13

Power Series and Differential Equations:

The Method of Frobenius

Its all well and good to be able to find power series representations for functions you know
via the standard computations for Taylor series. Even better is to be able to find power series
representations for functions you dont know, say, solutions to certain nasty differential equations
that cannot be solved by elementary methods. That is where the method of Frobenius is useful.
It is a procedure for finding power series formulas for solutions to a great many differential
equations of interest in mathematical physics. Actually, these formulas are often not true power
series, but modified power series. To be precise, the series obtained are power series multiplied
by relatively simple functions.
The basic method is described and illustrated in the first section. This will give you the tools
to start practicing the method. After than, we will discuss when the method is applicable and
what sort of solutions, in general, the method yields.

13.1 The Basic Method of Frobenius (Illustrated)

Preliminaries
Let us assume we have a differential equation of the form
d2 y dy
a(x) + b(x) + c(x)y = 0
dx 2 dx
where a , b , and c are suitable functions. Just what suitable means will be discussed later.
For now, we will assume they are suitable rational functions (i.e., polynomials divided by
polynomials) such as in Bessels equation of order 1/2
d2 y 1 dy
h 1
i
2
+ + 1 2
y = 0 . (13.1)
dx x dx 4x
This is the equation we will solve to illustrate the method.
The basic method of Frobenius seeks solutions of the form of power series about some given
point x0 multiplied by x x0 to some power. That is, it seeks solutions of the form

X
y(x) = (x x0 )r ak (x x0 )k
k=0
where r and the ak s are constants to be determined by the method.

2/3/2010
The Method of Frobenius Chapter & Page: 132

The Basic Steps

Before actually starting the method, there are two pre-steps:

Pre-step 1 Choose a value for x0 . If conditions are given for y(x) at some point, then use that
point for x0 . Otherwise, choose x0 as convenient which usually means you choose
x0 = 0 .
For our example, we have no initial values at any point, so we will choose x0
as simply as possibly; namely, x0 = 0 .

Pre-step 2 Get the differential equation into the form

d2 y dy
A(x) + B(x) + C(x)y = 0
dx 2 dx
where A , B , and C are polynomials.
To get the given differential equation (equation (13.1)) into the form desired,
we multiply the equation by 4x 2 . That gives us the differential equation

d2 y dy
4x 2 2
+ 4x + [4x 2 1]y = 0 . (13.2)
dx dx

(Yes, we could have just multiplied by x 2 , but getting rid of any fractions will
simplify computation.)

Now for the basic method of Frobenius:

Step 1: (a) Start by assuming a solution of the form

X
r
y = y(x) = (x x0 ) ak (x x0 )k (13.3a)
k=0

where a0 is an arbitrary constant. Since it is arbitrary, we can and will assume

a0 6= 0 in the following computations.1
(b) Then simplify the formula for the following computations by bringing the (x x0 )r
factor into the summation,

X
y = y(x) = ak (x x0 )k+r . (13.3b)
k=0

(c) And then compute the corresponding modified power series for y and y from
the assumed series for y by differentiating term-by-term.
Since weve already decided x0 = 0 , we assume

X
X
r k
y = y(x) = x ak x = ak x k+r (13.4)
k=0 k=0

1 Insisting that a 6 = 0 helps limit the possible values of r .

0

version: 2/3/2010
The Method of Frobenius Chapter & Page: 133

with a0 6= 0 . Differentiating this term-by-term, we see that

d X
y =
ak x k+r
dx
k=0

X d   X
= ak x k+r = ak (k + r )x k+r 1 ,
dx
k=0 k=0

d X
y
= ak (k + r )x k+r 1
dx
k=0

X d   X
= ak (k + r )x k+r 1 = ak (k + r )(k + r 1)x k+r 2
dx
k=0 k=0

Step 2: Plug these series for y , y , and y back into the differential equation, multiply things
out, and divide out the x r to get the left side of your equation in the form of the sum of
a few power series.
Some Notes:

ii. Absorb any xs in A , B and C (of the differential equation) into the series.

iii. Dividing out the x r isnt necessary, but it simplifies the expressions slightly and
reduces the chances of silly errors later.

iv. You may want to turn your paper sideways for more room!

