Anda di halaman 1dari 147

5CCP2255 Mathematical Methods in Physics II

Lecture Notes, Problems and Outline

Lecturer: L.N.Kantorovich
Physics, Kings College London
email: lev.kantorovitch@kcl.ac.uk

March 24, 2012


1

Recommended textbooks:
Two very good books covering all the material; it should be good for stronger students:

ISBN 1-891389-29-7 D. McQuarrie "Mathematical methods for scientists and engineers" Univ. Sci. Books
2003.
ISBN 0521679710 K. F. Riley, M. P. Hobson, and S. J. Bence, Mathematical Methods for Physics and
Engineering, Cambridge Univ. Press, 2006

This one should be ok for most students; however, the explanations are rather scarse and for some students
may be insucient:

ISBN 0-471-04409-1 M. Boas "Methematical methods in the physical sciences" Wiley 2nd Edition 1983

This one contains a lot of examples and very detailed explanations; should be ideal for students who do not
feel condent in maths:

ISBN 0-333-919394 K. Stroud "Engineering mathematics" Palgrave, 5th Edition 2001


Contents

I Lecture notes 6
1 Some Special Functions 7
1.1 Dirac Delta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Filtering Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.3 Alternative Delta Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.4 Connection With Heaviside Unit Step Function . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Denition and main properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Gaussian Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 Stirlings Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 The Beta Function [optional ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Fourier Series 13
2.1 Trigonometric series: an intuitive approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Dirichlet conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Integration and dierentiation of the Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Parcevals thorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Complex (exponential) form of the Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Application to dierential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 A more rigorous approach to the Fourier series [optional ] . . . . . . . . . . . . . . . . . . . . . . . . 23

3 Fourier Transform 26
3.1 Fourier series (recap) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.1 Periodic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 Fourier Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.3 Even and Odd Functions f (t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.4 Complex Form for F S(t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.5 Physical signicance of F S(t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 The Fourier Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Alternative Forms of the Fourier Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Fourier Transform of Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.6 Convolution Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Spatially Varying Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Parcevals Therorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.9 Application of the FT to the Poison equation [optional ] . . . . . . . . . . . . . . . . . . . . . . . . . 36

4 Laplace Transform 39
4.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 LT of derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Solution of Ordinary Dierential Equations (ODEs) . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4 Simplifying Complicated Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.5 Convolution Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.6 A more rigorous consideration of the LT, its relation to the FT and the inverse LT (not in the syllabus) 45

2
CONTENTS 3

5 Vector Calculus: curvilinear coordinates (Part I) 49


5.1 Scalar and Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.3 Coordinate Surfaces and Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.4 Unit Base Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.5 Orthogonal Coordinate Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.6 Line Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.7 Volume element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.8 Mechanics of a Point in General Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . 58

6 Vector calculus: Grad, Div and Curl in Orthogonal Curvilinear Coordinates


(Part II) 62
6.1 Gradient of a Scalar Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.2 Divergence of a Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Laplacian in Orthogonal Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.4 Curl of a Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

7 Series Solutions of Ordinary Dierential Equations 69


7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.2 Ordinary and Singular Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.3 Series Solutions about an Ordinary Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
7.4 Classication of Singular Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.5 Series Solutions About a Regular Singular Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.6 Special cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.7 More Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

8 Partial Dierential Equations 81


8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2 PDEs of Physical Importance: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2.1 Classical Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2.1.1 Transverse waves on a string . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2.1.2 Vibrations of a circular membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.2.2 Diusion Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.2.2.1 Diusion processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.2.2.2 Heat conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.3 Helmholtz equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.4 Laplace equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.5 Poisson equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.6 Schrdinger equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.3 PDEs of Physical Importance: derivation of some equations . . . . . . . . . . . . . . . . . . . . . . . 83
8.3.1 Oscillation of a one-dimensional string . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.3.2 Hydrodynamic equations of ideal liquid (gas) . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.4 Propagation of sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.5 Heat conduction equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.6 Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.7 Superposition Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.8 Fourier Method for Linear Homogenenous PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.9 Oscillations of a rectangular membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
8.10 Question of Separability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.11 Heat-Conduction Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

9 Legendre Polynomials Pn (x) 101


9.1 Generating function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.2 Recurrence relations 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.3 Other recurrence relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.4 Dierential equation for functions Pn (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
9.5 Special values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
CONTENTS 4

9.6 Orthogonality and normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105


9.7 Expansion of functions in Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
9.8 Rodrigues formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
9.9 Other orthogonal polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

10 Laplace Equation in Spherical Polar Coordinates 111


10.1 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
10.2 Solution of the equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
10.3 Solution of equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
10.4 Legendre equation (m = 0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
10.5 Associated Legendre equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.6 Spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

II Problems 117
11 Problem sheet 1: (x), H(x) and (x) 118
11.1 Problems to be done in the class: (x) and H(x) functions . . . . . . . . . . . . . . . . . . . . . . . . 118
11.2 Problems to be done in the class: (x) function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.3 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.4 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

12 Problem sheet 2: Fourier series 121


12.1 Class problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
12.2 Home work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

13 Problem sheet 3: Fourier Transform 123


13.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
13.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
13.3 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

14 Problem sheet 4: Laplace transform 126


14.1 Problems on the Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
14.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
14.3 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

15 Problem sheet 5: Vector calculus, Part I 128


15.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
15.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
15.3 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

16 Problem sheet 6: Vector calculus, Part II 131


16.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
16.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
16.3 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

17 Problem sheet 7: Frobenius method 133


17.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
17.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
17.3 This week CHALLENGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

18 Problem sheet 8-9: Separation of variables 135


18.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
18.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

19 Problem sheet 10: Legendre Polynomials 138


19.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
19.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
CONTENTS 5

20 Problem sheet 11: Associated Legendre Functions and Laplace Equation 140
20.1 Problems to be done in the class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
20.2 Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

III Syllabus 142


20.3 Some special functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
20.3.1 Delta function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
20.3.2 Heviside step function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
20.3.3 Gamma function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
20.4 Fourier series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
20.5 Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
20.6 Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
20.7 Vector calculus: Part I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
20.8 Vector calculus: Part II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
20.9 Frobenius method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
20.10Separation of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
20.11Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
20.12Laplace equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Part I

Lecture notes

6
Chapter 1

Some Special Functions

1.1 Dirac Delta Function


1.1.1 Denition
Consider a function n (x) which is dened by
{
n, if 2n
1
x 1
2n
n (x) = (1.1)
0, if |x| > 2n
1

where n is a positive integer. For this function the denite integral



n (x)dx (1.2)

has the value 1 for all values of n.


When the value of the parameter n is increased, the graph of n (x) becomes narrower and more peaked while
the area under the curve remains unchanged and equal to 1. This result suggests that it might be possible to dene
an idealized impulse function by taking the limit n . Formally, we write

lim n (x) (x), (1.3)


n

where (x) is called the Dirac Delta Function. Intuitively, one would expect the limit function (x) to have the
following properties: {
, if x = 0
(x) = (1.4)
0, if x = 0
and

(x)dx = 1 (1.5)

Unfortunately, from a strictly mathematical point of view the limit (x) does not exist as an ordinary point
function because n (x) , as n . However, it is possible to interpret (x) as a generalised function using
the sequence of well-dened functions n (x) (n = 1, 2, 3...).

1.1.2 Filtering Theorem


Consider the integral
1
2n
n (x)f (x)dx = n f (x)dx, (1.6)
2n
1

where f (x) is any well-behaved function of x. For large values of n the range of integration 2n
1
x 1
2n becomes
very small, and we can use the Maclaurin expansion about the point x = 0

f (x) = f (0) + xf (0) + 0(x2 ) (1.7)

7
CHAPTER 1. SOME SPECIAL FUNCTIONS 8

n(x)

n=3

n=2

n=1

o
Figure 1.1: Delta sequence.

to evaluate (1.6). We nd that ( )



1
n (x)f (x)dx = f (0) + O ,
n2
( 1
) 1
where O n2 means terms of the order of n2 .
Hence we obtain
[ ]
lim n (x)f (x)dx = f (0). (1.8)
n

From this important result we see that the integral operation acting on f (x) has ltered out the function value
f (0).
We can express the left hand side (LHS) of (1.8) in terms of (x) by formally interchanging the order of the
limit and integration operations, and applying (1.3). This procedure gives
[ ] [ ( ) ]
lim n (x)f (x)dx = lim n (x) f (x)dx = (x)f (x)dx. (1.9)
n n

The manilpulation in (1.9) is formal and non-rigorous. However, we can use the formula (1.9) to provide us with a
denition of a generalised function as the limit (1.3). It is now possible to express the ltering theorem in the nal
form:
(x)f (x)dx = f (0). (1.10)

From the basic result (1.10) we see that (x) essentially denes a ltering operation which maps a given function
f (x) into the value f (0). It is also helpful, at least formally, to visualise (x) as an innitely sharp impulse function
of unit area at x = 0.
If the above analysis is repeated for a delta sequence n (x a), n = 1, 2, 3..., which is centred on the point x = a,
then we obtain the more general ltering theorem

(x a)f (x)dx = f (a) (1.11)

NOTE : The limits in (1.11) are unimportant as long as a is inside the limits. Otherwise, the integral is zero.

1.1.3 Alternative Delta Sequences


The sequence of rectangular pulses n (x), (n = 1, 2, 3...) used to derive (1.8) is not unique. There are, in fact, many
delta sequences which will map f (x) f (0).
CHAPTER 1. SOME SPECIAL FUNCTIONS 9

w(x)

8
Figure 1.2: Graph of w(x).

In particular, it can be shown that if w(x) is a function of a general shape of Fig. 1.2 (i.e. it has a bell-like
shape with its tails at x = going to zero) and which satises

w(x)dx = 1 (1.12)

then nw(nx) n (x) (where n = 1, 2, 3...) is a delta sequence.


For example, the functions
1 1
w(1) (x) = , (1.13)
1 + x2
( )
1 1 2
w(2) (x) = 1 exp x , (1.14)
(2) 2 2
1
sin x 1
w(3) (x) = = eikx dk (1.15)
x 2 1
(the last passage in the equality is checked by direct integration) all satisfy the condition (1.12) and have the desired
shape as in Fig. 1.2. It follows, therefore, that
n 1
n(1) (x) = , (1.16)
1 + n 2 x2
( )
n 1
n(2) (x) = 1 exp n2 x2 , (1.17)
(2) 2 2
where (n = 1, 2, 3...), are alternative delta sequences which can be used to dene (x). The third sequence,
1 n
n substitution 1
(3)
n (x) = eiknx
dk = eixt dt
2 1 t = nk = 2
n

tends when n to the integral


1
n(3) (x) eixt dt = (x) (1.18)
2

which is a very frequently used integral representation of the delta function. We shall come across it again in the
later lectures.

1.1.4 Connection With Heaviside Unit Step Function


The Heaviside unit step function is dened as
{
0 if x < 0
H(x) = . (1.19)
1 if x > 0
CHAPTER 1. SOME SPECIAL FUNCTIONS 10

x
o
Figure 1.3: The Heaviside unit step function.

The derivative of H(x) is equal to zero for x = 0, and is not dened at x = 0. We might expect, therefore, (at
least intuitively) that
(x) = H (x). (1.20)
We can give a more convincing derivation of (1.20) by considering
x
(t)dt = (t)H(x t)dt = H(x). (1.21)

If we now dierentiate (1.21) with respect to x we obtain the result (1.20).


The derivative function H (x) should be considered to be a generalised funcion which is equal to (x).

1.2 The Gamma Function


1.2.1 Denition and main properties
The Gamma Function is a generalisation of the factorial function. In this section we shall derive a few of its basic
properties, solve a Gaussian integral and derive Stirlings approximation for n!.
We dene the Gamma Function by
(z) = tz1 et dt, (1.22)
0
where z is a real number. (The properties of (z) can be extended to complex z, but these will not be needed here.)
The fundamental property of (z) is the recursion relation
(z + 1) = z(z). (1.23)
This is proved by integrating (1.22) by parts: thus

z t
[ z t ]
(z + 1) = t e dt = t e 0 + z tz1 et dt = z(z) when z > 0.
0 0

When z = 1 we obtain (1) = 0 et dt = 1. When z is a positive integer it is easy to see, using induction, that
(n + 1) = n!. Hence we obtain, as a special case, 0! = (1) = 1.
The recursion relation can be used to extend (z) to the range z < 0.

1.2.2 Gaussian Integrals


One type of integrals one frequently encounters (e.g. in statistical physics) is of the form
+
ts eat dt
2
Is (a) = (1.24)

where a is a constant and s is a positive integer. We shall show that


{
a 2 ( s+1
s+1

Is (a) = 2 ) , when s is even (1.25)


0, when s is odd
CHAPTER 1. SOME SPECIAL FUNCTIONS 11

Proof When s is odd the integrand is an odd function and hence the integral must be zero. When s is even we
have
ts eat dt.
2
Is (a) = 2 (1.26)
0

Changing variables by dening x = at , whence t = a x 2 and dt = 12 a 2 x 2 dx, we obtain


12 1 1 1
2

( )
s+1
Is (a) = a 2 x 2 ex dx = a 2
s+1 s1 s+1
. Q.E.D. J
0 2

We shall now show that ( 21 ) = .

Proof Consider
[ ( )]2 + 2
1 (x2 +y 2 )
er rdrd
2
I0 (1) I0 (1) = = e dxdy =
2 0 0

on changing from cartesian to polar coordinates. Hence we obtain


[ ( )]2
1
er rdr = et dt = ,
2
= 2
2 0 0
( )
so that 12 = . Q.E.D. J ( )
Since, according to Eq. (1.25), 12 = I0 (1), then we obtain a very useful identity:
+

et dt =
2

1.2.3 Stirlings Approximation


Finally, we shall derive Stirlings approximation for (z), in the limit when z is very large, namely

(z + 1) 2z z z ez .

Proof
By denition we obtain
(z + 1) = tz et dt.
0
We now substitute t = zx to obtain

z zx z+1 z
(z + 1) = (zx) e zdx = z e exp(z(x ln x 1))dx
0 0

Now the function f (x) = x ln x 1 has a minimum at x = 1 and the value of f at the minimum is f (1) = 0.
Thus the function ezf (x) has a maximum at x = 1 and this maximum becomes very sharply peaked when z is
large. We can obtain the dominant contribution to the integral by expanding f (x) about x = 1 and retaining the
leading order term. That is, if we write x = 1 + y, we obtain
y2 ( )
f (1 + y) = 1 + y ln(1 + y) 1 = + O y3
2
if we expand it in the Taylor series about y = 0. Because the integrand in (z + 1) is so sharply peaked at y = 0,
no error is introduced by letting the lower limit of the integral be , instead of 1. Consequently, we obtain
+
zy 2
(z + 1) z z+1 ez e 2 dy = 2zz z ez .

As a consequence of Stirlings approximation



n! = (n + 1) 2nnn en ,
so that
ln (n!) n ln n n + O (ln n) .
In most applications the correlation terms of order ln n are negligible.
CHAPTER 1. SOME SPECIAL FUNCTIONS 12

f(x)

x
0 1
Figure 1.4: Function f (x) = x ln x 1.

1.3 The Beta Function [optional ]


The Beta function is closely related to the Gamma function and is used to evaluate some commonly occuring
integrals. It is dened by 1
B(p, q) = tp1 (1 t)q1 dt. (1.27)
0
We shall show that
(p)(q)
B(p, q) =
(p + q)

Proof. First let t = sin2 , so that dt = 2 sin cos d and using the denition (1.27) the Beta function can be
written as 2
B(p, q) = 2 (sin)(2p1) (cos)(2q1) d.
0

Second, consider (the corresponding substitutions are shown within the brackets at the end of each line)

p1 x
(p)(q) = x e dx y p1 ey dy
0 0

u2p1 v 2q1 e(u
2
+v 2 )
=4 dudv (where x = u2 , y = v 2 )
0 0

2

r2p+2q1 (sin )2q1 (cos )2p1 der dr (where u = r cos , v = r sin )
2
=4
0 0

tp+q1/2 et t 2 dt/2 (then, we use t = r2 )
1
= B(p, q)2
0

= B(p, q)(p + q)
whence the result follows. Q.E.D. J
Chapter 2

Fourier Series

2.1 Trigonometric series: an intuitive approach


We know that almost any good function can be expanded into a Taylor series about some point x0 :
2 3
f (x) = f0 + f1 (x x0 ) + f2 (x x0 ) + f3 (x x0 ) + . . .

The Taylor series has an innite number of terms. One can say that f (x) is represented by a linear combination of
n
an innite number of functions gn (x) = (x x0 ) , i.e.


f (x) = f0 g0 (x) + f1 g1 (x) + . . . = fn gn (x) (2.1)
n=0

The question one may ask is this: is it possible to nd other sets {gn (x)} of functions, may be containing an innite
number of functions, so that any good function f (x) would be possible to expand via these as in Eq. (2.1)?
In order to answer this question, we consider a very special set of functions f (x) which are periodic with period
of 2l, i.e.
f (x + 2l) = f (x) (2.2)
(the factor of two in the period is introduced for convenience). Then, we consider functions
nx nx
n (x) = cos and n (x) = sin (2.3)
l l
where n = 0, 1, 2, . . ., which also have the same periodicity for any n, i.e.
n(x + 2l) ( nx ) nx
cos = cos + 2n = cos
l l l
n(x + 2l) ( nx ) nx
sin = sin + 2n = sin
l l l
What we would like to do is to understand whether it is possible to express f (x) as a linear combination of all
these functions for all possible values of n from 0 to . We shall start our discussion by showing that the functions
(2.3) satisfy the following identities for n = m:
l l
n (x)m (x)dx = 0 and n (x)m (x)dx = 0 (2.4)
l l

Indeed, { }
nx mx 1 (n + m) x (n m) x
n (x)m (x) = cos cos = cos + cos
l l 2 l l
Note that for any non-equal indices n and m, the integer numbers k = n m are never equal to zero. Then, for
any k = 0
l l
kx l kx l
cos dx = sin = (sin k sin (k)) = 0
l l k l l k

13
CHAPTER 2. FOURIER SERIES 14

so that the rst integral in Eq. (2.4) is zero. Similarly, we proof the second identity in Eq. (2.4) using
{ }
nx mx 1 (n m) x (n + m) x
n (x)m (x) = sin sin = cos cos
l l 2 l l

Finally, we can also see that


l
n (x)m (x)dx = 0 (2.5)
l

for any n and m (even if they are equal), since


{ }
nx mx 1 (n + m) x (n m) x
n (x)m (x) = cos sin = sin + sin
l l 2 l l

and l
l
kx l kx l
sin dx = cos = (cos k cos (k)) = 0
l l k l l k
for any k (including zero). Thus, we nd that the integral between l and l of a product of any two dierent
functions taken from the set {n , n } is always equal to zero. It is said that these functions are orthogonal or form
an orthogonal set of functions.
Consider now similar integrals between two identical functions:
l l ( ) { ( )}
2 2 nx 1 l 2nx 1 l
n (x) dx = cos dx = 1 + cos dx = dx = l
l l l 2 l l 2 l
l l ( nx ) l { ( )} l
1 2nx 1
2
n (x) dx = 2
sin dx = 1 cos dx = dx = l
l l l 2 l l 2 l

All the found relations can now be conveniently rewritten using the Kroneker symbol nm (equal to zero if n = m
and to unity if n = m) as
l l
n (x)m (x)dx = lnm and n (x)m (x)dx = lnm (2.6)
l l

Let us now assume that some function f (x) that is periodic with period 2l, can be represented as a linear
combination of all functions of the set {n , n }, i.e. as an innite series

a0 { nx }

a0 nx
f (x) = + {an n (x) + bn n (x)} = + an cos + bn sin (2.7)
2 n=1
2 n=1
l l

Note that 0 (x) = 0 and can be dropped; also, 0 (x) = 1 and can be separated from the sum with its coecient
chosen for convenience as a20 .
What we would like to do now is to determine the coecients a0 , a1 , a2 , a3, , . . . and b1 , b2 , b3 , . . .. To this end,
let us rst integrate both sides of Eq. (2.7) from l to l:
l l
{ l }
a0 l
a0
f (x)dx = dx + an n (x)dx + bn n (x)dx = 2l = a0 l
l l 2 n=1 l l 2

so that l
1
a0 = f (x)dx (2.8)
l l

(any integral in the curly brakets is equal to zero). To obtain other coecients an for n = 0, we rst multiply both
sides of Eq. (2.7) by m (x) with some xed value of m = 0 and then integrate from l to l:
l
{ l }
a0 l l
f (x)m (x)dx = m (x)dx + an n (x)m (x)dx + bn n (x)m (x)dx
l 2 l n=1 l l
CHAPTER 2. FOURIER SERIES 15

The rst term in the RHS is zero since m = 0. Similarly, due to Eq. (2.5), the second integral in the curly brackets
is also equal to zero, and we are left with
l
l
f (x)m (x)dx = an n (x)m (x)dx
l n=1 l

In the RHS we have an innite sum of terms containing the same integrals as in Eq. (2.6) which are all equal to
zero except for the single one in which n = m, i.e. only a single term in the sum above survives:
l
l

f (x)m (x)dx = an n (x)m (x)dx = an lnm = am l
l n=1 l n=1

which gives immediately


l l
1 1 mx
am = f (x)m (x)dx = f (x) cos dx (2.9)
l l l l l
Note that a0 of Eq. (2.8) can also formally be obtained from Eq. (2.9) although the latter was, strictly speaking,
obtained for non-zero values of m only. This is because of the factor of 12 introduced earlier in Eq. (2.7). Therefore,
Eq. (2.9) gives all an coecients.
The coecients bn in Eq. (2.7) are obtained similarly by multiplying both sides of Eq. (2.7) by the function
m (x) with some xed value of m = 0 and then integrating between l to l:
l
{ l }
a0 l l
f (x)m (x)dx = m (x)dx + an n (x)m (x)dx + bn n (x)m (x)dx
l 2 l n=1 l l


l

= bn n (x)m (x)dx = bn lnm = bm l
n=1 l n=1

so that
l l
1 1 mx
bm = f (x)m (x)dx = f (x) sin dx (2.10)
l l l l l
Formulae (2.8)-(2.10) solve the problem: if the function f (x) is known, then we can calculate all the coecients in
its expansion of Eq. (2.7). The coecients an and bn are called Fourier coecients, and the innite series (2.7) -
Fourier series.

Example 1 Consider the function f (x) = x specied in the interval < x < . Calculate the expansion
coecients an and bn and thus write the corresponding Fourier series.
Solution: First, we periodically repeat our function, so that its period becomes equal to 2. In this case l ,
and the formulae (2.8)-(2.10) for the coecients an and bn are rewritten as:

1
am = x cos (mx) dx = 0, m = 0, 1, 2, . . . and

{ }
1 1 cos(mx) 1
bm = x sin (mx) dx = x +m cos (mx) dx
m




1 cos(m) cos(m) 1
2 2
= + 2 sin(mx) = (1)m = (1)m+1 , m = 0
m m m
m m

| {z }
equal to zero

(integration by parts was used). Note that am = 0 for any m because the function under the integral for am is
an odd function and we integrate over a symmetric interval. Thus, in this particular example the Fourier series
consists only of sine functions:

2
f (x) = (1)m+1 sin(mx) (2.11)
m=1
m
CHAPTER 2. FOURIER SERIES 16

2.5 n=3
0
-2.5

2.5 n=5
fn(x)

0
-2.5

2.5 n=10
0
-2.5

2.5 n=20
0
-2.5
-15 -10 -5 0 5 10 15
x
Figure 2.1: Graphs of fn (x) corresponding to the rst n terms in the series of Eq. (2.11).

The convergence of the series is demonstrated in Fig. 2.1: the rst n terms in the series are accounted for, i.e. the
functions
n
2
fn (x) = (1)m+1 sin(mx)
m=1
m
for several choices of the upper limit n = 3, 5, 10, 20 are plotted. It can be seen that the series converges very quickly
to the exact function between and . Beyond this interval the function is periodically repeated. J
The actual integration limits from l to +l in the above formulae were chosen only for simplicity; in fact, due
to periodicity of f (x), cos mx
l l , one can use any limits diering by 2l, i.e. from l + c to l + c for any
and sin mx
value of c. For instance, in some cases it is convenient to use the interval from 0 to 2l.
The Fourier expansion can be handy in summing up innite numerical series.

Example 2 Show that


1 1 1
S =1 + + ... = (2.12)
3 5 7 4
Solution: Consider the series (2.11) generated in Example 1 for f (x) = x, < x < , and set there x = 2 :


2 ( m )
= (1)m+1 sin
2 m=1
m 2
The sine functions are non-zero only for odd values of m = 1, 3, 5, . . ., so that we obtain calculating the rst few
terms explicitly: ( )
1 1 1
= 2 1 + + . . . = 2S
2 3 5 7
so that S = 4 as required. J

2.2 Dirichlet conditions


The discussion above was not rigorous: rstly, we assumed that the expansion (2.7) exists and, secondly, we
integrated the innite expansion term by term which also cannot be always done. A more rigorous formulation of
CHAPTER 2. FOURIER SERIES 17

f(x)

f1(x) f2(x)

x0 x
Figure 2.2: The function f (x) discontinues at x0 . However, nite limits exist on both sides of the jump correspond-
ing to the two dierent functions on both sides of the point x = x0 : on the left, f (x 0) = limxx0 f1 (x) = f1 (x0 ),
while on the right f (x + 0) = limxx0 f2 (x) = f2 (x0 ).

the problem is this: we are given a function f (x) specied in the interval l < x < l 1 ; we then form an innite
series (2.7) with the coecients calculated via Eqs. (2.8)-(2.10). We ask if the series converges for any l < x < l,
and if it does, would the result be f (x)? Also, are there any limitations on the function f (x) itself?
We rst give some denitions. Let the function f (x) be discontinuous at x = x0 , see Fig. 2.2, but has well-dened
limits x x0 from the left and from the right of x0 , i.e.

f (x0 + 0) = lim f (x0 + )


0

f (x0 0) = lim f (x0 )


0

with > 0 in both cases. It is then said that f (x) has a discontinuity of the 1st kind at the point x0 . Then,
the function f (x) is said to be piecewise continuous in the interval a < x < b, if it has a nite number n < of
discontinuities of the 1st kind there, but otherwise is continuous everywhere, i.e. it is continuous beetwen any two
adjacent points of discontinuity.
The function of Example 1 f (x) = x, < x < , when periodically repeated, represents an example of such
function: in any nite interval crossing points , 2, . . ., it has a nal number of discontinuities; however, at
each discontinuity nite limits exist from both sides. For instance, consider the point of discontinuity x = . Just
on the left of it f (x) = x and the limit from the left f ( 0) = , while on the right of it f ( + 0) = due to
periodicity of the f (x). Thus, x = is the discontinuity of the 1st kind.
Then, the following Dirichlet theorem addresses the fundamental questions about the expansion of the function
f (x) into a Fourier series:

Theorem
If f (x) is piecewise continuous in the interval l < x < l and have period 2l, then the Fourier series

a0 { nx }

nx
fF S (x) = + an cos + bn sin (2.13)
2 n=1
l l

converges to f (x) at any point x where f (x) is continuous, and to the mean value,
1
fF S (x0 ) = [f (x0 0) + f (x0 + 0)] (2.14)
2
at the points x = x0 of discontinuity.
Although the proof of this theorem is quite remarkable (and not very dicult albeit lengthy), it is not given
here. The curious students should refer to some explanations in Section 2.7 (optional). Functions f (x) satisfying
conditions of the Dirichlet theorem are said to satisfy Dirichlet conditions.
1 Or, which is the same, which is periodic with period 2l, the main part of the function which is periodically repeated can start

anywhere, one choice is between l and l.


CHAPTER 2. FOURIER SERIES 18

Example 3 Consider the function f (x) = x, < x < , with period 2 (Example 1) whose Foruier series is
given by Eq. (2.11):

2
fF S (x) = (1)m+1 sin(mx)
m=1
m

What values does fF S (x) converge to at the points x = , 0, 2 , , 3


2 ?
Solution: f (x) is continuous at x = 0, 2 , 3
2 and thus fFS (0) = f (0) = 0, fF S ( 2 ) = f ( 2 ) = 2 and fF S ( 3
2 )=
f ( 2 ) = f ( 2 2) = f ( 2 ) = 2 , while at x = the function f (x) has the discontinuity of the 1st kind
3 3

with the limits on both sides equal to + (from the left) and (right), respectively, so that the mean is zero, i.e.
fF S () = 0. This is also clearly seen in Fig. 2.1. J

2.3 Integration and dierentiation of the Fourier series


The Foruier series can be integrated term by term if f (x) satises Dirichlets conditions, no additional requirements
are to be imposed on f (x). This is because, when integrating cos nx l or sin nx
l with respect to x, an additional
1
factor of n arises that can only accelerate the convergence of the Fourier series.

Example 4 Obtain the Foruier series for the function f (x) = x2 , < x < , that is periodic with period 2.
Solution: This can be obtained by integrating term-by-term the series (2.11) for f1 (x) = x from 0 to some
0 < x < : x x

2 m+1
f1 (x1 )dx1 = (1) sin(mx1 )dx1
0 m=1
m 0

x2
Since f1 (x1 ) = x1 in the interval under consideration, the integral in the LHS gives 2 . Integrating the sine functions
in the RHS, we obtain:

x2 2 1
= (1)m+1 [ cos(mx) + 1]
2 m=1
m m

2(1)m+1 (1)m+1
= cos(mx) + 2
m=1
m2 m=1
m2

The numerical series (the 2nd term) can be shown (using the direct method for f (x) = x2 , i.e. expanding it into
2
the Foruier series2 ) to be equal to 12 ,

(1)m+1 2
= ,
m=1
m2 12
so that we nally obtain:

2 (1)m+1
x2 = 4 cos(mx) (2.15)
3 m=1
m2
The convergence of this series with dierent number of terms n in the sum is pretty remarkable as is demonstrated
in Fig. 2.3. J
The situation with term-by-term dierentiation of the Fourier series is more complex since each dierentiation
of either cos nx nx
l or sin l brings in an extra n in the sum which results is slower convergence or even divergence.
For example, if we dierentiate formula (2.11) for f (x) = x, < x < , we obtain:


1= 2(1)m+1 cos(mx)
m=1

which contains the diverging series. There are much more severe restrictions on the function f (x) that would enable
its Fourier series be dierentiable term-by-term.
2 See your home work.
CHAPTER 2. FOURIER SERIES 19

10

n=10
n=3
8 n=2

fn(x)
4

0
-10 0 10
X

Figure 2.3: The partial Fourier series of f (x) = x2 , see Eq. (2.15).

2.4 Parcevals thorem


Consider a periodic function f (x) with period 2l satisfying Dirichlet conditions, i.e. f (x) has a nite number of
discontinuities of the 1st kind in the interval l < x < l. Thus, it can be expanded in the Fourier series (2.13). If
we square both sides and integrate from l to l, we obtain:
l
2 a20 a0 l { nx nx }
fF S (x)dx = 2l + 2 an cos + bn sin dx
l 4 2 n=1 l l l
{ nx } { mx }

l
nx mx
+ an cos + bn sin am cos + bm sin dx
n=1 m=1 l l l l l

The integral in the LHS can be replaced by the integral of f 2 (x) if f (x) is continuous everywhere. However, if it has
discontinuities, this can also be done by splitting the integral into a sum of integrals over each region of continuity
of f (x), where fF S (x) f (x). In the RHS of the above equation, any integral in the second term is zero. Also,
due to the orthogonality of the sine and cos functions, Eqs. (2.5) and (2.6), in the term with the double sum only
integrals with equal indices n = m are non-zero if taken between two cosine or two sine functions:

1 l 2 a2 ( 2 )
f (x)dx = 0 + an + b2n (2.16)
l l 2 n=1

This equation is called Parsevals equality or theorem. It can be used e.g. to calculate innite numerical series.

Example 5 Write the Parcevals equality for the series (2.11) of f (x) = x, < x < , and then sum up the
innite numerical series:
1 1 1
1 + 2 + 2 + 2 ...
2 3 4
Solution: The integral in the LHS of the Parcevals equality (2.16) is simply (l = ):

1 2 1 2 1 x3 2 2
f (x)dx = x dx = =
3 3

In the RHS of (2.16) we then have bm = 2(1)m+1 /m and am = 0, see (2.11), i.e.
[ ]2
2 4 1
(1)m+1 = 2
= 4 2
m=1
m m=1
m m=1
m

Therefore,

1 2 2
4 2
=
m=1
m 3
CHAPTER 2. FOURIER SERIES 20

or

1 1 1 2
1 + 2 + 2 + ... = =
2 3 m=1
m2 6
as required.J

2.5 Complex (exponential) form of the Fourier series


Another form of the Fourier series (2.7) of a function f (x) that is periodic with period 2l, is the complex (exponential)
series:


f (x) = cn einx/l = cn n (x) (2.17)
n= n=

The Fourier coecients cn can be obtained in the same way as for the sine/cosine series noting that the functions
n (x) = einx/l also form an orthogonal set. Indeed, if integers n and m are dierent, n = m, then
l l l
n (x)m (x)dx = einx/l eimx/l dx = ei(nm)x/l dx
l l l

l l l ( )

= ei(nm)x/l = ei(nm) ei(nm)
i(n m) l i(n m)
l
= {2i sin [(n m)]} = 0
i(n m)
If n = m, however, then
l l l
n (x)n (x)dx =
2
|n (x)| dx = dx = 2l
l l l

so that we can generally write


l
n (x)m (x)dx = 2lnm (2.18)
l

Note that one of the functions is complex conjugate in the above equation.
Thus, we muplitply both sides of Eq. (2.17) by m (x) with a xed index m, and integrate from l to l:
l
l

f (x)m (x)dx = cn n (x)m (x)dx = cn 2lnm = cm 2l
l n= l n=

which gives for any m:


l
1
cm = f (x)m (x)dx (2.19)
2l l

The same expressions (2.17) and (2.19) can also be derived directly from the sine/cosine Fourier series. Indeed,
starting from Eq. (2.13) and replacing sine and cosine with complex exponentials by means of the Eulers formulae:
1 ( ix ) 1 ( ix )
sin x = e eix and cos x = e + eix
2i 2
Indeed, we have:
{ )}
a0 1 ( inx/l inx/l
) 1 ( inx/l inx/l
f (x) = + an e +e + bn e e
2 n=1
2 2i

{ } { }
a0 1 inx/l 1 inx/l 1 inx/l 1 inx/l
= + an e + bn e + an e bn e
2 n=1
2 2i n=1
2 2i
( ) ( )
a0 1 1 1 1
= + an + bn einx/l
+ an bn einx/l (2.20)
2 2 n=1 i 2 n=1 i
CHAPTER 2. FOURIER SERIES 21

Now look at the formulae (2.9) and (2.10) for an and bn :



1 l nx 1 l nx
an = f (x) cos dx and bn = f (x) sin dx
l l l l l l
Although these expressions were obtained for positive values of n, they can formally be extended for any values of
n inclusing negative and zero values. Then, we have: an = an , bn = bn and b0 = 0. These expressions allow us
to rewrite the last sum in (2.20) by means of summing over all negative integers n:
( ) ( ) ( )
1 1 inx/l 1 1 1 1
an bn e = an bn e inx/l
= an + bn einx/l
2 n=1 i 2 n=1 i 2 n=1 i

You see that this sum looks now exactly the same as the rst sum in (2.20) in which n is positive, so that we can
combine the two into a single sum in which n takes on all integer values from to + except for n = 0:

( )
a0 1 1
f (x) = + an + bn einx/l
2 2 i
n=,n=0

a0
Noting, that b0 = 0 and the 2 term can also be formally incorporated into the sum, we can nally write:


f (x) = cn einx/l
n=

where for any n we have for the coecients:


( ) [ ]
1 1 1 1 l nx 1 l nx
cn = an + bn = f (x) cos dx + f (x) sin dx
2 i 2 l l l il l l
( ) (
1 l
nx 1 nx 1 l nx nx )
= f (x) cos dx + sin dx = f (x) cos dx i sin dx
2l l l i l 2l l l l

1 l
= f (x)einx/l dx
2l l
The obtained equations are the same as Eqs. (2.17) and (2.19) derived above dierently. Thus, the two forms of the
Fourier series are completely equivalent to each other. The exponential (complex) form looks simplier and thus is
easier to remember. It is always possible, using the Eulers formula, to obtain any of the forms of the representation
as illustrated by the following example.

