You are on page 1of 6

Power series solutions to the Legendre

Equation and the Legendre polynomials

In the last few lectures on series solutions, we had mentioned the Legendre Equation quite a
few times while dealing with some examples. The aim of this lecture is to get a fresh perspective
on the Legendre equations, with special emphasis to the cases when they admit polynomial
solutions. These polynomials are called the Legendre polynomials. These polynomials have
some very interesting properties, due to which they are of significant practical importance.
Definition 1. The equation
(1 − x2 )y 00 − 2xy 0 + α(α + 1)y = 0, (1)
where α is any real constant, is called the Legendre equation. 
Remark 1. Solving the Legendre equation is equivalent to solving an eigen value problem
associated to a linear operator.
The Legendre equation can be rewritten as:
(x2 − 1)y 00 (x) + 2xy 0 (x) = α(α + 1)y(x)
d  2
(x − 1)y 0 (x) = α(α + 1)y(x).

⇒ (∗)
dx
Recall now that C 2 (R) is a vector space with respect to usual addition and scalar multiplication.
Consider the operator T : C 2 (R → C 2 (R) given by
 
d 2 dy
T (y) := (x − 1) ∀y ∈ C 2 (R).
dx dx
Verify that T is a linear operator on C 2 (R) (exercise!). It is now easy to see that equation (∗)
is equivalent to:
T (y(x)) = α(α + 1)y(x).
In other words, equation (∗) now looks like the equation T v = λv where v = y(x) and
λ = α(α + 1). This reminds us of an eigenvalue problem!
Hence, finding a non-zero solution of the Legendre equation is equivalent to finding an eigen-
vector v of T with respect to the eigenvalue α(α + 1). 
A general solution of the Legendre equation is a linear combination (a0 y1 (x) + a1 y2 (x)) of
two power series solutions y1 (x) and y2 (x) of the equation (about the ordinary point 0), where

X α(α − 2) · · · (α − 2n + 2) · (α + 1)(α + 3) · · · (α + 2n − 1) 2n
y1 (x) = 1 + (−1)n x ,
n=1
(2n)!

and

X (α − 1)(α − 3) · · ·(α − 2n + 1) · (α + 2)(α + 4) · · · (α + 2n) 2n+1
y2 (x) = x + (−1)n x .
n=1
(2n + 1)!

1
By the ratio test, the radii of convergence of the series y1 (x) and y2 (x) are both equal to 1. These
solutions y1 (x) and y2 (x) also satisfy the initial conditions

y1 (0) = 1, y10 (0) = 0, y2 (0) = 0, y20 (0) = 1.


So the Wronskain W (y1 , y2 )(0) = 1 and hence {y1 (x), y2 (x)} is a fundamental solution set
and the general solution of the Legendre equation over (−1, 1) is

y(x) = a0 y1 (x) + a1 y2 (x)


with arbitrary constants a0 and a1 .
Let us now look into the expressions for y1 (x) and y2 (x) more closely. Observe that if α = 0,
then y1 (x) = 1. And if α is a non-zero even integer (say, α = 2m), then in the expression for
y1 (x) above, all terms (inside the summation) for n ≥ m + 1 are zero. So, if α = 2m (for some
non-negative integer m), then y1 (x) is a polynomial of degree 2m. Similarly, if α = 2m + 1
(for some non-negative integer m), then in the expression for y2 (x) above, all terms (inside the
summation) for n ≥ m + 1 are zero. In that case, y2 (x) will be a polynomial of degree 2m + 1.
However, note that if α = 0 or 2m, then y2 (x) is not a polynomial. And if α = 2m + 1, then
y1 (x) is not a polynomial. In other words, if α is a non-negative integer, then exactly one of
y1 (x) and y2 (x) is a polynomial. Let us now take a closer look into these polynomials when α
takes non-negative integral values.
Case I: Let α = 0 or α = 2m for some integer m ≥ 1.
We have
m
X α(α − 2) · · · (α − 2n + 2) · (α + 1)(α + 3) · · · (α + 2n − 1) 2n
y1 (x) = 1 + (−1)n x .
n=1
(2n)!

