TMA4170 Fourier

Analysis
Spring 2017
Norwegian University of Science and
Technology Exercise set 5 solution
Department of Mathematical
Sciences

1 B&N: 3.3 We want to solve the differential equation

au 00 + bu 0 + cu = f ,

where f (t ) is the impulse function

f0
(
2h , −h ≤ t ≤ h
f (t ) =
0, otherwise,

taken in the limit h → 0. (Strictly speaking, this definition of f makes sense only when
considering it as a regular distribution, i.e. it can only be applied within an integral, but we
shall not dig deeper into the details of this here). We begin by taking the Fourier transform
of both sides of the ODE. Since F (y 0 ) = i λF (y), we get

−aλ2 û + i bλû + c û = fˆ

To find fˆ(λ), we use the definition of the Fourier transform and take the limit as h → 0:

Z∞
f0
fˆ(λ) = p f (t )e −i λt dt
2π
−∞
Zh
f0 1 −i λt
= lim p e dt
h→0 2π 2h
−h
f0 −1 −i λt h
· ¸
= lim p e
h→0 2π 2i hλ −h
f0 sin(λh)
=p lim
2πλ h→0 h
f0
=p λ
2πλ
f0
=p .
2π

Thus, we have

f0
µ ¶
1
û(λ) = p .
2π −aλ2 + i bλ + c

We find the zeros of the polynomial on the right hand side as

p
ib 4ac − b 2
λ± = ± = i µ ± ω,
2a 2a

February 14, 2017 Page 1 of 3

 Exercise set 5 solution

p
b 4ac−b 2
where µ = 2a and ω = 2a . We can then factorize and split the polynomial into partial
fractions to find
f0
µ ¶
1 1
û(λ) = p − .
2π2ωa −λ + ω + i µ −λ − ω + i µ
We could now use the inverse Fourier transform directly, which would require a contour
integral. However, it is easier to consult a table of Fourier transforms (keeping in mind that
tables may use different conventions for the 2π factor(s) in the transform) and find that if

e −βt e i γt , t ≥ 0
(
g (t ) =
0, t < 0,

then
1 i
ĝ (λ) = p .
2π −λ + γ + i β
We can thus recognise that
e i ωt −e −i ωt
³ ´
( f 0 −µt
2ωa e i , t ≥0
u(t ) =
0, t < 0,

that is,
f 0 −µt
(
ωa e sin(ωt ), t ≥ 0
u(t ) =
0, t < 0.

2 B&N: 3.10 The difference approximations to u 0 (t k ) and u 00 (t k ) are

u k − u k−1 u k+1 − 2u k + u k−1

u 0 (t k ) ' , u 00 (t k ) ' .
h h2
Inserting these into the differential equation evaluated at time t k gives
u k+1 − 2u k + u k−1 u k − u k−1
a 2
+b + cu k = f k
h h
⇒au k+1 + (ch 2 + bh − 2a)u k + (a − bh)u k−1 = h 2 f k ,

which is what we wanted to show.

3 B&N: 3.11 Let us consider the hint given in the exercise first. We want to show that there
can be no solution z with |z| = 1 to the quadratic equation

az 2 + βz + γ = 0, (1)

where β = ch 2 + bh − 2a, γ = a − bh, and a, b, c, h > 0. We shall show this by contradiction.

Assume that there does exist a z solving this such that |z| = 1. Then, |z|2 = zz = 1, so
multiplying (1) by z yields

0 = az + β + γz
= az + β + az − bhz
= (2a − bh)Re(z) + β + i bhIm(z).

February 14, 2017 Page 2 of 3

 Exercise set 5 solution

We must have equality for the real and imaginary part of this equation. Since a, b, c, h ∈ R,
we also have β ∈ R. Thus, the imaginary part gives Im(z) = 0 since b, h > 0, i.e. z = Re(z) ∈ R,
and so the problem reduces to

(2a − bh)z + β = 0.

Since |z| = 1, we must have z = 1 or z = −1. With z = 1, the above equation reduces to
ch 2 = 0, which is impossible. In the case z = −1, we need to have

bh − 2a + β = ch 2 + 2bh − 4a = 0.

This does seem to be possible; for some given b, c, h > 0 we can choose a to satisfy the
equality. Or, in other terms, given a, b, c > 0, z = −1 is a solution if one uses a step length h
satisfying the quadratic equation

ch 2 + 2bh − 4a = 0.

If one avoids this step length, there is no solution to (1) with |z| = 1. Now, what we want to
show is that

aw j + β + γw j 6= 0.

Multiplying by w j and using that |w| = 1, we want to show that

a(w j )2 + βw j + γ 6= 0.

Since we know that |w j | = 1, we know that this holds as long as ch 2 + 2bh − 4a 6= 0.

4 B&N: 3.12 Taking the DFT of both sides of equation (3.11) we find that

a ŷ k+1 + β ŷ k + γ ŷ k−1 = h 2 fˆk .

Next, we use the shift property of the DFT ( ŷ j +1 = w j ŷ k ), and the fact that w −1 = w to find

aw k ŷ k + β ŷ k + γw k ŷ k = h 2 fˆk ,

that is,

fˆk
ŷ k = h 2 .
aw k + β + γw k

February 14, 2017 Page 3 of 3