Partial Differential Equations (Week 9+10)

The wave equation

Gustav Holzegel
March 21, 2019

1 Introduction
This week we begin the study of the wave equation in R1+d :

u = −∂t2 u + ∂x21 u + ∂x22 u + . . . + ∂x2d u = 0 (1)

Just as the Laplace equation is the prototype of and elliptic equation, the wave
equation is the prototype of a hyperbolic1 equation. We already discussed the
case d = 1 of one spatial dimension in Week 4, when we derived d’Alembert’s
formula and introduced the notions of domain of dependence and domain of
influence. In the general case we computed the characteristic hypersurfaces and
saw that they were given as solutions of the eikonal equation. We also observed
that t = 0 is non-characteristic, which means that the Cauchy Kovalevskaya
theorem will apply for Cauchy data (specifying u and the normal derivative)
prescribed on t = 0.
Our plan of action is the following
1. Derive and understand geometrically the basic energy estimate for the
wave equation and the domain of dependence property from this point
of view (cf. our discussion of Fritz John’s global Holmgren’s theorem dis-
cussed previously); applications
2. Derive an explicit solution formula for (1) in physical space in the physi-
cally most interesting case d = 3, illustrating the strong Huygen’s principle
and the loss of regularity of solutions if measured at the level of C k ; decay
of solutions
3. Derive an explicit solution formula for (1) in Fourier space giving yet
another perspective on the conservation of energy and regularity at the
L2 level
1 There a various definitions of this. We use it here to mean that the Cauchy problem is well-

posed on a non-characteristic hypersurface. While an elliptic equation has no-characteristic

hypersurfaces, hyperbolic equations in some sense have as many as possible. You should
compare with Exercise 2.21 of Week 3.

1
 4. Prove well-posedness for general linear equations u = bi (t, x) ∂i u + c · u
After this we will hopefully be in a position to discuss some non-linear ap-
plications in the last week.

2 The energy estimate

Let us assume that we have a classical C 2 solution of u = 0 on [0, T ]× Rd with
“data” u (0, x) = u0 (x) and ∂t u (0, x) = u1 (x). Multiplying the wave equation
by −∂t u yields
1 2
0 = −u · ∂t u = ∂t (∂t u) − ∇x (∂t u∇x u) + ∇x ∂t u · ∇x u = 0
2
or
1 h 2
i
= ∂t (∂t u) + |∇x u|2 − ∇x (∂t u∇x u) (2)
2
If we integrate this over the spacetime slab [0, T ] × Rd , then assuming that u
decays sufficiently rapidly near infinity (more on this below!) we would obtain
the energy conservation law
Z h i Z h i
2 2
dd x (∂t u) + |∇x u|2 = dd x (∂t u) + |∇x u|2
t=T t=0

As this works for any τ ≤ T we obtain

k∂t u (τ, ·) kL2x + ku (τ, ·) kH̊ 1 = ku1 kL2 + ku0 kH̊ 1 (3)
x

In order to derive this identity we have assumed that u is C 2 and that it vanishes
sufficiently rapidly near spatial infinity in order to make the boundary term
arising from ∇x (∂t u∇x u) vanish. We will now see that we can do much better
if we suitably localize the estimate.
Fix T > 0, R > 0 and consider a region
[
K= {τ } × BR+T −τ (4)
τ ∈[0,T ]

where BR+T −τ is the ball of radius B + T − τ centered at the origin.

