Contents

1 Lecture 1 - Introduction to Partial Differential Equations 5
1.1 Modeling and Derivation of PDE: . . . . . . . . . . . . . . . . 6
1.2 The Wave Equation: . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 The Drunkard’s Walk - The Heat Equation: . . . . . . . . . . 10
2 Lecture 2 - Preliminaries 13
2.1 Sequences and Series of Numbers: . . . . . . . . . . . . . . . . 13
2.2 Absolute and Conditional Convergence: . . . . . . . . . . . . 16
2.3 Power Series: . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Lecture 3 - Review of Methods to Solve ODE 19
3.1 First Order ODE: . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Another Method - Series Solution: . . . . . . . . . . . . . . . 20
3.3 Second Order Constant Coefficient Linear Equations: . . . . . 21
3.4 Euler/Equidimensional Equations: . . . . . . . . . . . . . . . 23
4 Lectures 4,5 Ordinary Points and Singular Points 27
4.1 An Ordinary Point: . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 A Singular Point: . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 The Airy equation: . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4 The Hermite Equation: . . . . . . . . . . . . . . . . . . . . . . 31
5 Lecture 6 - Singular points 33
5.1 Radius of Convergence and Nearest Singular Points . . . . . . 33
5.2 Singular Points: . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Regular Singular Points: . . . . . . . . . . . . . . . . . . . . . 35
5.4 More General Definition of a Regular Singular Point: . . . . . 36
6 Lecture 7 - Frobenius Series about Regular Singular Points 39
6.1 Series Expansion Summary: . . . . . . . . . . . . . . . . . . . 41
1
CONTENTS
7 Bessel’s Equation 43
7.1 Bessel’s Function of Order ν / ∈ ¦. . . , −2, −1, 0, 1, 2 . . .¦: . . . . 43
7.2 Bessel’s Function of Order ν = 0 - repeated roots: . . . . . . . 44
7.3 Bessel’s Function of Order ν =
1
2
: . . . . . . . . . . . . . . . . 46
7.4 Example - the roots differ by an integer . . . . . . . . . . . . 48
8 Separation of Variables 49
8.1 Types of Boundary Value Problems: . . . . . . . . . . . . . . 49
8.2 Separation of Variables - Fourier sine Series: . . . . . . . . . . 51
8.3 Heat Eq on a Circular Ring - Full Fourier Series . . . . . . . 57
9 Lecture 13 - Fourier Series 61
9.1 It can be useful to shift the interval of integration from [−L, L]
to [c, c + 2L] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
9.2 Complex Form of Fourier Series . . . . . . . . . . . . . . . . . 65
10 Lecture 14 - Even and Odd Functions 67
10.1 Integrals of Even and Odd Functions . . . . . . . . . . . . . . 67
10.2 Consequences of Even/Odd Property for Fourier Series . . . . 68
10.3 Half-Range Expansions . . . . . . . . . . . . . . . . . . . . . . 70
11 Lecture 15 - Convergence of Fourier Series 73
11.1 Convergence of Fourier Series . . . . . . . . . . . . . . . . . . 76
11.2 Illustration of the Gibbs Phenomenon . . . . . . . . . . . . . 77
11.3 Now consider the sum of the first N terms . . . . . . . . . . . 78
12 Lecture 16 - Parseval’s Identity 81
12.1 Geometric Interpretation of Parseval’s Formula . . . . . . . . 82
13 Lecture 17 - Solving the heat equation using finite difference
methods 85
13.1 Approximating the Derivatives of a Function by Finite Dif-
ferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
13.2 Heat Equation solution by Finite Differences . . . . . . . . . 87
14 Lecture 18 - Solving Laplace’s Equation using finite differ-
ences 91
14.1 Finite Difference approximation . . . . . . . . . . . . . . . . . 91
14.2 Solving the System of Equations by Jacobi Iteration . . . . . 93
2
CONTENTS
15 Lecture 19 Further Heat Conduction Problems: Inhomoge-
neous BC 95
16 Lecture 20 - Inhomogeneous Derivative BC 101
17 Lecture 21 Distributed, Time Dependent Heat Sources -
eigenfunction expansions 105
18 Lecture 22 More Eigenfunction Expansions - Time Depen-
dent Boundary Conditions 111
19 Lecture 23 - 1D Wave Equation 117
19.1 Guitar String . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
20 Lecture 24 - Space-Time Interpretation of D’Alembert’s So-
lution 121
20.1 Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 121
20.2 Region of Influence . . . . . . . . . . . . . . . . . . . . . . . . 122
20.3 Domain of Dependence . . . . . . . . . . . . . . . . . . . . . . 122
21 Lecture 25 Solution by separation of variables 125
21.1 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
21.2 Now we can use the trigonometric identities . . . . . . . . . . 127
22 Lecture 26 - Laplace’s Equation 129
22.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
22.2 Laplace’s Equation . . . . . . . . . . . . . . . . . . . . . . . . 130
22.3 Rectangular Domains . . . . . . . . . . . . . . . . . . . . . . 130
22.4 Solution to Problem (1A) by Separation of Variables . . . . . 131
23 Lecture 27 - More Rectangular Domains and semi-infinite
strip problems 135
23.1 Solution to Problem (1B) by Separation of Variables . . . . . 135
23.2 Rectangular domains with mixed BC . . . . . . . . . . . . . . 136
23.3 Semi-infinite strip problems . . . . . . . . . . . . . . . . . . . 138
24 Lecture 28 - Neumann Problem - only flux BC and Circular
domains 141
24.1 Neumann Problem on a rectangle . . . . . . . . . . . . . . . . 141
24.2 General Analysis of Laplace’s Equation on Circular Domains: 144
24.3 R Equation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
3
CONTENTS
24.4 Θ Equation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
24.5 For Different Boundary Conditions: . . . . . . . . . . . . . . . 145
24.5.1 Notes: . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
25 Lecture 29 Wedge Problems 147
26 Lecture 30 Wedges with cut-outs, circles, holes and annuli 153
26.1 Special Case - Electrical Impedance Tomography . . . . . . . 157
26.2 Poisson’s Integral Formula: . . . . . . . . . . . . . . . . . . . 159
27 Lecture 31 Sturm-Liouville Theory 161
27.1 Boundary value problems and Sturm-Liouville theory: . . . . 161
27.2 The regular Sturm-Liouville problem: . . . . . . . . . . . . . 162
27.3 Properties of SL Problems . . . . . . . . . . . . . . . . . . . . 164
28 Lecture 32 Solving the heat equation with Robin BC 167
28.1 Expansion in Robin Eigenfunctions . . . . . . . . . . . . . . . 167
28.2 Solving the Heat Equation with Robin BC . . . . . . . . . . . 168
29 Lecture 33 Variable coefficient BVP - eigenfunctions involv-
ing solutions to the Euler Equation: 171
29.1 Cases: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
29.2 Solving the heat equation by expanding in eigenfunctions in-
volving solutions to an Euler Equation: . . . . . . . . . . . . 173
30 Lecture 34 — Sturm Liouville Theory 175
30.1 Properties of SL Problems: . . . . . . . . . . . . . . . . . . . 175
30.2 Lagrange’s Identity: . . . . . . . . . . . . . . . . . . . . . . . 176
30.3 Proofs to selected properties: . . . . . . . . . . . . . . . . . . 177
4
Chapter 1
Lecture 1 - Introduction to
Partial Differential Equations
ODE - Equations which define functions of a single independent variable
by prescribing a relationship between the values of the function and its
derivatives.
EG:
y

(x) +e
y(x)
= 0. (1.1)
PDE - Involve multivariable functions u(x, t), u(x, y) that are determined by
prescribing a relationship between the function value and its partial deriva-
tives.
EG 1:
a

∂x
u(x, y) +b

∂y
u(x, y) = c First Order PDE (1.2)
5
Lecture 1 - Introduction to Partial Differential Equations
EG 2: Some Classic Second Order PDEs:
Quadric Classification Eq. Name
T = X
2
Parabolic

∂t
u(x, t) =

2
∂x
2
u(x, t) Heat Equation or Diffusion Eq
X
2
+Y
2
= k Elliptic

2
∂x
2
u(x, y) +

2
∂y
2
u(x, y) = f(x, y)
Poisson Eq f ≡ 0
Laplace Eq f = 0
T
2
−c
2
X
2
= k Hyperbolic

2
∂t
2
u(x, t) −c
2 ∂
2
∂x
2
u(x, t) = 0 The Wave Eq
By analogy with quadric surfaces aX
2
+2bXY +c
2
Y
2
+ = k that can be
reduced to a standard form by coordinate rotation, the most general linear
2nd order PDE
au
xx
+ 2bu
xy
+cu
yy
+ (1.3)
can be reduced by a transformation of coordinates to one of the Heat,
Laplace or the Wave Eq.
1.1 Modeling and Derivation of PDE:
1D Conservation Law: Traffic flow on a highway.
Consider the traffic flow on a highway and let u(x, t) be the density of
cars at x at time t.
[u] = # of cars/unit length. (1.4)
Let q(x, t) be the flux of cars at x at time t.
[q] = # of cars/unit time. (1.5)
6
1.1. MODELING AND DERIVATION OF PDE:
¦u(x, t + ∆t) −u(x, t)¦∆x · ¦q(x, t) −q(x + ∆x, t)¦∆t (1.6)
Let ∆x →0 and ∆t →0:
∂u
∂t
+
∂q
∂x
= 0 (1.7)
• conservation of cars
• conservation of heat
• conservation of chemicals.
How does q change with u?
Convection - and the first order Wave Equation: Assume that q
varies linearly with u, i.e., q = cu from which it follows that
∂u
∂t
+c
∂u
∂x
= 0 (1.8)
But this is just a wave equation. To see this consider the following moving
coordinate system.
x

= x −ct transformation of coordinates (1.9)
Guess:
u(x, t) = f(x −ct) solves u
t
+cu
x
= 0
u
t
= −cf

u
x
= f

(1.10)
Therefore u
t
+cu
x
= −cf

+cf

= 0.
Thus u
t
+ cu
x
= 0 has solutions of the form u(x, t) = f(x − ct) which
represents a right moving wave. What happens if q = −cu in which case
∂u
∂t
−c
∂u
∂x
= 0 (1.11)
7
Lecture 1 - Introduction to Partial Differential Equations
Exercise: Show that (1.11) has a solution of the form u(x, t) = f(x + ct)
which represents a left moving wave.
Note:


∂t
+c

∂x


∂t
−c

∂x

u(x, t) =

2
u
∂t
2
−c
2

2
u
∂x
2
= 0 (1.12)
is the 2nd order wave equation that has both left and right moving wave
solutions.
Fourier’s Law: Heat flows from hotter regions to colder ones?
q = −α
2
∂u
∂x
(1.13)
In this case the conservation law reduces to the form:
∂u
∂t
= α
2

2
u
∂x
2
The Heat Equation (1.14)
2D Heat Equation:
∂u
∂t
= α
2


2
u
∂x
2
+

2
u
∂y
2

(1.15)
8
1.2. THE WAVE EQUATION:
1.2 The Wave Equation:
Consider an elastic rod having a density ρ and cross-sectional area A, and let
σ(x, t) be the pressure in the rod at x at time t and u(x, t) the displacement
of the rod from equilibrium.
Balance of Linear Momentum F = Ma.
σ(x + ∆x, t)A−σ(x, t)A = ρA∆x

2
u
∂t
2
σ(x + ∆x, t) −σ(x, t)
∆x
= ρ

2
u
∂t
2
(1.16)
∆x →0
∂σ
∂x
= ρ

2
u
∂t
2
(BLM) Balance of Linear Momentum
Hooke’s Law
σ = E
∂u
∂x
(1.17)
Plug into (BLM) to obtain the 2nd order wave equation.

2
u
∂t
2
=

E
ρ


2
u
∂x
2
= c
2

2
u
∂x
2
, where c =

E
ρ
(1.18)
9
Lecture 1 - Introduction to Partial Differential Equations
1.3 The Drunkard’s Walk - The Heat Equation:
Let u(x, t) be the density of fruit-flies at point x at time t. Find an equation
for the density of flies at t + ∆t.
u(x, t + ∆t) = pu(x + ∆x, t) + (1 −2p)u(x, t) +pu(x −∆x, t)
= u(x, t) +p∆x
[u(x + ∆x, t) −u(x, t)]
∆x

[u(x, t) −u(x −∆x, t)]
∆x
· u(x, t) +p∆x
2
¸
∂u
∂x
(x, t) −
∂u
∂x
(x −∆x, t)
¸
∆x
(1.19)
· u(x, t) +p∆x
2

2
u
∂x
2
u(x, t + ∆t) −u(x, t)
∆t
·

p
∆x
2
∆t


2
u
∂x
2

∂u
∂t
= α
2

2
u
∂x
2
The Heat Eq.
What is the Mean Absolute Deviation of the Drunkard?
s
j
= ±∆x
t
j
= j∆t
x
N
= s
1
+s
2
+ +s
N
∼ 0 Expected Value (1.20)
x
2
N
= (s
1
+ +s
N
)
2
= s
2
1
+ +s
2
N
+ 2(s
1
s
2
+ +s
N−1
s
N
)
∼ N∆x
2
Therefore
x
2
N
·

t
N
∆t

∆x
2
= k
2
t
N
[x
N
[ ∼ k

t
N
10
1.3. THE DRUNKARD’S WALK - THE HEAT EQUATION:
0 2 4 6 8 10 12 14 16 18 20
−4
−3
−2
−1
0
1
2
3
4
5
t
x
Drunkard Walk
Figure 1.1: Simulation with N = 1000 trajectories for 200 steps along with
the mean absolute deviation envelopes shown in red
11
Lecture 1 - Introduction to Partial Differential Equations
12
Chapter 2
Lecture 2 - Preliminaries
2.1 Sequences and Series of Numbers:
A Sequence of Numbers:
1,
1
2
,
1
3
, . . . ,
1
n
, . . .
1,
1
2
,
1
4
, . . . ,

1
2

n−1
, . . .
Notation
¦a
n
¦ a
n
=
1
n
¦b
n
¦ b
n
=

1
2

n−1
A series of numbers:
1 +
1
2
+
1
3
+ +
1
n
+ =

¸
n=0
1
n

¸
n=0
a
n
(2.1)
Does this infinite sum yield a finite result?
1 +

1
2

+

1
2

2
+ +

1
2

n−1
=

¸
n=1

1
2

n−1 ∞
¸
n=1
b
n
(2.2)
Note: In order to sum to a finite number the terms of the sequence must
tend to 0 as n →∞.
Divergence Test:
lima
n
= 0 ⇒

¸
0
a
n
diverges. (2.3)
EG: a
n
= 1 1 + 1 + + 1 + →∞.
13
Lecture 2 - Preliminaries
Integral Test: Does

¸
n=1
1
n
converge?
Consider

1
dx
x
< 1 +
1
2
+
1
3
+ +
1
n
+ =

¸
n=1
1
n
(2.4)
Now

1
dx
x
= lim
T→∞
T

1
dx
x
= lim
T→∞
(ln T −ln 1) = ∞
But

1
dx
x
<

¸
n=1
1
n
Therefore

¸
n=1
1
n
= ∞.
Example: For what values of p will the series

¸
n=1
1
n
p
(2.5)
converge?
We consider the integral:

1
1
x
p
dx = lim
T→∞
T

1
dx
x
p
=

x
1−p
1−p

T
1
p = 1
ln x

T
1
p = 1
=

1
p−1
p > 1
∞ p ≤ 1
14
2.1. SEQUENCES AND SERIES OF NUMBERS:
p > 1:

¸
n=2
1
n
p
<

1
dx
x
p


¸
n=1
1
n
p
< ∞, p > 1
p ≤ 1:

1
dx
x
p
<

¸
n=1
1
n
p


¸
n=1
1
n
p
= ∞, p ≤ 1
p-Series:

¸
n=1
1
n
p
< ∞ p > 1
diverges p ≤ 1
Geometric Series - ‘the G-Series’:

¸
n=0
r
n
= 1 +r +r
2
+ +r
n
+ (2.6)
For what values of r does the G-Series converge?
15
Lecture 2 - Preliminaries
Partial Sum:
S
N
=
N
¸
n=0
r
n
= 1 +r + +r
N
=
(1 +r + +r
N
)
(1 −r)
(1 −r)
=
1 +r + +r
N
−r −r
2
− −r
N
−r
N+1
1 −r
(2.7)
=
1 −r
N+1
1 −r
If [r[ < 1 then
lim
N→∞
N
¸
n=0
r
n
= lim
N→∞
1 −r
N+1
1 −r
=
1
1 −r
(2.8)
If [r[ ≥ 1, series diverges.
G-Series:

¸
n=0
r
n
=

1
1−r
[r[ < 1
∞ [r[ ≥ 1
(2.9)
EG:

¸
n=0
1
2
n
=
1
1 −1/2
= 2, r =
1
2
. (2.10)
2.2 Absolute and Conditional Convergence:
Alternating Series Test:
¸
(−1)
n
a
n
a
n
> 0
(a) a
n+1
≤ a
n
(b) lim
n→∞
a
n
= 0
¸

¸
(−1)
n
a
n
converges:
EG:

¸
n=1
(−1)
n
n−1
converges.
Consider a series

¸
n=0
(−1)
n
a
n
.
If
¸
[a
n
[ < ∞ then
¸
(−1)
n
a
n
is said to be absolutely convergent.
16
2.3. POWER SERIES:
If
¸
[a
n
[ = ∞ but
¸
(−1)
n
a
n
< ∞ is conditionally convergent.
Ratio Test:
Consider

¸
n=0
b
n
and let lim
n→∞

b
n+1
b
n

= L. Then
¸
b
n
converges abso-
lutely if L < 1, diverges if L > 1. Test is inconclusive if L = 1.
EG 1:

¸
n=1
(−1)
n−1
2
n
n
2
.

b
n+1
b
n

=
2
n+1
/(n + 1)
2
2
n
/n
2
= 2

1 +
1
n

2
→2 series diverges. (2.11)
EG 2:

¸
n=1
n
4
e
−n
2
.

b
n+1
b
n

=
(n + 1)
4
e
−(n+1)
2
n
4
e
−n
2
=

1 +
1
n

4
e
−2n−1
→0 < 1 converges absolutely.
(2.12)
2.3 Power Series:
f(x) = a
0
+a
1
x +a
2
x
2
+ +a
n
x
n
polynomial approximation.
Idea: Extend the polynomial to include ∞ # of terms.
f(x) = a
0
+a
1
x +a
2
x
2
+ +a
n
x
n
+ Power Series
=

¸
n=0
a
n
x
n
(2.13)
EG: e
x
= 1 +
x
1!
+
x
2
2!
+
x
3
3!
+ +
x
n
n!
+ =

¸
n=0
x
n
n!
.
More General Power Series:
f(x) =

¸
n=0
a
n
(x −x
0
)
n
= a
0
+a
1
(x −x
0
) +a
2
(x −x
0
)
2
+ (2.14)
17
Lecture 2 - Preliminaries
Taylor Series - matching all the derivatives at a point:
f(x) =

¸
n=0
a
n
(x −x
0
)
n
= a
0
+a
1
(x −x
0
) +a
2
(x −x
0
)
2
+
f

(x) = a
1
+ 2a
2
(x −x
0
) + 3a
3
(x −x
0
)
2
+ +na
n
(x −x
0
)
n
+
⇒f

(x
0
) = a
1
f

(x) = 2a
2
+ 3.2a
3
(x −x
0
) + +n(n −1)a
n
(x −x
0
)
n−2
+
⇒f

(x
0
) = 2a
2
f
(3)
(x) = 3!a
3
+ 4.3.2(x −x
0
) + +n(n −1)(n −2)a
n
(x −x
0
)
n−3
+
⇒f
(3)
(x
0
) = 3!a
3
f
(n)
(x
0
) = n!a
n
⇒a
n
=
f
(n)
(x
0
)
n!
Therefore f(x) =

¸
n=0
f
(n)
(x
0
)
n!
(x −x
0
)
n
(2.15)
Alternative Form of Taylor Series:
f(x
0
+h) =

¸
n=0
f
(n)
(x
0
)
n!
h
n
(2.16)
EG 1:
e
x
=

¸
n=0
x
n
n!
sin x =

¸
n=0
(−1)
n
x
2n+1
(2n + 1)!
sinh x =

¸
n=0
x
2n+1
(2n + 1)!
(2.17)
cos x =

¸
n=0
(−1)
n
x
2n
(2n)!
cosh x =

¸
n=0
x
2n
(2n)!
e

= 1 +iθ +
(iθ)
2
2!
+
(iθ)
3
3!
+
=

1 −
θ
2
2!
+
θ
4
4!

+i

θ −
θ
3
3!
+

(2.18)
= cos θ +i sin θ
18
Chapter 3
Lecture 3 - Review of
Methods to Solve ODE
3.1 First Order ODE:
Separable Equations:
dy
dx
= P(x)Q(y) (3.1)

dy
Q(y)
=

P(x) dx +C
EG:
dy
dx
=
4y
x(y −3)

y −3
y

dy =
4
x
dx
y −3 ln [y[ = 4 ln [x[ +C (3.2)
y = ln(x
4
y
3
) +C
Ax
4
y
3
= e
y
Linear First Order Eq. - The Integrating Factor:
y

(x) +P(x)y = Q(x) (3.3)
Can we find a function F(x) to multiply (4.3) by in order to turn the left
hand side into a derivative of a product:
Fy

+FPy = FQ (3.4)
19
Lecture 3 - Review of Methods to Solve ODE
(Fy)

= Fy

+F

y = FQ (3.5)
So let F

= FP which is a separable Eq.
dF
F(x)
= P(x) dx ⇒

dF
F
=

P(x) dx +C
Therefore ln F =

P(x) dx +C (3.6)
or F = Ae
R
P(x) dx
choose A = 1
F = e
R
P(x) dx
integrating factor
Therefore
e
R
P(x) dx
y

+ e
R
P(x) dx
P(x)y = e
R
P(x) dx
Q(x)
(e
R
P(x) dx
y)

= e
R
P(x) dx
Q(x)
y(x) = e

R
P(x) dx

e
R
x
P(t) dt
Q(x) dx +C
¸
(3.7)
EG: 1
y

+ 2y = 0 (3.8)
F(x) = e
2x
⇒e
2x
y

+ e
2x
2y = (e
2x
y)

= 0
e
2x
y =?c
y(x) = Ce
−2x
3.2 Another Method - Series Solution:
Since the unknown solution y(x) is defined implicitly by (3.8) let us look for
a series solution: y(x) =

¸
n=0
a
n
x
n
.
y

=

¸
n=1
a
n
nx
n−1
(3.9)
Therefore y

+ 2y =

¸
n=1
a
n
nx
n−1
+

¸
n=0
2a
n
x
n
= 0
In the first sum let
m = n −1 n = 1 ⇒m = 0
n = m+ 1
20
3.3. SECOND ORDER CONSTANT COEFFICIENT LINEAR
EQUATIONS:
Therefore

¸
m=0
a
m+1
(m+ 1)x
m
+

¸
n=0
2a
n
x
n
= 0
n ⇔m :

¸
m=0
¦a
m+1
(m+ 1) + 2a
m
¦ x
m
= 0
(3.10)
a
m+1
= −
2
(m+1)
a
m
a
1
= −2a
0
, a
2
= +
2
2
2
1
a
0
, a
3
= −
2
3

2
2

2
1
a
0
= (−1)
3 2
3
3!
a
0
,
. . . , a
m
= (−1)
m2
m
m!
a
0
Therefore y(x) = a
0

¸
m=0
(−2x)
m
m!
= a
0
e
−2x
(3.11)
EG 2: Solve
dy
dx
+ cot(x)y = 5e
cos x
, y(π/2) = −4.
P(x) = cot x Q(x) = 5e
cos x
F(x) = e
R
cot xdx
= e
ln(sin x)
= sin x
(3.12)
Therefore sin(x)y

+ cos(x)y = (sin(x)y)

= 5e
cos x
sin x
sin(x)y = −5e
cos x
+C
y(x) = −
5e
cos x
−C
sin x
−4 = y(π/2) = −
5−C
1
⇒C = 1
Therefore y(x) =
1−5e
cos x
sin x
(3.13)
3.3 Second Order Constant Coefficient Linear Equa-
tions:
Ly = ay

+by

+cy = 0
Guess y = e
rx
y

= re
rx
y

= r
2
e
rx
Ly = (ar
2
+br +c)e
rx
= 0
Indicial Eq.:
ar
2
+br +c = 0 r = −


b
2
−4ac
2a
a(r −r
1
)(r −r
2
) = 0
(3.14)
21
Lecture 3 - Review of Methods to Solve ODE
Case I: ∆ = b
2
− 4ac > 0, r
1
= r
2
, y(x) = c
1
e
r
1
x
+ c
2
e
r
2
x
is the general
solution.
Case II: ∆ = 0, r
1
= r
2
, repeated roots Ly = a(r − r
1
)
2
e
rx
= 0. Then we
have one solution y(x) = e
r
1
x
– what about the second solution. Let
y(r, x) = e
rx
Ly(r, x) = a(r −r
1
)
2
e
rx
L

∂y
∂r
(r, x)

r=r
1
= [2a(r −r
1
)e
rx
+ 2a(r −r
1
)xe
rx
]
r=r
1
= 0
Therefore

∂y
∂r
(r, x)

r=r
1
= xe
r
1
x
is also a solution.
(3.15)
Thus y(x) = c
1
e
r
1
x
+c
2
xe
r
1
x
is the general solution.
Another Method:
Consider a small perturbation to the double root case:

r −(r
1
+)

r −(r
1
−)

= 0 (3.16)
y(x) = c
1
e
(r
1
+)x
+c
2
e
(r
1
−)x
=
e
(r
1
+)x
−e
(r
1
−)x
2
c
1
=
1
2
= −c
2
= e
r
1
x

e
x
−e
−x
2

→0
−→ xe
r
1
x
=


∂r
e
rx

r=r
1
(3.17)
Case III: Complex Conjugate Roots: ∆ = b
2
−4ac < 0
r
±
= −
b
2a
±i

4ac −b
2

1/2
= λ ±iµ
y(x) = c
1
e
(λ+iµ)x
+c
2
e
(λ−iµ)x
(3.18)
= e
λx
[Acos µx +Bsin µx] .
EG 1:
Ly = y

+y

−6y = 0
y = e
rx
(r
2
+r −6) = (r + 3)(r −2) = 0 (3.19)
y(x) = c
1
e
−3x
+c
2
e
2x
22
3.4. EULER/EQUIDIMENSIONAL EQUATIONS:
EG 2:
Ly = y

+ 6y

+ 9y = 0
y = e
rx
(r + 3)
2
= 0 (3.20)
y(x) = c
1
e
−3x
+c
2
xe
−3x
EG 3:
Ly = y

−4y

+ 13y = 0
y = e
rx
: r
2
−4r + 13 = 0
r =


16−52
2
= 2 ±3i
Therefore y(x) = e
2n
[Acos 3x +Bsin 3x] .
(3.21)
3.4 Euler/Equidimensional Equations:
Ly = x
2
y

+αxy

+βy = 0. (3.22)
Aside: Note if we let t = ln x or x = e
t
then
d
dx
=
d
dt
dt
dx

d
dt
= x
d
dx
.
d
2
dt
2
= x
d
dx

x
d
dx

= x
2
d
2
dx
2
+x
d
dx
⇒x
2
d
2
dx
2
=
d
2
dt
2

d
dt
(3.23)
Therefore ¨ y − ˙ y +α˙ y +βy = 0
¨ y + (α −1) ˙ y +βy = 0
(3.24)
y = e
rt
⇒r
2
+ (α −1)r +β = 0 Characteristic Eq.
Back to (3.22): Guess y = x
r
, y

= rx
r−1
, and y

= r(r −1)x
r−2
.
Therefore ¦r(r −1) +αr +β¦ x
r
= 0
f(r) = r
2
+ (α −1)r +β = 0 as above.
(3.25)
r
±
=
1 −α ±

(α −1)
2
−4β
2
(3.26)
Case 1: ∆ = (α −1)
2
−4β > 0 Two Distinct Real Roots r
1
, r
2
.
y = c
1
x
r
1
+c
2
x
r
2
if r
1
or r
2
< 0 [y[ →∞ as x →0. (3.27)
23
Lecture 3 - Review of Methods to Solve ODE
Case 2: ∆ = 0 Double Root (r −r
1
)
2
= 0.
y = c
1
x
r
1

∂r
L[x
r
] = L


∂r
x
r

= L[x
r
log x]

∂r
¦f(r)x
r
¦ = f

(r)x
r
+f(r)x
r
log x = 0 since f(r) = (r −r
1
)
2
.
(3.28)
General Solution: y(x) = (c
1
+c
2
log x)x
r
1
.
Check:
L(x
r
1
log x) = x
2
(x
r
log x)

+αx(x
r
log x)

+β(x
r
log x) −
= x
2

r(r −1)x
r
log x +rx
r−2
+ (r −1)x
r−2

(3.29)
+αx

rx
r−1
log x +x
r−1

+β(x
r
log x)
=
¸
r
2
+ (α −1)r +β
¸
x
r
log x +¦2r −1 +α¦ x
r
= 0
Case 3: ∆ = (α −1)
2
−4β < 0.
r
±
=
(1 −α)
2
±i
[4β −(α −1)
2
]
1/2
2
= λ ±iµ
y(x) = c
1
x
(λ+iµ)
+c
2
x
(λ−iµ)
x
r
= e
r ln x
= c
1
e
(λ+iµ) ln x
+c
2
e
(λ−iµ) ln x
(3.30)
= x
λ

c
1
e
iµln x
+c
2
e
−iµln x
¸
= A
1
x
λ
cos(µln x) +A
2
x
λ
sin(µln x)
Notes:
(1) If x < 0 replace by [x[.
(2)
w(y
1
, y
2
) =

y
1
y
2
y

1
y

2

= y
1
y

2
−y

1
y
2
(see Section 3.2 p.143)
=

x
λ
cos(µln x)
¸
log xx
λ
sin(µln x) +x
λ−1
cos(µln x)µ
¸

x
λ
log xcos(µln x) −x
λ−1
sin(µln x)µ
¸
x
λ
sin(µln x)
¸
= µx
2λ−1
independent for x = 0.
24
3.4. EULER/EQUIDIMENSIONAL EQUATIONS:
EG 1:
x
2
y

−xy

−2y = 0 y(1) = 0 ˙ y(1) = 1
y = x
r
r(r −1) −r −2 = 0 r
2
−2r −2 = 0
(r −1)
2
= 3 r = 1 ±

3
(3.31)
y = c
1
x
1+

3
+c
2
x
1−

3
y(1) = c
1
+c
2
= 0 c
2
= −c
1
y(x) = c
1

x
1+

3
−x
1−

3

(3.32)
y

(x) = c
1

1 +

3

x

3

1 −

3

x


3

x=1
= c
1
2

3 = 1
Therefore y(x) =
1
2

3

x
1+

3
−x
1−

3

. (3.33)
EG 2:
x
2
y

−3xy

+ 4y = 0 y(1) = 1 y

(1) = 0
y = x
r
=⇒ r(r −1) −3r + 4 = r
2
−4r + 4 = 0 (r −2)
2
= 0
(3.34)
y(x) = c
1
x
2
+c
2
x
2
log x
y(1) = c
1
= 1 y

(x) = 2x +c
2
[2xlog x +x]
α=1
(3.35)
= 2 +c
2
= 0
Therefore y(x) = x
2
−2x
2
log x.
25
Lecture 3 - Review of Methods to Solve ODE
26
Chapter 4
Lectures 4,5 Ordinary Points
and Singular Points
Lecture 4
Consider
P(x)y

+Q(x)y

+R(x)y = 0 Homogeneous Eq. (4.1)
Divide through by P(x):
Ly = y

+p(x)y

+q(x)y = 0 p(x) = Q/P, R/P (4.2)
4.1 An Ordinary Point:
x
0
is said to be an ordinary point of (5.2) if p(x) = Q/P and q(x) = R/P
are analytic at x
0
.
i.e. p(x) = p
0
+p
1
(x −x
0
) + =

¸
k=0
p
k
(x −x
0
)
k
q(x) = q
0
+q
1
(x −x
0
) + =

¸
k=0
q
k
(x −x
0
)
k
Note:
(1) If P, Q and R are polynomials then a point x
0
such that P(x
0
) = 0 is
an ordinary point.
27
Lectures 4,5 Ordinary Points and Singular Points
(2) If x
0
= 0 is an ordinary point then we assume
y =

¸
n=0
c
n
x
n
, y

n
=

¸
n=1
c
n
nx
n−1
, y

n
=

¸
n=2
c
n
n(n −1)x
n−2
0 = Ly =

¸
n=2
c
n
n(n −1)x
n−2
+


¸
n=0
p
n
x
n


¸
n=1
nc
n
x
n−1
(4.3)
+


¸
n=0
q
n
x
n


¸
n=0
c
n
x
n


¸
m=0
¸
(m+ 2)(m+ 1)c
m+2
+

p
0
(m+ 1)c
m+1
+ +p
m
c
1

+ (q
0
c
m
+ +q
m
c
0
)¦ x
m
= 0 (4.4)
yields a non-degenerate recursion for the c
m
.
At an ordinary point x
0
we can obtain two linearly independent solu-
tions by power series expansion.
About x
0
:
y(x) =

¸
n=0
c
n
(x −x
0
)
n
. (4.5)
(3) The radius of convergence of (4.5) is at least as large as the radius of
convergence of each of the series p(x) = Q/P q(x) = R/P.
i.e. up to the closest singularity to x
0
.
4.2 A Singular Point:
If p(x) or q(x) are not analytic at x
0
, then x
0
is said to be a singular point
of (4.2). For example if P, Q and R are polynomials and P(x
0
) = 0 and
Q(x
0
) = 0 or R(x
0
) = 0 then x
0
is a singular point.
EG:
(x −1)y

+y

= 0 (4.6)
x = 0 is an ordinary point.
x = 1 is a singular point.
Expand around the ordinary point
y(x) =

¸
n=0
c
n
x
n
, y

=

¸
n=1
nc
n
x
n−1
, y

=

¸
n=2
c
n
n(n −1)x
n−2
(4.7)
28
4.2. A SINGULAR POINT:
(x −1)

¸
n=2
c
n
n(n −1)x
n−2
+

¸
n=1
nc
n
x
n−1
= 0


¸
n=2
c
n
n(n −1)x
n−2
+

¸
n=2
c
n
¦n(n −1) +n¦ x
n−1
+c
1
= 0 (4.8)
m−1 = n −2 ⇒m = n −1 n = 2 ⇒m = 1 n = m+ 1
−c
2
2 1 +c
1
+

¸
m=2

−c
m+1
(m+ 1)m+c
m
m
2

x
m−1
= 0
c
0
Arbitrary:
c
m+1
=
m
m+ 1
c
m
m ≥ 2 c
2
=
c
1
2
c
3
=
2
3
c
2
=
c
1
3
c
4
=
3
4
c
3
=
c
1
4
. . . c
n
=
c
1
n
(4.9)
Therefore y(x) = c
0
+c
1

¸
n=1
x
n
n
.
Recall
1
1 −x
= 1 +x +x
2
+

1
1 −x
dx = −ln [1 −x[ = x +
x
2
2
+
x
3
3
+
y(x) = A+Bln [x −1[ (4.10)
But (4.6) is also an Euler Equation:
y = (x −1)
r
⇒r(r −1) +r = r
2
= 0 r = 0, 0.
y(x) = A+Bln(x −1) (4.11)
29
Lectures 4,5 Ordinary Points and Singular Points
Lecture 5
4.3 The Airy equation:
Consider the Airy equation y

= xy.
x = 0 is an ordinary point.
y =

¸
n=0
c
n
x
n
, y

=

¸
n=1
c
n
nx
n−1
, y

=

¸
n=2
c
n
n(n −1)x
n−2

¸
n=2
c
n
n(n −1)x
n−2
=

¸
n=0
c
n
x
n+1
m+ 1 = n −2 n = m+ 3 n = 2 ⇒m = −1
c
2
2x
0
+

¸
m=0

c
m+3
(m+ 3)(m+ 2) −c
m

x
m+1
= 0
c
2
= 0 c
m+3
=
cm
(m+3)(m+2)
m = 0, 1, . . .
(4.12)
(1) c
0
→c
3
→c
6
.
c
3
=
c
0
2.2
, c
6
=
c
3
6.5
=
c
0
6.5.3.2
, c
9
=
c
0
9.8.6.5.3.2
c
3n
=
c
0
(3n)(3n −1)(3n −3)(3n −4) . . . 9.8.6.5.3.2
(4.13)
y
0
(x) = 1 +
x
3
3.2
+
x
6
6.5
+ +
x
3n
(3n)(3n −1) . . . 3.2
+. . .
(2) c
1
→c
4
→c
7
→.
c
4
=
c
1
4.3
c
7
=
c
1
7.64.3
c
10
=
c
1
(10.9)(7.6)(4.3)
(4.14)
c
3n+1
=
c
1
(3n + 1)(3n)(3n −2)(3n −3) . . . (7.6)(4.3)
y
1
(x) = x +
x
4
4.3
+
x
7
7.6.4.3
+ +
x
3n+1
(3n + 1)(3n) . . . 4.3
(4.15)
y(x) = c
0
y
0
(x) +c
1
y
1
(x)
Radius of Convergence:
lim
m→∞
c
m+3
c
m
[x[
3
= ln
m→∞
[x[
3
(m+ 3)(m+ 2)
= 0 < 1 ρ = ∞. (4.16)
See B&D for expansion of Airy Solution about x
0
= 1 y(x) =
¸
a
n
(x−1)
n
.
It is useful to write x = (x −1) + 1.
y

= (x −1)y +y (4.17)
30
4.4. THE HERMITE EQUATION:
4.4 The Hermite Equation:
Ly = y

−2xy

+λy = 0.
Since x = 0 is an ordinary point let y(x) =

¸
n=0
a
n
x
n
then
Ly =

¸
n=2
a
n
n(n −1)x
n−2
−2

¸
n=1
a
n
nx
n


¸
n=0
a
n
x
n
= 0. (4.18)
m = n −2 →n = m+ 2 m ←n m ←n
n = 2 ⇒m = 0
Therefore

¸
m=1

a
m+2
(m+ 2)(m+ 1) −2a
m
m+λa
m

x
m
+ [a
2
2 +λa
0
] x
0
= 0. (4.19)
x
0
:
a
2
= −λa
0
/2 (4.20)
x
m
:
a
m+2
=
(2m−λ)a
m
(m+ 1)(m+ 2)
m ≥ 1 (4.21)
a
0
:
a
2
= −
λ
2
a
0
, a
4
=
(4 −λ)
4.3
a
2
=
(4 −λ)(−λ)
4.3.2
a
0
, a
6
=
(8 −λ)(4 −λ)(−λ)
6.5.4.3.2
a
0
a
2k
=
[4(k −1) −λ][4(k −2) −λ] . . . (−1))?a
0
(2k)!
(4.22)
y
0
= a
0
¸
1 −
λ
2
x
2
+
(λ −4)λ
4!
x
4
+
(8 −λ)(4 −λ)(−λ)
6!
x
6
+

a
1
:
a
3
=
(2 −λ)
3.2
a
1
; a
5
=
(6 −λ)
5!
(2 −λ)a
1
; a
7
=
(10 −λ)(6 −λ)(2 −λ)
7!
a
1
, . . . (4.23)
y
1
= a
1
¸
x +
(2 −λ)
3!
x
3
+
(6 −λ)(2 −λ)
5!
x
5
+
(10 −λ)(6 −λ)(2 −λ)x
7
7!
+

The general solution is of the form
y(x) = Ay
0
(x) +By
1
(x) (4.24)
Note:
31
Lectures 4,5 Ordinary Points and Singular Points
(a) If λ = 2n then the recursion yields a
m+2
= 0 = a
m+4
= for m = n.
Thus if n is an even integer then the series solution y
0
will terminate
and become a polynomial of degree n.
In this case:
y
0
(x) = a
0
¸
1 −nx
2
+n(n −2)2
2
x
4
4!
−n(n −2)(n −4)
2
3
x
6
6!
+
+(−1)
n/2
n(n −2) . . . 2.

2
n/2

x
n
n!

. (4.25)
On the other hand if n is an odd integer then the series solution y
1
(x)
will terminate and become a polynomial of degree n. In this case
y
1
(x) = a
1
¸
x −2(n −1)
x
3
3!
+ 2
2
(n −1)(n −3)
x
5
5!
−(n −1)(n −3)(n −5)2
3
x
7
7!
+ (4.26)
+(n −1)(n −3) . . . 3.1(−2)
(n−1)
2
x
n
n!

(b) For example in the special case λ = 4 = 2n then n = 2.
y
0
(x) = a
0
[1 −2x
2
]. (4.27)
32
Chapter 5
Lecture 6 - Singular points
5.1 Radius of Convergence and Nearest Singular
Points
EG. 1: (1 +x
2
)y

+ 2xy

+ 4x
2
y = 0.
(1) If we were given y(0) = 0 and y

(0) = 1 then we would want a power
series expansion of the form
y =

¸
n=0
c
n
x
n
about x
0
= 0. (5.1)
Roots of 1+x
2
= 0 are x = ±i, so we expect the radius of convergence
of the TS for
1
1+x
2
to be 1 since
1
1+x
2
= 1 −x
2
+x
4
−1
lim

a
n+2
an

= 1 ρ = 1.
(5.2)
(2) If we were given y(1) = 1, y

(1) = 0 then a power series expansion of
the form
¸
c
n
(x −1)
n
is required. In this case ρ =

2.
EG. 2: (x −1)(2x −1)y

+ 2xy

−2y = 0.
x = 0 is an ordinary point. x = 1 and x =
1
2
are singular points. One
solution of this equation is
y(x) =
1
x −1
= −(1 +x +x
2
+ ) ρ = 1. (5.3)
33
Lecture 6 - Singular points
This TS solution about the ordinary point x = 0 converges beyond the
singular point x =
1
2
.
EG: (x
2
−2x)y

+ 5(x −1)y

+ 3y = 0 y(1) = 7 y

(1) = 3.
x = 1 is an ordinary point. x = 0 is a singular point

(x −1)
2
−1

y

+
5(x −1)y

+ 3y = 0.
Let t = x −1 so that
d
dt
=
d
dx
and the equation is transformed to
(t
2
−1)
˙
˙y + 5t ˙ y + 3y = 0
y =

¸
n=0
c
n
t
n
, y

=

¸
n=1
c
n
nt
n−1
, y

=

¸
n=2
c
n
n(n −1)t
n−2

¸
n=2
n(n −1)c
n
t
n


¸
n=2
n(n −1)c
n
t
n−2
+ 5

¸
n=1
nc
n
t
n
+ 3

¸
n=0
c
n
t
n
= 0
m = n −2 n = m+ 2 n = 2 = m = 0

¸
m=2
[−c
m+2
(m+ 2)(m+ 1) +¦m(m−1) + 5m+ 3¦ c
m
] t
m
−2c
2
+ 3c
0
+ [−c
3
3.2 + 5c
1
+ 3c
1
] t = 0
t
0
> c
2
=
3
2
c
0
t
1
> c
3
=
8
6
c
1
=
4
3
c
1
t
m
> c
m+2
=
cm(m+1)(m+3)
(m+1)(m+2)
m ≥ 2.
(5.4)
c
0
:
c
4
=
5c
2
4
=
5
4

3
2

c
0
, c
6
=
7
6
c
4
=
7
6
5
4
3
2
c
0
y
0
(x) =

¸
n=0
357...(2n+1)
246...(2n)
(x −1)
2n
(5.5)
c
1
:
c
5
=
6
5
c
3
=
6
5
4
3
c
1
c
2n+1
=
46...2n+2
35...2n+1
c
1
y
1
(x) =

¸
n=0
46...2n+2
35...2n+1
(x −1)
2n+1
lim
n→∞
c
m+2
c
m
=
m+ 3
m+ 1
= 1 ρ = 1 (5.6)
y(x) = c
0
y
0
(x) +c
1
y
1
(x)
y(1) = c
0
= 7 y

(1) = c
1
= 3.
(5.7)
34
5.2. SINGULAR POINTS:
5.2 Singular Points:
Consider
P(x)y

+Q(x)y

+R(x)y = 0. (5.8)
If P, Q and R are polynomials without common factors then singular points
are points x
0
at which P(x
0
) = 0.
Note: At singular points the solution is not necessarily analytic.
Examples:
1.
x
2
y

+xy

= 0
y = x
r
→r(r −1) +r = 0 →y = c
1
+c
2
ln x
The x
2
y

admits wild behaviour.
2.
x
2
y

−2y = 0
y = x
r
→r(r −1) −2 = 0 r = −2, 1 y = c
1
x
2
+c
2
x
−1
Again the x
2
y

admits wild behaviour.
3.
x
2
y

−2xy

+ 2y = 0
y = x
r
→r(r −1) −2r + 2 = 0 r = 1, 2 y = c
1
x +c
2
x
2
In this case both solutions are analytic.
5.3 Regular Singular Points:
Notice that all these cases are equidimensional equations for which we can
identify solutions of the form x
r
or x
r
log x. There is a special class of
singular points called regular singular points in which the singularities are
no worse than those in the equidimensional equations.
y

+
α
x
y

+
β
x
2
y = 0. (5.9)
If P, Q and R are polynomials and suppose P(x
0
) = 0 then α
0
is a regular
singular point if
lim
x→x
0
(x −x
0
)
Q(x)
P(x)
and lim
x→x
0
(x −x
0
)
2
R(x)
P(x)
are finite. (5.10)
35
Lecture 6 - Singular points
I.E.
Q(x)
P(x)
=
p
0
(x −x
0
)
+p
1
+p
2
(x −x
0
) +
→ singularity no worse than
1
x−x
0
(5.11)
R(x)
P(x)
=
q
0
(x −x
0
)
2
+
q
1
(x −x
0
)
+q
2
+
→ singularity no worse than
1
(x−x
0
)
2
Examples:
1.
(1 −x
2
)y

−2xy

+ 4y = 0
P = 1 −x
2
P(±1) = 0 Q = −2x R = 4
lim
x→1
(x −1)
(−2x)
(1−x)(1+x)
= 1 lim
x→1
(x −1)
2 4
(1+x)(1−x)
= 0
(5.12)
x = 1 is a R.S.P. (similarly for x = −1).
2.
x
3
y

−y = 0
P(x) = x
3
Q = 0 R = −1
lim
x→0
x
2

−1
x
3

= ∞
(5.13)
Thus x = 0 is an irregular singular point. Actually y ∼ x
3/4
e
±2/x
1/2
as
x →0+ which is much wilder than the simple power law x
r
or x
r
log x.
Note: Any singular point that is not a regular singular point is called
an irregular singular point.
3. 2(x −2)
2
xy

+3xy

+(x −2)y = 0. Singular points at x = 0, 2. x = 0
is a regular singular point. x = 2 is an irregular singular point.
5.4 More General Definition of a Regular Singular
Point:
If P, Q, and R are not limited to polynomials then consider
P(x)y

+Q(x)y

+R(x)y = 0
or
x
2
y

+x

xQ
P

y

+

x
2
R
P

y = 0
(5.14)
36
5.4. MORE GENERAL DEFINITION OF A REGULAR SINGULAR
POINT:
x = 0 is a regular singular point if

xQ
P

and

x
2
R
P

are analytic at
x = 0. I.E.
xQ
P
= p(x) = p
0
+p
1
x + and
x
2
R
P
= q(x) = q
0
+q
1
x + . (5.15)
In this case
Ly = x
2
y

+xp
0
y

+q
0
y +
small as x→0
. .. .
x
¸
p
1
xy

+q
1
y +
¸
= 0. (5.16)
Then as x → 0 x
2
y

+ xp
0
y

+ q
0
y ≈ 0 which is an Euler Equation which
has solutions of the form y = x
r
.
Thus about a regular singular point we look for solutions of the form
y = x
r

¸
n=0
a
n
x
n
=

¸
n=0
a
n
x
n+r
.
Our task is to determine;
(i) r
(ii) the coefficients a
n
(iii) the radius of convergence.
EG. 1: x
2
y

+ 2(e
x
− 1)y

+ e
−x
cos xy = 0 P = x
2
Q = 2(e
x
− 1) R =
e
−x
cos x.
x = 0 is a singular point.
lim
x→0
xQ
P
= lim
k→0
x
2(e
x
−1)
x
2
= lim
x→0
2(e
x
−1)
x
0
0
= lim
x→0
2e
x
1
= 2 L’Hopital
lim
x→0
x
2
R
P
= lim
x→0
x
2
e
−x
cos x
x
2
= 1 < ∞. (5.17)
Since the quotient functions p = xQ/P and q = x
2
R/P have Taylor Expan-
sions about x = 0, x = 0 is a regular singular point.
37
Lecture 6 - Singular points
38
Chapter 6
Lecture 7 - Frobenius Series
about Regular Singular
Points
Example 1:
Ly = 2x
2
y

−xy

+ (1 +x)y = 0 x = 0 is a RSP.
y =

¸
n=0
a
n
x
n+r
(6.1)
Ly = 2x
2

¸
n=0
a
n
(n +r)(n +r −1)x
n+r−2
−x

¸
n=0
a
n
(n +r)x
n+r−1
+(1 +x)

¸
n=0
a
n
x
n+r
= 0

¸
n=0
a
n
¦2(n +r)(n +r −1) −(n +r) + 1¦ x
n+r
+

¸
n=0
a
n
x
n+r+1
= 0 (6.2)
m = n + 1 n = 0 →m = 1
n = m−1
Therefore a
0
¦2r(r −1) −r + 1¦ x
r
+

¸
n=1
[a
n
¦2(n +r)(n +r −1)
− (n +r) + 1¦ +a
n−1
] x
n+r
= 0.
39
Lecture 7 - Frobenius Series about Regular Singular Points
x
r
> Indicial Equation: 2r
2
−3r +1 = (2r −1)(r −1) = 0 r =
1
2
, r = 1.
a
0
arbitrary
Recursion
a
n
=
−a
n−1
(2n + 2r −3)(n +r) + 1
(6.3)
Let r = 1/2:
a
n
=
−a
n−1
(2n −2)(n + 1/2) + 1
=
−a
n−1
(n −1)(2n + 1) + 1
=
−a
n−1
n(2n −1)
n = 1 : a
1
=
−a
0
1
; n = 2 : a
2
=
−a
1
2.3
=
+a
0
2.3
a
3
=
−a
2
3.5
=
−a
0
1.(2.3)(3.5)
; a
4
=
−a
3
4.7
=
+a
0
1(2.3)(3.5)(4.7)
(6.4)
a
n
=
(−1)
n
a
0
n!1.3.5.(2n −1)
=
(−1)
n
2
(n−1)
a
0
n(2n −1)!
y
1
(x) = x
1/2

