Sie sind auf Seite 1von 30

Orthogonal Polynomials

Indre Skripkauskaite
F10GP Project
supervised by
Dr. M. Dreher
March 31, 2016

CONTENTS

Contents
1 Introduction

2 Hilbert Spaces and Self-Adjoint Operators

2.1

Hermitian Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

The Space L2 [a, b] and Differential Operators . . . . . . . . . . . . . . . . . . . .

3 Legendre Polynomials

10

3.1

Generating Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

3.2

Recurrence Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

3.3

The Differential Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

3.4

Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17

4 Chebyshevs Polynomials

20

4.1

Generating Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

4.2

The Differential Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

4.3

Recurrence Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

4.4

Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

5 Conclusion

29

Introduction

In mathematics, a set of polynomials is said to be orthogonal under some inner product if any
two of the polynomials from the given set are orthogonal, i.e. their scalar product equals zero.
There are quite a few families of Orthogonal polynomials, but in this project we will be focusing
only on Legendre and Chebyshevs polynomials.
Legendre polynomials are widely used in physics and Chebyshevs polynomials are applicable
in finance. However, by the use of our knowledge of various topics from Complex Analysis,
Functional Analysis, Linear Algebra as well as Calculus, we will be focusing only on the mathematical, i.e. the theoretical part and understanding the behavior of such families of polynomials
in respective L2 Hilbert spaces, which is the aim of this project.

HILBERT SPACES AND SELF-ADJOINT OPERATORS

Hilbert Spaces and Self-Adjoint Operators

In this section we will be discussing properties of Hilbert spaces and Self-Adjoint Operators,
where the understanding of Hermitian which we will be recalling constantly in later chapters.
We will also mention the famous Fourier Series, which plays an important role in orthogonality
of functions.

2.1

Hermitian Matrices

Definition 2.1.1. Let A be an nn matrix and let AT be its transpose. A is said to be a


Hermitian matrix [2] if it is equal to its conjugate transpose. This means
AT = A.
and we write
AH = A.
Definition 2.1.2. Let u, v Cn and k N. We define a scalar product in C as
hu, vi =

n
X

uk vk .

(2.1)

k=1

Proposition 2.1.3. Let u, v Cn and let A = AH . Then A is said to be self-adjoint if we can


write hAu, vi = hu, Avi.
PROOF:
Assume A = AH . We claim that hAu, vi = hu, Avi, for all u, v Cn .
Then we have
hAu, vi =

n
X

(Au)k vk

k=1

n
X

n
X

akj uj vk
j=1

k=1

n
X
j=1
n
X
j=1

uj

n
X
k=1

AH v j

!
akj vk

2.1

Hermitian Matrices

Therefore if A = AH , then

hAu, vi = hu, Avi .

(2.2)


Proposition 2.1.4. Eigenvalues of a Hermitian matrix are real.

PROOF: Let A be a Hermitian matrix. Let be an eigenvalue of A and let u be an eigenvector


of A to the eigenvalue . Suppose C.
By (2.2) we have that
hAu, ui = hu, Aui ,
hAu, ui = hu, ui = hu, ui = kuk2 ,
hu, Aui = hu, ui = hu, ui = kuk2 ,

hence
kuk2 = kuk2 .

Note that kuk2 > 0 since u 6= 0, therefore = R.

Theorem 2.1.5. Let A = AH , , be eigenvalues of A, 6= . Let u, v Cn be eigenvectors


of A to the eigenvalues , , hence
Au = u
Av = v,

then hu, vi = 0.

PROOF: Consider hAu, vi. From (2.2) we already know that

HILBERT SPACES AND SELF-ADJOINT OPERATORS

hAu, vi = hu, Avi


hu, vi = hu, vi
hu, vi = hu, vi
= hu, vi
hu, vi = hu, vi
( ) hu, vi = 0
hu, vi = 0

since 6= .
Hence the eigenvectors of distinct eigenvalues of a hermitian matrix are orthogonal.
Theorem 2.1.6 (Spectral Theorem). [1] Let A be a Hermitian matrix and let V be a finite
dimensional inner product space. Then there exists an orthonormal basis of V consisting of
eigenvectors of A.

2.2

The Space L2 [a, b] and Differential Operators

Definition 2.2.1. Let a, b R and let f (x),g(x) be two functions in L2 [a, b]. Then the scalar
product in L2 [a, b] [2] is defined as

Z
hf, gi =

f (x)g(x)dx.

(2.3)

We can also set A to act as a differential operator.


Example 2.2.2. Let A be a differential operator A :=
gn (x) L2 [, ], n N, then we get Agn = n2 gn .

