You are on page 1of 14

Physics 116C Solutions to Homework Set #2 Fall 2012

1. Boas, problem p.578, 12.7-5


Show that
_
1
1
P

(x)dx = 0, for > 0.


This is immediate once we remember that P
0
(x) = 1. It then follows that
_
1
1
P

(x)dx =
_
1
1
P

(x)dx 1 =
_
1
1
P

(x)P
0
(x)dx = 0 , for = 0 ,
because of the orthogonality of the Legendre polynomials on the interval 1 x 1.
2. Boas, problem p.580, 12.8-6
This problem asks for you to provide an alternate proof of
_
1
1
[P

(x)]
2
dx =
2
2 + 1
. (1)
Multiplying the recursion relation,
(2 + 1)P

(x) = P

+1
(x) P

1
(x) ,
by P

(x) and integrating between 1 and 1, we have


(2 + 1)
_
1
1
[P

(x)]
2
dx =
_
1
1
P

(x)[P

+1
(x) P

1
(x)]dx =
_
1
1
P

(x)P

+1
(x)dx . (2)
The last step has obtained by noting that P
1
(x) is a polynomial of degree 1 (which
means that the highest power that appears in the polynomial is x
1
. Taking the derivative,
it follows that P

1
(x) is a polynomial of degree 2. Consequently, using eq. (7.6) on
p. 578 of Boas, it follows that
_
1
1
P

(x)P

1
(x)dx = 0 ,
which justies the last step taken in eq. (2).
Finally, we evaluate the last integral in eq. (2) using integration by parts, which yields
_
1
1
[P

(x)]
2
dx =
1
2 + 1
_
P

(x)P
+1
(x)

1
1

_
1
1
P

(x)P
+1
(x)dx
_
=
2
2 + 1
. (3)
The nal step above was obtained by again applying eq. (7.6) on p. 578 of Boas [which
implies that the integral on the right hand side of eq. (3) vanishes], and using P

(1) = 1
and P

(1) = (1)

. Hence, we have conrmed the result of eq. (1).


1
3. Boas, problem p.581, 12.9-2
Expand the following function in Legendre series:
f(x) =
_
0, 1 < x < 0 ,
x, 0 < x < 1 .
We expand f(x) in Legendre polynomials, f(x) =

=0
c

(x), with unknown coe-


cients c

. Because of the orthogonality of the Legendre polynomials, if we multiply f(x) by


P

(x) and integrate between 1 and 1, we obtain:


_
1
1
f(x)P

(x)dx =

m=0
c
m
_
1
1
P
m
(x)P

(x)dx = c

2
2 + 1
.
It follows that the c

are given by:


c

=
2 + 1
2
_
1
1
f(x)P

(x)dx =
2 + 1
2
_
1
0
xP

(x)dx.
We now use the recursion relation (5.8a) from page 570 of Boas,
P

(x) = (2 1)xP
1
(x) ( 1)P
2
(x) ,
so that
xP

(x) =
+ 1
2 + 1
P
+1
(x) +

2 + 1
P
1
(x) ,
and
c

=
1
2
_
( + 1)
_
1
0
P
+1
(x)dx +
_
1
0
P
1
(x)dx
_
.
Finally, we use the results of problem 12-23.3 (problem 12 of homework set 1):
_
1
0
P
2n
(x)dx = 0 , n > 0 and
_
1
0
P
2n+1
(x)dx =
(1)
n
(2n 1)!!
2
n+1
(n + 1)!
=
_
1/2
n
_
. (4)
For = 2n + 1, we use eq. (4) to obtain:
c
2n+1
=
1
2
_
(2n + 2)
_
1
0
P
2n+2
(x)dx + (2n + 1)
_
1
0
P
2n
(x)dx
_
=
n0
1
2
_
1
0
dx =
1
2

