THE METHOD OF FROBENIUS

We have studied how to solve many differential equations via series solutions. In this
section we learn how to extend series solutions to a class of differential equations that
appear at first glance to diverge in our region of interest.

Let’s consider the equation:

2 x 2 y ′′ + 7 x( x + 1) y ′ − 3 y = 0 (1)

and we are interested in finding the series solution to this equation in the vicinity of x0=0.

Your first instinct might be divide each term in (1) to produce:

7 x( x + 1) 3
y ′′ + 2
y′ − 2 y = 0 (2)
2x 2x

In the vicinity of x0=0, it appears that this equation is undefined and will not yield
meaningful solutions to the equation (1) near 0.

The necessary conditions for solving equations of the form of (2)

However, the method of Frobenius provides us with a method of adapting our series
solutions techniques to solve equations like this if certain conditions hold. Let’s consider
now those conditions.

First, we will write our second order differential equation as:

y ′′ + P ( x ) y ′ + Q ( x ) y = 0 (3)

When transformed into this form, our original equation in (1)

7 x( x + 1) 3
has P ( x) = 2
and Q ( x) = − 2 .
2x 2x

The theorem of Frobenius shows that if both (x-x0)P(x) and (x-x0)2Q(x) have meaningful
series solutions around x0, then a series solution to the differential equation can be found.

Let’s apply this theorem to eq. (2) to see if the conditions of this theorem hold:

We want to find a series solution in the neighborhood of x0=0, so (x-x0) = x. Then, we

construct the terms xP(x) and x2Q(x) and see if they have are well behaved at x= 0. It is
easy to show that both xP(x) and x2Q(x) are well behaved at x0=0 and will have power
series that converge near 0. Thus, the conditions of the Theorem of Frobenius are met,
and we can find a power series solution for (1).
The Method of Frobenius

If the conditions described in the previous section are met, then we can find at least one
solution to a second order differential equation by assuming a solution of the form:
∞ ∞
y = x r ∑ an x n = ∑ an x n+r (4)
n=0 n =0

where r and an are constants to be determined, and n = 0, 1, 2, 3,…While n is always an

integer, there is no such constraint on r. We will find that r may have negative or
fractional values.

Eq. (4) looks very familiar to our earlier work with series solutions, and we should expect
that we will have to derive recursion relations to find the exact values of an. The obvious
difference is that we also have to determine the values of the exponent r; it is these
factors of xr that allow us to find solutions to our differential equations. (Note that other
authors and sources will use a staggeringly diverse array of symbols to represent the
exponent that we have called r. There seems to be no standard for this; do not be
confused, they are all dummy variables referring to the same concept. But we will use r
throughout.)

The indicial equation and the values of r

The first step in using the method of Frobenius is to determine the values of r that allow
us to solve the differential equation. We do this by exploiting the fact that this method
produces a series where the first non zero term is a0.

We substitute our assumed form of the series solution (4) into our original equation and
determine the proper values of r by setting n=0; in other words, we find an expression
that will ensure the a0 term is non-zero. Let’s apply this technique to our original
equation (1). Substituting (4) into (1) gives us:

∞ ∞ ∞ ∞
2 x 2 ∑ (n + r )(n + r − 1)a n x n + r − 2 + 7 x 2 ∑ (n + r )a n x n + r −1 + 7 x ∑ (n + r )a n x n + r −1 − 3∑ a n x n + r = 0
n =0 n =0 n=0 n=0
(5)

Let’s look carefully at (5). Notice that unlike our earlier work with series solutions, we
do not change the lower limit of summation when we differentiate our solution. This is
because of the presence of the factor xr; as noted above, the value of r is chosen so that
the first non zero term of the expansion is the a0 term, so we do not change the lower
limit of summation upon differentiation as we did in our prior work with series solutions.
Now, let’s rewrite (5) by consolidating all powers of x inside the summation signs, and
get:
 ∞ ∞ ∞ ∞
2∑ ( n + r )(n + r − 1)a n x n + r + 7∑ (n + r )a n x n + r +1 + 7∑ ( n + r )a n x n + r − 3∑ a n x n + r = 0 (6)
n=0 n =0 n=0 n =0

We see immediately that three of the summations are to the power of n+r; and that only
one summation (the second one in (6)) is to a different power. We know that we want to
re-index this sum so that it is also to the power of n+r¸ and when we do this, eq (6)
becomes:
∞ ∞ ∞ ∞
2∑ ( n + r )(n + r − 1)a n x n + r + 7∑ (n + r − 1) a n −1 x n + r + 7∑ ( n + r )a n x n + r − 3∑ a n x n + r = 0
n=0 n =1 n =0 n =0

(7)

Now it is appropriate to change the lower limit of the second summation since we have to
re-index each term that involves n.

