Math 353 Lecture Notes

Power Series Solutions

J. Wong (Fall 2020)

Topics covered
• Review of power series:

◦ Basic properties, calculations with power series

◦ Radius of convergence

• Series solutions (2nd order linear ODEs)

◦ Motivation
◦ Process for computing power series solutions
◦ Simplifying the process ( ∞
P
n=−∞ )
◦ General solution / basis

1 Introduction
Earlier, we showed that solutions to homogeneous linear ODEs have the form

y = c1 φ1 + c2 φ2

where {φ1 , φ2 } is a basis for the solution space. The problem of ‘solving’ the ODE was
reduced finding a pair of coefficients, at the cost of having to obtain the basis functions -
which might be complicated or impossible to find exactly.

Instead, a more robust approach might be to choose the basis functions ourselves. To this
end, we might try a solution in the form of an infinite series:
∞
X
y= an φn
n=0

where the φn ’s are not solutions, but simple functions we choose. By representing functions
in this way, we can reduce the problem of solving an ODE to a sequence of equations for
the coefficients. Of course, we now have to deal with (potentially) an infinite number of
coefficients, and choose the functions φk in exactly the right way.

1
One possible choice for the φn ’s are polynomials, which leads to power series solutions.
In this section, we seek solutions to linear ODEs like

P (x)y 00 + Q(x)y 0 + R(x)y = 0

by representing the solution by a power series

∞
X
y(x) = an (x − x0 )n .
n=0

The disadvantages of the method are that solutions will be useful only near x0 where the
sum converges, and that the solution is an infinite series. On the other hand, the process is
straightforward when it works.

2 Power series: Definitions

Convergence of series: Let c0 , c1 , · · · be a sequence of numbers. The series

∞
X
cn
n=0

is said to be convergent if
m
X
lim cn exists
m→∞
n=0

and is said to be absolutely convergent if

m
X
lim |cn | exists.
m→∞
n=0

The ratio test says that if

|cn+1 |
lim <1 (1)
n→∞ |cn |

then the series converges absolutely.

We will not need to worry about the difference between the two definitions of convergence;
absolute convergence will be the relevant definition.

Definition: A power series centered at x0 is a series

∞
X
an (x − x0 )n . (2)
n=0

2
The series is well-defined only for x within some distance from x0 .The radius of conver-
gence ρ is the largest value such that
∞
X
an (x − x0 )n converges for all x such that |x − x0 | < ρ.
n=0

Within its radius of convergence, a power series converges and is therefore a function:
∞
X
f (x) = an (x − x0 )n , defined for |x − x0 | < ρ.
n=0

Typically, the ratio test is enough to find the radius of convergence. For instance, consider
∞
X (−1)n
ln(1 − x) = xn .
n=1
n

Taking the ratio of successive terms, we find that

(−1)n+1 xn+1 /(n + 1) n

lim = lim x = x.
n→∞ (−1)n xn /n n→∞ n + 1

The ratio test (1) says that the series converges for x if the limit is less than one (in absolute
value), so the series converges for |x| < 1. The radius of convergence is 1.

Note that one should take the ratio of successive non-zero terms, e.g. for
∞
X 1 2n
n
x ,
n=0
3
x2n+2 /3n+1 1 2 1 2 √
lim = lim x = x =⇒ ρ = 1/ 3.
n→∞ x2n /3n n→∞ 3 3

The radius of convergence can be zero! For instance, ∞ n

P
n=0 n!x converges only at x = 0.

The value of the power series (for our purposes) is that it provides ‘local’ approximations to
a function near x0 . For instance, suppose we have a power series for f (x) around zero:
∞
X
f (x) = an x n .
n=0

Define, for m = 0, 1, 2, · · · ,
m
X
Pm (x) = an x n
n=0

3
(the series with terms only up to degree m). Each Pm is a partial sum of the series1 . For
instance, for sin x, the first few partial sums for the series around zero are

x3 x3 x5
P1 = x, P3 = x − , P5 = x − + .
6 6 120
Adding more terms improves the approximation, but all are accurate only near x = 0.