n0
,
which implies that c
1
=
1
2
and c
2n+1
= 0 for n = 1, 2, 3, . . ..
For = 2n, we again use eq. (4) to obtain:
c
2n
=
1
2
_
(2n + 1)
_
1
0
P
2n+1
(x)dx + 2n
_
1
0
P
2n1
(x)dx
_
=
1
2
_
(2n + 1)
(1)
n
(2n 1)!!
2
n+1
(n + 1)!
+ 2n
(1)
n1
(2n 3)!!
2
n
n!
_
=
(1)
n
(2n 3)!!
2
n+2
(n + 1)!
(1 4n) .
2
The rst few coecients are
c
0
=
1
4
, c
1
=
1
2
, c
2
=
5
16
, c
3
= 0 , c
4
=
3
32
, .
In summary, we have obtained:
f(x) =
_
0, 1 < x < 0
x, 0 < x < 1
_
=
1
4
P
0
(x) +
1
2
P
1
(x) +
5
16
P
2
(x)
3
32
P
4
(x) + . . .
=
1
2
P
1
(x)

n=0
(1)
n
(2n 3)!!(4n + 1)
2
n+2
(n + 1)!
P
2n
(x) .
4. Boas, problem p.582, 12.9-16
Prove the least square approximation property of the Legendre polynomials. Namely,
given f(x) the function to be approximated and the orthonormal rescaled Legendre poly-
nomials, denoted by p

(x), we can expand f(x) in this basis:


f(x) = c
0
p
0
(x) + c
1
p
1
(x) + c
2
p
2
(x) + . . . =

=0
c

(x) .
The coecient c

are found by multiplying f(x) by p

and integrating between 1 and 1,


_
1
1
f(x)p

(x)dx =
_
1
1

m=0
c
m
p
m
(x)p

(x)dx =

m=0
c
m

m
= c

. (5)
Now let F(x) = b
0
p
0
(x) + b
1
p
1
(x) + b
2
p
2
(x) be the (unknown) quadratic polynomial
satisfying the least square condition, that is, such that
I =
_
1
1
[f(x) F(x)]
2
dx
is a minimum. Squaring the bracket and using the orthonormality of the p

s we can rewrite
I as
I =
_
1
1
[f
2
(x) + F
2
(x) 2f(x)F(x)]dx
=
_
1
1
_
f
2
(x)

dx + b
2
0
+ b
2
1
+ b
2
2
2b
0
c
0
2b
1
c
1
2b
2
c
2
=
_
1
1
f
2
(x)dx + (b
0
c
0
)
2
+ (b
1
c
1
)
2
+ (b
2
c
2
)
2
c
2
0
c
2
1
c
2
2
.
We are looking for the unknown coecients b

that minimize I. Note that there are only


three terms in I that depend on the bs, and they form a sum of squared numbers. It
3
follows that I is minimum when these terms are zero, i.e., when b

= c

. Thus, we have
found that the coecients of the quadratic polynomial that best approximates a function
f(x) are the coecients of the Legendre expansion of the function itself.
Now we can generalize this result to approximate any function by a polynomial of
degree n. Writing the polynomial as F(x) =

n
=0
b

(x) and minimizing the integral


I =
_
1
1
[f(x) F(x)]
2
dx, we nd as before terms that depend on the bs of the form:
(b
0
c
0
)
2
+ (b
1
c
1
)
2
+ . . . + (b
n
c
n
)
2
.
Again, this sum is minimized for b

= c

, i.e., when the approximated polynomial is given


by the Legendre series expansion of the function.
5. Boas, problem p.584, 12.10-3
Show that the functions P
m

(x) for each m are a set of orthogonal functions on (1, 1).


That is, show that
_
1
1
P
m

(x)P
m
n
(x)dx = 0 , = n. (6)
We recall the associated Legendre equation
(1 x
2
)P
m

2xP
m

+
_
( + 1)
m
2
1 x
2
_
P
m

= 0 , (7)
and rewrite it as
d
dx
_
(1 x
2
)P
m

+
_
( + 1)
m
2
1 x
2
_
P
m

= 0 . (8)
The same equation holds for P
m
n
(x) with the replacement n. Multiplying this latter
equation by P
m
n
(x) and the former equation [eq. (8)] by P
m

(x), and subtracting the two


resulting equations, we nd
P
m
n
d
dx
_
(1 x
2
)P
m

P
m

d
dx
_
(1 x
2
)P
m
n

+ [( + 1) n(n + 1)] P
m

P
m
n
= 0 ,
which can be rewritten as
d
dx
_
(1 x
2
)(P
m
n
P
m

P
m

P
m
n

+ [( + 1) n(n + 1)] P
m

P
m
n
= 0 .
Integrating between 1 and 1, we have
(1 x
2
)(P
m
n
P
m

P
m

P
m
n

1
1
+ [( + 1) n(n + 1)]
_
1
1
P
m

(x)P
m
n
(x)dx = 0 . (9)
The rst term in eq. (9) is zero, so we have proven the orthogonality relation given in eq. (6)
for = n.
4
6. Boas, problem p.584, 12.10-7
Show that
d
m
dx
m
(x
2
1)

