Post on 09-Apr-2018
8/8/2019 Math 316 Notes
1/179
Contents
1 Lecture 1 - Introduction to Partial Differential Equations 5
1.1 Modeling and Derivation of PDE: . . . . . . . . . . . . . . . . 6
1.2 The Wave Equation: . . . . . . . . . . . . . . . . . . . . . . . 91.3 The Drunkards Walk - The Heat Equation: . . . . . . . . . . 10
2 Lecture 2 - Preliminaries 132.1 Sequences and Series of Numbers: . . . . . . . . . . . . . . . . 132.2 Absolute and Conditional Convergence: . . . . . . . . . . . . 162.3 Power Series: . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Lecture 3 - Review of Methods to Solve ODE 193.1 First Order ODE: . . . . . . . . . . . . . . . . . . . . . . . . . 193.2 Another Method - Series Solution: . . . . . . . . . . . . . . . 20
3.3 Second Order Constant Coefficient Linear Equations: . . . . . 213.4 Euler/Equidimensional Equations: . . . . . . . . . . . . . . . 23
4 Lectures 4,5 Ordinary Points and Singular Points 274.1 An Ordinary Point: . . . . . . . . . . . . . . . . . . . . . . . . 274.2 A Singular Point: . . . . . . . . . . . . . . . . . . . . . . . . . 284.3 The Airy equation: . . . . . . . . . . . . . . . . . . . . . . . . 304.4 The Hermite Equation: . . . . . . . . . . . . . . . . . . . . . . 31
5 Lecture 6 - Singular points 335.1 Radius of Convergence and Nearest Singular Points . . . . . . 335.2 Singular Points: . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Regular Singular Points: . . . . . . . . . . . . . . . . . . . . . 355.4 More General Definition of a Regular Singular Point: . . . . . 36
6 Lecture 7 - Frobenius Series about Regular Singular Points 396.1 Series Expansion Summary: . . . . . . . . . . . . . . . . . . . 41
1
8/8/2019 Math 316 Notes
2/179
CONTENTS
7 Bessels Equation 43
7.1 Bessels Function of Order / {. . . , 2, 1, 0, 1, 2 . . .}: . . . . 437.2 Bessels Function of Order = 0 - repeated roots: . . . . . . . 44
7.3 Bessels Function of Order = 12 : . . . . . . . . . . . . . . . . 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 . . . . . . . . . . . . . 7711.3 Now consider the sum of the first N terms . . . . . . . . . . . 78
12 Lecture 16 - Parsevals Identity 81
12.1 Geometric Interpretation of Parsevals Formula . . . . . . . . 82
13 Lecture 17 - Solving the heat equation using finite differencemethods 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 Laplaces Equation using finite differ-ences 91
14.1 Finite Difference approximation . . . . . . . . . . . . . . . . . 91
14.2 Solving the System of Equations by Jacobi Iteration . . . . . 93
2
8/8/2019 Math 316 Notes
3/179
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 DAlemberts 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 - Laplaces Equation 129
22.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
22.2 Laplaces 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-infinitestrip 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 Circulardomains 141
24.1 Neumann Problem on a rectangle . . . . . . . . . . . . . . . . 141
24.2 General Analysis of Laplaces Equation on Circular Domains: 144
24.3 R Equation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
3
8/8/2019 Math 316 Notes
4/179
CONTENTS
24.4 Equation: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
24.5 For Different Boundary Conditions: . . . . . . . . . . . . . . . 14524.5.1 Notes: . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 45
25 Lecture 29 Wedge Problems 147
26 Lecture 30 Wedges with cut-outs, circles, holes and annuli 15326.1 Special Case - Electrical Impedance Tomography . . . . . . . 15726.2 Poissons Integral Formula: . . . . . . . . . . . . . . . . . . . 159
27 Lecture 31 Sturm-Liouville Theory 16127.1 Boundary value problems and Sturm-Liouville theory: . . . . 16127.2 The regular Sturm-Liouville problem: . . . . . . . . . . . . . 162
27.3 Properties of SL Problems . . . . . . . . . . . . . . . . . . . . 164
28 Lecture 32 Solving the heat equation with Robin BC 16728.1 Expansion in Robin Eigenfunctions . . . . . . . . . . . . . . . 16728.2 Solving the Heat Equation with Robin BC . . . . . . . . . . . 168
29 Lecture 33 Variable coefficient BVP - eigenfunctions involv-ing solutions to the Euler Equation: 17129.1 Cases: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17129.2 Solving the heat equation by expanding in eigenfunctions in-
volving solutions to an Euler Equation: . . . . . . . . . . . . 173
30 Lecture 34 Sturm Liouville Theory 17530.1 Properties of SL Problems: . . . . . . . . . . . . . . . . . . . 17530.2 Lagranges Identity: . . . . . . . . . . . . . . . . . . . . . . . 17630.3 Proofs to selected properties: . . . . . . . . . . . . . . . . . . 177
4
8/8/2019 Math 316 Notes
5/179
Chapter 1
Lecture 1 - Introduction toPartial Differential Equations
ODE - Equations which define functions of a single independent variableby prescribing a relationship between the values of the function and itsderivatives.
EG:y(x) + ey(x) = 0. (1.1)
PDE - Involve multivariable functions u(x, t), u(x, y) that are determined byprescribing a relationship between the function value and its partial deriva-tives.
EG 1:
a
xu(x, y) + b
yu(x, y) = c First Order PDE (1.2)
5
8/8/2019 Math 316 Notes
6/179
Lecture 1 - Introduction to Partial Differential Equations
EG 2: Some Classic Second Order PDEs:
Quadric Classification Eq. Name
T = X2 Parabolic t
u(x, t) = 2
x2u(x, t) Heat Equation or Diffusion Eq
X2 + Y2 = k Elliptic 2
x2u(x, y) +
2
y2u(x, y) = f(x, y)
Poisson Eq f 0Laplace Eq f = 0
T2
c2X2 = k Hyperbolic
2
t2 u(x, t)
c2
2
x2 u(x, t) = 0 The Wave Eq
By analogy with quadric surfaces aX2 + 2bXY + c2Y2 + = k that can bereduced to a standard form by coordinate rotation, the most general linear2nd order PDE
auxx + 2buxy + cuyy + (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
8/8/2019 Math 316 Notes
7/179
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 qvaries 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 movingcoordinate system.
x = x ct transformation of coordinates (1.9)Guess:
u(x, t) = f(x ct) solves ut + cux = 0ut = cf ux = f (1.10)
Therefore ut + cux =
cf + cf = 0.
Thus ut + cux = 0 has solutions of the form u(x, t) = f(x ct) whichrepresents a right moving wave. What happens if q = cu in which case
u
t cu
x= 0 (1.11)
7
8/8/2019 Math 316 Notes
8/179
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) =
2u
t2 c2
2u
x2= 0 (1.12)
is the 2nd order wave equation that has both left and right moving wavesolutions.Fouriers Law: Heat flows from hotter regions to colder ones?
q = 2 ux
(1.13)
In this case the conservation law reduces to the form:
u
t= 2
2u
x2The Heat Equation (1.14)
2D Heat Equation:u
t= 2
2u
x2+
2u
y 2
(1.15)
8
8/8/2019 Math 316 Notes
9/179
1.2. THE WAVE EQUATION:
1.2 The Wave Equation:
Consider an elastic rod having a density and cross-sectional area A, and let(x, t) be the pressure in the rod at x at time t and u(x, t) the displacementof the rod from equilibrium.
Balance of Linear Momentum F = Ma.
(x + x, t)A (x, t)A = Ax2u
t2
(x + x, t) (x, t)x
= 2u
t2(1.16)
x 0 x
= 2u
t2(BLM) Balance of Linear Momentum
Hookes Law
= Eu
x(1.17)
Plug into (BLM) to obtain the 2nd order wave equation.
2u
t2= E
2u
x2= c2
2u
x2, where c = E (1.18)
9
8/8/2019 Math 316 Notes
10/179
Lecture 1 - Introduction to Partial Differential Equations
1.3 The Drunkards Walk - The Heat Equation:
Let u(x, t) be the density of fruit-flies at point x at time t. Find an equationfor 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) +px
[u(x + x, t)
u(x, t)]
x [u(x, t)
u(x
x, t)]
x
u(x, t) +px2
ux (x, t) ux (x x, t)
x
(1.19)
u(x, t) +px2 2u
x2
u(x, t + t) u(x, t)t
px2
t
2u
x2 u
t= 2
2u
x2The Heat Eq.
What is the Mean Absolute Deviation of the Drunkard?
sj = xtj = jt
xN = s1 + s2 + + sN 0 Expected Value (1.20)x2N = (s1 + + sN)2 = s21 + + s2N + 2(s1s2 + + sN1sN)
Nx2
Therefore
x2N
tNt
x2 = k2tN
|xN| k
tN
10
8/8/2019 Math 316 Notes
11/179
1.3. THE DRUNKARDS WALK - THE HEAT EQUATION:
0 2 4 6 8 10 12 14 16 18 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 withthe mean absolute deviation envelopes shown in red
11
8/8/2019 Math 316 Notes
12/179
Lecture 1 - Introduction to Partial Differential Equations
12
8/8/2019 Math 316 Notes
13/179
Chapter 2
Lecture 2 - Preliminaries
2.1 Sequences and Series of Numbers:
A Sequence of Numbers:
1, 12 ,13 , . . . ,
1n , . . .
1, 12 ,14 , . . . ,
12
n1, . . .
Notation{an} an = 1n{bn} bn =
12
n1A series of numbers:
1 + 12
+ 13
+ + 1n
+ = n=0
1n
n=0
an (2.1)
Does this infinite sum yield a finite result?
1 +
1
2
+
1
2
2+ +
1
2
n1=
n=1
1
2
n1 n=1
bn (2.2)
Note: In order to sum to a finite number the terms of the sequence musttend to 0 as n .Divergence Test:
lim an = 0 0
an diverges. (2.3)
EG: an = 1 1 + 1 + + 1 + .
13
8/8/2019 Math 316 Notes
14/179
Lecture 2 - Preliminaries
Integral Test: Does
n=11n converge?