Example 6 Obtain the complex (exponential) form of the Fourier series for f (x) = x, < x < as in Example
1.
Solution: We start by calculating the Fourier coecients cn from Eq. (2.19) using l = . If n = 0, then:
{ }
1 inx/ 1 inx 1 1 inx 1 inx
cn = xe dx = x e dx = x e + in e dx
2 2 |{z} | {z } 2 in
u dv
{ }
1 ( in ) 1 ( in ) 1 1
= e + ein e ein = 2 cos (n) + 2i sin (n)
2 in (in)
2 2 in (in)
2 | {z }
=0

1 (1)n+1
= cos (n) =
in in
and, when n = 0,
1
1 x2
c0 = xdx = =0
2 2 2
so that the Fourier series is

(1)n+1 inx
f (x) = e (2.21)
in
n=,n=0
CHAPTER 2. FOURIER SERIES 22

Example 7 Show that the above expansion is equivalent to the series (2.11).
Solution: Since einx = cos(nx) + i sin(nx), we get by splitting the sum into two with negative and positive
summation indices:


(1)n+1 inx (1)n+1 inx (1)n+1 inx (1)n+1 inx
f (x) = e + e = e + e
n=1
in n=1
in n=1
in n=1
in

where in the second sum we replaced the summation index n n, so that the new index would run from 1 to
+ as in the other sum. Combining the two sums together, and noting that (1)n+1 = (1)n+1 , we get:

(1)n+1 ( inx ) (1)n+1 2(1)n+1
f (x) = e einx = 2i sin(nx) = sin(nx)
n=1
in n=1
in n=1
n

which is exactly the same as in Eq. (2.11) which was obtained using the sine/cosine formulae for the Fourier series.

2.6 Application to dierential equations


This method is frequently used to obtain solutions of dierential equations in the form of the Fourier series that
then can be used on a computer with a nal number of terms in the sum to obtain the solution numerically. As an
example, consider a harmonic oscillator subject to a periodic excitation (external) force f (t) of a general form:
y + 02 y = f (t) (2.22)
where f (t) is a periodic function of the period T = 2/. We would like to obtain the particular integral of this
DE.
Using the Fourier series, it is possible to write down the solution of (2.22) for a general f (t). Indeed, expand
f (t) in a complex Fourier series (we use l = /):



int/
f (t) = fn e = fn eint (2.23)
n= n=

with
T /2 T 2/
1 1
fn = f (t)eint dt = f (t)eint dt = f (t)eint dt (2.24)
T T /2 T 0 2 0

where the integration was shifted to the interval 0 < t < T = 2


.
To obtain y(t) that satises the DE above, we shall expand it into a Fourier series as well; this means that we
assume that the solution has the same periodicity as the external force:



y(t) = yn eint/ = yn eint (2.25)
n= n=

Substituting Eqs. (2.23) and (2.25) into Eq. (2.22), we obtain:





[ ]
(in)2 + 02 yn eint = fn eint
n= n=

or

{[ ] }
(in)2 + 02 yn fn eint = 0 (2.26)
n=

This equation is satised for all values of t if and only if all coecients of eint are equal to zero simultaneously for
all values of n, [ ]
(in)2 + 02 yn fn = 0 (2.27)
Indeed, upon multiplying both sides of Eq. (2.26) by eimt with some xed value of m and integrating between 0
and T , we get only the n = m term left in the LHS of Eq. (2.26) due to orthogonality of the functions n (t) = eint :

{[ ] } T int imt {[ ] }
(in)2 + 02 yn fn e e dt = (in)2 + 02 yn fn nm T
n= 0 n=
CHAPTER 2. FOURIER SERIES 23
{[ ] }
=T (im)2 + 02 ym fm = 0
which is Eq. (2.27). Thus, we get the uknown Fourier coecients of the solution
fn fn
yn = = 2 (2.28)
2 2
(in) + 0 0 (n)2
and the whole solution reads

fn
y(t) = eint (2.29)
2
n= 0
(n)2
We see that the harmonics of f (t) with frequencies n are greatly enhanced in the solution if they come close to
the fundamental frequency 0 of the harmonic oscillator.

2.7 A more rigorous approach to the Fourier series [optional ]


Before actually giving the rigorous formulation for the Fourier series, we note that in Eq. (2.7) the function f (x)
is expanded into a set of linearly independent functions {n (x)} and {n (x)}. Functions f1 (x), f2 (x),. . .,fk (x) are
linearly independent if any one of them cannot be expressed as a linear combination of the others (compare with
the linear independence of vectors in a vector space). In other words, the equation

k
1 f1 (x) + 2 f2 (x) + . . . + k fk (x) = i fi (x) = 0
i=1

which is valid for any x from a specied interval, has only a unique trivial solution for the coecients 1 = 2 =
. . . = k = 0.

Statement:
The functions {n (x)} and {n (x)} of Eq. (2.3) are linearly independent.
Proof : To prove the above statement, we construct the linear combination of all functions with the unknown
coecients i and i :

[i i (x) + i i (x)] = 0 (2.30)
i=0
Multiply out both sides by j (x) with some xed j and integrate over x between l and l:

[ l ]
l
i i (x)j (x)dx + i i (x)j (x)dx = 0
i=0 l l

Due to the orthogonality of the functions, see Eqs. (2.5) and (2.6), all the integrals between any i (x) and j (x)
will be equal to zero, while from all integrals involving both i (x) and j (x) (i = 0, 1, 2, . . .) only one with the value
of i = j will survive:
l
j j (x)j (x)dx = j l = 0
l
Therefore, j = 0. By taking dierent values of j, we nd that any of the coecients 1 , 2 , etc. is equal to zero.
Similarly, by multiplying both sides of Eq. (2.30) on j (x) with xed j and integrating over x between l and l,
we nd j = 0. Since j was chosen arbitrarily, all the coecients 1 , 2 , etc. are equal to zero. Q.E.D. J
The function f (x) contains an innite amount of information since the set of x values is continuous; there-
fore, when we expand it into an innite set of linearly independent functions, we provide an adequate amount of
information for it. Of course, this is not yet sucient for the f (x) to be expandable: the set of functions {n (x)}
and {n (x)} must also be complete to represent f (x) adequately. We shall not elaborate on this point, however.
In order to understand whether the Fourier series calculated using the formulae (2.9) and (2.10) for the an and
bn coecients, actually converges to the function f (x), let us consider a general linear combination

0
N
fN (x) = + (n n (x) + n n (x)) (2.31)
2 n=1

of the same type as the Fourier series (2.7) but with arbitrary coecients n and n . The sum above is constructed
out of the rst N functions n (x) and n (x) of the Fourier series.
CHAPTER 2. FOURIER SERIES 24

Theorem
The expansion (2.31) converges on average to the function f (x) for any N if the coecients n and
n of the linear combination coincide with the corresponding Fourier coecients an and bn dened
by Eqs. (2.9) and (2.10), i.e. when n = an and n = bn for any n = 0, 1, . . . , N . By average
convergence we mean the minimum of the mean square error
l
1 2
N = [f (x) fN (x)] dx (2.32)
l l

Note that N 0, i.e. cannot be negative.


Proof : Indeed, substituting the expansion (2.31) into the square error expression (2.32), we get three terms:
l l l
1 2 1
N = f (x)dx 2
f (x)fN (x)dx + 2
fN (x)dx (2.33)
l l l l l l

We use the orthogonality of the functions {n (x)} and {n (x)} to calculate the last term in Eq. (2.33):

2 0 l
l N
1 2 02
fN (x)dx = + (n n (x) + n n (x)) dx
l l 2 l n=1 2 l

N
1 l
N
+ (n n (x) + n n (x)) (m m (x) + m m (x)) dx
l n=1 m=1 l
The second term in the RHS is zero since the integrals of either n or n are zeros for any n. In the third term,
only integrals between two n or two n functions with equal indices survive, and thus we obtain

02 ( 2 )
l N
1 2
fN (x)dx = + n + n2 (2.34)
l l 2 n=1

Similarly, we can consider the second integral in Eq. (2.33):


( l )
2
N
2 l 0 l l
f (x)fN (x)dx = f (x)dx + n f (x)n (x)dx + n f (x)n (x)dx
l l l l l n=1 l l

Using Eqs. (2.9) and (2.10) for the Fourier coecients, we can rewrite the above expression in a simplied form:
l
N
2
f (x)fN (x)dx = 0 a0 + 2 (n an + n bn ) (2.35)
l l n=1

Thus, collecting all terms, we get:


02 ( 2 )
l N N
1
N = f 2 (x)dx 0 a0 2 (n an + n bn ) + + n + n2
l l n=1
2 n=1

( )
1 l
02
N
[( 2 ) ( )]
= 2
f (x)dx + 0 a0 + n 2n an + n2 2n bn
l l 2 n=1
l [] N [ ]
1 1 2 a2 2 2
= f 2 (x)dx + (0 a0 ) 0 + (n an ) a2n + (n bn ) b2n (2.36)
l l 2 2 n=1

It is seen that the minimum of N with respect to the coecients n and n of the trial expansion (2.31) is achieved
at n = an and n = bn , i.e. when the expansion (2.31) coincides with the partial Fourier series containing the rst
N terms. Q.E.D.J
CHAPTER 2. FOURIER SERIES 25

Theorem
The Fourier coecients an and bn dened by Eqs. (2.9) and (2.10) tend to zero as n .

Proof : The minimum error N is obtained from Eq. (2.36) by putting n = an and n = bn :

a20 [ 2 ]
l N
1
N = f 2 (x)dx an + b2n (2.37)
l l 2 n=1

Note that the values of the coecients n and n do not depend on the value of N ; for instance, if N is increased
by one, N N + 1, two new coecients are added to the expansion (2.31), N +1 and N +1 , however, the values of
the previous coecients remain the same. At the same time, the error (2.37), N +1 = N a2N +1 b2N +1 , gets two
extra negative terms, i.e. can only become smaller. As the number of terms N in the expansion is increased, the
error gets smaller and smaller. On the other hands, the error is always not negative by construction, i.e. N 0.
Therefore, from Eq. (2.37),

a20 [ 2 ] 1 l 2
N
+ an + b2n f (x)dx (2.38)
2 n=1
l l
As N is increased, the sum in the LHS is getting larger, but
will always remain smaller than the positive value of
[ ]
the integral in the RHS. This means that the innite series n=1 a2n + b2n is absolutely convergent, and we can
write:

a20 [ 2 ] 1 l 2
+ an + b2n f (x)dx (2.39)
2 n=1
l l

Thus, the innite series in the LHS is bound from above. Since the series converges, the terms of it a2n + b2n tend
to zero as n , i.e. each of the coecients an and bn tends separately to zero as n . Q.E.D.J
It can then be shown that the error N 0 as N . This means that actually we have the equal sign in the
above equation:

a20 [ 2 2
] 1 l 2
+ an + bn = f (x)dx (2.40)
2 n=1
l l

This is the familiar Parsevals equality, Eq. (2.16).


Chapter 3

Fourier Transform

3.1 Fourier series (recap)


3.1.1 Periodic Functions
A function f (t) is said to be periodic if there exists a positive number p such that1

f (t + p) = f (t), t (, ). (3.1)

The number p is called the period of f (t). It is readily shown that if p is a period then {kp; k = 2, 3, ...} are also
periods of f (t). The smallest period is called the fundamental period T of f (t).

Example 1: The function f (t) = sin(t) has periods p = {2m; m = 1, 2, ...}, with a fundamental period T = 2.

3.1.2 Fourier Theorem


It can be proven that any piecewise smooth periodic function f (t), with a fundamental period T , can be represented
by the Fourier series
[ ( ) ( )]
1 2nt 2nt
F S(t) = a0 + an cos + bn sin , (3.2)
2 n=1
T T
where the Fourier coecient an and bn are dened as
( )
2 t0 +T 2nt
an = f (t) cos dt, (n = 0, 1, 2, ...) (3.3)
T t0 T
( )
2 t0 +T 2nt
bn = f (t) sin dt, (n = 1, 2, ...) (3.4)
T t0 T
and t0 is an arbitrary real constant. The integrals in (3.3) and (3.4) are taken over any complete period T of f (t).
It should be noted that the values of these integrals do not depend on the value of t0 . In applications the evaluation
of the Fourier coecients can often be simplied by making a careful choice for the value of t0 . For example, it is
often convenient to take t0 to be equal to 0 or T2 .
We must now discuss the connection between F S(t) and the periodic function f (t). If f (t) is a continuous
function at a point t = t1 then it can be shown that F S(t1 ) = f (t1 ). However, if f (t) has a discontinuity at t = t1
then the Fourier series gives the mean value
1
F S(t1 ) = [f (t1 +) + f (t1 )] , (3.5)
2
where
f (t1 ) = lim f (t1 ). (3.6)
0+

1 The symbol means belongs to; in this particular case t (, ) reads for any t from the interval < t < .

26
CHAPTER 3. FOURIER TRANSFORM 27

3.1.3 Even and Odd Functions f (t)


If f (t) is an even function then f (t) = f (t) for all t. Under these circumstances, we see from (3.4) with t0 = T
2
that the Fourier coecient bn = 0 for all n = 1, 2, .... When f (t) is an odd function then f (t) = f (t) for all t.
In this case it follows from (3.3) with t0 = T2 that an = 0 for all n = 0, 1, 2, ....

Example 2: Obtain the Fourier series of a periodic function f (t) with a fundamental period T = 2 which is
dened in the interval [1, 1) by
f (t) = t, for 1 t < 1.

Solution. For this case f (t) is an odd function of t and we have an = 0 for all n = 0, 1, 2, .... We also nd from
(3.4) with t0 = T
2 and T = 2 that
1 1
bn = t sin(nt)dt = 2 t sin(nt)dt. (3.7)
1 0

After integrating (3.7) by parts we obtain the formula


2(1)n1
bn = for n = 1, 2, ... (3.8)
n
Hence the Fourier series for f (t) is given by

2 (1)n1
F S(t) = sin(nt). J (3.9)
n=1 n

3.1.4 Complex Form for F S(t)


If we substitute the Euler identities
( )
2nt 1 ( i2nt/T )
cos = e + ei2nt/T , (3.10)
T 2
( )
2nt 1 ( i2nt/T )
sin = e ei2nt/T , (3.11)
T 2i
in the Fourier series (3.2) it is found that

1 1 1
F S(t) = a0 + (an ibn )ei2nt/T + (an + ibn )ei2nt/T . (3.12)
2 n=1
2 n=1
2

Next the substitution n = m is made in the second summation in equation (3.12). Hence we obtain
1

1 1 1
F S(t) = a0 + (an ibn )ei2nt/T + (am + ibm ) ei2mt/T . (3.13)
2 n=1
2 m=
2

It follows from (3.3) and (3.4) that am = am , bm = bm and b0 = 0. The application of these results to (3.13)
gives the required complex form

F S(t) = cn ei2nt/T , (3.14)
n=

where
1
cn =(an ibn ) , n = 0, 1, 2, ... (3.15)
2
We can also use the formulae (3.3) and (3.4) to write the Fourier coecient (3.15) in the alternative integral form

1 t0 +T
cn = f (t)ei2nt/T dt. (3.16)
T t0
CHAPTER 3. FOURIER TRANSFORM 28

f(t)

T/2 0 T/2 t

Figure 3.1: Graph of a non-periodic function f (t).

Example 3: Use equations (3.14) and (3.16) to derive the Fourier series F S(t) for the periodic function f (t)
which is dened in Example 2.

3.1.5 Physical signicance of F S(t)


The Fourier series (3.2) for a periodic function f (t) can be written in the alternative form

1
F S(t) = a0 + An sin (2vn t + n ) , (3.17)
2 n=1

where ( )1
An = a2n + b2n 2 , (3.18)
an
sin n = 1/2
, (3.19)
(an + b2n )
2

bn
cos n = 1/2
, (3.20)
(an + b2n )
2

and vn = Tn , with n = 1, 2, .... If we suppose that t is a time variable then we see from (3.17) that any periodic
signal f (t) can be synthesised by forming a linear superposition of simple harmonic vibrations which have a discrete
frequency spectrum { }
n
vn = ; n = 1, 2, ... . (3.21)
T
The amplitude An and phase n for the nth harmonic are related to the Fourier coecients an and bn by equations
(3.18)-(3.20).

3.2 The Fourier Integral


Our main aim in this section is to show that it is also possible to synthesise a non-periodic signal f (t) by using
a Fourier integral. We begin by considering an arbitrary piecewise smooth non-periodic function f (t) as shown in
Figure 3.1. Next we introduce an associated periodic function fT (t), with a fundamental period T , which is dened
in the interval [ T2 , T2 ) by
T T
fT (t) = f (t), for t< . (3.22)
2 2
It is clear from (3.22)
[ and) Figure 3.2 that the periodic function fT (t) will provide an exact representation for f (t)
provided that t T2 , T2 . Beyond this interval these two functions are dierent. However, if we allow the period
T then we have
f (t) = lim fT (t) for t (, ). (3.23)
T
CHAPTER 3. FOURIER TRANSFORM 29

fT(t)

f(t)

T/2 0 T/2 t
Figure 3.2: The non-periodic function f (t) (the dashed line) is compared with its periodic (with the period T )
approximation fT (t) (the solid line).

We now substitute the Fourier series (3.14) in (3.23) and apply the formula (3.16) with t0 = T2 . This procedure
gives

f (t) = lim gT (vn )ei2vn t v, (3.24)
T
n=
n
where vn = T,
vn+1 vn = 1/T v, (3.25)
T /2
gT (vn ) f (t)ei2vn t dt. (3.26)
T /2

In the limit T , v 0 and the summation in (3.24) becomes an integral. Hence, we obtain

f (t) = g (v)ei2vt dv, (3.27)

where
g (v) = f (t)ei2vt dt. (3.28)

Finally, the formula (3.28), with t replaced by , is substituted in equation (3.27). Hence we nd that
[ ]
i2v
f (t) = f ( )e d ei2vt dv. (3.29)

This very important result is known as the Fourier Integral representation for the non-periodic function f (t). We
see from (3.29) that if we wish to represent a non-periodic function f (t) over the innite range (, ) then the
Fourier sum over the dicrete frequency spectrum {vn ; n = 1, 2, ...} must be replaced by an integral over a continuous
frequency spectrum. It can be proved that the Fourier integral
representation (3.29) is valid for all piecewise smooth
non-periodic functions f (t), provided that the integral |f (t)|dt exists. If the function f (t) has a discontinuity
at t = t1 then the Fourier integral (3.29) will give the mean value of f (t) at t = t1 (see equations (3.5) and (3.6)).

3.3 Alternative Forms of the Fourier Integral


If we apply the Euler identity
ei2v( t) = cos [2v( t)] i sin [2v( t)] (3.30)
to the formula (3.29) it is found that

f (t) = dv f ( ) cos [2v( t)] d i dv f ( ) sin [2v( t)] d. (3.31)

CHAPTER 3. FOURIER TRANSFORM 30

The second double integral in (3.31) is equal to zero because sin [2v( t)] is an odd function of v. Hence we
obtain
f (t) = 2 dv f ( ) cos [2v( t)] d. (3.32)
0

This trigonometric form for the Fourier integral can also be written in the form

f (t) = 2 dv f ( ) [cos(2v ) cos(2vt) + sin(2v ) sin(2vt)] d. (3.33)
0

If f (t) is either an even or odd function then further simplication of (3.33) is possible. For an even function fe (t)
we nd that [ ]
fe (t) = 2 2 fe ( ) cos(2v )d cos(2vt)dv. (3.34)
0 0

This result is known as the Fourier cosine integral for fe (t). For an odd function f0 (t) we obtain the Fourier sine
integral representation [ ]
fo (t) = 2 2 fo ( ) sin(2v )d sin(2vt)dv. (3.35)
0 0

for fo (t). It is also possible to use the Fourier cosine and sine integrals to represent a general function f (t) which
is only dened in the half-range [0, ). Outside this range the Fourier cosine and sine integrals will give even and
odd extensions of f (t), respectively.

Example 4: Determine the Fourier integral representation for the function


{
1, for |t| 1
(t) =
0, for |t| > 1.

Solution: Because the fucntion (t) is even we can write the Fourier integral in the form (3.34). Hence we nd
that [ 1 ]
(t) = 4 cos(2v )d cos(2vt)dv. (3.36)
0 0

It follows from (3.36) that


sin(2v)
(t) = 4 cos(2vt)dv. (3.37)
0 2v
When t = 1 the Fourier integral (3.37) will give the mean value 12 . You can see how the approximation
T
sin(2v)
T (t) = 4 cos(2vt)dv
0 2v

converges to the exact function (t) in Figs. 3.3 and 3.4. J

3.4 Fourier Transform


The Fourier transform of f (t) is dened as

F (v) f (t)ei2vt dt. (3.38)

This denition is useful because it simplies the structure of the Fourier integral representation (3.29). From (3.38)
we see that the function f (t) has been transformed by a process of integration into a spectral function F (v) in the
frequency domain. If we introduce a functional operator F acting on f (t) which converts f (t) 7 F (v) then we can
write (3.38) in the form
F (v) = F[f (t)] = f (t)ei2vt dt (3.39)

From a physical point of view the Fourier transform essentially gives a spectral analysis of the signal f (t).
CHAPTER 3. FOURIER TRANSFORM 31

T=1
T=4
1

Step function T(t)

0.5

0
-3 -2 -1 0 1 2 3
t

Figure 3.3: Fourier integral for (t) with the range 0 v T for T = 1 and T = 4.

T=10
T=100
Step function T(t)

0.5

-3 -2 -1 0 1 2 3
t

Figure 3.4: Fourier integral for (t) with the range 0 v T for T = 10 and T = 100.
CHAPTER 3. FOURIER TRANSFORM 32

It is also possible to convert F (v) 7 f (t) by using the formal relation f (t) = F 1 [F (v )], where F 1 denotes
the inverse functional operator. Fortunately, an explicit formula for this inversion procedure can be derived by
substituting (3.38), with t replaced by , in the Fourier integral (3.29). Hence, we nd that

f (t) = F 1 [F (v)] = F (v)ei2vt dv. (3.40)

This result is called the inverse Fourier transform of F (v). From a physical point of view the inverse transform
shows how a signal f (t) can be synthesised from its frequency spectrum.
If we have an even function fe (t) then it is seen from (3.38) and the Fourier cosine integral (3.34) that the
Fourier transform pair can be expressed in the alternative form

F (v) = F [fe (t)] = 2 fe (t) cos(2vt)dt, (3.41)
0

fe (v) = F 1 [F (v)] = 2 F (v) cos(2vt)dv. (3.42)
0

For an odd function fo (t) we nd from (3.38) and the Fourier sine integral (3.35) that the Fourier transforn pair is
given by
F (v) = F [fo (t)] = 2i fo (t) sin(2vt)dt (3.43)
0

fo (t) = F 1 [F (v)] = +2i F (v) sin(2vt)dv (3.44)
0

It should be noted that the Fourier transform of an even function fe (t) is a real function of v.

Example 5: Determine the Fourier transform F (v) for the unit impulse function
{
n, for |t| 2n
1
n (t) =
0, for |t| > 2n
1

where n = 1, 2, .... Hence nd the Fourier transform of the Dirac delta function (t) and obtain an integral
representation for (t).

Solution: In this case we have an even function fe (t) = n (t). It follows, therefore, from (3.41) that
1
2n sin(v/n)
F (v) = F [n (t)] = 2n cos(2vt)dt = (3.45)
0 (v/n)

We see that as n increases the width t of n (t) becomes smaller, while the spread v of F (v) becomes larger. In
particular, we nd vt 2 for all n.
In the limit n , we obtain
sin(v/n)
lim F [n (t)] = lim =1 (3.46)
n n (v/n)

Thus, the Fourier transform of the Dirac delta function is



F[(t)] = (t)e i2 vt dt = 1 . (3.47)

The application of the inverse transform (3.40) to the result (3.47) gives the important formal integral representation

(t) = ei2vt dv. (3.48)

This expression for the Dirac delta function we have already met in the rst lecture where it was derived dierently,
using general properties of the delta functions.
CHAPTER 3. FOURIER TRANSFORM 33

Example 6: Determine the Fourier transform for the Gaussian delta function
n
e 2 n
1 2 2
n(1) (t) = 1
t
, (3.49)
(2) 2

where n = 1, 2, ....

Solution: From the denition (3.39) we nd


[ ]
n
e 2 n ei2vt dt.
1 2 2
F (v) = F n(1) (t) = 1
t
(3.50)
(2) 2

The substitution t = y/n in (3.50) gives



1
e 2 y ei2vy/n dy.
1 2
F (v) = 1 (3.51)
(2) 2

From this result we obtain


dF i
ye 2 y ei2vy/n dy.
1 1 2
= (2) 2 (3.52)
dv n

If we integrate the RHS of (3.52) by parts, it is found that



dF v
e 2 y ei2vy/n dy.
3 1 2
= 2 (2) 2 (3.53)
dv n

We see from (3.51) and (3.53) that F (v) satises the dierential equation

dF
= (2/n)2 vF (3.54)
dv
which solution is
F (v) = F (0)e 2 (2/n)
1 2 2
v
, (3.55)
where, from Eq. (3.51),
1
e 2 y dy
1 2
F (0) = 1 (3.56)
(2) 2
( )
is a constant which is independent of the value of n. This integral is related to the Gamma function 12 (see
the rst lecture) and can be calculated to be unity, i.e. F (0) = 1. Below, we shall obtain the same result using a
dierent method.
The constant F (0) can be determined by rst taking the limit n in equation (3.55). This procedure gives

lim F (v) = F (0). (3.57)


n

However, we can also write



lim F (v) = lim n(1) (t)ei2vt dt = F [(t)] = 1. (3.58)
n n

Hence, we see that F (0) = 1 and


F (v) = e 2 (2/n
2
1
)v 2
. (3.59)
(1)
The Gaussian function n (t) has a width t 1/n, while its Fourier transform F (v) is another Gaussian
function with a width v n/(2). It follows, therefore, that
1
vt . (3.60)
2
If we write E = hv with E = hv then (3.60) gives the Heisenberg uncertainty relation Et ~. Nice!
CHAPTER 3. FOURIER TRANSFORM 34

3.5 Fourier Transform of Derivatives


The Fourier transform of the derivative f (t) is given by


F [f (t)] = f (t)ei2vt dt. (3.61)

If we integrate the RHS of (3.61) by parts we nd that



[ ]
F [f (t)] = f (t)ei2vt + i2v f (t)ei2vt dt. (3.62)


In the derivation of (3.62) it is necessary to assume that the integral
|f (t)|dt exists. It follows, therefore, that
limt f (t) = 0. Hence we obtain
F [f (t)] = i2vF (v), (3.63)
where F (v) = F [f (t)]. More generally, it is found that
[ ]
F f (m) (t) = (i2v)m F (v), (3.64)

provided that limt f (j) (t) = 0, where j = 0, 1, 2, ..., m 1. These results are particularly useful when solving
ordinary and partial dierential equations.

3.6 Convolution Theorem


Consider two functions f (t) and g(t) with Fourier fransforms F (v) and G(v), respectively. The convolution of f (t)
and g(t) is dened as
f (t) g(t) f ( )g(t )d. (3.65)

If the inversion formula


g(t ) = G(v)ei2v(t ) dv (3.66)

is substituted in (3.65) and the order of integration is interchanged then we obtain



f (t) g(t) = G(v)e i2vt
dv f ( )ei2v d. (3.67)

Hence we have
f (t) g(t) = F (v)G(v)ei2vt dv. (3.68)

It is clear from this result that


f (t) g(t) = F 1 [F (v)G(v)] , (3.69)
and
F [f (t) g(t)] = F (v)G(v). (3.70)
The formula (3.70) is known as the convolution theorem. We see that the Fourier transform of a convolution
f (t) g(t) is simply equal to the product of the Fourier transform of f (t) and g(t).
It is possible to use the convolution theorem to analyse the output response function for a linear electrical
citcuit. The theorem also plays an important role in diraction theory and image processing.

Example 7: Determine the convolution of the Dirac Delta function (t b) with the function f (t).
CHAPTER 3. FOURIER TRANSFORM 35

Solution: From the denition (3.65) we nd that



(t b) f (t) = ( b)f (t )d. (3.71)

We now apply the ltering theorem for the Dirac Delta function to the RHS of (3.71). Hence we obtain

(t b) f (t) = f (t b). (3.72)

This result is particularly useful for constructing aperture functions in the theory of Fraunhofer diraction. J

3.7 Spatially Varying Functions


Many physical applications involve the Fourier transform of a spacially varying function f (x). Under the circum-
stances it is convenient to change the notation from v u and t x, where u denotes a spatial frequency. The
Fourier transform of f (x) can now be writtten as

F (u) = F [f (x)] = f (x)ei2ux dx, (3.73)

while the inverse transform becomes



1
f (x) = F [F (u)] = F (u)ei2ux du. (3.74)

The Fraunhofer diraction pattern formed when light waves pass through an aperture system consisting of
N = 1, 2, ... parallel slits can be expressed in terms of the spatial Fourier transform of the aperture function A(x).
In particular, it can be shown that the intensity I() of light diracted through an angle is given by

I() = I(0)|F (u)/F (0)|2 , (3.75)

where
F (u) = F [A(x)] = A(x)ei2ux dx (3.76)

and
u = sin /. (3.77)
[In the derivation of (3.75) it has been assumed that the incident light is normal to the plane of the aperture system.]

Example 8: Use the Fourier transform method to calculate the intensity function I() for light which is diracted
through a single slit of width a. In this case we have the aperture function
{
1, for |x| a/2
a (x) =
= 0, for |x| > a/2

The Fourier transform of the aperture function a (x) is given by


a/2
sin(ua)
F (u) = 2 cos(2ux)dx = a . (3.78)
0 (ua)

From this result and equation (3.75) we nd that


[ ]2
sin(a sin /)
I() = I(0) , (3.79)
(a sin /)
CHAPTER 3. FOURIER TRANSFORM 36

3.8 Parcevals Therorem


Parcevals throrem in the context of the Fourier transforms is similar to the theorem which occurs in Fourier series.
It states that, if F () and G() are the Fourier transforms of the functions f (x) and g(x) respectively, then

I= F ()G()d = f (x)g(x)dx. (3.80)

Using the denition of the Fourier transform I can be written as


( ) ( )
2ix 2iy
I= f (x)e dx g(y)e dy d


= f (x)e2ix g(y)e2iy dxdyd

( )
= e2i(xy) d f (x)g(y)dxdy



= (x y)f (x)g(y)dxdy = f (x)g(x)dx. (3.81)

In the above (x) is the Dirac delta function, and we have used its integral representation (3.48) and the ltering
theorem.
A consequence of Parcevals theorem is that if f (x) = g(x), then

|f (x)| dx =
2
|F ()|2 d. (3.82)

3.9 Application of the FT to the Poison equation [optional ]


The electrostatic potential (
x ) in the 3D space due to an arbitrary charge distribution of density (

x ) is described
by the Poison equation
2 2 2
= + + = 4(
x) (3.83)
x21 x22 x23
where
x = (x1 , x2 , x3 ) is used for a vector in 3D. We would like to solve (3.83) for (
x ) for a general (
x ).



To this end, let us expand both ( x ) and ( x ) in the Fourier integral. Since these are functions of 3 variables,
we rst expand them with respect to the rst variable x1 , e.g.

(
x ) = (1 , x2 , x3 ) ei21 x1 d1 (3.84)

where (1 , x2 , x3 ) F [ (x1 , x2 , x3 )]. Similarly, the image (1 , x2 , x3 ), which still depends on x2 and x3 , can also
be expanded into the Fourier integral with respect to x2 :

(1 , x2 , x3 ) = (1 , 2 , x3 ) ei22 x2 d2 , (3.85)

where the image here is expanded with respect to x3 :



(1 , 2 , x3 ) = (1 , 2 , 3 ) ei23 x3 d3 . (3.86)

Combining equations (3.84) to (3.86), we get the original function (x1 , x2 , x3 ) expanded in a triple Fourier integral:

(x1 , x2 , x3 ) = (1 , 2 , 3 ) ei2(1 x1 +2 x2 +3 x3 ) d1 d2 d3 (3.87)

which can then be rewritten in the vector form as





(

x)= (

)ei2( x ) d

. (3.88)
CHAPTER 3. FOURIER TRANSFORM 37

Similarly,


(

x)= (

)ei2( x ) d

. (3.89)

Then, substitute (3.88) and (3.89) into (3.83). We should calculate


( 2 )
2 2

( (

)ei2( x ) d


x)= 2 + 2 + 2
x1 x2 x3
( 2 )

2 2


+ 2 ei2( x ) d

= ( ) 2 + 2
x1 x2 x3
It is easy to see that, e.g.

