Notes 7

Series Solutions

The method of serious solutions can mostly be illustrated by the very basic differential
dy
equation − y = 0, which has general solution y = Cex .
dx
Suppose we want to look for a solution in the form of a power series:

y(x) = a0 + a1 x + a2 x2 + a3 x3 + · · · + an xn + · · ·

By putting y(x) into the equation y ′ − y = 0, we have

(a1 +2a2 x+3a3 x2 +4a4 x3 +· · ·+(n+1)an+1 xn +· · · )−(a0 +a1 x+a2 x2 +a3 x3 +· · ·+an xn +· · · ) = 0

(a1 − a0 ) + (2a2 − a1 )x + (3a3 − a2 )x2 + (4a4 − a3 )x3 + · · · + ((n + 1)an+1 − an )xn + · · · = 0

In other words,
a1 a0 a2 a0 a3 a0 a0
a1 = a0 , a2 = = , a3 = = , a4 = = , · · · , an = .
2 2 3 3·2 4 4! n!
The series solution is therefore given by

y(x) = a0 (1 + x + x2 /2! + x3 /3! + · · · + xn /n! + · · · ) = a0 ex

as expected.
In the rest of these notes, we shall focus mainly on homogeneous 2nd order linear
ODEs with (simple) polynomial coefficient functions for the illustration of the series
method. Recall that nonhomogeneous linear equations can be solved by the method of
variation of parameters, once two fundamental solutions of the homogeneous equation
with non-vanishing Wronskian are found.

7.1 Homegeneous 2nd Order Linear ODEs with Power Se-

ries Coefficient Functions
Let’s consider the basic case of a 2nd order homogeneous linear ODE

d2 y dy
2
+ p(x) + q(x)y = 0
dx dx

37
 38

n n
P∞ P∞
where p(x) = n=0 pn (x − a) and q(x) = k=0 qn (x − a) are power series which
converge in an open interval centered at a given by |x − a| < R for some R > 0.
The we have the following fundamental result:
Theorem The initial value problem y ′′ + p(y)y ′ + q(x)y = 0, y(a) = a0 , y ′ (a) = a1 , has
a unique power series solution y(x) = ∞ n
P
n=0 an (x − a) in the interval |x − a| < R.

Proof For simplicity sake, let a = 0. (Or one may make a substitution: t = x − a.)
Then let p(x) = ∞ n n n
P P∞ P∞
n=0 pn x , q(x) = k=0 qn x , and y(x) = n=0 an (x − a) . We have
∞
X ∞
X
n−1
′
y (x) = nan x = (n + 1)an+1 xn ,
n=1 n=0
∞
X ∞
X
y ′′ (x) = n(n − 1)an xn−2 = (n + 2)(n + 1)an+2 xn
n=2 n=0
The differential equation can then be turned into a system of algebraic equations:
∞
X ∞
X ∞
X ∞
X ∞
X
(n + 2)(n + 1)an+2 xn + p n xn · (n + 1)an+1 xn + q n xn · an xn = 0
n=0 n=0 n=0 k=0 n=0
n n
∞
" #
X X X
(n + 2)(n + 1)an+2 + (k + 1)ak+1 pn−k + ak qn−k xn = 0
n=0 k=0 k=0
By taking n = 0, 1, 2, . . ., we have



 2a2 = −(a1 p0 + a0 q0 )
3 · 2a3 = −(a1 p1 + 2a2 p0 + a0 q1 + a1 q0 )





 4 · 3a4 = −(a1 p2 + 2a2 p1 + 3a3 p0 + a0 q2 + a1 q1 + a2 q0 )

..

 .


