ADVANCED PARTIAL DIFFERENTIAL EQUATIONS:

HOMEWORK 3

KELLER VANDEBOGERT

1. Chapter 2, Problem 7

We have by the triangle inequality |y| − |x| 6 |y − x| 6 |y| + |x|. For

y ∈ ∂B(0, r), |y| = r, so that
1 1 1
n
6 n
6
(r + |x|) |y − x| (r − |x|)n
On ∂B(0, r). Suppose now that u is harmonic, and in particular sat-
isfies the mean value property. Using the above and Poisson’s formula
for the ball:

ˆ ˆ
r2 − |x|2 u(y) r2 − |x|2 u(y)
n
dS(y) 6 u(x) 6 dS(y)
nα(n)r ∂B(0,r) (r + |x|) nα(n)r ∂B(0,r) (r − |x|)n
2 −|x|2 ´
Where we’ve used that u(x) = rnα(n)r u(y)
∂B(0,r) |y−x|n
by Poisson’s for-
mula. Multiplying the left most and right most terms in the above
inequality by rn−2 , we find:

ˆ ˆ
(r − |x|)(r + |x|)rn−2 (r − |x|)(r + |x|)rn−2
u(y)dS(y) 6 u(x) 6 u(y)dS(y)
nα(n)rn−1 (r + |x|)n ∂B(0,r) nα(n)rn−1 (r − |x|)n ∂B(0,r)

Which, by definition, becomes:

(r − |x|)rn−2 (r + |x|)rn−2
u(y)dS(y) 6 u(x) 6 u(y)dS(y)
(r + |x|)n−1 ∂B(0,r) (r − |x|)n−1 ∂B(0,r)

Date: September 3, 2017.

1
2 KELLER VANDEBOGERT
ffl
But u is harmonic, so by the mean value property, ∂B(0,r)
u(y) =
u(0). Using this, the above implies:

(r − |x|)rn−2 (r + |x|)rn−2
u(0) 6 u(x) 6 u(0)
(r + |x|)n−1 (r − |x|)n−1
As desired.

2. Chapter 2, Problem 8

Following this hint, we can see that since u ≡ 1 solves ∆u = 0 in

B(0, r) and u = 1 on ∂B(0, r), Poisson’s formula for the ball implies:

ˆ
1= K(x, y)dy
∂B(0,r)
Where K(x, y) denotes Poisson’s kernel. By definition, G(x, y) for
the ball is harmonic and smooth for x 6= y and hence given  > 0, we can
find δ such that whenever |x − x0 | < δ, |Dα K(x, y) − Dα K(x0 , y)| < 
for any order derivative. In particular, since ∂B(0, r) is compact we
´
see that Dα u(x) = ∂B(0,r) Dxα K(x, y)g(y)dy. Using this, let  > 0, and
note that since g is continuous on a compact set it is bounded:

(2.1) ˆ
α α
|D u(x) − D u(x0 )| 6 |Dα K(x, y) − Dα K(x0 , y)||g(y)|dy
∂B(0,r)
ˆ
< |g(y)|dy
∂B(0,r)
6 ||g||L∞ Sn → 0
Where Sn denotes the surface area of the n-sphere. Hence u is
smooth. Also, it is clear by the above that

ˆ
∆u(x) = ∆x K(x, y)g(y)dy = 0
∂B(0,r)
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 3

So that u is indeed harmonic. Finally, we must show that u(x) =

g(x) on ∂B(0, r). Equivalently, this implies that lim u(x) = g(x0 )
x→x0
when x0 ∈ ∂B(0, r). By continuity of g, we can find δ for any  > 0
such that |g(x) − g(x0 )| <  whenever |x − x0 | < δ. First, note that
´
∂B(0,r)
g(x0 )K(x, y)dy = g(x0 ). Then we can split up our integral as
´ ´ ´
∂B(0,r)
= ∂B(0,r)∩B(x0 ,δ) + ∂B(0,r)\B(x0 ,δ) . We have:

