21 Laplace’s Equation and Harmonic Func-

tions

21.1 Introductory Remarks on the Laplacian operator

Given a domain Ω ⊂ Rd , then

∇2 u = div(grad u) = 0 in Ω (1)

is Laplace’s equation defined in Ω. If d = 2, in cartesian coordinates

∂ 2u ∂ 2u
∇2 u = +
∂x2 ∂y 2
or, in polar coordinates
1 ∂ 2u
 
2 1 ∂ ∂u
∇ u= r + .
r ∂r ∂r r2 ∂θ2
If d = 3, in cartesian coordinates
∂ 2u ∂ 2u ∂ 2u
∇2 u = + +
∂x2 ∂y 2 ∂z 2
while in cylindrical coordinates
1 ∂ 2u ∂ 2u
 
2 1 ∂ ∂u
∇ u= r + 2 2 + 2
r ∂r ∂r r ∂θ ∂z
and in spherical coordinates
1 ∂ 2u 1 ∂ 2u
   
2 1 ∂ 2 ∂u ∂u
∇ u= 2 ρ + 2 + cot φ + .
ρ ∂ρ ∂ρ ρ ∂φ2 ∂φ sin2 φ ∂θ2
We deal with problems involving most of these cases within these Notes.

Remark: There is not a universally accepted angle notion for the Laplacian
in spherical coordinates. See Figure 1 for what θ, φ mean here. (In the rest
of these Notes we will use the notation r for the radial distance from the
origin, no matter what the dimension.)

1
Figure 1: Coordinate angle definitions that will be used for spherical coordi-
nates in these Notes.

Remark: In the math literature the Laplacian is more commonly written with
the symbol ∆; that is, Laplace’s equation becomes ∆u = 0. For these Notes
we write the equation as is done in equation (1) above. The non-homogeneous
version of Laplace’s equation, namely
∇2 u = f (x) (2)
is called Poisson’s equation. Another important equation that comes up
in studying electromagnetic waves is Helmholtz’s equation:
∇2 u + k 2 u = 0 k 2 is a real, positive parameter (3)
Again, Poisson’s equation is a non-homogeneous Laplace’s equation; Helm-
holtz’s equation is not.
Laplace introduced the notion of a potential as the gradient of forces on
a celestial body in 1785, and this potential turned out to satisfy Laplace’s
equation. Then other applications involving Laplaces’s equation came along,
including steady state heat flow (Fourier, 1822), theory of magnetism (Gauss
and Weber, 1839), electric field theory (Thomson, 1847), complex analysis
(Cauchy, 1825, Riemann, 1851), irrotational fluid motion in 2D (Helmholtz,
1858). For the fluid case let u = (u, v) be the fluid velocity vector. Incom-
pressibility gives the condition div(u) = ∇ · u = 0, and irrotationality gives
the condition curl(u) = ∇ × u = 0; this allows us to write u = −grad φ,
where φ is called a velocity potential. Thus, 0 = div(u) = −div(grad φ), so
the velocity potential satisfies Laplace’s equation in the fluid domain.

2
Figure 2: Domain with boundary partitioned into two disjoint pieces for the
probability example.

Complex variable theory is the study of analytic functions f = f (z) of

the complex variable z = x + iy (so f has a convergent power series in z).
We write f (z) = u(z) + iv(z), where u and v are real valued functions, and
thinking of them as functions of x and y, they satisfy the Cauchy-Riemann
equations ∂u
∂x
∂v ∂u
= ∂y ∂v
, ∂y = − ∂x . Note that this gives

uxx = vyx = vxy = −uyy ; that is, ∇2 u = 0 .

Similarly, ∇2 v = 0.

Definition: If u is twice differentiable in each variable in domain Ω, and

∇2 u = 0 in Ω, then u is called harmonic in Ω (or is a harmonic function
in Ω).