 Xn

 (n + 2)(n + 1)an+2 = − [(k + 1)ak+1 pn−k + ak qn−k ]



k=0

Consequently, all coefficients of the power series of y(x) can be successively and uniquely
determined, starting from the given initial values a0 , a1 , and the coefficients of p(x) and
q(x). The only remaining hard work is to show that the power series obtained indeed
converges absolutely for any x in the interval |x| < R.
For simplicity sake, we consider here the simpler case y ′′ + q(x)y = 0. Some additional
technical details for the general case y ′′ + p(x)y ′ + q(x)y = 0 can be found in a remark
following this special case.
As discussed above, the coefficients of the series solution y(x) with y(0) = a0 , y ′ (0) =
a1 are determined recursively by



 2a2 = −a0 q0
3 · 2a3 = −(a0 q1 + a1 q0 )





 4 · 3a4 = −(a0 q2 + a1 q1 + a2 q0 )

..

 .


 n
X




 (n + 2)(n + 1)an+2 = − ak qn−k
k=0
 39

∞
X
To show that the power series y(x) = an xn is convergent for all |x| < R, we only need
n=0
to show that it converges for all |x| < r for any 0 < r < R.
∞
X
To show the (absolute) convergence of y(x) = an xn , we could try to find a sequence
n=0
∞
X
bn such that |an | ≤ bn , where bn xn is absolutely convergent for any |x| < r. In fact,
n=0
this would show the absolute convergence of y(x) for any |x| < r:
∞
X ∞
X
|an ||x|n ≤ bn |x|n < ∞.
n=0 n=0
∞
X
Since the given power series q(x) = qn xn is convergent at x = r < R, we have
n=0

lim qn r n = 0 .
n→∞

Consequently, the sequence qn r n must be bounded, and hence we can find a constant M
such that |qn r n | ≤ M for all n ≥ 0; or equivalently |qn | ≤ rMn for all n ≥ 0.
Let b0 = |a0 |, b1 = |a1 |. Since the coefficients of y(x) are determined recursively by



 2a2 = −a0 q0
 3 · 2a3 = −(a0 q1 + a1 q0 )



4 · 3a4 = −(a0 q2 + a1 q1 + a2 q0 )

 Xn

 (n + 2)(n + 1)an+2 = − ak qn−k



k=0

we have

let
 2|a2 | ≤ b0 M ≤ b0 M + b1 rM = 2b2


let
3 · 2|a3 | ≤ b0 · M 1 2
r + b1 M ≤ r (b0 M + b1 rM + b2 r M ) = 3 · 2b3
 4 · 3|a | ≤ b M + b M + b M ≤ 1 (b M + b rM + b r 2 M + b r 3 M ) let


4 0 r2 1 r 2 r2 0 1 2 3 = 4 · 3b4
Note that
1 2b2 2 + r2M
3 · 2b3 = (b0 M + b1 rM ) + b2 rM = + b2 rM = b2 ·
r r r
1 3 · 2b3 3 · 2 + r2M
4 · 3b4 = 2
(b0 M + b1 rM + b2 r 2 M + b3 r 3 M ) = + b3 rM = b3 ·
r r r
In general, we have
1
(n + 2)(n + 1)|an+2 | ≤ (n + 2)(n + 1)bn+2 = (b0 M + b1 rM + b2 r 2 M + · · · + bn+1 r n+1 M )
rn
(n + 1)nbn+1 (n + 1)n + r 2 M
= + bn+1 rM = bn+1 ·
r r
Hence for any |x| < r, the ratio test implies the absolute convergence of ∞ n
P
n=0 bn x since

bn+1 |x|n+1 n(n − 1) + r 2 M |x|

lim n
= |x| lim = < 1.
n→∞ bn |x| n→∞ r(n + 1)n r
 40

∞
X
We can therefore conclude the absolute convergence of y(x) = an xn for any |x| < R.
n=0

Remark The general case with p(x) 6≡ 0 is basically the same. One only needs to
n
X
consider also the term (k+1)ak+1 pn−k , and define bn inductively as before by b0 = |a0 |,
k=0
b1 = |a1 |, and |an | ≤ bn for all n ≥ 0, where

