J.

Rauch Applied Complex Analysis

The Dirichlet Problem
Abstract. We solve, by simple formula, the Dirichlet Problem in a
half space with step function boundary data. Uniqueness is proved by
complex variable methods. The formula for the Poisson kernel follows.
The case of the disk follows by a fractional linear transformation.

1. Introduction
For a nice bounded domain G and function g : ∂G → R the Dirichlet
Problem is to show that there is one and only one harmonic function
u : G → R so that the restriction of u to the boundary is equal to g.
And to find as many interesting qualitative properties of the solution.
The classic formulation is to consider continuous functions g and seek
solutions u that are continuous up to the boundary. We will be con-
sidering g that are continuous except for jump discontinuities in which
case the continuity is only required up to closed subintervals disjoint
from the discontinuity points of g.
That this problem has a unique solution is suggested by physical intu-
ition as follows. Consider heat flow in G where the boundary is main-
tained at the time independent temperature g. It seems reasonable
that the time independent boundary condition will drive the solution
in the limit t → ∞ to a time independent solution. That solution must
then be a time independent solution of ut = ν∆u that is equal to g
on the boundary. Therefore a solution of the Dirichlet Problem. On
physical grounds again, one expects that throughout G,
min g ≤ u ≤ max g . (1.1)
∂G ∂G

For continuous solutions u on the closed domain this estimate was

proved in class as a corollary to the mean value properties of analytic
and harmonic functions in the plane. The fact that physical intuition
leads to a provable qualitative property gives one confidence in the
model and the intuition.
In addition, (1.1) implies the uniqueness of solutions u to the Dirichlet
problem that are continuous up to the boundary. Indeed, if there were
two such solution u1 and u2 then the difference u = u1 − u2 would be
a harmonic function continuous up to the boundary and vanishing on
the boundary. Then (1.1) implies that u = 0 proving uniqueness. This
shows that no additional data need be prescribed in order to determine
the steady state. Showing that arbitrary solutions tend to the steady
1
2

state is properly the domain of a more detailed investigation of Partial

Differental Equations.
The special case of the Dirichlet Problem in the disk can be attacked by
introducing polar coordinates u = u(r, θ) and expanding the periodic
function of θ in a Fourier series. This leads to an exact solution formula
involving a Greens’ function called the Poisson kernel. The key element
is that the real and imaginary parts of z n are harmonic showing that
rn e±inθ , n = 0, 1, 2, . . .
are harmonic functions so by superposition
X∞
cn r|n| einθ
−∞

is harmonic too. There is a separate handout discussing that approach

from a complex variables point of view.
A similar strategy using Fourier Integrals works for G equal to the up-
per half space. In that case there is no uniqueness unless one prescribes
the additional condition that g and the solution u are required to be
bounded. In that case the estimate (1.1) involves sups and infs instead
of maxs and mins.
This note attacks the half space Dirichlet Problem starting with a dif-
ferent set of exact solutions derived from Arg(z). This naturally leads
to solutions for data on the boundary consisting of functions that are
piecewise constant. Pursing this line using purely complex methods
leads to the Poisson kernel and also estimate (1.1).

2. Step data for the half plane and the disk

The Dirichlet problem in the half space {y > 0} asks one to find a
harmonic function u with u|y=0 = g(x) with g a prescribed function.
Since the function y is harmonic with zero boundary value, given any
solution the functions u + Cy are also solutions. To guarantee unique-
ness one requires that u is bounded in {y > 0}.
For any analytic F = u + iv, u and v are harmonic. For example the
function F (z) = ln z with −π < arg < 3π/2 yields the two harmonic
functions ln r and arg(z) on the open half space y > 0. The function
arg is bounded, that is
sup | arg z| < ∞ .
y>0

The boundary value on the x-axis of arg z is the discontinuous function

that is equal to 0 on ]0, ∞[ and to π on ] − ∞, 0[.
 3

Exercise 2.1. Show that the function arg is continuous, and in fact
infinitely differentiable on {y ≥ 0} \ 0.

There is no value for the argument at the origin and it is best left
undefined. The uniqueness part of the Theorem below shows that that
is justifiable.
Definition 2.2. A function g : R → R is called a step function when
there is an n ≥ 1 and subdivision of R into n + 1 intervals −∞ < x1 <
x2 < · · · < xn < ∞ so that the restriction of g to each open interval is
constant.
Example 2.3. A constant function is a step function. It suffices to
take n = 1 and place the division point x1 arbitrarily.
Theorem 2.4. If g(x) is a step function with discontinuities at {xj },
then there is one and only one bounded harmonic function u in {y >
0} that is continuous in {y ≥ 0} \ {xj } and so that u(x, 0) = g(x)
on R \ {xj }. In addition there are uniquely determined constants aj ,
1 ≤ j ≤ n + 1, so that
u = a1 arg(z − x1 ) + · · · + an arg(z − xn ) + an+1 . (2.1)
Proof. Uniqueness. If u1 and u2 are solutions, denote by u := u1 −u2 .
Choose a harmonic conjugate v to u in the simply connected set {y > 0}
and define F = u + iv.
Exercise 2.5. Show that v and therefore F is continuous on {y ≥
0} \ {xj }.

