Chengbo Wang
Chengbo Wang
Chengbo Wang
417S11)
LECTURE 12
CHENGBO WANG
Last time:
∂2u
• (Derive) wave equations: ∂t2 − k∆u = Q(x, t)
• Boundary conditions
Today: Maximum principle, Uniqueness
1. Introduction: Uniqueness
Recall that we have discussed two methods of solving PDEs (IBVP for wave/heat
equations, BVP for Laplace equation)
(1) method of separation of variables (MSV)
find the solutions in terms of Fourier series → existence of PDE
(2) method of eigenfunction expansion
• assuming existence of the solutions with certain smooth property,
then expand it as series of eigenfunctions, determine the coefficients
(uniqueness of PDE)
• after that, check existence and the assumption (existence of PDE)
So, we have discussed one method to prove uniqueness.
However, for the method of eigenfunction expansion,
• Pros
(1) prove existence/uniqueness at the same time
(2) express the solution explicitly
• Cons
(1) rely heavily on the knowledge of the eigenfunctions
(2) apply only to the problems with regular domains (like rectangle)
In this lecture, we want to present a general method to prove uniqueness for the
Laplace/heat equations. For simplicity, here, we will discuss only the 2-dimensional
Laplace equation.
2. Maximum Principle
Theorem 1 (Maximum Principle). Let D be a connected (regular) bounded open
set in R2 . Assuming u = u(x, y) is a solution of the Laplace equation ∆u = 0 in D
and continuous on D̄ = D ∪ ∂D (∂D is the boundary of D). Then the maximum
and minimum values of u are attained on ∂D, that is,
∃~xM , ~xm ∈ ∂D, s.t. u(~xm ) ≤ u(~x) ≤ u(~xM )∀~x ∈ D
1
2 CHENGBO WANG
Proof.
1) If we have maximum value in D at, say, (x0 , y0 ), then we have
uxx ≤ 0, and uyy ≤ 0, at (x0 , y0 )
Since ∆u = uxx + uyy = 0, we see that uxx = uyy = 0 at (x0 , y0 ). This does not
give us enough information.
2) Instead, for any > 0, consider
v(x, y) = u(x, y) + (x2 + y 2 )
∆v(x, y) = ∆u(x, y) + ∆(x2 + y 2 ) = 0 + 4 > 0
Recall that ∆v ≤ 0 at interior maximum points, this tells us that there is no interior
maximum points in D.
3) Since u is continuous on D̄ = D ∪ ∂D, v is also continuous on D̄ and so has
maximum value at ~x0 ∈ D̄. Thus ~x0 ∈ ∂D,
max v(x, y) ≤ v(~x0 ) ≤ max v(x, y)
(x,y)∈D (x,y)∈∂D
Remark 1. A similar proof will yield the stability of the problem with respect of
the boundary conditions. That is, if v is a solution for
∆v = F inD
v=g on∂D
then
|u − v| ≤ max |f − g| .
∂D
we find the solutions are of the form (by MSV or eigenfunction expansion)
X∞ ∞
X
(4.2) u(r, θ) = a0 + an rn cos(nθ) + bn rn sin(nθ)
n=1 n=1
with Z π Z π
1 1
a0 = f (θ)dθ an = f (θ) cos(nθ)dθ, n ≥ 1
2π −π πRn −π
Z π
1
bn = f (θ) sin(nθ)dθ n ≥ 1.
πRn −π
It turns out that we can simplify the formula (4.2).
Theorem 3 (Poisson’s formula). The solution of (4.1) is given by
R2 − r 2 π
Z
f (α)
(4.3) u(r, θ) = 2 − 2Rr cos(θ − α) + r 2
dα .
2π −π R
Proof. By the formulas of the coefficients and (4.2), we see that
Z π ∞ Z π
1 X 1
u(r, θ) = f (θ)dθ + ( n f (α) cos(nα)dα)rn cos(nθ)
2π −π n=1
πR −π
∞ Z π
X 1
+ ( n f (α) sin(nα)dα)rn sin(nθ)
n=1
πR −π
Z π ∞ Z π
1 X rn
= f (θ)dθ + f (α)[cos(nα) cos(nθ) + sin(nα) sin(nθ)]dα
2π −π n=1
πRn −π
∞
Z π " #
1 X 2rn
= f (α) 1 + cos(n(α − θ)) dα .
2π −π n=1
πRn
P∞ 2rn
Question: what is 1 + n=1 πR n cos(n(α − θ))?
thus
∞ ∞ n
X 2rn X r
1+ cos(nα) = 1+ (einα + e−inα )
n=1
πRn n=1
R
∞ n ∞ n
X r X r
= eiα + e−iα −1
n=0
R n=0
R
1 1
= + −1
1 − Rr eiα 1 − Rr e−iα
2 − 2 Rr cos(α)
= −1
|1 − Rr eiα |2
2 − 2 Rr cos(α)
= 2 − 1
1 − 2 Rr cos(α) + Rr
2
1 − Rr
= 2
1 − 2 Rr cos(α) + Rr
R2 − r 2
=
R2 − 2Rr cos(α) + r2
Substituting this identity to the previous identity gives us (4.3).
With Poisson’s formula, it will be easy to obtain a remarkable property of the
solution for the Laplace equation.
Corollary 4 (mean value property). Let ∆u = 0 in an open disk with radius R,
BR , and u be continuous function in BR ∪ ∂BR , then the value of u at the center
is the same as the average value of u on the circle ∂BR .
Proof. Without loss of generality, assume BR center at the origin. Then an
application of (4.3) with r = 0 gives us
Z π
1
u(0) = u(R, θ)dθ .
2π −π
Then there must exist a R0 > 0 such that B2R (x0 , y0 ) ⊂ D for any R < R0 . Notice
that
u(x0 , y0 ) = average of u on∂BR (x0 , y0 ) ≤ M
PDE (110.417S11) 5
with the equality if and only if u = M for any (x, y) ∈ ∂BR (x0 , y0 ). This forces
u = M for any (x, y) ∈ BR0 (x0 , y0 ). By connectivity of D, we see that u(x, y) = M
for any (x, y) ∈ D, which completes the proof.
In fact, for the mean value property, we can have another simple proof, without
using the Poisson’s formula.
Another simple proof of the mean value property. Let
Z Z π
1 1
φ(r) := u(~y )dS(~y ) = u((x, y) + r(cos α, sin α))dα .
2πr ∂B(~x,r) 2π −π
Then
Z π
0 1
φ (r) = (cos α, sin α) · (∇u)((x, y) + r(cos α, sin α))dα
2π −π
~y − ~x
Z
1
= · (∇u)(~y )dS(~y )
2πr ∂B(~x,r) r
Z
1 ∂u
= dS(~y )
2πr ∂B(~x,r) ∂~n
Z
1
= ∆u(~y )d~y = 0 .
2πr B(~x,r)
Hence φ is a constant, and so
Z
1
φ(r) = lim φ(t) = lim u(~y )dS(~y ) = u(~x) ,
t→0 t→0 2πt ∂B(~
x,t)
which is the mean value property.
Department of Mathematics, Johns Hopkins University, Baltimore, Maryland 21218
E-mail address: cwang@math.jhu.edu