Introduction to Orthogonal Polynomials

Lecture held at University Dortmund

in summer 2002/2003

H. Michael Möller

April 2006
Chapter 3

Special orthogonal families

In this chapter, we study orthogonal systems (OS) {pn }∞ n=0 for given inner
product hf, gi := L(f g) where L is a square positive functional.

pn will always denote an orthogonal polynomial of degree n. If pn is orthonor-

mal and has a positive leading coefficient, then we will often write vn instead
of pn . In case lc(pn ) = 1, we also use un instead of pn .

3.1 The Legendre polynomials

Let L : P → R be defined by
Z 1
L(f ) := f (x) dx for all f ∈ P.
−1

The orthogonal polynomials w.r.t. (the inner product induced by) L are called
Legendre polynomials.

A very elegant way of obtaining explicit expressions for the Legendre poly-
nomials is the use of higher order antiderivatives.

Let pn be a Legendre polynomial of degree n. ϕn+1 is the (uniquely defined)

antiderivative of pn which satisfies ϕn+1 (−1) = 0. Then let ϕk+1 be the
(uniquely defined) antiderivative of ϕk with ϕk+1 (−1) = 0, for k = n +
(k)
1, . . . , 2n − 1. By differentiation ϕ2n = ϕ2n−k , k = 0, . . . , n − 1, especially
(k)
ϕ2n (−1) = 0, k = 0, . . . , n − 1,

37
38 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

(n)
and pn = ϕ2n .

Since pn has degree n, ϕk has degree k, k = n+1, . . . , 2n. ϕ2n is a polynomial

of degree 2n with an n-fold zero at −1.

For arbitrary q ∈ Pn−1 we have by orthogonality and partial integration

Z 1
(n)
0 = L(pn q) = ϕ2n (x)q(x) dx (3.0)
−1
Z 1
(n−1) (n−1)
= ϕ2n (x)q(x)|1−1 − ϕ2n (x)q ′ (x) dx
−1
Z 1
(n−1) (n−1)
= ϕ2n (1)q(1) − ϕ2n (x)q ′ (x) dx
−1
1 Z 1
(n−1−ν) (n−2)
X
ν
= (−1) ϕ2n (1)q (ν) (1) + ϕ2n (x)q ′′ (x) dx
ν=0 −1
..
.
n−1 Z 1
(n−1−ν) (0)
X
ν
= (−1) ϕ2n (1)q (ν) (1) + (−1) n
ϕ2n (x)q (n) (x) dx.
ν=0 −1

Since q (n) = 0 for q ∈ Pn−1 , we get for arbitrary q ∈ Pn−1

n−1
(n−1−ν)
X
0 = L(pn q) = (−1)ν ϕ2n (1)q (ν) (1).
ν=0

Inserting qℓ (x) := ℓ!1 (x − 1)ℓ ∈ Pn−1 , ℓ = 0, . . . , n − 1, for q gives

(n−1−ℓ)
0 = L(pn qℓ ) = (−1)ℓ ϕ2n (1), ℓ = 0, . . . , n − 1,

(ν)
since qℓ (1) = δℓν (biorthogonality!). Hence ϕ2n has in 1 an n-fold zero, too.
Being a degree 2n polynomial, it has no other zeros. Therefore

ϕ2n (x) = Kn (x2 − 1)n

with a constant Kn which is fixed by the normalization of pn .

(n)
Since ϕ2n = pn , we obtain the so called Rodrigues formula

dn (x2 − 1)n
pn (x) = Kn . (3.1)
dxn
3.1. THE LEGENDRE POLYNOMIALS 39

The polynomial (x2 − 1)n has leading coefficient 1 and degree 2n. Hence the
dn (x2 − 1)n (2n)!
leading coefficient of is and
dxn n!
(2n)!
lc(pn ) = Kn .
n!
If we consider un , the n-th Legendre polynomial with leading coefficient 1,
then Kn = n! and
(2n)!
n! dn (x2 − 1)n
un (x) = . (3.2)
(2n)! dxn

We need also L(p2n ) for finding an explicit representation of the orthonormal

Legendre polynomials and for the three term recursion formula. The first cal-
culation steps are similar to those of (3.0). We obtain by partial integration,
since we already know that ϕ2n has an n-fold zero also in 1,
Z 1 Z 1
(n) n (0)
L(pn pn ) = ϕ2n (x)pn (x) dx = (−1) ϕ2n (x)p(n)
n (x) dx.
−1 −1
(0) (n)
Inserting ϕ2n = Kn (x2 − 1)n and pn = n! lc(pn ) = Kn (2n)! we get
Z 1
n 2
L(pn pn ) = (−1) Kn (2n)! (x2 − 1)n dx.
−1
R1
Let In := −1
(x2 − 1)n dx. Then
Z 1
In = x2 (x2 − 1)n−1 dx − In−1
−1

and by partial integration

Z 1 Z 1
2 2 n−1 1 2
1
n 1 1
x (x − 1) dx = x(x − 1)  − (x2 − 1)n dx = − In .
−1 2n −1 2n −1 2n
We combine the last two equations
−2n
In = In−1 .
2n + 1
Using the notation1 k!! := k(k −2) · · · 3·1 for odd k and k!! := k(k −2) · · · 4·2
for even k we obtain by induction
(2n)!! (2n)!!
In = (−1)n I0 = (−1)n 2.
(2n + 1)!! (2n + 1)!!
1
formally (−1)!! := 1, 0!! := 1, and then recursively k!! := k · (k − 2)!! for k ≥ 1.
40 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

This gives finally (using the trivial identity (2n)! = (2n)!!(2n − 1)!!)

(2n)!! 2
L(p2n ) = Kn2 (2n)! 2 = Kn2 (2n)!!2 . (3.3)
(2n + 1)!! 2n + 1

Since (2n)!! = 2n(2n − 2) · · · 4 · 2 = 2n n! one can write alternatively

2
L(p2n ) = Kn2 (2n n!)2 .
2n + 1

The orthonormal Legendre polynomial of degree n is therefore given if

r
1 −n 2n + 1
Kn = 2 .
n! 2
Hence r
1 2n + 1 dn (x2 − 1)n
vn (x) = 2−n . (3.4)
n! 2 dxn

A different normalization is given by pn (1) = 1.

