Dynamics: A Set of Notes on Theoretical Physical Chemistry

Jaclyn Steen, Kevin Range and Darrin M. York

December 5, 2003
1
Contents
1 Vector Calculus 6
1.1 Properties of vectors and vector space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Fundamental operations involving vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Linear Algebra 11
2.1 Matrices, Vectors and Scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Matrix Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Transpose of a Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Unit Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Trace of a (Square) Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5.1 Inverse of a (Square) Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 More on Trace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 More on [A, B] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.8 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.8.1 Laplacian expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.8.2 Applications of Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.9 Generalized Greens Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.10 Orthogonal Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.11 Symmetric/Antisymmetric Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.12 Similarity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.13 Hermitian (self-adjoint) Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.14 Unitary Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.15 Comments about Hermitian Matrices and Unitary Tranformations . . . . . . . . . . . . . . . . 18
2.16 More on Hermitian Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.17 Eigenvectors and Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.18 Anti-Hermitian Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.19 Functions of Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.20 Normal Marices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21 Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21.1 Real Symmetric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21.2 Hermitian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21.3 Normal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21.4 Orthogonal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.21.5 Unitary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2
CONTENTS CONTENTS
3 Calculus of Variations 22
3.1 Functions and Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Functional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Variational Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Functional Derivatives: Elaboration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4.1 Algebraic Manipulations of Functional Derivatives . . . . . . . . . . . . . . . . . . . . 25
3.4.2 Generalization to Functionals of Higher Dimension . . . . . . . . . . . . . . . . . . . . 26
3.4.3 Higher Order Functional Variations and Derivatives . . . . . . . . . . . . . . . . . . . . 27
3.4.4 Integral Taylor series expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4.5 The chain relations for functional derivatives . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.6 Functional inverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Homogeneity and convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5.1 Homogeneity properties of functions and functionals . . . . . . . . . . . . . . . . . . . 31
3.5.2 Convexity properties of functions and functionals . . . . . . . . . . . . . . . . . . . . . 32
3.6 Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7.1 Problem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7.2 Problem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7.3 Problem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7.3.1 Part A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7.3.2 Part B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.3.3 Part C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.3.4 Part D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.3.5 Part E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.4 Problem 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.4.1 Part F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.4.2 Part G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.7.4.3 Part H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.7.4.4 Part I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.7.4.5 Part J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.7.4.6 Part K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.7.4.7 Part L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.7.4.8 Part M . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Classical Mechanics 40
4.1 Mechanics of a system of particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.1 Newtons laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.1.2 Fundamental denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 DAlemberts principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.4 Velocity-dependent potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.5 Frictional forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5 Variational Principles 54
5.1 Hamiltons Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Comments about Hamiltons Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3 Conservation Theorems and Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3
CONTENTS CONTENTS
6 Central Potential and More 61
6.1 Galilean Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.2 Kinetic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3 Motion in 1-Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3.1 Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3.2 Generalized Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.4 Classical Viral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.5 Central Force Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.6 Conditions for Closed Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.7 Bertrands Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.8 The Kepler Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.9 The Laplace-Runge-Lenz Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
7 Scattering 73
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.2 Rutherford Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.2.1 Rutherford Scattering Cross Section . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.2.2 Rutherford Scattering in the Laboratory Frame . . . . . . . . . . . . . . . . . . . . . . 76
7.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
8 Collisions 78
8.1 Elastic Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9 Oscillations 82
9.1 Euler Angles of Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.2 Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.3 General Solution of Harmonic Oscillator Equation . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.3.1 1-Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.3.2 Many-Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
9.4 Forced Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.5 Damped Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10 Fourier Transforms 90
10.1 Fourier Integral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
10.2 Theorems of Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.3 Derivative Theorem Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
10.4 Convolution Theorem Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
10.5 Parsevals Theorem Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
11 Ewald Sums 95
11.1 Rate of Change of a Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
11.2 Rigid Body Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
11.3 Principal Axis Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
11.4 Solving Rigid Body Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
11.5 Eulers equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11.6 Torque-Free Motion of a Rigid Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
11.7 Precession in a Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
11.8 Derivation of the Ewald Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.9 Coulomb integrals between Gaussians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4
CONTENTS CONTENTS
11.10Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
11.11Linear-scaling Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
11.12Greens Function Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
11.13Discrete FT on a Regular Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
11.14FFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
11.15Fast Fourier Poisson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12 Dielectric 106
12.1 Continuum Dielectric Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
12.2 Gauss Law I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.3 Gauss Law II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.4 Variational Principles of Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.5 Electrostatics - Recap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
12.6 Dielectrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
13 Exapansions 115
13.1 Schwarz inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
13.2 Triangle inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
13.3 Schmidt Orthogonalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
13.4 Expansions of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
13.5 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
13.6 Convergence Theorem for Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
13.7 Fourier series for different intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
13.8 Complex Form of the Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
13.9 Uniform Convergence of Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
13.10Differentiation of Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
13.11Integration of Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
13.12Fourier Integral Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
13.13M-Test for Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
13.14Fourier Integral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
13.15Examples of the Fourier Integral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
13.16Parsevals Theorem for Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
13.17Convolution Theorem for Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
13.18Fourier Sine and Cosine Transforms and Representations . . . . . . . . . . . . . . . . . . . . . 143
5
Chapter 1
Vector Calculus
These are summary notes on vector analysis and vector calculus. The purpose is to serve as a review. Although
the discussion here can be generalized to differential forms and the introduction to tensors, transformations and
linear algebra, an in depth discussion is deferred to later chapters, and to further reading.
1, 2, 3, 4, 5
For the purposes of this review, it is assumed that vectors are real and represented in a 3-dimensional Carte-
sian basis ( x, y, z), unless otherwise stated. Sometimes the generalized coordinate notation x
1
, x
2
, x
3
will be
used generically to refer to x, y, z Cartesian components, respectively, in order to allow more concise formulas
to be written using using i, j, k indexes and cyclic permutations.
If a sum appears without specication of the index bounds, assume summation is over the entire range of the
index.
1.1 Properties of vectors and vector space
A vector is an entity that exists in a vector space. In order to take for (in terms of numerical values for its
components) a vector must be associated with a basis that spans the vector space. In 3-D space, for example, a
Cartesian basis can be dened ( x, y, z). This is an example of an orthonormal basis in that each component
basis vector is normalized x x = y y = z z = 1 and orthogonal to the other basis vectors x y = y z =
z x = 0. More generally, a basis (not necessarily the Cartesian basis, and not necessarily an orthonormal basis) is
denoted (e
1
, e
2
, e
3
. If the basis is normalized, this fact can be indicated by the hat symbol, and thus designated
( e
1
, e
2
, e
3
.
Here the properties of vectors and the vector space in which they reside are summarized. Although the
present chapter focuses on vectors in a 3-dimensional (3-D) space, many of the properties outlined here are more
general, as will be seen later. Nonetheless, in chemistry and physics, the specic case of vectors in 3-D is so
prevalent that it warrants special attention, and also serves as an introduction to more general formulations.
A 3-D vector is dened as an entity that has both magnitude and direction, and can be characterized, provided
a basis is specied, by an ordered triple of numbers. The vector x, then, is represented as x = (x
1
, x
2
, x
3
).
Consider the following denitions for operations on the vectors x and y given by x = (x
1
, x
2
, x
3
) and
y = (y
1
, y
2
, y
3
):
1. Vector equality: x = y if x
i
= y
i
i = 1, 2, 3
2. Vector addition: x +y = z if z
i
= x
i
+y
i
i = 1, 2, 3
3. Scalar multiplication: ax = (ax
1
, ax
2
, ax
3
)
4. Null vector: There exists a unique null vector 0 = (0, 0, 0)
6
CHAPTER 1. VECTOR CALCULUS 1.2. FUNDAMENTAL OPERATIONS INVOLVING VECTORS
Furthermore, assume that the following properties hold for the above dened operations:
1. Vector addition is commutative and associative:
x +y = y +x
(x +y) +z = x + (y +z)
2. Scalar multiplication is associative and distributive:
(ab)x = a(bx)
(a +b)(x +y) = ax +bx +ay +by
The collection of all 3-D vectors that satisfy the above properties are said to form a 3-D vector space.
1.2 Fundamental operations involving vectors
The following fundamental vector operations are dened.
Scalar Product:
a b = a
x
b
x
+a
y
b
y
+a
z
b
z
=

i
a
i
b
i
= [a[[b[cos()
= b a (1.1)
where [a[ =

a a, and is the angle between the vectors a and b.

Cross Product:
a b = x(a
y
b
z
a
z
b
y
) + y(a
z
b
x
a
x
b
z
) + z(a
x
b
y
a
y
b
x
) (1.2)
or more compactly
c = a b (1.3)
where (1.4)
c
i
= a
j
b
k
a
k
b
j
(1.5)
where i, j, k are x, y, z and the cyclic permutations z, x, y and y, z, x, respectively. The cross product can be
expressed as a determinant:
The norm of the cross product is
[a b[ = [a[[b[sin() (1.6)
where, again, is the angle between the vectors a and b The cross product of two vectors a and b results in a
vector that is perpendicular to both a and b, with magnitude equal to the area of the parallelogram dened by a
and b.
The Triple Scalar Product:
a b c = c a b = b c a (1.7)
and can also be expressed as a determinant
The triple scalar product is the volume of a parallelopiped dened by a, b, and c.
The Triple Vector Product:
a (b c) = b(a c) c(a b) (1.8)
The above equation is sometimes referred to as the BAC CAB rule.
Note: the parenthases need to be retained, i.e. a (b c) ,= (a b) c in general.
Lattices/Projection of a vector
a = (a
x
) x + (a
y
) y + (a
y
) y (1.9)
7
1.2. FUNDAMENTAL OPERATIONS INVOLVING VECTORS CHAPTER 1. VECTOR CALCULUS
a x = a
x
(1.10)
r = r
1
a
1
+r
2
a
2
+r
3
a
3
(1.11)
a
i
a
j
=
ij
(1.12)
a
i
=
a
j
a
k
a
i
(a
j
a
k
)
(1.13)
Gradient,
= x
_

x
_
+ y
_

y
_
+ z
_

z
_
(1.14)
f([r[) = r
_
f
r
_
(1.15)
dr = xdx + ydy + zdz (1.16)
d = () dr (1.17)
(uv) = (u)v +u(v) (1.18)
Divergence,
V =
_
V
x
x
_
+
_
V
y
y
_
+
_
V
z
z
_
(1.19)
r = 3 (1.20)
(rf(r)) = 3f(r) +r
df
dr
(1.21)
if f(r) = r
n1
then rr
n
= (n + 2)r
n1
(fv) = f v +f v (1.22)
Curl,
x(
_
V
z
y
_

_
V
y
z
_
) + y(
_
V
x
z
_

_
V
z
x
_
) + z(
_
V
y
x
_

_
V
x
y
_
) (1.23)
(fv) = fv + (f) v (1.24)
r = 0 (1.25)
(rf(r)) = 0 (1.26)
(a b) = (b )a + (a )b +b (a) +a (b) (1.27)
8
CHAPTER 1. VECTOR CALCULUS 1.2. FUNDAMENTAL OPERATIONS INVOLVING VECTORS
= =
2
=
_

2

2
x
_
+
_

2

2
y
_
+
_

2

2
z
_
(1.28)
Vector Integration
Divergence theorem (Gausss Theorem)
_
V
f (r)d
3
r =
_
S
f (r) d =
_
S
f (r) nda (1.29)
let f (r) = uv then
(uv) = u v +u
2
v (1.30)
_
V
u vd
3
r +
_
V
u
2
vd
3
r =
_
S
(uv) nda (1.31)
The above gives the second form of Greens theorem.
Let f (r) = uv vu then
_
V
u vd
3
r +
_
V
u
2
vd
3
r
_
V
u vd
3
r
_
V
v
2
ud
3
r =
_
S
(uv) nda
_
S
(vu) nda (1.32)
Above gives the rst form of Greens theorem.
Generalized Greens theorem
_
V
u

Lu u

Lvd
3
r =
_
S
p(vu uv)) nda (1.33)
where

L is a self-adjoint (Hermetian) Sturm-Lioville operator of the form:

L = [p] +q (1.34)
Stokes Theorem
_
S
(v) nda =
_
C
V d (1.35)
Generalized Stokes Theorem
_
S
(d ) [] =
_
C
d [] (1.36)
where = ,,
Vector Formulas
a (b c) = b (c a) = c (a b) (1.37)
a (b c) = (a c)b (a b)c (1.38)
(a b) (c d) = (a c)(b d) (a d)(b c) (1.39)
= 0 (1.40)
(a) = 0 (1.41)
9
1.2. FUNDAMENTAL OPERATIONS INVOLVING VECTORS CHAPTER 1. VECTOR CALCULUS
(a) = ( a)
2
a (1.42)
(a) = a + a (1.43)
(a) = a +a (1.44)
(a b) = (a )b + (b )a +a (b) +b (a) (1.45)
(a b) = b (a) a (b) (1.46)
(a b) = a( b) b( a) + (b )a (a )b (1.47)
If x is the coordinate of a point with magnitude r = [x[, and n = x/r is a unit radial vector
x = 3 (1.48)
x = 0 (1.49)
n = 2/r (1.50)
n = 0 (1.51)
(a )n = (1/r)[a n(a n)] (1.52)
10
Chapter 2
Linear Algebra
2.1 Matrices, Vectors and Scalars
Matrices - 2 indexes (2
nd
rank tensors) - A
ij
/A
Vectors - 1 index

(1
st
rank tensor) - a
i
/a
Scalar - 0 index

(0 rank tensor) - a
Note: for the purpose of writing linear algebraic equations, a vector can be written as an N 1 Column
vector (a type of matrix), and a scalar as a 1 1 matrix.
2.2 Matrix Operations
Multiplication by a scalar .
C = A
. .
NN NN
= A means C
ij
= A
ij
= A
ij

Addition/Subtraction
C = AB = BA means C
ij
= A
ij
B
ij
Multiplication (inner product)
C = AB
. .
NN NMMN
means C
ij
=

k
A
ik
B
kj
AB ,= BA in general
A (B C) = (AB)C = ABC associative, not always communitive
Multiplication (outer product/direct product)
C
..
nmnm
= AB means C

= A
ij
B
k
=n(i 1) +k =m(j 1) +
11
2.3. TRANSPOSE OF A MATRIX CHAPTER 2. LINEAR ALGEBRA
AB ,=BA A(BC) =(AB)C
Note, for vectors
C = a b
T
. .
N11N
means C
ij
= a
i
b
j
2.3 Transpose of a Matrix
A = B
T
. .
NM (MN)
T
=NM
means A
ij
= (B
ij
)
T
= B
ji
Note:
(A
T
)
T
= A
_
(A
ij
)
T

T
= [A
ji
]
T
= A
ij
(A B)
T
= B
T
A
T
C =(A B)
T
= B
T
A
T
C
ij
=
_

k
A
ik
B
kj
_
T
=

k
A
jk
B
ki
=

k
B
ki
A
jk
=

k
(B
ik
)
T
(A
kj
)
T
2.4 Unit Matrix
(identity matrix)
_
_
_
1 0 0
0
.
.
. 0
0 0 1
_
_
_ 1 (2.1)
1
ij

ij
1
1
=1
T
= 1 A1 =1A = A
Commutator: a linear operation
[A, B] ABBA
[A, B] = 0 if Aand B are diagonal matrices
Diagonal Matrices:
A
ij
= a
ii

ij
Jacobi Identity:
[A, [B, C]] = [B, [A, C]] [C, [A, B]]
12
CHAPTER 2. LINEAR ALGEBRA 2.5. TRACE OF A (SQUARE) MATRIX
2.5 Trace of a (Square) Matrix
(a linear operator)
T
r
(A) =

i
A
ii
= T
r
_
A
T
_
T
r
(A B) = T
r
(C) =

i
C
ii
=

k
A
ik
B
ki
=

i
B
ki
A
ik
= T
r
(BA)
Note T
r
([A, B]) = 0
T
r
(A+B) = T
r
(A) +T
r
(B)
2.5.1 Inverse of a (Square) Matrix
A
1
A = 1 = A A
1
Note 1 1 = 1, thus 1 = 1
1
(A B)
1
= B
1
A
1
prove
2.6 More on Trace
Tr(ABC) = Tr(CBA)
Tr(a b
T
) = a
T
b
Tr(U
+
AU) = Tr(A) U
+
U = 1 or Tr(B
1
AB) = Tr(A)
Tr(S
+
S) 0 Tr(BSB
1
BTB
1
= Tr(ST)
Tr(A) = TrA
+
Tr(S

)
Tr(AB) = Tr(BA) = Tr(B
+
A
+
)
Tr(ABC) = Tr(CAB) = Tr(BCA)
Tr([A, B]) = 0
Tr(AB) = 0 if A = A
T
and B = B
T
2.7 More on [A, B]
[S
x
, S
y
] = iS
z
Proof: (A B)
1
= B
1
A
1
If x y = 1 then y = x
1
associative (A B) (B
1
A
1
) = A (B B
1
) A
1
= (A A
1
) = 1 thus (B
1
A
1
) = (A B)
1
13
2.8. DETERMINANTS CHAPTER 2. LINEAR ALGEBRA
2.8 Determinants
det(A) =

A
11
A
12
A
1n
A
21
A
22
A
2n
.
.
.
.
.
.
.
.
. A
nn

i,k...

ijk...
A
1i
A
2j
A
3k
. . . (2.2)

ijk...
: Levi-Civita symbol (1 even/odd permutation of 1, 2, 3 . . ., otherwise 0. (has N! terms)
A is a square matrix, (N N)
2.8.1 Laplacian expansion
D =
N

i
(1)
i+j
M
ij
A
ij
=
N

i
c
ij
A
ij
M
ij
= minor ij
c
ij
= cofactor = (1)
i+j
M
ij
D =
k
A
kk
(2.3)

A
11
0
A
21
A
22

(2.4)
N

i
= A
ij
c
ik
= det(A)
jk
=

A
ji
c
ik
det(A) is an antisymmetrized product
Properties: for an N N matrix A
1. The value of det(A) = 0 if
any two rows (or columns) are equal
each element of a row (or column) is zero
any row (or column) can be represented by a linear combination of the other rows (or columns). In
this case, A is called a singular matrix, and will have one or more of its eigenvalues equal to zero.
2. The value of det(A) is unchanged if
two rows (or columns) are swapped, sign changes
a multiple of one row (or column) is added to another row (or column)
A is transposed det(A) = det(A
+
) or det(A
+
) = det(A

) = (det(A))

14
CHAPTER 2. LINEAR ALGEBRA 2.8. DETERMINANTS
A undergoes unitary transformation det(A) = det(U
+
AU) (including the unitary tranformation that
diagonalized A)
det(e
A
) = e
tr(A)
(2.5)
3. If any row (or column) of A is multiplied by a scalar , the value of the determinat is det(A). If the
whole matrix is multiplied by , then
det(A) =
N
det(A) (2.6)
4. If A = BC, det(A) = det(B) det(C), but if A = B +C, det(A) ,= det(B) +det(C). (det(A) is not
a linear operator)
det(A
N
) = [det(A)]
N
(2.7)
5. If A is diagonalized, det(A) =
i
A
ii
(also, det(1) = 1)
6. det(A)
1
) = (det(A))
1
7. det(A

) = (detA)

= det(A
+
)
8. det(A) = det(U
+
AU) U
+
U = 1
2.8.2 Applications of Determinants
Wave function:
HF
DFT
(x
1
, x
2
x
N
) =
1

N!

1
(x
1
)
2
(x
1
)
N
(x
1

1
(x
2
)
2
(x
2
)
.
.
.
.
.
.
.
.
.
N
(x
N
)

(2.8)
Evaluate:
_

(x
1
, x
2
, x
3
)(x
1
, x
2
, x
3
)d
(write all the terms). How does this expression reduce if
_

i
(x)
j
(x)d =
iJ
(orthonormal spin orbitals)
J=Jacobian [dx
k
>= J[dq
j
>=

i
[dq
i
>< dq
i
[dx
k
>
dx
1
dx
2
dx
3
= det(J)dq
1
dq
2
dq
3
, J
ik
=
x
i
q
k
J
_
x
q
_
: J
ij
=
x
i
q
j
J
_
q
x
_
ij
=
q
i
x
j
(2.9)
det
_
J
_
q
x
__
=

q
1
x
1
q
1
x
2

.
.
.
.
.
.

(2.10)
q
1
=2x q
2
=y q
3
=z
dxdydz =
1
2
dq
1
(2.11)
15
2.9. GENERALIZED GREENS THEOREM CHAPTER 2. LINEAR ALGEBRA
x

+2x
x
x

=
1
2
x

x
=2 dx =
1
2
dx

dx =
dx
dq
1
dq
1
=
1
2
dq
1
(2.12)
2.9 Generalized Greens Theorem
Use:
_
v
vd =
_
s
v d (2.13)
_
v
(vLu uLv)d =
_
s
p(xu uv) d (2.14)
L = [p] +q = p +p
2
+q
_
v
_
v(p u +
2
u +q)u u(p v +p
2
v +q)v

d
Note
_
vqud =
_
uqv d
_
v
(v [p]u +v (pu) u [p]v v (pu)) d
=
_
v
( vpu upv) d (2.15)
v pu +v [p]u u pv u [p]v
=
_
s
(vpu upv) d
=
_
s
p(vu uv) ds
dF
i
=

k
F
i
x
k
dx
k
+
F
i
t
dt =(dr )F
i
+
F
i
t
dt
=
_
dr +dt

t
_
F
i
(2.16)
L = r p = r mv, v = r
A(B C) = B(AC) C(AB) r = r r
L =m(r ( r))
=m[(r r) r(r )]
=m[(r
2
r r) r r(r r )]
=mr
2
[ r( r w)] (2.17)
I =mr
2
L =I if r = 0
16
CHAPTER 2. LINEAR ALGEBRA 2.10. ORTHOGONAL MATRICES
2.10 Orthogonal Matrices
(analogy to a
i
a
j
=
ij
= a
T
i
a
j
)
A
T
A = 1
_
A
T
= A
1
_
A
T
= A
1
, therefore A
T
A = 1 = 1
T
= (A
T
A)
T
Note det(A) = 1 if A is orthogonal. Also: A and B orthogonal, then (AB) orthogonal.
Application: Rotation matrices
x

i
=

j
A
ij
x
j
or [x

i
>=

j
[x
j
x
j
[x

i
>
example:
_
_
r

_
_
=
_
_
sin cos sin sin cos
cos cos cos sin sin
sin cos 0
_
_
_
_
x
y
z
_
_
(2.18)
_
_
r

_
_
= C
_
_
x
y
z
_
_
(2.19)
_
_
x
y
z
_
_
= C
1
_
_
r

_
_
= C
T
_
_
r

_
_
(2.20)
since C
1
= C
T
(C is an orthogonal matrix)
Also Euler angles (we will use later. . .)
2.11 Symmetric/Antisymmetric Matrices
Symmetric means A
ij
= A
ji
, or A = A
T
Antisymmetric means A
ij
= A
ji
, or A = A
T
A =
1
2
_
A+A
T
_
+
1
2
_
AA
T
_
(2.21)
symmetric antisymmetric
Note also:
_
AA
T
_
T
=
_
A
T
_
T
A
T
=AA
T
Note: T
r
(AB) = 0 if A is symmetric and B is antisymmetric. Thus AA
T
A
T
A are symmetric, but
A A
T
,= A A
T
Quiz: If A is an upper triangular matrix, use the Laplacian expansion to determine a formula for det(A) in
terms of the elements of A.
17
2.12. SIMILARITY TRANSFORMATION CHAPTER 2. LINEAR ALGEBRA
2.12 Similarity Transformation
A

= BAB

(2.22)
if B
1
= B
T
(B orthogonal) BAB
T
= orthogonal similarity transformation.
2.13 Hermitian (self-adjoint) Matrices
Note:
(AB)

= A

(AB)
+
= B
+
A
+
also (A
+
)
+
= A
H
+
= H where H
+
(H

)
T
= (H
T
)

A real symmetric matrix is Hermitian or a real Hermitian matrix is symmetric (if a matrix is real, Hermi-
tian=symmetric)
2.14 Unitary Matrix
U
+
= U
1
A real orthogonal matrix is unitary or a real unitary matrix is orthogonal (if a matrix is real, unitary=orthogonal)
2.15 Comments about Hermitian Matrices and Unitary Tranformations
1. Unitary tranformations are norm preserving
x

= Ux (x

)
+
x

= x
+
U
+
Ux = x
+
x
2. More generally, x

= Ux, A

= UAU
+
A

= UAU
+
Ux = U(Ax) and (y)
+
A

= y
+
U
+
UAU
+
Ux
operation in transformation coordinates = transformation in uniform coordinates = y
+
Ax (invariant)
3. If A
+
= A, then (Ay)
+
x = y
+
x = y
+
A
+
x = y
+
A x (Hermitian property)
4. If A
+
= A, then (A

)
+
= (UAU
+
)
+
= UAU
+
, or (A

)
+
= A

2.16 More on Hermitian Matrices

C =
1
2
(C+C
+
)
. .
Hermitian
+
1
2
(CC
+
)
. .
anti-Hermitian
=
1
2
(C+C
+
) +
1
2i
i(CC
+
)
. .
Hermitian!
Note: C = i[A, B] is Hermitian even if A and B are not, (or iC = [A, B]). C = AB is Hermition if
A = A
+
, B = B
+
, and [A, B] = 0. A consequence of this is that C
n
is Hermitian if C is Hermitian. Also,
C = e
A
is Hermitian if A = A
+
, but C = e
iA
is not Hermitian (C is unitary). A unitary matrix in general
is not Hermitian.
f(A) =

k
C
k
A
k
is Hermitian if C
k
are real
18
CHAPTER 2. LINEAR ALGEBRA 2.17. EIGENVECTORS AND EIGENVALUES
2.17 Eigenvectors and Eigenvalues
Solve Ac = ac, (Aa1)c = 0 det() = 0 secular equation.
A

i
= a
i
c

i
Eigenvalue problem
Ac
i
=
i
Bc
i
Generalized eigenvalue problem
Ac = Bc
If c
+
i
B c
j
=
ij
then c
+
i
A c
j
=
i

ij
relation: A

= B

1
2
AB

1
2
and c

i
= B
1
2
c
i
(Can always transform. . .Lowden) Example: Hartree-Fock/Kohn-
Shon Equations:
FC = SC, H
eff
C = SC
For Ac
i
=
i
c
i
if A = A
+
a
i
= a

i
, c
+
i
c
j
=
ij
can be chosen
c
i
form a complete set
If A
+
A = 1, a
i
= 1, c
+
i
c
j
=
ij
det(A) = 1
Hc
i
=
i
c
i
or Hc = c
c =
_
c
1
c
2
c
3
.
.
.
.
.
.
.
.
.
_
(2.23)
=
_
_
_

