk. The
general case, in fact, the solvability of Dirichlet problem for any domain biholomorphic to the
unit disk, follows from this. (Not quite true; see Remark 11 and Remark 19.)
Let S 1 = ∂B(0, 1) = {z : |z| = 1}. Given f ∈ C(S 1 ), we wish to find u ∈ C(B(0, 1)) such
that ∆u = 0 in B(0, 1) and u |S 1 = f .
To motivate the construction of such a u, assume that such a u exists. Then by the mean
value property we have Z 2π
1
u(0) = f (eiθ )dθ.
2π 0
Now we want to know u(z) for other values of z ∈ B(0, 1). Consider the autmorphism of the
w+z
unit disk which takes 0 to z, T w = 1+zw . Then T maps B(0, 1) onto B(0, 1) biholomorphically
1
and T is a homeomorphism of S . Thus u ◦ T (w) is continuous on Ω and harmonic in Ω.
Hence, by the mean value property of harmonic functions,
Z 2π
1 eiθ + z
u(z) = (u ◦ T )(0) = f( )dθ.
2π 0 1 + zeiθ

4
We make a change of variable eit = T (eiθ ) or eiθ = T −1 (eit ). A calculation shows that
1 − |z|2 1 − |z|2
dθ = |(T −1 )0 (eit )|dt = dt = dt.
|1 − zeit |2 |eit − z|2
Thus we get the equation
2π 2π
eiθ + z
Z Z
1 1
u(z) = f( iθ
)dθ = f (eiθ )K(eit , z)dt, (1)
2π 0 1 + ze 2π 0

called the Poisson Equation, where

1 − |z|2
K(eit , z) = .
|eit − z|2
K is called the Poisson Kernel, more often written as P (eit , z) a notation we shall use whenever
convenient.
it
It is easy to see that K(eit , z) is harmonic in z being the real part of the function eeit −z
+z

which is holomorphic in z. Now the right hand side of Eq. 1 makes sense whether u in question
exists or not. Its equality to the middle term is only an application of the change of variables
formula. The middle term is a continuous extension of f inside the unit disk and the right
hand side at once shows that the extension is harmonic there, as we see below.
Theorem 14. Let f ∈ C(S 1 ) and define, for z ∈ B(0, 1),
Z 2π
1 z + eit
u(z) = f( )dt.
2π 0 1 + zeit
Then u is harmonic in B(0, 1) and a continuous extension of f into the open unit disk.

Proof. As seen above, the function u is the real part of the function
Z 2π
1 eit + z
g(z) = f (eit ) it dt.
2π 0 e −z
Now g is holomorphic in the unit disk as we see by Morera’s theorem or by differentiating
under the integral sign. Hence u, being the real part of g, is harmonic in the unit disk. To see
z+eit iθ
that u is a continuous extension of f inside the unit disk, we note that limz→eiθ 1+ze it = e

except when eiθ = −eit . Since f is continuous on the compact set S 1 , it is bounded on it. By
bounded convergence theorem
Z 2π
1 z + eit
lim u(z) = lim f ( )dt
z→eiθ 2π 0 z→eiθ 1 + zeit
Z 2π
1
= f (eiθ )dt
2π 0
= f (eiθ
which shows that u is a continuous extension of f into the open unit disk.

The above proof of the solution of the Dirichlet problem for the disk seems simpler than the
usual proof which uses a further property of the Poisson Kernel, i.e., its being an approximate
identity, a property which we explain below.

