5.1. Derivation of The Wave Equation

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Chapter 5

The Wave Equation

In this chapter we investigate the wave equation

(5.1) utt − ∆u = 0

and the nonhomogeneous wave equation

(5.2) utt − ∆u = f (x, t)

subject to appropriate initial and boundary conditions. Here x ∈ Ω ⊂ Rn , t > 0; the


unknown function u = u(x, t) : Ω × [0, ∞) → R.
We shall discover that solutions to the wave equation behave quite differently from solu-
tions of Laplaces equation or the heat equation. For example, these solutions are generally
not C ∞ and exhibit the finite speed of propagation of given disturbances.

5.1. Derivation of the wave equation


The wave equation is a simplified model for a vibrating string (n = 1), membrane (n = 2),
or elastic solid (n = 3). In this physical interpretation u(x, t) represents the displacement
in some direction of the point at time t ≥ 0.
Let V represent any smooth subregion of Ω. The acceleration within V is then
d2
Z Z
udx = utt dx,
dt2 V V

and the net force is Z


F · νdS,
∂V
where F denoting the force acting on V through ∂V , ν is the unit outnormal on ∂V .
Newton’s law says (assume the mass is 1) that
Z Z
utt = F · νdS.
V ∂V

This identity is true for any region, hence the divergence theorem tells that

utt = div F.

1
2 5. The Wave Equation

For elastic bodies, F is a function of Du, i.e., F = F (Du). For small u and small Du,
we use the linearization aDu to approximate F (Du), and so
utt − a∆u = 0,
when a = 1, the resulting equation is the wave equation. The physical interpretation
strongly suggests it will be mathematically appropriate to specify two initial conditions,
u(x, 0) and ut (x, 0).

5.2. One-dimensional wave equations and d’Alembert’s formula


This section is devoted to solving the Cauchy problem for one-dimensional wave equation:
(
utt − uxx = 0 in R × (0, ∞),
(5.3)
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ R,
where g, h are given functions.
Note that
utt − uxx = (ut − ux )t + (ut − ux )x = vt + vx
where v = ut − ux . So v(x, 0) = ut (x, 0) − ux (x, 0) = h(x) − g 0 (x) := a(x). From vt + vx = 0,
we have v(x, t) = a(x − t). Then u solves the nonhomogeneous transport equation
ut − ux = v(x, t), u(x, 0) = g(x).
Solve this problem for u to obtain
Z t
u(x, t) = g(x + t) + v(x + t − s, s) ds
0
Z t
1 x+t
Z
= g(x + t) + a(x + t − 2s) ds = g(x + t) + a(y) dy
0 2 x−t
1 x+t
Z
= g(x + t) + (h(y) − g 0 (y)) dy,
2 x−t
from which we deduce d’Alembert’s formula:
x+t
g(x + t) + g(x − t) 1
Z
(5.4) u(x, t) = + h(y) dy.
2 2 x−t

This formula defines a classical solution u ∈ C 2 if and only if g ∈ C 2 and h ∈ C 1 .


Remark 5.1. (i) When the initial data g and h are not smooth, we still use formula (5.4)
to define a (weak) solution to the Cauchy problem. If g ∈ C k and h ∈ C k−1 , then u ∈ C k .
Thus the wave equation does not have the smoothing effect like the heat equation has.
(ii) Any solution to the wave equation utt = uxx has the form
u(x, t) = F (x + t) + G(x − t)
for appropriate functions F and G. Usually, F (x + t) is called a traveling wave to the
left with speed 1; G(x − t) is called a traveling wave to the right with speed 1.
(iii) The value u(x0 , t0 ) depends only on the values of initial data from x0 − t0 to x0 + t0 ;
in this sense, the interval [x0 − t0 , x0 + t0 ] is called the domain of dependence for the
point (x0 , t0 ). The Cauchy data at (x0 , 0) influence the value of u in the set
I(x0 , t0 ) = {(x, t) | x0 − t < x < x0 + t, t > 0},
5.2. One-dimensional wave equations and d’Alembert’s formula 3

which is called the domain of influence for (x0 , 0). The domain of determinacy of
[x1 , x2 ]; that is, the set of points (x, t) at which the solution is completely determined by
the values of initial data on [x1 , x2 ], is given by
  
x2 − x1
D(x1 , x2 ) = (x, t) x1 + t ≤ x ≤ x2 − t, t ∈ 0, .

2
Lemma 5.1 (Parallelogram property). Let Ω be an open set in R × R̄+ . Then any solution
u of the one-dimensional wave equation in Ω satisfies
(5.5) u(A) + u(C) = u(B) + u(D),
where ABCD is any parallelogram contained in Ω with the slope 1 or −1, with A and C
being two opposite points as shown in Figure 5.1.

0 x

Figure 5.1. Parallelogram ABCD in Lemma 5.1.

We may use this property to solve certain initial and boundary problems.
Example 5.2. Solve the initial boundary value problem:

utt − uxx = 0,
 x > 0, t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), x > 0,

u(0, t) = k(t), t > 0,

where g, h, k are given functions satisfying certain smoothness and compatibility conditions.

Solution. (See Figure 5.2.) If point E = (x, t) is in the region (I); that is, x ≥ t > 0, then
u(x, t) is given by (5.4):
g(x + t) + g(x − t) 1 x+t
Z
u(x, t) = + h(y) dy.
2 2 x−t
In particular, Z 2x
g(2x) + g(0) 1
u(x, x) = + h(y)dy.
2 2 0
4 5. The Wave Equation

Region (II)
A

E Region (I)
B

0 F G x

Figure 5.2. Mixed-value problem in x > 0, t > 0.

If point A = (x, t) is in the region (II); that is, 0 ≤ x < t, then we use the parallelogram
ABCD as shown in the figure to obtain u(x, t) = u(B) + u(D) − u(C), where
u(B) = u(0, t − x) = k(t − x),
Z x+t
x+t x+t g(x + t) + g(0) 1
u(D) = u( , )= + h(y) dy,
2 2 2 2 0
t−x t−x g(t − x) + g(0) 1 t−x
Z
u(C) = u( , )= + h(y) dy.
2 2 2 2 0
Hence, for 0 ≤ x < t,
g(x + t) − g(t − x) 1 x+t
Z
u(x, t) = k(t − x) + + h(y) dy.
2 2 t−x
Therefore the solution to this problem is given by

 g(x + t) + g(x − t) + 1 x+t h(y) dy
R
(0 ≤ t ≤ x),

2 2 x−t
(5.6) u(x, t) =
k(t − x) + g(x + t) − g(t − x) + 1 x+t h(y) dy (0 ≤ x < t).
 R
2 2 t−x

Of course, some smoothness and compatibility conditions on g, h, k are needed in order for
u(x, t) to be a true solution of the problem. Derive such conditions as an exercise. 
Example 5.3. Solve the initial-boundary value problem

utt − uxx = 0,
 x ∈ (0, π), t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ (0, π),

u(0, t) = u(π, t) = 0, t > 0.