Combining the above series formulas for y , y and y with our differential
equation (equation (13.2)), we get

0 = 4x 2 y + 4x y + [4x 2 1]y

X
X
= 4x 2 ak (k + r )(k + r 1)x k+r 2 + 4x ak (k + r )x k+r 1
k=0 k=0

X
+ [4x 2 1] ak x k+r
k=0

X
X
2 k+r 2
= 4x ak (k + r )(k + r 1)x + 4x ak (k + r )x k+r 1
k=0 k=0

X
X
+ 4x 2 ak x k+r 1 ak x k+r
k=0 k=0

X
X
= ak 4(k + r )(k + r 1)x k+r + ak 4(k + r )x k+r
k=0 k=0

X
X
+ ak 4x k+2+r + ak (1)x k+r .
k=0 k=0

version: 2/3/2010
The Method of Frobenius Chapter & Page: 134

Dividing out the x r from each term then yields

X
X
k
0 = ak 4(k + r )(k + r 1)x + ak 4(k + r )x k
k=0 k=0

X
X
+ ak 4x k+2 + ak (1)x k .
k=0 k=0

Step 3: For each series in your last equation, do a change of index so that each series looks like

X  
something not involving x (x x0 )n .
n=something

Be sure to appropriately adjust the lower limit in each series.

In all but the third series in the example, the change of index is trivial,
n = k . In the third series, we will set n = k + 2 (equivalently, n 2 = k ).
This means, in the third series, replacing k with n 2 , and replacing k = 0
with n = 0 + 2 = 2 :

X
X
k
0 = ak 4(k + r )(k + r 1)x + ak 4(k + r )x k
k=0 k=0
| {z } | {z }
n=k n=k

X
X
+ ak 4x k+2 + ak (1)x k
|k=0 {z } |k=0 {z }
n=k+2 n=k

X
X
= an 4(n + r )(n + r 1)x n + an 4(n + r )x n
n=0 n=0

X
X
+ an2 4x n + an (1)x n .
n=2 n=0

Step 4: Convert the sum of series in your last equation into one big series. The first few terms
will probably have to be written separately. Simplify what can be simplified.
Since one of the series in the last equation begins with n = 2 , we need to
separate out the terms corresponding to n = 0 and n = 1 in the other series
before combining series:

X
X
n
0 = an 4(n + r )(n + r 1)x + an 4(n + r )x n
n=0 n=0

X
X
+ an2 4x n + an (1)x n
n=2 n=0

version: 2/3/2010
The Method of Frobenius Chapter & Page: 135


= a0 4(0 + r )(0 + r 1) x 0 + a1 4(1 + r )(1 + r 1) x 1
| {z } | {z }
4r (r 1) 4(1+r )r

X 
+ an 4(n + r )(n + r 1)x n
n=2

X 
0 1 n
+ a0 4(0 + r ) x + a1 4(1 + r ) x + an 4(n + r )x
| {z } | {z }
4r 4(1+r ) n=0

X 
X 
n 0 1 n
+ an2 4x + a0 x a1 x + an (1)x
n=2 n=2

   
= a0 4r (r 1) + 4r 1 x 0 + a1 4(1 + r )r + 4(1 + r ) 1 x 1
| {z } | {z }
4r 2 4r +4r 1 4r +4r 2 +4+4r 1

X  
+ an 4(n + r )(n + r 1) + an 4(n + r ) + an2 4 + an (1) x n
| {z }
n=2 an [4(n+r )(n+r 1)+4(n+r )1]+4an2

   
= a0 4r 2 1 x 0 + a1 4r 2 + 8r + 3 x 1
h
X i
 
+ an 4(n + r )(n + r ) 1 + 4an2 x n .
n=2

So our differential equation reduces to

   
a0 4r 2 1 x 0 + a1 4r 2 + 8r + 3 x 1
h
X i
 
+ an 4(n + r )2 1 + 4an2 x n = 0 . (13.5)
n=2

Observe that the end result of this step will be an equation of the form

some big power series = 0 .

This, in turn, tells us that each term that big power series must be 0 .

Step 5: The first term in the last equation just derived will be of the form
 
a0 formula of r (x x0 )something .

But, remember, each term in that series must be 0 . So we must have

 
a0 formula of r = 0 .

Moreover, since a0 6= 0 (by assumption), the above must reduce to

formula of r = 0 .

version: 2/3/2010
The Method of Frobenius Chapter & Page: 136

This is the indicial equation for r . It will always be a quadratic equation for r (i.e., of
the form r 2 + r + = 0 ). Solve this equation for r . You will get two solutions
(sometimes called called either the exponents of the solution or the exponents of the
singularity). Denote them by r2 and r1 with r2 r1 .2
In our example, the first term in the big series is the first term in equation
(13.5),  
a0 4r 2 1 x 0 .
Since this must be zero (and a0 6= 0 by assumption) the indicial equation is

4r 2 1 = 0 .

