0% found this document useful (0 votes)
128 views198 pages

WWW - Math.iitb - Ac.in/ Swapneel/207: Partial Differential Equations

The document discusses power series and their properties. Some key points: 1. A power series is a series of terms involving powers of a variable centered at a point. It converges in an interval around this center point. 2. The radius of convergence is the largest interval where the power series converges. 3. A function defined by a convergent power series is infinitely differentiable in its interval of convergence. 4. Real analytic functions can be represented by power series in an open interval around each point in their domain. Solving differential equations using power series involves determining the coefficients to satisfy the equation.

Uploaded by

Spandan Patil
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
128 views198 pages

WWW - Math.iitb - Ac.in/ Swapneel/207: Partial Differential Equations

The document discusses power series and their properties. Some key points: 1. A power series is a series of terms involving powers of a variable centered at a point. It converges in an interval around this center point. 2. The radius of convergence is the largest interval where the power series converges. 3. A function defined by a convergent power series is infinitely differentiable in its interval of convergence. 4. Real analytic functions can be represented by power series in an open interval around each point in their domain. Solving differential equations using power series involves determining the coefficients to satisfy the equation.

Uploaded by

Spandan Patil
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 198

Partial differential equations

Swapneel Mahajan

www.math.iitb.ac.in/˜swapneel/207

1
1 Power series

For a real number x0 and a sequence (an ) of real


numbers, consider the expression

X
an (x−x0 )n = a0 +a1 (x−x0 )+a2 (x−x0 )2 +. . . .
n=0

This is called a power series in the real variable x.

The number an is called the n-th coefficient of the


series and x0 is called its center.

For instance,

X 1 1 1
(x−1) = 1+ (x−1)+ (x−1)3 +. . .
n

n=0
n+1 2 3

is a power series in x centered at 1 and with n-th


1
coefficient equal to n+1 .

2
A power series centered at 0 has the form

X
an xn = a0 + a1 x + a2 x2 + . . . .
n=0

For instance,

X 1 n 1 1 3
x = 1 + x + x + ...
n=0
n+1 2 3

is a power series in x centered at 0 and with n-th


1
coefficient equal to n+1 .

3
What can we do with a power series?

We can substitute a value for the variable x and get a


series of real numbers.

We say that a power series converges at a point x1 if


substituting x1 for x yields a convergent series, that is,
if the series

X
an (x1 − x0 )n
n=0
converges to a real number.

A power series always converges at its center x0 to the


constant term a0 since

X
an (x0 − x0 )n = a0 .
n=0

(By convention 00 = 1.)


We want to know the values of x where a power series
converges.

How far can we move away from the center?

4
Lemma. Suppose a power series centered at x0
converges for some real number x1 6= x0 . Let
|x1 − x0 | = r.
Then the power series is convergent for all x such that
|x − x0 | < r.
That is, the power series is convergent in the open
interval of radius r centered at x0 .

The radius of convergence of the power series is the


largest nonnegative number R such that the power
series converges in the open interval

{x | |x − x0 | < R}.

The latter is called the interval of convergence of the


power series.

A power series can converge at all points in which case


we take R to be ∞.

The radius of convergence can also be 0.

5
We assume from now on that R > 0.
A power series determines a function in its interval of
convergence.

Denoting this function by f , we may write



X
f (x) = an (x − x0 )n , |x − x0 | < R.
n=0

Fact. The function f is infinitely differentiable in the


interval of convergence.

The successive derivatives of f can be computed by


differentiating the power series on the right termwise.

6
Hence

f (x0 ) = a0 , f 0 (x0 ) = a1 , f 00 (x0 ) = 2a2 ,

and in general,

f (n) (x0 )
an = .
n!

Thus, knowing f (x) in a small open interval around x0


determines it everywhere in the open interval
|x − x0 | < R.

7
2 Real analytic functions

Let R denote the set of real numbers.

A subset U of R is said to be open if for each x0 ∈ U,


there is a r
> 0 such that the open interval
|x − x0 | < r is contained in U .
For example, open intervals are open sets.

The interval (0, ∞) is also an open set.

Union of open sets is again an open set. For example,

R \ {0} = (−∞, 0) ∪ (0, ∞)

The interval [0, 1] or (0, 1] is not an open set.

8
Let f : U → R be a real-valued function on an open
set U .

We say f is real analytic at a point x0 ∈U


if there exists a power series centered at x0 which
converges to f in some open interval around x0 ,

that is,

X
f (x) = an (x − x0 )n
n=0
holds in some open interval around x0 .

We say f is real analytic on U if it is real analytic at all


points of U .

In general, we can always consider the set of all points


in the domain where f is real analytic. This is called
the domain of analyticity of f .

9
Corollary. Suppose f is real analytic on U .

Then f is infinitely differentiable on U .

Its power series representation around x0 is


necessarily the Taylor series of f around x0 ,

that is, the coefficients an are given by

f (n) (x0 )
an = .
n!
Just like continuity or differentiability, real analyticity is a
local property.

That is, to know whether f is real analytic at a point,


we only need to know the values of f in an open
interval around that point.

10
Example. Polynomials such as x3 − 2x + 1 are real
analytic on all of R.

A polynomial is a truncated power series

(so there is no issue of convergence).

By writing x = x0 + (x − x0 ) and expanding, we can


rewrite any polynomial using powers of x − x0 .

This will be a truncated power series centered at x0 .

For instance,

1 + 2x + x2 = 4 + 4(x − 1) + (x − 1)2 .

11
Fact. A power series is real analytic in its interval of
convergence.

Thus, if a function is real analytic at a point x0 ,

then it is real analytic in some open interval around x0 .

Thus, the domain of analyticity of a function is an open


set.
Fact. If f and g are real analytic functions on U , then
so are
cf, f + g, f g and f /g.

Here cf is the multiple of f by the scalar c.

Also we wrote f /g above, it is implicit that g 6= 0 on


U.

12
Example. The sine, cosine and exponential functions
are real analytic on all of R.

Their Taylor series around 0 are

x3 x5
sin(x) = x − + + ...
3! 5!
x2 x4
cos(x) = 1 − + + ...
2! 4!
x 2 x3
ex = 1 + x + + + ....
2! 3!
One can show that these identities are valid for all x.

For example,

π3 π5
sin(π) = π − + + ....
3! 5!

13
Example. Consider the function
1
f (x) = 2
.
1+x
It is real analytic on all of R.

The Taylor series around 0 is


1 2 4 6
2
= (1 − x + x − x + . . . ).
1+x
This series has a radius of convergence 1 and the
identity only holds for |x| < 1.

At x = 1, note that f (1) = 1/2 while the series


oscillates between 1 and 0.

Thus from here we can only conclude that f is real


analytic in (−1, 1).

If we want to show that f is real analytic at 1,

then we need to need to find another power series


centered at 1 which converges to f in an interval
around 1.

This is possible.

14
Even though a function is real analytic, one may not be
able to represent it by just one power series.

15
Example. The function f (x) = x1/3 is defined for all
x.
It is not differentiable at 0 and hence not real analytic at
0.
However it is real analytic at all other points.

For instance,

1/3 1/3 1 1 1 (x − 1)2


x = (1+(x−1)) = 1+ (x−1)+ ( −1) +. . .
3 3 3 2!
is valid for |x − 1| < 1, showing analyticity in the
interval (0, 2).

(This is the binomial theorem which we will prove later.)

Analyticity at other nonzero points can be established


similarly.

Thus, the domain of analyticity of f (x) = x1/3 is


R \ {0}.

16
Example. The function

e−1/x2 if x =
6 0,
f (x) =
0 if x = 0,

is infinitely differentiable.

But it is not real analytic at 0.

This is because f (n) (0) = 0 for all n and the Taylor


series around 0 is identically 0. So the Taylor series
around 0 does not converge to f in any open interval
around 0.

The domain of analyticity of f is R \ {0}.

17
For a function f , never say “convergence of f ”, instead
say “convergence of the Taylor series of f ”.

18
3 Solving a linear ODE by the power
series method

Example. Consider the first order linear ODE

y 0 − y = 0, y(0) = 1.

We know that the solution is y = ex .


Let us use power series to arrive at this result.

We assume that the solution has the form



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

The initial condition y(0) = 1 implies a0 = 1.


Substitute y =
P n in the ODE y 0 = y and
n an x
compare the coefficient of xn on both sides.

y 0 = a1 + 2a2 x + 3a3 x2 + · · · + (n + 1)an+1 xn + . . .


y = a0 + a1 x + a2 x2 + · · · + an xn + . . . .

19
This yields the equations

a1 = a0 , 2a2 = a1 , 3a3 = a2 , . . . .

In general,

(n + 1)an+1 = an for n ≥ 0.

Such a set of equations is called a recursion.

Since a0 = 1, we see that

a1 = 1, a2 = 1/2, a3 = 1/6, . . . .

In general,
an = 1/n!.
Thus

X 1 n
y(x) = x
n=0
n!
which we know is ex .

20
Example. Consider the first order linear ODE

y 0 − 2xy = 0, y(0) = 1.

We proceed as in the previous example.

The initial condition y(0) = 1 implies a0 = 1.

y 0 = a1 + 2a2 x + 3a3 x2 + · · · + (n + 1)an+1 xn + . . .


2xy = 2a0 x + 2a1 x2 + 2a2 x3 + · · · + 2an−1 xn + . . . .

This time we get the recursion:

(n + 2)an+2 = 2an for n ≥ 0,

and a1 = 0.
So all odd coefficients are zero.

21
For the even coefficients, put n = 2k .
This yields

(2k + 2)a2k+2 = 2a2k for k ≥ 0.

This is the same as

(k + 1)a2k+2 = a2k for k ≥ 0.

This yields
a2k = 1/k!.
Thus

X 1 2k
y(x) = x
k=0
k!
2
which we know is ex .

22
Example. Consider the second order linear ODE

y 00 + y = 0.

(The initial conditions are left unspecified.)

Proceeding as before, we get

(n + 2)(n + 1)an+2 + an = 0 for n ≥ 0,

and a0 and a1 are arbitrary.

Solving we obtain

x2 x4 x3 x5
y(x) = a0 (1− + +. . . )+a1 (x− + +. . . ).
2! 4! 3! 5!
Thus,
y(x) = a0 cos(x) + a1 sin(x).

23
Consider the following initial value problem.

p(x)y 00 + q(x)y 0 + r(x)y = g(x),

y(x0 ) = y0 , y 0 (x0 ) = y1 .
Here x0 , y0 and y1 are fixed real numbers.

We assume that the functions p, q , r and g are real


analytic in an interval containing the point x0 .

Let r> 0 be less than the minimum of the radii of


convergence of the functions p, q , r and g expanded in
power series around x0 .

We also assume that p(x) 6= 0 for all x in the interval


(x0 − r, x0 + r).

