Identites and properties for associated Legendre functions

DBW

This note is a personal note with a personal history; it arose out off my incapacity to find
references on the internet that prove relations that exist between the associated Legendre
functions. The goal is to put notes on the internet that at least contain the 14 recurrence
relations and some other identites found on Wikipedia1 , without resorting to advanced
methods, such as generating functions; in these notes merely basic analysis and algebraic
manipulations are used. Wikipedia cites many references, such as Hilbert and Courant
(1953), or Abramowitz and Stegun (1983). But most of these were not at my disposal,
and some are very useful mathematical tables, but with very little proofs.
I start with the definition and some basic properties of Legendre polynomials Pn , then
introduce associated Legendre functions Plm . Then follows the main text, in which I give
proofs of a number of relations among the Plm . I then consider the number of zeroes of
the Pn and Plm . After that, I show proofs of some integral relations, which are cited in
the main text. On notation: I often omit the argument x of the functions and write d for
differentiation with respect to x on many occasions. Finally I give (part of) a screen shot
of the Wikipedia-website showing the 14 recurrence relations mentioned above.
I have little hope this text will be free of typos and more serious errors; please write an
email if you find one to westradennis at gmail dot com!

1 Legendre Polynomials

We define the Legendre polynomials Pl for l = 0, 1, 2, . . . by Rodriguez formula

1  d l 2
Pl (x) = (x − 1)l . (1)
2l l! dx
By inspection, Pl has degree l, and we have Pl (−x) = (−1)l Pl (x) as the lth derivative of
an even function has parity (−1)l . The Pl form an orthogonal set of polynomials on [−1; 1]
1
See the section recurrence relations from https://en.wikipedia.org/wiki/Associated_Legendre_
polynomials .

1
and proofs of the relation Z 1
δk,l
Pl (x)Pk (x)dx = , (2)
−1 2l + 1
can be found in many (online) resources, and we present one in section 4. The symbol δk,l
is the Kronecker delta, which equals 1 if k = l and zero otherwise.
Our first relation needs the orthogonality relation and some hard work, although a proof
using the generating function is faster (but then, one first has to show the generating
function has the right properties).

The Legendre polynomials satisfy the following recurrence relation

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

Proof: Consider the polynomial xPn (x). It has degree n + 1 and is thus in the linear span
of P0 , . . . , Pn+1 . We can hence write xPn (x) as a linear combination of the first n + 2
Legendre polynomials and the kth Legendre polynomial appears with coefficient
2k + 1 1
Z
ak = xPn (x)Pk (x)dx .
2 −1

We are interested in integrals of xPr (x)Ps (x) for general r and s. In section 4, we show
that these integrals vanish unless r = s ± 1 and for this case, we can use
Z 1
2r
xPr (x)Pr−1 (x)dx = .
−1 (2r − 1)(2r + 1)
Writing xPn (x) = αPn+1 (x) + βPn−1 (x), and first integrating the product with Pn+1 we
n+1 n
find α = 2n+1 and similarly β = 2n+1 . Hence
(2n + 1)xPn (x) = (n + 1)Pn+1 (x) + nPn−1 (x) .
This is what we wanted to prove.

The difference Pl+1 − Pl−1 satsifies the following differential relation:

d 
(2l + 1)Pl (x) = Pl+1 (x) − Pl−1 (x) . (4)
dx

Proof: The second derivative of (x2 − 1)l+1 equals

d2 (x2 − 1)l+1 = 2(l + 1)(x2 − 1)l + 4l(l + 1)x2 (x2 − 1)l−1
(5)
= 2(2l + 1)(l + 1)(x2 − 1)l + 4l(l + 1)(x2 − 1)l−1 .
1
Acting on this result with 2l+1 (l+1)!
dl gives the required result.