5
 Various forms of Poisson Kernel

The function
w+z |w|2 − |z|2
K(w, z) = <( )= ,
w−z |w − z|2
for w 6= z, is also called Poisson Kernel. For |w| = 1 it takes the form
1 − |z|2
K(eiθ , z) =
|eiθ − z|2
If we set w = eiθ and z = reit , 0 ≤ r < 1 Poisson kernel takes the form
1 − r2 1 − r2
K(eiθ , reit ) = = .
1 − 2rcos(θ − t) + r2 |1 − rei(θ−t) |2
When w = ei0 = 1, K takes the form
1 − r2
K(1, reit ) =
|1 − reit |2
1 + reit
= <( )
1 − reit
X ∞
= <(1 + 2 rn eint )
n=1
∞
X
= r|n| eint .
n=−∞

Poisson Kernel as an Approximate Identity

1−|z|2
(1) For 0 ≤ |z| < 1, K(eiθ , z) > 0 since K(eiθ , z) = |eiθ −z|2
.

(2) We also have

Z 2π
1
K(eiθ , z)dθ = 1.
2π 0
This follows from the residue theorem since the right hand side of the above equation is equal
to Z
1 w + z dw
<[ ] = 1.
2π |w|=1 w − z w
This may also be seen using the Fourier expansion
∞
X
iθ
K(e , re ) = it
r|n| eni(θ−t) .
−∞

(3) For 0 < δ < π2 ,

sup K(eiθ , reit ) → 0 as r → 1.
|θ−t|≥δ
1−r2
This is immediate since K(eiθ , reit ) ≤ cos2 δ/2
if |θ − t| ≥ δ.

6
Ex. 15. Prove Thm. 14 using the above three properties of the Poisson Kernel. Hint:
Observe that
Z 2π Z 2π 
1 iθ iθ 1 
u(z) − f (e )K(e , z) dθ = u(z) − f (eiθ ) K(eiθ , z) dθ.
2π 0 2π 0

Given ε choose δ by using the uniform continuity of f . Split domain of integration into two
parts: one over |θ| ≤ δ and the other over its complement in [−π, π]. Use the third property
of the kernel to estimate the integral over |θ| > δ.

Ex. 16. Let u be a continuous function on an open set Ω which satisfies the mean value
property, i.e., for every a ∈ Ω and for every r with B(a, r) ⊂ Ω,
Z 2π
1
u(a) = u(a + reit )dt.
2π 0

Show that u is harmonic in Ω. Hint: Let v be the solution of the Dirichlet problem for the
boundary data u |∂B(a,r) . Then v −u satisfies the minimum and maximum principles (because
of the mean value property).

Ex. 17. Let f be continuous on ∂B(a, r). Show that if |p − a| < r then
2π
|r|2 − |p − a|2
Z
1
u(p) = f (a + reit ) dt
2π 0 |a + reit − p|2

is the unique continuous and harmonic extension of f in B(a, r).

Ex. 18. Find a “Poisson integral formula” for the solution of the Dirichlet problem in the
upper half plane {z ∈ C : =(z) > 0} with a bounded continuous boundary data.

Remark 19. Go through Remark 11 (especially the first example) at the beginning of this
section. Try to see why we cannot conclude the uniqueness of the solutions in the upper half
plane even though the upper half plane is biholomorphic to the unit disk.

3 Harnack’s Inequalities and Harnack’s Principle

We have |w| − |z| ≤ |w − z| ≤ |w| + |z|. Using these we obtain estimates for K(w, z) (Recall
that |w| > |z|.):

|w|2 − |z|2 (|w| − |z|)(|w| + |z|) (|w − z|)(|w| + |z|) |w| + |z| |w| + |z|
K(w, z) = = ≤ = ≤ .
|w − z|2 |w − z|2 |w − z|2 |w − z| |w| − |z|

Also
|w|2 − |z|2 (|w| − |z|)(|w| + |z|) |w| − |z|
K(w, z) = ≥ = .
|w − z|2 (|w| + |z|)2 |w| + |z|
Hence
|w| − |z| |w| + |z|
≤ K(w, z) ≤ . (2)
|w| + |z| |w| − |z|