24
Theorem. Under the above conditions, there is a
unique solution to the initial value problem in the
interval (x0 − r, x0 + r), and moreover it can be
represented by a power series

X
y(x) = an (x − x0 )n
n=0

whose radius of convergence is at least r .

25
Moreover, there is an algorithm to compute the power
series representation of y .

It works as follows.

• Plug a general power series into the ODE.

• Take derivatives of the power series formally.

• Equate the coefficients of (x − x0 )n for each n.

• Obtain a recursive definition of the coefficients an .

• The an ’s are uniquely determined and we obtain


the desired power series solution.

In most of our examples, x0 = 0 and the functions p,


q , r and g will be polynomials of small degree.

26
This result generalizes to any n-th order linear ODE
with the first n − 1 derivatives at x0 specified.

For the first order linear ODE, the initial condition is


simply the value at x0 . Suppose the first order ODE is

p(x)y 0 + q(x)y = 0.

Then, by separation of variables, the general solution is


q(x)
R
− dx
ce p(x) ,

provided p(x) 6= 0 else the integral may not be


well-defined.

This is an indicator why such a condition is required in


the hypothesis of the theorem.

27
Example. Consider the function

f (x) = (1 + x)p

where |x| < 1 and p is any real number.


Note that it satisfies the linear ODE.

(1 + x)y 0 = py, y(0) = 1.

Let us solve this using the power series method around


x = 0.
Since 1 + x is zero at x = −1, we are guaranteed a
solution only for |x| < 1.
The initial condition y(0) = 1 implies a0 = 1.
To calculate the recursion, express each term as a
power series:

y 0 = a1 + 2a2 x + 3a3 x2 + · · · + (n + 1)an+1 xn + . . .


xy 0 = a1 x + 2a2 x2 + · · · + nan xn + . . .
py = pa0 + pa1 x + pa2 x2 + · · · + pan xn + . . . .

28
Comparing coefficients yields the recursion:

p−n
an+1 = an for n ≥ 0.
n+1
Since a0 = 1, we see that
p(p − 1) p(p − 1)(p − 2)
a1 = p, a2 = , a3 = ,....
2 6
This shows that

p p(p − 1) 2
(1 + x) = 1 + px + x + ....
2
This is the binomial theorem and we just proved it.

The most well-known case is when p is a positive


integer (in which case the power series terminates to a
polynomial of degree p).

29
Given a function f , it is useful to know whether it
satisfies an ODE.

30
Example. Consider the second order linear ODE

y 00 + y 0 − 2y = 0.

(The initial conditions are left unspecified.)

Proceeding as before, we get the recursion:

(n+2)(n+1)an+2 +(n+1)an+1 −2an = 0 for n ≥ 0,

and a0 and a1 are arbitrary.

The recursion involves three terms.

How do we solve it?

The key idea is to make the substitution

bn = n!an ,

so we obtain

bn+2 + bn+1 − 2bn = 0 for n ≥ 0,

31
We now guess that bn = λn is a solution.
This yields the quadratic

λ2 + λ − 2 = 0

(which can be written directly by looking at the


constant-coefficient ODE).

Its roots are 1 and −2.

Thus the general solution is

bn = α + β (−2)n ,

where α and β are arbitrary.

The general solution to the original recursion is

1 (−2)n
an = α + β .
n! n!
So the general solution to the ODE is

y(x) = αex + βe−2x .

32
4 Inner product spaces

4.1 Vector spaces

Recall the notion of a vector space V over R.


Elements of V are called vectors, and

elements of R are called scalars.

There are two operations in a vector space, namely,

• addition
v + w, v, w ∈ V,
and

• scalar multiplication

cv, c ∈ R, v ∈ V.

Any vector space V has a dimension, which may not


be finite.

33
4.2 Inner product spaces

Let V be a vector space over R (not necessarily


finite-dimensional).

A bilinear form on V is a map

h , i : V × V → R,

which is linear in both coordinates, that is,

hau + v, wi = ahu, wi + hv, wi

hu, av + wi = ahu, vi + hu, wi


for a ∈ R and u, v ∈ V .
An inner product on V is a bilinear form on V which is

• symmetric: hv, wi = hw, vi.


• positive definite: hv, vi ≥ 0 for all v , and
hv, vi = 0 iff v = 0.
A vector space with an inner product is called an inner
product space.

34
4.3 Orthogonality

In an inner product space, we have the notion of


orthogonality.

We say vectors u and v are orthogonal if hu, vi = 0.


More generally, a set of vectors forms an orthogonal
system if they are mutually orthogonal.

An orthogonal basis is an orthogonal system which is


also a basis.

35
Example. Consider the vector space Rn .

Vectors are given by n tuples (a1 , . . . , an ).

Addition and scalar multiplication are done


coordinatewise. That is,

(a1 , . . . , an )+(b1 , . . . , bn ) := (a1 +b1 , . . . , an +bn )

c(a1 , . . . , an ) := (ca1 , . . . , can ).


The rule
n
X
h(a1 , . . . , an ), (b1 , . . . , bn )i := ai bi
i=1

defines an inner product on Rn .

Let ei be the vector which is 1 in the i-th position, and


0 in all other positions.
Then {e1 , . . . , en } is an orthogonal basis.

For example,

{(1, 0, 0), (0, 1, 0), (0, 0, 1)}

is an orthogonal basis of R3 .

36
The previous example can be formulated more
abstractly as follows.
Example. Let V be a finite-dimensional vector space
with basis {e1 , . . . , en }.
Pn Pn
For u = i=1 ai ei and v = i=1 bi ei , let
n
X
hu, vi := ai bi .
i=1

This defines an inner product on V .

Further, {e1 , . . . , en } is an orthogonal basis.

37
Lemma. Suppose V is a finite-dimensional inner
product space, and e1 , . . . , en is an orthogonal basis.
Then for any v ∈V,
n
X hv, ei i
v= ei .
i=1
hei , ei i
Pn
Proof. To see this, write v = i=1 ai ei .

We want to find the coefficients ai .

For this we take inner product of v with ei one by one:


n
X
hv, ej i = h ai ei , ej i
i=1
n
X
= ai hei , ej i
i=1
= aj hej , ej i.

Thus,
hv, ej i
aj =
hej , ej i
as required.

38
Lemma. In a finite-dimensional inner product space,
there always exists an orthogonal basis.

You can start with any basis and modify it to an


orthogonal basis by Gram-Schmidt orthogonalization.

This result is not necessarily true in infinite-dimensional


inner product spaces.

In this generality, we can only talk of a maximal


orthogonal set.

39
4.4 Length of a vector and Pythagoras
theorem

In an inner product space, we define for any v ∈V,

kvk := hv, vi1/2 .

This is called the norm or length of the vector v .

It verifies the following properties.

k0k = 0 and kvk > 0 if v 6= 0,

kv + wk ≤ kvk + kwk,
kavk = |a|kvk,
for all v, w ∈ V and a ∈ R.

40
Theorem. For orthogonal vectors v and w in any inner
product space,

kv + wk2 = kvk2 + kwk2 .

This is the modern avatar of Pythagoras theorem.

Proof. The proof is as follows.

kv + wk2 = hv + w, v + wi
= hv, vi + hv, wi + hw, vi + hw, wi
= hv, vi + hw, wi
= kvk2 + kwk2 .

More generally, for any orthogonal system


{v1 , . . . , vn },

kv1 + · · · + vn k2 = kv1 k2 + · · · + kvn k2 .

41
5 Legendre equation

Consider the following second order linear ODE.

(1 − x2 )y 00 − 2xy 0 + p(p + 1)y = 0.

Here p denotes a fixed real number.

This is known as the Legendre equation.

This ODE is defined for all real numbers x.

The Legendre equation can also be written in the form

((1 − x2 )y 0 )0 + p(p + 1)y = 0.

We assume p ≥ −1/2.

42
5.1 General solution

The coefficients

(1 − x2 ), −2x and p(p + 1)

are polynomials (and in particular real analytic).

However 1 − x2 = 0 for x = ±1.


The points x = ±1 are the singular points of the
Legendre equation.

Our theorem guarantees a power series solution of the


Legendre equation around x = 0 in the interval
(−1, 1).

43
We apply the power series method to the Legendre
equation

(1 − x2 )y 00 − 2xy 0 + p(p + 1)y = 0.

The recursion obtained is

(n+2)(n+1)an+2 −n(n−1)an −2nan +p(p+1)an = 0.

That is,

(n+2)(n+1)an+2 −[n(n−1)+2n−p(p+1)]an = 0.

That is,

(n+2)(n+1)an+2 −[n(n+1)−p(p+1)]an = 0.

That is,

(n − p)(n + p + 1)
an+2 = an .
(n + 2)(n + 1)
This is valid for n ≥ 0, with a0 and a1 arbitrary.

44
Thus, the general solution to the Legendre equation in
the interval (−1, 1) is given by

p(p+1) 2 (p(p−2)(p+1)(p+3) 4
y(x)=a0 (1− 2!
x + 4!
x +... )
(p−1)(p+2) 3 (p−1)(p−3)(p+2)(p+4) 5
+a1 (x− 3!
x + 5!
x +... ).

It is called the Legendre function.

The first series is an even function while the second


series is an odd function.

45
5.2 Legendre polynomials

Now suppose the parameter p in the Legendre


equation is a nonnegative integer.

Then one of the two series in the general solution


terminates, and is a polynomial.

Thus, for each nonnegative integer m, we obtain a


polynomial Pm (x) (up to multiplication by a constant).

It is traditional to normalize the constants so that


Pm (1) = 1.
These are called the Legendre polynomials.

The m-th Legendre polynomial Pm (x) solves the


Legendre equation

(1 − x2 )y 00 − 2xy 0 + m(m + 1)y = 0.

This solution is valid for all x, not just for x ∈ (−1, 1).

46
The first few values are as follows.

m Pm (x)
0 1
1 x
1
2 (3x2 − 1)
2
1
3 (5x3 − 3x)
2
1
4 (35x4 − 30x2 + 3)
8
1
5 (63x5 − 70x3 + 15x)
8

47
Their graphs in the interval (−1, 1) are given below.

48
5.3 Second solution to the Legendre
equation when p is an integer

Now let us consider the second independent solution.


It is a honest power series (not a polynomial).

For p = 0, it is
x3 x5 1 Ä1 + xä
x+ + + · · · = log .
3 5 2 1−x
For p = 1, it is
x2 x4 x6 1 Ä1 + xä
1− − − − · · · = 1 − x log .
1 3 5 2 1−x
These nonpolynomial solutions for any nonnegative
integer p always have a log factor of the above kind
and hence are unbounded near both +1 and −1.

(Since they are either even or odd, the behavior at


x = 1 is reflected at x = −1.)
They are called the Legendre functions of the second
kind.

