The wave equation

Intoduction to PDE

1 The Wave Equation in one dimension

The equation is
∂ 2u 2
2∂ u
− c = 0. (1)
∂t2 ∂x2
Setting ξ1 =
x + ct, ξ2 = x − ct and looking at the function v(ξ1 , ξ2 ) =
2 ξ1 −ξ2
u ξ1 +ξ
2
, 2c , we see that if u satisfies (1) then v satisfies

∂ξ1 ∂ξ2 v = 0.

The “general” solution of this equation is

v = f (ξ1 ) + g(ξ2 )

with f, g arbitrary functions. In other words, the “general” solution of (1) is

u(x, t) = f (x + ct) + g(x − ct) (2)

The two functions w1 = f (x + ct) and w2 = g(x − ct) solve

∂t w1 − c∂x w1 = 0

and
∂t w2 + c∂x w2 = 0
describing propagation to the left (in positive time) and propagation to the
right of a wave of arbitrary constant shape. The solution to the initial value
problem 
u(x, 0) = u0 (x)
(3)
∂t u(x, 0) = u1 (x)

1
is obtained from the equations u(x, t) = f (x + ct) + g(x − ct) and ∂t u(x, t) =
cf 0 (x + ct) − cg 0 (x − ct) by setting t = 0. We obtain

u0 = f + g
1
u = f 0 − g0
c 1

and so, differentiating the first equation, solving and then integrating, we
obtain
f = 12 (u0 + 1c U1 ) + constant


g = 12 (u0 − 1c U1 ) + constant
where U1 is some primitive of u1 . Returning to the formula for u we obtain
ˆ
1 1 x+ct
u(x, t) = (u0 (x + ct) + u0 (x − ct)) + u1 (ξ)dξ (4)
2 2c x−ct
If u0 , u1 have compact support K then at any T > 0 the solution u(x, t) will
have compact support KT formed with the union of the points on the rays
x ± ct = constant starting at t = 0 in K and arriving at T in KT . The signal
travels with finite speed c. A point lying at distance larger than cT from K
cannot be reached by the signal in less than T time.

2 The wave equation in Rn

The equation is
∂ 2u
− c2 ∆u = 0 (5)
∂t2
with c > 0. We prescribe

u(x, 0) = f (x)
∂u (6)
∂t
(x, 0)
= g(x)

We use the method of spherical means due to Poisson. For a function h =

h(x1 , . . . , xn ) we define Mh (x, r) by
ˆ
1
Mh (x, r) := h(y)dS(y) (7)
ωn rn−1 |y−x|=r

Clearly ˆ
1
Mh (x, r) = h(x + rξ)dS(ξ)
ωn |ξ|=1

2
We can use this formula to extend the definition of Mh (x, r) for all r ∈ R.
Note that for fixed x this is an even function of r,

Mh (x, −r) = Mh (x, r)

and also
Mh (x, 0) = h(x)
if h is continuous. Let us compute ∂r Mh :

∂r Mh = ξ · ∇hdS = r−1 ξ · ∇ξ (h(x + rξ))dS

Sn−1 Sn−1

The vector ξ is the external normal to the unit ball. Thus, using the Gauss
formula
´ ´
∂r Mh (x, r) = ω1n r |ξ|<1 ∆ξ h(x + rξ)dξ = ωrn |ξ|<1 ∆x h(x + rξ)dx
´ 1−n ´
= ωrn ∆x |ξ|<1 h(x + rξ)dξ = ∆x rωn |x−y|<1 h(y)dy
1−n ´ r ´ ´r
= ∆x rωn 0 dρ |y−x|=ρ h(y)dS(y) = ∆x r1−n 0 ρn−1 Mh (x, ρ)dρ
 
´r
= r1−n 0 ρn−1 ∆x Mh (x, ρ)dρ

Multiplying by rn−1 and differentiating we obtain

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

which is the Darboux equation:

 2 
∂ n−1 ∂
+ Mh (x, r) = ∆x Mh (x, r) (8)
∂r2 r ∂r

Note that the left hand side is the radial part of the Laplacian. Moreover,
since ˆ
r
∂r Mh = ∆x h(x + rξ)dξ
ωn |ξ|<1
we have that
∂r Mh (x, 0) = 0 (9)
If u(x, t) solves (5) then