=
( m)!
( + m)!
(x
2
1)
m
d
+m
dx
+m
(x
2
1)

. (10)
Recall that Leibniz rule can be written as
d
k
dx
k
(uv) =
k

n=0
_
k
n
_
d
n
u
dx
n
d
kn
v
dv
kn
=
k

n=0
k!
n!(k n)!
d
n
u
dx
n
d
kn
v
dv
kn
. (11)
We will also need to compute p derivatives of x
q
where p and q are integers. For p q, we
have
d
p
dx
p
(x
q
) = q(q 1)(q 2) (q p + 1)x
qp
=
q!
(q p)!
x
qp
. (12)
In fact, eq. (12) holds for any non-negative integers p and q. If p > q, then (d
p
/dx
p
)x
q
= 0
since after q derivatives one obtains a constant, and any further derivatives then yield zero.
Recalling that n! = (n+1) and (x) is innite when x is a non-positive integer, it follows
that 1/(x) = 0 when x is a non-positive integer, or equivalently 1/n! = 0 when n is a
negative integer. Indeed, eq. (12) is consistent when p > q since then 1/(q p)! is the
inverse of the factorial of a negative integer which is zero. We can immediately generalize
eq. (12) to
d
p
dx
p
[(x + c)
q
] =
q!
(q p)!
(x + c)
qp
, (13)
for any constant c by noting that d(x + c) = dx.
To prove eq. (10), it is convenient to (x
2
1)

= (x+1)

(x1)

and then evaluate both


sides of eq. (10) with the help of Leibniz rule given in eq. (11). We rst compute the left
hand side of eq. (10),
d
m
dx
m
(x
2
1)

=
m

k=0
( m)!
k!( mk)!
d
k
dx
k
(x + 1)

d
mk
dx
mk
(x 1)

=
m

k=0
( m)!
k!( mk)!
!
( k)!
(x + 1)
k
!
(m + k)!
x
m+k
= ( m)![!]
2
m

k=0
(x + 1)
k
(x 1)
m+k
k!( mk)!( k)!(m + k)!
, (14)
after making use of eq. (13).
5
Likewise,
d
+m
dx
+m
(x
2
1)

=
+m

k=0
( + m)!
k!( + mk)!
d
k
dx
k
(x + 1)

d
+mk
dx
+mk
(x 1)

=
+m

k=0
( + m)!
k!( + mk)!
!
( k)!
(x + 1)
k
!
(k m)!
x
km
= ( + m)![!]
2
+m

k=0
(x + 1)
k
(x 1)
km
k!( + mk)!( k)!(k m)!
.
Note that 1/(k m)! = 0 if k < m and 1/( k)! = 0 if k > . Thus only the terms with
m k contribute to the last sum in eq. (15). Hence,
d
+m
dx
+m
(x
2
1)

= ( + m)![!]
2

k=m
(x + 1)
k
(x 1)
km
k!( + mk)!( k)!(k m)!
.
We can relabel the above sum by taking k k + m, in which case the new summation
index k now runs from k = 0 to k = m. Thus,
d
+m
dx
+m
(x
2
1)

= ( + m)![!]
2
m

k=0
(x + 1)
km
(x 1)
k
(k + m)!( k)!( k m)!k!
. (15)
Using eq. (15) , we conclude that
( m)!
( + m)!
(x
2
1)
m
d
+m
dx
+m
(x
2
1)