49
5.4 The vector space of polynomials

The set of all polynomials in the variable x is a vector


space.

(We can add polynomials and multiply a polynomial by


a scalar.)

The set
{1, x, x2 , . . . }
is a basis of the vector space of polynomials.

Note: This vector space is not finite-dimensional.

Every polynomial is a ’vector’ in this vector space.

50
The vector space of polynomials carries an inner
product defined by
Z 1
hf, gi := f (x)g(x)dx.
−1

Note that we are integrating only between −1 and 1.


This ensures that the integral is always finite.

The norm of a polynomial is defined by


å1/2
1
ÇZ
kf k := f (x)f (x)dx .
−1

51
5.5 Technique of derivative transfer

Note this simple consequence of integration by parts:


Fact. For differentiable functions f and g , if

(f g)(b) = (f g)(a),

then
Z b Z b
f g 0 dx = − f 0 gdx.
a a

(This process transfers the derivative from g to f .)

52
5.6 Orthogonality of the Legendre
polynomials

Since Pm (x) is a polynomial of degree m, it follows


that
{P0 (x), P1 (x), P2 (x), . . . }
is a basis of the vector space of polynomials.
Proposition. We have

6= n,
Z 1 0 if m
Pm (x)Pn (x)dx =
−1  2 if m = n.
2n+1

The first alternative says that Legendre polynomials


form an orthogonal basis for the vector space of
polynomials.

The second alternative can be rewritten as


2
kPn (x)k2 = .
2n + 1

53
Proof. We will only prove orthogonality.

We use the technique of derivative-transfer.

Suppose m 6= n.
Since Pm (x) solves the Legendre equation for
p = m, we have
0 0
((1 − x2 )Pm ) + m(m + 1)Pm = 0.

Multiply by Pn and integrate to get


Z 1 Z 1
0 0
((1 − x2 )Pm ) Pn + m(m + 1) Pm Pn = 0.
−1 −1

By derivative transfer, we get


Z 1 Z 1
0
− (1 − x2 )Pm Pn0 + m(m + 1) Pm Pn = 0.
−1 −1

54
Interchanging the roles of m and n,
Z 1 Z 1
0
− (1 − x2 )Pm Pn0 + n(n + 1) Pm Pn = 0.
−1 −1

Subtracting the two identities, we obtain


Z 1
[m(m + 1) − n(n + 1)] Pm Pn = 0.
−1

Since m 6= n, the scalar in front can be canceled and


we get
Z 1
Pm Pn = 0.
−1
Thus, Pm and Pn are orthogonal.

55
5.7 Rodrigues formula

Consider the sequence of polynomials


Ä d än
qn (x) := (x2 −1)n = Dn (x2 −1)n , n ≥ 0.
dx
Observe that qn (x) has degree n.

For instance,

d 2
q1 (x) = (x − 1) = 2x.
dx

The first few polynomials are 1, 2x, 4(3x2 − 1), . . . .


[Note the similarity with the Legendre polynomials.]

56
Proposition. We have

1 Ä d än 2
Pn (x) = n (x − 1)n .
2 n! dx
Equivalently,

1
Pn (x) = n
qn (x).
2 n!
This is known as Rodrigues formula.

57
Proof. We sketch a proof of Rodrigues formula.

We first claim that



6= n,
Z 1 0 if m
qm (x)qn (x)dx =
−1  2 (2n n!)2 if m = n.
2n+1

This can be established using the technique of


derivative-transfer.

We prove orthogonality. That is,


Z 1
qm (x)qn (x)dx = 0 if m 6= n.
−1

Assume without loss of generality that m < n.

58
To understand the method, let us take m = 2 and
n = 5.
Z 1
D2 (x2 − 1)2 D5 (x2 − 1)5 dx
−1
Z 1
=− D3 (x2 − 1)2 D4 (x2 − 1)5 dx
−1
Z 1
=+ D4 (x2 − 1)2 D3 (x2 − 1)5 dx
−1
Z 1
=− D5 (x2 − 1)2 D2 (x2 − 1)5 dx
−1
= 0.

Thus, q2 and q5 are orthogonal.

59
Transfer m + 1 of the n derivatives in qn (x) to qm (x).
Z 1
Dm (x2 − 1)m Dn (x2 − 1)n dx
−1
Z 1
=− Dm+1 (x2 − 1)m Dn−1 (x2 − 1)n dx
−1
Z 1
= · · · = (−1)m D2m (x2 −1)m Dn−m (x2 −1)n dx
−1
Z 1
= (−1)m+1 D2m+1 (x2 −1)m Dn−m−1 (x2 −1)n dx = 0,
−1

since D 2m+1 (x2 − 1)m ≡ 0.


Thus, qm and qn are orthogonal for m < n.

60
We deduce that Pn (x) and qn (x) are scalar multiples
of each other.

By convention Pn (1) = 1 while qn (1) = 2n n!. Why?


So Rodrigues formula is proved.

61
5.8 Square-integrable functions

A function f (x) on [−1, 1] is square-integrable if


Z 1
f (x)f (x)dx < ∞.
−1

For instance, polynomials, continuous functions,


piecewise continuous functions are square-integrable.

The set of all square-integrable functions on [−1, 1] is


a vector space.

The inner product on polynomials extends to


square-integrable functions. That is, for
square-integrable functions f and g , we define their
inner product by
Z 1
hf, gi := f (x)g(x)dx.
−1

62
5.9 Fourier-Legendre series

The Legendre polynomials no longer form a basis for


the vector space of square-integrable functions.

But they form a maximal orthogonal set in this larger


space.

This means that there is no nonzero square-integrable


function which is orthogonal to all Legendre
polynomials.

This is a nontrivial fact.

63
This allows us to expand any square-integrable function
f (x) on [−1, 1] in a series of Legendre polynomials
X
cn Pn (x),
n≥0

where

hf, Pn i
cn =
hPn , Pn i
Z 1
2n + 1
= f (x)Pn (x)dx.
2 −1

This is called the Fourier-Legendre series (or simply


the Legendre series) of f (x).

64
Example. Consider the function

1 if 0 < x < 1,
f (x) =
−1 if − 1 < x < 0.

The Legendre series of f (x) is

3 7 11
P1 (x) − P3 (x) + P5 (x) − . . . .
2 8 16
By the Legendre expansion theorem (stated below),
this series converges to f (x) for x 6= 0 and to 0 for
x = 0.

65
Let us go back to the general case.

The Fourier-Legendre series of f (x) converges in


norm to f (x), that is,
m
X
kf (x) − cn Pn (x)k → 0 as m → ∞.
n=0

Pointwise convergence is more delicate.

There are two issues here:

Does the series converge at x?

If yes, then does it converge to f (x)?

66
A useful result in this direction is the Legendre
expansion theorem:
Theorem. If both f (x) and f 0 (x) have at most a finite
number of jump discontinuities in the interval [−1, 1],

then the Legendre series converges to

1
(f (x− ) + f (x+ )) for − 1 < x < 1,
2
converges to f (−1+ ) at x = −1, and
converges to f (1− ) at x = 1.
In particular, the series converges to f (x) at every
point of continuity.

67
6 Ordinary and singular points

Consider the second-order linear ODE

p(x)y 00 + q(x)y 0 + r(x)y = 0,

where p, q , and r are real analytic functions with no


common zeroes.

A point x0 is called an ordinary point if p(x0 ) 6= 0, and


a singular point if p(x0 ) = 0.

68
A singular point x0 is called regular if after dividing
throughout by p(x) the ODE can be written in the form

00 b(x) 0 c(x)
y + y + 2
y = 0,
x − x0 (x − x0 )
where b(x) and c(x) are real analytic around x0 .

A singular point which is not regular is called irregular.

For instance, in the equation x3 y 00 + xy 0 + y = 0,


x = 0 is an irregular singular point.

69
7 Cauchy-Euler equation

Consider the Cauchy-Euler equation

x2 y 00 + b0 xy 0 + c0 y = 0,

where b0 and c0 are constants.

x = 0 is a regular singular point since we can rewrite


the ODE as
b0 0 c0
y 00 + y + 2 y = 0.
x x
The rest are all ordinary points.

Assume x > 0.
Note that y = xr solves the equation if

r(r − 1) + b0 r + c0 = 0,

that is,
r2 + (b0 − 1)r + c0 = 0.

70
Let r1 and r2 denote the roots of this quadratic
equation.

• If the roots are real and unequal, then

xr1 and xr2

are two independent solutions.

• If the roots are real and equal, then

xr1 and (log x)xr1

are two independent solutions.

• If the roots are complex (written as a ± ib), then

xa cos(b log x) and xa sin(b log x)

are two independent solutions.

This gives us a general idea of the behavior of the


solution of an ODE near a regular singular point.

71
8 Fuchs-Frobenius theory

Consider the ODE

x2 y 00 + xb(x) y 0 + c(x) y = 0,

where
X X
n
b(x) = bn x and c(x) = cn xn
n≥0 n≥0

are real analytic in a neighborhood of the origin.

Restrict to x > 0.
Assume a solution of the form
X
r
y(x) = x an xn , a0 6= 0,
n≥0

with r fixed.

72
Substitute in the ODE and equate the coefficient of xr
to obtain
r(r − 1) + b0 r + c0 = 0,
that is,
r2 + (b0 − 1)r + c0 = 0.
Let us denote the quadratic by I(r).

It is called the indicial equation of the ODE.

Let r1 and r2 be the roots of I(r) = 0 with r1 ≥ r2 .

73
Equating the coefficient of xr+1 , we obtain

[(r + 1)r + (r + 1)b0 + c0 ]a1 + a0 (rb1 + c1 ) = 0.

More generally, equating the coefficient of xr+n , we


obtain the recursion
n−1
X
I(r+n)an + aj ((r+j)bn−j +cn−j ) = 0, n ≥ 1.
j=0

This is a recursion for the an ’s.

One can solve it provided I(r + n) 6= 0 for any n.

74
Theorem. The ODE has as a solution for x >0
X
r1
y1 (x) = x (1 + an xn )
n≥1

where an ’s solve the recursion for r = r1 and a0 = 1.


The power series converges in the interval in which
both b(x) and c(x) converge.

The term fractional power series solution is used for


such a solution.

75
8.1 Roots not differing by an integer

Recall that r1 and r2 are the roots of I(r) = 0 with


r1 ≥ r2 .
Theorem. If r1 − r2 is not an integer, then a second
independent solution for x > 0 is given by
X
r2
y2 (x) = x (1 + An xn )
n≥1

where An ’s solve the recursion for r = r2 , a0 = 1.


The power series in these solutions converge in the
interval in which both b(x) and c(x) converge.

This solution is also a fractional power series.

76
Example. Consider the ODE

2x2 y 00 − xy 0 + (1 + x)y = 0.