2 i2(



x) 2 i2(1 x1 +2 x2 +3 x3 ) 2
e = e = (i21 ) ei2(1 x1 +2 x2 +3 x3 )
x21 x21
2 2
and the same for x22
and x23
. Summing all three contributions up:



( )


ei2( x ) = (i2) 12 + 22 + 32 ei2( x ) = (2) 2 ei2( x )
2 2

so that we obtain:


(v) 2 ei2( x ) d

2
= (2) .

On the other hand, it is equal to (


x ) given by (3.89):
[ ]



(2) (
) 2 ei2( x ) d

= 4 (

)ei2( x ) d

2
,

or [ ]

(2) (

) 2 + 4(

) ei2( x ) d

2
=0

We see that the zero function (


x ) 0 (in the RHS) is expanded in the triple Fourier integral (the LHS) as in
equations (3.88), (3.89). Therefore, its Fourier image is also zero:

(2) (

) 2 +4(

2
)=0

(

(
)
)= (3.90)
2
Thus, the required potential, from (3.90) and (3.88):


( ) i2(



x)
d
1
( ) = e . (3.91)
2
We substitute here


(

)= (

y )ei2( y ) d

y. (3.92)

that is the FT of (

x ) (inverse to equation (3.89)):


d i2(



(
y )ei2( y ) d

1 x)
(x) = 2
e y

[ ]
d i2
(
x

(
y )d
(
y )J ( x
y ) d

1 y) 1
= y e = y (3.93)
2
where
d

i2



J (

x

y)= 2
e (xy) (3.94)

CHAPTER 3. FOURIER TRANSFORM 38



The last integral, J( R ), can be actually calculated as follows. First, we note that it cannot depend on the actual




direction of R , only on its length R = | R |. Thus, we can choose R along the 3 axis:


d
i2


R d1 d2 d3 i2R cos
J( R ) = 2
e = e
2
( )



where = R , 3 is the angle between the axis and the vector R . Then, we introduce the spherical coordinates
(, , ):
d1 d2 d3 = 2 sin ddd
2 2

d i2R cos
J( R ) = e sin d d
0 2 0 0

= 2 d ei2R cos sin d (3.95)
0 0
In the d integral we make the following substitution: t = cos dt = sin d and
1 1
ei2R cos d = ei2Rt (dt) = ei2Rt dt
0 1 1

1 ( i2R ) 2i sin (2R) sin (2R)


= e ei2R = =
i2R i2R R
and from (3.95) then: [ ]


sin (2R) 2 sin x
J( R ) = 2 d = dx =
0 R R 0 x R


since the integral in the square brackets is equal to 2 . Therefore, from (3.93) (note, R =
x
y ):

(

(
y)
x)= d
|x

y
y|

which is the famous Coulomb formula!


Chapter 4

Laplace Transform

4.1 Denition
It is yet another example of the integral transform (FT is the one that you are already familiar with):

L (f (t)) = f (t)ept dt F (p) (4.1)
0

where f (t) is a function in the t-space, while its transform, F (p) = L(f ), is the corresponding function in the
p-space, where p is generally a complex number. It does not matter what f (t) is at t < 0; however, it is desirable
to set f (t) = 0 at t < 0. The LT is a linear operator:

L(f + g) L(f ) + L(g) (4.2)

L(cf ) cL(f ) (4.3)


where c is a complex number.

Example 1 Calculate the LT of the function f (t) = 1.

Solution:

1 1
L(f ) = ept dt = ept = , if Re(p) > 0
0 p 0 p
The condition Re(p) > 0 is needed for the convergence of the integral at the upper limit since for p = x + iy we
have
ept = ext eiyt 0 at t =
only if x > 0.

Example 2 Calculate the LT of f (t) = et .

Solution:

(p+)t e(p+)t 1
L(f ) = e dt = =
0 (p + ) 0 p+
provided that Re(p + ) > 0.

Example 3 Calculate the LT of the function f (t) = eit = cos t + i sin t.

39
CHAPTER 4. LAPLACE TRANSFORM 40

Solution: Use the previous formula with = i:


1 p + i p
L(f ) = = 2 = 2 +i 2 (4.4)
p i p + 2 p + 2 p + 2
Note that p here could also be chosen as real. Then, due to linearity of the LT, we also have:
p
L(cos t) = (4.5)
p2 + 2

L(sin t) = (4.6)
p2 + 2
Nevertheless, this result is valid for any complex p.
The necessary condition here is derived from that of example 2 above:

Re(p i) = Re(p) > 0

(assuming is real).

Example 4 Calculate the LT of the function f (t) = t sin t.

Solution:
1 ( it )
L(f ) = ept t sin tdt = e eit dtept t
0 0 2i
[ ]
1
= t e(pi)t e(p+i)t dt
2i 0
Use integration by parts in each term:


1 e(pi)t e(p+i)t 1
L(f ) = t t e(pi)t dt
2i (p i) 0 (p + i) 0 0 (p i)
| {z } | {z }
=0 =0
]
1
+ e(p+i)t dt
0 (p + i)
[ ]
1 1 ( ) 1 ( )
= L eit L eit
2i p i p + i
[ ]
1 1 p + i 1 p i
= 2
2i p i p + 2 p + i p2 + 2
1 (p + i)2 (p i)2 1 4ip 2p
= 2 = 2 = 2 (4.7)
2i 2 2
(p + ) 2i (p + )
2 2 (p + 2 )
2

Another, much simpler method: note that



p
L(cos t) = ept cos tdt
p2 + 2

and then dierentiate both sides with respect to :



2p
ept (t sin t) dt L (t sin t) 2,
0 (p2 + 2 )

which gives the same result.


CHAPTER 4. LAPLACE TRANSFORM 41

4.2 LT of derivatives
Let us assume that for some f (t) the LT is known and is

L (f (t)) = F (p) (4.8)

What is the LP of f (t)? Use the denition and then integrate by parts:

df
L (f (t)) = ept dt = f ept 0 f (t)(p)ept dt
0 dt 0

= f ()ep f (0) + p ept f (t)dt = f (0) + pL(f ),
0
So
L (f (t)) = pL (f (t)) f (0) (4.9)

For the 2nd derivative we can use the above formula since f (t) = g (t) with g(t) f (t):

L(f ) = L(g ) = pL(g) g(0)

= pL(f ) g(0) = p (pL(f ) f (0)) f (0),


L(f ) = p2 L(f ) pf (0) f (0) (4.10)
( )
This way it is possible to obtain a general formula for L f (n) , if needed.

4.3 Solution of Ordinary Dierential Equations (ODEs)


The LT may become very convenient when solving a certain class of ODEs with the right hand side (i.e. inhomoge-
nous ones). Consider, for instance, the following ODE problem:
{
y + 4y + 4y = t2 e2t
(4.11)
y(0) = 0, y (0) = 0

Note that normally we would look for a general solution of the homogeneous solution with two arbitrary constants;
then we would try to nd a partial integral, i.e. a function which satises the whole equation with the t2 e2t ;
nally, we would use the initial conditions
y(0) = y (0) = 0
to obtain the arbitrary constants. Using the LT method, all these can be done in one go: dene L (y(t)) = Y (p).
Then, take the LT of both sides of the ODE:
( )
L(y ) + 4L(y ) + 4L(y) = L t2 e2t

Here:
L(y ) = p2 L(y) py(0) y (0) = p2 L(y) = p2 Y (p),
L(y ) = pL(y) y(0) = pL(y) = pY (p),
L(y) = Y (p)
( )
and L t2 e2t is obtained by integrating by parts twice:

( )
L t2 e2t = pt 2 2t
e t e dt = t2 e(p+2)t dt
0 0

e(p+2)t 1
=t 2
2te(p+2)t dt
(p + 2) 0 (p + 2) 0
[ ]
2 e(p+2)t 1 (p+2)t
=+ t e dt
p+2 (p + 2) 0 (p + 2) 0
CHAPTER 4. LAPLACE TRANSFORM 42

2 ( ) 2 1 2
= 2
L e2t = 2
=
(p + 2) (p + 2) p + 2 (p + 2)3
so that we obtain the following algebraic equation for Y (p):
2
p2 Y (p) + 4pY (p) + 4Y (p) =
(p + 2)3

which yields:
2 1 2
Y (p) = = (4.12)
(p + 2)3 p2 + 4p + 4 (p + 2)5
What we want here is the inverse LT of this function. It can easily be seen to be
( )
1 2 1 4 2t
L 5
= t e (4.13)
(p + 2) 12

Problem 1: Check by direct calculation that


( ) 2
L t4 e2t = 12 (4.14)
(p + 2)5

Solution of ODEs continued... Therefore, the solution of the ODE above is


1 4 2t
y(t) = t e
12
It satises the ODE and the intital conditions.
Another example: {
y + 4y = sin 2t
(4.15)
y(0) = 10, y (0) = 0
y (t) L(y ) = p2 Y (p) py(0) y (0) = p2 Y 10p,
2
sin 2t L(sin 2t) = ,
p2 +4
so that we obtain
2
p2 Y 10p + 4Y =
p2 + 4
( ) 2
p2 + 4 Y = + 10p
p2 + 4
2 10p
Y (p) = 2 + .
(p2 + 4) p2 + 4
Inversely, {
2
(p2 +4)2
1
8 (sin 2t 2t cos 2t)
(4.16)
10p
p2 +4 10 cos 2t
and the nal solution is
1
y(t) = 10 cos 2t + (sin 2t 2t cos 2t)
8
which satises problem (4.15). (Note: see the tables of the Laplace Transform at the end of the lecture!)
CHAPTER 4. LAPLACE TRANSFORM 43

t2
Integration Area

t1

Figure 4.1: Interation area in Eq. (4.21).

4.4 Simplifying Complicated Fractions


Sometimes, Y (p) looks very complicated as there is no solution for the inverse LT in the LT tables. In many cases,
however, fractions can be split into a sum of simpler ones which might be in the table. Here is an example:
1 A B
= + (4.17)
(p + a)(p + b) (p + a) (p + b)

A and B are constants to be found:


(p + a)B + (p + b)A = 1
p(A + B) + (aB + bA) = 1
{ {
A+B =0 A = B 1 1 1
B= and A = =
aB + bA = 1 (a b)B = 1 ab ab ba
[ ]
1 1 1 1
= +
(p + a)(p + b) ab p+a p+b
so that ( ) [ ( ) ( )]
1 1 1 1 1 [ at ]
L1 = L1 + L1 = e + ebt (4.18)
(p + a)(p + b) ab p+a p+b ab

4.5 Convolution Theorem


Let
G(p) = L (g(t)) = ept1 g (t1 ) dt1 (4.19)
0

H(p) = L (h(t)) = ept2 h (t2 ) dt2 (4.20)
0
Then consider
G(p)H(p) = dt2 dt1 ep(t1 +t2 ) g (t1 ) h (t2 ) (4.21)
0 0

This is a double integral in the (t1 t2 ) plane, in the integration area is shown in Fig. 4.1. Then, we make the
following change of variables in the inner (with respect to t1 ) integral: t1 t = t1 + t2 , i.e. t1 = t t2 , dt1 = dt
and
t1 = 0 t = t2
t1 = t =
and we obtain: [ ]
pt
G(p)H(p) = dt2 dte g (t t2 ) h (t2 ) (4.22)
0 t2

The integration area here is shown in Fig. 4.2.


CHAPTER 4. LAPLACE TRANSFORM 44

t2 = t
t2

Figure 4.2: Integration area in Eq. (4.22).

t2 = t
t2

Figure 4.3: Integration area in Eq. (4.23).

What we want now is that the t2 -integral become an internal one. To accomplish this, we change the integration
pattern from horizontal to the equivalent vertical lines as shown in Fig. 4.3. This gives:
[ t ]
G(p)H(p) = dt dt2 ept g (t t2 ) h (t2 ) (4.23)
0 0
[ t ]
= dtept g (t t2 ) h (t2 ) dt2
0 0

Thus, the LT of t
y(t) = g (t t ) h(t )dt g h(t) (4.24)
0
is
L (y(t)) = G(p)H(p) (4.25)
The function y(t) is called the convolution of g and h (compare the FT lectures!). This result is extremely important
and is, in fact, similar to the FT of the convolution of two functions.

Example of application: {
y + 3y + 2y = f (t)
(4.26)
y(0) = y (0) = 0
If L (f (t)) = F (p) and L (y(t)) = Y (p), then making the LT we obtain:

p2 Y py(0) y (0) + 3 (pY y(0)) + 2Y = F (p)

p2 Y + 3pY + 2Y = F
1
Y (p) = F (p)
p2 + 3p + 2
is a product of two transforms. Since
L1 (F (p)) = f (t)
CHAPTER 4. LAPLACE TRANSFORM 45

) ( ( )
1 1
L1
= L 1
p2 + 3p + 2 (p + 1)(p + 2)
( ) ( ) ( )
1 1 1 1 1 1 1
=L =L L = et e2t ,
p+1 p+2 p+1 p+2
then we obtain, using (4.24) and (4.25):
t[ ] t [ ]

y(t) = e(tt ) e2(tt ) f (t )dt or f (t t ) et e2t dt
0 0

Either form is, of course, valid! We have obtained a general solution of the ODE (4.26) without even knowing the
actual form of the function f (t) in the RHS!

4.6 A more rigorous consideration of the LT, its relation to the FT and
the inverse LT (not in the syllabus)
The consideration provided above was somewhat formal and not really rigorous. In particular, it was not clear
what conditions the function f (t) must satisfy for the LT to exist. Also, no mention has been made of the direct
calculation of the inverse LT (ILT). Here we shall introduce the LT from the FT which would show the relationship
between the two and would also allow us to derive a direct formula for the ILT. At the same time, we shall also
derive a sucient condition for the LT to exist and be an analytical function in the complex plane.
Theorem: if f (t) is of an exponential growth, i.e. it goes to innity not faster than the exponential function
ep0 t with some positive p0 > 0, then the LT L[f (t)] = F (p) of f (t) is an analytical function of p in the complex
semiplane Re(p) > p0 .
Proof : If f (t) is of the exponential growth, this means that

|f (t)| < M ep0 t or |f (t)| ep0 t < M , (4.27)

where M > 0 is a positive constant. In other words, f (t) times the exponential ep0 t is always limited by some
positive M , i.e. one can see that this conditions indeed means that f (t) grows not faster than the exponential e+p0 t .
The real number p0 may be considered as a characteristic exponential of the function f (t). Now consider the LT
integral:

F (p) = f (t)ept dt
0

If this integral exists, then it should converge absolutely, i.e. |F (p)| < (must be nite). Let us estimate its
modulus:

|F (p)| = f (t)ept dt f (t)ept dt = |f (t)| ept dt
0 0 0

Here we used a well known inequality (see below) that the absolute value of a sum is always smaller or equal to the
sum of the absolute values. Since the integral is a sum, then we can always write what was written above. Then,
p = x + iy is a complex number, so
pt (x+iy)t xt iyt xt iyt
e = e = e e = e e = ext |cos (yt) i sin (yt)| = ext cos2 (yt) + sin2 (yt) = ext ,

so that


M
|f (t)| ept dt = |f (t)| ext dt M ep0 t ext dt = M e(xp0 )t dt = e(xp0 )t
0 0 0 0 x p0 0

If Re (x p0 ) = x Re(p0 ) > 0, i.e. if Re(p) = x > Re(p0 ), then the value of the expression above at t = is zero
and we obtain
M

|F (p)| = pt
f (t)e dt , (4.28)
0x p0
CHAPTER 4. LAPLACE TRANSFORM 46

which means that the LT F (p) is nite, i.e. the LT integral converges absolutely. Similarly, we can consider the
derivative
dF (p) d d ( pt )
= f (t)ept dt = f (t) e dt = f (t)tept dt
dp dp 0 0 dt 0
We can proceed in the same way as before to see if this derivative is nite:

dF
= pt
f (t)te dt = pt
f (t)te dt xt
|f (t)| te dt M p0 t xt
e te dt = M te(xp0 )t dt
dp
0 0 0 0 0

This integral can be calculated by parts in the usual way:



(xp0 )t 1
(xp0 )t 1
te dt = t e + x p0 e(xp0 )t dt
0 (x p0 ) 0 0

The rst term is equal to zero both at t = 0 (because of the t present there) and at t = if the condition
x = Re(p) > Re(p0 ) is satised, as above (note that the exponential et with > 0 tends to zero much faster
than any power of t when t ). Then, only the integral remains which is calculated in the same way as before:

1 1 (xp0 )t 1
te(xp0 )t dt = e(xp0 )t dt = e =
0 x p0 0 (x p0 )
2
0 (x p0 )
2

Finally, we conclude that


dF M
M te(xp0 )t dt =
dp 2 ,
0 (x p0 )
i.e. it is indeed nite. This means that the function F (p) is analytical (can be dierentiated). J

An example of a good function f (t) is, for instance, the exponential e2t . It goes to innity when t +,
however, this happens not faster than ep0 t with p0 > 2. Therefore, the LT of this function does exist in the semiplane
2
Re(p) > 2. However, the function f (t) = e2t grows much faster than the exponential function ep0 t with any p0 ,
and hence its LT does not exist.
Now let us briey consider the relationship between the FT and LT which will allow us to derive an expression
for the ILT as well. Consider a function f (t) which is zero for any t < 0. We complement f (t) with an extra
exponential factor ext with some real parameter x, i.e. we shall consider gx (t) = f (t)ext . The FT of it will
depend on both x and :

Fx () = gx (t)ei2t dt = f (t)ext ei2t dt = f (t)e(x+i2)t dt (4.29)
0 0

Note that we replaced the bottom integration limit by zero as f (t) = 0 for any negative t. The inverse FT is
{
i2t gx (t) = f (t)ext , t > 0
Fx ()e dt =
0 , t<0

Multiplying both sides of this equation by ext , we obtain:


{
(x+i2)t f (t) , t > 0
Fx ()e d = (4.30)
0 , t<0

The number x + i2 is some complex number p, so that Eqs. (4.29) and (4.30) can also be alternatively written
as:
Fx () = f (t)ept dt (4.31)
0
and
f (t) = Fx ()ept d (4.32)

One can recognise in Eq. (4.31) the LT of the function f (t), i.e. Fx () L [f (t)] = F (p). In the other equation
(4.32) we shall change the variables from to p = x + i2. This gives:
x+i
1
f (t) = F (p)ept dp (4.33)
2i xi
CHAPTER 4. LAPLACE TRANSFORM 47

This formula provides a recipe for the inverse LT. One can see that in order to calculate f (t) from its LT F (p),
one has to perform an integration in the complex plane of the function F (p)ept along the vertical line Re(p) = x
from to +. This calculation can only be done using certain results of complex calculus which are beyond the
current course. However, one can appreciate from this brief encounter that there is indeed a very close relationship
between the two integral transforms.

Here we shall prove the inequality we used above a number of times, which is that the absolute value of a sum of
complex numbers is always smaller or equal to the sum of their absolute values, i.e. for any set of complex numbers
{z1 , z2 , z3 , . . .}, one has:


zi |zi | (4.34)

i i

Firstly, it is sucient to prove this for only two complex numbers. Indeed, if this was valid for any two, then for
e.g. three
|z1 + z2 + z3 | = |z1 + (z2 + z3 )| |z1 | + |z2 + z3 | |z1 | + |z2 | + |z3 |
it would be valid as well, as required.
To prove it for two complex numbers, write z1 = x1 + iy1 and z2 = x2 + iy2 . Then,

2 2
|z1 + z2 | = |x1 + x2 + i (y1 + y2 )| = (x1 + x2 ) + (y1 + y2 ) = (x21 + y12 ) + (x22 + y22 ) + 2 (x1 x2 + y1 y2 ) ,

|z1 | = x21 + y12 and |z2 | = x22 + y22 .
Assume the opposite, i.e. that
|z1 + z2 | > |z1 | + |z2 | (4.35)
The RHS is obviously positive, and so is the LHS which is larger it. Then, we are allowed to square both sides:
2 ( ) ( ) 2 ( ) ( )
|z1 + z2 | x21 + y12 + x22 + y22 + 2 (x1 x2 + y1 y2 ) > (|z1 | + |z2 |) x21 + y12 + x22 + y22 + 2 |z1 | |z2 |

Cancelling terms in both sides and then deviding by 2, we obtain:



x1 x2 + y1 y2 > |z1 | |z2 | x21 + y12 x22 + y22

Again, both sides are positive (since the RHS obviously is), so we can sqaure them both again:
2 2
(x1 x2 ) + (y1 y2 ) + 2x1 x2 y1 y2 > x21 x22 + x21 y22 + y12 x22 + y12 y22

After obvious simplications, we obtain:


2
0 > x21 y22 + y12 x22 2x1 x2 y1 y2 (x1 y2 y1 x2 )

which is obviously wrong! Therefore, our assumption (4.35) was incorrect, and hence, the opposite is in fact true,
Q. E. D.
CHAPTER 4. LAPLACE TRANSFORM 48

Table 4.1: Laplace transforms of some functions


y = f (t), t > 0
Y = L(y) = F (p) = 0 ept f (t)dt Range of p in the complex plane
[y = f (t) = 0, t < 0]
1
L1 1 p Re p > 0
L2 eat 1
p+a Re (p + a) > 0
L3 sin at a
p2 +a2 Re p > |Im a|
L4 cos at p
p2 +a2 Re p > |Im a|
L5 tk , k > 1 k!
pk+1
or (k+1)
pk+1
Re p > 0
L6 tk eat , k > 1 k!
(p+a)k+1
or (k+1)
(p+a)k+1
Re (p + a) > 0
eat ebt 1
L7 ba (p+a)(p+b) Re (p + a) > 0
aeat bebt p
L8 ab (p+a)(p+b) Re (p + b) > 0
L9 sinh at a
p2 a2 Re p > |Re a|
L10 cosh at p
p2 a2 Re p > |Re a|
L11 t sin at 2ap
(p2 +a2 )2
Re p > |Im a|
p a2
2
L12 t cos at (p2 +a2 )2
Re p > |Im a|
at
L13 e sin bt b
(p+a)2 +b2 Re (p + a) > |Im b|
L14 eat cos bt p+a
(p+a)2 +b2 Re (p + a) > |Im b|
a2
L15 1 cos at p(p2 +a2 ) Re p > |Im a|
a3
L16 at sin at p2 (p2 +a2 ) Re p > |Im a|
2a3
L17 sin at at cos at (p2 +a2 )2
Re p > |Im a|
L18 eat (1 at) p
(p+a)2 Re (p + a) > 0
L19 sin at
arctan ap Re p > |Im a|
t ( )
1 1 a+b ab
L20 t sin at cos bt, a > 0, b > 0 2 arctan p + arctan p Re p > 0
eat ebt p+b
L21 ln p+a Re (p + a) > 0 and Re (p + b) > 0
( t )
1 a p
L22 1 erf 2a
t
, a>0 pe Re p > 0
( 2 )
2 1/2
L23 { J0 (at) p +a Re p > |Re a|, or for real a = 0, Re p 0
1, t > a > 0 1 pa
L24 f (t) = pe Re p > 0
0, t<a
Chapter 5

Vector Calculus: curvilinear coordinates


(Part I)

Example through cycildrical cord.

5.1 Scalar and Vector Fields


If to each point P in a region R there corresponds a scalar number (P ), then is called a scalar function of
position and we say that a scalar eld has been dened in R, see Fig. 5.1. For example, the temperature T at any
point within or on the surface of the Earth at a certain time denes a scalar eld, see Fig. 5.2.



If to each point P in a region R there corresponds a vector F (P ), then F is called a vector function of position


and we say that a vector eld F has been dened in R, see Fig. 5.3.
For example, if the velocity at any point P within a moving liquid is known at a certain time, then a vector
eld has been dened in R, see Fig. 5.4.

5.2 Coordinate Systems


The position of a point P can be specied by giving the Cartesian coordinates (x, y, z) for P , with respect to a
xed rectangular frame of reference.
In many problems it is often more convenient to change from Cartesian coordinate system (x, y, z) to a dierent
system of coordinates. Such a coordinate transformation can be carried out by using a well-dened procedure to
associate a new set of coordinates (q1 , q2 , q3 ) with each point P (x, y, z) in the region R. We shall assume that the

R .P
(P)

Figure 5.1: Scalar eld in a region R.

49
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 50

.P
.
T(P)

Figure 5.2: Example of a scalar eld.

P .

F(P)

Figure 5.3: Vector eld in region R.

P . v(P)

Figure 5.4: Velocities of particles in a liquid owing in a pipe represents as an example of a vector eld.

P x
. (x,y,z)
y

r
z
0

y

Figure 5.5: Cartesian, cylindrical and spherical coordinate systems representations for point P .
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 51

relationship between the two sets of coordinates can be described by transformation functions:

x = x (q1, q2 , q3 ) ,
y = y (q1, q2 , q3 ) , (5.1)
z = z (q1, q2 , q3 ) .

For each point P (x, y, z) in R we shall also suppose that the equations (5.1) can be solved to give a unique solution
set
qi = qi (x, y, z) , (i = 1, 2, 3). (5.2)
In practice, one often nds that for certain points P (x, y, z) the solutions (5.2) are not uniquely dened. Special
points of this type are called singular points of the coordinate transformation.
The new coordinates (q1 , q2 , q3 ) dened by the transformation functions (5.1) and (5.2) are called Curvilinear
Coordinates for the point P (x, y, z).

Example 1 For the case of cylindrical coordinates (q1 , q2 , q3 ) (, , z) we have the transformation functions:

x = cos ,
y = sin , (5.3)
z = z,

where 0 < , 0 < 2 and < z < +, see Fig. 5.5. Note that for points x = 0, y = 0 the angle is
indeterminate. Thus , the points on the z-axis are singular points of the transformation.
From equations (5.3) we readily obtain the inverse relations
( )1/2
= x2 + y 2 , (5.4)
z = z.

Problem 1 Determine the remaining inverse relation = (x, y, z).

Example 2

For the case of spherical polar coordiantes (q1 , q2 , q3 ) (r, , ) we have the transformation functions

x = r sin cos ,
y = r sin sin (5.5)
z = r cos ,

where 0 r < , 0 and 0 < 2, see Fig. 5.5. From these equations we readily obtain the inverse
relation ( )1/2
r = x2 + y 2 + z 2 . (5.6)

Problem
2 Determine the inverse relations = (x, y, z) and = (x, y, z). What are the singular points of the transfor-
mation (5.5)?

We see that it is possible to represent the position of any point P in a reagion R using an arbitrary Curvilinear
Coordinate system (q1 , q2 , q3 ). Our next task is to set up a general procedure for representing a general vector

F (P ) in terms of curvilinear coordinates (q1 , q2 , q3 ).


CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 52

q line
2
q1=c1
e2 F(p)
q2=c2

e1 e3

q line
q3=c3 q line
1 3
0 y

Figure 5.6: Coordinate lines and unit base vectors in a general curvilinear coordinate system (q1 , q2 , q3 ).
z ez

line
0 e

P( , ,z) e

0

y
Q

Figure 5.7: Coordinate lines and surfaces in cylindrical coordinates.

5.3 Coordinate Surfaces and Lines


Consider the subset of points in R which satisfy the constraint q1 = c1 , where c1 is a real constant, Fig. 5.6. We
see from (5.2) that these points must lie on a coordinate surface described by the equation

q1 (x, y, z) = c1 . (5.7)

In a similar manner the constraints q2 = c2 and q3 = c3 are satised by points on the coordinate surfaces

q2 (x, y, z) = c2 , (5.8)

q3 (x, y, z) = c3 , (5.9)
respectively.
It is clear that the three coordinate surfaces will all intersect at a point P which will have curvilinear coordinates
(c1 , c2 , c3 ). The coordinate surfaces q2 (x, y, z) = c2 and q3 (x, y, z) = c3 will intersect along a space curve which
passes through P . As one moves along this space curve the coordinate q1 will vary, with q2 = c2 and q3 = c3 both
constant. For this reason we call the intersection curve the q1 coordinate line through P . In a similar manner we
use the other pairs of coordinate surfaces to dene the q2 and q3 coordinate lines.

Example 3 (1st part) Consider the case of cylindrical coordinates as in Fig. 5.5. Determine all corrdinate lines
and surfaces.

Solution: The coordinate surfaces for this case are (see Fig. 5.7):
= c1 , (cylinder coaxial with z-axis)
= c2 , (plane hinged along z-axis)
z = c3 , (horizontal plane through P ).
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 53

The coordinate lines are:


line, (line O P extended outwards)
line, (horizontal circle of radius through P )
z line, (vertical line QP extended).

5.4 Unit Base Vectors


Next we introduce a set of vectors {
e1 ,

e2 ,

e3 } at the intersection point P , where
ei is the unit tangent vector to
the qi coordinate line in the direction of increasing qi , where i = 1, 2, 3. It is now possible to represent an arbitrary


vector F (P ) in the form


F (P ) = 1
e1 + 2
e2 + 3
e3 , (5.10)


where i (i = 1, 2, 3) are scalar quantities. The vectors
e1 ,

e2 ,

e3 are called unit base vectors for the vector F (P ).




It should be noted that the direction of the base vectors e , e , e will usually depend on the position of the point
1 2 3
P.
To determine the base vectors

e1 ,

e2 ,

e3 we consider the general position vector





r =x i +y j +zk, (5.11)

and introduce the transformation equations x = x (q1 , q2 , q3 ), y = y (q1 , q2 , q3 ) and z = z (q1 , q2 , q3 ). In this manner,
we obtain
r =r (q1 , q2 , q3 ). Now we evaluate the partial derivatives:



r x y z
= i + j + k , (i = 1, 2, 3), (5.12)
qi qi qi qi

at the point P . The derivative


r /qi is a tangent vector to the qi coordinate line at P in the direction of
increasing qi , see Fig. 5.6. Hence,


= hi

r
ei , (i = 1, 2, 3) (5.13)
qi
where hi is a positive number which usually depends on the position of the point P . From (5.13) it follows that:
( )

1 r
ei = , (5.14)
hi qi


r

hi = . (5.15)
qi
The quantity hi is called a scale factor. As you can see, it is simply chosen to make unit base vectors of the unit
length.

5.5 Orthogonal Coordinate Systems


If the set of vectors {

e1 ,

e2 ,

e3 } satises the relations


e1

e2 =

e2

e3 =

e3

e1 = 0 (5.16)

at every point P in R, then we say that the curvilinear coordinate system (q1 , q2 , q3 ) is orthogonal. For an orthogonal
system the coordinate surfaces through any point P will all intersect at right angles.

Example 3 (2nd part) Consider again the case of cylindrical coordinates, and determine the unit base vectors.
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 54

Solution: We should construct the tangent vectors


e ,

e ,

ez at the point P . It is seen from the diagram that the
coordinate system is orthogonal.
To derive formulae for

e ,

e ,

ez we use the transformation equations x = cos , y = sin , z = z to write






r =x i +y j +zk,

in the form





r =

r (, , z) = cos i + sin j + z k .
We now apply equations (5.14) and (5.15) to this result. It is found that:


r


r ( 2 )1/2
= cos i + sin j , h = cos + sin2 =1



r


r ( 2 2 )1/2
= sin i + cos j , h = sin + 2 cos2 =



r

r
= k , hz = 1.
z z
Hence, we have




e = cos i + sin j ,




e = sin i + cos j , (5.17)



e = k.
z

It is easily veried that ]




e

e =

e

e =

ez

ez = 1,


e
e =
e
e =
e
e = 0.
z z

These results show explicitly that the coordinate system is orthogonal.


It is interesting to note that equations (5.17) can be written in the matrix form:


e i

e = M j ,


ez

k

where M denotes the matrix


cos sin 0
M = sin cos 0 .
0 0 1
The matrix M is orthogonal with M1 = MT . Thus, we have the inverse relation

i
e


j = M e ,
1



ez
k

with
cos sin 0
M1 = sin cos 0 .
0 0 1

Example 4 Consider the case of Cartesian coordinates (x, y, z) as in Fig. 5.8. The coordinate surfaces for this
case are planes parallel to the yz, xzand xyplanes. The coordinate lines through P are parallel to the x, y




and zaxes. Thus the unit base vectors are
ex = i , ey = j and
ez = k for all points P . We also have
hx = hy = hz = 1.
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 55

ez

ey
P

ex

0 y

Figure 5.8: Cartesian coordinates.

Problem 3 Describe the coordinate surfaces and lines for spherical polar coordinates (r, , ). Determine formulae


for the unit base vectors er ,

e ,

e in terms of the unit vectors i , j , k . Use a matrix method to express the vectors


i , j , k in terms of

er ,

e ,

e .

5.6 Line Elements


If we make a small change in the coordinates of a point P from (q1 , q2 , q3 ) to (q1 + dq1 , q2 + dq2 , q3 + dq3 ) then to
rst order the change in the position vector

r (q1 , q2 , q3 ) is given by the dierential line element
3 ( )

3
d
hi dqi

r
r = dqi = ei . (5.18)
i=1
qi i=1

Note that (5.18) is valid for general non-orthogonal curvilinear coordinate systems.
It follows from (5.18) that the length ds of the displacement vector d
r is given by


3
3
(ds) = d

r d

2
r = gij dqi dqj , (5.19)
i=1 j=1

where ( ) ( )

gij hi hj (

ei
r r
ej ) = (5.20)
qi qj
is called the metric coecient. We see from (5.20) that the metric coecient satises the relation gij = gji . It is
also possible to write (5.19) in the matrix form

dq1
(ds) = (dq1 dq2 dq3 ) G dq2 ,
2
(5.21)
dq3

where
g11 g12 g13
G = g21 g22 g23 , (5.22)
g31 g32 g33
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 56

q 3 line

h 3 dq 3e3

h 1 dq 1e1

h 2 dq 2e2
P

q 1 line q 2 line

Figure 5.9: Diagram of line elements.

is a symmetric matrix whose elements gij from a metric tensor.


If we have a space curve C dened by the parametric representation

qi = qi (t), (i = 1, 2, 3) (5.23)

where t is a real parameter in the range t1 t t2 , then ds can be used to give the dierential arc length for the
curve C. The total arc length of C is
1/2
t2 3 3
dq i dq j
s= gij dt. (5.24)
t1 i=1 j=1
dt dt

If we move along the qi coordinate line through P by making a small coordinate change from qi to qi + dqi
(keeping the other curvilinear coordinates constant) then to rst order the change in the position vector

r (q1 , q2 , q3 )
is from Eq. (5.18):
d
ri = hi dqi

ei , (5.25)
as one can see from Fig. 5.9. The length of this dierential line element is

dsi = |d

ri | = hi dqi . (5.26)

It is clear from this equation why hi is called a scale factor.


