You are on page 1of 5

Why We Care about Bessel Functions

There really are good reasons why we need and use Bessel functions. The text’s
examples are fine, but they avoid the more common use, that of separable solutions
to partial differential equations. For starters, consider the two-dimensional Wave
Equation in polar coordinates,

1 ∂ 2ψ
∇2 ψ =
v 2 ∂t2
where ψ = ψ(ρ, θ, t) and v is a constant with dimensions of speed. The use of ρ as
the radius is a physics thing; r is reserved for three dimensions (see below and the
text’s Section 3.7). In polar coordinates, then,

1 ∂ 2ψ
 
2 1 ∂ ∂ψ
∇ ψ = ∇ · (∇ ψ) = ρ + 2 .
ρ ∂ρ ∂ρ ρ ∂θ 2

(If this is a surprise, you are invited to either rederive this expression or look it up.
A physics text might be of avail.) In three dimensions in cylindrical coordinates,
we would have ψ = ψ(ρ, θ, z, t) and

1 ∂ 2ψ ∂ 2ψ
 
2 1 ∂ ∂ψ
∇ ψ= ρ + 2 + .
ρ ∂ρ ∂ρ ρ ∂θ 2 ∂z 2
For now, let’s stick with the 2-d case.
What we do is look for solutions of the form

ψ(ρ, θ, t) = f (ρ) g(θ) eiωt. (A)

This represents a “good guess,” if not an inspired guess. The motivation is similar
to what you did in Chapter 2, looking for solutions of second-order homogeneous
equations that had a particular form. Here, what we are looking for, primarily, are
the possible values of the frequency ω, but in order to find ω we need to solve other
differential equations, not just algebra. For a complete solution, we would need
sums of solutions corresponding to different independent solutions to (A).
Substitution of the form given in (A) into the Wave Equation gives

f (ρ) d2 g ω2
  
1 d df
ρ g(θ) + 2 = − 2 f (ρ) g(θ).
ρ dρ dρ ρ dθ 2 v

Dividing by the product f (ρ) g(θ) gives

1 1 d2 g ω2
 
11 d df
ρ + = − .
f ρ dρ dρ g ρ2 dθ 2 v2

1
At this point, if you are concerned about a possible division by zero, you are entitled
to be concerned. However, it turns out not to matter. None of the texts used in
this class address the issue, and we won’t either.
What we have done is to “separate the variables” and this method is known as
“Separation of Variables.” We recogize that the right hand side is a constant, inde-
 
pendent of θ, as is the first term on the left. Therefore the term 1/g ρ2 d2 g/dθ 2

must be independent of θ, and so (1/g) d2 g/dθ 2 must be a function of ρ. But
by our assumption in (A), the second derivative is a function of θ only. So, if

(1/g) d2 g/dθ 2 is a funciton of neither ρ nor θ, it must be a constant. What we
will do is to skip some steps (they’re not to hard to fill in), using the fact that g(θ)
must be periodic with period 2π to state that

d2 g
g(θ) = cos(nθ + φ), = −n2 g(θ),
dθ 2
where a multiplicative constant has been ingnored in the first equation above, and
the fact that the phase φ will in general depend on n will not be addressed further.
The ODE for f (ρ) now is seen to be

n2 ω2
 
1 d df
ρ − f + f = 0.
ρ dρ dρ ρ2 v2

The change of variables x = ρ/L, α = ωL/v and y(x) = f (ρ), for L some constant
distance, gives
1 d n2
(xy ) − 2 y + α2 y = 0.

(B)
x dx x
The text’s Eq. 28 on Page 248 may be written in a form equivalent to (B), as

n2
 
1 d 1 ′ 2
(xy ′ ) = (xy ′ ) = − α y. (C)
x dx x x2

Suppose that we wish to consider this as an eigenvalue equation, with the boundary
conditions y(0) finite, y(L) = 0. The rest of the section in the text describes
how such problems are solved; what I would like to do is show how the above
form (Equation (C)) can be exploited. Specifically, suppose we have two solutions,
corresponding to two distinct values of α, so that
 2 
1 ′ ′ n 2
(xy1 ) = − α 1 y1
x x2
 2 
1 ′ ′ n 2
(xy2 ) = − α 2 y2
x x2

2
where y1 (0) and y2 (0) are finite and y1 (L) = y2 (L) = 0.
Now for the razzle-dazzle; consider
Z L L Z L

(xy1′ ) y2 dx = xy1′ y2 − y2′ (xy1′ ) dx,
0 0 0

where an integration by parts has been used. The vanishing of y2 (L) leads to
Z L Z L