= ( m)![!]
2
m

k=0
(x + 1)
k
(x 1)
m+k
(k + m)!( k)!( k m)!k!
. (16)
Comparing equations (14) and (16) we see that the identity in eq. (10) is correct.
7. Boas, problem p.590, 12.12-4
Use the series form of the Bessel function to show that
d
dx
J
0
(x) = J
1
(x) (17)
The Bessel function J
p
(x) has the following series representation [cf. eq. (12.9) on p. 590
of Boas]:
J
p
(x) =

n=0
(1)
n
(n + 1)(n + 1 + p)
_
x
2
_
2n+p
. (18)
6
Setting p = 0 and taking the derivative yields:
d
dx
J
0
(x) =
d
dx

n=0
(1)
n
(n + 1)(n + 1)
_
x
2
_
2n
=

n=1
(1)
n
n
(n + 1)(n + 1)
_
x
2
_
2n1
,
after noting that the n = 0 term does not contribute to the sum above. Since the sums
above converge uniformly for all x, it is permissible to dierentiate the series representation
of J
0
(x) term by term. Relabeling n n + 1, the new summation index n will now run
from n = 0 to innity, which yields
d
dx
J
0
(x) =

n=0
(1)
n+1
(n + 1)
(n + 2)(n + 2)
_
x
2
_
2n+1
=

n=0
(1)
n
(n + 1)(n + 2)
_
x
2
_
2n+1
= J
1
(x) .
In the obtaining the nal sum, we used (n + 2) = (n + 1)(n + 1), and recognized the
resulting expression for J
1
(x) [cf. eq. (18) with p = 1].
8. Boas, problem p.590, 12.12-9
Prove that
_
x
2
J
1/2
(x) = sin x.
We substitute p =
1
2
in eq. (18),
_
x
2
J
1/2
(x) =
_
x
2

n=0
(1)
n
(n + 1)(n +
3
2
)
_
x
2
_
2n+
1
2
=

n=0
(1)
n
(n + 1)(n +
3
2
)
_
x
2
_
2n+1
. (19)
To simplify the above result, we apply the duplication formula for the Gamma function
[cf. problem 11.79 on p. 545 of Boas],
(2n) =
1

2
2n1
(n)(n +
1
2
) , (20)
Substituting n n + 1 in eq. (20), we obtain
2
2n+1
(n + 1)(n +
3
2
) =

(2n + 2) ,
Inserting this result into eq. (19) and writing (2n + 2) = (2n + 1)!, we end up with
_
x
2
J
1/2
(x) =

n=0
(1)
n
(2n + 1)!
x
2n+1
= sin x.
7
9. Boas, problem p.591, 12.13-6
Show from
N
p
(x) =
cos(p)J
p
(x) J
p
(x)
sin(p)
, (21)
that
N
(2n+1)/2
(x) = (1)
n+1
J
(2n+1)/2
(x) .
For p =
1
2
(2n + 1) we have cos p = 0 and sin p = (1)
n
. Plugging those values back
in eq. (21), it follows that
N
(2n+1)/2
(x) = (1)
n
J
(2n+1)/2
(x) .
10. Boas, problem p.593, 12.15-8
Using J
p1
(x) J
p+1
(x) = 2J

p
(x) show that
_

0
J
1
(x)dx =
_

0
J
3
dx = . . . =
_

0
J
2n+1
(x)dx ,
and
_

0
J
0
(x)dx =
_

0
J
2
(x)dx = . . . =
_

0
J
2n
(x)dx ,
and prove that
_

0
J
n
(x)dx = 1 for all integral n.
For any odd p = 2n + 1, using the recursion formula given in eq. (15.4) on p. 592 of
Boas, it follows that
_

0
J
2n+2
(x)dx =
_

0
_
J
2n
(x) 2J

2n+1
(x)

dx =
_

0
J
2n
(x)dx 2J
2n+1
(x)

0
=
_

0
J
2n
(x)dx , (22)
where we have used that fact that lim
x0
J
2n+1
(x) = 0 and lim
x
J
2n+1
(x) = 0 for any
non-negative integer n. The x 0 limit follows from the series representation of the Bessel
function given in eq. (18), and the x limit follows from the observation derived in
class that the amplitude of the Bessel function graph falls o as 1/

x as x . The
latter behavior is also apparent from the asymptotic forms given in the table on p. 604 of
Boas. Since eq. (22) is true for any non-negative integer n, it immediately follows that
_