ˆ
|u(x) − g(x0 )| 6 K(x, y)|g(y) − g(x0 )|dy
∂B(0,r)∩B(x0 ,δ)
ˆ
(2.2)
+ K(x, y)|g(y) − g(x0 )|dy
∂B(0,r)\B(x0 ,δ)
:= I + J
´ ´
Then, firstly, since 1 = ∂B(0,r) K(x, y)dy, in particular U K(x, y) <
1 for any U ⊂ B(0, r), so we immediately find that

ˆ
I< K(x, y)dy < 
∂B(0,r)∩B(x0 ,δ)

For J, we see that by the triangle inequality |y−x0 | 6 |y−x|+|x−x0 |

and |g(x) − g(x0 ) 6 2||g||L∞ . In particular, it is clear that for y ∈
/
B(x0 , δ) and |x − x0 | < δ/2, that |y − x0 | 6 |y − x| + 12 |y − x0 | =⇒
1
2
|y − x0 | 6 |y − x|. Employing this in J:

(2.3)
ˆ ˆ
2(r2 − |x|2 )||g||L∞
K(x, y)|g(y) − g(x0 )|dy 6 |y − x0 |−n dy
∂B(0,r)\B(x0 ,δ) nα(n)r ∂B(0,r)\B(x0 ,δ)
→0

Since |x| → r as x → x0 , as x0 ∈ ∂B(0, r). Then we are done, since

we see that |u(x) − g(x0 )| → 0 as x → x0 , so that this is indeed a
solution.
4 KELLER VANDEBOGERT

3. Chapter 2, Problem 10

(a). Denoting by v for the odd reflection of u(x1 , . . . , xn ), when xn > 0,

u ≡ v so v is obviously C 2 on the upper half. For xn < 0:

∂v ∂u
=− i 6= n
∂xi ∂xi
∂v ∂u
=
∂xn ∂xn
Which are both continuous in the lower half sphere. Similarly,

∂ 2v ∂ 2u
= − i 6= n
∂x2i ∂x2i
∂ 2v ∂ 2u
=− 2
∂x2n ∂xn
Thus these derivatives are again continuous, since u ∈ C 2 (C 2 ). Using
this, in the lower half sphere, we see that ∆v = −∆u = 0, so v is
harmonic in both the upper and lower half ball, and hence everywhere.
Also, we see that on the intersection of their respective boundaries,
v = u(x1 , . . . , xn−1 , 0) = 0.

(b). We can employ Poisson’s formula for the ball to find that

ˆ
v(x) = K(x, y)v(y)dS(y)
∂B(0,1)

Where K(x, y) denotes Poisson’s kernel. We have shown that Pois-

son’s kernel is harmonic and that K(x, y) ∈ C 2 (B(0, 1)), and so we
find:

ˆ
∂v(x) ∂K(x, y)
= v(y)dS(y)
∂xi ∂B(0,1) ∂xi
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 5

And since we know that v(y) is continuous on ∂B(0, 1), we see that
∂K(x,y)
the product ∂xi
v(y) is continuous and hence so is the integral, so
that v(x) ∈ C 1 (U ). Similarly, the second derivative is easy to calculate:

ˆ
∂ 2 v(x) ∂ 2 K(x, y)
= v(y)dS(y)
∂x2i ∂B(0,1) ∂x2i
2
So that v ∈ C (U ). Finally, noting that K(x, y) is harmonic:

ˆ
∆v(x) = ∆x K(x, y)v(y)dS(y) = 0
∂B(0,1)
So that v is harmonic, and we are done.

4. Chapter 2, Problem 12

(a). Define uλ (x, t) := u(λx, λ2 t). Then, we see:

∂uλ ∂u
= λ2 (λx, λ2 t)
∂t ∂t

∂ 2 uλ 2
2∂ u
= λ (λx, λ2 t)
∂x2i ∂x2i
And hence summing over all i:

∂uλ
− ∆uλ = λ2 (ut − ∆u) = 0
∂t
So that uλ also satisfies the heat equation.