2
2 Associated Legendre Functions

We define the associated Legendre functions Plm for −l ≤ m ≤ l by

(−1)m 2 m/2 d
 l+m
Plm (x) = (1 − x ) (x2 − 1)l . (6)
2l l! dx
One immediately sees that Pl0 = Pl and that for m ≥ 0 we have
 d m
Plm (x) = (−1)m (1 − x2 )m/2 Pl (x) . (7)
dx
The m if m is odd, then Plm is
√ functions Pl are polynomials of degree l for m even and
1 − x2 times a polynomial of degre l − 1. The parity of Plm is (−1)l+m . If one tries to
apply the defining formula (6) for m > l one finds zero.

We have: p 1  m+1 m+1


1 − x2 Plm (x) = Pl−1 − Pl+1 , m ≥ 0. (8)
2l + 1

Proof: We act on identity (4) with (−1)m (1 − x2 )(m+1)/2 dm and the result follows. Later,
when we have shown (21) we will see that the condition m ≥ 0 can be relaxed.
The relation (8) is the sixth on the list of recurrence relations of Wikipedia.

For negative m we can use the following relation between Plm and Pl−m :

(l − m)! m
Pl−m = (−1)m P . (9)
(l + m)! l

Proof: By induction one easily shows that for k = 0, 1, 2, 3, . . . and some differentiable
functions f and g
k  
k
X k r
d (f g) = d (f ) · dk−r (g) .
r
r=0

We consider f = x − 1 and g = x + 1, so that and for p = 0, 1, 2, 3, . . .

d(f p ) = pf p−1 , dk (f r ) = (r)k f r−k

r!
where (r)k is defined to be 0 if r < k and (r)k = (r−k)! for r ≥ k.
We now show that
(l − m)! l+m l l
dl−m (f l g l ) = (f g)m d (f g ) .
(l + m)!

3
This will prove the claim about the associated Legendre polynomials.
To show the intermediate statement, we first rewrite
l+m
X  l  
l+m l l l+m r l l+m−r l
X l+m r l
d (f g ) = d (f ) · d (g ) = d (f ) · dl+m−r (g l )
r r=m
r
r=0

since the omitted terms in the sum vanish; dr (f s ) = 0 if r > s. Then we can write put
in the expressions for dr (f l ) and for dl+m−r (g l ), and change the summation variable to
s = r − m to obtain
l−m
l+m l l
X l!2 (l + m)!
d (f g ) = f l−s−m g s . (10)
(l − s − m)!s!(s + m)!(l − s)!
s=0

Second, we rewrite the other term dl−m (f l g l ) out as

l−m
X (l − m)! l! l!
dl−m (f l g l ) = f l−r g l−(l−m−r)
r!(l − m − r)! (l − r)! (l − (l − m − r))!
r=0

which by inspection is proportional to (f g)m times (10), and the constant of proportionality
(l−m)!
can be read off to be (−1)m (l+m)! . This proves that what we set out to show.
Equation (9) makes it clear why Plm = 0 if |m| > l.

The following relation is the first of the 14 recurrence relations listed by Wikipedia:
m
(2l + 1)xPlm = (l − m + 1)Pl+1 m
+ (l + m)Pl−1 . (11)

Proof: First we remark that it suffices to proof (11) for nonnegative m: If (11) holds for
(l+m)!
positive m, then one can insert the relation Plm = (l−m)! (−1)m Pl−m and one shows that it
follows that (2l + 1)xPl−m = (l + m + 1)Pl+1
−m −m
+ (l − m)Pl−1 , which is precisely (11) subject
to the substitution m 7→ −m.
Differentiating relation (3) m times and multiplying with (−1)m (1 − x2 )m/2 we find for
nonnegative m the relation
m
(l + 1)Pl+1 (x) = (2l + 1)xPlm (x) − m(2l + 1)(1 − x2 )1/2 Plm−1 (x) − lPl−1
m
(x) . (12)

Combining eqns.(12) and (8) one obtains the result.

The following two results are rather cheap.

