WAVE EQUATION

BY LI CHEN

Contents
1. Cauchy Problem 1
1.1. Solution formula and existence 1
1.2. Uniqueness — Energy method 6
2. Initial boundary value problem in 1D 9
2.1. Solution formula by separation of variables 10
2.2. Existence of solution for (2.1) 15
2.3. Uniqueness and Stability — Energy estimates 16
2.4. Resonance 16
3. Appendix-On Fourier Series 17
4. Problems 18
References 20

1. Cauchy Problem
1.1. Solution formula and existence.

1.1.1. D’Alembert formula — 1D. We will give the formal solution of Cauchy problem

utt − uxx = 0, in R+ × R
u|t=0 = g(x), (1.1)
ut |t=0 = h(x).

By factorizing the operator ∂tt − ∂xx = (∂t + ∂x )(∂t − ∂x ), we will solve the following two transport
equations,

vt + vx = 0, (1.2)
′
v|t=0 = h(x) − g (x)

and

ut − ux = v, (1.3)
u|t=0 = g(x)

By the solution of transport equation, (1.2) has solution

v(x, t) = h(x − t) + g ′ (x − t).

1
2 BY LI CHEN

(1.3) has solution

∫ t
u(x, t) = g(x + t) + v(x + (t − s), s)ds.
0
Thus the solution of (1.1) is
∫ t
u(x, t) = g(x + t) + h(x + t − 2s) + g ′ (x + t − 2s)ds
0
∫ x+t
1
= g(x + t) + (h(y) − g ′ (y))dy
2 x−t
∫
1 1 1 x+t
= g(x + t) − g(x + t) + g(x − t) + h(y)dy
2 2 2 x−t
∫
1 1 x+t
= (g(x + t) + g(x − t)) + h(y)dy.
2 2 x−t
D’Alembert formula refers to
∫
1 1 x+t
u(x, t) = (g(x + t) + g(x − t)) + h(y)dy. (1.4)
2 2 x−t
which is the formal solution of (1.1).
From this formula, we are ready to get the existence of solution for smooth initial data,
Theorem 1.1. If g ∈ C 2 (R), h ∈ C 1 (R), then u ∈ C 2 (R × [0, +∞)) satisfies wave equation
utt − uxx = 0, and
lim u(x, t) = g(x0 ), lim ut (x, t) = h(x0 ).
(x,t)→(x0 ,0) (x,t)→(x0 ,0)

Some properties of the solution. By using D’Alembert formula, there are some important
sets in (x, t) space.
(1) We call {y ∈ R||y − x| ≤ t} the domain of dependence of point (x, t).
(2) {(x, t) ∈ R × [0, +∞)|x ≥ x1 − t and x ≤ x2 + t} the range of influence of [x1 , x2 ].
(3) {(x, t) ∈ R × [0, +∞)|x ≥ x1 + t and x ≤ x2 − t} the determining region of [x1 , x2 ].
(4) x + t and x − t are characteristics of the wave equation.
Due to characteristics, wave equation has the property of Finite speed propagation of singu-
larity.
Nonhomogeneous problem By the same method we can also find the solution of nonhomoge-
neous problem
utt − uxx = f (x, t), x ∈ R, t > 0 (1.5)
u|t=0 = g(x), ut |t=0 = h(x)
which is,
∫ x+t ∫ t ∫ x+(t−s)
1 1 1
u(x, t) = (g(x + t) + g(x − t)) + h(y)dy + ds f (y, s)dy. (1.6)
2 2 x−t 2 0 x−(t−s)

Theorem 1.2. If g ∈ C 2 (R), h ∈ C 1 (R), f ∈ C 2 (R × [0, +∞)), then u in (1.6), a function in

C 2 (R × [0, +∞)), is a classical solution of (1.5).
By using the solution formula, it is easy to check that
Corollary 1.1. If g, h and f are odd (even, or periodic) in x, so is u.
 WAVE EQUATION 3

1.1.2. Half-line problem. By using extension, we will be able to give the solution formula of the
following problem
utt − uxx = 0, in R+ × R+
u|t=0 = g(x), ut |t=0 = h(x) (1.7)
u|x=0 = 0
For compatibility we need h(0) = g(0) = 0. Due to the homogeneous Dirichlet boundary condition
at x = 0, we use odd extension. Let
{
g(x) x≥0
g̃ =
−g(−x) x < 0

The same extensions for ũ(x, t) and h̃(x). Then we have that ũ satisfies
ũtt − ũxx = 0, in R+ × R
ũ|t=0 = g̃(x),
ũt |t=0 = h̃(x).
By D’Alembert formula, ũ has the representation
∫ x+t
1 1
ũ(x, t) = (g̃(x + t) + g̃(x − t)) + h̃(y)dy.
2 2 x−t

We need to get back to the domain {(x, t) : x > 0, t > 0} and drop the tildes in the formula. There
are two cases, in the case of x ≥ t, our solution has the representation
∫
1 1 x+t
u(x, t) = (g(x + t) + g(x − t)) + h(y)dy.
2 2 x−t
And in the case of 0 ≤ x < t, the solution is
∫ ∫ t−x
1 1 ( x+t )
u(x, t) = (g(x + t) − g(t − x)) + h(y)dy − h(−y)dy .
2 2 0 0
∫ x+t
1 1
= (g(x + t) − g(t − x)) + h(y)dy.
2 2 t−x
Remark 1.1. For nonhomogeneous boundary condition u|x=0 = uD (t), one can use new variable
v = u − uD (t) to do the same discussion, where vtt − vxx = −(uD )tt .

Remark 1.2. It is an easy exercise to get half-line problem with homogeneous Neumann boundary
condition ux |x=0 = 0 by using even extension.