(b). Take the partial with respect to λ:

∂uλ ∂u
= x · Du(λx, λ2 x) + 2λt (λx, λ2 x)
∂λ ∂t
However, by the above it is clear that v(x, t) = ∂u λ
| , v as de-
∂t λ=0

fined in the book. Using commutativity of mixed partial derivatives

(guaranteed by the smoothness of u):
6 KELLER VANDEBOGERT

∂2 ∂
vt − ∆v = uλ − ∆ uλ
(4.1) ∂t∂λ ∂λ
∂ ∂uλ
 
= − ∆uλ = 0
∂λ ∂t
∂uλ
Since ∂t
− ∆uλ = 0 by part (a). Hence we see that after setting
λ = 1, v satisfies the heat equation as well.

5. Chapter 2, Problem 14

Define v := ect u. Then, by the statement of the problem, we see that

v satisfies

vt − ∆u = ect f

v=g
2
In Rn+1
+ and on ∂Rn+1
+ , respectively. Letting Φ(x, t) :=
1
(4πt)n/2
e−|x| /4t
denote the heat kernel, we can solve for v:

ˆ ˆ tˆ
v(x, t) = Φ(x − y, t)g(y)dy + Φ(x − y, t − s)f (y, s)dyds
Rn 0 Rn

And using the fact that v(x, t) := ect u(x, t), multiply the above by
e−ct to find:

ˆ ˆ tˆ
−ct
u(x, t) = Φ(x−y, t)e g(y)dy+ Φ(x−y, t−s)e−ct f (y, s)dyds
Rn 0 Rn

And the above solves ut − ∆u + cu = f in Rn+1

+ and u = g on ∂Rn+1
+ ,

so we are done.
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 7

6. Chapter 2, Problem 15

Define v(x, t) := u(x, t) − g(t). Then, it is clear that v satisfies

vt − vxx = −g 0 in R+ × (0, ∞), v(x, 0) = g(0) = 0, and v(0, t) = 0. We
now extend v by odd reflection, so that for x < 0, v(x, t) := −v(−x, t).
Then, this extends v to satisfy vt −vxx = −g 0 (t) when x > 0, vt −vxx − =
g 0 (t) for x < 0, and v(x, 0) = 0. Then, using Duhamel’s principle we
can immediately solve for v:

ˆ t ˆ ∞ ˆ 0
1 −(x−y)2 −(x−y)2
 
0
v(x, t) = − e 4(t−s) g (s)dy+ e 4(t−s) g 0 (s)dy ds
0 (4π(t − s))1/2 0 −∞
´∞ −(x−y)2
Now, let us consider the term 0
e 4(t−s) g 0 (s)dy. We have:

(6.1)
ˆ ∞ ˆ ∞ ˆ 0
−(x−y)2 −(x−y)2 −(x−y)2
0 0
e 4(t−s) g (s)dy = e 4(t−s) g (s)dy − e 4(t−s) g 0 (s)dy
0 −∞ −∞
ˆ ∞ ˆ 0
−x2 −(x−y)2
= e 4(t−s) g 0 (s)dy − e 4(t−s) g 0 (s)dy
−∞ −∞
ˆ 0

1/2 −(x−y)2
0
= g (s) 4(t − s) Γ(1/2) − e 4(t−s) g 0 (s)dy
−∞

Plugging this back into our expression for v (and noting that Γ(1/2) =
√
π):

ˆ t ˆ t ˆ 0
1 −(x−y)2
0
v(x, t) = − g (s)ds + 2 e 4(t−s) g 0 (s)dyds
0 0 (4π(t − s))1/2 −∞
´t
Now, of course 0
g 0 (s) = g(t), so using the fact that v(x, t) =
u(x, t) − g(t), we have an expression for u(x, t):