4
 If we differentiate the defining relation (6) we obtain:

mx Plm+1 (x)
dPlm (x) = − P m
(x) − √ . (13)
1 − x2 l 1 − x2

Now it is a simple step for the following.

Multiplying (13) with −(1 − x2 ) one obtains nr. 13 of the list of recurrence relations from
Wikipedia: p
(x2 − 1)dPlm (x) = mxPlm (x) + 1 − x2 Plm+1 . (14)

The following relation gives an easy way to let l go one up:

p
m
Pl+1 (x) = xPlm (x) − (l + m) 1 − x2 Plm−1 (x) . (15)

Proof: We consider
 
dl+m+1 (x2 − 1)l+1 = dl+m 2(l + 1)x(x2 − 1)l ,
r
and use the binomial relation for derivatives dr (f g) = dk f dr−k g to rewrite the right-
P

k
hand side as
2(l + 1)(l + m)dl+m−1 (x2 − 1)l + 2(l + 1)xdl+m (x2 − 1)l .
(−1)m
Multiplying with 2l+1 (l+1)!
(1 − x2 )m/2 one obtains the result.

The next is the second in the list on Wikipedia:

p  
2mxPlm (x) = − 1 − x2 Plm+1 (x) + (l − m + 1)(l + m)Plm−1 (x) . (16)

Proof: We consider the following identity

 
dl+m+1 (x2 − 1)l+1 = dl+m+1 (x2 − 1)l (x2 − 1)

and use the binomial identity for derivatives rewrite the right-hand side as
(x2 − 1)dl+m+1 (x2 − 1)l + (l + m + 1)2xdl+m (x2 − 1)l + (l + m)(l + m + 1)dl+m−1 (x2 − 1)l .
(−1)m
Multiplying through by − x2 )m/2 and regrouping some terms we find
2l l!
(1
p  
m
2(l + 1)Pl+1 − 2x(l + m + 1)Plm = 1 − x2 Plm+1 − (l + m)(l + m + 1)Plm−1 . (17)

5
 m we obtain the (16).
If we insert relation (15) to eliminate Pl+1

This identity is the third on the list of Wikipedia:

1 1  m+1 m−1

√ Plm (x) = − Pl−1 (x) + (l + m − 1)(l + m)Pl−1 (x) . (18)
1 − x2 2m

√
m + 2m(l + m) 1 − x2 P m−1 ,
Proof: We start with (16) and substitute 2mxPlm = 2mPl+1 l
which follows from (15). Then it follows that
p  
m
2mPl+1 = − 1 − x2 Plm+1 + (l + m)(l + m + 1)Plm−1 ,

from which (18) follows immediately if we change l to l − 1.

The following identity is not on Wikipedia’s list, but is definitely usefull to relate some
identities. Its proof is rather cumbersome and tedious.

The following is rather surprising:

m+1 m−1 m+1 m−1
Pl−1 + (l + m)(l + m − 1)Pl−1 = Pl+1 + (l − m + 1)(l − m + 2)Pl+1 . (19)

Remark: Although at first glance the right-hand side seems to be of degree l + 1, the
highest-order terms cancel, and so is also of degree l − 1.
Proof: If we plug in the definitions of the Plm , we see that we are to prove that the two
terms  
L = 4l(l + 1) (1 − x2 )dl+m + (l + m)(l + m − 1)dl+m−2 (x2 − 1)l−1

and  
R = (1 − x2 )dl+m+2 + (l − m + 1)(l − m + 2)dl+m (x2 − 1)l+1

are equal. We distinguish between two cases: (I) l + m = 2n, (II) l + m = 2n + 1. We first
consider case (I), and eliminate all occurrences of m.
B!
We expand (x2 − 1)l±1 by the binomial theorem, apply dA xB = (B−A)! x
B−A and find that

l−n
X
L= 4l(l + 1)Lk x2k ,
k=0

where
 
l+n+k+1 l−1 (2k + 2n − 2)!  
Lk = 2(−1) (2n+2l−1)(l−n−k)+k(2k−1)−n(2n−1) .
n+k−1 (2k)!