Consider
1
dx
x< 1 +
1
2+
1
3+ + 1
n+ =
n=1
1
n(2.4)
Now
1
dx
x= lim
T
T1
dx
x= lim
T(ln T ln1) =
But
1
dx
x 1 p 1
14
8/8/2019 Math 316 Notes
15/179
2.1. SEQUENCES AND SERIES OF NUMBERS:
p > 1:
n=2
1
np 1
p 1:
1
dx
xp 1
diverges p 1
Geometric Series - the G-Series:n=0
rn = 1 + r + r2 + + rn + (2.6)
For what values of r does the G-Series converge?
15
8/8/2019 Math 316 Notes
16/179
Lecture 2 - Preliminaries
Partial Sum:
SN =N
n=0
rn = 1 + r + + rN
=(1 + r + + rN)
(1 r) (1 r)
=1 + r + + rN r r2 rN rN+1
1 r (2.7)
=1 rN+1
1 rIf
|r
|< 1 then
limN
Nn=0
rn = limN
1 rN+11 r =
1
1 r (2.8)
If |r| 1, series diverges.G-Series:
n=0
rn =
11r |r| < 1 |r| 1 (2.9)
EG:
n=01
2n=
1
1
1/2= 2, r =
1
2. (2.10)
2.2 Absolute and Conditional Convergence:
Alternating Series Test:(1)nan an > 0
(a) an+1 an(b) lim
n an = 0
(1)nan converges:
EG:
n=1
(
1)
n
n1
converges.
Consider a series
n=0
(1)nan.
If
|an| < then
(1)nan is said to be absolutely convergent.
16
8/8/2019 Math 316 Notes
17/179
2.3. POWER SERIES:
If |an| = but (1)nan < is conditionally convergent.
Ratio Test:
Consider
n=0
bn and let limn
bn+1bn = L. Then bn converges abso-
lutely if L < 1, diverges if L > 1. Test is inconclusive if L = 1.
EG 1:
n=1
(1)n1 2n
n2.
bn+1bn
=2n+1/(n + 1)2
2n/n2= 2
1 +
1
n
2 2 series diverges. (2.11)
EG 2:
n=1
n4en2
.
bn+1bn = (n + 1)4e(n+1)2n4en2 =
1 +
1
n
4e2n1 0 < 1 converges absolutely.
(2.12)
2.3 Power Series:
f(x) = a0 + a1x + a2x2 + + anxn polynomial approximation.Idea: Extend the polynomial to include # of terms.
f(x) = a0 + a1x + a2x2 + + anxn + Power Series
=
n=0
anxn (2.13)
EG: ex = 1 + x1! +x2
2! +x3
3! + + xn
n! + =
n=0xn
n! .
More General Power Series:
f(x) =
n=0
an(x x0)n = a0 + a1(x x0) + a2(x x0)2 + (2.14)
17
8/8/2019 Math 316 Notes
18/179
Lecture 2 - 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)n2 + f(x0) = 2a2
f(3)(x) = 3!a3 + 4.3.2(x x0) + + n(n 1)(n 2)an(x x0)n3 + f(3)(x0) = 3!a3
f(n)
(x0) = n!an an =f(n)(x0)
n!
Therefore f(x) =
n=0
f(n)(x0)
n!(x x0)n (2.15)
Alternative Form of Taylor Series:
f(x0 + h) =
n=0
f(n)(x0)
n!hn (2.16)
EG 1:
ex = n=0
xn
n!
sin x =
n=0
(1)n x2n+1
(2n + 1)!sinh x =
n=0
x2n+1
(2n + 1)!(2.17)
cos x =
n=0
(1)n x2n
(2n)!cosh x =
n=0
x2n
(2n)!
ei = 1 + i +(i)2
2!+
(i)3
3!+
= 1 22!
+ 4
4! + i 3
3!+ (2.18)
= cos + i sin
18
8/8/2019 Math 316 Notes
19/179
Chapter 3
Lecture 3 - Review ofMethods 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
xdx
y 3 ln |y| = 4 ln |x| + C (3.2)y = ln(x4y3) + C
Ax4y3 = ey
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 lefthand side into a derivative of a product:
F y + F P y = F Q (3.4)
19
8/8/2019 Math 316 Notes
20/179
Lecture 3 - Review of Methods to Solve ODE
(F y) = F y + Fy = F Q (3.5)
So let F = F P 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 = AeR
P(x) dx choose A = 1
F = eR
P(x) dx integrating factor
Therefore
eR
P(x) dxy + eR
P(x) dxP(x)y = eR
P(x) dxQ(x)(eR
P(x) dxy)
= eR
P(x) dxQ(x)
y(x) = eR
P(x) dx
eRx P(t) dtQ(x) dx + C
(3.7)EG: 1
y + 2y = 0 (3.8)
F(x) = e2x e2xy + e2x2y = (e2xy) = 0e2xy =?c
y(x) = Ce2x
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=0anx
n.
y =
n=1
annxn1 (3.9)
Therefore y + 2y =
n=1 annxn1 +
n=0 2anxn = 0
In the first sum let
m = n 1 n = 1 m = 0n = m + 1
20
8/8/2019 Math 316 Notes
21/179
3.3. SECOND ORDER CONSTANT COEFFICIENT LINEAREQUATIONS:
Therefore
m=0 am+1(m + 1)xm +
n=0 2anxn = 0
n m :
m=0{am+1(m + 1) + 2am} xm = 0
(3.10)
am+1 = 2(m+1) am
a1 = 2a0, a2 = + 22 21 a0, a3 = 23 22 21 a0 = (1)3 23
3! a0,
. . . , am = (1)m 2mm! a0
Therefore y(x) = a0
m=0(2x)m
m! = a0e2x
(3.11)
EG 2: Solvedy
dx+ cot(x)y = 5ecos x, y(/2) = 4.
P(x) = cot x Q(x) = 5ecos x
F(x) = eR
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
y(x) =
5ecosxCsin x
4 = y(/2) = 5C1 C = 1
Therefore y(x) = 15ecosx
sin x
(3.13)
3.3 Second Order Constant Coefficient Linear Equa-tions:
Ly = ay + by + cy = 0Guess y = erx y = rerx y = r2erx
Ly = (ar2 + br + c)erx = 0
Indicial Eq.:
ar2 + br + c = 0 r = b
b24ac2a
a(r r1)(r r2) = 0 (3.14)
21
8/8/2019 Math 316 Notes
22/179
Lecture 3 - Review of Methods to Solve ODE
Case I: = b2 4ac > 0, r1 = r2, y(x) = c1er1x + c2er2x is the generalsolution.Case II: = 0, r1 = r2, repeated roots Ly = a(r r1)2erx = 0. Then wehave one solution y(x) = er1x what about the second solution. Let
y(r, x) = erx
Ly(r, x) = a(r r1)2erxL
yr (r, x)
r=r1
= [2a(r r1)erx + 2a(r r1)xerx]r=r1 = 0Therefore
yr (r, x)r=r1 = xer1x is also a solution.(3.15)
Thus y(x) = c1er1x + c2xe
r1x is the general solution.
Another Method:
Consider a small perturbation to the double root case:
r (r1 + )
r (r1 )
= 0 (3.16)
y(x) = c1e(r1+)x + c2e
(r1)x
= e(r1+)xe(r1)x
2 c1 =1
2 = c2= er1x
exex
2
0 xer1x = r erxr=r1
(3.17)
Case III: Complex Conjugate Roots: = b2 4ac < 0
r = b2a
i4ac b21/2 = iy(x) = c1e
(+i)x + c2e(i)x (3.18)
= ex [A cos x + B sin x] .
EG 1:
Ly = y + y 6y = 0y = erx(r2 + r 6) = (r + 3)(r 2) = 0 (3.19)
y(x) = c1e3x + c2e2x
22
8/8/2019 Math 316 Notes
23/179
3.4. EULER/EQUIDIMENSIONAL EQUATIONS:
EG 2:
Ly = y + 6y + 9y = 0y = erx(r + 3)2 = 0 (3.20)
y(x) = c1e3x + c2xe3x
EG 3:
Ly = y 4y + 13y = 0y = erx : r2 4r + 13 = 0
r = 4
16522 = 2 3i
Therefore y(x) = e2n [A cos3x + B sin3x] .
(3.21)
3.4 Euler/Equidimensional Equations:
Ly = x2y + xy + y = 0. (3.22)
Aside: Note if we let t = ln x or x = et thend
dx=
d
dt
dt
dx d
dt= x
d
dx.
d2
dt2= x
d
dx
x
d
dx
= x2
d2
dx2+ x
d
dx x2 d
2
dx2=
d2
dt2 d
dt(3.23)
Therefore y y + y + y = 0y + ( 1)y + y = 0 (3.24)
y = ert r2 + ( 1)r + = 0 Characteristic Eq.Back to (3.22): Guess y = xr, y = rxr1, and y = r(r 1)xr2.
Therefore {r(r 1) + r + } xr = 0f(r) = r2 + ( 1)r + = 0 as above. (3.25)
r =1
( 1)2 42 (3.26)Case 1: = ( 1)2 4 > 0 Two Distinct Real Roots r1, r2.
y = c1xr1 + c2x
r2 if r1 or r2 < 0 |y| as x 0. (3.27)
23
8/8/2019 Math 316 Notes
24/179
Lecture 3 - Review of Methods to Solve ODE
Case 2: = 0 Double Root (r r1)2 = 0.
y = c1xr1
r L[x
r] = L
r x
r
= L[xr log x]
r {f(r)xr} = f(r)xr + f(r)xr log x = 0 since f(r) = (r r1)2.
(3.28)
General Solution: y(x) = (c1 + c2 log x)xr1 .
Check:
L(xr1 log x) = x2(xr log x) + x(xr log x) + (xr log x) = x
2 r(r 1)xr log x + rxr2 + (r 1)xr2 (3.29)+ x
rxr1 log x + xr1
+ (xr log x)
=
r2 + ( 1)r + xr log x + {2r 1 + } xr = 0Case 3: = ( 1)2 4 < 0.
r =(1 )
2 i [4 ( 1)
2]1/2
2= i
y(x) = c1x(+i) + c2x
(i) xr = er ln x
= c1e(+i) ln x + c2e
(i) ln x (3.30)
= x c1ei ln x + c2ei ln x= A1x
cos( ln x) + A2x sin( ln x)
Notes:
(1) Ifx < 0 replace by |x|.