Clearly y1 (x) = 1 when α = 0, . Let α = 2m, where m ∈ N. Then ∀n ∈ {1, . . . , m}, we have

α(α − 2) · · · (α − 2n + 2) = 2m(2m − 2) · · · (2m − 2n + 2)

m!
= 2n m(m − 1) · · · (m − n + 1) = 2n .
(m − n)!
And
(α + 1)(α + 3) · · · (α + 2n − 1) = (2m + 1)(2m + 3) · · · (2m + 2n − 1)
(2m)!(2m + 1)(2m + 2)(2m + 3)(2m + 4) · · · (2m + 2n − 1)(2m + 2n)
=
(2m)!(2m + 2)(2m + 4) · · · (2m + 2n)
(2m + 2n)!m! (2m + 2n)!m!
= = n .
2n (2m)!(m + 1)(m + 2) · · · (m + n)m! 2 (2m)!(m + n)!
If we denote the y1 (x) by y1,2m (x) when α = 2m, then we see that
m
X (−1)n 2n (m!)2 (2m + 2n)!
y1,2m (x) = 1 + x2n
n=1
2n (2n)!(2m)!(m + n)!(m − n)!

m
(m!)2 X (−1)n (2m + 2n)!
=1+ x2n .
(2m)! n=1 (2n)!(m + n)!(m − n)!
In particular, for α = 0, 2, 4 (m = 0, 1, 2), the corresponding polynomials are
35 4
y1,0 (x) = 1, y1,2 (x) = 1 − 3x2 , y1,4 (x) = 1 − 10x2 + x.
3

2
Case II: Let α = 2m + 1 for some integer m ≥ 0.
We have
m
X (α − 1)(α − 3) · · ·(α − 2n + 1) · (α + 2)(α + 4) · · · (α + 2n) 2n+1
y2 (x) = x + (−1)n x .
n=1
(2n + 1)!

Then ∀n ∈ {1, . . . , m}, we have

2n m!
(α − 1)(α − 3) · · · (α − 2n + 1) = (2m)(2m − 2) · · · (2m − 2n + 2) = ,
(m − n)!

and
(α + 2)(α + 4) · · · (α + 2n) = (2m + 3)(2m + 5) · · · (2m + 2n + 1)
(2m + 1)!(2m + 2)(2m + 3)(2m + 4)(2m + 5) · · · (2m + 2n)(2m + 2n + 1)
=
(2m + 1)!(2m + 2)(2m + 4) · · · (2m + 2n)
(2m + 2n + 1)!m!
= .
2n (2m + 1)!(m + n)!
If we denote the y2 (x) by y2,2m+1 (x) when α = 2m + 1, then we get
m
(m!)2 X (−1)n (2m + 2n + 1)!
y2,2m+1 (x) = x + x2n+1 .
(2m + 1)! n=1 (2n + 1)!(m + n)!(m − n)!

For example, when α = 1, 3, 5 (m = 0, 1, 2), the polynomials are


5 14 21
y2,1 (x) = x, y2,3 (x) = x − x3 , y2,5 (x) = x − x3 + x5 .
3 3 5
These polynomial solutions are scaled versions of Legendre polynomials. If we multiply
these polynomials by some numbers, then we will get the Legendre polynomials. Now the
following natural question arises:
Question: Why do we need to go to the Legendre polynomials? Why these scaled versions are
not enough for us?
Answer: In the above discussion, we have obtained 2 kinds of polynomials. We want to unify
the way of writing these two kinds of polynomials into a single format.

To obtain a single formula which contains both the polynomials in y1 (x) and y2 (x), let
[n/2]
1 X (−1)r (2n − 2r)! n−2r
Pn (x) = n x ,
2 r=0 r!(n − r)!(n − 2r)!

where [n/2] denotes the greatest integer ≤ n/2. [Note here that [n/2] = m when n = 2m and
n = 2m + 1 both.]

Theorem 1. Let fn (x) be the polynomial solution of the Legendre equation when α is a non
negative integer n. Then,

(−1)t (t!)2 22t


fn (x) = Pn (x), where t = [n/2].
n!