6
We also expand R = l−n+1 Rk x2k and find
P
k=0
 
l + 1 (2k + 2n)!  
Rk = 2(−1)l+n+k (2n+2k+1)(l+1−n−k)+k(2k−1)−(2l−2n+1)(l−n+1) .
n + k1 (2k)!

In particular, Rl−n+1 = 0. We can write

4(l + 1)!(2k + 2n − 2)!

4l(l + 1)Lk = (−1)l+n+l+1 · 2(n + 1)(l + 1 − n − k)ak
(n + k)!(l + 1 − n − k)!(2k)!

with
ak = (2n + 2l − 1)(l − n − k) + k(2k − 1) − n(2n − 1)
and similarly

4(l + 1)!(2k + 2n − 2)!

Rk = (−1)l+n+l+1 (n + k)(2k + 2n − 1)bk
(n + k)!(l + 1 − n − k)!(2k)!

with

bk = (2l − 2n + 1)(l − n + 1) − (2n + 2k + 1)(l + 1 − n − k) − k(2k − 1) .

By inspection, it suffices to show 2ak (l + 1 − n − k) = (2k + 2n − 1)bk , which is seen rather

easily if we rewrite ak and bk as

ak = (l − 2n)(2n + 2k − 1) , bk = 2(l − 2n)(l + 1 − n − k) .

This shows identity (19) for l + m is even.

For the case, where l + m is odd, we take l + m = 2n + 1, and it boils down to proving
that the terms  
L = (1 − x2 )d2n+1 + 2n(2n + 1)d2n−1 (x2 − 1)l−1

and  
2 2n+3 2n+1
R = (1 − x )d + (2l − 2n)(2l − 2n + 1)d (x2 − 1)l+1
Pl−n 2k−1 L
Pl−n+1
are related by 4l(l+1)L = R. Again we write L = k=1 x k and R = k=1 x2k−1 Rk .
We find with some algebra that

(l − 1)!(2n + 2k − 2)!
Lk = 2(−1)l+n+k+1 ak
(n + k − 1)!(l − n − k)!(2k − 1)!

where ak = (l − 2n − 1)(2n + 2k − 1). This leads to

(l + 1)!(2n + 2k)!(2n + 2k − 1)!

4l(l + 1)Lk = 4(−1)l+n+k+1 (l − 2n − 1) .
(n + k)!(l − n − k)!(2k − 1)!

7
Similar and similarly tedious algebra leads to

(l + 1)!(2n + 2k)!
Rk = 2(−1)l+n+k+1 bk
(n + k)!(l + 1 − n − k)!(2k − 1)!

with

bk = (l − n)(2l − 2n + 1) − (2k − 1)(k − 1) − (2n + 2k + 1)(l + 1 − n − k) ,

which can be brought into

bk = 2(l − 2n − 1)(l + 1 − n − k) .

Substituting this expression into Rk shows that 4l(l + 1)Lk = Rk and the proof is done –
indeed, also Rl−n+1 = 0.

Nr. 4 on Wikipedia’s list follows directly from (18) and (19)

1 −1  m+1 m−1

√ Plm = Pl+1 + (l − m + 1)(l − m + 2)Pl+1 . (20)
1 − x2 2m

Nr. 5 on Wikipedia’s with m ≥ 0 list follows directly from (8) and (19)
p 1 
m−1 m−1

1 − x2 Plm = (l − m + 1)(l − m + 2)Pl+1 − (l + m − 1)(l + m)Pl−1 . (21)
(2l + 1)

Let us now show: Equations (8) and (21) also hold for m ≤ 0. By virtue of (9) we substi-
tute Plm = (−1)m (l+m)!
(l−m)! Pl
−m m−1
, Pl+1 (l+m)!
= (−1)m+1 (l−m+2)! −m+1
Pl+1 m−1
and Pl−1 (l+m)! −m+1
= (−1)m (l−m)! Pl−1
√ −m 1 −m+1 −m+1
in (21), which leads directly to 1 − x2 Pl = 2l+1 (Pl−1 − Pl+1 ), which is precisely
(8) for negative m. The same substitutions in (8) show that (21) also holds for negative
m.
As nr. 6 was already proven, we now proceed to number 7 on Wikipedia’s list.