Solution. (See Figure 5.3.) We divide the strip (0, π) × (0, ∞) by line segments of slope ±1
starting first at (0, 0) and (π, 0) and then at all the intersection points with the boundaries.
5.2. One-dimensional wave equations and d’Alembert’s formula 5

(IV)

(II) (III)
B

(I)

0 π x

Figure 5.3. Mixed-value problem in 0 < x < π, t > 0.

We can solve u in the region (I) by formula (5.4). In all other regions we use the
parallelogram formula (5.5).
Another way to solve this problem is the method of separation of variables. First
try u(x, t) = X(x)T (t), then we should have
X 00 (x)T (t) = T 00 (t)X(x), X(0) = X(π) = 0,
which implies that
T 00 (t) X 00 (x)
= = λ,
T (t) X(x)
where λ is a constant. From
X 00 (x) = λX(x), X(0) = X(π) = 0,
we find that λ = −j 2 with all j = 1, 2, . . ., and
Xj (x) = sin(jx), Tj (t) = aj cos(jt) + bj sin(jt).
To make sure that u satisfies the initial condition, we consider
X∞
(5.7) u(x, t) = [aj cos(jt) + bj sin(jt)] sin(jx).
j=1

To determine coefficients aj and bj , we use



X ∞
X
u(x, 0) = g(x) = aj sin(jx), ut (x, 0) = h(x) = jbj sin(jx).
j=1 j=1

i.e., the aj and jbj are Fourier coefficients of functions g(x) and h(x). That is,
2 π
Z Z π
2
aj = g(x) sin(jx)dx, bj = h(x) sin(jx)dx.
π 0 jπ 0
Substitute these coefficients into (5.7) and we obtain a formal solution u in terms of trigono-
metric series; the issue of convergence will not be discussed here. 
6 5. The Wave Equation

5.3. The method of spherical means


Suppose u solves the Cauchy problem of n-dimensional wave equation
(
utt − ∆u = 0, (x, t) ∈ Rn × R+ ,
(5.8)
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ Rn .
The idea is to reduce the problem to a problem of one-dimensional wave equation. This
reduction requires the method of spherical means.
For x ∈ R, r > 0, t > 0, define
Z
U (x; r, t) = − u(y, t) dSy = Mu(·,t) (x, r),
∂B(x,r)
Z Z
G(x; r) = − g(y) dSy = Mg (x, r), H(x; r) = − h(y) dSy = Mh (x, r).
∂B(x,r) ∂B(x,r)
Note that
Z Z
1
U (x; r, t) = u(y, t)dSy = − u(x + rξ, t)dSξ .
nαn rn−1 ∂B(x,r) ∂B(0,1)

So, for fixed x, the function U (x; r, t) extends as a function of r ∈ R and t ∈ R+ . One can
recover u(x, t) from U (x; r, t) in terms of
u(x, t) = lim U (x; r, t).
r→0+

5.3.1. The Euler-Poisson-Darboux equation. Let u(x, t) be a smooth solution to


(5.8). We show that, for each x ∈ Rn , the function U (x; r, t) solves a PDE for (r, t) ∈
R+ × R+ .
Theorem 5.4 (Euler-Poisson-Darboux equation). Let u ∈ C m (Rn × [0, ∞)) with m ≥ 2
solve (5.8). Then, for each x ∈ Rn , the function U (x; r, t) ∈ C m ([0, ∞) × [0, ∞)) and solves
the Cauchy problem of the Euler-Poisson-Darboux equation
(
Utt − Urr − n−1
r Ur = 0 in (0, ∞) × (0, ∞),
(5.9)
U = G, Ut = H on (0, ∞) × {t = 0}.

Proof. 1. Note that, for each x ∈ Rn , the regularity of U on (r, t) follows easily as
Z
U (x; r, t) = − u(x + rξ, t)dSξ .
∂B(0,1)

2. Using the formula above, we have


Z Z
r 1
Ur = − ∆u(y, t) dy = utt (y, t) dy.
n B(x,r) nαn rn−1 B(x,r)
Hence !
Z r Z
n−1 1
r Ur = utt (y, t)dSy dρ
nαn 0 ∂B(x,ρ)
and so, differentiating with respect to r yields
Z
n−1 1
(r Ur ) r = utt (y, t)dSy = rn−1 Utt ,
nαn ∂B(x,r)
which expands into the Euler-Poisson-Darboux equation. The initial condition is satisfied
easily; this completes the proof of the theorem. 
5.3. The method of spherical means 7

5.3.2. Solutions in R3 and Kirchhoff ’s formula. For the most important case of three-
dimensional wave equations, we can easily see that the Euler-Poisson-Darboux equation
implies
1
(rU )rr = rUrr + 2Ur = (r2 Ur )r = rUtt = (rU )tt .
r
3
That is, for fixed x ∈ R , the function Ũ (r, t) = rU (x; r, t) solves the 1-D wave equation
Ũtt = Ũrr in r > 0, t > 0, with
Ũ (0, t) = 0, Ũ (r, 0) = rG := G̃, Ũ (r, 0) = rH := H̃.
This is the mixed-value problem studied in Example 5.2; hence, by (5.6) we have
1 t+r
Z
1
Ũ (r, t) = [G̃(r + t) − G̃(t − r)] + H̃(y) dy (0 ≤ r ≤ t).
2 2 t−r
We recover u(x, t) by
Ũ (r, t)
u(x, t) = lim U (x; r, t) = lim
r→0+ r→0+ r
" #
r+t
G̃(r + t) − G̃(t − r)
Z
1
= lim + H̃(y) dy = G̃0 (t) + H̃(t).
r→0+ 2r 2r −r+t

Therefore, we have obtained the so-called Kirchhoff ’s formula for 3-D wave equation:
Z ! Z

u(x, t) = t− g(y) dSy + t − h(y) dSy
∂t ∂B(x,t) ∂B(x,t)
Z
(5.10) = − (th(y) + g(y) + Dg(y) · (y − x)) dSy
∂B(x,t)
Z
1
= (th(y) + g(y) + Dg(y) · (y − x)) dSy (x ∈ R3 , t > 0).
4πt2 ∂B(x,t)

