You are on page 1of 14

Math221: HW# 8 solutions

Andy Royston
December 4, 2005

12.3.3 The Leibniz rule is


n    
(n)
X n n n!
(uv) = u(k) v (n−k) , = . (1)
k k (n − k)!k!
k=0

f (n) means nth derivative and the coefficients are the binomial coefficients, displayed in Pascal’s
triangle. Now apply with u(x) = x2 and v(x) = sin x. Note that derivatives of x2 higher than 2
will vanish. Hence
6·5
(x2 sin x)(6) = x2 (sin x)(6) + 6(2x)(sin x)(5) + (2)(sin x)(4) . (2)
2
Now note that the derivatives of sine repeat themselves with a cycle of four. (sin x)(6) = (sin x)(2) =
− sin x, (sin x)(5) = (sin x)(1) = cos x, and (sin x)(4) = sin x. Thus

(x2 sin x)(6) = −x2 sin x + 12x cos x + 30 sin x. (3)

12.3.5 Now let u(x) = x2 and v(x) = e−x . Again, only the first three terms survive. This time
note that the nth derivative of e−x just brings down n factors of −1. Therefore, (e−x )(n) = (−1)n e−x .
Then

100 · 99
(x2 e−x )(100) = x2 (−1)100 e−x + 100(2x)(−1)99 e−x + (2)(−1)98 e−x
2
= e−x (x2 − 200x + 9900). (4)

12.4.3 The Rodriguez formula is

1 dl 2
Pl (x) = (x − 1)l . (5)
2l l! dxl
The trick to evaluating the derivative for annoyingly large l is to use the Leibniz rule. We will need
to do this in the next problem, but here it is simpler to just multiply (x2 − 1)l out. We have

1
1 d0 2 0 1
P0 (x) = (x − 1) = 1 = 1, (6)
20 0! dx0 1·1
1 d1 2 1 d 2
P1 (x) = 1 1
(x − 1)1 = (x − 1) = x. (7)
2 1! dx 2 dx
1 d2 2 2 1 d2 4 1
P2 (x) = 2 2
(x − 1) = 2
(x − 2x2 + 1) = (4 · 3x2 − (2)2 · 1)
2 2! dx 8 dx 8
1
= (3x2 − 1). (8)
2
Then

1 d3 2 3 1 d3 6
P3 (x) = (x − 1) = (x − 3x4 + 3x2 − 1)
23 3! dx3 3 · 24 dx3
1 6·5 3 3 1
= 4
(6 · 5 · 4x3 − (3)4 · 3 · 2x) = 2
x − x = (5x3 − 3x) (9)
3·2 3·2 2 2
and

1 d4 2 4 1 d4 8
P4 (x) = (x − 1) = (x − 4x6 + 6x4 − 4x2 + 1)
24 4! dx4 4 · 3 · 25 dx4
1 7·5 4 3·5 3
= 5
(8 · 7 · 6 · 5x4 − (4)6 · 5 · 4 · 3x2 + (6)4 · 3 · 2) = x − 2 + 3
4·3·2 4·2 2 2
1
= (35x4 − 30x2 + 3). (10)
8

12.4.4 This is a very important result. Plug in Rodriguez’ formula for Pl (x) and integrate by
parts repeatedly. Each time the power of x will go down one, while the order of the derivative
dn dn−1
decreases by one (if dv = dx n f then v = dxn−1 v). Then, writing out the first few terms,

Z 1 Z 1 l
m 1 m d
x Pl (x)dx = l x l
(x2 − 1)l dx
−1 2 l! −1 dx
 l−1
1 Z 1 l−1

1 m d 2

l m−1 d 2 l
= l x (x − 1) − m x (x − 1) dx
2 l! dxl−1 −1 −1 dxl−1
 l−1
1 l−2
1 Z 1 l−2

1 m d 2

l m−1 d 2

l m−2 d 2 l
= l x (x − 1) − mx (x − 1) + m(m − 1) x (x − 1) dx
2 l! dxl−1 −1 dxl−2 −1 −1 dxl−2
= ....

Now, eventually this process must terminate, because eventually there will be no powers of x left.
(This will happen before we run out of derivatives on (x2 − 1) because we are assuming l > m).

