Lecture 20

1 Maximum principle
Theorem 1. (weak form) Let D be a bounded connected domain in Rn . For a
smooth solution u which satises 4u = 0. Then
max u = max u
D ∂D

and
min u = min u.
D ∂D

Proof. Given a small  > 0, let v(x) = u(x) + |x|2 . Then we have in D
4v = 4u + 2n = 2n > 0. (1)
If maximum of v attained at x0 ∈ D, we have
D2 v(x0 ) ≤ 0. (2)
Thus (2) contradicts to (1). So the maximum point of v must be attained
on the boundary which means max v = max v . So we get
D ∂D

max u ≤ max u ≤ max v = max v ≤ max u + l2 ,

∂D D D ∂D ∂D

where l is bounded by the diameter of the domain D.

Letting  → 0, we have
max u = max u.
D ∂D

Similarly, for the proof of minD u = min∂D u, we consider v(x) = u(x) −

|x|2 .
n
Suppose u satises the equation 4u+ ∂u
+cu = 0 in D ∈ Rn
P
Exercise 2. bi ∂xi
i=1
where c ≤ 0 and bi are bounded constants. Prove that
max u ≤ max{u, 0}.
D ∂D

1
 and
min u ≥ min{u, 0}.
D ∂D

Hint: let v(x) = u(x) + eαx1 for some large α.

Proposition 3. (Mean value property) Let u be a harmonic function in a disk
D, continuous in its closure D. Then from Poisson's formula, we have the
following mean value properties for any point x0 ∈ D and any ball Br (x0 ) ⊆ D
Z
1
u(x0 ) = u(x0 )ds0 .
2πr |x0 −x0 |=r

and
Z
1
u(x0 ) = u(x0 )dx0 .
πr2 |x0 −x0 |≤r

Remark. There are also mean value properties for n dimensional harmonic func-
tions.
Proof. Without loss of generality, we assume that x0 = 0.
Recall the Poisson's formula
a2 − |x|2 u(x0 )
Z
u(x) = ds0 .
2πa |x0 |=a |x − x0 |2

Let x = 0, we have
a2 u(x0 ) 0
Z
u(0) = ds
2πa |x0 |=a |x0 |2
Z
1
= u(x0 )ds0 .
2πa |x0 |=a

Mutiply both side by a and integrate from 0 to r

r Z rZ
r2
Z
1
u(0) = u(0)ada = u(x0 )ds0
2 0 2π 0 |x0 |=a
Z
1
= u(x0 )dx0 .
2π |x0 |≤r

So we get
Z
1
u(0) = u(x0 )dx0 .
πr2 |x0 |≤r

2
Theorem 4. (strong form) Let u(x) be harmonic in D which is a bounded
connected domain in Rn . Then the maximum point x0 ∈/ D unless u ≡ constant.
In other word, if maximum point x0 ∈ D, then u ≡ costant.
Proof. Denote M = max u. Set Σ = {x ∈ D; u(x) = M }. It is relatively closed
D
in D. If x0 ∈ D, We need to show Σ = D . From the mean value property, we
have for B r (x0 ) ⊆ D for some r > 0
Z
1
M = u(x0 ) = u(x)dx ≤ M.
|Br | Br (x0 )

Thus Br (x0 ) ⊆ Σ. This implies Σ is relatively open in D. In this way, using

the assumption that D is connected, we deduce that Σ = D.
Proposition 5. Let u be a continuous harmonic function in any open set D
of the plane. Then u(x) is smooth in D. This also true for n-dimensional
harmonic functions.
Proof. For any point x ∈ D, there is a ball Ba (x0 ) such that x ∈ Ba (x0 ) ⊆ D.
The mean value property is
a2 − |x − x0 |2 u(x0 )
Z
u(x) = ds0 .
2πa |x0 −x0 |=a |x0 − x|2

Because the denominator of the integrand |x0 − x|2 6= 0 when x ∈ Ba (x0 ).

It implies u is smooth in Ba (x0 ).
Proposition 6. The Dirichlet problem to the Laplace equation
4u = f in D
u=h on ∂D

is unique.
Proof. Suppose u and v are solutions all satisfy the above Dirichlet problem.
Let w = u − v which satises
4w = 0 in D
w = 0 on ∂D.

By the maximum principle we have

0 = min u ≤ wmin ≤ w(x) ≤ wmax ≤ max u = 0.
∂D ∂D

So we get w ≡ 0 which proved the uniqueness.