Thus, r
1 1
r = = .
4 2
Following the convention given,
1 1
r2 = and r1 = .
2 2

Step 6: Using r1 , the larger r just found:

(a) Plug r1 into the last series equation (and simplify, if possible). This will give you
an equation of the form

X  
n th formula of ak s (x x0 )n = 0 .
n=n 0

Since each term must vanish, we have

n th formula of ak s = 0 for n 0 n .

(b) Solve this for

ahighest index = formula of n and lower indexed ak s .

A few of these equations may need to be treated separately, but you will also
obtain a relatively simple formula that holds for all indices above some fixed value.
This formula is the recursion formula for computing each coefficient an from the
previously computed coefficients.
(c) To simplify things just a little, do another change of indices so that the recursion
formula just derived is rewritten as

ak = formula of k and lower-indexed coefficients .

2 We are assuming r and r are real numbers. This is the standard situation. Well briefly discuss the cases where
1 2
r1 and r2 are complex numbers later.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 137

Letting r = r1 = 1/2 in equation (13.5) yields

   
0 = a0 4r 2 1 x 0 + a1 4r 2 + 8r + 3 x 1
h
X i
 
+ an 4(n + r )2 1 + 4an2 x n
n=2

         
1 2 0 1 2 1
= a0 4 1 x + a1 4 +8 + 3 x1
2 2 2

     
X 1 2
+ an 4 n + 1 + 4an2 x n
2
n=2

X   2  
= a0 0x 0 + a1 8x 1 + an 4n + 4n + 1 1 + 4an2 x n .
n=2

The first term vanishes (as it should since r = 1/2 satisfies the indicial equation,
which came from making the first term vanish). Doing a little more simple
algebra, we see that, with r = 1/2 , equation (13.5) reduces to

X  
0a0 x 0 + 8a1 x 1 + 4 n(n + 1)an + an2 x n = 0 . (13.6)
n=2

From the above series, we must have

n(n + 1)an + an2 = 0 for n = 2, 3, 4 . . . .

Solving for an leads to the recursion formula

1
an = an2 for n = 2, 3, 4 . . . .
n(n + 1)

Using the trivial change of index, k = n , this is

1
ak = ak2 for k = 2, 3, 4 . . . . (13.7)
k(k + 1)

Step 7: Use the recursion formula (and any corresponding formulas for the lower-order terms)
to find all the ak s in terms of a0 and, possibly, one other am . Look for patterns!
From the first two terms in equation (13.6),

0a0 = 0 H a0 is arbitrary.

8a1 = 0 H a1 = 0 .

Using these values and the recursion formula (equation (13.7)) with k =
2, 3, 4, . . . (and looking for patterns):
1 1
a2 = a22 = a0 ,
2(2 + 1) 23

version: 2/3/2010
The Method of Frobenius Chapter & Page: 138

1 1 1
a3 = a32 = a1 = 0 = 0 ,
3(3 + 1) 34 34

1 1 1 1 (1)2 (1)2
a4 = a42 = a2 = a0 = a0 = a0 ,
4(4 + 1) 45 45 23 5432 5!
1 1
a5 = a52 = 0 = 0 ,
5(5 + 1) 56

1 1 1 (1)2 (1)3
a6 = a62 = a4 = a0 = a0 ,
6(6 + 1) 67 76 5! 7!
..
.
The patterns should be obvious here:
ak = 0 for k = 1, 3, 5, 7, . . . ,
and
(1)k/2
ak = a0 for k = 2, 4, 6, 8, . . . .
(k + 1)!
Using k = 2m , this be written more conveniently as
a0
a2m = (1)m for m = 1, 2, 3, 4, . . . .
(2m + 1)!
Moreover, this last equation reduces to the trivially true statement a0 = a0
if m = 0 . So, in fact, it gives all the even-indexed coefficients,
a0
a2m = (1)m for m = 0, 1, 2, 3, 4, . . . .
(2m + 1)!