7
Let us use these inequalities along with the mean value property of harmonic functions. If u
is harmonic in B(a, R1 ),then for R < R1 we have by the mean value property of harmonic
functions: Z 2π
1
u(a) = u(a + Reiθ )dθ.
2π 0
Let us assume that u is a nonnegative function, harmonic in B(a, R1 ). Let 0 < r < R < R1 .
Then Z 2π
it 1
u(a + re ) = K(Reiθ , reit )u(a + Reiθ )dθ.
2π 0
Applying Eq. 2 and the mean value property, we get
R−r R+r
u(a) ≤ u(a + reit ) ≤ u(a).
R+r R−r
Thus we have obtained:

Theorem 20 (Harnack Inequalities). If u is continuous and non-negative in B(a, R) and

harmonic in B(a, R) then for any z = a + reit ∈ B(a, R)
R−r R+r
u(a) ≤ u(z) = u(a + reit ) ≤ u(a).
R+r R−r

These inequalities govern the growth of u(z) in a neighbourhood of a.

Theorem 21 (Harnack’s Principle). Let (un ) be a sequence of real valued harmonic functions
on a connected open set Ω in C. Assume that un (z) ≤ un+1 (z) for all n ≥ 1 and all z ∈ Ω.
If (un (z)) is bounded for some z ∈ Ω, then (un ) converges uniformly on compact subsets of Ω
and the limit function is harmonic in Ω.

Proof. Let E = {a ∈ Ω : (un (a)) is bounded.}. By hypothesis E is not empty. We show first
that E is both open and closed in Ω.
(1) E is open: Let a ∈ E. Choose R > 0 such that B(a, R) ⊂ Ω. Then, by Harnack’s
inequalities (applied to the nonnegative harmonic function un − u1 ),
R−r R+r
(un (a) − u1 (a)) ≤ un (z) − u1 (z) ≤ (un (a) − u1 (a))
R+r R−r
for all z ∈ B(a, R). Thus (un (z)) is bounded for all z ∈ B(a, R) and E is open.
(2) E is closed in Ω: Let a ∈ Ω be a cluster point of E. To show that a ∈ E, choose R > 0
such that B(a, R) ⊂ Ω. Let z0 ∈ E ∩ B(a, R/2). By Harnack’s inequalities

R − 12 R R + 12 R
(un (a) − u1 (a)) ≤ un (z0 ) − u1 (z0 ) ≤ (un (a) − u1 (a)).
R + 12 R R − 12 R

Since z0 ∈ E, i.e.,(un (z0 )) is bounded, we see that (un (a)) is bounded and a ∈ E. Thus E is
closed. Since Ω is connected we conclude that E = Ω.
Thus for any z ∈ Ω, (un (z)) is bounded and monotone. Hence it is convergent to a finite
value. To prove the uniform convergence of the sequence on compact subsets of Ω it is enough

8
to prove the uniform convergence on a neighbourhood of any a ∈ Ω. Let R > 0 be chosen
such that B(a, R) ⊂ Ω. Let 0 < r < R. Then for all z ∈ B(a, r)

R−r R+r
(un (a) − um (a)) ≤ un (z) − um (z) ≤ (un (a) − um (a)).
R+r R−r
Since (un (a)) converges to a finite real number, we see that (un ) converges uniformly on
B(a, r). Since each un satisfies the mean value property (being harmonic) we see that the
limit function u, which is continuous, also satisfies the mean value property. By Exer. 16 u is
harmonic. This proves the theorem.

4 Subharmonic Functions and Their Properties

Theorem 22. Let Ω be a connected open set the plane. Let v : Ω → R be continuous. Then
the following are equivalent:
(1) For every region D ⊂ Ω whose closure is compact and contained in Ω and for every
function u continuous on D and harmonic in D, if v ≤ u on ∂D then v ≤ u on D (hence
also on D).
(2) For every point p ∈ Ω and for every disk B(p, r) whose closure is contained in Ω we
have Z 2π
1
v(p) ≤ v(p + reiθ ) dθ.
2π 0
(3) For every point p ∈ Ω and for some disk B(p, ε) which lies in Ω, for any r, 0 < r < ε,
Z 2π
1
v(p) ≤ v(p + reiθ ) dθ.
2π 0

(4) For every region D ⊂ Ω, for every u harmonic in D satisfying v ≤ u, either v < u in
D or v = u in D.