For the case of orthogonal coordinates we have


ei

ei = ij , (i, j = 1, 2, 3) (5.27)

where
ij = 1, f or i = j
(5.28)
ij = 0 f or i = j
is the Kronecker delta symbol. The application of (5.27) to the formulae (5.19) and (5.20) gives the simplied result

2

3
2

3
2

3
2
(ds) = gij (dqi ) = h2i (dqi ) = (dsi ) . (5.29)
i=1 i=1 i=1
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 57

since the matrix (5.22) has the diagonal form:



h21 0 0
G= 0 h22 0 (5.30)
0 0 h23

It is readily seen that (5.29) can also be derived geometrically by applying the Pythagoras theorem to the rectangular
box surrounding the vector d
r (see Fig. 5.9).

5.7 Volume element


As we know from the 1st semester, curvilinear coordinates could also be useful in performing volume integration,
and we derived there the conversion formula for the volume element dV = dxdydz between the Cartesian and any
other coordinates q1 , q2 , q3 (we did not call them curvilinear though at the time):

dV = |J| dq1 dq2 dq3

where J = (x, y, z) / (q1 , q2 , q3 ) is the Jacobian of the transformation. Exactly the same result can be derived
again using our new mathematical tools developed above.
Indeed, consider Fig. 5.9 again. The volume of the gure there is based on three vectors = h
x1

1 e1 dq1 ,





x2 = h2 e2 dq2 and x3 = h3 e3 dq3 , which form its sides as indicated on the Figure. These vectors will be orthogonal
to each other only for orthogonal curvilinear coordinates, in which case, obviously,

dV = x1 x2 x3 = h1 h2 h3 dq1 dq2 dq3

i.e. the Jacobian in this case J = h1 h2 h3 is simply given by the product of the all three scale factors. In a general
case, however, the sides of the gure in Fig. 5.9 are not orthogonal, and its volume is given by the mixed product
of all three vectors of the gure sides (see the note at the end of this Section):

dV = |(
[
x1

x2
])| = |h h h (
x3

1 2 3 e1 [ e2 e3 ])| dq1 dq2 dq3

This formula can already be used in practical calculations since it gives a general result for the Jacobian as J =
h1 h2 h3 (

e1 [

e2

e3 ]). However, it is instructive to demonstrate that this is actually the same result as the one
derived previously via derivatives. To this end, recall the actual expressions for the unit base vectors, Eq. (5.14):
it is seen that hi

ei = qri and, hence, our previous result can be written as
( [ ])


dV = |h1 h2 h3 (

e1 [

e2
r r r
e3 ])| dq1 dq2 dq3 = dq1 dq2 dq3 (5.31)
q1 q2 q3

It is not dicult to see now that the mixed product of derivatives above is exactly the Jacobian J = (x, y, z) / (q1 , q2 , q3 ).


Indeed, the mixed product of three vectors
a , b and
c can be written as a determinant (see the note at the end
of this Section):
( [ ]) ax ay az


a b c = bx by bz (5.32)
cx cy cz

so that the Jacobian in Eq. (5.31) can be written as



( [ ]) x/q1

y/q1 z/q1
r r r (x, y, z)
J= = x/q2 y/q2 z/q2 (q1 , q2 , q3 )
q1 q2 q3 x/q3
y/q3 z/q3

as required.

The Note: Here we shall show that the volume of a gure constructed using three (generally) non-orthogonal


vectors
a , b and
c is given by their mixed product. This gure is shown in Fig. 5.10(b) as ABCDEFGH. Its
volume is equal to the product of the area of its base ABCD to the height h as shown. Consider rst the drawing
(a) in Fig. 5.10 and let us calculate the area Sbase of this parallelogram. Obviously, Sbase = ad, where d makes
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 58

F
G
a b
B E
C
H
c
d
b h B
C
b

A a D A a D
(a) (b)


Figure 5.10: To the calculation of the volume of the gure ABCDEFGH based on three vectors

a , b and

c : (a)
the base ABCD of the gure; (b) the gure itself.



an angle =
2 with the vector b and is given by d = b cos = b sin , so that Sbase = ab sin . It follows


then that the area of the base is given by the length (the module)
of
the vector product of

a and b , as these two

vectors make exactly the angle with each other: Sbase = a b .
Next, moving to the calculation of the volume of the gure ABCDEFGH, we see that the volume V = Sbase h,
where h = c sin is the height of the gure, and is the angle the vector
c makes with the base plane. At the


same time, = 2 , where is the angle between the vector c and the vector product vector

a b , as shown.
Therefore, one can also write:
( )

V = Sbase h = Sbase c sin = cSbase cos = c a b cos
2



which is exactly the dot product of the[ vectors

] a b and c . In other words, the volume
( [ is given
])

by the mixed




product of the three vectors, V = c a b , more conventionally denoted as V = c a b .

Finally, let us show that the mixed product can be written via the determinant as in Eq. (5.32). This is actually
straightforward. Indeed,




i j k
a b = ax ay az ,
b by bz
x



so that the dot product of this vector with the vector c would be given by replacing i , j and k with the


components c , c and c of the vector c , as it follows from the denition of the dot product.
x y z

5.8 Mechanics of a Point in General Curvilinear Coordinates


Imagine, you want to solve Newtons equations for an electron in a eld which is conveniently given in some


curvilinear coordinates, e.g. the force is F i (q1 , q2 , q3 ), i = 1, 2, 3. The equations of motion

d2

r

m =F (5.33)
dt2
should be then transformed into the curvilinear coordinates in order to be solved. Of course, everything can be
formulated in the conventional Cartesian system, but that might be extremely (and unnecessarily!) complicated.
What we need to do is to calculate d2
r /dt2 in the curvilinear coordinates. First of all, if we start from point

P (q1 , q2 , q3 ) and move to P (q1 + dq1 , q2 + dq2 , q3 + dq3 ) then the displacement vector connecting the two points
will be given by Eq. (5.18):
d
r = hi dqi

ei (5.34)
i
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 59

and the velocity is thus



d
dqi



hi qi

r
v = = hi ei = ei (5.35)
dt i
dt i

where the dot above qi means its time derivative. The acceleration entering the Newtons equations (5.33) will be

d
[ d ( )]

a =
v
= (hi qi )

ei + hi qi
d
ei (5.36)
dt i
dt dt

The derivative of

ei is calculated via the known relationship between
ei and

r in Cartesian coordinates and taking
into account that i, j, k do not change:
( )

1 r 1 x 1 y 1 z



ei = = i + j + k = Mi1 i + Mi2 j + Mi3 k , (5.37)
hi qi h1 qi h2 qi h3 qi

where M = (Mij ) is the transformation matrix. Its elements generally depend only on q1 , q2 , q3 , not on x, y, z!
Therefore, the derivative
d
ei dMi1
dMi2 dMi3
= i + j + k (5.38)
dt dt dt dt


is easily calculated for each particular case of the curvilinear coordinates. Expressing back i , j , k via

e1 ,

e2 ,

e3
using M1 matrix, we arrive at:

d
ei dMi1 (( 1 ) ( ) ( ) ) dMi2 (( 1 ) ( ) ( ) )
= M 11
e1 + M1 12 e2 + M1 13 e3 + M 21
e1 + M1 22 e2 + M1 23 e3
dt dt dt
dMi3 (( 1 ) ( ) ( ) )
+ M 31
e1 + M1 32 e2 + M1 33 e3 (5.39)
dt
Substituting equation (5.39) into (5.36), we calculate the acceleration
a in the given curvilinear system.





Finally, if the force F (x, y, z) = Fx i + Fy j + Fz k is given in Cartesian coordinates, it can be transformed


into the q1 , q2 , q3 system using the transformation relations x = x (q1 , q2 , q3 ), etc. and expressions of i , j , k via

e1 ,

e2 ,

e3 .

Appendix. Cylindrical Coordinate System


r0
x = r cos , y = r sin , z = z, [0, 2]
< z <

1. Inverse Relation: y y
x2 + y 2 = r, r = x2 + y 2 , tan = , = arctan
x x
2. Singular point: is not uniquely dened for (x = y = 0) for any z.
3. a) Coordinate surfaces: see Fig. 5.11; coordinate lines: see Fig. 5.12.
4. a) Dierentiation
x y z
= cos , = sin , =0
r r r
x y z
= r sin , = r cos , =0

x y z
= 0, = 0, =1
z z t
b) Scale factors
( )2 ( )2 ( )2

r x y z
hr = =
+ + = cos2 + sin2 + 02 = 1
r r r r
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 60

z z z
y
=tan =constant
x

y y
r y

x x
x =constant z=constant
r=constant

Figure 5.11: Coordinate surfaces in cylindrical coordinate system.

z z z
rsurface
surface z surface rplane

surface
line
z

zplane
y y y
r
zline
x x
x rline: 1 , z are fixed r, z are fixed r, are fixed

Figure 5.12: Coordinate lines in cylindrical coordinate system.


h = r2 sin2 + r2 cos2 + 02 = r

hz = 02 + 02 + 12 = 1
c) Unit base vectors, see Fig. 5.13:
( )

1 r


er = = cos i + sin j
hr r


1(
)



e = r sin i + r cos j = sin i + cos j
r



e = k z

5. It is orthogonal:


er

er =

e

e =

ez

ez = 1 by construction


e

e = cos sin + sin cos = 0,

e

e =
e
e =0
r r z z

6. Transformation Matrix:

cos sin 0 cos sin 0
M = sin cos 0 , MT = sin cos 0 M1
0 0 1 0 0 1

since M MT = 1 by inspection.
7. Thus,




i = cos

er sin

e , j = sin

er + cos

e , k =
ez
CHAPTER 5. VECTOR CALCULUS: CURVILINEAR COORDINATES (PART I) 61

ez
P(r, ,z)
e
er
z

r
y

x

Figure 5.13: Unit base vectors of point P in cylindrical coordinates.

8. Metric tensor ( ) ( )

hi hj (

ei

r r
g = (gij ) , gij = ej ) = h2i ij
qi qj

1 0 0
g = 0 r2 0
0 0 1
is diagonal as it should for the orthogonal coordinate system.
9. Derivative wrt (with respect to) t:
d

er d (
)

(
)

= i cos + j sin = sin i + cos j = sin i + cos j
dt dt
= [ sin (cos
er sin
e ) + cos (sin
er + cos

e )]

( )
= er ( sin cos + cos sin ) +
e sin2 + cos2 =
e
| {z } | {z }
=0 =1
i.e. we obtain:
d

=

er
e (5.40)
dt
Similarly,
d
e

(
)

= cos i sin j = cos i + sin j
dt
= [cos (cos

er sin e ) + sin (sin er + cos e )]
[ ( 2 ) ]
= er cos + sin2 + e ( cos sin + sin cos ) = er
d

=
e
er (5.41)
dt
and, nally,
d

ez
=0
dt
10. Velocity:


v = hi qi

ei r

er + r

e + z

ez =

v
i
Acceleration: ( )


a =
v = r

er + r

er + r + r
e + r

e + z

ez

Using the expressions for



er ,

e , obtained above, we obtain:
( ) ( )

a = r r2
er + 2r + r
e + z

ez .
Chapter 6

Vector calculus: Grad, Div and Curl in


Orthogonal Curvilinear Coordinates
(Part II)

6.1 Gradient of a Scalar Field


Consider a scalar eld (P ) dened at each point P in a region R. Let be the change in (P ) when we move a
distance s from P in the direction of a unit vector

s , see Fig. 6.1. The directional derivative of (P ) is

d (s + s) (s)
= lim = lim , (6.1)
ds s0 s s0 s

The gradient of (P ) is a vector (denoted by grad ) which is dened by the relation

(grad)
d
s , (6.2)
ds
We shall see directly that it does not depend on the direction

s.
We see that the component of grad in the direction of the unit vector

s gives d
ds in that direction.
In order to determine grad we introduce an orthogonal curvilinear coordinate system (q1 , q2 , q3 ). We can now
write
3
grad = (grad)i
ei , (6.3)
i=1
where
(grad)i = (grad)
ei , (6.4)


is the component of grad in the direction of ei . The application of (6.2) to (6.4) gives
d
(grad)i = . (6.5)
dsi

grad (P )
s

s
(P )
Figure 6.1: The directional derivative of (P ) along the direction

s.

62
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I

q 3line e3
grad

P
e1 e2

q 1 line q line
2

Figure 6.2: grad(P ) in general curvilinear coordinates.

Next we consider a small change of from qi to qi + dqi along the qi -coordinate line through P . For this change
we have the dierential of ( )

d = dqi , (6.6)
qi
and the length (see Eq. (5.26) in Chapter 5):
dsi = hi dqi (6.7)
In the limit dqi 0 these equations yield
( )
d 1
= , (i = 1, 2, 3). (6.8)
dsi hi qi

From this result and Eq. (6.5) we nally obtain (see alos Fig. 6.2):
( )
1
(grad)i = (6.9)
hi qi

and thus from Eq. (6.3)


3 ( )
1
grad = ei . (6.10)
i=1
h i q i

We can write (6.10) in the alternative form


grad = , (6.11)
where ( )

3


ei
1
. (6.12)
i=1
h i qi

(
)
Example 1: For Cartesian coordinates we have (q1 , q2 , q3 ) (x, y, z), (

e1 ,

e2 ,

e3 ) i , j , k and h1 = h2 =
h3 = 1. Hence, we obtain from (6.10) the well-known result:
( ) ( ) ( )

grad = i + j + k.
x y z

For cylindrical coordinates we nd that (q1 , q2 , q3 ) (, , z), (



e1 ,

e2 ,

e3 ) (

e ,

e ,

ez ) and (h1 , h2 , h3 ) =
(1, , 1). It follows, therefore, that
( ) ( ) ( )
1
grad = e + e + ez .
z
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I

n
F

dS

P
S

Figure 6.3: The point P is surrounded by surface S of volume V . Here, S represents an element of S with the
outward normal
n.
q 3 line

K h2 q 2 coordinate surface
C q 12 q
1 1
G
B J
e3

P
e2 F
e1
h3 q 3
L
D

q 2 line
H
A I
q 1 line S coordinate surface
h1 q 1
q +12 q
E 1 1

Figure 6.4: To the calculation of the ux through the faces EFGH and IJKL of the closed surface S.

6.2 Divergence of a Vector Field




Consider a vector eld F (P ) dened in a region R. Surround the point P by a surface S which encloses a small


volume V , see Fig. 6.3. We dene the divergence of F (P ) at P to be

Flux(S, P )
div F lim , (6.13)
V 0 V
where I I



Flux(S, P ) = F dS = F
n dS, (6.14)
S S


is the ux of F across the surface S. Note that dS is an area element on S, and
n is a unit vector normal to dS
pointing in an outward direction as shown in Fig. 6.3.
An important feature of (6.13) is that it provides one with an intrinsic denition which makes no reference to
any particular coordinate system. It can also be shown that the limit (6.13) does not depend on the shape of the


surface S. We see, therefore, that div F (P ) is a well-dened scalar eld which only depends on the vector eld

F (P ).


To determine div F (P ) we introduce an orthogonal curvilinear coordinate system (q1 , q2 , q3 ) and surround the
point P (q1 , q2 , q3 ) with a small curvilinear box S formed by the six coordinate surfaces which have qi coordinate


equal to qi 21 qi (i = 1, 2, 3) as in Fig. 6.4. We also write the vector eld F in the component form


3
F (q1 , q2 , q3 ) = Fi (q1 , q2 , q3 )

ei . (6.15)
i=1
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I



To leading order, the ux of F across the coordinate surface ABCD (with q1 coordinate equal to q1 ) is

(F1 h2 h3 )P q2 q3 since

n

e1 , (6.16)

where
(F1 h2 h3 )P = F1 (q1 , q2 , q3 ) h2 (q1 , q2 , q3 ) h3 (q1 , q2 , q3 ) . (6.17)


Here, the unit normal vector
n is
e1 , so that F n = F1 . If we make the substitution q1 q1 + 21 q1 in (6.16) and
apply the Taylor theorem it is found that the outward ux across the coordinate surface EF GH (with q1 coordinate
equal to q1 + 12 q1 ) is [ ]
q1
(F1 h2 h3 )P q2 q3 + (F1 h2 h3 ) q2 q3 + ... (6.18)
q1 P 2
In a simmilar manner, we nd that the outward ux across the coordinate surface IJKL (with q1 coordinate equal
to q1 21 q1 ) is [ ]
q1
(F1 h2 h3 )P q2 q3 + (F1 h2 h3 ) q2 q3 + ... (6.19)
q1 P 2


because for this surface the outward normal n =
e1 and thus F n = F1 .
Hence the total outward ux across the opposite pair of surfaces EF GH and IJKL is
[ ]

(F1 h2 h3 ) q1 q2 q3 + ... (6.20)
q1 P

If the same analysis is repeated for the other two pairs of opposite faces we obtain
[ ]

Flux(S, P ) = (F1 h2 h3 ) + (F2 h3 h1 ) + (F3 h1 h2 ) q1 q2 q3 + ..., (6.21)
q1 q2 q3
where the partial derivatives are evaluated at the point P .
The volume V enclosed by S is given by (the system is orthogonal):

V = h1 h2 h3 q1 q2 q3 + ... (6.22)

Finally, we substitute (6.21) and (6.22) in the formula (6.13) and take the limit qi 0, (i = 1, 2, 3). This procedure
yields [ ]

1
div F = (F1 h2 h3 ) + (h1 F2 h3 ) + (h1 h2 F3 ) (6.23)
h1 h2 h3 q1 q2 q3

Example 2: For Cartesian coordinates (6.23) reduces to the expected result


Fx
Fy Fz
div F = + + .
x y z
For cylindrical coordinates (, , z) we nd that

1 1 F Fz
div F = (F ) + + .
z

It can be shown that (6.23) can be expressed in the form





div F = F , (6.24)


where is dened in (6.12). The evaluation of F for curvilinear coordinates requires great care because the




unit base vector ( e1 , e2 , e3 ) will depend on the position P (q1 , q2 , q3 ). In particular,
(

)


(F1

e1 + F2

e2 + F3

e1 e2 e3
F = + + e3 )
h1 q1 h2 q2 h3 q3
1 F1 1 F2 1 F3
= + + .
h1 q1 h2 q2 h3 q3
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I

curl F

P
C



Figure 6.5: To the denition of curl of vector eld F .

6.3 Laplacian in Orthogonal Curvilinear Coordinates


If we have a scalar eld (P ) we can generate a vector eld


F = grad = . (6.25)


We can now associate a further scalar eld with F by considering its divergence, i.e. the Laplacian of :


div F = div(grad) = = 2 . (6.26)

To evaluate 2 we rst introduce orthogonal curvilinear coordinates (q1 , q2 , q3 ) and write




F = F1

e1 + F2

e2 + F3

e3 , (6.27)

where ( )
1
Fi = , (i = 1, 2, 3). (6.28)
hi qi
Next we substitute (6.28) in the formula (6.23). This prodecure gives the important result:
[ ( ) ( ) ( )]
1 h2 h3 h3 h1 h1 h2
2 = + + . (6.29)
h1 h2 h3 q1 h1 q1 q2 h2 q2 q3 h3 q3

Example 3: For Cartesian coordinates (6.29) simplies to a well-known result

2 2 2
2 = + + .
q12 q22 q32
For culindrical coordinates (, , z) we nd that
( )
1 1 2 2
2 = + 2 + .
2 z 2

6.4 Curl of a Vector Field




Consider a vector eld F (P ) dened in a region R. Draw a unit vector n at the point P , and suppose that S is


any smooth surface through P which has a normal vector at P equal to n . Next construct a small closed curve C
on this surface which encloses P and an area S, see Fig. 6.4, and evaluate the line integral
I


Line Int(S, C, P ) = F d
r, (6.30)
C
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I

q 3line

C B

B coordinate surface q 1
e3

C
P
e2 h 3 dq 3
e1

A
D
h 2 dq 2
D
q 2 line

q 1 line
A

Figure 6.6: To the calculation of the line integral along ABCD.

where the direction of integration around C is related to



n by the right-hand screw rule. We can now dene the



component of curl of F at P in the direction n as the limit:
( )

curl F
Line Int(S, C, P )
n = lim . (6.31)
S0 S
An important feature of (6.31) is that it provides one with an intrinsic denition which makes no reference to any
particular coordinate system. It can also be shown that the limit (6.31) is independent of the particular choice of


the surface S and the shape of the embedded contour C. We see, therefore, that curl F (P ) is a well-dened vector


eld which only depends on F (P ).


To determine curl F (P ) we introduce an orthogonal curvilinear coordinate system (q1 , q2 , q3 ), and write


3
F (q1 , q2 , q3 ) = Fi (q1 , q2 , q3 )

ei . (6.32)
i=1
( )

We can evaluate curl F e1 at P (

q1 ,

q2 ,

q3 ) by taking S to be the coordinate surface q1 , and C to be the closed
path ABCD as shown in Fig. 6.6, where:
( )
1
AB is along the q3 -coordinate line through A point q1 , q2 + q2 , q3 ,
2
( )
1
BC is along the q2 -coordinate line through B at q1 , q2 , q3 + q3 ,
2
( )
1
CD is along the q3 -coordinate line through C at q1 , q2 q2 , q3 ,
2
( )
1
DA is along the q2 -coordinate line through D at q1 , q2 , q3 q3 .
2


The line integral of F along the q3 -coordinate line D B is:
B


F dr = (F3 h3 )P q3 + ..., (6.33)
D

where
(F3 h3 )P = F3 (q1 , q2 , q3 ) h3 (q1 , q2 , q3 ) . (6.34)


We can now calculate the line integrals of F along AB and CD by making the substitutions q2 q2 21 q2
respectively in (6.33), and applying the Taylor theorem. This procedure gives
B [ ]


F d
q2
r = (F3 h3 )P q3 + (F3 h3 ) q3 + ... along AB, (6.35)
A q2 P 2
CHAPTER 6. VECTOR CALCULUS: GRAD, DIV AND CURL IN ORTHOGONAL CURVILINEAR COORDINATES(PART I

[ ]
D


F d
q2
r = (F3 h3 )P q3 + (F3 h3 )q3 + ... along CD. (6.36)
C q2 P 2

Note that in the latter case we move in the direction opposite to the vector
e3 .
In a similar manner, we nd by considering the line integral along the q2 -coordinate line C A , that the line
integrals along DA and BC are, respectively:
A [ ]


F d
q3
r = (F2 h2 )P q2 (F2 h2 ) q2 + ..., (6.37)
D q3 P 2
[ ]
C


F d
q3
r = (F2 h2 )P q2 (F2 h2 ) q2 + ... (6.38)
B q3 P 2
Hence, we obtain for the whole closed-path line integral:
I [ ]


F dr = (F3 h3 ) (F2 h2 ) q2 q3 + ... (6.39)
ABCD q2 q3 P

Next, the area enclosed by ABCD is


S = h2 h3 q2 q3 + ... (6.40)


It follows from equations (6.31), (6.39) and (6.40) that (recall that n = e1 here):

( [ ]
)
( )

curl F
1
curl F e1 = (h3 F3 ) (h2 F2 ) . (6.41)
1 h2 h3 q2 q3
( )
( )

The other two components curl F e2 and curl F e3 can be readily obtained by applying a symmetry
argument to (6.41). This procedure yields the required result:

[ ]
[ ]

e1 e2
curl F = (h3 F3 ) (h2 F2 ) + (h1 F1 ) (h3 F3 )
h2 h3 q2 q3 h3 h1 q3 q1

[ ]
e3
+ (h2 F2 ) (h1 F1 ) . (6.42)
h1 h2 q1 q2
This formula can be expressed more compactly as a determinant:

h1

h3

e3

1 e1 h2e2
curl F =
q3 . (6.43)
h1 h2 h3 q1 q2
h1 F 1 h2 F 2 h3 F3
Chapter 7

Series Solutions of Ordinary Dierential


Equations

7.1 Introduction
When the partial dierential equations of mathematical physics, such as

2 = 0, (Laplace equation)

1 2
2 = , (wave equation)
c2 t2
1
2 = , (diusion equation)
D t
2m
2 + 2 (E V ) = 0, (Schrdinger equation) ,
~
are solved using the method of separation of variables, one obtains ordinary second-order dierential equations of
the type
y (x) + p(x)y (x) + q(x)y(x) = 0, (7.1)
where p(x) and q(x) are known functions of the independent variable x.
When the coecients p(x) and q(x) are both constants it is possible to solve (7.1) in terms of elementary
functions. However, if the coecients p(x) and q(x) are variable functions then the solutions of (7.1) usually dene
new transcendental functions which cannot be expressed in terms of a nite number of elementary functions. For
this general case we can use the series method to derive solutions of the Eq. (7.1).

7.2 Ordinary and Singular Points


Denitions:
The point x = a is said to be an ordinary point of the dierential equation (7.1) if both p(x) and q(x) are well-
behaved functions1 at x = a. If the point x = a is not an ordinary point then we shall call it a singular point.

Example 1: For the dierential equation

x2 (x + 3)y (x) (x + 3)y (x) + 3xy(x) = 0, (7.2)

we see that
1 3
p(x) = 2
, q(x) = .
x x(x + 3)
1 A functionf (x) which is well-behaved at x = a is single-valued, nite and continuous at x = a, and possesses derivatives f (k) (a)

of all orders k = 1, 2, 3, ....

69
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 70

It follows that the dierential equation (7.2) has singular points at x = 0 and 3. All other points in the interval
< x < are ordinary points of (7.2).

7.3 Series Solutions about an Ordinary Point


If x = a is an ordinary point of the Eq. (7.2) then we can, at least in principle, calculate all the derivatives y (k) (a),
(k = 1, 2, 3...) in terms of y(a) and y (a), by repeated dierentiation of (7.1) with respect to x. Hence, we can
formally construct a Taylor series solution about x = a:


y(x) = cr (x a)r , (7.3)
r=0

where
y (r) (a)
cr = . (7.4)
r!
This series contains two arbitrary constants y(a) and y (a) and gives us the general soluton of the Eq. (7.1) in
the neighbourhood of x = a, provided that 0 |x a| < R, where R is the radius of convergence of the power
series (7.3). Usually, the radius of convergence R is equal to the distance Ja from the point x = a to the nearest
singular point of the Eq. (7.1). However, in exceptional cases it is possible to also have R > Ja . It should be noted
that, in the neighbourhood of an ordinary point, the general solution of the Ed. (7.1) is always well-behaved for
0 |x a| < R.
In practice the coecients cr are most easily found by substituting the series (7.3) in the dierential equation
(7.1). After some rearrangement of the terms this procedure leads to a recurrence relation which can be used to
generate the coecients cr , (r = 2, 3, ...) given the values of the initial coecients c0 and c1 .

Example 2: Determine the series solution of the dierential equation


( 2 )
d
T [y(x)] x y(x) = 0, (7.5)
dx2

which satises the boundary conditions y(0) = 1 and y (0) = 0.


It is readily seen that all the points of the dierential equation (7.5) are ordinary points. Hence, we can write


y(x) = cr xr , |x| < . (7.6)
r=0

From (7.6) we nd that:




y (x) = rcr xr1 , (7.7)
r=1


y (x) = r(r 1)cr xr2 . (7.8)
r=2

Note that in y (x) the rst term (r = 0) in the sum disappears; the rst two terms (r = 0, 1) disappear in the sum
in the experession for y (x). The substitution of the series (7.6) and (7.8) in dierential eqution (7.5) gives



T [y(x)] = r(r 1)cr xr2 cr xr+1 ,
r=2 r=0




= 2c2 + r(r 1)cr xr2 cr xr+1 . (7.9)
r=3 r=0

We have specically separated out the rst term (r = 2) in the rst sum (the rst line), so that the sum starts
from r = 3 now, to make the two sums to have the same powers of x. Indeed, in the second line, the rst sum
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 71

contains terms x1 ,x2 , etc. and one can easily see that the same powers of x appear in the second sum. Once we
now have both sums starting from the same powers of x, we can make them look similar. To this end, we make the
replacement m = r 2 for the summation index in the rst summation of Eq. (7.9), while in the second summation
we make the replacement m = r + 1. Hence, we obtain


T [y(x)] = 2c2 + [(m + 2) (m + 1) cm+2 cm1 ] xm . (7.10)
m=1

The diernetial eqution T [y(x)] = 0 will be satised for all values of x, provided that

c2 = 0, (7.11)

(m + 1) (m + 2) cm+2 = cm1 , (m 1). (7.12)


From the two-term recurrence relation (7.12) with m = 1, 2, 3, ... we can generate the following cm coecients:

m=1 m=2 m=3


c0 c1 c2
c3 = 23 c4 = 34 c5 = 45 =0
m=4 m=5 m=6
c3 c0 c4 c1 c5
c6 = 56 = (25)(36) c7 = 67 = (36)(47) c8 = 78 =0

One can see that if we start from c0 , we generate the coecients c3 , c6 , etc. If we start from c1 , we generate
coecients c4 , c7 , etc. Finally, starting from c2 (which is zero), we generate all other coecients c5 , c8 , etc. which
are also all equal to zero. Since nothing can be said about the coecients c0 and c1 , we have to accept that these
can be arbitrary. Therefore, if we recall the general form of the solution,

y(x) = c0 x0 + c1 x1 + c2 x2 + . . .

then two solutions are obtained, one starting from c0 , and another - from c1 :
[ ]
3 6 x3 x6 x9
y1 (x) = c0 + c3 x + c6 x + . . . = c0 1 + + + + ...
2 3 (2 5)(3 6) (2 5 8)(3 6 9)
[ ]
4 7 x4 x7 x10
y2 (x) = c1 x + c4 x + c7 x + . . . = c1 x + + + + ...
3 4 (3 6)(4 7) (3 6 9)(4 7 10)
where we have used the explicit relationship between the coecients derived above from the recurrence relations
(see the Table).
One can see that the solutions are dened up to arbitary constants c0 and c1 . However, this is not a problem
as we have to combine the two solutions in a linear combination with arbitarry coecients anyway, so that c0 and
c1 will be absorbed by these constants (see below). Hence, we obtain the desired general series solution:

y(x) = Ay1 (x) + By2 (x)


[ ]
x3 x6 x9
= C1 1 + + + + ...
2 3 (2 5)(3 6) (2 5 8)(3 6 9)
[ ]
x4 x7 x10
+C2 x + + + + ... (7.13)
3 4 (3 6)(4 7) (3 6 9)(4 7 10)
with two new arbitrary constants C1 = Ac0 and C2 = Bc1 .
The application of the boundary conditions y(0) = C1 1, and y (0) = C2 0 gives the required solution:

x3 x6 x9
y(x) = 1 + + 2 + 3 + ... (7.14)
3 1!(2) 3 2!(2 5) 3 3!(2 5 8)

It is possible to write the series (7.14) in the general form:




x3k
y(x) = 1 + k
= 1 + ck x3k , |x| < (7.15)
k=1 3k k!
j=1 (3j 1) k=1
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 72

where 1

k
ck = 3k k! (3j 1) , k = 1, 2, 3, . . .
j=1

and the product sign means this:



k
f (j) = f (1) f (2) ... f (k). (7.16)
j=1

The convergence of
the series solution (7.14) can be investigated directly using the ratio test:

An innite series k=0 ak is convergent if

ak+1
lim < 1, (7.17)
k ak

and divergent if
ak+1
lim > 1. (7.18)
k ak

(The ratio test is inconclusive when limk |ak+1 /ak | = 1.)


The application of the ratio test to the series (7.14) with ak ck x3k gives

ak+1 ck+1 x3(k+1) ck+1 3 |x|
3
= = |x| = . (7.19)
ak ck x3k ck 3(k + 1)(3k + 2)
We see from this result that
ak+1
lim = 0, for all x. (7.20)
k ak
It follows, therefore, that the series solution (7.14) is convergent for all nite values of x.

7.4 Classication of Singular Points


Let us now suppose that the dierential equation (7.1) has a singular point at x = a. If the functions (x a)p(x)
and (x a)2 q(x) are both well-behaved at x = a, then we say that x = a is a regular singular point (RSP) of the
Eq. (7.1).
A singular point which is not a regular singular point is called an irregular singular point of the Eq. (7.1).

Example 3: We have seen (Example 1) that the dierential wquation

x2 (x + 3)y (x) (x + 3)y (x) + 3xy(x) = 0,

has singular points at x = 0 and x = 3. For the singular point x = 3 we have


(x + 3) (x + 3)
(x a)p(x) = , (x a)2 q(x) = 3 .
x2 x
Both these functions are well-behaved at x = 3. It follows, therefore, that x = 3 is a regular singular point.
For the singular point x = 0 we nd that
1 3x
(x a)p(x) = , (x a)2 q(x) = .
x (x + 3)
In this case it is clear that the function (x a)p(x) is not well-behaved at x = 0. It follows, therefore, that x = 0
is an irregular singular point.
This classication scheme for singular points is important because it is always possible to obtain the solution of
the Eq. (7.1) in the neighbourhood of a regular singular point x = a, by using a simple generalization of the series
method described above. The application of series methods to irregular singular points gives rise to major problems
such as divergent asymptotic series, which cannot be discussed in this introductory course.
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 73

7.5 Series Solutions About a Regular Singular Point


If x = a is a RSP of the Eq. (7.1) then the Taylor series solution (7.3) will break down because the derivatives
y (r) (a) (r = 2, 3, ...) will not exist. Under these circumstances, it can be shown that the Eq.(7.1) always has at least
one solution of the Frobenius type:


y(s, x) = (x a)s cr (x a)r (7.21)
r=0


= cr (x a)r+s , (7.22)
r=0

where the exponent s can have negative and non-integral values. The radius of convergence R of the series (7.21)
is at least as large as the distance to the nearest singular point of the dierential equation (7.1).
In practice the exponent s and the coecients cr are found by substituting the series (7.22) in the dierential
equation. After some rearrangement of the terms this procedure leads to an indicial equation for s, and a
recurrence relation for the coecients cr .

Example 4: Determine the general series solution of the dierential equation


( )
d2 d
T [y(x)] 2x 2 + + 1 y(x) = 0, (7.23)
dx dx

about x = 0.