1 ∂2
(∆x u)(x + rξ, t) = u(x + rξ, t)
c2 ∂t2

3
and so
1 ∂2
(∆x Mu ) (x, t, r) = 2 2 Mu (x, t, r)
c ∂t
On the other hand, from the Darboux equation
 2 
∂ n−1 ∂
(∆x Mu ) (x, t, r) = + Mu
∂r2 r ∂r

Thus, the spherical means of a solution of the wave equation solve the spher-
ically symmetric wave equation
 2  2 
∂ 2 ∂ n−1 ∂
−c + Mu = 0 (10)
∂t2 ∂r2 r ∂r
b(r)
We will continue the calculation now for n = 3 where the ansatz a(r) = r
transforms the Laplacian
 2 
∂ n−1 ∂ 1
2
+ a = ∂r2 b
∂r r ∂r r
1−n
n−1
In general, the trick a = r 2 b gets rid of the singular term r
∂r b but
(n−1)(n−3)
introduces the singular term r2
b. Thus, writing

1
Mu (x, t, r) = Nu (x, t, r)
r
we obtain that the function Nr (x, t, u) satisfies the one dimensional wave
equation  2 2

∂ 2 ∂
−c Nu (x, t, r) = 0 (11)
∂t2 ∂r2
Now
Nu (x, 0, r) = rMu (x, 0, r) = rMf (x, r) = Nf (x, r)
and
∂t Nu (x, 0, r) = r∂t Mu (x, 0, r) = rMg (x, r) = Ng (x, r)
It follows that
ˆ r+ct
1 1
Nu (x, t, r) = (Nf (x, r + ct) + Nf (x, r − ct)) + Ng (x, ξ)dξ
2 2c r−ct

4
Dividing by r we obtain

Mu (x, t, r) =
1 1
´ r+ct
2r
((r + ct)Mf (x, r + ct) + (r − ct)Mf (x, r − ct)) + 2cr r−ct
ξMg (x, ξ)dξ

Now Mf and Mg are even, and therefore it follows that

Mu (x, t, r) =
(r+ct)Mf (x,r+ct)−(ct−r)Mf (x,ct−r) 1
´ r+ct (12)
2r
+ 2cr ct−r
ξMg (x, ξ)dξ

Recalling that Mu (x, t, 0) = u(x, t), we let r → 0

Mu (x, t, 0) = d
dr
+ ct)Mf (x, r + ct))| r=0 + (r+ct)Mcg (x,r+ct) | r=0
((r
= ∂t (tMf (x, ct)) 
+ tMg (x, ct)
´ ´ 
= 4πc1 2 t |y−x|=ct g(y)dS(y) + ∂t 4πc1 2 t |y−x|=ct f (y)dS(y)

Carrying out the differentiation we have

1
´ 
1
´ 
u(x, t) = g(y)dS(y) + ∂ t 4πc2 t |y−x|=ct f (y)dS(y)
1
´ 4πc2 t
|y−x|=ct
1
´
= 4πc2 t |y−x|=ct g(y)dS(y) + 4πc2 t2 |y−x|=ct (f (y) + (y − x) · ∇f (y)) dS(y)
(13)
The formula (13) gives the unique solution of the problem (5), (6). The do-
main of inflence of a point is the set in space-time where a disturbance which
was initially situated at the point can occur. We see that if the disturbance
was initially (t = 0) located at x ∈ R3 then it will propagate on a spherical
front S(x, ct) = {y ∈ R3 | |x − y| = ct}. The union of these spheres is the
future-oriented light cone

Γ+
x = {(y, t)| |x − y| = ct, t > 0}

The domain of influence is the light cone. This is the (strong) Huyghens
principle. If you sit at a fixed location in R3 , and wait, the evolution is an
inflating sphere of radius ct, and after a while the signal passes. This is why
we can communicate (sound, light) in 3d.
Note that there is a loss of regularity because the solution sees derivatives
of f . In order to make this clear, let us note that if the initial data are
spherically symmetric, then the solution is spherically symmetric as well (the
equation commutes with rotations). Note that if h is spherically symmetric

5
then h(y) = Mh (0, |y|). Denoting |y| = ρ, looking at (12) and reading at
x = 0, r = ρ we obtain
ˆ ρ+ct
(ρ + ct)f (ρ + ct) + (ρ − ct)f (ρ − ct) 1
u(ρ, t) = + ξg(ξ)dξ
2ρ 2cρ ρ−ct