The following identity holds

p
1 − x2 Plm+1 = x(l − m)Plm − (l + m)Pl−1
m
. (22)

Proof: We use (16) to write

p p
1 − x2 Plm+1 = −(l + m)(l − m + 1) 1 − x2 Plm−1 − 2mxPlm .

8
 √
Then we insert (8) to eliminate 1 − x2 Plm−1 and find
p (l + m)(l − m + 1) m (l + m)(l − m + 1) m
1 − x2 Plm+1 = − Pl−1 + Pl+1 − 2mxPlm .
2l + 1 2l + 1
Inserting now (l − m + 1)Pl+1m = (2l + 1)xP m − (l + m)P m , a.k.a. (11), and doing the
l l−1
algebra then finishes the proof.

The following identity is nr. 8 on Wikipedia’s list:

p
1 − x2 Plm+1 = (l − m + 1)Pl+1
m
− (l + m + 1)xPlm . (23)

Proof: We rewrite the right-hand side of (22) as

x(l − m)Plm − (l + m)Pl−1
m
= (2l + 1)xPlm − (l + m + 1)xPlm − (l + m)Pl−1
m
.
The proof is finished by applying (11) to eliminate the term (2l + 1)xPlm .
We now turn to some differential properties of the associated Legendre functions.

We present nr. 9:
p 1 1
1 − x2 dPlm (x) = − Plm+1 (x) + (l + m)(l + 1 − m)Plm−1 (x) . (24)
2 2

We can rewrite equation (13) as

p mx
1 − x2 dPlm (x) = − √ Plm (x) − Plm+1 (x) .
1−x 2

Using equation (16) and eliminating mxPlm (x) leads to the result.
The remaining equations from the list of identities on Wikipedia are then mainly just
simple algebra and combining all the other results:

Number 10:
1  
(1 − x2 )dPlm = m
(l + 1)(l + m)Pl−1 m
− l(l − m + 1)Pl+1 (25)
2l + 1

Proof: Using eqn.(24) we write the left-hand side of equation to be proven as

1 p p 
(l + m)(l − m + 1) 1 − x2 Plm−1 − 1 − x2 Plm+1 .
2
√ √
Then we use equation (8) for 1 − x2 Plm−1 and equation (21) for 1 − x2 Plm+1 . This
finishes the proof.

9
 Number 11:
(x2 − 1)dPlm = lxPlm − (l + m)Pl−1
m
. (26)

Proof: If combine (11) and (25) this identity follows immediately.

Number 12:
(x2 − 1)dPlm = −(l + 1)xPlm + (l + 1 − m)Pl+1
m
. (27)

Proof: The proof follows if we use l = (2l + 1) − (l + 1) and write by virtue of (26)

(x2 − 1)dPlm = (2l + 1)xPlm − (l + 1)xPlm − (l + m)Pl−1

m

and then use relation (11).

As we already had number 13 before, we go on to the last on the list:

Number 14:
p
(x2 − 1)dPlm = −(l + m)(l − m + 1) 1 − x2 Plm−1 − mxPlm . (28)

First
√ we write mxPlm = 2mxPlm − mxPlm , then we insert (16) and subsequently we add
1 − x2 Plm+1 , which shows
p p
mxPlm + 1 − x2 Plm+1 = − 1 − x2 (l + m)(l − m + 1)Plm−1 − mxPlm ,

which is the right-hand side of the equation to be proven by virtue of (14).