Theorem 3.1If the n-th Legendre polynomial is normalized by pn (1) = 1

(and denoted in the forthcoming by Pn ), then

1 −n dn (x2 − 1)n
Pn (x) = 2 . (3.5)
n! dxn
These polynomials satisfy the three-term-recursion

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

with initial polynomials P0 (x) = 1, P1 (x) = x. In addition

Z 1
2
Pn2 (x) dx = .
−1 2n + 1

Proof. (Exercises).

From (3.1) we can derive an explicit representation for the coefficient of the
Legendre polynomials. The binomial
n  
2 n
X n
(x − 1) = (−1)k x2n−2k
k=0
k
3.1. THE LEGENDRE POLYNOMIALS 41

n times differentiated gives

n  
X n (2n − 2k)! n−2k
pn (x) = Kn (−1)k x . (3.6)
k=0
k (n − 2k)!

This formula shows, that pn is an even function if n is even, and an odd

function if n is odd.

We will derive now the three-term recursion for the Legendre polynomials
using theorem 2.5. Therefore is is most convenient to consider the normal-
ization lc(pn ) = 1. We write hence un for pn .

From Thm. 2.5 we know

un+1 (x) = (x − βn )un (x) − γn un−1(x), n ∈ N.

By normalization u0 = 1 and u1 has lc(u1 ) = 1. But, as remarked before, u1

is an odd function. Therefore u1 (x) = x.

βn = 0 since un+1 , xun , and un−1 are simultaneously even or odd and βn un
has the other parity. Hence βn un = 0. By thm. 2.5 γn = L(u2n )/L(u2n−1 ).
Using equation (3.3) and Kn = n!/(2n!), this gives

n!2 2
L(u2n ) = 2
(2n)!!2 ,
(2n)! 2n + 1
(n − 1)!2 2
L(u2n−1 ) = 2
(2n − 2)!!2 ,
(2n − 2)! 2n − 1

and hence
n2 2n − 1
γn = 2 2
(2n)2
(2n − 1) (2n) 2n + 1
n2
= .
(2n − 1)(2n + 1)

Thus u0 = 1, u1 (x) = x, and

n2
un+1 (x) = xun (x) − un−1 (x), n ∈ N. (3.7)
(2n − 1)(2n + 1)

The first Legendre polynomials with leading coefficient 1 are easily computed
42 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

by this recursion.

u2 (x) = x2 − 31
u3 (x) = x3 − 53 x
u4 (x) = x4 − 76 x2 + 35
3

u5 (x) = x5 − 109
5
x3 + 21 x
6 15 4 5 2 5
u6 (x) = x − 11 x + 11 x − 231
21 5
u7 (x) = x7 − 13 x + 105
143
35
x3 − 429 x

The orthonormal Legendre polynomials vn satisfy by (3.4) and (3.2)

r r
1 −n 2n + 1 dn (x2 − 1)n 1 −n 2n + 1 (2n)!
vn = 2 = 2 un =: cn un .
n! 2 dxn n! 2 n!
Hence substituting vk for uk in (3.7), k = n + 1, n, n − 1, gives

cn+1 n2 cn+1
vn+1 (x) = xvn − vn−1 (x)
cn (2n − 1)(2n + 1) cn−1
= An xvn (x) − Cn vn−1 (x)

with
√ √ √
1 1 2n + 3 (2n + 1)(2n + 2) 2n + 3 2n + 1
An = √ =
n + 1 2 2n + 1 n+1 n+1
cn+1 cn+1 cn
and using cn−1
= cn cn−1
= An An−1
√ √ √ √
n2 2n + 3 2n + 1 2n + 1 2n − 1
Cn =
(2n − 1)(2n + 1) n+1 n
√
n 2n + 3
= √ .
n + 1 2n − 1
This gives the three term recursion
√ √ √
2n + 3 2n + 1 n 2n + 3
vn+1 = xvn (x) − √ vn−1 (x), n ∈ N, (3.8)
n+1 n + 1 2n − 1
√
v0 = √12 , v1 (x) = √32 x. (The initial polynomials v0 and v1 are calculated by
normalizing 1 and x resp.)

Now, we present some formulas for expressing derivatives of the Legendre

polynomials. First the differential equation of second order.
3.1. THE LEGENDRE POLYNOMIALS 43

Theorem 3.2 The Legendre polynomial pn is a solution of the differential

equation
(1 − x2 )y ′′ − 2xy ′ + n(n + 1)y = 0.
Proof. The Leibniz rule for the (n + 1)-st derivative of a product of functions
reads
n+1  
(n+1)
X n + 1 (n+1−k) (k)
(uv) = u v .
k=0
k
If v is a polynomial of degree at most 2, then it simplifies to
1
(uv)(n+1) = u(n+1) v + (n + 1)u(n) v ′ + n(n + 1)u(n−1) v ′′ .
2
Now, let u(x) := (x2 − 1)n and pn (x) = Kn u(n) . Then

u′ (x)(x2 − 1) = 2n x u(x).

Differentiating both sides n + 1 times gives

u(n+2) (x2 − 1) + 2(n + 1)u(n+1) x + n(n + 1)u(n) = 2nxu(n+1) + 2n(n + 1)u(n)

or after reordering and replacing u(n) by y,

(x2 − 1)y ′′ + 2xy ′ − n(n + 1)y = 0. (3.9)

This means, that u(n) and its constant multiples Kn u(n) solve this differential
equation for y. 

Beginning from now, we consider only the Legendre polynomial normalized

by pn (1) = 1. We still use the notation Pn for this Legendre polynomial.
Then, we obtain for instance
′ ′
Pn+1 (x) − Pn+1 (x) = (2n + 1)Pn (x). (3.10)

The proof for this identity follows from

1
Pn = (q n )(n) with q(x) = x2 − 1,
(2n)!!
since then
1 1
Pn+1 − Pn−1 = (q n+1 )(n+1) − (q n−1 )(n−1)
(2n + 2)!! (2n − 2)!!
 1 1 (n−1)
= (q n+1 )(2) − q n−1
(2n + 2)!! (2n − 2)!!
44 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

and by differentiation

′ ′
 1 n+1 (2) 1 n−1
(n)
Pn+1 − Pn−1 = (q ) − q .
(2n + 2)!! (2n − 2)!!

Because of
 ′
n+1 (2) 2 n+1 ′′ 2 n
(q ) (x) = ((x − 1) ) = (n + 1)(x − 1) (x)2x
 
= (2n + 2) 2nx2 + x2 − 1 (x2 − 1)n−1

this gives finally the equation (3.10)

 (2n + 1)x2 − 1 1 (n)
′ ′
Pn+1 (x) − Pn−1 (x) = ( − )(x2 − 1)n−1
(2n)!! (2n − 2)!!
 2n + 1 (n)
= (x2 − 1)n = (2n + 1)Pn (x).
(2n)!!