(2)
w(y1, y2) =
y1 y2y1 y
2
= y1y2 y1y2 (see Section 3.2 p.143)
=
x cos( ln x)
log xx sin( ln x) + x1 cos( ln x)
x log x cos( ln x) x1 sin( ln x)
x sin( ln x)
= x21 independent for x = 0.
24
8/8/2019 Math 316 Notes
25/179
3.4. EULER/EQUIDIMENSIONAL EQUATIONS:
EG 1:
x2y xy 2y = 0 y(1) = 0 y(1) = 1y = xr r(r 1) r 2 = 0 r2 2r 2 = 0
(r 1)2 = 3 r = 1 3(3.31)
y = c1x1+
3 + c2x13
y(1) = c1 + c2 = 0 c2 = c1y(x) = c1
x1+
3 x1
3
(3.32)
y(x) = c1
1 +
3
x
3
1
3
x
3
x=1= c12
3 = 1
Therefore y(x) = 12
3x1+3 x13 . (3.33)
EG 2:
x2y 3xy + 4y = 0 y(1) = 1 y(1) = 0y = xr = r(r 1) 3r + 4 = r2 4r + 4 = 0 (r 2)2 = 0 (3.34)
y(x) = c1x2 + c2x
2 log x
y(1) = c1 = 1 y(x) = 2x + c2 [2x log x + x]=1 (3.35)
= 2 + c2 = 0
Therefore y(x) = x2 2x2 log x.
25
8/8/2019 Math 316 Notes
26/179
Lecture 3 - Review of Methods to Solve ODE
26
8/8/2019 Math 316 Notes
27/179
Chapter 4
Lectures 4,5 Ordinary Pointsand 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:
x0 is said to be an ordinary point of (5.2) if p(x) = Q/P and q(x) = R/Pare analytic at x0.
i.e. p(x) = p0 +p1(x x0) + =
k=0pk(x x0)k
q(x) = q0 + q1(x x0) + =
k=0qk(x x0)k
Note:
(1) If P, Q and R are polynomials then a point x0 such that P(x0) = 0 isan ordinary point.
27
8/8/2019 Math 316 Notes
28/179
Lectures 4,5 Ordinary Points and Singular Points
(2) Ifx0 = 0 is an ordinary point then we assume
y =
n=0
cnxn, yn =
n=1
cnnxn1, yn =
n=2
cnn(n 1)xn2
0 = Ly =
n=2
cnn(n 1)xn2 +
n=0
pnxn
n=1
ncnxn1 (4.3)
+
n=0
qnxn
n=0
cnxn
m=0(m + 2)(m + 1)cm+2 + p0(m + 1)cm+1 + +pmc1
+ (q0cm + + qmc0)} xm = 0 (4.4)yields a non-degenerate recursion for the cm.
At an ordinary point x0 we can obtain two linearly independent solu-tions 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 pointof (4.2). For example if P, Q and R are polynomials and P(x0) = 0 andQ(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
cnxn, y =
n=1
ncnxn1, y =
n=2
cnn(n 1)xn2 (4.7)
28
8/8/2019 Math 316 Notes
29/179
4.2. A SINGULAR POINT:
(x
1)
n=2
cnn(n
1)xn2 +
n=1
ncnxn1 = 0
n=2
cnn(n 1)xn2 +
n=2
cn {n(n 1) + n} xn1 + c1 = 0 (4.8)
m 1 = n 2 m = n 1 n = 2 m = 1 n = m + 1c2 2 1 + c1 +
m=2
cm+1(m + 1)m + cmm2xm1 = 0c0 Arbitrary:
cm+1 =m
m + 1cm m 2 c2 = c1
2
c3 =23
c2 =c13
c4 =34
c3 =c14
. . . cn =c1n
(4.9)
Therefore y(x) = c0 + c1
n=1
xn
n.
Recall
1
1 x = 1 + x + x2 +
1
1 x dx = ln |1 x| = x +x2
2+
x3
3+
y(x) = A + B ln |x 1| (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
8/8/2019 Math 316 Notes
30/179
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=0cnx
n, y =
n=1cnnx
n1, y =
n=2cnn(n 1)xn2
n=2
cnn(n 1)xn2 =
n=0cnx
n+1
m + 1 = n 2 n = m + 3 n = 2 m = 1c22x0 +
m=0
cm+3(m + 3)(m + 2) cmxm+1 = 0c2 = 0 cm+3 =
cm(m+3)(m+2) m = 0, 1, . . .
(4.12)
(1) c0 c3 c6.c3 =
c02.2
, c6 =c3
6.5=
c06.5.3.2
, c9 =c0
9.8.6.5.3.2
c3n =c0
(3n)(3n 1)(3n 3)(3n 4) . . . 9.8.6.5.3.2 (4.13)
y0(x) = 1 +x3
3.2+
x6
6.5+ + x
3n
(3n)(3n 1) . . . 3.2 + . . .
(2) c1
c4
c7
.
c4 =c1
4.3c7 =
c17.64.3
c10 =c1
(10.9)(7.6)(4.3)(4.14)
c3n+1 =c1
(3n + 1)(3n)(3n 2)(3n 3) . . . (7.6)(4.3)y1(x) = x +
x4
4.3+
x7
7.6.4.3+ + x
3n+1
(3n + 1)(3n) . . . 4.3(4.15)
y(x) = c0y0(x) + c1y1(x)
Radius of Convergence:
limmcm+3
cm |x|3
= lnm |x
|3
(m + 3)(m + 2) = 0 < 1 = . (4.16)See B&D for expansion of Airy Solution about x0 = 1 y(x) =
an(x 1)n.
It is useful to write x = (x 1) + 1.y = (x 1)y + y (4.17)
30
8/8/2019 Math 316 Notes
31/179
4.4. THE HERMITE EQUATION:
4.4 The Hermite Equation:
Ly = y 2xy + y = 0.Since x = 0 is an ordinary point let y(x) =
n=0
anxn then
Ly =
n=2
ann(n 1)xn2 2
n=1
annxn +
n=0
anxn = 0. (4.18)
m = n 2 n = m + 2 m n m nn = 2 m = 0
Thereforem=1
am+2(m + 2)(m + 1) 2amm + am
xm + [a22 + a0] x
0 = 0. (4.19)
x0 :a2 = a0/2 (4.20)
xm :
am+2 =(2m )am
(m + 1)(m + 2)m 1 (4.21)
a0:
a2 =
2 a0, a4 =(4
)
4.3 a2 =(4
)(
)
4.3.2 a0, a6 =(8
)(4
)(
)
6.5.4.3.2 a0
a2k =[4(k 1) ][4(k 2) ] . . . (1))?a0
(2k)!(4.22)
y0 = a0
1
2x2 +
( 4)4!
x4 +(8 )(4 )()
6!x6 +
a1:
a3 =(2 )
3.2a1; a5 =
(6 )5!
(2 )a1; a7 = (10 )(6 )(2 )7!
a1, . . .(4.23)
y1 = a1 x +(2 )
3!x3 +
(6 )(2 )5!
x5 +(10 )(6 )(2 )x7
7!+
The general solution is of the form
y(x) = Ay0(x) + By1(x) (4.24)
Note:
31
8/8/2019 Math 316 Notes
32/179
Lectures 4,5 Ordinary Points and Singular Points
(a) If = 2n then the recursion yields am+2 = 0 = am+4 = for m = n.Thus if n is an even integer then the series solution y0 will terminateand become a polynomial of degree n.
In this case:
y0(x) = a0
1 nx2 + n(n 2)22 x
4
4! n(n 2)(n 4) 2
3x6
6!+
+ (1)n/2n(n 2) . . . 2.2n/2xnn!
. (4.25)
On the other hand ifn is an odd integer then the series solution y1(x)will terminate and become a polynomial of degree n. In this case
y1(x) = a1
x 2(n 1) x
3
3!+ 22(n 1)(n 3) x
5
5!
(n 1)(n 3)(n 5)23 x7
7!+ (4.26)
+ (n 1)(n 3) . . . 3.1(2) (n1)2 xn
n!
(b) For example in the special case = 4 = 2n then n = 2.
y0(x) = a0[1 2x2]. (4.27)
32
8/8/2019 Math 316 Notes
33/179
Chapter 5
Lecture 6 - Singular points
5.1 Radius of Convergence and Nearest SingularPoints
EG. 1: (1 + x2)y + 2xy + 4x2y = 0.
(1) If we were given y(0) = 0 and y(0) = 1 then we would want a powerseries expansion of the form
y =
n=0
cnxn about x0 = 0. (5.1)
Roots of 1+ x2 = 0 are x = i, so we expect the radius of convergenceof the TS for 1
1+x2to be 1 since
11+x2 = 1 x2 + x4 1
liman+2an = 1 = 1.
(5.2)
(2) If we were given y(1) = 1, y(1) = 0 then a power series expansion ofthe form
cn(x 1)n is required. In this case =
2.
EG. 2: (x 1)(2x 1)y + 2xy 2y = 0.x = 0 is an ordinary point. x = 1 and x = 12 are singular points. Onesolution of this equation is
y(x) =1
x 1 = (1 + x + x2 + ) = 1. (5.3)
33
8/8/2019 Math 316 Notes
34/179
Lecture 6 - Singular points
This TS solution about the ordinary point x = 0 converges beyond the
singular point x =12 .
EG: (x2 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 1y +
5(x 1)y + 3y = 0.Let t = x 1 so that ddt = ddx and the equation is transformed to
(t2 1)y + 5ty + 3y = 0y =
n=0
cntn, y =
n=1
cnntn1, y =
n=2
cnn(n 1)tn2
n=2n(n 1)cntn
n=2
n(n 1)cntn2 + 5
n=1ncnt
n + 3
n=0cnt
n = 0
m = n
2 n = m + 2 n = 2 = m = 0
m=2
[cm+2(m + 2)(m + 1) + {m(m 1) + 5m + 3} cm] tm
2c2 + 3c0 + [c33.2 + 5c1 + 3c1] t = 0
t0 > c2 =32 c0
t1 > c3 =86 c1 =
43 c1
tm > cm+2 =cm(m+1)(m+3)
(m+1)(m+2) m 2.