3 Calculations with power series

Power series have many convenient properties. The properties below show how they can be
manipulated term by term. Here, a ‘term’ refers to a term of the sum, i.e. a coefficient times
a power of x. We will often refer to ‘the xj term’ to mean the term containing xj .

For simplicity, we will assume the power series is around x0 = 0 (but the results hold
for any x0 by just replacinh xn with (x − x0 )n ).

Important technical note: formulas hereafter are valid for x in the radius of conver-
gence of both series; this will be implied and not stated for each formula (the formulas are
correct where they are defined).

Equality
Two power series are equal,
∞
X ∞
X
an x n = bn x n ,
n=0 n=0

if all the terms are equal. The equality above holds, for instance, if an = bn for all n.
1
An ambiguity: there are several ways to count the number of terms. The ’m-th partial sum’ may refer
to the first m non-zero terms, the first m terms (including zero terms) or the terms up to degree m.

4
Addition
Power series can be added term-wise (by each power of x):
∞
X ∞
X ∞
X
n n
an x + bn x = (an + bn )xn .
n=0 n=0 n=0

Often the indexing is different for two series. To add, align powers of x in each sum. For
example,suppose we need to compute
∞
X ∞
X
n
an x + x bn x n .
n=0 n=0

Terms with the same power of x must be added together, so we need to shift the index to
be n + 1 instead of n. In full detail, here is the calculation:
∞
X ∞
X ∞
X ∞
X
n n n
an x + x bn x = an x + bn xn+1
n=0 n=0 n=0 n=0
∞
X ∞
X
= an x n + bm−1 xm (set m = n + 1)
n=0 m=1
X∞ X∞
= an x n + bn−1 xn (relabel m as n)
n=0 n=1
∞
X ∞
X
n
= a0 + an x + bn−1 xn (match starting index)
n=1 n=1
X∞
= a0 + (an + bn−1 )xn .
n=1

With the first few terms written out:

a0 + a1 x + a2 x 2 + · · ·
b0 x + b1 x 2 + · · ·
= a0 + (a1 + b0 )x + (a2 + b1 )x2 + · · ·
The second sum is written with a new index m = n+1 so that the power of x is xm , matching
the first sum. Then the two series can be added term by term. Note that the x0 term in the
first sum is alone, since the second sum starts at x1 .

The explicit change of index to m (and then back again) is often skipped. A terser ver-
sion of the calculation would look like the following:
∞
X ∞
X ∞
X ∞
X
n n+1 n
an x + bn x = an x + bn−1 xn
n=0 n=0 n=0 n=1
∞
X
= a0 + (an + bn−1 )xn .
n=1

5
Multiplication
The product of two power series
∞
X ∞
X
n
an x , bn x n
n=0 n=0

is itself a power series:

∞ ∞
! ∞
!
X X X
cn x n = an x n bn x n .
n=0 n=0 n=0
The right hand side will be a sum of all pairwise products of terms in the first sum with
terms in the second sum. To find the n-th coefficient cn , collect terms that multiply to give
xn . The terms aj xj and bn−j xn−j will combine to give aj bn−j xn . Writing things out:
(a0 +a1 x+a2 x2 +· · · )(b0 +b1 x+b2 x2 +· · · ) = a0 b0 +(a0 b1 +a1 b0 )x+(a2 b0 +a1 b1 +a0 b2 )x2 +· · ·
The result is that n
X
cn = aj bn−j .
j=0

The most important special case is when one power series is simple, in which case the formula
does not need to be invoked. For instance,
∞
X ∞
X ∞
X
2 n n+2
x an x = an x = an−2 xn .
n=0 n=0 n=2

If there are a few terms, it is not too bad to compute by computing each piece separately:
∞
X ∞
X ∞
X
(x2 − 3x) an x n = an xn+2 − 3an xn+1
n=0 n=0 n=0
∞
X ∞
X
= an−2 xn − 3an−1 xn
n=2 n=1
∞
X
= 3a0 x + (an−2 − 3an−1 )xn .
n=2

Note that the first sum is shifted by 2; the second one is shifted by 1. You could, of course,
only shift one of them by one instead and end up with xn+1 or xn+2 in the result.