5.3.3. Solutions in R2 by Hadamard’s method of descent. Assume u ∈ C 2 (R2 ×


[0, ∞)) solves problem (5.8) with n = 2. We would like to derive a formula of u in terms of
g and h. The trick is to consider u as a solution to a 3-dimensional wave problem with one
added dimension x3 and then to use Kirchhoff’s formula to find u. This is the well-known
Hadamard’s method of descent.
Define ũ(x̃, t) = u(x, t) for x̃ = (x, x3 ) ∈ R3 , t > 0, where x = (x1 , x2 ) ∈ R2 . Then ũ
solves (
˜ = 0,
ũtt − ∆ũ (x̃, t) ∈ R3 × (0, ∞),
ũ(x̃, 0) = g̃(x̃), ũt (x̃, 0) = h̃(x̃), x̃ ∈ R3 ,
where ∆˜ is the Laplacian in R3 , g̃(x̃) = g(x) and h̃(x̃) = h(x). Let x̄ = (x, 0) ∈ R3 . Then,
by Kirchhoff’s formula,
Z ! Z

u(x, t) = ũ(x̄, t) = t− g̃(ỹ) dSỹ + t − h̃(ỹ) dSỹ ,
∂t ∂ B̃(x̄,t) ∂ B̃(x,t)

where B̃(x̄, t) is the ball in R3 centered at x̄ of radius t. To evaluate this formula, we


parametrize ∂ B̃(x̄, t) by parameter z = (z1 , z2 ) ∈ B(x, t) ⊂ R2 as
p
ỹ = (z, ±γ(z)), γ(z) = t2 − |z − x|2 (z ∈ B(x, t)).
8 5. The Wave Equation

p
Note that Dγ(z) = (x − z)/ t2 − |z − x|2 and thus
p t dz
dSỹ = 1 + |Dγ(z)|2 dz = p ;
t2 − |z − x|2
hence, noting that ∂B(x̃, t) has the top and bottom parts with y3 = ±γ(z),
Z Z
1
− g̃(ỹ) dSỹ = g̃(ỹ) dSỹ
∂ B̃(x̄,t) 4πt2 ∂ B̃(x̄,t)
Z
2 g(z)t dz
= 2
p
4πt B(x,t) t2 − |z − x|2
Z
1 g(z) dz
= p
2πt B(x,t) t2 − |z − x|2
Z
t g(z) dz
= − p .
2 B(x,t) t2 − |z − x|2
R
Similarly, we obtain the formula for −∂ B̃(x̄,t) h̃(ỹ) dSỹ ; finally, we have obtained the so-called
Poisson’s formula for 2-D wave equation:
!
∂ t2 t2
Z Z
g(z) dz h(z) dz
u(x, t) = − p + − p
∂t 2 B(x,t) t2 − |z − x|2 2 B(x,t) t2 − |z − x|2
(5.11)
tg(z) + t2 h(z) + tDg(z) · (z − x)
Z
1
= dz (x ∈ R2 , t > 0).
2πt2 B(x,t)
p
t2 − |z − x|2
Remark 5.2. (i) There are some fundamental differences for the wave equation between
the one dimension and the dimensions n = 2, 3. In both Kirchhoff’s formula (n = 3) and
Poisson’s formula (n = 2), the solution u depends on the derivative Dg of the initial data
u(x, 0) = g(x). For example, if g ∈ C m , h ∈ C m−1 for some m ≥ 1 then u is only C m−1 and
hence ut is only C m−2 (but ut (x, 0) = h(x) ∈ C m−1 ); therefore, there is loss of regularity
for the wave equation when n = 2, 3 (in fact for all n ≥ 2). However, this does not happen
when n = 1, in which u, ut are at least as smooth as g, h.
(ii) There are also some fundamental differences between the 3-D wave equation and
the 2-D wave equation. In R2 , we need the information of initial data g, h in the whole disc
B(x, t) to compute the value u(x, t); that is, the domain of dependence for (x, t) is the
whole disc B(x, t), while in R3 we only need the information of g, h on the sphere ∂B(x, t)
to compute the value u(x, t); that is, the domain of dependence for (x, t) is the sphere
∂B(x, t), not the solid ball B(x, t). In R3 , a “disturbance” initiated at x0 propagates along
the sharp wavefront ∂B(x0 , t) and does not affect the value of u elsewhere; this is known
as the strong Huygens’s principle. In R2 , a “disturbance” initiated at x0 will affect the
values u(x, t) in the whole region |x−x0 | ≤ t. In both cases n = 2, 3 (in fact all cases n ≥ 1),
the domain of influence of the initial data grows (with time t) at speed 1; therefore, the
wave equation has the finite speed of propagation.
(iii) To illustrate the differences between n = 2 and n = 3 of the wave equation, imagine
you are at position x in Rn and there is a sharp initial disturbance at position x0 away from
you at time t = 0. If n = 3 then you will only feel the disturbance (e.g., hear a screaming)
once, exactly at time t = |x − x0 |; however, if n = 2, you will feel the disturbance (e.g., you
are on a boat in a large lake and feel the wave) at all times t ≥ |x − x0 |, although the effect
on you will die out at t → ∞.
5.4. Solutions of the wave equation for general dimensions 9

5.4. Solutions of the wave equation for general dimensions


We now try to find the solution u(x, t) of problem (5.8) in Rn × R+ by solving its spherical
mean U (x; r, t) from the Euler-Poisson-Darboux equation. We need the following useful
result.
Lemma 5.5. Let f : R → R be C m+1 , where m ≥ 1. Then for k = 1, 2, . . . , m,
d2 1 d k−1 2k−1 1 d k 2k 0
(i) ( )( ) (r f (r)) = ( ) (r f (r));
dr2 r dr r dr
k−1
1 d k−1 2k−1 X dj f
(ii) ( ) (r f (r)) = βjk rj+1 j (r),
r dr dr
j=0

where β0k = (2k − 1)!! = (2k − 1)(2k − 3) · · · 3 · 1 and βjk are independent of f .

Proof. Homework. 

5.4.1. Solutions for the odd-dimensional wave equation. Assume that n = 2k + 1


(k ≥ 1) and u ∈ C k+1 (Rn × [0, ∞)) is a solution to the problem (5.8) in Rn × R+ .
As above, let U (x; r, t) be the spherical mean of u(x, t). Then for each x ∈ Rn the
function U (x; r, t) is C k+1 in (r, t) ∈ [0, ∞) × [0, ∞) and solves the Euler-Poisson-Darboux
equation. Let
1 ∂ k−1 2k−1
V (r, t) = ( ) (r U (x; r, t)) (r > 0, t ≥ 0).
r ∂r
Lemma 5.6. We have that Vtt = Vrr and V (0+ , t) = 0 for r > 0, t > 0.

Proof. By part (i) of Lemma 5.5, we have


∂ 2 1 ∂ k−1 2k−1 1 ∂ k 2k
Vrr = ( 2
)( ) (r U) = ( ) (r Ur )
∂r r ∂r r ∂r
1 ∂ k−1 1 2k 1 ∂ k−1 2k−1
=( ) [ (r Ur )r ] = ( ) [r Utt ] = Vtt ,
r ∂r r r ∂r
where we have used the Euler-Poisson-Darboux equation (r2k Ur )r = r2k Utt . Finally, it
follows from part (ii) of Lemma 5.5 that V (0+ , t) = 0. 

