SOLUTIONS TO EXERCISES IN PARTIAL DIFFERENTIAL EQUATIONS

EDWARD CHERNYSH

In this document we solve multiple exercises in preparation for the midterm

exam in Math 475 at McGill university. The problems were taken from either the
course notes or the assignments. We focus on the following topics:
◦ The method of characteristics,
◦ 1D wave equation,
◦ 3D wave equation,
◦ distributions (generalized functions),
◦ Fourier transforms.

1. Method of Characteristics
Exercise 1.1. We know from the lectures that the general solution to the PDE
aux + buy + u = 0, a,0
ax+by
−
is u(x, y) = f (bx − ay)e a2 +b2 for a smooth function f of a single variable. Show that
this is equivalent to saying that there exists a function g such that
x
u(x, y) = g(bx − ay)e− a .
Solution. This follows at once from the observation that
ax + by x x −a2 x − bya + xa2 + xb2 x bx − ay x
− 2 2
+ − = 2 2
− =− 2 2
− .
a +b a a a(a + b ) a a(a + b ) a

Exercise 1.2. Suppose that every solution to aux +buy = 0 satisfies u(1, 2) = u(3, 6).
What is ba ?
Solution. The PDE may be rewritten as ∇u ·(a, b) = 0. This means that all solutions
are constant along curves with direction (a, b) at all points. These are precisely
curves of the form:
b
y(x) = x + c, c ∈ R .
a
Date: October 31, 2017 – Spooky!
1
2 E.CHERNYSH

Since u(1, 2) = u(3, 6) for all solutions, it follows that (1, 2) and (3, 6) live on the
same characteristic curve. Hence,
b
4 = 6 − 2 = (3 − 1)
a
which implies that ba = 2. 
Exercise 1.3. Using the methods of characteristics, solve
ux1 + x1 ux2 + ux3 = u, u(x1 , x2 , 0) = x1 + x2 .
Solution. The characteristic curves of the equation are given by:
ẋ1 (s) = 1, ẋ2 (s) = x1 (s), ẋ3 (s) = 1
with z(s) = u(x1 (s), x2 (s), x3 (s)). Along these curves, by the chain rule, ż(s) = z(s).
This means that
z(s) = Ces
along these curves. The characteristics can then be solves by taking
s2
x1 (s) = s + c1 , x3 (s) = s, + c1 s + c2 .
x2 (s) =
2
Here we have taken x3 (s) = s so that x3 (0) = 0. Therefore,
z(0) = Ce0 = C = x1 (0) + x2 (0) = c1 + c2 .
Hence, u(x1 , x2 , x3 ) = (c1 + c2 )ex3 along these curves. Clearly,
c1 = x1 − x3
so that
x32 x32
c2 = x2 − + (x3 − x1 )x3 = x2 + − x1 x3 .
2 2
Putting all this together yields
x32 x3
!
u(x1 , x2 , x3 ) = x1 − x3 + x2 − x1 x3 + e .
2

Exercise 1.4. Suppose u(x, t) solves ut +uux = 0 in R ×(0, ∞) with u(1, 1) = 5. Along
which curve in the x − t plane do we know the values of u?
Solution. Along the characteristics we know that the solution u will be constant.
Thus, we are asked to determine which characteristic curve the point (1, 1) lies.
Introducing a dummy variable s
ṫ(s) = s, z(s) = C, ẋ(s) = C
 SOLVED EXERCISES 3

where z(s) = u(x(s), t(s)). This gives t(s) = s + 1 and x(s) = Cs + 1 (where we have
chosen our arbitrary constants so that (x(0), t(0)) = (1, 1)). It follows that C = 5.
Hence, we know the values of u on
x(t) = 5(t − 1) + 1 = 5t − 4.