An other way of representing the Legendre polynomials is the use of gener-

ating functions.
P∞ n
Definition. The formal power series n=0 an z is called the generating
∞
function of {an }n=0 . In some applications, esp. in electrical engeneering, it
is also called the z-transformation of {an }∞
n=0 .

Consider the function

1
R(z, x) := √ .
1 − 2zx + z 2
From analysis, we know that for |t| < 1
∞
1 X (2n − 1)!!
√ = tn
1 − t n=0 (2n)!!

where 0!! := 1 and (−1)!! := 1. (In the following, we just need that the
coefficients of this power series are positive numbers.) Replacing t by 2zx−z 2
gives
X∞
R(z, x) = an (2zx − z 2 )n with an > 0.
n=0

We want to reorder this series. For this purpose

√ we have to be sure, that
the series converges absolutely. Consider 1/ 1 − t for t = |zx| + z 2 . This
3.1. THE LEGENDRE POLYNOMIALS 45

series converges and has positive coefficients, if (x is fixed and) |z| is suffi-
ciently small. Hence we have a converging majorizing series. This ensures
the absolute convergence required for reordering,
∞
X
R(z, x) = αn (x)z n .
n=0

Because of
1
1= √ = R(0, x) = α0 (x)z 0 = α0 (x)
1 − 2 · 0x + 0 2

and
∂R −1
(0, x) = (1 − 2zx + z 2 )−3/2 (−2x + 2z)|z=0 = x
∂z 2
∞
X
= nαn (x)z n−1 |z=0 = α1 (z)
n=1

we already have α0 (x) = 1(= P0 (x)) and α1 (x) = x(= P1 (x)). Our goal is to
show
αn (x) = Pn (x) for all n ∈ N0 .
∂R equals
∂z ∞
x−z X
√ 3 = nαn (x)z n−1 .
1 − 2zx + z 2
n=1

Multiplication by (1 − 2zx + z 2 ) gives

∞
X
2
(x − z)R(z, x) = (1 − 2zx + z ) nαn (x)z n−1 .
n=1

Hence
∞
X ∞
X
n
x αn (x)z − αn (x)z n+1 =
n=0 n=0
∞
X ∞
X ∞
X
n−1 n
nαn (x)z − 2x nαn (x)z + nαn (x)z n+1
n=1 n=1 n=1

or by resummation
∞
X ∞
X
x αn (x)z n − αn−1 (x)z n =
n=0 n=1
∞
X ∞
X ∞
X
n n
(n + 1)αn+1(x)z − 2x nαn (x)z + (n − 1)αn−1 (x)z n
n=0 n=1 n=1
46 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

Comparing the coefficients of z n one has for n ≥ 1

xαn (x) − αn−1 = (n + 1)αn+1 (x) − 2x nαn (x) + (n − 1)αn−1 (x)

or reordered

(n + 1)αn+1 (x) = (2n + 1)xαn (x) − nαn−1 .

This is exactly the three term recursion as for the Legendre polynomials Pn ,
given in Theorem 3.1. Since we already have α0 = P0 and α1 = P1 , the
(same) recursion gives thus αn = Pn for n = 2, 3, . . .

Theorem 3.3 The Legendre polynomials Pn normalized by Pn (1) = 1 satisfy

∞
1 X
√ = Pn (x)z n .
1 − 2zx + z 2 n=0

The generating function √ 1 encodes the sequence {Pn (x)}∞

n=0 in
1 − 2zx + z 2
a certain sense. It contains the full information on all Legendre polynomials
Pn . In many instances it can be used for deriving relations among the Pn ’s
and estimates etc.

Theorem 3.4 |Pn (x)| ≤ 1 for all x ∈ [−1, 1].

Proof. Since cos : [0, 2π] → [−1, 1] is surjective, there is for every
x ∈ [−1, 1] a ϕ ∈ [0, 2π] such that x = cos ϕ. Then
∞
1 X
R(z, cosϕ) = p = Pn (cos ϕ)z n .
1 − 2zcosϕ + z 2 n=0

The Euler-de Moivre formula cos ϕ = 21 eiϕ + 21 e−iϕ gives

1 − 2 z cosϕ + z 2 = (1 − zeiϕ )(1 − ze−iϕ ).

Hence
1 1
R(z, cosϕ) = √ √
1 − zeiϕ 1 − ze−iϕ
∞ ∞
X (2k − 1)!! k iϕk  X (2n − 1)!! n −iϕn 
= z e z e
k=0
(2k)!! n=0
(2n)!!
3.1. THE LEGENDRE POLYNOMIALS 47
P∞
The Cauchy product of these two power series gives n=0 Pn (cos ϕ)z n . So
the coefficient of z n is
n
X (2k − 1)!! (2n − 2k − 1)!!
Pn (cosϕ) = ei(2k−n)ϕ .
k=0
(2k)!! (2n − 2k)!!

This gives finally (using |ei(2k−n)ϕ | = 1)

n
X (2k − 1)!! (2n − 2k − 1)!!
|Pn (cosϕ)| ≤ .
(2k)!! (2n − 2k)!!
k=0

But the right hand term equals Pn (cos 0) = Pn (1) = 1. 

Finally, we derive some useful relations among the Pn and their derivatives
using the generating function. The partial derivatives of R are

∂R = √ 1 −1 (−2z) = √ z =
P∞ ′ n
∂x 3 2 3 n=1 Pn (x)z
1 − 2zx + x 2 1 − 2zx + z 2
∂R = √ x − z P∞ n−1
∂z 3 = n=1 Pn (x)nz
1 − 2zx + z 2

Hence (x − z) ∂R
∂x
= z ∂R
∂z
. Therefore
∞
X ∞
X ∞
X ∞
X
(x − z) Pn′ (x)z n =x Pn′ (x)z n − Pn′ (x)z n+1 = Pn (x)nz n .
n=1 n=1 n=1 n=1

Comparing the coefficients of the same power of z gives for n ≥ 1

xPn′ (x) − Pn−1

′
(x) = nPn (x). (3.11)
′ ′
This relation combined with (3.10) (Pn+1 − Pn−1 = (2n + 1)Pn ) gives
′
Pn+1 (x) − xPn′ (x) = (n + 1)Pn (x)

and (replacing n + 1 by n)

Pn′ (x) − xPn−1

′
(x) = nPn−1 (x).

