Orthogonal Basis
Orthogonal Basis
Orthogonal Basis
12.1:Orthogonal Functions
n
X
ui vi .
i=1
Orthogonal collections
ui
(v, ui )
v=
.
2
ku
k
i
i=1
Orthogonal Functions
f1 (x) f2 (x) dx
a
i (x) j (x) dx = 0,
a
when i 6= j.
An example
cos(A + B) + cos(A B)
.
2
1 sin((m + n)x) sin((m n)x)
=
+
=0
2
m+n
mn
Orthonormality
!1/2
f (x)2 dx
a
An example
R 2
2
We have k1k = 1 dx = 2 , so k1k = 2.
For m 1, we have
k cos(mx)k2 =
Z
cos2 (mx) dx =
x sin(2mx)
+
=
2
4m
Thus k cos(mx)k = .
1 + cos(2mx)
dx
2
= .
cos(2x) cos(3x)
x,
The collection { 12 , cos
, , . . . } is thus orthonormal
on [, ].
(1)
(f, n ) =
f (x) n (x) dx =
ck k (x) n (x) dx
a
k=0
ck
k (x) n (x) dx = cn
a
k=0
2n (x) dx = cn kn k2 .
Orthogonal systems
(f, n )
,
2
kn k
n0 .
X
(f, n )
f (x) =
n (x).
2
k
k
n
n=0
(2)
However, the expansion formula (2) does not hold in general for an
arbitrary orthogonal system on [a, b]. For example, it could happen
that f 6= 0 but f (x) is orthogonal to each function n (x) in the
system and thus the RHS of (2) would be 0 in that case while
f (x) 6= 0.
Completeness
X
(f, n )
f (x)
(3)
n (x).
2
kn k
n=0
on [a, b] in the L2 -sense which means that
lim
N
X
2
(f, n )
f (x)
n (x) dx = 0.
2
kn k
n=0
Some remarks
sin(mx) sin(nx) dx = 0,
m, n 1, m 6= n, (b)
cos(mx) sin(nx) dx = 0,
m 0, n 1. (c)
sin((m n)x) sin((m + n)x)
=
= 0.
mn
m+n
k cos(mx)k2 = ,
k sin(nx)k2 = .
a0 X
f (x)
+
{an cos(nx) + bn sin(nx)} (4),
2
n=1
where
a0
2
(f,1)
k1k2
1
2
f (x) dx = average of f on [, ],
1
(f, cos(nx))
=
an =
k cos(nx)k2
(f, sin(nx))
1
bn =
=
k sin(nx)k2
More generally, if p > 0 and f (x) is pwc on [p, p], then it will have
a Fourier series expansion on [p, p] given by
X
nx
nx
a0
(4),
+
an cos
f (x)
+ bn sin
2
p
p
n=1
where the Fourier coefficients an , bn are defined by
Z p
nx
1
f (x) cos
dx, n 0,
an =
p p
p
1
bn =
p
f (x) sin
p
nx
p
dx n 1.
An example
The function
f (x) =
0,
x,
< x 0
0<x<
sin(nx)
cos(nx) +
f (x) +
4
n2
n
n1
n odd
n1
2
2
2
cos(x)
cos(3x)
cos(5x) + . . .
4
9
25
1
1
1
sin(x) sin(2x) + sin(3x) sin(4x) + . . . ()
2
3
4
Periodic extension
Since
cos
sin
n(x + 2p)
p
n(x + 2p)
p
= cos
= sin
nx
+ 2n
p
nx
+ 2n
p
= cos
= sin
nx
p
nx
p
Piecewise continuity
xai
lim f (x) = f (a
i+1 )
xai+1
Pointwise convergence
Note that, in the theory of Fourier series, if f (x) is pwc, the value of
the function f (x) at the end points ai where f (x) is discontinuous
is unimportant (as they do not affect the integral to compute the
Fourier coefficients of f (x)).
f (x),
if f (x) is continuous at x,
f (x+ )+f (x )
,
2
f (x) sin
p
nx
p
dx = 0, n 1.
X
nx
a0
+
x [p, p],
an cos
f (x)
(6)
2
p
n=1
where an =
2
p
f (x) cos
0
nx
p
dx, n 0 .
X
nx
x [p, p],
f (x)
(7)
bn sin
p
n=1
2
where bn =
p
f (x) sin
0
nx
p
dx, n 1 .
It is given explicitly by
n
2 X
sin(nx),
(1)n+1 2
f (x)
n=1
n 1/4
x [, ].
1.0
0.5
15
10
K K
K
5
0.5
1.0
10
15
Suppose that f (x) is defined on the interval [0, p]. Then, f (x) can
be expanded in a Fourier series in several ways.
where, for n 0,
Z
Z
2 p
nx
nx
2 p
an =
dx =
dx
f (x) cos
fe (x) cos
p 0
p
p 0
p
In particular, since fe (x) = f (x) for 0 x p,
a0 X
f (x)
+
an cos
2
n=1
nx
p
x [0, p].
We can also consider the odd extension, fo (x), of f (x) on [p, p],
defined by
(
f (x),
0 < x < p,
fo (x) =
f (x), p < x < 0,
and compute its 2p-periodic sine Fourier series expansion. The
coefficients can be computed directly in terms of the original
function f (x).
P
for x [p, p], where,
We have fo (x) n=1 bn sin nx
p
bn =
2
p
f (x) sin
0
nx
p
dx =
2
p
fo (x) sin
0
nx
p
dx ,
n=1
bn sin
nx
p
x [0, p].
where
2nx
p
+ bn sin
f (x) cos
2nx
p
dx, n 0,
f (x) sin
2nx
p
dx, n 1.
a0 X
f (x)
+
an cos
2
n=1
p
2
an =
p
2
bn =
p
0
p
0
2nx
p
x [0, p].
a0
+
an cos
2
n=1
nx
p
+ bn sin
nx
p
x [p, p],
inx
p
inx
p
inx
p
inx
p
a0
+e
e
e
e
f (x)
+ bn
+
an
2
2
2i
n=1
X
X
inx
inx
an i bn
a0
an + i bn
=
+
e p +
e p
2
2
2
n=1
n=1
= c0 +
n=1
cn e
inx
p
n=1
inx
p
cn e
an i bn
cn =
2
Z p
Z p
1
nx
nx
1
dx i
dx
f (x) cos
f (x) sin
=
2p p
p
2p p
p
Z p
inx
1
f (x) e p dx, n 1,
=
2p p
cn
1
an + i bn
=
=
2
2p
f (x) e
p
inx
p
dx,
n 1.
cn e
f (x) e
inx
p
f (x)
inx
p
nZ
where
1
cn =
2p
p
p
dx,
nZ .