¸
n=0
(−1)
n
2
(n−1)
n(2n −1)!
x
n
r = 1:
a
n
=
−a
n−1
(2n −1)(n + 1) + 1
=
−a
n−1
(2n + 1)n
a
1
=
−a
0
3.1
, a
2
=
−a
1
5.2
=
+a
0
(1.3)(2.5)
; a
3
=
−a
2
3.7
=
−a
0
(1.3)(2.5)(3.7)
a
n
=
(−1)
n
a
0
n!3.5.7.(2n + 1)
=
(−1)
n
2
n
a
0
(2n + 1)!
(6.5)
y
2
(x) = x

¸
n=0
(−1)
n
2
n
(2n + 1)!
x
n
General Solution: y(x) = c
1
y
1
(x) +c
2
y
2
(x)
Radius of Convergence ∞.
40
6.1. SERIES EXPANSION SUMMARY:
6.1 Series Expansion Summary:
Consider
P(x)y

+Q(x)y

+R(x)y = 0 (6.6)
Divide by P(x):
y

+p(x)y

+q(x)y = 0, p(x) =
Q(x)
P(x)
, q(x) =
R(x)
P(x)
(6.7)
Ordinary Points:
x
0
is an ordinary point if p(x) and q(x) are analytic at x
0
. I.E.
p(x) = p
0
+p
1
(x −x
0
) +
q(x) = q
0
+q
1
(x −x
0
) + . (6.8)
About an ordinary point x
0
we can obtain 2 linearly independent solutions
of the form
y(x) =

¸
n=0
a
n
(x −x
0
)
n
(6.9)
whose radius of convergence is at least as large as those of p and q in (7.8)
- up to the singularity closest to x
0
.
Singular Points: If P(x
0
) = 0 then p(x) and q(x) may fail to be analytic
in which case x
0
is a singular point.
Regular Singular Points:
A point x
0
is a regular singular point if
(x −x
0
)
Q(x)
P(x)
= p
0
+p
1
(x −x
0
) +
(x −x
0
)
2
R(x)
P(x)
= q
0
+q
1
(x −x
0
) + (6.10)
are analytic at x
0
. In this case
(x −x
0
)
2
y

+ (x −x
0
)

(x −x
0
)
Q(x)
P(x)

y

+ (x −x
0
)
2
R(x)
P(x)
y = 0 (6.11)
has singularities no worse than the Euler Equation:
(x −x
0
)
2
y

+ (x −x
0
)p
0
y

+q
0
y = 0. (6.12)
In this case we look for solutions of the form
y(x) = (x −x
0
)
r

¸
n=0
a
n
(x −x
0
)
n
. (6.13)
41
Lecture 7 - Frobenius Series about Regular Singular Points
42
Chapter 7
Bessel’s Equation
Lecture 8
7.1 Bessel’s Function of Order ν / ∈ ¦. . . , −2, −1, 0, 1, 2 . . .¦:
Ly = x
2
y

+xy

+ (x
2
−ν
2
) y = 0 (7.1)
x = 0 is a regular Singular Point: therefore let y =

¸
n=0
a
n
x
n+r
.
0 =

¸
n=0
a
n

(n +r)(n +r −1) + (n +r) −ν
2

x
n+r
+

¸
n=0
a
n
x
n+r+2
(7.2)
m = n + 2 n = m−2
n = 0 ⇒m = 2
0 =

¸
m=2
¸
a
m

(m+r)
2
−ν
2

+a
m−2
¸
x
m+r
+ a
0
¸
r
2
−ν
2
¸
x
r
(7.3)
+a
1
¸
(1 +r)
2
−ν
2
¸
x
r+1
x
r
> a
0
= 0 ⇒r = ±ν Indicial Eq. Roots
x
r+1
> a
1
¸
(1 ±ν)
2
−ν
2
¸
= a
1
(1 ±2ν) = 0 provided ν =
1
2
.
x
m+r
> a
m
= −
a
m−2
(m+r)
2
−ν
2
m ≥ 2
(7.4)
43
Bessel’s Equation
r = ν:
a
m
= −
a
m−2
(m+ν)
2
−ν
2
= −
a
m−2
m
2
+ 2mν
= −
a
m−2
m(m+ 2ν)
a
2
= −
a
0
2(2 + 2ν)
= −
a
0
2
2
(1 +ν)
a
4
= −
a
2
4(4 + 2ν)
=
(−1)
2
a
0
2.2
4
(2 +ν)(1 +ν)
. . . a
2m
=
(−1)
m
a
0
m!2
2m
(1 +ν) . . . (m+ν)
(7.5)
y
1
(x) = x
ν

¸
m=0
(−1)
m
(x/2)
2m
m!(1 +ν)(2 +ν) . . . (m+ν)
x→0
→0
r = −ν:
a
m
= −
a
m−2
m(m−2ν)
a
2
= −
a
0
2(2 −2ν)
= −
a
0
2
2
(1 −ν)
, a
4
= −
a
2
4(4 −2ν)
=
(−1)
2
a
0
22
4
(1 −ν)(2 −ν)
. . . a
2m
=
(−1)
m
a
0
m!2
2m
(1 −ν) . . . (m−ν)
(7.6)
y
2
(x) = x
−ν

¸
m=0
(−1)
m
(x/2)
2m
m!(1 −ν) . . . (m−ν)
x→0
→∞
7.2 Bessel’s Function of Order ν = 0 - repeated
roots:
Ly = x
2
y +xy

+x
2
y = 0.
y =

¸
n=0
a
n
x
n+r
Ly =

¸
n=0
a
n
¸
(n +r)(n +r −1) + (n +r)
¸
x
n+r
+a
n
x
n+r+2
= 0
m = n + 2 n = m−2 (7.7)
0 =

¸
n=2

a
n
(n +r)
2
+a
n−2

x
n+r
+a
0

r(r −1) +r

x
r
+a
1

(r + 1)r +r + 1

x
r+1
= 0
The indicial equation is: a
0
r
2
= 0 r
1,2
= 0, 0 a double root.
r
1
= 0 ⇒a
1
.1 = 0 ⇒a
1
= 0.
44
7.2. BESSEL’S FUNCTION OF ORDER ν = 0 - REPEATED ROOTS:
Recursion: a
n
= −
a
n−2
(n +r)
2
n ≥ 2.
a
2
= −
a
0
2
2
; a
4
= −
a
2
4
2
=
a
0
2
2
4
2
; a
6
= −
a
4
6
2
= −
a
0
2
2
4
2
6
2
; a
8
=
a
0
2
2
4
2
6
2
8
2
(7.8)
a
2m
=
(−1)
m
2
2m
(m!)
2
a
0
(7.9)
y
1
(x) =

1 +

¸
m=1
(−1)
m
x
2m
2
2m
(m!)
2
¸
= J
0
(x)
0 5 10 15 20
−4
−3
−2
−1
0
1
x

J
0
(
x
)

a
n
d

Y
0
(
x
)



J
0
Y
0
Figure 7.1: Zeroth order bessel functions j
0
(x) and Y
0
(x)
To get a second solution
y(x, r) = a
0
x
r

1 −
x
2
(2 +r)
2
+
x
4
(2 +r)
2
(4 +r)
2
+ +
(−1)
m
x
2m
(2 +r)
2
(4 +r)
2
. . . (2m+r)
2
+

(7.10)
∂y
∂r
(x, r)

r=r
1
= a
0
log xy
1
(x) +a
0
x
r

¸
m=1
(−1)
m
x
2m

∂r

1
(2 +r)
2
. . . (2m+r)
2

.
45
Bessel’s Equation
Let
a
2m
(r) = ¦ ¦ ⇒ln a
2m
(r) = −2 ln(2 +r) −. . . −2 ln(2m+r) (7.11)
a

2m
(0) =


2
2 +r

2
4 +r

2
(2m+r)

r=0
a
2m
(0)
=

−1 −
1
2
−. . . −
1
m

a
2m
(0) = −H
m
a
2m
(0).
Let H
m
= 1 +
1
2
+ +
1
m
. Therefore
y
2
(x) = J
0
(x) ln x +

¸
m=1
(−1)
m+1
H
m
2
2m
(m!)
2
x
2m
x > 0. (7.12)
It is conventional to define
Y
0
(x) =
2
π

y
2
(x) + (γ −log 2)J
0
(x)

(7.13)
where
γ = lim
n→∞
(H
n
−log n) = 0.5772 Euler’s Constant
y(x) = c
1
J
0
(x) +c
2
Y
0
(x). (7.14)
Lecture 9
7.3 Bessel’s Function of Order ν =
1
2
:
Consider the case ν = 1/2 Ly = x
2
y

+xy

+

x
2

1
4

y = 0. Let
y =

¸
n=0
a
n
x
n+r
(7.15)
Ly =

¸
n=0
a
n

(n +r)
2

1
4

x
n+r
+

¸
n=0
a
n
x
n+r+2
= 0
m = n + 2
n = m−2
n = 0 ⇒m = 2
(7.16)
Ly = a
0

r
2

1
4

+a
1

(r + 1)
2

1
4

+

¸
n=2
¸
a
n

(n +r)
2

1
4

+a
n−2

x
n+r
= 0.
46
7.3. BESSEL’S FUNCTION OF ORDER ν =
1
2
:
Indicial Equation: r
2

1
4
= 0, r = ±
1
2
Roots differ by an integer.
Recurrence: a
n
= −
a
n−2
(n +r)
2

1
4
n ≥ 2.
r
1
= +1/2:
a
n
= −
a
n−2
(n +
1
2
)
2

1
4
= −
a
n−2
(n + 1)n
n ≥ 2;

9
4

1
4

a
1
= 0 ⇒a
1
= 0
a
2
= −
a
0
3.2
a
4
=
(−1)
2
a
0
5.4.3.2
. . . a
2n
=
(−1)
n
a
0
(2n + 1)!
y
1
(x) = x
1
2

¸
n=0
(−1)
n
x
2n
(2n + 1)!
= x

1
2

¸
n=0
(−1)
n
x
2n+1
(2n + 1)!
= x

1
2
sin x
(7.17)
r
2
= −
1
2
:
a
n
= −
a
n−2
(n −
1
2
)
2

1
4
= −
a
n−2
n(n −1)
, n ≥ 2,
n = 1 ⇒a
1


1
2
+ 1

2

1
4
¸
= a
1
.0 = 0 a
1
and a
0
arbitrary.
(7.18)
a
0
:
a
2
= −
a
0
2.1
a
4
=
(−1)
2
a
0
4.3.2.1
. . . a
2n
=
(−1)
n
a
0
(2n)!
(7.19)
a
1
:
a
3
= −
a
1
3.2
a
5
=
(−1)
2
a
1
5.4.3.2
a
2n+1
=
(−1)
n
a
1
(2n + 1)!
(7.20)
y
2
(x) = a
0
x

1
2

¸
n=0
(−1)
n
x
2n
(2n)!
+a
1
x

1
2

¸
n=0
(−1)
n
x
2n+1
(2n + 1)!
= a
0
x

1
2
cos x +a
1
x

1
2
sin x (7.21)
` included in y
1
(x).
47
Bessel’s Equation
7.4 Example - the roots differ by an integer
Let Ly = xy

−y = 0, x = 0 is a regular singular point.
y =

¸
n=0
c
n
x
n+α

¸
n=0
c
n
(n +α)(n +α −1)x
n+α−1
−c
n
x
n+α
= 0
↑ (7.22)
p −1 = n

¸
n=1
¦c
n
(x +α)(n +α −1) −c
n−1
¦ x
n+α−1
+c
0
(α −1)αx
α−1
= 0
Indicial Equation: (α −1)α = 0 α = 0, 1 differ by integer.
Recurrence Rel: c
n
=
c
n−1
(n +α)(n +α −1)
n ≥ 1.
Note: When α = 0, c
1
blows up!
Let α = 1 ⇒ c
1
=
c
0
2
, c
2
=
c
0
12
, . . ..
y
1
(x) = c
0
x

1 +
x
2
+
x
2
12
+

= c
0
u
1
(x). (7.23)
Second Solution:
¯ y(x, α) = αy(x, α) = c
0
x
α

α +
x
1 +α
+
x
2
(1 +α)(2 +α)(1 +α)
+

∂¯ y
∂α
= c
0
x
α
ln x

α +
x
1 +α
+

(7.24)
+c
0
x
α

1 −
x
(1 +α)
2

x
2
(1 +α)
2
(2 +α)
¸
2
(1 +α)
+
1
(2 +α)

+

∂¯ y
∂α

α=0
= c
0

x +
x
2
2
+
x
3
12
+

ln x +c
0

1 −x −
5
4
x
2

= c
0
u
2
.
Therefore y(x) = (A+Bln x)

x +
x
2
2
+
x
2
12
+

+B

1 −x −
5
4
x
3

.
48
Chapter 8
Separation of Variables
Lecture 10
8.1 Types of Boundary Value Problems:
Dirichlet Boundary Conditions
1. Heat Equation: α
2
= Thermal Conductivity.
• Heat Flow in a Bar •Heat Flow on a Disk
2. Wave Equation: c = Wave Speed.
• Vibration of a String
3. Laplace’s Equation:
49
Separation of Variables
Neuman Boundary Conditions: What do you expect the solution to
look like as t →∞?
Mixed Boundary Conditions:
Ice
Heat Bath u(0, t) = A u(L, t) = B Heat Bath 2.
Ice
50
8.2. SEPARATION OF VARIABLES - FOURIER SINE SERIES:
8.2 Separation of Variables - Fourier sine Series:
Consider the heat conduction in an insulated rod whose endpoints are held
at zero degree for all time and within which the initial temperature is given
by f(x).
Fourier’s Guess:
u(x, t) = X(x)T(t) (8.1)
u
t
= X(x)
˙
T(t) = α
2
u
xx
= α
2
X

(x)T(t)
÷α
2
XT:
X

(x)
X(x)
=
˙
T(t)
α
2
T(t)
= Constant = −α
2
. (8.2)
− >
˙
T(t) = −α
2
λ
2
T(t)
dT
T
= −α
2
λ
2
dt
ln [T[ = −α
2
λ
2
t +c
T(t) = De
−α
2
λ
2
t
.
(8.3)
x >
X

(x) +λ
2
X(x) = 0
Guess X(x) = e
rx
⇒(r
2

2
)e
rx
= 0 r = ±λi
(8.4)
X = c
1
e
iλx
+c
2?
e
−iλx
= Asin λx +Bcos λx.
(8.5)
Impose the boundary conditions:
0 = u(0, t) = X(0)T(t) = BT(t) ⇒B = 0
0 = u(L, t) = X(L)T(t) = (Asin λL)T(t).
(8.6)
51
Separation of Variables
Now we do not want the trivial solution so A = 0. Thus we look for values
of λ such that
sin λL = 0 ⇒λ =


L

n = 1, 2, . . . . (8.7)
Thus u
n
(x, t) = e
−α
2
(

L
)
2
t
sin

nπx
L

n = 1, 2, . . .
are all solutions of u
t
= α
2
u
xx
. (8.8)
Since (8.8) (above eq. number) is linear, a linear combination of solutions
is again a solution. Thus the most general solution is
u(x, t) =

¸
n=1
b
n
sin

nπx
L

e
−α
2
(

L
)
2
t. (8.9)
What about the initial condition u(x, 0) = f(x).
u(x, 0) = f(x) =

¸
n=1
b
n
sin

nπx
L

. (8.10)
Given f(x) we need to find the b
n
such that the infinite series of functions
¸
b
n
sin

nπx
L

agrees with f on [0, L].
Question: f(x) may not be periodic f(x + 2L) = f(x) but the series is
periodic since sin


L

(x + 2L) = sin

nπx
L

.
Answer: In fact they do agree on [0, L] and are different elsewhere.
52
8.2. SEPARATION OF VARIABLES - FOURIER SINE SERIES:
Lecture 11
How do we find the b
n
?
Observe that we have a new type of eigenvalue problem subject to
X(0) = 0 X(L) = 0. Just as in the case with matrices we obtain sequence
of eigenvalues which in this case is infinite:
λ
n
=


L

n = 1, 2, . . . (8.11)
and corresponding eigenfunctions
x
n
(x) = sin λ
n
x = sin

nπx
L

sin

πx
L

, sin

2πx
L

, sin

3πx
L

, . . .

.
(8.12)
Recall that for symmetric matrices the eigenvectors form a basis.
Aside: How do we expand a vector?
Express f in terms of the basis vectors
¸
v
1
, v
2
, v
3
¸
f = α
1
v
1

2
v
2

3
v
3
f v
k
= α
1
v
1
v
k

2
v
2
v
k

3
v
3
v
k

v
1
v
1
v
1
v
2
v
1
v
3
v
1
v
2
v
2
v
2
v
2
v
3
v
1
v
3
v
2
v
3
v
3
v
3
¸
¸

α
1
α
2
α
3
¸
¸
=

f v
1
f v
2
f v
3
¸
¸
(8.13)
If v
k
⊥ v

, k = i.e. the v
k
are orthogonal
α
k
=
f v
k
v
k
v
k
(8.14)
But functions are just infinite dimensional vectors:
53
Separation of Variables
f · [f
1
, f
2
, . . . , f
N
]
g · [g
1
, g
2
, . . . , g
N
]
f g = f
1
g
1
+f
2
g
2
+ +f
N
g
N
∆x =
L
N
(8.15)
=
N
¸
k=1
f(x
k
)g(x
k
).
Now
N
¸
k=1
f(x
k
)g(x
k
)∆x ·
L

0
f(x)g(x) dx = 'f, g`. (8.16)
Back to finding b
n
:
f(x) =

¸
n=1
b
n
sin

nπx
L

(8.17)
L

0
f(x) sin

kπx
L

dx =

¸
n=1
b
n
L

0
sin

nπx
L

sin

kπx
L

dx.
Recall sin(A) sin B =
1
2
¦cos(A−B) −cos(A+B)¦. Therefore
I
nk
=
L

0
sin

nπx
L

sin

kπx
L

dx
=
1
2
L

0
cos(n −k)
πx
L
−cos(n +k)
πx
L
dx n = k
=
1
2
¸
sin(n −k)πx/L
(n −k)π/L

sin(n +k)πx/L
(n +k)π/L

L
0
= 0 (8.18)
I
nn
=
L

0
sin
2

nπx
L

dx =
1
2
L

0
1 −cos

2nπx
L

dx
= L/2
Therefore b
k
=
2
L
L

0
f(x) sin

kπx
L

dx.
54
8.2. SEPARATION OF VARIABLES - FOURIER SINE SERIES:
Example 8.1
f(x) =

2x 0 < x <
1
2
L = 1
2(1 −x)
1
2
< x < 1
b
n
= 2

1
2

0
2xsin(nπx) dx +
1

1
2
2(1 −x) sin(nπx) dx

= 8
sin(nπ/2)
n
2
π
2
n = 1 2 3 4 5
sin


L

1 0 −1 0 1
Therefore u(x, t) =
8
π
2

¸
k=0
(−1)
k
(2k + 1)
2
sin

(2k + 1)πx

e
−(2k+1)
2
π
2
t
. (8.19)
• Observe as t →∞ u(x, t) →0 (all the heat leaks out).
• u(x, 0) =
8
π
2

¸
k=0
(−1)
k
(2k + 1)
2
sin

(2k + 1)πx

.

π
2
8
=

¸
k=0
1
(2k + 1)
2
by letting x =
1
2
⇒f(x) = 1.
−2 −1 0 1 2
−1
−0.5
0
0.5
1
1 terms of the Fourier Series
x
f
(
x
)
−2 −1 0 1 2
−1
−0.5
0
0.5
1
2 terms of the Fourier Series
x
f
(
x
)
Example 8.2
f(x) = x 0 < x < 1 L = 1
b
n
= 2
1

0
xsin(nπx) dx = −2
cos(nπ)

= 2
(−1)
n+1

Therefore u(x, t) =
2
π

¸
n=1
(−1)
n+1
n
sin(nπx)e
−(nπ)
2
t
. (8.20)
55
Separation of Variables
• As t →∞ u(x, t) →0.
• u(x, 0) =
2
π

¸
n=1
(−1)
n+1
n
sin(nπx).

u

1
2
, 0

=
1
2
=
2
π

¸
n=1
(−1)
n+1
n
sin(nπ/2)
=
2
π

¸
k=0
(−1)
k
(2k + 1)
π
4
= 1 −
1
3
+
1
5
−. . . .
k n sin


2

0 1 1
2 0
1 3 −1
4 0
2 5 1
(8.21)
−2 −1 0 1 2
−1
−0.5
0
0.5
1
1 terms of the Fourier Series
x
f
(
x
)
−2 −1 0 1 2
−1
−0.5
0
0.5
1
2 terms of the Fourier Series
x
f
(
x
)
56
8.3. HEAT EQ ON A CIRCULAR RING - FULL FOURIER SERIES
Lecture 12
8.3 Heat Eq on a Circular Ring - Full Fourier Se-
ries
Physical Interpretation: Consider a thin circular wire in which there is
no radial temperature dependence.
∂u
∂r
= 0.
u
t
= α
2
u
xx
(8.22)
BC:
u(−L, t) = u(L, t)
∂u
∂x
(−L, t) =
∂u
∂x
(L, t)
¸
Periodic BC
IC: u(x, 0) = f(x)
Assume u(x, t) = X(x)T(t).
As before:
X

(x)
X(x)
=
˙
T(t)
α
2
T(t)
= −λ
2
.
IVP:
˙
T(t)
α
2
T(t)
= −λ
2
⇒T(t) = ce
−λ
2
t
.
r =
2L

=
L
π
= Constant.
The Laplacian becomes
∆u =

2
u
∂r
2
+
1
r
∂u
∂r
+
1
r
2

2
u
∂θ
2
=

2
u
∂(rθ)
2
(8.23)
57
Separation of Variables
if we let x = rθ we obtain (1.1).
BVP:
X


2
X = 0
X(−L) = X(L)
X

(−L) = X

(L)

Eigenvalue Problem
look for λ such that
nontrivial x can be found.
X(x) = Acos(λx) +Bsin(λx)
X(−L) = Acos(λL) −Bsin(λL) = Acos(λL) +Bsin(λL) = X(L)
therefore 2Bsin(λL) = 0
X

(x) = −Aλsin λx +Bλcos(λx) (8.24)
X

(−L) = +Aλsin(λL) +Bλcos(λL) = −Aλsin(λL) +Bλcos(λL) = X

(L)
therefore 2Aλsin(λL) = 0
Therefore λ
n
L = (nπ) n = 0, 1, . . . .
Solutions to (1.1) that satisfy the BC are thus of the form
u
n
(x, t) = e
−(

L
)
2
α
2
t

A
n
cos

nπx
L

+B
n
sin

nπx
L
¸
. (8.25)
Superposition of all these solutions yields the general solution
u(x, t) = A
0
+

¸
n=1

A
n
cos

nπx
L

+B
n
sin

nπx
L
¸
e
−(

L
)
2
α
2
t
. (8.26)
In order to match the IC we have
f(x) = u(x, 0) = A
0
+

¸
n=1
A
n
cos

nπx
L

+B
n
sin

nπx
L

. (8.27)
As before we obtain expressions for the A
n
and B
n
by projecting f(x) onto
the basis functions sin

nπx
L

and cos

nπx
L

.
L

−L
f(x)

sin

mπx
L

cos

mπx
L

dx = A
0
L

−L

sin

mπx
L

cos

mπx
L

dx (8.28)
+

¸
n=1
A
n
L

−L
cos

nπx
L

sin

mπx
L

cos

mπx
L

dx
+

¸
n=1
B
n
L

−L
sin

nπx
L

sin

mπx
L

cos

mπx
L

dx.
58
8.3. HEAT EQ ON A CIRCULAR RING - FULL FOURIER SERIES
As in the previous example we use the orthogonality relations:
L

−L
sin

mπx
L

sin

nπx
L

dx = Lδ
mn
L

−L
cos

mπx
L

cos

nπx
L

dx = Lδ
mn
m and n = 0 (8.29)
= 2L m = n = 0
L

−L
sin

mπx
L

cos

nπx
L

dx = 0 ∀m, n.
Plugging these orthogonality conditions into (1.6) we obtain
A
0
=
1
2L
L

−L
f(x) dx = average value of f(x) on [−L, L]
A
n
=
1
L
L

−L
f(x) cos

nπx
L

dx and B
n
=
1
L
L

−L
f(x) sin

nπx
L

dx.

(8.30)
Note:
1. (1.6) and (1.9) [typist’s note: check re-numbering when equations are
re-labeled, these could be renamed to (9.6) and (9.9) respectively]
represent the full Fourier Series Expansion for f(x) on the interval
[−L, L].
2. By defining a
n
=
1
L
L

−L
f(x) cos

nπx
L

dx =

2A
0
A
n
and b
n
= B
n
the Fourier Series (1.6) is often written in the form
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

+b
n
sin

nπx
L

. (8.31)
59
Separation of Variables
60
Chapter 9
Lecture 13 - Fourier Series
We consider the expansion of the function f(x) of the form
f(x) ∼
a
0
2
+

¸
n=1
a
n
cos

nπx
L

+b
n
sin

nπx
L

= S(x) (9.1)
where
a
n
=
1
L
L

−L
f(x) cos

nπx
L

dx
a
0
2
=
1
2L
L

−L
f(x) dx = average value of f.
b
n
=
1
L
L

−L
f(x) sin

nπx
L

dx (9.2)
Note:
1. Note that cos


L
(x +λ)

= cos

nπx
L

provided
nπλ
L
= 2π, λ =
2L
n
and similarly sin


L
(x + 2L)

= sin

nπx
L

. Thus each of the terms
of the Fourier Series S(x) on the RHS of (10.1) is a periodic function
having a period 2L. As a result the function S(x) is also periodic.
How does this relate to f(x) which may not be periodic?
The function S(x) represented by the series is known as the periodic
extension of f on [−L, L].
61
Lecture 13 - Fourier Series
2. If f (or its periodic extension) is discontinuous at a point x
0
then S(x)
converges to the average value of f across the discontinuity.
S(x
0
) =
1
2
¸
f(x
+
0
) +f(x

0
)
¸
(9.3)
Example 9.1
f(x) =

0 −π < x < 0 L = π
x 0 ≤ x ≤ π
(9.4)
62
a
0
=
1
π
π

−π
f(x) dx =
1
π
π

0
xdx =
π
2
(9.5)
a
n
=
1
π
π

−π
f(x) cos(nx) dx
=
1
π
π

0
xcos(nx) dx
=
1
π

x
sin(nx)
n

π
0

1
n
π

0
1. sin(nx) dx

=
1
π

π sin
n
(nπ) +
1
n
2
cos(nx)

π
0

=
1
πn
2

(−1)
n
−1

n 1 2 3 4
(−1)
n
−1 −2 0 −2 0
(9.6)
a
2m+1
= −
2
π(2m+ 1)
2
m = 0, 1, 2, . . . (9.7)
b
n
=
1
π
π

−π
f(x) sin(nx) dx
=
1
π
π

0
xsin(nx) dx
=
1
π

−x
cos(nx)
n

π
0
+
1
n
π

0
1. cos(nx) dx

=
1
π

−π
cos(nπ)
n
+
0. cos 0
n
+
1
n
2
sin(nx)

π
0

= (−1)
n+1
/n (9.8)
f(x) =
a
0
2
+

¸
n=1
a
n
cos(nx) +b
n
sin(nx)
=
π
4

2
π

¸
m=0
cos

(2m+ 1)x

(2m+ 1)
2
+

¸
n=1
(−1)
n+1
sin(nx)
n
(9.9)
63
Lecture 13 - Fourier Series
9.1 It can be useful to shift the interval of inte-
gration from [−L, L] to [c, c + 2L]
Since the periodic extension f
e
(x) is periodic with period 2L (as are the
basis functions cos

nπx
L

and sin

nπx
L

).
a
n
=
1
L
L

−L
f(x) cos

nπx
L

dx =
1
L
c+2L

c
f
e
(x) cos

nπx
L

dx (9.10)
b
n
=
1
L
L

−L
f(x) sin

nπx
L

dx =
1
L
c+2L

c
f
e
(x) sin

nπx
L

dx. (9.11)
Example 9.2 Previous Example:
f(x) =

0 −π < x < 0
x 0 ≤ x ≤ π
(9.12)
On [π, 3π]
f
e
(x) =

0 π < x < 2π
x −2π 2π ≤ x ≤ 3π
(9.13)
a
n
=
1
π

π
f
e
(x) cos(nx) dx
=
1
π


(x −2π) cos(nx) dx
=
1
π
π

0
t cos(nt) dt.
t = x −2π dx = dt
x = t + 2π x = π ⇒t = −π
x = 3π ⇒t = π
since cos n(t + 2π) = cos t
(9.14)
64
9.2. COMPLEX FORM OF FOURIER SERIES
9.2 Complex Form of Fourier Series
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

+b
n
sin

nπx
L

cos

nπx
L

=
e
i(
nπx
L
)
+ e
−i(
nπx
L
)
2
; sin

nπx
L

=
e
i(
nπx
L
)
−e
−i(
nπx
L
)
2i
Therefore f(x) =
a
0
2
+

¸
n=1
a
n
2

e
i(
nπx
L
)
+ e
−i(
nπx
L
)
¸
+
b
n
2i

e
i(
nπx
L
)
−e
−i(
nπx
L
)
¸
=
a
0
2
+

¸
n=1

a
n
−ib
n
2

e
i(
nπx
L
)
+

a
n
+ib
n
2

e
−i(
nπx
L
)
(9.15)
↑ ↑ ↑
c
0
c
n
c
−n
=

¸
n=−∞
c
n
e
i(
nπx
L
)
c
n
=
a
n
−ib
n
2
=
1
2L
L

−L
f(x)

cos

nπx
L

−i sin

nπx
L
¸
dx (9.16)
=
1
2L
L

−L
f(x)e
−i(
nπx
L
)
dx b
−n
= −b
n
(9.17)
Therefore
f(x) =

¸
n=−∞
c
n
e
i(
nπx
L
)
(9.18)
c
n
=
1
2L
L

−L
f(x)e
−i(
nπx
L
)
dx. (9.19)
65
Lecture 13 - Fourier Series
Example 9.3
f(x) =

−1 −π ≤ x < 0
1 0 < x < π
L = π (9.20)
c
n
=
1


0

−L
e
−inx
dx +
π

0
e
−inx
dx

(9.21)
=
1


e
−inx

0
−π
(−in)
+
e
−inx

π
0
(−in)
¸
(9.22)
=
i
2πn
¸
−2 + e
+inπ
+ e
−inπ
¸
=

0 n even
(2/iπn) n odd
(9.23)
Therefore
f(x) =

¸
n=−∞
2
πi(2n + 1)
e
i

(2n+1)x

. (9.24)
66
Chapter 10
Lecture 14 - Even and Odd
Functions
Even: f(−x) = f(x)
Odd: f(−x) = −f(x)
10.1 Integrals of Even and Odd Functions
L

−L
f(x) dx =
0

−L
f(x) dx +
L

0
f(x) dx (10.1)
=
L

0

f(−x) +f(x)

dx (10.2)
=

2
L

0
f(x) dx f even
0 f odd.
(10.3)
Notes: Let E(x) represent an even function and O(x) an odd function.
1. If f(x) = E(x) O(x) then f(−x) = E(−x)O(−x) = −E(x)O(x) =
−f(x) ⇒f is odd.
2. E
1
(x) E
2
(x) → even.
67
Lecture 14 - Even and Odd Functions
3. O
1
(x) O
2
(x) → even.
4. Any function can be expressed as a sum of an even part and an odd
part:
f(x) =
1
2

f(x) +f(−x)

. .. .
even part
+
1
2

f(x) −f(−x)

. .. .
odd part
. (10.4)
Check: Let E(x) =
1
2

f(x) + f(−x)

. Then E(−x) =
1
2

f(−x) + f(x)

=
E(x) even. Similarly let
O(x) =
1
2

f(x) −f(−x)

(10.5)
O(−x) =
1
2

f(−x) −f(x)

= −O(x) odd. (10.6)
10.2 Consequences of Even/Odd Property for Fourier
Series
(I) Let f(x) be Even-Cosine Series:
a
n
=
1
L
L

−L
f(x) cos
. .. .
even

nπx
L

dx =
2
L
L

0
f(x) cos

nπx
L

dx(10.7)
b
n
=
1
L
L

−L
f(x) sin

nπx
L

. .. .
odd
dx = 0. (10.8)
Therefore
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

; a
n
=
2
L
L

0
f(x) cos

nπx
L

dx. (10.9)
68
10.2. CONSEQUENCES OF EVEN/ODD PROPERTY FOR FOURIER
SERIES
(II) Let f(x) be Odd-Sine Series:
a
n
=
1
L
L

−L
f(x) cos

nπx
L

. .. .
odd
dx = 0 (10.10)
b
n
=
1
L
L

−L
f(x) sin

nπx
L

. .. .
even
dx =
2
L
L

0
f(x) sin

nπx
L

dx
Therefore
f(x) =

¸
n=1
b
n
sin

nπx
L

; b
n
=
2
L
L

−0
f(x) sin

nπx
L

dx.
(III) Since any function can be written as the sum of an even and odd part,
we can interpret the cos and sin series as even/odd:
f(x) =
even odd
1
2

f(x) +f(−x)

+
1
2

f(x) −f(−x)

(10.11)
=

a
0
2
+

¸
n=1
a
n
cos

nπx
L

¸
+


¸
n=1
b
n
sin

nπx
L

¸
where
a
n
=
2
L
L

0
1
2

f(x) +f(−x)

cos

nπx
L

dx =
1
L
L

−L
f(x) cos

nπx
L

dx
b
n
=
2
L
L

0
1
2

f(x) −f(−x)

sin

nπx
L

dx =
1
L
L

−L
f(x) sin

nπx
L

dx.
69
Lecture 14 - Even and Odd Functions
10.3 Half-Range Expansions
If we are given a function f(x) on an interval [0, L] and we want to represent
f by a Fourier Series we have two choices - a Cosine Series or a Sine Series.
Cosine Series:
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

(10.12)
a
n
=
2
L
L

0
f(x) cos

nπx
L

dx. (10.13)
Sine Series:
f(x) =

¸
n=1
b
n
sin

nπx
L

(10.14)
b
n
=
2
L
L

0
f(x) sin

nπx
L

dx. (10.15)
Example 10.1 Expand f(x) = x 0 < x < 2 in a half-range (a) Sine Series,
(b) Cosine Series.
(a)
70
10.3. HALF-RANGE EXPANSIONS
b
n
=
2

0
f(t) sin

t dt (10.16)
=
2

0
t sin

2
t dt (10.17)
= −
t cos

2
t


2

2
0
+
2

2

0
cos

2
t dt (10.18)
= −
4

cos(nπ) +

2

2
sin


2
t

2
0
(10.19)
= −
4

(−1)
n
f(1) = 1 =
4
π

¸
n=1
(−1)
n+1
n
sin


2

π
4
= 1 −
1
3
+
1
5

1
7
+
(10.20)
Therefore
f(t) =
4
π

¸
n=1
(−1)
n+1
n
sin


2
t

. (10.21)
(b)
a
0
=
1
2
2

0
t dt =
1
2
t
2
2

2
0
= 1 (10.22)
a
n
=
2

0
t cos

2
t dt =

2

t sin

2
t

2
0

2

2

0
sin

2
t dt
= +

2

2
cos

2
t

2
0
=
4
n
2
π
2
¦cos nπ −1¦ (10.23)
71
Lecture 14 - Even and Odd Functions
Therefore
f(t) = 1 +
4
π
2

¸
n=1

(−1)
n
−1

n
2
cos

2
t (10.24)
= 1 −
8
π
2

¸
n=0
cos
(2n + 1)
2
πt/(2n + 1)
2
. (10.25)
The cosine series converges faster than Sine Series.
f(2) = 2 = 1 +
8
π
2

¸
n=0
1
(2n + 1)
2
π
2
8
= 1 +
1
3
2
+
1
5
2
+
72
Chapter 11
Lecture 15 - Convergence of
Fourier Series
Example 11.1 (Completion of problem illustrating Half-range Expansions)
Periodic Extension: Assume that f(x) = x, 0 < x < 2 represents one full
period of the function so that f(x + 2) = f(x). 2L = 2 ⇒L = 1.
a
0
=
1
L
L

−L
f(x) dx =
1

−1
f(x) dx =
2

0
xdx =
x
2
2

2
0
= 2 (11.1)
since f(x + 2) = f(2). (11.2)
73
Lecture 15 - Convergence of Fourier Series
n ≥ 1:
a
n
=
1
L
L

−L
f(x) cos

nπx
L

dx =
1

−1
f(x) cos(nπx) dx L = 1
=
2

0
xcos(nπx) dx
=

xsin(nπx)

2
0

1

2

0
sin(nπx) dx
¸
¸
=
1
(nπ)
2
cos(nπx)

2
0
=
1
(nπ)
2

cos(2nπ) −1

= 0 (11.3)
b
n
=
1
L
L

−L
f(x) sin

nπx
L

dx =
1

−1
f(x) sin(nπx) dx
=
2

0
xsin(nπx) dx =

−x
cos(nπx)

2
0
+
1
(nπ)
2

0
cos(nπx) dx
¸
¸
=
−2

+
sin(nπx)
(nπ)
2

2
0
=

−2

(11.4)
Therefore
f(x) =
2
2

2
π

¸
n=1
sin(nπx)
n
(11.5)
= 1 −
2
π

¸
n=1
sin(nπx)
n
(11.6)
74
−4 −2 0 2 4
−1
0
1
2
3
x
S
(
x
)
Figure 11.1: Full Range Expansion S
N
(x) = 1 −
2
π
N=20
¸
n=1
sin(nπx)
n
−4 −2 0 2 4
−2
−1
0
1
2
x
S
(
x
)

1
Figure 11.2: Full Range Expansion S
N
(x) −1 = −
2
π
N=20
¸
n=1
sin(nπx)
n
75
Lecture 15 - Convergence of Fourier Series
11.1 Convergence of Fourier Series
• What conditions do we need to impose on f to ensure that the Fourier
Series converges to f.
• We consider piecewise continuous functions:
Theorem 11.2 Let f and f

be piecewise continuous functions on [−L, L]
and let f be periodic with period 2L, then f has a Fourier Series
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

+b
n
sin

nπx
L

= S(x)
where
a
n
=
1
L
L

−L
f(x) cos

nπx
L

dx and b
n
=
1
L
L

−L
f(x) sin

nπx
L

dx.
(11.7)
The Fourier Series converges to f(x) at all points at which f is continuous
and to
1
2

f(x+) +f(x−)

at all points at which f is discontinuous.
• Thus a Fourier Series converges to the average value of the left and
right limits at a point of discontinuity of the function f(x).
76
11.2. ILLUSTRATION OF THE GIBBS PHENOMENON
11.2 Illustration of the Gibbs Phenomenon
• Near points of discontinuity truncated Fourier Series exhibit oscilla-
tions - overshoot.
−2 −1 0 1 2
−1.5
−1
−0.5
0
0.5
1
1.5
x/π
S
N
(
x
)

f
o
r

N
=
5
Figure 11.3: Fourier Series for a step function
Example 11.3 Consider the half-range sine series expansion of
f(x) = 1 on [0, π]. (11.8)
f(x) = 1 =

¸
n=1
b
n
sin(nx)
where b
n
=
2
π
π

0
sin(nx) dx =
2
π


cos nx
n

π
0
=
2
πn

1 −(−1)
n

=

4/πn n odd
0 n even
Therefore f(x) =
4
π

¸
n=1
n odd
sin(nx)
n
=
4
π

¸
m=0
sin(2m+1)x
(2m+1)
.
(11.9)
Note:
1. f(π/2) = 1 =
4
π

¸
m=0
sin

(2m+ 1)π/2

(2m+ 1)
=
4
π

1 −
1
3
+
1
5

. There-
fore
π
4
= 1 −
1
3
+
1
5
− .
77
Lecture 15 - Convergence of Fourier Series
2. Recall the complex Fourier Series example for the function
f(x) =

−1 −π ≤ x < 0
1 0 < x < π
(11.10)
which turns out to be equivalent to the odd extension of the above
function represented by the half-range sine expansion, which we can
see from the following calculation
f(x) =

¸
n=−∞
n odd
2
πin
e
inx
=
4
π

¸
n=1
n odd
e
inx
−e
−inx
2in
=
4
π

¸
n=1
n odd
sin(nx)
n
.
(11.11)
11.3 Now consider the sum of the first N terms
S
N
(x) =
4
π
N
¸
m=0
sin(2m+ 1)x
(2m+ 1)
=
4
π
Im

N
¸
m=0
e
i(2m+1)x
(2m+ 1)
¸
(11.12)
S

N
(x) =
4
π
Im

N
¸
m=0
ie
i(2m+1)x
¸
(11.13)
=
4
π
Im

ie
ix
N
¸
m=0

e
i2x

m
¸
(11.14)
=
4
π
Im

ie
ix

1 + e
i2x
+ +

e
i2x

N
1 −e
i2x

(1 −e
i2x
)
¸
(11.15)
=
4
π
Im

ie
ix

1 −e
i2(N+1)x
1 −e
i2x
¸
(11.16)
=
4
π
Im

i

1 −e
i2(N+1)x
e
ix
−e
−ix
¸
(11.17)
=
2
π
Im

e
i2(N+1)x
−1
sin x
¸
(11.18)
=
2
π
sin 2(N + 1)x
sin x
. (11.19)
78
11.3. NOW CONSIDER THE SUM OF THE FIRST N TERMS
Therefore
t = 2(N + 1)u du =
dt
2(N+1)

S
N
(x) =
2
π
x

0
sin 2(N+1)u
sin u
du ·
2
π
2(N+1)x

0
sin
t
t dt
(11.20)
0
0
.
5
1
1
.
5
2

1
0

5 0 5
1
0
x
/
π
(2/π) sin(2(N+1)x)/sin(x)
Figure 11.4: (2/π)sin(2(N + 1)x)/sin(x) for N = 5
Observe S

N
(x) =
2
π
sin 2(N + 1)x
sin x
= 0 when 2(N + 1)x
N
= π thus the
maximum value of S
N
(x) occurs at
x
N
=
π
2(N + 1)
(11.21)
79
Lecture 15 - Convergence of Fourier Series
0
0
.
5
1
1
.
5
2

1
.
5

1

0
.
5 0
0
.
5 1
1
.
5
x
/
π
(2/π) ∫
0
x
sin 2(N+1)u /sin u du for N= 5
Figure 11.5: Integral of (2/π)sin(2(N + 1)x)/sin(x)
80
Chapter 12
Lecture 16 - Parseval’s
Identity
Lemma 12.1 (A version of Parseval’s Identity)
Let f(x) =

¸
n=1
b
n
sin

nπx
L

0 < x < L. Then
2
L
L

0

f(x)

2
dx =

¸
n=1
b
2
n
.
Proof:
L

0

f(x)

2
dx =

¸
m=1

¸
n=1
b
m
b
n
L

0
sin

mπx
L

sin

nπx
L

dx (12.1)
=

¸
m=1

¸
n=1
b
m
b
n
δ
mn

L
2
=
L
2

¸
n=1
b
2
n
. (12.2)
For a full Fourier Series on [−L, L] Parseval’s Theorem assumes the form:
f(x) =
a
0
2
+

¸
n=1
a
n
cos

nπx
L

+b
n
sin

nπx
L

(12.3)
1
L
L

−L

f(x)

2
dx =
a
2
0
2
+

¸
n=1
a
2
n
+b
2
n
. (12.4)
Example 12.2 Recall for x ∈ [0, 2] f(x) = x =
4
π

¸
n=1
(−1)
n+1
n
sin

nπx
2

.
81
Lecture 16 - Parseval’s Identity
Therefore
2
L
L

0

f(x)

2
dx =
2
2
2

0
x
2
dx =

4
π

2

¸
n=1
1
n
2

x
3
3

2
0
=

4
π

2

¸
n=1
1
n
2
π
2
6
=

¸
n=1
1
n
2
(12.5)
Note:

¸
n=1
1
(2n)
2
=
1
2
2

¸
n=1
1
n
2
=
1
4

π
2
6

=
π
2
24
.
Also note that
evens odds
π
2
6
=

¸
n=1
1
n
2
=

¸
m=1
1
(2m)
2
+

¸
m=0
1
(2m+1)
2
=
π
2
24
+

¸
m=0
1
(2m+1)
2
Therefore

¸
m=0
1
(2m+ 1)
2
=
π
2
6

π
2
24
=
π
2
8
. (12.6)
12.1 Geometric Interpretation of Parseval’s For-
mula
f = b
1
ˆe
1
+b
2
ˆe
2
(12.7)
[f [
2
= f f = b
2
1
ˆe
1
ˆe
1
+b
2
2
ˆe
2
ˆe
2
(12.8)
= b
2
1
+b
2
2
Pythagoras’ Theorem (12.9)
For Fourier Sine Components:
2
L
L

0

f(x)

2
dx =

¸
n=1
b
2
n
. (12.10)
82
12.1. GEOMETRIC INTERPRETATION OF PARSEVAL’S FORMULA
Example 12.3 Consider f(x) = x
2
−π < x < π. The Fourier Series
Expansion is:
x
2
=
π
2
3
+ 4

¸
n=1
(−1)
n
n
2
cos(nx). (12.11)
n 1 2 3 4
cos


2

0 −1 0 1
Let
x =
π
2

π
2
4
=
π
2
3
+ 4

¸
n=1
(−1)
n
n
2
cos


2


π
2
12
= 4

¸
k=1
(−1)
k
(2k)
2
(12.12)
Therefore
π
2
12
=

¸
k=1
(−1)
k+1
k
2
. (12.13)
By Parseval’s Formula:
2
π
π

0
x
4
dx = 2

π
2
3

2
+ 16

¸
n=1
1
n
4
2
π
x
5
5

π
0
=

4
9
+ 16

¸
n=1
1
n
4
9−5
45
=
4
45
=
8
90
1
90
(12.14)
Therefore
π
4
90
=

¸
n=1
1
n
4
= δ?(4). (12.15)
83
Lecture 16 - Parseval’s Identity
84
Chapter 13
Lecture 17 - Solving the heat
equation using finite
difference methods
13.1 Approximating the Derivatives of a Function
by Finite Differences
Recall that the derivative of a function was defined by taking the limit of a
difference quotient:
f

(x) = lim
∆x→0
f(x + ∆x) −f(x)
∆x
. (13.1)
Now to use the computer to solve differential equations we go in the opposite
direction - we replace derivatives by appropriate difference quotients. If we
assume that the function can be differentiated many times then Taylor’s
Theorem is a very useful device in determining the appropriate difference
quotient to use. For example consider
f(x + ∆x) = f(x) + ∆xf

(x) +
∆x
2
2!
f

(x) +
∆x
3
3!
f
(3)
(x) +
∆x
4
4!
f
(4)
(x) +. . . (13.2)
Re-arranging terms in (2) and dividing by ∆x we obtain
f(x + ∆x) −f(x)
∆x
= f

(x) +
∆x
2
f

(x) +
∆x
2
3!
f
(3)
(x) +. . . .
85
Lecture 17 - Solving the heat equation using finite difference methods
If we take the limit ∆x →0 then we recover (1). But for our purposes it is
more useful to retain the approximation
f(x + ∆x) −f(x)
∆x
= f

(x) +
∆x
2
f

(ξ) (13.3)
= f

(x) +O(∆x).
We retain the term
∆x
2
f

(ξ) in (3) as a measure of the error involved when
we approximate f

(x) by the difference quotient

f(x + ∆x) − f(x)

/∆x.
Notice that this error depends on how large f

is in the interval [x, x +∆x]
(i.e. on the smoothness of f) and on the size of ∆x. Since we like to focus
on that part of the error we can control we say that the error term is of the
order ∆x – denoted by O(∆x). Technically a term or function E(∆x) is
O(∆x) if
E(∆x)
∆x
∆x→0
→ const.
Now the difference quotient (3) is not the only one that can be used to
approximate f

(x). Indeed if we consider the expansion of f(x −∆x):
f(x −∆x) = f(x) −∆xf

(x) +
∆x
2
2!
f

(x) −
∆x
3
3!
f
(3)
(x) +
∆x
4
4!
f
(4)
(x) +. . . . (13.4)
and we subtract (4) from (2) and divide by (2∆x) we obtain:
f(x + ∆x) −f(x −∆x)
2∆x
= f

(x) +
∆x
2
3!
f
(3)
(ξ). (13.5)
We notice that the error term associated with this form of difference ap-
proximation is O(∆x
2
), which converges more rapidly to zero as ∆x →0.
In order to obtain an approximation to f

(x) we add (2) to (4) which
upon re-arrangement and dividing by ∆x
2
leads to:
f(x + ∆x) −2f(x) +f(x −∆x)
∆x
2
= f

(x) +
1
12
∆x
2
f
(4)
(ξ). (13.6)
Due to the symmetry of the difference approximations (5) and (6) about the
expansion point x these are called central difference approximations. The
difference approximation (3) is known as a forward difference approximation.
We note that the central difference schemes (5) and (6) are second order
accurate while the forward difference scheme (3) is only O(∆x).
86
13.2. HEAT EQUATION SOLUTION BY FINITE DIFFERENCES
13.2 Heat Equation solution by Finite Differences
Consider the following initial-boundary value problem for the heat equation
∂u
∂t
= α
2

2
u
∂x
2
0 < x < 1, t > 0 (13.7)
BC: u(0, t) = 0 u(1, t) = 0 (13.8)
IC: u(x, 0) = f(x). (13.9)
The basic idea is to replace the derivatives in the heat equation by differ-
ence quotients. We consider the relationships between u at (x, t) and its
neighbours a distance ∆x apart and at a time ∆t later.
Corresponding to the difference quotient approximations introduced in
Section 1, we consider the following partial difference approximations.
Forward Difference in Time:
u(x, t + ∆t) = u(x, t) + ∆t
∂u
∂t
(x, t) +
∆t
2
2!

2
u
∂x
2
(x, t) + .
After re-arrangement and division by ∆t:
u(x, t + ∆t) −u(x, t)
∆t
=
∂u
∂t
(u, t) +O(∆t). (13.10)
Central Differences in Space:
u(x + ∆x, t) = u(x, t) + ∆x
∂u
∂x
(x, t) +
∆x
2
2!

2
u
∂x
2
(u, t) +
∆x
3
3!

3
u
∂x
3
(x, t) +
∆x
4
4!

4
u
∂x
2
(x, t) +
u(x −∆x, t) = u(x, t) −∆x
∂u
∂x
(x, t) +
∆x
2
2!

2
u
∂x
2
(x, t) −
∆x
3
3!

3
u
∂x
3
(x, t) +
∆x
4
4!