1 1
Solution In this dierential equation p(x) = 2x and q(x) = 2x . We should rst check what type is the point
1 2 1
x = 0. Since the functions xp(x) = 2 and x q(x) = 2 x are both well-behaved at x = 0, it follows, therefore, that
x = 0 is a RSP. We can now assume a series solution of the Frobenius type:


y(x) = cr xr+s . (7.24)
r=0

From (7.24) we nd that




y (x) = cr (r + s)xr+s1 , (7.25)
r=0


y (x) = cr (r + s)(r + s 1)xr+s2 . (7.26)
r=0

Note that, opposite to Example 2 considered above, in this case the rst terms in the sums do not disappear after
dierentiation as s may not be integer. Thus, we have to keep all terms under the sums, so that both sums still
start from r = 0. The substitution of these results in Eq.(7.23) gives



T [y(x)] = cr (r + s)(2r + 2s 1)xr+s1 + cr xr+s (7.27)
r=0 r=0




= c0 s(2s 1)xs1 + cr (r + s)(2r + 2s 1)xr+s1 + cr xr+s . (7.28)
r=1 r=0

Next, we make the index shift r 7 r + 1 in the rst summation in Eq. (7.28), i.e. we introduce a new index
r = r 1 in the sum and then write r instead of r for convenience:



cr (r + s)(2r + 2s 1)xr+s1 = cr +1 (r + s + 1)(2r + 2s + 1)xr +s
r=1 r =0



= cr+1 (r + s + 1)(2r + 2s + 1)xr+s
r=0
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 74

The two sums in (7.28) can now be combined into one. Hence we obtain for the LHS of the original equation (7.23):


T [y(x)] = c0 Q(s)xs1 + [(r + s + 1) (2r + 2s + 1) cr+1 + cr ] xr+s , (7.29)
r=0

where
Q(s) s(2s 1). (7.30)
The function T [y(x)] is to be equal to zero for all values of x. This means that each and every term in the expansion
of T [y(x)] should have zero coecient to the power of x, i.e. we should have c0 Q(s) = 0 and, at the same time,

(r + s + 1)(2r + 2s + 1)cr+1 + cr = 0 (7.31)

for every r = 0, 1, 2, . . ..
Concider rst the recurrence relation (7.31). One can see that cr+1 is proportional to cr ; more explicitly: c1 is
given by c0 ; c2 is given by c1 ; c3 by c2 , and so on. The rst equation, c0 s(2s 1) = 0, accepts as a solution also
c0 = 0. However, it is clear from what was said above that in this case all other coecients will also be equal to
zero: c1 = 0 due to c0 = 0; then, c2 = 0 due to c1 = 0, etc. Therefore, if we choose c0 = 0, then we shall obtain
a trivial solution y(x) = 0 of the dierential equaiton, quite not the one we wish. Hence, we have to assume that
c0 = 0.
Then, the quadratic equation Q(s) = s(2s 1) = 0, which is called the indicial equation, has to be considered
for the values of s and its roots will be denoted by s1 and s2 . We see that two values of s are possible:
1
s1 = , s2 = 0. (7.32)
2
Using the particular values of s, we can now generate the values of the coecients cr usikng the recurrence relations.
Consider rst s = 21 . we now assume that the coecients cr satisfy the recurrence relation:with c0 1. It
follows from (7.31) that

r1
1
cr = (1)r , (r 1). (7.33)
j=0
(j + s + 1)(2j + 2s + 1)

Hence the series




r1
1
y(s, x) = xs 1 + (x)r (7.34)
r=1 j=0
(j + s + 1)(2j + 2s + 1)

is a solution of the dierential equation


T [y(s, x)] = Q(s)xs1 . (7.35)
It is clear from Eqs.(7.34) and (7.35) that the series y(s1 , x) and y(s2 , x) provide us with two independent
solutions of the dierential equation (7.23). We conclude, therefore, that the general series solution of the dierential
equation T [y(x)] = 0 is
y(x) = Ay(s1 , x) + By(s2 , x), |x| < (7.36)
where

( )
(x)r
r1
1 , s1 = 1
y(s1 , x) = x 1 +
1
2 (7.37)
r=1
r! j=0
(2j + 3) 2


(x)r
r1
1
y(s2 , x) = 1 + , (s2 = 0) (7.38)
r=1
r! j=0
(2j + 1)

and A, B are arbitrary constants.

7.6 Special cases


The procedure described in Example 4 always gives the general series solution of the dierential equation (7.1)
about a RSP, provided that the roots of the indicial equation do not dier by an integer. For the special cases
s1 s2 = 0, 1, 2, ... the technique must be modied.
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 75

Special Case: s1 = s2 In this case we only obtain one series solution! But we do recall from the general theory
of linear second order dierential equations, that if one solution is known, the other one can always be constructed.
It can be shown then that the second solution always involves a logarithmic function.

Special Cases: s1 s2 = 1, 2, 3, ... When s1 s2 = n, (n = 1, 2, 3, ...) the larger root s1 always gives a series
solution of the dierential equation. If we attempt to generate a second solution using the recurrence relation with
s = s2 then two possibilities arise:
1. The coecient cn is not dened because it satises the relation 0 cn = 0. For this case the second solution
of the dierential equation involves logarithmic terms.
2. The coecient cn has an arbitrary form because it satises the relation 0 cn = 0. In this case the second
solution does not involve logarithmic terms.
Note that instead of using s2 in these cases, it is always possible to generate the second solution from the rst
one using the general formula mentioned above which allows the second solution to be expressed via the rst one
explicitly.

7.7 More Examples


Example 5: Solve the harmonic oscillator equation

y (x) + 2 y(x) = 0
using the series expansion method.

Solution: This equation does not have any singular points. We can expand around any point then. We shall use
the Frobenius method and expand around x = 0:


y= Cr xr+s
r=0

which gives:


y = (r + s) Cr xr+s1
r=0


y = (r + s)(r + s 1)Cr xr+s2
r=0
After substituting into the DE above:



Cr (r + s)(r + s 1)xr+s2 + 2 Cr xr+s = 0
r=0 r=0

The following powers of x are contained in the rst sum: x , x , xs , xs+1 , etc., corresponding to the summation
s2 s1

index r = 0, 1, 2, 3, etc., repsectively.At the same time, the second sum contains the terms with xs , xs+1 , etc.
Anticipating combining the two sums together later on, we separate out the rst two foreign terms in the rst
sum:



C0 s(s 1)xs2 + C1 (s + 1)sxs1 + Cr (r + s)(r + s 1)xr+s2 + 2 Cr xr+s = 0
r=2 r=0
In the rst sum we shift the summation index r 2 r, so that the two sums can be combined into one:

( )
C0 s(s 1)xs2 + C1 s(s + 1)xs1 + (r + s + 2)(r + s + 1)Cr+2 + 2 Cr xr+s = 0 (7.39)
r=0

This is a sum of powers of x which is equal to zero for all values of x. This can only be true if all coecients to
every power of x are all equal to zero at the same time. It is convenient to assume that C0 = 0. Then, from the
rst temr we conclude that s(s 1) = 0 (the indicial equation), which gives us two possible values of s: s1 = 0 and
s2 = 1. Therefore, we now consider two cases:
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 76

(1) s1 = 0 From the second term in (7.39), we see that the coecient C1 s(s + 1) to xs1 is zero anyway, so that
C1 is arbitary at this stage. From the last term in (7.39) we obtain the recurrence relation

2
Cr+2 = Cr , where r = 0, 1, 2, . . .
(r + 2)(r + 1)

Staring from C0 , we can construct from the recurrence relation all even coecients:

2 2
C2 = C0 = C0
21 2!
2 4 4
C4 = C2 = + C0 = C0
43 4321 4!
2 6
C6 =
C4 = (1)3 C0
65 6!
and so on. Note that in the end all even coecients are epxressed via C0 . In fact, one can easily see the general
rule:
2n
C2n = (1)n C0 , n = 1, 2, ...
(2n)!
which can be proven e.g. by the method of mathematical induction.2
Combining all terms with even coecients (and thus even powers of x), we obtain the rst solution of the
equation:

2n 2n
y1 (x) = C0 (1)n x C0 cos x
n=1
(2n)!
One can recognise here the Taylor expansion of the cosine function with some arbitary prefactor C0 .
Next, we consider odd terms (odd powers of x). These can be generated starting from arbitrary C1 and using
the same recurrence relation as above. This procedure gives:

2 2
C3 = C1 (1)1 C1
32 3!
2 4
C5 = (1) C3 = (1)2 C1
54 5!
2 6
C7 =
C5 = (1)3 C1
76 7!
Yet again, all odd coecients are expresed via C1 , and one can also work out the general rule:

2
C2n+1 = (1)n C1 , n = 1, 2, ...
(2n + 1)!

Hence, all terms with odd powers of x combined together form the 2nd solution:


2n C1 (x)2n+1 C1
y2 (x) = C1 (1)n x2n+1 (1)n = sin x
n=1
(2n + 1)! n=1 (2n + 1)!

which is to be easily identied with the sine function.


Either of the solutions converges at all values: < x < +.
2 Indeed, this is true for n = 1. We then assume that it is also true for some value of n. Then, using the recurrence relation, we get

for the next coecient:


2 2 2n 2n+2
C2(n+1) = C2n+2 = C2n = (1)n C0 = (1)n+1 C0 ,
(2n + 2)(2n + 1) (2n + 2)(2n + 1) (2n)! (2n + 2)!
which exactly the correct expression to be expected for the C2(n+1) . Q.E.D.
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 77

(2) s2 = 1 In this case C0 is again arbitary; however, in the second term of (7.39) we have s(s + 1) which is not
equal to zero, so that one has to set C1 to zero to ensure this term is zero for all values of x. The recurrence relation
is
2
Cr+2 = Cr , r 0
(r + 3)(r + 2)
which gives (C0 is arbitary):

2 2
C2 = C0 = (1)1 C0
32 3!
2 4
C4 = C2 = (1)2 C0
54 5!
2 6
C6 = C4 = (1)3 C0
76 7!
and so on, One can see that we arrive at y2 (x) again. Since C1 = 0, and using the recurrence relation, we see that
there will be no odd terms at all: C3 = C5 = ... = 0. Thus, by using the second value of s, a new solution is not
generated.
Concluding, the nal general solution of the harmonic oscillator equation is

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

as expected, where A and B are arbitrary constants.

Example 6: Solve the Ledendre equation


( )
1 x2 y (x) 2xy (x) + l(l + 1)y(x) = 0. (7.40)

using the Frobenius method.

Solution: First of all, we write p(x) and q(x) to nd all singular points of the DE and characterise them:
2x 2x
p(x) = =
1x 2 (1 + x)(1 x)

l(l + 1)
q(x) =
(1 + x)(1 x)
Points x = 1 are regular singular points (RSPs) since:
2x
(1 + x)p(x) = is regular at x = 1
1x
l(l + 1)
(1 + x)2 q(x) = (1 + x) is also regular at x = 1
1x
so that the point x = 1 is the RSP; similarly,
2x
(1 x)p(x) = is regular at x = +1
1+x
l(l + 1)
(1 x)2 q(x) = (1 x) is regular at x = 1,
1+x
so that x = 1 is also the RSP.
Then, we can seek the series expansion around x = 0 in which case it is supposed to converge for any x within
the interval 1 < x < 1. Consider


s
y(x) = x Cr xr
r=0
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 78

which is followed by


y (x) = Cr (r + s)xr+s1
r=0


y (x) = Cr (r + s)(r + s 1)xr+s2
r=0

Substituting into the dierential equation;






( )
Cr (r + s)(r + s 1) 1 x2 xr+s2 2 Cr (r + s)xr+s + l(l + 1) Cr xr+s = 0. (7.41)
r=0 r=0 r=0

The rst term contains two sums:




1st = Cr (r + s)(r + s 1)xr+s2 = C0 s(s 1)xs2 + C1 (s + 1)sxs1 + Cr (r + s)(r + s 1)xr+s2
r=0 r=2

which, after changing the summation index r 2 r, is worked out into




1st = C0 (s 1)sxs2 + C1 (s + 1)sxs1 + Cr+2 (r + s + 2)(r + s + 1)xr+s ,
r=0

while the second sum is




2nd = Cr (r + s)(r + s 1)xr+s
r=0

We separated out the rst two terms in the rst sum above to be able to combine the two sums together into a
single sum. Moreover, all other sums in Eq. (7.41) have now the same structure and can be combined together:

C0 (s 1)sxs2 + C1 (s + 1)sxs1


+ [(r + s + 2)(r + s + 1)Cr+2 (r + s)(r + s 1)Cr 2(r + s)Cr + l(l + 1)Cr ] xr+s = 0. (7.42)
r=0

After rearranging:


C0 s(s 1)xs2 + C1 s(s + 1)xs1 + {(r + s + 2)(r + s + 1)Cr+2 [(r + s)(r + s + 1) l(l + 1)] Cr } xr+s = 0.
r=0
(7.43)
Again, the rst two terms should go, so that, assuming C0 is arbitary, we have the following indicial equation:

s(s 1) = 0 (7.44)

We must consider two cases then.

(1) s1 = 0 Becuase of the second term in (7.43), C1 must be arbitrary and the recurrence relation (from the last
term with the sum sign) reads:
r(r + 1) l(l + 1)
Cr+2 = Cr (7.45)
(r + 2)(r + 1)
We can start either from C0 or C1 . If we rst start from C0 , then we can generate all the C2n coecients with even
indices:
l(l + 1) k
C2 = C0 C0 , with k = l(l + 1)
21 2!
23k (6 k)k
C4 = C2 = C0
43 4!
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 79

54k (20 k)(6 k)k


C6 = C4 = C0
65 6!
which results in a series solution
[ ]
k 2 (6 k)k 4
y1 (x) = C0 1 x x ... (7.46)
2! 4!

containing even powers of x.


Starting from C1 , we similarly obtain:
21k 2k
C3 = C1 = C1
32 3!
43k (12 k)(2 k)
C5 = C3 = C1
54 5!
which give rise to the second solution containing odd powers of x:
[ ]
2 k 3 (12 k)(2 k) 5
y2 (x) = C1 x + x + x + ... (7.47)
3! 5!

(2) s2 = 1 The recurrence relation in this casse is:

(r + 1)(r + 2) k
Cr+2 = Cr
(r + 3)(r + 2)

Since the second term in Eq.(7.43) should go, we have to set C1 = 0. Then, starting from C0 , we obtain C2 , C4 , ..
and so on, but it can easily be checked that this way we obtain an expansion which is the same as y2 (x) obtained
already. Thus, the second value of s does not lead to new solutions.

Convergence (see Eq.(7.45)):


[ ]
Cr+2 xr+2 r(r + 1) k 2 r k

y1 (x) = x == x2 x2 as r ,
Cr xr (r + 2)(r + 1) r + 2 (r + 1)(r + 2)

since
k
0
(r + 1)(r + 2)
and
r 1
= 2 1
r+2 1+ r

when r . Therefore, the series in y1 (x) converges for x2 < 1, 1 < x < 1, as expected. Similar analysis is
performed for y2 (x):
Cr+2 xr+2
y2 (x) = r(r + 1) k x2 x2
Cr xr (r + 1)(r + 2)
as r , i.e. we have 1 < x < 1 as well.
In general, either solutions diverge at x = 1. When l=integer, then either of the solutions become polynomials
which are nite at x = 1. Indeed, consider rst l=even. Then k = l(l + 1) = even, and thus for r = l we have for
y1 (x) that the coecient
r(r + 1) k
Cr+2 = Cr 0
(r + 1)(r + 2)
r(r + 1) k
Cr+4 = Cr+2 0
(r + 1)(r + 2)
and so on, i.e. all the coecients after Cr+2 will also be equal to zero, and, therefore, the series expansion terminates
at Cr , i.e. y1 (x) is a polynomial with a nite number of terms. For, example, if l = 2, then k = 2 3 = 6 and
6
C2 = C0 = 3C0
2!
CHAPTER 7. SERIES SOLUTIONS OF ORDINARY DIFFERENTIAL EQUATIONS 80

(6 6)6
C4 = C0 0
4!
C6 = C8 = ... = 0
and thus y1 (x) becomes simply [ ]
y1 (x) = C0 1 3x2 .
At the same time, it can be seen that the other solution, y2 (x), will not terminate for any even l.
If l=odd, then y2 (x) is of a polynomial form, while y1 (x) will be an innite series.
The solutions are called Legendre polynomials.
Chapter 8

Partial Dierential Equations

8.1 Introduction
Any relation between an unknown function (x, y, ...) and its partial derivatives is called a partial dierential
equation (PDE), with independent variables (x, y, ...) and dependent variable .
Examples:
2 2
+ = 0, (8.1)
x2 y 2
2 2
y 2
+ x 2 2 2 = 0, (8.2)
x y
2 2 2
+ + + xy = 0. (8.3)
x2 xy y 2
The order of a PDE is the highest order of the partial derivatives in the equation. It is clear that Eqs. (8.1), (8.2)
and (8.3) are all of second order.
A PDE is linear if and its partial derivatives occur only to the rst degree in the equation, and the products
of and its partial derivatives are absent. We see that Eqs. (8.1) and (8.3) are linear equations, while Eq.
(8.2) is non-linear because of the term 2 . Many of the partial dierential equations which are of importance in
mathematical physics are of the linear type.
A linear PDE is homogeneous if each term in the equation contains either the dependent variable , or one of its
partial derivatives. It is seen that Eq. (8.1) is a linear homogeneous PDE, while Eq. (8.3) is a linear inhomogeneous
PDE because of the term xy.

8.2 PDEs of Physical Importance: Introduction


8.2.1 Classical Wave Equation
In many physical applications the following, the so-called wave equations, is met:

1 2
2 = , (8.4)
c2 t2
where 2 is the Laplacian operator written in the appropriate number of variables (1D, 2D or 3D space) and
coordinate system (i.e. using cartesian or curvilinear coordinates) , t is a time variable and c is a constant. This
linear homogeneous PDE of second order describes classical wave motions with constant phase velocity c. Several
examples are considered below.

8.2.1.1 Transverse waves on a string


Consider a string of uniform line density which is stretched by a uniform tension T0 , see Fig. 8.1. If the string is
allowed to undergo small transverse vibrations then the displacement (x, t) satises the PDE

2 1 2
= , (8.5)
x2 c2 t2

81
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 82

x
0 x L

Figure 8.1: Transverse waves in a string of length L.

Figure 8.2: A circular membrane of radius a.

where the 1D Laplacian operator has been used, and


( )1/2
T0
c= . (8.6)

For large transverse displacements one nds that (x, t) satises a more complicated non-linear PDE.

8.2.1.2 Vibrations of a circular membrane


Consider a uniform membrane which is stretched over a rigid circular support of radius a, as shown in Fig. 8.2.
If the membrane is allowed to vibrate with a small amplitude then the transverse displacement (, , t) (in polar
coordinates (, )) satises the wave equation (8.4) with
( )
2 2 1 1 2
2 = + = + (8.7)
x2 y 2 2 2
written in the polar coordinates, and
()1/2
T
c= , (8.8)
d
where T is the surface tension and d is the surface density of the membrane.

8.2.2 Diusion Equation


The diusion equation is similar to the wave equation discussed above, but contains only the rst-order time
detrivative:
1
2 =, (8.9)
D t
where D is a constant. This linear homogeneous PDE of the second order is important for a number of physical
processes as outlined below.

8.2.2.1 Diusion processes


If we have a solute of density (

r , t) which is diusing through a solvent, then the behaviour of (

r , t) is described
by
Eq. (8.9). The quantity D is called the diusion constant.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 83

8.2.2.2 Heat conduction


If we have a process of heat conduction in a solid which has a density , thermal conductivity and specic heat
capacity CV , then the temperature (
r , t) satises Eq. (8.9) with

D= . (8.10)
CV
In this case D is called the thermal diusivity.

8.2.3 Helmholtz equation


This is the equation:

2 S + k 2 S = 0, (8.11)


where S = S( r ) and k is a constant. This equation can be obtained from the wave equation (8.4) by making the
substitution
(
r , t) = S(

r )eit . (8.12)


We then nd that the spatial function S( r ) satises (8.11) with k = /c. In a similar manner, we can also obtain
(8.11) by substituting
(
r , t) = S(
r )et , (8.13)
in the diusion equation. In this case, we nd that S( 1/2
r ) satises (8.11) with k = (/D) .

8.2.4 Laplace equation


2 = 0, (8.14)


where = ( r ). This equation plays a crucial role in potential theory. Any solution of the Laplace equation is
called a harmonic function. Examples:
1. The electrostatic potential (

r ) in a charge-free region satises Eq. (8.14).
2. The gravitational potential (
r ) in a mass-free region is a solution of Eq. (8.14).
3. The velocity potential (
r ) of an incompressible irrotational uid in the absence of sources and sinks also
satises the Laplace equation.
It should also be noted that in the steady state /t = 0, so that the diusion equation (8.9) reduces to the
Laplace equation.

8.2.5 Poisson equation


2 = f (

r ). (8.15)



The electrostatic potential ( r ) in a region with a continuous charge density ( r ) satises the inhomogeneous
PDE (8.15) with
f (

r ) = (
r )/40 . (8.16)

8.2.6 Schrdinger equation


The wave function of an electron placed in a potential V (

r ) satises the following DE:

~2 2
+ V (r ) = E, (8.17)
2m
where E is the energy of the electron corresponding to the wavefunction . This DE is an example of an eigenvector-
eigenvalue problem, similar to that in the matrix algebra: here not only , but also E are to be determined.

8.3 PDEs of Physical Importance: derivation of some equations


In this Section we shall derive some of the PDEs mentioned above.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 84

F(x)
u(x) tangent T0

A B

u(x) T0 u(x+dx)
dx
x x+dx x

Figure 8.3: Oscillating string. The tangent direction to the string at point x is indicated by a vector which makes
angle with the x axis.

8.3.1 Oscillation of a one-dimensional string


Consider a string which lies along the x axis and is subjected to a strong tension T0 in the same direction. In
equilibrium the string will be stretched along the x axis. If we apply a perpendicular external force F (per unit
length) and/or take the string out of its equilibrium position, it will start to oscillate vertically. Let the vertical
displacement of each point x be u(x, t), it is a function of both x and time, t.
Let us consider a small element AB of the string, with the point A being at x and the point B at x + dx, see
Fig. 8.3. The total force acting on this element is due to two tensions applied to points A and B which work in the
opposite directions, see Fig. 8.3, and an external force F (x), which is applied in the vertical direction. We assume
that the oscillations are small (i.e. the vertical displacement u(x, t) is small). This means that the tensions are
nearly horizontal, and cancell each other in this direction (if they did not, then the string would also oscillate in
the lateral direction!). Therefore, we should only care about the balance of the forces in the vertical direction.
Let (x) be the angle the tangent line to the string makes with the x axis at the point x as shown in the
Figure. Then, the dierence in the heights between points B and A, which is u(x + dx) u(x) = u x dx (the time
is omitted for convenience, so this is a partial change of u(x, t) with time kept constant), can be calculated as
x dx = dx tan (x) (x)dx, because for small oscillations the angle is small and therefore tan sin ;
u

x (x). The vertical component of the tension force acting downwards at the left point A is
hence u
( )
u
T0 sin (x) T0 (x) = T0
x A

On the other hand, the vertical component of the tension applied to the point B is similarly
( )
u
T0 sin (x + dx) T0 (x + dx) = T0
x B

Therefore, the total force acting on the element dx in the vertical direction will be
( ) ( ) [( ) ( ) ]
u u u u
F dx + T0 T0 = F dx + T0
x B x A x B x A

The expression in the square brackets gives a change of the function f (x) = u
x between the two points B and A
2u
which are separated by dx, i.e. it is f (x + dx) f (x) = x dx x2 dx. Therefore, the total force acting on the
f

element dx of the string in the vertical direction will be


( )
2u
F + T0 2 dx
x

On the other hand, due to Newtons equations of motion, this force should be equal to the element mass, dx (here
2
is the string line density), multiplied by the vertical acceleration, which is given by t2u . Therefore, the equation
of motion of the element dx will be this:
( )
2u 2u
dx 2 = F + T0 2 dx
t x
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 85

which, after cancelling out on dx, gives the nal equation of motion for the string:

1 2u 2u
= G(x) +
c2 t2 x2
where G(x) = F (x)/T0 and c2 = T0 /. When the external forces F are absent, we arrive at the familiar wave
equation (8.5) for the spring, with the velocity c indeed given by Eq. (8.6).

8.3.2 Hydrodynamic equations of ideal liquid (gas)


We consider an ideal liquid (it could also be a gas), i.e. we assume that the forces applied to its any nite volume
V with the surface S can be expressed via pressure P due to external (for the chosen volume) part of the liquid.
The latter exerts a force acting inside the volume, i.e. in the direction opposite to the surface normal
n assumed
to be directed out of the volume. In other words, the total force due to the external (for the volume) part of the
liquid is
Pn dS (8.18)
S
Consider now the Gausss theorem (the 1st semester). It states that for any volume V with the boundary surface


S and a vector eld F (x, y, z)



Fn dS = div F dV
S V


If we take the vector eld F = (P, 0, 0), we obtain from this:

P
P nx dS = dV
x



Similarly, by taking F = (0, P, 0) and F = (0, 0, P ) we obtain from this, respectively:

P P
P ny dS = dV and P nz dS = dV
y z



Multiplying each of these equations by the unit base vectors i , j and k , respectively, and then summing them
up, we obtain a useful integral formula:
( )

P P P
P n dS = i + j + k dV gradP dV (8.19)
S V x y z V

Using this nice result, which is a consequence of the Gausss theorem, the force due to pressure (8.18) can be also
written via the gradient of the pressure as



P n dS = gradP dV (8.20)
S V


We can also have an external force density F acting on the liquid, so that the total force acting on the volume
will be then ( )



F dV gradP dV = F gradP dV
V V V
According to the Newtons second law, this force results in the acceleration of the liquid volume given by

d
v
dV
V dt

where

v is the velocity of the liquid depending on time t and the point
r = (x, y, z). Therefore, one can write:
( )
d
v
dV = F gradP dV
V dt V

or ( )
d
v

F + gradP dV = 0
V dt
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 86

Since the choice of the volume is arbitrary, the integrand must be equal to zero, i.e. we obtain:
d
v

= F gradP
dt
Here the acceleration is given by the total derivative of the velocity. Indeed,
v is the function of time and the
coordinates, and the liquid particles move in space, so that their coordinates change as well. Therefore:
d
v
v

v x
v y
v z
v

v

v

v
= + + + + vx + vy + vz
dt t x t y t z t t x y z
The coordinate derivatives term is normally written in the following short form as a dot product:




vz =

v grad

v v v
vx + vy + v
x y z
where the gradient is understood to be taken separately for each component of the velocity eld. Finally, we arrive
at the following equations:

1
+
v grad

v
v = F gradP (8.21)
t
This equation is to be supplemented with the continuity equation

+ div (


v ) = 0 (8.22)
t
derived in the 1st semester course. Given the equation of states for the liquid, i.e. how the pressure eects the
density, P = f (, T ), one can solve these two equations to obtain the velocity eld in the liquid under the applied


external forces F and temperature T . Note that the divergence term in the continuity equation (8.22) can also be
written as:
div (
(vx ) (vy ) (vz )
v ) = + +
x y z
vx vy vz
= + vx + + vy + + vz
x x y y z z
( ) ( )
vz div
v +

vx vy vz
= + + + vx + vy + v grad
x y z x y z
so that the continuity equation may also be written as follows:

+ div

v +


v grad = 0 (8.23)
t

8.4 Propagation of sound


We assume that velocities of a gas change very little in space, i.e. we can neglect the v grad

v in Eq. (8.21).
Then, we have

v
1
= F gradP (8.24)
t
Similarly, the continuity equation (8.23) can be simplied by neglecting the

v grad term containing small velocities:

+ div


v =0 (8.25)
t
Let 0 be the density of the gas in equilibrium. Then, relative uctuation s of the gas density can be introduced
via = 0 (1 + s), so that d = 0 ds and
d ds
= ds
1+s
assuming uctuations of the density are small. Therefore, one can write:
t s
t and hence the continuity
equation (8.25) takes on a simpler form:
+ div

s
v =0 (8.26)
t
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 87

Next, we can assume that the gradient of the pressure is proportional to the gradient of density uctuations:

gradP = e grad s

where e is a material constant related to the elasticity of the media (the gas). Substituting this into the original
Eq. (8.24) and replacing with 0 there, we get:


v
e
= F grad s
t 0
Take the divergence of both parts of this equation. The left hand side:
( )
2s
(div

v
div = v)= 2
t t t

where we used Eq. (8.26) at the last step. The right hand side:
( )
e
e
e
div F grad s = div F div grad s = div F s
0 0 0

i.e. it contains the Laplacian of s. Equating the left and the right hand sides, we nally obtain the equation sought
for:
2s

= a2 s div F (8.27)
t2
where a2 = e/0 . If the external forces are absent, this equation turns into a familiar wave equation in 3D space:

2s
= a2 s (8.28)
t2

8.5 Heat conduction equation


We start by stating an experimental fact (the Fouriers law) about the heat conduction: the amount of heat passing
through a surface dS per unit time is proportional to the gradient of temperature there:
dQ T
= dS
dt n
where is the thermal conductivity, Tn is the directional derivative showing the change of temperature in the
direction perpendicular to the normal

n to the surface dS. According to the denition of the directional derivative,


n can be directly related to the temperatrure gradient as follows: n = n gradT .
T T

Consider now a nite volume V of some media (e.g. gas) with the surface S. If the energy ow goes across the
boundary surface S outside in the direction along its normal n (directed outwards), the temperature inside the
T
volume is reduced and hence obviously n < 0. We then dene the amount of heat given away by the volume V to
the environment around it,
gradT

T
dQ = dt dS = dt n dS
S n S
as positive. This is how we dene dQ, and for this the minus sign is necessary. Using the Gausss theorem, the
surface integral can be turned into the volume one:

dQ = dt div ( gradT ) dV
V

where may depend on the spatial position in general and hence cannot be taken out of the divergence. On the
other hand, the heat dQ is the one lost by the volume over
( time ) dt due to decrease of its temperature. If we take
a small volume dV inside V , then this volume lost CV T t dt dV of heat (again, this quantity is positive as the
time derivative of the temperature is negative). The total loss in the volume is
( )
T
dQ = CV dt dV
V t
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 88

(x,t)

x
0 x L

Figure 8.4: A stretched exible string of length L.

The two last expressions for dQ are to be equal, i.e. we obtain:


( )
T
CV dt dV = dt div ( gradT ) dV
V t V
or [ ]
T
CV div ( gradT ) dV = 0
V t
Again, due to the fact that the volume V inside the whole system was chosen is arbitrarily, the integrand should
be equal to zero:
T
CV = div ( gradT ) (8.29)
t
This is the required heat transport equation which, given the appropriate boundary and initial conditions, should
provide us with the temperature distribution in the system T (x, y, z, t) over time. For a homogeneous system the
constant does not depend on the spatial variables and can be taken out of the divergence. Since divergence of the
gradient is the Laplacian, we arrive at the more familliar form of the heat transport equation:
1 T
= T (8.30)
D t
where D = / (CV ) is the thermal diusivity. In stationary conditions, e.g. at long times t in the case of
constant in time sources of heat (for instance, when a rod is kept at constant temperatures T1 and T2 at its both
ends for indenitely long time), the temperature does not longer depend on time and we arrive at the stationary
distribution T (x, y, z) which satises the Laplace equation:

T = 0 (8.31)

This is to be solved under the given boundary conditions.

8.6 Boundary Value Problems


In physical applications one usually has to nd a particular solution (

r , t) of a PDE which is valid in some region
R and satises certain conditions at t = 0, and on the boundary of R.
A possible procedure for solving this type of boundary value problem is to determine the general solution of the
PDE using a direct integration method, and then impose the initial and boundary conditions.

Example 1: A stretched exible string with xed ends at x = 0 and x = L, as shown in Fig. 8.4, is given an
initial vertical displacement A sin(x/L), and then released from rest at t = 0. The displacement (x, t) for t > 0
is the solution of the wave equation:

2 1 2
2
= 2 2, (8.32)
x c t
which satises the boundary conditions

(0, t) = 0, (L, t) = 0 for t 0, (8.33)


CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 89

and the initial conditions


(x, 0) = A sin(x/L), t (x, 0) = 0 for 0 x L, (8.34)
where t denotes /t.

Solution: The general solution of (8.32) may be obtained by rst applying the coordinate transformation:

= x + ct, (8.35)

= x ct. (8.36)
Under this transformation we have
[ ]
1 1
(x, t) ( + ), ( ) (, ), (8.37)
2 2c

so that

= + = + = 1
x x x
( )
1 1
= + = = 2
c t c t t
so that
2 1 1 1 2 2 2
= = + = + 2 + , (8.38)
x2 x 2 2
1 2 1 2 2 2 2 2 2
2 2
= = = 2
2 + . (8.39)
c t c t 2
Hence the wave equation (8.32) becomes:
2
= 0. (8.40)

The general solution of this PDE is
(, ) = u() + v(), (8.41)
where u() and v() are arbitrary functions of and , respectively . We now see that the general solution of (8.32)
is
(x, t) = u(x + ct) + v(x ct) (8.42)
The particular solution (x, t) which satises the boundary conditions is found to be:
A [ ] A [ ]
(x, t) = sin (x + ct) + sin (x ct) , (8.43)
2 L 2 L
( x ) ( )
ct
= A sin cos . (8.44)
L L
To show this, we rst apply the initial conditions:
( )
(x, 0) = u(x) + v(x) = A sin x (8.45)
L
t (x, 0) = u (x)c v (x)c = c [u (x) v (x)] 0.
Integrating the 2nd equation, we obtain:
u(x) v(x) = B (8.46)
where B is a constant. Solving together with the rst equation (8.45), one obtains:

A ( ) B
u(x) = sin x + ,
2 L 2
A ( ) B
v(x) = sin x ,
2 L 2
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 90

so that [ ] A [ ]
A
(x, t) = u(x + ct) + v(x ct) = sin (x + ct) + sin (x ct)
2 L 2 L
since the constants cancel out.
Note that the boundary conditions are satised automatically:
A ( ) A ( )
(0, t) = sin ct + sin ct = 0
2 L 2 L
A [ ] A [ ]
(L, t) = sin (L + ct) + sin (L ct)
2 L 2 L
A ( ) A ( )
= sin + ct + sin ct
2 L 2 L
A ( ) A ( )
= sin ct + sin ct = 0
2 L 2 L
which proves Eq.(8.43).