2|a2 | ≤ b1 M + b0 M ≤ b1 M + b0 M + b1 rM = 2b2

M M
3 · 2|a3 | ≤ 2b2 M + b1 + b0 + b1 M + b2 rM = 3 · 2b3
r r
2b2
= 2b2 M + + b2 rM
r
M M M M
4 · 3|a4 | ≤ 3b3 M + 2b2 + b1 2 + b0 2 + b1 + b2 M + b3 rM = 4 · 3b4
r r r r
3 · 2b3
= 3b3 M + + b3 rM
r
n h
X M M i
(n + 2)(n + 1)bn+2 = (k + 1)bk+1 n−k + bk n−k + bn+1 rM,
r r
k=0
n−1
Xh M M i
= (n + 1)bn+1 M + (k + 1)bk+1 + bk + bn M + bn+1 rM
r n−k r n−k
k=0
1

= (n + 1)bn+1 M + (n + 1)nbn+1 − bn rM + bn M + bn+1 rM
r
(n + 1)rM + (n + 1)n + r 2 M
= bn+1 ·
r
n n
where |pn | ≤ M/r , |qn | ≤ M/r for all n ≥ 0. One can then proceed as before to show
the absolute convergence of ∞ n
P
n=0 bn x for all |x| < r < R.

Example 7.1.1 (Airy’s Equation) Find the solutions of y ′′ − xy = 0.

Let y(x) = ∞ n
P
n=0 an x . We have

∞
X ∞
X
y ′′ − y = (n + 2)(n + 1)an+2 xn − x an xn = 0
n=0 n=0

∞
X ∞
X
2a2 + (n + 3)(n + 2)an+3 xn+1 − an xn+1 = 0
n=0 n=0
∞ h
X i
2a2 + (n + 3)(n + 2)an+3 − an xn+1 = 0
n=0
1
i.e., a2 = 0, and an+3 = an for all n ≥ 0. In particular,
(n + 3)(n + 2)
a2 = 0 =⇒ 0 = a5 = a8 = · · · = a3n+2
1 1 a0
a3n = a3n−3 = a3n−6 = · · · =
3n(3n − 1) 3n(3n − 1)(3n − 3)(3n − 4) (2 · 3)(5 · 6) · · · (3n − 1)(3n)
 41

1 1 a1
a3n+1 = a3n−2 = a3n−5 = · · · =
(3n + 1)3n (3n + 1)3n(3n − 2)(3n − 3) (3 · 4)(6 · 7) · · · (3n)(3n + 1)
Hence
 1 3 1 1 
y(x) = a0 1 + x + x6 + · · · + x3n + · · ·
2·3 2·3·5·6 (2 · 3)(5 · 6) · · · (3n − 1)(3n)
 1 4 1 1 
+a1 x + x + x7 + · · · + x3n+1 + · · ·
3·4 3·4·6·7 (3 · 4)(6 · 7) · · · (3n)(3n + 1)
= a0 y1 (x) + a1 y2 (x)
Note that y1 (0) = 1, y1′ (0) = 0, y2 (0) = 0, y2′ (0) = 1, and W [y1 , y2 ](0) = 1.
It is also easy to see that both y1 (x) and y2 (x) have radius of convergence R = ∞.

Remark Consider the gamma function defined on the interval1 (0, ∞) by

Z ∞
Γ(s) = ts−1 e−t dt, s > 0,
0

which satisfies
Z ∞ ∞ Z ∞
Γ(s + 1) = ts e−t dt = −ts e−t + sts−1 e−t dt = sΓ(s)
0 0 0

i.e., s = Γ(s + 1)/Γ(s). We then have

Γ(s + n) Γ(s + n) Γ(s + n − 1) Γ(s + 1)

= · ··· = (s + n − 1)(s + n − 2) · · · (s) .
Γ(s) Γ(s + n − 1) Γ(s + n − 2) Γ(s)

In particular, we have

1 1 Γ( 23 )
= n 2 =
(2 · 3)(5 · 6) · · · (3n − 1)(3n) 9 n! · 3 · (1 + 23 ) · · · (n − 1 + 23 ) 9n n!Γ(n + 23 )

and

1 1 Γ( 43 )
= n =
(3 · 4)(6 · 7) · · · (3n)(3n + 1) 9 n! · 43 (1 + 43 ) · · · (n − 1 + 43 ) 9n n!Γ(n + 43 )