ˆ t ˆ 0
1 −(x−y)2
u(x, t) = 2 1/2
e 4(t−s) g 0 (s)dyds
0 (4π(t − s)) −∞
We must now integrate by parts:
8 KELLER VANDEBOGERT

(6.2)
ˆ t ˆ 0 ˆ 0
1 −(x−y)2 g(s) −(x−y)2 t
e 4(t−s) dydg(s) = e 4(t−s) dy

0 (4π(t − s))1/2 −∞ (4π(t − s))1/2 −∞ 0

ˆ t ˆ 0
g(s) −(x−y)2
− √ 3/2
e 4(t−s) dyds

0 4 π(t − s) −∞
ˆ t ˆ 0 −(x−y)2
g(s) (x − y)2 e 4(t−s)
+ √ dyds
0 4 π(t − s)
3/2
−∞ 2(t − s)
:= I − J + K

−(x−y)2
´0 (x−y)2 e 4(t−s) d
−(x−y)2
We can integrate the term −∞ 2(t−s)
dy by noticing that dy
e 4(t−s) =
−(x−y)2
(x−y)e 4(t−s)
2(t−s)
. Then:

ˆ 0 −(x−y)2 −(x−y)2
  h i0
(x − y)d e 4(t−s) = (x − y)e 4(t−s)

−∞ −∞
ˆ 0 −(x−y)2
(6.3) + e 4(t−s) dy
−∞
ˆ 0
−x2 −(x−y)2
= xe 4(t−s) + e 4(t−s) dy
−∞

Plugging the above into the last term, K, of (6.2):

(6.4)
ˆ t ˆ 0
−(x−y)2 ˆ t
−x2
g(s) (x − y)2 e 4(t−s) xe 4(t−s) g(s)
√ dyds = x √ ds
0 4 π(t − s)3/2 −∞ 2(t − s) 0 4 π(t − s)
3/2
ˆ t ˆ 0
g(s) −(x−y)2
+ √ 3/2
e 4(t−s) dyds

0 4 π(t − s) −∞

´t −x2
xe 4(t−s) g(s)
But this shows that K = x √
0 4 π(t−s)3/2
+ J, and hence, using this
in (6.2):
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 9

(6.5)
ˆ t ˆ 0 ˆ t
−x2
1 −(x−y)2 xe 4(t−s) g(s)
e 4(t−s) g 0 (s)dyds = I + x √ ds
0 (4π(t − s))1/2 −∞ 0 4 π(t − s)3/2
We proceed to show that I = 0. Since g(0) = 0, it is obvious that
the term inside I evaluated at 0 is 0, since all other terms remain finite.
To find the inside of I evaluated at t, suppose that t − s <  and then
−(x−y)2
+ e
let  → 0 . Then, it is clear that 4
1/2
→ 0 uniformly as  → 0+ ,
x 6= y, and hence evaluating at t also yields 0, since the only point for
which it does not tend to 0 is a singleton and hence integrates to 0.
Thus, I = 0 and combining all of the above with this, we can finally
conclude, using our expression for u(x, t):

ˆ t
−x2
x e 4(t−s) g(s)
u(x, t) = √ ds
4π 0 (t − s)3/2

7. Chapter 2, Problem 16

Defining u := u − t ( > 0), we see that ut − ∆u = − < 0. This

shows that u cannot attain a maximum on the interior of UT , since
if this were the case, we would see that ut > 0 (since locally, u must
be increasing) and ∆u 6 0 (locally convex) for some (x0 , t0 ) ∈ UT .
But this would then force ut − ∆u > 0. Hence, u cannot attain is
maximum on the interior.
Now, denote by M the maximum of u on the boundary ΓT , and
M := max u. By our definition, we certainly have that u 6 u. Suppose
UT
now for sake of contradiction that u attains its maximum on the interior
of UT . Then, M − t 6 M , since else u would attain its maximum on
the interior as well. Hence we have the following chain of inequalities:
10 KELLER VANDEBOGERT

M − t 6 M 6 M

Letting  → 0, we see that M0 = M . But M0 = max u, hence we

ΓT
conclude:

max u = max u
UT ΓT

Contradicting the fact that u attains its maximum on the interior.