τ =T

BT +R−τ

τ =0
BT +R

2
You may think
 of thisregion as a cut-off (at t = 0 and t = T ) past light cone
with tip at T + R, ~0 . We will denote the boundary of BR+T −τ in R3 by
SR+T −τ and the unit outward normal to this boundary by N .
Integrating (2) over the region K then yields
1
Z h i
dd x (∂t u)2 + |∇x u|2
2 {t=T }×BR
Z T Z  
1 2 1 2
+ dt (∂t u) + |∇x u| − ∂t u · N u dσSR+T −τ
0 {τ }×SR+T −τ 2 2
1
Z h i
2
= dd x (∂t u) + |∇x u|2 (5)
2 {t=0}×BR+T

It is not hard to see using Cauchy-Schwarz that the integrand in the second
line is non-negative. We can actually obtain something more quantitative. Let
us denote the induced gradient on the spheres SR+T −τ by ∇ / (i.e. the derivatives
tangent to these d − 2 dimensional spheres). We may decompose

∂t = N + V

where V is a derivative tangent to the wall of the cone2 Then, from the easily
verified identities
2
−∂t uN u = − (N u) − N u · V u
1 1 2 1 2
∂t u∂t u = (N u) + N u · V u + (V u)
2 2 2
1 1 1
|∇x u|2 = (N u)2 + |∇u|
/ 2
2 2 2
we see that (5) becomes
1
Z h i
dd x (∂t u)2 + |∇x u|2
2 {t=T }×BR
T  
1 1
Z Z
2 2
+ dt /
(V u) + |∇u| dσSR+T −τ
0 {τ }×SR+T −τ 2 2
1
Z h i
2
= dd x (∂t u) + |∇x u|2 (6)
2 {t=0}×BR+T

This identity is truly remarkable and illustrates the domain of dependence

property of the wave equation. Indeed, we certainly have
Z h i Z h i
2 2 2
d
d x (∂t u) + |∇x u| ≤ dd x (∂t u) + |∇x u|2 (7)
{t=T }×BR {t=0}×BR+T

and hence
2 In polar coordinates
q ∂t = ∂r + (∂t − ∂r ) since the wall of the cone is given by zero set of
x21 + . . . x2d − R − T = t − T + r − R, so that indeed (∂t − ∂r ) H = 0.
H (t, x1 , . . . , xd ) = t +

3
 S
Corollary 2.1. Suppose u = 0 in {t = 0} × BR+T . Then u = 0 in τ ∈[0,T ] {τ }×
BR+T −τ .
Corollary 2.2. Two C 2 solutions u and v in K = τ ∈[0,T ] {τ } × BR+T −τ that
S

satisfy u = v and ∂t u = ∂t v on {t = 0} × BR+T have to agree in all of K.

Exercise 2.3. Can you generalize this domain of dependence/ uniqueness prop-
erties to more general wave equations? Hint: Gronwall’s inequality. You should
also compare and contrast with the statement we obtained from Holmgren’s the-
orem.
Let us understand a bit better the underlying geometry of this computation.
The expression (2) is apparently a boundary term and it will induce different
expressions dependent on the geometry of the boundary hypersurfaces. What is
useful in the estimates is if the expressions induced are non-negative, as it was
the case for the hypersurfaces of constant t and the characteristic hypersurfaces
discussed above. More generally, we define
Definition 2.4. We will call a hypersurface spacelike if it can be represented
locally as f = 0 for f satisfying
2
− (∂t f ) + |∇x f |2 < 0 (8)
We call it timelike if “<” above is replaced by “>” and null if “<” is replaced
by “=”.
Remark 2.5. For thoseP familiar with Minkowski geometry and its inner-product
< x, y >= −x0 y0 + di=1 xi yi (giving rise to a notion of timelike, spacelike and
null vectors depending on the sign of < x, x >), note that the expression (8)
is precisely the Minkowski norm of the gradient of f . Hence a hypersurface is
spacelike if its normal vector is timelike etc.
Exercise 2.6. Obtain the energy estimate for two homologous spacelike hyper-
surfaces S1 and S2 , i.e. spacelike hypersurfaces with common boundary ∂S1 =
∂S2 bounding a region. Hint: Observe that integrating (2) over a spacelike
hypersurface produces after using (8) a non-negative expression controlling all
derivatives.
Note that given the domain of dependence property we can combine the
Cauchy-Kovalevskaya theorem with the energy estimate to obtain a notion of
generalized solution (how?).
It is remarkable that the energy estimate (which is at the level of L2 ) does
not lose regularity: It relates the H 1 × L2 norm for data to the same norm for
the solution at any later time. We will see that this property does not hold at
the C k level.
Finally, it’s important to realize the flexibility of the energy method. In prin-
ciple you can add lower order terms and apply Gronwall. Unlike in Holmgren’s
theorem (which you can now revisit and interpret the one-parameter family of
non-characteristic hypersurfaces as foliation of spacelike slices for which you
could apply the energy estimate), the method does not require analyticity of
the coefficients.