Hence, in the above example, u and v are harmonic real and imaginary parts
of the analytic function f . Note that any constant is harmonic (everywhere)
and that if u is harmonic in a domain, so is any constant multiple of u.
For another example suppose Ω is a bounded domain in either R2 or R3 ,
and suppose ∂Ω = C1 ∪ C2 , C1 , C2 are nonempty and non-intersecting. That
is, the boundary of Ω is made up of two disjoint nonempty pieces (see Figure
2). For the sake of argument we will consider the 3D space case and define
u(x, y, z) = probability that a particle that begins at (x, y, z) ∈ Ω and moves
as a random Brownian motion particle, will stop at a point of C2 . Then it
turns out that u satisfies the problem
 2
 ∇ u = 0 in Ω
u=1 on C2
u=0 on C1


3
 Here is another important example. Let E be the electric field vector and
H be the magnetic field vector. The Maxwell’s equations from electro-
magnetic field theory can be written as
∂
∇ × E = −µ H (4)
∂t
∇·H=0 (5)
∂
∇×H= E+J (6)
∂t
∇ · E = ρ/ (7)

where µ is the magnetic permeability, ρ is charge density,  is the dielectric

constant, and J is a (generally specified) current density. If the medium is
homogeneous, all the parameters are constant. Equation (6) is sometimes
called Faraday’s law, and (7) is sometimes called Coulomb’s law. In elec-
trostatics, the t-derivatives are zero, so from (4), ∇ × E = 0 (i.e. E is an
irrotational field). Therefore, there exists an electrical potential φ such that
E = −∇φ. So, div E = − div (∇φ) = −∇2 φ = ρ/, that is, the electric po-
tential satisfies Poisson’s equation. If we are in a vacuum and consider ρ = 0,
then φ satisfies Laplace’s equation. If J = 0, then H can be associated with
a magnetic potential, ψ, and ψ satisfies Laplace’s equation because of (5).
In this section we mostly are concerned with Laplace’s equation in 2D
spatial domains, and construct solutions in some special cases. Again we
do this via separation of variables method. This is mostly a change of no-
tation rather than any new ideas. We restrict our problems to rectangular
(cartesian) coordinates and polar coordinates, mainly because of computa-
tional “simplicity” since problems in 3 space variables can get really long
and messy without introducing any new technical ideas. (We do one case in
spherical coordinates in Appendix H.)

21.2 Laplace’s Equation in a Rectangle

Consider

∇2 u = 0 in Ω = {(x, y) : 0 < x < K, 0 < y < L}

along with Dirichlet boundary conditions as given in Figure 3. (There is

nothing special about using Dirichlet b.c.s on each of the four sides; we could

4
 Figure 3: Rectangular domain in R2 with Dirichlet data on all sides.

have picked Neumann or Robin conditions for any, or all, of the sides. So
we can consider this problem is one of 12 possible problems to discuss.) By
linearity we view the solution u as the sum of four functions (if we have non-
homogeneous b.c.s on all four sides), u = u1 + u2 + u3 + u4 , where uj satisfies
Laplace’s equation in Ω and satisfies one non-homogeneous b.c. while having
homogeneous b.c.s on the other 3 sides. For example, assume u2 be the
solution the the problem pictured in Figure 4. By separation of variables
method, let u2 = X(x)Y (y) and substitute this into the equation. Then
d2 X d2 Y 1 d2 X 1 d2 Y
Y + X = 0 ⇒ = − =λ
dx2 dy 2 X dx2 Y dy 2
so
d2 Y
+ λY = 0 , 0 < y < L , Y (0) = Y (L) = 0 .
dy 2
The eigenvalues are just λn = (nπ/L)2 , n = 1, 2, . . ., with associated eigen-
2 2
functions Yn (y) = sin( nπy L
). Thus, ddxX2 − λX = ddxX2 − (nπ/L)2 X = 0,
0 < x < K, X(0) = 0. This gives X(x) = Xn (x) = sinh( nπx L
), up
to a multiplicative constant. Therefore, each “mode” of u is made up of
sinh( nπx
L
) sin( nπy
L
), and adding up all contributions gives
∞
2
X nπx nπy
u (x, y) = bn sinh( ) sin( )
n=1
L L
Lastly, we consider the one non-homogeneous b.c., i.e.
∞
X nπK nπy
g2 (y) = u2 (K, y) = bn sinh( ) sin( ).
n=1
L L

5
 Figure 4: Rectangular domain in R2 with nonzero data only on one side.

Figure 5: Rectangular domain in R2 with nonzero data on a different side.

