III.

4 Classical Orthogonal Polynomials

Classical orthogonal polynomials are special families of orthogonal polynomials with a
number of beautiful properties, for example
1. Their derivatives are also OPs
2. They are eigenfunctions of simple differential operators
As stated above orthogonal polynomials are uniquely defined by the weight and w(x)

the constant . The classical orthogonal polynomials are꞉

kn

1. Chebyshev polynomials (1st kind) ꞉

Tn (x) w(x) = 1/√1 − x on . 2
[−1, 1]

2. Chebyshev polynomials (2nd kind) ꞉

Un (x) √1 − x2 on . [−1, 1]

3. Legendre polynomials ꞉
Pn (x) w(x) = 1 on .
[−1, 1]

4. Ultrapsherical polynomials (my fav!)꞉ Cn

(λ)
꞉ (x) w(x) = (1 − x ) on , 2 λ−1/2
[−1, 1]

,
λ ≠ 0 λ > −1/2 .
5. Jacobi polynomials꞉ Pn
(a,b)
꞉
(x) w(x) = (1 − x) (1 + x) on
a
, b
.[−1, 1] a, b > −1

6. Laguerre polynomials꞉ ꞉
Ln (x) w(x) = exp(−x) on . [0, ∞)

7. Hermite polynomials ꞉
Hn (x) w(x) = exp(−x ) on 2
.(−∞, ∞)

In the notes we will discuss꞉

1. Chebyshev polynomials꞉ These are closely linked to Fourier series and are one of the
most powerful tools in numerics.
2. Legendre polynomials꞉ These have no simple closed‑form expression but can be
defined in terms of a Rodriguez formula, a feature that
applies to all other classical families.
1. Chebyshev
Definition 1 (Chebyshev polynomials, 1st kind) Tn (x) are orthogonal with respect to
1/√1 − x and satisfy꞉
2

T0 (x) = 1,

n−1 n n−1
Tn (x) = 2 x + O(x )

Definition 2 (Chebyshev polynomials, 2nd kind) Un (x) are orthogonal with respect to
√1 − x .
2

n n n−1
Un (x) = 2 x + O(x )

Theorem 1 (Chebyshev T are cos) For −1 ≤ x ≤ 1

Tn (x) = cos n acos x.

In other words
Tn (cos θ) = cos nθ.

Proof
We need to show that are
pn (x) ꞉= cos nacos x

1. graded polynomials
2. orthogonal w.r.t. 1/√1 − x on 2
, and[−1, 1]

3. have the right normalisation constant kn = 2

n−1
for n = 2, … .
Property (2) follows under a change of variables꞉
1 π π
pn (x)pm (x) cos(nθ) cos(mθ)
∫ dx = ∫ sin θdθ = ∫ cos(nθ) cos(mθ)dx = 0
−1 √ 1 − x2 0 √1 − cos2 θ 0

if n ≠ m.
To see that they are graded we use the fact that
cos(n − 1)θ + cos(n + 1)θ pn−1 (x) + pn+1 (x)
xpn (x) = cos θ cos nθ = =
2 2

In other words pn+1 (x) = 2xpn (x) − pn−1 (x) . Since each time we multiply by and 2x

p0 (x) = 1 we have
n n−1
pn (x) = (2x) + O(x )

which completes the proof.

∎

Buried in the proof is the 3‑term recurrence꞉

Corollary 1 (Chebyshev 3‑term recurrence)
xT0 (x) = T1 (x)

Tn−1 (x) + Tn+1 (x)

xTn (x) =
2

Chebyshev polynomials are particularly powerful as their expansions are cosine series in
disguise꞉ for
∞
f (x) = ∑ f̌ k Tk (x)
k=0

we have
∞

f (cos θ) = ∑ f̌ k cos kθ.

k=0
Thus the coefficients can be recovered fast using FFT‑based techniques as discussed in
the problem sheet.
In the problem sheet we will also show the following꞉
Theorem 2 (Chebyshev U are sin) For ,
x = cos θ

sin(n + 1)θ
Un (x) =
sin θ

which satisfy꞉
xU0 (x) = U1 (x)/2

Un−1 (x) Un+1 (x)

xUn (x) = + .
2 2

2. Legendre
Definition 3 (Legendre) Legendre polynomials Pn (x) are orthogonal polynomials with
respect to w(x) = 1 on , with
[−1, 1]

1 2n (2n)!
kn = ( ) =
n n 2
2 n 2 (n!)