4
u
∂x
4
(x, t) + .
Adding and re-arranging:
u(x + ∆x, t) −2u(x, t) +u(x −∆x, t)
∆x
2
=

2
u
∂x
2
(x, t) +O(∆x
2
). (13.11)
Substituting (2) and (3) into (1a) we obtain
u(x, t + ∆t) −u(x, t)
∆t
= α
2

u(x + ∆x, t) −2u(x, t) +u(x −∆x, t)
∆x
2

+O(∆t, ∆x
2
).
Re-arranging:
u(x, t + ∆t) = u(x, t) +α
2

∆t
∆x
2

¦u(x + ∆x, t) −2u(x, t) +u(x −∆x, t)¦ . (13.12)
87
Lecture 17 - Solving the heat equation using finite difference methods
We subdivide the spatial interval [0, 1] into N + 1 equally spaced sample
points x
n
= n∆x. The time interval [0, T] is subdivided into M + 1 equal
time levels t
k
= k∆t. At each of these space-time sample points we introduce
approximations:
u(x
n
, t
k
) · u
k
n
.
u
u u
u
u u
u
u u

T
T
T
T
T
T
-
?
6
u
k
0
u
k
n−1
u
k
n
u
k
n+1
u
k
N−1
u
k
N
t
k+1
t
k
u
k+1
N
u
k+1
n
u
k+1
0
∆t
∆x
u
k+1
n
= u
k
n

2

∆t
∆x
2

u
k
n+1
−2u
k
n
+u
k
n−1

↑ This is implemented in the spread sheets Heat0 and Heat.
88
13.2. HEAT EQUATION SOLUTION BY FINITE DIFFERENCES
Implementing Derivative Boundary Conditions:
Assume that the boundary conditions (1b) are changed to
BC: u(0, t) = 0,
∂u
∂x
(1, t) = 0.
Consider a central difference approximation to
∂u
∂x
(1, t), where x
N
= N∆x =
1,
u(x
N
+ ∆x, t) −u(x
N
−∆x, t)
∆x
= 0.
Re-arranging we obtain:
u(x
N
+ ∆x, t) = u(x
N
−∆x, t) (∗)
Since x
N
= 1 we observe that x
N
+∆x is outside the domain we introduce
an extra column u
N+1
into which we copy the values u
N−1
. In the column
x
N
we implement the same difference approximation for the Heat Equation,
namely:
u
k+1
N
= u
k
N

2
(
∆t
∆x
2
)(u
k
N+1
−2u
k
N
+u
k
N−1
) (∗∗)
↑ This is implemented in the spread sheet Heat0f.
while u
k
N+1
= u
k
N−1
(see (*) ) since column u
k
N−1
is copied to column u
k
N+1
.
Note that this BC could be implemented another way without introducing
the additional column, by eliminating u
N+1
from (∗) and (∗∗):
u
k+1
N
= u
k
N
+ 2α
2

∆t
∆x
2

u
k
N−1
−u
k
N

.
If this latter equation is implemented at x
N
there is no need to introduce
an extra column U
N+1
or to implement the difference equation given in (**)
as the the derivative boundary condition is taken care of automatically.
Some EXERCISES and Observations: – Heat Equation
1. Change the ∆t in cell D1 from 0.001 to 0.05 and you will observe what
is known as a numerical instability. Now change ∆t to 0.00625, which
is known as the stability boundary and observe what happens. Now
let ∆t = 0.006 and observe the abrupt change in the solution - it is
much closer to what we would expect.
89
Lecture 17 - Solving the heat equation using finite difference methods
2. The instability noted in 1. above is not the only source of error in the
numerical approximation. Although numerical instability is evident for
a parameter choice that is unstable, the other type of error is present
in almost every type of numerical approximation scheme. This class of
error results from discarding the O(∆x
2
) and O(∆t) terms in (2) and
(3) when we replace derivatives in (1a) by difference quotients. This
error is known as the truncation error. To determine the truncation
error change the spread sheet to implement the initial condition
f(x) =

2x 0 < x < 1/2
2(1 −x) 1/2 ≤ x < 1
.
Now code up the Fourier Series (in another spread sheet) that is de-
rived on page 21 of the notes and compare the numerical solution to
the ‘exact’ Fourier Series solution with 50 terms. The difference be-
tween the two is mainly due to the truncation error since the round-off
error is about 10
−12
and does not grow if stable parameters are used.
3. Implement derivative boundary conditions on both endpoints x = 0 and
x = 1. Check the numerical solution against the problem solved in
HW1 #3.
90
Chapter 14
Lecture 18 - Solving
Laplace’s Equation using
finite differences
14.1 Finite Difference approximation
Consider the boundary value problem

2
u
∂x
2
+

2
u
∂y
2
= 0 0 < x, y < 1 (14.1)
BC: u(0, y) = 0; u(1, y) = 0; u(x, 0) = f(x); u(x, 1) = 0. (14.2)
-
6
6
-
-
6
?
1
u(1, y) = 0
1 x1 xn xN = 1
y0
y1
1 = yM
∆x
∆y
y
x x0
uxx + uyy = 0
u(x, 0) = f(x)
u(0, y) = 0
u(x, 1) = 0
91
Lecture 18 - Solving Laplace’s Equation using finite differences
As before we replace the second derivatives in (1a) by central difference
quotients that are second order accurate:
u(x + ∆x, y) −2u(x, y) +u(x −∆x, y)
∆x
2
=

2
u
∂x
2
(x, y) +O(∆x
2
)(14.3)
u(x, y + ∆y) −2u(x, y) +u(x, y −∆y)
∆y
2
=

2
u
∂y
2
(x, y) +O(∆y
2
). (14.4)
We partition the interval 0 ≤ x ≤ 1 into (N + 1) equally spaced nodes
x
n
= n∆x and the interval 0 ≤ y ≤ 1 into (M + 1) equally spaced nodes
y
m
= m∆y. Replacing the derivatives in (1a) by the difference quotients in
(3) and (4) and representing the mesh values at (x
n
, y
m
) by u
nm
· u(x
n
, y
m
)
we obtain:
u
n+1m
−2u
nm
+u
n−1m
∆x
2
+
u
nm+1
−2u
nm
+u
nm−1
∆y
2
= (u
xx
+u
yy
)
(xn,xm)
+O(∆x
2
, ∆y
2
).
If we choose ∆x = ∆y then we obtain
u
n+1m
+u
n−1m
+u
nm+1
+u
nm−1
−4u
nm
= 0 1 ≤ n, m ≤ (N −1), (M −1). (14.5)
u
u u
u
u
j j
j
j
j
1 1
1
1
-4
u
n−1m
unm u
n+1m
u
nm−1
u
nm+1
This is known as the finite difference ‘Stencil’ that relates u
nm
to its 4
nearest neighbours.
This is a system of (N − 1) (M − 1) unknowns for the values of u
nm
interior to the domain - recall the boundary values are already specified!
92
14.2. SOLVING THE SYSTEM OF EQUATIONS BY JACOBI
ITERATION
14.2 Solving the System of Equations by Jacobi
Iteration
This is a procedure to solve the system of Equation (3) by looping through
each of the mesh points and updating u
nm
according to (3) assuming that
the nearest neighbours already have values close to the exact solution. This
procedure is repeated until the changes that are made in each iteration falls
below a certain tolerance.
To implement this iterative procedure we observe that the discrete Laplace
Equation (5) can be re-written in the form:
u
k+1
nm
=
u
k
n+1m
+u
k
n−1m
+u
k
nm+1
+u
k
nm−1
4
(14.6)
t t t
t
t
-
?

6
average
Thus u
nm
is the average value of its nearest neighbours. Note that a new
superscript index k has been introduced to represent the nodal values at the
kth iteration. Thus iteration can be viewed as taking successive neighbour
averages until there is no change, at which point the value of u
mn
equals
the average of the values at its mesh neighbours. This mean value property
is a discrete form of a fundamental property of any solution to Laplace’s
Equation.
To implement the iterative procedure (6) on a spread sheet, go to the
Tools Menu at the top of the screen and click on the Options Tab. Then
select the Calculation Tab. Check the Iteration box. If you set the number
of iterations to 5 say, then if you start with zero values throughout the
interior of the domain (as you should if you cut and paste as demonstrated
in class), you will see the values percolate 5 cells into the domain from the
non zero boundary condition f(x) = sin(πx). You can choose a surface plot
to visualize the solution. Now hold down the F9 key and watch the solution
move to equilibrium. This iterative process essentially uses diffusion on a
pseudo time scale to take the solution to equilibrium.
EXERCISES and Notes for Laplace’s Equation:
93
Lecture 18 - Solving Laplace’s Equation using finite differences
1. Implement a 0 derivative BC along the lines x = 0 and x = 1. Plot a
cross section of the results along y = 1/2. To ensure that
∂u
∂x
(0, y) =
0 =
∂u
∂x
(1, y).
2. Implement an inhomogeneous term for Poisson’s Equation:

2
u
∂x
2
+

2
u
∂y
2
= f(x, y) 0 < x, y < 1.
Introduce finite difference quotients, assume ∆x = ∆y to arrive at the
iterative formula:
u
k+1
nm
=

u
k
n+1m
+u
k
n−1m
+u
k
nm+1
+u
k
nm−1
−∆x
2
f(x
n
, y
m
)

4
. (∗)
It may be useful to calculate the values of f
nm
on a separate sheet in
which the same cell values as those for u
nm
are maintained. Then the
values of f
nm
can be referenced in the calculation of u
nm
according to
(∗).
94
Chapter 15
Lecture 19 Further Heat
Conduction Problems:
Inhomogeneous BC
Example 15.1 Specified Temperatures
u
t
= α
2
u
xx
0 < x < L, t > 0 (15.1)
BC: u(0, t) = u
0
u(L, t) = u
1
u
0
, u
1
constants (15.2)
u(x, 0) = g(x). (15.3)
Firstly consider the steady-state solution (i.e., when u
t
= 0) which we denote
by u
α
(x). In this case (15.1) becomes
α
2
u


(x) = 0 ⇒u

(x) = A
0
x +B
0
u

(0) = B
0
= u
0
u

(L) = A
0
L +u
0
= u
1

u

(x) =

u
1
−u
0
L

x +u
0
steady state solution
.
Let u(x, t) = u

(x) +v(x, t). Substitute into (15.1)
u
t
=

u

(x) +v(x, t)

t
= α
2

u

(x) +v(x, t)

xx
⇒v
t
= α
2
v
xx
(15.4)
since

u

(x)

xx
= 0. Substitute into (15.2)
u(0, t) = u
0
= u

(0) +v(0, t) = u
0
+v(0, t) ⇒ v(0, t) = 0
u(L, t) = u
1
= u

(L) +v(L, t) = u
1
+v(L, t) ⇒ v(L, t) = 0.
95
Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC
Substitute into (15.3)
u(x, 0) = g(x) = u

(x) +v(x, 0) ⇒v(x, 0) = g(x) −u

(x).
Thus we have to solve a new problem for v which has zero BC:
v
t
= α
2
v
xx
v(0, t) = 0 = v(L, t)
v(x, 0) = g(x) −u

(x).
(15.5)
Separate variables: v(x, t) = X(x)T(t).
˙
T(t)
α
2
T(t)
=
X

(x)
X(x)
= −λ
2
= const
T(t) = ce
−λ
2
α
2
t
X


2
X = 0 X(0) = 0 = X(L) ⇒X
n
(x) = sin

nπx
L

L
λ
n
= nπ, n = 1, . . .
v(x, t) =

¸
n=1
b
n
e
−α
2
(

L
)
2
t
sin

nπx
L

(15.6)
v(x, 0) = g(x) −u

(x) =

¸
n=1
b
n
sin

nπx
L

⇒b
n
=
2
L
L

0
¦g(x) −u

(x)¦ sin

nπx
L

dx
Thus the solution to the inhomogeneous problem is:
u(x, t) = u

(x) +v(x, t) (15.7)
= u
0
+

u
1
−u
0
L

x +

¸
n=1
b
n
e
−α
2
(

L
)
2
t
sin

nπx
L

(15.8)
where
b
n
=
2
L
L

0
¦g(x) −u

(x)¦ sin

nπx
L

dx. (15.9)
Example 15.2 Alternative-eigenfunction expansion approach. In order to
solve the boundary value problem (15.1)-(15.3) we could recognize that

sin

nπx
L
¸

n=1
are eigenfunctions of the spatial operator:


2
∂x
2
(15.10)
96
along with the homogeneous Dirichlet BC v(0, t) = 0 = v(L, t). We therefore
assume an eigenfunction expansion of the form:
v(x, t) =

¸
n=1
ˆ v
n
(t) sin

nπx
L

(15.11)
∂v
∂t
=

¸
n=1
˙
ˆ v
n
(t) sin

nπx
L

and (15.12)

2
v
∂x
2
= −

¸
n−1
ˆ v
n
(t)


L

2
sin

nπx
L

(15.13)
v
t
= α
2
v
xx


¸
n=1

˙
ˆ v
n
(t) +α
2


L

2
ˆ v
n
(t)

sin

nπx
L

= 0.
(15.14)
Therefore
˙
ˆ v
n
(t) = −α
2


L

2
ˆ v
n
(t) A simple ODE for ˆ v
n
(t): (15.15)
⇒ ˆ v
n
(t) = ˆ v
n
(0)e
−α
2
(

L
)
2
t
. (15.16)
Therefore
v(x, t) =

¸
n=1
ˆ v
n
(0)e
−α
2
(

L
)
2
t
sin

nπx
L

(15.17)
v(x, 0) =

¸
n=1
ˆ v
n
(0) sin

nπx
L

= g(x) −u

(x) (15.18)
ˆ v
n
(0) =
2
L
L

0
¦g(x) −u

(x)¦ sin

nπx
L

dx (15.19)
which is the same solution as that in (15.8) above.
Example 15.3
u
t
= α
2
u
xx
0 < x < L, t > 0 (15.20)
BC: u(0, t) = u
0
u
x
(L, t) = 0 (15.21)
IC: u(x, 0) = g(x). (15.22)
Look for a steady solution: u


(x) = 0.
97
Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC
u

(x) = Ax +B u

(0) = B = u
0
u


(x) = A = 0 (15.23)
Therefore
u

(x) = u
0
. (15.24)
Let u(x, t) = u

(x) +v(x, t) = u
0
+v(x, t).
u
t
= α
2
u
xx
⇒v
t
= α
2
v
xx
(15.25)
u(0, t) = u
0
⇒u
0
= u
0
+v(0, t) ⇒v(0, t) = 0 (15.26)
u
x
(L, t) = 0 ⇒0 = v
x
(L, t) ⇒v
x
(L, t) = 0 (15.27)
u(x, 0) = u
0
+v(x, 0) = g(x) ⇒v(x, 0) = g(x) −u
0
(15.28)
Thus v(x, t) satisfies
v
t
= α
2
v
xx
(15.29)
v(0, t) = 0 = v
x
(L, t) (15.30)
v(x, 0) = g(x) −u
0
. (15.31)
u(x, 0) = u
0
+v(x, 0) = g(x) ⇒v(x, 0) = g(x) −u
0
.
We now need a solution v(x, t) = X(x)T(t) to (15.29):
˙
T(t)
α
2
T(t)
=
x

(x)
X(x)
= −λ
2
˙
T(t) = −λ
2
α
2
T(t) ⇒T(t) = ce
−λ
2
α
2
t
(15.32)
X


2
X = 0; X(0) = 0 = X

(L) (15.33)
⇒X(x) = Acos(λx) +Bsin(λx) X

(x) = −Aλsin(λx) (15.34)
+Bλcos(dx)
X(0) = A = 0 (15.35)
Therefore X

(L) = Bλcos(λL) = 0 ⇒λ
k
= (2k −1)
π
2L
(15.36)
k = 1, 2, 3, . . .
or λ = 0 which is the trivial solution. (15.37)
98
Therefore
v(x, t) =

¸
k=1
b
k
e
−λ
2
k
α
2
t
sin

(2k −1)
2L
πx

(15.38)
v(x, 0) =

¸
k=1
b
k
sin

(2k −1
2L
πx

= g(x) −u
0
(15.39)
⇒b
n
=
2
L
L

0
¦g(x) −u
0
¦ sin

(2k −1)
2L
πx

dx. (15.40)
Returning to u(x, t) = u
0
+v(x, t):
u(x, t) = u
0
+

¸
k=1
b
k
e
−α
2
λ
2
k
t
sin

(2k −1)
2L
πx

. (15.41)
Example 15.4 Heat Equation with some heat loss:
u
t
= α
2
u
xx
−u 0 < x < L, t > 0 (15.42)
BC: u(0, t) = 0 u(L, t) = u
1
(15.43)
IC: u(x, 0) = g(x). (15.44)
Look for the steady state solution u

(x):
α
2
u


−u

= 0
u

(x) = Acosh

x
α

+Bsinh

x
α

u

(0) = A = 0 u

(L) = Bsin h

L
α

= u
1
B =
u
1
sin h

L
α
.
(15.45)
Therefore
u

(x) = u
1
sinh

x
α

sinh

L
α

. (15.46)
99
Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC
Now let u(x, t) = u

(x) +v(x, t).
u
t
= α
2
u
xx
−u ⇒
u(0, t) = 0 ⇒
u(L, t) = u
1

u(x, 0) = g(x) ⇒
v
t
= α
2
v
xx
−v
0 = u

(0) +v(0, t)
u
1
= u

(L) +v(L, t) = u
1
+v(L, t)
u

(x) +v(x, 0) = g(x)

v
t
= α
2
v
xx
−v
v(0, t) = 0
v(L, t) = 0
v(x, 0) = g(x) −u

(x).
(15.47)
To solve (15.47) we separate variables v(x, t) = X(x)T(t). Therefore
˙
T(t)
T(t)
=
α
2
X

X
−1 ⇒
1
α
2

˙
T(t)
T(t)
+ 1

=
X

(x)
X(x)
= −λ
2
. (15.48)
Therefore
˙
T(t) = −(λ
2
α
2
+ 1)T(t) ⇒T(t) = ce
−(1+λ
2
α
2
)t
(15.49)
X


2
X = 0 ⇒X(x) = Acos λx +Bsin λx (15.50)
⇒X(0) = 0 ⇒A = 0 X(L) = Bsin(λL) = 0 ⇒λ
n
=


L

n = 1, 2, . . . . (15.51)
Therefore
v(x, t) =

¸
n=1
b
n
e
−(1+λ
2
n
α
2
)t
sin

πx
L

(15.52)
v(x, 0) = g(x) −u

(x) =

¸
n=1
b
n
sin

nπx
L

⇒b
n
(15.53)
=
2
L
L

0
¦g(x) −u

(x)¦ sin

nπx
L

dx. (15.54)
Therefore
u(x, t) = u
1
sinh

x
α

sin h

L
α

+

¸
n=1
b
n
e
−(1+λ
2
n
α
2
)t
sin

nπx
L

.(15.55)
Remark 15.5 Note: The −u term in the PDE is responsible for the e
−t
factor in the solution.
100
Chapter 16
Lecture 20 - Inhomogeneous
Derivative BC
Example 16.1 Inhomogeneous Derivative BC:
u
t
= α
2
u
xx
0 < x < L, t > 0 (16.1)
BC: u
x
(0, t) = A u
x
(L, t) = B (16.2)
IC: u(x, 0) = g(x). (16.3)
• Try for a steady solution: u


(x) = 0, u

(x) = αx + β, u
x
= α but
then we cannot match both BC unless A = B = α. This means that
if we are pumping and removing heat from the rod at different rates
then the temperature does not reach a steady state.
• Instead of subtracting off a steady solution we subtract a particular
solution which depends on x and t of the form:
w(x, t) = ax
2
+bx +ct (16.4)
w
t
= c = α
2
w
xx
= 2α
2
a ⇒c = 2α
2
a. (16.5)
101
Lecture 20 - Inhomogeneous Derivative BC
Then
w(x, t) = ax
2
+bx + 2α
2
at (16.6)
solves the heat equation.
Now we determine the constants a and b so that w(x, t) satisfies the
inhomogeneous BC:
w
x
= 2ax +b : w
x
(0, t) = b = A, w
x
(L, t) = 2aL +A = B. (16.7)
Therefore a = (B −A)/2L. Therefore
w(x, t) =
(B −A)
2L
x
2
+Ax +α
2

B −A
L

t. (16.8)
Now let
u(x, t) = w(x, t) +v(x, t). (16.9)
u
t
= w
t
+v
t
= α
2
(w
xx
+v
xx
) ⇒ v
t
= α
2
v
xx
u
x
(0, t) = A = w
x
(0, t) +v
x
(0, t) = A+v
x
(0, t) ⇒ v
x
(0, t) = 0
u
x
(L, t) = B = w
x
(L, t) +v
x
(L, t) = B +v
x
(L, t) ⇒ v
x
(L, t) = 0
u(x, 0) = g(x) = w(x, 0) +v(x, 0) ⇒ v(x, 0) = g(x) −w(x, 0)
. (16.10)
Equations (16.10) represent the homogeneous Neumann BVP seen pre-
viously. Therefore
u(x, t) =
(B −A)
2L
x
2
+Ax +α
2

B −A
L

t +
a
0
2
(16.11)
+

¸
n=1
a
n
cos

nπx
L

e
−α
2
(
nπx
L
)t
where
a
n
=
2
L
L

0

g(x) −
¸
(B −A)
2L
x
2
+Ax

cos

nπx
L

dx. (16.12)
102
Example 16.2 A bar with an external heat source s(x) = x.
u
t
= α
2
u
xx
+x 0 < x < L (16.13)
BC: u(0, t) = 0 u(L, t) = B (16.14)
IC: u(x, 0) = g(x). (16.15)
Steady state problem u
t
= 0:
0 = α
2
u


+x
u

(0) = 0 u

(L) = B
(16.16)
u


= −
x
α
2
u


= −
x
2

2
+a u

= −
x
3

2
+ax +b
u

(0) = b = 0 u

(L) = −
L
3

2
+aL = B ⇒a =
B
L
+
L
2

2
(16.17)
Therefore
u

(x) = −
x
3

2
+

B
L
+
L
2

2

x = x

B
L
+
1

2
(L
2
−x
2
)

. (16.18)
Let u(x, t) = u

(x) +v(x, t).
u
t
= α
2
u
xx
+x ⇒ (u

+v)
t
= α
2
(u↓

+v)
xx
+x↓ ⇒ v
t
= α
2
v
xx
u(0, t) = 0 ⇒ u

(0) +v(0, t) = 0 ⇒ v(0, t) = 0
u(L, t) = B ⇒ u

(L) +v(L, t) = B ⇒ v(L, t) = 0
u(x, 0) = g(x) ⇒ u

(x) +v(x, 0) = g(x) ⇒ v(x, 0) = g(x) −u

(x).
(16.19)
Separation of variables yields:
v(x, t) =

¸
n=1
b
n
e
−(

L
)
2
α
2
t
sin

nπx
L

(16.20)
103
Lecture 20 - Inhomogeneous Derivative BC
where
b
n
=
2
L
L

0
¦g(x) −u

(x)¦ sin

nπx
L

dx. (16.21)
Therefore
u(x, t) = x

B
L
+
1

2
(L
2
−x
2
)

+

¸
n=1
b
n
e
−α
2
(

L
)
2
t
sin

nπx
L

↑ ↑ (16.22)
steady transient
Note:
lim
x→∞
u(x, t) = x

B
L
+
1

2
(L
2
−x
2
)

. (16.23)
104
Chapter 17
Lecture 21 Distributed, Time
Dependent Heat Sources -
eigenfunction expansions
Example 17.1 A Bar with a Time-Varying External Heat Source:
u
t
= α
2
u
xx
+ e
−t
sin

2πx
L

0 < x < L, t > 0 (17.1)
BC: u(0, t) = 0; u(L, t) = L (17.2)
IC: u(x, 0) = x. (17.3)
Consider the function w(x) = x which satisfies the BC as well as the
homogeneous version of the PDE.
Now let u(x, t) = w(x) +v(x, t).
u
t
= (w+v)
t
= α
2
(w+v)
xx
+ e
−t
sin

2πx
L

(17.4)
⇒v
t
= α
2
v
xx
+ e
−t
sin

2πx
L

(17.5)
u(0, t) = w(0) +v(0, t) = 0 ⇒v(0, t) = 0 (17.6)
u(L, t) = w(L) +v(L, t) = L ⇒v(L, t) = 0 (17.7)
x = u(x, 0) = w(x) +v(x, 0) = x +v(x, 0) ⇒v(x, 0) = 0. (17.8)
105
Lecture 21 Distributed, Time Dependent Heat Sources - eigenfunction
expansions
Now assume that v(x, t) =

¸
n=1
ˆ v
n
(t) sin

nπx
L

.
∂v
∂t
=

¸
n=1
dˆ v
n
dt
(t) sin

nπx
L


2
v
∂x
2
=

¸
n=1
ˆ v
n
(t)


L

2
sin

nπx
L

.(17.9)
Therefore
v
t
−α
2
v
xx
=

¸
n=1

dˆ v
n
dt

2


L

2
ˆ v
n
−e
−t
δ
2n

sin

nπx
L

= 0. (17.10)
Therefore
dˆ v
n
dt

2


L

2
ˆ v
n
= e
−t
δ
2n
(17.11)
d
dt

e
α
2
(

L
)
2
t
ˆ v
n

= e
h
α
2
(

L
)
2
−1
i
t
δ
2n
. (17.12)
Therefore
e
−α
2
(

L
)
2
t
ˆ v
n
=
e
h
α
2
(

L
)
2
−1
i
t
α
2


L

2
−1
δ
2n
+c
n
c
n
arbitrary (17.13)
ˆ v
n
(t) =
e
−t
δ
2n
α
2


L

2
−1
+ e
−α
2
(

L
)
2
t
c
n
(17.14)
v(x, 0) = 0 ⇒ ˆ v
n
(0) = 0 =
δ
2n
α
2


L

2
−1
+c
n
⇒ (17.15)
c
n
=

0 n = 2

1
α
2
(

L
)
2
−1
n = 2
v(x, t) =
1
α
2


L

2
−1

e
−t
−e
−α
2
(

L
)
2
t
¸
sin

2πx
L

(17.16)
u(x, t) = x +v(x, t) = x +

e
−t
−e
−α
2
(

L
)
2
t
α
2


L

2
−1

sin

2πx
L

.
106
Example 17.2 A bar with a general external heat source s(x, t)
u
t
= α
2
u
xx
+s(x, t) (17.17)
BC: u(0, t) = A u(L, t) = B (17.18)
IC: u(x, 0) = f(x, t). (17.19)
We look for a particular solution: w(x, t) by expanding s(x, t) as a Sine
Series. Note that the sine functions are the eigenfunctions that correspond
to the homogeneous form of the BC in (17.18). Thus if we add w(x, t) to a
solution of (17.17)-(17.18) without the source (i.e. with s(x, t) = 0) we will
not affect the BC.
1. Eigenfunction Expansion:
Let
s(x, t) =

¸
n=1
ˆ s
n
(t) sin

nπx
L

(17.20)
where
ˆ s
n
(t) =
2
L
L

0
s(x, t) sin

nπx
L

dx. (17.21)
If we assume
w(x, t) =

¸
n=1
ˆ w
n
(t) sin

nπx
L

(17.22)
then
w
t
=

¸
n=1
ˆ w

n
(t) sin

nπx
L

(17.23)
w
xx
= −

¸
n=1
ˆ w
n


L

2
sin

nπx
L

. (17.24)
Therefore substituting these expansions into w
t
= α
2
w
xx
+ s(x, t) we
obtain:

¸
n=1

ˆ w

n

2


L

2
ˆ w
n
− ˆ s
n
(t)

sin

nπx
L

= 0. (17.25)
107
Lecture 21 Distributed, Time Dependent Heat Sources - eigenfunction
expansions
Therefore
ˆ w

n
(t) = −α
2


L

2
ˆ w
n
(t) + ˆ s
n
(t). (17.26)
This is a linear 1st order ODE with integrating factor e
α
2


L
«
2
t
.
Therefore
w
n
(t) =
t

0
e
−α
2
(

L
)
2
(t −τ)ˆ s
n
(τ) dτ +c
n
e
−α
2
(

L
)
2
t
(17.27)
where the c
n
are arbitrary constants. Since we are only looking for a
particular solution we choose c
n
≡ 0.
Therefore
w(x, t) =

¸
n=1

¸
t

0
e
−α
2
(

L
)
2
(t−τ)
ˆ s
n
(τ) dτ
¸

sin

nπx
L

. (17.28)
2. Now that we have a particular solution we exploit the fact that the
Problem (17.17)-(17.18) is linear and use superposition. Let
u(x, t) = w(x, t) +v(x, t) (17.29)
u
t
= w
t
+v
t
= α
2
(w
xx
+v
xx
) +s (x, t) (17.30)
⇒v
t
= α
2
v
xx
. (17.31)
A = u(0, t) = w(0, t) +v(0, t) = v(0, t) since w(0, t) = 0
B = u(0, t) = w(L, t) +v(L, t) = v(L, t) since w(L, t) = 0.
f(x) = u(x, 0) = w(x, 0) +v(x, 0) ⇒v(x, 0) = f(x) −w(x, 0) (17.32)
thus v(x, t) satisfies:
v
t
= α
2
v
xx
BC: v(0, t) = A v(L, t) = B
IC: v(x, 0) = f(x) −w(x, 0)

. (17.33)
Now the boundary value Problem (2) was solved on pg. 76 of the notes.
Therefore
u(x, t) =

B −A
L

x +A+

¸
n=1
e
−α
2
(

L
)
2
t

b
n
+
t

0
e
α
2
(

L
)
2
τ
ˆ s
n
(τ) dx

sin

nπx
L

(17.34)
108
where
b
n
=
2
L
L

0

f(x) −w(x, 0) −

(B −A)
x
L
+A
¸
sin

nπx
L

dx. (17.35)
109
Lecture 21 Distributed, Time Dependent Heat Sources - eigenfunction
expansions
110
Chapter 18
Lecture 22 More
Eigenfunction Expansions -
Time Dependent Boundary
Conditions
Example 18.1 Let us revisit the problem with inhomogeneous derivative
BC - but we will now use Eigenfunction Expansions.
u
t
= α
2
u
xx
0 < x < L, t > 0 (18.1)
BC: u
x
(0, t) = A u
x
(L, t) = B (18.2)
IC: u(x, 0) = g(x) (18.3)
First look for a function of the form h(x) = ax
2
+ bx that satisfies the
inhomogeneous BC:
h(x) = ax
2
+bx, h
x
(x) = 2ax +b
h
x
(0) = b = A h
x
(L) = 2aL +A = B ⇒a = (B −A)/2L
h(x) =

B −A
2L

x
2
+Ax.
Now let
u(x, t) = h(x) +v(x, t).
Substitute into the PDE:
u
t
=

h(x) +v(x, t)

t
= α
2
u
xx
= α
2

h(x) +v(x, t)

xx
= α
2
2a +α
2
v
xx
.
111
Lecture 22 More Eigenfunction Expansions - Time Dependent Boundary
Conditions
Therefore
v
t
= α
2
v
xx
+ 2aα
2
(18.4)
A = u
x
(0, t) = h
x
(0) +v
x
(0, t) = A+V
x
(0, t) ⇒ v
x
(0, t) = 0 (18.5)
B = u
x
(L, t) = h
x
(L) +V
x
(L, t) = B +V
x
(L, t) ⇒ v
x
(L, t) = 0 (18.6)
g(x) = u(x, 0) = h(x) +v(x, 0) ⇒ v(x, 0) = g(x) −h(x). (18.7)
We now use an Eigenfunction Expansion to solve the BVP (18.4)-(18.7).
Because of the homogeneous Neumann BC we assume an expansion of the
form
v(x, t) = ˆ v
0
(t)/2 +

¸
n=1
ˆ v
n
(t) cos

nπx
L

v
t
=
˙
ˆ v
0
(t)/2 +

¸
n=1
˙
ˆ v
n
(t) cos

nπx
L

v
x
=

¸
n=1
ˆ v
n
(t)


L
¸
sin

nπx
L

, v
xx
=

¸
n=1
ˆ v
n
(t)


L

2

cos

nπx
L

.
We also expand the inhomogeneous term in (1.4) in terms of the Eigenfunc-
tions:
2aα
2
= a
0
/2 +

¸
n=1
a
n
cos

nπx
L

a
0
= 4aα
2
, a
n
= 0 n ≥ 1.
Therefore
0 = v
t
−α
2
v
xx
−2aα
2
=
˙
ˆ v
0
(t)/2 −2aα
2
+

¸
n=1

˙
ˆ v
n

2


L

2
ˆ v
n

cos

nπx
L

.
Therefore
˙
ˆ v
0
(t) = 4aα
2
⇒ ˆ v
0
(t) = 4aα
2
t +c
0
˙
ˆ v
n
(t) = −α
2


L

2
ˆ v
n
⇒ ˆ v
n
(t) = ˆ v
n
(0)e
−α
2
(

L
)
2
t
.
Therefore
v(x, t) =
4aα
2
t +c
0
2
+

¸
n=1
ˆ v
n
(0)e
−α
2
(

L
)
2
t
cos

nπx
L

.
112
g(x) −h(x) = g(x) −

B −A
2L

x
2
+Ax

= v(x, 0) =
c
0
2
+

¸
n=1
ˆ v
n
(0) cos

nπx
L

c
0
=
2
L
1

0
¸
g(x) −

B −A
2L

x
2
+Ax

dx
ˆ v
n
(0) =
2
L
1

0
¸
g(x) −

B −A
2L

x
2
+Ax

cos

nπx
L

dx.
Thus
u(x, t) =

B −A
2L

x
2
+Ax + 2aα
2
t +
c
0
2
+

¸
n=1
ˆ v
n
(0)e
−α
2
(

L
)
2
t
cos

nπx
L

which is identical to the solution obtained in Example 16.1.
Example 18.2 Time Dependent Boundary Conditions - general case:
u
t
= α
2
u
xx
, 0 < x < L,
BC: u(0, t) = φ
0
(t) u(L, t) = φ
1
(t), (18.8)
IC: u(x, 0) = f(x). (18.9)
Let w(x, t) = φ
0
(t) +x

φ
1
(t) −φ
0
(t)
L

⇒w(0, t) = φ
0
(t); w(L, t) = φ
1
(t).
Now let u(x, t) = w(x, t) +v(x, t). Then
w
t
+v
t
= α
2
(w
xx
+v
xx
)
v
t
= α
2
v
xx
−w
t
w
t
=
˙
φ
0
+
x
L
(
˙
φ
1

˙
φ
0
)
BC: u(0, t) = φ
0
(t) = w(0, t) +v(0, t) = φ
0
(t) +v(0, t) ⇒v(0, t) = 0
u(L, t) = φ
1
(t) = w(L, t) +v(L, t) = φ
1
(t) +v(L, t) ⇒v(L, t) = 0
IC: u(x, 0) = f(x) = w(x, 0) +v(x, 0) ⇒v(x, 0) = f(x) −w(x, 0). (18.10)
Thus we need to solve the following BVP for v(x, t):
v
t
= α
2
v
xx
−w
t
BC: v(0, t) = 0 v(L, t) = 0 (18.11)
IC: v(x, 0) = f(x) −w(x, 0).
Now v(x, t) can be found using an eigenfunction expansion.
113
Lecture 22 More Eigenfunction Expansions - Time Dependent Boundary
Conditions
Specific case: Let φ
0
(t) = At; φ
1
(t) = 0 f(x) = 0. In this case
w(x, t) = At +
x
L
(0 −At) = At

1 −
x
L

. (18.12)
u
t
= α
2
u
xx
0 < x < L
BC: u(0, t) = At u(L, t) = 0 (18.13)
IC: u(x, t) = 0.
Let u(x, t) = w(x, t) +v(x, t) where w(x, t) = At

1 −
x
L

. Then
v
t
= α
2
v
xx
−A

1 −
x
L

v(0, t) = 0 = v(L, t) (18.14)
v(x, 0) = 0.
Let
s(x, γ) = −A

1 −
x
L

=

¸
n=1
ˆ s
n
(t) sin

nπx
L

ˆ s
n
=
2
L
L

0
A

x
L
−1

sin

nπx
L

dx (18.15)
= −
2A

.
Now let
v(x, t) =

¸
n=1
ˆ v
n
(t) sin

nπx
L

(18.16)
v
t
=

¸
n=1
˙
ˆ v
n
(t) sin

nπx
L

v
xx
= −

¸
n=1
ˆ v
n
(t)

nπx?
L

2
sin

nπx
L

.
Therefore
0 = v
t
−α
2
v
xx
−s(x, t) =

¸
n=1

˙
ˆ v
n
(t) +α
2


L

2
ˆ v
n
+
2A

sin

nπx
L

. (18.17)
114
Therefore
˙
ˆ v
n
(t) +α
2


L

2
ˆ v
n
(t) = −
2A

(18.18)

e

2
(

L
)
2
t
ˆ v
n
(t)

= −
2A

e
α
2
(

L
)
2
t
(18.19)
e
α
2
(

L
)
2
t
ˆ v
n
(t) = −
2AL
2
α
2
(nπ)
3
e
α
2
(

L
)
2
t
+B (18.20)
ˆ v
n
(t) = −
2AL
2
α
2
(nπ)
3
+Be
−α
2
(

L
)
2
t
(18.21)
0 = ˆ v
n
(0) = −
2AL
2
α
2
(nπ)
3
+B. (18.22)
Therefore
ˆ v
n
(t) =
2AL
2
α
2
(nπ)
3

e
−α
2
(

L
)
2
t
−1

. (18.23)
Therefore
u(x, t) = At

1 −
x
L

+
2AL
2
π
3
α
2

¸
n=1
(e
−α
2
(

L
)
2
t
−1)
n
3
sin

nπx
L

. (18.24)
115
Lecture 22 More Eigenfunction Expansions - Time Dependent Boundary
Conditions
116
Chapter 19
Lecture 23 - 1D Wave
Equation

2
u
∂t
2
= c
2

2
u
∂x
2
(19.1)

2
u
∂t
2
→ expect 2 initial conditions u(x, 0) = f(x)
∂u
∂t
(x, 0) = g(x)

2
u
∂x
2
→ expect 2 boundary conditions u(0, t) = 0
u(L, t) = 0.
(19.2)
19.1 Guitar String
Note:


∂t
+c

∂x


∂t
−c

∂x

u(x, t) = 0. (19.3)
Let w =


∂t
−c

∂x

u then
∂w
∂t
+c
∂w
∂x
= 0. 1st order wave eq. (19.4)
117
Lecture 23 - 1D Wave Equation
Thus

∂t
+c

∂x
→ right moving wave (19.5)

∂t
−c

∂x
→ left moving wave. (19.6)
Claim: u
1
(x, t) = G(x +ct) is a solution to (19.1)
u
t
= cG

u
tt
= c
2
G

(19.7)
u
x
= G

u
xx
= G

. (19.8)
Therefore
u
tt
−c
2
u
xx
= c
2
G

−c
2
G

= 0. (19.9)
Similarly u
2
(x, t) = F(x −ct) is also a solution to (19.1). Is the sum of two
solutions also a solution?
Claim u(x, t) = α
1
u
1
(x, t) + α
2
u
2
(x, t) is a solution of (19.1) if u
1
and
u
2
are solutions.

2
∂t
2

1
u
1

2
u
2
) = α
1

2
u
1
∂t
2

2

2
u
2
∂t
2
= α
1
c
2

2
u
1
∂x
2

2
c
2

2
u
2
∂x
2
since u
1
and u
2
solve (19.1)
Thus

2
∂t
2

1
u
1

2
u
2
) = c
2

2
∂x
2

1
u
1

2
u
2
).
Therefore
u(x, t) = F(x −ct) +G(x +ct) solves (19.1). (19.10)
Notes:
1. This property is due to the linearity of u
tt
= c
2
u
xx
(19.1).
2. Every solution for (19.1) on (−∞, ∞) is of this form.
118
19.1. GUITAR STRING
3. D’Alembert’s Solution Change of variables: Let
r = x +ct s = x −ct
x =
1
2
(r +s) t =
1
2c
(r −s)
. (19.11)

∂r
=

∂x
∂x
∂r
+

∂t
∂t
∂r
=
1
2c


∂t
+c

∂x

(19.12)

∂s
=

∂x
∂x
∂s
+

∂t
∂t
∂s
= −
1
2c


∂t
−c

∂x

(19.13)
Therefore
−4c
2

2
u
∂r∂s
=


∂t
+c

∂x


∂t
−c

∂x

u =

2
u
∂t
2
−c
2

2
u
∂x
2
= 0. (19.14)
Therefore

2
u
∂r∂s
(r, s) = 0 (19.15)

∂u
∂s
(r, s) =
¯
φ
1
(s) (19.16)
⇒u(r, s) =

¯
φ
1
(s) ds +φ
2
(r) = φ
1
(s) +φ
2
(r). (19.17)
Say we have the IC:
u(x, 0) = u
0
(x) displacement (19.18)
∂u
∂t
(x, 0) = v
0
(x) velocity (19.19)
u(x, t) = F(x −ct) +G(x +ct) (19.20)
u(x, 0) = F(x) +G(x) = u
0
(x) (19.21)
∂u
∂t
(x, 0) = −cF

(x) +cG

(x) = v
0
(x) (19.22)
−cF(x) +cG(x) =
x

0
v
0
(ξ) dξ +A (19.23)
¸
1 1
−c c
¸
F
G

=

u
0
x

0
v
0
(ξ) dξ +A
¸
¸
(19.24)
119
Lecture 23 - 1D Wave Equation
F =
1
2c

cu
0

¸
x

0
v
0
(ξ) dξ +A
¸

(19.25)
G =
1
2c

x

0
v
0
(ξ) dξ +A+cu
0

(19.26)
Therefore
u(x, t) =
1
2
[u
0
(x −ct) +u
0
(x +ct)] +
1
2c

x+ct
x−ct
v
0
(ξ) dξ (19.27)
D’Alembert’s Solution to the wave equation on (−∞, ∞).
120
Chapter 20
Lecture 24 - Space-Time
Interpretation of
D’Alembert’s Solution
u(x, t) =
1
2
[u
0
(x −ct) +u
0
(x +ct)] +
1
2c
x+ct

x−ct
v
0
(s)ds (20.1)
20.1 Characteristics
In the x −t plane the lines
x −ct = x
0
and x +ct = x
0
(20.2)
are called characteristics.
x −ct = x
0
⇒ t =
1
c
x −
1
c
x
0
x +ct = x
0
⇒ t = −
1
c
x +
1
c
x
0
(20.3)
x +ct = x
0
thus t = −
1
c
x
0
+
1
c
x
0

1
c
121
Lecture 24 - Space-Time Interpretation of D’Alembert’s Solution
x −ct = x
0
thus t =
1
c
x −
1
c
x
0
1
c
20.2 Region of Influence
The lines x +ct = x
0
and x −ct = x
0
bound the region of influence.
20.3 Domain of Dependence
The lines x = x
0
−ct
0
and x = x
0
+ct
0
that pass through the point (x
0
, t
0
)
bound the domain of dependence.
122
20.3. DOMAIN OF DEPENDENCE
Example 20.1 Special Case:
u(x, 0) =

1 [x[ < 1
0 [x[ > 1
(20.4)
u(x, t) =
1
2
[u
0
(x −ct) +u
0
(x +ct)] (20.5)
Let c = 1.
t =
1
2
:
x
r

1
2
= 1 ⇒ x
r
=
3
2
x
R
+
1
2
= 1 x
R
=
1
2
x


1
2
= −1 ⇒ x

= −
1
2
x
L
+
1
2
= −1 x
L
= −
3
2
(20.6)
t = 1:
x
r
−1 = 1 ⇒ x
r
= 2 x
R
+ 1 = 1 ⇒ x
R
= 1
x

−1 = −1 ⇒ x

= 0 x

+ 1 = −1 ⇒ x
L
= −2
(20.7)
t = 2:
x
r
−2 = 1 ⇒ x
r
= 3 x
R
+ 2 = 1 ⇒ x
R
= −1
x

−2 = −1 ⇒ x

= 1 x
L
+ 2 = −1 ⇒ x
L
= −3
(20.8)
123
Lecture 24 - Space-Time Interpretation of D’Alembert’s Solution
124
Chapter 21
Lecture 25 Solution by
separation of variables
Example 21.1
u
tt
= c
2
u
xx
0 < x < L, t > 0 (21.1)
BC: u(0, t) = 0, u(L, t) = 0 (21.2)
IC: u(x, 0) = f(x), u
t
(x, 0) = g(x) (21.3)
For a guitar string c =

T
0
ρ
0
.
Separate Variables u(x, t) = X(x)T(t)
¨
T(t)
c
2
T(t)
=
X

(x)
X(x)
= −λ
2
(21.4)
¨
T(t) +λ
2
c
2
T(t) = 0 ⇒ T(t) = c
1
cos(λct) +c
2
sin(λct) (21.5)
X


2
X = 0
X(0) = 0 = X(L)


X(x) = Acos(λx) +Bsin λx
X(0) = A = 0 X(L) = Bsin λL = 0

λ
n
=

L
n = 1, 2, . . .
X
n
= sin

nπx
L

.
125
Lecture 25 Solution by separation of variables
Therefore
u(x, t) =

¸
n=1
A
n
cos

nπct
L

sin

nπx
L

+B
n
sin

nπct
L

sin

nπx
L

(21.6)
u(x, 0) =

¸
n=1
A
n
sin

nπx
L

= f(x) ⇒ A
n
=
2
L
L

0
f(x) sin

nπx
L

(21.7)
u
t
(x, t) =

¸
n=1
−A
n

nπc
L

sin

nπct
L

sin

nπx
L

+B
n

nπc
L

cos

nπct
L

sin

nπx
L

(21.8)
u
t
(x, 0) =

¸
n=1
B
n

nπc
L

sin

nπx
L

= g(x) ⇒ B
n

nπc
L

=
2
L
L

0
g(x) sin

nπx
L

dx .
(21.9)
Therefore
u(x, t) =

¸
n=1

A
n
cos

nπct
L

+B
n
sin

nπct
L

sin

nπx
L

. (21.10)
21.1 Notes
1. Period and Frequency:
cos

nπc
L
(t +T)

= cos

nπct
L

provided
nπcT
L
= 2π (21.11)
thus T
n
=

2L
c

1
n
is the period (seconds per cycle) of mode n. f
n
=
1
T
n
= n

c
2L

are the natural frequencies of vibration.
2. Modes of Vibration: Standing waves of wavelength λ
n
=
2L
n
.
126
21.2. NOW WE CAN USE THE TRIGONOMETRIC IDENTITIES
21.2 Now we can use the trigonometric identities
sin(A±B) = sin Acos B ±cos Asin B; cos(A±B) = cos Acos B
∓sin Asin B (21.12)
to interpret the solution (21.10).
cos

nπct
L

sin

nπx
L

=
1
2

sin

L
(x +ct) + sin


L

(x −ct)
¸
(21.13)
sin

nπx
L

sin

nπct
L

=
1
2

cos

L
(x −ct) −cos

L
(x +ct)
¸
. (21.14)
Now

¸
n=1
A
n
cos

nπct
L

sin

nπx
L

=
1
2

¸
n=1
A
n

sin


L

(x +ct)
+sin


L

(x −ct)

(21.15)
=
1
2
[f
0
(x +ct) +f
0
(x −ct)] (21.16)
where f
0
is the odd periodic extension of f.