This is just the Fourier sine series for g2 (y) on (0, L) with Fourier coefficient
bn sinh( nπK
L
). Therefore,
Z L
2 nπy
bn = nπK
g2 (y) sin( )dy .
L sinh( L ) 0 L
The other three problems are done the same way. The ODE boundary-value
problem with homogeneous boundary conditions at both ends of their interval
determines the EVP. So, for the case in Figure 5, it is the x-variable problem
(u3 = X(x)Y (y)) that determines the eigenvalues and eigenfunctions. In this
case, λn = ( nπ
K
)2 , Xn (x) = sin( nπx
k
), so that Y 00 − λn Y = 0, Y (L) = 0. For
convenience, you can write Y (y) = Yn (y) = sinh( nπ K
(L − y)), rather than a
linear combination of sinh(nπy/L), cosh(nπy/l). Then
∞
3
X nπ nπx
u (x, y) = an sinh((L − y)) sin( ) with
n=1
K K
Z K
2 nπx
an = nπL
f1 (x) sin( )dx .
K sinh( K ) 0 K

6
 Figure 6: Square domain and boundary conditions for the Exercise.

Try the following problem:

Exercise: Let ∇2 u = 0 in the square Ω = {(x, y) : 0 < x, y < π} with

the boundary data given as in Figure 6. Find u(x, y).

21.3 Laplace’s Equation in a Disk: Poisson’s Formula

Consider
 2
∇ u=0 in Ω = {(x, y) : x2 + y 2 < a2 } = {(r, θ) : 0 ≤ r < a, 0 ≤ θ < 2π}
u(a, θ) = f (θ) 0 ≤ θ < 2π

Now, in polar coordinates

2 1 ∂ ∂u 1 ∂ 2u
∇ u= (r ) + 2 2 = 0 ,
r ∂r ∂r r ∂θ
so let u = Θ(θ)φ(r). Then, by separation of variables,
d
r dr (r dφ
dr
) 1 d2 Θ
=− =λ ⇒
φ Θ dθ2
d2 Θ
+ λΘ = 0 0 ≤ θ < 2π ⇒
dθ2 √ √
Θ(θ) = an cos( λθ) + bn sin( λθ) .

But Θ must be 2π-periodic, which forces λ = λn = n2 , for n = 0, 1, 2, . . .,

so Θ = Θn (θ) = an cos(nθ) + bn sin(nθ). Note that λ0 = 0 is an eigenvalue.

7
Also,
d2 φ dφ
r2 2
+r − n2 φ = 0 . (8)
dr dr
If n = 0, then

d2 φ dφ d dφ
r2 + r = r (r ) = 0 ⇒ φ = φ0 = c ln(r) + d .
dr2 dr dr dr
Since u, and hence φ, must be bounded at r = 0, then c = 0 ; i.e. φ0 =
constant. If n > 0, then (8) is a Cauchy-Euler equation. Therefore, let
φ = rα ; then the characteristic equation for (8) when this is substituted in
becomes α(α − 1) + α − n2 = α2 − n2 = 0, so α = ±n. The general solution,
for n > 0, is now
φ = φn (r) = αn rn + βn r−n . (9)
Again, since we require boundedness at r = 0, βn = 0 for all n ≥ 1. In
summary, φn (r) is a constant for n = 0 (so is Θ = Θ0 (θ)) and is a constant
times rn for n > 0. Combining this with Θn (θ) gives
∞
a0 X n
u(r, θ) = + r {an cos(nθ) + bn sin(nθ)} . (10)
2 n=1

Letting r → a gives
∞
a0 X n
f (θ) = + a {an cos(nθ) + bn sin(nθ)} .
2 n=1

This is the full Fourier series for f (θ). Using the orthogonality of the set
{1, cos(nθ), sin(nθ)}n≥1 on [0, 2π], we have
Z 2π
a0 1
= f (ψ)dψ = average of f on [0, 2π] , (11)
2 2π 0

and   Z 2π  
an 1 cos(nψ)
= n f (ψ) dψ . (12)
bn πa 0 sin(nψ)
This finishes the description of the solution (10) to our potential equation in
the disk. But let us push the result a bit further by substituting (11),(12)