The reason for this complicated normalisation constant is both historical and that it leads
to simpler formulae for recurrence relationships.
Classical orthogonal polynomials have Rodriguez formulae, defining orthogonal
polynomials as high order derivatives of simple functions. In this case we have꞉
Lemma 1 (Legendre Rodriguez formula)
n
1 d
2 n
Pn (x) = (1 − x )
n n
(−2) n! dx

Proof We need to verify꞉

1. graded polynomials
2. orthogonal to all lower degree polynomials on [−1, 1] , and
3. have the right normalisation constant kn =
2
1
n
(
2n

n
) .
(1) follows since its a degree polynomial (the ‑th derivative of a degree
n n 2n

polynomial). (2) follows by integration by parts. Note that and its first
(1 − x )
2 n
n − 1

derivatives vanish at . If is a degree

±1 rm polynomial we have꞉
m < n
 1 n 1 n−1
d d
2 n 2 n ′ n
∫ (1 − x ) rm (x)dx = − ∫ (1 − x ) rm (x)dx = ⋯ = (−)
n n−1
−1
dx −1
dx

2 n (n)
∫ (1 − x ) rm (x)dx = 0.
−1

(3) follows since꞉

n n−1
d d
n 2n 2n−1 n 2n−1 2n−1
[(−) x + O(x )] = (−) 2n x + O(x )]
dxn dxn−1
n−2

n
d 2n−2 2n−2
= (−) 2n(2n − 1) x + O(x )] = ⋯
n−2
dx

(2n)!
n n n−1 n
= (−) 2n(2n − 1) ⋯ (n + 1)x + O(x ) = (−)
n!

This allows us to determine the coefficients which are useful in proofs. In particular (λ)
kn

we will use ꞉ kn
(2)

Lemma 2 (Legendre monomial coefficients)

P0 (x) = 1

P1 (x) = x

(2n)! (2n − 2)!

n n−2 n−4
Pn (x) = x − x + O(x )
n n
2 (n!)2 2 (n − 2)!(n − 1)!
 
kn (2)
kn

(Here the is as O(x , which implies that the term is a polynomial of degree
n−4
) x → ∞

≤ n − 4 . For the term is therefore precisely zero.)

n = 2, 3 O(x
n−4
)

Proof
The and case are immediate. For the other case we expand
n = 0 1 to get꞉ (1 − x )
2 n

n n
d d
n 2 n 2n 2n−2 2n−4
(−) (1 − x ) = [x − nx + O(x )]
n n
dx dx
n n−2
= (2n) ⋯ (2n − n + 1)x − n(2n − 2) ⋯ (2n − 2 − n + 1)x

(2n)! n(2n − 2)!

n n−2 n−4
= x − x + O(x )
n! (n − 2)!

Multiplying through by n
1

2 (n!)
completes the derivation.
∎

Theorem 3 (Legendre 3‑term recurrence)

 xP0 (x) = P1 (x)

(2n + 1)xPn (x) = nPn−1 (x) + (n + 1)Pn+1 (x)

Proof The n = 0case is immediate (since w(x) = w(−x) an = 0 , from PS8). For the
other cases we match terms꞉
n+1
(2n + 1)xPn (x) − nPn−1 (x) − (n + 1)Pn+1 (x) = [(2n + 1)kn − (n + 1)kn+1 ]x

(2) (2)
n−1 n−3
+ [(2n + 1)kn − nkn−1 − (n + 1)k ]x + O(x )
n+1

Using the expressions for andkn kn

(2)
above we have (leaving the manipulations as an
exercise)꞉
(2n + 1)! (2n + 2)!
(2n + 1)kn − (n + 1)kn+1 = − (n + 1)
n
2 (n!)2 2
n+1
((n + 1)!)2

(2)
(2n − 2)! (2n − 2
(2)
(2n + 1)kn − nkn−1 − (n + 1)k = −(2n + 1) − n
n+1 n
n−1
2 (n − 2)!(n − 1)! 2 ((n −

(2n)!
+ (n + 1) = 0
n+1
2 (n − 1)!n!

Thus
n−3
(2n + 1)xPn (x) − nPn−1 (x) − (n + 1)Pn+1 (x) = O(x )

But as it is orthogonal to Pk (x) for 0 ≤ k ≤ n − 3 it must be zero. ∎