If we subtract from this equation the x-fold multiple of (3.11), then

(1 − x2 )Pn′ (x) = nPn−1 (x) − nxPn (x). (3.12)

48 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

3.2 The Chebyshev polynomials (first kind)

We start by the triginometric identities

cos((n + 1)t) = cos(t)cos(nt) − sin(t)sin(nt)

cos((n − 1)t) = cos(t)cos(nt) + sin(t)sin(nt)

If we add these two equations and dismantle for cos((n + 1)t), we obtain

cos((n + 1)t) = 2cos(t)cos(nt) − cos((n − 1)t).

Since cos : [0, 2π] → [−1, 1] is a bijection, we can substitute x for cos(t)
and get (t = arccos(x) and) for Tk (x) := cos(k arccos(x)) the three term
recursion
Tn+1 (x) = 2 x Tn (x) − Tn−1 (x) for x ∈ [−1, 1]. (3.13)
For n = 0 and n = 1 we can determine Tn directly,

T0 (x) = cos(0 arccos(x)) = cos(0) = 1,

T1 (x) = cos(arccos(x)) = x.

The forthcoming polynomials Tn can be computed easily by the recursion

(3.13)
T2 (x) = 2xT1 (x) − T0 (x) = 2x2 − 1
T3 (x) = 2xT2 (x) − T1 (x) = 4x3 − 3x
T4 (x) = 2xT3 (x) − T2 (x) = 8x4 − 8x2 + 1
T5 (x) = 2xT4 (x) − T3 (x) = 16x5 − 20x3 + 5x
etc.
We see (this can be proved easily by induction) that
– Tn is a polynomial of degree n in x.
– The leading coefficient of Tn is 2n−1 for n ≥ 1.
– Tn is even, if n is even, and odd, if n is odd. This is equivalent to

Tn (−x) = (−1)n Tn (x) for all n ∈ N0 .

Definition. The polynomials Tn given by T0 = 1, T1 (x) = x and by the three

term recursion (3.13) are called Chebyshev polynomials (of the first kind).

We have already found the identity for all x ∈ [−1, 1]

Tn (x) = cos(n arccos(x)), n ∈ N0 . (3.14)

3.2. THE CHEBYSHEV POLYNOMIALS (FIRST KIND) 49

For x = 1 we have arccos(x) = 0 because of cos(0) = 1. Therefore

Tn (1) = cos(n0) = 1, Tn (−1) = (−1)n .

We will derive now a representation for Tn (x), |x| > 1. Consider the linear
recursion with constant coefficients

an+1 − 2xan + an−1 = 0 n ∈ N.

A sequence {ak }∞ k=0 is called solution for this recursion, if any three consec-
utive elements ak−1 , ak , ak+1 satisfy ak+1 − 2xak + ak−1 = 0. Obviously, given
a0 and a1 , the solution of the recursion is uniquely determined.

On the one hand, we already know that {ak }∞ ∞

k=0 = {Tn (x)}n=0 is the (unique)
solution with a0 = 1, a1 = x. On the other hand, consider an := λn for all
n ∈ N0 . The sequence {λn }∞
n=0 is a solution, iff

λn+1 − 2xλn + λn−1 = λn−1 (λ2 − 2x λ + 1) = 0 for all n ≥ 1.

√
Hence the two values λ = x ± x2 − 1 lead to solutions of the recursion.
Obviously, also any linear combination
√ √
ak := c1 (x + x2 − 1)k + c2 (x − x2 − 1)k

gives a solution of the recursion, too. We choose c1 and c2 such that a0 = 1

and a1 = x holds true. c1 = c2 = 12 gives

1 √ 1 √
(x + x2 − 1)0 + (x − x2 − 1)0 = 1,
2 2
1 √ 1 √
(x + x2 − 1)1 + (x − x2 − 1)1 = x.
2 2
Then the sequence {an }n=0 coincides with {Tn (x)}∞
∞
n=0 . Hence

1 √ 1 √
Tn (x) = (x + x2 − 1)n + (x − x2 − 1)n for all n ∈ N0 (3.15)
2 2
and all x ∈ R, |x| > 1.

If we expand (3.15) we obtain

n   n  
1 X n n−k √ 2 k 1 X n n−k √ k
Tn (x) = x x −1 − x (−1)k x2 − 1
2 k=0 k 2 k=0 k
50 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

⌊n/2⌋  
X n
= xn−2k (x2 − 1)k
k=0
2k
⌊n/2⌋  k  
X n X k
n−2k
= x (−1)j x2k−2j
k=0
2k j=0
j
⌊n/2⌋
k   
XX n k
= (−1)j x2n−2j .
k=0 j=0
2k j

The summation is over all pairs (j, k) with 0 ≤ j ≤ k ≤ ⌊n/2⌋. Hence

⌊n/2⌋ ⌊n/2⌋   
X
j 2n−2j
X n k
Tn (x) = (−1) x .
j=0 k=j
2k j

The combinatorial identity (which we will not prove here)

⌊n/2⌋   
X n k (n − j − 1)!
= n2n−2j−1
2k j j!(n − 2j)!
k=j

gives finally the explicit representation

⌊n/2⌋
n X (n − j − 1)!
Tn (x) = (−1)j (2x)n−2j . (3.16)
2 j=0 j!(n − 2j)!

The Chebyshev polynomial Tn is an orthogonal degree n polynomial for a

suitable inner product, which we will derive now.

