0% found this document useful (0 votes)
122 views12 pages

Fourier Series

1. The Fourier series expansion of a function f(x) over the interval (-π, π) involves representing f(x) as the sum of orthogonal basis functions (sinusoidal terms) with coefficients that can be determined from integrals of f(x) and the basis functions. 2. The Fourier series converges to f(x) at points where f(x) is differentiable or monotonic over an interval. 3. Any interval (a, b) can be transformed to (-π, π) by substituting X = (2x - (a + b))/(b - a). The Fourier series then involves terms of the form sin(nX) and cos(nX).
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
122 views12 pages

Fourier Series

1. The Fourier series expansion of a function f(x) over the interval (-π, π) involves representing f(x) as the sum of orthogonal basis functions (sinusoidal terms) with coefficients that can be determined from integrals of f(x) and the basis functions. 2. The Fourier series converges to f(x) at points where f(x) is differentiable or monotonic over an interval. 3. Any interval (a, b) can be transformed to (-π, π) by substituting X = (2x - (a + b))/(b - a). The Fourier series then involves terms of the form sin(nX) and cos(nX).
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

Fourier Series and their Applications

The functions 1, cos x, sin


 x, cos 2x, sin 2x, · · · are orthogonal over (−π, π).
Z π  0 m 6= n

cos mx cos nxdx =  π m = n 6= 0
−π 
2π m = n = 0
Z π
cos mx sin nxdx = 0 for all m, n
−π (
Z π 0 m 6= n
sin mx sin nxdx =
−π π m = n 6= 0
In fact the functions satisfy these relations over any interval (α, α + 2π).
Assuming that f (x), defined and integrable in (−π, π), has an expansion.

1 X
a0 + (an cos mx + bn sin nx)
2 n=1
uniformly convergent over (−π, π)
Z π
f (x)dx = πa0
Z−π
π 1Zπ
f (x) cos nxdx = πan therefore an + ibn = f (x)einx dx
−π
Z π π −π

f (x) sin nxdx = πbn


−π
These coefficients exist irrespective of whether or not the series converges
and is equal to f (x), and they are called the Fourier coefficients.

Sufficient Conditions for convergence


¯ ¯
¯ f (x) − f (ξ) ¯
a) If f (x) is differentiable at ξ (or if ∃m, such that ¯¯
¯
¯ < m,
¯
x−ξ ¯
x²(ξ − h, ξ + h)) then the fourier series converges at ξ to f (ξ).

b) If f (x) is monotonic in ξ < x < ξ + h and in ξ − h < x < ξ for some


h > 0, then the fourier series converges at ξ to the value
1
2
{f (ξ − 0) + f (ξ + 0)}.

1
General Range ³ ´
a+b
π x− 2
The range a ≤ x ≤ b is standardised by substituting X = ³
b−a
´
2
then −π < X < π.

1 X
The series a0 + an cos nX + bn sin nX
2 n=1 Ã ! Ã !

1 X 2x − (a + b) 2x − (a + b)
becomes a0 + an cos 2nπ + bn sin 2nπ
2 n=1 b−a b−a
Periodicity of f (x)
We suppose that f (x) is represented by the series (when cgt.) for all x,
hence since the sum function of the series is periodic, with period 2π, we
have f (x + 2π) = f (x) which defines f (x) outside the original range.

Fourier Series for 12 − t (0 < t < 1)


First consider the identity
e(m+ 2 )ix − e− 2 ix
1 1
m
X
nix e(m+1)ix − 1
1+ e = =
n=1 ³
e ix − 1
´ ³
2i sin´ 21 x ³ ´
cos m + 21 x + i sin m + 12 x − cos 21 x − i sin 21 x
=
2i sin 12 x
Hence for x real (x 6= 0, ±2π, · · ·)³ taking´ real and imaginary parts:
1
Xm
1 1 sin m + 2 x
Re : 1 + cos nx = +
n=1 2 ³ 2 sin 1 x
´ 2
1
1 X m
1 sin m + 2 x
or + cos nx = (1)
2 n=1 2 sin 12 x ³ ´
1
1 1 Xm
1 cos m + 2 x
Im : − cot x + sin nx = − (2)
2 2 n=1 2 sin 12 x
Integrate (1) and (2) from x to π ³ ´
m Z π sin m + 1 t
1 X sin nx 1 2
(1): (π − x) − = 1 dt (3)
2 n=1 n 2 x sin 2 t ³ ´
¶¸π " m #π Z π cos m + 1 t
1 − cos nt 1
· µ X 2
(2): − log sin t + =− 1 dt (4)
2 x n=1 n x
2 x sin 2 t
Now suppose δ ≤ x ≤ 2π − δ.