Exercise 1.5. Use the method of characteristics to solve the equation
ut + uux = 1, u(x, 0) = x for x, t ≥ 0.
Solution. As usual, we take t(s) = s as a dummy parameter so that t(0) = 0. If
ẋ(s) = z(s) then ż(s) = 1 along these curves. Hence,
z(s) = s + C
s2
so that x(s) = 2 + Cs + D. Observe that for s = 0:
z(0) = u(x(0), t(0)) = x(0) = D.
Therefore C = D which allows us to write
t2
x(s) ↔ x(t) = + C(t + 1).
2
This implies that
2x − t
C= .
2(t + 1)
Thus, along these curves
2x − t
u(x, t) = t + .
2(t + 1)

Exercise 1.6. Employ the method of characteristics to find a general solution to
the PDE ux + uy + uz = 0. Find a particular solution when u(x, y, 0) = x2 + y 2 .
Solution. The characteristic curves are
x(s) = s + c1 , y(s) = s + c2 , z(s) = s + c3 .
Without loss of generality we take c3 = 0. The PDE then states that any C 1 (R 3 )
solution is constant along these curves. Hence, given a triple (x, y, z) in R 3 we
need only determine which characteristic curve it lies upon.1 Each of the curves
is determined completely, and uniquely, by c1 and c2 . Hence,
u(x, y, z) = f (c1 , c2 ) = f (x − z, y − z)
1It is easy to see that, even with our choice of c , every point in R 3 lies on a characteristic.
3
4 E.CHERNYSH

for some function f in two variables. Now assume that the solution u satisfies the
auxiliary condition u(x, y, 0) = x2 + y 2 . Then
u(x, y, 0) = f (x, y) = x2 + y 2 .
Therefore u(x, y, z) = (x − z)2 + (y − z)2 . 
Exercise 1.7. In each of the following, let u(x, t) be a smooth solution to the stated
PDE such that u(4, 1) = 1. In each case, give a point x ∈ R such that we know the
value of u(x, 0).
(i) ut + ux = 0,
(ii) ut + uux = 0,
(iii) ut + t 2 ux = 0,
(iv) ut + ux + u = 0.
Solution.
(i) The characteristics are of the form t(s) = s + c1 and x(s) = s + c2 . An appro-
priate choice of c1,2 yields
t(s) = s + 1, x(s) = s + 4.
Thus, at s = −1 we find that u(x(−1), t(−1)) = u(3, 0) = 1 since any solution
u must be constant along these characteristics.
(ii) Again, take t(s) = s + 1 and ẋ(s) = z(s). We know that z(s) = C for some con-
stant C (as implied by the PDE) so that x(s) = Cs + 4. The initial condition
u(4, 1) = 1 implies that c = 1. Therefore we are left with the characteristics
t(s) = s + 1, x(s) = s + 4.
Thus we know that u(3, 0) = 1.
(iii) Choose t(s) = s and ẋ(s) = s2 . We then find that
s3
x(s) =
+C
3
where we take C = 4 so that x(0) = 4. Hence,
s3
x(s) = + 4.
3
At s = −1:
1 11
x(−1) = 4 − = .
3 3
It follows (since u will be constant along such a characteristic and (4, 1) is
a point on this curve) that u(11/3, 0) = 1.
 SOLVED EXERCISES 5

(iv) Here we choose t(s) = s + 1 and x(s) = s + 4 so that (x(0), t(0)) = (4, 1) as per
our initial data. However, we instead have
ż(s) = −z(s) =⇒ z(s) = Ce−s .
At s = 0 we know that z(0) = u(x(0), t(0)) = Ce0 = C = 1. Therefore,
z(0) = e−s
which shows that u(3, 0) = e (take s = −1 in our parametrized curves).