Now that
1 ∂ k−1 2k−1 1 ∂ k−1 2k−1
V (r, 0) = ( ) (r U (x; r, 0)) = ( ) [r Mg (x, r)] := G̃(r)
r ∂r r ∂r
and
1 ∂ k−1 2k−1 1 ∂ k−1 2k−1
Vt (r, 0) = ( ) (r Ut (x; r, 0)) = ( ) [r Mh (x, r)] := H̃(r).
r ∂r r ∂r
Hence, by (5.6), we have
1 r+t
Z
1
(5.12) V (r, t) = [G̃(r + t) − G̃(t − r)] + H̃(y)dy (0 ≤ r ≤ t).
2 2 t−r

By (ii) of Lemma 5.5,


k−1
1 ∂ k−1 2k−1 X ∂j
V (r, t) = ( ) (r U (x; r, t)) = βjk rj+1 j U (x; r, t)
r ∂r ∂r
j=0
10 5. The Wave Equation

and hence
V (r, t)
u(x, t) = U (x; 0+ , t) = lim
r→0+ β0k r
" Z r+t #
1 G̃(r + t) − G̃(t − r) 1
= k lim + H̃(y)dy
β0 r→0+ 2r 2r t−r
1 0
= [G̃ (t) + H̃(t)].
β0k

Therefore, we obtain the formula for C k+1 -solution of (2k + 1)-dimensional wave equa-
tion:
   
1 ∂ 1 ∂ k−1  2k−1 1 ∂ k−1  2k−1
(5.13) u(x, t) = k ( ) t Mg (x, t) + ( ) t Mh (x, t) .
β0 ∂t t ∂t t ∂t
Note that when n = 3 (so k = 1) this formula agrees with Kirchhoff’s formula derived
earlier.
In fact, the formula (5.13) defines indeed a classical solution to problem (5.8) under
some smoothness assumption on initial data.
Theorem 5.7 (Solution of wave equation in odd-dimensions). If n = 2k + 1 ≥ 3, g ∈
C k+2 (Rn ) and h ∈ C k+1 (Rn ), then the function u(x, t) defined by (5.13) belongs to C 2 (Rn ×
(0, ∞)), solves the wave equation utt = ∆u in Rn ×(0, ∞), and satisfies the Cauchy condition
in the sense that, for each x0 ∈ Rn ,
lim u(x, t) = g(x0 ), lim ut (x, t) = h(x0 ).
x→x0 , t→0+ x→x0 , t→0+

Proof. We may separate the proof in two cases: (a) g ≡ 0, and (b) h ≡ 0. The proof in
case (a) is given in the text. Here we give a similar proof for case (b) by assuming h ≡ 0.
1. The function u(x, t) defined by (5.13) becomes
 
1 ∂ 1 ∂ k−1  2k−1
u(x, t) = k ( ) t G(x; t) , G(x; t) = Mg (x, t).
β0 ∂t t ∂t
By Lemma 5.5(ii),
k−1  j j+1 G

1 X k j∂ G j+1 ∂
u(x, t) = k β (j + 1)t +t → G(x0 , 0+ ) = g(x0 )
β0 j=0 j ∂tj ∂tj+1

as (x, t) → (x0 , 0+ ). Also from this formula,


2
lim ut (x, t) = lim Gt (x, t).
x→x0 ,t→0+ β0k x→x0 ,t→0+

Note that Z Z
t 1
Gt (x, t) = − ∆g(y) dy = ∆g(y) dy.
n B(x,t) nαn t2k B(x,t)
Hence ut (x, t) → 0 as (x, t) → (x0 , 0+ ).
2. By Lemma 5.5(i),
1  1 ∂ k 2k 1 ∂  1 ∂ k 2k
(5.14) ut (x, t) = (t Gt ), utt (x, t) = (t Gt ).
β0k t ∂t β0k ∂t t ∂t
5.4. Solutions of the wave equation for general dimensions 11

Since
!
Z Z t Z
1 1
Gt (x, t) = ∆g(y) dy = ∆g(y) dSy dρ,
nαn t2k B(x,t) nαn t2k 0 ∂B(x,ρ)

we have Z
2k 2k
(t Gt )t = t − ∆g(y) dSy .
∂B(x,t)
Hence
Z
1 ∂  1 ∂ k−1  1 2k  1 ∂  1 ∂ k−1  2k−1 
utt (x, t) = k (t Gt )t = k t − ∆g dS .
β0 ∂t t ∂t t β0 ∂t t ∂t ∂B(x,t)

On the other hand,


1 ∂  1 ∂ k−1  2k−1 
∆u(x, t) = t ∆G(x; t)
β0k ∂t t ∂t
and Z
∆G(x; t) = − ∆g(y) dSy .
∂B(x,t)
This proves utt = ∆u in Rn × (0, ∞). 
Remark 5.3. (i) In the above theorem, the solution u(x, t) defined by (5.13) can be ex-
tended to t ≤ 0 in the same way as Mg (x, t) and Mh (x, t). Then the extended function u ∈
C 2 (Rn × [0, ∞)) takes the initial data in the classical sense: u(x, 0) = g(x), ut (x, 0) = h(x).
(ii) Let n = 2k + 1 ≥ 3. To compute u(x, t) we need the information of g, Dg, · · · , Dk g
and that of h, Dh, · · · , Dk−1 h only on ∂B(x, t), not on the whole ball B̄(x, t). Therefore,
for the odd-dimensional wave equations, the domain of dependence for (x, t) is also the
sharp wavefront ∂B(x, t); so one still has the strong Huygens’ principle.
(iii) If n = 1, in order for u to be C 2 , we need g ∈ C 2 and h ∈ C 1 . However, if
n = 2k + 1 ≥ 3, in order for u to be C 2 , we need g ∈ C k+2 and h ∈ C k+1 . So, the solutions
in general lose k-orders of smoothness from the initial data.

5.4.2. Solutions for the even-dimensional wave equation. Assume that n = 2k is


even. Suppose u is a C k+1 solution to the Cauchy problem (5.8). Again we use Hadamard’s
method of descent similarly as in the case n = 2.
Let x̃ = (x, xn+1 ) ∈ Rn+1 , where x ∈ Rn . Set
ũ(x̃, t) = u(x, t), g̃(x̃) = g(x), h̃(x̃) = h(x).
Then ũ is a C k+1 solution to the wave equation in Rn+1 × (0, ∞) with initial data ũ(x̃, 0) =
g̃(x̃), ũt (x̃, 0) = h̃(x̃). Since n + 1 = 2k + 1 is odd, we use (5.13) to obtain ũ(x̃, t) and then
u(x, t) = ũ(x̄, t), where x̄ = (x, 0) ∈ Rn+1 . In this way, we obtain
   
1 ∂ 1 ∂ k−1  2k−1 1 ∂ k−1  2k−1
u(x, t) = ( ) t Mg̃ (x̄; t) + ( ) t Mh̃ (x̄; t) .
(2k − 1)!! ∂t t ∂t t ∂t
Note that
Z
1
Mg̃ (x̄, t) = g̃(y, yn+1 )dS
tn (n + 1)αn+1 2
yn+1 +|x−y|2 =t2
Z
2 g(y)
= n−1
p dy,
t (n + 1)αn+1 B(x,t) t − |y − x|2
2
12 5. The Wave Equation

t 2
since dS = √ 2 dy on the surface yn+1 + |y − x|2 = t2 . Similarly we have
t −|y−x|2
Z
2 h(y)
Mh̃ (x̄, t) = n−1
p dy.
t (n + 1)αn+1 B(x,t) t2 − |y − x|2
Therefore, we have the following representation formula for even n:
2 n ∂ h 1 ∂ n−2 Z g(y) dy i
u(x, t) = ( ) 2 p
(n + 1)!!αn+1 ∂t t ∂t B(x,t) t2 − |y − x|2
(5.15) Z
1 ∂ n−2 h(y) dy o
+( ) 2 p .
t ∂t B(x,t) t2 − |y − x|2
When n = 2, this reduces to Poisson’s formula for 2-D wave equation obtained above.