¸
n=1
B
n
sin

nπct
L

sin

nπx
L

=
1
2

¸
n=1
B
n

cos

L
(x −ct) −cos

L
(x +ct)

. (21.17)
Let
b
n
=
2
L
L

0
g(x) sin

nπx
L

dx ⇒g(x) =

¸
n=1
b
n
sin

nπx
L

. (21.18)
Then B
n
=

L
nπc

b
n
x

0
g(s) ds =

¸
n=1
b
n

−cos

nπs
L

x
0


L
=

¸
n=1
cB
n

1 −cos

nπx
L

(21.19)
1
c
x

0
g(s) ds =

¸
n=1
B
n
= −

¸
n=1
B
n
cos

nπx
L

. (21.20)
127
Lecture 25 Solution by separation of variables
Therefore

¸
n=1
B
n
sin

nπct
L

sin

nπx
L

=
1
2c

0

x−ct
g
0
(s) ds +
x+ct

0
g
0
(s) ds

. (21.21)
Therefore
u(x, t) =
1
2
[f
0
(x +ct) +f
0
(x −ct)] +
1
2c
x+ct

x−ct
g
0
(s) ds (21.22)
where f
0
and g
0
are the odd periodic extensions of f and g on [0, L] i.e.
f
0
(x) =

f(x) 0 < x < L and f
0
(x + 2L) = f
0
(x
0
)
−f(−x) −L < x < 0
(21.23)
g
0
(x) =

g(x) 0 < x < L and g
0
(x + 2L) = g
0
(x)
−g(−x) −L < x < 0
. (21.24)
Notes:
1. Equation (21.22) above shows that the Wave Equation Solution for
a string tied down at its ends is given by D’Alembert’s Solution (see
(19.27) of the notes) in which the initial displacement function is given
by the odd periodic extension f
0
of the initial displacement of the
string, and the initial velocity function is given by the odd periodic
extension of g
0
.
2. Information is carried along the characteristic curves x + ct = const
x −ct = const.
3. Observe that the time dependence of the solution involves sin

nπct
L

and cos

nπct
L

which do not decay with time. Thus the solutions
to the Wave Equation persist with time, whereas the solutions to the
Heat Equation typically decay exponentially with time.
128
Chapter 22
Lecture 26 - Laplace’s
Equation
Steady State Solutions of the Heat or Wave Equations that do not vary with
time so that
∂u
∂t
= 0 =

2
u
∂t
2
.
2D:
∆u =

2
u
∂x
2
+

2
u
∂y
2
= 0. (22.1)
3D:
∆u =

2
u
∂x
2
+

2
u
∂y
2
+

2
u
∂z
2
= 0. (22.2)
• No initial conditions required.
• Only boundary conditions.
In Polar Coordinates: ∆u =

2
u
∂r
2
+
1
r
∂u
∂r
+
1
r
2

2
u
∂θ
2
= 0.
129
Lecture 26 - Laplace’s Equation
22.1 Summary
In this course we have studied the solution of the second order linear PDE.
∂u
∂t
= α
2
∆u

2
u
∂t
2
= c
2
∆u
∆u = 0
Heat: Parabolic T = αX
2
Wave: Hyperbolic T
2
−c
2
X
2
= A
Laplace’s Eq.: Elliptic X
2
+Y
2
= A.
(22.3)
Important:
1. These equations are second order because they have at most 2nd par-
tial derivatives.
2. These equations are all linear so that a linear combination of solutions
is again a solution.
22.2 Laplace’s Equation
(1) 2D Steady-State Heat Conduction and (2) Static Deflection of a Mem-
brane.
u
t
= α
2
(u
xx
+u
yy
) −→u(x, y, t) inside a domain D. (22.4)
• Steady-State Solution satisfies:
∆u = u
xx
+u
yy
= 0 (x, y) ∈ D (22.5)
BC: u prescribed on ∂D. (22.6)
• We consider domains D that are rectangular, circular, pizza slices.
22.3 Rectangular Domains
u(0, y) = g
1
(y) ∆u = u
xx
+u
yy
= 0 u(a, y) = g
2
(y) (22.7)
Idea
130
22.4. SOLUTION TO PROBLEM (1A) BY SEPARATION OF
VARIABLES
• We want to use separation of variables so we need homogeneous bound-
ary conditions.
• Since the equation is linear we can break the problem into simpler
problems which do have sufficient homogeneous BC and use superpo-
sition to obtain the solution to (22.7).
Pictorially:
22.4 Solution to Problem (1A) by Separation of
Variables
(1A)
u
xx
+u
yy
= 0 (22.8)
u(0, y) = 0 = u(a, y) = u(x, b); u(x, 0) = f
1
(x). (22.9)
Let
u(x, y) = X(x)Y (y). (22.10)
X

(x)Y (y) +X(x)Y

(y) = 0 (22.11)
÷XY:
X

(x) = −Y

(y) = const = ±λ
2
X(x) Y (y) BC
(22.12)
−λ
2
:
X


2
X = 0
Y

−λ
2
Y = 0
X = Acos λx +Bsin λx
Y = C cosh λx +Dsinh λx
X(0) = 0 = X(a)
Y (0) = . . . Y (b) = 0
• Because sin and cos have an ∞ # of real roots the choice −λ
2
is
good for BC’s for Problems (A) and (C).
131
Lecture 26 - Laplace’s Equation

2
:
X

−λ
2
X = 0
Y


2
Y = 0
X = Acosh(λx) +Bsinh(λx)
Y = C cos(λy) +Dsin(λy)
X(0) = . . . X(a) = . . .
Y (0) = 0 = Y (b)
. (22.13)
• Again because sin and cos have an ∞ # of real roots the choice

2
is good for BC’s for Problems (B) and (D).
Back to Solving (1A):
X(0) = 0 ⇒A = 0 (22.14)
X(a) = Bsin(λa) = 0 ⇒
λ
n
=

a
n = 1, 2, . . .
X
n
(x) = sin

nπx
a

. (22.15)
u(x, b) = X(x)Y (b) = 0 ⇒Y (b) = 0 (22.16)
Y (b) = C cosh(λb) +Dsinh(λb) = 0 ⇒c = −Dtan h(λa) (22.17)
Y (y) = −Dtan h(λb) cosh(λy) +Dsinh(λy) (22.18)
= D

sinh(λy) cosh(λb) −cosh(λy) sinh(λb)
cosh(λb)

(22.19)
=
D
cosh(λb)
sinh λ(y −b) =
¯
Dsinh λ(y −b). (22.20)
Note: We could save ourselves the time by building the BC y(b) = 0
directly into the solution by letting
Y
n
(y) =
¯
Dsinh λ
n
(y −b) (22.21)
directly.
Now the functions: u
n
(x, y) = sin

nπx
a

sinh


a
(y −b)

n = 1, 2, . . .
satisfy all the homogeneous BC of Problem (1A). In order to match
the BC u(x, 0) = f
1
(x) we need to superimpose all these solutions.
u(x, y) =

¸
n=1
B
n
sin

nπx
a

sinh


a
(y −b)

(22.22)
f
1
(x) = u(x, 0) =

¸
n=1

−B
n
sinh

nπb
a

. .. .
bn
sin

nπx
a

(22.23)
132
22.4. SOLUTION TO PROBLEM (1A) BY SEPARATION OF
VARIABLES
where
−B
n
sinh

nπb
a

= b
n
=
2
a
a

0
f
1
(x) sin

nπx
a

dx. (22.24)
Therefore
u(x, y) =

¸
n=1
B
n
sinh


a
(y −b)

sin

nπx
a

;
where B
n
= −
2
a sinh

nπb
a

a

0
f
1
(x) sin

nπx
a

dx
(22.25)
Specific Example Let f
L
(x) = 1 =

¸
n=1
b
n
sin

nπx
a

.
b
n
=
2
anπ

1 + (−1)
n+1

= −B
n
sinh

nπb
a

. (22.26)
Therefore
u(x, y) =
1
a

¸
n=1

2

[1 + (−1)
n+1
]
sinh

nπb
a
sin

nπx
a

sinh


a
(y −b)

. (22.27)
133
Lecture 26 - Laplace’s Equation
134
Chapter 23
Lecture 27 - More
Rectangular Domains and
semi-infinite strip problems
23.1 Solution to Problem (1B) by Separation of
Variables
∆u = u
xx
+u
yy
= 0 (23.1)
0 = u(x, 0) = u(x, b) = u(0, y); u(b, y) = g
2
(y) (23.2)
Let
u(x, y) = X(x)Y (y) (23.3)
X

(x)
X(x)
= −
Y

(y)
Y (y)
= ±λ
2
. (23.4)
Since we have homogeneous BC at y = 0 and y = b we want the function
Y (y) to behave like sines and cosines. So we choose const = +λ
2
X

−λ
2
X = 0
Y


2
Y = 0
X = c
1
cosh λx +c
2
sinh λx
Y = Acos(λx) +Bsin(λx)
(23.5)
135
Lecture 27 - More Rectangular Domains and semi-infinite strip problems
u(x, 0) = X(x)Y (0) = 0 ⇒Y (0) = 0 ⇒Y (0) = A = 0 (23.6)
u(x, b) = X(x)Y (b) = 0 ⇒Y (b) = 0 ⇒Y = Bsin(λb) = 0, λ
n
=

b
n = 1, 2, . . .(23.7)
Y
n
= sin

nπy
b

u(0, y) = X(0)Y (y) = 0 ⇒X(0) = c
1
= 0.
Therefore X
n
(x) = c
2
sinh

nπx
b

.
Therefore u
n
(x, y) = sin

nπy
b

sinh

nπx
b

satisfy the homogeneous BC.
Therefore u(x, y) =

¸
n=1
c
n
sin h

nπx
b

sin

nπy
b

.
Now to satisfy the inhomogeneous BC
g
2
(y) = u(a, y) =

¸
n=1
c
n
sin h

nπa
b

. .. .
bn
sin

nπy
b

(23.8)
where
c
n
sinh

nπa
b

=
2
b
b

0
g
2
(y) sin

nπy
b

dy. (23.9)
Summarizing:
u(x, y) =

¸
n=1
c
n
sinh

nπx
b

sin

nπy
b

; c
n
=
2
b sin h

nπa
b

b

0
g
2
(y) sin

nπy
b

dy.(23.10)
23.2 Rectangular domains with mixed BC
Example 23.1 Insulating BC along two sides and specified temperatures
on the others:
∆u = u
xx
+u
yy
= 0 (23.11)
0 = u
x
(0, y) = u
x
(a, y) = u(x, 0) (23.12)
u(x, b) = f(x). (23.13)
136
23.2. RECTANGULAR DOMAINS WITH MIXED BC
Let u(x, y) = X(x)Y (y).
X

X
= −
Y

Y
= ±λ
2
. (23.14)
Since we have homogeneous BC on X

(0) = 0 = X

(a) choose −λ
2
.
1. X


2
X = 0 X

(0) = 0 = X

(a).
X(x) = Acos λx +Bsin λx
X

(0) = Bλ = 0 ⇒B = 0
X

(x) = −Aλsin(λx) +Bλcos(λx)
X

(a) = −Aλsin(λa) = 0
(23.15)
Therefore
λ
n
= (nπ/a) n = 0, 1, 2, . . . X
n
(x) = cos

nπy
a

(23.16)
are eigenfunctions and eigenvalues.
2. λ
n
= 0: Y

−λ
2
Y = 0 and Y (0) = 0 ⇒Y
n
(y) = Asinh

nπy
a

n = 0.
Thus
u
n
(x, y) = cos

nπx
a

sinh

nπy
a

(23.17)
satisfy homogeneous BC.
λ
0
= 0: In this case the ODE for Y
0
is:
Y

0
= 0 ⇒Y (y) = c
1
y +c
2
(23.18)
Y
0
(0) = c
2
= 0 ⇒Y
0
(y) = y (23.19)
and u
0
(x, y) = y 1 satisfies the homogeneous BC.
Therefore
u(x, y) = c
0
y +

¸
n=1
c
n
sinh

nπy
a

cos

nπx
a

(23.20)
u(x, b) =
(2c
0
b)
2
+

¸
n=1
c
n
sinh

nπb
a

cos

nπx
a

= f(x) (23.21)
(2c
0
b) =
2
a
a

0
f(x) dx c
n
sinh

nπb
a

=
2
a
a

0
f(x) cos

nπx
a

dx (23.22)
c
0
=
1
ab
a

0
f(x) dx c
n
=
2
a sinh(
n+b
a
)
a

0
f(x) cos

nπx
a

dx (23.23)
u(x, y) = c
0
y +

¸
n=1
c
n
sinh

nπy
a

cos

nπx
a

. (23.24)
137
Lecture 27 - More Rectangular Domains and semi-infinite strip problems
23.3 Semi-infinite strip problems
Example 23.2 A Semi-infinite strip with specified temperatures:
u
xx
+u
yy
= 0 0 < x < a, 0 < y < ∞ (23.25)
u(0, y) = 0 = u(a, y) (23.26)
u(x, 0) = f(x) u(x, y) →0 as y →∞. (23.27)
Let u(x, t) = X(x)T(t) and plug into (1a?):
X

(x)
X(x)
= −
Y

(y)
Y (y)
= −λ
2
since we have homogeneous BC on X. (23.28)
1.
X


2
X = 0
X(0) = 0 = X(a)

λ
n
= nπ/a n = 1, 2, . . .
X
n
= sin

nπx
a

(23.29)
2. Y

− λ
2
Y = 0 Y (y) = Ae
−λy
+ Be
λy
. Since u(x, y) → 0 as y → ∞
we require B = 0. Therefore
u
n
(x, y) = e
−λny
sin

nπx
a

(23.30)
satisfy the homogeneous BC and the BC at ∞. Thus
u(x, y) =

¸
n=1
c
n
e
−(

a
)y
sin

nπx
a

. (23.31)
f(x) = u(x, 0) =

¸
n=1
c
n
sin

nπx
a

⇒c
n
=
2
a
a

0
f(x) sin

nπx
a

dx.(23.32)
138
23.3. SEMI-INFINITE STRIP PROBLEMS
Example 23.3 Semi-infinite strip with inhomogeneous BC:
Look for a function v(x) for which v

= 0 and which satisfies the inho-
mogeneous BC.
v = αx +β v(0) = A = β v(a) = αa +A = B
Therefore v(x) =

B −A
a

x +A.
Now let u(x, y) = v(x) +w(x, y).
0 = u
xx
+u
yy
= v
xx
+w
xx
+v
yy
+w
yy
⇒∆w = 0 (23.33)
A = u(0, y) = v(0) +w(0, y) ⇒w(0, y) = 0 (23.34)
B = u(a, y) = v(a) +w(a, y) ⇒w(a, y) = 0 (23.35)
f(x) = u(x, 0) = v(x) +w(x, 0) ⇒w(x, 0) = f(x) −v(x). (23.36)
Thus w satisfies the same BVP as does u in Eg. 3 above.
Therefore
u(x, y) = (B −A)(x/a) +A+

¸
n=1
d
n
e
−(

a
)y
sin

nπx
a

(23.37)
where
d
n
=
2
a
a

0
¦f(x) −v(x)¦ sin

nπx
a

dx. (23.38)
139
Lecture 27 - More Rectangular Domains and semi-infinite strip problems
140
Chapter 24
Lecture 28 - Neumann
Problem - only flux BC and
Circular domains
24.1 Neumann Problem on a rectangle
Example 24.1 Neumann Problem:
u
xx
+u
yy
= 0, 0 < x < a 0 < y < b (24.1)
u
x
(0, y) = 0 u
x
(a, y) = f(y) (24.2)
u
y
(x, 0) = 0 = u
y
(x, b). (24.3)
Let u(x, y) = X(x)Y (y).
X

(x)
X(x)
= −
Y

(y)
Y (y)
= λ
2
(24.4)
Y

(y) +λ
2
Y (y) = 0
Y

(0) = 0 = Y

(b)

Y = Acos λy +Bsin λy
Y

= −Aλsin λy +Bλcos λy
(24.5)
141
Lecture 28 - Neumann Problem - only flux BC and Circular domains
Y

(0) = λB = 0 λ = 0 or B = 0. (24.6)
Y

(b) = −Aλsin λb = 0
λ
n
= (nπ/b) n = 0, 1, . . .
Y
n
= cos

nπy
b

, Y
0
= 1
(24.7)
X

n
−λ
2
X
n
= 0 (24.8)
X

n
(0) = 0 (24.9)
n = 0: X

0
= 0, X
0
= c
0
x +D
0
⇒X

0
= c
0
⇒X

0
(0) = c
0
= 0.
Choose D
0
= 1: X
0
= 1
n ≥ 1 X
n
= c
n
cosh(λ
n
x) +D
n
sinh(λ
n
x)
X

n
= c
n
λsinh(λ
n
x) +D
n
λcosh(λ
n
x)
X

n
(0) = λ
n
D
n
= 0
(24.10)
Choose c
n
= 1: X
n
= cosh(λ
n
x).
Thus
u
n
(x, y) = X
n
Y
n
= cosh(λ
n
x) cos(λ
n
y)
u
0
(x, y) = X
0
Y
0
= 1

satisfy homog. BC.(24.11)
Therefore
u(x, y) = A
0
+

¸
n=1
A
n
cosh

nπx
b

cos

nπy
b

. (24.12)
Now f(y) = u
x
(a, y).
u
x
(x, y) =

¸
n=1
A
n


b

sinh

nπx
b

cos

nπy
b

(24.13)
u
x
(a, y) =

¸
n=1

A
n


b

sinh

nπa
b
¸
cos

nπy
b

= f(y) . . . (24.14)
This is like a Fourier Cosine Series for f(y) but without the constant
term a
0
.
Recall
f(y) =
a
0
2
+

¸
n=1
a
n
cos

nπy
b

, a
n
=
2
b
b

0
f(y) cos

nπy
b

dy. (24.15)
142
24.1. NEUMANN PROBLEM ON A RECTANGLE
Thus the expansion (24.14) is consistent only if a
0
= 0. For this to be
true we require that
b

0
f(y) dy = 0 (24.16)
if
b

0
f(y) dy = 0 then there is no solution to the boundary value problem 1.
Note
1. If
b

0
f(y) dy = 0 there is a net flux into the domain through the right
hand boundary and, since the other boundaries are insulated, there
can be no steady solution – the temperature will continually change
with time.
2. If
b

0
f(y) dy = 0 there is no net flux through the boundary and a
steady state can exist. i.e. It is possible that u
xx
+ u
yy
= u
t
= 0. If
b

0
f(y) dy = 0 then
A
n


b

sinh

nπa
b

=
2
b
b

0
f(y) cos

nπy
b

dy. (24.17)
Therefore
A
n
=
2
nπ sinh

nπa
b

b

0
f(y) cos

nπy
b

dy n ≥ 1 (24.18)
and
u

(x, y) = A
0
+

¸
n=1
A
n
cosh

nπx
L

cos

nπy
b

(24.19)
where A
0
is undetermined. u(x, y) is said to be known up to an arbi-
trary constant.
143
Lecture 28 - Neumann Problem - only flux BC and Circular domains
3. If u

(x, y) is the steady state of a 2D Heat Equation u
t
= u
xx
+ u
yy
with u(x, y, 0) = u
0
(x, y) then

D
u
t
dxdy =

D
∇ ∇udxdy =

∂D
∂u
∂n
ds = 0. (24.20)
Therefore

∂t

¸

D
udxdy
¸

= 0 ⇒

D
udxdy = const for all time =

D
u
0
(x, y) dxdy.(24.21)
Now

∂D
u

(x, y) = A
0
area(D) =

∂D
u
0
(x, y) dx (24.22)
Which the condition that determines A
0
.
24.2 General Analysis of Laplace’s Equation on
Circular Domains:
Recall the Laplacian in polar coordinates:
0 = ∆u = u
xx
+u
yy
= u
rr
+
1
r
u
r
+
1
r
2
u
θθ
r = (x
2
+y
2
)
1/2
θ = tan
−1
(y/x)
. (24.23)
Let
u(r, θ) = R(r)Θ(θ) (24.24)
r
2
R

+rR

R(r)
= −
Θ

Θ(θ)
= λ
2
(24.25)
which leads to r
2
R

+rR

−λ
2
R = 0 and Θ


2
Θ = 0.
24.3 R Equation:
λ = 0: r
2
R

+rR

= 0, R = r
γ
⇒γ(γ−1)+γ = γ
2
= 0 ⇒R(r) = C+Dln r
λ = 0: r
2
R

+rR

−λ
2
= 0, R = r
γ
⇒γ(γ −1) +γ −λ
2
= γ
2
−λ
2
= 0 ⇒
R(r) = Cr
λ
+Dr
−λ
.
144
24.4. Θ EQUATION:
24.4 Θ Equation:
Θ


2
Θ = 0 Θ = Acos λθ +Bsin λθ Θ

= −Aλsin λθ +Bλcos λθ
24.5 For Different Boundary Conditions:
(I) Θ(0) = 0 = Θ(α) λ
n
= nπ/α n = 1, 2, . . . Θ
n
(θ) = sin λ
n
θ
(II) Θ

(0) = 0 = Θ

(α) λ
n
= nπ/α n = 0, 1, 2, . . . Θ
n
(θ) ∈ ¦1, cos λ
n
θ¦
(III) Θ(0) = 0 = Θ

(α) λ
n
= (2n −1)π/2α n = 1, 2, . . . Θ
n
(θ) = sin λ
n
θ
(IV) Θ

(0) = 0 = Θ(α) λ
n
= (2n −1)π/2α n = 1, 2, . . . Θ
n
(θ) = cos λ
n
θ
(V)
Θ(−π) = Θ(π)
Θ

(−π) = Θ

(π)

λ
n
= n n = 0, 1, 2, . . . Θ
n
(θ) ∈ ¦1, cos λ
n
θ, sin λ
n
θ¦.
The most general solution is thus of the form
u(r, θ) = ¦A
0

0
ln r¦ 1 +

¸
n=1

A
n
r
λn

n
r
−λn
¸
cos λ
n
θ(24.26)
+

¸
n=1

B
n
r
λn

n
r
−λn
¸
sin λ
n
θ. (24.27)
24.5.1 Notes:
• For problems that include the origin, the condition [u[ < ∞ as r →0
dictates that α
0
= 0, α
n
= 0 and β
n
= 0.
• For problems that involve infinite domains the condition [u[ < ∞ as
r →∞ dictates that A
n
= 0 and B
n
= 0.
• The values of λ
n
and the corresponding eigenfunctions depend on the
boundary conditions (I)–(V) that apply.
145
Lecture 28 - Neumann Problem - only flux BC and Circular domains
146
Chapter 25
Lecture 29 Wedge Problems
Example 25.1 Wedge with homogeneous BC on θ = 0, θ = α < 2π
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 0 < r < a, 0 < θ < α (25.1)
u(r, 0) = 0 u(r, α) = 0 (25.2)
u(r, θ) bounded as r →0 u(a, θ) = f(θ). (25.3)
Let u(r, θ) = R(r) Θ(θ).
r
2
(R

+
1
r
R

)
R
= −
Θ

(θ)
Θ(θ)
= λ
2

r
2
R

+rR

−λ
2
R = 0 Euler Eq.
Θ


2
Θ = 0
u(r, 0) = R(r)Θ(0) = 0 ⇒Θ(0) = 0; u(r, α) = R(r)Θ(α) = 0 ⇒Θ(α) = 0
Eigenvalue
Problem

Θ


2
Θ = 0
Θ(0) = 0 = Θ(α)
Θ = Acos λθ +Bsin(λθ)
Θ(0) = A = 0 Θ(α) = Bsin(λα) = 0
(25.4)
Therefore
λ
n
= (nπ/α) n = 1, 2, . . . Θ
n
= sin

nπθ
α

. (25.5)
147
Lecture 29 Wedge Problems
To solve the Euler Eq. let R(r) = r
γ
, R

= γr
γ−1
, R

= γ(γ − 1)r
γ−2
.
Therefore
γ(γ −1) +γ −λ
2
= γ
2
−λ
2
= 0 ⇒γ = ±λ. (25.6)
Therefore
R(r) = c
1
r
λ
+c
2
r
−λ
. (25.7)
Now since u(r, θ) < ∞ as r →0 we require c
2
= 0. Therefore
u(r, θ) =

¸
n=1
c
n
r
(

α
)
sin

nπθ
α

(25.8)
u(a, θ) = f(θ) =

¸
n=1

c
n
a
(

α
)
¸
sin

nπθ
α

. (25.9)
This is just a Fourier Sine Series for f(θ): Therefore
c
n
a
(

α
)
=
2
α
α

0
f(θ) sin

nπθ
α

dθ (25.10)
c
n
=
2
α
a
−(

α
)
α

0
f(θ) sin

nπθ
α

dθ. (25.11)
Therefore
u(x, θ) =

¸
n=1
c
n
r
(

α
)
sin

nπθ
α

. (25.12)
Example 25.2 A wedge with Inhomogeneous BC
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 0 < r < a, 0 < θ < α (25.13)
u(r, 0) = u
0
u(r, α) = u
1
u(r, θ) < ∞ as r →0 u(a, θ) = f(θ).(25.14)
148
Let us look for the simplest function of θ only that satisfies the inhomoge-
neous BC of the from: w(θ) = (u
1
−u
0
)
θ
α
+u
0
. Note that w
θθ
= 0 and that
w(0) = u
0
and w(α) = u
1
. Then let u(r, θ) = w(θ) +v(r, θ).
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= v
rr
+
1
r
v
r
+
1
r
2
v
θθ
= 0
v(r, 0) = 0 v(r, α) = 0
v(a, θ) = f(θ) −w(θ)

Essentially the problem
solved in Example 25.1
(25.15)
The solution is
u(r, θ) = (u
1
−u
0
)
θ
α
+u
0
+

¸
n=1
c
n
r
(

α
)
sin

nπθ
α

(25.16)
where
c
n
=
2
α
a
−(

α
)

0

f(θ) −w(θ)

sin

nπθ
α

dθ. (25.17)
Example 25.3 A wedge with insulating BC on θ = 0 and θ = α < 2π.
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0
u
θ
(r, 0) = 0 u
θ
(r, α) = 0
u(a, θ) = f(θ).
(25.18)
Let
u(r, θ) = R(r)Θ(θ) ⇒r
2

R

+
1
r
R

/R(r) = −Θ

/Θ = λ
2
(25.19)
Θ equation`
Θ


2
Θ = 0
Θ

(0) = 0 = Θ

(α)

Θ(θ) = Acos λθ +Bsin(λθ)
Θ

(0) = Bλ = 0 λ = 0 or B = 0;
(25.20)
Θ

(θ) = −Aλsin(λθ) +Bλcos(λθ)
Θ

(α) = −Aλsin(λα) = 0 λ
n
=

α
; n = 0, 1, . . .
(25.21)
149
Lecture 29 Wedge Problems
R equation` r
2
R

n
+rR

n
−λ
2
n
R
n
= 0.
n = 0: rR

0
+R

0
= (rR

0
)

= 0 ⇒rR

0
= a?
0
⇒R
0
(r) = a?
0
ln r +c
0
.
n ≥ 1: r
2
R

n
+rR

n
−λ
2
R
n
= 0 ⇒R
n
= c
n
r
λn
+D
n
r
−λn
.
Since u(r, θ) < ∞ (i.e. must be bounded) as r →0 we require d
0
= 0 =
D
n
. Therefore
u(r, θ) =
c
0
2
+

¸
n=1
c
n
r
(

α
)
cos

nπθ
α

(25.22)
f(θ) = u(a, θ) =
c
0
2
+

¸
n=1
c
n
a
(

α
)
cos

nπθ
α

(25.23)
c
0
=
2
α
α

0
f(θ)d?θ c
n
=
2
α
a
−(

α
)
α

0
f(θ) cos

nπθ
α

dθ (25.24)
u(r, θ) =
c
0
2
+

¸
n=1
c
n
r
(

α
)
cos

nπθ
α

. (25.25)
Example 25.4 Mixed BC - a ‘crack like’ problem.
∆u = u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 (25.26)
subject to
u(r, 0) = 0
∂u
∂θ
(r, π) = 0 (25.27)
u(a, θ) = f(θ). (25.28)
Let u(r, θ) = R(r)Θ(θ).
r
2

R

+
1
r
R

R
= −
Θ

(θ)
Θ(θ)
= λ
2
(25.29)
Θ equation`
Θ


2
Θ = 0
Θ(0) = 0 Θ

(π) = 0
Θ = Acos λθ +Bsin λθ Θ

= −Aλsin λθ +Bλcos λθ
Θ(0) = A = 0 Θ

(π) = Bλcos(λπ) = 0 ⇒πλ
1
=
π
2
,

2
, . . .
(25.30)
150
or λ
n
= (2n + 1)
1
2
n = 0, 1, . . . λ = 0 as this would be trivial.
R equation` r
2
R

+ rR

− λ
2
R = 0 R(r) = r
γ
⇒ γ
2
− λ
2
= 0 γ = ±λ.
Therefore
u
n
(r, θ) =

c
n
r
λn
+d
n
r
−λn

sin λ
n
θ. (25.31)
Since u should be bounded as r →0 we conclude that d
n
= 0. The general
solution is thus
u(r, θ) =

¸
n=0
c
n
r
(2n+1)/2
sin

(2n + 1)
2
θ

(25.32)
f(θ) = u(a, θ) =

¸
n=0
c
n
a
(2n+1)/2
sin

2n + 1
2

θ

. (25.33)
Check orthogonality
π

0
sin

2m+ 1
2

θ

sin

2n + 1
2

θ

dθ =

0 m = n
π/2 m = n
. (25.34)
Therefore
c
n
=
2a
−(n+
1
2
)
π
π

0
f(θ) sin

n +
1
2

θ

dθ (25.35)
u(r, θ) =

¸
n=0
c
n
r
(n+
1
2
)
sin

n +
1
2

θ

. (25.36)
151
Lecture 29 Wedge Problems
152
Chapter 26
Lecture 30 Wedges with
cut-outs, circles, holes and
annuli
Example 26.1 A circular wedge with a cut-out:
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 (26.1)
u
θ
(r, 0) = 0 u
θ
(r, α) = 0
u(b, θ) = 0 u(a, θ) = f(θ)
(26.2)
Let u(r, θ) = R(r)Θ(θ).
r
2
(R

+
1
r
R)
R(r)
= −
Θ

(θ)
Θ(θ)
= λ
2

r
2
R

+rR

−λ
2
R = 0
Θ


2
Θ = 0
(26.3)
Θ equation`
Θ


2
Θ = 0
Θ

(0) = 0 = Θ

(α)

Θ = Acos λθ +Bsin λθ
Θ

(0) = Bλ = 0 ⇒B or λ = 0,
(26.4)
Θ

= −Aλsin λθ +Bλcos λθ
Θ

(α) = −Aλsin λα = 0, λ =

α
n = 0, 1, . . .
(26.5)
153
Lecture 30 Wedges with cut-outs, circles, holes and annuli
R equation` n = 0: (rR

0
)

= 0 rR

0
= B
0
R
0
= A
0
+B
0
ln r. Note
u
0
(b, θ) = R
0
(b)Θ
0
(θ) = 0 ⇒R
0
(b) = A
0
+B
0
ln b = 0, A
0
= −B
0
ln b. (26.6)
Therefore R
0
= B
0
ln(r/b). Choose B
0
= 1.
n ≥ 1: r
2
R

n
+rR

n
−λ
2
R
n
= 0 R(r) = A
n
r
λn
+B
n
r
−λn
R
n
(b) = A
n
b
λn
+B
n
b
−λn
= 0 ⇒B
n
= −A
n
b
2λn
(26.7)
R
n
(r) = A
n
[r
λn
−b
2λn
r
−λn
] Choose A
n
= 1. (26.8)
u
n
(r, θ) =

r
(

α
)
−b
2(

α
)
r
−(

α
)

cos

nπθ
α

(26.9)
u
0
(r, θ) = ln

r
b

1 (26.10)
Therefore
u(r, θ) = c
0
ln

r
b

+

¸
n=1
c
n

r
(

α
)
−b
(
2nπ
α
)
r
−(

α
)

cos

nπθ
α

(26.11)
u(a, θ) = f(θ) = 2

c
0
ln

a
b

2
+

¸
n=1
c
n

a
(

α
)
−b
(
2nπ
α
)
r
−(

α
)

cos

nπθ
α

(26.12)
=
a
0
2
+

¸
n=1
a
n
cos

nπθ
α

. (26.13)
Therefore
2c
0
ln(a/b) =
2
α
α

0
f(θ) dθ. (26.14)
c
n
=
2
α

a
(

α
)
−b
(
2nπ
α
)
a


α

α

0
f(θ) cos

nπθ
α

dθ (26.15)
c
0
=
1
αln(a/b)
α

0
f(θ) dθ. (26.16)
Note: In the special case f(θ) = 1, c
0
=
1
log(a/b)
, and c
n
= 0 n ≥ 1.
By Fourier basis function orthogonality
α

0
1 cos

nπθ
α

dθ = 0 so that the
154
solution reduces to:
u(r, θ) =
log(r/b)
log(a/b)
(26.17)
which is purely radial i.e. has no θ dependence.
Example 26.2 Complete circle (interior circle)
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 0 < r < a, 0 < θ < 2π. (26.18)
BC: u(a, θ) = f(θ) u(r, θ) < ∞ r →0
Periodicity u(θ + 2π) = u(θ) periodic.
(26.19)
Let u(r, θ) = R(r)Θ(θ).
r
2

R

+
1
r
R

R(r)
= −
Θ

Θ
= +λ
2

r
2
R

+rR

−λ
2
R = 0 Euler Eq.
Θ


2
Θ = 0
(26.20)
Θ equation` Θ


2
Θ = 0 Θ = Acos(λθ) +Bsin(λθ)
θ(−π) = Acos(λπ) −Bsin(λπ) = Θ(π) = Acos(λπ) (26.21)
+Bsin(λπ) ⇒2Bsin(λπ) = 0 (26.22)
Θ

(θ) = −Aλsin(λθ) +Bλcos(λθ) (26.23)
Θ

(−π) = Aλsin(λπ) +Bλcos(λπ) = Θ

(π) = −Aλsin(λπ)(26.24)
+Bλcos(λπ) ⇒2Aλsin(λπ) = 0. (26.25)
Therefore λ
n
=

π
= n; n = 0, 1, 2, . . . Θ
n
(θ) = cos(nθ) and sin(nθ).
R equation` n = 0: λ
0
= 0 ⇒rR

0
+R

0
= (rR

0
)

= 0 R
0
= B
0
ln r +A
0
.
n ≥ 1: λ
n
= n:
r
2
R

n
+rR

n
−(n)
2
R
n
= 0 Euler Eq.
R
n
= r
γ
⇒γ(γ −1) +γ −n
2
= 0 γ = ±λ
n
= ±n
R
n
= c
n
r
−n
+d
n
r
n
.
(26.26)
155
Lecture 30 Wedges with cut-outs, circles, holes and annuli
Now since u(r, θ) < ∞ as r → 0 we must exclude solutions that blow up.
Thus B
0
= 0 and c
n
= 0. Therefore
u(r, θ) =
a
0
2
+

¸
n=1
¸
a
n
cos(nθ) +b
n
sin(nθ)
¸
r
n
(26.27)
u(a, θ) = f(θ) =
a
0
2
+

¸
n=1

a
n
cos(nθ) +b
n
sin(nθ)

a
n
(26.28)
a
0
=
1
π
π

−π
f(θ) dθ a
n
=
a
−n
π
π

−π
f(θ) cos(nθ) dθ (26.29)
b
n
=
a
−n
π
π

−π
f(θ) sin(nθ) dθ. (26.30)
Note:
1. For problem exterior to a circle we require that u(r, θ) < ∞as r →∞.
In this case we require that B
0
= 0 and that d
n
= 0 so that R
0
= A
0
and R
n
= r
−n

a
n
cos nθ +b
n
sin(nθ)

.
u(r, θ) =
a
0
2
+

¸
n=1

a
n
cos(nθ) +b
n
sin(nθ)

r
−n
(26.31)
a
n
=
a
n
π
π

−π
f(θ) cos(nθ) dθ, b
n
=
a
n
π
π

−π
f(θ) sin(nθ) dθ. (26.32)
Remark 26.3 Note 2: (Neumann BC)
u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0
∂u
∂r
(a, θ) = f(θ)
u 2π - periodic.
(26.33)
u(r, θ) =
a
0
2
+

¸
n=1
r
n

a
n
cos(nθ) +b
n
sin(nθ)

(26.34)
∂u
∂r

r=a
= f(θ) =

¸
n=1
nr
n−1

a
n
cos(nθ) +b
n
sin(nθ)

r=a
(26.35)
=

¸
n=1
na
n−1

a
n
cos(nθ) +b
n
sin(nθ)

. (26.36)
156
26.1. SPECIAL CASE - ELECTRICAL IMPEDANCE TOMOGRAPHY
A solution will not exist unless a
0
=
1

π

−π
f(θ) dθ = 0. Otherwise there is a
net flux of heat across the boundary and no steady state solution will exist.
26.1 Special Case - Electrical Impedance Tomog-
raphy
f(θ) = I
0
δ

θ −
π
2

−I
0
δ

θ +
π
2

.
f(−θ) = I
0
δ

θ +
π
2

−I
0
δ

θ −
π
2

(26.37)
= I
0
δ

θ +
π
2

−I
0
δ

θ −
π
2

= −f(θ). (26.38)
Thus f is odd ⇒a
0
= a
n
= 0.
na
n−1
b
n
=
2
π
π

0
I
0
δ

θ −
π
2

sin(nθ) dθ (26.39)
b
n
=
2I
0
πna
n−1
sin


2

(26.40)
u(r, θ) =
2aI
0
π

¸
n=1
sin(nθ)
n
sin


2

r
a

n
(26.41)
For enrichment ↓ =
2aI
0
π

¸
n=1

r
a

n
1
2
¸
cos n

θ −
π
2

n

cos n

θ +
π
2

n
¸
(26.42)
=
aI
0
π

¸
n=1

r
a

n
¸
cos n

θ −
π
2

n

cos n

θ +
π
2

n
¸
(26.43)
=
aI
0
π
Re
¸

¸
n=1
z
n
1
n


¸
n=1
z
n
2
n
¸
z
1
=

r
a

e
i(θ−
π
2
)
z
2
=

r
a

e
i(θ+
π
2
)
. (26.44)
157
Lecture 30 Wedges with cut-outs, circles, holes and annuli
Now for [z[ < 1:
1
1 −z
= 1 +z +z
2
+ =

¸
k=0
z
k
−ln(1 −z) = z +
z
2
2
+ =

¸
n=1
z
n
n
. (26.45)
Therefore
u(r, θ) = −
aI
0
π
Re
¸
ln

1 −z
1
1 −z
2

. (26.46)
Now if (1 −z) = Ae

then
Re

ln(1 −z)

= Re

ln

Ae

= Re[ln A+iφ] = ln A. (26.47)
Therefore
u(r, θ) = −
aI
0

ln

1 −z
1
1 −z
2

2
. (26.48)
Now
z
1
=

r
a

e
i(θ−
π
2
)
= ρe

1
(26.49)
| −z|
2
= (1 −z
1
)(1 −z
1
) =

1 −ρe

1

1 −ρe
−iφ
1

(26.50)
= 1 −ρ

e

1
+ e
−iφ
1


2
(26.51)
= 1 −2ρ cos φ
1

2
. (26.52)
Similarly z
2
=

r
a

e
i(φ+
π
2
)
= ρe

2
and [1 − z
2
[
2
= 1 − 2ρ cos φ
2
+ ρ
2
.
Therefore
u(r, θ) =
aI
0

ln
¸
1 −2(
r
a
) cos(θ +
π
2
) + (
r
a
)
2
1 −2(
r
a
) cos(θ −
π
2
) + (
r
a
)
2
¸
(26.53)
u(r, θ) =
aI
0

ln
¸
a
2
+ 2ar sin θ +r
2
a
2
−2ar sin θ +r
2

(26.54)
158
26.2. POISSON’S INTEGRAL FORMULA:
26.2 Poisson’s Integral Formula:
u(r, θ) =
a
0
2
+

¸
n=1

a
n
cos(nθ) +b
n
sin(nθ)

r
n
(26.55)
=
1

π

−π
f(φ)dφ +
1
π

¸
n=1

r
a

n
π

−π
f(φ)

cos nθ cos nφ + sin(nθ) sin(nφ)


(26.56)
=
1
π

1
2
π

−π
f(φ) dφ +

¸
n=1

r
a

n
π

−π
f(φ) cos n(θ −φ) dφ

(26.57)
=
1
π
π

−π
f(φ)

1
2
+

¸
n=1

r
a

n
cos n(θ −φ)
¸
dφ (26.58)
Now

¸
n=1

r
a

n
cos n(θ −φ) = Rρ

¸
n=1
z
n
= Rρ

z
1−z

= Rρ

z(1−¯ z)
(1−z)(1−¯ z)

=
(
r
a
) cos(θ−φ)−(
r
a
)
2
1−2(
r
a
) cos(θ−φ)+(
r
a
)
2
=
ar cos(θ−φ)−r
2
a
2
−2ar cos(θ−φ)+r
2
z =

r
a

ρ
i(θ−φ)
(1 −z)(1 − ¯ z) = 1 −(z + ¯ z) +[z[
2
= 1 −2

r
a

cos(θ −φ) + 1?

r
a

2
z(1 − ¯ z) =
r
a
ρ
i(θ−φ)

r
a

2
Therefore
u(r, θ) =
1
π
π

−π
f(φ)

1
2
+
ar cos(θ −φ) −r
2
a
2
−2ar cos(θ −φ) +r
2

dφ (26.59)
=
1
π
π

−π
f(φ)

1
2
a
2
−ar cos(θ −φ) +
1
2
r
2
+ar cos(θ −φ) −r
2
a
2
−2ar cos(θ −φ) +r
2
¸

(26.60)
u(r, θ) =
1

(a
2
−r
2
)
π

−π
f(φ)
a
2
−2ar cos(θ −φ) +r
2
dφ (26.61)
159
Lecture 30 Wedges with cut-outs, circles, holes and annuli
160
Chapter 27
Lecture 31 Sturm-Liouville
Theory
27.1 Boundary value problems and Sturm-Liouville
theory:
• Up till now we have been able to solve partial differential equations
by separating variables which typically reduces the PDE to solving an
eigenvalue problem and some initial value problem (for the Heat and
Wave Equations) or some inhomogeneous boundary value problem (in
the case of Laplace’s Equation).
• The eigenvalue problems thus far have been simple
1.
x


2
X = 0
X(0) = 0 = X(L)
→ λ
n
=

L
n = 1, 2, . . . X
n
= sin

nπx
L

.
2.
X


2
X = 0
X

(0) = 0 = X

(L)
→ λ
n
=

L
n = 0, 1, . . . X
n
= cos

nπx
L

.
3.
X


2
X = 0
X(x + 2π) = X(x)

periodic BC
→ λ
n
=

L
, n = 0, 1, . . .
X
n
=

sin

nπx
L

cos

nπx
L

X
0
= 1.
161
Lecture 31 Sturm-Liouville Theory
• In this section we abstract these problems to a general class of bound-
ary value problems which share a common set of properties. The so-
called Sturm-Liouville Problems define a class of eigenvalue prob-
lems which include many of the above problems as special cases. The
S −L Problem helps to identify those assumptions that are needed to
define an(delete the word an?) eigenvalue problems with the properties
that we require.
27.2 The regular Sturm-Liouville problem:

p(x)y

−q(x)y +λr(x)y = 0 0 < x <
α
1
y(0) +α
2
y

(0) = 0 β
1
y() +β
2
y

() = 0
(27.1)
where p, p

, q and r are continuous on 0 ≤ x ≤ and p(x) ≥ 0 and r(x) > 0
on 0 ≤ x ≤ .
We define the Sturm-Liouville eigenvalue problem as:
Ly = λry Ly = −(py

)

+qy
α
1
y(0) +α
2
y

(0) = 0 β
1
y() +β
2
y

() = 0
p(x) > 0 and r(x) > 0.

SL (27.2)
Remark 27.1 Note:
1. If p = 1 q = 0 r = 1 α
1
= 1 α
2
= 0 β
1
= 1 β
2
= 0 we obtain
Problem (1) above whereas if p = 1 q = 0 r = 1 α
1
= 0 α
2
= 1
β
1
= 0 β
2
= 1 we obtain Problem (2) above. Notice that the boundary
conditions for these two problems are specified at separate points and
are called separated BC. The periodic BC X(0) = X(2π) are not
separated so that Problem (3) is not a SL Problem.
2. If p > 0 and r > 0 and < ∞ then the SL Problem is said to be
regular. If p(x) or r(x) is zero for some x or the domain is [0, ∞) then
the problem is singular.
3. There is no loss of generality in the form of Ly = −(py

) +qy since it
is possible to convert a general 2nd order eigenvalue problem
−P(x)y

−Q(x)y

+R(x)y = λy (27.3)
to this form by multiplying by an integrating factor µ(x)
−µ(x)P(x)y

−µQ(x)y

+µ(x)R(x)y = λµ(x)y (27.4)
162
27.2. THE REGULAR STURM-LIOUVILLE PROBLEM:
but
Ly = −py

−p

y

+qy = λry. (27.5)
So we can identify that p = µP and p

= µQ ⇒ p

= µ

P + µP

= µQ
which is a linear 1st order ODE for µ with integrating factor ρ

P

P

Q
P
dx
µ

+

P

P

Q
P

µ = 0 ⇒



R
Q
P
dx
µ

= 0 µ =
ρ
R
Q
P
dx
P
. (27.6)
Example 27.2 Reducing a boundary value problem to SL form:
φ

+xφ

+λφ = 0 (27.7)
φ(0) = 0 = φ (27.8)
We bring (27.8) into SL form by multiplying by the integrating factor
µ =
1
p
ρ
R
Q
P
dx
= ρ
R
xdx
= ρ
x
2
/2
P(x) = 1 Q(x) = x R(x) = 1.
ρ
x
2
/2
φ


x
2
/2

+λρ
x
2
/2
φ = 0

ρ
x
2
/2
φ

= λρ
x
2
/2
φ
p(x) = ρ
x
2
/2
r(x) = ρ
x
2
/2
(27.9)
Example 27.3 Convert the equation −y

+x
4
y
1
= λy to SL form
P = 1 Q = −x
4
µ = ρ

R
x
4
dx
= ρ
−x
5
/5
(27.10)
Therefore −ρ
−x
5
/5
y


−x
5
/5
x
4
y

= λρ
−x
5
/5
(27.11)

ρ
−x
5
/5
y

= λρ
−x
5
/5
y. (27.12)
163
Lecture 31 Sturm-Liouville Theory
27.3 Properties of SL Problems
1. Eigenvalues:
(a) The eigenvalues λ are all real.
(b) There are an ∞# of eigenvalues λ
j
with λ
j
< λ
2
< . . . < λ
j
→∞
as j →∞.
(c) λ
j
> 0 provided
α
1
α
2
< 0,
β
1
β
2
> 0 q(x) > 0.
2. Eigenfunctions: For each λ
j
there is an eigenfunction φ
j
(x) that is
unique up to a multiplicative const. and which satisfy:
(a) φ
j
(x) are real and can be normalized so that

0
r(x)φ
2
j
(x) dx = 1.
(b) The eigenfunctions corresponding to different eigenvalues are or-
thogonal with respect to the weight function r(x):

0
r(x)φ
j
(x)φ
K
(x) dx = 0 j = k. (27.13)
(c) φ
j
(x) has exactly j −1 zeros on (0, ).
3. Expansion Property: ¦φ
j
(x)¦ are complete if f(x) is piecewise
smooth then
f(x) =

¸
n=1
c
n
φ
n
(x)
where c
n
=

R
0
r(x)f(x)φn(x) dx

R
0
r(x)φ
2
n
(x) dx
(27.14)
Example 27.4 Robin Boundary Conditions:
X

+λX = 0 α = µ
2
X

(0) = h
1
X(0) X

() = −h
2
X()
(27.15)
X(x) = Acos µx +Bsin µx (27.16)
X

(x) = −Aµsin µx +Bµcos µx (27.17)
164
27.3. PROPERTIES OF SL PROBLEMS
BC 1: X

(0) = Bµ = h
1
X(0) = A A = Bµ/h
1
.
BC 2:
X

() = −Aµsin(µ) +Bµcos(µ) = −h
2
X() = −h
2
[Acos µ +Bsin µ]
(27.18)
⇒B
¸

µ
2
h
1
sin(µ) +µcos(µ)

= −Bh
2
¸
µ
h
1
cos µ + sin µ

(27.19)
B


µ
2
h
1
+h
2

sin µ +

µ +
h
2
h
1
µ

cos µ

= 0. (27.20)
Therefore
tan(µ) =
¸
µ(h
1
+h
2
)
µ
2
−h
1
h
2

. (27.21)
h
1
and h
2
= 0: X
n
= µ
n
cos µ
n
x + sin µ
n
x
h
2
= 0 h
1
= 0:
X
n
=
µ
n
h
1
cos µ
n
x + sin µ
n
x (27.22)
=
cos µ
n
( −x)
sin µ
n

(27.23)
h
1
→∞ h
2
= 0:
X
n
= sin(µ
n
x) (27.24)
µ
n

¸
2n −1
2

π

n = 1, 2, . . . (27.25)
165
Lecture 31 Sturm-Liouville Theory
166
Chapter 28
Lecture 32 Solving the heat
equation with Robin BC
28.1 Expansion in Robin Eigenfunctions
Assume that we can expand f(x) in terms of φ
n
(x):
f(x) =

¸
n−1
c
n
φ
n
(x) (28.1)
1

0
f(x) sin(µ
n
x) dx = c
n
1

0

φ
n
(x)

2
dx (28.2)
= c
n
1
2

1 + cos
2
µ
n

(28.3)
Therefore
c
n
=
2
[1 + cos
2
µ
n
]
1

0
f(x) sin(µ
n
x) dx. (28.4)
If f(x) = x then
1

0
xsin(µ
n
x) dx = −
cos(µnx)
µn
−x

1
0
+
1
µn
1

0
cos µ
n
xdx
= −
cos(µn)
µn
+
sin µnx
µ
2
n

1
0
but −µ
n
cos µ
n
= sin µ
n
=
sin µn−µn cos µn
µ
2
n
= 2
sin µn
µ
2
n
.
(28.5)
167
Lecture 32 Solving the heat equation with Robin BC
Therefore
c
n
=
4 sin µ
n
µ
2
n
[1 + cos
2
µ
n
]
(28.6)
f(x) = 4

¸
n=1
sin µ
n
sin(µ
n
x)
µ
2
n
[1 + cos
2
µ
n
]
(28.7)
28.2 Solving the Heat Equation with Robin BC
u
t
= α
2
u
xx
0 < x < 1 (28.8)
u(0, t) = 1 u
x
(1, t) +u(1, t) = 0 (28.9)
u(x, 0) = f(x). (28.10)
Look for a steady state solution v(x)
v

(x) = 0
v(0) = 1 v

(1) +v(1) = 0

(28.11)
v = Ax +B v(0) = B = 1 v

(x) = A v

(1) +v(1) = A+ (A+ 1) = 0
A = −1/2
(28.12)
Therefore
v(x) = 1 −x/2. (28.13)
Now let u(x, t) = v(x) +w(x, t)
u
t
= w
t
= α
2
(v

+w
xx
) ⇒ w
t
= α
2
w
xx
1 = u(0, t) = v(0) +w(0, t) = 1 +w(0, t) ⇒ w(0, t) = 0
0 = u
x
(1, t) +u(1, t) = ¦v

(1)+v(1)¦ +w
x
(1, t) +w(1, t) ⇒ w
x
(1, t) +w(1, t) = 0
f(x) = u(x, 0) = v(x) +w(x, 0) ⇒ w(x, 0) = f(x) −v(x).
168
28.2. SOLVING THE HEAT EQUATION WITH ROBIN BC
Let
w(x, t) = X(x)T(t) (28.14)
˙
T(t)
α
2
T(t)
=
X

X
= −µ
2
(28.15)
T(t) = cρ
−α
2
µ
2
t
(28.16)
X


2
X = 0
X(0) = 0 X

(1) +X(1) = 0

The µ
n
are solutions of the transcendental
equation: tan µ
n
= −µ
n
.
(28.17)
X
n
(x) = sin(µ
n
x) (28.18)
w(x, t) =

¸
n=1
c
n
ρ
−α
2
µ
2
n
t
sin(µ
n
x) (28.19)
where
f(x) −v(x) = w(x, 0) =

¸
n=1
c
n
sin(µ
n
x) (28.20)
⇒c
n
=
2
[1 + cos
2
µ
n
]
1

0
[f(x) −v(x)] sin(µ
n
x) dx (28.21)
u(x, t) = 1 −
n?
2
+

¸
n=1
c
n
ρ
−α
2
µ
2
n
t
sin(µ
n
x). (28.22)
169
Lecture 32 Solving the heat equation with Robin BC
170
Chapter 29
Lecture 33 Variable
coefficient BVP -
eigenfunctions involving
solutions to the Euler
Equation:
Example 29.1 Eigenfunctions involving solutions to an Euler Equation:
(x
2
φ

)

+λφ = 0 1 < x < 2
φ(1) = 0 φ(2) = 0
x
2
φ

+ 2xφ

+λφ = 0 An Euler Eq.
(29.1)
Let
φ(x) = x
r
r(r −1) + 2r +λ = r
2
+r +λ = 0. (29.2)
Therefore
r =
−1 ±

1 −4λ
2
= r
1
, r
2
. (29.3)
29.1 Cases:
λ =
1
4
:
φ(x) = c
1
x

1
2
+c
2
x

1
2
log x (29.4)
φ(1) = c
1
= 0 φ(2) = c
2
2

1
2
log 2 = 0 ⇒c
2
= 0 (29.5)
171
Lecture 33 Variable coefficient BVP - eigenfunctions involving solutions to
the Euler Equation:
so there is no Eigenfunction for λ = 1/4.
λ =
1
4
:
φ(x) = c
1
x
r
1
+c
2
x
r
2
(29.6)
φ(1) = c
1
+c
2
= 0 c
2
= −c
1
(29.7)
φ(2) = c
1

2
r
1
−2
r
2

= 0 (29.8)
2
r
1
−r
2
= 1
ρ
(r
1
−r
2
) ln 2
= 1 = ρ
2πin
(r
1
−r
2
) ln 2 = 2πin
r
1
−r
2
=

1 −4λ = 2πni/ ln(2)
(29.9)
Thus to obtain nontrivial solutions we require 1−4λ < 0 λ >
1
4
. For λ >
1
4

1 −4λ = i

4λ −1 = 2πni/ ln(2). (29.10)
The Eigenvalues are:
λ
n
=
1
4
+
π
2
n
2
[ln(2)]
2
, 4λ
n
−1 =

2
n
2
[ln(2)]
2
= (2β
n
)
2
β
n
= (nπ/ ln 2). (29.11)
The corresponding roots r
1
and r
2
are as follows
(r
1
)
n
= −
1
2
+iβ
n
and (r
2
)
n
= −
1
2
−iβ
n
(29.12)
φ
n
(x) = c
n
x