Hence the assertion follows.

8. Chapter 2, Problem 17

(a). Let E(r) := E(0, 0; r), where u is a subsolution of the heat equa-
tion. Without loss of generality, we can assume (x, t) = (0, 0). Then,
2
define ψ(y, s) := − n2 log(−4πs) + |y|
4s
+ n log(r), where s 6 0, and note
that ψ restricted to ∂E(r)\(0, 0) vanishes. Define:

¨
1 |y|2
φ(r) := n u(y, s) dyds
r E(r) s2
(8.1) ¨
|y|2
= u(ry, r2 s) dyds
E(1) s2
The second equality comes from the natural change of variable y 7→
ry, s 7→ r2 s. Now we can differentiate φ with respect to r:

¨
0 |y|2 |y|2
φ (r) = Du · y + 2ru s 2 dyds
E(1) s2 s
¨
(8.2) 1 |y|2 |y|2
= Du · y + 2us dyds
rn+1 E(r) s2 s2
:= I + J
Now we want to consider the term J. Note that |y|2 /s2 = 2Dψ · y,
where ψ defined above. Using this,
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 11

¨
1
J= 4us Dψ · ydyds
rn+1 E(r)
Now integrate by parts with respect to y, and note that our boundary
term vanishes. Then, D(us y) = Dus · y + Dy · us . But Dy is merely a
vector of 1’s, and hence Dy · u = nus . Using this,

(8.3) ¨ ¨
1 1
4us Dψ · ydyds = − n+1 4ψDus · y + 4nus ψdyds
rn+1 E(r) r E(r)

Now integrate the first term by parts with respect to s. Then, ψs =

n |y|2
− 2s − 4s2
. Using this:

(8.4) ¨ ¨
1 1
− n+1 4ψDus · y + 4nus ψdyds = − n+1 −4ψs Du · y + 4nus ψdyds
r E(r) r E(r)
¨
1 2n |y|2
= − n+1 Du · y + 2 Du · y + 4nus ψdyds
r E(r) s s
¨
1 2n
= − n+1 Du · y + 4nus ψdyds − I
r E(r) s

Combining this with our expression for φ0 (r):

¨
0 12n
φ (r) = − Du · y + 4nus ψdyds
rn+1
E(r) s
But u is a subsolution. Hence, us 6 ∆u. We want to integrate this
by parts:

(8.5) ¨ ¨
1 2n 1 2n
− n+1 Du · y + 4nus ψdyds > − n+1 Du · y + 4n∆uψdyds
r E(r) s r E(r) s
¨
1 2n
= n+1 − Du · y + 4nDu · Dψdyds
r E(r) s
y
Noting the form of ψ, we see that Dψ = 2s
. Hence:
12 KELLER VANDEBOGERT

¨ ¨
1 2n 1 2n y

− Du·y+4nDu·Dψdyds = n+1 Du· − y+4n dyds = 0
rn+1 E(r) s r E(r) s 2s
Combining this with the above, we find that φ0 (r) > 0, so that φ
is an increasing function. This means that it attains its minimum as
r → 0.