1
0
.
.
.
0
N
_
_
_ (2.24)
since c
+
i
c
j
=
ij
, c
+
c = 1 and also c
+
H c = c
+
c = c
+
c = hence c is a unitary matrix and is the
unitary matrix that diagonalizes H. c
+
Hc = (eigenvalue spectrum).
2.18 Anti-Hermitian Matrices
Read about them. . .
2.19 Functions of Matrices
UAU
+
= a A = U
+
aU or AU
+
= U
+
a
f(A) = U
+
_
_
_
f(a
1
)
.
.
.
f(a
n
)
_
_
_U
19
2.19. FUNCTIONS OF MATRICES CHAPTER 2. LINEAR ALGEBRA
Power series e.g.
e
A
=

k=0
1
k!
A
k
sin(A) =

k=0
(1)
k
(2k + 1)!
A
2k+1
cos(A) =

k=0
(1)
k
(2k)!
A
2k
Note:
A
2
U
+
=AUU
+
=AU
+
Q
=U
+
QQ
=U
+
Q
2
Q
2

_
Q
2
11
0
0
.
.
.
_
A
k
U
+
= U
+
Q
K
Note, if f(A) =

k
C
k
A
k
and UAU
+
= Q, then
F =f(A)U
+
=

k
c
k
U
+
Q
k
=U
+

k
c
k
Q
k
=U
+
_
_
_
f(a
11
)
.
.
.
f(a
nn
)
_
_
_ = U
+
f (2.25)
so if UAU
+
= Q, then UFU
+
= f and f
ij
= f(a
i
)
ij
Also:
Trace Formula
det(e
A
) = e
Tr(A)
(special case of det[f(A)] =
i
f(a
ii
))
Baker-Hausdorff Formula
e
iG
He
iG
= H+ [iGH] +
1
2
[iG, [iG, H]] +
Note, if
AU
+
= U
+
QA() = A+1
so has some eigenvectors, but eigenvalues are shifted.
A()U
+
= (A+1)U
+
= U
+
Q+U
+
1 = U
+
(Q+1) = U
+
Q()
20
CHAPTER 2. LINEAR ALGEBRA 2.20. NORMAL MARICES
2.20 Normal Marices
[A, A
+
] = 0 (Hermitian and real symmetry are specic cases)
Ac
i
= a
i
c
i
, A
+
c
i
= a

i
c
i
, c
+
i
c
j
= 0
2.21 Matrix
2.21.1 Real Symmetric
A = A
T
= A
+
a
i
real c
T
i
c
j
=
ij
Hermitian normal
2.21.2 Hermitian
A = A
+
a
i
real c
+
i
c
j
=
ij
normal
2.21.3 Normal
[A, A
+
] = 0 if Ac
i
= a
i
c
i
c
+
i
c
j
=
ij
2.21.4 Orthogonal
U
T
U = 1 (U
T
= U
1
) a
i
(1) c
T
i
c
j
=
ij
unitary, normal
2.21.5 Unitary
U
+
U = 1 (U
+
= U
1
) a
i
real (1) c
+
i
c
j
=
ij
normal
If UAU
+
= a, and U
+
U = 1 then [A, A
+
] = 0 and conversely.
21
Chapter 3
Calculus of Variations
3.1 Functions and Functionals
Here we consider functions and functionals of a single argument (a variable and function, respectively) in
order to introduce the extension of the conventional function calculus to that of functional calculus.
A function f(x) is a prescription for transforming a numerical argument x into a number; e.g. f(x) =
1 +x
2
+e
x
.
A functional F[y] is a prescription for transforming a function argument y(x) into a number; e.g. F[y] =
_
x
2
x
1
y
2
(x) e
sx
dx.
Hence, a functional requires knowledge of its function argument (say y(x)) not at a single numerical point x, in
general, but rather over the entire domain of the functions numerical argument x (i.e., over all x in the case of
y(x)). Alternately stated, a functional is often written as some sort of integral (see below) where the argument of
y (we have been referring to it as x) is a dummy integration index that gets integrated out.
In general, a functional F[y] of the 1-dimensional function y(x) may be written
F[y] =
_
x
2
x
1
f
_
x, y(x), y

(x), y
(n)
(x),
_
dx (3.1)
where f is a (multidimensional) function, and y

dy/dx, y
n
d
n
y/dx
n
. For the purposes here, we will
consider the bounday conditions of the function argument y are such that y, y

y
(n1)
have xed values at
the endpoints; i.e.,
y
(j)
(x
1
) = y
(j)
1
, y
(j)
(x
2
) = y
(j)
2
for j = 0, (n 1) (3.2)
where y
(j)
1
and y
(j)
2
are constants, and y
(0)
y.
In standard function calculus, the derivative of a function f(x) with respect to x is dened as the limiting
process
_
df
dx
_
x
lim
0
f(x +) f(x)

(3.3)
(read the derivative of f with respect to x, evaluated at x). The function derivative indicates how f(x) changes
when x changes by an inntesimal amount from x to x +.
Analogously, we dene the functional derivative of F[y] with respect to the function y at a particular point
x
0
by
_
F
y(x)
_
y(x
0
)
lim
0
F[y +(x x
0
)] F[y]

(3.4)
22
CHAPTER 3. CALCULUS OF VARIATIONS 3.2. FUNCTIONAL DERIVATIVES
(read, the functional derivative of F with respect to y, evaluated at the point y(x
0
)). This functional derivative
indicates how F[y] changes when the function y(x) is changed by an inntesimal amount at the point x = x
0
from y(x) to y(x) +(x x
0
).
We now procede formally to derive these relations.
3.2 Functional Derivatives: Formal Development in 1-Dimension
Consider the problem of nding a function y(x) that corresponds to a stationary condition (an extremum
value) of the functional F[y] of Eq. 3.1, subject to the boundary conditions of Eq. 3.2; i.e., that the function y
and a sufcient number of its derivatives are xed at the boundary. For the purposes of describing variations, we
dene the function
y(x, ) = y(x) +(x) = y(x, 0) +(x) (3.5)
where (x) is an arbitrary differentiable function that satises the end conditions
(j)
(x
1
) =
(j)
(x
2
) = 0
for j = 1, (n 1) such that in any variation, the boundary conditions of Eq. 3.2 are preserved; i.e., that
y
(j)
(x
1
) = y
(j)
1
and y
(j)
(x
2
) = y
(j)
2
for j = 0, (n 1). It follows the derivative relations
y
(j)
(x, ) = y
(j)
(x) +
(j)
(x) (3.6)
d
d
y
(j)
(x, ) =
(j)
(x) (3.7)
where superscript j (j = 0, n) in parentheses indicates the order of the derivative with respect to x. To
remind, here n is the highest order derivative of y that enters the functional F[y]. For many examples in physics
n = 1 (such as we will see in classical mechanics), and only the xed end points of y itself are required; however,
in electrostatics and quantum mechanics often higher order derivatives are involved, so we consider the more
general case.
If y(x) is the function that corresponds to an extremum of F[y], we expect that any inntesimal variation
(x) away from y that is sufciently smooth and subject to the xed-endpoint boundary conditions of Eq. 3.2)
will have zero effect (to rst order) on the extremal value. The arbitrary function (x) of course has been
dened to satisfy the differentiability and boundary conditions, and the scale factor allows a mechanism for
effecting inntesimal variations through a limiting procedure approaching = 0. At = 0, the varied function
y(x, ) is the extremal value y(x). Mathematically, this implies
d
d
[F[y +]]
=0
= 0
=
d
d
__
x
2
x
1
f(x, y(x, ), y

(x, ), ) dx
_
=0
(3.8)
=
_
x
2
x
1
__
f
y
_
y(x, )

+
_
f
y

_
y

(x, )

+
_
dx
=
_
x
2
x
1
__
f
y
_
(x) +
_
f
y

(x) +
_
dx
where we have used y(x, )/ = (x), y

(x, )/ =

(x), etc... If we integrate by parts the term in Eq. 3.8

involving

(x) we obtain
_
x
2
x
1
_
f
y

(x) dx =
_
f
y

_
(x)

x
2
x
1

_
x
2
x
1
d
dx
_
f
y

_
(x) dx
=
_
x
2
x
1
d
dx
_
f
y

_
(x) dx (3.9)
23
3.3. VARIATIONAL NOTATION CHAPTER 3. CALCULUS OF VARIATIONS
where we have used the fact that (x
1
) = (x
2
) = 0 to cause the boundary term to vanish. More generally, for
all the derivative terms in Eq. 3.8 we obtain
_
x
2
x
1
_
f
y
(j)
_

(j)
(x) dx = (1)
(j)
_
x
2
x
1
d
j
dx
j
_
f
y
(j)
_
(x) dx (3.10)
for j = 1, n where we have used the fact that
(j)
(x
1
) =
(j)
(x
2
) = 0 for j = 0, (n 1) to cause the
boundary terms to vanish. Substituting Eq. 3.10 in Eq. 3.8 gives
d
d
[F[y +]]
=0
=
_
x
2
x
1
__
f
y
_

d
dx
_
f
y

_
+
+(1)
n
d
n
dx
n
_
f
y
(n)
__
(x) dx
= 0 (3.11)
Since (x) is an arbitrary differential function subject to the boundary conditions of Eq. 3.2, the terms in brackets
must vanish. This leads to a generalized form of the Euler equation in one dimension for the extremal value of
F[y] of Eq. 3.4
_
f
y
_

d
dx
_
f
y

_
+ + (1)
n
d
n
dx
n
_
f
y
(n)
_
= 0 (3.12)
In other words, for the function y(x) to correspond to an extremal value of F[y], Eq. 3.12 must be satised for
all y(x); i.e., over the entire domain of x of y(x). Consequently, solution of Eq. 3.12 requires solving for an
entire function - not just a particular value of the function argument x as in function calculus. Eq. 3.12 is referred
to as the Euler equation (1-dimensional, in this case), and typically results in a differential equation, the solution
of which (subject to the boundary conditions already discussed) provides the function y(x) that produces the
extremal value of the functional F[y].
We next dene a more condensed notation, and derive several useful techniques such as algebraic manipu-
lation of functionals, functional derivatives, chain relations and Taylor expansions. We also explicitly link the
functional calculus back to the traditional function calculus in certain limits.
3.3 Variational Notation
We dene F[y], the variation of the functional F[y], as
F[y]
d
d
[F[y +]]
=0

=
_
x
2
x
1
__
f
y
_

d
dx
_
f
y

_
+ + (1)
n
d
n
dx
n
_
f
y
(n)
__
(x) dx
=
_
x
2
x
1
F
y(x)
y(x) dx (3.13)
where (analagously) y(x), the variation of the function y(x), is dened by
y(x)
d
d
[y(x) +(x)]
=0

= (x)
= y(x, ) y(x, 0) (3.14)
where we have again used the identity y(x, 0) = y(x). Relating the notation of the preceding section, we have
y(x, ) = y(x) +(x) = y(x) +y(x) (3.15)
24
CHAPTER 3. CALCULUS OF VARIATIONS 3.4. FUNCTIONAL DERIVATIVES: ELABORATION
Note that the operators and d/dx commute; i.e., that (d/dx) = (d/dx):
y

(x) =
dy
dx
=
d
d
_
d
dx
(y +)
_
=0
(3.16)
=
d
d
__
y

_
=0
=

=
d
dx
() =
d
dx
y
Hence the (rst) functional variation of F[y] can be written
F[y] =
_
x
2
x
1
F
y(x)
y(x) dx (3.17)
where
F
y(x)
is dened as the (rst) functional derivative of F[y] with respect to y at position x, and assuming
F[y] in the form of Eq. 3.1, is given by
F
y(x)
=
_
f
y
_

d
dx
_
f
y

_
+ + (1)
n
d
n
dx
n
_
f
y
(n)
_
(3.18)
Note that the functional derivative dened above is itself a function of x. In fact, Eq. 3.17 can be considered as a
generalization of an exact differential of a discreet multivariable function in calculus, .e.g.,
dF(x
1
, x
2
, , x
N
) =

i
F
x
i
dx
i
(3.19)
where the summation over discreet variables x
i
has been replaced by integration over a continuous set of variables
x.
An extremal value of F[y] is a solution of the stationary condition
F[y] = 0 (3.20)
and can be found by solution of the Euler equation 3.12.
3.4 Functional Derivatives: Elaboration
3.4.1 Algebraic Manipulations of Functional Derivatives
The functional derivative has properties analogous to a normal function derivative

y(x)
(c
1
F
1
+c
2
F
2
) = c
1
F
1
y(x)
+c
2
F
2
y(x)
(3.21)

y(x)
(F
1
F
2
) =
F
1
y(x)
F
2
+F
1
F
2
y(x)
(3.22)

y(x)
_
F
1
F
2
_
=
_
F
1
y(x)
F
2
F
1
F
2
y(x)
_
/F
2
2
(3.23)
25
3.4. FUNCTIONAL DERIVATIVES: ELABORATION CHAPTER 3. CALCULUS OF VARIATIONS
3.4.2 Generalization to Functionals of Higher Dimension
The expression for the functional derivative in Eq. 3.18 can be generalized to multidimensional functions in a
straight forward manner.
F
(x
1
, x
N
)
=
_
f

_
+
n

j=1
(1)
j
N

i=1

j
x
j
i
_
f

(j)
x
i
_
(3.24)
The functional derivative in the 3-dimensional case is
F
(r)
=
_
f

_
f

_
+
2
_
f

_
(3.25)
Example: Variational principle for a 1-particle quantum system.
Consider the energy of a 1-particle system in quantum mechanics subject to an external potential v(r). The
system is described by the 1-particle Hamiltonian operator

H =
h
2
2m

2
+v(r) (3.26)
The expectation value of the energy given a trial wave function

(r) is given by
E[

] =
_

(r)
_

h
2
2m

2
+v(r)
_

(r)d
3
r
_

(r)

(r)d
3
r
(3.27)
=
_

)
Let us see what equation results when we require that the energy is an extremal value with respect to the wave
function; i.e.,
E[]
(r)
= 0 (3.28)
Where we denote the wave function that produces this extremal value . Since this is a 1-particle system in
the absence of magnetic elds, there is no need to consider spin explicitly or to enforce antisymmetry of the
wave function as we would in a many-particle system. Moreover, there is no loss of generality if we restrict
the wave function to be real (one can double the effort in this example by considering the complex case, but
the manipulations are redundant, and it does not add to the instructive value of the variational technique - and
a students time is valuable!). Finally, note that we have constructed the energy functional E[

] to take on an
un-normalized wave function and return a correct energy (that is to say, the normalization is built into the energy
expression), alleviating the need to explicitly constrain the wave function to be normalized in the variational
process. We return to this point in a later example using the method of Lagrange multipliers.
Recall from Eq. 3.23 we have
E[]
(r)
=
_
_

_
(r)
[)
[)
(r)
_

_
_
_
/[)
2
(3.29)
Let us consider in more detail the rst functional derivative term,

_
(r)
=

(r)
_

(r)
_

h
2
2m

2
+v(r)
_

(r)d
3
r (3.30)
26
CHAPTER 3. CALCULUS OF VARIATIONS 3.4. FUNCTIONAL DERIVATIVES: ELABORATION
where we have dropped the complex conjugation since we consider the case where is real. It is clear that the
integrand, as written above, depends explicitly only on and
2
. Using Eq. 3.25 (the specic 3-dimensional
case of Eq. 3.24), we have that
f
(r)
=
h
2
2m

2
(r) + 2v(r)(r) (3.31)
_
f

2
(r)
_
=
_
h
2
2m
(r)
_
and thus

_
(r)
= 2
_
h
2
2m

2
+v(r)
_
(r) = 2

H(r) = 2

H[) (3.32)
where the last equality simply reverts back to Bra-Ket notation. Similarly, we have that
[)
(r)
= 2(r) = 2[) (3.33)
This gives the Euler equation
E[]
(r)
=
_
2

H[)[) 2[)
_

__
/[)
2
(3.34)
= 2
_
_

H[)
[)

[)
_

_
[)
2
_
_
= 0
Multiplying through by [), dividing by 2 and substituting in the expression for E[] above we obtain

H[) =
_

_
[)
[) = E[][) (3.35)
which is, of course, just the stationary-state Schr odinger equation.
3.4.3 Higher Order Functional Variations and Derivatives
Higher order functional variations and functional derivatives follow in a straight forward manner from the cor-
responding rst order denitions. The second order variation is
2
F = (F), and similarly for higher order
variations. Second and higher order functional derivatives are dened in an analogous way.
The solutions of the Euler equations are extremals - i.e., stationary points that correspond to maxima, minima
or saddle points (of some order). The nature of the stationary point can be discerned (perhaps not completely) by
consideration of the second functional variation. For a functional F[f], suppose f
0
is the function that solves the
Euler equation; i.e., that satises F[f] = 0 or equivalently [F[f]/f(x)]
f=f
0
= 0, then

2
F =
1
2
_ _
f(x)
_

2
F
f(x)f(x

)
_
f=f
0
f(x

) dxdx

(3.36)
The stationary point of F[f
0
] at f
0
can be characterized by

2
F 0 : minimum

2
F = 0 : saddle point (order undetermined)

2
F 0 : maximum
(3.37)
27
3.4. FUNCTIONAL DERIVATIVES: ELABORATION CHAPTER 3. CALCULUS OF VARIATIONS
Example: Second functional derivatives.
Consider the functional for the classical electrostatic energy
J[] =
1
2
_ _
(r)(r

)
[r r

[
d
3
rd
3
r

(3.38)
what is the second functional derivative
_

2
J[]
(r)(r

)
_
?
The rst functional derivative with respect to (r) is
J[]
(r)
=
1
2
_
(r

)
[r r

[
d
3
r

+
1
2
_
(r

)
[r r

[
d
3
r

(3.39)
=
_
(r

)
[r r

[
d
3
r

Note, r

is merely a dummy integration index - it could have been called x, y, r

, etc... The important feature

is that after integration what results is a function of r - the same r that is indicated by the functional derivative
J[]
(r)
.
The second functional derivative with respect to (r

) is
_

2
J[]
(r)(r

)
_
=

(r

)
J[]
(r)
(3.40)
=

(r

)
_
(r

)
[r r

[
d
3
r

=
1
[r r

[
3.4.4 Integral Taylor series expansions
An integral Taylor expansion for F[f
0
+ f] is dened as
F[f
0
+ f] = F[f
0
]
+

n=1
1
n!
_ _

_
_

(n)
F
f(x
1
)f(x
2
) f(x
n
)
_
f
0
f(x
1
)f(x
2
) f(x
n
) dx
1
dx
2
dx
n
(3.41)
For functionals of more than one function, e.g. F[f, g], mixed derivatives can be dened. Typically, for
sufciently well behaved functionals, the order of the functional derivative operations is not important (i.e., they
commute),

2
F
f(x)g(x

)
=

2
F
g(x)f(x

)
(3.42)
28
CHAPTER 3. CALCULUS OF VARIATIONS 3.4. FUNCTIONAL DERIVATIVES: ELABORATION
An integral Taylor expansion for F[f
0
+ f, g
0
+ g] is dened as
F[f
0
+ f, g
0
+ g] = F[f
0
, g
0
]
+
_ _
F
f(x)
_
f
0
,g
0
f(x) dx
+
_ _
F
g(x)
_
f
0
,g
0
g(x) dx
+
1
2
_ _
f(x)
_

2
F
f(x)f(x

)
_
f
0
,g
0
f(x

) dxdx

+
_ _
f(x)
_

2
F
f(x)g(x

)
_
f
0
,g
0
g(x

) dxdx

+
1
2
_ _
g(x)
_

2
F
g(x)g(x

)
_
f
0
,g
0
g(x

) dxdx

+ (3.43)
3.4.5 The chain relations for functional derivatives
From Eq. 3.13, the variation of a functional F[f] is given by
F =
_
F
f(x)
f(x) dx (3.44)
(where it is understood we have dropped the denite integral notation with endpoinds x
1
and x
2
- it is also valid
that the boundaries be at plus or minus innity). If at each point x, f(x) itself is a functional of another function
g, we write f = f[g(x), x] (an example is the electrostatic potential (r) which at every point r is a functional
of the charge density at all points), we have
f(x) =
_
f(x)
g(x

)
g(x

) dx (3.45)
which gives the integral chain relation
F =
_ _
F
f(x)
f(x)
g(x

)
g(x

) dxdx

=
_
F
g(x

)
g(x

) dx

(3.46)
hence,
F
g(x

)
=
_
F
f(x)
f(x)
g(x

)
dx (3.47)
Suppose F[f] is really an ordinary function (a special case of a functional); i.e. F = F(f), then written as a
functional
F(f(x)) =
_
F(f(x

))(x

x)dx

(3.48)
it follows that
F(f(x))
f(x

)
=
dF
df
(x

x) (3.49)
If we take F(f) = f itself, we see that
f(x)
f(x

)
= (x

x) (3.50)
29
3.5. HOMOGENEITY AND CONVEXITY CHAPTER 3. CALCULUS OF VARIATIONS
If instead we have a function that takes a functional argument, e.g. g = g(F[f(x)]), then we have
g
f(x)
=
_
g
F[f, x

]
F[f, x

]
f(x)
dx

=
_
dg
dF[f, x

]
(x

x)
F[f, x

]
f(x)
dx

=
dg
dF
F
f(x)
(3.51)
If the argument f of the functional F[f] contains a parameter ; i.e., f = f(x; ), then the derivative of F[f]
with respect to the parameter is given by
F[f(x; )]

=
_
F
f(x; )
f(x; )

dx (3.52)
3.4.6 Functional inverses
For ordinary function derivatives, the inverse is dened as (dF/df)
1
= df/dF such that (dF/df)
1
(df/dF) =
1, and hence is unique. In the case of functional derivatives, the relation between F[f] and f(x) is in general a
reduced dimensional mapping; i.e., the scalar F[f] is determined from many (often an innite number) values of
the function argument f(x).
Suppose we have the case where we have a function f(x), each value of which is itself a functional of another
function g(x

). Moreover, assume that this relation is invertable; i.e., that each value of g(x

) can be written as
a functional of f(x). A simple example would be if f(x) were a smooth function, and g(x

) was the Fourier

transform of f(x) (usually the x

would be called k or or something...). For this case we can write

f(x) =
_
f(x)
g(x

)
g(x

) dx

(3.53)
g(x

) =
_
g(x

)
f(x

)
f(x

) dx

(3.54)
which leads to
f(x) =
_ _
f(x)
g(x

)
g(x

)
f(x

)
f(x

) dx

dx

(3.55)
providing the reciprocal relation
_
f(x)
g(x

)
g(x

)
f(x

)
dx

=
f(x)
f(x

)
= (x x

) (3.56)
We now dene the inverse as
_
f(x)
g(x

)
_
1
=
g(x

)
f(x)
(3.57)
from which we obtain
_
f(x)
g(x

)
_
f(x)
g(x

)
_
1
dx

= (x x

) (3.58)
3.5 Homogeneity and convexity properties of functionals
In this section we dene two important properties, homogeneity and convexity, and discuss some of the powerful
consequences and inferences that can be ascribed to functions and functionals that have these properties.
30
CHAPTER 3. CALCULUS OF VARIATIONS 3.5. HOMOGENEITY AND CONVEXITY
3.5.1 Homogeneity properties of functions and functionals
A function f(x
1
, x
2
) is said to be homogeneous of degree k (in all of its degrees of freedom) if
f(x
1
, x
2
, ) =
k
f(x
1
, x
2
) (3.59)
and similarly, a functional F[f] is said to be homogeneous of degree k if
F[f] =
k
F[f] (3.60)
Thus homogeneity is a type of scaling relationship between the value of the function or functional with unscaled
arguments and the corresponding values with scaled arguments. If we differentiate Eq. 3.59 with respect to we
obtain for the term on the left-hand side
df(x
1
, x
2
, )
d
=
f(x
1
, x
2
, )
(x
i
)
d(x
i
)
d
(3.61)
=

i
x
i
f(x
1
, x
2
, )
(x
i
)
(3.62)
(where we note that d(x
i
)/d = x
i
), and for the term on the right-hand side
d
d
_

k
f(x
1
, x
2
, )
_
= k
k1
f(x
1
, x
2
, ) (3.63)
Setting = 1 and equating the left and right-hand sides we obtain the important relation

i
x
i
f(x
1
, x
2
, )
x
i
= kf(x
1
, x
2
, ) (3.64)
Similarly, for homogeneous functionals we can derive an analogous relation
_
F
f(x)
f(x) dx = kF[f] (3.65)
Sometimes these formulas are referred to as Eulers theorem for homogeneous functions/functionals. These
relations have the important conseuqnce that, for homogeneous functions (functionals), the value of the function
(functional) can be derived from knowledge only of the function (functional) derivative.
For example, in thermodynamics, extensive quantities (such as the Energy, Enthalpy, Gibbs free energy,
etc...) are homogeneous functionals of degree 1 in their extensive variables (like entropy, volume, the number of
particles, etc...). and intensive quantities (such as the pressure, etc...) are homogeneous functionals of degree 0.
Consider the energy as a function of entropy S, volume V , and particle number n
i
for each type of particle (i
represents a type of particle). Then we have that E = E(S, V, n
1
, n
2
) and
E = E(S, V, n
1
, n
2
) (3.66)
dE =
_
E
S
_
V,n
i
dS +
_
E
V
_
S,n
i
dV +

i
_
E
n
i
_
S,V,n
j=i
dn
i
(3.67)
= TdS pdV +

i
dn
i
where we have used the identities
_
E
S
_
V,n
i
= T,
_
E
V
_
S,n
i
= p, and
_
E
n
i
_
S,V,n
j=i
=
i
. From the rst-order
homogeneity of the extensive quantity E(S, V, n
1
, n
2
) we have that
E(S, V, n
1
, n
2
, ) = E(S, V, n
1
, n
2
) (3.68)
31
3.5. HOMOGENEITY AND CONVEXITY CHAPTER 3. CALCULUS OF VARIATIONS
The Euler theorem for rst order homogeneous functionals then gives
E =
_
E
S
_
V,n
i
S +
_
E
V
_
S,n
i
V +

i
_
E
n
i
_
S,V,n
j=i
n
i
= TS pV +

i
n
i
(3.69)
Taking the total differential of the above equation yields
dE = TdS +SdT pdV V dp +

i
dn
i
+n
i
d
i
Comparison of Eqs. 3.68 and 3.70 gives the well-known Gibbs-Duhem equation
SdT V dp +

i
n
i
d
i
= 0 (3.70)
Another example is the classical and quantum mechanical virial theorem that uses homogeneity to relate the
kinetic and potential energy. The virial theorem (for 1 particle) can be stated as
_
x
V
x
+y
V
y
+z
V
z
_
= 2 K) (3.71)
Note that the factor 2 arises from the fact that the kinetic energy is a homogeneous functional of degree 2 in the
particle coordinates. If the potential energy is a central potential; i.e., V (r) = C r
n
(a homogeneous functional
of degree n) we obtain
nV ) = 2 K) (3.72)
In the case of atoms (or molecules - if the above is generalized slightly) V (r) is the Coulomb potential 1/r,
n = 1 and we have < V >= 2 < K > or E = (1/2) < V > since E =< V > + < K >.
3.5.2 Convexity properties of functions and functionals
Powerful relations can be derived for functions and functionals that possess certain convexity properties. We
dene convexity in three cases, starting with the most general: 1) general functions (functionals), 2) at least
once-differentiable functions (functionals), and 3) at least twice-differentiable functions (functionals).
For a general function (functional) to be convex on the interval I (for functions) or the domain T (for func-
tionals) if, for 0 1
f(x
1
+ (1 )x
2
) f(x
1
) + (1 )f(x
2
) (3.73)
F[f
1
+ (1 )f
2
] F[f
1
] + (1 )F[f
2
] (3.74)
for x
1
, x
2
I and f
1
, f
2
T. f(x) (F[f]) is said to be strictly convex on the interval if the equality holds only
for x
1
= x
2
(f
1
= f
2
). f(x) (F[f]) is said to be concave (strictly concave) if f(x) (-F[f]) is convex (strictly
convex).
A once-differentiable function f(x) (or functional F[f]) is convex if and only if
f(x
1
) f(x
2
) f

(x
2
) (x
1
x
2
) 0 (3.75)
F[f
1
] F[f
2
]
_ _
F[f]
f(x)
_
f=f
2
(f
1
(x) f
2
(x)) dx 0 (3.76)
32
CHAPTER 3. CALCULUS OF VARIATIONS 3.5. HOMOGENEITY AND CONVEXITY
for x
1
, x
2
I and f
1
, f
2
T.
A twice-differentiable function f(x) (or functional F[f]) is convex if and only if
f