(5.4)
c0:c4 =
5c24 =
54
32
c0, c6 =
76 c4 =
76
54
32 c0
y0(x) =
n=0
357...(2n+1)246...(2n) (x 1)2n
(5.5)
c1:
c5 =65 c3 =
65
43 c1 c2n+1 =
46...2n+235...2n+1 c1
y1(x) =
n=0
46...2n+235...2n+1 (x 1)2n+1
limncm+2
cm =
m + 3
m + 1 = 1 = 1 (5.6)
y(x) = c0y0(x) + c1y1(x)y(1) = c0 = 7 y
(1) = c1 = 3.(5.7)
34
8/8/2019 Math 316 Notes
35/179
5.2. SINGULAR POINTS:
5.2 Singular Points:
Consider
P(x)y + Q(x)y + R(x)y = 0. (5.8)
IfP, Q and R are polynomials without common factors then singular pointsare points x0 at which P(x0) = 0.
Note: At singular points the solution is not necessarily analytic.
Examples:
1.x2y + xy = 0y = xr
r(r
1) + r = 0
y = c1 + c2 ln x
The x2y admits wild behaviour.
2.x2y 2y = 0y = xr r(r 1) 2 = 0 r = 2, 1 y = c1x2 + c2x1
Again the x2y admits wild behaviour.
3.x2y 2xy + 2y = 0y = xr r(r 1) 2r + 2 = 0 r = 1, 2 y = c1x + c2x2
In this case both solutions are analytic.
5.3 Regular Singular Points:
Notice that all these cases are equidimensional equations for which we canidentify solutions of the form xr or xr log x. There is a special class ofsingular points called regular singular points in which the singularities areno worse than those in the equidimensional equations.
y +
xy +
x2y = 0. (5.9)
If P, Q and R are polynomials and suppose P(x0) = 0 then 0 is a regularsingular point if
limxx0
(x x0) Q(x)P(x)
and limxx0
(x x0)2 R(x)P(x)
are finite. (5.10)
35
8/8/2019 Math 316 Notes
36/179
Lecture 6 - Singular points
I.E.Q(x)
P(x)
=p0
(x x0)+p1 +p2(x
x0) +
singularity no worse than 1xx0 (5.11)R(x)
P(x)=
q0(x x0)2 +
q1(x x0) + q2 +
singularity no worse than 1(xx0)2
Examples:
1.
(1 x2)y 2xy + 4y = 0P = 1
x2 P(
1) = 0 Q =
2x R = 4
limx1
(x 1) (2x)(1x)(1+x) = 1 limx1(x 1)2 4
(1+x)(1x) = 0 (5.12)
x = 1 is a R.S.P. (similarly for x = 1).2.
x3y y = 0P(x) = x3 Q = 0 R = 1
limx0 x
21
x3
=
(5.13)
Thus x = 0 is an irregular singular point. Actually y
x3/4e
2/x1/2 asx 0+ which is much wilder than the simple power law xr or xr log x.Note: Any singular point that is not a regular singular point is calledan irregular singular point.
3. 2(x 2)2xy + 3xy + (x 2)y = 0. Singular points at x = 0, 2. x = 0is a regular singular point. x = 2 is an irregular singular point.
5.4 More General Definition of a Regular SingularPoint:
If P, Q, and R are not limited to polynomials then consider
P(x)y + Q(x)y + R(x)y = 0or
x2y + x
xQP
y +
x2RP
y = 0
(5.14)
36
8/8/2019 Math 316 Notes
37/179
5.4. MORE GENERAL DEFINITION OF A REGULAR SINGULARPOINT:
x = 0 is a regular singular point ifxQP and x2R
P are analytic atx = 0. I.E.xQ
P= p(x) = p0 +p1x + and x
2R
P= q(x) = q0 + q1x + . (5.15)
In this case
Ly = x2y + xp0y + q0y +
small as x0 xp1xy
+ q1y +
= 0. (5.16)
Then as x 0 x2y + xp0y + q0y 0 which is an Euler Equation whichhas solutions of the form y = xr.
Thus about a regular singular point we look for solutions of the form
y = xr
n=0
anxn =
n=0
anxn+r.
Our task is to determine;
(i) r
(ii) the coefficients an
(iii) the radius of convergence.
EG. 1: x2y + 2(ex 1)y + ex cos xy = 0 P = x2 Q = 2(ex 1) R =e
x
cos x.x = 0 is a singular point.
limx0
xQ
P= lim
k0x
2(ex 1)x2
= limx0
2(ex 1)x
00= lim
x02ex
1= 2 LHopital
limx0
x2R
P= lim
x0x2
ex cos xx2
= 1 < . (5.17)
Since the quotient functions p = xQ/P and q = x2R/P have Taylor Expan-sions about x = 0, x = 0 is a regular singular point.
37
8/8/2019 Math 316 Notes
38/179
Lecture 6 - Singular points
38
8/8/2019 Math 316 Notes
39/179
Chapter 6
Lecture 7 - Frobenius Seriesabout Regular SingularPoints
Example 1:
Ly = 2x2y xy + (1 + x)y = 0 x = 0 is a RSP.
y =
n=0
anxn+r (6.1)
Ly = 2x2 n=0
an(n + r)(n + r 1)xn+r2 x n=0
an(n + r)xn+r1
+ (1 + x)
n=0
anxn+r = 0
n=0
an {2(n + r)(n + r 1) (n + r) + 1} xn+r
+
n=0
anxn+r+1 = 0 (6.2)
m = n + 1 n = 0
m = 1
n = m 1Therefore a0 {2r(r 1) r + 1} xr +
n=1
[an {2(n + r)(n + r 1)
(n + r) + 1} + an1] xn+r = 0.
39
8/8/2019 Math 316 Notes
40/179
Lecture 7 - Frobenius Series about Regular Singular Points
xr > Indicial Equation: 2r2 3r + 1 = (2r 1)(r 1) = 0 r = 12 , r = 1.a0 arbitraryRecursion
an =an1
(2n + 2r 3)(n + r) + 1 (6.3)
Let r = 1/2:
an =an1
(2n 2)(n + 1/2) + 1 =an1
(n 1)(2n + 1) + 1 =an1
n(2n 1)n = 1 : a1 =
a01
; n = 2 : a2 =a12.3
=+a02.3
a3 =a23.5
=a0
1.(2.3)(3.5); a4 =
a34.7
=+a0
1(2.3)(3.5)(4.7)(6.4)
an =(1)na0
n!1.3.5.(2n 1) =(1)n2(n1)a0
n(2n 1)!
y1(x) = x1/2
n=0
(1)n2(n1)n(2n 1)! x
n
r = 1:
an =an1
(2n 1)(n + 1) + 1 =an1
(2n + 1)n
a1 =a03.1
, a2 =a15.2
=+a0
(1.3)(2.5); a3 =
a23.7
=a0
(1.3)(2.5)(3.7)
an = (1)na0n!3.5.7.(2n + 1)
= (1)n2na0(2n + 1)!
(6.5)
y2(x) = x
n=0
(1)n2n(2n + 1)!
xn
General Solution: y(x) = c1y1(x) + c2y2(x)Radius of Convergence .
40
8/8/2019 Math 316 Notes
41/179
6.1. SERIES EXPANSION SUMMARY:
6.1 Series Expansion Summary:
ConsiderP(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: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 solutionsof 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 analyticin which case x0 is a singular point.Regular Singular Points:
A point x0 is a regular singular point if
(x x0) Q(x)P(x)
= p0 +p1(x x0) +
(x x0)2 R(x)P(x)
= q0 + q1(x x0) + (6.10)
are analytic at x0. In this case
(x x0)2y + (x x0)
(x x0) Q(x)P(x)
y + (x x0)2 R(x)
P(x)y = 0 (6.11)
has singularities no worse than the Euler Equation:
(x
x0)
2y + (x
x0)p0y
+ q0y = 0. (6.12)
In this case we look for solutions of the form
y(x) = (x x0)r
n=0
an(x x0)n. (6.13)
41
8/8/2019 Math 316 Notes
42/179
Lecture 7 - Frobenius Series about Regular Singular Points
42
8/8/2019 Math 316 Notes
43/179
Chapter 7
Bessels Equation
Lecture 8
7.1 Bessels Function of Order / {. . . , 2, 1, 0, 1, 2 . . .}:Ly = x2y + xy + (x2 2) y = 0 (7.1)
x = 0 is a regular Singular Point: therefore let y =
n=0
anxn+r.
0 =
n=0
an
(n + r)(n + r 1) + (n + r) 2xn+r + n=0
anxn+r+2 (7.2)
m = n + 2 n = m 2n = 0 m = 2
0 =
m=2
am
(m + r)2 2 + am2 xm+r + a0 r2 2xr (7.3)+ a1
(1 + r)2 2
xr+1
xr > a0 = 0 r = Indicial Eq. Rootsxr+1 > a1
(1 )2 2 = a1(1 2) = 0 provided = 12 .
xm+r > am = am2(m + r)2 2 m 2
(7.4)
43
8/8/2019 Math 316 Notes
44/179
Bessels Equation
r = :
am = am2(m + )2 2 =
am2m2 + 2m
= am2m(m + 2)
a2 = a02(2 + 2)
= a022(1 + )
a4 = a24(4 + 2)
=(1)2a0
2.24(2 + )(1 + )
. . . a2m =(1)ma0
m!22m(1 + ) . . . (m + )(7.5)
y1(x) = x
m=0
(1)m(x/2)2mm!(1 + )(2 + ) . . . (m + )
x0 0
r = :
am = am2m(m 2)
a2 = a02(2 2) =
a022(1 ) , a4 =
a24(4 2) =
(1)2a0224(1 )(2 )
. . . a2m =(1)ma0
m!22m(1 ) . . . (m ) (7.6)
y2(x) = x
m=0
(1)m(x/2)2mm!(1 ) . . . (m )
x0
7.2 Bessels Function of Order = 0 - repeatedroots:
Ly = x2y + xy + x2y = 0.
y =
n=0
anxn+r
Ly =
n=0
an
(n + r)(n + r 1) + (n + r)xn+r + anxn+r+2 = 0m = n + 2 n = m 2 (7.7)
0 = n=2
an(n + r)2 + an2xn+r + a0r(r 1) + rxr + a1(r + 1)r + r + 1xr+1 = 0The indicial equation is: a0r
2 = 0 r1,2 = 0, 0 a double root.
r1 = 0 a1.1 = 0 a1 = 0.