2. Wave Equation
Exercise 2.1. Let E(x, y, z, t) and B(x, y, z, t) be smooth electric and magnetic fields,
respectively. They are governed by Maxwell’s equations:
∇ × E = −∂t B, ∇ × B = µε∂t E, ∇ · E = 0, ∇ · B = 0.
in the above µ and ε are constants (that you likely encountered in physics). Show
that if
E(x, y, z, t) = (0, 0, E(x, t)) and B(x, y, z, t) = (0, B(x, t), 0)
then both E(x, t) and B(x, t) satisfy the wave equation:
∂tt u − c2 ∂xx u ≡ 0, c = (µε)−1/2 .
Solution. We first relate B(x, t) to E(x, t) using Maxwell’s equations. Observe that
(0, −Bt (x, t), 0) = −∂t B(x, y, z, t) = ∇ × E(x, y, z, t) = (0, −Ex (x, t), 0).
Thus, Bt (x, t) = Ex (x, t) . In like,
µε(0, 0, Et (x, t)) = ∇ × B(x, y, z, t) = (0, 0, Bx (x, t)).
Thus, µεEt (x, t) = Bx (x, t). Finally,
Exx (x, t) = ∂x Bt (x, t) = Btx (x, t) = ∂t Bx (x, t) = µε∂Et (x, t) = µεEtt (x, t).
Similarly, B(x, t) satisfies the wave equation. 
Exercise 2.2. Fix a time t > 0 and assume φ1,2 , ψ1,2 are bounded functions defined
on R . Let ui , for i = 1, 2, denote the solution to
∂tt ui − c2 ∂xx u ≡ 0, ui (x, 0) = φi (x), ∂t ui (x, 0) = ψ(x).
Prove that for every ε > 0 there exists δ > 0 such that

φ1 − φ2 < δ and φ1 − φ2 < δ =⇒ ku1 − u2 k∞ < ε.
6 E.CHERNYSH

Solution. Applying d’Albembert’s formula, we find that

ˆ
1 1 x+ct
u1 (x, t) = (φ1 (x + ct) + φ1 (x − ct)) + ψ (σ ) dσ
2 2c x−ct 1
ˆ
1 1 x+ct
u2 (x, t) = (φ2 (x + ct) + φ2 (x − ct)) + ψ (σ ) dσ .
2 2c x−ct 2
This implies, in particular, that for t > 0 fixed:
   
φ1 (x + ct) − φ2 (x + ct) + φ1 (x − ct) − φ2 (x − ct)
|u1 (x, t) − u2 (x, t)| ≤
2
ˆ x+ct 
1 ψ (σ ) − ψ (σ ) dσ .
+ 1 2
2c x−ct
Especially,
ˆ
1 x+ct
|u1 (x, t) − u2 (x, t)| ≤ φ1 − φ2 ∞ + ψ1 − ψ2 dσ .
∞
2c x−ct
Fix ε > 0 and take δ > 0 such that δ(1 + t) < ε. Then, for each x:
δ
|u1 (x, t) − u2 (x, t)| ≤ δ + · 2ct = δ(1 + t) < ε.
2c
Take now the supremum over x ∈ R ; this yields
ku1 − u2 k∞ ≤ δ(1 + t) < ε.

Exercise 2.3. Suppose that the propagation due to a pressure disturbance in 3D
is modeled by the 3D wave equation: utt = ∆u. At time t = 0 an explosion occurs
at position x = 0 inducing the initial conditions:
φ(x) ≡ 0, ψ(x) = 1|x|≤1 (x).
(i) What is the value of u(x, 10) where x = (10, 0, 0)?
(ii) At t = 10, what is the value of u at the point x = (20, 8, 17)?
(iii) At what times will u((20, 20, 20, ), t) be non-zero?
Solution. Before we proceed, we apply Kirchoff’s formula to obtain an explicit
representation of u: ¨
1
u(x, t) = ψ(y) dS.
4πt ∂B(x,t)
(i). Directly taking x = (10, 0, 0) and t = 10:
ˆ
1
(x, t) 7→ dS
40π ∂B((10,0,0),10)∩B(0,1)
 SOLVED EXERCISES 7

(ii). We shall show that |y − x| = 10 implies that |y| > 1. This will imply that u
evaluates to zero at (x, 10). Indeed,
|x| − |y| ≤ 10 =⇒ |y| > 1.
(iii). It is easier to determine at which points the displacement vanishes. This is
equivalent to saying that
|y − x| = t =⇒ |y| ≥ 1.
This can occur only if
√ √
0 ≤ t <≤ 1200 − 1, t ≥ 1200 + 1.

