You are on page 1of 14

Some Special Functions Dr.

Suresh Kumar, BITS Pilani 1

Chapter 8

Some Special Functions

Note: This module is prepared from Chapter 8 of the text book (G.F. Simmons, Differential
Equations with Applications and Historical Notes, TMH, 2nd ed., 1991) just to help the students.
The study material is expected to be useful but not exhaustive. For detailed study, the students
are advised to attend the lecture/tutorial classes regularly, and consult the text book.

Appeal: Please do not print this e-module unless it is really necessary.

Dr. Suresh Kumar, Department of Mathematics, BITS Pilani, Pilani Campus


Contents

0.1 Legendre Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3


0.2 Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.3 Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
0.3.1 Second solution of Bessels DE . . . . . . . . . . . . . . . . . . . . . . . . . . 10
0.3.2 Properties of Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 12
0.4 Orthogonal properties of Bessel functions . . . . . . . . . . . . . . . . . . . . . . . . 12
0.4.1 Fourier-Bessel Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2
Some Special Functions

0.1 Legendre Polynomials


A DE of the form

(1 x2 )y 00 2xy 0 + n(n + 1)y = 0, (1)

where n is a constant, is called Legendres Equation. We observe that x = 0 is an ordinary point


of (1). So there exists a series solution of the form

X
y= ak xk = a0 + a1 x + a2 x2 + a3 x3 + ......... (2)
k=0

Substituting (2) into (1), we obtain



X
X
k2
ak k(k 1)x + ak (n k)(n + k + 1)xk = 0. (3)
k=0 k=0

Comparing coefficients of xk2 , we obtain

(n k + 2)(n + k 1)
ak k(k 1) + ak2 (n k + 2)(n + k 1) = 0 or ak = ak2 .
k(k 1)

a2 = n(n+1)
2!
a0 , a3 = (n1)(n+2)
3!
a1 a4 = (n2)n(n+1)(n+3)
4!
a0 , a5 = (n3)(n1)(n+2)(n+4)
5!
a1 , .......

Substituting these values into (2), we obtain the general solution of (1) as

y = c1 y 1 + c2 y 2

where
 
n(n + 1) 2 (n 2)n(n + 1)(n + 3) 4
y 1 = a0 1 x + x ......... ,
2! 4!
 
(n 1)(n + 2) 3 (n 3)(n 1)(n + 2)(n + 4) 5
y 2 = a1 x x + x ......... .
3! 5!
We observe that y1 and y2 are LI solutions of the Legendre equation (1), and these are analytic
in the range 1 < x < 1. However, the solutions most useful in the applications are those bounded
near x = 1. Notice that x = 1 is a regular singular point of the Legendre equation (1). We use

3
Some Special Functions Dr. Suresh Kumar, BITS Pilani 4

the transformation t = (1 x)/2 so that x = 1 corresponds to t = 0, and (1) transforms to the


hypergeometric DE

t(1 t)y 00 + (1 2t)y 0 + n(n + 1)y = 0, (4)

where the prime denote derivative with respect to t. Here, a = n, b = n + 1 and c = 1. So the
solution of (4) in the neighbourhood of t = 0 is given by

y1 = F (n, n + 1, 1, t). (5)

The other LI solution can be found as


Z
1 R P dt
y2 = y1 e dt = y1 (ln t + a1 t + ........). (6)
y12

However, this solution is not bounded near t = 0. So any solution of (4) bounded near t = 0 is a
constant multiple of y1 . Consequently, the constant multiples of F (n, n + 1, 1, (1 x)/2) are the
solutions of (1), which are bounded near x = 1.
If n is a non-negative integer, then F (n, n + 1, 1, (1 x)/2) defines a polynomial of degree n
known as Legendre polynomial, denoted by Pn (x). Therefore,

n(n + 1) n(n 1)(n + 1)(n + 2) (2n)!