2
There will be no powers left on x after we’ve applied integration by parts m times. Let us write
down what that last term will look like:

Z 1 
1
m
x Pl (x)dx = l . . .
−1 2 l!
1 1
dl−m 2 l−m
Z 
m−1
m d 2 l
+(−1) m(m − 1) · · · (2)x l−m (x − 1) + (−1) m! l−m
(x − 1) dx . (11)
dx −1 −1 dx

Now this last term can be integrated directly; it is just the integral of a derivative:
Z 1 l−m Z 1  l−m−1  l−m−1
1
d 2 l d d 2 l d 2

l
l−m
(x − 1) dx = l−m−1
(x − 1) dx = (x − 1) . (12)
−1 dx −1 dx dx dxl−m−1 −1

Thus, we have reduced the original integral to a sum of terms to be evaluated at ±1. It is not hard
to see the pattern, and write down the general term in the sum. We have
Z 1 m 1
m 1 X m! m−k dl−k−1 2
l
x Pl (x)dx = l x l−k−1
(x − 1) . (13)
−1 2 l! k=0 k! dx −1

Now what we will show is that each of these terms vanishes at the end points ±1. In order to
do this, we must evaluate the arbitrary order derivative of (x2 − 1)l . Actually, from looking at the
sum, the highest derivative is order l − 1. The lowest derivative is order l − m − 1. We know l > m.
In the most extreme case l = m + 1, the lowest order derivative is actually 0. Hence, what we need
to show is that
1
dn 2
l
n
(x − 1) = 0, ∀n = 0, 1, . . . , l − 1. (14)
dx −1

If we can show this, then every term in (13) vanishes and the result is obtained. We can do this
by using the Leibniz formula. Observe that

(x2 − 1)l = ((x − 1)(x + 1))l = (x − 1)l (x + 1)l (15)


so let u(x) = (x − 1)l and v(x) = (x + 1)l . Then it is extremely easy to take derivatives of u and
v. One would employ the chain rule, but the derivative of the stuff inside is just 1, so it’s just like
taking derivatives of xl . In general result is

dp l dp−1 l−1 dp−2


(x − 1) = l (x − 1) = l(l − 1) (x − 1)l−2 = · · ·
dxp dxp−1 dxp−2
l!
= l(l − 1) · (l − p + 1)(x − 1)l−p = (x − 1)l−p , (16)
p!
p
d l!
p
(x + 1)l = (x + 1)l−p . (17)
dx p!

3
Hence, the general result is

n
dn 2 l
X n! dp l d
n−p

n
(x − 1) = p
(x − 1) n−p (x + 1)l
dx p=0
p!(n − p)! dx dx
n
X n! l! l!
= · · (x − 1)l−p (x + 1)l−(n−p) . (18)
p=0
p!(n − p)! p! (n − p)!

Now, since n ≤ l − 1, every term in this sum has a product of (x − 1) to some non-zero power and
(x + 1) to some non-zero power. Thus, when we set x = ±1, one or the other factor will give zero
in every term. Therefore, we get the result (14). It follows from (13) that
Z 1
xm Pl (x)dx = 0, ∀m = 0, 1, . . . , l − 1. (19)
−1

12.5.3 The recursion relation is

lPl (x) = (2l − 1)xPl−1 (x) − (l − 1)Pl−2 (x). (20)


For l = 2 this gives
1
2P2 = 3xP1 − P0 = 3x2 − 1 ⇒ P2 (x) = (3x2 − 1). (21)
2
For l = 3,

5x 15 5 15 9
3P3 = 5xP2 − 2P1 = (3x2 − 1) − 2x = x3 − x − 2x = x2 − x
2 2 2 2 2
5 3 3 1
⇒ P3 (x) = x − x = (5x3 − 3x). (22)
2 2 2
For l = 4,

7x 3 35 21 9 3
4P4 = 7xP3 − 3P1 = (5x3 − 3x) − (3x2 − 1) = x4 − x2 − x2 +
2 2 2 2 2 2
35 4 2 3
= x − 15x +
2 2
35 4 15 2 3 1 4 2
⇒ P4 (x) = x − x + = (35x − 30x + 3). (23)
8 4 8 8
For l = 5,