¨
1 |y|2
lim φ(r) = lim n u(y, s) dyds
r→0 r→0 r E(r) s2
¨
|y|2
(8.6) = lim u(ry, r2 s) dyds
r→0 E(1) s2
¨
|y|2
= u(0, 0) dyds
E(1) s2
˜ |y|2
Our task is to compute E(1) s2
dyds. Recall that

|y|2
e− 4s
E(1) = {(y, s) : s 6 0, > 1}
(−4πs)n/2
Then, by basic manipulations we see that |y|2 6 2ns log(−4πs) in
1
E(1). Also, since |y|2 > 0, we see that s > − 4π . Hence,

¨ ˆ 0 ˆ
|y|2 1
dyds = |y|2 dyds
E(1) s2 −1/4π s2 |y|2 62ns log(−4πs)

Converting to spherical coordinates, |y| 7→ r:

(8.7)
ˆ 0 ˆ ˆ 0 ˆ (2ns log(−4πs))1/2
1 2 2π n/2 1
|y| dyds = rn+1 drds
−1/4π s2 |y|2 62ns log(−4πs) Γ(n/2) −1/4π s2 0
n/2 ˆ 0
2π
= (2ns log(−4πs))n/2+1 s−2 ds
(n + 2)Γ(n/2) −1/4π
n/2+1 n/2 ˆ 1/4π
2 · (2n) π  1
n/2+1
= sn/2−1 log ds
(n + 2)Γ(n/2) 0 4πs
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 13

1


Now make the change of variable u = log 4πs
. Our bounds will
´ 1/4π ´0
change as 0 → ∞:

ˆ
2 · (2n)n/2+1 π n/2 1/4π n/2−1  1
n/2+1
s log ds
(n + 2)Γ(n/2) 0 4πs
ˆ ∞
2 · (2n)n/2+1 π n/2
= (e−nu/2 )un/2+1 du
(n + 2)Γ(n/2)(4π)n/2 0
ˆ
2 · (2n)n/2+1 π n/2  2 n/2+2 ∞ −t n/2+1
= e t dt
(n + 2)Γ(n/2)(4π)n/2 n 0
2 · (2n)n/2+1 π n/2  2 n/2+2
= Γ(n/2 + 2)
(8.8) (n + 2)Γ(n/2)(4π)n/2 n
(2n)n/2+1  2 n/2+2
= (n/2 + 1)Γ(n/2 + 1)
(n/2 + 1)Γ(n/2)4n/2 n
(2n)n/2+1  2 n/2+2
= (n/2)Γ(n/2)
Γ(n/2)2n n
2n/2+1 nn/2+1 2n/2+2
= (n/2)
2n nn/2+2
2n+2
= n = 22 = 4
2
And hence we see limr→0 φ(r) = 4u(0, 0). Since (0, 0) can be trans-
lated to any general coordinate (x, t), we conclude:

¨
1 |x − y|2
u(x, t) 6 n u(y, s) dyds
4r E(x,t;r) (t − s)2
As asserted, and we are done.

(b). Suppose u attains its maximum at the point (x0 , t0 ). Then, by

part a, whenever 0 < r < dist((x, t), ∂UT ):

¨ ¨
1 |x − y|2 1 |x − y|2
M = u(x0 , t0 ) 6 n u(y, s) dyds 6 M dyds
4r E(x0 ,t0 ;r) (t − s)2 4rn E(r) (t − s)2
˜ 2
By the work of part (a), it is seen that 4r1n E(x0 ,t0 ;r) |x−y|
(t−s)2 dyds = 1
by a simple change of variable. We then conclude that u(x, t) = M
14 KELLER VANDEBOGERT

for all (x, t) ∈ E(x0 , t0 ; r). However, the set {(x, t) : u(x, t) = M } is a
closed set and the above shows that it is also an open set. Since UT is
connected, the only clopen subset of UT is the entire set itself. Then,
u(x, t) = M for all (x, t), implying u is constant.
Then, if we assumed that u were not constant and attained its max-
imum on the interior, we can apply the above argument to see that u
attains its maximum in some heat ball contained in UT . But this forces
u to be constant, a contradiction. Hence, u attains its maximum on
the boundary, and:

max u = max u
UT ΓT

As asserted.

