Section 8.

4 The Frobenius Method 463

example showing why To appreciate why an ordinary power series solution cannot be developed
no power series around a regular singular point, it will be sufcient to consider the CauchyEuler
solution exists about equation
a singular point
x 2 y + 3xy + 2y = 0,
which has a regular singular point at the origin. This CauchyEuler equation was
solved analytically in Example 6.10, where its solution was found to be
y(x) = C1 x 1 cos(ln |x|) + C2 x 1 sin(ln |x|).
The reason that no power series solution exists in this case is seen to be the presence
of the factor x 1 and the function ln |x| in the analytical solution, neither of which
can be expanded in a power series about the origin.

Summary The regular and singular points of a general homogeneous second order linear variable
coefcient differential equation were dened and illustrated by example. It was shown
how, if necessary, a singular point occurring at x = a could be shifted to the origin, and
an example was used to demonstrate why an ordinary power series solution cannot be
developed around a regular singular point.

EXERCISES 8.3
Identify the nature of the singular points in each of the 4. xy + (1 x)y + ny = 0 (n > 0).
following equations. 5. (x + 4)3 y + 2(x + 4)y + xy = 0.
1. (1 x)2 y + 2(x 1)y + y = 0. 6. (x 2 4)y + (x + 3)y 5(x + 1)y = 0.
2. x 2 y + 3x 2 y + (1 + x 2 )y = 0. 7. (3 x)2 y + 4y + cos x(3 x 2 )y = 0.
3. (1 + x)2 y + 2y + y = 0. 8. x 2 y + 8y + 3xy = 0.

8.4 The Frobenius Method

A generalization of the power series method that was introduced by Frobenius
(18491917) enables a solution of a homogeneous linear differential equation to be
developed about a regular singular point. He considered the differential equation
a(x)y + b(x)y + c(x)y = 0, (22)
and established the following result that is stated without proof.

THEOREM 8.1 Frobenius theorem Let x0 be a regular singular point of (22). Then, in some
interval 0 < x x0 < d, the equation will always possess at least one solution of
the form
 
y(x) = (x x0 )c a0 + a1 (x x0 ) + a2 (x x0 )2 +


= (x x0 )c an (x x0 )n ,
n=0

the Frobenius where a0 = 0 and c is a real or complex number. A second linearly independent
theorem and solution of similar form will exist that may contain a logarithmic term, though
method of solution with a different value of c and some other coefcients b0 , b1 , b2 , . . . in place of
464 Chapter 8 Series Solutions of Differential Equations, Special Functions, and SturmLiouville Equations

the coefcients a0 , a1 , a2 , . . . . Taken together, these two solutions form a basis of

solutions for the differential equation.

GEORG FERDINAND FROBENIUS (18491917)

A German mathematician whose main research was in group theory and analysis. He worked in
Zurich and Berlin and published his method for the series solution of linear ordinary differential
equations in 1873.

For simplicity, and because of their frequent occurrence, in what follows we will
develop the Frobenius method in terms of a slightly less general class of equations
by setting a(x) = x 2 in (22). So we will consider the equation
x 2 y + b(x)y + c(x)y = 0, (23)
and write it in the standard form
y + P(x)y + Q(x)y = 0, (24)
where
p(x) q(x)
P(x) = and Q(x) = , (25)
x x2
and assume that p(x) and q(x) are analytic functions at x = 0. So we will only
consider equations of the form (24) with regular singular points at the origin.
To determine the exponent c in Theorem 8.1 we substitute a solution of the
form


y(x) = x c an x n (26)
n=0

into equation (24), where c is to be determined along with the coefcients an .

When making this substitution we will need to use the following results obtained
by differentiation of (26):


y (x) = ca0 x c1 + (c + 1)a1 x c + (c + 2)a2 x c+1 + = (n + c)an x n+c1 (27)
n=0

and
y (x) = c(c 1)a0 x c2 + (c + 1)ca1 x c1 + (c + 2)(c + 1)a2 x c +


= (n + c)(n + c 1)an x n+c2 . (28)
n=0

As the functions p(x) and q(x) are assumed to be analytic at the origin, they
can be expanded as the Maclaurin series
p(x) = p0 + p1 x + p2 x 2 + and q(x) = q0 + q1 x + q2 x 2 + . (29)
Substituting (27) to (29) into (24) leads to the result
x c2 [c(c 1)a0 + (c + 1)ca1 x + ]
 
