Math 1280

Notes 8
More on series solutions, and an introduction to “orthogonal polynomials”

1 Ordinary points and singular points

We are considering general second order linear homogeneous equations

P (x) y 00 + Q (x) y 0 + R (x) y = 0; (1)

and looking for series solutions

X
1
y= an (x x0 )n :
n=0

We assume that P; Q and R all have power series expansions around x0 . (In most
cases that we will consider, P; Q and R are polynomials. Any polynomial has a
Taylor series around any point, and since eventually the derivatives of the function
are zero, the Taylor series is a …nite sum. This is the expansion we are referring to.

Example: P (x) = x2 + x + 1. Find the Taylor series around x = 2:

Method 1

P 0 (x) = 2x + 1
P 00 (x) = 2
P (j) (x) = 0 for j > 1:

Hence,

P (2) = 7
P 0 (2) = 5
P 00 (2) = 2

so
2
P (x) = 7 + 5 (x 2) + (x 2)2 + 0 x3 + ::
2!
with all the remaining terms 0:

1
 Method 2

x2 + x + 1 = (x 2)2 + 4x 4 + f(x 2) + 2g + 1
2
= (x 2) + f4 (x 2) + 8g 4 + (x 2) + 3
= (x 2)2 + 5 (x 2) + 7

Many other ways of algebraic manipulation could be used. The Taylor series is the
most straightforward, but not necessarily the quickest.

De…nition 1 Assume that P; Q;and R have no common factor. Then a point x0

is a “ordinary point” for (1) if P (x0 ) 6= 0: Otherwise x0 is a “singular” point for
(1) :

There are several examples given in the text. Be sure to reach Theorem 5.3.1.
It was illustrated at the end of the last section of notes.

2 Hermite’s equation
Hermite’s equation (problem 21, pg. 260 ) is

y 00 2xy 0 + y = 0;

where y is a function of x. Notice that there are no singular points. We can …nd a
series solution by the usual way:
X
1
y= an x n
n=0
X1
y0 = nan xn 1

n=1
X
1
y 00 = n (n 1) an xn 2 :
n=2

Substituting this into Hermite’s equation gives

X
1 X
1 X
1
n 2 n
n (n 1) an x 2 nan x + an x n = 0
n=2 n=1 n=0

2
Rearranging the indices, we let m = n 2 in the …rst sum and then change m back
to n; to get
X
1 X
1 X
1
n n
(n + 2) (n + 1) an+2 x 2 nan x + an x n = 0
n=0 n=1 n=0

The coe¢ cient of x0 then gives us

2a2 + a0 = 0
a0
a2 =
2
and the coe¢ cient of xn for n > 0 gives
( 2n)
an+2 = an
(n + 2) (n + 1)
Hence, the even terms are
4 ( 4)
a4 = a2 = a0
(4) (3) 4!
8 ( 4) ( 8)
a6 = a4 = a0
(6) (5) 6!
and so forth, while the odd numbered terms are
2
a3 = a1
3!
( 6) ( 2)
a5 = a1
5!
( 10) ( 6) ( 2)
a7 = a1
7!
and so forth. We get two linearly independent solutions by …rst setting a0 = 1;
a1 = 0; and then setting a0 = 0; a1 = 1. Notice that y (0) = a0 and y 0 (0) = a1 .
The solutions are
( 4) ( 4) ( 8)
y1 (x) = 1 x2 + x4 x6 +
2! 4! 6!
and
2 ( 2) ( 6) ( 2) ( 6) ( 10)
y2 (x) = x x3 + x5 x7 +
3! 5! 7!
3
 Since Hermite’s equation has no singular points, Theorem 5.3.1 tells us that the
series converges for all x: But the most interesting cases are for = 2; 4; 6; 8; etc.,
that is, any even positive integer. In these cases we can see that one or the other
of these functions is not an in…nite sum, but only a …nite sum.

For example, if = 10; then

8 3 32 5
y2 (x) = x x + x
3! 5!
and all the later terms are zero. The functions we get then are the so-called "Hermite
polynomials", or multiples thereof. For a reason we will give below, the Hermite
polynomial Hn , where the highest order term is a multiple of xn , is chosen so the
coe¢ cient of xn is 2n : For example, H5 is supposed to start o¤ with 32xn ; so we get
8
H5 (x) = 32x5 5!x3 + 5!x = 32x5 160x3 + 120x:
3!
The …rst four Hn are (as given in the text), pg. 260 )

H0 (x) = 1
H1 (x) = 2x
H2 (x) = 4x2 2
H3 (x) = 8x3 12
H4 (x) = 16x4 48x2 + 12

I will now list some of the properties of the Hermite polynomials. We will not
have time to prove these in general, but I will try to give some examples. You can
…nd these properties on Wikipedia, or on "MathWorld", another good math website.

1. They satisfy a "recurrence relation"

Hn+1 (x) = 2xHn (x) 2nHn 1 (x)

This is a quick way of …guring them out.

2. They have a “generating function", obtained from the power series for ex :
The relevant formula is
X1
Hn (x) n
2xt t2
e = t :
n=0
n!