By the use of (2.3) and integration by parts

d2
dx2

and gn (x) := sin(nx),

2.2

The Space L2 [a, b] and Differential Operators

gn00 (x)gm (x)dx

hAgn , gm i =

0
gn0 (x)gm
(x)dx

gn0 (x)gm (x)



00
0
gn (x)gm (x)dx gn (x)gm (x)

= hgn , Agm i ,
hAgn , gm i = hgn , Agm i
(2.4)
for n, m N.
Remark: If compare this result to the Proposition 2.1.3 in the previous section, we can see
that in this example A acts as a differential operator rather than being a Hermitian matrix and
gn , gm are being treated as eigenfunctions rather than being eigenvectors of A.
Continuing from 2.4 we have that


n gn , gm = gn , m2 gm ,
n2 hgn , gm i = m2 hgn , gm i ,
(n2 m2 ) hgn , gm i = 0,
hgn , gm i = 0
(2.5)
for n 6= m.

Definition 2.2.3. The set of functions {f1 (x), f2 (x), . . .} is orthogonal in L2 [a, b] [2] if:
hfn (x), fm (x)i = 0

(2.6)

for n 6= m.
Proposition 2.2.4. The set {sin(nx), cos(mx)}, n N+ , m N0 } is an orthogonal set of
functions [2] in the Hilbert space L2 [, ].

PROOF:
By use of (2.3) we can see that

HILBERT SPACES AND SELF-ADJOINT OPERATORS


sin(nx)
cos(nx)dx =
h1, cos(nx)i =
= 0,
n


Z
cos(nx)
sin(nx)dx =
h1, sin(nx)i =
= 0.
n
Z

(2.7)
By (2.5)
hsin(nx), sin(mx)i = 0

(2.8)

hcos(nx), cos(mx)i = 0

f or n 6= m.

Note that cos(x) is an even function and sin(x) is an odd function.


Hence
Z

hcos(nx), sin(mx)i =

cos(nx) sin(mx)dx = 0


Theorem 2.2.5. Let n N and let f(t) be 2L-periodic function in Hilbert space L2 [L, L].
Then the Fourier Series expansion [4] of f(t) is represented as

f (t) = a0 +


an cos

n=1

nt
L


+


bn sin

n=1

nt
L


,

(2.9)

where
L


nt
f (t) cos
dt,
L
L


Z
1 L
nt
bn =
dt
f (t) sin
L L
L
Z L
1
a0 =
f (t)dt.
2L L

1
an =
L

Remark: We can make a comparison of the above to the Theorem 2.1.5, i.e. we can say that


nt
cos nt
are the eigen functions to the eigen values an , bn , a0 in the Hilbert space
L , sin L
L2 [L, L]. Although, we do not claim that we can span full L2 [L, L] space, since it requires
deeper analysis, so we will not be focusing on that in this project.

2.2

The Space L2 [a, b] and Differential Operators

Example 2.2.6 (Heaviside step function).

1
f (t) =

for

< t < 0

for

0 < t < .

(2.10)

Our function f(t) is 2 periodic and it can be approximated by the use of (2.9) as

f (t) =

4c
4c
4c
4c
sin(t) +
sin(3t) +
sin(5t) + +
sin((2n + 1)t) + . . .

3
5
(2n + 1)

(2.11)

n N.

We can now compare the original function with the approximated result where the blue line
indicates the Heaviside step function f(t) and the red line indicated the Fourier approximation
with n = 6.

Figure 1: Fourier series approximation of the Heaviside step function (2.10) with n = 6

10

3
3.1

LEGENDRE POLYNOMIALS

Legendre Polynomials
Generating Function

Legendre Polynomials were introduced by Legendre in the theory of potential, where they are
related to the expansion of the reciprocal of the distance

1
R,

where R is the distance between two

points r and r0 .

Figure 2:
1

R = |r r0 | = (r2 + r02 2rr0 cos ) 2 ,


where is the angle between r and r0 . If we let t =

r0
r ,x

= cos() we have that

1
1
1
= (1 2xt + t2 ) 2
R
r

(3.1)

for 1 6 x 6 1. If we rewrite
1

(1 2xt + t2 ) 2
as
(t x

p
p
1
1
x2 1) 2 (t x + x2 1) 2

(3.2)

and treat it as a function of t, we see that (3.2) has two singularities at


x

p
x2 1,

for 1 6 x 6 1. Therefore, if |t| min |x

x2 1|, we have the following Taylor Series

expansion:

(1 2xt + t2 ) 2 =

Pn (x)tn .

n=0

t C, is is called the generating function [3] for some function Pn (x) which will be discussed
now.

3.1

Generating Function

11

Definition 3.1.1. The Legendre polynomials are defined by Rodrigues formula [5]
Pn (x) =

1 dn 2
(x 1)n ,
2n n! dxn

n N,

(3.3)

for arbitrary values of x.

Later we will see that these Pn s are the same as in section (3.3).
To obtain the general expression for the nth Legendre polynomial we will use the binomial
expression
(x2 1)n =

n
X
(1k n!) 2n2k
x
.
k!(n k)!

