MATH 551 LECTURE NOTES

MULTI-DIMENSIONAL PDES: SPECIAL FUNCTIONS

BESSEL FUNCTIONS ETC.

Topics covered
• References for looking up properties/numbers
• Useful properties of Bessel functions
◦ First and second kind; behavior for small |x|
◦ Bounded vs. unbounded solutions; zeros
◦ Modified bessel functions (the ‘negative eigenvalue’ case)
• Legendre functions
• Reminder: Cauchy-Euler equations
• A condensed ‘formula sheet’ of properties
1. Special functions
When the eigenvalue ODE cannot be solved directly, we can appeal to theory to define
solutions, then use other methods to derive relevant information.1 Such functions arise as
solutions to eigenvalue problems for PDEs like the heat equation in non-rectangular ge-
ometries (cylinder, sphere and so on).

There are several standard references for these ‘special functions’. Be very careful when
using these properties: make sure you are using the standard version of the function.
• The standard (trusted) book is Abramowitz & Stegun, Handbook of mathematical
functions, available online as a pdf. The information here is correct.
• The standard on-line database is at https://dlmf.nist.gov/.
• Most (good) computing software has the standard special functions built in; for
instance, besselj(ord,x) in Matlab for Jν (x). Zero-finders for such functions are
easy to find (e.g. on the Matlab file exchange)
• Wolfram has BesselJ[ord,x] and BesselJZero[ord,n] for Jν and its zeros (etc.)
1.1. Scaling. Often, it is necessary to rescale or change variables to convert the ODE
into ‘standard form’, i.e. one of the standard already-solved problems with known solutions.

It is important to recognize which parameters affect the structure of the solution and which
can be ‘scaled out’. For example when solving the ODE
−y 00 = m2 y
we can set ξ = mx (rescale) to get a ‘standard’ form
−yξξ = ξ =⇒ y = c1 cos(ξ) + c2 sin(ξ) = c1 cos(mx) + c2 sin(mx)
i.e. it suffices to derive properties of the equation (and the solution) for m = 1.
1Most of the tools come from asymptotics and perturbation methods; for the standard introduction,
see Bender & Orszag’s asymptotic and perturbation methods for scientists and engineers.
1
2 INTRO TO PDES

1.2. Bessel functions. The most notable example (appearing e.g. in cylindrical coordinates
for the r-direction) is Bessel’s equation
ν2
 
00 1 0
y + y + 1 − 2 y = 0. (1.1)
x x
The number ν ≥ 0 is usually called the order. Since it is a second-order linear ODE, the
general solution has the form
y(x) = c1 y1 (x) + c2 y2 (x)
for a pair of linarly independent solutions y1 , y2 . There is a standard choice for this basis
(with nice properties). We write the solution as
y(x) = c1 Jν (x) + c2 Yν (x)
where Jν and Yν are ‘Bessel functions’ of the ‘first’ and ‘second’ kind. It is true that
Yν = J−ν unless 2ν is an integer
so for the most part, the two basis functions are Jν and J−ν . The rule here is that J−ν must
be ‘adjusted’ in the special case where 2ν is an integer (in particular, for ν = 0).
1.3. Bounded vs. unbounded. The behavior for small |x| is as follows. If ν > 0 then
Jν (x) ∼ Cxν ,
Yν (x) ∼ Cx−ν as |x| → 0,
and for zero-th order,
J0 (x) ∼ 1, Y0 (x) ∼ C log |x| as |x| → 0.
This means that if ν ≥ 0 then Jν is bounded (as x → 0) and Yν is singuar at x = 0, so if
we need to apply a boundedness constraint then Jν should be kept.
1
1
0.5

0.5 0

-0.5
0
-1

-0.5 -1.5
0 5 10 0 5 10

1.4. Equivalent ODEs. The eigenvalue probem

ν2
 
00 1 0
y + y + λ− 2 y =0
x x
√
for λ > 0 is equivalent to Bessel’s equation by rescaling ξ = x λ. When λ < 0 it is instead
equivalent to the modified Bessel equation (see next page).
 INTRO TO PDES 3