4
9x
(35x4 − 30x2 + 3) − 2(5x3 − 3x)
5P5 = 9xP4 − 4P3 =
8
215 5 135 3 27 1
= x − x + x − 10x3 + 6x = (315x5 − 350x3 + 75x)
8 4 8 8
1
⇒ P5 (x) = (63x5 − 70x + 15x). (24)
8
For l = 6,

11x 5
(63x5 − 70x3 + 15x) − (35x4 − 30x2 + 3)
6P6 = 11xP5 − 5P4 =
8 8
693 6 770 4 175 4 165 2 150 2 15 693 6 945 4 315 2 15
= x − x − x + x + x − = x − x + x −
8 8 8 8 8 8 8 8 8 8
1
⇒ P6 (x) = (231x6 − 315x4 + 105x2 − 5). (25)
16

12.5.5 Differentiate equation (20), being careful to use the product rule with xPl−1 (x):

lPl0 = (2l − 1)Pl−1 + (2l − 1)xPl−1


0 0
− (l − 1)Pl−2 . (26)
Now use the recursion relation

xPl0 − Pl−1
0
= lPl (27)
0
to get an expression for Pl−2 . Do this by replacing l with l−1. Don’t forget to make the replacement
in the coefficient as well. Then

0 0 0 0
xPl−1 − Pl−2 = (l − 1)Pl−1 ⇒ Pl−2 = xPl−1 − (l − 1)Pl−1 . (28)
Now plug this into (26) and simplify:

lPl0 = (2l − 1)Pl−1 + (2l − 1)xPl−1


0 0
− (l − 1)(xPl−1 − (l − 1)Pl−1 )
0 2 0
= ((2l − 1)x − (l − 1)x)Pl−1 + ((l − 1) + (2l − 1))Pl−1 = lxPl−1 + l2 Pl−1 . (29)

Now divide through by l to get the result:

Pl0 = xPl−1
0
+ lPl−1 or Pl0 (x) − xPl−1
0
(x) = lPl−1 (x). (30)

5
12.5.7 Rewrite (30) with l → l + 1:

0
Pl+1 − xPl0 = (l + 1)Pl ⇒ 0
xPl0 = Pl+1 − (l + 1)Pl . (31)
Then plug this result into (27).

0 0
Pl+1 − (l + 1)Pl − Pl−1 = lPl . (32)
Rearranging terms, one finds

0 0
(2l + 1)Pl (x) = Pl+1 (x) − Pl−1 (x). (33)

12.5.9 Follow the hint: start with the highest power of x, taking a multiple of the Legendre
polynomial that will have highest power matching this one. We have 3x2 + x − 1. Since P2 (x) =
1
2
(3x2 − 1), take

2P2 (x) = 3x2 − 1. (34)


Then this matches the highest term, and incidentally, the last term as well. The only term left is
x, which is just P1 (x). Hence,

3x2 + x − 1 = 2P2 (x) + P1 (x). (35)

12.6.4 Let h(x) = f (x)g(x). If f (x) is even and g(x) is odd, then h(x) is odd:

h(−x) = f (−x)g(−x) = f (x)(−g(x)) = −f (x)g(x) = −h(x). (36)


Therefore, the result follows from the fact that the integral of an odd function over a symmetric
interval is zero. We always use this result, but here I suppose I should prove it, otherwise there is
no content to this problem at all. The proof goes as follows:

Z a Z 0 Z a
h(x)dx = h(x)dx + h(x)dx. (37)
−a −a 0

Now change variables in the first integral to t = −x ⇒ x = −t. Then dx = −dt and when x = −a,
t = a; x = 0 ⇒ t = 0. Then
Z 0 Z 0 Z a Z a Z a
h(x)dx = h(−t)(−dt) = h(−t)dt = − h(t)dt = − h(x)dx. (38)
−a a 0 0 0
In the third step I used the fact that h is odd, and in the forth step I just renamed the dummy
variable. Then
Z a Z 0 Z a Z a Z a
h(x)dx = h(x)dx + h(x)dx = − h(x)dx + h(x)dx = 0. (39)
−a −a 0 0 0