The general solution of the Airy equation found above can thus be written as

Γ( 23 )z 3n Γ( 43 )z 3n+1
∞
X X∞
y = a0 y1 (x) + a1 y2 (x) = a0 + a1
n=0
9n !Γ(n + 23 ) n=0
9n !Γ(n + 43 )

The Airy function of the first kind Ai(x), and second kind Bi(x) respectively, are then
given by
∞ ∞
2 X z 3n − 43
X z 3n+1
Ai(x) = 3− 3 2
n n!Γ(n + )
− 3 n n!Γ(n + 4 )
= Ai(0)y1 (x) + Ai′ (0)y2 (x)
n=0
9 3 n=0
9 3
1
In complex analysis, this gamma function can be “analytically continued” to a “meromorphic func-
tion” in the whole complex plane with only “simple poles” at the non-positive integers.
 42

∞ ∞
1 X z 3n − 56
X z 3n+1
Bi(x) = 3− 6 n n!Γ(n + 2 )
+ 3 n n!Γ(n + 4 )
= Bi(0)y1 (x) + Bi′ (0)y2 (x)
n=0
9 3 n=0
9 3

where
1 1
Ai(0) = 2 , Ai′ (0) =− 1 ,
32/3 Γ( 3 ) 31/3 Γ( 3 )
1 31/6
Bi(0) = 2 , Bi′ (0) = 1 .
31/6 Γ( 3 ) Γ( 3 )

Many “special functions” are arising from solutions of 2nd order linear ODEs with
polynomial coefficients.

7.2 Ordinary and Regular Singular Points

Consider the more general 2nd order linear differential equation

P (x)y ′′ + Q(x)y ′ + R(x)y = 0

where P, Q, R are power series in (x − x0 ) on some open interval |x − x0 | < R.

• x0 is called an ordinary point, or a non-singular point, of the differential equa-

tion if P (x0 ) 6= 0.

• x0 is called a singular point of the differential equation if P (x0 ) = 0.

• A singular point x0 is called a regular singular point if the following limits are
finite:
(x − x0 )Q(x) (x − x0 )2 R(x)
lim , lim
x→x0 P (x) x→x0 P (x)
Roughly speaking, that means we could find series expansions of the form
∞
Q(x) b−1 X
= + bn (x − x0 )n (has at worst a pole of order 1 at x0 )
P (x) x − x0
n=0

∞
R(x) c−2 c−1 X
= 2
+ + cn (x−x0 )n (has at worst a pole of order 2 at x0 )
P (x) (x − x0 ) x − x0
n=0

• In order words, the differential equation has a regular singular point at x = x0 if

the equation can be written as

(x − x0 )2 y ′′ + (x − x0 )P̃ (x)y ′ + Q̃(x)y = 0

where P̃ (x) and Q̃(x) can be written as power series in (x−x0 ) in some neighborhood
of x0 .

Suppose x0 is an ordinary point, and P (x) 6= 0 in |x − x0 | < R; otherwise choose a

smaller radius radius of the interval. One can then turn the equation into the standard
form
Q(x) ′ R(x)
y ′′ + y + y=0
P (x) P (x)
 43

where the coefficient functions can be written as power series in the interval |x − x0 | < R.
The existence and uniqueness of solution of IVP then follows from previous discussion.
The simplest example of 2nd order linear differential equations with regular singular
points is the following Euler differential equation:

ax2 y ′′ + bxy ′ + cy = 0.

where a 6= 0, b, c are constants.

Note that one may expect that the Euler equation should have a solution of the form
y = xr for x > 0, since by putting xr into the equation, we have

[ar(r − 1) + br + c]xr = 0

i.e., y = xr is a solution of the Euler differential equation if and only if r is a root of the
indicial equation
ar(r − 1) + br + c = 0
Remark To take care of the solution of the Euler differential equation for x < 0, just
consider y = |x|r = (−x)r . The same conclusion follows.