2
Then using the Riemann Lebesgue theorem, we have, letting m → ∞ in (3)
and (4)

1 X sin nx
(π − x) = (5)
2 n=1 n
1 X∞
(−1)n X ∞
cos nx
log sin x − + =0
2 n=1 n n=1 n

1 X cos nx
Therefore log 2 sin x = − (6)
2 n=1 n
Alternative Proof of (5) and (6)
Z ξ
dt
log(1 − ξ) = −
0 1−t
where we take a cut along the positive real axis in the t-plane from 1 to ∞.
The branch of log(1 − ξ) chosen is that which is real when ξ is real, and is
one-valued in the cut plane. In particular this vanishes at ξ = 0
1 tm
= 1 + t + · · · + tm−1 +
1−t 1−t
m
Z ξ
dt X ξ n Z ξ tm
therefore = + dt
0 1−t n=1 n 0 1−t
where the path is taken along the radius 0 − ξ.
DIAGRAM
For all t on the radius through ξ
|1 − t| ≥ | sin θ| Re(ξ) > 0
≥ 1 otherwise
Hence in all¯Z cases |1 − ¯ t| ≥
¯Z sin δ when δ ≤¯ arg ξ ≤ 2π − δ (0 < δ < π)
¯ ξ tm ¯ ¯ r (peiθ )m eiθ ¯
Therefore ¯¯ dt¯¯ = ¯¯ dp¯¯
¯ ¯ ¯ ¯
0 1−t 0 1−t
Z r
pm 1 rm+1 1
≤ dp = ≤ 0≤r≤1
0 | sin δ| ¯ sin δ¯ m + 1 (m + 1) sin δ
¯Z ξ tm dt ¯
Hence m→∞lim ¯¯ ¯=0 δ ≤ arg ξ ≤ 2π − δ
¯ ¯
0 1 − t¯
0≤r≤1
When r = 1 the convergence is uniform with respect to θ. Hence we have
X∞
ξn
log(1 − ξ) = −
n=1 n
Where the series converges on |ξ| = 1 except at ξ = 1, uniformly in
δ ≤ arg ξ ≤ 2π − δ

3
log(1 − ξ) = log |1 −Ãξ| + i arg(1
! − ξ)
1 π θ
µ ¶
= log 2 sin θ − i −
2 2 2
Taking real and imaginary parts gives

1 X sin nθ
π−θ =
2 1 n

1 cos nθ
µ ¶ X
log 2 sin θ = − 0 < θ < 2π
2 1 n
Convergence being uniform in δ ≤ θ ≤ 2π − δ.

Fourier Expansion of the Bernoulli polynomials in 0 ≤ t ≤ 1. Values of the


Bernoulli numbers. ∞
1 X sin nx
Put x = 2πt in (π − x) =
2 1 n

1 1 X sin 2πnt
Therefore t − = −
2 π 1 n

X sin 2πnt
= −2 (1) 0<t<1
1 2πn

X sin 2πnt
Therefore P1 (t) = −2
1 2πn
P20 (t) = P1 (t) P
Z 2
(0) = 0
t
Therefore P2 (t) = P1 (s)ds
0
The series (1) converges uniformly in ² ≤ t ≤ 1 − ²
∞ Z t
Z t X sin 2nπs
P1 (s)ds = −2 ds ²≤t≤1−²
² 1 ² 2nπ

X cos 2nπ² − cos 2nπt
= −2
1 (2nπ)2
The
¯ series ¯on the right converges absolutely and uniformly since
¯ cos(2nπt) ¯ 1 X 1
¯
¯ (2nπ)2 ¯
¯
≤ and converges.
(2nπ)2 n2
¯ ¯

Z t X 1 − cos 2nπt
Hence P1 (s)ds = −2 0≤t≤1
0 1 (2nπ)2
(using continuity)

cos 2nπt
Hence P2 (t) = P¯2 + 2
X
(2)
1 (2nπ)2

4

1 −2
P¯2 = −2
X
2
= S2
1 (2nπ) (2π)2
Next P30 (t) = P2 (t) − P¯2

X sin 2nπt
Therefore P3 (t) = 2 3
1 (2nπ)
and generally we have

cos 2nπt
P2m (t) − P¯2m = (−1)m−1 2
X
2m
m=1 (2nπ)