Theorem 5.8 (Solution of wave equation in even-dimensions). If n = 2k ≥ 2, g ∈ C k+2 (Rn )


and h ∈ C k+1 (Rn ), then the function u(x, t) defined by (5.15) belongs to C 2 (Rn × (0, ∞)),
solves the wave equation utt = ∆u in Rn × (0, ∞), and satisfies the Cauchy condition in the
sense that, for each x0 ∈ Rn ,

lim u(x, t) = g(x0 ), lim ut (x, t) = h(x0 ).


x→x0 , t→0+ x→x0 , t→0+

Proof. This follows from the theorem in odd dimensions. 

Remark 5.4. (i) In the theorem, note that


Z Z
g(y) dy n−1 g(x + tz) dz
p =t p
B(x,t) t2 − |y − x|2 B(0,1) 1 − |z|2
g(x+tz) dz
√ is in C k+2 (Rn × R). Hence, the solution
R
and that the function G(x, t) = B(0,1) 1−|z|2
u(x, t) defined by (5.15) can be extended to t ≤ 0 and the extended function u ∈ C 2 (Rn ×
[0, ∞)) takes the initial data in the classical sense: u(x, 0) = g(x), ut (x, 0) = h(x).
(ii) Let n = 2k ≥ 2. To compute u(x, t) we need the information of g, Dg, · · · , Dk g and
that of h, Dh, · · · , Dk−1 h in the solid ball B(x, t). Therefore, for the even-dimensional wave
equations, the domain of dependence for (x, t) is the whole solid ball B̄(x, t).
(iii) If n = 2k ≥ 2, in order for u to be C 2 , we need g ∈ C k+2 and h ∈ C k+1 . So, again,
the solutions in general lose k-orders of smoothness from the initial data.

5.4.3. Solution of the wave equation from the heat equation*. We study another
method of solving the odd-dimensional wave equations by the heat equation.
Suppose u is a bounded, smooth solution to the Cauchy problem
(
utt − ∆u = 0 in Rn × (0, ∞)
(5.16)
u(x, 0) = g(x), ut (x, 0) = 0 on x ∈ Rn ,

where n is odd and g is smooth with nice decay at ∞. We extend u to negative times by
even extension of t and then define
Z ∞
1 s2
v(x, t) = 1/2
u(x, s)e− 4t ds (x ∈ Rn , t > 0).
(4πt) −∞
5.4. Solutions of the wave equation for general dimensions 13

Then v is bounded,
Z ∞
1 2
− s4t
∆v(x, t) = ∆u(x, s)e ds
(4πt)1/2 −∞
Z ∞
1 s2
= 1/2
uss (x, s)e− 4t ds
(4πt) −∞
Z ∞
1 s − s2
= us (x, s) e 4t ds
(4πt)1/2 −∞ 2t
Z ∞  s2
1 1  − s2
= u(x, s) − e 4t ds
(4πt)1/2 −∞ 4t2 2t
and Z ∞  s2
1 1  − s2
vt (x, t) = u(x, s) − e 4t ds.
(4πt)1/2 −∞ 4t2 2t
Moreover
lim v(x, t) = g(x) (x ∈ Rn ).
t→0+
Therefore, v solves the Cauchy problem for the heat equation:
(
vt − ∆v = 0 in Rn × (0, ∞)
v(x, 0) = g(x) on x ∈ Rn .
As v is bounded, by uniqueness, we have
Z
1 |y−x|2
− 4t
v(x, t) = g(y)e dy (x ∈ Rn , t > 0).
(4πt)1/2 Rn
We have two formulas for v(x, t) and take 4t = 1/λ in the two formulas to obtain
Z ∞
1  λ  n−1
Z
−λs2 2
e−λ|y−x| g(y) dy
2
u(x, s)e ds =
0 2 π Rn
(5.17) n−1 Z ∞
nαn λ 2
  2
= e−λr rn−1 G(x; r) dr
2 π 0
for all λ > 0, where Z
G(x; r) = Mg (x, r) = − g(y) dSy .
∂B(x,r)
So far, we have not used the odd dimension assumption. We will solve for u from (5.17)
when n = 2k + 1 ≥ 3 is odd. Noticing that − 2r 1 d −λr2 ) = λe−λr2 , we have
dr (e
Z ∞ Z ∞
n−1
−λr2 n−1 2
λ 2 e r G(x; r) dr = λk e−λr r2k G(x; r) dr
0 0
(−1)k ∞ h 1 d k −λr2 i 2k
Z
= (e ) r G(x; r) dr
2k 0 r dr
Z ∞ h
1 1 ∂ k 2k−1 i 2
= k r (r G(x; r)) e−λr dr,
2 0 r ∂r
where we integrated by parts k times (be careful with the operator ( 1r dr
d k
) ). We can then
write (5.17) as
Z ∞ Z ∞ h
2 nαn 1 ∂ k 2k−1 i 2
u(x, r)e−λr dr = k k+1 r (r G(x; r)) e−λr dr
0 π 2 0 r ∂r
14 5. The Wave Equation

for all λ > 0. If we think of r2 as τ , then this equation says that the Laplace transforms of
two functions of τ are the same; therefore, the two functions of τ must be the same, which
also implies the two functions of r are also the sam. So we obtain
 
nαn  1 ∂ k 2k−1 nαn ∂ 1 ∂ k−1  2k−1
(5.18) u(x, t) = k k+1 t (t G(x; t)) = k k+1 ( ) t G(x; t) ,
π 2 t ∂t π 2 ∂t t ∂t
which, except for the constant, agrees with the formula (5.13) with h ≡ 0. In fact the
n/2
constant here, using αn = Γ(πn+2 ) , Γ( 12 ) = π 1/2 and Γ(s + 1) = sΓ(s) for all s > 0,
2

nαn (2k + 1)π 1/2 1 1


k k+1
= 1 = = k
π 2 k+1
2 Γ(k + 1 + 2 ) (2k − 1)!! β0
is also in agreement with the constant in (5.13).