0
J
0
(x)dx =
_

0
J
2
(x)dx = . . . =
_

0
J
2n
(x)dx ,
8
as required.
For any even p = 2n > 0, the a similar analysis applies,
_

0
J
2n+1
(x)dx =
_

0
(J
2n1
(x) 2J

2n
(x))dx =
_

0
J
2n1
(x)dx 2J
2n
(x)

0
=
_

0
J
2n1
(x)dx , (23)
for any non-negative integer n. Hence,
_

0
J
1
(x)dx =
_

0
J
3
dx = . . . =
_

0
J
2n+1
(x)dx ,
as required.
To complete the problem, we only have to compute two integrals. First, we use eq. (17)
to obtain
_

0
J
1
(x)dx =
_

0
d
dx
J
0
(x)dx = J
0
(x)

0
= 1 .
In obtaining this result, we used the fact that lim
x0
J
0
(x) = 1 [a fact that is evident from
the series representation of J
0
(x) given in eq. (18)], and lim
x
J
0
(x) = 0, due to the 1/

x
behavior as x . Second, we employ the Laplace transform of the Bessel function given
by entry L23 of the Laplace Transform Table on p. 470 of Boas,
L[J
0
(ax)]
_

0
J
0
(ax)e
px
dx =
1
_
p
2
+ a
2
.
Setting p = 0 and a = 1 then yields
_

0
J
0
(x)dx = 1 .
Combining all the results above, we have successfully veried that
_

0
J
n
(x)dx = 1 , (24)
for all non-negative integers n. Since J
n
(x) = (1)
n
J
n
(x) for any integer n [cf. eq. (13.2)
on p. 590 of Boas], we conclude that eq. (24) actually holds for any integer n.
9
11. Boas, problem p.606, 12.21-9
Solve the following dierential equation by the Frobenius method.
x
2
y

+ (x
2
3x)y

+ (4 2x)y = 0 . (25)
Verify that the conditions of Fuchss theorem are satised. Observing that the Frobenius
method only yields one solution in this case, nd the second solution, knowing that the
second solution is equal to ln x times the rst solution plus another Frobenius series.
First, we observe that the conditions of Fuchss theorem are satised. These conditions
correspond to the requirement that for the dierential equation y

+ f(x)y

+ g(x)y = 0,
the quantities xf(x) and x
2
g(x) are non-singular at x = 0. If we divide eq. (25) by x
2
, we
can identify f(x) = 1 3/x and g(x) = (4 2x)/x
2
, which clearly satisfy the requirements
of Fuchss theorem.
Next, we insert the generalized series y(x) =

n=0
a
n
x
n+s
into eq. (25):

n=0
(n + s)(n + s 1)a
n
x
n+s
+

n=0
[(n + s)a
n
x
n+s+1
3(n + s)a
n
x
n+s
]
+

n=0
[4a
n
x
n+s
2a
n
x
n+s+1
] = 0 ,
which we can rewrite as
[s(s 1) 3s + 4]a
0
x
s
+

n=0
_
[(n + s + 1)(n + s 3) + 4]a
n+1
+ (n + s 2)a
n
_
x
n+s+1
= 0 . (26)
The indicial equation is obtained by setting the coecient of x
s
to be zero,
s
2
4s + 4 = 0 = (s 2)
2
= 0 = s = 2 . (27)
We see that we only have one (double) root for s, which implies that the Frobenius method
yields only one series solution. The rst solution y
1
(x) is found using the value s = 2 in
the recursion relation, which is obtained by setting the coecient of x
n+s+1
above (after
setting s = 2) to zero,
[(n + 3)(n 1) + 4]a
n+1
+ na
n
= 0 , for n = 0, 1, 2, 3, . . . .
Noting that (n + 3)(n 1) + 4 = (n + 1)
2
, it follows that,
a
n+1
=
na
n
(n + 1)
2
, for n = 0, 1, 2, 3, . . . . (28)
Eq. (28) implies that a
1
= a
2
= . . . = a
n
= 0 (for any positive integer n). Thus, the only
non-zero coecient is a
0
, and we conclude that the rst solution of eq. (25) is proportional
to
y
1
(x) = x
2
. (29)
It is a simple matter to verify this result by direct substitution into eq. (25).
10
The second solution must be of the form y(x) = a
0
y
1
(x) ln x+

n=0
b
n
x
n+s
, where s = 2
as before. Substituting y
1
(x) = x
2
as determined in eq. (29), we write
y(x) = a
0
x
2
ln x +