Pn (x) = F (n, n+1, 1, (1x)/2) = 1+ 2
(x1)+ 2 2
(x1)2 +....+ (x1)n .
(1!) 2 (2!) 2 (n!)2 2n

Notice that Pn (1) = 1 for all n. Next, after a sequence of algebraic manipulations, we can obtain

1 dn
Pn (x) = [(x2 1)n ],
2n n! dxn
known as Rodrigues formula. The following theorem provides the alternative approach to obtain
the Rodrigues formula.

Theorem 0.1.1. (Rodrigues Formula)


1 dn
Prove that Pn (x) = [(x2 1)n ].
2n n! dxn
Proof. Let v = (x2 1)n . Then we have

dv
v1 = 2nx(x2 1)n1 where v1 = .
dx
= (x2 1)v1 = 2nx(x2 1)n .
= (1 x2 )v1 + 2nxv = 0.
Differentiating it n + 1 times with respect to x using the Leibnitz theorem, we get

(n + 1)n
(1 x2 )vn+2 + (n + 1)(2x)vn+1 + (2)vn + 2n[xvn+1 + (n + 1)vn ] = 0.
2!
= (1 x2 )vn00 2xvn0 + n(n + 1)vn = 0.
Some Special Functions Dr. Suresh Kumar, BITS Pilani 5

This shows that cvn (c is an arbitrary constant) is a solution of the Legendres equation (1). Also
cvn is a polynomial of degree n. But we know that the nth degree polynomial Pn (x) is a solution
of the Legendres equation. It follows that
dn
Pn (x) = cvn = c [(x2 1)n ]. (7)
dxn
To find c, we put x = 1 into (7) to get
 n 
d 2 n
Pn (1) = c [(x 1) ] .
dxn x=1
 n 
d n n
= 1=c n
[(x 1) (x + 1) ] = c[n!(x+1)n +Terms containing the factor (x1)]x=1 .
dx x=1
1
= 1 = c.n!2n or c = .
n!2n
Thus, (7) becomes

1 dn
Pn (x) = n n
[(x2 1)n ].
2 n! dx
This completes the proof.

Remark 0.1.1. Using Rodrigues formula, we get


P0 (x) = 1, P1 (x) = x, P2 (x) = 12 (3x2 1), P3 (x) = 12 (5x3 3x) , P4 (x) = 35 4
8
x 15 2
4
x + 3
8
etc.

Ex. 0.1.1. Express the polynomial x4 + 3x3 x2 + 5x 2 in terms of Legendres polynomials.

Sol. 0.1.1. Since P4 (x) = 358


x4 15
4
x2 + 38 , so x4 = 35
8
P4 (x)+ 67 x2 35
3
. Similarly, x3 = 25 P3 (x)+ 35 x,
2 2 1
x = 3 P2 (x) + 3 , x = P1 (x), 1 = P0 (x). Using all these, we get

8 6 2 34 224
x4 + 3x3 x2 + 5x 2 = P4 (x) + P3 (x) P2 (x) + P1 (x) P0 (x).
35 5 21 5 105

Orthogonality of Legendre Polynomials


Z 1
Ex. 0.1.2. Show that Pm (x)Pn (x)dx = 0 for m 6= n.
1

Sol. 0.1.2. We know that y = Pm (x) is a solution of

(1 x2 )y 00 2xy 0 + m(m + 1)y = 0, (8)

and z = Pn (x) is a solution of

(1 x2 )z 00 2xz 0 + n(n + 1)z = 0. (9)

Multiplying (8) by z and (9) by y, and subtracting, we get


Some Special Functions Dr. Suresh Kumar, BITS Pilani 6

(1 x2 )(y 00 z yz 00 ) 2x(y 0 z yz 0 ) + [m(m + 1) n(n + 1)]yz = 0.

d
= [(1 x2 )(y 0 z yz 0 )] + (m n)(m + n + 1)yz = 0. (10)
dx
Integrating (10) from 1 to 1, we have
Z 1
(m n)(m + n + 1) yzdx = 0
1

Also, m 6= n. So it gives
Z 1
Pm (x)Pn (x)dx = 0.
1

This is known as the orthogonality property of Legendre polynomials.