Exercise 2.4. Consider a vibrating infinite string with initial disturbance at t = 0
in the intervals [1, 2] and [4, 5]. At t = 10, at which positions will one feel these
disturbances?
Solution. Applying d’Alembert’s formula with t = 10 shows that we wish to know
for which x:
[x − 10, x + 10] ∩ ([1, 2] ∩ [4, 5]) , ∅.
First, note that [x − 10, x + 10] intersects [1, 2] if and only if
x + 10 ≥ 1 and x − 10 ≤ 2.
That is, if and only if x ∈ [−9, 12]. Similarly, [x − 10, x + 10] intersects [4, 5] if and
only if x ∈ [−6, 15]. Hence, we feel the disturbance for x ∈ [−9, 15]. 
Exercise 2.5. Consider the 1D wave equation ut t = ux x for (x, t) ∈ R × (0, ∞) to-
gether with the initial constraints
φ(x) = 0, ut (x, 0) = ψ(x)
where (
1, if |x − 3| ≤ 1 or |x + 3| ≤ 1,
ψ(x) =
0, else.
(i) At the time t = 1 where on the string is the displacement non-zero?
(ii) At time t = 10 which points have maximal displacement?
(iii) Compute u(0, t).
Solution. Before we proceed, we note that by d’Alembert’s formula any solution is
of the form ˆ
1 x+t
u(x, t) = ψ(σ ) dσ .
2 x−t
´ x+t
(i). At time t = 1 we obtain u(x, 1) = 21 x−t ψ(σ ) dσ which is non-zero if and only if
(x − 1, x + 1) ∩ {x : |x − 3| ≤ 1 or |x + 3| ≤ 1} , Ø.
8 E.CHERNYSH

Note that |x − 3| ≤ 1 if and only if−1 ≤ x − 3 ≤ 1 which is equivalent to saying

2 ≤ x ≤ 4. Similarly, |x + 3| ≤ 1 if and only if −4 ≤ x ≤ −2. Therefore, we seek
points x such that
(x − 1, x + 1) ∩ [[−4, −2] t [2, 4]] , Ø.
Clearly, (x − 1, x + 1) ∩ (−4, −2) is non-empty if and only if x − 1 < −2 and x + 1 > −4.
That is, if and only if x ∈ (−5, −1). In like, (x − 1, x + 1) intersects (2, 4) if and only
if x − 1 < 4 and x + 1 > 2. Thus, if and only if x ∈ (1, 5).

(ii) The solution u(x, 10) will have maximal displacement whenever (x − 10, x + 10)
covers [[−4, −2] t [2, 4]]. This will occur if and only if x − 10 ≤ −4 and x + 10 ≥ 4.
Thus, if and only if x ∈ [−6, 6].
´t
(iii). The “difficulty” lies in computing −t ψ(σ ) dσ . If t ≤ 2 then clearly (−t, t) does
not intersect (−4, −2) t (2, 4) which implies that u(0, t) = 0. On the other-hand, if
t ≥ 4 then (−t, t) engulfs (−4, −2) t (2, 4) which would imply that
ˆ
1 t 2+2
ψ(σ ) dσ = = 2.
2 −t 2
Suppose now that t ∈ (2, 4). Then,
t − 2 + (−2 + t)
u(0, t) = = t − 2.
2
That is, 


0, if t ≤ 2,

u(0, t) = t − 2, if 2 < t < 4,



2,

if t ≥ 4.

Exercise 2.6. Assume a phenomenon propagates in 3-dimensional space accord-
ing to the equation utt − ∆u ≡ 0. Suppose a disturbance at x = 0 creates an initial
change in velocity according to
(
1, if |x| ≤ 1,
ψ(x) = .
0, else.
Evaluate u(0, t) for t ≥ 0
Solution. Kirchoff’s formula yields that for (x, t) ∈ R × (0, ∞):
¨ ¨
1 1
u(0, t) = ψ(y) dS = dS
4πt ∂B(0,t) 4πt ∂B(0,t)∩B(0,1)
 SOLVED EXERCISES 9

Thus, since ∂(0, t) does not intersect B(0, 1) for t > 1, it follows that u(0, t) vanishes
for t > 1. Otherwise, if 0 < t < 1:
4πt 2
u(0, t) = = t.
4πt


3. Distributions
Henceforth, we shall denote by D(R ) the space of all smooth (infinitely differ-
entiable) functions of compact support defined on R . The functions in D(R ) are
allowed to take complex values. We endow D(R ) with the metric topology by
defining
 
k·k∞ : D(R ) −→ [0, ∞), φ 7→ sup φ(x) .
x∈R

It is easy to show that k·k∞ is indeed a valid norm on D(R ), and it therefore induces
a metric on the space. A distribution is then defined to be a continuous linear
functional
F : D(R ) −→ C.
If f ∈ L1loc (R ) it induces a distribution:

hFf , φi.