3 On the zeros of (associated) Legendre polynomials

In this section we prove the following claims:

(1) The Legendre polynomials Pl (x) have l distinct simple zeros on the interval (−1; +1);

(2) The associated Legendre functions Plm (x) have l − m distinct simple zeros on the
interval (−1; +1).

The case l = 0 is trivial, since P0 (x) = 1. Claim (1) is claim (2) restricted to m = 0. We
first consider this case. To make the notation easier, we introduce the short-hand
Z +1
hf, gi = f (x)g(x)dx .
−1

10
Then the integral of Pn over the interval [−1; +1] equals zero as this integral equals hPl , 1i =
hPl , P0 i = 0. Thus Pl , being continuous, must change sign somewhere on (−1; +1) and
thus has at least one zero of odd multiplicity. Let ξ1 , . . . , ξk be all the different zeros of
odd multiplicity of Pl , so that 1 ≤ k ≤ l and the continuous function ZPl with Z =
(x − ξ1 ) · (x − ξ2 ) · · · (x − ξk ) does not change on (−1; +1). Thus hZ, Pl i =
6 0.
If k < l then Z is a polynomial of degree smaller than l and thus can be written as a linear
combination of P0 , P1 , . . . , Pl−1 . But then hZ, Pl i vanishes, which cannot be, and thus
k = l. Thus Pl has l distinct zeros.
For the second claim, we need a preliminary result, which is
Z +1
m m 2 (l + m)!
hPl , Pl0 i = Plm (x)Plm
0 dx = δl,l0 (29)
−1 2l + 1 (l − m)!

which is proved in section 4.

Since we may assume l > 0 and m > 0, we first consider l = m. Then Pm m is proportional

to (1 − x2 )m/2 and thus has zero zeros on (−1; +1). Thus in this case the claim is true.
For the remaining we fix l > m > 0. For each k ≥ 0 the functions Pkm is of the form
(1 − x2 )m · pk−m (x), where pk−m is a polynomial of degree k − m. Thus for given N > m
m (x), (1 − x2 )−m/2 · P m (x), (1 − x2 )−m/2 · P m (x) span
the polynomials (1 − x2 )−m/2 · Pm m+1 N
the space of polynomials with degree at most N − m.
Since hPlm , Pm
m i = 0 as l > m and P m does not change sign, so does P m have to change
m l
sign. Thus there are k ≥ 1 zeros ξ1 , . . . , ξk of Plm of odd multiplicity on (−1; +1). We
introduce Z = (1 − x2 )m/2 · (x − ξ1 ) · · · (x − ξk ), so that ZPlm does not change sign on
(−1; +1) and therefore hZ, Plm i =
6 0.
If l < k then Z can be written as a linear sum of less then l associated Legendre functions
Pmm , . . . P m , so that in this case we should have hZ, P m i = 0, which we already argued
k l
cannot be. Hence l = k and the claim is proved.

4 Proof of integral relations

We consider the function B(a, b) defined by

Z 1
B(a, b) = ta (1 − t)b dt .
0

A partial integration shows

Z 1 Z 1
b b
ta (1 − t)b dt = ta+1 (1 − t)b−1 dt = B(a + 1, b − 1) .
0 a+1 0 a+1

11
 1
Since B(a, 0) = a+1 , it then follows rather easily

a!b!
B(a, b) = . (30)
(a + b + 1)!

Using the substitution x = 2t − 1 we deduce

Z 1
l!l!
(1 − x2 )l dx = 22l+1 B(l, l) = 22l+1 . (31)
−1 (2l + 1)!