Differentiation
Within its radius of convergence, a power series may be differentiated any number of times
and this can be done termwise:
∞
! ∞ ∞
d X n
X d n
X
an x = (an x ) = nan xn−1 .
dx n=0 n=0
dx n=1

Note that since the a0 term is a constant, it vanishes when taking the derivative. The radius
of convergence of all the derivatives is the same as the original series.

6
4 A useful trick (series arithmetic, the easy way)
Note that the ‘extra’ terms (like 3a0 x above) are exceptions that must be outside the sum.
An alternate approach simplifies this process (and is suggested for calculations, if you want).
Instead, we write a power series as a ‘doubly infinite’ sum:
∞
X ∞
X
n
an x → an x n
n=0 n=−∞

where we have extended the coefficients by zeros:

an = 0 for n < 0.

Now, there are no exceptional terms! To illustrate by example:

Addition of series: Take the example from before,

∞
X ∞
X
an x n + x b n xn .
n=0 n=0

Extend both coefficients by zeros:

an , bn = 0 for n < 0.

Now index-shifting the second term is easier. Below, all sums are from −∞ to ∞ (for
‘all integers n’).
∞
X X
x bn x n = bn xn+1
n=−∞ n
X
= bn−1 xn (shift n → n − 1)
n

Note that since the range is −∞ to ∞, the range doe snot change!

The addition then looks like:

∞
X ∞
X X X
n
an x + x b n xn . = an x n + bn xn+1
n=0 n=0 n n
X X
n
= an x + bn−1 xn (shift n → n − 1)
n n
X
= (an + bn−1 )xn .
n

We see that when n ≤ −1, the coefficient is zero, leaving

(a0 + b−1 )x0 + (a1 + b0 )x1 + · · ·

7
which gives a0 x0 + (a1 + b0 )x1 as calculated before.

Differentiation: This method is useul here also, but remember to pay attention to terms
that are zero. For example,

d2 X n
X X
2
a n x = an n(n − 1)xn−2 = an+2 (n + 2)(n + 1)xn .
dx n n
|{z}
n
shift n → n+2

Note that this sum (the non-zero part) starts at n = 0 since

an+2 (n + 2)(n + 1) = 0 if n = −1, n − 2

because of the (n + 2), (n + 1) factors, so this is really a power series

2a2 x0 + 6a3 x1 + · · · .

4.1 Note on arbitrary coefficients

The missing coefficients matter! Consider the equation

y 00 = 7.

If y is a power series X
y(x) = an x n
n

then we get X X
an+2 (n + 2)(n + 1)xn = 7 + 0 · xn .
n n≥1

Equating terms of each power, we get

xn term : an+2 = 0 for n ≥ 1

x0 term : 2a2 = 7
x−1 term : 0 · a1 = 0
x−2 term : 0 · a0 = 0

Thus there are two arbitrary coefficients, and we get

7
y(x) = x2 + a1 x + a0 , for any a0 , a1 ∈ R.
2

8
5 More on combining power series
Taylor series
Suppose
∞
X
f (x) = an x n .
n=0

Differentiating k times, we find that

∞
!
k
d X
f (k) (x) = an x n = k!ak + · · ·
dxk n=0

The first k − 1 terms vanish; all the others in the ellipsis have an x in them. Evaluating at
zero, we get
f (k) (0) = k!ak
which gives the familiar formula for the Taylor series,
1 (k)
ak = f (0). (3)
k!
Thus we can easily compute power series for functions when we know their derivatives.