The Schwarz Reflection Theorem shows that defining F (z) := F (z ∗ )∗

when Im z ≤ 0 yields a bounded function that is analytic in C \ {xj }.
Each of the excluded points is an isolated singularity. Abuse notation
by denoting by F that analytic continuation.
Though the function u is bounded the harmonic conjugate will in gen-
eral not be bounded. The function F has isolated singularitiwa at the
points xj . The real part of F is bounded. Both essential singulari-
ties and poles have real parts unbounded in any neighborhood of each
singularity so the only possibility is a removable singularity.
Exercise 2.6. Give details for the last sentence.

Therefore defining F appropriately at each xj yields a entire function

with bounded real part. The function eF (z) is therefore bounded and
entire. Liouville’s Theorem implies that eF (z) is constant function.
Therefore F (z) is constant.
4

Exercise 2.7. Give details for the last sentence.

Therefore u is constant. Since u = 0 on R \ {xj } it follows that u = 0

everywhere. This proves uniqueness.
Existence. We show that there is a choice of the constants aj so that
the formula (2.1) is a solution.
R \ {xj } is the disjoint union of n + 1 open intervals denoted Ij for
1 ≤ j ≤ n + 1 the n + 1. Number so that Ij+1 is to the right of Ij .
Denote by gj the values of g on the corresponding interval. The key
observation is that
arg(z − xj ) = 0, on Ik for k ≥ j .
Thus (2.1) holds if and only if
π(aj + aj+1 + · · · + an ) + an+1 = gj , 1 ≤ j ≤ n + 1.
The case j = n + 1 holds if and only if
an+1 = gn+1 .
Then the case n holds if an only if
π an + an+1 = gn
detemining an . Continuing, the values of the aj are uniquely deter-
mined so that the function (2.1) satisfies the boundary value prob-
lem. 
Example 2.8. If g(x) is the characteristic function of ]0, ∞[ (a.k.a.
the heaviside function), the solution is
u = 1 − (arg z)/π . (2.2)
Exercise 2.9. Explain why the unique determination of the aj does
NOT prove uniqueness of the solution.
Exercise 2.10. Find the solution so that for x ∈]j − 1, j[ with 1 ≤
j ≤ N , u(x, 0) = j, and, u(x, 0) = 0 otherwise. Sketch the boundary
values.
Exercise 2.11. Denote by τh the operator that translates a function h
units in the x direction
(τh u)(x, y) := u(x − h, y) .
For a ∈ R and h > 0, show that the unique solution with g equal to the
characteristic function of the interval ]a, a + h[ is equal to
τ−h (arg(z − a)) − arg(z − a)
− .
π
 5

Conclude that the solution is strictly positive in the open upper half
plane.
Exercise 2.12. If g is a nonnegative step function not identically equal
to zero, show that the solution is strictly positive in the open upper half
plane. Hint. Use the preceding exercise. Discussion. This recovers
Estimate 1.1. The present proof does not depend on the maximum and
minimum theorems for harmonic functions that depended on continuity
up to the boundary. The present solutions are not continuous up to
the boundary.
Exercise 2.13. Show that the solution constructed in the Theorem
takes values for y > 0 strictly between the minimum and maximum
values of g(x). Hint. Use the preceding exercise.

Suppose now that g(x) is continuous on R and that the two limits
limx→±∞ g(x) exist. Using the solvability of the Dirichlet problem for
step function data, and approximating g uniformly by such data, it is
not hard to prove that the Dirichlet problem with boundary value g
has a unique bounded solution. A alternative strategy is given in the
next section.
Corollary 2.14. Suppose that 0 ≤ θ1 < θ2 < · · · < θn < 2π and Pj :=
eiθj the finite set of corresponding points on the unit circle. Suppose
that g : S 1 \ {Pj } is constant on each of the intervals in S 1 \ {Pj }.
Then there is one and only one harmonic function u : {|z| < 1} that is
uniformly bounded and continuous on {|z| ≤ 1} \ {Pj } and equal to g
on S 1 \ {Pj }.

Proof. Choose a fractional linear transformation L(ζ) conformally map-

ping the upper half plane to the unit disk. Then u solves the problem
of the Corollary if and only if w(ζ) := u(L(ζ) solves a Dirichlet problem
of the same type for the upper half plane. Existence and uniqueness
follow from Theorem 2.4. 