(x) 0 (3.77)
_

2
F[f]
f(x)f(x

)
_
0 (3.78)
for x I and f T.
An important property of convex functions (functionals) is known as Jensens inequality: For a convext
function f(x) (functional F[f])
f(x)) f(x)) (3.79)
F[f)] F[f]) (3.80)
where ) denotes an average (either discreet of continuous) over a positive semi-denite set of weights. In
fact, Jensens inequality can be extended to a convex function of a Hermitian operator
f
__

O
__

_
f(

O)
_
(3.81)
where in this case ) denotes the quantum mechanical expectation value [ [ ). Hence, Jensens inequal-
ity is valid for averages of the form
f) =

i
P
i
f
i
where P
i
0,

i
P
i
= 1 (3.82)
f) =
_
P(x)f(x) dx; where P(x) 0,
_
P(x) dx = 1 (3.83)
f) =
_

(x)

f(x) dx; where
_

(x)(x) dx = 1 (3.84)
The proof is elementary but I dont feel like typing it at 3:00 in the morning. Instead, here is an example of
an application - maybe in a later version...
Example: Convex functionals in statistical mechanics.
Consider the convect function e
x
(it is an innitely differentiable function, and has d
2
(e
x
)/dx
2
= e
x
, a positive denite second
derivative). Hence e
<x>
exp(x).
e
<x>
exp(x) (3.85)
This is a useful relation in statistical mechanics.
Similarly the function xln(x) is convex for x 0. If we consider two sets of probabilities Pi and P

i
such that Pi, P

i
0 and

i
Pi =

i
P

i
= 1, then we obtain from Jensens inequality
x ln(x) xln(x) (3.86)

i
P

i

Pi
P

ln

i
P

i

Pi
P

i
P

i

Pi
P

i
ln

Pi
P

(3.87)
Note that

N
i
P

i
Pi/P

i
=

N
i
Pi = 1 and hence the left hand side of the above inequality is zero since 1 ln(1) = 0. The right hand
side can be reduced by canceling out the P

i
factors in the numerator and denominator, which results in
N

i
Pi ln

Pi
P

0 (3.88)
which is a famous inequality derived by Gibbs, and is useful in providing a lower bound on the entropy: let P

i
= 1/N then, taking minus
the above equation, we get

i
Pi ln(Pi) ln(N) (3.89)
If the entropy is dened as kB

N
i
Pi ln(Pi) where kB is the Boltzmann constant (a positive quantity), then the largest value
the entropy can have (for an ensemble of N discreet states) is kB ln(N), which occurs when all probabilities are equal (the innite
temperature limit).
33
3.6. LAGRANGE MULTIPLIERS CHAPTER 3. CALCULUS OF VARIATIONS
3.6 Lagrange Multipliers
In this section, we outline an elegant method to introduce constraints into the variational procedure. We begin
with the case of a discreet constraint condition, and then outline the generalization to a continuous (pointwise)
set of constraints.
Consider the problem to extremize the functional F[f] subject to a functional constraint condition
G[f] = 0 (3.90)
In this case, the method of Lagrange multipliers can be used. We dene the auxiliary function
[f] F[f] G[f] (3.91)
where is a parameter that is yet to be determined. We then solve the variational condition

f(x)
=
F
f(x)

G
f(x)
= 0 (3.92)
Solution of the above equation results, in general, in a innite set of solutions depending on the continuous
parameter lambda. We then have the freedom to choose the particular value of that satises the constraint
requirement. Hopefully there exists such a value of , and that value is unique - but sometimes this is not the
case. Note that, if f
0
(x) is a solution of the constrained variational equation (Eq. 3.92), then
=
_
F
f(x)
_
f=f
0
/
_
G
f(x)
_
f=f
0
(3.93)
for any and all values of f
0
(x)! Often in chemistry and physics the Lagrange multiplier itself has a physical
meaning (interpretation), and is sometimes referred to as a sensitivity coefcient.
Constraints can be discreet, such as the above example, or continuous. A continuous (or pointwise) set of
constraints can be written
g[f, x] = 0 (3.94)
where the notation g[f, x] is used to represent a simultaneous functional of f and function of x - alternately
stated, g[f, x] is a functional of f at every point x. We desire to impose a continuous set of constraints at every
point x, and for this purpose, we require a Lagrange multiplier (x) that is itself a continuous function of x. We
then dene (similar to the discreet constraint case above) the auxiliary function
[f] F[f]
_
(x)g[f, x] dx (3.95)
and then solve

f(x)
=
F
f(x)

_
(x

)
g[f, x

]
f(x)
dx

= 0 (3.96)
As before, the Lagrange multiplier (x) is determined to satisy the constraint condition of Eq. 3.94. One can
consider (x) to be a continuous (innite) set of Lagrange multipliers associated with a constraint condition at
each point x.
34
CHAPTER 3. CALCULUS OF VARIATIONS 3.7. PROBLEMS
3.7 Problems
3.7.1 Problem 1
Consider the functional
T
W
[] =
1
8
_
(r) (r)
(r)
dddr (3.97)
a. Evaluate the functional derivative T
W
/(r).
b. Let (r) = [(r)[
2
(assume is real), and rewrite T[] = T
W
[[(r)[
2
]. What does this functional
represent in quantum mechanics?
c. Evaluate the functional derivative T[]/(r) directly and verify that it is identical to the functional
derivative obtained using the chain relation
T[]
(r)
=
T
W
[]
(r)
=
_
T
W
[]
(r

)

(r

)
(r)
d
3
r

3.7.2 Problem 2
Consider the functional
E
x
[] =
_

4/3
(r)
_
c b
2
x
3/2
(r)
_
d
3
r (3.98)
where c and b are constants, and
x(r) =
[(r)[

4/3
(r)
(3.99)
evaluate the functional derivative
Ex[]
(r)
.
3.7.3 Problem 3
This problem is an illustrative example of variational calculus, vector calculus and linear algebra surrounding an
important area of physics: classical electrostatics.
The classical electrostatic energy of a charge distribution (r) is given by
J[] =
1
2
_ _
(r)(r

)
[r r

[
d
3
rd
3
r

(3.100)
This is an example of a homogeneous functional of degree 2, that is to say J[] =
2
J[].
3.7.3.1 Part A
Show that
J[] =
1
k
_ _
J
(r)
_

(r)d
3
r (3.101)
where k = 2. Show that the quantity
_
J
(r)
_

= (r) where (r) is the electrostatic potential due to the charge

distribution (r).
35
3.7. PROBLEMS CHAPTER 3. CALCULUS OF VARIATIONS
3.7.3.2 Part B
The charge density (r) and the the electrostatic potential (r) are related by formula derived above. With
appropriate boundary conditions, an equivalent condition relating (r) and (r) is given by the Poisson equation

2
(r) = 4(r)
Using the Poisson equation and Eq. 3.101 above, write a new functional W
1
[] for the electrostatic energy with
as the argument instead of and calculate the functional derivative
W
1
(r)
.
3.7.3.3 Part C
Rewrite W
1
[] above in terms of the electrostatic (time independant) eld E = assuming that the quantity
(r)(r) vanishes at the boundary (e.g., at [r[ = ). Denote this new functional W
2
[]. W
2
[] should have
no explicit dependence on itself, only through terms involving .
3.7.3.4 Part D
Use the results of the Part C to show that
J[] = W
1
[] = W
2
[] 0
for any and connected by the Poisson equation and subject to the boundary conditions described in Part C.
Note: (r) can be either positive OR negative or zero at different r.
3.7.3.5 Part E
Show explicitly
W
1
(r)
=
W
2
(r)
3.7.4 Problem 4
This problem is a continuation of problem 3.
3.7.4.1 Part F
Perform an integral Taylor expansion of J[] about the reference charge density
0
(r). Let (r) = (r)

0
(r). Similarly, let (r),
0
(r) and (r) be the electrostatic potentials associated with (r),
0
(r) and (r),
respectively.
Write out the Taylor expansion to innite order. This is not an innite problem! At what order does the
Taylor expansion become exact for any
0
(r) that is sufciently smooth?
3.7.4.2 Part G
Suppose you have a density (r), but you do not know the associated electrostatic potential (r). In other words,
for some reason it is not convenient to calculate (r) via
(r) =
_
(r

)
[r r

[
d
3
r
However, suppose you know a density
0
(r) that closely resembles (r), and for which you do know the asso-
ciated electrostatic potential
0
(r). So the knowns are (r),
0
(r) and
0
(r) (but NOT (r)!) and the goal is
to approximate J[] in the best way possible from the knowns. Use the results of Part F to come up with a new
functional W
3
[,
0
,
0
] for the approximate electrostatic energy in terms of the known quantities.
36
CHAPTER 3. CALCULUS OF VARIATIONS 3.7. PROBLEMS
3.7.4.3 Part H
Consider the functional
U

] =
_
(r)

(r)d
3
r +
1
8
_

(r)
2

(r)d
3
r (3.102)
where

(r) is not necessarily the electrostatic potential corresponding to (r), but rather a trial function indepen-
dent of (r). Show that
U

(r)
= 0
leads to the Poisson equation; i.e., the

(r) that produces an extremum of U

] is, in fact, the electrostatic

potential (r) corresponding to (r).
Note: we could also have written the functional U

] in terms of the trial density (r) as

U

[ ] =
_ _
(r) (r

)
[r r

[
d
3
rd
3
r

1
2
_ _
(r) (r

)
[r r

[
d
3
rd
3
r

(3.103)
withe variational condition
U

[ ]
(r)
= 0
3.7.4.4 Part I
Show that U

[
0
] and U

[
0
] of Part H are equivalent to the expression for W
3
[,
0
,
0
] in Part G. In other words,
the functional W
3
[,
0
,
0
] shows you how to obtain the best electrostatic energy approximation for J[]
given a reference density
0
(r) for which the electrostatic potential
0
(r) is known. The variational condition
U

(r)
= 0 or
U[ ]
(r)
= 0 of Part H provides a prescription for obtaining the best possible model density and
model potential

(r). This is really useful!!
3.7.4.5 Part J
We now turn toward casting the variational principle in Part H into linear-algebraic form. We rst expand the
trial density (r) as
(r) =
N
f

k
c
k

k
(r)
where the
k
(r) are just a set of N
f
analytic functions (say Gaussians, for example) for which it assumed we can
solve or in some other way conveniently obtain the matrix elements
A
i,j
=
_ _

i
(r)
j
(r

)
[r r

[
d
3
rd
3
r

and
b
i
=
_ _
(r)
i
(r

)
[r r

[
d
3
rd
3
r

so A is an N
f
N
f
square, symmetric matrix and b is an N
f
1 column vector. Rewrite U

[ ] of Part I as a
matrix equation U

[c] involving the Amatrix and b vector dened above. Solve the equation
U

[c]
c
= 0
for the coefcient vector c to give the best model density (r) (in terms of the electrostatic energy).
37
3.7. PROBLEMS CHAPTER 3. CALCULUS OF VARIATIONS
3.7.4.6 Part K
Repeat the excercise in Part J with an additional constraint that the model density (r) have the same normaliza-
tion as the real density (r); i.e., that
_
(r)d
3
r =
_
(r)d
3
r = N (3.104)
or in vector form
c
T
d = N (3.105)
where d
i
=
_

i
(r)d
3
r. In other words, solve

_
U

[c] (c
T
d N)
_
= 0
for c

() in terms of the parameter , and determine what value of the Lagrange multiplier satises the
constraint condition of Eq. 3.105.
3.7.4.7 Part L
In Part J you were asked to solve an unconstrained variational equation, and in Part K you were asked to solve
for the more general case of a variation with a single constraint. 1) Show that the general solution of c

()
(the superscript indicates that c

() variational solution and not just an arbitrary vector) of Part K reduces to

the unconstrained solution of Part J for a particular value of (which value?). 2) Express c

() as c

() =
c

(0) + c

() where c

() is the unconstrained variational solution c

(0) when the constraint condition is

turned on. Show that indeed, U

[c

(0)] U

[c

()]. Note that this implies the extremum condition corresponds

to a maximum. 3) Suppose that the density (r) you wish to model by (r) can be represented by
(r) =

k
x
k

k
(r) (3.106)
where the functions
k
(r) are the same functions that were used to expand (r). Explicitly solve for c

(0), and
c

() for this particular (r).

3.7.4.8 Part M
In Part J you were asked to solve an unconstrained variational equation, and in Part K you were asked to solve
for the more general case of a variation with a single constraint. You guessed it - now we generalize the solution
to an arbitrary number of constraints (so long as the number of constraints N
c
does not exceed the number of
variational degrees of freedom N
f
- which we henceforth will assume). For example, we initially considered a
single normalization constraint that the model density (r) integrate to the same number as the reference density
(r), in other words
_
(r)d
3
r =
_
(r)d
3
r N
This constraint is a specic case of more general form of linear constraint
_
(r)f
n
(r)d
3
r =
_
(r)f
n
(r)d
3
r y
n
For example, if f
1
(r) = 1 we recover the original constraint condition with y
1
= N, which was that the monopole
moment of equal that of . As further example, if f
2
(r) = x, f
3
(r) = y, and f
4
(r) = z, then we would also
require that each component of the dipole moment of equal those of , and in general, if f
n
(r) = r
l
Y
lm
(r)
38
CHAPTER 3. CALCULUS OF VARIATIONS 3.7. PROBLEMS
where the functions Y
lm
are spherical harmonics, we could constrain an arbitrary number of multipole moments
to be identical.
This set of N
c
constraint conditions can be written in matrix form as
D
T
c = y
where the matrix Dis an N
f
N
c
matrix dened as
D
i,j
=
_

i
(r)f
j
(r)d
3
r
and the y is an N
c
1 column vector dened by
y
j
=
_
(r)f
j
(r)d
3
r
Solve the general constrained variation

_
U

[c]
T
(D
T
c y)
_
= 0
for the coefcients c

() and the N
c
1 vector of Lagrange multipliers . Verify that 1) if = 0 one recovers
the unconstrained solution of Part J, and 2) = 1 (where 1 is just a vector of 1s) recovers the solution for the
single constraint condition of Part K.
39
Chapter 4
Elementary Principles of Classical Mechanics
4.1 Mechanics of a system of particles
4.1.1 Newtons laws
1. Every object in a state of uniform motion tends to remain in that state unless acted on by an external force.
v
i
= r
i
=
d
dt
r
i
(4.1)
p
i
= m
i
r
i
= m
i
v
i
(4.2)
2. The force is equal to the change in momentum per change in time.
F
i
=
d
dt
p
i
p
i
(4.3)
3. For every action there is an equal and opposite reaction.
F
ij
= F
ij
(weak law), not always true
e.g. F
ij
=
i
V (r
ij
), e.g. Biot-Savart law moving e

F
ij
= f
ij
r
ij
= F
ij
= f
ji
r
ji
(strong law)
e.g. F
ij
=
i
V ([r
ij
[), e.g. Central force problem
40
CHAPTER 4. CLASSICAL MECHANICS 4.1. MECHANICS OF A SYSTEM OF PARTICLES
4.1.2 Fundamental denitions
v
i
=
d
dt
r
i
r
i
(4.4)
a
i
=
d
dt
v
i
=
d
2
dt
2
r
i
= r
i
(4.5)
p
i
=m
i
v
i
= m
i
r
i
(4.6)
L
i
=r
i
p
i
(4.7)
N
i
=r
i
F
i
=

L
i
(4.8)
F
i
= p
i
=
d
dt
(m
i
r
i
) (4.9)
N
i
= r
i
p
i
(4.10)
= r
i

d
dt
(m
i
r
i
)
=
d
dt
(r
i
p
i
)
=
d
dt
L
i
=

L
i
Proof:
d
dt
(r
i
p
i
) = r
i
p
i
+r
i
p
i
(4.11)
=v
i
mv
i
+r
i
p
i
=0 +r
i
p
i
If
d
dt
A(t) = 0, At =constant, and A is conserved.
A conservative force eld (or system) is one for which the work required to move a particle in a closed
loop vanishes.
_
F ds = 0 note: F = V (r), then
_
c
V (r) ds =
_
s
(V (r)
. .

) nda (4.12)
_
c
A ds =
_
s
(A) nda (Stokes Theorem) (4.13)
System of particles m
i
,= m
i
(t)
Fixed mass, strong law of a and r on internal forces.
Suppose F
i
=

j
F
ji
+F
(e)
i
and f
ij
= f
ji
F
ij
= 0
41
4.1. MECHANICS OF A SYSTEM OF PARTICLES CHAPTER 4. CLASSICAL MECHANICS
r
ij
F
ij
= r
ji
F
ji
= 0 (4.14)
Strong law on F
ij
F
i
= P
i
=
d
dt
_
m
i
d
dt
r
i
_
(4.15)
=
d
2
dt
2
m
i
r
i
, for m
i
,= m
i
(t)
let R =

i
m
i
r
i
M
and M =

i
m
i

i
F
i
=
d
2
dt
2

i
m
i
r
i
(4.16)
=M
d
2
dt
2

i
m
i
r
i
M
=M
d
2
R
dt
2
=

i
F
(e)
i
+

j
F
ij
Note:

i

j
F
ij
=

j>i
(F
ij
+F
ji
) = 0

P M
d
2
R
dt
2
=

i
F
(e)
i
F
(e)
(4.17)

P =

i
M
dr
i
dt
(4.18)
=
d
dt
P
N
i
=r
i
F
i
(4.19)
=r
i

P
i
=

L
i
N
i
=

i
N
i
=

L
i
=

L (4.20)
=

i
r
i
F
i
=

i
r
i

_
_

j
F
ji
+F
(e)
i
_
_
=

ij
r
i
F
ji
+

i
r
i
F
(e)
i
=

i
r
i
F
(e)
i
= N
(e)
=

L
i
42
CHAPTER 4. CLASSICAL MECHANICS 4.1. MECHANICS OF A SYSTEM OF PARTICLES

j<i
r
i
F
ij
+

j>i
r
i
F
ji
= 0 (4.21)

j<i
r
i
F
ij
=

j>i
r
i
F
ji
Note, although
P =

i
P
i
(4.22)
=

i
m
i
d
dt
r
i
=m
d
dt
(4.23)
V
dR
dt
,=

i
v
i
=

i
dr
i
dt
L =

i
L
i
(4.24)
=

i
r
i
P
i
r
i
=(r
i
R) +R = r

i
+R (4.25)
r
i
= r

i
+

R (4.26)
p
i
=p

i
+m
i
V (4.27)
(V
i
=V

i
+V) (4.28)
L =

i
r

i
P

i
+

i
R
i
P

i
+

i
r

i
m
i
V
i
+

i
R
i
m
i
V
i
(4.29)
Note

i
m
i
r

i
=

i
m
i
r
i

i
m
i
R (4.30)
=MRMR = 0
hence

i
r
i
m
i
V =
_

i
m
i
r
i
_
V = 0 (4.31)

i
RP

i
=

i
Rm
i
V

i
(4.32)
=R
d
dt
_

i
m
i
r

i
_
=0
43
4.1. MECHANICS OF A SYSTEM OF PARTICLES CHAPTER 4. CLASSICAL MECHANICS
L =

i
r

i
P

i
. .
L about C.O.M.
+ RMV
. .
L of C.O.M.
(4.33)
For a system of particles obeying Newtons equations of motion, the work done by the system equals the
difference in kinetic energy.
F
i
= m
i

V
i
, dr
i
=
dr
i
dt
dt (4.34)
W
12
=

i
_
2
1
F
i
dr
i
(4.35)
=

i
_
2
1
m
i

V
i
V
i
dt
=

i
_
2
1
1
2
m
i
d
dt
(V
i
V
i
)dt
=

i
1
2
m
i
V
i
V
i

2
1
=
1
2
m
i
V
2
i

2
1
= T
2
T
1
The kinetic energy can be expressed as a kinetic energy of the center of mass and a T of the particles relative to
the center of mass.
T =
1
2

i
m
i
(V+V

i
) (V+V

i
) (4.36)
=
1
2

i
m
i
V
2
. .
1
2
MV
2
+

i
m
i
V

i
V
. .
(

i
m
i
V

i
)V
+
1
2

i
m
i
(V

i
)
2
. .
internal
Note

i
m
i
V

i
=

i
m
i
V
i
MV = 0
Proof

i
m
i
V

i
=
d
dt

i
m
i
r
i
(4.37)
=M
d
dt

i
m
i
r
i
M
=M
dR
dt
=MV
W
12
= T
2
T
1
=

i
_
2
1
F
i
dr
i
(4.38)
In the special case that
F
i
= F
(e)
i
+

j
F
ij
where F
(e)
i
=
i
V
i
and F
ij
=
i
V
ij
([r
i
r
j
[)
44
CHAPTER 4. CLASSICAL MECHANICS 4.1. MECHANICS OF A SYSTEM OF PARTICLES
Note

i
V
ij
(r
ij
) =
j
V
i
, V
ij
(r
ij
) (4.39)
=V
ji
(r
ij
)
=
1
r
ij
_
dV
dr
ij
_
(r
i
r
j
)
=F
ji
W
12
=T
2
T
1
(4.40)
=

i
_
2
1

i
V
i
dr
i
+

j
_
2
1

i
V
ij
(r
ij
) fr
i
Note
_
2
1

i
V
i
dr
i
=
_
2
1
(dr
i

i
)V
i
(4.41)
=
_
2
1
dV
i
= V
i

2
1
where
dr
i

i
= dx
d
dx
+dy
d
dy
+dz
d
dz
(4.42)

j
A
ij
=

j<i
A
ij
+

j>i
A
ji
+

i
A
ii
(4.43)

j
_
2
1

i
V
ij
r
ij
dr
i
=

j<i
_
2
1

i
V
ij
(r
ij
) dr
i

j
V
ji
(r
ij
) dr
j
(4.44)
=

j<i
_
2
1

ij
V
ij
(r
ij
)dr
ij
=

j<i
_
2
1
(dr
ij

ij
)V
ij
=

j<i
_
2
1
dV
ij
=

j<i
V
ij

2
1
=
1
2

j
V
ij

2
1
Hence we can dene a potential energy
V =

i
V
i
. .
external
+
1
2

ij
V
ij
. .
internal
(4.45)
45
4.2. CONSTRAINTS CHAPTER 4. CLASSICAL MECHANICS
W
12
=T
2
T
1
(4.46)
=(V
2
V
1
)
=V
1
V
2
4.2 Constraints
Constraints are scleronomous if they do not depend explicitly on time, and are rheonomous if they do.
(fr
1
, r
2
, . . ., r
N
, t)= 0 (holonomic)
e.g. rigid body (r
i
r
j
)
2
= C
2
ij
Nonholonomic: r
2
a
2
0, e.g. container boundary
Nonholonomic constaints cannot be used to eliminate dependent variables.
r
1
. . . r
N
q
1
. . . q
NK
; q
NK+1
. . . q
N
For holonomic systems with applied forces derivable from a scalar potential with workless constraints, a La-
grangian can always be dened.
Constraints are articial...name one that is not...?
r
i
(q
1
, q
2
. . . q
NK
, t; q
NK+1
, q
N
)
4.3 DAlemberts principle
F
i
= F
(a)
i
+f
i
F
(a)
i
= applied force
f
i
= constraint force
r
i
(t) = virtual (inntesimal) displacement, so small that F
i
does not change, consistent with the forces and
constraints at the time t.
We consider only constraint forces f
i
that do no net virtual work on the system (

i
f
i
v
i
= 0) since:
W
12
=
_
2
1
f
i
dr
i
(4.47)
=
_
2
1
f
i
v
i
dt
In other words, inntesimally,

i
f
i
r
i
= 0.
Not all constraint forces obey this condition, like a frictional force due to sliding on a surface.
So at equilibrium, F
i
= 0, thus

i
F
i
r
i
=

i
F
(a)
i
r
i
+

i
f
i
r
i
(4.48)
=

i
F
(a)
i
r
i
46
CHAPTER 4. CLASSICAL MECHANICS 4.3. DALEMBERTS PRINCIPLE
The principle of virtual work for a system at equilibrium:

i
F
(a)
i
r
i
= 0 (4.49)
We want to derive Lagranges equations for a set of generalized corrdinates that can be used to eliminate a
set of holonomic constraints. Here we go. . . we start with Newtons equations (valid for Cartesian Coordinates)
F
i
=

P
i
, or (F
i

P
i
) = 0

i
(F
i

P
i
) r
i
= 0 (4.50)
Recall r
i
is not an independent variation-it must obey the constraints.
DAlemberts principle:

i
_
F
(a)
i


P
i
_
r
i
+

i
f
i
r
i
=

i
_
F
(a)
i


P
i
_
r
i
= 0 (4.51)
(for constant forces that do no net virtual work)
Not useful yet. . .
r
i
= r
i
(q
1
, q
2
, . . . , q
N
, t)
v
i
=
d
dt
r
i
(4.52)
=

j
r
i
q
j
dq
j
dt
+
r
i
t
=

j
r
i
q
j
q
j
+
r
i
t
r
i
=

j
r
i
q
j
q
j
(4.53)

i
F
i
r
i
=

j
F
i

r
i
q
j
(4.54)
=

j
Q
j
q
j
Where Q
j

i
F
i

r
i
q
j
= generalized force.