Step 8: Using r = r1 along with the formulas just derived for the coefficients, write out the
resulting series for y . Try to simplify it and factor out the arbitrary constant(s).
Plugging r = 1/2 and the formulas just derived for the an s into the formula
originally assumed for y (equation (13.4) on page 132), we get

X
y = xr ak x k
k=0

X
X 
r k k
= x ak x + ak x
k=0 k=0
k odd k even

X 
1/ X a0
= x 2 0 xk + (1)m x 2m
(2m + 1)!
k=0 m=0
k odd
 
1 X 1
= x /2 0 + a0 (1)m x 2m .
(2m + 1)!
m=0
1/
So one solution to Bessels equation of order 2 (equation (13.1) on page
131) is given by

1 X (1)m
y = a0 x /2 x 2m . (13.8)
(2m + 1)!
m=0

version: 2/3/2010
The Method of Frobenius Chapter & Page: 139

Step 9: If the indical equation had two distinct solutions, now repeat steps 6 through 8 with
the smaller r , r2 . Sometimes (but not always) this will give you a second independent
solution to the differential equation. Sometimes, also, the series formula derived in this
mega-step will include the series formula already derived.
Letting r = r2 = 1/2 in equation (13.5) yields
   
0 = a0 4r 2 1 x 0 + a1 4r 2 + 8r + 3 x 1
h
X i
 
+ an 4(n + r )2 1 + 4an2 x n
n=2

         
1 2 0 1 2 1
= a0 4 1 x + a1 4 8 + 3 x1
2 2 2

     
X 1 2
+ an 4 n 1 + 4an2 x n
2
n=2

X
0
  2
1
 
= a0 0x + a1 0x + an 4n 4n + 1 1 + 4an2 x n
n=2

Then
..
.

yielding

1 X (1)m 1 X (1)m
y = a0 x /2 x 2m + a1 x /2 x 2m+1 . (13.9)
(2m)! (2m + 1)!
m=0 m=0

Note that the second series term is the same series (slightly rewritten) as in
1 1
equation (13.8) (since x /2 x 2m+1 = x /2 x 2m ).

?Exercise 13.1: Fill in the dots in the last statement. That is, do all the computations
that were omitted.

Step 10 If the last step yielded y as an arbitrary linear combination of two different series, then
that is the general solution to the original differential equation. If the last step yielded
y as just one arbitrary constant times a series, then the general solution to the original
differential equation is the linear combination of the two series obtained at the end of
steps 8 and 9. Either way, write down the general solution (using different symbols for
the two different arbitrary constants!). If step 9 did not yield a new series solution, then
at least write down the one solution previously derived, noting that a second solution is
still needed for the general solution to the differential equation.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1310

We are in luck. In the last step we obtained y as the linear combination of

two different series (equation (13.9)). So

1 X (1)m 1 X (1)m
y = a0 x /2 x 2m + a1 x /2 x 2m+1 ,
(2m)! (2m + 1)!
m=0 m=0

is the general solution to our original differential equation (equation (13.1)

Bessels equation of order 1/2 ).

Last Step See if you recognize the series as the series for some well-known function (you
probably wont!).
Our luck continues! The two series are easily recognized as the series for the
sine and the cosine functions:

1/2
X (1)m 2m 1/2
X (1)m
y = a0 x x + a1 x x 2m+1
(2m)! (2m + 1)!
m=0 m=0
1 1
= a0 x /2 cos(x) + a1 x /2 sin(x) .
1/
So the general solution to Bessels equation of order 2 is
cos(x) sin(x)
y = a0 + a1 . (13.10)
x x

13.2 Practical Advice and Possibly Helpful Comments

1. If you get something like
2a1 = 0 ,
then you know a1 = 0 . On the other hand, if you get something like

0a1 = 0 ,

then you have an equation that tells you nothing about a1 . This means that a1 is an
arbitrary constant (unless something else tells you otherwise).

2. If the recursion formula blows up at some point, then some of the coefficients must be
zero. For example, if
3
an = an2 ,
(n + 2)(n 5)
then, for n = 5 ,
3
a5 = a3 = a3 ,
(7)(0)
which can only make sense if a3 = 0 . Note also, that, unless otherwise indicated, a5
here would be arbitrary. (Remember, the last equation is equivalent to (7)(0)a5 = 3a3 .)

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1311

3. If you get a coefficient being zero, it is a good idea to check back using the recursion
formula to see if any of the previous coefficients must also be zero, or if many of the fol-
lowing coefficients are zero. In some cases, you may find that an infinite series solution
only contains a finite number of nonzero terms, in which case we have a terminating
series; i.e., a solution which is simply a polynomial.