5.5. Nonhomogeneous wave equations and Duhamel’s principle


We now turn to the initial value problem for nonhomogeneous wave equation
(
utt − ∆u = f (x, t) in Rn × (0, ∞)
(5.19)
u(x, 0) = 0, ut (x, 0) = 0 on x ∈ Rn ,
where f (x, t) is a given function.
Motivated by Duhamel’s principle used to solve the nonhomogeneous heat equations,
for each s ≥ 0, let U (x, t; s) be the solution to the homogeneous Cauchy problem
(
Utt (x, t; s) − ∆U (x, t; s) = 0 in Rn × (s, ∞),
(5.20)
U (x, s; s) = 0, Ut (x, s; s) = f (x, s) on x ∈ Rn .
Define
Z t
(5.21) u(x, t) = U (x, t; s)ds
0
Note that if v = v(x, t; s) is the solution to the Cauchy problem
(
vtt − ∆v = 0 in Rn × (0, ∞),
(5.22)
v(x, 0) = 0, vt (x, 0) = f (x, s) on x ∈ Rn ,
then U (x, t; s) = v(x, t − s; s) for all x ∈ Rn , t ≥ s. Therefore,
Z t Z t
u(x, t) = U (x, t; s) ds = v(x, t − s; s) ds (x ∈ Rn , t > 0).
0 0
n
Theorem 5.9. Assume n ≥ 2 and f ∈ C [ 2 ]+1 (Rn × [0, ∞)). Let U (x, t; s) be the solution
of (5.20). Then the function u defined by (5.21) is in C 2 (Rn × [0, ∞)) and a solution to
(5.19).

Proof. 1. The regularity of f guarantees a solution U (x, t; s) is given by (5.13) if n is odd


or (5.15) if n is even. In either case, u ∈ C 2 (Rn × [0, ∞)).
2. A direct computation shows that
Z t Z t
ut (x, t) = U (x, t; t) + Ut (x, t; s)ds = Ut (x, t; s)ds,
0 0
Z t Z t
utt (x, t) = Ut (x, t; t) + Utt (x, t; s)ds = f (x, t) + Utt (x, t; s)ds,
0 0
5.5. Nonhomogeneous wave equations and Duhamel’s principle 15

Z t
∆u(x, t) = ∆U (x, t; s)ds.
0
Hence utt − ∆u = f (x, t). Clearly u(x, 0) = ut (x, 0) = 0. 
Example 5.10. Find a solution of the following problem
(
utt − uxx = tex , (x, t) ∈ R × (0, ∞),
u(x, 0) = 0, ut (x, 0) = 0.

Solution. First, for each s ≥ 0 solve


(
vtt = vxx in R × (0, ∞),
v(x, 0) = 0, vt (x, 0) = sex .
From d’Alembert’s formula, we have
Z x+t
1 1
v = v(x, t; s) = sey dy = s(ex+t − ex−t ).
2 x−t 2
1 x+t−s
Hence U (x, t; s) = v(x, t − s; s) = 2 s(e − ex+s−t ) and so
Z t
1 t
Z
u(x, t) = U (x, t; s)ds = s(ex+t−s − ex+s−t ) ds
0 2 0
1
= [ex+t (−te−t − e−t + 1) − ex−t (tet − et + 1)]
2
1
= (−2tex + ex+t − ex−t ).
2

Example 5.11. Find a solution of the Cauchy problem
(
utt − ∆u = f (x, t), x ∈ R3 , t > 0,
u(x, 0) = 0, ut (x, 0) = 0, x ∈ R3 .

Solution. By Kirchhoff’s formula, the solution v of the Cauchy problem


(
vtt − ∆v = 0, x ∈ R3 , t > 0,
u(x, 0) = 0, ut (x, 0) = f (x, s), x ∈ R3
is given by Z
1
v = v(x, t; s) = f (y, s) dSy (x ∈ R3 , t > 0).
4πt ∂B(x,t)
Hence
Z t Z tZ
1 f (y, s)
u(x, t) = v(x, t − s; s) ds = dSy ds
0 4π 0 ∂B(x,t−s) t − s
Z tZ
1 f (y, t − r)
= dSy dr
4π 0 ∂B(x,r) r
f (y, t − |y − x|)
Z
1
= dy.
4π B(x,t) |y − x|
Note that the domain of dependence (on f ) is the finite set {(y, t − |y − x|) | y ∈ B̄(x, t)},
which is the boundary of a solid cone in R3 × R+ ; the integrand on the right is called a
retarded potential. 
16 5. The Wave Equation

5.6. Energy method and the uniqueness


There are some subtle issues about the uniqueness of the Cauchy problem of wave equations.
The formulas (5.13) for odd n or (5.15) for even n hold under more and more smoothness
conditions of the initial data g, h as the dimension n gets larger and larger. For initial
data not too smooth we cannot use such formulas to claim the uniqueness. Instead, we
use certain quantities that behave nicely for the wave equation. One such quantity is the
energy.

5.6.1. Domain of dependence. Let u ∈ C 2 be a solution to the wave equation utt = ∆u


in Rn × (0, ∞). Fix x0 ∈ Rn , t0 > 0, and consider the backward wave cone with apex
(x0 , t0 ):
K(x0 , t0 ) = {(x, t) | 0 ≤ t ≤ t0 , |x − x0 | ≤ t0 − t}.
Theorem 5.12 (Domain of dependence). Let u ∈ C 2 (Rn × [0, ∞)) solve the wave equation
utt = ∆u in Rn × (0, ∞). If u = ut = 0 on B(x0 , t0 ) × {t = 0}, then u ≡ 0 within K(x0 , t0 ).

Proof. Define the local energy


Z
1  
e(t) = u2t (x, t) + |Du(x, t)|2 dx (0 ≤ t ≤ t0 ).
2 B(x0 ,t0 −t)
Then
1 t0 −t
Z Z  
e(t) = u2t (x, t) + |Du(x, t)|2 dSx dρ
2 0 ∂B(x0 ,ρ)
and so, by the divergence theorem,
Z t0 −t Z  
0
e (t) = ut utt (x, t) + Du · Dut (x, t) dSx dρ
0 ∂B(x0 ,ρ)
Z
1  
− u2t (x, t) + |Du(x, t)|2 dSx
2 ∂B(x0 ,t0 −t)
Z Z
1 
2 2

= (ut utt + Du · Dut )dx − u (x, t) + |Du(x, t)| dS
B(x0 ,t0 −t) 2 ∂B(x0 ,t0 −t) t
(5.23) Z Z
∂u
= ut (utt − ∆u)dx + ut dS
B(x0 ,t0 −t) ∂B(x0 ,t0 −t) ∂ν
Z
1  
− u2 (x, t) + |Du(x, t)|2 dS
2 ∂B(x0 ,t0 −t) t
Z  ∂u 1 1 
= ut − u2t (x, t) − |Du(x, t)|2 dS ≤ 0
∂B(x0 ,t0 −t) ∂ν 2 2
because the last integrand is less than zero; in fact,
∂u 1 2 1 ∂u 1 1 1
ut − ut − |Du|2 ≤ |ut || | − u2t − |Du|2 ≤ − (|ut | − |Du|)2 ≤ 0.
∂ν 2 2 ∂ν 2 2 2
0
Now that e (t) ≤ 0 implies that e(t) ≤ e(0) = 0 for all 0 ≤ t ≤ t0 . Thus ut = Du = 0, and
consequently u ≡ 0 in K(x0 , t0 ). 
Theorem 5.13 (Uniqueness of Cauchy problem for wave equation). Given any f, g, h the
Cauchy problem
(
utt − ∆u = f (x, t), x ∈ Rn , t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ Rn
5.6. Energy method and the uniqueness 17

can have at most one solution u in C 2 (Rn × [0, ∞)).