The spherical Bessel equation

 
1 2 0 0 ν(ν + 1)
(x y ) + 1 − y = 0. (1.2)
x2 x2
is equivalent to Bessel’s equation of order ν + 1/2 under the transform w = x1/2 y, i.e.
(ν + 1/2)2
 
1/2 00 1 0
w(x) = x y(x) =⇒ w + w + 1 − w = 0.
x x2
Thus the ‘spherical Bessel functions’ solving (1.2) are
Jν+1/2 Yν+1/2
√ , √
x x

1.5. Oscillation and zeros. The Bessel function Jν (x) (also Yν ) has the property that
Jν (x) has a sequence of positive zeros 0 < γν,1 < γν,2 < · · · (1.3)
√
due the decaying oscillations of the function for x > 0 (see graph). Equations like Jν (a λ) =
0 can be solved in terms of these zeros (in this case, (γn /a)2 for n = 1, 2, · · · ).

1.6. Modified Bessel functions. Note that Bessel’s equation requires that the coefficient
on the 1/x2 term is positive (ν 2 ). When solving for negative eigenvalues, one sometimes
runs into the opposite case.
The modified Bessel equation has the form
ν2
 
00 1 0
y + y − 1+ 2 y =0 (1.4)
x x
and the two modified Bessel functions of the first and second kind are Iν and Kν , with
y = c1 Iν (x) + c2 Kν (x).
Unlike the Bessel functions, the modified versions do not oscillate (see plot below):
Iν (x), Kν (x) > 0 for x > 0.
Precisely, Iν is increasing (and Iν (0) = 0); Kν (x) is decreasing (and Kν (x) → ∞ as x & 0).
4 4

3 3

2 2

1 1

0 0
0 1 2 3 4 0 1 2 3 4
4 INTRO TO PDES

1.7. Legendre ‘polynomials’. In spherical coordinates, we need to solve the problem

1 m2
(sin φΦ0 )0 + (λ − )Φ = 0.
sin φ sin2 φ
Define the transformation
ξ = cos φ, y(ξ) = Φ(φ).
Then
∂ ∂
= sin φ , sin2 φ = 1 − ξ 2
∂φ ∂ξ
we obtain the (generalized) Legendre equation
m2
((1 − x2 )y 0 )0 + (λ − )y = 0. (1.5)
1 − x2
For a given n, this eigenvalue problem has bounded solutions only for eigenvalues
λ = n(n + 1), n an integer ≥ m.
The eigenvalues/bounded eigenfunctions are the Legendre functions
λ = n(n + 1), y = Pmn (x), n ≥ m.
It is true that
m even =⇒ Pmn is a polynomial of deg. m
in which case the function is called an associated Legendre polynomial.

Further relations can be derived using Rodgrigues’ formula, which will not be detailed
here (see Haberman, Chapter 7).
 INTRO TO PDES 5

Properties of special functions

Bessel functions (first/second kind):
• Equation/solution (ν ≥ 0 is a constant: the ‘order’):
001 0 ν2
y + y + (1 − 2 )y = 0 =⇒ y = c1 Jν (x) + c2 Yν (x)
x x
• If 2ν ∈
/ Z then Yν = J−ν (in particular, if ν = 0)
• Jν is bounded and Yν is unbounded (at zero). Precisely, if ν > 0 then
Jν ∼ Cν |x|ν as x → 0,
Yν ∼ |x|−ν
and if ν = 0then Jν ∼ 1 and Yν ∼ C log |x|.
• Values at zero (J0 is ‘normalized’):
J0 (0) = 1, Jν (0) = 0 for ν > 0, Jν0 (0) = 0 for ν > 1, Yν (0) = ∞
• Both Jν and Yν oscillate for positive x; in particular
Jν (x) =⇒ positive zeros at γν,j with 0 < γν,1 < γν,2 < · · · → ∞.
Jν0 (x) and Yν0 (x) also have a sequence of positive zeros.