P
i
r
i
=

i
m
i
r
i
(4.55)
=

i,j
m
i
r
i

r
i
q
j
q
j
47
4.3. DALEMBERTS PRINCIPLE CHAPTER 4. CLASSICAL MECHANICS
Note:

i
d
dt
_
m
i
r
i

r
i
q
j
_
=

i
m
i
r
i

r
i
q
j
+m
i
r
i

d
dt
_
r
i
q
j
_
and,
d
dt
_
r
i
q
j
_
=
v
i
q
j
; recall
v
i
q
j
=
r
i
q
j

P
i
r
i
=

i , j
_

_
d
dt
_
m
i
v
i

v
i
q
j
_
. .

q
j
(
1
2
m
i
v
2
i
)
m
i
v
i

v
i
q
j
. .

q
(
1
2
m
i
v
2
i
)
_

_
q
j
(4.56)
=

j
_
d
dt
_
T
q
j
_

T
q
_
q
j
So DAlemberts principle can be written:

i
_
F
(a)
i


P
i
_
r
i
=

j
_
d
dt
_
T
q
j
_

T
q
j
Q
j
_
q
j
(4.57)
Note: the only restriction on the constraints is that the net virtual work of the constraint forces vanish. (In-
cludes some non-holonomic constraints!)
In the case of holonomic constraints, it is possible to nd a set of independent coordinates that obey the
constraint conditions.
r
i
= r
i
(q
1
, . . . q
N
, t)
F
i
=
i
V (r
1
, . . . r
N
)

q
j
=

j
r
i
q
j

r
i
(4.58)
=

i
r
i
q
j

i
Generalized forces:
Q
j
=

i
F
i

r
i
q
j
(4.59)
=

i
r
i
q
j

i
V
=
V
q
j
d
dt
_
T
q
j
_

q
j
(T V ) = 0 (4.60)
If V = V (r
1
, . . . r
N
) as above, then:
V
q
j
= 0
48
CHAPTER 4. CLASSICAL MECHANICS 4.4. VELOCITY-DEPENDENT POTENTIALS
and
d
dt
_
L
q
j
_

L
q
j
= 0 (4.61)
where L(q, q, t) = T( q, q) V (q, t)
Note:
L

(q, q, t) = L(q, q, t) +
d
dt
F(q, t) (4.62)
d
dt
_
L

q
i
_

q
i
=
d
dt
_

q
i
_
L +
dF
dt
__

q
i
_
L +
dF
dt
_
(4.63)
d
dt
F(q, t) =

i
q
i
F
q
i
+
F
t
(4.64)

q
i
_
dF
dt
_
=
F
q
i
_

q
i
_
d
dt
=
d
dt
_

q
i
_
d
dt
_
L
q
i
_
+
d
dt
_
F
q
i
_

L
q
i

_

q
i
_
d
dt
F =
d
dt
_
L
q
i
_

L
q
i
(4.65)
4.4 Velocity-dependent potentials and dissipation functions
Recall:
Q
j
=

i
f
i

r
i
q
i
(4.66)
if F
i
=
i
v(r
1
, . . . r
N
), and Q
j
=
v
q
j
If there is a U(q, q, t) s.t.
Q
j
=
U
q
j
+
d
dt
_
U
q
j
_
(4.67)
Then, clearly DAlemberts principle:
d
dt
_
T
q
j
_

q
j
T = Q
j
=
d
dt
_
U
q
j
_

U
q
j
(4.68)
or with L = T U
d
dt
_
L
q
j
_

L
q
j
= 0
U = generalized potential
F =q[E+ (v B)] (4.69)
=q[
dA
dt
+(A v)]
E(r, t) = (r, t)
A
t
(4.70)
49
4.5. FRICTIONAL FORCES CHAPTER 4. CLASSICAL MECHANICS
B(r, t) = A(r, t) (4.71)
F =U +
d
dt
_
U
v
_
(4.72)
=q[E+ (v B)]
U(r, t) = q[ A v] (4.73)
Proof:
U(r, t) = q[ +(A v)] (4.74)
d
dt
_
U
v
_
= q
dA
dt
Where:
(A v) = (A )v + (v )A+A(v) +v (A)
Note:
r = 0, so v = 0 also,

x
(v) =
d
dt

x
(r) (4.75)
=
d
dt
x
=0
(A v) = (v )A+v (A) = (v )A+v B
Note:
dA
dt
=
A
t
+ (v )A (4.76)
=
_
dA
dt

A
t
_
+v B
F =U +
d
dt
_
U
v
_
(4.77)
=q
_
+
_
dA
dt

A
t
_
+v B
dA
dt
_
=q[E+v B]
4.5 Frictional forces
F =
1
2

i
v
i

k v
i
(4.78)
=
1
2

i
(k
x
v
2
ix
+k
y
v
2
iy
+k
z
v
2
iz
)
50
CHAPTER 4. CLASSICAL MECHANICS 4.5. FRICTIONAL FORCES
Suppose a force cannot be derived from a scalar potential U(q, q, t) by the prescription: Q
j
=
U
q
j
+
d
dt
_
U
q
j
_
for example, the frictional force F
fix
= k
x
v
ix
. In this case we cannot use Lagranges equations in
their form, but rather in the form:
d
dt
_
L
q
j
_

L
q
j
= Q
j

_
d
dt
_
U
q
j
_

U
q
j
_
= Q
fi
(4.79)
In the case of a frictional force, Q
fi
can be derived F
fi
=
v
i
F
where F =
1
2

i
(k
x
v
2
ix
+k
y
v
2
iy
+k
z
v
2
iz
) (Rayleighs dissipation function)
Work done on the system = -work done by the system.
Note:
W
f
=
_
F
f
dr (4.80)
=
_
F v
. .
2F
dt
=
_
dW
f
dt
dt
(4.81)
2F = the rate of energy dissipation due to friction.
Transforming to generalized coordinates:
recall:
r
i
q
j
=
v
i
q
j
Q
j
=

i
F
fi

r
i
q
j
(4.82)
=

v
i
F
r
i
q
j
=

v
i
q
j

v
i
F
=
F
q
j
Q
j
=
F
q
j
= the generalized force arising from friction
Derived from F
fi
=
v
i
F
The Lagrange equations become:
d
dt
_
L
q
i
_

L
q
i
=
F
q
j
(4.83)
Transformation of the kinetic energy to generalized coordinates
51
4.5. FRICTIONAL FORCES CHAPTER 4. CLASSICAL MECHANICS
T =

i
1
2
m
i
v
2
i
(4.84)
=

i
1
2
m
i
_
_

j
r
i
q
j
q
j
+
r
i
t
_
_
2
=m
o
+

j
m
j
q
j
+
1
2

j,k
m
jk
q
j
q
k
=T
o
+T
1
+T
2
m
o
=

i
1
2
m
i
_
r
i
t
_
2
(4.85)
m
j
=

i
m
i
r
i
t

r
i
q
j
(4.86)
m
jk
=

i
m
i
r
i
q
j

r
i
q
k
(4.87)
m
o
and m
j
= 0, if
r
i
t
= 0
Problem 23 of Goldstein
L =T V (4.88)
=
1
2
mv
2
z
mgz (4.89)
F =
1
2
kv
2
z
(4.90)
Q
z
=
F
v
z
(4.91)
=kv
z
Lagranges equation:
d
dt
_
L
v
z
_
kv
z
(4.92)
d
dt
(mv
z
) +mg = kv
z
or
dv
z
(t)
dt
+
_
k
m
_
v
z
(t) = g
This is of the form:
dy(t)
dt
+p(t)y(t) = q(t)
52
CHAPTER 4. CLASSICAL MECHANICS 4.5. FRICTIONAL FORCES
or
d
dt
[(t)y(t)] = (t)
where:
(t) = e

t
p(t

)dt

d
dt
= (t)p(t)
In our case:
(t) =e

t k
m
t

dt

(4.93)
=e
k
m
t
d
dt
_
e
k
m
t
V
z
(t)
_
=
_
dV
z
(t)
dt
+
k
m
V
z
(t)
_
e
k
m
t
(4.94)
=ge
k
m
t
e
k
m
t
V
z
(t) =
_
t _
ge
k
m
t

_
dt

(4.95)
=
_
mg
k
_
e
k
m
t
+c
V
z
(t) =
mg
k
+ce

k
m
t
(4.96)
=
mg
k
_
1 e

k
m
t
_
V
z
(0) =
mg
k
+c (4.97)
=0
c =
mg
k
(4.98)
dV
z
dt
=0 t

=
V

z
= lim
t
V
z
(t) (4.99)
=
mg
k
lim
x0
1 e
xt
x
= lim
x0
te
x
x =
k
m
=t
Note:
lim
k
m
0
V
z
(t) = gt (4.100)
53
Chapter 5
Variational Principles
Particle sliding down a movable wedge (frictionless)
x
m
= s cos +x x
m
= s cos + x
y
m
= sin y
m
= s sin
x
M
= x x
M
= x
y
M
= 0 y
M
= 0
T =
1
2
m( x
2
m
+ y
2
m
) +
1
2
M( x
2
M
+ y
2
M
) (5.1)
=
1
2
m( s
2
+ 2 x s cos + x
2
) +
1
2
M x
2
V =mgy (5.2)
=mg sin
L =
1
2
m( s
2
+ 2 x s cos + x
2
) +
1
2
M x
2
+mgs sin (5.3)
d
dt
_
L
s
_

L
s
= 0
d
dt
_
L
x
_

L
x
= 0
L
s
= mg sin
L
x
= 0
L
s
= m s +m x cos
L
x
= (m+M) x +m s cos
d
dt
[m s +m x cos ] = mg sin
d
dt
[(m+M) x +m s cos
. .
constant of motion
] = 0
1. s(m) + x(m cos ) = mg sin
2. s(m cos ) + x(m+M) = 0
54
CHAPTER 5. VARIATIONAL PRINCIPLES 5.1. HAMILTONS PRINCIPLE
(1.)
(2.)
cos
gives:
x
_
m cos
(m+M)
cos
_
= mg sin (5.4)
or
x =mg sin
_
cos
mcos
2
(m+M)
_
(5.5)
=A
x
x(t) = x
o
+ x
o
t +
1
2
A
x
t
2
(5.6)
(1.)
mcos
(m+M)
(2.) gives:
s
_
m
(m cos )
2
(m+M)
_
= mg sin (5.7)
or
s =mg sin
_
(m+M)
m(m+M) m
2
cos
2

_
(5.8)
=A
s
s(t) = s
o
+ s
o
t +
1
2
A
s
t
2
(5.9)
5.1 Hamiltons Principle
A monogenic system is one where all forces (except for those of constraint) are derivable from a generalized
scalar potential U(q, q, t).
If the scalar potential is only a function of the coordinates, U(q), a monogenic system is also conservative.
Hamiltons Principle (The principle of least action)
Fundamental postulate of classical mechanics for monogenic systems under holonomic constraints to replace
Newtons equations of motion.
For monogenic systems, the classical motion of a system (i.e., its path through phase space) between time t
a
and t
b
is the one for which the action integral:
S(a, b) =
_
t
b
ta
L(q, q, t)dt (5.10)
L = T U (5.11)
has a stationary value, that is to say S = 0, which leads to the Lagrange equations:
d
dt
_
L
q
_

L
q
= 0, and the
generalized forces are:
Q
j
=
d
dt
_
U
q
_

U
q
(5.12)
55
5.2. COMMENTS ABOUT HAMILTONS PRINCIPLE CHAPTER 5. VARIATIONAL PRINCIPLES
5.2 Comments about Hamiltons Principle
S = 0 poses no restrictions to the particular set of generalized coordinates used to represent the motion of the
system, and therefore are automatical, invariant to transformations

between sets of generalized coordinates.

Transformations with det(J) ,= 0, i.e. that span the same space.

Note: Hamiltons Principle takes the form of an un-constrained variation of the action integral.
If the system constraints are holonomic, i.e. can be written as :
f

(q
1
, q
2
, . . . q
N
) = 0
= 1, . . . K
Then a set of generalized coordinates can always be found, q

1
, q

2
. . . q

NK
, that satisfy the constraint conditions,
in which case Hamiltons Principle is both a necessary and sufcient condition for Lagranges equations.
If nonholonomic constraints are present such that a set of generalized coordinates cannot be dened that
satisfy the constraints, sometimes these constraints can be introduced through a constrained variation of the
action integral with the method of Lagrange multipliers.
Suppose we consider a more general form of constraint, formally a type of nonholonomic constraint, called
semi-holonomic:
f

(q
1
, q
2
, . . . q
N
, q
1
, q
2
, . . . q
N
, t) = 0
= 1, . . . K
Consider the constrained variation of the action:

_
_
L(q, q, t) +
k

<1

(q, q, t)f

(q, q)
_
dt = 0 (5.13)
d
dt
_
L
q
k
_

L
q
k
+
d
dt
_

q
k
_

__

q
k
_

_
= 0 (5.14)
or
Q

k
=
d
dt
_
L
q
k
_

L
q
k
(5.15)
where
Q

k
=

q
k
_

d
dt
_

q
k
_

__
(5.16)
if the generalized force of constraint, in terms of the yet-to-be-determined Lagrange multipliers

. So here, the
forces of constraint will be supplied as part of the solution to the problem!
By solving explicitly for the constraint forces, there is no need to try and nd a generalized set of coordinates
(which cannot be done for semi-holonomic constraints).
Expanding out the expression for Q
k
we obtain:
Q

k
=

_
f

q
k

d
dt
_
f

q
k
__
+

q
k

d
dt
_

q
k
__
f

q
k
df

dt
+
d

dt
f

q
k
_
(5.17)
56
CHAPTER 5. VARIATIONAL PRINCIPLES 5.2. COMMENTS ABOUT HAMILTONS PRINCIPLE
Note, in the case of the semi-holonomic constraint condition:
f

(q
1
, . . . q
N
, q
1
, . . . q
N
) = 0
which must be imposed in order to solve for the

, also implies that

df
dt
= 0. Imposition of these conditions on
Q

k
leads to:
Q

k
=

_
f

q
k
d
dt
_
f

q
k
__

dt
f

q
k
(5.18)
which is the same result we would have obtained if we assumed from the start that

(t) only. Note further

that if the f

were actually holonomic constraints, f

(q
1
, . . . q
N
) = 0,
Q

k
=

q
k
(5.19)
which is often seen in other texts.
Example of a hoop on a wedge
M = mass of hoop
T =
1
2
Mv
2
+
1
2
I
2
(5.20)
v = x (5.21)
I =Mr
2
(5.22)
=

(5.23)
T =
1
2
M x
2
+
1
2
Mr
2

2
(5.24)
V =mgy (5.25)
=Mg( x) sin
L =
1
2
M x
2
+
1
2
Mr
2

2
Mg( x) sin (5.26)
Unconstrained variation of the action leads to:
d
dt
_
L
x
_

L
x
= 0
d
dt
_
L

= 0
d
dt
(M x) Mg sin = 0
d
dt
(Mr
2

) 0 = 0
x = g sin

= 0
x(t) = x
o
+ x
o
t +
1
2
g sin t
2
(t) =
o
+

o
t
57
5.2. COMMENTS ABOUT HAMILTONS PRINCIPLE CHAPTER 5. VARIATIONAL PRINCIPLES
If starting at rest at top of wedge (x
o
= 0, x
o
= 0) the time it would take to go down would be:
x(t
f
) = =
1
2
g sin t
2
f
(5.27)
t
f
=

2
g sin
Note, t
f
does not depend on
o
or

o
. If
o
= 0 and

o
= 0, then (t) = 0, and the hoop completely slips down
the wedge.
Consider now the constraint the hoop must roll down the wedge without slipping. The constraint condition is
that rd = dx, or r

= x, hence:
f
1
= r

x = 0
The constrained variation
_
(L +
1
f
1
)dt = 0 gives:
Q

x
=
d
dt
_
L
x
_

L
x
Q

=
d
dt
_
L

x
=
1
_
f
1
x

d
dt
_
f
1
x
__
Q

=
1
_
f
1

d
dt
_
f
1

__
=

1
f
1
x
=

1
f
1

= 0

1
(1) =

1
= 0

1
r =

1
r
r

x = 0 (5.28)
d
dt
(M x) Mg sin =

1
(5.29)
d
dt
(Mr
2

) =

1
r (5.30)
Solving for x,

and

we obtain:
Mr
2

1
r (5.31)
=Mr x
(5.32)
or
M x =

1
(5.33)
=

1
Mg sin
(5.34)
or

1
=
Mg sin
2
(5.35)
x =
g sin
2
(5.36)
58
CHAPTER 5. VARIATIONAL PRINCIPLES 5.2. COMMENTS ABOUT HAMILTONS PRINCIPLE
1
2
the acceleration with slipping

=
g sin
2r
(5.37)
t
f
=

4
g sin
(5.38)
Note: we pretended in this problem we had a semi-holonomic constraint
f

= r

x = 0 = f

( x,

)
but really we could have stated this as a holonomic constraint
f

= r x = 0 = f

(x, )
in which case the Q

x
s are in holonomic form:
Q

x
=
1
f

x
Q

=
1
f

=
1
=
1
r
Carrying through, we again obtain the equations of motion.
x =
g sin
2

=
g sin
2r
but is given by:
= Mg
sin
2
(5.39)
In this case, amounts to a frictional force of the non-slipping constraint,
Mgsin
2
, that is
1
2
the magnitude
of gravitational force along the wedge
d
dy
(Mgy) = Mg sin, and in the opposite direction.
The rst integral of Lagranges equations
Note:
d
dt
_
L(q, q, t)

k
q
k
L
q
k
_
=
L
t
+

k
q
k
L
q
k
+

k
q
k
L
q
k

k
q
k
L
q
k

k
q
k
d
dt
_
L
q
k
_
=
L
t

k
q
k
_
d
dt
_
L
q
k
_

L
q
k
_
(5.40)
Hence, if
L
t
= 0, then

k
q
k
_
d
dt
_
L
q
k
_

L
q
k
_
= 0 (5.41)
If the Lagrange equations
d
dt
_
L
q
k
_

L
q
k
= 0 are satised, this gives the rst integral equation:
d
dt
_
L(q, q, t)

k
q
k
L
q
k
_
=
L
t
(5.42)
59
5.3. CONSERVATION THEOREMS AND SYMMETRY CHAPTER 5. VARIATIONAL PRINCIPLES
5.3 Conservation Theorems and Symmetry
We dene a generalized momentum as P
j

L
q
j
that is said to be conjugate to q
j
. Sometimes this is called the
canonical momentum.
It is clear by the rst integral equation that if
L
t
= 0, then:

k
q
k
L
q
k
L(q, q, t)
_
= constant energy function
d
dt
h(q
1
, . . . q
N
, q
1
. . . q
N
, t) =
L
t
If there is an ignorable coordinate in that
L
q
k
= 0, then clearly:
d
d
_
L
q
k
_
= 0 =
d
dt
P
k
hence P
k
=constant
Also more generally if:

k
W
k
L
q
k
= 0, where W
k
,= W
k
(t)
then

k
W
k
P
k
= constant
Symmetry: Finally note that time t is reversible in the equations of motion: reversing the velocities q(t) sends
the trajectory backwards along the same path.
60
Chapter 6
Central Potential and More
6.1 Galilean Transformation
A Galilean transformation is one such that:
r(t) = r

(t

) +v t

= r

(t) +v t
since t = t

and relates two reference frames to one another, where the r

frame moves with constant velocity

relative the r frame.
A fundamental assumption of classical mechanics (that breaks down in relativistic mechanics) is that the
classical equations of motion are invariant to Galilean transformation.
6.2 Kinetic Energy
The kinetic energy of a particle is:
T =
1
2
m( x
2
+ y
2
+ z
2
) Cartesian (6.1)
=
1
2
m( r
2
+r
2

2
+r
2
sin

2
) Spherical (6.2)
=
1
2
m( r
2
+r
2

2
+ z
2
) Cylindrical (6.3)
6.3 Motion in 1-Dimension
6.3.1 Cartesian Coordinates
L =
1
2
m x
2
U(x) (6.4)
d
dt
(m x)
L
x
= 0 (6.5)
or

d
dt
_
L x
L
x
_
= 0 (6.6)
61
6.3. MOTION IN 1-DIMENSION CHAPTER 6. CENTRAL POTENTIAL AND MORE

_
1
2
m x
2
U(x) m x
2
_
=
1
2
m x
2
+U(x) (6.7)
=E = constant
x =
dx
dt
=
_
2
m
(E U(x)) (6.8)
dt =
dx
_
2
m
(E U(x))
t =
_
dx
_
2
m
(E U(x))
c
Since E = T +U and T 0 and E U. When E = U(x
E
)T = 0, so the x
E
are turning points.
Calculating the period T(E) between the turing points x
1
(E) and x
2
(E):
T(E) =
_
x
2
(E)
x
1
(E)
dt +
_
x
1
(E)
x
2
(E)
(dt) (6.9)
=2
_
x
2
(E)
x
1
(E)
dt
=2
_
x
2
(E)
x
1
(E)
dx
_
2
m
(E U(x))
6.3.2 Generalized Coordinates
L =
1
2
m x
2
U(x) (6.10)
=T( x) U(x)
in general, T = T
o
(q) +T
1
(q, q) +T
2
(q, q) if x = x(q) and not x = x(q, t), then:
T(q, q) =T
2
(q, q) (6.11)
=
1
2
ma(q) q
2
L =
1
2
m a(q) q
2
U(q) (6.12)
E = q
L
q
L (6.13)
=
1
2
ma(q) q
2
+U(q)
_
2
m
a(q) [E U(q)] = q =
dq
dt
(6.14)
62
CHAPTER 6. CENTRAL POTENTIAL AND MORE 6.4. CLASSICAL VIRAL THEOREM
dt =
dq
_
2
m
a(q) [E U(q)]
t =
_
m
2
_
dq
_
a(q) [E U(q)]
+C
T(E) = 2
_
M
2
_
x
2
(E)
x
1
(E)
dq
_
a(q) [E U(x)]
(6.15)
6.4 Classical Viral Theorem
(for central forces)
Consider G =

i
p
i
r
i
for a system of particles:
dG
dt
=

i
p
i
r
i
+

i
p
i
r
i
from
d
dt
p
i
= F
i
dG
dt
=

i
F
i
r
i
+

i
m
i
r
2
i
=

i
F
i
r
i
+ 2T
Consider:
A) = lim

_

0
A(t)dt (6.16)
_
dG
dt
_
= lim

_

0
dG
dt
dt (6.17)
=
1

lim

[G() G(0)]
=0, when lim

G() < constant

In this case,
_
dG
dt
_
= 0 = 2T) +
_

i
F
i
r
i
_
or
2 T) =
_

i
F
i
r
i
_
Viral theorem (6.18)
If F
i
=
i
V (r
1
, . . . r
N
) and V (r
1
, . . . r
N
) is a homogeneous function of degree n + 1 in the r
i
,
2 T) =
_

i
V (r
1
, . . . r
N
) r
i
_
(6.19)
=(n + 1) V )
For a central force:
V (r) =ar
n+1
F
r
=(n + 1)ar
n
r
n
2 T) = (n + 1) V ) or T) =
(n + 1)
2
V )
63
6.5. CENTRAL FORCE PROBLEM CHAPTER 6. CENTRAL POTENTIAL AND MORE
6.5 Central Force Problem
A monogenic system of 2 mass particles interacting via V (r
2
r
1
), we have:
L =T(

R, r) U(r, r, . . .) (6.20)
=
1
2
M

R
2
+
1
2
r
2
U(r, r, . . .)
R
1
M

i
m
i
r
i
=
m
1
r
1
+m
2
r
2
m
1
+m
2
, r r
2
r
1
(6.21)
M =

i
m
i
= m
1
+m
2
=
m
1
m
2
m
1
+m
2
or
1

=
1
m
1
+
1
m
2
d
dt
_
L

R
_

L
R
= M

R = 0 (6.22)
R(t) = R
o
+

R
o
t (6.23)
For V = V (r):
d
dt
_
L
r
_

L
r
=r
V
r
(6.24)
=r
_
1
r
dV
dr
_
r
For V = V (r) we have the following constants of motion:
M

R = constant (total linear momentum)
L = r p = constant (angular momentum about the center of mass)
L =
1
2
( r
2
+r
2

2
) V (r) (6.25)
Since L is conserved (a constant), and L r = 0, (L r) motion must occur in a plane.
d
dt
_
L

=
d
dt
(r
2

)
. .
P

0 = 0 (6.26)
r
2

=0 (6.27)
I
2
=

P

(6.28)
64
CHAPTER 6. CENTRAL POTENTIAL AND MORE 6.5. CENTRAL FORCE PROBLEM
1. r
2

= P

= (6.29)
constant (angular momentum)
(Keplers 2
nd
Law)
Note: r
2

= 0 implies also
_
1
2
r (r

)
_
. .
(sectorial velocity)
= constant
2.
d
dt
_
L
r
_

L
r
=
d
dt
( r) r

2
+
V
r
..
dV
dr
= 0 (6.30)
=2
_
dA
dt
_
(6.31)
dA
dt
= constant (6.32)
r

2
r
3
=
dV
dr
r =
dV
dr
_
V +
1
2
_

r
_
2
_
Multiply both sides by
dr
dt
, we obtain:
r r =
d
dt
_
1
2
r
2
_
(6.33)
=
dr
dt
d
dr
_
V +
1
2
_

r
_
2
_
=
d
dt
_
V +
1
2
_

r
_
2
_
Combining, we obtain the conservation of the energy function (which we knew should be!)
d
dt
_
1
2
r
2
+
1
2
_

r
_
2
+V
_
= 0 (6.34)
or
1
2
r
2
+
1
2
_

r
_
2
+V = E = constant (6.35)
Note: this is 1
st
integral.

r
2

+ r r
1
2
r
2

1
2
r
2
+V (r)
We begin with the rst integral 6.35, since it does not couple r and . Solving for r:
r =

_
2

(E V (r)

2
2r
2
. .
V

(r)
) =
dr
dt
(6.36)
65
6.5. CENTRAL FORCE PROBLEM CHAPTER 6. CENTRAL POTENTIAL AND MORE
V

(r) = V +

2
2r
2
dt =
dr
_
2

_
E V (r)

2
2r
2
_
(6.37)
let r(t
o
) r
o
then, V

(r) =V (r) +

2
2r
2
t(r) =
_
r
ro
dr
_
2

_
E V (r)

2
2r
2
_
(6.38)
=
_
r
o
dr
_
2

(E V

(r))
Solving for t(r) and inverting gives r(t), then rst integral r

= = r
d
dt
gives:
(t) =
_
t
0
dt
r
2
(t)
+
o
(6.39)
Note, this is equivalent to the problem of a particle of mass moving under the inuence of an effective
potential:
E =
1
2
r
2
+V

(r) (6.40)
Example:
V

(r) = V (r) +

2
2r
2
with
F

r
=
dV
dr
+

2
r
3
..
(6.41)
centrifugal force
E V

V
V =
k
r
(6.42)
f =
k
r
2
(6.43)
(gravitational electrostatic)
Sometimes we are interested not necessarily in r(t) and (t), but rather on r() or (r) or some other relation
between r and that describe to orbit. Recall:
d
dt
=

r
2
(6.44)
dr
dt
=

_
E V

2
2r
2
_
(6.45)
66
CHAPTER 6. CENTRAL POTENTIAL AND MORE 6.6. CONDITIONS FOR CLOSED ORBITS
dr
dt
=
d
dt
_
dr
d
_
dr
dt
=

r
2
_
dr
d
_
(6.46)
d
r
2

= dt =
dr
_
2

_
E V

2
2r
2
_
(6.47)
d =
dr
r
2
_
2

_
E V

2
2r
2
_
(6.48)
=
_
r
ro
dr
r
2
_
2E

2

2V

2

1
r
2
+
o
(6.49)
Indroducing a sometimes useful transformation u =
1
r
:
=
o

_
u
uo
du
_
2(EV )

2
u
2
(6.50)
This can be solved for many cases when:
power law of force
V =ar
n+1
(6.51)
F
r
=(n + 1)ar
n
(6.52)
n = 1, 2, 3 (trigonometric) and n = 5, 3, 0, 4, 5, 7 (elliptic)
6.6 Conditions for Closed Orbits
V

(r) = V (r) +

2
2r
2
(6.53)
For circular orbit:
E = V (r
o
) +

2
2r
2
o
+
1
2
r
2
(6.54)

dV

dr

ro
= 0 =f

(r
o
)
=f(r
o
) +

2
r
3
o
for circular orbit, equivalent to f(r
o
) =

2
r
3
o
if
d
2
V

dr
2

ro
>0 (stable orbit)
<0 (unstable orbit)
67
6.7. BERTRANDS THEOREM CHAPTER 6. CENTRAL POTENTIAL AND MORE

d
2
V

dr
2

ro
=
df
dr

ro

3
2
r
4
o
For a stable orbit:
df
dr
<
3
r
o
f(r
o
)
or:
1
(
1
ro
)
a
f(r
o
)
_
df
dr
_

ro
=
d lnf
d lnr

ro
> 3
(assume f(r
o
) < 0)
If f = kr
n
, a stable orbit requires n > 3 if k > 0, or n > 0 for k < 0.
For small perturbations about stable circular orbits:
U
1
r
= U
o
+a cos
Condition for closed orbits(eventually retraces itself), and we have:
d lnf
d lnr
=
2
3 f(r) =
k
r
3
2
(6.55)
were is a rational number.
It turns out, that for arbitrary perturbations (not only small ones), stable closed orbits are possible only for

2
= 2 and
2
= 4. This was proved by Bertrand.
6.7 Bertrands Theorem
The only central forces that result in closed orbits for all bound particles are for:
f(r) =
k
r
3

2
(6.56)
where:
1.
2
= 1 and f(r) =
k
r
2
(inverse square law)
2.
2
= 4 and f(r) = k r (Hookes law)
6.8 The Kepler Problem
V =
k
r
F =
k
r
2
_
r
r
_
The equation for the orbit in the Kepler problem:
68
CHAPTER 6. CENTRAL POTENTIAL AND MORE 6.8. THE KEPLER PROBLEM
=

_
du
_
2

2
_
E V
_
1
u
_
u
2
(indenite integral) (6.57)
=

_
du
_
2

2
[E +ku] u
2
V = k u
Note:
_
du
_
+u +u
2
=
1

arccos
_

+ 2u

q
_
(6.58)
where q
2
4
let:

2E

2
=
2k
l
2
=1
q =
_
2k

2
_
2
_
1 +
2E
2
k
2
_
(6.59)
Note: cos(a) = cos(a), thus arccos can give
=

arccos
_
_
_

2
U
k1
_
_
1 +
2E
2
k
2
_
_
(6.60)

1 +
2E
2
k
2
= e
cos(

) =
_
1

2
u
k
_
e
(6.61)
e cos(

) + 1 =

2
k
u
Only 3 of 4 constants appear in the orbit equation
u =
1
r
=
k

2
_
1 +e cos(

)
_
(6.62)

= turning point of orbit

Constants: E, ,

,
o
[
o
= initial position]
r =
a(1 e
2
)
[1 +e cos(

)]
(6.63)
where a is the semimajor axis
a =
1
2
(r
1
+r
2
) (6.64)
=
k
2E
69
6.8. THE KEPLER PROBLEM CHAPTER 6. CENTRAL POTENTIAL AND MORE
Equation for a conic with one focus at the origin:
e = eccentricity and ellipse axis: a(1 e) and a(1 +e)
e =

1 +
2E
2
k
2
(6.65)
=

1

2
ka
Aside: In the Bohr model for the atom:
E = T +V =
1
2

2
r
2

_
e
2
4
o
_
1
r
(6.66)
=
1
2
_

_
u
2

_
e
2
4
o
_
u
dE
du
=

2

u
_
e
2
4
o
_
= 0 (6.67)
leads to:
U

2
_
e
2
4
o
_
=
1
r

(6.68)
r

_
4
o
e
2
_
(6.69)
E

=
1
2
_
e
2
4
o
r

_
(6.70)
If is chosen to be quantized: = nh, n = 1,2 . . .
Note:
a
o

4
o
h
2
e
2
m
e
The Bohr atom radii and energy states:
r
n
=
n
2
h
2

_
4
o
e
2
_
(6.71)
E
n
=
1
2

e
2
4
o

1
r
n
(6.72)
The equation for the motion it time for the Kepler problem:
t =
_

2
_
r
ro
dr
_
E +
k
r


2
2r
2
(6.73)
=

3
k
2
_

o
d
[1 +e cos(

)]
2
(6.74)
=

3
2k
2
_
tan(

2
)
0
(1 +x
2
)dx for e = 1 (6.75)
70
CHAPTER 6. CENTRAL POTENTIAL AND MORE 6.9. THE LAPLACE-RUNGE-LENZ VECTOR
or:
t =

3
2k
2
_
tan

2
+
1
3
tan
3

2
_
(6.76)
Parabolic motion. Hard to invert!
Introduce auxiliary variable (the eccentric anomaly)
r = a(1 e cos )
t() =
_
a
3
k
_

0
(1 e cos )d (6.77)
=
_

k
a
3
2
(e sin)
=t(2) (6.78)
=2
_

k
a
3
2
(period)
(6.79)
Note: Keplers 3
rd
law, K = +G(m
1
m
2
), for all planets m
2
= mass of the sun. (6.80)
=
2a
3
2
_
G(m
1
m
2
)
C a
3
2
=
2

(Revolution frequency) (6.81)

=

k
a
3
=

_
1
a
3
2
_
6.9 The Laplace-Runge-Lenz Vector
A = p L k
r
r
(6.82)
V =
k
r
(6.83)
(6.84)
f (r) =
k
r
2
_
r
r
_
(6.85)
=f(r)
r
r
d
dt
(p L) = mf(r)r
2
d
dt
_
r
r
_
(6.86)
71
6.9. THE LAPLACE-RUNGE-LENZ VECTOR CHAPTER 6. CENTRAL POTENTIAL AND MORE
for:
f(r) =
k
r
2
(6.87)
=
d
dt
_
kr
r
_
Note: A L = 0 and A p = 0, A? in plane of orbit.
72
Chapter 7
Scattering
7.1 Introduction
I = intensity (ux density) of incident particles
()d = scattering cross section (differential scattering cross section)
= number of particles scattered into solid angle d
(about ) per unit time
Incident intensity (Symmetry about axis of incident beam)
d 2 sind
= scattering angle
s = impact parameter
=

mv
o
=

2mE
= S(, E)
= angular momentum and E = energy
r
m
= distance of periapsis
= 2 (7.1)
sin()[d[ = s[ds[ (7.2)
() =
s
sin

ds
d

(7.3)
(Assume one-to-one mapping s )
73
7.1. INTRODUCTION CHAPTER 7. SCATTERING
Recall:
=
_
r
ro
dr
r
2
_
2E

2

2V

2

1
r
2
+
o
..

(7.4)
= n when r r
m
(for
o
= and r
o
= )
=
_

rm
dr
r
2
_
2E

2

2V

2

1
r
2
(7.5)
=
_

rm
dr
r
2
_
2

2
[E V (r)]
1
r
2
= (E, ) (7.6)
Note:

=

r
2
= constant, hence it cannot change sign (is monotonic in time)
Want in terms of S and E.
E =
1
2
V
2

(7.7)
[[ = = [r p[ = constant
r p = x y z
s 0
V

0 0
=V

s z
=V

s E =
1
2
V
2

=
_
2E
= s
_
2E (7.8)
(s) = 2 (7.9)
= 2
_

rm
sdr
r
_
r
2
_
1
V (r)
E
_
s
2
= 2
_
um
0
sdu
_
1
V (
1
u
)
E
s
2
u
2
2

2
=
2
s
2
2E
=
1
s
2
E
74
CHAPTER 7. SCATTERING 7.1. INTRODUCTION
With = S

2E and (S) = 2
(s) = 2
_

rm
s dr
r
_
r
2
_
1
V (r)
E
_
s
2
(7.10)
u =
1
r
r =
1
u
dr =
du
u
2
(7.11)
u
m
=
1
r
m
u

=0 (7.12)
(s) = 2
_
um
0
s du
_
1
V (
1
u
)
E
s
2
u
2
(7.13)
Scattering in a Coulomb eld
V =
ZZ

r
(atomic units)
For E > 0,
e =

1 +
2El
2
(ZZ

)
2
(7.14)
=

1 +
_
2Es
ZZ

_
2
1
r
=
_
ZZ

2
_
[ cos(

) 1], let

=
lim
r
1
r
= 0, hence cos

..
( ) 1 = 0
cos() = cos =
1

=cos
_

2


2
_
(7.15)
=cos
_

2


2
_
=sin

2
=
1

Solving for:

2
= cot
2

2
+ 1 cos
2

2
(7.16)
since:
1
sin
2
x
=
cos
2
x + sin
2
x
sin
2
x
= cot
2
x + 1
cot

2
=
2Es
ZZ

(7.17)
75
7.2. RUTHERFORD SCATTERING CHAPTER 7. SCATTERING
this gives:
S =
ZZ

2E
cot

2
(7.18)
recall:
() =
S
sin
[ds[
[d[
(7.19)
=
1
4
_
ZZ

2E
_
2
csc
4

2
7.2 Rutherford Scattering
7.2.1 Rutherford Scattering Cross Section
(Rutherford experiment...)
Total scattering cross section:

T
=
_
d() (7.20)
=2
_

0
dsin()
Really, not a single-values function of s
() =

i
s
i
sin

ds
d

(7.21)
Rainbow scattering (E of particle exceeds V
m
, goes through)
Glory scattering (orbiting/spinaling occurs...)
7.2.2 Rutherford Scattering in the Laboratory Frame
tan
1
=
m
2
sin
m
1
+m
2
cos()
(7.22)

2
=
1
2
( )
. .
=2
2
(7.23)
Scattering Particle:

2
(
2
) =
_
ZZ

2E
_
2
sec
3

2
(7.24)

1
(
1
) = difcult . . . (7.25)
76
CHAPTER 7. SCATTERING 7.3. EXAMPLES
Case I:
m
2
>> m
1

1
m
1

1
(
1
) =
1
u
_
ZZ

2E
1
_
2
csc
4
_

1
2
_
(same result) (7.26)
Case II:
m
1
=m
2
= m =
m
2
=2
1
x in g.
1
+
2
=

2

1
(
1
) =
_
ZZ

E
1
_
2
cos
1
sin
4

1
(7.27)
For identical particles, cannot tell which is scattered, so:

1
() =
_
ZZ

E
1
_
2
_
1
sin
4

+
1
cos
4

_
cos (7.28)
=
_
ZZ

E
1
_
2
_
csc
4
+ sec
4

cos
7.3 Examples
E V

V
E
4
= circular (7.29)
E
3
= elliptic (7.30)
E
2
= parabolic (7.31)
E
1
= hyperbolic (7.32)
r
1
, r
2
= apsidal distances (turning points) for bounded motion.
r
o
is solution of:
dV

dr

ro
= 0
(Circular orbit)
r =
a(1 e
2
)
1 +e cos(

)
e =

1 +
2E
2
K
2
77
Chapter 8
Collisions
Collisions...
P

1
=V
o
+
P
m
2
(8.1)
P

2
=V
o
+
P
m
1
(8.2)
AB =P OC =V
o
AO
OB
=
m
1
m
2
AO =
P
m
2
OB =
P
m
1
If m
2
is initially at rest ([OB[ =
o
)
tan
1
=
m
2
sinx
m
1
+m
2
cos x

2
=
1
2
( x) (8.3)
In center of mass reference frame (i.e. moving with the center of mass) - total P = 0 for this frame...
E
i
=
_
E
1i
+
P
2
o
2m
1
_
+
_
E
2i
+
P
2
o
2m
2
_
(8.4)
E
i
= internal energy
P
1
= m
1
v
1
(8.5)
T =
P
2
1
2m
1
=
P
2
o
2m
1
(8.6)
[P
1
[ = [P
2
[ = P
o
(8.7)
P
2
= m
2
v
2
= P
1
(8.8)
78
CHAPTER 8. COLLISIONS 8.1. ELASTIC COLLISIONS
T =
P
2
2
2m
2
=
P
2
o
2m
2
(8.9)
=E
i
(E
1i
+E
2i
) (8.10)
= disintegration energy > 0 (8.11)
=
1
2
P
2
o
_
1
m
1
+
1
m
2
_
=
P
2
o
2
=
m
1
m
2
m
1
+m
2
In laboratory center of mass reference frame, V = velocity of primary particle prior to disintegration.
V =v v
o
(8.12)
v =V+v
o
(8.13)
v
o
=v V (8.14)
v = velocity of one of the particles (in L frame) after disintegration
v
o
= velocity of one of the particles (in C frame) after disintegration
8.1 Elastic Collisions
No change in the internal energy of colliding particles.
In center of mass reference frame (P = 0), let:
P
o
n
o
=v
o
(8.15)
v
o

P
o

n
o
(8.16)
P

o1
=m
1
v

o1
(8.17)
=P

o2
(8.18)
=P
o
n
o
(8.19)
[P

o1
[ = [P

o2
[ = P
o
(8.20)
m
1
v

o1
= m
2
v

o2
(8.21)
P

o2
= m
2
v

o2
= P

o1
= P
o
n
o
Primes after collision
m
1
v

o1
=P
o
n
2
m
1
v

o1
=P
o
v

o1
=
P
o
m
1
(8.22)
79
8.1. ELASTIC COLLISIONS CHAPTER 8. COLLISIONS
v

o1
=
m
2
v
o
m
1
+m
2
=
P
o
m
1
n
o
(8.23)
v

o2
=
m
1
v
o
m
1
+m
2
=
P
o
m
2
n
o
(8.24)
In L reference frame:
P =P
1
+P
2
v =v
1
v
2
(8.25)
v

1
=v

10
+ (8.26)
v

2
=v

20
+ (8.27)
=
P
m
=
m
1
v
1
+m
2
v
2
m
1
+m
2
(8.28)
P

1
=P
o
n
o
+ (P
1
+P
2
)
m
1
m
1
+m
2
=
_
v
o
+
P
m
2
_
(8.29)
P

2
=P
o
n
o
+ (P
1
+P
2
)
m
2
m
1
+m
2
=
_
v
o
+
P
m
1
_
(8.30)
If m
2
is initially at rest (P
2
= 0) then:
tan
1
=
m
2
sinx
m
1
+m
2
cos x

2
=
1
2
( x) (8.31)

1
=
[P

1
[
m
1
=
_
(m
2
1
+m
2
2
+ 2m
1
m
2
cos x)
m
1
+m
2
(8.32)

2
=
[P

2
[
m
2
=
2m
1

m
1
+m
2
sin
1
2
x (8.33)
For head on collisions, x = and P
1
and P
2
are in the same (m
1
> m
2
) or opposite (m
1
< m
2
)
directions.
In this case (head-on collision):
v

1
=
m
1
m
2
m
1
+m
2
v v

2
=
2m
1
m
1
+m
2
v (8.34)
E
1
=
1
2
m
1
v
2
E

2max
=
4m
1
m
2
(m
1
+m
2
)
2
E
1
(8.35)
For head-on collision, if m
1
= m
2
, then m
1
stops and transfers all of its energy to m
2
.
If m
1
< m
2
then m
1
can bounce off m
2
and have velocity in any direction.
80
CHAPTER 8. COLLISIONS 8.1. ELASTIC COLLISIONS
If m
1
>
2
, it cannot bounce back (become reected) and there is an angle
max
for which the deected
angle
1
must lie within
max

1

max
sin
max
=
m
2
m
1
(8.36)
If m
1
= m
2
, then
1
+
2
=

2
:

1
=
1
2
x
2
=
1
2
( x) (8.37)

1
= cos
1
2
x

2
= sin
1
2
x (8.38)
and after collision, particles move at right angles to one-another.
81
Chapter 9
Oscillations
9.1 Euler Angles of Rotation
Eulers Theorem: The general displacement of a rigid body with one point xed is a rotation about some axis.
Chasles Theorem: The most general displacement of a rigid body is a translation plus a rotation.
u(x) =
1
2
kx
2
P(x) e
u(x)
x(t) = Asin(t +) (9.1)
p(t)dt =
dx
2

A
2
x
2
(Classical, microcanonical) (9.2)
[
o
[
2
e
x
2
=
_
ku
h
2
_ 2

o
p(t)dt =
1
2
sin
1
_
x
A
_
(9.3)
9.2 Oscillations
Consider a conservative system with generalized coordinates whose transformation are independent of time (i.e.
constraints are independent of time and T = T
2
quadratic).
At equilibrium:
Q
i
=
_
V
q
i
_
o
= 0 (9.4)
V (q) = V
(o)
(q
o
) +q
T
V
(1)
+
1
2
q
T
V
(2)
q +. . .
_
V
(1)
o
_
i
=
V
q
i

o
_
V
(2)
o
_
ij
=

2
V
q
i
q
j

o
82
CHAPTER 9. OSCILLATIONS 9.2. OSCILLATIONS
So, for small displacements q about the equilibrium q
o
, we have:
V (q) V (q
o
) = V q =
1
2
q
T
V
(2)
q (9.5)
T =
1
2

ij
T
(2)
ij
q
i
q
j
(9.6)
where:
T
(2)
ij

k
m
k
r
i
q
i

r
j
q
j
= m
ij
(q
1
. . . q
N
) (9.7)
(

q q
o
) = q = q
T =
1
2
q
T
T
(2)
q (9.8)
=
1
2
q
T
T
(2)
q

1
2
q
T
T
(2)
o
q (9.9)
where:
T
(2)
o
= T
(2)
(q
o
)
Note:
T
(2)
= T
(2)
(q)
T
(2)
ij
(q) =T
(2)
ij
(q
o
) +q
T

_
T
(2)
ij
q
_
+
1
2
q
T

2
T
(2)
ij
qq
_
q +. . .
T
(2)
ij
(q
o
)
. .
T
(2)
o
for small q q q
o
So,
V =V (q) V (q
o
) (9.10)
=
1
2
q
T
V
(2)
o
q
Note: the zero of the potential is arbitrary
T =
1
2
q
T
T
(2)
o
q (9.11)
L =T V (9.12)
=
1
2
_
q
T
T
(2)
o
q + q
T
V
(2)
o
q
_
83
9.2. OSCILLATIONS CHAPTER 9. OSCILLATIONS
dL
dt
_
L
q
_

L
q
= T
(2)
o
q +V
(2)
o
q = 0 (9.13)
Note:
d
dt
_
T
(2)
o
_
= 0 since the equation of transformation are independent of time, and T
(2)
o
= T
(2)
(q
o
)
Note: in Cartesian coordinates,
_
q
_
=
_
_
_
_
_
_
_
_
_
_
_
_
x
1
y
1
z
1
.
.
.
xN
3
yN
3
zN
3
_
_
_
_
_
_
_
_
_
_
_
_
=
_
_
_
x
1
.
.
.
x
N
_
_
_
_
T
(2)
o
_
ij
= m

i

ij
where
i
= moD(i, 3)
e.g.
_
_
_
_
_
_
_
_
_
_
_
m
1
m
1
m
1
m
2
m
2
m
2
.
.
.
_
_
_
_
_
_
_
_
_
_
_
and in mass-weighted Cartesian coordinates:
x

i
=
x
i

M
i
_
T
(2)
o
_
ij
=
ij
let:
q
k
=C a
k
cos(t ) q
k
=
2
q
k
Note: the q
k
s are functions of time, and the a
k
s are linear algebraic coefcients to be solved for.
_
V
(2)
o

2
T
(2)
o
_
q = 0 (9.14)
V
(2)
o
a
k
=
2
k
T
(2)
o
a
k
(9.15)
V
(2)
o
A = T
(2)
o
A
2
(9.16)
(A generalized eigenvector problem that have non-trivial solutions when
2
is a root of the secular equation.
Note that V and T are real symmetric.)
84
CHAPTER 9. OSCILLATIONS 9.3. GENERAL SOLUTION OF HARMONIC OSCILLATOR EQUATION
We learned previously that the solution of a generalized eigenvector problem involving Hermitian matrices
results in a set of eigenvectors that are bi-orthogonal with real eigenvalues. Note:
V
(2)
o
A = T
(2)
o
A
2
(9.17)
Dropping superscrips and subscripts.
(T

1
2
V T

1
2
)T
1
2
A = T

1
2
TT

1
2
T A
2
(9.18)
Regular eigenvector problem:
V

A = A

2
= A

T
V

if A

is normalized such that A

T
A = 1 in which case:
A

T
= A
1
(Ais orthogonal)
or

2
k
=
a

k
T
V

a
k
a

k
T
a
k
if all
2
k
> 0, all
k
> 0 minimum.
if one
2
k
< 0, that
k
is imaginary transition state (1
st
order saddle point)
if n of the
2
k
< 0, those
k
are imaginary, n
th
order saddle point
V
(2)
o
A = T
(2)
o
A
2
(9.19)
A = (a
1
, a
2
. . . a
N
)a
k
Choose normalization such that:
A
T
V
(2)
o
A =
2
(diagonal) A
T
T
(2)
o
A = 1
9.3 General Solution of Harmonic Oscillator Equation
9.3.1 1-Dimension
d
2
dt
2
q(t) =
2
q(t) (9.20)
85
9.3. GENERAL SOLUTION OF HARMONIC OSCILLATOR EQUATION CHAPTER 9. OSCILLATIONS
q(t) =C
1
e
it
+C
2
e
it
_
C
3
_
e
it
e
it

2i
+C
4
. . .
_
(9.21)
=C
3
sin(t) +C
4
cos(t)
=C
5
sin[t +C
6
]
=C
7
cos[t +C
7
]
.
.
.
e.g.
q(0) =q
o
q(0) =0 (9.22)
C
4
=q
o
C
3
=0 C
3
=0
q(t) = q
o
cos t
e.g.
q(t
o
) =0 q(t
o
) =v
o
C

=t
o
C
5
=v
o
C
5
=
v
o

q(t) =
v
o

sin[(t t
o
)]
9.3.2 Many-Dimension
Normal coordinates
q
i
(t) =

k
C
k
a
ik
cos[t +
k
] or equivalent
let:
q

= A
T
q
. .
q

k
=(a
T
k
q)a
k
(Projection in orthogonal coordinates)
or
q = A q

86
CHAPTER 9. OSCILLATIONS 9.4. FORCED VIBRATIONS
V =
1
2
q
T
V
(2)
o
q (9.23)
=
1
2
q

T
A
T
V
(2)
o
A
. .

2
q

=
1
2
q

T

2
q

=
1
2

2
k
q

k
2
T =
1
2
q
T
T
(2)
o
q (9.24)
=
1
2
q

T
A
T
T
(2)
o
A
. .
?
q

=
1
2
q

T
q

=
1
2

k
q

k
2
9.4 Forced Vibrations
Q

i
=

i
A
ji
Q
j
( Q

= A
T
Q, Q = A Q

)
q

i
+
2
i
q

i
= Q

i
Suppose Q

i
= Q

io
cos(t +
i
) (e.g. the oscillating electromagnetic eld of a monochromatic light source
on a polyatomic molecule, q

i
are the molecular normal Raman vibrational modes)
Note:

i
=
2
Q

io
cos(t +
i
) =
2
Q

i
let q

i
=
Q

2
i

2
then q

i
=
2
q

i
q

i
+
2
i
q

i
=
2
q

i
+
2
i
q

i
(9.25)
=(
2
i

2
)q

i
=(
2
i

2
)
Q

2
i

2
=Q

i
(9.26)
87
9.5. DAMPED OSCILLATIONS CHAPTER 9. OSCILLATIONS
Which is a solution of the desired equation. Hence:
q

i
(t) =
Q

i
(t)

2
i

2
(9.27)
=
Q

io
cos(t +
i
)

2
i

2
q
i
(t) =

j
A
ij
q

j
(t) (9.28)
=

j
A
ij
Q

jo
cos(t +
j
)

2
j

2
9.5 Damped Oscillations
Recall:
F =
1
2

j
F
ij
q
i
q
j
(9.29)
=
1
2
q
T
F q

1
2
q
T
F
(2)
o
q
Lagranges equations are:
d
dt
_
L
q
j
_

L
q
j
= Q
j
=
F
q
j
(9.30)
or in matrix form.
T
(2)
o
q +F
(2)
o
q +V
(2)
o
q = 0 (9.31)
It is not possible, in general, to nd a principle axis transformation that simultaneously diagonalizes three
matices in order to decouple the equations.
In some cases it is possible, however. One example is when frictional forces are proportional to both the
particles mass and velocity.
More generally, anytime the transformation that diagonolizes T
(2)
o
and V
(2)
o
also diagonalizes F
(2)
o
with
eigenvalues F
i
, we have:
q

k
+F
k
q

k
+
2
k
q

k
= 0 (9.32)
In that case (use of complex exponentials useful)
q

k
(t) = C
k
e
i

k
t
leads to:
_

k
2
+i

k
F
k

2
i
_
q

k
(t) = 0 (9.33)
88
CHAPTER 9. OSCILLATIONS 9.5. DAMPED OSCILLATIONS

k
=

2
k

F
2
k
4
i
F
k
2
(9.34)
ax
2
+bx +c
x =
b

b
2
4ac
2a
a =1 b =iF
k
c =
2
k
Which gives:
q

k
(t) = C
k
e
F
k
t
2
e
i
k
t
(9.35)
The general solution is harder. . .
89
Chapter 10
Fourier Transforms
10.1 Fourier Integral Theorem
f(r) piecewise smooth, absolutely integrable, and if:
f(r) =
1
2
_
f(r
+
) +f(r

(10.1)
at points of discontinuity, then:
f(r) =
1
(2)
3
2
_

f(k)e
ikr
d
3
k (10.2)
where

f(k) =F
k
(f(r)) (10.3)
=
1
(2)
3
2
_

f(r

)e
ikr

d
3
r

Note, this implies:

f(r) =
1
(2)
3
_

d
3
ke
ikr
_

d
3
r

e
ikr

f(r

) (10.4)
=
_

d
3
r

f(r

)
1
(2)
3
_

e
ik(rr

)
d
3
k
. .
(rr

)
=
_

d
3
r

f(r

)(r r

)
(r r

) =
1
(2)
3
_

e
ik(rr

)
d
3
k (10.5)
This is the spectral resolution of the identity.
f(r) =
1
(2)
3
2
_

d
3
ke
ik(r)

f(k)

f(k) =
1
(2)
3
2
_

d
3
re
ik(r)
f(r)
90
CHAPTER 10. FOURIER TRANSFORMS 10.2. THEOREMS OF FOURIER TRANSFORMS
10.2 Theorems of Fourier Transforms
Derivative Theorem:
F
k
f(r) = ik

f(k) (10.6)
Convolution Theorem:
(f g)(r)
_
g(r

)f(r r

)d
3
r

(10.7)
F
k
f g =
1
(2)
3
2

f(k) g(k) (10.8)

Translation Theorem:
F
k
(r R
A
) =
e
ikR
A
(2)
3
2
(10.9)
and hence:
F
k
f(r R
A
) =
e
ikR
A
(2)
3
2

f(k) (10.10)
Parsevals Theorem:
_

(r)g(r)d
3
r =
_

f(k)

g(k)d
3
k (10.11)
Note: there are other conventions for dening things, for example:
f(r) =
_

d
3
ke
ikr

f(k) (10.12)
f(k) =
1
(2)
3
_

d
3
re
ikr

f(r) . . . (10.13)
10.3 Derivative Theorem Proof
F
k

r
f(r) =
1
(2)
3
2
_

d
3
re
ikr

r
f(r) (10.14)
Note:

r
_
e
ikr
f(r)
_
=
_

r
e
ikr
_
f(r) +e
ikr

r
f(r) (10.15)
hence
_

d
3
re
ikr

r
f(r) = (10.16)
_

d
3
r
r
_
e
ikr
f(r)
_
. .

d
3
r
_

r
e
ikr
_
f(r) (10.17)
91
10.4. CONVOLUTION THEOREM PROOF CHAPTER 10. FOURIER TRANSFORMS
recall the generalized divergence theorem
_
V
dV
r
f =
_
S
d f (10.18)
_

d
3
r
r
_
e
ikr
f(r)
_
=
_
S
d
_
e
ikr
f(r)
_
= 0 (10.19)
if
_
[e
ikr
[ = 1
_
f(r) 0 at any boundarty s.t.
_
S
d[f(r)[ = 0 then we have:
F
k

r
f(r) =
1
(2)
3
2
_

d
3
r
r
e
ikr
f(r) (10.20)
= +ik
1
(2)
3
2
_

d
3
re
ikr
f(r)
= +ik

f(k)
10.4 Convolution Theorem Proof
h(r) =
1
(2)
3
2
_

d
3
ke
ikr

f(k) g(k)
. .

h(k)
(10.21)
=
1
(2)
3
2
_

d
3
ke
ikr
g(k)
1
(2)
3
2
_

d
3
r

e
ikr
f(r

)
. .

f(k)
=
1
(2)
3
2
_

d
3
r

f(r

)
1
(2)
3
2
_

d
3
ke
ik(rr

)
g(k)
. .
g(rr

)
=
1
(2)
3
2
_

d
3
r

f(r

)g(r r

)
=
1
(2)
3
2
f g
92
CHAPTER 10. FOURIER TRANSFORMS 10.5. PARSEVALS THEOREM PROOF
10.5 Parsevals Theorem Proof
_

d
3
k

(E)g(k) =
_

d
3
k
_
1
(2)
3
2
_

d
3
re
ikr

(r)
_
g(k) (10.22)
=
_

d
3
k
1
(2)
3
2
_

d
3
r

(r)
_

d
3
r

g(r

)e
ik(rr

)
=
_

d
3
rf(r)
_

d
3
r

g(r

)
1
(2)
3
2
_

d
3
ke
ik(rr

)
. .
(rr

)
=
_

d
3
rf(r)g(r)

2
(r) = 4(r)
k
2

(k) =4 (k)

(k) =4
(k)
k
2
for Gaussian,

(k) = 4
e
k
2
4
2
(2)
3
2
k
2
(10.23)
(r) =
1
(2)
3
2

4
(2)
3
2
_

d
3
k
e
k
2
4
2
k
2
e
ikr
(10.24)
k r = kr cos (10.25)
d
d
e
ikr cos
= ikr sine
ikr cos
(10.26)
(r) =
4
8
3
_

0
dkk
2
_
2
0
d
_

0
d sin e
ikr cos
e
k
2
4
2
k
2
(10.27)
=
1
2
2
2
_

0
dk e
k
2
4
2
_
e
ikr cos
ikr

0
_
. .
e
ikr
ikr
+
e
ikr
ikr
=
2 sin(kr)
kr
=
2

1
r
_

0
dk
e
k
2
4
2
k
sin(kr)
=
1
r

2

1
2
( 4
2
)
1
2
_
r
0
dy e

2
y
2
93
10.5. PARSEVALS THEOREM PROOF CHAPTER 10. FOURIER TRANSFORMS
let:
t = y
dt = dy =
1
r

2

1
2

1
2
2
1

_
r
0
dt e
t
2
erf(r)
r
where erf(x)
2

_
x
0
dt e
t
2
Application:
g

(r) =
_

_
3
e

2
r
2
_
g

(r)d
3
r = 1
q

(k) =
1
(2)
3
2

d
3
r
_

_
3
e

2
r
2
e
ikr
(10.28)
=
1
(2)
3
2

_
3
_

d
3
re
(ri
k
2
)
2
e

k
2
4
2
Since:
_
r i
k
2
_
2
=
2
r
2
2ir
k
2

k
2
4
2
(10.29)
=
1
(2)
3
2
e

k
2
4
2
_

_
3
_

d
3
r

2
r
2
. .
1
r

=r
ik
2
2
=
e

k
2
4
2
(2)
3
2
Note:
F
k

1
r
=
(2)
3
2
k
2
(10.30)
So

(k) =
1
(2)
3
2

(2)
3
2
k
2
e

k
2
4
2
(2)
3
2
(10.31)
=
e

k
2
4
2
(2)
3
2
k
2
for a Gaussian

g

(r)
94
Chapter 11
Ewald Sums
11.1 Rate of Change of a Vector
(dG)
s
. .
(space)
= (dG)
body
. .
(body r)
+(dG)
rotational
. .
(rotational)
= (dG)
r
+dG
_
d
dt
_
s
=
_
d
dt
_
r
+x rotating coordinate system (11.1)
_
d
dt
r
_
=
_
dr
dt
_
r
+ r v
s
=v
r
+ r
Newtons equations F
s
= mQ
s
= m
_
d
dt
v
s
_
s
F =m
__
d
dt
v
r
+ r
_
s
_
(11.2)
=m
_ _
dv
r
dt
_
r
. .
ar
+ v
r
+
_
d
dt
( r)
_
r
+ ( r)
_
=ma
r
+m[ v
r
+ v
r
+ ( r)]
=ma
r
+ 2m( v
r
) +m[ ( r)]
F 2m( v
r
)
. .
Coriolis effect
m[ ( r) ]
. .
Centrifugal force
= F
eff
= ma
r
(Only if v
r
,= 0, so v
r
= 0 or v
r
| ), and Centrifugal force 0.3% of gravitational force.
11.2 Rigid Body Equations of Motion
Inertia tensor: I(3 3)
I
jk
=
_
v
(r)(r
2

jk
x
j
x
k
)d
3
r (11.3)
95
11.2. RIGID BODY EQUATIONS OF MOTION CHAPTER 11. EWALD SUMS
x
i
= x, y, z for i = 1, 2, 3
e.g. for a system of point particles
I
jk
=

i
m
i
(r
2
i

jk
x
j
x
k
) (11.4)
Note: Typically the origin is taken to be the center of mass.
a (b c) = (a c)b (a b)c
L = I
L =

i
r
i
p
i
(11.5)
=

i
m
i
(r
i
v
i
) v
i
= r
i
=

i
m
i
(r
i
( r
i
))
=

i
m
i
[r
2
i
r
i
(r
i
)]
=

i
[m
i
r
2
i

1 r
i
r
i
]
. .