1
2

x
iβn
−x
−iβn

(29.13)
= c
n
x

1
2

ρ
iβn ln x
−ρ
−iβn ln x

(29.14)
= d
n
x

1
2
sin (β
n
ln x) (29.15)
= d
n
x

1
2
sin
¸

ln x
ln(2)

(29.16)
choose d
n
= 1 or normalize so that
2

1
φ
2
n
(x) dx = 1.
Example 29.2 Solving a variable coefficient Heat Conduction Problem:
u
t
= D(x
2
u
x
)
x
−u 1 < x < 2 t > 0
u(1, t) = 0 = u(2; t) u(x, 0) = f(x).
(29.17)
172
29.2. SOLVING THE HEAT EQUATION BY EXPANDING IN
EIGENFUNCTIONS INVOLVING SOLUTIONS TO AN EULER
EQUATION:
Let
u(x, t) = X(x)T(t) (29.18)
˙
T(t)
DT(t)
=
(x
2
X

)

X
−1 (29.19)
˙
T(t)
DT(t)
+ 1 =
(x
2
X

)

X
= −λ (29.20)
˙
T +D(1 +λ)T = 0 T(t) = cρ
−D(1+λ)t
(x
2
X

)

+λX = 0
X(1) = 0 = X(2)

λ
n
=
1
4
+
(πn)
2
[ln(2)]
2
; X
n
(x) = x

1
2
sin


ln x
ln 2

n = 1, 2, . . .
(29.21)
u(x, t) = x

1
2

¸
n=1
c
n
ρ
−D(1+λn)t
sin


ln x
ln 2

(29.22)
f(x) = u(x, 0) = x

1
2

¸
n=1
c
n
sin


ln x
ln 2

(29.23)
29.2 Solving the heat equation by expanding in
eigenfunctions involving solutions to an Euler
Equation:
∆u = u
rr
+
1
r
u
r
+
1
r
2
u
θθ
= 0 1 < r < 2, 0 < θ < α (29.24)
u(r, 0) = 0, u(r, α) = f(r) (29.25)
u(1, θ) = 0, u(2, θ) = 0 (29.26)
u(r, θ) = R(r)Θ(θ) (29.27)
r
2
R

+rR

R(r)
= −
Θ

Θ
= −λ
2
(because of Homog. BC) (29.28)
Θ:
Θ

−λ
2
Θ = 0 Θ = c cosh λθ +Dsinh λθ
Θ(0) = 0 Θ(0) = c = 0 ⇒Θ(θ) = Dsin hλθ
173
Lecture 33 Variable coefficient BVP - eigenfunctions involving solutions to
the Euler Equation:
R: r
2
R

+ rR

+ λ
2
R = 0() R(1) = 0 = R(2). Although we can easily
see that dividing through by r we can reduce () to S-L form, let us use the
integrating factor
µ() =
1
P
ρ
R
Q
P
dr
=
1
r
2
ρ
R
r
r
2
dr
=
1
r
2
ρ
ln r
=
1
r
. (29.29)
Therefore

1
r
() ⇒rR

R

= −(rR

)

=
λ
2
r
R. (29.30)
Now let us look for Eigenvalues and Eigenfunctions to (). Let
R(r) = r
γ
⇒γ(γ −1) +γ +λ
2
= γ
2

2
= 0 γ = ±iλ. (29.31)
R(r) = c
1
r

+c
2
r
−iλ
r

= ρ
iλln r
(29.32)
= Acos(λln r) +Bsin(λln r) (29.33)
R(1) = Acos [λ(ln 1)] +Bsin(λln 1) = A = 0 (29.34)
R(2) = Bsin [λln 2] = 0 ⇒λ
n
ln 2 = nπ n = 1, 2, . . . (29.35)
and the corresponding Eigenfunctions are R
n
= sin


ln r
ln 2

. Therefore
u(r, θ) =

¸
n=1
B
n
sinh(λ
n
θ) sin(λ
n
ln r). (29.36)
Now match BC:
f(r) = u(r, α) =

¸
n=1
B
n
sinh(λ
n
α) sin


ln r
ln 2

x = ln r
dx
0
? =
1
r
dr
.(29.37)
Now
2

1
1
r
sin

mπ ln r
ln 2

sin

nπ ln r
ln 2

dr =
ln 2

0
sin

mπx
ln 2

sin

nπx
ln 2

dx =

0 m = n
ln 2
2
m = n
. (29.38)
Therefore
B
n
=
2
ln 2 sinh

nπα
ln 2

2

1
f(r)
r
sin

m?n?π ln r
r

dr. (29.39)
174
Chapter 30
Lecture 34 — Sturm
Liouville Theory
30.1 Properties of SL Problems:
1. Eigenvalue Properties:
• (a) the eigenvalues λ are all real.
• (b) there are an infinite number of eigenvalues λ
j
with λ
1
<
λ
2
< . . . λ
j
→∞ as j →∞.
• (c) λ
j
> 0 provided
α
1
α
2
< 0,
β
1
β
2
> 0 and q(x) > 0.
2. Eigenfunction Properties: Corresponding to each eigenvalue λ
j
there
is an eigenfunction φ
j
(x) that is unique up to multiplication by a
constant, and which satisfy:
• (a) φ
j
(x) are real and can be normalized

0
r(x)φ
2
j
(x) dx = 1.
• (b) the eigenfunctions corresponding to different eigenvalues are
orthogonal with respect to the weight function r(x):

0
r(x)φ
j
(x)φ
k
(x) dx = 0 if j = k. (30.1)
• (c) φ
j
(x) has exactly j −1 zeros on 0 < x < .
175
Lecture 34 — Sturm Liouville Theory
3. Expansion Property: The eigenfunctions φ
j
(x) form a complete set
so that any piecewise smooth function f(x) can be expanded as a
generalized Fourier Series:
f(x) =

¸
n=1
c
n
φ
n
(x) (30.2)
by orthogonality:
c
n
=

0
r(x)f(x)φ
n
(x) dx

0
r(x)φ
2
n
(x) dx
. (30.3)
30.2 Lagrange’s Identity:
Lagrange’s Identity

0
(vLu −uLv) dx = −p(x)u

v

0
+ p(x)uv

0
.
is fundamental to the development of S-L Theory.
Proof: Let u and v be any sufficiently differentiable functions; then

0
vLudx =

0
v

−(pu

)

+qu
¸
dx (30.4)
= −vpu

0
+

0
u

pv

dx +

0
uqv dx (30.5)
= −vpu

0
+ upv

0
+

0
u

−(pv

)

+qv
¸
dx(30.6)
Therefore

0
vLudx = −pvu

0
+ puv

0
+

0
uLv dx. (30.7)
Now suppose that u and v both satisfy the SL boundary conditions. I.E.
α
1
u(0) +α
2
u

(0) = 0
α
1
v(0) +α
2
v

(0) = 0
β
1
u() +β
2
u

() = 0
β
1
v() +β
2
v

() = 0
(30.8)
176
30.3. PROOFS TO SELECTED PROPERTIES:
then

0
vLudx −

0
uLv dx = −p()u

()v() +p()u()v

() (30.9)
+p(0)u

(0)v(0) −p(0)u(0)v

(0) (30.10)
= p()

+
β
1
β
2
u()v() +u()


β
1
β
2
v()

(30.11)
+p(0)


α
1
α
2
u(0)v(0) −u(0)


α
1
α
2
v(0)

(30.12)
= 0. (30.13)
Thus

0
vLudx =

0
uL?v dx whenever u and v satisfy the SL boundary
condition.
Note:
• If L and BC are such that

0
vLudx =

0
uLv dx then L is said to be
self-adjoint. Notation if we define (f, g) =

0
f(x)g(x) dx then we
may write (v, Lv) = (u, Lv).
30.3 Proofs to selected properties:
(1a) λ is real: Let Ly = λry (1) α
1
y(0) +α
2
y

(0) = 0 β
1
y() +β
2
y

() = 0.
Take the conjugate of (1) L¯ y =
¯
λr¯ y. By Lagrange’s Identity:
0 = (¯ y, Ly) −(y, L¯ y) (30.14)
= (¯ y, rλy) −(y, r
¯
λ¯ y) (30.15)
=

0
¯ y(x)rλy(x) dx −

0
y(x)r(x)
¯
λ¯ y(x) dx (30.16)
= (λ −
¯
λ)

0
r(x)

y(x)

2
dx (30.17)
177
Lecture 34 — Sturm Liouville Theory
Since r(x)[y(x)[
2
≥ 0 it follows that λ =
¯
λ ⇒λ is real.
(1c) λ
j
> 0 provided α
1

2
< 0 β
1

2
> 0 and q(x) > 0. Consider
Ly = −(py

)

+ qy = λry (SL) and multiply (SL) by y and integrate from 0
to :
(y, Ly) =

0
−(py

)

y +qy
2
dx = λ

0
r(x)

y(x)

2
dx (30.18)
Therefore λ =

0
−(py

)

y +qy
2
dx

0
ry
2
dx
this is known as Rayleigh’s Quotient.
=
[−py

y]

0
+

0
p(y

)
2
+qy
2
dx

0
ry
2
dx
(30.19)
=
+p()
β
1
β
2

y()

2
−p(0)
α
1
α
2

y(0)

2
+

0
p(y

)
2
+qy
2
dx

0
ry
2
dx
. (30.20)
Therefore λ > 0 since the RHS is all positive.
Note: If q(x) ≡ 0 and α
1
= 0 = β
1
then with y

(0) = 0 = y

() we have
nontrivial eigenfunction y(x) = 1 and eigenvalue λ = 0.
(2b) Eigenfunctions corresponding to different eigenvalues are or-
thogonal. Consider two distinct eigenvalues λ
j
= λ
k
λ
j
: Lφ
j
= rλ
j
φ
j
and
λ
k
: Lφ
k
= rλ
k
φ
k
. Then
0 = (φ
K
, Lφ
j
) −(φ
j
, Lφ
K
) by Lagrange’s Identity (30.21)
= (φ
K
, rλ
j
φ
j
) −(φ
j
, rλ
K
φ
K
) (30.22)
= (λ
j
−λ
K
)

0
r(x)φ
K
(x)φ
j
(x) dx (30.23)
now λ
j
= λ
k?K
implies that

0
r(x)φ
k
(x)φ
j
(x) dx = 0. (30.24)
178
30.3. PROOFS TO SELECTED PROPERTIES:
(3) Expansion Property It is difficult to prove the convergence of the
eigenfunction series expansion for f(x) that is piecewise smooth. However,
if we assume the expansion converges then it is a simple matter to use
orthogonality to determine the coefficients in the expansion: Let f(x) =

¸
n=1
c
n
φ
n
(x).

0
f(x)φ
m
(x)r(x) dx =

¸
n=1
c
n

0
r(x)φ
m
(x)φ
n
(x) dx (30.25)
orthogonality implies
c
m
=

0
r(x)f(x)φ
m
(x) dx

0
r(x)

φ
m
(x)

2
dx
. (30.26)
179

CONTENTS 7 Bessel’s Equation 7.1 Bessel’s Function of Order ν ∈ {. . . , −2, −1, 0, 1, 2 . . .}: / 7.2 Bessel’s Function of Order ν = 0 - repeated roots: . . . 7.3 Bessel’s Function of Order ν = 1 : . . . . . . . . . . . . 2 7.4 Example - the roots differ by an integer . . . . . . . . 43 43 44 46 48

. . . .

. . . .

. . . .

. . . .

8 Separation of Variables 49 8.1 Types of Boundary Value Problems: . . . . . . . . . . . . . . 49 8.2 Separation of Variables - Fourier sine Series: . . . . . . . . . . 51 8.3 Heat Eq on a Circular Ring - Full Fourier Series . . . . . . . 57 9 Lecture 13 - Fourier Series 61 9.1 It can be useful to shift the interval of integration from [−L, L] to [c, c + 2L] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 9.2 Complex Form of Fourier Series . . . . . . . . . . . . . . . . . 65 10 Lecture 14 - Even and Odd Functions 67 10.1 Integrals of Even and Odd Functions . . . . . . . . . . . . . . 67 10.2 Consequences of Even/Odd Property for Fourier Series . . . . 68 10.3 Half-Range Expansions . . . . . . . . . . . . . . . . . . . . . . 70 11 Lecture 15 - Convergence of Fourier Series 73 11.1 Convergence of Fourier Series . . . . . . . . . . . . . . . . . . 76 11.2 Illustration of the Gibbs Phenomenon . . . . . . . . . . . . . 77 11.3 Now consider the sum of the first N terms . . . . . . . . . . . 78 12 Lecture 16 - Parseval’s Identity 81 12.1 Geometric Interpretation of Parseval’s Formula . . . . . . . . 82 13 Lecture 17 - Solving the heat equation using finite difference methods 85 13.1 Approximating the Derivatives of a Function by Finite Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 13.2 Heat Equation solution by Finite Differences . . . . . . . . . 87 14 Lecture 18 - Solving Laplace’s Equation using finite differences 91 14.1 Finite Difference approximation . . . . . . . . . . . . . . . . . 91 14.2 Solving the System of Equations by Jacobi Iteration . . . . . 93 2

CONTENTS 15 Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC 95 16 Lecture 20 - Inhomogeneous Derivative BC 101

17 Lecture 21 Distributed, Time Dependent Heat Sources eigenfunction expansions 105 18 Lecture 22 More Eigenfunction Expansions - Time Dependent Boundary Conditions 111 19 Lecture 23 - 1D Wave Equation 117 19.1 Guitar String . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 20 Lecture 24 - Space-Time Interpretation of lution 20.1 Characteristics . . . . . . . . . . . . . . . 20.2 Region of Influence . . . . . . . . . . . . . 20.3 Domain of Dependence . . . . . . . . . . . D’Alembert’s So121 . . . . . . . . . . . 121 . . . . . . . . . . . 122 . . . . . . . . . . . 122

21 Lecture 25 Solution by separation of variables 125 21.1 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 21.2 Now we can use the trigonometric identities . . . . . . . . . . 127 22 Lecture 26 - Laplace’s Equation 22.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . 22.2 Laplace’s Equation . . . . . . . . . . . . . . . . . . . 22.3 Rectangular Domains . . . . . . . . . . . . . . . . . 22.4 Solution to Problem (1A) by Separation of Variables 23 Lecture 27 - More Rectangular Domains strip problems 23.1 Solution to Problem (1B) by Separation of 23.2 Rectangular domains with mixed BC . . . 23.3 Semi-infinite strip problems . . . . . . . . 129 130 130 130 131

. . . .

. . . .

. . . .

. . . .

. . . .

and semi-infinite 135 Variables . . . . . 135 . . . . . . . . . . . 136 . . . . . . . . . . . 138

24 Lecture 28 - Neumann Problem - only flux BC and Circular domains 141 24.1 Neumann Problem on a rectangle . . . . . . . . . . . . . . . . 141 24.2 General Analysis of Laplace’s Equation on Circular Domains: 144 24.3 R Equation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 3

1 Notes: .CONTENTS 24. . . . . . . . . . . . . . . . . .Electrical Impedance Tomography . . . . . . . . . . . . . 162 27. . . . . . . .2 Poisson’s Integral Formula: . . . . . . . . circles. 173 30 Lecture 34 — Sturm Liouville Theory 175 30. . . . .5 For Different Boundary Conditions: . 167 28. . . . . . . . . . . . . .4 Θ Equation: . . . . . . 145 25 Lecture 29 Wedge Problems 147 26 Lecture 30 Wedges with cut-outs. . .1 Cases: . . . . . . . . . . . . . . . . . . . . .1 Special Case .2 Solving the heat equation by expanding in eigenfunctions involving solutions to an Euler Equation: . . . . . . 161 27. . . . . . . . . . . . . . . . . 177 4 . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . .eigenfunctions involving solutions to the Euler Equation: 171 29.1 Boundary value problems and Sturm-Liouville theory: . . . . . . . . . . . . . .3 Proofs to selected properties: . . .1 Expansion in Robin Eigenfunctions . . . . . 176 30. . . . . . 145 24. . . . . . . . . .2 Solving the Heat Equation with Robin BC . . .2 Lagrange’s Identity: . . . . . .2 The regular Sturm-Liouville problem: . . . 164 28 Lecture 32 Solving the heat equation with Robin BC 167 28. . 159 27 Lecture 31 Sturm-Liouville Theory 161 27. . . . . . . . . . . . . . . . . . 171 29. . 157 26. . . . . . . . 168 29 Lecture 33 Variable coefficient BVP . . . . . . . . . . . 175 30. . . . . . . . . . . . holes and annuli 153 26. . . . . . .1 Properties of SL Problems: . . 145 24. . . . . . . . . . .3 Properties of SL Problems . . . . .

Chapter 1 Lecture 1 .2) 5 . t). EG 1: a ∂ ∂ u(x. y) that are determined by prescribing a relationship between the function value and its partial derivatives. y) = c ∂x ∂y First Order PDE (1. EG: y (x) + ey(x) = 0.Equations which define functions of a single independent variable by prescribing a relationship between the values of the function and its derivatives.Introduction to Partial Differential Equations ODE .Involve multivariable functions u(x.1) PDE . (1. u(x. y) + b u(x.

t) Heat Equation or Diffusion Eq X2 + Y 2 = k Elliptic ∂2 ∂x2 u(x. t) be the flux of cars at x at time t. y) + ∂2 ∂y 2 u(x.Introduction to Partial Differential Equations EG 2: Some Classic Second Order PDEs: Quadric Classification Eq. Name T = X2 Parabolic ∂ ∂t u(x. Consider the traffic flow on a highway and let u(x. [q] = # of cars/unit time. t) = ∂2 ∂x2 u(x. Laplace or the Wave Eq. y) = f (x. t) = 0 2 The Wave Eq By analogy with quadric surfaces aX 2 + 2bXY + c2 Y 2 + · · · = k that can be reduced to a standard form by coordinate rotation.1 Modeling and Derivation of PDE: 1D Conservation Law: Traffic flow on a highway.Lecture 1 .3) can be reduced by a transformation of coordinates to one of the Heat. y) Poisson Eq f ≡ 0 Laplace Eq f = 0 T 2 − c2 X 2 = k Hyperbolic ∂2 ∂t2 u(x. t) be the density of cars at x at time t. Let q(x.4) 6 . 1. the most general linear 2nd order PDE auxx + 2buxy + cuyy + · · · (1.5) (1. [u] = # of cars/unit length. (1. t) ∂ − c2 ∂x2 u(x.

x = x − ct transformation of coordinates Guess: u(x.1. t)}∆x Let ∆x → 0 and ∆t → 0: {q(x. t) − q(x + ∆x. t + ∆t) − u(x.and the first order Wave Equation: Assume that q varies linearly with u.6) ∂u ∂q + =0 ∂t ∂x • conservation of cars • conservation of heat • conservation of chemicals.e.. t)}∆t (1. What happens if q = −cu in which case ∂u ∂u −c =0 ∂t ∂x (1.9) (1. To see this consider the following moving coordinate system.10) Thus ut + cux = 0 has solutions of the form u(x.7) How does q change with u? Convection . t) = f (x − ct) which represents a right moving wave. ut + cux = 0 (1. (1. t) = f (x − ct) solves ut = −cf Therefore ux = f ut + cux = −cf + cf = 0.11) 7 .1. MODELING AND DERIVATION OF PDE: {u(x.8) But this is just a wave equation. q = cu from which it follows that ∂u ∂u +c =0 ∂t ∂x (1. i.

11) has a solution of the form u(x. t) = ∂2u ∂2u − c2 2 = 0 ∂t2 ∂x (1.12) is the 2nd order wave equation that has both left and right moving wave solutions.Lecture 1 . t) = f (x + ct) which represents a left moving wave.Introduction to Partial Differential Equations Exercise: Show that (1.13) In this case the conservation law reduces to the form: ∂u ∂2u = α2 2 ∂t ∂x 2D Heat Equation: ∂u = α2 ∂t ∂2u ∂2u + 2 ∂x2 ∂y (1. Fourier’s Law: Heat flows from hotter regions to colder ones? q = −α2 ∂u ∂x (1.14) 8 .15) The Heat Equation (1. Note: ∂ ∂ +c ∂t ∂x ∂ ∂ −c ∂t ∂x u(x.

2. t)A = ρA∆x σ(x + ∆x. t) the displacement of the rod from equilibrium. σ(x + ∆x.17) (1.1. t) be the pressure in the rod at x at time t and u(x. THE WAVE EQUATION: 1. t) − σ(x. t) ∆x ∂σ ∆x → 0 ∂x Hooke’s Law ∂2u ∂t2 ∂2u = ρ 2 ∂t = ρ ∂2u ∂t2 (1. σ=E ∂2u = ∂t2 E ρ ∂2u ∂2u = c2 2 . and let σ(x.2 The Wave Equation: Consider an elastic rod having a density ρ and cross-sectional area A. t)A − σ(x. Balance of Linear Momentum F = M a. where ∂x2 ∂x c= E ρ (1.18) 9 .16) (BLM) Balance of Linear Momentum ∂u ∂x Plug into (BLM) to obtain the 2nd order wave equation.

t) + p∆x2 ∂x (1. t)] = u(x. t) + (1 − 2p)u(x. Find an equation for the density of flies at t + ∆t. t) p ⇒ = α2 2 The Heat Eq.3 The Drunkard’s Walk .Lecture 1 . t) + p∆x − ∆x ∆x ∂u ∂u (x. t) − ∂x (x − ∆x. t) + pu(x − ∆x.19) ∆x ∂2u u(x.20) = s1 + s2 + · · · + sN ∼ 0 Expected Value = (s1 + · · · + sN ) = 2 s2 1 + ··· + 2 s2 N + 2(s1 s2 + · · · + sN −1 sN ) ∼ N ∆x Therefore tN ∆x2 = k 2 tN ∆t √ |xN | ∼ k tN x2 N 10 . t) − u(x.The Heat Equation: Let u(x. t) − u(x − ∆x. u(x. t) [u(x + ∆x. t) u(x. ∆t ∆t ∂x2 ∂t ∂x What is the Mean Absolute Deviation of the Drunkard? sj tj xN x2 N = ±∆x = j∆t (1. t)] [u(x.Introduction to Partial Differential Equations 1. t + ∆t) − u(x. t) be the density of fruit-flies at point x at time t. t) + p∆x2 2 ∂x ∆x2 ∂ 2 u ∂u ∂2u u(x. t + ∆t) = pu(x + ∆x.

THE DRUNKARD’S WALK .3.1.1: Simulation with N = 1000 trajectories for 200 steps along with the mean absolute deviation envelopes shown in red 11 .THE HEAT EQUATION: 5 4 3 2 1 x 0 −1 −2 −3 −4 Drunkard Walk 0 2 4 6 8 10 t 12 14 16 18 20 Figure 1.

Lecture 1 .Introduction to Partial Differential Equations 12 .

1 . n . 13 . Divergence Test: ∞ 0 lim an = 0 ⇒ an diverges. .1) + ··· + 1 2 n−1 ∞ = n=1 1 2 n−1 ∞ bn n=1 (2. . . . (2. . . ..Preliminaries 2.Chapter 2 Lecture 2 . 1 .3) EG: an = 1 1 + 1 + · · · + 1 + · · · → ∞. 2 . .. 1 . .2) Note: In order to sum to a finite number the terms of the sequence must tend to 0 as n → ∞. 1 .1 Sequences and Series of Numbers: A Sequence of Numbers: Notation {an } an = {bn } bn = 1 1. . . 2 3 n−1 1 1. 4 2 1 n 1 n−1 2 A series of numbers: 1 1 1 1 + + + ··· + + ··· = 2 3 n Does this infinite sum yield a finite result? 1+ 1 2 + 1 2 2 ∞ n=0 1 n ∞ an n=0 (2..

5) converge? We consider the integral: ∞ 1 1 xp T dx = lim T →∞ 1 dx xp = x1−p 1−p ln x 1 p−1 T 1 T 1 p=1 p=1 = 14 ∞ p>1 p≤1 .4) Now 1 ∞ dx x dx x 1 ∞ = T →∞ 1 ∞ lim dx = lim (ln T − ln 1) = ∞ T →∞ x But < n=1 1 n Therefore n=1 1 n = ∞. Example: For what values of p will the series ∞ n=1 1 np (2.Lecture 2 .Preliminaries ∞ 1 n Integral Test: Does n=1 converge? Consider ∞ dx x 1 ∞ < 1+ 1 1 1 + + ··· + + ··· = 2 3 n T ∞ n=1 1 n (2.

1. np p>1 p ≤ 1: ∞ dx < xp 1 ∞ n=1 1 ⇒ np ∞ n=1 1 = ∞. np p≤1 p-Series: ∞ n=1 1 np <∞ p>1 diverges p ≤ 1 Geometric Series . SEQUENCES AND SERIES OF NUMBERS: p > 1: ∞ n=2 ∞ 1 < np 1 dx ⇒ xp ∞ n=1 1 < ∞.6) For what values of r does the G-Series converge? 15 .‘the G-Series’: ∞ rn = 1 + r + r2 + · · · + rn + · · · n=0 (2.2.

7) 1−r 1 − rN +1 1−r If |r| < 1 then N N →∞ lim rn = lim n=0 1 − rN +1 1 = N →∞ 1−r 1−r (2. G-Series: ∞ rn = n=0 1 1−r ∞ |r| < 1 |r| ≥ 1 1 r= . series diverges.9) EG: ∞ n=0 1 1 = = 2.2 Absolute and Conditional Convergence: Alternating Series Test: (−1)n an an > 0 (a) an+1 ≤ an (b) lim an = 0 n→∞ ∞ ⇒ (−1)n an converges: EG: n=1 (−1) n−1 converges.10) 2. n ∞ Consider a series n=0 (−1)n an . (−1)n an is said to be absolutely convergent. 2 (2. n 2 1 − 1/2 (2.Preliminaries Partial Sum: N SN = n=0 rn = 1 + r + · · · + rN = = = (1 + r + · · · + rN ) (1 − r) (1 − r) 1 + r + · · · + rN − r − r2 − · · · − rN − rN +1 (2.Lecture 2 .8) If |r| ≥ 1. If 16 |an | < ∞ then .

bn+1 (n + 1)4 e−(n+1) = = bn n4 e−n2 2 1+ 1 n 4 e−2n−1 → 0 < 1 converges absolutely. ∞ 2n EG 1: (−1)n−1 2 . n bn and let lim n=1 bn+1 1 2n+1 /(n + 1)2 =2 1+ = n /n2 bn 2 n ∞ 2 2 → 2 series diverges. Test is inconclusive if L = 1.12) 2. POWER SERIES: If |an | = ∞ but ∞ (−1)n an < ∞ is conditionally convergent. diverges if L > 1. Idea: Extend the polynomial to include ∞ # of terms. Ratio Test: Consider bn+1 = L. Then bn converges abson→∞ bn n=0 lutely if L < 1. (2.3.3 Power Series: f (x) = a0 + a1 x + a2 x2 + · · · + an xn polynomial approximation. More General Power Series: ∞ f (x) = n=0 an (x − x0 )n = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · ·(2.14) 17 .2. f (x) = a0 + a1 x + a2 x2 + · · · + an xn + · · · Power Series ∞ = n=0 an xn (2.11) EG 2: n=1 n4 e−n .13) EG: ex = 1 + x 1! + x2 2! + x3 3! + ··· + xn n! ∞ + ··· = n=0 xn n! . (2.

17) cos x = n=0 (−1)n n=0 eiθ = 1 + iθ + (iθ)2 (iθ)3 + + ··· 2! 3! θ2 θ4 θ3 = 1− + − ··· + i θ − + ··· 2! 4! 3! = cos θ + i sin θ (2.2(x − x0 ) + · · · + n(n − 1)(n − 2)an (x − x0 )n−3 + · · · ⇒ f (3) (x0 ) = 3!a3 f (n) (x0 ) n! (2.Lecture 2 .15) f (n) (x0 ) = n!an ⇒ an = ∞ Therefore f (x) = n=0 f (n) (x0 ) (x − x0 )n n! Alternative Form of Taylor Series: ∞ f (x0 + h) = n=0 f (n) (x0 ) n h n! (2.16) EG 1: ∞ ex = n=0 ∞ xn n! (−1)n x2n+1 (2n + 1)! x2n (2n)! ∞ sin x = n=0 ∞ sinh x = cosh x = n=0 ∞ x2n+1 (2n + 1)! x2n (2n)! (2.18) 18 .Preliminaries Taylor Series .matching all the derivatives at a point: ∞ f (x) = n=0 an (x − x0 )n = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · f (x) = a1 + 2a2 (x − x0 ) + 3a3 (x − x0 )2 + · · · + nan (x − x0 )n + · · · ⇒ f (x0 ) = a1 f (x) = 2a2 + 3.2a3 (x − x0 ) + · · · + n(n − 1)an (x − x0 )n−2 + · · · ⇒ f (x0 ) = 2a2 f (3) (x) = 3!a3 + 4.3.

Review of Methods to Solve ODE 3.4) 19 .1) dy dx (3. .The Integrating Factor: y (x) + P (x)y = Q(x) (3.Chapter 3 Lecture 3 .1 First Order ODE: dy dx dy Q(y) EG: 4y x(y − 3) 4 y−3 dy = dx y x y − 3 ln |y| = 4 ln |x| + C = y = ln(x y ) + C Ax y 4 3 4 3 Separable Equations: = P (x)Q(y) = P (x) dx + C (3.2) = ey Linear First Order Eq.3) Can we find a function F (x) to multiply (4.3) by in order to turn the left hand side into a derivative of a product: Fy + FPy = FQ (3.

6) P (x) dx + C R P (x) dx = Ae = e R choose A = 1 integrating factor P (x) dx +e R R (3.8) 3.9) + n=0 y + 2y = n=1 an nx n−1 2an xn = 0 n=1⇒m=0 .Review of Methods to Solve ODE (F y) = F y + F y = F Q So let F = F P which is a separable Eq.Lecture 3 .8) let us look for a series solution: y(x) = n=0 ∞ an xn .2 Another Method .5) = P (x) dx ⇒ = dF = F P (x) dx + C (3. dF F (x) Therefore ln F or F F Therefore e R P (x) dx y R P (x) dx P (x)y = e P (x) dx Q(x) R (e P (x) dx y) = e P (x) dx Q(x) R Rx e− P (x) dx e P (t) dt Q(x) dx + C (3.Series Solution: ∞ Since the unknown solution y(x) is defined implicitly by (3.7) y(x) = EG: 1 y + 2y = 0 F (x) = e2x ⇒ e2x y + e2x 2y = (e2x y) = 0 e2x y =?c y(x) = Ce−2x (3. y Therefore In the first sum let m=n−1 n=m+1 20 = n=1 ∞ an nxn−1 ∞ (3.

am = (−1)m 2 a0 m! ∞ Therefore y(x) = a0 m=0 (−2x)m m! = a0 e−2x EG 2: Solve dy + cot(x)y = 5ecos x .11) 3 . . dx P (x) = cot x Q(x) = 5ecos x R F (x) = e cot x dx = eln(sin x) = sin x (3.12) Therefore sin(x)y + cos(x)y = (sin(x)y) = 5ecos x sin x sin(x)y = −5ecos x + C −C y(x) = − 5e sin x cos x (3.13) −4 = y(π/2) = − 5−C ⇒ C = 1 1 Therefore y(x) = 1−5ecos x sin x 3. .3. a3 = − 2 · 1 3 2 2 m · 2 a0 = (−1)3 2 a0 .3. 1 3! (3.3 Second Order Constant Coefficient Linear Equations: Ly = ay + by + cy = 0 Guess y = erx y = rerx y = r2 erx Ly = (ar2 + br + c)erx = 0 Indicial Eq. .14) 21 . y(π/2) = −4.10) n⇔m: {am+1 (m + 1) + 2am } xm = 0 2 am+1 = − (m+1) am 2 a1 = −2a0 . a2 = + 2 2 a0 . SECOND ORDER CONSTANT COEFFICIENT LINEAR EQUATIONS: ∞ Therefore m=0 am+1 (m + 1)xm + ∞ m=0 ∞ n=0 2an xn = 0 (3.: ar2 + br + c = 0 a(r − r1 )(r − r2 ) = 0 r = − b± √ b2 −4ac 2a (3.

Thus y(x) = c1 er1 x + c2 xer1 x is the general solution.18) [A cos µx + B sin µx] . r1 = r2 . Case II: ∆ = 0.Lecture 3 . x) = erx Ly(r.19) + c2 e 2x . y(x) = c1 er1 x + c2 er2 x is the general solution. Another Method: Consider a small perturbation to the double root case: r − (r1 + ) r − (r1 − ) = 0 y(x) = c1 e(r1 + )x + c2 e(r1 − (r + )x −e(r − )x = e 1 2 1 = er1 x e x −e− x (3. Let y(r.Review of Methods to Solve ODE Case I: ∆ = b2 − 4ac > 0. repeated roots Ly = a(r − r1 )2 erx = 0. x) r=r 1 ∂y (r. x) = a(r − r1 )2 erx L Therefore ∂y ∂r (r. (3.17) Case III: Complex Conjugate Roots: ∆ = b2 − 4ac < 0 b 1/2 ± i 4ac − b2 = λ ± iµ 2a y(x) = c1 e(λ+iµ)x + c2 e(λ−iµ)x r± = − = e EG 1: Ly = y + y − 6y = 0 y = erx (r2 + r − 6) = (r + 3)(r − 2) = 0 y(x) = c1 e 22 −3x λx (3. r1 = r2 . x) ∂r r=r 1 (3. Then we have one solution y(x) = er1 x – what about the second solution.16) )x c1 = xer1 x = 2 −→ →0 1 2 = −c2 ∂ rx ∂r e r=r1 (3.15) = [2a(r − r1 )erx + 2a(r − r1 )xerx ]r=r1 = 0 = xer1 x is also a solution.

3.20) + c2 xe −3x (3.26) Case 1: ∆ = (α − 1)2 − 4β > 0 Two Distinct Real Roots r1 . −3x (3.4 Euler/Equidimensional Equations: Ly = x2 y + αxy + βy = 0.25) as above. Back to (3. y = c1 xr1 + c2 xr2 if r1 or r2 < 0 |y| → ∞ as x → 0.21) 3.22) Aside: Note if we let t = ln x or x = et then d2 d =x dt2 dx d dx = x2 d d dt d d = ⇒ =x . Therefore {r(r − 1) + αr + β} xr = 0 f (r) = r2 + (α − 1)r + β = 0 1−α± (α − 1)2 − 4β 2 (3. dx dt dx dt dx (3. (3. r± = (3.23) x d2 d d2 d2 d +x ⇒ x2 2 = 2 − dx2 dx dx dt dt Therefore y − y + αy + βy = 0 ¨ ˙ ˙ y + (α − 1)y + βy = 0 ¨ ˙ (3.24) y = ert ⇒ r2 + (α − 1)r + β = 0 Characteristic Eq.22): Guess y = xr . y = rxr−1 .27) 23 .4. r2 . and y = r(r − 1)xr−2 . EULER/EQUIDIMENSIONAL EQUATIONS: EG 2: Ly = y + 6y + 9y = 0 y = erx (r + 3)2 = 0 y(x) = c1 e EG 3: Ly = y − 4y + 13y = 0 y = erx : r2√ 4r + 13 = 0 − 4± 16−52 r= = 2 ± 3i 2 Therefore y(x) = e2n [A cos 3x + B sin 3x] . (3.

28) since f (r) = (r − r1 )2 . .2 p. y = c1 xr1 ∂ r ∂r L[x ] ∂ ∂r = L ∂ r ∂r x = L[xr log x] (3. [4β − (α − 1)2 ] (1 − α) ±i = λ ± iµ 2 2 y(x) = c1 x(λ+iµ) + c2 x(λ−iµ) xr = er ln x r± = = c1 e(λ+iµ) ln x + c2 e(λ−iµ) ln x = xλ c1 eiµ ln x + c2 e−iµ ln x = A1 xλ cos(µ ln x) + A2 xλ sin(µ ln x) Notes: (1) If x < 0 replace by |x|. Check: L(xr1 log x) = x2 (xr log x) + αx(xr log x) + β(xr log x) − = x2 r(r − 1)xr log x + rxr−2 + (r − 1)xr−2 + αx rx = 2 r−1 (3. (2) w(y1 .29) log x + x r r−1 + β(x log x) r r + (α − 1)r + β x log x + {2r − 1 + α} xr = 0 Case 3: ∆ = (α − 1)2 − 4β < 0. y2 ) = = y 1 y2 y 1 y2 = y1 y 2 − y 1 y 2 (see Section 3.Lecture 3 . {f (r)xr } = f (r)xr + f (r)xr log x = 0 General Solution: y(x) = (c1 + c2 log x)xr1 .143) (3.30) 1/2 xλ cos(µ ln x) log xxλ sin(µ ln x) + xλ−1 cos(µ ln x)µ xλ sin(µ ln x) − xλ log x cos(µ ln x) − xλ−1 sin(µ ln x)µ = µx2λ−1 24 independent for x = 0.Review of Methods to Solve ODE Case 2: ∆ = 0 Double Root (r − r1 )2 = 0.

33) (3.34) 25 . 2 3 x=1 (3.35) = 2 + c2 = 0 Therefore y(x) = x − 2x log x.3.32) √ = c1 2 3 = 1 (3.31) + c2 x1− 3 √ 3 y(1) = c1 + c2 = 0 c2 = −c1 y(x) = c1 x1+ y (x) = c1 Therefore EG 2: y= xr x2 y − 3xy + 4y = 0 y(1) = 1 y (1) = 0 =⇒ r(r − 1) − 3r + 4 = r2 − 4r + 4 = 0 (r − 2)2 = 0 y(x) = c1 x2 + c2 x2 log x y(1) = c1 = 1 y (x) = 2x + c2 [2x log x + x]α=1 (3.4. 2 2 − x1− 3 √ √ √ √ 1 + 3 x 3 − 1 − 3 x− 3 √ √ y(x) = √ √ 1 √ x1+ 3 − x1− 3 . EULER/EQUIDIMENSIONAL EQUATIONS: EG 1: x2 y − xy − 2y = 0 y(1) = 0 y(1) = 1 ˙ y = xr r(r − 1) − r − 2 = 0 r2√ 2r − 2 = 0 − (r − 1)2 = 3 r = 1 ± 3 y = c1 x1+ √ 3 (3.

Review of Methods to Solve ODE 26 .Lecture 3 .

Chapter 4

Lectures 4,5 Ordinary Points and Singular Points
Lecture 4 Consider P (x)y + Q(x)y + R(x)y = 0 Homogeneous Eq. Divide through by P (x): Ly = y + p(x)y + q(x)y = 0 p(x) = Q/P, R/P (4.2) (4.1)

4.1

An Ordinary Point:

x0 is said to be an ordinary point of (5.2) if p(x) = Q/P and q(x) = R/P are analytic at x0 .

i.e.

p(x) = p0 + p1 (x − x0 ) + · · · q(x) = q0 + q1 (x − x0 ) + · · ·

=
k=0 ∞

pk (x − x0 )k qk (x − x0 )k

=
k=0

Note: (1) If P , Q and R are polynomials then a point x0 such that P (x0 ) = 0 is an ordinary point. 27

Lectures 4,5 Ordinary Points and Singular Points (2) If x0 = 0 is an ordinary point then we assume
∞ ∞ ∞

y =
n=0 ∞

cn xn , yn =
n=1

cn nxn−1 , yn =

cn n(n − 1)xn−2
n=2 ∞

0 = Ly =
n=2

cn n(n − 1)xn−2 +
n=0 ∞ ∞

pn xn
n=1

ncn xn−1

(4.3)

+
n=0 ∞

qn xn
n=0

cn xn

(m + 2)(m + 1)cm+2 + p0 (m + 1)cm+1 + · · · + pm c1
m=0

+ (q0 cm + · · · + qm c0 )} xm = 0 yields a non-degenerate recursion for the cm .

(4.4)

At an ordinary point x0 we can obtain two linearly independent solutions by power series expansion. About x0 : y(x) =
n=0 ∞

cn (x − x0 )n .

(4.5)

(3) The radius of convergence of (4.5) is at least as large as the radius of convergence of each of the series p(x) = Q/P q(x) = R/P . i.e. up to the closest singularity to x0 .

4.2

A Singular Point:

If p(x) or q(x) are not analytic at x0 , then x0 is said to be a singular point of (4.2). For example if P , Q and R are polynomials and P (x0 ) = 0 and Q(x0 ) = 0 or R(x0 ) = 0 then x0 is a singular point. EG: (x − 1)y + y = 0 (4.6) x = 0 is an ordinary point. x = 1 is a singular point. Expand around the ordinary point
∞ ∞ ∞

y(x) =
n=0

cn x ,

n

y =
n=1

ncn x

n−1

,

y =
n=2

cn n(n − 1)xn−2

(4.7)

28

4.2. A SINGULAR POINT:
∞ ∞

(x − 1)
n=2 ∞

cn n(n − 1)xn−2 +
n=1 ∞

ncn xn−1 = 0 (4.8)


n=2

cn n(n − 1)xn−2 +
n=2 ∞

cn {n(n − 1) + n} xn−1 + c1 = 0 n=2⇒m=1 n=m+1

m−1=n−2⇒m=n−1 −c2 · 2 · 1 + c1 +
m=2

−cm+1 (m + 1)m + cm m2 xm−1 = 0

c0 Arbitrary: cm+1 = c3 = m cm m+1 2 c1 c2 = 3 3

m ≥ 2 c2 =

c1 2 (4.9)

Therefore y(x) = c0 + c1
n=1

3 c1 c1 c4 = c3 = . . . cn = 4 4 n n x . n

Recall 1 = 1 + x + x2 + · · · 1−x y(x) = A + B ln |x − 1| 1 x2 x3 dx = − ln |1 − x| = x + + + ··· 1−x 2 3 (4.10)

But (4.6) is also an Euler Equation: y = (x − 1)r ⇒ r(r − 1) + r = r2 = 0 r = 0, 0. y(x) = A + B ln(x − 1) (4.11)

29

y = (x − 1)y + y 30 . . .8.5 6.3.15) 4.6. c6 = = .Lectures 4.12) n = m + 3 n = 2 ⇒ m = −1 cm (m+3)(m+2) cm+3 (m + 3)(m + 2) − cm xm+1 = 0 m = 0.. (7.6.6)(4.5. 4. .4.2 6.3 The Airy equation: Consider the Airy equation y = xy.8.5 Ordinary Points and Singular Points Lecture 5 4.6.3) (4.2 6. (4.9)(7.3 c10 = c1 (10. .64. x = 0 is an ordinary point.2 c0 c3n = (4.3.5. 9. c4 = c3n+1 = c1 4. c2 = 0 cm+3 = (1) c0 → c3 → c6 .2 9.3 y(x) = c0 y0 (x) + c1 y1 (x) Radius of Convergence: cm+3 3 |x|3 |x| = lnm→∞ =0<1 m→∞ cm (m + 3)(m + 2) lim ρ = ∞.3 (3n + 1)(3n) . . ∞ y= n=0 cn xn . .3 7. c9 = 2. c0 c3 c0 c0 .17) See B&D for expansion of Airy Solution about x0 = 1 y(x) = It is useful to write x = (x − 1) + 1.5 (3n)(3n − 1) . .16) an (x − 1)n . 1)xn−2 = ∞ y = ∞ n=0 n=2 cn n(n − 1)xn−2 cn n(n − m+1=n−2 c2 2x0 + ∞ m=0 cn xn+1 (4. 3. . .5. .3 c7 = c1 7. ∞ y = ∞ n=2 n=1 cn nxn−1 .6)(4..3. 3.3) x4 x7 x3n+1 y1 (x) = x + + + ··· + (4. . 1. (4.13) (3n)(3n − 1)(3n − 3)(3n − 4) .14) c1 (3n + 1)(3n)(3n − 2)(3n − 3) .2 c3 = (2) c1 → c4 → c7 →.2 x6 x3n x3 y0 (x) = 1 + + + ··· + + .

Since x = 0 is an ordinary point let y(x) = an xn then n=0 ∞ ∞ ∞ Ly = n=2 an n(n − 1)xn−2 − 2 n=1 an nxn + λ n=0 an xn = 0. a5 = (2 − λ)a1 . (4. . .5. a6 = a0 2 4.24) . (4. a4 = a2 = a0 .18) m=n−2→n=m+2 n=2⇒m=0 Therefore ∞ m←n m←n am+2 (m + 2)(m + 1) − 2am m + λam xm + [a2 2 + λa0 ] x0 = 0.3 4.23) .21) The general solution is of the form y(x) = Ay0 (x) + By1 (x) Note: 31 (4. a7 = a1 .4 The Hermite Equation: ∞ Ly = y − 2xy + λy = 0. .2 6.19) m=1 x0 : a2 = −λa0 /2 xm : am+2 = a0 : λ (4 − λ) (4 − λ)(−λ) (8 − λ)(4 − λ)(−λ) a2 = − a0 .2 5! 7! (2 − λ) 3 (6 − λ)(2 − λ) 5 (10 − λ)(6 − λ)(2 − λ)x7 = a1 x + x + x + + ··· 3! 5! 7! (2m − λ)am (m + 1)(m + 2) m≥1 (4.4.20) (4.3.(4.22) (2k)! λ (λ − 4)λ 4 (8 − λ)(4 − λ)(−λ) 6 y0 = a0 1 − x2 + x + x + ··· 2 4! 6! a1 : a3 = y1 (2 − λ) (6 − λ) (10 − λ)(6 − λ)(2 − λ) a1 .4. (−1))?a0 a2k = (4.4. THE HERMITE EQUATION: 4.2 [4(k − 1) − λ][4(k − 2) − λ] . 3. .3.

.1(−2) 2 n! (4.27) 32 .26) (b) For example in the special case λ = 4 = 2n then n = 2. . In this case: y0 (x) = a0 1 − nx2 + n(n − 2)22 x4 23 x6 − n(n − 2)(n − 4) + ··· 4! 6! xn . y0 (x) = a0 [1 − 2x2 ]. 2n/2 n! On the other hand if n is an odd integer then the series solution y1 (x) will terminate and become a polynomial of degree n. (4. 2. . Thus if n is an even integer then the series solution y0 will terminate and become a polynomial of degree n. (4. In this case y1 (x) = a1 x − 2(n − 1) x3 x5 + 22 (n − 1)(n − 3) 3! 5! 7 x − (n − 1)(n − 3)(n − 5)23 + · · · 7! (n−1) xn + (n − 1)(n − 3) .25) + (−1)n/2 n(n − 2) . 3.5 Ordinary Points and Singular Points (a) If λ = 2n then the recursion yields am+2 = 0 = am+4 = · · · for m = n. .Lectures 4.

In this case ρ = 2.Singular points 5.3) 33 . x = 1 and x = solution of this equation is y(x) = 1 2 are singular points. EG.1 Radius of Convergence and Nearest Singular Points EG. x = 0 is an ordinary point. y (1) = 0 then a power series expansion of √ the form cn (x − 1)n is required.2) = 1 ρ = 1. x−1 (5. (1) If we were given y(0) = 0 and y (0) = 1 then we would want a power series expansion of the form ∞ y= n=0 cn xn about x0 = 0. 2: (x − 1)(2x − 1)y + 2xy − 2y = 0.1) Roots of 1 + x2 = 0 are x = ±i. lim an+2 an (2) If we were given y(1) = 1. 1: (1 + x2 )y + 2xy + 4x2 y = 0. so we expect the radius of convergence 1 of the TS for 1+x2 to be 1 since 1 1+x2 = 1 − x2 + x4 − 1 (5. One 1 = −(1 + x + x2 + · · · ) ρ = 1. (5.Chapter 5 Lecture 6 .

4) m ≥ 2. d d Let t = x − 1 so that dt = dx and the equation is transformed to (t2 − 1)˙ + 5ty + 3y = 0 ˙y ˙ ∞ ∞ ∞ y= ∞ n=2 ∞ m=2 n=0 cn tn .6) y(x) = c0 y0 (x) + c1 y1 (x) y(1) = c0 = 7 y (1) = c1 = 3. y = ∞ n=2 cn n(n − 1)tn−2 ∞ n=0 n(n − 1)cn tn − n(n − 1)cn tn−2 + 5 ncn tn + 3 cn tn = 0 n=1 m=n−2 n=m+2 n=2=m=0 [−cm+2 (m + 2)(m + 1) + {m(m − 1) + 5m + 3} cm ] tm −2c2 + 3c0 + [−c3 3..(2n) (x c6 = 7 c4 = 6 − 1)2n 753 6 4 2 c0 (5..... 34 (5. x = 1 is an ordinary point.2n+2 35..2 + 5c1 + 3c1 ] t = 0 t0 > t1 > c2 = c3 = 3 2 c0 8 6 c1 4 = 3 c1 (5..Lecture 6 . 2 EG: (x2 − 2x)y + 5(x − 1)y + 3y = 0 y(1) = 7 y (1) = 3..Singular points This TS solution about the ordinary point x = 0 converges beyond the singular point x = 1 .7) ..2n+1 c1 n→∞ lim cm+2 m+3 = =1 ρ=1 cm m+1 (5.2n+1 (x = c2n+1 = − 1)2n+1 46.. x = 0 is a singular point (x − 1)2 − 1 y + 5(x − 1)y + 3y = 0.. 357..(2n+1) 246.2n+2 35. y = ∞ n=2 n=1 cn ntn−1 .5) y1 (x) = 64 5 3 c1 46. tm > cm+2 = c0 : c4 = 5c2 4 ∞ n=0 cm (m+1)(m+3) (m+1)(m+2) = y0 (x) = c1 : c5 = 6 5 c3 ∞ n=0 5 3 4 2 c0 .

1 y = c1 x2 + c2 x−1 Again the x2 y admits wild behaviour. y + α β y + 2 y = 0.2. SINGULAR POINTS: 5. Examples: 1.2 Singular Points: P (x)y + Q(x)y + R(x)y = 0. x2 y − 2xy + 2y = 0 y = xr → r(r − 1) − 2r + 2 = 0 r = 1. There is a special class of singular points called regular singular points in which the singularities are no worse than those in the equidimensional equations. 2. x x (5. x2 y + xy = 0 y = xr → r(r − 1) + r = 0 → y = c1 + c2 ln x The x2 y admits wild behaviour.3 Regular Singular Points: Notice that all these cases are equidimensional equations for which we can identify solutions of the form xr or xr log x. Q and R are polynomials without common factors then singular points are points x0 at which P (x0 ) = 0.10) 35 x→x0 x→x0 . (5. Note: At singular points the solution is not necessarily analytic. x2 y − 2y = 0 y = xr → r(r − 1) − 2 = 0 r = −2. 5.8) Consider If P . Q and R are polynomials and suppose P (x0 ) = 0 then α0 is a regular singular point if lim (x − x0 ) Q(x) P (x) and lim (x − x0 )2 R(x) P (x) are finite. 2 y = c1 x + c2 x2 In this case both solutions are analytic. 3. (5.5.9) If P .