4
2.1 Pointwise bounds
We can obtain pointwise bounds for solutions to the wave equation using Sobolev
embedding. We give two different ways of doing this:

Method 1: Commute the equation with a basis of the tangent space of Rn ,

i.e. with ∂x1 , . . . , ∂xd . We obviously have

[, ∂xi ] = 0

The underlying is the translation invariance of the wave operator. Applying the
energy estimate to each of the commuted equations we immediately obtain

ku (τ, ·) kH̊ 1 + ku (τ, ·) kH̊ 2 ≤ ku1 kH 1 + ku0 kH̊ 1 + ku0 kH̊ 2 (9)

Now in R3 we have the Sobolev embedding

sup |u| ≤ C (ku (τ, ·) kL6 + kDu (τ, ·) kL6 ) ≤ C ku (τ, ·) kH̊ 1 + ku (τ, ·) kH̊ 2

for u of compact support, which implies

sup |u| ≤ C ku1 kH 1 + ku0 kH̊ 1 + ku0 kH̊ 2

global pointwise control on the solution.

Method 2: Commute with ∂t . We obviously have

[, ∂t ] = 0

The underlying is the translation time translation invariance of the wave opera-
tor. Applying the energy estimate to the ∂t -commuted equation we immediately
obtain
Z Z Z Z
|∆u|2 = |∂t2 u|2 ≤ |∂t2 u|2 + |∇x ∂t u|2 = |∆u|2 + |∇x u1 |2
t=τ t=τ t=0 t=0
≤ |u0 |2H̊ 2 + |u1 |2H̊ 1
(10)

Now using elliptic theory we know that for u of compact support the right hand
side controls all second derivatives. Combining this with the energy estimate
as the uncommuted level yields the same pointwise bound as in Method 1 after
applying the Sobolev embedding.

3 Kirchhoff’s formula for the wave equation

We now switch gears a bit and leave the estimates behind to obtain an explicit
representation formula for solution to (1) in terms of initial data. The formula

5
will be useful to illustrate the Huygen’s principle and also the loss of regularity
at the level of C k regularity.
We consider the Cauchy problem

utt − c2 ∆u = 0
u|t=0 = f (11)
ut |t=0 = g

For h (x) = h (x1 , . . . , xd ) continuous from Rd → R we define

1
Z
Mh (x, r) = h (y) dSy (12)
ωd rn−1 |y−x|=r

the average over a sphere of radius r around x (ωd denoting the area element of
the unit-sphere in d dimensions). Writing y = x + rξ with |ξ| = 1 we have

1
Z
Mh (x, r) = h (x + rξ) dSξ .
ωd |ξ|=1

We can extend Mh (x, r) to an even function defined for all real

r (change
k d k d+1


ξ → −ξ). Observe also that h ∈ C R implies M h ∈ C R .
Now, for h ∈ C 2 Rd we find using the divergence theorem the identity

∂2
 
d−1 ∂
+ Mh (x, r) = ∆x Mh (x, r) (13)
∂r2 r ∂r

This is known as the Darboux equation. Note the “initial conditions”, Mh (x, 0) =
h (x) and ∂r Mh (x, r) = 0 since Mh is even in r.
r=0
To derive (13), note

