7.

Application

solving differential equations

Definition 1. A (ordinary) differential equation is linear if it can be written in the form

L(y) = f (x) where L is a linear operator, that is

L(αy1 + βy2 ) = αL(y1 ) + βL(y2 ), for all α, β ∈ R

and functions y1 , y2 are ’reasonably nice’.

PN dk y
Example: L[y] = k=0 ak (x)y (k) (x) where y (k) = dxk
for k ∈ N and y (0) = y. More
generally, if f : Rn → R is differentiable, we can consider the linear partial differential
operator of order 1
n
X ∂f
L[f ] = ai (x)∂i f (x), where ∂i f = (7)
i=1 ∂xi

Examples of higher order partial differential operator can be much more complicated and
I will just mention one simple one that we will use: the Laplacian

n
X ∂ 2f
∆f = 2
.
i=1 ∂xi

Please note that in your textbook ∆f is written as ∇2 f (as ∆ has been used for other
purpose in the textbook).

We could say linear operators act on C ∞ (Ω), infinitely differentiable functions on the
domain Ω (a special open set, to be exact, an open connected set, you do not have to
worry about this term if you have no idea about being ’connected’) in Rn . Clearly, a linear
combination of linear operators is again a linear operator. In particular, the collection
of linear operators on C ∞ (Ω) is a vector space (defined in linear algebra). For example,
if L1 , L2 are linear operators, then so is αL1 + βL2 , α, β ∈ R. In particular, αL + β∆,
α, β ∈ R is a linear operator where L is defined in equation (7).
68
 Mar 9, 23 69

Remark Depending on the order of the differential operators, we could replace C ∞ (Ω)
by C k (Ω), the space of functions with continuous differentials up to order k.

Linear operators and superposition

Recall that if L[y] = a(x)y 00 + b(x)y 0 + c(x)y, then L is a linear operator. In particular,
if L[y1 ] = L[y2 ] = 0, then L[y] = 0 if y(x) = αy1 (x) + βy2 (x) for all x, where α, β are fixed
real numbers.
Here is one quick example: if L[y] = y 00 − 2y 0 + y, then it is easy to check that L[ex ] =
L[xex ] = 0. Hence L[αex + βxex ] = 0 for all α, β ∈ R. Similarly if L1 [y] = y 00 + y, then
L1 [α sin x + β cos x] = 0 for all α.β ∈ R.
However, if L[u] = ut −uxx , without certain convergence assumption, it may not be true
P∞
that L[ k=0 αk Xk (x)Tk (t)] = 0 (see page 3 for Xk and Tk ) even though L[Xk (x)Tk (t)] = 0
for all k. However, in this module, we will assume that it is usually true.

Part A: partial differential equations

Fourier series is an essential tool in the method of separation of variables. First let us
recall the following example that we have done before.
Example

ut = α2 uxx , 0 < x < 1, t > 0; u(0, t) = u(1, t) = 0 for all t ≥ 0, (1)

u(x, 0) = f (x), 0 < x < 1. (2)

Let us quickly recall how we did it. First we start by guessing u(x, t) = T (t)X(x), then
70 Mar 9, 23

ut = T 0 (t)X(x) and uxx = T (t)X 00 (x)

and hence
T 0 (t) X 00 (x)
= .
α2 T (t) X(x)

However, notice that LHS depends only on t and RHS depends only on x. For that to
happen, they must both equal to a constant. Thus, there exists λ such that

T 0 (t)
=λ (3)
α2 T (t)
X 00 (x)
= λ. (4)
X(x)
Recall from (1) that we must have X(0) = X(1) = 0.
We then consider three cases.
Case (1) λ > 0: (4) has no nontrivial solution. This is because its general solution is of
the form
√ √
X(x) = C1 e λx
+ C2 e− λx

and hence X(0) = X(1) = 0 =⇒ C1 = C2 = 0.

Case (2) λ = 0: again the only solution of (4) such that X(0) = X(1) = 0 is X(x) = 0.
Case (3) λ = −β 2 , β > 0: then

X(x) = C1 cos βx + C2 sin βx

and X(0) = 0 =⇒ C1 = 0. Hence

0 = X(1) = C2 sin β =⇒ β = kπ, k ∈ N if we want C2 6= 0.

Note that if λ = −(kπ)2 , k ∈ N, then

X(x) = C2 sin kπx ( let Xk (x) = sin kπx.)