For example, if we know f (x) has a power series

1
f (x) = x + x4 + · · ·
6
Then f 0 (x) = 1 + 23 x3 + · · · so f 0 (0) = 1. From this direct calculation (differentiating more
times) or Taylor’s formula we can compute
1
f (0) = 0, f 0 (0) = 1, f 00 (0) = f 000 (0) = 0, f (4) (0) = 4! · = 4.
6
Conversely, of course, we can get the power series for known functions using Taylor’s formula
(3), e.g.
1 n!
f (x) = =⇒ f (n) (x) = (−1)n =⇒ f (n) (0) = −n!
x−1 (x − 1)n+1
so Taylor’s formula yields the familiar geometric series
∞
1 X
= 1 + x + x2 + x3 + · · · = xn .
1−x n=0

5.1 Combining series, examples

We will need to use these rules to evaluate expressions involving power series. It is important
to make sure to align powers of x and deal with ’stray’ terms in one sum but not the other.
In particular, note that differentiating removes terms and tends to shift the index (depending

9
on how you write it).

Some examples will illustrate the typical calculations.

Example 1: Suppose
∞
X
f (x) = an x n
n=0

and we want to compute the series centered at x = 0 for

f 00 (x) − xf (x).

Differentiating term-wise and then aligning indices, we get

∞
X ∞
X
f 00 (x) − xf (x) = n(n − 1)an xn−2 − an xn+1
n=2 n=0
∞
X ∞
X
n
= (n + 2)(n + 1)an+2 x − an−1 xn
n=0 n=1
∞
X
= 2a2 + ((n + 2)(n + 1)an+2 − an−1 ).
n=1

For the second step, the first index is shifted so that n − 2 becomes n (m = n − 2) and the
second index is shifted so that n + 1 becomes n (m = n + 1). We’ll use this result shortly to
solve an ODE (see next section).

Example 2: Suppose
∞
X
f (x) = an x n
n=0

and we want to compute the series centered at x = 0 for

g(x) = f 00 (x) + x3 f 0 (x).

Plugging in the series for f and computing term-wise, we get

∞
X ∞
X
g= an n(n − 1)xn−2 + x3 an nxn−1
n=2 n=1
∞
X ∞
X
= an n(n − 1)xn−2 + an nxn+2 .
n=2 n=1

Note that when taking f 00 , we lose the x0 and x1 terms, so the indexing starts at 2. Now
replace the index in the first term with m = n − 2 and the index in the second term with

10
m = n + 2 (relabeling the new index as n again) to combine:
∞
X ∞
X
g= an n(n − 1)xn−2 + an nxn+2 .
n=2 n=1
∞
X ∞
X
n
= an+2 (n + 2)(n + 1)x + an−2 (n − 2)xn
n=0 n=3
∞
X
= 2a2 + 6a3 x + 12a4 x2 + [an+2 (n + 2)(n + 1) + an−2 (n − 2)] xn .
n=3

Note that the second sum starts with the x3 term, so the terms up to x2 in the first sum do
not get combined with anything in the second sum (so they get written separately)

It is worth noting that you can sanity check your results at any step by just writing out
a few terms of each sum. For instance,
∞
X
an n(n − 1)xn−2 = (a2 · 2 · 1 · x0 + a3 · 3 · 2 · x1 + · · ·
| {z } | {z }
n=2 n=2 n=3

and the shifted version is

∞
X
an+2 (n + 2)(n + 1)xn = a2 · 2 · 1x0 + a3 · 3 · 2 · x1 + · · ·
| {z } | {z }
n=0 n=0 n=1

so they are indeed equal. Writing out a few terms catches most errors in computation (which
are easy to make because of the bookkeeping involved).