I (old notation)

T
r
=

i
1
2
m
i
v
2
i
(11.6)
=

i
1
2
m
i
v
i
v
i
=

i
1
2
m
i
v
i
( r
i
)
=

i
1
2
m
i
(r
i
v
i
)
=
1
2

i
m
i
(r
i
v
i
)
=
1
2
L
or
T
r
=
1
2

T
I (11.7)
=
1
2
I
2
where = n (n = axis of rotation) and I = n
T
I n
96
CHAPTER 11. EWALD SUMS 11.3. PRINCIPAL AXIS TRANSFORMATION
11.3 Principal Axis Transformation
UTU
T
=I
D
(diagonal) U U
T
=1
I
D
=
_
_
I
xx
0
I
yy
0 I
zz
_
_
I
xx
, I
yy
, and I
zz
are the principal moments (eigenvalues) of I.
Radius of gyration:
R
o
=
1
m

i
m
i
r
2
i
(11.8)
or
1
m
_
(r) r
2
d
3
r (11.9)
or
_
I
m
(11.10)
or
1
n
T

m
(11.11)
Moment of inertia about axis n
= n
1

I
(11.12)
I = n
T
I n (11.13)
1 =
T
I (11.14)
11.4 Solving Rigid Body Problems
T =
1
2
Mv
2
+
1
2
I
2
= T
t
+T
r
T
r
=
1
2

T
T in general
=
1
2

i
I
ii
in the principal axis system
(11.15)
97
11.5. EULERS EQUATIONS OF MOTION CHAPTER 11. EWALD SUMS
11.5 Eulers equations of motion
_
dL
dt
_
s
=
_
dL
dt
_
r
+wL = N
..
torque
(11.16)
space body (11.17)
system rotating system (11.18)
Recall:
_
d
dt
_
s
=
_
d
dt
_
b
+x (11.19)
In the principal axis system (which depends on time, since the body is rotating) takes the form:
I
i
d
i
dt
+
ijk

k
I
k
= N
i
(11.20)
I
i

i

k
(I
j
I
k
) = N
i
(11.21)
i, j, k = 1, 2, 3 3, 1, 2 2, 3, 1
11.6 Torque-Free Motion of a Rigid Body
_
dL
dt
_
r
+ L = N = 0 (11.22)
Example: Symmertical (I
1
= I
2
) rigid body motion.
_
_
_
I
1

1
=
2

3
(I
2
I
3
) I
1

1
= (I
1
I
3
)
2

3
I
2

2
=
3

1
(I
3
I
1
) I
1
= I
2

I
1

2
= (I
3
I
1
)
1

3
I
3

3
=
1

2
(I
1
I
2
) I
3

3
= 0
_
_
_
I
3

3
=0 I
3

3
=L
3
= constant
= = z
1

1
=
(I
1
I
3
)
I
1

3
=
2
(11.23)

2
=
1
(11.24)
where:
=
_
I
3
I
1
I
1
_

3
Note:

= 0 since
3
= constant
98
CHAPTER 11. EWALD SUMS 11.7. PRECESSION IN A MAGNETIC FIELD

1
=
2

2
=
1
Harmonic!
1
=Acos t

2
=
1
=
2
Asint (
1
=Asint)
So:

2
= Asint
Note:

2
1
+
2
2
= A
2
[cos
2
t + sin
2
t] = A
2
= constant
This is an equation for a circle. Thus = constant,
3
= constant, and precesses around z axis with
11.7 Precession of a System of Charges in a Magnetic Field
m =
1
2

i
q
i
(r
i
v
i
) magnetic moment
L =

i
m
i
(r
i
v
i
) (11.25)
Consider the case when:
m =L =
q
2m
(gyromagnetic ratio)
Example but often left unspecied.
V = (m B) B = magnetic eld (11.26)
N =MB (torque) =L B
_
dL
dt
_
s
=L B (11.27)
=L (B)
Same as
_
dL
dt
_
r
+ L = 0 or
_
dL
dt
_
r
= L (formula for torque free motion!)
=
q
2m
B = Larmor frequency
=B (more general)
99
11.8. DERIVATION OF THE EWALD SUM CHAPTER 11. EWALD SUMS
11.8 Derivation of the Ewald Sum
E =

i<j
q
i
q
j
r
1
ij
=
1
2

j=i
q
i
q
j
r
1
ij
analogous
1
2
_ _
(r
1
)(r
2
)r
1
12
dr
1
dr
2
For periodic systems:
E =
1
2

j
q
i
q
j

[r
ij
+n[
1
(11.28)
=
1
2

j
q
i
q
j

E
(r
ij
)
=
1
2

i
q
i
(r
i
) where (r
i
) =

j
q
j

E
(r
ij
)
Consider the split:
(r
i
) = [(r
i
)
s
(r
i
)] +
s
(r
i
)
= potential of charges

s
= screening of potential
For example, choose
s
to come from a sum of spherical Gaussian functions:
g

(r) =
_

_
3
e
(r)
2
(11.29)

s
(r) =

j
q
j

n
g

(r +n R
j
) (11.30)
Note
s
is a smooth periodic function.

s
(r) = 4
s
(r) (11.31)
11.9 Coulomb integrals between Gaussians
_ _
g

(r
1
R
A
)g

(r
2
R
B
)r
1
12
dr
1
dr
2
(11.32)
=
erf
_

2
+
2
R
AB
_
R
AB
where:
erf(x)
2

_
x
0
e
t
2
dt
1. lim

,
erf(R
AB
)
R
AB
100
CHAPTER 11. EWALD SUMS 11.10. FOURIER TRANSFORMS
2. lim

,
erf

R
AB

R
AB

real
(r
i
) =(r
i
)
s
(r
i
) (11.33)
=

j
q
j

_
1
[r
ij
+n[

erf([r
ij
+n[)
[r
ij
+n[
_
=

j
q
j

n
erfc([r
ij
+n[)
[r
ij
+n[
Short-ranged and can be altervated by
One can choose s.t. only the n = 0 term is signicant.

real
(r
i
) =

j=i
q
j
erfc(r
ij
)
r
ij
q
i

. .
self term
and Energy contribution
E
real
=
1
2

j
q
i
q
j
_
erfc(r
ij
)
r
ij

ij
_
(11.34)
This choice of is sometimes termed the minimum image , and leads to a (N
2
) procedure for the real
space term.
11.10 Fourier Transforms
For a periodic system described by lattice vectors:
n = n
1
a
1
+n
2
a
2
+n
3
a
3
The Fourier transform must be taken over a reciprocal lattice.
m = m
1
a

1
+m
2
a

2
+m
3
a

3
where the reciprocal space lattice vectors satisfy:
a

i
a
j
=
ij
i, j = 1, 2, 3
or
PQ =1 = QP Q =P
1
P = (a
1
a
2
a
3
)

Q =
_
_
a

1
a

2
a

3
_
_
101
11.10. FOURIER TRANSFORMS CHAPTER 11. EWALD SUMS
a

1
=
a
2
a
3
a
1
a
2
a
3
=
a
2
a
3
v
= f (a
1
, a
2
, a
3
) (11.35)
a

1
=f (a
3
, a
1
, a
2
) (11.36)
a

1
=f (a
2
, a
3
, a
1
) cyclic permutation (11.37)
(11.38)
The discrete Fourier transform representation of a function over the reciprocal lattice is:
f(r) =
1
v

f(k)e
ikr
(11.39)
where:
k = 2m = 2(m
1
a

1
+m
2
a

2
+m
3
a

3
)
We now return to:
1.
2

s
(r) = 4
s
(r)
where

s
(r) =

j
q
j
g

(r R
j
) (11.40)
Note:
Fg

(r) = e

k
2
4
2
(11.41)
Fourier transform of 1. gives:
(ik)
2

s
(k) =4
s
(k) (11.42)

s
(k) =4

s
(k)
k
2
(11.43)
by inverse transforming we obtain:
(r) =
1
v

k
q
j
4
e

k
2
4
2
k
2
e
ik(rR
j
)
(11.44)
Note: this blows up at k = 0! Can we throw this term out?

s
(r) =
1
v

k
4
(k)
k
2
(11.45)
but
s
(0) =
_

s
(r)dr = 0 for neutral system!
It turns out we can throw out the k = 0 term if:
1. system is neutral
2. system has no dipole moment
102
CHAPTER 11. EWALD SUMS 11.11. LINEAR-SCALING ELECTROSTATICS
Under these conditions we have:

s
(r) =
4
v

k=0

s
(k)
k
2
(11.46)
If there is a dipole, an extra term in the energy must be added of the form:
J(P, D, E) [

D[
2
D = dipole
P = shape
E = dielectric
What do we do in solution?
(What errors are introduced due to periodicity?)
11.11 Linear-scaling Electrostatics
E =
1
2
_
(r)(r)d
3
r (11.47)
=
1
2
_ _
(r)G(r, r

)(r

)d
3
r

d
3
r
(r) =
_
G(r, r

)(r

)d
3
r

(11.48)

2
(r) = 4(r) (r) = 1 gas-phase

2
G(r, r

) = 4(r, r

)
G(r, r

) =
1
[r r

[
(non periodic) (11.49)
4
v

k=0
e
ik(rr

)
k
2
(periodic) (11.50)
(k = 2m)
11.12 Greens Function Expansion
G(r, r

) =

k
N
k
(r)M
k
(r

) (11.51)
Seperable!
103
11.13. DISCRETE FT ON A REGULAR GRID CHAPTER 11. EWALD SUMS
11.13 Discrete FT on a Regular Grid
f(x) f(x
n
) = f
n
x
n
=n n =o, . . . N 1

f(k)

f(k
m
) =

f
m
k
m
=
2m
N
m =
N
2
, . . . 0, . . .
N
2
(11.52)

f
(k)
=
_
f(x)e
ikx
dx
N1

n=o
f
n
e
i2
mn
N
=

f
m
(11.53)
similarly
f
n
=
1
N
N1

m=0
f
m
e
i2
mn
N
(11.54)
11.14 FFT
Danielson-Lanczos Lemma
W
N
e
i2
N
(11.55)
f
m
=
N1

n=0
W
nm
N
f
n
(11.56)
=
N
2
1

n=0
W
nm
N
2
f
2n
+W
m
N
N
2
1

n=0
W
nm
f
2n+1
=F
e
m
+W
m
N
f
o
m
2 FTs of dim.
N
2
(11.57)
Used recursively with bit reversal to obtain the full FFT in (N log N)!
Ewald Sums (Linear Scaling)
(r) =

i
q
i
(r R
i
) (11.58)

s
(r) =

i
q
i
g(r R
i
) (11.59)
e.g. g(r) =
_

_
3
e

2
r
2
(r) =[(r)
s
(r)] +
s
(r) (11.60)
(r) =[(r)
s
(r)] +
s
(r) (11.61)
104
CHAPTER 11. EWALD SUMS 11.15. FAST FOURIER POISSON

s
(r) = 4
s
(r) (11.62)
Fourier transform
k
2

s
(k) = 4
s
(k) (11.63)
Particle-Mesh, P
3
M Methods
FFT g(r),

i
q
i
(r R
i
) to obtain
s
(k)
Note:
s
(k) is the transform of
s
(r) (a plane wave expansion)
11.15 Fast Fourier Poisson
Evaluate
2
(r) directly on grid. (No interpolation)
Solve for
s
(r) on grid. Modify real space term accordingly.
E =
1
2

j=i
q
i
q
j
erfc(

r
ij
)
r
ij
(11.64)

i
q
2
i

+
1
2
_

s
(r)
s
(r)d
3
r
. .
1
2

ijk
s(i,j,k)s(i,j,k)
+J(D, P, )

2
which equals the exact integral for the plane-wave projected density.
105
Chapter 12
Dielectric
12.1 Continuum Dielectric Models
Gas phase:

o
(r) =4
o
(r)
o
(r) =
_
(r

)
[r r

[
d
3
r

E
o
=4
o
Suppose in addition to
o
(r) there is a dipole polarization p(r), then:
(r) =
_ _

o
(r

)
[r r

[
+
P(r

) (r r

)
[r r

[
3
_
d
3
r (12.1)
=
_ _

o
(r

)
[r r

[
+P(r

)
r
1
_
1
[r r

[
__
d
3
r

=
_ _

o
(r

)
r
1 P(r

)
[r r

[
_
d
3
r

=
_ _

o
(r

) +
pol
(r

)
[r r

[
_
d
3
r

=
1
4
_ _

r
1 (E
o
(r

) 4P(r

))
[r r

[
_
d
3
r

[
pol
(r) =
r
P(r)]

2
(r) = E(r) (12.2)
=4(
o
(r) +
pol
(r))
D(r) E(r) + 4(r) (12.3)
D = E+ 4 (12.4)
=4
o
+ 4
pol
4
pol
=4
o
106
CHAPTER 12. DIELECTRIC 12.1. CONTINUUM DIELECTRIC MODELS
D = 4
o
So where does come from?
=x
e
E x
e
= electric susceptibility
D = E+ 4 = (1 + 4x
e
)E = E (LPIM)
(r) 1 + 4x
e
static dielectric function
D =4
o
(12.5)
= (E)
=4
o
() = 4
o
Poisson equation for lin. isotropic pol. med.
(r) =
o
(r) +
pol
(r) (12.6)
(
o
+
pol
) = 4
o
(12.7)
(
o
+
pol
) +
2

o
. .
4o
+
2

pol
= 4
o
(12.8)

pol
= 4
o
[1 ] (12.9)

pol
=4
o
[1 ]

(12.10)
=4
pol
107
12.2. GAUSS LAW I CHAPTER 12. DIELECTRIC
12.2 Gauss Law I
_
v=

pol
(r)d
3
r =
_

o
_
1

_
d
3
r +
1
4
_

d
3
r (12.11)
=
_

o
_
1

1
_
d
3
r +
1
4
_

_
1

_
Ed
3
r
=
_

o
_
1

1
_
d
3
r
1
4
_ _
1

_
Dd
3
r +
1
4
_
s=
_
1

_
D nda
=
_

o
_
1

1
1

_
d
3
r +
1
4
_
s=
_
1

_
D nda
=
_

o
d
3
r +
1
4

1

2
_
s=
D nda
=
_

o
d
3
r +
1
2
_

o
d
3
r
=
_
1
2

2
__

o
d
3
r

_
1

_
=

2
let
2
are s = (constant). Note:
_
D nda =
_
Dd
3
r = 4
_

o
d
3
r
_
v=

pol
(r)d
3
r =
_

2
1

2
__
v=

o
(r)d
3
r (12.12)
Gauss Law I: Volume integral over all space.
If changes from
1
to
2
discontinuously at a dielectric boundary dening the surface s
12
, then care must be
taken to insure the boundary conditions.
(D
2
D
1
) n
21
= 4
o
where
o
is a static surface charge density (part of
o
)
(E
2
E
1
) n
21
= 0 (12.13)
This implies:

pol
= (
2

1
) n
21
(12.14)
where

i
=
_

i
1
4
_
E
i
(12.15)
=
_

i
1
4
i
_
D
i
Recall:
D =E = 4 =
1
4
(DE)
dielectric dicont.
1

2
and s
12
108
CHAPTER 12. DIELECTRIC 12.3. GAUSS LAW II
If
o
= 0, D
2
n
21
= D
1
n
21
= D n
21

pol
=(
2

1
) n
21
(12.16)
=
_

2
1
4
2

1
1
4
1
D
1
_
n
21
=
1
4
_

1
_
D
12.3 Gauss Law II

pol
=
1
4
_

1
_
D n
21
(12.17)
_
s
12

pol
da =
1
4
_
s
12
_

2
_
D n
21
da (12.18)
=
1
4
_

2
__
v
1
Dd
3
r
=
_

2
__
v
1

o
d
3
r
=
_

2
_
Q
o
(v
1
)
_
s
12

pol
da =
_

2
_
Q
o
(v
1
) (12.19)
Guass Law II for integral over surface at dielectric discontinuity.
12.4 Variational Principles of Electrostatics
Electrostatic energy:
W =
1
4
_
d
3
r
_
D
0
E D (Nonlinear dielectric medium) (12.20)
if medium is linear,
_
D
0
E D =
1
2
E D, then:
W =
1
8
_
d
3
rE D (12.21)
=
1
8
_
d
3
r() D
=
1
4
_
d
3
r D
1
4
_
s=
D nda
=
1
2
_
d
3
r
o

W[,
o
, ]
_

o
(r)(r)d
3
r
1
8
_
d
3
r (12.22)
109
12.5. ELECTROSTATICS - RECAP CHAPTER 12. DIELECTRIC
W
(r)
=
o
(r) +
2
8
[] = 0 (12.23)
[] = 4 (The Poisson equation with dielectric)
The variational prinicple allows to be solved for using optimization methods. Do you minimize or maxi-
mize?
W
(r)
=
o
(r) +
1
4
[] (12.24)
=
_
d
3
r

o
(r) +
1
4

r
[(r)
r
(r

)]
_
(r r

)d
3
r

2
W
(r)(r

)
=
1
4

r
[(r

)
r
(r r

)] (12.25)
if > D, the operator is negative in that:
_ _
f(r)

2
W
(r)(r)
f(r

)d
3
rd
3
r

=
_ _
f(r)
r
[(r

)
r
(r r

)]f(r

)d
3
rd
3
r

(12.26)
=
_ _
f(r)[(r

)
r
(r r

)]
r
f(r

)d
3
rd
3
r

=
_ _
f(r)[(r

)
r
(r r

)]
r
f(r

)d
3
rd
3
r

=
_
(r

)[f(r

)[
2
d
3
r

0
hence W[] is maximized to obtain (r).
=1 W = [,
o
, = 1] =
_

o
d
3
r
1
8
_
d
3
r
12.5 Electrostatics - Recap
Gas-phase in vacuo = 1

2
r

o
(r) =4(r) (

L =4)

o
(r) =
_
G
o
(r, r

)
o
(r

)d
3
r

(12.27)
where:

r
2G
o
(r, r

) = 4(r r

) (12.28)
depends on boundary conditions:
G
o
(r, r

) =
_

_
1
|rr

|
real space, = 0, r
4
v

k=o
e
ik(rr

)
k
2
k = 2m
m = m
1
a

1
+m
2
a

2
+m
3
a

3
110
CHAPTER 12. DIELECTRIC 12.5. ELECTROSTATICS - RECAP
Sometimes we can use the Greens function and analytically obtain a solution for the potential.
In real space, for example Gaussian:

1
(r) =
_

_
3
e
(
1
|rR
1
|)
2

2
(r) =
_

_
3
e
(
2
|rR
2
|)
2

1
(r) =
_

1
(r

)
[r r

[
d
3
r

(12.29)
=
erf(, r)
r
_ _

1
(r)
2
(r

)
[r r

[
d
3
rd
3
r

=
erf(

12
R
12
)
R
12
(12.30)

12
=

1

2
_

2
1
+
2
2
(12.31)
In reciprocal space, Ewald sum:
For minimum image

s.t. n = 0 only term.

E =
1
2

j
q
i
q
j
_
_
erfc(r
ij
)
r
ij

ij
+
4
v

k=0
e

k
2
4
2
k
2
e
ik(r
i
r
j
)
_
_
(12.32)
=

j<i
q
i
q
j
_
erfc(r
ij
)
r
ij

ij
_
+
2
v

k=0
e

k
2
4
2
k
2
[s(k)[
2
s(k) =

i
q
i
e
ikr
i
(12.33)
Sometimes, however, we need to break down G
o
(r, r

) into a product form, and found it useful to expand

G(r, r

) in terms of eigenvectors of its associated operator. This leads to:

Real space:
(r) = 4

=0
1
2 + 1

m=
y
m

..
(, )
m
(r) (12.34)

m
(r) =
1
r
+1
_
r
0
x
(2+)

m
(x)dx +r

_

0
x
(1)

m
(x)dx (12.35)
where

m
(r) =
_
dy

m
()(r) (12.36)
or Fourier transforms:
k
2

(k) =4 (k) (12.37)

(r) =
1
(2)
3
2
_
d
3
ke
ikr
_
4 (k)
k
2
_
(12.38)
(k) =
1
(2)
3
2
_
d
3
k
_
e
ikr
_

(r) (12.39)
111
12.6. DIELECTRICS CHAPTER 12. DIELECTRIC
Reciprocal space (lattice):
(r) =
4
v

k=0
(k)
k
2
(12.40)
for:
k =2m m =
3

i=1
m
i
a

i
12.6 Dielectrics
D = 4
o
(12.41)
D =E (12.42)
=E+ 4
E = (12.43)
Poisson equation for a linear isotropic polarizable medium characterized by (r)
[] = 4
o
(12.44)

pol
(r) (12.45)
(r) =
_
(
o
(r

) =
pol
(r

))
[r r

[
d
3
r

(12.46)

pol
(r) =
o
[1 ]

+
1
4

(12.47)
Gauss Law I:
_
v=

pol
(r)d
3
r =
_

2
1

2
__
v=

o
(r)d
3
r (12.48)
Gauss Law II:
_
s
12

pol
(r)da =
_

2
__
v
1

o
(r)d
3
r (12.49)

ion
(r) =

s
c
s
q
s
e
qs(r)
(c
s
= conc.)

s
c
s
q
s

s
c
s
q
2
s
(linearized)
112
CHAPTER 12. DIELECTRIC 12.6. DIELECTRICS
For Born ion, LPB equation leads to:

2
k
2
= 4
o
(r) (12.50)
k
2
=
4

s
q
2
s
c
s
(12.51)
(Note: I =
1
2

s
q
2
s
c
s
ion conc.)
(r) =q
ion
e
kr
r
(pt. chg.) (12.52)
or
(r) =
q
ion
e
k(ra)
r(1 +ka)
(ion of radius a) (12.53)
x
2
E
pol
[
pol
;
o
, ] =
1
2
_

_
E
pol
..

pol
+
_
1

_
E
o
..
o
_
2
d
3
r (12.54)
x
2
E
pol

pol
=
_
2
_

pol
+
_
1

o
__
= 0 (12.55)

_

pol
+
1

o
_
+
2

pol
. .
4
pol
+
_
1

_

o
+
o
+
_
1

o
. .
4o
= 0 (12.56)
leads to:

pol
=
_
1

o
+
1
4

(12.57)
1
4

pol
=
_
1

o
+
1
4

pol
+
1

o
_
+
1
4

_
1

_

o
(12.58)
Note:

_
1

_
=

+ ( 1)
_
1

_
(12.59)
=

( 1)

2
=

( 1)

2
=

(12.60)
113
12.6. DIELECTRICS CHAPTER 12. DIELECTRIC

pol
=
_
1

o
+
1
4

pol
+
o
. .