1.1.3. Kirchhoff formula in 3D and Poisson formula in 2D. We will reduce the multi-dimension
problem into a half-line problem by using Spherical mean of the solution.
The spherical mean of a function u(x, t) on ∂B(x, r) is given by
∫
U (x; r, t) = − u(y, t)dSy (1.8)
∂B(x,r)

Lemma 1.1. If u ∈ C m ([0, +∞) × Rn ) is a solution of

utt − ∆u = 0, in (0, +∞) × Rn (1.9)
u|t=0 = g, ut |t=0 = h.
4 BY LI CHEN

Then the spherical mean of u, U (x; r, t) ∈ C m ([0, +∞) × [0, +∞)) satisfies
n−1
Utt − Urr − Ur = 0, in (0, +∞) × R+
r
U |t=0 = G, Ut |t=0 = H,
which is called the Euler-Poisson-Darboux equation.

Proof. By direct calculations, we have

∫ ∫
∂ ∂
Ur (x; r, t) = − u(y, t)dSy = − u(x + rz, t)dSz
∂r ∂B(x,r) ∂r ∂B(0,1)
∫ ∫
y−x
= − ∇u(x + rz, t) · zdSz = − ∇u(y, t) · dSy
∂B(0,1) ∂B(x,r) r
∫ ∫
r
= − ∇u · γdSy = − ∆u(y, t)dy.
∂B(x,r) n B(x,r)

As a consequence,
lim Ur (x; r, t) = 0.
r→0+
If we take one derivative more,
∫ ∫
∂ (r ) 1 ∂ ( 1−n )
Urr (x; r, t) = − ∆u(y, t)dy = r ∆u(y, t)dy
∂r n B(x,r) nα(n) ∂r B(x,r)
∫ ∫
1−n 1 1 ∂
= ∆u(y, t)dy + ∆u(y, t)dy
n α(n)rn B(x,r) nα(n)rn−1 ∂r B(x,r)
∫ ∫
1
= ( − 1) − ∆udy + − ∆udSy .
n B(x,r) ∂B(x,r)

and
1
lim Urr (x; r, t) = ∆u(x, t).
r→0+ n
Then by iteration, if u ∈ C m , we have U ∈ C m .
Back to the first order derivative, by using the wave equation utt − ∆u = 0, we have
∫ ∫
r 1
Ur = − utt dy = utt dy.
n B(x,r) nα(n)rn−1 B(x,r)
Multiplication of it by rn−1 gives
∫
1
rn−1 Ur = utt dy.
nα(n) B(x,r)

The the desired equation follows from Taking one more derivative of it, i.e.
∫ ∫
1
(r n−1
Ur )r = utt dSy = r n−1
− utt dSy = rn−1 Utt .
nα(n) ∂B(x,r) ∂B(x,r)


In the case n = 3, we will get Kirchhoff’s formula by using Euler-Poisson-Darboux equation.
Let Ue = rU , G
e = rG and H e = rH, we have
er = U + rUr ,
U
and moreover,
ett = rUtt = rUrr + 2Ur = (U + rUr )r = U
U err .
 WAVE EQUATION 5

e solves the half-line problem

Now U

ett − U
U err = 0, in R+ × R+
e |t=0
U e U
= G, et |t=0 = H
e
e |r=0 = 0.
U

By solution representation in half-line problem, we have

∫ r+t
e (x; r, t) = 1 (G(r
U e − r)) + 1
e + t) − G(t e
H(y)dy, ∀0 < r < t.
2 2 t−r

Since u(x, t) is a continuous function, its value on (x, t) is exactly the limit of its spherical mean.
Thus we have
e (x; r, t)
U
u(x, t) = lim = lim U (x; r, t)
r→0+ r r→0+
[ ∫ r+t ]
e + t) − G(t
1 G(r e − r) 1
= lim + e
H(y)dy
r→0+ 2 r 2r t−r
= e ′ (t) + H(t)
G e

e and H,
By definition of G e going back to variables g and h, we arrive at
∫ ∫
∂[ ]
u(x, t) = t − gdSy + t − hdSy .
∂t ∂B(x,t) ∂B(x,t)

Then if we do calculation one step more,

∫ ∫
∂ ∂
− g(y)dSy = − g(x + tz)dSz
∂t ∂B(x,t) ∂t ∂B(0,1)
∫ ∫
= − ∇g(x + tz) · zdSz = − ∇g(y) · (y − x)dSy ,
∂B(0,1) ∂B(x,t)

we will have the 3-D Kirchhoff formula,

∫
u(x, t) = − [g(y) + ∇g(y) · (y − x) + th(y)]dSy . (1.10)
∂B(x,t)

In the case n = 2, we will get Poisson’s formula by the method of descent.

If u(x1 , x2 , t) is a solution in 2-D. Let ū(x1 , x2 , x3 , t) = u(x1 , x2 , t), then ū solves the wave equation
in 3-D,

ūtt − ∆ū = 0 in R3 × (0, ∞)

ū|t=0 = ḡ, ūt |t=0 = h̄

where ḡ(x1 , x2 , x3 ) = g(x1 , x2 ) and h̄(x1 , x2 , x3 ) = h(x1 , x2 ). Then by Kirchhoff’s formula in 3-D,
we have
∫ ∫
∂[ ]
u(x, t) = t − ḡ(y)dSy + t h̄(y)dSy ,
∂t ∂B(x̄,t) ∂B(x̄,t)
6 BY LI CHEN

where x̄ = (x1 , x2 , 0). Due to the fact that ḡ(y1 , y2 , y3 ) = g(y1 , y2 ), we can simplify the integral on
∂B(x̄, t) by
∫ ∫
1
− ḡ(y)dSy = ḡdSy
∂B(x̄,t) 4πt2 ∂B(x̄,t)
∫
2 1
= g(y)(1 + |Dγ(y)|2 ) 2 dy,
4πt2 B(x,t)
√
where γ(y) = t2 − (y − x)2 and (1 + |Dγ(y)|2 ) 2 = t(t2 − |y − x|2 )− 2 . Therefore,
1 1