4
This is seen from
t2 1 1 2
e2xt =1+ 2xt t2 + 2xt t2 +
1! 2!
1
= 1 + 2xt t2 +
4x2 t2 + t4 4xt3 + ::
2!
2x 4x2 2 2
=1+ t+ t +
1! 2!
Notice that the …rst three terms agree with what we got earlier, and this formula
gives a reason for choosing Hn as we did, with 2n as the leading coe¢ cient.

3. They satisfy an “orthogonality relation":

Z 1
2
Hm (x) Hn (x) e x dx = 0:
1

Recall from math 1180 that polynomials can be orthogonal with respect to an inner
2
product. Here the inner product is weighted by the factor e x ; which is necessary
for the inde…nite integrals to converge.

Many more properties are listed on the websites.

The Hermite polynomials are examples of what are called “special functions”,
which are functions, usually de…ned by either polynomials or in…nite series, which
are important in physics and other applications. Doing a Google search for "Hermite
polynomials" yielded about 117,000 links.

We shall run into several more special functions in this chapter. The next comes
from:

3 Chebyshev’s equation
Chebyshev’s equation (problem 10, pg. 265 ) is

1 x2 y 00 xy 0 + 2
y = 0;

where y is a function of x. Now there are singular points at x = 1. However

x0 = 0 is an ordinary point.

5
 We can …nd these solutions in the usual way:
X
1
y= an xn ; etc. for the derivatives.
n=0

Substituting this into Chebyshev’s equation results eventually in

2
2a2 = a0
2
a0
a2 =
2
The coe¢ cient of x1 gives
2
6a3 + 1 a1 = 0
2
1
a3 = a1 :
3!
and the coe¢ cient of xn for n > 0 gives
2
n (n 1) + n n2 2
an+2 = an = an :
(n + 2) (n + 1) (n + 2) (n + 1)
Hence, the even terms are
2 2 2
4 (4 )
a4 = a2 = a0
(4) (3) 4!
2 2 2 2
16 (16 ) (4 )
a6 = a4 = a0
(6) (5) 6!
and so forth, while the odd numbered terms are
2
1
a3 = a1
3!
(32 2
) (12 2
)
a5 = a1
5!
(52 2
) (32 2
) (12 2
)
a7 = a1
7!
and so forth. We get two linearly independent solutions as usual:
2 2 2 2 2 2
(4 ) (4 ) (16 )
y1 (x) = 1 x2 x2 x4
2! 2! 4!
6
and
2
1 (32 2
) (12 2
) (52 2
) (32 2
) (12 2
)
y2 (x) = x + x3 + x5 + x7 +
3! 5! 7!
From Theorem 5.3.1 it follows that the radius of convergence in each case is at
least 1; because the series for 1 1x2 has radius of convergence equal to 1.
But we can also see that if is an integer, then one of the series stops after a
certain point, and we get a polynomial, as before, with Hermite’s equation. In this
case, there is no issue of convergence. We can say that r = 1. This emphasizes
that Theorem 5.3.1 gives a minimum value for r: It could be larger, as it is in this
example if is an integer.

The Chebyshev polynomials are denoted by Tn (x) : (Perhaps someone who has
studied Russian can explain why.) They are de…ned to be the polynomial solution of
Chebyshev’s equation, with = n; normalized so that Tn (1) = 1: They give another
class of special functions, and have the same sorts of properties as the Hermite
polynomials:

1. There is a “generating function”, with

1 t2 X1
= T0 (x) + 2Tn (x) tn
1 2xt + t2 n=1

2. They have an orthogonality property

Z 1
1
Tn (x) Tm (x) p dx = 0 if n 6= m:
1 1 x2
Note the weight function.

3. There is a recurrence relation:

Tn+1 (x) = 2xTn (x) Tn 1 (x)

4. Here is a direct formula for Tn (x) :

1h p n p ni
Tn (x) = x + x2 1 + x x2 1
2

7
This seems to involve complex numbers for jxj < 1; but the complex parts will cancel.

5. Here is another neat formula: If x = cos , then

Tn (x) = cos (n ) :

In other words,
Tn (x) = cos (n arccos (x))
Notice that this formula cannot be used if jxj > 1; since then we can’t have x = cos .
(Or can we? It turns out that we can if we allow to be complex, but we are not
considering that here.)

The Chebyshev polynomials turn out to be important in numerical analysis, be-

cause when they are used to approximate non-polynomial functions, they are e¢ cient
in giving a good approximation. .We can see that there is something unusual about
them from the two formulas involving trig functions. Their values lie in [ 1; 1] if
jxj 1. (“Obviously”, this cannot be true for all x: Why not?)

In this sense, they resemble trigonometric functions, over this interval, and yet
they are polynomials, not in…nite sums. Here is the plot of T20 (x):;

8
 Another quite surprising property is that Tn has all of its maxima and minima
in [ 1; 1], and the maximum and minimum values are 1. It is perhaps surprising
that any polynomial could have such a property. Plus, as property (2) above states,
they are orthogonal with respect to a particular weight function.

These properties make the Chebyshev polynomials perhaps even more important
than those of Hermite. A Google search for "Chebysheve polynomials" gives about
130,000 links.

4 Homework
Section 5.2, # 2,7,15
Section 5.3, # 6, 23 (you can use the result from 22).