We are now in a position to determine the values of r for this equation. Remember that r
must be chosen so that the a0 term is non zero. We can “strip out” the n=0 term in (7)
and solve for the values of r. Setting n= 0 in (7) gives us:

[2r ( r − 1) + 7 r − 3]a 0 = 0 ⇒ ( 2r 2 + 5r − 3) a 0 = 0 (8)

Since we know that a0 is non-zero, eq. (8) implies:

2 r 2 + 5r − 3 = 0 (9)

Equation (9) is called the indicial equation, and its solution gives us the values of r we
use in finding our solution in the form of (4).

Solving (9) is easily done by factoring:

( 2r − 1)( r + 3) = 0 (10)

and our values of r are r= ½ and r=-3; we will see that our solutions to the differential
equation will involve factors of x1/2 and x – 3.

In fact, we can derive the indicial equation even before re-indexing summations. We
look for those summations that involve the lowest power of x, and use those terms with
n=0 to derive the indicial equation. In equation (6) above, the first, third and fourth sums
are the terms with the lowest power of x, and these are the terms that we use to derive the
indicial equation.

Finding the recursion relation

The next step in this method is a familiar one to us, namely, finding the recursion relation
that will enable us to determine the values of an. We begin by referring back to eq. (7):
 ∞ ∞ ∞ ∞
2∑ ( n + r )(n + r − 1)a n x n + r + 7∑ (n + r − 1) a n −1 x n + r + 7∑ ( n + r )a n x n + r − 3∑ a n x n + r = 0
n=0 n =1 n =0 n =0

We know that since we have all our exponents to the same power, we can combine like
terms to produce:

[2(n + r )(n + r − 1) + 7(n + r ) − 3]a n + 7(n + r − 1)a n −1 = 0 (11)

or equivalently:

− 7(n + r − 1)a n −1
an = (12)
[2(n + r ) − 1][(n + r ) + 3]

And this is the recursion relation for this particular differential equation.

We can make a couple of important points about recursion relations in the method of
Frobenius. First, you will have to use this recursion relation twice; once to determine
values of an when r = ½ , and a second time to determine values of an when r = -3.
Second, notice the denominator in (12). It is highly recommended that you factor the
coefficient of an in writing your final recursion relation.

Review the steps in deriving the indicial equation (10). You set n = 0 in deriving (10). If
you replace r with n+r , you will have determined the coefficient of an for all values of n
(including non zero values). Look now at the denominator in (12); the factors in the
denominator are those you would derive by replacing r with (n+r) in (10).

Of course, it is critical that you determine the indicial equation correctly, else you will
propagate your errors all the way through the problem.

Finding the values of an

Now that we have a recursion relation, we can calculate the coefficients an. As noted
above, we have to do this separately for each value of r.

It is standard to start with the greater value of r.

When r= ½ :

− 7(n + r − 1) a n −1 − 7(n − 1 / 2)a n −1 − 7(2n − 1)a n −1

an = = = for n ≥ 1 (13)
[ 2( n + r ) − 1][(n + r ) + 3] [2(n + 1 / 2) − 1][n + 7 / 2] 2n( 2n + 7)

This recursion relation gives us the following values:

− 7a0 21a1 147

a1 = , a2 = − = a0 (14)
18 44 792

And we can write the first solution to the differential equation as:
 7 147 2
y1 = a 0 x1 / 2 (1 − x+ x + ...) (15)
18 792

When r = -3:

We go back to the original recursion relation (12) and begin as:

− 7( n + r − 1) a n −1 − 7( n − 4) a n −1 − 7( n − 4) a n −1
an = = = for n ≥ 1 (16)
[ 2( n + r ) − 1][(n + r ) + 3] [2( n − 3) − 1][ n] n( 2n − 7)

This recursion relation yields the following coefficients:

− 21 7a 49 − 7 a0 − 343
a1 = a0 , a 2 = − 1 = a 0 , a3 = = a0 (17)
5 3 5 3 15

And the solution corresponding to this value of r is then given by:

21 49 2 343 3
y 2 = a 0 x −3 (1 − x+ x − x ...) (18)
5 5 15

The general solution to the differential equation

We have in essence found the complete solution to our original equation in (1). We then
write the general solution as:

7 147 2 21 49 2 343 3
y = c1 y1 + c 2 y 2 = c1 x 1 / 2 (1 − x+ x + ...) + c 2 x −3 (1 − x + x − x ) (19)
18 792 5 5 15

There is not much new in (19); the only point to make is to take explicit note of the
coefficients. Above we calculated coefficients in the two cases (i.e., where r = ½ and
when r = -3) and called the coefficients in both cases an as is our custom. In writing the
final solution (19), we just have to remember that y1 and y2 are independent solutions to
the original differential equation, so we need to keep in mind that the coefficients c1 and
c2 are likely to be different.