∫ ∫
1 g(y)
− ḡ(y)dSy = dy
∂B(x̄,t) 2πt B(x,t) (t − |y − x|2 ) 12
2
∫
t g(y)
= − dy.
2 B(x,t) (t2 − |y − x|2 ) 12
Then taking derivative with respect to t, we have
[ ∫ ] [ ∫ ]
∂ 2 g(y) ∂ g(x + tz)
t − 1 dy = t− 1 dz
∂t B(x,t) (t2 − |y − x|2 ) 2 ∂t B(0,1) (1 − |z|2 ) 2
∫ ∫
g(x + tz) ∇g(x + tz) · z
= − 1 dz + t − 1 dz
B(0,1) (1 − |z| ) 2
2
B(0,1) (1 − |z|2 ) 2
∫ ∫
g(y) ∇g(y) · (y − x)
= t− dy + t − dy
(t 2 − |y − x|2 ) 12 (t 2 − |y − x|2 ) 12
B(x,t) B(x,t)

Thus the 2-D Poisson’s formula is

∫
1 tg(y) + t2 h(y) + t∇g(y) · (y − x)
u(x, t) = − 1 dy. (1.11)
2 B(x,t) (t2 − |y − x|2 ) 2
By Kirchhoff’s and Poisson’s formula in 3-D and 2-D, we have the following existence result.
Theorem 1.3. n = 2, 3. If g ∈ C 3 (Rn ), h ∈ C 2 (Rn ) with Q = Rn × (0, +∞), then u ∈ C 2 (Q̄) is a
classical solution of (1.9).
Difference of Solution behavior between 3-D and 2-D If we have a closed look at the
Kirchhoff formula (1.10) and the Poisson formula (1.11), we can easily find the main difference is
the integral. Integral over the sphere (which is the boundary of a domain) in Kirchhoff formula and
integral over the ball in Poisson formula.
Let’s assume that the initial data has compact support Ω, where Ω is connected and regular
enough. ∀x0 ̸∈ Ω and d1 = dist(x0 , Ω) > 0, d2 = max{dist(x0 , x) : x ∈ Ω}, then the possible
nonzero point of u(x0 , t) can only be interval [d1 , d2 ]. While in 2-D, the possible nonzero point of
u(x0 , t) must be the half line [d1 , +∞). That’s explained why one can hear the others’ voice in 3-D,
and the water wave diffuse the whole space in 2-D. A picture is needed here.
1.2. Uniqueness — Energy method. With existence theory at hand, uniqueness is a natural
question to ask. Is the existed classical solution unique? Furthermore, is it stable? in which sense?
We first introduce a useful lemma.
Lemma 1.2. (Gronwall’s inequality) Assume G(τ ) ≥ 0, G′ (τ ) ∈ C[0, T ], G(0) = 0 and ∀τ ∈ [0, T ],
the following inequality holds
dG(τ )
≤ CG(τ ) + F (τ )
dτ
 WAVE EQUATION 7

where C is a constant, F (τ ) ≥ 0 nondecreasing in τ . Then

dG(τ )
≤ eCτ F (τ ),
dτ
and
G(τ ) ≤ C −1 (eCτ − 1)F (τ ).

Proof. Multiplying the given inequality by e−Cτ and integrate it on [0, τ ], we have
∫ τ
e−Cτ G(τ ) ≤ e−Ct F (t)dt ≤ F (τ )C −1 (1 − e−Cτ ).
0


The Cauchy problem we considered is revisited here

utt − uxx = f, in R+ × R := Q
u|t=0 = g(x), (1.12)
ut |t=0 = h(x).
The energy inequality of 1-D Cauchy problem (1.12) is

Theorem 1.4. If u ∈ C 1 (R × [0, +∞)) ∩ C 2 (R × (0, +∞)) is a solution of (1.12), then ∀(x0 , t0 ) ∈
R × (0, +∞), we have
∫ (∫ ∫ ∫ )
[u2t (x, τ ) + u2x (x, τ )]dx ≤ M (h2 + gx2 )dx + f 2 (x, t)dxdt ,
Ω Ω0 K
∫ ∫ τ (∫ ∫ ∫ τ )
[u2t (x, t) + u2x (x, t)]dxdt ≤ M (h2 + gx2 )dx + f 2 (x, t)dxdt .
Kτ Ω0 Kτ

where K = {(x, t) ∈ R × [0, +∞) : |x − x0 | < t0 − t}, Kτ = K ∩ {0 ≤ t ≤ τ }, Ωτ = K ∩ {t = τ },

M = e t0 .

Proof. Multiply the equation by ut and integrated on Kτ , we have

∫ ∫ ∫ ∫
(ut utt − ut uxx )dxdt = ut f dxdt
Kτ Kτ

Notice that the boundary of Kτ is ∂Kτ = Ω0 ∪ Ωτ ∪ Γ1τ ∪ Γ2τ , we can calculate the left hand side by
using divergence theorem,
∫ ∫ ∫ ∫
1 2
(ut + u2x )t dxdt − (ut ux )x dxdt
Kτ 2 Kτ
∫
1
= ( (u2t + u2x ), −ut ux )T · γdl
∂K 2
∫ τ ∫
1 2 1 2
= (ut + u2x )dx − (ut + u2x )dx
Ωτ 2 Ω0 2
∫ ∫
1 1 2 1 1
+ √ ( (ut + ux ) + ut ux )dl +
2
√ ( (u2t + u2x ) − ut ux )dl
Γ 1 2 2 2
Γτ 2 2
∫ τ ∫
1 2 1 2
≥ (ut + u2x )dx − (h + gx2 )dx.
Ωτ 2 Ω0 2

where γ is the exterior unit normal vector of ∂Kτ , has values γ = (−1, 0) on Ω0 , γ = (1, 0) on Ωτ ,
γ = √12 (1, −1) on Γ1τ and γ = √12 (1, 1) on Γ2τ .
8 BY LI CHEN