Exercise 3.1. Give an example of a continuous function f who derivatives f 0 , f 00

in the sense of distributions are generated by functions such that f 000 = δ0 .2

Solution. We define
x2 /2, if x ≥ 0,
(
f (x) :=
0, else.
This function is clearly continuous on R and has derivative (in the classical sense):
(
0 x, if x ≥ 0,
f (x) =
0, else.

2We denote by δ the Dirac delta “function”. As a distribution, this is the functional given by
0
hδ0 , φi = φ(0) for all φ ∈ D(R ).
10 E.CHERNYSH

Now, for each φ ∈ D(R ):

ˆ ˆ ∞
x2 00
hFf00 , φi = hFf , φ00 i = 0
f (x)φ (x) dx = φ (x) dx
0 2
R
ˆ ∞
=− xφ0 (x) dx
0
ˆ ∞
= φ(x) dx.
0
In the sense of distributions, f 00 = H0 (x); where H0 (x) = 1[0,∞) (x) is the Heaviside
function. Since H00 (x) = δ0 in the sense of distributions, we are done. 
Exercise 3.2. Prove that the sequence of hat functions:
, if |x| ≤ n1 ,
(n
fn (x) := 2
0, else,
converges to δ0 in the sense of distributions.
Solution. Let ε > 0 be given and fix φ ∈ D(R ). We must show that
ˆ
lim fn (x)φ(x) dx = φ(0).
n→∞ R

For each n ∈ N
ˆ  ˆ 1/n ˆ 
 f (x)φ(x) dx − φ(0) =  n n 1/n 
 R n   −1/n 2 φ(x) dx − 2 −1/n φ(0) dx
ˆ
n 1/n  
≤ φ(x) − φ(0) dx
2 −1/n
 
≤ sup φ(x) − φ(0) .
|x|≤1/n
 
Since φ is continuous, there exists δ > 0 such that φ(x) − φ(0) < ε whenever |x| < δ.
Now, for all n sufficiently large it is clear that |x| < 1/n < δ whence
ˆ 
 f (x)φ(x) dx − φ(0) ≤ ε.
 R n 
Letting n → ∞ we obtain
ˆ 
lim sup  fn (x)φ(x) dx − φ(0) ≤ ε

n→∞ R
for each ε > 0. It follows that
ˆ
n→∞
fn (x)φ(x) dx −−−−−→ φ(0).
R
 SOLVED EXERCISES 11


Exercise 3.3. Define
e−x , if x > 0,
(
f (x) :=
−ex , if x ≤ 0.
Determine f 0 in the sense of distributions and show that as distributions:
f 00 = 2δ00 + f .
Solution. Fix a test function φ and evaluate
ˆ 0 ˆ ∞
0 0 x 0
hFf , φi = −hFf , φ i = e φ (x) dx − e−x φ0 (x) dx
−∞ 0
ˆ 0 ˆ ∞
x
= 2φ(0) − e φ(x) dx − e−x φ(x) dx.
ˆ−∞ 0

= 2φ(0) + g(x)φ(x) dx
R
where g ∈ L1loc (R ) is given by
−e−x , if x > 0,
(
g(x) :=
−ex , if x ≤ 0.
Taking, again, the derivative in the sense of distributions we find that
hFf00 , φi = −hFf0 , φ0 i = −hδ0 , φ0 i − hFg , φ0 i
where
ˆ ∞ ˆ 0 ˆ ∞
0 −x 0 x 0
−hFg , φ i = e φ (x) dx + e φ (x) dx = −φ(x) + e−x φ(x) dx + φ(0)
0 −∞ 0
ˆ 0
− ex φ(x) dx.
−∞
This completes the problem. 
Exercise 3.4. Find a locally integrable function f (x) such that
f 0 (x) = x2 + 4x + δ2
in the sense of distributions.
Solution. The approach here is classical. Consider the function
x3
f (x) := + 2x2 + H2 (x), H2 (x) := 1[2,∞) (x).
3
12 E.CHERNYSH

