1 Last Time
• Bessel functions are oscillatory and damped
∞
X (−1)k x 2k±µ
J±µ (x) = . (1)
k! Γ(k + 1 ± µ) 2
k=0
2 Today
• Membrane close-out
• Bessel function properties
3 Membrane wrap-up
The general solution is:
∞ X
∞
X r
u(r, θ, t) = Jm αmn [Amn cos mθ + Bmn sin mθ] cos ωmn t (2a)
m=0 n=1
a
∞ X
∞
X r
+ Jm αmn [Cmn cos mθ + Dmn sin mθ] sin ωmn t , (2b)
m=0 n=1
a
1
4 Orthogonality relations
We earlier showed that Bessel’s equation is of the Sturm-Liouville form with weight function
w(r) = r and eigenvalues k 2 , so with appropriate BCs the solutions (Bessel functions) are
orthogonal. In this case we have the Dirichlet condition at one boundary (the cylinder
surface), Jm (a) = 0, but it is not necessarily true that Jm (0) = 0. However, at the other
‘boundary’, the point at the origin r = 0, the function p(r) = r that appears inside the
derivative vanishes, which is sufficient to ensure orthogonality.
Let’s do the orthogonality derivation again for the specific case of Bessel functions. Consider
m2
d dJm (kr)
r + k2 r − Jm (kr) = 0 (4a)
dr dr r
m2
d dJm (qr)
r + q2 r − Jm (qr) = 0 (4b)
dr dr r
Jm (ka) = Jm (qa) = 0 . (4c)
Then
ˆ a
d dJm d dJm
dr Jm (qr) r (kr) − Jm (kr) r (qr) (5a)
0 dr dr dr dr
ˆ a
= (q 2 − k 2 ) dr rJm (kr)Jm (qr) . (5b)
0
αm`
and k = kmn = αmn 2 2
a ; q = km` = a , ` 6= n, so q − k 6= 0.
So for our model problem, we have the orthogonality relation
ˆa
dr rJm (kmn r)Jm (kmj r) ∝ δnj . (7)
0
More generally,
ˆa
dr r Jm (kmn r)J` (k`j r) = Nmn δnj δm` , (8)
0
where Nmn is the normalization integral that we’ll need to calculate. The integral can be
evaluated analytically, using recursion relations that we’ll derive in a bit. But let’s pause and
first solve our model problem, with Nmn to be determined afterwards.
2
4.1 Back to membrane
So based on the Sturm-Liouville theorem, to invert,
´ 2π multiply both sidesP by a trig function
of one particular mode m0 and integrate π1 0 dθ, which collapses the m . And multiply
both sides by the m0 th-order Bessel function J´m0 (km0 j r) with km0 j corresponding to one
a
particular zero of the function, and integrate N 1 0 0 dr r, which collapses the n .
P
m j
ˆ a ˆ 2π
1 1
Am0 j = dr rJm0 (km0 j r) dθ u(r, θ, 0) cos m0 θ , (9)
Nm0 j 0 π 0
and similarly for Bm0 j , Cm0 j , Dm0 j , where we used orthogonality properties of Bessel functions
and trigonometric functions.
If instead we had free boundaries, ∂u ∂r r=a = 0, then
We call this a generating function because coefficients for its power series in t ‘generate’ the
Bessel functions.
To see this, let ` − j = m, i.e.
`=j+m . (12)
Note that m runs from −∞ to ∞, but for m < 0, we must have j ≥ −m since ` is non-negative.
Thus,
∞
∞ X
X (−1)j x 2j+m
G= tm (13a)
m=0 j=0
(j + m)!j! 2
−1 ∞
X (−1)j
X x 2j+m
+ tm . (13b)
m=−∞ j=−m
(j + m)!j! 2
In the sum over negative m, write j = n − m and make use of the relation (from last lecture’s
3
notes) between negative- and positive-integer Bessel functions:
∞ −1 ∞
X X X (−1)n x 2n−m
G= tm Jm (x) + tm (−1)−m (14a)
m=0 m=−∞ n=0
n!(n − m)! 2
∞
X −1
X
= tm Jm (x) + tm (−1)m J−m (x) (14b)
m=0 m=−∞
∞
X
= tm Jm (x) , (14c)
m=−∞
4
Here’s another one valid for any order (integer or not):
∞
d µ d X 2µ x 2k+2µ
[x Jµ (x)] = (−1)k (17a)
dx dx k! Γ(k + µ + 1) 2
k=0
∞
X 2µ (k + µ) x 2k+2µ−1
= (−1)k (17b)
k! Γ(k + µ + 1) 2
k=0
∞ x 2k+µ−1
X 1
= xµ (−1)k (17c)
k! Γ(k + µ) 2
k=0
d µ
[x Jµ (x)] = xµ Jµ−1 (x) . (17d)
dx
And one more that we’ll use to derive the normalization integral:
d Jµ (x) Jµ+1 (x)
µ
=− . (18)
dx x xµ
and make another copy of the ODE (following the same procedure to establish Sturm-Liouville
orthogonality in general), except that for now we’ll leave the separation constant k to be
arbitrary (i.e., not an eigenvalue of Jµ (ka) = 0 just yet)
µ2
d d
r Jµ (kr) + k 2 r − Jµ (kr) = 0 . (20)
dr dr r
Multiply (19) by Jµ (kr), subtract (20) multiplied by Jµ (kµn r), and integrate, to
arrive at
a
dJµ (kµn r) dJ
µ (kr)
r Jµ (kr) − Jµ
(k r)
µn
dr dr 0
ˆa
= (k 2 − kµn
2
) dr r Jµ (kµn r)Jµ (kr) . (21a)
0
On the LHS, only the 1st term survives, since Jµ (kµn a) = 0 for Dirichlet boundary conditions.
Now differentiate both sides with respect to k (the only variable at this point, since
we’ve integrated over r), and then set k = kµn . On the RHS, when the derivative acts on
the function Jµ (kr) inside the integral, setting k = kµn gives zero, owing to the factor in
5
parentheses, so
ˆa
d d
a Jµ (ka) Jµ (kµn r) = 2kµn dr rJµ2 (kµn r) . (22)
dk k=kµn dr r=a
0
8 Summary
• Membrane solutions for fixed BCs and for free BCs
9 Next Time
• Finish Bessel normalization integral
• Cylinder problems
• Modified Bessel functions