The general solution of a second-order ODE involves two arbitrary constants. However, we see from Eq. (8.42)
that the general solution of a second-order PDE will involve two arbitrary functions. In most cases the general
solution of a PDE will not be known. Even when a general solution is available it will usually not be very useful,
because of the diculty of choosing the arbitrary functions in such a way that the boundary conditions are satised.
It is clear, therefore, that we need to alternate methods for solving PDEs which enable one to construct the
required solution around the appropriate boundary conditions.
In the following notes we shall discuss specialised methods for solving the linear homogeneous PDEs which occur
in mathematical physics, see Eqs. (8.4), (8.5), (8.9), (8.11), (8.14) and (8.17).

8.7 Superposition Principle


If each of the functions 1 , 2 , ..., n are solutions of a linear homogeneous PDE then any linear combination of
these functions
n
= Ci i , (8.47)
i=1

is also a solution of the PDE, where C1 , C2 , ..., Cn are arbitrary constants. This result is readily veried for the
linear homogeneous PDEs of physical interest by substituting (8.47) in the equation. For example, if 1 , 2 , ..., n
are solutions of the Laplace equation we have 2 i = 0 (i = 1, 2, ..., n). However, it is also clear that:
( n )
n
2
Ci i = Ci 2 i = 0. (8.48)
i=1 i=1

We shall now show that the superposition principle plays a crucial role in the Fourier method for solving boundary
value problems.

8.8 Fourier Method for Linear Homogenenous PDEs


The various stages in this method are:
1. Substitute in the PDE the product form:

(x, y, ...) = X(x)Y (y)..., (8.49)

where X(x), Y (y), ... are functions of just one independent variable x, y, ..., respectively, and then attempt to
separate the variables in the resulting equation.
2. If this procedure is successful, one obtains a set of separated ordinary dierential equations which can be solved
for the functions X(x), Y (y), .... It is now possible to write down a set of product solutions i (i = 1, 2, ...) of
the type (8.49) for the PDE.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 91

3. Next we use the superposition principle to construct the general solution of the PDE in the form


= Ci i . (8.50)
i=1

Finally, the values of the constant coecients Ci are determined by imposing the inital and boundary condi-
tions.

Example 2: Use the Fourier method to determine the solution of the wave equation

2 1 2
= , (8.51)
x2 c2 t2
which satises the boundary conditions

(0, t) = 0, (L, t) = 0, for t 0 (8.52)

and the initial conditions


(x, 0) = f (x), t (x, 0) = g(x) for 0 x L. (8.53)
The stretched string problem discussed in Example 1 is a special case of this boundary value problem with f (x) =
A sin(x/L), and g(x) = 0.

Solution: The substitution of the product form,

(x, t) = X(x)T (t), (8.54)


in Eq. (8.51) gives:
d2 X 1 d2 T
T = X . (8.55)
dx2 c2 dt2
Hence we have
1 d2 X 1 d2 T
= . (8.56)
X dx2 c2 T dt2
The left hand side of (8.56) is a function of x only, while the right hand side is a function of t only. It follows
that Eq. (8.56) can only remain valid for all values of x and t if both sides of the equation are equal to the same
constant K. Hence we must have:
d2 X
= KX, (8.57)
dx2
d2 T
= c2 KT, (8.58)
dt2
which are two ordinary DEs for the functions X(x) and T (t), respectively. The constant K is called the separation
constant.
We shall now consider Eq. (8.57) for dierent values of K:
1. When K > 0, we can write K = p2 . For this case the solution of (8.57) is

X(x) = Aepx + Bepx (8.59)

If the boundary conditions at x = 0 and x = L are applied to Eq. (8.59) one nds that we must have A = 0
and B = 0. This leads to the trivial solution (x, t) 0, which is not of physical interest!

2. When K = 0, we nd that
X(x) = Ax + B. (8.60)
This solution is also of no physical interest, because the boundary conditions at x = 0 and x = L can only be
satised when A = 0 and B = 0.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 92

3. When K < 0, we can write K = k 2 . For this case we have

X(x) = A sin(kx) + B cos(kx). (8.61)

The application of the boundary conditions at x = 0 and x = L to (8.61) gives:

B = 0 (when using x = 0 condition), (8.62)

A sin(kL) = 0 (when using x = L condition). (8.63)


In Eq. (8.63) the constant A = 0 since otherwise we arrive again at the trivial solution. Hence, Eq. (8.63)
will be satised if k takes one of the discrete values:
n
kn = , n = 1, 2, ..., (8.64)
L
which follows from the roots of the equation sin(kL) = 0. Note that n = 0 as zero value of n would result in
zero value of k and hence in the trivial solution. Also, negative values of n do not give anything new ( as these
)
result in
( the same
) values of the separation constant K = k 2
n and the solutions X |n| (x) =
( A sin
) k|n| x =
A sin k|n| x , which dier only by the sign from the positive n solutions X|n| (x) = A sin k|n| x .
From the above analysis we see that the boundary conditions can only be satised when the separation constant
K takes certain discrete eigenvalues Kn = kn2 (n = 1, 2, ...). This type of situation is of frequent occurence in
the theory of PDEs, and in the Schrodinger equations it gives a mathematical mechanism for the quantisation of
energy!
Associated with the eigenvalue kn we have a physically acceptable eigenfunction

Xn (x) = An sin(nx/L), n = 1, 2, ..., (8.65)

where An are some constants, which may be dierent for dierent n, so that we gave them n as an index. These
eigenfunctions satisfy the orthogonality relation:
L
sin(nx/L) sin(mx/L)dx = (L/2)nm , (8.66)
0

where nm denotes the Kronecker delta symbol (nm = 1 if n = m and nm = 0 if n = m). Derivation of Eq. (8.65)
is given in Appendix 1 at the end of this Section.
Next we solve the dierential equation (8.58) with K = kn2 . It is found that

Tn (t) = Dn sin(n t) + En cos(n t), (8.67)

where
n = n(c/L), n = 1, 2, .... (8.68)
We can now use Eqs. (8.65) and (8.67) to obtain a set of product solutions for the PDE:

n (x, t) = An sin(nx/L) [Dn sin(n t) + En cos(n t)] = sin(nx/L) [Bn sin(n t) + Cn cos(n t)] (8.69)

where n = 1, 2, ... and we combined products An Dn and An En of arbitrary constants into new arbitrary constants
Bn and Cn . Each such elementary product solution satises the boundary conditions! It should be noted that the
special motion n (x, t) is called the n-th normal mode of vibration for the system, and its angular frequency of
vibration is n = n(c/L). The n = 1 normal mode is called the fundamental, and has an angular frequency
( )1/2
T
1 = , (8.70)
L

where (8.6) was used.


At that point, however, initial conditions may still not be satised! Therefore, in the nal stage of the analysis
we use the superposition principle to construct a general solution of the PDE:


(x, t) = sin(nx/L) [Bn sin(n t) + Cn cos(n t)] , (8.71)
n=1
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 93

which is a linear combination of the elementary solutions (8.69). The arbitrary constants Bn and Cn must be chosen
so that the initial conditions are satised. Thus, we must have:


(x, 0) = Cn sin(nx/L) f (x), (8.72)
n=1



t (x, 0) = Bn n sin(nx/L) g(x). (8.73)
n=1

Equations (8.72) and (8.73) are just Fourier sine series for f (x) and g(x), respectively. It follows, therefore, that
(see Chapter 2 concerning the method of nding the coecients in the expnasion over the functions forming an
orthonormal set):

2 L
Cn = f (x) sin(nx/L)dx, (8.74)
L 0
L
2
Bn = g(x) sin(nx/L)dx, (8.75)
n L 0
where n = 1, 2, ....

Special Cases:

1. Initial displacement is zero, i.e. f (x) 0. For this case we have Cn = 0 for all n.

2. Initial velocity is zero, i.e.we have g(x) 0. For this case we have Bn = 0 for all n.

Appendix 1. Orthogonality Relation


The easiest way to prove Eq. (8.66) is to rst show that
1 ( i ) 1 ( i )
e ei
sin sin = e ei
2i 2i
(
1 i(+) ) (
1 i() )
= e + e(+) + e + ei()
4 4
1 1 1
= cos( + ) + cos( ) = [cos( ) cos( + )] .
2 2 2
Therefore, { }

L
nx mx 1 L
(n m)x (n + m)x
sin sin dx = cos cos dx.
0 L L 2 0 L L
If n = m, we obtain
{ } L
1 L
2nx L 1 L 2nx L
1 cos dx = sin = .
2 0 L 2 2 2n L 0 2
If now n = m, we obtain instead:
L L


1 L (n m)x 1 L (n + m)x
sin
2 (n m) | 2 (n + m) sin =0
{zL } | {zL }
=0 0 =0 0

so that the result is zero. Note the (nm) in the denomitator in the last expression! This means that the [ n( = m1case
) ]
must be considered separately. Note also, that n can also be a half-integer, i.e the functions n (x) = sin L n+ 2 x
also satisfy the orthogonality relation
L
L
n (x)m (x)dx = nm .
0 2
The proof given above is valid for this case as well!
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 94

fixed

y
fixed
fixed
x
fixed
Figure 8.5: 2D rectangular membrane.

8.9 Oscillations of a rectangular membrane


Let (x, y, t) be the tansverse displacement of the point (x, y) of the membrane, see Fig. 8.5. Then, the corre-
sponding wave equation to solve is:
2 2 1 2
2
+ 2
= 2 2 (8.76)
x y c t
It should satisfy the boundary conditions:
{
(x, 0, t) = (x, L, t) = 0, x [0, L]
(8.77)
(0, y, t) = (L, y, t) = 0, y [0, L]

Initial conditions:

(x, y, 0) = 1 (x, y), = t (x, y, 0) = 2 (x, y) (8.78)
t t=0
Separation of variables,
(x, y, t) X(x)Y (y)T (t) (8.79)
gives, after substituting into the equation:
1
X Y T + XY T = XY T
c2
Divide by XY T :
X Y 1 T
+ = (8.80)
X
|{z} Y
|{z} c2 T
|{z}
depends only on x depends only on y depends only on t
Therefore, we can write:
X Y 1 T
= k1 , = k2 , 2 = k1 + k2
X Y c T
where k1 and k2 are separation constants. We have three equations (ODE):

X = k1 X with respect to x, (8.81)

Y = k2 Y with respect to y, (8.82)


T = c2 (k1 + k2 ) T with respect to t. (8.83)
What are the permissible values of k1 , k2 ? Consider k1 and apply the boundary conditions:
1. k1 = 0 gives X = 0, X(x) = Ax + B which satises conditions (8.77) only if A = B = 0, i.e. this is of no
interest!
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 95

2. k1 = 2 > 0 gives X = 2 X with the following solution:

X(x) = Aex + Bex

Again, the second line of the Eq.(8.77) is only satised if A = B = 0 which is of no interest to us again!
3. k1 = 2 < 0 yields X + 2 X = 0,

X(x) = A sin(x) + B cos(x)

Apply the boundary conditions (8.77):

X(0) = 0 = B B = 0

X(L) = A sin L = 0 sin L = 0 or



n = n, n = (1, 2, ...).
L
Thus, allowed solutions for X(x) are given by eigenfunctions
( n )
Xn (x) = sin x , n = 1, 2, ... (8.84)
L
with eigenvalues k1 = 2n , n = 1, 2, .... We do not need to bother about the constant amplitude (prefactor) A
here as it will be absorbed by other constants in T (t) in the product elementary solution (see below). So we simply
choose it as unity.
The boundary condition (8.77) applied to (8.82) gives a similar result for k2 , namely: k2 = 2m with m = 1, 2, ...
being another positive integer, and the corresponding eigenfunctions are
( m )
Ym (y) = sin y , m = 1, 2, ... (8.85)
L
where again we do not keep the amplitude (prefactor) to the sine function as it will be absorbed by constants in
T (t).
Next we consider Eq. (8.83) for T (t):
( )
T + c2 2n + 2m T = 0 (8.86)

Solution:
T (t) = A sin (nm t) + B cos (nm t) (8.87)
with the frequency
c 2
nm = c 2n + m 2 = n + m2 . (8.88)
L
The general solution then is the linear combination of all possible elementary solutions:

( n ) ( m )
(x, y, t) = (Anm sin nm t + Bnm cos nm t) sin x sin y . (8.89)
n,m=1
L L

As you can see, there was no need to keep the prefactors in Xn (x) and Ym (y): when constructing the linear
combination above, these constants would simply be absorbed by the contstants Anm and Bnm already contained
in T (t).
The constants Anm and Bnm are to be obtained from the initial conditions in the same way as for the 1D string.
To obtain Bnm , we apply the boundary condition for (x, y, 0):
( n ) ( m )
(x, y, 0) = Bnm sin x sin y 1 (x, y)
nm
L L

Here (x, y, 0) is expanded into a double Fourier series with respect to the sine functions, so that the expansion
coecients, Bnm , are found from it in the usual way:
( )2 L L ( n ) ( m )
2
Bnm = dx dy1 (x, y) sin x sin y (8.90)
L 0 0 L L
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 96

To obtain Anm , we apply the condition for the time derivative at t = 0. Since

( n ) ( m )
t (x, y, t) = = nm (Anm cos nm t Bnm sin nm t) sin x sin y ,
t n,m=1
L L

then ( n ) ( m )
t (x, y, 0) = nm Anm sin x sin y 2 (x, y)
nm
L L
and, therefore,
( )2 L L ( n ) ( m )
1 2
Anm = dx dy2 (x, y) sin x sin y .
nm L 0 0 L L

8.10 Question of Separability


No general rules exist to tell us whether a given PDE is separable or not in a particular coordinate system - we
must just test and see! For example, consider the linear homogeneous PDE of second order:
( )2
2 + x2 + y 2 = 0, (8.91)

2 2
with 2 = 2x + 2y . The substitution
(x, y) = X(x)Y (y) (8.92)
gives ( ) ( )
1 d2 X 1 d2 Y
+ x4 + + y4 + 2x2 y 2 = 0. (8.93)
X dx2 Y dy 2
We see that in Cartesian coordinates Eq. (8.91) is not separable, because of the term 2x2 y 2 .
Suppose now we transform (8.91) to plane polar coordinates, dened by x = cos , y = sin . This procedure
gives: ( )
1 1 2
+ 2 + 4 = 0. (8.94)
2
Now make the substitution
(, ) = R()(), (8.95)
in the transformed Eq. (8.94). We nd that:
[ ( ) ]
1 d dR 1 d2
+ 6 R + = 0. (8.96)
R d d d2

This result can be separated into two ODEs!


We see that a PDE may be separable in one coordinate system, but not in another! This shows importance of
knowing various coordinate systems.

8.11 Heat-Conduction Equation


Here we shall consider solution of the Heat-conduction partial dierential equation
1
= 2
t
where (x, y, z, t) is the temperature and the thermal diusivity, t - time. This problem has some peculiar
points specic to this problem. First of all, we shall consider the situation when a source of heat is constant (the
temperature at the boundaries of the sample does not change with time). In this case the distribution of temperature
in the sample would eventualy (at t ) stabilise, i.e. it would reach a stationary distribution.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 97

Stationary solution
The stationary solution of the heat-conduction equation, satises the condition:


= 0.
t t=
Therefore, the stationary solution thus satises the Laplace equation

(x, y, z, t ) = 0

as it is just a function of spatial coordinates.

Example 4 Find the stationary solution for the temperature distribution in a 1D rod of length L, ends of which
are kept at constant temperatures
(0, t) = 0
(L, t) = 1

d2
Solution: In 1D we have the equation dx2 = 0 with the solution

(x) = Ax + B

where A, B are found from the boundary conditions


{
AL + B = 1 A = 1 0
L
B = 0 B = 0

and the nal solution is the linear distribution of temperature:


x
(x) = 0 + (1 0 ) .
L
It does not depend on time t. Thus, the distribution of temperature in a rod kept at dierent temperatures at both
ends, is linear.

Heat ow problem
Suppose, the initial distribution of the temperature in a bar is described by some function (x):

(x, 0) = (x)

Suppose then, that starting from t = 0 the two ends at x = 0 and x = L are maintained at constant temperatures
0 and 1 . Find the (x, t) at any time t > 0.
We have to solve the following problem:
1 2
=
t x2

(x, 0) = (x)
(0, t) = 0

(L, t) = 1 .
It is convenient to simplify rst the boundary conditions, since our method of separation of variables will only work
if we have zero boundary conditions. So, we have to reformulate the problem to have the zero boundary conditions
instead. To this end, we recall that will tend to the stationary distribution
x
(x, t = ) = + (1 0 )
L
at t which satises the same boundary conditions. Therefore, it is advisable to make a substitution:
[ x]
= 0 + (1 0 ) + U (x, t)
L
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 98

and formulate the problem for the new function U (x, t) instead:

1 U 2U
=
t x2
which satises
U (0, t) = 0 and U (L, t) = 0
The problem is simpler now since the boundary conditions are zero boundary conditions. Therefore, we can use the
method of separation of variables as with the wave equation.

Example 5 A sphere of radius S has temperature 0 . At t = 0 it was placed in water tank of temperature 1 .
Find the stationary distribution of the temperature in the sphere (i.e. at t ).

Solution: We solve the stationary problem using the spherical coordinates:


( )
1 d 2 d
r = 0,
r2 dr dr

d d A A
r2 = A, = 2 , d = 2 dr
dr dr r r

Adr A
(r) = = +B
r2 r
The solution should be nite at all r including the r = 0, i.e. it then follows that A == 0. Thus,

(r) = B at all 0 r S.

Since (S) = 1 (r) = 1 , the result to be expected: the whole sphere would have the same temperature as
the water in the tank around it.

Example 6 Find the complete solution for the sphere of temperature 0 placed in a bath of water with temperature
1 .

Solution: We should solve the heat-conduction equation


( )
1 1
= 2 r2
t r r r
with the following initial,
(r, 0) = 0 , 0 r < S,
and boundary
(g, t) = 1
conditions.
First of all, we separate out the stationary solution (r, t) = 1 from the solution, i.e we introduce a new
function
(r, t) = (r, t) 1
which satises: ( )
1 1
= 2 r2
t r r r
(r, 0) = 0 1
(S, t) = 0.
Next, we use the method of separation of variables:

(r, t) = R(r)T (t)


CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 99

1 T 1 1 d ( 2 )
= r R k 2 .
T R r2 dr
Note that k 2 should be used here since T (t) should decay:
1 T
= k 2 T (t) = T0 ek t .
2

T
A positive value of the separation constant +k 2 would give a non physical increase of to .
Consider now the equation for R(r):
1 ( 2 )
r R = k 2 R
r2
2
R + R + k 2 R = 0.
r
It is solved by using R1 = rR which gives for R1 :
1
R= R1
r
1 1
R =
R1 + R1
r2 r
2 2 1
R = 2 R1 + 3 R1 + R1
r r r
and we obtain ( ) ( )
2 2 1 2 1 1 1
2 R1 + 3 R1 + R1 + 2 R1 + R1 + k 2 R1 = 0
r r r r r r r
Simplifying, we obtain:
1 [ ]
R1 + k 2 R1 = 0
r
R1 + k 2 R1 = 0
R1 (r) = A sin (kr) + B cos (kr)
R1 sin (kr) cos (kr)
R(r) =
=A +B
r r r
Similarly to the stationary case considered in the previous example, B = 0 since otherwise R(r = 0) = (note
that sinrkr is nite ar r = 0)1 . Thus:
sin (kr)
R(r) = A
r
Use now the boundary conditions:
sin (kS)
R(S) = A =0
S

sin (kS) = 0 k kn = n, n = 1, 2, ...
S
Therefore, we obtain the following general solution:

sin (kn r) kn2 t
(r, t) = An e , kn = n
n=1
r S

The functions {sin kn r} form an orthogonal set,


S
S
sin (kn r) sin (km r) dr = nm ,
0 2
so that to obtain the soecients An we can use the initial conditions:

sin (kn r)
(r, 0) = 0 1 An
n=1
r
1 You sin x
may recall that limx0 x
= 1.
CHAPTER 8. PARTIAL DIFFERENTIAL EQUATIONS 100



r (0 1 ) = An sin (kn r)
n=1

Multiplying both sides on sin (km r) and integrating over r between 0 and S, we obtain:
S S
2 2
An = (0 1 ) r sin (kn r) dr = (0 1 ) r sin (kn r) dr
|{z}
S 0 S 0 | {z }
u dv

(calculating by parts) [ ]
S S
2 1 1

= (0 1 ) r cos (kn r) + cos (kn r) dr
S kn 0 kn 0
[ S ]
2 S 1 2
= (0 1 ) cos (kn S) + 2 sin (kn r) = (0 1 ) cos (kn S)
S kn kn 0 kn

since sin (kn S) = sin (n) = 0. Next,


( )
cos (kn S) = cos nS = cos (n) = (1)n
S
and hence
2S
An = (1)n (1 0 )
n
and we nally obtain:

2S (1)n sin (kn r) kn2 t
(r, t) = (1 0 ) e
n=1 n r
so that, nally,

2S (1)n sin (kn r) kn2 t
(r, t) = 1 + (r, t) = 1 + (1 0 ) e
n=1 n r
At t it tends to 1 as expected.
Chapter 9

Legendre Polynomials Pn(x)

9.1 Generating function


Consider expansion of

1
G(x, t) = Pn (x)tn (9.1)
1 2xt + t2 n=0

with respect to t in a power (Taylor) series. Here 1 < x < 1, otherwise the square root is complex. Indeed,
( )
f (t) = 1 2xt + t2 = (t x)2 + 1 x2

is a parabola, see Fig. 9.1. It is positive for all values of t only if 1 x2 > 0, i. e. when 1 < x < 1.

9.2 Recurrence relations 1


The generating function (9.1) can help in constructing the recurrence relations for Pn (x). Dierentiating the Taylor
expansion for G(x, t) with respect to t:

G
= Pn (x)ntn1 . (9.2)
t n=0

On the other hand,



G 1 xt xt xt
= = = G(x, t) = Pn (x)tn . (9.3)
t t 1 2xt + t2 (1 2xt + t )
2 3/2 1 2xt + t 2 1 2xt + t 2
n=0

1x2

t
0 x
Figure 9.1: Function f (t).

101
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 102

The two expressions should be identical. Hence, we obtain:



( )
1 2xt + t2 nPn (x)tn1 = (x t) Pn (x)tn
n=0 n=0

or






nPn tn1 2xnPn tn + nPn tn+1 = xPn tn Pn tn+1 .
n=0 n=0 n=0 n=0 n=0

Collect sums with similar powers of t:






nPn tn1 (2n + 1)xPn tn + (n + 1)Pn tn+1 = 0.
n=0 n=0 n=0
| {z } | {z }
n1n n+1n

Change summation indices as indicated above; we then get:






(n + 1)Pn+1 tn (2n + 1)xPn tn + nPn1 tn = 0.
n=1 n=0 n=1

The n = 1 term in the 1st sum does not contribute and the summation can start from n = 0; in the 3rd term we
can add the n = 0 term since it is zero anyway. Then, all 3 summations would run from n = 0 and can be combined
into one:

[(n + 1)Pn+1 x(2n + 1)Pn + nPn1 ] tn = 0.
n=0

Since it is valid for any t, then

(n + 1)Pn+1 + nPn1 = (2n + 1)xPn , n 1. (9.4)

Note that this recurrence relation is formally valid for n = 0 as well if we postulate that P1 = 0, in which case we
simply get P1 (x) = xP0 (x).
This recurrent relation can be used to generate the functions Pn (x). To show how this can be done, let us rst
use Eq. (9.1) to obtain the rst two functions by expanding G(x, t) explicitly into the power series:

n [ n ]
1 t G(x, t) G 1 2 G
G(x, t) = = = G(x, 0) + t+ 2
t2 + . . . (9.5)
1 2xt + t2 n=0
n! t n
t=0 t t=0 2 t t=0

G(x, 0) = 1.

G xt
x
= = =x
(1 2xt + t2 ) t=0
t t=0 3/2 (1)3/2

2G 1 3 (x t)(2t 2x)
=
t2 (1 2xt + t )
2 3/2 2 (1 2xt + t2 )5/2

2 G 3
2 = 1 x(2x) = 3x2 1
t t=0 2
Therefore, explicitly from (9.5):
3x2 1 2
G(x, t) = |{z}
1 + |{z}
x t+ t + ...
2 }
| {z
P0 (x) P1 (x)
P2 (x)

We conclude:
P0 (x) = 1 (9.6)
P1 (x) = x (9.7)
1( 2 )
P2 (x) = 3x 1 (9.8)
2
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 103

This can be continued, however, the calculation becomes increasingly cumbersome. On the other hand, already
P2 (x) can be generated directly from Eq. (9.4) with much less eort once P0 is known. Indeed, since P0 = 1 as
given by direct calculation above, then P1 = xP0 = x in accordance with ( (9.7). )Then, using n = 1 in the recurrence
relation, we get 2P2 + P0 = 3xP1 which gives P2 = 12 (3xP1 P0 ) = 12 3x2 1 , i.e. the same expression as above.
All higher order functions (with large values of n) are obtained in exactly the same way, i.e. using a very simple
algebra.
It can easily be seen that functions Pn (x) are polynomials of the order n.

9.3 Other recurrence relations


A number of other recurrence relations relating the function Pn with dierent values of n can also be established.
Before, we dierentiated G with respect to t. This time, we shall dierentiate it with respect to x. On the once
hand,

G
= Pn (x)tn ,
x n=0

while on the other,



G t t t
= = G(x, t) = Pn tn
x (1 2xt + t2 )
3/2 1 2xt + t2 1 2xt + t2 n=0

The two expressions must be equal, so that



( )
1 2xt + t2 Pn tn = Pn tn+1
n=0 n=0

or





Pn tn 2x Pn tn+1 + Pn tn+2 = Pn tn+1
n=0 n=0 n=0 n=0
| {z } | {z }
nn+1 n+1n

which transforms into:







Pn+1 tn+1 (2xPn + Pn ) tn+1 +
Pn1 tn+1 = 0.
n=1 n=0 n=1

In the 1st sum the n = 1 term does not contribute since P0


= 0. Separating then out the n = 0 terms in the rst
and second sums and collecting other terms together, we obtain:


[ ] n+1
P1 t 2x P0 +P0 t + Pn+1 2xPn Pn + Pn1

t =0
|{z}
n=1
=0
| {z }
=0 since P1 =1.

Thus,

Pn+1 2xPn + Pn1

= Pn n 1, (9.9)
which is the desired relation. Note that it contains the Legendre polynomials with three consequitive indices. Other
identities can be also obtained via additional dierentiations as described explicitly below. In doing this, we shall
try to aim at obtaining such identities which relate only Legendre polynomials with two consequitive indices.
To this end, we rst dierentiate (9.4) with respect to x and multiply by 2:

2(n + 1)Pn+1 + 2nPn1 = (2n + 1)2xPn + 2(2n + 1)Pn . (9.10)
Solve (9.10) with respect to 2xPn and substitute into (9.9):

(2n + 1)Pn = Pn+1 Pn1 (9.11)
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 104


Solving (9.11) for Pn+1 and using in (9.9) gives:

Pn1 = xPn nPn (9.12)

Solving (9.11) for Pn1 and using in (9.9) gives:

Pn+1 = (n + 1)Pn + xPn (9.13)

which, when written for n + 1 n, yields also

Pn = nPn1 + xPn1

. (9.14)

Now we have both Pn1 and Pn+1 expressed via Pn and Pn by means of Eqs. (9.12) and (9.13), respectively.
This should allow us to formulate a dierential equation for the polynomials Pn (x).

9.4 Dierential equation for functions Pn (x)


From (9.12)

Pn1 = xPn nPn (9.15)
which, after dierentiation, becomes

Pn1 = xPn + Pn nPn (9.16)
Dierentiate (9.14):
Pn = nPn1

+ xPn1
+ Pn1
= (n + 1)Pn1
+ xPn1 (9.17)

Substituting and
Pn1 from (9.15) and (9.16), respectively, into (9.17), gives an equation containing functions
Pn1
Pn with the same index n:
Pn = (xPn nPn ) (n + 1) + x (xPn + Pn nPn )
or ( )
1 x2 Pn 2xPn (x) + n(n + 1)Pn (x) = 0 (9.18)
which is called Legendres Equation and functions Pn (x) - Legendre polynomials.

9.5 Special values


At x = 1 and t < 1:


1 1 1
G(1, t) = = = = tn
1 2xt + t2 (1 t)2 1 t n=0
On the other hand,


G(1, t) = Pn (1)tn ,
n=0

so it follows then, that


Pn (1) = 1 (9.19)
for any n.
Similarly, at x = 1


1
G(1, t) = = (1) t
n n
Pn (1)tn ,
1 + t n=0 n=0

so that
Pn (1) = (1)n (9.20)
In fact, since
1 1
G(x, t) = = = G(x, t)
1 + 2xt + t 2 1 2x(t) + (t)2
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 105

it follows that



Pn (x)t n
Pn (x)(1)n tn ,
n=0 n=0

and, therefore,
Pn (x) = (1)n Pn (x) (9.21)
In particular, Pn (0) = (1) Pn (0) P(n (0) = 0 for
n
) odd n. In other words, the polynomials Pn (x) for odd n do not
contain constant terms, e.g. P3 (x) = 12 5x3 3x .

9.6 Orthogonality and normalisation


First, we rewrite Eq. (9.18) as follows:
d [( ) ]
1 x2 Pn + n(n + 1)Pn = 0 (9.22)
dx
Multiplying on Pm (x) and integrating between -1 and 1 gives:
1 1
d [( ) ]
Pm 1 x2 Pn dx +n(n + 1) Pn Pm dx = 0
1 |{z} |dx {z } 1
u
dv

The 1st integral is taken by parts as shown:


1 1
( ) 1 ( )
Pm (x) 1 x2 Pn (x) 1
1 x2 Pn (x)Pm (x)dx + n(n + 1) Pn (x)Pm (x)dx = 0
| {z } 1 1
=0

or
1 1 ( )
n(n + 1) Pn (x)Pm (x)dx =
1 x2 Pn Pm dx (9.23)
1 1

Alternatively, if we started from Eq. (9.22) for Pm (x) instead and multiplied it by Pn (x), we would obtain the same
result in which n and m interchange:
1 1
( )
m(m + 1) Pm (x)Pn (x)dx = 1 x2 P m Pn dx
1 1

Both integrals are symmetrical with respect to the m n permutation and the right hand sides are identical.
Therefore, after subtracting one equation from another yields:
1
[n(n + 1) m(m + 1)] Pn (x)Pm (x)dx = 0
1

If n = m, then 1
Pn (x)Pm (x)dx = 0, n = m (9.24)
1

It is said, that Pn (x) and Pm (x) are orthogonal.


Consider now the special case of equal n and m by calculating
1 1
1
2 n+m 2n
G (x, t)dx = t Pn (x)Pm (x)dx = t Pn2 (x)dx (9.25)
1 n=0 m=0 1 n=0 1

where use has been made of the already established orthogonality of the Legendre polynomials. On the other hand,
1 1
dx 1 1 dx
2
G (x, t)dx = =
1 1 2xt + t 2t 1 x t22t
2 +1
1

1
1 t2 + 1 1 1 t 2t+1
1 2t t2 1 1 t2 2t + 1
2

= ln x = ln = ln = ln
2t 2t 1 2t 1 t22t +1 2t 2t t2 1 2t t2 + 2t + 1
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 106

1 (t 1)2 1 1 t 1
= ln = ln = [ln(1 t) ln(1 + t)]
2t (t + 1)2 t 1 + t t
But, using the Taylor expansion for the logarithm:

tk
ln(1 + t) = (1)k+1 ,
k
k=1


tk tk
(t)k
ln(1 t) = + (1)k+1 = (1)k (1)k+1 = ,
k k k
k=1 k=1 k=1

so that
k
t [ ] [ ] tk t2n+1
ln(1 t) ln(1 + t) = 1 (1)k+1 = 1 + (1)k+1 = 2
k=1
k
k=1
| {z } k n=0
2n + 1
k is odd, k=2n+1

and thus 1
1 2
G2 (x, t)dx = [ln(1 t) ln(1 + t)] = t2n
1 t n=0
2n + 1

Comparing this with Eq. (9.25), we conclude that


1
2
Pn2 (x)dx = (9.26)
1 2n + 1

and thus, generally: 1


2
Pn (x)Pm (x)dx = mn (9.27)
1 2n + 1
where nm is the Kroneker delta symbol.

9.7 Expansion of functions in Legendre polynomials


Pn (x) are orthogonal, but not normalised to unity. Introduce

2n + 1
n (x) = Pn (x), n 0 (9.28)
2
The new functions form an orthogonal set:
1
n |m = n (x)m (x)dx = nm (9.29)
1

as follows from (9.27).