Proof. (1) implies (2): For let B(p, r) be a disk whose closure lies in Ω and let w be the
harmonic function in B(p, r) which agrees with v on the boundary of B(p, r). Then
Z 2π
1
v(p) ≤ w(p) = v(p + reiθ ) dθ
2π 0

where the inequality on the left follows from (1) and the equality on th right is due to the
mean value property of the harmonic w.
(2) implies (3) is trivial.
(3) implies (4): Let u be as in (4) and let w = v − u. Then w ≤ 0 in Ω. Assume that
w(a) = 0 for some a ∈ Ω. We show that w = 0 in all of Ω. The set E = {p ∈ D : w(p) = 0}
is clearly non-empty and closed in Ω. We show that E is also open. Let ε > 0 be as in (3).
Since u is harmonic in Ω it satisfies the mean value property. In view of (3) we have for any
r, 0 < r < ε,

9
 Z 2π
1
w(p) = 0 = v(p) − u(p) ≤ [v(p + reiθ ) − u(p + reiθ )] dθ
2π 0
Z 2π
1
= w(p + reiθ ) dθ ≤ 0,
2π 0

since w ≤ 0 in D. Clearly w(p + reiθ ) = 0 for all θ and all r < ε, whence w = 0 in B(p, ε)
and E is open. Since Ω is connected, we see that E = Ω.
(4) implies (1): Let v, u and D be as in (1). Let w = v − u. Then w ≤ 0 on the ∂D. Let
m = max{w(p) : p ∈ D}. Since w is continuous on compact D, m is attained on this set. If
it is attained on the boundary of D then clearly v ≤ u on D since v ≤ u on ∂D. Otherwise
there is a point p0 ∈ D with w(p0 ) = m. If we replace u by U = u + m we see that U is
harmonic in D, v ≤ U in D and v(p0 ) = U (p0 ) for p0 ∈ D. By (4), v = U in all of D, hence
on D. Thus v − u = m on ∂D, and since v ≤ u on ∂D, m ≤ 0 whence v ≤ u on D. The
theorem is proved.

Definition 23. Any real valued continuous function u defined on a connected open set Ω is
said to be subharmonic if it satisfies any one (hence all) of the four conditions of the theorem
above.

The following theorem and proposition are immediate consequences of Def. 23 and Thm. 22.

Theorem 24 ( Maximum Principle for Subharmonic Functions). (1) (Weak Form) If v is

subharmonic in Ω and v ≤ M in Ω then either v < M or v = M .
(2) (Strong Form) If Ω is compact, v is continuous on Ω and subharmonic in Ω then v
attains its supremum on the boundary of Ω.

Ex. 25. A continuous function u : Ω → R is subharmonic iff for every domain D ⊂ Ω and
every harmonic function v on Ω, the function u + v has no maximum in D unless u + v is a
constant.

Proposition 26. We have:

(1) sum of two subharmonic functions is subharmonic, in particular the sum of a harmonic
function and a subharmonic function is subharmonic.
(2) nonnegative multiple of a subharmonic function is subharmonic.
(3) max{u, v} is subharmonic whenever u and v are subharmonic.

Poisson Modification

Let v be subharmonic in Ω. Let D := B(a, r) be a disk whose closure lies in Ω. Define Pv on

Ω, called the Poisson modification of v, as follows:
(
v(p), p ∈ Ω − B(a, r),
Pv (p) = 1 R 2π r2 −|a−p|2 it
2π 0 |a+reit −p|2
v(a + re ) dt, p ∈ B(a, r).

10
Pv depends on the disk D. When we wish to emphasize its dependence on D we shall write
PD,v or PD v in place of Pv .

Theorem 27. Poisson modification of v is subharmonic on Ω.