(3.4)

k=0

Substituting (3.4) into (3.3) implies


[n/2]

Pn (x) =

X
k=0

(1)k (2n 2k)!


xn2k .
2n k!(n k!)(n 2k)!

By the use of (3.5) we can see that the first eleven Legendre polynomials [8] are:

P0 (x) = 1,
P1 (x) = x,
1
P2 (x) = (3x2 1),
2
1
P3 (x) = (5x3 3x),
2
1
P4 (x) = (35x4 30x2 + 3),
8
1
P5 (x) = (63x5 70x3 5),
8
1
P6 (x) = (231x6 315x4 + 105x2 5),
16
1
P7 (x) = (429x7 693x5 + 315x3 35x),
16
1
P8 (x) =
(6435x8 12012x6 + 6930x4 1260x2 + 35),
128
1
P9 (x) =
(12155x9 25740x7 + 18018x5 4620x3 + 315x),
128
1
P10 (x) =
(46189x10 109395x8 + 90090x6 30030x4 + 3465x2 63).
256

(3.5)

12

LEGENDRE POLYNOMIALS

Figure 3: Legendre polynomials of degree 0 through 10


We can also approach the generating function via Cauchy Integral Formula in the following way:
Theorem 3.1.2 (Cauchy Integral Formula [5]). Let f (z) be analytic in a simply connected
domain D. Let C be a closed contour going in the counter-clockwise direction inside D, and let
z be an interior point of D. Then
1
2i

n!
2i

Z
C

Z
C

f ()
d = f (z).
z

f ()
dn
d
=
f (z)
( z)n+1
dz n

(3.6)

Proposition 3.1.3. For n N, t C, |x| 6 1, x R, (1 2xt + t2 ) 6= 0, hence |t| < 13 ,


1

g(x, t) = (1 2xt + t2 ) 2

(3.7)

is the generating function for Legendre polynomials.


PROOF: We wish to show that

X
n=0

Pn (x)tn = g(x, t).

(3.8)

3.1

Generating Function

13

Replace Pn (x) by the Rodrigues formulas and use the Cauchy integral theorem (3.6) we obtain
Z
(1 z 2 )n
(1)n dn
(1)n 1
2 n
Pn (x) = n
(1

x
)
=
dz,
(3.9)
2 n! dxn
2n 2i C (z x)n+1

where C is any curve enclosing x, going in a counter-clockwise direction. Now inserting (3.9)
into (3.8) we get

X
n=0

X
(1 z 2 )n
(1)n 1 n
dz.
t
Pn (x)t =
n+1
2n 2i
C (z x)
n

(3.10)

n=0

Now interchanging summation and integration in (3.10) we get

X
n=0

1
Pn (x)t =
2i
n

Z
C

dz X (1)n n
t
zx
2n

n=0

(1 z 2 )
zx

n
.

(3.11)

The resulting series is a simple geometric series, which is readily summed. We then obtain the
following integral:
1

2i

Z
C

2
dz
.
t z 2 2t z 1 2t x

(3.12)

Now we evaluate this integral by using the familiar residue integration technique. We can clearly
see that the denominator of (3.12) has two roots:
r
1
2
1 1p
1

x
+
1
=
+
1 2xt + t2
z1 = +
t
t2
t
t
t
and
1
z2 =
t

1
2
1 1p
x
+
1
=

1 2xt + t2 .
t2
t
t
t

We can rewrite z2 as
z2 =

1
t2
1
t

1
x
t2
1
t2

+1

2t x + 1

=
1
t

qt

x1

1
t2

(3.13)

2t x + 1
.

Now if we multiply both numerator and denominator of (3.13) by t we obtain the following:
z2 =

2x t

x
1 + 1 2xt + t2

(3.14)

14

LEGENDRE POLYNOMIALS

for t 0.
We now choose C to be a path surrounding the point x and z2 and by applying Residue Theorem
we see that

2
lim
t zz2


2t x + 1
1

.
=
2
2
2
z tz 1 tx
1 2xt + t2

1
t

1
t2

Consequently,

Pn (x)tn =

n=0

1
.
1 2xt + t2

(3.15)


3.2

Recurrence Relation

Proposition 3.2.1. Legendre Polynomials satisfy the Recurrence Relation [5]


(n + 1)Pn+1 (x) (2n + 1)xPn (x) + nPn1 (x) = 0

(3.16)

for n N+ .
PROOF:
We are going to show it by first differentiating the generating function (3.7) with respect to t:

X
3
1
xt
g
= (1 + t2 2xt) 2 (2t 2x) =
Pn (x)tn1
3 =
2
t
2
(1 + t 2xt) 2
n=0

Multiply (3.17) by (1 + t2 2xt) to obtain


(1 + t2 2xt)

nPn (x)tn1 = (x t)(1 + t2 2xt) 2 = (x t)

n=0

(1 + t2 2xt)

nPn (x)tn1 = (x t)

n=0

(1 + t2 2xt)

X
n=0

nPn (x)tn1 (x t)

X
n=0

X
n=0

Pn (x)tn = 0.

n=0

Setting the coefficient of tn equal to zero, we find that


(n + 1)Pn+1 (x) 2nxPn (x) + (n 1)Pn1 (x) xPn (x) = 0,

Pn (x)tn ,
Pn (x)tn ,

(3.17)

3.3

The Differential Equation

15

or equivalently
(n + 1)Pn+1 (x) (2n + 1)xPn (x) + nPn1 (x) = 0,

n N.