Observe that x = 0 is a regular singular point since


the ODE can be written as
1 0 (1 + x)
y 00 − y + 2
y = 0.
2x 2x
(The functions b(x) = −1/2 and c(x) = (1 + x)/2
are in fact polynomials, so they converge everywhere.)

Let us apply the Frobenius method.

The indicial equation is

2 3r 1
r − + = 0.
2 2
The roots are r1 = 1 and r2 = 1/2.
Their difference is not an integer.

So we will get two fractional power series solutions


converging everywhere.

Let us write them down explicitly.

77
The general recursion is

(2(r+n)(r+n−1)−(r+n)+1)an +an−1 = 0, n ≥ 1.

That is,
−1
an = an−1 , n ≥ 1
(r + n − 1)(2r + 2n − 1)
For the root r = 1, the recursion simplifies to
−1
an = an−1 , n ≥ 1
(2n + 1)n
leading to the solution for x >0
Ç å
P (−1)n xn
y1 (x)=x 1+ n≥1 (2n+1)n(2n−1)(n−1)...(5·2)(3·1)
.

Similarly, for the root r = 1/2, the recursion simplifies


to
−1
an = an−1 , n ≥ 1
2n(n − 1/2)
leading to the second solution for x > 0
Ç å
(−1)n xn
y2 (x)=x1/2
P
1+ n≥1 2n(n−1/2)(2n−2)(n−3/2)...(4·3/2)(2·1/2)
.

78
8.2 Repeated roots

Theorem. If the indicial equation has repeated roots,


then there is a second solution of the form
X
r1
y2 (x) = y1 (x) log x + x An xn
n≥1

with y1 (x) as before.

The power series converges in the interval in which


both b(x) and c(x) converge.

Let us see where this solution comes from.

In doing so, we will also derive a formula for An .

79
Treating r as a variable, one can uniquely solve

n−1
X
I(r+n)an + aj ((r+j)bn−j +cn−j ) = 0, n ≥ 1
j=0

starting with a0 = 1.
Since the an depend on r , let us write an (r).

Now consider
Ä X ä
r n
ϕ(r, x) := x 1 + an (r)x .
n≥1

Note that the first solution is ϕ(r1 , x).

80
For the second solution, take partial derivative of
ϕ(r, x) with respect to r, and then put r = r1 .

∂ϕ(r, x)
y2 (x) =
∂r r=r
1
Ç å
∂ X
= xr an (r)xn

∂r n≥0

r=r1

a0n (r1 )xn


X X
r1 n r1
= x log x an (r1 )x + x
n≥0 n≥1

a0n (r1 )xn .


X
r1
= y1 (x) log x + x
n≥1

In particular,
An = a0n (r1 ).

81
Example. Consider the ODE

x2 y 00 + 3xy 0 + (1 − 2x)y = 0.

This has a regular singularity at 0 with b(x) = 3 and


c(x) = 1 − 2x.
The indicial equation is r 2 + 2r + 1 = 0 which has a
repeated root at −1.

One may check that the recursion is

(n + r + 1)2 an = 2an−1 , n ≥ 1.

Hence
2n
an (r) = 2
a0 .
[(r + 2)(r + 3) . . . (r + n + 1)]
Setting r = −1 (and a0 = 1) yields the fractional
power series solution

1 X 2n n
y1 (x) = 2
x .
x n≥0 (n!)

The power series converges everywhere.

82
For the second solution:
1 1 1
a0n (r)
Ä ä
= −2an (r) + +· · ·+ , n ≥ 1.
r+2 r+3 r+n+1
Evaluating at r = −1,
2n+1 Hn
a0n (−1) =− ,
(n!)2
where
1 1
Hn = 1 + + ··· + .
2 n
(These are the partial sums of the harmonic series.)

So the second solution is

1 X 2n+1 Hn n
y2 (x) = y1 (x) log x − 2
x .
x n≥1 (n!)

The power series converges everywhere.

83
8.3 Roots differing by an integer

Theorem. If r1 − r2 is a positive integer, say N , then


there is a second solution of the form
X
r2
y2 (x) = Ky1 (x) log x + x (1 + An xn ),
n≥1

with
d
An = ((r − r2 )an (r)) , n ≥ 1.

dr r=r2

and
K = lim (r − r2 )aN (r).
r→r2
The power series converges in the interval in which
both b(x) and c(x) converge.

It is possible that K = 0.

84
Example. Consider the ODE

xy 00 − (4 + x)y 0 + 2y = 0.

The indicial equation is r(r − 5) = 0, with the roots


differing by a positive integer.

The recursion is

(n + r)(n + r − 5)an = (n + r − 3)an−1 , n ≥ 1.

Hence

(n + r − 3) . . . (r − 2)
an (r) = a0 .
(n + r) . . . (1 + r)(n + r − 5) . . . (r − 4)
Setting r = 5 (and a0 = 1) yields the fractional power
series solution
X 60
y1 (x) = xn+5 .
n≥0
n!(n + 5)(n + 4)(n + 3)

85
For the second solution, the ‘critical’ function is

(r + 2)(r + 1)r(r − 1)(r − 2)


a5 (r) = .
(r + 5) . . . (r + 1)r(r − 1) . . . (r − 4)
Note that there is a factor of r in the numerator also.

So this function does not have a singularity at r = 0,


and K = 0, and a5 (0) = 1/720.
Ä ä
1 2
y2 (x)=(1+ 12 x+ 12 x5 + 60
xn
P
x )+a5 (0) n≥6 (n−5)!n(n−1)(n−2)
.

Here the first solution y1 (x) is clearly visible.

Thus,
1 1 2
1+ x+ x
2 12
also solves the ODE.

86
8.4 Summary

While solving an ODE around a regular singular point


by the Frobenius method, the cases encountered are

• roots not differing by an integer

• repeated roots

• roots differing by a positive integer with no log term

• roots differing by a positive integer with log term

The larger root always yields a fractional power series


solution.

In the first and third cases, the smaller root also yields
a fractional power series solution.

In the second and fourth cases, the second solution


involves a log term.

87
9 Gamma function

Define for all p > 0,


Z ∞
Γ(p) := tp−1 e−t dt.
0

There is a problem at p = 0 since 1/t is not integrable


in an interval containing 0.

The same problem persists for p < 0.


For large values of p, there is no problem because e−t
is rapidly decreasing.

88
Z ∞
Γ(1) = e−t dt = 1.
0
For any integer n ≥ 1,
Z x
Γ(n + 1) = lim tn e−t dt
x→∞ 0
Ç Z x å
=n lim tn−1 e−t dt
x→∞ 0

= nΓ(n).

This yields
Γ(n) = (n − 1)!.
Thus the gamma function extends the factorial function
to all positive real numbers.

89
The above calculation is valid for any real p > 0, so

Γ(p + 1) = p Γ(p).

We use this identity to extend the gamma function to all


real numbers except 0 and the negative integers:

First extend it to the interval (−1, 0), then to


(−2, −1), and so on.
The graph is shown below.

90
Though the gamma function is now defined for all real
numbers (except the nonpositive integers), remember
that the integral representation is valid only for p > 0.
It is useful to rewrite
1 p
= .
Γ(p) Γ(p + 1)
This holds for all p if we impose the natural condition
that the reciprocal of Γ evaluated at a nonpositive
integer is 0.

91
A well-known value of the gamma function at a
non-integer point is
Z ∞ Z ∞
−1/2 −t −s2 √
Γ(1/2) = t e dt = 2 e ds = π.
0 0

(We used the substitution t = s2 .) By translating,


4√
Γ(−3/2) = 3 π ≈ 2.363

Γ(−1/2) = −2 π ≈ −3.545
1√
Γ(3/2) =2 π ≈ 0.886
3√
Γ(5/2) =4 π ≈ 1.329
15 √
Γ(7/2) = 8 π ≈ 3.323.

92
10 Bessel functions

Consider the second-order linear ODE

x2 y 00 + xy 0 + (x2 − p2 )y = 0.

This is known as the Bessel equation.

Here p denotes a fixed real number.

We may assume p ≥ 0.
There is a regular singularity at x = 0.
All other points are ordinary.

93
10.1 Bessel functions of the first kind

We apply the Frobenius method to find the solutions.

In previous notation, b(x) = 1 and c(x) = x2 − p2 .


The indicial equation is

r2 − p2 = 0.

The roots are r1 = p and r2 = −p.


The recursion is

(r + n + p)(r + n − p)an + an−2 = 0, n ≥ 2,

and a1 = 0.
So all odd terms are 0.

94
Let us solve this recursion with r as a variable.

The even terms are


1
a2 (r) = − 2 2
a0 ,
(r + 2) − p
1
a4 (r) = 2 2 2 2
a0 ,
((r + 2) − p )((r + 4) − p )
and in general

(−1)n
a2n (r) = a0 .
((r + 2)2 − p2 )((r + 4)2 − p2 ) . . . ((r + 2n)2 − p2 )

95
The fractional power series solution for the larger root
r1 = p obtained by setting a0 = 1 and r = p is
(−1)n
(x)=xp x2n
P
y1 n≥0 ((p+2)2 −p2 )((p+4)2 −p2 )...((p+2n)2 −p2 )

(−1)n
=xp x2n .
P
n≥0 22n n!(1+p)...(n+p)

The power series converges everywhere.


1
Multiplying by 2p Γ(1+p) , we define

(−1)n
Å ãp X Å ã2n
x x
Jp (x) := , x > 0.
2 n≥0
n! Γ(n + p + 1) 2

This is called the Bessel function of the first kind of


order p.

It is a solution of the Bessel equation.

96
The Bessel function of order 0 is

x2 x4 x6
J0 (x) = 1 − 2 + 2 2 − 2 2 2 + . . .
2 2 4 2 4 6
Å ã2 Å ã4 Å ã6
x 1 x 1 x
=1− + − + ...
2 2!2! 2 3!3! 2
This is similar to cos x.

The Bessel function of order 1 is


Å ã3 Å ã5
x 1 x 1 x
J1 (x) = − + + ....
2 1!2! 2 2!3! 2
This is similar to sin x.

97
Both J0 (x) and J1 (x) have a damped oscillatory
behavior having an infinite number of zeroes, and
these zeroes occur alternately.

Further, they satisfy derivative identities similar to


cos x and sin x.

J00 (x) = −J1 (x) and [xJ1 (x)]0 = xJ0 (x).

These functions are real analytic representable by a


single power series centered at 0.

98
10.2 Second independent solution

Observe that r1 − r2 = 2p.


The analysis to get a second independent solution of
the Bessel equation splits into four cases.

• 2p is not an integer:
We get a second fractional power series solution.

• p = 0:
We get a log term.

• p is a positive half-integer 1/2, 3/2, 5/2, . . . :


We get a second fractional power series solution.