If
pg = 0 and f is continuous but is not differentiable, like for instance f (x) =
1 − |x|2 for |x| ≤ 1, f (x) = 0 for |x| > 1, then the solution u(ρ, t) becomes
infinite at ρ = 0 at time t = c−1 . We have a linear focusing effect in L∞ . In
L2 we have conservation. Let
ˆ
E(t) = (|∇x u|2 + (∂t u)2 )dx
R3

Then
d
E(t) = 0
dt
We took c = 1. The energy balance is true in any dimension. Indeed,
ˆ
d
E(t) = u2udx
2dt R3

where
2u = (∂t2 − ∆)u
is the d’Alembertian.

3 Hadamard’s method of descent

This method finds solutions of a PDE by considering them as particular
solutions of equations in more variables. We seek solutions of

∂t2 u − c2 (∂x21 + ∂x22 )u = 0 (14)

as solutions of the 3D wave equation in x3 = 0. We assume thus that

f (x1 , x2 , x3 ) = f (x1 , x2 , 0) and g(x1 , gx2 , x3 ) = g(x1 , x2 , 0) with a slight abuse
of notation. We use (13) at x3 = 0 noting that

{y| |y − x| = ct} = {y| (y1 − x1 )2 + (y2 − x2 )2 + y32 = c2 t2 }

and that
dS(y) = (cos α)−1 dy1 dy2

6
where
|y3 |
cos α = .
ct
´ ´
The integral |y−x|=ct g(y1 , y2 , y3 )dS(y) is clearly equal to 2 |y−x|=ct, y3 >0 gdS
because g does not depend on y3 . Introducing
p
ρ = (y1 − x1 )2 + (y2 − x2 )2

the integral becomes

ˆ ˆ
1
g(y)dS = 2ct g(y1 , y2 ) p dy1 dy2
|y−x|=ct Dx (ct) c2 t2 − ρ2
where
Dx (ct) = {(y1 , y2 )| ρ < ct}
The formula (13) becomes
˜
1
u(x1 , x2 , t) = 2πc ρ<ct
√g(y1 ,y2 ) dy1 dy2
c2 t2 −ρ2 
˜ (15)

+∂t 2πc 1
ρ<ct
√f (y1 ,y2 ) dy1 dy2
2 2
c t −ρ 2

We see that in n = 2 the strong Huyghens principle does not hold. The
domain of influence of a point is the solid disk Dx (ct). Once received, the
signal never dies: communication is impossible in 2d.

4 The inhomogeneous Cauchy problem

We wish to solve
2u = F (16)
with initial conditions given by (6). Because of linearity of the equation, we
can solve separately the initial value and the forcing:

u(x, t) = u1 (x, t) + u2 (x, t)

with
2u1 = 0, u(x, 0) = f (x), ∂t u(x, 0) = g(x)
and
2u2 = F, u(x, 0) = 0, ∂t u(x, 0) = 0.

7
Thus, WLOG f = g = 0. Let us denote by WF (x, t, s) the solution of the
homogeneous problem,

2w = 0

(17)
w(x, s) = 0, ∂t w(x, s) = F (x, s)

In n = 3 we have
ˆ
1
WF (x, t, s) = 2
F (y, s)dS(y) (18)
4πc (t − s) |y−x|=c(t−s)

for t > s. Now we claim that the solution of (16) is given by

ˆ t
u(x, t) = WF (x, t, s)ds (19)
0

This is Duhamel’s priciple. It is the analogue of Laplace’s “variation of

constants” formula from ODEs. Let us check that it works.
ˆ t ˆ t
∂t u(x, t) = WF (x, t, t − 0) + ∂t WF (x, t, s)ds = ∂t WF (x, t, s)ds
0 0

because WF (x, t, t − 0) = 0. Then

ˆ t
∂ 2 WF
∂t2 u = ∂t WF (x, t, t − 0) + (x, t, s)ds
0 ∂t2
Now ˆ t
∆x u = (∆x WF )(x, t, s)ds
0
and because
∂t WF (x, t, t − 0) = F (x, t)
we verified that u solves the inhomogeneous equation.