6
12.6.6 Note that Pl (x) is a sum of even powers of x if l is even, and sum of odd powers of x if l
is odd. It follows that

even, l even
Pl (x) is . (40)
odd, l odd
Now, if we take a derivative of an even power of x, it becomes and odd power, and vice versa.
Hence,

0 odd, l even
Pl (x) is . (41)
even, l odd
Thus, Pl (x)Pl0 (x) is always the product of an even and odd function for any l. Therefore, using the
result of the previous problem, with a = 1,
Z 1
Pl (x)Pl0 (x)dx = 0. (42)
−1

The other way to show this is by problem (12.4.4). Pl (x) is a polynomial of degree l. Therefore,
Pl0 (x) is a polynomial of degree l − 1. Hence, we may generally write
l−1
X
Pl0 (x) = am x m , (43)
m=0

where the am are some coefficients that we don’t need to determine. Now just use linearity of the
integral, and the result of problem (12.4.4):
Z 1 Z 1 l−1
X  l−1
X Z 1
Pl (x)Pl0 (x)dx = Pl (x) m
am x dx = am Pl (x)xm = 0. (44)
−1 −1 m=0 m=0 −1

Every term in the sum vanishes because we always have m < l.

12.6.8 We have the set of functions {cos ((n + 1/2)x)|n = 0, 1, . . .}. To show that they are
orthogonal on (0, π), pick any two m, n and compute
Z π
cos ((m + 1/2)x) cos ((n + 1/2)x)dx. (45)
0
We want to show that the result is zero if m 6= n. Use the trig identity
1
cos α cos β = (cos (α + β) + cos (α − β)). (46)
2
Then

7
π
1 π
Z Z
cos ((m + 1/2)x) cos ((n + 1/2)x)dx = (cos ((m + n + 1)x) + cos ((m − n)x))dx
0 2 0

 π  π, n=m 
1 sin ((m + n + 1)x) π π
= + sin ((m−n)x)
= δmn . (47)
2 m+n+1 , n 6= m 2

0
 m−n
0

Everything vanishes except when m = n. In this case we get π/2.


Now we expand the function f (x) = 1 in terms of these functions. Write

X
f (x) = 1 = an cos ((n + 1/2)x). (48)
n=0

The goal is to find the coefficients an . Using the orthogonality of the cosine functions, multiply
both sides by cos ((m + 1/2)x) and integrate:

Z π Z π ∞
X 
1 · cos ((m + 1/2)x) = cos ((m + 1/2)x) an cos ((n + 1/2)x) dx
0 0 n=0
∞ Z π ∞
X X π π
= an cos ((m + 1/2)x) cos ((n + 1/2)x)dx = an δmn = am
n=0 0 n=0
2 2
Z π
2
⇒ am = cos ((m + 1/2)x)dx. (49)
π 0

Hence we find
Z π
π
2 2 sin ((n + 1/2)x) 2 sin (nπ + π/2)
an = cos ((n + 1/2)x)dx = · = · . (50)
π 0 π n + 1/2
0 π n + 1/2
When n is even we have sin (π/2) = sin (5π/2) = · · · = 1. For odd n we get −1. Hence,

2 (−1)n 4 (−1)n
an = · = · . (51)
π n + 1/2 π 2n + 1
Thus

4 X (−1)n
f (x) = 1 = cos ((n + 1/2)x). (52)
π n=0 2n + 1
This example might lead one to guess that these cosines are a complete set on (0, π), though this
is certainly no proof. In fact, this set is a complete set, as Boas comments at the end of chapter 7,
section 11. Problem 7.11.11 outlines a proof, which ultimately relies on Dirichlet’s theorem.

8
12.7.3 Pm (x) is a polynomial of degree m. Therefore, it may generally be written as
m
X
Pm (x) = ak xk (53)
k=0

for some coefficients ak . Then


Z 1 m
X Z 1
Pm (x)Pl (x)dx = ak xk Pl (x) = 0, (54)
−1 k=0 −1

since k ≤ m and assuming m < l, as follows from problem (12.4.4) and linearity of the integral.