• p is a positive integer:
We get a log term.

We discuss the first three cases.

99
Case : Suppose 2p is not an integer.

Solving the recursion uniquely for a0 = 1 and


r = −p, we obtain

−p
X (−1)n 2n
y2 (x) = x 2n
x .
n≥0
2 n!(1 − p) . . . (n − p)

1
Normalizing by 2−p Γ(1−p) , define

Å ã−p X
(−1)n
Å ã2n
x x
J−p (x) := , x > 0.
2 n≥0
n! Γ(n − p + 1) 2

This is a second solution of the Bessel equation


linearly independent of Jp (x).

It is clearly unbounded near x = 0, behaving like x−p


as x approaches 0.

100
Case : Suppose p is a positive half-integer.

The same solution as in the previous case works in the


half-integer case because N = 2p is an odd integer,
and the recursion only involves the even coefficients
a0 , a2 , a4 , etc.
Remark. J−p (x) make sense for all p. However,
when p is an integer, J−p (x) = (−1)p Jp (x).

101
Case : Suppose p = 0.
In this case,
(−1)n
a2n (r) = 2 2 2
,
(r + 2) (r + 4) . . . (r + 2n)
and an (r) = 0 for n odd. The first solution is a power
series with coefficients an (0):
X (−1)n
2n
y1 (x) = J0 (x) = 2n 2
x , x > 0.
n≥0
2 (n!)

Differentiating the a2n (r) with respect to r , we get


1 1 1 ä
a02n (r)
Ä
= −2a2n (r) + +· · ·+ .
r+2 r+4 r + 2n
Now setting r = 0, we obtain

(−1)n−1 Hn 1 1
a02n (0) = 2n 2
, Hn = 1+ +· · ·+ .
2 (n!) 2 n
Thus, the second solution is
X (−1)n Hn
2n
y2 (x) = J0 (x) log x− 2n 2
x , x > 0.
n≥1
2 (n!)

102
10.3 Summary of p = 0 and p = 1/2

For p = 0, we found two independent solutions.


The first function J0 (x) is a real analytic function for all
R, while the second function has a logarithmic
singularity at 0.

For p= 1/2, two independent solutions are J1/2 (x)


and J−1/2 (x).

These can be expressed in terms of the trigonometric


functions:
   
2 2
J 1 (x) = sin x and J− 1 (x) = cos x.
2 πx 2 πx
Both exhibit singular behavior at 0.

The first function is bounded near 0 but does not have


a bounded derivative near 0, while the second function
is unbounded near 0.

103

The substitution u(x) = x y(x) transforms the
Bessel equation into

1 − 4p 2ä
u00 + 1 +
Ä
2
u = 0.
4x
For p = 1/2, this specializes to

u00 + u = 0,

whose solutions are sin x and cos x.

104
10.4 Bessel identities

Proposition. For any real number p,

d p
[x Jp (x)] = xp Jp−1 (x)
dx
d î −p
x Jp (x) = −x−p Jp+1 (x)
ó
dx
2p
Jp−1 (x) + Jp+1 (x) = Jp (x)
x
Jp−1 (x) − Jp+1 (x) = 2Jp0 (x)

The first two identities can be directly established by


manipulating the respective power series.

These can then be used to prove the next two


identities.

105
We prove the first identity.

Recall

(−1)n
Å ãp X Å ã2n
x x
Jp (x) = .
2 n≥0
n! Γ(n + p + 1) 2

Ñ é0
(−1)n
Å ã2n+2p
p 0 p
X x
(x Jp (x)) = 2
n≥0
n! Γ(n + p + 1) 2
X (−1)n (2n + 2p) 1 Å x ã2n+2p−1
= 2p
n≥0
n! Γ(n + p + 1) 2 2
(−1)n
Å ã2n+2p−1
p
X x
=2
n≥0
n! Γ(n + p) 2
(−1)n
Å ãp−1 X Å ã2n
p x x
=x
2 n≥0
n! Γ(n + p) 2
= xp Jp−1 (x).

106
Using the first two identities:

Jp−1 (x) + Jp+1 (x) = x−p [xp Jp (x)]0 − xp [x−p Jp (x)]0


p p
= Jp0 (x) + Jp (x) − [Jp0 (x) − Jp (x)]
x x
2p
= Jp (x).
x

107
The identity

2p
Jp−1 (x) + Jp+1 (x) = Jp (x)
x
can be thought of as a recursion in p.

In other words, Jp+n (x) can be computed


algorithmically from Jp (x) and Jp+1 (x) for all integer
n.
Recall that we had nice formulas for J 1 (x) and
2
J −1 (x).
2

Using them, we get formulas for Jp (x) for half-integer


values of p.

108
For instance:
 
1 2  sin x 
J 3 (x) = J 1 (x)−J− 1 (x) = −cos x
2 x 2 2 πx x
 
1 2  cos x 
J −3 (x) = − J− 1 (x)−J 1 (x) = − +sin x
2 x 2 2 πx x
 
3 2  3 sin x 3 cos x 
J 5 (x) = J 3 (x)−J 1 (x) = 2
− −sin x
2 x 2 2 πx x x

109
10.5 Zeroes of Bessel function

Fix p ≥ 0.
Let Z (p) denote the set of zeroes of Jp (x).
Fact. The set of zeroes is a sequence increasing to
infinity.

Let x1 and x2 be successive positive zeroes of Jp (x).

• If 0 ≤ p < 1/2, then x2 − x1 is less than π and


approaches π as x1 → ∞.

• If p = 1/2, then x2 − x1 = π .

• If p > 1/2, then x2 − x1 is greater than π and


approaches π as x1 → ∞.

This fact can be proved using the Sturm Comparison


theorem.

110
The first few zeroes are tabulated below.

J0 (x) J1 (x) J2 (x) J3 (x) J4 (x) J


1 2.4048 3.8317 5.1356 6.3802 7.5883 8
2 5.5201 7.0156 8.4172 9.7610 11.0647 12
3 8.6537 10.1735 11.6198 13.0152 14.3725 15
4 11.7915 13.3237 14.7960 16.2235 17.6160 18
5 14.9309 16.4706 17.9598 19.4094 20.8269 22

111
10.6 Scaled Bessel equation

Consider the second-order linear ODE

x2 y 00 + xy 0 + (a2 x2 − p2 )y = 0.

This is known as the scaled Bessel equation.

The usual Bessel equation has been scaled by the


scalar a.

The function Jp (ax) solves this equation.

It is called the scaled Bessel function.

112
10.7 Orthogonality

Define an inner product on square-integrable functions


on [0, 1] by
Z 1
hf, gi := xf (x)g(x)dx.
0

This is similar to the previous inner product except that


f (x)g(x) is now multiplied by x, and the interval of
integration is from 0 to 1.

The multiplying factor x is called a weight function.

113
Fix p ≥ 0.
The set of scaled Bessel functions

{Jp (zx) | z ∈ Z (p) }

indexed by the zero set Z (p) form an orthogonal family:


Proposition. If k and ` are any two positive zeroes of
the Bessel function Jp (x), then

Z 1  1 [J 0 (k)]2 if k = `,
2 p
xJp (kx)Jp (`x)dx =
0 0 if k 6= `.

In this result, p is fixed.

114
Proof. We prove orthogonality.

For any positive constants a and b, the functions


u(x) = Jp (ax) and v(x) = Jp (bx) satisfy the
scaled Bessel equations

1 p2ä 1 p 2ä
u00 + u0 + a2 − 2 u = 0 and v 00 + v 0 + b2 − 2 v = 0
Ä Ä
x x x x
respectively.

Multiplying the first equation by v , the second by u and


subtracting,

1 0
(u0 v − v 0 u)0 + (u v − v 0 u) = (b2 − a2 )uv.
x
Multiplying by x, we obtain

[x(u0 v − v 0 u)]0 = (b2 − a2 )xuv.

115
Integrating from 0 to 1, we get
Z 1
(b2 − a2 ) xuv dx = u0 (1)v(1) − v 0 (1)u(1).
0

Suppose a = k and b = ` are distinct zeroes of


Jp (x).
Then u(1) = v(1) = 0, so the rhs is zero.
Further, b2 − a2 6= 0, so the integral in the lhs is zero.
This proves orthogonality.

116
10.8 Fourier-Bessel series

Fix p ≥ 0.
Any square-integrable function f (x) on [0, 1] can be
expanded in a series of scaled Bessel functions
Jp (zx) as
X
cz Jp (zx),
z∈Z (p)
where
Z 1
2
cz = 0 x f (x) Jp (zx) dx.
[Jp (z)]2 0

This is called the Fourier-Bessel series of f (x) for the


parameter p.

117
Example. Let us compute the Fourier-Bessel series
(for p
= 0) of the function f (x) = 1 in the interval
0 ≤ x ≤ 1.
Z 1
1 1 J1 (z)
x J0 (zx) dx = x J1 (zx) = ,

0 z 0 z
so
2
cz = .
zJ1 (z)
Thus, the Fourier-Bessel series is
X 2
J0 (zx).
zJ1 (z)
z∈Z (0)

By theorem below, this converges to 1 for all


0 < x < 1.

118
The Fourier-Bessel series of f (x) converges to f (x)
in norm.

For pointwise convergence, we have the Bessel


expansion theorem:
Theorem. If both f (x) and f 0 (x) have at most a
finite number of jump discontinuities in the interval
[0, 1], then the Bessel series converges to
1
(f (x− ) + f (x+ ))
2
for 0 < x < 1.

119
11 Fourier series

11.1 Orthogonality of the trigonometric family

Consider the space of square-integrable functions on


[−π, π].
Define an inner product by
Z π
1
hf, gi := f (x)g(x)dx.
2π −π

The norm of a function is then given by


Ç Z π å1/2
1
kf k = f (x)f (x)dx .
2π −π

120
Proposition. The set {1, cos nx, sin nx}n≥1 is a
maximal orthogonal set with respect to this inner
product.

Explicitly,
h1, 1i = 1.

0 if m 6= n,
hcos mx, cos nxi =
1/2 if m = n.

0 if m 6= n,
hsin mx, sin nxi =
1/2 if m = n.
hsin mx, cos nxi = h1, cos nxi = h1, sin mxi = 0.

This can be proved by a direct calculation.

121
11.2 Fourier series

Any square-integrable function f (x) on [−π, π] can


be expanded in a series of the trigonometric functions

X
a0 + (an cos nx + bn sin nx)
n=1

where the an and bn are given by


Z π
1
a0 = f (x)dx,
2π −π

1 π
Z
an = f (x) cos nxdx, n ≥ 1,
π −π
Z π
1
bn = f (x) sin nx dx, n ≥ 1.
π −π
This is called the Fourier series of f (x).

The an and bn are called the Fourier coefficients.

The above formulas are sometimes called the Euler


formulas.