5 Decay of solutions
We discuss solutions of the homogeneous wave equation

2u = 0 (20)

8
in Rn with initial data (6). From the representation (13) it is clear that, when
n = 3 and f, g have compact support, say f ∈ C01 (B(0, R)), g ∈ C0 (B(0, R))
for some R > 0, then it follows that

|u(x, t)| ≤ Ct−1

for large t, uniformly for all (x, t) ∈ R4 . Indeed, the term involving g is
bounded by
ˆ 
t−1  g(y)dS  ≤ Ct−1 H 2 ({y| |x − y| = t, |y| < R})
 
|x−y|=t

where H 2 (S) is the two dimensional Hausdorff measure (area in our case).
Because the area of any part of a sphere inside B(0, R) is bounded by a
constant depending only on R, this term is done. Similarly, the second term
is bounded
´ 
t−2  |x−y|=t (f (y) + (y − x) · ∇f (y))dS(y)
 
 
≤ Ct−1 |x| t
+ t−1 H 2 ({y| |x − y| = t, |y| < R})

If |x|
t
≤ 2 we are done, and if |x|
t
≥ 2 then |y| ≥ |x| − t ≥ t, so H 2 ({y| |x − y| =
t, |y| < R}) = 0 for t > R. We see also that the decay is sharp: we can
choose g ≥ 0, g > 1 on some fixed region and f = 0 and then there are
nontrivial contributions of the order Ct−1 . (Of course, we cannot fix x,
that would contradict the strong Huyghens principle; we are talking about
uniform bounds, i.e. we are allowed to move x around). We present now two
approaches to prove the same result for general dimensions, namely
n−1
|u(x, t)| ≤ Ct− 2

for t large, uniformly in Rn+1 . The first appoach is based on ocillatory

integrals; the second, due to Klainerman, uses the symmetries of the wave
equation.

5.1 Oscillatory integrals

For the first approach, we write
ˆ h i
−n
u(x, t) = (2π) eix·ξ sin(t|ξ|)|ξ|−1 gb(ξ) + cos(t|ξ|)fb(ξ) dξ
Rn

9
which we rewrite as

u(x, t) = u+ (x, t) + u− (x, t)

with ˆ
−n
u± (x, t) = (2π) ei(x·ξ±t|ξ|) A± (ξ)dξ
Rn
and with
1 b 
A± (ξ) = f (ξ) ± i−1 |ξ|−1 gb(ξ)
2
Both terms u± decay. Let us look at u+ , the other term is treated the same
way. Without loss of generality we may take coordinates aligned with the
direction of x, so, without loss of generality we may assume

x = Ren

where en = (0, 0, . . . , 1), R = |x|. Thus

ˆ
−n
u+ (x, t) = (2π) ei(Rξn +t|ξ|) A+ (ξ)dξ
Rn

Now
 we take a small number  and take a smooth cutoff on the unit sphere
ξ
χ |ξ| supported in |ξn |2 ≥ (1 − 2 )|ξ|2 . We write

u+ (x, t) = a(x, t) + b(x, t)

where ˆ
−n
a(x, t) = (2π) ei(Rξn +t|ξ|) α(ξ)dξ
Rn
with  
ξ
α(ξ) = A+ (ξ)χ
|ξ|
and ˆ
−n
b(x, t) = (2π) ei(Rξn +t|ξ|) β(ξ)dξ
Rn
with   
ξ
β(ξ) = A+ (ξ) 1 − χ
|ξ|

10
 p
Now on the support of β we have that ρ = ξ12 + . . . ξn−1 2
satisfies ρ ≥ |ξ|.
In the integral representing b we write
 N
it|ξ| |ξ| d it|ξ| |ξ| d
e = e = eit|ξ| = LN (eit|ξ| ),
itρ dρ itρ dρ
|ξ| d
where L = itρ dρ
and N is arbitrary. We integrate by parts, taking advantage
of the fact that eiRξn does not depend on ρ:
ˆ
−n
b(x, t) = (2π) (−1) N
ei(Rξn +t|ξ|) (L∗ )N (β(ξ))dξ
Rn

We assume that f, g are smooth and decay fast enough. Then A± (ξ) are
smooth and decay fast enough. In order to have this behavior for small ξ as
well, we need to assume that gb(ξ) = O(|ξ|), for instance that g = Dg1 with
D some directional derivative and g1 smooth and decaying at infinity. Then
we can assure that on the support of β we have
X j
|(L∗ )N (β)(ξ)| ≤ (t|ξ|)−N CN, |∂ξ A+ (ξ)|
|j|≤N

with CN, depending on , N and χ. We have thus

ˆ X j
−N
|b(x, t)| ≤ CN, t |ξ|−N |∂ξ A+ (ξ)|dξ
Rn |j|≤N

This integral converges if N ≤ n − 1 (because of the behavior near zero; if

we excise a region near zero, then we have arbitrary decay), so

|b(x, t)| ≤ Ct−n+1

n−1
a better decay than t− 2 . Let us consider now a(x, t). We denote
R
λ= , k = |ξ|.
t
and write the phase as

