Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abdollah Jalilian
www.razi.ac.ir/jalilian/
Department of Statistics
Razi University
Spring 2012
Spring 2012
1 / 42
Course Outline
Fourier analysis
I
I
Fourier series
Fourier transform
Stochastic processes
I
I
I
Poisson process
Markov chains and Birth and Death process
Renewal theory
Queuing theory
I
I
Queuing system
Queuing models
Spring 2012
2 / 42
Evaluation:
Midterm Exam 40%
1391/01/29
Spring 2012
3 / 42
Fourier Analysis
Periodic phenomena mathematical modeling by periodic functions
(periodic signals)
Fourier analysis : approximating complicated periodic functions with series
of simple periodic functions (sines and cosines)
Periodic function :
f : R R is a periodic function if for some p > 0
f (x + p) = f (x)
p is called a period of f
for any n = 1, 2, . . ., np is also a period of f ; i.e.
f (x + np) = f (x)
Spring 2012
4 / 42
Theorem
The above functions are orthogonal on [, ];
Z
cos(nx) cos(mx)dx =
0
Z
sin(nx) sin(mx)dx =
0
Z
cos(nx) sin(mx)dx = 0,
i.e.
m=n
,
n 6= m
m=n
,
n 6= m
Proof?
Jalilian (Department of Statistics)
Spring 2012
5 / 42
Fourier Series
Let f be a piecewise continuous function of period p = 2, then
f (x) = a0 +
an cos(nx) + bn sin(nx)
n=1
where
Z
1
f (x)dx
2
Z
1
an =
f (x) cos(nx)dx n = 1, 2, . . .
Z
1
bn =
f (x) sin(nx)dx n = 1, 2, . . .
a0 =
The above series is called the Fourier series of f with Fourier coefficients
a0 , a1 , b1 , a2 , b2 , . . ..
Jalilian (Department of Statistics)
Spring 2012
6 / 42
Fourier Series
Does the Fourier series converges for any x? Is the sum equal to f (x)?
Partial sums of the Fourier series of a function
Sn (x) = a0 +
n
X
aj cos(jx) + bj sin(jx)
n = 1, 2, . . .
j=1
Convergence means
a0 +
an cos(nx) + bn sin(nx) = lim Sn (x)
n
n=1
Theorem
The Fourier series converges if f has a left-hand and a right-hand
derivative at each point in [, ]. Its sum is f (x) except at points x0
where f is discontinuous. At x0 , the sum of the series is the average of the
left-hand and the right-hand limits of f at x0 .
Jalilian (Department of Statistics)
Spring 2012
7 / 42
Fourier Series
Example: Let k > 0 be a constant and
k
0<x <
f (x) =
k < x < 0
f (x + 2) = f (x)
n = 1, 2, . . .
4k
n
n = 1, 3, 5, . . .
n = 2, 4, 6, . . .
i
4k h
1
1
sin(x) + sin(3x) + sin(5x) +
3
5
Sn (0) = Sn () = Sn () = 0 n = 1, 2, . . .
Jalilian (Department of Statistics)
Spring 2012
8 / 42
Exercises
f (x) = |x|
f (x) = x2
f (x) =
x
x
< x < 0
0<x <
Spring 2012
9 / 42
X
n=1
an cos(
n
n
x) + bn sin( x)
L
L
Spring 2012
10 / 42
1<x <2
0
k 1 < x < 1
f (x) =
0 2 < x < 1
p = 4, L = 2
k
2
2k
n
2k
n
n = 1, 5, 9, . . .
n = 3, 7, 11, . . .
n = 2, 4, 6, . . .
bn = 0 n = 1, 2, . . .
f (x) =
i
k
2k h
1
3
1
5
+
cos( x) cos( x) + cos( x) +
2
2
3
2
5
2
Spring 2012
11 / 42
p=
,L =
a0 =
an =
2E
(n1)(n+1)
bn =
u(t) =
0
E
2
n = 1, 3, 5, . . .
n = 2, 4, 6, . . .
n=1
n = 2, 3, 4, . . .
i
E
E
2E h 1
1
+ sin(t)
cos(2t) +
cos(4t) +
2
1.3
3.5
Spring 2012
12 / 42
an cos(
n=1
a0 =
1
L
f (x)dx
an =
2
L
n
x)
L
f (x) cos(
0
n
x)dx
L
n = 1, 2, . . .
bn sin(
n=1
bn =
Jalilian (Department of Statistics)
2
L
f (x) sin(
0
n
x)
L
n
x)dx
L
n = 1, 2, . . .
Spring 2012
13 / 42
Theorem
Let f and g be functions of period p = 2L. The Fourier coefficients of
c1 f + c2 g are
c1 a1n + c2 a2n
n = 0, 1, 2, . . . ,
c1 b1n + c2 b2n
n = 1, 2, . . .
where a1n and b1n are Fourier coefficients of f and a2n and b2n are Fourier
coefficients of g .