Any function f (x), 1 < x < 1, can be expanded in n (x):


f (x) = Cn n (x) (9.30)
n=0

The coecients Cn are obtained from (9.29) using the same method as we used when deriving coecients of the
Fourier series: multiply (9.30) by m (x) and integrate from 1 and 1:
1
1
f (x)m (x)dx = Cn n (x)m (x)dx = Cn nm = Cm
1 1
n=0 | {z } n=0
nm

i.e. 1
Cm = f (x)m (x)dx (9.31)
1
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 107

Thus, f (x) can be expanded in Pn (x) as




f (x) = an Pn (x) (9.32)
n=0

with 1
2n + 1
an = f (x)Pn (x)dx (9.33)
2 1

Example 1: Expansion of the Dirac delta function in Legendre polynomials:




(x) = an Pn (x)
n=0
1
2n + 1 2n + 1
an = (x)Pn (x)dx = Pn (0)
2 1 2

Example 2: Calculate Pn (0) using the generating function:



1 (2n)! 2n
G(0, t) = = (1)n 2 22n
t which should also to be equal to = Pn (0)tn
1 + t2 n=0
(n!) n=0

From Eq. (9.21) it follows that Pn (0) = (1)n Pn (0), i.e. Pn (0) = 0 for odd values of n. Therefore,



Pn (0)tn can be rewritten as P2n (0)t2n
n=0 n=0

and thus
(2n)!
P2n (0) = (1)n , P2n+1 (0) = 0
22n (n!)2

9.8 Rodrigues formula


We shall prove that
1 dn ( 2 )n
Pn (x) = x 1 (9.34)
2n n! dxn
( )n
Proof: Let (x) = x2 1 . Then
d ( )n1
= n x2 1 2x
dx
( 2 ) d
x 1 = 2xn(x) (9.35)
dx
Use now Leibnitz formula (Handout 1) to dierentiate (9.35) n + 1 times:
[ ] ( n + 1 )(
dn+1 ( 2 ) d dn+1 [( 2 ) (1) ] n+1 )(k) (nk+2)
LHS = x 1 = x 1 = x2 1
dx n+1 dx dx n+1 k
k=0

Only k = 0, 1, 2 terms are left since


( 2 )(0) ( )(1) ( )(2)
x 1 = x2 1, x2 1 = 2x, x2 1 = 2 and
( )(n)
x2 1 = 0 for any n 3
Thus, ( ) ( ) ( )
n+1 ( ) n+1 n+1
LHS = x2 1 (n+2) + 2x(n+1) + 2(n)
0 1 2
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 108

( ) n(n + 1) (n)
= x2 1 (n+2) + (n + 1)2x(n+1) + 2 (9.36)
2
since ( )
n+1 (n + 1)!
= =1
0 0!(n + 1)!
( )
n+1 (n + 1)! (n + 1)!
= = =n+1
1 1!(n + 1 1)! n!
( )
n+1 (n + 1)! (n + 1)! n(n + 1)
= = =
2 2!(n + 1 2)! 2(n 1)! 2
On the other hand, the RHS of (9.35) is

n+1 ( )
dn+1 n+1
2n (x) = 2n x(k) (n+1k)
dx n+1 k
k=0
( ) ( )
n+1 n+1
= 2n x(n+1) + 2n (n) = 2nx(n+1) + 2n(n + 1)(n)
0 1
Since RHS = LHS, we obtain:
( 2 )
x 1 (n+2) + 2(n + 1)x(n+1) + n(n + 1)(n) == 2nx(n+1) + 2n(n + 1)(n) ,
( )
x2 1 (n+2) + 2x(n+1) n(n + 1)(n) = 0
( )
1 x2 (n+2) 2x(n+1) + n(n + 1)(n) = 0
which is the Legendre equation (9.18) for U (x) = (n) .
1
The Rodrigues formula has been proven up to a constant factor. To verify that the factor is 2n n! , we shall
calculate the Pn (1) = 1 and compare it with U (1):

dn ( 2 )n dn n n
U (x) = (n) = x 1 = [(x + 1) (x 1) ]
dxn dxn
Use Leibnitz formula again:
n (
)
n n (k) n (nk)
U (x) = [(x + 1) ] [(x 1) ]
k
k=0

But:
n (k) dk n n!
[(x + 1) ] = (x + 1) = n(n 1)...(n k + 1)(x + 1)nk = (x + 1)nk
dxk (n k)!
n (k) n!
[(x 1) ] = (x 1)nk
(n k)!
so that, if we replace k n k:
n (nk) n!
[(x 1) ] = (x 1)k
k!
and we obtain:
n (
)
n n! n!
U (x) = (x + 1)nk (x 1)k
k (n k)! k!
k=0

At x = 1 only one term with k = 0 is left:


( )
n n! n!
U (1) = (x + 1)n = n!2n
0 n! 0! x=1

which proves the normalisation factor in Eq. (9.34).


CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 109

9.9 Other orthogonal polynomials


There are many other functions-polynomials dened in a similar way. Some of them are given in the Table below.
There, polynomials n (x) are dened on the interval (a, b) such that
b
n (x)m (x)W (x)dx = Nn nm
a

where W (x) is called the weighting function. All polynomials can be dened via a specic generating function


G(x, t) = n (x)tn
n=0

as shown with the only exception of the Hermite polynomials in which case the relationship with the generating
function is slightly dierent:

2 tn
e2xtt = Hn
n=0
n!

Polynomial (a, b) Weight Nn G(x, t)


( )1/2
Legendre Pn (x) 1 x 1 1 2
1 2xt + t2
2n+1

ex
2 2
Hermite Hn (x) - < x < n
{ 2 n! e2xtt
Chebyshev Tn (x) 1 x 1 1 2, n > 0 1tx
1x2 , n = 0 12xt+t2
exp( 1t
xt
)
Laguerre Ln (x) 0x< ex 1 1t

Handout. Leibnitz theorem


We would like to obtain a simple formula for dierentiating a product of two functions arbitrary number of times.
Let f (x) = u(x)v(x). Then
( ) ( ) 2 n2 ( ) k nk
dn f dn (uv) dn v n du dn1 v n d ud v n d ud v dn u
= = u + + + ... + + ... + v
dxn dxn dxn 1 dx dxn1 2 dx2 dxn2 k dxk dxnk dxn
n (
) k nk n (
)
n d ud v n
u(k) v (nk) (9.37)
k dxk dxnk k
k=0 k=0
( )
n n!
where = k!(nk)! are the binomial coecients. In fact, the Leibnitz formula looks extremely similar to the
k
well-known binomial theorem
n ( )
n
(a + b)n = ak bnk (9.38)
k
k=0

and, therefore, is sometimes written in the following symbolic form:

dn (uv)
= (u + v)(n) (9.39)
dxn
where the superscript (n) means dierentiate with respect to x n times.

Proof:
d
(uv) = u(1) v + uv (1) = u(1) v (0) + u(0) v (1) (i.e. familiar u v + uv )
dx
d2
(uv) = u(2) v (0) + 2u(1) v (1) + u(0) v (2)
dx2
d3 ( ) ( ) ( )
(3) (0) (2) (1) (2) (1) (1) (2) (1) (2) (0) (3)
(uv) = u v + u v + 2 u v + u v + u v + u v
dx3
CHAPTER 9. LEGENDRE POLYNOMIALS PN (X) 110

= u(3) v (0) + 3u(2) v (1) + 3u(1) v (2) + u(0) v (3)


which are all consistent with Eqs. (9.37) or (9.39), since:
( )
3 3!
= =1
0 0!3!
( ) ( )
3 3! 3 3!
= = 3, = =3
1 1!2! 2 2!1!
( )
3 3!
= =1
3 0!3!
The general proof is based on the method of mathematical induction. We assume that
( )
dn (uv)
n
n
= u(k) v (nk) (9.40)
dxn k
k=0

is valid for some n. We should now prove from it that Eq. (9.37) holds also for the next value of n, i.e. for n + 1.
Dierentiating both sides of Eq. (9.40), we get:
[ n ( ) ] n ( )
dn+1 (uv) d n (k) (nk)
n [ (k+1) (nk) (k) (nk+1)
]
= u v = u v + u v
dxn+1 dx k k
k=0 k=0

In the 1st term we make a substitution of the summation index, k + 1 k, which gives:
n+1 ( ) n ( )
dn+1 (uv) n (k) (nk+1)
n
= u v + u(k) v (nk+1)
dxn+1 k1 k
k=1 k=0

Then, separate out the k = n + 1 term in the rst sum and the k = 0 term in the second; combine the others
together:
( ) n [(
) ( )] ( )
dn+1 (uv) n n n n
= u(n+1) v (0) + + u(k) v (nk+1) + u(0) v (n+1) (9.41)
dxn+1 n k1 k 0
k=1

Since ( ) ( )
n n+1
=1
n n+1
( ) ( )
n n+1
=1
0 0
( ) ( ) [ ]
n n n! n! n! 1 1
+ = + = +
k1 (k 1)!(n k + 1)! k!(n k)!
k (k 1)!(n k)! n k + 1 k
( )
n! k+nk+1 n! n+1 (n + 1)! n+1
= = = ,
(k 1)!(n k)! (n k + 1)k (k 1)!(n k)! (n k + 1)k k!(n k + 1)! k
we nally obtain from Eq. (9.41):
( ) n (
) ( ) ( n+1 )
n+1
dn+1 (uv) n+1 n+1 n+1
= u (0) (n+1)
v + (k) (n+1k)
u v + u v
(n+1) (0)
u(k) v (n+1k)
dxn+1 0 k n+1 k
k=1 k=0

which is the desired result. Indeed, we know that Eq. (9.37) holds for n = 1. Then, we also know that it holds for
the next value of n, which is n = 2. As it holds for the n = 2, it should also hold for the next one, n = 3, and so
on we can continue this argument indenitely to arrive at the nal conclusion that it is valid for any integer value
of n.
Chapter 10

Laplace Equation in Spherical Polar


Coordinates

10.1 Separation of Variables


The Laplace equation 2 = 0 in spherical polar coordinates (r, , ) can be written as:
( ) ( )
1 2 1 1 2
r + sin + = 0. (10.1)
r2 r r r2 sin r2 sin2 2

We attempt to separate the variables in (10.1) by making the substitution

(r, , ) = R(r)()(). (10.2)

After multiplying through the resulting equation by r2 sin2 (R)1 we nd that


( ) ( )
2 1 d 2 dR sin d d 1 d2
sin r + sin = . (10.3)
R dr dr d d d2

The LHS of (10.3) depends only on (r, ), while the RHS depends only on . For this to be possible both sides
must be equal to the same constant V . Hence, we have two equations:

d2
= V , (10.4)
d2
and ( ) ( )
1 d dR 1 d d V
r2 = sin + . (10.5)
R dr dr sin d d sin2
The LHS of (10.5) depends only on r, while the RHS depends only on . It follows, therefore, that bost sides must
be equal to the same constant U . Hence, we obtain two nal equations:
( )
d 2 dR
r U R = 0, (10.6)
dr dr
( ) [ ]
1 d d V
sin + U = 0. (10.7)
sin d d sin2
The method of separation of variables has succeeded, and we now have to solve three ordinary dierential
equations. Note that we have two separation constants, U and V .

10.2 Solution of the equation


We consider Eq. (10.4) for dierent values of V :

111
CHAPTER 10. LAPLACE EQUATION IN SPHERICAL POLAR COORDINATES 112

(a) When V > 0, we can write V = m2 . For this case the solution of (10.4) is

() = A sin(m) + B cos(m). (10.8)

(b) When V = 0, we nd that


() = A + B. (10.9)
(c) When V < 0, we can write V = p2 . For this case we have

() = A sinh(p) + B cosh(p). (10.10)

If we increase by 2 we come back to the same point in real space, i. e. (r, , ) (r, , + 2). In most
problems is a single-valued function of position, and we require, therefore, that

(r, , + 2) = (r, , ), (10.11)

and hence
( + 2) = (). (10.12)
This result implies that () must be a periodic function with a period of 2.
It is readily seen that this is only possible in the case (a), i.e. we must have the eigenvalues

V = m2 , (m = 0, 1, 2, ...) (10.13)

and corresponding eigenfunctions


() = A sin(m) + B cos(m). (10.14)
For quantum-mechanical applications it is usually more convenient to use the complex eigenfunctions:

() = A+ eim + B eim . (10.15)

10.3 Solution of equation


The substitution of Eq. (10.13) in (10.7) gives:
( )
d d [ ]
sin sin + U sin2 m2 = 0. (10.16)
d d

In order to simplify this equation we apply the transformation

= cos , 1 < 1. (10.17)

For any function f () we can write


df df d df
= = sin . (10.18)
d d d d
Hence we obtain
df ( ) df
sin = 1 2 . (10.19)
d d
and ( ) ( ) ( )
d d d ( ) d ( ) d ( ) d
sin sin = sin 1 2 = 1 2 1 2 . (10.20)
d d d d d d
The substitution of (10.20) in Eq.(10.16) gives:
[ ]
( ) d2 d m2
1 2 2 + U = 0. (10.21)
d2 d (1 2 )

This dierential equation has regular singular points at = +1 and = 1.


CHAPTER 10. LAPLACE EQUATION IN SPHERICAL POLAR COORDINATES 113

10.4 Legendre equation (m = 0)


For the special case of m = 0, Eq. (10.21) reduces to the Legendre equation:

( ) d2 d
1 2 2
2 + U = 0. (10.22)
d d
The origin = 0 is an ordinary point of this dierential equation and can be chosen as the point around which the
solution of the above equation ia obtained using the Frobenius method (Chapter 7). Since we are only interested
in the region of 1 1, this is the most convenient choice. Thus, we can write


() = cr r , || < 1. (10.23)
r=0

and then try to solve Eq. (10.22) using the series method. It is found then that the coecients cr satisfy the
recurrence relation (see also Chapter 7)

[r(r + 1) U ]
cr+2 = cr , (r 0). (10.24)
(r + 1)(r + 2)

It is seen that two sequences of coecients are produced in this way: (i) starting from c0 , we generate coecients
cr with even indices r, all proprotional to the c0 ; (ii) starting from c1 , we generate coecients cr with odd r, all
proportional to the c1 . The rst series will therefore contain only even powers of , while the other series - only
odd powers. Hence, the general solution of (10.22) is:


2k k1

2k+1 k1

() = c0 1 + {2j(2j + 1) U } + c1 1 + {(2j + 1)(2j + 2) U } , || < 1.
(2k)! j=0 (2k + 1)! j=0
k=1 k=1
(10.25)
where c0 and c1 are arbitrary constants.
In physical problems we require that (r, , ) is nite along the z-axis ( = 0 and = ). We see, therefore,
that the solution (10.25) must yield a nite value when = +1 and = 1. It is readily seen, using the ratio test,
that the two series in (10.25) are absolutely convergent for || < 1. However, a more sensitive test for convergence
shows that, for general values of U , both series solutions in (10.25) are logarithmically divergent at = 1!
Physically acceptable solutions are only possible if the separation constant U takes on either of the following
special values:
U = n(n + 1) (10.26)
with any value of n = 0, 1, 2, 3, . . .. We cannot prove this here, however, it can readily be seen that for this particular
choice of the constant U one of the two series becomes a nite polynomial and thus is well dened at = 1.
Indeed, let us choose some even value of n; then the coecients cr+2 for r < n will all be nonzero; however, for
r = n the coecient cn+2 contains precisely the dierence n(n + 1) U which is zero. Therefore, cn+2 = 0 and,
consequently, all the following coecients cr with even values of r > n + 2 will all be equal to zero. Under the
circumstances, the rst series in (10.25) terminates and becomes a polynomial. The other series for the same value
of n diverges for = 1 and should thus we rejected.
Similarly, if n is odd, the second series becomes a polynomial of degree n, and the rst innite series must be
rejected. Thus, for each value of U = n(n + 1), n = 0, 1, 2, ... we have only one physically acceptable solution of the
Legendre equation. These will be nothing but Legendre polynomials Pn of the previous Chapter (with the proper
choice of the constants c0 or c1 to ensure the correct prefactor).
From the series (10.25) with U = n(n + 1) we nd that:
1( 2 ) 1( 3 )
P0 () = 1, P1 () = , P2 () = 3 1 , P3 () = 5 3 ,
2 2
1( ) 1( )
P4 () = 354 302 + 3 , P5 () = 635 703 + 15 .
8 8
CHAPTER 10. LAPLACE EQUATION IN SPHERICAL POLAR COORDINATES 114

10.5 Associated Legendre equation


The substitution U = n(n + 1) in (10.21) gives the associated Legendre equation
[ ]
( ) d2 d m2
1 2 2 + n(n + 1) = 0. (10.27)
d2 d 1 2
This equation can be solved using series methods. However, there is an easier way! We make the substitution:
( )m
() = 1 2 2 () (10.28)

in (10.27), which gives (see handout 1):


( ) d2 d
1 2 2(m + 1) + (n m)(n + m + 1) = 0. (10.29)
d2 d
On the other hand, consider the Legendre polynomial
1 dn ( 2 )n
Pn () = 1 (10.30)
2n n! d n

which satises the dierential equation


( ) d2 Pn () dPn ()
1 2 2 + n(n + 1)Pn () = 0. (10.31)
d2 d
Let us dierentiate this equation m times using the Leibnitz formula. We have 3 terms:
[ ] m ( )
dm ( ) 2
2 d Pn m ( )
2 (k) d
mk+2
Pn
1st = 1 = 1
d m d 2 k d mk+2
k=0
( ) ( ) ( )
m ( ) m+2 Pn
2 d m dm+1 Pn m dm Pn
= 1 + (2) + (2)
0 dm+2 1 d m+1 2 dm
( ) dm+2 Pn dm+1 Pn dm Pn
= 1 2 m+2
2m m+1
m(m 1) ; (10.32)
d d dm

[ ] ( ) ( )
dm dPn m dm+1 Pn m dm Pn dm+1 Pn d m Pn
2nd = (2) = (2) + (2) = 2 2m ; (10.33)
dm d 0 dm+1 1 dm dm+1 dm

dm Pn
3rd = n(n + 1)
dm
So, our dierential equation (10.31) after m dierentiations becomes (the sum of all three contributions)
( ) dm+2 Pn dm+1 Pn dm Pn
1 2 2(m + 1) + [n(n + 1) m(m + 1)] = 0. (10.34)
dm+2 dm+1 dm
(10.34) should be compared with (10.29), and one can see that

n(n + 1) m(m + 1) = (n m)(n + m + 1),

so that
dm
() Pn ().
dm
In other words,
( ) m dm ( ) m dm+n ( 2 )n
() 1 2 2 m
Pn 1 2 2 m+n
1 .
d d
The proportionality constant is usually taken such that the solution of the associated Legendre equation is:
(1)m ( ) m dn+m ( 2 )n
() = n
1 2 2 n+m
1 (10.35)
2 n! d
CHAPTER 10. LAPLACE EQUATION IN SPHERICAL POLAR COORDINATES 115

( ) m dm
(1)m 1 2 2 Pn () Pnm (). (10.36)
dm
Since Pn is a polynomial of the order n, it is clear that 0 m n; otherwise, if m > n,
dm
Pn = 0.
dm
The functions Pnm () are called associated Legendre functions.
NOTE: Pnm () is not the only solution of the dierential equation (10.27). However, other solutions become
innite as 1 and thus are not physically acceptable.
It can be shown that the associated Legendre functions satisfy the following orthogonality condition:
1
Pnm ()Pnm ()d = 0, n = n
1

10.6 Spherical harmonics


The functions Ynm (, ) (also frequently written as Ynm (, )) dened by

m m 2n + 1 (n |m|)! |m|
Yn (, ) = (1) P (cos )eim , (10.37)
4 (n + |m|)! n

where n m n and n = 0, 1, 2..., are known as spherical harmonics. They represent the complete solution of
the angular part of the Laplace equation in spherical polar coordinates, and are normalised so that
2

(Ynm (, )) Yjk (, ) sin dd = nj mk . (10.38)
0 0

Here the integration is performed over the solid angle d = sin dd (which integrates to 4 over the whole sphere).
Note also that only a single exponential is chosen above so that two dierent functions can be dened for negative,
m = |m|, and positive, m = |m|, values of m.
The rst few spherical harmonics are:
0 1
Y0 (, ) =
4

0 3
Y1 (, ) = cos
4

3 1 3
Y1 (, ) =
1
sin e , Y1 (, ) =
i
sin ei
8 8

0 5
Y2 (, ) = P2 (cos )
4

5 1 5
1
Y2 (, ) = i
3 sin cos e , Y2 (, ) = 3 sin cos ei
24 24

5 2 5
2
Y2 (, ) = 2 2i
3 sin e , Y2 (, ) = 3 sin2 e2i
96 96
Real spherical harmonics are obtained my mixing (for m = 0) Ynm and Ynm . Note that Yn0 is already real!
Spherical harmonics Ynm are useful in expressing angular dependence of functions of vectors as an expansion
over them: any function f (, ) can be expanded as

n
f (, ) = fnm Ynm (, ) (10.39)
n=0 m=n

where (using orthogonality of the spherical functions, Eq. (10.38))


2

fnm = (Ynm (, )) f (, ) sin dd (10.40)
0 0
CHAPTER 10. LAPLACE EQUATION IN SPHERICAL POLAR COORDINATES 116

One important results is, for instance, that if there are two unit vectors n1 and n2 with the angle between them,
then one can show:
4 n
Pn (cos ) = Y m (1 , 1 )Ynm (2 , 2 )
2n + 1 m=n n

where and angles (1 , 1 ) and (2 , 2 ) correspond to the orientation of the rst and the second vectors.
Spherical functions are also used in quantum chemistry to express angluar dependence of atomic orbitals centreed
on atoms of molecules and crystals when solving the Shrdinger equation.

Handout. Proof of (10.29) of the main lecture text


Denote
m
X() = 1 2 , D = X , =
2
( )m
() = 1 2 2 () = X = D
Using the Leibnitz formula:
= D

= D + D = X 1
(2) + X = 2X 1 + X ;
= D + 2D + D
D = X 1 (2)
D = 2X 1 + ( 1)X 2 (2)2
= 2X 1 + 4( 1)X 2 2
Therefore, the rst term in (10.27) is:
( ) [ ]
= 1 2 = X = 2X + 4( 1)X 1 2 + 2 X (2) + X +1
[( ) ]
= X 2 + 42 ( 1)X 1 4 + X (10.41)
The second term in (10.27) is:
( ) [ ]
= 2 = 2 2X 1 + X = X 42 X 1 2 (10.42)

The third term in (10.27):


( )
m2 [ ]
= n(n + 1) = n(n + 1) X 1 m2 X (10.43)
1 2

Combining all terms together, we obtain:


{ [ ] }
X X (4 + 2) + 42 X 1 2 + 42 ( 1)X 1 + n(n + 1) m2 X 1 = 0

or, omitting X which is a common factor, and simplifying:


[ ( ) ]
X 2(2 + 1) + n(n + 1) 2 + 42 2 m2 X 1 = 0

Recall now that, in fact, = m/2. Then:


2(2 + 1) = 2(m + 1),
( ) ( )
n(n + 1) 2 + 42 2 m2 X 1 = n(n + 1) m + m2 2 1 X 1
= n(n + 1) m m2 = n(n + 1) (m + 1)m = (n m)(n + m + 1),
i.e. we obtain exactly Eq. (10.29).
Part II

Problems

117
Chapter 11

Problem sheet 1: (x), H(x) and (x)

11.1 Problems to be done in the class: (x) and H(x) functions


The Dirac delta function (x) is a special generalised function that satises the following ltering theorem:
b {
f (x0 ), if a < x0 < b
f (x)(x x0 )dx =
a 0 otherwise
The Heaviside function H(x) is dened in the following way:
{
1, if x 0
H(x) =
0, if x < 0
1. Calculate the integral: 20
(5x + 6)e10(x+1) dx
2

10

2. Calculate the integral:


(5x + 6)e10(x+1) dx
2

3. By considering the integral


f (x)H (x)dx

prove that the derivative of the Heaviside function, H (x), is equal to the Dirac delta function (x). [Hint:
use integration by parts.]
4. Calculate the following integral:

e2x [(2x + 1) + 5H(x + 1)] dx

5. Let f (x) be some continous function. What eect does the Heaviside function H(x) have on it in the following
combinations (assume that a > 0):

H(x)f (x)
H(x a)f (x)
H(x a)f (x a)

Sketch your ndings using some arbitrary function f (x) (e.g. an exponential function).
6. Express the function {
0, if |x| > 1
f (x) =
1, if |x| 1
via the Heaviside function H(x). [Hint: using H(x) with a shifted argument, create two functions which step
at 0 and 1, respectively; then, combine the two.]

118
CHAPTER 11. PROBLEM SHEET 1: (X), H(X) AND (X) 119

11.2 Problems to be done in the class: (x) function


The Gamma function satises the recurrence relation (z + 1) = z(z) is dened as an integral:

(z) = tz1 et dt
0

The following Gaussian integrals are expressed via the amma function:
( )
s at2 (s+1)/2 s+1
t e dt = a
2
where s is even; the integral is equal to zero for odd s.

1. Evaluate (3/2) and (5/2) using the recurrence relation for the -function and the identity (1/2) = .
Then prove for any n = 2, 3, ... ( )
1 1 3 5 ... (2n 1)
n+ =
2 2n
2. Express the integral
t2 et dt
2


via the -function and then calculate it using the results of the previous problem.
3. Calculate the following integral ( ) 2
1 + 2x2 e2xx dx

11.3 Homework
Please, hand in your solutions next Thursday by 11 am. Late papers will be marked down.
1. Using a sequence of functions
{
n, when 2n
1
t 1
2n
n (t) = , n = 1, 2, 3, ...,
0, when |t| > 2n
1

explain the signicance of the general ltering theorem for the Dirac delta function (t). [10 points]
2. Prove the integral identity
f (x) (x)dx = f (0),

which can be used as a dention of the impulse function (x). [Hint: use the integration by parts.] [10 points]
3. Prove the following formal identity: x(x) = 0. (Hint: multiply this by some f (x) and then integrate around
x = 0 using the ltering theorem). [5 points]
4. Calculate the following integral:

I= e3t [3(3t 2) + 3H(t 2)] dt

2 6
(Answer : I = e +e ) [10 points]
5. Express the sign function {
1, if x > 0
sgn(x) =
1, if x < 0
in terms of the Heaviside function. [5 points]
6. Calculate the integral
t4 et dt
2
I=


3
by expressing it rst via the - function. (Answer : I = 45/2
) [10 points]
CHAPTER 11. PROBLEM SHEET 1: (X), H(X) AND (X) 120

f(t)

... ...

t
4 2 0 2 4
Figure 11.1: A signal represented as a innite sequence of unit height steps.

11.4 This week CHALLENGE


The solutions, written separately, should be handed in together with the home work. Please, put them in the correct
pile!!!

1. Express the function {


x2 , if 0 < x < 3
f (x) =
5x, if x > 3
via the Heaviside function H(x). [5 points]
2. Prove the following formal identity:
( ) 1
x2 a2 = [(x + a) + (x a)]
2a
assuming that a > 0. [10 points]

3. Generalise the method of the previous problem and derive a formal representation for the delta function of
an arbitrary (but otherwise well-behaved) function f (x), i.e. (f (x)). [10 points]
4. Express the signal shown in the Fig. 11.1 in terms of the Heaviside function. You are allowed to use the
summation sign and as many of the Heaviside functions as necessary! [15 points]
5. Calculate the 1st and 2nd derivatives of f (x) = |x|. [10 points]

The student (or students?) that scored most points will get a surprise gift. The elegance of the solutions will also
be taken into account! The maximum score is 50 points.
Chapter 12

Problem sheet 2: Fourier series

12.1 Class problems


1. Show that the sine/cosine Fourier series (FS) expansion of the function
{
1, 0 < x <
f (x) =
0, < x < 0

is

1 1 1 (1)n
fF S (x) = + sin(nx) (12.1)
2 n=1 n
What does the FS converge to at x = 0?

2. Using the FS expansion for the f (x) dened in the previous problem, obtain the FS for
{
1, 0 < x <
g(x) =
1, < x < 0

3. Use the FS for f (x) of problem 1 to sum the numerical series


1 1
S =1 + ...
3 5

4. Show that the complex (exponential) FS of the function



0, < x < 0
f (x) = 1, 0 < x < /2

0, /2 < x <

is


1 1 i sin n
2 inx
fF S (x) = + e (12.2)
4 2in
n=,n=0

5. Use x = 0 in the FS of the previous problem to obtain the sum of the numerical series of problem 3.
6. Use the Parcevals theorem applied to the series (12.1) of problem 1 to show that

1 1 2
1+ + + ... =
32 52 8

121
CHAPTER 12. PROBLEM SHEET 2: FOURIER SERIES 122

12.2 Home work


1. Consider the periodic function f (x) one period of which is specied as f (x) = x2 , < x < . [16 points]

sketch the fuction in the interval 3 < x < 3


is this function continuous everywhere?
show that the FS of this function is

2 4(1)n
fF S (x) = + cos(nx) (12.3)
3 n=1
n2

state if the FS converges everywhere to x2

2. Show that if f (x) = f (x), i.e. the function f (x) is odd, then its FS will only contain sine functions. [7 points]

3. Use the FS (12.3) of problem 1 to nd the sum of the numerical series


1 1 1
S =1 + 2 2 + ...
22 3 4
2
(Answer : S = 12 ) [4 points]
4. Show that the expansion of the function
{
sin x, 0 x <
f (x) =
0, < x < 0

into the complex (exponential) FS is




1 ( ix ) 1 (1)n+1 inx
fF S (x) = e eix + e
4i 2(1 n2 )
n=,n=1

[Hint: you should consider the cn coecients for n = 1 separately.] [15 points]
5. Applying the Parcevals theorem to the series (12.3) of problem 1, nd the sum of
1 1
S =1+ + 4 + ...
24 3
4
(Answer : S = 90 ) [6 points]
Chapter 13

Problem sheet 3: Fourier Transform

The Fourier transform (FT) F[f (t)] of the function f (t) is dened as:

F[f (t)] F () = f (t)ei2t dt and F 1 [F ()] f (t) = F ()ei2t d,

respectively. Note that the signs in the exponents may be reversed, but must be dierent in the two formulae.

13.1 Problems to be done in the class


1. Calculate the FT of the function:
1
f (t) = [(t a) + (t + a)]
2
2. Show that the FT of the function:
1, 1 < t < 0
f (t) = 1, 0 < t < 1

0, otherwise
is given by
2i sin2
F () =

3. Using the inverse FT, express the function f (t) of the previous problem as an integral from to +.

4. Using the result of the previous problem, evaluate the integral:



sin3 t cos t
dt
0 t

5. Calculate the FT of the function



(t) = f ( )f (t )d f f

where f (t) is given in problem 2.


6. Show that the FT of an odd function can be written as:

F () = 2i f (t) sin(2t)dt
0

123
CHAPTER 13. PROBLEM SHEET 3: FOURIER TRANSFORM 124

13.2 Homework
Please, hand in your solutions next Thursday by 11 am. Late papers will be marked down.
1. What conditions should a function f (t) satisfy for its FT to exist? Hence, explain why the FT does not
formally exist for the Heaviside unit step function H(x). [5 points]
2. Show that the FT of the function f (t) = e|t| (where > 0) is given by
2
F () =
2 + (2)2
[5 points]
3. Thus prove the following integral representation of it:

2 cos(ut)
e|t| = du
0 2 + u2
[10 points]
4. Using the result of the previous problem, calculate the integral

du
I=
0 1 + u2

(Answer : I = 2 ) Then check your result by calculating the integral directly using a dierent method. (Hint:
look up for the integral representation of the arctan.) [5 points]
5. Show that the FT of an even function can also be written as:

F () = 2 f (t) cos(2t)dt
0
[5 points]
6. The function f (t) is dened as et for t 0 ( > 0) and zero otherwise. Show by direct calculation of the
convolution integral, considering separately the cases of positive and negative t, that the convolution of this
function with itself g(t) = f (t) f (t) is g(t) = tf (t). [5 points]
7. Find the FTs of the functions f (t) and g(t) dened in the previous problem. (Answer : F [f (t)] = F () =
1 2
( + i2) and F [g(t)] = G() = ( + i2) ) Then state the convolution theorem and check its validity
in the particular case of the functions f (t) and g(t). [5 points]

13.3 This week CHALLENGE


The solutions, written separately, should be handed in together with the home work. Please, put them in the correct
pile!!!
1. The FT of a function f (t) is F (). Using the method of mathematical induction, derive a general expression
for the FT of the nth derivative, f (n) (t), of the function f (t). [5 points]
2. The FT of a function f (r) in the 3D space can be dened as the following 3D integral:

1
F[f (r)] = F (k) = f (r)eikr dr
(2)3/2

Prove then the following identities:

(a) F[f (r)]=-ikF (k) [5 points]


(b) F[ f (r)]=k F (k)
2 2
[5 points]
(c) F[ g(r)]=-ikG(k), where g(r) is some vector function in the 3D space and G(k) is its FT (also a
vector function) as given above. [10 points]
CHAPTER 13. PROBLEM SHEET 3: FOURIER TRANSFORM 125

3. Consider an innitely long string along the x-axis that is pulled up at x = 0 and then let go. The initial shape
of the string is described by the function f (x) = he|x|/a (h and a are some parameters). Show by using the
FT method that the solution of the equation of motion for the string (the wave equation),

1 2y 2y
2 2
=
v t x2
can be written as
ha cos(vkt) ikx
y(x, t) = e dk
1 + k 2 a2
[15 points]
The student (or students?) that scored the most points will get a surprise gift.
Chapter 14

Problem sheet 4: Laplace transform

The Laplace tranform (LT) of f (t) is dened as an integral:



L(f ) = f (t)ept dt F (p)
0

where p is a complex number. The LT of the rst and second derivatives of the function f (t) are:

L(f ) = pL(f ) f (0) and L(f ) = p2 L(f ) f (0) pf (0)

If F (p) and G(p) are the LT of the functions f (t) and g(t), respectively, then the LT of their convolution,
t
c(t) = f (t )g( )d
0

is F (p)G(p). A table of required LTs can be found in books (a copy from Boas was given to you together with the
lecture notes).

14.1 Problems on the Laplace transform


1. Calculate the LT of f (t) = eat cos(t)
2. Calculate the LT of {
1, t > a > 0
f (t) = H(t a)
0, t<a

3. Solve the following DE subject to initial conditions:

y y = 2et , y(0) = 3

Use a table of LTs when necessary.


4. Solve the following DE subject to initial conditions:

y + 9y = cos3t, y(0) = 0, y (0) = 6

Use a table of LTs whn necessary.


5. Represent the fraction
3p + 2
F (p) =
3p2 + 5p 2
as a sum of partial fractions and then calculate the inverse LT of it using a table of LTs as required.

126
CHAPTER 14. PROBLEM SHEET 4: LAPLACE TRANSFORM 127

(a)
f(t) ... (b) f(t)
4A
A
3A

2A
t
A
A
a 2a 3a 4a 5a
0 a 2a 3a 4a t

Figure 14.1: (a) Ladder function and (b) periodic rectangular impulse.

14.2 Homework
Please, hand in your solutions next Thursday by 11am. Late papers will be marked down.
1. Formulate and prove the convolution theorem for the LT. [10 points]
2. Calculate the LT of f (t) = et cos2 (t). [Hint: use appropriate trigonometric identities to simplify f (t) and
then apply required entries in the table of LTs.] (Answer : L [f (t)] = F (p) = 2(p+)
1 p+
+ 12 (p+) 2 +4 2 ) [5 points]

3. Let the LT of the function g(t) is L[g] = G(p). Calculate the LT of the following function:
{
g(t a), t > a > 0
f (t) =
0, t < a

(Answer : L [f (t)] = F (p) = epa G(p) ) [5 points]


4. Calculate the inverse LT of
1p
F (p) =
p2 + 4p + 13
[Hint: split f (t) into simple partial fractions rst and then use the table of the LTs]. (Answer : L1 [F (p)] =
e2t (sin 3t cos 3t) ) [7 points]
5. By usign the LT, nd the solution of the following dierential equation subject to the given initial conditions:

y + y 5y = e2t , y(0) = 1, y (0) = 2.