Z 1
2
Ex. 0.1.3. Show that Pn2 (x)dx = .
1 2n + 1
Sol. 0.1.3. The Rodrigues formula is
1 dn 1
Pn (x) = n n
[(x2 1)n ] = n Dn (x2 1)n .
2 n! dx 2 n!
Therefore, we have
Z 1 Z 1
n 2 2
(2 n!) Pn (x)dx = Dn (x2 1)n Dn (x2 1)n dx
1 1
Z 1
n x=1
 n 2 n n1 2
Dn+1 (x2 1)n Dn1 (x2 1)n dx

= D (x 1) D (x 1) x=1
1
Z 1
= 0 Dn+1 (x2 1)n Dn1 (x2 1)n dx
1
Z 1
n
= (1) D2n (x2 1)n (x2 1)n dx (Integrating (n 1) times more)
1
Z 1
= (1)n (2n)! (x2 1)n dx (Put x = sin )
1
Z/2
= 2(2n)! cos2n+1 d
0
2n(2n 2).......4.2
= 2(2n)!
(2n + 1)(2n 1)........3.1
[2n(2n 2).......4.2]2
= 2(2n)!
(2n + 1)!
2
= (2n n!)2
2n + 1
Z 1
2
Pn2 (x)dx = .
1 2n + 1
Some Special Functions Dr. Suresh Kumar, BITS Pilani 7

Legendre Series
Let f (x) be a function defined from x = 1 to x = 1. Then we can write,

X
f (x) = cn Pn (x), (11)
n=0

where cn s are constants to be determined. Multiplying both sides of (11) by Pn (x) and integrating
from 1 to 1, we get
Z 1 Z 1
2
f (x)Pn (x)dx = cn Pn2 (x)dx = cn .
1 1 2n + 1
Z 1
2n + 1
= cn = f (x)Pn (x)dx.
2 1

Using the values of cn into (11), we get the expansion of f (x) in terms of Legendre polynomials,
known as the Legendre series of f (x).

Ex. 0.1.4. If f (x) = x for 0 < x < 1 otherwise 0, then show that f (x) = 14 P0 (x) + 21 P1 (x) +
5
P (x) + .......
16 2

1 1 1 1
Z Z
1
Sol. 0.1.4. c0 = f (x)P0 (x)dx = x.1dx = , etc.
2 1 2 0 4

X
2 1/2
Ex. 0.1.5. Prove that (1 2xt + t ) = tn Pn (x), and
n=0
hence prove the recurrence relation nPn (x) = (2n 1)xPn1 (x) (n 1)Pn2 (x).

Sol. 0.1.5. Please try yourself.

Note: The function (1 2xt + t2 )1/2 is called generating function of the Legendre polynomials.
Note that the Legendre polynomials Pn (x) appear as coefficients of tn in the expansion of the
function
(1 2xt + t2 )1/2 .

0.2 Gamma Function


The gamma function is defined as
Z
(n) = ex xn1 dx, (n > 0) (12)
0

The condition n >Z0 is necessary in order to guarantee the convergence of the integral.

Note that (1) = ex dx = 1.
0
Next, we have
Z 
Z Z
x n n x n1 x
ex xn1 dx.

(n + 1) = e x dx = x e (1) 0 n nx e (1)dx = n
0 0 0
Some Special Functions Dr. Suresh Kumar, BITS Pilani 8

(n + 1) = n(n).
It is the recurrence relation for gamma function. Using this relation recursively, we have
(2) = 1.(1) = 1,
(3) = 2.(2) = 2.1 = 2!,
(4) = 3.(3) = 3.2! = 3!,
......
(n + 1) = n.(n) = n.(n 1)! = n!.
Thus, (n)
takes positive integer values for positive integer values of n. It can be proved that
1
( 2 ) = . For,
  Z Z
1 t 1/2 2
= e t dx = 2 ex dx, where t1/2 = x.
2 0 0