Fix a function φ ∈ D(R ) and compute

ˆ 2 3 ! ˆ ∞ 3 !
0 0 x 2 0 x 2
hFf , φi = −hFf , φ i = − + 2x φ (x) dx − + 2x + 1 φ0 (x) dx
−∞ 3 2 3
where we label these integrals by I1 and I2 respectively. Clearly,
ˆ 2 3 ! ˆ 2
x 8
 
2
I1 = − 0
+ 2x φ (x) dx = − 8 + φ(2) + (x2 + 4x)φ(x) dx
−∞ 3 3 −∞
whilst
ˆ ∞
8
 
I2 = 8 + + 1 φ(2) + (x2 + 4x)φ(x) dx.
3 2
Thus, ˆ
hFf0 , φi = hδ2 , φi + (x2 + 4x)φ(x) dx.
R

Exercise 3.5. Suppose {Fn }n∈N is a sequence of distributions converging (in distri-
(k)
bution) to a distribution F. Show that for each k ∈ N one has that Fn → F (k) in
distribution.
Proof. It suffices to prove the claim for k = 1. Note that for each φ ∈ D(R ):
lim hFn0 , φi = − lim hFn , φ0 i = −hF, φ0 i = hF 0 , φi.
n→∞ n→∞

Exercise 3.6. Let f : R −→ C be a continuously differentiable function. Show that
f 0 is a derivative of f in the sense of distributions.
Proof. Fix a function φ ∈ D(R ) and observe that
ˆ ˆ
hFf , φi = − f (x)φ (x) dx = f 0 (x)φ(x) dx.
0 0
R R

Exercise 3.7. Define
x3 ,
(
if x ≥ 0,
g(x) =
−x3 , if x < 0.
(i) Compute g (4) in the sense of distributions.
(ii) Define ˆ
g(x − y)
f (x) = 6
dy.
R 1+y
Accept that f (x) is smooth and compute f (4) (x).
 SOLVED EXERCISES 13

Solution. (i). Let Fg denote the distribution induced by g (this makes sense since
g ∈ L1loc (R )). For a function φ ∈ D(R ) we find that
ˆ
(4)
hFg , φi = hFg , φ i = g(x)φ(4) (x) dx
(4)

ˆ ∞ ˆ 0
R
3 (4)
= x φ (x) dx − x3 φ(4) (x) dx
0 −∞
ˆ ∞ ˆ 0
2 (3)
= −3 x φ (x) dx + 3 x2 φ(3) (x) dx
0 −∞
ˆ ∞ ˆ 0
=6 xφ(2) (x) dx − 6 xφ(2) (x) dx
0 −∞
ˆ ∞ ˆ 0
= −6 φ0 (x) dx + 6 φ0 (x) dx
0 −∞

which reduces to 12φ(0). Hence, g (4) = 12δ0 in the sense of distributions.

(ii). Using the fact that convolution is commutative:

ˆ
g(y)
f (x) = 6
dy
R 1 + (x − y)

which implies that

ˆ ˆ
(4) d4 g(y) ∂4 g(y)
f (x) = 6
dy = 6
dy.
dx4 R 1 + (x − y)
4
R ∂x 1 + (x − y)

By symmetry,
ˆ ˆ
∂4 g(y) ∂4 1
6
dy = g(y) dy
4
R ∂x 1 + (x − y) R ∂y 1 + (x − y)6
4

which is by very definition

(4)
hFg , φi
1
where φ(y) = 1+(x−y)6 with x fixed. We know from the first part that this is pre-

cisely 12δ0 which implies that

12
f (4) (x) = 12φ(0) = .
1 + x6

14 E.CHERNYSH

4. The Fourier Transform

Exercise 4.1. Suppose that y(x) is a smooth function that satisfies
y (4) + 4y (2) + y = f (x), f ∈ Cc (R ).
Using the Fourier transform, write an expression for ŷ in-terms of fˆ.
Solution. Using the linearity of F [·], we take the Fourier transform of both sides
of the ODE to obtain
fˆ(ξ) = yd
(4) (ξ) + 4y
d (2) (ξ) + ŷ(ξ) = i 4 ξ 4 ŷ(ξ) + 4i 2 ξ 2 ŷ(ξ) + ŷ(ξ).