3
P ROOF : Recall that
[n/2]
1 X (−1)r (2n − 2r)! n−2r
Pn (x) = n x .
2 r=0 r!(n − r)!(n − 2r)!
Case I: When n = 2m.
m
(m!)2 X (−1)n (2m + 2n)!
fn (x) = y1,2m (x) = 1 + x2n
(2m)! n=1 (2n)!(m + n)!(m − n)!
(m!)2 (2m)! (−1)m (4m)! 2m
 
(2m + 2)! 2 (2m + 4)! 4
= − x + x + ··· + x
(2m)! (m!)2 2!(m + 1)!(m − 1)! 4!(m + 2)!(m − 2)! {(2m)!}2
For n = 2m,
m
1 X (−1)r (4m − 2r)!
Pn (x) = x2m−2r
22m r=0
r!(2m − r)!(2m − 2r)!
(4m)!x2m (4m − 2)!x2m−2 (4m − 4)!x2m−4 (−1)m (2m)!
 
1
= − + + ··· + .
22m {(2m)!}2 (2m − 1)!(2m − 2)! 2!(2m − 2)!(2m − 4)! (m!)2
Then we can easily see that
n n
(−1)m (m!)2 22m (−1)[ 2 ] ([ n2 ]!)2 22[ 2 ]
fn (x) = y1,2m (x) = P2m (x) = Pn (x).
(2m)! n!

Case II: When n = 2m + 1.


m
(m!)2 X (−1)n (2m + 2n + 1)!
fn (x) = y2,2m+1 (x) = x + x2n+1
(2m + 1)! n=1 (2n + 1)!(m + n)!(m − n)!
(m!)2 (2m + 3)!x3
 
(2m + 1)!x
= − + ···
(2m + 1)! (m!)2 (m − 1)!(m + 1)!3!
(m!)2 (−1)m−1 (4m − 1)!x2m−1 (−1)m (4m + 1)!x2m+1
 
+ ··· + +
(2m + 1)! {(2m − 1)!}2 (2m)!(2m + 1)!
(−1)m (m!)2 (−1)m (4m + 2)!x2m+1 (−1)m (4m)!x2m−1
 
= + + ···
2(2m + 1)! {(2m + 1)!}2 (2m)!(2m − 1)!
(−1)m (m!)2 (−1)m−1 (2m + 4)!x3 (−1)m (2m + 2)!x
 
+ ··· + +
2(2m + 1)! (m + 2)!(m − 1)!3! (m + 1)!m!
" m #
(−1)m (m!)2 2m+1 1 X (−1)r (4m + 2 − 2r)!
= 2 xn−2r
2(2m + 1)! 22m+1 r=0 r!(2m + 1 − r)!(2m + 1 − 2r)!
n n
(−1)m (m!)2 22m (−1)[ 2 ] ([ n2 ]!)2 22[ 2 ]
= P2m+1 (x) = Pn (x).
(2m + 1)! n!


Definition 2. Pn (x) are solutions of the Legendre equations for α = n and are called the
Legendre polynomials. 

4
1
P0 (x)

P2 (x) P1 (x)
0.5 P3 (x)
P4 (x)
P5 (x)
0

−0.5

−1

−1 −0.5 0 0.5 1

Figure 1: The first 6 Legendre polynomials over the interval [−1, 1]

The graphs of the first six Legendre polynomials P0 (x) = 1, P1 (x) = x, P2 (x) = 21 (3x2 −1),
P3 (x) = 12 (5x3 − 3x), P4 (x) = 18 (35x4 − 30x2 + 3), P5 (x) = 81 (63x5 − 70x3 + 15x), are plotted
in Figure 1. Let us now write the formula for Pn (x) in a more compact form which is easy to
remember. Recall that
[n/2]
1 X (−1)r (2n − 2r)! n−2r
Pn (x) = n x .
2 r=0 r!(n − r)!(n − 2r)!