And the right hand side can be estimated by

∫ ∫ ∫ ∫ ∫ ∫
1 1
ut f ≤ 2
ut + f 2.
Kτ 2 Kτ 2 Kτ
Combined the above discussions together, we have
∫ ∫ ∫ ∫ ∫ ∫
(ut + ux ) ≤
2 2 2 2
(h + gx ) + u2t + f 2.
Ωτ Ω0 Kτ Kτ
Now let
∫ ∫ ∫ τ ∫ x0 +(t0 −t)
G(τ ) = (u2t + u2x )dxdt = (u2t + u2x )dxdt,
Kτ 0 x0 −(t0 −t)
∫ ∫ ∫
F (τ ) = (h2 + gx2 )dx + f 2 dxdt.
Ω0 Kτ
Our above estimates is equivalently
dG(τ )
≤ G(τ ) + F (τ )
dτ
where F (τ ) is increasing in τ . Then Gronwall’s inequality implies that
G(τ ) ≤ (eτ − 1)F (τ ) ≤ et0 F (τ ).

We can also get the L2 estimate from the energy estimate.
Theorem 1.5. If u ∈ C 1 (R × [0, +∞)) ∩ C 2 (R × (0, +∞)) is a solution of (1.12), then ∀(x0 , t0 ) ∈
R × (0, +∞),
∫ (∫ ∫ ∫ )
u (x, τ )dx ≤ M1
2 2 2 2
(g + h + gx )dx + f 2 dxdt
Ω Ω0 K
∫ ∫ τ (∫ ∫ ∫ τ )
u2 (x, t)dxdt ≤ M1 (g 2 + h2 + gx2 )dx + f 2 dxdt
Kτ Ω0 Kτ

where M1 = e (e + 1), τ ∈ [0, t0 ] and the definition of domains Kτ , Ωτ and Ω0 are the same as
t0 t0

before.
Proof. We only need to prove that ∥u∥L2 (Ωτ ) and ∥u∥L2 (Kτ ) can be controlled by ∥ut ∥L2 (Kτ ) . In
fact,
∫ ∫ ∫ τ ∫ ∫
(u (x, τ ) − u (x, 0))dx =
2 2
∂t u (x, t)dtdx ≤
2
(u2 + u2t )dxdt.
Ωτ Ωτ 0 Kτ
By Gronwall’s inequality, we have
∫ (∫ ∫ ∫ )
u (x, τ )dx ≤ e
2 t0 2
g (x)dx + u2t dxdt .
Ω Ω0 K
∫ ∫ τ (∫ ∫ ∫ τ )
u2 (x, t)dxdt ≤ et0 g 2 (x)dx + u2t dxdt .
Kτ Ω0 Kτ
2
Thus the L estimate is a direct consequence by energy estimates. 
Uniqueness is a direct corollary of energy estimates. Let Q = R × (0, +∞).
Corollary 1.2. If u1 and u2 are two C 2 (Q) ∩ C 1 (Q̄) solutions of the Cauchy problem (1.12), then
u1 = u2 in Q.
 WAVE EQUATION 9

Proof. Let w = u1 − u2 , then wtt − wxx = 0 in Q and w|t=0 = wt |t=0 = 0. Then energy estimates
for w gives that ∀(x0 , t0 ) ∈ Q, 0 < τ < t0 ,
∫
(w2 + wt2 + wx2 )dx ≤ 0,
Ωτ

which implies w = 0 in Ωτ . Since (x0 , t0 ) is arbitrary, the uniqueness is proved. 

Stability in the sense of H 1 norm, a H 1 norm of a function is defined by ∥u∥H 1 = ∥u∥L2 +∥∇u∥L2 .

Corollary 1.3. If u1 , u2 are C 2 (Q) ∩ C 1 (Q̄) solutions of the Cauchy problem (1.12) with different
data f1 , g1 , h1 and f2 , g2 , h2 . Then
( )
∥u1 − u2 ∥H 1 (Kτ ) ≤ M ∥g1 − g2 ∥H 1 (Ω0 ) + ∥h1 − h2 ∥L2 (Ω0 ) + ∥f1 − f2 ∥L2 (Kτ ) ,

where
∫ ∫
∥u∥2H 1 (Kτ ) = [u2 (x, τ ) + u2t (x, τ ) + u2x (x, τ )]dxdt
Kτ
∫
∥g∥2H 1 (Ω0 ) = [g 2 (x) + gx2 (x)]dx
Ω0
∫
∥h∥2L2 (Ω0 ) = h2 (x)dx
Ω0
∫ ∫
∥f ∥2L2 (Kτ ) = f 2 (x, t)dxdt.
Kτ

Proof. Just notice that w = u1 − u2 is a solution of

wtt − wxx = f1 − f2 in Qm
w|t=0 = g1 − g2 , wt |t=0 = h1 − h2 ,

then using the energy estimates and L2 estimates. 

2. Initial boundary value problem in 1D

We will consider the initial boundary value problem 1-D

utt − uxx = 0 x ∈ (0, 1), t > 0 (2.1)

u|t=0 = g(x), ut |t=0 = h(x)
u|x=0 = u|x=1 = 0.