4. It is possible to end up with a three term recursion formula, say,

1 2
an = an1 + an2 .
n2 + 1 3n(n + 3)

This, naturally, makes finding patterns rather difficult.

5. Keep in mind that, even if you find that finding patterns and describing them by nice
formulas is beyond you, you can always use the recursion formulas to compute (or have
a computer compute) as many terms as you wish of the series solutions.

6. The computations can become especially messy and confusing when x0 = 0 . In this
case, simplify matters by using the substitutions

Y (X) = y(x) with X = x x0 .

You can then easily verify that, under these substitutions,

dY dy
Y (X) = = = y (x)
dX dx
and the differential equation

d2 y dy
A(x) + B(x) + C(x)y = 0 (13.11)
dx 2 dx
becomes
d 2Y dY
A1 (X) + B1 (X) + C1 (X)Y = 0 (13.11 )
d X2 dX
with

A1 (X) = A(X + x0 ) , B1 (X) = B(X + x0 ) and C1 (X) = C(X + x0 ) .

Use the method of Frobenious to find the modified power series solutions

X
Y (X) = X r ak X k
k=0

for equation (13.11 ). The corresponding solutions to the original differential equation,
equation (13.11), are then given from this via the above substitution,

X
X
r k r
y(x) = Y (X) = X ak X = (x x0 ) ak (x x0 )k .
k=0 k=0

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1312

13.3 Theory Behind the Method (Without Proofs)

Basic Theory and Ordinary Points
Let us consider any differential equation of the form

d2 y dy
a(x) + b(x) + c(x)y = 0 (13.12)
dx 2 dx
where a , b and c are known functions. Remember that the general solution over some interval
(, ) to such a differential equation is given by

y(x) = c1 y1 (x) + c2 y2 (x)

where y1 and y2 is a linearly independent pair of particular solutions (no arbitrary constants)
over (, ) , and c1 and c2 are two arbitrary constants.
For describing theory (though rarely for simplifying computations), it helps to divide the
differential equation by a(x) , obtaining the equivalent equation

d2 y dy
+ P(x) + Q(x)y = 0 (13.12 )
dx 2 dx

(with P = b/a and Q = c/a ). Now, for each point z 0 in the complex plane, we say that:

1. z 0 is an ordinary point for the differential equation if and only if both P(z) and Q(z)
(viewed as functions of a complex variable) are analytic at z 0 .3

2. z 0 is a singular point for the differential equation if and only if it is not an ordinary point
for the equation.

Keep in mind that P = b/a and Q = c/a . So if a(z) vanishes at some point z 0 where either
b(z) or c(z) is nonzero, then that point, z 0 , will be a singular point for the differential equation.
The importance of whether a point is ordinary or singular is indicated in the following
theorem.

Theorem 13.1 (Big Theorem on Solving DEs about Ordinary Points)

Let x0 be an ordinary point (on the real line) for the differential equation

d2 y dy
a(x) 2
+ b(x) + c(x)y = 0 .
dx dx
Then a general solution
y(x) = c1 y1 (x) + c2 y2 (x)
exists on some interval containing x0 , with y1 and y2 being analytic at x0 . These functions,
y1 and y2 , can be obtained by the basic method of Frobenius. Moreover:

1. The indicial equation from the Frobenius method is r (r 1) = 0 . (Hence r1 = 1 and

r2 = 0 .)
3 That is, P(z) and Q(z) can each be described as a power series about z over some neighborhood of that point.
0
(We are anticipating some stuff from our study of complex variables.)

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1313

and
2. The series formulas for both y1 and y2 obtained by the method of Frobenius are valid
at least on the interval (x0 R, x0 + R) where R is the distance from x0 to the closest
point on the complex plane at which either P or Q is not analytic.

For the proof of this theorem, see a good intermediate or higher level text on differential
equations.4 We wont attempt it here. Our interest is simply that it assures us that someone
has confirmed that the basic Frobenius method works when we attempt it using power series
about ordinary points, and has even given us an idea as to the interval over which the series will
converge. It even states that the series will be true power series.
Keep in mind that, while the theorem describes cases where the method of Frobenius works,
it does not say that it is always the best method. For example, it assures you that you can find
infinite series solutions to
d2 y dy
a + b + cy = 0
dx 2 dx
when a , b and c are constants but why bother? It would be much easier to simply use
a method described in section A.3 of the appendix reviewing elementary ordinary differential
equations.
In fact, whenever x0 is an ordinary point for the differential equation, using the full Frobenius
method is excessive. With a little thought, youll realize that the above theorem says that you can
skip the whole indicial equation part of the Frobenius method. Instead, just use the relevant
parts of the method after assuming

X
y(x) = ak (x x0 )k
k=0

with a0 and a1 being arbitrary constants.