We remember from analysis

π if m = 0
(
Z π
cos(mt) dt = 1 π
sin(mt) = 0 if m 6= 0

0
m 0

Hence by a triginometric identity

Z π
1 π 1 π
Z Z
cos(nt)cos(kt) dt = cos((n + k)t) dt + cos((n − k)t) dt
0 2 0 2 0

 π if k = n = 0
π
= if k = n 6= 0
 2
0 if k 6= n
3.2. THE CHEBYSHEV POLYNOMIALS (FIRST KIND) 51
√
Therefore the substitution x = cos(t) ( ⇒ sin(t) = 1 − x2 ) gives

Z 1
1  π if k = n = 0
π
Tn (x)Tk (x) √ dx = 2
if k = n 6= 0
1−x2
−1 
0 if k 6= n

Thus {Tn } is an orthogonal sequence for the inner product

1
1
Z
hf, gi = f (x)g(x) √ dx
−1 1 − x2

and {vn }∞
n=0 is an ONS for the same inner product where

r
1 2
v0 := √ , vn := Tn n ≥ 1.
π π

Remark. The representation (3.14) shows |Tn (x)| ≤ 1 for all x ∈ [−1, 1] and
allows to determine the n zeros ξk of Tn explicitely,

2k − 1 2k − 1
ξk = cos( π) ⇒ Tn (ξk ) = cos( π) = 0, k = 1, . . . , n,
2n 2

and also all extremal points η ∈ [−1, 1], i.e., those with |Tn (η)| = 1,

kπ
ηk = cos( ) ⇒ Tn (ηk ) = cos(kπ) = (−1)k , k = 0, . . . , n.
n

For obtaining a generating function for {Tn (x)}∞n=0 we need to reorder the
power series
X∞  
Tk (x) − 2xTk−1 (x) + Tk−2 (x) z k
k=2

(although this series is 0). The power series converges absolutely for |x| ≤ 1
and |z| < 1 because for such values of x and z

|(Tk (x) − 2xTk−1 (x) + Tk−2 (x))z k | ≤ (1 + 2 + 1)|z|k ≤ 4|z|k

P∞ k 4|z|2
and k=2 4|z| = 1−|z|
< ∞.
52 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

Therefore we can write

∞ 
X 
0 = (T0 − 1) + (T1 (x) − x)z + Tk (x) − 2xTk−1 (x) + Tk−2 (x) z k
k=2
∞
X X∞ ∞
X
k k−1 2
= Tk (x)z − 1 − xz − 2xz Tk−1 z +z Tk−2 z k−2
k=0 k=2 k=2
∞
X ∞
X ∞
X
= Tk (x)z k − 1 − xz − 2xz Tk z k + z 2 Tk z k
k=0 k=1 k=0
∞
X  
= Tk (x)z k 1 − 2xz + z 2 − 1 − xz + 2xz.
k=0

Hence we obtain the generating function for {Tn (x)}∞

n=0

∞
1 − xz X
= Tk (x)z k .
1 − 2xz + z 2 k=0

There are many interesting and even fascinating results on Chebyshev poly-
nomials (see the Exercises for examples). A good monograph devoted to
these polynomials and their properties is
Th. J. Rivlin: The Chebyshev Polynomials. John Wiley Sons, Toronto 1974.
We will just mention one property, which shows, that Chebyshev polynomi-
als play an important rôle in approximation theory.

Theorem. Consider the space C[−1, 1] with norm kf k∞ := max−1≤x≤1 |f (x)|.

Then 0 is the best approximation to Tn in Pn−1 ,

1 = kTn k∞ ≤ kTn − pk∞ ∀p ∈ Pn−1 .

Since Tn has leading coefficient 2n−1 for n ≥ 1, this theorem gives

21−n = minp∈Pn−1 k21−n Tn − pk∞

n−1
X
n
= mina0 ,...,an−1 ∈R kx − ak xk k∞
k=0

This shows, that the function xn is approximated best in Pn−1 by the poly-
nomial xn − 21−n Tn (x) ∈ Pn−1 and the approximation error is then 21−n .
3.3. THE CHEBYSHEV POLYNOMIALS (SECOND KIND) 53

3.3 The Chebyshev polynomials (second kind)

We start with the trigonometric identity

sin((n + 1)t) = sin(nt)cos(t) + cos(nt)sin(t).

We divide by sin(t) and let x = cos(t). This gives

sin((n + 1)t) sin(nt)

=x + Tn (x).
sin(t) sin(t)
√
Because of sin(t) = 1 − x2 if x = cos(t) and t ∈ [0, π], we obtain therefore
in analogy to the Chebyshev polynomials of the first kind the Chebyshev
polynomials of the second kind
sin((n + 1) arccos(x))
Un (x) := √ ,
1 − x2
which satisfy

Un (x) = xUn−1 (x) + Tn (x) for all n ∈ N. (3.17)

First of all, we have to show, that Un is a polynomial (of degree n). For
n = 0 we have by definition
sin(t)
U0 (x) = = 1 ( cos(t) = x ).
sin(t)

Then (3.17) gives successively

U1 (x) = xU0 (x) + T1 (x) = 2x,

U2 (x) = xU1 (x) + T2 (x) = 2x2 + 2x2 − 1 = 4x2 − 1,
U3 (x) = xU2 (x) + T3 (x) = 4x3 − x + 4x3 − 3x = 8x3 − 4x,
U4 (x) = xU3 (x) + T4 (x) = 8x4 − 4x2 + 8x4 − 8x2 + 1 = 16x4 − 12x2 + 1,
etc.

By induction, one can show easily that Un is a polynomial of degree n with

lc(Un ) = 2n for all n ∈ N0 and, since Tn (−x) = (−1)n Tn (x) the same holds
for Un . In other words, Un is an even polynomial if n is even, and an odd
one if n is odd.

Lemma. Un+1 (x) = 2xUn (x) − Un−1 (x) for all n ∈ N.

54 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

Proof. We sum up the following two trigonometric equalities

sin((n + 1)t) = sin(nt)cos(t) + cos(nt)sin(t)
sin((n − 1)t) = sin(nt)cos(t) − cos(nt)sin(t)
and divide by sin(t). The substitution x = cos(t) gives
Un (x) + Un−2 (x) = 2xUn−1 (x).
Subtracting Un−2 (x) on both sides and renumbering gives the assertion. 

The lemma shows that {Tn (x)}∞ ∞

n=0 and {Un (x)}n=0 satisfy the same three
term recursion formula. The polynomial systems differ by the initial condi-
tions
T0 (x) = 1, U0 (x) = 1,
T1 (x) = x, U1 (x) = 2x.
This tiny difference only in the definition of the linear polynomials causes
the difference of the two systems!