11) = 1 (x−x0 )2 Examples: 1. (similarly for x = −1).12) x→1 x = 1 is a R. 1/2 5.Singular points Q(x) P (x) R(x) P (x) p0 + p1 + p2 (x − x0 ) + · · · (x − x0 ) → singularity no worse than q0 q1 + + q2 + · · · 2 (x − x0 ) (x − x0 ) → singularity no worse than I.P. 2(x − 2)2 xy + 3xy + (x − 2)y = 0. 3. x = 0 is a regular singular point. Singular points at x = 0.13) Thus x = 0 is an irregular singular point. x3 y − y = 0 P (x) = x3 Q = 0 R = −1 lim x2 −1 = ∞ x3 x→0 (5. and R are not limited to polynomials then consider P (x)y + Q(x)y + R(x)y = 0 or x2 y + x 36 xQ P (5. Q. x = 2 is an irregular singular point. = 1 x−x0 (5.14) y + x2 R P y = 0 . Note: Any singular point that is not a regular singular point is called an irregular singular point. 2.Lecture 6 .4 More General Definition of a Regular Singular Point: If P . 2. Actually y ∼ x3/4 e±2/x as x → 0+ which is much wilder than the simple power law xr or xr log x.E.S. (1 − x2 )y − 2xy + 4y = 0 P = 1 − x2 P (±1) = 0 Q = −2x R = 4 (−2x) 4 lim (x − 1) (1−x)(1+x) = 1 lim (x − 1)2 (1+x)(1−x) = 0 x→1 (5.

x = 0 is a regular singular point. x = 0 is a singular point.16) Then as x → 0 x2 y + xp0 y + q0 y ≈ 0 which is an Euler Equation which has solutions of the form y = xr . Thus about a regular singular point we look for solutions of the form ∞ ∞ y=x r n=0 an x = n=0 n an xn+r . xQ x→0 P x2 R lim x→0 P lim = lim x P = x2 Q = 2(ex − 1) R = 2ex 2(ex − 1) 2(ex − 1) 0 0 = lim = lim = 2 L’Hopital x→0 1 x→0 k→0 x2 x e−x cos x (5. x→0 x2 Since the quotient functions p = xQ/P and q = x2 R/P have Taylor Expansions about x = 0.5. xQ = p(x) = p0 + p1 x + · · · P In this case small as 2 x→0 xQ P and x2 R P are analytic at and x2 R = q(x) = q0 + q1 x + · · · . I.17) = lim x2 = 1 < ∞. Our task is to determine.E.15) P Ly = x y + xp0 y + q0 y + x p1 xy + q1 y + · · · = 0. MORE GENERAL DEFINITION OF A REGULAR SINGULAR POINT: x = 0 is a regular singular point if x = 0. EG. 37 . 1: x2 y + 2(ex − 1)y + e−x cos xy = 0 e−x cos x. (5. (i) r (ii) the coefficients an (iii) the radius of convergence. (5.4.

Lecture 6 .Singular points 38 .

Chapter 6 Lecture 7 . ∞ y = n=0 ∞ an xn+r ∞ (6. + r)(n + r − 1) 39 .Frobenius Series about Regular Singular Points Example 1: Ly = 2x2 y − xy + (1 + x)y = 0 x = 0 is a RSP.1) Ly = 2x 2 n=0 an (n + r)(n + r − 1)x ∞ n+r−2 −x n=0 an (n + r)xn+r−1 + (1 + x) n=0 ∞ an xn+r = 0 an {2(n + r)(n + r − 1) − (n + r) + 1} xn+r n=0 ∞ + n=0 an xn+r+1 = 0 n=0→m=1 ∞ (6.2) m=n+1 n=m−1 Therefore a0 {2r(r − 1) − r + 1} xr + − (n + r) + 1} + an−1 ] x [an {2(n n=1 n+r = 0.

Lecture 7 - Frobenius Series about Regular Singular Points xr > Indicial Equation: 2r2 − 3r + 1 = (2r − 1)(r − 1) = 0 a0 arbitrary Recursion −an−1 an = (2n + 2r − 3)(n + r) + 1 Let r = 1/2: an = n=1: a1 = a3 = an = −an−1 −an−1 −an−1 = = (2n − 2)(n + 1/2) + 1 (n − 1)(2n + 1) + 1 n(2n − 1) −a0 −a1 +a0 ; n = 2 : a2 = = 1 2.3 2.3 −a2 −a0 −a3 +a0 (6.4) = ; a4 = = 3.5 1.(2.3)(3.5) 4.7 1(2.3)(3.5)(4.7) (−1)n 2(n−1) a0 (−1)n a0 = n!1.3.5.(2n − 1) n(2n − 1)!

r = 1, 2

r = 1.

(6.3)

y1 (x) = x1/2
n=0

(−1)n 2(n−1) n x n(2n − 1)!

r = 1: an = a1 = an = −an−1 −an−1 = (2n − 1)(n + 1) + 1 (2n + 1)n −a0 −a1 +a0 −a2 −a0 , a2 = = ; a3 = = 3.1 5.2 (1.3)(2.5) 3.7 (1.3)(2.5)(3.7) (−1)n a0 (−1)n 2n a0 (6.5) = n!3.5.7.(2n + 1) (2n + 1)!

y2 (x) = x
n=0

(−1)n 2n n x (2n + 1)!

General Solution: y(x) = c1 y1 (x) + c2 y2 (x) Radius of Convergence ∞.

40

6.1. SERIES EXPANSION SUMMARY:

6.1

Series Expansion Summary:
P (x)y + Q(x)y + R(x)y = 0 (6.6)

Consider Divide by P (x): y + p(x)y + q(x)y = 0, p(x) = Q(x) R(x) , q(x) = P (x) P (x) (6.7)

Ordinary Points: x0 is an ordinary point if p(x) and q(x) are analytic at x0 . I.E. p(x) = p0 + p1 (x − x0 ) + · · · q(x) = q0 + q1 (x − x0 ) + · · · . (6.8) About an ordinary point x0 we can obtain 2 linearly independent solutions of the form

y(x) =
n=0

an (x − x0 )n

(6.9)

whose radius of convergence is at least as large as those of p and q in (7.8) - up to the singularity closest to x0 . Singular Points: If P (x0 ) = 0 then p(x) and q(x) may fail to be analytic in which case x0 is a singular point. Regular Singular Points: A point x0 is a regular singular point if Q(x) P (x) R(x) (x − x0 )2 P (x) (x − x0 ) are analytic at x0 . In this case (x − x0 )2 y + (x − x0 ) (x − x0 ) Q(x) P (x) y + (x − x0 )2 R(x) y=0 P (x) (6.11) = p0 + p1 (x − x0 ) + · · · = q0 + q1 (x − x0 ) + · · · (6.10)

has singularities no worse than the Euler Equation: (x − x0 )2 y + (x − x0 )p0 y + q0 y = 0. In this case we look for solutions of the form

(6.12)

y(x) = (x − x0 )

r n=0

an (x − x0 )n .

(6.13)

41

Lecture 7 - Frobenius Series about Regular Singular Points

42

2) n=m−2 n=0⇒m=2 ∞ 0= m=2 am (m + r)2 − ν 2 + am−2 xm+r + a0 r2 − ν 2 xr (7.1) x = 0 is a regular Singular Point: therefore let y = n=0 ∞ an xn+r . ∞ 0= n=0 an (n + r)(n + r − 1) + (n + r) − ν 2 xn+r + n=0 an xn+r+2 m=n+2 (7.Chapter 7 Bessel’s Equation Lecture 8 7. 2 . am−2 xm+r > am = − m≥2 (m + r)2 − ν 2 (7.}: / Ly = x2 y + xy + (x2 − ν 2 ) y = 0 ∞ (7. . . −2. . 1.4) 43 . Roots 1 r+1 > x a1 (1 ± ν)2 − ν 2 = a1 (1 ± 2ν) = 0 provided ν = 2 . −1. .1 Bessel’s Function of Order ν ∈ {. . 0.3) + a1 (1 + r)2 − ν 2 xr+1 xr > a0 = 0 ⇒ r = ±ν Indicial Eq.

∞ y = n=0 ∞ an xn+r an (n + r)(n + r − 1) + (n + r) xn+r + an xn+r+2 = 0 n=0 Ly = m=n+2 ∞ n=m−2 (7. 44 . .2 Bessel’s Function of Order ν = 0 .repeated roots: Ly = x2 y + xy + x2 y = 0. . . . .24 (2 + ν)(1 + ν) ma (−1) 0 . (m + ν) m!2 ∞ y1 (x) = xν m=0 (−1)m (x/2)2m m!(1 + ν)(2 + ν) . . .2 = 0. .7) 0 = n=2 an (n + r)2 + an−2 xn+r + a0 r(r − 1) + r xr + a1 (r + 1)r + r + 1 xr+1 = 0 a double root. .6) m!22m (1 − ν) . a2m = (7. (m − ν) ∞ y2 (x) = x−ν m=0 (−1)m (x/2)2m m!(1 − ν) .Bessel’s Equation r = ν: am = − a2 = − a0 2(2 + 2ν) am−2 am−2 am−2 =− 2 =− 2 − ν2 (m + ν) m + 2mν m(m + 2ν) a0 a2 (−1)2 a0 = − 2 a4 = − = 2 (1 + ν) 4(4 + 2ν) 2. The indicial equation is: a0 r2 = 0 r1. . 0 r1 = 0 ⇒ a1 . a4 = − = 4 2 (1 − ν) 4(4 − 2ν) 22 (1 − ν)(2 − ν) (−1)m a0 . a2m = (7. . .5) 2m (1 + ν) . (m + ν) →0 x→0 r = −ν: am = − a2 = − a0 2(2 − 2ν) am−2 m(m − 2ν) a0 a2 (−1)2 a0 = − 2 .1 = 0 ⇒ a1 = 0. (m − ν) →∞ x→0 7.

8) a2m = y1 (x) = (−1)m a0 22m (m!)2 ∞ (7. . r) = a0 xr 1 − + ∂y (x.1: Zeroth order bessel functions j0 (x) and Y0 (x) To get a second solution y(x. = a0 log xy1 (x) + a0 x r=r1 .9) = J0 (x) 1+ m=1 (−1)m x2m 22m (m!)2 1 J0(x) and Y0(x) 0 −1 −2 −3 −4 0 5 10 x 15 J0 Y0 20 Figure 7. a4 a0 = − 2 2 2. 2 6 2 4 6 a0 2 42 62 82 2 a2 = − a4 = − a2 a0 = 2 2.REPEATED ROOTS: Recursion: an = − a0 . 22 an−2 (n + r)2 n ≥ 2.10) (−1)m x2m ∂ ∂r 1 (2 + r)2 . BESSEL’S FUNCTION OF ORDER ν = 0 . 2 4 2 4 a6 = − a8 = (7. (2m + r)2 (7. .2.7. . (2m + r)2 45 . . r) ∂r ··· ∞ r m=1 x2 x4 (−1)m x2m + + ··· + (2 + r)2 (2 + r)2 (4 + r)2 (2 + r)2 (4 + r)2 .

.11) 2 2 2 a2m (0) = − − ··· − a2m (0) 2+r 4+r (2m + r) r=0 1 1 = −1 − − . Therefore 2 m ∞ y2 (x) = J0 (x) ln x + m=1 (−1)m+1 Hm 2m x 22m (m!)2 x > 0.14) y(x) = c1 J0 (x) + c2 Y0 (x).16) Ly = a0 1 r − 4 + a1 1 (r + 1) − 4 2 ∞ + n=2 an (n + r)2 − + an−2 xn+r = 0. .Bessel’s Equation Let a2m (r) = { } ⇒ ln a2m (r) = −2 ln(2 + r) − . 2 m Let Hm = 1 + 1 1 + · · · + .15) ∞ Ly = n=0 an 2 1 (n + r) − 4 2 ∞ x n+r + n=0 an x n+r+2 m = n+2 =0 n = m−2 n = 0⇒m=2 1 4 (7. Lecture 9 7. .12) It is conventional to define Y0 (x) = where γ = n→∞ 2 y2 (x) + (γ − log 2)J0 (x) π (7. − 2 ln(2m + r) (7. .5772 Euler’s Constant (7. 46 . (7. Let Consider the case ν = 1/2 Ly = x2 y + xy + x2 − ∞ y= n=0 an xn+r (7.3 Bessel’s Function of Order ν = 1 : 2 1 4 y = 0. − a2m (0) = −Hm a2m (0).13) lim (Hn − log n) = 0.

21) cos x + a1 x −1 2 sin x included in y1 (x)...2 5. 47 . BESSEL’S FUNCTION OF ORDER ν = 2 : 1 = 0. 4 an−2 Recurrence: an = − (n + r)2 − r1 = +1/2: Indicial Equation: r2 − an−2 (n + 1 )2 − 2 r=± 1 4 1 2 n ≥ 2.1 a2n = (−1)n a0 (2n)! (7.4.2 (2n + 1)! ∞ ∞ 1 1 1 (−1)n x2n (−1)n x2n+1 = x− 2 = x− 2 sin x y1 (x) = x 2 (2n + 1)! (2n + 1)! 1 4 =− n=0 n=0 1 r2 = − : 2 an = − n = 1 ⇒ a1 an−2 an−2 . 4. .1 a4 = (−1)2 a0 .18) 1 1 = a1 . − (n + 1)n 4 4 a0 (−1)2 a0 (−1)n a0 a2 = − a4 = .17) 3.2 a5 = (−1)2 a1 5. .3. n ≥ 2. a2n = (7.3. − +1 − 2 4 a0 : a2 = − a1 : a3 = − a1 3.4. =− n(n − 1) (n − 1 )2 − 1 2 4 2 (7.19) ∞ y2 (x) = a0 x = a0 x 1 −2 n=0 −1 2 1 (−1)n x2n + a1 x− 2 (2n)! ∞ n=0 (−1)n x2n+1 (2n + 1)! (7. Roots differ by an integer.3. an = − an−2 9 1 a1 = 0 ⇒ a1 = 0 n ≥ 2.2 a2n+1 = (−1)n a1 (2n + 1)! (7.3.0 = 0 a1 and a0 arbitrary.1 7.20) a0 2.2.

y= n=0 ∞ cn xn+α cn (n + α)(n + α − 1)xn+α−1 − cn xn+α = 0 n=0 ↑ p−1=n ∞ (7. α) = c0 xα α + ¯ ∂y ¯ ∂α = c0 xα ln x α + x x2 + + ··· 1 + α (1 + α)(2 + α)(1 + α) (7.4 Example . Recurrence Rel: cn = (n + α)(n + α − 1) Note: When α = 0.24) 2 1 + + ··· (1 + α) (2 + α) 5 2 x − · · · = c0 u2 ..the roots differ by an integer x = 0 is a regular singular point. . cn−1 n ≥ 1. c1 blows up! c0 c0 Let α = 1 ⇒ c1 = .22) {cn (x + α)(n + α − 1) − cn−1 } xn+α−1 + c0 (α − 1)αxα−1 = 0 n=1 Indicial Equation: (α − 1)α = 0 α = 0. . c2 = . (7. α) = αy(x. 2 12 y1 (x) = c0 x 1 + Second Solution: y (x.Bessel’s Equation 7. 4 .23) ∂y ¯ ∂α α=0 x + ··· 1+α x x2 + c0 xα 1 − − 2 (1 + α) (1 + α)2 (2 + α) x2 x3 = c0 x + + + · · · ln x + c0 1 − x − 2 12 x2 x2 + + ··· 2 12 Therefore y(x) = (A + B ln x) x + 5 + B 1 − x − x3 − · · · 4 48 . ∞ Let Ly = xy − y = 0. 1 differ by integer. x x2 + + ··· 2 12 = c0 u1 (x). .

Wave Equation: c = Wave Speed. Laplace’s Equation: 49 . • Heat Flow in a Bar 2. Heat Equation: α2 = Thermal Conductivity.1 Types of Boundary Value Problems: Dirichlet Boundary Conditions 1.Chapter 8 Separation of Variables Lecture 10 8. •Heat Flow on a Disk • Vibration of a String 3.

Ice 50 . t) = B Heat Bath 2. t) = A u(L.Separation of Variables Neuman Boundary Conditions: What do you expect the solution to look like as t → ∞? Mixed Boundary Conditions: Ice Heat Bath u(0.

t) = X(0)T (t) = BT (t) ⇒ B = 0 0 = u(L.3) x> Guess X (x) + λ2 X(x) = 0 X(x) = erx ⇒ (r2 + λ2 )erx = 0 r = ±λi X = c1 eiλx + c2? e−iλx = A sin λx + B cos λx. Fourier’s Guess: u(x. X(x) α T (t) dT = −α2 λ2 dt T ln |T | = −α2 λ2 t + c 2 2 T (t) = De−α λ t . t) = X(x)T (t) ˙ ut = X(x)T (t) = α2 uxx = α2 X (x)T (t) ÷α2 XT : ˙ T (t) X (x) = 2 = Constant = −α2 .2) −> ˙ T (t) = −α2 λ2 T (t) (8.Fourier sine Series: Consider the heat conduction in an insulated rod whose endpoints are held at zero degree for all time and within which the initial temperature is given by f (x).2 Separation of Variables .2.4) (8. SEPARATION OF VARIABLES . (8. t) = X(L)T (t) = (A sin λL)T (t).6) 51 (8.FOURIER SINE SERIES: 8. (8. Impose the boundary conditions: 0 = u(0.5) .8.1) (8.

a linear combination of solutions is again a solution. L L Answer: In fact they do agree on [0. . Thus we look for values of λ such that sin λL = 0 ⇒ λ = nπ L n = 1. L (8.10) Given f (x) we need to find the bn such that the infinite series of functions nπx agrees with f on [0. Thus n = 1.9) What about the initial condition u(x. (8. . . t) = n=1 bn sin nπx −α2 ( nπ )2 L e t. . number) is linear.7) 2 nπ 2 nπx un (x. .Separation of Variables Now we do not want the trivial solution so A = 0. 2.8) Since (8. L].8) (above eq. . bn sin L Question: f (x) may not be periodic f (x + 2L) = f (x) but the series is nπ nπx periodic since sin (x + 2L) = sin . (8. t) = e−α ( L ) t sin L are all solutions of ut = α2 uxx . ∞ u(x. L (8. 0) = f (x) = n=1 bn sin nπx . 0) = f (x). . Thus the most general solution is ∞ u(x. 2. L] and are different elsewhere. 52 .

12) Express f in terms of the basis vectors v1 . k = i. Just as in the case with matrices we obtain sequence of eigenvalues which in this case is infinite: λn = nπ L n = 1. .. .. the vk are orthogonal αk = f · vk vk · vk (8.14) But functions are just infinite dimensional vectors: 53 . Aside: How do we expand a vector? (8. . v2 . 2.. .11) and corresponding eigenfunctions nπx xn (x) = sin λn x = sin L πx 2πx 3πx sin . sin .2.e. SEPARATION OF VARIABLES . v3 f = α1 v1 + α2 v2 + α3 v3 f ·  k = α1 v1 · vk + α2 v2 · vk  α3 v3 · v v +  k   v1 · v1 v1 · v2 v1 · v3 α1 f · v1  v1 · v2 v2 · v2 v2 · v3   α2  =  f · v2  v1 · v3 v2 · v3 v3 · v3 α3 f · v3 If vk ⊥ v . L L L Recall that for symmetric matrices the eigenvectors form a basis.FOURIER SINE SERIES: Lecture 11 How do we find the bn ? Observe that we have a new type of eigenvalue problem subject to X(0) = 0 X(L) = 0.8. (8.13) (8. sin .

fN ] [g1 . f2 . . .17) nπx sin L kπx L f (x) sin 0 dx = n=1 bn 0 sin dx. .18) dx Inn = 0 sin2 nπx L dx = 1 2 0 1 − cos 2nπx L = L/2 L Therefore bk = 2 L 0 f (x) sin kπx L dx. g . gN ] ∆x = L N (8. 54 . . . . Therefore 2 sin nπx sin L kπx L dx Ink = 0 L = 1 2 0 cos(n − k) πx πx − cos(n + k) dx L L L 0 n=k 1 sin(n − k)πx/L sin(n + k)πx/L − 2 (n − k)π/L (n + k)π/L = 0 = L L (8. . Recall sin(A) sin B = L 1 {cos(A − B) − cos(A + B)}.15) f · g = f1 g1 + f2 g2 + · · · + fN gN N = k=1 f (xk )g(xk ). Now N L f (xk )g(xk )∆x k=1 0 f (x)g(x) dx = f.16) Back to finding bn : ∞ f (x) = n=1 L bn sin kπx L nπx L ∞ L (8. (8.Separation of Variables f g [f1 . g2 . .

n 55 .5 −1 −2 1 terms of the Fourier Series 1 0.20) bn = 2 0 x sin(nπx) dx = −2 ∞ n=1 Therefore u(x. (8. 0) = 2 π • π2 = 8 ∞ k=0 ∞ k=0 (−1)k sin (2k + 1)πx . t) → 0 (all the heat leaks out).1 f (x) = 2x 0<x< 1 L=1 2 1 2(1 − x) 2 < x < 1   1 1   2   2x sin(nπx) dx + 2(1 − x) sin(nπx) dx = 2     1 0 2 bn sin(nπ/2) = 8 n2 π 2 Therefore u(x.8.19) (2k + 1)2 • Observe as t → ∞ u(x.5 f(x) 0 −0. SEPARATION OF VARIABLES .5 −1 0 x 1 2 −1 −2 −1 0 x 1 2 Example 8. t) = 8 π2 ∞ k=0 sin nπ L n = 1 2 3 4 5 1 0 −1 0 1 (−1)k 2 2 sin (2k + 1)πx e−(2k+1) π t .5 f(x) 0 −0. 2 2 terms of the Fourier Series 1 (2k + 1)2 1 0. (2k + 1)2 by letting x = 1 ⇒ f (x) = 1.2 f (x) = x 1 0<x<1 L=1 cos(nπ) (−1)n+1 =2 nπ nπ (8. 8 • u(x. t) = 2 π (−1)n+1 2 sin(nπx)e−(nπ) t .2.FOURIER SINE SERIES: Example 8.

.5 −0. 2 π ∞ n=1 • u(x.. t) → 0.5 −1 −2 −1 −2 −1 0 x 1 2 56 ..5 0.0 2 (−1)n+1 sin(nπx). n 1 2 = = 2 π ∞ n=1 ∞ k=0 (−1)n+1 n u • = sin(nπ/2) 2 π (−1)k (2k + 1) π 4 = 1− 1 1 + − . 3 5 k n sin nπ 2 0 1 1 2 0 1 3 −1 4 0 2 5 1 (8. 0) = 1 .5 f(x) 0 f(x) −1 0 x 1 2 0 −0.21) 1 1 terms of the Fourier Series 1 2 terms of the Fourier Series 0.Separation of Variables • As t → ∞ u(x.

t) ∂u ∂u (−L. ∂r ut = α2 uxx u(−L.22) BC: Periodic BC IC: Assume u(x. t) ∂x ∂x u(x.FULL FOURIER SERIES Lecture 12 8. ˙ T (t) X (x) = 2 = −λ2 . HEAT EQ ON A CIRCULAR RING . As before: X(x) α T (t) ˙ T (t) 2 IVP: 2 = −λ2 ⇒ T (t) = ce−λ t . t) = X(x)T (t). t) = u(L. 0) = f (x) (8.23) 57 .8. t) = (L. = 0. 2π π The Laplacian becomes ∆u = = ∂ 2 u 1 ∂u 1 ∂2u + + 2 2 ∂r2 r ∂r r ∂θ 2u ∂ ∂(rθ)2 (8. α T (t) 2L L r= = = Constant.3.Full Fourier Series Physical Interpretation: Consider a thin circular wire in which there is ∂u no radial temperature dependence.3 Heat Eq on a Circular Ring .

 0  X(L)  X (L) Eigenvalue Problem look for λ such that nontrivial x can be found.25) Superposition of all these solutions yields the general solution ∞ u(x.27) As before we obtain expressions for the An and Bn by projecting f (x) onto nπx nπx the basis functions sin and cos . X(x) = A cos(λx) + B sin(λx) X(−L) = A cos(λL) − B sin(λL) = A cos(λL) + B sin(λL) = X(L) therefore 2B sin(λL) = 0 X (x) = −Aλ sin λx + Bλ cos(λx) therefore 2Aλ sin(λL) = 0 Therefore λn L = (nπ) n = 0. . t) = A0 + n=1 An cos nπx nπx + Bn sin L L e−( L ) nπ 2 2 α t . . 0) = A0 + n=1 An cos nπx nπx + Bn sin . L L (8.24) X (−L) = +Aλ sin(λL) + Bλ cos(λL) = −Aλ sin(λL) + Bλ cos(λL) = X (L) An cos nπx nπx + Bn sin L L . .1) that satisfy the BC are thus of the form un (x.Separation of Variables if we let x = rθ we obtain X + λ2 X = = BVP: X(−L) X (−L) = (1. (8. t) = e−( L ) nπ 2 2 α t (8. Solutions to (1.1).28) + n=1 ∞ An −L L cos nπx L nπx L sin cos sin cos mπx L mπx L mπx L mπx L dx + n=1 Bn −L sin dx. (8. 58 . L L L f (x) −L sin cos mπx L mπx L L dx = A0 −L ∞ sin cos mπx L mπx L L dx (8. 1.26) In order to match the IC we have ∞ f (x) = u(x. .

    Note: 1. HEAT EQ ON A CIRCULAR RING .9) respectively] represent the full Fourier Series Expansion for f (x) on the interval [−L.8.3. Plugging these orthogonality conditions into (1. (1.6) we obtain L A0 = 1 2L −L L f (x) dx = average value of f (x) on [−L.FULL FOURIER SERIES As in the previous example we use the orthogonality relations: L sin −L L mπx nπx sin L L mπx nπx cos L L dx = Lδmn cos −L dx = Lδmn m and n = 0 (8.6) and (1. L L (8. these could be renamed to (9.6) is often written in the form f (x) = a0 + 2 ∞ an cos n=1 nπx nπx + bn sin .29) = 2L m = n = 0 L sin −L mπx nπx cos L L dx = 0 ∀m.  L  2A0 1 nπx dx = and bn = Bn 2. n.30)     dx. L] L An = 1 L −L nπx f (x) cos L 1 dx and Bn = L −L f (x) sin nπx L          (8.9) [typist’s note: check re-numbering when equations are re-labeled.6) and (9. By defining an = f (x) cos  L L An −L the Fourier Series (1. L].31) 59 .

Separation of Variables 60 .

How does this relate to f (x) which may not be periodic? The function S(x) represented by the series is known as the periodic extension of f on [−L. 61 . λ = L L L n nπ nπx and similarly sin (x + 2L) = sin .2) Note: 1. As a result the function S(x) is also periodic. L].1) is a periodic function having a period 2L.Fourier Series We consider the expansion of the function f (x) of the form a0 f (x) ∼ + 2 where L L ∞ an cos n=1 nπx nπx + bn sin = S(x) L L (9. bn = 1 L −L f (x) sin dx (9. Note that cos nπ nπx nπλ 2L provided (x + λ) = cos = 2π.1) an = 1 L −L L nπx f (x) cos L nπx L dx a0 1 = 2 2L −L f (x) dx = average value of f . Thus each of the terms L L of the Fourier Series S(x) on the RHS of (10.Chapter 9 Lecture 13 .

Lecture 13 .3) Example 9.Fourier Series 2.4) 62 . If f (or its periodic extension) is discontinuous at a point x0 then S(x) converges to the average value of f across the discontinuity.1 f (x) = 0 x −π < x < 0 0≤x≤π L=π (9. S(x0 ) = 1 f (x+ ) + f (x− ) 0 0 2 (9.

π π a0 = 1 π −π π 1 f (x) dx = π 0 x dx = π 2 (9. 2. sin(nx) dx    a2m+1 π 1 π sin 1 (nπ) 2 cos(nx) + π n n 0 1 n 1 2 3 4 = (−1)n − 1 n − 1 −2 0 −2 0 2 (−1) πn 2 = − m = 0. cos 0 1 −π + + 2 sin(nx) n n n n+1 ∞ = (−1) f (x) = = /n an cos(nx) + bn sin(nx) (9.6) (9. π(2m + 1)2 = (9.8) a0 + 2 π 2 − 4 π n=1 ∞ m=0 cos (2m + 1)x + (2m + 1)2 ∞ (−1)n+1 n=1 sin(nx) n (9. . .5) an = 1 π −π π f (x) cos(nx) dx = 1 π x cos(nx) dx π π = 0  1  sin(nx) x π n − 0 1 n 0 1.9) 63 . 1. cos(nx) dx    π 0 = 1 π cos(nπ) 0.7) π bn = 1 π −π π f (x) sin(nx) dx = 1 π x sin(nx) dx π 0 π = 0  1 cos(nx) −x π n 1 + n 0 1. .

0 t = x − 2π dx = dt x = t + 2π x = π ⇒ t = −π (9.10) bn = 1 L −L f (x) sin dx = 1 L c fe (x) sin dx. 3π] fe (x) = 0 −π < x < 0 x 0≤x≤π (9.13) an = = = 1 π 1 π 1 π 3π fe (x) cos(nx) dx π 3π (x − 2π) cos(nx) dx 2π π t cos(nt) dt.11) Example 9. L L L c+2L an = 1 L −L L nπx f (x) cos L nπx L 1 dx = L c c+2L fe (x) cos nπx L nπx L dx (9.12) 0 π < x < 2π x − 2π 2π ≤ x ≤ 3π (9.1 It can be useful to shift the interval of integration from [−L.Fourier Series 9. c + 2L] Since the periodic extension fe (x) is periodic with period 2L (as are the nπx nπx basis functions cos and sin ).14) x = 3π ⇒ t = π since cos n(t + 2π) = cos t 64 . L] to [c. (9.2 Previous Example: f (x) = On [π.Lecture 13 .

18) cn = f (x)e−i( nπx L ) dx.16) = 1 2L −L f (x)e−i( nπx L b−n = −bn (9.2 Complex Form of Fourier Series f (x) = nπx cos L a0 + 2 e ∞ an cos n=1 nπx nπx + bn sin L L = Therefore f (x) = = nπx nπx nπx nπx ei( L ) − e−i( L ) + e−i( L ) . sin = 2 L 2i ∞ nπx nπx a0 an i( nπx ) bn i( nπx ) + e L + e−i( L ) + e L − e−i( L ) 2 2 2i i( nπx ) L a0 + 2 ↑ c0 ∞ n=1 ∞ n=1 an − ibn 2 ↑ cn nπx L ei( nπx L )+ an + ibn 2 ↑ c−n e−i( nπx L ) (9. COMPLEX FORM OF FOURIER SERIES 9.9.19) 65 .2. (9.15) = n=−∞ cn ei( ) an − ibn cn = 2 L = 1 2L −L L f (x) cos nπx nπx − i sin L L ) dx dx (9.17) Therefore ∞ f (x) = n=−∞ L cn ei( 1 2L −L nπx L ) (9.

3 f (x) = −1 −π ≤ x < 0 L=π 1 0<x<π   0 π   1 − e−inx dx + e−inx dx  2π  −L 0 (9. (9.Fourier Series Example 9.21) = = Therefore 1 2π − 0 e−inx −π (−in) + π e−inx 0 (−in) 0 n even (2/iπn) n odd (9.24) 66 .Lecture 13 .20) cn = (9.23) i −2 + e+inπ + e−inπ = 2πn f (x) = 2 ei πi(2n + 1) n=−∞ ∞ (2n+1)x .22) (9.

If f (x) = E(x) · O(x) then f (−x) = E(−x)O(−x) = −E(x)O(x) = −f (x) ⇒ f is odd. 67 . E1 (x) · E2 (x) → even.3) 0 Notes: Let E(x) represent an even function and O(x) an odd function. 1.Chapter 10 Lecture 14 .Even and Odd Functions Even: f (−x) = f (x) Odd: f (−x) = −f (x) 10.1 Integrals of Even and Odd Functions L 0 L f (x) dx = −L −L L f (x) dx + 0 f (x) dx (10.2) = 2 0 f (x) dx f even f odd. (10.1) = 0 f (−x) + f (x) dx    L (10. 2.

Even and Odd Functions 3. (10. L 2 an = L 0 f (x) cos nπx L dx. O1 (x) · O2 (x) → even.Lecture 14 . Any function can be expressed as a sum of an even part and an odd part: f (x) = 1 1 f (x) + f (−x) + f (x) − f (−x) . 2 (10. 4.5) (10.6) 10. Similarly let O(x) = O(−x) = 1 f (x) − f (−x) 2 1 f (−x) − f (x) = −O(x) odd.8) odd Therefore a0 f (x) = + 2 ∞ n=1 L nπx an cos .4) Check: Let E(x) = 1 1 f (x) + f (−x) . Then E(−x) = f (−x) + f (x) = 2 2 E(x) even. 2 2 even part odd part (10. (10.2 Consequences of Even/Odd Property for Fourier Series (I) Let f (x) be Even-Cosine Series: L L an = 1 L −L L nπx f (x) cos L even f (x) sin nπx L 2 dx = L 0 f (x) cos nπx L (10.9) 68 .7) dx bn = 1 L −L dx = 0.

69 .10. CONSEQUENCES OF EVEN/ODD PROPERTY FOR FOURIER SERIES (II) Let f (x) be Odd-Sine Series: L an = 1 L −L L f (x) cos odd nπx L dx = 0 (10.11) 1 1 f (x) + f (−x) + f (x) − f (−x) 2 2 ∞ ∞ a0 nπx nπx + an cos + bn sin 2 L L n=1 n=1 = where L L an = 2 L 0 L 1 nπx f (x) + f (−x) cos 2 L 1 nπx f (x) − f (−x) sin 2 L 0 1 dx = L −L L f (x) cos nπx L nπx L dx bn = 2 L 1 dx = L −L f (x) sin dx. we can interpret the cos and sin series as even/odd: f (x) = even odd (10.10) L bn = 1 L −L nπx f (x) sin L even 2 dx = L 0 f (x) sin nπx L dx Therefore ∞ L f (x) = n=1 nπx . bn sin L 2 bn = L −0 f (x) sin nπx L dx.2. (III) Since any function can be written as the sum of an even and odd part.

15) Example 10. (a) 70 . Cosine Series: a0 + 2 L ∞ f (x) = an cos n=1 nπx L dx.14) bn = 2 L 0 f (x) sin (10. L] and we want to represent f by a Fourier Series we have two choices .1 Expand f (x) = x 0 < x < 2 in a half-range (a) Sine Series.12) an = 2 L 0 f (x) cos nπx L (10. (10.13) Sine Series: ∞ f (x) = n=1 bn sin L nπx L nπx L dx.Lecture 14 .Even and Odd Functions 10.a Cosine Series or a Sine Series. (b) Cosine Series. (10.3 Half-Range Expansions If we are given a function f (x) on an interval [0.

18) = − 4 cos(nπ) + nπ 2 nπ sin (10.3. n 2 (10. HALF-RANGE EXPANSIONS bn = 2 0 2 f (t) sin nπ t dt (10.16) = 0 t sin nπ t dt 2 2 2 (10.10.20) Therefore 4 π ∞ n=1 f (t) = (−1)n+1 nπ sin t .17) = − t cos nπ t 2 nπ 2 0 2 + nπ 0 cos 2 nπ t dt 2 nπ t 2 ∞ 2 (10.21) (b) 2 a0 = 1 2 0 2 1 t2 t dt = 2 2 2 =1 0 2 (10.19) 0 n+1 (−1) 4 sin f (1) = 1 = π 4 n = − (−1)n n=1 nπ π = 1 − 1 + 1 − 1 + ··· 4 3 5 7 nπ 2 (10.22) an = 0 nπ t cos t dt = 2 2 nπ 2 2 nπ 2 nπ 2 t sin t − 2 0 2 nπ 0 sin nπ t dt 2 (10.23) = + cos nπ t 2 = 0 4 {cos nπ − 1} n2 π 2 71 .

25) The cosine series converges faster than Sine Series.24) (10. 8 f (2) = 2 = 1 + 2 π ∞ n=0 1 (2n + 1)2 π2 1 1 = 1 + 2 + 2 + ··· 8 3 5 72 . 2 (10.Lecture 14 .Even and Odd Functions Therefore f (t) = 1 + = 1− 4 π2 8 π2 ∞ n=1 ∞ n=0 (−1)n − 1 nπ cos t 2 n 2 cos (2n + 1) πt/(2n + 1)2 .

Chapter 11 Lecture 15 . 2L = 2 ⇒ L = 1. .1 (Completion of problem illustrating Half-range Expansions) Periodic Extension: Assume that f (x) = x.1) (11. 0 < x < 2 represents one full period of the function so that f (x + 2) = f (x).2) 73 since f (x + 2) = f (2).Convergence of Fourier Series Example 11. L 1 2 a0 = 1 L −L f (x) dx = −1 f (x) dx = 0 x dx = x2 2 2 =2 0 (11.

3) bn = 1 L −L 2 nπx f (x) sin L  dx = −1 f (x) sin(nπx) dx 2  cos(nπx) dx = 0 x sin(nπx) dx = −x sin(nπx) −2 + nπ (nπ)2 2 cos(nπx) nπ 2 0 + 1 (nπ) 0 = Therefore = 0 −2 nπ (11.4) f (x) = 2 2 − 2 π 2 π ∞ n=1 ∞ n=1 sin(nπx) n sin(nπx) n (11.Convergence of Fourier Series n ≥ 1: L 1 an = 1 L −L 2 nπx f (x) cos L dx = −1 f (x) cos(nπx) dx L = 1 = 0 x cos(nπx) dx  2 2  sin(nπx) dx x sin(nπx) =  nπ = − 0 2 0 1 nπ 0 1 cos(nπx) (nπ)2 L 1 = cos(2nπ) − 1 = 0 (nπ)2 1 (11.Lecture 15 .6) 74 .5) = 1− (11.

3 2 S(x) 1 0 −1 −4 −2 0 x 2 4 N =20 n=1 Figure 11.2: Full Range Expansion SN (x) − 1 = − π sin(nπx) n 75 .1: Full Range Expansion SN (x) = 1 − 2 π sin(nπx) n 2 1 S(x)−1 0 −1 −2 −4 −2 0 x 2 4 N =20 n=1 2 Figure 11.

7) L f (x) cos −L nπx L dx and bn = 1 L f (x) sin −L nπx L dx.Lecture 15 .2 Let f and f be piecewise continuous functions on [−L.Convergence of Fourier Series 11. The Fourier Series converges to f (x) at all points at which f is continuous 1 and to f (x+) + f (x−) at all points at which f is discontinuous. then f has a Fourier Series f (x) = where an = 1 L L a0 2 ∞ + n=1 an cos nπx L + bn sin nπx L = S(x) (11. 2 • Thus a Fourier Series converges to the average value of the left and right limits at a point of discontinuity of the function f (x). L] and let f be periodic with period 2L. • We consider piecewise continuous functions: Theorem 11.1 Convergence of Fourier Series • What conditions do we need to impose on f to ensure that the Fourier Series converges to f . 76 .

5 −2 −1 0 x/π 1 2 Figure 11. (11.9) sin(nx) n = 4 π ∞ m=0 sin(2m+1)x (2m+1) .2.8) f (x) = 1 = where bn = = Therefore f (x) = bn sin(nx) n=1 π 2 sin(nx) dx π 0 4/πn n odd 0 n even 4 π n ∞ n=1 = 2 π − cos nx n π 0 = 2 πn 1 − (−1)n (11. There3 5 77 .5 0 −0.overshoot.5 −1 −1. π].5 1 S (x) for N=5 0.2 Illustration of the Gibbs Phenomenon • Near points of discontinuity truncated Fourier Series exhibit oscillations . 4 3 5 ∞ 1− 1 1 + − · · · . ILLUSTRATION OF THE GIBBS PHENOMENON 11.11.3: Fourier Series for a step function Example 11.3 Consider the half-range sine series expansion of f (x) = 1 ∞ N on [0. odd Note: sin (2m + 1)π/2 4 4 1. 1. f (π/2) = 1 = = π (2m + 1) π m=0 π 1 1 fore = 1 − + − · · · .

3 Now consider the sum of the first N terms SN (x) = 4 π N m=0 sin(2m + 1)x 4 = Im (2m + 1) π N N m=0 ei(2m+1)x (2m + 1) (11.Lecture 15 .19) 1 − ei2(N +1)x eix − e−ix ei2(N +1)x − 1 sin x 2 sin 2(N + 1)x . π sin x . Recall the complex Fourier Series example for the function f (x) = −1 −π ≤ x < 0 1 0<x<π (11.Convergence of Fourier Series 2.13) m 4 Im ieix π 4 Im ieix π 4 Im ieix π 4 Im i π 2 Im π ei2x m=0 (11.12) SN (x) = = 4 Im π iei(2m+1)x m=0 N (11.17) (11.16) (11.10) which turns out to be equivalent to the odd extension of the above function represented by the half-range sine expansion.11) n=1 odd 11.14) N = = = = = 78 1 + ei2x + · · · + ei2x 1 − ei2x 1 − ei2(N +1)x 1 − ei2x (1 − ei2x ) (11. which we can see from the following calculation ∞ f (x) = n=−∞ 2 inx πin e sin(nx) .18) (11.15) (11. n = 4 π n ∞ n=1 einx −e−inx 2in = n 4 π odd n ∞ odd (11.

NOW CONSIDER THE SUM OF THE FIRST N TERMS Therefore t = 2(N + 1)u x 0 2(N +1)x du = dt 2(N +1) (11.5 2 79 .5 1 x/π 1.4: (2/π)sin(2(N + 1)x)/sin(x) for N = 5 2 sin 2(N + 1)x = 0 when 2(N + 1)xN = π thus the π sin x maximum value of SN (x) occurs at xN = π 2(N + 1) (11.21) 0 0.3.20) sin t SN (x) = 2 π sin 2(N +1)u sin u du 2 π t dt 0 (2/π) sin(2(N+1)x)/sin(x) −10 Observe SN (x) = −5 10 0 5 Figure 11.11.

Convergence of Fourier Series (2/π) ∫0 sin 2(N+1)u /sin u du for N= 5 −1.5 0 Figure 11.5 2 80 .5 1 x/π 1.5: Integral of (2/π)sin(2(N + 1)x)/sin(x) x −0.Lecture 15 .5 0.5 −1 0 1 0.5 1.

4) 4 Example 12. n n n=1 ∞ n=1 (12.2) For a full Fourier Series on [−L.1) = m=1 n=1 bm bn · δmn · b2 . n Proof: L ∞ ∞ L f (x) 0 2 dx = m=1 n=1 ∞ ∞ bm bn 0 sin mπx nπx sin L L L L = 2 2 ∞ dx (12. n 2 81 . Then L L 0 f (x) 2 dx = n=1 b2 . n n=1 (12. 2] f (x) = x = π (−1)n+1 nπx sin .Parseval’s Identity Lemma 12. L] Parseval’s Theorem assumes the form: f (x) = L a0 + 2 a2 0 + 2 ∞ an cos n=1 ∞ nπx nπx + bn sin L L (12.2 Recall for x ∈ [0.Chapter 12 Lecture 16 .1 (A version of Parseval’s Identity) ∞ L ∞ Let f (x) = n=1 nπx 2 bn sin 0 < x < L.3) 1 L −L f (x) 2 dx = a2 + b2 .

5) 1 1 Note: = 2 (2n)2 2 n=1 Also note that π2 6 ∞ ∞ ∞ n=1 1 1 = n2 4 π2 6 = evens = n=1 1 n2 ∞ odds ∞ = = m=1 π2 24 1 (2m)2 + m=0 ∞ 1 (2m+1)2 1 (2m+1)2 + m=0 Therefore ∞ m=0 1 π2 π2 π2 = − = .1 Geometric Interpretation of Parseval’s Formula f = b1ˆ1 + b2ˆ2 e e |f |2 = f · f = b2ˆ1 · ˆ1 + b2ˆ2 · ˆ2 e e 1e 2e = b2 1 + b2 2 Pythagoras’ Theorem (12.Parseval’s Identity Therefore 2 L L f (x) 0 2 dx = ⇒ 2 2 2 0 x2 dx = 2 x3 3 0 π2 6 4 2 π ∞ n=1 ∞ 1 n2 = = π2 . 24 4 2 1 π n2 n=1 ∞ 1 n2 n=1 (12. n (12.10) 82 .7) (12.9) For Fourier Sine Components: L 2 L 0 ∞ f (x) 2 dx = n=1 b2 . (2m + 1)2 6 24 8 (12.Lecture 16 .6) 12.8) (12.

n2 (12. GEOMETRIC INTERPRETATION OF PARSEVAL’S FORMULA Example 12.3 Consider f (x) = x2 −π < x < π.15) 83 . k2 (12. The Fourier Series Expansion is: x2 = π2 +4 3 ∞ n=1 (−1)n cos(nx). n4 (12.1.11) cos Let x= π 2 nπ 2 n 1 2 3 4 0 −1 0 1 ⇒ π2 4 = 2 −π 12 Therefore = n=1 ∞ (−1)k 4 (2k)2 k=1 π2 3 ∞ +4 (−1)n n2 cos nπ 2 (12.12) π2 = 12 By Parseval’s Formula: 2 π π ∞ k=1 (−1)k+1 .14) Therefore π4 90 ∞ = n=1 1 = δ?(4).12.13) x4 dx = 2 = 0 π 2 x5 π 5 0 π2 3 2 ∞ + 16 ∞ n=1 1 n4 1 n4 9−5 45 = 2π 4 9 + 16 n=1 4 45 1 90 = 8 90 (12.

Lecture 16 .Parseval’s Identity 84 .

∆x→0 ∆x (13. .1 Approximating the Derivatives of a Function by Finite Differences Recall that the derivative of a function was defined by taking the limit of a difference quotient: f (x) = lim f (x + ∆x) − f (x) . ∆x 2 3! 85 .1) Now to use the computer to solve differential equations we go in the opposite direction . . If we assume that the function can be differentiated many times then Taylor’s Theorem is a very useful device in determining the appropriate difference quotient to use.2) . .Solving the heat equation using finite difference methods 13. For example consider f (x + ∆x) = f (x) + ∆xf (x) + ∆x2 ∆x3 (3) ∆x4 (4) f (x) + f (x) + f (x) + . 2! 3! 4! Re-arranging terms in (2) and dividing by ∆x we obtain ∆x ∆x2 (3) f (x + ∆x) − f (x) = f (x) + f (x) + f (x) + . (13..we replace derivatives by appropriate difference quotients.Chapter 13 Lecture 17 .

3) ∆x f (ξ) in (3) as a measure of the error involved when 2 we approximate f (x) by the difference quotient f (x + ∆x) − f (x) /∆x. We note that the central difference schemes (5) and (6) are second order accurate while the forward difference scheme (3) is only O(∆x). 2! 3! 4! and we subtract (4) from (2) and divide by (2∆x) we obtain: f (x + ∆x) − f (x − ∆x) ∆x2 (3) = f (x) + f (ξ). on the smoothness of f ) and on the size of ∆x.6) Due to the symmetry of the difference approximations (5) and (6) about the expansion point x these are called central difference approximations.. Technically a term or function E(∆x) is O(∆x) if We retain the term E(∆x) ∆x ∆x→0 → const. 2 ∆x 12 (13. 2∆x 3! (13. Now the difference quotient (3) is not the only one that can be used to approximate f (x). In order to obtain an approximation to f (x) we add (2) to (4) which upon re-arrangement and dividing by ∆x2 leads to: f (x + ∆x) − 2f (x) + f (x − ∆x) 1 = f (x) + ∆x2 f (4) (ξ). The difference approximation (3) is known as a forward difference approximation. But for our purposes it is more useful to retain the approximation f (x + ∆x) − f (x) ∆x ∆x f (ξ) 2 = f (x) + O(∆x).Lecture 17 .5) We notice that the error term associated with this form of difference approximation is O(∆x2 ).Solving the heat equation using finite difference methods If we take the limit ∆x → 0 then we recover (1). .4) .e. = f (x) + (13. Indeed if we consider the expansion of f (x − ∆x): f (x − ∆x) = f (x) − ∆xf (x) + ∆x2 ∆x3 (3) ∆x4 (4) f (x) − f (x) + f (x) + . x + ∆x] (i. which converges more rapidly to zero as ∆x → 0. (13. 86 . Since we like to focus on that part of the error we can control we say that the error term is of the order ∆x – denoted by O(∆x). Notice that this error depends on how large f is in the interval [x.

we consider the following partial difference approximations. t + ∆t) − u(x. t) = 0 u(1. t) + (x. ∂t 2! ∂x2 After re-arrangement and division by ∆t: u(x. HEAT EQUATION SOLUTION BY FINITE DIFFERENCES 13. 0) = f (x). t) + · · · 4! ∂x2 ∆x4 ∂ 4 u (x.13. t) + ∆x ∂u (x. Corresponding to the difference quotient approximations introduced in Section 1.9) The basic idea is to replace the derivatives in the heat equation by difference quotients.12) {u(x + ∆x. t)} . t) + u(x − ∆x. t) − 2u(x. t) − 2u(x. 87 u(x + ∆x. t) − 2! ∂x2 ∆x3 ∂ 3 u (x. t) + u(x − ∆x. t) + O(∆x2 ). t) = u(x. t) ∂u = (u. (13. Forward Difference in Time: u(x. ∆x2 ).8) (13. t) + 3! ∂x3 ∆x3 ∂ 3 u (x. t > 0 ∂t ∂x BC: u(0. t) + · · · . t) + α2 ∆t ∆x2 (13. 4! ∂x4 (13. t) − ∆x (x. Consider the following initial-boundary value problem for the heat equation (13. . t) + 2! ∂x2 ∆x2 ∂ 2 u (x. t) = α2 ∆t Re-arranging: u(x. t) − 2u(x. t) + O(∆t).7) (13. t) = u(x. We consider the relationships between u at (x. t) + u(x − ∆x. t) + · · · .2. t) and its neighbours a distance ∆x apart and at a time ∆t later. t + ∆t) = u(x. t + ∆t) = u(x.11) 2 ∆x ∂x2 Substituting (2) and (3) into (1a) we obtain u(x. t) + ∆t ∂u ∆t2 ∂ 2 u (x. t) + ∂x ∆x2 ∂ 2 u (u. ∆t ∂t Central Differences in Space: u(x + ∆x. t) ∆x2 + O(∆t.10) Adding and re-arranging: u(x + ∆x. t + ∆t) − u(x. t) = 0 IC: u(x.2 Heat Equation solution by Finite Differences ∂u ∂2u = α2 2 0 < x < 1. t) + ∂x ∂u u(x − ∆x. t) + 3! ∂x3 ∆x4 ∂ 4 u (x. t) ∂2u = (x.