X sin 2nπt
P2m+1 (t) = (−1)m−1 2 2m+1
m=1 (2nπ)
2S2m
P¯2m = (−1)m
(2π)2m
We also have
φm (t)
= Pm (t) m = 2, 3, · · ·
m!
Bm
= (−1)m P¯2m
(2m)!
For k ≥ 2 it can be shown that
1
1 ≤ Sk ≤ 1 + k−2 (S2 − 1)
2
Therefore Sk = 1 + o(k)
2
Also P2m+1 (t) ∼ (−1)m−1 sin 2πt
(2π)2m+1
2
P2m (t) ∼ (−1)m−1 (1 − cos 2πt)
(2π)m
Fourier Series of the Square Wave

1 X sin nx
We have (π − x) = 0 < x < 2π
2 n=1 n
Write y = x − π

1 sin ny
(−1)n
X
− y= −π <y <π
2 n=1 n

sin nx
(−1)n−1
X
x=2 −π <x<π
n=1 n

sin nx
(−1)n−1
X
Write f (x) = 2
n=1 n

5
qqqqqqqqqq qqqqqqqqqq qqqqqqqqqq
­
­ ­ ­
­ ­ ­ ­
­ ­ ­
­ ­ ­
­ ­ ­
­ ­
­ ­ ­
­ ­ ­ ­
­ ­
­ ­ ­ ­
­ ­
­ qqqqqqqqqq­
­ q­q q q q q q q q q

Graph of f (x) is shown by solid lines.


Graph of f (x + π) is shown by broken lines.
Graph of f (x) − f (x + π) is shown by dotted lines.
The fourier series of f (x) − f (x + π) is then
∞ ∞
sin nx sin nx
(−1)n (−1)n−1 (−1)n
X X
2 −2
n=1 n n=1 n

X sin(2n + 1)x
=4
n=0 2n + 1 (

4 X sin(2n + 1)x +1 0<x<π
=
π n=0 2n + 1 −1 −π < x < 0
Find coefficients by direct integration.

Gibbs’ Phenomenon m
4X sin(2n + 1)x
Write Sm (x) =
π n=0 2n + 1
m
π d X sin(2m + 2)x
Sm (x) = cos(2n + 1)x =
4 dx n=0 2 sin x
π (2m + 1)π
This vanishes in o < x < π at x = ···
2m + 2 2m + 2
Sm (x) is symmetrical about π2 .
Hence consider the value of Sm for 0 < x < π2 , and in particular at
π
x= , the first max.
2m
µ +2 ¶ Z π
π π 2m+2 sin(2m + 2)t
Sm = dt
4 2m + 2 0 2 sin t
s
Put t = then we have
2m + 2

6
π π 1Z π sin sds 1 Z π sins s
µ ¶ µ ¶
sin = s = φ ds
4 2m + 2 2 0 (2m + 2) sin 2m+2 2 0 s 2m + 2
u
where φ(u) = .
sin u
s
Now 1 ≤ φ(u) ≤ φ(δ) 0 ≤ u ≤ δ < π and 0 ≤ ≤ π2m + 2
2m + 2
s π
µ ¶ µ ¶
So 1 ≤ φ ≤φ
2m + 2 2m + 2
sin s
1≥ ≥0 in 0 ≤ s ≤ π
sZ
π sin s Z π
sin s
µ
s

Hence ds ≤ φ ds
µ 0 s¶ Z 0 s 2m + 2
π π sin s
≤φ ds
2m + 2µ 0 s ¶
π
Since m→∞lim φ = 1 we have
2m
µ +2 ¶
Z π
sin s s Z π
sin s
lim φ ds = ds
s µ2m + 2 ¶ Z 0π s
m→∞ 0
π 2 sin s
Hence m→∞ lim Sm = ds ≈ 1.179 > 1
2m + 2 π 0 s
Dirichlet’s Formula (sufficient conditions for convergence)
Assume that f (x) is bounded and integrable over [−π, π], and
f (x + 2π) = f (x)
m
1 X
Write Sm (x) = a0 + (an cos nx + bn sin nx)
2 n=1
m
" #
1Zπ 1 X
= f (t) + (cos nt cos nx + sin nt sin nx) dt
π −π 2 n=1
m
" #
1Zπ 1 X
= f (t) + cos n(t − x) dt
π −π 2 ³ n=1 ´
1 Zπ sin m + 21 (t − x)
= f (t) dt
2π −π sin 12 (t³− x) ´
1 Z π−x sin m + 12 s
= f (x + s) ds
2π −π−x ³
sin 12 ´s
1 Zπ sin m + 21 s
= f (x + s) ds by periodicity
2π −π sin 21 s ³ ´ ³ ´
1 Zπ sin m + 12 t sin m + 21 t
= [f (x + t) + f (x − t)] dt as is even.
2π 0 sin 12 t sin 21 t