Proposition 7. Suppose that u ∈ C 2 (B R (x0 )) is harmonic. Then there holds
n
|Du(x0 )| ≤ max|u|.
R BR

3
Proof. Because ∂u
∂xi satises

∂u
4 = 0.
∂xi

Hence ∂u
∂xi has the mean value inequality
Z
∂u 1 ∂u
(x0 ) = (y)dy
∂xi |BR (x0 )| BR (x0 ) ∂xi
Z
1 yi
= u(y) dSy .
|BR (x0 )| ∂BR (x0 ) R

The last equation is due to divergence theorem. So we have

X ∂u 1 XZ yi
2
|Du(x0 )| = | |2 (x0 ) ≤ ( u(y) dSy )2
i
∂xi |BR (x0 )|2 i ∂BR (x0 ) R
|∂BR (x0 )|
Z X yi
≤ 2
u2 (y) ( )2 dSy
|BR (x0 )| ∂BR (x0 ) i
R
|∂BR (x0 )|
= ( max|u|)2
|BR (x0 )| B R
n
= ( max|u|)2 .
R BR

Exercise 8. A bounded harmonic function in Rn is constant.

Exercise 9. Suppose u ∈ C 2 (D) satises
4u + cu = f (x) in D
u = ϕ(x) on ∂D

for some f ∈ C(D) and ϕ ∈ C(∂D). If c ≤ 0, then show that

|u(x)| ≤ max|ϕ| + C max f
∂D D

for any x ∈ D . Where C is a positive constant which depends on diameter

of D.
Example 10. Solve laplace equation in a Wedge
42 u = 0 in W = {(r, θ); 0 < r < a, 0 < θ < β}
u(r, 0) = 0 = u(r, β)
∂u
(a, θ) = h(θ).
∂r

4
Proof. By separation of variable in polar coordinate, we get the solution
∞
X nπθ
u(r, θ) = An rnπ/β sin
n=1
β

with coecients given by

Z β
1−nπ/β 2 nπθ
An = a h(θ) sin dθ.
nπ 0 β

Example 11. Solve laplace equation in annulus

42 u = 0 in A = {0 < a2 < x2 + y 2 < b2 }
u = g(θ) on x2 + y 2 = a2
u = h(θ). on x 2 + y 2 = b2 .
Proof. By separation of variable in polar coordinate, we get the solution
∞
1 X
u(r, θ) = (C0 + D0 log r) + (Cn rn + Dn r−n ) cos nθ + (An rn + Bn r−n ) sin nθ.
2 n=1

Note that in this case we don't throw out the function r−n and log r.
From the boundary condition, the coecients need to satisfy
Z 2π
1
C0 + D0 log a = g(θ)dθ
π 0
Z 2π
1
C0 + D0 log b = h(θ)dθ,
π 0

and
Z 2π
1
Cn an + Dn a−n = g(θ) cos nθdθ
π 0
Z 2π
n −n 1
Cn b + Dn b = h(θ) cos nθdθ,
π 0

and
Z 2π
1
An an + Bn a−n = g(θ) sin nθdθ
π 0
Z 2π
1
An bn + Bn b−n = h(θ) sin nθdθ.
π 0

Thus
Z 2π
1
D0 = (h(θ) − g(θ))dθ
π log ab 0
1 2π
Z
log b log a
C0 = [ g(θ) − h(θ)]dθ,
π 0 log b − log a log b − log a

5
 and
−1 " R 2π #
an a−n 1
  
Cn π g(θ) cos nθdθ
= R 02π ,
Dn bn b−n 1
h(θ) cos nθdθ
π 0

and
−1 " R 2π #
an a−n 1
  
An π 0
g(θ) sin nθdθ
= .
b−n
R 2π
Bn bn 1
h(θ) sin nθdθ
π 0

Example 12. Solve laplace equation in the exterior of a disk

42 u = 0 in x2 + y 2 > a2
u = h(θ) on x2 + y 2 = a2
u bounded as x2 + y 2 → ∞.

Proof. Because u is bounded as r → ∞, in this case we throw out the function

rn and log r. By separation of variable in polar coordinate, we get the solution
∞
1 X
u(r, θ) = A0 + r−n (An cos nθ + Bn sin nθ)
2 n=1

with the coecients given by

π
an
Z
An = h(θ) cos nθdθ
π −π

and
π
an
Z
Bn = h(θ) sin nθdθ.
π −π

So we get Poisson's formula in this case

Z 2π
2 2 h(φ) dφ
u(r, θ) = (r − a )
0 a2 2
− 2ar cos(θ − φ) + r 2π

for r > a.