44
8/8/2019 Math 316 Notes
45/179
7.2. BESSELS FUNCTION OF ORDER = 0 - REPEATED ROOTS:
Recursion: an = an2(n + r)2
n 2.
a2 = a022
; a4 = a242
=a0
2242; a6 = a4
62= a0
224262; a8 =
a022426282
(7.8)
a2m =(1)m
22m(m!)2a0 (7.9)
y1(x) =
1 +
m=1
(1)mx2m22m(m!)2
= J0(x)
0 5 10 15 204
3
2
1
0
1
x
J
0(x)and
Y0
(x)
J0
Y
0
Figure 7.1: Zeroth order bessel functions j0(x) and Y0(x)
To get a second solution
y(x, r) = a0xr
1 x2
(2 + r)2+
x4
(2 + r)2(4 + r)2+ + (1)
mx2m
(2 + r)2(4 + r)2 . . . (2m + r)2
+
(7.10
y
r(x, r)
r=r1
= a0 log xy1(x) + a0xr
m=1
(1)mx2m r
1
(2 + r)2 . . . (2m + r)2
.
45
8/8/2019 Math 316 Notes
46/179
Bessels Equation
Let
a2m(r) = { } ln a2m(r) = 2 ln(2 + r) . . . 2ln(2m + r) (7.11)a2m(0) =
2
2 + r 2
4 + r 2
(2m + r)
r=0
a2m(0)
=
1 1
2 . . . 1
m
a2m(0) = Hma2m(0).
Let Hm = 1 +1
2+ + 1
m. Therefore
y2(x) = J0(x) ln x +
m=1(1)m+1Hm
22m(m!)2x2m x > 0. (7.12)
It is conventional to define
Y0(x) =2
y2(x) + ( log 2)J0(x)
(7.13)
where
= limn(Hn log n) = 0.5772 Eulers Constant
y(x) = c1J0(x) + c2Y0(x). (7.14)
Lecture 9
7.3 Bessels Function of Order = 12:
Consider the case = 1/2 Ly = x2y + xy +
x2 14
y = 0. Let
y =
n=0
anxn+r (7.15)
Ly = n=0
an
(n + r)2 14
xn+r + n=0
anxn+r+2 = 0m = n + 2n = m 2n = 0 m = 2
(7.16)
Ly = a0
r2 1
4
+ a1
(r + 1)2 1
4
+
n=2
an
(n + r)2 1
4
+ an2
xn+r = 0.
46
8/8/2019 Math 316 Notes
47/179
7.3. BESSELS FUNCTION OF ORDER = 12 :
Indicial Equation: r2
1
4
= 0, r =
1
2
Roots differ by an integer.
Recurrence: an = an2(n + r)2 14
n 2.
r1 = +1/2:
an = an2(n + 12 )
2 14= an2
(n + 1)nn 2;
9
4 1
4
a1 = 0 a1 = 0
a2 = a03.2
a4 =(1)2a05.4.3.2
. . . a2n =(1)na0(2n + 1)!
y1(x) = x12
n=0(1)nx2n(2n + 1)!
= x12
n=0(1)nx2n+1
(2n + 1)!= x
12 sin x
(7.17)
r2 = 12
:
an = an2(n 12 )2 14
= an2n(n 1) , n 2,
n = 1 a1
12
+ 1
2 1
4
= a1.0 = 0 a1 and a0 arbitrary.
(7.18)
a0:
a2 = a02.1
a4 =(1)2a04.3.2.1
. . . a2n =(1)na0
(2n)!(7.19)
a1:
a3 = a13.2
a5 =(1)2a15.4.3.2
a2n+1 =(1)na1(2n + 1)!
(7.20)
y2(x) = a0x 12 n=0
(1)n
x2n
(2n)!+ a1x 12
n=0
(1)n
x2n+1
(2n + 1)!
= a0x 1
2 cos x + a1x 1
2 sin x (7.21)
included in y1(x).
47
8/8/2019 Math 316 Notes
48/179
Bessels 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
cnxn+
n=0
cn(n + )(n + 1)xn+1 cnxn+ = 0
(7.22)p 1 = n
n=1 {cn(x + )(n + 1) cn1} xn+1 + c0( 1)x1 = 0Indicial Equation: ( 1) = 0 = 0, 1 differ by integer.Recurrence Rel: cn =
cn1(n + )(n + 1) n 1.
Note: When = 0, c1 blows up!
Let = 1 c1 = c02
, c2 =c012
, . . ..
y1(x) = c0x
1 +
x
2+
x2
12+
= c0u1(x). (7.23)
Second Solution:
y(x, ) = y(x, ) = c0x
+
x
1 + +
x2
(1 + )(2 + )(1 + )+
y
= c0x
ln x
+
x
1 + +
(7.24)
+ c0x
1 x
(1 + )2 x
2
(1 + )2(2 + )
2
(1 + )+
1
(2 + )
+
y
=0
= c0
x +
x2
2+
x3
12+
ln x + c0
1 x 5
4x2
= c0u2.
Therefore y(x) = (A + B ln x)x +x2
2
+x2
12
+
+ B 1 x 5
4
x3
.
48
8/8/2019 Math 316 Notes
49/179
Chapter 8
Separation of Variables
Lecture 10
8.1 Types of Boundary Value Problems:
Dirichlet Boundary Conditions
1. Heat Equation: 2 = Thermal Conductivity.
Heat Flow in a Bar Heat Flow on a Disk
2. Wave Equation: c = Wave Speed.
Vibration of a String
3. Laplaces Equation:
49
8/8/2019 Math 316 Notes
50/179
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
8/8/2019 Math 316 Notes
51/179
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 heldat zero degree for all time and within which the initial temperature is givenby f(x).
Fouriers Guess:
u(x, t) = X(x)T(t) (8.1)
ut = X(x)T(t) = 2uxx =
2X(x)T(t)
2XT:X(x)X(x)
=T(t)
2T(t)= Constant = 2. (8.2)
>
T(t) = 22T(t) dTT
= 22 dtln |T| = 22t + c
T(t) = De22t.
(8.3)
x >
X(x) + 2X(x) = 0Guess X(x) = erx (r2 + 2)erx = 0 r = i (8.4)
X = c1eix + c2?e
ix
= A sin x + B cos x.(8.5)
Impose the boundary conditions:
0 = u(0, t) = X(0)T(t) = BT(t) B = 00 = u(L, t) = X(L)T(t) = (A sin L)T(t).
(8.6)
51
8/8/2019 Math 316 Notes
52/179
Separation of Variables
Now we do not want the trivial solution so A = 0. Thus we look for valuesof such that
sin L = 0 =n
L
n = 1, 2, . . . . (8.7)
Thus un(x, t) = e2(nL )
2t sin
nxL
n = 1, 2, . . .
are all solutions of ut = 2uxx. (8.8)
Since (8.8) (above eq. number) is linear, a linear combination of solutionsis again a solution. Thus the most general solution is
u(x, t) =
n=1
bn sinnx
L
e
2(nL )2
t. (8.9)
What about the initial condition u(x, 0) = f(x).
u(x, 0) = f(x) =
n=1
bn sinnx
L
. (8.10)
Given f(x) we need to find the bn such that the infinite series of functionsbn sin
nxL
agrees with f on [0, L].
Question: f(x) may not be periodic f(x + 2L) = f(x) but the series isperiodic since sin
nL
(x + 2L) = sin
nxL
.
Answer: In fact they do agree on [0, L] and are different elsewhere.
52
8/8/2019 Math 316 Notes
53/179
8.2. SEPARATION OF VARIABLES - FOURIER SINE SERIES:
Lecture 11
How do we find the bn?
Observe that we have a new type of eigenvalue problem subject toX(0) = 0 X(L) = 0. Just as in the case with matrices we obtain sequenceof eigenvalues which in this case is infinite:
n =n
L
n = 1, 2, . . . (8.11)
and corresponding eigenfunctions
xn(x) = sin nx = sinnx
L
sinx
L
, sin
2x
L
, sin
3x
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
v1, v2, v3
f = 1v1 + 2v2 + 3v3f vk = 1v1 vk + 2v2 vk + 3v3 vk v1 v1 v1 v2 v1 v3v1 v2 v2 v2 v2 v3
v1 v3 v2 v3 v3 v3
12
3
=
f v1f v2
f v3
(8.13)
If vk v, k = i.e. the vk are orthogonal
k =f vk
vk vk (8.14)
But functions are just infinite dimensional vectors:
53
8/8/2019 Math 316 Notes
54/179
Separation of Variables
f [f1, f2, . . . , f N]g [g1, g2, . . . , gN]
f g = f1g1 + f2g2 + + fNgN x = LN
(8.15)
=N
k=1
f(xk)g(xk).
NowN
k=1
f(xk)g(xk)x L
0
f(x)g(x) dx = f, g. (8.16)
Back to finding bn:
f(x) =
n=1
bn sinnx
L
(8.17)
L0
f(x)sin
kx
L
dx =
n=1
bn
L0
sinnx
L
sin
kx
L
dx.
Recall sin(A)sin B =1
2{cos(A B) cos(A + B)}. Therefore
Ink =
L
0
sinnxL sinkx
L dx=
1
2
L0
cos(n k) xL
cos(n + k) xL
dx n = k
=1
2
sin(n k)x/L
(n k)/L sin(n + k)x/L
(n + k)/L
L0
= 0 (8.18)
Inn =
L0
sin2
nx
L dx =
1
2
L0
1 cos
2nx
L
dx
= L/2
Therefore bk =2
L
L0
f(x)sin
kx
L
dx.
54
8/8/2019 Math 316 Notes
55/179
8.2. SEPARATION OF VARIABLES - FOURIER SINE SERIES:
Example 8.1
f(x) = 2x 0 < x < 12 L = 1
2(1 x) 12 < x < 1
bn = 2
12
0
2x sin(nx) dx +
112
2(1 x)sin(nx) dx
= 8sin(n/2)
n22n = 1 2 3 4 5
sin
nL
1 0 1 0 1
Therefore u(x, t) =8
2
k=0
(1)k(2k + 1)2
sin
(2k + 1)x
e(2k+1)
22t. (8.19)
Observe as t u(x, t) 0 (all the heat leaks out).
u(x, 0) = 82
k=0
(1)k(2k + 1)2
sin
(2k + 1)x
.