(Answer : y(t) = e2t ) [8 points]

14.3 This week CHALLENGE


The solutions, written separately, should be handed in together with the homework. Please put them in the correct
pile!
1. Show that [ t ]
1
L f ( )d = L [f ]
0 p
[5 points]
2. Hence, calculate the LT of the so-called Frenel integrals dened as follows:
t t
cos sin
C(t) = d and S(t) = d
0 2 0 2
[10 points]
3. Calculate the LT of the ladder functions shown in Fig. 1 (a). [15 points]
4. The same for the periodic rectangular impulse shown in Fig. 1 (b). [15 points]
As usual, there is a bonus for the best solutions!
Chapter 15

Problem sheet 5: Vector calculus, Part I

15.1 Problems to be done in the class


1. The parabolic cylinder coordinates (u, v, z) are dened by the relations.
1( 2 )
x= u v 2 , y = uv, z = z
2
(a) Sketch the coordinate surfaces and lines. [Hint: it may help if you rst try to write the corresponding
equations in analytical form.]
(b) Obtain scale factors and express the unit base vectors eu , ev and ez via the Cartesian unit vectors i, j
and k. Do the unit base vectors change their direction from point to point? Sketch the direction of the
unit vectors for a general point P (u, v, z).
(c) Check that the coordinate system is orthogonal.
(d) Write the corresponding transformation matrix M between the two sets of the unit base vectors and
check that the parabolic cylinder system is orthogonal. Check then that the inverse, M1 , is equal to
f
the transpose matrix M.
(e) Hence, express the Cartesian unit vectors i, j and k back via eu , ev and ez .
(f) A point particle is specied by the coordinates (u(t), v(t), z(t)) at each time moment t. Calculate the
derivatives of the unit vectors eu , ev and ez with respect to time t.
(g) Hence, calculate the velocity and the acceleration vectors of the point in the parabolic cylinder coordi-
nates.

2. Express the scalar eld


V (x, y, z) = ex + ey + ez
2 2 2

in cylindrical coordinates (r, , z) dened via relations:


x = r cos , y = r sin , z = z

3. Also rewrite the vector eld


x bi + y bj + z k
b
F(x, y, z) =
x2 + y 2 x2 + y 2
entirely in cylindrical coordinates.

15.2 Homework
1. A general curvilinear orthogonal coordinate system (q1 , q2 , q3 ) is given by he transformation functions: x =
x (q1 , q2 , q3 ), y = y (q1 , q2 , q3 )and z = z (q1 , q2 , q3 ). Show that the unit base vectors e1 , e2 and e3 and scale
factors h1 , h2 and h3 in this system are given, respectively, by the following relations:
( )
1 r r
ei =
, hi =
hi qi qi

128
CHAPTER 15. PROBLEM SHEET 5: VECTOR CALCULUS, PART I 129

bank bank

v
0 A
river
a
Figure 15.1: A schematic of a boat crossing a river.

2. The spherical coordinate system (r, , ) is dened by the following transformation relations:

x = r sin cos , y = r sin sin , z = r cos

(a) Sketch the coordinate lines and surfaces. [5 points]


(b) Obtain the scale factors hr , h , h (Answer : hr = 1, h = r, h = r sin ) and show that the unit base
vectors er , e and e are expressed via the Cartesian unit vectors i, j and k as follows:

er = sin (cos i + sin j) + cos k

e = cos (cos i + sin j) sin k


e = sin i + cos j
Sketch the direction of the unit base vectors for a general point P (r, , ). [10 points]
(c) Check if this system is orthogonal. (Answer : yes) [5 points]
(d) Hence, express i, j and k via the vectors er , e and e . [5 points]
(e) Assuming that coordinates r = r(t), = (t) and = (t) change with time t, show that the derivatives
of the unit base vectors with respect to time are:

er = e + sin e

e = er + cos e
e = (sin er + cos e )
[10 points]
CHAPTER 15. PROBLEM SHEET 5: VECTOR CALCULUS, PART I 130

(f) Show that the velocity and acceleration vectors in this system are:

v = r = rer + re + r sin e
( ) ( )
a = r = r r2 r2 sin2 er + 2r + r r2 sin cos e +
( )
+ 2r sin + r sin + 2r cos e

Note that the velocity can be found either from the expression for the ds or simply by dierentiating the
position vector r =rer with respect to time t. As usual, dots above the symbol correspond to the time
derivative. [10 points]

3. Consider classically an electron of mass m moving in a potential U (r) = 1r of the nucleus of the hydrogen
atom, where r is the distance between the two charged particles. Why the most appropriate curvilinear
coordinate system in this case is the spherical system? Show that the classical Newtons equations of motion
for the electron in the chosen coordinate system are:
1
r r2 r2 sin2 =
mr2

2r + r r2 sin cos = 0
2r sin + r sin + 2r cos = 0
[10 points]

15.3 This week CHALLENGE


The solutions, written separately, should be handed in together with the home work. Please, put them in the correct
pile!!!

1. A charge density of a point charge q that is positioned at point r0 is described by the distribution function

(r) = q (r r0 ) q (x x0 ) (y y0 ) (z z0 )

Rewrite the distribution function in the spherical coordinates (r, , ). [Hint: start from one of the delta
function sequences and then take the limit.] [15 points]

2. Show that the 3D FT


1
F (k) = f (r)eikr dr
(2)3/2
( 2
)
2 1
of the function f (r) = r + can be written as an integral

1 2 r sin(kr)
F (k) = dr
k 0 r 2 + 2
[10 points]
3. A boat crosses a river of width a with a constant speed 0 = |v|. Assume that the river ows with a constant
velocity V at any point across its width. The boat starts at a point A (see Fig.1) and along the whole
journey its velocity v is kept directed towards the point 0 on the bank exactly opposite the point A. Find the
trajectory of the boat. [Hint: use polar coordinates with the origin at point 0.] [20 points]
The student (or students?) scored most points will get a surprise gift. The elegancy of the solutions will also be
taken into account! The maximum scoreis 40 points.
Chapter 16

Problem sheet 6: Vector calculus, Part II

The expressions for the gradient, divergence and curl in a general curvilinear coordinate system were given in the
lectures. Please use them here!

16.1 Problems to be done in the class


The spherical coordinate system (r, , ) are dened by the following transformation relations:

x = r sin cos y = r sin sin , z = r cos

The scale factors are: hr = 1, h = r h = r sin .


1. Write in the spherical coordinate system, where = (r, , ) is a scalar eld.

2. Using the spherical coordinates, calculate for


( )
(x, y, z) = 10 exp x2 y 2 z 2
(
)
at the point P r = 1, = 2, = 2 .
3. Check your result by calculating the gradient of the same function in Cartesian coordinates.

4. Derive the Laplacian 2 of a scalar eld = (r, , ) in the spherical coordinates.

5. Calculate 2 for = er cos .


2

6. Show that the volume element in the spherical coordinate system is

dV = r2 dr sin dd

Hence, calculate the volume of a sphere of radius R.

16.2 Homework
Please, hand in your solutions next Thursday by 11am. Late papers will be marked down.
1. Let (q1 , q2 , q3 ) be a scalar eld dened in a general curvilinear coordinate system (q1 , q2 , q3 ). Show that the
gradient of is given as
3
1
grad = b
ei
h qi
i=1 i
[10 points]

2. Derive divF in the spherical coordinate system, where F=F(r, , ) is a vector eld. [5 points]

131
CHAPTER 16. PROBLEM SHEET 6: VECTOR CALCULUS, PART II 132
(
)
3. Calculate divF at the point P r = 1, = 2, = 2 for the vector eld
ebr
F= + r sin b
e
r2
(Answer : divF = 2 cos = 0) [5 points]
4. Calculate curl of a general vector eld F = Fr ebr + F eb + F eb in the spherical coordinates. [5 points]
5. Calculate curl F of the vector eld F = r2 ebr + r2 sin (b
e . Show that only)the second term in the vector eld
contributes. Hence, calculate curl F at the point P r = 1, = 2 , = 2 (Answer : curl F = 3r sin b e =
e ) and sketch the direction of this vector on this 3D Cartesian coordinate system.
3b [10 points]
6. Consider a particle of unit mass moving within the z = 0 plane in a central potential eld U (r) = 1/r, where
r is the distance from the centre.

(a) why the cylindrical coordinate system (r, , z) is the most convenient one for this problem? [1 point]
(b) derive an expression for the force eld, F = gradU , acting on the particle in this coordinate system. It
is known from the lectures that the scale factors for this system are: hr = 1, h = r, hz = 1. (Answer :
F = b er /r2 ) [3 points]
(c) using the fact that the acceleration vector in the cylindrical coordinates is
( ) ( )
a = r r2 ebr + 2r + r eb + zb
ez ,

nd equations of motion for both r(t) and (t). Hence, show that (t) will change linearly with time if
the particle moves along a circular trajectory within the plane.

16.3 This week CHALLENGE


The solutions, written separately, should be handed in together with the home work. Please, put them in the correct
pile!!!
1. Consider a small volume of an incompressible stationary uid of density . For the stationary uid the velocity
of the volume v(r(t)) depends only on its position r(t). First, show that the acceleration of the volume is
1 ( 2)
a = (v )v = v v ( v)
2
If the pressure in the uid is P , then show that the force density (the force per unit volume) acting on the small
volume can be recast as P . Then, by writing down the 2nd Newtons law for the volume and multiplying
it on a unit vector e directed along the velocity v (i.e. along the lines of uid current), derive the Bernullis
equation for the stationary uid:
1 2
v + P = constant along the line of ow (along the current)
2
[20 points]
2. Consider a propagation of electro-magnetic waves in a homogeneous conductiong media (e.g. a metal) with
conductivity along the direction x. Maxwell equations in the absence of external charges for the media are:
4 0 E
H= (E) +
c c t
1 H
E= , E = H = 0,
c t
where c is the speed of light and 0 the permittivity of free space. Assuming that the elds H(r, t) eit
and E(r, t) eit change with time harmonically, solve the Maxwell equations and obtain the relationship
betweent he elds H and E. Will the elds decay inside the media (attenuate)? Also show that, as in the
vacuum, the elds are still tangenital. [20 points]
The student(s) scored most points will get a surprise gift. The elegancy of the solutions will also be taken into
account! The maximum score is 40 points.
Chapter 17

Problem sheet 7: Frobenius method

2nd order linear homogeneous dierential equations

a(x)y + b(x)y + c(x)y = 0 (17.1)

can be solved using a generalised power series expansion (the Frobenius method)



y(x) = cr (x x0 )r+s = ar (x) (17.2)
r=0 r=0

with, in general, some non-integer s. The point x0 can be chosen arbitrary, but it is advised to choose it as one of
the singular points (if present) to have the largest radius of convergence of the series. To nd and characterise the
singular points of the dierential equation, rewrite it in a standard form:

b(x) c(x)
y + p(x)y + q(x)y = 0, where p(x) = and q(x) = (17.3)
a(x) a(x)

and then check the functions a(x), p(x) and q(x) as described in the lectures.

For all second order dierential equations below (problems 2 and 3):
nd and characterise singular points,

obtain two independent solutions y1 (x) and y2 (x) using the Frobenius method:

substitute the series expansion into the equation;


collect all terms into a single sum and a few extra terms;
obtain an indicial equation for s and nd its solutions;
obtain a recurrence relation between the coecients cr ;
for each value of s nd the rst four values of the coecients cr (i.e. for r=1,2,3,4) and then try to
establish an expression for a general r (the latter is not compulsory though!);

uding the recurrence relation for the coecients cr , investigate the radius of convergence for either of the two
solutions by solving for x the inequality
ar+1 (x)
lim <1
r ar (x)

17.1 Problems to be done in the class


1. Specify and classify the singular points of the dierential equation:
( 2 )( )
x + 1 x2 1 y + 3(x 1)y + 2(x + 1)y = 0

133
CHAPTER 17. PROBLEM SHEET 7: FROBENIUS METHOD 134

2. Obtain two independent series solutions of the dierential equation:

x2 y (x) + xy (x) 9y(x) = 0

3. Obtain two independent series solutions of the dierential equation:

3xy (x) + (3x + 1)y (x) + y(x) = 0

17.2 Homework
Please, hand in your solutions next Thursday by 11am. Late papers will be marked down.

1. Specify and classify the singular points of the dierential equation:


( )
(x 4) x2 1 y + (x 1)y + (x + 1)y = 0

[5 points]
2. Obtain two independent series solutions of the dierential equation:

x2 y (x) 6y(x) = 0

(Answer : y1 (x) = x2 and y2 (x) = x3 ) [10 points]


3. Obtain two independent series solutions of the dierential equation:

x2 y (x) + 2x2 y (x) 2y(x) = 0


( )
(Answer : y1 (x) = x1 (1 x) and y2 (x) = x2 1 x + 35 x2 15 4 3
x + ... ) [20 points]

17.3 This week CHALLENGE


The solutions, written separately, should be handed in together with the home work. Please, put them in the correct
pile!!!

1. Consider a general dierential equation (17.3). Show that, if we know one solution of the dierential equation,
y1 (x) , then the second solution can always be obtained as follows:
x [ x ]
Q(x1 )
y2 (x) = y1 (x) 2 dx1 , where Q(x) = exp p(x1 )dx1 (17.4)
x0 [y1 (x1 )] x0

[Hint: nd a rst order dierential equation for the function (x) = y1 (x)y2 (x) y1 (x)y2 (x).] [10 points]
2. Obtain one independent solution J0 (x) of the dierential equation

xy + y + xy = 0

using a generalised power series. As the indicial equation gives the same roots for s (in fact, s = 0), the second
solution of this dierential equation cannot be obtained directly. Instead, use the following trial function for
the second solution,


K0 (x) = J0 (x) ln x + br xr
r=1

and hence obtain the unknown coecients br . [25 points]


Chapter 18

Problem sheet 8-9: Separation of variables

18.1 Problems to be done in the class


1. Consider a partial dierential equation:
2 2
+ e(x +y ) = 0
2 2
+
x2 y 2
written in the Cartesian coordinates.

(a) Using (x, y) = X(x)Y (y), try to separate the variables x, y in the equation. Why do you think it does
not work?
(b) Choose now the polar coordinate system (r, ) (it is the cylindrical system without z) and rewrite the
equation in this system for (x, y) (r, ) employing the (r, )-part of the Laplacian in the cylindrical
coordinates: ( )
2 2 1 1 2
+ = r +
x2 y 2 r r r r2 2
(c) Attempt to separate the variables by writing (r, ) = R(r)(). State two ODEs for R(r) and ().

2. The sound in a pipe is described by the following one-dimensional wave equation


1 2 2
2 2
= (18.1)
v t x2
where v is the sound velocity. Initial conditions are: (x, 0) = 0 and t (x, 0) = v0 , which correspond to
somebody blowing into one end (at x = 0) of the pipe with velocity v0 at the initial moment only. The
boundary condition at the mouth end of the pipe is (0, t) = 0, i.e. the mouth is attached to the pipe all the
time and hence there are no vibrations of the air present, while the(other end,) x = L, of the pipe is openned,

i.e. the corresponding boundary condition there can be written as x (x, t) x=L = 0.

(a) Using (x, t) = X(x)T (t) in equation (18.1), separate the variables by introducing a separation constant
k and nally obtain two ODEs fot X(x) and T (t).
(b) Explain, by solving the ODE for X(x), why k should be negative, k = p2 , and other possibilities lead
to the trivial solution of no interest to us.
(c) Applying the boundary conditions to X(x), show that p takes on a discrete innite set of values
( )
1
p pn = n+ , n = 0, 1, 2, 3, ...
L 2

(d) The general solution of the equation is obtained as a superposition of all sets of products Xn (x)Tn (t):


(x, t) = [An sin (pn vt) + Bn cos (pn vt)] sin(pn x)
n=0

Why are there two arbitrary constants associated with every term in the sum?

135
CHAPTER 18. PROBLEM SHEET 8-9: SEPARATION OF VARIABLES 136

(e) Now you should apply the initial conditions to obtain the values of the arbitrary constants An and Bn for
all n = 0, 1, 2, .... Show that Bn = 0 and derive the expression for An . Finally, write down the particular
solution of the original dierential equation which satises both the boundary and initial conditions.

3. The method of separation of variables in the case of equations with more than two variables can be done in
stages, when each variable is separated individually from all others, i.e. one after the other, until one ends
up with a set of ordinary dierential equations (ODEs). As an example, consider the Laplace equation in the
Cartesian coordinate system:
2 2 2
+ + =0
x2 y 2 z 2

(a) By substituting (x, y, z) = (x, y)Z(z) into the equation, separate the variables (x, y) from z by
introducing the separation constant k1 . Obtain two equations: one for Z(z) and another for (x, y).
(b) Consider then the equation for (x, y) and attempt to separate the variables there considering (x, y) =
X(x)Y (y) and introducing another separation constant k2 . Find the corresponding ODEs for X(x) and
Y (y).
(c) What is the nal form of (x, y, z)?

4. The heat ow in a bar of length L can be described by the 1D heat transport equation

1 2
= (18.2)
t x2
where is the thermal diusivity. Initially the distribution of temperature in the bar is (x, 0) = 0 sin 3x
2L
(note that since the equation is of the rst order with respect to the time derivative - second order overall
- only one initial condition is sucient). The boundary condition at x = 0 end is (0, t) = 0, i.e. the bar
is maintained at zero temperature, ( while at)the other end, x = L, there is no heat loss, i.e. the boundary
condition there can be written as x (x, t) x=L = 0.

(a) Using (x, t) = X(x)T (t) in equation (18.2), separate the variables by introducing a separation constant
k and nally obtain two ODEs fot X(x) and T (t).
(b) Explain, by solving the ODE for T (t), why k should be negative and other possibilities lead to unphysical
results.
(c) Choose k = p2 and solve the ODE for X(x). There should be two arbitrary constants in the solution.
(d) Applying the boundary conditions to X(x), show that p takes on a discrete innite set of values
( )
1
p pn = n+ , n = 0, 1, 2, 3, ...
L 2

(e) The general solution of the equation is obtained as a superposition of all sets of products Xn (x)Tn (t):


An epn t sin(pn x)
2
(x, t) =
n=0

Why is there only one arbitrary constant associated with every term in the sum?
(f) Now you should apply the initial conditions to obtain the values of the arbitrary constants An for all
n = 0, 1, 2, .... Show that only one constant is not equal to zero which corresponds to n = 1. Finally,
write down the particular solution of the original dierential equation which satises both the boundary
and initial conditions.
(g) What will the steady state solution (at t ) be? Would you expect this result on the physical grounds?
CHAPTER 18. PROBLEM SHEET 8-9: SEPARATION OF VARIABLES 137

18.2 Homework
1. Consider the one-dimentional wave equation (30 points for the whole equation)

2y 1 2y
= (18.3)
x2 c2 t2
for a string of length L xed at both ends. 30 points

(a) Formulate the boundary conditions for this problem.


(b) Use the method of separation of variables to determine two ODEs, one involving x and another involving t.
Consider all the possibilities for the separation constant and nd the values consistent with the boundary
conditions. (Answer : the separation constant k = p2n , where pn = n/L with n = 1, 2, 3, . . .)
(c) Hence, write down a general solution of the wave equation that satises the boundary conditions.
( )
(d) Prove that the functions n (x) = sin nx
L for dierent n satisfy the orthonormality conditions:
L {
L 1, if n = m
n (x)m (x)dx = nm , where nm =
0 2 0, if n = m

L
(e) Now assume that the string is intially (at t = 0) pulled by 0.06 at x = 5 and then released. Determine
the corresponding partial solution of the wave equation. Answer :
(
)
3 n nx cnt
y(x, t) = 2 2
sin sin cos
n=1
4 n 5 L L
Chapter 19

Problem sheet 10: Legendre Polynomials

The Legendre polynomials Pn (x) can be dened as:


a solution of the Legendre dierential equation:

(1 x2 )Pn (x) 2xPn (x) + n(n + 1)Pn (x) = 0 (19.1)

coecients in the expansion of the generating function



1
G(x, t) = = Pn (x)tn (19.2)
1 2xt + t2 n=0

from the Rodrigues formula:


1 dn 2
Pn (x) = (x 1)n (19.3)
2n n! dxn
Legendre polynomials satisfy the following recurrence relations:

(n + 1)Pn+1 (x) (2n + 1)xPn (x) + nPn1 (x) = 0 (19.4)

An arbitrary well-behaved function f (x) can be expanded into a series with respect to the Legendre polynomials:

1
2n + 1
f (x) = cn Pn (x), where cn = f (x)Pn (x)dx (19.5)
n=0
2 1

19.1 Problems to be done in the class


1. Using Eq. (19.2), show that explicit expressions for Pn (x) with n = 0, 1, 2, 3 are given by:
1 1
P0 = 1, P1 = x, P2 = (3x2 1), P3 = (5x3 3x)
2 2

2. Check that the same expressions are obtained from the Rodrigues formula (19.3).

3. Using the recurrence relations (19.4), show that


1
P4 (x) = (35x4 30x2 + 3)
8

4. Show, using explicit calculation, that P3 (x) is orthogonal to P4 (x) (the expressions for those are given
above) and that it is properly normalised, i.e.
1 1
2 2
P3 (x)P4 (x)dx = 0 and P32 (x)dx = =
1 1 23+1 7

138
CHAPTER 19. PROBLEM SHEET 10: LEGENDRE POLYNOMIALS 139

5. Calculate the integrals:


1 1
Pn (x)dx and xPn (x)dx
1 1

6. Function f (x) = 1 + x + 2x2 can be expanded into a series with respect to the Legendre polynomials:


2
1 + x + 2x = cn Pn (x)
n=0

Since f (x) is a polynomial of the order two, all the Legendre polynomials Pn (x) with n > 2 can be excluded,
i.e. cn = 0 for any n > 2. Calculate then the coecients c0 , c1 and c2 using the general method of Eq. (19.5).
Using explicit expressions for the several rst Legendre polynomials from problem 1, verify your expansion.

19.2 Homework
1. Using both Rodrigues formula and the recurrence relation (the necessary polynomials of lower orders you
would need are given above), show that
1( )
P5 (x) = 63x5 70x3 + 15x
8
10 points

2. Check that P4 (x) is orthogonal to P1 (x) and P2 (x). 5 points


3. Expand f (x) = x3 via the appropriate Legendre polynomials. (Answer : x3 = 35 P1 (x) + 52 P3 (x)) 10 points

4. Hence, calculate the integral: 1


In = Pn (x)x3 dx
1

(Answer : In = 25 n1 + 4
35 n3 ) 5 points
Chapter 20

Problem sheet 11: Associated Legendre


Functions and Laplace Equation

The associated Legendre functions are given by

dm (1)m dn+m
Pnm (x) = (1)m (1 x2 )m/2 m
Pn (x) = n (1 x2 )m/2 n+m (x2 1)n (20.1)
dx 2 n! dx
where m = 0, . . . , n. The functions Pnm (x) and Pnm (x) with the same m but dierent n = n are orthogonal:
1
Pnm (x)Pnm (x)dx = 0 for n = n (20.2)
1

The Leibnitz formula for the multiple dierentiation of a product of two functions:
n ( )
(n) (n) n (n)
(uv) = (u(x)v(x)) = u(k) v (nk) (u + v) (20.3)
k
k=0

where the binomial coecients ( )


n n!
= (20.4)
k k!(n k)!
Note also that
n (
)
n n
(a + b) = ak bnk (20.5)
k
k=0

The Laplacian of in the spherical coordinates reads:


( ) ( )
1 2 1 1 2
2
2
r + 2 sin + 2 2 (20.6)
r r r r sin r sin 2

20.1 Problems to be done in the class


1. Calculate Pnm (x) for n = 1, 2, 3 and all values of m = 0, . . . , n.
2. Show that the functions Pnm (x) calculated in problem 1 (i.e. for n = 1, 2, 3) with the same m but dierent n
are orthogonal, Eq. (20.2).
3. Calculate all binomial coecients which are needed for (a + b)4 or (uv)(4) . Thus write (a + b)4 and (uv)(4) in
the expanded form using Eqs. (20.5) and (20.3).
4. Consider a function (x) = (1 x3 )f (x). The nthe derivative of (x) can be written in the following form:

(n) (x) = a0 (1 x3 )f (n) + a1 x2 f (n1) + a2 xf (n2) + a3 f (n3)

where f (k) is the kth derivative of f (x). Using the Leibnitz formula (20.3), calculate the numerical
coecients a0 , a1 , a2 and a3 .

140
CHAPTER 20. PROBLEM SHEET 11: ASSOCIATED LEGENDRE FUNCTIONS AND LAPLACE EQUATION141

5. Consider the Schrdinger equation for the hydrogen atom in atomic units (all fundamental constants
disappear in these units):
1
2 + V (r) = E
2
where V (r) = 1r is the central interaction between the electron and the nucleus and E is the electron
energy.

(a) Use the method of separation of variables to nd three ordinary dierential equations (ODE) for the
three functions in the product solution, (r, , ) = R(r)()().

(b) Show that the solutions for () and () result in the spherical harmonics, i.e.
{ }
m sin m
()() = Pn (cos ) (20.7)
cos m

where n = 0, 1, 2, . . . and m = 0, . . . , n.

20.2 Homework
1. Calculate Pnm (x) for n = 4 and m = 0, 1. Answer :
1( )
P40 (x) = 35x4 30x2 + 3
8
5 ( )
P41 (x) = 1 x2 7x3 3x
2
7 points
( )
2. Consider the function (x) = 2x 5x4 f (x). The nderivative of (x) can be written in the following form: 10 points

(n) (x) = a0 (x)f (n) + a1 (x)f (n1) + a2 (x)f (n2) + a3 (x)f (n3) + a4 (x)f (n4)

where f (k) is the kth derivative of f (x). Using the Leibnitz formula (20.3), obtain explicit expressions for
the functions ai (x), i = 1, ..., 4. Answer :

a0 (x) = 2x 5x4 ,

( )
a1 (x) = 2n 1 10x3 ,

a2 (x) = 30n(n 1)x2 ,

a3 (x) = 20n(n 1)(n 2)x

a4 (x) = 5n(n 1)(n 2)(n 3)

3. Consider the wave equation in the spherical coordinates: 16 points

1 2
= 2
c2 t2
(a) Use the method of separation of variables to nd four separate ODE for the four functions in the
product solution (r, , , t) = R(r)()()T (t).
(b) Show that the solutions of the ODEs for () and () also involve the spherical harmonics, Eq. (20.7).
Part III

Syllabus

142
143

The best preparation to the exam is in going through:


lecture notes, paying special attention to the points highlighted in the sections below
problems solved in the lecture notes
problems we did during the problem class (you have their solutions!)
your marked home work sheets
We may also browse through some books I recommended for this course. You should also be familiar with the
method of mathematica induction and how it can be used to prove some simple formulae.

20.3 Some special functions


20.3.1 Delta function
Denition; ltering theorem {
b
f (0), if a < x < b
(x)f (x)dx =
a 0 otherwise
and ability to solve problems with some particular f (x) using it.

20.3.2 Heviside step function


Denition; calculation of integrals with it, e.g.

H(x a)f (x)dx = f (x)dx
a

20.3.3 Gamma function


Denition, express integrals of the type

tn et dt and tn et
2 2
+t+c
dt

via the gamma function (t). [In the 2nd integral you have to make up the complete square, t2 + t + c =
(t t0 )2 + , and then change the variables.]

20.4 Fourier series


Orthogonal sets of functions; sine and cosine functions as an example of those; Fourier cos/sin series

a0 { nx }

nx
f (x) = + an cos + bn sin (20.8)
2 n=1
l l

for periodic functions f (x), where


l
1
a0 = f (x)dx (20.9)
l l
l l
1 1 mx
am = f (x)m (x)dx = f (x) cos dx (20.10)
l l l l l
l l
1 1 mx
bm = f (x)m (x)dx = f (x) sin dx (20.11)
l l l l l
understanding of the Dirichlet conditions; understanding if one can dierentiate and/or integrate (term by term)
the FS; calculation of innite numerical series from the FS of a function; Parcevals theorem

1 l 2 a2 ( 2 )
f (x)dx = 0 + an + b2n (20.12)
l l 2 n=1
144

(derivation is not required); calculation of innite numerical series from the above expansion.
Complex form of the FS,

f (x) = cn einx/l (20.13)
n=

with the Fourier coecients l


1
cn = f (x)einx/l dx
2l l

You should be able to derive Complex from from the sin/cos form and vise versa!
Practical skills: ability to expand a periodic function into the FS; calculation of a numerical series from the
FS of a function; FS for odd (f (x) = f (x)) and even (f (x) = +f (x)) functions.

20.5 Fourier transform


Denition of the FT of a function f (t):

F[f (t)] = F () = f (t)ei2t dt

and its inverse:


F 1 [F ()] = f (t) = F ()ei2t d

(derivation from the FS is not required); Dirichlet conditions; FT of a derivative of f (t); denition of the convolution
of two functions
f (t) g(t) = f (t )g( )d

and its FT:


F[f (t) g(t)] = F[f (t)]F[g(t)]
Practical skills: ability to calculate F () from f (t) and write f (t) avia its FT F (); calculation of denite intgerals
from the inverse FT at some particular values of t; FT of odd and even functions f (t).

20.6 Laplace transform


Denition, linearity; simple examples of LT of functions such as f (t) = 1, f (t) = et , eit , cos(t) and sin(t);
LT of a derivatives f (t) and f (t) (derivation is required); solution of ODEs and systems of ODEs using the LT
method; convolution theorem (derivation is required);
Practical skills: ability to obtain LT of a particular function f (t) given; ability to use given LTs of functions
(a Table of transforms) to calculation other transforms; ability to use the convolution theorem; ability to solve
linear ODEs and systems of those; simplication of fractions in the LT space and then do the inverse LT on those.

20.7 Vector calculus: Part I


Transformation equations between Cartesian (x, y, z) and curvilinear (q1 , q2 , q3 ) representations; coordinate lines
and surfaces; examples (polar, cyclindrical and spherical coordinates); derivation of the unit base vectors
( )
1 r r
ei =
, hi =
hi qi qi

via i, j, k Cartesian vectors. Orthogonal curvilinear systems and how this is checked. Line element


3
dr = hi dqi ei
i=1
145

and the volume element


( [ ])
r r r
dV = dq1 dq2 dq3 = h1 h2 h3 (e1 [e2 e3 ]) dq1 dq2 dq3
q1 q2 q3
volume element for orthogonal systems is simply
dV = h1 h2 h3 dq1 dq2 dq3
Mechanics of a point in curvilinear coordinates: calculation of the velocity v(t) and acceleration a(t) via curvilinear
coordinates and the unit base vectors.
Practical skills: ability to use the above equations for a particular curvilinear coordinate system, ability
to obtain (sketch) coordinate lines and surfaces; check orthogonality of unit base vectors; express i, j, k of the
Cartesian system via the ei vectors for orthogonal systems (the inverse problem); volume elements of the spherical
and cylindrical systems; ability to obtain v(t) and a(t) for the given curvilinear coordinates; ability to write a given
scalar and vector els entirely in the given curvilinear system.

20.8 Vector calculus: Part II


Denition of grad = of a scalar eld and its expression in orthogonal curvilinear coordinates (derivation is
required); understanding of the corresponding expressions for div, curl and Laplacian (derivation and memorising
of them are not required).
Practical skills: ability to apply given general expressions for grad, curl, div and to the particular curvilinear
system; calculation of grad, curl, div and at particular points.

20.9 Frobenius method


ODEs in the standard form
y + p(x)y + q(x)y = 0;
denition of the singular points; their characterisation (RSP and IRSP); how to obtain a series solution around a
given point (not IRSP) using this method:


y(x) = an (x a)n+s
n=0

Here s is important if a is a RSP. Be aware of special cases, but dealing with them in practice is not required.
Practical skills: nd singular points and characterise them; use the Frobenius method to solve gievn ODE of
the type above.

20.10 Separation of variables


Types of partial DE (PDE) in physics; general solution
(x, t) = u(x + ct) + v(x ct)
of the 1D wave equation for a string xed at two ends; interpretation of the two components of the solution;
superposition principle for linear PDEs; initial and boundary conditions.
Fourier method for solving PDEs:
separation of variables using an elementary product trial solution leading to several ODEs for each of the
functions in the product;
application of the boundary conditions to obtain the appropriate values for arbitrary constants introduced
when separating variables;
construction of the general solution;
calculation of the expansion coecients from the initial conditions.
Practical skills: how to rewrite the given partial DE in the given curvilinear system assuming that general
expressions for grad, div, curl and are available; follow the scheme above to obtain a particular solution of the
given PDE satisfying given boundary and initial conditions; specics of the wave and the heat transport PDEs.
146

20.11 Legendre polynomials


Denition of the Legendre polynomials Pn (x) via the generating function


1
G(x, t) = = Pn (x)tn
1 2xt + t2 n=0

Awareness of various recurrence relations connecting dierent Pn with each other (derivations are not required),

(n + 1)Pn+1 + nPn1 = (2n + 1)xPn

in particular; orthogonality of the polynomials (derivation is not required)


1
2
Pn (x)Pm (x)dx = nm
1 2n + 1

expansion of a function f (x) in a series via the polynomials



1
2n + 1
f (x) = cn Pn (x), cn = Pn (x)f (x)dx
n=0
2 1

Rodrigues formula
1 dn ( 2 )n
Pn (x) = x 1
2n n! dxn
(derivation is not required); Leibnitz theorem for (u(x)v(x))(n) via the binomial coecients (derivation is not
required).
Practical skills: derivation of several rst polynomials from the generating function, Rodrigues formula and
the recurrence relations; ability to expand simple functions (e.g. given polynomials) in a series withe respect to
Pn s; ability to apply Leibnitz theorem to products of functions.

20.12 Laplace equation


Laplace equation in spherical coordinates (r, , ); separation of variables, limitations of the arbitrary constants to
be introduced from the periodicity of the function depending on (derivation is required); be aware what is the
solution for the ODE involving the function depending on (derivation is not required), i.e. associated Legendre
function
(1)m ( )m/2 dn+m ( 2 )n ( )
2 m/2 d
m
Pnm (x) = n 1 x2 x 1 = (1) m
1 x Pn (x)
2 n! dxn+m dxm
Practical skills: ability to derive several rst functions from the formula above; separation of variables in the
Laplace equation or related PDEs written in spherical coordinates.

Anda mungkin juga menyukai