+ p0 + p1 x + p2 x 2 + x c2 (ca0 + (c + 1)a1 x + )
  
+ x c2 q0 + q1 x + q2 x 2 + a0 + a1 x + a2 x 2 + = 0.
 Section 8.4 The Frobenius Method 465

If (26) is to be a solution of (24), the coefcient of each power of x in this last

result must vanish to make it an identity. Collecting terms involving the same power
of x and equating their coefcients to zero will lead to a sequence of equations
connecting the coefcients an in (26), and equating the coefcient of the lowest
power of x to zero will give an equation from which c can be determined.
The lowest power of x in the preceding result is x c2 , so collecting terms in-
volving x c2 and equating the coefcient of x c2 to zero gives

[c(c 1) + p0 c + q0 ]a0 = 0.

As Theorem 8.1 requires a0 = 0, it follows that c is determined by the equation

c(c 1) + p0 c + q0 = 0. (30)

indicial equation This equation is called the indicial equation associated with differential equation
(24), because it determines the permissible values of the index c to be used in the
solution given in Theorem 8.1.
The indicial equation of differential equation (24) can be constructed without
the need to make the substitution (26), because it is easily seen that

p0 = lim [x P(x)] and q0 = lim [x 2 Q(x)]. (31)

x0 x0

For the class of equations of type (24) that all have a regular singular point at the
origin, the appropriate form of the Frobenius theorem follows from Theorem 8.1 if
we set x0 = 0.
It is important to notice that for a general equation (22) in which a(x) = x 2 the
indicial equation does not take the form given in (30). When this situation arises
the indicial equation must be obtained by substituting (26) into (22) and equating
to zero the coefcient of the lowest power of x that occurs in the expansion.
As the indicial equation is a quadratic equation in c, the following relationships
between its roots c1 and c2 are possible:

(a) Roots c1 and c2 are real and distinct and do not differ by an integer
(b) Roots c1 and c2 are real and differ by an integer
(c) Roots c1 and c2 are real and equal
(d) Roots c1 and c2 are complex conjugates

The reason for identifying these different cases is to be found in the following
theorem, which is stated without proof in terms of a differential equation with a
regular singular point located at the origin (see references [3.3] and [3.5]).

THEOREM 8.2 Forms of Frobenius solution depending on the nature of c1 and c2 Let a differen-
tial equation of the form

x 2 y + x[x P(x)]y + [x 2 Q(x)]y = 0

have a regular singular point at x = 0. Let x P(x) and x 2 Q(x) each be capable of
expansion as convergent power series in an interval |x| < d, where d > 0 is the
466 Chapter 8 Series Solutions of Differential Equations, Special Functions, and SturmLiouville Equations

smaller of the two radii of convergence, and suppose that

p0 = lim [x P(x)] and q0 = lim [x 2 Q(x)].

x0 x0

Then in terms of the exponent c in (26), and the coefcients p0 and q0 , the indicial
equation for the differential equation is

c(c 1) + p0 c + q0 = 0,

with two roots c1 and c2 that may be real or complex conjugates.

The two linearly independent solutions of the differential equation that exist
depend on the relationship between the roots of the indicial equation, and they
take the following forms.

Case (a) Real roots with c1 > c2 and c1 c2 neither zero nor a positive integer
In the intervals d < x < 0 and 0 < x < d the differential equation has two linearly
different forms of independent solutions of the form
Frobenius solution
and examples    



y1 (x) = |x|c1 1 + an x n and y2 (x) = |x|c2 1 + bn x n ,
n=1 n=1

where the coefcients an are obtained by substituting c = c1 in the recurrence re-

lation connecting coefcients and then setting a0 = 1, and the coefcients bn are
obtained in similar fashion by substituting c = c2 in the recurrence relation, replac-
ing an by bn and setting b0 = 1.

Case (b) Real roots with c1 c2 equal to a positive integer

In the intervals d < x < 0 and 0 < x < d the differential equation has two linearly
independent solutions of the form
 



y1 (x) = |x|c1
1+ an x n
and y2 (x) = Ay1 (x) ln |x| + |x|c2 n x n ,
n=1 n=1

where the coefcients an are determined as in Case (a), and the coefcients A
and n are found by substituting y(x) = y2 (x) in the differential equation. Some
differential equations for which c1 c2 is a positive integer have no logarithmic
term in their solution y2 (x), in which case A = 0.