12.7.5 Note that P0 (x) = 1. Hence, again by problem (12.4.4), if l > 0,


Z 1 Z 1
Pl (x)dx = Pl (x)P0 (x)dx = 0. (55)
−1 −1

12.7.6 Note that the square of an odd function is even, and the square of an even function is
even. Therefore Pl2 is even for any l. On the other hand P1 (x) = x is odd. Therefore, from problem
(12.6.6),
Z 1
(Pl (x))2 P1 (x)dx = 0. (56)
−1

12.8.6 Multiplying each side of the recursion relation (33) by Pl gives

0 0
(2l + 1)Pl2 = Pl Pl+1 − Pl Pl−1 . (57)
Now integrate both sides over (−1, 1).

Z 1 Z 1 Z 1 Z 1
2 0 0 0
(2l + 1) (Pl (x)) dx = Pl (x)Pl+1 (x)dx − Pl (x)Pl−1 (x)dx = Pl (x)Pl+1 (x)dx. (58)
−1 −1 −1 −1

0
The last term vanishes because Pl−1 is a polynomial of degree less than l. It then follows from
problem (12.4.4) that integral vanishes. For the middle term integrate by parts. With the remaining
integral, again use (12.4.4) to conclude it vanishes:
Z 1 1 Z 1 1
0 0

Pl (x)Pl+1 (x)dx = Pl (x)Pl+1 (x) −
Pl (x)Pl+1 (x)dx = Pl (x)Pl+1 (x) . (59)
−1 −1 −1 −1

Now recall the result

Pm (1) = 1, Pm (−1) = (−1)m . (60)

9
Then

Z 1
1

(2l + 1) (Pl (x)) dx = Pl (x)Pl+1 (x) = (1)(1) − (−1)l (−1)l+1 = 1 − (−1)2l+1 = 2.
2
(61)
−1 −1

This follows because 2l + 1 is odd for any l. Thus we obtain the result
Z 1
2
(Pl (x))2 dx = . (62)
−1 2l + 1

12.9.2 We wish to expand f (x) as a sum of Legendre polynomials. Write



X
f (x) = cl Pl (x) (63)
l=0

and use the orthogonality to solve for the cl :

Z 1 ∞ Z 1 ∞
X X 2 2
f (x)Pm (x) = cl Pl (x)Pm (x)dx = cl δlm = cm
−1 l=0 −1 l=0
2l + 1 2m + 1
Z 1
2m + 1
⇒ cm = f (x)Pm (x)dx. (64)
2 −1

f (x) is given by

0, −1 ≤ x ≤ 0
f (x) = . (65)
x, 0 < x < 1
I don’t know of any way to evaluate the integral for general m, so we will just compute the first
few terms, as is done in the examples.

1
1 1 1 1
Z Z Z
1 1
c0 = f (x)P0 (x)dx = 0 + xPO (x)dx = dx = ; (66)
2 −1 2 0 2 0 4
Z 1 Z 1
3 3 1
c1 = xP1 (x)dx = x2 dx = ; (67)
2 0 2 0 2
Z 1
5 x 5 3 1 5
c2 = (3x2 − 1)dx = ( − )= ; (68)
2 0 2 2 2·4 2·2 16
Z 1
7 x 7 5 3
c3 = (5x3 − 3x)dx = ( − ) = 0; (69)
2 0 2 2 2·5 2·3
1  
9 −1
Z
9 x 4 2 9 35 15 3 3
c4 = (35x − 30x + 3)dx = ( − + )= =− . (70)
2 0 8 2 48 16 16 2 48 32

10
Hence,
1 1 5 3
f (x) = P0 (x) + P1 (x) + P2 (x) − P4 (x) + · · · . (71)
4 2 16 32

12.9.10 Do the same calculations as the previous problem but for f (x) = 3x2 + x − 1:

1
1 1 1 3 x2
Z
2
1
c0 = (3x + x − 1)dx = (x + − x) = (2 + 0 − 2) = 0; (72)
2 −1 2 2 −1 2
Z 1 1
3 3 3 1 1 3 1
c1 = x(3x2 + x − 1)dx = ( x4 + x3 − x2 ) = · · 2 = 1; (73)
2 −1 2 4 3 2 −1 2 3
Z 1 Z 1
5 1 5
c2 = (3x2 + x − 1) (3x2 − 1)dx = (9x4 + 3x3 − 6x2 − x + 1)dx
2 −1 2 4 −1
1
5 9 5 3 4 3 1 2
= ( x + x − 2x − x + x)
4 5 4 2 −1
5 18 18 5 9 − 10 + 5
= ( − 4 + 2) = −5+ = = 2. (74)
4 5 4 2 2
Now,
Z 1
2l + 1
cl = (3x2 + x − 1)Pl (x)dx = 0, for l ≥ 3 (75)
2 −1