Rξn + t|ξ| = tk(λ cos θ + 1)

where
ξn
cos θ =
|ξ|

11
We introduce polar coordinates and have

´∞´ ´π a(x, t) =
−n ikt(λ cos θ+1)
(2π) 0 Sn−2 0
e α(k, θ, ω)(sin θ)n−2 k n−1 dθdkdSn−2 (ω)
√
On the support of α we have | cos θ| ≥ 1 − 2 , so it is close to 1. We
disitinguish two situations: if the phase is bounded away from zero, or if
the phase could vanish. The first situation occurs if cos θ > 0 (and hence it
is about 1) or if cos θ < 0 (and hence it is about −1) and λ < 1 −  or if
cos θ < 0, and λ > (1 − 22 )−1 . In these situations we write
d ikt(λ cos θ+1)
eikt(λ cos θ+1) = it(λ cos1 θ+1) dk e
h iN
= it(λ cos1 θ+1) dk
d
eikt(λ cos θ+1) = LN (eikt(λ cos θ+1) )

After cutting off and integrating by parts, the resulting expression decays
faster than any power of t. We are left with the behavior of

´∞ ´ a1 (x,
´ πt) =ikt(λ cos θ+1)
(2π)−n 0
k n−1 dk Sn−2
dSn−2 (ω) π−δ
e α(k, θ, ω)(sin θ)n−2 dθ

for δ = δ() small and with 1 −  ≤ λ ≤ (1 − 22 )−1 . Because λ is close to 1,

we write

´∞ ´ a1 (x, ´t) =
π
(2π)−n 0
eikt(1−λ) k n−1 dk Sn−2
dS(ω) π−δ
eikλt(cos θ+1) α(k, θ, ω)(sin θ)n−2 dθ

z2
Changing variables to 2
= 1 + cos θ, the inner integral is of the form
ˆ γ
z2
eikλt 2 α1 (k, z, ω)z n−2 dz
−γ
√ √
with γ = 2(1 + 1 + 2 )−1  and for an appropriate α1 . Thus
ˆ γ 
 z2  1−n
ikλt n−2

 e 2 α (k, z, ω)z
1 dz  ≤ C(kλt) 2
−γ

holds with C depending only on n, , χ a and angular derivatives of A+ (ξ).

12
5.2 Method of commuting vector fields of Klainerman
We start by the observation that if f is a radial function in Rn , i.e. f (x) =
F (|x|), then
ˆ ∞
2
f (x) = −2 F (r)F 0 (r)dr ≤ C|x|−(n−1) kf kL2 (Rn ) k∇f kL2 (Rn ) .
|x|

n−1
Radial functions in H 1 (Rn ) decay like |x|− 2 . Of course, we can’t expect
this to be true for all functions in H 1 (Rn ). Radial functions belong to the
kernel of the angular momentum operators
Ωij = xi ∂j − xj ∂i .
It turns out that a general “decay” estimate, of the kind obeyed by the radial
functions exists, provided an account is made of the size of the momentum
operators. Let A be a Lie algebra of first order operators with smooth co-
efficients (vector fields), finitely generated and let X1 , . . . XN be generators.
We use the notation
k N
! 21
X X
|u(x)|A,k = |Xii Xi2 . . . Xil u(x)|2 (21)
l=0 i1 ,...il =1

and denote ˆ  12
kukA,k = |u(x)|2A,k dx (22)
Rn
Let O0 be the Lie algebra generated by the angular momentum operators
Ωij , 1 ≤ i, j ≤ n.
Lemma 1. Let m ≥ [ n−1 2
] + 1. There exists a constant C = C(m, n) such
that for every smooth function f on Sn−1 and all η ∈ Sn−1 we have
ˆ  21
2
|f (η)| ≤ C |f (ξ)|O0 ,m dS(ξ) . (23)
Sn−1

This follows from the Sobolev embedding.