For each k ∈ N, we have T 0 = −(kπ)2 α2 T and hence

2t 2
T (t) = ce−(kαπ) let Tk (t) = e−(kαπ) t .
2
Thus for each k ∈ N, Tk (t)Xk (x) = e−(kαπ) t sin kπx is a solution of (1). We then note
that for any {ck } ⊂ R,
∞
2
ck e−(kαπ) t sin kπx
X

k=1
 Mar 9, 23 71

satisfies (1) (provided the series converges and we can do term by term differentiations).
We now make use of (2) and we have
∞
X
f (x) = u(x, 0) = ck sin kπx.
k=1

We will then find ck by using the Fourier sine series of f . That is

Z 1
ck = 2 f (x) sin kπxdx.
0

Next, we will show that the solution make sense. We will need a fact from MA3110
which is about term by term differentiation.
It follows from that fact that whenever f is piecewise continuous function on [0, 1], then
the function
∞
2
ck e−(kαπ) t sin kπx
X
u(x, t) =
k=1
satisfies both (1) and (2).
We will provide a sketch of the proof
First, clearly the series converges for any t > 0. Thus, u(x, t) is well defined for any
t > 0.
Step (1): for any t0 > 0, the series (of term by term derivation w.r.t t)
∞
2
−(kαπ)2 ck e−(kαπ) t sin kπx
X

k=1
2
converges uniformly on [t0 /2, 2t0 ] and each of the term ck e−(kαπ) t sin kπx has a continuous
derivative on [t0 /2, 2t0 ]. It then follows that
∞
2
−(kαπ)2 ck e−(kαπ) t sin kπx.
X
ut (x, t0 ) =
k=1

Step (2): By similar argument (we now need to apply the theorem twice), we conclude
that
∞
2
−(kπ)2 ck e−(kαπ) t sin kπx.
X
uxx (x, t0 ) =
k=1
Step (3): It follows from Parseval’s identity that
Z 1 N ∞
ck sin kπx|2 = lim |ck |2 → 0.
X X
lim |f (x) −
N →∞ 0 N →∞
k=1 k=N +1

Thus,
N
X
ck sin kπx converges to f in the means on [0,1]
k=1
or converges to f in L2 [0, 1].
72 Mar 9, 23

By similar argument, we have u(., t) → f as t → 0 in L2 [0, 1].

However, if f 0 is piecewise continuous and continuous such that f (1) = f (0) = 0, then
u(., t) → f as t → 0 uniformly on [0,1].

[Exercise]

 ux (0, t) = ux (2, t) = 0, t > 0
utt = uxx , .
 u(x, 0) = f (x), ut (x, 0) = g(x), 0 < x < 2
 Mar 9, 23 73

[Exercise]

uxx − 4uyy = u, 0 < x < 1, u(x, 0) = 0, u(x, 1) = 2, 0 < x < 1

u(0, y) = 1, uy (0, y) = cos2 πy, y > 0.

74 Mar 9, 23

[Exercise]

uxx + uyy = 0, 0 < x, y < 1, u(x, 0) = u(x, 1) = 0, 0 < x < 1

u(0, y) = 0, u(1, y) = sin2 πy, 0 < y < 1.

 Mar 9, 23 75

[Exercise]

uxx + 4uyy = 0, 0 < x, y < 1, u(x, 0) = u(x, 1) = 1, 0 < x < 1

u(0, y) = 1, u(1, y) = cos2 πy, 0 < y < 1.

76 Mar 9, 23

Part B: Ordinary differential equations

First, let us look at one example:

y 00 + y = sin 2x, 0 < x < π, y(0) = y(π) = 0.

Let L(y) = y 00 + y. Instead of solving the above mentioned problem, we will try to find
nontrivial solution yk such that L(yk ) = λk yk for some λk ∈ R.
(We will call λk the eigenvalue and yk is an eigenfunction with respect to λk ).
The problem becomes: finding λ ∈ R such that

y 00 + y = λy, y(0) = y(π) = 0

has a nontrivial solution.

We can of course use the same technique to find the solution of

y 00 + y = f (x), 0 < x < π, y(0) = y(π) = 0

for any piecewise smooth function f on [0, π].

 Mar 9, 23 77

[Exercise]

y 00 + 4y = x(1 − x), 0 < x < 1, y(0) = y(1) = 0.

[Exercise]

y 00 + y = cos2 x, 0 < x < π, y 0 (0) = y 0 (π) = 0.

78 Mar 9, 23

[Exercise]

 1, if 0 < x < 1
y 00 − y =
 −1, if 1 < x < 2
y 0 (0) = y 0 (2) = 0.
Nonhomogeneous boundary conditions

y 00 + y = sin 2x, 0 < x < π, y(0) = 0, y(π) = 1.

8. Sturm-Liouville Problems

(§ 67)
Self-adjoint differential equation:

(p(x)y 0 )0 + q(x)y + λr(x)y = P2 [y] + λr(x)y = 0,

a0 y(a) + a1 y 0 (a) = b0 y(b) + b1 y 0 (b) = 0 (a20 + a21 6= 0, b20 + b21 6= 0).

Note that λ is a parameter. Such a problem is called a Sturm-Liouville Problem. We say
that the problem is regular if

p, r > 0 on [a, b], p, q, r are continuous and p is continuously differentiable.

Solving such a problem means finding values of λ (eigenvalues) and the corresponding
nontrivial solutions φλ (eigenfunctions). The set of all eigenvalues of a regular problem is
called its spectrum.
 Mar 9, 23 79

First, let us look at some simple examples.