Summary of the technique

We can use the method of Frobenius whenever we have a differential equation that meets
the conditions described above.

Once you have verified that the method will apply to the equation under scrutiny:
∞ ∞
1) Assume a solution of the form y = x r ∑ a n x n = ∑ a n x n + r
n=0 n =0

2) Substitute into equation

3) Re-index as needed so that all summations contain xn+r

4) Derive the indicial equation by setting n = 0; this step tells you the needed values of r
in order that the first non zero term be the a0 term.

5) Derive recursion relation for equation.

6) Sequentially substitute values of r into recursion relation to determine coefficients for

each solution to the differential equation

∞
7) Substitute appropriate values of r and an into y = x r ∑ a n x n for each value of r.
n =0
8) Write the general solution of the differential equation.

The Second Solution in Frobenius’ Method—Three Cases

When we began discussion of Frobenius’ method, we said that this technique provides at
least one solution to a differential equation that meets the conditions that (x-x0) P(x) and
(x-x0)2 Q(x) are well defined in the neighborhood of x0. In the section immediately above,
we glossed over any nuance implied by this statement and found both solutions to the
differential equation.

There are actually three different possible cases that arise in using Frobenius’ method,
and the nature of the second solution to the equation is dependent upon which case we
encounter.

The method of Frobenius will always produce a solution for the larger value of r. The
second solution depends on:

CASE I:

If the two roots of the indicial equation, r1 and r2 are such that the quantity (r1 – r2) is not
an integer, then the second solution to the differential equation is found in exactly the
manner we have just used.

In short, we can find both solutions to the differential equation from:

y1 = x r1 ∑ a n x n and y 2 = x r2 ∑ bn x n (20)

where r1 and r2 are the solutions to the indicial equation, and I call the coefficients an and
bn just to remind us that the sets of coefficients produced for the two summations are
almost certain to be different from each other.
CASE II:

If the two roots of the indicial equation are equal, i.e., r1 = r2, then the method of
Frobenius will only provide one solution to the differential equation.

The second solution will be of the form:

x
exp[− ∫ P( x ′)dx ′]
y 2 = y1 ∫ dx (21)
( y12 )

Where P(x) is as defined earlier, y1 is the solution of the form of eq. (4), and y2 is the
second solution to the differential equation.

We will show examples of how to use (21) to solve equations a little later in the
document.

CASE III:

If the two roots of the indicial equation are different and are such that (r1 – r2) is an
integer, then the method of Frobenius may or may not produce the second solution to the
differential equation. (Don’t you love this case…)

The obvious question is how you can determine whether you the method of Frobenius
will produce the second solution or not. The way to proceed is to find the roots of the
indicial equation and the recursion relation. Use these data to find the solution for the
larger root of the indicial equation.

Then, substitute the value of the lesser root of the indicial equation into the recursion
formula and see if it produces a complete set of coefficients. If it does, then the method
of Frobenius is valid for the second solution of the differential equation, and the second
∞
solution comes simply from y = x r ∑ a n x n where now r is the lesser root of the indicial
n =0
equation.

However, if the recursion relation produces one or more infinite results when using the r2
root, then the method of Frobenius will not yield the second solution, and you have to
apply eq. (21) to find that solution.

EXAMPLES

Below is an example showing of a CASE III equation which shows how to use eq. (21) to
find the second solution.

Let’s start with the differential equation:

 x ( x − 1) y ′′ + 3 xy ′ + y = 0 (22)

Check that this equation satisfies the conditions for using the Frobenius method. Here,
P(x)=3/(x-1) and Q(x)=1/x(x-1).
∞ ∞
Substituting y = x r ∑ a n x n = ∑ a n x n + r into (22) yields:
n=0 n =0

y = ∑ (n + r )(n + r − 1)a n x n + r − ∑ (n + r )(n + r − 1)a n x n + r −1 + 3∑ (n + r )a n x n + r + ∑ a n x n + r = 0

n =0 n=0 n=0 n =0
(23)

The indicial equation:

We use eq. (23) to find the indicial equation. As described before, the indicial equation
provides us with the values of r that will solve the differential equation and will ensure
that the first non zero term of the power series solution is the a0 term. We can find the
indicial equation by setting n equal to zero in those terms of (23) bearing the lowest
power of x. In this case, there is only one such term: the second one. (The exponent in
the second term is n+r-1, in all other terms the exponent is n+r.)