follows from problem (12.4.4). Hence, the series terminates and we find

f (x) = 3x2 + x − 1 = P1 (x) + 2P2 (x) (76)


which is the result we had before.

12.11.2 The equation is x2 y 00 + xy 0 − 9y = 0. Try a solution of the form



X
y(x) = an xs+n , (77)
n=0

where we assume a0 6= 0. Then

y = z0 xs + a1 xs+1 + a2 xs+2 + · · · , (78)


y 0 = a0 sxs−1 + a1 (s + 1)xs + a2 (s + 2)xs+1 + · · · , (79)
y 00 = a0 s(s − 1)xs−2 + a1 (s + 1)sxs−1 + a2 (s + 2)(s + 1)xs + · · · , (80)

so

11
xy 0 = a0 sxs + a1 (s + 1)x( s + 1) + a2 (s + 2)xs+2 + · · · , (81)
x2 y 00 = a0 s(s − 1)xs + a1 (s + 1)sxs+1 + a2 (s + 2)(s + 1)xs+2 + · · · . (82)

Grouping terms, the differential equation becomes

x2 y 00 + xy 0 − 9y = (s(s − 1) + s − 9)a0 xs + ((s + 1)s + (s + 1) − 9)a1 xs+1 + · · ·


+((s + n)(s + n − 1) + (s + n) − 9)an xs+n + · · · = 0 (83)

Hence we must have

((s + n)(s + n − 1) + (s + n) − 9)an = ((s + n)(s + n − 1 + 1) − 9)an = ((s + n)2 − 9)an


= (s + n + 3)(s + n − 3)an = 0, ∀n = 0, 1, 2, . . . . (84)

For n = 0, this is the indicial equation. We learn

(s + 3)(s − 3)a0 = 0 ⇒ s = ±3, (85)


since we assume a0 6= 0. Either s is a solution, so we must consider both separately. If we call
the resulting power series that we find from each y±3 (x), then the general solution to the equation
will be the sum of the two power series, y = y−3 + y3 . Let us first suppose s = 3. Then go to the
general term and plug this in:

an undetermined, n = 0
(3 + n + 3)(3 + n − 3)an = (n + 6)nan = 0 ⇒ . (86)
an = 0, n>0
Hence the series terminates after one term:

X
y3 = an x3+n = a0 x3 . (87)
n=0

Now we go back to the general term and solve for the coefficients when s = −3. These coefficients
will have nothing to do with the ones we just solved for, so I will relabel them ãn to avoid confusion.
Then


ãn undetermined, n = 0, 6
(−3 + n + 3)(−3 + n − 3)ãn = n(n − 6)ãn = 0 ⇒ . (88)
ãn = 0, n=6 0, 6
The solution for y−3 is

X
y−3 = ãn x−3+n = ã0 x−3 + ã6 x3 . (89)
n=0

12
The general solution is a sum of y±3 :

y(x) = a0 x3 + ã0 x−3 + ã6 x3 = Ax−3 + Bx3 , (90)


where I renamed arbitrary coefficients. Note that there are two linearly independent solutions,
x3 , x−3 , as there must be because this is a second order equation. Hence, there are really two
arbitrary coefficients; I can always call B = a0 + ã6 .