2
8=
k=0
1
(2k + 1)2by letting x =
1
2 f(x) = 1.
2 1 0 1 21
0.5
0
0.5
1
1 terms of the Fourier Series
x
f(x)
2 1 0 1 21
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
bn = 2
10
x sin(nx) dx = 2 cos(n)n
= 2 (1)n+1n
Therefore u(x, t) =2
n=1
(1)n+1n
sin(nx)e(n)2t. (8.20)
55
8/8/2019 Math 316 Notes
56/179
Separation of Variables
As t u(x, t) 0.
u(x, 0) = 2
n=1
(1)n+1n
sin(nx).
u
1
2, 0
=
1
2= 2
n=1
(1)n+1n sin(n/2)
=2
k=0
(1)k(2k + 1)
4= 1 1
3+
1
5 . . . .
k n sin
n2
0 1 1
2 01 3 1
4 02 5 1
(8.21)
2 1 0 1 21
0.5
0
0.5
1
1 terms of the Fourier Series
x
f(x)
2 1 0 1 21
0.5
0
0.5
1
2 terms of the Fourier Series
x
f(x)
56
8/8/2019 Math 316 Notes
57/179
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.
ut = 2uxx (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)
2T(t)= 2.
IVP:T(t)
2T(t)= 2 T(t) = ce2t.
r =2L
2=
L
= Constant.
The Laplacian becomes
u =2u
r2+
1
r
u
r+
1
r22u
2
=2u
(r)2(8.23)
57
8/8/2019 Math 316 Notes
58/179
Separation of Variables
if we let x = r we obtain (1.1).
BVP:X +
2
X = 0X(L) = X(L)X(L) = X(L)
Eigenvalue Problemlook for such thatnontrivial x can be found.
X(x) = A cos(x) + B sin(x)
X(L) = A cos(L) B sin(L) = A cos(L) + B sin(L) = X(L)therefore 2B sin(L) = 0
X(x) = A sin x + B cos(x) (8.24)X(L) = +A sin(L) + B cos(L) = A sin(L) + B cos(L) = X(L)
therefore 2A sin(L) = 0
Therefore nL = (n) n = 0, 1, . . . .Solutions to (1.1) that satisfy the BC are thus of the form
un(x, t) = e(nL )
22t
An cos
nxL
+ Bn sin
nxL
. (8.25)
Superposition of all these solutions yields the general solution
u(x, t) = A0 +
n=1
An cos
nxL
+ Bn sin
nxL
e(
nL )
22t. (8.26)
In order to match the IC we have
f(x) = u(x, 0) = A0 +
n=1
An cosnx
L
+ Bn sin
nxL
. (8.27)
As before we obtain expressions for the An and Bn by projecting f(x) onto
the basis functions sinnx
L
and cos
nxL
.
LL
f(x)
sin
mx
L
cos
mx
L
dx = A0L
L
sin
mx
L
cos
mx
L
dx (8.28)
+
n=1 A
n
L
L
cosnxL sin mxL cos mxL dx
+
n=1
Bn
LL
sinnx
L
sin mxL cos
mx
L
dx.58
8/8/2019 Math 316 Notes
59/179
8.3. HEAT EQ ON A CIRCULAR RING - FULL FOURIER SERIES
As in the previous example we use the orthogonality relations:
LL
sinmx
L
sin
nxL
dx = Lmn
LL
cosmx
L
cos
nxL
dx = Lmn m and n = 0 (8.29)
= 2L m = n = 0L
L
sinmx
L
cos
nxL
dx = 0 m,n.
Plugging these orthogonality conditions into (1.6) we obtain
A0 =1
2L
LL
f(x) dx = average value off(x) on [L, L]
An =1
L
LL
f(x)cosnx
L
dx and Bn =
1
L
LL
f(x)sinnx
L
dx.
(8.30)
Note:
1. (1.6) and (1.9) [typists note: check re-numbering when equations arere-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 an =1
L
LL
f(x)cosnx
L
dx =
2A0
An
and bn = Bn
the Fourier Series (1.6) is often written in the form
f(x) =a02
+
n=1 an cosnx
L + bn sinnx
L . (8.31)
59
8/8/2019 Math 316 Notes
60/179
Separation of Variables
60
8/8/2019 Math 316 Notes
61/179
Chapter 9
Lecture 13 - Fourier Series
We consider the expansion of the function f(x) of the form
f(x) a02
+
n=1
an cosnx
L
+ bn sin
nxL
= S(x) (9.1)
where
an =1
L
LL
f(x)cosnx
L
dx
a02
=1
2L
LL
f(x) dx = average value off.
bn = 1L
LL
f(x)sinnx
L
dx (9.2)
Note:
1. Note that cosn
L(x + )
= cos
nxL
provided
n
L= 2, =
2L
nand similarly sin
nL
(x + 2L)
= sinnx
L
. Thus each of the terms
of the Fourier Series S(x) on the RHS of (10.1) is a periodic functionhaving 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 periodicextension of f on [L, L].
61
8/8/2019 Math 316 Notes
62/179
Lecture 13 - Fourier Series
2. Iff (or its periodic extension) is discontinuous at a point x0 then S(x)
converges to the average value of f across the discontinuity.
S(x0) =1
2
f(x+0 ) + f(x
0 )
(9.3)
Example 9.1
f(x) =
0 < x < 0 L = x 0 x (9.4)
62
8/8/2019 Math 316 Notes
63/179
a0 =
1
f(x) dx =
1
0
x dx =
2 (9.5)
an =1
f(x)cos(nx) dx
=1
0
x cos(nx) dx
=1
xsin(nx)
n
0
1n
0
1. sin(nx) dx
=1
sinn
(n) + 1n2
cos(nx)
0
=
1
n2
(1)n 1 n 1 2 3 4(1)n 1 2 0 2 0 (9.6)
a2m+1 = 2(2m + 1)2
m = 0, 1, 2, . . . (9.7)
bn =1
f(x)sin(nx) dx
=
1
0
x sin(nx) dx
=1
x cos(nx)n
0
+1
n
0
1. cos(nx) dx
=1
cos(n)
n+
0. cos0
n+
1
n2sin(nx)
0
= (1)n+1/n (9.8)
f(x) =a02
+
n=1an cos(nx) + bn sin(nx)
=
4 2
m=0
cos (2m + 1)x(2m + 1)2
+
n=1
(1)n+1 sin(nx)n
(9.9)
63
8/8/2019 Math 316 Notes
64/179
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 fe(x) is periodic with period 2L (as are the
basis functions cosnx
L
and sin
nxL
).
an = 1L
LL
f(x)cosnxL
dx = 1L
c+2Lc
fe(x)cosnxL
dx (9.10)bn =
1
L
LL
f(x)sinnx
L
dx =
1
L
c+2Lc
fe(x)sinnx
L
dx. (9.11)
Example 9.2 Previous Example:
f(x) =
0 < x < 0x 0 x (9.12)
On [, 3]
fe(x) =
0 < x < 2x 2 2 x 3 (9.13)
an =1
3
fe(x)cos(nx) dx= 1
32
(x 2)cos(nx) dx
= 1
0
t cos(nt) dt.
t = x 2 dx = dtx = t + 2 x = t = x = 3 t =
since cos n(t + 2) = cos t
(9.14)
64
8/8/2019 Math 316 Notes
65/179
9.2. COMPLEX FORM OF FOURIER SERIES
9.2 Complex Form of Fourier Series
f(x) =a02
+
n=1
an cosnx
L
+ bn sin
nxL
cosnx
L
=
ei(nxL ) + ei(
nxL )
2; sin
nxL
=
ei(nxL ) ei(nxL )
2i
Therefore f(x) =a02
+
n=1
an2
ei(
nxL ) + ei(
nxL )
+
bn2i
ei(
nxL ) ei(nxL )
= a02 +
n=1
an ibn2 ei(nxL ) +an + ibn2 ei(nxL ) (9.15) c0 cn cn
=
n=
cnei(nxL )
cn = an ibn2
= 12L
LL
f(x)cosnxL
i sinnxL
dx (9.16)
=1
2L
LL
f(x)ei(nxL ) dx bn = bn (9.17)
Therefore
f(x) =
n=
cn
ei(nxL ) (9.18)
cn =1
2L
LL
f(x)ei(nxL ) dx. (9.19)
65
8/8/2019 Math 316 Notes
66/179
Lecture 13 - Fourier Series
Example 9.3
f(x) =
1 x < 01 0 < x <
L = (9.20)
cn =1
2
0L
einx dx +
0
einx dx
(9.21)
=1
2
e
inx0
(in) +einx
0
(in)
(9.22)
=i
2n 2 + e+in + ein
=
0 n even
(2/in) n odd(9.23)
Therefore
f(x) =
n=
2
i(2n + 1)ei
(2n+1)x
. (9.24)
66
8/8/2019 Math 316 Notes
67/179
Chapter 10
Lecture 14 - Even and OddFunctions
Even: f(x) = f(x)Odd: f(x) = f(x)
10.1 Integrals of Even and Odd Functions
LL
f(x) dx =
0L
f(x) dx +
L0
f(x) dx (10.1)
=
L0
f(x) + f(x) dx (10.2)
=
2L0
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. E1(x) E2(x) even.
67
8/8/2019 Math 316 Notes
68/179
Lecture 14 - Even and Odd Functions
3. O1(x) O2(x) even.
4. Any function can be expressed as a sum of an even part and an oddpart:
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
2f(x) + f(x)
. Then E(x) = 1
2f(x) + f(x)
=
E(x) even. Similarly let
O(x) =1
2
f(x) f(x) (10.5)
O(x) = 12
f(x) f(x) = O(x) odd. (10.6)
10.2 Consequences of Even/Odd Property for FourierSeries
(I) Let f(x) be Even-Cosine Series:
an =1
L
LL
f(x)cos even
nxL
dx =
2
L
L0
f(x)cosnx
L
dx(10.7)
bn =1
L
LL
f(x)sinnx
L
odd
dx = 0. (10.8)
Therefore
f(x) =a02
+
n=1
an cosnx
L
; an =
2
L
L0
f(x)cosnx
L
dx.(10.9)
68
8/8/2019 Math 316 Notes
69/179
10.2. CONSEQUENCES OF EVEN/ODD PROPERTY FOR FOURIERSERIES
(II) Let f(x) be Odd-Sine Series:
an =1
L
LL
f(x)cosnx
L
odd
dx = 0 (10.10)
bn =1
L
LL
f(x)sinnx
L
even
dx =2
L
L0
f(x)sinnx
L
dx
Therefore
f(x) =
n=1
bn sinnxL ; bn = 2LL
0f(x)sinnxL 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)
=
a02
+
n=1an cos
nx
L
+
n=1
bn sin
nx
L
where
an =2
L
L0
1
2
f(x) + f(x) cosnx
L
dx =
1
L
LL
f(x)cosnx
L
dx
bn =2
L
L0
1
2
f(x) f(x) sinnx
L
dx =
1
L
LL
f(x)sinnx
L
dx.