Note that
dn
 
(2n − 2r)! n−2r 1 1 n
x = n x2n−2r and = .
(n − 2r)! dx r!(n − r)! n! r
Therefore
 
[n/2] n n [n/2]
1 X d 1 d  X
Pn (x) = n (−1)r n Cr n x2n−2r = n (−1)r n Cr x2n−2r  .
2 n! r=0 dx 2 n! dxn r=0

Note that 2n − 2r < n when [ n2 ] < r ≤ n. So when [n/2] < r ≤ n, the term x2n−2r has degree
less than n, so its nth derivative is zero. Thus, Pn (x) above can be expressed as
n
1 dn X
 
r n
Pn (x) = n (−1) x2n−2r .
2 n! dx n r
r=0

This gives (using binomial theorem),


1 dn 2
Pn (x) = (x − 1)n ,
2n n! dxn
which is known as Rodrigues’s formula. The Legendre polynomials have some very interesting
properties as listed below:
• For each n ≥ 0, Pn (1) = 1. This is illustrated in Figure 1 for the first six Legendre poly-
nomials. Moreover, Pn (x) is the only polynomial which satisfies the Legendre equation

(1 − x2 )y 00 − 2xy 0 + n(n + 1)y = 0

and the extra condition y(1) = 1. In other words, Pn (x) is the unique solution of the
above IVP.

5
• For each n ≥ 0, Pn (−x) = (−1)n Pn (x).
A proof of this follows immediately from Rodrigues’s formula. Just replace x by −x in
it. So, the odd degree Legendre polynomials are odd functions of x, and the even degree
Legendre polynomials are even functions of x.
• The Legendre polynomials can be formed via a linear algebra based approach (without
going into the method of finding solutions of the Legendre equation)! Let Pn (R) be the
inner product space of real polynomials of degree atmost n, with respect to the inner
product Z 1
hp, qi = p(x)q(x)dx, ∀p, q ∈ Pn (R),
−1
For each n = N∪{0}, it may be easily verified that (Pn (R, h · i) is indeed an inner product
space. The space Pn (R) has a standard basis, namely, {1, x, . . . , xn }, which is NOT an
orthonormal basis. Executing Gram Schmidt orthonormalisation
nq on the canonicalobasis
2k+1
{1, x, . . . , xn } of Pn (R), gives the orthonormal basis 2
Pk (x) : k = 0, . . . , n .
nq o
2k+1
• Since 2
Pk (x) : k = 0, . . . , n is an orthonormal basis of Pn (R), we have:
Z 1 
0 if m 6= n,
Pn (x)Pm (x)dx = 2
−1 2n+1
if m = n.

• If f (x) is a polynomial of degree at most n, we have


n
2k + 1 1
X Z
f (x) = ck Pk (x), where ck = f (x)Pk (x)dx.
k=0
2 −1
nq o
2k+1
To see this observe that clearly, f (x) ∈ Pn (R). Since, 2
P k (x) : k = 0, . . . , n is
an orthonormal basis of Pn (R), we have,
n
* r +r
X 2k + 1 2k + 1
f (x) = f (x), Pk (x) Pk (x)
k=0
2 2
n 
2k + 1 1
X Z 
= f (x)Pk (x)dx Pk (x)
k=0
2 −1
n
2k + 1 1
X Z
= ck Pk (x), where ck = f (x)Pk (x)dx.
k=0
2 −1

Hence the proof.


• For every polynomial g(x) with deg(g(x)) < n, we have
Z 1
g(x)Pn (x)dx = 0.
−1
Pn−1
The proof of this is as follows: Here deg(g(x)) ≤ n−1. Then clearly g(x) = k=0 ck Pk (x).
From this, it is immediately clear that
Z 1 * n−1 + n−1
X X
g(x)Pn (x)dx = hg(x), Pn (x)i = ck Pk (x), Pn (x) = ck hPk (x), Pn (x)i = 0.
−1 k=0 k=0

Since the Legendre polynomials form an orthonormal basis for Pn (R), there are many other
important applications of the Legendre polynomials, which is not the case if we look at the
standard basis {1, x, . . . , xn } of Pn (R).

You might also like