Proof. Note that Pv is harmonic in B(a, r) and the values of Pv on the boundary of B(a, r)
agree with those of v. (See Exer. 17. It is easy to see that Pv satisfies condition (3) of Thm. 22
on Ω−∂Ω. Moreover v−PV subharmonic on B(a, r), continuous up to the boundary of B(a, r)
where it vanishes. So by the strong maximum principle for subharmonic functions we conclude
that v ≤ Pv on B(a, r). Since v = Pv outside B(a, r), we see that v ≤ Pv on Ω. If p is a point
in ∂Ω and B(p, ρ) a disk with centre p and closure lying in Ω, then v ≤ Pv on the boundary
of this disk and we have:
Z 2π Z 2π
1 it 1
Pv (p) = v(p) ≤ v(p + ρe )dt ≤ Pv (a + ρeit )dt.
2π 0 2π 0

Thus Pv satisfies condition (3) of Thm. 22 at all points of Ω and Pv is subharmonic in Ω.

Ex. 28. Show that u : Ω → R is subharmonic iff u ≤ PD,u for every disk D ⊂ Ω.

Proposition 29. Let u : Ω → R be C 2 . Then u is subharmonic iff ∆u := uxx + uyy ≥ 0 on

Ω.

Proof. Since subharmonicity is a local property we may assume, without loss of generality,
that u is C 2 and ∆u ≥ 0 on Ω := B(0, 1).
First we assume that ∆u > 0. Then u cannot have a maximum in Ω. If false, let a ∈ Ω
2 2
be a point of maximum of u. Then, at a, ∂∂xu2 ≤ 0 and ∂∂yu2 ≤ 0 so that ∆u ≤ 0 at this point
a — a contradiction.
Now, if h is harmonic in Ω then ∆(u + h) > 0 and hence u + h has no maximum in
Ω. Again employing the arguments in the proof of Thm. 22 (or Exer. 25) we see that u is
subharmonic in Ω.
To deal with the general case, for any ε > 0 consider the function vε (x, y) := u(x, y) +
ε(x2 + y 2 ). Then ∆vε = ∆u + 4ε > 0. Hence vε is subharmonic by the first part. Hence for
every disk D ⊂ Ω we have PDvε ≥ vε . That is, we have

PD u + ε(x2 + y 2 ) ≥ PD u + PD (x2 + y 2 ) ≥ u + ε(x2 + y 2 ).

Letting ε → 0 we see that PD u ≥ u. Hence by Exer. 28, u is subharmonic.

The other way implication is left as an exercise.

This Proposition brings out the analogy between the harmonic functions and the “linear”
functions (from R to R) of the form ax + b and that of the subharmonic functions and the
(twice differentiable) convex functions characterised by f 00 ≥ 0. It is worthwhile keeping this
analogy in mind. In particular, notice that condition 1 of Thm. 22 is the geometric definition
of a convex function on R.

11
5 Subharmonic Functions on a Riemann Surface

Since the property of being harmonic or subharmonic is entirely a local property of a function,
it is possible to define such functions on any Riemann surface. We have already defined
harmonic functions on a Riemann surface above. A subharmonic function on a Riemann
surface is defined as follows.
Let X be a Riemann surface, and Ω a connected open subset of X. A continuous real
valued function v on Ω is said to be subharmonic if for every co-ordinate chart (U, z) with
U ⊂ Ω, v ◦ z −1 is subharmonic in z(U ). Let us call a co-ordinate chart (U, z) a parametric
disk with centre p ∈ U and radius r if z(U ) is a disk in the complex plane with centre z(p)
and radius r. The following theorem lists some useful equivalent definitions of a subharmonic
function on a Riemann Surface. The proof is similar to the proof of Thm. 22. We need a
notation. Let (U, z) be a parametric disk with centre p and radius r with closure of U lying
in Ω. By mean value of v on ∂U , denoted by Mv,U , we mean
Z 2π
1
v ◦ (z −1 z(p) + reit )dt.