Setting n=0 in the second term produces the indicial equation:

r ( r − 1) = 0 (24)

for which the roots are easily seen to be r = 0, 1. This is an example of CASE III, where
the difference in the roots of the indicial equation is an integer.

The recursion relation

We know that the method of Frobenius will produce at least one solution for the original
equation (22). To produce the recursion relation we need to re-index only the second
summation. Doing so in the typical way, we can proceed by writing the recursion
relation:

(n + r + 1)(n + r + 1) (n + r + 1)
a n +1 = an = an (25)
(n + r )(n + r + 1) (n + r )

For the greater root, we have that:

(n + r + 1) (n + 2)
a n +1 = an = an (26)
(n + r ) (n + 1)

The first solution

By now we are very adept at using recursion relations to find coefficients, and doing so
allows us to write:
 3 4
a1 = 2a 0 ; a 2 = a1 = 3a 0 ; a3 = a 2 = 4a 0 (27)
2 3

which leads immediately to:

∞
y1 = x r ∑ a n x n = a 0 x1 (1 + 2 x + 3 x 2 + 4 x 3 + ...nx n −1 ) (28)
n =0

This is the first solution to the differential equation. Now we have to test whether
Frobenius will provide us with the second solution.

The second solution

Since (r1-r2) is an integer (with unequal roots), we do not know a priori whether the
method of Frobenius will admit both or only one solution. To find out, we return to the
recursion relation (25) and now substitute the lesser root of the indicial equation.
Substituting r2 =0 into (25) yields:

(n + r + 1)(n + r + 1) (n + r + 1) (n + 1)
a n +1 = an = an = an (29)
(n + r )(n + r + 1) (n + r ) n

We run into trouble immediately in trying to compute the a1 term. We find the a1 term by
setting n = 0 in (25). This is clearly not a valid operation, since we know the solution
must converge in some neighborhood around x0=0, and an infinite term for a1 is clearly
impossible. This is an example where we cannot use Frobenius for the second solution,
and we must try our hand at the technique suggested in eq. (21).

This equation requires that we know the first solution (which we have in eq. (28)), and
also know the expression for P(x), which in this case is 3/(x – 1). Recalling (21):

exp[− ∫ P ( x ′)dx ′]
y 2 = y1 ∫ dx
y12
We start by computing the integral in the numerator:

3
exp[− ∫ dx ′] = exp[−3 ln( x − 1)] = exp[ln( x − 1) −3 ] = ( x − 1) −3 (30)
x′ − 1

Now, our next step is to substitute the square of the first solution into the denominator of
the integral above and integrate to find the second solution. The difficulty of this
depends on how nicely the y1 solution adapts to integration.

We could take eq. (28) as written, expand 1/y2 as a power series, multiply by (x-1)-3 and
integrate term by term; and in some cases this is what we have to do. However, in some
cases we can study the power series expansion for y1 and recognize it as the expansion of
some well known function. Let’s try this for our solution (28):

∞
y1 = x r ∑ a n x n = a 0 x 1 (1 + 2 x + 3 x 2 + 4 x 3 + ...nx n −1 )
n=0

The series expansion portion of this looks a little familiar; it might remind you of the
expansion:

1
1− x
= 1 + x + x 2 + x 3 + ... = ∑x n
, x <1 (31)

This is close, but not exact, since our solution in (28) has the form ∑ nx n −1
. But we

immediately recognize nxn-1 as the result of differentiating xn, so we differentiate both

sides of (31) and find:

d ⎛ 1 ⎞ 1
⎜ ⎟= = 1 + 2 x + 3x 2 + ... = ∑ nx n −1 (32)
dx ⎝ 1 − x ⎠ (1 − x) 2

x
With this realization, we can write our solution y1 as . Now, we go back to our
(1 − x) 2
equation (21), and apply it to this case and find:

( x − 1) −3 ( x − 1) −3 (1 − x) 4 1− x
y 2 ( x) = y1 ( x) ∫ 2
dx = y1 ( x) ∫ 2
dx = − y1 ( x) ∫ 2 dx (33)
⎛ x ⎞ x x
⎜⎜ ⎟
2 ⎟
⎝ (1 − x) ⎠
The minus sign is introduced in the last step of (33) by noting (x-1)-3 = [(-1)(1-x)]-3

=(-1)-3(1-x)-3.= (-1)(1-x)-3.