5.2 Alternate method: starting with −∞

The alternate approach greatly simplifies the process, and is recommended for these calcu-
lations. Consider Example 2 again; we have
∞
X
f (x) = an x n
n=0

and wish to compute f 00 + x3 f 0 . Extending the an ’s by zeros, we have

∞
X
f (x) = an x n .
−∞
P P∞
Write sums n to mean n=−∞ . Then
X X
f 0 (x) = nan xn−1 f 00 (x) = n(n − 1)an xn−2
n n

11
Continuing with the example, we get
X X
f 00 (x) + x3 f 0 (x) = n(n − 1)an xn−2 + nan xn+2
n n
X X
= (n + 2)(n + 1)an+2 xn + (n − 2)an−2 xn
n n
X
= [(n + 2)(n + 1)an+2 + (n − 2)an−2 ] xn
n
X
= bn x n
n

where
bn = (n + 2)(n + 1)an+2 + (n − 2)an−2 .
You should check that the bn ’s are zero for n < 0, so this result really is a power series as
expected.

Now suppose we want to solve the ODE

f 00 + x3 f 0 = 0.
Then X X
bn x n = 0 · xn
n n

=⇒ (n + 2)(n + 1)an+2 + (n − 2)an−2 = 0 for all n.

The first few relevant equations are:
n = −2 =⇒ 0 · a0 = 0
n = −1 =⇒ 0 · a1 = 0
−(n − 2)
n ≥ 0 =⇒ an+2 = an−2 .
(n + 2)(n + 1)
This gives that
1
a2 = 0, a3 = 0, a4 = 0, a5 = −
a1 · · ·
20
and so on. You can then check from the formula that only a0 and the coefficients a1 , a5 , a9 , · · ·
are non-zero, and all multiples of a1 . This gives
1 5
f (x) = a0 + a1 (x − x + c9 x 9 + · · · )
20
for coefficients c5 , · · · you could calculate. The result here makes sense because the ODE is
a first order ODE for f 0 , and the solution is
Z
4
f = c1 + c2 e−x /4 dx

The process here suggests that we can use power series to get solutions to ODEs, the subject
of the next section.

12
6 Series solutions: Basics
Our goal is to obtain local approximations to a linear, second order ODE

P (x)y 00 + Q(x)y 0 + R(x)y = 0 (4)

near x = 0.2 In particular, we seek two linearly independent solutions as power series:
X ∞
X
n
y1 = bn x , y2 = cn x n .
n=0 n=0

We will use the following straightforward process:

1) Assume y(x) has the form of a power series:

X
y(x) = an x n . (5)
n=0

2) Write P, Q and R as power series as well:

∞
X ∞
X ∞
X
n n
P = pn x , Q= qn x , R= rn xn .
n=0 n=0 n=0

Most of the time, P, Q and R are simple (polynomials).

3) Substitute y into the ODE and use the multiplication/addition/differentiation rules to

write the result as a single sum:
X
0= (expression involving an ’s)xn .

4) Conclude that the coefficient of xn in the above must be zero for each n

5) Solve the equations to get the an ’s.

6) From the general solution (5), identify a solution basis and verify that they are linearly
independent.

Practically, this means we can approximate solutions by a few terms, and then add more as
needed to improve accuracy or understand finer properties of solutions. The process only
works under certain conditions (which we address later).
2
The methods herein work equally well for inhomogeneous ODEs, for higher order linear ODEs and
solutions around other points x0 ; we focus on (4) for simplicity as the general cases does not give any more
insight into the idea.

13
6.1 Illustrative example:
To see how the method works, we find a series solution for

y 00 + y = 0. (6)

Get equations for coefficients: We assume a solution series of the form

∞
X
y= an x n (7)
n=0

and substitute into the ODE:

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

Now shift the index on the first term by two:

∞
X
(an+2 (n + 2)(n + 1) + an )xn = 0.
n=0

For this equation to be true, we need the coefficients of xn to be zero for all n. Setting the
coefficients to zero yields a recurrence relation:
an
an+2 = − , n = 0, 1, 2 · · · . (8)
(n + 2)(n + 1)

Notice that there is no equation for a0 or a1 . We have shown that any sequence satisfying
(8) will make the series (7) a solution to the ODE (6). There is no restriction on a0 or a1 ;
they are arbitrary.