(c). Define v := φ(u), where φ is smooth and convex, implying φ00 > 0.
Then, we compute:

∂ ∂φ ∂u
φ(u) =
∂xi ∂u ∂xi

∂ ∂φ ∂u ∂ 2 φ  ∂u 2 ∂φ ∂ 2 u
= +
∂xi ∂u ∂xi ∂u2 ∂xi ∂u ∂x2i
P ∂ 2 φ ∂u 2
 
Summing over i, ∆v = i ∂u2 ∂xi + ∂φ
∂u
∆u. Also, vt = ∂φ ∂u
∂u ∂t
.
Putting this all together:

∂φ ∂u X ∂ 2 φ  ∂u 2 ∂φ
vt − ∆v = − − ∆u
∂u ∂t i
∂u2 ∂xi ∂u
∂φ   X ∂ 2 φ  ∂u 2
(8.9) = ut − ∆u −
∂u i
∂u2 ∂xi
X ∂ 2 φ  ∂u 2
=− 2 ∂x
60
i
∂u i
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 15

Where the last step uses convexity of φ. Then we see that this is
indeed a subsolution.

(d). Define v := |Du|2 + u2t , where ut solves the heat equation. Then,
we first calculate:

∂
|Du|2 = 2Du · Duxi
∂xi
∂
2Du · Duxi = 2|Duxi |2 + 2Du · Duxi xi
∂xi
And hence, from the above, ∆|Du|2 = 2 i |Duxi xi |2 + 2Du · D(∆u).
P

∂
Likewise, it is obvious that ∂t
|Du|2 = 2Du · Dut . We also see:

∂ 2
u = 2ut utxi
∂xi t
∂
2ut utxi = 2u2txi + 2ut ut xi xi
∂xi
And similarly, summing over i, ∆u2t = 2 i u2txi + 2ut (∆u)t . Also,
P

∂ 2
u
∂t t
= 2ut utt . Using all of the above:

X X
vt − ∆v = 2Du · Dut + 2ut utt − 2 |Duxi xi |2 − 2Du · D(∆u) − 2 u2txi − 2ut (∆u)t
i i
X 
= 2Du · D(ut − ∆u) + 2ut (ut − ∆u)t − 2 |Duxi xi |2 + u2txi
i
X 
2
= −2 |Duxi xi | + u2txi 60
i

So we conclude that v is indeed a subsolution as asserted.

9. Chapter 2, Problem 18

Suppose that utt − ∆u = 0 in Rn × (0, ∞), with u(x, 0) = 0 and

ut (x, 0) = h. Then, set v := ut .
16 KELLER VANDEBOGERT

It is clear that vtt − ∆v = (utt − ∆u)t = 0 in Rn × (0, ∞). Similarly,

v(x, 0) = ut (x, 0) = h.
Finally, vt (x, 0) = utt (x, 0) = ∆u(x, 0). But u(x, 0) = 0, and hence
∆u(x, 0) = 0, so that vt (x, 0) = 0, as desired.

10. Chapter 2, Problem 19

(a). Suppose uxy = 0. Then, certainly ux = f (x), where f is any

´x
function. Integrating again, we see that u(x, y) = x0 f (x)dx + g(y).
Redefining our functions, we conclude that u(x, y) = F (x) + G(y) for
arbitrary functions F and G.

(b). Set ξ = x + t and η = x − t. Then, for u(ξ, η), employ the chain
rule to find:

ut (ξ, η) = uξ − uη

utt (ξ, η) = uξξ − 2uξη + uηη

Similarly,

ux (ξ, η) = uξ + uη

uxx (ξ, η) = uξξ + 2uξη + uηη

Assuming first that utt − uxx = 0, subtracting the above becomes

−4uξη = 0, so uξη = 0. Conversely, if uξη = 0, then, noting that
x = η/2 + ξ/2 and t = ξ/2 − η/2:

uξ = ux /2 + ut /2

uξη = uxx /4 − utt /4

And hence utt − uxx = 0.