Case (c) Real roots with c1 = c2

In the intervals d < x < 0 and 0 < x < d the differential equation has two linearly
independent solutions of the form
 



y1 (x) = |x| c1
1+ an x n
and y2 (x) = y1 (x) ln |x| + |x|c1 n x n ,
n=1 n=1
 Section 8.4 The Frobenius Method 467

where the coefcients an are determined as in Case (a), and the coefcients n are
found by substituting y(x) = y2 (x) into the differential equation.

Case (d) Complex conjugate roots

If c1 = + i and c2 = i with = 0, then in the intervals d < x < 0 and
0 < x < d the two linearly independent solutions of the differential equation are
the real and imaginary parts of
 


+i
y(x) = |x| 1+ an x n
,
n=1

where the coefcients an are determined as in Case (a).

It is important to recognize that the solutions in cases (a) to (d) of Theorem 8.2
all lie in intervals of the form 0 < x < d that do not contain the origin. A solution
in the interval d < x < 0 can be obtained from the above results by replacing x
by x and, depending on the relationship between the roots c1 and c2 , seeking a
solution in the manner indicated in the illustrative examples that follow.

Case (a) Roots c1 and c2 Are Distinct

and Do Not Differ by an Integer
EXAMPLE 8.8 Find the solution of
2xy + (x + 1)y + y = 0
in some interval 0 < x < d.

Solution As the coefcient of y vanishes at x = 0 the origin must be a singular

point of this equation. When the differential equation is written in standard form
we nd that P(x) = (x + 1)/(2x) and Q(x) = 1/(2x), so p0 = limx0 x P(x) = 1/2
and q0 = limx0 x 2 Q(x) = 0, showing that the origin is a regular singular point of
the differential equation.
From (30) the indicial equation is seen to be
 
1 1
c(c 1) + c = 0, or c c = 0,
2 2
showing that the permissible values of c are c = 0 and c = 1/2. As these values of
c are distinct and do not differ by an integer, the solution will be of the type given
in Theorem 8.2(a).
Setting


y(x) = an x n+c
n=0

and substituting into the differential equation in the usual way leads to the result




2 (n + c)(n + c 1)an x n+c1 + (n + c)an x n+c + (n + c)an x n+c1
n=0 n=0 n=0


+ an x n+c = 0.
n=0
468 Chapter 8 Series Solutions of Differential Equations, Special Functions, and SturmLiouville Equations

Shifting the summation index in the rst and third summations gives




2 (n + c + 1)(n + c)an+1 x n+c + (n + c)an x n+c + (n + c + 1)an+1 x n+c
n=1 n=0 n=1



+ an x n+c = 0,
n=0

and, nally, combining terms we arrive at the result




[2(n + c + 1)(n + c) + (n + c + 1)]an+1 x n+c + (n + c + 1)an x n+c = 0.
n=1 n=0

Separating out the term corresponding to n = 1 allows this to be written



[2c(c 1) + c]a0 x c1 + {[2(n + c + 1)(n + c) + (n + c + 1)]an+1
n=0

+ (n + c + 1)an }x n+c
= 0.
To proceed further we must now equate to zero the coefcient of each power of
x. Equating to zero the coefcient of x c1 simply gives the indicial equation, but
equating to zero the coefcient of x n+c for n = 0, 1, 2, . . . gives
(n + c + 1)(2n + 2c + 1)an+1 + (n + c + 1)an = 0.
As n + c + 1 = 0 this recurrence relation can be written
an
an+1 = .
2n + 2c + 1
Starting with the value c = 0, we nd that
an
an+1 = ,
2n + 1
so
a1 a0 a2 a0 a3 a0
a1 = a0 , a2 = = , a3 = = , a4 = = ,
3 3 5 35 7 357
a4 a0 a5 a0
a5 = = , a6 = = ,....
9 3579 11 3 5 7 9 11
Examination of a5 and a6 shows they can be written
2468 24 4!
a5 = a0 = a0
9! (2 4 + 1)!
and
25 5!
a6 = a0 .
(2 5 + 1)!
These expressions suggest that the coefcient of the general term in the series is
(1)n+1 2n n!
an+1 = a0 for n = 0, 1, 2, . . . ,
(2n + 1)!
and this is easily veried by mathematical induction. As we are considering the case
 Section 8.4 The Frobenius Method 469

in which c = 0, it follows from Theorem 8.2(a) that for some d1 > 0 one solution is
 