Solve for the coefficients: To find the coefficients, we iterate the recurrence relation.
It tells us that we solve for the an ’s as follows:

a0 → a2 → a4 → a6 , · · ·
a1 → a3 → a5 → a7 , · · ·

For even coefficients, using the recurrence to get a2 and a4 gives

a0 a0 a2 a0
a2 = − =− , a4 = − = .
1·2 2! 3·4 4!
This suggests that
a0
a2n = (−1)n , n = 1, 2, · · · (9)
(2n)!
In general, for even indices,
a2n−2
a2n = −
2n(2n − 1)

14
and it is straightforward to show from this that our guess (9) is correct.

For the odd coefficients, we have

a1 a3 a1
a3 = − , a5 = − =
3! 4·5 5!
which suggests that
a1
a2n+1 = (−1)n , n = 1, 2, · · · . (10)
(2n + 1)!
Note that the odd coefficients are all multiples of a1 and the even ones are all multiples of a0 .

We can therefore split the power series for y into two parts:
∞
X ∞
X ∞
X
n 2n
y= an x = a2n x + a2n+1 x2n+1 .
n=0 n=0 n=1

Now use the formulas (9) and (10) for the coeficients:
∞ ∞ ∞
X
n
X (−1)n 2n
X (−1)n+1 2n+1
y= an x = a0 x + a1 x .
n=0 n=0
(2n)! n=1
(2n + 1)!

This is in the form of the span of two solutions,

y = a0 y1 + a1 y2 ,
where ∞
X (−1)n
y1 = x2n = cos x,
n=0
(2n)!
∞
X (−1)n 2n+1
y2 = x = sin x
n=0
(2n + 1)!
(by recognizing the power series for cos x and sin x). The result makes sense since, of course,
cos x, sin x are a basis for solutions to y 00 + y = 0.

Checking linear independence: Last, we need to verify that y1 , y2 are linearly inde-
pendent. This can be done from the series form. The first few terms of each are
1 1
y1 = 1 − x2 + · · · , y2 = x − x3 + · · ·
2 6
From this we can easily compute

y1 (0) = 1, y10 (0) = −x + · · ·  =0

x=0

and
1 
y2 (0) = 0, y20 (0) = 1 − x2 + · · ·  =1

2 x=0
so the two solutions are linearly independent. (You could also just use Taylor’s theorem
directly instead of computing f 0 (x) then plugging in 0).

15
7 A more difficult example
We find the general series solution to the Airy equation

y 00 = xy

which arises in various applications in physics (e.g. optics, quantum mechanics). Unlike the
previous example, the Airy equation does not have an exact solution, so the series solution
will give information we cannot otherwise obtain!

Again, look for a solution of the form

∞
X
y= an x n .
n=0

Here, let’s use the extension trick and write

∞
X
y= an x n , an = 0 for n < 0.
n=−∞

Now substitute into the ODE:

X X
an n(n − 1)xn−2 − an xn+1 = 0.
n n

Now shift the index of the first sum up by three:

X
(an+3 (n + 3)(n + 2) − an ) xn+1 = 0.
n

Plugging in n = −2 and n = −3 we find that

0 · a0 = 0, 0 · a1 = 0 =⇒ a0 , a1 arbitrary.

Now for n = −1,

2a2 = 0 =⇒ a2 = 0
so a2 must be zero (not arbitrary!). For the rest,
an
an+3 = , n ≥ 0. (11)
(n + 3)(n + 2)

Since an depends on an−3 , (11) produces three independent sequences of coefficients.

First, the one starting with a2 . Since a2 = 0, we also have

0 = a2 = a5 = a8 = · · ·

16
so a3m+2 = 0 for all m.