  2  Z  Z 
1 x2 y 2
= 2 e dx 2 e dy
2 0 0
Z Z
2 2
= e(x +y ) dxdy
Z0 2 Z0
2
= er rdrd, x = r cos , y = r sin
0 0
=
Having known the precise value of 21 , we can calculate the values of gamma function at positive


fractions with denominator 2. For instance,


5 3 1
   
7 5 3 1 1
= . . = . . .
2 2 2 2 2 2 2 2
For values of gamma function at positive fractions with denominator different from 2, we have to
rely upon the numerically approximated value of the integral arising in gamma function.
Note that (n) given by (12) is not defined for n 0. We extend the definition of gamma
function by the relation
(n + 1)
(n) = (13)
n
Then (n) is defined for all n except when n is any non-positive integer. If we agree (n) to be
for non-positive integer values of n, then 1/(n) is defined for all n. Such an agreement is useful
while dealing with Bessel functions. The Gamma function is, thus, defined as

Z


ex xn1 dx , n>0
0





(n) = (n + 1)
, n < 0 but not an integer


n



, n = 0, 1, 2, .......

Some Special Functions Dr. Suresh Kumar, BITS Pilani 9

Note that the gamma function generalizes the concept of factorial from non-negative integers
to any real number via the formula

n! = (n + 1).

0.3 Bessel Functions


The DE

x2 y 00 + xy 0 + (x2 p2 )y = 0, (14)

where p is a non-negative constant, is called Bessels DE. We see that x = 0 is a regular singular
point of (14). So there exists at least one Frobenious series solution of the form

X
y= an xn+r , (a0 6= 0). (15)
n=0

Using (15) into (14), we get



X
X
an [(n + r)2 p2 ]xn+r + an xn+r+2 = 0. (16)
n=0 n=0

Equating to 0 the coefficient of xr , the lowest degree term in x, we obtain

a0 (r2 p2 ) = 0 or r2 p2 = 0.

Therefore, roots of the indicial equation are r = p, p.


Next equating to 0 the coefficient of xr+1 , we find

a1 [(r + 1)2 p2 ] = 0 or a1 = 0 for r = p.

Now equating to 0 the coefficient of xn+r , we have the recurrence relation


an2
an = , (17)
(n + r)2 p2
where n = 2, 3, 4....
For r = p, we get the solution in the form

X (1)n (x/2)2n
y = a0 x p . (18)
n=0
n!(p + 1)(p + 2)....(p + n)

1
The Bessel function of first kind of order p, denoted by Jp (x) is defined by putting a0 = 2p p!
into
(18) so that

X (1)n (x/2)2n+p
Jp (x) = , (19)
n=0
n!(n + p + 1)

which is well defined for all real values of p in accordance with the definition of gamma function.
Some Special Functions Dr. Suresh Kumar, BITS Pilani 10

1.0

0.8

0.6

0.4

0.2

x
2 4 6 8 10
- 0.2

- 0.4

Figure 1: Plots of J0 (x) (Blue curve) and J1 (x) (Red curve).

From applications point of view, the most useful Bessel functions are of order 0 and 1, given
by
x2 x4 x6
J0 (x) = 1 + + ..........
22 22 .42 22 .42 .62
x 1  x 3 1  x 5
J1 (x) = + ..........
2 1!2! 2 2!3! 2
Plots of J0 (x) (Blue curve) and J1 (x) (Red curve) are shown in Figure 1. It may be seen that J0 (x)
and J1 (x) vanish alternatively, and have infinitely many zeros on positive x-axis, as expected, since
J0 (x) and J1 (x) are two particular LI solutions of the Bessels DE (14). Later, we shall show that
J00 (x) = J1 (x). Thus, J0 (x) and J1 (x) behave just like cos x and sin x. This analogy may also be
observed by the fact that the normal form of Bessels DE (14) given by

1 4p2
 
00
u + 1+ u = 0,
4x2
behaves as

u00 + u = 0,

for large values of x, with solutions cos x and sin x. It means J0 (x) and J1 (x) behave more precisely
like cos x and sin x for larger values of x.