7
³ ´
1 X m sin m + 21 x
Since + cos nx = 1 ,
2 n=1³ ´
2 sin 2
x
Z π sin m + 1 x
1 2
π= dx
2 0 2 sin 21 x ³ ´
1 1 Zπ sin m + 21 t
Therefore [f (x + 0) + f (x − 0)] = [f (x + 0) + f (x − 0)]
2 2π 0 sin 12 t
1
Therefore Sm (x) − [f (x + 0) + f (x − 0)]
2 ³ ´
1 Z π sin m + 12 t
= [f (x + t) − f (x + 0)] dt
2π 0 sin³ 12 t ´
1 Zπ sin m + 12 t
+ [f (x − t) − f (x − 0)] dt (1)
2π 0 sin 21 t
When f (x + 0) = f (x − 0) = f (x) (1) becomes ³ ´
1 Z π sin m + 21 t
Sm (x) − f (x) = [f (x + t) + f (x − t) − 2f (x)] dt (1a)
2π 0 sin 12 t
The integrals appearing in (1) and (1a) are all of the form
Z b
1
φ(t) sin λtdt where a = 0, b = π, λ = m + 2
a
f (x + t) − f (x + 0) f (x − t) − f (x − 0)
φ(t) = , , or
sin 12 t sin 12 t
f (x + t) + f (x − t) − 2f (x)
sin 12 t
Hence if φ(t) is bounded and integrable over [0, π], then by the Riemann
Z π
Lebesgue theorem, φ(t) sin λt → 0 as λ → ∞.
0
In fact in the above cases φ(t) is bounded and integrable over [h, π] h > 0
and so the convergence depends only on the behaviour of the function in a
sufficiently small interval [0, h].

Integration of a Fourier Series Z π


1
µ ¶
If f (x) is bounded and integrable in [−π, π] and F (x) = f (t) − a0 dt
−π 2
Where 12 a0 = f¯ is the constant term in the Fourier series for f , then F (x)
has a Fourier series, convergent everywhere to F (x), obtained by integrating
the Fourier series for f (x) − 12 a0 term by term. [This holds even if the Fourier
series for f does not converge.]

8
F (x) is an absolutely continuous function and hence possesses a Fourier series
converging everywhere to F (x).

1 X
F (x) = A0 + (An cos nx + Bn sin nx)
2 n=1
Assuming that f is continuous on (−π, π) ensures the existence of F 0 (x), and
1 Z π
An = F (x) cos nxdx n = 1, 2, · · ·
π ·π ¸π
1 sin nx 11 Z π
= F (x) − F 0 (x) sin nxdx
π nµ −π n π¶ −π
1 Zπ 1
=0− f (x) − a0 sin nxdx [F (π) = F (−π) = 0]

Z π −π 2
1 bn
=− f (x) sin nxdx = −
nπ −π n
an
Similarly Bn =
n ∞
1 X bn an
Therefore F (x) = A0 + − cos nx + sin nx
2 n=1 n n

1 bn
(−1)n
X
Putting x = π gives (a0 ) =
2 n=1 n
X∞
an sin nx + bn ((−1)n − cos nx)
Therefore F (x) =
n=1 n
∞ Z
X x
= {an cos nt + bn sin nt}dt
n=1 −π

Differentiation of a Fourier Series


This is not always valid.

X sin nx 1
e.g. = (π − x) 0 ≤ x ≤ 2π
n=1 n 2
∞ ∞
X d sin nx X
= cos nx which does not converge.
n=1 dx n n=1

Sufficient Conditions
If f (x) is continuous and f 0 (x) exists except at a finite number of points,
and both f (x) and f 0 (x) have Fourier series which converge, then the series
for f 0 (x) is obtained by term by term differentiation of the Fourier series for
f (x).

1 X
i.e. f (x) = a0 + an cos nx + bn sin nx
2 n=1

9

1 X
⇒ [f 0 (x + 0) + f 0 (x − 0)] = nbn cos nx − nan sin nx
2 n=1
[This is really just the same as the result for integration, with slightly weaker
conditions.]