Mv,U =
2π 0
Theorem 30. Let Ω be a connected open set of a Riemann surface X. Let v : Ω → R be a
continuous. Then the following are equivalent:
(1) For every region D ⊂ Ω whose closure is compact and contained in Ω and for every
function u continuous on D and harmonic in D, if v ≤ u on ∂D then v ≤ u on D (hence
also on D).
(2) For every point p ∈ Ω and for every parametric disk (U, z) with centre p, radius r and
whose closure is contained in Ω we have
v(p) ≤ Mv,U .

(3) For every point p ∈ Ω and for some parametric disk (U, z) with centre p and radius ε
which lies in Ω, for any r, 0 < r < ε, and W := z −1 ({ζ : |ζ − z(p)| < r}), we have
v(p) ≤ Mv,W .

(4) For every region D ⊂ Ω, for every u harmonic in D satisfying v ≤ u, either v < u in
D or v = u in D.

If v is subharmonic in Ω, we can define its Poisson modification as follows: Let (U, z)

be a parametric disk with centre p and r whose closure lies in Ω. Let u be the continuous
harmonic extension of v ◦ z −1 from the boundary of z(U ) into z(U ). We now define the
Poisson modification Pv of v as follows:
(
v(q), q ∈ Ω \ U;
PU,v (q) := PU v := Pv (q) =
u ◦ z(q), q ∈ U .
As before v ≤ Pv and Pv is subharmonic.
Ex. 31. Note that Pv can be defined even for a continuous v which is not subharmonic.
Show that v is subharmonic if for all parametric disks (U, z) with closure of U lying in Ω,
v ≤ Pv,U .

12
6 Perron’s Method

Theorem 32. Let Ω be an open connected set in a Riemann Surface X. Let F be a uniformly
bounded non-empty family of subharmonic functions on Ω with the following properties:
(1) u, v ∈ F ⇒ max{u, v} ∈ F.
(2) u ∈ F ⇒ Pu,U ∈ F for any parametric disk (U, z) whose closure lies in Ω.
Then the function h(x) = sup{u(x) : u ∈ F} is harmonic in Ω.

Proof. Let a ∈ Ω, (U, z) a parametric disk at a whose closure lies in Ω. Choose a sequence
(un ) ∈ F such that un (a) → h(a) as n → ∞. If we replace un by vn = max{u1 , u2 , · · · , un }
and further vn by Pvn ,U , in view of hypothesis (1) and (2) of the theorem, we still remain
within the family F. Further the resulting sequence still converges, at a, to h(a). Therefore,
without loss of generality, we assume that the chosen sequence (un ) is increasing and that
un = Pun ,U .
Let u(p) = limn→ un (p), p ∈ Ω. Then u(a) = h(a) and by Harnack’s theorem u is harmonic
in U . We claim that u(p) = h(p) for all p ∈ U . To see this let p ∈ U and let (vn ) be an
increasing sequence in F satisfying (i) un ≤ vn = Pvn ,U , (ii) vn (p) → h(p). If v = limn→∞ vn ,
then v is harmonic in U by Harnack’s theorem. Also u ≤ v. u and v are harmonic in U with
u(a) = v(a). Maximum principle applied to u − v shows that u = v in all U . Thus u = h in
U and h is harmonic in U . Since (U, z) is an arbitrary parametric disk in Ω we conclude that
h is harmonic in Ω.

Solution of Dirichlet Problem

Let Ω be a connected open set in a Riemann surface X and f : ∂Ω → R be a bounded con-