Example 1
y 00 + λy = 0, y(0) = y(π) = 0.
Check that p(x) = 1, q(x) = 0 and r(x) = 1. Clearly p(x), r(x) > 0 and p(x) is continu-
ously differentiable. Hence this is a regular Sturm-Liouville problem.
When λ = k 2 , k ∈ N, then it has an eigenfunction φλ (x) = sin kx.

Example 2
y 00 + λy = 0, y(1) = y 0 (2) = 0.
First of all, note that it is easier to look at the problem

y 00 + λy = 0, y(0) = y 0 (1) = 0.

It can be done by changing of variable: t = x − 1.

 2
2k−1
Solution: λ = 2
π is an eigenvalue and its eigenfunction is φλ (x) = sin 2k−1
2
πt.

[Exercise] Find the eigenvalues and their corresponding eigenfunctions of the following
Sturm-Liouville problem:

y 00 + λy = 0, y 0 (0) = y 0 (π) = 0.
80 Mar 9, 23

Let us now make some observations from the above examples.

(1) If ψ1 (x) and ψ2 (x) are both eigenfunctions with respect to the same eigenvalue,
then ψ1 (x) = kψ2 (x), k is a nonzero constant.

(2) If φ1 (x) and φ2 (x) are eigenfunctions correspond to eigenvalues λ1 and λ2 respec-
tively, λ1 6= λ2 , then
Z π Z π
φ1 (x)φ2 (x)dx = 0 or φ1 (x)φ2 (x)r(x)dx = 0.
0 0

Moreover, φ1 (x) and φ2 (x) are linearly independent.

(3) All eigenvalues are real.

(4) There are infinite number of eigenvalues. Moreover, it can be arranged in an

increasing order λ1 < λ2 < · · · < λn < · · · such that limn→∞ λn → ∞.
 Mar 9, 23 81

It turns out that all the above observations are true for any regular Sturm-Liouville
problem.
Proof.
82 Mar 9, 23

Next, let us note that the idea of eigenvalues and eigenfunctions can be used for other
linear BVP: L[y] + λr(x)y = 0, l1 (y) = l2 (y) = 0. For example,

p0 (x)y 00 + p1 (x)y 0 + q(x)y + λr(x)y = 0, l1 (y) = 0, l2 (y) = 0.

In particular, we can consider

y 00 + 2y 0 + λy = 0, y(0) = y(1) = 0.

Finally, let us look at some examples of singular Sturm-Liouville problems (optional).

We say that the self-adjoint BVP: (p(x)y 0 )0 + q(x)y + λr(x)y = 0 is singular if it is not
regular. For example, if p(x) = 0 for some x ∈ [a, b], or a0 = a1 = 0, or r(x) ≤ 0 for some
x ∈ [a, b]. Let us look at some examples:

Example y 00 + λy = 0, y(0) = 0, |y(x)| ≤ M < ∞ for all x ∈ (0, ∞).

√
Note that any λ ∈ (0, ∞) is an eigenvalue with eigenfunction sin λx. Note that it is
now impossible to arrange all the eigenvalues as {λ1 , λ2 , · · · }, such that λ1 < λ2 < · · · .

Example § 103: ((1 − x2 )y 0 )0 + λy = 0 (Legendre’s differential equations)

lim y(x) < ∞, lim y(x) < ∞.

x→−1 x→1
 Mar 9, 23 83

The eigenvalues of this problem are λn = n(n − 1); n = 1, 2, · · · , and the corresponding
eigenfunctions are the Legendre’s polynomials Pn−1 .
Fourier (Legendre) series
{Pn (x) : n ∈ N} is a family of orthogonal functions on [−1, 1]. Hence if we write
P∞
f (x) = k=0 cn Pn (x), then
2n + 1 Z 1
cn = f (x)Pn (x)dx,
2 −1

since Z 1
2
< Pn , Pn >= Pn (x)2 dx = .
−1 2n + 1

1 dn 2
Note that P0 = 1, P1 = x, P2 = 12 (3x2 −1), P3 = 12 (5x3 −3x), Pn = (x − 1)n , · · ·
2n n! dxn

Exercise Find the Fourier (Legendre) series of x2 and x3 .

84 Mar 9, 23

Let us now collect some useful theorems.

Theorem If an orthogonal set {φn } is complete in P C[a, b], then any function f ∈
P C[a, b] that is orthogonal to all φn (that is < f, φn >= 0 for all n), must be zero except
finite number of points in [a, b].

Unfortunately, there is no simple procedure to establish the completeness of a given

orthogonal family of functions. However, there are some known results.
Theorem The orthogonal set {φn : n = 0, 1, 2, · · · } in [a, b] with respect to r(x) is
complete in P C[a, b] if φn is a polynomial of degree n.

For example {Pn : n = 0, 1, 2, · · · } is complete in P C[a, b].

Theorem
Let {φn (x) : n ∈ N} be the set of all eigenfunctions of the regular Sturm-Liouville
problem. Then it is complete in P C[a, b] and the (general) Fourier series of f (x) converges
1
to (f (x+ ) + f (x− )) at each point in (a, b) if f is piecewise smooth on [a, b].
2