1 1
Z Z
∂r Mh (x, r) = ξ · ∇y h (x + rξ) dSξ = ξ · ∇ξ h (x + rξ) dSξ
ωd |ξ|=1 ωd r |ξ|=1
1 r
Z Z
= ∆ξ h (x + rξ) dξ = ∆x h (x + rξ) dξ
ωd r |ξ|<1 ωd |ξ|<1
Z 1 Z 1
r
Z
= ∆x dρρd−1 h (x + rρξ) dSξ = r∆x dρρd−1 Mh (x, rρ)
ωd 0 |ξ|=1 0

and after a further change of variables

Z r
∂r Mh (x, r) = r−d+1 ∆x dρ ρd−1 Mh (x, ρ) , (14)
0

which yields

∂r rd−1 ∂r Mh (x, r) = ∆x rd−1 Mh (x, r)

(15)

and hence (13).

6
 The idea to solve (1) is to write down an equation for the spherical means
of u. This will be a 1 + 1 dimensional PDE which we can solve explicitly.
Conversely, we shall be able to recover the solution from its spherical means.
We define
1
Z
Mu (x, r, t) = u (x + rξ, t) dSξ (16)
ωd |ξ|=1
Note that u (x, t) = Mu (x, 0, t), recovering u from its means. Now by the Wave
equation and by Darboux we know that we have
 2 
∂ d−1 ∂
∂t2 Mu = ∆x Mu = + Mu .
∂r2 r ∂r
Let us restrict to d = 3. In this case we have
∂2 ∂2
2
(rMu ) = c2 2 (rMu ) .
∂t ∂r
Hence we have that rMu satisfies the 1 + 1 dimensional wave equation with
initial values:
rMu = rMf (x, r) and ∂t (rMu ) = rMg (x, r) at t = 0 (17)
and hence d’Alembert’s formula applies. Therefore,
1
rMu (x, r, t) = [(r + ct) Mf (x, r + ct) + (r − ct) Mf (x, r − ct)]
2
1 r+ct
Z
+ ξMg (x, ξ) dξ (18)
2c r−ct
Dividing by r and taking the limit r → 0 (Exercise – use that Mg (x, r) is even!)
we finally find
u (x, t) = tMg (x, ct) + ∂t (tMf (x, ct))
!
1 ∂ 1
Z Z
= g (y) dSy + f (y) dSy (19)
4πc2 t |y−x|=ct ∂t 4πc2 t |y−x|=ct

We conclude:
Proposition 3.1. Any solution u of the initial value problem (11) which is C 2
for t ≥ 0 in n = 3 space dimensions
is given by formula (19) and is hence
unique. Conversely, given f ∈ C 3 R3 and g ∈ C 2 R3 the u (x, t) defined by

the above formula is C 2 and satisfies (11).

Note the loss of regularity! To make this loss more manifest, we compute
!
1
Z
∂t (tMf (x, ct)) = Mf (x, ct) + t∂t f (x + ctξ) dSξ
ωd |ξ|=1
1 y−x
Z
= Mf (x, ct) + t Df (x + ctξ) · dSξ (20)
ωd |ξ|=1 ct

7
and obtain
!
1
Z X
u (x, t) = tg (y) + f (y) + fyi (yi − xi ) dSy . (21)
4πc2 t2 |y−x|=ct i

Besides the aforementioned loss of regularity, many more things can be read
off from the above formula (19). One is the strong Huygen’s principle. The
solution at a point (x, t > 0) depends only on the “data” of the surface S (x, ct)
the intersection of the past light cone with the hypersurface t = 0. It does not
depend on the values inside the ball B (x, ct). This sharp propagation of signals
is special for the wave equation in odd spatial dimensions (with the exception
d = 1). In even dimensions one only has the weak Huygen’s principle (value at
(t, x) depends on the values in the entire ball B (x, ct)) and for most hyperbolic
equations one also only has the weak form. Conversely, the data near a point y
on the initial hyperplane t = 0 only influence the solution at points (t, x) near
the cone |x − y| = ct emanating from y.
Finally, the formula (19) allows us to show that the solution decays in time
(we already know it spreads over larger and larger regions of space). Suppose
you have initial data of compact support at t = 0, the support being contained
in a large ball B (0, ρ). Now consider a point (t, x) for some large t. It is clear
that the past light cone from (t, x) can only intersect the ball B (0, ρ) is a set
of area 4πρ2 . Thus the solution satisfies |u (t, x) | ≤ Ct−1 with the constant
depending on the size of the support.
An important goal is to establish such decay estimates for more general
(possibly non-linear) wave equations. This will require more stable methods
than the explicit solution formula derived above and we’ll mention some in the
last week of the course.