More Advanced Theory

If you go back and check our example in the section 13.1, you will find that our series expansion
was not done about an ordinary point for the differential equation. The point x0 = 0 turns out
to be a singular point for Bessels equation of order 1/2 . Indeed, mathematical physics is full
of situations where series expansions are attempted about singular points. Fortunately, most of
these singular points are not badly singular.
Time for more terminology:
Suppose z 0 is a singular point for
d2 y dy
a(x) 2
+ b(x) + c(x)y = 0 .
dx dx
Divide through by a(x) to get
d2 y dy
2
+ P(x) + Q(x)y = 0
dx dx
(with P = b/a and Q = c/
a ). We say that z 0 is a regular singular point for the differential
equation if and only if
p(z) q(z)
P(z) = and Q(z) =
z z0 (z z 0 )2
4 say, Differential Equations: Theory and Applications by R. Redheffer, Jones & Bartlett Publishers, 1991.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1314

where p(z) and q(z) are analytic functions at z 0 . In other words, a singular point z 0 is a
regular singular point for the differential equation if and only if the differential equation can be
rewritten as
d2 y p(x) d y q(x)
+ + = 0
dx 2 x z 0 dx (x z 0 )2
or, equivalently, as
d2 y dy
(x z 0 )2 2
+ (x z 0 ) p(x) + q(x) = 0
dx dx
for some functions p(x) and q(x) analytic at x0 .
If you go back and check, you will find that x0 = 0 is a regular singular point for Bessels
equation of order 1/2 . As the next theorem shows, regular singular points are not that badly
singular.

Theorem 13.2 (The Big Theorem on the Frobenius Method)

Let x0 be a regular singular point (on the real line) for
d2 y dy
a(x) 2
+ b(x) + c(x)y = 0 .
dx dx
Then the indicial equation arising in the basic method of Frobenius exists and is a quadratic
equation with two solutions r1 and r2 (which may be one solution, repeated).5 If r2 and r1 are
real, assume r2 r1 . Then:
1. The basic method of Frobenius will yield at least one solution of the form

X
y1 (x) = (x x0 )r1 ak (x x0 )k
k=0

where a0 is the one and only arbitrary constant.

2. If r1 r2 is not an integer, then the basic method of Frobenius will yield a second
independent solution of the form

X
y2 (x) = (x x0 )r2 ak (x x0 )k
k=0

where a0 is an arbitrary constant.

3. If r1 r2 = N is positive integer, then the method of Frobenius might yield a second
independent solution of the form

X
r2
y2 (x) = (x x0 ) ak (x x0 )k
k=0

where a0 is an arbitrary constant. If it doesnt, then a second independent solution exists

of the form

X
r2
y2 (x) = y1 (x) ln |x x0 | + (x x0 ) bk (x x0 )k
k=0

5 See the first note after this theorem regarding the use of the basic method when the coefficients are not rational
functions.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1315

or, equivalently,

X 
N k
y2 (x) = y1 (x) ln |x x0 | + (x x0 ) ck (x x0 )
k=0

where b0 and c0 are nonzero constants.

4. If r1 = r2 , then there is a second solution of the form

X
1+r1
y2 (x) = y1 (x) ln |x x0 | + (x x0 ) bk (x x0 )k
k=0
or, equivalently,

X 
y2 (x) = y1 (x) ln |x x0 | + (x x0 )1 ck (x x0 )k
k=0

where b0 , at least, is nonzero.