 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 17

(c). Using the previous two parts, we can immediately deduce that if
utt − uxx = 0, then u = F (x + t) + G(x − t) for some functions F ,
G. Suppose that u(x, 0) = g, ut (x, 0) = h. Then, this implies that
g(x) = F (x) + G(x) and h(x) = F 0 (x) − G0 (x). Solving for F and G,
we find:

1 0
F 0 (x) =


g (x) + h(x)
2
1
G0 (x) = g 0 (x) − h(x)


2
And hence, after integrating we find:

(10.1)
ˆ ˆ x−t
1  x+t 0 
u(x, t) = g (y) + h(y)dy + g 0 (y) − h(y)dy
2 0 0
ˆ ˆ x−t
1   1  x+t 
= g(x + t) + g(x − t) + h(y)dy − h(y)dy
2 2 0 0
1  1 ˆ x+t
= g(x + t) + g(x − t) + h(y)dy
2 2 x−t
Which is precisely D’Alembert’s solution, so we are done.

(d). Suppose first that u is a left moving wave. In particular, this

means that ux + ut = 0, where u(x, 0) = g(x). Then, this is merely the
solution to the transport equation, so u(x, t) = g(x − t). Using this, we
also know that ut (x, 0) = h(x). But ut = −g 0 (x − t) =⇒ ut (x, 0) =
−g 0 (x). Then, we see that h(x) = −g 0 (x).
Similarly, if we have that u is a right moving wave, u satisfies ut −
ux = 0, with u(x, 0) = g(x). Again, using the solution to the transport
equation, u(x, t) = g(x + t). This then shows that ut = g 0 (x + t) and
hence g 0 (x) = h(x), and these are the conditions for the solution being
a left and right moving wave.
18 KELLER VANDEBOGERT

11. Chapter 2, Problem 24

(a). Suppose u solves the wave equation in R × (0, ∞). Then, with
´∞ ´∞
k(t) := 12 −∞ u2t (x, t)dx and p(t) = 12 −∞ u2x (x, t)dx. Then, u can of
course be solved using D’Alembert’s solution. Firstly, however, note:

ˆ ∞
d
k(t) = ut utt dx
dt −∞
ˆ ∞
d
p(t) = ux uxt dx
dt −∞
´∞
Recalling that utt = uxx , we can now integrate the expression −∞ ut utt dx =
´∞
−∞
ut uxx dx by parts.

ˆ ∞ ˆ ∞
∞
ut uxx dx = ut ux −∞ − ux uxt dx
−∞ −∞
Since g and h have compact support and noticing the form of D’Alembert’s
∞
solution, this immediately implies that ut ux −∞
= 0. Then,

ˆ ∞ ˆ ∞
d d
k(t) + p(t) = − ux uxt dx + ux uxt dx = 0
dt dt −∞ −∞
And hence k(t) + p(t) is a constant with respect to t.

(b). Using parts (a) and (b) of the previous problem, we know that
u(x, t) = F (x + t) + G(x − t) for some functions F , G. Noting the
dependence on our initial conditions for D’Alembert’s solution, we can
also conclude immediately that F and G must have compact support.
Now, it is simple to see:

ut = F 0 (x + t) − G0 (x − t)

ux = F 0 (x + t) + G0 (x − t)

And hence:
 ADVANCED PARTIAL DIFFERENTIAL EQUATIONS: HOMEWORK 3 19

ˆ ∞ ˆ ∞
u2t − u2x dx = −4 F 0 (x + t)G0 (x − t)dx
−∞ −∞
But for any compactly supported function, its derivative must also
be compactly supported. Suppose that the support of both of these is
contained in the interval [−M, M ] for sufficiently large M . Then for
any x, it is never possible for x − 2M and x + 2M to both be contained
in [−M, M ], since else the length of this interval would be at least 4M .
In other words, for t > 2M , either G0 (x − t) or F 0 (x + t) must vanish,
implying that

k(t) = p(t)

For sufficiently large t.