The Fourier series of f (x) converges to f (x) in norm.

122
11.3 Pythagoras theorem or Parseval’s
identity

Suppose V is a finite-dimensional inner product space


and {v1 , . . . , vk } is an orthogonal basis.
Pk 2 Pk 2 kv k2 .
If v = i=1 a i v i , then kvk = a
i=1 i i

This is the Pythagoras theorem.

There is an infinite-dimensional analogue which says


that the square of the norm of f is the sum of the
squares of the norms of its components with respect to
any maximal orthogonal set.

Thus, we have

2 1X 2
kf k = a20 + (an + b2n ).
2 n≥1

This is known as the Parseval identity.

123
11.4 Pointwise convergence

A function f : R → R is periodic (of period 2π ) if

f (x + 2π) = f (x) for all x.

Theorem. Let f (x) be a periodic function of period


2π which is integrable on [−π, π]. Then at a point x, if
the left and right derivative exist, then the Fourier
series of f converges to

1
[f (x+ ) + f (x− )].
2

124
Example. Consider the function

1 if 0 < x < π,
f (x) =
−1 if − π < x < 0.
The value at 0, π and −π is left unspecified. Its
periodic extension is the square-wave.

Since f is an odd function, a0 and all the an are zero.


The bn for n ≥ 1 can be calculated as follows.
Z π
1
bn = f (x) sin nx dx
π −π
Z π
2
= sin nx dx
π 0
2
= (1 − cos nπ)


 4 if n is odd,

=
0 if n is even.

Thus the Fourier series of f (x) is

4Ä sin 3x sin 5x ä
sin x + + + ... .
π 3 5

125
This series converges to f (x) at all points except
integer multiples of π where it converges to 0.

The partial sums of the Fourier series wiggle around


the square wave.

126
In particular, evaluating at x = π/2,
π 4Ä 1 1 1 ä
f( ) = 1 = 1 − + − + ... .
2 π 3 5 7
Rewriting,

1 1 1 π
1 − + − + ··· = .
3 5 7 4

127
Example. Consider the function

f (x) = x2 , −π ≤ x ≤ π.

Since f is an even function, the bn are zero.

1
Z π
π 2
a0 = x2 dx = .
2π −π 3

1 π
Z
4  42 if n is even,
2
an = x cos nx dx = 2 cos nπ = n
π −π n − 4 if n is odd.
n2

Thus the Fourier series of f (x) is

π2 Ä cos 2x cos 3x ä
− 4 cos x − + − ... .
3 4 9
This series converges to f (x) at all points.

128
Evaluating at x = π,
π 2 Ä 1 1
2
π = + 4 1 + + + . . . ).
3 4 9
This yields the identity

X 1 π2
2
= .
n≥1
n 6

129
11.5 Fourier sine and cosine series

If f is an odd (even) function, then its Fourier series


has only sine (cosine) terms.

This allows us to do something interesting.

Suppose f is defined on the interval (0, π).

Then we can extend it as an odd function on (−π, π)


and expand it in a Fourier sine series,

or extend it as an even function on (−π, π) and


expand it in a Fourier cosine series.

130
For instance, consider the function

f (x) = x, 0 < x < π.

The Fourier sine series of f (x) is


Ä sin 2x sin 3x ä
2 sin x − + − ...
2 3
while the Fourier cosine series of f (x) is

π 4 Ä cos x cos 3x ä
− 2
+ 2
+ ...
2 π 1 3
The two series are equal on 0 <x<π
(but different on −π < x < 0).

131
The Fourier cosine series above is the same as the
Fourier series of g(x)= |x|. Note that g 0 (x) equals
the square wave function f (x) discussed before and
the Fourier series of the square wave is precisely the
term-by-term derivative of the Fourier series of g(x).

This is a general fact which can be seen by applying


derivative transfer on the Euler formulas.

For instance,
Z π Z π
f 0 (x) sin nxdx = − f (x)(sin nx)0 dx
−π −π
Z π
= −n f (x) cos nxdx
−π

shows that the bn of f 0 (x) are related to the an of


f (x).

132
11.6 Fourier series for arbitrary periodic
functions

One can also consider Fourier series for functions of


any period not necessarily 2π .

Suppose the period is 2`.

Then the Fourier series is of the form



X nπx nπx
a0 + an cos + bn sin .
n=1
` `

The Fourier coefficients are given by


Z `
1 1 ` nπx
Z
a0 = f (x)dx, an = f (x) cos dx, a
2` −` ` −` `
1 ` nπx
Z
bn = f (x) sin dx, n ≥ 1.
` −` `
By scaling the independent variable, one can transform
the given periodic function to a 2π -periodic function,
and then apply the standard theory.

133
12 One-dimensional heat equation

The one-dimensional heat equation is given by

ut = kuxx , 0 < x < `, t > 0,

where k is a fixed positive constant.

There are two variables, x is the space variable and t


is the time variable.

This PDE describes the temperature evolution of a thin


rod of length `.

More precisely, u(x, t) is the temperature at point x at


time t.

134
The temperature at t = 0 is specified.
This is the initial condition.

We write it as
u(x, 0) = u0 (x).
In addition to the initial condition, there are conditions
specified at the two endpoints of the rod.

These are the boundary conditions.

135
We consider four different kinds of boundary conditions
one by one.

In each case, we employ the method of separation of


variables:

Suppose u(x, t) = X(x)T (t).


Substituting this in the PDE, we get

T 0 (t)X(x) = kX 00 (x)T (t).

Observe that we can now separate the variables:

X 00 (x) T 0 (t)
= = λ (say).
X(x) kT (t)
The equality is between a function of x and a function
of t, so both must be constant. We have denoted this
constant by λ.

136
12.1 Dirichlet boundary conditions

u(0, t) = u(`, t) = 0.
In other words, the endpoints of the rod are maintained
at temperature 0 at all times t.

(The rod is isolated from the surroundings except at the


endpoints from where heat will be lost to the
surroundings.)

We need to consider three cases:

137
1. λ > 0:
Write λ = µ2 with µ > 0.
X 00 (x) T 0 (t)
= = µ2 .
X(x) kT (t)
Then

µx −µx µ2 kt
X(x) = Ae +Be and T (t) = Ce .

Hence

µ2 kt
u(x, t) = e (Aeµx + Be−µx ),

where the constant C has been absorbed in A and


B.
The boundary conditions imply that A = 0 = B.
So there is no nontrivial solution of this form.

138
2. λ = 0:
In this case,

X 00 (x) T 0 (t)
= = 0.
X(x) kT (t)
Thus, we have X(x) = Ax + B and T (t) = C .
Hence
u(x, t) = Ax + B.
The boundary conditions give A = 0 = B.
Thus this case also does not yield a nontrivial
solution.

139
3. λ < 0:
Write λ = −µ2 with µ > 0.
X 00 (x) T 0 (t)
= = −µ2 .
X(x) kT (t)
Then

−µ2 kt
X(x) = A cos µx+B sin µx and T (t) = Ce .

Hence

−µ2 kt
u(x, t) = e [A cos µx + B sin µx].

The boundary condition at x = 0 yields A = 0,


and at x = ` yields

B sin µ` = 0.

Thus B = 0 unless µ = nπ/`, n = 1, 2, 3, . . . .


Hence
2 (π/`)2 kt nπx
un (x, t) = e−n sin , n = 1, 2, 3, . . .
`
are the nontrivial solutions.

140
The general solution is obtained by taking an infinite
linear combination of these solutions:

X
−n2 (π/`)2 kt nπx
u(x, t) = bn e sin .
n=1
`

The coefficients bn remain to be found.

For this we finally make use of the initial condition:



X nπx
u(x, 0) = u0 (x) = bn sin , 0 < x < `.
n=1
`

It is natural to let the rhs be the Fourier sine series of


u0 (x) over the interval (0, `).

141
Conclusion: The solution for Dirichlet boundary
conditions is

2 (π/`)2 kt nπx
bn e−n
X
u(x, t) = sin ,
n=1
`

where
Z `
2 nπx
bn = u0 (x) sin dx.
` 0 `
As t increases, the temperature of the rod rapidly
approaches 0 everywhere.

142
12.2 Neumann boundary conditions

ux (0, t) = 0 = ux (`, t).


In other words, there is no heat loss at the endpoints.
Thus the rod is completely isolated from the
surroundings.

As in the Dirichlet case, we need to consider three


cases:

143
1. λ > 0:
Write λ = µ2 with µ > 0.
Then

µx −µx µ2 kt
X(x) = Ae +Be and T (t) = Ce .

The boundary conditions imply that A = 0 = B.


So there is no nontrivial solution of this form.

144
2. λ = 0:
In this case we have X(x) = Ax + B and
T (t) = C .
Hence
u(x, t) = Ax + B.
The boundary conditions give A = 0.
Hence this case contributes the solution
u(x, t) = constant.

145
3. λ < 0:
Write λ = −µ2 with µ > 0.
It follows that

−µ2 kt
u(x, t) = e [A cos µx + B sin µx].

The boundary conditions now imply that B = 0.


Also A = 0 unless µ = nπ/`, n = 1, 2, 3, . . . .
Thus

−n2 (π/`)2 kt nπx


un (x, t) = e cos , n = 1, 2, 3, . . .
`
are the nontrivial solutions.

146
The general solution will now be of the form

X
−n2 (π/`)2 kt nπx
u(x, t) = a0 + an e cos .
n=1
`

The coefficients an remain to be determined.

For this we finally make use of the initial condition:



X nπx
u(x, 0) = u0 (x) = a0 + an cos , 0 < x < `.
n=1
`

Now we let the rhs must be the Fourier cosine series of


u0 (x) over the interval (0, `).

147
Conclusion: The solution for Neumann boundary
conditions is

X
−n2 (π/`)2 kt nπx
u(x, t) = a0 + an e cos ,
n=1
`

where
Z ` Z `
1 2 nπx
a0 = u0 (x)dx, an = u0 (x) cos dx.
` 0 ` 0 `

All terms except for the first one tend rapidly to zero as
t → ∞.
So one is left with a0 , which is the mean or average
value of u0 .

Physically, this means that an isolated rod will


eventually assume a constant temperature, which is
the mean of the initial temperature distribution.

148
12.3 Mixed boundary conditions

u(0, t) = 0 = ux (`, t).


Thus, the left endpoint is maintained at temperature 0
(so there will be heat loss from that end), while there is
no heat loss at the right endpoint.

The solution is
X
−(n+1/2)2 (π/`)2 kt (n + 1/2)πx
u(x, t) = bn e sin ,
n≥0
`

where
Z `
2 (n + 1/2)πx
bn = u0 (x) sin dx.
` 0 `

For the mixed boundary conditions

ux (0, t) = 0 = u(`, t),

the solution is similar with a cosine series instead.