Example 7.2.1 Solve the differential equation 2x2 y ′′ + 5xy ′ + y = 0, x > 0.

Obviously, x = 0 is a regular singular point of the differential equation.
The equation 2r(r − 1) + 5r + 1 = 2r 2 + 3r + 1 = (2r + 1)(r + 1) = 0 has two roots,
r = −1 or r = −1/2.
Hence we have two solutions y1 = x−1 and y2 = x−1/2 . Their Wronskian is
1 1
W [y1 , y2 ](x) = x−1 (− x−3/2 ) − (−x−2 )x−1/2 = x−5/2 6= 0
2 2
i.e., y1 , y2 form a set of fundamental solutions, and hence the general solution is

y = C1 x−1 + C2 x−1/2 .

Remark Generally speaking, the indicial equation ar(r − 1) + br + c = 0 of the Euler

differential equation may have just one repeated real root r, or two complex roots α ± iβ.

• If r is a repeated real root such that y1 = xr is a solution of ax2 y ′′ + bxy ′ + cy = 0

for x > 0, then the method of reduction of order can provide another solution of
the the equation which has the form y2 = u(x)xr for some u(x) (Exercise: Find this
u(x); or check directly that y2 = xr ln x is another solution of the equation in this
case, such that W [y1 , y2 ](x) 6= 0.)

• If α ± iβ are two complex roots of the indicial equation, then

α+iβ
xα+iβ = eln x = e(α+iβ) ln x = eα ln x eiβ ln x

= eα ln x (cos(β ln x) + i(sin(β ln x)) = xα cos(β ln x) + ixr sin(β ln x)

The real part and imaginary part of the complex-valued solution then form a set
of fundamental solutions of the equation, namely, xα cos(β ln x) and xα sin(β ln x).
(Exercise: Compute the Wronskian of these two solutions.)
 44

Remark Note that a 2nd order homogeneous linear ODEs with a regular singular point
at x = 0 is essentially a differential equation close to the Euler differential equation near
x = 0:
P (x)y ′′ + Q(x)y ′ + R(x)y = 0
xQ(x) ′ x2 R(x)
x2 y ′′ + x y + y=0
P (x) P (x)
which is closed to the Euler differential equation

x2 y ′′ + p0 xy ′ + q0 y = 0

near x = 0, with
xQ(x) x2 R(x)
lim = p0 , lim = q0
x→0 P (x) x→0 P (x)

Example 7.2.2 The Legendre equation

(1 − x2 )y ′′ − 2xy ′ + α(α + 1)y = 0

has two regular singular points at x = ±1. Any point x 6= ±1 is an ordinary point of the
equation.
Just note that
−2x α(α + 1)
lim (x − 1) = 1, lim (x − 1)2 = 0,
x→1 (1 − x)(1 + x) x→1 (1 − x)(1 + x)
and
−2x α(α + 1)
lim (x + 1) = 1, lim (x + 1)2 = 0.
x→−1 (1 − x)(1 + x) x→−1 (1 − x)(1 + x)

Exercise Since x = 0 is an ordinary point of the Legendre equation, and 1 − x2 6=

0 for |x| < 1. Find the recursive relation which determines the coefficients of series
∞
X
solutions an xn of the equation in |x| < 1. Show also that the fundamental solutions
n=0
corresponding to initial conditions a0 = 1, a1 = 0, and a0 = 0, a1 = 1 respectively, have
radius of convergence R = 1. Answer: an+2 = − (α+n+1)(α−n)
(n+2)(n+1) an .
(Of particular interest is the case when α is a positive integer, which gives rise to
a polynomial solution of the Legendre equation. (Why?) These so called Legendre
polynomials appear in many application problems, and have very rich properties. In fact,
many other interesting “orthogonal polynomials” are closely related to linear differential
equations with polynomial coefficients.)