(1)n+1 2n n! n+1
y(x) = a0 1 + x .
n=0
(2n + 1)!
As the constant a0 = 0 is arbitrary, we set a0 = 1 and take for a fundamental solution
of the differential equation


(1)n+1 2n n!
y1 (x) = 1 + x n+1 for 0 < x < d1 .
n=0
(2n + 1)!
A second fundamental (linearly independent) solution follows by using the
other value c = 1/2, for which the recurrence relation becomes
an
an+1 = .
2n + 2
Using this result and recognizing that the coefcients an are not the same as the
ones in y1 (x), we nd that
a0 a1 a0 a2 a0
a1 = , a2 = = 2 , a3 = = 3 ,
2 22 2 2! 23 2 3!
a3 a0
a4 = = 4 ,....
24 2 4!
This pattern of coefcients suggests that the coefcient of the general term in
the series is
(1)n
an = n a0 ,
2 n!
and this also is easily veried by using an inductive argument. Setting the arbitrary
constant a0 = 1, it follows from Theorem 8.2(a) that for some d2 > 0 a second
fundamental solution is given by


(1)n
y2 (x) = x 1/2 x n = x 1/2 ex/2 , for 0 < x < d2 .
n=0
2n n!
The solutions y1 (x) and y2 (x) form a basis for solutions of the differential equation in
an interval of the form 0 < x < d, where d = min{d1 , d2 }. Thus, the general solution is
y(x) = C1 y1 (x) + C2 y2 (x), for 0 < x < d,
where C1 and C2 are arbitrary constants. The value of d is
d = min{R1 , R2 },
where R1 and R2 are the radii of convergence of the series solutions for y1 (x) and
y2 (x), respectively. In this case R1 = R2 = , so the general solution is valid for
x > 0.

Case (b) Roots c1 and c2 Are Real

and Differ by an Integer
EXAMPLE 8.9 Find the solution of
x 2 y + x(2 + x)y 2y = 0
in some interval 0 < x < d.
470 Chapter 8 Series Solutions of Differential Equations, Special Functions, and SturmLiouville Equations

Solution The equation has a singular point at the origin, and writing it in standard
form shows that P(x) = (2 + x)/x and Q(x) = 2/x 2 . Thus,

p0 = lim x P(x) = 2 and q0 = lim x 2 Q(x) = 2,

x0 x0

so the equation has a regular singular point at the origin. It follows from (30) that
the indicial equation is

c(c 1) + 2c 2 = 0, or c2 + c 2 = 0.

The permissible values of c are thus c = 2 and c = 1, and these differ by an integer.
Substituting the series


y(x) = an x n+c
n=0

into the differential equation gives





(n + c)(n + c 1)an x n+c + 2 (n + c)an x n+c + (n + c)an x n+c+1
n=0 n=0 n=0


2 an x n+c = 0.
n=0

Shifting the index in the third summation so it starts from n = 1 and separating out
the terms multiplied by x c enables the equation to be written


a0 (c2 + c 2)x c + {[(n + c)(n + c + 1) 2]an + (n + c 1)an1 }x n+c = 0.
n=1

Proceeding as usual and equating the coefcient of x c to zero simply gives

the indicial equation, whereas equating the coefcient of x n+c to zero gives the
recurrence relation
(n + c 1)
an = an1 for n = 1, 2, . . . .
[2 (n + c)(n + c + 1)]
Considering the larger root c = 1, as required by Theorem 8.2(b), we nd that
n
an = an1 for n = 1, 2, . . . .
[2 (1 + n)(2 + n)]
So the rst few coefcients are
a0 2 a0 3a2 a0
a1 = , a2 = a1 = , a3 = = ,
4 [2 3 4] 45 [2 4 5] 456
4a3 a0
a4 = = ,....
[2 5 6] 4567
As c = 1, setting the arbitrary constant a0 = 1, it follows from Theorem 8.2(b)
that for some d1 > 0 a fundamental solution of the differential equation is
 
x x2 x3 x4
y1 (x) = x 1 + + ,
4 45 456 4567