The Method of Frobenius

Department of Mathematics
IIT Guwahati
SHB/SU

SHB/SU MA-102 (2020)

If either p(x) or q(x) in
y 00 + p(x)y 0 + q(x)y = 0

is not analytic near x0 , then power series solutions near x0 may

or may not exist.
Example: Try to find a power series solution of
x 2 y 00 − y 0 − y = 0 (1)
about the point x0 = 0.
Assume that a solution
∞
X
y (x) = an x n
n=0

exists.

SHB/SU MA-102 (2020)

Substituting this series in (1) leads to the recursion formula
n2 − n − 1
an+1 = an .
n+1
The ratio test shows that this power series converges only for
x = 0. Thus, there is no power series solution valid in any
open interval about x0 = 0.
On the other hand consider the following example.
Example:
x 2 y 00 (x) − 2xy 0 (x) + 2y (x) = 0.
It has a fundamental solution set {x, x 2 } although x = 0 is a
singular point of the equation. So every solution is of the form
y (x) = C1 x + C2 x 2
which is a polynomial in x and therefore a power series
solution about x = 0 with infinite radius of convergence.
The Method of Frobenius is a useful method to treat such
equations.
SHB/SU MA-102 (2020)
Cauchy-Euler equations revisited

Recall that a second order homogeneous Cauchy-Euler

equation has the form
ax 2 y 00 (x) + bxy 0 (x) + cy (x) = 0, x > 0, (2)
where a(6= 0), b, c are real constants. Writing (2) in the
standard form as
b c
y 00 + p(x)y 0 + q(x)y = 0, where p(x) = , q(x) = 2 .
ax ax

Note that x = 0 is a singular point for (2). We seek solutions

of the form
y (x) = x r
and then try to determine the values for r .

SHB/SU MA-102 (2020)

Set
L(y )(x) := ax 2 y 00 (x) + bxy 0 (x) + cy (x) and w (r , x) := x r .
Now

L(w )(x) = ax 2 r (r − 1)x r −2 + bxrx r −1 + cx r

= {ar 2 + (b − a)r + c}x r .

Thus,
w = x r is a solution ⇐⇒ r satisfies
ar 2 + (b − a)r + c = 0. (3)
The equation (3) is known as the auxiliary or indicial equation
for (2).

SHB/SU MA-102 (2020)

Case I: When (3) has two distinct roots r1 , r2 . Then
L(w )(x) = a(r − r1 )(r − r2 )x r .
The two linearly independent solutions are

y1 (x) = w (r1 , x) = x r1 , y2 (x) = w (r2 , x) = x r2 for x > 0.

Case II: When r1 = α + iβ, r2 = α − iβ. Then
x α+iβ = e (α+iβ) ln x = e α ln x cos(β ln x) + i e α ln x sin(β ln x)
= x α cos(β ln x) + i x α sin(β ln x).
Thus, two linearly independent real-valued solutions are

y1 (x) = x α cos(β ln x), y2 (x) = x α sin(β ln x).

SHB/SU MA-102 (2020)

Case III: When r1 = r2 = r0 is a repeated roots. Then
L(w )(x) = a(r − r0 )2 x r .
Setting r = r0 yields the solution

y1 (x) = w (r0 , x) = x r0 , x > 0.

To find the second linearly independent solution, we note that
∂
{L(w )(x)}|r =r0 = {a(r − r0 )2 x r ln x + 2a(r − r0 )x r }|r =r0 = 0.
∂r
Since ∂r∂ L(w ) = L ∂w
  ∂
∂r
( ∂r and L commute), we obtain
 
∂w 
L = 0.
∂r r =r0

SHB/SU MA-102 (2020)

A second linearly independent solution is

∂w ∂ r 
y2 (x) = (r0 , x) = (x ) = x r0 ln x, x > 0.
∂r ∂r r =r0

Example: Find a general solution to

4x 2 y 00 (x) + y (x) = 0, x > 0.