Legendre polynomials can be calculated step by step, starting from P0 (x) = 1, P1 (x) = x.

3.3

The Differential Equation

Proposition 3.3.1. Legendre Polynomials solve the differential equation [7]


((1 x2 )Pn0 (x))0 + n(n + 1)Pn (x) = 0,

n N0 .

(3.18)

PROOF:
By the use of the generating function (3.15)

1
2t
t
X 0
:
Pn (x)tn =
3 =
3
x
2 (1 2xt + t2 ) 2
(1 2xt + t2 ) 2
n=0

X
1
2x + 2t
xt
:
nPn (x)tn1 =
(3.19)
3 =
3
t
2 (1 2xt + t2 ) 2
(1 2xt + t2 ) 2
n=0
 

1
3
2x + 2t
2 X
n2
:
n(n 1)Pn (x)t
=

3 + (x t)
2
t
2 (1 2xt + t2 ) 52
(1 2xt + t2 ) 2
n=0

1
(1 2xt + t2 )

3
2

3(x t)2
5

(1 2xt + t2 ) 2

(3.20)

X
t(1 x2 )
:
(1 x2 )Pn0 (x)tn =
3
x
1 2xt + t2 ) 2
n=0


 
0
0 X
2xt
3
2t
2
:
(1 x2 )Pn0 tn =
+
t(1

x
)

(1 x2 )
3
5
2
x
2
(1 2xt + t ) 2
(1 2xt + t2 ) 2
n=0

(1 x2 )

2xt
3

(1 2xt + t2 ) 2

3t2 (1 x2 )
5

(1 2xt + t2 ) 2

(3.21)

Now we multiply (3.19) by t and obtain

X
n=0

nPn (x)tn =

t(x t)
3

(1 2xt + t2 ) 2

(3.22)

16

LEGENDRE POLYNOMIALS

If we multiply (3.20) by t2 we get

n(n 1)Pn (x)tn =

n=0

t2
(1 2xt + t2 )

3
2

3t2 (x t)2

(3.23)

(1 2xt + t2 ) 2

Now if we multiply (3.22) by 2 an add it to (3.23) we obtain the following:

2nPn (x)t +

n=0

n(n 1)Pn (x)tn =

n=0

(2n + n(n 1))Pn (x)tn =

n=0

n(n + 1)Pn (x)tn =

n=0

2t(x t)
(1 2xt + t2 )

3
2

2xt 3t2

(1 2xt + t2 ) 2
2xt 3t2
(1 2xt + t2 ) 2

t2
(1 2xt + t2 )

3
2

3t2 (x t)2
5

(1 2xt + t2 ) 2

3t2 (x t)2
5

(1 2xt + t2 ) 2
3t2 (x t)2

(3.24)

(1 2xt + t2 ) 2

Now by adding (3.21) to (3.24) we see that

X
0
(1 x2 )Pn0 (x) tn +
n(n + 1)Pn (x)tn =

n=0

n=0

X
n=0

2xt
3

(1 2xt + t2 ) 2
2xt 3t2

(1 2xt + t2 )

3
2

3t2 (1 x2 )
5

(1 2xt + t2 ) 2
3t2 (x t)2
5

(1 2xt + t2 ) 2


2xt + 2xt 3t2 3t2 ((1 x2 ) + (x t)2 )
+
(1 x2 )Pn0 (x)0 tn + n(n + 1)Pn (x)tn =
3
5
(1 2xt + t2 ) 2
(1 2xt + t2 ) 2
=
=

3t2

(1 2xt + t) 2
3t2 + 3t2
3

(1 2xt + t2 ) 2

3t2 (1 2xt + t2 )
5

(1 2xt + t2 ) 2

= 0.