Here we give the Dirichlet boundary condition. One can also give Neumann boundary condition
and Robin boundary condition.
For derivation of the equation and physical meaning of boundary conditions, check Salsa’s book
[1] Page. 227-228. Page. 230.
The problem (2.1) can be solved by separation of variables. We first give a formal calculation.
Suppose that our solution has a factorized form u(x, t) = X(x)T (t), once we put it into the
equation we will get immediately
X ′′ T ′′
XT ′′ − X ′′ T = 0 ⇒ =
X T
10 BY LI CHEN

Since both side of this equation are functions of different variables x and t, once they are equal,
they must be constant independent of x and t. We denote the constant by −λ. Taking account of
the boundary condition we have
X ′′ + λX = 0, X(0) = X(1) = 0,
′′
T + λT = 0.
Then the solutions with some undetermined constants A, B, C and D are (Later on, we will prove
that λ ≥ 0 so that one can take square root of it)
√ √
X(x) = C cos λx + D sin λx
√ √
T (t) = A cos λt + B sin λt
Boundary conditoin for X shows that
√
C = 0, D sin λ = 0, ⇒ λ = (nπ)2 , n = 1, 2, 3, · · ·
Thus for any fixed n we can have a solution with undetermined constants An and Bn ,
√ √ √
un (x, t) = (An cos λn t + Bn sin λn t) sin λn x
= (An cos nπt + Bn sin nπt) sin nπx.
By superposition principle, we know the finite summation of solutions is still a solution. For any
fixed N , we denote
∑
N ∑
N
uN (x, t) = un (x, t) = (An cos nπt + Bn sin nπt) sin nπx.
n=1 n=1

Now if initial data g(x) and h(x) has the same form, say
∑
N ∑
N
g(x) = gn sin nπx, h(x) = hn sin nπx,
n=1 n=1

then uN must be a solution with this initial data. One could naturally ask how about the case with
general initial data, what is u(x, t) then? Can we use ∞ to replace N ? The problem went directly
to the theory of Fourier Series. We first write initial data g and h into Fourier Series, which can be
done for L2 functions, then the corresponding series lim uN (x, t) can be expected to be a solution
N →∞
for smooth enough initial data. We will prove this seriously later. Before that, we introduce the
so called Sturm-Liouville problem, which is the theoretical basis of the method of separation of
variables.

2.1. Solution formula by separation of variables.

2.1.1. Eigenvalue problem. We will study the following eigenvalue problem

X ′′ + λX = 0, 0<x<1 (2.2)
′
−α1 X (0) + β1 X(0) = 0 α i , βi ≥ 0
′
α2 X (1) + β2 X(1) = 0 αi + βi > 0.

Theorem 2.1. (Sturm-Liouville theorem)

(1) All eigenvalues of (2.2) are nonnegative. In addition, if β1 + β2 > 0, then all eigenvalues
are positive.
 WAVE EQUATION 11

(2) Eigenvalues are countable and increasing to infinity, i.e.

0 ≤ λ1 < λ2 < · · · < λn < · · · , lim λn = ∞.

n→∞

(3) Eigenfunctions of different eigenvalues are orthoganal in the following sense

∫ 1
Xλ Xµ dx = 0, for λ ̸= µ,
0

(4) ∀f ∈ L (0, 1), it holds that

2

∞ ∫1
∑ f (x)Xn (x)dx
2 0
f (x) = Cn Xn (x), in the sense of L (0, 1), Cn = ∫1 ,
n=1 0
Xn2 dx
or
lim ∥f (x) − fn (x)∥L2 = 0
n→∞
∑
n
where fn (x) = Ci Xi (x) is called the generalized Fourier Series.
i=1

Proof. We will only proof the first three statements. The last one can be found in functional analysis
in the part of compact self-adjoint operators.
(1) Multiply the equation by Xλ and integrate it on (0, 1), we have
1 ∫ 1 ∫ 1
′ ′ 2
Xλ Xλ  − (Xλ ) dx + λ Xλ2 dx = 0.
0 0 0

Boundary conditions show that

−α1 Xλ′ (0)Xλ (0) + β1 Xλ2 (0) = 0, −α1 (Xλ′ (0))2 + β1 Xλ (0)Xλ′ (0) = 0
α2 Xλ′ (1)Xλ (1) + β2 Xλ2 (1) = 0, α2 (Xλ′ (1))2 + β2 Xλ (1)Xλ′ (1) = 0.

From these, we get

1
Xλ′ (0)Xλ (0) = (α1 (Xλ′ (0))2 + β1 Xλ2 (0))
α 1 + β1
−1
Xλ′ (1)Xλ (1) = (β2 Xλ2 (1) + α2 (Xλ′ (1))2 ).
α 2 + β2
As a consequence we know the nonnegativity of
∫ 1 ∫ 1
λ Xλ2 dx = (Xλ′ )2 dx − Xλ′ (1)Xλ (1) + Xλ′ (0)Xλ (0) ≥ 0.
0 0

Thus we have λ ≥ 0 and and furthermore

β1 β2
λ = 0 if and only if Xλ′ ≡ 0 and X 2 (0) + X 2 (1) = 0,
α1 + β1 λ α 2 + β2 λ
or equivalently
β1 ( β2 ) 2
Xλ ≡ C and + C = 0.
α1 + β1 α2 + β2
We can see from this expression that if β1 + β2 > 0, then Xλ ≡ 0. In the end, we get that
in the case of β1 + β2 > 0, λ must be positive.
12 BY LI CHEN

(2) We have already λ ≥ 0 and λ = 0 iff β1 = β2 = 0. From X ′′ + λX = 0, we know that there

exist A and B such that
√ √
X(x) = A cos λx + B sin λx,
√ √ √ √
X ′ (x) = −A λ sin λx + B λ cos λx.
We will study it in the following three cases:
(a) Dirichlet boundary√ condition. α1 = α2 = 0, we have X(0) = X(1) = 0. So we have
A = 0 and B sin λ = 0. As a consequence,
λn = (nπ)2 , Xn (x) = sin nπx, n = 1, 2, · · · .
Obviously, in this case, λn is monotone and increasing to ∞.
′ ′
(b) Neumann
√ boundary condition. β1 = β2 = 0, we have X (0) = X (1) = 0, then B = 0,
A sin λ = 0. Thus we have
λn = (nπ)2 , Xn (x) = cos nπx, n = 0, 1, 2, · · · .
(c) Robin boundary condition. α1 β2 + α2 β1 > 0, from the boundary condition we have
√
β1 A − α1 B λ = 0,
√ √ √ √ √
β2 (A cos λ + B sin λ) − α2 λ(A sin λ − B cos λ) = 0.
√
A α1 λ
By calculations, = , and
B β1
√ 1 √ 1
β2 α 1 λ √ + β2 β1 − α1 α2 λ + β1 α2 λ √ = 0.
tan λ tan λ
√
Let ξ = λ, then
(β2 α1 + β1 α2 )ξ
tan ξ = .
α1 α2 ξ 2 − β1 β2
From this formulation, we know that λn is increasing to ∞.
(3) Multiply the equation by Xµ and Xλ separately and integrate on (0, 1), we have
1 ∫ 1 ∫ 1