Now, the integral in (33) is not so bad, and we can evaluate it almost by inspection, so
that finally we have our second solution:

−1 ⎡1 + x ln x ⎤ x ⎡1 + x ln x ⎤ 1 + x ln x
y 2 ( x) = − y1 ( x)[ − ln x] = y1 ( x) ⎢ ⎥ = 2 ⎢ ⎥ = (1 − x) 2 (34)
x ⎣ x ⎦ (1 − x) ⎣ x ⎦

where we make explicit use of the analytic form of the solution y1(x).

In most cases, we will not have such a nice expression for the first solution to plug into
(21), and we will have to expand 1/y2 in a power series inside the integral and then
integrate term by term.
I will tend to give you problems in which you can use Frobenius for both solutions; if a
problem I give you has only one solution, then I will likely specify to provide one
solution.

One more example….

I refer you to Boas, pp. 585-6. She introduces the method of Frobenius with the
differential equation:

x 2 y ′′ + 4 xy ′ + ( x 2 + 2) y = 0

If you look at this equation term by term, you should be able to see that there will be four
terms in this summation, and that three of them (the first, second and last) will all involve
powers of xn+r; the penultimate sum will have xn+r+2. The indicial equation is then:

r (r − 1) + 4r + 2 = r 2 + 3r + 2 = (r + 2)(r + 1) = 0 ⇒ r = −1, − 2

This is an example of CASE III, since the difference of the roots is an integer. The
recursion relation for this equation is:

− an−2
an = for n ≥ 2
(n + r + 1)(n + r + 2)

We know the method of Frobenius will provide one solution with r set to the larger root
of the indicial equation; substituting r = -1, we get the recursion relation:

− a n−2
an = for n ≥ 2
n(n + 1)

We test whether Frobenius can give us the second solution by substituting r = -2:

− a n−2
an = for n ≥ 2
n(n − 1)

Since we begin our evaluation of an at n = 2, this final recursion relation will yield valid
values for an (since the denominator is never zero for n ≥ 2 .) This case is an example of
a CASE III equation where the method of Frobenius will yield both solutions to the
differential equation.

And…one last example…

We consider finally an example of CASE II which will also allow us to see how to find
the second solution if we cannot easily write the first solution in a simple closed analytic
form.

Consider the differential equation:

 x 2 y ′′ + xy ′ − xy = 0

The indicial equation for this equation becomes:

r ( r − 1) + r = 0 ⇒ r = 0, 0

Since the two roots of the indicial equation are zero, we know we can find one solution of
the form:

y1 ( x) = x 0 ∑ a n x n

The exponent of x arises from r = 0¸ and we have to find the values of an from the
recursion relation:

a n −1 a
an = = n2−1 when r = 0
(n + r ) 2
n

This recursion relation leads to a first solution of:

x2 x3
y1 ( x) = a0 (1 + x + + + ...)
4 36

To find the second solution, we make use of eq. (21) with P(x) = 1/x, so we have that:

1 1
exp[− ∫ P ( x)dx] = exp[− ∫ dx] = exp[ − ln x] =
x x

and eq. (21) becomes:

⎡ 1/ x ⎤
y 2 ( x) = y1 ( x) ⎢ ∫ 2 dx ⎥
⎣ a 0 (1 + x + x / 4 + x / 36 + ...)
2 3 2
⎦

This looks pretty ugly.

We begin by expanding 1/(1+x+x2/4+x3/36+…)2 in a power series:

We recall that:

1
= 1 − 2 x + 3 x 2 − 4 x 3 ...
(1 + x) 2

So that we can expand:

1 1 5 23 3
= = 1 − 2x + x 2 − x + ...
( y1 ( x)) 2
(1 + x + x / 4 + x / 36 + ..)
2 3 2
2 9
With this expansion, our second solution becomes:

⎡ 1 − 2 x + 5 x 2 / 2 − 23x 3 / 9... ⎤
y 2 ( x) = y1 ( x) ⎢ ∫ dx ⎥ = y1 ( x)[ln x − 2 x + 5 x 2 / 4 − 23x 3 / 27...]
⎣ x ⎦

= y1 ( x ) ln x + a 0 ( −2 x − 3 x 2 / 4 − 11x 3 / 108 − ...)

and this is the end….for now.