69
8/8/2019 Math 316 Notes
70/179
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 representf by a Fourier Series we have two choices - a Cosine Series or a Sine Series.
Cosine Series:
f(x) =a02
+
n=1
an cosnx
L
(10.12)
an =2
L
L0
f(x)cosnx
L
dx. (10.13)
Sine Series:
f(x) =
n=1
bn sinnx
L
(10.14)
bn =2
L
L0
f(x)sinnx
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
8/8/2019 Math 316 Notes
71/179
10.3. HALF-RANGE EXPANSIONS
bn =
2
0
f(t)sin
n
t dt (10.16)
=
20
t sinn
2t dt (10.17)
= t cosn2 t
n2
2
0
+2
n
20
cosn
2t dt (10.18)
= 4n
cos(n) +
2
n
2sin
n2
t
2
0(10.19)
= 4n
(1)n f(1) = 1 = 4 n=1 (1)n+1n sin n2 4 = 1 13 + 15 17 +
(10.20)
Therefore
f(t) =4
n=1
(1)n+1n
sinn
2t
. (10.21)
(b)
a0 =1
2
20
t dt =1
2
t2
2
20
= 1 (10.22)
an =
20
t cos n2
t dt = 2n t sin n
2t2
0 2
n 2
0
sin n2
t dt
= +
2
n
2cos
n
2t
2
0
=4
n22{cos n 1} (10.23)
71
8/8/2019 Math 316 Notes
72/179
Lecture 14 - Even and Odd Functions
Therefore
f(t) = 1 +4
2
n=1
(1)n 1
n2cos
n
2t (10.24)
= 1 82
n=0
cos(2n + 1)
2t/(2n + 1)2. (10.25)
The cosine series converges faster than Sine Series.
f(2) = 2 = 1 +8
2
n=0
1
(2n + 1)22
8= 1 +
1
32+
1
52+
72
8/8/2019 Math 316 Notes
73/179
Chapter 11
Lecture 15 - Convergence ofFourier Series
Example 11.1 (Completion of problem illustrating Half-range Expansions)Periodic Extension: Assume that f(x) = x, 0 < x < 2 represents one fullperiod of the function so that f(x + 2) = f(x). 2L = 2
L = 1.
a0 =1
L
LL
f(x) dx =
11
f(x) dx =
20
x dx =x2
2
20
= 2 (11.1)
since f(x + 2) = f(2). (11.2)
73
8/8/2019 Math 316 Notes
74/179
Lecture 15 - Convergence of Fourier Series
n 1:
an =1
L
LL
f(x)cosnx
L
dx =
11
f(x) cos(nx) dx L = 1
=
20
x cos(nx) dx
=
x sin(nx)
n
20
1
n
20
sin(nx) dx
=
1
(n)2 cos(nx)2
0=
1
(n)2 cos(2n) 1 = 0 (11.3)bn =
1
L
LL
f(x)sinnx
L
dx =
11
f(x)sin(nx) dx
=
20
x sin(nx) dx =
xcos(nx)
n
20
+1
(n)
20
cos(nx) dx
=2n
+sin(nx)
(n)2
2
0=
2n
(11.4)
Therefore
f(x) =2
2 2
n=1
sin(nx)
n(11.5)
= 1 2
n=1
sin(nx)
n
(11.6)
74
8/8/2019 Math 316 Notes
75/179
4 2 0 2 4
1
0
1
2
3
x
S(x)
Figure 11.1: Full Range Expansion SN(x) = 1 2N=20n=1
sin(nx)n
4 2 0 2 42
1
0
1
2
x
S(x)1
Figure 11.2: Full Range Expansion SN(x) 1 = 2N=20n=1
sin(nx)n
75
8/8/2019 Math 316 Notes
76/179
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 FourierSeries 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) = a02 +
n=1an cos
nxL
+ bn sin
nxL
= S(x)
where
an = 1L
LL
f(x)cos nxL dx and bn = 1L LL f(x)sin nxL dx.(11.7)
The Fourier Series converges to f(x) at all points at which f is continuous
and to1
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 andright limits at a point of discontinuity of the function f(x).
76
8/8/2019 Math 316 Notes
77/179
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 21.5
1
0.5
0
0.5
1
1.5
x/
SN
(x)forN=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, ]. (11.8)
f(x) = 1 =
n=1bn sin(nx)
where bn =2
0
sin(nx) dx = 2 cos nxn 0 = 2n 1 (1)n
=
4/n n odd0 n even
Therefore f(x) = 4
n=1
n odd
sin(nx)n =
4
m=0
sin(2m+1)x(2m+1) .
(11.9)
Note:
1. f(/2) = 1 =4
m=0
sin
(2m + 1)/2
(2m + 1)=
4
1 1
3+
1
5
. There-
fore
4= 1 1
3+
1
5 .
77
8/8/2019 Math 316 Notes
78/179
Lecture 15 - Convergence of Fourier Series
2. Recall the complex Fourier Series example for the function
f(x) =
1 x < 01 0 < x <
(11.10)
which turns out to be equivalent to the odd extension of the abovefunction represented by the half-range sine expansion, which we cansee from the following calculation
f(x) =
n=
n odd
2in e
inx = 4
n=1
n odd
einxeinx2in
= 4n=1
n odd
sin(nx)n .
(11.11)
11.3 Now consider the sum of the first N terms
SN(x) =4
Nm=0
sin(2m + 1)x
(2m + 1)=
4
Im
N
m=0
ei(2m+1)x
(2m + 1)
(11.12)
SN(x) =4
Im
N
m=0 iei(2m+1)x
(11.13)=
4
Im
ieix
Nm=0
ei2x
m(11.14)
=4
Im
ieix
1 + ei2x + + ei2xN
1 ei2x
(1 ei2x)
(11.15)
=4
Im
ieix
1 ei2(N+1)x
1 ei2x
(11.16)
=4
Im
i
1 ei2(N+1)x
eix
eix
(11.17)
=2
Im
ei2(N+1)x 1
sin x
(11.18)
=2
sin2(N + 1)x
sin x. (11.19)
78
8/8/2019 Math 316 Notes
79/179
11.3. NOW CONSIDER THE SUM OF THE FIRST N TERMS
Therefore
t = 2(N + 1)u du = dt2(N+1)
SN(x) =
2
x0
sin 2(N+1)usin u du 2
2(N+1)x0
sint t dt
(11.20)
0
0.5
1
1.
5
2
1
0
5 0 5 1
0
x/
(2/) sin(2(N+1)x)/sin(x)
Figure 11.4: (2/)sin(2(N + 1)x)/sin(x) for N = 5
Observe SN(x) =2
sin2(N + 1)x
sin x= 0 when 2(N + 1)xN = thus the
maximum value of SN(x) occurs at
xN =
2(N + 1)(11.21)
79
8/8/2019 Math 316 Notes
80/179
Lecture 15 - Convergence of Fourier Series
0
0.
5
1
1
.5
2
1.
51
0.
5 0
0.
5 1
1.
5
x/
(2/) 0
xsin 2(N+1)u /sin u du for N= 5
Figure 11.5: Integral of (2/)sin(2(N + 1)x)/sin(x)
80
8/8/2019 Math 316 Notes
81/179
Chapter 12
Lecture 16 - ParsevalsIdentity
Lemma 12.1 (A version of Parsevals Identity)
Letf(x) =
n=1
bn sinnx
L
0 < x < L. Then
2
L
L0
f(x)
2dx =
n=1
b2n.
Proof:
L
0
f(x)2 dx =
m=1
n=1
bmbn
L
0
sinmxL sinnxL dx (12.1)=
m=1
n=1
bmbn mn L2
=L
2
n=1
b2n. (12.2)
For a full Fourier Series on [L, L] Parsevals Theorem assumes the form:
f(x) =a02
+
n=1
an cosnx
L
+ bn sin
nxL
(12.3)
1
L
L
L
f(x)2 dx = a20
2 +
n=1 a
2
n + b2
n. (12.4)
Example 12.2 Recall for x [0, 2] f(x) = x = 4
n=1
(1)n+1n
sinnx
2
.
81
8/8/2019 Math 316 Notes
82/179
Lecture 16 - Parsevals Identity
Therefore
2L
L0
f(x)
2dx = 22
20
x2 dx = 4
2 n=1
1n2
x332
0=
4
2 n=1
1n2
2
6 =
n=1
1n2
(12.5)
Note:
n=1
1
(2n)2=
1
22
n=1
1
n2=
1
4
2
6
=
2
24.