Xµ Xλ′  − Xµ′ Xλ′ + λ Xλ Xµ = 0
0 0 0
1 ∫ 1 ∫ 1

Xλ Xµ′  − Xµ′ Xλ′ + µ Xλ Xµ = 0
0 0 0

The difference between this two equations shows that

∫ 1 1 1
 
(λ − µ) Xλ Xµ = −Xµ Xλ′  + Xλ Xµ′  .
0 0 0

Now the boundary condition for Xλ and Xµ are

−α1 Xλ′ (0) + β1 Xλ (0) = 0, α2 Xλ′ (1) + β2 Xλ (1) = 0
−α1 Xµ′ (0) + β1 Xµ (0) = 0, α2 Xµ′ (1) + β2 Xµ (1) = 0
These algebraic systems have non zero solutions, thus the coefficient determinents are 0, i.e.
 ′   ′ 
 Xλ (0) Xλ (0)   Xλ (1) Xλ (1) 
  = 0,  
 X ′ (0) Xµ (0)   X ′ (1) Xµ (1)  = 0.
µ µ
 WAVE EQUATION 13

Thus we have ∫ 1
(λ − µ) Xµ Xλ dx = 0.
0
Since λ ̸= µ, we know Xλ and Xµ are authogonal.

Remark 2.1. When β1 = β2 = 0, it is Neumann boundary condition. In this case, λ = 0 is an
eigenvalue, its eigenfunction is X0 = 1.
Remark 2.2. {Xn (x)} is a complete authogonal basis of L2 (0, 1), after normalization it is
Xn (x)
Xn∗ (x) = .
∥Xn (x)∥L2
Then ∀f ∈ L2 (0, 1), the fourier coefficient Cn∗ is
∫1
∗ f (x)Xn (x)dx
Cn = 0
∥xn (x)∥L2
which is the inner product of f (x) and Xn∗ (x).
2.1.2. Separation of variable. Formally the solution of (2.1) is
∞
∑
u(x, t) = (An cos nπt + Bn sin nπt) sin nπx. (2.3)
n=1
Next we will give the method on determining coefficients An and Bn by using initial data.
Take t = 0 in (2.3),
∞
∑ ∞
∑
u(x, 0) = An sin nπx, and ut (x, 0) = nπBn sin nπx.
n=1 n=1
Assume initial data g and h has the following Fourier expansion by sine functions
∞
∑ ∫ 1
g(x) = gn sin nπx, gn = 2 g(x) sin nπxdx,
n=1 0
∞
∑ ∫ 1
h(x) = hn sin nπx, hn = 2 h(x) sin nπxdx.
n=1 0

Then a natural choice of the coefficients are

1
An = gn , Bn = hn .
nπ
Thus the solution expression is
∞
∑ hn
u(x, t) = (gn cos nπt + sin nπt) sin nπx. (2.4)
n=1
nπ
Summary. There are three main steps in separation of variables
(1) Separation of variable formally and set up the eigenvalue problem,
(2) Solve eigenvalue problem, and solve the ODE for T (t),
(3) Summation, fixed the coefficients from initial data.
Questions remained.
(1) How about the solution for other boundary conditions? Neumann and Robin?
(2) Inhomogeneous boundary condition? Homogenization.
14 BY LI CHEN

(3) Inhomogeneous equation utt − uxx = f ?

(4) Under what condition is u(x, t) a solution?
For Neumann and Robin boundary conditions, the related eigenvalue problem is already studied
in Sturm-Liouville theorem. We will answer the remaining three questions in the following three
subsections.

2.1.3. Inhomogeneous equation. We briefly explain how to deal with the non homogeneous equations.
Here we use (0, l) instead of (0, 1).
utt − uxx = f x ∈ (0, l), t > 0 (2.5)
u|x=0 = u|x=l = 0
u|t=0 = g(x) ut |t=0 = h(x).
nπx
Firstly we know that the eigenfunctions are sin , n = 1, 2, · · · . Then assume that
l
∑∞
nπ
u(x, t) = Tn (t) sin x
n=1
l
∞
∑ nπ
f (x, t) = fn (t) sin x
n=1
l
∞
∑ nπ
g(x) = gn sin x
n=1
l
∑∞
nπ
h(x) = hn sin x.
n=1
l

Then solve the ODE for Tn (t),

nπ 2
Tn′′ (t) + ( ) Tn (t) = fn (t)
l
Tn (0) = gn , Tn′ (0) = hn .
One can get that the solution is
∫ t
nπ l nπ l nπ
Tn (t) = gn cos t+ hn sin t+ fn (τ ) sin (t − τ )dτ.
l nπ l nπ 0 l
Then by replacing Tn (t) in the solution u(x, t) by this, we get the solution for inhomogeneous
equation.

2.1.4. Inhomogeneous boundary conditions. The problem with inhomogeneous boundary condition
is
utt − uxx = f x ∈ (0, l), t > 0 (2.6)
u|x=0 = u0 (t) u|x=l = u1 (t)
u|t=0 = g(x) ut |t=0 = h(x).
We will use homogenization technic. Introduce a new function v(x, t) such that the homogeneous
boundary conditions are true for v(x, t), more precisely, let
x l−x
u(x, t) = v(x, t) + u1 (t) + u0 (t),
l l
 WAVE EQUATION 15

then v(x, t) solves

x ′′ l − x ′′
vtt − vxx = f (x, t) − u − u0
l 1 l
v|x=0 = v|x=l = 0
x l−x
v|t=0 = g(x) − u1 (0) − u0 (0)
l l
x l−x ′
vt |t=0 = h(x) − u′1 (0) − u0 (0)
l l
By the method of dealing with inhomogeneous equations, we can get a formula for v(x, t), which
gives the solution formula for u(x, t).