7.3 The Method of Frobenius

It suffices to say that the Frobenius type solution at a regular singular point x = x0 of a
differential equation is inspired by the case of Euler differential equation. The main idea
is to look for series solutions of the form
∞
X ∞
X
y(x) = (x − x0 )ν an (x − x0 )n = an (x − x0 )n+ν , where a0 6= 0
n=0 n=0
 45

Note that the exponent ν may not be an integer.

Let’s take a look at a couple of simple examples first.

Example 7.3.1 Find the Frobenius type solutions of the equation 4xy ′′ + 2y ′ + y = 0,
x > 0, which has a regular singular point at x = 0.
X∞
Putting y(x) = an xn+ν into the equation, we have
n=0
∞
X ∞
X
y′ = (n + ν)an xn+ν−1 , y ′′ = (n + ν)(n + ν − 1)an xn+ν−2
n=0 n=0

and the differential equation is turned into

∞
X ∞
X ∞
X
4 (n + ν)(n + ν − 1)an xn+ν−1 + 2 (n + ν)an xn+ν−1 + an xn+ν = 0
n=0 n=0 n=0
∞
X ∞
X ∞
X
(4a0 ν(ν−1)+2a0 ν)xν−1 + 4(n+ν)(n+ν−1)an xn+ν−1 +2 (n+ν)an xn+ν−1 + an xn+ν = 0
n=1 n=1 n=0
∞
X ∞
X ∞
X
ν−1 n+ν n+ν
(4a0 ν(ν−1)+2a0 ν)x + 4(n+ν+1)(n+ν)an+1 x +2 (n+ν+1)an+1 x + an xn+ν = 0
n=0 n=0 n=0
(
(4ν(ν − 1) + 2ν)a0 = 0
4(n + ν + 1)(n + ν)an+1 + 2(n + ν + 1)an+1 + an = 0
Hence
an
2ν(2ν − 1)a0 = 0, an+1 = −
2(n + ν + 1)(2n + 2ν + 1)
i.e,. ν = 0 or ν = 12 .
an−1
Case ν = 0: an = −
2n(2n − 1)
a0 a1 a0 a2 a0 a0
a1 = − , a2 = − = , a3 = − = − , · · · an = (−1)n
2·1 4·3 4! 6·5 6! (2n)!
Thus we have the solution
 x x2 x3 xn  √
y1 = a0 1 − + − + · · · + (−1)n + · · · = a0 cos x.
2! 4! 6! ()!
an−1
Case ν = 21 : an = −
(2n + 1)2n
a0 a1 a0 a2 a0 a0
a1 = − , a2 = − = , a3 = − = − , · · · an = (−1)n
3·2 5·4 5! 7·6 7! (2n + 1)!
Thus we have the solution

1/2
 x x2 x3 n xn  √
y2 = a0 x 1− + − + · · · + (−1) + · · · = a0 sin x.
3! 7! 7! (2n + 1)!
The general solution of the equation is
√ √
y = C1 cos x + C2 sin x .
 46

Exercise Find a set of fundamental solutions of 2x2 − xy ′ + (1 + x)y = 0, x > 0, which

are Frobenius type solutions.

Example 7.3.2 The differential equation x2 y ′′ − 2x2 y ′ + (x2 + 41 )y = 0, x > 0, has a

regular singular point at x = 0. Find solutions by the method of Frobenius.
X∞
Let y(x) = an xn+r , where a0 6= 0. Then by putting the series into the equation,
n=0
we have
∞ ∞ ∞ ∞
X
n+r
X
n+r+1
X
n+r+2 1X
(n + r)(n + r − 1)an x − 2(n + r)an x + an x + an xn+r = 0
n=0 n=0 n=0
4 n=0

∞ ∞ ∞
X 1 n+r
X
n+r+1
X
((n + r)(n + r − 1) + )an x − 2(n + r)an x + an xn+r+2
n=0
4 n=0 n=0
1 1
(r(r − 1) + )a0 xr + (((r + 1)r + )a1 − 2ra0 )xr+1
4 4
∞
X 1
+ ((n + r + 2)(n + r + 1) + )an+2 − 2(n + r + 1)an+1 + an )xn+r+2 = 0
4
n=0
1 1
(r − )2 a0 xr + ((r + )2 a1 − 2ra0 )xr+1
2 2
∞
X 1
+ ((n + r + 2)(n + r + 1) + )an+2 − 2(n + r + 1)an+1 + an )xn+r+2 = 0
4
n=0