(xy1′ ) y2 dx = − y2′ (xy1′ ) dx.
0 0

Similarly,
Z L Z L

(xy2′ ) y1 dx = − y1′ (xy2′ ) dx,
0 0

with the result that Z L h i


′ ′
(xy1′ ) y2 − y1 (xy2′ ) dx = 0.
0

Now, from the above differential equations for y1 and y2 ,

n2 n2
   
′ 2 ′ 2
(xy1′ ) = − α1 xy1 , (xy2′ ) = − α2 xy2 ,
x2 x2

and substitution into the integral with the square brackets gives

L
n2
  2  
n
Z
2 2
0= x − α1 − − α2 y1 y2 dx
0 x2 x2
 L
Z
2 2
= α2 − α1 x y1 y2 dx.
0

Thus, we see that if α1 6= α2 ,


Z L
x y1 y2 dx = 0.
0

This is often phrased as “eigenfunctions corresponding to distinct eigenvalues are


orthogonal.” The kicker is that orthogonality is with respect to the above inner
product (yes, it is analogous to a dot product), which came from the form used in
(B) and (C) above, which came from the cylindrical geometry (keep this in mind a
while).
A subtle point (well, maybe not so subtle) in the above is that n had to be the
same for both y1 and y2 . If n1 6= n2 , are y1 and y2 still orthogonal? Well, no, but

3
remember that our motivation for the form of Bessel’s equation was that we sought
something of the form
ψ(ρ, θ, t) = f (ρ) g(θ) eiωt,

with
g(θ) = cos(nθ + φ),

with n an integer. If two possible ψs of the above form with different |n| are
multiplied and integrated over θ, the result is zero (try it). Actually, this is a simpler
example of eigenfunctions (cos(nθ + φ)) corresponding to distinct eigenvalues (the
ns) being orthogonal. We can combine the results for f (or y) and g by noting that
if, with f1 (ρ) = y1 (ρ/L), f2 (ρ) = y2 (ρ/L), and integrating over a disc of radius L,

η1 (ρ, θ) = f1 (ρ) g1 (θ) = f1 (ρ) cos(n1 θ + φ1 ), η2 = f2 (ρ) cos(n2 θ + φ2 ),


Z Z 2πZ L
η1 η2 dA = η1 η2 ρ dρ dθ = 0
0 0

if |n1 | =
6 |n2 |. Thus, the geometrical factor (the factor of x or ρ) is reproduced in
the orthogonality relation.
Well, you may say, that’s cute, but what else can we do with the formulation
represented by (C)? If we’re in spherical coordinates, and we seek a spherically
symmetric solution to the scalar wave equation (an example would be a sound wave
from a monaural speaker modeled as a point source), the equation describing the
wave for a given frequency would be

1 2 ′ ′ ω 2
r f + 2 f = 0. (D)
r2 c
(Note that this is not an eigenvalue problem, yet; there are no boundaries, and any
frequency is possible.) This can be expanded into

f′ ω2
f ′′ + 2 + 2 f = 0.
r c
Making the change of variable x = ωr c and letting y(x) = f (r(x)), the equation for
y in terms of the dimensionless variable x is

y′
y ′′ + 2 + y = 0.
x
At this point, we can use the transformation indicated on Pages 250-251. In Equa-
tion (3), we have (after multiplying by x2 ), A = 2, B = 0, C = 1, q = 2, so that

4
α = − 12 , β = 1, k = 1 and p = 12 . Our general solution, as given in Equation (6),
the highlighted equation on Page 251, is
 
y = x−1/2 c1 J1/2 (x) + c2 J−1/2 (x)
 
−1/2 sin x cos x
=x c1 √ + c2 √ (E)
x x
sin x cos x
= c1 + c2 ,
x x
a very convenient result.
With such an amazing result, you might well suspect that there is a simpler
way. There is, but it’s far from obvious (even though all physicists know it or
pretend to know it). The thing is to see that

d2 d ′ ′ ′′ 1 d 2 ′
(rf ) = (rf + f ) = 2f + rf = r f ,
dr 2 dr r dr

and so the original expression for f becomes

1 d2 ω2
(rf ) + f = 0, or
r dr 2 c2
ω2
h′′ + 2 h = 0,
c
where h = rf . Thus, h satisfies a fairly simple differential equation.
To show the consistency with the geometric interpretation, consider solutions
of the form in (E) corresponding to different frequencies, with f finite at the origin
and f = 0 at r = a (which is y ωa

c
= 0; now this is an eigenvalue problem). Then,
we have (apart from the multiplicative constants)

sin (n1 πr/a) sin (n2 πr/a)


y1 = , y2 =
r r

for integer n1 , n2 and it can be shown (give it a try) that


Z a
r 2 y1 y2 dr = 0
0

6 |n2 |. The factor of r 2 is again the same factor in r12 (r 2 f ′ )′ in (D), and
if |n1 | =
R
indeed is part of the volume integral if we consider y1 y2 dV (although the angular
part is not so simple as in cylindrical coordinates).

You might also like