tinuous function. To avoid triviality we assume that f is non-constant. Let M = sup{|f (x)| :
x ∈ ∂Ω}. Let F denote the set of all (real valued) continuous functions on Ω such that
(i) u is subharmonic in Ω and |u| ≤ M , (ii) u |∂Ω ≤ f . Then F is not an empty family
since the constant function −M is in F. Also F satisfies the conditions of Thm. 32. Hence
h(p) = sup{u(p) : u ∈ F} for p ∈ Ω is harmonic in Ω.
For this h to be a solution of the Dirichlet problem for f we must have limy→p h(y) = f (p)
for all p ∈ ∂Ω. Under certain conditions on Ω we can ensure this but not always. Le us call
h the Perron function corresponding to the boundary functionf .
Let us observe, however, that if m ≤ f ≤ M and u is any solution of the Dirichlet
problem for this f , then u = h. For then, u is harmonic and m ≤ u ≤ M in Ω by the
maximum principle, hence belongs to F. We therefore have u ≤ h. Let v ∈ F. Since u is
harmonic v − u is subharmonic in Ω and ≤ 0 on ∂Ω. Hence v − u ≤ 0 on Ω by subharmonicity
of v − u or v ≤ u in Ω. Clearly h ≤ u in Ω. Thus u = h.
In the next section we give conditions on ∂Ω which ensure that the Perron function h is
indeed a solution for the Dirichlet problem for the boundary data f .

13
7 Boundary Behaviour

Definition 33. A point a ∈ ∂Ω is called a peak point or a regular point if there exists an
open set V with a ∈ V and a continuous real valued function P on Ω ∩ V with the following
properties: (i) P is subharmonic in Ω ∩ V , (ii) P (a) = 0, (iii) P (x) < 0 for all x ∈ Ω ∩ V \ {a}.
Such a function P is called a peaking function at a. (In old terminology a peaking function
is called a barrier at a.)

We will show presently that if b ∈ ∂Ω is a peak point then limp→b h(p) = f (b), so that h
constructed above has continuous extension to every peak point of ∂Ω, and the value of h at
the peak point agrees with the value of f at that point. In particular if every point of ∂Ω
is a peak point then the Dirichlet problem has a solution for any continuous bounded f on
∂Ω. This is the main theorem of this section. A converse of this which is easy to prove is the
following:
Theorem 34. If the Dirichlet problem admits a solution for every bounded continuous f on
∂Ω then every point of ∂Ω is a peak point.

Proof. Let b ∈ ∂Ω. Choose a continuous f on ∂Ω such that −1 ≤ f ≤ 0 with f (p) = 0 if and
only of p = b. Let u the solution of the Dirichlet problem for this f . Then −1 ≤ u ≤ 0. By
maximum principle u < 0 in Ω. Clearly u is a peaking function at b and b is a peak point.

There are easy geometric conditions which ensure that a boundary point is a peak point.
We give two such conditions below:

Condition 1. Let Ω be a connected open set in the complex plane and let b ∈ ∂Ω.
Assume that Ω satisfies the exterior sphere condition at b, i.e.,there exists a disk B(z0 , r) such
that b ∈ ∂B(z0 , r) and B(z0 , r) ∩ Ω = ∅. Then b is a peak point.

b+z0
Proof. Let a = 2 . Then P (z) = log 2r − log|z − a| is a peaking function at b, hence b is a
peak point.

Condition 2. If b ∈ ∂Ω can be reached by an analytic arc (i.e., the image of a line

segment under a holomorphic map) which has no points in common with Ω \ {b},then b is a
peak point.

Proof. This is a local problem. We may assume that Ω is a subset of the complex plane and
that b can be reached by a line segment with no point in common with Ω\{b}. We may simplify
further by assuming that b = 0 and the line segment is {(x, y) : y = 0 & x ≤ 0}. Choose a
√ √
single valued branch of z in the complement of the negative real axis. Set −P (z) = <( z).
√
In polar co-ordinates −P (z) = − rcosθ/2 with −π ≤ θ ≤ π. P peaks at b, and b is therefore
peak point of ∂Ω.

Ex. 35. Show that if ∂Ω is a C 1 (e.g., given locally as a level surface with nonvanishing
gradient) then every point of ∂Ω is a peak point.

Before we prove the main theorem of this section we need a lemma.