Hence

0
(1 x2 )Pn0 (x) tn + n(n + 1)Pn (x)tn = 0.

n=0

Now setting the coefficients of tn equal to zero, we finally obtain


0
(1 x2 )Pn0 (x) + n(n + 1)Pn (x) = 0,

n N,


(3.25)

Remark: Let us compare the above result to Theorem 2.1.5, which states the following:
Au = u

(3.26)

3.4

Orthogonality

17

for all u Cn .
We shal prove the following proposition first:

Proposition 3.3.2. A =

d
dx


d
(1 x2 ) dx
is a self-adjoint operator in Hilbert space L2 [1, 1],

i.e. hAu, vi = hu, Avi, for all twice differentiable functions u(x), v(x), x R.
PROOF: Using differentiation by parts we see that
Z 1
dx
(Au) v
hAu, vi =
1 x2
1

Z 1 p
d p
d
1
2
2
=
1x
( 1x
u(x)) v(x)
dx
dx
dx
1 x2
1
Z 1 p
0
1 x2 u0 vdx
=
1

p

1
Z

=0

x2 u0 v

Z
 1


1

p
1 x2 u0 v 0 dx

0
p
1 x2 v 0 dx
u0

Z
 p
 1

= u 1 x2 v 0 +
1

0
p
1 x2 v 0 dx

1
=0+
u (Av)
dx
1 x2
1
= hu, Avi ,

hence A is indeed self-adjoint

If we set = n(n + 1), u = Pn (x) and treat A as a differential operator in L2 , i.e.



d
d
A = dx
(1 x2 ) dx
in (3.26), then comparing to Theorem 2.1.5 we can write (3.25) as


d
2 d
(1 x ) Pn (x) = n(n + 1)Pn (x),
dx
dx

(3.27)

Consequently,
((1 x2 )Pn0 (x))0 = n(n + 1)Pn (x)

(3.28)

which is the self-adjoint form of Legendre Differential Equation.

3.4

Orthogonality

Proposition 3.4.1. Legendre polynomials are orthogonal in the Hilbert space L2 [(1, 1), dt].

18

LEGENDRE POLYNOMIALS

PROOF:
R1

We wish to show

1 Pn (x)Pm (x)dx

= 0 for n 6= m. If we multiply both sides by Pm (x) and take

the scalar product as defined in (2.3) of both sides we get


Z

0
Pm (x) (1 x2 )Pn0 (x) dx

Pn (x)Pm (x) =

n(n + 1)

= Pm (x) (1 x
Z

)Pn0 (x)

 1
+
1


0
Pm
(x) (1 x2 )Pn0 (x) dx

0
0
Pm
(x)(1 x2 ) Pn (x)dx

=
1

Pm (x)Pn (x)dx

= m(m + 1)
1

Pm (x)Pn (x)dx = 0

(3.29)

whenever n 6= m.

Proposition 3.4.2. If n = m, then the scalar product or two Legendre polynomials satisfy

PROOF:

Pn2 (x)dx =

2
.
2n + 1

We substitute (3.15) into L2 ([1, 1]) scalar product by the use of (2.3):
*

Pn (x)t
n=0
X

+
Pm (x)t

1
1

,
2
1 2xt + t
1 2xt + t2

m=0

hPn (x), Pm (x)i t

n=0 m=0

n+m

=
1

1
dx.
1 2xt + t2

By use of (3.34) we drop the terms where n 6= m and consider only the cases where n = m

3.4

Orthogonality

19

obtaining the following:

2n

hPn (x), Pn (x)i t

n=0

=
1

t2n

n=0

Pn2 (x)dx =

1
dx
1 2xt + t2
1
dx
1 2xt + t2

1

1
2
= ln(1 2xt + t )
2t
1

1
2
ln(1 2t + t ) ln(1 + 2t + t2 ) .
=
2t

Note that ln(t) = ln(t)

X
n=0

2n

Pn2 (x)dx =


1
ln(1 t)2 ln(1 + t)2
2t

1
= (ln(1 t) ln(1 + t)) .
t

Note that ln(a) ln(b) = ln( ab )

2n

n=0

1 1t
Pn2 (x)dx = ln
t 1+t
1


1
1 t 1
= ln
t
1+t
1 1+t
= ln
.
t 1t

Using Taylor expansion we obtain

ln

X 2t2n+1
1+t
2t3 2t5 2t7
2t2n+1
=2+
+
+
+ +
+ =
1t
3
5
7
2n + 1
2n + 1

(3.30)

n=1

and so

X 2t2n
2t2 2t4 2t6
2t2n
1 1+t
ln
=2+
+
+
+ +
+ =
.
t 1t
3
5
7
2n + 1
2n + 1
n=1

Finally
Z 1

X
X
2t2n
2n
=
t
Pn2 (x)dx,
2n + 1
1

n=0

n=0

(3.31)

20

CHEBYSHEVS POLYNOMIALS

hence
Z

Pn2 (x)dx =

2
2n + 1


Remark:
We have shown that Legendre polynomials are orthogonal, hence, linearly independent. We can
refer back to Theorem 2.1.6. and make a comparison. We can say that in the Hilbert space
L2 [(1, 1), dt], Legendre polynomials act as eigen functions. However, we do not claim that Pn0 s
span full L2 [(1, 1), dt] space, because proving it would require much deeper understanding so
we will not be focusing on that in this project.