_
(12.61)
Suppose: elds cancel! = 1 =
W =
1
2
_ _
(
o
(r) +
pol
(r)) (
o
(r

) +
pol
(r

))
[r r

[
d
3
rd
3
r

(12.62)
W

pol
= 0 (12.63)
W

pol
=
1
2

T
o
G
o

o
+
T
pol
B
o
+
1
2

T
pol
A
pol
(12.64)
W

pol
= B
o
+A
pol
= 0 (12.65)

pol
=

pol,k
f
k
(r)
o
=

o,
g
k
(r)

pol
= A
1
B
o
(12.66)
sq. sym.
A
ij
=
_ _
f
i
(r)f
j
(r

)
[r r

[
d
3
rd
3
r

(12.67)
Not sq. or sym.
B
ij
=
_ _
f
i
(r)g
j
(r

)
[r r

[
d
3
rd
3
r

(12.68)
W =
1
2

T
o
G
o

o

T
B
T
A
1
B
o
+
1
2

T
o
B
T
A
1
B
o
(12.69)
=
1
2

T
o
G
o

o

1
2

T
o
B
T
A
1
B
o
. .
G
pol
=
1
2

T
o
[G
o
+G
pol
]
o
G
pol
= B
T
A
1
B
This is the Conductor like screening model COSMO
114
Chapter 13
Expansions in Orthogonal Functions
The scalar product (or inner product) of two functions (in general complex) over the interval a x b is
denoted (f, g) and is dened by
(f, g)
_
b
a
f

(x)g(x)dx
For functions of more than one variable,
(f, g)
_
v
f

(r)g(r)dr
The scalar product with respect to weighting function w(x) or w(r) is
(f, g)
w

_
b
a
f

(x)g(x)w(x)dx = (f, gw)

or
(f, g)
w

_
v
f

(r)g(r)w(r)dr = (f, gw)

where the weighting function w is real, nonnegative and integrable over the region involved, and can vanish at
only a nite number of points in the region.
The above deinitions of scalar products, both with and without a weighting function, satisfy the basic require-
ments of a scalr product in any linear vector space, namely:
1. (f, g) is a compex number, and (f, g)

= (g, f)
2. (f, c
1
g
1
+c
2
g
2
) = c
1
(f, g
1
) +c
2
(f, g
2
)
3. (f, f) 0, real, and (f, f) = 0 f = 0 almost everywhere,
i.e., f = 0 except at a nite number of points in the region.
From these properties other important relations follow. For example
(c
1
f
1
+c
2
f
2
, g) = c

1
(f
1
, g) +c

2
(f
2
, g)
Proof:
(c
1
f
1
+c
2
f
2
, g) =(g, c
1
f
1
+c
2
f
2
)

from 1.
=[c
1
(g, f
1
) +c
2
(g, f
2
)]

from 2.
=c

1
(g, f
1
)

+c

2
(g, f
2
)

= c

1
(f
1
, g) +c

2
(f
2
, g) from 1. Q.E.D.
115
13.1. SCHWARZ INEQUALITY CHAPTER 13. EXAPANSIONS
13.1 Schwarz inequality
An important consequence of the basic properties of a scalar product is the Schwarz inequality:
[(f, g)[
2
(f, f)(g, g)
or
[(f, g)[ |f| |g|
where |f| (f, f)
1
2
is called the norm of f.
Proof of the Schwarz inequality:
(f +g, f +g) 0 for any complex number , and = 0 if and only if f = g, proportional
(f, f) +(f, g) +

(g, f) +

(g, g) 0
If (g, g) ,= 0, choose =
(f,g)

(g,g)
, so
(f, f)
(f, g)

(g, g)
(f, g)
(f, g)
(g, g)
(g, f) +
(f, g)
(g, g)
(f, g)

(g, g)
(g, g) 0
(f, f) = (f, f)

(f, f)(g, g) (f, g)(g, f) = [(f, g)[

2
(13.1)
If (g, g) = 0, then g = 0, so (f, g) = (f, 0) = 0, and the equality is clearly satised.
13.2 Triangle inequality
|f +g| |f| +|g| and |f g| [ |f| |g| [
Follow from the basic properties of a scalar product, and the Schwarz inequality.
Proof:
0 |f +g|
2
=(f +g, f +g)
=(f, f) +(f, g) +

(g, f) +

(g, g)
For = 1,
|f +g|
2
=(f, f) + (g, g) + (f, g) + (f, g)

= |f|
2
+|g|
2
+ 2Re(f, g)
|f|
2
+|g|
2
+ 2[(f, g)[, since Re z [z[ for any complex z
|f|
2
+|g|
2
+ 2|f| |g|, from the Schwarz inequality
(|f| +|g|)
2
|f +g| |f| +|g|
116
CHAPTER 13. EXAPANSIONS 13.3. SCHMIDT ORTHOGONALIZATION
For = 1,
|f g|
2
=|f|
2
+|g|
2
2Re(f, g)
|f|
2
+|g|
2
2[(f, g)[
|f|
2
+|g|
2
2|f| |g|
=(|f| |g|g|)
2
|f g| [ |f| |g| [, Q.E.D.
Denition 1 f is normalized if and only if |f| = (f, f)
1
2
= 1
Denition 2 f and g are orthogonal if and only if (f, g) = 0
Denition 3 The set u
i
, i = 1, 2, . . . is said to be orthonormal if each u
i
is normalized and is orthogonal to the
other u
i
Denition 4 For an orthonormal set u
i
, (u
i
, u
j
) =
ij
13.3 Schmidt Orthogonalization Procedure
From any set of linearly independent functions (or vectors) v
i
, i = 1, 2, . . . one can construct the same number
of orthonormal functions u
i
as follows:
u
1
=
v
i
|v
1
|
, normalized
u
2
=
v
2
(u
1
, v
2
)u
1
|

|
, so normalized, at (u
2
, u
1
) = 0
u
3
=
v
3
(v
3
, u
1
)u
1
(v
3
, u
2
)u
2
|

|
, normalized, at (u
3
.u
1
) = 0, (u
3
, u
2
) = 0
u
i
=
v
1

i1
j=1
(v
i
, u
j
)u
j
|

|
= 0, normalized, at (u
1
, u
k
) = 0, k < i,
since
(u
i
, u
k
) =
(v
i
, u
k
)

i1
j=1
(v
i
, u
j
)

jk
..
(u
j
, u
k
)
| |
= 0, k < 1
None of the u
i
vanish (and hence they are normalizable) because the v
i
are linearly independent. Otherwise
the procedure might fail. The procedure is not unique, because the order in which one uses (lables) the original
v
i
is arbitrary.
(u
1
, v
2
(v
2
, u
1
)u
1
) =(u
1
, v
2
) (v
2
, u
1
)(u
1
, u
1
)
=(u
1
, v
2
) (u
1
, v
2
) = 0
117
13.4. EXPANSIONS OF FUNCTIONS CHAPTER 13. EXAPANSIONS
13.4 Expansions of Functions
Given an enumerably innite orthonormal set u
i
, i = 1, 2, . . ., it is sometimes possible to expand an arbitrary
function f (from a particular class of functions) in the innite series:
f(x) =

i=1
c
i
u
i
(x) for some range of x (or r).
The principal topic of this chapter is the investigation of the nature of such expansions, and the conditions
under which such a series converges to the function f, or represents it in some other useful way.
If there is a set of c
i
such that the series

i=1
c
i
u
i
(x) converges uniformly, and hence denes a continuous
function, and that function equals f(x), then the c
i
must be related to f by c
i
= (u
i
, f), since, if f =

i=1
c
i
u
i
,
and the series converges uniformly, so it can be integrated term-by-term,
_
b
a
u

j
fdx =(u
j
, f)
=

i=1
c
i
_
b
a
u

j
u
i
dx
=

i=1
c
i
(u
j
, u
i
. .

ji
)
=c
j
c
i
= (u
i
, f), called the Fourier coefcient or f with respect to the u
i
. If there is a weighting function w,
c
i
= (u
i
, f)
w
=
_
b
a
u

i
fwdx
This expression for the coefcients c
i
in a series

i
c
i
u
i
representing f can be arrived at in different mannor.
Suppose one wants to approximate f by a nite series

n
i=1
a
1
u
i
in the least squares sense, or in the mean,
by determining the a
i
such that
M (f
n

i=1
a
i
u
i
, f
n

i=1
a
i
u
i
)
w
=
_
b
a

f
n

i=1
a
i
u
i

2
wdx
is a minimum. For any choice of the a
i
, M 0, so
0 M =(f, f)
w

_
f,
n

i=1
a
i
u
i
_
w

_
n

i=1
a
i
u
i
, f
_
w
+
_
_
n

j=1
a
i
u
i
,
n

j=1
a
j
u
j
_
_
w
=(f, f)
w

i=1
a
i
(f, u
i
)
w

i=1
a

i
(u
i
, f)
w
+
n

i=1
n

j=1
a

i
a
j
(u
i
, u
j
)
w
. .

ij
Since (u
i
, u
j
)
w
=
ij
, (u
i
, f)
w
= c
i
, and (f, u
i
)
w
= (u
i
, f)

w
= c

i
,
0 M =(f, f)
w

i=1
a
i
c

i

n

i=1
a

i
c
i
+
n

i=1
a

i
a
i
=(f, f)
w
+
n

i=1
[a
i
c
i
[
2

i=1
[c
i
[
2
118
CHAPTER 13. EXAPANSIONS 13.4. EXPANSIONS OF FUNCTIONS
Clearly M is a minimum if a
i
= c
i
= (u
i
, f)
w
, the Fourier coefcient.
The minimum value is:
M
min
=|f
n

i=1
c
i
u
i
|
2
=
_
b
a

f
n

i=1
c
i
u
i

2
wdx
=(f, f)
w

i=1
[c
i
[
2
0
So
(f, f)
w

_
b
a
[f[
2
wdx
n

i=1
[c
i
[
2
Taking the limit as h :
_
b
a
[f[
2
wdx lim
n
n

i=1
[c
i
[
2
=

i=1
[c
i
[
2
Bessels inequality
Hence, if
_
b
a
[f[wdx exists, then the series

i=1
[c
i
[
2
converges, so lim
h
[c
i
[
2
= 0. Therefore, lim
h
c
i
=
0.
This remarkable result will be used later in the proof of the convergence of Fourier series.
Denition 5 The set u
i
, i = 1, 2, . . ., is said to be complete with respect to some class of functions if for any
function f of the class.
lim
n
|f
n

i=1
c
i
u
i
|
2
= 0
or equivalently if:
lim
n
_
b
a

f
n

i=1
c
i
u
i

2
wdx = 0 (13.2)
(Bessels Inequality becomes an equality) or
(f, f)
w
=

c=1
[c
i
[
2
(13.3)
The last equation is called a completeness relation.
Denition 6 The completeness relation is a special case of a more general realtion called Parsevals relation:
If the set u
i
is complete, and

f c
i
(u
i
, f)
w
, b
i
(u
i
, g)
w
, then (f, g)
w
=

i=1
c

i
b
i
.
u
1
complete means:
lim
n
|f
n

i
c
i
u
i
|
2
= 0
119
13.4. EXPANSIONS OF FUNCTIONS CHAPTER 13. EXAPANSIONS
Proof:
[(f, g)
w

i=1
c

i
b
i
[ =[(f, g)
w

i=1
(f, u
i
)
w
b
i
[
=[(f, g)
w
(f,
n

i=1
b
i
u
i
)
w
[
=[(f, g
n

i=1
b
i
u
i
)
w
[
|f| |g
n

i=1
b
i
u
i
|
from the Schwarz inequality [(f, g)[ |f| |g|
Hence:
lim
n
[(f, g)
w

i=1
c

i
b
i
[ |f| lim
n
|g
n

i=1
b
i
u
i
| = 0
lim
n
_
(f, g)
n

i=1
c

i
b
i
_
= 0
because
lim
n
|g
n

i=1
b
i
u
i
| = 0 since the u
i
are complete.
Hence (f, g)
w
=

i=1
c

i
b
i
It is important to note that, even if the set of u
i
is complete, so that
lim
n
_
b
a

f
n

i=1
c
i
u
i

2
wdx = 0 (13.4)
it does not necessarily mean that f =

i=1
c
i
u
i
, even almost everywhere. Sufcient conditions for the latters to
be true, in addition to the set u
i
being complete, are that the f and u
i
be continuous, and the series f =

i=1
c
i
u
i
converge uniformly.
Proof:
_
b
a

i=1
c
i
u
i

2
wdx =
_
b
a
_
_
f

f f

i=1
c
i
u
i

i=1
c

i
u

i
f +
_

i=1
c

i
u

i
_
_
_

j=1
c
j
u
j
_
_
_
_
wdx
=(f, f)w

i=1
c
i
(f, u
i
)

i=1
c

i
(u
i
, f) +

i=1

j=1
c

i
c
j
(u
i
, u
j
)
because

i=1
c
i
u
i
converges uniformly, so it is a continuous function, and the series can be integrated term-by-
term. Since c
i
= (u
i
, f), (u
i
, u
j
) = J
ij
,
_
a
b
[f

i=1
c
i
u
i
[
2
wdx = (f, f)
w

i=1
[c
i
[
2
if the system converges uniformly.
120
CHAPTER 13. EXAPANSIONS 13.4. EXPANSIONS OF FUNCTIONS
But also we know that
lim
n
_
b
a

i=1
c
i
u
i

2
wdx = (f, f)
w

i=1
[c
i
[
2
f

i=1
c
i
u
i
if the series converges uniformly and lim can then be moved inside the integration.
Hence, if the series converges uniformly,
_
b
a

i=1
c
i
u
i

2
wdx = lim
n
_
b
a

f
n

i=1
c
i
u
i

2
wdx
=0 if the set is complete
But if the series converges uniformly, it is a continuous function. If f is also continuous, f =

i=1
c
i
u
i
is
continuous, and the integral can vanish only if f

i=1
c
i
u
i
= 0, or f =

i=1
c
i
u
i
, a x b Q.E.D.
Theorem 1 (Dettman p.335) A uniformly convergent sequence of continuous functions converges to a continuous
function
Denition 7 The set of vectors (functions) u
i
, i = 1, 2, . . ., is said to be closed if any function orthogonal to all
the u
i
must have zero norm, and hence is zero almost everywhere. That is, if:
(u
i
, f) = 0 , then |f| = (f, f)
1
2
w
= 0
Theorem 2 If a set of u
i
is complete, then it is closed.
Proof: Since the set is complete,
lim
n
_
b
a

f
n

i=1
(u
i
, f)u
i

2
wdx = 0
So any function f orthogonal to all the u
i
satises:
lim
n
_
b
a
[f 0[
2
wdx =0
=
_
b
a
[f[
2
wdx
=|f|
2
|f| = 0
The converse is also true for square-integrable functions, but the proof requires some preliminary discussion.
121
13.4. EXPANSIONS OF FUNCTIONS CHAPTER 13. EXAPANSIONS
Denition 8 Denition: An innite sequence q
n
, n = 1, 2, . . . of vecotrs in a linear vector space (such as a
sequence of square-integrable functions) is said to form a Cauchy sequence if, given any > 0, there is an
integer N such that |g
n
g
m
| < if n, m > N.
Denition 9 A linear vector space is said to be complete if every Cauchy sequence of vectors in the space has a
limit vector in the space, such that lim
n
|g g
n
| = 0, where g is called the limit of the sequence g
n
.
Denition 10 A Hilbert space is an innitely-dimensional complex complete linear vector spece with a scaler
product.
Theorem 3 (Riesz-Fischer theorem) Every Cauchy sequence of square-integrable functions has a limit, which is
a square-integrable function.
Hence the space of square-integrable functions is a Hilbert space. We are now ready to prove the following:
Theorem 4 If the set of orthonormal functions u
i
, i = 1, 2, . . . is closed, then it is complete.
Proof: Let f be any square-integrable function, and g
n
f

n
i=1
c
i
u
i
, where c
i
= (u
i
, f). Then:
(u
k
, g
n
) =(u
k
, f
n

i=1
c
i
u
i
)
=(u
k
, f)
n

i=1
c
i
(u
k
, u
i
)
=(u
k
, f)
n

i=1
c
i

ki
, since (n
k
, u
i
) = J
ki
=0 if k n
Next consider:
|g
n
g
m
|
2
=|f
n

i=1
c
i
u
i
f +
m

i=1
c
i
u
i
|
2
=|
m

i=n+1
c
i
u
i
|
2
, where m > n
=
_
_
m

i=n+1
c
i
u
i
,
m

j=n+1
c
j
u
j
_
_
=
m

i=n+1
m

j=n+1
c

i
c
j
(u
i
, u
j
. .

ij
)
=
m

i=n+1
[c
i
[
2
But from Bessels inequality (f, f)

i=1
[c
i
[
2
, it follows that the series converges, so for any > 0
there is an integer N such that

m
i=n+1
[c
i
[
2
< for m > n > N. Hence the g
n
from a Cauchy sequence of
square-integrable functions, and hence there is a square-integrable function g such that
lim
n
|g g
n
| = 0 (Riesz-Fischer)
122
CHAPTER 13. EXAPANSIONS 13.4. EXPANSIONS OF FUNCTIONS
Therefore:
lim
n
|g
n
| = lim
n
|g
n
g +g|
lim
n
|g
n
g| +|g|
=|g|
Furthermore, for given k and n > k,
[(u
k
, g)[ =[(u
k
, g g
n
)[ since (u
k
, g
n
) = 0 for n > k
|u
k
| |g g
n
|
=|g g
n
| from the Schwarz inequality, and |u
k
| = 1
Hence:
lim
n
[(u
k
, g)[ lim
n
|g g
n
| = 0
so [(u
k
, g)[ = 0 and (u
k
, g) = 0
So the limit function g is orthogonal to all the u
k
(hence closed). Hence, if the set of u
k
is closed, then
|g| = 0. But then
lim
n
|g
n
| = |g| = 0
or
lim
n
_
b
a

f
n

i=1
c
i
u
i

2
wdx = 0 (means u
i
is complete)
and we have proved that the set of orthonormal u
i
is complete if it is closed.
Denition 11 A real function f(x) is piecewise continuous (or sectionally continuous) on an interval a x b
if the interval can be subdivided into a nite number of intervals in each of which f(x) is continuous and
approaches nite limits at the end.
If f(x) is a complex function of the real variable x, it is piecewise continuous and both its real and imaginary
parts are piecewise continuous.
Denition 12 A function f(x) is piecewise smooth if both f(x) and its derivative f

(x) are piecewise continu-

ous.
Example: f(x) is continuous, and f

(x) is piecewise continuous on a x b so f(x) is piecewise smooth.

Example: both f(x) and f

(x) are piecewise continuous, so f(x) is piecewise smooth.

Denition 13
f(x+) lim
0<0
f(x +)
Denition 14
f(x) lim
0<0
f(x )
Denition 15
f

(x) lim
x0
f(x + x) f(x)
x
which means that, given any > 0, there is a J > 0 such that f

(x) exists so that

(x)
f(x + x) f(x)
x

< if [x[ <

123
13.5. FOURIER SERIES CHAPTER 13. EXAPANSIONS
So in the denition f

(x) lim
x0
f(x+x)f(x)
x
, the limit must exist and be independent of whether
x 0 through positive or negative values.
Denition 16 The right-hand derivative of f(x) at x is
f

R
(x) lim
0<x0
f(x + x) f(x+)
x
Denition 17 The left-hand derivative is
f

R
(x) lim
0<x0
f(x) (x x)
x
At a point at which f

(x) exists, f

R
(x) = f

L
(x) = f

(x) If f(x) is piecewise smooth, then f

R
(x) = f

(x+)
and f

L
(x) = f

(x), and these exist at each point.

13.5 Fourier Series
Reference: Churchill, Fourier Series and Boundary Value Problems
The set of functions
1

2
,
1

cos nx,
1

sinnx, n = 1, 2, . . ., is orthonormal over any interval of length

2. Consider the interval x . The series best representing f(x) in the least-squares sense has
coefcients c
n
= (u
m
f), so the series is
_
1

2
, f
_
1

2
+

n=1
__
1

cos nx, f
_
1

cos nx +
_
1

sinnx, f
_
1

sinnx
_
=
1

2
(1, f) +

n=1
__
1

cos nx, f
_
cos nx +
1

(sinnx, f) sinnx
_
=
A
o
2
+

n=1
[A
n
cos nx +B
n
sinnx]
where
A
n
=
1

(cos nx, f) =
1

f(x) cos nxdx, n = 0, 1, 2, . . .

B
n
=
1

(sinnx, f) =
1

f(x) sinnxdx
This Trignometric series with the coefcients A
n
and B
n
is called the Fourier series for f(x) over x
13.6 Convergence Theorem for Fourier Series
If f(x) is piecewise continuous and periodic with period 2 for all x, and if, at points of discontinuity f(x) is
dened by:
f(x)
1
2
[f(x+) +f(x)] (13.5)
then
f(x) =
A
o
2

n=1
[A
n
cos nx +B
n
sinnx] (13.6)
124
CHAPTER 13. EXAPANSIONS 13.6. CONVERGENCE THEOREM FOR FOURIER SERIES
where
A
n
=
1

f(x) cos nxdx B

n
=
1

f(x) sinnxdx
at all points where f(x) has nite right and left derivatives. This includes all points if f(x) is piecewise smooth.
If f(x) is not periodic, the expansion is valid for x , with the series converging to
1
2
[f()+f(+)]
at x = .
Lemma 1 If
_

0
[g(x)[
2
dx exists, then
lim
n
_

0
g(x) cos Nxdx = 0 and lim
NR
_

0
g(x) sinNxdx = 0
Proof: The functions
_
2

sinNx, N = 1, 2, . . . are orthonormal over 0 x . Hence Bassels inequality

gives:
_

0
[g(x)[
2
dx

N=1

_

0
_
2

sinNxg(x)dx

2
so the series converges.
Hence
lim
n
_

0
sinNxg(x) dx = 0
Similarily, the set
1

,
_
2

cos Nx is orthonormal over 0 x , so

lim
n
_

0
cos Nxg(x) dx = 0 Q.E.D.
A
n
=
1

f(x) cos nxdx B

n
=
1

f(x) sinnxdx
We are now ready to prove the convergence Th. for Fourier series. The partial sum of the Fourier series is
S
N
(x) =
1
2
A
o
+
N

n=1
[A
n
cos nx +B
n
sinnx]
=
1
2
_

f(x

)dx

+
N

n=1
__
1

f(x

) cos nx

dx

_
cos nx +
_
1

f(x

) sinnx

dx

_
sinnx
_
=
1
2
_

dx

f(x

)
_
1 + 2
N

n=1
[cos nx

cos nx +sinnx

sinnx]
_
=
1
2
_

dx

f(x

)
_
1 + 2
N

n=1
cos ny
_
Change the variable of integration to y = x

x, so dy = dx

, and
S
N
(x) =
1
2
_
x
x
dy f(y +x)
_
1 + 2
N

n=1
cos ny
_
125
13.6. CONVERGENCE THEOREM FOR FOURIER SERIES CHAPTER 13. EXAPANSIONS
Since the integrand is periodic with period 2,
_
x
x
=
_

. Also, the series can be summed as follows:

F
N
=1 + 2
N

n=1
cos ny
=1 + 2
N

n=1
(e
iny
+e
iny
)
2
=
N

n=N
e
iny
(13.7)
so
e
iny
F
N
F
N
= e
i(N+1)y
e
iNy
F
N
=
e
i(N+1)y
e
iNy
e
iy
1
=
e
i
y
2
(e
i(N+
1
2
)y
e
i(N+
1
2
)y
)
e
i
y
2
(e
i
y
2 e
i
y
2 )
=
sin(N +
1
2
)y
sin
y
2
(13.8)
Hence
1 + 2
N

n=1
cos ny =
sin(N +
1
2
)y
sin
y
2
(13.9)
Therefore also
1

_

0
sin(N +
1
2
)y
sin
y
2
dy =
1

_

0
_
1 + 2
N

n=1
cos ny
_
dy
=
1

_
y + 2
N

n=1
sinny
n
_

0
=1
so
1

_

0
sin(N +
1
2
)y
sin
y
2
dy = 1 (13.10)
Using the Eq. (13.9) above in the last expression for S
N
(x) gives
S
N
(x) =
1
2
_

dy f(y +x)
sin(N +
1
2
)y
sin
y
2
=
1
2
_

0
dy f(y +x)
sin(N +
1
2
)y
sin
y
2
dy +
1
2
_
0

dy

f(y

+x)
sin(N +
1
2
)y

sin
y

2
126
CHAPTER 13. EXAPANSIONS 13.6. CONVERGENCE THEOREM FOR FOURIER SERIES
Replacing y

by y in the second integral,

S
N
(x) =
1
2
_

0
dy f(y +x)
sin(N +
1
2
)y
sin
y
2
dy +
1
2
_

0
dy f(y +x)
sin(N +
1
2
)y
sin
y
2
=
1
2
_

0
[f(x +y) +f(x y)]
sin(N +
1
2
)y
sin
y
2
dy (13.11)
From Eq. (13.10) it follows that
1
2
[f(x+) +f(x)] =
1
2
_

0
[f(x+) +f(x)]
sin(N +
1
2
)y
sin
y
2
dy (13.12)
Subtracting Eq. (13.12) from Eq. (13.11) we obtain:
S
N
(X)
1
2
[f(x+) +f(x)] =
1
2
_

0
[f(x +y) f(x+) +f(x y) f(x)]
sin(N +
1
2
)y
sin
y
2
dy
=
1

_

0
(y) sin(N +
1
2
)y dy (13.13)
where
(y)
[f(x +y) f(x+) +f(x y) f(x)]
2 sin
y
2
=
_
[f(x +y) f(x+)]
y

[f(x) f(x y)]
y
_
_
y
2
_
sin
y
2
(13.14)
From Eq. (13.13),
lim
N
_
S
N
(x)
1
2
[f(x+) +f(x)]
_
= lim
N
1

_

0
(y) sin(N +
1
2
)y dy
= lim
N
1

_

0
(y)
_
sinNy cos
1
2
y + cos Ny sin
1
2
y
_
dy
= lim
N
1

_

0
_
(y) cos
1
2
y
_
sinNydy + lim
N
1

_

0
_
(y) sin
1
2
y
_
cos Ny dy
=0 from the lemma (13.15)
if (y) cos
1
2
y and (y) sin
1
2
y are square integrable over 0 y , which is the case if (y) is piecewise
continuous for 0 y . From the rst of expressions Eq. (13.14) for (y) it is apparent that (y) is piecewise
continuous on 0 y except possibly at y = 0 where sin
y
2
vanishes. But from the second expression
Eq. (13.14),
lim
0<y0
(y) = lim
0<y0
_
[f(x +y) f(x+)]
y

[f(x) f(x y)]
y
_
(
y
2
)
sin(
y
2
)
=f

R
(x) f

L
(x) (13.16)
So, at points x for which f

R
(x) and f

L
(x) exist,
127
13.7. FOURIER SERIES FOR DIFFERENT INTERVALS CHAPTER 13. EXAPANSIONS
1
2
[f(x+) +f(x)] = lim
N
S
N
(x)
=
1
2
A
o
+

n=1
[A
n
cos nx +B
n
sinnx] (13.17)
This completes the proof of the convergence theorem.
f(x)
1
2
[f(x+) +f(x)] =
A
o
2
+

n=1

n=1
[A
n
cos nx +B
n
sinnx]
if f(x) is piecewise continuous and periodic (period 2)
13.7 Fourier series for different intervals
If g(t) is piecewise smooth and perioc with period 2, it has the Fourier series:
g(t) =
1
2
A
o
+

n=1
[A
n
cos nt +B
n
sinnt] (13.18)
where
A
n
=
1

g(t) cos nt dt B
n
=
1

g(t) sinnt dt
The function f(x) g(

L
x) is piecewise smooth and periodic with period 2L. Replacing t by
x
L
in the
Fourier series for g(t), we obtain
f(x) =
1
2
A
o
+