Spring 2012
14 / 42
N
X
An cos(nx) + Bn sin(nx)
i=1
Spring 2012
15 / 42
h =f +g
h = af , a R
Subspace: any subset of a vector space that is closed under addition and
scalar multiplication is also a vector space.
subspace of all continuous functions on [a, b]
subspace of all polynomial functions on [a, b]
subspace of all functions of period p = 2L on [a, b]
A basis of an infinite dimensional space?
Linear combinations of elements of a basis?
Jalilian (Department of Statistics)
Spring 2012
16 / 42
f 2 (x)r (x)dx 0
Orthogonality: hf , g i = 0
Triangle inequality: kf + g k kf k + kg k
(kf g k2 = kf k2 + kg k2 2hf , g i)
Convergence: fn f means kfn f k 0 as n
Jalilian (Department of Statistics)
Spring 2012
17 / 42
orthonormal sequence
b
Z
hym , yn i =
ym (x)yn (x)r (x)dx = mn =
1 m=n
0 m=
6 n
n a
where
Sn (x) =
n
X
am ym (x)
m=0
Jalilian (Department of Statistics)
Spring 2012
18 / 42
Theorem
Let {y0 , y1 , y2 , . . .} be a complete orthonormal set of functions on [a, b] in
a subspace S. If there is a f S where hf , ym i = 0 for every
m = 0, 1, 2, . . . then kf k = 0.
Spring 2012
19 / 42
1
m=0
m+1
ym (x) =
cos( 2 x) m = 1, 3, 5, . . .
m = 2, 4, 6, . . .
sin( m2 x)
Then y0 , y1 ,
y2 , . . . are orthogonal: hym , yn i = 0 m 6= n
am ym (x) = a0 +
m=0
X
am cos(mx) + bm sin(mx)
m=1
Spring 2012
20 / 42
m=1
Spring 2012
21 / 42
b2c
i
1 dn h 2
1 X
2n 2j n2j
n
j n
(x 1) = n
Pn (x) = n
(1)
x
2 n! dx n
2
j
n
j=0
m=1
with
2m + 1
am =
2
Jalilian (Department of Statistics)
f (x)Pm (x)dx
1
Spring 2012
22 / 42
m=0
hence
2
kf Sk k = kf k + kSk k 2hf , Sk i = kf k
k
X
2
am
m=0
X
2
am
= kf k2 =
f 2 (x)r (x)dx
m=0
Jalilian (Department of Statistics)
a
Advanced Mathematics for CE
Spring 2012
23 / 42
Fourier Integral
A nonperiodic function on R = (, ) is a function of period p =
Let fL be a function of period p = 2L and define wn = n
L .
1
fL (x) =
2L
Z L
1X
fL (x) cos(wn x)dx cos(wn x)
fL (x)dx +
L
L
n=1
Z L
+
fL (x) sin(wn x)dx sin(wn x) wn
If
|f (x)|dx
1
lim fL (x) =
L
Z
f (x) cos(wx)dx cos(wx)
f (x) sin(wx)dx sin(wx) dw
Z
Spring 2012
24 / 42
Fourier Integral
Theorem
Let f be a pciewise continuous function on every finite interval and has a
right- and a left-hand derivative at every point. If the integral
Z
|f (x)|dx
exists, then
Z
f (x) =
A(w ) cos(wx) + B(w ) sin(wx) dw
1
A(w ) =
f (v ) cos(wv )dv
1
B(w ) =
f (v ) sin(wv )dv
Spring 2012
25 / 42
2
A(w ) =
f (x) cos(wx)dx
0
2
B(w ) =
f (x) sin(wx)dx
0
Spring 2012
26 / 42
Fourier Integral
Example: Find the Fourier integral representation of the function
1 |x| < 1
f (x) =
0 |x| > 1
2
A(w ) =
2 cos(wv ) 1
2 sin(w )
f (v ) cos(wv )dv =
=
w
w
0
Z
2 sin(w ) cos(wx)
dw
0
w
Z
1 sin(w (1 + x)) + sin(w (1 x))
=
dw
0
w
f (x) =
Spring 2012
27 / 42
Fourier Integral
Example: Find the Fourier integral representation of the function
f (x) = e kx
A(w ) =
k > 0, x > 0
f (v ) cos(wv )dv
0
h
i
k
w
2
kv
e
cos(wv
)
sin(wv
)
k2 + w 2
k
0
2k
=
(k 2 + w 2 )
2k
f (x) =
Z
0
cos(wx)
dw
k2 + w2
Z
0
cos(wx)
kx
dw =
e
k2 + w 2
2k
Spring 2012
28 / 42
Fourier Transform
Transforming a function: to obtain a new function
An integral transformation: transforming by integrating; e.g. Laplace
transform
Let f (x) be an even function. The Fourier cosine transform of f is
r Z
2
Fc (f )(w ) = fc (w ) =
f (x) cos(wx)dx
0
with the inverse Fourier cosine transform
r Z
2
1
Fc (fc )(x) = f (x) =
fc (w ) cos(wx)dw
0
Let f (x) be an odd function. The Fourier sine transform of f is
r Z
2
Fs (f )(w ) = fs (w ) =
f (x) sin(wx)dx
0
with the inverse Fourier cosine transform
r Z
2
1
fs (w ) sin(wx)dw
Fs (fs )(x) = f (x) =
0
Jalilian (Department of Statistics)
Spring 2012
29 / 42
2 k sin(aw )
w
r
2 k 1 cos(aw )
Fs (f )(w ) = fs (w ) =
Fc (f )(w ) = fc (w ) =
2 1
1 + w2
r
2 w
Fs (g )(w ) = gs (w ) =
1 + w2
Fc (g )(w ) = gc (w ) =
Spring 2012
30 / 42
Fs (f 0 )(w ) = w Fc (f )(w )
Why?