14
Lemma 36. Let a ∈ ∂Ω be a peak point. Let m ≤ M be given. Let U be a neighbourhood at
a with compact closure. Then there exists a u ∈ C(Ω, R) with the following properties:
u is subharmonic in Ω, u(a) = M ,
u(p) ≤ M on Ω ∩ U ,
u = m on Ω \ U .

Proof. Let V be a neighbourhood of a which admits a peak function P at a. The number

sup{P (z) : z ∈ Ω ∩ ∂U } is either not defined (which can happen if the Ω ∩ ∂U is empty)
or it is less than an negative number −δ for some δ > 0. Choose N large enough so that
−N δ < m − M . Define
(
m, if z ∈ Ω − U
u(z) =
M + max{m − M, N P (z)}, if z ∈ Ω ∩ U .

This u is continuous and subharmonic on Ω∩U and equals m in a neighbourhood of Ω\(Ω∩U )

and so u is subharmonic in Ω.

Theorem 37. Let Ω be a connected open set in a Riemann surface X. Let h be the Perron
function associated to a real valued continuous bounded function f on ∂Ω. Then for every
peak point b ∈ ∂Ω we have
lim h(y) = f (b).
y→b

Proof. Since f is bounded, assume that m ≤ f ≤ M .Let ε > 0 be given. We plan to show
that
f (b) − ε ≤ lim inf h(y) ≤ lim sup h(y) ≤ f (b) + ε.
y→b y→b

For the given ε > 0, we choose neighbourhood V of b with compact closure such that

f (b) − ε ≤ f (y) ≤ f (b) + ε,

for all y ∈ ∂Ω ∩ V .
Step 1. Using the previous lemma, choose v ∈ C(Ω) with the following properties: (i) it
is subharmonic in Ω, (ii) v(b) = f (b) − ε, (iii) v ≤ f (b) − ε on Ω ∩ V and (iv) v = m − ε on
Ω\V.
For y ∈ ∂Ω \ (Ω ∩ V ),
v(y) = m − ε < m ≤ f (y).
For y ∈ ∂Ω ∩ V ,
f (b) − ε ≤ f (y) ≤ f (b) + ε.
But v(y) ≤ f (b) − ε on Ω ∩ V . Thus v ≤ f on ∂Ω. Thus v ∈ F. Hence v ≤ h, or,

f (b) − ε = v(b) = lim v(y) ≤ lim inf h(y).

y→b y→b

Step 2. We again use the previous lemma to find a w ∈ C(Ω) such that (i) w is subhar-
monic in Ω, (ii) w(b) = −f (b), (iii) w ≤ −f (b) on Ω ∩ V (iv) w = −M on Ω \ V .

15
 Take any u ∈ F and y ∈ ∂Ω ∩ V . Then

u(y) ≤ f (y) ≤ f (b) + ε.

Hence on ∂Ω ∩ V ,
w(y) + u(y) ≤ −f (b) + f (b) + ε = ε.
On Ω ∩ ∂V ,
w(y) + u(y) ≤ −M + M = 0.
Thus the function w + u which is subharmonic in Ω ∩ V satisfies w + u ≤ ε on Ω ∩ V by
property (1) of Thm. 30. Thus
u ≤ ε − w on Ω ∩ V
for all u ∈ F. Therefore
h(y) ≤ ε − w(y)
for all y ∈ Ω ∩ V . This implies that

lim sup h(y) ≤ ε − w(b) = ε + f (b).

y→b

The theorem is proved.

Acknowledgements: These notes were based on lectures given by me in an Instructional

Conference on Riemann Surfaces, December 7-28, 1995 at Department of Mathematics, Uni-
versity of Bombay, Bombay. I would like to thank Professor Nadkarni for reasons best known
to both of us!

References

(1) L.V. Ahlfors, Complex Analysis, Third Ed., McGraw-Hill (1979).

(2) H.M. Farkas and I. Kra, Riemann Surfaces, Springer-Verlag, (GTM # 17).

(3) O. Forster, Lectures on Riemann Surfaces, Springer-Verlag, (GTM # 81).

16