Example 5.14. (a) Show that there exists a constant K such that
K
|u(x, t)| ≤U (0) ∀ 0 < t < T
t
whenever T > 0 and u is a smooth solution to the Cauchy problem of 3-D wave equation
(
utt − ∆u = 0, x ∈ R3 , 0 < t < T,
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ R3 ,

where U (0) = R3 (|g| + |h| + |Dg| + |Dh| + |D2 g|) dy.


R

(b) Let u be a smooth solution to the wave equation utt = ∆u in R3 × (0, ∞) satisfying
Z
1
lim (|u(x, t)| + |Du(x, t)| + |ut (x, t)| + |Dut (x, t)| + |D2 u(x, t)|) dx = 0.
t→∞ t R3

Show that u ≡ 0 on R3 × (0, ∞).

Proof. The details are left as Homework. Part (b) follows from (a) by considering ũ(x, t) =
u(x, T − t) on Rn × (0, T ). 

5.6.2. Energy method for mixed-value problem of wave equation. Let Ω be a


bounded smooth domain in Rn and let ΩT = Ω × (0, T ]. Let
Γ T = ∂ 0 ΩT = ΩT \ ΩT .
We are interested in the uniqueness of initial-boundary value problem (mixed-value problem)

utt − ∆u = f in ΩT ,

(5.24) u=g on ΓT ,

ut = h on Ω × {t = 0}.

Theorem 5.15 (Uniqueness of mixed-value problem for wave equation). There exists at
most one solution u ∈ C 2 (ΩT ) of mixed-value problem (5.24).

Proof. Let u1 , u2 be any two solutions; then v = u1 − u2 is in C 2 (ΩT ) and solves



vtt − ∆v = 0 in ΩT ,

v = 0 on ΓT ,

vt = 0 on Ω × {t = 0}.

Define the energy


Z 
1 
e(t) = vt2 (x, t) 2
+ |Dv(x, t)| dx (0 ≤ t ≤ T ).
2 Ω

Then, by the divergence theorem


Z Z Z
0 ∂v
e (t) = (vt vtt + Dv · Dvt )dx = vt (vtt − ∆v)dx + vt dS = 0
Ω Ω ∂Ω ∂ν
since v = 0 on ∂Ω for all 0 < t < T implies that vt = 0 on ∂Ω for all 0 < t < T. Therefore
v ≡ 0 from v = 0 on ΓT . 
18 5. The Wave Equation

5.6.3. Other initial and boundary value problems. Uniqueness of wave equation can
be used to find the solutions to some mixed-value problems. Since solution is unique, any
solution found in special forms will be the unique solution.
Example 5.16. Solve the Cauchy problem of the wave equation
(
utt − ∆u = 0 in R3 × (0, ∞),
u(x, 0) = 0, ut (x, 0) = h(|x|),
where h(r) is a given function.

Solution. In theory, we could use Kirchhoff’s formula to find the solution; however, the
computation would be too complicated. Instead, we can try to find a solution in the form
of u(x, t) = v(|x|, t) by solving an equation for v, which becomes exactly the Euler-Poisson-
Darboux equation that can be solved easily when n = 3; some condition on h is needed in
order to have a classical solution. Details are left as an exercise. 
Example 5.17. Let λ ∈ R and λ 6= 0. Solve
(
utt − ∆u + λu = 0 in Rn × (0, ∞),
u(x, 0) = g(x), ut (x, 0) = h(x).

Solution. We use the idea of Hadamard’s descent method. We first make u a solution to
the wave equation in Rn+1 × (0, ∞) and recover u by this solution.
If λ = µ2 > 0 (the equation is called the Klein-Gordon equation), let v(x̃, t) =
u(x, t) cos(µxn+1 ), where x̃ = (x, xn+1 ) ∈ Rn+1 and x ∈ Rn .
If λ = −µ2 < 0, let v(x̃, t) = u(x, t)eµxn+1 . Then, in both cases, v(x̃, t) solves the wave
equation and can be solved by using the formula (5.13) or (5.15). Then we recover u(x, t)
in both cases by u(x, t) = v(x̄, t), where x̄ = (x, 0) ∈ Rn+1 . 
Example 5.18. Solve

utt (x, t) − ∆u(x, t) = 0,
 x = (x0 , xn ) ∈ Rn−1 × R+ , t > 0,
u(x, 0) = g(x), ut (x, 0) = h(x), xn > 0,

u(x0 , 0, t) = 0, x0 ∈ Rn−1 .

Solution. We extend the functions g, h to odd functions g̃ and h̃ in xn ; e.g., g̃(x0 , −xn ) =
−g̃(x0 , xn ) for all xn ∈ R and g̃(x0 , xn ) = g(x0 , xn ) when xn > 0. We then solve
(
ũtt − ∆ũ = 0, x ∈ Rn , t > 0,
ũ(x, 0) = g̃(x), ũt (x, 0) = h̃(x), x ∈ Rn .
Since V (x, t) = ũ(x0 , xn , t) + ũ(x0 , −xn , t) solves
Vtt − ∆V = 0, V (x, 0) = 0, Vt (x, 0) = 0,
the uniqueness result implies V ≡ 0, i.e., ũ is an odd function in xn . Hence u = ũ|xn >0 is
the solution to the original problem. 
Example 5.19. Let Ω be a bounded domain. Solve

utt − ∆u = f (x, t)
 in Ω × (0, ∞),
u(x, 0) = g(x), ut (x, 0) = h(x), x ∈ Ω,

u(x, t) = 0, x ∈ ∂Ω, t ≥ 0.

5.7. Finite speed of propagation for second-order linear hyperbolic equations 19

Solution. Use the method of separation variables and try to find the solution of the form

X
u= uj (x)Tj (t),
j=1

where
−∆uj = λj uj , uj |∂Ω = 0,
Tj00 (t) − λj T (t) = wj (t), Tj (0) = aj , Tj0 (0) = bj ,
By the elliptic theory, eigenfunctions {uj (x)}∞ 2
j=1 form an orthonormal basis of L (Ω), and
wj (t), aj , bj are the Fourier coefficients of w(x, t), g(x) and h(x) with respect to uj (x),
respectively.
The question is whether the series gives indeed a true solution; we do not study such
questions in this course.


