LEGENDRE POLYNOMIALS - ORTHOGONALITY

Link to: physicspages home page.

To leave a comment or report an error, please use the auxiliary blog and
include the title or URL of this post in your comment.
Post date: 28 May 2021.
In another post, we found that the Legendre polynomials could be written
as an explicit sum as follows

[n/2]
(2n − 2k)!
Pn (x) = ∑ (−1)k xn−2k (1)
k=0 2n (n − k)!k!(n − 2k)!
This can be written as a derivative if we observe that if k ≤ [n/2]

dn 2n−2k (2n − 2k)! n−2k

x = x (2)
dxn (n − 2k)!
Note that if k > [n/2], the nth derivative of x2n−2k is zero, since after
[n/2] derivatives, the term x2n−2k is reduced to a constant. We can use this
fact to rewrite the sum above as

[n/2]
(−1)k dn 2n−2k
Pn (x) = ∑ x (3)
k=0 2n (n − k)!k! dxn
n
(−1)k dn 2n−2k
= ∑ n n
x (4)
k=0 2 (n − k)!k! dx
1 n (−1)k n! dn 2n−2k
= ∑ (n − k)!k! dxn x (5)
2n n! k=0
1 dn n (−1)k n! 2n−2k
= ∑ (n − k)!k! x (6)
2n n! dxn k=0
2 1)n (since the
 −
The sum in the last line is the binomial expansion of (x
n
factorials within the sum form the binomial coefficient ), so we can
k
write this as

1 dn 2
Pn (x) = (x − 1)n (7)
2n n! dxn
1
 LEGENDRE POLYNOMIALS - ORTHOGONALITY 2

This is known as the Rodrigues formula for Legendre polynomials. Al-

though it’s not all that convenient for calculating the polynomials them-
selves, it can be used to prove various properties about them. One of the
most important theorems is that the polynomials are orthogonal. This means
that if n 6= m, we have

Z 1
Pm (x)Pn (x)dx = 0 (8)
−1

This property turns out to be of vital importance in quantum mechanics,

where the polynomials form the basis of the associated Legendre functions,
which in turn form part of the solution of the three-dimensional Schrödinger
equation. We’ll run through the proof here.
Using the Rodrigues formula, we have

Z 1 m
dn 2
Z 1
1 d
Pm (x)Pn (x)dx = (x2 −1)m (x −1)n dx (9)
−1 2m+n m!n! −1 dx m dxn
Before doing the integration, we note that all derivatives of the function
(x2 − 1)m up to the (m − 1)th derivative have x2 − 1 as a factor (we can
see this by applying the chain rule), and are therefore zero at x = ±1. Now
assume that m < n for the purposes of being definite (it won’t matter if
m > n since we can just swap the two indexes throughout the argument).
If we integrate 9 by parts we get

Z 1 m
d dn 2 dn−1 2 n d
m 1
m
(x2 − 1)m (x − 1)n
dx = (x − 1) (x2 − 1)m −
−1 dx dxn dxn−1 dxm −1
(10)
Z 1 m+1
d dn−1 2
m+1
(x2 − 1)m (x − 1)n dx
−1 dx dxn−1
(11)

Because of the condition just stated, the boundary term at the start is zero,
so we can continue by integrating the remaining integral by parts, throwing
away the boundary term until we have done n integrations. At this point we
will have

dm+n 2
Z 1 Z 1
1
Pm (x)Pn (x)dx = m+n (−1)n (x2 − 1)n (x − 1)m dx
−1 2 m!n! −1 dxm+n
(12)
 LEGENDRE POLYNOMIALS - ORTHOGONALITY 3

Since m < n, the derivative inside the integral is zero, since the largest
power of x in (x2 − 1)m is x2m and 2m < m + n. Therefore, the over-
all integral is zero, and we have shown that the Legendre polynomials are
orthogonal (that is, 8 is true).
What if n = m? In that case, the integration by parts technique won’t
work, since we can’t count on the final integral being zero. However, after
n integrations by parts, we get to the formula

(−1)n d2n 2
Z 1 Z 1
Pn (x)Pn (x)dx = (x2 − 1)n (x − 1)n dx (13)
−1 (n!)2 22n −1 dx2n
The derivative inside the integral will kill off all terms within (x2 − 1)n
except for the highest power of x2n and the derivative of that is

d2n 2 n d2n 2n
(x − 1) = x (14)
dx2n dx2n
= (2n)! (15)
We can therefore write

(−1)n (2n)!
Z 1 Z 1
Pn (x)Pn (x)dx = (x2 − 1)n dx (16)
−1 (n!)2 22n −1
The final integral can be done by using a trigonometric substitution (e.g.
x = sin θ so x2 − 1 = − cos θ and dx = cos θdθ). This still requires the
integration of a high power of cos θ so we can take the easy way out and
use mathematical software such as Maple or Mathematica to do the integral
directly. In that case we find
√
(−1)n Γ(n + 1) π
Z 1
2 n
(x − 1) dx = (17)
−1 Γ(n + 3/2)
The gamma function Γ(x) is a generalization of the factorial function and
has a number of convenient properties we can use to simplify this (we’ll
leave the derivation of these properties to another post). In particular, we
have

Γ(n + 1) = n! (18)
√
Γ(z)Γ(z + 1/2) = 21−2z πΓ(2z) (19)
The second of these formulas (known as the duplication formula) can be
used to show that
 LEGENDRE POLYNOMIALS - ORTHOGONALITY 4

√
π(2n + 1)!
Γ(n + 3/2) = Γ((n + 1) + 1/2) = (20)
21+2n n!
so plugging this into 17 we find that

(n!)2 21+2n
Z 1
(x2 − 1)n dx = (−1)n (21)
−1 (2n + 1)!

so plugging this back into 16 we get finally

Z 1 2 1+2n
1 2n (n!) 2
Pn (x)Pn (x)dx = (2n)!(−1) (22)
−1 (n!)2 22n (2n + 1)!
2
= (23)
2n + 1
We can combine the results on integration of the Legendre polynomials
to get the overall orthogonality condition:
Z 1
2
Pn (x)Pm (x)dx = δnm (24)
−1 2n + 1
P INGBACKS
Pingback: Laplace’s equation in spherical coordinates - surface charge
Pingback: Associated Legendre functions - orthogonality