Jalilian (Department of Statistics)
Spring 2012
31 / 42
a>0
Fc (f 00 )(w ) = a2 Fc (f )(w )
but
r
00
2 0
f (0)
r
2
= w 2 Fc (f )(w ) + a
Fc (f )(w ) = w Fs (f )(w )
r
2
w Fc (f )(w ) + a
2
= a2 Fc (f )(w )
Fc (f )(w ) =
2
a
2
a + w2
Spring 2012
32 / 42
i
A(w ) cos(wx) + B(w ) sin(wx) dw
0
Z Z h
i
1
cos(wv ) cos(wx) + sin(wv ) sin(wx) f (v )dv dw
0
Z Z
1
cos(wx wv )f (v )dv dw
0
Z Z
1
cos(w (x v ))f (v )dv dw
2
Z Z
1
cos(w (x v )) + i sin(w (x v )) f (v )dv dw
2
Z Z
1
iw (xv )
e
f (v )dv dw
2
Z
Z
1
1
iwv
e
f (v )dv e iwx dw
2
2
Spring 2012
33 / 42
e iwx f (x)dx
1
(f)(x) = f (x) =
2
Theorem
If f is absolutely integrable and piecewise continuous on every finite
interval, then its Fourier transform exists.
Spring 2012
34 / 42
a>0
sin(w )
2 w
1
F(g )(w ) =
2(a + iw )
F(f )(w ) =
Spring 2012
35 / 42
(f g )(x) =
f (t)g (x t)dt =
f (x t)g (t)dt
Why?
Jalilian (Department of Statistics)
Spring 2012
36 / 42
2k
N
k = 0, 1, . . . , N 1
N1
X
cn e inxk
k = 0, 1, 2 . . . , N 1
n=0
To find cn s
N1
X
fk e
imxk
k=0
N1
X N1
X
k=0 n=0
N1
X N1
X
cn
n=0
Jalilian (Department of Statistics)
cn e i(nm)xk
e i(nm)
2k
N
k=0
Spring 2012
37 / 42
cn
N1
X
i(nm) 2k
N
N1
X
cn
n=0
k=0
= Ncm +
N1
Xh
k=0
N1
X
i(nm) 2
N
cn
n6=m
= Ncm +
N1
X
N1
Xh
ik
e i(nm) N
ik
k=0
h
i
2 N
1 e i(nm) N
cn
n6=m
1 e i(nm) N
= Ncm
Thus
N1
1 X
cn =
fk e inxk
N
n = 0, 1, . . . , N 1
k=0
Spring 2012
38 / 42
N1
X
fk e inxk
n = 0, 1, . . . , N 1
k=0
f = (f0 , f1 , . . . , fN1 )T ,
f = (f0 , f1 , . . . , fN1 )T ,
enk = e inxk = e in
Then
f0
f1
f2
..
.
fN1
2k
N
= w nk
w0
w0
w0
..
.
w0
w1
w2
..
.
w0
w2
w4
..
.
w0
w N1
w 2(N1)
N1
FN = enk n,k=0
w = wN = e
w0
w N1
w 2(N1)
..
.
w (N1)
2i
N
f0
f1
f2
..
.
fN1
f = FN f
Jalilian (Department of Statistics)
Spring 2012
39 / 42
Inverse of DFT
Since
f = FN f
Then
f = FN1f
FN1 =?
N1
Complex conjugate of FN : FN = N1 w
nk n,k=0
FN FN = FN FN = NI
FN1 =
1
FN
N
Spring 2012
40 / 42
0
14
1 1
1
1
1 i 1 i 1 4 + 8i
1 1 1 1 4 = 6
9
4 8i
1 i 1 i
1 1
1
1
14
4 + 8i
1
1
i
1
i
4 1 1 1 1 6
1 i 1 i
4 8i
0
1
=
4
9
Spring 2012
41 / 42
Since
k=0
h 2i i2
2i
wN2 = e N
= e M = wM
fn =
Split f =
k=0
M1
X
kn
wM
f2k + wNn
k=0
(f0 , f1 , . . . , fN1 )T into
kn
wM
f2k+1
k=0
,
,
M1
X
Spring 2012
42 / 42