Modified Bessel functions: (Bessel, opposite sign)

1 ν2
y 00 + y 0 − (1 + 2 )y = 0
x x
=⇒ y = c1 Iν (x) + c2 Kν (x)
Both K and I are positive for x > 0 and
I0 (0) = 1, Iν (0) = 0 for ν > 0, K(0) = ∞.
Related to J and Y by imaginary arguments, e.g. In (x) = i−n Jn (ix).

Bessel eigenvalue problem (both signs for λ):

1 ν2 √ √
y 00 + y 0 + (λ − 2 )y = 0, λ ≥ 0 =⇒ y = c1 Jν (x λ) + c2 Yν (x λ)
x x
1 ν2 √ √
y 00 + y 0 + (λ − 2 )y = 0, λ < 0 =⇒ y = c1 Iν (x λ) + c2 Kν (x λ)
x x
• Can check by other means that λ ≥ 0 (Rayleigh quotient)
• x(a) = 0 or x0 (a) = 0 =⇒ only λ ≥ 0 has solutions
• bounded at x = 0 =⇒ no Yν , Kν term

Spherical Bessel’s equation

• Equivalent to Bessel with ‘plus 1/2 order’
√ √
1 2 0 0 ν(ν + 1) Jν+1/2 (x λ) Yν+1/2 (x λ)
(x y ) + (λ − )y = 0 =⇒ y = c1 √ + c2 √
x2 x2 x x
6 INTRO TO PDES

• Special case (order 1/2 exactly):

√ √
sin x λ cos x λ
ν = 0 =⇒ y = c1 + c2 .
x x
Legendre polynomials:
n2
((1 − x2 )y 0 )0 + (λ − )y = 0.
1 − x2
• Bounded solutions only for λ = m(m + 1) where m ≥ n is an integer.
• Bounded solution (‘Legendre function/polynomial’): y = Pnm (x). Top letter: ‘order’ m.
Bottom letter: eigenvalue index (λn = n(n + 1)).
• If m is even then Pnm is a polynomial of degree n.
• Given λn , total of n + 1 eigenfunctions Pn0 , · · · Pnn
m
• Given order m, sequence Pmm , Pm+1 m
, Pm+2 , · · · for eigenvalues λm , λm+1 , · · ·
Other useful facts
Cauchy-Euler equations: For the ODE
p q
y 00 + y 0 + 2 y 00 = 0
x x
r
with p, q, constants, guess a trial solution x :
xr is a solution ⇐⇒ 0 = p(r) = r(r − 1) + pr + q.
Three cases for the solution:
r1 6= r2 , real =⇒ y = c1 |x|r1 + c2 |x|r2
r1 = r2 =⇒ y = c1 |x|r1 + c2 |x|r1 log |x|
r = a ± ωi =⇒ y = c1 |x|a cos(ω log |x|) + c2 |x|a sin(ω log |x|).
Indicial polynomial: For the ODE
p(x) 0 q(x) 00
y 00 +y + 2 y =0 (1.6)
x x
where p, q are continuous, define the indicial polynomial
f (r) = r(r − 1) + p(0)r + q(0).
Its roots determine the behavior of the solutions as x → 0:
r1 6= r2 , real =⇒ y1 ∼ |x|r1 , y2 ∼ |x|r2
r1 = r2 =⇒ y1 ∼ |x|r1 , y2 ∼ |x|r1 log |x|
r = a ± ωi =⇒ y1 ∼ |x|a cos(log |x|), y2 ∼ |x|a sin(log |x|)
In short: near x = 0, (1.6) behaves ‘like a Cauchy-Euler equation’ with p(x), q(x) replaced
by p(0) and q(0). For the general theory (messy), see Frobenius’ method.