For the other two parts, observe that a3 , a6 , a9 , · · · are all multiples of a0 :
a3m = c3m a0 for certain coefficients c3m
and similarly
a3m+1 = c3m+1 a1 for certain coefficients c3m+1
Plugging all this into the series and breaking it up into the three parts (the zeros, the a0 and
a1 parts), we find that
∞
X ∞
X
y(x) = a0 c3m x3m + a1 c3m+1 x3m+1
m=0 m=0
= a0 y1 (x) + a1 y2 (x)
which is exactly the desired structure. To verify linear independence, it’s not hard to compute
from the recurrence that the first two terms of y1 , y2 are
1 1 4
y 1 = 1 + x3 + · · · , y2 = x + x + ···
6 12
Since
y1 (0) = 1, y10 (0) = 0, y2 (0) = 0, y20 (0) = 1
the Wronskian is 1 at x = 0 so the solutions are indeed linearly independent.

Extra (formula for the coeffs): To compute the coefficients, it’s convenient to rewrite
n−2
an = an−3 .
n(n − 1)(n − 2)
The reason is that the n(n − 1)(n − 2) in the denominator means that we will get n! in the
denominator for an . The numerator will give factors 1 · 4 · · · and 2 · 5 · · · (each factor goes
up by three each iteration). The result is that
1 · 4 · · · (3m − 2) 2 · 5 · · · (3m − 1)
a3m = a0 , a3m+1 = a1 .
(3m)! (3m + 1)!
Setting a0 = 1 and a1 = 0 we get a solution with only the x3m terms:
∞  
X 1 · 4 · · · (3m − 2)
y1 = x3m
m=0
(3m)!

and setting a0 = 0 and a1 = 1 we get a solution with only x3m+1 terms:

∞  
X 2 · 5 · · · (3m − 1)
y2 = x3m+1 .
m=0
(3m + 1)!

The general solution is

y = a0 y1 + a1 y2 .

17
8 Theory for series solutions: Ordinary points
We have found, for a few examples, power series solutions to linear equations

P (x)y 00 + Q(x)y 0 + R(x)y = 0.

The main questions are:

• Under what conditions can we find a power series solution to the ODE?

• If such a solution exists, where is it valid (what is the radius of convergence?)

• What can we do if a power series does not exist?

8.1 Analytic functions

Definition: A function f (x) is called analytic3 at a point x0 if it has a power series

expansion at x0 with a non-zero radius of convergence ρ:
∞
X
f (x) = an (x − x0 )n for |x − x0 | < ρ.
n=0

Informally: a function is analytic at x0 if it has a power series at x0 .

For some examples, consider, around x = 0,

cos x
(i) sin x/x, (ii) , (iii) x5/2
x
i) Analytic at x = 0 because it has a power series

sin x/x = (x − x3 /6 + · · · )/x = 1 − x2 /6 + · · ·

ii) Not analytic because it has a singularity at x = 0. Attempting to use the approach
from (i) gives
1
cos x/x = (1 − x2 /2 + · · · )/x = + · · ·
x
so there is a 1/x-like singularity at x = 0.

iii) Not analytic at x = 0. To have a power series, the function must have derivatives of
all orders at x = 0, but
(x5/2 )000 = Cx−1/2
which is infinite at x = 0.
3
Technicality: This really should be ’real analytic’ for a real-valued function, or the condition should be
for complex numbers with |z − z0 | < ρ, but the distinction is irrelevant here.

18
8.2 Ordinary points (when do power series solutions exist?)
Whether the power series method can be used on a linear ODE depends on whether the
coefficient functions are analytic. For simplicity, we consider second order ODEs of the form

y 00 + p(x)y 0 + q(x)y = 0 (12)

Definition: We call a point x0 ...

• ...an ordinary point of the ODE (12) if p and q are analytic at x0 ,

• ...a singular point of the ODE if either p or q fail to to be analytic x0 ,

For example, the ODE

(x2 − 1)2 y 00 + (sin x)y = 0
has an ordinary point at x = 0. In the ‘standard form,
sin x
y 00 + y = 0.
x2 − 1
The coefficient of y 0 is zero and the coefficient of y is
sin x sin x
q= =
(x2 − 1)2 (x − 1)2 (x + 1)2
which has singularities at x = ±1, but is otherwise analytic. So in fact, is has ordinary
points for all x0 except x0 = ±1.
(Note: the fact that the functions are analytic away from the singularity is known since
sin x, 1/(x ± 1)2 have power series and so their product does as well).