1] into N + 1 equally spaced sample points xn = n∆x. T ] is subdivided into M + 1 equal time levels tk = k∆t. tk ) uk+1 0 uk+1 n uk . The time interval [0.Solving the heat equation using finite difference methods We subdivide the spatial interval [0. 88 . At each of these space-time sample points we introduce approximations: u(xn . n uk+1 N tk+1 u 6 ∆t tk ? u T u    u   u uk n−1 uk -n ∆x T T u T T Tu uk n+1 u uk −1 N u uk N uk 0 uk+1 = uk + α2 n n ∆t ∆x2 uk − 2uk + uk n+1 n n−1 ↑ This is implemented in the spread sheets Heat0 and Heat.Lecture 17 .

where xN = N ∆x = ∂x 1. u(xN + ∆x. N N N N Note that this BC could be implemented another way without introducing the additional column. t) (∗) Since xN = 1 we observe that xN +∆x is outside the domain we introduce an extra column uN +1 into which we copy the values uN −1 . which is known as the stability boundary and observe what happens. t) = 0. Some EXERCISES and Observations: – Heat Equation 1. Change the ∆t in cell D1 from 0.it is much closer to what we would expect.00625. t) = u(xN − ∆x. namely: uk+1 = uk + α2 ( N N ∆t )(uk +1 − 2uk + uk −1 ) (∗∗) N N N ∆x2 ↑ This is implemented in the spread sheet Heat0f. Now change ∆t to 0. HEAT EQUATION SOLUTION BY FINITE DIFFERENCES Implementing Derivative Boundary Conditions: Assume that the boundary conditions (1b) are changed to BC: u(0.2.001 to 0. 89 . by eliminating uN +1 from (∗) and (∗∗): uk+1 = uk + 2α2 N N ∆t ∆x2 uk −1 − uk .05 and you will observe what is known as a numerical instability. t) − u(xN − ∆x. N N If this latter equation is implemented at xN there is no need to introduce an extra column UN +1 or to implement the difference equation given in (**) as the the derivative boundary condition is taken care of automatically. ∂u (1. Now let ∆t = 0. t). t) = 0. t) = 0. ∂x ∂u Consider a central difference approximation to (1. ∆x Re-arranging we obtain: u(xN + ∆x.13. In the column xN we implement the same difference approximation for the Heat Equation. while uk +1 = uk −1 (see (*) ) since column uk −1 is copied to column uk +1 .006 and observe the abrupt change in the solution .

Check the numerical solution against the problem solved in HW1 #3. 3. This class of error results from discarding the O(∆x2 ) and O(∆t) terms in (2) and (3) when we replace derivatives in (1a) by difference quotients. the other type of error is present in almost every type of numerical approximation scheme. To determine the truncation error change the spread sheet to implement the initial condition f (x) = 2x 0 < x < 1/2 . Although numerical instability is evident for a parameter choice that is unstable. The difference between the two is mainly due to the truncation error since the round-off error is about 10−12 and does not grow if stable parameters are used.Lecture 17 . 2(1 − x) 1/2 ≤ x < 1 Now code up the Fourier Series (in another spread sheet) that is derived on page 21 of the notes and compare the numerical solution to the ‘exact’ Fourier Series solution with 50 terms. This error is known as the truncation error. The instability noted in 1.Solving the heat equation using finite difference methods 2. above is not the only source of error in the numerical approximation. 90 . Implement derivative boundary conditions on both endpoints x = 0 and x = 1.

y) = 0. y) = 0 uxx + uyy = 0 u(1.Solving Laplace’s Equation using finite differences 14. u(x. 0) = f (x) 1 - x x0 x1 xn xN = 1 - 91 . 1) = 0. (14. y 6 1 u(x.Chapter 14 Lecture 18 . (14. 1) = 0 6 1 = yM ∆x  - u(0.1) u(x.1 Finite Difference approximation Consider the boundary value problem ∂2u ∂2u + 2 = 0 0 < x. 0) = f (x). y < 1 ∂y ∂x2 u(1. y) = 0 y1 ∆y 6 ? y0 u(x.2) BC: u(0. y) = 0.

y − ∆y) ∆y 2 = = ∂2u (14. (14. (14. y) − 2u(x. y) + u(x. (M − 1).4) ∂y 2 We partition the interval 0 ≤ x ≤ 1 into (N + 1) equally spaced nodes xn = n∆x and the interval 0 ≤ y ≤ 1 into (M + 1) equally spaced nodes ym = m∆y.xm ) + O(∆x2 .5) unm+1 u1j un−1m u 1j u unm un+1m 1j j -4 u u1j unm−1 This is known as the finite difference ‘Stencil’ that relates unm to its 4 nearest neighbours. ∆x2 ∆y 2 If we choose ∆x = ∆y then we obtain un+1m + un−1m + unm+1 + unm−1 − 4unm = 0 1 ≤ n. m ≤ (N − 1). ∆y 2 ).recall the boundary values are already specified! 92 . ym ) we obtain: un+1m − 2unm + un−1m unm+1 − 2unm + unm−1 + = (uxx + uyy )(xn . y) ∆x2 u(x.Lecture 18 . y) + u(x − ∆x. y + ∆y) − 2u(x. y) + O(∆y 2 ). This is a system of (N − 1) × (M − 1) unknowns for the values of unm interior to the domain .Solving Laplace’s Equation using finite differences As before we replace the second derivatives in (1a) by central difference quotients that are second order accurate: u(x + ∆x. ym ) by unm u(xn . y) + O(∆x2 ) ∂x2 ∂2u (x. Replacing the derivatives in (1a) by the difference quotients in (3) and (4) and representing the mesh values at (xn .3) (x.

EXERCISES and Notes for Laplace’s Equation: 93 . You can choose a surface plot to visualize the solution.2. Then select the Calculation Tab.2 Solving the System of Equations by Jacobi Iteration This is a procedure to solve the system of Equation (3) by looping through each of the mesh points and updating unm according to (3) assuming that the nearest neighbours already have values close to the exact solution. Now hold down the F9 key and watch the solution move to equilibrium.t t 6 average t Thus unm is the average value of its nearest neighbours.14. To implement the iterative procedure (6) on a spread sheet. If you set the number of iterations to 5 say. This procedure is repeated until the changes that are made in each iteration falls below a certain tolerance. Thus iteration can be viewed as taking successive neighbour averages until there is no change. go to the Tools Menu at the top of the screen and click on the Options Tab. To implement this iterative procedure we observe that the discrete Laplace Equation (5) can be re-written in the form: uk+1 = nm k k k uk n+1m + un−1m + unm+1 + unm−1 4 (14. This mean value property is a discrete form of a fundamental property of any solution to Laplace’s Equation. Note that a new superscript index k has been introduced to represent the nodal values at the kth iteration. at which point the value of umn equals the average of the values at its mesh neighbours. SOLVING THE SYSTEM OF EQUATIONS BY JACOBI ITERATION 14. This iterative process essentially uses diffusion on a pseudo time scale to take the solution to equilibrium. then if you start with zero values throughout the interior of the domain (as you should if you cut and paste as demonstrated in class). Check the Iteration box.6) t t ? . you will see the values percolate 5 cells into the domain from the non zero boundary condition f (x) = sin(πx).

(∗) 4 It may be useful to calculate the values of fnm on a separate sheet in which the same cell values as those for unm are maintained. Implement a 0 derivative BC along the lines x = 0 and x = 1. To ensure that (0. 94 .Lecture 18 . ym ) . ∂x 2. ∂x2 ∂y Introduce finite difference quotients. Implement an inhomogeneous term for Poisson’s Equation: ∂2u ∂2u + 2 = f (x. y < 1. y) = ∂x ∂u 0= (1.Solving Laplace’s Equation using finite differences 1. Plot a ∂u cross section of the results along y = 1/2. y). Then the values of fnm can be referenced in the calculation of unm according to (∗). assume ∆x = ∆y to arrive at the iterative formula: uk+1 = nm k k k 2 uk n+1m + un−1m + unm+1 + unm−1 − ∆x f (xn . y) 0 < x.

2) u(0. t) since u∞ (x) xx t = α2 u∞ (x) + v(x. t) = u1 t>0 (15. t) = u1 + v(L. t) ⇒ v(0. steady state solution Let u(x.1 Specified Temperatures ut = α2 uxx BC: u(0.1) (15.3) u0 . 0 < x < L. t) = u1 = u∞ (L) + v(L.1) ut = u∞ (x) + v(x. In this case (15. t) = u0 = u∞ (0) + v(0. t) xx ⇒ vt = α2 vxx (15.1) becomes α2 u∞ (x) = 0 ⇒ u∞ (x) = A0 x + B0 u∞ (0) = B0 = u0 u∞ (L) = A0 L + u0 = u1 ⇒ u∞ (x) = u1 −u0 x + u0 L . t) = 0 u(L. u1 constants Firstly consider the steady-state solution (i. t) ⇒ v(L.e.Chapter 15 Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC Example 15. t) = u0 u(x. t) = u∞ (x) + v(x. 95 .4) = 0.. u(L. Substitute into (15. t) = u0 + v(0. when ut = 0) which we denote by uα (x).2) (15. 0) = g(x). t). Substitute into (15. t) = 0.

t) = X(x)T (t).7) ∞ u1 − u0 L x+ n=1 2 nπ 2 nπx bn e−α ( L ) t sin L (15. n = 1. . t) = 0 = v(L. 0) = g(x) − u∞ (x) = n=1 nπx 2 bn sin ⇒ bn = L L 0 {g(x) − u∞ (x)} sin nπx L dx Thus the solution to the inhomogeneous problem is: u(x. 0) = g(x) − u∞ (x). t) = n=1 (15.9) Example 15.8) bn = 2 L 0 {g(x) − u∞ (x)} sin nπx L dx.2 Alternative-eigenfunction expansion approach. In order to solve the boundary value problem (15. (15. Thus we have to solve a new problem for v which has zero BC: vt = α2 vxx v(0. 0) = g(x) = u∞ (x) + v(x.10) .Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC Substitute into (15.1)-(15.5) T (t) = ce−λ X + λ2 X = 0 X(0) = 0 = X(L) ⇒ Xn (x) = sin ∞ 2 nπ 2 nπx bn e−α ( L ) t sin L nπx L L λn = nπ. 0) = g(x) − u∞ (x). Separate variables: v(x. 0) ⇒ v(x. v(x. t) v(x.3) u(x. t) = u0 + where L (15. .3) we could recognize that nπx ∞ sin are eigenfunctions of the spatial operator: L n=1 − 96 ∂2 ∂x2 (15. .6) L ∞ v(x. t) = u∞ (x) + v(x. ˙ T (t) α2 T (t) = X (x) = −λ2 = const X(x) 2 α2 t (15.

t) = u0 IC: u(x. t) = 0 .22) ux (L.8) above.20) (15.16) Therefore ∞ v(x. 97 0 < x < L. Look for a steady solution: u∞ (x) = 0.along with the homogeneous Dirichlet BC v(0. 0) = g(x).18) v(x.3 ut = α2 uxx BC: u(0.21) (15.11) (15.17) (15. t) = ∂v ∂t ∂2v ∂x2 n=1 ∞ vn (t) sin ˆ nπx L and sin nπx L 2 (15. t>0 (15. 0) = n=1 vn (0) sin ˆ L nπx = g(x) − u∞ (x) L nπx L dx vn (0) = ˆ 2 L 0 {g(x) − u∞ (x)} sin (15. ˆ ˆ (15. t) = n=1 ∞ 2 nπ 2 nπx vn (0)e−α ( L ) t sin ˆ L (15. We therefore assume an eigenfunction expansion of the form: ∞ v(x. L (15. t).14) = = − n=1 ∞ nπx ˙ vn (t) sin ˆ L vn (t) ˆ nπ L 2 n−1 ∞ vt = α2 vxx ⇒ n=1 nπ ˙ vn (t) + α2 ˆ L vn (t) sin ˆ Therefore nπ ˙ vn (t) = −α2 ˆ L 2 vn (t) A simple ODE for vn (t): (15.12) (15.19) which is the same solution as that in (15. t) = 0 = v(L. Example 15.15) ˆ ˆ 2 2 nπ ⇒ vn (t) = vn (0)e−α ( L ) t .13) nπx = 0.

t) ⇒ vx (L. 2.35) π Therefore X (L) = Bλ cos(λL) = 0 ⇒ λk = (2k − 1) (15. t) satisfies vt = α2 vxx v(0. 0) = g(x) − u0 . We now need a solution v(x. t) = 0 u(x. t) ⇒ v(0.28) (15. t) = u0 + v(x. t) = u∞ (x) + v(x. t) v(x. t) = 0 = vx (L. t) = 0 ux (L. 98 (15.30) (15. 0) = g(x) − u0 Thus v(x. ut = α2 uxx ⇒ vt = α2 vxx u(0.37) X(0) = A = 0 . t) = X(x)T (t) to (15.26) (15.25) (15. 0) = g(x) − u0 . 0) = g(x) ⇒ v(x.24) (15. t) = 0 ⇒ 0 = vx (L. or λ = 0 which is the trivial solution.31) (15. .33) +Bλ cos(dx) X(0) = 0 = X (L) ⇒ X(x) = A cos(λx) + B sin(λx) X (x) = −Aλ sin(λx) (15.23) (15. 2 u∞ (0) = B = u0 u∞ (x) = A = 0 (15.36) 2L k = 1.27) (15. t) = u0 ⇒ u0 = u0 + v(0. 0) = u0 + v(x. 0) = g(x) ⇒ v(x. t).32) (15.34) (15. 0) = u0 + v(x.29) (15. . Let u(x. u(x. 3.Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC u∞ (x) = Ax + B Therefore u∞ (x) = u0 . .29): ˙ T (t) x (x) = = −λ2 2 T (t) α X(x) 2 2 ˙ T (t) = −λ2 α2 T (t) ⇒ T (t) = ce−λ α t X + λ X = 0.

38) (15.41) Example 15. t): ∞ u(x.44) u(L. t) = u1 α 2 u∞ − u∞ = 0 u∞ (x) = A cosh x x + B sinh α α u∞ (0) = A = 0 u∞ (L) = B sin h L α = u1 u1 B= sin h (15.46) 99 . 0) = 2 L 0 (2k − 1 πx 2L ⇒ bn = {g(x) − u0 } sin dx. Therefore u∞ (x) = u1 sinh x α sinh L α . t) = u0 + k=1 bk e−α 2 λ2 t k sin (2k − 1) πx . t) = 0 IC: u(x. t) = u0 + v(x. t) = k=1 ∞ bk e−λk α t sin bk sin k=1 L 2 2 (2k − 1) πx 2L = g(x) − u0 (2k − 1) πx 2L (15.39) v(x.4 Heat Equation with some heat loss: ut = α2 uxx − u BC: u(0.40) Returning to u(x. 0) = g(x).45) L α .42) (15.43) (15. 2L (15.Therefore ∞ v(x. Look for the steady state solution u∞ (x): 0 < x < L. (15. t>0 (15. (15.

47) we separate variables v(x. t) u(L. t) = 0 (15. t) = 0 ⇒ u1 = u∞ (L) + v(L.51) 2 Therefore ∞ v(x. t) = u1 + v(L. vt = α2 vxx − v ut = α2 uxx − u ⇒ 0 = u∞ (0) + v(0. t) = X(x)T (t).52) (15.(15. (15. 0) = g(x) u(x. (15. t) ⇒ u(0. t) = u∞ (x) + v(x.50) nπ ⇒ X(0) = 0 ⇒ A = 0 X(L) = B sin(λL) = 0 ⇒ λn = L n = 1. t) = u1 sinh α sin h L α ∞ + n=1 bn e−(1+λn α 2 2 )t sin nπx .Lecture 19 Further Heat Conduction Problems: Inhomogeneous BC Now let u(x. t).54) Therefore x u(x.55) L Remark 15. . t) = n=1 bn e−(1+λn α 2 2 )t sin ∞ πx L bn sin nπx ⇒ bn L nπx L dx. 0) = g(x) ⇒  vt = α2 vxx − v    v(0. .48) (15. X(x) (15. 0) = g(x) − u∞ (x) = n=1 L 2 = L 0 {g(x) − u∞ (x)} sin (15. . t) = u1 ⇒ u∞ (x) + v(x. 100 . . 2. To solve (15.53) v(x.5 Note: The −u term in the PDE is responsible for the e−t factor in the solution. t) = 0   v(x. Therefore ˙ T (t) α2 X 1 = −1⇒ 2 T (t) X α Therefore 2 2 ˙ T (t) = −(λ2 α2 + 1)T (t) ⇒ T (t) = ce−(1+λ α )t ˙ T (t) +1 T (t) = X (x) = −λ2 .47)  v(L.49) X + λ X = 0 ⇒ X(x) = A cos λx + B sin λx (15. 0) = g(x) − u∞ (x).

2) (16. t) = B • Try for a steady solution: u∞ (x) = 0.4) (16.1) (16. t) = A ux (L.5) 101 . This means that if we are pumping and removing heat from the rod at different rates then the temperature does not reach a steady state.Chapter 16 Lecture 20 .3) BC: ux (0. • Instead of subtracting off a steady solution we subtract a particular solution which depends on x and t of the form: w(x.Inhomogeneous Derivative BC Example 16. 0) = g(x). t) = ax2 + bx + ct wt = c = α wxx = 2α a ⇒ c = 2α a.1 Inhomogeneous Derivative BC: ut = α2 uxx IC: u(x. 2 2 2 (16. u∞ (x) = αx + β. t>0 (16. 0 < x < L. ux = α but then we cannot match both BC unless A = B = α.

0) = g(x) − w(x. ut ux (0. 0) (B − A) 2 x + Ax + α2 2L B−A L t.8) Equations (16. t) = 0 v(x.12) 102 . 0) + v(x. t) + vx (L. t) = (B − A) 2 x + Ax + α2 2L ∞ B−A L t+ a0 2 (16. t) = b = A.11) + n=1 an cos nπx −α2 ( nπx )t L e L where L 2 an = L 0 g(x) − (B − A) 2 x + Ax 2L cos nπx L dx. Therefore u(x. t) = ax2 + bx + 2α2 at solves the heat equation. t) = 0 (16. t). t) u(x. t) = w(x. t) + vx (0. t) = A + vx (0.10) .9) vt = α2 vxx vx (0. Now we determine the constants a and b so that w(x. t) g(x) = w(x.Lecture 20 . t) = 2aL + A = B. t) satisfies the inhomogeneous BC: wx = 2ax + b : wx (0. 0) ⇒ ⇒ ⇒ ⇒ (16. 0) = = = = wt + vt = α2 (wxx + vxx ) A = wx (0.10) represent the homogeneous Neumann BVP seen previously. t) ux (L. t) = Now let u(x.6) Therefore a = (B − A)/2L. t) = B + vx (L. Therefore w(x. (16.Inhomogeneous Derivative BC Then w(x. wx (L. t) B = wx (L.7) (16. (16. (16. t) + v(x. vx (L.

t) = B Steady state problem ut = 0: 0 = α 2 u∞ + x u∞ (0) = 0 u∞ (L) = B x2 x3 + a u∞ = − 2 + ax + b 2α2 6α L3 B L2 u∞ (0) = b = 0 u∞ (L) = − 2 + aL = B ⇒ a = + 2 6α L 6α u∞ = − x α2 u∞ = − Therefore u∞ (x) = − x3 + 6α2 B L2 + 2 L 6α x=x B 1 + 2 (L2 − x2 ) . t) = 0 IC: u(x. ut = α2 uxx + x u(0. 0) = g(x) ⇒ ⇒ ⇒ ⇒ (u∞ +v)t = α2 (u ∞ +v)xx + x ↓ ↓ u∞ (0) + v(0. L 6α (16. 0) = g(x) − u∞ (x). 0) = g(x).20) 103 . 0<x<L (16.19) v(L. t) = u∞ (x) + v(x.15) u(L.2 A bar with an external heat source s(x) = x. t). Separation of variables yields: ∞ v(x. 0) = g(x) ⇒ ⇒ ⇒ ⇒ vt = α2 vxx v(0. t) = 0 u(L.16) (16. t) = 0 v(x. ut = α2 uxx + x BC: u(0. t) = 0 u∞ (L) + v(L. t) = B u∞ (x) + v(x.17) Let u(x.14) (16. t) = n=1 bn e − ( L ) nπ 2 2 α t sin nπx L (16. t) = B u(x.Example 16.13) (16.18) (16. t) = 0 (16.

Lecture 20 .22) lim u(x. L 6α (16. (16. t) = x 1 B + 2 (L2 − x2 ) + L 6α ↑ steady Note: x→∞ ∞ n=1 2 nπ 2 nπx bn e−α ( L ) t sin L ↑ transient (16.21) Therefore u(x.23) 104 .Inhomogeneous Derivative BC where L 2 bn = L 0 {g(x) − u∞ (x)} sin nπx L dx. t) = x B 1 + 2 (L2 − x2 ) .

t) = 0.Chapter 17 Lecture 21 Distributed. 0) = 0. Time Dependent Heat Sources eigenfunction expansions Example 17. t) = 0 ⇒ v(0. 0) = w(x) + v(x. t). Now let u(x. t) = L ⇒ v(L.1 A Bar with a Time-Varying External Heat Source: ut = α2 uxx + e−t sin BC: u(0. 0) ⇒ v(x. ut = (w +v)t = α2 (w +v)xx + e−t sin 2πx L u(0.7) x = u(x. u(L. 0) = x.3) Consider the function w(x) = x which satisfies the BC as well as the homogeneous version of the PDE. t) = 0 ⇒ vt = α2 vxx + e−t sin u(L.1) (17. t) = 0 2πx L (17. IC: u(x.8) 105 . t) = w(L) +v(L.4) (17. t) = L 2πx L 0 < x < L. (17. 0) = x + v(x.5) (17. t) = w(0) +v(0.2) (17. t > 0 (17.6) (17. t) = w(x) + v(x.

t) = n=1 vn (t) sin ˆ ∂2v = ∂x2 ∞ nπx .15) n=2 n=2 (17. Time Dependent Heat Sources .eigenfunction expansions ∞ Now assume that v(x. L vn (t) − ˆ nπ L 2 ∂v = ∂t ∞ n=1 dˆn v nπx (t) sin dt L sin n=1 nπx L .9) Therefore ∞ vt − α2 vxx = n=1 dˆn v nπ + α2 dt L 2 vn − e−t δ2n sin ˆ nπx = 0.(17.Lecture 21 Distributed. t) = x + 2 2π e−t − e−α ( L ) t α2 2π 2 L −1 .13) (17.10) L Therefore nπ 2 dˆn v + α2 vn = e−t δ2n ˆ dt L i h 2 d α2 ( nπ )2 t α2 ( nπ ) −1 t L L vn = e ˆ e δ2n .14) v(x. t) = x + v(x. t) = 1 2 α2 2π L 2 2π e−t − e−α ( L ) 2 1 α2 ( 2π ) −1 L 2 t −1 sin 2 2πx L sin u(x. 106 . dt Therefore e 2 −α2 nπ t L (17.16) 2πx L cn = v(x. (17.11) (17.12) ( ) v = ˆn e i h 2 α2 ( nπ ) −1 t L vn (t) = ˆ δ2n + cn cn nπ 2 −1 L 2 nπ 2 e−t δ2n + e−α ( L ) t cn 2 α2 nπ − 1 L α2 δ2n 2 nπ 2 α L arbitrary (17. 0) = 0 ⇒ vn (0) = 0 = ˆ −1 + cn ⇒ 0 − (17.

17)-(17.20) where 2 sn (t) = ˆ L s(x.25) 107 . 0) = f (x. t) to a solution of (17.17) (17. with s(x.2 A bar with a general external heat source s(x. t) BC: u(0.Example 17. t) = ∞ wn (t) sin ˆ n=1 (17. t) = B IC: u(x.19) We look for a particular solution: w(x.21) If we assume w(x. t).23) nπx . (17.e.18) (17. (17. Note that the sine functions are the eigenfunctions that correspond to the homogeneous form of the BC in (17. L (17.18). t) by expanding s(x. t) = n=1 ∞ sn (t) sin ˆ L nπx L nπx L nπx L (17. t) we obtain: ∞ wn + α2 ˆ n=1 nπ L 2 wn − sn (t) sin ˆ ˆ nπx = 0.18) without the source (i.24) wn ˆ n=1 nπ L sin Therefore substituting these expansions into wt = α2 wxx + s(x. t) sin 0 dx. L (17. t) = 0) we will not affect the BC. t) ut = α2 uxx + s(x. t) as a Sine Series.22) then ∞ wt = wxx = − wn (t) sin ˆ n=1 ∞ nπx L 2 (17. 1. Thus if we add w(x. Eigenfunction Expansion: Let s(x. t) = A u(L.

26) „ nπ «2 This is a linear 1st order ODE with integrating factor e Therefore t L t .eigenfunction expansions Therefore wn (t) = −α2 ˆ nπ L 2 wn (t) + sn (t). t) = v(L. L (17.28) 2. t) since B = u(0. t) = v(0. 0) ⇒ v(x. t) = w(0. t) + v(0.29) (17. t) = 0 w(L.31) w(0. t) = 0. Time Dependent Heat Sources . A = u(0. t) satisfies: BC: IC:  vt = α2 vxx  v(0. t) ut = w t +vt = α (w xx +vxx ) + s (x. 0) = f (x) − w(x. t) = A v(L. t) + v(x. ˆ ˆ α2 (17. Now that we have a particular solution we exploit the fact that the Problem (17. 0) + v(x.33) Now the boundary value Problem (2) was solved on pg. t) + v(L. t) = B−A L ∞ x+A+ n=1 e 2 −α2 nπ t L (   ) b +  n t e ( 0 2 α2 nπ τ L ) s (τ ) dx ˆn    sin 108 nπx L (17. f (x) = u(x.34) . 0) (17.  v(x. Therefore ∞   0 t  e 2 −α2 nπ (t−τ ) L w(x.18) is linear and use superposition. Since we are only looking for a particular solution we choose cn ≡ 0. Let u(x. t) = n=1 ( ) sn (τ ) dτ  sin ˆ nπx . 0) (17. t) since 2 2 (17.30) (17.Lecture 21 Distributed. Therefore u(x. t) = B . 0) = w(x.27) where the cn are arbitrary constants. t) = w(L.32) thus v(x.17)-(17. 0) = f (x) − w(x. wn (t) = 0 2 nπ 2 nπ e−α ( L ) (t − τ )ˆn (τ ) dτ + cn e−α ( L ) s 2 2 t (17. 76 of the notes. t) ⇒ vt = α vxx . t) = w(x.

35) 109 .where L 2 bn = L 0 f (x) − w(x. (17. 0) − (B − A) x +A L sin nπx L dx.

eigenfunction expansions 110 .Lecture 21 Distributed. Time Dependent Heat Sources .

1) (18. t). 2L Now let u(x. 0) = g(x) 0 < x < L. ut = α2 uxx IC: u(x.Chapter 18 Lecture 22 More Eigenfunction Expansions Time Dependent Boundary Conditions Example 18. t) xx = α2 · 2a + α2 vxx .1 Let us revisit the problem with inhomogeneous derivative BC . t) t = α2 uxx = α2 h(x) + v(x. 111 . Substitute into the PDE: ut = h(x) +v(x. hx (x) = 2ax + b hx (0) = b = A hx (L) = 2aL + A = B ⇒ a = (B − A)/2L B−A h(x) = x2 + Ax.3) BC: ux (0.but we will now use Eigenfunction Expansions. t) = h(x) + v(x. t) = B First look for a function of the form h(x) = ax2 + bx that satisfies the inhomogeneous BC: h(x) = ax2 + bx. t>0 (18.2) (18. t) = A ux (L.

0) ⇒ v(x. Because of the homogeneous Neumann BC we assume an expansion of the form ∞ v(x.7) We now use an Eigenfunction Expansion to solve the BVP (18.7).Lecture 22 More Eigenfunction Expansions . L We also expand the inhomogeneous term in (1. t) = B + Vx (L. 0) = g(x) − h(x). an = 0 n ≥ 1. L . ˆ ˆ ˆ L Therefore 4aα2 t + c0 v(x.5) (18.4)-(18. t) ⇒ vx (L.6) g(x) = u(x.4) (18. (18. vxx = L ∞ vx = n=1 vn (t) − ˆ vn (t) − ˆ n=1 nπ L 2 cos nπx . 0) = h(x) + v(x. t) = 0 (18. t) = + 2 112 ∞ n=1 2 nπ 2 nπx vn (0)e−α ( L ) t cos ˆ . Therefore ∞ ˙ 0 = vt − α2 vxx − 2aα2 = v0 (t)/2 − 2aα2 + ˆ n=1 nπ ˙ vn + α2 ˆ L 2 vn cos ˆ nπx . t) = v0 (t)/2 + ˆ n=1 ∞ vn (t) cos ˆ nπx L ˙ vt = v0 (t)/2 + ˆ n=1 ∞ nπx ˙ vn (t) cos ˆ L nπ L sin nπx .4) in terms of the Eigenfunctions: ∞ 2aα2 = a0 /2 + n=1 an cos nπx L a0 = 4aα2 .Time Dependent Boundary Conditions Therefore vt = α2 vxx + 2aα2 A = ux (0. t) = A + Vx (0. t) = hx (L) + Vx (L. t) = hx (0) + vx (0. t) ⇒ vx (0. t) = 0 B = ux (L. L Therefore ˙ v0 (t) = 4aα2 ⇒ v0 (t) = 4aα2 t + c0 ˆ ˆ 2 nπ 2 nπ 2 ˙ vn (t) = −α2 ˆ vn ⇒ vn (t) = vn (0)e−α ( L ) t .

0) = f (x) = w(x. L Now let u(x. t) = φ1 (t) + v(L. 0) ⇒ v(x. x2 + Ax + 2aα2 t + c0 + 2 ∞ n=1 2 nπ 2 nπx vn (0)e−α ( L ) t cos ˆ L which is identical to the solution obtained in Example 16. Now v(x. t) = φ0 (t). 113 . 0) = f (x) − w(x. t) + v(0. (18. t) ⇒ v(L. t) = 0 IC: u(x. t) = φ0 (t) = w(0. 0) = c0 + 2 ∞ vn (0) cos ˆ n=1 nπx L c0 = 2 L 0 1 g(x) − B−A 2L B−A 2L x2 + Ax dx vn (0) = ˆ Thus u(x. t) = B−A 2L 2 L 0 g(x) − x2 + Ax cos nπx L dx.2 Time Dependent Boundary Conditions . 0) = f (x) − w(x. t) can be found using an eigenfunction expansion.10) Thus we need to solve the following BVP for v(x. w(L. Then wt + vt = α2 (wxx +vxx ) x ˙ ˙ ˙ wt = φ0 + (φ1 − φ0 ) L BC: u(0. t) = φ0 (t) + v(0. t) = φ1 (t) = w(L. IC: u(x. t) = 0 vt = α2 vxx − wt u(L.general case: ut = α2 uxx . φ1 (t) − φ0 (t) ⇒ w(0. t) + v(L. t) = φ1 (t).1.9) BC: u(0. 0) = f (x). t) = φ0 (t) + x 0 < x < L.11) IC: v(x. 0) + v(x. Let w(x.8) (18. t) ⇒ v(0. t). t) = φ1 (t). 0). (18.g(x) − h(x) = g(x) − 1 B−A 2L x2 + Ax = v(x. Example 18. t) = 0 v(L. t): vt = α2 vxx − wt BC: v(0. t) = φ0 (t) u(L. t) + v(x. t) = 0 (18. 0). t) = w(x.

t) = 0. L 114 . In this case w(x. 0) = 0. t) = 0 Let u(x. Let s(x. t) = At + x x (0 − At) = At 1 − .17) .16) ∞ vt = n=1 nπx ˙ vn (t) sin ˆ L vxx = − n=1 vn (t) ˆ nπx? L 2 sin nπx . γ) = −A 1 − L x . t) = 0 = v(L. φ1 (t) = 0 f (x) = 0. L Therefore ∞ 0 = vt − α vxx − s(x.Time Dependent Boundary Conditions Specific case: Let φ0 (t) = At. t) = w(x. Then L x L (18. t) = n=1 2 nπ ˙ vn (t) + α2 ˆ L 2 vn + ˆ 2A nπ sin nπx (18. t) v(x. t) = At u(L.15) sn = ˆ 2 L 0 A x nπx − 1 sin L L 2A = − .12) ut = α2 uxx IC: u(x. t) = At 1 − vt = α2 vxx − A 1 − v(0. t) = n=1 ∞ vn (t) sin ˆ nπx L (18. t) where w(x.Lecture 22 More Eigenfunction Expansions .14) x = L ∞ sn (t) sin ˆ n=1 nπx L dx (18. L L 0<x<L (18. t) + v(x. nπ Now let ∞ v(x.13) (18. BC: u(0.

Therefore nπ 2 2A ˙ vn (t) + α2 ˆ vn (t) = − ˆ L nπ 2 nπ 2 t 2A α2 ( nπ )2 t L e+α ( L ) vn (t) ˆ = − e nπ 2 nπ 2 2AL2 α2 ( nπ )2 t L eα ( L ) t vn (t) = − 2 ˆ +B e α (nπ)3 2 nπ 2 2AL2 vn (t) = − 2 ˆ + Be−α ( L ) t α (nπ)3 2AL2 0 = vn (0) = − 2 ˆ + B. α (nπ)3 Therefore vn (t) = ˆ Therefore x 2AL2 u(x.19) (18.20) (18.23) 115 .22) (18.21) (18. α2 (nπ)3 (18. (18.18) (18.24) n3 L 2 2 nπ 2 2AL2 e−α ( L ) t − 1 . t) = At 1 − + 3 2 L π α ∞ n=1 2 nπ (e−α ( L ) t − 1) nπx sin .

Lecture 22 More Eigenfunction Expansions .Time Dependent Boundary Conditions 116 .

t) = 0.2) → expect 2 boundary conditions u(0. 0) = f (x) ∂u ∂t (x.1) = g(x) (19.1 Guitar String Note: ∂ ∂ +c ∂t ∂x Let w = ∂ ∂ −c ∂t ∂x u then (19.Chapter 19 Lecture 23 . 0) (19. (19. t) = 0 u(L. 1st order wave eq. t) = 0.1D Wave Equation ∂2u ∂2u = c2 2 ∂t2 ∂x ∂2u ∂t2 ∂2u ∂x2 → expect 2 initial conditions u(x. ∂t ∂x . 19.3) ∂w ∂w +c = 0.4) 117 ∂ ∂ −c ∂t ∂x u(x.

(19.1) 1.9) G utt = c2 G uxx = G . This property is due to the linearity of utt = c2 uxx (19.7) (19. (19. ∂t2 ∂x Therefore u(x. t) = F (x − ct) + G(x + ct) Notes: solves (19.5) (19. Every solution for (19. Is the sum of two solutions also a solution? Claim u(x. 118 .1) if u1 and u2 are solutions.1). t) = F (x − ct) is also a solution to (19. ∂2 ∂ 2 u1 ∂ 2 u2 (α1 u1 + α2 u2 ) = α1 2 + α2 2 ∂t2 ∂t ∂t ∂ 2 u2 ∂ 2 u1 + α2 c2 = α1 c2 ∂x2 ∂x2 Thus ∂2 ∂2 (α1 u1 + α2 u2 ) = c2 2 (α1 u1 + α2 u2 ). t) + α2 u2 (x.1) ut = cG ux = Therefore utt − c2 uxx = c2 G − c2 G = 0. t) = α1 u1 (x.1). t) is a solution of (19. (19.1). ∞) is of this form.10) since u1 and u2 solve (19.Lecture 23 .8) Similarly u2 (x.1) on (−∞. 2.6) Claim: u1 (x.1D Wave Equation Thus ∂ ∂ +c ∂t ∂x ∂ ∂ −c ∂t ∂x → right moving wave → left moving wave. t) = G(x + ct) is a solution to (19. (19.

19.1. GUITAR STRING 3. D’Alembert’s Solution Change of variables: Let r = x + ct 1 x = 2 (r + s) ∂ ∂r ∂ ∂s Therefore −4c2 ∂2u = ∂r∂s ∂ ∂ +c ∂t ∂x ∂ ∂ −c ∂t ∂x u= ∂2u ∂2u − c2 2 = 0. (19.14) ∂t2 ∂x s = x − ct . 1 t = 2c (r − s) (19.11)

= =

∂ ∂ ∂x ∂ ∂t 1 ∂ + = +c ∂x ∂r ∂t ∂r 2c ∂t ∂x ∂ ∂ ∂x ∂ ∂t 1 ∂ + =− −c ∂x ∂s ∂t ∂s 2c ∂t ∂x

(19.12) (19.13)

Therefore ∂2u (r, s) = 0 (19.15) ∂r∂s ∂u ¯ ⇒ (r, s) = φ1 (s) (19.16) ∂s ¯ ⇒ u(r, s) = φ1 (s) ds + φ2 (r) = φ1 (s) + φ2 (r). (19.17) Say we have the IC: u(x, 0) = u0 (x) displacement ∂u (x, 0) = v0 (x) velocity ∂t u(x, t) = F (x − ct) + G(x + ct) u(x, 0) = F (x) + G(x) = u0 (x) ∂u (x, 0) = −cF (x) + cG (x) = v0 (x) ∂t
x

(19.18) (19.19) (19.20) (19.21) (19.22) (19.23)

−cF (x) + cG(x) =
0

v0 (ξ) dξ + A

 1 1 −c c F G =

u0
x

  (19.24)

v0 (ξ) dξ + A
0

119

Lecture 23 - 1D Wave Equation    x   1  v0 (ξ) dξ + A cu0 −  2c  0  x    1 v0 (ξ) dξ + A + cu0  2c 
0

F

=

(19.25)

G =

(19.26)

Therefore u(x, t) = 1 1 [u0 (x − ct) + u0 (x + ct)] + 2 2c
x+ct

v0 (ξ) dξ (19.27)
x−ct

D’Alembert’s Solution to the wave equation on (−∞, ∞).

120

Chapter 20

Lecture 24 - Space-Time Interpretation of D’Alembert’s Solution
x+ct

1 1 u(x, t) = [u0 (x − ct) + u0 (x + ct)] + 2 2c
x−ct

v0 (s)ds

(20.1)

20.1

Characteristics
x − ct = x0 and x + ct = x0 (20.2)

In the x − t plane the lines

are called characteristics. x − ct = x0 ⇒ t = x + ct = x0 1 1 x − x0 c c

(20.3)

1 1 ⇒ t = − x + x0 c c

1 1 1 x + ct = x0 thus t = − x0 + x0 − c c c 121

t0 ) bound the domain of dependence. 20.3 Domain of Dependence The lines x = x0 − ct0 and x = x0 + ct0 that pass through the point (x0 .Lecture 24 .Space-Time Interpretation of D’Alembert’s Solution 1 1 1 x − ct = x0 thus t = x − x0 c c c 20. 122 .2 Region of Influence The lines x + ct = x0 and x − ct = x0 bound the region of influence.

t) = Let c = 1. DOMAIN OF DEPENDENCE Example 20.20. t = 1: 2 xr − 1 =1 2 1 [u0 (x − ct) + u0 (x + ct)] 2 (20.8) 123 .1 Special Case: u(x.3.7) (20.5) ⇒ xr = 3 2 xR + 1 =1 2 xR = 1 2 (20.4) u(x.6) 1 1 x − = −1 ⇒ x = − 2 2 t = 1: 1 3 xL + = −1 xL = − 2 2 xr − 1 = 1 ⇒ xr = 2 xR + 1 = 1 ⇒ xR = 1 x − 1 = −1 ⇒ x = 0 x + 1 = −1 ⇒ xL = −2 t = 2: xr − 2 = 1 ⇒ xr = 3 xR + 2 = 1 ⇒ xR = −1 x − 2 = −1 ⇒ x = 1 xL + 2 = −1 ⇒ xL = −3 (20. 0) = 1 |x| < 1 0 |x| > 1 (20.

Lecture 24 .Space-Time Interpretation of D’Alembert’s Solution 124 .

2) (21. .1) (21. T0 .5) 2X = 0 X +λ X(x) = A cos(λx) + B sin λx ⇒ X(0) = 0 = X(L) X(0) = A = 0 X(L) = B sin λL = 0 nπ λn = n = 1. 0) = g(x) t>0 (21. t) = X(x)T (t) ¨ X (x) T (t) = = −λ2 c2 T (t) X(x) (21. 0) = f (x). .3) u(L. Xn = sin L 125 . t) = 0 For a guitar string c = Separate Variables u(x. L nπx .Chapter 21 Lecture 25 Solution by separation of variables Example 21. t) = 0. IC: u(x. 2. . ρ0 0 < x < L. ut (x.1 utt = c2 uxx BC: u(0.4) ¨ T (t) + λ2 c2 T (t) = 0 ⇒ T (t) = c1 cos(λct) + c2 sin(λct) (21.

10) L 21. t) = n=1 An cos nπct L + Bn sin nπct L sin nπx . (21. 0) = n=1 Bn nπx nπc sin = g(x) ⇒ Bn L L = g(x) sin 0 Therefore ∞ u(x. t) = nπx nπc + Bn L L cos nπct L nπc L sin 2 L L nπx L nπx L (21.Lecture 25 Solution by separation of variables Therefore ∞ u(x. t) = n=1 ∞ An cos nπct L sin nπx + Bn sin L 2 L L nπct L f (x) sin sin nπx L nπx L (21. Period and Frequency: cos nπc (t + T ) = cos L 2L c nπct L provided nπcT = 2π L (21. (21. Modes of Vibration: Standing waves of wavelength λn = 126 .9) ∞ ut (x.6) u(x. 2L . fn = n are the natural frequencies of vibration.7) 0 ut (x. n 2.8) dx .1 Notes 1.11) thus Tn = 1 c =n Tn 2L 1 is the period (seconds per cycle) of mode n. 0) = n=1 ∞ An sin −An n=1 nπx = f (x) ⇒ An = L nπc sin L nπct L sin (21.

14) Now ∞ An cos n=1 nπct L sin nπx L = 1 2 ∞ An sin n=1 nπ (x + ct) L (21. L (21.15) + sin = nπ (x − ct) L 1 [f0 (x + ct) + f0 (x − ct)] (21.19) L (21. 2 L L (21.2.18) Then Bn = x bn ∞ g(s) ds = 0 x n=1 ∞ bn − cos nπs L nπ L ∞ x 0 ∞ = n=1 cBn 1 − cos nπx .21. L L Let L 2 bn = L 0 nπx g(x) sin L L nπc ∞ dx ⇒ g(x) = n=1 bn sin nπx . ∞ Bn sin n=1 nπct L sin nπx 1 = L 2 ∞ Bn cos n=1 nπ nπ (x − ct) − cos (x + ct) (21. to interpret the solution (21.13) (21.17) . NOW WE CAN USE THE TRIGONOMETRIC IDENTITIES 21.20) 1 c 0 g(s) ds = n=1 Bn = − n=1 Bn cos 127 .16) 2 where f0 is the odd periodic extension of f .2 Now we can use the trigonometric identities cos(A ± B) = cos A cos B sin A sin B (21. L nπx (21.10).12) sin(A ± B) = sin A cos B ± cos A sin B. cos nπct nπx sin L L nπx nπct sin sin L L = = 1 nπ nπ sin (x + ct) + sin (x − ct) 2 L L 1 nπ nπ cos (x − ct) − cos (x + ct) .

2. whereas the solutions to the Heat Equation typically decay exponentially with time. 3.e.Lecture 25 Solution by separation of variables Therefore ∞ Bn sin n=1 nπct L  nπx 1  sin = L 2c  0 x+ct g0 (s) ds + 0 g0 (s) ds    . and the initial velocity function is given by the odd periodic extension of g0 . f (x) 0<x<L −f (−x) −L < x < 0 g(x) 0<x<L −g(−x) −L < x < 0 and and f0 (x + 2L) = f0 (x0 ) (21.27) of the notes) in which the initial displacement function is given by the odd periodic extension f0 of the initial displacement of the string.21) x−ct Therefore x+ct 1 1 u(x. 128 .24) . Thus the solutions L to the Wave Equation persist with time.23) g0 (x + 2L) = g0 (x) (21.22) where f0 and g0 are the odd periodic extensions of f and g on [0. Equation (21. L] i. Information is carried along the characteristic curves x + ct = const x − ct = const. t) = [f0 (x + ct) + f0 (x − ct)] + 2 2c x−ct g0 (s) ds (21. (21. f0 (x) = g0 (x) = Notes: 1.22) above shows that the Wave Equation Solution for a string tied down at its ends is given by D’Alembert’s Solution (see (19. Observe that the time dependence of the solution involves sin and cos nπct L nπct which do not decay with time.

∂t ∂t 2D: ∂2u ∂2u + 2 = 0.Chapter 22 Lecture 26 .1) 3D: ∂2u ∂2u ∂2u + 2 + 2 = 0. ∂ 2 u 1 ∂u 1 ∂2u + + 2 2 = 0. ∂x2 ∂y ∂z ∆u = (22. ∂r2 r ∂r r ∂θ 129 In Polar Coordinates: ∆u = . ∂x2 ∂y ∆u = (22.2) • No initial conditions required. • Only boundary conditions.Laplace’s Equation Steady State Solutions of the Heat or Wave Equations that do not vary with ∂u ∂2u time so that =0= 2.

y) = g2 (y) Idea 130 (22.6) (22.1 Summary In this course we have studied the solution of the second order linear PDE.2 Laplace’s Equation (1) 2D Steady-State Heat Conduction and (2) Static Deflection of a Membrane. ∂u = α2 ∆u Heat: Parabolic T = αX 2 ∂t ∂2u ∂t2 ∆u = c2 ∆u = 0 Wave: Hyperbolic T 2 − c2 X 2 = A(22.: Elliptic Important: 1. y) ∈ D BC: u prescribed on ∂D. t) inside a domain D. These equations are all linear so that a linear combination of solutions is again a solution. 22.3) X 2 + Y 2 = A. y.7) . Laplace’s Eq. circular.5) (22. pizza slices.Laplace’s Equation 22. 22.Lecture 26 . ut = α2 (uxx + uyy ) −→ u(x. • Steady-State Solution satisfies: ∆u = uxx + uyy = 0 (x. y) = g1 (y) ∆u = uxx + uyy = 0 u(a. (22.3 Rectangular Domains u(0. 2. These equations are second order because they have at most 2nd partial derivatives.4) • We consider domains D that are rectangular.

7).10) (22. 131 . • Since the equation is linear we can break the problem into simpler problems which do have sufficient homogeneous BC and use superposition to obtain the solution to (22. y) = u(x.4. b). Pictorially: 22.9) (22.22.11) (22. 0) = f1 (x). . Y (b) = 0 • Because sin and cos have an ∞ # of real roots the choice −λ2 is good for BC’s for Problems (A) and (C). .4 (1A) Solution to Problem (1A) by Separation of Variables uxx + uyy = 0 u(0. y) = 0 = u(a. (22.8) u(x. X (x)Y (y) + X(x)Y (y) = 0 ÷XY: X (x) = −Y (y) = const = ±λ2 X(x) Y (y) BC −λ2 : (22. y) = X(x)Y (y). SOLUTION TO PROBLEM (1A) BY SEPARATION OF VARIABLES • We want to use separation of variables so we need homogeneous boundary conditions. Let u(x.12) X + λ2 X = 0 X = A cos λx + B sin λx X(0) = 0 = X(a) Y − λ2 Y = 0 Y = C cosh λx + D sinh λx Y (0) = .

20) Note: We could save ourselves the time by building the BC y(b) = 0 directly into the solution by letting ¯ Yn (y) = D sinh λn (y − b) directly. . .17) Y (y) = −D tan h(λb) cosh(λy) + D sinh(λy) sinh(λy) cosh(λb) − cosh(λy) sinh(λb) = D cosh(λb) D ¯ sinh λ(y − b) = D sinh λ(y − b). .16) u(x. (22. . X(a) = . y) = n=1 ∞ Bn sin nπx nπ sinh (y − b) (22.15) Xn (x) = sin nπx a (22. In order to match the BC u(x. . . nπx nπ sinh (y − b) n = 1. 2. 0) = f1 (x) we need to superimpose all these solutions. . (22.23) a f1 (x) = u(x. 2.19) (22. . = cosh(λb) (22. a a satisfy all the homogeneous BC of Problem (1A).14) λn = nπ n = 1.Lecture 26 . Now the functions: un (x. y) = sin ∞ (22.18) (22. 0) = n=1 −Bn sinh bn 132 .Laplace’s Equation +λ2 : X − λ2 X = 0 Y + λ2 Y = 0 X = A cosh(λx) + B sinh(λx) Y = C cos(λy) + D sin(λy) X(0) = . Back to Solving (1A): X(0) = 0 ⇒ A = 0 X(a) = B sin(λa) = 0 ⇒ (22. .22) a a nπb a sin nπx (22.13) . Y (0) = 0 = Y (b) • Again because sin and cos have an ∞ # of real roots the choice +λ2 is good for BC’s for Problems (B) and (D). b) = X(x)Y (b) = 0 ⇒ Y (b) = 0 Y (b) = C cosh(λb) + D sinh(λb) = 0 ⇒ c = −D tan h(λa) (22. .21) u(x. a .

a a a where Bn 2 = − a sinh nπb a f1 (x) sin 0 nπx a dx (22.4. y) = 1 a ∞ 2 1 + (−1)n+1 = −Bn sinh anπ − n=1 nπx 2 [1 + (−1)n+1 ] nπ sin sinh (y − b) . (22. SOLUTION TO PROBLEM (1A) BY SEPARATION OF VARIABLES where a −Bn sinh nπb a 2 = bn = a 0 f1 (x) sin nπx a dx.24) Therefore ∞ u(x.22. a nπb a .26) bn = Therefore u(x. (22.27) nπb nπ sinh a a a 133 .25) ∞ Specific Example Let fL (x) = 1 = n=1 bn sin nπx . (22. y) = n=1 Bn sinh nπ nπx (y − b) sin .

Lecture 26 .Laplace’s Equation 134 .

1 Solution to Problem (1B) by Separation of Variables ∆u = uxx + uyy = 0 0 = u(x.3) (23. y) = X(x)Y (y) X (x) Y (y) = − = ±λ2 . y) = g2 (y) (23. X(x) Y (y) u(b. y).Chapter 23 Lecture 27 .5) 135 . So we choose const = +λ2 X − λ2 X = 0 Y + λ2 Y = 0 X = c1 cosh λx + c2 sinh λx Y = A cos(λx) + B sin(λx) (23.1) (23.4) Since we have homogeneous BC at y = 0 and y = b we want the function Y (y) to behave like sines and cosines.More Rectangular Domains and semi-infinite strip problems 23. b) = u(0. 0) = u(x. Let u(x.2) (23.