Note that
L(w )(x) = (4r 2 − 4r + 1)x r .
The indicial equation has repeated roots r0 = 1/2. Thus, the
general solution is
√ √
y (x) = c1 x + c2 x ln x, x > 0.

SHB/SU MA-102 (2020)

What happens to the solution when x < 0?
If any root of the indicial equation is a positive integer, then
we get a solution that is in fact valid on the entire real line.
But when the roots are not integers or ln x appears in the
solution derived above, then they are not valid for x < 0.
To find solutions valid for x < 0, the Cauchy-Euler equation
needs to be transformed by setting x 7→ −ξ. The resulting
Cauchy-Euler equation is exactly the same as before but in the
dependent variable ξ. Therefore its general solution for ξ > 0
is exactly the same as the one for positive values of x but
written in terms of ξ.
The solution for the original equation are then obtained by
replacing ξ by −x = |x| as x < 0.

SHB/SU MA-102 (2020)

The process leads to the following facts:
Let r1 , r2 be the roots of the indicial equation associated with
a second order Cauchy-Euler equation. Its general solution in
any deleted neigbourhood (ρ, 0) ∪ (0, ρ), ρ > 0, of 0 is given by
• y (x) = c1 |x|r1 + c2 |x|r2 if r1 and r2 are distinct real roots.
• y (x) = (c1 + c2 ln|x|)|x|r1 if r1 = r2 .
• y (x) = |x|α [c1 cos(β ln|x|) + c2 sin(β ln|x|)] if
r1 = α + iβ = r̄2 , β 6= 0.

Example: A general solution to

4x 2 y 00 (x) + y (x) = 0, x ∈ (ρ, 0) ∪ (0, −ρ), ρ > 0,

is given by p
y (x) = (c1 + c2 ln |x|) |x|.

SHB/SU MA-102 (2020)

The Method of Frobenius

To motivate the procedure, recall the Cauchy-Euler equation

in the standard form

y 00 (x) + p(x)y 0 (x) + q(x)y (x) = 0, (4)

where
p0 q0
p(x) = , q(x) = 2 with p0 = b/a q0 = c/a.
x x
The indicial equation is of the form
r (r − 1) + p0 r + q0 = 0. (5)

If r = r1 is a root of (5), then w (r1 , x) = x r1 is a solution to

(4).

SHB/SU MA-102 (2020)

Assume that xp(x) and x 2 q(x) (instead of being constants)
are analytic functions. Then
∞
X
xp(x) = p0 + p1 x + p2 x + · · · = pn x n , (6)
n=0

∞
X
2 2
x q(x) = q0 + q1 x + q2 x + · · · = qn x n (7)
n=0

in some neighborhood of x = 0. Then, it follows that

lim xp(x) = p0 and lim x 2 q(x) = q0 .
x→0 x→0

Therefore, it is reasonable to expect that the solutions to (2)

will behave (for x near 0) like the solutions to the
Cauchy-Euler equation

x 2 y 00 (x) + p0 xy 0 (x) + q0 y (x) = 0.

SHB/SU MA-102 (2020)

When p(x) and q(x) satisfy (6) and (7), we say that the
singular point x = 0 is regular. This observation leads to the
following definition.
Definition: A singular point x0 of
y 00 (x) + p(x)y 0 (x) + q(x)y (x) = 0
is said to be a regular singular point if both (x − x0 )p(x) and
(x − x0 )2 q(x) are analytic at x0 . Otherwise x0 is called an
irregular singular point.
Example: Classify the singular points of the equation
(x 2 − 1)2 y 00 (x) + (x + 1)y 0 (x) − y (x) = 0.
The singular points are 1 and −1. Note that x = 1 is an
irregular singular point and x = −1 is a regular singular point.