Moreover, if we let R be the distance between x0 and the nearest singular point (other than x0 )
in the complex plane (with R = if x0 is the only singular point), then the series solutions
described above converge at least on the intervals (x0 R, x0 ) and (x0 , x0 + R) .6

Again, for the proof, go to some decent intermediate differential equation text.7
Some notes regarding this theorem:

1. In first describing the method of Frobenious to solve

d2 y dy
a(x) + b(x) + c(x)y = 0
ds 2 dx
we assumed a , b and c were rational functions so that we could get the differential
equation into the form
d2 y dy
A(x) 2
+ B(x) + C(x)y = 0
ds dx
where A(x) , B(x) and C(x) are polynomials (Pre-step 2). More generally, you want to
get the equation into the above form where A(x) , B(x) and C(x) are analytic functions
about x0 , and then express these functions as power series about x0 . After all, a power
series is just a really big polynomial, isnt it?
Fortunately, it seems safe to say that the use of the Frobenius method when a(x) ,
b(x) and c(x) are not rational functions is rare.

2. In case 3 (where r1 r2 is a positive integer), if the method of Frobenius does yield

a second solution, then you will end up rederiving the first one again as you derive the
second.

3. The formulas for y2 (x) involving ln |x x0 | can be derived via the method of reduction
of order for finding second solutions to second order differential equations (see your old
DE text, not A&W). This is of theoretical, not practical, significance. It would usually be
6 The power series parts will converge on the interval (x R, x + R) , indeed, on the disk of radius R in the
0 0
complex plane about x 0 . However, the (x x 0 )r will behave badly at x = x 0 if r is not a positive real number.
7 such as the one by Redheffer mentioned in a previous footnote.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1316

a mistake to attempt to actually compute these series using reduction of order. Instead,
just plug these series formulas into the differential equation and find recursion formulas
for the coefficients. At some point, you will probably have to write out the series formula
for y1 (x) and multiply two series together. It will not be pretty.

4. In practice, one major application of many of these series to determine the general behavior
of solutions near x0 . For example, the logarithmic singularity in some of the solutions
may allow us to reject them in certain applications.

5. Again, keep in mind that simpler methods might be available. For instance, you can use
the Frobenius method on Cauchy-Euler equations (see section A.3 of the appendix on
ordinary differential equations), but that would be silly.

6. Obvious variations of the Frobenius method can be applied to first-order and third-order
(and fourth-order, etc.) differential equations.

7. As long as we allow complex numbers, we will have only a little difficulty with expressions
of the form (x x0 )r when x x0 < 0 and, say, r = 1/2 . When complex numbers are
being avoided, it may be best to write your final answer using

|x x0 |r instead of (x x0 )r .

But we will be allowing complex numbers. In fact, x and x0 may be complex in future
work, in which case we will see that the above mentioned replacement of (x x0 )r with
|x x0 |r would be BAD.

Behavior of Solutions Near Regular Singular Points

Power Series Near Their Centers
Assume f is given by a power series about some point x0 ,

X
f (x) = ak (x x0 )k
k=0

where a0 6= 0 . Observe that

f (x) = a0 + a1 (x x0 ) + a2 (x x0 )2 + a3 (x x0 )3 + a4 (x x0 )4
 
= a0 + (x x0 ) a1 + a2 (x x0 ) + a3 (x x0 )2 + a4 (x x0 )3 +

X
= a0 + (x x0 ) ak+1 (x x0 )k .
k=0

Now, if x x0 , then x x0 is small, and

f (x) = a0 + something small a0 .

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1317

Behavior of the Solutions

Suppose weve used the method of Frobenius to solve
d2 y dy
a(x) + b(x) + c(x)y = 0
dx 2 dx
about a regular singular point x0 . Let r1 and r2 be the solutions to the indicial equation (we
will assume them to be real with r1 r2 ), and let y1 (x) and y2 (x) be the solutions as described
in our big theorem on the Frobenius method (theorem 13.2).
Lets look at those solutions when x x0 .
No matter what, we know that

X
y1 (x) = (x x0 )r1 ak (x x0 )k .
k=0

So, for x x0
y1 (x) (x x0 )r1 a0 .
In particular, then,

0 if r1 > 0


lim |y1 (x)| = lim |x x0 |r1 |a0 | = |a0 | if r1 = 0 .
xx 0 xx 0

+ if r1 < 0

Now assume r2 6= r1 . With luck, our second solution is similar to the first,

X
y2 (x) = (x x0 )r2 ak (x x0 )k .
k=0

So, for x x0 ,
y2 (x) a0 (x x0 )r2 .
Still, it is possible that our second solution is of the form

X 
N k
y2 (x) = y1 (x) ln |x x0 | + (x x0 ) ck (x x0 )
k=0

where N = r1 r2 is a positive integer and c0 6= 0 . Then, for x x0 ,

 
r1 N
y2 (x) a0 (x x0 ) ln |x x0 | + (x x0 ) c0 .