0.3.1 Second solution of Bessels DE


To obtain second solution it is natural to try the second root r = p of the indicial equation. We
assume that p is not an integer otherwise the difference of the indicial equation roots p and p
would be the integer 2p. For r = p, the equation a1 [(r + 1)2 p2 ] = 0 becomes a1 (1 2p) = 0,
which lets a1 arbitrary for p = 1/2. So there is no compulsion to choose a1 = 0. However, we fix
a1 = 0 after all we are interested in a particular solution. Also, for r = p, the recurrence relation
(17) reduces to
an2 an2
an = 2 2
= ,
(n p) p n(n 2p)
where n = 2, 3, 4....
For n = 3, we get 3(3 2p)a3 = a1 = 0. This lets a3 arbitrary for p = 3/2. We choose a3 = 0.
Some Special Functions Dr. Suresh Kumar, BITS Pilani 11

Likewise, we choose a5 = 0, a7 = 0, ........ for the sake of particular solution, and thus obtain the
following particular solution of (14):

X (1)n (x/2)2np
Jp (x) = , (20)
n=0
n!(n p + 1)

which is same as if we replace p by p in (19).


Notice that Jp (x) and Jp (x) are LI since Jp (x) is bounded near x = 0 but Jp (x) is not so.
Thus, when p is not an integer, the general solution of Bessels equation (14) is

y = c1 Jp (x) + c2 Jp (x). (21)

Now let us see what happens when p is a non-negative integer say m. We have

X (1)n (x/2)2nm
Jm (x) =
n=0
n!(m + n)!

(1)n (x/2)2nm
 
X 1
= = 0, n = 0, 1, 2, ...., m 1
n=m
n!(m + n)! (m + n)!

X (1)n+m (x/2)2(n+m)m
= (Replacing the dummy variable n by n + m)
n=0
(n + m)!(m + n + m)!

X (1)n (x/2)2n+m
= (1)m
n=0
n!(m + n)!
= (1)m Jm (x)

This shows that Jp (x) and Jp (x) are not LI when p is an integer.
When p is not an integer, any function of the form (21) with c2 6= 0 is a Bessel function of
second kind. The standard Bessel function of second kind is defined as
Jp (x) cos p Jp (x)
Yp (x) = . (22)
sin p

One can write (21) in the equivalent form

y = c1 Jp (x) + c2 Yp (x), (23)

which is general solution of (14) when p is not an integer. One may observe that Yp (x) is not
defined when p is an integer say m. However, it can be shown that

Ym (x) = lim Yp (x)


pm

exists, and it is taken as the Bessel function of second kind. Thus, it follows that (23) is general
solution of Bessels equation (14) in all cases. It is found that Yp (x) is not bounded near x = 0 for
p 0. Accordingly, if we are interested in solutions of Bessels equation near x = 0, which is often
the case in applications, then we must take c2 = 0 in (23).
Some Special Functions Dr. Suresh Kumar, BITS Pilani 12

0.3.2 Properties of Bessel Functions


It is easy to prove the following:
d p
(1) [x Jp (x)] = xp Jp1 (x).
dx
d  p
x Jp (x) = xp Jp+1 (x).

(2)
dx
1
(3) Jp0 (x) = [Jp1 (x) Jp+1 (x)].
2
2p
(4) Jp+1 (x) = Jp (x) Jp1 (x).
x
Z
From (1), we have xp Jp1 (x)dx = xp Jp (x) + C.
Z
Similarly, (2) gives xp Jp+1 (x)dx = xp Jp (x) + C.
Also, notice that (4) is the recurrence relation for Bessel functions. By definition of Bessel function,
it can be shown that
r r
2 2
J1/2 (x) = sin x, J1/2 (x) = cos x.
x x
So by property (4)
r  
1 2 sin x
J3/2 (x) = J1/2 (x) J1/2 (x) = cos x .
x x x

Again, by property (4)


r
1 2  cos x 
J3/2 (x) = J1/2 (x) J1/2 (x) = sin x .
x x x
Thus, every Bessel function Jm+ 1 (x), where m is any integer, is elementary as it is expressible in
2
terms of elementary functions.

0.4 Orthogonal properties of Bessel functions


If m and n are positive zeros of Jp (x), then
Z 1 
0, m 6= n
xJp (m x)Jp (n x)dx = 1 2
0 J ( ) ,
2 p+1 m
m = n.