Chebyshevs Polynomials

Definition 4.0.1. Chebyshevs polynomials are defined as


Tn (x) = cos(n arccos(x)),

n N+ , 1 x 1.

The first eleven Chebyshevs polynomials [9] are:


T0 (x) = 1
T1 (x) = x
T2 (x) = 2x2 1
T3 (x) = 4x3 3x
T4 (x) = 8x4 8x2 + 1
T5 (x) = 16x5 20x3 + 5x
T6 (x) = 32x6 48x4 + 18x2 1
T7 (x) = 64x7 112x5 + 56x3 7x
T8 (x) = 128x8 256x6 + 160x4 32x2 + 1
T9 (x) = 256x9 576x7 + 432x5 120x3 + 9x
T10 (x) = 512x1 0 1280x8 + 1120x6 100x4 + 50x2 1

(4.1)

4.1

Generating Function

21

Figure 4: Chebyshevs polynomials of degree 0 through 10

4.1

Generating Function

Proposition 4.1.1. Let x = cos() and Tn (x) = cos(n). Then the generating function [5] for
Chebyshevs polynomials is

X
1 xt
=
Tn (x)tn
2
1 2xt + t
n=0

for t C, |t|  1, n N0 .

PROOF:
Let |t| < 1, then

X
n=0

Tn (x)t =

cos(n)tn .

(4.2)

n=0

By the use of Eulers formulas:


ein = cos n + i sin n

(4.3)

ein = cos n i sin n

(4.4)

ein + ein = 2 cos(n)

(4.5)

If we add (4.3) to (4.4) we get

22

CHEBYSHEVS POLYNOMIALS

Now dividing both sides of (4.5) by 2, we can see that

Tn (x)tn =

n=0

i
1 X h i n
(te ) + (tei )n
2

1
=
2

Note that

n=0 (te

i )n

1
1tei

X
n=0

and

n=0

1
Tn (x)t =
2

(te ) +

n=0

n=0 (te

i n

!
(te

i n

n=0

i )n

1
,
1tei

so

1
1
+
i
1 te
1 tei



1 1 tei + 1 tei
=
2 (1 tei )(1 tei )


1 2 2t(ei + ei )
=
2 1 (ei + ei ) + t2
1 t cos()
=
.
1 2t cos() + t2

And since x = cos()

Tn (x)tn =

n=0

1 xt
.
1 2xt + t2

(4.6)

4.2

The Differential Equation

Proposition 4.2.1. Chebyshevs polynomials solve Chebyshevs differential equation [6]

(1 x2 )Tn (x)00 xTn (x)0 + n2 Tn (x) = 0,

PROOF:

1 x 1

(4.7)

4.2

The Differential Equation

23

We differentiate the generating f unction (4.6) with respect to x and t:

X 0
2t(1 tx)
t
:
Tn (x)tn =

2
2
x
(1 2xt + t )
1 2xt + t2
n=0

t(1 t2 )
(1 2xt + t2 )2

(4.8)

2 X 00
4t2 (1 t2 )
n
:
T
(x)t
=
n
x2
(1 2xt + t2 )3

(4.9)

n=0

X
(2t 2x)(1 xt)
x

:
nTn (x)tn1 =
2
t
1 2xt + t
(t2 2xt + 1)2
n=0

xt2 2t + x
(1 2xt + t2 )2

(4.10)

2 X
2(2t 2x)(xt2 2t + x)
2xt 2
n2
:

n(n

1)T
(x)t
=
n
t2
(t2 2xt + 1)2
(1 2xt + t2 )3
n=0

(1 x2 )Tn00 (x)tn =

n=0

xTn0 (x)tn =

n=0

2(xt3 3t2 + 3xt 2x2 + 1)


(t2 2xt + 1)3

(4.11)

4t2 (1 x2 )(1 t2 )
(1 2xt + t2 )3

(4.12)

xt(1 t2 )
.
(1 2xt + t2 )2

(4.13)

Multiplying (4.10) by t we get

nTn (x)tn =

n=0

t(xt2 2t + x)
,
(1 2xt + t2 )2

(4.14)

then multiplying (4.11) by t2

n(n 1)Tn (x)tn =

n=0

2t2 (xt3 3t2 + 3xt 2x2 + 1)


(1 2xt + t2 )3

(4.15)

Now adding (4.14) to (4.15) we obtain the following:

X
n=0

n2 Tn (x)tn =

t(xt2 2t + x) 2t2 (xt3 3t2 + 3xt 2x2 + 1)

(1 2xt + t2 )2
(1 2xt + t2 )3

(4.16)

24

CHEBYSHEVS POLYNOMIALS

Adding (4.12), subtracting (4.14) and then adding (4.16) we see that

X
n=0


4t2 (1 x2 )(1 t2 )
(1 x2 )Tn00 (x) xTn0 (x) + n2 Tn (x) tn =
(1 2xt + t2 )3

(4.17)

xt(1 t2 )
t(xt2 2t + x)
+
(1 2xt + t2 )2 (1 2xt + t2 )2
2t2 (xt3 3t2 + 3xt 2x2 + 1)