Let us now show that Z 1

2
Pk (x)Pl (x)dx = . (32)
−1 2l + 1
Plugging in the definition, we have
Z 1 Z 1
1
Pk (x)Pl (x)dx = k+l
dk (x2 − 1)k · dl (x2 − 1)l dx .
−1 2 l!k! −1

Since da (x2 − 1)b vanishes at x = ±1 if a < b, partial integration leads to

Z 1 Z 1
(−1)l
Pk (x)Pl (x)dx = k+l dk+l (x2 − 1)k · (x2 − 1)l dx .
−1 2 l!k! −1

If k 6= l we may assume l > k and then dk+l (x2 − 1)k = 0. Hence the integral vanishes
unless k = l. In this case we use d2l (x2 − 1)l = d2l x2l = (2l)! and (x2 − 1)l = (−1)l (1 − x2 )l
to find Z 1 Z 1
(2l)!
Pk (x)Pl (x)dx = δk,l 2l (x2 − 1)l dx ,
−1 2 l!l! −1
which directly leads to the result using (31).
Let us now show that
Z 1
2r 2s
xPs (x)Pr (x)dx = δr,s+1 + δs,r+1 . (33)
−1 (2r − 1)(2r + 1) (2s + 1)(2s − 1)

First we consider the case that r = s. In this case the integral is easily found to be zero
– Pr has parity (−1)r , and thus x(Pr )2 has parity −1. Therefore the integral over the
symmetric interval [−1; 1] vanishes.
Now we consider the case r > s. WeR find, using Rodriguez formula (1), partial integration
1
and (xf )(n) = xf (n) +nf (n−1) , that −1 xPr (x)Ps (x)dx equals up to some numerical factors
the expression
Z 1 h  d r+s  d r+s−1 i
(x2 − 1)r x (x2 − 1)s + r (x2 − 1)s dx .
−1 dx dx

12
We see that the first term vanishes since r > s; the second term only survives if r ≤ s + 1.
Hence we need only consider the case s = r − 1.
Writing cr = 2r1r! we find, again using Rodriguez formula, partial integration and (xf )(n) =
xf (n) + nf (n−1) but also that the (2r − 2)th derivative of (x2 − 1)r−1 equals (2r − 2)!, that
Z 1 Z 1
xPr (x)Pr−1 (x)dx = (−1)r cr cr−1 (2r − 2)! (x2 − 1)r dx .
−1 −1
R1 2r
Putting in the numbers, we find −1 xPr (x)Pr−1 (x)dx = (2r−1)(2r+1) . The case s > r leads
to the same expression, but s and r interchanged.
Finally, we show Z +1
2 (l + m)!
Pkm (x)Plm (x)dx = δk,l . (34)
−1 2l + 1 (l − m)!
Plugging in the definition, we have
Z +1 Z 1
m m 1
Pk (x)Pl (x)dx = k+l (1 − x2 )m dk+m (x2 − 1)k · dl+m (x2 − 1)l dx .
−1 2 k!l! −1

The degree of g(x) = (1−x2 )m dk+m (x2 −1)k is k+m. The degree of h(x) = dl+m (x2 −1)l is
l−m. If we first assume k 6= l, we may as well take l > k. The first m−1 derivatives vanish
at x = ±1 and the first l − 1 derivatives of (x2 − 1)l vanish at well at x = ±. Therefire,
if we perform partial integrations, moving the derivatives from h to g, all boundary terms
vanish. Since dl+m g(x) = 0 if l > k, the integral (34) vanishes if k 6= l.
If k = l, we find
Z +1
(−1)l+m 1 2
Z  
Pkm (x)Plm (x)dx = k+l (x − 1)l dl+m (1 − x2 )m dl+m (x2 − 1)l dx .
−1 2 l!l! −1

Since the term in brackets has degree l + m, we only need to consider the highest order
term;
  (2l)!(l + m)!
dl+m (1 − x2 )m dl+m (x2 − 1)l = (−1)m dl+m (x2m dl+m x2l ) = (−1)m .
(l − m)!

Combining this result and the previous integrals one easily finds the stated result.

13
5 The list of recurrence relations from Wikipedia

14