Therefore, for ξ ∈ R :  
fˆ(ξ) = ŷ(ξ) ξ 4 − 4ξ 2 + 1
so that
fˆ(ξ)
ŷ(ξ) = .
ξ 4 − 4ξ 2 + 1

Exercise 4.2. Define f (x) := e−|x| /2. Compute the Fourier transform of f .
Solution. For fixed ξ ∈ R we have
ˆ ∞ −|x| ˆ ˆ
ˆ e −ixξ 1 0 x(1−iξ) 1 ∞ −x(1+iξ)
f (ξ) = e dx = e dx + e dxs
−∞ 2 2 −∞ 2 0
0 ∞
ex(1−iξ)  −e−x(1+iξ) 
=  +
2(1 − iξ) −∞ 2(1 + iξ) 0

1 1 1
 
= +
2 1 − iξ 1 + iξ
1 1
= = .
(1 − iξ)(1 + iξ) 1 + ξ 2

Exercise 4.3. Using the Fourier transform, solve the ODE
y 00 (x) − y(x) = f (x), f ∈ C(R ) ∩ L1 (R ).
Solution. By linearity of the Fourier transform, for fixed ξ ∈ R
fˆ(ξ) = fc00 (ξ) − ŷ(ξ) = i 2 ξ 2 ŷ(ξ) − ŷ(ξ) = −(1 + ξ 2 )ŷ(ξ).

Hence,
fˆ(ξ)
ŷ(ξ) = − .
1 + ξ2
 SOLVED EXERCISES 15

By virtue of the previous problem, we know that g(x) = e−|x| 2 is the Fourier inverse
1
to 1+ξ 2 . We may therefore write,
ˆ
∨ ∨ 1
f (y)e−|x−y | dy.
 
ˆ
y(x) = − f (ξ) · ĝ(ξ) = − f ∗ g = −(f ∗ g)(x) = −
d
2 R

Prove the Riemann-Lebesgue lemma:
Lemma 1. Let f ∈ L1 (R ), then
lim fˆ(n) = 0.
n→∞

Proof. We proceed in 2 steps.

Step 1. The claim holds when f is the characteristic function of an interval. Sup-
pose that f (x) = 1[a,b] (x) where a, b are real numbers. By direct calculation,
ˆ ˆ b −inx b

−inb − e−ina )
fˆ(n) = f (x)e−inx dx = e−inx dx =
ie  = i(e .
R a n a n
Thus,
  2 n→∞
fˆ(ξ) ≤ −−−−−→ 0.
n
Therefore, this must also hold for finite linear combination of characteristic func-
tions of intervals (i.e. step functions).

Step 2. This holds for general f ∈ L1 (R ). Certainly, let ε > 0 be given. By their
density in L1 (R ), there exists a step function ϕ = N
P
j=1 aj 1[aj ,bj ] such that
ˆ
ε
|f (x) − ϕ(x)| dx < .
R 2
Therefore,
ˆ ˆ ˆ
−inx −inx
f (x)e dx = (f (x) − ϕ(x))e dx + ϕ(x)e−inx dx.
R R R
Which then implies
ˆ  ˆ  ˆ 
 f (x)e−inx dx ≤  (f (x) − ϕ(x))e−inx dx +  ϕ(x)e−inx dx
ˆ ˆ
 R   R   R 

≤ |f (x) − ϕ(x)| dx +  ϕ(x)e−inx dx .
R R
16 E.CHERNYSH

Now, there exists K ∈ N such that n ≥ N implies, by step 1,

ˆ 
 ϕ(x)e−inx dx < ε .
 R  2
Putting all this together:
ˆ 
 f (x)e−inx dx < ε, ∀n ≥ K.

R

This is, by definition, equivalent to saying that
lim fˆ(n) = 0.
n→∞