There are many connections between first and second kind polynomials. For
instance, if we dismantle xUn (x) in the three term recursion and make and
index transformation,
1 1
xUn−1 (x) = Un (x) + Un−2 (x),
2 2
then inserting into (3.17) and dismantling for Tn gives
1 1
Tn = Un − Un−2 . (3.18)
2 2
An other nice relation between the two systems is obtained by differentiating
Tn ,
Tn′ (x) = nUn−1 (x). (3.19)
Here we used x = cos(t) and
d cos(nt) dt 1 sin(nt)
= −nsin(nt) =n .
dt dx d cos(t) sin(t)
dt
We can estimate easily Un and obtain by (3.19) an estimate for Tn′ . If x =
cos(t), then
sin((n + 1)t) ei(n+1)t − e−i(n+1)t
Un (x) = =
sin(t) eit − e−it
Xn Xn
i(n−k)t −ikt
= e e = ei(n−2k)t .
k=0 k=0
3.3. THE CHEBYSHEV POLYNOMIALS (SECOND KIND) 55

Hence
n
X
|Un (x)| ≤ |ei(n−2k)t | ≤ n + 1 for all x ∈ [−1, 1]. (3.20)
k+0

and for t → 0, i.e., x → 1,

n
X
Un (1) = ei(n−2k)0 = n + 1.
k=0

We combine (3.19) and (3.20)

|Tn′ (x)| = n|Un−1 (x)| ≤ |Tn′ (1)| = nUn−1 (1) = n2 ∀ x ∈ [−1, 1]. (3.21)

Remark. By V.A.Markov the inequality

kf ′k∞ ≤ n2

holds true, if f ∈ Pn and kf k∞ := max−1≤x≤1 |f (x)| ≤ 1. This upper bound

can not be improved as (3.21) shows.

For obtaining
P∞ the generating function for {Un (x)}∞ n=0 , we need to reorder the
series n=0 Un (x)z n , i.e., we need the absolute convergence of it for instance
for |x| ≤ 1 and |z| ≤ a < 1. This holds because
∞ ∞
X
n
X 1
|Un (x)||z | ≤ (n + 1)an = < ∞.
n=0 n=0
(1 − a)2

By similar arguments, we may reorder

∞
X
0 = (Un (x) − 2xUn−1 (x) + Un−2 )z n
n=2
∞
X ∞
X ∞
X
n n−1 2
= Un (x)z − 2xz Un−1 (x)z +z Un−2 (x)z n−2
n=2 n=2 n=2
X∞  
= Un (x)z n 1 − 2xz + z 2 − U0 (x)z 0 − U1 (x)z 1 + 2xzU0 (x)z 0
n=0

Hence
∞
1 X
= Un (x)z n . (3.22)
1 − 2xz + z 2 n=0
56 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

We will now show the orthogonality of the Chebyshev polynomials (second

kind). For sake of completeness, we formulate the following results also for
the Chebyshev polynomials of the first kind.

Theorem 3.6. If the inner product is defined by

Z 1
1
hf, gi := f (x)g(x) √ dx,
−1 1 − x2
q q q
then { √1π , π2 T1 , π2 T2 , π2 T3 , . . .} is an ONS, where Tn denotes the n-th
Chebyshev polynomial of the first kind. If the inner product is defined by
Z 1 √
hf, gi := f (x)g(x) 1 − x2 dx,
−1
q
then { π2 Un }∞
n=0 is an ONS, where Un denotes the n-th Chebyshev polyno-
mial of the second kind.

Proof. The statement for the Tn is already proved before. We have to show
only Z 1 √

2 0 if m 6= n
Un (x)Um (x) 1 − x dx = π
−1 2
if m = n
√
The substitution x = cos(t) ( ⇒ sin(t) = 1 − x2 ) and a trigonometric
identity gives
R1 √ R π sin((n + 1)t) sin((m + 1)t)
U (x)Um (x) 1 − x2 dx = 0
−1 n
sin(t)2 dt
Rπ Rπ sin(t) sin(t)
= 21 0 cos((n − m)t) dt − 21 0 cos((n + m + 2)t) dt.
Rπ
Because of 0 cos(at)dt = π if a = 0 and = a1 (sin(aπ) − sin(0) = 0 if
a ∈ Z \ {0}, we obtain
Z 1 √

2 0 if m 6= n,
Un (x)Um (x) 1 − x dx = π
−1 2
if m = n.
q
2
Hence {Un }∞n=0 is an orthogonal system and { U }∞ an ONS.
π n n=0


The zeros of Un are the inner extremal point of Tn+1 in [−1, 1] because of
′
Tn+1 = (n + 1)Un , i.e.,
kπ
Un (cos( )) = 0, k = 1, . . . , n,
n+1
3.4. THE JACOBI POLYNOMIALS 57

and since |sin((n + 1) arccos(x)| ≤ 1 we have for −1 < x < 1 in addition to

(3.20) the estimate for −1 < x < 1

1
|Un (x)| ≤ √ . (3.23)
1 − x2

Remark. In analogy to the best approximation property of the polynomial

Tn with respect to the norm kf k∞ := max−1≤x≤1 |f (x)| the Chebyshev poly-
nomial URn has a best approximation property with respect to the norm
1
kf k1 := −1 |f (x)| dx. As shown for instance in the above mentioned book
of Th. J. Rivlin, 0 is the best approximation to Un in Pn−1 with respect to
the norm kf k1 , i.e.,

kUn k1 ≤ kUn − pk1 for all p ∈ Pn−1 .

3.4 The Jacobi polynomials

We consider now the more general inner product
Z 1
hf, gi := f (x)g(x) (1 − x)α (1 + x)β dx
−1

Here one requires α > −1 and β > −1, because otherwise the inner product
is not defined for all f, g ∈ P.

In the subsections before we have already studied the cases α = β = 0, (the

Legendre polynomials), the case α = β = − 21 , (Chebyshev polynomials of
the first kind), and α = β = 21 , (Chebyshev polynomials of the second kind).
We will not consider separately the case α = β (so called Gegenbauer poly-
nomials or also hyperspheric polynomials), but consider the general case of
Jacobi polynomials for arbitrary α > −1, β > −1.

a


First we need here the binomial coefficients k
for noninteger a ∈ R.

Definition. For a ∈ R and k ∈ N we define

 
a a(a − 1) · · · (a − k + 1)
:= .
k k!

This definition coincides obviously with the known one for k ∈ N.

58 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

An other way of expressing a product a(a − 1) · · · (a − k + 1) is possible by

using the Gamma function
Z ∞
Γ(x) := e−t tx−1 dt for x > 0.
0

The known functional equation Γ(x + 1) = x · Γ(x) gives

Γ(a + 1)
a(a − 1) · · · (a − k + 1) = .
Γ(a − k + 1)
Since Γ(1) = 1 (implying Γ(k + 1) = k!) one can also write
 
a Γ(a + 1)
=
k Γ(k + 1)Γ(a − k + 1)
for a > −1 and k
∈ N. This identity can also be used for defining the bino-
mial coefficient ka for arbitrary real k with −1 < k < a + 1.