12.13.3 We wish to show that


r
πx
J−1/2 (x) = cos x. (91)
2
Begin with the general series expression for J−1/2 ; this is equation (13.1) of the text since p < 0.
We have

∞  2n−1/2 r X ∞  2n
X (−1)n x 2 (−1)n x
J−1/2 (x) = = .
n=0
Γ(n + 1)Γ(n − 1/2 + 1) 2 x n=0
Γ(n + 1)Γ(n − 1/2 + 1) 2
p (92)
−1/2
Now, when we multiply this by πx/2, it will remove the factor of x out front and what we
have left is a regular power series in x. This power series contains only even powers in x. It should
be clear now what must happen: this must be the Taylor series of cos x.
We need to get rid of the Gamma functions. Recall Γ(n + 1) = n!, and from a previous exercise,

(2n)! √
Γ(n − 1/2 + 1) = Γ(n + 1/2) = π. (93)
4n n!
Plugging these results in

r ∞  2n r ∞
2X (−1)n x 2 X (−1)n 4n n! x2n
= · 2n
x n=0 Γ(n + 1)Γ(n − 1/2 + 1) 2 πx n=0 n!(2n)! 2
r ∞
2 X (−1)n 2n
= ·x . (94)
πx n=0 (2n)!

In the last line, the 4n canceled against the 22n = (22 )n . As you can check, (formula (13.2) of
chapter one), or should remember, the infinite sum is exactly the Taylor series for cos x, which
converges to cos x for all x. Hence,

r r r ∞ ∞
πx πx 2 X (−1)n 2n X (−1)n 2n
J−1/2 (x) = ·x = · x = cos x. (95)
2 2 πx n=0 (2n)! n=0
(2n)!

13
12.15.1 O.k. so here’s the one I couldn’t do in office hours the other day; my excuse was sleep
deprivation. (And there is no typo in the book). The proof is not hard once you make one key
observation. I think the problem is that the book hint throws you off. The proof only starts out
like the example in the book for the first couple lines.
Begin by writing down the left hand side, using the power series for Jp (x) and taking the
derivative:

∞  2n+p 
(−1)n

d −p d −p
X x
[x Jp (x)] = x
dx dx n=0
Γ(n + 1)Γ(n + p + 1) 2
∞ ∞
(−1)n x2n (−1)n (2n) x2n−1
X 
d X
= · 2n+p = · 2n+p . (96)
dx n=0 Γ(n + 1)Γ(n + p + 1) 2 n=0
Γ(n + 1)Γ(n + p + 1) 2

This is where the analogy with the book proof ends. The key observation to make is that the first
term of this sum, corresponding to n = 0, vanishes because of the n in the numerator. Hence, the
sum really doesn’t start until n = 1:
∞ ∞
X (−1)n (2n) x2n−1 X (−1)n (2n) x2n−1
· = · . (97)
n=0
Γ(n + 1)Γ(n + p + 1) 22n+p n=1
Γ(n + 1)Γ(n + p + 1) 22n+p
Since we want sums that start at zero, (because we are trying to get a Bessel function on the right
hand side), let us change the summation index. Let m = n−1 ⇒ n = m+1. Make this replacement
everywhere you see an n. Then when n = 1, m = 0, so the sum starts from 0 and still goes to ∞.
Thus

∞ ∞
X (−1)n (2n) x2n−1 X (−1)m+1 · 2(m + 1) x2(m+1)−1
· 2n+p = · 2(m+1)+p . (98)
n=1
Γ(n + 1)Γ(n + p + 1) 2 m=0
Γ(m + 2)Γ(m + p + 2) 2

We want Γ(m + 1) in the denominator, so use Γ(m + 2) = (m + 1)Γ(m + 1). The factor of (m + 1)
will cancel with the numerator. Bring the extra factor of 2 in the numerator into the denominator
by subtracting one from the power. Then multiply and divide by xp . Thus we obtain

∞ ∞
X (−1)m+1 · 2 x2m+1 −p
X (−1)m+1 x2m+p+1
= x
m=0
Γ(m + 1)Γ(m + p + 2) 22m+p+2 m=0
Γ(m + 1)Γ(m + p + 2) 22m+p+1
∞  2m+(p+1)
−p
X (−1)m x
= −x . (99)
m=0
Γ(m + 1)Γ(m + 1 + (p + 1)) 2

Now, the last sum is exactly Jp+1 (x). Hence,


d −p
[x Jp (x)] = −x−p Jp+1 (x). (100)
dx

14

You might also like