Thus r = 21 , a1 = a0 , and (n + 2)2 an+2 = (2n + 3)an+1 − an ; i.e.,

(2n + 3)an+1 − an
an+2 =
(n + 2)2

Hence
5a0
3a0 − a0 a0 2 − a0 a0
a2 = 2
= , a3 = 2
=
2 2 3 3!
and inductively
(2n+3)a0 (n+1)a0
(n+1)! − (n+1)! a0
an+2 = =
(n + 2)2 (n + 2)!
Hence the Frobenius type solutions are given by

x x2
y1 (x) = a0 x1/2 (1 + + + · · · ) = a0 x1/2 ex
1! 2!
Exercise Find a solution of the form y2 (x) = y1 (x)u(x) of the differential equation
by the method of reduction of order such that the general general solution is given by
y = C1 y1 (x) + C2 y2 (x). (Answer: x1/2 ex ln x.)
 47

In general, after rewriting a 2nd order homogeneous linear ODE with 0 as a regular
singular point into the form
∞
X ∞
X
x2 y ′′ + x p n xn y ′ + q 0 xn y = 0
n=0 n=0

∞
X
y(x) = an xn+r where a0 6= 0 is a solution of the equation if
n=0

∞
X ∞
X ∞
X ∞
X ∞
X
(n + r)(n + r − 1)an xn+r + p n xn (n + r)an xn+r + q n xn an xn+r = 0
n=0 n=0 n=0 n=0 n=0

∞
X ∞
X ∞
X ∞
X ∞
X
(n + r)(n + r − 1)an xn + p n xn (n + r)an xn + q n xn an xn = 0
n=0 n=0 n=0 n=0 n=0
i.e.,

x0 -term : F (r) = r(r − 1) + p0 r + q0 = 0 Indicial Equation

x-term : ((1 + r)r + p0 (1 + r) + q0 )a1 + (rp1 + q1 )a0
i.e., F (1 + r)a1 + (rp1 + q1 )a0
2
x -term : (2 + r)r + p0 (2 + r) + q0 )a2 + ((1 + r)p0 + q0 )a1 + (rp1 + q1 )a0
i.e., F (2 + r)a2 + (rp1 + q1 )a0 + ((1 + r)p0 + q0 )a1
..
.
n−1
X
xn -term : F (n + r)a2 + ((k + r)pn−k + qn−k )ak
k=0

Hence r must be a root of the indicial equation

r(r − 1)p0 r + q0 = 0

and the recursive relation to determine the coefficients is

n−1
X
F (n + r)an = − ((k + r)pn−k + qn−k )ak
k=0
Pn−1
k=0 ((k + r)pn−k + qn−k )ak
an = −
F (n + r)
which works as long as F (n + r) 6= 0 for all n ≥ 1.

• If r1 ≥ r2 are two real roots of the indicial equation, then n + r1 cannot be a root
of the indicial equation for n ≥ 1, i.e., F (n + r1 ) 6= 0, and hence a Frobenius type
solution of the following form
∞
X
r1
y1 (x) = x an xn+r
n=0

must exist.
 48

If in addition r1 − r2 is not a non-negative integer, i.e., r2 6= n + r1 for any integer

n ≥ 1, such that F (n + r1 ) 6= 0, then a second Frobenius type solution linearly
independent of the previous one
∞
X
y1 (x) = xr2 bn xn+r
n=0

can be found.