+n = 1

_
A
n
cos
nx
L
+B
n
sin
nx
L
_
(13.19)
where
A
n
=
1

_
L
L
g
_
x
L
__
cos
nx
L
_

L
dx
=
1

_
L
L
f(x) cos
nx
L
dx (13.20)
and
B
n
=
1

_
L
L
g
_
x
L
__
sin
nx
L
_

L
dx
=
1

_
L
L
f(x) sin
nx
L
dx (13.21)
For a piecewise smooth function f(x), which is periodic with period 2L, this Fourier series converges to
f(x)
1
2
[f(x) +f(x+)] at all x.
128
CHAPTER 13. EXAPANSIONS 13.7. FOURIER SERIES FOR DIFFERENT INTERVALS
A piecewise smooth function f(x) which is not periodic in x can be expanded in a Fourier series in the form
of Eqn 1 above, and the series will converge to f(x)
1
2
[f(x) + f(x+)] at all points on any open interval of
length 2L, say x
o
L x x
o
+L, if the coefcients are determined by:
A
n
=
1
L
_
xo+L
xoL
f(x) cos
nx
L
dx B
n
=
1
L
_
xo+L
xoL
f(x) sin
nx
L
dx (13.22)
This is because one can dene a function F(x) which is perioc with period 2L and which equals f(x) for
x
o
L < x < x
o
+L. F(x) has a Fourier series Eq. (13.19) which equals f(x) for x
o
L < x < x
o
+L. Since
F(x) is periodic with period 2L, the integrals in A
n
and B
n
can be taken over any interval of length 2L, and in
particular for x
o
L < x < x
o
+ L, where F(x) = f(x), giving the expressions (13.22). At x = x
o
L and
x = x
o
+L, the series converges to
1
2
[f(x
o
L+) +f(x
o
+L)].
In particular, for the interval L x L, a piecewise smooth f(x) can be expanded as:
f(x) =
1
2
A
o
+

n=1
_
A
n
cos
nx
L
+B
n
sin
nx
L
_
, (L x L) (13.23)
where
A
n
=
1

_
L
L
f(x) cos
nx
L
dx B
n
=
1

_
L
L
f(x) sin
nx
L
dx
If f(x) is an even function, all the B
n
= 0, and if it is an off function, all the A
n
= 0. A function which is
piecewise smooth for 0 x L can be expanded in a Fourier sine series.
f(x)
1
2
[f(x) +f(x+)]
=

n=1
B
n
sin
nx
L
(13.24)
where
B
n
=
2
L
_
L
0
f(x) sin
nx
L
dx (13.25)
and where the series vanishes at x = 0 and x = L, or the function can be expanded in a Fourier cosine series
f(x) =
1
2
A
o
+

n=1
A
n
cos
nx
L
(13.26)
where
A
n
=
2
L
_
L
0
f(x) cos
nx
L
dx (13.27)
and the series converge to
1
2
[f(x) +f(x+)] for 0 x L, to f(0+) for x = 0, and to f(L) at x = L.
These results follow from the fact that a piecewise smooth function dened on 0 x L can have its
denition extended to L x L in such a way that it is either an odd or an even function of x.
129
13.8. COMPLEX FORM OF THE FOURIER SERIES CHAPTER 13. EXAPANSIONS
13.8 Complex Form of the Fourier Series
Suppose that f(x) is either a real or complex function which is piecewise smooth on x
o
L x x
o
+L. The
Fourier series for f(x), valid on that interval, can be written:
f(x) =
1
2
A
o
+

n=1
_
A
n
cos
nx
L
+B
n
sin
nx
L
_
=
1
2
A
o
+

n=1
_
A
n
_
e
inx
L
+e
inx
L
2
_
+B
n
_
e
inx
L
e
inx
L
2
__
=
1
2
A
o
+

n=1
_
1
2
(A
n
iB
n
)e
inx
L
+
1
2
(A
n
+iB
n
)e

inx
L
_
(13.28)
Let
c
n

1
2
(A
n
iB
n
) =
1
2L
_
xo+L
xoL
_
cos
nx
L
c sin
nx
L
_
f(x)dx
c
n
=
1
2L
_
xo+L
xoL
f(x)e

inx
L
dx (13.29)
Then
1
2
A
o
=
1
2L
_
xo+L
xoL
f(x)dx = c
o
(13.30)
and
1
2
(A
n
+iB
n
) =
1
2L
_
xo+L
xoL
f(x)
_
cos
nx
L
+i sin
nx
L
_
dx
=
1
2L
_
xo+L
xoL
f(x)e
inx
L
dx
=C
n
(13.31)
Hence
f(x) = c
o
+

n=1
_
c
n
e
inx
L
+c
n
e

inx
L
_
(13.32)
or
f(x) =

n=
c
n
e
inx
L
x
o
L x x
o
+L
This is called the complex form of the Fourier series for f(x), whether or not f(x) itself is complex. It
converges to
1
2
[f(x) +f(x+)] for x
o
L < x < x
o
L
At x = x
o
L and x = x
o
+L, it converges to
1
2
[f(x
o
L+) +f(x
o
+L)].
130
CHAPTER 13. EXAPANSIONS 13.9. UNIFORM CONVERGENCE OF FOURIER SERIES
13.9 Uniform Convergence of Fourier Series
f(x) =
A
o
2
+

n=1
[A
n
cos nx +B
n
sinnx] (13.33)
([A[ +[B[)
2
=[A[
2
+[B[
2
+ 2[A[ [B[
([A[ [B[)
2
=[A[
2
+[B[
2
2[A[ [B[
([A[ +[B[)
2
+ ([A[ [B[)
2
=2([A[
2
+[B[
2
([A[ +[B[)
2
hence [A[ +[B[

2([A[
2
+[B[
2
)
1
2
[A
n
cos nx +B
n
sinnx[ [A
n
cos nx[ +[B
n
sinnx[
[A
n
[ +[B
n
[
[[A
n
[
2
+[B
n
[
2
]
1
2
=

2[A
2
n
+B
2
n
]
1
2
taking f(x) to be real, so the A
n
and B
n
are real. Hence the Fourier series for f(x) converges absolutely and
uniformly if the series

N
n=1
[A
2
n
+B
2
n
]
1
2
converges.
Let
S
N

N

n=1
[A
2
n
+B
2
n
]
1
2
=
N

n=1
1
n
[n
2
(A
2
n
+B
2
n
)]
1
2
=
N

n=1
_
1
n
2
..
|cn|
_1
2
[n
2
(A
2
n
+B
2
n
)
. .
|Dn|
]
1
2

_
N

n=1
1
n
2
__
N

n=1
n
2
(A
2
n
+B
2
n
)
_
1
2
[(c, D)[ |c| |D|
from the Schwarz inequality in a vector space in which vectors are N-Tuples of real numbers, such as c
(c
1
, c
2
. . . c
N
), d (d
1
, d
2
. . . d
N
, and the scalar product is dened by (c, d)

N
n=1
c
n
d
n
. The Schwarz
inequality is [(c, d)[ (c, c)
1
2
(d, d)
1
2
, or
N

n=1
c
n
d
n

_
N

n=1
c
2
n
N

n=1
d
2
n
_
1
2
(13.34)
With the choice
c
n
=
1
n
d
n
=[n
2
(A
2
n
+B
2
n
)]
1
2
= n(A
2
n
+B
2
n
)
1
2
(13.35)
131
13.10. DIFFERENTIATION OF FOURIER SERIES CHAPTER 13. EXAPANSIONS
N

n=1
(A
2
n
+B
2
n
)
1
2

_
N

n=1
1
n
2
N

n=1
n
2
(A
2
n
+B
2
n
)
_
1
2
(13.36)
If, for x , f(x) is continuous, f

(x) is piecewise continuous, and f() = f(), then the series

n=1
n
2
(A
2
n
+ B
2
n
) converges, as shown in problem (9). Since the series

n=1
1
n
2
also converges, it follows
that

n=1
(A
2
n
+B
2
n
)
1
2
converges, so the Fourier series for f(x) converges uniformly and absolutely.

n=1
1
n
2
=

2
6

n=1
1
n
p
=(p) (Reimann Zeta)
Hence we have proven:
Theorem 5 If f(x) is continuous and if it is only piecewise continuous and f

(x) is piecewise continuous for

x , and f() = f(), then the Fourier series for f(x) converges uniformly and absolutely on
x (or any interval that is continuous).
It can be shown, by a more complicated proof, that the Fourier series for a piecewise smooth function converges
uniformly and absolutely in any closed subinterval in which the function is continuous. Piecewise smooth
f(x), f

(x) are piecewise continuous.

13.10 Differentiation of Fourier Series
If f(x) is periodic and continuous, and if f

(x) and f

(x) are piecewise continuous, then the term-by-term

derivative of the Fourier series for f(x) equals f

(x) except at points where f

(x) is discontinuous.
Under these conditions both f(x) and f

(x) have convergent Fourier series, and the term-by-term derivation

of the series for f(x) equals the series for f

(x). Details are worked out in problem (12.)

13.11 Integration of Fourier Series
If f(x) is piecewise continuous for L x L, then, whether or not the Fourier series
Ao
2
+

n=1
_
A
n
cos
nx
L
+B
n
sin
nx
L

corresponding to f(x) converges, the term-by-term integral of the series from L to x L equals
_
x
L
f(x)dx.
Proof:
F(x)
_
x
L
f(x)dx
1
2
A
o
x
is a continuous function, with derivative F

(x) = f(x)
1
2
A
o
which is piecewise continuous on L x L
for f(x) piecewise continuous on L x L
Show F(x) periodic.
132
CHAPTER 13. EXAPANSIONS 13.11. INTEGRATION OF FOURIER SERIES
Also F(L) =
1
2
A
o
L, and
F(L) =
_
L
L
f(x)dx
1
2
A
o
L
=LA
o

1
2
A
o
L
=
1
2
A
o
L = F(L)
Hence F(x) has a Fourier series
F(x) =
a
o
2
+

n=1
_
a
n
cos
nx
L
+b
n
sin
nx
L
_
(13.37)
which converges to f(x) for all points on L x L. At x = L,
F(L) =
1
2
A
o
L
=
a
o
2
+

n=1
_
a
n
cos n
. .
(1)
n
+b
n
sinn
. .
0
_
=
a
o
2
+

n=1
a
n
(1)
n
Hence
a
o
2
=
1
2
A
o
L

n=1
a
n
(1)
n
(13.38)
Also, for n > 0
a
n
=
1
L
_
L
L
F(x) cos
nx
L
dx
=
1
L
__
F(x) sin
nx
L
n
L
_
L
L

_
L
L
F

(x) sin
nx
L
n
L
dx
_
=
1
n
_
L
L
F

(x) sin
nx
L
dx
=
1
n
_
L
L
_
f(x)
1
2
A
o
_
sin
nx
L
dx
=
L
n
1
L
_
L
L
f(x) sin
nx
L
dx
=
L
n
B
n
for n > 0
133
13.11. INTEGRATION OF FOURIER SERIES CHAPTER 13. EXAPANSIONS
And
b
n
=
1
L
_
L
L
F(x) sin
nx
L
dx
=
1
L
__
F(x)
cos
nx
L
n
L
_
L
L

_
L
L
F

(x)
cos
nx
L
n
L
dx
_
=
L
n
1
L
_
L
L
F

(x) cos
nx
L
dx
=
L
n
1
L
_
L
L
_
f(x)
1
2
A
o
_
cos
nx
L
dx
=
L
n
1
L
__
L
L
f(x) cos
nx
L
dx
1
2
A
o
_
sin
nx
L

L
L
n
L
_
=
L
n
A
n
for n > 0
Hence the Fourier series
F(x) =
a
o
2
+

n=1
_
a
n
cos
nx
L
+b
n
sin
nx
L
_
(13.39)
becomes
F(x) =
_
1
2
A
o
L

n=1
a
n
(1)
n
_
+

n=1
_
a
n
cos
nx
L
+b
n
sin
nx
L
_
=
1
2
A
o
L +

n=1
_
a
n
_
cos
nx
L
(1)
n
_
+b
n
sin
nx
L
_
=
1
2
A
o
L +

n=1
_

L
n
B
n
_
cos
nx
L
(1)
n
_
+
L
n
A
n
sin
nx
L
_
(13.40)
But F(x)
_
x
L
f(x)dx
1
2
A
o
x, so
_
x
L
f(x)dx =
1
2
A
o
(x +L) +

n=1
_
A
n
L
n
sin
nx
L
B
n
L
n
_
cos
nx
L
(1)
n
_
_
=
_
x
L
1
2
A
o
dx +

n=1
_
A
n
_
x
L
cos
nx
L
dx +B
n
_
x
L
sin
nx
L
dx
_
=Term-by-term integral of the Fourier series for f(x).
_
f(x) =
_
fourier series for f f(x) = four. ser.
f(x)
x
=

x
f. s.
f(x)piecewise continuous f(x)continuous f(x)periodic and
or piecewise continuous
f

(x)p.c. f

(x)p.c.
f

(x)p.c.
134
CHAPTER 13. EXAPANSIONS 13.12. FOURIER INTEGRAL REPRESENTATION
13.12 Fourier Integral Representation
If f(x) is piecewise smooth for L x L, it has a complex Fourier series.
F(x) =

n=
c
n
e
in
x
L
c
n

1
2L
_
L
L
f(x)e
in
x
L
dx
which converges to
1
2
[f(x+) +f(x)] for L < x < L.
We wish to consider what happens as L . We cannot naively put l = for c
n
because that gives
c
n
=
1
2
_

f(x)dx for all n, and the Fourier series becomes just

n=
c
n
. Putting c
n
in the series for f(x):
f(x) =

n=
_
1
2L
_
L
L
f(x

)
_
e
i
nx
L
=

n=
1
2L
_
L
L
dx

f(x

)e
i
n
L
(xx

)
(13.41)
Let k

L
. Then,
f(x) =

n=
k
2
_
L
L
dx

f(x

)e
i nk(xx

)
(13.42)
But
lim
k0

n
F(nk)k =
_

F(k)dk (13.43)
Hence, as L , k

L
0. Fourier integral expression for f(x):
f(x) =
1
2
_

dk
_

dx

f(x

)e
ik(xx

)
(13.44)
This proof is heuristic (suggestive, but not rigorous). The result can be written
f(x) =
1

2
_

dkF(k)e
ikx
(13.45)
where
F(k)
1

2
_

dx

f(x

)e
ikx

Fourier transform of f(x)

Before providing rigorously the Fourier integral theorem, we need to consider improper integrals of the
form:
I(k) =
_

a
f(x, k)dx
= lim
R
_
R
a
f(x, k)dx
. .
I
R
(k)
= lim
R
I
R
(k) (13.46)
or
135
13.13. M-TEST FOR UNIFORM CONVERGENCE CHAPTER 13. EXAPANSIONS
I(k) =
_

f(x, k)dx
= lim
R
_
R
R
F(x, k)dx
= lim
R
I
R
(k) (13.47)
I(k) is said to converge uniformly in k, for A k B, if given > o, one can nd Q, independent of k for
a k b, such that
[I(k) I
R
(k)[ < for R > Q
13.13 M-Test for Uniform Convergence
If [f(x, k)[ M(x) for A k B and if
_

a
M(x) exists, then
_

a
f(x, k)dx converges uniformly. A
uniformly convergent improper integral I(k) can be integrated under the integral sign:
_
B
A
dk
_

f(x, k)dx =
_

__
B
A
f(x, k)dk
_
dx (13.48)
These properties are analogous to those of a uniformly convergent innite series:
S(k) =

n=0
F
n
(k)
= lim
N
N

n=0
F
n
(k) (13.49)
An important simpler example is
I(k) =
_

F(x)e
ikx
dx (13.50)
If
_

[f(x)[dx exists, then I(k) converges uniformly in k, since [f(x)e

ikx
[ = [f(x)[, so one can take M(x) =
[f(x)[.
13.14 Fourier Integral Theorem
If f(x) is piecewise smooth for all nite intervals, if
_

[f(x)[dx exists, and if

F(k)
1

2
_

f(x)e
ikx
dx (13.51)
then
f(x)
1
2
[f(x+) +f(x)] =
1

2
_

F(k)e
ikx
dk (13.52)
F =
1

2
_

dxe
ikx

F
1
=
1

2
_

dke
ikx

Proof: Reference: Dettman, Chapter 8, or Churchill.
136
CHAPTER 13. EXAPANSIONS 13.14. FOURIER INTEGRAL THEOREM
Lemma 2 Riemanns Lemma. If f(x) is piecewise continuous for a x b, a and b nite, then lim
R
_
b
a
f(x) sinRxdx =
0
Proof: Since the integral can be split up into integrals of a continuous fct, Lemma 2 is true if it si true for
a continuous f(x), which we now assume. If we change the variable from x to t such that Rx = Rt + , so
x = t +

R
,
_
b
a
f(x) sinRxdx =
_
b

R
a

R
f
_
t +

R
_
sin(Rt +)
. .
sin Rt
dt
=
_
b

R
a

R
f
_
t +

R
_
sinRt dt (13.53)
Hence, adding these expressions:
2
_
b
a
f(x) sinRxdx =
_
b
a
f(x) sinRxdx
_
b

R
a

R
f
_
x +

R
_
sinRxdx
=
_
a
a

R
f
_
x +

R
_
sinRxdx

_
b

R
a
_
f
_
x +

R
_
f(x)
_
sinRxdx
+
_
b
b

R
f(x) sinRxdx
2

_
b
a
f(x) sinRxdx

_
a
a

R
f
_
x +

R
_
sinRxdx

_
b

R
a
_
f
_
x +

R
_
f(x)
_

_
b
b

R
f(x) sinRxdx

R
M +

b

R
a

f
_
x +

R
_
f(x)

max
+

R
M
for a x b

R
where [f(x)[ M for a x b. M is nite, and

f
_
x +

R
_
f(x)

max
0 as R
because f(x) is continuous. Hence
lim
R
2

_
b
a
f(x) sinRxdx

= 0
lim
R
_
b
a
f(x) sinRxdx = 0
Lemma 3 If f(x) is piecewise smooth in any nite interval and
_

[f(x)[dx < , then

lim
R
_
T
T
f(x +t)
sinRt
t
dt =

2
[f(x+) +f(x)] 0 < T
137
13.14. FOURIER INTEGRAL THEOREM CHAPTER 13. EXAPANSIONS
Proof: First consider nite T
I
T
lim
R
_
T
T
f(x +t)
sinRt
t
dt

2
[f(x+) +f(x)]
= lim
R
_
T
0
[f(x +t) +f(x t)]
sinRt
t
dt

2
[f(x+) +f(x)]
But
lim
R
_
T
0
sinRt
t
dt = lim
R
_
RT
0
sint

dt

=
_

0
sint
t
dt
=

2
as we showed last semester by contour integration.
Hence
I
T
= lim
R
_
T
0
_
f(x +t) f(x+)
t
+
f(x t) f(x)
t
_
sinRt dt
=0 from Lemma 2, since the integrand is piecewise
continuous for all nite t 0
lim
t0
_
f(x +t) f(x+)
t
+
f(x t) f(x)
t
_
= f

R
(x) f

L
(x) (13.54)
which exists nce f

(x) is piecewise continuous. This proves the Lemma for nite T. For T =
But
_

[f(x +t)[dt =
_

[f(t)[dt
so given any > 0 can nd T large enough so
_
_
T
R
+
_

T
_
f(t)dt <
Hence
lim
R

f(x +t)
sinRt
t
dt

2
[f(x+) +f(x)]

< + lim
R

_
T
T
f(x +t)
sinRt
t
dt

2
[f(x+) +f(x)]

. .
0, from proof for nite T, Lemma 2
=0, arbitrary, since
Q.E.D.
lim
R
_

f(x +t)
sinRt
t
dt =

2
[f(x+) +fx]
Proof of the Fourier Integral Theorem:
1

2
_

F(k)e
ikx
dk =
1

2
_

dk
_
1

2
_

f(x

)e
ikx

dx

_
e
ikx
=
1

2
_

dk
_

dx

f(x

)e
ik(xx

)
= lim
R
1

2
_
R
R
dk
_

dx

f(x

)e
ik(xx

)
138
CHAPTER 13. EXAPANSIONS 13.15. EXAMPLES OF THE FOURIER INTEGRAL THEOREM
But [f(x

)e
ik(xx

)
[ = [f(x

)[ and
_

dx

[f(x

)[ < , so the integral over x

converges uniformly in k, so
the order of the integration can be exchanged.
1

2
_

F(x)e
ikx
dk = lim
R
1
2
_

dx

f(x

)
_
R
R
dk e
ik(xx

)
= lim
R
1
2
_

dx

f(x

)
_
sink(x x

)
(x x

)
_
k=R
k=R
= lim
R
1

dx

f(x

)
sinR(x x

)
x x

= lim
R
1

dt f(x +t)
sinRt
t
=
1
2
[f(x+) +f(x)] from Lemma 3. Q.E.D.
We note for future reference that the Fourier transform F(k) of a piecewise smooth function f(x) statisfying
_

[f(x)[dx < is bounded:

F(k)
1

2
_

f(x)e
ikx
dx
so
[F(k)[
1

2
_

[f(x)e
ikx
[dx
=
1

2
_

[f(x)[dx
=B <
since
_

[f(x)[dx <
[F(k)[ B <
13.15 Examples of the Fourier Integral Theorem
f(x) = ce

x
2

2
, called a gaussian function. (13.55)
1

2
=
2
=
1

=
1

139
13.15. EXAMPLES OF THE FOURIER INTEGRAL THEOREM CHAPTER 13. EXAPANSIONS
F(k) =
1

2
_

ce

x
2

2
e
ikx
dx
=
c

2
_

2
_
x
2
+ik
2
x

dx
=
c

2
_

2
_
x
2
+ik
2
x +
_
ik
2
2
_
2

_
ik
2
2
_
2
_
dx
=
c

2
_

2
_
x +
ik
2
2
_
2
dxe

2
k
2

4
4
=
c

2
e

k
2

4
4
__

x
2

2
dx

_
. .

=
c

2
e

2
4
k
2
which is a gaussian function of k.
Note that
_

f(x)dx =

2F(0) =

2
c

2
= 1
if c =
1

.
So f(x) =
1

e
x
2

2
had F(k) =
1
sqrt2
e

2
4
k
2
x k
2

xk =
2

= 2, for all (13.56)

The narrower f(x), the broader F(k). This is typical of all functions and their Fourier transforms.
As 0, f(x) =
1

x
2

2
approaches the Dirac delta function, (x), which has the properties (x) = 0
if x ,= 0, but
_

(x)dx = 1, and
f(x) =
_

f(x

)(x

x)dx

if f(x) is continuous at x. The last property follows from

_

f(x

)(x

x)dx

=
_
x+
x
f(x

)(x

x)dx

= f(x)
_
x+
x
(x

x)dx

. .
1
= f(x)
The Fourier transform of (x x
o
) is
F(k) =
1

2
_

(x x
o
)e
ikx
dx =
1

2
e
ikxo
so the Fourier integral representation is:
(x x
o
) =
1

2
_

F(k)e
ikx
dk =
1
2
_

e
ck(xxo)
dk
140
CHAPTER 13. EXAPANSIONS 13.16. PARSEVALS THEOREM FOR FOURIER TRANSFORMS
This important result follows also from putting
F(k) =
1

2
_

f(x

)e
ikx

dx

(13.57)
in
f(x) =
1

2
_

F(k)e
ikx
dk
=
1

2
_

_
1

2
_

f(x

)e
ikx

dx

_
e
ikx
dk
=
1

2
_

dk
_

dx

f(x

)e
ck(xx

)
=
_

dx

f(x

)
_
1
2
_

e
ik
(x x

)dk
_
, exchanging order of integration
=
_

f(x

)(x

x)dx
(x

x) =
1
2
_

e
ik(xx

)
dk
=
1
2
_

0
cos k(x x

)dk
=12
_

e
ik(x

x)
dk
As another example, consider
f(x) =
_
e
ikox
, x
0, x < , x >
(13.58)
F(k) =
1
2
_

e
iko
e
ikx
dx
=
1
2
_

e
i(kko)x
dx =
1
2
_

cos(k k
o
)xdx
=
1
2
sin(k k
o
)x
(k k
o
)
[

=
_
2

sin(k k
o
)
(k k
o
)
=
_
2

sin(k k
o
)
(k k
o
)
xk =

=
13.16 Parsevals Theorem for Fourier Transforms
Suppose f(x) and g(x) are piecewise smooth for all nite intervals, and
_

[f(x)[dx and
_

[g(x)[dx exist, so
f(x) and g(x) have Fourier integral representations with Fourier transforms F(k) and G(k) which are bounded,
so in particular [G(k)[ < B for all k.
The Parseval theorem is that:
_

(x)f(x)dx =
_

(k)F(k)dk (13.59)
141
13.17. CONVOLUTION THEOREM FOR FOURIER TRANSFORMS CHAPTER 13. EXAPANSIONS
Proof:
_

(k)F(k)dk =
_

dkG

(k)
_
1

2
_

f(x)e
ikx
dx
_
=
1

2
_

dk
__

f(x)G

(k)e
ikx
dx
_
But the integral over x converges uniformly in k, since
[f(x)G

(k)e
ikx
[ [f(x)[B, M(x)
and
_

M(x)dx = B
_

[f(x)[dx <
Hence the order of integration can be changed:
_

(k)F(k)dk =
1

2
_

dxF(x)
__

(k)e
ikx
dk
_
(13.60)
=
_

dxF(x)
_
1

2
_

G(k)e
ikx
dk
_

=
_

dxf(x)g(x)

, Q.E.D.
For the special case g(x) = f(x),
_

[f(x)[
2
dx =
_

[F(k)[
2
dk (13.61)
13.17 Convolution Theorem for Fourier Transforms
If H(k) = G(k)F(k), where
G(k) =
1

2
_

g(x)e
ikx
dx and F(k) =
1

2
_

f(x)e
ikx
dx,
then
h(x) =
1

2
_

H(k)e
ikx
dk
=
1

2
_

G(k)F(k)e
ikx
dk
=
1

2
_

dk
_
1

2
_

g(x

)e
ikx

dx

_
F(k)e
ikx
=
1

2
_

dk
_

dx

g(x

)e
ik(xx

)
(k) (13.62)
The integral over x

converges uniformly in k, since

_

infty
dx

[g(x

)[
142
CHAPTER 13. EXAPANSIONS 13.18. FOURIER SINE AND COSINE TRANSFORMS AND REPRESENTATIONS
exists, and [F(k)[ B, so the order of integration can be exchanged.
h(x) =
1
2
_

dx

dkF(k)e
ik(xx

)
g(x

)
=
1

2
_

dx

g(x

)
_
1

2
_

F(k)e
ik(xx

)
dk
_
(13.63)
h(x) =
1

2
_

g(x

)f(x x

)dx

; Called the Fourier Convolution of f(x) and g(x)

If we change the variable of integration to x

= x x

, dx

= dx

h(x) =
1

2
_

g(x x

)f(x

)(dx

)
=
1

2
_

f(x

)g(x x

)dx

=
1

2
_

f(x

)g(x x

)dx

(13.64)
Hence the convolution is symmetric in the two functions f and g. The signicance of the convolution h(x)
is that it is the function whose Fourier transform is the product of the transforms of f and g.
Fh(x) =

h(k) = g(k)

f(k)
13.18 Fourier Sine and Cosine Transforms and Representations
F(k)
1

2
_

f(x)e
ikx
dx =
1

2
_

f(x)[cos kx sinkx]dx
Hence, if f(x) = f(x), an even function,
143
Bibliography
(1) ARFKEN, G. B., AND WEBER, H. J. Mathematical methods for physicists, 5 ed. Academic Press, San Diego, 2001.
(2) BAMBERG, P., AND STERNBERG, S. A course in mathematics for students of physics: 1. Cambridge University Press,
Cambridge, 1991.
(3) BAMBERG, P., AND STERNBERG, S. A course in mathematics for students of physics: 2. Cambridge University Press,
Cambridge, 1991.
(4) CHOQUET-BRUHAT, Y., DEWITT-MORETTE, C., AND DILLARD-BLEICK, M. Analysis, manifolds and physics. Part
I: basics, revised ed. North-Holland, Amsterdam, 1991.
(5) CHOQUET-BRUHAT, Y., AND DEWITT-MORETTE, C. Analysis, manifolds and physics. Part II: 92 applications.
North-Holland, Amsterdam, 1989.
(6) ALLEN, M., AND TILDESLEY, D. Computer Simulation of Liquids. Oxford University Press, 1987.
144