Half-Range Series
Let f (x) be bounded and integrable in [0, π]

(1) Cosine Series


(
f (x) 0≤x≤π
define fc (x) =
f (−x) −π ≤ x ≤ 0
Then fc (x) is an even function, which has a Fourier series in which
bn ≡ 0
1Zπ
an = fc (x) cos nxdx
π −π
2Zπ 2Zπ
= fc (x) cos nxdx = f (x) cos nxdx
π 0 π 0
(2) Sine Series
(
f (x) 0<x<π
define fs (x) =
−f (−x) −π < x < 0
If f (0) 6= 0 fs is discontinuous at 0.
If f (π) 6= 0 fs is discontinuous at π.
Then fs (x) is an odd function, and has a Fourier series in which an ≡ 0
1Zπ
bn = fs (x) sin nxdx
π −π
2Zπ 2Zπ
= fs (x) sin nxdx = f (x) sin nxdx
π 0 π 0

Order of magnitude of Fourier coefficients


1Zπ
an − ibn = f (x)e−inx dx = cn (1)
π −π
Suppose f (x) and all its derivatives are bounded and continuous in
(−π, α1 ), (α1 , αZ2 ) · · · (αk , π)
1 π (m)
Write cm
n = f (x)e−inx dx
π −π

10
k Z αr+1
1X
= f (m) (x)e−inx dx (2)
π 0 −αr
Integrating (1) by parts gives#
k
" α
0 f (x) −inx r+1 1 Z αr+1 0
f (x)einx dx
X
πcn = πcn = − e +
in in α r
" k r=0 αr #
1 X −inαr −inαr+1
Z αr+1
0 −inx
= f (αr + 0)e − f (αr+1 − 0)e + f (x)e dx
in r=0 αr
1
·
= f (−π + 0)e−inπ − f (π − 0)e−inπ
in
k
#
{f (αr + 0) − f (αr − 0)}e−inαr + πc(1)
X
+ n
r=1
f (−π + 0) = f (π + 0) by periodicity
Therefore f (−π + 0)einπ − f (π − 0)e−inπ
= [f (π + 0) − f (π − 0)]e−inπ = [f (αk+1 + 0) − f (αk+1 − 0)]einαk+1
Hence we have
πc(1) 1 k+1
πc(0) n
X
n = + {f (αr + 0) − f (αr − 0)}e−inαr
ni ni r=1
1 k+1
Write Jn(m) = {f (m) (αr + 0) − f (m) (αr − 0)}e−inαr
X
π r=1
c(1) Jn(0)
Therefore c(0)n = n
+
ni ni
(2) (1)
c J
Similarly c(1)
n = n
+ n
ni ni
(0)
J Jn(1) c(2)
Therefore c(0)n = n
+ 2
+ n
2
ni (ni) (ni)
1 Z π
Since c(2)
n = f 00 (x)e−inx dx is bounded,
π −π ³ ´
1
if Jn(0) = 0 for all n ≥ m then cn = c(0) n is O n2 as n → ∞
³ ´
1
If also Jn(1) = 0 for all n ≥ m then cn = c(0) n is O n3 as n → ∞
(1) (r)
In particular
³ ´if f, f , · · · f are continuous but f (r+1) is not continuous then
1
cn = O nr+2 as n → ∞
(0)
In fact Jn vanishes only if f is continuous for if we write
k+1
f (αr + 0) − f (αr − 0) = jr then πJn(0) =
X
jr e−inαr .
r=1
If Jn(0) = 0 for n ≥ m then

11
k+1
X
jr e−inαr = 0 n = m, m + 1, · · ·
r=1
−iαr
Taking n =m, m + 1, · · · , m + k, we write  e = zr
m m m
z1 z2 · · · zk+1 j1
.
.
 . 
  ..  = 0

Therefore  .    

z1m+k z2m+k · · · zk+1 m+k


jk+1
The determinant ¯ of the matrix ¯is
¯ 1 ··· 1 ¯¯
¯
¯ z1 · · · zk+1 ¯
¯ ¯
(z1 z2 · · · zk+1 )m ¯¯ .. ¯ = (z1 z2 · · · zk+1 )m
Y
(zr − zs )
¯ .
¯
¯ k
¯ r>s
k
¯
¯ z
1 · · · zk+1 ¯
zr − zs 6= 0 for r 6= s
Therefore the determinant is non zero.
Therefore j1 = j2 = · · · = jk+1 = 0
Therefore f is continuous.

Parseval’s Theorem
If f (x) is bounded and integrable in (−π, π)

1Zπ 1
(f (x))2 dx = a20 + (a2n + b2n ).
X
then
π −π 2 n=1
[Note that this is true even though f (x) does not equal the sum of its Fourier
series.]
If we assume that f (x) is continuous
" and the Fourier series #converges to f (x),

1Zπ 2 1Zπ 1 X
f (x) = f (x) a0 + (an cos nx + bn sin nx) dx
n −π π −π 2 n=1
Since uniformity of convergence is not affected by multiplying by f (x) we
can integrate term by term

1 1Zπ 1X Z π Z π
RHS= a0 f (x)dx + an f (x) cos nxdx + bn f (x) sin nxdx
2 π −π π n=1 −π −π

1
= a20 + a2n + b2n
X
2 n=1

12

You might also like