• If r = α ± iβ are two complex roots of F (r) = 0, then F (n + α + iβ) 6= 0 for all

n ≥ 1, and hence a complex-valued Frobenius type solution must exist
∞
X
y(x) = xα+iβ an xn+r
n=0

Taking the real part and imaginary part, two linearly independent solutions of the
following form can be found:
∞
X ∞
X
eαx cos(ln x) an xn , eαx sin(ln x) bn xn
n=0 n=0

• If r1 − r2 = 0, then reduction of order shows that the differential equation has a

solution of the form
∞
X
y2 (x) = y1 (x) ln x + xr1 bn xn
n=0

• If r1 −r2 = N is a positive integer, then reduction of order shows that the differential
equation has a solution of the form
∞
X
y2 (x) = Cy1 (x) ln x + xr2 bn xn
n=0

where the constant C may sometimes turn out to be zero.

In the last two cases, one may put y2 (x) into the equation to determine the coefficients
bn . However, the algebra may not be tidy, and the general pattern for bn may not be
recognizable.

Remark We add some details here about the reduction of order process here.
Let y1 = xr ∞ n r
P
n=0 an x = x h(x), h(0) = a0 6= 0, be a Frobenius type solution of the
equation
∞
X X ∞
2 ′′ n ′
x y +x pn x y + q0 xn uy1 = 0
n=0 n=0

To find another solution of the form y = y1 (x)u(x), just put it into the equation to see
that
∞
X ∞
X
x2 (u′′ y1 + 2u′ y1′ + uy1′′ ) + x pn xn (u′ y1 + uy1′ ) + q0 xn uy1 = 0
n=0 n=0
 49

∞
X ∞
X ∞
X
x2 (u′′ y1 + 2u′ y1′ ) + x pn xn y1 u′ + (x2 y1′′ + x pn xn y1′ + q0 xn y1 )u = 0
n=0 n=0 n=0
∞
X
x2 y1 u′′ + (2x2 y1′ + x pn xn y1 )u′ = 0
n=0

2y1′
u′′ + ( + p0 x−1 + p1 + p2 x−2 + · · · )u′ = 0
y1
which is a first order linear ODE in u′ . Thus
 R 2y′ 
d 1 −1 −2
′ ( y1 +p0 x +p1 +p2 x +··· )dx
ue =0
dx
R 2y ′
− ( y 1 +p0 x−1 +p1 +p2 x+p3 x2 +··· )dx
u′ (x) = Ce 1

For simplicity sake, just tak C = 1, and therefore

∞
2 1 −(p1 x+p2 x2 /2+··· ) 1 X
u′ (x) = e−(ln y1 +p0 ln x+p1 x+··· ) = e = cn xn
xp0 y12 xp0 +2r
n=0

where c0 6= 0, recalling that y1 = xr h(x), where h(0) 6= 0.

Note that if r1 and r2 are the roots of the indicial equation r(r − 1) − p0 r + q0 = 0,
i.e.,
r 2 + (p0 − 1)r + q0 = 0
then r1 + r2 = 1 − p0 .
Case r1 = r2 :
In this case, let r = r1 = r2 , hence p0 + 2r = 1, and then u′ (x) looks like
∞
1 X
u′ (x) = · cn xn
x n=0

where c0 6= 0. Then by integrating the series, one must have a ln x term

u(x) = c0 ln x + c1 x + · · ·

After dividing by c0 , one can reach a second solution of the differential equation in the
following form:
∞
X
y2 (x) = y1 ln x + xr bn xn
n=0
Case r1 = r2 :
Suppose r1 − r2 = N is a positive integer. Then r1 = r2 + N , and

2r1 + p0 = r1 + r2 + N + p0 = 1 + N

Thus
∞
!
c0 −N 1
Z X Z
n−1−N
u(x) = cn x dx = − x −· · ·−cN −1 x−1 +cN dx+ higher order terms
N x
n=0
 50

which may or may not have a logarithmic term, depending on whether cN is non-zero or
not.
Note that y1 (x)x−N +k = xr1 x−N xk h(x) = xr2 xk h(x) for all nonnegative integer k in
this case. Thus a second solution of the follow form can be found:
∞
X
y2 (x) = Cy1 (x) ln x + xr2 bn xn
n=0