(1 2xt + t2 )3

Now writing everything under common denominator we obtain the following:

X
n=0


4t2 (1 t2 x2 + x2 t2 )
(1 x2 )Tn00 (x) xTn0 (x) + n2 Tn (x) tn =
(1 2xt + t2 )3
xt(1 t2 )(1 2xt + t2 )
(1 t2 + x2 t2 )3
2
t(xt 2t + x)(1 2xt + t2 )
+
(1 2xt + t2 )3
2t2 (xt3 3t2 + 3xt + 2x2 + 1)

(1 2xt + t2 )3

(4.18)

= (4t2 4t4 4t2 x2 + 4x2t4 xt + 2x2 t2 xt3 xt3 2x2 t4 + xt5 + xt3 3x2 t4 + xt5 2t2 + 4xt3
2t4 + tx 2x2 t2 + t3 x 2xt5 t4 6xt3 + 4x2 t2 2t2 )/(1 2xt + 2x2 + 1)3
0/(1 2xt + 2x2 + 1) = 0
Hence


(1 x2 )Tn00 (x) xTn0 (x) + n2 Tn (x) tn = 0

n=0

Now setting the coefficients of tn equal to zero we obtain


(1 x2 )Tn00 (x) xTn0 (x) + n2 Tn (x) = 0

Proposition 4.2.2. Let u(x), v(x) be continuous and twice differentiable functions, x R. Let
A be the differential operator
p

x2

d
dx

p

d
2
1x
.
dx

Then
hAu, vi = hu, Avi ,


dx
in L2 [1, 1], 1x
2

i.e. A is self-adjoint.

4.2

The Differential Equation

25

PROOF:

dx
(Au(x)) v(x)
1 x2
1

Z 1 p
d p
d
dx
2
2
1x
( 1x
u(x) v(x)
=
dx
dx
1 x2
1
Z 1 p
0
=
1 x2 u0 (x) v(x)dx

hAu, vi =

Z 1p
 1
p

0
2
1 x u (x)v(x)
1 x2 u0 (x)v 0 (x)dx
=
1
1
Z 1
p
0
=0
u0 (x)
1 x2 v 0 (x) dx
1
1

Z
=0+

u(x) (Av(x))

dx
1 x2

= hu, Avi

i.e. A is self-adjoint.

Remark:
If we divide (4.7) by

1 x2 we obtain the following:

p
x
n2
1 x2 Tn00 (x)
Tn0 (x) +
Tn (x) = 0
1 x2
1 x2

(4.19)

Proposition 4.2.3. The expression


p
0
n2
1 x2 Tn0 (x) +
Tn (x) = 0
1 x2

(4.20)

is the self-adjoint form of the Chebyshev differential equation.

PROOF:
If we multiply (4.20) by
(x) = n2 , we obtain

1 x2 , apply A as in Proposition 4.2.2 and let u(x) =

T (x)
n
1x2

and

26

CHEBYSHEVS POLYNOMIALS

Au = u,
p

d
d
n2
2
1x
Tn (x)
Tn (x) =
dx
dx
1 x2
p
0
n2
Tn (x),
1 x2 Tn0 (x) =
1 x2

(4.21)

hence A is a self-adjoint operator and (4.21) is a self-adjoint form of Chebyshevs differential


equation.

Remark: We compare the above to Theorem 2.1.5 and see that the same idea repeats
once again, but instead of having an eigenvector u, we have an eigenfunction u(x) and an
eigenvalue = n2 R.

4.3

Recurrence Relation

Proposition 4.3.1. Chebyshevs Polynomials satisfy the following recurrence relation [7]:
Tn+1 (x) = 2xTn (x) Tn1 (x)

PROOF:
Lets introduce the notation = arccos(x).
Then (4.1) becomes Tn ((x)) = Tn () = cos(n), where 0 6 6 2.
We observe that replacing n by n + 1

Tn+1 () = cos((n + 1)) = cos(n) cos(n) sin(n) sin(n),

(4.22)

Tn1 () = cos((n 1)) = cos(n) cos() + sin(n) sin().

(4.23)

Now adding (4.22) to (4.23) we obtain the following


Tn+1 + Tn1 = 2 cos(n)cos() + sin(n) sin(),
Tn+1 () = 2 cos(n) cos() Tn1 (),
Tn+1 (x) = 2xTn (x) Tn1 (x),

4.4

Orthogonality

27

or equivalently
Tn+1 (x) = 2xTn (x) Tn1 (x),

(4.24)

which is the recurrence relation for Chebyshevs Polynomials.

Observation: Now we know that Tn s are indeed polynomials.