Lemma 2. There exists a constant C = C(n) such that, for every smooth
function f , and all x 6= 0 ∈ Rn
n−1 1 1
|f (x)| ≤ C|x|− 2 kf kO2 n−1 k∇f kO2 n−1 (24)
0 ,1+[ 2 ] 0 ,1+[ 2
]

13
 Let x ∈ Rn be fixed. We write x = rω with ω ∈ Sn−1 . We fix ω and write
ˆ ∞  12 ˆ ∞  21
2 1−n 2 n−1 2 n−1
|f (rω)| ≤ 2r |f (ρω)| ρ dρ |∂ρ f (ρω)| ρ dρ
0 0

From Lemma 1 it follows that

ˆ
2
|f (ρω)| ≤ C |f (ρξ)|2O0 ,m dS(ξ)
Sn−1

with m = 1 + [ n−1
2
]. Then note also that
ˆ ∞ ˆ ˆ
n−1 2
ρ dρ |f (ρξ)|O0 ,m dS(ξ) = |f (x)|2O0 ,m dx
0 Sn−1 Rn

We note the similar inequality for ∂ρ f (ρω), and we are done.

We want to use now additional vector fields that commute with the wave
operator. For simplicity of notation we will use x0 = t = −x0 , xj = xj for
coordinates. Indices running from 0 to 3 will be denoted by greek letters,
indices running from 1 to 3 by latin ones. We denote by ηαβ the Lorentz
metric, η αβ its inverse (it is the same matrix) recall that η00 = −1, ηij = δij ,
all the rest of the entries are zero. The translation operators are
∂
Tµ = .
∂xµ
The angular momentum operators are
∂ ∂
Ωµν = xµ ν
− xν µ .
∂x ∂x
We denote by S the dilation
∂
S = xµ .
∂xµ
We use summation convention. The commutation relations are

[Tµ , Tν ] = 0, [Tµ , Ωαβ ] = ηµα Tβ − ηµβ Tα ,

[Tµ , S] = Tµ ,
[Ωµν , Ωαβ ] = ηµα Ωβν − ηµβ Ωαν + ηνα Ωµβ − ηνβ Ωµα ,

14
 [Ωµν , S] = 0.
Thus, (Tµ ) generate a Lie algebra denoted T , (Ωµν ) generate a Lie algebra
denoted Ω, (Ωµν , S) generate a Lie algebra denoted L (for Lorentz) and (Tµ ,
Ωµν , S) generate a Lie algebra denoted Π (for Poincaré). Note that

[2, Tµ ] = 0, [2, Ωµν ] = 0

and
[2, S] = 22.
Thus the dilation of a solution of 2 = 0 is again a solution, and obviously
the same is true for all elements in Π. We will use conservation of energy of
solutions and their Π derivatives to obtain decay. We begin by recalling the
classical local Sobolev lemma in Rn :

Lemma 3. There exists a constant C = C(m, n) so that, for m ≥ 1 + [ n2 ],

every R > 0, every smooth u and all x ∈ B(0, R) we have
 21
ˆ

m
n
X X
|u(x)| ≤ C Rj− 2  |∂ α u(x)|2 dx (25)
j=0 B(0,R) |α|=j

This is easily seen done by reducing the problem to R = 1 using a dila-

tion, and then using an extension theorem in H m (B(0, 1)), extending u to a
function in v ∈ H m (Rn ) of comparable H m norm, and supported in B(0, 2).
We take a smooth compactly supported cutoff function χ equal identically
to 1 in a neighborhood of B(0, 1) and write the Fourier inversion formula for
m
χv. Using the weights (1 + |ξ|2 ) 2 we see that the L1 (Rn ) norm of the Fourier
transform of χv is bounded by a constant multiple of the H m (B(0, 1)) norm
of u.
Now we note that for (t, x) = (x0 , x) not on the wave cone |x|2 = t2 we
can express regular ∂i derivatives in terms of the vector fields generating L:
1
Tν = (xµ Ωµν + xν S) (26)
r2 − t2
with obvious notation r2 = |(x1 , x2 , . . . , xn )|2 , r2 − t2 = ηαβ xα xβ . We denote
1
σ− = (1 + |r − t|2 ) 2 .