3.1 General dimensions

A similar formula to (19) can be derived for general d via Hadamard’s method
of decent. As mentioned, in even dimensions the solution at (t, x) will depend
on the values in the entire ball |x − y| ≤ ct (not only the sphere |x − y| = ct) of
the data. See the books of Evans or John.

3.2 Duhamels principle

Consider the inhomogeneous wave equation with trivial data

utt − ∆u = f (t, x) in Rd × (0, ∞)

(22)
u = 0 , ut = 0 for Rd × {0}

We define ũ = ũ (x, t; s) to be the solution of

ũtt (·, s) − ∆ũ (·; s) = 0 in Rd × (s, ∞)

ũ (·, s) = 0 , ũt (·, s) = f (·, s) for Rd × {t = s} (23)

8
Now set
Z t
u (x, t) := ũ (x, t; s) ds f or x ∈ Rd and t ≥ 0 (24)
0

This is a solution of the problem (22):

Proposition 3.2. Let d = 3 and f ∈ C 2 Rd × (0, ∞) . Then u defined by (24)

solves (22).
Proof. Note first that by our well-posedness result for the homogeneous wave
2
equation, the ũ (x, t; s) are well-defined and
C in all its arguments for 0 ≤ s ≤ t
2 d
(why?). Hence u (x, t) is C R × [0, ∞) . Clearly, the trivial initial conditions
of (22) are also satisfied. Clearly, the trivial initial conditions of (22) are also
satisfied. To see that it also solves the inhomogenous wave equation ,we compute
Z t Z t
ut (x, t) = ũ (x, t; t) + ũt (x, t; s) ds = ũt (x, t; s) ds , (25)
0 0

Z t Z t
utt (x, t) = ũt (x, t; t) + ũtt (x, t; s) ds = f (x, t) + utt (x, t; s) ds . (26)
0 0

Combining this with

Z t
∆u (x, t) = ∆ũ (x, t; s) ds (27)
0

yields the result.

4 Fourier synthesis
As for the Schroedinger and heat equation, we can follow a Fourier based ap-
proach. Assuming we have a solution of the problem (11) in the Schwartz space,
we obtain the second order ODE

ûtt = −c2 |ξ|2 û (28)

which is solved by
sin (c|ξ|t)
û (t, ξ) = fˆ (ξ) cos (c|ξ|t) + ĝ (ξ) (29)
c|ξ|

Theorem 4.1. If f, g ∈ S Rd , then there exists a unique u ∈ C ∞ R, S Rd

solving the initial value problem (11).
The proof of this theorem is as in the case of the Schroedinger and the heat
equation. Note that taking a time derivative only leads to polynomial growth
in ξ which is easily compensated for by the Schwartz property of fˆ and ĝ.

9
 It is a straightforward exercise to derive the identity

|ût |2 + c2 |ξ|2 |û|2 = |ĝ|2 + c2 |ξ|2 |fˆ (ξ) |2 (30)

which when integrated over al ξ yields the familiar energy conservation (in
particular, the Schwartz space property ensures the “vanishing sufficiently fast
near infinity property” stated loosely in the first section above).
Of course more is true in the sense that the identity (30) holds for any
frequency ξ in itself!
Exercise 4.2. Show that for s ≥ 1 real the expression

kut k2H(s−1) + c2 k∇x uk2H(s−1) (31)

is independent of t. Relate this to the conservation laws derived in physical space

(for integer s).
Using these estimates one can – again as for the Schroedinger and heat
equation – derive a notion of generalized solutions associated with data f, g ∈
H(s) × H(s−1) ....
Exercise 4.3. Use the domain of dependence property to remove the assumption
that the data are Schwartz from

Theorem 4.1 to obtain existence and uniqueness
of a solution u ∈ C ∞ Rd+1 from data f, g ∈ C ∞ (R).