149
12.4 Periodic boundary conditions

u(0, t) = u(`, t), ux (0, t) = ux (`, t).

The solution is

2 (π/`)2 kt
e−4n [an cos 2nπx +bn sin 2nπx
P
` ]
u(x,t)=a0 + n≥1 `
,

where
Z ` Z `
1 2 2nπx
a0 = u0 (x)dx, an = u0 (x) cos dx
` 0 ` 0 `
and
Z `
2 2nπx
bn = u0 (x) sin dx.
` 0 `

150
12.5 Dirichlet boundary conditions with heat
source

We now solve

ut − kuxx = f (x, t), 0 < x < `, t > 0,

with u(x, 0) = u0 specified, and with Dirichlet


boundary conditions.

For simplicity, we assume ` = 1.

151
Expand everything in a Fourier sine series over (0, 1):
X
u(x, t) = Yn (t) sin nπx,
n≥1
X
f (x, t) = Bn (t) sin nπx,
n≥1
X
u0 (x) = bn sin nπx,
n≥1

where the Bn (t) and the bn are known while the


Yn (t) are to be determined.
Substituting, we obtain
X X
2 2
[Ẏn (t)+kn π Yn (t)] sin nπx = Bn (t) sin nπx.
n≥1 n≥1

This implies that Yn (t) solves the ODE

Ẏn (t) + kn2 π 2 Yn (t) = Bn (t), Yn (0) = bn .

152
Example. Suppose

f (x, t) = sin πx sin πt and u0 (x) = 0.

Thus, the initial temperature of the rod is 0 everywhere


but there is a heat source.

Then bn = 0 for all n ≥ 1, and



0 for n 6= 1,
Bn (t) =
sin πt for n = 1.

Therefore, for n 6= 1,

Ẏn (t) + kn2 π 2 Yn (t) = 0, Yn (0) = 0

which implies Yn ≡ 0.
For n = 1,

Ẏ1 (t) + kπ 2 Y1 (t) = sin πt, Y1 (0) = 0.

Solving, we get
î 2
ó
1
u(x,t)= e−π kt −cos πt+kπ sin πt sin πx.
π(k π 2 +1)
2

153
13 One-dimensional wave equation

The one-dimensional wave equation is given by

utt = c2 uxx , 0 < x < `, t > 0,

where c is a fixed constant.

This describes the vibrations of a string of length `.

More precisely, u(x, t) represents the deflection from


the mean position of the string at position x at time t.

154
The initial position and velocity is specified.

These are the initial conditions.

We write them as

u(x, 0) = u0 (x) and ut (x, 0) = u1 (x).

In addition to the initial conditions, there are conditions


specified at the two endpoints of the string.

These are the boundary conditions.

155
Adopting the method of separation of variables, let
u(x, t) = X(x)T (t).
Substituting this in the PDE, we get

T 00 (t)X(x) = c2 X 00 (x)T (t).

Observe that we can now separate the variables:

X 00 (x) T 00 (t)
= 2 = λ (say).
X(x) c T (t)

156
13.1 Dirichlet boundary conditions

u(0, t) = u(`, t) = 0.

The boundary conditions can be rewritten as

X(0) = X(`) = 0.

As in the case of the heat equation we consider three


cases.

157
1. λ > 0: Write λ = µ2 with µ > 0. Then

X(x) = Aeµx + Be−µx .

The boundary conditions imply that A = 0 = B.


So there is no nontrivial solution of this form.

2. λ = 0: In this case we have X(x) = Ax + B .


Again the boundary conditions give A = 0 = B .
Thus this case also does not yield a nontrivial
solution.

158
3. λ < 0: Write λ = −µ2 with µ > 0.
X 00 (x) T 00 (t)
= 2 = −µ2 .
X(x) c T (t)
Then

X(x) = A cos µx + B sin µx

and
T (t) = C cos cµt + D sin cµt.
The boundary conditions now imply that A
= 0.
Also B = 0 unless µ = nπ/`, n = 1, 2, 3, . . . .
Thus
cnπt cnπt nπx
un (x, t) = [C cos +D sin ] sin , n = 1, 2, 3, . . .
` ` `
are the nontrivial solutions.

These are called the pure harmonics or simply the


harmonics or the modes of vibration.
cnπ
` is called the frequency of the n-th harmonic.

159
The general solution (ignoring initial conditions) is
X cnπt cnπt nπx
u(x, t) = [Cn cos +Dn sin ] sin .
n≥1
` ` `

Now from the initial conditions,


Z `
2 nπx
Cn = u0 (x) sin dx
` 0 `
and
Z `
2 nπx
Dn = u1 (x) sin dx.
cnπ 0 `
Thus the Fourier sine series of u0 (x) and u1 (x) enter
into the solution.

160
Conclusion: The solution is
X cnπt cnπt nπx
u(x, t) = [Cn cos +Dn sin ] sin ,
n≥1
` ` `

where
Z `
2 nπx
Cn = u0 (x) sin dx
` 0 `
and
Z `
2 nπx
Dn = u1 (x) sin dx.
cnπ 0 `

161
13.2 Neumann boundary conditions

ux (0, t) = 0 = ux (`, t).


They describe a a vibrating string free to slide vertically
at both ends,

OR shaft vibrating torsionally in which the ends are


held in place by frictionless bearings so that rotation at
the ends is permitted but all other motion is prevented,

OR longitudinal waves in an air column open at both


ends.

162
Conclusion: The solution is
X cnπt cnπt nπx
u(x, t) = C0 +D0 t+ [Cn cos +Dn sin ] cos ,
n≥1
` ` `

where
Z ` Z `
1 2 nπx
C0 = u0 (x)dx, Cn = u0 (x) cos dx, n ≥ 1
` 0 ` 0 `
and
Z ` Z `
1 2 nπx
D0 = u1 (x)dx, Dn = u1 (x) cos dx, n ≥ 1
` 0 cnπ 0 `

163
Note the presence of the linear term D0 t.

It says that the whole wave or the vibrating interval


drifts in the direction of u-axis at a uniform rate.
R`
Imposing the condition 0 u1 (x)dx = 0 can prevent
this drift.

(Note that the integral represents the “net” velocity. So


things are at rest in an average sense when this
integral is zero.)

At the other extreme, if u0 (x) = 0 and u1 (x) is a


constant, then u(x, t) = D0 t, in the shaft example,
the shaft will rotate with uniform angular velocity D0 ,
there will be no vibrational motion.

164
13.3 Mixed boundary conditions

u(0, t) = 0 = ux (`, t).

The solution is
P
u(x,t)= n≥0
[Cn cos c(n+1/2)πt+Dn sin c(n+1/2)πt] sin(n+1/2)πx,

where
Z `
2 π
Cn = u0 (x) sin(n + 1/2) xdx
` 0 `
and
Z `
2 π
Dn = u1 (x) sin(n + 1/2) xdx.
c(n + 1/2)π 0 `

165
13.4 Periodic boundary conditions

u(0, t) = u(`, t) and ux (0, t) = ux (`, t).

The solution is

2cnπt 2cnπt 2nπx 2nπx


P
u(x,t)=C0 +D0 t+ n≥1
[Cn cos `
+Dn sin `
][An cos `
+Bn sin `
]

where
Z `
1
C0 = u0 (x)dx,
` 0
R` R`
Cn An = 2` u0 (x) cos 2nπx
`
dx, and Cn Bn = 2` u0 (x) sin 2nπx
`
dx.
0 0

And also
Z `
1
D0 = u1 (x)dx,
` 0
and

1
R` 2nπx 1
R` 2nπx
Dn An = cnπ u1 (x) cos `
dx, and Dn Bn = cnπ u1 (x) sin `
dx.
0 0

Note that An , Bn , Cn and Dn are not uniquely


defined, but the above products are.

166
14 Nonhomogeneous case

Consider the nonhomogeneous wave equation

utt − uxx = f (x, t), 0 < x < 1, t > 0

with Dirichlet boundary conditions.

Expand everything in a Fourier sine series on (0, 1):


P P
f (x,t)= B (t) sin nπx
n≥1 n
and u(x,t)= n≥1
Yn (t) sin nπx.

Then the functions Yn (t) must satisfy

Ÿn (t) + n2 π 2 Yn (t) = Bn (t), n = 1, 2, 3, . . . .

Also let
P P
u0 (x)= b sin nπx
n≥1 n
and u1 (x)= b
n≥1 1n
sin nπx.

These lead to the initial conditions

Yn (0) = bn and Ẏn (0) = b1n .

They determine the Yn (t) uniquely.

167
Example. Suppose

f (x, t) = sin πx sin πt and u0 (x) = u1 (x) = 0.

This problem has homogeneous Dirichlet boundary


conditions and zero initial conditions. Thus
bn = 0 = b1n for all n ≥ 1, and

0 for n 6= 1,
Bn (t) =
sin πt for n = 1.

Therefore we have Yn (t) = 0 for n ≥ 2 while Y1 (t)


is the solution to the IVP

Ÿ1 (t) + π 2 Y1 (t) = sin πt, Y1 (0) = 0 = Ẏ1 (0).


sin πt−πt cos πt
Solving we find Y1 (t) = 2π 2
and hence

(sin πt − πt cos πt)


u(x, t) = 2
sin πx.

168
15 Vibrations of a circular membrane

The two-dimensional wave equation is given by

utt = c2 (uxx + uyy ).

There are three variables, x and y are the space


variables and t is the time variable.

169
Recall the polar coordinates (r, θ) in R2 :

x = r cos θ, y = r sin θ.

The wave equation written in polar coordinates is

utt = c2 (urr + r−1 ur + r−2 uθθ ).

170
We consider the wave equation for a circular
membrane of radius R.

The initial conditions are

u(r, θ, 0) = f (r, θ) and ut (r, θ, 0) = g(r, θ).

We assume Dirichlet boundary conditions

u(R, θ, t) = 0.

Physically u(r, θ, t) represents the displacement of


the point (r, θ) at time t in the z -direction. These are
transverse vibrations.

171
15.1 Radially symmetric solutions

We first find solutions which are radially symmetric, that


is, independent of θ .

Thus u = u(r, t) with initial and boundary conditions

u(r, 0) = f (r), ut (r, 0) = g(r) and u(R, t) = 0.

The wave equation simplifies to

utt = c2 (urr + r−1 ur ).

Substituting u(r, t) = X(r)Z(t), we get

Z 00 (t)X(r) = c2 (X 00 (r) + r−1 X 0 (r))Z(t).

Separating variables, we obtain

Z 00 (t) X 00 (r) + r−1 X 0 (r)


2
= = λ.
c Z(t) X(r)
What can λ be?

172
It has to be negative. (We will address this point later.)
Put λ = −µ2 .
The equation in the r variable is

r2 X 00 (r) + rX 0 (r) + µ2 r2 X(r) = 0.