15
Lemma 4. For every k ≥ 0 there exists C = C(n, k) so that for every smooth
function and every point x = (x0 , x1 , . . . , xn ) we have
k
 
 ∂  ≤ C(σ− )−k |u(x)|Π,k

 α
 ∂x 0 . . . ∂xαn u(x) (27)
0 n

holds for all α0 + · · · + αn = k.

The proof is done by induction. For k = 1 we saw that
∂u 1 1
ν
= (xµ Ωµν + xν S)
∂x r−tr+t
so  
 ∂u  C
 ∂xν  ≤ |r − t| |u(x)|L,1 .
 

In order to take care of the region |r − t| < 1 we augment to Π. We obtain,

for any function u,  
 ∂u  C
 ∂xν  ≤ σ− |u(x)|Π,1 . (28)
 

For higher derivatives we use the commutation relations, which look symbol-
ically like [T , Π] = T . For instance, by (28), we have

∂ 2u C
| µ ν
|≤ |Tµ u(x)|Π,1
∂x ∂x σ−
and a term of the form XTµ u with X ∈ Π ∪ {I} is of the form

XTµ u = Tµ X(u) + T

with T ∈ T , so we apply (28) to deduce

C
|XTµ u| ≤ |u(x)|Π,2
σ−
Now we consider a point (t, x) in the interior of the wave cone
t
|x| ≤ .
2
Then, clearly,
t
σ− (t, x) ≥ . (29)
2
16
 t
Let R = 2
and apply Lemma 3:
 21
ˆ

m
n
X X
|u(t, x)| ≤ Rj− 2  |∂ α u(y)|2 dy 
j=0 |y|≤R |α|≤j

Using, for each y the inequality (27) of Lemma 4 and recalling (29) which is
valid for (t, y), we obtain
´  21
j− n
Pm −j 2
|u(t, x)| ≤ C j=0 R 2R
|y|≤R
|u(t, y)| Π,j dy
n P
´  1
2
= Ct− 2 m j=0 |y|≤ t
|u(t, y)|2Π,j dy
2
n
≤ Ct− 2 ku(t, ·)kΠ,m

We proved:

Lemma 5. Let m ≥ 1 + n2 . There exists a constant C = C(m, n) so that,

 

for all smooth functions u and all points (t, x) with t > 0, t ≥ 2|x|, we have
n
|u(t, x)| ≤ Ct− 2 ku(t, ·)kΠ,m . (30)
t
On the other hand, for |x| ≥ 2
we can use (24):
n−1
|u(t, x)| ≤ Ct− 2 ku(t, ·)kΠ,2+[ n−1 ] (31)
2

Combining (30) and (31) we have

Proposition 1. Let m ≥ [ n2 ] + 2. There exists a constant C = C(m, n) so

that for all t > 0, x ∈ Rn and smooth functions u we have
n−1
|u(t, x)| ≤ C(1 + t)− 2 ku(t, ·)kΠ,m (32)

We control the region t < 1, |x| ≤ 12 using usual Sobolev inequalities on

each time slice. If t < 1 but |x| ≥ 12 or if t ≥ 1 and |x| ≥ 2t we use (31) and
if t ≥ 1, |x| ≤ 2t we use (30).
Just as Lemma 2, Proposition 1 is not a true decay result, it is just a
convenient tautology. However, for solutions of the wave equation, we can
easily control the right hand side.

17
Theorem 1. Let m ≥ [ n2 ] + 2. Let u(t, x) solve

2u = 0

and assume that the functions

u0 (x) = u(0, x), u1 (x) = ∂t u(0, x)

are such that

k∇x u(0, ·)kΠ,m < ∞, k∂t u(0, ·)kΠ,m < ∞

Then
n−1
|∇x u(t, x)| ≤ C(1 + t)− 2 [k∇x u(0, ·)kΠ,m + k∂t u(0, ·)kΠ,m ]

Note that we need to use the equation in order to express the high Π, m
norms at time t = 0 in terms of spatial derivatives of the initial data.
The proof follows by considering v = ∂j Γu where Γ ∈ Π. Note that
1
kv(t, ·)kL2 (Rn ) ≤ C[kv(0, ·)kL2 (Rn ) + k(−∆)− 2 ∂t v(0, ·)kL2 (Rn ) ]

holds for any solution of the wave equation. So we have

k∇x Γu(t, ·)kL2 (Rn ) ≤ C[k(∇x Γu)(0, ·)kL2 (Rn ) + k(∂t Γu)(0, ·)kL2 (Rn )

18