8
into (10):
Z 2π
1
u(r, θ) = f (ψ)dψ +
2π 0
∞
1 X  r n 2π
Z
f (ψ){cos(nψ) cos(nθ) + sin(ψ) sin(θ)}dψ
π n=1 a 0
Z 2π ( ∞  
)
1 X r n
= f (ψ) 1 + 2 cos n(θ − ψ) dψ
2π 0 1
a

Consider now the expression in the brackets with ξ := θ − ψ:

∞   ∞   ∞  
X r n X r n X r n
1+2 cos(nξ) = 1 + einξ + e−inξ .
1
a 1
a 1
a
P∞ z
Recall the geometric series n=1 zn = 1−z
if |z| < 1. Since r < a, |z| =
| ar e±iξ | = | ar | < 1, so
∞  ∞ 
X r n X r n
1+ eiξ + e−iξ
1
a 1
a
(r/a)eiξ (r/a)e−iξ
=1+ +
1 − (r/a)eiξ 1 − (r/a)e−iξ
reiξ re−iξ
=1+ +
a − reiξ a − re−iξ
re (a − re−iξ ) + re−iξ (a − reiξ )
iξ
=1+
a2 − ar(eiξ + e−iξ ) + r2
a2 − r 2
= 2 .
a − 2ar cos(ξ) + r2
Hence
2π
a2 − r2
Z
f (ψ) dψ
u(r, θ) = . (13)
2π 0 a2 − 2ar cos(θ − ψ) + r2
This is Poisson’s Formula for any harmonic function inside a circle of ra-
dius a > 0 (centered at the origin), with boundary values given by f (ψ).

Theorem: Let f (θ) be any continuous function on the boundary of the

disk {r = a}. Then (13) provides the only harmonic function in Ω = {r < a}

9
for which for any x0 = (a, θ0 ) ∈ ∂Ω = {r = a}, limx→x0 u(x) = f (x0 ) from
inside Ω.

This means that u is at least continuous in Ω and its boundary, and twice
continuously differentiable inside Ω. In actuality, u is infinitely differentiable
inside Ω.

Exercise: Derive the series solution and analogue to Poisson’s formula for
the exterior problem:
 2
∇ u=0 in Ω = {(r, θ) : r > a, 0 ≤ θ < 2π}
u(a, θ) = f (θ) 0 ≤ θ < 2π

21.4 Some Comments and Consequences

1. Let x = (r cos(θ), r sin(θ)), x0 = (a cos(ψ), a sin(ψ)), then |x − x0 |2 =
(r cos(θ) − a cos(ψ))2 + (r sin(θ) − a sin(ψ))2 = r2 − 2ar cos(θ − ψ) + a2 ,
so Poisson’s formula (13) can be rewritten as

a2 − |x|2 u(x0 )
Z
u(x) = 0 2
ds0 , (14)
2πa |x0 |=a |x − x |

where arclength on the circle is ds0 = a dθ.

2. Mean Value Property: Let x = 0 (r = 0) in (13) (or (14)):

Z 2π
1
u(0, θ) = u(0, 0) = f (ψ)dψ = average of f
2π 0
That is, the value of a harmonic function u at the center of a disk equals
the average of u on its circumference (if the disk is contained in a region
where u is harmonic). This also means minr=a u ≤ u(0) ≤ maxr=a u
because of the mean value property, and equality holds for one, hence
both, sides of the inequality only if u = constant. It also follows that if
u is harmonic in domain Ω, and if C is any circle contained in Ω with
center located at x ∈ Ω, then u(x) = average of values of u around C.
This leads to

10
3. Maximum Principle: For any bounded, connected domain Ω, let u
be continuous on Ω̄ = Ω ∪ ∂Ω, harmonic in Ω. Then the maximum
(and minimum) of u is attained on ∂Ω, and nowhere else (unless u =
constant).
Suppose u attained a maximum at some x∗ ∈ Ω̄, that is u(x) ≤ u(x∗ ) =
M for all x ∈ Ω. Then the claim is that x∗ ∈ / Ω, unless u = constant.
∗
But if x ∈ Ω it is the center for a disk C contained entirely in Ω. By
the mean value property, u(x∗ ) is the average of the values of u around
the boundary of the C, and since the average can not be greater than
M, then M = u(x∗ ) = average of u on the circle ≤ M , which implies
u ≡ M along the whole circumference of C. This makes u ≡ M in the
whole disk C. Repeating this argument for another overlapping disk
forces u to be M in the union of the disks. Continue to move to new
disks, one obtain u ≡ M throughout Ω; hence u is a constant in its
domain. This argument can be repeated for −u since it would also be
harmonic in Ω, and so the minimum of u is attained on ∂Ω (unless u
is a constant).