2.2. Existence of solution for (2.1). Now we have the solution formula (2.4). Under what con-
ditions is u a classical solution of (2.1)? We need that u is at least twice differentiable in both x
and t. According to the theory on function Series, we need that
∞
∑ ∞
∑ ∞
∑
un , Dun , D 2 un
n=1 n=1 n=1

are uniformly convergent in (0, 1) × (0, T ).

To have classical solution of (2.1), we also need some compatibility conditions,

g(0) = g(1) = 0, h(0) = h(1) = 0, g ′′ (0) = g ′′ (1) = 0. (2.7)

Theorem 2.2. g ∈ C 3 [0, 1], h ∈ C 2 [0, 1] and they satisfies the compatibility condition (2.7), then
∑∞
u(x, t) = n=1 un (x, t) ∈ C 2 (Q̄) is a solution of (2.1).

Proof. Integral by parts on the coefficients of g and h by using compatibility conditions,

∫ 1 ∫ 1
hn 2 2 2
= h(x) sin nπxdx = − h′′ (x) sin nπxdx := − an
nπ nπ 0 (nπ)3 0 (nπ)3
∫ 1 ∫ 1
2 2
gn = 2 g(x) sin nπxdx = 3
g ′′′ (x) cos nπxdx := bn .
0 (nπ) 0 (nπ)3
Then we have
∞ (
∑ )
2 2
u(x, t) = 3
bn cos nπt − 3
an sin nπt sin nπx.
n=1
(nπ) (nπ)
Moreover the following estimates holds
C C
|un | ≤ , |Dun | ≤ 2
n3 n
C 1
|D2 un | ≤ (|an | + |bn |) ≤ C( 2 + |an |2 + |bn |2 ),
n n
where the right hand side of the last inequality can be bounded by Bessel inequality,
∑∞ ∫ 1 ∞
∑ ∫ 1
|an |2 ≤ 2 |h′′ |2 dx, |bn |2 ≤ 2 |g ′′′ |2 dx.
n=1 0 n=1 0


16 BY LI CHEN

2.3. Uniqueness and Stability — Energy estimates. Let Qτ = (0, 1) × (0, τ ), we have the
following energy estimates for initial boundary value problem of wave equation (2.1) in Qτ . Then
uniqueness and stability can be obtained from that.

Theorem 2.3. Assume u ∈ C 2 (Qτ ) ∩ C 1 (Q̄τ ), then

∫ 1 ∫ 1 ∫
(u2 + u2t + u2x )dx ≤ M ( (h2 + g 2 + gx2 )dx + f 2 dxdt).
0 0 Qτ

Proof. Multiply the wave equation by ut and integrate it on Qτ ,

∫ ∫ ∫
∂ 2
(u + u2x ) ≤ f2 + u2t .
Qτ ∂t t Qτ Qτ

∫ ∫τ ∫1
Notice that Qτ
= 0 0
, we have

∫ 1 ∫ 1 ∫ ∫
(u2t + u2x )|t=τ ≤ 2
(h + gx2 ) + f + 2
u2t .
0 0 Qτ Qτ

By Gronwall’s inequality,
∫ 1 (∫ 1 ∫ )
(u2t + u2x )|t=τ ≤M 2
(h + gx2 ) + f2 .
0 0 Qτ

Similar to the discussion in Cauchy problem, we have the L2 estimates.



2.4. Resonance. Consider initial boundary value problem

utt − uxx = A(x) sin ωt, x ∈ (0, 1), t > 0

u|x=0,1 = 0,
u|t0 = ut |t=0 = 0.

Compatibility conditions A(0) = A(1) = 0 is needed for existence of classical solution. We assume
A ∈ C 1 . The solution formula from separation of variable is

∑∞ ∫ t
1
u(x, t) = fn (τ ) sin nπ(t − τ )dτ · sin nπx
n=1
nπ 0
∑∞ ∫ t
an
= sin nπx sin ωτ · sin nπ(t − τ )dτ,
n=1
nπ 0

where
∫ 1
an = 2 A(x) sin nπxdx.
0
 WAVE EQUATION 17

If we calculate further, we will see

∫ t
sin ωτ · sin nπ(t − τ )dτ
0
∫ t
1
= − (cos(nπt + (ω − nπ)τ ) − cos(−(ω + nπ)τ − nπt))dτ
0 2
∫ t ∫
1 1 t
= cos((ω + nπ)τ − nπt)dτ − cos((ω − nπ)τ + nπt)dτ
2 0 2 0
 t
ω̸=nπ 1 t 1 
= sin((ω + nπ)τ − nπt) − sin((ω − nπ)τ + nπt)
2(ω + nπ) 0 2(ω − nπ) 0
ω̸=nπ 1 1
= (sin ωt + sin nπt) − (sin ωt − sin nπt).
2(ω + nπ) 2(ω − nπ)
If ω = kπ for some k, then
∫
ak ( 2 sin kπt 1 t )
uk (x, t) = − cos kπtdτ sin kπx
kπ kπ + kπ 2 0
( a ak )
k
= sin kπt − t · cos kπt sin kπx.
(kπ)2 2kπ
Thus in the case of ω = kπ, we have
∑( 1 1 )
u(x, t) = (sin ωt + sin nπt) − (sin ωt − sin nπt) sin nπx
2(ω + nπ) 2(ω − nπ)
n̸=k
( a ak )
k
+ sin kπt − t · cos kπt sin nπx
(kπ)2 2kπ
As t → ∞, we have uk (x, t) must blow up at some point.