n=0
b
n
x
n+s
, (30)
y

(x) = 2a
0
xln x + a
0
x +

n=0
(n + s)b
n
x
n+s1
,
y

(x) = 2a
0
ln x + 3a
0
+

n=0
(n + s)(n + s 1)b
n
x
n+s2
.
Inserting these results into eq. (25) yields
(2a
0
lnx + 3a
0
)x
2
+ (x
2
3x)(2a
0
xlnx + a
0
x) + (4 2x)a
0
x
2
ln x
+

n=0
(n + s)(n + s 1)b
n
x
n+s
+ (x
2
3x)

n=0
(n + s)b
n
x
n+s1
+ (4 2x)

n=0
b
n
x
n+s
= 0 .
Indeed the terms proportional to ln x cancel exactly as they should. What remains is:
a
0
x
3
+[s(s1)3s+4]b
0
x
s
+

n=0
_
[(n+s+1)(n+s3)+4]b
n+1
+(n+s2)b
n
_
x
n+s+1
= 0 .
The only dierence from our previous calculation of y
1
(x) is the presence of a
0
x
3
above.
The coecient of b
0
x
s
vanishes since the indicial equation is satised, which yields s = 2
as expected. Inserting s = 2 into the above result then yields
a
0
x
3
+

n=0
_
[(n + 3)(n 1) + 4]b
n+1
+ nb
n
_
x
n+3
= 0 .
Noting that (n + 3)(n 1) + 4 = (n + 1)
2
, and separating out the n = 0 term from the
sum, we obtain
(a
0
+ b
1
)x
3
+

n=1
_
(n + 1)
2
b
n+1
+ nb
n
_
x
n+3
= 0 .
The rst term above yields b
1
= a
0
, and the resulting recursion relation is
b
n+1
=
nb
n
(n + 1)
2
, for n = 1, 2, 3, . . . .
It is not dicult to deduce the general term,
b
n
=
(1)
n
a
0
(n 1)!
[n!]
2
, for n = 1, 2, 3, . . . .
11
Writing n! = n (n 1)!, the above result can be rewritten as
b
n
=
(1)
n
a
0
n n!
, for n = 1, 2, 3, . . . . (31)
Note that b
0
= 0 is an arbitrary constant that is not xed by the above procedure. Inserting
these results back into eq. (30) [after setting s = 2] then yields,
y(x) = x
2
(b
0
+ a
0
ln x) + a
0

n=1
(1)
n
n n!
x
n+2
.
This solution depends on two independent constants, and thus provides the most general
solution to eq. (25). Since the rst solution was obtained in eq. (29) [and corresponds to
the b
0
x
2
term above], we conclude that the second solution is proportional to
y
2
(x) = x
2
ln x +

n=1
(1)
n
n n!
x
n+2
. (32)
Expanding out the sum yields,
y
2
(x) = x
2
log x x
3
+
1
4
x
4

1
18
x
5
+
1
96
x
6
+ . . . .
ADDED NOTE:
One can employ an alternative method for obtaining y
2
(x), which makes use of Abels
formula. Our strategy is to use eqs. (9) and (14) of the class handout entitled Applications
of the Wronskian to ordinary linear dierential equations. First we compute the Wronskian,
W(x) = exp
_

_
x
_
1
3
x
_
dx
_
= cx
3
e
x
,
where we have dropped the overall constant, as it is not needed for this computation. Then,
it follows that
y
2
(x) = y
1
(x)
_
x
W(x)
[y
1
(x)]
2
= x
2
_
x
e
x
x
dx. (33)
The indenite integral above can be evaluated by inserting above the power series for e
x
,
_
x
e
x
x
dx =

n=0
_
x
(1)
n
x
n1
n!
dx = ln x +

n=1
(1)
n
x
n
n n!
. (34)
Using this result in eq. (33) yields:
y
2
(x) = x
2
ln x +

n=1
(1)
n
x
n+2
n n!
,
which conrms the result obtained in eq. (32).
12
Note that eq. (34) allows us to write the second solution to eq. (25) in terms of a special
function introduced in problem 11.104 on p. 552 of Boas. Namely, the exponential integral
is dened as
Ei(x) =
_
x

e
t
t
dt .
A related exponential integral is dened by
E
1
(x) =
_

x
e
t
t
dt = Ei(x) .
The denite integral E
1
(x) diers from the indenite integral that appears in eq. (34) by
an additive constant of integration. One can show that
1
E
1
(x) = + ln x +

