11.3.

Fourier series and forced oscillation

11.4 Approximation by trigonometric polynomials

Eugenia Malinnikova, NTNU

September 12, 2017

1
Forced oscillation, Fourier series

We consider linear second order differential equations with

constant coefficients:

ay 00 + by 0 + cy = f (x)

The idea is to expand f into a Fourier series and solve equations

ay 00 + by 0 + cy = a0
ay 00 + by 0 + cy = an cos nx + bn sin nx

using the method of undetermined coefficients.

2
Example
Let f (x) = |x|, −π ≤ x ≤ π and f is extended to a 2π-periodic
function. The Fourier series of f is
∞ ∞
π 4 X cos(2k − 1)x X
Sf (x) = − = a0 + an cos nx
2 π (2k − 1)2
k=1 n=1

We want to solve the equation y 00 + ω 2 y = f (x). Let solve it for

each term of the form an cos nx by the method of undetermined
coefficients. The particular solution is
an
yn (x) = cos nx
ω2 − n2
Then the solution is given by the series
∞
π 4X cos(2k − 1)x
y (x) = 2
−
2ω π (2k − 1)2 (ω 2 − (2k − 1)2 )
k=1

3
Separating frequencies

The formula from P the last example gives:

If f (x) = a0 + an cos nx + bn sin nx and we solve the equation
y 00 + ω 2 y = 0 then
∞
a0 X an cos nx + bn sin nx
y (x) = +
ω02 n=1 ω 2 − n2

In this sum the nth coefficients are divided by ω 2 − n2 . When

ω ≈ n0 we get y (x) with the coefficient on place n0 multiplied by a
large number and other coefficients comparable or smaller then
those of f .

This method of Fourier analysis of signals is built in our ears

(cochlea).

4
Approximation by trigonometric polynomials

Let f be a function on [−π, π]. We are looking for the best

approximation of f by a trigonometric polynomial of degree n, i.e.
function of the form
n
X
Tn (x) = A0 + (Ak cos kx + Bk sin kx),
k=1

where the error is calculated as

Z π
2
E = |f (x) − Tn (x)|2 dx.
−π

5
Best approximation

Theorem
Let Sf = a0 + ∞
P
k=1 (ak cos kx + bk sin kx) be the Fourier series of
piece-wise continuous function f . Then the best approximation by a
trigonometric polynomial of degree n is achieved with the
polynomial
n
X
Pn (x) = a0 + (ak cos kx + bk sin kx).
k=1

In this case the error is given by

Z π n
X
En2 = |f (x)|2 dx − 2πa02 − π (ak2 + bk2 )
−π k=1

6
Bessel’s inequality and Parseval’s identity
Since En ≥ 0 the theorem above implies that
Z π n
X
2
|f (x)| dx ≥ 2πa02 +π (ak2 + bk2 )
−π k=1

It turns out that as n goes to infinity the error tends to zero and the
following identity holds
Z π ∞
X ∞
X
|f (x)|2 dx = 2πa02 + π (ak2 + bk2 ) = 2π .|ck |2 .
−π k=1 k=−∞

The Parseval’s identity gives also the following formula for the error
∞
X
En2 = π (ak2 + bk2 ).
k=n+1