4. Smoothness: Let u be harmonic in Ω ⊂ R2 . Then u(x) possesses

partial derivatives of all orders in Ω.

Remark: We mentioned earlier that this smoothness property and the

maximum principle property holds for the heat equation. Actually,
in that case, they are a consequence of the properties holding for the
Laplacian (but they absolutely do not hold for the wave equation).

5. Harnack’s inequality: Let u be harmonic and non-negative in the

disk Ω = {|x| < a}. Then, for any x = (r, θ) ∈ Ω,
a−r a+r
u(0) ≤ u(x) ≤ u(0) .
a+r a−r

Exercise: This is just a consequence of Poisson’s formula and the mean

value property. Try to prove it.

6. Liouville’s Theorem: A function that is harmonic in the whole plane,

and bounded either from above, or below, is a constant.

11
 Suppose u is harmonic in R2 and u(x) < M for all x ∈ R2 . Then M −
u(x) is harmonic and non-negative in R2 . Apply Harnack’s inequality
for a disk centered at the origin and having radius a:
a−r a+r
(M − u(0)) ≤ M − u(x) ≤ (M − u(0)) .
a+r a−r
Now let a → ∞. this gives M − u(0) ≤ M − u(x) ≤ M − u(0); that is,
u(x) ≡ u(0) ⇒ u is a constant. For the case where u(x) > m for all
x ∈ R2 , use the same argument for u(x) − m.
7. Dirichlet’s Principle: A general principle in physics is that a system
prefers going into a state of lowest energy (the ‘ground state’). Dirich-
let’s principle codifies this mathematically for Laplace’s equation on a
bounded domain (in Rn ) with specific Dirichlet boundary conditions,
and states that of all smooth functions defined on the domain and
satisfying the boundary conditions, it is the harmonic function on the
domain that has minimum potential energy. The Dirichlet principle is
discussed in Appendix I.

21.5 Laplace’s Equation on a Wedge

Consider

∇2 u = 0 in Ω = {(r, θ) : 0 < r < a, 0 < θ < θ0 } 
u(a, θ) = f (θ) 0 < θ < θ0 (15)
u(r, 0) = u(r, θ0 ) = 0 0 < r < a


As before, let u(r, θ) = φ(r)Θ(θ). Then

d2 Θ
+ λΘ = 0 , 0 < θ < θ0 , Θ(0) = Θ(θ0 ) = 0 .
dθ2
and  
d dφ
r r − λφ = 0 , 0 < r < a .
dr dr
√ √ √
Now Θ(θ) = a cos( λθ)+b sin( λθ); with Θ(0) = a = 0, Θ(θ0 ) = b sin( λθ0 ) =
 2
0, then we have λ = λn = nπ θ0
, n = 1, 2, . . ., and so Θ(θ) = Θn (θ) =
sin(nπθ/θ0 ). Also,
2
 2
2d φ dφ nπ
r 2
+r − φ=0.
dr dr θ0

12
Again, this is a Cauchy-Euler equation, with general solution φ(r) = φn (r) =
arnπ/θ0 + br−nπ/θ0 because the characteristic polynomial
√ for
√ the equation is
α2 − λn = 0, so a fundamental set of solutions is r λn , r− λn . But we want
boundedness of u, hence φ as r → 0 within the wedge, so we need b = 0.
Therefore,
∞
X
u(r, θ) = An rnπ/θ0 sin(nπθ/θ0 ) .
n=1

Letting r → a we obtain the Fourier sine series for f :

∞
X
f (θ) = An anπ/θ0 sin(nπθ/θ0 ) ⇒
n=1
−nπ/θ0 Z θ0
2a
An = f (θ) sin(nπθ/θ0 )dθ .
θ0 0