3. Appendix-On Fourier Series

∀f ∈ L1 (−l, l), it can be written into a Series by using sines and cosines functions
∞
A0 ∑ nπx nπx
f (x) ∼ + (An cos + Bn sin ).
2 n=1
l l
where
∫ l
1 nπx
An = f (x) cos dx, n = 0, 1, 2, · · ·
l −l l
∫ l
1 nπx
Bn = f (x) sin dx, n = 1, 2, · · ·
l −l l

are called Fourier coefficients of f .

If f (x) is an odd function, then Bn = 0, and
∞ ∫
A0 ∑ nπx 2 l
nπx
f (x) ∼ + An cos , An = f (x) cos dx
2 n=1
l l 0 l
If f (x) is an even function, then An = 0, and
∞
∑ ∫ l
nπx 2 nπx
f (x) ∼ Bn sin , Bn = f (x) sin dx.
n=1
l l 0 l
18 BY LI CHEN

A0 ∑
N
nπx nπx
For any fixed N ≥ 1, (SN f )(x) = + (An cos + Bn sin ) is called Trigonometric
2 n=1
l l
polynomials.
nπx mπx
sin , cos , n, m = 1, 2, · · · are orthogonal in the sense that
l l
∫
1 l mπx nπx
cos cos dx = δmn
l −l l l
∫
1 l mπx nπx
sin sin dx = δmn
l −l l l
∫
1 l mπx nπx
sin cos dx = 0
l −l l l
√ nπx √ nπx ∞
Moreover, {1, 2 cos , 2 sin }n=1 is an orthonormal basis in L2 (−l, l), where the inner
l ∫l l
1
product in L2 (−l, l) is defined by f (x)ḡ(x)dx.
2l −l
Theorem 3.1. (Convergence in L2 norm)
lim ∥f (x) − (SN f )(x)∥L2 = 0, for f ∈ L2 (−l, l).
N →∞

Theorem 3.2. (Bessel inequality) For f ∈ L2 (−l, l), it holds

∞ ∫
A20 ∑ 2 1 l 2
+ (An + Bn ) ≤
2
f dx.
2 n=1
l −l

Theorem 3.3. (Parseval’s equality) For f ∈ L2 (−l, l), it holds

∞ ∫
A20 ∑ 2 1 l 2
+ (An + Bn2 ) = f dx.
2 n=1
l −l

4. Problems
F (x − at) + G(x + at)
(1) Verify that u(x, t) = is a solution of
h−x
( x )2 ∂ 2 u ∂ [( x )2 ∂u ]
1− 2
= a2 1−
h ∂t ∂x h ∂x
where h > 0, a > 0 are constants, F, G are any function in C 2 .
(2) (a) Show the general solution of the PDE uxy = 0 is u(x, t) = F (x) + G(y) for arbitrary
function F, G.
(b) Using the change of variables ξ = x + t, η = x − t, show utt − uxx = 0 if and only if
uξη = 0.
(c) Use the above two facts to derive d’Alembert’s formula.
(3) Give energy estimates for half-line problem and the Cauchy problem in Multi-D case.
(4) (Equal partition of energy) Suppose that u ∈ C 2 (R × [0, ∞)) is a solution of the following
Cauchy problem
utt − uxx = 0 (x, t) ∈ R × (0, ∞)
u|t=0 = g, ut |t=0 = h x ∈ R.
 WAVE EQUATION 19

∫ ∞
1
where g, h have compact support. Let kinetic energy be k(t) = u2t (x, t)dx, potential
∫ 2 −∞
1 ∞ 2
energy be p(t) = u (x, t)dx. Try to prove
2 −∞ x
(a) k(t) + p(t) is a constant independent of t.
(b) k(t) = p(t) for large enough t.
(5)
utt − uxx = 0, x ∈ (0, +∞), t ∈ (0, +∞)
u|x=0 = cos ωt
u|t=0 = Ae−x ,
2
ut |t=0 = 0.
Find the condition for A and ω such that solution u ∈ C 2 (R̄+ × R̄+ ), and give this solution
formula.
(6) If u is a classical solution of
utt − uxx = 0, x ∈ (0, 1), t ∈ (0, +∞)
u|x=0 = u|x=1 = 0
u|t=0 = 0, ut |t=0 = x2 (1 − x).
what is the limit ∫ 1
lim (u2t + u2x )dx.
t→+∞ 0
(7) Solve eigenvalue problem
X ′′ (x) + λX(x) = 0, x ∈ (0, l)
X(0) = X ′ (l) = 0.
(8)
X ′′ (x) + λX(x) = 0, x ∈ (0, 1)
′
X (0) + X(0) = 0, X(1) = 0.
(a) Find an eigenfunction with eigenvalue zero. Call it X0 (x).
(b) Find an equation for the positive eigenvalues λ = β 2 .
(c) Show graphically from part (8b) that there are an infinite number of positive eigenval-
ues.
(d) Is there a negative eigenvalue?
(9) Apply separation of variables to get formal solution of
utt − uxx = 0 (x, t) ∈ (0, 1) × (0, ∞)
ux |x=0 = A sin ωt, u|x=1 = 0 t ≥ 0
u|t=0 = 1, ut |t=0 = 0 x ∈ [0, 1].
(10)
utt − uxx = 0, x ∈ (0, 1), t ∈ (0, +∞)
u|x=0 = u|x=1 = 0
u|t=0 = αx4 + βx3 + sin x, ut |t=0 = γ cos x.
20 BY LI CHEN

Solve the problem and give the conditions on α, β and γ such that the solution you gave is
a classical one.
(11) Find the solution of initial boundary values for heat equation by separation of variables.
ut − uxx = sin xπ, x ∈ (0, 1), t ∈ (0, +∞)
u|x=0 = u|x=1 = 0
u|t=0 = 0.
(12) Discussions One can get solution formula of (2.1) by D’Alembert and Fourier series, are
they the same?

References
[1] S. Salsa, Partial differential equations in Action.
[2] L. C. Evans, Partial differential equations.

Department of Mathematical Sciences, Tsinghua University, Beijing, 100084, People’s Republic of

China
E-mail address: [email protected]