As a second example, the Bessel equation of order α

x2 y 00 + xy 0 + (x2 − α2 )y = 0

(with α a constant) has a regular singular point at x = 0, since it is equivalently

α2
 
00 1 0
y + y + 1 − 2 y = 0.
x x

8.3 A hint at what happens at singular points

A simple example illustrates what can go wrong at singular points. Consider

4x2 y 00 + 4xy 0 − y = 0

We know this as an Euler equation; recall that plugging in xr yields solutions when

0 = 4r2 − 1 = (r − 1/2)(r + 1/2)

19
so the solution is
y(x) = c1 x1/2 + c2 x−1/2
This ODE has a singular point at x = 0 since
1 1
y 00 + y 0 − 2 y = 0
x 4x
has singular coefficients there. Notice that the solution has no power series at x = 0 (because
x1/2 ) has no power series at x = 0), so the power series method cannot work.

You can check that looking for a solution

∞
X
y(x) = an x n
n=0

and plugging in via the usual procedure will not yield two linearly independent solutions!

For the details: Plug in and move the 1/x, 1/x2 coefficients inside the sum to get
X X 1X
an n(n − 1)xn−2 + an nxn−2 − an xn−2 = 0
n n
4 n

which does not require any shifting! Combine to get

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

The equations to solve for the coefficients are then

1
(n2 − )an = 0 for integers n ≥ 0.
4
But for the power series, the n’s are integers so the factor n2 − 1/4 is never zero, which yields

an = 0 for all n =⇒ y(x) = 0.

THus, the only power series solution is the trivial one.

8.4 Main result: ordinary points

Near an ordinary point, the coefficients p(x) and q(x) in the ODE

y 00 + p(x)y 0 + q(x)y = 0 (13)

can be expanded in a power series to obtain

X  X 
y 00 + pn (x − x0 )n y 0 + qn (x − x0 )n y = 0.
n=0 n=0

20
 an (x − x0 )n , all the operations (derivatives, multiplication, addition)
P
If we can plug in y =
on power series yield another power series, so the result will be
∞
X
(· · · )(x − x0 )n = 0
n=0
n
where each coefficent of (x−x0 ) is an expression involving the unknowns a0 , · · · , an , leading
to a recurrence that can be solved. It turns out that the resulting recurrence will always have
a solution at an ordinary point. The main result (not proven here) confirms this observation:

Theorem on ordinary points: (i) At an ordinary point x0 , The ODE (13) has a power
series solution with a non-zero radius of convergence, which will have the form
∞
X
y= an (x − x0 )n = a0 y1 + a1 y2
n=0

for linearly independent series solutions y1 , y2 .

(ii) Moreover, the series for y1 , y2 are guaranteed to converge in the interval
where the series for p and q both converge.

That is, the radius of convergence of y1 (and y2 ) is at least as large as the mini-
mum of the radii of convergence for p and q.

The second part gives us a lower bound on the region where solutions will converge just
by looking at the ODE. We simply need to find where p and q have convergent power series
and take the intersection.

For example, the Airy equation

y 00 = xy
has an ordinary point at every x0 since x is a polynomial. Since
∞
X
x= an x n , a1 = 1, an = 0 otherwise
n=0

is already a power series, it converges everywhere (the ’sum’ is just one term). Thus the
series solutions to the Airy equation also converge everywhere.

On the other hand, consider a series around x0 = 0 for

(1 − x)y 00 + xy 0 + y = 0. (14)
The coefficients p(x) = x/(1 − x) and q = 1/(1 − x) both have P series around zero with
radius of convergence 1 (check this!) so any series solution y = an xn to (14) has radius
of convergence at least one. We do not know that the radius of convergence for a solution
will be exactly one (it could be larger!).

21