5 Back to the estimates

Our final goal is to prove the following well-posedness theorem, which we will
state in the smooth category. The proof will suggest appropriate Sobolev ver-
sions.
Theorem 5.1. Let u0 ∈ C ∞ Rd , u1 ∈ C ∞ Rd
and bα , c, F ∈ C ∞ Rd+1 .

Then there exists a unique solution u ∈ C ∞ Rd+1 of the wave equation

bµ (x, t) ∂µ u + c (x, t) u + F (x, t) = 0

P
u +
(32)
u (0, ·) = u0 , ∂t u (0, ·) = u1 .

Moreover, any solution v ∈ C 2 Rd+1 of (32) satisfying v (0, x) = u0 (x),

∂t v (0, x) = u1 (x) for all x ∈ X ⊂ Rd satisfies u = v in the domain of de-

pendence of X.
Note that we already proved the domain of dependence property.

5.1 The inhomogenous wave equation

Recall the Duhamel formula (24). Let us apply the energy estimate in the
inhomogeneous case,
u = F

10
for F a given function of x and t. Multiplying again by −∂t u and integrating
over the cut-off cone K (4) now produces (Exercise) in view of
Z Z Z T Z
∂t u · F ≤ |∂t u|2 + |F |2 ≤ dt |∂t u|2 + |F |2 (33)
K K 0 {t=τ }×BR+T −τ

the estimate
Z τ
E (τ ) ≤ dτ̃ E (τ̃ ) + (E (0) + G (τ )) (34)
0

where
1 1
Z
E (τ ) = (∂t u)2 + |∇x u|2 (35)
{t=τ }×BR+T −τ 2 2

and
Z τ Z
G (τ ) = |F |2 . (36)
0 {t=τ }×BR+T −τ

Gronwall’s inequality yields the estimate

E (τ ) ≤ [E (0) + G (τ )] eτ . (37)

Note that for trivial data there is a gain of one derivative compared with the
inhomogeneity measured here at the level of L2 . Again, at the C k level there is
no gain (we needed F ∈ C 2 to get u ∈ C 2 when deriving Duhamel’s formula).
You should compare with the case of elliptic regularity, where one could gain
two derivatives (at Sobolev or Hoelder level).

5.2 Lower order perturbations

Let us now try to prove an energy estimate for the operator (32). The main
part of the estimate will be the same as for the box, except that now we have
to control an additional term of the form
Z hX i
(∂t u) bµ (x, t) ∂µ u + c (x, t) u + F (x, t)
K
Z h i Z Z
2
≤C (∂t u) + |∇x u|2 + C u2 + C F2
K K K
Z h i Z Z
2 2 2
≤C (∂t u) + |∇x u| + C F + u2 (38)
K K {t=0}×BR+T

Here the last step follows from a Poincare inequality (Exercise). The constants
depend on the T of the region K. With the usual definitions of energy (35) and
the inhomogeneous term (36) we find
Z Z τ
E (τ ) = E (0) + u2 + G (τ ) + dτ̃ E (τ̃ ) (39)
{t=0}×BR+T 0

11
for which Gronwall’s inequality yields
Z !
E (τ ) ≤ E (0) + u2 + G (τ ) eCτ (40)
{t=0}×BR+T

We actually have (Exercise – this is a byproduct of the previous Exercise on the

Poincare inequality)
Z Z !
u2 + E (τ ) ≤ E (0) + u2 + G (τ ) eCτ (41)
{t=τ }×BR+T −τ {t=0}×BR+T

which yields
h p i
ku (τ, ·) kH 1 (BR+T −τ ) ≤ C ku (0, ·) kH 1 (BR+T ) + G (τ ) eCτ (42)

Exercise 5.2. Obtain higher order Sobolev estimates as well as pointwise bounds
for u! Hint: Use commutation as in Section 2.1.