Starting point is the Rodrigues formula

dn  
Jn(α,β) (x) = Kn · (1 − x) −α
(1 + x) −β α+n
(1 − x) (1 + x) β+n
.
dxn
(α,β)
We show first that Jn is a polynomial of degree n. By direct calculation
dnn (1 − x)α+n (1 + x)β+n
 
dx
α+n−k β + n

α + n

= nk=0 n k β+k
P

k k k!(−1) (1 − x) n − k (n − k)!(1 + x)
= (1 − x)α (1 + x)β nk=0 n! α+n

β+n

(−1)k (1 − x)n−k (1 + x)k
P
k n−k

we see that
n   
X α+n β+n
Jn(α,β) (x) = n!Kn (−1) n
(−1)n−k (1 − x)n−k (1 + x)k
k n−k
k=0
(3.24)
and its leading coefficient is
n   
X α+n β+n
lc(Jn(α,β) ) = n!Kn (−1) n
. (3.25)
k=0
k n − k

In the Rodrigues formula, we needed the n-th derivative of ϕ,

ϕ(x) := (1 − x)α+n (1 + x)β+n .

3.4. THE JACOBI POLYNOMIALS 59

Its lower order derivatives are (with 0 ≤ m < n)

m   
(m)
X α+n β+n
ϕ (x) = m! (−1)k (1 − x)α+n−k (1 + x)β+n−m+k .
k=0
k m − k

Each of the m + 1 summands is zero in 1 and in −1 because of

α + n − k ≥ α + n − m ≥ α + 1 > 0,
β + n − m + k ≥ β + n − m ≥ β + 1 > 0.

Hence ϕ(m) is zero in ±1. Therefore the inner product

Z 1
(α,β)
hq, Jn i = q(x)Jn(α,β) (1 − x)α (1 + x)β dx
−1

can be transformed by partial integration to

Z 1
(α,β)
hq, Jn i = Kn q(x)ϕ(n) (x) dx
−1
Z 1
= −Kn q ′ (x)ϕ(n−1) (x) dx
−1
= ...
Z 1
n
= (−1) Kn q (n) ϕ(x) dx
−1

Especially if q ∈ Pn−1 , then

hq, Jn(α,β) i = 0.

Theorem 3.8. Every orthogonal polynomial of degree n for the inner product
Z 1
hf, gi := f (x)g(x) (1 − x)α (1 + x)β dx
−1

is of type
n   
X α+n β+n
Jn(α,β) (x) = n!Kn (−1)n
(−1)n−k (1 − x)n−k (1 + x)k .
k=0
k n − k

α+β+2n −1 (α,β)
If Kn = (−1) (−1)n −n
n

n!
n
then lc(J n ) = 1. If K n := n!
2 then
(α,β) α+n
Jn (1) = n . If
s
Γ(α + β + n + 1) 2−α−β−2n−1
Kn = (α + β + 2n + 1)
Γ(α + n + 1)Γ(β + n + 1) n!
60 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

(α,β)
then Jn is orthonormal.
(α,β)
Proof. A representation for lc(Jn ) is already shown in (3.25),
n   
(α,β) n
X α β
lc(Jn ) = Kn n!(−1) .
k=0
k n−k

In Analysis one shows for real a, b, and |x| < 1

∞   ∞  
a
X a k b
X b n
(1 + x) = x and (1 + x) = x .
k=0
k n=0
n

Therefore
∞ X n    ∞  
α+β
X α β n
X α+β n
(1 + x) = x = x .
n=0 k=0
k n − k n=0
n

Comparing the coefficients of the power series gives

n     
X α β α+β
= .
k=0
k n − k n

Hence  
α+β
lc(Jn(α,β) ) = Kn n!(−1) n
.
n
(α,β)
This gives the representation for lc(Jn ) = 1.

Direct computation shows

 
a+n n
Jn(α,β) (1) = Kn n!(−1) n
2 .
n
(α,β)
The polynomial Jn with
Jn(α,β) (x) = cn xn + lower degree terms
is orthogonal. Hence
hJn(α,β) , Jn(α,β) i = cn hxn , Jn(α,β) i.
Therefore
Z 1
hJn(α,β) , Jn(α,β) i = cn xn Jn(α,β) (x)(1 − x)α (1 + x)β dx
−1
Z 1 n 
n d

α+n β+n
= c n Kn x (1 − x) (1 + x) dx
−1 dxn
3.4. THE JACOBI POLYNOMIALS 61

and by partial integration and by using a standard integral formula (Beta

function)
Z 1
(α,β) (α,β) n
hJn , Jn i = cn Kn (−1) n! (1 − x)α+n (1 + x)β+n dx
−1
Γ(α + n + 1)Γ(β + n + 1) α+β+2n+2
= cn Kn (−1)n n! 2 .
Γ(α + β + 2n + 2)
We substitute now the leading coefficient
 
(α,β) n α + β + 2n Γ(α + β + 2n + 1)
cn = lc(Jn ) = Kn n!(−1) = Kn (−1)n
n Γ(α + β + n + 1)
and obtain finally
Γ(α + n + 1)Γ(β + n + 1) 1
hJn(α,β) , Jn(α,β) i = Kn2 n! 2α+β+2n+2 .
Γ(α + β + n + 1) α + β + 2n + 1
Hence Kn chosen as in the assertion gives the orthonormal polynomial. 