Also note that
evens odds
2
6 = n=1
1n2 = m=1
1(2m)2 + m=0
1(2m+1)2
= 2
24 +
m=0
1(2m+1)2
Therefore
m=0
1
(2m + 1)2=
2
6
2
24=
2
8. (12.6)
12.1 Geometric Interpretation of Parsevals For-mula
f = b1e1 + b2e2 (12.7)
|f|2 = f f = b21e1 e1 + b22e2 e2 (12.8)= b21 + b
22 Pythagoras Theorem (12.9)
For Fourier Sine Components:
2
L
L0
f(x)
2dx =
n=1
b2n. (12.10)
82
8/8/2019 Math 316 Notes
83/179
12.1. GEOMETRIC INTERPRETATION OF PARSEVALS FORMULA
Example 12.3 Consider f(x) = x2 < x < . The Fourier SeriesExpansion is:
x2 =2
3+ 4
n=1
(1)nn2
cos(nx). (12.11)
n 1 2 3 4cos
n2
0 1 0 1
Let
x = 2 2
4 =2
3 + 4
n=1
(1)nn2 cos
n2
212 = 4
k=1
(1)k(2k)2
(12.12)
Therefore
2
12 =
k=1
(
1)k+1
k2 . (12.13)
By Parsevals Formula:
2
0
x4 dx = 2
2
3
2+ 16
n=1
1n4
2
x5
5
0
= 24
9 + 16
n=1
1n4
9545 =
445 =
890
190
(12.14)
Therefore
4
90=
n=1
1
n4= ?(4). (12.15)
83
8/8/2019 Math 316 Notes
84/179
Lecture 16 - Parsevals Identity
84
8/8/2019 Math 316 Notes
85/179
Chapter 13
Lecture 17 - Solving the heatequation using finitedifference methods
13.1 Approximating the Derivatives of a Functionby Finite Differences
Recall that the derivative of a function was defined by taking the limit of adifference quotient:
f(x) = limx0
f(x + x) f(x)x
. (13.1)
Now to use the computer to solve differential equations we go in the oppositedirection - we replace derivatives by appropriate difference quotients. If weassume that the function can be differentiated many times then TaylorsTheorem is a very useful device in determining the appropriate differencequotient to use. For example consider
f(x + x) = f(x) + xf(x) +x2
2!f(x) +
x3
3!f(3)(x) +
x4
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
2f(x) +
x2
3!f(3)(x) + . . . .
85
8/8/2019 Math 316 Notes
86/179
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 ismore useful to retain the approximation
f(x + x) f(x)x
= f(x) +x
2f() (13.3)
= f(x) + O(x).
We retain the termx
2f() 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 focuson 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) isO(x) if
E(x)
x
x0 const.
Now the difference quotient (3) is not the only one that can be used toapproximate f(x). Indeed if we consider the expansion of f(x x):
f(x x) = f(x) xf(x) + x2
2!f(x) x
3
3!f(3)(x) +
x4
4!f(4)(x) + . . . .(13.4)
and we subtract (4) from (2) and divide by (2x) we obtain:
f(x + x) f(x x)2x
= f(x) +x2
3!f(3)(). (13.5)
We notice that the error term associated with this form of difference ap-proximation is O(x2), which converges more rapidly to zero as x 0.
In order to obtain an approximation to f(x) we add (2) to (4) whichupon re-arrangement and dividing by x2 leads to:
f(x + x) 2f(x) + f(x x)x2
= f(x) +1
12x2f(4)(). (13.6)
Due to the symmetry of the difference approximations (5) and (6) about theexpansion point x these are called central difference approximations. Thedifference approximation (3) is known as a forward difference approximation.We note that the central difference schemes (5) and (6) are second orderaccurate while the forward difference scheme (3) is only O(x).
86
8/8/2019 Math 316 Notes
87/179
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= 2
2u
x20 < 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 itsneighbours 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) + tu
t(x, t) +
t2
2!
2u
x2(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) + xux
(x, t) + x22!
2ux2
(u, t) + x33!
3ux3
(x, t) + x44!
4ux2
(x, t) +
u(x x, t) = u(x, t) xux
(x, t) +x2
2!
2u
x2(x, t) x
3
3!
3u
x3(x, t) +
x4
4!
4u
x4(x, t) +
Adding and re-arranging:
u(x + x, t) 2u(x, t) + u(x x, t)x2
=2u
x2(x, t) + O(x2). (13.11)
Substituting (2) and (3) into (1a) we obtain
u(x, t + t) u(x, t)t
= 2 u(x + x, t) 2u(x, t) + u(x x, t)x2 + O(t, x2).Re-arranging:
u(x, t + t) = u(x, t) + 2
t
x2
{u(x + x, t) 2u(x, t) + u(x x, t)} .(13.12)
87
8/8/2019 Math 316 Notes
88/179
Lecture 17 - Solving the heat equation using finite difference methods
We subdivide the spatial interval [0, 1] into N + 1 equally spaced sample
points xn = nx. The time interval [0, T] is subdivided into M + 1 equaltime levels tk = kt. At each of these space-time sample points we introduceapproximations:
u(xn, tk) ukn.
u
u u
u
u u
u
uu
TTTTTT
-?
6
uk0 ukn1 u
kn u
kn+1 u
kN1 u
kN
tk+1
tk
uk+1Nuk+1nu
k+10
t
x
uk+1n = ukn +
2
t
x2
ukn+1 2ukn + ukn1
This is implemented in the spread sheets Heat0 and Heat.
88
8/8/2019 Math 316 Notes
89/179
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 tou
x(1, t), where xN = Nx =
1,
u(xN + x, t) u(xN x, t)x
= 0.
Re-arranging we obtain:
u(xN + x, t) = u(xN x, t) ()
Since xN = 1 we observe that xN+x is outside the domain we introducean extra column uN+1 into which we copy the values uN1. In the columnxN we implement the same difference approximation for the Heat Equation,namely:
uk+1N = ukN +
2(t
x2)(ukN+1 2ukN + ukN1) ()
This is implemented in the spread sheet Heat0f.while u
kN+1 = u
kN1 (see (*) ) since column u
kN1 is copied to column u
kN+1.
Note that this BC could be implemented another way without introducingthe additional column, by eliminating uN+1 from () and ():
uk+1N = ukN + 2
2
t
x2
ukN1 ukN
.
If this latter equation is implemented at xN there is no need to introducean extra column UN+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 whatis known as a numerical instability. Now change t to 0.00625, whichis known as the stability boundary and observe what happens. Nowlet t = 0.006 and observe the abrupt change in the solution - it ismuch closer to what we would expect.
89
8/8/2019 Math 316 Notes
90/179
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 fora parameter choice that is unstable, the other type of error is presentin almost every type of numerical approximation scheme. This class oferror results from discarding the O(x2) and O(t) terms in (2) and(3) when we replace derivatives in (1a) by difference quotients. Thiserror is known as the truncation error. To determine the truncationerror change the spread sheet to implement the initial condition
f(x) =
2x 0 < x < 1/22(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 tothe exact Fourier Series solution with 50 terms. The difference be-tween the two is mainly due to the truncation error since the round-offerror is about 1012 and does not grow if stable parameters are used.
3. Implement derivative boundary conditions on both endpoints x = 0 andx = 1. Check the numerical solution against the problem solved inHW1 #3.
90
8/8/2019 Math 316 Notes
91/179
Chapter 14
Lecture 18 - Solving
Laplaces Equation usingfinite differences
14.1 Finite Difference approximation
Consider the boundary value problem
2u
x2+
2u
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 = 1y0
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
8/8/2019 Math 316 Notes
92/179
Lecture 18 - Solving Laplaces 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)x2
=2u
x2(x, y) + O(x2)(14.3)
u(x, y + y) 2u(x, y) + u(x, y y)y2
=2u
y2(x, y) + O(y2).(14.4)
We partition the interval 0 x 1 into (N + 1) equally spaced nodesxn = nx and the interval 0 y 1 into (M + 1) equally spaced nodesym = my. Replacing the derivatives in (1a) by the difference quotients in(3) and (4) and representing the mesh values at (xn, ym) by unm u(xn, ym)we obtain:
un+1m 2unm + un1mx2
+unm+1 2unm + unm1
y2= (uxx + uyy)(xn,xm) + O(x
2, y2).
If we choose x = y then we obtain
un+1m + un1m + unm+1 + unm1 4unm = 0 1 n, m (N 1), (M 1).(14.5)
u
u u
u
u jj
j
j
j
1 1
1
1
-4
un1m unm un+1m
unm1
unm+1
This is known as the finite difference Stencil that relates unm to its 4nearest neighbours.
This is a system of (N 1) (M 1) unknowns for the values of unminterior to the domain - recall the boundary values are already specified!
92
8/8/2019 Math 316 Notes
93/179
14.2. SOLVING THE SYSTEM OF EQUATIONS BY JACOBIITERATION
14.2 Solving the System of Equations by Jacobi
Iteration
This is a procedure to solve the system of Equation (3) by looping througheach of the mesh points and updating unm according to (3) assuming thatthe nearest neighbours already have values close to the exact solution. Thisprocedure is repeated until the changes that are made in each iteration fallsbelow a certain tolerance.
To implement this iterative procedure we observe that the discrete LaplaceEquation (5) can be re-written in the form:
u
k+1
nm =
ukn+1m + ukn
1m + u
knm+1 + u
knm
1
4 (14.6)
t t t
t
t
- ?
6 average
Thus unm is the average value of its nearest neighbours. Note that a newsuperscript index k has been introduced to represent the nodal values at thekth iteration. Thus iteration can be viewed as taking successive neighbouraverages until there is no change, at which point the value of umn equalsthe average of the values at its mesh neighbours. This mean value propertyis a discrete form of a fundamental property of any solution to LaplacesEquation.
To implement the iterative procedure (6) on a spread sheet, go to theTools Menu at the top of the screen and click on the Options Tab. Thenselect the Calculation Tab. Check the Iteration box. If you set the numberof iterations to 5 say, then if you start with zero values throughout theinterior of the domain (as you should if you cut and paste as demonstratedin class), you will see the values percolate 5 cells into the domain from thenon zero boundary condition f(x) = sin(x). You can choose a surface plot
to visualize the solution. Now hold down the F9key and watch the solutionmove to equilibrium. This iterative process essentially uses diffusion on apseudo time scale to take the solution to equilibrium.
EXERCISES and Notes for Laplaces Equation:
93
8/8/2019 Math 316 Notes
94/179
Lecture 18 - Solving Laplaces 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 ux (0, y) =
0 =u
x(1, y).
2. Implement an inhomogeneous term for Poissons Equation:
2u
x2+
2u
y2= f(x, y) 0 < x, y < 1.
Introduce finite difference quotients, assume x = y to arrive at theiterative formula:
uk+1nm = ukn+1m + ukn1m + uknm+1 + uknm1 x2f(xn, ym)4 . ()It may be useful to calculate the values of fnm on a separate sheet inwhich the same cell values as those for unm are maintained. Then thevalues offnm can be referenced in the calculation ofunm according to().
94
8/8/2019 Math 316 Notes
95/179
Chapter 15
Lecture 19 Further HeatConduction Problems:Inhomogeneous BC
Example 15.1 Specified