This is the scaled Bessel equation of order 0.

This implies that X(r) is a multiple of the scaled


Bessel function of the first kind J0 (µr).

(The second solution of the Bessel equation is


unbounded at r = 0. Since we require u(0, t) to be
finite, this solution has to be discarded.)

173
The Dirichlet boundary condition now implies
J0 (µR) = 0.
So µ must be 1/R times one of the countably many
positive zeroes of J0 .

These are the fundamental modes of vibration of the


membrane.

For any such µ (there are countably many of them),


Z(t) is given by

Z(t) = A cos cµt + B sin cµt.

The elementary solutions or pure harmonics are

un (r, t) = (An cos cµn t + Bn sin cµn t)J0 (µn r),

where µn R is the n-th positive zero of J0 .

174
The center of the membrane vibrates with the
maximum amplitude.

For n = 1, the membrane moves fully up and then fully


down. All points except those on the boundary vibrate.
µ1
For n = 2, there is a circle of radius µ2 R which does
not move at all. This is called a nodal line. When the
portion inside the nodal line moves up, the portion
outside moves down, and vice versa.

What happens for n = 3?


There are two nodal lines.

Visualize this vibration.

175
Conclusion: The solution is given by
X
u(r, t) = (An cos cµn t+Bn sin cµn t)J0 (µn r).
n≥1

The constants can be calculated by expanding the


initial position f (r) and initial velocity g(r) into
Fourier-Bessel series. Thus,
Z R
2
An = 2 2 rf (r)J0 (µn r)dr
R J1 (µn R) 0

and
Z R
2
Bn = rg(r)J0 (µn r)dr.
cµn R2 J12 (µn R) 0

176
15.2 Reason for not allowing λ to be zero or
positive

• Suppose λ = 0.
Then we need to solve

r2 X 00 (r) + rX 0 (r) = 0.

This is a Cauchy-Euler equation.

The general solution is A log r + B.


It must be finite at r
= 0 which gives A = 0, while
it is zero at r = R which gives B = 0.

Thus, A = B = 0.

177
• Suppose λ = µ2 > 0.
Then we get the Bessel equation scaled by the
imaginary number iµ.

Since u must be bounded as r → 0, discarding


the unbounded solution, we are left with

X (µr)2k
J0 (iµr) = .
k≥0
4k (k!)2

But since this is a series of positive terms only, it


cannot satisfy X(R) = 0.

178
15.3 General solution

We solve for u(r, θ, t) = X(r)Y (θ)Z(t).


Substituting in utt = c2 (urr + r−1 ur + r−2 uθθ ),
we get

Z 00 (t)X(r)Y (θ)
= c2 (X 00 (r)Y (θ)+r−1 X 0 (r)Y (θ)+r−2 X(r)Y 00 (θ))Z(t).

The method of separation of variables now proceeds in


two steps:

We have a total of 3 variables and we need to separate


one variable at a time.

Separating θ ,

Y 00 (θ) r2 X 00 (r) + rX 0 (r) r2 Z 00 (t)


=− + 2 .
Y (θ) X(r) c Z(t)
What can this constant be?

179
Periodic boundary conditions in θ force the constant to
be −n2 for n = 0, 1, 2, 3, . . . . We have

Y (θ) = A cos nθ + B sin nθ.

(n = 0 recovers the radially symmetric case.)


Putting the constant equal to −n2 and separating r
and t,

Z 00 (t) X 00 (r) + r−1 X 0 (r) n2


2
= − 2.
c Z(t) X(r) r

180
For reasons similar to the radially symmetric case, this
constant must be negative. So we write it as −µ2 .

The equation in the r variable

r2 X 00 (r) + rX 0 (r) + (µ2 r2 − n2 )X(r) = 0.

This is the scaled Bessel equation of order n.

The elementary solutions are

(A cos nθ+B sin nθ)(C cos cµt+D sin cµt)Jn (µr),

where µR is a positive zero of Jn .

181
First visualize the amplitude cos θJ1 (µr), where µR
is the first positive zero of J1 .

Now visualize the corresponding vibration.

There is a nodal like along a diameter, when the


portion on one side of the diameter is up, the other
portion is down, and vice versa.

182
16 Coordinate systems

Polar coordinates (r, θ) in R2 :

x = r cos θ, y = r sin θ.

Cylindrical coordinates (r, θ, z) in R3 :

x = r cos θ, y = r sin θ, z = z.

Spherical polar coordinates (r, θ, ϕ) in R3 :

x = r cos θ sin ϕ, y = r sin θ sin ϕ, z = r cos ϕ.

In all three cases, 0 ≤ θ ≤ 2π , while in the last case


0 ≤ ϕ ≤ π.

183
17 Laplacian operator

The Laplacian operator on the line is ∆R (u) = u00 .


The Laplacian operator on the plane is

∆R2 (u) = uxx + uyy


1 1
= urr + ur + 2 uθθ .
r r
These are standard coordinates and polar coordinates
respectively.

184
The Laplacian operator on three-dimensional space is

∆R3 (u) = uxx + uyy + uzz ,


1 1
= (urr + ur + 2 uθθ ) + uzz ,
r r
2 1 1
= urr + ur + 2 (uϕϕ + cot ϕuϕ + 2 uθθ ).
r r sin ϕ
These are standard coordinates, cylindrical coordinates
and spherical polar coordinates respectively.

185
The Laplacian operator on the sphere of radius R is

1 1
∆(u) = 2 (uϕϕ + cot ϕuϕ + 2 uθθ ).
R sin ϕ
When R = 1, we write ∆S2 (u) instead of ∆(u).

186
17.1 General heat and wave equation

The Laplacian operator can be used to define the heat


equation in general:

ut = k∆(u),

where ∆ is the Laplacian operator on the domain


under consideration.

Similarly, the general wave equation can be written as

utt = c2 ∆(u).

187
17.2 Eigenfunctions for Laplacian operator
on unit sphere

Consider the eigenvalue problem on the unit sphere:

∆S2 (u) = µ u.

Any solution u is called an eigenfunction and the


corresponding µ is called its eigenvalue.

188
Explicitly, we want to solve

∂2u ∂u 1 ∂2u
2
+ cot ϕ + 2 2
= µ u.
∂ϕ ∂ϕ sin ϕ ∂θ

Put u(θ, ϕ) = X(θ)Y (ϕ).


Substituting in the PDE,

1
X(θ)Y 00 (ϕ) + cot ϕX(θ)Y 0 (ϕ) + 2 X 00
(θ)Y (ϕ)
sin ϕ
= µ X(θ)Y (ϕ).

Dividing by X(θ)Y (ϕ) and separating variables,

Y 00 (ϕ) Y 0 (ϕ) X 00 (θ)


sin2 ϕ[ + cot ϕ − µ] = − .
Y (ϕ) Y (ϕ) X(θ)

The constant is forced to be m2 , for some nonnegative


integer m because of periodic boundary conditions in
θ.

189
The equation in ϕ becomes

m 2
Y 00 (ϕ) + cot ϕY 0 (ϕ) − (µ + 2 )Y (ϕ) = 0.
sin ϕ

The change of variable x = cos ϕ yields

2 00 0 m2
(1 − x )y − 2xy − (µ + 2
)y = 0.
1−x
(Here x is the coordinate along the z -axis.)

This is the associated Legendre equation.

Putting m = 0 recovers the Legendre equation.

190
The associated Legendre equation has a bounded
solution iff µ = −n(n + 1) for some nonnegative
integer n.

The bounded solution is given by

y(x) = (1 − x2 )m/2 Pn(m) (x).

Going back to the ϕ coordinate,

Y (ϕ) = sinm ϕ Pn(m) (cos ϕ).

191
Thus

cos mθ sinm ϕ Pn(m) (cos ϕ), m = 0, 1, 2, . . . n


sin mθ sinm ϕ Pn(m) (cos ϕ), m = 1, 2, . . . n

are eigenfunctions for the Laplacian operator on the


unit sphere with eigenvalues −n(n + 1).

In particular,

cos ϕ, sin θ sin ϕ, cos θ sin ϕ

are eigenfunctions with eigenvalue −2.

(These are nothing but z , y and x in spherical-polar


coordinates.)

192
17.3 Vibrations of a spherical membrane

The wave equation on the unit sphere is

utt = ∆S2 u.

Putting u(θ, ϕ, t) = X(θ)Y (ϕ)Z(t), we obtain


Z 00 (t) ∆S2 (X(θ)Y (ϕ))
= = µ.
Z(t) X(θ)Y (ϕ)
What can µ be?

193
Since µ is an eigenvalue of the Laplacian operator on
the sphere, it has the form

µ = −n(n + 1).

The pure harmonics are

(m)
√ √
(A cos mθ+B sin mθ) sinm ϕPn (cos ϕ)(C cos( n(n+1)t)+D sin( n(n+1)t)),

for nonnegative integers 0 ≤ m ≤ n and n 6= 0.

194
The case m = 0 corresponds to the harmonics which
do not depend on θ . These are called latitudinally
symmetric solutions. We elaborate on these below.

The maximum amplitudes of vibrations are at the north


and south poles. For n = 1, the equator is a nodal
line, the maximum amplitudes are at the poles.

What happens for n = 2?


General case: There are n values of ϕ for which
Pn (cos ϕ) = 0, the corresponding latitudes are the
nodal lines.

What happens for n = 0?

Z(t) = A + Bt

195
17.4 Laplace equation in three space

Recall
2 1 1
∆R3 (u) = urr + ur + 2 (uϕϕ +cot ϕuϕ + 2 uθθ ).
r r sin ϕ
Consider the Laplace equation in three space:

∆R3 (u) = 0.

Putting u(r, θ, ϕ) = Z(r)X(θ)Y (ϕ), we obtain

r2 Z 00 (r) 2rZ 0 (r) ∆S2 (X(θ)Y (ϕ))


+ =− .
Z(r) Z(r) X(θ)Y (ϕ)

196
From the eigenvalue problem for the Laplacian
operator, we know that this constant must be n(n + 1)
for some nonnegative integer n.

The equation in the r variable is

r2 Z 00 (r) + 2rZ 0 (r) − n(n + 1)Z(r) = 0.

This is a Cauchy-Euler equation with solutions r n and


1
rn+1
.

The elementary solutions are

1 (m)
(Crn +D n+1 )(A cos mθ+B sin mθ) sinm ϕPn (cos ϕ),
r

where 0 ≤ m ≤ n are nonnegative integers.

197
17.5 Other domains

In the circular membrane case, we did not consider the


second solution to the Bessel equation since it is
unbounded at r = 0. However, this solution would
have to be considered if we are looking at an annulus.

Similar remark applies while working with a spherical


cap (such as the upper hemisphere) as opposed to the
full sphere.

198

You might also like