5.7. Finite speed of propagation for second-order linear hyperbolic


equations
We study a class of special second-order linear partial differential equations of the form
utt + Lu = 0 (x ∈ Rn , t > 0),
where L has a special form
n
X
Lu = − aij (x)Dij u,
i,j=1
with smooth symmetric coefficients ij uniform ellipticity condition on Rn .
(a (x)) satisfying
In this case, we say the operator ∂tt + L is uniformly hyperbolic.
Let (x0 , t0 ) ∈ Rn × (0, ∞). Assume q(x) is a continuous function on Rn , positive and
smooth in Rn \ {x0 } and q(x0 ) = 0. Consider a curved backward cone
C = {(x, t) ∈ Rn × (0, t0 ) | q(x) < t0 − t}
and for each 0 ≤ t ≤ t0 let
Ct = {x ∈ Rn | q(x) < t0 − t}.
Assume ∂Ct = {x ∈ Rn | q(x) = t0 − t} is a smooth surface for each t ∈ [0, t0 ). In addition,
we assume
Xn
(5.25) aij (x)Di q(x)Dj q(x) ≤ 1 (x ∈ Rn \ {x0 }).
i,j=1

Lemma 5.20. Let β(x, t) be a smooth function and


Z
α(t) = β(x, t) dx (0 < t < t0 ).
Ct
Then Z Z
0 β(x, t)
α (t) = βt (x, t) dx − dSx .
Ct ∂Ct |Dq(x)|
Proof. This follows from the co-area formula. 
Theorem 5.21 (Domain of dependence). Let u be a smooth solution to utt + Lu = 0 in
Rn × [0, ∞). If u = ut = 0 on C0 , then u ≡ 0 with the cone C.
20 5. The Wave Equation

Proof. Define the local energy


Z n
1  X 
e(t) = u2t + aij Di uDj u dx (0 ≤ t ≤ t0 ).
2 Ct i,j=1

Then the lemma above implies


Z n Z n
0
 X
ij 1   X  1
e (t) = ut utt + a Di uDj ut dx − u2t + aij Di uDj u dS
Ct 2 ∂Ct |Dq|
i,j=1 i,j=1

:= A − B.
Note that aij Di uDj ut = Dj (aij ut Di u)−ut Dj (aij Di u) (no sum); hence, integrating by parts
and by the equation utt + Lu = 0, we have
Z  n
X  Z n
X
A= ut utt − Dj (aij Di u) dx + ut aij (Di u)νj dS
Ct i,j=1 ∂Ct i,j=1
Z n
X Z n
X
=− ut Di uDj aij dx + ut aij (Di u)νj dS,
Ct i,j=1 ∂Ct i,j=1

where ν = (µ1 , ν2 , · · · , νn ) is the outer unit normal on ∂Ct . In fact,


Dj q
νj = (j = 1, 2, · · · , n) on ∂Ct .
|Dq|
Pn ij
Since hv, wi = i,j=1 a vi wj defines an inner product on v, w ∈ Rn , by Cauchy-Schwartz’s
inequality,
Xn Xn n
1/2  X 1/2
aij (Di u)νj ≤ aij Di uDj u aij νi νj .


i,j=1 i,j=1 i,j=1

Therefore
Z n
X n
1/2  X 1/2
|A| ≤ Ce(t) + |ut | aij Di uDj u aij νi νj dS
∂Ct i,j=1 i,j=1
Z n n
1  X  X 1/2
≤ Ce(t) + u2t + ij
a Di uDj u ij
a νi νj dS.
2 ∂Ct i,j=1 i,j=1

However, since νj = (Dj q)/|Dq|, by (5.25), we have


n
X 1/2 1
aij νi νj ≤ on ∂Ct .
|Dq|
i,j=1

Consequently, we derive that |A| ≤ Ce(t) + B and thus

e0 (t) ≤ Ce(t) (0 < t < t0 ).

Sine e(0) = 0, this gives e(t) ≡ 0. Hence u ≡ 0 within C. 


5.7. Finite speed of propagation for second-order linear hyperbolic equations 21

5.7.1. Energy method for mixed-value problems. Let Ω be a bounded smooth do-
main in Rn and let ΩT = Ω × (0, T ]. Let
ΓT = ∂ 0 ΩT = ΩT \ ΩT .
We are interested in the uniqueness of initial-boundary value problem (mixed-value problem)
(
utt + Lu + B(x, t) · Du + c(x, t)u = f in ΩT ,
(5.26)
u = g on ΓT , ut = h on Ω × {t = 0},
where Lu is defined as above, B and c are bounded functions on ΩT , and g, h are given
functions.
Theorem 5.22 (Uniqueness of mixed-value problem). There exists at most one solution
u ∈ C 2 (ΩT ) of mixed-value problem (5.26).

Proof. Let u1 , u2 be any two solutions; then v = u1 − u2 is in C 2 (ΩT ) and solves


(
vtt + Lv + B(x, t) · Dv + c(x, t)v = 0 in ΩT ,
v = 0 on ΓT , vt = 0 on Ω × {t = 0}.
Define the energy
Z h n
1 X i
e(t) = vt2 (x, t) + v 2 (x, t) + aij (x)vxi vxj dx (0 ≤ t ≤ T ).
2 Ω i,j=1
Then
Z  n
X 
0
(5.27) e (t) = vt vtt + vvt + aij vxi vxj t dx.
Ω i,j=1

Since aij vxi vxj t =(aij vxi vt )xj − aij ij


xj vxi vt − a vxi xj vt
and since v = 0 on ∂Ω for all 0 < t < T
implies that vt = 0 on ∂Ω for all 0 < t < T , it follows by the divergence theorem that
Z Z Z
ij
a (x)vxi vxj t dx = a vxi vt νj dS − (aij
ij ij
xj vxi vt + a vxi xj vt ) dx
Ω ∂Ω Ω
Z
= − (aij ij
xj vxi vt + a vxi xj vt ) dx.

Therefore, using the equation vtt + Lv = −B · Dv − cv, by (5.27), we obtain that
Z  n
X 
e0 (t) = vt v − B · Dv − cv − aij
xj xi dx
v
Ω i,j=1
Z Z  n
X 
≤C (vt2 + v 2 + |Dv|2 ) dx ≤ C vt2 + v 2 + aij vxi vxj dx,
Ω Ω i,j=1
where the last inequality follows from the uniform positivity of matrix (aij ). Hence e0 (t) ≤
Ce(t) on (0, T ). Note that e(t) ≥ 0 and e(0) = 0; this implies e(t) ≡ 0 on [0, T ], proving
v ≡ 0 on ΩT . 

You might also like