Remark: On the more general wedge problem

Consider the problem

∇2 u = 0 in Ω = {(r, θ) : 0 < r < a, 0 < θ < θ0 } 

u(a, θ) = f (θ) 0 < θ < θ0

(16)
u(r, 0) = A(r) 

u(r, θ0 ) = B(r) 0 < r < a


It is natural to break this problem up into two pieces, one being problem
(15) with solution u(1) given above, and one being

∇2 u = 0 in Ω = {(r, θ) : 0 < r < a, 0 < θ < θ0 } 

u(a, θ) = 0 0 < θ < θ0

(17)
u(r, 0) = A(r) 

u(r, θ0 ) = B(r) 0 < r < a


with solution u(2) . Then u = u(1) + u(2) . Unfortunately, problem (17) takes
more advanced techniques to solve than what we are willing to do at this
subject level.

Remark: In three space dimensions we would be interested in Laplace’s equa-

tion in the ball. This is discussed in Appendix H, where another historically
important equation and special functions are introduced, namely Legendre’s
equation and Legendre polynomials.

13
Summary: You need to know how to compute eigenfunction expansions for
harmonic functions on specific domains, like rectangles, disks, and wedges.
You need to know properties of harmonic functions as given in this section.

Exercises:

1. Let ∇2 u = 0 in the square Ω = {(x, y) : 0 < x, y < π}, with ux (π, y) +

au(π, y) = G, and u = 0 on the other three sides of the square. Assume
G and a are constants, with a 6= −1. Determine the eigenfunction
expansion for the solution u(x, y).
P∞ sinh(nx)
(Answer: u(x, y) = 4Gπ n=odd n cosh(nπ)+a sinh(nπ) sin(ny).)

2. Consider ∇2 u = 0 in the unit

R 2πdisk Ω = {(r, θ) : r < 1, 0 < θ ≤ 2π},
∂u
with ∂r (1, θ) = g(θ), where 0 g(θ) dθ = 0. Find a formal solution.

3. Let ∇2 u = 0 in the rectangle Ω = {(x, y) : 0 < x < K, 0 < y < L},

with boundary conditions ux (0, y) = 0, uy (x, 0) = uy (x, L) = 0, and
u(K, y) = A = constant. Determine the harmonic solution in Ω.
(Answer: u(x, y) ≡ A.)

4. Consider ∇2 u = c in the unit disk Ω = {(r, θ) : r < 1, 0 < θ ≤ 2π},

where c is a constant, and ∂u/∂r = 1 on the boundary ∂Ω = {r = 1}.
Determine all solutions. (Hint: boundary condition is independent of
the angle θ, so so is the solution.)

5. Consider
∂ 2u ∂ 2u ∂u
+ − α = 0 |x| < l , z > 0 , α > 0 is a constant
∂x2 ∂z 2 ∂z
Assume u is to remain finite as z → ∞, and has the form
∞
A0 X nπx
u(x, z) = + An (z) cos( ).
2 n=1
l

Show that ∞
X nπx
u(x, z) = a0 /2 + an eβn z cos( )
n=1
l
where the a0n s, βn0 s are constants to be determined.

14
 Now assume the boundary condition
∂u
− (x, 0) + αu(x, 0) = RH(ω − |x|)
∂z
where R > 0 is constant, ω is some fixed constant in (0, l), and H(·) is
the Heaviside function. Find the solution u(x, z)1 .

6. Consider the Neumann problem

 2
∇ u = f (x, y) in Ω ⊂ R2
∂u
∂ν
=0 on ∂Ω

where ν is the unit outward pointing vector defined on the boundary

of domain Ω.

(a) What can we add to any solution to get another solution?

R
(b) Use the divergence theorem and the PDE to show that Ω f (x, y)dxdy =
0 is a necessary condition for the Neumann problem to have a so-
lution.
(c) Can you give a physical interpretation of (a) and (b) for heat flow
or general diffusion?

(If the boundary is completely insulated, then the non-homogeneity

has average (over Ω) zero, meaning the internal sources and sinks must
“balance.”)

1
This problem is associated with a certain irrigation problem.

15