4.4

Orthogonality



1
Proposition 4.4.1. Chebyshevs Polynomials are orthogonal in L2 [1, 1], 1x
.
2
PROOF:
We shall prove this by using (4.20) for

o
d np
n2
1 x2 Tn0 +
Tn = 0
dx
1 x2
o
d np
m2
0
+
1 x2 Tm
Tm = 0
dx
1 x2

(4.25)
(4.26)

Multiplying (4.25) by Tm and (4.26) by Tn and then subtracting the results we obtain
o
o
d np
d np
1 x2 Tn0 Tm
1 x2 Tm Tn +
dx
dx
n
o
p
d
0
1 x2 (Tn0 Tm Tm
Tn )
dx

n2 m2

Tm Tn = 0
1 x2
n2 m2

Tm Tn = 0
1 x2

(4.27)

Now if we integrate (4.27) over the interval [1, 1] with respect to x we get
Z



Tm Tn
1 x2

0
0

dx = 2
T
T

T
T

m
n
m

n m2 n
1 x2
1
Z 1
Tm Tn

dx = 0
1 x2
1

for n 6= m.

We can clearly see that the value of (4.28) is zero if m = n 6= 0.


when m = n = 0 and m = n 6= 0?

(4.28)

But what happens

28

CHEBYSHEVS POLYNOMIALS

Let
= arccos(x)
d =

dx
.
1 x2

Tn (x) = cos(n arccos(x)) = cos(n)

(4.29)

Tm (x) = cos(m arccos(x)) = cos(m)

(4.30)

where [0, ], n, m N.
Now if we take a scalar product in L2 [0, ] of (4.29) and (4.30) we obtain the following:
Z
cos(n) cos(m)d
(4.31)
hTn (x), Tm (x)i =
0
Z
cos(n + m) cos(n m)
=
d
(4.32)
2
0

Now if n = m 6= 0 we can see that (4.32) equals to




Z
1
1 sin(2n) x

(cos((2n) cos(0)) d =
=
2 0
2
2n
2
0

and if n = m = 0 we have


Z
1 sin()
1
(cos((2n) cos(0)) d =
2x =
2 0
2
2n
0

Now summarizing (4.28), (4.33) and (4.34) we finally get

for
0,
Z 1
Tn (x)Tm (x)

=
for
2,

1 x2
1

,
for

(4.33)

(4.34)

m 6= n,
m = n 6= 0,
m=n=0


Remark:
We can refer back to Theorem 2.1.6 once again and and to summarize the above we can say that
Tn s are eigen functions, since they are orthogonal, i.e. linearly independent. However, similarly


dx
, since
as for Legendre polynomials, we do not claim that they span entire L2 [1, 1], 1x
2
showing it would require much further analysis and we shall not be focusing on that in this
project.

29

Conclusion

We can see that we have been following the same pattern through out the paper. Either we are
examining Hermite matrices, Legendre polynomials or Chebyshevs polynomials, in each case we
have some scalar product and orthogonality in some Hilbert space. We have also noticed a strong
connection between Hermite matrices and Orthogonal polynomials in general, where Hermmite
matrices are self-adjoint and Orthogonal polynomials can be also expressed in their self adjoint
form. It is also important to stress that the Spectral Theorem plays an important role in the
analysis of Orthogonal polynomials.

30

REFERENCES

References
[1] Wikipedia, Spectral Theorem, (https://en.wikipedia.org/wiki/Spectraltheorem), (Accessed
on 30/03/2016).
[2] M. Youngson, Functional Analysis Lecture Notes, Heriot Watt University, Edinburgh, 2015.
[3] Z. X. Wang, D. R. Guo. Special Functions, chapter 4. Hypergeometric Function, page 176.
Singapore.
[4] H. D. Sterck, Week 4 - Discrete Fourier Methods, Introduction to Computational Mathematics Course Notes, University of Waterloo, Waterloo, 2015.
[5] Holt, Rinehart and Winston, Special Functions of Mathematical Physics, New York, 1961.
[6] G. Szego, American Mathematical Society. Orthogonal Polynomials. Colloquium Publication.
Volume XXIII. Chapter II. Definition of Orthogonal Polynomials; Principal Examples, pages
23-29. Providence, Rhode Island, 1939.
[7] R. A. Silverman (ed.). Special Functions and Their Applications, chapter 4, Orthogonal
Polynomials, pages 43-50. PRENTICE-HALL, INC. Englewood Cliffs, N.J., 1965.
[8] Wikipedia, Legendre Polynomials (https://en.wikipedia.org/wiki/Legendrepolynomials),
(Accessed on 17/03/2016)
[9] Wikipecia, Chebyshev Polynomials, (https://en.wikipedia.org/wiki/Chebyshevpolynomials),
(Accessed on 21/03/2016).
[10] M. Dreher, Mathematics for Physics III, (https://sites.google.com/site/michaeldreher7/home/
lecture-notes), (Accessed on 15/03/2016).

Das könnte Ihnen auch gefallen