3. The Green’s function for the upper half space

3.1. Elementary version. Suppose that h > 0 and g = 1]a,a+h[ is
the characteristic function of an interval. Exercise 2.11 shows that the
solution of the Dirichlet Problem is
θ(x + h, y) − θ(x, y)
u(x, y) = .
π
6

The Fundament Theorem of Calculus implies that with

1 ∂θ
G(x, y) := − ,
π ∂x
Z a+h Z ∞
u(x, y) = G(x, y) dy = G(x − s, y) g(s) ds .
a −∞

To compute a formula for G, differentiate using θ = Im ln z. Therefore

∂θ ∂ ln z ∂ ln z 1 z −y
= Im = Im = Im = Im 2 = 2 .
∂x ∂x ∂z z |z| x + y2
Therefore
1 y
G(x, y) = .
π x + y2
2

The resulting formula

Z ∞
1 y
u(x, y) = g(s) ds
π −∞ (x − s)2 + y 2
for the bounded solution of the Dirichlet Problem in the half space is
called Poisson’s Integral Formula. It works for for example when
g(x) is continuous tends to zero as |x| → ∞.

3.2. Version using Dirac’s delta function. There is an alternate

derivation of the formula for G(x, y) namely to solve the Dirichlet prob-
lem with initial temperature equal to 0 for x < 0 and x > ε and tem-
perature 1/ε on ]0, ε[. This initial temperature converges to δ(x) as
ε → 0. Call the solution uε (x, y). Then

G(x, y) = lim uε (x, y) .

ε→0

To see that the formulas are the same, define θ to be the standard
argument with values in ] − π, π[. Then Exercise 2.11 shows that
θ(x − ε, y) − θ(x, y)
uε (x, y) = .
επ
Therefore
1 ∂θ(x, y)
lim uε (x, y) = − .
ε→0 π ∂x
This agrees with the formula from the preceding subsection.
 7

4. Thermal equilibrium and electrostatics

When F (z) = u+iv is analytic and u represents temperature at thermal

equilibrium, the level curves of u are isotherms and −∇u is orthogonal
to those curves and is the direction of heat flow. The level curves of v,
that are orthogonal to those of u, therefore represent heat flow lines.
If the level curves of v are sketched with equal increments in the con-
stant values, then the density of level lines crossing a surface measures
the flux of heat per unit time.
If u is the electrostatic potential then ∇u is the electric field so the
level curves of u are orthogonal to the field. Thus the level curves of
v are parallel to the field. They are the field lines. If the level curves
of v are sketched with constant increments, then the density of lines
represents the electric field strength.
The energy density for the heat flow problems is u dxdy while in elec-
trostatics it is |E|2 dxdy = |∇u|2 dxdy. For heat flow the harmonic
function arg z has finite energy in each bounded set. For electrostat-
ics the energy in any bounded set containing 0 is infinite. If one tries
to arrange a laboratory realization of an electrostatic experiment for
the solutions of section 1, there will be spectacular arcing from the
adjacent intervals at different potentials. These boundary values are
unattainable.
On the other hand, intervals of different electrostatic potential sepa-
rated by an insulator is easily realized and the corresponding boundary
value problems solved using the conformal mapping sin z. In contrast
to the results of §1 these interesting solutions do not generalize to sim-
ple formulas for more than two intervals of insulator.

5. Related examples
Uniqueness for some other boundary value problems can be proved
using reflection. The next result is applied to the difference of two
solutions.
Theorem 5.1. If u is a uniformly bounded and continuously differen-
tialbe function on 0 ≤ y ≤ L which is harmonic in 0 < y < L satifies
u(x, 0) = 0 and either
∂u(x, L)
for all x, u(x, L) = 0 or, for all x, = 0,
∂y
then u = 0.
8

Proof. First treat u(x, L) = 0. Choose a harmonic conjugate v in

0 ≤ y ≤ L and let F (z) = u + iv. Shwartz reflection across y = L
yields an analytic function in 0 < y < 2L with u(x, 2L) = 0 and the
same uniform bound on u as in the initial strip. Repeating extends to
0 < y < ∞ with the same uniform bound on u. A final relfection in
y = 0 yields an entire F with uniformly bounded u. Therefore F is
constant so u is constant. Since u = 0 on the x-axis it follows that
u = 0.
For the case uy (x, L) = 0 the Cauchy-Riemann equations imply that
vx (x, L) = 0 so v is constant on y = L Adding a constant to v we may
assume that v(x, L) = 0. A Shwartz reflection across y = L then yields
an analytic continuation to 0 < y < 2L with the same bound on u and
u(x, 2L) = 0. The first result then applies to prove that u = 0. 

This result treats the case of the electric field between two infinite
capacitor plates. One of the important examples not covered is the
field describing edge effects in a semi-infinite or finite capacitor. These
may be included in a future version.