More Rectangular Domains and semi-infinite strip problems (23. y) = sin sinh satisfy the homogeneous BC. nπx Therefore Xn (x) = c2 sinh . . .6) nπ u(x.7) b nπy Yn = sin b u(0. 23. y) = cn sinh b b b sin h n=1 ∞ b nπa b g2 (y) sin 0 nπy b dy.Lecture 27 . 2.2 Rectangular domains with mixed BC Example 23.13) .1 Insulating BC along two sides and specified temperatures on the others: ∆u = uxx + uyy = 0 0 = ux (0.11) (23. cn = u(x. (23.10) ( 23.8) where b nπa 2 cn sinh = b b 0 g2 (y) sin nπy b dy. b b ∞ nπy nπx sin . y) = X(0)Y (y) = 0 ⇒ X(0) = c1 = 0. . y) = cn sin h b b n=1 Now to satisfy the inhomogeneous BC ∞ u(x. y) = n=1 cn sin h bn nπy nπa sin b b (23. Therefore u(x. y) = ux (a. b nπy nπx Therefore un (x. 0) = X(x)Y (0) = 0 ⇒ Y (0) = 0 ⇒ Y (0) = A = 0 g2 (y) = u(a. b) = X(x)Y (b) = 0 ⇒ Y (b) = 0 ⇒ Y = B sin(λb) = 0. 0) u(x. λn = ( n = 1.9) Summarizing: nπy 2 nπx sin . 23.12) (23. b) = f (x). y) = u(x. 136 (23.

a a (23.2. 1.20) (23. .22) 0 a c0 = 1 ab 0 f (x) dx cn = ∞ 2 a sinh( n+b ) a f (x) cos 0 nπx a dx (23. y) = y · 1 satisfies the homogeneous BC.15) X (0) = Bλ = 0 ⇒ B = 0 X (a) = −Aλ sin(λa) = 0 Therefore λn = (nπ/a) n = 0. b) = (2c0 b) + 2 a cn sinh n=1 cos 2 a a nπx = f (x) a f (x) cos nπx a (2c0 b) = 2 a 0 f (x) dx cn sinh a = dx (23. 2. y) = c0 y + n=1 cn sinh nπy nπx cos .18) (23.17) u(x. λn = 0: Y − λ2 Y = 0 and Y (0) = 0 ⇒ Yn (y) = A sinh Thus nπy nπx un (x.23. y) = c0 y + n=1 cn sinh ∞ nπy nπx cos a a nπb a nπb a (23. (23.16) n = 0. 2. X(x) = A cos λx + B sin λx X (x) = −Aλ sin(λx) + Bλ cos(λx) (23.23) u(x. λ0 = 0: In this case the ODE for Y0 is: Y0 = 0 ⇒ Y (y) = c1 y + c2 (23. are eigenfunctions and eigenvalues.24) 137 . RECTANGULAR DOMAINS WITH MIXED BC Let u(x. . X Y =− = ±λ2 .14) X Y Since we have homogeneous BC on X (0) = 0 = X (a) choose −λ2 . y) = X(x)Y (y). 1. y) = cos sinh a a satisfy homogeneous BC.21) u(x. Therefore ∞ Xn (x) = cos nπy a nπy a (23. X + λ2 X = 0 X (0) = 0 = X (a).19) Y0 (0) = c2 = 0 ⇒ Y0 (y) = y and u0 (x. . (23.

X + λ2 X = 0 X(0) = 0 = X(a) λn Xn = nπ/a n = 1.31) ∞ f (x) = u(x. Let u(x.More Rectangular Domains and semi-infinite strip problems 23.30) satisfy the homogeneous BC and the BC at ∞. .28) X(x) Y (y) 1.3 Semi-infinite strip problems Example 23. Therefore un (x. y) u(x. 0) = f (x) 0 < x < a. y) → 0 as y → ∞. y) = n=1 cn e−( nπ a )y sin nπx . a a (23. . t) = X(x)T (t) and plug into (1a?): Y (y) X (x) =− = −λ2 since we have homogeneous BC on X.26) (23. (23.Lecture 27 . 0) = n=1 nπx 2 cn sin ⇒ cn = a a 0 f (x) sin nπx a dx. Y − λ2 Y = 0 Y (y) = Ae−λy + Beλy .32) ( 138 . y) → 0 as y → ∞ we require B = 0.25) (23. y) = e−λn y sin nπx a (23. . 23. nπx = sin a (23. y) = 0 = u(a.27) u(0. Thus ∞ u(x. 2.2 A Semi-infinite strip with specified temperatures: uxx + uyy = 0 u(x. 0<y<∞ (23.29) 2. Since u(x.

3 above.36) Thus w satisfies the same BVP as does u in Eg. v = αx + β v(0) = A = β v(a) = αa + A = B B−A Therefore v(x) = x + A. y) ⇒ w(a. 0 = uxx + uyy = vxx +wxx + vyy +wyy ⇒ ∆w = 0 A = u(0. SEMI-INFINITE STRIP PROBLEMS Example 23.3.34) (23. y) = 0 (23.3 Semi-infinite strip with inhomogeneous BC: Look for a function v(x) for which v = 0 and which satisfies the inhomogeneous BC. 0) = v(x) + w(x.23.33) (23. y) = v(x) + w(x.38) 139 .37) where a 2 dn = a 0 {f (x) − v(x)} sin nπx a dx.35) f (x) = u(x. y) ⇒ w(0. y) = v(a) + w(a. y) = v(0) + w(0. (23. Therefore ∞ u(x. y) = 0 B = u(a. 0) = f (x) − v(x). y). a Now let u(x. (23. 0) ⇒ w(x. y) = (B − A)(x/a) + A + n=1 d n e −( nπ a )y sin nπx a (23.

More Rectangular Domains and semi-infinite strip problems 140 .Lecture 27 .

y) = X(x)Y (y). Let u(x.5) 141 . y) = 0 0<x<a ux (a. b). X (x) Y (y) =− = λ2 X(x) Y (y) Y (y) + λ2 Y (y) = 0 Y (0) = 0 = Y (b) (24.Chapter 24 Lecture 28 . ux (0.only flux BC and Circular domains 24.1 Neumann Problem on a rectangle Example 24. y) = f (y) 0<y<b (24.1) (24.3) uy (x.1 Neumann Problem: uxx + uyy = 0.2) (24. 0) = 0 = uy (x.4) Y Y = A cos λy + B sin λy = −Aλ sin λy + Bλ cos λy (24.Neumann Problem .

ux (a.only flux BC and Circular domains Y (0) = λB = 0 λ = 0 or B = 0.15) ( 142 . .24. y) = A0 + n=1 An cosh nπy nπx cos . y) = X0 Y0 = 1 Therefore ∞ satisfy homog. an = b b 0 f (y) cos nπy b dy. ∞ ux (x. y) = n=1 ∞ An An n=1 nπ nπx nπy sinh cos b b b nπ nπa sinh b b cos nπy b (24.14) .13) = f (y) .Neumann Problem . y). nπy . .8) (24. Y0 = 1 Yn = cos b (24. b b (24.11) ( u(x. Choose D0 = 1: X0 = 1 n≥1 Xn = cn cosh(λn x) + Dn sinh(λn x) Xn = cn λ sinh(λn x) + Dn λ cosh(λn x) Xn (0) = λn Dn = 0 (24. λn = (nπ/b) n = 0.9) (24.6) Y (b) = −Aλ sin λb = 0 (24. X0 = c0 x + D0 ⇒ X0 = c0 ⇒ X0 (0) = c0 = 0. y) = This is like a Fourier Cosine Series for f (y) but without the constant term a0 .(24. Recall a0 f (y) = + 2 ∞ n=1 b nπy 2 an cos . . Thus un (x.7) X n − λ2 X n = 0 Xn (0) = 0 n = 0: X0 = 0.10) Choose cn = 1: Xn = cosh(λn x). 1. .12) Now f (y) = ux (a. 24.Lecture 28 . BC. y) = Xn Yn = cosh(λn x) cos(λn y) u0 (x.

19) where A0 is undetermined.24. (24. Note b 1.e. u(x. y) = A0 + n=1 An cosh nπx nπy cos L b (24.17) Therefore b 2 An = nπ sinh and nπa b f (y) cos 0 nπy b dy n≥1 (24. 143 . y) is said to be known up to an arbitrary constant.16) if 0 f (y) dy = 0 then there is no solution to the boundary value problem 1.18) ∞ u∞ (x. b 2. If 0 f (y) dy = 0 there is a net flux into the domain through the right hand boundary and. since the other boundaries are insulated.14) is consistent only if a0 = 0. there can be no steady solution – the temperature will continually change with time.1. i. It is possible that uxx + uyy = ut = 0. NEUMANN PROBLEM ON A RECTANGLE Thus the expansion (24. If 0 f (y) dy = 0 there is no net flux through the boundary and a steady state can exist. If b f (y) dy = 0 then 0 b An nπa 2 nπ sinh = b b b 0 f (y) cos nπy b dy. For this to be true we require that b f (y) dy = 0 0 b (24.

y) is the steady state of a 2D Heat Equation ut = uxx + uyy with u(x.21) ( Now u∞ (x. y) then ut dx dy = D D · u dx dy = ∂D ∂u ds = 0. y) dx (24. y.25) r = (x2 + y 2 )1/2 . R = rγ ⇒ γ(γ − 1) + γ − λ2 = γ 2 − λ2 = 0 ⇒ R(r) = Crλ + Dr−λ . 0) = u0 (x. y) dx dy.20) Therefore   ∂  u dx dy  = 0 ⇒ ∂t D D u dx dy = const for all time = D u0 (x.Lecture 28 . 24. 24.2 General Analysis of Laplace’s Equation on Circular Domains: Recall the Laplacian in polar coordinates: 1 1 0 = ∆u = uxx + uyy = urr + ur + 2 uθθ r r Let u(r. R = rγ ⇒ γ(γ −1)+γ = γ 2 = 0 ⇒ R(r) = C +D ln r λ = 0: r2 R + rR − λ2 = 0.3 R Equation: λ = 0: r2 R +rR = 0.Neumann Problem .24) (24. θ) = R(r)Θ(θ) + rR Θ = − = λ2 R(r) Θ(θ) (24.23) θ = tan−1 (y/x) r2 R which leads to r2 R + rR − λ2 R = 0 and Θ + λ2 Θ = 0. 144 .22) Which the condition that determines A0 . y) = A0 × area(D) = ∂D ∂D u0 (x. ∂n (24.only flux BC and Circular domains 3. (24. If u∞ (x. 24.

αn = 0 and βn = 0. . Θn (θ) ∈ {1.27) ( n=1 + 24. 145 .26) Bn rλn + βn r−λn sin λn θ. . 2. 2. 1. 24. Θn (θ) ∈ {1. • The values of λn and the corresponding eigenfunctions depend on the boundary conditions (I)–(V) that apply. 2. . . .5. . 1. . the condition |u| < ∞ as r → 0 dictates that α0 = 0.1 Notes: • For problems that include the origin. . . . 2. The most general solution is thus of the form ∞ u(r. sin λn θ}. θ) = {A0 + α0 ln r} · 1 + n=1 ∞ An rλn + αn r−λn cos λn θ (24. 2.24. . Θn (θ) = cos λn θ (V) Θ(−π) = Θ(π) Θ (−π) = Θ (π) λn = n n = 0. . cos λn θ. .4 Θ Equation: Θ + λ2 Θ = 0 Θ = A cos λθ + B sin λθ Θ = −Aλ sin λθ + Bλ cos λθ 24. . • For problems that involve infinite domains the condition |u| < ∞ as r → ∞ dictates that An = 0 and Bn = 0. .5 For Different Boundary Conditions: (I) Θ(0) = 0 = Θ(α) λn = nπ/α n = 1. Θn (θ) = sin λn θ (II) Θ (0) = 0 = Θ (α) λn = nπ/α n = 0. cos λn θ} (III) Θ(0) = 0 = Θ (α) λn = (2n − 1)π/2α n = 1.4. Θ EQUATION: 24. Θn (θ) = sin λn θ (IV) Θ (0) = 0 = Θ(α) λn = (2n − 1)π/2α n = 1.

Neumann Problem .Lecture 28 .only flux BC and Circular domains 146 .

r =− = λ2 ⇒ Θ + λ2 Θ = 0 R Θ(θ) u(r. u(r. 0 < θ < α (25. α) = R(r)Θ(α) = 0 ⇒ Θ(α) = 0 Eigenvalue Problem Therefore λn = (nπ/α) n = 1.1 Wedge with homogeneous BC on θ = 0. θ = α < 2π 1 1 urr + ur + 2 uθθ = 0 r r u(r. 2.3) u(a.2) (25. α) = 0 u(r.4) Θ(0) = A = 0 Θ(α) = B sin(λα) = 0 . . (25. 0) = R(r)Θ(0) = 0 ⇒ Θ(0) = 0. . 0) = 0 u(r. .1) (25. (R + 1 R ) Θ (θ) r2 R + rR − λ2 R = 0 Euler Eq.5) 147 Θ + λ2 Θ = 0 Θ(0) = 0 = Θ(α) Θ = A cos λθ + B sin(λθ) (25. θ) bounded as r → 0 Let u(r. Θn = sin nπθ α .Chapter 25 Lecture 29 Wedge Problems Example 25. r2 0 < r < a. θ) = R(r) · Θ(θ). θ) = f (θ).

(25. α) = u1 0 < r < a. θ) < ∞ as r → 0 we require c2 = 0. θ) = f (θ) = n=1 nπ cn a( α ) sin This is just a Fourier Sine Series for f (θ): Therefore α nπ cn a( α ) = 2 α 0 f (θ) sin α nπθ α dθ nπθ α (25.7) u(r.8) (25.6) (25.Lecture 29 Wedge Problems To solve the Euler Eq. R = γ(γ − 1)rγ−2 . (25. Therefore ∞ (25. (25.10) cn = Therefore 2 −( nπ ) a α α 0 f (θ) sin dθ. 0<θ<α (25. Therefore γ(γ − 1) + γ − λ2 = γ 2 − λ2 = 0 ⇒ γ = ±λ. θ) = f (θ). θ) = n=1 ∞ cn r( nπ α ) sin nπθ α nπθ α .14) . let R(r) = rγ . 0) = u0 148 u(r.13) u(r. (25. θ) < ∞ as r → 0 u(a.12) Example 25.2 A wedge with Inhomogeneous BC 1 1 urr + ur + 2 uθθ = 0 r r u(r.11) ∞ u(x. R = γrγ−1 . Therefore R(r) = c1 rλ + c2 r−λ .9) u(a. Now since u(r. θ) = n=1 cn r( nπ α ) sin nπθ α .

α) = 0  solved in Example 25. θ) = w(θ) + v(r. θ) = R(r)Θ(θ) ⇒ r2 R + R r Θ equation Θ + λ2 Θ = 0 Θ (0) = 0 = Θ (α) Θ(θ) = A cos λθ + B sin(λθ) Θ (0) = Bλ = 0 λ = 0 or B = 0.20) Θ (θ) = −Aλ sin(λθ) + Bλ cos(λθ) (25.16) f (θ) − w(θ) sin 0 nπθ α dθ.15) v(r.Let us look for the simplest function of θ only that satisfies the inhomogeθ neous BC of the from: w(θ) = (u1 − u0 ) + u0 . 0) = 0 uθ (r. . . θ) = (u1 − u0 ) where ∞ nπ 2 cn = a−( α ) α θ + u0 + α ∞ cn r( n=1 nπ α ) sin nπθ α (25. Let 1 u(r. 1. . Then let u(r.21) Θ (α) = −Aλ sin(λα) = 0 λn = nπ .  1 1 1 1 urr + ur + 2 uθθ = vrr + vr + 2 vθθ = 0   Essentially the problem r r r r (25.18) (25. 0) = 0 v(r.17) Example 25.3 A wedge with insulating BC on θ = 0 and θ = α < 2π. Note that wθθ = 0 and that α w(0) = u0 and w(α) = u1 . 1 1 urr + ur + 2 uθθ = 0 r r uθ (r. θ) = f (θ). α 149 . α) = 0 u(a. /R(r) = −Θ /Θ = λ2 (25. θ) = f (θ) − w(θ) The solution is u(r. n = 0.19) (25.1  v(a. θ). (25.

(25. .27) (25. Θ(0) = 0 Θ (π) = 0 2 2 Θ = A cos λθ + B sin λθ 150 . 1 1 ∆u = urr + ur + 2 uθθ = 0 r r subject to ∂u (r. n ≥ 1: r2 Rn + rRn − λ2 Rn = 0 ⇒ Rn = cn rλn + Dn r−λn . 0) = 0 Let u(r.e. Since u(r.26) (25.30) Θ(0) = A = 0 Θ (π) = Bλ cos(λπ) = 0 ⇒ πλ1 = .28) (25. . π) = 0 ∂θ u(a. u(r. . θ) = c0 + 2 c0 + 2 α ∞ cn r( n=1 ∞ nπ α ) cos nπθ α nπθ α α (25. θ) < ∞ (i.23) nπθ α nπ cn a( α ) cos n=1 c0 = 2 α 0 f (θ)d?θ ∞ nπ 2 cn = a−( α ) α f (θ) cos 0 dθ (25.25) Example 25. θ) = R(r)Θ(θ).24) u(r.22) (25. must be bounded) as r → 0 we require d0 = 0 = Dn . θ) = f (θ) = u(a. n n = 0: rR0 + R0 = (rR0 ) = 0 ⇒ rR0 = a?0 ⇒ R0 (r) = a?0 ln r + c0 . θ) = c0 + 2 cn r( n=1 nπ α ) cos nπθ α .4 Mixed BC . .Lecture 29 Wedge Problems R equation r2 Rn + rRn − λ2 Rn = 0. θ) = f (θ). r2 Θ equation Θ + λ2 Θ = 0 R + 1R r R =− Θ (θ) = λ2 Θ(θ) (25.a ‘crack like’ problem. Therefore u(r.29) Θ = −Aλ sin λθ + Bλ cos λθ π 3π (25.

(25.31) Since u should be bounded as r → 0 we conclude that dn = 0.or λn = (2n + 1) 1 n = 0. .34) π/2 m = n Therefore cn = 2a−(n+ 2 ) π 0 ∞ 1 π f (θ) sin 1 n+ 1 2 1 2 θ dθ (25. The general solution is thus ∞ u(r.32) (25. 2 2 R + rR − λ2 R = 0 R(r) = r γ ⇒ γ 2 − λ2 = 0 γ = ±λ. 1. θ) = n=0 cn a(2n+1)/2 sin Check orthogonality π sin 0 2m + 1 2 θ sin 2n + 1 2 θ dθ = 0 m=n . θ) = n=0 cn r(2n+1)/2 sin ∞ (2n + 1) θ 2 2n + 1 2 θ .36) 151 . R equation r Therefore un (r. θ) = cn rλn + dn r−λn sin λn θ. .35) u(r. (25. (25. (25. λ = 0 as this would be trivial. .33) f (θ) = u(a. θ) = n=0 cn r(n+ 2 ) sin n+ θ .

Lecture 29 Wedge Problems 152 .

θ) = 0 u(a. 0) = 0 uθ (r. . (26. θ) = f (θ) Let u(r. λ = nπ n = 0. θ) = R(r)Θ(θ). α) = 0 u(b.4) Θ (0) = Bλ = 0 ⇒ B or λ = 0.5) 153 Θ = −Aλ sin λθ + Bλ cos λθ Θ (α) = −Aλ sin λα = 0. circles. α . . r2 (R + 1 R) Θ (θ) r =− = λ2 ⇒ R(r) Θ(θ) Θ equation Θ + λ2 Θ = 0 Θ (0) = 0 = Θ (α) r2 R + rR − λ2 R = 0 Θ + λ2 Θ = 0 (26. holes and annuli Example 26.Chapter 26 Lecture 30 Wedges with cut-outs. .2) (26.3) Θ = A cos λθ + B sin λθ (26. 1.1) (26.1 A circular wedge with a cut-out: 1 1 urr + ur + 2 uθθ = 0 r r uθ (r.

and cn = 0 n ≥ 1. holes and annuli R equation n = 0: (rR0 ) = 0 rR0 = B0 R0 = A0 + B0 ln r.15) c0 = 1 α ln(a/b) 0 f (θ) dθ.7) (26. θ) = c0 ln + b u(a.10) Rn (r) = An [r λn −b 2λn −λn r ] cn r( n=1 a b nπ α ) − b( 2nπ ) r−( nπ ) cos α α nπθ α (26. θ) = ln b Therefore r u(r.6) Therefore R0 = B0 ln(r/b). θ) = f (θ) = 2 = Therefore α ∞ (26. θ) = R0 (b)Θ0 (θ) = 0 ⇒ R0 (b) = A0 + B0 ln b = 0.Lecture 30 Wedges with cut-outs. θ) = r( α ) − b2( α ) r−( α ) cos α r ·1 u0 (r. circles.8) (26.14) α cn = 2 2nπ nπ nπ α a( α ) − b( α ) a− α α f (θ) cos 0 nπθ α dθ (26.12) α (26. (26. A0 = −B0 ln b. Note u0 (b. n ≥ 1: r2 Rn + rRn − λ2 Rn = 0 R(r) = An rλn + Bn r−λn Rn (b) = An bλn + Bn b−λn = 0 ⇒ Bn = −An b2λn Choose An = 1. (26.13) c0 ln 2 ∞ ∞ + n=1 nπ 2nπ nπ cn a( α ) − b( α ) r−( α ) cos a0 + 2 an cos n=1 nπθ α . log(a/b) α nπθ By Fourier basis function orthogonality 1 · cos dθ = 0 so that the α 0 154 . nπ nπ nπ nπθ un (r.11) nπθ (26.16) Note: In the special case f (θ) = 1. c0 = 1 . Choose B0 = 1.9) (26. 2 2c0 ln(a/b) = α 0 f (θ) dθ. (26.

n ≥ 1: λn = n: r2 Rn + rRn − (n)2 Rn = 0 Euler Eq. θ) = log(r/b) log(a/b) (26.26) 155 . 1.solution reduces to: u(r.20) Θ + λ2 Θ = 0 Θ Θ equation Θ + λ2 Θ = 0 Θ = A cos(λθ) + B sin(λθ) θ(−π) = A cos(λπ) − B sin(λπ) = Θ(π) = A cos(λπ) +B sin(λπ) ⇒ 2B sin(λπ) = 0 Θ (θ) = −Aλ sin(λθ) + Bλ cos(λθ) +Bλ cos(λπ) ⇒ 2Aλ sin(λπ) = 0.2 Complete circle (interior circle) 1 1 urr + ur + 2 uθθ = 0 r r 0 < r < a. Example 26.17) which is purely radial i. 2. = +λ2 ⇒ (26. θ) = R(r)Θ(θ). .21) (26. . θ) = f (θ) u(r. Therefore λn = (26.24) nπ = n.25) Θ (−π) = Aλ sin(λπ) + Bλ cos(λπ) = Θ (π) = −Aλ sin(λπ)(26. (26. 0 < θ < 2π. .18) BC: u(a. (26. Θn (θ) = cos(nθ) and sin(nθ). π R equation n = 0: λ0 = 0 ⇒ rR0 + R0 = (rR0 ) = 0 R0 = B0 ln r + A0 . Let u(r.22) (26. has no θ dependence.19) Θ r2 R + rR − λ2 R = 0 Euler Eq.23) (26. Rn = rγ ⇒ γ(γ − 1) + γ − n2 = 0 γ = ±λn = ±n Rn = cn r−n + dn rn . n = 0.e. r2 R + 1 R r R(r) =− (26. θ) < ∞ r → 0 Periodicity u(θ + 2π) = u(θ) periodic.

θ) = a0 + 2 an π π ∞ an cos(nθ) + bn sin(nθ) r−n n=1 (26.34) (26.28) u(a. θ) = ∂u ∂r = f (θ) r=a (26. (26.31) an = an f (θ) cos(nθ) dθ.29) −π π a−n π −π f (θ) sin(nθ) dθ. u(r. Therefore u(r.30) Note: 1.Lecture 30 Wedges with cut-outs. holes and annuli Now since u(r. For problem exterior to a circle we require that u(r. θ) < ∞ as r → 0 we must exclude solutions that blow up.32) −π −π Remark 26. (26. In this case we require that B0 = 0 and that dn = 0 so that R0 = A0 and Rn = r−n · an cos nθ + bn sin(nθ) .36) r=a n=1 ∞ = n=1 nan−1 an cos(nθ) + bn sin(nθ) .27) (26. θ) < ∞ as r → ∞.33) a0 2 = ∞ + n=1 ∞ rn an cos(nθ) + bn sin(nθ) nrn−1 an cos(nθ) + bn sin(nθ) (26. θ) = f (θ) = π a0 + 2 ∞ an cos(nθ) + bn sin(nθ) an n=1 a0 = 1 π −π f (θ) dθ a−n an = π bn = π f (θ) cos(nθ) dθ (26. θ) = a0 + 2 ∞ an cos(nθ) + bn sin(nθ) rn n=1 (26. bn = π π f (θ) sin(nθ) dθ. u(r.periodic. 156 .35) (26. circles.3 Note 2: (Neumann BC) 1 1 urr + ur + 2 uθθ = 0 r r ∂u (a. θ) = f (θ) ∂r u 2π . Thus B0 = 0 and cn = 0.

26.1. SPECIAL CASE - ELECTRICAL IMPEDANCE TOMOGRAPHY
π

1 A solution will not exist unless a0 = 2π
−π

f (θ) dθ = 0. Otherwise there is a

net flux of heat across the boundary and no steady state solution will exist.

26.1

Special Case - Electrical Impedance Tomography
π π − I0 δ θ + . 2 2

f (θ) = I0 δ θ −

f (−θ) = I0 δ − θ + = I0 δ θ + Thus f is odd ⇒ a0 = an = 0.
π

π π − I0 δ θ − 2 2

π 2

− I0 δ − θ −

π 2

(26.37) (26.38)

= −f (θ).

na

n−1

bn = bn =

2 π
0

I0 δ θ −

π sin(nθ) dθ 2

(26.39) (26.40) r a
n

u(r, θ) =

2I0 nπ sin πnan−1 2 ∞ 2aI0 sin(nθ) nπ sin π n 2
n=1

(26.41)

For enrichment ↓ =

2aI0 π aI0 π

∞ n=1 ∞

r a r a
∞ n

n

1 cos n θ − 2 n cos n θ − n
∞ n=1 n z2 n π 2

π 2

cos n θ + n
π 2

π 2

(26.42)

=

n=1

cos n θ + n
r a r a

(26.43)

=

aI0 Re π

n=1

n z1 − n

z1 = z2 =

π ei(θ− 2 ) . π ei(θ+ 2 )

(26.44)

157

Lecture 30 Wedges with cut-outs, circles, holes and annuli Now for |z| < 1: 1 1−z

= 1 + z + z2 + · · · z2 + ··· z+ 2

=
k=0 ∞

zk zn n . (26.45)

− ln(1 − z) =

=
n=1

Therefore u(r, θ) = − Now if (1 − z) = Aeiφ then Re ln(1 − z) = Re ln Aeiφ Therefore u(r, θ) = − Now z1 = −z
2

aI0 Re ln π

1 − z1 1 − z2

.

(26.46)

= Re[ln A + iφ] = ln A.

(26.47)

aI0 1 − z1 2 ln . 2π 1 − z2

(26.48)

r i(θ− π ) 2 e = ρeiφ1 a = (1 − z1 )(1 − z1 ) = 1 − ρeiφ1 = 1−ρ e
iφ1

(26.49) 1 − ρe−iφ1 (26.50) (26.51) (26.52)

+e

−iφ1

2

= 1 − 2ρ cos φ1 + ρ2 . Similarly z2 = Therefore

r i(φ+ π ) 2 e = ρeiφ2 and |1 − z2 |2 = 1 − 2ρ cos φ2 + ρ2 . a
r r 1 − 2( a ) cos(θ + π ) + ( a )2 aI0 2 ln r r 2π 1 − 2( a ) cos(θ − π ) + ( a )2 2

u(r, θ) = u(r, θ) =

(26.53) (26.54)

aI0 a2 + 2ar sin θ + r2 ln 2 2π a − 2ar sin θ + r2

158

26.2. POISSON’S INTEGRAL FORMULA:

26.2

Poisson’s Integral Formula:
a0 + 2 1 2π
−π ∞

u(r, θ) =

an cos(nθ) + bn sin(nθ) rn 1 f (φ)dφ + π
π ∞ n=1 π

(26.55)

n=1 π

=

r a

n

f (φ) cos nθ cos nφ + sin(nθ) sin(nφ) dφ
−π

(26.56) =  1 1 π 2
π ∞ π

f (φ) dφ +
−π n=1

r a
n

n −π

f (φ) cos n(θ − φ) dφ

  

(26.57)

= Now
∞ n=1

1 π
−π

f (φ)

1 + 2

∞ n=1

r a

cos n(θ − φ)

(26.58)

r n cos n(θ a

− φ) = Rρ = = = =

zn

n=1 z Rρ 1−z z(1−¯) z Rρ (1−z)(1−¯) z
r r ( a ) cos(θ−φ)−( a ) r 2 r 1−2( a ) cos(θ−φ)+( a ) 2

r z = a ρi(θ−φ) (1 − z)(1 − z ) = 1 − (z + z ) + |z|2 ¯ ¯ r = 1 − 2 a cos(θ − φ) + 1? r 2 r z(1 − z ) = a ρi(θ−φ) − a ¯

r 2 a

ar cos(θ−φ)−r2 a2 −2ar cos(θ−φ)+r 2

Therefore
π

u(r, θ) =

1 π
−π π

f (φ)

1 ar cos(θ − φ) − r2 + 2 2 a − 2ar cos(θ − φ) + r2
1 2 2a

(26.59)

=

1 π
−π

f (φ)

− ar cos(θ − φ) + 1 r2 + ar cos(θ − φ) − r2 2 a2 − 2ar cos(θ − φ) + r2

(26.60)
π

1 2 u(r, θ) = (a − r2 ) 2π
−π

f (φ) dφ a2 − 2ar cos(θ − φ) + r2

(26.61)

159

Lecture 30 Wedges with cut-outs. circles. holes and annuli 160 .

2.Chapter 27 Lecture 31 Sturm-Liouville Theory 27. 1. 2. . 1. x + λ2 X = 0 X(0) = 0 = X(L) X + λ2 X = 0 X (0) = 0 = X (L) X + λ2 X = 0 3. . . Xn = sin n = 0. • The eigenvalue problems thus far have been simple 1. L → λn = X(x + 2π) = X(x) periodic BC → λn = nπ . Xn = cos nπx . . . L Xn = X0 = 1. . n = 0.1 Boundary value problems and Sturm-Liouville theory: • Up till now we have been able to solve partial differential equations by separating variables which typically reduces the PDE to solving an eigenvalue problem and some initial value problem (for the Heat and Wave Equations) or some inhomogeneous boundary value problem (in the case of Laplace’s Equation). 2π → λn = nπ L nπ L n = 1. L nπx . . sin cos nπx L nπx L 161 . . .

1) where p. If p(x) or r(x) is zero for some x or the domain is [0. Remark 27. The periodic BC X(0) = X(2π) are not separated so that Problem (3) is not a SL Problem.1 Note: 1. ∞) then the problem is singular. The socalled Sturm-Liouville Problems define a class of eigenvalue problems which include many of the above problems as special cases. We define the Sturm-Liouville eigenvalue problem as:  Ly = λry Ly = −(py ) + qy  SL (27. q and r are continuous on 0 ≤ x ≤ and p(x) ≥ 0 and r(x) > 0 on 0 ≤ x ≤ .2 The regular Sturm-Liouville problem: p(x)y − q(x)y + λr(x)y = 0 0 < x < α1 y(0) + α2 y (0) = 0 β1 y( ) + β2 y ( ) = 0 (27. p . 27. If p = 1 q = 0 r = 1 α1 = 1 α2 = 0 β1 = 1 β2 = 0 we obtain Problem (1) above whereas if p = 1 q = 0 r = 1 α1 = 0 α2 = 1 β1 = 0 β2 = 1 we obtain Problem (2) above. 2.2) α1 y(0) + α2 y (0) = 0 β1 y( ) + β2 y ( ) = 0  p(x) > 0 and r(x) > 0. 3. There is no loss of generality in the form of Ly = −(py ) + qy since it is possible to convert a general 2nd order eigenvalue problem −P (x)y − Q(x)y + R(x)y = λy to this form by multiplying by an integrating factor µ(x) −µ(x)P (x)y − µQ(x)y + µ(x)R(x)y = λµ(x)y 162 (27.Lecture 31 Sturm-Liouville Theory • In this section we abstract these problems to a general class of boundary value problems which share a common set of properties. The S − L Problem helps to identify those assumptions that are needed to define an(delete the word an?) eigenvalue problems with the properties that we require.3) . Notice that the boundary conditions for these two problems are specified at separate points and are called separated BC.4) (27. If p > 0 and r > 0 and < ∞ then the SL Problem is said to be regular.

(27.2.7) (27.5) So we can identify that p = µP and p = µQ ⇒ p = µ P + µP = µQ P Q which is a linear 1st order ODE for µ with integrating factor ρ − dx P P µ + P Q − P P µ = 0 ⇒ Pρ − R Q P dx µ =0 µ= ρ R Q P dx P .9) − ρx p(x) = ρx 2 /2 2 /2 φ = λρx 2 /2 φ r(x) = ρx 2 /2 Example 27.12) y +ρ −x5 /5 4 −x5 /5 x y y −x5 /5 −x5 /5 163 .3 Convert the equation −y + x4 y 1 = λy to SL form P = 1 Q = −x4 Therefore − ρ −x5 /5 µ = ρ− − ρ R x4 dx = ρ−x = λρ = λρ 5 /5 (27. THE REGULAR STURM-LIOUVILLE PROBLEM: but Ly = −py − p y + qy = λry.10) (27.8) into SL form by multiplying by the integrating factor R 1 RQ 2 µ = ρ P dx = ρ x dx = ρx /2 P (x) = 1 p 2 2 2 ρx /2 φ + ρx /2 xφ + λρx /2 φ = 0 Q(x) = x R(x) = 1.27. (27.8) We bring (27.6) Example 27.11) y. (27.2 Reducing a boundary value problem to SL form: φ + xφ + λφ = 0 φ(0) = 0 = φ (27. (27.

Expansion Property: {φj (x)} are complete if f (x) is piecewise smooth then ∞ f (x) = where cn = cn φn (x) n=1 R r(x)f (x)φn (x) dx 0 (27. ). . (b) There are an ∞ # of eigenvalues λj with λj < λ2 < . (c) λj > 0 provided α2 β2 2.16) (27.Lecture 31 Sturm-Liouville Theory 27. j (b) The eigenfunctions corresponding to different eigenvalues are orthogonal with respect to the weight function r(x): r(x)φj (x)φK (x) dx = 0 j = k. β1 α1 < 0. Eigenvalues: (a) The eigenvalues λ are all real.3 Properties of SL Problems 1. 3.14) R 0 r(x)φ2 (x) dx n Example 27. < λj → ∞ as j → ∞.4 Robin Boundary Conditions: X + λX = 0 α = µ2 X (0) = h1 X(0) X ( ) = −h2 X( ) X(x) = A cos µx + B sin µx X (x) = −Aµ sin µx + Bµ cos µx 164 (27.13) (c) φj (x) has exactly j − 1 zeros on (0. and which satisfy: (a) φj (x) are real and can be normalized so that 0 r(x)φ2 (x) dx = 1. . > 0 q(x) > 0. 0 (27.15) (27.17) . Eigenfunctions: For each λj there is an eigenfunction φj (x) that is unique up to a multiplicative const.

(27. BC 2: X ( ) = −Aµ sin(µ ) + Bµ cos(µ ) = −h2 X( ) = −h2 [A cos µ + B sin µ ] (27. (27.25) 165 . µ2 − h1 h2 (27.18) ⇒B − µ2 h1 sin(µ ) + µ cos(µ ) = −Bh2 µ cos µ + sin µ h1 = 0. .21) h1 and h2 = 0: Xn = µn cos µn x + sin µn x h2 = 0 h1 = 0: Xn = = h1 → ∞ h2 = 0: µn cos µn x + sin µn x h1 cos µn ( − x) sin µn (27.23) Xn = sin(µn x) 2n − 1 µn ∼ 2 (27.24) π n = 1. .19) B Therefore − h2 µ2 + h2 sin µ + µ + µ cos µ h1 h1 µ(h1 + h2 ) .3.20) tan(µ ) = (27.27.22) (27. 2. PROPERTIES OF SL PROBLEMS BC 1: X (0) = Bµ = h1 X(0) = A A = Bµ/h1 . .

Lecture 31 Sturm-Liouville Theory 166 .

µ n 167 .4) If f (x) = x then 1 x sin(µn x) dx = 0 − cos(µn x) − x + µn 0 1 1 µn 1 cos µn x dx 0 = − cos(µn ) + µn but − µn cos µn = sin µn = 1 sin µn x 2 µn 0 (28.3) 1 = cn 1 + cos2 µn ·2 Therefore 1 2 cn = [1 + cos2 µn ] 0 f (x) sin(µn x) dx. (28.1 Expansion in Robin Eigenfunctions ∞ Assume that we can expand f (x) in terms of φn (x): f (x) = n−1 1 1 cn φn (x) (28.1) f (x) sin(µn x) dx = cn 0 0 φn (x) 2 dx (28.Chapter 28 Lecture 32 Solving the heat equation with Robin BC 28.2) (28.5) sin µn −µn cos µn µ2 n µ = 2 sin 2 n .

t) = 0 f (x) = u(x.2 Solving the Heat Equation with Robin BC ut = α2 uxx u(0.13) 1 = u(0.11) v = Ax + B v(0) = B = 1 v (x) = A v (1) + v(1) = A + (A + 1) = 0 (28.Lecture 32 Solving the heat equation with Robin BC Therefore cn = 4 sin µn µ2 [1 + cos2 µn ] n ∞ (28. t) = 1 + w(0. 0) = v(x) + w(x.8) (28. t) = 0 v (x) = 0 v(0) = 1 v (1) + v(1) = 0 (28.9) (28. t) = v(0) + w(0. t) + w(1. t) + u(1.10) ux (1.6) (28. t) = 0 0 = ux (1. t) ⇒ w(0. t) ⇒ wx (1. 0) = f (x) − v(x). t) + u(1. t) ut = wt = α2 (v +wxx ) ⇒ wt = α2 wxx (28. Look for a steady state solution v(x) 0<x<1 (28.12) A = −1/2 Therefore v(x) = 1 − x/2. t) + w(1. Now let u(x. t) = 1 u(x. 168 .7) f (x) = 4 n=1 sin µn sin(µn x) µ2 [1 + cos2 µn ] n 28. t) = {v (1)+ v(1)} +wx (1. t) = v(x) + w(x. 0) ⇒ w(x. 0) = f (x).

17) equation: tan µn = −µn . (28.19) where ∞ f (x) − v(x) = w(x.21) cn ρ−α 2 µ2 t n u(x. t) = X(x)T (t) ˙ T (t) X = = −µ2 2 T (t) α X T (t) = cρ−α X + µ2 X = 0 X(0) = 0 X (1) + X(1) = 0 2 µ2 t (28. t) = n=1 cn ρ−α sin(µn x) (28.22) n=1 169 .20) ⇒ cn = 2 [1 + cos2 µn ] 0 [f (x) − v(x)] sin(µn x) dx (28.14) (28.18) 2 µ2 t n Xn (x) = sin(µn x) ∞ w(x.16) The µn are solutions of the transcendental (28. 0) = n=1 cn sin(µn x) 1 (28. (28.28. t) = 1 − n? + 2 ∞ sin(µn x).15) (28.2. SOLVING THE HEAT EQUATION WITH ROBIN BC Let w(x.

Lecture 32 Solving the heat equation with Robin BC 170 .

1 Eigenfunctions involving solutions to an Euler Equation: (x2 φ ) + λφ = 0 1<x<2 φ(1) = 0 φ(2) = 0 x2 φ + 2xφ + λφ = 0 An Euler Eq. r 2 .1 λ = 1: 4 Cases: φ(x) = c1 x− 2 + c2 x− 2 log x φ(1) = c1 = 0 φ(2) = c2 2 −1 2 1 1 (29. (29.4) log 2 = 0 ⇒ c2 = 0 (29.3) 29. Let φ(x) = xr Therefore r= −1 ± r(r − 1) + 2r + λ = r2 + r + λ = 0.5) 171 .1) = r1 . √ 2 1 − 4λ (29.Chapter 29 Lecture 33 Variable coefficient BVP eigenfunctions involving solutions to the Euler Equation: Example 29.2) (29.

13) (29.16) = dn x− 2 sin (βn ln x) ln x sin nπ ln(2) 2 choose dn = 1 or normalize so that 1 φ2 (x) dx = 1.Lecture 33 Variable coefficient BVP .14) (29.7) (29. 1 4 (29. t) 172 1<x<2 t>0 u(x. For λ > 1 4 (29. n Example 29. t) = 0 = u(2.6) (29.8) =0 2r1 −r2 = 1 ρ(r1 −r2 ) ln 2 = 1 = ρ2πin (r1 − r2 ) ln 2 = 2πin √ r1 − r2 = 1 − 4λ = 2πni/ ln(2) Thus to obtain nontrivial solutions we require 1 − 4λ < 0 λ > √ √ 1 − 4λ = i 4λ − 1 = 2πni/ ln(2).11) The corresponding roots r1 and r2 are as follows 1 1 (r1 )n = − + iβn and (r2 )n = − − iβn 2 2 φn (x) = cn x− 2 xiβn − x−iβn = cn x− 2 ρiβn ln x − ρ−iβn ln x = dn x −1 2 1 1 1 (29. 4λn − 1 = = (2βn )2 4 [ln(2)]2 [ln(2)]2 βn = (nπ/ ln 2). λ = 1: 4 φ(x) = c1 xr1 + c2 xr2 φ(1) = c1 + c2 = 0 c2 = −c1 φ(2) = c1 2 − 2 r1 r2 (29.15) (29. (29.9) .eigenfunctions involving solutions to the Euler Equation: so there is no Eigenfunction for λ = 1/4.12) (29. (29. 0) = f (x).2 Solving a variable coefficient Heat Conduction Problem: ut = D(x2 ux )x − u u(1.17) .10) The Eigenvalues are: λn = 1 π 2 n2 4π 2 n2 + .

20) T (t) = cρ−D(1+λ)t 1 (πn)2 λn = 1 + [ln(2)]2 . u(2.21) 4 ln 2 n = 1.22) (29.25) u(1. u(r. α) = f (r) (29.2.26) (29.23) f (x) = u(x.19) (29. θ) = 0 r2 R u(r. 0) = x cn sin nπ n=1 ln x ln 2 29. t) = x cn ρ−D(1+λn )t sin nπ 1 −2 ln x ln 2 (29. BC) (29.24) r r u(r. SOLVING THE HEAT EQUATION BY EXPANDING IN EIGENFUNCTIONS INVOLVING SOLUTIONS TO AN EULER EQUATION: Let u(x. . . 0) = 0. 0 < θ < α (29.27) (because of Homog. 2. t) = X(x)T (t) ˙ T (t) (x2 X ) = −1 DT (t) X ˙ T (t) +1 = DT (t) ˙ T + D(1 + λ)T = 0 (x2 X ) + λX = 0 X(1) = 0 = X(2) (x2 X ) = −λ X (29. ∞ −1 2 n=1 ∞ u(x. Xn (x) = x− 2 sin nπ ln x (29. θ) = R(r)Θ(θ) + rR Θ = − = −λ2 R(r) Θ (29.28) Θ: Θ − λ2 Θ = 0 Θ(0) = 0 Θ = c cosh λθ + D sinh λθ Θ(0) = c = 0 ⇒ Θ(θ) = D sin hλθ 173 . .18) (29.29.2 Solving the heat equation by expanding in eigenfunctions involving solutions to an Euler Equation: 1 1 ∆u = urr + ur + 2 uθθ = 0 1 < r < 2. θ) = 0.

. θ) = n=1 Bn sinh(λn θ) sin(λn ln r). (29. . Therefore ∞ u(r.33) (29.38) .(29.37) dx0 ? = 1 dr r Now 2 ln 2 1 sin r 1 mπ ln r ln 2 sin nπ ln r ln 2 dr = 0 sin mπx nπx sin ln 2 ln 2 dx = 0 ln 2 2 m=n (29. r r Now let us look for Eigenvalues and Eigenfunctions to ( ).32) (29. α) = n=1 Bn sinh(λn α) sin nπ ln r ln 2 x = ln r .36) Now match BC: ∞ f (r) = u(r.29) ln r and the corresponding Eigenfunctions are Rn = sin nπ ln 2 .30) 1 R ρ P Q dr P = 1 R ρ r2 r dr r2 = 1 ln r 1 ρ = . (29. . Although we can easily see that dividing through by r we can reduce ( ) to S-L form. m=n Therefore 2 2 Bn = ln 2 sinh nπα ln 2 f (r) sin r 1 m?n?π ln r r dr.eigenfunctions involving solutions to the Euler Equation: R: r2 R + rR + λ2 R = 0( ) R(1) = 0 = R(2).35) n = 1.39) 174 . Let R(r) = rγ ⇒ γ(γ − 1) + γ + λ2 = γ 2 + λ2 = 0 γ = ±iλ. 2.Lecture 33 Variable coefficient BVP . (29.31) R(r) = c1 riλ + c2 r−iλ riλ = ρiλ ln r (29. = A cos(λ ln r) + B sin(λ ln r) R(1) = A cos [λ(ln 1)] + B sin(λ ln 1) = A = 0 R(2) = B sin [λ ln 2] = 0 ⇒ λn ln 2 = nπ (29. r2 r (29.34) (29. let us use the integrating factor µ( ) = Therefore 1 λ2 − · ( ) ⇒ rR R = −(rR ) = R.

1 Properties of SL Problems: 1. β1 β2 > 0 and q(x) > 0. j • (b) the eigenfunctions corresponding to different eigenvalues are orthogonal with respect to the weight function r(x): r(x)φj (x)φk (x) dx = 0 if j = k.Chapter 30 Lecture 34 — Sturm Liouville Theory 30. 2.1) • (c) φj (x) has exactly j − 1 zeros on 0 < x < . • (c) λj > 0 provided α1 α2 < 0. Eigenvalue Properties: • (a) the eigenvalues λ are all real. 175 . 0 (30. Eigenfunction Properties: Corresponding to each eigenvalue λj there is an eigenfunction φj (x) that is unique up to multiplication by a constant. and which satisfy: • (a) φj (x) are real and can be normalized 0 r(x)φ2 (x) dx = 1. • (b) there are an infinite number of eigenvalues λj with λ1 < λ2 < . . . λj → ∞ as j → ∞.

2) by orthogonality: r(x)f (x)φn (x) dx cn = 0 . Expansion Property: The eigenfunctions φj (x) form a complete set so that any piecewise smooth function f (x) can be expanded as a generalized Fourier Series: ∞ f (x) = n=1 cn φn (x) (30.E. I. r(x)φ2 (x) dx n 0 (30.5) = −vpu 0 + upv 0 + 0 u −(pv ) + qv dx(30. (30.Lecture 34 — Sturm Liouville Theory 3.8) . then vLu dx = 0 0 v −(pu ) + qu dx (30.2 Lagrange’s Identity: Lagrange’s Identity (vLu − uLv) dx = −p(x)u v 0 0 + p(x)uv 0 .4) = −vpu 0 + 0 u pv dx + 0 uqv dx (30. is fundamental to the development of S-L Theory.6) Therefore 0 vLu dx = −pvu 0 + puv 0 + 0 uLv dx.3) 30. Proof: Let u and v be any sufficiently differentiable functions. α1 u(0) + α2 u (0) = 0 α1 v(0) + α2 v (0) = 0 176 β1 u( ) + β2 u ( ) = 0 β1 v( ) + β2 v ( ) = 0 (30.7) Now suppose that u and v both satisfy the SL boundary conditions.

rλ¯) y = 0 (30.30.3 Proofs to selected properties: (1a) λ is real: Let Ly = λry (1) α1 y(0) + α2 y (0) = 0 β1 y( ) + β2 y ( ) = 0. Ly) − (y.3. g) = 0 f (x)g(x) dx then we may write (v. Notation if we define (f.13) Thus 0 vLu dx = 0 uL?v dx whenever u and v satisfy the SL boundary condition. Take the conjugate of (1) L¯ = λr¯. Lv) = (u.12) (30. Note: • If L and BC are such that 0 vLu dx = 0 uLv dx then L is said to be self-adjoint. By Lagrange’s Identity: y ¯ y 0 = (¯. rλy) − (y.17) 177 .14) (30.11) (30. (30.9) (30.10) (30. Lv).15) ¯y y(x)r(x)λ¯(x) dx 0 y (x)rλy(x) dx − ¯ (30. L¯) y y ¯y = (¯. PROOFS TO SELECTED PROPERTIES: then vLu dx − 0 0 uLv dx = −p( )u ( )v( ) + p( )u( )v ( ) +p(0)u (0)v(0) − p(0)u(0)v (0) β1 β1 = p( ) + u( )v( ) + u( ) − v( ) β2 β2 α1 α1 +p(0) − u(0)v(0) − u(0) − v(0) α2 α2 = 0. 30.16) ¯ = (λ − λ) 0 r(x) y(x) dx 2 (30.

Lecture 34 — Sturm Liouville Theory ¯ Since r(x)|y(x)|2 ≥ 0 it follows that λ = λ ⇒ λ is real. ry 2 dx 0 [−py y]0 + = 0 0 p(y )2 + qy 2 dx (30.18) −(py ) y + qy 2 dx Therefore λ = 0 this is known as Rayleigh’s Quotient. Then 0 = (φK . (1c) λj > 0 provided α1 /α2 < 0 β1 /β2 > 0 and q(x) > 0. Consider two distinct eigenvalues λj = λk λj : Lφj = rλj φj and λk : Lφk = rλk φk . (30.21) (30. rλj φj ) − (φj . (2b) Eigenfunctions corresponding to different eigenvalues are orthogonal. rλK φK ) = (λj − λK ) 0 (30. Note: If q(x) ≡ 0 and α1 = 0 = β1 then with y (0) = 0 = y ( ) we have nontrivial eigenfunction y(x) = 1 and eigenvalue λ = 0. 0 (30. LφK ) by Lagrange’s Identity = (φK . Lφj ) − (φj .20) Therefore λ > 0 since the RHS is all positive.22) (30.24) 178 . Ly) = 0 −(py ) y + qy 2 dx = λ 0 r(x) y(x) 2 dx (30. Consider Ly = −(py ) + qy = λry (SL) and multiply (SL) by y and integrate from 0 to : (y.19) ry 2 dx +p( ) β1 y( ) β2 = 2 − p(0) α1 y(0) α2 ry 2 dx 0 2 + 0 p(y )2 + qy 2 dx .23) r(x)φK (x)φj (x) dx now λj = λk?K implies that r(x)φk (x)φj (x) dx = 0.

30.26) dx 0 179 . However. n=1 ∞ f (x)φm (x)r(x) dx = 0 n=1 cn 0 r(x)φm (x)φn (x) dx (30.3. PROOFS TO SELECTED PROPERTIES: (3) Expansion Property It is difficult to prove the convergence of the eigenfunction series expansion for f (x) that is piecewise smooth. if we assume the expansion converges then it is a simple matter to use orthogonality to determine the coefficients in the expansion: Let f (x) = ∞ cn φn (x).25) orthogonality implies r(x)f (x)φm (x) dx cm = 0 . r(x) φm (x) 2 (30.

Sign up to vote on this title
UsefulNot useful