SHB/SU MA-102 (2020)

Series solutions about a regular singular point: Frobenius
Method
Assume that x = x0 is a regular singular point of
y 00 (x) + p(x)y 0 (x) + q(x)y (x) = 0.
In the method of Frobenius, we seek solutions of the form
∞
X
r
w (r , x) = (x − x0 ) an (x − x0 )n , x > x0 .
n=0

Assume that a0 6= 0. We now determine r and an , n ≥ 1.

Clearly w (r , x) solves the given equation for x > x0 if and only
if it solves

(x − x0 )2 y 00 (x) + (x − x0 )[(x − x0 )p(x)]y 0 (x)

+ (x − x0 )2 q(x)y (x) = 0 (8)
SHB/SU MA-102 (2020)
As x0 is a regular singular point,
∞
X ∞
X
(x − x0 )p(x) = pn (x − x0 )n , (x − x0 )2 q(x) = qn (x − x0 )n .
n=0 n=0

Differentiating w (r , x) with respect to x, we have

∞
X
0
w (r , x) = (n + r )an (x − x0 )n+r −1 ,
n=0
∞
X
w 00 (r , x) = (n + r )(n + r − 1)an (x − x0 )n+r −2 .
n=0
0 00
Substituting w , w , w , (x − x0 )p(x) and (x − x0 )2 q(x) into (8),
∞
X
(n + r )(n + r − 1)an (x − x0 )n+r
n=0
∞ ∞
! !
X X
n+1 n+r −1
+ pn (x − x0 ) (n + r )an (x − x0 )
n=0 n=0
∞ ∞
! !
X X
n n+r
+ qn (x − x0 ) an (x − x0 ) = 0.
n=0 n=0

SHB/SU MA-102 (2020)

Grouping like powers of x − x0 , starting with the lowest power, r , gives

[r (r − 1) + p0 r + q0 ]a0 (x − x0 )r + [(r + 1)ra1 + (r + 1)p0 a1

+q0 a1 + p1 ra0 + q1 a0 ](x − x0 )r +1 + · · · = 0.

which can be written in compact form as

a0 F (r )(x − x0 )r +
∞
( n−1
)
X X
F (r + n)an + ak [(r + k)pn−k + qn−k ] (x − x0 )r +n = 0 (9)
n=1 k=0

where F (r ) = r (r − 1) + p0 r + q0 .
Equating coefficient of (x − x0 )r to zero, as a0 6= 0, we have the indicial
equation.
{r (r − 1) + p0 r + q0 } = 0 (10)
where p0 = lim (x − x0 )p(x), and lim (x − x0 )2 q(x) = q0 .
x→x0 x→x0

SHB/SU MA-102 (2020)

Suppose r1 , r2 are the roots of the indicial equation. Putting r = r1 and
equating coefficients of higher powers of x − x0 to zero gives the
recurrence relation
n−1
X
F (r + n)an + ak [(r + k)pn−k + qn−k ] = 0, n ≥ 1, (11)
k=0

from which the indices an , n ≥ 1 may be found in terms of a0 and r1

provided r2 6= r1 + n for all n = 1, 2, . . . .
Repeating the process with r = r2 , gives two solutions
∞
!
X
r1 n
y1 (x) = (x − x0 ) a0 + an (r1 )(x − x0 ) , a0 6= 0,
n=1
∞
!
X
r2 n
y2 (x) = (x − x0 ) a0 + an (r2 )(x − x0 ) a0 6= 0.
n=1

These form a fundamental solution set for x > x0 , if r1 , r2 ∈ R, r1 > r2

and r1 − r2 is not a positive integer.