But, as you can easily verify yourself, |ln |x x0 || is much smaller than |x x0 |N when
x x0 . So, when x x0 ,
 
r1 N
y2 (x) a0 (x x0 ) ln |x x0 | + (x x0 ) c0

a0 (x x0 )r1 (x x0 )N c0
= A0 (x x0 )r1 N
= A0 (x x0 )r2 .

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1318

So, if r1 > r2 and x x0 , then

y2 (x) A0 (x x0 )r2

for some (nonzero) constant A0 . In particular, then,

0 if r2 > 0


lim |y2 (x)| = lim |x x0 |r2 |A0 | = |A0 | if r2 = 0 .
xx 0 xx 0

+ if r2 < 0

Finally, if r1 = r2 , then the second solution can be written as


X 
1 k
y2 (x) = y1 (x) ln |x x0 | + (x x0 ) ck (x x0 ) .
k=0

Since ln |x x0 | blows up as x 0 and the other terms in the brackets go to zero as x 0 ,

we have, for x x0 ,

y2 (x) a0 (x x0 )r1 [ln |x x0 | + 0] = a0 (x x0 )r1 ln |x x0 | .

Using LHpitals rule, you can then easily show that

(
0 if r1 > 0
lim |y2 (x)| = lim |a0 | |x x0 |r1 |ln |x x0 || = .
xx 0 xx 0 + if r1 0

More on the Indicial Equation and Complex Solutions

Again, assume x0 is a regular singular point for

d2 y dy
a(x) 2
+ b(x) + c(x)y = 0 ,
dx dx
and let p and q be the analytic functions given by
b(x) p(x) c(x) q(x)
= and = ,
a(x) x x0 a(x) (x x 0 )2

Its not hard to show that the indicial equation is actually given by

r (r 1) + p(x0 )r + q(x0 ) = 0 .

Knowing this, and solving for the r s early can save a little work in finding the series solutions,
especially if you can make use of the big theorem on the Frobenius method (Theorem 13.2).
In practice, the functions a , b , and c in the differential equation are always real valued,
and you expect to be able to obtain real-valued solutions to the differential equation. And in
almost all (if not all) examples of the Frobenius method in textbooks, the roots of the indicial
equation turn out to be real. In fact, few of the references Ive had time to check even mention the
possibility of the r s being complex, and the comments that I did find appear to be last minute
changes because someone said What about complex roots?8
8 In one case the changes were rather poorly considered.

version: 2/3/2010
The Method of Frobenius Chapter & Page: 1319

Still, even if we require a , b , and c to be real-valued functions, it is quite possible to

construct a differential equation whose indicial equation yields nontrivial complex roots r =
i . The result of using the Frobenius method on this differential equation will be that
described in case 2 of Theorem 13.2 because the difference between the r s would not be an
integer. In this case it would not matter which r you refer to as r1 . Moreover, if a , b , and c
are real-valued functions, then you can easily show that, if y1 is a complex-valued solution to
the differential equation, then so is its complex conjugate y2 = y1 and the two form a linearly
independent pair. Hence an arbitrary linear combination of y1 and its conjugate will serve as
a general solution to the differential equation. It then follows that the real and imaginary parts
of y1 are also a linearly independent pair of solutions. Hence, a general solution in terms of
real-valued functions can also be given by

y(x) = c1 Re[y1 (x)] + c2 Im[y2 (x)] .

What all this means is that, in theory, if r is complex and your first series solution obtained
via Frobenius has complex terms, then you need not find the other series solution. Just use the
real and imaginary parts of the series solution you already have. Just how practical this is depends
on how easy it is to extract those real and imaginary parts.
If you want, you can play with an equation whose indical equation has complex roots by
doing the next exercise.

?Exercise 13.2 (OPTIONAL): Consider the Cauchy-Euler equation

d2 y dy
x2 + x + 1 = 0 .
dx 2 dx

a: Solve it using the methods described for Cauchy-Euler equations in the appendix on
elementary ordinary differential equations. (The answer will be

y(x) = c1 cos(ln |x|) + c2 sin(ln |x|) .

To get this, you may have to use the fact that x = e ln|x| for x > 0 .)
b: Solve it using the method of Frobenius. (Warning: This may get messy I have not
tried it yet.)

version: 2/3/2010