Proof. Since y = Jp (x) is a solution of

p2
 
00 1 0
y + y + 1 2 y = 0,
x x

it follows that u(x) = Jp (m x) and v(x) = Jp (n x) satisfy the equations

p2
 
00 1 0 2
u + u + m 2 u = 0, (24)
x x
Some Special Functions Dr. Suresh Kumar, BITS Pilani 13

p2
 
001 0 2
v + v + n 2 v = 0, (25)
x x

Multiplying (24) by v and (25) by u, and subtracting the resulting equations, we obtain

d 0 1
(u v v 0 u) + (u0 v v 0 u) = (2n 2m )uv.
dx x
After multiplication by x, it becomes
d
[x(u0 v v 0 u)] = (2n 2m )xuv.
dx
Now, integrating with respect to x from 0 to 1, we have
Z 1
1
2 2
(n m ) uv = [x(u0 v v 0 u)]0 = 0,
0

since u(1) = Jp (m ) = 0 and v(1) = Jp (n ) = 0.


Z 1
xJp (m x)Jp (n x)dx = 0, (m 6= n).
0

Next, we consider the case m = n. Multiplying (24) by 2x2 u0 , we get

2x2 u0 u00 + 2xu02 + 22m x2 uu0 2p2 uu0 = 0.

d
x2 u02 + 2m x2 u2 p2 u2 = 22m xu2 .

=
dx
Integrating from 0 to 1 with respect to x, we get
Z 1
1
2
xu2 dx = x2 u02 + 2m x2 u2 p2 u2 0 = 2m Jp02 (m ) + (2m p2 )Jp2 (m ).

2m
0

(Notice that u(0) = Jp (0) = 0 for p > 0. So p2 u2 (0) = 0 for p 0).


Z 1
1 1 2
xJp2 (m x)dx = Jp02 (m ) = Jp+1 (m ).
0 2 2
d  p
x Jp (x) = xp Jp+1 (x) leads to

For,
dx
p
Jp0 (x) = Jp (x) Jp+1 (x),
x
p
Jp0 (m ) = Jp (m ) Jp+1 (m ) = Jp+1 (m ).
m
Some Special Functions Dr. Suresh Kumar, BITS Pilani 14

0.4.1 Fourier-Bessel Series


In mathematical physics, it is often necessary to expand a given function in terms of Bessel func-
tions. The simplest and most useful expansions are of the form

X
f (x) = an Jp (n x) = a1 Jp (1 x) + a2 Jp (2 x) + ..........., (26)
n=1

where f (x) is defined on the interval 0 x 1 and n are positive zeros of some fixed Bessel
function Jp (x) with p 0. Now multiplying (26) by xJp (n x) and integrating from x = 0 to x = 1,
we get
Z 1
1 2
xf (x)Jp (n x)dx = an Jp+1 (n ),
0 2
which gives
Z 1
2
an = 2 xf (x)Jp (n x)dx.
Jp+1 (n ) 0

Ex. 0.4.1. Express f (x) = 1 in terms of the functions J0 (n x)

Sol. 0.4.1. We have


Z 1
2
an = 2
xf (x)J0 (n x)dx
J1 (n ) 0
 1  
2 1 d
= xJ1 (n x) , [xJ1 (x)] = xJ0 (x)
J12 (n ) n 0 dx
2 J1 (n )
= 2

J1 (n ) n
2
=
n J1 (n )
So the required Fourier-Bessel series is

X 2
1= J0 (n x).
n=1
n J1 (n )

Convergence of Fourier-Bessel Series


Assume that f (x) and f 0 (x) have at most finite number of jump discontinuities on [0, 1]. If
x (0, 1), then the Bessel series converges to f (x) when x is a point of continuity of f (x), and
converges to 12 [f (x) + f (x+)] when x is a point of discontinuity. At x = 1, the series converges
to 0 regardless of the nature of f (x) because every Jp (n ) = 0. The series converges to 0 if p > 0
and to f (0+) if p = 0.

You might also like