SHB/SU MA-102 (2020)

Example: Find the roots of the indicial equation at the regular
singular point x = −1 of

(x 2 − 1)2 y 00 (x) + (x + 1)y 0 (x) − y (x) = 0,

and find the first four terms of two linearly independent series
solutions about x = −1 valid for all x + 1 > 0.
We find that
1
p0 = lim (x + 1)p(x) = lim (x − 1)−2 = ,
x→−1 x→−1 4
1
q0 = lim (x + 1)2 q(x) = lim [−(x − 1)−2 ] = − .
x→−1 x→−1 4
Thus, the indicial equation is given by
1 1
r (r − 1) + r − = 0.
4 4
and r1 = 1 and r2 = − 41 are the roots.
SHB/SU MA-102 (2020)
Here r1 − r2 = 54 not an integer.
Therefore the method provides two linearly independent
solutions of the form
∞
X ∞
X
n+r1
y1 (x) = an (x+1) ; a0 6= 0, and y2 (x) = bn (x+1)n+r2 ,
n=0 n=0

where the coefficients an ’s are to be determined in terms of a0

and r1 and bn ’s are to determined in terms of b0 and r2 by
using the recurrence relations
n−1
X
F (r + n)an + ak [(r + k)pn−k + qn−k ] = 0, n ≥ 1,
k=0

where F (r + n) = (r + n)(r + n − 1) + 14 (r + n) − 14 and

pn , qn , n ≥ 0, are the coefficients of the Taylor’s expansion of
1 2 1
(x + 1)p(x) = (x−1) 2 and (x + 1) q(x) = − (x−1)2 about

x = −1 respectively.
SHB/SU MA-102 (2020)
Noting the Taylor’s expansion
1 1 1 3 1
= + (x + 1) + (x + 1)2 + (x + 1)3 + · · · ,
(x − 1)2 4 4 16 8

we have p1 = −q1 = 41 , p2 = −q2 = 3

16 , and p3 = −q3 = 18 .
Using these with a0 = 1, r1 = 1, in the recurrence relation gives an = 0
for all n ≥ 1. Therefore one solution is

y1 (x) = x + 1.

Repeating with b0 = 1 and r2 = − 14 in the recurrence relations gives the

other solution as
 
1 5 1 2 445 3
y2 (x) = 1 1 − (x + 1) − (x + 1) + (x + 1) + · · · .
(x + 1) 4 4 4 162 × 84

Hence the GS is y (x) = c1 y1 (x) + c2 y2 (x).

SHB/SU MA-102 (2020)

If r1 = r2 , then there exist two linearly independent solutions of the form
∞
X
y1 (x) = an (r1 )(x − x0 )n+r1 ; a0 6= 0, x > x0
n=0
∞
X
y2 (x) = y1 (x) ln(x − x0 ) + bn (r1 )(x − x0 )n+r1 , x > x0 .
n=1

If r1 − r2 is a positive integer, then there exist two linearly independent

solutions of the form
∞
X
y1 (x) = an (r1 )(x − x0 )n+r1 ; a0 6= 0, x > x0 ,
n=0
∞
X
y2 (x) = Cy1 (x) ln(x − x0 ) + bn (r2 )(x − x0 )n+r2 , b0 6= 0, x > x0 .
n=0

where C is a constant that could be zero. Using y1 (x) and y2 (x) and
their derivatives in the equation and setting coefficients of powers of
x − x0 to zero gives the recurrence relations from which the coefficients
bn (r1 ), bn (r2 ), n = 1, 2, . . . , and the constant C in the second case are
determined.
SHB/SU MA-102 (2020)
If r1 ∈ C then r2 = r¯1 and the two linearly independent solutions are the
real and imaginary parts of
∞
X
y (x) = an (r1 )(x − x0 )n+r1 .
n=0

In the following example, we shall illustrate the Frobenius method for the
case that the indicial equation has repeated roots. The regular singular
point is x = 0 and the first few terms of a series expansion in powers of x
valid for x > 0 are obtained.
Example 2: Consider the DE
x 2 y 00 (x) − xy 0 (x) + (1 − x)y (x) = 0, x > 0.
We have lim xp(x) = p0 = −1 and lim x 2 q(x) = q0 = 1.
x→0 x→0
The indicial equation is r (r − 1) + p0 r + q0 = 0 ⇒ (r − 1)2 = 0 has two
equal roots r1 = r2 = 1.
To obtain the first series solution, take
∞
X
y1 (x) = an x n+1 .
n=0