Finally, we summarize without proof the differential equation, the three term
recursion and the generating function for the Jacobi polynomials. (The proofs
are along the same scheme as the proofs shown in the preceding subsections.)
(α,β)
The Jacobi polynomial Jn solves the differential equation
 
(1 − x )y + β − α − (α + β + 2)x y ′ + n(n + α + β + 1)y = 0.
2 ′′
(3.26)

(α,β) n+α


Let Pn denote the Jacobi polynomial Jn normed by Pn (1) = n
. Then
Pn+1 (x) = (An x + Bn )Pn (x) − Cn Pn (x), n ∈ N, (3.27)
with
(2n + α + β + 1)(2n + α + β + 2)
An = ,
2(n + 1)(n + α + β + 1)
(2n + α + β + 1)(α2 − β 2 )
Bn = ,
2(n + 1)(n + α + β + 1)(2n + α + β)
(n + α)(n + β)(2n + α + β + 2)
Cn = ,
(n + 1)(n + α + β + 1)(2n + α + β)
α+β α−β
and P0 = 1, P1 (x) = (1 + 2
)x + 2
.
√
With the same normalization and with R(x, z) := 1 − 2xz + z 2
∞
2α+β X
= Pn (x)z n . (3.28)
R(x, z)(1 − z + R(x, z))α (1 + z + R(x, z))β
k=0
62 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

3.5 The Laguerre polynomials

The inner product is here
Z ∞
hf, gi := f (x)g(x) e−x dx.
0

The Rodrigues’ formula

dn n −x
Ln (x) := Kn ex (x e ) (3.29)
dxn
defines a polynomial Ln of degree n, because by Leibniz’ formula
n  
dn n −x X n n!
n
(x e ) = xn−k (−1)n−k e−x
dx k=0
k (n − k)!

and hence Ln has the explicit representation

n   
X n n
Ln (x) = Kn k!(−1)n−k xn−k . (3.30)
k=0
k k

The summand for k = 0 gives the leading coefficient

lc(Ln ) = Kn (−1)n .

The first n derivatives of ϕ,

ϕ(x) = xn e−x

are ϕ(n) (x) = Kn−1 e−x Ln (x) and

m  
(m)
X m n!
ϕ (x) = (−1)n−k xn−k e−x , 0 ≤ m < n.
k=0
k (n − k)!

Especially for 0 ≤ m < n

ϕ(m) (0) = 0 and ϕ(m) (x) → 0 for x → ∞.

Therefore by partial integration

Z ∞ Z ∞
−1 −1 −x
Kn hq, Ln i = Kn q(x)Ln (x)e dx = q(x)ϕ(n) (x)dx
Z ∞0 0
Z ∞
′ (n−1) n
= − q (x)ϕ (x)dx = . . . = (−1) q (n) (x)ϕ(x)dx.
0 0
3.5. THE LAGUERRE POLYNOMIALS 63

Hence
hq, Ln i = 0 if q ∈ Pn−1 ,
i.e., Ln is orthogonal of degree n, and if q = Ln (= Kn (−1)n xn + lower order
terms) then
Z ∞
−1 n n
Kn hLn , Ln i = (−1) hx , Ln i = n!ϕ(x)dx = n!Γ(n + 1) = (n!)2 .
0

Consequently

Kn = (−1)n ⇒ lc(Ln ) = 1,
1
Kn = (−1)n ⇒ hLn , Ln i = 1, lc(Ln ) > 0.
n!

For the Laguerre polynomials with leading coefficient 1, i.e. Kn = (−1)n ,

computation gives

L0 = 1, L1 = −ex (xe−x )′ = x − 1.

The three term recursion is

 
Ln+1 (x) = x − (2n + 1) Ln (x) − n2 Ln−1 (x). (3.31)

Hence
L2 (x) = (x − 3)L1 (x) − L0 (x) = x2 − 4x + 2
L3 (x) = (x − 5)L2 (x) − 4L1 (x) = x3 − 9x2 + 18x − 6
L4 (x) = (x − 7)L3 (x) − 9L2 (x) = x4 − 16x3 + 72x2 − 96x + 24
L5 (x) = (x − 9)L4 − 16L3 = x5 − 25x4 + 200x3 − 600x2 + 600x − 120
etc.

The generating function for the orthonormalized Laguerre polynomials, i.e.,

Kn = (−1)n /n!, is
∞
1 −xz X
exp( )= Ln (x)z n . (3.32)
1+z 1+z n=0

The second order differential equation solved by y = Ln is

xy ′′ + (1 − x)y ′ + ny = 0.
64 CHAPTER 3. SPECIAL ORTHOGONAL FAMILIES

If one introduces an additional factor xα , α > 0, into the inner product,

Z ∞
hf, giα := f (x)g(x) xα e−x dx, α > 0,
0

(α)
then one obtains the so called generalized Laguerre polynomial Ln . This
polynomial satisfies the Rodrigues formula

dn n+α −x
L(α)
n (x) = Kn x −α x
e (x e ), (3.33)
dxn

is orthogonal (with respect to the inner product h , iα ), has the explicit

representation
n  
X n+α 1
L(α)
n (x) = n!Kn (−x)n−k , (3.34)
k=0
k (n − k)!

and therefore
lc(L(α) n
n ) = Kn (−1) .

If Kn = (−1)n , then the three term recurrence is

 
(α) (α)
Ln+1 (x) = x − (2n + 1 − α) L(α)
n (x) − n(n + α)Ln−1 (x), n ∈ N. (3.35)

(α) (α)
The initial values are here L0 = 1 and L1 (x) = x − (1 + α).
(α)
If we choose Kn = (−1)n (n!Γ(n + α + 1))−1/2 , then Ln is, as computation
shows, orthonormal with positive leading coefficient.

The generating function for the generalized Laguerre polynomials with lead-
ing coefficient 1 is

 1 α+1 ∞
xz X 1 (α)
exp(− )= Ln (x)z n (3.36)
1+z 1+z n=0
n!

(α)
and the second order differential equation solved by y = Ln is

xy ′′ + (α + 1 − x)y ′ + ny = 0. (3.37)
3.6. THE HERMITE POLYNOMIALS 65

3.6 The Hermite polynomials

The inner product Z ∞
2
hf, gi := f (x)g(x) e−x dx
−∞

leads to the so called Hermite polynomials Hn . They are defined by the

Rodrigues’ formula

n −x2 dn −x2
Hn (x) = (−1) e (e ) (3.38)
dxn
and have the explicit representation
⌊n/2⌋
X 1
Hn (x) = n! (−1)k (2x)n−2k . (3.39)
k=0
k!(n − 2k)!

They are orthogonal with respect to the given inner product and satisfy

lc(Hn ) = 2n .

Computation gives √
hHn , Hn i = π2n n!
p√ −1
Therefore the polynomials π2n n! Hn are orthonormal (with positive
leading coefficient).

Beginning with H0 = 1 and H1 (x) = 2x, the remaining Hermite polynomials

can easily be computed by the three term recurrence formula

Hn+1 (x) = 2xHn (x) − 2nHn (x), n ∈ N. (3.40)

y = Hn solves the differential equation

y ′′ − 2xy ′ + 2ny = 0.

The generating function for the Hermite polynomials is

∞
X
2 1
exp(2xz − z ) = Hn (x)z n . (3.41)
n=0
n!