n=1
(1)
n
x
n
n n!
,
where is Eulers constant. Thus, one can choose the second solution to eq. (25) to be
2
y
2
(x) = x
2
E
1
(x) . (35)
Note that E
1
(x) = (0, x) [cf. the solution to problem 11.105(a) on pp. 552 and 802 of
Boas]. Thus, one can also express the second solution to eq. (25) in terms of the incomplete
Gamma function, (p, x) [dened in problem 11.102 on p. 551 of Boas], with p = 0.
BONUS MATERIAL:
If one sets the coecient of x
n+s+1
in eq. (26) to zero before setting s = 2, one obtains
(after a little algebraic manipulation) the recursion relation,
(n + s 1)
2
a
n+1
(s) + (n + s 2)a
n
(s) = 0 , for n = 0, 1, 2, 3, . . . ,
where the explicit s dependence of the coecients is indicated by writing a
n
(s) above. It
is not dicult to work out the general coecient for arbitary n,
a
n
(s) =
(1)
n
(s 2)(s 1)s(s + 1) (s + n 3)a
0
(s 1)
2
s
2
(s + 1)
2
(s + n 3)
2
(s + n 2)
2
=
(1)
n
(s 2)a
0
(s 1)s(s + 1) (s + n 3)(s + n 2)
2
, for n = 1, 2, 3, . . . , (36)
where a
0
= 0 is independent of s by denition. In particular, we see that by setting s = 2,
we get a
n
(s)

s=2
= 0 for n = 1, 2, 3, . . ., which conrms the result obtained below eq. (28).
This immediately implies that the rst solution of eq. (25) is y
1
(x) = a
0
x
2
.
1
See, e.g., http://en.wikipedia.org/wiki/Exponential integral and references therein.
2
We could have chosen y
2
(x) = x
2
[ + E
1
(x)], which would coincide with the result obtained in
eq. (32). But, x
2
is proportional to y
1
(x), so there is no harm in dening a second solution to eq. (25)
that omits this latter term. The overall minus sign is also irrelevant, since any constant times x
2
E
1
(x) is
also a solution to eq. (25).
13
Referring to eq. (15) of the class handout entitled Series solutions to a second order
linear dierential equation with regular singular points, we can determine the coecients
b
n
that enter into eq. (30) by using eq. (36) and computing
b
n
=
_
a
n
s
_
s=2
=
(1)
n
a
0
n n!
, for n = 1, 2, 3, . . . ,
which reproduces eq. (31). Thus, eq. (15) of the class handout just cited implies that the
second solution of eq. (25) is given by
y
2
(x) = y
1
(x) ln x + x
2

n=1
b
n
x
n
,
with y
1
(x) = a
0
x
2
and the b
n
given by eq. (31). This is precisely the result obtained in
eq. (32) up to the overall multiplicative constant a
0
[which we are free to include or omit,
as the overall normalization of the solution is a matter of convention].
12. Boas, problem p.615, 12.23-9
Evaluate
_
1
1
xP

(x)P
n
(x)dx, for n . (37)
We use the recursion relation given in eq. (5.8a) on p. 570 of Boas with + 1:
( + 1)P
+1
(x) = (2 + 1)xP

(x) P
1
(x) . (38)
Multiplying by P
n
(x) and integrating from 1 to 1, we get
_
1
1
xP

(x)P
n
(x)dx =
1
2 + 1
_
1
1
[( + 1)P
+1
(x) + P
1
(x)]P
n
(x)dx
=
1
2 + 1
_
( + 1)
2
2n + 1

+1,n
+
2
2n + 1

1,n
_
, (39)
where we have used the orthogonality relation of the Legendre polynomials [cf. eq. (8.4) on
p. 580 of Boas],
_
1
1
P

(x)P
n
(x) dx =
2
2 + 1

n
.
By assumption, n . Hence,
+1,n
= 0, since this Kronecker delta is non-vanishing only
when n = + 1 which contradicts our assumption that n . Thus, we are left with
_
1
1
xP

(x)P
n
(x)dx =
2
(2 + 1)(2n + 1)

1,n
=
_

_
2
(2 + 1)(2 1)
, for n = 1 ,
0 , for any other n .
14

You might also like