SHB/SU MA-102 (2020)

Determine the coeffcients an ’s as in the previous examples and
obtain the first series solution as
∞
1 1 1 5 X 1
2
y1 (x) = x + x + x 3 + x 4 + x +··· = 2
x k+1 .
4 36 576 k=0
(k!)

The second linearly indpendent solution is of the form

∞
X
y2 (x) = y1 (x) ln x + bn x n+1 ,
n=1

where the coeffcients bn are to be

Pdetermined. Compute
0 0 −1 ∞
y2 (x) = y1 (x) ln x + x y1 (x) + n=1 (n + 1)bn x n .
∞
X
y200 (x) = y100 (x) ln x −x −2 y1 (x)+2x −1 y10 (x)+ n(n+1)bn x n−1 .
n=1

SHB/SU MA-102 (2020)

Substituting y2 (x), y20 (x) and y200 (x) in the differential equation
and after simplication it leads to
{x 2 y100 (x) − xy10 (x) + (1 − x)y1 (x)} ln x − 2y1 (x) + 2xy10 (x)
X ∞ X∞ X ∞
n+1 n+1
+ n(n + 1)bn x − (n + 1)bn x + bn x n+1
n=1 n=1 n=1
X∞
− bn x n+2 = 0.
n=1

Using the fact y1 is a solution and a shift in the indices of

summation gives
∞
X
2xy10 (x) 2
− 2y1 (x) + b1 x + (k 2 bk − bk−1 )x k+1 = 0.
k=2

SHB/SU MA-102 (2020)

Substituting the series expansions for y1 (x) and
∞
0
X xk
y1 (x) = (k + 1) , we have
k=0
(k!)2
∞ h
2
X 2k i
(2 + b1 )x + 2
+ k bk − bk−1 x k+1 = 0.
2

k=2
(k!)
Equating the coefficients equal to zero, we get
1h 2k i
b1 = −2, bk = 2 bk−1 − , k ≥ 2.
k (k!)2
Taking k = 2 and 3, compute
1 −3 1h 3 6 i −11
b2 = (b1 − 1) = , b3 = − − = .
22 4 9 4 36 108
Thus, a second linearly independent solution is
3 11 4
y2 (x) = y1 (x) ln x − 2x 2 − x 3 − x + ··· .
4 108
SHB/SU MA-102 (2020)
A GS is given by y (x) = c1 y1 (x) + c2 y2 (x), x > 0, where y1 (x)
and y2 (x) are two series solutions obtained as above.

The next example illustrates the method of Frobenius when

the indicial equation has roots that differ by a positive integer.
Example 3: Consider the DE

xy 00 (x) + 4y 0 (x) − xy (x) = 0, x > 0.

The roots of the indicial equations r1 = 0 and r2 = −3. Here
r1 − r2 is a positive integer. With r1 = 0, the first series
solution is given by
1 2 1 4
y1 (x) = 1 + x + x + ··· .
10 280

SHB/SU MA-102 (2020)

Since r1 − r2 = 3 is a positive integer, the second linearly
independent solution is of the form
∞
X
y2 (x) = Cy1 (x) ln x + bn x n−3 .
n=0

Subsititute the expression for y2 (x), y20 (x) and y200 (x) in the
differential equation leads to
{xy100 (x) + 4y10 (x) − xy1 (x)}C ln x + 3Cx −1 y1 (x) + 2Cy10 (x) +
∞
X ∞
X X∞
n−4 n−4
(n − 3)(n − 4)bn x + 4(n − 3)bn x − bn x n−2 = 0.
n=0 n=0 n=0