5.3 The proof of Theorem 5.1

With these estimates at hand, we can prove Theorem 5.1. The idea is to use
the fact that we know how to solve the inhomogeneous equation

u = G (x, t)

for a given G (x, t) and trivial data using the Duhamel formula.3 We let

Φ : C ∞ [0, T ] × Rd → C ∞ [0, T ] × Rd

G 7→ u (43)

be the map which maps the smooth inhomogeneity to a smooth solution of the
inhomogeneous wave equation with trivial data. In view of the energy estimate
for any integer k ≥ 1

ku (τ, ·) kH k (Rd ) ≤ CkGkH k−1 ([0,T ]×Rd )

this map extends to a bounded map (also denoted by Φ)

Φ : H k [0, T ] × Rd → H k+1 [0, T ] × Rd

We can write the wave equation as

X 
u=Φ bµ (x, t) ∂µ u + c (x, t) u + F (x, t) (44)

We now show the following (Exercise)

3 Note that the case of general initial data can be reduced to the case of trivial data by

considering the equation for v = u − f − g · t.

12
 1. The map v 7→ Φ (
bµ (x, t) ∂µ v + c (x, t)
P

v + F (x, t)) is a bounded map
from H 1 [0, T ] × Rd to H 1 [0, T ] × Rd
2. The map is a contraction (for sufficiently small T ).
By Banach’s fixed point theorem there exists a unique fixed point u ∈ H 1 which
is our solution to the wave equation. The fact that it is more regular (in fact
smooth) is obtained a posteriori from commutation.

5.4 Cauchy Kovalevskaya

Find an alternative proof of Theorem 5.1 using Cauchy Kovalevkaya, Holmgren’s
theorem and the energy estimate. Hint: Approximate first the data and then
the coefficients in the equation by analytic functions.

6 Exercises
1. (Fritz John, Problem 1, Chapter 5.1(c))
Let S denote a space like hyperplane with equation t = γx1 in xt-space.
Show that the Cauchy problem for u = 0 with data on S can be reduced
to the initial value problem (i.e. posed at t′ = 0) for the same equation by
introducing new independent variables x′ , t′ by the Lorentz transformation
x1 − γc2 t t − γx1
x′1 = p , x′2 = x2 , x′3 = x3 , t′ = p .
1 − γ 2 c2 1 − γ 2 c2

2. (Fritz John, Problem 1, Chapter 5.1(a))

(a) Show that for n = 3 the general solution of (11) with spherical sym-
metry about the origin has the form (r = |x|)
F (r + ct) + G (r − ct)
u (t, r) =
r
with suitable F, G. (Hint: See Week 4!)
(b) Show that the solution with initial data of the form
u=0 , ut = g (r)
with g an even function of r is given by
Z r+ct
1
u (t, r) = ρg (ρ) dρ (45)
2cr r−ct
(c) For 
1 for 0 < r < a
g (r) = (46)
0 for r > a
find u explicitly from (45) in the different regions bounded
 by the
cone r = a ± ct in R4 . Show that u is discontinuous at ~0, t = a . c

13
 Note that the above exercise illustrates the loss of regularity at the C k
level due to focussing: u is perfectly smooth (in fact trivial) initially.
3. This exercise illustrates that the L2 energy estimate for the wave equation
is very special in the sense that it is generally not possible to control the
Lp norm at some later time from the Lp norm of the data if p 6= 2. To
establish this, we will use Exercise 2, i.e. work with spherically symmetric
solutions and initial data f = 0 and g supported in a small ball around the
origin. Then use that the support spreads to a (large) annulus to establish
Littman’s theorem: For d = 3, p 6= 2 and t 6= 0 we have

kut (t) kLp

sup = ∞.
g∈S(Rd )\0 kut (0) kLp

You can find more on this in Rauch’s book, end of Section 4.7.

14