As y1 (x) is a solution to the differential equation, combine the

summations and simplify to obtain
∞
X
−1
3Cx y1 (x)+2Cy10 (x)−2b1 x −3 + [k(k−3)bk −bk−2 ]x k−4 = 0.
k=2

SHB/SU MA-102 (2020)

Substituting the series for y1 (x) and equating the coefficients
equal to zero, we have
1 1 1 1
b1 = 0, b2 = − b0 , C = b1 = 0, b4 = b2 = − b0 ,
2 3 4 8
7
b3 − 10 C 1 1 1
b5 = = b3 , b6 = b4 = − b0 ,
10 10 18 144
11
b5 − 280 C 1
b7 = = b3 .
28 280
Collecting the values of for the bn ’s and C = 0 gives
n 1 1 1 3 o
y2 (x) = b0 x −3 − x −1 − x − x + ···
2 8 144
n 1 2 1 4 o
+b3 1 + x + x + ··· ,
10 280
n 1 1 1 3 o
= b0 x −3 − x −1 − x − x + · · · + b3 y1 (x),
2 8 144

SHB/SU MA-102 (2020)

where b0 and b1 are arbitrary constants. In order to obtain a
second linearly independent solution, choose b0 to be nonzero.
Taking b0 = 1 and b3 = 0 gives
1 1 1 3
y2 (x) = x −3 − x −1 − x − x + ··· .
2 8 144
Thus, a GS is y (x) = c1 y1 (x) + c2 y2 (x), x > 0, with y1 (x)
and y2 (x) obtained as above.

SHB/SU MA-102 (2020)

To find a solution valid for x < x0 , the equation needs to be
transformed by using x 7→ x0 − ξ. The transformed equation
has the independent variable ξ and can be solved by assuming
that there is a solution of the form
∞
X
r
y (ξ) = ξ an ξ n , ξ > 0
n=0

via the method described above.

This is equivalent to assuming that the given equation has a
solution of the form
∞
X
r
y (x) = |x − x0 | an (x − x0 )n , for x < x0 .
n=0

SHB/SU MA-102 (2020)

Therefore if r1 and r2 are the roots of the indicial equation associated
with a regular singular point x0 of

y 00 (x) + p(x)y 0 (x) + q(x)y (x) = 0,

then two linearly independent solutions valid for all

x ∈ (x0 − ρ, x0 ) ∪ (x0 , x0 + ρ) for some ρ > 0 is of the form

∞
!
X
r1 n
y1 (x) = |x − x0 | a0 + an (r1 )(x − x0 ) , a0 6= 0
n=1
∞
!
X
r2 n
y2 (x) = |x − x0 | a0 + an (r2 )(x − x0 ) , a0 6= 0,
n=1

if r1 , r2 ∈ R, r1 > r2 and r1 − r2 is not a positive integer;

SHB/SU MA-102 (2020)

of the form
∞
X
y1 (x) = |x − x0 |r1 an (r1 )(x − x0 )n ; a0 6= 0,
n=0
∞
X
y2 (x) = y1 (x) ln(|x − x0 |) + |x − x0 |r1 bn (r1 )(x − x0 )n ,
n=1

if r1 = r2 ;
and of the form
∞
X
y1 (x) = |x − x0 |r1 an (r1 )(x − x0 )n ; a0 6= 0,
n=0
∞
X
y2 (x) = Cy1 (x) ln(|x − x0 |) + |x − x0 |r2 bn (r2 )(x − x0 )n , b0 6= 0,
n=0

if r1 − r2 is a positive integer. Here C is a constant that could be zero.

SHB/SU MA-102 (2020)

In all cases the power series involved in the solutions have a
positive radius of convergence bounded below by the minimum
of the radii of convergences of the Taylor’s expansions of
(x − x0 )p(x) and (x − x0 )2 q(x) about x = x0 .

SHB/SU MA-102 (2020)